1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
8 #include "xfs_shared.h"
9 #include "xfs_format.h"
10 #include "xfs_log_format.h"
11 #include "xfs_trans_resv.h"
12 #include "xfs_mount.h"
13 #include "xfs_inode.h"
14 #include "xfs_trans.h"
15 #include "xfs_inode_item.h"
17 #include "xfs_bmap_util.h"
19 #include "xfs_dir2_priv.h"
20 #include "xfs_ioctl.h"
21 #include "xfs_trace.h"
23 #include "xfs_icache.h"
25 #include "xfs_iomap.h"
26 #include "xfs_reflink.h"
28 #include <linux/falloc.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mman.h>
31 #include <linux/fadvise.h>
32 #include <linux/mount.h>
34 static const struct vm_operations_struct xfs_file_vm_ops;
37 * Decide if the given file range is aligned to the size of the fundamental
38 * allocation unit for the file.
41 xfs_is_falloc_aligned(
46 struct xfs_mount *mp = ip->i_mount;
49 if (XFS_IS_REALTIME_INODE(ip)) {
50 if (!is_power_of_2(mp->m_sb.sb_rextsize)) {
54 rextbytes = XFS_FSB_TO_B(mp, mp->m_sb.sb_rextsize);
55 div_u64_rem(pos, rextbytes, &mod);
58 div_u64_rem(len, rextbytes, &mod);
61 mask = XFS_FSB_TO_B(mp, mp->m_sb.sb_rextsize) - 1;
63 mask = mp->m_sb.sb_blocksize - 1;
66 return !((pos | len) & mask);
70 xfs_update_prealloc_flags(
72 enum xfs_prealloc_flags flags)
77 error = xfs_trans_alloc(ip->i_mount, &M_RES(ip->i_mount)->tr_writeid,
82 xfs_ilock(ip, XFS_ILOCK_EXCL);
83 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
85 if (!(flags & XFS_PREALLOC_INVISIBLE)) {
86 VFS_I(ip)->i_mode &= ~S_ISUID;
87 if (VFS_I(ip)->i_mode & S_IXGRP)
88 VFS_I(ip)->i_mode &= ~S_ISGID;
89 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
92 if (flags & XFS_PREALLOC_SET)
93 ip->i_diflags |= XFS_DIFLAG_PREALLOC;
94 if (flags & XFS_PREALLOC_CLEAR)
95 ip->i_diflags &= ~XFS_DIFLAG_PREALLOC;
97 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
98 return xfs_trans_commit(tp);
102 * Fsync operations on directories are much simpler than on regular files,
103 * as there is no file data to flush, and thus also no need for explicit
104 * cache flush operations, and there are no non-transaction metadata updates
105 * on directories either.
114 struct xfs_inode *ip = XFS_I(file->f_mapping->host);
116 trace_xfs_dir_fsync(ip);
117 return xfs_log_force_inode(ip);
122 struct xfs_inode *ip,
125 if (!xfs_ipincount(ip))
127 if (datasync && !(ip->i_itemp->ili_fsync_fields & ~XFS_ILOG_TIMESTAMP))
129 return ip->i_itemp->ili_commit_seq;
133 * All metadata updates are logged, which means that we just have to flush the
134 * log up to the latest LSN that touched the inode.
136 * If we have concurrent fsync/fdatasync() calls, we need them to all block on
137 * the log force before we clear the ili_fsync_fields field. This ensures that
138 * we don't get a racing sync operation that does not wait for the metadata to
139 * hit the journal before returning. If we race with clearing ili_fsync_fields,
140 * then all that will happen is the log force will do nothing as the lsn will
141 * already be on disk. We can't race with setting ili_fsync_fields because that
142 * is done under XFS_ILOCK_EXCL, and that can't happen because we hold the lock
143 * shared until after the ili_fsync_fields is cleared.
147 struct xfs_inode *ip,
154 xfs_ilock(ip, XFS_ILOCK_SHARED);
155 seq = xfs_fsync_seq(ip, datasync);
157 error = xfs_log_force_seq(ip->i_mount, seq, XFS_LOG_SYNC,
160 spin_lock(&ip->i_itemp->ili_lock);
161 ip->i_itemp->ili_fsync_fields = 0;
162 spin_unlock(&ip->i_itemp->ili_lock);
164 xfs_iunlock(ip, XFS_ILOCK_SHARED);
175 struct xfs_inode *ip = XFS_I(file->f_mapping->host);
176 struct xfs_mount *mp = ip->i_mount;
180 trace_xfs_file_fsync(ip);
182 error = file_write_and_wait_range(file, start, end);
186 if (xfs_is_shutdown(mp))
189 xfs_iflags_clear(ip, XFS_ITRUNCATED);
192 * If we have an RT and/or log subvolume we need to make sure to flush
193 * the write cache the device used for file data first. This is to
194 * ensure newly written file data make it to disk before logging the new
195 * inode size in case of an extending write.
197 if (XFS_IS_REALTIME_INODE(ip))
198 blkdev_issue_flush(mp->m_rtdev_targp->bt_bdev);
199 else if (mp->m_logdev_targp != mp->m_ddev_targp)
200 blkdev_issue_flush(mp->m_ddev_targp->bt_bdev);
203 * Any inode that has dirty modifications in the log is pinned. The
204 * racy check here for a pinned inode while not catch modifications
205 * that happen concurrently to the fsync call, but fsync semantics
206 * only require to sync previously completed I/O.
208 if (xfs_ipincount(ip))
209 error = xfs_fsync_flush_log(ip, datasync, &log_flushed);
212 * If we only have a single device, and the log force about was
213 * a no-op we might have to flush the data device cache here.
214 * This can only happen for fdatasync/O_DSYNC if we were overwriting
215 * an already allocated file and thus do not have any metadata to
218 if (!log_flushed && !XFS_IS_REALTIME_INODE(ip) &&
219 mp->m_logdev_targp == mp->m_ddev_targp)
220 blkdev_issue_flush(mp->m_ddev_targp->bt_bdev);
228 unsigned int lock_mode)
230 struct xfs_inode *ip = XFS_I(file_inode(iocb->ki_filp));
232 if (iocb->ki_flags & IOCB_NOWAIT) {
233 if (!xfs_ilock_nowait(ip, lock_mode))
236 xfs_ilock(ip, lock_mode);
247 struct xfs_inode *ip = XFS_I(file_inode(iocb->ki_filp));
250 trace_xfs_file_direct_read(iocb, to);
252 if (!iov_iter_count(to))
253 return 0; /* skip atime */
255 file_accessed(iocb->ki_filp);
257 ret = xfs_ilock_iocb(iocb, XFS_IOLOCK_SHARED);
260 ret = iomap_dio_rw(iocb, to, &xfs_read_iomap_ops, NULL, 0, 0);
261 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
266 static noinline ssize_t
271 struct xfs_inode *ip = XFS_I(iocb->ki_filp->f_mapping->host);
274 trace_xfs_file_dax_read(iocb, to);
276 if (!iov_iter_count(to))
277 return 0; /* skip atime */
279 ret = xfs_ilock_iocb(iocb, XFS_IOLOCK_SHARED);
282 ret = dax_iomap_rw(iocb, to, &xfs_read_iomap_ops);
283 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
285 file_accessed(iocb->ki_filp);
290 xfs_file_buffered_read(
294 struct xfs_inode *ip = XFS_I(file_inode(iocb->ki_filp));
297 trace_xfs_file_buffered_read(iocb, to);
299 ret = xfs_ilock_iocb(iocb, XFS_IOLOCK_SHARED);
302 ret = generic_file_read_iter(iocb, to);
303 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
313 struct inode *inode = file_inode(iocb->ki_filp);
314 struct xfs_mount *mp = XFS_I(inode)->i_mount;
317 XFS_STATS_INC(mp, xs_read_calls);
319 if (xfs_is_shutdown(mp))
323 ret = xfs_file_dax_read(iocb, to);
324 else if (iocb->ki_flags & IOCB_DIRECT)
325 ret = xfs_file_dio_read(iocb, to);
327 ret = xfs_file_buffered_read(iocb, to);
330 XFS_STATS_ADD(mp, xs_read_bytes, ret);
335 * Common pre-write limit and setup checks.
337 * Called with the iolocked held either shared and exclusive according to
338 * @iolock, and returns with it held. Might upgrade the iolock to exclusive
339 * if called for a direct write beyond i_size.
342 xfs_file_write_checks(
344 struct iov_iter *from,
347 struct file *file = iocb->ki_filp;
348 struct inode *inode = file->f_mapping->host;
349 struct xfs_inode *ip = XFS_I(inode);
351 size_t count = iov_iter_count(from);
352 bool drained_dio = false;
356 error = generic_write_checks(iocb, from);
360 if (iocb->ki_flags & IOCB_NOWAIT) {
361 error = break_layout(inode, false);
362 if (error == -EWOULDBLOCK)
365 error = xfs_break_layouts(inode, iolock, BREAK_WRITE);
372 * For changing security info in file_remove_privs() we need i_rwsem
375 if (*iolock == XFS_IOLOCK_SHARED && !IS_NOSEC(inode)) {
376 xfs_iunlock(ip, *iolock);
377 *iolock = XFS_IOLOCK_EXCL;
378 error = xfs_ilock_iocb(iocb, *iolock);
387 * If the offset is beyond the size of the file, we need to zero any
388 * blocks that fall between the existing EOF and the start of this
389 * write. If zeroing is needed and we are currently holding the iolock
390 * shared, we need to update it to exclusive which implies having to
391 * redo all checks before.
393 * We need to serialise against EOF updates that occur in IO completions
394 * here. We want to make sure that nobody is changing the size while we
395 * do this check until we have placed an IO barrier (i.e. hold the
396 * XFS_IOLOCK_EXCL) that prevents new IO from being dispatched. The
397 * spinlock effectively forms a memory barrier once we have the
398 * XFS_IOLOCK_EXCL so we are guaranteed to see the latest EOF value and
399 * hence be able to correctly determine if we need to run zeroing.
401 * We can do an unlocked check here safely as IO completion can only
402 * extend EOF. Truncate is locked out at this point, so the EOF can
403 * not move backwards, only forwards. Hence we only need to take the
404 * slow path and spin locks when we are at or beyond the current EOF.
406 if (iocb->ki_pos <= i_size_read(inode))
409 spin_lock(&ip->i_flags_lock);
410 isize = i_size_read(inode);
411 if (iocb->ki_pos > isize) {
412 spin_unlock(&ip->i_flags_lock);
414 if (iocb->ki_flags & IOCB_NOWAIT)
418 if (*iolock == XFS_IOLOCK_SHARED) {
419 xfs_iunlock(ip, *iolock);
420 *iolock = XFS_IOLOCK_EXCL;
421 xfs_ilock(ip, *iolock);
422 iov_iter_reexpand(from, count);
425 * We now have an IO submission barrier in place, but
426 * AIO can do EOF updates during IO completion and hence
427 * we now need to wait for all of them to drain. Non-AIO
428 * DIO will have drained before we are given the
429 * XFS_IOLOCK_EXCL, and so for most cases this wait is a
432 inode_dio_wait(inode);
437 trace_xfs_zero_eof(ip, isize, iocb->ki_pos - isize);
438 error = iomap_zero_range(inode, isize, iocb->ki_pos - isize,
439 NULL, &xfs_buffered_write_iomap_ops);
443 spin_unlock(&ip->i_flags_lock);
446 return file_modified(file);
450 xfs_dio_write_end_io(
456 struct inode *inode = file_inode(iocb->ki_filp);
457 struct xfs_inode *ip = XFS_I(inode);
458 loff_t offset = iocb->ki_pos;
459 unsigned int nofs_flag;
461 trace_xfs_end_io_direct_write(ip, offset, size);
463 if (xfs_is_shutdown(ip->i_mount))
472 * Capture amount written on completion as we can't reliably account
473 * for it on submission.
475 XFS_STATS_ADD(ip->i_mount, xs_write_bytes, size);
478 * We can allocate memory here while doing writeback on behalf of
479 * memory reclaim. To avoid memory allocation deadlocks set the
480 * task-wide nofs context for the following operations.
482 nofs_flag = memalloc_nofs_save();
484 if (flags & IOMAP_DIO_COW) {
485 error = xfs_reflink_end_cow(ip, offset, size);
491 * Unwritten conversion updates the in-core isize after extent
492 * conversion but before updating the on-disk size. Updating isize any
493 * earlier allows a racing dio read to find unwritten extents before
494 * they are converted.
496 if (flags & IOMAP_DIO_UNWRITTEN) {
497 error = xfs_iomap_write_unwritten(ip, offset, size, true);
502 * We need to update the in-core inode size here so that we don't end up
503 * with the on-disk inode size being outside the in-core inode size. We
504 * have no other method of updating EOF for AIO, so always do it here
507 * We need to lock the test/set EOF update as we can be racing with
508 * other IO completions here to update the EOF. Failing to serialise
509 * here can result in EOF moving backwards and Bad Things Happen when
512 * As IO completion only ever extends EOF, we can do an unlocked check
513 * here to avoid taking the spinlock. If we land within the current EOF,
514 * then we do not need to do an extending update at all, and we don't
515 * need to take the lock to check this. If we race with an update moving
516 * EOF, then we'll either still be beyond EOF and need to take the lock,
517 * or we'll be within EOF and we don't need to take it at all.
519 if (offset + size <= i_size_read(inode))
522 spin_lock(&ip->i_flags_lock);
523 if (offset + size > i_size_read(inode)) {
524 i_size_write(inode, offset + size);
525 spin_unlock(&ip->i_flags_lock);
526 error = xfs_setfilesize(ip, offset, size);
528 spin_unlock(&ip->i_flags_lock);
532 memalloc_nofs_restore(nofs_flag);
536 static const struct iomap_dio_ops xfs_dio_write_ops = {
537 .end_io = xfs_dio_write_end_io,
541 * Handle block aligned direct I/O writes
543 static noinline ssize_t
544 xfs_file_dio_write_aligned(
545 struct xfs_inode *ip,
547 struct iov_iter *from)
549 int iolock = XFS_IOLOCK_SHARED;
552 ret = xfs_ilock_iocb(iocb, iolock);
555 ret = xfs_file_write_checks(iocb, from, &iolock);
560 * We don't need to hold the IOLOCK exclusively across the IO, so demote
561 * the iolock back to shared if we had to take the exclusive lock in
562 * xfs_file_write_checks() for other reasons.
564 if (iolock == XFS_IOLOCK_EXCL) {
565 xfs_ilock_demote(ip, XFS_IOLOCK_EXCL);
566 iolock = XFS_IOLOCK_SHARED;
568 trace_xfs_file_direct_write(iocb, from);
569 ret = iomap_dio_rw(iocb, from, &xfs_direct_write_iomap_ops,
570 &xfs_dio_write_ops, 0, 0);
573 xfs_iunlock(ip, iolock);
578 * Handle block unaligned direct I/O writes
580 * In most cases direct I/O writes will be done holding IOLOCK_SHARED, allowing
581 * them to be done in parallel with reads and other direct I/O writes. However,
582 * if the I/O is not aligned to filesystem blocks, the direct I/O layer may need
583 * to do sub-block zeroing and that requires serialisation against other direct
584 * I/O to the same block. In this case we need to serialise the submission of
585 * the unaligned I/O so that we don't get racing block zeroing in the dio layer.
586 * In the case where sub-block zeroing is not required, we can do concurrent
587 * sub-block dios to the same block successfully.
589 * Optimistically submit the I/O using the shared lock first, but use the
590 * IOMAP_DIO_OVERWRITE_ONLY flag to tell the lower layers to return -EAGAIN
591 * if block allocation or partial block zeroing would be required. In that case
592 * we try again with the exclusive lock.
594 static noinline ssize_t
595 xfs_file_dio_write_unaligned(
596 struct xfs_inode *ip,
598 struct iov_iter *from)
600 size_t isize = i_size_read(VFS_I(ip));
601 size_t count = iov_iter_count(from);
602 int iolock = XFS_IOLOCK_SHARED;
603 unsigned int flags = IOMAP_DIO_OVERWRITE_ONLY;
607 * Extending writes need exclusivity because of the sub-block zeroing
608 * that the DIO code always does for partial tail blocks beyond EOF, so
609 * don't even bother trying the fast path in this case.
611 if (iocb->ki_pos > isize || iocb->ki_pos + count >= isize) {
613 if (iocb->ki_flags & IOCB_NOWAIT)
615 iolock = XFS_IOLOCK_EXCL;
616 flags = IOMAP_DIO_FORCE_WAIT;
619 ret = xfs_ilock_iocb(iocb, iolock);
624 * We can't properly handle unaligned direct I/O to reflink files yet,
625 * as we can't unshare a partial block.
627 if (xfs_is_cow_inode(ip)) {
628 trace_xfs_reflink_bounce_dio_write(iocb, from);
633 ret = xfs_file_write_checks(iocb, from, &iolock);
638 * If we are doing exclusive unaligned I/O, this must be the only I/O
639 * in-flight. Otherwise we risk data corruption due to unwritten extent
640 * conversions from the AIO end_io handler. Wait for all other I/O to
643 if (flags & IOMAP_DIO_FORCE_WAIT)
644 inode_dio_wait(VFS_I(ip));
646 trace_xfs_file_direct_write(iocb, from);
647 ret = iomap_dio_rw(iocb, from, &xfs_direct_write_iomap_ops,
648 &xfs_dio_write_ops, flags, 0);
651 * Retry unaligned I/O with exclusive blocking semantics if the DIO
652 * layer rejected it for mapping or locking reasons. If we are doing
653 * nonblocking user I/O, propagate the error.
655 if (ret == -EAGAIN && !(iocb->ki_flags & IOCB_NOWAIT)) {
656 ASSERT(flags & IOMAP_DIO_OVERWRITE_ONLY);
657 xfs_iunlock(ip, iolock);
658 goto retry_exclusive;
663 xfs_iunlock(ip, iolock);
670 struct iov_iter *from)
672 struct xfs_inode *ip = XFS_I(file_inode(iocb->ki_filp));
673 struct xfs_buftarg *target = xfs_inode_buftarg(ip);
674 size_t count = iov_iter_count(from);
676 /* direct I/O must be aligned to device logical sector size */
677 if ((iocb->ki_pos | count) & target->bt_logical_sectormask)
679 if ((iocb->ki_pos | count) & ip->i_mount->m_blockmask)
680 return xfs_file_dio_write_unaligned(ip, iocb, from);
681 return xfs_file_dio_write_aligned(ip, iocb, from);
684 static noinline ssize_t
687 struct iov_iter *from)
689 struct inode *inode = iocb->ki_filp->f_mapping->host;
690 struct xfs_inode *ip = XFS_I(inode);
691 int iolock = XFS_IOLOCK_EXCL;
692 ssize_t ret, error = 0;
695 ret = xfs_ilock_iocb(iocb, iolock);
698 ret = xfs_file_write_checks(iocb, from, &iolock);
704 trace_xfs_file_dax_write(iocb, from);
705 ret = dax_iomap_rw(iocb, from, &xfs_direct_write_iomap_ops);
706 if (ret > 0 && iocb->ki_pos > i_size_read(inode)) {
707 i_size_write(inode, iocb->ki_pos);
708 error = xfs_setfilesize(ip, pos, ret);
712 xfs_iunlock(ip, iolock);
717 XFS_STATS_ADD(ip->i_mount, xs_write_bytes, ret);
719 /* Handle various SYNC-type writes */
720 ret = generic_write_sync(iocb, ret);
726 xfs_file_buffered_write(
728 struct iov_iter *from)
730 struct file *file = iocb->ki_filp;
731 struct address_space *mapping = file->f_mapping;
732 struct inode *inode = mapping->host;
733 struct xfs_inode *ip = XFS_I(inode);
735 bool cleared_space = false;
738 if (iocb->ki_flags & IOCB_NOWAIT)
742 iolock = XFS_IOLOCK_EXCL;
743 xfs_ilock(ip, iolock);
745 ret = xfs_file_write_checks(iocb, from, &iolock);
749 /* We can write back this queue in page reclaim */
750 current->backing_dev_info = inode_to_bdi(inode);
752 trace_xfs_file_buffered_write(iocb, from);
753 ret = iomap_file_buffered_write(iocb, from,
754 &xfs_buffered_write_iomap_ops);
755 if (likely(ret >= 0))
759 * If we hit a space limit, try to free up some lingering preallocated
760 * space before returning an error. In the case of ENOSPC, first try to
761 * write back all dirty inodes to free up some of the excess reserved
762 * metadata space. This reduces the chances that the eofblocks scan
763 * waits on dirty mappings. Since xfs_flush_inodes() is serialized, this
764 * also behaves as a filter to prevent too many eofblocks scans from
765 * running at the same time. Use a synchronous scan to increase the
766 * effectiveness of the scan.
768 if (ret == -EDQUOT && !cleared_space) {
769 xfs_iunlock(ip, iolock);
770 xfs_blockgc_free_quota(ip, XFS_ICWALK_FLAG_SYNC);
771 cleared_space = true;
773 } else if (ret == -ENOSPC && !cleared_space) {
774 struct xfs_icwalk icw = {0};
776 cleared_space = true;
777 xfs_flush_inodes(ip->i_mount);
779 xfs_iunlock(ip, iolock);
780 icw.icw_flags = XFS_ICWALK_FLAG_SYNC;
781 xfs_blockgc_free_space(ip->i_mount, &icw);
785 current->backing_dev_info = NULL;
788 xfs_iunlock(ip, iolock);
791 XFS_STATS_ADD(ip->i_mount, xs_write_bytes, ret);
792 /* Handle various SYNC-type writes */
793 ret = generic_write_sync(iocb, ret);
801 struct iov_iter *from)
803 struct file *file = iocb->ki_filp;
804 struct address_space *mapping = file->f_mapping;
805 struct inode *inode = mapping->host;
806 struct xfs_inode *ip = XFS_I(inode);
808 size_t ocount = iov_iter_count(from);
810 XFS_STATS_INC(ip->i_mount, xs_write_calls);
815 if (xfs_is_shutdown(ip->i_mount))
819 return xfs_file_dax_write(iocb, from);
821 if (iocb->ki_flags & IOCB_DIRECT) {
823 * Allow a directio write to fall back to a buffered
824 * write *only* in the case that we're doing a reflink
825 * CoW. In all other directio scenarios we do not
826 * allow an operation to fall back to buffered mode.
828 ret = xfs_file_dio_write(iocb, from);
833 return xfs_file_buffered_write(iocb, from);
840 struct xfs_inode *ip = XFS_I(inode);
842 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
844 xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
848 xfs_break_dax_layouts(
854 ASSERT(xfs_isilocked(XFS_I(inode), XFS_MMAPLOCK_EXCL));
856 page = dax_layout_busy_page(inode->i_mapping);
861 return ___wait_var_event(&page->_refcount,
862 atomic_read(&page->_refcount) == 1, TASK_INTERRUPTIBLE,
863 0, 0, xfs_wait_dax_page(inode));
870 enum layout_break_reason reason)
875 ASSERT(xfs_isilocked(XFS_I(inode), XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL));
881 error = xfs_break_dax_layouts(inode, &retry);
886 error = xfs_break_leased_layouts(inode, iolock, &retry);
892 } while (error == 0 && retry);
897 #define XFS_FALLOC_FL_SUPPORTED \
898 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
899 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE | \
900 FALLOC_FL_INSERT_RANGE | FALLOC_FL_UNSHARE_RANGE)
909 struct inode *inode = file_inode(file);
910 struct xfs_inode *ip = XFS_I(inode);
912 uint iolock = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL;
914 bool do_file_insert = false;
916 if (!S_ISREG(inode->i_mode))
918 if (mode & ~XFS_FALLOC_FL_SUPPORTED)
921 xfs_ilock(ip, iolock);
922 error = xfs_break_layouts(inode, &iolock, BREAK_UNMAP);
927 * Must wait for all AIO to complete before we continue as AIO can
928 * change the file size on completion without holding any locks we
929 * currently hold. We must do this first because AIO can update both
930 * the on disk and in memory inode sizes, and the operations that follow
931 * require the in-memory size to be fully up-to-date.
933 inode_dio_wait(inode);
936 * Now AIO and DIO has drained we flush and (if necessary) invalidate
937 * the cached range over the first operation we are about to run.
939 * We care about zero and collapse here because they both run a hole
940 * punch over the range first. Because that can zero data, and the range
941 * of invalidation for the shift operations is much larger, we still do
942 * the required flush for collapse in xfs_prepare_shift().
944 * Insert has the same range requirements as collapse, and we extend the
945 * file first which can zero data. Hence insert has the same
946 * flush/invalidate requirements as collapse and so they are both
947 * handled at the right time by xfs_prepare_shift().
949 if (mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE |
950 FALLOC_FL_COLLAPSE_RANGE)) {
951 error = xfs_flush_unmap_range(ip, offset, len);
956 error = file_modified(file);
960 if (mode & FALLOC_FL_PUNCH_HOLE) {
961 error = xfs_free_file_space(ip, offset, len);
964 } else if (mode & FALLOC_FL_COLLAPSE_RANGE) {
965 if (!xfs_is_falloc_aligned(ip, offset, len)) {
971 * There is no need to overlap collapse range with EOF,
972 * in which case it is effectively a truncate operation
974 if (offset + len >= i_size_read(inode)) {
979 new_size = i_size_read(inode) - len;
981 error = xfs_collapse_file_space(ip, offset, len);
984 } else if (mode & FALLOC_FL_INSERT_RANGE) {
985 loff_t isize = i_size_read(inode);
987 if (!xfs_is_falloc_aligned(ip, offset, len)) {
993 * New inode size must not exceed ->s_maxbytes, accounting for
994 * possible signed overflow.
996 if (inode->i_sb->s_maxbytes - isize < len) {
1000 new_size = isize + len;
1002 /* Offset should be less than i_size */
1003 if (offset >= isize) {
1007 do_file_insert = true;
1009 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
1010 offset + len > i_size_read(inode)) {
1011 new_size = offset + len;
1012 error = inode_newsize_ok(inode, new_size);
1017 if (mode & FALLOC_FL_ZERO_RANGE) {
1019 * Punch a hole and prealloc the range. We use a hole
1020 * punch rather than unwritten extent conversion for two
1023 * 1.) Hole punch handles partial block zeroing for us.
1024 * 2.) If prealloc returns ENOSPC, the file range is
1025 * still zero-valued by virtue of the hole punch.
1027 unsigned int blksize = i_blocksize(inode);
1029 trace_xfs_zero_file_space(ip);
1031 error = xfs_free_file_space(ip, offset, len);
1035 len = round_up(offset + len, blksize) -
1036 round_down(offset, blksize);
1037 offset = round_down(offset, blksize);
1038 } else if (mode & FALLOC_FL_UNSHARE_RANGE) {
1039 error = xfs_reflink_unshare(ip, offset, len);
1044 * If always_cow mode we can't use preallocations and
1045 * thus should not create them.
1047 if (xfs_is_always_cow_inode(ip)) {
1048 error = -EOPNOTSUPP;
1053 if (!xfs_is_always_cow_inode(ip)) {
1054 error = xfs_alloc_file_space(ip, offset, len,
1055 XFS_BMAPI_PREALLOC);
1061 /* Change file size if needed */
1065 iattr.ia_valid = ATTR_SIZE;
1066 iattr.ia_size = new_size;
1067 error = xfs_vn_setattr_size(file_mnt_user_ns(file),
1068 file_dentry(file), &iattr);
1074 * Perform hole insertion now that the file size has been
1075 * updated so that if we crash during the operation we don't
1076 * leave shifted extents past EOF and hence losing access to
1077 * the data that is contained within them.
1079 if (do_file_insert) {
1080 error = xfs_insert_file_space(ip, offset, len);
1085 if (file->f_flags & O_DSYNC)
1086 error = xfs_log_force_inode(ip);
1089 xfs_iunlock(ip, iolock);
1100 struct xfs_inode *ip = XFS_I(file_inode(file));
1105 * Operations creating pages in page cache need protection from hole
1106 * punching and similar ops
1108 if (advice == POSIX_FADV_WILLNEED) {
1109 lockflags = XFS_IOLOCK_SHARED;
1110 xfs_ilock(ip, lockflags);
1112 ret = generic_fadvise(file, start, end, advice);
1114 xfs_iunlock(ip, lockflags);
1118 /* Does this file, inode, or mount want synchronous writes? */
1119 static inline bool xfs_file_sync_writes(struct file *filp)
1121 struct xfs_inode *ip = XFS_I(file_inode(filp));
1123 if (xfs_has_wsync(ip->i_mount))
1125 if (filp->f_flags & (__O_SYNC | O_DSYNC))
1127 if (IS_SYNC(file_inode(filp)))
1134 xfs_file_remap_range(
1135 struct file *file_in,
1137 struct file *file_out,
1140 unsigned int remap_flags)
1142 struct inode *inode_in = file_inode(file_in);
1143 struct xfs_inode *src = XFS_I(inode_in);
1144 struct inode *inode_out = file_inode(file_out);
1145 struct xfs_inode *dest = XFS_I(inode_out);
1146 struct xfs_mount *mp = src->i_mount;
1147 loff_t remapped = 0;
1148 xfs_extlen_t cowextsize;
1151 if (remap_flags & ~(REMAP_FILE_DEDUP | REMAP_FILE_ADVISORY))
1154 if (!xfs_has_reflink(mp))
1157 if (xfs_is_shutdown(mp))
1160 /* Prepare and then clone file data. */
1161 ret = xfs_reflink_remap_prep(file_in, pos_in, file_out, pos_out,
1163 if (ret || len == 0)
1166 trace_xfs_reflink_remap_range(src, pos_in, len, dest, pos_out);
1168 ret = xfs_reflink_remap_blocks(src, pos_in, dest, pos_out, len,
1174 * Carry the cowextsize hint from src to dest if we're sharing the
1175 * entire source file to the entire destination file, the source file
1176 * has a cowextsize hint, and the destination file does not.
1179 if (pos_in == 0 && len == i_size_read(inode_in) &&
1180 (src->i_diflags2 & XFS_DIFLAG2_COWEXTSIZE) &&
1181 pos_out == 0 && len >= i_size_read(inode_out) &&
1182 !(dest->i_diflags2 & XFS_DIFLAG2_COWEXTSIZE))
1183 cowextsize = src->i_cowextsize;
1185 ret = xfs_reflink_update_dest(dest, pos_out + len, cowextsize,
1190 if (xfs_file_sync_writes(file_in) || xfs_file_sync_writes(file_out))
1191 xfs_log_force_inode(dest);
1193 xfs_iunlock2_io_mmap(src, dest);
1195 trace_xfs_reflink_remap_range_error(dest, ret, _RET_IP_);
1196 return remapped > 0 ? remapped : ret;
1201 struct inode *inode,
1204 if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1206 if (xfs_is_shutdown(XFS_M(inode->i_sb)))
1208 file->f_mode |= FMODE_NOWAIT | FMODE_BUF_RASYNC;
1214 struct inode *inode,
1217 struct xfs_inode *ip = XFS_I(inode);
1221 error = xfs_file_open(inode, file);
1226 * If there are any blocks, read-ahead block 0 as we're almost
1227 * certain to have the next operation be a read there.
1229 mode = xfs_ilock_data_map_shared(ip);
1230 if (ip->i_df.if_nextents > 0)
1231 error = xfs_dir3_data_readahead(ip, 0, 0);
1232 xfs_iunlock(ip, mode);
1238 struct inode *inode,
1241 return xfs_release(XFS_I(inode));
1247 struct dir_context *ctx)
1249 struct inode *inode = file_inode(file);
1250 xfs_inode_t *ip = XFS_I(inode);
1254 * The Linux API doesn't pass down the total size of the buffer
1255 * we read into down to the filesystem. With the filldir concept
1256 * it's not needed for correct information, but the XFS dir2 leaf
1257 * code wants an estimate of the buffer size to calculate it's
1258 * readahead window and size the buffers used for mapping to
1261 * Try to give it an estimate that's good enough, maybe at some
1262 * point we can change the ->readdir prototype to include the
1263 * buffer size. For now we use the current glibc buffer size.
1265 bufsize = (size_t)min_t(loff_t, XFS_READDIR_BUFSIZE, ip->i_disk_size);
1267 return xfs_readdir(NULL, ip, ctx, bufsize);
1276 struct inode *inode = file->f_mapping->host;
1278 if (xfs_is_shutdown(XFS_I(inode)->i_mount))
1283 return generic_file_llseek(file, offset, whence);
1285 offset = iomap_seek_hole(inode, offset, &xfs_seek_iomap_ops);
1288 offset = iomap_seek_data(inode, offset, &xfs_seek_iomap_ops);
1294 return vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
1298 * Locking for serialisation of IO during page faults. This results in a lock
1302 * sb_start_pagefault(vfs, freeze)
1303 * invalidate_lock (vfs/XFS_MMAPLOCK - truncate serialisation)
1305 * i_lock (XFS - extent map serialisation)
1308 __xfs_filemap_fault(
1309 struct vm_fault *vmf,
1310 enum page_entry_size pe_size,
1313 struct inode *inode = file_inode(vmf->vma->vm_file);
1314 struct xfs_inode *ip = XFS_I(inode);
1317 trace_xfs_filemap_fault(ip, pe_size, write_fault);
1320 sb_start_pagefault(inode->i_sb);
1321 file_update_time(vmf->vma->vm_file);
1324 if (IS_DAX(inode)) {
1327 xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1328 ret = dax_iomap_fault(vmf, pe_size, &pfn, NULL,
1329 (write_fault && !vmf->cow_page) ?
1330 &xfs_direct_write_iomap_ops :
1331 &xfs_read_iomap_ops);
1332 if (ret & VM_FAULT_NEEDDSYNC)
1333 ret = dax_finish_sync_fault(vmf, pe_size, pfn);
1334 xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1337 xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1338 ret = iomap_page_mkwrite(vmf,
1339 &xfs_buffered_write_iomap_ops);
1340 xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1342 ret = filemap_fault(vmf);
1347 sb_end_pagefault(inode->i_sb);
1353 struct vm_fault *vmf)
1355 return (vmf->flags & FAULT_FLAG_WRITE) &&
1356 (vmf->vma->vm_flags & VM_SHARED);
1361 struct vm_fault *vmf)
1363 /* DAX can shortcut the normal fault path on write faults! */
1364 return __xfs_filemap_fault(vmf, PE_SIZE_PTE,
1365 IS_DAX(file_inode(vmf->vma->vm_file)) &&
1366 xfs_is_write_fault(vmf));
1370 xfs_filemap_huge_fault(
1371 struct vm_fault *vmf,
1372 enum page_entry_size pe_size)
1374 if (!IS_DAX(file_inode(vmf->vma->vm_file)))
1375 return VM_FAULT_FALLBACK;
1377 /* DAX can shortcut the normal fault path on write faults! */
1378 return __xfs_filemap_fault(vmf, pe_size,
1379 xfs_is_write_fault(vmf));
1383 xfs_filemap_page_mkwrite(
1384 struct vm_fault *vmf)
1386 return __xfs_filemap_fault(vmf, PE_SIZE_PTE, true);
1390 * pfn_mkwrite was originally intended to ensure we capture time stamp updates
1391 * on write faults. In reality, it needs to serialise against truncate and
1392 * prepare memory for writing so handle is as standard write fault.
1395 xfs_filemap_pfn_mkwrite(
1396 struct vm_fault *vmf)
1399 return __xfs_filemap_fault(vmf, PE_SIZE_PTE, true);
1403 xfs_filemap_map_pages(
1404 struct vm_fault *vmf,
1405 pgoff_t start_pgoff,
1408 struct inode *inode = file_inode(vmf->vma->vm_file);
1411 xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1412 ret = filemap_map_pages(vmf, start_pgoff, end_pgoff);
1413 xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1417 static const struct vm_operations_struct xfs_file_vm_ops = {
1418 .fault = xfs_filemap_fault,
1419 .huge_fault = xfs_filemap_huge_fault,
1420 .map_pages = xfs_filemap_map_pages,
1421 .page_mkwrite = xfs_filemap_page_mkwrite,
1422 .pfn_mkwrite = xfs_filemap_pfn_mkwrite,
1428 struct vm_area_struct *vma)
1430 struct inode *inode = file_inode(file);
1431 struct xfs_buftarg *target = xfs_inode_buftarg(XFS_I(inode));
1434 * We don't support synchronous mappings for non-DAX files and
1435 * for DAX files if underneath dax_device is not synchronous.
1437 if (!daxdev_mapping_supported(vma, target->bt_daxdev))
1440 file_accessed(file);
1441 vma->vm_ops = &xfs_file_vm_ops;
1443 vma->vm_flags |= VM_HUGEPAGE;
1447 const struct file_operations xfs_file_operations = {
1448 .llseek = xfs_file_llseek,
1449 .read_iter = xfs_file_read_iter,
1450 .write_iter = xfs_file_write_iter,
1451 .splice_read = generic_file_splice_read,
1452 .splice_write = iter_file_splice_write,
1453 .iopoll = iomap_dio_iopoll,
1454 .unlocked_ioctl = xfs_file_ioctl,
1455 #ifdef CONFIG_COMPAT
1456 .compat_ioctl = xfs_file_compat_ioctl,
1458 .mmap = xfs_file_mmap,
1459 .mmap_supported_flags = MAP_SYNC,
1460 .open = xfs_file_open,
1461 .release = xfs_file_release,
1462 .fsync = xfs_file_fsync,
1463 .get_unmapped_area = thp_get_unmapped_area,
1464 .fallocate = xfs_file_fallocate,
1465 .fadvise = xfs_file_fadvise,
1466 .remap_file_range = xfs_file_remap_range,
1469 const struct file_operations xfs_dir_file_operations = {
1470 .open = xfs_dir_open,
1471 .read = generic_read_dir,
1472 .iterate_shared = xfs_file_readdir,
1473 .llseek = generic_file_llseek,
1474 .unlocked_ioctl = xfs_file_ioctl,
1475 #ifdef CONFIG_COMPAT
1476 .compat_ioctl = xfs_file_compat_ioctl,
1478 .fsync = xfs_dir_fsync,