linux-stable/fs/xfs/xfs_aops.c

1678 lines
43 KiB
C
Raw Normal View History

/*
* Copyright (c) 2000-2005 Silicon Graphics, Inc.
* All Rights Reserved.
*
* This program is free software; you can redistribute it and/or
* modify it under the terms of the GNU General Public License as
* published by the Free Software Foundation.
*
* This program is distributed in the hope that it would be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program; if not, write the Free Software Foundation,
* Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
*/
#include "xfs.h"
#include "xfs_shared.h"
#include "xfs_format.h"
#include "xfs_log_format.h"
#include "xfs_trans_resv.h"
#include "xfs_sb.h"
#include "xfs_ag.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>
2009-12-14 23:14:59 +00:00
#include "xfs_trace.h"
#include "xfs_bmap.h"
#include "xfs_bmap_util.h"
#include "xfs_bmap_btree.h"
#include "xfs_dinode.h"
#include <linux/aio.h>
include cleanup: Update gfp.h and slab.h includes to prepare for breaking implicit slab.h inclusion from percpu.h percpu.h is included by sched.h and module.h and thus ends up being included when building most .c files. percpu.h includes slab.h which in turn includes gfp.h making everything defined by the two files universally available and complicating inclusion dependencies. percpu.h -> slab.h dependency is about to be removed. Prepare for this change by updating users of gfp and slab facilities include those headers directly instead of assuming availability. As this conversion needs to touch large number of source files, the following script is used as the basis of conversion. http://userweb.kernel.org/~tj/misc/slabh-sweep.py The script does the followings. * Scan files for gfp and slab usages and update includes such that only the necessary includes are there. ie. if only gfp is used, gfp.h, if slab is used, slab.h. * When the script inserts a new include, it looks at the include blocks and try to put the new include such that its order conforms to its surrounding. It's put in the include block which contains core kernel includes, in the same order that the rest are ordered - alphabetical, Christmas tree, rev-Xmas-tree or at the end if there doesn't seem to be any matching order. * If the script can't find a place to put a new include (mostly because the file doesn't have fitting include block), it prints out an error message indicating which .h file needs to be added to the file. The conversion was done in the following steps. 1. The initial automatic conversion of all .c files updated slightly over 4000 files, deleting around 700 includes and adding ~480 gfp.h and ~3000 slab.h inclusions. The script emitted errors for ~400 files. 2. Each error was manually checked. Some didn't need the inclusion, some needed manual addition while adding it to implementation .h or embedding .c file was more appropriate for others. This step added inclusions to around 150 files. 3. The script was run again and the output was compared to the edits from #2 to make sure no file was left behind. 4. Several build tests were done and a couple of problems were fixed. e.g. lib/decompress_*.c used malloc/free() wrappers around slab APIs requiring slab.h to be added manually. 5. The script was run on all .h files but without automatically editing them as sprinkling gfp.h and slab.h inclusions around .h files could easily lead to inclusion dependency hell. Most gfp.h inclusion directives were ignored as stuff from gfp.h was usually wildly available and often used in preprocessor macros. Each slab.h inclusion directive was examined and added manually as necessary. 6. percpu.h was updated not to include slab.h. 7. Build test were done on the following configurations and failures were fixed. CONFIG_GCOV_KERNEL was turned off for all tests (as my distributed build env didn't work with gcov compiles) and a few more options had to be turned off depending on archs to make things build (like ipr on powerpc/64 which failed due to missing writeq). * x86 and x86_64 UP and SMP allmodconfig and a custom test config. * powerpc and powerpc64 SMP allmodconfig * sparc and sparc64 SMP allmodconfig * ia64 SMP allmodconfig * s390 SMP allmodconfig * alpha SMP allmodconfig * um on x86_64 SMP allmodconfig 8. percpu.h modifications were reverted so that it could be applied as a separate patch and serve as bisection point. Given the fact that I had only a couple of failures from tests on step 6, I'm fairly confident about the coverage of this conversion patch. If there is a breakage, it's likely to be something in one of the arch headers which should be easily discoverable easily on most builds of the specific arch. Signed-off-by: Tejun Heo <tj@kernel.org> Guess-its-ok-by: Christoph Lameter <cl@linux-foundation.org> Cc: Ingo Molnar <mingo@redhat.com> Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
2010-03-24 08:04:11 +00:00
#include <linux/gfp.h>
#include <linux/mpage.h>
#include <linux/pagevec.h>
#include <linux/writeback.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>
2009-12-14 23:14:59 +00:00
void
xfs_count_page_state(
struct page *page,
int *delalloc,
int *unwritten)
{
struct buffer_head *bh, *head;
*delalloc = *unwritten = 0;
bh = head = page_buffers(page);
do {
if (buffer_unwritten(bh))
(*unwritten) = 1;
else if (buffer_delay(bh))
(*delalloc) = 1;
} while ((bh = bh->b_this_page) != head);
}
STATIC 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;
}
/*
* We're now finished for good with this ioend structure.
* Update the page state via the associated buffer_heads,
* release holds on the inode and bio, and finally free
* up memory. Do not use the ioend after this.
*/
STATIC void
xfs_destroy_ioend(
xfs_ioend_t *ioend)
{
struct buffer_head *bh, *next;
for (bh = ioend->io_buffer_head; bh; bh = next) {
next = bh->b_private;
bh->b_end_io(bh, !ioend->io_error);
}
mempool_free(ioend, xfs_ioend_pool);
}
/*
* 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;
tp = xfs_trans_alloc(mp, XFS_TRANS_FSYNC_TS);
error = xfs_trans_reserve(tp, &M_RES(mp)->tr_fsyncts, 0, 0);
if (error) {
xfs_trans_cancel(tp, 0);
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 02:01:00 +00:00
* We may pass freeze protection with a transaction. So tell lockdep
* we released it.
*/
rwsem_release(&ioend->io_inode->i_sb->s_writers.lock_map[SB_FREEZE_FS-1],
1, _THIS_IP_);
/*
* We hand off the transaction to the completion thread now, so
* clear the flag here.
*/
current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
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 03:49:46 +00:00
/*
* 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 03:49:46 +00:00
*/
STATIC int
[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 03:49:46 +00:00
xfs_setfilesize(
struct xfs_ioend *ioend)
[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 03:49:46 +00:00
{
struct xfs_inode *ip = XFS_I(ioend->io_inode);
struct xfs_trans *tp = 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 03:49:46 +00:00
xfs_fsize_t isize;
/*
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 02:01:00 +00:00
* The transaction may have been allocated in the I/O submission thread,
* thus we need to mark ourselves as beeing in a transaction manually.
* Similarly for freeze protection.
*/
current_set_flags_nested(&tp->t_pflags, PF_FSTRANS);
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 02:01:00 +00:00
rwsem_acquire_read(&VFS_I(ip)->i_sb->s_writers.lock_map[SB_FREEZE_FS-1],
0, 1, _THIS_IP_);
xfs_ilock(ip, XFS_ILOCK_EXCL);
isize = xfs_new_eof(ip, ioend->io_offset + ioend->io_size);
if (!isize) {
xfs_iunlock(ip, XFS_ILOCK_EXCL);
xfs_trans_cancel(tp, 0);
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 03:49:46 +00:00
}
trace_xfs_setfilesize(ip, ioend->io_offset, ioend->io_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, 0);
}
/*
* Schedule IO completion handling on the final put of an ioend.
*
* If there is no work to do we might as well call it a day and free the
* ioend right now.
*/
STATIC void
xfs_finish_ioend(
struct xfs_ioend *ioend)
{
if (atomic_dec_and_test(&ioend->io_remaining)) {
struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount;
if (ioend->io_type == XFS_IO_UNWRITTEN)
queue_work(mp->m_unwritten_workqueue, &ioend->io_work);
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 02:01:00 +00:00
else if (ioend->io_append_trans ||
(ioend->io_isdirect && xfs_ioend_is_append(ioend)))
queue_work(mp->m_data_workqueue, &ioend->io_work);
else
xfs_destroy_ioend(ioend);
}
[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 03:49:46 +00:00
}
/*
* IO write completion.
*/
STATIC void
xfs_end_io(
struct work_struct *work)
{
xfs_ioend_t *ioend = container_of(work, xfs_ioend_t, io_work);
struct xfs_inode *ip = XFS_I(ioend->io_inode);
int error = 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 03:49:46 +00:00
if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
ioend->io_error = -EIO;
goto done;
}
if (ioend->io_error)
goto done;
/*
* For unwritten extents we need to issue transactions to convert a
* range to normal written extens after the data I/O has finished.
*/
if (ioend->io_type == XFS_IO_UNWRITTEN) {
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 02:01:00 +00:00
error = xfs_iomap_write_unwritten(ip, ioend->io_offset,
ioend->io_size);
} else if (ioend->io_isdirect && xfs_ioend_is_append(ioend)) {
/*
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 02:01:00 +00:00
* For direct I/O we do not know if we need to allocate blocks
* or not so we can't preallocate an append transaction as that
* results in nested reservations and log space deadlocks. Hence
* allocate the transaction here. While this is sub-optimal and
* can block IO completion for some time, we're stuck with doing
* it this way until we can pass the ioend to the direct IO
* allocation callbacks and avoid nesting that way.
*/
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 02:01:00 +00:00
error = xfs_setfilesize_trans_alloc(ioend);
if (error)
goto done;
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 02:01:00 +00:00
error = xfs_setfilesize(ioend);
} else if (ioend->io_append_trans) {
error = xfs_setfilesize(ioend);
} else {
ASSERT(!xfs_ioend_is_append(ioend));
}
[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 03:49:46 +00:00
done:
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 02:01:00 +00:00
if (error)
ioend->io_error = -error;
xfs_destroy_ioend(ioend);
}
/*
* Call IO completion handling in caller context on the final put of an ioend.
*/
STATIC void
xfs_finish_ioend_sync(
struct xfs_ioend *ioend)
{
if (atomic_dec_and_test(&ioend->io_remaining))
xfs_end_io(&ioend->io_work);
}
/*
* Allocate and initialise an IO completion structure.
* We need to track unwritten extent write completion here initially.
* We'll need to extend this for updating the ondisk inode size later
* (vs. incore size).
*/
STATIC xfs_ioend_t *
xfs_alloc_ioend(
struct inode *inode,
unsigned int type)
{
xfs_ioend_t *ioend;
ioend = mempool_alloc(xfs_ioend_pool, GFP_NOFS);
/*
* Set the count to 1 initially, which will prevent an I/O
* completion callback from happening before we have started
* all the I/O from calling the completion routine too early.
*/
atomic_set(&ioend->io_remaining, 1);
ioend->io_isdirect = 0;
ioend->io_error = 0;
ioend->io_list = NULL;
ioend->io_type = type;
ioend->io_inode = inode;
ioend->io_buffer_head = NULL;
ioend->io_buffer_tail = NULL;
ioend->io_offset = 0;
ioend->io_size = 0;
ioend->io_append_trans = NULL;
INIT_WORK(&ioend->io_work, xfs_end_io);
return ioend;
}
STATIC int
xfs_map_blocks(
struct inode *inode,
loff_t offset,
struct xfs_bmbt_irec *imap,
int type,
int nonblocking)
{
struct xfs_inode *ip = XFS_I(inode);
struct xfs_mount *mp = ip->i_mount;
ssize_t count = 1 << inode->i_blkbits;
xfs_fileoff_t offset_fsb, end_fsb;
int error = 0;
int bmapi_flags = XFS_BMAPI_ENTIRE;
int nimaps = 1;
if (XFS_FORCED_SHUTDOWN(mp))
return -XFS_ERROR(EIO);
if (type == XFS_IO_UNWRITTEN)
bmapi_flags |= XFS_BMAPI_IGSTATE;
if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {
if (nonblocking)
return -XFS_ERROR(EAGAIN);
xfs_ilock(ip, XFS_ILOCK_SHARED);
}
ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
(ip->i_df.if_flags & XFS_IFEXTENTS));
ASSERT(offset <= mp->m_super->s_maxbytes);
if (offset + count > mp->m_super->s_maxbytes)
count = mp->m_super->s_maxbytes - offset;
end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
offset_fsb = XFS_B_TO_FSBT(mp, offset);
error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb,
imap, &nimaps, bmapi_flags);
xfs_iunlock(ip, XFS_ILOCK_SHARED);
if (error)
return -XFS_ERROR(error);
if (type == XFS_IO_DELALLOC &&
(!nimaps || isnullstartblock(imap->br_startblock))) {
error = xfs_iomap_write_allocate(ip, offset, imap);
if (!error)
trace_xfs_map_blocks_alloc(ip, offset, count, type, imap);
return -XFS_ERROR(error);
}
#ifdef DEBUG
if (type == XFS_IO_UNWRITTEN) {
ASSERT(nimaps);
ASSERT(imap->br_startblock != HOLESTARTBLOCK);
ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
}
#endif
if (nimaps)
trace_xfs_map_blocks_found(ip, offset, count, type, imap);
return 0;
}
STATIC int
xfs_imap_valid(
struct inode *inode,
struct xfs_bmbt_irec *imap,
xfs_off_t offset)
{
offset >>= inode->i_blkbits;
return offset >= imap->br_startoff &&
offset < imap->br_startoff + imap->br_blockcount;
}
/*
* BIO completion handler for buffered IO.
*/
STATIC void
xfs_end_bio(
struct bio *bio,
int error)
{
xfs_ioend_t *ioend = bio->bi_private;
ASSERT(atomic_read(&bio->bi_cnt) >= 1);
ioend->io_error = test_bit(BIO_UPTODATE, &bio->bi_flags) ? 0 : error;
/* Toss bio and pass work off to an xfsdatad thread */
bio->bi_private = NULL;
bio->bi_end_io = NULL;
bio_put(bio);
xfs_finish_ioend(ioend);
}
STATIC void
xfs_submit_ioend_bio(
struct writeback_control *wbc,
xfs_ioend_t *ioend,
struct bio *bio)
{
atomic_inc(&ioend->io_remaining);
bio->bi_private = ioend;
bio->bi_end_io = xfs_end_bio;
submit_bio(wbc->sync_mode == WB_SYNC_ALL ? WRITE_SYNC : WRITE, bio);
}
STATIC struct bio *
xfs_alloc_ioend_bio(
struct buffer_head *bh)
{
int nvecs = bio_get_nr_vecs(bh->b_bdev);
struct bio *bio = bio_alloc(GFP_NOIO, nvecs);
ASSERT(bio->bi_private == NULL);
bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9);
bio->bi_bdev = bh->b_bdev;
return bio;
}
STATIC void
xfs_start_buffer_writeback(
struct buffer_head *bh)
{
ASSERT(buffer_mapped(bh));
ASSERT(buffer_locked(bh));
ASSERT(!buffer_delay(bh));
ASSERT(!buffer_unwritten(bh));
mark_buffer_async_write(bh);
set_buffer_uptodate(bh);
clear_buffer_dirty(bh);
}
STATIC void
xfs_start_page_writeback(
struct page *page,
int clear_dirty,
int buffers)
{
ASSERT(PageLocked(page));
ASSERT(!PageWriteback(page));
if (clear_dirty)
clear_page_dirty_for_io(page);
set_page_writeback(page);
unlock_page(page);
writeback: remove pages_skipped accounting in __block_write_full_page() Miklos Szeredi <miklos@szeredi.hu> and me identified a writeback bug: > The following strange behavior can be observed: > > 1. large file is written > 2. after 30 seconds, nr_dirty goes down by 1024 > 3. then for some time (< 30 sec) nothing happens (disk idle) > 4. then nr_dirty again goes down by 1024 > 5. repeat from 3. until whole file is written > > So basically a 4Mbyte chunk of the file is written every 30 seconds. > I'm quite sure this is not the intended behavior. It can be produced by the following test scheme: # cat bin/test-writeback.sh grep nr_dirty /proc/vmstat echo 1 > /proc/sys/fs/inode_debug dd if=/dev/zero of=/var/x bs=1K count=204800& while true; do grep nr_dirty /proc/vmstat; sleep 1; done # bin/test-writeback.sh nr_dirty 19207 nr_dirty 19207 nr_dirty 30924 204800+0 records in 204800+0 records out 209715200 bytes (210 MB) copied, 1.58363 seconds, 132 MB/s nr_dirty 47150 nr_dirty 47141 nr_dirty 47142 nr_dirty 47142 nr_dirty 47142 nr_dirty 47142 nr_dirty 47205 nr_dirty 47214 nr_dirty 47214 nr_dirty 47214 nr_dirty 47214 nr_dirty 47214 nr_dirty 47215 nr_dirty 47216 nr_dirty 47216 nr_dirty 47216 nr_dirty 47154 nr_dirty 47143 nr_dirty 47143 nr_dirty 47143 nr_dirty 47143 nr_dirty 47143 nr_dirty 47142 nr_dirty 47142 nr_dirty 47142 nr_dirty 47142 nr_dirty 47134 nr_dirty 47134 nr_dirty 47135 nr_dirty 47135 nr_dirty 47135 nr_dirty 46097 <== -1038 nr_dirty 46098 nr_dirty 46098 nr_dirty 46098 [...] nr_dirty 46091 nr_dirty 46092 nr_dirty 46092 nr_dirty 45069 <== -1023 nr_dirty 45056 nr_dirty 45056 nr_dirty 45056 [...] nr_dirty 37822 nr_dirty 36799 <== -1023 [...] nr_dirty 36781 nr_dirty 35758 <== -1023 [...] nr_dirty 34708 nr_dirty 33672 <== -1024 [...] nr_dirty 33692 nr_dirty 32669 <== -1023 % ls -li /var/x 847824 -rw-r--r-- 1 root root 200M 2007-08-12 04:12 /var/x % dmesg|grep 847824 # generated by a debug printk [ 529.263184] redirtied inode 847824 line 548 [ 564.250872] redirtied inode 847824 line 548 [ 594.272797] redirtied inode 847824 line 548 [ 629.231330] redirtied inode 847824 line 548 [ 659.224674] redirtied inode 847824 line 548 [ 689.219890] redirtied inode 847824 line 548 [ 724.226655] redirtied inode 847824 line 548 [ 759.198568] redirtied inode 847824 line 548 # line 548 in fs/fs-writeback.c: 543 if (wbc->pages_skipped != pages_skipped) { 544 /* 545 * writeback is not making progress due to locked 546 * buffers. Skip this inode for now. 547 */ 548 redirty_tail(inode); 549 } More debug efforts show that __block_write_full_page() never has the chance to call submit_bh() for that big dirty file: the buffer head is *clean*. So basicly no page io is issued by __block_write_full_page(), hence pages_skipped goes up. Also the comment in generic_sync_sb_inodes(): 544 /* 545 * writeback is not making progress due to locked 546 * buffers. Skip this inode for now. 547 */ and the comment in __block_write_full_page(): 1713 /* 1714 * The page was marked dirty, but the buffers were 1715 * clean. Someone wrote them back by hand with 1716 * ll_rw_block/submit_bh. A rare case. 1717 */ do not quite agree with each other. The page writeback should be skipped for 'locked buffer', but here it is 'clean buffer'! This patch fixes this bug. Though I'm not sure why __block_write_full_page() is called only to do nothing and who actually issued the writeback for us. This is the two possible new behaviors after the patch: 1) pretty nice: wait 30s and write ALL:) 2) not so good: - during the dd: ~16M - after 30s: ~4M - after 5s: ~4M - after 5s: ~176M The next patch will fix case (2). Cc: David Chinner <dgc@sgi.com> Cc: Ken Chen <kenchen@google.com> Signed-off-by: Fengguang Wu <wfg@mail.ustc.edu.cn> Signed-off-by: David Chinner <dgc@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-10-17 06:30:42 +00:00
/* If no buffers on the page are to be written, finish it here */
if (!buffers)
end_page_writeback(page);
}
static inline int xfs_bio_add_buffer(struct bio *bio, struct buffer_head *bh)
{
return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh));
}
/*
* Submit all of the bios for all of the ioends we have saved up, covering the
* initial writepage page and also any probed pages.
*
* Because we may have multiple ioends spanning a page, we need to start
* writeback on all the buffers before we submit them for I/O. If we mark the
* buffers as we got, then we can end up with a page that only has buffers
* marked async write and I/O complete on can occur before we mark the other
* buffers async write.
*
* The end result of this is that we trip a bug in end_page_writeback() because
* we call it twice for the one page as the code in end_buffer_async_write()
* assumes that all buffers on the page are started at the same time.
*
* The fix is two passes across the ioend list - one to start writeback on the
* buffer_heads, and then submit them for I/O on the second pass.
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 11:09:45 +00:00
*
* 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 ioend chain rather
* than submit it to IO. This typically only happens on a filesystem shutdown.
*/
STATIC void
xfs_submit_ioend(
struct writeback_control *wbc,
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 11:09:45 +00:00
xfs_ioend_t *ioend,
int fail)
{
xfs_ioend_t *head = ioend;
xfs_ioend_t *next;
struct buffer_head *bh;
struct bio *bio;
sector_t lastblock = 0;
/* Pass 1 - start writeback */
do {
next = ioend->io_list;
for (bh = ioend->io_buffer_head; bh; bh = bh->b_private)
xfs_start_buffer_writeback(bh);
} while ((ioend = next) != NULL);
/* Pass 2 - submit I/O */
ioend = head;
do {
next = ioend->io_list;
bio = NULL;
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 11:09:45 +00:00
/*
* 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 (fail) {
ioend->io_error = -fail;
xfs_finish_ioend(ioend);
continue;
}
for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) {
if (!bio) {
retry:
bio = xfs_alloc_ioend_bio(bh);
} else if (bh->b_blocknr != lastblock + 1) {
xfs_submit_ioend_bio(wbc, ioend, bio);
goto retry;
}
if (xfs_bio_add_buffer(bio, bh) != bh->b_size) {
xfs_submit_ioend_bio(wbc, ioend, bio);
goto retry;
}
lastblock = bh->b_blocknr;
}
if (bio)
xfs_submit_ioend_bio(wbc, ioend, bio);
xfs_finish_ioend(ioend);
} while ((ioend = next) != NULL);
}
/*
* Cancel submission of all buffer_heads so far in this endio.
* Toss the endio too. Only ever called for the initial page
* in a writepage request, so only ever one page.
*/
STATIC void
xfs_cancel_ioend(
xfs_ioend_t *ioend)
{
xfs_ioend_t *next;
struct buffer_head *bh, *next_bh;
do {
next = ioend->io_list;
bh = ioend->io_buffer_head;
do {
next_bh = bh->b_private;
clear_buffer_async_write(bh);
unlock_buffer(bh);
} while ((bh = next_bh) != NULL);
mempool_free(ioend, xfs_ioend_pool);
} while ((ioend = next) != NULL);
}
/*
* Test to see if we've been building up a completion structure for
* earlier buffers -- if so, we try to append to this ioend if we
* can, otherwise we finish off any current ioend and start another.
* Return true if we've finished the given ioend.
*/
STATIC void
xfs_add_to_ioend(
struct inode *inode,
struct buffer_head *bh,
xfs_off_t offset,
unsigned int type,
xfs_ioend_t **result,
int need_ioend)
{
xfs_ioend_t *ioend = *result;
if (!ioend || need_ioend || type != ioend->io_type) {
xfs_ioend_t *previous = *result;
ioend = xfs_alloc_ioend(inode, type);
ioend->io_offset = offset;
ioend->io_buffer_head = bh;
ioend->io_buffer_tail = bh;
if (previous)
previous->io_list = ioend;
*result = ioend;
} else {
ioend->io_buffer_tail->b_private = bh;
ioend->io_buffer_tail = bh;
}
bh->b_private = NULL;
ioend->io_size += bh->b_size;
}
STATIC void
xfs_map_buffer(
struct inode *inode,
struct buffer_head *bh,
struct xfs_bmbt_irec *imap,
xfs_off_t offset)
{
sector_t bn;
struct xfs_mount *m = XFS_I(inode)->i_mount;
xfs_off_t iomap_offset = XFS_FSB_TO_B(m, imap->br_startoff);
xfs_daddr_t iomap_bn = xfs_fsb_to_db(XFS_I(inode), imap->br_startblock);
ASSERT(imap->br_startblock != HOLESTARTBLOCK);
ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
bn = (iomap_bn >> (inode->i_blkbits - BBSHIFT)) +
((offset - iomap_offset) >> inode->i_blkbits);
ASSERT(bn || XFS_IS_REALTIME_INODE(XFS_I(inode)));
bh->b_blocknr = bn;
set_buffer_mapped(bh);
}
STATIC void
xfs_map_at_offset(
struct inode *inode,
struct buffer_head *bh,
struct xfs_bmbt_irec *imap,
xfs_off_t offset)
{
ASSERT(imap->br_startblock != HOLESTARTBLOCK);
ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
xfs_map_buffer(inode, bh, imap, offset);
set_buffer_mapped(bh);
clear_buffer_delay(bh);
clear_buffer_unwritten(bh);
}
/*
* Test if a given page is suitable for writing as part of an unwritten
* or delayed allocate extent.
*/
STATIC int
xfs_check_page_type(
struct page *page,
unsigned int type)
{
if (PageWriteback(page))
return 0;
if (page->mapping && page_has_buffers(page)) {
struct buffer_head *bh, *head;
int acceptable = 0;
bh = head = page_buffers(page);
do {
if (buffer_unwritten(bh))
acceptable += (type == XFS_IO_UNWRITTEN);
else if (buffer_delay(bh))
acceptable += (type == XFS_IO_DELALLOC);
else if (buffer_dirty(bh) && buffer_mapped(bh))
acceptable += (type == XFS_IO_OVERWRITE);
else
break;
} while ((bh = bh->b_this_page) != head);
if (acceptable)
return 1;
}
return 0;
}
/*
* Allocate & map buffers for page given the extent map. Write it out.
* except for the original page of a writepage, this is called on
* delalloc/unwritten pages only, for the original page it is possible
* that the page has no mapping at all.
*/
STATIC int
xfs_convert_page(
struct inode *inode,
struct page *page,
loff_t tindex,
struct xfs_bmbt_irec *imap,
xfs_ioend_t **ioendp,
struct writeback_control *wbc)
{
struct buffer_head *bh, *head;
xfs_off_t end_offset;
unsigned long p_offset;
unsigned int type;
int len, page_dirty;
int count = 0, done = 0, uptodate = 1;
xfs_off_t offset = page_offset(page);
if (page->index != tindex)
goto fail;
if (!trylock_page(page))
goto fail;
if (PageWriteback(page))
goto fail_unlock_page;
if (page->mapping != inode->i_mapping)
goto fail_unlock_page;
if (!xfs_check_page_type(page, (*ioendp)->io_type))
goto fail_unlock_page;
/*
* page_dirty is initially a count of buffers on the page before
* EOF and is decremented as we move each into a cleanable state.
*
* Derivation:
*
* End offset is the highest offset that this page should represent.
* If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1))
* will evaluate non-zero and be less than PAGE_CACHE_SIZE and
* hence give us the correct page_dirty count. On any other page,
* it will be zero and in that case we need page_dirty to be the
* count of buffers on the page.
*/
end_offset = min_t(unsigned long long,
(xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT,
i_size_read(inode));
xfs: fix sub-page blocksize data integrity writes FSX on 512 byte block size filesystems has been failing for some time with corrupted data. The fault dates back to the change in the writeback data integrity algorithm that uses a mark-and-sweep approach to avoid data writeback livelocks. Unfortunately, a side effect of this mark-and-sweep approach is that each page will only be written once for a data integrity sync, and there is a condition in writeback in XFS where a page may require two writeback attempts to be fully written. As a result of the high level change, we now only get a partial page writeback during the integrity sync because the first pass through writeback clears the mark left on the page index to tell writeback that the page needs writeback.... The cause is writing a partial page in the clustering code. This can happen when a mapping boundary falls in the middle of a page - we end up writing back the first part of the page that the mapping covers, but then never revisit the page to have the remainder mapped and written. The fix is simple - if the mapping boundary falls inside a page, then simple abort clustering without touching the page. This means that the next ->writepage entry that write_cache_pages() will make is the page we aborted on, and xfs_vm_writepage() will map all sections of the page correctly. This behaviour is also optimal for non-data integrity writes, as it results in contiguous sequential writeback of the file rather than missing small holes and having to write them a "random" writes in a future pass. With this fix, all the fsx tests in xfstests now pass on a 512 byte block size filesystem on a 4k page machine. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Ben Myers <bpm@sgi.com> (cherry picked from commit 49b137cbbcc836ef231866c137d24f42c42bb483)
2013-05-19 23:51:08 +00:00
/*
* If the current map does not span the entire page we are about to try
* to write, then give up. The only way we can write a page that spans
* multiple mappings in a single writeback iteration is via the
* xfs_vm_writepage() function. Data integrity writeback requires the
* entire page to be written in a single attempt, otherwise the part of
* the page we don't write here doesn't get written as part of the data
* integrity sync.
*
* For normal writeback, we also don't attempt to write partial pages
* here as it simply means that write_cache_pages() will see it under
* writeback and ignore the page until some point in the future, at
* which time this will be the only page in the file that needs
* writeback. Hence for more optimal IO patterns, we should always
* avoid partial page writeback due to multiple mappings on a page here.
*/
if (!xfs_imap_valid(inode, imap, end_offset))
goto fail_unlock_page;
len = 1 << inode->i_blkbits;
p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1),
PAGE_CACHE_SIZE);
p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE;
page_dirty = p_offset / len;
bh = head = page_buffers(page);
do {
if (offset >= end_offset)
break;
if (!buffer_uptodate(bh))
uptodate = 0;
if (!(PageUptodate(page) || buffer_uptodate(bh))) {
done = 1;
continue;
}
if (buffer_unwritten(bh) || buffer_delay(bh) ||
buffer_mapped(bh)) {
if (buffer_unwritten(bh))
type = XFS_IO_UNWRITTEN;
else if (buffer_delay(bh))
type = XFS_IO_DELALLOC;
else
type = XFS_IO_OVERWRITE;
if (!xfs_imap_valid(inode, imap, offset)) {
done = 1;
continue;
}
lock_buffer(bh);
if (type != XFS_IO_OVERWRITE)
xfs_map_at_offset(inode, bh, imap, offset);
xfs_add_to_ioend(inode, bh, offset, type,
ioendp, done);
page_dirty--;
count++;
} else {
done = 1;
}
} while (offset += len, (bh = bh->b_this_page) != head);
if (uptodate && bh == head)
SetPageUptodate(page);
if (count) {
if (--wbc->nr_to_write <= 0 &&
wbc->sync_mode == WB_SYNC_NONE)
done = 1;
}
xfs_start_page_writeback(page, !page_dirty, count);
return done;
fail_unlock_page:
unlock_page(page);
fail:
return 1;
}
/*
* Convert & write out a cluster of pages in the same extent as defined
* by mp and following the start page.
*/
STATIC void
xfs_cluster_write(
struct inode *inode,
pgoff_t tindex,
struct xfs_bmbt_irec *imap,
xfs_ioend_t **ioendp,
struct writeback_control *wbc,
pgoff_t tlast)
{
struct pagevec pvec;
int done = 0, i;
pagevec_init(&pvec, 0);
while (!done && tindex <= tlast) {
unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1);
if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len))
break;
for (i = 0; i < pagevec_count(&pvec); i++) {
done = xfs_convert_page(inode, pvec.pages[i], tindex++,
imap, ioendp, wbc);
if (done)
break;
}
pagevec_release(&pvec);
cond_resched();
}
}
STATIC void
xfs_vm_invalidatepage(
struct page *page,
unsigned int offset,
unsigned int length)
{
trace_xfs_invalidatepage(page->mapping->host, page, offset,
length);
block_invalidatepage(page, offset, length);
}
/*
* If the page has delalloc buffers 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 a BUG() in xfs_get_blocks() later on if a direct IO read
* is done on that same region - the delalloc extent is returned when none is
* supposed to be there.
*
* We prevent this by truncating away the delalloc regions on the page before
* invalidating it. 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).
*
* This is not a performance critical path, so for now just do the punching a
* buffer head at a time.
*/
STATIC void
xfs_aops_discard_page(
struct page *page)
{
struct inode *inode = page->mapping->host;
struct xfs_inode *ip = XFS_I(inode);
struct buffer_head *bh, *head;
loff_t offset = page_offset(page);
if (!xfs_check_page_type(page, XFS_IO_DELALLOC))
goto out_invalidate;
if (XFS_FORCED_SHUTDOWN(ip->i_mount))
goto out_invalidate;
xfs_alert(ip->i_mount,
"page discard on page %p, inode 0x%llx, offset %llu.",
page, ip->i_ino, offset);
xfs_ilock(ip, XFS_ILOCK_EXCL);
bh = head = page_buffers(page);
do {
int error;
xfs_fileoff_t start_fsb;
if (!buffer_delay(bh))
goto next_buffer;
start_fsb = XFS_B_TO_FSBT(ip->i_mount, offset);
error = xfs_bmap_punch_delalloc_range(ip, start_fsb, 1);
if (error) {
/* something screwed, just bail */
if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
xfs_alert(ip->i_mount,
"page discard unable to remove delalloc mapping.");
}
break;
}
next_buffer:
offset += 1 << inode->i_blkbits;
} while ((bh = bh->b_this_page) != head);
xfs_iunlock(ip, XFS_ILOCK_EXCL);
out_invalidate:
xfs_vm_invalidatepage(page, 0, PAGE_CACHE_SIZE);
return;
}
/*
* 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.
* For any other dirty buffer heads on the page we should flush them.
*/
STATIC int
xfs_vm_writepage(
struct page *page,
struct writeback_control *wbc)
{
struct inode *inode = page->mapping->host;
struct buffer_head *bh, *head;
struct xfs_bmbt_irec imap;
xfs_ioend_t *ioend = NULL, *iohead = NULL;
loff_t offset;
unsigned int type;
__uint64_t end_offset;
pgoff_t end_index, last_index;
ssize_t len;
int err, imap_valid = 0, uptodate = 1;
int count = 0;
int nonblocking = 0;
trace_xfs_writepage(inode, page, 0, 0);
ASSERT(page_has_buffers(page));
/*
* 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(current->flags & PF_FSTRANS))
goto redirty;
/* Is this page beyond the end of the file? */
offset = i_size_read(inode);
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);
/*
* 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.
*/
if (page->index >= end_index + 1 || 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
* 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."
*/
zero_user_segment(page, offset_into_page, PAGE_CACHE_SIZE);
}
end_offset = min_t(unsigned long long,
(xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT,
offset);
len = 1 << inode->i_blkbits;
bh = head = page_buffers(page);
offset = page_offset(page);
type = XFS_IO_OVERWRITE;
if (wbc->sync_mode == WB_SYNC_NONE)
nonblocking = 1;
do {
int new_ioend = 0;
if (offset >= end_offset)
break;
if (!buffer_uptodate(bh))
uptodate = 0;
/*
* set_page_dirty dirties all buffers in a page, independent
* of their state. The dirty state however is entirely
* meaningless for holes (!mapped && uptodate), so skip
* buffers covering holes here.
*/
if (!buffer_mapped(bh) && buffer_uptodate(bh)) {
imap_valid = 0;
continue;
}
if (buffer_unwritten(bh)) {
if (type != XFS_IO_UNWRITTEN) {
type = XFS_IO_UNWRITTEN;
imap_valid = 0;
}
} else if (buffer_delay(bh)) {
if (type != XFS_IO_DELALLOC) {
type = XFS_IO_DELALLOC;
imap_valid = 0;
}
} else if (buffer_uptodate(bh)) {
if (type != XFS_IO_OVERWRITE) {
type = XFS_IO_OVERWRITE;
imap_valid = 0;
}
} else {
xfs: xfs_vm_writepage clear iomap_valid when !buffer_uptodate (REV2) On filesytems with a block size smaller than PAGE_SIZE we currently have a problem with unwritten extents. If a we have multi-block page for which an unwritten extent has been allocated, and only some of the buffers have been written to, and they are not contiguous, we can expose stale data from disk in the blocks between the writes after extent conversion. Example of a page with unwritten and real data. buffer content 0 empty b_state = 0 1 DATA b_state = 0x1023 Uptodate,Dirty,Mapped,Unwritten 2 DATA b_state = 0x1023 Uptodate,Dirty,Mapped,Unwritten 3 empty b_state = 0 4 empty b_state = 0 5 DATA b_state = 0x1023 Uptodate,Dirty,Mapped,Unwritten 6 DATA b_state = 0x1023 Uptodate,Dirty,Mapped,Unwritten 7 empty b_state = 0 Buffers 1, 2, 5, and 6 have been written to, leaving 0, 3, 4, and 7 empty. Currently buffers 1, 2, 5, and 6 are added to a single ioend, and when IO has completed, extent conversion creates a real extent from block 1 through block 6, leaving 0 and 7 unwritten. However buffers 3 and 4 were not written to disk, so stale data is exposed from those blocks on a subsequent read. Fix this by setting iomap_valid = 0 when we find a buffer that is not Uptodate. This ensures that buffers 5 and 6 are not added to the same ioend as buffers 1 and 2. Later these blocks will be converted into two separate real extents, leaving the blocks in between unwritten. Signed-off-by: Alain Renaud <arenaud@sgi.com> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2012-06-08 19:34:46 +00:00
if (PageUptodate(page))
ASSERT(buffer_mapped(bh));
xfs: xfs_vm_writepage clear iomap_valid when !buffer_uptodate (REV2) On filesytems with a block size smaller than PAGE_SIZE we currently have a problem with unwritten extents. If a we have multi-block page for which an unwritten extent has been allocated, and only some of the buffers have been written to, and they are not contiguous, we can expose stale data from disk in the blocks between the writes after extent conversion. Example of a page with unwritten and real data. buffer content 0 empty b_state = 0 1 DATA b_state = 0x1023 Uptodate,Dirty,Mapped,Unwritten 2 DATA b_state = 0x1023 Uptodate,Dirty,Mapped,Unwritten 3 empty b_state = 0 4 empty b_state = 0 5 DATA b_state = 0x1023 Uptodate,Dirty,Mapped,Unwritten 6 DATA b_state = 0x1023 Uptodate,Dirty,Mapped,Unwritten 7 empty b_state = 0 Buffers 1, 2, 5, and 6 have been written to, leaving 0, 3, 4, and 7 empty. Currently buffers 1, 2, 5, and 6 are added to a single ioend, and when IO has completed, extent conversion creates a real extent from block 1 through block 6, leaving 0 and 7 unwritten. However buffers 3 and 4 were not written to disk, so stale data is exposed from those blocks on a subsequent read. Fix this by setting iomap_valid = 0 when we find a buffer that is not Uptodate. This ensures that buffers 5 and 6 are not added to the same ioend as buffers 1 and 2. Later these blocks will be converted into two separate real extents, leaving the blocks in between unwritten. Signed-off-by: Alain Renaud <arenaud@sgi.com> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2012-06-08 19:34:46 +00:00
/*
* This buffer is not uptodate and will not be
* written to disk. Ensure that we will put any
* subsequent writeable buffers into a new
* ioend.
*/
imap_valid = 0;
continue;
}
if (imap_valid)
imap_valid = xfs_imap_valid(inode, &imap, offset);
if (!imap_valid) {
/*
* If we didn't have a valid mapping then we need to
* put the new mapping into a separate ioend structure.
* This ensures non-contiguous extents always have
* separate ioends, which is particularly important
* for unwritten extent conversion at I/O completion
* time.
*/
new_ioend = 1;
err = xfs_map_blocks(inode, offset, &imap, type,
nonblocking);
if (err)
goto error;
imap_valid = xfs_imap_valid(inode, &imap, offset);
}
if (imap_valid) {
lock_buffer(bh);
if (type != XFS_IO_OVERWRITE)
xfs_map_at_offset(inode, bh, &imap, offset);
xfs_add_to_ioend(inode, bh, offset, type, &ioend,
new_ioend);
count++;
}
if (!iohead)
iohead = ioend;
} while (offset += len, ((bh = bh->b_this_page) != head));
if (uptodate && bh == head)
SetPageUptodate(page);
xfs_start_page_writeback(page, 1, count);
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 11:09:45 +00:00
/* if there is no IO to be submitted for this page, we are done */
if (!ioend)
return 0;
ASSERT(iohead);
/*
* Any errors from this point onwards need tobe reported through the IO
* completion path as we have marked the initial page as under writeback
* and unlocked it.
*/
if (imap_valid) {
xfs_off_t end_index;
end_index = imap.br_startoff + imap.br_blockcount;
/* to bytes */
end_index <<= inode->i_blkbits;
/* to pages */
end_index = (end_index - 1) >> PAGE_CACHE_SHIFT;
/* check against file size */
if (end_index > last_index)
end_index = last_index;
xfs_cluster_write(inode, page->index + 1, &imap, &ioend,
wbc, end_index);
}
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 11:09:45 +00:00
/*
* Reserve log space if we might write beyond the on-disk inode size.
*/
err = 0;
if (ioend->io_type != XFS_IO_UNWRITTEN && xfs_ioend_is_append(ioend))
err = xfs_setfilesize_trans_alloc(ioend);
xfs_submit_ioend(wbc, iohead, err);
return 0;
error:
if (iohead)
xfs_cancel_ioend(iohead);
if (err == -EAGAIN)
goto redirty;
xfs_aops_discard_page(page);
ClearPageUptodate(page);
unlock_page(page);
return err;
redirty:
redirty_page_for_writepage(wbc, page);
unlock_page(page);
return 0;
}
STATIC int
xfs_vm_writepages(
struct address_space *mapping,
struct writeback_control *wbc)
{
xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED);
return generic_writepages(mapping, wbc);
}
/*
* Called to move a page into cleanable state - and from there
* to be released. The page should already be clean. We always
* have buffer heads in this call.
*
* Returns 1 if the page is ok to release, 0 otherwise.
*/
STATIC int
xfs_vm_releasepage(
struct page *page,
gfp_t gfp_mask)
{
int delalloc, unwritten;
trace_xfs_releasepage(page->mapping->host, page, 0, 0);
xfs_count_page_state(page, &delalloc, &unwritten);
if (WARN_ON(delalloc))
return 0;
if (WARN_ON(unwritten))
return 0;
return try_to_free_buffers(page);
}
STATIC int
__xfs_get_blocks(
struct inode *inode,
sector_t iblock,
struct buffer_head *bh_result,
int create,
int direct)
{
struct xfs_inode *ip = XFS_I(inode);
struct xfs_mount *mp = ip->i_mount;
xfs_fileoff_t offset_fsb, end_fsb;
int error = 0;
int lockmode = 0;
struct xfs_bmbt_irec imap;
int nimaps = 1;
xfs_off_t offset;
ssize_t size;
int new = 0;
if (XFS_FORCED_SHUTDOWN(mp))
return -XFS_ERROR(EIO);
offset = (xfs_off_t)iblock << inode->i_blkbits;
ASSERT(bh_result->b_size >= (1 << inode->i_blkbits));
size = bh_result->b_size;
if (!create && direct && offset >= i_size_read(inode))
return 0;
/*
* Direct I/O is usually done on preallocated files, so try getting
* a block mapping without an exclusive lock first. For buffered
* writes we already have the exclusive iolock anyway, so avoiding
* a lock roundtrip here by taking the ilock exclusive from the
* beginning is a useful micro optimization.
*/
if (create && !direct) {
lockmode = XFS_ILOCK_EXCL;
xfs_ilock(ip, lockmode);
} else {
lockmode = xfs_ilock_map_shared(ip);
}
ASSERT(offset <= mp->m_super->s_maxbytes);
if (offset + size > mp->m_super->s_maxbytes)
size = mp->m_super->s_maxbytes - offset;
end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + size);
offset_fsb = XFS_B_TO_FSBT(mp, offset);
error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb,
&imap, &nimaps, XFS_BMAPI_ENTIRE);
if (error)
goto out_unlock;
if (create &&
(!nimaps ||
(imap.br_startblock == HOLESTARTBLOCK ||
imap.br_startblock == DELAYSTARTBLOCK))) {
xfs: Use preallocation for inodes with extsz hints xfstest 229 exposes a problem with buffered IO, delayed allocation and extent size hints. That is when we do delayed allocation during buffered IO, we reserve space for the extent size hint alignment and allocate the physical space to align the extent, but we do not zero the regions of the extent that aren't written by the write(2) syscall. The result is that we expose stale data in unwritten regions of the extent size hints. There are two ways to fix this. The first is to detect that we are doing unaligned writes, check if there is already a mapping or data over the extent size hint range, and if not zero the page cache first before then doing the real write. This can be very expensive for large extent size hints, especially if the subsequent writes fill then entire extent size before the data is written to disk. The second, and simpler way, is simply to turn off delayed allocation when the extent size hint is set and use preallocation instead. This results in unwritten extents being laid down on disk and so only the written portions will be converted. This matches the behaviour for direct IO, and will also work for the real time device. The disadvantage of this approach is that for small extent size hints we can get file fragmentation, but in general extent size hints are fairly large (e.g. stripe width sized) so this isn't a big deal. Implement the second approach as it is simple and effective. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Mark Tinguely <tinguely@sgi.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2012-04-23 05:58:44 +00:00
if (direct || xfs_get_extsz_hint(ip)) {
/*
* Drop the ilock in preparation for starting the block
* allocation transaction. It will be retaken
* exclusively inside xfs_iomap_write_direct for the
* actual allocation.
*/
xfs_iunlock(ip, lockmode);
error = xfs_iomap_write_direct(ip, offset, size,
&imap, nimaps);
if (error)
return -error;
xfs: punch new delalloc blocks out of failed writes inside EOF. When a partial write inside EOF fails, it can leave delayed allocation blocks lying around because they don't get punched back out. This leads to assert failures like: XFS: Assertion failed: XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0, file: fs/xfs/xfs_super.c, line: 847 when evicting inodes from the cache. This can be trivially triggered by xfstests 083, which takes between 5 and 15 executions on a 512 byte block size filesystem to trip over this. Debugging shows a failed write due to ENOSPC calling xfs_vm_write_failed such as: [ 5012.329024] ino 0xa0026: vwf to 0x17000, sze 0x1c85ae and no action is taken on it. This leaves behind a delayed allocation extent that has no page covering it and no data in it: [ 5015.867162] ino 0xa0026: blks: 0x83 delay blocks 0x1, size 0x2538c0 [ 5015.868293] ext 0: off 0x4a, fsb 0x50306, len 0x1 [ 5015.869095] ext 1: off 0x4b, fsb 0x7899, len 0x6b [ 5015.869900] ext 2: off 0xb6, fsb 0xffffffffe0008, len 0x1 ^^^^^^^^^^^^^^^ [ 5015.871027] ext 3: off 0x36e, fsb 0x7a27, len 0xd [ 5015.872206] ext 4: off 0x4cf, fsb 0x7a1d, len 0xa So the delayed allocation extent is one block long at offset 0x16c00. Tracing shows that a bigger write: xfs_file_buffered_write: size 0x1c85ae offset 0x959d count 0x1ca3f ioflags allocates the block, and then fails with ENOSPC trying to allocate the last block on the page, leading to a failed write with stale delalloc blocks on it. Because we've had an ENOSPC when trying to allocate 0x16e00, it means that we are never goinge to call ->write_end on the page and so the allocated new buffer will not get marked dirty or have the buffer_new state cleared. In other works, what the above write is supposed to end up with is this mapping for the page: +------+------+------+------+------+------+------+------+ UMA UMA UMA UMA UMA UMA UND FAIL where: U = uptodate M = mapped N = new A = allocated D = delalloc FAIL = block we ENOSPC'd on. and the key point being the buffer_new() state for the newly allocated delayed allocation block. Except it doesn't - we're not marking buffers new correctly. That buffer_new() problem goes back to the xfs_iomap removal days, where xfs_iomap() used to return a "new" status for any map with newly allocated blocks, so that __xfs_get_blocks() could call set_buffer_new() on it. We still have the "new" variable and the check for it in the set_buffer_new() logic - except we never set it now! Hence that newly allocated delalloc block doesn't have the new flag set on it, so when the write fails we cannot tell which blocks we are supposed to punch out. WHy do we need the buffer_new flag? Well, that's because we can have this case: +------+------+------+------+------+------+------+------+ UMD UMD UMD UMD UMD UMD UND FAIL where all the UMD buffers contain valid data from a previously successful write() system call. We only want to punch the UND buffer because that's the only one that we added in this write and it was only this write that failed. That implies that even the old buffer_new() logic was wrong - because it would result in all those UMD buffers on the page having set_buffer_new() called on them even though they aren't new. Hence we shoul donly be calling set_buffer_new() for delalloc buffers that were allocated (i.e. were a hole before xfs_iomap_write_delay() was called). So, fix this set_buffer_new logic according to how we need it to work for handling failed writes correctly. Also, restore the new buffer logic handling for blocks allocated via xfs_iomap_write_direct(), because it should still set the buffer_new flag appropriately for newly allocated blocks, too. SO, now we have the buffer_new() being set appropriately in __xfs_get_blocks(), we can detect the exact delalloc ranges that we allocated in a failed write, and hence can now do a walk of the buffers on a page to find them. Except, it's not that easy. When block_write_begin() fails, it unlocks and releases the page that we just had an error on, so we can't use that page to handle errors anymore. We have to get access to the page while it is still locked to walk the buffers. Hence we have to open code block_write_begin() in xfs_vm_write_begin() to be able to insert xfs_vm_write_failed() is the right place. With that, we can pass the page and write range to xfs_vm_write_failed() and walk the buffers on the page, looking for delalloc buffers that are either new or beyond EOF and punch them out. Handling buffers beyond EOF ensures we still handle the existing case that xfs_vm_write_failed() handles. Of special note is the truncate_pagecache() handling - that only should be done for pages outside EOF - pages within EOF can still contain valid, dirty data so we must not punch them out of the cache. That just leaves the xfs_vm_write_end() failure handling. The only failure case here is that we didn't copy the entire range, and generic_write_end() handles that by zeroing the region of the page that wasn't copied, we don't have to punch out blocks within the file because they are guaranteed to contain zeros. Hence we only have to handle the existing "beyond EOF" case and don't need access to the buffers on the page. Hence it remains largely unchanged. Note that xfs_getbmap() can still trip over delalloc blocks beyond EOF that are left there by speculative delayed allocation. Hence this bug fix does not solve all known issues with bmap vs delalloc, but it does fix all the the known accidental occurances of the problem. Signed-off-by: Dave Chinner <david@fromorbit.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2012-04-27 09:45:21 +00:00
new = 1;
} else {
/*
* Delalloc reservations do not require a transaction,
xfs: punch new delalloc blocks out of failed writes inside EOF. When a partial write inside EOF fails, it can leave delayed allocation blocks lying around because they don't get punched back out. This leads to assert failures like: XFS: Assertion failed: XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0, file: fs/xfs/xfs_super.c, line: 847 when evicting inodes from the cache. This can be trivially triggered by xfstests 083, which takes between 5 and 15 executions on a 512 byte block size filesystem to trip over this. Debugging shows a failed write due to ENOSPC calling xfs_vm_write_failed such as: [ 5012.329024] ino 0xa0026: vwf to 0x17000, sze 0x1c85ae and no action is taken on it. This leaves behind a delayed allocation extent that has no page covering it and no data in it: [ 5015.867162] ino 0xa0026: blks: 0x83 delay blocks 0x1, size 0x2538c0 [ 5015.868293] ext 0: off 0x4a, fsb 0x50306, len 0x1 [ 5015.869095] ext 1: off 0x4b, fsb 0x7899, len 0x6b [ 5015.869900] ext 2: off 0xb6, fsb 0xffffffffe0008, len 0x1 ^^^^^^^^^^^^^^^ [ 5015.871027] ext 3: off 0x36e, fsb 0x7a27, len 0xd [ 5015.872206] ext 4: off 0x4cf, fsb 0x7a1d, len 0xa So the delayed allocation extent is one block long at offset 0x16c00. Tracing shows that a bigger write: xfs_file_buffered_write: size 0x1c85ae offset 0x959d count 0x1ca3f ioflags allocates the block, and then fails with ENOSPC trying to allocate the last block on the page, leading to a failed write with stale delalloc blocks on it. Because we've had an ENOSPC when trying to allocate 0x16e00, it means that we are never goinge to call ->write_end on the page and so the allocated new buffer will not get marked dirty or have the buffer_new state cleared. In other works, what the above write is supposed to end up with is this mapping for the page: +------+------+------+------+------+------+------+------+ UMA UMA UMA UMA UMA UMA UND FAIL where: U = uptodate M = mapped N = new A = allocated D = delalloc FAIL = block we ENOSPC'd on. and the key point being the buffer_new() state for the newly allocated delayed allocation block. Except it doesn't - we're not marking buffers new correctly. That buffer_new() problem goes back to the xfs_iomap removal days, where xfs_iomap() used to return a "new" status for any map with newly allocated blocks, so that __xfs_get_blocks() could call set_buffer_new() on it. We still have the "new" variable and the check for it in the set_buffer_new() logic - except we never set it now! Hence that newly allocated delalloc block doesn't have the new flag set on it, so when the write fails we cannot tell which blocks we are supposed to punch out. WHy do we need the buffer_new flag? Well, that's because we can have this case: +------+------+------+------+------+------+------+------+ UMD UMD UMD UMD UMD UMD UND FAIL where all the UMD buffers contain valid data from a previously successful write() system call. We only want to punch the UND buffer because that's the only one that we added in this write and it was only this write that failed. That implies that even the old buffer_new() logic was wrong - because it would result in all those UMD buffers on the page having set_buffer_new() called on them even though they aren't new. Hence we shoul donly be calling set_buffer_new() for delalloc buffers that were allocated (i.e. were a hole before xfs_iomap_write_delay() was called). So, fix this set_buffer_new logic according to how we need it to work for handling failed writes correctly. Also, restore the new buffer logic handling for blocks allocated via xfs_iomap_write_direct(), because it should still set the buffer_new flag appropriately for newly allocated blocks, too. SO, now we have the buffer_new() being set appropriately in __xfs_get_blocks(), we can detect the exact delalloc ranges that we allocated in a failed write, and hence can now do a walk of the buffers on a page to find them. Except, it's not that easy. When block_write_begin() fails, it unlocks and releases the page that we just had an error on, so we can't use that page to handle errors anymore. We have to get access to the page while it is still locked to walk the buffers. Hence we have to open code block_write_begin() in xfs_vm_write_begin() to be able to insert xfs_vm_write_failed() is the right place. With that, we can pass the page and write range to xfs_vm_write_failed() and walk the buffers on the page, looking for delalloc buffers that are either new or beyond EOF and punch them out. Handling buffers beyond EOF ensures we still handle the existing case that xfs_vm_write_failed() handles. Of special note is the truncate_pagecache() handling - that only should be done for pages outside EOF - pages within EOF can still contain valid, dirty data so we must not punch them out of the cache. That just leaves the xfs_vm_write_end() failure handling. The only failure case here is that we didn't copy the entire range, and generic_write_end() handles that by zeroing the region of the page that wasn't copied, we don't have to punch out blocks within the file because they are guaranteed to contain zeros. Hence we only have to handle the existing "beyond EOF" case and don't need access to the buffers on the page. Hence it remains largely unchanged. Note that xfs_getbmap() can still trip over delalloc blocks beyond EOF that are left there by speculative delayed allocation. Hence this bug fix does not solve all known issues with bmap vs delalloc, but it does fix all the the known accidental occurances of the problem. Signed-off-by: Dave Chinner <david@fromorbit.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2012-04-27 09:45:21 +00:00
* we can go on without dropping the lock here. If we
* are allocating a new delalloc block, make sure that
* we set the new flag so that we mark the buffer new so
* that we know that it is newly allocated if the write
* fails.
*/
xfs: punch new delalloc blocks out of failed writes inside EOF. When a partial write inside EOF fails, it can leave delayed allocation blocks lying around because they don't get punched back out. This leads to assert failures like: XFS: Assertion failed: XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0, file: fs/xfs/xfs_super.c, line: 847 when evicting inodes from the cache. This can be trivially triggered by xfstests 083, which takes between 5 and 15 executions on a 512 byte block size filesystem to trip over this. Debugging shows a failed write due to ENOSPC calling xfs_vm_write_failed such as: [ 5012.329024] ino 0xa0026: vwf to 0x17000, sze 0x1c85ae and no action is taken on it. This leaves behind a delayed allocation extent that has no page covering it and no data in it: [ 5015.867162] ino 0xa0026: blks: 0x83 delay blocks 0x1, size 0x2538c0 [ 5015.868293] ext 0: off 0x4a, fsb 0x50306, len 0x1 [ 5015.869095] ext 1: off 0x4b, fsb 0x7899, len 0x6b [ 5015.869900] ext 2: off 0xb6, fsb 0xffffffffe0008, len 0x1 ^^^^^^^^^^^^^^^ [ 5015.871027] ext 3: off 0x36e, fsb 0x7a27, len 0xd [ 5015.872206] ext 4: off 0x4cf, fsb 0x7a1d, len 0xa So the delayed allocation extent is one block long at offset 0x16c00. Tracing shows that a bigger write: xfs_file_buffered_write: size 0x1c85ae offset 0x959d count 0x1ca3f ioflags allocates the block, and then fails with ENOSPC trying to allocate the last block on the page, leading to a failed write with stale delalloc blocks on it. Because we've had an ENOSPC when trying to allocate 0x16e00, it means that we are never goinge to call ->write_end on the page and so the allocated new buffer will not get marked dirty or have the buffer_new state cleared. In other works, what the above write is supposed to end up with is this mapping for the page: +------+------+------+------+------+------+------+------+ UMA UMA UMA UMA UMA UMA UND FAIL where: U = uptodate M = mapped N = new A = allocated D = delalloc FAIL = block we ENOSPC'd on. and the key point being the buffer_new() state for the newly allocated delayed allocation block. Except it doesn't - we're not marking buffers new correctly. That buffer_new() problem goes back to the xfs_iomap removal days, where xfs_iomap() used to return a "new" status for any map with newly allocated blocks, so that __xfs_get_blocks() could call set_buffer_new() on it. We still have the "new" variable and the check for it in the set_buffer_new() logic - except we never set it now! Hence that newly allocated delalloc block doesn't have the new flag set on it, so when the write fails we cannot tell which blocks we are supposed to punch out. WHy do we need the buffer_new flag? Well, that's because we can have this case: +------+------+------+------+------+------+------+------+ UMD UMD UMD UMD UMD UMD UND FAIL where all the UMD buffers contain valid data from a previously successful write() system call. We only want to punch the UND buffer because that's the only one that we added in this write and it was only this write that failed. That implies that even the old buffer_new() logic was wrong - because it would result in all those UMD buffers on the page having set_buffer_new() called on them even though they aren't new. Hence we shoul donly be calling set_buffer_new() for delalloc buffers that were allocated (i.e. were a hole before xfs_iomap_write_delay() was called). So, fix this set_buffer_new logic according to how we need it to work for handling failed writes correctly. Also, restore the new buffer logic handling for blocks allocated via xfs_iomap_write_direct(), because it should still set the buffer_new flag appropriately for newly allocated blocks, too. SO, now we have the buffer_new() being set appropriately in __xfs_get_blocks(), we can detect the exact delalloc ranges that we allocated in a failed write, and hence can now do a walk of the buffers on a page to find them. Except, it's not that easy. When block_write_begin() fails, it unlocks and releases the page that we just had an error on, so we can't use that page to handle errors anymore. We have to get access to the page while it is still locked to walk the buffers. Hence we have to open code block_write_begin() in xfs_vm_write_begin() to be able to insert xfs_vm_write_failed() is the right place. With that, we can pass the page and write range to xfs_vm_write_failed() and walk the buffers on the page, looking for delalloc buffers that are either new or beyond EOF and punch them out. Handling buffers beyond EOF ensures we still handle the existing case that xfs_vm_write_failed() handles. Of special note is the truncate_pagecache() handling - that only should be done for pages outside EOF - pages within EOF can still contain valid, dirty data so we must not punch them out of the cache. That just leaves the xfs_vm_write_end() failure handling. The only failure case here is that we didn't copy the entire range, and generic_write_end() handles that by zeroing the region of the page that wasn't copied, we don't have to punch out blocks within the file because they are guaranteed to contain zeros. Hence we only have to handle the existing "beyond EOF" case and don't need access to the buffers on the page. Hence it remains largely unchanged. Note that xfs_getbmap() can still trip over delalloc blocks beyond EOF that are left there by speculative delayed allocation. Hence this bug fix does not solve all known issues with bmap vs delalloc, but it does fix all the the known accidental occurances of the problem. Signed-off-by: Dave Chinner <david@fromorbit.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2012-04-27 09:45:21 +00:00
if (nimaps && imap.br_startblock == HOLESTARTBLOCK)
new = 1;
error = xfs_iomap_write_delay(ip, offset, size, &imap);
if (error)
goto out_unlock;
xfs_iunlock(ip, lockmode);
}
trace_xfs_get_blocks_alloc(ip, offset, size, 0, &imap);
} else if (nimaps) {
trace_xfs_get_blocks_found(ip, offset, size, 0, &imap);
xfs_iunlock(ip, lockmode);
} else {
trace_xfs_get_blocks_notfound(ip, offset, size);
goto out_unlock;
}
if (imap.br_startblock != HOLESTARTBLOCK &&
imap.br_startblock != DELAYSTARTBLOCK) {
/*
* For unwritten extents do not report a disk address on
* the read case (treat as if we're reading into a hole).
*/
if (create || !ISUNWRITTEN(&imap))
xfs_map_buffer(inode, bh_result, &imap, offset);
if (create && ISUNWRITTEN(&imap)) {
if (direct) {
bh_result->b_private = inode;
set_buffer_defer_completion(bh_result);
}
set_buffer_unwritten(bh_result);
}
}
/*
* If this is a realtime file, data may be on a different device.
* to that pointed to from the buffer_head b_bdev currently.
*/
bh_result->b_bdev = xfs_find_bdev_for_inode(inode);
/*
* If we previously allocated a block out beyond eof and we are now
* coming back to use it then we will need to flag it as new even if it
* has a disk address.
*
* With sub-block writes into unwritten extents we also need to mark
* the buffer as new so that the unwritten parts of the buffer gets
* correctly zeroed.
*/
if (create &&
((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) ||
(offset >= i_size_read(inode)) ||
(new || ISUNWRITTEN(&imap))))
set_buffer_new(bh_result);
if (imap.br_startblock == DELAYSTARTBLOCK) {
BUG_ON(direct);
if (create) {
set_buffer_uptodate(bh_result);
set_buffer_mapped(bh_result);
set_buffer_delay(bh_result);
}
}
/*
* If this is O_DIRECT or the mpage code calling tell them how large
* the mapping is, so that we can avoid repeated get_blocks calls.
*/
if (direct || size > (1 << inode->i_blkbits)) {
xfs_off_t mapping_size;
mapping_size = imap.br_startoff + imap.br_blockcount - iblock;
mapping_size <<= inode->i_blkbits;
ASSERT(mapping_size > 0);
if (mapping_size > size)
mapping_size = size;
if (mapping_size > LONG_MAX)
mapping_size = LONG_MAX;
bh_result->b_size = mapping_size;
}
return 0;
out_unlock:
xfs_iunlock(ip, lockmode);
return -error;
}
int
xfs_get_blocks(
struct inode *inode,
sector_t iblock,
struct buffer_head *bh_result,
int create)
{
return __xfs_get_blocks(inode, iblock, bh_result, create, 0);
}
STATIC int
xfs_get_blocks_direct(
struct inode *inode,
sector_t iblock,
struct buffer_head *bh_result,
int create)
{
return __xfs_get_blocks(inode, iblock, bh_result, create, 1);
}
/*
* Complete a direct I/O write request.
*
* If the private argument is non-NULL __xfs_get_blocks signals us that we
* need to issue a transaction to convert the range from unwritten to written
* extents. In case this is regular synchronous I/O we just call xfs_end_io
* to do this and we are done. But in case this was a successful AIO
* request this handler is called from interrupt context, from which we
* can't start transactions. In that case offload the I/O completion to
* the workqueues we also use for buffered I/O completion.
*/
STATIC void
xfs_end_io_direct_write(
struct kiocb *iocb,
loff_t offset,
ssize_t size,
void *private)
{
struct xfs_ioend *ioend = iocb->private;
/*
* While the generic direct I/O code updates the inode size, it does
* so only after the end_io handler is called, which means our
* end_io handler thinks the on-disk size is outside the in-core
* size. To prevent this just update it a little bit earlier here.
*/
if (offset + size > i_size_read(ioend->io_inode))
i_size_write(ioend->io_inode, offset + size);
/*
* blockdev_direct_IO can return an error even after the I/O
* completion handler was called. Thus we need to protect
* against double-freeing.
*/
iocb->private = NULL;
[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 03:49:46 +00:00
ioend->io_offset = offset;
ioend->io_size = size;
if (private && size > 0)
ioend->io_type = XFS_IO_UNWRITTEN;
xfs_finish_ioend_sync(ioend);
}
STATIC ssize_t
xfs_vm_direct_IO(
int rw,
struct kiocb *iocb,
const struct iovec *iov,
loff_t offset,
unsigned long nr_segs)
{
struct inode *inode = iocb->ki_filp->f_mapping->host;
struct block_device *bdev = xfs_find_bdev_for_inode(inode);
struct xfs_ioend *ioend = NULL;
ssize_t ret;
if (rw & WRITE) {
size_t size = iov_length(iov, nr_segs);
/*
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 02:01:00 +00:00
* We cannot preallocate a size update transaction here as we
* don't know whether allocation is necessary or not. Hence we
* can only tell IO completion that one is necessary if we are
* not doing unwritten extent conversion.
*/
iocb->private = ioend = xfs_alloc_ioend(inode, XFS_IO_DIRECT);
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 02:01:00 +00:00
if (offset + size > XFS_I(inode)->i_d.di_size)
ioend->io_isdirect = 1;
ret = __blockdev_direct_IO(rw, iocb, inode, bdev, iov,
offset, nr_segs,
xfs_get_blocks_direct,
xfs_end_io_direct_write, NULL, 0);
if (ret != -EIOCBQUEUED && iocb->private)
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 02:01:00 +00:00
goto out_destroy_ioend;
} else {
ret = __blockdev_direct_IO(rw, iocb, inode, bdev, iov,
offset, nr_segs,
xfs_get_blocks_direct,
NULL, NULL, 0);
}
return ret;
out_destroy_ioend:
xfs_destroy_ioend(ioend);
return ret;
}
xfs: punch new delalloc blocks out of failed writes inside EOF. When a partial write inside EOF fails, it can leave delayed allocation blocks lying around because they don't get punched back out. This leads to assert failures like: XFS: Assertion failed: XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0, file: fs/xfs/xfs_super.c, line: 847 when evicting inodes from the cache. This can be trivially triggered by xfstests 083, which takes between 5 and 15 executions on a 512 byte block size filesystem to trip over this. Debugging shows a failed write due to ENOSPC calling xfs_vm_write_failed such as: [ 5012.329024] ino 0xa0026: vwf to 0x17000, sze 0x1c85ae and no action is taken on it. This leaves behind a delayed allocation extent that has no page covering it and no data in it: [ 5015.867162] ino 0xa0026: blks: 0x83 delay blocks 0x1, size 0x2538c0 [ 5015.868293] ext 0: off 0x4a, fsb 0x50306, len 0x1 [ 5015.869095] ext 1: off 0x4b, fsb 0x7899, len 0x6b [ 5015.869900] ext 2: off 0xb6, fsb 0xffffffffe0008, len 0x1 ^^^^^^^^^^^^^^^ [ 5015.871027] ext 3: off 0x36e, fsb 0x7a27, len 0xd [ 5015.872206] ext 4: off 0x4cf, fsb 0x7a1d, len 0xa So the delayed allocation extent is one block long at offset 0x16c00. Tracing shows that a bigger write: xfs_file_buffered_write: size 0x1c85ae offset 0x959d count 0x1ca3f ioflags allocates the block, and then fails with ENOSPC trying to allocate the last block on the page, leading to a failed write with stale delalloc blocks on it. Because we've had an ENOSPC when trying to allocate 0x16e00, it means that we are never goinge to call ->write_end on the page and so the allocated new buffer will not get marked dirty or have the buffer_new state cleared. In other works, what the above write is supposed to end up with is this mapping for the page: +------+------+------+------+------+------+------+------+ UMA UMA UMA UMA UMA UMA UND FAIL where: U = uptodate M = mapped N = new A = allocated D = delalloc FAIL = block we ENOSPC'd on. and the key point being the buffer_new() state for the newly allocated delayed allocation block. Except it doesn't - we're not marking buffers new correctly. That buffer_new() problem goes back to the xfs_iomap removal days, where xfs_iomap() used to return a "new" status for any map with newly allocated blocks, so that __xfs_get_blocks() could call set_buffer_new() on it. We still have the "new" variable and the check for it in the set_buffer_new() logic - except we never set it now! Hence that newly allocated delalloc block doesn't have the new flag set on it, so when the write fails we cannot tell which blocks we are supposed to punch out. WHy do we need the buffer_new flag? Well, that's because we can have this case: +------+------+------+------+------+------+------+------+ UMD UMD UMD UMD UMD UMD UND FAIL where all the UMD buffers contain valid data from a previously successful write() system call. We only want to punch the UND buffer because that's the only one that we added in this write and it was only this write that failed. That implies that even the old buffer_new() logic was wrong - because it would result in all those UMD buffers on the page having set_buffer_new() called on them even though they aren't new. Hence we shoul donly be calling set_buffer_new() for delalloc buffers that were allocated (i.e. were a hole before xfs_iomap_write_delay() was called). So, fix this set_buffer_new logic according to how we need it to work for handling failed writes correctly. Also, restore the new buffer logic handling for blocks allocated via xfs_iomap_write_direct(), because it should still set the buffer_new flag appropriately for newly allocated blocks, too. SO, now we have the buffer_new() being set appropriately in __xfs_get_blocks(), we can detect the exact delalloc ranges that we allocated in a failed write, and hence can now do a walk of the buffers on a page to find them. Except, it's not that easy. When block_write_begin() fails, it unlocks and releases the page that we just had an error on, so we can't use that page to handle errors anymore. We have to get access to the page while it is still locked to walk the buffers. Hence we have to open code block_write_begin() in xfs_vm_write_begin() to be able to insert xfs_vm_write_failed() is the right place. With that, we can pass the page and write range to xfs_vm_write_failed() and walk the buffers on the page, looking for delalloc buffers that are either new or beyond EOF and punch them out. Handling buffers beyond EOF ensures we still handle the existing case that xfs_vm_write_failed() handles. Of special note is the truncate_pagecache() handling - that only should be done for pages outside EOF - pages within EOF can still contain valid, dirty data so we must not punch them out of the cache. That just leaves the xfs_vm_write_end() failure handling. The only failure case here is that we didn't copy the entire range, and generic_write_end() handles that by zeroing the region of the page that wasn't copied, we don't have to punch out blocks within the file because they are guaranteed to contain zeros. Hence we only have to handle the existing "beyond EOF" case and don't need access to the buffers on the page. Hence it remains largely unchanged. Note that xfs_getbmap() can still trip over delalloc blocks beyond EOF that are left there by speculative delayed allocation. Hence this bug fix does not solve all known issues with bmap vs delalloc, but it does fix all the the known accidental occurances of the problem. Signed-off-by: Dave Chinner <david@fromorbit.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2012-04-27 09:45:21 +00:00
/*
* Punch out the delalloc blocks we have already allocated.
*
* Don't bother with xfs_setattr given that nothing can have made it to disk yet
* as the page is still locked at this point.
*/
STATIC void
xfs_vm_kill_delalloc_range(
struct inode *inode,
loff_t start,
loff_t end)
{
struct xfs_inode *ip = XFS_I(inode);
xfs_fileoff_t start_fsb;
xfs_fileoff_t end_fsb;
int error;
start_fsb = XFS_B_TO_FSB(ip->i_mount, start);
end_fsb = XFS_B_TO_FSB(ip->i_mount, end);
if (end_fsb <= start_fsb)
return;
xfs_ilock(ip, XFS_ILOCK_EXCL);
error = xfs_bmap_punch_delalloc_range(ip, start_fsb,
end_fsb - start_fsb);
if (error) {
/* something screwed, just bail */
if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
xfs_alert(ip->i_mount,
"xfs_vm_write_failed: unable to clean up ino %lld",
ip->i_ino);
}
}
xfs_iunlock(ip, XFS_ILOCK_EXCL);
}
STATIC void
xfs_vm_write_failed(
xfs: punch new delalloc blocks out of failed writes inside EOF. When a partial write inside EOF fails, it can leave delayed allocation blocks lying around because they don't get punched back out. This leads to assert failures like: XFS: Assertion failed: XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0, file: fs/xfs/xfs_super.c, line: 847 when evicting inodes from the cache. This can be trivially triggered by xfstests 083, which takes between 5 and 15 executions on a 512 byte block size filesystem to trip over this. Debugging shows a failed write due to ENOSPC calling xfs_vm_write_failed such as: [ 5012.329024] ino 0xa0026: vwf to 0x17000, sze 0x1c85ae and no action is taken on it. This leaves behind a delayed allocation extent that has no page covering it and no data in it: [ 5015.867162] ino 0xa0026: blks: 0x83 delay blocks 0x1, size 0x2538c0 [ 5015.868293] ext 0: off 0x4a, fsb 0x50306, len 0x1 [ 5015.869095] ext 1: off 0x4b, fsb 0x7899, len 0x6b [ 5015.869900] ext 2: off 0xb6, fsb 0xffffffffe0008, len 0x1 ^^^^^^^^^^^^^^^ [ 5015.871027] ext 3: off 0x36e, fsb 0x7a27, len 0xd [ 5015.872206] ext 4: off 0x4cf, fsb 0x7a1d, len 0xa So the delayed allocation extent is one block long at offset 0x16c00. Tracing shows that a bigger write: xfs_file_buffered_write: size 0x1c85ae offset 0x959d count 0x1ca3f ioflags allocates the block, and then fails with ENOSPC trying to allocate the last block on the page, leading to a failed write with stale delalloc blocks on it. Because we've had an ENOSPC when trying to allocate 0x16e00, it means that we are never goinge to call ->write_end on the page and so the allocated new buffer will not get marked dirty or have the buffer_new state cleared. In other works, what the above write is supposed to end up with is this mapping for the page: +------+------+------+------+------+------+------+------+ UMA UMA UMA UMA UMA UMA UND FAIL where: U = uptodate M = mapped N = new A = allocated D = delalloc FAIL = block we ENOSPC'd on. and the key point being the buffer_new() state for the newly allocated delayed allocation block. Except it doesn't - we're not marking buffers new correctly. That buffer_new() problem goes back to the xfs_iomap removal days, where xfs_iomap() used to return a "new" status for any map with newly allocated blocks, so that __xfs_get_blocks() could call set_buffer_new() on it. We still have the "new" variable and the check for it in the set_buffer_new() logic - except we never set it now! Hence that newly allocated delalloc block doesn't have the new flag set on it, so when the write fails we cannot tell which blocks we are supposed to punch out. WHy do we need the buffer_new flag? Well, that's because we can have this case: +------+------+------+------+------+------+------+------+ UMD UMD UMD UMD UMD UMD UND FAIL where all the UMD buffers contain valid data from a previously successful write() system call. We only want to punch the UND buffer because that's the only one that we added in this write and it was only this write that failed. That implies that even the old buffer_new() logic was wrong - because it would result in all those UMD buffers on the page having set_buffer_new() called on them even though they aren't new. Hence we shoul donly be calling set_buffer_new() for delalloc buffers that were allocated (i.e. were a hole before xfs_iomap_write_delay() was called). So, fix this set_buffer_new logic according to how we need it to work for handling failed writes correctly. Also, restore the new buffer logic handling for blocks allocated via xfs_iomap_write_direct(), because it should still set the buffer_new flag appropriately for newly allocated blocks, too. SO, now we have the buffer_new() being set appropriately in __xfs_get_blocks(), we can detect the exact delalloc ranges that we allocated in a failed write, and hence can now do a walk of the buffers on a page to find them. Except, it's not that easy. When block_write_begin() fails, it unlocks and releases the page that we just had an error on, so we can't use that page to handle errors anymore. We have to get access to the page while it is still locked to walk the buffers. Hence we have to open code block_write_begin() in xfs_vm_write_begin() to be able to insert xfs_vm_write_failed() is the right place. With that, we can pass the page and write range to xfs_vm_write_failed() and walk the buffers on the page, looking for delalloc buffers that are either new or beyond EOF and punch them out. Handling buffers beyond EOF ensures we still handle the existing case that xfs_vm_write_failed() handles. Of special note is the truncate_pagecache() handling - that only should be done for pages outside EOF - pages within EOF can still contain valid, dirty data so we must not punch them out of the cache. That just leaves the xfs_vm_write_end() failure handling. The only failure case here is that we didn't copy the entire range, and generic_write_end() handles that by zeroing the region of the page that wasn't copied, we don't have to punch out blocks within the file because they are guaranteed to contain zeros. Hence we only have to handle the existing "beyond EOF" case and don't need access to the buffers on the page. Hence it remains largely unchanged. Note that xfs_getbmap() can still trip over delalloc blocks beyond EOF that are left there by speculative delayed allocation. Hence this bug fix does not solve all known issues with bmap vs delalloc, but it does fix all the the known accidental occurances of the problem. Signed-off-by: Dave Chinner <david@fromorbit.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2012-04-27 09:45:21 +00:00
struct inode *inode,
struct page *page,
loff_t pos,
unsigned len)
{
xfs: fix assertion failure in xfs_vm_write_failed() In xfs_vm_write_failed(), we evaluate the block_offset of pos with PAGE_MASK which is an unsigned long. That is fine on 64-bit platforms regardless of whether the request pos is 32-bit or 64-bit. However, on 32-bit platforms the value is 0xfffff000 and so the high 32 bits in it will be masked off with (pos & PAGE_MASK) for a 64-bit pos. As a result, the evaluated block_offset is incorrect which will cause this failure ASSERT(block_offset + from == pos); and potentially pass the wrong block to xfs_vm_kill_delalloc_range(). In this case, we can get a kernel panic if CONFIG_XFS_DEBUG is enabled: XFS: Assertion failed: block_offset + from == pos, file: fs/xfs/xfs_aops.c, line: 1504 ------------[ cut here ]------------ kernel BUG at fs/xfs/xfs_message.c:100! invalid opcode: 0000 [#1] SMP ........ Pid: 4057, comm: mkfs.xfs Tainted: G O 3.9.0-rc2 #1 EIP: 0060:[<f94a7e8b>] EFLAGS: 00010282 CPU: 0 EIP is at assfail+0x2b/0x30 [xfs] EAX: 00000056 EBX: f6ef28a0 ECX: 00000007 EDX: f57d22a4 ESI: 1c2fb000 EDI: 00000000 EBP: ea6b5d30 ESP: ea6b5d1c DS: 007b ES: 007b FS: 00d8 GS: 00e0 SS: 0068 CR0: 8005003b CR2: 094f3ff4 CR3: 2bcb4000 CR4: 000006f0 DR0: 00000000 DR1: 00000000 DR2: 00000000 DR3: 00000000 DR6: ffff0ff0 DR7: 00000400 Process mkfs.xfs (pid: 4057, ti=ea6b4000 task=ea5799e0 task.ti=ea6b4000) Stack: 00000000 f9525c48 f951fa80 f951f96b 000005e4 ea6b5d7c f9494b34 c19b0ea2 00000066 f3d6c620 c19b0ea2 00000000 e9a91458 00001000 00000000 00000000 00000000 c15c7e89 00000000 1c2fb000 00000000 00000000 1c2fb000 00000080 Call Trace: [<f9494b34>] xfs_vm_write_failed+0x74/0x1b0 [xfs] [<c15c7e89>] ? printk+0x4d/0x4f [<f9494d7d>] xfs_vm_write_begin+0x10d/0x170 [xfs] [<c110a34c>] generic_file_buffered_write+0xdc/0x210 [<f949b669>] xfs_file_buffered_aio_write+0xf9/0x190 [xfs] [<f949b7f3>] xfs_file_aio_write+0xf3/0x160 [xfs] [<c115e504>] do_sync_write+0x94/0xd0 [<c115ed1f>] vfs_write+0x8f/0x160 [<c115e470>] ? wait_on_retry_sync_kiocb+0x50/0x50 [<c115f017>] sys_write+0x47/0x80 [<c15d860d>] sysenter_do_call+0x12/0x28 ............. EIP: [<f94a7e8b>] assfail+0x2b/0x30 [xfs] SS:ESP 0068:ea6b5d1c ---[ end trace cdd9af4f4ecab42f ]--- Kernel panic - not syncing: Fatal exception In order to avoid this, we can evaluate the block_offset of the start of the page by using shifts rather than masks the mismatch problem. Thanks Dave Chinner for help finding and fixing this bug. Reported-by: Michael L. Semon <mlsemon35@gmail.com> Reviewed-by: Dave Chinner <david@fromorbit.com> Reviewed-by: Mark Tinguely <tinguely@sgi.com> Signed-off-by: Jie Liu <jeff.liu@oracle.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2013-07-16 05:11:16 +00:00
loff_t block_offset;
xfs: punch new delalloc blocks out of failed writes inside EOF. When a partial write inside EOF fails, it can leave delayed allocation blocks lying around because they don't get punched back out. This leads to assert failures like: XFS: Assertion failed: XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0, file: fs/xfs/xfs_super.c, line: 847 when evicting inodes from the cache. This can be trivially triggered by xfstests 083, which takes between 5 and 15 executions on a 512 byte block size filesystem to trip over this. Debugging shows a failed write due to ENOSPC calling xfs_vm_write_failed such as: [ 5012.329024] ino 0xa0026: vwf to 0x17000, sze 0x1c85ae and no action is taken on it. This leaves behind a delayed allocation extent that has no page covering it and no data in it: [ 5015.867162] ino 0xa0026: blks: 0x83 delay blocks 0x1, size 0x2538c0 [ 5015.868293] ext 0: off 0x4a, fsb 0x50306, len 0x1 [ 5015.869095] ext 1: off 0x4b, fsb 0x7899, len 0x6b [ 5015.869900] ext 2: off 0xb6, fsb 0xffffffffe0008, len 0x1 ^^^^^^^^^^^^^^^ [ 5015.871027] ext 3: off 0x36e, fsb 0x7a27, len 0xd [ 5015.872206] ext 4: off 0x4cf, fsb 0x7a1d, len 0xa So the delayed allocation extent is one block long at offset 0x16c00. Tracing shows that a bigger write: xfs_file_buffered_write: size 0x1c85ae offset 0x959d count 0x1ca3f ioflags allocates the block, and then fails with ENOSPC trying to allocate the last block on the page, leading to a failed write with stale delalloc blocks on it. Because we've had an ENOSPC when trying to allocate 0x16e00, it means that we are never goinge to call ->write_end on the page and so the allocated new buffer will not get marked dirty or have the buffer_new state cleared. In other works, what the above write is supposed to end up with is this mapping for the page: +------+------+------+------+------+------+------+------+ UMA UMA UMA UMA UMA UMA UND FAIL where: U = uptodate M = mapped N = new A = allocated D = delalloc FAIL = block we ENOSPC'd on. and the key point being the buffer_new() state for the newly allocated delayed allocation block. Except it doesn't - we're not marking buffers new correctly. That buffer_new() problem goes back to the xfs_iomap removal days, where xfs_iomap() used to return a "new" status for any map with newly allocated blocks, so that __xfs_get_blocks() could call set_buffer_new() on it. We still have the "new" variable and the check for it in the set_buffer_new() logic - except we never set it now! Hence that newly allocated delalloc block doesn't have the new flag set on it, so when the write fails we cannot tell which blocks we are supposed to punch out. WHy do we need the buffer_new flag? Well, that's because we can have this case: +------+------+------+------+------+------+------+------+ UMD UMD UMD UMD UMD UMD UND FAIL where all the UMD buffers contain valid data from a previously successful write() system call. We only want to punch the UND buffer because that's the only one that we added in this write and it was only this write that failed. That implies that even the old buffer_new() logic was wrong - because it would result in all those UMD buffers on the page having set_buffer_new() called on them even though they aren't new. Hence we shoul donly be calling set_buffer_new() for delalloc buffers that were allocated (i.e. were a hole before xfs_iomap_write_delay() was called). So, fix this set_buffer_new logic according to how we need it to work for handling failed writes correctly. Also, restore the new buffer logic handling for blocks allocated via xfs_iomap_write_direct(), because it should still set the buffer_new flag appropriately for newly allocated blocks, too. SO, now we have the buffer_new() being set appropriately in __xfs_get_blocks(), we can detect the exact delalloc ranges that we allocated in a failed write, and hence can now do a walk of the buffers on a page to find them. Except, it's not that easy. When block_write_begin() fails, it unlocks and releases the page that we just had an error on, so we can't use that page to handle errors anymore. We have to get access to the page while it is still locked to walk the buffers. Hence we have to open code block_write_begin() in xfs_vm_write_begin() to be able to insert xfs_vm_write_failed() is the right place. With that, we can pass the page and write range to xfs_vm_write_failed() and walk the buffers on the page, looking for delalloc buffers that are either new or beyond EOF and punch them out. Handling buffers beyond EOF ensures we still handle the existing case that xfs_vm_write_failed() handles. Of special note is the truncate_pagecache() handling - that only should be done for pages outside EOF - pages within EOF can still contain valid, dirty data so we must not punch them out of the cache. That just leaves the xfs_vm_write_end() failure handling. The only failure case here is that we didn't copy the entire range, and generic_write_end() handles that by zeroing the region of the page that wasn't copied, we don't have to punch out blocks within the file because they are guaranteed to contain zeros. Hence we only have to handle the existing "beyond EOF" case and don't need access to the buffers on the page. Hence it remains largely unchanged. Note that xfs_getbmap() can still trip over delalloc blocks beyond EOF that are left there by speculative delayed allocation. Hence this bug fix does not solve all known issues with bmap vs delalloc, but it does fix all the the known accidental occurances of the problem. Signed-off-by: Dave Chinner <david@fromorbit.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2012-04-27 09:45:21 +00:00
loff_t block_start;
loff_t block_end;
loff_t from = pos & (PAGE_CACHE_SIZE - 1);
loff_t to = from + len;
struct buffer_head *bh, *head;
xfs: fix assertion failure in xfs_vm_write_failed() In xfs_vm_write_failed(), we evaluate the block_offset of pos with PAGE_MASK which is an unsigned long. That is fine on 64-bit platforms regardless of whether the request pos is 32-bit or 64-bit. However, on 32-bit platforms the value is 0xfffff000 and so the high 32 bits in it will be masked off with (pos & PAGE_MASK) for a 64-bit pos. As a result, the evaluated block_offset is incorrect which will cause this failure ASSERT(block_offset + from == pos); and potentially pass the wrong block to xfs_vm_kill_delalloc_range(). In this case, we can get a kernel panic if CONFIG_XFS_DEBUG is enabled: XFS: Assertion failed: block_offset + from == pos, file: fs/xfs/xfs_aops.c, line: 1504 ------------[ cut here ]------------ kernel BUG at fs/xfs/xfs_message.c:100! invalid opcode: 0000 [#1] SMP ........ Pid: 4057, comm: mkfs.xfs Tainted: G O 3.9.0-rc2 #1 EIP: 0060:[<f94a7e8b>] EFLAGS: 00010282 CPU: 0 EIP is at assfail+0x2b/0x30 [xfs] EAX: 00000056 EBX: f6ef28a0 ECX: 00000007 EDX: f57d22a4 ESI: 1c2fb000 EDI: 00000000 EBP: ea6b5d30 ESP: ea6b5d1c DS: 007b ES: 007b FS: 00d8 GS: 00e0 SS: 0068 CR0: 8005003b CR2: 094f3ff4 CR3: 2bcb4000 CR4: 000006f0 DR0: 00000000 DR1: 00000000 DR2: 00000000 DR3: 00000000 DR6: ffff0ff0 DR7: 00000400 Process mkfs.xfs (pid: 4057, ti=ea6b4000 task=ea5799e0 task.ti=ea6b4000) Stack: 00000000 f9525c48 f951fa80 f951f96b 000005e4 ea6b5d7c f9494b34 c19b0ea2 00000066 f3d6c620 c19b0ea2 00000000 e9a91458 00001000 00000000 00000000 00000000 c15c7e89 00000000 1c2fb000 00000000 00000000 1c2fb000 00000080 Call Trace: [<f9494b34>] xfs_vm_write_failed+0x74/0x1b0 [xfs] [<c15c7e89>] ? printk+0x4d/0x4f [<f9494d7d>] xfs_vm_write_begin+0x10d/0x170 [xfs] [<c110a34c>] generic_file_buffered_write+0xdc/0x210 [<f949b669>] xfs_file_buffered_aio_write+0xf9/0x190 [xfs] [<f949b7f3>] xfs_file_aio_write+0xf3/0x160 [xfs] [<c115e504>] do_sync_write+0x94/0xd0 [<c115ed1f>] vfs_write+0x8f/0x160 [<c115e470>] ? wait_on_retry_sync_kiocb+0x50/0x50 [<c115f017>] sys_write+0x47/0x80 [<c15d860d>] sysenter_do_call+0x12/0x28 ............. EIP: [<f94a7e8b>] assfail+0x2b/0x30 [xfs] SS:ESP 0068:ea6b5d1c ---[ end trace cdd9af4f4ecab42f ]--- Kernel panic - not syncing: Fatal exception In order to avoid this, we can evaluate the block_offset of the start of the page by using shifts rather than masks the mismatch problem. Thanks Dave Chinner for help finding and fixing this bug. Reported-by: Michael L. Semon <mlsemon35@gmail.com> Reviewed-by: Dave Chinner <david@fromorbit.com> Reviewed-by: Mark Tinguely <tinguely@sgi.com> Signed-off-by: Jie Liu <jeff.liu@oracle.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2013-07-16 05:11:16 +00:00
/*
* The request pos offset might be 32 or 64 bit, this is all fine
* on 64-bit platform. However, for 64-bit pos request on 32-bit
* platform, the high 32-bit will be masked off if we evaluate the
* block_offset via (pos & PAGE_MASK) because the PAGE_MASK is
* 0xfffff000 as an unsigned long, hence the result is incorrect
* which could cause the following ASSERT failed in most cases.
* In order to avoid this, we can evaluate the block_offset of the
* start of the page by using shifts rather than masks the mismatch
* problem.
*/
block_offset = (pos >> PAGE_CACHE_SHIFT) << PAGE_CACHE_SHIFT;
xfs: punch new delalloc blocks out of failed writes inside EOF. When a partial write inside EOF fails, it can leave delayed allocation blocks lying around because they don't get punched back out. This leads to assert failures like: XFS: Assertion failed: XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0, file: fs/xfs/xfs_super.c, line: 847 when evicting inodes from the cache. This can be trivially triggered by xfstests 083, which takes between 5 and 15 executions on a 512 byte block size filesystem to trip over this. Debugging shows a failed write due to ENOSPC calling xfs_vm_write_failed such as: [ 5012.329024] ino 0xa0026: vwf to 0x17000, sze 0x1c85ae and no action is taken on it. This leaves behind a delayed allocation extent that has no page covering it and no data in it: [ 5015.867162] ino 0xa0026: blks: 0x83 delay blocks 0x1, size 0x2538c0 [ 5015.868293] ext 0: off 0x4a, fsb 0x50306, len 0x1 [ 5015.869095] ext 1: off 0x4b, fsb 0x7899, len 0x6b [ 5015.869900] ext 2: off 0xb6, fsb 0xffffffffe0008, len 0x1 ^^^^^^^^^^^^^^^ [ 5015.871027] ext 3: off 0x36e, fsb 0x7a27, len 0xd [ 5015.872206] ext 4: off 0x4cf, fsb 0x7a1d, len 0xa So the delayed allocation extent is one block long at offset 0x16c00. Tracing shows that a bigger write: xfs_file_buffered_write: size 0x1c85ae offset 0x959d count 0x1ca3f ioflags allocates the block, and then fails with ENOSPC trying to allocate the last block on the page, leading to a failed write with stale delalloc blocks on it. Because we've had an ENOSPC when trying to allocate 0x16e00, it means that we are never goinge to call ->write_end on the page and so the allocated new buffer will not get marked dirty or have the buffer_new state cleared. In other works, what the above write is supposed to end up with is this mapping for the page: +------+------+------+------+------+------+------+------+ UMA UMA UMA UMA UMA UMA UND FAIL where: U = uptodate M = mapped N = new A = allocated D = delalloc FAIL = block we ENOSPC'd on. and the key point being the buffer_new() state for the newly allocated delayed allocation block. Except it doesn't - we're not marking buffers new correctly. That buffer_new() problem goes back to the xfs_iomap removal days, where xfs_iomap() used to return a "new" status for any map with newly allocated blocks, so that __xfs_get_blocks() could call set_buffer_new() on it. We still have the "new" variable and the check for it in the set_buffer_new() logic - except we never set it now! Hence that newly allocated delalloc block doesn't have the new flag set on it, so when the write fails we cannot tell which blocks we are supposed to punch out. WHy do we need the buffer_new flag? Well, that's because we can have this case: +------+------+------+------+------+------+------+------+ UMD UMD UMD UMD UMD UMD UND FAIL where all the UMD buffers contain valid data from a previously successful write() system call. We only want to punch the UND buffer because that's the only one that we added in this write and it was only this write that failed. That implies that even the old buffer_new() logic was wrong - because it would result in all those UMD buffers on the page having set_buffer_new() called on them even though they aren't new. Hence we shoul donly be calling set_buffer_new() for delalloc buffers that were allocated (i.e. were a hole before xfs_iomap_write_delay() was called). So, fix this set_buffer_new logic according to how we need it to work for handling failed writes correctly. Also, restore the new buffer logic handling for blocks allocated via xfs_iomap_write_direct(), because it should still set the buffer_new flag appropriately for newly allocated blocks, too. SO, now we have the buffer_new() being set appropriately in __xfs_get_blocks(), we can detect the exact delalloc ranges that we allocated in a failed write, and hence can now do a walk of the buffers on a page to find them. Except, it's not that easy. When block_write_begin() fails, it unlocks and releases the page that we just had an error on, so we can't use that page to handle errors anymore. We have to get access to the page while it is still locked to walk the buffers. Hence we have to open code block_write_begin() in xfs_vm_write_begin() to be able to insert xfs_vm_write_failed() is the right place. With that, we can pass the page and write range to xfs_vm_write_failed() and walk the buffers on the page, looking for delalloc buffers that are either new or beyond EOF and punch them out. Handling buffers beyond EOF ensures we still handle the existing case that xfs_vm_write_failed() handles. Of special note is the truncate_pagecache() handling - that only should be done for pages outside EOF - pages within EOF can still contain valid, dirty data so we must not punch them out of the cache. That just leaves the xfs_vm_write_end() failure handling. The only failure case here is that we didn't copy the entire range, and generic_write_end() handles that by zeroing the region of the page that wasn't copied, we don't have to punch out blocks within the file because they are guaranteed to contain zeros. Hence we only have to handle the existing "beyond EOF" case and don't need access to the buffers on the page. Hence it remains largely unchanged. Note that xfs_getbmap() can still trip over delalloc blocks beyond EOF that are left there by speculative delayed allocation. Hence this bug fix does not solve all known issues with bmap vs delalloc, but it does fix all the the known accidental occurances of the problem. Signed-off-by: Dave Chinner <david@fromorbit.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2012-04-27 09:45:21 +00:00
ASSERT(block_offset + from == pos);
xfs: punch new delalloc blocks out of failed writes inside EOF. When a partial write inside EOF fails, it can leave delayed allocation blocks lying around because they don't get punched back out. This leads to assert failures like: XFS: Assertion failed: XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0, file: fs/xfs/xfs_super.c, line: 847 when evicting inodes from the cache. This can be trivially triggered by xfstests 083, which takes between 5 and 15 executions on a 512 byte block size filesystem to trip over this. Debugging shows a failed write due to ENOSPC calling xfs_vm_write_failed such as: [ 5012.329024] ino 0xa0026: vwf to 0x17000, sze 0x1c85ae and no action is taken on it. This leaves behind a delayed allocation extent that has no page covering it and no data in it: [ 5015.867162] ino 0xa0026: blks: 0x83 delay blocks 0x1, size 0x2538c0 [ 5015.868293] ext 0: off 0x4a, fsb 0x50306, len 0x1 [ 5015.869095] ext 1: off 0x4b, fsb 0x7899, len 0x6b [ 5015.869900] ext 2: off 0xb6, fsb 0xffffffffe0008, len 0x1 ^^^^^^^^^^^^^^^ [ 5015.871027] ext 3: off 0x36e, fsb 0x7a27, len 0xd [ 5015.872206] ext 4: off 0x4cf, fsb 0x7a1d, len 0xa So the delayed allocation extent is one block long at offset 0x16c00. Tracing shows that a bigger write: xfs_file_buffered_write: size 0x1c85ae offset 0x959d count 0x1ca3f ioflags allocates the block, and then fails with ENOSPC trying to allocate the last block on the page, leading to a failed write with stale delalloc blocks on it. Because we've had an ENOSPC when trying to allocate 0x16e00, it means that we are never goinge to call ->write_end on the page and so the allocated new buffer will not get marked dirty or have the buffer_new state cleared. In other works, what the above write is supposed to end up with is this mapping for the page: +------+------+------+------+------+------+------+------+ UMA UMA UMA UMA UMA UMA UND FAIL where: U = uptodate M = mapped N = new A = allocated D = delalloc FAIL = block we ENOSPC'd on. and the key point being the buffer_new() state for the newly allocated delayed allocation block. Except it doesn't - we're not marking buffers new correctly. That buffer_new() problem goes back to the xfs_iomap removal days, where xfs_iomap() used to return a "new" status for any map with newly allocated blocks, so that __xfs_get_blocks() could call set_buffer_new() on it. We still have the "new" variable and the check for it in the set_buffer_new() logic - except we never set it now! Hence that newly allocated delalloc block doesn't have the new flag set on it, so when the write fails we cannot tell which blocks we are supposed to punch out. WHy do we need the buffer_new flag? Well, that's because we can have this case: +------+------+------+------+------+------+------+------+ UMD UMD UMD UMD UMD UMD UND FAIL where all the UMD buffers contain valid data from a previously successful write() system call. We only want to punch the UND buffer because that's the only one that we added in this write and it was only this write that failed. That implies that even the old buffer_new() logic was wrong - because it would result in all those UMD buffers on the page having set_buffer_new() called on them even though they aren't new. Hence we shoul donly be calling set_buffer_new() for delalloc buffers that were allocated (i.e. were a hole before xfs_iomap_write_delay() was called). So, fix this set_buffer_new logic according to how we need it to work for handling failed writes correctly. Also, restore the new buffer logic handling for blocks allocated via xfs_iomap_write_direct(), because it should still set the buffer_new flag appropriately for newly allocated blocks, too. SO, now we have the buffer_new() being set appropriately in __xfs_get_blocks(), we can detect the exact delalloc ranges that we allocated in a failed write, and hence can now do a walk of the buffers on a page to find them. Except, it's not that easy. When block_write_begin() fails, it unlocks and releases the page that we just had an error on, so we can't use that page to handle errors anymore. We have to get access to the page while it is still locked to walk the buffers. Hence we have to open code block_write_begin() in xfs_vm_write_begin() to be able to insert xfs_vm_write_failed() is the right place. With that, we can pass the page and write range to xfs_vm_write_failed() and walk the buffers on the page, looking for delalloc buffers that are either new or beyond EOF and punch them out. Handling buffers beyond EOF ensures we still handle the existing case that xfs_vm_write_failed() handles. Of special note is the truncate_pagecache() handling - that only should be done for pages outside EOF - pages within EOF can still contain valid, dirty data so we must not punch them out of the cache. That just leaves the xfs_vm_write_end() failure handling. The only failure case here is that we didn't copy the entire range, and generic_write_end() handles that by zeroing the region of the page that wasn't copied, we don't have to punch out blocks within the file because they are guaranteed to contain zeros. Hence we only have to handle the existing "beyond EOF" case and don't need access to the buffers on the page. Hence it remains largely unchanged. Note that xfs_getbmap() can still trip over delalloc blocks beyond EOF that are left there by speculative delayed allocation. Hence this bug fix does not solve all known issues with bmap vs delalloc, but it does fix all the the known accidental occurances of the problem. Signed-off-by: Dave Chinner <david@fromorbit.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2012-04-27 09:45:21 +00:00
head = page_buffers(page);
block_start = 0;
for (bh = head; bh != head || !block_start;
bh = bh->b_this_page, block_start = block_end,
block_offset += bh->b_size) {
block_end = block_start + bh->b_size;
xfs: punch new delalloc blocks out of failed writes inside EOF. When a partial write inside EOF fails, it can leave delayed allocation blocks lying around because they don't get punched back out. This leads to assert failures like: XFS: Assertion failed: XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0, file: fs/xfs/xfs_super.c, line: 847 when evicting inodes from the cache. This can be trivially triggered by xfstests 083, which takes between 5 and 15 executions on a 512 byte block size filesystem to trip over this. Debugging shows a failed write due to ENOSPC calling xfs_vm_write_failed such as: [ 5012.329024] ino 0xa0026: vwf to 0x17000, sze 0x1c85ae and no action is taken on it. This leaves behind a delayed allocation extent that has no page covering it and no data in it: [ 5015.867162] ino 0xa0026: blks: 0x83 delay blocks 0x1, size 0x2538c0 [ 5015.868293] ext 0: off 0x4a, fsb 0x50306, len 0x1 [ 5015.869095] ext 1: off 0x4b, fsb 0x7899, len 0x6b [ 5015.869900] ext 2: off 0xb6, fsb 0xffffffffe0008, len 0x1 ^^^^^^^^^^^^^^^ [ 5015.871027] ext 3: off 0x36e, fsb 0x7a27, len 0xd [ 5015.872206] ext 4: off 0x4cf, fsb 0x7a1d, len 0xa So the delayed allocation extent is one block long at offset 0x16c00. Tracing shows that a bigger write: xfs_file_buffered_write: size 0x1c85ae offset 0x959d count 0x1ca3f ioflags allocates the block, and then fails with ENOSPC trying to allocate the last block on the page, leading to a failed write with stale delalloc blocks on it. Because we've had an ENOSPC when trying to allocate 0x16e00, it means that we are never goinge to call ->write_end on the page and so the allocated new buffer will not get marked dirty or have the buffer_new state cleared. In other works, what the above write is supposed to end up with is this mapping for the page: +------+------+------+------+------+------+------+------+ UMA UMA UMA UMA UMA UMA UND FAIL where: U = uptodate M = mapped N = new A = allocated D = delalloc FAIL = block we ENOSPC'd on. and the key point being the buffer_new() state for the newly allocated delayed allocation block. Except it doesn't - we're not marking buffers new correctly. That buffer_new() problem goes back to the xfs_iomap removal days, where xfs_iomap() used to return a "new" status for any map with newly allocated blocks, so that __xfs_get_blocks() could call set_buffer_new() on it. We still have the "new" variable and the check for it in the set_buffer_new() logic - except we never set it now! Hence that newly allocated delalloc block doesn't have the new flag set on it, so when the write fails we cannot tell which blocks we are supposed to punch out. WHy do we need the buffer_new flag? Well, that's because we can have this case: +------+------+------+------+------+------+------+------+ UMD UMD UMD UMD UMD UMD UND FAIL where all the UMD buffers contain valid data from a previously successful write() system call. We only want to punch the UND buffer because that's the only one that we added in this write and it was only this write that failed. That implies that even the old buffer_new() logic was wrong - because it would result in all those UMD buffers on the page having set_buffer_new() called on them even though they aren't new. Hence we shoul donly be calling set_buffer_new() for delalloc buffers that were allocated (i.e. were a hole before xfs_iomap_write_delay() was called). So, fix this set_buffer_new logic according to how we need it to work for handling failed writes correctly. Also, restore the new buffer logic handling for blocks allocated via xfs_iomap_write_direct(), because it should still set the buffer_new flag appropriately for newly allocated blocks, too. SO, now we have the buffer_new() being set appropriately in __xfs_get_blocks(), we can detect the exact delalloc ranges that we allocated in a failed write, and hence can now do a walk of the buffers on a page to find them. Except, it's not that easy. When block_write_begin() fails, it unlocks and releases the page that we just had an error on, so we can't use that page to handle errors anymore. We have to get access to the page while it is still locked to walk the buffers. Hence we have to open code block_write_begin() in xfs_vm_write_begin() to be able to insert xfs_vm_write_failed() is the right place. With that, we can pass the page and write range to xfs_vm_write_failed() and walk the buffers on the page, looking for delalloc buffers that are either new or beyond EOF and punch them out. Handling buffers beyond EOF ensures we still handle the existing case that xfs_vm_write_failed() handles. Of special note is the truncate_pagecache() handling - that only should be done for pages outside EOF - pages within EOF can still contain valid, dirty data so we must not punch them out of the cache. That just leaves the xfs_vm_write_end() failure handling. The only failure case here is that we didn't copy the entire range, and generic_write_end() handles that by zeroing the region of the page that wasn't copied, we don't have to punch out blocks within the file because they are guaranteed to contain zeros. Hence we only have to handle the existing "beyond EOF" case and don't need access to the buffers on the page. Hence it remains largely unchanged. Note that xfs_getbmap() can still trip over delalloc blocks beyond EOF that are left there by speculative delayed allocation. Hence this bug fix does not solve all known issues with bmap vs delalloc, but it does fix all the the known accidental occurances of the problem. Signed-off-by: Dave Chinner <david@fromorbit.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2012-04-27 09:45:21 +00:00
/* skip buffers before the write */
if (block_end <= from)
continue;
/* if the buffer is after the write, we're done */
if (block_start >= to)
break;
if (!buffer_delay(bh))
continue;
if (!buffer_new(bh) && block_offset < i_size_read(inode))
continue;
xfs_vm_kill_delalloc_range(inode, block_offset,
block_offset + bh->b_size);
}
xfs: punch new delalloc blocks out of failed writes inside EOF. When a partial write inside EOF fails, it can leave delayed allocation blocks lying around because they don't get punched back out. This leads to assert failures like: XFS: Assertion failed: XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0, file: fs/xfs/xfs_super.c, line: 847 when evicting inodes from the cache. This can be trivially triggered by xfstests 083, which takes between 5 and 15 executions on a 512 byte block size filesystem to trip over this. Debugging shows a failed write due to ENOSPC calling xfs_vm_write_failed such as: [ 5012.329024] ino 0xa0026: vwf to 0x17000, sze 0x1c85ae and no action is taken on it. This leaves behind a delayed allocation extent that has no page covering it and no data in it: [ 5015.867162] ino 0xa0026: blks: 0x83 delay blocks 0x1, size 0x2538c0 [ 5015.868293] ext 0: off 0x4a, fsb 0x50306, len 0x1 [ 5015.869095] ext 1: off 0x4b, fsb 0x7899, len 0x6b [ 5015.869900] ext 2: off 0xb6, fsb 0xffffffffe0008, len 0x1 ^^^^^^^^^^^^^^^ [ 5015.871027] ext 3: off 0x36e, fsb 0x7a27, len 0xd [ 5015.872206] ext 4: off 0x4cf, fsb 0x7a1d, len 0xa So the delayed allocation extent is one block long at offset 0x16c00. Tracing shows that a bigger write: xfs_file_buffered_write: size 0x1c85ae offset 0x959d count 0x1ca3f ioflags allocates the block, and then fails with ENOSPC trying to allocate the last block on the page, leading to a failed write with stale delalloc blocks on it. Because we've had an ENOSPC when trying to allocate 0x16e00, it means that we are never goinge to call ->write_end on the page and so the allocated new buffer will not get marked dirty or have the buffer_new state cleared. In other works, what the above write is supposed to end up with is this mapping for the page: +------+------+------+------+------+------+------+------+ UMA UMA UMA UMA UMA UMA UND FAIL where: U = uptodate M = mapped N = new A = allocated D = delalloc FAIL = block we ENOSPC'd on. and the key point being the buffer_new() state for the newly allocated delayed allocation block. Except it doesn't - we're not marking buffers new correctly. That buffer_new() problem goes back to the xfs_iomap removal days, where xfs_iomap() used to return a "new" status for any map with newly allocated blocks, so that __xfs_get_blocks() could call set_buffer_new() on it. We still have the "new" variable and the check for it in the set_buffer_new() logic - except we never set it now! Hence that newly allocated delalloc block doesn't have the new flag set on it, so when the write fails we cannot tell which blocks we are supposed to punch out. WHy do we need the buffer_new flag? Well, that's because we can have this case: +------+------+------+------+------+------+------+------+ UMD UMD UMD UMD UMD UMD UND FAIL where all the UMD buffers contain valid data from a previously successful write() system call. We only want to punch the UND buffer because that's the only one that we added in this write and it was only this write that failed. That implies that even the old buffer_new() logic was wrong - because it would result in all those UMD buffers on the page having set_buffer_new() called on them even though they aren't new. Hence we shoul donly be calling set_buffer_new() for delalloc buffers that were allocated (i.e. were a hole before xfs_iomap_write_delay() was called). So, fix this set_buffer_new logic according to how we need it to work for handling failed writes correctly. Also, restore the new buffer logic handling for blocks allocated via xfs_iomap_write_direct(), because it should still set the buffer_new flag appropriately for newly allocated blocks, too. SO, now we have the buffer_new() being set appropriately in __xfs_get_blocks(), we can detect the exact delalloc ranges that we allocated in a failed write, and hence can now do a walk of the buffers on a page to find them. Except, it's not that easy. When block_write_begin() fails, it unlocks and releases the page that we just had an error on, so we can't use that page to handle errors anymore. We have to get access to the page while it is still locked to walk the buffers. Hence we have to open code block_write_begin() in xfs_vm_write_begin() to be able to insert xfs_vm_write_failed() is the right place. With that, we can pass the page and write range to xfs_vm_write_failed() and walk the buffers on the page, looking for delalloc buffers that are either new or beyond EOF and punch them out. Handling buffers beyond EOF ensures we still handle the existing case that xfs_vm_write_failed() handles. Of special note is the truncate_pagecache() handling - that only should be done for pages outside EOF - pages within EOF can still contain valid, dirty data so we must not punch them out of the cache. That just leaves the xfs_vm_write_end() failure handling. The only failure case here is that we didn't copy the entire range, and generic_write_end() handles that by zeroing the region of the page that wasn't copied, we don't have to punch out blocks within the file because they are guaranteed to contain zeros. Hence we only have to handle the existing "beyond EOF" case and don't need access to the buffers on the page. Hence it remains largely unchanged. Note that xfs_getbmap() can still trip over delalloc blocks beyond EOF that are left there by speculative delayed allocation. Hence this bug fix does not solve all known issues with bmap vs delalloc, but it does fix all the the known accidental occurances of the problem. Signed-off-by: Dave Chinner <david@fromorbit.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2012-04-27 09:45:21 +00:00
}
xfs: punch new delalloc blocks out of failed writes inside EOF. When a partial write inside EOF fails, it can leave delayed allocation blocks lying around because they don't get punched back out. This leads to assert failures like: XFS: Assertion failed: XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0, file: fs/xfs/xfs_super.c, line: 847 when evicting inodes from the cache. This can be trivially triggered by xfstests 083, which takes between 5 and 15 executions on a 512 byte block size filesystem to trip over this. Debugging shows a failed write due to ENOSPC calling xfs_vm_write_failed such as: [ 5012.329024] ino 0xa0026: vwf to 0x17000, sze 0x1c85ae and no action is taken on it. This leaves behind a delayed allocation extent that has no page covering it and no data in it: [ 5015.867162] ino 0xa0026: blks: 0x83 delay blocks 0x1, size 0x2538c0 [ 5015.868293] ext 0: off 0x4a, fsb 0x50306, len 0x1 [ 5015.869095] ext 1: off 0x4b, fsb 0x7899, len 0x6b [ 5015.869900] ext 2: off 0xb6, fsb 0xffffffffe0008, len 0x1 ^^^^^^^^^^^^^^^ [ 5015.871027] ext 3: off 0x36e, fsb 0x7a27, len 0xd [ 5015.872206] ext 4: off 0x4cf, fsb 0x7a1d, len 0xa So the delayed allocation extent is one block long at offset 0x16c00. Tracing shows that a bigger write: xfs_file_buffered_write: size 0x1c85ae offset 0x959d count 0x1ca3f ioflags allocates the block, and then fails with ENOSPC trying to allocate the last block on the page, leading to a failed write with stale delalloc blocks on it. Because we've had an ENOSPC when trying to allocate 0x16e00, it means that we are never goinge to call ->write_end on the page and so the allocated new buffer will not get marked dirty or have the buffer_new state cleared. In other works, what the above write is supposed to end up with is this mapping for the page: +------+------+------+------+------+------+------+------+ UMA UMA UMA UMA UMA UMA UND FAIL where: U = uptodate M = mapped N = new A = allocated D = delalloc FAIL = block we ENOSPC'd on. and the key point being the buffer_new() state for the newly allocated delayed allocation block. Except it doesn't - we're not marking buffers new correctly. That buffer_new() problem goes back to the xfs_iomap removal days, where xfs_iomap() used to return a "new" status for any map with newly allocated blocks, so that __xfs_get_blocks() could call set_buffer_new() on it. We still have the "new" variable and the check for it in the set_buffer_new() logic - except we never set it now! Hence that newly allocated delalloc block doesn't have the new flag set on it, so when the write fails we cannot tell which blocks we are supposed to punch out. WHy do we need the buffer_new flag? Well, that's because we can have this case: +------+------+------+------+------+------+------+------+ UMD UMD UMD UMD UMD UMD UND FAIL where all the UMD buffers contain valid data from a previously successful write() system call. We only want to punch the UND buffer because that's the only one that we added in this write and it was only this write that failed. That implies that even the old buffer_new() logic was wrong - because it would result in all those UMD buffers on the page having set_buffer_new() called on them even though they aren't new. Hence we shoul donly be calling set_buffer_new() for delalloc buffers that were allocated (i.e. were a hole before xfs_iomap_write_delay() was called). So, fix this set_buffer_new logic according to how we need it to work for handling failed writes correctly. Also, restore the new buffer logic handling for blocks allocated via xfs_iomap_write_direct(), because it should still set the buffer_new flag appropriately for newly allocated blocks, too. SO, now we have the buffer_new() being set appropriately in __xfs_get_blocks(), we can detect the exact delalloc ranges that we allocated in a failed write, and hence can now do a walk of the buffers on a page to find them. Except, it's not that easy. When block_write_begin() fails, it unlocks and releases the page that we just had an error on, so we can't use that page to handle errors anymore. We have to get access to the page while it is still locked to walk the buffers. Hence we have to open code block_write_begin() in xfs_vm_write_begin() to be able to insert xfs_vm_write_failed() is the right place. With that, we can pass the page and write range to xfs_vm_write_failed() and walk the buffers on the page, looking for delalloc buffers that are either new or beyond EOF and punch them out. Handling buffers beyond EOF ensures we still handle the existing case that xfs_vm_write_failed() handles. Of special note is the truncate_pagecache() handling - that only should be done for pages outside EOF - pages within EOF can still contain valid, dirty data so we must not punch them out of the cache. That just leaves the xfs_vm_write_end() failure handling. The only failure case here is that we didn't copy the entire range, and generic_write_end() handles that by zeroing the region of the page that wasn't copied, we don't have to punch out blocks within the file because they are guaranteed to contain zeros. Hence we only have to handle the existing "beyond EOF" case and don't need access to the buffers on the page. Hence it remains largely unchanged. Note that xfs_getbmap() can still trip over delalloc blocks beyond EOF that are left there by speculative delayed allocation. Hence this bug fix does not solve all known issues with bmap vs delalloc, but it does fix all the the known accidental occurances of the problem. Signed-off-by: Dave Chinner <david@fromorbit.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2012-04-27 09:45:21 +00:00
/*
* This used to call block_write_begin(), but it unlocks and releases the page
* on error, and we need that page to be able to punch stale delalloc blocks out
* on failure. hence we copy-n-waste it here and call xfs_vm_write_failed() at
* the appropriate point.
*/
STATIC int
xfs_vm_write_begin(
struct file *file,
struct address_space *mapping,
loff_t pos,
unsigned len,
unsigned flags,
struct page **pagep,
void **fsdata)
{
xfs: punch new delalloc blocks out of failed writes inside EOF. When a partial write inside EOF fails, it can leave delayed allocation blocks lying around because they don't get punched back out. This leads to assert failures like: XFS: Assertion failed: XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0, file: fs/xfs/xfs_super.c, line: 847 when evicting inodes from the cache. This can be trivially triggered by xfstests 083, which takes between 5 and 15 executions on a 512 byte block size filesystem to trip over this. Debugging shows a failed write due to ENOSPC calling xfs_vm_write_failed such as: [ 5012.329024] ino 0xa0026: vwf to 0x17000, sze 0x1c85ae and no action is taken on it. This leaves behind a delayed allocation extent that has no page covering it and no data in it: [ 5015.867162] ino 0xa0026: blks: 0x83 delay blocks 0x1, size 0x2538c0 [ 5015.868293] ext 0: off 0x4a, fsb 0x50306, len 0x1 [ 5015.869095] ext 1: off 0x4b, fsb 0x7899, len 0x6b [ 5015.869900] ext 2: off 0xb6, fsb 0xffffffffe0008, len 0x1 ^^^^^^^^^^^^^^^ [ 5015.871027] ext 3: off 0x36e, fsb 0x7a27, len 0xd [ 5015.872206] ext 4: off 0x4cf, fsb 0x7a1d, len 0xa So the delayed allocation extent is one block long at offset 0x16c00. Tracing shows that a bigger write: xfs_file_buffered_write: size 0x1c85ae offset 0x959d count 0x1ca3f ioflags allocates the block, and then fails with ENOSPC trying to allocate the last block on the page, leading to a failed write with stale delalloc blocks on it. Because we've had an ENOSPC when trying to allocate 0x16e00, it means that we are never goinge to call ->write_end on the page and so the allocated new buffer will not get marked dirty or have the buffer_new state cleared. In other works, what the above write is supposed to end up with is this mapping for the page: +------+------+------+------+------+------+------+------+ UMA UMA UMA UMA UMA UMA UND FAIL where: U = uptodate M = mapped N = new A = allocated D = delalloc FAIL = block we ENOSPC'd on. and the key point being the buffer_new() state for the newly allocated delayed allocation block. Except it doesn't - we're not marking buffers new correctly. That buffer_new() problem goes back to the xfs_iomap removal days, where xfs_iomap() used to return a "new" status for any map with newly allocated blocks, so that __xfs_get_blocks() could call set_buffer_new() on it. We still have the "new" variable and the check for it in the set_buffer_new() logic - except we never set it now! Hence that newly allocated delalloc block doesn't have the new flag set on it, so when the write fails we cannot tell which blocks we are supposed to punch out. WHy do we need the buffer_new flag? Well, that's because we can have this case: +------+------+------+------+------+------+------+------+ UMD UMD UMD UMD UMD UMD UND FAIL where all the UMD buffers contain valid data from a previously successful write() system call. We only want to punch the UND buffer because that's the only one that we added in this write and it was only this write that failed. That implies that even the old buffer_new() logic was wrong - because it would result in all those UMD buffers on the page having set_buffer_new() called on them even though they aren't new. Hence we shoul donly be calling set_buffer_new() for delalloc buffers that were allocated (i.e. were a hole before xfs_iomap_write_delay() was called). So, fix this set_buffer_new logic according to how we need it to work for handling failed writes correctly. Also, restore the new buffer logic handling for blocks allocated via xfs_iomap_write_direct(), because it should still set the buffer_new flag appropriately for newly allocated blocks, too. SO, now we have the buffer_new() being set appropriately in __xfs_get_blocks(), we can detect the exact delalloc ranges that we allocated in a failed write, and hence can now do a walk of the buffers on a page to find them. Except, it's not that easy. When block_write_begin() fails, it unlocks and releases the page that we just had an error on, so we can't use that page to handle errors anymore. We have to get access to the page while it is still locked to walk the buffers. Hence we have to open code block_write_begin() in xfs_vm_write_begin() to be able to insert xfs_vm_write_failed() is the right place. With that, we can pass the page and write range to xfs_vm_write_failed() and walk the buffers on the page, looking for delalloc buffers that are either new or beyond EOF and punch them out. Handling buffers beyond EOF ensures we still handle the existing case that xfs_vm_write_failed() handles. Of special note is the truncate_pagecache() handling - that only should be done for pages outside EOF - pages within EOF can still contain valid, dirty data so we must not punch them out of the cache. That just leaves the xfs_vm_write_end() failure handling. The only failure case here is that we didn't copy the entire range, and generic_write_end() handles that by zeroing the region of the page that wasn't copied, we don't have to punch out blocks within the file because they are guaranteed to contain zeros. Hence we only have to handle the existing "beyond EOF" case and don't need access to the buffers on the page. Hence it remains largely unchanged. Note that xfs_getbmap() can still trip over delalloc blocks beyond EOF that are left there by speculative delayed allocation. Hence this bug fix does not solve all known issues with bmap vs delalloc, but it does fix all the the known accidental occurances of the problem. Signed-off-by: Dave Chinner <david@fromorbit.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2012-04-27 09:45:21 +00:00
pgoff_t index = pos >> PAGE_CACHE_SHIFT;
struct page *page;
int status;
xfs: punch new delalloc blocks out of failed writes inside EOF. When a partial write inside EOF fails, it can leave delayed allocation blocks lying around because they don't get punched back out. This leads to assert failures like: XFS: Assertion failed: XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0, file: fs/xfs/xfs_super.c, line: 847 when evicting inodes from the cache. This can be trivially triggered by xfstests 083, which takes between 5 and 15 executions on a 512 byte block size filesystem to trip over this. Debugging shows a failed write due to ENOSPC calling xfs_vm_write_failed such as: [ 5012.329024] ino 0xa0026: vwf to 0x17000, sze 0x1c85ae and no action is taken on it. This leaves behind a delayed allocation extent that has no page covering it and no data in it: [ 5015.867162] ino 0xa0026: blks: 0x83 delay blocks 0x1, size 0x2538c0 [ 5015.868293] ext 0: off 0x4a, fsb 0x50306, len 0x1 [ 5015.869095] ext 1: off 0x4b, fsb 0x7899, len 0x6b [ 5015.869900] ext 2: off 0xb6, fsb 0xffffffffe0008, len 0x1 ^^^^^^^^^^^^^^^ [ 5015.871027] ext 3: off 0x36e, fsb 0x7a27, len 0xd [ 5015.872206] ext 4: off 0x4cf, fsb 0x7a1d, len 0xa So the delayed allocation extent is one block long at offset 0x16c00. Tracing shows that a bigger write: xfs_file_buffered_write: size 0x1c85ae offset 0x959d count 0x1ca3f ioflags allocates the block, and then fails with ENOSPC trying to allocate the last block on the page, leading to a failed write with stale delalloc blocks on it. Because we've had an ENOSPC when trying to allocate 0x16e00, it means that we are never goinge to call ->write_end on the page and so the allocated new buffer will not get marked dirty or have the buffer_new state cleared. In other works, what the above write is supposed to end up with is this mapping for the page: +------+------+------+------+------+------+------+------+ UMA UMA UMA UMA UMA UMA UND FAIL where: U = uptodate M = mapped N = new A = allocated D = delalloc FAIL = block we ENOSPC'd on. and the key point being the buffer_new() state for the newly allocated delayed allocation block. Except it doesn't - we're not marking buffers new correctly. That buffer_new() problem goes back to the xfs_iomap removal days, where xfs_iomap() used to return a "new" status for any map with newly allocated blocks, so that __xfs_get_blocks() could call set_buffer_new() on it. We still have the "new" variable and the check for it in the set_buffer_new() logic - except we never set it now! Hence that newly allocated delalloc block doesn't have the new flag set on it, so when the write fails we cannot tell which blocks we are supposed to punch out. WHy do we need the buffer_new flag? Well, that's because we can have this case: +------+------+------+------+------+------+------+------+ UMD UMD UMD UMD UMD UMD UND FAIL where all the UMD buffers contain valid data from a previously successful write() system call. We only want to punch the UND buffer because that's the only one that we added in this write and it was only this write that failed. That implies that even the old buffer_new() logic was wrong - because it would result in all those UMD buffers on the page having set_buffer_new() called on them even though they aren't new. Hence we shoul donly be calling set_buffer_new() for delalloc buffers that were allocated (i.e. were a hole before xfs_iomap_write_delay() was called). So, fix this set_buffer_new logic according to how we need it to work for handling failed writes correctly. Also, restore the new buffer logic handling for blocks allocated via xfs_iomap_write_direct(), because it should still set the buffer_new flag appropriately for newly allocated blocks, too. SO, now we have the buffer_new() being set appropriately in __xfs_get_blocks(), we can detect the exact delalloc ranges that we allocated in a failed write, and hence can now do a walk of the buffers on a page to find them. Except, it's not that easy. When block_write_begin() fails, it unlocks and releases the page that we just had an error on, so we can't use that page to handle errors anymore. We have to get access to the page while it is still locked to walk the buffers. Hence we have to open code block_write_begin() in xfs_vm_write_begin() to be able to insert xfs_vm_write_failed() is the right place. With that, we can pass the page and write range to xfs_vm_write_failed() and walk the buffers on the page, looking for delalloc buffers that are either new or beyond EOF and punch them out. Handling buffers beyond EOF ensures we still handle the existing case that xfs_vm_write_failed() handles. Of special note is the truncate_pagecache() handling - that only should be done for pages outside EOF - pages within EOF can still contain valid, dirty data so we must not punch them out of the cache. That just leaves the xfs_vm_write_end() failure handling. The only failure case here is that we didn't copy the entire range, and generic_write_end() handles that by zeroing the region of the page that wasn't copied, we don't have to punch out blocks within the file because they are guaranteed to contain zeros. Hence we only have to handle the existing "beyond EOF" case and don't need access to the buffers on the page. Hence it remains largely unchanged. Note that xfs_getbmap() can still trip over delalloc blocks beyond EOF that are left there by speculative delayed allocation. Hence this bug fix does not solve all known issues with bmap vs delalloc, but it does fix all the the known accidental occurances of the problem. Signed-off-by: Dave Chinner <david@fromorbit.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2012-04-27 09:45:21 +00:00
ASSERT(len <= PAGE_CACHE_SIZE);
xfs: prevent stack overflows from page cache allocation Page cache allocation doesn't always go through ->begin_write and hence we don't always get the opportunity to set the allocation context to GFP_NOFS. Failing to do this means we open up the direct relcaim stack to recurse into the filesystem and consume a significant amount of stack. On RHEL6.4 kernels we are seeing ra_submit() and generic_file_splice_read() from an nfsd context recursing into the filesystem via the inode cache shrinker and evicting inodes. This is causing truncation to be run (e.g EOF block freeing) and causing bmap btree block merges and free space btree block splits to occur. These btree manipulations are occurring with the call chain already 30 functions deep and hence there is not enough stack space to complete such operations. To avoid these specific overruns, we need to prevent the page cache allocation from recursing via direct reclaim. We can do that because the allocation functions take the allocation context from that which is stored in the mapping for the inode. We don't set that right now, so the default is GFP_HIGHUSER_MOVABLE, which is effectively a GFP_KERNEL context. We need it to be the equivalent of GFP_NOFS, so when we initialise an inode, set the mapping gfp mask appropriately. This makes the use of AOP_FLAG_NOFS redundant from other parts of the XFS IO path, so get rid of it. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Ben Myers <bpm@sgi.com>
2013-10-29 11:11:57 +00:00
page = grab_cache_page_write_begin(mapping, index, flags);
xfs: punch new delalloc blocks out of failed writes inside EOF. When a partial write inside EOF fails, it can leave delayed allocation blocks lying around because they don't get punched back out. This leads to assert failures like: XFS: Assertion failed: XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0, file: fs/xfs/xfs_super.c, line: 847 when evicting inodes from the cache. This can be trivially triggered by xfstests 083, which takes between 5 and 15 executions on a 512 byte block size filesystem to trip over this. Debugging shows a failed write due to ENOSPC calling xfs_vm_write_failed such as: [ 5012.329024] ino 0xa0026: vwf to 0x17000, sze 0x1c85ae and no action is taken on it. This leaves behind a delayed allocation extent that has no page covering it and no data in it: [ 5015.867162] ino 0xa0026: blks: 0x83 delay blocks 0x1, size 0x2538c0 [ 5015.868293] ext 0: off 0x4a, fsb 0x50306, len 0x1 [ 5015.869095] ext 1: off 0x4b, fsb 0x7899, len 0x6b [ 5015.869900] ext 2: off 0xb6, fsb 0xffffffffe0008, len 0x1 ^^^^^^^^^^^^^^^ [ 5015.871027] ext 3: off 0x36e, fsb 0x7a27, len 0xd [ 5015.872206] ext 4: off 0x4cf, fsb 0x7a1d, len 0xa So the delayed allocation extent is one block long at offset 0x16c00. Tracing shows that a bigger write: xfs_file_buffered_write: size 0x1c85ae offset 0x959d count 0x1ca3f ioflags allocates the block, and then fails with ENOSPC trying to allocate the last block on the page, leading to a failed write with stale delalloc blocks on it. Because we've had an ENOSPC when trying to allocate 0x16e00, it means that we are never goinge to call ->write_end on the page and so the allocated new buffer will not get marked dirty or have the buffer_new state cleared. In other works, what the above write is supposed to end up with is this mapping for the page: +------+------+------+------+------+------+------+------+ UMA UMA UMA UMA UMA UMA UND FAIL where: U = uptodate M = mapped N = new A = allocated D = delalloc FAIL = block we ENOSPC'd on. and the key point being the buffer_new() state for the newly allocated delayed allocation block. Except it doesn't - we're not marking buffers new correctly. That buffer_new() problem goes back to the xfs_iomap removal days, where xfs_iomap() used to return a "new" status for any map with newly allocated blocks, so that __xfs_get_blocks() could call set_buffer_new() on it. We still have the "new" variable and the check for it in the set_buffer_new() logic - except we never set it now! Hence that newly allocated delalloc block doesn't have the new flag set on it, so when the write fails we cannot tell which blocks we are supposed to punch out. WHy do we need the buffer_new flag? Well, that's because we can have this case: +------+------+------+------+------+------+------+------+ UMD UMD UMD UMD UMD UMD UND FAIL where all the UMD buffers contain valid data from a previously successful write() system call. We only want to punch the UND buffer because that's the only one that we added in this write and it was only this write that failed. That implies that even the old buffer_new() logic was wrong - because it would result in all those UMD buffers on the page having set_buffer_new() called on them even though they aren't new. Hence we shoul donly be calling set_buffer_new() for delalloc buffers that were allocated (i.e. were a hole before xfs_iomap_write_delay() was called). So, fix this set_buffer_new logic according to how we need it to work for handling failed writes correctly. Also, restore the new buffer logic handling for blocks allocated via xfs_iomap_write_direct(), because it should still set the buffer_new flag appropriately for newly allocated blocks, too. SO, now we have the buffer_new() being set appropriately in __xfs_get_blocks(), we can detect the exact delalloc ranges that we allocated in a failed write, and hence can now do a walk of the buffers on a page to find them. Except, it's not that easy. When block_write_begin() fails, it unlocks and releases the page that we just had an error on, so we can't use that page to handle errors anymore. We have to get access to the page while it is still locked to walk the buffers. Hence we have to open code block_write_begin() in xfs_vm_write_begin() to be able to insert xfs_vm_write_failed() is the right place. With that, we can pass the page and write range to xfs_vm_write_failed() and walk the buffers on the page, looking for delalloc buffers that are either new or beyond EOF and punch them out. Handling buffers beyond EOF ensures we still handle the existing case that xfs_vm_write_failed() handles. Of special note is the truncate_pagecache() handling - that only should be done for pages outside EOF - pages within EOF can still contain valid, dirty data so we must not punch them out of the cache. That just leaves the xfs_vm_write_end() failure handling. The only failure case here is that we didn't copy the entire range, and generic_write_end() handles that by zeroing the region of the page that wasn't copied, we don't have to punch out blocks within the file because they are guaranteed to contain zeros. Hence we only have to handle the existing "beyond EOF" case and don't need access to the buffers on the page. Hence it remains largely unchanged. Note that xfs_getbmap() can still trip over delalloc blocks beyond EOF that are left there by speculative delayed allocation. Hence this bug fix does not solve all known issues with bmap vs delalloc, but it does fix all the the known accidental occurances of the problem. Signed-off-by: Dave Chinner <david@fromorbit.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2012-04-27 09:45:21 +00:00
if (!page)
return -ENOMEM;
status = __block_write_begin(page, pos, len, xfs_get_blocks);
if (unlikely(status)) {
struct inode *inode = mapping->host;
xfs_vm_write_failed(inode, page, pos, len);
unlock_page(page);
if (pos + len > i_size_read(inode))
truncate_pagecache(inode, i_size_read(inode));
xfs: punch new delalloc blocks out of failed writes inside EOF. When a partial write inside EOF fails, it can leave delayed allocation blocks lying around because they don't get punched back out. This leads to assert failures like: XFS: Assertion failed: XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0, file: fs/xfs/xfs_super.c, line: 847 when evicting inodes from the cache. This can be trivially triggered by xfstests 083, which takes between 5 and 15 executions on a 512 byte block size filesystem to trip over this. Debugging shows a failed write due to ENOSPC calling xfs_vm_write_failed such as: [ 5012.329024] ino 0xa0026: vwf to 0x17000, sze 0x1c85ae and no action is taken on it. This leaves behind a delayed allocation extent that has no page covering it and no data in it: [ 5015.867162] ino 0xa0026: blks: 0x83 delay blocks 0x1, size 0x2538c0 [ 5015.868293] ext 0: off 0x4a, fsb 0x50306, len 0x1 [ 5015.869095] ext 1: off 0x4b, fsb 0x7899, len 0x6b [ 5015.869900] ext 2: off 0xb6, fsb 0xffffffffe0008, len 0x1 ^^^^^^^^^^^^^^^ [ 5015.871027] ext 3: off 0x36e, fsb 0x7a27, len 0xd [ 5015.872206] ext 4: off 0x4cf, fsb 0x7a1d, len 0xa So the delayed allocation extent is one block long at offset 0x16c00. Tracing shows that a bigger write: xfs_file_buffered_write: size 0x1c85ae offset 0x959d count 0x1ca3f ioflags allocates the block, and then fails with ENOSPC trying to allocate the last block on the page, leading to a failed write with stale delalloc blocks on it. Because we've had an ENOSPC when trying to allocate 0x16e00, it means that we are never goinge to call ->write_end on the page and so the allocated new buffer will not get marked dirty or have the buffer_new state cleared. In other works, what the above write is supposed to end up with is this mapping for the page: +------+------+------+------+------+------+------+------+ UMA UMA UMA UMA UMA UMA UND FAIL where: U = uptodate M = mapped N = new A = allocated D = delalloc FAIL = block we ENOSPC'd on. and the key point being the buffer_new() state for the newly allocated delayed allocation block. Except it doesn't - we're not marking buffers new correctly. That buffer_new() problem goes back to the xfs_iomap removal days, where xfs_iomap() used to return a "new" status for any map with newly allocated blocks, so that __xfs_get_blocks() could call set_buffer_new() on it. We still have the "new" variable and the check for it in the set_buffer_new() logic - except we never set it now! Hence that newly allocated delalloc block doesn't have the new flag set on it, so when the write fails we cannot tell which blocks we are supposed to punch out. WHy do we need the buffer_new flag? Well, that's because we can have this case: +------+------+------+------+------+------+------+------+ UMD UMD UMD UMD UMD UMD UND FAIL where all the UMD buffers contain valid data from a previously successful write() system call. We only want to punch the UND buffer because that's the only one that we added in this write and it was only this write that failed. That implies that even the old buffer_new() logic was wrong - because it would result in all those UMD buffers on the page having set_buffer_new() called on them even though they aren't new. Hence we shoul donly be calling set_buffer_new() for delalloc buffers that were allocated (i.e. were a hole before xfs_iomap_write_delay() was called). So, fix this set_buffer_new logic according to how we need it to work for handling failed writes correctly. Also, restore the new buffer logic handling for blocks allocated via xfs_iomap_write_direct(), because it should still set the buffer_new flag appropriately for newly allocated blocks, too. SO, now we have the buffer_new() being set appropriately in __xfs_get_blocks(), we can detect the exact delalloc ranges that we allocated in a failed write, and hence can now do a walk of the buffers on a page to find them. Except, it's not that easy. When block_write_begin() fails, it unlocks and releases the page that we just had an error on, so we can't use that page to handle errors anymore. We have to get access to the page while it is still locked to walk the buffers. Hence we have to open code block_write_begin() in xfs_vm_write_begin() to be able to insert xfs_vm_write_failed() is the right place. With that, we can pass the page and write range to xfs_vm_write_failed() and walk the buffers on the page, looking for delalloc buffers that are either new or beyond EOF and punch them out. Handling buffers beyond EOF ensures we still handle the existing case that xfs_vm_write_failed() handles. Of special note is the truncate_pagecache() handling - that only should be done for pages outside EOF - pages within EOF can still contain valid, dirty data so we must not punch them out of the cache. That just leaves the xfs_vm_write_end() failure handling. The only failure case here is that we didn't copy the entire range, and generic_write_end() handles that by zeroing the region of the page that wasn't copied, we don't have to punch out blocks within the file because they are guaranteed to contain zeros. Hence we only have to handle the existing "beyond EOF" case and don't need access to the buffers on the page. Hence it remains largely unchanged. Note that xfs_getbmap() can still trip over delalloc blocks beyond EOF that are left there by speculative delayed allocation. Hence this bug fix does not solve all known issues with bmap vs delalloc, but it does fix all the the known accidental occurances of the problem. Signed-off-by: Dave Chinner <david@fromorbit.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2012-04-27 09:45:21 +00:00
page_cache_release(page);
page = NULL;
}
*pagep = page;
return status;
}
xfs: punch new delalloc blocks out of failed writes inside EOF. When a partial write inside EOF fails, it can leave delayed allocation blocks lying around because they don't get punched back out. This leads to assert failures like: XFS: Assertion failed: XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0, file: fs/xfs/xfs_super.c, line: 847 when evicting inodes from the cache. This can be trivially triggered by xfstests 083, which takes between 5 and 15 executions on a 512 byte block size filesystem to trip over this. Debugging shows a failed write due to ENOSPC calling xfs_vm_write_failed such as: [ 5012.329024] ino 0xa0026: vwf to 0x17000, sze 0x1c85ae and no action is taken on it. This leaves behind a delayed allocation extent that has no page covering it and no data in it: [ 5015.867162] ino 0xa0026: blks: 0x83 delay blocks 0x1, size 0x2538c0 [ 5015.868293] ext 0: off 0x4a, fsb 0x50306, len 0x1 [ 5015.869095] ext 1: off 0x4b, fsb 0x7899, len 0x6b [ 5015.869900] ext 2: off 0xb6, fsb 0xffffffffe0008, len 0x1 ^^^^^^^^^^^^^^^ [ 5015.871027] ext 3: off 0x36e, fsb 0x7a27, len 0xd [ 5015.872206] ext 4: off 0x4cf, fsb 0x7a1d, len 0xa So the delayed allocation extent is one block long at offset 0x16c00. Tracing shows that a bigger write: xfs_file_buffered_write: size 0x1c85ae offset 0x959d count 0x1ca3f ioflags allocates the block, and then fails with ENOSPC trying to allocate the last block on the page, leading to a failed write with stale delalloc blocks on it. Because we've had an ENOSPC when trying to allocate 0x16e00, it means that we are never goinge to call ->write_end on the page and so the allocated new buffer will not get marked dirty or have the buffer_new state cleared. In other works, what the above write is supposed to end up with is this mapping for the page: +------+------+------+------+------+------+------+------+ UMA UMA UMA UMA UMA UMA UND FAIL where: U = uptodate M = mapped N = new A = allocated D = delalloc FAIL = block we ENOSPC'd on. and the key point being the buffer_new() state for the newly allocated delayed allocation block. Except it doesn't - we're not marking buffers new correctly. That buffer_new() problem goes back to the xfs_iomap removal days, where xfs_iomap() used to return a "new" status for any map with newly allocated blocks, so that __xfs_get_blocks() could call set_buffer_new() on it. We still have the "new" variable and the check for it in the set_buffer_new() logic - except we never set it now! Hence that newly allocated delalloc block doesn't have the new flag set on it, so when the write fails we cannot tell which blocks we are supposed to punch out. WHy do we need the buffer_new flag? Well, that's because we can have this case: +------+------+------+------+------+------+------+------+ UMD UMD UMD UMD UMD UMD UND FAIL where all the UMD buffers contain valid data from a previously successful write() system call. We only want to punch the UND buffer because that's the only one that we added in this write and it was only this write that failed. That implies that even the old buffer_new() logic was wrong - because it would result in all those UMD buffers on the page having set_buffer_new() called on them even though they aren't new. Hence we shoul donly be calling set_buffer_new() for delalloc buffers that were allocated (i.e. were a hole before xfs_iomap_write_delay() was called). So, fix this set_buffer_new logic according to how we need it to work for handling failed writes correctly. Also, restore the new buffer logic handling for blocks allocated via xfs_iomap_write_direct(), because it should still set the buffer_new flag appropriately for newly allocated blocks, too. SO, now we have the buffer_new() being set appropriately in __xfs_get_blocks(), we can detect the exact delalloc ranges that we allocated in a failed write, and hence can now do a walk of the buffers on a page to find them. Except, it's not that easy. When block_write_begin() fails, it unlocks and releases the page that we just had an error on, so we can't use that page to handle errors anymore. We have to get access to the page while it is still locked to walk the buffers. Hence we have to open code block_write_begin() in xfs_vm_write_begin() to be able to insert xfs_vm_write_failed() is the right place. With that, we can pass the page and write range to xfs_vm_write_failed() and walk the buffers on the page, looking for delalloc buffers that are either new or beyond EOF and punch them out. Handling buffers beyond EOF ensures we still handle the existing case that xfs_vm_write_failed() handles. Of special note is the truncate_pagecache() handling - that only should be done for pages outside EOF - pages within EOF can still contain valid, dirty data so we must not punch them out of the cache. That just leaves the xfs_vm_write_end() failure handling. The only failure case here is that we didn't copy the entire range, and generic_write_end() handles that by zeroing the region of the page that wasn't copied, we don't have to punch out blocks within the file because they are guaranteed to contain zeros. Hence we only have to handle the existing "beyond EOF" case and don't need access to the buffers on the page. Hence it remains largely unchanged. Note that xfs_getbmap() can still trip over delalloc blocks beyond EOF that are left there by speculative delayed allocation. Hence this bug fix does not solve all known issues with bmap vs delalloc, but it does fix all the the known accidental occurances of the problem. Signed-off-by: Dave Chinner <david@fromorbit.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2012-04-27 09:45:21 +00:00
/*
* On failure, we only need to kill delalloc blocks beyond EOF because they
* will never be written. For blocks within EOF, generic_write_end() zeros them
* so they are safe to leave alone and be written with all the other valid data.
*/
STATIC int
xfs_vm_write_end(
struct file *file,
struct address_space *mapping,
loff_t pos,
unsigned len,
unsigned copied,
struct page *page,
void *fsdata)
{
int ret;
xfs: punch new delalloc blocks out of failed writes inside EOF. When a partial write inside EOF fails, it can leave delayed allocation blocks lying around because they don't get punched back out. This leads to assert failures like: XFS: Assertion failed: XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0, file: fs/xfs/xfs_super.c, line: 847 when evicting inodes from the cache. This can be trivially triggered by xfstests 083, which takes between 5 and 15 executions on a 512 byte block size filesystem to trip over this. Debugging shows a failed write due to ENOSPC calling xfs_vm_write_failed such as: [ 5012.329024] ino 0xa0026: vwf to 0x17000, sze 0x1c85ae and no action is taken on it. This leaves behind a delayed allocation extent that has no page covering it and no data in it: [ 5015.867162] ino 0xa0026: blks: 0x83 delay blocks 0x1, size 0x2538c0 [ 5015.868293] ext 0: off 0x4a, fsb 0x50306, len 0x1 [ 5015.869095] ext 1: off 0x4b, fsb 0x7899, len 0x6b [ 5015.869900] ext 2: off 0xb6, fsb 0xffffffffe0008, len 0x1 ^^^^^^^^^^^^^^^ [ 5015.871027] ext 3: off 0x36e, fsb 0x7a27, len 0xd [ 5015.872206] ext 4: off 0x4cf, fsb 0x7a1d, len 0xa So the delayed allocation extent is one block long at offset 0x16c00. Tracing shows that a bigger write: xfs_file_buffered_write: size 0x1c85ae offset 0x959d count 0x1ca3f ioflags allocates the block, and then fails with ENOSPC trying to allocate the last block on the page, leading to a failed write with stale delalloc blocks on it. Because we've had an ENOSPC when trying to allocate 0x16e00, it means that we are never goinge to call ->write_end on the page and so the allocated new buffer will not get marked dirty or have the buffer_new state cleared. In other works, what the above write is supposed to end up with is this mapping for the page: +------+------+------+------+------+------+------+------+ UMA UMA UMA UMA UMA UMA UND FAIL where: U = uptodate M = mapped N = new A = allocated D = delalloc FAIL = block we ENOSPC'd on. and the key point being the buffer_new() state for the newly allocated delayed allocation block. Except it doesn't - we're not marking buffers new correctly. That buffer_new() problem goes back to the xfs_iomap removal days, where xfs_iomap() used to return a "new" status for any map with newly allocated blocks, so that __xfs_get_blocks() could call set_buffer_new() on it. We still have the "new" variable and the check for it in the set_buffer_new() logic - except we never set it now! Hence that newly allocated delalloc block doesn't have the new flag set on it, so when the write fails we cannot tell which blocks we are supposed to punch out. WHy do we need the buffer_new flag? Well, that's because we can have this case: +------+------+------+------+------+------+------+------+ UMD UMD UMD UMD UMD UMD UND FAIL where all the UMD buffers contain valid data from a previously successful write() system call. We only want to punch the UND buffer because that's the only one that we added in this write and it was only this write that failed. That implies that even the old buffer_new() logic was wrong - because it would result in all those UMD buffers on the page having set_buffer_new() called on them even though they aren't new. Hence we shoul donly be calling set_buffer_new() for delalloc buffers that were allocated (i.e. were a hole before xfs_iomap_write_delay() was called). So, fix this set_buffer_new logic according to how we need it to work for handling failed writes correctly. Also, restore the new buffer logic handling for blocks allocated via xfs_iomap_write_direct(), because it should still set the buffer_new flag appropriately for newly allocated blocks, too. SO, now we have the buffer_new() being set appropriately in __xfs_get_blocks(), we can detect the exact delalloc ranges that we allocated in a failed write, and hence can now do a walk of the buffers on a page to find them. Except, it's not that easy. When block_write_begin() fails, it unlocks and releases the page that we just had an error on, so we can't use that page to handle errors anymore. We have to get access to the page while it is still locked to walk the buffers. Hence we have to open code block_write_begin() in xfs_vm_write_begin() to be able to insert xfs_vm_write_failed() is the right place. With that, we can pass the page and write range to xfs_vm_write_failed() and walk the buffers on the page, looking for delalloc buffers that are either new or beyond EOF and punch them out. Handling buffers beyond EOF ensures we still handle the existing case that xfs_vm_write_failed() handles. Of special note is the truncate_pagecache() handling - that only should be done for pages outside EOF - pages within EOF can still contain valid, dirty data so we must not punch them out of the cache. That just leaves the xfs_vm_write_end() failure handling. The only failure case here is that we didn't copy the entire range, and generic_write_end() handles that by zeroing the region of the page that wasn't copied, we don't have to punch out blocks within the file because they are guaranteed to contain zeros. Hence we only have to handle the existing "beyond EOF" case and don't need access to the buffers on the page. Hence it remains largely unchanged. Note that xfs_getbmap() can still trip over delalloc blocks beyond EOF that are left there by speculative delayed allocation. Hence this bug fix does not solve all known issues with bmap vs delalloc, but it does fix all the the known accidental occurances of the problem. Signed-off-by: Dave Chinner <david@fromorbit.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2012-04-27 09:45:21 +00:00
ASSERT(len <= PAGE_CACHE_SIZE);
ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata);
xfs: punch new delalloc blocks out of failed writes inside EOF. When a partial write inside EOF fails, it can leave delayed allocation blocks lying around because they don't get punched back out. This leads to assert failures like: XFS: Assertion failed: XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0, file: fs/xfs/xfs_super.c, line: 847 when evicting inodes from the cache. This can be trivially triggered by xfstests 083, which takes between 5 and 15 executions on a 512 byte block size filesystem to trip over this. Debugging shows a failed write due to ENOSPC calling xfs_vm_write_failed such as: [ 5012.329024] ino 0xa0026: vwf to 0x17000, sze 0x1c85ae and no action is taken on it. This leaves behind a delayed allocation extent that has no page covering it and no data in it: [ 5015.867162] ino 0xa0026: blks: 0x83 delay blocks 0x1, size 0x2538c0 [ 5015.868293] ext 0: off 0x4a, fsb 0x50306, len 0x1 [ 5015.869095] ext 1: off 0x4b, fsb 0x7899, len 0x6b [ 5015.869900] ext 2: off 0xb6, fsb 0xffffffffe0008, len 0x1 ^^^^^^^^^^^^^^^ [ 5015.871027] ext 3: off 0x36e, fsb 0x7a27, len 0xd [ 5015.872206] ext 4: off 0x4cf, fsb 0x7a1d, len 0xa So the delayed allocation extent is one block long at offset 0x16c00. Tracing shows that a bigger write: xfs_file_buffered_write: size 0x1c85ae offset 0x959d count 0x1ca3f ioflags allocates the block, and then fails with ENOSPC trying to allocate the last block on the page, leading to a failed write with stale delalloc blocks on it. Because we've had an ENOSPC when trying to allocate 0x16e00, it means that we are never goinge to call ->write_end on the page and so the allocated new buffer will not get marked dirty or have the buffer_new state cleared. In other works, what the above write is supposed to end up with is this mapping for the page: +------+------+------+------+------+------+------+------+ UMA UMA UMA UMA UMA UMA UND FAIL where: U = uptodate M = mapped N = new A = allocated D = delalloc FAIL = block we ENOSPC'd on. and the key point being the buffer_new() state for the newly allocated delayed allocation block. Except it doesn't - we're not marking buffers new correctly. That buffer_new() problem goes back to the xfs_iomap removal days, where xfs_iomap() used to return a "new" status for any map with newly allocated blocks, so that __xfs_get_blocks() could call set_buffer_new() on it. We still have the "new" variable and the check for it in the set_buffer_new() logic - except we never set it now! Hence that newly allocated delalloc block doesn't have the new flag set on it, so when the write fails we cannot tell which blocks we are supposed to punch out. WHy do we need the buffer_new flag? Well, that's because we can have this case: +------+------+------+------+------+------+------+------+ UMD UMD UMD UMD UMD UMD UND FAIL where all the UMD buffers contain valid data from a previously successful write() system call. We only want to punch the UND buffer because that's the only one that we added in this write and it was only this write that failed. That implies that even the old buffer_new() logic was wrong - because it would result in all those UMD buffers on the page having set_buffer_new() called on them even though they aren't new. Hence we shoul donly be calling set_buffer_new() for delalloc buffers that were allocated (i.e. were a hole before xfs_iomap_write_delay() was called). So, fix this set_buffer_new logic according to how we need it to work for handling failed writes correctly. Also, restore the new buffer logic handling for blocks allocated via xfs_iomap_write_direct(), because it should still set the buffer_new flag appropriately for newly allocated blocks, too. SO, now we have the buffer_new() being set appropriately in __xfs_get_blocks(), we can detect the exact delalloc ranges that we allocated in a failed write, and hence can now do a walk of the buffers on a page to find them. Except, it's not that easy. When block_write_begin() fails, it unlocks and releases the page that we just had an error on, so we can't use that page to handle errors anymore. We have to get access to the page while it is still locked to walk the buffers. Hence we have to open code block_write_begin() in xfs_vm_write_begin() to be able to insert xfs_vm_write_failed() is the right place. With that, we can pass the page and write range to xfs_vm_write_failed() and walk the buffers on the page, looking for delalloc buffers that are either new or beyond EOF and punch them out. Handling buffers beyond EOF ensures we still handle the existing case that xfs_vm_write_failed() handles. Of special note is the truncate_pagecache() handling - that only should be done for pages outside EOF - pages within EOF can still contain valid, dirty data so we must not punch them out of the cache. That just leaves the xfs_vm_write_end() failure handling. The only failure case here is that we didn't copy the entire range, and generic_write_end() handles that by zeroing the region of the page that wasn't copied, we don't have to punch out blocks within the file because they are guaranteed to contain zeros. Hence we only have to handle the existing "beyond EOF" case and don't need access to the buffers on the page. Hence it remains largely unchanged. Note that xfs_getbmap() can still trip over delalloc blocks beyond EOF that are left there by speculative delayed allocation. Hence this bug fix does not solve all known issues with bmap vs delalloc, but it does fix all the the known accidental occurances of the problem. Signed-off-by: Dave Chinner <david@fromorbit.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2012-04-27 09:45:21 +00:00
if (unlikely(ret < len)) {
struct inode *inode = mapping->host;
size_t isize = i_size_read(inode);
loff_t to = pos + len;
if (to > isize) {
truncate_pagecache(inode, isize);
xfs: punch new delalloc blocks out of failed writes inside EOF. When a partial write inside EOF fails, it can leave delayed allocation blocks lying around because they don't get punched back out. This leads to assert failures like: XFS: Assertion failed: XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0, file: fs/xfs/xfs_super.c, line: 847 when evicting inodes from the cache. This can be trivially triggered by xfstests 083, which takes between 5 and 15 executions on a 512 byte block size filesystem to trip over this. Debugging shows a failed write due to ENOSPC calling xfs_vm_write_failed such as: [ 5012.329024] ino 0xa0026: vwf to 0x17000, sze 0x1c85ae and no action is taken on it. This leaves behind a delayed allocation extent that has no page covering it and no data in it: [ 5015.867162] ino 0xa0026: blks: 0x83 delay blocks 0x1, size 0x2538c0 [ 5015.868293] ext 0: off 0x4a, fsb 0x50306, len 0x1 [ 5015.869095] ext 1: off 0x4b, fsb 0x7899, len 0x6b [ 5015.869900] ext 2: off 0xb6, fsb 0xffffffffe0008, len 0x1 ^^^^^^^^^^^^^^^ [ 5015.871027] ext 3: off 0x36e, fsb 0x7a27, len 0xd [ 5015.872206] ext 4: off 0x4cf, fsb 0x7a1d, len 0xa So the delayed allocation extent is one block long at offset 0x16c00. Tracing shows that a bigger write: xfs_file_buffered_write: size 0x1c85ae offset 0x959d count 0x1ca3f ioflags allocates the block, and then fails with ENOSPC trying to allocate the last block on the page, leading to a failed write with stale delalloc blocks on it. Because we've had an ENOSPC when trying to allocate 0x16e00, it means that we are never goinge to call ->write_end on the page and so the allocated new buffer will not get marked dirty or have the buffer_new state cleared. In other works, what the above write is supposed to end up with is this mapping for the page: +------+------+------+------+------+------+------+------+ UMA UMA UMA UMA UMA UMA UND FAIL where: U = uptodate M = mapped N = new A = allocated D = delalloc FAIL = block we ENOSPC'd on. and the key point being the buffer_new() state for the newly allocated delayed allocation block. Except it doesn't - we're not marking buffers new correctly. That buffer_new() problem goes back to the xfs_iomap removal days, where xfs_iomap() used to return a "new" status for any map with newly allocated blocks, so that __xfs_get_blocks() could call set_buffer_new() on it. We still have the "new" variable and the check for it in the set_buffer_new() logic - except we never set it now! Hence that newly allocated delalloc block doesn't have the new flag set on it, so when the write fails we cannot tell which blocks we are supposed to punch out. WHy do we need the buffer_new flag? Well, that's because we can have this case: +------+------+------+------+------+------+------+------+ UMD UMD UMD UMD UMD UMD UND FAIL where all the UMD buffers contain valid data from a previously successful write() system call. We only want to punch the UND buffer because that's the only one that we added in this write and it was only this write that failed. That implies that even the old buffer_new() logic was wrong - because it would result in all those UMD buffers on the page having set_buffer_new() called on them even though they aren't new. Hence we shoul donly be calling set_buffer_new() for delalloc buffers that were allocated (i.e. were a hole before xfs_iomap_write_delay() was called). So, fix this set_buffer_new logic according to how we need it to work for handling failed writes correctly. Also, restore the new buffer logic handling for blocks allocated via xfs_iomap_write_direct(), because it should still set the buffer_new flag appropriately for newly allocated blocks, too. SO, now we have the buffer_new() being set appropriately in __xfs_get_blocks(), we can detect the exact delalloc ranges that we allocated in a failed write, and hence can now do a walk of the buffers on a page to find them. Except, it's not that easy. When block_write_begin() fails, it unlocks and releases the page that we just had an error on, so we can't use that page to handle errors anymore. We have to get access to the page while it is still locked to walk the buffers. Hence we have to open code block_write_begin() in xfs_vm_write_begin() to be able to insert xfs_vm_write_failed() is the right place. With that, we can pass the page and write range to xfs_vm_write_failed() and walk the buffers on the page, looking for delalloc buffers that are either new or beyond EOF and punch them out. Handling buffers beyond EOF ensures we still handle the existing case that xfs_vm_write_failed() handles. Of special note is the truncate_pagecache() handling - that only should be done for pages outside EOF - pages within EOF can still contain valid, dirty data so we must not punch them out of the cache. That just leaves the xfs_vm_write_end() failure handling. The only failure case here is that we didn't copy the entire range, and generic_write_end() handles that by zeroing the region of the page that wasn't copied, we don't have to punch out blocks within the file because they are guaranteed to contain zeros. Hence we only have to handle the existing "beyond EOF" case and don't need access to the buffers on the page. Hence it remains largely unchanged. Note that xfs_getbmap() can still trip over delalloc blocks beyond EOF that are left there by speculative delayed allocation. Hence this bug fix does not solve all known issues with bmap vs delalloc, but it does fix all the the known accidental occurances of the problem. Signed-off-by: Dave Chinner <david@fromorbit.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2012-04-27 09:45:21 +00:00
xfs_vm_kill_delalloc_range(inode, isize, to);
}
}
return ret;
}
STATIC sector_t
xfs_vm_bmap(
struct address_space *mapping,
sector_t block)
{
struct inode *inode = (struct inode *)mapping->host;
struct xfs_inode *ip = XFS_I(inode);
trace_xfs_vm_bmap(XFS_I(inode));
xfs_ilock(ip, XFS_IOLOCK_SHARED);
filemap_write_and_wait(mapping);
xfs_iunlock(ip, XFS_IOLOCK_SHARED);
return generic_block_bmap(mapping, block, xfs_get_blocks);
}
STATIC int
xfs_vm_readpage(
struct file *unused,
struct page *page)
{
return mpage_readpage(page, xfs_get_blocks);
}
STATIC int
xfs_vm_readpages(
struct file *unused,
struct address_space *mapping,
struct list_head *pages,
unsigned nr_pages)
{
return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks);
}
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,
.releasepage = xfs_vm_releasepage,
.invalidatepage = xfs_vm_invalidatepage,
.write_begin = xfs_vm_write_begin,
.write_end = xfs_vm_write_end,
.bmap = xfs_vm_bmap,
.direct_IO = xfs_vm_direct_IO,
.migratepage = buffer_migrate_page,
xfs: pagecache usage optimization Hi. I introduced "is_partially_uptodate" aops for XFS. A page can have multiple buffers and even if a page is not uptodate, some buffers can be uptodate on pagesize != blocksize environment. This aops checks that all buffers which correspond to a part of a file that we want to read are uptodate. If so, we do not have to issue actual read IO to HDD even if a page is not uptodate because the portion we want to read are uptodate. "block_is_partially_uptodate" function is already used by ext2/3/4. With the following patch random read/write mixed workloads or random read after random write workloads can be optimized and we can get performance improvement. I did a performance test using the sysbench. #sysbench --num-threads=4 --max-requests=100000 --test=fileio --file-num=1 \ --file-block-size=8K --file-total-size=1G --file-test-mode=rndrw \ --file-fsync-freq=0 --file-rw-ratio=0.5 run -2.6.29-rc6 Test execution summary: total time: 123.8645s total number of events: 100000 total time taken by event execution: 442.4994 per-request statistics: min: 0.0000s avg: 0.0044s max: 0.3387s approx. 95 percentile: 0.0118s -2.6.29-rc6-patched Test execution summary: total time: 108.0757s total number of events: 100000 total time taken by event execution: 417.7505 per-request statistics: min: 0.0000s avg: 0.0042s max: 0.3217s approx. 95 percentile: 0.0118s arch: ia64 pagesize: 16k blocksize: 4k Signed-off-by: Hisashi Hifumi <hifumi.hisashi@oss.ntt.co.jp> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Felix Blyakher <felixb@sgi.com>
2009-03-29 07:53:38 +00:00
.is_partially_uptodate = block_is_partially_uptodate,
.error_remove_page = generic_error_remove_page,
};