linux_dsm_epyc7002/fs/xfs/xfs_aops.c

1197 lines
34 KiB
C
Raw Normal View History

// SPDX-License-Identifier: GPL-2.0
/*
* Copyright (c) 2000-2005 Silicon Graphics, Inc.
* Copyright (c) 2016-2018 Christoph Hellwig.
* All Rights Reserved.
*/
#include "xfs.h"
#include "xfs_shared.h"
#include "xfs_format.h"
#include "xfs_log_format.h"
#include "xfs_trans_resv.h"
#include "xfs_mount.h"
#include "xfs_inode.h"
#include "xfs_trans.h"
#include "xfs_inode_item.h"
#include "xfs_alloc.h"
#include "xfs_error.h"
#include "xfs_iomap.h"
xfs: event tracing support Convert the old xfs tracing support that could only be used with the out of tree kdb and xfsidbg patches to use the generic event tracer. To use it make sure CONFIG_EVENT_TRACING is enabled and then enable all xfs trace channels by: echo 1 > /sys/kernel/debug/tracing/events/xfs/enable or alternatively enable single events by just doing the same in one event subdirectory, e.g. echo 1 > /sys/kernel/debug/tracing/events/xfs/xfs_ihold/enable or set more complex filters, etc. In Documentation/trace/events.txt all this is desctribed in more detail. To reads the events do a cat /sys/kernel/debug/tracing/trace Compared to the last posting this patch converts the tracing mostly to the one tracepoint per callsite model that other users of the new tracing facility also employ. This allows a very fine-grained control of the tracing, a cleaner output of the traces and also enables the perf tool to use each tracepoint as a virtual performance counter, allowing us to e.g. count how often certain workloads git various spots in XFS. Take a look at http://lwn.net/Articles/346470/ for some examples. Also the btree tracing isn't included at all yet, as it will require additional core tracing features not in mainline yet, I plan to deliver it later. And the really nice thing about this patch is that it actually removes many lines of code while adding this nice functionality: fs/xfs/Makefile | 8 fs/xfs/linux-2.6/xfs_acl.c | 1 fs/xfs/linux-2.6/xfs_aops.c | 52 - fs/xfs/linux-2.6/xfs_aops.h | 2 fs/xfs/linux-2.6/xfs_buf.c | 117 +-- fs/xfs/linux-2.6/xfs_buf.h | 33 fs/xfs/linux-2.6/xfs_fs_subr.c | 3 fs/xfs/linux-2.6/xfs_ioctl.c | 1 fs/xfs/linux-2.6/xfs_ioctl32.c | 1 fs/xfs/linux-2.6/xfs_iops.c | 1 fs/xfs/linux-2.6/xfs_linux.h | 1 fs/xfs/linux-2.6/xfs_lrw.c | 87 -- fs/xfs/linux-2.6/xfs_lrw.h | 45 - fs/xfs/linux-2.6/xfs_super.c | 104 --- fs/xfs/linux-2.6/xfs_super.h | 7 fs/xfs/linux-2.6/xfs_sync.c | 1 fs/xfs/linux-2.6/xfs_trace.c | 75 ++ fs/xfs/linux-2.6/xfs_trace.h | 1369 +++++++++++++++++++++++++++++++++++++++++ fs/xfs/linux-2.6/xfs_vnode.h | 4 fs/xfs/quota/xfs_dquot.c | 110 --- fs/xfs/quota/xfs_dquot.h | 21 fs/xfs/quota/xfs_qm.c | 40 - fs/xfs/quota/xfs_qm_syscalls.c | 4 fs/xfs/support/ktrace.c | 323 --------- fs/xfs/support/ktrace.h | 85 -- fs/xfs/xfs.h | 16 fs/xfs/xfs_ag.h | 14 fs/xfs/xfs_alloc.c | 230 +----- fs/xfs/xfs_alloc.h | 27 fs/xfs/xfs_alloc_btree.c | 1 fs/xfs/xfs_attr.c | 107 --- fs/xfs/xfs_attr.h | 10 fs/xfs/xfs_attr_leaf.c | 14 fs/xfs/xfs_attr_sf.h | 40 - fs/xfs/xfs_bmap.c | 507 +++------------ fs/xfs/xfs_bmap.h | 49 - fs/xfs/xfs_bmap_btree.c | 6 fs/xfs/xfs_btree.c | 5 fs/xfs/xfs_btree_trace.h | 17 fs/xfs/xfs_buf_item.c | 87 -- fs/xfs/xfs_buf_item.h | 20 fs/xfs/xfs_da_btree.c | 3 fs/xfs/xfs_da_btree.h | 7 fs/xfs/xfs_dfrag.c | 2 fs/xfs/xfs_dir2.c | 8 fs/xfs/xfs_dir2_block.c | 20 fs/xfs/xfs_dir2_leaf.c | 21 fs/xfs/xfs_dir2_node.c | 27 fs/xfs/xfs_dir2_sf.c | 26 fs/xfs/xfs_dir2_trace.c | 216 ------ fs/xfs/xfs_dir2_trace.h | 72 -- fs/xfs/xfs_filestream.c | 8 fs/xfs/xfs_fsops.c | 2 fs/xfs/xfs_iget.c | 111 --- fs/xfs/xfs_inode.c | 67 -- fs/xfs/xfs_inode.h | 76 -- fs/xfs/xfs_inode_item.c | 5 fs/xfs/xfs_iomap.c | 85 -- fs/xfs/xfs_iomap.h | 8 fs/xfs/xfs_log.c | 181 +---- fs/xfs/xfs_log_priv.h | 20 fs/xfs/xfs_log_recover.c | 1 fs/xfs/xfs_mount.c | 2 fs/xfs/xfs_quota.h | 8 fs/xfs/xfs_rename.c | 1 fs/xfs/xfs_rtalloc.c | 1 fs/xfs/xfs_rw.c | 3 fs/xfs/xfs_trans.h | 47 + fs/xfs/xfs_trans_buf.c | 62 - fs/xfs/xfs_vnodeops.c | 8 70 files changed, 2151 insertions(+), 2592 deletions(-) Signed-off-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2009-12-15 06:14:59 +07:00
#include "xfs_trace.h"
#include "xfs_bmap.h"
#include "xfs_bmap_util.h"
#include "xfs_bmap_btree.h"
#include "xfs_reflink.h"
#include <linux/writeback.h>
xfs: Introduce writeback context for writepages xfs_vm_writepages() calls generic_writepages to writeback a range of a file, but then xfs_vm_writepage() clusters pages itself as it does not have any context it can pass between->writepage calls from __write_cache_pages(). Introduce a writeback context for xfs_vm_writepages() and call __write_cache_pages directly with our own writepage callback so that we can pass that context to each writepage invocation. This encapsulates the current mapping, whether it is valid or not, the current ioend and it's IO type and the ioend chain being built. This requires us to move the ioend submission up to the level where the writepage context is declared. This does mean we do not submit IO until we packaged the entire writeback range, but with the block plugging in the writepages call this is the way IO is submitted, anyway. It also means that we need to handle discontiguous page ranges. If the pages sent down by write_cache_pages to the writepage callback are discontiguous, we need to detect this and put each discontiguous page range into individual ioends. This is needed to ensure that the ioend accurately represents the range of the file that it covers so that file size updates during IO completion set the size correctly. Failure to take into account the discontiguous ranges results in files being too small when writeback patterns are non-sequential. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-02-15 13:21:19 +07:00
/*
* structure owned by writepages passed to individual writepage calls
*/
struct xfs_writepage_ctx {
struct xfs_bmbt_irec imap;
int fork;
xfs: validate writeback mapping using data fork seq counter The writeback code caches the current extent mapping across multiple xfs_do_writepage() calls to avoid repeated lookups for sequential pages backed by the same extent. This is known to be slightly racy with extent fork changes in certain difficult to reproduce scenarios. The cached extent is trimmed to within EOF to help avoid the most common vector for this problem via speculative preallocation management, but this is a band-aid that does not address the fundamental problem. Now that we have an xfs_ifork sequence counter mechanism used to facilitate COW writeback, we can use the same mechanism to validate consistency between the data fork and cached writeback mappings. On its face, this is somewhat of a big hammer approach because any change to the data fork invalidates any mapping currently cached by a writeback in progress regardless of whether the data fork change overlaps with the range under writeback. In practice, however, the impact of this approach is minimal in most cases. First, data fork changes (delayed allocations) caused by sustained sequential buffered writes are amortized across speculative preallocations. This means that a cached mapping won't be invalidated by each buffered write of a common file copy workload, but rather only on less frequent allocation events. Second, the extent tree is always entirely in-core so an additional lookup of a usable extent mostly costs a shared ilock cycle and in-memory tree lookup. This means that a cached mapping reval is relatively cheap compared to the I/O itself. Third, spurious invalidations don't impact ioend construction. This means that even if the same extent is revalidated multiple times across multiple writepage instances, we still construct and submit the same size ioend (and bio) if the blocks are physically contiguous. Update struct xfs_writepage_ctx with a new field to hold the sequence number of the data fork associated with the currently cached mapping. Check the wpc seqno against the data fork when the mapping is validated and reestablish the mapping whenever the fork has changed since the mapping was cached. This ensures that writeback always uses a valid extent mapping and thus prevents lost writebacks and stale delalloc block problems. Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Allison Henderson <allison.henderson@oracle.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>
2019-02-02 00:14:23 +07:00
unsigned int data_seq;
unsigned int cow_seq;
xfs: Introduce writeback context for writepages xfs_vm_writepages() calls generic_writepages to writeback a range of a file, but then xfs_vm_writepage() clusters pages itself as it does not have any context it can pass between->writepage calls from __write_cache_pages(). Introduce a writeback context for xfs_vm_writepages() and call __write_cache_pages directly with our own writepage callback so that we can pass that context to each writepage invocation. This encapsulates the current mapping, whether it is valid or not, the current ioend and it's IO type and the ioend chain being built. This requires us to move the ioend submission up to the level where the writepage context is declared. This does mean we do not submit IO until we packaged the entire writeback range, but with the block plugging in the writepages call this is the way IO is submitted, anyway. It also means that we need to handle discontiguous page ranges. If the pages sent down by write_cache_pages to the writepage callback are discontiguous, we need to detect this and put each discontiguous page range into individual ioends. This is needed to ensure that the ioend accurately represents the range of the file that it covers so that file size updates during IO completion set the size correctly. Failure to take into account the discontiguous ranges results in files being too small when writeback patterns are non-sequential. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-02-15 13:21:19 +07:00
struct xfs_ioend *ioend;
};
struct block_device *
xfs_find_bdev_for_inode(
struct inode *inode)
{
struct xfs_inode *ip = XFS_I(inode);
struct xfs_mount *mp = ip->i_mount;
if (XFS_IS_REALTIME_INODE(ip))
return mp->m_rtdev_targp->bt_bdev;
else
return mp->m_ddev_targp->bt_bdev;
}
struct dax_device *
xfs_find_daxdev_for_inode(
struct inode *inode)
{
struct xfs_inode *ip = XFS_I(inode);
struct xfs_mount *mp = ip->i_mount;
if (XFS_IS_REALTIME_INODE(ip))
return mp->m_rtdev_targp->bt_daxdev;
else
return mp->m_ddev_targp->bt_daxdev;
}
static void
xfs_finish_page_writeback(
struct inode *inode,
struct bio_vec *bvec,
int error)
{
struct iomap_page *iop = to_iomap_page(bvec->bv_page);
if (error) {
SetPageError(bvec->bv_page);
mapping_set_error(inode->i_mapping, -EIO);
}
ASSERT(iop || i_blocksize(inode) == PAGE_SIZE);
ASSERT(!iop || atomic_read(&iop->write_count) > 0);
if (!iop || atomic_dec_and_test(&iop->write_count))
end_page_writeback(bvec->bv_page);
}
/*
* We're now finished for good with this ioend structure. Update the page
* state, release holds on bios, and finally free up memory. Do not use the
* ioend after this.
*/
STATIC void
xfs_destroy_ioend(
struct xfs_ioend *ioend,
int error)
{
struct inode *inode = ioend->io_inode;
struct bio *bio = &ioend->io_inline_bio;
struct bio *last = ioend->io_bio, *next;
u64 start = bio->bi_iter.bi_sector;
bool quiet = bio_flagged(bio, BIO_QUIET);
for (bio = &ioend->io_inline_bio; bio; bio = next) {
struct bio_vec *bvec;
struct bvec_iter_all iter_all;
/*
* For the last bio, bi_private points to the ioend, so we
* need to explicitly end the iteration here.
*/
if (bio == last)
next = NULL;
else
next = bio->bi_private;
/* walk each page on bio, ending page IO on them */
bio_for_each_segment_all(bvec, bio, iter_all)
xfs_finish_page_writeback(inode, bvec, error);
bio_put(bio);
}
if (unlikely(error && !quiet)) {
xfs_err_ratelimited(XFS_I(inode)->i_mount,
"writeback error on sector %llu", start);
}
}
/*
* Fast and loose check if this write could update the on-disk inode size.
*/
static inline bool xfs_ioend_is_append(struct xfs_ioend *ioend)
{
return ioend->io_offset + ioend->io_size >
XFS_I(ioend->io_inode)->i_d.di_size;
}
STATIC int
xfs_setfilesize_trans_alloc(
struct xfs_ioend *ioend)
{
struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount;
struct xfs_trans *tp;
int error;
error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0,
XFS_TRANS_NOFS, &tp);
if (error)
return error;
ioend->io_append_trans = tp;
/*
xfs: fix direct IO nested transaction deadlock. The direct IO path can do a nested transaction reservation when writing past the EOF. The first transaction is the append transaction for setting the filesize at IO completion, but we can also need a transaction for allocation of blocks. If the log is low on space due to reservations and small log, the append transaction can be granted after wating for space as the only active transaction in the system. This then attempts a reservation for an allocation, which there isn't space in the log for, and the reservation sleeps. The result is that there is nothing left in the system to wake up all the processes waiting for log space to come free. The stack trace that shows this deadlock is relatively innocuous: xlog_grant_head_wait xlog_grant_head_check xfs_log_reserve xfs_trans_reserve xfs_iomap_write_direct __xfs_get_blocks xfs_get_blocks_direct do_blockdev_direct_IO __blockdev_direct_IO xfs_vm_direct_IO generic_file_direct_write xfs_file_dio_aio_writ xfs_file_aio_write do_sync_write vfs_write This was discovered on a filesystem with a log of only 10MB, and a log stripe unit of 256k whih increased the base reservations by 512k. Hence a allocation transaction requires 1.2MB of log space to be available instead of only 260k, and so greatly increased the chance that there wouldn't be enough log space available for the nested transaction to succeed. The key to reproducing it is this mkfs command: mkfs.xfs -f -d agcount=16,su=256k,sw=12 -l su=256k,size=2560b $SCRATCH_DEV The test case was a 1000 fsstress processes running with random freeze and unfreezes every few seconds. Thanks to Eryu Guan (eguan@redhat.com) for writing the test that found this on a system with a somewhat unique default configuration.... cc: <stable@vger.kernel.org> Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Andrew Dahl <adahl@sgi.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2012-11-28 09:01:00 +07:00
* We may pass freeze protection with a transaction. So tell lockdep
* we released it.
*/
__sb_writers_release(ioend->io_inode->i_sb, SB_FREEZE_FS);
/*
* We hand off the transaction to the completion thread now, so
* clear the flag here.
*/
current_restore_flags_nested(&tp->t_pflags, PF_MEMALLOC_NOFS);
return 0;
}
[XFS] Fix to prevent the notorious 'NULL files' problem after a crash. The problem that has been addressed is that of synchronising updates of the file size with writes that extend a file. Without the fix the update of a file's size, as a result of a write beyond eof, is independent of when the cached data is flushed to disk. Often the file size update would be written to the filesystem log before the data is flushed to disk. When a system crashes between these two events and the filesystem log is replayed on mount the file's size will be set but since the contents never made it to disk the file is full of holes. If some of the cached data was flushed to disk then it may just be a section of the file at the end that has holes. There are existing fixes to help alleviate this problem, particularly in the case where a file has been truncated, that force cached data to be flushed to disk when the file is closed. If the system crashes while the file(s) are still open then this flushing will never occur. The fix that we have implemented is to introduce a second file size, called the in-memory file size, that represents the current file size as viewed by the user. The existing file size, called the on-disk file size, is the one that get's written to the filesystem log and we only update it when it is safe to do so. When we write to a file beyond eof we only update the in- memory file size in the write operation. Later when the I/O operation, that flushes the cached data to disk completes, an I/O completion routine will update the on-disk file size. The on-disk file size will be updated to the maximum offset of the I/O or to the value of the in-memory file size if the I/O includes eof. SGI-PV: 958522 SGI-Modid: xfs-linux-melb:xfs-kern:28322a Signed-off-by: Lachlan McIlroy <lachlan@sgi.com> Signed-off-by: David Chinner <dgc@sgi.com> Signed-off-by: Tim Shimmin <tes@sgi.com>
2007-05-08 10:49:46 +07:00
/*
* Update on-disk file size now that data has been written to disk.
[XFS] Fix to prevent the notorious 'NULL files' problem after a crash. The problem that has been addressed is that of synchronising updates of the file size with writes that extend a file. Without the fix the update of a file's size, as a result of a write beyond eof, is independent of when the cached data is flushed to disk. Often the file size update would be written to the filesystem log before the data is flushed to disk. When a system crashes between these two events and the filesystem log is replayed on mount the file's size will be set but since the contents never made it to disk the file is full of holes. If some of the cached data was flushed to disk then it may just be a section of the file at the end that has holes. There are existing fixes to help alleviate this problem, particularly in the case where a file has been truncated, that force cached data to be flushed to disk when the file is closed. If the system crashes while the file(s) are still open then this flushing will never occur. The fix that we have implemented is to introduce a second file size, called the in-memory file size, that represents the current file size as viewed by the user. The existing file size, called the on-disk file size, is the one that get's written to the filesystem log and we only update it when it is safe to do so. When we write to a file beyond eof we only update the in- memory file size in the write operation. Later when the I/O operation, that flushes the cached data to disk completes, an I/O completion routine will update the on-disk file size. The on-disk file size will be updated to the maximum offset of the I/O or to the value of the in-memory file size if the I/O includes eof. SGI-PV: 958522 SGI-Modid: xfs-linux-melb:xfs-kern:28322a Signed-off-by: Lachlan McIlroy <lachlan@sgi.com> Signed-off-by: David Chinner <dgc@sgi.com> Signed-off-by: Tim Shimmin <tes@sgi.com>
2007-05-08 10:49:46 +07:00
*/
STATIC int
__xfs_setfilesize(
struct xfs_inode *ip,
struct xfs_trans *tp,
xfs_off_t offset,
size_t size)
[XFS] Fix to prevent the notorious 'NULL files' problem after a crash. The problem that has been addressed is that of synchronising updates of the file size with writes that extend a file. Without the fix the update of a file's size, as a result of a write beyond eof, is independent of when the cached data is flushed to disk. Often the file size update would be written to the filesystem log before the data is flushed to disk. When a system crashes between these two events and the filesystem log is replayed on mount the file's size will be set but since the contents never made it to disk the file is full of holes. If some of the cached data was flushed to disk then it may just be a section of the file at the end that has holes. There are existing fixes to help alleviate this problem, particularly in the case where a file has been truncated, that force cached data to be flushed to disk when the file is closed. If the system crashes while the file(s) are still open then this flushing will never occur. The fix that we have implemented is to introduce a second file size, called the in-memory file size, that represents the current file size as viewed by the user. The existing file size, called the on-disk file size, is the one that get's written to the filesystem log and we only update it when it is safe to do so. When we write to a file beyond eof we only update the in- memory file size in the write operation. Later when the I/O operation, that flushes the cached data to disk completes, an I/O completion routine will update the on-disk file size. The on-disk file size will be updated to the maximum offset of the I/O or to the value of the in-memory file size if the I/O includes eof. SGI-PV: 958522 SGI-Modid: xfs-linux-melb:xfs-kern:28322a Signed-off-by: Lachlan McIlroy <lachlan@sgi.com> Signed-off-by: David Chinner <dgc@sgi.com> Signed-off-by: Tim Shimmin <tes@sgi.com>
2007-05-08 10:49:46 +07:00
{
xfs_fsize_t isize;
xfs_ilock(ip, XFS_ILOCK_EXCL);
isize = xfs_new_eof(ip, offset + size);
if (!isize) {
xfs_iunlock(ip, XFS_ILOCK_EXCL);
xfs_trans_cancel(tp);
return 0;
[XFS] Fix to prevent the notorious 'NULL files' problem after a crash. The problem that has been addressed is that of synchronising updates of the file size with writes that extend a file. Without the fix the update of a file's size, as a result of a write beyond eof, is independent of when the cached data is flushed to disk. Often the file size update would be written to the filesystem log before the data is flushed to disk. When a system crashes between these two events and the filesystem log is replayed on mount the file's size will be set but since the contents never made it to disk the file is full of holes. If some of the cached data was flushed to disk then it may just be a section of the file at the end that has holes. There are existing fixes to help alleviate this problem, particularly in the case where a file has been truncated, that force cached data to be flushed to disk when the file is closed. If the system crashes while the file(s) are still open then this flushing will never occur. The fix that we have implemented is to introduce a second file size, called the in-memory file size, that represents the current file size as viewed by the user. The existing file size, called the on-disk file size, is the one that get's written to the filesystem log and we only update it when it is safe to do so. When we write to a file beyond eof we only update the in- memory file size in the write operation. Later when the I/O operation, that flushes the cached data to disk completes, an I/O completion routine will update the on-disk file size. The on-disk file size will be updated to the maximum offset of the I/O or to the value of the in-memory file size if the I/O includes eof. SGI-PV: 958522 SGI-Modid: xfs-linux-melb:xfs-kern:28322a Signed-off-by: Lachlan McIlroy <lachlan@sgi.com> Signed-off-by: David Chinner <dgc@sgi.com> Signed-off-by: Tim Shimmin <tes@sgi.com>
2007-05-08 10:49:46 +07:00
}
trace_xfs_setfilesize(ip, offset, size);
ip->i_d.di_size = isize;
xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
return xfs_trans_commit(tp);
}
int
xfs_setfilesize(
struct xfs_inode *ip,
xfs_off_t offset,
size_t size)
{
struct xfs_mount *mp = ip->i_mount;
struct xfs_trans *tp;
int error;
error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp);
if (error)
return error;
return __xfs_setfilesize(ip, tp, offset, size);
}
STATIC int
xfs_setfilesize_ioend(
struct xfs_ioend *ioend,
int error)
{
struct xfs_inode *ip = XFS_I(ioend->io_inode);
struct xfs_trans *tp = ioend->io_append_trans;
/*
* The transaction may have been allocated in the I/O submission thread,
* thus we need to mark ourselves as being in a transaction manually.
* Similarly for freeze protection.
*/
current_set_flags_nested(&tp->t_pflags, PF_MEMALLOC_NOFS);
__sb_writers_acquired(VFS_I(ip)->i_sb, SB_FREEZE_FS);
/* we abort the update if there was an IO error */
if (error) {
xfs_trans_cancel(tp);
return error;
}
return __xfs_setfilesize(ip, tp, ioend->io_offset, ioend->io_size);
}
/*
* IO write completion.
*/
STATIC void
xfs: implement per-inode writeback completion queues When scheduling writeback of dirty file data in the page cache, XFS uses IO completion workqueue items to ensure that filesystem metadata only updates after the write completes successfully. This is essential for converting unwritten extents to real extents at the right time and performing COW remappings. Unfortunately, XFS queues each IO completion work item to an unbounded workqueue, which means that the kernel can spawn dozens of threads to try to handle the items quickly. These threads need to take the ILOCK to update file metadata, which results in heavy ILOCK contention if a large number of the work items target a single file, which is inefficient. Worse yet, the writeback completion threads get stuck waiting for the ILOCK while holding transaction reservations, which can use up all available log reservation space. When that happens, metadata updates to other parts of the filesystem grind to a halt, even if the filesystem could otherwise have handled it. Even worse, if one of the things grinding to a halt happens to be a thread in the middle of a defer-ops finish holding the same ILOCK and trying to obtain more log reservation having exhausted the permanent reservation, we now have an ABBA deadlock - writeback completion has a transaction reserved and wants the ILOCK, and someone else has the ILOCK and wants a transaction reservation. Therefore, we create a per-inode writeback io completion queue + work item. When writeback finishes, it can add the ioend to the per-inode queue and let the single worker item process that queue. This dramatically cuts down on the number of kworkers and ILOCK contention in the system, and seems to have eliminated an occasional deadlock I was seeing while running generic/476. Testing with a program that simulates a heavy random-write workload to a single file demonstrates that the number of kworkers drops from approximately 120 threads per file to 1, without dramatically changing write bandwidth or pagecache access latency. Note that we leave the xfs-conv workqueue's max_active alone because we still want to be able to run ioend processing for as many inodes as the system can handle. Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Brian Foster <bfoster@redhat.com>
2019-04-16 03:13:20 +07:00
xfs_end_ioend(
struct xfs_ioend *ioend)
{
struct list_head ioend_list;
struct xfs_inode *ip = XFS_I(ioend->io_inode);
xfs_off_t offset = ioend->io_offset;
size_t size = ioend->io_size;
int error;
[XFS] Fix to prevent the notorious 'NULL files' problem after a crash. The problem that has been addressed is that of synchronising updates of the file size with writes that extend a file. Without the fix the update of a file's size, as a result of a write beyond eof, is independent of when the cached data is flushed to disk. Often the file size update would be written to the filesystem log before the data is flushed to disk. When a system crashes between these two events and the filesystem log is replayed on mount the file's size will be set but since the contents never made it to disk the file is full of holes. If some of the cached data was flushed to disk then it may just be a section of the file at the end that has holes. There are existing fixes to help alleviate this problem, particularly in the case where a file has been truncated, that force cached data to be flushed to disk when the file is closed. If the system crashes while the file(s) are still open then this flushing will never occur. The fix that we have implemented is to introduce a second file size, called the in-memory file size, that represents the current file size as viewed by the user. The existing file size, called the on-disk file size, is the one that get's written to the filesystem log and we only update it when it is safe to do so. When we write to a file beyond eof we only update the in- memory file size in the write operation. Later when the I/O operation, that flushes the cached data to disk completes, an I/O completion routine will update the on-disk file size. The on-disk file size will be updated to the maximum offset of the I/O or to the value of the in-memory file size if the I/O includes eof. SGI-PV: 958522 SGI-Modid: xfs-linux-melb:xfs-kern:28322a Signed-off-by: Lachlan McIlroy <lachlan@sgi.com> Signed-off-by: David Chinner <dgc@sgi.com> Signed-off-by: Tim Shimmin <tes@sgi.com>
2007-05-08 10:49:46 +07:00
/*
* Just clean up the in-memory strutures if the fs has been shut down.
*/
if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
error = -EIO;
goto done;
}
/*
* Clean up any COW blocks on an I/O error.
*/
error = blk_status_to_errno(ioend->io_bio->bi_status);
if (unlikely(error)) {
if (ioend->io_fork == XFS_COW_FORK)
xfs_reflink_cancel_cow_range(ip, offset, size, true);
goto done;
}
/*
* Success: commit the COW or unwritten blocks if needed.
*/
if (ioend->io_fork == XFS_COW_FORK)
error = xfs_reflink_end_cow(ip, offset, size);
else if (ioend->io_state == XFS_EXT_UNWRITTEN)
error = xfs_iomap_write_unwritten(ip, offset, size, false);
else
ASSERT(!xfs_ioend_is_append(ioend) || ioend->io_append_trans);
[XFS] Fix to prevent the notorious 'NULL files' problem after a crash. The problem that has been addressed is that of synchronising updates of the file size with writes that extend a file. Without the fix the update of a file's size, as a result of a write beyond eof, is independent of when the cached data is flushed to disk. Often the file size update would be written to the filesystem log before the data is flushed to disk. When a system crashes between these two events and the filesystem log is replayed on mount the file's size will be set but since the contents never made it to disk the file is full of holes. If some of the cached data was flushed to disk then it may just be a section of the file at the end that has holes. There are existing fixes to help alleviate this problem, particularly in the case where a file has been truncated, that force cached data to be flushed to disk when the file is closed. If the system crashes while the file(s) are still open then this flushing will never occur. The fix that we have implemented is to introduce a second file size, called the in-memory file size, that represents the current file size as viewed by the user. The existing file size, called the on-disk file size, is the one that get's written to the filesystem log and we only update it when it is safe to do so. When we write to a file beyond eof we only update the in- memory file size in the write operation. Later when the I/O operation, that flushes the cached data to disk completes, an I/O completion routine will update the on-disk file size. The on-disk file size will be updated to the maximum offset of the I/O or to the value of the in-memory file size if the I/O includes eof. SGI-PV: 958522 SGI-Modid: xfs-linux-melb:xfs-kern:28322a Signed-off-by: Lachlan McIlroy <lachlan@sgi.com> Signed-off-by: David Chinner <dgc@sgi.com> Signed-off-by: Tim Shimmin <tes@sgi.com>
2007-05-08 10:49:46 +07:00
done:
if (ioend->io_append_trans)
error = xfs_setfilesize_ioend(ioend, error);
list_replace_init(&ioend->io_list, &ioend_list);
xfs_destroy_ioend(ioend, error);
while (!list_empty(&ioend_list)) {
ioend = list_first_entry(&ioend_list, struct xfs_ioend,
io_list);
list_del_init(&ioend->io_list);
xfs_destroy_ioend(ioend, error);
}
}
/*
* We can merge two adjacent ioends if they have the same set of work to do.
*/
static bool
xfs_ioend_can_merge(
struct xfs_ioend *ioend,
int ioend_error,
struct xfs_ioend *next)
{
int next_error;
next_error = blk_status_to_errno(next->io_bio->bi_status);
if (ioend_error != next_error)
return false;
if ((ioend->io_fork == XFS_COW_FORK) ^ (next->io_fork == XFS_COW_FORK))
return false;
if ((ioend->io_state == XFS_EXT_UNWRITTEN) ^
(next->io_state == XFS_EXT_UNWRITTEN))
return false;
if (ioend->io_offset + ioend->io_size != next->io_offset)
return false;
if (xfs_ioend_is_append(ioend) != xfs_ioend_is_append(next))
return false;
return true;
}
/* Try to merge adjacent completions. */
STATIC void
xfs_ioend_try_merge(
struct xfs_ioend *ioend,
struct list_head *more_ioends)
{
struct xfs_ioend *next_ioend;
int ioend_error;
int error;
if (list_empty(more_ioends))
return;
ioend_error = blk_status_to_errno(ioend->io_bio->bi_status);
while (!list_empty(more_ioends)) {
next_ioend = list_first_entry(more_ioends, struct xfs_ioend,
io_list);
if (!xfs_ioend_can_merge(ioend, ioend_error, next_ioend))
break;
list_move_tail(&next_ioend->io_list, &ioend->io_list);
ioend->io_size += next_ioend->io_size;
if (ioend->io_append_trans) {
error = xfs_setfilesize_ioend(next_ioend, 1);
ASSERT(error == 1);
}
}
}
/* list_sort compare function for ioends */
static int
xfs_ioend_compare(
void *priv,
struct list_head *a,
struct list_head *b)
{
struct xfs_ioend *ia;
struct xfs_ioend *ib;
ia = container_of(a, struct xfs_ioend, io_list);
ib = container_of(b, struct xfs_ioend, io_list);
if (ia->io_offset < ib->io_offset)
return -1;
else if (ia->io_offset > ib->io_offset)
return 1;
return 0;
}
xfs: implement per-inode writeback completion queues When scheduling writeback of dirty file data in the page cache, XFS uses IO completion workqueue items to ensure that filesystem metadata only updates after the write completes successfully. This is essential for converting unwritten extents to real extents at the right time and performing COW remappings. Unfortunately, XFS queues each IO completion work item to an unbounded workqueue, which means that the kernel can spawn dozens of threads to try to handle the items quickly. These threads need to take the ILOCK to update file metadata, which results in heavy ILOCK contention if a large number of the work items target a single file, which is inefficient. Worse yet, the writeback completion threads get stuck waiting for the ILOCK while holding transaction reservations, which can use up all available log reservation space. When that happens, metadata updates to other parts of the filesystem grind to a halt, even if the filesystem could otherwise have handled it. Even worse, if one of the things grinding to a halt happens to be a thread in the middle of a defer-ops finish holding the same ILOCK and trying to obtain more log reservation having exhausted the permanent reservation, we now have an ABBA deadlock - writeback completion has a transaction reserved and wants the ILOCK, and someone else has the ILOCK and wants a transaction reservation. Therefore, we create a per-inode writeback io completion queue + work item. When writeback finishes, it can add the ioend to the per-inode queue and let the single worker item process that queue. This dramatically cuts down on the number of kworkers and ILOCK contention in the system, and seems to have eliminated an occasional deadlock I was seeing while running generic/476. Testing with a program that simulates a heavy random-write workload to a single file demonstrates that the number of kworkers drops from approximately 120 threads per file to 1, without dramatically changing write bandwidth or pagecache access latency. Note that we leave the xfs-conv workqueue's max_active alone because we still want to be able to run ioend processing for as many inodes as the system can handle. Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Brian Foster <bfoster@redhat.com>
2019-04-16 03:13:20 +07:00
/* Finish all pending io completions. */
void
xfs_end_io(
struct work_struct *work)
{
struct xfs_inode *ip;
struct xfs_ioend *ioend;
struct list_head completion_list;
unsigned long flags;
ip = container_of(work, struct xfs_inode, i_ioend_work);
spin_lock_irqsave(&ip->i_ioend_lock, flags);
list_replace_init(&ip->i_ioend_list, &completion_list);
spin_unlock_irqrestore(&ip->i_ioend_lock, flags);
list_sort(NULL, &completion_list, xfs_ioend_compare);
xfs: implement per-inode writeback completion queues When scheduling writeback of dirty file data in the page cache, XFS uses IO completion workqueue items to ensure that filesystem metadata only updates after the write completes successfully. This is essential for converting unwritten extents to real extents at the right time and performing COW remappings. Unfortunately, XFS queues each IO completion work item to an unbounded workqueue, which means that the kernel can spawn dozens of threads to try to handle the items quickly. These threads need to take the ILOCK to update file metadata, which results in heavy ILOCK contention if a large number of the work items target a single file, which is inefficient. Worse yet, the writeback completion threads get stuck waiting for the ILOCK while holding transaction reservations, which can use up all available log reservation space. When that happens, metadata updates to other parts of the filesystem grind to a halt, even if the filesystem could otherwise have handled it. Even worse, if one of the things grinding to a halt happens to be a thread in the middle of a defer-ops finish holding the same ILOCK and trying to obtain more log reservation having exhausted the permanent reservation, we now have an ABBA deadlock - writeback completion has a transaction reserved and wants the ILOCK, and someone else has the ILOCK and wants a transaction reservation. Therefore, we create a per-inode writeback io completion queue + work item. When writeback finishes, it can add the ioend to the per-inode queue and let the single worker item process that queue. This dramatically cuts down on the number of kworkers and ILOCK contention in the system, and seems to have eliminated an occasional deadlock I was seeing while running generic/476. Testing with a program that simulates a heavy random-write workload to a single file demonstrates that the number of kworkers drops from approximately 120 threads per file to 1, without dramatically changing write bandwidth or pagecache access latency. Note that we leave the xfs-conv workqueue's max_active alone because we still want to be able to run ioend processing for as many inodes as the system can handle. Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Brian Foster <bfoster@redhat.com>
2019-04-16 03:13:20 +07:00
while (!list_empty(&completion_list)) {
ioend = list_first_entry(&completion_list, struct xfs_ioend,
io_list);
list_del_init(&ioend->io_list);
xfs_ioend_try_merge(ioend, &completion_list);
xfs: implement per-inode writeback completion queues When scheduling writeback of dirty file data in the page cache, XFS uses IO completion workqueue items to ensure that filesystem metadata only updates after the write completes successfully. This is essential for converting unwritten extents to real extents at the right time and performing COW remappings. Unfortunately, XFS queues each IO completion work item to an unbounded workqueue, which means that the kernel can spawn dozens of threads to try to handle the items quickly. These threads need to take the ILOCK to update file metadata, which results in heavy ILOCK contention if a large number of the work items target a single file, which is inefficient. Worse yet, the writeback completion threads get stuck waiting for the ILOCK while holding transaction reservations, which can use up all available log reservation space. When that happens, metadata updates to other parts of the filesystem grind to a halt, even if the filesystem could otherwise have handled it. Even worse, if one of the things grinding to a halt happens to be a thread in the middle of a defer-ops finish holding the same ILOCK and trying to obtain more log reservation having exhausted the permanent reservation, we now have an ABBA deadlock - writeback completion has a transaction reserved and wants the ILOCK, and someone else has the ILOCK and wants a transaction reservation. Therefore, we create a per-inode writeback io completion queue + work item. When writeback finishes, it can add the ioend to the per-inode queue and let the single worker item process that queue. This dramatically cuts down on the number of kworkers and ILOCK contention in the system, and seems to have eliminated an occasional deadlock I was seeing while running generic/476. Testing with a program that simulates a heavy random-write workload to a single file demonstrates that the number of kworkers drops from approximately 120 threads per file to 1, without dramatically changing write bandwidth or pagecache access latency. Note that we leave the xfs-conv workqueue's max_active alone because we still want to be able to run ioend processing for as many inodes as the system can handle. Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Brian Foster <bfoster@redhat.com>
2019-04-16 03:13:20 +07:00
xfs_end_ioend(ioend);
}
}
STATIC void
xfs_end_bio(
struct bio *bio)
{
struct xfs_ioend *ioend = bio->bi_private;
xfs: implement per-inode writeback completion queues When scheduling writeback of dirty file data in the page cache, XFS uses IO completion workqueue items to ensure that filesystem metadata only updates after the write completes successfully. This is essential for converting unwritten extents to real extents at the right time and performing COW remappings. Unfortunately, XFS queues each IO completion work item to an unbounded workqueue, which means that the kernel can spawn dozens of threads to try to handle the items quickly. These threads need to take the ILOCK to update file metadata, which results in heavy ILOCK contention if a large number of the work items target a single file, which is inefficient. Worse yet, the writeback completion threads get stuck waiting for the ILOCK while holding transaction reservations, which can use up all available log reservation space. When that happens, metadata updates to other parts of the filesystem grind to a halt, even if the filesystem could otherwise have handled it. Even worse, if one of the things grinding to a halt happens to be a thread in the middle of a defer-ops finish holding the same ILOCK and trying to obtain more log reservation having exhausted the permanent reservation, we now have an ABBA deadlock - writeback completion has a transaction reserved and wants the ILOCK, and someone else has the ILOCK and wants a transaction reservation. Therefore, we create a per-inode writeback io completion queue + work item. When writeback finishes, it can add the ioend to the per-inode queue and let the single worker item process that queue. This dramatically cuts down on the number of kworkers and ILOCK contention in the system, and seems to have eliminated an occasional deadlock I was seeing while running generic/476. Testing with a program that simulates a heavy random-write workload to a single file demonstrates that the number of kworkers drops from approximately 120 threads per file to 1, without dramatically changing write bandwidth or pagecache access latency. Note that we leave the xfs-conv workqueue's max_active alone because we still want to be able to run ioend processing for as many inodes as the system can handle. Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Brian Foster <bfoster@redhat.com>
2019-04-16 03:13:20 +07:00
struct xfs_inode *ip = XFS_I(ioend->io_inode);
struct xfs_mount *mp = ip->i_mount;
unsigned long flags;
if (ioend->io_fork == XFS_COW_FORK ||
xfs: implement per-inode writeback completion queues When scheduling writeback of dirty file data in the page cache, XFS uses IO completion workqueue items to ensure that filesystem metadata only updates after the write completes successfully. This is essential for converting unwritten extents to real extents at the right time and performing COW remappings. Unfortunately, XFS queues each IO completion work item to an unbounded workqueue, which means that the kernel can spawn dozens of threads to try to handle the items quickly. These threads need to take the ILOCK to update file metadata, which results in heavy ILOCK contention if a large number of the work items target a single file, which is inefficient. Worse yet, the writeback completion threads get stuck waiting for the ILOCK while holding transaction reservations, which can use up all available log reservation space. When that happens, metadata updates to other parts of the filesystem grind to a halt, even if the filesystem could otherwise have handled it. Even worse, if one of the things grinding to a halt happens to be a thread in the middle of a defer-ops finish holding the same ILOCK and trying to obtain more log reservation having exhausted the permanent reservation, we now have an ABBA deadlock - writeback completion has a transaction reserved and wants the ILOCK, and someone else has the ILOCK and wants a transaction reservation. Therefore, we create a per-inode writeback io completion queue + work item. When writeback finishes, it can add the ioend to the per-inode queue and let the single worker item process that queue. This dramatically cuts down on the number of kworkers and ILOCK contention in the system, and seems to have eliminated an occasional deadlock I was seeing while running generic/476. Testing with a program that simulates a heavy random-write workload to a single file demonstrates that the number of kworkers drops from approximately 120 threads per file to 1, without dramatically changing write bandwidth or pagecache access latency. Note that we leave the xfs-conv workqueue's max_active alone because we still want to be able to run ioend processing for as many inodes as the system can handle. Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Brian Foster <bfoster@redhat.com>
2019-04-16 03:13:20 +07:00
ioend->io_state == XFS_EXT_UNWRITTEN ||
ioend->io_append_trans != NULL) {
spin_lock_irqsave(&ip->i_ioend_lock, flags);
if (list_empty(&ip->i_ioend_list))
WARN_ON_ONCE(!queue_work(mp->m_unwritten_workqueue,
&ip->i_ioend_work));
list_add_tail(&ioend->io_list, &ip->i_ioend_list);
spin_unlock_irqrestore(&ip->i_ioend_lock, flags);
} else
xfs_destroy_ioend(ioend, blk_status_to_errno(bio->bi_status));
}
xfs: validate writeback mapping using data fork seq counter The writeback code caches the current extent mapping across multiple xfs_do_writepage() calls to avoid repeated lookups for sequential pages backed by the same extent. This is known to be slightly racy with extent fork changes in certain difficult to reproduce scenarios. The cached extent is trimmed to within EOF to help avoid the most common vector for this problem via speculative preallocation management, but this is a band-aid that does not address the fundamental problem. Now that we have an xfs_ifork sequence counter mechanism used to facilitate COW writeback, we can use the same mechanism to validate consistency between the data fork and cached writeback mappings. On its face, this is somewhat of a big hammer approach because any change to the data fork invalidates any mapping currently cached by a writeback in progress regardless of whether the data fork change overlaps with the range under writeback. In practice, however, the impact of this approach is minimal in most cases. First, data fork changes (delayed allocations) caused by sustained sequential buffered writes are amortized across speculative preallocations. This means that a cached mapping won't be invalidated by each buffered write of a common file copy workload, but rather only on less frequent allocation events. Second, the extent tree is always entirely in-core so an additional lookup of a usable extent mostly costs a shared ilock cycle and in-memory tree lookup. This means that a cached mapping reval is relatively cheap compared to the I/O itself. Third, spurious invalidations don't impact ioend construction. This means that even if the same extent is revalidated multiple times across multiple writepage instances, we still construct and submit the same size ioend (and bio) if the blocks are physically contiguous. Update struct xfs_writepage_ctx with a new field to hold the sequence number of the data fork associated with the currently cached mapping. Check the wpc seqno against the data fork when the mapping is validated and reestablish the mapping whenever the fork has changed since the mapping was cached. This ensures that writeback always uses a valid extent mapping and thus prevents lost writebacks and stale delalloc block problems. Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Allison Henderson <allison.henderson@oracle.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>
2019-02-02 00:14:23 +07:00
/*
* Fast revalidation of the cached writeback mapping. Return true if the current
* mapping is valid, false otherwise.
*/
static bool
xfs_imap_valid(
struct xfs_writepage_ctx *wpc,
struct xfs_inode *ip,
xfs_fileoff_t offset_fsb)
{
if (offset_fsb < wpc->imap.br_startoff ||
offset_fsb >= wpc->imap.br_startoff + wpc->imap.br_blockcount)
return false;
/*
* If this is a COW mapping, it is sufficient to check that the mapping
* covers the offset. Be careful to check this first because the caller
* can revalidate a COW mapping without updating the data seqno.
*/
if (wpc->fork == XFS_COW_FORK)
xfs: validate writeback mapping using data fork seq counter The writeback code caches the current extent mapping across multiple xfs_do_writepage() calls to avoid repeated lookups for sequential pages backed by the same extent. This is known to be slightly racy with extent fork changes in certain difficult to reproduce scenarios. The cached extent is trimmed to within EOF to help avoid the most common vector for this problem via speculative preallocation management, but this is a band-aid that does not address the fundamental problem. Now that we have an xfs_ifork sequence counter mechanism used to facilitate COW writeback, we can use the same mechanism to validate consistency between the data fork and cached writeback mappings. On its face, this is somewhat of a big hammer approach because any change to the data fork invalidates any mapping currently cached by a writeback in progress regardless of whether the data fork change overlaps with the range under writeback. In practice, however, the impact of this approach is minimal in most cases. First, data fork changes (delayed allocations) caused by sustained sequential buffered writes are amortized across speculative preallocations. This means that a cached mapping won't be invalidated by each buffered write of a common file copy workload, but rather only on less frequent allocation events. Second, the extent tree is always entirely in-core so an additional lookup of a usable extent mostly costs a shared ilock cycle and in-memory tree lookup. This means that a cached mapping reval is relatively cheap compared to the I/O itself. Third, spurious invalidations don't impact ioend construction. This means that even if the same extent is revalidated multiple times across multiple writepage instances, we still construct and submit the same size ioend (and bio) if the blocks are physically contiguous. Update struct xfs_writepage_ctx with a new field to hold the sequence number of the data fork associated with the currently cached mapping. Check the wpc seqno against the data fork when the mapping is validated and reestablish the mapping whenever the fork has changed since the mapping was cached. This ensures that writeback always uses a valid extent mapping and thus prevents lost writebacks and stale delalloc block problems. Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Allison Henderson <allison.henderson@oracle.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>
2019-02-02 00:14:23 +07:00
return true;
/*
* This is not a COW mapping. Check the sequence number of the data fork
* because concurrent changes could have invalidated the extent. Check
* the COW fork because concurrent changes since the last time we
* checked (and found nothing at this offset) could have added
* overlapping blocks.
*/
if (wpc->data_seq != READ_ONCE(ip->i_df.if_seq))
return false;
if (xfs_inode_has_cow_data(ip) &&
wpc->cow_seq != READ_ONCE(ip->i_cowfp->if_seq))
return false;
return true;
}
/*
* Pass in a dellalloc extent and convert it to real extents, return the real
* extent that maps offset_fsb in wpc->imap.
*
* The current page is held locked so nothing could have removed the block
* backing offset_fsb, although it could have moved from the COW to the data
* fork by another thread.
*/
static int
xfs_convert_blocks(
struct xfs_writepage_ctx *wpc,
struct xfs_inode *ip,
xfs_fileoff_t offset_fsb)
{
int error;
/*
* Attempt to allocate whatever delalloc extent currently backs
* offset_fsb and put the result into wpc->imap. Allocate in a loop
* because it may take several attempts to allocate real blocks for a
* contiguous delalloc extent if free space is sufficiently fragmented.
*/
do {
error = xfs_bmapi_convert_delalloc(ip, wpc->fork, offset_fsb,
&wpc->imap, wpc->fork == XFS_COW_FORK ?
&wpc->cow_seq : &wpc->data_seq);
if (error)
return error;
} while (wpc->imap.br_startoff + wpc->imap.br_blockcount <= offset_fsb);
return 0;
}
STATIC int
xfs_map_blocks(
struct xfs_writepage_ctx *wpc,
struct inode *inode,
loff_t offset)
{
struct xfs_inode *ip = XFS_I(inode);
struct xfs_mount *mp = ip->i_mount;
ssize_t count = i_blocksize(inode);
xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset);
xfs_fileoff_t end_fsb = XFS_B_TO_FSB(mp, offset + count);
xfs_fileoff_t cow_fsb = NULLFILEOFF;
struct xfs_bmbt_irec imap;
struct xfs_iext_cursor icur;
int retries = 0;
int error = 0;
xfs: validate writeback mapping using data fork seq counter The writeback code caches the current extent mapping across multiple xfs_do_writepage() calls to avoid repeated lookups for sequential pages backed by the same extent. This is known to be slightly racy with extent fork changes in certain difficult to reproduce scenarios. The cached extent is trimmed to within EOF to help avoid the most common vector for this problem via speculative preallocation management, but this is a band-aid that does not address the fundamental problem. Now that we have an xfs_ifork sequence counter mechanism used to facilitate COW writeback, we can use the same mechanism to validate consistency between the data fork and cached writeback mappings. On its face, this is somewhat of a big hammer approach because any change to the data fork invalidates any mapping currently cached by a writeback in progress regardless of whether the data fork change overlaps with the range under writeback. In practice, however, the impact of this approach is minimal in most cases. First, data fork changes (delayed allocations) caused by sustained sequential buffered writes are amortized across speculative preallocations. This means that a cached mapping won't be invalidated by each buffered write of a common file copy workload, but rather only on less frequent allocation events. Second, the extent tree is always entirely in-core so an additional lookup of a usable extent mostly costs a shared ilock cycle and in-memory tree lookup. This means that a cached mapping reval is relatively cheap compared to the I/O itself. Third, spurious invalidations don't impact ioend construction. This means that even if the same extent is revalidated multiple times across multiple writepage instances, we still construct and submit the same size ioend (and bio) if the blocks are physically contiguous. Update struct xfs_writepage_ctx with a new field to hold the sequence number of the data fork associated with the currently cached mapping. Check the wpc seqno against the data fork when the mapping is validated and reestablish the mapping whenever the fork has changed since the mapping was cached. This ensures that writeback always uses a valid extent mapping and thus prevents lost writebacks and stale delalloc block problems. Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Allison Henderson <allison.henderson@oracle.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>
2019-02-02 00:14:23 +07:00
if (XFS_FORCED_SHUTDOWN(mp))
return -EIO;
/*
* COW fork blocks can overlap data fork blocks even if the blocks
* aren't shared. COW I/O always takes precedent, so we must always
* check for overlap on reflink inodes unless the mapping is already a
* COW one, or the COW fork hasn't changed from the last time we looked
* at it.
*
* It's safe to check the COW fork if_seq here without the ILOCK because
* we've indirectly protected against concurrent updates: writeback has
* the page locked, which prevents concurrent invalidations by reflink
* and directio and prevents concurrent buffered writes to the same
* page. Changes to if_seq always happen under i_lock, which protects
* against concurrent updates and provides a memory barrier on the way
* out that ensures that we always see the current value.
*/
xfs: validate writeback mapping using data fork seq counter The writeback code caches the current extent mapping across multiple xfs_do_writepage() calls to avoid repeated lookups for sequential pages backed by the same extent. This is known to be slightly racy with extent fork changes in certain difficult to reproduce scenarios. The cached extent is trimmed to within EOF to help avoid the most common vector for this problem via speculative preallocation management, but this is a band-aid that does not address the fundamental problem. Now that we have an xfs_ifork sequence counter mechanism used to facilitate COW writeback, we can use the same mechanism to validate consistency between the data fork and cached writeback mappings. On its face, this is somewhat of a big hammer approach because any change to the data fork invalidates any mapping currently cached by a writeback in progress regardless of whether the data fork change overlaps with the range under writeback. In practice, however, the impact of this approach is minimal in most cases. First, data fork changes (delayed allocations) caused by sustained sequential buffered writes are amortized across speculative preallocations. This means that a cached mapping won't be invalidated by each buffered write of a common file copy workload, but rather only on less frequent allocation events. Second, the extent tree is always entirely in-core so an additional lookup of a usable extent mostly costs a shared ilock cycle and in-memory tree lookup. This means that a cached mapping reval is relatively cheap compared to the I/O itself. Third, spurious invalidations don't impact ioend construction. This means that even if the same extent is revalidated multiple times across multiple writepage instances, we still construct and submit the same size ioend (and bio) if the blocks are physically contiguous. Update struct xfs_writepage_ctx with a new field to hold the sequence number of the data fork associated with the currently cached mapping. Check the wpc seqno against the data fork when the mapping is validated and reestablish the mapping whenever the fork has changed since the mapping was cached. This ensures that writeback always uses a valid extent mapping and thus prevents lost writebacks and stale delalloc block problems. Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Allison Henderson <allison.henderson@oracle.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>
2019-02-02 00:14:23 +07:00
if (xfs_imap_valid(wpc, ip, offset_fsb))
return 0;
/*
* If we don't have a valid map, now it's time to get a new one for this
* offset. This will convert delayed allocations (including COW ones)
* into real extents. If we return without a valid map, it means we
* landed in a hole and we skip the block.
*/
retry:
xfs_ilock(ip, XFS_ILOCK_SHARED);
ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
(ip->i_df.if_flags & XFS_IFEXTENTS));
/*
* Check if this is offset is covered by a COW extents, and if yes use
* it directly instead of looking up anything in the data fork.
*/
if (xfs_inode_has_cow_data(ip) &&
xfs_iext_lookup_extent(ip, ip->i_cowfp, offset_fsb, &icur, &imap))
cow_fsb = imap.br_startoff;
if (cow_fsb != NULLFILEOFF && cow_fsb <= offset_fsb) {
wpc->cow_seq = READ_ONCE(ip->i_cowfp->if_seq);
xfs_iunlock(ip, XFS_ILOCK_SHARED);
wpc->fork = XFS_COW_FORK;
goto allocate_blocks;
}
/*
xfs: validate writeback mapping using data fork seq counter The writeback code caches the current extent mapping across multiple xfs_do_writepage() calls to avoid repeated lookups for sequential pages backed by the same extent. This is known to be slightly racy with extent fork changes in certain difficult to reproduce scenarios. The cached extent is trimmed to within EOF to help avoid the most common vector for this problem via speculative preallocation management, but this is a band-aid that does not address the fundamental problem. Now that we have an xfs_ifork sequence counter mechanism used to facilitate COW writeback, we can use the same mechanism to validate consistency between the data fork and cached writeback mappings. On its face, this is somewhat of a big hammer approach because any change to the data fork invalidates any mapping currently cached by a writeback in progress regardless of whether the data fork change overlaps with the range under writeback. In practice, however, the impact of this approach is minimal in most cases. First, data fork changes (delayed allocations) caused by sustained sequential buffered writes are amortized across speculative preallocations. This means that a cached mapping won't be invalidated by each buffered write of a common file copy workload, but rather only on less frequent allocation events. Second, the extent tree is always entirely in-core so an additional lookup of a usable extent mostly costs a shared ilock cycle and in-memory tree lookup. This means that a cached mapping reval is relatively cheap compared to the I/O itself. Third, spurious invalidations don't impact ioend construction. This means that even if the same extent is revalidated multiple times across multiple writepage instances, we still construct and submit the same size ioend (and bio) if the blocks are physically contiguous. Update struct xfs_writepage_ctx with a new field to hold the sequence number of the data fork associated with the currently cached mapping. Check the wpc seqno against the data fork when the mapping is validated and reestablish the mapping whenever the fork has changed since the mapping was cached. This ensures that writeback always uses a valid extent mapping and thus prevents lost writebacks and stale delalloc block problems. Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Allison Henderson <allison.henderson@oracle.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>
2019-02-02 00:14:23 +07:00
* No COW extent overlap. Revalidate now that we may have updated
* ->cow_seq. If the data mapping is still valid, we're done.
*/
xfs: validate writeback mapping using data fork seq counter The writeback code caches the current extent mapping across multiple xfs_do_writepage() calls to avoid repeated lookups for sequential pages backed by the same extent. This is known to be slightly racy with extent fork changes in certain difficult to reproduce scenarios. The cached extent is trimmed to within EOF to help avoid the most common vector for this problem via speculative preallocation management, but this is a band-aid that does not address the fundamental problem. Now that we have an xfs_ifork sequence counter mechanism used to facilitate COW writeback, we can use the same mechanism to validate consistency between the data fork and cached writeback mappings. On its face, this is somewhat of a big hammer approach because any change to the data fork invalidates any mapping currently cached by a writeback in progress regardless of whether the data fork change overlaps with the range under writeback. In practice, however, the impact of this approach is minimal in most cases. First, data fork changes (delayed allocations) caused by sustained sequential buffered writes are amortized across speculative preallocations. This means that a cached mapping won't be invalidated by each buffered write of a common file copy workload, but rather only on less frequent allocation events. Second, the extent tree is always entirely in-core so an additional lookup of a usable extent mostly costs a shared ilock cycle and in-memory tree lookup. This means that a cached mapping reval is relatively cheap compared to the I/O itself. Third, spurious invalidations don't impact ioend construction. This means that even if the same extent is revalidated multiple times across multiple writepage instances, we still construct and submit the same size ioend (and bio) if the blocks are physically contiguous. Update struct xfs_writepage_ctx with a new field to hold the sequence number of the data fork associated with the currently cached mapping. Check the wpc seqno against the data fork when the mapping is validated and reestablish the mapping whenever the fork has changed since the mapping was cached. This ensures that writeback always uses a valid extent mapping and thus prevents lost writebacks and stale delalloc block problems. Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Allison Henderson <allison.henderson@oracle.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>
2019-02-02 00:14:23 +07:00
if (xfs_imap_valid(wpc, ip, offset_fsb)) {
xfs_iunlock(ip, XFS_ILOCK_SHARED);
return 0;
}
/*
* If we don't have a valid map, now it's time to get a new one for this
* offset. This will convert delayed allocations (including COW ones)
* into real extents.
*/
if (!xfs_iext_lookup_extent(ip, &ip->i_df, offset_fsb, &icur, &imap))
imap.br_startoff = end_fsb; /* fake a hole past EOF */
xfs: validate writeback mapping using data fork seq counter The writeback code caches the current extent mapping across multiple xfs_do_writepage() calls to avoid repeated lookups for sequential pages backed by the same extent. This is known to be slightly racy with extent fork changes in certain difficult to reproduce scenarios. The cached extent is trimmed to within EOF to help avoid the most common vector for this problem via speculative preallocation management, but this is a band-aid that does not address the fundamental problem. Now that we have an xfs_ifork sequence counter mechanism used to facilitate COW writeback, we can use the same mechanism to validate consistency between the data fork and cached writeback mappings. On its face, this is somewhat of a big hammer approach because any change to the data fork invalidates any mapping currently cached by a writeback in progress regardless of whether the data fork change overlaps with the range under writeback. In practice, however, the impact of this approach is minimal in most cases. First, data fork changes (delayed allocations) caused by sustained sequential buffered writes are amortized across speculative preallocations. This means that a cached mapping won't be invalidated by each buffered write of a common file copy workload, but rather only on less frequent allocation events. Second, the extent tree is always entirely in-core so an additional lookup of a usable extent mostly costs a shared ilock cycle and in-memory tree lookup. This means that a cached mapping reval is relatively cheap compared to the I/O itself. Third, spurious invalidations don't impact ioend construction. This means that even if the same extent is revalidated multiple times across multiple writepage instances, we still construct and submit the same size ioend (and bio) if the blocks are physically contiguous. Update struct xfs_writepage_ctx with a new field to hold the sequence number of the data fork associated with the currently cached mapping. Check the wpc seqno against the data fork when the mapping is validated and reestablish the mapping whenever the fork has changed since the mapping was cached. This ensures that writeback always uses a valid extent mapping and thus prevents lost writebacks and stale delalloc block problems. Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Allison Henderson <allison.henderson@oracle.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>
2019-02-02 00:14:23 +07:00
wpc->data_seq = READ_ONCE(ip->i_df.if_seq);
xfs_iunlock(ip, XFS_ILOCK_SHARED);
wpc->fork = XFS_DATA_FORK;
/* landed in a hole or beyond EOF? */
if (imap.br_startoff > offset_fsb) {
imap.br_blockcount = imap.br_startoff - offset_fsb;
imap.br_startoff = offset_fsb;
imap.br_startblock = HOLESTARTBLOCK;
imap.br_state = XFS_EXT_NORM;
}
xfs: make xfs_writepage_map extent map centric xfs_writepage_map() iterates over the bufferheads on a page to decide what sort of IO to do and what actions to take. However, when it comes to reflink and deciding when it needs to execute a COW operation, we no longer look at the bufferhead state but instead we ignore than and look up internal state held in the COW fork extent list. This means xfs_writepage_map() is somewhat confused. It does stuff, then ignores it, then tries to handle the impedence mismatch by shovelling the results inside the existing mapping code. It works, but it's a bit of a mess and it makes it hard to fix the cached map bug that the writepage code currently has. To unify the two different mechanisms, we first have to choose a direction. That's already been set - we're de-emphasising bufferheads so they are no longer a control structure as we need to do taht to allow for eventual removal. Hence we need to move away from looking at bufferhead state to determine what operations we need to perform. We can't completely get rid of bufferheads yet - they do contain some state that is absolutely necessary, such as whether that part of the page contains valid data or not (buffer_uptodate()). Other state in the bufferhead is redundant: BH_dirty - the page is dirty, so we can ignore this and just write it BH_delay - we have delalloc extent info in the DATA fork extent tree BH_unwritten - same as BH_delay BH_mapped - indicates we've already used it once for IO and it is mapped to a disk address. Needs to be ignored for COW blocks. The BH_mapped flag is an interesting case - it's supposed to indicate that it's already mapped to disk and so we can just use it "as is". In theory, we don't even have to do an extent lookup to find where to write it too, but we have to do that anyway to determine we are actually writing over a valid extent. Hence it's not even serving the purpose of avoiding a an extent lookup during writeback, and so we can pretty much ignore it. Especially as we have to ignore it for COW operations... Therefore, use the extent map as the source of information to tell us what actions we need to take and what sort of IO we should perform. The first step is to have xfs_map_blocks() set the io type according to what it looks up. This means it can easily handle both normal overwrite and COW cases. The only thing we also need to add is the ability to return hole mappings. We need to return and cache hole mappings now for the case of multiple blocks per page. We no longer use the BH_mapped to indicate a block over a hole, so we have to get that info from xfs_map_blocks(). We cache it so that holes that span two pages don't need separate lookups. This allows us to avoid ever doing write IO over a hole, too. Now that we have xfs_map_blocks() returning both a cached map and the type of IO we need to perform, we can rewrite xfs_writepage_map() to drop all the bufferhead control. It's also much simplified because it doesn't need to explicitly handle COW operations. Instead of iterating bufferheads, it iterates blocks within the page and then looks up what per-block state is required from the appropriate bufferhead. It then validates the cached map, and if it's not valid, we get a new map. If we don't get a valid map or it's over a hole, we skip the block. At this point, we have to remap the bufferhead via xfs_map_at_offset(). As previously noted, we had to do this even if the buffer was already mapped as the mapping would be stale for XFS_IO_DELALLOC, XFS_IO_UNWRITTEN and XFS_IO_COW IO types. With xfs_map_blocks() now controlling the type, even XFS_IO_OVERWRITE types need remapping, as converted-but-not-yet- written delalloc extents beyond EOF can be reported at XFS_IO_OVERWRITE. Bufferheads that span such regions still need their BH_Delay flags cleared and their block numbers calculated, so we now unconditionally map each bufferhead before submission. But wait! There's more - remember the old "treat unwritten extents as holes on read" hack? Yeah, that means we can have a dirty page with unmapped, unwritten bufferheads that contain data! What makes these so special is that the unwritten "hole" bufferheads do not have a valid block device pointer, so if we attempt to write them xfs_add_to_ioend() blows up. So we make xfs_map_at_offset() do the "realtime or data device" lookup from the inode and ignore what was or wasn't put into the bufferhead when the buffer was instantiated. The astute reader will have realised by now that this code treats unwritten extents in multiple-blocks-per-page situations differently. If we get any combination of unwritten blocks on a dirty page that contain valid data in the page, we're going to convert them to real extents. This can actually be a win, because it means that pages with interleaving unwritten and written blocks will get converted to a single written extent with zeros replacing the interspersed unwritten blocks. This is actually good for reducing extent list and conversion overhead, and it means we issue a contiguous IO instead of lots of little ones. The downside is that we use up a little extra IO bandwidth. Neither of these seem like a bad thing given that spinning disks are seek sensitive, and SSDs/pmem have bandwidth to burn and the lower Io latency/CPU overhead of fewer, larger IOs will result in better performance on them... As a result of all this, the only state we actually care about from the bufferhead is a single flag - BH_Uptodate. We still use the bufferhead to pass some information to the bio via xfs_add_to_ioend(), but that is trivial to separate and pass explicitly. This means we really only need 1 bit of state per block per page from the buffered write path in the writeback path. Everything else we do with the bufferhead is purely to make the buffered IO front end continue to work correctly. i.e we've pretty much marginalised bufferheads in the writeback path completely. Signed-off-By: Dave Chinner <dchinner@redhat.com> [hch: forward port, refactor and split off bits into other commits] Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2018-07-12 12:26:00 +07:00
/*
* Truncate to the next COW extent if there is one. This is the only
* opportunity to do this because we can skip COW fork lookups for the
* subsequent blocks in the mapping; however, the requirement to treat
* the COW range separately remains.
*/
if (cow_fsb != NULLFILEOFF &&
cow_fsb < imap.br_startoff + imap.br_blockcount)
imap.br_blockcount = cow_fsb - imap.br_startoff;
/* got a delalloc extent? */
if (imap.br_startblock != HOLESTARTBLOCK &&
isnullstartblock(imap.br_startblock))
goto allocate_blocks;
wpc->imap = imap;
trace_xfs_map_blocks_found(ip, offset, count, wpc->fork, &imap);
return 0;
allocate_blocks:
error = xfs_convert_blocks(wpc, ip, offset_fsb);
if (error) {
/*
* If we failed to find the extent in the COW fork we might have
* raced with a COW to data fork conversion or truncate.
* Restart the lookup to catch the extent in the data fork for
* the former case, but prevent additional retries to avoid
* looping forever for the latter case.
*/
if (error == -EAGAIN && wpc->fork == XFS_COW_FORK && !retries++)
goto retry;
ASSERT(error != -EAGAIN);
return error;
}
/*
* Due to merging the return real extent might be larger than the
* original delalloc one. Trim the return extent to the next COW
* boundary again to force a re-lookup.
*/
if (wpc->fork != XFS_COW_FORK && cow_fsb != NULLFILEOFF &&
cow_fsb < wpc->imap.br_startoff + wpc->imap.br_blockcount)
wpc->imap.br_blockcount = cow_fsb - wpc->imap.br_startoff;
ASSERT(wpc->imap.br_startoff <= offset_fsb);
ASSERT(wpc->imap.br_startoff + wpc->imap.br_blockcount > offset_fsb);
trace_xfs_map_blocks_alloc(ip, offset, count, wpc->fork, &imap);
return 0;
}
/*
* Submit the bio for an ioend. We are passed an ioend with a bio attached to
* it, and we submit that bio. The ioend may be used for multiple bio
* submissions, so we only want to allocate an append transaction for the ioend
* once. In the case of multiple bio submission, each bio will take an IO
* reference to the ioend to ensure that the ioend completion is only done once
* all bios have been submitted and the ioend is really done.
xfs: fix broken error handling in xfs_vm_writepage When we shut down the filesystem, it might first be detected in writeback when we are allocating a inode size transaction. This happens after we have moved all the pages into the writeback state and unlocked them. Unfortunately, if we fail to set up the transaction we then abort writeback and try to invalidate the current page. This then triggers are BUG() in block_invalidatepage() because we are trying to invalidate an unlocked page. Fixing this is a bit of a chicken and egg problem - we can't allocate the transaction until we've clustered all the pages into the IO and we know the size of it (i.e. whether the last block of the IO is beyond the current EOF or not). However, we don't want to hold pages locked for long periods of time, especially while we lock other pages to cluster them into the write. To fix this, we need to make a clear delineation in writeback where errors can only be handled by IO completion processing. That is, once we have marked a page for writeback and unlocked it, we have to report errors via IO completion because we've already started the IO. We may not have submitted any IO, but we've changed the page state to indicate that it is under IO so we must now use the IO completion path to report errors. To do this, add an error field to xfs_submit_ioend() to pass it the error that occurred during the building on the ioend chain. When this is non-zero, mark each ioend with the error and call xfs_finish_ioend() directly rather than building bios. This will immediately push the ioends through completion processing with the error that has occurred. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Mark Tinguely <tinguely@sgi.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2012-11-12 18:09:45 +07:00
*
* If @fail is non-zero, it means that we have a situation where some part of
* the submission process has failed after we have marked paged for writeback
* and unlocked them. In this situation, we need to fail the bio and ioend
* rather than submit it to IO. This typically only happens on a filesystem
* shutdown.
*/
xfs: don't chain ioends during writepage submission Currently we can build a long ioend chain during ->writepages that gets attached to the writepage context. IO submission only then occurs when we finish all the writepage processing. This means we can have many ioends allocated and pending, and this violates the mempool guarantees that we need to give about forwards progress. i.e. we really should only have one ioend being built at a time, otherwise we may drain the mempool trying to allocate a new ioend and that blocks submission, completion and freeing of ioends that are already in progress. To prevent this situation from happening, we need to submit ioends for IO as soon as they are ready for dispatch rather than queuing them for later submission. This means the ioends have bios built immediately and they get queued on any plug that is current active. Hence if we schedule away from writeback, the ioends that have been built will make forwards progress due to the plug flushing on context switch. This will also prevent context switches from creating unnecessary IO submission latency. We can't completely avoid having nested IO allocation - when we have a block size smaller than a page size, we still need to hold the ioend submission until after we have marked the current page dirty. Hence we may need multiple ioends to be held while the current page is completely mapped and made ready for IO dispatch. We cannot avoid this problem - the current code already has this ioend chaining within a page so we can mostly ignore that it occurs. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-02-15 13:23:12 +07:00
STATIC int
xfs_submit_ioend(
struct writeback_control *wbc,
struct xfs_ioend *ioend,
xfs: don't chain ioends during writepage submission Currently we can build a long ioend chain during ->writepages that gets attached to the writepage context. IO submission only then occurs when we finish all the writepage processing. This means we can have many ioends allocated and pending, and this violates the mempool guarantees that we need to give about forwards progress. i.e. we really should only have one ioend being built at a time, otherwise we may drain the mempool trying to allocate a new ioend and that blocks submission, completion and freeing of ioends that are already in progress. To prevent this situation from happening, we need to submit ioends for IO as soon as they are ready for dispatch rather than queuing them for later submission. This means the ioends have bios built immediately and they get queued on any plug that is current active. Hence if we schedule away from writeback, the ioends that have been built will make forwards progress due to the plug flushing on context switch. This will also prevent context switches from creating unnecessary IO submission latency. We can't completely avoid having nested IO allocation - when we have a block size smaller than a page size, we still need to hold the ioend submission until after we have marked the current page dirty. Hence we may need multiple ioends to be held while the current page is completely mapped and made ready for IO dispatch. We cannot avoid this problem - the current code already has this ioend chaining within a page so we can mostly ignore that it occurs. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-02-15 13:23:12 +07:00
int status)
{
xfs: mark speculative prealloc CoW fork extents unwritten Christoph Hellwig pointed out that there's a potentially nasty race when performing simultaneous nearby directio cow writes: "Thread 1 writes a range from B to c " B --------- C p "a little later thread 2 writes from A to B " A --------- B p [editor's note: the 'p' denote cowextsize boundaries, which I added to make this more clear] "but the code preallocates beyond B into the range where thread "1 has just written, but ->end_io hasn't been called yet. "But once ->end_io is called thread 2 has already allocated "up to the extent size hint into the write range of thread 1, "so the end_io handler will splice the unintialized blocks from "that preallocation back into the file right after B." We can avoid this race by ensuring that thread 1 cannot accidentally remap the blocks that thread 2 allocated (as part of speculative preallocation) as part of t2's write preparation in t1's end_io handler. The way we make this happen is by taking advantage of the unwritten extent flag as an intermediate step. Recall that when we begin the process of writing data to shared blocks, we create a delayed allocation extent in the CoW fork: D: --RRRRRRSSSRRRRRRRR--- C: ------DDDDDDD--------- When a thread prepares to CoW some dirty data out to disk, it will now convert the delalloc reservation into an /unwritten/ allocated extent in the cow fork. The da conversion code tries to opportunistically allocate as much of a (speculatively prealloc'd) extent as possible, so we may end up allocating a larger extent than we're actually writing out: D: --RRRRRRSSSRRRRRRRR--- U: ------UUUUUUU--------- Next, we convert only the part of the extent that we're actively planning to write to normal (i.e. not unwritten) status: D: --RRRRRRSSSRRRRRRRR--- U: ------UURRUUU--------- If the write succeeds, the end_cow function will now scan the relevant range of the CoW fork for real extents and remap only the real extents into the data fork: D: --RRRRRRRRSRRRRRRRR--- U: ------UU--UUU--------- This ensures that we never obliterate valid data fork extents with unwritten blocks from the CoW fork. Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Christoph Hellwig <hch@lst.de>
2017-02-03 06:14:02 +07:00
/* Convert CoW extents to regular */
if (!status && ioend->io_fork == XFS_COW_FORK) {
/*
* Yuk. This can do memory allocation, but is not a
* transactional operation so everything is done in GFP_KERNEL
* context. That can deadlock, because we hold pages in
* writeback state and GFP_KERNEL allocations can block on them.
* Hence we must operate in nofs conditions here.
*/
unsigned nofs_flag;
nofs_flag = memalloc_nofs_save();
xfs: mark speculative prealloc CoW fork extents unwritten Christoph Hellwig pointed out that there's a potentially nasty race when performing simultaneous nearby directio cow writes: "Thread 1 writes a range from B to c " B --------- C p "a little later thread 2 writes from A to B " A --------- B p [editor's note: the 'p' denote cowextsize boundaries, which I added to make this more clear] "but the code preallocates beyond B into the range where thread "1 has just written, but ->end_io hasn't been called yet. "But once ->end_io is called thread 2 has already allocated "up to the extent size hint into the write range of thread 1, "so the end_io handler will splice the unintialized blocks from "that preallocation back into the file right after B." We can avoid this race by ensuring that thread 1 cannot accidentally remap the blocks that thread 2 allocated (as part of speculative preallocation) as part of t2's write preparation in t1's end_io handler. The way we make this happen is by taking advantage of the unwritten extent flag as an intermediate step. Recall that when we begin the process of writing data to shared blocks, we create a delayed allocation extent in the CoW fork: D: --RRRRRRSSSRRRRRRRR--- C: ------DDDDDDD--------- When a thread prepares to CoW some dirty data out to disk, it will now convert the delalloc reservation into an /unwritten/ allocated extent in the cow fork. The da conversion code tries to opportunistically allocate as much of a (speculatively prealloc'd) extent as possible, so we may end up allocating a larger extent than we're actually writing out: D: --RRRRRRSSSRRRRRRRR--- U: ------UUUUUUU--------- Next, we convert only the part of the extent that we're actively planning to write to normal (i.e. not unwritten) status: D: --RRRRRRSSSRRRRRRRR--- U: ------UURRUUU--------- If the write succeeds, the end_cow function will now scan the relevant range of the CoW fork for real extents and remap only the real extents into the data fork: D: --RRRRRRRRSRRRRRRRR--- U: ------UU--UUU--------- This ensures that we never obliterate valid data fork extents with unwritten blocks from the CoW fork. Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Christoph Hellwig <hch@lst.de>
2017-02-03 06:14:02 +07:00
status = xfs_reflink_convert_cow(XFS_I(ioend->io_inode),
ioend->io_offset, ioend->io_size);
memalloc_nofs_restore(nofs_flag);
xfs: mark speculative prealloc CoW fork extents unwritten Christoph Hellwig pointed out that there's a potentially nasty race when performing simultaneous nearby directio cow writes: "Thread 1 writes a range from B to c " B --------- C p "a little later thread 2 writes from A to B " A --------- B p [editor's note: the 'p' denote cowextsize boundaries, which I added to make this more clear] "but the code preallocates beyond B into the range where thread "1 has just written, but ->end_io hasn't been called yet. "But once ->end_io is called thread 2 has already allocated "up to the extent size hint into the write range of thread 1, "so the end_io handler will splice the unintialized blocks from "that preallocation back into the file right after B." We can avoid this race by ensuring that thread 1 cannot accidentally remap the blocks that thread 2 allocated (as part of speculative preallocation) as part of t2's write preparation in t1's end_io handler. The way we make this happen is by taking advantage of the unwritten extent flag as an intermediate step. Recall that when we begin the process of writing data to shared blocks, we create a delayed allocation extent in the CoW fork: D: --RRRRRRSSSRRRRRRRR--- C: ------DDDDDDD--------- When a thread prepares to CoW some dirty data out to disk, it will now convert the delalloc reservation into an /unwritten/ allocated extent in the cow fork. The da conversion code tries to opportunistically allocate as much of a (speculatively prealloc'd) extent as possible, so we may end up allocating a larger extent than we're actually writing out: D: --RRRRRRSSSRRRRRRRR--- U: ------UUUUUUU--------- Next, we convert only the part of the extent that we're actively planning to write to normal (i.e. not unwritten) status: D: --RRRRRRSSSRRRRRRRR--- U: ------UURRUUU--------- If the write succeeds, the end_cow function will now scan the relevant range of the CoW fork for real extents and remap only the real extents into the data fork: D: --RRRRRRRRSRRRRRRRR--- U: ------UU--UUU--------- This ensures that we never obliterate valid data fork extents with unwritten blocks from the CoW fork. Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Christoph Hellwig <hch@lst.de>
2017-02-03 06:14:02 +07:00
}
xfs: don't chain ioends during writepage submission Currently we can build a long ioend chain during ->writepages that gets attached to the writepage context. IO submission only then occurs when we finish all the writepage processing. This means we can have many ioends allocated and pending, and this violates the mempool guarantees that we need to give about forwards progress. i.e. we really should only have one ioend being built at a time, otherwise we may drain the mempool trying to allocate a new ioend and that blocks submission, completion and freeing of ioends that are already in progress. To prevent this situation from happening, we need to submit ioends for IO as soon as they are ready for dispatch rather than queuing them for later submission. This means the ioends have bios built immediately and they get queued on any plug that is current active. Hence if we schedule away from writeback, the ioends that have been built will make forwards progress due to the plug flushing on context switch. This will also prevent context switches from creating unnecessary IO submission latency. We can't completely avoid having nested IO allocation - when we have a block size smaller than a page size, we still need to hold the ioend submission until after we have marked the current page dirty. Hence we may need multiple ioends to be held while the current page is completely mapped and made ready for IO dispatch. We cannot avoid this problem - the current code already has this ioend chaining within a page so we can mostly ignore that it occurs. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-02-15 13:23:12 +07:00
/* Reserve log space if we might write beyond the on-disk inode size. */
if (!status &&
(ioend->io_fork == XFS_COW_FORK ||
ioend->io_state != XFS_EXT_UNWRITTEN) &&
xfs_ioend_is_append(ioend) &&
!ioend->io_append_trans)
xfs: don't chain ioends during writepage submission Currently we can build a long ioend chain during ->writepages that gets attached to the writepage context. IO submission only then occurs when we finish all the writepage processing. This means we can have many ioends allocated and pending, and this violates the mempool guarantees that we need to give about forwards progress. i.e. we really should only have one ioend being built at a time, otherwise we may drain the mempool trying to allocate a new ioend and that blocks submission, completion and freeing of ioends that are already in progress. To prevent this situation from happening, we need to submit ioends for IO as soon as they are ready for dispatch rather than queuing them for later submission. This means the ioends have bios built immediately and they get queued on any plug that is current active. Hence if we schedule away from writeback, the ioends that have been built will make forwards progress due to the plug flushing on context switch. This will also prevent context switches from creating unnecessary IO submission latency. We can't completely avoid having nested IO allocation - when we have a block size smaller than a page size, we still need to hold the ioend submission until after we have marked the current page dirty. Hence we may need multiple ioends to be held while the current page is completely mapped and made ready for IO dispatch. We cannot avoid this problem - the current code already has this ioend chaining within a page so we can mostly ignore that it occurs. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-02-15 13:23:12 +07:00
status = xfs_setfilesize_trans_alloc(ioend);
ioend->io_bio->bi_private = ioend;
ioend->io_bio->bi_end_io = xfs_end_bio;
ioend->io_bio->bi_opf = REQ_OP_WRITE | wbc_to_write_flags(wbc);
xfs: don't chain ioends during writepage submission Currently we can build a long ioend chain during ->writepages that gets attached to the writepage context. IO submission only then occurs when we finish all the writepage processing. This means we can have many ioends allocated and pending, and this violates the mempool guarantees that we need to give about forwards progress. i.e. we really should only have one ioend being built at a time, otherwise we may drain the mempool trying to allocate a new ioend and that blocks submission, completion and freeing of ioends that are already in progress. To prevent this situation from happening, we need to submit ioends for IO as soon as they are ready for dispatch rather than queuing them for later submission. This means the ioends have bios built immediately and they get queued on any plug that is current active. Hence if we schedule away from writeback, the ioends that have been built will make forwards progress due to the plug flushing on context switch. This will also prevent context switches from creating unnecessary IO submission latency. We can't completely avoid having nested IO allocation - when we have a block size smaller than a page size, we still need to hold the ioend submission until after we have marked the current page dirty. Hence we may need multiple ioends to be held while the current page is completely mapped and made ready for IO dispatch. We cannot avoid this problem - the current code already has this ioend chaining within a page so we can mostly ignore that it occurs. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-02-15 13:23:12 +07:00
/*
* If we are failing the IO now, just mark the ioend with an
* error and finish it. This will run IO completion immediately
* as there is only one reference to the ioend at this point in
* time.
*/
if (status) {
ioend->io_bio->bi_status = errno_to_blk_status(status);
bio_endio(ioend->io_bio);
xfs: don't chain ioends during writepage submission Currently we can build a long ioend chain during ->writepages that gets attached to the writepage context. IO submission only then occurs when we finish all the writepage processing. This means we can have many ioends allocated and pending, and this violates the mempool guarantees that we need to give about forwards progress. i.e. we really should only have one ioend being built at a time, otherwise we may drain the mempool trying to allocate a new ioend and that blocks submission, completion and freeing of ioends that are already in progress. To prevent this situation from happening, we need to submit ioends for IO as soon as they are ready for dispatch rather than queuing them for later submission. This means the ioends have bios built immediately and they get queued on any plug that is current active. Hence if we schedule away from writeback, the ioends that have been built will make forwards progress due to the plug flushing on context switch. This will also prevent context switches from creating unnecessary IO submission latency. We can't completely avoid having nested IO allocation - when we have a block size smaller than a page size, we still need to hold the ioend submission until after we have marked the current page dirty. Hence we may need multiple ioends to be held while the current page is completely mapped and made ready for IO dispatch. We cannot avoid this problem - the current code already has this ioend chaining within a page so we can mostly ignore that it occurs. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-02-15 13:23:12 +07:00
return status;
}
ioend->io_bio->bi_write_hint = ioend->io_inode->i_write_hint;
submit_bio(ioend->io_bio);
xfs: don't chain ioends during writepage submission Currently we can build a long ioend chain during ->writepages that gets attached to the writepage context. IO submission only then occurs when we finish all the writepage processing. This means we can have many ioends allocated and pending, and this violates the mempool guarantees that we need to give about forwards progress. i.e. we really should only have one ioend being built at a time, otherwise we may drain the mempool trying to allocate a new ioend and that blocks submission, completion and freeing of ioends that are already in progress. To prevent this situation from happening, we need to submit ioends for IO as soon as they are ready for dispatch rather than queuing them for later submission. This means the ioends have bios built immediately and they get queued on any plug that is current active. Hence if we schedule away from writeback, the ioends that have been built will make forwards progress due to the plug flushing on context switch. This will also prevent context switches from creating unnecessary IO submission latency. We can't completely avoid having nested IO allocation - when we have a block size smaller than a page size, we still need to hold the ioend submission until after we have marked the current page dirty. Hence we may need multiple ioends to be held while the current page is completely mapped and made ready for IO dispatch. We cannot avoid this problem - the current code already has this ioend chaining within a page so we can mostly ignore that it occurs. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-02-15 13:23:12 +07:00
return 0;
}
static struct xfs_ioend *
xfs_alloc_ioend(
struct inode *inode,
int fork,
xfs_exntst_t state,
xfs_off_t offset,
struct block_device *bdev,
sector_t sector)
{
struct xfs_ioend *ioend;
struct bio *bio;
bio = bio_alloc_bioset(GFP_NOFS, BIO_MAX_PAGES, &xfs_ioend_bioset);
bio_set_dev(bio, bdev);
bio->bi_iter.bi_sector = sector;
ioend = container_of(bio, struct xfs_ioend, io_inline_bio);
INIT_LIST_HEAD(&ioend->io_list);
ioend->io_fork = fork;
ioend->io_state = state;
ioend->io_inode = inode;
ioend->io_size = 0;
ioend->io_offset = offset;
ioend->io_append_trans = NULL;
ioend->io_bio = bio;
return ioend;
}
/*
* Allocate a new bio, and chain the old bio to the new one.
*
* Note that we have to do perform the chaining in this unintuitive order
* so that the bi_private linkage is set up in the right direction for the
* traversal in xfs_destroy_ioend().
*/
static void
xfs_chain_bio(
struct xfs_ioend *ioend,
struct writeback_control *wbc,
struct block_device *bdev,
sector_t sector)
{
struct bio *new;
new = bio_alloc(GFP_NOFS, BIO_MAX_PAGES);
bio_set_dev(new, bdev);
new->bi_iter.bi_sector = sector;
bio_chain(ioend->io_bio, new);
bio_get(ioend->io_bio); /* for xfs_destroy_ioend */
ioend->io_bio->bi_opf = REQ_OP_WRITE | wbc_to_write_flags(wbc);
ioend->io_bio->bi_write_hint = ioend->io_inode->i_write_hint;
submit_bio(ioend->io_bio);
ioend->io_bio = new;
}
/*
* Test to see if we have an existing ioend structure that we could append to
* first, otherwise finish off the current ioend and start another.
*/
STATIC void
xfs_add_to_ioend(
struct inode *inode,
xfs_off_t offset,
struct page *page,
struct iomap_page *iop,
xfs: don't chain ioends during writepage submission Currently we can build a long ioend chain during ->writepages that gets attached to the writepage context. IO submission only then occurs when we finish all the writepage processing. This means we can have many ioends allocated and pending, and this violates the mempool guarantees that we need to give about forwards progress. i.e. we really should only have one ioend being built at a time, otherwise we may drain the mempool trying to allocate a new ioend and that blocks submission, completion and freeing of ioends that are already in progress. To prevent this situation from happening, we need to submit ioends for IO as soon as they are ready for dispatch rather than queuing them for later submission. This means the ioends have bios built immediately and they get queued on any plug that is current active. Hence if we schedule away from writeback, the ioends that have been built will make forwards progress due to the plug flushing on context switch. This will also prevent context switches from creating unnecessary IO submission latency. We can't completely avoid having nested IO allocation - when we have a block size smaller than a page size, we still need to hold the ioend submission until after we have marked the current page dirty. Hence we may need multiple ioends to be held while the current page is completely mapped and made ready for IO dispatch. We cannot avoid this problem - the current code already has this ioend chaining within a page so we can mostly ignore that it occurs. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-02-15 13:23:12 +07:00
struct xfs_writepage_ctx *wpc,
struct writeback_control *wbc,
xfs: don't chain ioends during writepage submission Currently we can build a long ioend chain during ->writepages that gets attached to the writepage context. IO submission only then occurs when we finish all the writepage processing. This means we can have many ioends allocated and pending, and this violates the mempool guarantees that we need to give about forwards progress. i.e. we really should only have one ioend being built at a time, otherwise we may drain the mempool trying to allocate a new ioend and that blocks submission, completion and freeing of ioends that are already in progress. To prevent this situation from happening, we need to submit ioends for IO as soon as they are ready for dispatch rather than queuing them for later submission. This means the ioends have bios built immediately and they get queued on any plug that is current active. Hence if we schedule away from writeback, the ioends that have been built will make forwards progress due to the plug flushing on context switch. This will also prevent context switches from creating unnecessary IO submission latency. We can't completely avoid having nested IO allocation - when we have a block size smaller than a page size, we still need to hold the ioend submission until after we have marked the current page dirty. Hence we may need multiple ioends to be held while the current page is completely mapped and made ready for IO dispatch. We cannot avoid this problem - the current code already has this ioend chaining within a page so we can mostly ignore that it occurs. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-02-15 13:23:12 +07:00
struct list_head *iolist)
{
struct xfs_inode *ip = XFS_I(inode);
struct xfs_mount *mp = ip->i_mount;
struct block_device *bdev = xfs_find_bdev_for_inode(inode);
unsigned len = i_blocksize(inode);
unsigned poff = offset & (PAGE_SIZE - 1);
sector_t sector;
sector = xfs_fsb_to_db(ip, wpc->imap.br_startblock) +
((offset - XFS_FSB_TO_B(mp, wpc->imap.br_startoff)) >> 9);
if (!wpc->ioend ||
wpc->fork != wpc->ioend->io_fork ||
wpc->imap.br_state != wpc->ioend->io_state ||
sector != bio_end_sector(wpc->ioend->io_bio) ||
offset != wpc->ioend->io_offset + wpc->ioend->io_size) {
xfs: don't chain ioends during writepage submission Currently we can build a long ioend chain during ->writepages that gets attached to the writepage context. IO submission only then occurs when we finish all the writepage processing. This means we can have many ioends allocated and pending, and this violates the mempool guarantees that we need to give about forwards progress. i.e. we really should only have one ioend being built at a time, otherwise we may drain the mempool trying to allocate a new ioend and that blocks submission, completion and freeing of ioends that are already in progress. To prevent this situation from happening, we need to submit ioends for IO as soon as they are ready for dispatch rather than queuing them for later submission. This means the ioends have bios built immediately and they get queued on any plug that is current active. Hence if we schedule away from writeback, the ioends that have been built will make forwards progress due to the plug flushing on context switch. This will also prevent context switches from creating unnecessary IO submission latency. We can't completely avoid having nested IO allocation - when we have a block size smaller than a page size, we still need to hold the ioend submission until after we have marked the current page dirty. Hence we may need multiple ioends to be held while the current page is completely mapped and made ready for IO dispatch. We cannot avoid this problem - the current code already has this ioend chaining within a page so we can mostly ignore that it occurs. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-02-15 13:23:12 +07:00
if (wpc->ioend)
list_add(&wpc->ioend->io_list, iolist);
wpc->ioend = xfs_alloc_ioend(inode, wpc->fork,
wpc->imap.br_state, offset, bdev, sector);
}
if (!__bio_try_merge_page(wpc->ioend->io_bio, page, len, poff, true)) {
if (iop)
atomic_inc(&iop->write_count);
if (bio_full(wpc->ioend->io_bio))
xfs_chain_bio(wpc->ioend, wbc, bdev, sector);
bio_add_page(wpc->ioend->io_bio, page, len, poff);
}
wpc->ioend->io_size += len;
}
STATIC void
xfs_vm_invalidatepage(
struct page *page,
unsigned int offset,
unsigned int length)
{
trace_xfs_invalidatepage(page->mapping->host, page, offset, length);
iomap_invalidatepage(page, offset, length);
}
/*
* If the page has delalloc blocks on it, we need to punch them out before we
* invalidate the page. If we don't, we leave a stale delalloc mapping on the
* inode that can trip up a later direct I/O read operation on the same region.
*
* We prevent this by truncating away the delalloc regions on the page. Because
* they are delalloc, we can do this without needing a transaction. Indeed - if
* we get ENOSPC errors, we have to be able to do this truncation without a
* transaction as there is no space left for block reservation (typically why we
* see a ENOSPC in writeback).
*/
STATIC void
xfs_aops_discard_page(
struct page *page)
{
struct inode *inode = page->mapping->host;
struct xfs_inode *ip = XFS_I(inode);
struct xfs_mount *mp = ip->i_mount;
loff_t offset = page_offset(page);
xfs_fileoff_t start_fsb = XFS_B_TO_FSBT(mp, offset);
int error;
if (XFS_FORCED_SHUTDOWN(mp))
goto out_invalidate;
xfs_alert(mp,
"page discard on page "PTR_FMT", inode 0x%llx, offset %llu.",
page, ip->i_ino, offset);
error = xfs_bmap_punch_delalloc_range(ip, start_fsb,
PAGE_SIZE / i_blocksize(inode));
if (error && !XFS_FORCED_SHUTDOWN(mp))
xfs_alert(mp, "page discard unable to remove delalloc mapping.");
out_invalidate:
mm, fs: get rid of PAGE_CACHE_* and page_cache_{get,release} macros PAGE_CACHE_{SIZE,SHIFT,MASK,ALIGN} macros were introduced *long* time ago with promise that one day it will be possible to implement page cache with bigger chunks than PAGE_SIZE. This promise never materialized. And unlikely will. We have many places where PAGE_CACHE_SIZE assumed to be equal to PAGE_SIZE. And it's constant source of confusion on whether PAGE_CACHE_* or PAGE_* constant should be used in a particular case, especially on the border between fs and mm. Global switching to PAGE_CACHE_SIZE != PAGE_SIZE would cause to much breakage to be doable. Let's stop pretending that pages in page cache are special. They are not. The changes are pretty straight-forward: - <foo> << (PAGE_CACHE_SHIFT - PAGE_SHIFT) -> <foo>; - <foo> >> (PAGE_CACHE_SHIFT - PAGE_SHIFT) -> <foo>; - PAGE_CACHE_{SIZE,SHIFT,MASK,ALIGN} -> PAGE_{SIZE,SHIFT,MASK,ALIGN}; - page_cache_get() -> get_page(); - page_cache_release() -> put_page(); This patch contains automated changes generated with coccinelle using script below. For some reason, coccinelle doesn't patch header files. I've called spatch for them manually. The only adjustment after coccinelle is revert of changes to PAGE_CAHCE_ALIGN definition: we are going to drop it later. There are few places in the code where coccinelle didn't reach. I'll fix them manually in a separate patch. Comments and documentation also will be addressed with the separate patch. virtual patch @@ expression E; @@ - E << (PAGE_CACHE_SHIFT - PAGE_SHIFT) + E @@ expression E; @@ - E >> (PAGE_CACHE_SHIFT - PAGE_SHIFT) + E @@ @@ - PAGE_CACHE_SHIFT + PAGE_SHIFT @@ @@ - PAGE_CACHE_SIZE + PAGE_SIZE @@ @@ - PAGE_CACHE_MASK + PAGE_MASK @@ expression E; @@ - PAGE_CACHE_ALIGN(E) + PAGE_ALIGN(E) @@ expression E; @@ - page_cache_get(E) + get_page(E) @@ expression E; @@ - page_cache_release(E) + put_page(E) Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> Acked-by: Michal Hocko <mhocko@suse.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-04-01 19:29:47 +07:00
xfs_vm_invalidatepage(page, 0, PAGE_SIZE);
}
xfs: don't chain ioends during writepage submission Currently we can build a long ioend chain during ->writepages that gets attached to the writepage context. IO submission only then occurs when we finish all the writepage processing. This means we can have many ioends allocated and pending, and this violates the mempool guarantees that we need to give about forwards progress. i.e. we really should only have one ioend being built at a time, otherwise we may drain the mempool trying to allocate a new ioend and that blocks submission, completion and freeing of ioends that are already in progress. To prevent this situation from happening, we need to submit ioends for IO as soon as they are ready for dispatch rather than queuing them for later submission. This means the ioends have bios built immediately and they get queued on any plug that is current active. Hence if we schedule away from writeback, the ioends that have been built will make forwards progress due to the plug flushing on context switch. This will also prevent context switches from creating unnecessary IO submission latency. We can't completely avoid having nested IO allocation - when we have a block size smaller than a page size, we still need to hold the ioend submission until after we have marked the current page dirty. Hence we may need multiple ioends to be held while the current page is completely mapped and made ready for IO dispatch. We cannot avoid this problem - the current code already has this ioend chaining within a page so we can mostly ignore that it occurs. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-02-15 13:23:12 +07:00
/*
* We implement an immediate ioend submission policy here to avoid needing to
* chain multiple ioends and hence nest mempool allocations which can violate
* forward progress guarantees we need to provide. The current ioend we are
* adding blocks to is cached on the writepage context, and if the new block
xfs: don't chain ioends during writepage submission Currently we can build a long ioend chain during ->writepages that gets attached to the writepage context. IO submission only then occurs when we finish all the writepage processing. This means we can have many ioends allocated and pending, and this violates the mempool guarantees that we need to give about forwards progress. i.e. we really should only have one ioend being built at a time, otherwise we may drain the mempool trying to allocate a new ioend and that blocks submission, completion and freeing of ioends that are already in progress. To prevent this situation from happening, we need to submit ioends for IO as soon as they are ready for dispatch rather than queuing them for later submission. This means the ioends have bios built immediately and they get queued on any plug that is current active. Hence if we schedule away from writeback, the ioends that have been built will make forwards progress due to the plug flushing on context switch. This will also prevent context switches from creating unnecessary IO submission latency. We can't completely avoid having nested IO allocation - when we have a block size smaller than a page size, we still need to hold the ioend submission until after we have marked the current page dirty. Hence we may need multiple ioends to be held while the current page is completely mapped and made ready for IO dispatch. We cannot avoid this problem - the current code already has this ioend chaining within a page so we can mostly ignore that it occurs. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-02-15 13:23:12 +07:00
* does not append to the cached ioend it will create a new ioend and cache that
* instead.
*
* If a new ioend is created and cached, the old ioend is returned and queued
* locally for submission once the entire page is processed or an error has been
* detected. While ioends are submitted immediately after they are completed,
* batching optimisations are provided by higher level block plugging.
*
* At the end of a writeback pass, there will be a cached ioend remaining on the
* writepage context that the caller will need to submit.
*/
static int
xfs_writepage_map(
struct xfs_writepage_ctx *wpc,
xfs: don't chain ioends during writepage submission Currently we can build a long ioend chain during ->writepages that gets attached to the writepage context. IO submission only then occurs when we finish all the writepage processing. This means we can have many ioends allocated and pending, and this violates the mempool guarantees that we need to give about forwards progress. i.e. we really should only have one ioend being built at a time, otherwise we may drain the mempool trying to allocate a new ioend and that blocks submission, completion and freeing of ioends that are already in progress. To prevent this situation from happening, we need to submit ioends for IO as soon as they are ready for dispatch rather than queuing them for later submission. This means the ioends have bios built immediately and they get queued on any plug that is current active. Hence if we schedule away from writeback, the ioends that have been built will make forwards progress due to the plug flushing on context switch. This will also prevent context switches from creating unnecessary IO submission latency. We can't completely avoid having nested IO allocation - when we have a block size smaller than a page size, we still need to hold the ioend submission until after we have marked the current page dirty. Hence we may need multiple ioends to be held while the current page is completely mapped and made ready for IO dispatch. We cannot avoid this problem - the current code already has this ioend chaining within a page so we can mostly ignore that it occurs. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-02-15 13:23:12 +07:00
struct writeback_control *wbc,
struct inode *inode,
struct page *page,
uint64_t end_offset)
{
xfs: don't chain ioends during writepage submission Currently we can build a long ioend chain during ->writepages that gets attached to the writepage context. IO submission only then occurs when we finish all the writepage processing. This means we can have many ioends allocated and pending, and this violates the mempool guarantees that we need to give about forwards progress. i.e. we really should only have one ioend being built at a time, otherwise we may drain the mempool trying to allocate a new ioend and that blocks submission, completion and freeing of ioends that are already in progress. To prevent this situation from happening, we need to submit ioends for IO as soon as they are ready for dispatch rather than queuing them for later submission. This means the ioends have bios built immediately and they get queued on any plug that is current active. Hence if we schedule away from writeback, the ioends that have been built will make forwards progress due to the plug flushing on context switch. This will also prevent context switches from creating unnecessary IO submission latency. We can't completely avoid having nested IO allocation - when we have a block size smaller than a page size, we still need to hold the ioend submission until after we have marked the current page dirty. Hence we may need multiple ioends to be held while the current page is completely mapped and made ready for IO dispatch. We cannot avoid this problem - the current code already has this ioend chaining within a page so we can mostly ignore that it occurs. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-02-15 13:23:12 +07:00
LIST_HEAD(submit_list);
struct iomap_page *iop = to_iomap_page(page);
unsigned len = i_blocksize(inode);
xfs: don't chain ioends during writepage submission Currently we can build a long ioend chain during ->writepages that gets attached to the writepage context. IO submission only then occurs when we finish all the writepage processing. This means we can have many ioends allocated and pending, and this violates the mempool guarantees that we need to give about forwards progress. i.e. we really should only have one ioend being built at a time, otherwise we may drain the mempool trying to allocate a new ioend and that blocks submission, completion and freeing of ioends that are already in progress. To prevent this situation from happening, we need to submit ioends for IO as soon as they are ready for dispatch rather than queuing them for later submission. This means the ioends have bios built immediately and they get queued on any plug that is current active. Hence if we schedule away from writeback, the ioends that have been built will make forwards progress due to the plug flushing on context switch. This will also prevent context switches from creating unnecessary IO submission latency. We can't completely avoid having nested IO allocation - when we have a block size smaller than a page size, we still need to hold the ioend submission until after we have marked the current page dirty. Hence we may need multiple ioends to be held while the current page is completely mapped and made ready for IO dispatch. We cannot avoid this problem - the current code already has this ioend chaining within a page so we can mostly ignore that it occurs. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-02-15 13:23:12 +07:00
struct xfs_ioend *ioend, *next;
uint64_t file_offset; /* file offset of page */
int error = 0, count = 0, i;
ASSERT(iop || i_blocksize(inode) == PAGE_SIZE);
ASSERT(!iop || atomic_read(&iop->write_count) == 0);
xfs: make xfs_writepage_map extent map centric xfs_writepage_map() iterates over the bufferheads on a page to decide what sort of IO to do and what actions to take. However, when it comes to reflink and deciding when it needs to execute a COW operation, we no longer look at the bufferhead state but instead we ignore than and look up internal state held in the COW fork extent list. This means xfs_writepage_map() is somewhat confused. It does stuff, then ignores it, then tries to handle the impedence mismatch by shovelling the results inside the existing mapping code. It works, but it's a bit of a mess and it makes it hard to fix the cached map bug that the writepage code currently has. To unify the two different mechanisms, we first have to choose a direction. That's already been set - we're de-emphasising bufferheads so they are no longer a control structure as we need to do taht to allow for eventual removal. Hence we need to move away from looking at bufferhead state to determine what operations we need to perform. We can't completely get rid of bufferheads yet - they do contain some state that is absolutely necessary, such as whether that part of the page contains valid data or not (buffer_uptodate()). Other state in the bufferhead is redundant: BH_dirty - the page is dirty, so we can ignore this and just write it BH_delay - we have delalloc extent info in the DATA fork extent tree BH_unwritten - same as BH_delay BH_mapped - indicates we've already used it once for IO and it is mapped to a disk address. Needs to be ignored for COW blocks. The BH_mapped flag is an interesting case - it's supposed to indicate that it's already mapped to disk and so we can just use it "as is". In theory, we don't even have to do an extent lookup to find where to write it too, but we have to do that anyway to determine we are actually writing over a valid extent. Hence it's not even serving the purpose of avoiding a an extent lookup during writeback, and so we can pretty much ignore it. Especially as we have to ignore it for COW operations... Therefore, use the extent map as the source of information to tell us what actions we need to take and what sort of IO we should perform. The first step is to have xfs_map_blocks() set the io type according to what it looks up. This means it can easily handle both normal overwrite and COW cases. The only thing we also need to add is the ability to return hole mappings. We need to return and cache hole mappings now for the case of multiple blocks per page. We no longer use the BH_mapped to indicate a block over a hole, so we have to get that info from xfs_map_blocks(). We cache it so that holes that span two pages don't need separate lookups. This allows us to avoid ever doing write IO over a hole, too. Now that we have xfs_map_blocks() returning both a cached map and the type of IO we need to perform, we can rewrite xfs_writepage_map() to drop all the bufferhead control. It's also much simplified because it doesn't need to explicitly handle COW operations. Instead of iterating bufferheads, it iterates blocks within the page and then looks up what per-block state is required from the appropriate bufferhead. It then validates the cached map, and if it's not valid, we get a new map. If we don't get a valid map or it's over a hole, we skip the block. At this point, we have to remap the bufferhead via xfs_map_at_offset(). As previously noted, we had to do this even if the buffer was already mapped as the mapping would be stale for XFS_IO_DELALLOC, XFS_IO_UNWRITTEN and XFS_IO_COW IO types. With xfs_map_blocks() now controlling the type, even XFS_IO_OVERWRITE types need remapping, as converted-but-not-yet- written delalloc extents beyond EOF can be reported at XFS_IO_OVERWRITE. Bufferheads that span such regions still need their BH_Delay flags cleared and their block numbers calculated, so we now unconditionally map each bufferhead before submission. But wait! There's more - remember the old "treat unwritten extents as holes on read" hack? Yeah, that means we can have a dirty page with unmapped, unwritten bufferheads that contain data! What makes these so special is that the unwritten "hole" bufferheads do not have a valid block device pointer, so if we attempt to write them xfs_add_to_ioend() blows up. So we make xfs_map_at_offset() do the "realtime or data device" lookup from the inode and ignore what was or wasn't put into the bufferhead when the buffer was instantiated. The astute reader will have realised by now that this code treats unwritten extents in multiple-blocks-per-page situations differently. If we get any combination of unwritten blocks on a dirty page that contain valid data in the page, we're going to convert them to real extents. This can actually be a win, because it means that pages with interleaving unwritten and written blocks will get converted to a single written extent with zeros replacing the interspersed unwritten blocks. This is actually good for reducing extent list and conversion overhead, and it means we issue a contiguous IO instead of lots of little ones. The downside is that we use up a little extra IO bandwidth. Neither of these seem like a bad thing given that spinning disks are seek sensitive, and SSDs/pmem have bandwidth to burn and the lower Io latency/CPU overhead of fewer, larger IOs will result in better performance on them... As a result of all this, the only state we actually care about from the bufferhead is a single flag - BH_Uptodate. We still use the bufferhead to pass some information to the bio via xfs_add_to_ioend(), but that is trivial to separate and pass explicitly. This means we really only need 1 bit of state per block per page from the buffered write path in the writeback path. Everything else we do with the bufferhead is purely to make the buffered IO front end continue to work correctly. i.e we've pretty much marginalised bufferheads in the writeback path completely. Signed-off-By: Dave Chinner <dchinner@redhat.com> [hch: forward port, refactor and split off bits into other commits] Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2018-07-12 12:26:00 +07:00
/*
* Walk through the page to find areas to write back. If we run off the
* end of the current map or find the current map invalid, grab a new
* one.
xfs: make xfs_writepage_map extent map centric xfs_writepage_map() iterates over the bufferheads on a page to decide what sort of IO to do and what actions to take. However, when it comes to reflink and deciding when it needs to execute a COW operation, we no longer look at the bufferhead state but instead we ignore than and look up internal state held in the COW fork extent list. This means xfs_writepage_map() is somewhat confused. It does stuff, then ignores it, then tries to handle the impedence mismatch by shovelling the results inside the existing mapping code. It works, but it's a bit of a mess and it makes it hard to fix the cached map bug that the writepage code currently has. To unify the two different mechanisms, we first have to choose a direction. That's already been set - we're de-emphasising bufferheads so they are no longer a control structure as we need to do taht to allow for eventual removal. Hence we need to move away from looking at bufferhead state to determine what operations we need to perform. We can't completely get rid of bufferheads yet - they do contain some state that is absolutely necessary, such as whether that part of the page contains valid data or not (buffer_uptodate()). Other state in the bufferhead is redundant: BH_dirty - the page is dirty, so we can ignore this and just write it BH_delay - we have delalloc extent info in the DATA fork extent tree BH_unwritten - same as BH_delay BH_mapped - indicates we've already used it once for IO and it is mapped to a disk address. Needs to be ignored for COW blocks. The BH_mapped flag is an interesting case - it's supposed to indicate that it's already mapped to disk and so we can just use it "as is". In theory, we don't even have to do an extent lookup to find where to write it too, but we have to do that anyway to determine we are actually writing over a valid extent. Hence it's not even serving the purpose of avoiding a an extent lookup during writeback, and so we can pretty much ignore it. Especially as we have to ignore it for COW operations... Therefore, use the extent map as the source of information to tell us what actions we need to take and what sort of IO we should perform. The first step is to have xfs_map_blocks() set the io type according to what it looks up. This means it can easily handle both normal overwrite and COW cases. The only thing we also need to add is the ability to return hole mappings. We need to return and cache hole mappings now for the case of multiple blocks per page. We no longer use the BH_mapped to indicate a block over a hole, so we have to get that info from xfs_map_blocks(). We cache it so that holes that span two pages don't need separate lookups. This allows us to avoid ever doing write IO over a hole, too. Now that we have xfs_map_blocks() returning both a cached map and the type of IO we need to perform, we can rewrite xfs_writepage_map() to drop all the bufferhead control. It's also much simplified because it doesn't need to explicitly handle COW operations. Instead of iterating bufferheads, it iterates blocks within the page and then looks up what per-block state is required from the appropriate bufferhead. It then validates the cached map, and if it's not valid, we get a new map. If we don't get a valid map or it's over a hole, we skip the block. At this point, we have to remap the bufferhead via xfs_map_at_offset(). As previously noted, we had to do this even if the buffer was already mapped as the mapping would be stale for XFS_IO_DELALLOC, XFS_IO_UNWRITTEN and XFS_IO_COW IO types. With xfs_map_blocks() now controlling the type, even XFS_IO_OVERWRITE types need remapping, as converted-but-not-yet- written delalloc extents beyond EOF can be reported at XFS_IO_OVERWRITE. Bufferheads that span such regions still need their BH_Delay flags cleared and their block numbers calculated, so we now unconditionally map each bufferhead before submission. But wait! There's more - remember the old "treat unwritten extents as holes on read" hack? Yeah, that means we can have a dirty page with unmapped, unwritten bufferheads that contain data! What makes these so special is that the unwritten "hole" bufferheads do not have a valid block device pointer, so if we attempt to write them xfs_add_to_ioend() blows up. So we make xfs_map_at_offset() do the "realtime or data device" lookup from the inode and ignore what was or wasn't put into the bufferhead when the buffer was instantiated. The astute reader will have realised by now that this code treats unwritten extents in multiple-blocks-per-page situations differently. If we get any combination of unwritten blocks on a dirty page that contain valid data in the page, we're going to convert them to real extents. This can actually be a win, because it means that pages with interleaving unwritten and written blocks will get converted to a single written extent with zeros replacing the interspersed unwritten blocks. This is actually good for reducing extent list and conversion overhead, and it means we issue a contiguous IO instead of lots of little ones. The downside is that we use up a little extra IO bandwidth. Neither of these seem like a bad thing given that spinning disks are seek sensitive, and SSDs/pmem have bandwidth to burn and the lower Io latency/CPU overhead of fewer, larger IOs will result in better performance on them... As a result of all this, the only state we actually care about from the bufferhead is a single flag - BH_Uptodate. We still use the bufferhead to pass some information to the bio via xfs_add_to_ioend(), but that is trivial to separate and pass explicitly. This means we really only need 1 bit of state per block per page from the buffered write path in the writeback path. Everything else we do with the bufferhead is purely to make the buffered IO front end continue to work correctly. i.e we've pretty much marginalised bufferheads in the writeback path completely. Signed-off-By: Dave Chinner <dchinner@redhat.com> [hch: forward port, refactor and split off bits into other commits] Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2018-07-12 12:26:00 +07:00
*/
for (i = 0, file_offset = page_offset(page);
i < (PAGE_SIZE >> inode->i_blkbits) && file_offset < end_offset;
i++, file_offset += len) {
if (iop && !test_bit(i, iop->uptodate))
continue;
error = xfs_map_blocks(wpc, inode, file_offset);
if (error)
break;
if (wpc->imap.br_startblock == HOLESTARTBLOCK)
continue;
xfs_add_to_ioend(inode, file_offset, page, iop, wpc, wbc,
&submit_list);
count++;
xfs: make xfs_writepage_map extent map centric xfs_writepage_map() iterates over the bufferheads on a page to decide what sort of IO to do and what actions to take. However, when it comes to reflink and deciding when it needs to execute a COW operation, we no longer look at the bufferhead state but instead we ignore than and look up internal state held in the COW fork extent list. This means xfs_writepage_map() is somewhat confused. It does stuff, then ignores it, then tries to handle the impedence mismatch by shovelling the results inside the existing mapping code. It works, but it's a bit of a mess and it makes it hard to fix the cached map bug that the writepage code currently has. To unify the two different mechanisms, we first have to choose a direction. That's already been set - we're de-emphasising bufferheads so they are no longer a control structure as we need to do taht to allow for eventual removal. Hence we need to move away from looking at bufferhead state to determine what operations we need to perform. We can't completely get rid of bufferheads yet - they do contain some state that is absolutely necessary, such as whether that part of the page contains valid data or not (buffer_uptodate()). Other state in the bufferhead is redundant: BH_dirty - the page is dirty, so we can ignore this and just write it BH_delay - we have delalloc extent info in the DATA fork extent tree BH_unwritten - same as BH_delay BH_mapped - indicates we've already used it once for IO and it is mapped to a disk address. Needs to be ignored for COW blocks. The BH_mapped flag is an interesting case - it's supposed to indicate that it's already mapped to disk and so we can just use it "as is". In theory, we don't even have to do an extent lookup to find where to write it too, but we have to do that anyway to determine we are actually writing over a valid extent. Hence it's not even serving the purpose of avoiding a an extent lookup during writeback, and so we can pretty much ignore it. Especially as we have to ignore it for COW operations... Therefore, use the extent map as the source of information to tell us what actions we need to take and what sort of IO we should perform. The first step is to have xfs_map_blocks() set the io type according to what it looks up. This means it can easily handle both normal overwrite and COW cases. The only thing we also need to add is the ability to return hole mappings. We need to return and cache hole mappings now for the case of multiple blocks per page. We no longer use the BH_mapped to indicate a block over a hole, so we have to get that info from xfs_map_blocks(). We cache it so that holes that span two pages don't need separate lookups. This allows us to avoid ever doing write IO over a hole, too. Now that we have xfs_map_blocks() returning both a cached map and the type of IO we need to perform, we can rewrite xfs_writepage_map() to drop all the bufferhead control. It's also much simplified because it doesn't need to explicitly handle COW operations. Instead of iterating bufferheads, it iterates blocks within the page and then looks up what per-block state is required from the appropriate bufferhead. It then validates the cached map, and if it's not valid, we get a new map. If we don't get a valid map or it's over a hole, we skip the block. At this point, we have to remap the bufferhead via xfs_map_at_offset(). As previously noted, we had to do this even if the buffer was already mapped as the mapping would be stale for XFS_IO_DELALLOC, XFS_IO_UNWRITTEN and XFS_IO_COW IO types. With xfs_map_blocks() now controlling the type, even XFS_IO_OVERWRITE types need remapping, as converted-but-not-yet- written delalloc extents beyond EOF can be reported at XFS_IO_OVERWRITE. Bufferheads that span such regions still need their BH_Delay flags cleared and their block numbers calculated, so we now unconditionally map each bufferhead before submission. But wait! There's more - remember the old "treat unwritten extents as holes on read" hack? Yeah, that means we can have a dirty page with unmapped, unwritten bufferheads that contain data! What makes these so special is that the unwritten "hole" bufferheads do not have a valid block device pointer, so if we attempt to write them xfs_add_to_ioend() blows up. So we make xfs_map_at_offset() do the "realtime or data device" lookup from the inode and ignore what was or wasn't put into the bufferhead when the buffer was instantiated. The astute reader will have realised by now that this code treats unwritten extents in multiple-blocks-per-page situations differently. If we get any combination of unwritten blocks on a dirty page that contain valid data in the page, we're going to convert them to real extents. This can actually be a win, because it means that pages with interleaving unwritten and written blocks will get converted to a single written extent with zeros replacing the interspersed unwritten blocks. This is actually good for reducing extent list and conversion overhead, and it means we issue a contiguous IO instead of lots of little ones. The downside is that we use up a little extra IO bandwidth. Neither of these seem like a bad thing given that spinning disks are seek sensitive, and SSDs/pmem have bandwidth to burn and the lower Io latency/CPU overhead of fewer, larger IOs will result in better performance on them... As a result of all this, the only state we actually care about from the bufferhead is a single flag - BH_Uptodate. We still use the bufferhead to pass some information to the bio via xfs_add_to_ioend(), but that is trivial to separate and pass explicitly. This means we really only need 1 bit of state per block per page from the buffered write path in the writeback path. Everything else we do with the bufferhead is purely to make the buffered IO front end continue to work correctly. i.e we've pretty much marginalised bufferheads in the writeback path completely. Signed-off-By: Dave Chinner <dchinner@redhat.com> [hch: forward port, refactor and split off bits into other commits] Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2018-07-12 12:26:00 +07:00
}
xfs: don't chain ioends during writepage submission Currently we can build a long ioend chain during ->writepages that gets attached to the writepage context. IO submission only then occurs when we finish all the writepage processing. This means we can have many ioends allocated and pending, and this violates the mempool guarantees that we need to give about forwards progress. i.e. we really should only have one ioend being built at a time, otherwise we may drain the mempool trying to allocate a new ioend and that blocks submission, completion and freeing of ioends that are already in progress. To prevent this situation from happening, we need to submit ioends for IO as soon as they are ready for dispatch rather than queuing them for later submission. This means the ioends have bios built immediately and they get queued on any plug that is current active. Hence if we schedule away from writeback, the ioends that have been built will make forwards progress due to the plug flushing on context switch. This will also prevent context switches from creating unnecessary IO submission latency. We can't completely avoid having nested IO allocation - when we have a block size smaller than a page size, we still need to hold the ioend submission until after we have marked the current page dirty. Hence we may need multiple ioends to be held while the current page is completely mapped and made ready for IO dispatch. We cannot avoid this problem - the current code already has this ioend chaining within a page so we can mostly ignore that it occurs. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-02-15 13:23:12 +07:00
ASSERT(wpc->ioend || list_empty(&submit_list));
ASSERT(PageLocked(page));
ASSERT(!PageWriteback(page));
/*
* On error, we have to fail the ioend here because we may have set
* pages under writeback, we have to make sure we run IO completion to
* mark the error state of the IO appropriately, so we can't cancel the
* ioend directly here. That means we have to mark this page as under
* writeback if we included any blocks from it in the ioend chain so
* that completion treats it correctly.
*
xfs: don't chain ioends during writepage submission Currently we can build a long ioend chain during ->writepages that gets attached to the writepage context. IO submission only then occurs when we finish all the writepage processing. This means we can have many ioends allocated and pending, and this violates the mempool guarantees that we need to give about forwards progress. i.e. we really should only have one ioend being built at a time, otherwise we may drain the mempool trying to allocate a new ioend and that blocks submission, completion and freeing of ioends that are already in progress. To prevent this situation from happening, we need to submit ioends for IO as soon as they are ready for dispatch rather than queuing them for later submission. This means the ioends have bios built immediately and they get queued on any plug that is current active. Hence if we schedule away from writeback, the ioends that have been built will make forwards progress due to the plug flushing on context switch. This will also prevent context switches from creating unnecessary IO submission latency. We can't completely avoid having nested IO allocation - when we have a block size smaller than a page size, we still need to hold the ioend submission until after we have marked the current page dirty. Hence we may need multiple ioends to be held while the current page is completely mapped and made ready for IO dispatch. We cannot avoid this problem - the current code already has this ioend chaining within a page so we can mostly ignore that it occurs. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-02-15 13:23:12 +07:00
* If we didn't include the page in the ioend, the on error we can
* simply discard and unlock it as there are no other users of the page
* now. The caller will still need to trigger submission of outstanding
* ioends on the writepage context so they are treated correctly on
* error.
*/
if (unlikely(error)) {
if (!count) {
xfs_aops_discard_page(page);
ClearPageUptodate(page);
unlock_page(page);
goto done;
}
/*
* If the page was not fully cleaned, we need to ensure that the
* higher layers come back to it correctly. That means we need
* to keep the page dirty, and for WB_SYNC_ALL writeback we need
* to ensure the PAGECACHE_TAG_TOWRITE index mark is not removed
* so another attempt to write this page in this writeback sweep
* will be made.
*/
set_page_writeback_keepwrite(page);
xfs: don't chain ioends during writepage submission Currently we can build a long ioend chain during ->writepages that gets attached to the writepage context. IO submission only then occurs when we finish all the writepage processing. This means we can have many ioends allocated and pending, and this violates the mempool guarantees that we need to give about forwards progress. i.e. we really should only have one ioend being built at a time, otherwise we may drain the mempool trying to allocate a new ioend and that blocks submission, completion and freeing of ioends that are already in progress. To prevent this situation from happening, we need to submit ioends for IO as soon as they are ready for dispatch rather than queuing them for later submission. This means the ioends have bios built immediately and they get queued on any plug that is current active. Hence if we schedule away from writeback, the ioends that have been built will make forwards progress due to the plug flushing on context switch. This will also prevent context switches from creating unnecessary IO submission latency. We can't completely avoid having nested IO allocation - when we have a block size smaller than a page size, we still need to hold the ioend submission until after we have marked the current page dirty. Hence we may need multiple ioends to be held while the current page is completely mapped and made ready for IO dispatch. We cannot avoid this problem - the current code already has this ioend chaining within a page so we can mostly ignore that it occurs. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-02-15 13:23:12 +07:00
} else {
clear_page_dirty_for_io(page);
set_page_writeback(page);
}
xfs: don't chain ioends during writepage submission Currently we can build a long ioend chain during ->writepages that gets attached to the writepage context. IO submission only then occurs when we finish all the writepage processing. This means we can have many ioends allocated and pending, and this violates the mempool guarantees that we need to give about forwards progress. i.e. we really should only have one ioend being built at a time, otherwise we may drain the mempool trying to allocate a new ioend and that blocks submission, completion and freeing of ioends that are already in progress. To prevent this situation from happening, we need to submit ioends for IO as soon as they are ready for dispatch rather than queuing them for later submission. This means the ioends have bios built immediately and they get queued on any plug that is current active. Hence if we schedule away from writeback, the ioends that have been built will make forwards progress due to the plug flushing on context switch. This will also prevent context switches from creating unnecessary IO submission latency. We can't completely avoid having nested IO allocation - when we have a block size smaller than a page size, we still need to hold the ioend submission until after we have marked the current page dirty. Hence we may need multiple ioends to be held while the current page is completely mapped and made ready for IO dispatch. We cannot avoid this problem - the current code already has this ioend chaining within a page so we can mostly ignore that it occurs. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-02-15 13:23:12 +07:00
unlock_page(page);
/*
* Preserve the original error if there was one, otherwise catch
* submission errors here and propagate into subsequent ioend
* submissions.
*/
list_for_each_entry_safe(ioend, next, &submit_list, io_list) {
int error2;
list_del_init(&ioend->io_list);
error2 = xfs_submit_ioend(wbc, ioend, error);
if (error2 && !error)
error = error2;
}
/*
* We can end up here with no error and nothing to write only if we race
* with a partial page truncate on a sub-page block sized filesystem.
*/
if (!count)
end_page_writeback(page);
done:
mapping_set_error(page->mapping, error);
return error;
}
/*
* Write out a dirty page.
*
* For delalloc space on the page we need to allocate space and flush it.
* For unwritten space on the page we need to start the conversion to
* regular allocated space.
*/
STATIC int
xfs: Introduce writeback context for writepages xfs_vm_writepages() calls generic_writepages to writeback a range of a file, but then xfs_vm_writepage() clusters pages itself as it does not have any context it can pass between->writepage calls from __write_cache_pages(). Introduce a writeback context for xfs_vm_writepages() and call __write_cache_pages directly with our own writepage callback so that we can pass that context to each writepage invocation. This encapsulates the current mapping, whether it is valid or not, the current ioend and it's IO type and the ioend chain being built. This requires us to move the ioend submission up to the level where the writepage context is declared. This does mean we do not submit IO until we packaged the entire writeback range, but with the block plugging in the writepages call this is the way IO is submitted, anyway. It also means that we need to handle discontiguous page ranges. If the pages sent down by write_cache_pages to the writepage callback are discontiguous, we need to detect this and put each discontiguous page range into individual ioends. This is needed to ensure that the ioend accurately represents the range of the file that it covers so that file size updates during IO completion set the size correctly. Failure to take into account the discontiguous ranges results in files being too small when writeback patterns are non-sequential. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-02-15 13:21:19 +07:00
xfs_do_writepage(
struct page *page,
xfs: Introduce writeback context for writepages xfs_vm_writepages() calls generic_writepages to writeback a range of a file, but then xfs_vm_writepage() clusters pages itself as it does not have any context it can pass between->writepage calls from __write_cache_pages(). Introduce a writeback context for xfs_vm_writepages() and call __write_cache_pages directly with our own writepage callback so that we can pass that context to each writepage invocation. This encapsulates the current mapping, whether it is valid or not, the current ioend and it's IO type and the ioend chain being built. This requires us to move the ioend submission up to the level where the writepage context is declared. This does mean we do not submit IO until we packaged the entire writeback range, but with the block plugging in the writepages call this is the way IO is submitted, anyway. It also means that we need to handle discontiguous page ranges. If the pages sent down by write_cache_pages to the writepage callback are discontiguous, we need to detect this and put each discontiguous page range into individual ioends. This is needed to ensure that the ioend accurately represents the range of the file that it covers so that file size updates during IO completion set the size correctly. Failure to take into account the discontiguous ranges results in files being too small when writeback patterns are non-sequential. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-02-15 13:21:19 +07:00
struct writeback_control *wbc,
void *data)
{
xfs: Introduce writeback context for writepages xfs_vm_writepages() calls generic_writepages to writeback a range of a file, but then xfs_vm_writepage() clusters pages itself as it does not have any context it can pass between->writepage calls from __write_cache_pages(). Introduce a writeback context for xfs_vm_writepages() and call __write_cache_pages directly with our own writepage callback so that we can pass that context to each writepage invocation. This encapsulates the current mapping, whether it is valid or not, the current ioend and it's IO type and the ioend chain being built. This requires us to move the ioend submission up to the level where the writepage context is declared. This does mean we do not submit IO until we packaged the entire writeback range, but with the block plugging in the writepages call this is the way IO is submitted, anyway. It also means that we need to handle discontiguous page ranges. If the pages sent down by write_cache_pages to the writepage callback are discontiguous, we need to detect this and put each discontiguous page range into individual ioends. This is needed to ensure that the ioend accurately represents the range of the file that it covers so that file size updates during IO completion set the size correctly. Failure to take into account the discontiguous ranges results in files being too small when writeback patterns are non-sequential. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-02-15 13:21:19 +07:00
struct xfs_writepage_ctx *wpc = data;
struct inode *inode = page->mapping->host;
loff_t offset;
uint64_t end_offset;
pgoff_t end_index;
trace_xfs_writepage(inode, page, 0, 0);
/*
* Refuse to write the page out if we are called from reclaim context.
*
* This avoids stack overflows when called from deeply used stacks in
* random callers for direct reclaim or memcg reclaim. We explicitly
* allow reclaim from kswapd as the stack usage there is relatively low.
*
* This should never happen except in the case of a VM regression so
* warn about it.
*/
if (WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) ==
PF_MEMALLOC))
goto redirty;
/*
* Given that we do not allow direct reclaim to call us, we should
* never be called while in a filesystem transaction.
*/
if (WARN_ON_ONCE(current->flags & PF_MEMALLOC_NOFS))
goto redirty;
xfs: fix infinite loop at xfs_vm_writepage on 32bit system Write to a file with an offset greater than 16TB on 32-bit system and then trigger page write-back via sync(1) will cause task hang. # block_size=4096 # offset=$(((2**32 - 1) * $block_size)) # xfs_io -f -c "pwrite $offset $block_size" /storage/test_file # sync INFO: task sync:2590 blocked for more than 120 seconds. "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. sync D c1064a28 0 2590 2097 0x00000000 ..... Call Trace: [<c1064a28>] ? ttwu_do_wakeup+0x18/0x130 [<c1066d0e>] ? try_to_wake_up+0x1ce/0x220 [<c1066dbf>] ? wake_up_process+0x1f/0x40 [<c104fc2e>] ? wake_up_worker+0x1e/0x30 [<c15b6083>] schedule+0x23/0x60 [<c15b3c2d>] schedule_timeout+0x18d/0x1f0 [<c12a143e>] ? do_raw_spin_unlock+0x4e/0x90 [<c10515f1>] ? __queue_delayed_work+0x91/0x150 [<c12a12ef>] ? do_raw_spin_lock+0x3f/0x100 [<c12a143e>] ? do_raw_spin_unlock+0x4e/0x90 [<c15b5b5d>] wait_for_completion+0x7d/0xc0 [<c1066d60>] ? try_to_wake_up+0x220/0x220 [<c116a4d2>] sync_inodes_sb+0x92/0x180 [<c116fb05>] sync_inodes_one_sb+0x15/0x20 [<c114a8f8>] iterate_supers+0xb8/0xc0 [<c116faf0>] ? fdatawrite_one_bdev+0x20/0x20 [<c116fc21>] sys_sync+0x31/0x80 [<c15be18d>] sysenter_do_call+0x12/0x28 This issue can be triggered via xfstests/generic/308. The reason is that the end_index is unsigned long with maximum value '2^32-1=4294967295' on 32-bit platform, and the given offset cause it wrapped to 0, so that the following codes will repeat again and again until the task schedule time out: end_index = offset >> PAGE_CACHE_SHIFT; last_index = (offset - 1) >> PAGE_CACHE_SHIFT; if (page->index >= end_index) { unsigned offset_into_page = offset & (PAGE_CACHE_SIZE - 1); /* * Just skip the page if it is fully outside i_size, e.g. due * to a truncate operation that is in progress. */ if (page->index >= end_index + 1 || offset_into_page == 0) { ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ unlock_page(page); return 0; } In order to check if a page is fully outsids i_size or not, we can fix the code logic as below: if (page->index > end_index || (page->index == end_index && offset_into_page == 0)) Secondly, there still has another similar issue when calculating the end offset for mapping the filesystem blocks to the file blocks for delalloc. With the same tests to above, run unmount(8) will cause kernel panic if CONFIG_XFS_DEBUG is enabled: XFS: Assertion failed: XFS_FORCED_SHUTDOWN(ip->i_mount) || \ ip->i_delayed_blks == 0, file: fs/xfs/xfs_super.c, line: 964 kernel BUG at fs/xfs/xfs_message.c:108! invalid opcode: 0000 [#1] SMP task: edddc100 ti: ec6ee000 task.ti: ec6ee000 EIP: 0060:[<f83d87cb>] EFLAGS: 00010296 CPU: 1 EIP is at assfail+0x2b/0x30 [xfs] .............. Call Trace: [<f83d9cd4>] xfs_fs_destroy_inode+0x74/0x120 [xfs] [<c115ddf1>] destroy_inode+0x31/0x50 [<c115deff>] evict+0xef/0x170 [<c115dfb2>] dispose_list+0x32/0x40 [<c115ea3a>] evict_inodes+0xca/0xe0 [<c1149706>] generic_shutdown_super+0x46/0xd0 [<c11497b9>] kill_block_super+0x29/0x70 [<c1149a14>] deactivate_locked_super+0x44/0x70 [<c114a427>] deactivate_super+0x47/0x60 [<c1161c3d>] mntput_no_expire+0xcd/0x120 [<c1162ae8>] SyS_umount+0xa8/0x370 [<c1162dce>] SyS_oldumount+0x1e/0x20 [<c15be18d>] sysenter_do_call+0x12/0x28 That because the end_offset is evaluated to 0 which is the same reason to above, hence the mapping and covertion for dealloc file blocks to file system blocks did not happened. This patch just fixed both issues. Reported-by: Michael L. Semon <mlsemon35@gmail.com> Signed-off-by: Jie Liu <jeff.liu@oracle.com> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2014-05-20 05:24:26 +07:00
/*
* Is this page beyond the end of the file?
*
xfs: fix infinite loop at xfs_vm_writepage on 32bit system Write to a file with an offset greater than 16TB on 32-bit system and then trigger page write-back via sync(1) will cause task hang. # block_size=4096 # offset=$(((2**32 - 1) * $block_size)) # xfs_io -f -c "pwrite $offset $block_size" /storage/test_file # sync INFO: task sync:2590 blocked for more than 120 seconds. "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. sync D c1064a28 0 2590 2097 0x00000000 ..... Call Trace: [<c1064a28>] ? ttwu_do_wakeup+0x18/0x130 [<c1066d0e>] ? try_to_wake_up+0x1ce/0x220 [<c1066dbf>] ? wake_up_process+0x1f/0x40 [<c104fc2e>] ? wake_up_worker+0x1e/0x30 [<c15b6083>] schedule+0x23/0x60 [<c15b3c2d>] schedule_timeout+0x18d/0x1f0 [<c12a143e>] ? do_raw_spin_unlock+0x4e/0x90 [<c10515f1>] ? __queue_delayed_work+0x91/0x150 [<c12a12ef>] ? do_raw_spin_lock+0x3f/0x100 [<c12a143e>] ? do_raw_spin_unlock+0x4e/0x90 [<c15b5b5d>] wait_for_completion+0x7d/0xc0 [<c1066d60>] ? try_to_wake_up+0x220/0x220 [<c116a4d2>] sync_inodes_sb+0x92/0x180 [<c116fb05>] sync_inodes_one_sb+0x15/0x20 [<c114a8f8>] iterate_supers+0xb8/0xc0 [<c116faf0>] ? fdatawrite_one_bdev+0x20/0x20 [<c116fc21>] sys_sync+0x31/0x80 [<c15be18d>] sysenter_do_call+0x12/0x28 This issue can be triggered via xfstests/generic/308. The reason is that the end_index is unsigned long with maximum value '2^32-1=4294967295' on 32-bit platform, and the given offset cause it wrapped to 0, so that the following codes will repeat again and again until the task schedule time out: end_index = offset >> PAGE_CACHE_SHIFT; last_index = (offset - 1) >> PAGE_CACHE_SHIFT; if (page->index >= end_index) { unsigned offset_into_page = offset & (PAGE_CACHE_SIZE - 1); /* * Just skip the page if it is fully outside i_size, e.g. due * to a truncate operation that is in progress. */ if (page->index >= end_index + 1 || offset_into_page == 0) { ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ unlock_page(page); return 0; } In order to check if a page is fully outsids i_size or not, we can fix the code logic as below: if (page->index > end_index || (page->index == end_index && offset_into_page == 0)) Secondly, there still has another similar issue when calculating the end offset for mapping the filesystem blocks to the file blocks for delalloc. With the same tests to above, run unmount(8) will cause kernel panic if CONFIG_XFS_DEBUG is enabled: XFS: Assertion failed: XFS_FORCED_SHUTDOWN(ip->i_mount) || \ ip->i_delayed_blks == 0, file: fs/xfs/xfs_super.c, line: 964 kernel BUG at fs/xfs/xfs_message.c:108! invalid opcode: 0000 [#1] SMP task: edddc100 ti: ec6ee000 task.ti: ec6ee000 EIP: 0060:[<f83d87cb>] EFLAGS: 00010296 CPU: 1 EIP is at assfail+0x2b/0x30 [xfs] .............. Call Trace: [<f83d9cd4>] xfs_fs_destroy_inode+0x74/0x120 [xfs] [<c115ddf1>] destroy_inode+0x31/0x50 [<c115deff>] evict+0xef/0x170 [<c115dfb2>] dispose_list+0x32/0x40 [<c115ea3a>] evict_inodes+0xca/0xe0 [<c1149706>] generic_shutdown_super+0x46/0xd0 [<c11497b9>] kill_block_super+0x29/0x70 [<c1149a14>] deactivate_locked_super+0x44/0x70 [<c114a427>] deactivate_super+0x47/0x60 [<c1161c3d>] mntput_no_expire+0xcd/0x120 [<c1162ae8>] SyS_umount+0xa8/0x370 [<c1162dce>] SyS_oldumount+0x1e/0x20 [<c15be18d>] sysenter_do_call+0x12/0x28 That because the end_offset is evaluated to 0 which is the same reason to above, hence the mapping and covertion for dealloc file blocks to file system blocks did not happened. This patch just fixed both issues. Reported-by: Michael L. Semon <mlsemon35@gmail.com> Signed-off-by: Jie Liu <jeff.liu@oracle.com> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2014-05-20 05:24:26 +07:00
* The page index is less than the end_index, adjust the end_offset
* to the highest offset that this page should represent.
* -----------------------------------------------------
* | file mapping | <EOF> |
* -----------------------------------------------------
* | Page ... | Page N-2 | Page N-1 | Page N | |
* ^--------------------------------^----------|--------
* | desired writeback range | see else |
* ---------------------------------^------------------|
*/
offset = i_size_read(inode);
mm, fs: get rid of PAGE_CACHE_* and page_cache_{get,release} macros PAGE_CACHE_{SIZE,SHIFT,MASK,ALIGN} macros were introduced *long* time ago with promise that one day it will be possible to implement page cache with bigger chunks than PAGE_SIZE. This promise never materialized. And unlikely will. We have many places where PAGE_CACHE_SIZE assumed to be equal to PAGE_SIZE. And it's constant source of confusion on whether PAGE_CACHE_* or PAGE_* constant should be used in a particular case, especially on the border between fs and mm. Global switching to PAGE_CACHE_SIZE != PAGE_SIZE would cause to much breakage to be doable. Let's stop pretending that pages in page cache are special. They are not. The changes are pretty straight-forward: - <foo> << (PAGE_CACHE_SHIFT - PAGE_SHIFT) -> <foo>; - <foo> >> (PAGE_CACHE_SHIFT - PAGE_SHIFT) -> <foo>; - PAGE_CACHE_{SIZE,SHIFT,MASK,ALIGN} -> PAGE_{SIZE,SHIFT,MASK,ALIGN}; - page_cache_get() -> get_page(); - page_cache_release() -> put_page(); This patch contains automated changes generated with coccinelle using script below. For some reason, coccinelle doesn't patch header files. I've called spatch for them manually. The only adjustment after coccinelle is revert of changes to PAGE_CAHCE_ALIGN definition: we are going to drop it later. There are few places in the code where coccinelle didn't reach. I'll fix them manually in a separate patch. Comments and documentation also will be addressed with the separate patch. virtual patch @@ expression E; @@ - E << (PAGE_CACHE_SHIFT - PAGE_SHIFT) + E @@ expression E; @@ - E >> (PAGE_CACHE_SHIFT - PAGE_SHIFT) + E @@ @@ - PAGE_CACHE_SHIFT + PAGE_SHIFT @@ @@ - PAGE_CACHE_SIZE + PAGE_SIZE @@ @@ - PAGE_CACHE_MASK + PAGE_MASK @@ expression E; @@ - PAGE_CACHE_ALIGN(E) + PAGE_ALIGN(E) @@ expression E; @@ - page_cache_get(E) + get_page(E) @@ expression E; @@ - page_cache_release(E) + put_page(E) Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> Acked-by: Michal Hocko <mhocko@suse.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-04-01 19:29:47 +07:00
end_index = offset >> PAGE_SHIFT;
xfs: fix infinite loop at xfs_vm_writepage on 32bit system Write to a file with an offset greater than 16TB on 32-bit system and then trigger page write-back via sync(1) will cause task hang. # block_size=4096 # offset=$(((2**32 - 1) * $block_size)) # xfs_io -f -c "pwrite $offset $block_size" /storage/test_file # sync INFO: task sync:2590 blocked for more than 120 seconds. "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. sync D c1064a28 0 2590 2097 0x00000000 ..... Call Trace: [<c1064a28>] ? ttwu_do_wakeup+0x18/0x130 [<c1066d0e>] ? try_to_wake_up+0x1ce/0x220 [<c1066dbf>] ? wake_up_process+0x1f/0x40 [<c104fc2e>] ? wake_up_worker+0x1e/0x30 [<c15b6083>] schedule+0x23/0x60 [<c15b3c2d>] schedule_timeout+0x18d/0x1f0 [<c12a143e>] ? do_raw_spin_unlock+0x4e/0x90 [<c10515f1>] ? __queue_delayed_work+0x91/0x150 [<c12a12ef>] ? do_raw_spin_lock+0x3f/0x100 [<c12a143e>] ? do_raw_spin_unlock+0x4e/0x90 [<c15b5b5d>] wait_for_completion+0x7d/0xc0 [<c1066d60>] ? try_to_wake_up+0x220/0x220 [<c116a4d2>] sync_inodes_sb+0x92/0x180 [<c116fb05>] sync_inodes_one_sb+0x15/0x20 [<c114a8f8>] iterate_supers+0xb8/0xc0 [<c116faf0>] ? fdatawrite_one_bdev+0x20/0x20 [<c116fc21>] sys_sync+0x31/0x80 [<c15be18d>] sysenter_do_call+0x12/0x28 This issue can be triggered via xfstests/generic/308. The reason is that the end_index is unsigned long with maximum value '2^32-1=4294967295' on 32-bit platform, and the given offset cause it wrapped to 0, so that the following codes will repeat again and again until the task schedule time out: end_index = offset >> PAGE_CACHE_SHIFT; last_index = (offset - 1) >> PAGE_CACHE_SHIFT; if (page->index >= end_index) { unsigned offset_into_page = offset & (PAGE_CACHE_SIZE - 1); /* * Just skip the page if it is fully outside i_size, e.g. due * to a truncate operation that is in progress. */ if (page->index >= end_index + 1 || offset_into_page == 0) { ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ unlock_page(page); return 0; } In order to check if a page is fully outsids i_size or not, we can fix the code logic as below: if (page->index > end_index || (page->index == end_index && offset_into_page == 0)) Secondly, there still has another similar issue when calculating the end offset for mapping the filesystem blocks to the file blocks for delalloc. With the same tests to above, run unmount(8) will cause kernel panic if CONFIG_XFS_DEBUG is enabled: XFS: Assertion failed: XFS_FORCED_SHUTDOWN(ip->i_mount) || \ ip->i_delayed_blks == 0, file: fs/xfs/xfs_super.c, line: 964 kernel BUG at fs/xfs/xfs_message.c:108! invalid opcode: 0000 [#1] SMP task: edddc100 ti: ec6ee000 task.ti: ec6ee000 EIP: 0060:[<f83d87cb>] EFLAGS: 00010296 CPU: 1 EIP is at assfail+0x2b/0x30 [xfs] .............. Call Trace: [<f83d9cd4>] xfs_fs_destroy_inode+0x74/0x120 [xfs] [<c115ddf1>] destroy_inode+0x31/0x50 [<c115deff>] evict+0xef/0x170 [<c115dfb2>] dispose_list+0x32/0x40 [<c115ea3a>] evict_inodes+0xca/0xe0 [<c1149706>] generic_shutdown_super+0x46/0xd0 [<c11497b9>] kill_block_super+0x29/0x70 [<c1149a14>] deactivate_locked_super+0x44/0x70 [<c114a427>] deactivate_super+0x47/0x60 [<c1161c3d>] mntput_no_expire+0xcd/0x120 [<c1162ae8>] SyS_umount+0xa8/0x370 [<c1162dce>] SyS_oldumount+0x1e/0x20 [<c15be18d>] sysenter_do_call+0x12/0x28 That because the end_offset is evaluated to 0 which is the same reason to above, hence the mapping and covertion for dealloc file blocks to file system blocks did not happened. This patch just fixed both issues. Reported-by: Michael L. Semon <mlsemon35@gmail.com> Signed-off-by: Jie Liu <jeff.liu@oracle.com> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2014-05-20 05:24:26 +07:00
if (page->index < end_index)
mm, fs: get rid of PAGE_CACHE_* and page_cache_{get,release} macros PAGE_CACHE_{SIZE,SHIFT,MASK,ALIGN} macros were introduced *long* time ago with promise that one day it will be possible to implement page cache with bigger chunks than PAGE_SIZE. This promise never materialized. And unlikely will. We have many places where PAGE_CACHE_SIZE assumed to be equal to PAGE_SIZE. And it's constant source of confusion on whether PAGE_CACHE_* or PAGE_* constant should be used in a particular case, especially on the border between fs and mm. Global switching to PAGE_CACHE_SIZE != PAGE_SIZE would cause to much breakage to be doable. Let's stop pretending that pages in page cache are special. They are not. The changes are pretty straight-forward: - <foo> << (PAGE_CACHE_SHIFT - PAGE_SHIFT) -> <foo>; - <foo> >> (PAGE_CACHE_SHIFT - PAGE_SHIFT) -> <foo>; - PAGE_CACHE_{SIZE,SHIFT,MASK,ALIGN} -> PAGE_{SIZE,SHIFT,MASK,ALIGN}; - page_cache_get() -> get_page(); - page_cache_release() -> put_page(); This patch contains automated changes generated with coccinelle using script below. For some reason, coccinelle doesn't patch header files. I've called spatch for them manually. The only adjustment after coccinelle is revert of changes to PAGE_CAHCE_ALIGN definition: we are going to drop it later. There are few places in the code where coccinelle didn't reach. I'll fix them manually in a separate patch. Comments and documentation also will be addressed with the separate patch. virtual patch @@ expression E; @@ - E << (PAGE_CACHE_SHIFT - PAGE_SHIFT) + E @@ expression E; @@ - E >> (PAGE_CACHE_SHIFT - PAGE_SHIFT) + E @@ @@ - PAGE_CACHE_SHIFT + PAGE_SHIFT @@ @@ - PAGE_CACHE_SIZE + PAGE_SIZE @@ @@ - PAGE_CACHE_MASK + PAGE_MASK @@ expression E; @@ - PAGE_CACHE_ALIGN(E) + PAGE_ALIGN(E) @@ expression E; @@ - page_cache_get(E) + get_page(E) @@ expression E; @@ - page_cache_release(E) + put_page(E) Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> Acked-by: Michal Hocko <mhocko@suse.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-04-01 19:29:47 +07:00
end_offset = (xfs_off_t)(page->index + 1) << PAGE_SHIFT;
xfs: fix infinite loop at xfs_vm_writepage on 32bit system Write to a file with an offset greater than 16TB on 32-bit system and then trigger page write-back via sync(1) will cause task hang. # block_size=4096 # offset=$(((2**32 - 1) * $block_size)) # xfs_io -f -c "pwrite $offset $block_size" /storage/test_file # sync INFO: task sync:2590 blocked for more than 120 seconds. "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. sync D c1064a28 0 2590 2097 0x00000000 ..... Call Trace: [<c1064a28>] ? ttwu_do_wakeup+0x18/0x130 [<c1066d0e>] ? try_to_wake_up+0x1ce/0x220 [<c1066dbf>] ? wake_up_process+0x1f/0x40 [<c104fc2e>] ? wake_up_worker+0x1e/0x30 [<c15b6083>] schedule+0x23/0x60 [<c15b3c2d>] schedule_timeout+0x18d/0x1f0 [<c12a143e>] ? do_raw_spin_unlock+0x4e/0x90 [<c10515f1>] ? __queue_delayed_work+0x91/0x150 [<c12a12ef>] ? do_raw_spin_lock+0x3f/0x100 [<c12a143e>] ? do_raw_spin_unlock+0x4e/0x90 [<c15b5b5d>] wait_for_completion+0x7d/0xc0 [<c1066d60>] ? try_to_wake_up+0x220/0x220 [<c116a4d2>] sync_inodes_sb+0x92/0x180 [<c116fb05>] sync_inodes_one_sb+0x15/0x20 [<c114a8f8>] iterate_supers+0xb8/0xc0 [<c116faf0>] ? fdatawrite_one_bdev+0x20/0x20 [<c116fc21>] sys_sync+0x31/0x80 [<c15be18d>] sysenter_do_call+0x12/0x28 This issue can be triggered via xfstests/generic/308. The reason is that the end_index is unsigned long with maximum value '2^32-1=4294967295' on 32-bit platform, and the given offset cause it wrapped to 0, so that the following codes will repeat again and again until the task schedule time out: end_index = offset >> PAGE_CACHE_SHIFT; last_index = (offset - 1) >> PAGE_CACHE_SHIFT; if (page->index >= end_index) { unsigned offset_into_page = offset & (PAGE_CACHE_SIZE - 1); /* * Just skip the page if it is fully outside i_size, e.g. due * to a truncate operation that is in progress. */ if (page->index >= end_index + 1 || offset_into_page == 0) { ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ unlock_page(page); return 0; } In order to check if a page is fully outsids i_size or not, we can fix the code logic as below: if (page->index > end_index || (page->index == end_index && offset_into_page == 0)) Secondly, there still has another similar issue when calculating the end offset for mapping the filesystem blocks to the file blocks for delalloc. With the same tests to above, run unmount(8) will cause kernel panic if CONFIG_XFS_DEBUG is enabled: XFS: Assertion failed: XFS_FORCED_SHUTDOWN(ip->i_mount) || \ ip->i_delayed_blks == 0, file: fs/xfs/xfs_super.c, line: 964 kernel BUG at fs/xfs/xfs_message.c:108! invalid opcode: 0000 [#1] SMP task: edddc100 ti: ec6ee000 task.ti: ec6ee000 EIP: 0060:[<f83d87cb>] EFLAGS: 00010296 CPU: 1 EIP is at assfail+0x2b/0x30 [xfs] .............. Call Trace: [<f83d9cd4>] xfs_fs_destroy_inode+0x74/0x120 [xfs] [<c115ddf1>] destroy_inode+0x31/0x50 [<c115deff>] evict+0xef/0x170 [<c115dfb2>] dispose_list+0x32/0x40 [<c115ea3a>] evict_inodes+0xca/0xe0 [<c1149706>] generic_shutdown_super+0x46/0xd0 [<c11497b9>] kill_block_super+0x29/0x70 [<c1149a14>] deactivate_locked_super+0x44/0x70 [<c114a427>] deactivate_super+0x47/0x60 [<c1161c3d>] mntput_no_expire+0xcd/0x120 [<c1162ae8>] SyS_umount+0xa8/0x370 [<c1162dce>] SyS_oldumount+0x1e/0x20 [<c15be18d>] sysenter_do_call+0x12/0x28 That because the end_offset is evaluated to 0 which is the same reason to above, hence the mapping and covertion for dealloc file blocks to file system blocks did not happened. This patch just fixed both issues. Reported-by: Michael L. Semon <mlsemon35@gmail.com> Signed-off-by: Jie Liu <jeff.liu@oracle.com> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2014-05-20 05:24:26 +07:00
else {
/*
* Check whether the page to write out is beyond or straddles
* i_size or not.
* -------------------------------------------------------
* | file mapping | <EOF> |
* -------------------------------------------------------
* | Page ... | Page N-2 | Page N-1 | Page N | Beyond |
* ^--------------------------------^-----------|---------
* | | Straddles |
* ---------------------------------^-----------|--------|
*/
mm, fs: get rid of PAGE_CACHE_* and page_cache_{get,release} macros PAGE_CACHE_{SIZE,SHIFT,MASK,ALIGN} macros were introduced *long* time ago with promise that one day it will be possible to implement page cache with bigger chunks than PAGE_SIZE. This promise never materialized. And unlikely will. We have many places where PAGE_CACHE_SIZE assumed to be equal to PAGE_SIZE. And it's constant source of confusion on whether PAGE_CACHE_* or PAGE_* constant should be used in a particular case, especially on the border between fs and mm. Global switching to PAGE_CACHE_SIZE != PAGE_SIZE would cause to much breakage to be doable. Let's stop pretending that pages in page cache are special. They are not. The changes are pretty straight-forward: - <foo> << (PAGE_CACHE_SHIFT - PAGE_SHIFT) -> <foo>; - <foo> >> (PAGE_CACHE_SHIFT - PAGE_SHIFT) -> <foo>; - PAGE_CACHE_{SIZE,SHIFT,MASK,ALIGN} -> PAGE_{SIZE,SHIFT,MASK,ALIGN}; - page_cache_get() -> get_page(); - page_cache_release() -> put_page(); This patch contains automated changes generated with coccinelle using script below. For some reason, coccinelle doesn't patch header files. I've called spatch for them manually. The only adjustment after coccinelle is revert of changes to PAGE_CAHCE_ALIGN definition: we are going to drop it later. There are few places in the code where coccinelle didn't reach. I'll fix them manually in a separate patch. Comments and documentation also will be addressed with the separate patch. virtual patch @@ expression E; @@ - E << (PAGE_CACHE_SHIFT - PAGE_SHIFT) + E @@ expression E; @@ - E >> (PAGE_CACHE_SHIFT - PAGE_SHIFT) + E @@ @@ - PAGE_CACHE_SHIFT + PAGE_SHIFT @@ @@ - PAGE_CACHE_SIZE + PAGE_SIZE @@ @@ - PAGE_CACHE_MASK + PAGE_MASK @@ expression E; @@ - PAGE_CACHE_ALIGN(E) + PAGE_ALIGN(E) @@ expression E; @@ - page_cache_get(E) + get_page(E) @@ expression E; @@ - page_cache_release(E) + put_page(E) Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> Acked-by: Michal Hocko <mhocko@suse.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-04-01 19:29:47 +07:00
unsigned offset_into_page = offset & (PAGE_SIZE - 1);
/*
* Skip the page if it is fully outside i_size, e.g. due to a
* truncate operation that is in progress. We must redirty the
* page so that reclaim stops reclaiming it. Otherwise
* xfs_vm_releasepage() is called on it and gets confused.
xfs: fix infinite loop at xfs_vm_writepage on 32bit system Write to a file with an offset greater than 16TB on 32-bit system and then trigger page write-back via sync(1) will cause task hang. # block_size=4096 # offset=$(((2**32 - 1) * $block_size)) # xfs_io -f -c "pwrite $offset $block_size" /storage/test_file # sync INFO: task sync:2590 blocked for more than 120 seconds. "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. sync D c1064a28 0 2590 2097 0x00000000 ..... Call Trace: [<c1064a28>] ? ttwu_do_wakeup+0x18/0x130 [<c1066d0e>] ? try_to_wake_up+0x1ce/0x220 [<c1066dbf>] ? wake_up_process+0x1f/0x40 [<c104fc2e>] ? wake_up_worker+0x1e/0x30 [<c15b6083>] schedule+0x23/0x60 [<c15b3c2d>] schedule_timeout+0x18d/0x1f0 [<c12a143e>] ? do_raw_spin_unlock+0x4e/0x90 [<c10515f1>] ? __queue_delayed_work+0x91/0x150 [<c12a12ef>] ? do_raw_spin_lock+0x3f/0x100 [<c12a143e>] ? do_raw_spin_unlock+0x4e/0x90 [<c15b5b5d>] wait_for_completion+0x7d/0xc0 [<c1066d60>] ? try_to_wake_up+0x220/0x220 [<c116a4d2>] sync_inodes_sb+0x92/0x180 [<c116fb05>] sync_inodes_one_sb+0x15/0x20 [<c114a8f8>] iterate_supers+0xb8/0xc0 [<c116faf0>] ? fdatawrite_one_bdev+0x20/0x20 [<c116fc21>] sys_sync+0x31/0x80 [<c15be18d>] sysenter_do_call+0x12/0x28 This issue can be triggered via xfstests/generic/308. The reason is that the end_index is unsigned long with maximum value '2^32-1=4294967295' on 32-bit platform, and the given offset cause it wrapped to 0, so that the following codes will repeat again and again until the task schedule time out: end_index = offset >> PAGE_CACHE_SHIFT; last_index = (offset - 1) >> PAGE_CACHE_SHIFT; if (page->index >= end_index) { unsigned offset_into_page = offset & (PAGE_CACHE_SIZE - 1); /* * Just skip the page if it is fully outside i_size, e.g. due * to a truncate operation that is in progress. */ if (page->index >= end_index + 1 || offset_into_page == 0) { ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ unlock_page(page); return 0; } In order to check if a page is fully outsids i_size or not, we can fix the code logic as below: if (page->index > end_index || (page->index == end_index && offset_into_page == 0)) Secondly, there still has another similar issue when calculating the end offset for mapping the filesystem blocks to the file blocks for delalloc. With the same tests to above, run unmount(8) will cause kernel panic if CONFIG_XFS_DEBUG is enabled: XFS: Assertion failed: XFS_FORCED_SHUTDOWN(ip->i_mount) || \ ip->i_delayed_blks == 0, file: fs/xfs/xfs_super.c, line: 964 kernel BUG at fs/xfs/xfs_message.c:108! invalid opcode: 0000 [#1] SMP task: edddc100 ti: ec6ee000 task.ti: ec6ee000 EIP: 0060:[<f83d87cb>] EFLAGS: 00010296 CPU: 1 EIP is at assfail+0x2b/0x30 [xfs] .............. Call Trace: [<f83d9cd4>] xfs_fs_destroy_inode+0x74/0x120 [xfs] [<c115ddf1>] destroy_inode+0x31/0x50 [<c115deff>] evict+0xef/0x170 [<c115dfb2>] dispose_list+0x32/0x40 [<c115ea3a>] evict_inodes+0xca/0xe0 [<c1149706>] generic_shutdown_super+0x46/0xd0 [<c11497b9>] kill_block_super+0x29/0x70 [<c1149a14>] deactivate_locked_super+0x44/0x70 [<c114a427>] deactivate_super+0x47/0x60 [<c1161c3d>] mntput_no_expire+0xcd/0x120 [<c1162ae8>] SyS_umount+0xa8/0x370 [<c1162dce>] SyS_oldumount+0x1e/0x20 [<c15be18d>] sysenter_do_call+0x12/0x28 That because the end_offset is evaluated to 0 which is the same reason to above, hence the mapping and covertion for dealloc file blocks to file system blocks did not happened. This patch just fixed both issues. Reported-by: Michael L. Semon <mlsemon35@gmail.com> Signed-off-by: Jie Liu <jeff.liu@oracle.com> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2014-05-20 05:24:26 +07:00
*
* Note that the end_index is unsigned long, it would overflow
* if the given offset is greater than 16TB on 32-bit system
* and if we do check the page is fully outside i_size or not
* via "if (page->index >= end_index + 1)" as "end_index + 1"
* will be evaluated to 0. Hence this page will be redirtied
* and be written out repeatedly which would result in an
* infinite loop, the user program that perform this operation
* will hang. Instead, we can verify this situation by checking
* if the page to write is totally beyond the i_size or if it's
* offset is just equal to the EOF.
*/
xfs: fix infinite loop at xfs_vm_writepage on 32bit system Write to a file with an offset greater than 16TB on 32-bit system and then trigger page write-back via sync(1) will cause task hang. # block_size=4096 # offset=$(((2**32 - 1) * $block_size)) # xfs_io -f -c "pwrite $offset $block_size" /storage/test_file # sync INFO: task sync:2590 blocked for more than 120 seconds. "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. sync D c1064a28 0 2590 2097 0x00000000 ..... Call Trace: [<c1064a28>] ? ttwu_do_wakeup+0x18/0x130 [<c1066d0e>] ? try_to_wake_up+0x1ce/0x220 [<c1066dbf>] ? wake_up_process+0x1f/0x40 [<c104fc2e>] ? wake_up_worker+0x1e/0x30 [<c15b6083>] schedule+0x23/0x60 [<c15b3c2d>] schedule_timeout+0x18d/0x1f0 [<c12a143e>] ? do_raw_spin_unlock+0x4e/0x90 [<c10515f1>] ? __queue_delayed_work+0x91/0x150 [<c12a12ef>] ? do_raw_spin_lock+0x3f/0x100 [<c12a143e>] ? do_raw_spin_unlock+0x4e/0x90 [<c15b5b5d>] wait_for_completion+0x7d/0xc0 [<c1066d60>] ? try_to_wake_up+0x220/0x220 [<c116a4d2>] sync_inodes_sb+0x92/0x180 [<c116fb05>] sync_inodes_one_sb+0x15/0x20 [<c114a8f8>] iterate_supers+0xb8/0xc0 [<c116faf0>] ? fdatawrite_one_bdev+0x20/0x20 [<c116fc21>] sys_sync+0x31/0x80 [<c15be18d>] sysenter_do_call+0x12/0x28 This issue can be triggered via xfstests/generic/308. The reason is that the end_index is unsigned long with maximum value '2^32-1=4294967295' on 32-bit platform, and the given offset cause it wrapped to 0, so that the following codes will repeat again and again until the task schedule time out: end_index = offset >> PAGE_CACHE_SHIFT; last_index = (offset - 1) >> PAGE_CACHE_SHIFT; if (page->index >= end_index) { unsigned offset_into_page = offset & (PAGE_CACHE_SIZE - 1); /* * Just skip the page if it is fully outside i_size, e.g. due * to a truncate operation that is in progress. */ if (page->index >= end_index + 1 || offset_into_page == 0) { ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ unlock_page(page); return 0; } In order to check if a page is fully outsids i_size or not, we can fix the code logic as below: if (page->index > end_index || (page->index == end_index && offset_into_page == 0)) Secondly, there still has another similar issue when calculating the end offset for mapping the filesystem blocks to the file blocks for delalloc. With the same tests to above, run unmount(8) will cause kernel panic if CONFIG_XFS_DEBUG is enabled: XFS: Assertion failed: XFS_FORCED_SHUTDOWN(ip->i_mount) || \ ip->i_delayed_blks == 0, file: fs/xfs/xfs_super.c, line: 964 kernel BUG at fs/xfs/xfs_message.c:108! invalid opcode: 0000 [#1] SMP task: edddc100 ti: ec6ee000 task.ti: ec6ee000 EIP: 0060:[<f83d87cb>] EFLAGS: 00010296 CPU: 1 EIP is at assfail+0x2b/0x30 [xfs] .............. Call Trace: [<f83d9cd4>] xfs_fs_destroy_inode+0x74/0x120 [xfs] [<c115ddf1>] destroy_inode+0x31/0x50 [<c115deff>] evict+0xef/0x170 [<c115dfb2>] dispose_list+0x32/0x40 [<c115ea3a>] evict_inodes+0xca/0xe0 [<c1149706>] generic_shutdown_super+0x46/0xd0 [<c11497b9>] kill_block_super+0x29/0x70 [<c1149a14>] deactivate_locked_super+0x44/0x70 [<c114a427>] deactivate_super+0x47/0x60 [<c1161c3d>] mntput_no_expire+0xcd/0x120 [<c1162ae8>] SyS_umount+0xa8/0x370 [<c1162dce>] SyS_oldumount+0x1e/0x20 [<c15be18d>] sysenter_do_call+0x12/0x28 That because the end_offset is evaluated to 0 which is the same reason to above, hence the mapping and covertion for dealloc file blocks to file system blocks did not happened. This patch just fixed both issues. Reported-by: Michael L. Semon <mlsemon35@gmail.com> Signed-off-by: Jie Liu <jeff.liu@oracle.com> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2014-05-20 05:24:26 +07:00
if (page->index > end_index ||
(page->index == end_index && offset_into_page == 0))
goto redirty;
/*
* The page straddles i_size. It must be zeroed out on each
* and every writepage invocation because it may be mmapped.
* "A file is mapped in multiples of the page size. For a file
xfs: fix infinite loop at xfs_vm_writepage on 32bit system Write to a file with an offset greater than 16TB on 32-bit system and then trigger page write-back via sync(1) will cause task hang. # block_size=4096 # offset=$(((2**32 - 1) * $block_size)) # xfs_io -f -c "pwrite $offset $block_size" /storage/test_file # sync INFO: task sync:2590 blocked for more than 120 seconds. "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. sync D c1064a28 0 2590 2097 0x00000000 ..... Call Trace: [<c1064a28>] ? ttwu_do_wakeup+0x18/0x130 [<c1066d0e>] ? try_to_wake_up+0x1ce/0x220 [<c1066dbf>] ? wake_up_process+0x1f/0x40 [<c104fc2e>] ? wake_up_worker+0x1e/0x30 [<c15b6083>] schedule+0x23/0x60 [<c15b3c2d>] schedule_timeout+0x18d/0x1f0 [<c12a143e>] ? do_raw_spin_unlock+0x4e/0x90 [<c10515f1>] ? __queue_delayed_work+0x91/0x150 [<c12a12ef>] ? do_raw_spin_lock+0x3f/0x100 [<c12a143e>] ? do_raw_spin_unlock+0x4e/0x90 [<c15b5b5d>] wait_for_completion+0x7d/0xc0 [<c1066d60>] ? try_to_wake_up+0x220/0x220 [<c116a4d2>] sync_inodes_sb+0x92/0x180 [<c116fb05>] sync_inodes_one_sb+0x15/0x20 [<c114a8f8>] iterate_supers+0xb8/0xc0 [<c116faf0>] ? fdatawrite_one_bdev+0x20/0x20 [<c116fc21>] sys_sync+0x31/0x80 [<c15be18d>] sysenter_do_call+0x12/0x28 This issue can be triggered via xfstests/generic/308. The reason is that the end_index is unsigned long with maximum value '2^32-1=4294967295' on 32-bit platform, and the given offset cause it wrapped to 0, so that the following codes will repeat again and again until the task schedule time out: end_index = offset >> PAGE_CACHE_SHIFT; last_index = (offset - 1) >> PAGE_CACHE_SHIFT; if (page->index >= end_index) { unsigned offset_into_page = offset & (PAGE_CACHE_SIZE - 1); /* * Just skip the page if it is fully outside i_size, e.g. due * to a truncate operation that is in progress. */ if (page->index >= end_index + 1 || offset_into_page == 0) { ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ unlock_page(page); return 0; } In order to check if a page is fully outsids i_size or not, we can fix the code logic as below: if (page->index > end_index || (page->index == end_index && offset_into_page == 0)) Secondly, there still has another similar issue when calculating the end offset for mapping the filesystem blocks to the file blocks for delalloc. With the same tests to above, run unmount(8) will cause kernel panic if CONFIG_XFS_DEBUG is enabled: XFS: Assertion failed: XFS_FORCED_SHUTDOWN(ip->i_mount) || \ ip->i_delayed_blks == 0, file: fs/xfs/xfs_super.c, line: 964 kernel BUG at fs/xfs/xfs_message.c:108! invalid opcode: 0000 [#1] SMP task: edddc100 ti: ec6ee000 task.ti: ec6ee000 EIP: 0060:[<f83d87cb>] EFLAGS: 00010296 CPU: 1 EIP is at assfail+0x2b/0x30 [xfs] .............. Call Trace: [<f83d9cd4>] xfs_fs_destroy_inode+0x74/0x120 [xfs] [<c115ddf1>] destroy_inode+0x31/0x50 [<c115deff>] evict+0xef/0x170 [<c115dfb2>] dispose_list+0x32/0x40 [<c115ea3a>] evict_inodes+0xca/0xe0 [<c1149706>] generic_shutdown_super+0x46/0xd0 [<c11497b9>] kill_block_super+0x29/0x70 [<c1149a14>] deactivate_locked_super+0x44/0x70 [<c114a427>] deactivate_super+0x47/0x60 [<c1161c3d>] mntput_no_expire+0xcd/0x120 [<c1162ae8>] SyS_umount+0xa8/0x370 [<c1162dce>] SyS_oldumount+0x1e/0x20 [<c15be18d>] sysenter_do_call+0x12/0x28 That because the end_offset is evaluated to 0 which is the same reason to above, hence the mapping and covertion for dealloc file blocks to file system blocks did not happened. This patch just fixed both issues. Reported-by: Michael L. Semon <mlsemon35@gmail.com> Signed-off-by: Jie Liu <jeff.liu@oracle.com> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2014-05-20 05:24:26 +07:00
* that is not a multiple of the page size, the remaining
* memory is zeroed when mapped, and writes to that region are
* not written out to the file."
*/
mm, fs: get rid of PAGE_CACHE_* and page_cache_{get,release} macros PAGE_CACHE_{SIZE,SHIFT,MASK,ALIGN} macros were introduced *long* time ago with promise that one day it will be possible to implement page cache with bigger chunks than PAGE_SIZE. This promise never materialized. And unlikely will. We have many places where PAGE_CACHE_SIZE assumed to be equal to PAGE_SIZE. And it's constant source of confusion on whether PAGE_CACHE_* or PAGE_* constant should be used in a particular case, especially on the border between fs and mm. Global switching to PAGE_CACHE_SIZE != PAGE_SIZE would cause to much breakage to be doable. Let's stop pretending that pages in page cache are special. They are not. The changes are pretty straight-forward: - <foo> << (PAGE_CACHE_SHIFT - PAGE_SHIFT) -> <foo>; - <foo> >> (PAGE_CACHE_SHIFT - PAGE_SHIFT) -> <foo>; - PAGE_CACHE_{SIZE,SHIFT,MASK,ALIGN} -> PAGE_{SIZE,SHIFT,MASK,ALIGN}; - page_cache_get() -> get_page(); - page_cache_release() -> put_page(); This patch contains automated changes generated with coccinelle using script below. For some reason, coccinelle doesn't patch header files. I've called spatch for them manually. The only adjustment after coccinelle is revert of changes to PAGE_CAHCE_ALIGN definition: we are going to drop it later. There are few places in the code where coccinelle didn't reach. I'll fix them manually in a separate patch. Comments and documentation also will be addressed with the separate patch. virtual patch @@ expression E; @@ - E << (PAGE_CACHE_SHIFT - PAGE_SHIFT) + E @@ expression E; @@ - E >> (PAGE_CACHE_SHIFT - PAGE_SHIFT) + E @@ @@ - PAGE_CACHE_SHIFT + PAGE_SHIFT @@ @@ - PAGE_CACHE_SIZE + PAGE_SIZE @@ @@ - PAGE_CACHE_MASK + PAGE_MASK @@ expression E; @@ - PAGE_CACHE_ALIGN(E) + PAGE_ALIGN(E) @@ expression E; @@ - page_cache_get(E) + get_page(E) @@ expression E; @@ - page_cache_release(E) + put_page(E) Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> Acked-by: Michal Hocko <mhocko@suse.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-04-01 19:29:47 +07:00
zero_user_segment(page, offset_into_page, PAGE_SIZE);
xfs: fix infinite loop at xfs_vm_writepage on 32bit system Write to a file with an offset greater than 16TB on 32-bit system and then trigger page write-back via sync(1) will cause task hang. # block_size=4096 # offset=$(((2**32 - 1) * $block_size)) # xfs_io -f -c "pwrite $offset $block_size" /storage/test_file # sync INFO: task sync:2590 blocked for more than 120 seconds. "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. sync D c1064a28 0 2590 2097 0x00000000 ..... Call Trace: [<c1064a28>] ? ttwu_do_wakeup+0x18/0x130 [<c1066d0e>] ? try_to_wake_up+0x1ce/0x220 [<c1066dbf>] ? wake_up_process+0x1f/0x40 [<c104fc2e>] ? wake_up_worker+0x1e/0x30 [<c15b6083>] schedule+0x23/0x60 [<c15b3c2d>] schedule_timeout+0x18d/0x1f0 [<c12a143e>] ? do_raw_spin_unlock+0x4e/0x90 [<c10515f1>] ? __queue_delayed_work+0x91/0x150 [<c12a12ef>] ? do_raw_spin_lock+0x3f/0x100 [<c12a143e>] ? do_raw_spin_unlock+0x4e/0x90 [<c15b5b5d>] wait_for_completion+0x7d/0xc0 [<c1066d60>] ? try_to_wake_up+0x220/0x220 [<c116a4d2>] sync_inodes_sb+0x92/0x180 [<c116fb05>] sync_inodes_one_sb+0x15/0x20 [<c114a8f8>] iterate_supers+0xb8/0xc0 [<c116faf0>] ? fdatawrite_one_bdev+0x20/0x20 [<c116fc21>] sys_sync+0x31/0x80 [<c15be18d>] sysenter_do_call+0x12/0x28 This issue can be triggered via xfstests/generic/308. The reason is that the end_index is unsigned long with maximum value '2^32-1=4294967295' on 32-bit platform, and the given offset cause it wrapped to 0, so that the following codes will repeat again and again until the task schedule time out: end_index = offset >> PAGE_CACHE_SHIFT; last_index = (offset - 1) >> PAGE_CACHE_SHIFT; if (page->index >= end_index) { unsigned offset_into_page = offset & (PAGE_CACHE_SIZE - 1); /* * Just skip the page if it is fully outside i_size, e.g. due * to a truncate operation that is in progress. */ if (page->index >= end_index + 1 || offset_into_page == 0) { ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ unlock_page(page); return 0; } In order to check if a page is fully outsids i_size or not, we can fix the code logic as below: if (page->index > end_index || (page->index == end_index && offset_into_page == 0)) Secondly, there still has another similar issue when calculating the end offset for mapping the filesystem blocks to the file blocks for delalloc. With the same tests to above, run unmount(8) will cause kernel panic if CONFIG_XFS_DEBUG is enabled: XFS: Assertion failed: XFS_FORCED_SHUTDOWN(ip->i_mount) || \ ip->i_delayed_blks == 0, file: fs/xfs/xfs_super.c, line: 964 kernel BUG at fs/xfs/xfs_message.c:108! invalid opcode: 0000 [#1] SMP task: edddc100 ti: ec6ee000 task.ti: ec6ee000 EIP: 0060:[<f83d87cb>] EFLAGS: 00010296 CPU: 1 EIP is at assfail+0x2b/0x30 [xfs] .............. Call Trace: [<f83d9cd4>] xfs_fs_destroy_inode+0x74/0x120 [xfs] [<c115ddf1>] destroy_inode+0x31/0x50 [<c115deff>] evict+0xef/0x170 [<c115dfb2>] dispose_list+0x32/0x40 [<c115ea3a>] evict_inodes+0xca/0xe0 [<c1149706>] generic_shutdown_super+0x46/0xd0 [<c11497b9>] kill_block_super+0x29/0x70 [<c1149a14>] deactivate_locked_super+0x44/0x70 [<c114a427>] deactivate_super+0x47/0x60 [<c1161c3d>] mntput_no_expire+0xcd/0x120 [<c1162ae8>] SyS_umount+0xa8/0x370 [<c1162dce>] SyS_oldumount+0x1e/0x20 [<c15be18d>] sysenter_do_call+0x12/0x28 That because the end_offset is evaluated to 0 which is the same reason to above, hence the mapping and covertion for dealloc file blocks to file system blocks did not happened. This patch just fixed both issues. Reported-by: Michael L. Semon <mlsemon35@gmail.com> Signed-off-by: Jie Liu <jeff.liu@oracle.com> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2014-05-20 05:24:26 +07:00
/* Adjust the end_offset to the end of file */
end_offset = offset;
}
return xfs_writepage_map(wpc, wbc, inode, page, end_offset);
redirty:
redirty_page_for_writepage(wbc, page);
unlock_page(page);
return 0;
}
xfs: Introduce writeback context for writepages xfs_vm_writepages() calls generic_writepages to writeback a range of a file, but then xfs_vm_writepage() clusters pages itself as it does not have any context it can pass between->writepage calls from __write_cache_pages(). Introduce a writeback context for xfs_vm_writepages() and call __write_cache_pages directly with our own writepage callback so that we can pass that context to each writepage invocation. This encapsulates the current mapping, whether it is valid or not, the current ioend and it's IO type and the ioend chain being built. This requires us to move the ioend submission up to the level where the writepage context is declared. This does mean we do not submit IO until we packaged the entire writeback range, but with the block plugging in the writepages call this is the way IO is submitted, anyway. It also means that we need to handle discontiguous page ranges. If the pages sent down by write_cache_pages to the writepage callback are discontiguous, we need to detect this and put each discontiguous page range into individual ioends. This is needed to ensure that the ioend accurately represents the range of the file that it covers so that file size updates during IO completion set the size correctly. Failure to take into account the discontiguous ranges results in files being too small when writeback patterns are non-sequential. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-02-15 13:21:19 +07:00
STATIC int
xfs_vm_writepage(
struct page *page,
struct writeback_control *wbc)
{
struct xfs_writepage_ctx wpc = { };
xfs: Introduce writeback context for writepages xfs_vm_writepages() calls generic_writepages to writeback a range of a file, but then xfs_vm_writepage() clusters pages itself as it does not have any context it can pass between->writepage calls from __write_cache_pages(). Introduce a writeback context for xfs_vm_writepages() and call __write_cache_pages directly with our own writepage callback so that we can pass that context to each writepage invocation. This encapsulates the current mapping, whether it is valid or not, the current ioend and it's IO type and the ioend chain being built. This requires us to move the ioend submission up to the level where the writepage context is declared. This does mean we do not submit IO until we packaged the entire writeback range, but with the block plugging in the writepages call this is the way IO is submitted, anyway. It also means that we need to handle discontiguous page ranges. If the pages sent down by write_cache_pages to the writepage callback are discontiguous, we need to detect this and put each discontiguous page range into individual ioends. This is needed to ensure that the ioend accurately represents the range of the file that it covers so that file size updates during IO completion set the size correctly. Failure to take into account the discontiguous ranges results in files being too small when writeback patterns are non-sequential. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-02-15 13:21:19 +07:00
int ret;
ret = xfs_do_writepage(page, wbc, &wpc);
xfs: don't chain ioends during writepage submission Currently we can build a long ioend chain during ->writepages that gets attached to the writepage context. IO submission only then occurs when we finish all the writepage processing. This means we can have many ioends allocated and pending, and this violates the mempool guarantees that we need to give about forwards progress. i.e. we really should only have one ioend being built at a time, otherwise we may drain the mempool trying to allocate a new ioend and that blocks submission, completion and freeing of ioends that are already in progress. To prevent this situation from happening, we need to submit ioends for IO as soon as they are ready for dispatch rather than queuing them for later submission. This means the ioends have bios built immediately and they get queued on any plug that is current active. Hence if we schedule away from writeback, the ioends that have been built will make forwards progress due to the plug flushing on context switch. This will also prevent context switches from creating unnecessary IO submission latency. We can't completely avoid having nested IO allocation - when we have a block size smaller than a page size, we still need to hold the ioend submission until after we have marked the current page dirty. Hence we may need multiple ioends to be held while the current page is completely mapped and made ready for IO dispatch. We cannot avoid this problem - the current code already has this ioend chaining within a page so we can mostly ignore that it occurs. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-02-15 13:23:12 +07:00
if (wpc.ioend)
ret = xfs_submit_ioend(wbc, wpc.ioend, ret);
return ret;
xfs: Introduce writeback context for writepages xfs_vm_writepages() calls generic_writepages to writeback a range of a file, but then xfs_vm_writepage() clusters pages itself as it does not have any context it can pass between->writepage calls from __write_cache_pages(). Introduce a writeback context for xfs_vm_writepages() and call __write_cache_pages directly with our own writepage callback so that we can pass that context to each writepage invocation. This encapsulates the current mapping, whether it is valid or not, the current ioend and it's IO type and the ioend chain being built. This requires us to move the ioend submission up to the level where the writepage context is declared. This does mean we do not submit IO until we packaged the entire writeback range, but with the block plugging in the writepages call this is the way IO is submitted, anyway. It also means that we need to handle discontiguous page ranges. If the pages sent down by write_cache_pages to the writepage callback are discontiguous, we need to detect this and put each discontiguous page range into individual ioends. This is needed to ensure that the ioend accurately represents the range of the file that it covers so that file size updates during IO completion set the size correctly. Failure to take into account the discontiguous ranges results in files being too small when writeback patterns are non-sequential. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-02-15 13:21:19 +07:00
}
STATIC int
xfs_vm_writepages(
struct address_space *mapping,
struct writeback_control *wbc)
{
struct xfs_writepage_ctx wpc = { };
xfs: Introduce writeback context for writepages xfs_vm_writepages() calls generic_writepages to writeback a range of a file, but then xfs_vm_writepage() clusters pages itself as it does not have any context it can pass between->writepage calls from __write_cache_pages(). Introduce a writeback context for xfs_vm_writepages() and call __write_cache_pages directly with our own writepage callback so that we can pass that context to each writepage invocation. This encapsulates the current mapping, whether it is valid or not, the current ioend and it's IO type and the ioend chain being built. This requires us to move the ioend submission up to the level where the writepage context is declared. This does mean we do not submit IO until we packaged the entire writeback range, but with the block plugging in the writepages call this is the way IO is submitted, anyway. It also means that we need to handle discontiguous page ranges. If the pages sent down by write_cache_pages to the writepage callback are discontiguous, we need to detect this and put each discontiguous page range into individual ioends. This is needed to ensure that the ioend accurately represents the range of the file that it covers so that file size updates during IO completion set the size correctly. Failure to take into account the discontiguous ranges results in files being too small when writeback patterns are non-sequential. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-02-15 13:21:19 +07:00
int ret;
xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED);
xfs: Introduce writeback context for writepages xfs_vm_writepages() calls generic_writepages to writeback a range of a file, but then xfs_vm_writepage() clusters pages itself as it does not have any context it can pass between->writepage calls from __write_cache_pages(). Introduce a writeback context for xfs_vm_writepages() and call __write_cache_pages directly with our own writepage callback so that we can pass that context to each writepage invocation. This encapsulates the current mapping, whether it is valid or not, the current ioend and it's IO type and the ioend chain being built. This requires us to move the ioend submission up to the level where the writepage context is declared. This does mean we do not submit IO until we packaged the entire writeback range, but with the block plugging in the writepages call this is the way IO is submitted, anyway. It also means that we need to handle discontiguous page ranges. If the pages sent down by write_cache_pages to the writepage callback are discontiguous, we need to detect this and put each discontiguous page range into individual ioends. This is needed to ensure that the ioend accurately represents the range of the file that it covers so that file size updates during IO completion set the size correctly. Failure to take into account the discontiguous ranges results in files being too small when writeback patterns are non-sequential. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-02-15 13:21:19 +07:00
ret = write_cache_pages(mapping, wbc, xfs_do_writepage, &wpc);
xfs: don't chain ioends during writepage submission Currently we can build a long ioend chain during ->writepages that gets attached to the writepage context. IO submission only then occurs when we finish all the writepage processing. This means we can have many ioends allocated and pending, and this violates the mempool guarantees that we need to give about forwards progress. i.e. we really should only have one ioend being built at a time, otherwise we may drain the mempool trying to allocate a new ioend and that blocks submission, completion and freeing of ioends that are already in progress. To prevent this situation from happening, we need to submit ioends for IO as soon as they are ready for dispatch rather than queuing them for later submission. This means the ioends have bios built immediately and they get queued on any plug that is current active. Hence if we schedule away from writeback, the ioends that have been built will make forwards progress due to the plug flushing on context switch. This will also prevent context switches from creating unnecessary IO submission latency. We can't completely avoid having nested IO allocation - when we have a block size smaller than a page size, we still need to hold the ioend submission until after we have marked the current page dirty. Hence we may need multiple ioends to be held while the current page is completely mapped and made ready for IO dispatch. We cannot avoid this problem - the current code already has this ioend chaining within a page so we can mostly ignore that it occurs. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-02-15 13:23:12 +07:00
if (wpc.ioend)
ret = xfs_submit_ioend(wbc, wpc.ioend, ret);
return ret;
}
STATIC int
xfs_dax_writepages(
struct address_space *mapping,
struct writeback_control *wbc)
{
xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED);
return dax_writeback_mapping_range(mapping,
xfs_find_bdev_for_inode(mapping->host), wbc);
}
STATIC int
xfs_vm_releasepage(
struct page *page,
gfp_t gfp_mask)
{
trace_xfs_releasepage(page->mapping->host, page, 0, 0);
return iomap_releasepage(page, gfp_mask);
}
STATIC sector_t
xfs_vm_bmap(
struct address_space *mapping,
sector_t block)
{
struct xfs_inode *ip = XFS_I(mapping->host);
trace_xfs_vm_bmap(ip);
/*
* The swap code (ab-)uses ->bmap to get a block mapping and then
* bypasses the file system for actual I/O. We really can't allow
* that on reflinks inodes, so we have to skip out here. And yes,
* 0 is the magic code for a bmap error.
*
* Since we don't pass back blockdev info, we can't return bmap
* information for rt files either.
*/
xfs: introduce an always_cow mode Add a mode where XFS never overwrites existing blocks in place. This is to aid debugging our COW code, and also put infatructure in place for things like possible future support for zoned block devices, which can't support overwrites. This mode is enabled globally by doing a: echo 1 > /sys/fs/xfs/debug/always_cow Note that the parameter is global to allow running all tests in xfstests easily in this mode, which would not easily be possible with a per-fs sysfs file. In always_cow mode persistent preallocations are disabled, and fallocate will fail when called with a 0 mode (with our without FALLOC_FL_KEEP_SIZE), and not create unwritten extent for zeroed space when called with FALLOC_FL_ZERO_RANGE or FALLOC_FL_UNSHARE_RANGE. There are a few interesting xfstests failures when run in always_cow mode: - generic/392 fails because the bytes used in the file used to test hole punch recovery are less after the log replay. This is because the blocks written and then punched out are only freed with a delay due to the logging mechanism. - xfs/170 will fail as the already fragile file streams mechanism doesn't seem to interact well with the COW allocator - xfs/180 xfs/182 xfs/192 xfs/198 xfs/204 and xfs/208 will claim the file system is badly fragmented, but there is not much we can do to avoid that when always writing out of place - xfs/205 fails because overwriting a file in always_cow mode will require new space allocation and the assumption in the test thus don't work anymore. - xfs/326 fails to modify the file at all in always_cow mode after injecting the refcount error, leading to an unexpected md5sum after the remount, but that again is expected Signed-off-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>
2019-02-19 00:38:49 +07:00
if (xfs_is_cow_inode(ip) || XFS_IS_REALTIME_INODE(ip))
return 0;
return iomap_bmap(mapping, block, &xfs_iomap_ops);
}
STATIC int
xfs_vm_readpage(
struct file *unused,
struct page *page)
{
trace_xfs_vm_readpage(page->mapping->host, 1);
return iomap_readpage(page, &xfs_iomap_ops);
}
STATIC int
xfs_vm_readpages(
struct file *unused,
struct address_space *mapping,
struct list_head *pages,
unsigned nr_pages)
{
trace_xfs_vm_readpages(mapping->host, nr_pages);
return iomap_readpages(mapping, pages, nr_pages, &xfs_iomap_ops);
xfs: don't dirty buffers beyond EOF generic/263 is failing fsx at this point with a page spanning EOF that cannot be invalidated. The operations are: 1190 mapwrite 0x52c00 thru 0x5e569 (0xb96a bytes) 1191 mapread 0x5c000 thru 0x5d636 (0x1637 bytes) 1192 write 0x5b600 thru 0x771ff (0x1bc00 bytes) where 1190 extents EOF from 0x54000 to 0x5e569. When the direct IO write attempts to invalidate the cached page over this range, it fails with -EBUSY and so any attempt to do page invalidation fails. The real question is this: Why can't that page be invalidated after it has been written to disk and cleaned? Well, there's data on the first two buffers in the page (1k block size, 4k page), but the third buffer on the page (i.e. beyond EOF) is failing drop_buffers because it's bh->b_state == 0x3, which is BH_Uptodate | BH_Dirty. IOWs, there's dirty buffers beyond EOF. Say what? OK, set_buffer_dirty() is called on all buffers from __set_page_buffers_dirty(), regardless of whether the buffer is beyond EOF or not, which means that when we get to ->writepage, we have buffers marked dirty beyond EOF that we need to clean. So, we need to implement our own .set_page_dirty method that doesn't dirty buffers beyond EOF. This is messy because the buffer code is not meant to be shared and it has interesting locking issues on the buffer dirty bits. So just copy and paste it and then modify it to suit what we need. Note: the solutions the other filesystems and generic block code use of marking the buffers clean in ->writepage does not work for XFS. It still leaves dirty buffers beyond EOF and invalidations still fail. Hence rather than play whack-a-mole, this patch simply prevents those buffers from being dirtied in the first place. cc: <stable@kernel.org> Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2014-09-02 09:12:51 +07:00
}
static int
xfs_iomap_swapfile_activate(
struct swap_info_struct *sis,
struct file *swap_file,
sector_t *span)
{
sis->bdev = xfs_find_bdev_for_inode(file_inode(swap_file));
return iomap_swapfile_activate(sis, swap_file, span, &xfs_iomap_ops);
}
const struct address_space_operations xfs_address_space_operations = {
.readpage = xfs_vm_readpage,
.readpages = xfs_vm_readpages,
.writepage = xfs_vm_writepage,
.writepages = xfs_vm_writepages,
.set_page_dirty = iomap_set_page_dirty,
.releasepage = xfs_vm_releasepage,
.invalidatepage = xfs_vm_invalidatepage,
.bmap = xfs_vm_bmap,
.direct_IO = noop_direct_IO,
.migratepage = iomap_migrate_page,
.is_partially_uptodate = iomap_is_partially_uptodate,
.error_remove_page = generic_error_remove_page,
.swap_activate = xfs_iomap_swapfile_activate,
};
const struct address_space_operations xfs_dax_aops = {
.writepages = xfs_dax_writepages,
.direct_IO = noop_direct_IO,
.set_page_dirty = noop_set_page_dirty,
.invalidatepage = noop_invalidatepage,
.swap_activate = xfs_iomap_swapfile_activate,
};