linux_dsm_epyc7002/fs/xfs/xfs_file.c

1210 lines
31 KiB
C
Raw Normal View History

// SPDX-License-Identifier: GPL-2.0
/*
* Copyright (c) 2000-2005 Silicon Graphics, Inc.
* All Rights Reserved.
*/
#include "xfs.h"
#include "xfs_fs.h"
#include "xfs_shared.h"
#include "xfs_format.h"
#include "xfs_log_format.h"
#include "xfs_trans_resv.h"
#include "xfs_mount.h"
#include "xfs_da_format.h"
#include "xfs_da_btree.h"
#include "xfs_inode.h"
#include "xfs_trans.h"
#include "xfs_inode_item.h"
#include "xfs_bmap.h"
#include "xfs_bmap_util.h"
#include "xfs_error.h"
#include "xfs_dir2.h"
#include "xfs_dir2_priv.h"
#include "xfs_ioctl.h"
#include "xfs_trace.h"
#include "xfs_log.h"
xfs: run an eofblocks scan on ENOSPC/EDQUOT From: Brian Foster <bfoster@redhat.com> Speculative preallocation and and the associated throttling metrics assume we're working with large files on large filesystems. Users have reported inefficiencies in these mechanisms when we happen to be dealing with large files on smaller filesystems. This can occur because while prealloc throttling is aggressive under low free space conditions, it is not active until we reach 5% free space or less. For example, a 40GB filesystem has enough space for several files large enough to have multi-GB preallocations at any given time. If those files are slow growing, they might reserve preallocation for long periods of time as well as avoid the background scanner due to frequent modification. If a new file is written under these conditions, said file has no access to this already reserved space and premature ENOSPC is imminent. To handle this scenario, modify the buffered write ENOSPC handling and retry sequence to invoke an eofblocks scan. In the smaller filesystem scenario, the eofblocks scan resets the usage of preallocation such that when the 5% free space threshold is met, throttling effectively takes over to provide fair and efficient preallocation until legitimate ENOSPC. The eofblocks scan is selective based on the nature of the failure. For example, an EDQUOT failure in a particular quota will use a filtered scan for that quota. Because we don't know which quota might have caused an allocation failure at any given time, we include each applicable quota determined to be under low free space conditions in the scan. Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2014-07-24 16:49:28 +07:00
#include "xfs_icache.h"
#include "xfs_pnfs.h"
#include "xfs_iomap.h"
#include "xfs_reflink.h"
#include <linux/dcache.h>
#include <linux/falloc.h>
#include <linux/pagevec.h>
#include <linux/backing-dev.h>
#include <linux/mman.h>
static const struct vm_operations_struct xfs_file_vm_ops;
int
xfs_update_prealloc_flags(
struct xfs_inode *ip,
enum xfs_prealloc_flags flags)
{
struct xfs_trans *tp;
int error;
error = xfs_trans_alloc(ip->i_mount, &M_RES(ip->i_mount)->tr_writeid,
0, 0, 0, &tp);
if (error)
return error;
xfs_ilock(ip, XFS_ILOCK_EXCL);
xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
if (!(flags & XFS_PREALLOC_INVISIBLE)) {
VFS_I(ip)->i_mode &= ~S_ISUID;
if (VFS_I(ip)->i_mode & S_IXGRP)
VFS_I(ip)->i_mode &= ~S_ISGID;
xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
}
if (flags & XFS_PREALLOC_SET)
ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
if (flags & XFS_PREALLOC_CLEAR)
ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
if (flags & XFS_PREALLOC_SYNC)
xfs_trans_set_sync(tp);
return xfs_trans_commit(tp);
}
/*
* Fsync operations on directories are much simpler than on regular files,
* as there is no file data to flush, and thus also no need for explicit
* cache flush operations, and there are no non-transaction metadata updates
* on directories either.
*/
STATIC int
xfs_dir_fsync(
struct file *file,
loff_t start,
loff_t end,
int datasync)
{
struct xfs_inode *ip = XFS_I(file->f_mapping->host);
struct xfs_mount *mp = ip->i_mount;
xfs_lsn_t lsn = 0;
trace_xfs_dir_fsync(ip);
xfs_ilock(ip, XFS_ILOCK_SHARED);
if (xfs_ipincount(ip))
lsn = ip->i_itemp->ili_last_lsn;
xfs_iunlock(ip, XFS_ILOCK_SHARED);
if (!lsn)
return 0;
return xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, NULL);
}
STATIC int
xfs_file_fsync(
struct file *file,
loff_t start,
loff_t end,
int datasync)
{
struct inode *inode = file->f_mapping->host;
struct xfs_inode *ip = XFS_I(inode);
struct xfs_mount *mp = ip->i_mount;
int error = 0;
int log_flushed = 0;
xfs_lsn_t lsn = 0;
trace_xfs_file_fsync(ip);
error = file_write_and_wait_range(file, start, end);
if (error)
return error;
if (XFS_FORCED_SHUTDOWN(mp))
return -EIO;
xfs_iflags_clear(ip, XFS_ITRUNCATED);
/*
* If we have an RT and/or log subvolume we need to make sure to flush
* the write cache the device used for file data first. This is to
* ensure newly written file data make it to disk before logging the new
* inode size in case of an extending write.
*/
if (XFS_IS_REALTIME_INODE(ip))
xfs_blkdev_issue_flush(mp->m_rtdev_targp);
else if (mp->m_logdev_targp != mp->m_ddev_targp)
xfs_blkdev_issue_flush(mp->m_ddev_targp);
/*
xfs: optimise away log forces on timestamp updates for fdatasync xfs: timestamp updates cause excessive fdatasync log traffic Sage Weil reported that a ceph test workload was writing to the log on every fdatasync during an overwrite workload. Event tracing showed that the only metadata modification being made was the timestamp updates during the write(2) syscall, but fdatasync(2) is supposed to ignore them. The key observation was that the transactions in the log all looked like this: INODE: #regs: 4 ino: 0x8b flags: 0x45 dsize: 32 And contained a flags field of 0x45 or 0x85, and had data and attribute forks following the inode core. This means that the timestamp updates were triggering dirty relogging of previously logged parts of the inode that hadn't yet been flushed back to disk. There are two parts to this problem. The first is that XFS relogs dirty regions in subsequent transactions, so it carries around the fields that have been dirtied since the last time the inode was written back to disk, not since the last time the inode was forced into the log. The second part is that on v5 filesystems, the inode change count update during inode dirtying also sets the XFS_ILOG_CORE flag, so on v5 filesystems this makes a timestamp update dirty the entire inode. As a result when fdatasync is run, it looks at the dirty fields in the inode, and sees more than just the timestamp flag, even though the only metadata change since the last fdatasync was just the timestamps. Hence we force the log on every subsequent fdatasync even though it is not needed. To fix this, add a new field to the inode log item that tracks changes since the last time fsync/fdatasync forced the log to flush the changes to the journal. This flag is updated when we dirty the inode, but we do it before updating the change count so it does not carry the "core dirty" flag from timestamp updates. The fields are zeroed when the inode is marked clean (due to writeback/freeing) or when an fsync/datasync forces the log. Hence if we only dirty the timestamps on the inode between fsync/fdatasync calls, the fdatasync will not trigger another log force. Over 100 runs of the test program: Ext4 baseline: runtime: 1.63s +/- 0.24s avg lat: 1.59ms +/- 0.24ms iops: ~2000 XFS, vanilla kernel: runtime: 2.45s +/- 0.18s avg lat: 2.39ms +/- 0.18ms log forces: ~400/s iops: ~1000 XFS, patched kernel: runtime: 1.49s +/- 0.26s avg lat: 1.46ms +/- 0.25ms log forces: ~30/s iops: ~1500 Reported-by: Sage Weil <sage@redhat.com> Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2015-11-03 09:14:59 +07:00
* All metadata updates are logged, which means that we just have to
* flush the log up to the latest LSN that touched the inode. If we have
* concurrent fsync/fdatasync() calls, we need them to all block on the
* log force before we clear the ili_fsync_fields field. This ensures
* that we don't get a racing sync operation that does not wait for the
* metadata to hit the journal before returning. If we race with
* clearing the ili_fsync_fields, then all that will happen is the log
* force will do nothing as the lsn will already be on disk. We can't
* race with setting ili_fsync_fields because that is done under
* XFS_ILOCK_EXCL, and that can't happen because we hold the lock shared
* until after the ili_fsync_fields is cleared.
*/
xfs_ilock(ip, XFS_ILOCK_SHARED);
if (xfs_ipincount(ip)) {
if (!datasync ||
xfs: optimise away log forces on timestamp updates for fdatasync xfs: timestamp updates cause excessive fdatasync log traffic Sage Weil reported that a ceph test workload was writing to the log on every fdatasync during an overwrite workload. Event tracing showed that the only metadata modification being made was the timestamp updates during the write(2) syscall, but fdatasync(2) is supposed to ignore them. The key observation was that the transactions in the log all looked like this: INODE: #regs: 4 ino: 0x8b flags: 0x45 dsize: 32 And contained a flags field of 0x45 or 0x85, and had data and attribute forks following the inode core. This means that the timestamp updates were triggering dirty relogging of previously logged parts of the inode that hadn't yet been flushed back to disk. There are two parts to this problem. The first is that XFS relogs dirty regions in subsequent transactions, so it carries around the fields that have been dirtied since the last time the inode was written back to disk, not since the last time the inode was forced into the log. The second part is that on v5 filesystems, the inode change count update during inode dirtying also sets the XFS_ILOG_CORE flag, so on v5 filesystems this makes a timestamp update dirty the entire inode. As a result when fdatasync is run, it looks at the dirty fields in the inode, and sees more than just the timestamp flag, even though the only metadata change since the last fdatasync was just the timestamps. Hence we force the log on every subsequent fdatasync even though it is not needed. To fix this, add a new field to the inode log item that tracks changes since the last time fsync/fdatasync forced the log to flush the changes to the journal. This flag is updated when we dirty the inode, but we do it before updating the change count so it does not carry the "core dirty" flag from timestamp updates. The fields are zeroed when the inode is marked clean (due to writeback/freeing) or when an fsync/datasync forces the log. Hence if we only dirty the timestamps on the inode between fsync/fdatasync calls, the fdatasync will not trigger another log force. Over 100 runs of the test program: Ext4 baseline: runtime: 1.63s +/- 0.24s avg lat: 1.59ms +/- 0.24ms iops: ~2000 XFS, vanilla kernel: runtime: 2.45s +/- 0.18s avg lat: 2.39ms +/- 0.18ms log forces: ~400/s iops: ~1000 XFS, patched kernel: runtime: 1.49s +/- 0.26s avg lat: 1.46ms +/- 0.25ms log forces: ~30/s iops: ~1500 Reported-by: Sage Weil <sage@redhat.com> Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2015-11-03 09:14:59 +07:00
(ip->i_itemp->ili_fsync_fields & ~XFS_ILOG_TIMESTAMP))
lsn = ip->i_itemp->ili_last_lsn;
}
xfs: optimise away log forces on timestamp updates for fdatasync xfs: timestamp updates cause excessive fdatasync log traffic Sage Weil reported that a ceph test workload was writing to the log on every fdatasync during an overwrite workload. Event tracing showed that the only metadata modification being made was the timestamp updates during the write(2) syscall, but fdatasync(2) is supposed to ignore them. The key observation was that the transactions in the log all looked like this: INODE: #regs: 4 ino: 0x8b flags: 0x45 dsize: 32 And contained a flags field of 0x45 or 0x85, and had data and attribute forks following the inode core. This means that the timestamp updates were triggering dirty relogging of previously logged parts of the inode that hadn't yet been flushed back to disk. There are two parts to this problem. The first is that XFS relogs dirty regions in subsequent transactions, so it carries around the fields that have been dirtied since the last time the inode was written back to disk, not since the last time the inode was forced into the log. The second part is that on v5 filesystems, the inode change count update during inode dirtying also sets the XFS_ILOG_CORE flag, so on v5 filesystems this makes a timestamp update dirty the entire inode. As a result when fdatasync is run, it looks at the dirty fields in the inode, and sees more than just the timestamp flag, even though the only metadata change since the last fdatasync was just the timestamps. Hence we force the log on every subsequent fdatasync even though it is not needed. To fix this, add a new field to the inode log item that tracks changes since the last time fsync/fdatasync forced the log to flush the changes to the journal. This flag is updated when we dirty the inode, but we do it before updating the change count so it does not carry the "core dirty" flag from timestamp updates. The fields are zeroed when the inode is marked clean (due to writeback/freeing) or when an fsync/datasync forces the log. Hence if we only dirty the timestamps on the inode between fsync/fdatasync calls, the fdatasync will not trigger another log force. Over 100 runs of the test program: Ext4 baseline: runtime: 1.63s +/- 0.24s avg lat: 1.59ms +/- 0.24ms iops: ~2000 XFS, vanilla kernel: runtime: 2.45s +/- 0.18s avg lat: 2.39ms +/- 0.18ms log forces: ~400/s iops: ~1000 XFS, patched kernel: runtime: 1.49s +/- 0.26s avg lat: 1.46ms +/- 0.25ms log forces: ~30/s iops: ~1500 Reported-by: Sage Weil <sage@redhat.com> Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2015-11-03 09:14:59 +07:00
if (lsn) {
error = xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, &log_flushed);
xfs: optimise away log forces on timestamp updates for fdatasync xfs: timestamp updates cause excessive fdatasync log traffic Sage Weil reported that a ceph test workload was writing to the log on every fdatasync during an overwrite workload. Event tracing showed that the only metadata modification being made was the timestamp updates during the write(2) syscall, but fdatasync(2) is supposed to ignore them. The key observation was that the transactions in the log all looked like this: INODE: #regs: 4 ino: 0x8b flags: 0x45 dsize: 32 And contained a flags field of 0x45 or 0x85, and had data and attribute forks following the inode core. This means that the timestamp updates were triggering dirty relogging of previously logged parts of the inode that hadn't yet been flushed back to disk. There are two parts to this problem. The first is that XFS relogs dirty regions in subsequent transactions, so it carries around the fields that have been dirtied since the last time the inode was written back to disk, not since the last time the inode was forced into the log. The second part is that on v5 filesystems, the inode change count update during inode dirtying also sets the XFS_ILOG_CORE flag, so on v5 filesystems this makes a timestamp update dirty the entire inode. As a result when fdatasync is run, it looks at the dirty fields in the inode, and sees more than just the timestamp flag, even though the only metadata change since the last fdatasync was just the timestamps. Hence we force the log on every subsequent fdatasync even though it is not needed. To fix this, add a new field to the inode log item that tracks changes since the last time fsync/fdatasync forced the log to flush the changes to the journal. This flag is updated when we dirty the inode, but we do it before updating the change count so it does not carry the "core dirty" flag from timestamp updates. The fields are zeroed when the inode is marked clean (due to writeback/freeing) or when an fsync/datasync forces the log. Hence if we only dirty the timestamps on the inode between fsync/fdatasync calls, the fdatasync will not trigger another log force. Over 100 runs of the test program: Ext4 baseline: runtime: 1.63s +/- 0.24s avg lat: 1.59ms +/- 0.24ms iops: ~2000 XFS, vanilla kernel: runtime: 2.45s +/- 0.18s avg lat: 2.39ms +/- 0.18ms log forces: ~400/s iops: ~1000 XFS, patched kernel: runtime: 1.49s +/- 0.26s avg lat: 1.46ms +/- 0.25ms log forces: ~30/s iops: ~1500 Reported-by: Sage Weil <sage@redhat.com> Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2015-11-03 09:14:59 +07:00
ip->i_itemp->ili_fsync_fields = 0;
}
xfs_iunlock(ip, XFS_ILOCK_SHARED);
/*
* If we only have a single device, and the log force about was
* a no-op we might have to flush the data device cache here.
* This can only happen for fdatasync/O_DSYNC if we were overwriting
* an already allocated file and thus do not have any metadata to
* commit.
*/
if (!log_flushed && !XFS_IS_REALTIME_INODE(ip) &&
mp->m_logdev_targp == mp->m_ddev_targp)
xfs_blkdev_issue_flush(mp->m_ddev_targp);
return error;
}
STATIC ssize_t
xfs_file_dio_aio_read(
struct kiocb *iocb,
struct iov_iter *to)
{
struct xfs_inode *ip = XFS_I(file_inode(iocb->ki_filp));
size_t count = iov_iter_count(to);
ssize_t ret;
trace_xfs_file_direct_read(ip, count, iocb->ki_pos);
if (!count)
return 0; /* skip atime */
file_accessed(iocb->ki_filp);
xfs_ilock(ip, XFS_IOLOCK_SHARED);
ret = iomap_dio_rw(iocb, to, &xfs_iomap_ops, NULL);
xfs_iunlock(ip, XFS_IOLOCK_SHARED);
return ret;
}
static noinline ssize_t
xfs_file_dax_read(
struct kiocb *iocb,
struct iov_iter *to)
{
struct xfs_inode *ip = XFS_I(iocb->ki_filp->f_mapping->host);
size_t count = iov_iter_count(to);
ssize_t ret = 0;
trace_xfs_file_dax_read(ip, count, iocb->ki_pos);
if (!count)
return 0; /* skip atime */
if (iocb->ki_flags & IOCB_NOWAIT) {
if (!xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED))
return -EAGAIN;
} else {
xfs_ilock(ip, XFS_IOLOCK_SHARED);
}
ret = dax_iomap_rw(iocb, to, &xfs_iomap_ops);
xfs_iunlock(ip, XFS_IOLOCK_SHARED);
file_accessed(iocb->ki_filp);
return ret;
}
STATIC ssize_t
xfs_file_buffered_aio_read(
struct kiocb *iocb,
struct iov_iter *to)
{
struct xfs_inode *ip = XFS_I(file_inode(iocb->ki_filp));
ssize_t ret;
trace_xfs_file_buffered_read(ip, iov_iter_count(to), iocb->ki_pos);
if (iocb->ki_flags & IOCB_NOWAIT) {
if (!xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED))
return -EAGAIN;
} else {
xfs_ilock(ip, XFS_IOLOCK_SHARED);
}
ret = generic_file_read_iter(iocb, to);
xfs_iunlock(ip, XFS_IOLOCK_SHARED);
return ret;
}
STATIC ssize_t
xfs_file_read_iter(
struct kiocb *iocb,
struct iov_iter *to)
{
struct inode *inode = file_inode(iocb->ki_filp);
struct xfs_mount *mp = XFS_I(inode)->i_mount;
ssize_t ret = 0;
XFS_STATS_INC(mp, xs_read_calls);
if (XFS_FORCED_SHUTDOWN(mp))
return -EIO;
if (IS_DAX(inode))
ret = xfs_file_dax_read(iocb, to);
else if (iocb->ki_flags & IOCB_DIRECT)
ret = xfs_file_dio_aio_read(iocb, to);
else
ret = xfs_file_buffered_aio_read(iocb, to);
if (ret > 0)
XFS_STATS_ADD(mp, xs_read_bytes, ret);
return ret;
}
/*
* Common pre-write limit and setup checks.
*
* Called with the iolocked held either shared and exclusive according to
* @iolock, and returns with it held. Might upgrade the iolock to exclusive
* if called for a direct write beyond i_size.
*/
STATIC ssize_t
xfs_file_aio_write_checks(
struct kiocb *iocb,
struct iov_iter *from,
int *iolock)
{
struct file *file = iocb->ki_filp;
struct inode *inode = file->f_mapping->host;
struct xfs_inode *ip = XFS_I(inode);
ssize_t error = 0;
size_t count = iov_iter_count(from);
xfs: always drain dio before extending aio write submission XFS supports and typically allows concurrent asynchronous direct I/O submission to a single file. One exception to the rule is that file extending dio writes that start beyond the current EOF (e.g., potentially create a hole at EOF) require exclusive I/O access to the file. This is because such writes must zero any pre-existing blocks beyond EOF that are exposed by virtue of now residing within EOF as a result of the write about to be submitted. Before EOF zeroing can occur, the current file i_size must be stabilized to avoid data corruption. In this scenario, XFS upgrades the iolock to exclude any further I/O submission, waits on in-flight I/O to complete to ensure i_size is up to date (i_size is updated on dio write completion) and restarts the various checks against the state of the file. The problem is that this protection sequence is triggered only when the iolock is currently held shared. While this is true for async dio in most cases, the caller may upgrade the lock in advance based on arbitrary circumstances with respect to EOF zeroing. For example, the iolock is always acquired exclusively if the start offset is not block aligned. This means that even though the iolock is already held exclusive for such I/Os, pending I/O is not drained and thus EOF zeroing can occur based on an unstable i_size. This problem has been reproduced as guest data corruption in virtual machines with file-backed qcow2 virtual disks hosted on an XFS filesystem. The virtual disks must be configured with aio=native mode and the must not be truncated out to the maximum file size (as some virt managers will do). Update xfs_file_aio_write_checks() to unconditionally drain in-flight dio before EOF zeroing can occur. Rather than trigger the wait based on iolock state, use a new flag and upgrade the iolock when necessary. Note that this results in a full restart of the inode checks even when the iolock was already held exclusive when technically it is only required to recheck i_size. This should be a rare enough occurrence that it is preferable to keep the code simple rather than create an alternate restart jump target. Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Eric Sandeen <sandeen@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2015-10-12 12:02:05 +07:00
bool drained_dio = false;
loff_t isize;
xfs: don't serialise adjacent concurrent direct IO appending writes For append write workloads, extending the file requires a certain amount of exclusive locking to be done up front to ensure sanity in things like ensuring that we've zeroed any allocated regions between the old EOF and the start of the new IO. For single threads, this typically isn't a problem, and for large IOs we don't serialise enough for it to be a problem for two threads on really fast block devices. However for smaller IO and larger thread counts we have a problem. Take 4 concurrent sequential, single block sized and aligned IOs. After the first IO is submitted but before it completes, we end up with this state: IO 1 IO 2 IO 3 IO 4 +-------+-------+-------+-------+ ^ ^ | | | | | | | \- ip->i_new_size \- ip->i_size And the IO is done without exclusive locking because offset <= ip->i_size. When we submit IO 2, we see offset > ip->i_size, and grab the IO lock exclusive, because there is a chance we need to do EOF zeroing. However, there is already an IO in progress that avoids the need for IO zeroing because offset <= ip->i_new_size. hence we could avoid holding the IO lock exlcusive for this. Hence after submission of the second IO, we'd end up this state: IO 1 IO 2 IO 3 IO 4 +-------+-------+-------+-------+ ^ ^ | | | | | | | \- ip->i_new_size \- ip->i_size There is no need to grab the i_mutex of the IO lock in exclusive mode if we don't need to invalidate the page cache. Taking these locks on every direct IO effective serialises them as taking the IO lock in exclusive mode has to wait for all shared holders to drop the lock. That only happens when IO is complete, so effective it prevents dispatch of concurrent direct IO writes to the same inode. And so you can see that for the third concurrent IO, we'd avoid exclusive locking for the same reason we avoided the exclusive lock for the second IO. Fixing this is a bit more complex than that, because we need to hold a write-submission local value of ip->i_new_size to that clearing the value is only done if no other thread has updated it before our IO completes..... Signed-off-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Alex Elder <aelder@sgi.com>
2011-08-25 14:17:02 +07:00
restart:
error = generic_write_checks(iocb, from);
if (error <= 0)
return error;
xfs: prepare xfs_break_layouts() for another layout type When xfs is operating as the back-end of a pNFS block server, it prevents collisions between local and remote operations by requiring a lease to be held for remotely accessed blocks. Local filesystem operations break those leases before writing or mutating the extent map of the file. A similar mechanism is needed to prevent operations on pinned dax mappings, like device-DMA, from colliding with extent unmap operations. BREAK_WRITE and BREAK_UNMAP are introduced as two distinct levels of layout breaking. Layouts are broken in the BREAK_WRITE case to ensure that layout-holders do not collide with local writes. Additionally, layouts are broken in the BREAK_UNMAP case to make sure the layout-holder has a consistent view of the file's extent map. While BREAK_WRITE breaks can be satisfied be recalling FL_LAYOUT leases, BREAK_UNMAP breaks additionally require waiting for busy dax-pages to go idle while holding XFS_MMAPLOCK_EXCL. After this refactoring xfs_break_layouts() becomes the entry point for coordinating both types of breaks. Finally, xfs_break_leased_layouts() becomes just the BREAK_WRITE handler. Note that the unlock tracking is needed in a follow on change. That will coordinate retrying either break handler until both successfully test for a lease break while maintaining the lock state. Cc: Ross Zwisler <ross.zwisler@linux.intel.com> Cc: "Darrick J. Wong" <darrick.wong@oracle.com> Reported-by: Dave Chinner <david@fromorbit.com> Reported-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2018-03-21 04:42:38 +07:00
error = xfs_break_layouts(inode, iolock, BREAK_WRITE);
if (error)
return error;
/*
* For changing security info in file_remove_privs() we need i_rwsem
* exclusively.
*/
if (*iolock == XFS_IOLOCK_SHARED && !IS_NOSEC(inode)) {
xfs_iunlock(ip, *iolock);
*iolock = XFS_IOLOCK_EXCL;
xfs_ilock(ip, *iolock);
goto restart;
}
/*
* If the offset is beyond the size of the file, we need to zero any
* blocks that fall between the existing EOF and the start of this
* write. If zeroing is needed and we are currently holding the
* iolock shared, we need to update it to exclusive which implies
* having to redo all checks before.
xfs: DIO write completion size updates race xfs_end_io_direct_write() can race with other IO completions when updating the in-core inode size. The IO completion processing is not serialised for direct IO - they are done either under the IOLOCK_SHARED for non-AIO DIO, and without any IOLOCK held at all during AIO DIO completion. Hence the non-atomic test-and-set update of the in-core inode size is racy and can result in the in-core inode size going backwards if the race if hit just right. If the inode size goes backwards, this can trigger the EOF zeroing code to run incorrectly on the next IO, which then will zero data that has successfully been written to disk by a previous DIO. To fix this bug, we need to serialise the test/set updates of the in-core inode size. This first patch introduces locking around the relevant updates and checks in the DIO path. Because we now have an ioend in xfs_end_io_direct_write(), we know exactly then we are doing an IO that requires an in-core EOF update, and we know that they are not running in interrupt context. As such, we do not need to use irqsave() spinlock variants to protect against interrupts while the lock is held. Hence we can use an existing spinlock in the inode to do this serialisation and so not need to grow the struct xfs_inode just to work around this problem. This patch does not address the test/set EOF update in generic_file_write_direct() for various reasons - that will be done as a followup with separate explanation. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2015-04-16 19:03:07 +07:00
*
* We need to serialise against EOF updates that occur in IO
* completions here. We want to make sure that nobody is changing the
* size while we do this check until we have placed an IO barrier (i.e.
* hold the XFS_IOLOCK_EXCL) that prevents new IO from being dispatched.
* The spinlock effectively forms a memory barrier once we have the
* XFS_IOLOCK_EXCL so we are guaranteed to see the latest EOF value
* and hence be able to correctly determine if we need to run zeroing.
*/
xfs: DIO write completion size updates race xfs_end_io_direct_write() can race with other IO completions when updating the in-core inode size. The IO completion processing is not serialised for direct IO - they are done either under the IOLOCK_SHARED for non-AIO DIO, and without any IOLOCK held at all during AIO DIO completion. Hence the non-atomic test-and-set update of the in-core inode size is racy and can result in the in-core inode size going backwards if the race if hit just right. If the inode size goes backwards, this can trigger the EOF zeroing code to run incorrectly on the next IO, which then will zero data that has successfully been written to disk by a previous DIO. To fix this bug, we need to serialise the test/set updates of the in-core inode size. This first patch introduces locking around the relevant updates and checks in the DIO path. Because we now have an ioend in xfs_end_io_direct_write(), we know exactly then we are doing an IO that requires an in-core EOF update, and we know that they are not running in interrupt context. As such, we do not need to use irqsave() spinlock variants to protect against interrupts while the lock is held. Hence we can use an existing spinlock in the inode to do this serialisation and so not need to grow the struct xfs_inode just to work around this problem. This patch does not address the test/set EOF update in generic_file_write_direct() for various reasons - that will be done as a followup with separate explanation. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2015-04-16 19:03:07 +07:00
spin_lock(&ip->i_flags_lock);
isize = i_size_read(inode);
if (iocb->ki_pos > isize) {
xfs: DIO write completion size updates race xfs_end_io_direct_write() can race with other IO completions when updating the in-core inode size. The IO completion processing is not serialised for direct IO - they are done either under the IOLOCK_SHARED for non-AIO DIO, and without any IOLOCK held at all during AIO DIO completion. Hence the non-atomic test-and-set update of the in-core inode size is racy and can result in the in-core inode size going backwards if the race if hit just right. If the inode size goes backwards, this can trigger the EOF zeroing code to run incorrectly on the next IO, which then will zero data that has successfully been written to disk by a previous DIO. To fix this bug, we need to serialise the test/set updates of the in-core inode size. This first patch introduces locking around the relevant updates and checks in the DIO path. Because we now have an ioend in xfs_end_io_direct_write(), we know exactly then we are doing an IO that requires an in-core EOF update, and we know that they are not running in interrupt context. As such, we do not need to use irqsave() spinlock variants to protect against interrupts while the lock is held. Hence we can use an existing spinlock in the inode to do this serialisation and so not need to grow the struct xfs_inode just to work around this problem. This patch does not address the test/set EOF update in generic_file_write_direct() for various reasons - that will be done as a followup with separate explanation. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2015-04-16 19:03:07 +07:00
spin_unlock(&ip->i_flags_lock);
xfs: always drain dio before extending aio write submission XFS supports and typically allows concurrent asynchronous direct I/O submission to a single file. One exception to the rule is that file extending dio writes that start beyond the current EOF (e.g., potentially create a hole at EOF) require exclusive I/O access to the file. This is because such writes must zero any pre-existing blocks beyond EOF that are exposed by virtue of now residing within EOF as a result of the write about to be submitted. Before EOF zeroing can occur, the current file i_size must be stabilized to avoid data corruption. In this scenario, XFS upgrades the iolock to exclude any further I/O submission, waits on in-flight I/O to complete to ensure i_size is up to date (i_size is updated on dio write completion) and restarts the various checks against the state of the file. The problem is that this protection sequence is triggered only when the iolock is currently held shared. While this is true for async dio in most cases, the caller may upgrade the lock in advance based on arbitrary circumstances with respect to EOF zeroing. For example, the iolock is always acquired exclusively if the start offset is not block aligned. This means that even though the iolock is already held exclusive for such I/Os, pending I/O is not drained and thus EOF zeroing can occur based on an unstable i_size. This problem has been reproduced as guest data corruption in virtual machines with file-backed qcow2 virtual disks hosted on an XFS filesystem. The virtual disks must be configured with aio=native mode and the must not be truncated out to the maximum file size (as some virt managers will do). Update xfs_file_aio_write_checks() to unconditionally drain in-flight dio before EOF zeroing can occur. Rather than trigger the wait based on iolock state, use a new flag and upgrade the iolock when necessary. Note that this results in a full restart of the inode checks even when the iolock was already held exclusive when technically it is only required to recheck i_size. This should be a rare enough occurrence that it is preferable to keep the code simple rather than create an alternate restart jump target. Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Eric Sandeen <sandeen@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2015-10-12 12:02:05 +07:00
if (!drained_dio) {
if (*iolock == XFS_IOLOCK_SHARED) {
xfs_iunlock(ip, *iolock);
xfs: always drain dio before extending aio write submission XFS supports and typically allows concurrent asynchronous direct I/O submission to a single file. One exception to the rule is that file extending dio writes that start beyond the current EOF (e.g., potentially create a hole at EOF) require exclusive I/O access to the file. This is because such writes must zero any pre-existing blocks beyond EOF that are exposed by virtue of now residing within EOF as a result of the write about to be submitted. Before EOF zeroing can occur, the current file i_size must be stabilized to avoid data corruption. In this scenario, XFS upgrades the iolock to exclude any further I/O submission, waits on in-flight I/O to complete to ensure i_size is up to date (i_size is updated on dio write completion) and restarts the various checks against the state of the file. The problem is that this protection sequence is triggered only when the iolock is currently held shared. While this is true for async dio in most cases, the caller may upgrade the lock in advance based on arbitrary circumstances with respect to EOF zeroing. For example, the iolock is always acquired exclusively if the start offset is not block aligned. This means that even though the iolock is already held exclusive for such I/Os, pending I/O is not drained and thus EOF zeroing can occur based on an unstable i_size. This problem has been reproduced as guest data corruption in virtual machines with file-backed qcow2 virtual disks hosted on an XFS filesystem. The virtual disks must be configured with aio=native mode and the must not be truncated out to the maximum file size (as some virt managers will do). Update xfs_file_aio_write_checks() to unconditionally drain in-flight dio before EOF zeroing can occur. Rather than trigger the wait based on iolock state, use a new flag and upgrade the iolock when necessary. Note that this results in a full restart of the inode checks even when the iolock was already held exclusive when technically it is only required to recheck i_size. This should be a rare enough occurrence that it is preferable to keep the code simple rather than create an alternate restart jump target. Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Eric Sandeen <sandeen@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2015-10-12 12:02:05 +07:00
*iolock = XFS_IOLOCK_EXCL;
xfs_ilock(ip, *iolock);
xfs: always drain dio before extending aio write submission XFS supports and typically allows concurrent asynchronous direct I/O submission to a single file. One exception to the rule is that file extending dio writes that start beyond the current EOF (e.g., potentially create a hole at EOF) require exclusive I/O access to the file. This is because such writes must zero any pre-existing blocks beyond EOF that are exposed by virtue of now residing within EOF as a result of the write about to be submitted. Before EOF zeroing can occur, the current file i_size must be stabilized to avoid data corruption. In this scenario, XFS upgrades the iolock to exclude any further I/O submission, waits on in-flight I/O to complete to ensure i_size is up to date (i_size is updated on dio write completion) and restarts the various checks against the state of the file. The problem is that this protection sequence is triggered only when the iolock is currently held shared. While this is true for async dio in most cases, the caller may upgrade the lock in advance based on arbitrary circumstances with respect to EOF zeroing. For example, the iolock is always acquired exclusively if the start offset is not block aligned. This means that even though the iolock is already held exclusive for such I/Os, pending I/O is not drained and thus EOF zeroing can occur based on an unstable i_size. This problem has been reproduced as guest data corruption in virtual machines with file-backed qcow2 virtual disks hosted on an XFS filesystem. The virtual disks must be configured with aio=native mode and the must not be truncated out to the maximum file size (as some virt managers will do). Update xfs_file_aio_write_checks() to unconditionally drain in-flight dio before EOF zeroing can occur. Rather than trigger the wait based on iolock state, use a new flag and upgrade the iolock when necessary. Note that this results in a full restart of the inode checks even when the iolock was already held exclusive when technically it is only required to recheck i_size. This should be a rare enough occurrence that it is preferable to keep the code simple rather than create an alternate restart jump target. Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Eric Sandeen <sandeen@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2015-10-12 12:02:05 +07:00
iov_iter_reexpand(from, count);
}
/*
* We now have an IO submission barrier in place, but
* AIO can do EOF updates during IO completion and hence
* we now need to wait for all of them to drain. Non-AIO
* DIO will have drained before we are given the
* XFS_IOLOCK_EXCL, and so for most cases this wait is a
* no-op.
*/
inode_dio_wait(inode);
xfs: always drain dio before extending aio write submission XFS supports and typically allows concurrent asynchronous direct I/O submission to a single file. One exception to the rule is that file extending dio writes that start beyond the current EOF (e.g., potentially create a hole at EOF) require exclusive I/O access to the file. This is because such writes must zero any pre-existing blocks beyond EOF that are exposed by virtue of now residing within EOF as a result of the write about to be submitted. Before EOF zeroing can occur, the current file i_size must be stabilized to avoid data corruption. In this scenario, XFS upgrades the iolock to exclude any further I/O submission, waits on in-flight I/O to complete to ensure i_size is up to date (i_size is updated on dio write completion) and restarts the various checks against the state of the file. The problem is that this protection sequence is triggered only when the iolock is currently held shared. While this is true for async dio in most cases, the caller may upgrade the lock in advance based on arbitrary circumstances with respect to EOF zeroing. For example, the iolock is always acquired exclusively if the start offset is not block aligned. This means that even though the iolock is already held exclusive for such I/Os, pending I/O is not drained and thus EOF zeroing can occur based on an unstable i_size. This problem has been reproduced as guest data corruption in virtual machines with file-backed qcow2 virtual disks hosted on an XFS filesystem. The virtual disks must be configured with aio=native mode and the must not be truncated out to the maximum file size (as some virt managers will do). Update xfs_file_aio_write_checks() to unconditionally drain in-flight dio before EOF zeroing can occur. Rather than trigger the wait based on iolock state, use a new flag and upgrade the iolock when necessary. Note that this results in a full restart of the inode checks even when the iolock was already held exclusive when technically it is only required to recheck i_size. This should be a rare enough occurrence that it is preferable to keep the code simple rather than create an alternate restart jump target. Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Eric Sandeen <sandeen@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2015-10-12 12:02:05 +07:00
drained_dio = true;
xfs: don't serialise adjacent concurrent direct IO appending writes For append write workloads, extending the file requires a certain amount of exclusive locking to be done up front to ensure sanity in things like ensuring that we've zeroed any allocated regions between the old EOF and the start of the new IO. For single threads, this typically isn't a problem, and for large IOs we don't serialise enough for it to be a problem for two threads on really fast block devices. However for smaller IO and larger thread counts we have a problem. Take 4 concurrent sequential, single block sized and aligned IOs. After the first IO is submitted but before it completes, we end up with this state: IO 1 IO 2 IO 3 IO 4 +-------+-------+-------+-------+ ^ ^ | | | | | | | \- ip->i_new_size \- ip->i_size And the IO is done without exclusive locking because offset <= ip->i_size. When we submit IO 2, we see offset > ip->i_size, and grab the IO lock exclusive, because there is a chance we need to do EOF zeroing. However, there is already an IO in progress that avoids the need for IO zeroing because offset <= ip->i_new_size. hence we could avoid holding the IO lock exlcusive for this. Hence after submission of the second IO, we'd end up this state: IO 1 IO 2 IO 3 IO 4 +-------+-------+-------+-------+ ^ ^ | | | | | | | \- ip->i_new_size \- ip->i_size There is no need to grab the i_mutex of the IO lock in exclusive mode if we don't need to invalidate the page cache. Taking these locks on every direct IO effective serialises them as taking the IO lock in exclusive mode has to wait for all shared holders to drop the lock. That only happens when IO is complete, so effective it prevents dispatch of concurrent direct IO writes to the same inode. And so you can see that for the third concurrent IO, we'd avoid exclusive locking for the same reason we avoided the exclusive lock for the second IO. Fixing this is a bit more complex than that, because we need to hold a write-submission local value of ip->i_new_size to that clearing the value is only done if no other thread has updated it before our IO completes..... Signed-off-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Alex Elder <aelder@sgi.com>
2011-08-25 14:17:02 +07:00
goto restart;
}
trace_xfs_zero_eof(ip, isize, iocb->ki_pos - isize);
error = iomap_zero_range(inode, isize, iocb->ki_pos - isize,
NULL, &xfs_iomap_ops);
if (error)
return error;
xfs: DIO write completion size updates race xfs_end_io_direct_write() can race with other IO completions when updating the in-core inode size. The IO completion processing is not serialised for direct IO - they are done either under the IOLOCK_SHARED for non-AIO DIO, and without any IOLOCK held at all during AIO DIO completion. Hence the non-atomic test-and-set update of the in-core inode size is racy and can result in the in-core inode size going backwards if the race if hit just right. If the inode size goes backwards, this can trigger the EOF zeroing code to run incorrectly on the next IO, which then will zero data that has successfully been written to disk by a previous DIO. To fix this bug, we need to serialise the test/set updates of the in-core inode size. This first patch introduces locking around the relevant updates and checks in the DIO path. Because we now have an ioend in xfs_end_io_direct_write(), we know exactly then we are doing an IO that requires an in-core EOF update, and we know that they are not running in interrupt context. As such, we do not need to use irqsave() spinlock variants to protect against interrupts while the lock is held. Hence we can use an existing spinlock in the inode to do this serialisation and so not need to grow the struct xfs_inode just to work around this problem. This patch does not address the test/set EOF update in generic_file_write_direct() for various reasons - that will be done as a followup with separate explanation. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2015-04-16 19:03:07 +07:00
} else
spin_unlock(&ip->i_flags_lock);
/*
* Updating the timestamps will grab the ilock again from
* xfs_fs_dirty_inode, so we have to call it after dropping the
* lock above. Eventually we should look into a way to avoid
* the pointless lock roundtrip.
*/
if (likely(!(file->f_mode & FMODE_NOCMTIME))) {
error = file_update_time(file);
if (error)
return error;
}
/*
* If we're writing the file then make sure to clear the setuid and
* setgid bits if the process is not being run by root. This keeps
* people from modifying setuid and setgid binaries.
*/
if (!IS_NOSEC(inode))
return file_remove_privs(file);
return 0;
}
static int
xfs_dio_write_end_io(
struct kiocb *iocb,
ssize_t size,
unsigned flags)
{
struct inode *inode = file_inode(iocb->ki_filp);
struct xfs_inode *ip = XFS_I(inode);
loff_t offset = iocb->ki_pos;
int error = 0;
trace_xfs_end_io_direct_write(ip, offset, size);
if (XFS_FORCED_SHUTDOWN(ip->i_mount))
return -EIO;
if (size <= 0)
return size;
/*
* Capture amount written on completion as we can't reliably account
* for it on submission.
*/
XFS_STATS_ADD(ip->i_mount, xs_write_bytes, size);
if (flags & IOMAP_DIO_COW) {
error = xfs_reflink_end_cow(ip, offset, size);
if (error)
return error;
}
/*
* Unwritten conversion updates the in-core isize after extent
* conversion but before updating the on-disk size. Updating isize any
* earlier allows a racing dio read to find unwritten extents before
* they are converted.
*/
if (flags & IOMAP_DIO_UNWRITTEN)
return xfs_iomap_write_unwritten(ip, offset, size, true);
/*
* We need to update the in-core inode size here so that we don't end up
* with the on-disk inode size being outside the in-core inode size. We
* have no other method of updating EOF for AIO, so always do it here
* if necessary.
*
* We need to lock the test/set EOF update as we can be racing with
* other IO completions here to update the EOF. Failing to serialise
* here can result in EOF moving backwards and Bad Things Happen when
* that occurs.
*/
spin_lock(&ip->i_flags_lock);
if (offset + size > i_size_read(inode)) {
i_size_write(inode, offset + size);
spin_unlock(&ip->i_flags_lock);
error = xfs_setfilesize(ip, offset, size);
} else {
spin_unlock(&ip->i_flags_lock);
}
return error;
}
/*
* xfs_file_dio_aio_write - handle direct IO writes
*
* Lock the inode appropriately to prepare for and issue a direct IO write.
* By separating it from the buffered write path we remove all the tricky to
* follow locking changes and looping.
*
* If there are cached pages or we're extending the file, we need IOLOCK_EXCL
* until we're sure the bytes at the new EOF have been zeroed and/or the cached
* pages are flushed out.
*
* In most cases the direct IO writes will be done holding IOLOCK_SHARED
* allowing them to be done in parallel with reads and other direct IO writes.
* However, if the IO is not aligned to filesystem blocks, the direct IO layer
* needs to do sub-block zeroing and that requires serialisation against other
* direct IOs to the same block. In this case we need to serialise the
* submission of the unaligned IOs so that we don't get racing block zeroing in
* the dio layer. To avoid the problem with aio, we also need to wait for
* outstanding IOs to complete so that unwritten extent conversion is completed
* before we try to map the overlapping block. This is currently implemented by
* hitting it with a big hammer (i.e. inode_dio_wait()).
*
* Returns with locks held indicated by @iolock and errors indicated by
* negative return values.
*/
STATIC ssize_t
xfs_file_dio_aio_write(
struct kiocb *iocb,
struct iov_iter *from)
{
struct file *file = iocb->ki_filp;
struct address_space *mapping = file->f_mapping;
struct inode *inode = mapping->host;
struct xfs_inode *ip = XFS_I(inode);
struct xfs_mount *mp = ip->i_mount;
ssize_t ret = 0;
int unaligned_io = 0;
int iolock;
size_t count = iov_iter_count(from);
struct xfs_buftarg *target = XFS_IS_REALTIME_INODE(ip) ?
mp->m_rtdev_targp : mp->m_ddev_targp;
xfs: allow logical-sector sized O_DIRECT Some time ago, mkfs.xfs started picking the storage physical sector size as the default filesystem "sector size" in order to avoid RMW costs incurred by doing IOs at logical sector size alignments. However, this means that for a filesystem made with i.e. a 4k sector size on an "advanced format" 4k/512 disk, 512-byte direct IOs are no longer allowed. This means that XFS has essentially turned this AF drive into a hard 4K device, from the filesystem on up. XFS's mkfs-specified "sector size" is really just controlling the minimum size & alignment of filesystem metadata. There is no real need to tightly couple XFS's minimal metadata size to the minimum allowed direct IO size; XFS can continue doing metadata in optimal sizes, but still allow smaller DIOs for apps which issue them, for whatever reason. This patch adds a new field to the xfs_buftarg, so that we now track 2 sizes: 1) The metadata sector size, which is the minimum unit and alignment of IO which will be performed by metadata operations. 2) The device logical sector size The first is used internally by the file system for metadata alignment and IOs. The second is used for the minimum allowed direct IO alignment. This has passed xfstests on filesystems made with 4k sectors, including when run under the patch I sent to ignore XFS_IOC_DIOINFO, and issue 512 DIOs anyway. I also directly tested end of block behavior on preallocated, sparse, and existing files when we do a 512 IO into a 4k file on a 4k-sector filesystem, to be sure there were no unexpected behaviors. Signed-off-by: Eric Sandeen <sandeen@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2014-01-22 05:46:23 +07:00
/* DIO must be aligned to device logical sector size */
if ((iocb->ki_pos | count) & target->bt_logical_sectormask)
return -EINVAL;
xfs: don't serialise adjacent concurrent direct IO appending writes For append write workloads, extending the file requires a certain amount of exclusive locking to be done up front to ensure sanity in things like ensuring that we've zeroed any allocated regions between the old EOF and the start of the new IO. For single threads, this typically isn't a problem, and for large IOs we don't serialise enough for it to be a problem for two threads on really fast block devices. However for smaller IO and larger thread counts we have a problem. Take 4 concurrent sequential, single block sized and aligned IOs. After the first IO is submitted but before it completes, we end up with this state: IO 1 IO 2 IO 3 IO 4 +-------+-------+-------+-------+ ^ ^ | | | | | | | \- ip->i_new_size \- ip->i_size And the IO is done without exclusive locking because offset <= ip->i_size. When we submit IO 2, we see offset > ip->i_size, and grab the IO lock exclusive, because there is a chance we need to do EOF zeroing. However, there is already an IO in progress that avoids the need for IO zeroing because offset <= ip->i_new_size. hence we could avoid holding the IO lock exlcusive for this. Hence after submission of the second IO, we'd end up this state: IO 1 IO 2 IO 3 IO 4 +-------+-------+-------+-------+ ^ ^ | | | | | | | \- ip->i_new_size \- ip->i_size There is no need to grab the i_mutex of the IO lock in exclusive mode if we don't need to invalidate the page cache. Taking these locks on every direct IO effective serialises them as taking the IO lock in exclusive mode has to wait for all shared holders to drop the lock. That only happens when IO is complete, so effective it prevents dispatch of concurrent direct IO writes to the same inode. And so you can see that for the third concurrent IO, we'd avoid exclusive locking for the same reason we avoided the exclusive lock for the second IO. Fixing this is a bit more complex than that, because we need to hold a write-submission local value of ip->i_new_size to that clearing the value is only done if no other thread has updated it before our IO completes..... Signed-off-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Alex Elder <aelder@sgi.com>
2011-08-25 14:17:02 +07:00
/*
* Don't take the exclusive iolock here unless the I/O is unaligned to
* the file system block size. We don't need to consider the EOF
* extension case here because xfs_file_aio_write_checks() will relock
* the inode as necessary for EOF zeroing cases and fill out the new
* inode size as appropriate.
xfs: don't serialise adjacent concurrent direct IO appending writes For append write workloads, extending the file requires a certain amount of exclusive locking to be done up front to ensure sanity in things like ensuring that we've zeroed any allocated regions between the old EOF and the start of the new IO. For single threads, this typically isn't a problem, and for large IOs we don't serialise enough for it to be a problem for two threads on really fast block devices. However for smaller IO and larger thread counts we have a problem. Take 4 concurrent sequential, single block sized and aligned IOs. After the first IO is submitted but before it completes, we end up with this state: IO 1 IO 2 IO 3 IO 4 +-------+-------+-------+-------+ ^ ^ | | | | | | | \- ip->i_new_size \- ip->i_size And the IO is done without exclusive locking because offset <= ip->i_size. When we submit IO 2, we see offset > ip->i_size, and grab the IO lock exclusive, because there is a chance we need to do EOF zeroing. However, there is already an IO in progress that avoids the need for IO zeroing because offset <= ip->i_new_size. hence we could avoid holding the IO lock exlcusive for this. Hence after submission of the second IO, we'd end up this state: IO 1 IO 2 IO 3 IO 4 +-------+-------+-------+-------+ ^ ^ | | | | | | | \- ip->i_new_size \- ip->i_size There is no need to grab the i_mutex of the IO lock in exclusive mode if we don't need to invalidate the page cache. Taking these locks on every direct IO effective serialises them as taking the IO lock in exclusive mode has to wait for all shared holders to drop the lock. That only happens when IO is complete, so effective it prevents dispatch of concurrent direct IO writes to the same inode. And so you can see that for the third concurrent IO, we'd avoid exclusive locking for the same reason we avoided the exclusive lock for the second IO. Fixing this is a bit more complex than that, because we need to hold a write-submission local value of ip->i_new_size to that clearing the value is only done if no other thread has updated it before our IO completes..... Signed-off-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Alex Elder <aelder@sgi.com>
2011-08-25 14:17:02 +07:00
*/
if ((iocb->ki_pos & mp->m_blockmask) ||
((iocb->ki_pos + count) & mp->m_blockmask)) {
unaligned_io = 1;
/*
* We can't properly handle unaligned direct I/O to reflink
* files yet, as we can't unshare a partial block.
*/
if (xfs_is_reflink_inode(ip)) {
trace_xfs_reflink_bounce_dio_write(ip, iocb->ki_pos, count);
return -EREMCHG;
}
iolock = XFS_IOLOCK_EXCL;
} else {
iolock = XFS_IOLOCK_SHARED;
}
if (iocb->ki_flags & IOCB_NOWAIT) {
if (!xfs_ilock_nowait(ip, iolock))
return -EAGAIN;
} else {
xfs_ilock(ip, iolock);
}
ret = xfs_file_aio_write_checks(iocb, from, &iolock);
if (ret)
goto out;
count = iov_iter_count(from);
/*
* If we are doing unaligned IO, wait for all other IO to drain,
* otherwise demote the lock if we had to take the exclusive lock
* for other reasons in xfs_file_aio_write_checks.
*/
if (unaligned_io) {
/* If we are going to wait for other DIO to finish, bail */
if (iocb->ki_flags & IOCB_NOWAIT) {
if (atomic_read(&inode->i_dio_count))
return -EAGAIN;
} else {
inode_dio_wait(inode);
}
} else if (iolock == XFS_IOLOCK_EXCL) {
xfs_ilock_demote(ip, XFS_IOLOCK_EXCL);
iolock = XFS_IOLOCK_SHARED;
}
trace_xfs_file_direct_write(ip, count, iocb->ki_pos);
ret = iomap_dio_rw(iocb, from, &xfs_iomap_ops, xfs_dio_write_end_io);
out:
xfs_iunlock(ip, iolock);
/*
* No fallback to buffered IO on errors for XFS, direct IO will either
* complete fully or fail.
*/
ASSERT(ret < 0 || ret == count);
return ret;
}
static noinline ssize_t
xfs_file_dax_write(
struct kiocb *iocb,
struct iov_iter *from)
{
struct inode *inode = iocb->ki_filp->f_mapping->host;
struct xfs_inode *ip = XFS_I(inode);
int iolock = XFS_IOLOCK_EXCL;
ssize_t ret, error = 0;
size_t count;
loff_t pos;
if (iocb->ki_flags & IOCB_NOWAIT) {
if (!xfs_ilock_nowait(ip, iolock))
return -EAGAIN;
} else {
xfs_ilock(ip, iolock);
}
ret = xfs_file_aio_write_checks(iocb, from, &iolock);
if (ret)
goto out;
pos = iocb->ki_pos;
count = iov_iter_count(from);
trace_xfs_file_dax_write(ip, count, pos);
ret = dax_iomap_rw(iocb, from, &xfs_iomap_ops);
if (ret > 0 && iocb->ki_pos > i_size_read(inode)) {
i_size_write(inode, iocb->ki_pos);
error = xfs_setfilesize(ip, pos, ret);
}
out:
xfs_iunlock(ip, iolock);
if (error)
return error;
if (ret > 0) {
XFS_STATS_ADD(ip->i_mount, xs_write_bytes, ret);
/* Handle various SYNC-type writes */
ret = generic_write_sync(iocb, ret);
}
return ret;
}
STATIC ssize_t
xfs_file_buffered_aio_write(
struct kiocb *iocb,
struct iov_iter *from)
{
struct file *file = iocb->ki_filp;
struct address_space *mapping = file->f_mapping;
struct inode *inode = mapping->host;
struct xfs_inode *ip = XFS_I(inode);
ssize_t ret;
int enospc = 0;
xfs: sync eofblocks scans under iolock are livelock prone The xfs_eofblocks.eof_scan_owner field is an internal field to facilitate invoking eofb scans from the kernel while under the iolock. This is necessary because the eofb scan acquires the iolock of each inode. Synchronous scans are invoked on certain buffered write failures while under iolock. In such cases, the scan owner indicates that the context for the scan already owns the particular iolock and prevents a double lock deadlock. eofblocks scans while under iolock are still livelock prone in the event of multiple parallel scans, however. If multiple buffered writes to different inodes fail and invoke eofblocks scans at the same time, each scan avoids a deadlock with its own inode by virtue of the eof_scan_owner field, but will never be able to acquire the iolock of the inode from the parallel scan. Because the low free space scans are invoked with SYNC_WAIT, the scan will not return until it has processed every tagged inode and thus both scans will spin indefinitely on the iolock being held across the opposite scan. This problem can be reproduced reliably by generic/224 on systems with higher cpu counts (x16). To avoid this problem, simplify the semantics of eofblocks scans to never invoke a scan while under iolock. This means that the buffered write context must drop the iolock before the scan. It must reacquire the lock before the write retry and also repeat the initial write checks, as the original state might no longer be valid once the iolock was dropped. Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-01-28 14:22:56 +07:00
int iolock;
if (iocb->ki_flags & IOCB_NOWAIT)
return -EOPNOTSUPP;
xfs: sync eofblocks scans under iolock are livelock prone The xfs_eofblocks.eof_scan_owner field is an internal field to facilitate invoking eofb scans from the kernel while under the iolock. This is necessary because the eofb scan acquires the iolock of each inode. Synchronous scans are invoked on certain buffered write failures while under iolock. In such cases, the scan owner indicates that the context for the scan already owns the particular iolock and prevents a double lock deadlock. eofblocks scans while under iolock are still livelock prone in the event of multiple parallel scans, however. If multiple buffered writes to different inodes fail and invoke eofblocks scans at the same time, each scan avoids a deadlock with its own inode by virtue of the eof_scan_owner field, but will never be able to acquire the iolock of the inode from the parallel scan. Because the low free space scans are invoked with SYNC_WAIT, the scan will not return until it has processed every tagged inode and thus both scans will spin indefinitely on the iolock being held across the opposite scan. This problem can be reproduced reliably by generic/224 on systems with higher cpu counts (x16). To avoid this problem, simplify the semantics of eofblocks scans to never invoke a scan while under iolock. This means that the buffered write context must drop the iolock before the scan. It must reacquire the lock before the write retry and also repeat the initial write checks, as the original state might no longer be valid once the iolock was dropped. Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-01-28 14:22:56 +07:00
write_retry:
iolock = XFS_IOLOCK_EXCL;
xfs_ilock(ip, iolock);
ret = xfs_file_aio_write_checks(iocb, from, &iolock);
if (ret)
goto out;
/* We can write back this queue in page reclaim */
current->backing_dev_info = inode_to_bdi(inode);
trace_xfs_file_buffered_write(ip, iov_iter_count(from), iocb->ki_pos);
ret = iomap_file_buffered_write(iocb, from, &xfs_iomap_ops);
if (likely(ret >= 0))
iocb->ki_pos += ret;
xfs: run an eofblocks scan on ENOSPC/EDQUOT From: Brian Foster <bfoster@redhat.com> Speculative preallocation and and the associated throttling metrics assume we're working with large files on large filesystems. Users have reported inefficiencies in these mechanisms when we happen to be dealing with large files on smaller filesystems. This can occur because while prealloc throttling is aggressive under low free space conditions, it is not active until we reach 5% free space or less. For example, a 40GB filesystem has enough space for several files large enough to have multi-GB preallocations at any given time. If those files are slow growing, they might reserve preallocation for long periods of time as well as avoid the background scanner due to frequent modification. If a new file is written under these conditions, said file has no access to this already reserved space and premature ENOSPC is imminent. To handle this scenario, modify the buffered write ENOSPC handling and retry sequence to invoke an eofblocks scan. In the smaller filesystem scenario, the eofblocks scan resets the usage of preallocation such that when the 5% free space threshold is met, throttling effectively takes over to provide fair and efficient preallocation until legitimate ENOSPC. The eofblocks scan is selective based on the nature of the failure. For example, an EDQUOT failure in a particular quota will use a filtered scan for that quota. Because we don't know which quota might have caused an allocation failure at any given time, we include each applicable quota determined to be under low free space conditions in the scan. Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2014-07-24 16:49:28 +07:00
/*
xfs: run an eofblocks scan on ENOSPC/EDQUOT From: Brian Foster <bfoster@redhat.com> Speculative preallocation and and the associated throttling metrics assume we're working with large files on large filesystems. Users have reported inefficiencies in these mechanisms when we happen to be dealing with large files on smaller filesystems. This can occur because while prealloc throttling is aggressive under low free space conditions, it is not active until we reach 5% free space or less. For example, a 40GB filesystem has enough space for several files large enough to have multi-GB preallocations at any given time. If those files are slow growing, they might reserve preallocation for long periods of time as well as avoid the background scanner due to frequent modification. If a new file is written under these conditions, said file has no access to this already reserved space and premature ENOSPC is imminent. To handle this scenario, modify the buffered write ENOSPC handling and retry sequence to invoke an eofblocks scan. In the smaller filesystem scenario, the eofblocks scan resets the usage of preallocation such that when the 5% free space threshold is met, throttling effectively takes over to provide fair and efficient preallocation until legitimate ENOSPC. The eofblocks scan is selective based on the nature of the failure. For example, an EDQUOT failure in a particular quota will use a filtered scan for that quota. Because we don't know which quota might have caused an allocation failure at any given time, we include each applicable quota determined to be under low free space conditions in the scan. Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2014-07-24 16:49:28 +07:00
* If we hit a space limit, try to free up some lingering preallocated
* space before returning an error. In the case of ENOSPC, first try to
* write back all dirty inodes to free up some of the excess reserved
* metadata space. This reduces the chances that the eofblocks scan
* waits on dirty mappings. Since xfs_flush_inodes() is serialized, this
* also behaves as a filter to prevent too many eofblocks scans from
* running at the same time.
*/
xfs: run an eofblocks scan on ENOSPC/EDQUOT From: Brian Foster <bfoster@redhat.com> Speculative preallocation and and the associated throttling metrics assume we're working with large files on large filesystems. Users have reported inefficiencies in these mechanisms when we happen to be dealing with large files on smaller filesystems. This can occur because while prealloc throttling is aggressive under low free space conditions, it is not active until we reach 5% free space or less. For example, a 40GB filesystem has enough space for several files large enough to have multi-GB preallocations at any given time. If those files are slow growing, they might reserve preallocation for long periods of time as well as avoid the background scanner due to frequent modification. If a new file is written under these conditions, said file has no access to this already reserved space and premature ENOSPC is imminent. To handle this scenario, modify the buffered write ENOSPC handling and retry sequence to invoke an eofblocks scan. In the smaller filesystem scenario, the eofblocks scan resets the usage of preallocation such that when the 5% free space threshold is met, throttling effectively takes over to provide fair and efficient preallocation until legitimate ENOSPC. The eofblocks scan is selective based on the nature of the failure. For example, an EDQUOT failure in a particular quota will use a filtered scan for that quota. Because we don't know which quota might have caused an allocation failure at any given time, we include each applicable quota determined to be under low free space conditions in the scan. Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2014-07-24 16:49:28 +07:00
if (ret == -EDQUOT && !enospc) {
xfs: sync eofblocks scans under iolock are livelock prone The xfs_eofblocks.eof_scan_owner field is an internal field to facilitate invoking eofb scans from the kernel while under the iolock. This is necessary because the eofb scan acquires the iolock of each inode. Synchronous scans are invoked on certain buffered write failures while under iolock. In such cases, the scan owner indicates that the context for the scan already owns the particular iolock and prevents a double lock deadlock. eofblocks scans while under iolock are still livelock prone in the event of multiple parallel scans, however. If multiple buffered writes to different inodes fail and invoke eofblocks scans at the same time, each scan avoids a deadlock with its own inode by virtue of the eof_scan_owner field, but will never be able to acquire the iolock of the inode from the parallel scan. Because the low free space scans are invoked with SYNC_WAIT, the scan will not return until it has processed every tagged inode and thus both scans will spin indefinitely on the iolock being held across the opposite scan. This problem can be reproduced reliably by generic/224 on systems with higher cpu counts (x16). To avoid this problem, simplify the semantics of eofblocks scans to never invoke a scan while under iolock. This means that the buffered write context must drop the iolock before the scan. It must reacquire the lock before the write retry and also repeat the initial write checks, as the original state might no longer be valid once the iolock was dropped. Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-01-28 14:22:56 +07:00
xfs_iunlock(ip, iolock);
xfs: run an eofblocks scan on ENOSPC/EDQUOT From: Brian Foster <bfoster@redhat.com> Speculative preallocation and and the associated throttling metrics assume we're working with large files on large filesystems. Users have reported inefficiencies in these mechanisms when we happen to be dealing with large files on smaller filesystems. This can occur because while prealloc throttling is aggressive under low free space conditions, it is not active until we reach 5% free space or less. For example, a 40GB filesystem has enough space for several files large enough to have multi-GB preallocations at any given time. If those files are slow growing, they might reserve preallocation for long periods of time as well as avoid the background scanner due to frequent modification. If a new file is written under these conditions, said file has no access to this already reserved space and premature ENOSPC is imminent. To handle this scenario, modify the buffered write ENOSPC handling and retry sequence to invoke an eofblocks scan. In the smaller filesystem scenario, the eofblocks scan resets the usage of preallocation such that when the 5% free space threshold is met, throttling effectively takes over to provide fair and efficient preallocation until legitimate ENOSPC. The eofblocks scan is selective based on the nature of the failure. For example, an EDQUOT failure in a particular quota will use a filtered scan for that quota. Because we don't know which quota might have caused an allocation failure at any given time, we include each applicable quota determined to be under low free space conditions in the scan. Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2014-07-24 16:49:28 +07:00
enospc = xfs_inode_free_quota_eofblocks(ip);
if (enospc)
goto write_retry;
enospc = xfs_inode_free_quota_cowblocks(ip);
if (enospc)
goto write_retry;
xfs: sync eofblocks scans under iolock are livelock prone The xfs_eofblocks.eof_scan_owner field is an internal field to facilitate invoking eofb scans from the kernel while under the iolock. This is necessary because the eofb scan acquires the iolock of each inode. Synchronous scans are invoked on certain buffered write failures while under iolock. In such cases, the scan owner indicates that the context for the scan already owns the particular iolock and prevents a double lock deadlock. eofblocks scans while under iolock are still livelock prone in the event of multiple parallel scans, however. If multiple buffered writes to different inodes fail and invoke eofblocks scans at the same time, each scan avoids a deadlock with its own inode by virtue of the eof_scan_owner field, but will never be able to acquire the iolock of the inode from the parallel scan. Because the low free space scans are invoked with SYNC_WAIT, the scan will not return until it has processed every tagged inode and thus both scans will spin indefinitely on the iolock being held across the opposite scan. This problem can be reproduced reliably by generic/224 on systems with higher cpu counts (x16). To avoid this problem, simplify the semantics of eofblocks scans to never invoke a scan while under iolock. This means that the buffered write context must drop the iolock before the scan. It must reacquire the lock before the write retry and also repeat the initial write checks, as the original state might no longer be valid once the iolock was dropped. Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-01-28 14:22:56 +07:00
iolock = 0;
xfs: run an eofblocks scan on ENOSPC/EDQUOT From: Brian Foster <bfoster@redhat.com> Speculative preallocation and and the associated throttling metrics assume we're working with large files on large filesystems. Users have reported inefficiencies in these mechanisms when we happen to be dealing with large files on smaller filesystems. This can occur because while prealloc throttling is aggressive under low free space conditions, it is not active until we reach 5% free space or less. For example, a 40GB filesystem has enough space for several files large enough to have multi-GB preallocations at any given time. If those files are slow growing, they might reserve preallocation for long periods of time as well as avoid the background scanner due to frequent modification. If a new file is written under these conditions, said file has no access to this already reserved space and premature ENOSPC is imminent. To handle this scenario, modify the buffered write ENOSPC handling and retry sequence to invoke an eofblocks scan. In the smaller filesystem scenario, the eofblocks scan resets the usage of preallocation such that when the 5% free space threshold is met, throttling effectively takes over to provide fair and efficient preallocation until legitimate ENOSPC. The eofblocks scan is selective based on the nature of the failure. For example, an EDQUOT failure in a particular quota will use a filtered scan for that quota. Because we don't know which quota might have caused an allocation failure at any given time, we include each applicable quota determined to be under low free space conditions in the scan. Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2014-07-24 16:49:28 +07:00
} else if (ret == -ENOSPC && !enospc) {
struct xfs_eofblocks eofb = {0};
enospc = 1;
xfs: xfs_sync_data is redundant. We don't do any data writeback from XFS any more - the VFS is completely responsible for that, including for freeze. We can replace the remaining caller with a VFS level function that achieves the same thing, but without conflicting with current writeback work. This means we can remove the flush_work and xfs_flush_inodes() - the VFS functionality completely replaces the internal flush queue for doing this writeback work in a separate context to avoid stack overruns. This does have one complication - it cannot be called with page locks held. Hence move the flushing of delalloc space when ENOSPC occurs back up into xfs_file_aio_buffered_write when we don't hold any locks that will stall writeback. Unfortunately, writeback_inodes_sb_if_idle() is not sufficient to trigger delalloc conversion fast enough to prevent spurious ENOSPC whent here are hundreds of writers, thousands of small files and GBs of free RAM. Hence we need to use sync_sb_inodes() to block callers while we wait for writeback like the previous xfs_flush_inodes implementation did. That means we have to hold the s_umount lock here, but because this call can nest inside i_mutex (the parent directory in the create case, held by the VFS), we have to use down_read_trylock() to avoid potential deadlocks. In practice, this trylock will succeed on almost every attempt as unmount/remount type operations are exceedingly rare. Note: we always need to pass a count of zero to generic_file_buffered_write() as the previously written byte count. We only do this by accident before this patch by the virtue of ret always being zero when there are no errors. Make this explicit rather than needing to specifically zero ret in the ENOSPC retry case. Signed-off-by: Dave Chinner <dchinner@redhat.com> Tested-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Ben Myers <bpm@sgi.com>
2012-10-08 17:56:04 +07:00
xfs_flush_inodes(ip->i_mount);
xfs: sync eofblocks scans under iolock are livelock prone The xfs_eofblocks.eof_scan_owner field is an internal field to facilitate invoking eofb scans from the kernel while under the iolock. This is necessary because the eofb scan acquires the iolock of each inode. Synchronous scans are invoked on certain buffered write failures while under iolock. In such cases, the scan owner indicates that the context for the scan already owns the particular iolock and prevents a double lock deadlock. eofblocks scans while under iolock are still livelock prone in the event of multiple parallel scans, however. If multiple buffered writes to different inodes fail and invoke eofblocks scans at the same time, each scan avoids a deadlock with its own inode by virtue of the eof_scan_owner field, but will never be able to acquire the iolock of the inode from the parallel scan. Because the low free space scans are invoked with SYNC_WAIT, the scan will not return until it has processed every tagged inode and thus both scans will spin indefinitely on the iolock being held across the opposite scan. This problem can be reproduced reliably by generic/224 on systems with higher cpu counts (x16). To avoid this problem, simplify the semantics of eofblocks scans to never invoke a scan while under iolock. This means that the buffered write context must drop the iolock before the scan. It must reacquire the lock before the write retry and also repeat the initial write checks, as the original state might no longer be valid once the iolock was dropped. Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-01-28 14:22:56 +07:00
xfs_iunlock(ip, iolock);
xfs: run an eofblocks scan on ENOSPC/EDQUOT From: Brian Foster <bfoster@redhat.com> Speculative preallocation and and the associated throttling metrics assume we're working with large files on large filesystems. Users have reported inefficiencies in these mechanisms when we happen to be dealing with large files on smaller filesystems. This can occur because while prealloc throttling is aggressive under low free space conditions, it is not active until we reach 5% free space or less. For example, a 40GB filesystem has enough space for several files large enough to have multi-GB preallocations at any given time. If those files are slow growing, they might reserve preallocation for long periods of time as well as avoid the background scanner due to frequent modification. If a new file is written under these conditions, said file has no access to this already reserved space and premature ENOSPC is imminent. To handle this scenario, modify the buffered write ENOSPC handling and retry sequence to invoke an eofblocks scan. In the smaller filesystem scenario, the eofblocks scan resets the usage of preallocation such that when the 5% free space threshold is met, throttling effectively takes over to provide fair and efficient preallocation until legitimate ENOSPC. The eofblocks scan is selective based on the nature of the failure. For example, an EDQUOT failure in a particular quota will use a filtered scan for that quota. Because we don't know which quota might have caused an allocation failure at any given time, we include each applicable quota determined to be under low free space conditions in the scan. Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2014-07-24 16:49:28 +07:00
eofb.eof_flags = XFS_EOF_FLAGS_SYNC;
xfs_icache_free_eofblocks(ip->i_mount, &eofb);
xfs_icache_free_cowblocks(ip->i_mount, &eofb);
xfs: xfs_sync_data is redundant. We don't do any data writeback from XFS any more - the VFS is completely responsible for that, including for freeze. We can replace the remaining caller with a VFS level function that achieves the same thing, but without conflicting with current writeback work. This means we can remove the flush_work and xfs_flush_inodes() - the VFS functionality completely replaces the internal flush queue for doing this writeback work in a separate context to avoid stack overruns. This does have one complication - it cannot be called with page locks held. Hence move the flushing of delalloc space when ENOSPC occurs back up into xfs_file_aio_buffered_write when we don't hold any locks that will stall writeback. Unfortunately, writeback_inodes_sb_if_idle() is not sufficient to trigger delalloc conversion fast enough to prevent spurious ENOSPC whent here are hundreds of writers, thousands of small files and GBs of free RAM. Hence we need to use sync_sb_inodes() to block callers while we wait for writeback like the previous xfs_flush_inodes implementation did. That means we have to hold the s_umount lock here, but because this call can nest inside i_mutex (the parent directory in the create case, held by the VFS), we have to use down_read_trylock() to avoid potential deadlocks. In practice, this trylock will succeed on almost every attempt as unmount/remount type operations are exceedingly rare. Note: we always need to pass a count of zero to generic_file_buffered_write() as the previously written byte count. We only do this by accident before this patch by the virtue of ret always being zero when there are no errors. Make this explicit rather than needing to specifically zero ret in the ENOSPC retry case. Signed-off-by: Dave Chinner <dchinner@redhat.com> Tested-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Ben Myers <bpm@sgi.com>
2012-10-08 17:56:04 +07:00
goto write_retry;
}
current->backing_dev_info = NULL;
out:
xfs: sync eofblocks scans under iolock are livelock prone The xfs_eofblocks.eof_scan_owner field is an internal field to facilitate invoking eofb scans from the kernel while under the iolock. This is necessary because the eofb scan acquires the iolock of each inode. Synchronous scans are invoked on certain buffered write failures while under iolock. In such cases, the scan owner indicates that the context for the scan already owns the particular iolock and prevents a double lock deadlock. eofblocks scans while under iolock are still livelock prone in the event of multiple parallel scans, however. If multiple buffered writes to different inodes fail and invoke eofblocks scans at the same time, each scan avoids a deadlock with its own inode by virtue of the eof_scan_owner field, but will never be able to acquire the iolock of the inode from the parallel scan. Because the low free space scans are invoked with SYNC_WAIT, the scan will not return until it has processed every tagged inode and thus both scans will spin indefinitely on the iolock being held across the opposite scan. This problem can be reproduced reliably by generic/224 on systems with higher cpu counts (x16). To avoid this problem, simplify the semantics of eofblocks scans to never invoke a scan while under iolock. This means that the buffered write context must drop the iolock before the scan. It must reacquire the lock before the write retry and also repeat the initial write checks, as the original state might no longer be valid once the iolock was dropped. Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-01-28 14:22:56 +07:00
if (iolock)
xfs_iunlock(ip, iolock);
if (ret > 0) {
XFS_STATS_ADD(ip->i_mount, xs_write_bytes, ret);
/* Handle various SYNC-type writes */
ret = generic_write_sync(iocb, ret);
}
return ret;
}
STATIC ssize_t
xfs_file_write_iter(
struct kiocb *iocb,
struct iov_iter *from)
{
struct file *file = iocb->ki_filp;
struct address_space *mapping = file->f_mapping;
struct inode *inode = mapping->host;
struct xfs_inode *ip = XFS_I(inode);
ssize_t ret;
size_t ocount = iov_iter_count(from);
XFS_STATS_INC(ip->i_mount, xs_write_calls);
if (ocount == 0)
return 0;
if (XFS_FORCED_SHUTDOWN(ip->i_mount))
return -EIO;
if (IS_DAX(inode))
return xfs_file_dax_write(iocb, from);
if (iocb->ki_flags & IOCB_DIRECT) {
/*
* Allow a directio write to fall back to a buffered
* write *only* in the case that we're doing a reflink
* CoW. In all other directio scenarios we do not
* allow an operation to fall back to buffered mode.
*/
ret = xfs_file_dio_aio_write(iocb, from);
if (ret != -EREMCHG)
return ret;
}
return xfs_file_buffered_aio_write(iocb, from);
}
xfs, dax: introduce xfs_break_dax_layouts() xfs_break_dax_layouts(), similar to xfs_break_leased_layouts(), scans for busy / pinned dax pages and waits for those pages to go idle before any potential extent unmap operation. dax_layout_busy_page() handles synchronizing against new page-busy events (get_user_pages). It invalidates all mappings to trigger the get_user_pages slow path which will eventually block on the xfs inode lock held in XFS_MMAPLOCK_EXCL mode. If dax_layout_busy_page() finds a busy page it returns it for xfs to wait for the page-idle event that will fire when the page reference count reaches 1 (recall ZONE_DEVICE pages are idle at count 1, see generic_dax_pagefree()). While waiting, the XFS_MMAPLOCK_EXCL lock is dropped in order to not deadlock the process that might be trying to elevate the page count of more pages before arranging for any of them to go idle. I.e. the typical case of submitting I/O is that iov_iter_get_pages() elevates the reference count of all pages in the I/O before starting I/O on the first page. The process of elevating the reference count of all pages involved in an I/O may cause faults that need to take XFS_MMAPLOCK_EXCL. Although XFS_MMAPLOCK_EXCL is dropped while waiting, XFS_IOLOCK_EXCL is held while sleeping. We need this to prevent starvation of the truncate path as continuous submission of direct-I/O could starve the truncate path indefinitely if the lock is dropped. Cc: Dave Chinner <david@fromorbit.com> Cc: Ross Zwisler <ross.zwisler@linux.intel.com> Reported-by: Jan Kara <jack@suse.cz> Reviewed-by: Jan Kara <jack@suse.cz> Reviewed-by: Christoph Hellwig <hch@lst.de> Acked-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2018-05-10 05:47:49 +07:00
static void
xfs_wait_dax_page(
struct inode *inode,
bool *did_unlock)
{
struct xfs_inode *ip = XFS_I(inode);
*did_unlock = true;
xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
schedule();
xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
}
static int
xfs_break_dax_layouts(
struct inode *inode,
uint iolock,
bool *did_unlock)
{
struct page *page;
ASSERT(xfs_isilocked(XFS_I(inode), XFS_MMAPLOCK_EXCL));
page = dax_layout_busy_page(inode->i_mapping);
if (!page)
return 0;
return ___wait_var_event(&page->_refcount,
atomic_read(&page->_refcount) == 1, TASK_INTERRUPTIBLE,
0, 0, xfs_wait_dax_page(inode, did_unlock));
}
xfs: prepare xfs_break_layouts() for another layout type When xfs is operating as the back-end of a pNFS block server, it prevents collisions between local and remote operations by requiring a lease to be held for remotely accessed blocks. Local filesystem operations break those leases before writing or mutating the extent map of the file. A similar mechanism is needed to prevent operations on pinned dax mappings, like device-DMA, from colliding with extent unmap operations. BREAK_WRITE and BREAK_UNMAP are introduced as two distinct levels of layout breaking. Layouts are broken in the BREAK_WRITE case to ensure that layout-holders do not collide with local writes. Additionally, layouts are broken in the BREAK_UNMAP case to make sure the layout-holder has a consistent view of the file's extent map. While BREAK_WRITE breaks can be satisfied be recalling FL_LAYOUT leases, BREAK_UNMAP breaks additionally require waiting for busy dax-pages to go idle while holding XFS_MMAPLOCK_EXCL. After this refactoring xfs_break_layouts() becomes the entry point for coordinating both types of breaks. Finally, xfs_break_leased_layouts() becomes just the BREAK_WRITE handler. Note that the unlock tracking is needed in a follow on change. That will coordinate retrying either break handler until both successfully test for a lease break while maintaining the lock state. Cc: Ross Zwisler <ross.zwisler@linux.intel.com> Cc: "Darrick J. Wong" <darrick.wong@oracle.com> Reported-by: Dave Chinner <david@fromorbit.com> Reported-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2018-03-21 04:42:38 +07:00
int
xfs_break_layouts(
struct inode *inode,
uint *iolock,
enum layout_break_reason reason)
{
bool retry;
xfs, dax: introduce xfs_break_dax_layouts() xfs_break_dax_layouts(), similar to xfs_break_leased_layouts(), scans for busy / pinned dax pages and waits for those pages to go idle before any potential extent unmap operation. dax_layout_busy_page() handles synchronizing against new page-busy events (get_user_pages). It invalidates all mappings to trigger the get_user_pages slow path which will eventually block on the xfs inode lock held in XFS_MMAPLOCK_EXCL mode. If dax_layout_busy_page() finds a busy page it returns it for xfs to wait for the page-idle event that will fire when the page reference count reaches 1 (recall ZONE_DEVICE pages are idle at count 1, see generic_dax_pagefree()). While waiting, the XFS_MMAPLOCK_EXCL lock is dropped in order to not deadlock the process that might be trying to elevate the page count of more pages before arranging for any of them to go idle. I.e. the typical case of submitting I/O is that iov_iter_get_pages() elevates the reference count of all pages in the I/O before starting I/O on the first page. The process of elevating the reference count of all pages involved in an I/O may cause faults that need to take XFS_MMAPLOCK_EXCL. Although XFS_MMAPLOCK_EXCL is dropped while waiting, XFS_IOLOCK_EXCL is held while sleeping. We need this to prevent starvation of the truncate path as continuous submission of direct-I/O could starve the truncate path indefinitely if the lock is dropped. Cc: Dave Chinner <david@fromorbit.com> Cc: Ross Zwisler <ross.zwisler@linux.intel.com> Reported-by: Jan Kara <jack@suse.cz> Reviewed-by: Jan Kara <jack@suse.cz> Reviewed-by: Christoph Hellwig <hch@lst.de> Acked-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2018-05-10 05:47:49 +07:00
int error;
xfs: prepare xfs_break_layouts() for another layout type When xfs is operating as the back-end of a pNFS block server, it prevents collisions between local and remote operations by requiring a lease to be held for remotely accessed blocks. Local filesystem operations break those leases before writing or mutating the extent map of the file. A similar mechanism is needed to prevent operations on pinned dax mappings, like device-DMA, from colliding with extent unmap operations. BREAK_WRITE and BREAK_UNMAP are introduced as two distinct levels of layout breaking. Layouts are broken in the BREAK_WRITE case to ensure that layout-holders do not collide with local writes. Additionally, layouts are broken in the BREAK_UNMAP case to make sure the layout-holder has a consistent view of the file's extent map. While BREAK_WRITE breaks can be satisfied be recalling FL_LAYOUT leases, BREAK_UNMAP breaks additionally require waiting for busy dax-pages to go idle while holding XFS_MMAPLOCK_EXCL. After this refactoring xfs_break_layouts() becomes the entry point for coordinating both types of breaks. Finally, xfs_break_leased_layouts() becomes just the BREAK_WRITE handler. Note that the unlock tracking is needed in a follow on change. That will coordinate retrying either break handler until both successfully test for a lease break while maintaining the lock state. Cc: Ross Zwisler <ross.zwisler@linux.intel.com> Cc: "Darrick J. Wong" <darrick.wong@oracle.com> Reported-by: Dave Chinner <david@fromorbit.com> Reported-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2018-03-21 04:42:38 +07:00
ASSERT(xfs_isilocked(XFS_I(inode), XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL));
xfs, dax: introduce xfs_break_dax_layouts() xfs_break_dax_layouts(), similar to xfs_break_leased_layouts(), scans for busy / pinned dax pages and waits for those pages to go idle before any potential extent unmap operation. dax_layout_busy_page() handles synchronizing against new page-busy events (get_user_pages). It invalidates all mappings to trigger the get_user_pages slow path which will eventually block on the xfs inode lock held in XFS_MMAPLOCK_EXCL mode. If dax_layout_busy_page() finds a busy page it returns it for xfs to wait for the page-idle event that will fire when the page reference count reaches 1 (recall ZONE_DEVICE pages are idle at count 1, see generic_dax_pagefree()). While waiting, the XFS_MMAPLOCK_EXCL lock is dropped in order to not deadlock the process that might be trying to elevate the page count of more pages before arranging for any of them to go idle. I.e. the typical case of submitting I/O is that iov_iter_get_pages() elevates the reference count of all pages in the I/O before starting I/O on the first page. The process of elevating the reference count of all pages involved in an I/O may cause faults that need to take XFS_MMAPLOCK_EXCL. Although XFS_MMAPLOCK_EXCL is dropped while waiting, XFS_IOLOCK_EXCL is held while sleeping. We need this to prevent starvation of the truncate path as continuous submission of direct-I/O could starve the truncate path indefinitely if the lock is dropped. Cc: Dave Chinner <david@fromorbit.com> Cc: Ross Zwisler <ross.zwisler@linux.intel.com> Reported-by: Jan Kara <jack@suse.cz> Reviewed-by: Jan Kara <jack@suse.cz> Reviewed-by: Christoph Hellwig <hch@lst.de> Acked-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2018-05-10 05:47:49 +07:00
do {
retry = false;
switch (reason) {
case BREAK_UNMAP:
error = xfs_break_dax_layouts(inode, *iolock, &retry);
if (error || retry)
break;
/* fall through */
case BREAK_WRITE:
error = xfs_break_leased_layouts(inode, iolock, &retry);
break;
default:
WARN_ON_ONCE(1);
error = -EINVAL;
}
} while (error == 0 && retry);
return error;
xfs: prepare xfs_break_layouts() for another layout type When xfs is operating as the back-end of a pNFS block server, it prevents collisions between local and remote operations by requiring a lease to be held for remotely accessed blocks. Local filesystem operations break those leases before writing or mutating the extent map of the file. A similar mechanism is needed to prevent operations on pinned dax mappings, like device-DMA, from colliding with extent unmap operations. BREAK_WRITE and BREAK_UNMAP are introduced as two distinct levels of layout breaking. Layouts are broken in the BREAK_WRITE case to ensure that layout-holders do not collide with local writes. Additionally, layouts are broken in the BREAK_UNMAP case to make sure the layout-holder has a consistent view of the file's extent map. While BREAK_WRITE breaks can be satisfied be recalling FL_LAYOUT leases, BREAK_UNMAP breaks additionally require waiting for busy dax-pages to go idle while holding XFS_MMAPLOCK_EXCL. After this refactoring xfs_break_layouts() becomes the entry point for coordinating both types of breaks. Finally, xfs_break_leased_layouts() becomes just the BREAK_WRITE handler. Note that the unlock tracking is needed in a follow on change. That will coordinate retrying either break handler until both successfully test for a lease break while maintaining the lock state. Cc: Ross Zwisler <ross.zwisler@linux.intel.com> Cc: "Darrick J. Wong" <darrick.wong@oracle.com> Reported-by: Dave Chinner <david@fromorbit.com> Reported-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2018-03-21 04:42:38 +07:00
}
#define XFS_FALLOC_FL_SUPPORTED \
(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE | \
FALLOC_FL_INSERT_RANGE | FALLOC_FL_UNSHARE_RANGE)
STATIC long
xfs_file_fallocate(
struct file *file,
int mode,
loff_t offset,
loff_t len)
{
struct inode *inode = file_inode(file);
struct xfs_inode *ip = XFS_I(inode);
long error;
enum xfs_prealloc_flags flags = 0;
uint iolock = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL;
loff_t new_size = 0;
bool do_file_insert = false;
if (!S_ISREG(inode->i_mode))
return -EINVAL;
if (mode & ~XFS_FALLOC_FL_SUPPORTED)
return -EOPNOTSUPP;
xfs_ilock(ip, iolock);
xfs: prepare xfs_break_layouts() for another layout type When xfs is operating as the back-end of a pNFS block server, it prevents collisions between local and remote operations by requiring a lease to be held for remotely accessed blocks. Local filesystem operations break those leases before writing or mutating the extent map of the file. A similar mechanism is needed to prevent operations on pinned dax mappings, like device-DMA, from colliding with extent unmap operations. BREAK_WRITE and BREAK_UNMAP are introduced as two distinct levels of layout breaking. Layouts are broken in the BREAK_WRITE case to ensure that layout-holders do not collide with local writes. Additionally, layouts are broken in the BREAK_UNMAP case to make sure the layout-holder has a consistent view of the file's extent map. While BREAK_WRITE breaks can be satisfied be recalling FL_LAYOUT leases, BREAK_UNMAP breaks additionally require waiting for busy dax-pages to go idle while holding XFS_MMAPLOCK_EXCL. After this refactoring xfs_break_layouts() becomes the entry point for coordinating both types of breaks. Finally, xfs_break_leased_layouts() becomes just the BREAK_WRITE handler. Note that the unlock tracking is needed in a follow on change. That will coordinate retrying either break handler until both successfully test for a lease break while maintaining the lock state. Cc: Ross Zwisler <ross.zwisler@linux.intel.com> Cc: "Darrick J. Wong" <darrick.wong@oracle.com> Reported-by: Dave Chinner <david@fromorbit.com> Reported-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2018-03-21 04:42:38 +07:00
error = xfs_break_layouts(inode, &iolock, BREAK_UNMAP);
if (error)
goto out_unlock;
if (mode & FALLOC_FL_PUNCH_HOLE) {
error = xfs_free_file_space(ip, offset, len);
if (error)
goto out_unlock;
} else if (mode & FALLOC_FL_COLLAPSE_RANGE) {
unsigned int blksize_mask = i_blocksize(inode) - 1;
if (offset & blksize_mask || len & blksize_mask) {
error = -EINVAL;
goto out_unlock;
}
/*
* There is no need to overlap collapse range with EOF,
* in which case it is effectively a truncate operation
*/
if (offset + len >= i_size_read(inode)) {
error = -EINVAL;
goto out_unlock;
}
new_size = i_size_read(inode) - len;
error = xfs_collapse_file_space(ip, offset, len);
if (error)
goto out_unlock;
} else if (mode & FALLOC_FL_INSERT_RANGE) {
unsigned int blksize_mask = i_blocksize(inode) - 1;
loff_t isize = i_size_read(inode);
if (offset & blksize_mask || len & blksize_mask) {
error = -EINVAL;
goto out_unlock;
}
/*
* New inode size must not exceed ->s_maxbytes, accounting for
* possible signed overflow.
*/
if (inode->i_sb->s_maxbytes - isize < len) {
error = -EFBIG;
goto out_unlock;
}
new_size = isize + len;
/* Offset should be less than i_size */
if (offset >= isize) {
error = -EINVAL;
goto out_unlock;
}
do_file_insert = true;
} else {
flags |= XFS_PREALLOC_SET;
if (!(mode & FALLOC_FL_KEEP_SIZE) &&
offset + len > i_size_read(inode)) {
new_size = offset + len;
error = inode_newsize_ok(inode, new_size);
if (error)
goto out_unlock;
}
if (mode & FALLOC_FL_ZERO_RANGE)
error = xfs_zero_file_space(ip, offset, len);
else {
if (mode & FALLOC_FL_UNSHARE_RANGE) {
error = xfs_reflink_unshare(ip, offset, len);
if (error)
goto out_unlock;
}
error = xfs_alloc_file_space(ip, offset, len,
XFS_BMAPI_PREALLOC);
}
if (error)
goto out_unlock;
}
if (file->f_flags & O_DSYNC)
flags |= XFS_PREALLOC_SYNC;
error = xfs_update_prealloc_flags(ip, flags);
if (error)
goto out_unlock;
/* Change file size if needed */
if (new_size) {
struct iattr iattr;
iattr.ia_valid = ATTR_SIZE;
iattr.ia_size = new_size;
error = xfs_vn_setattr_size(file_dentry(file), &iattr);
if (error)
goto out_unlock;
}
/*
* Perform hole insertion now that the file size has been
* updated so that if we crash during the operation we don't
* leave shifted extents past EOF and hence losing access to
* the data that is contained within them.
*/
if (do_file_insert)
error = xfs_insert_file_space(ip, offset, len);
out_unlock:
xfs_iunlock(ip, iolock);
return error;
}
STATIC int
xfs_file_clone_range(
struct file *file_in,
loff_t pos_in,
struct file *file_out,
loff_t pos_out,
u64 len)
{
return xfs_reflink_remap_range(file_in, pos_in, file_out, pos_out,
len, false);
}
STATIC ssize_t
xfs_file_dedupe_range(
struct file *src_file,
u64 loff,
u64 len,
struct file *dst_file,
u64 dst_loff)
{
struct inode *srci = file_inode(src_file);
u64 max_dedupe;
int error;
/*
* Since we have to read all these pages in to compare them, cut
* it off at MAX_RW_COUNT/2 rounded down to the nearest block.
* That means we won't do more than MAX_RW_COUNT IO per request.
*/
max_dedupe = (MAX_RW_COUNT >> 1) & ~(i_blocksize(srci) - 1);
if (len > max_dedupe)
len = max_dedupe;
error = xfs_reflink_remap_range(src_file, loff, dst_file, dst_loff,
len, true);
if (error)
return error;
return len;
}
STATIC int
xfs_file_open(
struct inode *inode,
struct file *file)
{
if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
return -EFBIG;
if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb)))
return -EIO;
file->f_mode |= FMODE_NOWAIT;
return 0;
}
STATIC int
xfs_dir_open(
struct inode *inode,
struct file *file)
{
struct xfs_inode *ip = XFS_I(inode);
int mode;
int error;
error = xfs_file_open(inode, file);
if (error)
return error;
/*
* If there are any blocks, read-ahead block 0 as we're almost
* certain to have the next operation be a read there.
*/
mode = xfs_ilock_data_map_shared(ip);
if (ip->i_d.di_nextents > 0)
error = xfs_dir3_data_readahead(ip, 0, -1);
xfs_iunlock(ip, mode);
return error;
}
STATIC int
xfs_file_release(
struct inode *inode,
struct file *filp)
{
return xfs_release(XFS_I(inode));
}
STATIC int
xfs_file_readdir(
struct file *file,
struct dir_context *ctx)
{
struct inode *inode = file_inode(file);
xfs_inode_t *ip = XFS_I(inode);
size_t bufsize;
/*
* The Linux API doesn't pass down the total size of the buffer
* we read into down to the filesystem. With the filldir concept
* it's not needed for correct information, but the XFS dir2 leaf
* code wants an estimate of the buffer size to calculate it's
* readahead window and size the buffers used for mapping to
* physical blocks.
*
* Try to give it an estimate that's good enough, maybe at some
* point we can change the ->readdir prototype to include the
* buffer size. For now we use the current glibc buffer size.
*/
bufsize = (size_t)min_t(loff_t, XFS_READDIR_BUFSIZE, ip->i_d.di_size);
return xfs_readdir(NULL, ip, ctx, bufsize);
}
STATIC loff_t
xfs_file_llseek(
struct file *file,
loff_t offset,
int whence)
{
struct inode *inode = file->f_mapping->host;
if (XFS_FORCED_SHUTDOWN(XFS_I(inode)->i_mount))
return -EIO;
switch (whence) {
default:
return generic_file_llseek(file, offset, whence);
case SEEK_HOLE:
offset = iomap_seek_hole(inode, offset, &xfs_iomap_ops);
break;
case SEEK_DATA:
offset = iomap_seek_data(inode, offset, &xfs_iomap_ops);
break;
}
if (offset < 0)
return offset;
return vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
}
/*
* Locking for serialisation of IO during page faults. This results in a lock
* ordering of:
*
* mmap_sem (MM)
* sb_start_pagefault(vfs, freeze)
* i_mmaplock (XFS - truncate serialisation)
* page_lock (MM)
* i_lock (XFS - extent map serialisation)
*/
static vm_fault_t
__xfs_filemap_fault(
struct vm_fault *vmf,
enum page_entry_size pe_size,
bool write_fault)
{
struct inode *inode = file_inode(vmf->vma->vm_file);
struct xfs_inode *ip = XFS_I(inode);
vm_fault_t ret;
trace_xfs_filemap_fault(ip, pe_size, write_fault);
if (write_fault) {
sb_start_pagefault(inode->i_sb);
file_update_time(vmf->vma->vm_file);
}
xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
if (IS_DAX(inode)) {
pfn_t pfn;
ret = dax_iomap_fault(vmf, pe_size, &pfn, NULL, &xfs_iomap_ops);
if (ret & VM_FAULT_NEEDDSYNC)
ret = dax_finish_sync_fault(vmf, pe_size, pfn);
} else {
if (write_fault)
ret = iomap_page_mkwrite(vmf, &xfs_iomap_ops);
else
ret = filemap_fault(vmf);
}
xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
if (write_fault)
sb_end_pagefault(inode->i_sb);
return ret;
}
static vm_fault_t
xfs_filemap_fault(
struct vm_fault *vmf)
{
/* DAX can shortcut the normal fault path on write faults! */
return __xfs_filemap_fault(vmf, PE_SIZE_PTE,
IS_DAX(file_inode(vmf->vma->vm_file)) &&
(vmf->flags & FAULT_FLAG_WRITE));
}
static vm_fault_t
mm,fs,dax: change ->pmd_fault to ->huge_fault Patch series "1G transparent hugepage support for device dax", v2. The following series implements support for 1G trasparent hugepage on x86 for device dax. The bulk of the code was written by Mathew Wilcox a while back supporting transparent 1G hugepage for fs DAX. I have forward ported the relevant bits to 4.10-rc. The current submission has only the necessary code to support device DAX. Comments from Dan Williams: So the motivation and intended user of this functionality mirrors the motivation and users of 1GB page support in hugetlbfs. Given expected capacities of persistent memory devices an in-memory database may want to reduce tlb pressure beyond what they can already achieve with 2MB mappings of a device-dax file. We have customer feedback to that effect as Willy mentioned in his previous version of these patches [1]. [1]: https://lkml.org/lkml/2016/1/31/52 Comments from Nilesh @ Oracle: There are applications which have a process model; and if you assume 10,000 processes attempting to mmap all the 6TB memory available on a server; we are looking at the following: processes : 10,000 memory : 6TB pte @ 4k page size: 8 bytes / 4K of memory * #processes = 6TB / 4k * 8 * 10000 = 1.5GB * 80000 = 120,000GB pmd @ 2M page size: 120,000 / 512 = ~240GB pud @ 1G page size: 240GB / 512 = ~480MB As you can see with 2M pages, this system will use up an exorbitant amount of DRAM to hold the page tables; but the 1G pages finally brings it down to a reasonable level. Memory sizes will keep increasing; so this number will keep increasing. An argument can be made to convert the applications from process model to thread model, but in the real world that may not be always practical. Hopefully this helps explain the use case where this is valuable. This patch (of 3): In preparation for adding the ability to handle PUD pages, convert vm_operations_struct.pmd_fault to vm_operations_struct.huge_fault. The vm_fault structure is extended to include a union of the different page table pointers that may be needed, and three flag bits are reserved to indicate which type of pointer is in the union. [ross.zwisler@linux.intel.com: remove unused function ext4_dax_huge_fault()] Link: http://lkml.kernel.org/r/1485813172-7284-1-git-send-email-ross.zwisler@linux.intel.com [dave.jiang@intel.com: clear PMD or PUD size flags when in fall through path] Link: http://lkml.kernel.org/r/148589842696.5820.16078080610311444794.stgit@djiang5-desk3.ch.intel.com Link: http://lkml.kernel.org/r/148545058784.17912.6353162518188733642.stgit@djiang5-desk3.ch.intel.com Signed-off-by: Matthew Wilcox <mawilcox@microsoft.com> Signed-off-by: Dave Jiang <dave.jiang@intel.com> Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com> Cc: Dave Hansen <dave.hansen@linux.intel.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: Jan Kara <jack@suse.com> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> Cc: Nilesh Choudhury <nilesh.choudhury@oracle.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Dave Jiang <dave.jiang@intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-02-25 05:56:59 +07:00
xfs_filemap_huge_fault(
struct vm_fault *vmf,
enum page_entry_size pe_size)
{
if (!IS_DAX(file_inode(vmf->vma->vm_file)))
return VM_FAULT_FALLBACK;
/* DAX can shortcut the normal fault path on write faults! */
return __xfs_filemap_fault(vmf, pe_size,
(vmf->flags & FAULT_FLAG_WRITE));
}
static vm_fault_t
xfs_filemap_page_mkwrite(
struct vm_fault *vmf)
{
return __xfs_filemap_fault(vmf, PE_SIZE_PTE, true);
}
/*
* pfn_mkwrite was originally intended to ensure we capture time stamp updates
* on write faults. In reality, it needs to serialise against truncate and
* prepare memory for writing so handle is as standard write fault.
*/
static vm_fault_t
xfs_filemap_pfn_mkwrite(
struct vm_fault *vmf)
{
return __xfs_filemap_fault(vmf, PE_SIZE_PTE, true);
}
static const struct vm_operations_struct xfs_file_vm_ops = {
.fault = xfs_filemap_fault,
mm,fs,dax: change ->pmd_fault to ->huge_fault Patch series "1G transparent hugepage support for device dax", v2. The following series implements support for 1G trasparent hugepage on x86 for device dax. The bulk of the code was written by Mathew Wilcox a while back supporting transparent 1G hugepage for fs DAX. I have forward ported the relevant bits to 4.10-rc. The current submission has only the necessary code to support device DAX. Comments from Dan Williams: So the motivation and intended user of this functionality mirrors the motivation and users of 1GB page support in hugetlbfs. Given expected capacities of persistent memory devices an in-memory database may want to reduce tlb pressure beyond what they can already achieve with 2MB mappings of a device-dax file. We have customer feedback to that effect as Willy mentioned in his previous version of these patches [1]. [1]: https://lkml.org/lkml/2016/1/31/52 Comments from Nilesh @ Oracle: There are applications which have a process model; and if you assume 10,000 processes attempting to mmap all the 6TB memory available on a server; we are looking at the following: processes : 10,000 memory : 6TB pte @ 4k page size: 8 bytes / 4K of memory * #processes = 6TB / 4k * 8 * 10000 = 1.5GB * 80000 = 120,000GB pmd @ 2M page size: 120,000 / 512 = ~240GB pud @ 1G page size: 240GB / 512 = ~480MB As you can see with 2M pages, this system will use up an exorbitant amount of DRAM to hold the page tables; but the 1G pages finally brings it down to a reasonable level. Memory sizes will keep increasing; so this number will keep increasing. An argument can be made to convert the applications from process model to thread model, but in the real world that may not be always practical. Hopefully this helps explain the use case where this is valuable. This patch (of 3): In preparation for adding the ability to handle PUD pages, convert vm_operations_struct.pmd_fault to vm_operations_struct.huge_fault. The vm_fault structure is extended to include a union of the different page table pointers that may be needed, and three flag bits are reserved to indicate which type of pointer is in the union. [ross.zwisler@linux.intel.com: remove unused function ext4_dax_huge_fault()] Link: http://lkml.kernel.org/r/1485813172-7284-1-git-send-email-ross.zwisler@linux.intel.com [dave.jiang@intel.com: clear PMD or PUD size flags when in fall through path] Link: http://lkml.kernel.org/r/148589842696.5820.16078080610311444794.stgit@djiang5-desk3.ch.intel.com Link: http://lkml.kernel.org/r/148545058784.17912.6353162518188733642.stgit@djiang5-desk3.ch.intel.com Signed-off-by: Matthew Wilcox <mawilcox@microsoft.com> Signed-off-by: Dave Jiang <dave.jiang@intel.com> Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com> Cc: Dave Hansen <dave.hansen@linux.intel.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: Jan Kara <jack@suse.com> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> Cc: Nilesh Choudhury <nilesh.choudhury@oracle.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Dave Jiang <dave.jiang@intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-02-25 05:56:59 +07:00
.huge_fault = xfs_filemap_huge_fault,
.map_pages = filemap_map_pages,
.page_mkwrite = xfs_filemap_page_mkwrite,
.pfn_mkwrite = xfs_filemap_pfn_mkwrite,
};
STATIC int
xfs_file_mmap(
struct file *filp,
struct vm_area_struct *vma)
{
/*
* We don't support synchronous mappings for non-DAX files. At least
* until someone comes with a sensible use case.
*/
if (!IS_DAX(file_inode(filp)) && (vma->vm_flags & VM_SYNC))
return -EOPNOTSUPP;
file_accessed(filp);
vma->vm_ops = &xfs_file_vm_ops;
if (IS_DAX(file_inode(filp)))
vma->vm_flags |= VM_MIXEDMAP | VM_HUGEPAGE;
return 0;
}
const struct file_operations xfs_file_operations = {
.llseek = xfs_file_llseek,
.read_iter = xfs_file_read_iter,
.write_iter = xfs_file_write_iter,
.splice_read = generic_file_splice_read,
.splice_write = iter_file_splice_write,
.unlocked_ioctl = xfs_file_ioctl,
#ifdef CONFIG_COMPAT
.compat_ioctl = xfs_file_compat_ioctl,
#endif
.mmap = xfs_file_mmap,
.mmap_supported_flags = MAP_SYNC,
.open = xfs_file_open,
.release = xfs_file_release,
.fsync = xfs_file_fsync,
.get_unmapped_area = thp_get_unmapped_area,
.fallocate = xfs_file_fallocate,
.clone_file_range = xfs_file_clone_range,
.dedupe_file_range = xfs_file_dedupe_range,
};
const struct file_operations xfs_dir_file_operations = {
.open = xfs_dir_open,
.read = generic_read_dir,
.iterate_shared = xfs_file_readdir,
.llseek = generic_file_llseek,
.unlocked_ioctl = xfs_file_ioctl,
#ifdef CONFIG_COMPAT
.compat_ioctl = xfs_file_compat_ioctl,
#endif
.fsync = xfs_dir_fsync,
};