linux/fs/iomap.c

2199 lines
55 KiB
C
Raw Normal View History

// SPDX-License-Identifier: GPL-2.0
/*
* Copyright (C) 2010 Red Hat, Inc.
* Copyright (c) 2016-2018 Christoph Hellwig.
*/
#include <linux/module.h>
#include <linux/compiler.h>
#include <linux/fs.h>
#include <linux/iomap.h>
#include <linux/uaccess.h>
#include <linux/gfp.h>
#include <linux/migrate.h>
#include <linux/mm.h>
#include <linux/mm_inline.h>
#include <linux/swap.h>
#include <linux/pagemap.h>
#include <linux/pagevec.h>
#include <linux/file.h>
#include <linux/uio.h>
#include <linux/backing-dev.h>
#include <linux/buffer_head.h>
#include <linux/task_io_accounting_ops.h>
#include <linux/dax.h>
#include <linux/sched/signal.h>
#include "internal.h"
/*
* Execute a iomap write on a segment of the mapping that spans a
* contiguous range of pages that have identical block mapping state.
*
* This avoids the need to map pages individually, do individual allocations
* for each page and most importantly avoid the need for filesystem specific
* locking per page. Instead, all the operations are amortised over the entire
* range of pages. It is assumed that the filesystems will lock whatever
* resources they require in the iomap_begin call, and release them in the
* iomap_end call.
*/
loff_t
iomap_apply(struct inode *inode, loff_t pos, loff_t length, unsigned flags,
const struct iomap_ops *ops, void *data, iomap_actor_t actor)
{
struct iomap iomap = { 0 };
loff_t written = 0, ret;
/*
* Need to map a range from start position for length bytes. This can
* span multiple pages - it is only guaranteed to return a range of a
* single type of pages (e.g. all into a hole, all mapped or all
* unwritten). Failure at this point has nothing to undo.
*
* If allocation is required for this range, reserve the space now so
* that the allocation is guaranteed to succeed later on. Once we copy
* the data into the page cache pages, then we cannot fail otherwise we
* expose transient stale data. If the reserve fails, we can safely
* back out at this point as there is nothing to undo.
*/
ret = ops->iomap_begin(inode, pos, length, flags, &iomap);
if (ret)
return ret;
if (WARN_ON(iomap.offset > pos))
return -EIO;
if (WARN_ON(iomap.length == 0))
return -EIO;
/*
* Cut down the length to the one actually provided by the filesystem,
* as it might not be able to give us the whole size that we requested.
*/
if (iomap.offset + iomap.length < pos + length)
length = iomap.offset + iomap.length - pos;
/*
* Now that we have guaranteed that the space allocation will succeed.
* we can do the copy-in page by page without having to worry about
* failures exposing transient data.
*/
written = actor(inode, pos, length, data, &iomap);
/*
* Now the data has been copied, commit the range we've copied. This
* should not fail unless the filesystem has had a fatal error.
*/
if (ops->iomap_end) {
ret = ops->iomap_end(inode, pos, length,
written > 0 ? written : 0,
flags, &iomap);
}
return written ? written : ret;
}
static sector_t
iomap_sector(struct iomap *iomap, loff_t pos)
{
return (iomap->addr + pos - iomap->offset) >> SECTOR_SHIFT;
}
static struct iomap_page *
iomap_page_create(struct inode *inode, struct page *page)
{
struct iomap_page *iop = to_iomap_page(page);
if (iop || i_blocksize(inode) == PAGE_SIZE)
return iop;
iop = kmalloc(sizeof(*iop), GFP_NOFS | __GFP_NOFAIL);
atomic_set(&iop->read_count, 0);
atomic_set(&iop->write_count, 0);
bitmap_zero(iop->uptodate, PAGE_SIZE / SECTOR_SIZE);
/*
* migrate_page_move_mapping() assumes that pages with private data have
* their count elevated by 1.
*/
get_page(page);
set_page_private(page, (unsigned long)iop);
SetPagePrivate(page);
return iop;
}
static void
iomap_page_release(struct page *page)
{
struct iomap_page *iop = to_iomap_page(page);
if (!iop)
return;
WARN_ON_ONCE(atomic_read(&iop->read_count));
WARN_ON_ONCE(atomic_read(&iop->write_count));
ClearPagePrivate(page);
set_page_private(page, 0);
put_page(page);
kfree(iop);
}
/*
* Calculate the range inside the page that we actually need to read.
*/
static void
iomap_adjust_read_range(struct inode *inode, struct iomap_page *iop,
loff_t *pos, loff_t length, unsigned *offp, unsigned *lenp)
{
iomap: readpages doesn't zero page tail beyond EOF When we read the EOF page of the file via readpages, we need to zero the region beyond EOF that we either do not read or should not contain data so that mmap does not expose stale data to user applications. However, iomap_adjust_read_range() fails to detect EOF correctly, and so fsx on 1k block size filesystems fails very quickly with mapreads exposing data beyond EOF. There are two problems here. Firstly, when calculating the end block of the EOF byte, we have to round the size by one to avoid a block aligned EOF from reporting a block too large. i.e. a size of 1024 bytes is 1 block, which in index terms is block 0. Therefore we have to calculate the end block from (isize - 1), not isize. The second bug is determining if the current page spans EOF, and so whether we need split it into two half, one for the IO, and the other for zeroing. Unfortunately, the code that checks whether we should split the block doesn't actually check if we span EOF, it just checks if the read spans the /offset in the page/ that EOF sits on. So it splits every read into two if EOF is not page aligned, regardless of whether we are reading the EOF block or not. Hence we need to restrict the "does the read span EOF" check to just the page that spans EOF, not every page we read. This patch results in correct EOF detection through readpages: xfs_vm_readpages: dev 259:0 ino 0x43 nr_pages 24 xfs_iomap_found: dev 259:0 ino 0x43 size 0x66c00 offset 0x4f000 count 98304 type hole startoff 0x13c startblock 1368 blockcount 0x4 iomap_readpage_actor: orig pos 323584 pos 323584, length 4096, poff 0 plen 4096, isize 420864 xfs_iomap_found: dev 259:0 ino 0x43 size 0x66c00 offset 0x50000 count 94208 type hole startoff 0x140 startblock 1497 blockcount 0x5c iomap_readpage_actor: orig pos 327680 pos 327680, length 94208, poff 0 plen 4096, isize 420864 iomap_readpage_actor: orig pos 331776 pos 331776, length 90112, poff 0 plen 4096, isize 420864 iomap_readpage_actor: orig pos 335872 pos 335872, length 86016, poff 0 plen 4096, isize 420864 iomap_readpage_actor: orig pos 339968 pos 339968, length 81920, poff 0 plen 4096, isize 420864 iomap_readpage_actor: orig pos 344064 pos 344064, length 77824, poff 0 plen 4096, isize 420864 iomap_readpage_actor: orig pos 348160 pos 348160, length 73728, poff 0 plen 4096, isize 420864 iomap_readpage_actor: orig pos 352256 pos 352256, length 69632, poff 0 plen 4096, isize 420864 iomap_readpage_actor: orig pos 356352 pos 356352, length 65536, poff 0 plen 4096, isize 420864 iomap_readpage_actor: orig pos 360448 pos 360448, length 61440, poff 0 plen 4096, isize 420864 iomap_readpage_actor: orig pos 364544 pos 364544, length 57344, poff 0 plen 4096, isize 420864 iomap_readpage_actor: orig pos 368640 pos 368640, length 53248, poff 0 plen 4096, isize 420864 iomap_readpage_actor: orig pos 372736 pos 372736, length 49152, poff 0 plen 4096, isize 420864 iomap_readpage_actor: orig pos 376832 pos 376832, length 45056, poff 0 plen 4096, isize 420864 iomap_readpage_actor: orig pos 380928 pos 380928, length 40960, poff 0 plen 4096, isize 420864 iomap_readpage_actor: orig pos 385024 pos 385024, length 36864, poff 0 plen 4096, isize 420864 iomap_readpage_actor: orig pos 389120 pos 389120, length 32768, poff 0 plen 4096, isize 420864 iomap_readpage_actor: orig pos 393216 pos 393216, length 28672, poff 0 plen 4096, isize 420864 iomap_readpage_actor: orig pos 397312 pos 397312, length 24576, poff 0 plen 4096, isize 420864 iomap_readpage_actor: orig pos 401408 pos 401408, length 20480, poff 0 plen 4096, isize 420864 iomap_readpage_actor: orig pos 405504 pos 405504, length 16384, poff 0 plen 4096, isize 420864 iomap_readpage_actor: orig pos 409600 pos 409600, length 12288, poff 0 plen 4096, isize 420864 iomap_readpage_actor: orig pos 413696 pos 413696, length 8192, poff 0 plen 4096, isize 420864 iomap_readpage_actor: orig pos 417792 pos 417792, length 4096, poff 0 plen 3072, isize 420864 iomap_readpage_actor: orig pos 420864 pos 420864, length 1024, poff 3072 plen 1024, isize 420864 As you can see, it now does full page reads until the last one which is split correctly at the block aligned EOF, reading 3072 bytes and zeroing the last 1024 bytes. The original version of the patch got this right, but it got another case wrong. The EOF detection crossing really needs to the the original length as plen, while it starts at the end of the block, will be shortened as up-to-date blocks are found on the page. This means "orig_pos + plen" no longer points to the end of the page, and so will not correctly detect EOF crossing. Hence we have to use the length passed in to detect this partial page case: xfs_filemap_fault: dev 259:1 ino 0x43 write_fault 0 xfs_vm_readpage: dev 259:1 ino 0x43 nr_pages 1 xfs_iomap_found: dev 259:1 ino 0x43 size 0x2cc00 offset 0x2c000 count 4096 type hole startoff 0xb0 startblock 282 blockcount 0x4 iomap_readpage_actor: orig pos 180224 pos 181248, length 4096, poff 1024 plen 2048, isize 183296 xfs_iomap_found: dev 259:1 ino 0x43 size 0x2cc00 offset 0x2cc00 count 1024 type hole startoff 0xb3 startblock 285 blockcount 0x1 iomap_readpage_actor: orig pos 183296 pos 183296, length 1024, poff 3072 plen 1024, isize 183296 Heere we see a trace where the first block on the EOF page is up to date, hence poff = 1024 bytes. The offset into the page of EOF is 3072, so the range we want to read is 1024 - 3071, and the range we want to zero is 3072 - 4095. You can see this is split correctly now. This fixes the stale data beyond EOF problem that fsx quickly uncovers on 1k block size filesystems. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2018-11-22 00:06:37 +08:00
loff_t orig_pos = *pos;
loff_t isize = i_size_read(inode);
unsigned block_bits = inode->i_blkbits;
unsigned block_size = (1 << block_bits);
unsigned poff = offset_in_page(*pos);
unsigned plen = min_t(loff_t, PAGE_SIZE - poff, length);
unsigned first = poff >> block_bits;
unsigned last = (poff + plen - 1) >> block_bits;
/*
* If the block size is smaller than the page size we need to check the
* per-block uptodate status and adjust the offset and length if needed
* to avoid reading in already uptodate ranges.
*/
if (iop) {
unsigned int i;
/* move forward for each leading block marked uptodate */
for (i = first; i <= last; i++) {
if (!test_bit(i, iop->uptodate))
break;
*pos += block_size;
poff += block_size;
plen -= block_size;
first++;
}
/* truncate len if we find any trailing uptodate block(s) */
for ( ; i <= last; i++) {
if (test_bit(i, iop->uptodate)) {
plen -= (last - i + 1) * block_size;
last = i - 1;
break;
}
}
}
/*
* If the extent spans the block that contains the i_size we need to
* handle both halves separately so that we properly zero data in the
* page cache for blocks that are entirely outside of i_size.
*/
iomap: readpages doesn't zero page tail beyond EOF When we read the EOF page of the file via readpages, we need to zero the region beyond EOF that we either do not read or should not contain data so that mmap does not expose stale data to user applications. However, iomap_adjust_read_range() fails to detect EOF correctly, and so fsx on 1k block size filesystems fails very quickly with mapreads exposing data beyond EOF. There are two problems here. Firstly, when calculating the end block of the EOF byte, we have to round the size by one to avoid a block aligned EOF from reporting a block too large. i.e. a size of 1024 bytes is 1 block, which in index terms is block 0. Therefore we have to calculate the end block from (isize - 1), not isize. The second bug is determining if the current page spans EOF, and so whether we need split it into two half, one for the IO, and the other for zeroing. Unfortunately, the code that checks whether we should split the block doesn't actually check if we span EOF, it just checks if the read spans the /offset in the page/ that EOF sits on. So it splits every read into two if EOF is not page aligned, regardless of whether we are reading the EOF block or not. Hence we need to restrict the "does the read span EOF" check to just the page that spans EOF, not every page we read. This patch results in correct EOF detection through readpages: xfs_vm_readpages: dev 259:0 ino 0x43 nr_pages 24 xfs_iomap_found: dev 259:0 ino 0x43 size 0x66c00 offset 0x4f000 count 98304 type hole startoff 0x13c startblock 1368 blockcount 0x4 iomap_readpage_actor: orig pos 323584 pos 323584, length 4096, poff 0 plen 4096, isize 420864 xfs_iomap_found: dev 259:0 ino 0x43 size 0x66c00 offset 0x50000 count 94208 type hole startoff 0x140 startblock 1497 blockcount 0x5c iomap_readpage_actor: orig pos 327680 pos 327680, length 94208, poff 0 plen 4096, isize 420864 iomap_readpage_actor: orig pos 331776 pos 331776, length 90112, poff 0 plen 4096, isize 420864 iomap_readpage_actor: orig pos 335872 pos 335872, length 86016, poff 0 plen 4096, isize 420864 iomap_readpage_actor: orig pos 339968 pos 339968, length 81920, poff 0 plen 4096, isize 420864 iomap_readpage_actor: orig pos 344064 pos 344064, length 77824, poff 0 plen 4096, isize 420864 iomap_readpage_actor: orig pos 348160 pos 348160, length 73728, poff 0 plen 4096, isize 420864 iomap_readpage_actor: orig pos 352256 pos 352256, length 69632, poff 0 plen 4096, isize 420864 iomap_readpage_actor: orig pos 356352 pos 356352, length 65536, poff 0 plen 4096, isize 420864 iomap_readpage_actor: orig pos 360448 pos 360448, length 61440, poff 0 plen 4096, isize 420864 iomap_readpage_actor: orig pos 364544 pos 364544, length 57344, poff 0 plen 4096, isize 420864 iomap_readpage_actor: orig pos 368640 pos 368640, length 53248, poff 0 plen 4096, isize 420864 iomap_readpage_actor: orig pos 372736 pos 372736, length 49152, poff 0 plen 4096, isize 420864 iomap_readpage_actor: orig pos 376832 pos 376832, length 45056, poff 0 plen 4096, isize 420864 iomap_readpage_actor: orig pos 380928 pos 380928, length 40960, poff 0 plen 4096, isize 420864 iomap_readpage_actor: orig pos 385024 pos 385024, length 36864, poff 0 plen 4096, isize 420864 iomap_readpage_actor: orig pos 389120 pos 389120, length 32768, poff 0 plen 4096, isize 420864 iomap_readpage_actor: orig pos 393216 pos 393216, length 28672, poff 0 plen 4096, isize 420864 iomap_readpage_actor: orig pos 397312 pos 397312, length 24576, poff 0 plen 4096, isize 420864 iomap_readpage_actor: orig pos 401408 pos 401408, length 20480, poff 0 plen 4096, isize 420864 iomap_readpage_actor: orig pos 405504 pos 405504, length 16384, poff 0 plen 4096, isize 420864 iomap_readpage_actor: orig pos 409600 pos 409600, length 12288, poff 0 plen 4096, isize 420864 iomap_readpage_actor: orig pos 413696 pos 413696, length 8192, poff 0 plen 4096, isize 420864 iomap_readpage_actor: orig pos 417792 pos 417792, length 4096, poff 0 plen 3072, isize 420864 iomap_readpage_actor: orig pos 420864 pos 420864, length 1024, poff 3072 plen 1024, isize 420864 As you can see, it now does full page reads until the last one which is split correctly at the block aligned EOF, reading 3072 bytes and zeroing the last 1024 bytes. The original version of the patch got this right, but it got another case wrong. The EOF detection crossing really needs to the the original length as plen, while it starts at the end of the block, will be shortened as up-to-date blocks are found on the page. This means "orig_pos + plen" no longer points to the end of the page, and so will not correctly detect EOF crossing. Hence we have to use the length passed in to detect this partial page case: xfs_filemap_fault: dev 259:1 ino 0x43 write_fault 0 xfs_vm_readpage: dev 259:1 ino 0x43 nr_pages 1 xfs_iomap_found: dev 259:1 ino 0x43 size 0x2cc00 offset 0x2c000 count 4096 type hole startoff 0xb0 startblock 282 blockcount 0x4 iomap_readpage_actor: orig pos 180224 pos 181248, length 4096, poff 1024 plen 2048, isize 183296 xfs_iomap_found: dev 259:1 ino 0x43 size 0x2cc00 offset 0x2cc00 count 1024 type hole startoff 0xb3 startblock 285 blockcount 0x1 iomap_readpage_actor: orig pos 183296 pos 183296, length 1024, poff 3072 plen 1024, isize 183296 Heere we see a trace where the first block on the EOF page is up to date, hence poff = 1024 bytes. The offset into the page of EOF is 3072, so the range we want to read is 1024 - 3071, and the range we want to zero is 3072 - 4095. You can see this is split correctly now. This fixes the stale data beyond EOF problem that fsx quickly uncovers on 1k block size filesystems. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2018-11-22 00:06:37 +08:00
if (orig_pos <= isize && orig_pos + length > isize) {
unsigned end = offset_in_page(isize - 1) >> block_bits;
if (first <= end && last > end)
plen -= (last - end) * block_size;
}
*offp = poff;
*lenp = plen;
}
static void
iomap_set_range_uptodate(struct page *page, unsigned off, unsigned len)
{
struct iomap_page *iop = to_iomap_page(page);
struct inode *inode = page->mapping->host;
unsigned first = off >> inode->i_blkbits;
unsigned last = (off + len - 1) >> inode->i_blkbits;
unsigned int i;
bool uptodate = true;
if (iop) {
for (i = 0; i < PAGE_SIZE / i_blocksize(inode); i++) {
if (i >= first && i <= last)
set_bit(i, iop->uptodate);
else if (!test_bit(i, iop->uptodate))
uptodate = false;
}
}
if (uptodate && !PageError(page))
SetPageUptodate(page);
}
static void
iomap_read_finish(struct iomap_page *iop, struct page *page)
{
if (!iop || atomic_dec_and_test(&iop->read_count))
unlock_page(page);
}
static void
iomap_read_page_end_io(struct bio_vec *bvec, int error)
{
struct page *page = bvec->bv_page;
struct iomap_page *iop = to_iomap_page(page);
if (unlikely(error)) {
ClearPageUptodate(page);
SetPageError(page);
} else {
iomap_set_range_uptodate(page, bvec->bv_offset, bvec->bv_len);
}
iomap_read_finish(iop, page);
}
static void
iomap_read_end_io(struct bio *bio)
{
int error = blk_status_to_errno(bio->bi_status);
struct bio_vec *bvec;
struct bvec_iter_all iter_all;
bio_for_each_segment_all(bvec, bio, iter_all)
iomap_read_page_end_io(bvec, error);
bio_put(bio);
}
struct iomap_readpage_ctx {
struct page *cur_page;
bool cur_page_in_bio;
bool is_readahead;
struct bio *bio;
struct list_head *pages;
};
static void
iomap_read_inline_data(struct inode *inode, struct page *page,
struct iomap *iomap)
{
size_t size = i_size_read(inode);
void *addr;
if (PageUptodate(page))
return;
BUG_ON(page->index);
BUG_ON(size > PAGE_SIZE - offset_in_page(iomap->inline_data));
addr = kmap_atomic(page);
memcpy(addr, iomap->inline_data, size);
memset(addr + size, 0, PAGE_SIZE - size);
kunmap_atomic(addr);
SetPageUptodate(page);
}
static loff_t
iomap_readpage_actor(struct inode *inode, loff_t pos, loff_t length, void *data,
struct iomap *iomap)
{
struct iomap_readpage_ctx *ctx = data;
struct page *page = ctx->cur_page;
struct iomap_page *iop = iomap_page_create(inode, page);
bool same_page = false, is_contig = false;
loff_t orig_pos = pos;
unsigned poff, plen;
sector_t sector;
if (iomap->type == IOMAP_INLINE) {
WARN_ON_ONCE(pos);
iomap_read_inline_data(inode, page, iomap);
return PAGE_SIZE;
}
/* zero post-eof blocks as the page may be mapped */
iomap_adjust_read_range(inode, iop, &pos, length, &poff, &plen);
if (plen == 0)
goto done;
if (iomap->type != IOMAP_MAPPED || pos >= i_size_read(inode)) {
zero_user(page, poff, plen);
iomap_set_range_uptodate(page, poff, plen);
goto done;
}
ctx->cur_page_in_bio = true;
/*
* Try to merge into a previous segment if we can.
*/
sector = iomap_sector(iomap, pos);
if (ctx->bio && bio_end_sector(ctx->bio) == sector)
is_contig = true;
if (is_contig &&
__bio_try_merge_page(ctx->bio, page, plen, poff, &same_page)) {
if (!same_page && iop)
atomic_inc(&iop->read_count);
goto done;
}
/*
* If we start a new segment we need to increase the read count, and we
* need to do so before submitting any previous full bio to make sure
* that we don't prematurely unlock the page.
*/
if (iop)
atomic_inc(&iop->read_count);
if (!ctx->bio || !is_contig || bio_full(ctx->bio)) {
gfp_t gfp = mapping_gfp_constraint(page->mapping, GFP_KERNEL);
int nr_vecs = (length + PAGE_SIZE - 1) >> PAGE_SHIFT;
if (ctx->bio)
submit_bio(ctx->bio);
if (ctx->is_readahead) /* same as readahead_gfp_mask */
gfp |= __GFP_NORETRY | __GFP_NOWARN;
ctx->bio = bio_alloc(gfp, min(BIO_MAX_PAGES, nr_vecs));
ctx->bio->bi_opf = REQ_OP_READ;
if (ctx->is_readahead)
ctx->bio->bi_opf |= REQ_RAHEAD;
ctx->bio->bi_iter.bi_sector = sector;
bio_set_dev(ctx->bio, iomap->bdev);
ctx->bio->bi_end_io = iomap_read_end_io;
}
bio_add_page(ctx->bio, page, plen, poff);
done:
/*
* Move the caller beyond our range so that it keeps making progress.
* For that we have to include any leading non-uptodate ranges, but
* we can skip trailing ones as they will be handled in the next
* iteration.
*/
return pos - orig_pos + plen;
}
int
iomap_readpage(struct page *page, const struct iomap_ops *ops)
{
struct iomap_readpage_ctx ctx = { .cur_page = page };
struct inode *inode = page->mapping->host;
unsigned poff;
loff_t ret;
for (poff = 0; poff < PAGE_SIZE; poff += ret) {
ret = iomap_apply(inode, page_offset(page) + poff,
PAGE_SIZE - poff, 0, ops, &ctx,
iomap_readpage_actor);
if (ret <= 0) {
WARN_ON_ONCE(ret == 0);
SetPageError(page);
break;
}
}
if (ctx.bio) {
submit_bio(ctx.bio);
WARN_ON_ONCE(!ctx.cur_page_in_bio);
} else {
WARN_ON_ONCE(ctx.cur_page_in_bio);
unlock_page(page);
}
/*
* Just like mpage_readpages and block_read_full_page we always
* return 0 and just mark the page as PageError on errors. This
* should be cleaned up all through the stack eventually.
*/
return 0;
}
EXPORT_SYMBOL_GPL(iomap_readpage);
static struct page *
iomap_next_page(struct inode *inode, struct list_head *pages, loff_t pos,
loff_t length, loff_t *done)
{
while (!list_empty(pages)) {
struct page *page = lru_to_page(pages);
if (page_offset(page) >= (u64)pos + length)
break;
list_del(&page->lru);
if (!add_to_page_cache_lru(page, inode->i_mapping, page->index,
GFP_NOFS))
return page;
/*
* If we already have a page in the page cache at index we are
* done. Upper layers don't care if it is uptodate after the
* readpages call itself as every page gets checked again once
* actually needed.
*/
*done += PAGE_SIZE;
put_page(page);
}
return NULL;
}
static loff_t
iomap_readpages_actor(struct inode *inode, loff_t pos, loff_t length,
void *data, struct iomap *iomap)
{
struct iomap_readpage_ctx *ctx = data;
loff_t done, ret;
for (done = 0; done < length; done += ret) {
if (ctx->cur_page && offset_in_page(pos + done) == 0) {
if (!ctx->cur_page_in_bio)
unlock_page(ctx->cur_page);
put_page(ctx->cur_page);
ctx->cur_page = NULL;
}
if (!ctx->cur_page) {
ctx->cur_page = iomap_next_page(inode, ctx->pages,
pos, length, &done);
if (!ctx->cur_page)
break;
ctx->cur_page_in_bio = false;
}
ret = iomap_readpage_actor(inode, pos + done, length - done,
ctx, iomap);
}
return done;
}
int
iomap_readpages(struct address_space *mapping, struct list_head *pages,
unsigned nr_pages, const struct iomap_ops *ops)
{
struct iomap_readpage_ctx ctx = {
.pages = pages,
.is_readahead = true,
};
loff_t pos = page_offset(list_entry(pages->prev, struct page, lru));
loff_t last = page_offset(list_entry(pages->next, struct page, lru));
loff_t length = last - pos + PAGE_SIZE, ret = 0;
while (length > 0) {
ret = iomap_apply(mapping->host, pos, length, 0, ops,
&ctx, iomap_readpages_actor);
if (ret <= 0) {
WARN_ON_ONCE(ret == 0);
goto done;
}
pos += ret;
length -= ret;
}
ret = 0;
done:
if (ctx.bio)
submit_bio(ctx.bio);
if (ctx.cur_page) {
if (!ctx.cur_page_in_bio)
unlock_page(ctx.cur_page);
put_page(ctx.cur_page);
}
/*
* Check that we didn't lose a page due to the arcance calling
* conventions..
*/
WARN_ON_ONCE(!ret && !list_empty(ctx.pages));
return ret;
}
EXPORT_SYMBOL_GPL(iomap_readpages);
/*
* iomap_is_partially_uptodate checks whether blocks within a page are
* uptodate or not.
*
* Returns true if all blocks which correspond to a file portion
* we want to read within the page are uptodate.
*/
int
iomap_is_partially_uptodate(struct page *page, unsigned long from,
unsigned long count)
{
struct iomap_page *iop = to_iomap_page(page);
struct inode *inode = page->mapping->host;
unsigned len, first, last;
unsigned i;
/* Limit range to one page */
len = min_t(unsigned, PAGE_SIZE - from, count);
/* First and last blocks in range within page */
first = from >> inode->i_blkbits;
last = (from + len - 1) >> inode->i_blkbits;
if (iop) {
for (i = first; i <= last; i++)
if (!test_bit(i, iop->uptodate))
return 0;
return 1;
}
return 0;
}
EXPORT_SYMBOL_GPL(iomap_is_partially_uptodate);
int
iomap_releasepage(struct page *page, gfp_t gfp_mask)
{
/*
* mm accommodates an old ext3 case where clean pages might not have had
* the dirty bit cleared. Thus, it can send actual dirty pages to
* ->releasepage() via shrink_active_list(), skip those here.
*/
if (PageDirty(page) || PageWriteback(page))
return 0;
iomap_page_release(page);
return 1;
}
EXPORT_SYMBOL_GPL(iomap_releasepage);
void
iomap_invalidatepage(struct page *page, unsigned int offset, unsigned int len)
{
/*
* If we are invalidating the entire page, clear the dirty state from it
* and release it to avoid unnecessary buildup of the LRU.
*/
if (offset == 0 && len == PAGE_SIZE) {
WARN_ON_ONCE(PageWriteback(page));
cancel_dirty_page(page);
iomap_page_release(page);
}
}
EXPORT_SYMBOL_GPL(iomap_invalidatepage);
#ifdef CONFIG_MIGRATION
int
iomap_migrate_page(struct address_space *mapping, struct page *newpage,
struct page *page, enum migrate_mode mode)
{
int ret;
ret = migrate_page_move_mapping(mapping, newpage, page, mode, 0);
if (ret != MIGRATEPAGE_SUCCESS)
return ret;
if (page_has_private(page)) {
ClearPagePrivate(page);
get_page(newpage);
set_page_private(newpage, page_private(page));
set_page_private(page, 0);
put_page(page);
SetPagePrivate(newpage);
}
if (mode != MIGRATE_SYNC_NO_COPY)
migrate_page_copy(newpage, page);
else
migrate_page_states(newpage, page);
return MIGRATEPAGE_SUCCESS;
}
EXPORT_SYMBOL_GPL(iomap_migrate_page);
#endif /* CONFIG_MIGRATION */
static void
iomap_write_failed(struct inode *inode, loff_t pos, unsigned len)
{
loff_t i_size = i_size_read(inode);
/*
* Only truncate newly allocated pages beyoned EOF, even if the
* write started inside the existing inode size.
*/
if (pos + len > i_size)
truncate_pagecache_range(inode, max(pos, i_size), pos + len);
}
static int
iomap_read_page_sync(struct inode *inode, loff_t block_start, struct page *page,
unsigned poff, unsigned plen, unsigned from, unsigned to,
struct iomap *iomap)
{
struct bio_vec bvec;
struct bio bio;
if (iomap->type != IOMAP_MAPPED || block_start >= i_size_read(inode)) {
zero_user_segments(page, poff, from, to, poff + plen);
iomap_set_range_uptodate(page, poff, plen);
return 0;
}
bio_init(&bio, &bvec, 1);
bio.bi_opf = REQ_OP_READ;
bio.bi_iter.bi_sector = iomap_sector(iomap, block_start);
bio_set_dev(&bio, iomap->bdev);
__bio_add_page(&bio, page, plen, poff);
return submit_bio_wait(&bio);
}
static int
__iomap_write_begin(struct inode *inode, loff_t pos, unsigned len,
struct page *page, struct iomap *iomap)
{
struct iomap_page *iop = iomap_page_create(inode, page);
loff_t block_size = i_blocksize(inode);
loff_t block_start = pos & ~(block_size - 1);
loff_t block_end = (pos + len + block_size - 1) & ~(block_size - 1);
unsigned from = offset_in_page(pos), to = from + len, poff, plen;
int status = 0;
if (PageUptodate(page))
return 0;
do {
iomap_adjust_read_range(inode, iop, &block_start,
block_end - block_start, &poff, &plen);
if (plen == 0)
break;
if ((from > poff && from < poff + plen) ||
(to > poff && to < poff + plen)) {
status = iomap_read_page_sync(inode, block_start, page,
poff, plen, from, to, iomap);
if (status)
break;
}
} while ((block_start += plen) < block_end);
return status;
}
static int
iomap_write_begin(struct inode *inode, loff_t pos, unsigned len, unsigned flags,
struct page **pagep, struct iomap *iomap)
{
const struct iomap_page_ops *page_ops = iomap->page_ops;
pgoff_t index = pos >> PAGE_SHIFT;
struct page *page;
int status = 0;
BUG_ON(pos + len > iomap->offset + iomap->length);
fs: break out of iomap_file_buffered_write on fatal signals Tetsuo has noticed that an OOM stress test which performs large write requests can cause the full memory reserves depletion. He has tracked this down to the following path __alloc_pages_nodemask+0x436/0x4d0 alloc_pages_current+0x97/0x1b0 __page_cache_alloc+0x15d/0x1a0 mm/filemap.c:728 pagecache_get_page+0x5a/0x2b0 mm/filemap.c:1331 grab_cache_page_write_begin+0x23/0x40 mm/filemap.c:2773 iomap_write_begin+0x50/0xd0 fs/iomap.c:118 iomap_write_actor+0xb5/0x1a0 fs/iomap.c:190 ? iomap_write_end+0x80/0x80 fs/iomap.c:150 iomap_apply+0xb3/0x130 fs/iomap.c:79 iomap_file_buffered_write+0x68/0xa0 fs/iomap.c:243 ? iomap_write_end+0x80/0x80 xfs_file_buffered_aio_write+0x132/0x390 [xfs] ? remove_wait_queue+0x59/0x60 xfs_file_write_iter+0x90/0x130 [xfs] __vfs_write+0xe5/0x140 vfs_write+0xc7/0x1f0 ? syscall_trace_enter+0x1d0/0x380 SyS_write+0x58/0xc0 do_syscall_64+0x6c/0x200 entry_SYSCALL64_slow_path+0x25/0x25 the oom victim has access to all memory reserves to make a forward progress to exit easier. But iomap_file_buffered_write and other callers of iomap_apply loop to complete the full request. We need to check for fatal signals and back off with a short write instead. As the iomap_apply delegates all the work down to the actor we have to hook into those. All callers that work with the page cache are calling iomap_write_begin so we will check for signals there. dax_iomap_actor has to handle the situation explicitly because it copies data to the userspace directly. Other callers like iomap_page_mkwrite work on a single page or iomap_fiemap_actor do not allocate memory based on the given len. Fixes: 68a9f5e7007c ("xfs: implement iomap based buffered write path") Link: http://lkml.kernel.org/r/20170201092706.9966-2-mhocko@kernel.org Signed-off-by: Michal Hocko <mhocko@suse.com> Reported-by: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp> Reviewed-by: Christoph Hellwig <hch@lst.de> Cc: Al Viro <viro@zeniv.linux.org.uk> Cc: <stable@vger.kernel.org> [4.8+] Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-02-04 05:13:26 +08:00
if (fatal_signal_pending(current))
return -EINTR;
if (page_ops && page_ops->page_prepare) {
status = page_ops->page_prepare(inode, pos, len, iomap);
if (status)
return status;
}
page = grab_cache_page_write_begin(inode->i_mapping, index, flags);
if (!page) {
status = -ENOMEM;
goto out_no_page;
}
if (iomap->type == IOMAP_INLINE)
iomap_read_inline_data(inode, page, iomap);
else if (iomap->flags & IOMAP_F_BUFFER_HEAD)
status = __block_write_begin_int(page, pos, len, NULL, iomap);
else
status = __iomap_write_begin(inode, pos, len, page, iomap);
if (unlikely(status))
goto out_unlock;
*pagep = page;
return 0;
out_unlock:
unlock_page(page);
put_page(page);
iomap_write_failed(inode, pos, len);
out_no_page:
if (page_ops && page_ops->page_done)
page_ops->page_done(inode, pos, 0, NULL, iomap);
return status;
}
int
iomap_set_page_dirty(struct page *page)
{
struct address_space *mapping = page_mapping(page);
int newly_dirty;
if (unlikely(!mapping))
return !TestSetPageDirty(page);
/*
* Lock out page->mem_cgroup migration to keep PageDirty
* synchronized with per-memcg dirty page counters.
*/
lock_page_memcg(page);
newly_dirty = !TestSetPageDirty(page);
if (newly_dirty)
__set_page_dirty(page, mapping, 0);
unlock_page_memcg(page);
if (newly_dirty)
__mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
return newly_dirty;
}
EXPORT_SYMBOL_GPL(iomap_set_page_dirty);
static int
__iomap_write_end(struct inode *inode, loff_t pos, unsigned len,
unsigned copied, struct page *page, struct iomap *iomap)
{
flush_dcache_page(page);
/*
* The blocks that were entirely written will now be uptodate, so we
* don't have to worry about a readpage reading them and overwriting a
* partial write. However if we have encountered a short write and only
* partially written into a block, it will not be marked uptodate, so a
* readpage might come in and destroy our partial write.
*
* Do the simplest thing, and just treat any short write to a non
* uptodate page as a zero-length write, and force the caller to redo
* the whole thing.
*/
if (unlikely(copied < len && !PageUptodate(page)))
return 0;
iomap_set_range_uptodate(page, offset_in_page(pos), len);
iomap_set_page_dirty(page);
return copied;
}
static int
iomap_write_end_inline(struct inode *inode, struct page *page,
struct iomap *iomap, loff_t pos, unsigned copied)
{
void *addr;
WARN_ON_ONCE(!PageUptodate(page));
BUG_ON(pos + copied > PAGE_SIZE - offset_in_page(iomap->inline_data));
addr = kmap_atomic(page);
memcpy(iomap->inline_data + pos, addr + pos, copied);
kunmap_atomic(addr);
mark_inode_dirty(inode);
return copied;
}
static int
iomap_write_end(struct inode *inode, loff_t pos, unsigned len,
unsigned copied, struct page *page, struct iomap *iomap)
{
const struct iomap_page_ops *page_ops = iomap->page_ops;
int ret;
if (iomap->type == IOMAP_INLINE) {
ret = iomap_write_end_inline(inode, page, iomap, pos, copied);
} else if (iomap->flags & IOMAP_F_BUFFER_HEAD) {
ret = block_write_end(NULL, inode->i_mapping, pos, len, copied,
page, NULL);
} else {
ret = __iomap_write_end(inode, pos, len, copied, page, iomap);
}
__generic_write_end(inode, pos, ret, page);
if (page_ops && page_ops->page_done)
page_ops->page_done(inode, pos, copied, page, iomap);
put_page(page);
if (ret < len)
iomap_write_failed(inode, pos, len);
return ret;
}
static loff_t
iomap_write_actor(struct inode *inode, loff_t pos, loff_t length, void *data,
struct iomap *iomap)
{
struct iov_iter *i = data;
long status = 0;
ssize_t written = 0;
unsigned int flags = AOP_FLAG_NOFS;
do {
struct page *page;
unsigned long offset; /* Offset into pagecache page */
unsigned long bytes; /* Bytes to write to page */
size_t copied; /* Bytes copied from user */
offset = offset_in_page(pos);
bytes = min_t(unsigned long, PAGE_SIZE - offset,
iov_iter_count(i));
again:
if (bytes > length)
bytes = length;
/*
* Bring in the user page that we will copy from _first_.
* Otherwise there's a nasty deadlock on copying from the
* same page as we're writing to, without it being marked
* up-to-date.
*
* Not only is this an optimisation, but it is also required
* to check that the address is actually valid, when atomic
* usercopies are used, below.
*/
if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
status = -EFAULT;
break;
}
status = iomap_write_begin(inode, pos, bytes, flags, &page,
iomap);
if (unlikely(status))
break;
if (mapping_writably_mapped(inode->i_mapping))
flush_dcache_page(page);
copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
flush_dcache_page(page);
status = iomap_write_end(inode, pos, bytes, copied, page,
iomap);
if (unlikely(status < 0))
break;
copied = status;
cond_resched();
iov_iter_advance(i, copied);
if (unlikely(copied == 0)) {
/*
* If we were unable to copy any data at all, we must
* fall back to a single segment length write.
*
* If we didn't fallback here, we could livelock
* because not all segments in the iov can be copied at
* once without a pagefault.
*/
bytes = min_t(unsigned long, PAGE_SIZE - offset,
iov_iter_single_seg_count(i));
goto again;
}
pos += copied;
written += copied;
length -= copied;
balance_dirty_pages_ratelimited(inode->i_mapping);
} while (iov_iter_count(i) && length);
return written ? written : status;
}
ssize_t
iomap_file_buffered_write(struct kiocb *iocb, struct iov_iter *iter,
const struct iomap_ops *ops)
{
struct inode *inode = iocb->ki_filp->f_mapping->host;
loff_t pos = iocb->ki_pos, ret = 0, written = 0;
while (iov_iter_count(iter)) {
ret = iomap_apply(inode, pos, iov_iter_count(iter),
IOMAP_WRITE, ops, iter, iomap_write_actor);
if (ret <= 0)
break;
pos += ret;
written += ret;
}
return written ? written : ret;
}
EXPORT_SYMBOL_GPL(iomap_file_buffered_write);
static struct page *
__iomap_read_page(struct inode *inode, loff_t offset)
{
struct address_space *mapping = inode->i_mapping;
struct page *page;
page = read_mapping_page(mapping, offset >> PAGE_SHIFT, NULL);
if (IS_ERR(page))
return page;
if (!PageUptodate(page)) {
put_page(page);
return ERR_PTR(-EIO);
}
return page;
}
static loff_t
iomap_dirty_actor(struct inode *inode, loff_t pos, loff_t length, void *data,
struct iomap *iomap)
{
long status = 0;
ssize_t written = 0;
do {
struct page *page, *rpage;
unsigned long offset; /* Offset into pagecache page */
unsigned long bytes; /* Bytes to write to page */
offset = offset_in_page(pos);
bytes = min_t(loff_t, PAGE_SIZE - offset, length);
rpage = __iomap_read_page(inode, pos);
if (IS_ERR(rpage))
return PTR_ERR(rpage);
status = iomap_write_begin(inode, pos, bytes,
AOP_FLAG_NOFS, &page, iomap);
put_page(rpage);
if (unlikely(status))
return status;
WARN_ON_ONCE(!PageUptodate(page));
status = iomap_write_end(inode, pos, bytes, bytes, page, iomap);
if (unlikely(status <= 0)) {
if (WARN_ON_ONCE(status == 0))
return -EIO;
return status;
}
cond_resched();
pos += status;
written += status;
length -= status;
balance_dirty_pages_ratelimited(inode->i_mapping);
} while (length);
return written;
}
int
iomap_file_dirty(struct inode *inode, loff_t pos, loff_t len,
const struct iomap_ops *ops)
{
loff_t ret;
while (len) {
ret = iomap_apply(inode, pos, len, IOMAP_WRITE, ops, NULL,
iomap_dirty_actor);
if (ret <= 0)
return ret;
pos += ret;
len -= ret;
}
return 0;
}
EXPORT_SYMBOL_GPL(iomap_file_dirty);
static int iomap_zero(struct inode *inode, loff_t pos, unsigned offset,
unsigned bytes, struct iomap *iomap)
{
struct page *page;
int status;
status = iomap_write_begin(inode, pos, bytes, AOP_FLAG_NOFS, &page,
iomap);
if (status)
return status;
zero_user(page, offset, bytes);
mark_page_accessed(page);
return iomap_write_end(inode, pos, bytes, bytes, page, iomap);
}
static int iomap_dax_zero(loff_t pos, unsigned offset, unsigned bytes,
struct iomap *iomap)
{
return __dax_zero_page_range(iomap->bdev, iomap->dax_dev,
iomap_sector(iomap, pos & PAGE_MASK), offset, bytes);
}
static loff_t
iomap_zero_range_actor(struct inode *inode, loff_t pos, loff_t count,
void *data, struct iomap *iomap)
{
bool *did_zero = data;
loff_t written = 0;
int status;
/* already zeroed? we're done. */
if (iomap->type == IOMAP_HOLE || iomap->type == IOMAP_UNWRITTEN)
return count;
do {
unsigned offset, bytes;
offset = offset_in_page(pos);
bytes = min_t(loff_t, PAGE_SIZE - offset, count);
if (IS_DAX(inode))
status = iomap_dax_zero(pos, offset, bytes, iomap);
else
status = iomap_zero(inode, pos, offset, bytes, iomap);
if (status < 0)
return status;
pos += bytes;
count -= bytes;
written += bytes;
if (did_zero)
*did_zero = true;
} while (count > 0);
return written;
}
int
iomap_zero_range(struct inode *inode, loff_t pos, loff_t len, bool *did_zero,
const struct iomap_ops *ops)
{
loff_t ret;
while (len > 0) {
ret = iomap_apply(inode, pos, len, IOMAP_ZERO,
ops, did_zero, iomap_zero_range_actor);
if (ret <= 0)
return ret;
pos += ret;
len -= ret;
}
return 0;
}
EXPORT_SYMBOL_GPL(iomap_zero_range);
int
iomap_truncate_page(struct inode *inode, loff_t pos, bool *did_zero,
const struct iomap_ops *ops)
{
unsigned int blocksize = i_blocksize(inode);
unsigned int off = pos & (blocksize - 1);
/* Block boundary? Nothing to do */
if (!off)
return 0;
return iomap_zero_range(inode, pos, blocksize - off, did_zero, ops);
}
EXPORT_SYMBOL_GPL(iomap_truncate_page);
static loff_t
iomap_page_mkwrite_actor(struct inode *inode, loff_t pos, loff_t length,
void *data, struct iomap *iomap)
{
struct page *page = data;
int ret;
if (iomap->flags & IOMAP_F_BUFFER_HEAD) {
ret = __block_write_begin_int(page, pos, length, NULL, iomap);
if (ret)
return ret;
block_commit_write(page, 0, length);
} else {
WARN_ON_ONCE(!PageUptodate(page));
iomap_page_create(inode, page);
iomap: set page dirty after partial delalloc on mkwrite The iomap page fault mechanism currently dirties the associated page after the full block range of the page has been allocated. This leaves the page susceptible to delayed allocations without ever being set dirty on sub-page block sized filesystems. For example, consider a page fault on a page with one preexisting real (non-delalloc) block allocated in the middle of the page. The first iomap_apply() iteration performs delayed allocation on the range up to the preexisting block, the next iteration finds the preexisting block, and the last iteration attempts to perform delayed allocation on the range after the prexisting block to the end of the page. If the first allocation succeeds and the final allocation fails with -ENOSPC, iomap_apply() returns the error and iomap_page_mkwrite() fails to dirty the page having already performed partial delayed allocation. This eventually results in the page being invalidated without ever converting the delayed allocation to real blocks. This problem is reliably reproduced by generic/083 on XFS on ppc64 systems (64k page size, 4k block size). It results in leaked delalloc blocks on inode reclaim, which triggers an assert failure in xfs_fs_destroy_inode() and filesystem accounting inconsistency. Move the set_page_dirty() call from iomap_page_mkwrite() to the actor callback, similar to how the buffer head implementation works. The actor callback is called iff ->iomap_begin() returns success, so ensures the page is dirtied as soon as possible after an allocation. Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2018-09-29 11:51:01 +08:00
set_page_dirty(page);
}
return length;
}
vm_fault_t iomap_page_mkwrite(struct vm_fault *vmf, const struct iomap_ops *ops)
{
struct page *page = vmf->page;
struct inode *inode = file_inode(vmf->vma->vm_file);
unsigned long length;
loff_t offset, size;
ssize_t ret;
lock_page(page);
size = i_size_read(inode);
if ((page->mapping != inode->i_mapping) ||
(page_offset(page) > size)) {
/* We overload EFAULT to mean page got truncated */
ret = -EFAULT;
goto out_unlock;
}
/* page is wholly or partially inside EOF */
if (((page->index + 1) << PAGE_SHIFT) > size)
length = offset_in_page(size);
else
length = PAGE_SIZE;
offset = page_offset(page);
while (length > 0) {
ret = iomap_apply(inode, offset, length,
IOMAP_WRITE | IOMAP_FAULT, ops, page,
iomap_page_mkwrite_actor);
if (unlikely(ret <= 0))
goto out_unlock;
offset += ret;
length -= ret;
}
wait_for_stable_page(page);
return VM_FAULT_LOCKED;
out_unlock:
unlock_page(page);
return block_page_mkwrite_return(ret);
}
EXPORT_SYMBOL_GPL(iomap_page_mkwrite);
struct fiemap_ctx {
struct fiemap_extent_info *fi;
struct iomap prev;
};
static int iomap_to_fiemap(struct fiemap_extent_info *fi,
struct iomap *iomap, u32 flags)
{
switch (iomap->type) {
case IOMAP_HOLE:
/* skip holes */
return 0;
case IOMAP_DELALLOC:
flags |= FIEMAP_EXTENT_DELALLOC | FIEMAP_EXTENT_UNKNOWN;
break;
case IOMAP_MAPPED:
break;
case IOMAP_UNWRITTEN:
flags |= FIEMAP_EXTENT_UNWRITTEN;
break;
case IOMAP_INLINE:
flags |= FIEMAP_EXTENT_DATA_INLINE;
break;
}
if (iomap->flags & IOMAP_F_MERGED)
flags |= FIEMAP_EXTENT_MERGED;
if (iomap->flags & IOMAP_F_SHARED)
flags |= FIEMAP_EXTENT_SHARED;
return fiemap_fill_next_extent(fi, iomap->offset,
iomap->addr != IOMAP_NULL_ADDR ? iomap->addr : 0,
iomap->length, flags);
}
static loff_t
iomap_fiemap_actor(struct inode *inode, loff_t pos, loff_t length, void *data,
struct iomap *iomap)
{
struct fiemap_ctx *ctx = data;
loff_t ret = length;
if (iomap->type == IOMAP_HOLE)
return length;
ret = iomap_to_fiemap(ctx->fi, &ctx->prev, 0);
ctx->prev = *iomap;
switch (ret) {
case 0: /* success */
return length;
case 1: /* extent array full */
return 0;
default:
return ret;
}
}
int iomap_fiemap(struct inode *inode, struct fiemap_extent_info *fi,
loff_t start, loff_t len, const struct iomap_ops *ops)
{
struct fiemap_ctx ctx;
loff_t ret;
memset(&ctx, 0, sizeof(ctx));
ctx.fi = fi;
ctx.prev.type = IOMAP_HOLE;
ret = fiemap_check_flags(fi, FIEMAP_FLAG_SYNC);
if (ret)
return ret;
if (fi->fi_flags & FIEMAP_FLAG_SYNC) {
ret = filemap_write_and_wait(inode->i_mapping);
if (ret)
return ret;
}
while (len > 0) {
ret = iomap_apply(inode, start, len, IOMAP_REPORT, ops, &ctx,
iomap_fiemap_actor);
/* inode with no (attribute) mapping will give ENOENT */
if (ret == -ENOENT)
break;
if (ret < 0)
return ret;
if (ret == 0)
break;
start += ret;
len -= ret;
}
if (ctx.prev.type != IOMAP_HOLE) {
ret = iomap_to_fiemap(fi, &ctx.prev, FIEMAP_EXTENT_LAST);
if (ret < 0)
return ret;
}
return 0;
}
EXPORT_SYMBOL_GPL(iomap_fiemap);
/*
* Seek for SEEK_DATA / SEEK_HOLE within @page, starting at @lastoff.
* Returns true if found and updates @lastoff to the offset in file.
*/
static bool
page_seek_hole_data(struct inode *inode, struct page *page, loff_t *lastoff,
int whence)
{
const struct address_space_operations *ops = inode->i_mapping->a_ops;
unsigned int bsize = i_blocksize(inode), off;
bool seek_data = whence == SEEK_DATA;
loff_t poff = page_offset(page);
if (WARN_ON_ONCE(*lastoff >= poff + PAGE_SIZE))
return false;
if (*lastoff < poff) {
/*
* Last offset smaller than the start of the page means we found
* a hole:
*/
if (whence == SEEK_HOLE)
return true;
*lastoff = poff;
}
/*
* Just check the page unless we can and should check block ranges:
*/
if (bsize == PAGE_SIZE || !ops->is_partially_uptodate)
return PageUptodate(page) == seek_data;
lock_page(page);
if (unlikely(page->mapping != inode->i_mapping))
goto out_unlock_not_found;
for (off = 0; off < PAGE_SIZE; off += bsize) {
if (offset_in_page(*lastoff) >= off + bsize)
continue;
if (ops->is_partially_uptodate(page, off, bsize) == seek_data) {
unlock_page(page);
return true;
}
*lastoff = poff + off + bsize;
}
out_unlock_not_found:
unlock_page(page);
return false;
}
/*
* Seek for SEEK_DATA / SEEK_HOLE in the page cache.
*
* Within unwritten extents, the page cache determines which parts are holes
* and which are data: uptodate buffer heads count as data; everything else
* counts as a hole.
*
* Returns the resulting offset on successs, and -ENOENT otherwise.
*/
static loff_t
page_cache_seek_hole_data(struct inode *inode, loff_t offset, loff_t length,
int whence)
{
pgoff_t index = offset >> PAGE_SHIFT;
pgoff_t end = DIV_ROUND_UP(offset + length, PAGE_SIZE);
loff_t lastoff = offset;
struct pagevec pvec;
if (length <= 0)
return -ENOENT;
pagevec_init(&pvec);
do {
unsigned nr_pages, i;
nr_pages = pagevec_lookup_range(&pvec, inode->i_mapping, &index,
end - 1);
if (nr_pages == 0)
break;
for (i = 0; i < nr_pages; i++) {
struct page *page = pvec.pages[i];
if (page_seek_hole_data(inode, page, &lastoff, whence))
goto check_range;
lastoff = page_offset(page) + PAGE_SIZE;
}
pagevec_release(&pvec);
} while (index < end);
/* When no page at lastoff and we are not done, we found a hole. */
if (whence != SEEK_HOLE)
goto not_found;
check_range:
if (lastoff < offset + length)
goto out;
not_found:
lastoff = -ENOENT;
out:
pagevec_release(&pvec);
return lastoff;
}
static loff_t
iomap_seek_hole_actor(struct inode *inode, loff_t offset, loff_t length,
void *data, struct iomap *iomap)
{
switch (iomap->type) {
case IOMAP_UNWRITTEN:
offset = page_cache_seek_hole_data(inode, offset, length,
SEEK_HOLE);
if (offset < 0)
return length;
/* fall through */
case IOMAP_HOLE:
*(loff_t *)data = offset;
return 0;
default:
return length;
}
}
loff_t
iomap_seek_hole(struct inode *inode, loff_t offset, const struct iomap_ops *ops)
{
loff_t size = i_size_read(inode);
loff_t length = size - offset;
loff_t ret;
/* Nothing to be found before or beyond the end of the file. */
if (offset < 0 || offset >= size)
return -ENXIO;
while (length > 0) {
ret = iomap_apply(inode, offset, length, IOMAP_REPORT, ops,
&offset, iomap_seek_hole_actor);
if (ret < 0)
return ret;
if (ret == 0)
break;
offset += ret;
length -= ret;
}
return offset;
}
EXPORT_SYMBOL_GPL(iomap_seek_hole);
static loff_t
iomap_seek_data_actor(struct inode *inode, loff_t offset, loff_t length,
void *data, struct iomap *iomap)
{
switch (iomap->type) {
case IOMAP_HOLE:
return length;
case IOMAP_UNWRITTEN:
offset = page_cache_seek_hole_data(inode, offset, length,
SEEK_DATA);
if (offset < 0)
return length;
/*FALLTHRU*/
default:
*(loff_t *)data = offset;
return 0;
}
}
loff_t
iomap_seek_data(struct inode *inode, loff_t offset, const struct iomap_ops *ops)
{
loff_t size = i_size_read(inode);
loff_t length = size - offset;
loff_t ret;
/* Nothing to be found before or beyond the end of the file. */
if (offset < 0 || offset >= size)
return -ENXIO;
while (length > 0) {
ret = iomap_apply(inode, offset, length, IOMAP_REPORT, ops,
&offset, iomap_seek_data_actor);
if (ret < 0)
return ret;
if (ret == 0)
break;
offset += ret;
length -= ret;
}
if (length <= 0)
return -ENXIO;
return offset;
}
EXPORT_SYMBOL_GPL(iomap_seek_data);
/*
* Private flags for iomap_dio, must not overlap with the public ones in
* iomap.h:
*/
iomap: Use FUA for pure data O_DSYNC DIO writes If we are doing direct IO writes with datasync semantics, we often have to flush metadata changes along with the data write. However, if we are overwriting existing data, there are no metadata changes that we need to flush. In this case, optimising the IO by using FUA write makes sense. We know from the IOMAP_F_DIRTY flag as to whether a specific inode requires a metadata flush - this is currently used by DAX to ensure extent modification as stable in page fault operations. For direct IO writes, we can use it to determine if we need to flush metadata or not once the data is on disk. Hence if we have been returned a mapped extent that is not new and the IO mapping is not dirty, then we can use a FUA write to provide datasync semantics. This allows us to short-cut the generic_write_sync() call in IO completion and hence avoid unnecessary operations. This makes pure direct IO data write behaviour identical to the way block devices use REQ_FUA to provide datasync semantics. On a FUA enabled device, a synchronous direct IO write workload (sequential 4k overwrites in 32MB file) had the following results: # xfs_io -fd -c "pwrite -V 1 -D 0 32m" /mnt/scratch/boo kernel time write()s write iops Write b/w ------ ---- -------- ---------- --------- (no dsync) 4s 2173/s 2173 8.5MB/s vanilla 22s 370/s 750 1.4MB/s patched 19s 420/s 420 1.6MB/s The patched code clearly doesn't send cache flushes anymore, but instead uses FUA (confirmed via blktrace), and performance improves a bit as a result. However, the benefits will be higher on workloads that mix O_DSYNC overwrites with other write IO as we won't be flushing the entire device cache on every DSYNC overwrite IO anymore. Signed-Off-By: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2018-05-03 03:54:53 +08:00
#define IOMAP_DIO_WRITE_FUA (1 << 28)
#define IOMAP_DIO_NEED_SYNC (1 << 29)
#define IOMAP_DIO_WRITE (1 << 30)
#define IOMAP_DIO_DIRTY (1 << 31)
struct iomap_dio {
struct kiocb *iocb;
iomap_dio_end_io_t *end_io;
loff_t i_size;
loff_t size;
atomic_t ref;
unsigned flags;
int error;
bool wait_for_completion;
union {
/* used during submission and for synchronous completion: */
struct {
struct iov_iter *iter;
struct task_struct *waiter;
struct request_queue *last_queue;
blk_qc_t cookie;
} submit;
/* used for aio completion: */
struct {
struct work_struct work;
} aio;
};
};
int iomap_dio_iopoll(struct kiocb *kiocb, bool spin)
{
struct request_queue *q = READ_ONCE(kiocb->private);
if (!q)
return 0;
return blk_poll(q, READ_ONCE(kiocb->ki_cookie), spin);
}
EXPORT_SYMBOL_GPL(iomap_dio_iopoll);
static void iomap_dio_submit_bio(struct iomap_dio *dio, struct iomap *iomap,
struct bio *bio)
{
atomic_inc(&dio->ref);
if (dio->iocb->ki_flags & IOCB_HIPRI)
bio_set_polled(bio, dio->iocb);
dio->submit.last_queue = bdev_get_queue(iomap->bdev);
dio->submit.cookie = submit_bio(bio);
}
static ssize_t iomap_dio_complete(struct iomap_dio *dio)
{
struct kiocb *iocb = dio->iocb;
struct inode *inode = file_inode(iocb->ki_filp);
fs: invalidate page cache after end_io() in dio completion Commit 332391a9935d ("fs: Fix page cache inconsistency when mixing buffered and AIO DIO") moved page cache invalidation from iomap_dio_rw() to iomap_dio_complete() for iomap based direct write path, but before the dio->end_io() call, and it re-introdued the bug fixed by commit c771c14baa33 ("iomap: invalidate page caches should be after iomap_dio_complete() in direct write"). I found this because fstests generic/418 started failing on XFS with v4.14-rc3 kernel, which is the regression test for this specific bug. So similarly, fix it by moving dio->end_io() (which does the unwritten extent conversion) before page cache invalidation, to make sure next buffer read reads the final real allocations not unwritten extents. I also add some comments about why should end_io() go first in case we get it wrong again in the future. Note that, there's no such problem in the non-iomap based direct write path, because we didn't remove the page cache invalidation after the ->direct_IO() in generic_file_direct_write() call, but I decided to fix dio_complete() too so we don't leave a landmine there, also be consistent with iomap_dio_complete(). Fixes: 332391a9935d ("fs: Fix page cache inconsistency when mixing buffered and AIO DIO") Signed-off-by: Eryu Guan <eguan@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Jan Kara <jack@suse.cz> Reviewed-by: Lukas Czerner <lczerner@redhat.com>
2017-10-14 00:47:46 +08:00
loff_t offset = iocb->ki_pos;
ssize_t ret;
if (dio->end_io) {
ret = dio->end_io(iocb,
dio->error ? dio->error : dio->size,
dio->flags);
} else {
ret = dio->error;
}
if (likely(!ret)) {
ret = dio->size;
/* check for short read */
fs: invalidate page cache after end_io() in dio completion Commit 332391a9935d ("fs: Fix page cache inconsistency when mixing buffered and AIO DIO") moved page cache invalidation from iomap_dio_rw() to iomap_dio_complete() for iomap based direct write path, but before the dio->end_io() call, and it re-introdued the bug fixed by commit c771c14baa33 ("iomap: invalidate page caches should be after iomap_dio_complete() in direct write"). I found this because fstests generic/418 started failing on XFS with v4.14-rc3 kernel, which is the regression test for this specific bug. So similarly, fix it by moving dio->end_io() (which does the unwritten extent conversion) before page cache invalidation, to make sure next buffer read reads the final real allocations not unwritten extents. I also add some comments about why should end_io() go first in case we get it wrong again in the future. Note that, there's no such problem in the non-iomap based direct write path, because we didn't remove the page cache invalidation after the ->direct_IO() in generic_file_direct_write() call, but I decided to fix dio_complete() too so we don't leave a landmine there, also be consistent with iomap_dio_complete(). Fixes: 332391a9935d ("fs: Fix page cache inconsistency when mixing buffered and AIO DIO") Signed-off-by: Eryu Guan <eguan@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Jan Kara <jack@suse.cz> Reviewed-by: Lukas Czerner <lczerner@redhat.com>
2017-10-14 00:47:46 +08:00
if (offset + ret > dio->i_size &&
!(dio->flags & IOMAP_DIO_WRITE))
fs: invalidate page cache after end_io() in dio completion Commit 332391a9935d ("fs: Fix page cache inconsistency when mixing buffered and AIO DIO") moved page cache invalidation from iomap_dio_rw() to iomap_dio_complete() for iomap based direct write path, but before the dio->end_io() call, and it re-introdued the bug fixed by commit c771c14baa33 ("iomap: invalidate page caches should be after iomap_dio_complete() in direct write"). I found this because fstests generic/418 started failing on XFS with v4.14-rc3 kernel, which is the regression test for this specific bug. So similarly, fix it by moving dio->end_io() (which does the unwritten extent conversion) before page cache invalidation, to make sure next buffer read reads the final real allocations not unwritten extents. I also add some comments about why should end_io() go first in case we get it wrong again in the future. Note that, there's no such problem in the non-iomap based direct write path, because we didn't remove the page cache invalidation after the ->direct_IO() in generic_file_direct_write() call, but I decided to fix dio_complete() too so we don't leave a landmine there, also be consistent with iomap_dio_complete(). Fixes: 332391a9935d ("fs: Fix page cache inconsistency when mixing buffered and AIO DIO") Signed-off-by: Eryu Guan <eguan@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Jan Kara <jack@suse.cz> Reviewed-by: Lukas Czerner <lczerner@redhat.com>
2017-10-14 00:47:46 +08:00
ret = dio->i_size - offset;
iocb->ki_pos += ret;
}
fs: invalidate page cache after end_io() in dio completion Commit 332391a9935d ("fs: Fix page cache inconsistency when mixing buffered and AIO DIO") moved page cache invalidation from iomap_dio_rw() to iomap_dio_complete() for iomap based direct write path, but before the dio->end_io() call, and it re-introdued the bug fixed by commit c771c14baa33 ("iomap: invalidate page caches should be after iomap_dio_complete() in direct write"). I found this because fstests generic/418 started failing on XFS with v4.14-rc3 kernel, which is the regression test for this specific bug. So similarly, fix it by moving dio->end_io() (which does the unwritten extent conversion) before page cache invalidation, to make sure next buffer read reads the final real allocations not unwritten extents. I also add some comments about why should end_io() go first in case we get it wrong again in the future. Note that, there's no such problem in the non-iomap based direct write path, because we didn't remove the page cache invalidation after the ->direct_IO() in generic_file_direct_write() call, but I decided to fix dio_complete() too so we don't leave a landmine there, also be consistent with iomap_dio_complete(). Fixes: 332391a9935d ("fs: Fix page cache inconsistency when mixing buffered and AIO DIO") Signed-off-by: Eryu Guan <eguan@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Jan Kara <jack@suse.cz> Reviewed-by: Lukas Czerner <lczerner@redhat.com>
2017-10-14 00:47:46 +08:00
/*
* Try again to invalidate clean pages which might have been cached by
* non-direct readahead, or faulted in by get_user_pages() if the source
* of the write was an mmap'ed region of the file we're writing. Either
* one is a pretty crazy thing to do, so we don't support it 100%. If
* this invalidation fails, tough, the write still worked...
*
* And this page cache invalidation has to be after dio->end_io(), as
* some filesystems convert unwritten extents to real allocations in
* end_io() when necessary, otherwise a racing buffer read would cache
* zeros from unwritten extents.
*/
if (!dio->error &&
(dio->flags & IOMAP_DIO_WRITE) && inode->i_mapping->nrpages) {
int err;
err = invalidate_inode_pages2_range(inode->i_mapping,
offset >> PAGE_SHIFT,
(offset + dio->size - 1) >> PAGE_SHIFT);
if (err)
dio_warn_stale_pagecache(iocb->ki_filp);
fs: invalidate page cache after end_io() in dio completion Commit 332391a9935d ("fs: Fix page cache inconsistency when mixing buffered and AIO DIO") moved page cache invalidation from iomap_dio_rw() to iomap_dio_complete() for iomap based direct write path, but before the dio->end_io() call, and it re-introdued the bug fixed by commit c771c14baa33 ("iomap: invalidate page caches should be after iomap_dio_complete() in direct write"). I found this because fstests generic/418 started failing on XFS with v4.14-rc3 kernel, which is the regression test for this specific bug. So similarly, fix it by moving dio->end_io() (which does the unwritten extent conversion) before page cache invalidation, to make sure next buffer read reads the final real allocations not unwritten extents. I also add some comments about why should end_io() go first in case we get it wrong again in the future. Note that, there's no such problem in the non-iomap based direct write path, because we didn't remove the page cache invalidation after the ->direct_IO() in generic_file_direct_write() call, but I decided to fix dio_complete() too so we don't leave a landmine there, also be consistent with iomap_dio_complete(). Fixes: 332391a9935d ("fs: Fix page cache inconsistency when mixing buffered and AIO DIO") Signed-off-by: Eryu Guan <eguan@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Jan Kara <jack@suse.cz> Reviewed-by: Lukas Czerner <lczerner@redhat.com>
2017-10-14 00:47:46 +08:00
}
/*
* If this is a DSYNC write, make sure we push it to stable storage now
* that we've written data.
*/
if (ret > 0 && (dio->flags & IOMAP_DIO_NEED_SYNC))
ret = generic_write_sync(iocb, ret);
inode_dio_end(file_inode(iocb->ki_filp));
kfree(dio);
return ret;
}
static void iomap_dio_complete_work(struct work_struct *work)
{
struct iomap_dio *dio = container_of(work, struct iomap_dio, aio.work);
struct kiocb *iocb = dio->iocb;
iocb->ki_complete(iocb, iomap_dio_complete(dio), 0);
}
/*
* Set an error in the dio if none is set yet. We have to use cmpxchg
* as the submission context and the completion context(s) can race to
* update the error.
*/
static inline void iomap_dio_set_error(struct iomap_dio *dio, int ret)
{
cmpxchg(&dio->error, 0, ret);
}
static void iomap_dio_bio_end_io(struct bio *bio)
{
struct iomap_dio *dio = bio->bi_private;
bool should_dirty = (dio->flags & IOMAP_DIO_DIRTY);
if (bio->bi_status)
iomap_dio_set_error(dio, blk_status_to_errno(bio->bi_status));
if (atomic_dec_and_test(&dio->ref)) {
if (dio->wait_for_completion) {
struct task_struct *waiter = dio->submit.waiter;
WRITE_ONCE(dio->submit.waiter, NULL);
blk_wake_io_task(waiter);
} else if (dio->flags & IOMAP_DIO_WRITE) {
struct inode *inode = file_inode(dio->iocb->ki_filp);
INIT_WORK(&dio->aio.work, iomap_dio_complete_work);
queue_work(inode->i_sb->s_dio_done_wq, &dio->aio.work);
} else {
iomap_dio_complete_work(&dio->aio.work);
}
}
if (should_dirty) {
bio_check_pages_dirty(bio);
} else {
if (!bio_flagged(bio, BIO_NO_PAGE_REF)) {
struct bvec_iter_all iter_all;
struct bio_vec *bvec;
bio_for_each_segment_all(bvec, bio, iter_all)
put_page(bvec->bv_page);
}
bio_put(bio);
}
}
static void
iomap_dio_zero(struct iomap_dio *dio, struct iomap *iomap, loff_t pos,
unsigned len)
{
struct page *page = ZERO_PAGE(0);
int flags = REQ_SYNC | REQ_IDLE;
struct bio *bio;
bio = bio_alloc(GFP_KERNEL, 1);
bio_set_dev(bio, iomap->bdev);
bio->bi_iter.bi_sector = iomap_sector(iomap, pos);
bio->bi_private = dio;
bio->bi_end_io = iomap_dio_bio_end_io;
get_page(page);
__bio_add_page(bio, page, len, 0);
bio_set_op_attrs(bio, REQ_OP_WRITE, flags);
iomap_dio_submit_bio(dio, iomap, bio);
}
static loff_t
iomap_dio_bio_actor(struct inode *inode, loff_t pos, loff_t length,
struct iomap_dio *dio, struct iomap *iomap)
{
unsigned int blkbits = blksize_bits(bdev_logical_block_size(iomap->bdev));
unsigned int fs_block_size = i_blocksize(inode), pad;
unsigned int align = iov_iter_alignment(dio->submit.iter);
struct iov_iter iter;
struct bio *bio;
bool need_zeroout = false;
iomap: Use FUA for pure data O_DSYNC DIO writes If we are doing direct IO writes with datasync semantics, we often have to flush metadata changes along with the data write. However, if we are overwriting existing data, there are no metadata changes that we need to flush. In this case, optimising the IO by using FUA write makes sense. We know from the IOMAP_F_DIRTY flag as to whether a specific inode requires a metadata flush - this is currently used by DAX to ensure extent modification as stable in page fault operations. For direct IO writes, we can use it to determine if we need to flush metadata or not once the data is on disk. Hence if we have been returned a mapped extent that is not new and the IO mapping is not dirty, then we can use a FUA write to provide datasync semantics. This allows us to short-cut the generic_write_sync() call in IO completion and hence avoid unnecessary operations. This makes pure direct IO data write behaviour identical to the way block devices use REQ_FUA to provide datasync semantics. On a FUA enabled device, a synchronous direct IO write workload (sequential 4k overwrites in 32MB file) had the following results: # xfs_io -fd -c "pwrite -V 1 -D 0 32m" /mnt/scratch/boo kernel time write()s write iops Write b/w ------ ---- -------- ---------- --------- (no dsync) 4s 2173/s 2173 8.5MB/s vanilla 22s 370/s 750 1.4MB/s patched 19s 420/s 420 1.6MB/s The patched code clearly doesn't send cache flushes anymore, but instead uses FUA (confirmed via blktrace), and performance improves a bit as a result. However, the benefits will be higher on workloads that mix O_DSYNC overwrites with other write IO as we won't be flushing the entire device cache on every DSYNC overwrite IO anymore. Signed-Off-By: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2018-05-03 03:54:53 +08:00
bool use_fua = false;
int nr_pages, ret = 0;
size_t copied = 0;
if ((pos | length | align) & ((1 << blkbits) - 1))
return -EINVAL;
if (iomap->type == IOMAP_UNWRITTEN) {
dio->flags |= IOMAP_DIO_UNWRITTEN;
need_zeroout = true;
}
if (iomap->flags & IOMAP_F_SHARED)
dio->flags |= IOMAP_DIO_COW;
if (iomap->flags & IOMAP_F_NEW) {
need_zeroout = true;
} else if (iomap->type == IOMAP_MAPPED) {
/*
* Use a FUA write if we need datasync semantics, this is a pure
* data IO that doesn't require any metadata updates (including
* after IO completion such as unwritten extent conversion) and
* the underlying device supports FUA. This allows us to avoid
* cache flushes on IO completion.
*/
if (!(iomap->flags & (IOMAP_F_SHARED|IOMAP_F_DIRTY)) &&
(dio->flags & IOMAP_DIO_WRITE_FUA) &&
blk_queue_fua(bdev_get_queue(iomap->bdev)))
use_fua = true;
}
/*
* Operate on a partial iter trimmed to the extent we were called for.
* We'll update the iter in the dio once we're done with this extent.
*/
iter = *dio->submit.iter;
iov_iter_truncate(&iter, length);
nr_pages = iov_iter_npages(&iter, BIO_MAX_PAGES);
if (nr_pages <= 0)
return nr_pages;
if (need_zeroout) {
/* zero out from the start of the block to the write offset */
pad = pos & (fs_block_size - 1);
if (pad)
iomap_dio_zero(dio, iomap, pos - pad, pad);
}
do {
size_t n;
if (dio->error) {
iov_iter_revert(dio->submit.iter, copied);
return 0;
}
bio = bio_alloc(GFP_KERNEL, nr_pages);
bio_set_dev(bio, iomap->bdev);
bio->bi_iter.bi_sector = iomap_sector(iomap, pos);
bio->bi_write_hint = dio->iocb->ki_hint;
bio->bi_ioprio = dio->iocb->ki_ioprio;
bio->bi_private = dio;
bio->bi_end_io = iomap_dio_bio_end_io;
ret = bio_iov_iter_get_pages(bio, &iter);
if (unlikely(ret)) {
/*
* We have to stop part way through an IO. We must fall
* through to the sub-block tail zeroing here, otherwise
* this short IO may expose stale data in the tail of
* the block we haven't written data to.
*/
bio_put(bio);
goto zero_tail;
}
n = bio->bi_iter.bi_size;
if (dio->flags & IOMAP_DIO_WRITE) {
iomap: Use FUA for pure data O_DSYNC DIO writes If we are doing direct IO writes with datasync semantics, we often have to flush metadata changes along with the data write. However, if we are overwriting existing data, there are no metadata changes that we need to flush. In this case, optimising the IO by using FUA write makes sense. We know from the IOMAP_F_DIRTY flag as to whether a specific inode requires a metadata flush - this is currently used by DAX to ensure extent modification as stable in page fault operations. For direct IO writes, we can use it to determine if we need to flush metadata or not once the data is on disk. Hence if we have been returned a mapped extent that is not new and the IO mapping is not dirty, then we can use a FUA write to provide datasync semantics. This allows us to short-cut the generic_write_sync() call in IO completion and hence avoid unnecessary operations. This makes pure direct IO data write behaviour identical to the way block devices use REQ_FUA to provide datasync semantics. On a FUA enabled device, a synchronous direct IO write workload (sequential 4k overwrites in 32MB file) had the following results: # xfs_io -fd -c "pwrite -V 1 -D 0 32m" /mnt/scratch/boo kernel time write()s write iops Write b/w ------ ---- -------- ---------- --------- (no dsync) 4s 2173/s 2173 8.5MB/s vanilla 22s 370/s 750 1.4MB/s patched 19s 420/s 420 1.6MB/s The patched code clearly doesn't send cache flushes anymore, but instead uses FUA (confirmed via blktrace), and performance improves a bit as a result. However, the benefits will be higher on workloads that mix O_DSYNC overwrites with other write IO as we won't be flushing the entire device cache on every DSYNC overwrite IO anymore. Signed-Off-By: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2018-05-03 03:54:53 +08:00
bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_IDLE;
if (use_fua)
bio->bi_opf |= REQ_FUA;
else
dio->flags &= ~IOMAP_DIO_WRITE_FUA;
task_io_account_write(n);
} else {
iomap: Use FUA for pure data O_DSYNC DIO writes If we are doing direct IO writes with datasync semantics, we often have to flush metadata changes along with the data write. However, if we are overwriting existing data, there are no metadata changes that we need to flush. In this case, optimising the IO by using FUA write makes sense. We know from the IOMAP_F_DIRTY flag as to whether a specific inode requires a metadata flush - this is currently used by DAX to ensure extent modification as stable in page fault operations. For direct IO writes, we can use it to determine if we need to flush metadata or not once the data is on disk. Hence if we have been returned a mapped extent that is not new and the IO mapping is not dirty, then we can use a FUA write to provide datasync semantics. This allows us to short-cut the generic_write_sync() call in IO completion and hence avoid unnecessary operations. This makes pure direct IO data write behaviour identical to the way block devices use REQ_FUA to provide datasync semantics. On a FUA enabled device, a synchronous direct IO write workload (sequential 4k overwrites in 32MB file) had the following results: # xfs_io -fd -c "pwrite -V 1 -D 0 32m" /mnt/scratch/boo kernel time write()s write iops Write b/w ------ ---- -------- ---------- --------- (no dsync) 4s 2173/s 2173 8.5MB/s vanilla 22s 370/s 750 1.4MB/s patched 19s 420/s 420 1.6MB/s The patched code clearly doesn't send cache flushes anymore, but instead uses FUA (confirmed via blktrace), and performance improves a bit as a result. However, the benefits will be higher on workloads that mix O_DSYNC overwrites with other write IO as we won't be flushing the entire device cache on every DSYNC overwrite IO anymore. Signed-Off-By: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2018-05-03 03:54:53 +08:00
bio->bi_opf = REQ_OP_READ;
if (dio->flags & IOMAP_DIO_DIRTY)
bio_set_pages_dirty(bio);
}
iov_iter_advance(dio->submit.iter, n);
dio->size += n;
pos += n;
copied += n;
nr_pages = iov_iter_npages(&iter, BIO_MAX_PAGES);
iomap_dio_submit_bio(dio, iomap, bio);
} while (nr_pages);
/*
* We need to zeroout the tail of a sub-block write if the extent type
* requires zeroing or the write extends beyond EOF. If we don't zero
* the block tail in the latter case, we can expose stale data via mmap
* reads of the EOF block.
*/
zero_tail:
if (need_zeroout ||
((dio->flags & IOMAP_DIO_WRITE) && pos >= i_size_read(inode))) {
/* zero out from the end of the write to the end of the block */
pad = pos & (fs_block_size - 1);
if (pad)
iomap_dio_zero(dio, iomap, pos, fs_block_size - pad);
}
return copied ? copied : ret;
}
static loff_t
iomap_dio_hole_actor(loff_t length, struct iomap_dio *dio)
{
length = iov_iter_zero(length, dio->submit.iter);
dio->size += length;
return length;
}
static loff_t
iomap_dio_inline_actor(struct inode *inode, loff_t pos, loff_t length,
struct iomap_dio *dio, struct iomap *iomap)
{
struct iov_iter *iter = dio->submit.iter;
size_t copied;
BUG_ON(pos + length > PAGE_SIZE - offset_in_page(iomap->inline_data));
if (dio->flags & IOMAP_DIO_WRITE) {
loff_t size = inode->i_size;
if (pos > size)
memset(iomap->inline_data + size, 0, pos - size);
copied = copy_from_iter(iomap->inline_data + pos, length, iter);
if (copied) {
if (pos + copied > size)
i_size_write(inode, pos + copied);
mark_inode_dirty(inode);
}
} else {
copied = copy_to_iter(iomap->inline_data + pos, length, iter);
}
dio->size += copied;
return copied;
}
static loff_t
iomap_dio_actor(struct inode *inode, loff_t pos, loff_t length,
void *data, struct iomap *iomap)
{
struct iomap_dio *dio = data;
switch (iomap->type) {
case IOMAP_HOLE:
if (WARN_ON_ONCE(dio->flags & IOMAP_DIO_WRITE))
return -EIO;
return iomap_dio_hole_actor(length, dio);
case IOMAP_UNWRITTEN:
if (!(dio->flags & IOMAP_DIO_WRITE))
return iomap_dio_hole_actor(length, dio);
return iomap_dio_bio_actor(inode, pos, length, dio, iomap);
case IOMAP_MAPPED:
return iomap_dio_bio_actor(inode, pos, length, dio, iomap);
case IOMAP_INLINE:
return iomap_dio_inline_actor(inode, pos, length, dio, iomap);
default:
WARN_ON_ONCE(1);
return -EIO;
}
}
/*
* iomap_dio_rw() always completes O_[D]SYNC writes regardless of whether the IO
iomap: Use FUA for pure data O_DSYNC DIO writes If we are doing direct IO writes with datasync semantics, we often have to flush metadata changes along with the data write. However, if we are overwriting existing data, there are no metadata changes that we need to flush. In this case, optimising the IO by using FUA write makes sense. We know from the IOMAP_F_DIRTY flag as to whether a specific inode requires a metadata flush - this is currently used by DAX to ensure extent modification as stable in page fault operations. For direct IO writes, we can use it to determine if we need to flush metadata or not once the data is on disk. Hence if we have been returned a mapped extent that is not new and the IO mapping is not dirty, then we can use a FUA write to provide datasync semantics. This allows us to short-cut the generic_write_sync() call in IO completion and hence avoid unnecessary operations. This makes pure direct IO data write behaviour identical to the way block devices use REQ_FUA to provide datasync semantics. On a FUA enabled device, a synchronous direct IO write workload (sequential 4k overwrites in 32MB file) had the following results: # xfs_io -fd -c "pwrite -V 1 -D 0 32m" /mnt/scratch/boo kernel time write()s write iops Write b/w ------ ---- -------- ---------- --------- (no dsync) 4s 2173/s 2173 8.5MB/s vanilla 22s 370/s 750 1.4MB/s patched 19s 420/s 420 1.6MB/s The patched code clearly doesn't send cache flushes anymore, but instead uses FUA (confirmed via blktrace), and performance improves a bit as a result. However, the benefits will be higher on workloads that mix O_DSYNC overwrites with other write IO as we won't be flushing the entire device cache on every DSYNC overwrite IO anymore. Signed-Off-By: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2018-05-03 03:54:53 +08:00
* is being issued as AIO or not. This allows us to optimise pure data writes
* to use REQ_FUA rather than requiring generic_write_sync() to issue a
* REQ_FLUSH post write. This is slightly tricky because a single request here
* can be mapped into multiple disjoint IOs and only a subset of the IOs issued
* may be pure data writes. In that case, we still need to do a full data sync
* completion.
*/
ssize_t
iomap_dio_rw(struct kiocb *iocb, struct iov_iter *iter,
const struct iomap_ops *ops, iomap_dio_end_io_t end_io)
{
struct address_space *mapping = iocb->ki_filp->f_mapping;
struct inode *inode = file_inode(iocb->ki_filp);
size_t count = iov_iter_count(iter);
iomap: invalidate page caches should be after iomap_dio_complete() in direct write After XFS switching to iomap based DIO (commit acdda3aae146 ("xfs: use iomap_dio_rw")), I started to notice dio29/dio30 tests failures from LTP run on ppc64 hosts, and they can be reproduced on x86_64 hosts with 512B/1k block size XFS too. dio29 diotest3 -b 65536 -n 100 -i 1000 -o 1024000 dio30 diotest6 -b 65536 -n 100 -i 1000 -o 1024000 The failure message is like: bufcmp: offset 0: Expected: 0x62, got 0x0 diotest03 1 TPASS : Read with Direct IO, Write without diotest03 2 TFAIL : diotest3.c:142: comparsion failed; child=98 offset=1425408 diotest03 3 TFAIL : diotest3.c:194: Write Direct-child 98 failed Direct write wrote 0x62 but buffer read got zero. This is because, when doing direct write to a hole or preallocated file, we invalidate the page caches before converting the extent from unwritten state to normal state, which is done by iomap_dio_complete(), thus leave a window for other buffer reader to cache the unwritten state extent. Consider this case, with sub-page blocksize XFS, two processes are direct writing to different blocksize-aligned regions (say 512B) of the same preallocated file, and reading the region back via buffered I/O to compare contents. process A, region [0,512] process B, region [512,1024] xfs_file_write_iter xfs_file_aio_dio_write iomap_dio_rw iomap_apply invalidate_inode_pages2_range xfs_file_write_iter xfs_file_aio_dio_write iomap_dio_rw iomap_apply invalidate_inode_pages2_range iomap_dio_complete xfs_file_read_iter xfs_file_buffered_aio_read generic_file_read_iter do_generic_file_read <readahead fills pagecache with 0> iomap_dio_complete xfs_file_read_iter <read gets 0 from pagecache> Process A first invalidates page caches, at this point the underlying extent is still in unwritten state (iomap_dio_complete not called yet), and process B finishs direct write and populates page caches via readahead, which caches zeros in page for region A, then process A reads zeros from page cache, instead of the actual data. Fix it by invalidating page caches after converting unwritten extent to make sure we read content from disk after extent state changed, as what we did before switching to iomap based dio. Also introduce a new 'start' variable to save the original write offset (iomap_dio_complete() updates iocb->ki_pos), and a 'err' variable for invalidating caches result, cause we can't reuse 'ret' anymore. Signed-off-by: Eryu Guan <eguan@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-03-03 07:02:06 +08:00
loff_t pos = iocb->ki_pos, start = pos;
loff_t end = iocb->ki_pos + count - 1, ret = 0;
unsigned int flags = IOMAP_DIRECT;
bool wait_for_completion = is_sync_kiocb(iocb);
struct blk_plug plug;
struct iomap_dio *dio;
lockdep_assert_held(&inode->i_rwsem);
if (!count)
return 0;
dio = kmalloc(sizeof(*dio), GFP_KERNEL);
if (!dio)
return -ENOMEM;
dio->iocb = iocb;
atomic_set(&dio->ref, 1);
dio->size = 0;
dio->i_size = i_size_read(inode);
dio->end_io = end_io;
dio->error = 0;
dio->flags = 0;
dio->submit.iter = iter;
dio->submit.waiter = current;
dio->submit.cookie = BLK_QC_T_NONE;
dio->submit.last_queue = NULL;
if (iov_iter_rw(iter) == READ) {
if (pos >= dio->i_size)
goto out_free_dio;
if (iter_is_iovec(iter) && iov_iter_rw(iter) == READ)
dio->flags |= IOMAP_DIO_DIRTY;
} else {
iomap: Use FUA for pure data O_DSYNC DIO writes If we are doing direct IO writes with datasync semantics, we often have to flush metadata changes along with the data write. However, if we are overwriting existing data, there are no metadata changes that we need to flush. In this case, optimising the IO by using FUA write makes sense. We know from the IOMAP_F_DIRTY flag as to whether a specific inode requires a metadata flush - this is currently used by DAX to ensure extent modification as stable in page fault operations. For direct IO writes, we can use it to determine if we need to flush metadata or not once the data is on disk. Hence if we have been returned a mapped extent that is not new and the IO mapping is not dirty, then we can use a FUA write to provide datasync semantics. This allows us to short-cut the generic_write_sync() call in IO completion and hence avoid unnecessary operations. This makes pure direct IO data write behaviour identical to the way block devices use REQ_FUA to provide datasync semantics. On a FUA enabled device, a synchronous direct IO write workload (sequential 4k overwrites in 32MB file) had the following results: # xfs_io -fd -c "pwrite -V 1 -D 0 32m" /mnt/scratch/boo kernel time write()s write iops Write b/w ------ ---- -------- ---------- --------- (no dsync) 4s 2173/s 2173 8.5MB/s vanilla 22s 370/s 750 1.4MB/s patched 19s 420/s 420 1.6MB/s The patched code clearly doesn't send cache flushes anymore, but instead uses FUA (confirmed via blktrace), and performance improves a bit as a result. However, the benefits will be higher on workloads that mix O_DSYNC overwrites with other write IO as we won't be flushing the entire device cache on every DSYNC overwrite IO anymore. Signed-Off-By: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2018-05-03 03:54:53 +08:00
flags |= IOMAP_WRITE;
dio->flags |= IOMAP_DIO_WRITE;
iomap: Use FUA for pure data O_DSYNC DIO writes If we are doing direct IO writes with datasync semantics, we often have to flush metadata changes along with the data write. However, if we are overwriting existing data, there are no metadata changes that we need to flush. In this case, optimising the IO by using FUA write makes sense. We know from the IOMAP_F_DIRTY flag as to whether a specific inode requires a metadata flush - this is currently used by DAX to ensure extent modification as stable in page fault operations. For direct IO writes, we can use it to determine if we need to flush metadata or not once the data is on disk. Hence if we have been returned a mapped extent that is not new and the IO mapping is not dirty, then we can use a FUA write to provide datasync semantics. This allows us to short-cut the generic_write_sync() call in IO completion and hence avoid unnecessary operations. This makes pure direct IO data write behaviour identical to the way block devices use REQ_FUA to provide datasync semantics. On a FUA enabled device, a synchronous direct IO write workload (sequential 4k overwrites in 32MB file) had the following results: # xfs_io -fd -c "pwrite -V 1 -D 0 32m" /mnt/scratch/boo kernel time write()s write iops Write b/w ------ ---- -------- ---------- --------- (no dsync) 4s 2173/s 2173 8.5MB/s vanilla 22s 370/s 750 1.4MB/s patched 19s 420/s 420 1.6MB/s The patched code clearly doesn't send cache flushes anymore, but instead uses FUA (confirmed via blktrace), and performance improves a bit as a result. However, the benefits will be higher on workloads that mix O_DSYNC overwrites with other write IO as we won't be flushing the entire device cache on every DSYNC overwrite IO anymore. Signed-Off-By: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2018-05-03 03:54:53 +08:00
/* for data sync or sync, we need sync completion processing */
if (iocb->ki_flags & IOCB_DSYNC)
dio->flags |= IOMAP_DIO_NEED_SYNC;
iomap: Use FUA for pure data O_DSYNC DIO writes If we are doing direct IO writes with datasync semantics, we often have to flush metadata changes along with the data write. However, if we are overwriting existing data, there are no metadata changes that we need to flush. In this case, optimising the IO by using FUA write makes sense. We know from the IOMAP_F_DIRTY flag as to whether a specific inode requires a metadata flush - this is currently used by DAX to ensure extent modification as stable in page fault operations. For direct IO writes, we can use it to determine if we need to flush metadata or not once the data is on disk. Hence if we have been returned a mapped extent that is not new and the IO mapping is not dirty, then we can use a FUA write to provide datasync semantics. This allows us to short-cut the generic_write_sync() call in IO completion and hence avoid unnecessary operations. This makes pure direct IO data write behaviour identical to the way block devices use REQ_FUA to provide datasync semantics. On a FUA enabled device, a synchronous direct IO write workload (sequential 4k overwrites in 32MB file) had the following results: # xfs_io -fd -c "pwrite -V 1 -D 0 32m" /mnt/scratch/boo kernel time write()s write iops Write b/w ------ ---- -------- ---------- --------- (no dsync) 4s 2173/s 2173 8.5MB/s vanilla 22s 370/s 750 1.4MB/s patched 19s 420/s 420 1.6MB/s The patched code clearly doesn't send cache flushes anymore, but instead uses FUA (confirmed via blktrace), and performance improves a bit as a result. However, the benefits will be higher on workloads that mix O_DSYNC overwrites with other write IO as we won't be flushing the entire device cache on every DSYNC overwrite IO anymore. Signed-Off-By: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2018-05-03 03:54:53 +08:00
/*
* For datasync only writes, we optimistically try using FUA for
* this IO. Any non-FUA write that occurs will clear this flag,
* hence we know before completion whether a cache flush is
* necessary.
*/
if ((iocb->ki_flags & (IOCB_DSYNC | IOCB_SYNC)) == IOCB_DSYNC)
dio->flags |= IOMAP_DIO_WRITE_FUA;
}
if (iocb->ki_flags & IOCB_NOWAIT) {
if (filemap_range_has_page(mapping, start, end)) {
ret = -EAGAIN;
goto out_free_dio;
}
flags |= IOMAP_NOWAIT;
}
fs: fix data invalidation in the cleancache during direct IO Patch series "Properly invalidate data in the cleancache", v2. We've noticed that after direct IO write, buffered read sometimes gets stale data which is coming from the cleancache. The reason for this is that some direct write hooks call call invalidate_inode_pages2[_range]() conditionally iff mapping->nrpages is not zero, so we may not invalidate data in the cleancache. Another odd thing is that we check only for ->nrpages and don't check for ->nrexceptional, but invalidate_inode_pages2[_range] also invalidates exceptional entries as well. So we invalidate exceptional entries only if ->nrpages != 0? This doesn't feel right. - Patch 1 fixes direct IO writes by removing ->nrpages check. - Patch 2 fixes similar case in invalidate_bdev(). Note: I only fixed conditional cleancache_invalidate_inode() here. Do we also need to add ->nrexceptional check in into invalidate_bdev()? - Patches 3-4: some optimizations. This patch (of 4): Some direct IO write fs hooks call invalidate_inode_pages2[_range]() conditionally iff mapping->nrpages is not zero. This can't be right, because invalidate_inode_pages2[_range]() also invalidate data in the cleancache via cleancache_invalidate_inode() call. So if page cache is empty but there is some data in the cleancache, buffered read after direct IO write would get stale data from the cleancache. Also it doesn't feel right to check only for ->nrpages because invalidate_inode_pages2[_range] invalidates exceptional entries as well. Fix this by calling invalidate_inode_pages2[_range]() regardless of nrpages state. Note: nfs,cifs,9p doesn't need similar fix because the never call cleancache_get_page() (nor directly, nor via mpage_readpage[s]()), so they are not affected by this bug. Fixes: c515e1fd361c ("mm/fs: add hooks to support cleancache") Link: http://lkml.kernel.org/r/20170424164135.22350-2-aryabinin@virtuozzo.com Signed-off-by: Andrey Ryabinin <aryabinin@virtuozzo.com> Reviewed-by: Jan Kara <jack@suse.cz> Acked-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com> Cc: Alexander Viro <viro@zeniv.linux.org.uk> Cc: Ross Zwisler <ross.zwisler@linux.intel.com> Cc: Jens Axboe <axboe@kernel.dk> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Alexey Kuznetsov <kuznet@virtuozzo.com> Cc: Christoph Hellwig <hch@lst.de> Cc: Nikolay Borisov <n.borisov.lkml@gmail.com> Cc: <stable@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-05-04 05:55:59 +08:00
ret = filemap_write_and_wait_range(mapping, start, end);
if (ret)
goto out_free_dio;
/*
* Try to invalidate cache pages for the range we're direct
* writing. If this invalidation fails, tough, the write will
* still work, but racing two incompatible write paths is a
* pretty crazy thing to do, so we don't support it 100%.
*/
fs: fix data invalidation in the cleancache during direct IO Patch series "Properly invalidate data in the cleancache", v2. We've noticed that after direct IO write, buffered read sometimes gets stale data which is coming from the cleancache. The reason for this is that some direct write hooks call call invalidate_inode_pages2[_range]() conditionally iff mapping->nrpages is not zero, so we may not invalidate data in the cleancache. Another odd thing is that we check only for ->nrpages and don't check for ->nrexceptional, but invalidate_inode_pages2[_range] also invalidates exceptional entries as well. So we invalidate exceptional entries only if ->nrpages != 0? This doesn't feel right. - Patch 1 fixes direct IO writes by removing ->nrpages check. - Patch 2 fixes similar case in invalidate_bdev(). Note: I only fixed conditional cleancache_invalidate_inode() here. Do we also need to add ->nrexceptional check in into invalidate_bdev()? - Patches 3-4: some optimizations. This patch (of 4): Some direct IO write fs hooks call invalidate_inode_pages2[_range]() conditionally iff mapping->nrpages is not zero. This can't be right, because invalidate_inode_pages2[_range]() also invalidate data in the cleancache via cleancache_invalidate_inode() call. So if page cache is empty but there is some data in the cleancache, buffered read after direct IO write would get stale data from the cleancache. Also it doesn't feel right to check only for ->nrpages because invalidate_inode_pages2[_range] invalidates exceptional entries as well. Fix this by calling invalidate_inode_pages2[_range]() regardless of nrpages state. Note: nfs,cifs,9p doesn't need similar fix because the never call cleancache_get_page() (nor directly, nor via mpage_readpage[s]()), so they are not affected by this bug. Fixes: c515e1fd361c ("mm/fs: add hooks to support cleancache") Link: http://lkml.kernel.org/r/20170424164135.22350-2-aryabinin@virtuozzo.com Signed-off-by: Andrey Ryabinin <aryabinin@virtuozzo.com> Reviewed-by: Jan Kara <jack@suse.cz> Acked-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com> Cc: Alexander Viro <viro@zeniv.linux.org.uk> Cc: Ross Zwisler <ross.zwisler@linux.intel.com> Cc: Jens Axboe <axboe@kernel.dk> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Alexey Kuznetsov <kuznet@virtuozzo.com> Cc: Christoph Hellwig <hch@lst.de> Cc: Nikolay Borisov <n.borisov.lkml@gmail.com> Cc: <stable@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-05-04 05:55:59 +08:00
ret = invalidate_inode_pages2_range(mapping,
start >> PAGE_SHIFT, end >> PAGE_SHIFT);
if (ret)
dio_warn_stale_pagecache(iocb->ki_filp);
fs: fix data invalidation in the cleancache during direct IO Patch series "Properly invalidate data in the cleancache", v2. We've noticed that after direct IO write, buffered read sometimes gets stale data which is coming from the cleancache. The reason for this is that some direct write hooks call call invalidate_inode_pages2[_range]() conditionally iff mapping->nrpages is not zero, so we may not invalidate data in the cleancache. Another odd thing is that we check only for ->nrpages and don't check for ->nrexceptional, but invalidate_inode_pages2[_range] also invalidates exceptional entries as well. So we invalidate exceptional entries only if ->nrpages != 0? This doesn't feel right. - Patch 1 fixes direct IO writes by removing ->nrpages check. - Patch 2 fixes similar case in invalidate_bdev(). Note: I only fixed conditional cleancache_invalidate_inode() here. Do we also need to add ->nrexceptional check in into invalidate_bdev()? - Patches 3-4: some optimizations. This patch (of 4): Some direct IO write fs hooks call invalidate_inode_pages2[_range]() conditionally iff mapping->nrpages is not zero. This can't be right, because invalidate_inode_pages2[_range]() also invalidate data in the cleancache via cleancache_invalidate_inode() call. So if page cache is empty but there is some data in the cleancache, buffered read after direct IO write would get stale data from the cleancache. Also it doesn't feel right to check only for ->nrpages because invalidate_inode_pages2[_range] invalidates exceptional entries as well. Fix this by calling invalidate_inode_pages2[_range]() regardless of nrpages state. Note: nfs,cifs,9p doesn't need similar fix because the never call cleancache_get_page() (nor directly, nor via mpage_readpage[s]()), so they are not affected by this bug. Fixes: c515e1fd361c ("mm/fs: add hooks to support cleancache") Link: http://lkml.kernel.org/r/20170424164135.22350-2-aryabinin@virtuozzo.com Signed-off-by: Andrey Ryabinin <aryabinin@virtuozzo.com> Reviewed-by: Jan Kara <jack@suse.cz> Acked-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com> Cc: Alexander Viro <viro@zeniv.linux.org.uk> Cc: Ross Zwisler <ross.zwisler@linux.intel.com> Cc: Jens Axboe <axboe@kernel.dk> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Alexey Kuznetsov <kuznet@virtuozzo.com> Cc: Christoph Hellwig <hch@lst.de> Cc: Nikolay Borisov <n.borisov.lkml@gmail.com> Cc: <stable@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-05-04 05:55:59 +08:00
ret = 0;
if (iov_iter_rw(iter) == WRITE && !wait_for_completion &&
iomap_dio_rw: Allocate AIO completion queue before submitting dio Executing xfs/104 test in a loop on Linux-v4.13 kernel on a ppc64 machine can cause the following NULL pointer dereference, .queue_work_on+0x4c/0x80 .iomap_dio_bio_end_io+0xbc/0x1f0 .bio_endio+0x118/0x1f0 .blk_update_request+0xd0/0x470 .blk_mq_end_request+0x24/0xc0 .lo_complete_rq+0x40/0xe0 .__blk_mq_complete_request_remote+0x28/0x40 .flush_smp_call_function_queue+0xc4/0x1e0 .smp_ipi_demux_relaxed+0x8c/0x100 .icp_hv_ipi_action+0x54/0xa0 .__handle_irq_event_percpu+0x84/0x2c0 .handle_irq_event_percpu+0x28/0x80 .handle_percpu_irq+0x78/0xc0 .generic_handle_irq+0x40/0x70 .__do_irq+0x88/0x200 .call_do_irq+0x14/0x24 .do_IRQ+0x84/0x130 This occurs due to the following sequence of events, 1. Allocate dio for Direct I/O write. 2. Invoke iomap_apply() until iov_iter_count() bytes have been submitted. - Assume that we have submitted atleast one bio. Hence iomap_dio->ref value will be >= 2. - If during the second iteration, iomap_apply() ends up returning -ENOSPC, we would break out of the loop and since the 'ret' value is a negative number we end up not allocating memory for super_block->s_dio_done_wq. 3. Meanwhile, iomap_dio_bio_end_io() is invoked for bios that have been submitted and here the code ends up dereferencing the NULL pointer stored at super_block->s_dio_done_wq. This commit fixes the bug by allocating memory for super_block->s_dio_done_wq before iomap_apply() is invoked. Reported-by: Eryu Guan <eguan@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Tested-by: Eryu Guan <eguan@redhat.com> Signed-off-by: Chandan Rajendra <chandan@linux.vnet.ibm.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-09-23 02:47:33 +08:00
!inode->i_sb->s_dio_done_wq) {
ret = sb_init_dio_done_wq(inode->i_sb);
if (ret < 0)
goto out_free_dio;
}
inode_dio_begin(inode);
blk_start_plug(&plug);
do {
ret = iomap_apply(inode, pos, count, flags, ops, dio,
iomap_dio_actor);
if (ret <= 0) {
/* magic error code to fall back to buffered I/O */
if (ret == -ENOTBLK) {
wait_for_completion = true;
ret = 0;
}
break;
}
pos += ret;
iomap_dio_rw: Prevent reading file data beyond iomap_dio->i_size On a ppc64 machine executing overlayfs/019 with xfs as the lower and upper filesystem causes the following call trace, WARNING: CPU: 2 PID: 8034 at /root/repos/linux/fs/iomap.c:765 .iomap_dio_actor+0xcc/0x420 Modules linked in: CPU: 2 PID: 8034 Comm: fsstress Tainted: G L 4.11.0-rc5-next-20170405 #100 task: c000000631314880 task.stack: c0000003915d4000 NIP: c00000000035a72c LR: c00000000035a6f4 CTR: c00000000035a660 REGS: c0000003915d7570 TRAP: 0700 Tainted: G L (4.11.0-rc5-next-20170405) MSR: 800000000282b032 <SF,VEC,VSX,EE,FP,ME,IR,DR,RI> CR: 24004284 XER: 00000000 CFAR: c0000000006f7190 SOFTE: 1 GPR00: c00000000035a6f4 c0000003915d77f0 c0000000015a3f00 000000007c22f600 GPR04: 000000000022d000 0000000000002600 c0000003b2d56360 c0000003915d7960 GPR08: c0000003915d7cd0 0000000000000002 0000000000002600 c000000000521cc0 GPR12: 0000000024004284 c00000000fd80a00 000000004b04ae64 ffffffffffffffff GPR16: 000000001000ca70 0000000000000000 c0000003b2d56380 c00000000153d2b8 GPR20: 0000000000000010 c0000003bc87bac8 0000000000223000 000000000022f5ff GPR24: c0000003b2d56360 000000000000000c 0000000000002600 000000000022d000 GPR28: 0000000000000000 c0000003915d7960 c0000003b2d56360 00000000000001ff NIP [c00000000035a72c] .iomap_dio_actor+0xcc/0x420 LR [c00000000035a6f4] .iomap_dio_actor+0x94/0x420 Call Trace: [c0000003915d77f0] [c00000000035a6f4] .iomap_dio_actor+0x94/0x420 (unreliable) [c0000003915d78f0] [c00000000035b9f4] .iomap_apply+0xf4/0x1f0 [c0000003915d79d0] [c00000000035c320] .iomap_dio_rw+0x230/0x420 [c0000003915d7ae0] [c000000000512a14] .xfs_file_dio_aio_read+0x84/0x160 [c0000003915d7b80] [c000000000512d24] .xfs_file_read_iter+0x104/0x130 [c0000003915d7c10] [c0000000002d6234] .__vfs_read+0x114/0x1a0 [c0000003915d7cf0] [c0000000002d7a8c] .vfs_read+0xac/0x1a0 [c0000003915d7d90] [c0000000002d96b8] .SyS_read+0x58/0x100 [c0000003915d7e30] [c00000000000b8e0] system_call+0x38/0xfc Instruction dump: 78630020 7f831b78 7ffc07b4 7c7ce039 40820360 a13d0018 2f890003 419e0288 2f890004 419e00a0 2f890001 419e02a8 <0fe00000> 3b80fffb 38210100 7f83e378 The above problem can also be recreated on a regular xfs filesystem using the command, $ fsstress -d /mnt -l 1000 -n 1000 -p 1000 The reason for the call trace is, 1. When 'reserving' blocks for delayed allocation , XFS reserves more blocks (i.e. past file's current EOF) than required. This is done because XFS assumes that userspace might write more data and hence 'reserving' more blocks might lead to the file's new data being stored contiguously on disk. 2. The in-memory 'struct xfs_bmbt_irec' mapping the file's last extent would then cover the prealloc-ed EOF blocks in addition to the regular blocks. 3. When flushing the dirty blocks to disk, we only flush data till the file's EOF. But before writing out the dirty data, we allocate blocks on the disk for holding the file's new data. This allocation includes the blocks that are part of the 'prealloc EOF blocks'. 4. Later, when the last reference to the inode is being closed, XFS frees the unused 'prealloc EOF blocks' in xfs_inactive(). In step 3 above, When allocating space on disk for the delayed allocation range, the space allocator might sometimes allocate less blocks than required. If such an allocation ends right at the current EOF of the file, We will not be able to clear the "delayed allocation" flag for the 'prealloc EOF blocks', since we won't have dirty buffer heads associated with that range of the file. In such a situation if a Direct I/O read operation is performed on file range [X, Y] (where X < EOF and Y > EOF), we flush dirty data in the range [X, Y] and invalidate page cache for that range (Refer to iomap_dio_rw()). Later for performing the Direct I/O read, XFS obtains the extent items (which are still cached in memory) for the file range. When doing so we are not supposed to get an extent item with IOMAP_DELALLOC flag set, since the previous "flush" operation should have converted any delayed allocation data in the range [X, Y]. Hence we end up hitting a WARN_ON_ONCE(1) statement in iomap_dio_actor(). This commit fixes the bug by preventing the read operation from going beyond iomap_dio->i_size. Reported-by: Santhosh G <santhog4@linux.vnet.ibm.com> Signed-off-by: Chandan Rajendra <chandan@linux.vnet.ibm.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-04-13 02:03:20 +08:00
if (iov_iter_rw(iter) == READ && pos >= dio->i_size)
break;
} while ((count = iov_iter_count(iter)) > 0);
blk_finish_plug(&plug);
if (ret < 0)
iomap_dio_set_error(dio, ret);
iomap: Use FUA for pure data O_DSYNC DIO writes If we are doing direct IO writes with datasync semantics, we often have to flush metadata changes along with the data write. However, if we are overwriting existing data, there are no metadata changes that we need to flush. In this case, optimising the IO by using FUA write makes sense. We know from the IOMAP_F_DIRTY flag as to whether a specific inode requires a metadata flush - this is currently used by DAX to ensure extent modification as stable in page fault operations. For direct IO writes, we can use it to determine if we need to flush metadata or not once the data is on disk. Hence if we have been returned a mapped extent that is not new and the IO mapping is not dirty, then we can use a FUA write to provide datasync semantics. This allows us to short-cut the generic_write_sync() call in IO completion and hence avoid unnecessary operations. This makes pure direct IO data write behaviour identical to the way block devices use REQ_FUA to provide datasync semantics. On a FUA enabled device, a synchronous direct IO write workload (sequential 4k overwrites in 32MB file) had the following results: # xfs_io -fd -c "pwrite -V 1 -D 0 32m" /mnt/scratch/boo kernel time write()s write iops Write b/w ------ ---- -------- ---------- --------- (no dsync) 4s 2173/s 2173 8.5MB/s vanilla 22s 370/s 750 1.4MB/s patched 19s 420/s 420 1.6MB/s The patched code clearly doesn't send cache flushes anymore, but instead uses FUA (confirmed via blktrace), and performance improves a bit as a result. However, the benefits will be higher on workloads that mix O_DSYNC overwrites with other write IO as we won't be flushing the entire device cache on every DSYNC overwrite IO anymore. Signed-Off-By: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2018-05-03 03:54:53 +08:00
/*
* If all the writes we issued were FUA, we don't need to flush the
* cache on IO completion. Clear the sync flag for this case.
*/
if (dio->flags & IOMAP_DIO_WRITE_FUA)
dio->flags &= ~IOMAP_DIO_NEED_SYNC;
WRITE_ONCE(iocb->ki_cookie, dio->submit.cookie);
WRITE_ONCE(iocb->private, dio->submit.last_queue);
/*
* We are about to drop our additional submission reference, which
* might be the last reference to the dio. There are three three
* different ways we can progress here:
*
* (a) If this is the last reference we will always complete and free
* the dio ourselves.
* (b) If this is not the last reference, and we serve an asynchronous
* iocb, we must never touch the dio after the decrement, the
* I/O completion handler will complete and free it.
* (c) If this is not the last reference, but we serve a synchronous
* iocb, the I/O completion handler will wake us up on the drop
* of the final reference, and we will complete and free it here
* after we got woken by the I/O completion handler.
*/
dio->wait_for_completion = wait_for_completion;
if (!atomic_dec_and_test(&dio->ref)) {
if (!wait_for_completion)
return -EIOCBQUEUED;
for (;;) {
set_current_state(TASK_UNINTERRUPTIBLE);
if (!READ_ONCE(dio->submit.waiter))
break;
if (!(iocb->ki_flags & IOCB_HIPRI) ||
!dio->submit.last_queue ||
!blk_poll(dio->submit.last_queue,
dio->submit.cookie, true))
io_schedule();
}
__set_current_state(TASK_RUNNING);
}
return iomap_dio_complete(dio);
out_free_dio:
kfree(dio);
return ret;
}
EXPORT_SYMBOL_GPL(iomap_dio_rw);
/* Swapfile activation */
#ifdef CONFIG_SWAP
struct iomap_swapfile_info {
struct iomap iomap; /* accumulated iomap */
struct swap_info_struct *sis;
uint64_t lowest_ppage; /* lowest physical addr seen (pages) */
uint64_t highest_ppage; /* highest physical addr seen (pages) */
unsigned long nr_pages; /* number of pages collected */
int nr_extents; /* extent count */
};
/*
* Collect physical extents for this swap file. Physical extents reported to
* the swap code must be trimmed to align to a page boundary. The logical
* offset within the file is irrelevant since the swapfile code maps logical
* page numbers of the swap device to the physical page-aligned extents.
*/
static int iomap_swapfile_add_extent(struct iomap_swapfile_info *isi)
{
struct iomap *iomap = &isi->iomap;
unsigned long nr_pages;
uint64_t first_ppage;
uint64_t first_ppage_reported;
uint64_t next_ppage;
int error;
/*
* Round the start up and the end down so that the physical
* extent aligns to a page boundary.
*/
first_ppage = ALIGN(iomap->addr, PAGE_SIZE) >> PAGE_SHIFT;
next_ppage = ALIGN_DOWN(iomap->addr + iomap->length, PAGE_SIZE) >>
PAGE_SHIFT;
/* Skip too-short physical extents. */
if (first_ppage >= next_ppage)
return 0;
nr_pages = next_ppage - first_ppage;
/*
* Calculate how much swap space we're adding; the first page contains
* the swap header and doesn't count. The mm still wants that first
* page fed to add_swap_extent, however.
*/
first_ppage_reported = first_ppage;
if (iomap->offset == 0)
first_ppage_reported++;
if (isi->lowest_ppage > first_ppage_reported)
isi->lowest_ppage = first_ppage_reported;
if (isi->highest_ppage < (next_ppage - 1))
isi->highest_ppage = next_ppage - 1;
/* Add extent, set up for the next call. */
error = add_swap_extent(isi->sis, isi->nr_pages, nr_pages, first_ppage);
if (error < 0)
return error;
isi->nr_extents += error;
isi->nr_pages += nr_pages;
return 0;
}
/*
* Accumulate iomaps for this swap file. We have to accumulate iomaps because
* swap only cares about contiguous page-aligned physical extents and makes no
* distinction between written and unwritten extents.
*/
static loff_t iomap_swapfile_activate_actor(struct inode *inode, loff_t pos,
loff_t count, void *data, struct iomap *iomap)
{
struct iomap_swapfile_info *isi = data;
int error;
switch (iomap->type) {
case IOMAP_MAPPED:
case IOMAP_UNWRITTEN:
/* Only real or unwritten extents. */
break;
case IOMAP_INLINE:
/* No inline data. */
pr_err("swapon: file is inline\n");
return -EINVAL;
default:
pr_err("swapon: file has unallocated extents\n");
return -EINVAL;
}
/* No uncommitted metadata or shared blocks. */
if (iomap->flags & IOMAP_F_DIRTY) {
pr_err("swapon: file is not committed\n");
return -EINVAL;
}
if (iomap->flags & IOMAP_F_SHARED) {
pr_err("swapon: file has shared extents\n");
return -EINVAL;
}
/* Only one bdev per swap file. */
if (iomap->bdev != isi->sis->bdev) {
pr_err("swapon: file is on multiple devices\n");
return -EINVAL;
}
if (isi->iomap.length == 0) {
/* No accumulated extent, so just store it. */
memcpy(&isi->iomap, iomap, sizeof(isi->iomap));
} else if (isi->iomap.addr + isi->iomap.length == iomap->addr) {
/* Append this to the accumulated extent. */
isi->iomap.length += iomap->length;
} else {
/* Otherwise, add the retained iomap and store this one. */
error = iomap_swapfile_add_extent(isi);
if (error)
return error;
memcpy(&isi->iomap, iomap, sizeof(isi->iomap));
}
return count;
}
/*
* Iterate a swap file's iomaps to construct physical extents that can be
* passed to the swapfile subsystem.
*/
int iomap_swapfile_activate(struct swap_info_struct *sis,
struct file *swap_file, sector_t *pagespan,
const struct iomap_ops *ops)
{
struct iomap_swapfile_info isi = {
.sis = sis,
.lowest_ppage = (sector_t)-1ULL,
};
struct address_space *mapping = swap_file->f_mapping;
struct inode *inode = mapping->host;
loff_t pos = 0;
loff_t len = ALIGN_DOWN(i_size_read(inode), PAGE_SIZE);
loff_t ret;
/*
* Persist all file mapping metadata so that we won't have any
* IOMAP_F_DIRTY iomaps.
*/
ret = vfs_fsync(swap_file, 1);
if (ret)
return ret;
while (len > 0) {
ret = iomap_apply(inode, pos, len, IOMAP_REPORT,
ops, &isi, iomap_swapfile_activate_actor);
if (ret <= 0)
return ret;
pos += ret;
len -= ret;
}
if (isi.iomap.length) {
ret = iomap_swapfile_add_extent(&isi);
if (ret)
return ret;
}
*pagespan = 1 + isi.highest_ppage - isi.lowest_ppage;
sis->max = isi.nr_pages;
sis->pages = isi.nr_pages - 1;
sis->highest_bit = isi.nr_pages - 1;
return isi.nr_extents;
}
EXPORT_SYMBOL_GPL(iomap_swapfile_activate);
#endif /* CONFIG_SWAP */
static loff_t
iomap_bmap_actor(struct inode *inode, loff_t pos, loff_t length,
void *data, struct iomap *iomap)
{
sector_t *bno = data, addr;
if (iomap->type == IOMAP_MAPPED) {
addr = (pos - iomap->offset + iomap->addr) >> inode->i_blkbits;
if (addr > INT_MAX)
WARN(1, "would truncate bmap result\n");
else
*bno = addr;
}
return 0;
}
/* legacy ->bmap interface. 0 is the error return (!) */
sector_t
iomap_bmap(struct address_space *mapping, sector_t bno,
const struct iomap_ops *ops)
{
struct inode *inode = mapping->host;
loff_t pos = bno << inode->i_blkbits;
unsigned blocksize = i_blocksize(inode);
if (filemap_write_and_wait(mapping))
return 0;
bno = 0;
iomap_apply(inode, pos, blocksize, 0, ops, &bno, iomap_bmap_actor);
return bno;
}
EXPORT_SYMBOL_GPL(iomap_bmap);