linux/fs/gfs2/log.c

1352 lines
37 KiB
C
Raw Normal View History

// SPDX-License-Identifier: GPL-2.0-only
/*
* Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
* Copyright (C) 2004-2007 Red Hat, Inc. All rights reserved.
*/
#include <linux/sched.h>
#include <linux/slab.h>
#include <linux/spinlock.h>
#include <linux/completion.h>
#include <linux/buffer_head.h>
#include <linux/gfs2_ondisk.h>
#include <linux/crc32.h>
#include <linux/crc32c.h>
#include <linux/delay.h>
#include <linux/kthread.h>
#include <linux/freezer.h>
#include <linux/bio.h>
#include <linux/blkdev.h>
#include <linux/writeback.h>
#include <linux/list_sort.h>
#include "gfs2.h"
#include "incore.h"
#include "bmap.h"
#include "glock.h"
#include "log.h"
#include "lops.h"
#include "meta_io.h"
#include "util.h"
#include "dir.h"
#include "trace_gfs2.h"
#include "trans.h"
static void gfs2_log_shutdown(struct gfs2_sbd *sdp);
/**
* gfs2_struct2blk - compute stuff
* @sdp: the filesystem
* @nstruct: the number of structures
*
* Compute the number of log descriptor blocks needed to hold a certain number
* of structures of a certain size.
*
* Returns: the number of blocks needed (minimum is always 1)
*/
unsigned int gfs2_struct2blk(struct gfs2_sbd *sdp, unsigned int nstruct)
{
unsigned int blks;
unsigned int first, second;
/* The initial struct gfs2_log_descriptor block */
blks = 1;
first = sdp->sd_ldptrs;
if (nstruct > first) {
/* Subsequent struct gfs2_meta_header blocks */
second = sdp->sd_inptrs;
blks += DIV_ROUND_UP(nstruct - first, second);
}
return blks;
}
/**
* gfs2_remove_from_ail - Remove an entry from the ail lists, updating counters
* @bd: The gfs2_bufdata to remove
*
* The ail lock _must_ be held when calling this function
*
*/
void gfs2_remove_from_ail(struct gfs2_bufdata *bd)
{
bd->bd_tr = NULL;
list_del_init(&bd->bd_ail_st_list);
list_del_init(&bd->bd_ail_gl_list);
atomic_dec(&bd->bd_gl->gl_ail_count);
brelse(bd->bd_bh);
}
/**
* gfs2_ail1_start_one - Start I/O on a transaction
* @sdp: The superblock
* @wbc: The writeback control structure
* @tr: The transaction to start I/O on
* @plug: The block plug currently active
*/
static int gfs2_ail1_start_one(struct gfs2_sbd *sdp,
struct writeback_control *wbc,
struct gfs2_trans *tr, struct blk_plug *plug)
__releases(&sdp->sd_ail_lock)
__acquires(&sdp->sd_ail_lock)
{
struct gfs2_glock *gl = NULL;
struct address_space *mapping;
struct gfs2_bufdata *bd, *s;
struct buffer_head *bh;
int ret = 0;
list_for_each_entry_safe_reverse(bd, s, &tr->tr_ail1_list, bd_ail_st_list) {
bh = bd->bd_bh;
gfs2_assert(sdp, bd->bd_tr == tr);
if (!buffer_busy(bh)) {
if (buffer_uptodate(bh)) {
list_move(&bd->bd_ail_st_list,
&tr->tr_ail2_list);
continue;
}
if (!cmpxchg(&sdp->sd_log_error, 0, -EIO)) {
gfs2_io_error_bh(sdp, bh);
gfs2_withdraw_delayed(sdp);
}
}
if (gfs2_withdrawn(sdp)) {
gfs2_remove_from_ail(bd);
continue;
}
if (!buffer_dirty(bh))
continue;
if (gl == bd->bd_gl)
continue;
gl = bd->bd_gl;
list_move(&bd->bd_ail_st_list, &tr->tr_ail1_list);
mapping = bh->b_page->mapping;
if (!mapping)
continue;
spin_unlock(&sdp->sd_ail_lock);
ret = generic_writepages(mapping, wbc);
if (need_resched()) {
blk_finish_plug(plug);
cond_resched();
blk_start_plug(plug);
}
spin_lock(&sdp->sd_ail_lock);
if (ret == -ENODATA) /* if a jdata write into a new hole */
ret = 0; /* ignore it */
if (ret || wbc->nr_to_write <= 0)
break;
return -EBUSY;
}
return ret;
}
static void dump_ail_list(struct gfs2_sbd *sdp)
{
struct gfs2_trans *tr;
struct gfs2_bufdata *bd;
struct buffer_head *bh;
list_for_each_entry_reverse(tr, &sdp->sd_ail1_list, tr_list) {
list_for_each_entry_reverse(bd, &tr->tr_ail1_list,
bd_ail_st_list) {
bh = bd->bd_bh;
fs_err(sdp, "bd %p: blk:0x%llx bh=%p ", bd,
(unsigned long long)bd->bd_blkno, bh);
if (!bh) {
fs_err(sdp, "\n");
continue;
}
fs_err(sdp, "0x%llx up2:%d dirt:%d lkd:%d req:%d "
"map:%d new:%d ar:%d aw:%d delay:%d "
"io err:%d unwritten:%d dfr:%d pin:%d esc:%d\n",
(unsigned long long)bh->b_blocknr,
buffer_uptodate(bh), buffer_dirty(bh),
buffer_locked(bh), buffer_req(bh),
buffer_mapped(bh), buffer_new(bh),
buffer_async_read(bh), buffer_async_write(bh),
buffer_delay(bh), buffer_write_io_error(bh),
buffer_unwritten(bh),
buffer_defer_completion(bh),
buffer_pinned(bh), buffer_escaped(bh));
}
}
}
/**
* gfs2_ail1_flush - start writeback of some ail1 entries
* @sdp: The super block
* @wbc: The writeback control structure
*
* Writes back some ail1 entries, according to the limits in the
* writeback control structure
*/
void gfs2_ail1_flush(struct gfs2_sbd *sdp, struct writeback_control *wbc)
{
struct list_head *head = &sdp->sd_ail1_list;
struct gfs2_trans *tr;
struct blk_plug plug;
int ret;
unsigned long flush_start = jiffies;
trace_gfs2_ail_flush(sdp, wbc, 1);
blk_start_plug(&plug);
spin_lock(&sdp->sd_ail_lock);
restart:
ret = 0;
if (time_after(jiffies, flush_start + (HZ * 600))) {
fs_err(sdp, "Error: In %s for ten minutes! t=%d\n",
__func__, current->journal_info ? 1 : 0);
dump_ail_list(sdp);
goto out;
}
list_for_each_entry_reverse(tr, head, tr_list) {
if (wbc->nr_to_write <= 0)
break;
ret = gfs2_ail1_start_one(sdp, wbc, tr, &plug);
if (ret) {
if (ret == -EBUSY)
goto restart;
break;
}
}
out:
spin_unlock(&sdp->sd_ail_lock);
blk_finish_plug(&plug);
if (ret) {
gfs2_lm(sdp, "gfs2_ail1_start_one (generic_writepages) "
"returned: %d\n", ret);
gfs2_withdraw(sdp);
}
trace_gfs2_ail_flush(sdp, wbc, 0);
}
/**
* gfs2_ail1_start - start writeback of all ail1 entries
* @sdp: The superblock
*/
static void gfs2_ail1_start(struct gfs2_sbd *sdp)
{
struct writeback_control wbc = {
.sync_mode = WB_SYNC_NONE,
.nr_to_write = LONG_MAX,
.range_start = 0,
.range_end = LLONG_MAX,
};
return gfs2_ail1_flush(sdp, &wbc);
}
static void gfs2_log_update_flush_tail(struct gfs2_sbd *sdp)
{
unsigned int new_flush_tail = sdp->sd_log_head;
struct gfs2_trans *tr;
if (!list_empty(&sdp->sd_ail1_list)) {
tr = list_last_entry(&sdp->sd_ail1_list,
struct gfs2_trans, tr_list);
new_flush_tail = tr->tr_first;
}
sdp->sd_log_flush_tail = new_flush_tail;
}
static void gfs2_log_update_head(struct gfs2_sbd *sdp)
{
unsigned int new_head = sdp->sd_log_flush_head;
if (sdp->sd_log_flush_tail == sdp->sd_log_head)
sdp->sd_log_flush_tail = new_head;
sdp->sd_log_head = new_head;
}
/*
* gfs2_ail_empty_tr - empty one of the ail lists of a transaction
*/
static void gfs2_ail_empty_tr(struct gfs2_sbd *sdp, struct gfs2_trans *tr,
struct list_head *head)
{
struct gfs2_bufdata *bd;
while (!list_empty(head)) {
bd = list_first_entry(head, struct gfs2_bufdata,
bd_ail_st_list);
gfs2_assert(sdp, bd->bd_tr == tr);
gfs2_remove_from_ail(bd);
}
}
/**
* gfs2_ail1_empty_one - Check whether or not a trans in the AIL has been synced
* @sdp: the filesystem
gfs2: Issue revokes more intelligently Before this patch, function gfs2_write_revokes would call gfs2_ail1_empty, then traverse the sd_ail1_list looking for transactions that had bds which were no longer queued to a glock. And if it found some, it would try to issue revokes for them, up to a predetermined maximum. There were two problems with how it did this. First was the fact that gfs2_ail1_empty moves transactions which have nothing remaining on the ail1 list from the sd_ail1_list to the sd_ail2_list, thus making its traversal of sd_ail1_list miss them completely, and therefore, never issue revokes for them. Second was the fact that there were three traversals (or partial traversals) of the sd_ail1_list, each of which took and then released the sd_ail_lock lock: First inside gfs2_ail1_empty, second to determine if there are any revokes to be issued, and third to actually issue them. All this taking and releasing of the sd_ail_lock meant other processes could modify the lists and the conditions in which we're working. This patch simplies the whole process by adding a new parameter to function gfs2_ail1_empty, max_revokes. For normal calls, this is passed in as 0, meaning we don't want to issue any revokes. For function gfs2_write_revokes, we pass in the maximum number of revokes we can, thus allowing gfs2_ail1_empty to add the revokes where needed. This simplies the code, allows for a single holding of the sd_ail_lock, and allows gfs2_ail1_empty to add revokes for all the necessary bd items without missing any. Signed-off-by: Bob Peterson <rpeterso@redhat.com> Reviewed-by: Andreas Gruenbacher <agruenba@redhat.com>
2019-02-22 05:28:07 +08:00
* @tr: the transaction
* @max_revokes: If nonzero, issue revokes for the bd items for written buffers
*
* returns: the transaction's count of remaining active items
*/
static int gfs2_ail1_empty_one(struct gfs2_sbd *sdp, struct gfs2_trans *tr,
gfs2: Issue revokes more intelligently Before this patch, function gfs2_write_revokes would call gfs2_ail1_empty, then traverse the sd_ail1_list looking for transactions that had bds which were no longer queued to a glock. And if it found some, it would try to issue revokes for them, up to a predetermined maximum. There were two problems with how it did this. First was the fact that gfs2_ail1_empty moves transactions which have nothing remaining on the ail1 list from the sd_ail1_list to the sd_ail2_list, thus making its traversal of sd_ail1_list miss them completely, and therefore, never issue revokes for them. Second was the fact that there were three traversals (or partial traversals) of the sd_ail1_list, each of which took and then released the sd_ail_lock lock: First inside gfs2_ail1_empty, second to determine if there are any revokes to be issued, and third to actually issue them. All this taking and releasing of the sd_ail_lock meant other processes could modify the lists and the conditions in which we're working. This patch simplies the whole process by adding a new parameter to function gfs2_ail1_empty, max_revokes. For normal calls, this is passed in as 0, meaning we don't want to issue any revokes. For function gfs2_write_revokes, we pass in the maximum number of revokes we can, thus allowing gfs2_ail1_empty to add the revokes where needed. This simplies the code, allows for a single holding of the sd_ail_lock, and allows gfs2_ail1_empty to add revokes for all the necessary bd items without missing any. Signed-off-by: Bob Peterson <rpeterso@redhat.com> Reviewed-by: Andreas Gruenbacher <agruenba@redhat.com>
2019-02-22 05:28:07 +08:00
int *max_revokes)
{
struct gfs2_bufdata *bd, *s;
struct buffer_head *bh;
int active_count = 0;
list_for_each_entry_safe_reverse(bd, s, &tr->tr_ail1_list,
bd_ail_st_list) {
bh = bd->bd_bh;
gfs2_assert(sdp, bd->bd_tr == tr);
/*
* If another process flagged an io error, e.g. writing to the
* journal, error all other bhs and move them off the ail1 to
* prevent a tight loop when unmount tries to flush ail1,
* regardless of whether they're still busy. If no outside
* errors were found and the buffer is busy, move to the next.
* If the ail buffer is not busy and caught an error, flag it
* for others.
*/
if (!sdp->sd_log_error && buffer_busy(bh)) {
active_count++;
continue;
}
if (!buffer_uptodate(bh) &&
!cmpxchg(&sdp->sd_log_error, 0, -EIO)) {
gfs2_io_error_bh(sdp, bh);
gfs2_withdraw_delayed(sdp);
}
gfs2: Issue revokes more intelligently Before this patch, function gfs2_write_revokes would call gfs2_ail1_empty, then traverse the sd_ail1_list looking for transactions that had bds which were no longer queued to a glock. And if it found some, it would try to issue revokes for them, up to a predetermined maximum. There were two problems with how it did this. First was the fact that gfs2_ail1_empty moves transactions which have nothing remaining on the ail1 list from the sd_ail1_list to the sd_ail2_list, thus making its traversal of sd_ail1_list miss them completely, and therefore, never issue revokes for them. Second was the fact that there were three traversals (or partial traversals) of the sd_ail1_list, each of which took and then released the sd_ail_lock lock: First inside gfs2_ail1_empty, second to determine if there are any revokes to be issued, and third to actually issue them. All this taking and releasing of the sd_ail_lock meant other processes could modify the lists and the conditions in which we're working. This patch simplies the whole process by adding a new parameter to function gfs2_ail1_empty, max_revokes. For normal calls, this is passed in as 0, meaning we don't want to issue any revokes. For function gfs2_write_revokes, we pass in the maximum number of revokes we can, thus allowing gfs2_ail1_empty to add the revokes where needed. This simplies the code, allows for a single holding of the sd_ail_lock, and allows gfs2_ail1_empty to add revokes for all the necessary bd items without missing any. Signed-off-by: Bob Peterson <rpeterso@redhat.com> Reviewed-by: Andreas Gruenbacher <agruenba@redhat.com>
2019-02-22 05:28:07 +08:00
/*
* If we have space for revokes and the bd is no longer on any
* buf list, we can just add a revoke for it immediately and
* avoid having to put it on the ail2 list, where it would need
* to be revoked later.
*/
if (*max_revokes && list_empty(&bd->bd_list)) {
gfs2_add_revoke(sdp, bd);
(*max_revokes)--;
continue;
}
list_move(&bd->bd_ail_st_list, &tr->tr_ail2_list);
}
return active_count;
}
/**
* gfs2_ail1_empty - Try to empty the ail1 lists
* @sdp: The superblock
gfs2: Issue revokes more intelligently Before this patch, function gfs2_write_revokes would call gfs2_ail1_empty, then traverse the sd_ail1_list looking for transactions that had bds which were no longer queued to a glock. And if it found some, it would try to issue revokes for them, up to a predetermined maximum. There were two problems with how it did this. First was the fact that gfs2_ail1_empty moves transactions which have nothing remaining on the ail1 list from the sd_ail1_list to the sd_ail2_list, thus making its traversal of sd_ail1_list miss them completely, and therefore, never issue revokes for them. Second was the fact that there were three traversals (or partial traversals) of the sd_ail1_list, each of which took and then released the sd_ail_lock lock: First inside gfs2_ail1_empty, second to determine if there are any revokes to be issued, and third to actually issue them. All this taking and releasing of the sd_ail_lock meant other processes could modify the lists and the conditions in which we're working. This patch simplies the whole process by adding a new parameter to function gfs2_ail1_empty, max_revokes. For normal calls, this is passed in as 0, meaning we don't want to issue any revokes. For function gfs2_write_revokes, we pass in the maximum number of revokes we can, thus allowing gfs2_ail1_empty to add the revokes where needed. This simplies the code, allows for a single holding of the sd_ail_lock, and allows gfs2_ail1_empty to add revokes for all the necessary bd items without missing any. Signed-off-by: Bob Peterson <rpeterso@redhat.com> Reviewed-by: Andreas Gruenbacher <agruenba@redhat.com>
2019-02-22 05:28:07 +08:00
* @max_revokes: If non-zero, add revokes where appropriate
*
* Tries to empty the ail1 lists, starting with the oldest first
*/
gfs2: Issue revokes more intelligently Before this patch, function gfs2_write_revokes would call gfs2_ail1_empty, then traverse the sd_ail1_list looking for transactions that had bds which were no longer queued to a glock. And if it found some, it would try to issue revokes for them, up to a predetermined maximum. There were two problems with how it did this. First was the fact that gfs2_ail1_empty moves transactions which have nothing remaining on the ail1 list from the sd_ail1_list to the sd_ail2_list, thus making its traversal of sd_ail1_list miss them completely, and therefore, never issue revokes for them. Second was the fact that there were three traversals (or partial traversals) of the sd_ail1_list, each of which took and then released the sd_ail_lock lock: First inside gfs2_ail1_empty, second to determine if there are any revokes to be issued, and third to actually issue them. All this taking and releasing of the sd_ail_lock meant other processes could modify the lists and the conditions in which we're working. This patch simplies the whole process by adding a new parameter to function gfs2_ail1_empty, max_revokes. For normal calls, this is passed in as 0, meaning we don't want to issue any revokes. For function gfs2_write_revokes, we pass in the maximum number of revokes we can, thus allowing gfs2_ail1_empty to add the revokes where needed. This simplies the code, allows for a single holding of the sd_ail_lock, and allows gfs2_ail1_empty to add revokes for all the necessary bd items without missing any. Signed-off-by: Bob Peterson <rpeterso@redhat.com> Reviewed-by: Andreas Gruenbacher <agruenba@redhat.com>
2019-02-22 05:28:07 +08:00
static int gfs2_ail1_empty(struct gfs2_sbd *sdp, int max_revokes)
{
struct gfs2_trans *tr, *s;
int oldest_tr = 1;
int ret;
spin_lock(&sdp->sd_ail_lock);
list_for_each_entry_safe_reverse(tr, s, &sdp->sd_ail1_list, tr_list) {
if (!gfs2_ail1_empty_one(sdp, tr, &max_revokes) && oldest_tr)
list_move(&tr->tr_list, &sdp->sd_ail2_list);
else
oldest_tr = 0;
}
gfs2_log_update_flush_tail(sdp);
ret = list_empty(&sdp->sd_ail1_list);
spin_unlock(&sdp->sd_ail_lock);
if (test_bit(SDF_WITHDRAWING, &sdp->sd_flags)) {
gfs2_lm(sdp, "fatal: I/O error(s)\n");
gfs2_withdraw(sdp);
}
return ret;
}
static void gfs2_ail1_wait(struct gfs2_sbd *sdp)
{
struct gfs2_trans *tr;
struct gfs2_bufdata *bd;
struct buffer_head *bh;
spin_lock(&sdp->sd_ail_lock);
list_for_each_entry_reverse(tr, &sdp->sd_ail1_list, tr_list) {
list_for_each_entry(bd, &tr->tr_ail1_list, bd_ail_st_list) {
bh = bd->bd_bh;
if (!buffer_locked(bh))
continue;
get_bh(bh);
spin_unlock(&sdp->sd_ail_lock);
wait_on_buffer(bh);
brelse(bh);
return;
}
}
spin_unlock(&sdp->sd_ail_lock);
}
static void __ail2_empty(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
{
gfs2_ail_empty_tr(sdp, tr, &tr->tr_ail2_list);
list_del(&tr->tr_list);
gfs2_assert_warn(sdp, list_empty(&tr->tr_ail1_list));
gfs2_assert_warn(sdp, list_empty(&tr->tr_ail2_list));
gfs2_trans_free(sdp, tr);
}
static void ail2_empty(struct gfs2_sbd *sdp, unsigned int new_tail)
{
struct list_head *ail2_list = &sdp->sd_ail2_list;
unsigned int old_tail = sdp->sd_log_tail;
struct gfs2_trans *tr, *safe;
spin_lock(&sdp->sd_ail_lock);
if (old_tail <= new_tail) {
list_for_each_entry_safe(tr, safe, ail2_list, tr_list) {
if (old_tail <= tr->tr_first && tr->tr_first < new_tail)
__ail2_empty(sdp, tr);
}
} else {
list_for_each_entry_safe(tr, safe, ail2_list, tr_list) {
if (old_tail <= tr->tr_first || tr->tr_first < new_tail)
__ail2_empty(sdp, tr);
}
}
spin_unlock(&sdp->sd_ail_lock);
}
/**
* gfs2_log_is_empty - Check if the log is empty
* @sdp: The GFS2 superblock
*/
bool gfs2_log_is_empty(struct gfs2_sbd *sdp) {
return atomic_read(&sdp->sd_log_blks_free) == sdp->sd_jdesc->jd_blocks;
}
static bool __gfs2_log_try_reserve_revokes(struct gfs2_sbd *sdp, unsigned int revokes)
{
unsigned int available;
available = atomic_read(&sdp->sd_log_revokes_available);
while (available >= revokes) {
if (atomic_try_cmpxchg(&sdp->sd_log_revokes_available,
&available, available - revokes))
return true;
}
return false;
}
/**
* gfs2_log_release_revokes - Release a given number of revokes
* @sdp: The GFS2 superblock
* @revokes: The number of revokes to release
*
* sdp->sd_log_flush_lock must be held.
*/
void gfs2_log_release_revokes(struct gfs2_sbd *sdp, unsigned int revokes)
{
if (revokes)
atomic_add(revokes, &sdp->sd_log_revokes_available);
}
GFS2: remove transaction glock GFS2 has a transaction glock, which must be grabbed for every transaction, whose purpose is to deal with freezing the filesystem. Aside from this involving a large amount of locking, it is very easy to make the current fsfreeze code hang on unfreezing. This patch rewrites how gfs2 handles freezing the filesystem. The transaction glock is removed. In it's place is a freeze glock, which is cached (but not held) in a shared state by every node in the cluster when the filesystem is mounted. This lock only needs to be grabbed on freezing, and actions which need to be safe from freezing, like recovery. When a node wants to freeze the filesystem, it grabs this glock exclusively. When the freeze glock state changes on the nodes (either from shared to unlocked, or shared to exclusive), the filesystem does a special log flush. gfs2_log_flush() does all the work for flushing out the and shutting down the incore log, and then it tries to grab the freeze glock in a shared state again. Since the filesystem is stuck in gfs2_log_flush, no new transaction can start, and nothing can be written to disk. Unfreezing the filesytem simply involes dropping the freeze glock, allowing gfs2_log_flush() to grab and then release the shared lock, so it is cached for next time. However, in order for the unfreezing ioctl to occur, gfs2 needs to get a shared lock on the filesystem root directory inode to check permissions. If that glock has already been grabbed exclusively, fsfreeze will be unable to get the shared lock and unfreeze the filesystem. In order to allow the unfreeze, this patch makes gfs2 grab a shared lock on the filesystem root directory during the freeze, and hold it until it unfreezes the filesystem. The functions which need to grab a shared lock in order to allow the unfreeze ioctl to be issued now use the lock grabbed by the freeze code instead. The freeze and unfreeze code take care to make sure that this shared lock will not be dropped while another process is using it. Signed-off-by: Benjamin Marzinski <bmarzins@redhat.com> Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
2014-05-02 11:26:55 +08:00
/**
* gfs2_log_release - Release a given number of log blocks
* @sdp: The GFS2 superblock
* @blks: The number of blocks
*
*/
void gfs2_log_release(struct gfs2_sbd *sdp, unsigned int blks)
{
atomic_add(blks, &sdp->sd_log_blks_free);
trace_gfs2_log_blocks(sdp, blks);
gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <=
sdp->sd_jdesc->jd_blocks);
if (atomic_read(&sdp->sd_log_blks_needed))
wake_up(&sdp->sd_log_waitq);
GFS2: remove transaction glock GFS2 has a transaction glock, which must be grabbed for every transaction, whose purpose is to deal with freezing the filesystem. Aside from this involving a large amount of locking, it is very easy to make the current fsfreeze code hang on unfreezing. This patch rewrites how gfs2 handles freezing the filesystem. The transaction glock is removed. In it's place is a freeze glock, which is cached (but not held) in a shared state by every node in the cluster when the filesystem is mounted. This lock only needs to be grabbed on freezing, and actions which need to be safe from freezing, like recovery. When a node wants to freeze the filesystem, it grabs this glock exclusively. When the freeze glock state changes on the nodes (either from shared to unlocked, or shared to exclusive), the filesystem does a special log flush. gfs2_log_flush() does all the work for flushing out the and shutting down the incore log, and then it tries to grab the freeze glock in a shared state again. Since the filesystem is stuck in gfs2_log_flush, no new transaction can start, and nothing can be written to disk. Unfreezing the filesytem simply involes dropping the freeze glock, allowing gfs2_log_flush() to grab and then release the shared lock, so it is cached for next time. However, in order for the unfreezing ioctl to occur, gfs2 needs to get a shared lock on the filesystem root directory inode to check permissions. If that glock has already been grabbed exclusively, fsfreeze will be unable to get the shared lock and unfreeze the filesystem. In order to allow the unfreeze, this patch makes gfs2 grab a shared lock on the filesystem root directory during the freeze, and hold it until it unfreezes the filesystem. The functions which need to grab a shared lock in order to allow the unfreeze ioctl to be issued now use the lock grabbed by the freeze code instead. The freeze and unfreeze code take care to make sure that this shared lock will not be dropped while another process is using it. Signed-off-by: Benjamin Marzinski <bmarzins@redhat.com> Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
2014-05-02 11:26:55 +08:00
}
/**
* __gfs2_log_try_reserve - Try to make a log reservation
* @sdp: The GFS2 superblock
* @blks: The number of blocks to reserve
* @taboo_blks: The number of blocks to leave free
*
* Try to do the same as __gfs2_log_reserve(), but fail if no more log
* space is immediately available.
*/
static bool __gfs2_log_try_reserve(struct gfs2_sbd *sdp, unsigned int blks,
unsigned int taboo_blks)
{
unsigned wanted = blks + taboo_blks;
unsigned int free_blocks;
free_blocks = atomic_read(&sdp->sd_log_blks_free);
while (free_blocks >= wanted) {
if (atomic_try_cmpxchg(&sdp->sd_log_blks_free, &free_blocks,
free_blocks - blks)) {
trace_gfs2_log_blocks(sdp, -blks);
return true;
}
}
return false;
}
/**
* __gfs2_log_reserve - Make a log reservation
* @sdp: The GFS2 superblock
* @blks: The number of blocks to reserve
* @taboo_blks: The number of blocks to leave free
*
* @taboo_blks is set to 0 for logd, and to GFS2_LOG_FLUSH_MIN_BLOCKS
* for all other processes. This ensures that when the log is almost full,
* logd will still be able to call gfs2_log_flush one more time without
* blocking, which will advance the tail and make some more log space
* available.
*
* We no longer flush the log here, instead we wake up logd to do that
* for us. To avoid the thundering herd and to ensure that we deal fairly
* with queued waiters, we use an exclusive wait. This means that when we
* get woken with enough journal space to get our reservation, we need to
* wake the next waiter on the list.
*/
static void __gfs2_log_reserve(struct gfs2_sbd *sdp, unsigned int blks,
unsigned int taboo_blks)
{
unsigned wanted = blks + taboo_blks;
unsigned int free_blocks;
atomic_add(blks, &sdp->sd_log_blks_needed);
for (;;) {
if (current != sdp->sd_logd_process)
wake_up(&sdp->sd_logd_waitq);
io_wait_event(sdp->sd_log_waitq,
(free_blocks = atomic_read(&sdp->sd_log_blks_free),
free_blocks >= wanted));
do {
if (atomic_try_cmpxchg(&sdp->sd_log_blks_free,
&free_blocks,
free_blocks - blks))
goto reserved;
} while (free_blocks >= wanted);
}
reserved:
trace_gfs2_log_blocks(sdp, -blks);
if (atomic_sub_return(blks, &sdp->sd_log_blks_needed))
wake_up(&sdp->sd_log_waitq);
}
/**
* gfs2_log_try_reserve - Try to make a log reservation
* @sdp: The GFS2 superblock
* @tr: The transaction
* @extra_revokes: The number of additional revokes reserved (output)
*
* This is similar to gfs2_log_reserve, but sdp->sd_log_flush_lock must be
* held for correct revoke accounting.
*/
bool gfs2_log_try_reserve(struct gfs2_sbd *sdp, struct gfs2_trans *tr,
unsigned int *extra_revokes)
{
unsigned int blks = tr->tr_reserved;
unsigned int revokes = tr->tr_revokes;
unsigned int revoke_blks = 0;
*extra_revokes = 0;
if (revokes && !__gfs2_log_try_reserve_revokes(sdp, revokes)) {
revoke_blks = DIV_ROUND_UP(revokes, sdp->sd_inptrs);
*extra_revokes = revoke_blks * sdp->sd_inptrs - revokes;
blks += revoke_blks;
GFS2: remove transaction glock GFS2 has a transaction glock, which must be grabbed for every transaction, whose purpose is to deal with freezing the filesystem. Aside from this involving a large amount of locking, it is very easy to make the current fsfreeze code hang on unfreezing. This patch rewrites how gfs2 handles freezing the filesystem. The transaction glock is removed. In it's place is a freeze glock, which is cached (but not held) in a shared state by every node in the cluster when the filesystem is mounted. This lock only needs to be grabbed on freezing, and actions which need to be safe from freezing, like recovery. When a node wants to freeze the filesystem, it grabs this glock exclusively. When the freeze glock state changes on the nodes (either from shared to unlocked, or shared to exclusive), the filesystem does a special log flush. gfs2_log_flush() does all the work for flushing out the and shutting down the incore log, and then it tries to grab the freeze glock in a shared state again. Since the filesystem is stuck in gfs2_log_flush, no new transaction can start, and nothing can be written to disk. Unfreezing the filesytem simply involes dropping the freeze glock, allowing gfs2_log_flush() to grab and then release the shared lock, so it is cached for next time. However, in order for the unfreezing ioctl to occur, gfs2 needs to get a shared lock on the filesystem root directory inode to check permissions. If that glock has already been grabbed exclusively, fsfreeze will be unable to get the shared lock and unfreeze the filesystem. In order to allow the unfreeze, this patch makes gfs2 grab a shared lock on the filesystem root directory during the freeze, and hold it until it unfreezes the filesystem. The functions which need to grab a shared lock in order to allow the unfreeze ioctl to be issued now use the lock grabbed by the freeze code instead. The freeze and unfreeze code take care to make sure that this shared lock will not be dropped while another process is using it. Signed-off-by: Benjamin Marzinski <bmarzins@redhat.com> Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
2014-05-02 11:26:55 +08:00
}
if (!blks)
return true;
if (__gfs2_log_try_reserve(sdp, blks, GFS2_LOG_FLUSH_MIN_BLOCKS))
return true;
if (!revoke_blks)
gfs2_log_release_revokes(sdp, revokes);
return false;
}
/**
* gfs2_log_reserve - Make a log reservation
* @sdp: The GFS2 superblock
* @tr: The transaction
* @extra_revokes: The number of additional revokes reserved (output)
*
* sdp->sd_log_flush_lock must not be held.
*/
void gfs2_log_reserve(struct gfs2_sbd *sdp, struct gfs2_trans *tr,
unsigned int *extra_revokes)
{
unsigned int blks = tr->tr_reserved;
unsigned int revokes = tr->tr_revokes;
unsigned int revoke_blks;
*extra_revokes = 0;
if (revokes) {
revoke_blks = DIV_ROUND_UP(revokes, sdp->sd_inptrs);
*extra_revokes = revoke_blks * sdp->sd_inptrs - revokes;
blks += revoke_blks;
}
__gfs2_log_reserve(sdp, blks, GFS2_LOG_FLUSH_MIN_BLOCKS);
}
/**
* log_distance - Compute distance between two journal blocks
* @sdp: The GFS2 superblock
* @newer: The most recent journal block of the pair
* @older: The older journal block of the pair
*
* Compute the distance (in the journal direction) between two
* blocks in the journal
*
* Returns: the distance in blocks
*/
static inline unsigned int log_distance(struct gfs2_sbd *sdp, unsigned int newer,
unsigned int older)
{
int dist;
dist = newer - older;
if (dist < 0)
dist += sdp->sd_jdesc->jd_blocks;
return dist;
}
[GFS2] assertion failure after writing to journaled file, umount This patch passes all my nasty tests that were causing the code to fail under one circumstance or another. Here is a complete summary of all changes from today's git tree, in order of appearance: 1. There are now separate variables for metadata buffer accounting. 2. Variable sd_log_num_hdrs is no longer needed, since the header accounting is taken care of by the reserve/refund sequence. 3. Fixed a tiny grammatical problem in a comment. 4. Added a new function "calc_reserved" to calculate the reserved log space. This isn't entirely necessary, but it has two benefits: First, it simplifies the gfs2_log_refund function greatly. Second, it allows for easier debugging because I could sprinkle the code with calls to this function to make sure the accounting is proper (by adding asserts and printks) at strategic point of the code. 5. In log_pull_tail there apparently was a kludge to fix up the accounting based on a "pull" parameter. The buffer accounting is now done properly, so the kludge was removed. 6. File sync operations were making a call to gfs2_log_flush that writes another journal header. Since that header was unplanned for (reserved) by the reserve/refund sequence, the free space had to be decremented so that when log_pull_tail gets called, the free space is be adjusted properly. (Did I hear you call that a kludge? well, maybe, but a lot more justifiable than the one I removed). 7. In the gfs2_log_shutdown code, it optionally syncs the log by specifying the PULL parameter to log_write_header. I'm not sure this is necessary anymore. It just seems to me there could be cases where shutdown is called while there are outstanding log buffers. 8. In the (data)buf_lo_before_commit functions, I changed some offset values from being calculated on the fly to being constants. That simplified some code and we might as well let the compiler do the calculation once rather than redoing those cycles at run time. 9. This version has my rewritten databuf_lo_add function. This version is much more like its predecessor, buf_lo_add, which makes it easier to understand. Again, this might not be necessary, but it seems as if this one works as well as the previous one, maybe even better, so I decided to leave it in. 10. In databuf_lo_before_commit, a previous data corruption problem was caused by going off the end of the buffer. The proper solution is to have the proper limit in place, rather than stopping earlier. (Thus my previous attempt to fix it is wrong). If you don't wrap the buffer, you're stopping too early and that causes more log buffer accounting problems. 11. In lops.h there are two new (previously mentioned) constants for figuring out the data offset for the journal buffers. 12. There are also two new functions, buf_limit and databuf_limit to calculate how many entries will fit in the buffer. 13. In function gfs2_meta_wipe, it needs to distinguish between pinned metadata buffers and journaled data buffers for proper journal buffer accounting. It can't use the JDATA gfs2_inode flag because it's sometimes passed the "real" inode and sometimes the "metadata inode" and the inode flags will be random bits in a metadata gfs2_inode. It needs to base its decision on which was passed in. Signed-off-by: Bob Peterson <rpeterso@redhat.com> Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
2007-06-19 03:50:20 +08:00
/**
* calc_reserved - Calculate the number of blocks to keep reserved
[GFS2] assertion failure after writing to journaled file, umount This patch passes all my nasty tests that were causing the code to fail under one circumstance or another. Here is a complete summary of all changes from today's git tree, in order of appearance: 1. There are now separate variables for metadata buffer accounting. 2. Variable sd_log_num_hdrs is no longer needed, since the header accounting is taken care of by the reserve/refund sequence. 3. Fixed a tiny grammatical problem in a comment. 4. Added a new function "calc_reserved" to calculate the reserved log space. This isn't entirely necessary, but it has two benefits: First, it simplifies the gfs2_log_refund function greatly. Second, it allows for easier debugging because I could sprinkle the code with calls to this function to make sure the accounting is proper (by adding asserts and printks) at strategic point of the code. 5. In log_pull_tail there apparently was a kludge to fix up the accounting based on a "pull" parameter. The buffer accounting is now done properly, so the kludge was removed. 6. File sync operations were making a call to gfs2_log_flush that writes another journal header. Since that header was unplanned for (reserved) by the reserve/refund sequence, the free space had to be decremented so that when log_pull_tail gets called, the free space is be adjusted properly. (Did I hear you call that a kludge? well, maybe, but a lot more justifiable than the one I removed). 7. In the gfs2_log_shutdown code, it optionally syncs the log by specifying the PULL parameter to log_write_header. I'm not sure this is necessary anymore. It just seems to me there could be cases where shutdown is called while there are outstanding log buffers. 8. In the (data)buf_lo_before_commit functions, I changed some offset values from being calculated on the fly to being constants. That simplified some code and we might as well let the compiler do the calculation once rather than redoing those cycles at run time. 9. This version has my rewritten databuf_lo_add function. This version is much more like its predecessor, buf_lo_add, which makes it easier to understand. Again, this might not be necessary, but it seems as if this one works as well as the previous one, maybe even better, so I decided to leave it in. 10. In databuf_lo_before_commit, a previous data corruption problem was caused by going off the end of the buffer. The proper solution is to have the proper limit in place, rather than stopping earlier. (Thus my previous attempt to fix it is wrong). If you don't wrap the buffer, you're stopping too early and that causes more log buffer accounting problems. 11. In lops.h there are two new (previously mentioned) constants for figuring out the data offset for the journal buffers. 12. There are also two new functions, buf_limit and databuf_limit to calculate how many entries will fit in the buffer. 13. In function gfs2_meta_wipe, it needs to distinguish between pinned metadata buffers and journaled data buffers for proper journal buffer accounting. It can't use the JDATA gfs2_inode flag because it's sometimes passed the "real" inode and sometimes the "metadata inode" and the inode flags will be random bits in a metadata gfs2_inode. It needs to base its decision on which was passed in. Signed-off-by: Bob Peterson <rpeterso@redhat.com> Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
2007-06-19 03:50:20 +08:00
* @sdp: The GFS2 superblock
*
* This is complex. We need to reserve room for all our currently used
* metadata blocks (e.g. normal file I/O rewriting file time stamps) and
* all our journaled data blocks for journaled files (e.g. files in the
[GFS2] assertion failure after writing to journaled file, umount This patch passes all my nasty tests that were causing the code to fail under one circumstance or another. Here is a complete summary of all changes from today's git tree, in order of appearance: 1. There are now separate variables for metadata buffer accounting. 2. Variable sd_log_num_hdrs is no longer needed, since the header accounting is taken care of by the reserve/refund sequence. 3. Fixed a tiny grammatical problem in a comment. 4. Added a new function "calc_reserved" to calculate the reserved log space. This isn't entirely necessary, but it has two benefits: First, it simplifies the gfs2_log_refund function greatly. Second, it allows for easier debugging because I could sprinkle the code with calls to this function to make sure the accounting is proper (by adding asserts and printks) at strategic point of the code. 5. In log_pull_tail there apparently was a kludge to fix up the accounting based on a "pull" parameter. The buffer accounting is now done properly, so the kludge was removed. 6. File sync operations were making a call to gfs2_log_flush that writes another journal header. Since that header was unplanned for (reserved) by the reserve/refund sequence, the free space had to be decremented so that when log_pull_tail gets called, the free space is be adjusted properly. (Did I hear you call that a kludge? well, maybe, but a lot more justifiable than the one I removed). 7. In the gfs2_log_shutdown code, it optionally syncs the log by specifying the PULL parameter to log_write_header. I'm not sure this is necessary anymore. It just seems to me there could be cases where shutdown is called while there are outstanding log buffers. 8. In the (data)buf_lo_before_commit functions, I changed some offset values from being calculated on the fly to being constants. That simplified some code and we might as well let the compiler do the calculation once rather than redoing those cycles at run time. 9. This version has my rewritten databuf_lo_add function. This version is much more like its predecessor, buf_lo_add, which makes it easier to understand. Again, this might not be necessary, but it seems as if this one works as well as the previous one, maybe even better, so I decided to leave it in. 10. In databuf_lo_before_commit, a previous data corruption problem was caused by going off the end of the buffer. The proper solution is to have the proper limit in place, rather than stopping earlier. (Thus my previous attempt to fix it is wrong). If you don't wrap the buffer, you're stopping too early and that causes more log buffer accounting problems. 11. In lops.h there are two new (previously mentioned) constants for figuring out the data offset for the journal buffers. 12. There are also two new functions, buf_limit and databuf_limit to calculate how many entries will fit in the buffer. 13. In function gfs2_meta_wipe, it needs to distinguish between pinned metadata buffers and journaled data buffers for proper journal buffer accounting. It can't use the JDATA gfs2_inode flag because it's sometimes passed the "real" inode and sometimes the "metadata inode" and the inode flags will be random bits in a metadata gfs2_inode. It needs to base its decision on which was passed in. Signed-off-by: Bob Peterson <rpeterso@redhat.com> Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
2007-06-19 03:50:20 +08:00
* meta_fs like rindex, or files for which chattr +j was done.)
* If we don't reserve enough space, corruption will follow.
[GFS2] assertion failure after writing to journaled file, umount This patch passes all my nasty tests that were causing the code to fail under one circumstance or another. Here is a complete summary of all changes from today's git tree, in order of appearance: 1. There are now separate variables for metadata buffer accounting. 2. Variable sd_log_num_hdrs is no longer needed, since the header accounting is taken care of by the reserve/refund sequence. 3. Fixed a tiny grammatical problem in a comment. 4. Added a new function "calc_reserved" to calculate the reserved log space. This isn't entirely necessary, but it has two benefits: First, it simplifies the gfs2_log_refund function greatly. Second, it allows for easier debugging because I could sprinkle the code with calls to this function to make sure the accounting is proper (by adding asserts and printks) at strategic point of the code. 5. In log_pull_tail there apparently was a kludge to fix up the accounting based on a "pull" parameter. The buffer accounting is now done properly, so the kludge was removed. 6. File sync operations were making a call to gfs2_log_flush that writes another journal header. Since that header was unplanned for (reserved) by the reserve/refund sequence, the free space had to be decremented so that when log_pull_tail gets called, the free space is be adjusted properly. (Did I hear you call that a kludge? well, maybe, but a lot more justifiable than the one I removed). 7. In the gfs2_log_shutdown code, it optionally syncs the log by specifying the PULL parameter to log_write_header. I'm not sure this is necessary anymore. It just seems to me there could be cases where shutdown is called while there are outstanding log buffers. 8. In the (data)buf_lo_before_commit functions, I changed some offset values from being calculated on the fly to being constants. That simplified some code and we might as well let the compiler do the calculation once rather than redoing those cycles at run time. 9. This version has my rewritten databuf_lo_add function. This version is much more like its predecessor, buf_lo_add, which makes it easier to understand. Again, this might not be necessary, but it seems as if this one works as well as the previous one, maybe even better, so I decided to leave it in. 10. In databuf_lo_before_commit, a previous data corruption problem was caused by going off the end of the buffer. The proper solution is to have the proper limit in place, rather than stopping earlier. (Thus my previous attempt to fix it is wrong). If you don't wrap the buffer, you're stopping too early and that causes more log buffer accounting problems. 11. In lops.h there are two new (previously mentioned) constants for figuring out the data offset for the journal buffers. 12. There are also two new functions, buf_limit and databuf_limit to calculate how many entries will fit in the buffer. 13. In function gfs2_meta_wipe, it needs to distinguish between pinned metadata buffers and journaled data buffers for proper journal buffer accounting. It can't use the JDATA gfs2_inode flag because it's sometimes passed the "real" inode and sometimes the "metadata inode" and the inode flags will be random bits in a metadata gfs2_inode. It needs to base its decision on which was passed in. Signed-off-by: Bob Peterson <rpeterso@redhat.com> Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
2007-06-19 03:50:20 +08:00
*
* We can have metadata blocks and jdata blocks in the same journal. Each
* type gets its own log descriptor, for which we need to reserve a block.
* In fact, each type has the potential for needing more than one log descriptor
* in cases where we have more blocks than will fit in a log descriptor.
[GFS2] assertion failure after writing to journaled file, umount This patch passes all my nasty tests that were causing the code to fail under one circumstance or another. Here is a complete summary of all changes from today's git tree, in order of appearance: 1. There are now separate variables for metadata buffer accounting. 2. Variable sd_log_num_hdrs is no longer needed, since the header accounting is taken care of by the reserve/refund sequence. 3. Fixed a tiny grammatical problem in a comment. 4. Added a new function "calc_reserved" to calculate the reserved log space. This isn't entirely necessary, but it has two benefits: First, it simplifies the gfs2_log_refund function greatly. Second, it allows for easier debugging because I could sprinkle the code with calls to this function to make sure the accounting is proper (by adding asserts and printks) at strategic point of the code. 5. In log_pull_tail there apparently was a kludge to fix up the accounting based on a "pull" parameter. The buffer accounting is now done properly, so the kludge was removed. 6. File sync operations were making a call to gfs2_log_flush that writes another journal header. Since that header was unplanned for (reserved) by the reserve/refund sequence, the free space had to be decremented so that when log_pull_tail gets called, the free space is be adjusted properly. (Did I hear you call that a kludge? well, maybe, but a lot more justifiable than the one I removed). 7. In the gfs2_log_shutdown code, it optionally syncs the log by specifying the PULL parameter to log_write_header. I'm not sure this is necessary anymore. It just seems to me there could be cases where shutdown is called while there are outstanding log buffers. 8. In the (data)buf_lo_before_commit functions, I changed some offset values from being calculated on the fly to being constants. That simplified some code and we might as well let the compiler do the calculation once rather than redoing those cycles at run time. 9. This version has my rewritten databuf_lo_add function. This version is much more like its predecessor, buf_lo_add, which makes it easier to understand. Again, this might not be necessary, but it seems as if this one works as well as the previous one, maybe even better, so I decided to leave it in. 10. In databuf_lo_before_commit, a previous data corruption problem was caused by going off the end of the buffer. The proper solution is to have the proper limit in place, rather than stopping earlier. (Thus my previous attempt to fix it is wrong). If you don't wrap the buffer, you're stopping too early and that causes more log buffer accounting problems. 11. In lops.h there are two new (previously mentioned) constants for figuring out the data offset for the journal buffers. 12. There are also two new functions, buf_limit and databuf_limit to calculate how many entries will fit in the buffer. 13. In function gfs2_meta_wipe, it needs to distinguish between pinned metadata buffers and journaled data buffers for proper journal buffer accounting. It can't use the JDATA gfs2_inode flag because it's sometimes passed the "real" inode and sometimes the "metadata inode" and the inode flags will be random bits in a metadata gfs2_inode. It needs to base its decision on which was passed in. Signed-off-by: Bob Peterson <rpeterso@redhat.com> Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
2007-06-19 03:50:20 +08:00
* Metadata journal entries take up half the space of journaled buffer entries.
*
* Also, we need to reserve blocks for revoke journal entries and one for an
* overall header for the lot.
*
* Returns: the number of blocks reserved
*/
static unsigned int calc_reserved(struct gfs2_sbd *sdp)
{
unsigned int reserved = GFS2_LOG_FLUSH_MIN_BLOCKS;
unsigned int blocks;
struct gfs2_trans *tr = sdp->sd_log_tr;
[GFS2] assertion failure after writing to journaled file, umount This patch passes all my nasty tests that were causing the code to fail under one circumstance or another. Here is a complete summary of all changes from today's git tree, in order of appearance: 1. There are now separate variables for metadata buffer accounting. 2. Variable sd_log_num_hdrs is no longer needed, since the header accounting is taken care of by the reserve/refund sequence. 3. Fixed a tiny grammatical problem in a comment. 4. Added a new function "calc_reserved" to calculate the reserved log space. This isn't entirely necessary, but it has two benefits: First, it simplifies the gfs2_log_refund function greatly. Second, it allows for easier debugging because I could sprinkle the code with calls to this function to make sure the accounting is proper (by adding asserts and printks) at strategic point of the code. 5. In log_pull_tail there apparently was a kludge to fix up the accounting based on a "pull" parameter. The buffer accounting is now done properly, so the kludge was removed. 6. File sync operations were making a call to gfs2_log_flush that writes another journal header. Since that header was unplanned for (reserved) by the reserve/refund sequence, the free space had to be decremented so that when log_pull_tail gets called, the free space is be adjusted properly. (Did I hear you call that a kludge? well, maybe, but a lot more justifiable than the one I removed). 7. In the gfs2_log_shutdown code, it optionally syncs the log by specifying the PULL parameter to log_write_header. I'm not sure this is necessary anymore. It just seems to me there could be cases where shutdown is called while there are outstanding log buffers. 8. In the (data)buf_lo_before_commit functions, I changed some offset values from being calculated on the fly to being constants. That simplified some code and we might as well let the compiler do the calculation once rather than redoing those cycles at run time. 9. This version has my rewritten databuf_lo_add function. This version is much more like its predecessor, buf_lo_add, which makes it easier to understand. Again, this might not be necessary, but it seems as if this one works as well as the previous one, maybe even better, so I decided to leave it in. 10. In databuf_lo_before_commit, a previous data corruption problem was caused by going off the end of the buffer. The proper solution is to have the proper limit in place, rather than stopping earlier. (Thus my previous attempt to fix it is wrong). If you don't wrap the buffer, you're stopping too early and that causes more log buffer accounting problems. 11. In lops.h there are two new (previously mentioned) constants for figuring out the data offset for the journal buffers. 12. There are also two new functions, buf_limit and databuf_limit to calculate how many entries will fit in the buffer. 13. In function gfs2_meta_wipe, it needs to distinguish between pinned metadata buffers and journaled data buffers for proper journal buffer accounting. It can't use the JDATA gfs2_inode flag because it's sometimes passed the "real" inode and sometimes the "metadata inode" and the inode flags will be random bits in a metadata gfs2_inode. It needs to base its decision on which was passed in. Signed-off-by: Bob Peterson <rpeterso@redhat.com> Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
2007-06-19 03:50:20 +08:00
if (tr) {
blocks = tr->tr_num_buf_new - tr->tr_num_buf_rm;
reserved += blocks + DIV_ROUND_UP(blocks, buf_limit(sdp));
blocks = tr->tr_num_databuf_new - tr->tr_num_databuf_rm;
reserved += blocks + DIV_ROUND_UP(blocks, databuf_limit(sdp));
}
[GFS2] assertion failure after writing to journaled file, umount This patch passes all my nasty tests that were causing the code to fail under one circumstance or another. Here is a complete summary of all changes from today's git tree, in order of appearance: 1. There are now separate variables for metadata buffer accounting. 2. Variable sd_log_num_hdrs is no longer needed, since the header accounting is taken care of by the reserve/refund sequence. 3. Fixed a tiny grammatical problem in a comment. 4. Added a new function "calc_reserved" to calculate the reserved log space. This isn't entirely necessary, but it has two benefits: First, it simplifies the gfs2_log_refund function greatly. Second, it allows for easier debugging because I could sprinkle the code with calls to this function to make sure the accounting is proper (by adding asserts and printks) at strategic point of the code. 5. In log_pull_tail there apparently was a kludge to fix up the accounting based on a "pull" parameter. The buffer accounting is now done properly, so the kludge was removed. 6. File sync operations were making a call to gfs2_log_flush that writes another journal header. Since that header was unplanned for (reserved) by the reserve/refund sequence, the free space had to be decremented so that when log_pull_tail gets called, the free space is be adjusted properly. (Did I hear you call that a kludge? well, maybe, but a lot more justifiable than the one I removed). 7. In the gfs2_log_shutdown code, it optionally syncs the log by specifying the PULL parameter to log_write_header. I'm not sure this is necessary anymore. It just seems to me there could be cases where shutdown is called while there are outstanding log buffers. 8. In the (data)buf_lo_before_commit functions, I changed some offset values from being calculated on the fly to being constants. That simplified some code and we might as well let the compiler do the calculation once rather than redoing those cycles at run time. 9. This version has my rewritten databuf_lo_add function. This version is much more like its predecessor, buf_lo_add, which makes it easier to understand. Again, this might not be necessary, but it seems as if this one works as well as the previous one, maybe even better, so I decided to leave it in. 10. In databuf_lo_before_commit, a previous data corruption problem was caused by going off the end of the buffer. The proper solution is to have the proper limit in place, rather than stopping earlier. (Thus my previous attempt to fix it is wrong). If you don't wrap the buffer, you're stopping too early and that causes more log buffer accounting problems. 11. In lops.h there are two new (previously mentioned) constants for figuring out the data offset for the journal buffers. 12. There are also two new functions, buf_limit and databuf_limit to calculate how many entries will fit in the buffer. 13. In function gfs2_meta_wipe, it needs to distinguish between pinned metadata buffers and journaled data buffers for proper journal buffer accounting. It can't use the JDATA gfs2_inode flag because it's sometimes passed the "real" inode and sometimes the "metadata inode" and the inode flags will be random bits in a metadata gfs2_inode. It needs to base its decision on which was passed in. Signed-off-by: Bob Peterson <rpeterso@redhat.com> Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
2007-06-19 03:50:20 +08:00
return reserved;
}
static void log_pull_tail(struct gfs2_sbd *sdp)
{
unsigned int new_tail = sdp->sd_log_flush_tail;
unsigned int dist;
if (new_tail == sdp->sd_log_tail)
return;
dist = log_distance(sdp, new_tail, sdp->sd_log_tail);
ail2_empty(sdp, new_tail);
gfs2_log_release(sdp, dist);
sdp->sd_log_tail = new_tail;
}
void log_flush_wait(struct gfs2_sbd *sdp)
{
DEFINE_WAIT(wait);
if (atomic_read(&sdp->sd_log_in_flight)) {
do {
prepare_to_wait(&sdp->sd_log_flush_wait, &wait,
TASK_UNINTERRUPTIBLE);
if (atomic_read(&sdp->sd_log_in_flight))
io_schedule();
} while(atomic_read(&sdp->sd_log_in_flight));
finish_wait(&sdp->sd_log_flush_wait, &wait);
}
}
static int ip_cmp(void *priv, const struct list_head *a, const struct list_head *b)
{
struct gfs2_inode *ipa, *ipb;
ipa = list_entry(a, struct gfs2_inode, i_ordered);
ipb = list_entry(b, struct gfs2_inode, i_ordered);
if (ipa->i_no_addr < ipb->i_no_addr)
return -1;
if (ipa->i_no_addr > ipb->i_no_addr)
return 1;
return 0;
}
static void __ordered_del_inode(struct gfs2_inode *ip)
{
if (!list_empty(&ip->i_ordered))
list_del_init(&ip->i_ordered);
}
static void gfs2_ordered_write(struct gfs2_sbd *sdp)
{
struct gfs2_inode *ip;
LIST_HEAD(written);
spin_lock(&sdp->sd_ordered_lock);
list_sort(NULL, &sdp->sd_log_ordered, &ip_cmp);
while (!list_empty(&sdp->sd_log_ordered)) {
ip = list_first_entry(&sdp->sd_log_ordered, struct gfs2_inode, i_ordered);
if (ip->i_inode.i_mapping->nrpages == 0) {
__ordered_del_inode(ip);
continue;
}
list_move(&ip->i_ordered, &written);
spin_unlock(&sdp->sd_ordered_lock);
filemap_fdatawrite(ip->i_inode.i_mapping);
spin_lock(&sdp->sd_ordered_lock);
}
list_splice(&written, &sdp->sd_log_ordered);
spin_unlock(&sdp->sd_ordered_lock);
}
static void gfs2_ordered_wait(struct gfs2_sbd *sdp)
{
struct gfs2_inode *ip;
spin_lock(&sdp->sd_ordered_lock);
while (!list_empty(&sdp->sd_log_ordered)) {
ip = list_first_entry(&sdp->sd_log_ordered, struct gfs2_inode, i_ordered);
__ordered_del_inode(ip);
if (ip->i_inode.i_mapping->nrpages == 0)
continue;
spin_unlock(&sdp->sd_ordered_lock);
filemap_fdatawait(ip->i_inode.i_mapping);
spin_lock(&sdp->sd_ordered_lock);
}
spin_unlock(&sdp->sd_ordered_lock);
}
void gfs2_ordered_del_inode(struct gfs2_inode *ip)
{
struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
spin_lock(&sdp->sd_ordered_lock);
__ordered_del_inode(ip);
spin_unlock(&sdp->sd_ordered_lock);
}
void gfs2_add_revoke(struct gfs2_sbd *sdp, struct gfs2_bufdata *bd)
{
struct buffer_head *bh = bd->bd_bh;
struct gfs2_glock *gl = bd->bd_gl;
sdp->sd_log_num_revoke++;
if (atomic_inc_return(&gl->gl_revokes) == 1)
gfs2_glock_hold(gl);
bh->b_private = NULL;
bd->bd_blkno = bh->b_blocknr;
gfs2_remove_from_ail(bd); /* drops ref on bh */
bd->bd_bh = NULL;
set_bit(GLF_LFLUSH, &gl->gl_flags);
list_add(&bd->bd_list, &sdp->sd_log_revokes);
}
void gfs2_glock_remove_revoke(struct gfs2_glock *gl)
{
if (atomic_dec_return(&gl->gl_revokes) == 0) {
clear_bit(GLF_LFLUSH, &gl->gl_flags);
gfs2_glock_queue_put(gl);
}
}
gfs2: Issue revokes more intelligently Before this patch, function gfs2_write_revokes would call gfs2_ail1_empty, then traverse the sd_ail1_list looking for transactions that had bds which were no longer queued to a glock. And if it found some, it would try to issue revokes for them, up to a predetermined maximum. There were two problems with how it did this. First was the fact that gfs2_ail1_empty moves transactions which have nothing remaining on the ail1 list from the sd_ail1_list to the sd_ail2_list, thus making its traversal of sd_ail1_list miss them completely, and therefore, never issue revokes for them. Second was the fact that there were three traversals (or partial traversals) of the sd_ail1_list, each of which took and then released the sd_ail_lock lock: First inside gfs2_ail1_empty, second to determine if there are any revokes to be issued, and third to actually issue them. All this taking and releasing of the sd_ail_lock meant other processes could modify the lists and the conditions in which we're working. This patch simplies the whole process by adding a new parameter to function gfs2_ail1_empty, max_revokes. For normal calls, this is passed in as 0, meaning we don't want to issue any revokes. For function gfs2_write_revokes, we pass in the maximum number of revokes we can, thus allowing gfs2_ail1_empty to add the revokes where needed. This simplies the code, allows for a single holding of the sd_ail_lock, and allows gfs2_ail1_empty to add revokes for all the necessary bd items without missing any. Signed-off-by: Bob Peterson <rpeterso@redhat.com> Reviewed-by: Andreas Gruenbacher <agruenba@redhat.com>
2019-02-22 05:28:07 +08:00
/**
* gfs2_flush_revokes - Add as many revokes to the system transaction as we can
gfs2: Issue revokes more intelligently Before this patch, function gfs2_write_revokes would call gfs2_ail1_empty, then traverse the sd_ail1_list looking for transactions that had bds which were no longer queued to a glock. And if it found some, it would try to issue revokes for them, up to a predetermined maximum. There were two problems with how it did this. First was the fact that gfs2_ail1_empty moves transactions which have nothing remaining on the ail1 list from the sd_ail1_list to the sd_ail2_list, thus making its traversal of sd_ail1_list miss them completely, and therefore, never issue revokes for them. Second was the fact that there were three traversals (or partial traversals) of the sd_ail1_list, each of which took and then released the sd_ail_lock lock: First inside gfs2_ail1_empty, second to determine if there are any revokes to be issued, and third to actually issue them. All this taking and releasing of the sd_ail_lock meant other processes could modify the lists and the conditions in which we're working. This patch simplies the whole process by adding a new parameter to function gfs2_ail1_empty, max_revokes. For normal calls, this is passed in as 0, meaning we don't want to issue any revokes. For function gfs2_write_revokes, we pass in the maximum number of revokes we can, thus allowing gfs2_ail1_empty to add the revokes where needed. This simplies the code, allows for a single holding of the sd_ail_lock, and allows gfs2_ail1_empty to add revokes for all the necessary bd items without missing any. Signed-off-by: Bob Peterson <rpeterso@redhat.com> Reviewed-by: Andreas Gruenbacher <agruenba@redhat.com>
2019-02-22 05:28:07 +08:00
* @sdp: The GFS2 superblock
*
* Our usual strategy is to defer writing revokes as much as we can in the hope
* that we'll eventually overwrite the journal, which will make those revokes
* go away. This changes when we flush the log: at that point, there will
* likely be some left-over space in the last revoke block of that transaction.
* We can fill that space with additional revokes for blocks that have already
* been written back. This will basically come at no cost now, and will save
* us from having to keep track of those blocks on the AIL2 list later.
*/
void gfs2_flush_revokes(struct gfs2_sbd *sdp)
{
gfs2: Issue revokes more intelligently Before this patch, function gfs2_write_revokes would call gfs2_ail1_empty, then traverse the sd_ail1_list looking for transactions that had bds which were no longer queued to a glock. And if it found some, it would try to issue revokes for them, up to a predetermined maximum. There were two problems with how it did this. First was the fact that gfs2_ail1_empty moves transactions which have nothing remaining on the ail1 list from the sd_ail1_list to the sd_ail2_list, thus making its traversal of sd_ail1_list miss them completely, and therefore, never issue revokes for them. Second was the fact that there were three traversals (or partial traversals) of the sd_ail1_list, each of which took and then released the sd_ail_lock lock: First inside gfs2_ail1_empty, second to determine if there are any revokes to be issued, and third to actually issue them. All this taking and releasing of the sd_ail_lock meant other processes could modify the lists and the conditions in which we're working. This patch simplies the whole process by adding a new parameter to function gfs2_ail1_empty, max_revokes. For normal calls, this is passed in as 0, meaning we don't want to issue any revokes. For function gfs2_write_revokes, we pass in the maximum number of revokes we can, thus allowing gfs2_ail1_empty to add the revokes where needed. This simplies the code, allows for a single holding of the sd_ail_lock, and allows gfs2_ail1_empty to add revokes for all the necessary bd items without missing any. Signed-off-by: Bob Peterson <rpeterso@redhat.com> Reviewed-by: Andreas Gruenbacher <agruenba@redhat.com>
2019-02-22 05:28:07 +08:00
/* number of revokes we still have room for */
unsigned int max_revokes = atomic_read(&sdp->sd_log_revokes_available);
gfs2: Issue revokes more intelligently Before this patch, function gfs2_write_revokes would call gfs2_ail1_empty, then traverse the sd_ail1_list looking for transactions that had bds which were no longer queued to a glock. And if it found some, it would try to issue revokes for them, up to a predetermined maximum. There were two problems with how it did this. First was the fact that gfs2_ail1_empty moves transactions which have nothing remaining on the ail1 list from the sd_ail1_list to the sd_ail2_list, thus making its traversal of sd_ail1_list miss them completely, and therefore, never issue revokes for them. Second was the fact that there were three traversals (or partial traversals) of the sd_ail1_list, each of which took and then released the sd_ail_lock lock: First inside gfs2_ail1_empty, second to determine if there are any revokes to be issued, and third to actually issue them. All this taking and releasing of the sd_ail_lock meant other processes could modify the lists and the conditions in which we're working. This patch simplies the whole process by adding a new parameter to function gfs2_ail1_empty, max_revokes. For normal calls, this is passed in as 0, meaning we don't want to issue any revokes. For function gfs2_write_revokes, we pass in the maximum number of revokes we can, thus allowing gfs2_ail1_empty to add the revokes where needed. This simplies the code, allows for a single holding of the sd_ail_lock, and allows gfs2_ail1_empty to add revokes for all the necessary bd items without missing any. Signed-off-by: Bob Peterson <rpeterso@redhat.com> Reviewed-by: Andreas Gruenbacher <agruenba@redhat.com>
2019-02-22 05:28:07 +08:00
gfs2_log_lock(sdp);
gfs2_ail1_empty(sdp, max_revokes);
gfs2_log_unlock(sdp);
}
/**
* gfs2_write_log_header - Write a journal log header buffer at lblock
* @sdp: The GFS2 superblock
* @jd: journal descriptor of the journal to which we are writing
* @seq: sequence number
* @tail: tail of the log
* @lblock: value for lh_blkno (block number relative to start of journal)
* @flags: log header flags GFS2_LOG_HEAD_*
* @op_flags: flags to pass to the bio
*
* Returns: the initialized log buffer descriptor
*/
void gfs2_write_log_header(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd,
u64 seq, u32 tail, u32 lblock, u32 flags,
int op_flags)
{
struct gfs2_log_header *lh;
u32 hash, crc;
struct page *page;
struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
struct timespec64 tv;
struct super_block *sb = sdp->sd_vfs;
u64 dblock;
if (gfs2_withdrawn(sdp))
return;
page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
lh = page_address(page);
clear_page(lh);
lh->lh_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
lh->lh_header.mh_type = cpu_to_be32(GFS2_METATYPE_LH);
lh->lh_header.__pad0 = cpu_to_be64(0);
lh->lh_header.mh_format = cpu_to_be32(GFS2_FORMAT_LH);
lh->lh_header.mh_jid = cpu_to_be32(sdp->sd_jdesc->jd_jid);
lh->lh_sequence = cpu_to_be64(seq);
lh->lh_flags = cpu_to_be32(flags);
lh->lh_tail = cpu_to_be32(tail);
lh->lh_blkno = cpu_to_be32(lblock);
hash = ~crc32(~0, lh, LH_V1_SIZE);
lh->lh_hash = cpu_to_be32(hash);
ktime_get_coarse_real_ts64(&tv);
lh->lh_nsec = cpu_to_be32(tv.tv_nsec);
lh->lh_sec = cpu_to_be64(tv.tv_sec);
if (!list_empty(&jd->extent_list))
dblock = gfs2_log_bmap(jd, lblock);
else {
unsigned int extlen;
int ret;
extlen = 1;
ret = gfs2_get_extent(jd->jd_inode, lblock, &dblock, &extlen);
if (gfs2_assert_withdraw(sdp, ret == 0))
return;
}
lh->lh_addr = cpu_to_be64(dblock);
lh->lh_jinode = cpu_to_be64(GFS2_I(jd->jd_inode)->i_no_addr);
/* We may only write local statfs, quota, etc., when writing to our
own journal. The values are left 0 when recovering a journal
different from our own. */
if (!(flags & GFS2_LOG_HEAD_RECOVERY)) {
lh->lh_statfs_addr =
cpu_to_be64(GFS2_I(sdp->sd_sc_inode)->i_no_addr);
lh->lh_quota_addr =
cpu_to_be64(GFS2_I(sdp->sd_qc_inode)->i_no_addr);
spin_lock(&sdp->sd_statfs_spin);
lh->lh_local_total = cpu_to_be64(l_sc->sc_total);
lh->lh_local_free = cpu_to_be64(l_sc->sc_free);
lh->lh_local_dinodes = cpu_to_be64(l_sc->sc_dinodes);
spin_unlock(&sdp->sd_statfs_spin);
}
BUILD_BUG_ON(offsetof(struct gfs2_log_header, lh_crc) != LH_V1_SIZE);
crc = crc32c(~0, (void *)lh + LH_V1_SIZE + 4,
sb->s_blocksize - LH_V1_SIZE - 4);
lh->lh_crc = cpu_to_be32(crc);
gfs2_log_write(sdp, jd, page, sb->s_blocksize, 0, dblock);
gfs2_log_submit_bio(&jd->jd_log_bio, REQ_OP_WRITE | op_flags);
}
/**
* log_write_header - Get and initialize a journal header buffer
* @sdp: The GFS2 superblock
* @flags: The log header flags, including log header origin
*
* Returns: the initialized log buffer descriptor
*/
static void log_write_header(struct gfs2_sbd *sdp, u32 flags)
{
int op_flags = REQ_PREFLUSH | REQ_FUA | REQ_META | REQ_SYNC;
enum gfs2_freeze_state state = atomic_read(&sdp->sd_freeze_state);
gfs2_assert_withdraw(sdp, (state != SFS_FROZEN));
if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags)) {
gfs2_ordered_wait(sdp);
log_flush_wait(sdp);
op_flags = REQ_SYNC | REQ_META | REQ_PRIO;
}
sdp->sd_log_idle = (sdp->sd_log_flush_tail == sdp->sd_log_flush_head);
gfs2_write_log_header(sdp, sdp->sd_jdesc, sdp->sd_log_sequence++,
sdp->sd_log_flush_tail, sdp->sd_log_flush_head,
flags, op_flags);
gfs2_log_incr_head(sdp);
log_flush_wait(sdp);
log_pull_tail(sdp);
gfs2_log_update_head(sdp);
}
/**
* gfs2_ail_drain - drain the ail lists after a withdraw
* @sdp: Pointer to GFS2 superblock
*/
void gfs2_ail_drain(struct gfs2_sbd *sdp)
{
struct gfs2_trans *tr;
spin_lock(&sdp->sd_ail_lock);
/*
* For transactions on the sd_ail1_list we need to drain both the
* ail1 and ail2 lists. That's because function gfs2_ail1_start_one
* (temporarily) moves items from its tr_ail1 list to tr_ail2 list
* before revokes are sent for that block. Items on the sd_ail2_list
* should have already gotten beyond that point, so no need.
*/
while (!list_empty(&sdp->sd_ail1_list)) {
tr = list_first_entry(&sdp->sd_ail1_list, struct gfs2_trans,
tr_list);
gfs2_ail_empty_tr(sdp, tr, &tr->tr_ail1_list);
gfs2_ail_empty_tr(sdp, tr, &tr->tr_ail2_list);
list_del(&tr->tr_list);
gfs2_trans_free(sdp, tr);
}
while (!list_empty(&sdp->sd_ail2_list)) {
tr = list_first_entry(&sdp->sd_ail2_list, struct gfs2_trans,
tr_list);
gfs2_ail_empty_tr(sdp, tr, &tr->tr_ail2_list);
list_del(&tr->tr_list);
gfs2_trans_free(sdp, tr);
}
gfs2_drain_revokes(sdp);
spin_unlock(&sdp->sd_ail_lock);
}
/**
* empty_ail1_list - try to start IO and empty the ail1 list
* @sdp: Pointer to GFS2 superblock
*/
static void empty_ail1_list(struct gfs2_sbd *sdp)
{
unsigned long start = jiffies;
for (;;) {
if (time_after(jiffies, start + (HZ * 600))) {
fs_err(sdp, "Error: In %s for 10 minutes! t=%d\n",
__func__, current->journal_info ? 1 : 0);
dump_ail_list(sdp);
return;
}
gfs2_ail1_start(sdp);
gfs2_ail1_wait(sdp);
if (gfs2_ail1_empty(sdp, 0))
return;
}
}
/**
* trans_drain - drain the buf and databuf queue for a failed transaction
* @tr: the transaction to drain
*
* When this is called, we're taking an error exit for a log write that failed
* but since we bypassed the after_commit functions, we need to remove the
* items from the buf and databuf queue.
*/
static void trans_drain(struct gfs2_trans *tr)
{
struct gfs2_bufdata *bd;
struct list_head *head;
if (!tr)
return;
head = &tr->tr_buf;
while (!list_empty(head)) {
bd = list_first_entry(head, struct gfs2_bufdata, bd_list);
list_del_init(&bd->bd_list);
if (!list_empty(&bd->bd_ail_st_list))
gfs2_remove_from_ail(bd);
kmem_cache_free(gfs2_bufdata_cachep, bd);
}
head = &tr->tr_databuf;
while (!list_empty(head)) {
bd = list_first_entry(head, struct gfs2_bufdata, bd_list);
list_del_init(&bd->bd_list);
if (!list_empty(&bd->bd_ail_st_list))
gfs2_remove_from_ail(bd);
kmem_cache_free(gfs2_bufdata_cachep, bd);
}
}
/**
* gfs2_log_flush - flush incore transaction(s)
* @sdp: The filesystem
* @gl: The glock structure to flush. If NULL, flush the whole incore log
* @flags: The log header flags: GFS2_LOG_HEAD_FLUSH_* and debug flags
*
*/
void gfs2_log_flush(struct gfs2_sbd *sdp, struct gfs2_glock *gl, u32 flags)
{
struct gfs2_trans *tr = NULL;
unsigned int reserved_blocks = 0, used_blocks = 0;
enum gfs2_freeze_state state = atomic_read(&sdp->sd_freeze_state);
unsigned int first_log_head;
unsigned int reserved_revokes = 0;
down_write(&sdp->sd_log_flush_lock);
trace_gfs2_log_flush(sdp, 1, flags);
repeat:
/*
* Do this check while holding the log_flush_lock to prevent new
* buffers from being added to the ail via gfs2_pin()
*/
if (gfs2_withdrawn(sdp) || !test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))
goto out;
/* Log might have been flushed while we waited for the flush lock */
if (gl && !test_bit(GLF_LFLUSH, &gl->gl_flags))
goto out;
first_log_head = sdp->sd_log_head;
sdp->sd_log_flush_head = first_log_head;
tr = sdp->sd_log_tr;
if (tr || sdp->sd_log_num_revoke) {
if (reserved_blocks)
gfs2_log_release(sdp, reserved_blocks);
reserved_blocks = sdp->sd_log_blks_reserved;
reserved_revokes = sdp->sd_log_num_revoke;
if (tr) {
sdp->sd_log_tr = NULL;
tr->tr_first = first_log_head;
if (unlikely (state == SFS_FROZEN)) {
if (gfs2_assert_withdraw_delayed(sdp,
!tr->tr_num_buf_new && !tr->tr_num_databuf_new))
goto out_withdraw;
}
}
} else if (!reserved_blocks) {
unsigned int taboo_blocks = GFS2_LOG_FLUSH_MIN_BLOCKS;
reserved_blocks = GFS2_LOG_FLUSH_MIN_BLOCKS;
if (current == sdp->sd_logd_process)
taboo_blocks = 0;
if (!__gfs2_log_try_reserve(sdp, reserved_blocks, taboo_blocks)) {
up_write(&sdp->sd_log_flush_lock);
__gfs2_log_reserve(sdp, reserved_blocks, taboo_blocks);
down_write(&sdp->sd_log_flush_lock);
goto repeat;
}
BUG_ON(sdp->sd_log_num_revoke);
}
if (flags & GFS2_LOG_HEAD_FLUSH_SHUTDOWN)
clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
if (unlikely(state == SFS_FROZEN))
if (gfs2_assert_withdraw_delayed(sdp, !reserved_revokes))
goto out_withdraw;
gfs2_ordered_write(sdp);
if (gfs2_withdrawn(sdp))
goto out_withdraw;
lops_before_commit(sdp, tr);
if (gfs2_withdrawn(sdp))
goto out_withdraw;
gfs2_log_submit_bio(&sdp->sd_jdesc->jd_log_bio, REQ_OP_WRITE);
if (gfs2_withdrawn(sdp))
goto out_withdraw;
if (sdp->sd_log_head != sdp->sd_log_flush_head) {
log_write_header(sdp, flags);
} else if (sdp->sd_log_tail != sdp->sd_log_flush_tail && !sdp->sd_log_idle) {
log_write_header(sdp, flags);
[GFS2] assertion failure after writing to journaled file, umount This patch passes all my nasty tests that were causing the code to fail under one circumstance or another. Here is a complete summary of all changes from today's git tree, in order of appearance: 1. There are now separate variables for metadata buffer accounting. 2. Variable sd_log_num_hdrs is no longer needed, since the header accounting is taken care of by the reserve/refund sequence. 3. Fixed a tiny grammatical problem in a comment. 4. Added a new function "calc_reserved" to calculate the reserved log space. This isn't entirely necessary, but it has two benefits: First, it simplifies the gfs2_log_refund function greatly. Second, it allows for easier debugging because I could sprinkle the code with calls to this function to make sure the accounting is proper (by adding asserts and printks) at strategic point of the code. 5. In log_pull_tail there apparently was a kludge to fix up the accounting based on a "pull" parameter. The buffer accounting is now done properly, so the kludge was removed. 6. File sync operations were making a call to gfs2_log_flush that writes another journal header. Since that header was unplanned for (reserved) by the reserve/refund sequence, the free space had to be decremented so that when log_pull_tail gets called, the free space is be adjusted properly. (Did I hear you call that a kludge? well, maybe, but a lot more justifiable than the one I removed). 7. In the gfs2_log_shutdown code, it optionally syncs the log by specifying the PULL parameter to log_write_header. I'm not sure this is necessary anymore. It just seems to me there could be cases where shutdown is called while there are outstanding log buffers. 8. In the (data)buf_lo_before_commit functions, I changed some offset values from being calculated on the fly to being constants. That simplified some code and we might as well let the compiler do the calculation once rather than redoing those cycles at run time. 9. This version has my rewritten databuf_lo_add function. This version is much more like its predecessor, buf_lo_add, which makes it easier to understand. Again, this might not be necessary, but it seems as if this one works as well as the previous one, maybe even better, so I decided to leave it in. 10. In databuf_lo_before_commit, a previous data corruption problem was caused by going off the end of the buffer. The proper solution is to have the proper limit in place, rather than stopping earlier. (Thus my previous attempt to fix it is wrong). If you don't wrap the buffer, you're stopping too early and that causes more log buffer accounting problems. 11. In lops.h there are two new (previously mentioned) constants for figuring out the data offset for the journal buffers. 12. There are also two new functions, buf_limit and databuf_limit to calculate how many entries will fit in the buffer. 13. In function gfs2_meta_wipe, it needs to distinguish between pinned metadata buffers and journaled data buffers for proper journal buffer accounting. It can't use the JDATA gfs2_inode flag because it's sometimes passed the "real" inode and sometimes the "metadata inode" and the inode flags will be random bits in a metadata gfs2_inode. It needs to base its decision on which was passed in. Signed-off-by: Bob Peterson <rpeterso@redhat.com> Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
2007-06-19 03:50:20 +08:00
}
if (gfs2_withdrawn(sdp))
goto out_withdraw;
lops_after_commit(sdp, tr);
gfs2_log_lock(sdp);
sdp->sd_log_blks_reserved = 0;
spin_lock(&sdp->sd_ail_lock);
if (tr && !list_empty(&tr->tr_ail1_list)) {
list_add(&tr->tr_list, &sdp->sd_ail1_list);
tr = NULL;
}
spin_unlock(&sdp->sd_ail_lock);
gfs2_log_unlock(sdp);
GFS2: remove transaction glock GFS2 has a transaction glock, which must be grabbed for every transaction, whose purpose is to deal with freezing the filesystem. Aside from this involving a large amount of locking, it is very easy to make the current fsfreeze code hang on unfreezing. This patch rewrites how gfs2 handles freezing the filesystem. The transaction glock is removed. In it's place is a freeze glock, which is cached (but not held) in a shared state by every node in the cluster when the filesystem is mounted. This lock only needs to be grabbed on freezing, and actions which need to be safe from freezing, like recovery. When a node wants to freeze the filesystem, it grabs this glock exclusively. When the freeze glock state changes on the nodes (either from shared to unlocked, or shared to exclusive), the filesystem does a special log flush. gfs2_log_flush() does all the work for flushing out the and shutting down the incore log, and then it tries to grab the freeze glock in a shared state again. Since the filesystem is stuck in gfs2_log_flush, no new transaction can start, and nothing can be written to disk. Unfreezing the filesytem simply involes dropping the freeze glock, allowing gfs2_log_flush() to grab and then release the shared lock, so it is cached for next time. However, in order for the unfreezing ioctl to occur, gfs2 needs to get a shared lock on the filesystem root directory inode to check permissions. If that glock has already been grabbed exclusively, fsfreeze will be unable to get the shared lock and unfreeze the filesystem. In order to allow the unfreeze, this patch makes gfs2 grab a shared lock on the filesystem root directory during the freeze, and hold it until it unfreezes the filesystem. The functions which need to grab a shared lock in order to allow the unfreeze ioctl to be issued now use the lock grabbed by the freeze code instead. The freeze and unfreeze code take care to make sure that this shared lock will not be dropped while another process is using it. Signed-off-by: Benjamin Marzinski <bmarzins@redhat.com> Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
2014-05-02 11:26:55 +08:00
if (!(flags & GFS2_LOG_HEAD_FLUSH_NORMAL)) {
GFS2: remove transaction glock GFS2 has a transaction glock, which must be grabbed for every transaction, whose purpose is to deal with freezing the filesystem. Aside from this involving a large amount of locking, it is very easy to make the current fsfreeze code hang on unfreezing. This patch rewrites how gfs2 handles freezing the filesystem. The transaction glock is removed. In it's place is a freeze glock, which is cached (but not held) in a shared state by every node in the cluster when the filesystem is mounted. This lock only needs to be grabbed on freezing, and actions which need to be safe from freezing, like recovery. When a node wants to freeze the filesystem, it grabs this glock exclusively. When the freeze glock state changes on the nodes (either from shared to unlocked, or shared to exclusive), the filesystem does a special log flush. gfs2_log_flush() does all the work for flushing out the and shutting down the incore log, and then it tries to grab the freeze glock in a shared state again. Since the filesystem is stuck in gfs2_log_flush, no new transaction can start, and nothing can be written to disk. Unfreezing the filesytem simply involes dropping the freeze glock, allowing gfs2_log_flush() to grab and then release the shared lock, so it is cached for next time. However, in order for the unfreezing ioctl to occur, gfs2 needs to get a shared lock on the filesystem root directory inode to check permissions. If that glock has already been grabbed exclusively, fsfreeze will be unable to get the shared lock and unfreeze the filesystem. In order to allow the unfreeze, this patch makes gfs2 grab a shared lock on the filesystem root directory during the freeze, and hold it until it unfreezes the filesystem. The functions which need to grab a shared lock in order to allow the unfreeze ioctl to be issued now use the lock grabbed by the freeze code instead. The freeze and unfreeze code take care to make sure that this shared lock will not be dropped while another process is using it. Signed-off-by: Benjamin Marzinski <bmarzins@redhat.com> Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
2014-05-02 11:26:55 +08:00
if (!sdp->sd_log_idle) {
empty_ail1_list(sdp);
if (gfs2_withdrawn(sdp))
goto out_withdraw;
log_write_header(sdp, flags);
GFS2: remove transaction glock GFS2 has a transaction glock, which must be grabbed for every transaction, whose purpose is to deal with freezing the filesystem. Aside from this involving a large amount of locking, it is very easy to make the current fsfreeze code hang on unfreezing. This patch rewrites how gfs2 handles freezing the filesystem. The transaction glock is removed. In it's place is a freeze glock, which is cached (but not held) in a shared state by every node in the cluster when the filesystem is mounted. This lock only needs to be grabbed on freezing, and actions which need to be safe from freezing, like recovery. When a node wants to freeze the filesystem, it grabs this glock exclusively. When the freeze glock state changes on the nodes (either from shared to unlocked, or shared to exclusive), the filesystem does a special log flush. gfs2_log_flush() does all the work for flushing out the and shutting down the incore log, and then it tries to grab the freeze glock in a shared state again. Since the filesystem is stuck in gfs2_log_flush, no new transaction can start, and nothing can be written to disk. Unfreezing the filesytem simply involes dropping the freeze glock, allowing gfs2_log_flush() to grab and then release the shared lock, so it is cached for next time. However, in order for the unfreezing ioctl to occur, gfs2 needs to get a shared lock on the filesystem root directory inode to check permissions. If that glock has already been grabbed exclusively, fsfreeze will be unable to get the shared lock and unfreeze the filesystem. In order to allow the unfreeze, this patch makes gfs2 grab a shared lock on the filesystem root directory during the freeze, and hold it until it unfreezes the filesystem. The functions which need to grab a shared lock in order to allow the unfreeze ioctl to be issued now use the lock grabbed by the freeze code instead. The freeze and unfreeze code take care to make sure that this shared lock will not be dropped while another process is using it. Signed-off-by: Benjamin Marzinski <bmarzins@redhat.com> Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
2014-05-02 11:26:55 +08:00
}
if (flags & (GFS2_LOG_HEAD_FLUSH_SHUTDOWN |
GFS2_LOG_HEAD_FLUSH_FREEZE))
GFS2: remove transaction glock GFS2 has a transaction glock, which must be grabbed for every transaction, whose purpose is to deal with freezing the filesystem. Aside from this involving a large amount of locking, it is very easy to make the current fsfreeze code hang on unfreezing. This patch rewrites how gfs2 handles freezing the filesystem. The transaction glock is removed. In it's place is a freeze glock, which is cached (but not held) in a shared state by every node in the cluster when the filesystem is mounted. This lock only needs to be grabbed on freezing, and actions which need to be safe from freezing, like recovery. When a node wants to freeze the filesystem, it grabs this glock exclusively. When the freeze glock state changes on the nodes (either from shared to unlocked, or shared to exclusive), the filesystem does a special log flush. gfs2_log_flush() does all the work for flushing out the and shutting down the incore log, and then it tries to grab the freeze glock in a shared state again. Since the filesystem is stuck in gfs2_log_flush, no new transaction can start, and nothing can be written to disk. Unfreezing the filesytem simply involes dropping the freeze glock, allowing gfs2_log_flush() to grab and then release the shared lock, so it is cached for next time. However, in order for the unfreezing ioctl to occur, gfs2 needs to get a shared lock on the filesystem root directory inode to check permissions. If that glock has already been grabbed exclusively, fsfreeze will be unable to get the shared lock and unfreeze the filesystem. In order to allow the unfreeze, this patch makes gfs2 grab a shared lock on the filesystem root directory during the freeze, and hold it until it unfreezes the filesystem. The functions which need to grab a shared lock in order to allow the unfreeze ioctl to be issued now use the lock grabbed by the freeze code instead. The freeze and unfreeze code take care to make sure that this shared lock will not be dropped while another process is using it. Signed-off-by: Benjamin Marzinski <bmarzins@redhat.com> Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
2014-05-02 11:26:55 +08:00
gfs2_log_shutdown(sdp);
if (flags & GFS2_LOG_HEAD_FLUSH_FREEZE)
atomic_set(&sdp->sd_freeze_state, SFS_FROZEN);
GFS2: remove transaction glock GFS2 has a transaction glock, which must be grabbed for every transaction, whose purpose is to deal with freezing the filesystem. Aside from this involving a large amount of locking, it is very easy to make the current fsfreeze code hang on unfreezing. This patch rewrites how gfs2 handles freezing the filesystem. The transaction glock is removed. In it's place is a freeze glock, which is cached (but not held) in a shared state by every node in the cluster when the filesystem is mounted. This lock only needs to be grabbed on freezing, and actions which need to be safe from freezing, like recovery. When a node wants to freeze the filesystem, it grabs this glock exclusively. When the freeze glock state changes on the nodes (either from shared to unlocked, or shared to exclusive), the filesystem does a special log flush. gfs2_log_flush() does all the work for flushing out the and shutting down the incore log, and then it tries to grab the freeze glock in a shared state again. Since the filesystem is stuck in gfs2_log_flush, no new transaction can start, and nothing can be written to disk. Unfreezing the filesytem simply involes dropping the freeze glock, allowing gfs2_log_flush() to grab and then release the shared lock, so it is cached for next time. However, in order for the unfreezing ioctl to occur, gfs2 needs to get a shared lock on the filesystem root directory inode to check permissions. If that glock has already been grabbed exclusively, fsfreeze will be unable to get the shared lock and unfreeze the filesystem. In order to allow the unfreeze, this patch makes gfs2 grab a shared lock on the filesystem root directory during the freeze, and hold it until it unfreezes the filesystem. The functions which need to grab a shared lock in order to allow the unfreeze ioctl to be issued now use the lock grabbed by the freeze code instead. The freeze and unfreeze code take care to make sure that this shared lock will not be dropped while another process is using it. Signed-off-by: Benjamin Marzinski <bmarzins@redhat.com> Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
2014-05-02 11:26:55 +08:00
}
out_end:
used_blocks = log_distance(sdp, sdp->sd_log_flush_head, first_log_head);
reserved_revokes += atomic_read(&sdp->sd_log_revokes_available);
atomic_set(&sdp->sd_log_revokes_available, sdp->sd_ldptrs);
gfs2_assert_withdraw(sdp, reserved_revokes % sdp->sd_inptrs == sdp->sd_ldptrs);
if (reserved_revokes > sdp->sd_ldptrs)
reserved_blocks += (reserved_revokes - sdp->sd_ldptrs) / sdp->sd_inptrs;
out:
if (used_blocks != reserved_blocks) {
gfs2_assert_withdraw_delayed(sdp, used_blocks < reserved_blocks);
gfs2_log_release(sdp, reserved_blocks - used_blocks);
}
up_write(&sdp->sd_log_flush_lock);
gfs2_trans_free(sdp, tr);
if (gfs2_withdrawing(sdp))
gfs2_withdraw(sdp);
trace_gfs2_log_flush(sdp, 0, flags);
return;
out_withdraw:
trans_drain(tr);
/**
* If the tr_list is empty, we're withdrawing during a log
* flush that targets a transaction, but the transaction was
* never queued onto any of the ail lists. Here we add it to
* ail1 just so that ail_drain() will find and free it.
*/
spin_lock(&sdp->sd_ail_lock);
if (tr && list_empty(&tr->tr_list))
list_add(&tr->tr_list, &sdp->sd_ail1_list);
spin_unlock(&sdp->sd_ail_lock);
tr = NULL;
goto out_end;
}
/**
* gfs2_merge_trans - Merge a new transaction into a cached transaction
* @sdp: the filesystem
* @new: New transaction to be merged
*/
static void gfs2_merge_trans(struct gfs2_sbd *sdp, struct gfs2_trans *new)
{
struct gfs2_trans *old = sdp->sd_log_tr;
WARN_ON_ONCE(!test_bit(TR_ATTACHED, &old->tr_flags));
old->tr_num_buf_new += new->tr_num_buf_new;
old->tr_num_databuf_new += new->tr_num_databuf_new;
old->tr_num_buf_rm += new->tr_num_buf_rm;
old->tr_num_databuf_rm += new->tr_num_databuf_rm;
old->tr_revokes += new->tr_revokes;
old->tr_num_revoke += new->tr_num_revoke;
list_splice_tail_init(&new->tr_databuf, &old->tr_databuf);
list_splice_tail_init(&new->tr_buf, &old->tr_buf);
spin_lock(&sdp->sd_ail_lock);
list_splice_tail_init(&new->tr_ail1_list, &old->tr_ail1_list);
list_splice_tail_init(&new->tr_ail2_list, &old->tr_ail2_list);
spin_unlock(&sdp->sd_ail_lock);
}
static void log_refund(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
{
[GFS2] assertion failure after writing to journaled file, umount This patch passes all my nasty tests that were causing the code to fail under one circumstance or another. Here is a complete summary of all changes from today's git tree, in order of appearance: 1. There are now separate variables for metadata buffer accounting. 2. Variable sd_log_num_hdrs is no longer needed, since the header accounting is taken care of by the reserve/refund sequence. 3. Fixed a tiny grammatical problem in a comment. 4. Added a new function "calc_reserved" to calculate the reserved log space. This isn't entirely necessary, but it has two benefits: First, it simplifies the gfs2_log_refund function greatly. Second, it allows for easier debugging because I could sprinkle the code with calls to this function to make sure the accounting is proper (by adding asserts and printks) at strategic point of the code. 5. In log_pull_tail there apparently was a kludge to fix up the accounting based on a "pull" parameter. The buffer accounting is now done properly, so the kludge was removed. 6. File sync operations were making a call to gfs2_log_flush that writes another journal header. Since that header was unplanned for (reserved) by the reserve/refund sequence, the free space had to be decremented so that when log_pull_tail gets called, the free space is be adjusted properly. (Did I hear you call that a kludge? well, maybe, but a lot more justifiable than the one I removed). 7. In the gfs2_log_shutdown code, it optionally syncs the log by specifying the PULL parameter to log_write_header. I'm not sure this is necessary anymore. It just seems to me there could be cases where shutdown is called while there are outstanding log buffers. 8. In the (data)buf_lo_before_commit functions, I changed some offset values from being calculated on the fly to being constants. That simplified some code and we might as well let the compiler do the calculation once rather than redoing those cycles at run time. 9. This version has my rewritten databuf_lo_add function. This version is much more like its predecessor, buf_lo_add, which makes it easier to understand. Again, this might not be necessary, but it seems as if this one works as well as the previous one, maybe even better, so I decided to leave it in. 10. In databuf_lo_before_commit, a previous data corruption problem was caused by going off the end of the buffer. The proper solution is to have the proper limit in place, rather than stopping earlier. (Thus my previous attempt to fix it is wrong). If you don't wrap the buffer, you're stopping too early and that causes more log buffer accounting problems. 11. In lops.h there are two new (previously mentioned) constants for figuring out the data offset for the journal buffers. 12. There are also two new functions, buf_limit and databuf_limit to calculate how many entries will fit in the buffer. 13. In function gfs2_meta_wipe, it needs to distinguish between pinned metadata buffers and journaled data buffers for proper journal buffer accounting. It can't use the JDATA gfs2_inode flag because it's sometimes passed the "real" inode and sometimes the "metadata inode" and the inode flags will be random bits in a metadata gfs2_inode. It needs to base its decision on which was passed in. Signed-off-by: Bob Peterson <rpeterso@redhat.com> Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
2007-06-19 03:50:20 +08:00
unsigned int reserved;
unsigned int unused;
unsigned int maxres;
gfs2_log_lock(sdp);
if (sdp->sd_log_tr) {
gfs2_merge_trans(sdp, tr);
} else if (tr->tr_num_buf_new || tr->tr_num_databuf_new) {
gfs2_assert_withdraw(sdp, !test_bit(TR_ONSTACK, &tr->tr_flags));
sdp->sd_log_tr = tr;
set_bit(TR_ATTACHED, &tr->tr_flags);
}
[GFS2] assertion failure after writing to journaled file, umount This patch passes all my nasty tests that were causing the code to fail under one circumstance or another. Here is a complete summary of all changes from today's git tree, in order of appearance: 1. There are now separate variables for metadata buffer accounting. 2. Variable sd_log_num_hdrs is no longer needed, since the header accounting is taken care of by the reserve/refund sequence. 3. Fixed a tiny grammatical problem in a comment. 4. Added a new function "calc_reserved" to calculate the reserved log space. This isn't entirely necessary, but it has two benefits: First, it simplifies the gfs2_log_refund function greatly. Second, it allows for easier debugging because I could sprinkle the code with calls to this function to make sure the accounting is proper (by adding asserts and printks) at strategic point of the code. 5. In log_pull_tail there apparently was a kludge to fix up the accounting based on a "pull" parameter. The buffer accounting is now done properly, so the kludge was removed. 6. File sync operations were making a call to gfs2_log_flush that writes another journal header. Since that header was unplanned for (reserved) by the reserve/refund sequence, the free space had to be decremented so that when log_pull_tail gets called, the free space is be adjusted properly. (Did I hear you call that a kludge? well, maybe, but a lot more justifiable than the one I removed). 7. In the gfs2_log_shutdown code, it optionally syncs the log by specifying the PULL parameter to log_write_header. I'm not sure this is necessary anymore. It just seems to me there could be cases where shutdown is called while there are outstanding log buffers. 8. In the (data)buf_lo_before_commit functions, I changed some offset values from being calculated on the fly to being constants. That simplified some code and we might as well let the compiler do the calculation once rather than redoing those cycles at run time. 9. This version has my rewritten databuf_lo_add function. This version is much more like its predecessor, buf_lo_add, which makes it easier to understand. Again, this might not be necessary, but it seems as if this one works as well as the previous one, maybe even better, so I decided to leave it in. 10. In databuf_lo_before_commit, a previous data corruption problem was caused by going off the end of the buffer. The proper solution is to have the proper limit in place, rather than stopping earlier. (Thus my previous attempt to fix it is wrong). If you don't wrap the buffer, you're stopping too early and that causes more log buffer accounting problems. 11. In lops.h there are two new (previously mentioned) constants for figuring out the data offset for the journal buffers. 12. There are also two new functions, buf_limit and databuf_limit to calculate how many entries will fit in the buffer. 13. In function gfs2_meta_wipe, it needs to distinguish between pinned metadata buffers and journaled data buffers for proper journal buffer accounting. It can't use the JDATA gfs2_inode flag because it's sometimes passed the "real" inode and sometimes the "metadata inode" and the inode flags will be random bits in a metadata gfs2_inode. It needs to base its decision on which was passed in. Signed-off-by: Bob Peterson <rpeterso@redhat.com> Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
2007-06-19 03:50:20 +08:00
reserved = calc_reserved(sdp);
maxres = sdp->sd_log_blks_reserved + tr->tr_reserved;
gfs2_assert_withdraw(sdp, maxres >= reserved);
unused = maxres - reserved;
if (unused)
gfs2_log_release(sdp, unused);
sdp->sd_log_blks_reserved = reserved;
gfs2_log_unlock(sdp);
}
/**
* gfs2_log_commit - Commit a transaction to the log
* @sdp: the filesystem
* @tr: the transaction
*
* We wake up gfs2_logd if the number of pinned blocks exceed thresh1
* or the total number of used blocks (pinned blocks plus AIL blocks)
* is greater than thresh2.
*
* At mount time thresh1 is 2/5ths of journal size, thresh2 is 4/5ths of
* journal size.
*
* Returns: errno
*/
void gfs2_log_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
{
log_refund(sdp, tr);
if (atomic_read(&sdp->sd_log_pinned) > atomic_read(&sdp->sd_log_thresh1) ||
((sdp->sd_jdesc->jd_blocks - atomic_read(&sdp->sd_log_blks_free)) >
atomic_read(&sdp->sd_log_thresh2)))
wake_up(&sdp->sd_logd_waitq);
}
/**
* gfs2_log_shutdown - write a shutdown header into a journal
* @sdp: the filesystem
*
*/
static void gfs2_log_shutdown(struct gfs2_sbd *sdp)
{
gfs2_assert_withdraw(sdp, !sdp->sd_log_blks_reserved);
gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
gfs2_assert_withdraw(sdp, list_empty(&sdp->sd_ail1_list));
log_write_header(sdp, GFS2_LOG_HEAD_UNMOUNT | GFS2_LFC_SHUTDOWN);
log_pull_tail(sdp);
gfs2_assert_warn(sdp, sdp->sd_log_head == sdp->sd_log_tail);
gfs2_assert_warn(sdp, list_empty(&sdp->sd_ail2_list));
}
static inline int gfs2_jrnl_flush_reqd(struct gfs2_sbd *sdp)
{
return (atomic_read(&sdp->sd_log_pinned) +
atomic_read(&sdp->sd_log_blks_needed) >=
atomic_read(&sdp->sd_log_thresh1));
}
static inline int gfs2_ail_flush_reqd(struct gfs2_sbd *sdp)
{
unsigned int used_blocks = sdp->sd_jdesc->jd_blocks - atomic_read(&sdp->sd_log_blks_free);
if (test_and_clear_bit(SDF_FORCE_AIL_FLUSH, &sdp->sd_flags))
return 1;
return used_blocks + atomic_read(&sdp->sd_log_blks_needed) >=
atomic_read(&sdp->sd_log_thresh2);
}
/**
* gfs2_logd - Update log tail as Active Items get flushed to in-place blocks
* @data: Pointer to GFS2 superblock
*
* Also, periodically check to make sure that we're using the most recent
* journal index.
*/
int gfs2_logd(void *data)
{
struct gfs2_sbd *sdp = data;
unsigned long t = 1;
DEFINE_WAIT(wait);
while (!kthread_should_stop()) {
if (gfs2_withdrawn(sdp)) {
msleep_interruptible(HZ);
continue;
}
/* Check for errors writing to the journal */
if (sdp->sd_log_error) {
gfs2_lm(sdp,
"GFS2: fsid=%s: error %d: "
"withdrawing the file system to "
"prevent further damage.\n",
sdp->sd_fsname, sdp->sd_log_error);
gfs2_withdraw(sdp);
continue;
}
if (gfs2_jrnl_flush_reqd(sdp) || t == 0) {
gfs2: Issue revokes more intelligently Before this patch, function gfs2_write_revokes would call gfs2_ail1_empty, then traverse the sd_ail1_list looking for transactions that had bds which were no longer queued to a glock. And if it found some, it would try to issue revokes for them, up to a predetermined maximum. There were two problems with how it did this. First was the fact that gfs2_ail1_empty moves transactions which have nothing remaining on the ail1 list from the sd_ail1_list to the sd_ail2_list, thus making its traversal of sd_ail1_list miss them completely, and therefore, never issue revokes for them. Second was the fact that there were three traversals (or partial traversals) of the sd_ail1_list, each of which took and then released the sd_ail_lock lock: First inside gfs2_ail1_empty, second to determine if there are any revokes to be issued, and third to actually issue them. All this taking and releasing of the sd_ail_lock meant other processes could modify the lists and the conditions in which we're working. This patch simplies the whole process by adding a new parameter to function gfs2_ail1_empty, max_revokes. For normal calls, this is passed in as 0, meaning we don't want to issue any revokes. For function gfs2_write_revokes, we pass in the maximum number of revokes we can, thus allowing gfs2_ail1_empty to add the revokes where needed. This simplies the code, allows for a single holding of the sd_ail_lock, and allows gfs2_ail1_empty to add revokes for all the necessary bd items without missing any. Signed-off-by: Bob Peterson <rpeterso@redhat.com> Reviewed-by: Andreas Gruenbacher <agruenba@redhat.com>
2019-02-22 05:28:07 +08:00
gfs2_ail1_empty(sdp, 0);
gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
GFS2_LFC_LOGD_JFLUSH_REQD);
}
if (gfs2_ail_flush_reqd(sdp)) {
gfs2_ail1_start(sdp);
gfs2_ail1_wait(sdp);
gfs2: Issue revokes more intelligently Before this patch, function gfs2_write_revokes would call gfs2_ail1_empty, then traverse the sd_ail1_list looking for transactions that had bds which were no longer queued to a glock. And if it found some, it would try to issue revokes for them, up to a predetermined maximum. There were two problems with how it did this. First was the fact that gfs2_ail1_empty moves transactions which have nothing remaining on the ail1 list from the sd_ail1_list to the sd_ail2_list, thus making its traversal of sd_ail1_list miss them completely, and therefore, never issue revokes for them. Second was the fact that there were three traversals (or partial traversals) of the sd_ail1_list, each of which took and then released the sd_ail_lock lock: First inside gfs2_ail1_empty, second to determine if there are any revokes to be issued, and third to actually issue them. All this taking and releasing of the sd_ail_lock meant other processes could modify the lists and the conditions in which we're working. This patch simplies the whole process by adding a new parameter to function gfs2_ail1_empty, max_revokes. For normal calls, this is passed in as 0, meaning we don't want to issue any revokes. For function gfs2_write_revokes, we pass in the maximum number of revokes we can, thus allowing gfs2_ail1_empty to add the revokes where needed. This simplies the code, allows for a single holding of the sd_ail_lock, and allows gfs2_ail1_empty to add revokes for all the necessary bd items without missing any. Signed-off-by: Bob Peterson <rpeterso@redhat.com> Reviewed-by: Andreas Gruenbacher <agruenba@redhat.com>
2019-02-22 05:28:07 +08:00
gfs2_ail1_empty(sdp, 0);
gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
GFS2_LFC_LOGD_AIL_FLUSH_REQD);
}
t = gfs2_tune_get(sdp, gt_logd_secs) * HZ;
try_to_freeze();
do {
prepare_to_wait(&sdp->sd_logd_waitq, &wait,
TASK_INTERRUPTIBLE);
if (!gfs2_ail_flush_reqd(sdp) &&
!gfs2_jrnl_flush_reqd(sdp) &&
!kthread_should_stop())
t = schedule_timeout(t);
} while(t && !gfs2_ail_flush_reqd(sdp) &&
!gfs2_jrnl_flush_reqd(sdp) &&
!kthread_should_stop());
finish_wait(&sdp->sd_logd_waitq, &wait);
}
return 0;
}