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 * Fsync operations on directories are much simpler than on regular files,
71 * as there is no file data to flush, and thus also no need for explicit
72 * cache flush operations, and there are no non-transaction metadata updates
73 * on directories either.
82 struct xfs_inode *ip = XFS_I(file->f_mapping->host);
84 trace_xfs_dir_fsync(ip);
85 return xfs_log_force_inode(ip);
93 if (!xfs_ipincount(ip))
95 if (datasync && !(ip->i_itemp->ili_fsync_fields & ~XFS_ILOG_TIMESTAMP))
97 return ip->i_itemp->ili_commit_seq;
101 * All metadata updates are logged, which means that we just have to flush the
102 * log up to the latest LSN that touched the inode.
104 * If we have concurrent fsync/fdatasync() calls, we need them to all block on
105 * the log force before we clear the ili_fsync_fields field. This ensures that
106 * we don't get a racing sync operation that does not wait for the metadata to
107 * hit the journal before returning. If we race with clearing ili_fsync_fields,
108 * then all that will happen is the log force will do nothing as the lsn will
109 * already be on disk. We can't race with setting ili_fsync_fields because that
110 * is done under XFS_ILOCK_EXCL, and that can't happen because we hold the lock
111 * shared until after the ili_fsync_fields is cleared.
115 struct xfs_inode *ip,
122 xfs_ilock(ip, XFS_ILOCK_SHARED);
123 seq = xfs_fsync_seq(ip, datasync);
125 error = xfs_log_force_seq(ip->i_mount, seq, XFS_LOG_SYNC,
128 spin_lock(&ip->i_itemp->ili_lock);
129 ip->i_itemp->ili_fsync_fields = 0;
130 spin_unlock(&ip->i_itemp->ili_lock);
132 xfs_iunlock(ip, XFS_ILOCK_SHARED);
143 struct xfs_inode *ip = XFS_I(file->f_mapping->host);
144 struct xfs_mount *mp = ip->i_mount;
148 trace_xfs_file_fsync(ip);
150 error = file_write_and_wait_range(file, start, end);
154 if (xfs_is_shutdown(mp))
157 xfs_iflags_clear(ip, XFS_ITRUNCATED);
160 * If we have an RT and/or log subvolume we need to make sure to flush
161 * the write cache the device used for file data first. This is to
162 * ensure newly written file data make it to disk before logging the new
163 * inode size in case of an extending write.
165 if (XFS_IS_REALTIME_INODE(ip))
166 blkdev_issue_flush(mp->m_rtdev_targp->bt_bdev);
167 else if (mp->m_logdev_targp != mp->m_ddev_targp)
168 blkdev_issue_flush(mp->m_ddev_targp->bt_bdev);
171 * Any inode that has dirty modifications in the log is pinned. The
172 * racy check here for a pinned inode while not catch modifications
173 * that happen concurrently to the fsync call, but fsync semantics
174 * only require to sync previously completed I/O.
176 if (xfs_ipincount(ip))
177 error = xfs_fsync_flush_log(ip, datasync, &log_flushed);
180 * If we only have a single device, and the log force about was
181 * a no-op we might have to flush the data device cache here.
182 * This can only happen for fdatasync/O_DSYNC if we were overwriting
183 * an already allocated file and thus do not have any metadata to
186 if (!log_flushed && !XFS_IS_REALTIME_INODE(ip) &&
187 mp->m_logdev_targp == mp->m_ddev_targp)
188 blkdev_issue_flush(mp->m_ddev_targp->bt_bdev);
196 unsigned int lock_mode)
198 struct xfs_inode *ip = XFS_I(file_inode(iocb->ki_filp));
200 if (iocb->ki_flags & IOCB_NOWAIT) {
201 if (!xfs_ilock_nowait(ip, lock_mode))
204 xfs_ilock(ip, lock_mode);
215 struct xfs_inode *ip = XFS_I(file_inode(iocb->ki_filp));
218 trace_xfs_file_direct_read(iocb, to);
220 if (!iov_iter_count(to))
221 return 0; /* skip atime */
223 file_accessed(iocb->ki_filp);
225 ret = xfs_ilock_iocb(iocb, XFS_IOLOCK_SHARED);
228 ret = iomap_dio_rw(iocb, to, &xfs_read_iomap_ops, NULL, 0, 0);
229 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
234 static noinline ssize_t
239 struct xfs_inode *ip = XFS_I(iocb->ki_filp->f_mapping->host);
242 trace_xfs_file_dax_read(iocb, to);
244 if (!iov_iter_count(to))
245 return 0; /* skip atime */
247 ret = xfs_ilock_iocb(iocb, XFS_IOLOCK_SHARED);
250 ret = dax_iomap_rw(iocb, to, &xfs_read_iomap_ops);
251 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
253 file_accessed(iocb->ki_filp);
258 xfs_file_buffered_read(
262 struct xfs_inode *ip = XFS_I(file_inode(iocb->ki_filp));
265 trace_xfs_file_buffered_read(iocb, to);
267 ret = xfs_ilock_iocb(iocb, XFS_IOLOCK_SHARED);
270 ret = generic_file_read_iter(iocb, to);
271 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
281 struct inode *inode = file_inode(iocb->ki_filp);
282 struct xfs_mount *mp = XFS_I(inode)->i_mount;
285 XFS_STATS_INC(mp, xs_read_calls);
287 if (xfs_is_shutdown(mp))
291 ret = xfs_file_dax_read(iocb, to);
292 else if (iocb->ki_flags & IOCB_DIRECT)
293 ret = xfs_file_dio_read(iocb, to);
295 ret = xfs_file_buffered_read(iocb, to);
298 XFS_STATS_ADD(mp, xs_read_bytes, ret);
303 * Common pre-write limit and setup checks.
305 * Called with the iolocked held either shared and exclusive according to
306 * @iolock, and returns with it held. Might upgrade the iolock to exclusive
307 * if called for a direct write beyond i_size.
310 xfs_file_write_checks(
312 struct iov_iter *from,
315 struct file *file = iocb->ki_filp;
316 struct inode *inode = file->f_mapping->host;
317 struct xfs_inode *ip = XFS_I(inode);
319 size_t count = iov_iter_count(from);
320 bool drained_dio = false;
324 error = generic_write_checks(iocb, from);
328 if (iocb->ki_flags & IOCB_NOWAIT) {
329 error = break_layout(inode, false);
330 if (error == -EWOULDBLOCK)
333 error = xfs_break_layouts(inode, iolock, BREAK_WRITE);
340 * For changing security info in file_remove_privs() we need i_rwsem
343 if (*iolock == XFS_IOLOCK_SHARED && !IS_NOSEC(inode)) {
344 xfs_iunlock(ip, *iolock);
345 *iolock = XFS_IOLOCK_EXCL;
346 error = xfs_ilock_iocb(iocb, *iolock);
355 * If the offset is beyond the size of the file, we need to zero any
356 * blocks that fall between the existing EOF and the start of this
357 * write. If zeroing is needed and we are currently holding the iolock
358 * shared, we need to update it to exclusive which implies having to
359 * redo all checks before.
361 * We need to serialise against EOF updates that occur in IO completions
362 * here. We want to make sure that nobody is changing the size while we
363 * do this check until we have placed an IO barrier (i.e. hold the
364 * XFS_IOLOCK_EXCL) that prevents new IO from being dispatched. The
365 * spinlock effectively forms a memory barrier once we have the
366 * XFS_IOLOCK_EXCL so we are guaranteed to see the latest EOF value and
367 * hence be able to correctly determine if we need to run zeroing.
369 * We can do an unlocked check here safely as IO completion can only
370 * extend EOF. Truncate is locked out at this point, so the EOF can
371 * not move backwards, only forwards. Hence we only need to take the
372 * slow path and spin locks when we are at or beyond the current EOF.
374 if (iocb->ki_pos <= i_size_read(inode))
377 spin_lock(&ip->i_flags_lock);
378 isize = i_size_read(inode);
379 if (iocb->ki_pos > isize) {
380 spin_unlock(&ip->i_flags_lock);
382 if (iocb->ki_flags & IOCB_NOWAIT)
386 if (*iolock == XFS_IOLOCK_SHARED) {
387 xfs_iunlock(ip, *iolock);
388 *iolock = XFS_IOLOCK_EXCL;
389 xfs_ilock(ip, *iolock);
390 iov_iter_reexpand(from, count);
393 * We now have an IO submission barrier in place, but
394 * AIO can do EOF updates during IO completion and hence
395 * we now need to wait for all of them to drain. Non-AIO
396 * DIO will have drained before we are given the
397 * XFS_IOLOCK_EXCL, and so for most cases this wait is a
400 inode_dio_wait(inode);
405 trace_xfs_zero_eof(ip, isize, iocb->ki_pos - isize);
406 error = xfs_zero_range(ip, isize, iocb->ki_pos - isize, NULL);
410 spin_unlock(&ip->i_flags_lock);
413 return file_modified(file);
417 xfs_dio_write_end_io(
423 struct inode *inode = file_inode(iocb->ki_filp);
424 struct xfs_inode *ip = XFS_I(inode);
425 loff_t offset = iocb->ki_pos;
426 unsigned int nofs_flag;
428 trace_xfs_end_io_direct_write(ip, offset, size);
430 if (xfs_is_shutdown(ip->i_mount))
439 * Capture amount written on completion as we can't reliably account
440 * for it on submission.
442 XFS_STATS_ADD(ip->i_mount, xs_write_bytes, size);
445 * We can allocate memory here while doing writeback on behalf of
446 * memory reclaim. To avoid memory allocation deadlocks set the
447 * task-wide nofs context for the following operations.
449 nofs_flag = memalloc_nofs_save();
451 if (flags & IOMAP_DIO_COW) {
452 error = xfs_reflink_end_cow(ip, offset, size);
458 * Unwritten conversion updates the in-core isize after extent
459 * conversion but before updating the on-disk size. Updating isize any
460 * earlier allows a racing dio read to find unwritten extents before
461 * they are converted.
463 if (flags & IOMAP_DIO_UNWRITTEN) {
464 error = xfs_iomap_write_unwritten(ip, offset, size, true);
469 * We need to update the in-core inode size here so that we don't end up
470 * with the on-disk inode size being outside the in-core inode size. We
471 * have no other method of updating EOF for AIO, so always do it here
474 * We need to lock the test/set EOF update as we can be racing with
475 * other IO completions here to update the EOF. Failing to serialise
476 * here can result in EOF moving backwards and Bad Things Happen when
479 * As IO completion only ever extends EOF, we can do an unlocked check
480 * here to avoid taking the spinlock. If we land within the current EOF,
481 * then we do not need to do an extending update at all, and we don't
482 * need to take the lock to check this. If we race with an update moving
483 * EOF, then we'll either still be beyond EOF and need to take the lock,
484 * or we'll be within EOF and we don't need to take it at all.
486 if (offset + size <= i_size_read(inode))
489 spin_lock(&ip->i_flags_lock);
490 if (offset + size > i_size_read(inode)) {
491 i_size_write(inode, offset + size);
492 spin_unlock(&ip->i_flags_lock);
493 error = xfs_setfilesize(ip, offset, size);
495 spin_unlock(&ip->i_flags_lock);
499 memalloc_nofs_restore(nofs_flag);
503 static const struct iomap_dio_ops xfs_dio_write_ops = {
504 .end_io = xfs_dio_write_end_io,
508 * Handle block aligned direct I/O writes
510 static noinline ssize_t
511 xfs_file_dio_write_aligned(
512 struct xfs_inode *ip,
514 struct iov_iter *from)
516 int iolock = XFS_IOLOCK_SHARED;
519 ret = xfs_ilock_iocb(iocb, iolock);
522 ret = xfs_file_write_checks(iocb, from, &iolock);
527 * We don't need to hold the IOLOCK exclusively across the IO, so demote
528 * the iolock back to shared if we had to take the exclusive lock in
529 * xfs_file_write_checks() for other reasons.
531 if (iolock == XFS_IOLOCK_EXCL) {
532 xfs_ilock_demote(ip, XFS_IOLOCK_EXCL);
533 iolock = XFS_IOLOCK_SHARED;
535 trace_xfs_file_direct_write(iocb, from);
536 ret = iomap_dio_rw(iocb, from, &xfs_direct_write_iomap_ops,
537 &xfs_dio_write_ops, 0, 0);
540 xfs_iunlock(ip, iolock);
545 * Handle block unaligned direct I/O writes
547 * In most cases direct I/O writes will be done holding IOLOCK_SHARED, allowing
548 * them to be done in parallel with reads and other direct I/O writes. However,
549 * if the I/O is not aligned to filesystem blocks, the direct I/O layer may need
550 * to do sub-block zeroing and that requires serialisation against other direct
551 * I/O to the same block. In this case we need to serialise the submission of
552 * the unaligned I/O so that we don't get racing block zeroing in the dio layer.
553 * In the case where sub-block zeroing is not required, we can do concurrent
554 * sub-block dios to the same block successfully.
556 * Optimistically submit the I/O using the shared lock first, but use the
557 * IOMAP_DIO_OVERWRITE_ONLY flag to tell the lower layers to return -EAGAIN
558 * if block allocation or partial block zeroing would be required. In that case
559 * we try again with the exclusive lock.
561 static noinline ssize_t
562 xfs_file_dio_write_unaligned(
563 struct xfs_inode *ip,
565 struct iov_iter *from)
567 size_t isize = i_size_read(VFS_I(ip));
568 size_t count = iov_iter_count(from);
569 int iolock = XFS_IOLOCK_SHARED;
570 unsigned int flags = IOMAP_DIO_OVERWRITE_ONLY;
574 * Extending writes need exclusivity because of the sub-block zeroing
575 * that the DIO code always does for partial tail blocks beyond EOF, so
576 * don't even bother trying the fast path in this case.
578 if (iocb->ki_pos > isize || iocb->ki_pos + count >= isize) {
580 if (iocb->ki_flags & IOCB_NOWAIT)
582 iolock = XFS_IOLOCK_EXCL;
583 flags = IOMAP_DIO_FORCE_WAIT;
586 ret = xfs_ilock_iocb(iocb, iolock);
591 * We can't properly handle unaligned direct I/O to reflink files yet,
592 * as we can't unshare a partial block.
594 if (xfs_is_cow_inode(ip)) {
595 trace_xfs_reflink_bounce_dio_write(iocb, from);
600 ret = xfs_file_write_checks(iocb, from, &iolock);
605 * If we are doing exclusive unaligned I/O, this must be the only I/O
606 * in-flight. Otherwise we risk data corruption due to unwritten extent
607 * conversions from the AIO end_io handler. Wait for all other I/O to
610 if (flags & IOMAP_DIO_FORCE_WAIT)
611 inode_dio_wait(VFS_I(ip));
613 trace_xfs_file_direct_write(iocb, from);
614 ret = iomap_dio_rw(iocb, from, &xfs_direct_write_iomap_ops,
615 &xfs_dio_write_ops, flags, 0);
618 * Retry unaligned I/O with exclusive blocking semantics if the DIO
619 * layer rejected it for mapping or locking reasons. If we are doing
620 * nonblocking user I/O, propagate the error.
622 if (ret == -EAGAIN && !(iocb->ki_flags & IOCB_NOWAIT)) {
623 ASSERT(flags & IOMAP_DIO_OVERWRITE_ONLY);
624 xfs_iunlock(ip, iolock);
625 goto retry_exclusive;
630 xfs_iunlock(ip, iolock);
637 struct iov_iter *from)
639 struct xfs_inode *ip = XFS_I(file_inode(iocb->ki_filp));
640 struct xfs_buftarg *target = xfs_inode_buftarg(ip);
641 size_t count = iov_iter_count(from);
643 /* direct I/O must be aligned to device logical sector size */
644 if ((iocb->ki_pos | count) & target->bt_logical_sectormask)
646 if ((iocb->ki_pos | count) & ip->i_mount->m_blockmask)
647 return xfs_file_dio_write_unaligned(ip, iocb, from);
648 return xfs_file_dio_write_aligned(ip, iocb, from);
651 static noinline ssize_t
654 struct iov_iter *from)
656 struct inode *inode = iocb->ki_filp->f_mapping->host;
657 struct xfs_inode *ip = XFS_I(inode);
658 int iolock = XFS_IOLOCK_EXCL;
659 ssize_t ret, error = 0;
662 ret = xfs_ilock_iocb(iocb, iolock);
665 ret = xfs_file_write_checks(iocb, from, &iolock);
671 trace_xfs_file_dax_write(iocb, from);
672 ret = dax_iomap_rw(iocb, from, &xfs_direct_write_iomap_ops);
673 if (ret > 0 && iocb->ki_pos > i_size_read(inode)) {
674 i_size_write(inode, iocb->ki_pos);
675 error = xfs_setfilesize(ip, pos, ret);
679 xfs_iunlock(ip, iolock);
684 XFS_STATS_ADD(ip->i_mount, xs_write_bytes, ret);
686 /* Handle various SYNC-type writes */
687 ret = generic_write_sync(iocb, ret);
693 xfs_file_buffered_write(
695 struct iov_iter *from)
697 struct file *file = iocb->ki_filp;
698 struct address_space *mapping = file->f_mapping;
699 struct inode *inode = mapping->host;
700 struct xfs_inode *ip = XFS_I(inode);
702 bool cleared_space = false;
705 if (iocb->ki_flags & IOCB_NOWAIT)
709 iolock = XFS_IOLOCK_EXCL;
710 xfs_ilock(ip, iolock);
712 ret = xfs_file_write_checks(iocb, from, &iolock);
716 /* We can write back this queue in page reclaim */
717 current->backing_dev_info = inode_to_bdi(inode);
719 trace_xfs_file_buffered_write(iocb, from);
720 ret = iomap_file_buffered_write(iocb, from,
721 &xfs_buffered_write_iomap_ops);
722 if (likely(ret >= 0))
726 * If we hit a space limit, try to free up some lingering preallocated
727 * space before returning an error. In the case of ENOSPC, first try to
728 * write back all dirty inodes to free up some of the excess reserved
729 * metadata space. This reduces the chances that the eofblocks scan
730 * waits on dirty mappings. Since xfs_flush_inodes() is serialized, this
731 * also behaves as a filter to prevent too many eofblocks scans from
732 * running at the same time. Use a synchronous scan to increase the
733 * effectiveness of the scan.
735 if (ret == -EDQUOT && !cleared_space) {
736 xfs_iunlock(ip, iolock);
737 xfs_blockgc_free_quota(ip, XFS_ICWALK_FLAG_SYNC);
738 cleared_space = true;
740 } else if (ret == -ENOSPC && !cleared_space) {
741 struct xfs_icwalk icw = {0};
743 cleared_space = true;
744 xfs_flush_inodes(ip->i_mount);
746 xfs_iunlock(ip, iolock);
747 icw.icw_flags = XFS_ICWALK_FLAG_SYNC;
748 xfs_blockgc_free_space(ip->i_mount, &icw);
752 current->backing_dev_info = NULL;
755 xfs_iunlock(ip, iolock);
758 XFS_STATS_ADD(ip->i_mount, xs_write_bytes, ret);
759 /* Handle various SYNC-type writes */
760 ret = generic_write_sync(iocb, ret);
768 struct iov_iter *from)
770 struct file *file = iocb->ki_filp;
771 struct address_space *mapping = file->f_mapping;
772 struct inode *inode = mapping->host;
773 struct xfs_inode *ip = XFS_I(inode);
775 size_t ocount = iov_iter_count(from);
777 XFS_STATS_INC(ip->i_mount, xs_write_calls);
782 if (xfs_is_shutdown(ip->i_mount))
786 return xfs_file_dax_write(iocb, from);
788 if (iocb->ki_flags & IOCB_DIRECT) {
790 * Allow a directio write to fall back to a buffered
791 * write *only* in the case that we're doing a reflink
792 * CoW. In all other directio scenarios we do not
793 * allow an operation to fall back to buffered mode.
795 ret = xfs_file_dio_write(iocb, from);
800 return xfs_file_buffered_write(iocb, from);
807 struct xfs_inode *ip = XFS_I(inode);
809 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
811 xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
815 xfs_break_dax_layouts(
821 ASSERT(xfs_isilocked(XFS_I(inode), XFS_MMAPLOCK_EXCL));
823 page = dax_layout_busy_page(inode->i_mapping);
828 return ___wait_var_event(&page->_refcount,
829 atomic_read(&page->_refcount) == 1, TASK_INTERRUPTIBLE,
830 0, 0, xfs_wait_dax_page(inode));
837 enum layout_break_reason reason)
842 ASSERT(xfs_isilocked(XFS_I(inode), XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL));
848 error = xfs_break_dax_layouts(inode, &retry);
853 error = xfs_break_leased_layouts(inode, iolock, &retry);
859 } while (error == 0 && retry);
864 /* Does this file, inode, or mount want synchronous writes? */
865 static inline bool xfs_file_sync_writes(struct file *filp)
867 struct xfs_inode *ip = XFS_I(file_inode(filp));
869 if (xfs_has_wsync(ip->i_mount))
871 if (filp->f_flags & (__O_SYNC | O_DSYNC))
873 if (IS_SYNC(file_inode(filp)))
879 #define XFS_FALLOC_FL_SUPPORTED \
880 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
881 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE | \
882 FALLOC_FL_INSERT_RANGE | FALLOC_FL_UNSHARE_RANGE)
891 struct inode *inode = file_inode(file);
892 struct xfs_inode *ip = XFS_I(inode);
894 uint iolock = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL;
896 bool do_file_insert = false;
898 if (!S_ISREG(inode->i_mode))
900 if (mode & ~XFS_FALLOC_FL_SUPPORTED)
903 xfs_ilock(ip, iolock);
904 error = xfs_break_layouts(inode, &iolock, BREAK_UNMAP);
909 * Must wait for all AIO to complete before we continue as AIO can
910 * change the file size on completion without holding any locks we
911 * currently hold. We must do this first because AIO can update both
912 * the on disk and in memory inode sizes, and the operations that follow
913 * require the in-memory size to be fully up-to-date.
915 inode_dio_wait(inode);
918 * Now AIO and DIO has drained we flush and (if necessary) invalidate
919 * the cached range over the first operation we are about to run.
921 * We care about zero and collapse here because they both run a hole
922 * punch over the range first. Because that can zero data, and the range
923 * of invalidation for the shift operations is much larger, we still do
924 * the required flush for collapse in xfs_prepare_shift().
926 * Insert has the same range requirements as collapse, and we extend the
927 * file first which can zero data. Hence insert has the same
928 * flush/invalidate requirements as collapse and so they are both
929 * handled at the right time by xfs_prepare_shift().
931 if (mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE |
932 FALLOC_FL_COLLAPSE_RANGE)) {
933 error = xfs_flush_unmap_range(ip, offset, len);
938 error = file_modified(file);
942 if (mode & FALLOC_FL_PUNCH_HOLE) {
943 error = xfs_free_file_space(ip, offset, len);
946 } else if (mode & FALLOC_FL_COLLAPSE_RANGE) {
947 if (!xfs_is_falloc_aligned(ip, offset, len)) {
953 * There is no need to overlap collapse range with EOF,
954 * in which case it is effectively a truncate operation
956 if (offset + len >= i_size_read(inode)) {
961 new_size = i_size_read(inode) - len;
963 error = xfs_collapse_file_space(ip, offset, len);
966 } else if (mode & FALLOC_FL_INSERT_RANGE) {
967 loff_t isize = i_size_read(inode);
969 if (!xfs_is_falloc_aligned(ip, offset, len)) {
975 * New inode size must not exceed ->s_maxbytes, accounting for
976 * possible signed overflow.
978 if (inode->i_sb->s_maxbytes - isize < len) {
982 new_size = isize + len;
984 /* Offset should be less than i_size */
985 if (offset >= isize) {
989 do_file_insert = true;
991 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
992 offset + len > i_size_read(inode)) {
993 new_size = offset + len;
994 error = inode_newsize_ok(inode, new_size);
999 if (mode & FALLOC_FL_ZERO_RANGE) {
1001 * Punch a hole and prealloc the range. We use a hole
1002 * punch rather than unwritten extent conversion for two
1005 * 1.) Hole punch handles partial block zeroing for us.
1006 * 2.) If prealloc returns ENOSPC, the file range is
1007 * still zero-valued by virtue of the hole punch.
1009 unsigned int blksize = i_blocksize(inode);
1011 trace_xfs_zero_file_space(ip);
1013 error = xfs_free_file_space(ip, offset, len);
1017 len = round_up(offset + len, blksize) -
1018 round_down(offset, blksize);
1019 offset = round_down(offset, blksize);
1020 } else if (mode & FALLOC_FL_UNSHARE_RANGE) {
1021 error = xfs_reflink_unshare(ip, offset, len);
1026 * If always_cow mode we can't use preallocations and
1027 * thus should not create them.
1029 if (xfs_is_always_cow_inode(ip)) {
1030 error = -EOPNOTSUPP;
1035 if (!xfs_is_always_cow_inode(ip)) {
1036 error = xfs_alloc_file_space(ip, offset, len);
1042 /* Change file size if needed */
1046 iattr.ia_valid = ATTR_SIZE;
1047 iattr.ia_size = new_size;
1048 error = xfs_vn_setattr_size(file_mnt_user_ns(file),
1049 file_dentry(file), &iattr);
1055 * Perform hole insertion now that the file size has been
1056 * updated so that if we crash during the operation we don't
1057 * leave shifted extents past EOF and hence losing access to
1058 * the data that is contained within them.
1060 if (do_file_insert) {
1061 error = xfs_insert_file_space(ip, offset, len);
1066 if (xfs_file_sync_writes(file))
1067 error = xfs_log_force_inode(ip);
1070 xfs_iunlock(ip, iolock);
1081 struct xfs_inode *ip = XFS_I(file_inode(file));
1086 * Operations creating pages in page cache need protection from hole
1087 * punching and similar ops
1089 if (advice == POSIX_FADV_WILLNEED) {
1090 lockflags = XFS_IOLOCK_SHARED;
1091 xfs_ilock(ip, lockflags);
1093 ret = generic_fadvise(file, start, end, advice);
1095 xfs_iunlock(ip, lockflags);
1100 xfs_file_remap_range(
1101 struct file *file_in,
1103 struct file *file_out,
1106 unsigned int remap_flags)
1108 struct inode *inode_in = file_inode(file_in);
1109 struct xfs_inode *src = XFS_I(inode_in);
1110 struct inode *inode_out = file_inode(file_out);
1111 struct xfs_inode *dest = XFS_I(inode_out);
1112 struct xfs_mount *mp = src->i_mount;
1113 loff_t remapped = 0;
1114 xfs_extlen_t cowextsize;
1117 if (remap_flags & ~(REMAP_FILE_DEDUP | REMAP_FILE_ADVISORY))
1120 if (!xfs_has_reflink(mp))
1123 if (xfs_is_shutdown(mp))
1126 /* Prepare and then clone file data. */
1127 ret = xfs_reflink_remap_prep(file_in, pos_in, file_out, pos_out,
1129 if (ret || len == 0)
1132 trace_xfs_reflink_remap_range(src, pos_in, len, dest, pos_out);
1134 ret = xfs_reflink_remap_blocks(src, pos_in, dest, pos_out, len,
1140 * Carry the cowextsize hint from src to dest if we're sharing the
1141 * entire source file to the entire destination file, the source file
1142 * has a cowextsize hint, and the destination file does not.
1145 if (pos_in == 0 && len == i_size_read(inode_in) &&
1146 (src->i_diflags2 & XFS_DIFLAG2_COWEXTSIZE) &&
1147 pos_out == 0 && len >= i_size_read(inode_out) &&
1148 !(dest->i_diflags2 & XFS_DIFLAG2_COWEXTSIZE))
1149 cowextsize = src->i_cowextsize;
1151 ret = xfs_reflink_update_dest(dest, pos_out + len, cowextsize,
1156 if (xfs_file_sync_writes(file_in) || xfs_file_sync_writes(file_out))
1157 xfs_log_force_inode(dest);
1159 xfs_iunlock2_io_mmap(src, dest);
1161 trace_xfs_reflink_remap_range_error(dest, ret, _RET_IP_);
1162 return remapped > 0 ? remapped : ret;
1167 struct inode *inode,
1170 if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1172 if (xfs_is_shutdown(XFS_M(inode->i_sb)))
1174 file->f_mode |= FMODE_NOWAIT | FMODE_BUF_RASYNC;
1180 struct inode *inode,
1183 struct xfs_inode *ip = XFS_I(inode);
1187 error = xfs_file_open(inode, file);
1192 * If there are any blocks, read-ahead block 0 as we're almost
1193 * certain to have the next operation be a read there.
1195 mode = xfs_ilock_data_map_shared(ip);
1196 if (ip->i_df.if_nextents > 0)
1197 error = xfs_dir3_data_readahead(ip, 0, 0);
1198 xfs_iunlock(ip, mode);
1204 struct inode *inode,
1207 return xfs_release(XFS_I(inode));
1213 struct dir_context *ctx)
1215 struct inode *inode = file_inode(file);
1216 xfs_inode_t *ip = XFS_I(inode);
1220 * The Linux API doesn't pass down the total size of the buffer
1221 * we read into down to the filesystem. With the filldir concept
1222 * it's not needed for correct information, but the XFS dir2 leaf
1223 * code wants an estimate of the buffer size to calculate it's
1224 * readahead window and size the buffers used for mapping to
1227 * Try to give it an estimate that's good enough, maybe at some
1228 * point we can change the ->readdir prototype to include the
1229 * buffer size. For now we use the current glibc buffer size.
1231 bufsize = (size_t)min_t(loff_t, XFS_READDIR_BUFSIZE, ip->i_disk_size);
1233 return xfs_readdir(NULL, ip, ctx, bufsize);
1242 struct inode *inode = file->f_mapping->host;
1244 if (xfs_is_shutdown(XFS_I(inode)->i_mount))
1249 return generic_file_llseek(file, offset, whence);
1251 offset = iomap_seek_hole(inode, offset, &xfs_seek_iomap_ops);
1254 offset = iomap_seek_data(inode, offset, &xfs_seek_iomap_ops);
1260 return vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
1264 * Locking for serialisation of IO during page faults. This results in a lock
1268 * sb_start_pagefault(vfs, freeze)
1269 * invalidate_lock (vfs/XFS_MMAPLOCK - truncate serialisation)
1271 * i_lock (XFS - extent map serialisation)
1274 __xfs_filemap_fault(
1275 struct vm_fault *vmf,
1276 enum page_entry_size pe_size,
1279 struct inode *inode = file_inode(vmf->vma->vm_file);
1280 struct xfs_inode *ip = XFS_I(inode);
1283 trace_xfs_filemap_fault(ip, pe_size, write_fault);
1286 sb_start_pagefault(inode->i_sb);
1287 file_update_time(vmf->vma->vm_file);
1290 if (IS_DAX(inode)) {
1293 xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1294 ret = dax_iomap_fault(vmf, pe_size, &pfn, NULL,
1295 (write_fault && !vmf->cow_page) ?
1296 &xfs_direct_write_iomap_ops :
1297 &xfs_read_iomap_ops);
1298 if (ret & VM_FAULT_NEEDDSYNC)
1299 ret = dax_finish_sync_fault(vmf, pe_size, pfn);
1300 xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1303 xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1304 ret = iomap_page_mkwrite(vmf,
1305 &xfs_buffered_write_iomap_ops);
1306 xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1308 ret = filemap_fault(vmf);
1313 sb_end_pagefault(inode->i_sb);
1319 struct vm_fault *vmf)
1321 return (vmf->flags & FAULT_FLAG_WRITE) &&
1322 (vmf->vma->vm_flags & VM_SHARED);
1327 struct vm_fault *vmf)
1329 /* DAX can shortcut the normal fault path on write faults! */
1330 return __xfs_filemap_fault(vmf, PE_SIZE_PTE,
1331 IS_DAX(file_inode(vmf->vma->vm_file)) &&
1332 xfs_is_write_fault(vmf));
1336 xfs_filemap_huge_fault(
1337 struct vm_fault *vmf,
1338 enum page_entry_size pe_size)
1340 if (!IS_DAX(file_inode(vmf->vma->vm_file)))
1341 return VM_FAULT_FALLBACK;
1343 /* DAX can shortcut the normal fault path on write faults! */
1344 return __xfs_filemap_fault(vmf, pe_size,
1345 xfs_is_write_fault(vmf));
1349 xfs_filemap_page_mkwrite(
1350 struct vm_fault *vmf)
1352 return __xfs_filemap_fault(vmf, PE_SIZE_PTE, true);
1356 * pfn_mkwrite was originally intended to ensure we capture time stamp updates
1357 * on write faults. In reality, it needs to serialise against truncate and
1358 * prepare memory for writing so handle is as standard write fault.
1361 xfs_filemap_pfn_mkwrite(
1362 struct vm_fault *vmf)
1365 return __xfs_filemap_fault(vmf, PE_SIZE_PTE, true);
1369 xfs_filemap_map_pages(
1370 struct vm_fault *vmf,
1371 pgoff_t start_pgoff,
1374 struct inode *inode = file_inode(vmf->vma->vm_file);
1377 xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1378 ret = filemap_map_pages(vmf, start_pgoff, end_pgoff);
1379 xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1383 static const struct vm_operations_struct xfs_file_vm_ops = {
1384 .fault = xfs_filemap_fault,
1385 .huge_fault = xfs_filemap_huge_fault,
1386 .map_pages = xfs_filemap_map_pages,
1387 .page_mkwrite = xfs_filemap_page_mkwrite,
1388 .pfn_mkwrite = xfs_filemap_pfn_mkwrite,
1394 struct vm_area_struct *vma)
1396 struct inode *inode = file_inode(file);
1397 struct xfs_buftarg *target = xfs_inode_buftarg(XFS_I(inode));
1400 * We don't support synchronous mappings for non-DAX files and
1401 * for DAX files if underneath dax_device is not synchronous.
1403 if (!daxdev_mapping_supported(vma, target->bt_daxdev))
1406 file_accessed(file);
1407 vma->vm_ops = &xfs_file_vm_ops;
1409 vma->vm_flags |= VM_HUGEPAGE;
1413 const struct file_operations xfs_file_operations = {
1414 .llseek = xfs_file_llseek,
1415 .read_iter = xfs_file_read_iter,
1416 .write_iter = xfs_file_write_iter,
1417 .splice_read = generic_file_splice_read,
1418 .splice_write = iter_file_splice_write,
1419 .iopoll = iocb_bio_iopoll,
1420 .unlocked_ioctl = xfs_file_ioctl,
1421 #ifdef CONFIG_COMPAT
1422 .compat_ioctl = xfs_file_compat_ioctl,
1424 .mmap = xfs_file_mmap,
1425 .mmap_supported_flags = MAP_SYNC,
1426 .open = xfs_file_open,
1427 .release = xfs_file_release,
1428 .fsync = xfs_file_fsync,
1429 .get_unmapped_area = thp_get_unmapped_area,
1430 .fallocate = xfs_file_fallocate,
1431 .fadvise = xfs_file_fadvise,
1432 .remap_file_range = xfs_file_remap_range,
1435 const struct file_operations xfs_dir_file_operations = {
1436 .open = xfs_dir_open,
1437 .read = generic_read_dir,
1438 .iterate_shared = xfs_file_readdir,
1439 .llseek = generic_file_llseek,
1440 .unlocked_ioctl = xfs_file_ioctl,
1441 #ifdef CONFIG_COMPAT
1442 .compat_ioctl = xfs_file_compat_ioctl,
1444 .fsync = xfs_dir_fsync,