linux/fs/xfs/xfs_log_recover.c

3517 lines
97 KiB
C
Raw Normal View History

// SPDX-License-Identifier: GPL-2.0
/*
* Copyright (c) 2000-2006 Silicon Graphics, Inc.
* All Rights Reserved.
*/
#include "xfs.h"
#include "xfs_fs.h"
#include "xfs_shared.h"
#include "xfs_format.h"
#include "xfs_log_format.h"
#include "xfs_trans_resv.h"
#include "xfs_bit.h"
#include "xfs_sb.h"
#include "xfs_mount.h"
xfs: log recovery should replay deferred ops in order As part of testing log recovery with dm_log_writes, Amir Goldstein discovered an error in the deferred ops recovery that lead to corruption of the filesystem metadata if a reflink+rmap filesystem happened to shut down midway through a CoW remap: "This is what happens [after failed log recovery]: "Phase 1 - find and verify superblock... "Phase 2 - using internal log " - zero log... " - scan filesystem freespace and inode maps... " - found root inode chunk "Phase 3 - for each AG... " - scan (but don't clear) agi unlinked lists... " - process known inodes and perform inode discovery... " - agno = 0 "data fork in regular inode 134 claims CoW block 376 "correcting nextents for inode 134 "bad data fork in inode 134 "would have cleared inode 134" Hou Tao dissected the log contents of exactly such a crash: "According to the implementation of xfs_defer_finish(), these ops should be completed in the following sequence: "Have been done: "(1) CUI: Oper (160) "(2) BUI: Oper (161) "(3) CUD: Oper (194), for CUI Oper (160) "(4) RUI A: Oper (197), free rmap [0x155, 2, -9] "Should be done: "(5) BUD: for BUI Oper (161) "(6) RUI B: add rmap [0x155, 2, 137] "(7) RUD: for RUI A "(8) RUD: for RUI B "Actually be done by xlog_recover_process_intents() "(5) BUD: for BUI Oper (161) "(6) RUI B: add rmap [0x155, 2, 137] "(7) RUD: for RUI B "(8) RUD: for RUI A "So the rmap entry [0x155, 2, -9] for COW should be freed firstly, then a new rmap entry [0x155, 2, 137] will be added. However, as we can see from the log record in post_mount.log (generated after umount) and the trace print, the new rmap entry [0x155, 2, 137] are added firstly, then the rmap entry [0x155, 2, -9] are freed." When reconstructing the internal log state from the log items found on disk, it's required that deferred ops replay in exactly the same order that they would have had the filesystem not gone down. However, replaying unfinished deferred ops can create /more/ deferred ops. These new deferred ops are finished in the wrong order. This causes fs corruption and replay crashes, so let's create a single defer_ops to handle the subsequent ops created during replay, then use one single transaction at the end of log recovery to ensure that everything is replayed in the same order as they're supposed to be. Reported-by: Amir Goldstein <amir73il@gmail.com> Analyzed-by: Hou Tao <houtao1@huawei.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Tested-by: Amir Goldstein <amir73il@gmail.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-11-22 12:53:02 +08:00
#include "xfs_defer.h"
#include "xfs_inode.h"
#include "xfs_trans.h"
#include "xfs_log.h"
#include "xfs_log_priv.h"
#include "xfs_log_recover.h"
#include "xfs_trans_priv.h"
#include "xfs_alloc.h"
#include "xfs_ialloc.h"
xfs: event tracing support Convert the old xfs tracing support that could only be used with the out of tree kdb and xfsidbg patches to use the generic event tracer. To use it make sure CONFIG_EVENT_TRACING is enabled and then enable all xfs trace channels by: echo 1 > /sys/kernel/debug/tracing/events/xfs/enable or alternatively enable single events by just doing the same in one event subdirectory, e.g. echo 1 > /sys/kernel/debug/tracing/events/xfs/xfs_ihold/enable or set more complex filters, etc. In Documentation/trace/events.txt all this is desctribed in more detail. To reads the events do a cat /sys/kernel/debug/tracing/trace Compared to the last posting this patch converts the tracing mostly to the one tracepoint per callsite model that other users of the new tracing facility also employ. This allows a very fine-grained control of the tracing, a cleaner output of the traces and also enables the perf tool to use each tracepoint as a virtual performance counter, allowing us to e.g. count how often certain workloads git various spots in XFS. Take a look at http://lwn.net/Articles/346470/ for some examples. Also the btree tracing isn't included at all yet, as it will require additional core tracing features not in mainline yet, I plan to deliver it later. And the really nice thing about this patch is that it actually removes many lines of code while adding this nice functionality: fs/xfs/Makefile | 8 fs/xfs/linux-2.6/xfs_acl.c | 1 fs/xfs/linux-2.6/xfs_aops.c | 52 - fs/xfs/linux-2.6/xfs_aops.h | 2 fs/xfs/linux-2.6/xfs_buf.c | 117 +-- fs/xfs/linux-2.6/xfs_buf.h | 33 fs/xfs/linux-2.6/xfs_fs_subr.c | 3 fs/xfs/linux-2.6/xfs_ioctl.c | 1 fs/xfs/linux-2.6/xfs_ioctl32.c | 1 fs/xfs/linux-2.6/xfs_iops.c | 1 fs/xfs/linux-2.6/xfs_linux.h | 1 fs/xfs/linux-2.6/xfs_lrw.c | 87 -- fs/xfs/linux-2.6/xfs_lrw.h | 45 - fs/xfs/linux-2.6/xfs_super.c | 104 --- fs/xfs/linux-2.6/xfs_super.h | 7 fs/xfs/linux-2.6/xfs_sync.c | 1 fs/xfs/linux-2.6/xfs_trace.c | 75 ++ fs/xfs/linux-2.6/xfs_trace.h | 1369 +++++++++++++++++++++++++++++++++++++++++ fs/xfs/linux-2.6/xfs_vnode.h | 4 fs/xfs/quota/xfs_dquot.c | 110 --- fs/xfs/quota/xfs_dquot.h | 21 fs/xfs/quota/xfs_qm.c | 40 - fs/xfs/quota/xfs_qm_syscalls.c | 4 fs/xfs/support/ktrace.c | 323 --------- fs/xfs/support/ktrace.h | 85 -- fs/xfs/xfs.h | 16 fs/xfs/xfs_ag.h | 14 fs/xfs/xfs_alloc.c | 230 +----- fs/xfs/xfs_alloc.h | 27 fs/xfs/xfs_alloc_btree.c | 1 fs/xfs/xfs_attr.c | 107 --- fs/xfs/xfs_attr.h | 10 fs/xfs/xfs_attr_leaf.c | 14 fs/xfs/xfs_attr_sf.h | 40 - fs/xfs/xfs_bmap.c | 507 +++------------ fs/xfs/xfs_bmap.h | 49 - fs/xfs/xfs_bmap_btree.c | 6 fs/xfs/xfs_btree.c | 5 fs/xfs/xfs_btree_trace.h | 17 fs/xfs/xfs_buf_item.c | 87 -- fs/xfs/xfs_buf_item.h | 20 fs/xfs/xfs_da_btree.c | 3 fs/xfs/xfs_da_btree.h | 7 fs/xfs/xfs_dfrag.c | 2 fs/xfs/xfs_dir2.c | 8 fs/xfs/xfs_dir2_block.c | 20 fs/xfs/xfs_dir2_leaf.c | 21 fs/xfs/xfs_dir2_node.c | 27 fs/xfs/xfs_dir2_sf.c | 26 fs/xfs/xfs_dir2_trace.c | 216 ------ fs/xfs/xfs_dir2_trace.h | 72 -- fs/xfs/xfs_filestream.c | 8 fs/xfs/xfs_fsops.c | 2 fs/xfs/xfs_iget.c | 111 --- fs/xfs/xfs_inode.c | 67 -- fs/xfs/xfs_inode.h | 76 -- fs/xfs/xfs_inode_item.c | 5 fs/xfs/xfs_iomap.c | 85 -- fs/xfs/xfs_iomap.h | 8 fs/xfs/xfs_log.c | 181 +---- fs/xfs/xfs_log_priv.h | 20 fs/xfs/xfs_log_recover.c | 1 fs/xfs/xfs_mount.c | 2 fs/xfs/xfs_quota.h | 8 fs/xfs/xfs_rename.c | 1 fs/xfs/xfs_rtalloc.c | 1 fs/xfs/xfs_rw.c | 3 fs/xfs/xfs_trans.h | 47 + fs/xfs/xfs_trans_buf.c | 62 - fs/xfs/xfs_vnodeops.c | 8 70 files changed, 2151 insertions(+), 2592 deletions(-) Signed-off-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2009-12-15 07:14:59 +08:00
#include "xfs_trace.h"
#include "xfs_icache.h"
#include "xfs_error.h"
xfs: update metadata LSN in buffers during log recovery Log recovery is currently broken for v5 superblocks in that it never updates the metadata LSN of buffers written out during recovery. The metadata LSN is recorded in various bits of metadata to provide recovery ordering criteria that prevents transient corruption states reported by buffer write verifiers. Without such ordering logic, buffer updates can be replayed out of order and lead to false positive transient corruption states. This is generally not a corruption vector on its own, but corruption detection shuts down the filesystem and ultimately prevents a mount if it occurs during log recovery. This requires an xfs_repair run that clears the log and potentially loses filesystem updates. This problem is avoided in most cases as metadata writes during normal filesystem operation update the metadata LSN appropriately. The problem with log recovery not updating metadata LSNs manifests if the system happens to crash shortly after log recovery itself. In this scenario, it is possible for log recovery to complete all metadata I/O such that the filesystem is consistent. If a crash occurs after that point but before the log tail is pushed forward by subsequent operations, however, the next mount performs the same log recovery over again. If a buffer is updated multiple times in the dirty range of the log, an earlier update in the log might not be valid based on the current state of the associated buffer after all of the updates in the log had been replayed (before the previous crash). If a verifier happens to detect such a problem, the filesystem claims corruption and immediately shuts down. This commonly manifests in practice as directory block verifier failures such as the following, likely due to directory verifiers being particularly detailed in their checks as compared to most others: ... Mounting V5 Filesystem XFS (dm-0): Starting recovery (logdev: internal) XFS (dm-0): Internal error XFS_WANT_CORRUPTED_RETURN at line ... of \ file fs/xfs/libxfs/xfs_dir2_data.c. Caller xfs_dir3_data_verify ... ... Update log recovery to update the metadata LSN of recovered buffers. Since metadata LSNs are already updated by write verifer functions via attached log items, attach a dummy log item to the buffer during validation and explicitly set the LSN of the current transaction. This ensures that the metadata LSN of a buffer is updated based on whether the recovery I/O actually completes, and if so, that subsequent recovery attempts identify that the buffer is already up to date with respect to the current transaction. Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-09-26 06:34:27 +08:00
#include "xfs_buf_item.h"
#include "xfs_ag.h"
#include "xfs_quota.h"
xfs: only run COW extent recovery when there are no live extents As part of multiple customer escalations due to file data corruption after copy on write operations, I wrote some fstests that use fsstress to hammer on COW to shake things loose. Regrettably, I caught some filesystem shutdowns due to incorrect rmap operations with the following loop: mount <filesystem> # (0) fsstress <run only readonly ops> & # (1) while true; do fsstress <run all ops> mount -o remount,ro # (2) fsstress <run only readonly ops> mount -o remount,rw # (3) done When (2) happens, notice that (1) is still running. xfs_remount_ro will call xfs_blockgc_stop to walk the inode cache to free all the COW extents, but the blockgc mechanism races with (1)'s reader threads to take IOLOCKs and loses, which means that it doesn't clean them all out. Call such a file (A). When (3) happens, xfs_remount_rw calls xfs_reflink_recover_cow, which walks the ondisk refcount btree and frees any COW extent that it finds. This function does not check the inode cache, which means that incore COW forks of inode (A) is now inconsistent with the ondisk metadata. If one of those former COW extents are allocated and mapped into another file (B) and someone triggers a COW to the stale reservation in (A), A's dirty data will be written into (B) and once that's done, those blocks will be transferred to (A)'s data fork without bumping the refcount. The results are catastrophic -- file (B) and the refcount btree are now corrupt. In the first patch, we fixed the race condition in (2) so that (A) will always flush the COW fork. In this second patch, we move the _recover_cow call to the initial mount call in (0) for safety. As mentioned previously, xfs_reflink_recover_cow walks the refcount btree looking for COW staging extents, and frees them. This was intended to be run at mount time (when we know there are no live inodes) to clean up any leftover staging events that may have been left behind during an unclean shutdown. As a time "optimization" for readonly mounts, we deferred this to the ro->rw transition, not realizing that any failure to clean all COW forks during a rw->ro transition would result in catastrophic corruption. Therefore, remove this optimization and only run the recovery routine when we're guaranteed not to have any COW staging extents anywhere, which means we always run this at mount time. While we're at it, move the callsite to xfs_log_mount_finish because any refcount btree expansion (however unlikely given that we're removing records from the right side of the index) must be fed by a per-AG reservation, which doesn't exist in its current location. Fixes: 174edb0e46e5 ("xfs: store in-progress CoW allocations in the refcount btree") Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Chandan Babu R <chandan.babu@oracle.com> Reviewed-by: Dave Chinner <dchinner@redhat.com>
2021-12-16 03:52:23 +08:00
#include "xfs_reflink.h"
#define BLK_AVG(blk1, blk2) ((blk1+blk2) >> 1)
STATIC int
xlog_find_zeroed(
struct xlog *,
xfs_daddr_t *);
STATIC int
xlog_clear_stale_blocks(
struct xlog *,
xfs_lsn_t);
STATIC int
xlog_do_recovery_pass(
struct xlog *, xfs_daddr_t, xfs_daddr_t, int, xfs_daddr_t *);
/*
* Sector aligned buffer routines for buffer create/read/write/access
*/
/*
xfs: more robust recovery xlog buffer validation mkfs has a historical problem where it can format very small filesystems with too small of a physical log. Under certain conditions, log recovery of an associated filesystem can end up passing garbage parameter values to some of the cycle and log record verification functions due to bugs in log recovery not dealing with such filesystems properly. This results in attempts to read from bogus/underflowed log block addresses. Since the buffer read may ultimately succeed, log recovery can proceed with bogus data and otherwise go off the rails and crash. One example of this is a negative last_blk being passed to xlog_find_verify_log_record() causing us to skip the loop, pass a NULL head pointer to xlog_header_check_mount() and crash. Improve the xlog buffer verification to address this problem. We already verify xlog buffer length, so update this mechanism to also sanity check for a valid log relative block address and otherwise return an error. Pass a fixed, valid log block address from xlog_get_bp() since the target address will be validated when the buffer is read. This ensures that any bogus log block address/length calculations lead to graceful mount failure rather than risking a crash or worse if recovery proceeds with bogus data. Reported-by: Zorro Lang <zlang@redhat.com> Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-10-27 00:31:15 +08:00
* Verify the log-relative block number and length in basic blocks are valid for
* an operation involving the given XFS log buffer. Returns true if the fields
* are valid, false otherwise.
*/
xfs: more robust recovery xlog buffer validation mkfs has a historical problem where it can format very small filesystems with too small of a physical log. Under certain conditions, log recovery of an associated filesystem can end up passing garbage parameter values to some of the cycle and log record verification functions due to bugs in log recovery not dealing with such filesystems properly. This results in attempts to read from bogus/underflowed log block addresses. Since the buffer read may ultimately succeed, log recovery can proceed with bogus data and otherwise go off the rails and crash. One example of this is a negative last_blk being passed to xlog_find_verify_log_record() causing us to skip the loop, pass a NULL head pointer to xlog_header_check_mount() and crash. Improve the xlog buffer verification to address this problem. We already verify xlog buffer length, so update this mechanism to also sanity check for a valid log relative block address and otherwise return an error. Pass a fixed, valid log block address from xlog_get_bp() since the target address will be validated when the buffer is read. This ensures that any bogus log block address/length calculations lead to graceful mount failure rather than risking a crash or worse if recovery proceeds with bogus data. Reported-by: Zorro Lang <zlang@redhat.com> Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-10-27 00:31:15 +08:00
static inline bool
xlog_verify_bno(
struct xlog *log,
xfs: more robust recovery xlog buffer validation mkfs has a historical problem where it can format very small filesystems with too small of a physical log. Under certain conditions, log recovery of an associated filesystem can end up passing garbage parameter values to some of the cycle and log record verification functions due to bugs in log recovery not dealing with such filesystems properly. This results in attempts to read from bogus/underflowed log block addresses. Since the buffer read may ultimately succeed, log recovery can proceed with bogus data and otherwise go off the rails and crash. One example of this is a negative last_blk being passed to xlog_find_verify_log_record() causing us to skip the loop, pass a NULL head pointer to xlog_header_check_mount() and crash. Improve the xlog buffer verification to address this problem. We already verify xlog buffer length, so update this mechanism to also sanity check for a valid log relative block address and otherwise return an error. Pass a fixed, valid log block address from xlog_get_bp() since the target address will be validated when the buffer is read. This ensures that any bogus log block address/length calculations lead to graceful mount failure rather than risking a crash or worse if recovery proceeds with bogus data. Reported-by: Zorro Lang <zlang@redhat.com> Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-10-27 00:31:15 +08:00
xfs_daddr_t blk_no,
int bbcount)
{
xfs: more robust recovery xlog buffer validation mkfs has a historical problem where it can format very small filesystems with too small of a physical log. Under certain conditions, log recovery of an associated filesystem can end up passing garbage parameter values to some of the cycle and log record verification functions due to bugs in log recovery not dealing with such filesystems properly. This results in attempts to read from bogus/underflowed log block addresses. Since the buffer read may ultimately succeed, log recovery can proceed with bogus data and otherwise go off the rails and crash. One example of this is a negative last_blk being passed to xlog_find_verify_log_record() causing us to skip the loop, pass a NULL head pointer to xlog_header_check_mount() and crash. Improve the xlog buffer verification to address this problem. We already verify xlog buffer length, so update this mechanism to also sanity check for a valid log relative block address and otherwise return an error. Pass a fixed, valid log block address from xlog_get_bp() since the target address will be validated when the buffer is read. This ensures that any bogus log block address/length calculations lead to graceful mount failure rather than risking a crash or worse if recovery proceeds with bogus data. Reported-by: Zorro Lang <zlang@redhat.com> Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-10-27 00:31:15 +08:00
if (blk_no < 0 || blk_no >= log->l_logBBsize)
return false;
if (bbcount <= 0 || (blk_no + bbcount) > log->l_logBBsize)
return false;
return true;
}
/*
* Allocate a buffer to hold log data. The buffer needs to be able to map to
* a range of nbblks basic blocks at any valid offset within the log.
*/
static char *
xlog_alloc_buffer(
struct xlog *log,
int nbblks)
{
xfs: more robust recovery xlog buffer validation mkfs has a historical problem where it can format very small filesystems with too small of a physical log. Under certain conditions, log recovery of an associated filesystem can end up passing garbage parameter values to some of the cycle and log record verification functions due to bugs in log recovery not dealing with such filesystems properly. This results in attempts to read from bogus/underflowed log block addresses. Since the buffer read may ultimately succeed, log recovery can proceed with bogus data and otherwise go off the rails and crash. One example of this is a negative last_blk being passed to xlog_find_verify_log_record() causing us to skip the loop, pass a NULL head pointer to xlog_header_check_mount() and crash. Improve the xlog buffer verification to address this problem. We already verify xlog buffer length, so update this mechanism to also sanity check for a valid log relative block address and otherwise return an error. Pass a fixed, valid log block address from xlog_get_bp() since the target address will be validated when the buffer is read. This ensures that any bogus log block address/length calculations lead to graceful mount failure rather than risking a crash or worse if recovery proceeds with bogus data. Reported-by: Zorro Lang <zlang@redhat.com> Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-10-27 00:31:15 +08:00
/*
* Pass log block 0 since we don't have an addr yet, buffer will be
* verified on read.
*/
if (XFS_IS_CORRUPT(log->l_mp, !xlog_verify_bno(log, 0, nbblks))) {
xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
nbblks);
return NULL;
}
/*
* We do log I/O in units of log sectors (a power-of-2 multiple of the
* basic block size), so we round up the requested size to accommodate
* the basic blocks required for complete log sectors.
*
* In addition, the buffer may be used for a non-sector-aligned block
* offset, in which case an I/O of the requested size could extend
* beyond the end of the buffer. If the requested size is only 1 basic
* block it will never straddle a sector boundary, so this won't be an
* issue. Nor will this be a problem if the log I/O is done in basic
* blocks (sector size 1). But otherwise we extend the buffer by one
* extra log sector to ensure there's space to accommodate this
* possibility.
*/
if (nbblks > 1 && log->l_sectBBsize > 1)
nbblks += log->l_sectBBsize;
nbblks = round_up(nbblks, log->l_sectBBsize);
return kvzalloc(BBTOB(nbblks), GFP_KERNEL | __GFP_RETRY_MAYFAIL);
}
/*
* Return the address of the start of the given block number's data
* in a log buffer. The buffer covers a log sector-aligned region.
*/
static inline unsigned int
xlog_align(
struct xlog *log,
xfs_daddr_t blk_no)
{
return BBTOB(blk_no & ((xfs_daddr_t)log->l_sectBBsize - 1));
}
static int
xlog_do_io(
struct xlog *log,
xfs_daddr_t blk_no,
unsigned int nbblks,
char *data,
enum req_op op)
{
int error;
if (XFS_IS_CORRUPT(log->l_mp, !xlog_verify_bno(log, blk_no, nbblks))) {
xfs: more robust recovery xlog buffer validation mkfs has a historical problem where it can format very small filesystems with too small of a physical log. Under certain conditions, log recovery of an associated filesystem can end up passing garbage parameter values to some of the cycle and log record verification functions due to bugs in log recovery not dealing with such filesystems properly. This results in attempts to read from bogus/underflowed log block addresses. Since the buffer read may ultimately succeed, log recovery can proceed with bogus data and otherwise go off the rails and crash. One example of this is a negative last_blk being passed to xlog_find_verify_log_record() causing us to skip the loop, pass a NULL head pointer to xlog_header_check_mount() and crash. Improve the xlog buffer verification to address this problem. We already verify xlog buffer length, so update this mechanism to also sanity check for a valid log relative block address and otherwise return an error. Pass a fixed, valid log block address from xlog_get_bp() since the target address will be validated when the buffer is read. This ensures that any bogus log block address/length calculations lead to graceful mount failure rather than risking a crash or worse if recovery proceeds with bogus data. Reported-by: Zorro Lang <zlang@redhat.com> Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-10-27 00:31:15 +08:00
xfs_warn(log->l_mp,
"Invalid log block/length (0x%llx, 0x%x) for buffer",
blk_no, nbblks);
return -EFSCORRUPTED;
}
blk_no = round_down(blk_no, log->l_sectBBsize);
nbblks = round_up(nbblks, log->l_sectBBsize);
ASSERT(nbblks > 0);
error = xfs_rw_bdev(log->l_targ->bt_bdev, log->l_logBBstart + blk_no,
BBTOB(nbblks), data, op);
if (error && !xlog_is_shutdown(log)) {
xfs_alert(log->l_mp,
"log recovery %s I/O error at daddr 0x%llx len %d error %d",
op == REQ_OP_WRITE ? "write" : "read",
blk_no, nbblks, error);
}
return error;
}
STATIC int
xlog_bread_noalign(
struct xlog *log,
xfs_daddr_t blk_no,
int nbblks,
char *data)
{
return xlog_do_io(log, blk_no, nbblks, data, REQ_OP_READ);
}
STATIC int
xlog_bread(
struct xlog *log,
xfs_daddr_t blk_no,
int nbblks,
char *data,
char **offset)
{
int error;
error = xlog_do_io(log, blk_no, nbblks, data, REQ_OP_READ);
if (!error)
*offset = data + xlog_align(log, blk_no);
return error;
}
STATIC int
xlog_bwrite(
struct xlog *log,
xfs_daddr_t blk_no,
int nbblks,
char *data)
{
return xlog_do_io(log, blk_no, nbblks, data, REQ_OP_WRITE);
}
#ifdef DEBUG
/*
* dump debug superblock and log record information
*/
STATIC void
xlog_header_check_dump(
xfs_mount_t *mp,
xlog_rec_header_t *head)
{
xfs_debug(mp, "%s: SB : uuid = %pU, fmt = %d",
__func__, &mp->m_sb.sb_uuid, XLOG_FMT);
xfs_debug(mp, " log : uuid = %pU, fmt = %d",
&head->h_fs_uuid, be32_to_cpu(head->h_fmt));
}
#else
#define xlog_header_check_dump(mp, head)
#endif
/*
* check log record header for recovery
*/
STATIC int
xlog_header_check_recover(
xfs_mount_t *mp,
xlog_rec_header_t *head)
{
ASSERT(head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM));
/*
* IRIX doesn't write the h_fmt field and leaves it zeroed
* (XLOG_FMT_UNKNOWN). This stops us from trying to recover
* a dirty log created in IRIX.
*/
if (XFS_IS_CORRUPT(mp, head->h_fmt != cpu_to_be32(XLOG_FMT))) {
xfs_warn(mp,
"dirty log written in incompatible format - can't recover");
xlog_header_check_dump(mp, head);
return -EFSCORRUPTED;
}
if (XFS_IS_CORRUPT(mp, !uuid_equal(&mp->m_sb.sb_uuid,
&head->h_fs_uuid))) {
xfs_warn(mp,
"dirty log entry has mismatched uuid - can't recover");
xlog_header_check_dump(mp, head);
return -EFSCORRUPTED;
}
return 0;
}
/*
* read the head block of the log and check the header
*/
STATIC int
xlog_header_check_mount(
xfs_mount_t *mp,
xlog_rec_header_t *head)
{
ASSERT(head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM));
if (uuid_is_null(&head->h_fs_uuid)) {
/*
* IRIX doesn't write the h_fs_uuid or h_fmt fields. If
* h_fs_uuid is null, we assume this log was last mounted
* by IRIX and continue.
*/
xfs_warn(mp, "null uuid in log - IRIX style log");
} else if (XFS_IS_CORRUPT(mp, !uuid_equal(&mp->m_sb.sb_uuid,
&head->h_fs_uuid))) {
xfs_warn(mp, "log has mismatched uuid - can't recover");
xlog_header_check_dump(mp, head);
return -EFSCORRUPTED;
}
return 0;
}
/*
* This routine finds (to an approximation) the first block in the physical
* log which contains the given cycle. It uses a binary search algorithm.
* Note that the algorithm can not be perfect because the disk will not
* necessarily be perfect.
*/
STATIC int
xlog_find_cycle_start(
struct xlog *log,
char *buffer,
xfs_daddr_t first_blk,
xfs_daddr_t *last_blk,
uint cycle)
{
char *offset;
xfs_daddr_t mid_blk;
xfs_daddr_t end_blk;
uint mid_cycle;
int error;
end_blk = *last_blk;
mid_blk = BLK_AVG(first_blk, end_blk);
while (mid_blk != first_blk && mid_blk != end_blk) {
error = xlog_bread(log, mid_blk, 1, buffer, &offset);
if (error)
return error;
mid_cycle = xlog_get_cycle(offset);
if (mid_cycle == cycle)
end_blk = mid_blk; /* last_half_cycle == mid_cycle */
else
first_blk = mid_blk; /* first_half_cycle == mid_cycle */
mid_blk = BLK_AVG(first_blk, end_blk);
}
ASSERT((mid_blk == first_blk && mid_blk+1 == end_blk) ||
(mid_blk == end_blk && mid_blk-1 == first_blk));
*last_blk = end_blk;
return 0;
}
/*
* Check that a range of blocks does not contain stop_on_cycle_no.
* Fill in *new_blk with the block offset where such a block is
* found, or with -1 (an invalid block number) if there is no such
* block in the range. The scan needs to occur from front to back
* and the pointer into the region must be updated since a later
* routine will need to perform another test.
*/
STATIC int
xlog_find_verify_cycle(
struct xlog *log,
xfs_daddr_t start_blk,
int nbblks,
uint stop_on_cycle_no,
xfs_daddr_t *new_blk)
{
xfs_daddr_t i, j;
uint cycle;
char *buffer;
xfs_daddr_t bufblks;
char *buf = NULL;
int error = 0;
/*
* Greedily allocate a buffer big enough to handle the full
* range of basic blocks we'll be examining. If that fails,
* try a smaller size. We need to be able to read at least
* a log sector, or we're out of luck.
*/
bufblks = roundup_pow_of_two(nbblks);
while (bufblks > log->l_logBBsize)
bufblks >>= 1;
while (!(buffer = xlog_alloc_buffer(log, bufblks))) {
bufblks >>= 1;
if (bufblks < log->l_sectBBsize)
return -ENOMEM;
}
for (i = start_blk; i < start_blk + nbblks; i += bufblks) {
int bcount;
bcount = min(bufblks, (start_blk + nbblks - i));
error = xlog_bread(log, i, bcount, buffer, &buf);
if (error)
goto out;
for (j = 0; j < bcount; j++) {
cycle = xlog_get_cycle(buf);
if (cycle == stop_on_cycle_no) {
*new_blk = i+j;
goto out;
}
buf += BBSIZE;
}
}
*new_blk = -1;
out:
kmem_free(buffer);
return error;
}
static inline int
xlog_logrec_hblks(struct xlog *log, struct xlog_rec_header *rh)
{
if (xfs_has_logv2(log->l_mp)) {
int h_size = be32_to_cpu(rh->h_size);
if ((be32_to_cpu(rh->h_version) & XLOG_VERSION_2) &&
h_size > XLOG_HEADER_CYCLE_SIZE)
return DIV_ROUND_UP(h_size, XLOG_HEADER_CYCLE_SIZE);
}
return 1;
}
/*
* Potentially backup over partial log record write.
*
* In the typical case, last_blk is the number of the block directly after
* a good log record. Therefore, we subtract one to get the block number
* of the last block in the given buffer. extra_bblks contains the number
* of blocks we would have read on a previous read. This happens when the
* last log record is split over the end of the physical log.
*
* extra_bblks is the number of blocks potentially verified on a previous
* call to this routine.
*/
STATIC int
xlog_find_verify_log_record(
struct xlog *log,
xfs_daddr_t start_blk,
xfs_daddr_t *last_blk,
int extra_bblks)
{
xfs_daddr_t i;
char *buffer;
char *offset = NULL;
xlog_rec_header_t *head = NULL;
int error = 0;
int smallmem = 0;
int num_blks = *last_blk - start_blk;
int xhdrs;
ASSERT(start_blk != 0 || *last_blk != start_blk);
buffer = xlog_alloc_buffer(log, num_blks);
if (!buffer) {
buffer = xlog_alloc_buffer(log, 1);
if (!buffer)
return -ENOMEM;
smallmem = 1;
} else {
error = xlog_bread(log, start_blk, num_blks, buffer, &offset);
if (error)
goto out;
offset += ((num_blks - 1) << BBSHIFT);
}
for (i = (*last_blk) - 1; i >= 0; i--) {
if (i < start_blk) {
/* valid log record not found */
xfs_warn(log->l_mp,
"Log inconsistent (didn't find previous header)");
ASSERT(0);
error = -EFSCORRUPTED;
goto out;
}
if (smallmem) {
error = xlog_bread(log, i, 1, buffer, &offset);
if (error)
goto out;
}
head = (xlog_rec_header_t *)offset;
if (head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM))
break;
if (!smallmem)
offset -= BBSIZE;
}
/*
* We hit the beginning of the physical log & still no header. Return
* to caller. If caller can handle a return of -1, then this routine
* will be called again for the end of the physical log.
*/
if (i == -1) {
error = 1;
goto out;
}
/*
* We have the final block of the good log (the first block
* of the log record _before_ the head. So we check the uuid.
*/
if ((error = xlog_header_check_mount(log->l_mp, head)))
goto out;
/*
* We may have found a log record header before we expected one.
* last_blk will be the 1st block # with a given cycle #. We may end
* up reading an entire log record. In this case, we don't want to
* reset last_blk. Only when last_blk points in the middle of a log
* record do we update last_blk.
*/
xhdrs = xlog_logrec_hblks(log, head);
if (*last_blk - i + extra_bblks !=
BTOBB(be32_to_cpu(head->h_len)) + xhdrs)
*last_blk = i;
out:
kmem_free(buffer);
return error;
}
/*
* Head is defined to be the point of the log where the next log write
* could go. This means that incomplete LR writes at the end are
* eliminated when calculating the head. We aren't guaranteed that previous
* LR have complete transactions. We only know that a cycle number of
* current cycle number -1 won't be present in the log if we start writing
* from our current block number.
*
* last_blk contains the block number of the first block with a given
* cycle number.
*
* Return: zero if normal, non-zero if error.
*/
STATIC int
xlog_find_head(
struct xlog *log,
xfs_daddr_t *return_head_blk)
{
char *buffer;
char *offset;
xfs_daddr_t new_blk, first_blk, start_blk, last_blk, head_blk;
int num_scan_bblks;
uint first_half_cycle, last_half_cycle;
uint stop_on_cycle;
int error, log_bbnum = log->l_logBBsize;
/* Is the end of the log device zeroed? */
error = xlog_find_zeroed(log, &first_blk);
if (error < 0) {
xfs_warn(log->l_mp, "empty log check failed");
return error;
}
if (error == 1) {
*return_head_blk = first_blk;
/* Is the whole lot zeroed? */
if (!first_blk) {
/* Linux XFS shouldn't generate totally zeroed logs -
* mkfs etc write a dummy unmount record to a fresh
* log so we can store the uuid in there
*/
xfs_warn(log->l_mp, "totally zeroed log");
}
return 0;
}
first_blk = 0; /* get cycle # of 1st block */
buffer = xlog_alloc_buffer(log, 1);
if (!buffer)
return -ENOMEM;
error = xlog_bread(log, 0, 1, buffer, &offset);
if (error)
goto out_free_buffer;
first_half_cycle = xlog_get_cycle(offset);
last_blk = head_blk = log_bbnum - 1; /* get cycle # of last block */
error = xlog_bread(log, last_blk, 1, buffer, &offset);
if (error)
goto out_free_buffer;
last_half_cycle = xlog_get_cycle(offset);
ASSERT(last_half_cycle != 0);
/*
* If the 1st half cycle number is equal to the last half cycle number,
* then the entire log is stamped with the same cycle number. In this
* case, head_blk can't be set to zero (which makes sense). The below
* math doesn't work out properly with head_blk equal to zero. Instead,
* we set it to log_bbnum which is an invalid block number, but this
* value makes the math correct. If head_blk doesn't changed through
* all the tests below, *head_blk is set to zero at the very end rather
* than log_bbnum. In a sense, log_bbnum and zero are the same block
* in a circular file.
*/
if (first_half_cycle == last_half_cycle) {
/*
* In this case we believe that the entire log should have
* cycle number last_half_cycle. We need to scan backwards
* from the end verifying that there are no holes still
* containing last_half_cycle - 1. If we find such a hole,
* then the start of that hole will be the new head. The
* simple case looks like
* x | x ... | x - 1 | x
* Another case that fits this picture would be
* x | x + 1 | x ... | x
* In this case the head really is somewhere at the end of the
* log, as one of the latest writes at the beginning was
* incomplete.
* One more case is
* x | x + 1 | x ... | x - 1 | x
* This is really the combination of the above two cases, and
* the head has to end up at the start of the x-1 hole at the
* end of the log.
*
* In the 256k log case, we will read from the beginning to the
* end of the log and search for cycle numbers equal to x-1.
* We don't worry about the x+1 blocks that we encounter,
* because we know that they cannot be the head since the log
* started with x.
*/
head_blk = log_bbnum;
stop_on_cycle = last_half_cycle - 1;
} else {
/*
* In this case we want to find the first block with cycle
* number matching last_half_cycle. We expect the log to be
* some variation on
* x + 1 ... | x ... | x
* The first block with cycle number x (last_half_cycle) will
* be where the new head belongs. First we do a binary search
* for the first occurrence of last_half_cycle. The binary
* search may not be totally accurate, so then we scan back
* from there looking for occurrences of last_half_cycle before
* us. If that backwards scan wraps around the beginning of
* the log, then we look for occurrences of last_half_cycle - 1
* at the end of the log. The cases we're looking for look
* like
* v binary search stopped here
* x + 1 ... | x | x + 1 | x ... | x
* ^ but we want to locate this spot
* or
* <---------> less than scan distance
* x + 1 ... | x ... | x - 1 | x
* ^ we want to locate this spot
*/
stop_on_cycle = last_half_cycle;
error = xlog_find_cycle_start(log, buffer, first_blk, &head_blk,
last_half_cycle);
if (error)
goto out_free_buffer;
}
/*
* Now validate the answer. Scan back some number of maximum possible
* blocks and make sure each one has the expected cycle number. The
* maximum is determined by the total possible amount of buffering
* in the in-core log. The following number can be made tighter if
* we actually look at the block size of the filesystem.
*/
num_scan_bblks = min_t(int, log_bbnum, XLOG_TOTAL_REC_SHIFT(log));
if (head_blk >= num_scan_bblks) {
/*
* We are guaranteed that the entire check can be performed
* in one buffer.
*/
start_blk = head_blk - num_scan_bblks;
if ((error = xlog_find_verify_cycle(log,
start_blk, num_scan_bblks,
stop_on_cycle, &new_blk)))
goto out_free_buffer;
if (new_blk != -1)
head_blk = new_blk;
} else { /* need to read 2 parts of log */
/*
* We are going to scan backwards in the log in two parts.
* First we scan the physical end of the log. In this part
* of the log, we are looking for blocks with cycle number
* last_half_cycle - 1.
* If we find one, then we know that the log starts there, as
* we've found a hole that didn't get written in going around
* the end of the physical log. The simple case for this is
* x + 1 ... | x ... | x - 1 | x
* <---------> less than scan distance
* If all of the blocks at the end of the log have cycle number
* last_half_cycle, then we check the blocks at the start of
* the log looking for occurrences of last_half_cycle. If we
* find one, then our current estimate for the location of the
* first occurrence of last_half_cycle is wrong and we move
* back to the hole we've found. This case looks like
* x + 1 ... | x | x + 1 | x ...
* ^ binary search stopped here
* Another case we need to handle that only occurs in 256k
* logs is
* x + 1 ... | x ... | x+1 | x ...
* ^ binary search stops here
* In a 256k log, the scan at the end of the log will see the
* x + 1 blocks. We need to skip past those since that is
* certainly not the head of the log. By searching for
* last_half_cycle-1 we accomplish that.
*/
ASSERT(head_blk <= INT_MAX &&
(xfs_daddr_t) num_scan_bblks >= head_blk);
start_blk = log_bbnum - (num_scan_bblks - head_blk);
if ((error = xlog_find_verify_cycle(log, start_blk,
num_scan_bblks - (int)head_blk,
(stop_on_cycle - 1), &new_blk)))
goto out_free_buffer;
if (new_blk != -1) {
head_blk = new_blk;
goto validate_head;
}
/*
* Scan beginning of log now. The last part of the physical
* log is good. This scan needs to verify that it doesn't find
* the last_half_cycle.
*/
start_blk = 0;
ASSERT(head_blk <= INT_MAX);
if ((error = xlog_find_verify_cycle(log,
start_blk, (int)head_blk,
stop_on_cycle, &new_blk)))
goto out_free_buffer;
if (new_blk != -1)
head_blk = new_blk;
}
validate_head:
/*
* Now we need to make sure head_blk is not pointing to a block in
* the middle of a log record.
*/
num_scan_bblks = XLOG_REC_SHIFT(log);
if (head_blk >= num_scan_bblks) {
start_blk = head_blk - num_scan_bblks; /* don't read head_blk */
/* start ptr at last block ptr before head_blk */
error = xlog_find_verify_log_record(log, start_blk, &head_blk, 0);
if (error == 1)
error = -EIO;
if (error)
goto out_free_buffer;
} else {
start_blk = 0;
ASSERT(head_blk <= INT_MAX);
error = xlog_find_verify_log_record(log, start_blk, &head_blk, 0);
if (error < 0)
goto out_free_buffer;
if (error == 1) {
/* We hit the beginning of the log during our search */
start_blk = log_bbnum - (num_scan_bblks - head_blk);
new_blk = log_bbnum;
ASSERT(start_blk <= INT_MAX &&
(xfs_daddr_t) log_bbnum-start_blk >= 0);
ASSERT(head_blk <= INT_MAX);
error = xlog_find_verify_log_record(log, start_blk,
&new_blk, (int)head_blk);
if (error == 1)
error = -EIO;
if (error)
goto out_free_buffer;
if (new_blk != log_bbnum)
head_blk = new_blk;
} else if (error)
goto out_free_buffer;
}
kmem_free(buffer);
if (head_blk == log_bbnum)
*return_head_blk = 0;
else
*return_head_blk = head_blk;
/*
* When returning here, we have a good block number. Bad block
* means that during a previous crash, we didn't have a clean break
* from cycle number N to cycle number N-1. In this case, we need
* to find the first block with cycle number N-1.
*/
return 0;
out_free_buffer:
kmem_free(buffer);
if (error)
xfs_warn(log->l_mp, "failed to find log head");
return error;
}
/*
* Seek backwards in the log for log record headers.
*
* Given a starting log block, walk backwards until we find the provided number
* of records or hit the provided tail block. The return value is the number of
* records encountered or a negative error code. The log block and buffer
* pointer of the last record seen are returned in rblk and rhead respectively.
*/
STATIC int
xlog_rseek_logrec_hdr(
struct xlog *log,
xfs_daddr_t head_blk,
xfs_daddr_t tail_blk,
int count,
char *buffer,
xfs_daddr_t *rblk,
struct xlog_rec_header **rhead,
bool *wrapped)
{
int i;
int error;
int found = 0;
char *offset = NULL;
xfs_daddr_t end_blk;
*wrapped = false;
/*
* Walk backwards from the head block until we hit the tail or the first
* block in the log.
*/
end_blk = head_blk > tail_blk ? tail_blk : 0;
for (i = (int) head_blk - 1; i >= end_blk; i--) {
error = xlog_bread(log, i, 1, buffer, &offset);
if (error)
goto out_error;
if (*(__be32 *) offset == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
*rblk = i;
*rhead = (struct xlog_rec_header *) offset;
if (++found == count)
break;
}
}
/*
* If we haven't hit the tail block or the log record header count,
* start looking again from the end of the physical log. Note that
* callers can pass head == tail if the tail is not yet known.
*/
if (tail_blk >= head_blk && found != count) {
for (i = log->l_logBBsize - 1; i >= (int) tail_blk; i--) {
error = xlog_bread(log, i, 1, buffer, &offset);
if (error)
goto out_error;
if (*(__be32 *)offset ==
cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
*wrapped = true;
*rblk = i;
*rhead = (struct xlog_rec_header *) offset;
if (++found == count)
break;
}
}
}
return found;
out_error:
return error;
}
/*
* Seek forward in the log for log record headers.
*
* Given head and tail blocks, walk forward from the tail block until we find
* the provided number of records or hit the head block. The return value is the
* number of records encountered or a negative error code. The log block and
* buffer pointer of the last record seen are returned in rblk and rhead
* respectively.
*/
STATIC int
xlog_seek_logrec_hdr(
struct xlog *log,
xfs_daddr_t head_blk,
xfs_daddr_t tail_blk,
int count,
char *buffer,
xfs_daddr_t *rblk,
struct xlog_rec_header **rhead,
bool *wrapped)
{
int i;
int error;
int found = 0;
char *offset = NULL;
xfs_daddr_t end_blk;
*wrapped = false;
/*
* Walk forward from the tail block until we hit the head or the last
* block in the log.
*/
end_blk = head_blk > tail_blk ? head_blk : log->l_logBBsize - 1;
for (i = (int) tail_blk; i <= end_blk; i++) {
error = xlog_bread(log, i, 1, buffer, &offset);
if (error)
goto out_error;
if (*(__be32 *) offset == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
*rblk = i;
*rhead = (struct xlog_rec_header *) offset;
if (++found == count)
break;
}
}
/*
* If we haven't hit the head block or the log record header count,
* start looking again from the start of the physical log.
*/
if (tail_blk > head_blk && found != count) {
for (i = 0; i < (int) head_blk; i++) {
error = xlog_bread(log, i, 1, buffer, &offset);
if (error)
goto out_error;
if (*(__be32 *)offset ==
cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
*wrapped = true;
*rblk = i;
*rhead = (struct xlog_rec_header *) offset;
if (++found == count)
break;
}
}
}
return found;
out_error:
return error;
}
/*
xfs: fix log recovery corruption error due to tail overwrite If we consider the case where the tail (T) of the log is pinned long enough for the head (H) to push and block behind the tail, we can end up blocked in the following state without enough free space (f) in the log to satisfy a transaction reservation: 0 phys. log N [-------HffT---H'--T'---] The last good record in the log (before H) refers to T. The tail eventually pushes forward (T') leaving more free space in the log for writes to H. At this point, suppose space frees up in the log for the maximum of 8 in-core log buffers to start flushing out to the log. If this pushes the head from H to H', these next writes overwrite the previous tail T. This is safe because the items logged from T to T' have been written back and removed from the AIL. If the next log writes (H -> H') happen to fail and result in partial records in the log, the filesystem shuts down having overwritten T with invalid data. Log recovery correctly locates H on the subsequent mount, but H still refers to the now corrupted tail T. This results in log corruption errors and recovery failure. Since the tail overwrite results from otherwise correct runtime behavior, it is up to log recovery to try and deal with this situation. Update log recovery tail verification to run a CRC pass from the first record past the tail to the head. This facilitates error detection at T and moves the recovery tail to the first good record past H' (similar to truncating the head on torn write detection). If corruption is detected beyond the range possibly affected by the max number of iclogs, the log is legitimately corrupted and log recovery failure is expected. Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-08-09 09:21:52 +08:00
* Calculate distance from head to tail (i.e., unused space in the log).
*/
static inline int
xlog_tail_distance(
struct xlog *log,
xfs_daddr_t head_blk,
xfs_daddr_t tail_blk)
{
if (head_blk < tail_blk)
return tail_blk - head_blk;
return tail_blk + (log->l_logBBsize - head_blk);
}
/*
* Verify the log tail. This is particularly important when torn or incomplete
* writes have been detected near the front of the log and the head has been
* walked back accordingly.
*
* We also have to handle the case where the tail was pinned and the head
* blocked behind the tail right before a crash. If the tail had been pushed
* immediately prior to the crash and the subsequent checkpoint was only
* partially written, it's possible it overwrote the last referenced tail in the
* log with garbage. This is not a coherency problem because the tail must have
* been pushed before it can be overwritten, but appears as log corruption to
* recovery because we have no way to know the tail was updated if the
* subsequent checkpoint didn't write successfully.
*
xfs: fix log recovery corruption error due to tail overwrite If we consider the case where the tail (T) of the log is pinned long enough for the head (H) to push and block behind the tail, we can end up blocked in the following state without enough free space (f) in the log to satisfy a transaction reservation: 0 phys. log N [-------HffT---H'--T'---] The last good record in the log (before H) refers to T. The tail eventually pushes forward (T') leaving more free space in the log for writes to H. At this point, suppose space frees up in the log for the maximum of 8 in-core log buffers to start flushing out to the log. If this pushes the head from H to H', these next writes overwrite the previous tail T. This is safe because the items logged from T to T' have been written back and removed from the AIL. If the next log writes (H -> H') happen to fail and result in partial records in the log, the filesystem shuts down having overwritten T with invalid data. Log recovery correctly locates H on the subsequent mount, but H still refers to the now corrupted tail T. This results in log corruption errors and recovery failure. Since the tail overwrite results from otherwise correct runtime behavior, it is up to log recovery to try and deal with this situation. Update log recovery tail verification to run a CRC pass from the first record past the tail to the head. This facilitates error detection at T and moves the recovery tail to the first good record past H' (similar to truncating the head on torn write detection). If corruption is detected beyond the range possibly affected by the max number of iclogs, the log is legitimately corrupted and log recovery failure is expected. Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-08-09 09:21:52 +08:00
* Therefore, CRC check the log from tail to head. If a failure occurs and the
* offending record is within max iclog bufs from the head, walk the tail
* forward and retry until a valid tail is found or corruption is detected out
* of the range of a possible overwrite.
*/
STATIC int
xlog_verify_tail(
struct xlog *log,
xfs_daddr_t head_blk,
xfs: fix log recovery corruption error due to tail overwrite If we consider the case where the tail (T) of the log is pinned long enough for the head (H) to push and block behind the tail, we can end up blocked in the following state without enough free space (f) in the log to satisfy a transaction reservation: 0 phys. log N [-------HffT---H'--T'---] The last good record in the log (before H) refers to T. The tail eventually pushes forward (T') leaving more free space in the log for writes to H. At this point, suppose space frees up in the log for the maximum of 8 in-core log buffers to start flushing out to the log. If this pushes the head from H to H', these next writes overwrite the previous tail T. This is safe because the items logged from T to T' have been written back and removed from the AIL. If the next log writes (H -> H') happen to fail and result in partial records in the log, the filesystem shuts down having overwritten T with invalid data. Log recovery correctly locates H on the subsequent mount, but H still refers to the now corrupted tail T. This results in log corruption errors and recovery failure. Since the tail overwrite results from otherwise correct runtime behavior, it is up to log recovery to try and deal with this situation. Update log recovery tail verification to run a CRC pass from the first record past the tail to the head. This facilitates error detection at T and moves the recovery tail to the first good record past H' (similar to truncating the head on torn write detection). If corruption is detected beyond the range possibly affected by the max number of iclogs, the log is legitimately corrupted and log recovery failure is expected. Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-08-09 09:21:52 +08:00
xfs_daddr_t *tail_blk,
int hsize)
{
struct xlog_rec_header *thead;
char *buffer;
xfs_daddr_t first_bad;
int error = 0;
bool wrapped;
xfs: fix log recovery corruption error due to tail overwrite If we consider the case where the tail (T) of the log is pinned long enough for the head (H) to push and block behind the tail, we can end up blocked in the following state without enough free space (f) in the log to satisfy a transaction reservation: 0 phys. log N [-------HffT---H'--T'---] The last good record in the log (before H) refers to T. The tail eventually pushes forward (T') leaving more free space in the log for writes to H. At this point, suppose space frees up in the log for the maximum of 8 in-core log buffers to start flushing out to the log. If this pushes the head from H to H', these next writes overwrite the previous tail T. This is safe because the items logged from T to T' have been written back and removed from the AIL. If the next log writes (H -> H') happen to fail and result in partial records in the log, the filesystem shuts down having overwritten T with invalid data. Log recovery correctly locates H on the subsequent mount, but H still refers to the now corrupted tail T. This results in log corruption errors and recovery failure. Since the tail overwrite results from otherwise correct runtime behavior, it is up to log recovery to try and deal with this situation. Update log recovery tail verification to run a CRC pass from the first record past the tail to the head. This facilitates error detection at T and moves the recovery tail to the first good record past H' (similar to truncating the head on torn write detection). If corruption is detected beyond the range possibly affected by the max number of iclogs, the log is legitimately corrupted and log recovery failure is expected. Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-08-09 09:21:52 +08:00
xfs_daddr_t tmp_tail;
xfs_daddr_t orig_tail = *tail_blk;
buffer = xlog_alloc_buffer(log, 1);
if (!buffer)
return -ENOMEM;
/*
xfs: fix log recovery corruption error due to tail overwrite If we consider the case where the tail (T) of the log is pinned long enough for the head (H) to push and block behind the tail, we can end up blocked in the following state without enough free space (f) in the log to satisfy a transaction reservation: 0 phys. log N [-------HffT---H'--T'---] The last good record in the log (before H) refers to T. The tail eventually pushes forward (T') leaving more free space in the log for writes to H. At this point, suppose space frees up in the log for the maximum of 8 in-core log buffers to start flushing out to the log. If this pushes the head from H to H', these next writes overwrite the previous tail T. This is safe because the items logged from T to T' have been written back and removed from the AIL. If the next log writes (H -> H') happen to fail and result in partial records in the log, the filesystem shuts down having overwritten T with invalid data. Log recovery correctly locates H on the subsequent mount, but H still refers to the now corrupted tail T. This results in log corruption errors and recovery failure. Since the tail overwrite results from otherwise correct runtime behavior, it is up to log recovery to try and deal with this situation. Update log recovery tail verification to run a CRC pass from the first record past the tail to the head. This facilitates error detection at T and moves the recovery tail to the first good record past H' (similar to truncating the head on torn write detection). If corruption is detected beyond the range possibly affected by the max number of iclogs, the log is legitimately corrupted and log recovery failure is expected. Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-08-09 09:21:52 +08:00
* Make sure the tail points to a record (returns positive count on
* success).
*/
error = xlog_seek_logrec_hdr(log, head_blk, *tail_blk, 1, buffer,
xfs: fix log recovery corruption error due to tail overwrite If we consider the case where the tail (T) of the log is pinned long enough for the head (H) to push and block behind the tail, we can end up blocked in the following state without enough free space (f) in the log to satisfy a transaction reservation: 0 phys. log N [-------HffT---H'--T'---] The last good record in the log (before H) refers to T. The tail eventually pushes forward (T') leaving more free space in the log for writes to H. At this point, suppose space frees up in the log for the maximum of 8 in-core log buffers to start flushing out to the log. If this pushes the head from H to H', these next writes overwrite the previous tail T. This is safe because the items logged from T to T' have been written back and removed from the AIL. If the next log writes (H -> H') happen to fail and result in partial records in the log, the filesystem shuts down having overwritten T with invalid data. Log recovery correctly locates H on the subsequent mount, but H still refers to the now corrupted tail T. This results in log corruption errors and recovery failure. Since the tail overwrite results from otherwise correct runtime behavior, it is up to log recovery to try and deal with this situation. Update log recovery tail verification to run a CRC pass from the first record past the tail to the head. This facilitates error detection at T and moves the recovery tail to the first good record past H' (similar to truncating the head on torn write detection). If corruption is detected beyond the range possibly affected by the max number of iclogs, the log is legitimately corrupted and log recovery failure is expected. Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-08-09 09:21:52 +08:00
&tmp_tail, &thead, &wrapped);
if (error < 0)
goto out;
xfs: fix log recovery corruption error due to tail overwrite If we consider the case where the tail (T) of the log is pinned long enough for the head (H) to push and block behind the tail, we can end up blocked in the following state without enough free space (f) in the log to satisfy a transaction reservation: 0 phys. log N [-------HffT---H'--T'---] The last good record in the log (before H) refers to T. The tail eventually pushes forward (T') leaving more free space in the log for writes to H. At this point, suppose space frees up in the log for the maximum of 8 in-core log buffers to start flushing out to the log. If this pushes the head from H to H', these next writes overwrite the previous tail T. This is safe because the items logged from T to T' have been written back and removed from the AIL. If the next log writes (H -> H') happen to fail and result in partial records in the log, the filesystem shuts down having overwritten T with invalid data. Log recovery correctly locates H on the subsequent mount, but H still refers to the now corrupted tail T. This results in log corruption errors and recovery failure. Since the tail overwrite results from otherwise correct runtime behavior, it is up to log recovery to try and deal with this situation. Update log recovery tail verification to run a CRC pass from the first record past the tail to the head. This facilitates error detection at T and moves the recovery tail to the first good record past H' (similar to truncating the head on torn write detection). If corruption is detected beyond the range possibly affected by the max number of iclogs, the log is legitimately corrupted and log recovery failure is expected. Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-08-09 09:21:52 +08:00
if (*tail_blk != tmp_tail)
*tail_blk = tmp_tail;
/*
xfs: fix log recovery corruption error due to tail overwrite If we consider the case where the tail (T) of the log is pinned long enough for the head (H) to push and block behind the tail, we can end up blocked in the following state without enough free space (f) in the log to satisfy a transaction reservation: 0 phys. log N [-------HffT---H'--T'---] The last good record in the log (before H) refers to T. The tail eventually pushes forward (T') leaving more free space in the log for writes to H. At this point, suppose space frees up in the log for the maximum of 8 in-core log buffers to start flushing out to the log. If this pushes the head from H to H', these next writes overwrite the previous tail T. This is safe because the items logged from T to T' have been written back and removed from the AIL. If the next log writes (H -> H') happen to fail and result in partial records in the log, the filesystem shuts down having overwritten T with invalid data. Log recovery correctly locates H on the subsequent mount, but H still refers to the now corrupted tail T. This results in log corruption errors and recovery failure. Since the tail overwrite results from otherwise correct runtime behavior, it is up to log recovery to try and deal with this situation. Update log recovery tail verification to run a CRC pass from the first record past the tail to the head. This facilitates error detection at T and moves the recovery tail to the first good record past H' (similar to truncating the head on torn write detection). If corruption is detected beyond the range possibly affected by the max number of iclogs, the log is legitimately corrupted and log recovery failure is expected. Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-08-09 09:21:52 +08:00
* Run a CRC check from the tail to the head. We can't just check
* MAX_ICLOGS records past the tail because the tail may point to stale
* blocks cleared during the search for the head/tail. These blocks are
* overwritten with zero-length records and thus record count is not a
* reliable indicator of the iclog state before a crash.
*/
xfs: fix log recovery corruption error due to tail overwrite If we consider the case where the tail (T) of the log is pinned long enough for the head (H) to push and block behind the tail, we can end up blocked in the following state without enough free space (f) in the log to satisfy a transaction reservation: 0 phys. log N [-------HffT---H'--T'---] The last good record in the log (before H) refers to T. The tail eventually pushes forward (T') leaving more free space in the log for writes to H. At this point, suppose space frees up in the log for the maximum of 8 in-core log buffers to start flushing out to the log. If this pushes the head from H to H', these next writes overwrite the previous tail T. This is safe because the items logged from T to T' have been written back and removed from the AIL. If the next log writes (H -> H') happen to fail and result in partial records in the log, the filesystem shuts down having overwritten T with invalid data. Log recovery correctly locates H on the subsequent mount, but H still refers to the now corrupted tail T. This results in log corruption errors and recovery failure. Since the tail overwrite results from otherwise correct runtime behavior, it is up to log recovery to try and deal with this situation. Update log recovery tail verification to run a CRC pass from the first record past the tail to the head. This facilitates error detection at T and moves the recovery tail to the first good record past H' (similar to truncating the head on torn write detection). If corruption is detected beyond the range possibly affected by the max number of iclogs, the log is legitimately corrupted and log recovery failure is expected. Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-08-09 09:21:52 +08:00
first_bad = 0;
error = xlog_do_recovery_pass(log, head_blk, *tail_blk,
XLOG_RECOVER_CRCPASS, &first_bad);
while ((error == -EFSBADCRC || error == -EFSCORRUPTED) && first_bad) {
xfs: fix log recovery corruption error due to tail overwrite If we consider the case where the tail (T) of the log is pinned long enough for the head (H) to push and block behind the tail, we can end up blocked in the following state without enough free space (f) in the log to satisfy a transaction reservation: 0 phys. log N [-------HffT---H'--T'---] The last good record in the log (before H) refers to T. The tail eventually pushes forward (T') leaving more free space in the log for writes to H. At this point, suppose space frees up in the log for the maximum of 8 in-core log buffers to start flushing out to the log. If this pushes the head from H to H', these next writes overwrite the previous tail T. This is safe because the items logged from T to T' have been written back and removed from the AIL. If the next log writes (H -> H') happen to fail and result in partial records in the log, the filesystem shuts down having overwritten T with invalid data. Log recovery correctly locates H on the subsequent mount, but H still refers to the now corrupted tail T. This results in log corruption errors and recovery failure. Since the tail overwrite results from otherwise correct runtime behavior, it is up to log recovery to try and deal with this situation. Update log recovery tail verification to run a CRC pass from the first record past the tail to the head. This facilitates error detection at T and moves the recovery tail to the first good record past H' (similar to truncating the head on torn write detection). If corruption is detected beyond the range possibly affected by the max number of iclogs, the log is legitimately corrupted and log recovery failure is expected. Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-08-09 09:21:52 +08:00
int tail_distance;
/*
* Is corruption within range of the head? If so, retry from
* the next record. Otherwise return an error.
*/
tail_distance = xlog_tail_distance(log, head_blk, first_bad);
if (tail_distance > BTOBB(XLOG_MAX_ICLOGS * hsize))
break;
xfs: fix log recovery corruption error due to tail overwrite If we consider the case where the tail (T) of the log is pinned long enough for the head (H) to push and block behind the tail, we can end up blocked in the following state without enough free space (f) in the log to satisfy a transaction reservation: 0 phys. log N [-------HffT---H'--T'---] The last good record in the log (before H) refers to T. The tail eventually pushes forward (T') leaving more free space in the log for writes to H. At this point, suppose space frees up in the log for the maximum of 8 in-core log buffers to start flushing out to the log. If this pushes the head from H to H', these next writes overwrite the previous tail T. This is safe because the items logged from T to T' have been written back and removed from the AIL. If the next log writes (H -> H') happen to fail and result in partial records in the log, the filesystem shuts down having overwritten T with invalid data. Log recovery correctly locates H on the subsequent mount, but H still refers to the now corrupted tail T. This results in log corruption errors and recovery failure. Since the tail overwrite results from otherwise correct runtime behavior, it is up to log recovery to try and deal with this situation. Update log recovery tail verification to run a CRC pass from the first record past the tail to the head. This facilitates error detection at T and moves the recovery tail to the first good record past H' (similar to truncating the head on torn write detection). If corruption is detected beyond the range possibly affected by the max number of iclogs, the log is legitimately corrupted and log recovery failure is expected. Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-08-09 09:21:52 +08:00
/* skip to the next record; returns positive count on success */
error = xlog_seek_logrec_hdr(log, head_blk, first_bad, 2,
buffer, &tmp_tail, &thead, &wrapped);
xfs: fix log recovery corruption error due to tail overwrite If we consider the case where the tail (T) of the log is pinned long enough for the head (H) to push and block behind the tail, we can end up blocked in the following state without enough free space (f) in the log to satisfy a transaction reservation: 0 phys. log N [-------HffT---H'--T'---] The last good record in the log (before H) refers to T. The tail eventually pushes forward (T') leaving more free space in the log for writes to H. At this point, suppose space frees up in the log for the maximum of 8 in-core log buffers to start flushing out to the log. If this pushes the head from H to H', these next writes overwrite the previous tail T. This is safe because the items logged from T to T' have been written back and removed from the AIL. If the next log writes (H -> H') happen to fail and result in partial records in the log, the filesystem shuts down having overwritten T with invalid data. Log recovery correctly locates H on the subsequent mount, but H still refers to the now corrupted tail T. This results in log corruption errors and recovery failure. Since the tail overwrite results from otherwise correct runtime behavior, it is up to log recovery to try and deal with this situation. Update log recovery tail verification to run a CRC pass from the first record past the tail to the head. This facilitates error detection at T and moves the recovery tail to the first good record past H' (similar to truncating the head on torn write detection). If corruption is detected beyond the range possibly affected by the max number of iclogs, the log is legitimately corrupted and log recovery failure is expected. Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-08-09 09:21:52 +08:00
if (error < 0)
goto out;
*tail_blk = tmp_tail;
first_bad = 0;
error = xlog_do_recovery_pass(log, head_blk, *tail_blk,
XLOG_RECOVER_CRCPASS, &first_bad);
}
if (!error && *tail_blk != orig_tail)
xfs_warn(log->l_mp,
"Tail block (0x%llx) overwrite detected. Updated to 0x%llx",
orig_tail, *tail_blk);
out:
kmem_free(buffer);
return error;
}
/*
* Detect and trim torn writes from the head of the log.
*
* Storage without sector atomicity guarantees can result in torn writes in the
* log in the event of a crash. Our only means to detect this scenario is via
* CRC verification. While we can't always be certain that CRC verification
* failure is due to a torn write vs. an unrelated corruption, we do know that
* only a certain number (XLOG_MAX_ICLOGS) of log records can be written out at
* one time. Therefore, CRC verify up to XLOG_MAX_ICLOGS records at the head of
* the log and treat failures in this range as torn writes as a matter of
* policy. In the event of CRC failure, the head is walked back to the last good
* record in the log and the tail is updated from that record and verified.
*/
STATIC int
xlog_verify_head(
struct xlog *log,
xfs_daddr_t *head_blk, /* in/out: unverified head */
xfs_daddr_t *tail_blk, /* out: tail block */
char *buffer,
xfs_daddr_t *rhead_blk, /* start blk of last record */
struct xlog_rec_header **rhead, /* ptr to last record */
bool *wrapped) /* last rec. wraps phys. log */
{
struct xlog_rec_header *tmp_rhead;
char *tmp_buffer;
xfs_daddr_t first_bad;
xfs_daddr_t tmp_rhead_blk;
int found;
int error;
bool tmp_wrapped;
/*
* Check the head of the log for torn writes. Search backwards from the
* head until we hit the tail or the maximum number of log record I/Os
* that could have been in flight at one time. Use a temporary buffer so
* we don't trash the rhead/buffer pointers from the caller.
*/
tmp_buffer = xlog_alloc_buffer(log, 1);
if (!tmp_buffer)
return -ENOMEM;
error = xlog_rseek_logrec_hdr(log, *head_blk, *tail_blk,
XLOG_MAX_ICLOGS, tmp_buffer,
&tmp_rhead_blk, &tmp_rhead, &tmp_wrapped);
kmem_free(tmp_buffer);
if (error < 0)
return error;
/*
* Now run a CRC verification pass over the records starting at the
* block found above to the current head. If a CRC failure occurs, the
* log block of the first bad record is saved in first_bad.
*/
error = xlog_do_recovery_pass(log, *head_blk, tmp_rhead_blk,
XLOG_RECOVER_CRCPASS, &first_bad);
if ((error == -EFSBADCRC || error == -EFSCORRUPTED) && first_bad) {
/*
* We've hit a potential torn write. Reset the error and warn
* about it.
*/
error = 0;
xfs_warn(log->l_mp,
"Torn write (CRC failure) detected at log block 0x%llx. Truncating head block from 0x%llx.",
first_bad, *head_blk);
/*
* Get the header block and buffer pointer for the last good
* record before the bad record.
*
* Note that xlog_find_tail() clears the blocks at the new head
* (i.e., the records with invalid CRC) if the cycle number
* matches the current cycle.
*/
found = xlog_rseek_logrec_hdr(log, first_bad, *tail_blk, 1,
buffer, rhead_blk, rhead, wrapped);
if (found < 0)
return found;
if (found == 0) /* XXX: right thing to do here? */
return -EIO;
/*
* Reset the head block to the starting block of the first bad
* log record and set the tail block based on the last good
* record.
*
* Bail out if the updated head/tail match as this indicates
* possible corruption outside of the acceptable
* (XLOG_MAX_ICLOGS) range. This is a job for xfs_repair...
*/
*head_blk = first_bad;
*tail_blk = BLOCK_LSN(be64_to_cpu((*rhead)->h_tail_lsn));
if (*head_blk == *tail_blk) {
ASSERT(0);
return 0;
}
}
if (error)
return error;
xfs: fix log recovery corruption error due to tail overwrite If we consider the case where the tail (T) of the log is pinned long enough for the head (H) to push and block behind the tail, we can end up blocked in the following state without enough free space (f) in the log to satisfy a transaction reservation: 0 phys. log N [-------HffT---H'--T'---] The last good record in the log (before H) refers to T. The tail eventually pushes forward (T') leaving more free space in the log for writes to H. At this point, suppose space frees up in the log for the maximum of 8 in-core log buffers to start flushing out to the log. If this pushes the head from H to H', these next writes overwrite the previous tail T. This is safe because the items logged from T to T' have been written back and removed from the AIL. If the next log writes (H -> H') happen to fail and result in partial records in the log, the filesystem shuts down having overwritten T with invalid data. Log recovery correctly locates H on the subsequent mount, but H still refers to the now corrupted tail T. This results in log corruption errors and recovery failure. Since the tail overwrite results from otherwise correct runtime behavior, it is up to log recovery to try and deal with this situation. Update log recovery tail verification to run a CRC pass from the first record past the tail to the head. This facilitates error detection at T and moves the recovery tail to the first good record past H' (similar to truncating the head on torn write detection). If corruption is detected beyond the range possibly affected by the max number of iclogs, the log is legitimately corrupted and log recovery failure is expected. Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-08-09 09:21:52 +08:00
return xlog_verify_tail(log, *head_blk, tail_blk,
be32_to_cpu((*rhead)->h_size));
}
/*
* We need to make sure we handle log wrapping properly, so we can't use the
* calculated logbno directly. Make sure it wraps to the correct bno inside the
* log.
*
* The log is limited to 32 bit sizes, so we use the appropriate modulus
* operation here and cast it back to a 64 bit daddr on return.
*/
static inline xfs_daddr_t
xlog_wrap_logbno(
struct xlog *log,
xfs_daddr_t bno)
{
int mod;
div_s64_rem(bno, log->l_logBBsize, &mod);
return mod;
}
/*
* Check whether the head of the log points to an unmount record. In other
* words, determine whether the log is clean. If so, update the in-core state
* appropriately.
*/
static int
xlog_check_unmount_rec(
struct xlog *log,
xfs_daddr_t *head_blk,
xfs_daddr_t *tail_blk,
struct xlog_rec_header *rhead,
xfs_daddr_t rhead_blk,
char *buffer,
bool *clean)
{
struct xlog_op_header *op_head;
xfs_daddr_t umount_data_blk;
xfs_daddr_t after_umount_blk;
int hblks;
int error;
char *offset;
*clean = false;
/*
* Look for unmount record. If we find it, then we know there was a
* clean unmount. Since 'i' could be the last block in the physical
* log, we convert to a log block before comparing to the head_blk.
*
* Save the current tail lsn to use to pass to xlog_clear_stale_blocks()
* below. We won't want to clear the unmount record if there is one, so
* we pass the lsn of the unmount record rather than the block after it.
*/
hblks = xlog_logrec_hblks(log, rhead);
after_umount_blk = xlog_wrap_logbno(log,
rhead_blk + hblks + BTOBB(be32_to_cpu(rhead->h_len)));
if (*head_blk == after_umount_blk &&
be32_to_cpu(rhead->h_num_logops) == 1) {
umount_data_blk = xlog_wrap_logbno(log, rhead_blk + hblks);
error = xlog_bread(log, umount_data_blk, 1, buffer, &offset);
if (error)
return error;
op_head = (struct xlog_op_header *)offset;
if (op_head->oh_flags & XLOG_UNMOUNT_TRANS) {
/*
* Set tail and last sync so that newly written log
* records will point recovery to after the current
* unmount record.
*/
xlog_assign_atomic_lsn(&log->l_tail_lsn,
log->l_curr_cycle, after_umount_blk);
xlog_assign_atomic_lsn(&log->l_last_sync_lsn,
log->l_curr_cycle, after_umount_blk);
*tail_blk = after_umount_blk;
*clean = true;
}
}
return 0;
}
static void
xlog_set_state(
struct xlog *log,
xfs_daddr_t head_blk,
struct xlog_rec_header *rhead,
xfs_daddr_t rhead_blk,
bool bump_cycle)
{
/*
* Reset log values according to the state of the log when we
* crashed. In the case where head_blk == 0, we bump curr_cycle
* one because the next write starts a new cycle rather than
* continuing the cycle of the last good log record. At this
* point we have guaranteed that all partial log records have been
* accounted for. Therefore, we know that the last good log record
* written was complete and ended exactly on the end boundary
* of the physical log.
*/
log->l_prev_block = rhead_blk;
log->l_curr_block = (int)head_blk;
log->l_curr_cycle = be32_to_cpu(rhead->h_cycle);
if (bump_cycle)
log->l_curr_cycle++;
atomic64_set(&log->l_tail_lsn, be64_to_cpu(rhead->h_tail_lsn));
atomic64_set(&log->l_last_sync_lsn, be64_to_cpu(rhead->h_lsn));
xlog_assign_grant_head(&log->l_reserve_head.grant, log->l_curr_cycle,
BBTOB(log->l_curr_block));
xlog_assign_grant_head(&log->l_write_head.grant, log->l_curr_cycle,
BBTOB(log->l_curr_block));
}
/*
* Find the sync block number or the tail of the log.
*
* This will be the block number of the last record to have its
* associated buffers synced to disk. Every log record header has
* a sync lsn embedded in it. LSNs hold block numbers, so it is easy
* to get a sync block number. The only concern is to figure out which
* log record header to believe.
*
* The following algorithm uses the log record header with the largest
* lsn. The entire log record does not need to be valid. We only care
* that the header is valid.
*
* We could speed up search by using current head_blk buffer, but it is not
* available.
*/
STATIC int
xlog_find_tail(
struct xlog *log,
xfs_daddr_t *head_blk,
xfs_daddr_t *tail_blk)
{
xlog_rec_header_t *rhead;
char *offset = NULL;
char *buffer;
int error;
xfs_daddr_t rhead_blk;
xfs_lsn_t tail_lsn;
bool wrapped = false;
bool clean = false;
/*
* Find previous log record
*/
if ((error = xlog_find_head(log, head_blk)))
return error;
ASSERT(*head_blk < INT_MAX);
buffer = xlog_alloc_buffer(log, 1);
if (!buffer)
return -ENOMEM;
if (*head_blk == 0) { /* special case */
error = xlog_bread(log, 0, 1, buffer, &offset);
if (error)
goto done;
if (xlog_get_cycle(offset) == 0) {
*tail_blk = 0;
/* leave all other log inited values alone */
goto done;
}
}
/*
* Search backwards through the log looking for the log record header
* block. This wraps all the way back around to the head so something is
* seriously wrong if we can't find it.
*/
error = xlog_rseek_logrec_hdr(log, *head_blk, *head_blk, 1, buffer,
&rhead_blk, &rhead, &wrapped);
if (error < 0)
goto done;
if (!error) {
xfs_warn(log->l_mp, "%s: couldn't find sync record", __func__);
error = -EFSCORRUPTED;
goto done;
}
*tail_blk = BLOCK_LSN(be64_to_cpu(rhead->h_tail_lsn));
/*
* Set the log state based on the current head record.
*/
xlog_set_state(log, *head_blk, rhead, rhead_blk, wrapped);
tail_lsn = atomic64_read(&log->l_tail_lsn);
/*
* Look for an unmount record at the head of the log. This sets the log
* state to determine whether recovery is necessary.
*/
error = xlog_check_unmount_rec(log, head_blk, tail_blk, rhead,
rhead_blk, buffer, &clean);
if (error)
goto done;
/*
xfs: only run torn log write detection on dirty logs XFS uses CRC verification over a sub-range of the head of the log to detect and handle torn writes. This torn log write detection currently runs unconditionally at mount time, regardless of whether the log is dirty or clean. This is problematic in cases where a filesystem might end up being moved across different, incompatible (i.e., opposite byte-endianness) architectures. The problem lies in the fact that log data is not necessarily written in an architecture independent format. For example, certain bits of data are written in native endian format. Further, the size of certain log data structures differs (i.e., struct xlog_rec_header) depending on the word size of the cpu. This leads to false positive crc verification errors and ultimately failed mounts when a cleanly unmounted filesystem is mounted on a system with an incompatible architecture from data that was written near the head of the log. Update the log head/tail discovery code to run torn write detection only when the log is not clean. This means something other than an unmount record resides at the head of the log and log recovery is imminent. It is a requirement to run log recovery on the same type of host that had written the content of the dirty log and therefore CRC failures are legitimate corruptions in that scenario. Reported-by: Jan Beulich <JBeulich@suse.com> Tested-by: Jan Beulich <JBeulich@suse.com> Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-03-07 05:22:22 +08:00
* Verify the log head if the log is not clean (e.g., we have anything
* but an unmount record at the head). This uses CRC verification to
* detect and trim torn writes. If discovered, CRC failures are
* considered torn writes and the log head is trimmed accordingly.
*
xfs: only run torn log write detection on dirty logs XFS uses CRC verification over a sub-range of the head of the log to detect and handle torn writes. This torn log write detection currently runs unconditionally at mount time, regardless of whether the log is dirty or clean. This is problematic in cases where a filesystem might end up being moved across different, incompatible (i.e., opposite byte-endianness) architectures. The problem lies in the fact that log data is not necessarily written in an architecture independent format. For example, certain bits of data are written in native endian format. Further, the size of certain log data structures differs (i.e., struct xlog_rec_header) depending on the word size of the cpu. This leads to false positive crc verification errors and ultimately failed mounts when a cleanly unmounted filesystem is mounted on a system with an incompatible architecture from data that was written near the head of the log. Update the log head/tail discovery code to run torn write detection only when the log is not clean. This means something other than an unmount record resides at the head of the log and log recovery is imminent. It is a requirement to run log recovery on the same type of host that had written the content of the dirty log and therefore CRC failures are legitimate corruptions in that scenario. Reported-by: Jan Beulich <JBeulich@suse.com> Tested-by: Jan Beulich <JBeulich@suse.com> Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-03-07 05:22:22 +08:00
* Note that we can only run CRC verification when the log is dirty
* because there's no guarantee that the log data behind an unmount
* record is compatible with the current architecture.
*/
xfs: only run torn log write detection on dirty logs XFS uses CRC verification over a sub-range of the head of the log to detect and handle torn writes. This torn log write detection currently runs unconditionally at mount time, regardless of whether the log is dirty or clean. This is problematic in cases where a filesystem might end up being moved across different, incompatible (i.e., opposite byte-endianness) architectures. The problem lies in the fact that log data is not necessarily written in an architecture independent format. For example, certain bits of data are written in native endian format. Further, the size of certain log data structures differs (i.e., struct xlog_rec_header) depending on the word size of the cpu. This leads to false positive crc verification errors and ultimately failed mounts when a cleanly unmounted filesystem is mounted on a system with an incompatible architecture from data that was written near the head of the log. Update the log head/tail discovery code to run torn write detection only when the log is not clean. This means something other than an unmount record resides at the head of the log and log recovery is imminent. It is a requirement to run log recovery on the same type of host that had written the content of the dirty log and therefore CRC failures are legitimate corruptions in that scenario. Reported-by: Jan Beulich <JBeulich@suse.com> Tested-by: Jan Beulich <JBeulich@suse.com> Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-03-07 05:22:22 +08:00
if (!clean) {
xfs_daddr_t orig_head = *head_blk;
error = xlog_verify_head(log, head_blk, tail_blk, buffer,
xfs: only run torn log write detection on dirty logs XFS uses CRC verification over a sub-range of the head of the log to detect and handle torn writes. This torn log write detection currently runs unconditionally at mount time, regardless of whether the log is dirty or clean. This is problematic in cases where a filesystem might end up being moved across different, incompatible (i.e., opposite byte-endianness) architectures. The problem lies in the fact that log data is not necessarily written in an architecture independent format. For example, certain bits of data are written in native endian format. Further, the size of certain log data structures differs (i.e., struct xlog_rec_header) depending on the word size of the cpu. This leads to false positive crc verification errors and ultimately failed mounts when a cleanly unmounted filesystem is mounted on a system with an incompatible architecture from data that was written near the head of the log. Update the log head/tail discovery code to run torn write detection only when the log is not clean. This means something other than an unmount record resides at the head of the log and log recovery is imminent. It is a requirement to run log recovery on the same type of host that had written the content of the dirty log and therefore CRC failures are legitimate corruptions in that scenario. Reported-by: Jan Beulich <JBeulich@suse.com> Tested-by: Jan Beulich <JBeulich@suse.com> Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-03-07 05:22:22 +08:00
&rhead_blk, &rhead, &wrapped);
if (error)
goto done;
xfs: only run torn log write detection on dirty logs XFS uses CRC verification over a sub-range of the head of the log to detect and handle torn writes. This torn log write detection currently runs unconditionally at mount time, regardless of whether the log is dirty or clean. This is problematic in cases where a filesystem might end up being moved across different, incompatible (i.e., opposite byte-endianness) architectures. The problem lies in the fact that log data is not necessarily written in an architecture independent format. For example, certain bits of data are written in native endian format. Further, the size of certain log data structures differs (i.e., struct xlog_rec_header) depending on the word size of the cpu. This leads to false positive crc verification errors and ultimately failed mounts when a cleanly unmounted filesystem is mounted on a system with an incompatible architecture from data that was written near the head of the log. Update the log head/tail discovery code to run torn write detection only when the log is not clean. This means something other than an unmount record resides at the head of the log and log recovery is imminent. It is a requirement to run log recovery on the same type of host that had written the content of the dirty log and therefore CRC failures are legitimate corruptions in that scenario. Reported-by: Jan Beulich <JBeulich@suse.com> Tested-by: Jan Beulich <JBeulich@suse.com> Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-03-07 05:22:22 +08:00
/* update in-core state again if the head changed */
if (*head_blk != orig_head) {
xlog_set_state(log, *head_blk, rhead, rhead_blk,
wrapped);
tail_lsn = atomic64_read(&log->l_tail_lsn);
error = xlog_check_unmount_rec(log, head_blk, tail_blk,
rhead, rhead_blk, buffer,
xfs: only run torn log write detection on dirty logs XFS uses CRC verification over a sub-range of the head of the log to detect and handle torn writes. This torn log write detection currently runs unconditionally at mount time, regardless of whether the log is dirty or clean. This is problematic in cases where a filesystem might end up being moved across different, incompatible (i.e., opposite byte-endianness) architectures. The problem lies in the fact that log data is not necessarily written in an architecture independent format. For example, certain bits of data are written in native endian format. Further, the size of certain log data structures differs (i.e., struct xlog_rec_header) depending on the word size of the cpu. This leads to false positive crc verification errors and ultimately failed mounts when a cleanly unmounted filesystem is mounted on a system with an incompatible architecture from data that was written near the head of the log. Update the log head/tail discovery code to run torn write detection only when the log is not clean. This means something other than an unmount record resides at the head of the log and log recovery is imminent. It is a requirement to run log recovery on the same type of host that had written the content of the dirty log and therefore CRC failures are legitimate corruptions in that scenario. Reported-by: Jan Beulich <JBeulich@suse.com> Tested-by: Jan Beulich <JBeulich@suse.com> Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-03-07 05:22:22 +08:00
&clean);
if (error)
goto done;
}
}
/*
* Note that the unmount was clean. If the unmount was not clean, we
* need to know this to rebuild the superblock counters from the perag
* headers if we have a filesystem using non-persistent counters.
*/
if (clean)
set_bit(XFS_OPSTATE_CLEAN, &log->l_mp->m_opstate);
/*
* Make sure that there are no blocks in front of the head
* with the same cycle number as the head. This can happen
* because we allow multiple outstanding log writes concurrently,
* and the later writes might make it out before earlier ones.
*
* We use the lsn from before modifying it so that we'll never
* overwrite the unmount record after a clean unmount.
*
* Do this only if we are going to recover the filesystem
*
* NOTE: This used to say "if (!readonly)"
* However on Linux, we can & do recover a read-only filesystem.
* We only skip recovery if NORECOVERY is specified on mount,
* in which case we would not be here.
*
* But... if the -device- itself is readonly, just skip this.
* We can't recover this device anyway, so it won't matter.
*/
if (!xfs_readonly_buftarg(log->l_targ))
error = xlog_clear_stale_blocks(log, tail_lsn);
done:
kmem_free(buffer);
if (error)
xfs_warn(log->l_mp, "failed to locate log tail");
return error;
}
/*
* Is the log zeroed at all?
*
* The last binary search should be changed to perform an X block read
* once X becomes small enough. You can then search linearly through
* the X blocks. This will cut down on the number of reads we need to do.
*
* If the log is partially zeroed, this routine will pass back the blkno
* of the first block with cycle number 0. It won't have a complete LR
* preceding it.
*
* Return:
* 0 => the log is completely written to
* 1 => use *blk_no as the first block of the log
* <0 => error has occurred
*/
STATIC int
xlog_find_zeroed(
struct xlog *log,
xfs_daddr_t *blk_no)
{
char *buffer;
char *offset;
uint first_cycle, last_cycle;
xfs_daddr_t new_blk, last_blk, start_blk;
xfs_daddr_t num_scan_bblks;
int error, log_bbnum = log->l_logBBsize;
*blk_no = 0;
/* check totally zeroed log */
buffer = xlog_alloc_buffer(log, 1);
if (!buffer)
return -ENOMEM;
error = xlog_bread(log, 0, 1, buffer, &offset);
if (error)
goto out_free_buffer;
first_cycle = xlog_get_cycle(offset);
if (first_cycle == 0) { /* completely zeroed log */
*blk_no = 0;
kmem_free(buffer);
return 1;
}
/* check partially zeroed log */
error = xlog_bread(log, log_bbnum-1, 1, buffer, &offset);
if (error)
goto out_free_buffer;
last_cycle = xlog_get_cycle(offset);
if (last_cycle != 0) { /* log completely written to */
kmem_free(buffer);
return 0;
}
/* we have a partially zeroed log */
last_blk = log_bbnum-1;
error = xlog_find_cycle_start(log, buffer, 0, &last_blk, 0);
if (error)
goto out_free_buffer;
/*
* Validate the answer. Because there is no way to guarantee that
* the entire log is made up of log records which are the same size,
* we scan over the defined maximum blocks. At this point, the maximum
* is not chosen to mean anything special. XXXmiken
*/
num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
ASSERT(num_scan_bblks <= INT_MAX);
if (last_blk < num_scan_bblks)
num_scan_bblks = last_blk;
start_blk = last_blk - num_scan_bblks;
/*
* We search for any instances of cycle number 0 that occur before
* our current estimate of the head. What we're trying to detect is
* 1 ... | 0 | 1 | 0...
* ^ binary search ends here
*/
if ((error = xlog_find_verify_cycle(log, start_blk,
(int)num_scan_bblks, 0, &new_blk)))
goto out_free_buffer;
if (new_blk != -1)
last_blk = new_blk;
/*
* Potentially backup over partial log record write. We don't need
* to search the end of the log because we know it is zero.
*/
error = xlog_find_verify_log_record(log, start_blk, &last_blk, 0);
if (error == 1)
error = -EIO;
if (error)
goto out_free_buffer;
*blk_no = last_blk;
out_free_buffer:
kmem_free(buffer);
if (error)
return error;
return 1;
}
/*
* These are simple subroutines used by xlog_clear_stale_blocks() below
* to initialize a buffer full of empty log record headers and write
* them into the log.
*/
STATIC void
xlog_add_record(
struct xlog *log,
char *buf,
int cycle,
int block,
int tail_cycle,
int tail_block)
{
xlog_rec_header_t *recp = (xlog_rec_header_t *)buf;
memset(buf, 0, BBSIZE);
recp->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM);
recp->h_cycle = cpu_to_be32(cycle);
recp->h_version = cpu_to_be32(
xfs_has_logv2(log->l_mp) ? 2 : 1);
recp->h_lsn = cpu_to_be64(xlog_assign_lsn(cycle, block));
recp->h_tail_lsn = cpu_to_be64(xlog_assign_lsn(tail_cycle, tail_block));
recp->h_fmt = cpu_to_be32(XLOG_FMT);
memcpy(&recp->h_fs_uuid, &log->l_mp->m_sb.sb_uuid, sizeof(uuid_t));
}
STATIC int
xlog_write_log_records(
struct xlog *log,
int cycle,
int start_block,
int blocks,
int tail_cycle,
int tail_block)
{
char *offset;
char *buffer;
int balign, ealign;
int sectbb = log->l_sectBBsize;
int end_block = start_block + blocks;
int bufblks;
int error = 0;
int i, j = 0;
/*
* Greedily allocate a buffer big enough to handle the full
* range of basic blocks to be written. If that fails, try
* a smaller size. We need to be able to write at least a
* log sector, or we're out of luck.
*/
bufblks = roundup_pow_of_two(blocks);
while (bufblks > log->l_logBBsize)
bufblks >>= 1;
while (!(buffer = xlog_alloc_buffer(log, bufblks))) {
bufblks >>= 1;
if (bufblks < sectbb)
return -ENOMEM;
}
/* We may need to do a read at the start to fill in part of
* the buffer in the starting sector not covered by the first
* write below.
*/
balign = round_down(start_block, sectbb);
if (balign != start_block) {
error = xlog_bread_noalign(log, start_block, 1, buffer);
if (error)
goto out_free_buffer;
j = start_block - balign;
}
for (i = start_block; i < end_block; i += bufblks) {
int bcount, endcount;
bcount = min(bufblks, end_block - start_block);
endcount = bcount - j;
/* We may need to do a read at the end to fill in part of
* the buffer in the final sector not covered by the write.
* If this is the same sector as the above read, skip it.
*/
ealign = round_down(end_block, sectbb);
if (j == 0 && (start_block + endcount > ealign)) {
error = xlog_bread_noalign(log, ealign, sectbb,
buffer + BBTOB(ealign - start_block));
if (error)
break;
}
offset = buffer + xlog_align(log, start_block);
for (; j < endcount; j++) {
xlog_add_record(log, offset, cycle, i+j,
tail_cycle, tail_block);
offset += BBSIZE;
}
error = xlog_bwrite(log, start_block, endcount, buffer);
if (error)
break;
start_block += endcount;
j = 0;
}
out_free_buffer:
kmem_free(buffer);
return error;
}
/*
* This routine is called to blow away any incomplete log writes out
* in front of the log head. We do this so that we won't become confused
* if we come up, write only a little bit more, and then crash again.
* If we leave the partial log records out there, this situation could
* cause us to think those partial writes are valid blocks since they
* have the current cycle number. We get rid of them by overwriting them
* with empty log records with the old cycle number rather than the
* current one.
*
* The tail lsn is passed in rather than taken from
* the log so that we will not write over the unmount record after a
* clean unmount in a 512 block log. Doing so would leave the log without
* any valid log records in it until a new one was written. If we crashed
* during that time we would not be able to recover.
*/
STATIC int
xlog_clear_stale_blocks(
struct xlog *log,
xfs_lsn_t tail_lsn)
{
int tail_cycle, head_cycle;
int tail_block, head_block;
int tail_distance, max_distance;
int distance;
int error;
tail_cycle = CYCLE_LSN(tail_lsn);
tail_block = BLOCK_LSN(tail_lsn);
head_cycle = log->l_curr_cycle;
head_block = log->l_curr_block;
/*
* Figure out the distance between the new head of the log
* and the tail. We want to write over any blocks beyond the
* head that we may have written just before the crash, but
* we don't want to overwrite the tail of the log.
*/
if (head_cycle == tail_cycle) {
/*
* The tail is behind the head in the physical log,
* so the distance from the head to the tail is the
* distance from the head to the end of the log plus
* the distance from the beginning of the log to the
* tail.
*/
if (XFS_IS_CORRUPT(log->l_mp,
head_block < tail_block ||
head_block >= log->l_logBBsize))
return -EFSCORRUPTED;
tail_distance = tail_block + (log->l_logBBsize - head_block);
} else {
/*
* The head is behind the tail in the physical log,
* so the distance from the head to the tail is just
* the tail block minus the head block.
*/
if (XFS_IS_CORRUPT(log->l_mp,
head_block >= tail_block ||
head_cycle != tail_cycle + 1))
return -EFSCORRUPTED;
tail_distance = tail_block - head_block;
}
/*
* If the head is right up against the tail, we can't clear
* anything.
*/
if (tail_distance <= 0) {
ASSERT(tail_distance == 0);
return 0;
}
max_distance = XLOG_TOTAL_REC_SHIFT(log);
/*
* Take the smaller of the maximum amount of outstanding I/O
* we could have and the distance to the tail to clear out.
* We take the smaller so that we don't overwrite the tail and
* we don't waste all day writing from the head to the tail
* for no reason.
*/
max_distance = min(max_distance, tail_distance);
if ((head_block + max_distance) <= log->l_logBBsize) {
/*
* We can stomp all the blocks we need to without
* wrapping around the end of the log. Just do it
* in a single write. Use the cycle number of the
* current cycle minus one so that the log will look like:
* n ... | n - 1 ...
*/
error = xlog_write_log_records(log, (head_cycle - 1),
head_block, max_distance, tail_cycle,
tail_block);
if (error)
return error;
} else {
/*
* We need to wrap around the end of the physical log in
* order to clear all the blocks. Do it in two separate
* I/Os. The first write should be from the head to the
* end of the physical log, and it should use the current
* cycle number minus one just like above.
*/
distance = log->l_logBBsize - head_block;
error = xlog_write_log_records(log, (head_cycle - 1),
head_block, distance, tail_cycle,
tail_block);
if (error)
return error;
/*
* Now write the blocks at the start of the physical log.
* This writes the remainder of the blocks we want to clear.
* It uses the current cycle number since we're now on the
* same cycle as the head so that we get:
* n ... n ... | n - 1 ...
* ^^^^^ blocks we're writing
*/
distance = max_distance - (log->l_logBBsize - head_block);
error = xlog_write_log_records(log, head_cycle, 0, distance,
tail_cycle, tail_block);
if (error)
return error;
}
return 0;
}
/*
* Release the recovered intent item in the AIL that matches the given intent
* type and intent id.
*/
void
xlog_recover_release_intent(
struct xlog *log,
unsigned short intent_type,
uint64_t intent_id)
{
struct xfs_ail_cursor cur;
struct xfs_log_item *lip;
struct xfs_ail *ailp = log->l_ailp;
spin_lock(&ailp->ail_lock);
for (lip = xfs_trans_ail_cursor_first(ailp, &cur, 0); lip != NULL;
lip = xfs_trans_ail_cursor_next(ailp, &cur)) {
if (lip->li_type != intent_type)
continue;
if (!lip->li_ops->iop_match(lip, intent_id))
continue;
spin_unlock(&ailp->ail_lock);
lip->li_ops->iop_release(lip);
spin_lock(&ailp->ail_lock);
break;
}
xfs_trans_ail_cursor_done(&cur);
spin_unlock(&ailp->ail_lock);
}
int
xlog_recover_iget(
struct xfs_mount *mp,
xfs_ino_t ino,
struct xfs_inode **ipp)
{
int error;
error = xfs_iget(mp, NULL, ino, 0, 0, ipp);
if (error)
return error;
error = xfs_qm_dqattach(*ipp);
if (error) {
xfs_irele(*ipp);
return error;
}
if (VFS_I(*ipp)->i_nlink == 0)
xfs_iflags_set(*ipp, XFS_IRECOVERY);
return 0;
}
/******************************************************************************
*
* Log recover routines
*
******************************************************************************
*/
static const struct xlog_recover_item_ops *xlog_recover_item_ops[] = {
&xlog_buf_item_ops,
&xlog_inode_item_ops,
&xlog_dquot_item_ops,
&xlog_quotaoff_item_ops,
&xlog_icreate_item_ops,
&xlog_efi_item_ops,
&xlog_efd_item_ops,
&xlog_rui_item_ops,
&xlog_rud_item_ops,
&xlog_cui_item_ops,
&xlog_cud_item_ops,
&xlog_bui_item_ops,
&xlog_bud_item_ops,
xfs: Set up infrastructure for log attribute replay Currently attributes are modified directly across one or more transactions. But they are not logged or replayed in the event of an error. The goal of log attr replay is to enable logging and replaying of attribute operations using the existing delayed operations infrastructure. This will later enable the attributes to become part of larger multi part operations that also must first be recorded to the log. This is mostly of interest in the scheme of parent pointers which would need to maintain an attribute containing parent inode information any time an inode is moved, created, or removed. Parent pointers would then be of interest to any feature that would need to quickly derive an inode path from the mount point. Online scrub, nfs lookups and fs grow or shrink operations are all features that could take advantage of this. This patch adds two new log item types for setting or removing attributes as deferred operations. The xfs_attri_log_item will log an intent to set or remove an attribute. The corresponding xfs_attrd_log_item holds a reference to the xfs_attri_log_item and is freed once the transaction is done. Both log items use a generic xfs_attr_log_format structure that contains the attribute name, value, flags, inode, and an op_flag that indicates if the operations is a set or remove. [dchinner: added extra little bits needed for intent whiteouts] Signed-off-by: Allison Henderson <allison.henderson@oracle.com> Reviewed-by: Chandan Babu R <chandanrlinux@gmail.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Dave Chinner <david@fromorbit.com>
2022-05-04 10:41:02 +08:00
&xlog_attri_item_ops,
&xlog_attrd_item_ops,
};
static const struct xlog_recover_item_ops *
xlog_find_item_ops(
struct xlog_recover_item *item)
{
unsigned int i;
for (i = 0; i < ARRAY_SIZE(xlog_recover_item_ops); i++)
if (ITEM_TYPE(item) == xlog_recover_item_ops[i]->item_type)
return xlog_recover_item_ops[i];
return NULL;
}
/*
* Sort the log items in the transaction.
*
* The ordering constraints are defined by the inode allocation and unlink
* behaviour. The rules are:
*
* 1. Every item is only logged once in a given transaction. Hence it
* represents the last logged state of the item. Hence ordering is
* dependent on the order in which operations need to be performed so
* required initial conditions are always met.
*
* 2. Cancelled buffers are recorded in pass 1 in a separate table and
* there's nothing to replay from them so we can simply cull them
* from the transaction. However, we can't do that until after we've
* replayed all the other items because they may be dependent on the
* cancelled buffer and replaying the cancelled buffer can remove it
* form the cancelled buffer table. Hence they have tobe done last.
*
* 3. Inode allocation buffers must be replayed before inode items that
* read the buffer and replay changes into it. For filesystems using the
* ICREATE transactions, this means XFS_LI_ICREATE objects need to get
* treated the same as inode allocation buffers as they create and
* initialise the buffers directly.
*
* 4. Inode unlink buffers must be replayed after inode items are replayed.
* This ensures that inodes are completely flushed to the inode buffer
* in a "free" state before we remove the unlinked inode list pointer.
*
* Hence the ordering needs to be inode allocation buffers first, inode items
* second, inode unlink buffers third and cancelled buffers last.
*
* But there's a problem with that - we can't tell an inode allocation buffer
* apart from a regular buffer, so we can't separate them. We can, however,
* tell an inode unlink buffer from the others, and so we can separate them out
* from all the other buffers and move them to last.
*
* Hence, 4 lists, in order from head to tail:
* - buffer_list for all buffers except cancelled/inode unlink buffers
* - item_list for all non-buffer items
* - inode_buffer_list for inode unlink buffers
* - cancel_list for the cancelled buffers
*
* Note that we add objects to the tail of the lists so that first-to-last
* ordering is preserved within the lists. Adding objects to the head of the
* list means when we traverse from the head we walk them in last-to-first
* order. For cancelled buffers and inode unlink buffers this doesn't matter,
* but for all other items there may be specific ordering that we need to
* preserve.
*/
STATIC int
xlog_recover_reorder_trans(
struct xlog *log,
struct xlog_recover *trans,
int pass)
{
struct xlog_recover_item *item, *n;
int error = 0;
LIST_HEAD(sort_list);
LIST_HEAD(cancel_list);
LIST_HEAD(buffer_list);
LIST_HEAD(inode_buffer_list);
LIST_HEAD(item_list);
list_splice_init(&trans->r_itemq, &sort_list);
list_for_each_entry_safe(item, n, &sort_list, ri_list) {
enum xlog_recover_reorder fate = XLOG_REORDER_ITEM_LIST;
item->ri_ops = xlog_find_item_ops(item);
if (!item->ri_ops) {
xfs_warn(log->l_mp,
"%s: unrecognized type of log operation (%d)",
__func__, ITEM_TYPE(item));
ASSERT(0);
/*
* return the remaining items back to the transaction
* item list so they can be freed in caller.
*/
if (!list_empty(&sort_list))
list_splice_init(&sort_list, &trans->r_itemq);
error = -EFSCORRUPTED;
break;
}
if (item->ri_ops->reorder)
fate = item->ri_ops->reorder(item);
switch (fate) {
case XLOG_REORDER_BUFFER_LIST:
list_move_tail(&item->ri_list, &buffer_list);
break;
case XLOG_REORDER_CANCEL_LIST:
trace_xfs_log_recover_item_reorder_head(log,
trans, item, pass);
list_move(&item->ri_list, &cancel_list);
break;
case XLOG_REORDER_INODE_BUFFER_LIST:
list_move(&item->ri_list, &inode_buffer_list);
break;
case XLOG_REORDER_ITEM_LIST:
trace_xfs_log_recover_item_reorder_tail(log,
trans, item, pass);
list_move_tail(&item->ri_list, &item_list);
break;
}
}
ASSERT(list_empty(&sort_list));
if (!list_empty(&buffer_list))
list_splice(&buffer_list, &trans->r_itemq);
if (!list_empty(&item_list))
list_splice_tail(&item_list, &trans->r_itemq);
if (!list_empty(&inode_buffer_list))
list_splice_tail(&inode_buffer_list, &trans->r_itemq);
if (!list_empty(&cancel_list))
list_splice_tail(&cancel_list, &trans->r_itemq);
return error;
}
void
xlog_buf_readahead(
struct xlog *log,
xfs_daddr_t blkno,
uint len,
const struct xfs_buf_ops *ops)
{
if (!xlog_is_buffer_cancelled(log, blkno, len))
xfs_buf_readahead(log->l_mp->m_ddev_targp, blkno, len, ops);
}
STATIC int
xlog_recover_items_pass2(
struct xlog *log,
struct xlog_recover *trans,
struct list_head *buffer_list,
struct list_head *item_list)
{
struct xlog_recover_item *item;
int error = 0;
list_for_each_entry(item, item_list, ri_list) {
trace_xfs_log_recover_item_recover(log, trans, item,
XLOG_RECOVER_PASS2);
if (item->ri_ops->commit_pass2)
error = item->ri_ops->commit_pass2(log, buffer_list,
item, trans->r_lsn);
if (error)
return error;
}
return error;
}
/*
* Perform the transaction.
*
* If the transaction modifies a buffer or inode, do it now. Otherwise,
* EFIs and EFDs get queued up by adding entries into the AIL for them.
*/
STATIC int
xlog_recover_commit_trans(
struct xlog *log,
struct xlog_recover *trans,
int pass,
struct list_head *buffer_list)
{
int error = 0;
int items_queued = 0;
struct xlog_recover_item *item;
struct xlog_recover_item *next;
LIST_HEAD (ra_list);
LIST_HEAD (done_list);
#define XLOG_RECOVER_COMMIT_QUEUE_MAX 100
xfs: free uncommitted transactions during log recovery Log recovery allocates in-core transaction and member item data structures on-demand as it processes the on-disk log. Transactions are allocated on first encounter on-disk and stored in a hash table structure where they are easily accessible for subsequent lookups. Transaction items are also allocated on demand and are attached to the associated transactions. When a commit record is encountered in the log, the transaction is committed to the fs and the in-core structures are freed. If a filesystem crashes or shuts down before all in-core log buffers are flushed to the log, however, not all transactions may have commit records in the log. As expected, the modifications in such an incomplete transaction are not replayed to the fs. The in-core data structures for the partial transaction are never freed, however, resulting in a memory leak. Update xlog_do_recovery_pass() to first correctly initialize the hash table array so empty lists can be distinguished from populated lists on function exit. Update xlog_recover_free_trans() to always remove the transaction from the list prior to freeing the associated memory. Finally, walk the hash table of transaction lists as the last step before it goes out of scope and free any transactions that may remain on the lists. This prevents a memory leak of partial transactions in the log. Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-06-25 01:11:41 +08:00
hlist_del_init(&trans->r_list);
error = xlog_recover_reorder_trans(log, trans, pass);
if (error)
return error;
list_for_each_entry_safe(item, next, &trans->r_itemq, ri_list) {
trace_xfs_log_recover_item_recover(log, trans, item, pass);
xfs: on-stack delayed write buffer lists Queue delwri buffers on a local on-stack list instead of a per-buftarg one, and write back the buffers per-process instead of by waking up xfsbufd. This is now easily doable given that we have very few places left that write delwri buffers: - log recovery: Only done at mount time, and already forcing out the buffers synchronously using xfs_flush_buftarg - quotacheck: Same story. - dquot reclaim: Writes out dirty dquots on the LRU under memory pressure. We might want to look into doing more of this via xfsaild, but it's already more optimal than the synchronous inode reclaim that writes each buffer synchronously. - xfsaild: This is the main beneficiary of the change. By keeping a local list of buffers to write we reduce latency of writing out buffers, and more importably we can remove all the delwri list promotions which were hitting the buffer cache hard under sustained metadata loads. The implementation is very straight forward - xfs_buf_delwri_queue now gets a new list_head pointer that it adds the delwri buffers to, and all callers need to eventually submit the list using xfs_buf_delwi_submit or xfs_buf_delwi_submit_nowait. Buffers that already are on a delwri list are skipped in xfs_buf_delwri_queue, assuming they already are on another delwri list. The biggest change to pass down the buffer list was done to the AIL pushing. Now that we operate on buffers the trylock, push and pushbuf log item methods are merged into a single push routine, which tries to lock the item, and if possible add the buffer that needs writeback to the buffer list. This leads to much simpler code than the previous split but requires the individual IOP_PUSH instances to unlock and reacquire the AIL around calls to blocking routines. Given that xfsailds now also handle writing out buffers, the conditions for log forcing and the sleep times needed some small changes. The most important one is that we consider an AIL busy as long we still have buffers to push, and the other one is that we do increment the pushed LSN for buffers that are under flushing at this moment, but still count them towards the stuck items for restart purposes. Without this we could hammer on stuck items without ever forcing the log and not make progress under heavy random delete workloads on fast flash storage devices. [ Dave Chinner: - rebase on previous patches. - improved comments for XBF_DELWRI_Q handling - fix XBF_ASYNC handling in queue submission (test 106 failure) - rename delwri submit function buffer list parameters for clarity - xfs_efd_item_push() should return XFS_ITEM_PINNED ] Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Mark Tinguely <tinguely@sgi.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2012-04-23 13:58:39 +08:00
switch (pass) {
case XLOG_RECOVER_PASS1:
if (item->ri_ops->commit_pass1)
error = item->ri_ops->commit_pass1(log, item);
xfs: on-stack delayed write buffer lists Queue delwri buffers on a local on-stack list instead of a per-buftarg one, and write back the buffers per-process instead of by waking up xfsbufd. This is now easily doable given that we have very few places left that write delwri buffers: - log recovery: Only done at mount time, and already forcing out the buffers synchronously using xfs_flush_buftarg - quotacheck: Same story. - dquot reclaim: Writes out dirty dquots on the LRU under memory pressure. We might want to look into doing more of this via xfsaild, but it's already more optimal than the synchronous inode reclaim that writes each buffer synchronously. - xfsaild: This is the main beneficiary of the change. By keeping a local list of buffers to write we reduce latency of writing out buffers, and more importably we can remove all the delwri list promotions which were hitting the buffer cache hard under sustained metadata loads. The implementation is very straight forward - xfs_buf_delwri_queue now gets a new list_head pointer that it adds the delwri buffers to, and all callers need to eventually submit the list using xfs_buf_delwi_submit or xfs_buf_delwi_submit_nowait. Buffers that already are on a delwri list are skipped in xfs_buf_delwri_queue, assuming they already are on another delwri list. The biggest change to pass down the buffer list was done to the AIL pushing. Now that we operate on buffers the trylock, push and pushbuf log item methods are merged into a single push routine, which tries to lock the item, and if possible add the buffer that needs writeback to the buffer list. This leads to much simpler code than the previous split but requires the individual IOP_PUSH instances to unlock and reacquire the AIL around calls to blocking routines. Given that xfsailds now also handle writing out buffers, the conditions for log forcing and the sleep times needed some small changes. The most important one is that we consider an AIL busy as long we still have buffers to push, and the other one is that we do increment the pushed LSN for buffers that are under flushing at this moment, but still count them towards the stuck items for restart purposes. Without this we could hammer on stuck items without ever forcing the log and not make progress under heavy random delete workloads on fast flash storage devices. [ Dave Chinner: - rebase on previous patches. - improved comments for XBF_DELWRI_Q handling - fix XBF_ASYNC handling in queue submission (test 106 failure) - rename delwri submit function buffer list parameters for clarity - xfs_efd_item_push() should return XFS_ITEM_PINNED ] Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Mark Tinguely <tinguely@sgi.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2012-04-23 13:58:39 +08:00
break;
case XLOG_RECOVER_PASS2:
if (item->ri_ops->ra_pass2)
item->ri_ops->ra_pass2(log, item);
list_move_tail(&item->ri_list, &ra_list);
items_queued++;
if (items_queued >= XLOG_RECOVER_COMMIT_QUEUE_MAX) {
error = xlog_recover_items_pass2(log, trans,
buffer_list, &ra_list);
list_splice_tail_init(&ra_list, &done_list);
items_queued = 0;
}
xfs: on-stack delayed write buffer lists Queue delwri buffers on a local on-stack list instead of a per-buftarg one, and write back the buffers per-process instead of by waking up xfsbufd. This is now easily doable given that we have very few places left that write delwri buffers: - log recovery: Only done at mount time, and already forcing out the buffers synchronously using xfs_flush_buftarg - quotacheck: Same story. - dquot reclaim: Writes out dirty dquots on the LRU under memory pressure. We might want to look into doing more of this via xfsaild, but it's already more optimal than the synchronous inode reclaim that writes each buffer synchronously. - xfsaild: This is the main beneficiary of the change. By keeping a local list of buffers to write we reduce latency of writing out buffers, and more importably we can remove all the delwri list promotions which were hitting the buffer cache hard under sustained metadata loads. The implementation is very straight forward - xfs_buf_delwri_queue now gets a new list_head pointer that it adds the delwri buffers to, and all callers need to eventually submit the list using xfs_buf_delwi_submit or xfs_buf_delwi_submit_nowait. Buffers that already are on a delwri list are skipped in xfs_buf_delwri_queue, assuming they already are on another delwri list. The biggest change to pass down the buffer list was done to the AIL pushing. Now that we operate on buffers the trylock, push and pushbuf log item methods are merged into a single push routine, which tries to lock the item, and if possible add the buffer that needs writeback to the buffer list. This leads to much simpler code than the previous split but requires the individual IOP_PUSH instances to unlock and reacquire the AIL around calls to blocking routines. Given that xfsailds now also handle writing out buffers, the conditions for log forcing and the sleep times needed some small changes. The most important one is that we consider an AIL busy as long we still have buffers to push, and the other one is that we do increment the pushed LSN for buffers that are under flushing at this moment, but still count them towards the stuck items for restart purposes. Without this we could hammer on stuck items without ever forcing the log and not make progress under heavy random delete workloads on fast flash storage devices. [ Dave Chinner: - rebase on previous patches. - improved comments for XBF_DELWRI_Q handling - fix XBF_ASYNC handling in queue submission (test 106 failure) - rename delwri submit function buffer list parameters for clarity - xfs_efd_item_push() should return XFS_ITEM_PINNED ] Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Mark Tinguely <tinguely@sgi.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2012-04-23 13:58:39 +08:00
break;
default:
ASSERT(0);
}
if (error)
xfs: on-stack delayed write buffer lists Queue delwri buffers on a local on-stack list instead of a per-buftarg one, and write back the buffers per-process instead of by waking up xfsbufd. This is now easily doable given that we have very few places left that write delwri buffers: - log recovery: Only done at mount time, and already forcing out the buffers synchronously using xfs_flush_buftarg - quotacheck: Same story. - dquot reclaim: Writes out dirty dquots on the LRU under memory pressure. We might want to look into doing more of this via xfsaild, but it's already more optimal than the synchronous inode reclaim that writes each buffer synchronously. - xfsaild: This is the main beneficiary of the change. By keeping a local list of buffers to write we reduce latency of writing out buffers, and more importably we can remove all the delwri list promotions which were hitting the buffer cache hard under sustained metadata loads. The implementation is very straight forward - xfs_buf_delwri_queue now gets a new list_head pointer that it adds the delwri buffers to, and all callers need to eventually submit the list using xfs_buf_delwi_submit or xfs_buf_delwi_submit_nowait. Buffers that already are on a delwri list are skipped in xfs_buf_delwri_queue, assuming they already are on another delwri list. The biggest change to pass down the buffer list was done to the AIL pushing. Now that we operate on buffers the trylock, push and pushbuf log item methods are merged into a single push routine, which tries to lock the item, and if possible add the buffer that needs writeback to the buffer list. This leads to much simpler code than the previous split but requires the individual IOP_PUSH instances to unlock and reacquire the AIL around calls to blocking routines. Given that xfsailds now also handle writing out buffers, the conditions for log forcing and the sleep times needed some small changes. The most important one is that we consider an AIL busy as long we still have buffers to push, and the other one is that we do increment the pushed LSN for buffers that are under flushing at this moment, but still count them towards the stuck items for restart purposes. Without this we could hammer on stuck items without ever forcing the log and not make progress under heavy random delete workloads on fast flash storage devices. [ Dave Chinner: - rebase on previous patches. - improved comments for XBF_DELWRI_Q handling - fix XBF_ASYNC handling in queue submission (test 106 failure) - rename delwri submit function buffer list parameters for clarity - xfs_efd_item_push() should return XFS_ITEM_PINNED ] Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Mark Tinguely <tinguely@sgi.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2012-04-23 13:58:39 +08:00
goto out;
}
out:
if (!list_empty(&ra_list)) {
if (!error)
error = xlog_recover_items_pass2(log, trans,
buffer_list, &ra_list);
list_splice_tail_init(&ra_list, &done_list);
}
if (!list_empty(&done_list))
list_splice_init(&done_list, &trans->r_itemq);
return error;
}
STATIC void
xlog_recover_add_item(
struct list_head *head)
{
struct xlog_recover_item *item;
item = kmem_zalloc(sizeof(struct xlog_recover_item), 0);
INIT_LIST_HEAD(&item->ri_list);
list_add_tail(&item->ri_list, head);
}
STATIC int
xlog_recover_add_to_cont_trans(
struct xlog *log,
struct xlog_recover *trans,
char *dp,
int len)
{
struct xlog_recover_item *item;
char *ptr, *old_ptr;
int old_len;
/*
* If the transaction is empty, the header was split across this and the
* previous record. Copy the rest of the header.
*/
if (list_empty(&trans->r_itemq)) {
ASSERT(len <= sizeof(struct xfs_trans_header));
if (len > sizeof(struct xfs_trans_header)) {
xfs_warn(log->l_mp, "%s: bad header length", __func__);
return -EFSCORRUPTED;
}
xlog_recover_add_item(&trans->r_itemq);
ptr = (char *)&trans->r_theader +
sizeof(struct xfs_trans_header) - len;
memcpy(ptr, dp, len);
return 0;
}
/* take the tail entry */
item = list_entry(trans->r_itemq.prev, struct xlog_recover_item,
ri_list);
old_ptr = item->ri_buf[item->ri_cnt-1].i_addr;
old_len = item->ri_buf[item->ri_cnt-1].i_len;
mm: Add kvrealloc() During log recovery of an XFS filesystem with 64kB directory buffers, rebuilding a buffer split across two log records results in a memory allocation warning from krealloc like this: xfs filesystem being mounted at /mnt/scratch supports timestamps until 2038 (0x7fffffff) XFS (dm-0): Unmounting Filesystem XFS (dm-0): Mounting V5 Filesystem XFS (dm-0): Starting recovery (logdev: internal) ------------[ cut here ]------------ WARNING: CPU: 5 PID: 3435170 at mm/page_alloc.c:3539 get_page_from_freelist+0xdee/0xe40 ..... RIP: 0010:get_page_from_freelist+0xdee/0xe40 Call Trace: ? complete+0x3f/0x50 __alloc_pages+0x16f/0x300 alloc_pages+0x87/0x110 kmalloc_order+0x2c/0x90 kmalloc_order_trace+0x1d/0x90 __kmalloc_track_caller+0x215/0x270 ? xlog_recover_add_to_cont_trans+0x63/0x1f0 krealloc+0x54/0xb0 xlog_recover_add_to_cont_trans+0x63/0x1f0 xlog_recovery_process_trans+0xc1/0xd0 xlog_recover_process_ophdr+0x86/0x130 xlog_recover_process_data+0x9f/0x160 xlog_recover_process+0xa2/0x120 xlog_do_recovery_pass+0x40b/0x7d0 ? __irq_work_queue_local+0x4f/0x60 ? irq_work_queue+0x3a/0x50 xlog_do_log_recovery+0x70/0x150 xlog_do_recover+0x38/0x1d0 xlog_recover+0xd8/0x170 xfs_log_mount+0x181/0x300 xfs_mountfs+0x4a1/0x9b0 xfs_fs_fill_super+0x3c0/0x7b0 get_tree_bdev+0x171/0x270 ? suffix_kstrtoint.constprop.0+0xf0/0xf0 xfs_fs_get_tree+0x15/0x20 vfs_get_tree+0x24/0xc0 path_mount+0x2f5/0xaf0 __x64_sys_mount+0x108/0x140 do_syscall_64+0x3a/0x70 entry_SYSCALL_64_after_hwframe+0x44/0xae Essentially, we are taking a multi-order allocation from kmem_alloc() (which has an open coded no fail, no warn loop) and then reallocating it out to 64kB using krealloc(__GFP_NOFAIL) and that is then triggering the above warning. This is a regression caused by converting this code from an open coded no fail/no warn reallocation loop to using __GFP_NOFAIL. What we actually need here is kvrealloc(), so that if contiguous page allocation fails we fall back to vmalloc() and we don't get nasty warnings happening in XFS. Fixes: 771915c4f688 ("xfs: remove kmem_realloc()") Signed-off-by: Dave Chinner <dchinner@redhat.com> Acked-by: Mel Gorman <mgorman@techsingularity.net> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-08-10 01:10:00 +08:00
ptr = kvrealloc(old_ptr, old_len, len + old_len, GFP_KERNEL);
if (!ptr)
return -ENOMEM;
memcpy(&ptr[old_len], dp, len);
item->ri_buf[item->ri_cnt-1].i_len += len;
item->ri_buf[item->ri_cnt-1].i_addr = ptr;
trace_xfs_log_recover_item_add_cont(log, trans, item, 0);
return 0;
}
/*
* The next region to add is the start of a new region. It could be
* a whole region or it could be the first part of a new region. Because
* of this, the assumption here is that the type and size fields of all
* format structures fit into the first 32 bits of the structure.
*
* This works because all regions must be 32 bit aligned. Therefore, we
* either have both fields or we have neither field. In the case we have
* neither field, the data part of the region is zero length. We only have
* a log_op_header and can throw away the header since a new one will appear
* later. If we have at least 4 bytes, then we can determine how many regions
* will appear in the current log item.
*/
STATIC int
xlog_recover_add_to_trans(
struct xlog *log,
struct xlog_recover *trans,
char *dp,
int len)
{
struct xfs_inode_log_format *in_f; /* any will do */
struct xlog_recover_item *item;
char *ptr;
if (!len)
return 0;
if (list_empty(&trans->r_itemq)) {
/* we need to catch log corruptions here */
if (*(uint *)dp != XFS_TRANS_HEADER_MAGIC) {
xfs_warn(log->l_mp, "%s: bad header magic number",
__func__);
ASSERT(0);
return -EFSCORRUPTED;
}
if (len > sizeof(struct xfs_trans_header)) {
xfs_warn(log->l_mp, "%s: bad header length", __func__);
ASSERT(0);
return -EFSCORRUPTED;
}
/*
* The transaction header can be arbitrarily split across op
* records. If we don't have the whole thing here, copy what we
* do have and handle the rest in the next record.
*/
if (len == sizeof(struct xfs_trans_header))
xlog_recover_add_item(&trans->r_itemq);
memcpy(&trans->r_theader, dp, len);
return 0;
}
ptr = kmem_alloc(len, 0);
memcpy(ptr, dp, len);
in_f = (struct xfs_inode_log_format *)ptr;
/* take the tail entry */
item = list_entry(trans->r_itemq.prev, struct xlog_recover_item,
ri_list);
if (item->ri_total != 0 &&
item->ri_total == item->ri_cnt) {
/* tail item is in use, get a new one */
xlog_recover_add_item(&trans->r_itemq);
item = list_entry(trans->r_itemq.prev,
struct xlog_recover_item, ri_list);
}
if (item->ri_total == 0) { /* first region to be added */
if (in_f->ilf_size == 0 ||
in_f->ilf_size > XLOG_MAX_REGIONS_IN_ITEM) {
xfs_warn(log->l_mp,
"bad number of regions (%d) in inode log format",
in_f->ilf_size);
ASSERT(0);
kmem_free(ptr);
return -EFSCORRUPTED;
}
item->ri_total = in_f->ilf_size;
item->ri_buf =
kmem_zalloc(item->ri_total * sizeof(xfs_log_iovec_t),
0);
}
if (item->ri_total <= item->ri_cnt) {
xfs_warn(log->l_mp,
"log item region count (%d) overflowed size (%d)",
item->ri_cnt, item->ri_total);
ASSERT(0);
kmem_free(ptr);
return -EFSCORRUPTED;
}
/* Description region is ri_buf[0] */
item->ri_buf[item->ri_cnt].i_addr = ptr;
item->ri_buf[item->ri_cnt].i_len = len;
item->ri_cnt++;
trace_xfs_log_recover_item_add(log, trans, item, 0);
return 0;
}
/*
* Free up any resources allocated by the transaction
*
* Remember that EFIs, EFDs, and IUNLINKs are handled later.
*/
STATIC void
xlog_recover_free_trans(
struct xlog_recover *trans)
{
struct xlog_recover_item *item, *n;
int i;
xfs: free uncommitted transactions during log recovery Log recovery allocates in-core transaction and member item data structures on-demand as it processes the on-disk log. Transactions are allocated on first encounter on-disk and stored in a hash table structure where they are easily accessible for subsequent lookups. Transaction items are also allocated on demand and are attached to the associated transactions. When a commit record is encountered in the log, the transaction is committed to the fs and the in-core structures are freed. If a filesystem crashes or shuts down before all in-core log buffers are flushed to the log, however, not all transactions may have commit records in the log. As expected, the modifications in such an incomplete transaction are not replayed to the fs. The in-core data structures for the partial transaction are never freed, however, resulting in a memory leak. Update xlog_do_recovery_pass() to first correctly initialize the hash table array so empty lists can be distinguished from populated lists on function exit. Update xlog_recover_free_trans() to always remove the transaction from the list prior to freeing the associated memory. Finally, walk the hash table of transaction lists as the last step before it goes out of scope and free any transactions that may remain on the lists. This prevents a memory leak of partial transactions in the log. Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-06-25 01:11:41 +08:00
hlist_del_init(&trans->r_list);
list_for_each_entry_safe(item, n, &trans->r_itemq, ri_list) {
/* Free the regions in the item. */
list_del(&item->ri_list);
for (i = 0; i < item->ri_cnt; i++)
kmem_free(item->ri_buf[i].i_addr);
/* Free the item itself */
kmem_free(item->ri_buf);
kmem_free(item);
}
/* Free the transaction recover structure */
kmem_free(trans);
}
/*
* On error or completion, trans is freed.
*/
STATIC int
xlog_recovery_process_trans(
struct xlog *log,
struct xlog_recover *trans,
char *dp,
unsigned int len,
unsigned int flags,
int pass,
struct list_head *buffer_list)
{
int error = 0;
bool freeit = false;
/* mask off ophdr transaction container flags */
flags &= ~XLOG_END_TRANS;
if (flags & XLOG_WAS_CONT_TRANS)
flags &= ~XLOG_CONTINUE_TRANS;
/*
* Callees must not free the trans structure. We'll decide if we need to
* free it or not based on the operation being done and it's result.
*/
switch (flags) {
/* expected flag values */
case 0:
case XLOG_CONTINUE_TRANS:
error = xlog_recover_add_to_trans(log, trans, dp, len);
break;
case XLOG_WAS_CONT_TRANS:
error = xlog_recover_add_to_cont_trans(log, trans, dp, len);
break;
case XLOG_COMMIT_TRANS:
error = xlog_recover_commit_trans(log, trans, pass,
buffer_list);
/* success or fail, we are now done with this transaction. */
freeit = true;
break;
/* unexpected flag values */
case XLOG_UNMOUNT_TRANS:
/* just skip trans */
xfs_warn(log->l_mp, "%s: Unmount LR", __func__);
freeit = true;
break;
case XLOG_START_TRANS:
default:
xfs_warn(log->l_mp, "%s: bad flag 0x%x", __func__, flags);
ASSERT(0);
error = -EFSCORRUPTED;
break;
}
if (error || freeit)
xlog_recover_free_trans(trans);
return error;
}
/*
* Lookup the transaction recovery structure associated with the ID in the
* current ophdr. If the transaction doesn't exist and the start flag is set in
* the ophdr, then allocate a new transaction for future ID matches to find.
* Either way, return what we found during the lookup - an existing transaction
* or nothing.
*/
STATIC struct xlog_recover *
xlog_recover_ophdr_to_trans(
struct hlist_head rhash[],
struct xlog_rec_header *rhead,
struct xlog_op_header *ohead)
{
struct xlog_recover *trans;
xlog_tid_t tid;
struct hlist_head *rhp;
tid = be32_to_cpu(ohead->oh_tid);
rhp = &rhash[XLOG_RHASH(tid)];
hlist_for_each_entry(trans, rhp, r_list) {
if (trans->r_log_tid == tid)
return trans;
}
/*
* skip over non-start transaction headers - we could be
* processing slack space before the next transaction starts
*/
if (!(ohead->oh_flags & XLOG_START_TRANS))
return NULL;
ASSERT(be32_to_cpu(ohead->oh_len) == 0);
/*
* This is a new transaction so allocate a new recovery container to
* hold the recovery ops that will follow.
*/
trans = kmem_zalloc(sizeof(struct xlog_recover), 0);
trans->r_log_tid = tid;
trans->r_lsn = be64_to_cpu(rhead->h_lsn);
INIT_LIST_HEAD(&trans->r_itemq);
INIT_HLIST_NODE(&trans->r_list);
hlist_add_head(&trans->r_list, rhp);
/*
* Nothing more to do for this ophdr. Items to be added to this new
* transaction will be in subsequent ophdr containers.
*/
return NULL;
}
STATIC int
xlog_recover_process_ophdr(
struct xlog *log,
struct hlist_head rhash[],
struct xlog_rec_header *rhead,
struct xlog_op_header *ohead,
char *dp,
char *end,
int pass,
struct list_head *buffer_list)
{
struct xlog_recover *trans;
unsigned int len;
int error;
/* Do we understand who wrote this op? */
if (ohead->oh_clientid != XFS_TRANSACTION &&
ohead->oh_clientid != XFS_LOG) {
xfs_warn(log->l_mp, "%s: bad clientid 0x%x",
__func__, ohead->oh_clientid);
ASSERT(0);
return -EFSCORRUPTED;
}
/*
* Check the ophdr contains all the data it is supposed to contain.
*/
len = be32_to_cpu(ohead->oh_len);
if (dp + len > end) {
xfs_warn(log->l_mp, "%s: bad length 0x%x", __func__, len);
WARN_ON(1);
return -EFSCORRUPTED;
}
trans = xlog_recover_ophdr_to_trans(rhash, rhead, ohead);
if (!trans) {
/* nothing to do, so skip over this ophdr */
return 0;
}
/*
* The recovered buffer queue is drained only once we know that all
* recovery items for the current LSN have been processed. This is
* required because:
*
* - Buffer write submission updates the metadata LSN of the buffer.
* - Log recovery skips items with a metadata LSN >= the current LSN of
* the recovery item.
* - Separate recovery items against the same metadata buffer can share
* a current LSN. I.e., consider that the LSN of a recovery item is
* defined as the starting LSN of the first record in which its
* transaction appears, that a record can hold multiple transactions,
* and/or that a transaction can span multiple records.
*
* In other words, we are allowed to submit a buffer from log recovery
* once per current LSN. Otherwise, we may incorrectly skip recovery
* items and cause corruption.
*
* We don't know up front whether buffers are updated multiple times per
* LSN. Therefore, track the current LSN of each commit log record as it
* is processed and drain the queue when it changes. Use commit records
* because they are ordered correctly by the logging code.
*/
if (log->l_recovery_lsn != trans->r_lsn &&
ohead->oh_flags & XLOG_COMMIT_TRANS) {
error = xfs_buf_delwri_submit(buffer_list);
if (error)
return error;
log->l_recovery_lsn = trans->r_lsn;
}
return xlog_recovery_process_trans(log, trans, dp, len,
ohead->oh_flags, pass, buffer_list);
}
/*
* There are two valid states of the r_state field. 0 indicates that the
* transaction structure is in a normal state. We have either seen the
* start of the transaction or the last operation we added was not a partial
* operation. If the last operation we added to the transaction was a
* partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS.
*
* NOTE: skip LRs with 0 data length.
*/
STATIC int
xlog_recover_process_data(
struct xlog *log,
struct hlist_head rhash[],
struct xlog_rec_header *rhead,
char *dp,
int pass,
struct list_head *buffer_list)
{
struct xlog_op_header *ohead;
char *end;
int num_logops;
int error;
end = dp + be32_to_cpu(rhead->h_len);
num_logops = be32_to_cpu(rhead->h_num_logops);
/* check the log format matches our own - else we can't recover */
if (xlog_header_check_recover(log->l_mp, rhead))
return -EIO;
trace_xfs_log_recover_record(log, rhead, pass);
while ((dp < end) && num_logops) {
ohead = (struct xlog_op_header *)dp;
dp += sizeof(*ohead);
ASSERT(dp <= end);
/* errors will abort recovery */
error = xlog_recover_process_ophdr(log, rhash, rhead, ohead,
dp, end, pass, buffer_list);
if (error)
return error;
dp += be32_to_cpu(ohead->oh_len);
num_logops--;
}
return 0;
}
xfs: log recovery should replay deferred ops in order As part of testing log recovery with dm_log_writes, Amir Goldstein discovered an error in the deferred ops recovery that lead to corruption of the filesystem metadata if a reflink+rmap filesystem happened to shut down midway through a CoW remap: "This is what happens [after failed log recovery]: "Phase 1 - find and verify superblock... "Phase 2 - using internal log " - zero log... " - scan filesystem freespace and inode maps... " - found root inode chunk "Phase 3 - for each AG... " - scan (but don't clear) agi unlinked lists... " - process known inodes and perform inode discovery... " - agno = 0 "data fork in regular inode 134 claims CoW block 376 "correcting nextents for inode 134 "bad data fork in inode 134 "would have cleared inode 134" Hou Tao dissected the log contents of exactly such a crash: "According to the implementation of xfs_defer_finish(), these ops should be completed in the following sequence: "Have been done: "(1) CUI: Oper (160) "(2) BUI: Oper (161) "(3) CUD: Oper (194), for CUI Oper (160) "(4) RUI A: Oper (197), free rmap [0x155, 2, -9] "Should be done: "(5) BUD: for BUI Oper (161) "(6) RUI B: add rmap [0x155, 2, 137] "(7) RUD: for RUI A "(8) RUD: for RUI B "Actually be done by xlog_recover_process_intents() "(5) BUD: for BUI Oper (161) "(6) RUI B: add rmap [0x155, 2, 137] "(7) RUD: for RUI B "(8) RUD: for RUI A "So the rmap entry [0x155, 2, -9] for COW should be freed firstly, then a new rmap entry [0x155, 2, 137] will be added. However, as we can see from the log record in post_mount.log (generated after umount) and the trace print, the new rmap entry [0x155, 2, 137] are added firstly, then the rmap entry [0x155, 2, -9] are freed." When reconstructing the internal log state from the log items found on disk, it's required that deferred ops replay in exactly the same order that they would have had the filesystem not gone down. However, replaying unfinished deferred ops can create /more/ deferred ops. These new deferred ops are finished in the wrong order. This causes fs corruption and replay crashes, so let's create a single defer_ops to handle the subsequent ops created during replay, then use one single transaction at the end of log recovery to ensure that everything is replayed in the same order as they're supposed to be. Reported-by: Amir Goldstein <amir73il@gmail.com> Analyzed-by: Hou Tao <houtao1@huawei.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Tested-by: Amir Goldstein <amir73il@gmail.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-11-22 12:53:02 +08:00
/* Take all the collected deferred ops and finish them in order. */
static int
xlog_finish_defer_ops(
xfs: proper replay of deferred ops queued during log recovery When we replay unfinished intent items that have been recovered from the log, it's possible that the replay will cause the creation of more deferred work items. As outlined in commit 509955823cc9c ("xfs: log recovery should replay deferred ops in order"), later work items have an implicit ordering dependency on earlier work items. Therefore, recovery must replay the items (both recovered and created) in the same order that they would have been during normal operation. For log recovery, we enforce this ordering by using an empty transaction to collect deferred ops that get created in the process of recovering a log intent item to prevent them from being committed before the rest of the recovered intent items. After we finish committing all the recovered log items, we allocate a transaction with an enormous block reservation, splice our huge list of created deferred ops into that transaction, and commit it, thereby finishing all those ops. This is /really/ hokey -- it's the one place in XFS where we allow nested transactions; the splicing of the defer ops list is is inelegant and has to be done twice per recovery function; and the broken way we handle inode pointers and block reservations cause subtle use-after-free and allocator problems that will be fixed by this patch and the two patches after it. Therefore, replace the hokey empty transaction with a structure designed to capture each chain of deferred ops that are created as part of recovering a single unfinished log intent. Finally, refactor the loop that replays those chains to do so using one transaction per chain. Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de>
2020-09-26 08:39:37 +08:00
struct xfs_mount *mp,
struct list_head *capture_list)
xfs: log recovery should replay deferred ops in order As part of testing log recovery with dm_log_writes, Amir Goldstein discovered an error in the deferred ops recovery that lead to corruption of the filesystem metadata if a reflink+rmap filesystem happened to shut down midway through a CoW remap: "This is what happens [after failed log recovery]: "Phase 1 - find and verify superblock... "Phase 2 - using internal log " - zero log... " - scan filesystem freespace and inode maps... " - found root inode chunk "Phase 3 - for each AG... " - scan (but don't clear) agi unlinked lists... " - process known inodes and perform inode discovery... " - agno = 0 "data fork in regular inode 134 claims CoW block 376 "correcting nextents for inode 134 "bad data fork in inode 134 "would have cleared inode 134" Hou Tao dissected the log contents of exactly such a crash: "According to the implementation of xfs_defer_finish(), these ops should be completed in the following sequence: "Have been done: "(1) CUI: Oper (160) "(2) BUI: Oper (161) "(3) CUD: Oper (194), for CUI Oper (160) "(4) RUI A: Oper (197), free rmap [0x155, 2, -9] "Should be done: "(5) BUD: for BUI Oper (161) "(6) RUI B: add rmap [0x155, 2, 137] "(7) RUD: for RUI A "(8) RUD: for RUI B "Actually be done by xlog_recover_process_intents() "(5) BUD: for BUI Oper (161) "(6) RUI B: add rmap [0x155, 2, 137] "(7) RUD: for RUI B "(8) RUD: for RUI A "So the rmap entry [0x155, 2, -9] for COW should be freed firstly, then a new rmap entry [0x155, 2, 137] will be added. However, as we can see from the log record in post_mount.log (generated after umount) and the trace print, the new rmap entry [0x155, 2, 137] are added firstly, then the rmap entry [0x155, 2, -9] are freed." When reconstructing the internal log state from the log items found on disk, it's required that deferred ops replay in exactly the same order that they would have had the filesystem not gone down. However, replaying unfinished deferred ops can create /more/ deferred ops. These new deferred ops are finished in the wrong order. This causes fs corruption and replay crashes, so let's create a single defer_ops to handle the subsequent ops created during replay, then use one single transaction at the end of log recovery to ensure that everything is replayed in the same order as they're supposed to be. Reported-by: Amir Goldstein <amir73il@gmail.com> Analyzed-by: Hou Tao <houtao1@huawei.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Tested-by: Amir Goldstein <amir73il@gmail.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-11-22 12:53:02 +08:00
{
xfs: proper replay of deferred ops queued during log recovery When we replay unfinished intent items that have been recovered from the log, it's possible that the replay will cause the creation of more deferred work items. As outlined in commit 509955823cc9c ("xfs: log recovery should replay deferred ops in order"), later work items have an implicit ordering dependency on earlier work items. Therefore, recovery must replay the items (both recovered and created) in the same order that they would have been during normal operation. For log recovery, we enforce this ordering by using an empty transaction to collect deferred ops that get created in the process of recovering a log intent item to prevent them from being committed before the rest of the recovered intent items. After we finish committing all the recovered log items, we allocate a transaction with an enormous block reservation, splice our huge list of created deferred ops into that transaction, and commit it, thereby finishing all those ops. This is /really/ hokey -- it's the one place in XFS where we allow nested transactions; the splicing of the defer ops list is is inelegant and has to be done twice per recovery function; and the broken way we handle inode pointers and block reservations cause subtle use-after-free and allocator problems that will be fixed by this patch and the two patches after it. Therefore, replace the hokey empty transaction with a structure designed to capture each chain of deferred ops that are created as part of recovering a single unfinished log intent. Finally, refactor the loop that replays those chains to do so using one transaction per chain. Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de>
2020-09-26 08:39:37 +08:00
struct xfs_defer_capture *dfc, *next;
xfs: log recovery should replay deferred ops in order As part of testing log recovery with dm_log_writes, Amir Goldstein discovered an error in the deferred ops recovery that lead to corruption of the filesystem metadata if a reflink+rmap filesystem happened to shut down midway through a CoW remap: "This is what happens [after failed log recovery]: "Phase 1 - find and verify superblock... "Phase 2 - using internal log " - zero log... " - scan filesystem freespace and inode maps... " - found root inode chunk "Phase 3 - for each AG... " - scan (but don't clear) agi unlinked lists... " - process known inodes and perform inode discovery... " - agno = 0 "data fork in regular inode 134 claims CoW block 376 "correcting nextents for inode 134 "bad data fork in inode 134 "would have cleared inode 134" Hou Tao dissected the log contents of exactly such a crash: "According to the implementation of xfs_defer_finish(), these ops should be completed in the following sequence: "Have been done: "(1) CUI: Oper (160) "(2) BUI: Oper (161) "(3) CUD: Oper (194), for CUI Oper (160) "(4) RUI A: Oper (197), free rmap [0x155, 2, -9] "Should be done: "(5) BUD: for BUI Oper (161) "(6) RUI B: add rmap [0x155, 2, 137] "(7) RUD: for RUI A "(8) RUD: for RUI B "Actually be done by xlog_recover_process_intents() "(5) BUD: for BUI Oper (161) "(6) RUI B: add rmap [0x155, 2, 137] "(7) RUD: for RUI B "(8) RUD: for RUI A "So the rmap entry [0x155, 2, -9] for COW should be freed firstly, then a new rmap entry [0x155, 2, 137] will be added. However, as we can see from the log record in post_mount.log (generated after umount) and the trace print, the new rmap entry [0x155, 2, 137] are added firstly, then the rmap entry [0x155, 2, -9] are freed." When reconstructing the internal log state from the log items found on disk, it's required that deferred ops replay in exactly the same order that they would have had the filesystem not gone down. However, replaying unfinished deferred ops can create /more/ deferred ops. These new deferred ops are finished in the wrong order. This causes fs corruption and replay crashes, so let's create a single defer_ops to handle the subsequent ops created during replay, then use one single transaction at the end of log recovery to ensure that everything is replayed in the same order as they're supposed to be. Reported-by: Amir Goldstein <amir73il@gmail.com> Analyzed-by: Hou Tao <houtao1@huawei.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Tested-by: Amir Goldstein <amir73il@gmail.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-11-22 12:53:02 +08:00
struct xfs_trans *tp;
xfs: proper replay of deferred ops queued during log recovery When we replay unfinished intent items that have been recovered from the log, it's possible that the replay will cause the creation of more deferred work items. As outlined in commit 509955823cc9c ("xfs: log recovery should replay deferred ops in order"), later work items have an implicit ordering dependency on earlier work items. Therefore, recovery must replay the items (both recovered and created) in the same order that they would have been during normal operation. For log recovery, we enforce this ordering by using an empty transaction to collect deferred ops that get created in the process of recovering a log intent item to prevent them from being committed before the rest of the recovered intent items. After we finish committing all the recovered log items, we allocate a transaction with an enormous block reservation, splice our huge list of created deferred ops into that transaction, and commit it, thereby finishing all those ops. This is /really/ hokey -- it's the one place in XFS where we allow nested transactions; the splicing of the defer ops list is is inelegant and has to be done twice per recovery function; and the broken way we handle inode pointers and block reservations cause subtle use-after-free and allocator problems that will be fixed by this patch and the two patches after it. Therefore, replace the hokey empty transaction with a structure designed to capture each chain of deferred ops that are created as part of recovering a single unfinished log intent. Finally, refactor the loop that replays those chains to do so using one transaction per chain. Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de>
2020-09-26 08:39:37 +08:00
int error = 0;
xfs: log recovery should replay deferred ops in order As part of testing log recovery with dm_log_writes, Amir Goldstein discovered an error in the deferred ops recovery that lead to corruption of the filesystem metadata if a reflink+rmap filesystem happened to shut down midway through a CoW remap: "This is what happens [after failed log recovery]: "Phase 1 - find and verify superblock... "Phase 2 - using internal log " - zero log... " - scan filesystem freespace and inode maps... " - found root inode chunk "Phase 3 - for each AG... " - scan (but don't clear) agi unlinked lists... " - process known inodes and perform inode discovery... " - agno = 0 "data fork in regular inode 134 claims CoW block 376 "correcting nextents for inode 134 "bad data fork in inode 134 "would have cleared inode 134" Hou Tao dissected the log contents of exactly such a crash: "According to the implementation of xfs_defer_finish(), these ops should be completed in the following sequence: "Have been done: "(1) CUI: Oper (160) "(2) BUI: Oper (161) "(3) CUD: Oper (194), for CUI Oper (160) "(4) RUI A: Oper (197), free rmap [0x155, 2, -9] "Should be done: "(5) BUD: for BUI Oper (161) "(6) RUI B: add rmap [0x155, 2, 137] "(7) RUD: for RUI A "(8) RUD: for RUI B "Actually be done by xlog_recover_process_intents() "(5) BUD: for BUI Oper (161) "(6) RUI B: add rmap [0x155, 2, 137] "(7) RUD: for RUI B "(8) RUD: for RUI A "So the rmap entry [0x155, 2, -9] for COW should be freed firstly, then a new rmap entry [0x155, 2, 137] will be added. However, as we can see from the log record in post_mount.log (generated after umount) and the trace print, the new rmap entry [0x155, 2, 137] are added firstly, then the rmap entry [0x155, 2, -9] are freed." When reconstructing the internal log state from the log items found on disk, it's required that deferred ops replay in exactly the same order that they would have had the filesystem not gone down. However, replaying unfinished deferred ops can create /more/ deferred ops. These new deferred ops are finished in the wrong order. This causes fs corruption and replay crashes, so let's create a single defer_ops to handle the subsequent ops created during replay, then use one single transaction at the end of log recovery to ensure that everything is replayed in the same order as they're supposed to be. Reported-by: Amir Goldstein <amir73il@gmail.com> Analyzed-by: Hou Tao <houtao1@huawei.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Tested-by: Amir Goldstein <amir73il@gmail.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-11-22 12:53:02 +08:00
xfs: proper replay of deferred ops queued during log recovery When we replay unfinished intent items that have been recovered from the log, it's possible that the replay will cause the creation of more deferred work items. As outlined in commit 509955823cc9c ("xfs: log recovery should replay deferred ops in order"), later work items have an implicit ordering dependency on earlier work items. Therefore, recovery must replay the items (both recovered and created) in the same order that they would have been during normal operation. For log recovery, we enforce this ordering by using an empty transaction to collect deferred ops that get created in the process of recovering a log intent item to prevent them from being committed before the rest of the recovered intent items. After we finish committing all the recovered log items, we allocate a transaction with an enormous block reservation, splice our huge list of created deferred ops into that transaction, and commit it, thereby finishing all those ops. This is /really/ hokey -- it's the one place in XFS where we allow nested transactions; the splicing of the defer ops list is is inelegant and has to be done twice per recovery function; and the broken way we handle inode pointers and block reservations cause subtle use-after-free and allocator problems that will be fixed by this patch and the two patches after it. Therefore, replace the hokey empty transaction with a structure designed to capture each chain of deferred ops that are created as part of recovering a single unfinished log intent. Finally, refactor the loop that replays those chains to do so using one transaction per chain. Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de>
2020-09-26 08:39:37 +08:00
list_for_each_entry_safe(dfc, next, capture_list, dfc_list) {
struct xfs_trans_res resv;
struct xfs_defer_resources dres;
/*
* Create a new transaction reservation from the captured
* information. Set logcount to 1 to force the new transaction
* to regrant every roll so that we can make forward progress
* in recovery no matter how full the log might be.
*/
resv.tr_logres = dfc->dfc_logres;
resv.tr_logcount = 1;
resv.tr_logflags = XFS_TRANS_PERM_LOG_RES;
error = xfs_trans_alloc(mp, &resv, dfc->dfc_blkres,
dfc->dfc_rtxres, XFS_TRANS_RESERVE, &tp);
if (error) {
xfs: log shutdown triggers should only shut down the log We've got a mess on our hands. 1. xfs_trans_commit() cannot cancel transactions because the mount is shut down - that causes dirty, aborted, unlogged log items to sit unpinned in memory and potentially get written to disk before the log is shut down. Hence xfs_trans_commit() can only abort transactions when xlog_is_shutdown() is true. 2. xfs_force_shutdown() is used in places to cause the current modification to be aborted via xfs_trans_commit() because it may be impractical or impossible to cancel the transaction directly, and hence xfs_trans_commit() must cancel transactions when xfs_is_shutdown() is true in this situation. But we can't do that because of #1. 3. Log IO errors cause log shutdowns by calling xfs_force_shutdown() to shut down the mount and then the log from log IO completion. 4. xfs_force_shutdown() can result in a log force being issued, which has to wait for log IO completion before it will mark the log as shut down. If #3 races with some other shutdown trigger that runs a log force, we rely on xfs_force_shutdown() silently ignoring #3 and avoiding shutting down the log until the failed log force completes. 5. To ensure #2 always works, we have to ensure that xfs_force_shutdown() does not return until the the log is shut down. But in the case of #4, this will result in a deadlock because the log Io completion will block waiting for a log force to complete which is blocked waiting for log IO to complete.... So the very first thing we have to do here to untangle this mess is dissociate log shutdown triggers from mount shutdowns. We already have xlog_forced_shutdown, which will atomically transistion to the log a shutdown state. Due to internal asserts it cannot be called multiple times, but was done simply because the only place that could call it was xfs_do_force_shutdown() (i.e. the mount shutdown!) and that could only call it once and once only. So the first thing we do is remove the asserts. We then convert all the internal log shutdown triggers to call xlog_force_shutdown() directly instead of xfs_force_shutdown(). This allows the log shutdown triggers to shut down the log without needing to care about mount based shutdown constraints. This means we shut down the log independently of the mount and the mount may not notice this until it's next attempt to read or modify metadata. At that point (e.g. xfs_trans_commit()) it will see that the log is shutdown, error out and shutdown the mount. To ensure that all the unmount behaviours and asserts track correctly as a result of a log shutdown, propagate the shutdown up to the mount if it is not already set. This keeps the mount and log state in sync, and saves a huge amount of hassle where code fails because of a log shutdown but only checks for mount shutdowns and hence ends up doing the wrong thing. Cleaning up that mess is an exercise for another day. This enables us to address the other problems noted above in followup patches. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2022-03-30 09:22:01 +08:00
xlog_force_shutdown(mp->m_log, SHUTDOWN_LOG_IO_ERROR);
xfs: proper replay of deferred ops queued during log recovery When we replay unfinished intent items that have been recovered from the log, it's possible that the replay will cause the creation of more deferred work items. As outlined in commit 509955823cc9c ("xfs: log recovery should replay deferred ops in order"), later work items have an implicit ordering dependency on earlier work items. Therefore, recovery must replay the items (both recovered and created) in the same order that they would have been during normal operation. For log recovery, we enforce this ordering by using an empty transaction to collect deferred ops that get created in the process of recovering a log intent item to prevent them from being committed before the rest of the recovered intent items. After we finish committing all the recovered log items, we allocate a transaction with an enormous block reservation, splice our huge list of created deferred ops into that transaction, and commit it, thereby finishing all those ops. This is /really/ hokey -- it's the one place in XFS where we allow nested transactions; the splicing of the defer ops list is is inelegant and has to be done twice per recovery function; and the broken way we handle inode pointers and block reservations cause subtle use-after-free and allocator problems that will be fixed by this patch and the two patches after it. Therefore, replace the hokey empty transaction with a structure designed to capture each chain of deferred ops that are created as part of recovering a single unfinished log intent. Finally, refactor the loop that replays those chains to do so using one transaction per chain. Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de>
2020-09-26 08:39:37 +08:00
return error;
}
xfs: log recovery should replay deferred ops in order As part of testing log recovery with dm_log_writes, Amir Goldstein discovered an error in the deferred ops recovery that lead to corruption of the filesystem metadata if a reflink+rmap filesystem happened to shut down midway through a CoW remap: "This is what happens [after failed log recovery]: "Phase 1 - find and verify superblock... "Phase 2 - using internal log " - zero log... " - scan filesystem freespace and inode maps... " - found root inode chunk "Phase 3 - for each AG... " - scan (but don't clear) agi unlinked lists... " - process known inodes and perform inode discovery... " - agno = 0 "data fork in regular inode 134 claims CoW block 376 "correcting nextents for inode 134 "bad data fork in inode 134 "would have cleared inode 134" Hou Tao dissected the log contents of exactly such a crash: "According to the implementation of xfs_defer_finish(), these ops should be completed in the following sequence: "Have been done: "(1) CUI: Oper (160) "(2) BUI: Oper (161) "(3) CUD: Oper (194), for CUI Oper (160) "(4) RUI A: Oper (197), free rmap [0x155, 2, -9] "Should be done: "(5) BUD: for BUI Oper (161) "(6) RUI B: add rmap [0x155, 2, 137] "(7) RUD: for RUI A "(8) RUD: for RUI B "Actually be done by xlog_recover_process_intents() "(5) BUD: for BUI Oper (161) "(6) RUI B: add rmap [0x155, 2, 137] "(7) RUD: for RUI B "(8) RUD: for RUI A "So the rmap entry [0x155, 2, -9] for COW should be freed firstly, then a new rmap entry [0x155, 2, 137] will be added. However, as we can see from the log record in post_mount.log (generated after umount) and the trace print, the new rmap entry [0x155, 2, 137] are added firstly, then the rmap entry [0x155, 2, -9] are freed." When reconstructing the internal log state from the log items found on disk, it's required that deferred ops replay in exactly the same order that they would have had the filesystem not gone down. However, replaying unfinished deferred ops can create /more/ deferred ops. These new deferred ops are finished in the wrong order. This causes fs corruption and replay crashes, so let's create a single defer_ops to handle the subsequent ops created during replay, then use one single transaction at the end of log recovery to ensure that everything is replayed in the same order as they're supposed to be. Reported-by: Amir Goldstein <amir73il@gmail.com> Analyzed-by: Hou Tao <houtao1@huawei.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Tested-by: Amir Goldstein <amir73il@gmail.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-11-22 12:53:02 +08:00
xfs: proper replay of deferred ops queued during log recovery When we replay unfinished intent items that have been recovered from the log, it's possible that the replay will cause the creation of more deferred work items. As outlined in commit 509955823cc9c ("xfs: log recovery should replay deferred ops in order"), later work items have an implicit ordering dependency on earlier work items. Therefore, recovery must replay the items (both recovered and created) in the same order that they would have been during normal operation. For log recovery, we enforce this ordering by using an empty transaction to collect deferred ops that get created in the process of recovering a log intent item to prevent them from being committed before the rest of the recovered intent items. After we finish committing all the recovered log items, we allocate a transaction with an enormous block reservation, splice our huge list of created deferred ops into that transaction, and commit it, thereby finishing all those ops. This is /really/ hokey -- it's the one place in XFS where we allow nested transactions; the splicing of the defer ops list is is inelegant and has to be done twice per recovery function; and the broken way we handle inode pointers and block reservations cause subtle use-after-free and allocator problems that will be fixed by this patch and the two patches after it. Therefore, replace the hokey empty transaction with a structure designed to capture each chain of deferred ops that are created as part of recovering a single unfinished log intent. Finally, refactor the loop that replays those chains to do so using one transaction per chain. Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de>
2020-09-26 08:39:37 +08:00
/*
* Transfer to this new transaction all the dfops we captured
* from recovering a single intent item.
*/
list_del_init(&dfc->dfc_list);
xfs_defer_ops_continue(dfc, tp, &dres);
xfs: proper replay of deferred ops queued during log recovery When we replay unfinished intent items that have been recovered from the log, it's possible that the replay will cause the creation of more deferred work items. As outlined in commit 509955823cc9c ("xfs: log recovery should replay deferred ops in order"), later work items have an implicit ordering dependency on earlier work items. Therefore, recovery must replay the items (both recovered and created) in the same order that they would have been during normal operation. For log recovery, we enforce this ordering by using an empty transaction to collect deferred ops that get created in the process of recovering a log intent item to prevent them from being committed before the rest of the recovered intent items. After we finish committing all the recovered log items, we allocate a transaction with an enormous block reservation, splice our huge list of created deferred ops into that transaction, and commit it, thereby finishing all those ops. This is /really/ hokey -- it's the one place in XFS where we allow nested transactions; the splicing of the defer ops list is is inelegant and has to be done twice per recovery function; and the broken way we handle inode pointers and block reservations cause subtle use-after-free and allocator problems that will be fixed by this patch and the two patches after it. Therefore, replace the hokey empty transaction with a structure designed to capture each chain of deferred ops that are created as part of recovering a single unfinished log intent. Finally, refactor the loop that replays those chains to do so using one transaction per chain. Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de>
2020-09-26 08:39:37 +08:00
error = xfs_trans_commit(tp);
xfs_defer_resources_rele(&dres);
xfs: proper replay of deferred ops queued during log recovery When we replay unfinished intent items that have been recovered from the log, it's possible that the replay will cause the creation of more deferred work items. As outlined in commit 509955823cc9c ("xfs: log recovery should replay deferred ops in order"), later work items have an implicit ordering dependency on earlier work items. Therefore, recovery must replay the items (both recovered and created) in the same order that they would have been during normal operation. For log recovery, we enforce this ordering by using an empty transaction to collect deferred ops that get created in the process of recovering a log intent item to prevent them from being committed before the rest of the recovered intent items. After we finish committing all the recovered log items, we allocate a transaction with an enormous block reservation, splice our huge list of created deferred ops into that transaction, and commit it, thereby finishing all those ops. This is /really/ hokey -- it's the one place in XFS where we allow nested transactions; the splicing of the defer ops list is is inelegant and has to be done twice per recovery function; and the broken way we handle inode pointers and block reservations cause subtle use-after-free and allocator problems that will be fixed by this patch and the two patches after it. Therefore, replace the hokey empty transaction with a structure designed to capture each chain of deferred ops that are created as part of recovering a single unfinished log intent. Finally, refactor the loop that replays those chains to do so using one transaction per chain. Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de>
2020-09-26 08:39:37 +08:00
if (error)
return error;
}
ASSERT(list_empty(capture_list));
return 0;
xfs: log recovery should replay deferred ops in order As part of testing log recovery with dm_log_writes, Amir Goldstein discovered an error in the deferred ops recovery that lead to corruption of the filesystem metadata if a reflink+rmap filesystem happened to shut down midway through a CoW remap: "This is what happens [after failed log recovery]: "Phase 1 - find and verify superblock... "Phase 2 - using internal log " - zero log... " - scan filesystem freespace and inode maps... " - found root inode chunk "Phase 3 - for each AG... " - scan (but don't clear) agi unlinked lists... " - process known inodes and perform inode discovery... " - agno = 0 "data fork in regular inode 134 claims CoW block 376 "correcting nextents for inode 134 "bad data fork in inode 134 "would have cleared inode 134" Hou Tao dissected the log contents of exactly such a crash: "According to the implementation of xfs_defer_finish(), these ops should be completed in the following sequence: "Have been done: "(1) CUI: Oper (160) "(2) BUI: Oper (161) "(3) CUD: Oper (194), for CUI Oper (160) "(4) RUI A: Oper (197), free rmap [0x155, 2, -9] "Should be done: "(5) BUD: for BUI Oper (161) "(6) RUI B: add rmap [0x155, 2, 137] "(7) RUD: for RUI A "(8) RUD: for RUI B "Actually be done by xlog_recover_process_intents() "(5) BUD: for BUI Oper (161) "(6) RUI B: add rmap [0x155, 2, 137] "(7) RUD: for RUI B "(8) RUD: for RUI A "So the rmap entry [0x155, 2, -9] for COW should be freed firstly, then a new rmap entry [0x155, 2, 137] will be added. However, as we can see from the log record in post_mount.log (generated after umount) and the trace print, the new rmap entry [0x155, 2, 137] are added firstly, then the rmap entry [0x155, 2, -9] are freed." When reconstructing the internal log state from the log items found on disk, it's required that deferred ops replay in exactly the same order that they would have had the filesystem not gone down. However, replaying unfinished deferred ops can create /more/ deferred ops. These new deferred ops are finished in the wrong order. This causes fs corruption and replay crashes, so let's create a single defer_ops to handle the subsequent ops created during replay, then use one single transaction at the end of log recovery to ensure that everything is replayed in the same order as they're supposed to be. Reported-by: Amir Goldstein <amir73il@gmail.com> Analyzed-by: Hou Tao <houtao1@huawei.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Tested-by: Amir Goldstein <amir73il@gmail.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-11-22 12:53:02 +08:00
}
xfs: proper replay of deferred ops queued during log recovery When we replay unfinished intent items that have been recovered from the log, it's possible that the replay will cause the creation of more deferred work items. As outlined in commit 509955823cc9c ("xfs: log recovery should replay deferred ops in order"), later work items have an implicit ordering dependency on earlier work items. Therefore, recovery must replay the items (both recovered and created) in the same order that they would have been during normal operation. For log recovery, we enforce this ordering by using an empty transaction to collect deferred ops that get created in the process of recovering a log intent item to prevent them from being committed before the rest of the recovered intent items. After we finish committing all the recovered log items, we allocate a transaction with an enormous block reservation, splice our huge list of created deferred ops into that transaction, and commit it, thereby finishing all those ops. This is /really/ hokey -- it's the one place in XFS where we allow nested transactions; the splicing of the defer ops list is is inelegant and has to be done twice per recovery function; and the broken way we handle inode pointers and block reservations cause subtle use-after-free and allocator problems that will be fixed by this patch and the two patches after it. Therefore, replace the hokey empty transaction with a structure designed to capture each chain of deferred ops that are created as part of recovering a single unfinished log intent. Finally, refactor the loop that replays those chains to do so using one transaction per chain. Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de>
2020-09-26 08:39:37 +08:00
/* Release all the captured defer ops and capture structures in this list. */
static void
xlog_abort_defer_ops(
struct xfs_mount *mp,
struct list_head *capture_list)
{
struct xfs_defer_capture *dfc;
struct xfs_defer_capture *next;
list_for_each_entry_safe(dfc, next, capture_list, dfc_list) {
list_del_init(&dfc->dfc_list);
xfs_defer_ops_capture_free(mp, dfc);
xfs: proper replay of deferred ops queued during log recovery When we replay unfinished intent items that have been recovered from the log, it's possible that the replay will cause the creation of more deferred work items. As outlined in commit 509955823cc9c ("xfs: log recovery should replay deferred ops in order"), later work items have an implicit ordering dependency on earlier work items. Therefore, recovery must replay the items (both recovered and created) in the same order that they would have been during normal operation. For log recovery, we enforce this ordering by using an empty transaction to collect deferred ops that get created in the process of recovering a log intent item to prevent them from being committed before the rest of the recovered intent items. After we finish committing all the recovered log items, we allocate a transaction with an enormous block reservation, splice our huge list of created deferred ops into that transaction, and commit it, thereby finishing all those ops. This is /really/ hokey -- it's the one place in XFS where we allow nested transactions; the splicing of the defer ops list is is inelegant and has to be done twice per recovery function; and the broken way we handle inode pointers and block reservations cause subtle use-after-free and allocator problems that will be fixed by this patch and the two patches after it. Therefore, replace the hokey empty transaction with a structure designed to capture each chain of deferred ops that are created as part of recovering a single unfinished log intent. Finally, refactor the loop that replays those chains to do so using one transaction per chain. Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de>
2020-09-26 08:39:37 +08:00
}
}
xfs: shutdown in intent recovery has non-intent items in the AIL generic/388 triggered a failure in RUI recovery due to a corrupted btree record and the system then locked up hard due to a subsequent assert failure while holding a spinlock cancelling intents: XFS (pmem1): Corruption of in-memory data (0x8) detected at xfs_do_force_shutdown+0x1a/0x20 (fs/xfs/xfs_trans.c:964). Shutting down filesystem. XFS (pmem1): Please unmount the filesystem and rectify the problem(s) XFS: Assertion failed: !xlog_item_is_intent(lip), file: fs/xfs/xfs_log_recover.c, line: 2632 Call Trace: <TASK> xlog_recover_cancel_intents.isra.0+0xd1/0x120 xlog_recover_finish+0xb9/0x110 xfs_log_mount_finish+0x15a/0x1e0 xfs_mountfs+0x540/0x910 xfs_fs_fill_super+0x476/0x830 get_tree_bdev+0x171/0x270 ? xfs_init_fs_context+0x1e0/0x1e0 xfs_fs_get_tree+0x15/0x20 vfs_get_tree+0x24/0xc0 path_mount+0x304/0xba0 ? putname+0x55/0x60 __x64_sys_mount+0x108/0x140 do_syscall_64+0x35/0x80 entry_SYSCALL_64_after_hwframe+0x44/0xae Essentially, there's dirty metadata in the AIL from intent recovery transactions, so when we go to cancel the remaining intents we assume that all objects after the first non-intent log item in the AIL are not intents. This is not true. Intent recovery can log new intents to continue the operations the original intent could not complete in a single transaction. The new intents are committed before they are deferred, which means if the CIL commits in the background they will get inserted into the AIL at the head. Hence if we shut down the filesystem while processing intent recovery, the AIL may have new intents active at the current head. Hence this check: /* * We're done when we see something other than an intent. * There should be no intents left in the AIL now. */ if (!xlog_item_is_intent(lip)) { #ifdef DEBUG for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur)) ASSERT(!xlog_item_is_intent(lip)); #endif break; } in both xlog_recover_process_intents() and log_recover_cancel_intents() is simply not valid. It was valid back when we only had EFI/EFD intents and didn't chain intents, but it hasn't been valid ever since intent recovery could create and commit new intents. Given that crashing the mount task like this pretty much prevents diagnosing what went wrong that lead to the initial failure that triggered intent cancellation, just remove the checks altogether. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2022-03-30 09:22:00 +08:00
/*
* When this is called, all of the log intent items which did not have
xfs: shutdown in intent recovery has non-intent items in the AIL generic/388 triggered a failure in RUI recovery due to a corrupted btree record and the system then locked up hard due to a subsequent assert failure while holding a spinlock cancelling intents: XFS (pmem1): Corruption of in-memory data (0x8) detected at xfs_do_force_shutdown+0x1a/0x20 (fs/xfs/xfs_trans.c:964). Shutting down filesystem. XFS (pmem1): Please unmount the filesystem and rectify the problem(s) XFS: Assertion failed: !xlog_item_is_intent(lip), file: fs/xfs/xfs_log_recover.c, line: 2632 Call Trace: <TASK> xlog_recover_cancel_intents.isra.0+0xd1/0x120 xlog_recover_finish+0xb9/0x110 xfs_log_mount_finish+0x15a/0x1e0 xfs_mountfs+0x540/0x910 xfs_fs_fill_super+0x476/0x830 get_tree_bdev+0x171/0x270 ? xfs_init_fs_context+0x1e0/0x1e0 xfs_fs_get_tree+0x15/0x20 vfs_get_tree+0x24/0xc0 path_mount+0x304/0xba0 ? putname+0x55/0x60 __x64_sys_mount+0x108/0x140 do_syscall_64+0x35/0x80 entry_SYSCALL_64_after_hwframe+0x44/0xae Essentially, there's dirty metadata in the AIL from intent recovery transactions, so when we go to cancel the remaining intents we assume that all objects after the first non-intent log item in the AIL are not intents. This is not true. Intent recovery can log new intents to continue the operations the original intent could not complete in a single transaction. The new intents are committed before they are deferred, which means if the CIL commits in the background they will get inserted into the AIL at the head. Hence if we shut down the filesystem while processing intent recovery, the AIL may have new intents active at the current head. Hence this check: /* * We're done when we see something other than an intent. * There should be no intents left in the AIL now. */ if (!xlog_item_is_intent(lip)) { #ifdef DEBUG for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur)) ASSERT(!xlog_item_is_intent(lip)); #endif break; } in both xlog_recover_process_intents() and log_recover_cancel_intents() is simply not valid. It was valid back when we only had EFI/EFD intents and didn't chain intents, but it hasn't been valid ever since intent recovery could create and commit new intents. Given that crashing the mount task like this pretty much prevents diagnosing what went wrong that lead to the initial failure that triggered intent cancellation, just remove the checks altogether. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2022-03-30 09:22:00 +08:00
* corresponding log done items should be in the AIL. What we do now is update
* the data structures associated with each one.
*
xfs: shutdown in intent recovery has non-intent items in the AIL generic/388 triggered a failure in RUI recovery due to a corrupted btree record and the system then locked up hard due to a subsequent assert failure while holding a spinlock cancelling intents: XFS (pmem1): Corruption of in-memory data (0x8) detected at xfs_do_force_shutdown+0x1a/0x20 (fs/xfs/xfs_trans.c:964). Shutting down filesystem. XFS (pmem1): Please unmount the filesystem and rectify the problem(s) XFS: Assertion failed: !xlog_item_is_intent(lip), file: fs/xfs/xfs_log_recover.c, line: 2632 Call Trace: <TASK> xlog_recover_cancel_intents.isra.0+0xd1/0x120 xlog_recover_finish+0xb9/0x110 xfs_log_mount_finish+0x15a/0x1e0 xfs_mountfs+0x540/0x910 xfs_fs_fill_super+0x476/0x830 get_tree_bdev+0x171/0x270 ? xfs_init_fs_context+0x1e0/0x1e0 xfs_fs_get_tree+0x15/0x20 vfs_get_tree+0x24/0xc0 path_mount+0x304/0xba0 ? putname+0x55/0x60 __x64_sys_mount+0x108/0x140 do_syscall_64+0x35/0x80 entry_SYSCALL_64_after_hwframe+0x44/0xae Essentially, there's dirty metadata in the AIL from intent recovery transactions, so when we go to cancel the remaining intents we assume that all objects after the first non-intent log item in the AIL are not intents. This is not true. Intent recovery can log new intents to continue the operations the original intent could not complete in a single transaction. The new intents are committed before they are deferred, which means if the CIL commits in the background they will get inserted into the AIL at the head. Hence if we shut down the filesystem while processing intent recovery, the AIL may have new intents active at the current head. Hence this check: /* * We're done when we see something other than an intent. * There should be no intents left in the AIL now. */ if (!xlog_item_is_intent(lip)) { #ifdef DEBUG for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur)) ASSERT(!xlog_item_is_intent(lip)); #endif break; } in both xlog_recover_process_intents() and log_recover_cancel_intents() is simply not valid. It was valid back when we only had EFI/EFD intents and didn't chain intents, but it hasn't been valid ever since intent recovery could create and commit new intents. Given that crashing the mount task like this pretty much prevents diagnosing what went wrong that lead to the initial failure that triggered intent cancellation, just remove the checks altogether. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2022-03-30 09:22:00 +08:00
* Since we process the log intent items in normal transactions, they will be
* removed at some point after the commit. This prevents us from just walking
* down the list processing each one. We'll use a flag in the intent item to
* skip those that we've already processed and use the AIL iteration mechanism's
* generation count to try to speed this up at least a bit.
*
xfs: shutdown in intent recovery has non-intent items in the AIL generic/388 triggered a failure in RUI recovery due to a corrupted btree record and the system then locked up hard due to a subsequent assert failure while holding a spinlock cancelling intents: XFS (pmem1): Corruption of in-memory data (0x8) detected at xfs_do_force_shutdown+0x1a/0x20 (fs/xfs/xfs_trans.c:964). Shutting down filesystem. XFS (pmem1): Please unmount the filesystem and rectify the problem(s) XFS: Assertion failed: !xlog_item_is_intent(lip), file: fs/xfs/xfs_log_recover.c, line: 2632 Call Trace: <TASK> xlog_recover_cancel_intents.isra.0+0xd1/0x120 xlog_recover_finish+0xb9/0x110 xfs_log_mount_finish+0x15a/0x1e0 xfs_mountfs+0x540/0x910 xfs_fs_fill_super+0x476/0x830 get_tree_bdev+0x171/0x270 ? xfs_init_fs_context+0x1e0/0x1e0 xfs_fs_get_tree+0x15/0x20 vfs_get_tree+0x24/0xc0 path_mount+0x304/0xba0 ? putname+0x55/0x60 __x64_sys_mount+0x108/0x140 do_syscall_64+0x35/0x80 entry_SYSCALL_64_after_hwframe+0x44/0xae Essentially, there's dirty metadata in the AIL from intent recovery transactions, so when we go to cancel the remaining intents we assume that all objects after the first non-intent log item in the AIL are not intents. This is not true. Intent recovery can log new intents to continue the operations the original intent could not complete in a single transaction. The new intents are committed before they are deferred, which means if the CIL commits in the background they will get inserted into the AIL at the head. Hence if we shut down the filesystem while processing intent recovery, the AIL may have new intents active at the current head. Hence this check: /* * We're done when we see something other than an intent. * There should be no intents left in the AIL now. */ if (!xlog_item_is_intent(lip)) { #ifdef DEBUG for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur)) ASSERT(!xlog_item_is_intent(lip)); #endif break; } in both xlog_recover_process_intents() and log_recover_cancel_intents() is simply not valid. It was valid back when we only had EFI/EFD intents and didn't chain intents, but it hasn't been valid ever since intent recovery could create and commit new intents. Given that crashing the mount task like this pretty much prevents diagnosing what went wrong that lead to the initial failure that triggered intent cancellation, just remove the checks altogether. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2022-03-30 09:22:00 +08:00
* When we start, we know that the intents are the only things in the AIL. As we
* process them, however, other items are added to the AIL. Hence we know we
* have started recovery on all the pending intents when we find an non-intent
* item in the AIL.
*/
STATIC int
xlog_recover_process_intents(
struct xlog *log)
{
xfs: proper replay of deferred ops queued during log recovery When we replay unfinished intent items that have been recovered from the log, it's possible that the replay will cause the creation of more deferred work items. As outlined in commit 509955823cc9c ("xfs: log recovery should replay deferred ops in order"), later work items have an implicit ordering dependency on earlier work items. Therefore, recovery must replay the items (both recovered and created) in the same order that they would have been during normal operation. For log recovery, we enforce this ordering by using an empty transaction to collect deferred ops that get created in the process of recovering a log intent item to prevent them from being committed before the rest of the recovered intent items. After we finish committing all the recovered log items, we allocate a transaction with an enormous block reservation, splice our huge list of created deferred ops into that transaction, and commit it, thereby finishing all those ops. This is /really/ hokey -- it's the one place in XFS where we allow nested transactions; the splicing of the defer ops list is is inelegant and has to be done twice per recovery function; and the broken way we handle inode pointers and block reservations cause subtle use-after-free and allocator problems that will be fixed by this patch and the two patches after it. Therefore, replace the hokey empty transaction with a structure designed to capture each chain of deferred ops that are created as part of recovering a single unfinished log intent. Finally, refactor the loop that replays those chains to do so using one transaction per chain. Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de>
2020-09-26 08:39:37 +08:00
LIST_HEAD(capture_list);
struct xfs_ail_cursor cur;
xfs: log recovery should replay deferred ops in order As part of testing log recovery with dm_log_writes, Amir Goldstein discovered an error in the deferred ops recovery that lead to corruption of the filesystem metadata if a reflink+rmap filesystem happened to shut down midway through a CoW remap: "This is what happens [after failed log recovery]: "Phase 1 - find and verify superblock... "Phase 2 - using internal log " - zero log... " - scan filesystem freespace and inode maps... " - found root inode chunk "Phase 3 - for each AG... " - scan (but don't clear) agi unlinked lists... " - process known inodes and perform inode discovery... " - agno = 0 "data fork in regular inode 134 claims CoW block 376 "correcting nextents for inode 134 "bad data fork in inode 134 "would have cleared inode 134" Hou Tao dissected the log contents of exactly such a crash: "According to the implementation of xfs_defer_finish(), these ops should be completed in the following sequence: "Have been done: "(1) CUI: Oper (160) "(2) BUI: Oper (161) "(3) CUD: Oper (194), for CUI Oper (160) "(4) RUI A: Oper (197), free rmap [0x155, 2, -9] "Should be done: "(5) BUD: for BUI Oper (161) "(6) RUI B: add rmap [0x155, 2, 137] "(7) RUD: for RUI A "(8) RUD: for RUI B "Actually be done by xlog_recover_process_intents() "(5) BUD: for BUI Oper (161) "(6) RUI B: add rmap [0x155, 2, 137] "(7) RUD: for RUI B "(8) RUD: for RUI A "So the rmap entry [0x155, 2, -9] for COW should be freed firstly, then a new rmap entry [0x155, 2, 137] will be added. However, as we can see from the log record in post_mount.log (generated after umount) and the trace print, the new rmap entry [0x155, 2, 137] are added firstly, then the rmap entry [0x155, 2, -9] are freed." When reconstructing the internal log state from the log items found on disk, it's required that deferred ops replay in exactly the same order that they would have had the filesystem not gone down. However, replaying unfinished deferred ops can create /more/ deferred ops. These new deferred ops are finished in the wrong order. This causes fs corruption and replay crashes, so let's create a single defer_ops to handle the subsequent ops created during replay, then use one single transaction at the end of log recovery to ensure that everything is replayed in the same order as they're supposed to be. Reported-by: Amir Goldstein <amir73il@gmail.com> Analyzed-by: Hou Tao <houtao1@huawei.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Tested-by: Amir Goldstein <amir73il@gmail.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-11-22 12:53:02 +08:00
struct xfs_log_item *lip;
struct xfs_ail *ailp;
xfs: proper replay of deferred ops queued during log recovery When we replay unfinished intent items that have been recovered from the log, it's possible that the replay will cause the creation of more deferred work items. As outlined in commit 509955823cc9c ("xfs: log recovery should replay deferred ops in order"), later work items have an implicit ordering dependency on earlier work items. Therefore, recovery must replay the items (both recovered and created) in the same order that they would have been during normal operation. For log recovery, we enforce this ordering by using an empty transaction to collect deferred ops that get created in the process of recovering a log intent item to prevent them from being committed before the rest of the recovered intent items. After we finish committing all the recovered log items, we allocate a transaction with an enormous block reservation, splice our huge list of created deferred ops into that transaction, and commit it, thereby finishing all those ops. This is /really/ hokey -- it's the one place in XFS where we allow nested transactions; the splicing of the defer ops list is is inelegant and has to be done twice per recovery function; and the broken way we handle inode pointers and block reservations cause subtle use-after-free and allocator problems that will be fixed by this patch and the two patches after it. Therefore, replace the hokey empty transaction with a structure designed to capture each chain of deferred ops that are created as part of recovering a single unfinished log intent. Finally, refactor the loop that replays those chains to do so using one transaction per chain. Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de>
2020-09-26 08:39:37 +08:00
int error = 0;
#if defined(DEBUG) || defined(XFS_WARN)
xfs_lsn_t last_lsn;
#endif
ailp = log->l_ailp;
spin_lock(&ailp->ail_lock);
#if defined(DEBUG) || defined(XFS_WARN)
last_lsn = xlog_assign_lsn(log->l_curr_cycle, log->l_curr_block);
#endif
xfs: proper replay of deferred ops queued during log recovery When we replay unfinished intent items that have been recovered from the log, it's possible that the replay will cause the creation of more deferred work items. As outlined in commit 509955823cc9c ("xfs: log recovery should replay deferred ops in order"), later work items have an implicit ordering dependency on earlier work items. Therefore, recovery must replay the items (both recovered and created) in the same order that they would have been during normal operation. For log recovery, we enforce this ordering by using an empty transaction to collect deferred ops that get created in the process of recovering a log intent item to prevent them from being committed before the rest of the recovered intent items. After we finish committing all the recovered log items, we allocate a transaction with an enormous block reservation, splice our huge list of created deferred ops into that transaction, and commit it, thereby finishing all those ops. This is /really/ hokey -- it's the one place in XFS where we allow nested transactions; the splicing of the defer ops list is is inelegant and has to be done twice per recovery function; and the broken way we handle inode pointers and block reservations cause subtle use-after-free and allocator problems that will be fixed by this patch and the two patches after it. Therefore, replace the hokey empty transaction with a structure designed to capture each chain of deferred ops that are created as part of recovering a single unfinished log intent. Finally, refactor the loop that replays those chains to do so using one transaction per chain. Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de>
2020-09-26 08:39:37 +08:00
for (lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
lip != NULL;
lip = xfs_trans_ail_cursor_next(ailp, &cur)) {
xfs: avoid a UAF when log intent item recovery fails KASAN reported a UAF bug when I was running xfs/235: BUG: KASAN: use-after-free in xlog_recover_process_intents+0xa77/0xae0 [xfs] Read of size 8 at addr ffff88804391b360 by task mount/5680 CPU: 2 PID: 5680 Comm: mount Not tainted 6.0.0-xfsx #6.0.0 77e7b52a4943a975441e5ac90a5ad7748b7867f6 Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 1.15.0-1 04/01/2014 Call Trace: <TASK> dump_stack_lvl+0x34/0x44 print_report.cold+0x2cc/0x682 kasan_report+0xa3/0x120 xlog_recover_process_intents+0xa77/0xae0 [xfs fb841c7180aad3f8359438576e27867f5795667e] xlog_recover_finish+0x7d/0x970 [xfs fb841c7180aad3f8359438576e27867f5795667e] xfs_log_mount_finish+0x2d7/0x5d0 [xfs fb841c7180aad3f8359438576e27867f5795667e] xfs_mountfs+0x11d4/0x1d10 [xfs fb841c7180aad3f8359438576e27867f5795667e] xfs_fs_fill_super+0x13d5/0x1a80 [xfs fb841c7180aad3f8359438576e27867f5795667e] get_tree_bdev+0x3da/0x6e0 vfs_get_tree+0x7d/0x240 path_mount+0xdd3/0x17d0 __x64_sys_mount+0x1fa/0x270 do_syscall_64+0x2b/0x80 entry_SYSCALL_64_after_hwframe+0x46/0xb0 RIP: 0033:0x7ff5bc069eae Code: 48 8b 0d 85 1f 0f 00 f7 d8 64 89 01 48 83 c8 ff c3 66 2e 0f 1f 84 00 00 00 00 00 90 f3 0f 1e fa 49 89 ca b8 a5 00 00 00 0f 05 <48> 3d 01 f0 ff ff 73 01 c3 48 8b 0d 52 1f 0f 00 f7 d8 64 89 01 48 RSP: 002b:00007ffe433fd448 EFLAGS: 00000246 ORIG_RAX: 00000000000000a5 RAX: ffffffffffffffda RBX: 0000000000000000 RCX: 00007ff5bc069eae RDX: 00005575d7213290 RSI: 00005575d72132d0 RDI: 00005575d72132b0 RBP: 00005575d7212fd0 R08: 00005575d7213230 R09: 00005575d7213fe0 R10: 0000000000000000 R11: 0000000000000246 R12: 0000000000000000 R13: 00005575d7213290 R14: 00005575d72132b0 R15: 00005575d7212fd0 </TASK> Allocated by task 5680: kasan_save_stack+0x1e/0x40 __kasan_slab_alloc+0x66/0x80 kmem_cache_alloc+0x152/0x320 xfs_rui_init+0x17a/0x1b0 [xfs] xlog_recover_rui_commit_pass2+0xb9/0x2e0 [xfs] xlog_recover_items_pass2+0xe9/0x220 [xfs] xlog_recover_commit_trans+0x673/0x900 [xfs] xlog_recovery_process_trans+0xbe/0x130 [xfs] xlog_recover_process_data+0x103/0x2a0 [xfs] xlog_do_recovery_pass+0x548/0xc60 [xfs] xlog_do_log_recovery+0x62/0xc0 [xfs] xlog_do_recover+0x73/0x480 [xfs] xlog_recover+0x229/0x460 [xfs] xfs_log_mount+0x284/0x640 [xfs] xfs_mountfs+0xf8b/0x1d10 [xfs] xfs_fs_fill_super+0x13d5/0x1a80 [xfs] get_tree_bdev+0x3da/0x6e0 vfs_get_tree+0x7d/0x240 path_mount+0xdd3/0x17d0 __x64_sys_mount+0x1fa/0x270 do_syscall_64+0x2b/0x80 entry_SYSCALL_64_after_hwframe+0x46/0xb0 Freed by task 5680: kasan_save_stack+0x1e/0x40 kasan_set_track+0x21/0x30 kasan_set_free_info+0x20/0x30 ____kasan_slab_free+0x144/0x1b0 slab_free_freelist_hook+0xab/0x180 kmem_cache_free+0x1f1/0x410 xfs_rud_item_release+0x33/0x80 [xfs] xfs_trans_free_items+0xc3/0x220 [xfs] xfs_trans_cancel+0x1fa/0x590 [xfs] xfs_rui_item_recover+0x913/0xd60 [xfs] xlog_recover_process_intents+0x24e/0xae0 [xfs] xlog_recover_finish+0x7d/0x970 [xfs] xfs_log_mount_finish+0x2d7/0x5d0 [xfs] xfs_mountfs+0x11d4/0x1d10 [xfs] xfs_fs_fill_super+0x13d5/0x1a80 [xfs] get_tree_bdev+0x3da/0x6e0 vfs_get_tree+0x7d/0x240 path_mount+0xdd3/0x17d0 __x64_sys_mount+0x1fa/0x270 do_syscall_64+0x2b/0x80 entry_SYSCALL_64_after_hwframe+0x46/0xb0 The buggy address belongs to the object at ffff88804391b300 which belongs to the cache xfs_rui_item of size 688 The buggy address is located 96 bytes inside of 688-byte region [ffff88804391b300, ffff88804391b5b0) The buggy address belongs to the physical page: page:ffffea00010e4600 refcount:1 mapcount:0 mapping:0000000000000000 index:0xffff888043919320 pfn:0x43918 head:ffffea00010e4600 order:2 compound_mapcount:0 compound_pincount:0 flags: 0x4fff80000010200(slab|head|node=1|zone=1|lastcpupid=0xfff) raw: 04fff80000010200 0000000000000000 dead000000000122 ffff88807f0eadc0 raw: ffff888043919320 0000000080140010 00000001ffffffff 0000000000000000 page dumped because: kasan: bad access detected Memory state around the buggy address: ffff88804391b200: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc ffff88804391b280: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc >ffff88804391b300: fa fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb ^ ffff88804391b380: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb ffff88804391b400: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb ================================================================== The test fuzzes an rmap btree block and starts writer threads to induce a filesystem shutdown on the corrupt block. When the filesystem is remounted, recovery will try to replay the committed rmap intent item, but the corruption problem causes the recovery transaction to fail. Cancelling the transaction frees the RUD, which frees the RUI that we recovered. When we return to xlog_recover_process_intents, @lip is now a dangling pointer, and we cannot use it to find the iop_recover method for the tracepoint. Hence we must store the item ops before calling ->iop_recover if we want to give it to the tracepoint so that the trace data will tell us exactly which intent item failed. Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de>
2022-10-17 08:54:40 +08:00
const struct xfs_item_ops *ops;
xfs: shutdown in intent recovery has non-intent items in the AIL generic/388 triggered a failure in RUI recovery due to a corrupted btree record and the system then locked up hard due to a subsequent assert failure while holding a spinlock cancelling intents: XFS (pmem1): Corruption of in-memory data (0x8) detected at xfs_do_force_shutdown+0x1a/0x20 (fs/xfs/xfs_trans.c:964). Shutting down filesystem. XFS (pmem1): Please unmount the filesystem and rectify the problem(s) XFS: Assertion failed: !xlog_item_is_intent(lip), file: fs/xfs/xfs_log_recover.c, line: 2632 Call Trace: <TASK> xlog_recover_cancel_intents.isra.0+0xd1/0x120 xlog_recover_finish+0xb9/0x110 xfs_log_mount_finish+0x15a/0x1e0 xfs_mountfs+0x540/0x910 xfs_fs_fill_super+0x476/0x830 get_tree_bdev+0x171/0x270 ? xfs_init_fs_context+0x1e0/0x1e0 xfs_fs_get_tree+0x15/0x20 vfs_get_tree+0x24/0xc0 path_mount+0x304/0xba0 ? putname+0x55/0x60 __x64_sys_mount+0x108/0x140 do_syscall_64+0x35/0x80 entry_SYSCALL_64_after_hwframe+0x44/0xae Essentially, there's dirty metadata in the AIL from intent recovery transactions, so when we go to cancel the remaining intents we assume that all objects after the first non-intent log item in the AIL are not intents. This is not true. Intent recovery can log new intents to continue the operations the original intent could not complete in a single transaction. The new intents are committed before they are deferred, which means if the CIL commits in the background they will get inserted into the AIL at the head. Hence if we shut down the filesystem while processing intent recovery, the AIL may have new intents active at the current head. Hence this check: /* * We're done when we see something other than an intent. * There should be no intents left in the AIL now. */ if (!xlog_item_is_intent(lip)) { #ifdef DEBUG for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur)) ASSERT(!xlog_item_is_intent(lip)); #endif break; } in both xlog_recover_process_intents() and log_recover_cancel_intents() is simply not valid. It was valid back when we only had EFI/EFD intents and didn't chain intents, but it hasn't been valid ever since intent recovery could create and commit new intents. Given that crashing the mount task like this pretty much prevents diagnosing what went wrong that lead to the initial failure that triggered intent cancellation, just remove the checks altogether. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2022-03-30 09:22:00 +08:00
if (!xlog_item_is_intent(lip))
break;
/*
* We should never see a redo item with a LSN higher than
* the last transaction we found in the log at the start
* of recovery.
*/
ASSERT(XFS_LSN_CMP(last_lsn, lip->li_lsn) >= 0);
xfs: log recovery should replay deferred ops in order As part of testing log recovery with dm_log_writes, Amir Goldstein discovered an error in the deferred ops recovery that lead to corruption of the filesystem metadata if a reflink+rmap filesystem happened to shut down midway through a CoW remap: "This is what happens [after failed log recovery]: "Phase 1 - find and verify superblock... "Phase 2 - using internal log " - zero log... " - scan filesystem freespace and inode maps... " - found root inode chunk "Phase 3 - for each AG... " - scan (but don't clear) agi unlinked lists... " - process known inodes and perform inode discovery... " - agno = 0 "data fork in regular inode 134 claims CoW block 376 "correcting nextents for inode 134 "bad data fork in inode 134 "would have cleared inode 134" Hou Tao dissected the log contents of exactly such a crash: "According to the implementation of xfs_defer_finish(), these ops should be completed in the following sequence: "Have been done: "(1) CUI: Oper (160) "(2) BUI: Oper (161) "(3) CUD: Oper (194), for CUI Oper (160) "(4) RUI A: Oper (197), free rmap [0x155, 2, -9] "Should be done: "(5) BUD: for BUI Oper (161) "(6) RUI B: add rmap [0x155, 2, 137] "(7) RUD: for RUI A "(8) RUD: for RUI B "Actually be done by xlog_recover_process_intents() "(5) BUD: for BUI Oper (161) "(6) RUI B: add rmap [0x155, 2, 137] "(7) RUD: for RUI B "(8) RUD: for RUI A "So the rmap entry [0x155, 2, -9] for COW should be freed firstly, then a new rmap entry [0x155, 2, 137] will be added. However, as we can see from the log record in post_mount.log (generated after umount) and the trace print, the new rmap entry [0x155, 2, 137] are added firstly, then the rmap entry [0x155, 2, -9] are freed." When reconstructing the internal log state from the log items found on disk, it's required that deferred ops replay in exactly the same order that they would have had the filesystem not gone down. However, replaying unfinished deferred ops can create /more/ deferred ops. These new deferred ops are finished in the wrong order. This causes fs corruption and replay crashes, so let's create a single defer_ops to handle the subsequent ops created during replay, then use one single transaction at the end of log recovery to ensure that everything is replayed in the same order as they're supposed to be. Reported-by: Amir Goldstein <amir73il@gmail.com> Analyzed-by: Hou Tao <houtao1@huawei.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Tested-by: Amir Goldstein <amir73il@gmail.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-11-22 12:53:02 +08:00
/*
* NOTE: If your intent processing routine can create more
xfs: proper replay of deferred ops queued during log recovery When we replay unfinished intent items that have been recovered from the log, it's possible that the replay will cause the creation of more deferred work items. As outlined in commit 509955823cc9c ("xfs: log recovery should replay deferred ops in order"), later work items have an implicit ordering dependency on earlier work items. Therefore, recovery must replay the items (both recovered and created) in the same order that they would have been during normal operation. For log recovery, we enforce this ordering by using an empty transaction to collect deferred ops that get created in the process of recovering a log intent item to prevent them from being committed before the rest of the recovered intent items. After we finish committing all the recovered log items, we allocate a transaction with an enormous block reservation, splice our huge list of created deferred ops into that transaction, and commit it, thereby finishing all those ops. This is /really/ hokey -- it's the one place in XFS where we allow nested transactions; the splicing of the defer ops list is is inelegant and has to be done twice per recovery function; and the broken way we handle inode pointers and block reservations cause subtle use-after-free and allocator problems that will be fixed by this patch and the two patches after it. Therefore, replace the hokey empty transaction with a structure designed to capture each chain of deferred ops that are created as part of recovering a single unfinished log intent. Finally, refactor the loop that replays those chains to do so using one transaction per chain. Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de>
2020-09-26 08:39:37 +08:00
* deferred ops, you /must/ attach them to the capture list in
* the recover routine or else those subsequent intents will be
xfs: log recovery should replay deferred ops in order As part of testing log recovery with dm_log_writes, Amir Goldstein discovered an error in the deferred ops recovery that lead to corruption of the filesystem metadata if a reflink+rmap filesystem happened to shut down midway through a CoW remap: "This is what happens [after failed log recovery]: "Phase 1 - find and verify superblock... "Phase 2 - using internal log " - zero log... " - scan filesystem freespace and inode maps... " - found root inode chunk "Phase 3 - for each AG... " - scan (but don't clear) agi unlinked lists... " - process known inodes and perform inode discovery... " - agno = 0 "data fork in regular inode 134 claims CoW block 376 "correcting nextents for inode 134 "bad data fork in inode 134 "would have cleared inode 134" Hou Tao dissected the log contents of exactly such a crash: "According to the implementation of xfs_defer_finish(), these ops should be completed in the following sequence: "Have been done: "(1) CUI: Oper (160) "(2) BUI: Oper (161) "(3) CUD: Oper (194), for CUI Oper (160) "(4) RUI A: Oper (197), free rmap [0x155, 2, -9] "Should be done: "(5) BUD: for BUI Oper (161) "(6) RUI B: add rmap [0x155, 2, 137] "(7) RUD: for RUI A "(8) RUD: for RUI B "Actually be done by xlog_recover_process_intents() "(5) BUD: for BUI Oper (161) "(6) RUI B: add rmap [0x155, 2, 137] "(7) RUD: for RUI B "(8) RUD: for RUI A "So the rmap entry [0x155, 2, -9] for COW should be freed firstly, then a new rmap entry [0x155, 2, 137] will be added. However, as we can see from the log record in post_mount.log (generated after umount) and the trace print, the new rmap entry [0x155, 2, 137] are added firstly, then the rmap entry [0x155, 2, -9] are freed." When reconstructing the internal log state from the log items found on disk, it's required that deferred ops replay in exactly the same order that they would have had the filesystem not gone down. However, replaying unfinished deferred ops can create /more/ deferred ops. These new deferred ops are finished in the wrong order. This causes fs corruption and replay crashes, so let's create a single defer_ops to handle the subsequent ops created during replay, then use one single transaction at the end of log recovery to ensure that everything is replayed in the same order as they're supposed to be. Reported-by: Amir Goldstein <amir73il@gmail.com> Analyzed-by: Hou Tao <houtao1@huawei.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Tested-by: Amir Goldstein <amir73il@gmail.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-11-22 12:53:02 +08:00
* replayed in the wrong order!
xfs: avoid a UAF when log intent item recovery fails KASAN reported a UAF bug when I was running xfs/235: BUG: KASAN: use-after-free in xlog_recover_process_intents+0xa77/0xae0 [xfs] Read of size 8 at addr ffff88804391b360 by task mount/5680 CPU: 2 PID: 5680 Comm: mount Not tainted 6.0.0-xfsx #6.0.0 77e7b52a4943a975441e5ac90a5ad7748b7867f6 Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 1.15.0-1 04/01/2014 Call Trace: <TASK> dump_stack_lvl+0x34/0x44 print_report.cold+0x2cc/0x682 kasan_report+0xa3/0x120 xlog_recover_process_intents+0xa77/0xae0 [xfs fb841c7180aad3f8359438576e27867f5795667e] xlog_recover_finish+0x7d/0x970 [xfs fb841c7180aad3f8359438576e27867f5795667e] xfs_log_mount_finish+0x2d7/0x5d0 [xfs fb841c7180aad3f8359438576e27867f5795667e] xfs_mountfs+0x11d4/0x1d10 [xfs fb841c7180aad3f8359438576e27867f5795667e] xfs_fs_fill_super+0x13d5/0x1a80 [xfs fb841c7180aad3f8359438576e27867f5795667e] get_tree_bdev+0x3da/0x6e0 vfs_get_tree+0x7d/0x240 path_mount+0xdd3/0x17d0 __x64_sys_mount+0x1fa/0x270 do_syscall_64+0x2b/0x80 entry_SYSCALL_64_after_hwframe+0x46/0xb0 RIP: 0033:0x7ff5bc069eae Code: 48 8b 0d 85 1f 0f 00 f7 d8 64 89 01 48 83 c8 ff c3 66 2e 0f 1f 84 00 00 00 00 00 90 f3 0f 1e fa 49 89 ca b8 a5 00 00 00 0f 05 <48> 3d 01 f0 ff ff 73 01 c3 48 8b 0d 52 1f 0f 00 f7 d8 64 89 01 48 RSP: 002b:00007ffe433fd448 EFLAGS: 00000246 ORIG_RAX: 00000000000000a5 RAX: ffffffffffffffda RBX: 0000000000000000 RCX: 00007ff5bc069eae RDX: 00005575d7213290 RSI: 00005575d72132d0 RDI: 00005575d72132b0 RBP: 00005575d7212fd0 R08: 00005575d7213230 R09: 00005575d7213fe0 R10: 0000000000000000 R11: 0000000000000246 R12: 0000000000000000 R13: 00005575d7213290 R14: 00005575d72132b0 R15: 00005575d7212fd0 </TASK> Allocated by task 5680: kasan_save_stack+0x1e/0x40 __kasan_slab_alloc+0x66/0x80 kmem_cache_alloc+0x152/0x320 xfs_rui_init+0x17a/0x1b0 [xfs] xlog_recover_rui_commit_pass2+0xb9/0x2e0 [xfs] xlog_recover_items_pass2+0xe9/0x220 [xfs] xlog_recover_commit_trans+0x673/0x900 [xfs] xlog_recovery_process_trans+0xbe/0x130 [xfs] xlog_recover_process_data+0x103/0x2a0 [xfs] xlog_do_recovery_pass+0x548/0xc60 [xfs] xlog_do_log_recovery+0x62/0xc0 [xfs] xlog_do_recover+0x73/0x480 [xfs] xlog_recover+0x229/0x460 [xfs] xfs_log_mount+0x284/0x640 [xfs] xfs_mountfs+0xf8b/0x1d10 [xfs] xfs_fs_fill_super+0x13d5/0x1a80 [xfs] get_tree_bdev+0x3da/0x6e0 vfs_get_tree+0x7d/0x240 path_mount+0xdd3/0x17d0 __x64_sys_mount+0x1fa/0x270 do_syscall_64+0x2b/0x80 entry_SYSCALL_64_after_hwframe+0x46/0xb0 Freed by task 5680: kasan_save_stack+0x1e/0x40 kasan_set_track+0x21/0x30 kasan_set_free_info+0x20/0x30 ____kasan_slab_free+0x144/0x1b0 slab_free_freelist_hook+0xab/0x180 kmem_cache_free+0x1f1/0x410 xfs_rud_item_release+0x33/0x80 [xfs] xfs_trans_free_items+0xc3/0x220 [xfs] xfs_trans_cancel+0x1fa/0x590 [xfs] xfs_rui_item_recover+0x913/0xd60 [xfs] xlog_recover_process_intents+0x24e/0xae0 [xfs] xlog_recover_finish+0x7d/0x970 [xfs] xfs_log_mount_finish+0x2d7/0x5d0 [xfs] xfs_mountfs+0x11d4/0x1d10 [xfs] xfs_fs_fill_super+0x13d5/0x1a80 [xfs] get_tree_bdev+0x3da/0x6e0 vfs_get_tree+0x7d/0x240 path_mount+0xdd3/0x17d0 __x64_sys_mount+0x1fa/0x270 do_syscall_64+0x2b/0x80 entry_SYSCALL_64_after_hwframe+0x46/0xb0 The buggy address belongs to the object at ffff88804391b300 which belongs to the cache xfs_rui_item of size 688 The buggy address is located 96 bytes inside of 688-byte region [ffff88804391b300, ffff88804391b5b0) The buggy address belongs to the physical page: page:ffffea00010e4600 refcount:1 mapcount:0 mapping:0000000000000000 index:0xffff888043919320 pfn:0x43918 head:ffffea00010e4600 order:2 compound_mapcount:0 compound_pincount:0 flags: 0x4fff80000010200(slab|head|node=1|zone=1|lastcpupid=0xfff) raw: 04fff80000010200 0000000000000000 dead000000000122 ffff88807f0eadc0 raw: ffff888043919320 0000000080140010 00000001ffffffff 0000000000000000 page dumped because: kasan: bad access detected Memory state around the buggy address: ffff88804391b200: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc ffff88804391b280: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc >ffff88804391b300: fa fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb ^ ffff88804391b380: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb ffff88804391b400: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb ================================================================== The test fuzzes an rmap btree block and starts writer threads to induce a filesystem shutdown on the corrupt block. When the filesystem is remounted, recovery will try to replay the committed rmap intent item, but the corruption problem causes the recovery transaction to fail. Cancelling the transaction frees the RUD, which frees the RUI that we recovered. When we return to xlog_recover_process_intents, @lip is now a dangling pointer, and we cannot use it to find the iop_recover method for the tracepoint. Hence we must store the item ops before calling ->iop_recover if we want to give it to the tracepoint so that the trace data will tell us exactly which intent item failed. Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de>
2022-10-17 08:54:40 +08:00
*
* The recovery function can free the log item, so we must not
* access lip after it returns.
xfs: log recovery should replay deferred ops in order As part of testing log recovery with dm_log_writes, Amir Goldstein discovered an error in the deferred ops recovery that lead to corruption of the filesystem metadata if a reflink+rmap filesystem happened to shut down midway through a CoW remap: "This is what happens [after failed log recovery]: "Phase 1 - find and verify superblock... "Phase 2 - using internal log " - zero log... " - scan filesystem freespace and inode maps... " - found root inode chunk "Phase 3 - for each AG... " - scan (but don't clear) agi unlinked lists... " - process known inodes and perform inode discovery... " - agno = 0 "data fork in regular inode 134 claims CoW block 376 "correcting nextents for inode 134 "bad data fork in inode 134 "would have cleared inode 134" Hou Tao dissected the log contents of exactly such a crash: "According to the implementation of xfs_defer_finish(), these ops should be completed in the following sequence: "Have been done: "(1) CUI: Oper (160) "(2) BUI: Oper (161) "(3) CUD: Oper (194), for CUI Oper (160) "(4) RUI A: Oper (197), free rmap [0x155, 2, -9] "Should be done: "(5) BUD: for BUI Oper (161) "(6) RUI B: add rmap [0x155, 2, 137] "(7) RUD: for RUI A "(8) RUD: for RUI B "Actually be done by xlog_recover_process_intents() "(5) BUD: for BUI Oper (161) "(6) RUI B: add rmap [0x155, 2, 137] "(7) RUD: for RUI B "(8) RUD: for RUI A "So the rmap entry [0x155, 2, -9] for COW should be freed firstly, then a new rmap entry [0x155, 2, 137] will be added. However, as we can see from the log record in post_mount.log (generated after umount) and the trace print, the new rmap entry [0x155, 2, 137] are added firstly, then the rmap entry [0x155, 2, -9] are freed." When reconstructing the internal log state from the log items found on disk, it's required that deferred ops replay in exactly the same order that they would have had the filesystem not gone down. However, replaying unfinished deferred ops can create /more/ deferred ops. These new deferred ops are finished in the wrong order. This causes fs corruption and replay crashes, so let's create a single defer_ops to handle the subsequent ops created during replay, then use one single transaction at the end of log recovery to ensure that everything is replayed in the same order as they're supposed to be. Reported-by: Amir Goldstein <amir73il@gmail.com> Analyzed-by: Hou Tao <houtao1@huawei.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Tested-by: Amir Goldstein <amir73il@gmail.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-11-22 12:53:02 +08:00
*/
spin_unlock(&ailp->ail_lock);
xfs: avoid a UAF when log intent item recovery fails KASAN reported a UAF bug when I was running xfs/235: BUG: KASAN: use-after-free in xlog_recover_process_intents+0xa77/0xae0 [xfs] Read of size 8 at addr ffff88804391b360 by task mount/5680 CPU: 2 PID: 5680 Comm: mount Not tainted 6.0.0-xfsx #6.0.0 77e7b52a4943a975441e5ac90a5ad7748b7867f6 Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 1.15.0-1 04/01/2014 Call Trace: <TASK> dump_stack_lvl+0x34/0x44 print_report.cold+0x2cc/0x682 kasan_report+0xa3/0x120 xlog_recover_process_intents+0xa77/0xae0 [xfs fb841c7180aad3f8359438576e27867f5795667e] xlog_recover_finish+0x7d/0x970 [xfs fb841c7180aad3f8359438576e27867f5795667e] xfs_log_mount_finish+0x2d7/0x5d0 [xfs fb841c7180aad3f8359438576e27867f5795667e] xfs_mountfs+0x11d4/0x1d10 [xfs fb841c7180aad3f8359438576e27867f5795667e] xfs_fs_fill_super+0x13d5/0x1a80 [xfs fb841c7180aad3f8359438576e27867f5795667e] get_tree_bdev+0x3da/0x6e0 vfs_get_tree+0x7d/0x240 path_mount+0xdd3/0x17d0 __x64_sys_mount+0x1fa/0x270 do_syscall_64+0x2b/0x80 entry_SYSCALL_64_after_hwframe+0x46/0xb0 RIP: 0033:0x7ff5bc069eae Code: 48 8b 0d 85 1f 0f 00 f7 d8 64 89 01 48 83 c8 ff c3 66 2e 0f 1f 84 00 00 00 00 00 90 f3 0f 1e fa 49 89 ca b8 a5 00 00 00 0f 05 <48> 3d 01 f0 ff ff 73 01 c3 48 8b 0d 52 1f 0f 00 f7 d8 64 89 01 48 RSP: 002b:00007ffe433fd448 EFLAGS: 00000246 ORIG_RAX: 00000000000000a5 RAX: ffffffffffffffda RBX: 0000000000000000 RCX: 00007ff5bc069eae RDX: 00005575d7213290 RSI: 00005575d72132d0 RDI: 00005575d72132b0 RBP: 00005575d7212fd0 R08: 00005575d7213230 R09: 00005575d7213fe0 R10: 0000000000000000 R11: 0000000000000246 R12: 0000000000000000 R13: 00005575d7213290 R14: 00005575d72132b0 R15: 00005575d7212fd0 </TASK> Allocated by task 5680: kasan_save_stack+0x1e/0x40 __kasan_slab_alloc+0x66/0x80 kmem_cache_alloc+0x152/0x320 xfs_rui_init+0x17a/0x1b0 [xfs] xlog_recover_rui_commit_pass2+0xb9/0x2e0 [xfs] xlog_recover_items_pass2+0xe9/0x220 [xfs] xlog_recover_commit_trans+0x673/0x900 [xfs] xlog_recovery_process_trans+0xbe/0x130 [xfs] xlog_recover_process_data+0x103/0x2a0 [xfs] xlog_do_recovery_pass+0x548/0xc60 [xfs] xlog_do_log_recovery+0x62/0xc0 [xfs] xlog_do_recover+0x73/0x480 [xfs] xlog_recover+0x229/0x460 [xfs] xfs_log_mount+0x284/0x640 [xfs] xfs_mountfs+0xf8b/0x1d10 [xfs] xfs_fs_fill_super+0x13d5/0x1a80 [xfs] get_tree_bdev+0x3da/0x6e0 vfs_get_tree+0x7d/0x240 path_mount+0xdd3/0x17d0 __x64_sys_mount+0x1fa/0x270 do_syscall_64+0x2b/0x80 entry_SYSCALL_64_after_hwframe+0x46/0xb0 Freed by task 5680: kasan_save_stack+0x1e/0x40 kasan_set_track+0x21/0x30 kasan_set_free_info+0x20/0x30 ____kasan_slab_free+0x144/0x1b0 slab_free_freelist_hook+0xab/0x180 kmem_cache_free+0x1f1/0x410 xfs_rud_item_release+0x33/0x80 [xfs] xfs_trans_free_items+0xc3/0x220 [xfs] xfs_trans_cancel+0x1fa/0x590 [xfs] xfs_rui_item_recover+0x913/0xd60 [xfs] xlog_recover_process_intents+0x24e/0xae0 [xfs] xlog_recover_finish+0x7d/0x970 [xfs] xfs_log_mount_finish+0x2d7/0x5d0 [xfs] xfs_mountfs+0x11d4/0x1d10 [xfs] xfs_fs_fill_super+0x13d5/0x1a80 [xfs] get_tree_bdev+0x3da/0x6e0 vfs_get_tree+0x7d/0x240 path_mount+0xdd3/0x17d0 __x64_sys_mount+0x1fa/0x270 do_syscall_64+0x2b/0x80 entry_SYSCALL_64_after_hwframe+0x46/0xb0 The buggy address belongs to the object at ffff88804391b300 which belongs to the cache xfs_rui_item of size 688 The buggy address is located 96 bytes inside of 688-byte region [ffff88804391b300, ffff88804391b5b0) The buggy address belongs to the physical page: page:ffffea00010e4600 refcount:1 mapcount:0 mapping:0000000000000000 index:0xffff888043919320 pfn:0x43918 head:ffffea00010e4600 order:2 compound_mapcount:0 compound_pincount:0 flags: 0x4fff80000010200(slab|head|node=1|zone=1|lastcpupid=0xfff) raw: 04fff80000010200 0000000000000000 dead000000000122 ffff88807f0eadc0 raw: ffff888043919320 0000000080140010 00000001ffffffff 0000000000000000 page dumped because: kasan: bad access detected Memory state around the buggy address: ffff88804391b200: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc ffff88804391b280: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc >ffff88804391b300: fa fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb ^ ffff88804391b380: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb ffff88804391b400: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb ================================================================== The test fuzzes an rmap btree block and starts writer threads to induce a filesystem shutdown on the corrupt block. When the filesystem is remounted, recovery will try to replay the committed rmap intent item, but the corruption problem causes the recovery transaction to fail. Cancelling the transaction frees the RUD, which frees the RUI that we recovered. When we return to xlog_recover_process_intents, @lip is now a dangling pointer, and we cannot use it to find the iop_recover method for the tracepoint. Hence we must store the item ops before calling ->iop_recover if we want to give it to the tracepoint so that the trace data will tell us exactly which intent item failed. Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de>
2022-10-17 08:54:40 +08:00
ops = lip->li_ops;
error = ops->iop_recover(lip, &capture_list);
spin_lock(&ailp->ail_lock);
if (error) {
trace_xlog_intent_recovery_failed(log->l_mp, error,
xfs: avoid a UAF when log intent item recovery fails KASAN reported a UAF bug when I was running xfs/235: BUG: KASAN: use-after-free in xlog_recover_process_intents+0xa77/0xae0 [xfs] Read of size 8 at addr ffff88804391b360 by task mount/5680 CPU: 2 PID: 5680 Comm: mount Not tainted 6.0.0-xfsx #6.0.0 77e7b52a4943a975441e5ac90a5ad7748b7867f6 Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 1.15.0-1 04/01/2014 Call Trace: <TASK> dump_stack_lvl+0x34/0x44 print_report.cold+0x2cc/0x682 kasan_report+0xa3/0x120 xlog_recover_process_intents+0xa77/0xae0 [xfs fb841c7180aad3f8359438576e27867f5795667e] xlog_recover_finish+0x7d/0x970 [xfs fb841c7180aad3f8359438576e27867f5795667e] xfs_log_mount_finish+0x2d7/0x5d0 [xfs fb841c7180aad3f8359438576e27867f5795667e] xfs_mountfs+0x11d4/0x1d10 [xfs fb841c7180aad3f8359438576e27867f5795667e] xfs_fs_fill_super+0x13d5/0x1a80 [xfs fb841c7180aad3f8359438576e27867f5795667e] get_tree_bdev+0x3da/0x6e0 vfs_get_tree+0x7d/0x240 path_mount+0xdd3/0x17d0 __x64_sys_mount+0x1fa/0x270 do_syscall_64+0x2b/0x80 entry_SYSCALL_64_after_hwframe+0x46/0xb0 RIP: 0033:0x7ff5bc069eae Code: 48 8b 0d 85 1f 0f 00 f7 d8 64 89 01 48 83 c8 ff c3 66 2e 0f 1f 84 00 00 00 00 00 90 f3 0f 1e fa 49 89 ca b8 a5 00 00 00 0f 05 <48> 3d 01 f0 ff ff 73 01 c3 48 8b 0d 52 1f 0f 00 f7 d8 64 89 01 48 RSP: 002b:00007ffe433fd448 EFLAGS: 00000246 ORIG_RAX: 00000000000000a5 RAX: ffffffffffffffda RBX: 0000000000000000 RCX: 00007ff5bc069eae RDX: 00005575d7213290 RSI: 00005575d72132d0 RDI: 00005575d72132b0 RBP: 00005575d7212fd0 R08: 00005575d7213230 R09: 00005575d7213fe0 R10: 0000000000000000 R11: 0000000000000246 R12: 0000000000000000 R13: 00005575d7213290 R14: 00005575d72132b0 R15: 00005575d7212fd0 </TASK> Allocated by task 5680: kasan_save_stack+0x1e/0x40 __kasan_slab_alloc+0x66/0x80 kmem_cache_alloc+0x152/0x320 xfs_rui_init+0x17a/0x1b0 [xfs] xlog_recover_rui_commit_pass2+0xb9/0x2e0 [xfs] xlog_recover_items_pass2+0xe9/0x220 [xfs] xlog_recover_commit_trans+0x673/0x900 [xfs] xlog_recovery_process_trans+0xbe/0x130 [xfs] xlog_recover_process_data+0x103/0x2a0 [xfs] xlog_do_recovery_pass+0x548/0xc60 [xfs] xlog_do_log_recovery+0x62/0xc0 [xfs] xlog_do_recover+0x73/0x480 [xfs] xlog_recover+0x229/0x460 [xfs] xfs_log_mount+0x284/0x640 [xfs] xfs_mountfs+0xf8b/0x1d10 [xfs] xfs_fs_fill_super+0x13d5/0x1a80 [xfs] get_tree_bdev+0x3da/0x6e0 vfs_get_tree+0x7d/0x240 path_mount+0xdd3/0x17d0 __x64_sys_mount+0x1fa/0x270 do_syscall_64+0x2b/0x80 entry_SYSCALL_64_after_hwframe+0x46/0xb0 Freed by task 5680: kasan_save_stack+0x1e/0x40 kasan_set_track+0x21/0x30 kasan_set_free_info+0x20/0x30 ____kasan_slab_free+0x144/0x1b0 slab_free_freelist_hook+0xab/0x180 kmem_cache_free+0x1f1/0x410 xfs_rud_item_release+0x33/0x80 [xfs] xfs_trans_free_items+0xc3/0x220 [xfs] xfs_trans_cancel+0x1fa/0x590 [xfs] xfs_rui_item_recover+0x913/0xd60 [xfs] xlog_recover_process_intents+0x24e/0xae0 [xfs] xlog_recover_finish+0x7d/0x970 [xfs] xfs_log_mount_finish+0x2d7/0x5d0 [xfs] xfs_mountfs+0x11d4/0x1d10 [xfs] xfs_fs_fill_super+0x13d5/0x1a80 [xfs] get_tree_bdev+0x3da/0x6e0 vfs_get_tree+0x7d/0x240 path_mount+0xdd3/0x17d0 __x64_sys_mount+0x1fa/0x270 do_syscall_64+0x2b/0x80 entry_SYSCALL_64_after_hwframe+0x46/0xb0 The buggy address belongs to the object at ffff88804391b300 which belongs to the cache xfs_rui_item of size 688 The buggy address is located 96 bytes inside of 688-byte region [ffff88804391b300, ffff88804391b5b0) The buggy address belongs to the physical page: page:ffffea00010e4600 refcount:1 mapcount:0 mapping:0000000000000000 index:0xffff888043919320 pfn:0x43918 head:ffffea00010e4600 order:2 compound_mapcount:0 compound_pincount:0 flags: 0x4fff80000010200(slab|head|node=1|zone=1|lastcpupid=0xfff) raw: 04fff80000010200 0000000000000000 dead000000000122 ffff88807f0eadc0 raw: ffff888043919320 0000000080140010 00000001ffffffff 0000000000000000 page dumped because: kasan: bad access detected Memory state around the buggy address: ffff88804391b200: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc ffff88804391b280: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc >ffff88804391b300: fa fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb ^ ffff88804391b380: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb ffff88804391b400: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb ================================================================== The test fuzzes an rmap btree block and starts writer threads to induce a filesystem shutdown on the corrupt block. When the filesystem is remounted, recovery will try to replay the committed rmap intent item, but the corruption problem causes the recovery transaction to fail. Cancelling the transaction frees the RUD, which frees the RUI that we recovered. When we return to xlog_recover_process_intents, @lip is now a dangling pointer, and we cannot use it to find the iop_recover method for the tracepoint. Hence we must store the item ops before calling ->iop_recover if we want to give it to the tracepoint so that the trace data will tell us exactly which intent item failed. Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de>
2022-10-17 08:54:40 +08:00
ops->iop_recover);
xfs: proper replay of deferred ops queued during log recovery When we replay unfinished intent items that have been recovered from the log, it's possible that the replay will cause the creation of more deferred work items. As outlined in commit 509955823cc9c ("xfs: log recovery should replay deferred ops in order"), later work items have an implicit ordering dependency on earlier work items. Therefore, recovery must replay the items (both recovered and created) in the same order that they would have been during normal operation. For log recovery, we enforce this ordering by using an empty transaction to collect deferred ops that get created in the process of recovering a log intent item to prevent them from being committed before the rest of the recovered intent items. After we finish committing all the recovered log items, we allocate a transaction with an enormous block reservation, splice our huge list of created deferred ops into that transaction, and commit it, thereby finishing all those ops. This is /really/ hokey -- it's the one place in XFS where we allow nested transactions; the splicing of the defer ops list is is inelegant and has to be done twice per recovery function; and the broken way we handle inode pointers and block reservations cause subtle use-after-free and allocator problems that will be fixed by this patch and the two patches after it. Therefore, replace the hokey empty transaction with a structure designed to capture each chain of deferred ops that are created as part of recovering a single unfinished log intent. Finally, refactor the loop that replays those chains to do so using one transaction per chain. Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de>
2020-09-26 08:39:37 +08:00
break;
}
}
xfs: proper replay of deferred ops queued during log recovery When we replay unfinished intent items that have been recovered from the log, it's possible that the replay will cause the creation of more deferred work items. As outlined in commit 509955823cc9c ("xfs: log recovery should replay deferred ops in order"), later work items have an implicit ordering dependency on earlier work items. Therefore, recovery must replay the items (both recovered and created) in the same order that they would have been during normal operation. For log recovery, we enforce this ordering by using an empty transaction to collect deferred ops that get created in the process of recovering a log intent item to prevent them from being committed before the rest of the recovered intent items. After we finish committing all the recovered log items, we allocate a transaction with an enormous block reservation, splice our huge list of created deferred ops into that transaction, and commit it, thereby finishing all those ops. This is /really/ hokey -- it's the one place in XFS where we allow nested transactions; the splicing of the defer ops list is is inelegant and has to be done twice per recovery function; and the broken way we handle inode pointers and block reservations cause subtle use-after-free and allocator problems that will be fixed by this patch and the two patches after it. Therefore, replace the hokey empty transaction with a structure designed to capture each chain of deferred ops that are created as part of recovering a single unfinished log intent. Finally, refactor the loop that replays those chains to do so using one transaction per chain. Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de>
2020-09-26 08:39:37 +08:00
xfs_trans_ail_cursor_done(&cur);
spin_unlock(&ailp->ail_lock);
xfs: proper replay of deferred ops queued during log recovery When we replay unfinished intent items that have been recovered from the log, it's possible that the replay will cause the creation of more deferred work items. As outlined in commit 509955823cc9c ("xfs: log recovery should replay deferred ops in order"), later work items have an implicit ordering dependency on earlier work items. Therefore, recovery must replay the items (both recovered and created) in the same order that they would have been during normal operation. For log recovery, we enforce this ordering by using an empty transaction to collect deferred ops that get created in the process of recovering a log intent item to prevent them from being committed before the rest of the recovered intent items. After we finish committing all the recovered log items, we allocate a transaction with an enormous block reservation, splice our huge list of created deferred ops into that transaction, and commit it, thereby finishing all those ops. This is /really/ hokey -- it's the one place in XFS where we allow nested transactions; the splicing of the defer ops list is is inelegant and has to be done twice per recovery function; and the broken way we handle inode pointers and block reservations cause subtle use-after-free and allocator problems that will be fixed by this patch and the two patches after it. Therefore, replace the hokey empty transaction with a structure designed to capture each chain of deferred ops that are created as part of recovering a single unfinished log intent. Finally, refactor the loop that replays those chains to do so using one transaction per chain. Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de>
2020-09-26 08:39:37 +08:00
if (error)
goto err;
xfs: log recovery should replay deferred ops in order As part of testing log recovery with dm_log_writes, Amir Goldstein discovered an error in the deferred ops recovery that lead to corruption of the filesystem metadata if a reflink+rmap filesystem happened to shut down midway through a CoW remap: "This is what happens [after failed log recovery]: "Phase 1 - find and verify superblock... "Phase 2 - using internal log " - zero log... " - scan filesystem freespace and inode maps... " - found root inode chunk "Phase 3 - for each AG... " - scan (but don't clear) agi unlinked lists... " - process known inodes and perform inode discovery... " - agno = 0 "data fork in regular inode 134 claims CoW block 376 "correcting nextents for inode 134 "bad data fork in inode 134 "would have cleared inode 134" Hou Tao dissected the log contents of exactly such a crash: "According to the implementation of xfs_defer_finish(), these ops should be completed in the following sequence: "Have been done: "(1) CUI: Oper (160) "(2) BUI: Oper (161) "(3) CUD: Oper (194), for CUI Oper (160) "(4) RUI A: Oper (197), free rmap [0x155, 2, -9] "Should be done: "(5) BUD: for BUI Oper (161) "(6) RUI B: add rmap [0x155, 2, 137] "(7) RUD: for RUI A "(8) RUD: for RUI B "Actually be done by xlog_recover_process_intents() "(5) BUD: for BUI Oper (161) "(6) RUI B: add rmap [0x155, 2, 137] "(7) RUD: for RUI B "(8) RUD: for RUI A "So the rmap entry [0x155, 2, -9] for COW should be freed firstly, then a new rmap entry [0x155, 2, 137] will be added. However, as we can see from the log record in post_mount.log (generated after umount) and the trace print, the new rmap entry [0x155, 2, 137] are added firstly, then the rmap entry [0x155, 2, -9] are freed." When reconstructing the internal log state from the log items found on disk, it's required that deferred ops replay in exactly the same order that they would have had the filesystem not gone down. However, replaying unfinished deferred ops can create /more/ deferred ops. These new deferred ops are finished in the wrong order. This causes fs corruption and replay crashes, so let's create a single defer_ops to handle the subsequent ops created during replay, then use one single transaction at the end of log recovery to ensure that everything is replayed in the same order as they're supposed to be. Reported-by: Amir Goldstein <amir73il@gmail.com> Analyzed-by: Hou Tao <houtao1@huawei.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Tested-by: Amir Goldstein <amir73il@gmail.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-11-22 12:53:02 +08:00
xfs: proper replay of deferred ops queued during log recovery When we replay unfinished intent items that have been recovered from the log, it's possible that the replay will cause the creation of more deferred work items. As outlined in commit 509955823cc9c ("xfs: log recovery should replay deferred ops in order"), later work items have an implicit ordering dependency on earlier work items. Therefore, recovery must replay the items (both recovered and created) in the same order that they would have been during normal operation. For log recovery, we enforce this ordering by using an empty transaction to collect deferred ops that get created in the process of recovering a log intent item to prevent them from being committed before the rest of the recovered intent items. After we finish committing all the recovered log items, we allocate a transaction with an enormous block reservation, splice our huge list of created deferred ops into that transaction, and commit it, thereby finishing all those ops. This is /really/ hokey -- it's the one place in XFS where we allow nested transactions; the splicing of the defer ops list is is inelegant and has to be done twice per recovery function; and the broken way we handle inode pointers and block reservations cause subtle use-after-free and allocator problems that will be fixed by this patch and the two patches after it. Therefore, replace the hokey empty transaction with a structure designed to capture each chain of deferred ops that are created as part of recovering a single unfinished log intent. Finally, refactor the loop that replays those chains to do so using one transaction per chain. Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de>
2020-09-26 08:39:37 +08:00
error = xlog_finish_defer_ops(log->l_mp, &capture_list);
if (error)
goto err;
return 0;
err:
xlog_abort_defer_ops(log->l_mp, &capture_list);
return error;
}
/*
xfs: shutdown in intent recovery has non-intent items in the AIL generic/388 triggered a failure in RUI recovery due to a corrupted btree record and the system then locked up hard due to a subsequent assert failure while holding a spinlock cancelling intents: XFS (pmem1): Corruption of in-memory data (0x8) detected at xfs_do_force_shutdown+0x1a/0x20 (fs/xfs/xfs_trans.c:964). Shutting down filesystem. XFS (pmem1): Please unmount the filesystem and rectify the problem(s) XFS: Assertion failed: !xlog_item_is_intent(lip), file: fs/xfs/xfs_log_recover.c, line: 2632 Call Trace: <TASK> xlog_recover_cancel_intents.isra.0+0xd1/0x120 xlog_recover_finish+0xb9/0x110 xfs_log_mount_finish+0x15a/0x1e0 xfs_mountfs+0x540/0x910 xfs_fs_fill_super+0x476/0x830 get_tree_bdev+0x171/0x270 ? xfs_init_fs_context+0x1e0/0x1e0 xfs_fs_get_tree+0x15/0x20 vfs_get_tree+0x24/0xc0 path_mount+0x304/0xba0 ? putname+0x55/0x60 __x64_sys_mount+0x108/0x140 do_syscall_64+0x35/0x80 entry_SYSCALL_64_after_hwframe+0x44/0xae Essentially, there's dirty metadata in the AIL from intent recovery transactions, so when we go to cancel the remaining intents we assume that all objects after the first non-intent log item in the AIL are not intents. This is not true. Intent recovery can log new intents to continue the operations the original intent could not complete in a single transaction. The new intents are committed before they are deferred, which means if the CIL commits in the background they will get inserted into the AIL at the head. Hence if we shut down the filesystem while processing intent recovery, the AIL may have new intents active at the current head. Hence this check: /* * We're done when we see something other than an intent. * There should be no intents left in the AIL now. */ if (!xlog_item_is_intent(lip)) { #ifdef DEBUG for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur)) ASSERT(!xlog_item_is_intent(lip)); #endif break; } in both xlog_recover_process_intents() and log_recover_cancel_intents() is simply not valid. It was valid back when we only had EFI/EFD intents and didn't chain intents, but it hasn't been valid ever since intent recovery could create and commit new intents. Given that crashing the mount task like this pretty much prevents diagnosing what went wrong that lead to the initial failure that triggered intent cancellation, just remove the checks altogether. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2022-03-30 09:22:00 +08:00
* A cancel occurs when the mount has failed and we're bailing out. Release all
* pending log intent items that we haven't started recovery on so they don't
* pin the AIL.
*/
STATIC void
xlog_recover_cancel_intents(
struct xlog *log)
{
struct xfs_log_item *lip;
struct xfs_ail_cursor cur;
struct xfs_ail *ailp;
ailp = log->l_ailp;
spin_lock(&ailp->ail_lock);
lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
while (lip != NULL) {
xfs: shutdown in intent recovery has non-intent items in the AIL generic/388 triggered a failure in RUI recovery due to a corrupted btree record and the system then locked up hard due to a subsequent assert failure while holding a spinlock cancelling intents: XFS (pmem1): Corruption of in-memory data (0x8) detected at xfs_do_force_shutdown+0x1a/0x20 (fs/xfs/xfs_trans.c:964). Shutting down filesystem. XFS (pmem1): Please unmount the filesystem and rectify the problem(s) XFS: Assertion failed: !xlog_item_is_intent(lip), file: fs/xfs/xfs_log_recover.c, line: 2632 Call Trace: <TASK> xlog_recover_cancel_intents.isra.0+0xd1/0x120 xlog_recover_finish+0xb9/0x110 xfs_log_mount_finish+0x15a/0x1e0 xfs_mountfs+0x540/0x910 xfs_fs_fill_super+0x476/0x830 get_tree_bdev+0x171/0x270 ? xfs_init_fs_context+0x1e0/0x1e0 xfs_fs_get_tree+0x15/0x20 vfs_get_tree+0x24/0xc0 path_mount+0x304/0xba0 ? putname+0x55/0x60 __x64_sys_mount+0x108/0x140 do_syscall_64+0x35/0x80 entry_SYSCALL_64_after_hwframe+0x44/0xae Essentially, there's dirty metadata in the AIL from intent recovery transactions, so when we go to cancel the remaining intents we assume that all objects after the first non-intent log item in the AIL are not intents. This is not true. Intent recovery can log new intents to continue the operations the original intent could not complete in a single transaction. The new intents are committed before they are deferred, which means if the CIL commits in the background they will get inserted into the AIL at the head. Hence if we shut down the filesystem while processing intent recovery, the AIL may have new intents active at the current head. Hence this check: /* * We're done when we see something other than an intent. * There should be no intents left in the AIL now. */ if (!xlog_item_is_intent(lip)) { #ifdef DEBUG for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur)) ASSERT(!xlog_item_is_intent(lip)); #endif break; } in both xlog_recover_process_intents() and log_recover_cancel_intents() is simply not valid. It was valid back when we only had EFI/EFD intents and didn't chain intents, but it hasn't been valid ever since intent recovery could create and commit new intents. Given that crashing the mount task like this pretty much prevents diagnosing what went wrong that lead to the initial failure that triggered intent cancellation, just remove the checks altogether. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2022-03-30 09:22:00 +08:00
if (!xlog_item_is_intent(lip))
break;
spin_unlock(&ailp->ail_lock);
lip->li_ops->iop_release(lip);
spin_lock(&ailp->ail_lock);
lip = xfs_trans_ail_cursor_next(ailp, &cur);
}
xfs_trans_ail_cursor_done(&cur);
spin_unlock(&ailp->ail_lock);
}
/*
* This routine performs a transaction to null out a bad inode pointer
* in an agi unlinked inode hash bucket.
*/
STATIC void
xlog_recover_clear_agi_bucket(
struct xfs_perag *pag,
int bucket)
{
struct xfs_mount *mp = pag->pag_mount;
struct xfs_trans *tp;
struct xfs_agi *agi;
struct xfs_buf *agibp;
int offset;
int error;
error = xfs_trans_alloc(mp, &M_RES(mp)->tr_clearagi, 0, 0, 0, &tp);
if (error)
goto out_error;
error = xfs_read_agi(pag, tp, &agibp);
if (error)
goto out_abort;
agi = agibp->b_addr;
agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
offset = offsetof(xfs_agi_t, agi_unlinked) +
(sizeof(xfs_agino_t) * bucket);
xfs_trans_log_buf(tp, agibp, offset,
(offset + sizeof(xfs_agino_t) - 1));
error = xfs_trans_commit(tp);
if (error)
goto out_error;
return;
out_abort:
xfs_trans_cancel(tp);
out_error:
xfs_warn(mp, "%s: failed to clear agi %d. Continuing.", __func__,
pag->pag_agno);
return;
}
static int
xlog_recover_iunlink_bucket(
struct xfs_perag *pag,
struct xfs_agi *agi,
int bucket)
{
struct xfs_mount *mp = pag->pag_mount;
xfs: double link the unlinked inode list Now we have forwards traversal via the incore inode in place, we now need to add back pointers to the incore inode to entirely replace the back reference cache. We use the same lookup semantics and constraints as for the forwards pointer lookups during unlinks, and so we can look up any inode in the unlinked list directly and update the list pointers, forwards or backwards, at any time. The only wrinkle in converting the unlinked list manipulations to use in-core previous pointers is that log recovery doesn't have the incore inode state built up so it can't just read in an inode and release it to finish off the unlink. Hence we need to modify the traversal in recovery to read one inode ahead before we release the inode at the head of the list. This populates the next->prev relationship sufficient to be able to replay the unlinked list and hence greatly simplify the runtime code. This recovery algorithm also requires that we actually remove inodes from the unlinked list one at a time as background inode inactivation will result in unlinked list removal racing with the building of the in-memory unlinked list state. We could serialise this by holding the AGI buffer lock when constructing the in memory state, but all that does is lockstep background processing with list building. It is much simpler to flush the inodegc immediately after releasing the inode so that it is unlinked immediately and there is no races present at all. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de>
2022-07-14 09:46:43 +08:00
struct xfs_inode *prev_ip = NULL;
struct xfs_inode *ip;
xfs: double link the unlinked inode list Now we have forwards traversal via the incore inode in place, we now need to add back pointers to the incore inode to entirely replace the back reference cache. We use the same lookup semantics and constraints as for the forwards pointer lookups during unlinks, and so we can look up any inode in the unlinked list directly and update the list pointers, forwards or backwards, at any time. The only wrinkle in converting the unlinked list manipulations to use in-core previous pointers is that log recovery doesn't have the incore inode state built up so it can't just read in an inode and release it to finish off the unlink. Hence we need to modify the traversal in recovery to read one inode ahead before we release the inode at the head of the list. This populates the next->prev relationship sufficient to be able to replay the unlinked list and hence greatly simplify the runtime code. This recovery algorithm also requires that we actually remove inodes from the unlinked list one at a time as background inode inactivation will result in unlinked list removal racing with the building of the in-memory unlinked list state. We could serialise this by holding the AGI buffer lock when constructing the in memory state, but all that does is lockstep background processing with list building. It is much simpler to flush the inodegc immediately after releasing the inode so that it is unlinked immediately and there is no races present at all. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de>
2022-07-14 09:46:43 +08:00
xfs_agino_t prev_agino, agino;
int error = 0;
agino = be32_to_cpu(agi->agi_unlinked[bucket]);
while (agino != NULLAGINO) {
error = xfs_iget(mp, NULL,
XFS_AGINO_TO_INO(mp, pag->pag_agno, agino),
0, 0, &ip);
if (error)
xfs: double link the unlinked inode list Now we have forwards traversal via the incore inode in place, we now need to add back pointers to the incore inode to entirely replace the back reference cache. We use the same lookup semantics and constraints as for the forwards pointer lookups during unlinks, and so we can look up any inode in the unlinked list directly and update the list pointers, forwards or backwards, at any time. The only wrinkle in converting the unlinked list manipulations to use in-core previous pointers is that log recovery doesn't have the incore inode state built up so it can't just read in an inode and release it to finish off the unlink. Hence we need to modify the traversal in recovery to read one inode ahead before we release the inode at the head of the list. This populates the next->prev relationship sufficient to be able to replay the unlinked list and hence greatly simplify the runtime code. This recovery algorithm also requires that we actually remove inodes from the unlinked list one at a time as background inode inactivation will result in unlinked list removal racing with the building of the in-memory unlinked list state. We could serialise this by holding the AGI buffer lock when constructing the in memory state, but all that does is lockstep background processing with list building. It is much simpler to flush the inodegc immediately after releasing the inode so that it is unlinked immediately and there is no races present at all. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de>
2022-07-14 09:46:43 +08:00
break;
ASSERT(VFS_I(ip)->i_nlink == 0);
ASSERT(VFS_I(ip)->i_mode != 0);
xfs_iflags_clear(ip, XFS_IRECOVERY);
agino = ip->i_next_unlinked;
xfs: double link the unlinked inode list Now we have forwards traversal via the incore inode in place, we now need to add back pointers to the incore inode to entirely replace the back reference cache. We use the same lookup semantics and constraints as for the forwards pointer lookups during unlinks, and so we can look up any inode in the unlinked list directly and update the list pointers, forwards or backwards, at any time. The only wrinkle in converting the unlinked list manipulations to use in-core previous pointers is that log recovery doesn't have the incore inode state built up so it can't just read in an inode and release it to finish off the unlink. Hence we need to modify the traversal in recovery to read one inode ahead before we release the inode at the head of the list. This populates the next->prev relationship sufficient to be able to replay the unlinked list and hence greatly simplify the runtime code. This recovery algorithm also requires that we actually remove inodes from the unlinked list one at a time as background inode inactivation will result in unlinked list removal racing with the building of the in-memory unlinked list state. We could serialise this by holding the AGI buffer lock when constructing the in memory state, but all that does is lockstep background processing with list building. It is much simpler to flush the inodegc immediately after releasing the inode so that it is unlinked immediately and there is no races present at all. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de>
2022-07-14 09:46:43 +08:00
if (prev_ip) {
ip->i_prev_unlinked = prev_agino;
xfs_irele(prev_ip);
xfs: double link the unlinked inode list Now we have forwards traversal via the incore inode in place, we now need to add back pointers to the incore inode to entirely replace the back reference cache. We use the same lookup semantics and constraints as for the forwards pointer lookups during unlinks, and so we can look up any inode in the unlinked list directly and update the list pointers, forwards or backwards, at any time. The only wrinkle in converting the unlinked list manipulations to use in-core previous pointers is that log recovery doesn't have the incore inode state built up so it can't just read in an inode and release it to finish off the unlink. Hence we need to modify the traversal in recovery to read one inode ahead before we release the inode at the head of the list. This populates the next->prev relationship sufficient to be able to replay the unlinked list and hence greatly simplify the runtime code. This recovery algorithm also requires that we actually remove inodes from the unlinked list one at a time as background inode inactivation will result in unlinked list removal racing with the building of the in-memory unlinked list state. We could serialise this by holding the AGI buffer lock when constructing the in memory state, but all that does is lockstep background processing with list building. It is much simpler to flush the inodegc immediately after releasing the inode so that it is unlinked immediately and there is no races present at all. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de>
2022-07-14 09:46:43 +08:00
/*
* Ensure the inode is removed from the unlinked list
* before we continue so that it won't race with
* building the in-memory list here. This could be
* serialised with the agibp lock, but that just
* serialises via lockstepping and it's much simpler
* just to flush the inodegc queue and wait for it to
* complete.
*/
xfs: collect errors from inodegc for unlinked inode recovery Unlinked list recovery requires errors removing the inode the from the unlinked list get fed back to the main recovery loop. Now that we offload the unlinking to the inodegc work, we don't get errors being fed back when we trip over a corruption that prevents the inode from being removed from the unlinked list. This means we never clear the corrupt unlinked list bucket, resulting in runtime operations eventually tripping over it and shutting down. Fix this by collecting inodegc worker errors and feed them back to the flush caller. This is largely best effort - the only context that really cares is log recovery, and it only flushes a single inode at a time so we don't need complex synchronised handling. Essentially the inodegc workers will capture the first error that occurs and the next flush will gather them and clear them. The flush itself will only report the first gathered error. In the cases where callers can return errors, propagate the collected inodegc flush error up the error handling chain. In the case of inode unlinked list recovery, there are several superfluous calls to flush queued unlinked inodes - xlog_recover_iunlink_bucket() guarantees that it has flushed the inodegc and collected errors before it returns. Hence nothing in the calling path needs to run a flush, even when an error is returned. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Dave Chinner <david@fromorbit.com>
2023-06-05 12:48:15 +08:00
error = xfs_inodegc_flush(mp);
if (error)
break;
xfs: double link the unlinked inode list Now we have forwards traversal via the incore inode in place, we now need to add back pointers to the incore inode to entirely replace the back reference cache. We use the same lookup semantics and constraints as for the forwards pointer lookups during unlinks, and so we can look up any inode in the unlinked list directly and update the list pointers, forwards or backwards, at any time. The only wrinkle in converting the unlinked list manipulations to use in-core previous pointers is that log recovery doesn't have the incore inode state built up so it can't just read in an inode and release it to finish off the unlink. Hence we need to modify the traversal in recovery to read one inode ahead before we release the inode at the head of the list. This populates the next->prev relationship sufficient to be able to replay the unlinked list and hence greatly simplify the runtime code. This recovery algorithm also requires that we actually remove inodes from the unlinked list one at a time as background inode inactivation will result in unlinked list removal racing with the building of the in-memory unlinked list state. We could serialise this by holding the AGI buffer lock when constructing the in memory state, but all that does is lockstep background processing with list building. It is much simpler to flush the inodegc immediately after releasing the inode so that it is unlinked immediately and there is no races present at all. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de>
2022-07-14 09:46:43 +08:00
}
xfs: double link the unlinked inode list Now we have forwards traversal via the incore inode in place, we now need to add back pointers to the incore inode to entirely replace the back reference cache. We use the same lookup semantics and constraints as for the forwards pointer lookups during unlinks, and so we can look up any inode in the unlinked list directly and update the list pointers, forwards or backwards, at any time. The only wrinkle in converting the unlinked list manipulations to use in-core previous pointers is that log recovery doesn't have the incore inode state built up so it can't just read in an inode and release it to finish off the unlink. Hence we need to modify the traversal in recovery to read one inode ahead before we release the inode at the head of the list. This populates the next->prev relationship sufficient to be able to replay the unlinked list and hence greatly simplify the runtime code. This recovery algorithm also requires that we actually remove inodes from the unlinked list one at a time as background inode inactivation will result in unlinked list removal racing with the building of the in-memory unlinked list state. We could serialise this by holding the AGI buffer lock when constructing the in memory state, but all that does is lockstep background processing with list building. It is much simpler to flush the inodegc immediately after releasing the inode so that it is unlinked immediately and there is no races present at all. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de>
2022-07-14 09:46:43 +08:00
prev_agino = agino;
prev_ip = ip;
}
xfs: double link the unlinked inode list Now we have forwards traversal via the incore inode in place, we now need to add back pointers to the incore inode to entirely replace the back reference cache. We use the same lookup semantics and constraints as for the forwards pointer lookups during unlinks, and so we can look up any inode in the unlinked list directly and update the list pointers, forwards or backwards, at any time. The only wrinkle in converting the unlinked list manipulations to use in-core previous pointers is that log recovery doesn't have the incore inode state built up so it can't just read in an inode and release it to finish off the unlink. Hence we need to modify the traversal in recovery to read one inode ahead before we release the inode at the head of the list. This populates the next->prev relationship sufficient to be able to replay the unlinked list and hence greatly simplify the runtime code. This recovery algorithm also requires that we actually remove inodes from the unlinked list one at a time as background inode inactivation will result in unlinked list removal racing with the building of the in-memory unlinked list state. We could serialise this by holding the AGI buffer lock when constructing the in memory state, but all that does is lockstep background processing with list building. It is much simpler to flush the inodegc immediately after releasing the inode so that it is unlinked immediately and there is no races present at all. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de>
2022-07-14 09:46:43 +08:00
if (prev_ip) {
xfs: collect errors from inodegc for unlinked inode recovery Unlinked list recovery requires errors removing the inode the from the unlinked list get fed back to the main recovery loop. Now that we offload the unlinking to the inodegc work, we don't get errors being fed back when we trip over a corruption that prevents the inode from being removed from the unlinked list. This means we never clear the corrupt unlinked list bucket, resulting in runtime operations eventually tripping over it and shutting down. Fix this by collecting inodegc worker errors and feed them back to the flush caller. This is largely best effort - the only context that really cares is log recovery, and it only flushes a single inode at a time so we don't need complex synchronised handling. Essentially the inodegc workers will capture the first error that occurs and the next flush will gather them and clear them. The flush itself will only report the first gathered error. In the cases where callers can return errors, propagate the collected inodegc flush error up the error handling chain. In the case of inode unlinked list recovery, there are several superfluous calls to flush queued unlinked inodes - xlog_recover_iunlink_bucket() guarantees that it has flushed the inodegc and collected errors before it returns. Hence nothing in the calling path needs to run a flush, even when an error is returned. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Dave Chinner <david@fromorbit.com>
2023-06-05 12:48:15 +08:00
int error2;
xfs: double link the unlinked inode list Now we have forwards traversal via the incore inode in place, we now need to add back pointers to the incore inode to entirely replace the back reference cache. We use the same lookup semantics and constraints as for the forwards pointer lookups during unlinks, and so we can look up any inode in the unlinked list directly and update the list pointers, forwards or backwards, at any time. The only wrinkle in converting the unlinked list manipulations to use in-core previous pointers is that log recovery doesn't have the incore inode state built up so it can't just read in an inode and release it to finish off the unlink. Hence we need to modify the traversal in recovery to read one inode ahead before we release the inode at the head of the list. This populates the next->prev relationship sufficient to be able to replay the unlinked list and hence greatly simplify the runtime code. This recovery algorithm also requires that we actually remove inodes from the unlinked list one at a time as background inode inactivation will result in unlinked list removal racing with the building of the in-memory unlinked list state. We could serialise this by holding the AGI buffer lock when constructing the in memory state, but all that does is lockstep background processing with list building. It is much simpler to flush the inodegc immediately after releasing the inode so that it is unlinked immediately and there is no races present at all. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de>
2022-07-14 09:46:43 +08:00
ip->i_prev_unlinked = prev_agino;
xfs_irele(prev_ip);
xfs: collect errors from inodegc for unlinked inode recovery Unlinked list recovery requires errors removing the inode the from the unlinked list get fed back to the main recovery loop. Now that we offload the unlinking to the inodegc work, we don't get errors being fed back when we trip over a corruption that prevents the inode from being removed from the unlinked list. This means we never clear the corrupt unlinked list bucket, resulting in runtime operations eventually tripping over it and shutting down. Fix this by collecting inodegc worker errors and feed them back to the flush caller. This is largely best effort - the only context that really cares is log recovery, and it only flushes a single inode at a time so we don't need complex synchronised handling. Essentially the inodegc workers will capture the first error that occurs and the next flush will gather them and clear them. The flush itself will only report the first gathered error. In the cases where callers can return errors, propagate the collected inodegc flush error up the error handling chain. In the case of inode unlinked list recovery, there are several superfluous calls to flush queued unlinked inodes - xlog_recover_iunlink_bucket() guarantees that it has flushed the inodegc and collected errors before it returns. Hence nothing in the calling path needs to run a flush, even when an error is returned. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Dave Chinner <david@fromorbit.com>
2023-06-05 12:48:15 +08:00
error2 = xfs_inodegc_flush(mp);
if (error2 && !error)
return error2;
xfs: double link the unlinked inode list Now we have forwards traversal via the incore inode in place, we now need to add back pointers to the incore inode to entirely replace the back reference cache. We use the same lookup semantics and constraints as for the forwards pointer lookups during unlinks, and so we can look up any inode in the unlinked list directly and update the list pointers, forwards or backwards, at any time. The only wrinkle in converting the unlinked list manipulations to use in-core previous pointers is that log recovery doesn't have the incore inode state built up so it can't just read in an inode and release it to finish off the unlink. Hence we need to modify the traversal in recovery to read one inode ahead before we release the inode at the head of the list. This populates the next->prev relationship sufficient to be able to replay the unlinked list and hence greatly simplify the runtime code. This recovery algorithm also requires that we actually remove inodes from the unlinked list one at a time as background inode inactivation will result in unlinked list removal racing with the building of the in-memory unlinked list state. We could serialise this by holding the AGI buffer lock when constructing the in memory state, but all that does is lockstep background processing with list building. It is much simpler to flush the inodegc immediately after releasing the inode so that it is unlinked immediately and there is no races present at all. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de>
2022-07-14 09:46:43 +08:00
}
return error;
}
/*
xfs: prevent CIL push holdoff in log recovery generic/530 on a machine with enough ram and a non-preemptible kernel can run the AGI processing phase of log recovery enitrely out of cache. This means it never blocks on locks, never waits for IO and runs entirely through the unlinked lists until it either completes or blocks and hangs because it has run out of log space. It runs out of log space because the background CIL push is scheduled but never runs. queue_work() queues the CIL work on the current CPU that is busy, and the workqueue code will not run it on any other CPU. Hence if the unlinked list processing never yields the CPU voluntarily, the push work is delayed indefinitely. This results in the CIL aggregating changes until all the log space is consumed. When the log recoveyr processing evenutally blocks, the CIL flushes but because the last iclog isn't submitted for IO because it isn't full, the CIL flush never completes and nothing ever moves the log head forwards, or indeed inserts anything into the tail of the log, and hence nothing is able to get the log moving again and recovery hangs. There are several problems here, but the two obvious ones from the trace are that: a) log recovery does not yield the CPU for over 4 seconds, b) binding CIL pushes to a single CPU is a really bad idea. This patch addresses just these two aspects of the problem, and are suitable for backporting to work around any issues in older kernels. The more fundamental problem of preventing the CIL from consuming more than 50% of the log without committing will take more invasive and complex work, so will be done as followup work. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2019-09-06 12:35:39 +08:00
* Recover AGI unlinked lists
*
xfs: prevent CIL push holdoff in log recovery generic/530 on a machine with enough ram and a non-preemptible kernel can run the AGI processing phase of log recovery enitrely out of cache. This means it never blocks on locks, never waits for IO and runs entirely through the unlinked lists until it either completes or blocks and hangs because it has run out of log space. It runs out of log space because the background CIL push is scheduled but never runs. queue_work() queues the CIL work on the current CPU that is busy, and the workqueue code will not run it on any other CPU. Hence if the unlinked list processing never yields the CPU voluntarily, the push work is delayed indefinitely. This results in the CIL aggregating changes until all the log space is consumed. When the log recoveyr processing evenutally blocks, the CIL flushes but because the last iclog isn't submitted for IO because it isn't full, the CIL flush never completes and nothing ever moves the log head forwards, or indeed inserts anything into the tail of the log, and hence nothing is able to get the log moving again and recovery hangs. There are several problems here, but the two obvious ones from the trace are that: a) log recovery does not yield the CPU for over 4 seconds, b) binding CIL pushes to a single CPU is a really bad idea. This patch addresses just these two aspects of the problem, and are suitable for backporting to work around any issues in older kernels. The more fundamental problem of preventing the CIL from consuming more than 50% of the log without committing will take more invasive and complex work, so will be done as followup work. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2019-09-06 12:35:39 +08:00
* This is called during recovery to process any inodes which we unlinked but
* not freed when the system crashed. These inodes will be on the lists in the
* AGI blocks. What we do here is scan all the AGIs and fully truncate and free
* any inodes found on the lists. Each inode is removed from the lists when it
* has been fully truncated and is freed. The freeing of the inode and its
* removal from the list must be atomic.
*
* If everything we touch in the agi processing loop is already in memory, this
* loop can hold the cpu for a long time. It runs without lock contention,
* memory allocation contention, the need wait for IO, etc, and so will run
* until we either run out of inodes to process, run low on memory or we run out
* of log space.
*
* This behaviour is bad for latency on single CPU and non-preemptible kernels,
* and can prevent other filesystem work (such as CIL pushes) from running. This
xfs: prevent CIL push holdoff in log recovery generic/530 on a machine with enough ram and a non-preemptible kernel can run the AGI processing phase of log recovery enitrely out of cache. This means it never blocks on locks, never waits for IO and runs entirely through the unlinked lists until it either completes or blocks and hangs because it has run out of log space. It runs out of log space because the background CIL push is scheduled but never runs. queue_work() queues the CIL work on the current CPU that is busy, and the workqueue code will not run it on any other CPU. Hence if the unlinked list processing never yields the CPU voluntarily, the push work is delayed indefinitely. This results in the CIL aggregating changes until all the log space is consumed. When the log recoveyr processing evenutally blocks, the CIL flushes but because the last iclog isn't submitted for IO because it isn't full, the CIL flush never completes and nothing ever moves the log head forwards, or indeed inserts anything into the tail of the log, and hence nothing is able to get the log moving again and recovery hangs. There are several problems here, but the two obvious ones from the trace are that: a) log recovery does not yield the CPU for over 4 seconds, b) binding CIL pushes to a single CPU is a really bad idea. This patch addresses just these two aspects of the problem, and are suitable for backporting to work around any issues in older kernels. The more fundamental problem of preventing the CIL from consuming more than 50% of the log without committing will take more invasive and complex work, so will be done as followup work. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2019-09-06 12:35:39 +08:00
* can lead to deadlocks if the recovery process runs out of log reservation
* space. Hence we need to yield the CPU when there is other kernel work
* scheduled on this CPU to ensure other scheduled work can run without undue
* latency.
*/
static void
xlog_recover_iunlink_ag(
struct xfs_perag *pag)
{
struct xfs_agi *agi;
struct xfs_buf *agibp;
int bucket;
int error;
error = xfs_read_agi(pag, NULL, &agibp);
if (error) {
/*
* AGI is b0rked. Don't process it.
*
* We should probably mark the filesystem as corrupt after we've
* recovered all the ag's we can....
*/
return;
}
/*
* Unlock the buffer so that it can be acquired in the normal course of
* the transaction to truncate and free each inode. Because we are not
* racing with anyone else here for the AGI buffer, we don't even need
* to hold it locked to read the initial unlinked bucket entries out of
* the buffer. We keep buffer reference though, so that it stays pinned
* in memory while we need the buffer.
*/
agi = agibp->b_addr;
xfs_buf_unlock(agibp);
for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++) {
error = xlog_recover_iunlink_bucket(pag, agi, bucket);
if (error) {
/*
* Bucket is unrecoverable, so only a repair scan can
* free the remaining unlinked inodes. Just empty the
* bucket and remaining inodes on it unreferenced and
* unfreeable.
*/
xlog_recover_clear_agi_bucket(pag, bucket);
}
}
xfs: per-cpu deferred inode inactivation queues Move inode inactivation to background work contexts so that it no longer runs in the context that releases the final reference to an inode. This will allow process work that ends up blocking on inactivation to continue doing work while the filesytem processes the inactivation in the background. A typical demonstration of this is unlinking an inode with lots of extents. The extents are removed during inactivation, so this blocks the process that unlinked the inode from the directory structure. By moving the inactivation to the background process, the userspace applicaiton can keep working (e.g. unlinking the next inode in the directory) while the inactivation work on the previous inode is done by a different CPU. The implementation of the queue is relatively simple. We use a per-cpu lockless linked list (llist) to queue inodes for inactivation without requiring serialisation mechanisms, and a work item to allow the queue to be processed by a CPU bound worker thread. We also keep a count of the queue depth so that we can trigger work after a number of deferred inactivations have been queued. The use of a bound workqueue with a single work depth allows the workqueue to run one work item per CPU. We queue the work item on the CPU we are currently running on, and so this essentially gives us affine per-cpu worker threads for the per-cpu queues. THis maintains the effective CPU affinity that occurs within XFS at the AG level due to all objects in a directory being local to an AG. Hence inactivation work tends to run on the same CPU that last accessed all the objects that inactivation accesses and this maintains hot CPU caches for unlink workloads. A depth of 32 inodes was chosen to match the number of inodes in an inode cluster buffer. This hopefully allows sequential allocation/unlink behaviours to defering inactivation of all the inodes in a single cluster buffer at a time, further helping maintain hot CPU and buffer cache accesses while running inactivations. A hard per-cpu queue throttle of 256 inode has been set to avoid runaway queuing when inodes that take a long to time inactivate are being processed. For example, when unlinking inodes with large numbers of extents that can take a lot of processing to free. Signed-off-by: Dave Chinner <dchinner@redhat.com> [djwong: tweak comments and tracepoints, convert opflags to state bits] Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-08-07 02:05:39 +08:00
xfs_buf_rele(agibp);
}
static void
xlog_recover_process_iunlinks(
struct xlog *log)
{
struct xfs_perag *pag;
xfs_agnumber_t agno;
for_each_perag(log->l_mp, agno, pag)
xlog_recover_iunlink_ag(pag);
}
STATIC void
xlog_unpack_data(
struct xlog_rec_header *rhead,
char *dp,
struct xlog *log)
{
int i, j, k;
for (i = 0; i < BTOBB(be32_to_cpu(rhead->h_len)) &&
i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) {
*(__be32 *)dp = *(__be32 *)&rhead->h_cycle_data[i];
dp += BBSIZE;
}
if (xfs_has_logv2(log->l_mp)) {
xlog_in_core_2_t *xhdr = (xlog_in_core_2_t *)rhead;
for ( ; i < BTOBB(be32_to_cpu(rhead->h_len)); i++) {
j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
*(__be32 *)dp = xhdr[j].hic_xheader.xh_cycle_data[k];
dp += BBSIZE;
}
}
}
/*
* CRC check, unpack and process a log record.
*/
STATIC int
xlog_recover_process(
struct xlog *log,
struct hlist_head rhash[],
struct xlog_rec_header *rhead,
char *dp,
int pass,
struct list_head *buffer_list)
{
__le32 old_crc = rhead->h_crc;
__le32 crc;
crc = xlog_cksum(log, rhead, dp, be32_to_cpu(rhead->h_len));
/*
* Nothing else to do if this is a CRC verification pass. Just return
* if this a record with a non-zero crc. Unfortunately, mkfs always
* sets old_crc to 0 so we must consider this valid even on v5 supers.
* Otherwise, return EFSBADCRC on failure so the callers up the stack
* know precisely what failed.
*/
if (pass == XLOG_RECOVER_CRCPASS) {
if (old_crc && crc != old_crc)
return -EFSBADCRC;
return 0;
}
/*
* We're in the normal recovery path. Issue a warning if and only if the
* CRC in the header is non-zero. This is an advisory warning and the
* zero CRC check prevents warnings from being emitted when upgrading
* the kernel from one that does not add CRCs by default.
*/
if (crc != old_crc) {
if (old_crc || xfs_has_crc(log->l_mp)) {
xfs_alert(log->l_mp,
"log record CRC mismatch: found 0x%x, expected 0x%x.",
le32_to_cpu(old_crc),
le32_to_cpu(crc));
xfs_hex_dump(dp, 32);
}
/*
* If the filesystem is CRC enabled, this mismatch becomes a
* fatal log corruption failure.
*/
if (xfs_has_crc(log->l_mp)) {
XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, log->l_mp);
return -EFSCORRUPTED;
}
}
xlog_unpack_data(rhead, dp, log);
return xlog_recover_process_data(log, rhash, rhead, dp, pass,
buffer_list);
}
STATIC int
xlog_valid_rec_header(
struct xlog *log,
struct xlog_rec_header *rhead,
xfs_daddr_t blkno,
int bufsize)
{
int hlen;
if (XFS_IS_CORRUPT(log->l_mp,
rhead->h_magicno != cpu_to_be32(XLOG_HEADER_MAGIC_NUM)))
return -EFSCORRUPTED;
if (XFS_IS_CORRUPT(log->l_mp,
(!rhead->h_version ||
(be32_to_cpu(rhead->h_version) &
(~XLOG_VERSION_OKBITS))))) {
xfs_warn(log->l_mp, "%s: unrecognised log version (%d).",
__func__, be32_to_cpu(rhead->h_version));
return -EFSCORRUPTED;
}
/*
* LR body must have data (or it wouldn't have been written)
* and h_len must not be greater than LR buffer size.
*/
hlen = be32_to_cpu(rhead->h_len);
if (XFS_IS_CORRUPT(log->l_mp, hlen <= 0 || hlen > bufsize))
return -EFSCORRUPTED;
if (XFS_IS_CORRUPT(log->l_mp,
blkno > log->l_logBBsize || blkno > INT_MAX))
return -EFSCORRUPTED;
return 0;
}
/*
* Read the log from tail to head and process the log records found.
* Handle the two cases where the tail and head are in the same cycle
* and where the active portion of the log wraps around the end of
* the physical log separately. The pass parameter is passed through
* to the routines called to process the data and is not looked at
* here.
*/
STATIC int
xlog_do_recovery_pass(
struct xlog *log,
xfs_daddr_t head_blk,
xfs_daddr_t tail_blk,
int pass,
xfs_daddr_t *first_bad) /* out: first bad log rec */
{
xlog_rec_header_t *rhead;
xfs: fix recovery failure when log record header wraps log end The high-level log recovery algorithm consists of two loops that walk the physical log and process log records from the tail to the head. The first loop handles the case where the tail is beyond the head and processes records up to the end of the physical log. The subsequent loop processes records from the beginning of the physical log to the head. Because log records can wrap around the end of the physical log, the first loop mentioned above must handle this case appropriately. Records are processed from in-core buffers, which means that this algorithm must split the reads of such records into two partial I/Os: 1.) from the beginning of the record to the end of the log and 2.) from the beginning of the log to the end of the record. This is further complicated by the fact that the log record header and log record data are read into independent buffers. The current handling of each buffer correctly splits the reads when either the header or data starts before the end of the log and wraps around the end. The data read does not correctly handle the case where the prior header read wrapped or ends on the physical log end boundary. blk_no is incremented to or beyond the log end after the header read to point to the record data, but the split data read logic triggers, attempts to read from an invalid log block and ultimately causes log recovery to fail. This can be reproduced fairly reliably via xfstests tests generic/047 and generic/388 with large iclog sizes (256k) and small (10M) logs. If the record header read has pushed beyond the end of the physical log, the subsequent data read is actually contiguous. Update the data read logic to detect the case where blk_no has wrapped, mod it against the log size to read from the correct address and issue one contiguous read for the log data buffer. The log record is processed as normal from the buffer(s), the loop exits after the current iteration and the subsequent loop picks up with the first new record after the start of the log. Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-08-09 09:21:51 +08:00
xfs_daddr_t blk_no, rblk_no;
xfs_daddr_t rhead_blk;
char *offset;
char *hbp, *dbp;
xfs: detect and handle invalid iclog size set by mkfs XFS log records have separate fields for the record size and the iclog size used to write the record. mkfs.xfs zeroes the log and writes an unmount record to generate a clean log for the subsequent mount. The userspace record logging code has a bug where the iclog size (h_size) field of the log record is hardcoded to 32k, even if a log stripe unit is specified. The log record length is correctly extended to the stripe unit. Since the kernel log recovery code uses the h_size field to determine the log buffer size, this means that the kernel can attempt to read/process records larger than the buffer size and overrun the buffer. This has historically not been a problem because the kernel doesn't actually run through log recovery in the clean unmount case. Instead, the kernel detects that a single unmount record exists between the head and tail and pushes the tail forward such that the log is viewed as clean (head == tail). Once CRC verification is enabled, however, all records at the head of the log are verified for CRC errors and thus we are susceptible to overrun problems if the iclog field is not correct. While the core problem must be fixed in userspace, this is historical behavior that must be detected in the kernel to avoid severe side effects such as memory corruption and crashes. Update the log buffer size calculation code to detect this condition, warn the user and resize the log buffer based on the log stripe unit. Return a corruption error in cases where this does not look like a clean filesystem (i.e., the log record header indicates more than one operation). Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-01-04 12:55:10 +08:00
int error = 0, h_size, h_len;
int error2 = 0;
int bblks, split_bblks;
int hblks, split_hblks, wrapped_hblks;
xfs: free uncommitted transactions during log recovery Log recovery allocates in-core transaction and member item data structures on-demand as it processes the on-disk log. Transactions are allocated on first encounter on-disk and stored in a hash table structure where they are easily accessible for subsequent lookups. Transaction items are also allocated on demand and are attached to the associated transactions. When a commit record is encountered in the log, the transaction is committed to the fs and the in-core structures are freed. If a filesystem crashes or shuts down before all in-core log buffers are flushed to the log, however, not all transactions may have commit records in the log. As expected, the modifications in such an incomplete transaction are not replayed to the fs. The in-core data structures for the partial transaction are never freed, however, resulting in a memory leak. Update xlog_do_recovery_pass() to first correctly initialize the hash table array so empty lists can be distinguished from populated lists on function exit. Update xlog_recover_free_trans() to always remove the transaction from the list prior to freeing the associated memory. Finally, walk the hash table of transaction lists as the last step before it goes out of scope and free any transactions that may remain on the lists. This prevents a memory leak of partial transactions in the log. Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-06-25 01:11:41 +08:00
int i;
struct hlist_head rhash[XLOG_RHASH_SIZE];
LIST_HEAD (buffer_list);
ASSERT(head_blk != tail_blk);
blk_no = rhead_blk = tail_blk;
xfs: free uncommitted transactions during log recovery Log recovery allocates in-core transaction and member item data structures on-demand as it processes the on-disk log. Transactions are allocated on first encounter on-disk and stored in a hash table structure where they are easily accessible for subsequent lookups. Transaction items are also allocated on demand and are attached to the associated transactions. When a commit record is encountered in the log, the transaction is committed to the fs and the in-core structures are freed. If a filesystem crashes or shuts down before all in-core log buffers are flushed to the log, however, not all transactions may have commit records in the log. As expected, the modifications in such an incomplete transaction are not replayed to the fs. The in-core data structures for the partial transaction are never freed, however, resulting in a memory leak. Update xlog_do_recovery_pass() to first correctly initialize the hash table array so empty lists can be distinguished from populated lists on function exit. Update xlog_recover_free_trans() to always remove the transaction from the list prior to freeing the associated memory. Finally, walk the hash table of transaction lists as the last step before it goes out of scope and free any transactions that may remain on the lists. This prevents a memory leak of partial transactions in the log. Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-06-25 01:11:41 +08:00
for (i = 0; i < XLOG_RHASH_SIZE; i++)
INIT_HLIST_HEAD(&rhash[i]);
/*
* Read the header of the tail block and get the iclog buffer size from
* h_size. Use this to tell how many sectors make up the log header.
*/
if (xfs_has_logv2(log->l_mp)) {
/*
* When using variable length iclogs, read first sector of
* iclog header and extract the header size from it. Get a
* new hbp that is the correct size.
*/
hbp = xlog_alloc_buffer(log, 1);
if (!hbp)
return -ENOMEM;
error = xlog_bread(log, tail_blk, 1, hbp, &offset);
if (error)
goto bread_err1;
rhead = (xlog_rec_header_t *)offset;
xfs: detect and handle invalid iclog size set by mkfs XFS log records have separate fields for the record size and the iclog size used to write the record. mkfs.xfs zeroes the log and writes an unmount record to generate a clean log for the subsequent mount. The userspace record logging code has a bug where the iclog size (h_size) field of the log record is hardcoded to 32k, even if a log stripe unit is specified. The log record length is correctly extended to the stripe unit. Since the kernel log recovery code uses the h_size field to determine the log buffer size, this means that the kernel can attempt to read/process records larger than the buffer size and overrun the buffer. This has historically not been a problem because the kernel doesn't actually run through log recovery in the clean unmount case. Instead, the kernel detects that a single unmount record exists between the head and tail and pushes the tail forward such that the log is viewed as clean (head == tail). Once CRC verification is enabled, however, all records at the head of the log are verified for CRC errors and thus we are susceptible to overrun problems if the iclog field is not correct. While the core problem must be fixed in userspace, this is historical behavior that must be detected in the kernel to avoid severe side effects such as memory corruption and crashes. Update the log buffer size calculation code to detect this condition, warn the user and resize the log buffer based on the log stripe unit. Return a corruption error in cases where this does not look like a clean filesystem (i.e., the log record header indicates more than one operation). Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-01-04 12:55:10 +08:00
/*
* xfsprogs has a bug where record length is based on lsunit but
* h_size (iclog size) is hardcoded to 32k. Now that we
* unconditionally CRC verify the unmount record, this means the
* log buffer can be too small for the record and cause an
* overrun.
*
* Detect this condition here. Use lsunit for the buffer size as
* long as this looks like the mkfs case. Otherwise, return an
* error to avoid a buffer overrun.
*/
h_size = be32_to_cpu(rhead->h_size);
xfs: detect and handle invalid iclog size set by mkfs XFS log records have separate fields for the record size and the iclog size used to write the record. mkfs.xfs zeroes the log and writes an unmount record to generate a clean log for the subsequent mount. The userspace record logging code has a bug where the iclog size (h_size) field of the log record is hardcoded to 32k, even if a log stripe unit is specified. The log record length is correctly extended to the stripe unit. Since the kernel log recovery code uses the h_size field to determine the log buffer size, this means that the kernel can attempt to read/process records larger than the buffer size and overrun the buffer. This has historically not been a problem because the kernel doesn't actually run through log recovery in the clean unmount case. Instead, the kernel detects that a single unmount record exists between the head and tail and pushes the tail forward such that the log is viewed as clean (head == tail). Once CRC verification is enabled, however, all records at the head of the log are verified for CRC errors and thus we are susceptible to overrun problems if the iclog field is not correct. While the core problem must be fixed in userspace, this is historical behavior that must be detected in the kernel to avoid severe side effects such as memory corruption and crashes. Update the log buffer size calculation code to detect this condition, warn the user and resize the log buffer based on the log stripe unit. Return a corruption error in cases where this does not look like a clean filesystem (i.e., the log record header indicates more than one operation). Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-01-04 12:55:10 +08:00
h_len = be32_to_cpu(rhead->h_len);
if (h_len > h_size && h_len <= log->l_mp->m_logbsize &&
rhead->h_num_logops == cpu_to_be32(1)) {
xfs_warn(log->l_mp,
xfs: detect and handle invalid iclog size set by mkfs XFS log records have separate fields for the record size and the iclog size used to write the record. mkfs.xfs zeroes the log and writes an unmount record to generate a clean log for the subsequent mount. The userspace record logging code has a bug where the iclog size (h_size) field of the log record is hardcoded to 32k, even if a log stripe unit is specified. The log record length is correctly extended to the stripe unit. Since the kernel log recovery code uses the h_size field to determine the log buffer size, this means that the kernel can attempt to read/process records larger than the buffer size and overrun the buffer. This has historically not been a problem because the kernel doesn't actually run through log recovery in the clean unmount case. Instead, the kernel detects that a single unmount record exists between the head and tail and pushes the tail forward such that the log is viewed as clean (head == tail). Once CRC verification is enabled, however, all records at the head of the log are verified for CRC errors and thus we are susceptible to overrun problems if the iclog field is not correct. While the core problem must be fixed in userspace, this is historical behavior that must be detected in the kernel to avoid severe side effects such as memory corruption and crashes. Update the log buffer size calculation code to detect this condition, warn the user and resize the log buffer based on the log stripe unit. Return a corruption error in cases where this does not look like a clean filesystem (i.e., the log record header indicates more than one operation). Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-01-04 12:55:10 +08:00
"invalid iclog size (%d bytes), using lsunit (%d bytes)",
h_size, log->l_mp->m_logbsize);
h_size = log->l_mp->m_logbsize;
xfs: detect and handle invalid iclog size set by mkfs XFS log records have separate fields for the record size and the iclog size used to write the record. mkfs.xfs zeroes the log and writes an unmount record to generate a clean log for the subsequent mount. The userspace record logging code has a bug where the iclog size (h_size) field of the log record is hardcoded to 32k, even if a log stripe unit is specified. The log record length is correctly extended to the stripe unit. Since the kernel log recovery code uses the h_size field to determine the log buffer size, this means that the kernel can attempt to read/process records larger than the buffer size and overrun the buffer. This has historically not been a problem because the kernel doesn't actually run through log recovery in the clean unmount case. Instead, the kernel detects that a single unmount record exists between the head and tail and pushes the tail forward such that the log is viewed as clean (head == tail). Once CRC verification is enabled, however, all records at the head of the log are verified for CRC errors and thus we are susceptible to overrun problems if the iclog field is not correct. While the core problem must be fixed in userspace, this is historical behavior that must be detected in the kernel to avoid severe side effects such as memory corruption and crashes. Update the log buffer size calculation code to detect this condition, warn the user and resize the log buffer based on the log stripe unit. Return a corruption error in cases where this does not look like a clean filesystem (i.e., the log record header indicates more than one operation). Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-01-04 12:55:10 +08:00
}
error = xlog_valid_rec_header(log, rhead, tail_blk, h_size);
if (error)
goto bread_err1;
hblks = xlog_logrec_hblks(log, rhead);
if (hblks != 1) {
kmem_free(hbp);
hbp = xlog_alloc_buffer(log, hblks);
}
} else {
ASSERT(log->l_sectBBsize == 1);
hblks = 1;
hbp = xlog_alloc_buffer(log, 1);
h_size = XLOG_BIG_RECORD_BSIZE;
}
if (!hbp)
return -ENOMEM;
dbp = xlog_alloc_buffer(log, BTOBB(h_size));
if (!dbp) {
kmem_free(hbp);
return -ENOMEM;
}
memset(rhash, 0, sizeof(rhash));
if (tail_blk > head_blk) {
/*
* Perform recovery around the end of the physical log.
* When the head is not on the same cycle number as the tail,
* we can't do a sequential recovery.
*/
while (blk_no < log->l_logBBsize) {
/*
* Check for header wrapping around physical end-of-log
*/
offset = hbp;
split_hblks = 0;
wrapped_hblks = 0;
if (blk_no + hblks <= log->l_logBBsize) {
/* Read header in one read */
error = xlog_bread(log, blk_no, hblks, hbp,
&offset);
if (error)
goto bread_err2;
} else {
/* This LR is split across physical log end */
if (blk_no != log->l_logBBsize) {
/* some data before physical log end */
ASSERT(blk_no <= INT_MAX);
split_hblks = log->l_logBBsize - (int)blk_no;
ASSERT(split_hblks > 0);
error = xlog_bread(log, blk_no,
split_hblks, hbp,
&offset);
if (error)
goto bread_err2;
}
/*
* Note: this black magic still works with
* large sector sizes (non-512) only because:
* - we increased the buffer size originally
* by 1 sector giving us enough extra space
* for the second read;
* - the log start is guaranteed to be sector
* aligned;
* - we read the log end (LR header start)
* _first_, then the log start (LR header end)
* - order is important.
*/
wrapped_hblks = hblks - split_hblks;
error = xlog_bread_noalign(log, 0,
wrapped_hblks,
offset + BBTOB(split_hblks));
if (error)
goto bread_err2;
}
rhead = (xlog_rec_header_t *)offset;
error = xlog_valid_rec_header(log, rhead,
split_hblks ? blk_no : 0, h_size);
if (error)
goto bread_err2;
bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
blk_no += hblks;
xfs: fix recovery failure when log record header wraps log end The high-level log recovery algorithm consists of two loops that walk the physical log and process log records from the tail to the head. The first loop handles the case where the tail is beyond the head and processes records up to the end of the physical log. The subsequent loop processes records from the beginning of the physical log to the head. Because log records can wrap around the end of the physical log, the first loop mentioned above must handle this case appropriately. Records are processed from in-core buffers, which means that this algorithm must split the reads of such records into two partial I/Os: 1.) from the beginning of the record to the end of the log and 2.) from the beginning of the log to the end of the record. This is further complicated by the fact that the log record header and log record data are read into independent buffers. The current handling of each buffer correctly splits the reads when either the header or data starts before the end of the log and wraps around the end. The data read does not correctly handle the case where the prior header read wrapped or ends on the physical log end boundary. blk_no is incremented to or beyond the log end after the header read to point to the record data, but the split data read logic triggers, attempts to read from an invalid log block and ultimately causes log recovery to fail. This can be reproduced fairly reliably via xfstests tests generic/047 and generic/388 with large iclog sizes (256k) and small (10M) logs. If the record header read has pushed beyond the end of the physical log, the subsequent data read is actually contiguous. Update the data read logic to detect the case where blk_no has wrapped, mod it against the log size to read from the correct address and issue one contiguous read for the log data buffer. The log record is processed as normal from the buffer(s), the loop exits after the current iteration and the subsequent loop picks up with the first new record after the start of the log. Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-08-09 09:21:51 +08:00
/*
* Read the log record data in multiple reads if it
* wraps around the end of the log. Note that if the
* header already wrapped, blk_no could point past the
* end of the log. The record data is contiguous in
* that case.
*/
if (blk_no + bblks <= log->l_logBBsize ||
blk_no >= log->l_logBBsize) {
rblk_no = xlog_wrap_logbno(log, blk_no);
xfs: fix recovery failure when log record header wraps log end The high-level log recovery algorithm consists of two loops that walk the physical log and process log records from the tail to the head. The first loop handles the case where the tail is beyond the head and processes records up to the end of the physical log. The subsequent loop processes records from the beginning of the physical log to the head. Because log records can wrap around the end of the physical log, the first loop mentioned above must handle this case appropriately. Records are processed from in-core buffers, which means that this algorithm must split the reads of such records into two partial I/Os: 1.) from the beginning of the record to the end of the log and 2.) from the beginning of the log to the end of the record. This is further complicated by the fact that the log record header and log record data are read into independent buffers. The current handling of each buffer correctly splits the reads when either the header or data starts before the end of the log and wraps around the end. The data read does not correctly handle the case where the prior header read wrapped or ends on the physical log end boundary. blk_no is incremented to or beyond the log end after the header read to point to the record data, but the split data read logic triggers, attempts to read from an invalid log block and ultimately causes log recovery to fail. This can be reproduced fairly reliably via xfstests tests generic/047 and generic/388 with large iclog sizes (256k) and small (10M) logs. If the record header read has pushed beyond the end of the physical log, the subsequent data read is actually contiguous. Update the data read logic to detect the case where blk_no has wrapped, mod it against the log size to read from the correct address and issue one contiguous read for the log data buffer. The log record is processed as normal from the buffer(s), the loop exits after the current iteration and the subsequent loop picks up with the first new record after the start of the log. Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-08-09 09:21:51 +08:00
error = xlog_bread(log, rblk_no, bblks, dbp,
&offset);
if (error)
goto bread_err2;
} else {
/* This log record is split across the
* physical end of log */
offset = dbp;
split_bblks = 0;
if (blk_no != log->l_logBBsize) {
/* some data is before the physical
* end of log */
ASSERT(!wrapped_hblks);
ASSERT(blk_no <= INT_MAX);
split_bblks =
log->l_logBBsize - (int)blk_no;
ASSERT(split_bblks > 0);
error = xlog_bread(log, blk_no,
split_bblks, dbp,
&offset);
if (error)
goto bread_err2;
}
/*
* Note: this black magic still works with
* large sector sizes (non-512) only because:
* - we increased the buffer size originally
* by 1 sector giving us enough extra space
* for the second read;
* - the log start is guaranteed to be sector
* aligned;
* - we read the log end (LR header start)
* _first_, then the log start (LR header end)
* - order is important.
*/
error = xlog_bread_noalign(log, 0,
bblks - split_bblks,
offset + BBTOB(split_bblks));
if (error)
goto bread_err2;
}
2012-11-12 19:54:24 +08:00
error = xlog_recover_process(log, rhash, rhead, offset,
pass, &buffer_list);
2012-11-12 19:54:24 +08:00
if (error)
goto bread_err2;
blk_no += bblks;
rhead_blk = blk_no;
}
ASSERT(blk_no >= log->l_logBBsize);
blk_no -= log->l_logBBsize;
rhead_blk = blk_no;
}
/* read first part of physical log */
while (blk_no < head_blk) {
error = xlog_bread(log, blk_no, hblks, hbp, &offset);
if (error)
goto bread_err2;
rhead = (xlog_rec_header_t *)offset;
error = xlog_valid_rec_header(log, rhead, blk_no, h_size);
if (error)
goto bread_err2;
/* blocks in data section */
bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
error = xlog_bread(log, blk_no+hblks, bblks, dbp,
&offset);
if (error)
goto bread_err2;
error = xlog_recover_process(log, rhash, rhead, offset, pass,
&buffer_list);
if (error)
goto bread_err2;
blk_no += bblks + hblks;
rhead_blk = blk_no;
}
bread_err2:
kmem_free(dbp);
bread_err1:
kmem_free(hbp);
/*
* Submit buffers that have been added from the last record processed,
* regardless of error status.
*/
if (!list_empty(&buffer_list))
error2 = xfs_buf_delwri_submit(&buffer_list);
if (error && first_bad)
*first_bad = rhead_blk;
xfs: free uncommitted transactions during log recovery Log recovery allocates in-core transaction and member item data structures on-demand as it processes the on-disk log. Transactions are allocated on first encounter on-disk and stored in a hash table structure where they are easily accessible for subsequent lookups. Transaction items are also allocated on demand and are attached to the associated transactions. When a commit record is encountered in the log, the transaction is committed to the fs and the in-core structures are freed. If a filesystem crashes or shuts down before all in-core log buffers are flushed to the log, however, not all transactions may have commit records in the log. As expected, the modifications in such an incomplete transaction are not replayed to the fs. The in-core data structures for the partial transaction are never freed, however, resulting in a memory leak. Update xlog_do_recovery_pass() to first correctly initialize the hash table array so empty lists can be distinguished from populated lists on function exit. Update xlog_recover_free_trans() to always remove the transaction from the list prior to freeing the associated memory. Finally, walk the hash table of transaction lists as the last step before it goes out of scope and free any transactions that may remain on the lists. This prevents a memory leak of partial transactions in the log. Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-06-25 01:11:41 +08:00
/*
* Transactions are freed at commit time but transactions without commit
* records on disk are never committed. Free any that may be left in the
* hash table.
*/
for (i = 0; i < XLOG_RHASH_SIZE; i++) {
struct hlist_node *tmp;
struct xlog_recover *trans;
hlist_for_each_entry_safe(trans, tmp, &rhash[i], r_list)
xlog_recover_free_trans(trans);
}
return error ? error : error2;
}
/*
* Do the recovery of the log. We actually do this in two phases.
* The two passes are necessary in order to implement the function
* of cancelling a record written into the log. The first pass
* determines those things which have been cancelled, and the
* second pass replays log items normally except for those which
* have been cancelled. The handling of the replay and cancellations
* takes place in the log item type specific routines.
*
* The table of items which have cancel records in the log is allocated
* and freed at this level, since only here do we know when all of
* the log recovery has been completed.
*/
STATIC int
xlog_do_log_recovery(
struct xlog *log,
xfs_daddr_t head_blk,
xfs_daddr_t tail_blk)
{
int error;
ASSERT(head_blk != tail_blk);
/*
* First do a pass to find all of the cancelled buf log items.
* Store them in the buf_cancel_table for use in the second pass.
*/
error = xlog_alloc_buf_cancel_table(log);
if (error)
return error;
error = xlog_do_recovery_pass(log, head_blk, tail_blk,
XLOG_RECOVER_PASS1, NULL);
if (error != 0)
goto out_cancel;
/*
* Then do a second pass to actually recover the items in the log.
* When it is complete free the table of buf cancel items.
*/
error = xlog_do_recovery_pass(log, head_blk, tail_blk,
XLOG_RECOVER_PASS2, NULL);
if (!error)
xlog_check_buf_cancel_table(log);
out_cancel:
xlog_free_buf_cancel_table(log);
return error;
}
/*
* Do the actual recovery
*/
STATIC int
xlog_do_recover(
struct xlog *log,
xfs_daddr_t head_blk,
xfs_daddr_t tail_blk)
{
struct xfs_mount *mp = log->l_mp;
struct xfs_buf *bp = mp->m_sb_bp;
struct xfs_sb *sbp = &mp->m_sb;
int error;
trace_xfs_log_recover(log, head_blk, tail_blk);
/*
* First replay the images in the log.
*/
error = xlog_do_log_recovery(log, head_blk, tail_blk);
xfs: on-stack delayed write buffer lists Queue delwri buffers on a local on-stack list instead of a per-buftarg one, and write back the buffers per-process instead of by waking up xfsbufd. This is now easily doable given that we have very few places left that write delwri buffers: - log recovery: Only done at mount time, and already forcing out the buffers synchronously using xfs_flush_buftarg - quotacheck: Same story. - dquot reclaim: Writes out dirty dquots on the LRU under memory pressure. We might want to look into doing more of this via xfsaild, but it's already more optimal than the synchronous inode reclaim that writes each buffer synchronously. - xfsaild: This is the main beneficiary of the change. By keeping a local list of buffers to write we reduce latency of writing out buffers, and more importably we can remove all the delwri list promotions which were hitting the buffer cache hard under sustained metadata loads. The implementation is very straight forward - xfs_buf_delwri_queue now gets a new list_head pointer that it adds the delwri buffers to, and all callers need to eventually submit the list using xfs_buf_delwi_submit or xfs_buf_delwi_submit_nowait. Buffers that already are on a delwri list are skipped in xfs_buf_delwri_queue, assuming they already are on another delwri list. The biggest change to pass down the buffer list was done to the AIL pushing. Now that we operate on buffers the trylock, push and pushbuf log item methods are merged into a single push routine, which tries to lock the item, and if possible add the buffer that needs writeback to the buffer list. This leads to much simpler code than the previous split but requires the individual IOP_PUSH instances to unlock and reacquire the AIL around calls to blocking routines. Given that xfsailds now also handle writing out buffers, the conditions for log forcing and the sleep times needed some small changes. The most important one is that we consider an AIL busy as long we still have buffers to push, and the other one is that we do increment the pushed LSN for buffers that are under flushing at this moment, but still count them towards the stuck items for restart purposes. Without this we could hammer on stuck items without ever forcing the log and not make progress under heavy random delete workloads on fast flash storage devices. [ Dave Chinner: - rebase on previous patches. - improved comments for XBF_DELWRI_Q handling - fix XBF_ASYNC handling in queue submission (test 106 failure) - rename delwri submit function buffer list parameters for clarity - xfs_efd_item_push() should return XFS_ITEM_PINNED ] Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Mark Tinguely <tinguely@sgi.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2012-04-23 13:58:39 +08:00
if (error)
return error;
if (xlog_is_shutdown(log))
return -EIO;
/*
* We now update the tail_lsn since much of the recovery has completed
* and there may be space available to use. If there were no extent
* or iunlinks, we can free up the entire log and set the tail_lsn to
* be the last_sync_lsn. This was set in xlog_find_tail to be the
* lsn of the last known good LR on disk. If there are extent frees
* or iunlinks they will have some entries in the AIL; so we look at
* the AIL to determine how to set the tail_lsn.
*/
xlog_assign_tail_lsn(mp);
/*
* Now that we've finished replaying all buffer and inode updates,
* re-read the superblock and reverify it.
*/
xfs_buf_lock(bp);
xfs_buf_hold(bp);
error = _xfs_buf_read(bp, XBF_READ);
if (error) {
if (!xlog_is_shutdown(log)) {
xfs_buf_ioerror_alert(bp, __this_address);
ASSERT(0);
}
xfs_buf_relse(bp);
return error;
}
/* Convert superblock from on-disk format */
xfs_sb_from_disk(sbp, bp->b_addr);
xfs_buf_relse(bp);
/* re-initialise in-core superblock and geometry structures */
mp->m_features |= xfs_sb_version_to_features(sbp);
xfs_reinit_percpu_counters(mp);
xfs: Pre-calculate per-AG agbno geometry There is a lot of overhead in functions like xfs_verify_agbno() that repeatedly calculate the geometry limits of an AG. These can be pre-calculated as they are static and the verification context has a per-ag context it can quickly reference. In the case of xfs_verify_agbno(), we now always have a perag context handy, so we can store the AG length and the minimum valid block in the AG in the perag. This means we don't have to calculate it on every call and it can be inlined in callers if we move it to xfs_ag.h. Move xfs_ag_block_count() to xfs_ag.c because it's really a per-ag function and not an XFS type function. We need a little bit of rework that is specific to xfs_initialise_perag() to allow growfs to calculate the new perag sizes before we've updated the primary superblock during the grow (chicken/egg situation). Note that we leave the original xfs_verify_agbno in place in xfs_types.c as a static function as other callers in that file do not have per-ag contexts so still need to go the long way. It's been renamed to xfs_verify_agno_agbno() to indicate it takes both an agno and an agbno to differentiate it from new function. Future commits will make similar changes for other per-ag geometry validation functions. Further: $ size --totals fs/xfs/built-in.a text data bss dec hex filename before 1483006 329588 572 1813166 1baaae (TOTALS) after 1482185 329588 572 1812345 1ba779 (TOTALS) This rework reduces the binary size by ~820 bytes, indicating that much less work is being done to bounds check the agbno values against on per-ag geometry information. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <djwong@kernel.org>
2022-07-07 17:13:02 +08:00
error = xfs_initialize_perag(mp, sbp->sb_agcount, sbp->sb_dblocks,
&mp->m_maxagi);
if (error) {
xfs_warn(mp, "Failed post-recovery per-ag init: %d", error);
return error;
}
mp->m_alloc_set_aside = xfs_alloc_set_aside(mp);
/* Normal transactions can now occur */
clear_bit(XLOG_ACTIVE_RECOVERY, &log->l_opstate);
return 0;
}
/*
* Perform recovery and re-initialize some log variables in xlog_find_tail.
*
* Return error or zero.
*/
int
xlog_recover(
struct xlog *log)
{
xfs_daddr_t head_blk, tail_blk;
int error;
/* find the tail of the log */
xfs: validate metadata LSNs against log on v5 superblocks Since the onset of v5 superblocks, the LSN of the last modification has been included in a variety of on-disk data structures. This LSN is used to provide log recovery ordering guarantees (e.g., to ensure an older log recovery item is not replayed over a newer target data structure). While this works correctly from the point a filesystem is formatted and mounted, userspace tools have some problematic behaviors that defeat this mechanism. For example, xfs_repair historically zeroes out the log unconditionally (regardless of whether corruption is detected). If this occurs, the LSN of the filesystem is reset and the log is now in a problematic state with respect to on-disk metadata structures that might have a larger LSN. Until either the log catches up to the highest previously used metadata LSN or each affected data structure is modified and written out without incident (which resets the metadata LSN), log recovery is susceptible to filesystem corruption. This problem is ultimately addressed and repaired in the associated userspace tools. The kernel is still responsible to detect the problem and notify the user that something is wrong. Check the superblock LSN at mount time and fail the mount if it is invalid. From that point on, trigger verifier failure on any metadata I/O where an invalid LSN is detected. This results in a filesystem shutdown and guarantees that we do not log metadata changes with invalid LSNs on disk. Since this is a known issue with a known recovery path, present a warning to instruct the user how to recover. Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2015-10-12 12:59:25 +08:00
error = xlog_find_tail(log, &head_blk, &tail_blk);
if (error)
return error;
xfs: validate metadata LSNs against log on v5 superblocks Since the onset of v5 superblocks, the LSN of the last modification has been included in a variety of on-disk data structures. This LSN is used to provide log recovery ordering guarantees (e.g., to ensure an older log recovery item is not replayed over a newer target data structure). While this works correctly from the point a filesystem is formatted and mounted, userspace tools have some problematic behaviors that defeat this mechanism. For example, xfs_repair historically zeroes out the log unconditionally (regardless of whether corruption is detected). If this occurs, the LSN of the filesystem is reset and the log is now in a problematic state with respect to on-disk metadata structures that might have a larger LSN. Until either the log catches up to the highest previously used metadata LSN or each affected data structure is modified and written out without incident (which resets the metadata LSN), log recovery is susceptible to filesystem corruption. This problem is ultimately addressed and repaired in the associated userspace tools. The kernel is still responsible to detect the problem and notify the user that something is wrong. Check the superblock LSN at mount time and fail the mount if it is invalid. From that point on, trigger verifier failure on any metadata I/O where an invalid LSN is detected. This results in a filesystem shutdown and guarantees that we do not log metadata changes with invalid LSNs on disk. Since this is a known issue with a known recovery path, present a warning to instruct the user how to recover. Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2015-10-12 12:59:25 +08:00
/*
* The superblock was read before the log was available and thus the LSN
* could not be verified. Check the superblock LSN against the current
* LSN now that it's known.
*/
if (xfs_has_crc(log->l_mp) &&
xfs: validate metadata LSNs against log on v5 superblocks Since the onset of v5 superblocks, the LSN of the last modification has been included in a variety of on-disk data structures. This LSN is used to provide log recovery ordering guarantees (e.g., to ensure an older log recovery item is not replayed over a newer target data structure). While this works correctly from the point a filesystem is formatted and mounted, userspace tools have some problematic behaviors that defeat this mechanism. For example, xfs_repair historically zeroes out the log unconditionally (regardless of whether corruption is detected). If this occurs, the LSN of the filesystem is reset and the log is now in a problematic state with respect to on-disk metadata structures that might have a larger LSN. Until either the log catches up to the highest previously used metadata LSN or each affected data structure is modified and written out without incident (which resets the metadata LSN), log recovery is susceptible to filesystem corruption. This problem is ultimately addressed and repaired in the associated userspace tools. The kernel is still responsible to detect the problem and notify the user that something is wrong. Check the superblock LSN at mount time and fail the mount if it is invalid. From that point on, trigger verifier failure on any metadata I/O where an invalid LSN is detected. This results in a filesystem shutdown and guarantees that we do not log metadata changes with invalid LSNs on disk. Since this is a known issue with a known recovery path, present a warning to instruct the user how to recover. Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2015-10-12 12:59:25 +08:00
!xfs_log_check_lsn(log->l_mp, log->l_mp->m_sb.sb_lsn))
return -EINVAL;
if (tail_blk != head_blk) {
/* There used to be a comment here:
*
* disallow recovery on read-only mounts. note -- mount
* checks for ENOSPC and turns it into an intelligent
* error message.
* ...but this is no longer true. Now, unless you specify
* NORECOVERY (in which case this function would never be
* called), we just go ahead and recover. We do this all
* under the vfs layer, so we can get away with it unless
* the device itself is read-only, in which case we fail.
*/
if ((error = xfs_dev_is_read_only(log->l_mp, "recovery"))) {
return error;
}
/*
* Version 5 superblock log feature mask validation. We know the
* log is dirty so check if there are any unknown log features
* in what we need to recover. If there are unknown features
* (e.g. unsupported transactions, then simply reject the
* attempt at recovery before touching anything.
*/
if (xfs_sb_is_v5(&log->l_mp->m_sb) &&
xfs_sb_has_incompat_log_feature(&log->l_mp->m_sb,
XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN)) {
xfs_warn(log->l_mp,
"Superblock has unknown incompatible log features (0x%x) enabled.",
(log->l_mp->m_sb.sb_features_log_incompat &
XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN));
xfs_warn(log->l_mp,
"The log can not be fully and/or safely recovered by this kernel.");
xfs_warn(log->l_mp,
"Please recover the log on a kernel that supports the unknown features.");
return -EINVAL;
}
xfs: export log_recovery_delay to delay mount time log recovery XFS log recovery has been discovered to have race conditions with buffers when I/O errors occur. External tools are available to simulate I/O errors to XFS, but this alone is not sufficient for testing log recovery. XFS unconditionally resets the inactive region of the log prior to log recovery to avoid confusion over processing any partially written log records that might have been written before an unclean shutdown. Therefore, unconditional write I/O failures at mount time are caught by the reset sequence rather than log recovery and hinder the ability to test the latter. The device-mapper dm-flakey module uses an up/down timer to define a cycle for when to fail I/Os. Create a pre log recovery delay tunable that can be used to coordinate XFS log recovery with I/O errors simulated by dm-flakey. This facilitates coordination in userspace that allows the reset of stale log blocks to succeed and writes due to log recovery to fail. For example, define a dm-flakey instance with an uptime long enough to allow log reset to succeed and a log recovery delay long enough to allow the dm-flakey uptime to expire. The 'log_recovery_delay' sysfs tunable is exported under /sys/fs/xfs/debug and is only enabled for kernels compiled in XFS debug mode. The value is exported in units of seconds and allows for a delay of up to 60 seconds. Note that this is for XFS debug and test instrumentation purposes only and should not be used by applications. No delay is enabled by default. Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2014-09-09 09:56:13 +08:00
/*
* Delay log recovery if the debug hook is set. This is debug
* instrumentation to coordinate simulation of I/O failures with
xfs: export log_recovery_delay to delay mount time log recovery XFS log recovery has been discovered to have race conditions with buffers when I/O errors occur. External tools are available to simulate I/O errors to XFS, but this alone is not sufficient for testing log recovery. XFS unconditionally resets the inactive region of the log prior to log recovery to avoid confusion over processing any partially written log records that might have been written before an unclean shutdown. Therefore, unconditional write I/O failures at mount time are caught by the reset sequence rather than log recovery and hinder the ability to test the latter. The device-mapper dm-flakey module uses an up/down timer to define a cycle for when to fail I/Os. Create a pre log recovery delay tunable that can be used to coordinate XFS log recovery with I/O errors simulated by dm-flakey. This facilitates coordination in userspace that allows the reset of stale log blocks to succeed and writes due to log recovery to fail. For example, define a dm-flakey instance with an uptime long enough to allow log reset to succeed and a log recovery delay long enough to allow the dm-flakey uptime to expire. The 'log_recovery_delay' sysfs tunable is exported under /sys/fs/xfs/debug and is only enabled for kernels compiled in XFS debug mode. The value is exported in units of seconds and allows for a delay of up to 60 seconds. Note that this is for XFS debug and test instrumentation purposes only and should not be used by applications. No delay is enabled by default. Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2014-09-09 09:56:13 +08:00
* log recovery.
*/
if (xfs_globals.log_recovery_delay) {
xfs_notice(log->l_mp,
"Delaying log recovery for %d seconds.",
xfs_globals.log_recovery_delay);
msleep(xfs_globals.log_recovery_delay * 1000);
}
xfs_notice(log->l_mp, "Starting recovery (logdev: %s)",
log->l_mp->m_logname ? log->l_mp->m_logname
: "internal");
error = xlog_do_recover(log, head_blk, tail_blk);
set_bit(XLOG_RECOVERY_NEEDED, &log->l_opstate);
}
return error;
}
/*
* In the first part of recovery we replay inodes and buffers and build up the
* list of intents which need to be processed. Here we process the intents and
* clean up the on disk unlinked inode lists. This is separated from the first
* part of recovery so that the root and real-time bitmap inodes can be read in
* from disk in between the two stages. This is necessary so that we can free
* space in the real-time portion of the file system.
*/
int
xlog_recover_finish(
struct xlog *log)
{
int error;
error = xlog_recover_process_intents(log);
if (error) {
/*
* Cancel all the unprocessed intent items now so that we don't
* leave them pinned in the AIL. This can cause the AIL to
* livelock on the pinned item if anyone tries to push the AIL
* (inode reclaim does this) before we get around to
* xfs_log_mount_cancel.
*/
xlog_recover_cancel_intents(log);
xfs_alert(log->l_mp, "Failed to recover intents");
xfs: log shutdown triggers should only shut down the log We've got a mess on our hands. 1. xfs_trans_commit() cannot cancel transactions because the mount is shut down - that causes dirty, aborted, unlogged log items to sit unpinned in memory and potentially get written to disk before the log is shut down. Hence xfs_trans_commit() can only abort transactions when xlog_is_shutdown() is true. 2. xfs_force_shutdown() is used in places to cause the current modification to be aborted via xfs_trans_commit() because it may be impractical or impossible to cancel the transaction directly, and hence xfs_trans_commit() must cancel transactions when xfs_is_shutdown() is true in this situation. But we can't do that because of #1. 3. Log IO errors cause log shutdowns by calling xfs_force_shutdown() to shut down the mount and then the log from log IO completion. 4. xfs_force_shutdown() can result in a log force being issued, which has to wait for log IO completion before it will mark the log as shut down. If #3 races with some other shutdown trigger that runs a log force, we rely on xfs_force_shutdown() silently ignoring #3 and avoiding shutting down the log until the failed log force completes. 5. To ensure #2 always works, we have to ensure that xfs_force_shutdown() does not return until the the log is shut down. But in the case of #4, this will result in a deadlock because the log Io completion will block waiting for a log force to complete which is blocked waiting for log IO to complete.... So the very first thing we have to do here to untangle this mess is dissociate log shutdown triggers from mount shutdowns. We already have xlog_forced_shutdown, which will atomically transistion to the log a shutdown state. Due to internal asserts it cannot be called multiple times, but was done simply because the only place that could call it was xfs_do_force_shutdown() (i.e. the mount shutdown!) and that could only call it once and once only. So the first thing we do is remove the asserts. We then convert all the internal log shutdown triggers to call xlog_force_shutdown() directly instead of xfs_force_shutdown(). This allows the log shutdown triggers to shut down the log without needing to care about mount based shutdown constraints. This means we shut down the log independently of the mount and the mount may not notice this until it's next attempt to read or modify metadata. At that point (e.g. xfs_trans_commit()) it will see that the log is shutdown, error out and shutdown the mount. To ensure that all the unmount behaviours and asserts track correctly as a result of a log shutdown, propagate the shutdown up to the mount if it is not already set. This keeps the mount and log state in sync, and saves a huge amount of hassle where code fails because of a log shutdown but only checks for mount shutdowns and hence ends up doing the wrong thing. Cleaning up that mess is an exercise for another day. This enables us to address the other problems noted above in followup patches. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2022-03-30 09:22:01 +08:00
xlog_force_shutdown(log, SHUTDOWN_LOG_IO_ERROR);
return error;
}
/*
* Sync the log to get all the intents out of the AIL. This isn't
* absolutely necessary, but it helps in case the unlink transactions
* would have problems pushing the intents out of the way.
*/
xfs_log_force(log->l_mp, XFS_LOG_SYNC);
/*
* Now that we've recovered the log and all the intents, we can clear
* the log incompat feature bits in the superblock because there's no
* longer anything to protect. We rely on the AIL push to write out the
* updated superblock after everything else.
*/
if (xfs_clear_incompat_log_features(log->l_mp)) {
error = xfs_sync_sb(log->l_mp, false);
if (error < 0) {
xfs_alert(log->l_mp,
"Failed to clear log incompat features on recovery");
return error;
}
}
xlog_recover_process_iunlinks(log);
xfs: only run COW extent recovery when there are no live extents As part of multiple customer escalations due to file data corruption after copy on write operations, I wrote some fstests that use fsstress to hammer on COW to shake things loose. Regrettably, I caught some filesystem shutdowns due to incorrect rmap operations with the following loop: mount <filesystem> # (0) fsstress <run only readonly ops> & # (1) while true; do fsstress <run all ops> mount -o remount,ro # (2) fsstress <run only readonly ops> mount -o remount,rw # (3) done When (2) happens, notice that (1) is still running. xfs_remount_ro will call xfs_blockgc_stop to walk the inode cache to free all the COW extents, but the blockgc mechanism races with (1)'s reader threads to take IOLOCKs and loses, which means that it doesn't clean them all out. Call such a file (A). When (3) happens, xfs_remount_rw calls xfs_reflink_recover_cow, which walks the ondisk refcount btree and frees any COW extent that it finds. This function does not check the inode cache, which means that incore COW forks of inode (A) is now inconsistent with the ondisk metadata. If one of those former COW extents are allocated and mapped into another file (B) and someone triggers a COW to the stale reservation in (A), A's dirty data will be written into (B) and once that's done, those blocks will be transferred to (A)'s data fork without bumping the refcount. The results are catastrophic -- file (B) and the refcount btree are now corrupt. In the first patch, we fixed the race condition in (2) so that (A) will always flush the COW fork. In this second patch, we move the _recover_cow call to the initial mount call in (0) for safety. As mentioned previously, xfs_reflink_recover_cow walks the refcount btree looking for COW staging extents, and frees them. This was intended to be run at mount time (when we know there are no live inodes) to clean up any leftover staging events that may have been left behind during an unclean shutdown. As a time "optimization" for readonly mounts, we deferred this to the ro->rw transition, not realizing that any failure to clean all COW forks during a rw->ro transition would result in catastrophic corruption. Therefore, remove this optimization and only run the recovery routine when we're guaranteed not to have any COW staging extents anywhere, which means we always run this at mount time. While we're at it, move the callsite to xfs_log_mount_finish because any refcount btree expansion (however unlikely given that we're removing records from the right side of the index) must be fed by a per-AG reservation, which doesn't exist in its current location. Fixes: 174edb0e46e5 ("xfs: store in-progress CoW allocations in the refcount btree") Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Chandan Babu R <chandan.babu@oracle.com> Reviewed-by: Dave Chinner <dchinner@redhat.com>
2021-12-16 03:52:23 +08:00
/*
* Recover any CoW staging blocks that are still referenced by the
* ondisk refcount metadata. During mount there cannot be any live
* staging extents as we have not permitted any user modifications.
* Therefore, it is safe to free them all right now, even on a
* read-only mount.
*/
error = xfs_reflink_recover_cow(log->l_mp);
if (error) {
xfs_alert(log->l_mp,
"Failed to recover leftover CoW staging extents, err %d.",
error);
/*
* If we get an error here, make sure the log is shut down
* but return zero so that any log items committed since the
* end of intents processing can be pushed through the CIL
* and AIL.
*/
xfs: log shutdown triggers should only shut down the log We've got a mess on our hands. 1. xfs_trans_commit() cannot cancel transactions because the mount is shut down - that causes dirty, aborted, unlogged log items to sit unpinned in memory and potentially get written to disk before the log is shut down. Hence xfs_trans_commit() can only abort transactions when xlog_is_shutdown() is true. 2. xfs_force_shutdown() is used in places to cause the current modification to be aborted via xfs_trans_commit() because it may be impractical or impossible to cancel the transaction directly, and hence xfs_trans_commit() must cancel transactions when xfs_is_shutdown() is true in this situation. But we can't do that because of #1. 3. Log IO errors cause log shutdowns by calling xfs_force_shutdown() to shut down the mount and then the log from log IO completion. 4. xfs_force_shutdown() can result in a log force being issued, which has to wait for log IO completion before it will mark the log as shut down. If #3 races with some other shutdown trigger that runs a log force, we rely on xfs_force_shutdown() silently ignoring #3 and avoiding shutting down the log until the failed log force completes. 5. To ensure #2 always works, we have to ensure that xfs_force_shutdown() does not return until the the log is shut down. But in the case of #4, this will result in a deadlock because the log Io completion will block waiting for a log force to complete which is blocked waiting for log IO to complete.... So the very first thing we have to do here to untangle this mess is dissociate log shutdown triggers from mount shutdowns. We already have xlog_forced_shutdown, which will atomically transistion to the log a shutdown state. Due to internal asserts it cannot be called multiple times, but was done simply because the only place that could call it was xfs_do_force_shutdown() (i.e. the mount shutdown!) and that could only call it once and once only. So the first thing we do is remove the asserts. We then convert all the internal log shutdown triggers to call xlog_force_shutdown() directly instead of xfs_force_shutdown(). This allows the log shutdown triggers to shut down the log without needing to care about mount based shutdown constraints. This means we shut down the log independently of the mount and the mount may not notice this until it's next attempt to read or modify metadata. At that point (e.g. xfs_trans_commit()) it will see that the log is shutdown, error out and shutdown the mount. To ensure that all the unmount behaviours and asserts track correctly as a result of a log shutdown, propagate the shutdown up to the mount if it is not already set. This keeps the mount and log state in sync, and saves a huge amount of hassle where code fails because of a log shutdown but only checks for mount shutdowns and hence ends up doing the wrong thing. Cleaning up that mess is an exercise for another day. This enables us to address the other problems noted above in followup patches. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2022-03-30 09:22:01 +08:00
xlog_force_shutdown(log, SHUTDOWN_LOG_IO_ERROR);
xfs: only run COW extent recovery when there are no live extents As part of multiple customer escalations due to file data corruption after copy on write operations, I wrote some fstests that use fsstress to hammer on COW to shake things loose. Regrettably, I caught some filesystem shutdowns due to incorrect rmap operations with the following loop: mount <filesystem> # (0) fsstress <run only readonly ops> & # (1) while true; do fsstress <run all ops> mount -o remount,ro # (2) fsstress <run only readonly ops> mount -o remount,rw # (3) done When (2) happens, notice that (1) is still running. xfs_remount_ro will call xfs_blockgc_stop to walk the inode cache to free all the COW extents, but the blockgc mechanism races with (1)'s reader threads to take IOLOCKs and loses, which means that it doesn't clean them all out. Call such a file (A). When (3) happens, xfs_remount_rw calls xfs_reflink_recover_cow, which walks the ondisk refcount btree and frees any COW extent that it finds. This function does not check the inode cache, which means that incore COW forks of inode (A) is now inconsistent with the ondisk metadata. If one of those former COW extents are allocated and mapped into another file (B) and someone triggers a COW to the stale reservation in (A), A's dirty data will be written into (B) and once that's done, those blocks will be transferred to (A)'s data fork without bumping the refcount. The results are catastrophic -- file (B) and the refcount btree are now corrupt. In the first patch, we fixed the race condition in (2) so that (A) will always flush the COW fork. In this second patch, we move the _recover_cow call to the initial mount call in (0) for safety. As mentioned previously, xfs_reflink_recover_cow walks the refcount btree looking for COW staging extents, and frees them. This was intended to be run at mount time (when we know there are no live inodes) to clean up any leftover staging events that may have been left behind during an unclean shutdown. As a time "optimization" for readonly mounts, we deferred this to the ro->rw transition, not realizing that any failure to clean all COW forks during a rw->ro transition would result in catastrophic corruption. Therefore, remove this optimization and only run the recovery routine when we're guaranteed not to have any COW staging extents anywhere, which means we always run this at mount time. While we're at it, move the callsite to xfs_log_mount_finish because any refcount btree expansion (however unlikely given that we're removing records from the right side of the index) must be fed by a per-AG reservation, which doesn't exist in its current location. Fixes: 174edb0e46e5 ("xfs: store in-progress CoW allocations in the refcount btree") Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Chandan Babu R <chandan.babu@oracle.com> Reviewed-by: Dave Chinner <dchinner@redhat.com>
2021-12-16 03:52:23 +08:00
}
return 0;
}
void
xlog_recover_cancel(
struct xlog *log)
{
if (xlog_recovery_needed(log))
xlog_recover_cancel_intents(log);
}