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 if (flags & XFS_PREALLOC_SYNC)
99 xfs_trans_set_sync(tp);
100 return xfs_trans_commit(tp);
104 * Fsync operations on directories are much simpler than on regular files,
105 * as there is no file data to flush, and thus also no need for explicit
106 * cache flush operations, and there are no non-transaction metadata updates
107 * on directories either.
116 struct xfs_inode *ip = XFS_I(file->f_mapping->host);
118 trace_xfs_dir_fsync(ip);
119 return xfs_log_force_inode(ip);
124 struct xfs_inode *ip,
127 if (!xfs_ipincount(ip))
129 if (datasync && !(ip->i_itemp->ili_fsync_fields & ~XFS_ILOG_TIMESTAMP))
131 return ip->i_itemp->ili_commit_seq;
135 * All metadata updates are logged, which means that we just have to flush the
136 * log up to the latest LSN that touched the inode.
138 * If we have concurrent fsync/fdatasync() calls, we need them to all block on
139 * the log force before we clear the ili_fsync_fields field. This ensures that
140 * we don't get a racing sync operation that does not wait for the metadata to
141 * hit the journal before returning. If we race with clearing ili_fsync_fields,
142 * then all that will happen is the log force will do nothing as the lsn will
143 * already be on disk. We can't race with setting ili_fsync_fields because that
144 * is done under XFS_ILOCK_EXCL, and that can't happen because we hold the lock
145 * shared until after the ili_fsync_fields is cleared.
149 struct xfs_inode *ip,
156 xfs_ilock(ip, XFS_ILOCK_SHARED);
157 seq = xfs_fsync_seq(ip, datasync);
159 error = xfs_log_force_seq(ip->i_mount, seq, XFS_LOG_SYNC,
162 spin_lock(&ip->i_itemp->ili_lock);
163 ip->i_itemp->ili_fsync_fields = 0;
164 spin_unlock(&ip->i_itemp->ili_lock);
166 xfs_iunlock(ip, XFS_ILOCK_SHARED);
177 struct xfs_inode *ip = XFS_I(file->f_mapping->host);
178 struct xfs_mount *mp = ip->i_mount;
182 trace_xfs_file_fsync(ip);
184 error = file_write_and_wait_range(file, start, end);
188 if (xfs_is_shutdown(mp))
191 xfs_iflags_clear(ip, XFS_ITRUNCATED);
194 * If we have an RT and/or log subvolume we need to make sure to flush
195 * the write cache the device used for file data first. This is to
196 * ensure newly written file data make it to disk before logging the new
197 * inode size in case of an extending write.
199 if (XFS_IS_REALTIME_INODE(ip))
200 blkdev_issue_flush(mp->m_rtdev_targp->bt_bdev);
201 else if (mp->m_logdev_targp != mp->m_ddev_targp)
202 blkdev_issue_flush(mp->m_ddev_targp->bt_bdev);
205 * Any inode that has dirty modifications in the log is pinned. The
206 * racy check here for a pinned inode while not catch modifications
207 * that happen concurrently to the fsync call, but fsync semantics
208 * only require to sync previously completed I/O.
210 if (xfs_ipincount(ip))
211 error = xfs_fsync_flush_log(ip, datasync, &log_flushed);
214 * If we only have a single device, and the log force about was
215 * a no-op we might have to flush the data device cache here.
216 * This can only happen for fdatasync/O_DSYNC if we were overwriting
217 * an already allocated file and thus do not have any metadata to
220 if (!log_flushed && !XFS_IS_REALTIME_INODE(ip) &&
221 mp->m_logdev_targp == mp->m_ddev_targp)
222 blkdev_issue_flush(mp->m_ddev_targp->bt_bdev);
230 unsigned int lock_mode)
232 struct xfs_inode *ip = XFS_I(file_inode(iocb->ki_filp));
234 if (iocb->ki_flags & IOCB_NOWAIT) {
235 if (!xfs_ilock_nowait(ip, lock_mode))
238 xfs_ilock(ip, lock_mode);
249 struct xfs_inode *ip = XFS_I(file_inode(iocb->ki_filp));
252 trace_xfs_file_direct_read(iocb, to);
254 if (!iov_iter_count(to))
255 return 0; /* skip atime */
257 file_accessed(iocb->ki_filp);
259 ret = xfs_ilock_iocb(iocb, XFS_IOLOCK_SHARED);
262 ret = iomap_dio_rw(iocb, to, &xfs_read_iomap_ops, NULL, 0);
263 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
268 static noinline ssize_t
273 struct xfs_inode *ip = XFS_I(iocb->ki_filp->f_mapping->host);
276 trace_xfs_file_dax_read(iocb, to);
278 if (!iov_iter_count(to))
279 return 0; /* skip atime */
281 ret = xfs_ilock_iocb(iocb, XFS_IOLOCK_SHARED);
284 ret = dax_iomap_rw(iocb, to, &xfs_read_iomap_ops);
285 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
287 file_accessed(iocb->ki_filp);
292 xfs_file_buffered_read(
296 struct xfs_inode *ip = XFS_I(file_inode(iocb->ki_filp));
299 trace_xfs_file_buffered_read(iocb, to);
301 ret = xfs_ilock_iocb(iocb, XFS_IOLOCK_SHARED);
304 ret = generic_file_read_iter(iocb, to);
305 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
315 struct inode *inode = file_inode(iocb->ki_filp);
316 struct xfs_mount *mp = XFS_I(inode)->i_mount;
319 XFS_STATS_INC(mp, xs_read_calls);
321 if (xfs_is_shutdown(mp))
325 ret = xfs_file_dax_read(iocb, to);
326 else if (iocb->ki_flags & IOCB_DIRECT)
327 ret = xfs_file_dio_read(iocb, to);
329 ret = xfs_file_buffered_read(iocb, to);
332 XFS_STATS_ADD(mp, xs_read_bytes, ret);
337 * Common pre-write limit and setup checks.
339 * Called with the iolocked held either shared and exclusive according to
340 * @iolock, and returns with it held. Might upgrade the iolock to exclusive
341 * if called for a direct write beyond i_size.
344 xfs_file_write_checks(
346 struct iov_iter *from,
349 struct file *file = iocb->ki_filp;
350 struct inode *inode = file->f_mapping->host;
351 struct xfs_inode *ip = XFS_I(inode);
353 size_t count = iov_iter_count(from);
354 bool drained_dio = false;
358 error = generic_write_checks(iocb, from);
362 if (iocb->ki_flags & IOCB_NOWAIT) {
363 error = break_layout(inode, false);
364 if (error == -EWOULDBLOCK)
367 error = xfs_break_layouts(inode, iolock, BREAK_WRITE);
374 * For changing security info in file_remove_privs() we need i_rwsem
377 if (*iolock == XFS_IOLOCK_SHARED && !IS_NOSEC(inode)) {
378 xfs_iunlock(ip, *iolock);
379 *iolock = XFS_IOLOCK_EXCL;
380 error = xfs_ilock_iocb(iocb, *iolock);
389 * If the offset is beyond the size of the file, we need to zero any
390 * blocks that fall between the existing EOF and the start of this
391 * write. If zeroing is needed and we are currently holding the iolock
392 * shared, we need to update it to exclusive which implies having to
393 * redo all checks before.
395 * We need to serialise against EOF updates that occur in IO completions
396 * here. We want to make sure that nobody is changing the size while we
397 * do this check until we have placed an IO barrier (i.e. hold the
398 * XFS_IOLOCK_EXCL) that prevents new IO from being dispatched. The
399 * spinlock effectively forms a memory barrier once we have the
400 * XFS_IOLOCK_EXCL so we are guaranteed to see the latest EOF value and
401 * hence be able to correctly determine if we need to run zeroing.
403 * We can do an unlocked check here safely as IO completion can only
404 * extend EOF. Truncate is locked out at this point, so the EOF can
405 * not move backwards, only forwards. Hence we only need to take the
406 * slow path and spin locks when we are at or beyond the current EOF.
408 if (iocb->ki_pos <= i_size_read(inode))
411 spin_lock(&ip->i_flags_lock);
412 isize = i_size_read(inode);
413 if (iocb->ki_pos > isize) {
414 spin_unlock(&ip->i_flags_lock);
416 if (iocb->ki_flags & IOCB_NOWAIT)
420 if (*iolock == XFS_IOLOCK_SHARED) {
421 xfs_iunlock(ip, *iolock);
422 *iolock = XFS_IOLOCK_EXCL;
423 xfs_ilock(ip, *iolock);
424 iov_iter_reexpand(from, count);
427 * We now have an IO submission barrier in place, but
428 * AIO can do EOF updates during IO completion and hence
429 * we now need to wait for all of them to drain. Non-AIO
430 * DIO will have drained before we are given the
431 * XFS_IOLOCK_EXCL, and so for most cases this wait is a
434 inode_dio_wait(inode);
439 trace_xfs_zero_eof(ip, isize, iocb->ki_pos - isize);
440 error = iomap_zero_range(inode, isize, iocb->ki_pos - isize,
441 NULL, &xfs_buffered_write_iomap_ops);
445 spin_unlock(&ip->i_flags_lock);
448 return file_modified(file);
452 xfs_dio_write_end_io(
458 struct inode *inode = file_inode(iocb->ki_filp);
459 struct xfs_inode *ip = XFS_I(inode);
460 loff_t offset = iocb->ki_pos;
461 unsigned int nofs_flag;
463 trace_xfs_end_io_direct_write(ip, offset, size);
465 if (xfs_is_shutdown(ip->i_mount))
474 * Capture amount written on completion as we can't reliably account
475 * for it on submission.
477 XFS_STATS_ADD(ip->i_mount, xs_write_bytes, size);
480 * We can allocate memory here while doing writeback on behalf of
481 * memory reclaim. To avoid memory allocation deadlocks set the
482 * task-wide nofs context for the following operations.
484 nofs_flag = memalloc_nofs_save();
486 if (flags & IOMAP_DIO_COW) {
487 error = xfs_reflink_end_cow(ip, offset, size);
493 * Unwritten conversion updates the in-core isize after extent
494 * conversion but before updating the on-disk size. Updating isize any
495 * earlier allows a racing dio read to find unwritten extents before
496 * they are converted.
498 if (flags & IOMAP_DIO_UNWRITTEN) {
499 error = xfs_iomap_write_unwritten(ip, offset, size, true);
504 * We need to update the in-core inode size here so that we don't end up
505 * with the on-disk inode size being outside the in-core inode size. We
506 * have no other method of updating EOF for AIO, so always do it here
509 * We need to lock the test/set EOF update as we can be racing with
510 * other IO completions here to update the EOF. Failing to serialise
511 * here can result in EOF moving backwards and Bad Things Happen when
514 * As IO completion only ever extends EOF, we can do an unlocked check
515 * here to avoid taking the spinlock. If we land within the current EOF,
516 * then we do not need to do an extending update at all, and we don't
517 * need to take the lock to check this. If we race with an update moving
518 * EOF, then we'll either still be beyond EOF and need to take the lock,
519 * or we'll be within EOF and we don't need to take it at all.
521 if (offset + size <= i_size_read(inode))
524 spin_lock(&ip->i_flags_lock);
525 if (offset + size > i_size_read(inode)) {
526 i_size_write(inode, offset + size);
527 spin_unlock(&ip->i_flags_lock);
528 error = xfs_setfilesize(ip, offset, size);
530 spin_unlock(&ip->i_flags_lock);
534 memalloc_nofs_restore(nofs_flag);
538 static const struct iomap_dio_ops xfs_dio_write_ops = {
539 .end_io = xfs_dio_write_end_io,
543 * Handle block aligned direct I/O writes
545 static noinline ssize_t
546 xfs_file_dio_write_aligned(
547 struct xfs_inode *ip,
549 struct iov_iter *from)
551 int iolock = XFS_IOLOCK_SHARED;
554 ret = xfs_ilock_iocb(iocb, iolock);
557 ret = xfs_file_write_checks(iocb, from, &iolock);
562 * We don't need to hold the IOLOCK exclusively across the IO, so demote
563 * the iolock back to shared if we had to take the exclusive lock in
564 * xfs_file_write_checks() for other reasons.
566 if (iolock == XFS_IOLOCK_EXCL) {
567 xfs_ilock_demote(ip, XFS_IOLOCK_EXCL);
568 iolock = XFS_IOLOCK_SHARED;
570 trace_xfs_file_direct_write(iocb, from);
571 ret = iomap_dio_rw(iocb, from, &xfs_direct_write_iomap_ops,
572 &xfs_dio_write_ops, 0);
575 xfs_iunlock(ip, iolock);
580 * Handle block unaligned direct I/O writes
582 * In most cases direct I/O writes will be done holding IOLOCK_SHARED, allowing
583 * them to be done in parallel with reads and other direct I/O writes. However,
584 * if the I/O is not aligned to filesystem blocks, the direct I/O layer may need
585 * to do sub-block zeroing and that requires serialisation against other direct
586 * I/O to the same block. In this case we need to serialise the submission of
587 * the unaligned I/O so that we don't get racing block zeroing in the dio layer.
588 * In the case where sub-block zeroing is not required, we can do concurrent
589 * sub-block dios to the same block successfully.
591 * Optimistically submit the I/O using the shared lock first, but use the
592 * IOMAP_DIO_OVERWRITE_ONLY flag to tell the lower layers to return -EAGAIN
593 * if block allocation or partial block zeroing would be required. In that case
594 * we try again with the exclusive lock.
596 static noinline ssize_t
597 xfs_file_dio_write_unaligned(
598 struct xfs_inode *ip,
600 struct iov_iter *from)
602 size_t isize = i_size_read(VFS_I(ip));
603 size_t count = iov_iter_count(from);
604 int iolock = XFS_IOLOCK_SHARED;
605 unsigned int flags = IOMAP_DIO_OVERWRITE_ONLY;
609 * Extending writes need exclusivity because of the sub-block zeroing
610 * that the DIO code always does for partial tail blocks beyond EOF, so
611 * don't even bother trying the fast path in this case.
613 if (iocb->ki_pos > isize || iocb->ki_pos + count >= isize) {
615 if (iocb->ki_flags & IOCB_NOWAIT)
617 iolock = XFS_IOLOCK_EXCL;
618 flags = IOMAP_DIO_FORCE_WAIT;
621 ret = xfs_ilock_iocb(iocb, iolock);
626 * We can't properly handle unaligned direct I/O to reflink files yet,
627 * as we can't unshare a partial block.
629 if (xfs_is_cow_inode(ip)) {
630 trace_xfs_reflink_bounce_dio_write(iocb, from);
635 ret = xfs_file_write_checks(iocb, from, &iolock);
640 * If we are doing exclusive unaligned I/O, this must be the only I/O
641 * in-flight. Otherwise we risk data corruption due to unwritten extent
642 * conversions from the AIO end_io handler. Wait for all other I/O to
645 if (flags & IOMAP_DIO_FORCE_WAIT)
646 inode_dio_wait(VFS_I(ip));
648 trace_xfs_file_direct_write(iocb, from);
649 ret = iomap_dio_rw(iocb, from, &xfs_direct_write_iomap_ops,
650 &xfs_dio_write_ops, flags);
653 * Retry unaligned I/O with exclusive blocking semantics if the DIO
654 * layer rejected it for mapping or locking reasons. If we are doing
655 * nonblocking user I/O, propagate the error.
657 if (ret == -EAGAIN && !(iocb->ki_flags & IOCB_NOWAIT)) {
658 ASSERT(flags & IOMAP_DIO_OVERWRITE_ONLY);
659 xfs_iunlock(ip, iolock);
660 goto retry_exclusive;
665 xfs_iunlock(ip, iolock);
672 struct iov_iter *from)
674 struct xfs_inode *ip = XFS_I(file_inode(iocb->ki_filp));
675 struct xfs_buftarg *target = xfs_inode_buftarg(ip);
676 size_t count = iov_iter_count(from);
678 /* direct I/O must be aligned to device logical sector size */
679 if ((iocb->ki_pos | count) & target->bt_logical_sectormask)
681 if ((iocb->ki_pos | count) & ip->i_mount->m_blockmask)
682 return xfs_file_dio_write_unaligned(ip, iocb, from);
683 return xfs_file_dio_write_aligned(ip, iocb, from);
686 static noinline ssize_t
689 struct iov_iter *from)
691 struct inode *inode = iocb->ki_filp->f_mapping->host;
692 struct xfs_inode *ip = XFS_I(inode);
693 int iolock = XFS_IOLOCK_EXCL;
694 ssize_t ret, error = 0;
697 ret = xfs_ilock_iocb(iocb, iolock);
700 ret = xfs_file_write_checks(iocb, from, &iolock);
706 trace_xfs_file_dax_write(iocb, from);
707 ret = dax_iomap_rw(iocb, from, &xfs_direct_write_iomap_ops);
708 if (ret > 0 && iocb->ki_pos > i_size_read(inode)) {
709 i_size_write(inode, iocb->ki_pos);
710 error = xfs_setfilesize(ip, pos, ret);
714 xfs_iunlock(ip, iolock);
719 XFS_STATS_ADD(ip->i_mount, xs_write_bytes, ret);
721 /* Handle various SYNC-type writes */
722 ret = generic_write_sync(iocb, ret);
728 xfs_file_buffered_write(
730 struct iov_iter *from)
732 struct file *file = iocb->ki_filp;
733 struct address_space *mapping = file->f_mapping;
734 struct inode *inode = mapping->host;
735 struct xfs_inode *ip = XFS_I(inode);
737 bool cleared_space = false;
740 if (iocb->ki_flags & IOCB_NOWAIT)
744 iolock = XFS_IOLOCK_EXCL;
745 xfs_ilock(ip, iolock);
747 ret = xfs_file_write_checks(iocb, from, &iolock);
751 /* We can write back this queue in page reclaim */
752 current->backing_dev_info = inode_to_bdi(inode);
754 trace_xfs_file_buffered_write(iocb, from);
755 ret = iomap_file_buffered_write(iocb, from,
756 &xfs_buffered_write_iomap_ops);
757 if (likely(ret >= 0))
761 * If we hit a space limit, try to free up some lingering preallocated
762 * space before returning an error. In the case of ENOSPC, first try to
763 * write back all dirty inodes to free up some of the excess reserved
764 * metadata space. This reduces the chances that the eofblocks scan
765 * waits on dirty mappings. Since xfs_flush_inodes() is serialized, this
766 * also behaves as a filter to prevent too many eofblocks scans from
767 * running at the same time. Use a synchronous scan to increase the
768 * effectiveness of the scan.
770 if (ret == -EDQUOT && !cleared_space) {
771 xfs_iunlock(ip, iolock);
772 xfs_blockgc_free_quota(ip, XFS_ICWALK_FLAG_SYNC);
773 cleared_space = true;
775 } else if (ret == -ENOSPC && !cleared_space) {
776 struct xfs_icwalk icw = {0};
778 cleared_space = true;
779 xfs_flush_inodes(ip->i_mount);
781 xfs_iunlock(ip, iolock);
782 icw.icw_flags = XFS_ICWALK_FLAG_SYNC;
783 xfs_blockgc_free_space(ip->i_mount, &icw);
787 current->backing_dev_info = NULL;
790 xfs_iunlock(ip, iolock);
793 XFS_STATS_ADD(ip->i_mount, xs_write_bytes, ret);
794 /* Handle various SYNC-type writes */
795 ret = generic_write_sync(iocb, ret);
803 struct iov_iter *from)
805 struct file *file = iocb->ki_filp;
806 struct address_space *mapping = file->f_mapping;
807 struct inode *inode = mapping->host;
808 struct xfs_inode *ip = XFS_I(inode);
810 size_t ocount = iov_iter_count(from);
812 XFS_STATS_INC(ip->i_mount, xs_write_calls);
817 if (xfs_is_shutdown(ip->i_mount))
821 return xfs_file_dax_write(iocb, from);
823 if (iocb->ki_flags & IOCB_DIRECT) {
825 * Allow a directio write to fall back to a buffered
826 * write *only* in the case that we're doing a reflink
827 * CoW. In all other directio scenarios we do not
828 * allow an operation to fall back to buffered mode.
830 ret = xfs_file_dio_write(iocb, from);
835 return xfs_file_buffered_write(iocb, from);
842 struct xfs_inode *ip = XFS_I(inode);
844 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
846 xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
850 xfs_break_dax_layouts(
856 ASSERT(xfs_isilocked(XFS_I(inode), XFS_MMAPLOCK_EXCL));
858 page = dax_layout_busy_page(inode->i_mapping);
863 return ___wait_var_event(&page->_refcount,
864 atomic_read(&page->_refcount) == 1, TASK_INTERRUPTIBLE,
865 0, 0, xfs_wait_dax_page(inode));
872 enum layout_break_reason reason)
877 ASSERT(xfs_isilocked(XFS_I(inode), XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL));
883 error = xfs_break_dax_layouts(inode, &retry);
888 error = xfs_break_leased_layouts(inode, iolock, &retry);
894 } while (error == 0 && retry);
899 #define XFS_FALLOC_FL_SUPPORTED \
900 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
901 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE | \
902 FALLOC_FL_INSERT_RANGE | FALLOC_FL_UNSHARE_RANGE)
911 struct inode *inode = file_inode(file);
912 struct xfs_inode *ip = XFS_I(inode);
914 enum xfs_prealloc_flags flags = 0;
915 uint iolock = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL;
917 bool do_file_insert = false;
919 if (!S_ISREG(inode->i_mode))
921 if (mode & ~XFS_FALLOC_FL_SUPPORTED)
924 xfs_ilock(ip, iolock);
925 error = xfs_break_layouts(inode, &iolock, BREAK_UNMAP);
930 * Must wait for all AIO to complete before we continue as AIO can
931 * change the file size on completion without holding any locks we
932 * currently hold. We must do this first because AIO can update both
933 * the on disk and in memory inode sizes, and the operations that follow
934 * require the in-memory size to be fully up-to-date.
936 inode_dio_wait(inode);
939 * Now AIO and DIO has drained we flush and (if necessary) invalidate
940 * the cached range over the first operation we are about to run.
942 * We care about zero and collapse here because they both run a hole
943 * punch over the range first. Because that can zero data, and the range
944 * of invalidation for the shift operations is much larger, we still do
945 * the required flush for collapse in xfs_prepare_shift().
947 * Insert has the same range requirements as collapse, and we extend the
948 * file first which can zero data. Hence insert has the same
949 * flush/invalidate requirements as collapse and so they are both
950 * handled at the right time by xfs_prepare_shift().
952 if (mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE |
953 FALLOC_FL_COLLAPSE_RANGE)) {
954 error = xfs_flush_unmap_range(ip, offset, len);
959 if (mode & FALLOC_FL_PUNCH_HOLE) {
960 error = xfs_free_file_space(ip, offset, len);
963 } else if (mode & FALLOC_FL_COLLAPSE_RANGE) {
964 if (!xfs_is_falloc_aligned(ip, offset, len)) {
970 * There is no need to overlap collapse range with EOF,
971 * in which case it is effectively a truncate operation
973 if (offset + len >= i_size_read(inode)) {
978 new_size = i_size_read(inode) - len;
980 error = xfs_collapse_file_space(ip, offset, len);
983 } else if (mode & FALLOC_FL_INSERT_RANGE) {
984 loff_t isize = i_size_read(inode);
986 if (!xfs_is_falloc_aligned(ip, offset, len)) {
992 * New inode size must not exceed ->s_maxbytes, accounting for
993 * possible signed overflow.
995 if (inode->i_sb->s_maxbytes - isize < len) {
999 new_size = isize + len;
1001 /* Offset should be less than i_size */
1002 if (offset >= isize) {
1006 do_file_insert = true;
1008 flags |= XFS_PREALLOC_SET;
1010 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
1011 offset + len > i_size_read(inode)) {
1012 new_size = offset + len;
1013 error = inode_newsize_ok(inode, new_size);
1018 if (mode & FALLOC_FL_ZERO_RANGE) {
1020 * Punch a hole and prealloc the range. We use a hole
1021 * punch rather than unwritten extent conversion for two
1024 * 1.) Hole punch handles partial block zeroing for us.
1025 * 2.) If prealloc returns ENOSPC, the file range is
1026 * still zero-valued by virtue of the hole punch.
1028 unsigned int blksize = i_blocksize(inode);
1030 trace_xfs_zero_file_space(ip);
1032 error = xfs_free_file_space(ip, offset, len);
1036 len = round_up(offset + len, blksize) -
1037 round_down(offset, blksize);
1038 offset = round_down(offset, blksize);
1039 } else if (mode & FALLOC_FL_UNSHARE_RANGE) {
1040 error = xfs_reflink_unshare(ip, offset, len);
1045 * If always_cow mode we can't use preallocations and
1046 * thus should not create them.
1048 if (xfs_is_always_cow_inode(ip)) {
1049 error = -EOPNOTSUPP;
1054 if (!xfs_is_always_cow_inode(ip)) {
1055 error = xfs_alloc_file_space(ip, offset, len,
1056 XFS_BMAPI_PREALLOC);
1062 if (file->f_flags & O_DSYNC)
1063 flags |= XFS_PREALLOC_SYNC;
1065 error = xfs_update_prealloc_flags(ip, flags);
1069 /* Change file size if needed */
1073 iattr.ia_valid = ATTR_SIZE;
1074 iattr.ia_size = new_size;
1075 error = xfs_vn_setattr_size(file_mnt_user_ns(file),
1076 file_dentry(file), &iattr);
1082 * Perform hole insertion now that the file size has been
1083 * updated so that if we crash during the operation we don't
1084 * leave shifted extents past EOF and hence losing access to
1085 * the data that is contained within them.
1088 error = xfs_insert_file_space(ip, offset, len);
1091 xfs_iunlock(ip, iolock);
1102 struct xfs_inode *ip = XFS_I(file_inode(file));
1107 * Operations creating pages in page cache need protection from hole
1108 * punching and similar ops
1110 if (advice == POSIX_FADV_WILLNEED) {
1111 lockflags = XFS_IOLOCK_SHARED;
1112 xfs_ilock(ip, lockflags);
1114 ret = generic_fadvise(file, start, end, advice);
1116 xfs_iunlock(ip, lockflags);
1120 /* Does this file, inode, or mount want synchronous writes? */
1121 static inline bool xfs_file_sync_writes(struct file *filp)
1123 struct xfs_inode *ip = XFS_I(file_inode(filp));
1125 if (xfs_has_wsync(ip->i_mount))
1127 if (filp->f_flags & (__O_SYNC | O_DSYNC))
1129 if (IS_SYNC(file_inode(filp)))
1136 xfs_file_remap_range(
1137 struct file *file_in,
1139 struct file *file_out,
1142 unsigned int remap_flags)
1144 struct inode *inode_in = file_inode(file_in);
1145 struct xfs_inode *src = XFS_I(inode_in);
1146 struct inode *inode_out = file_inode(file_out);
1147 struct xfs_inode *dest = XFS_I(inode_out);
1148 struct xfs_mount *mp = src->i_mount;
1149 loff_t remapped = 0;
1150 xfs_extlen_t cowextsize;
1153 if (remap_flags & ~(REMAP_FILE_DEDUP | REMAP_FILE_ADVISORY))
1156 if (!xfs_has_reflink(mp))
1159 if (xfs_is_shutdown(mp))
1162 /* Prepare and then clone file data. */
1163 ret = xfs_reflink_remap_prep(file_in, pos_in, file_out, pos_out,
1165 if (ret || len == 0)
1168 trace_xfs_reflink_remap_range(src, pos_in, len, dest, pos_out);
1170 ret = xfs_reflink_remap_blocks(src, pos_in, dest, pos_out, len,
1176 * Carry the cowextsize hint from src to dest if we're sharing the
1177 * entire source file to the entire destination file, the source file
1178 * has a cowextsize hint, and the destination file does not.
1181 if (pos_in == 0 && len == i_size_read(inode_in) &&
1182 (src->i_diflags2 & XFS_DIFLAG2_COWEXTSIZE) &&
1183 pos_out == 0 && len >= i_size_read(inode_out) &&
1184 !(dest->i_diflags2 & XFS_DIFLAG2_COWEXTSIZE))
1185 cowextsize = src->i_cowextsize;
1187 ret = xfs_reflink_update_dest(dest, pos_out + len, cowextsize,
1192 if (xfs_file_sync_writes(file_in) || xfs_file_sync_writes(file_out))
1193 xfs_log_force_inode(dest);
1195 xfs_iunlock2_io_mmap(src, dest);
1197 trace_xfs_reflink_remap_range_error(dest, ret, _RET_IP_);
1198 return remapped > 0 ? remapped : ret;
1203 struct inode *inode,
1206 if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1208 if (xfs_is_shutdown(XFS_M(inode->i_sb)))
1210 file->f_mode |= FMODE_NOWAIT | FMODE_BUF_RASYNC;
1216 struct inode *inode,
1219 struct xfs_inode *ip = XFS_I(inode);
1223 error = xfs_file_open(inode, file);
1228 * If there are any blocks, read-ahead block 0 as we're almost
1229 * certain to have the next operation be a read there.
1231 mode = xfs_ilock_data_map_shared(ip);
1232 if (ip->i_df.if_nextents > 0)
1233 error = xfs_dir3_data_readahead(ip, 0, 0);
1234 xfs_iunlock(ip, mode);
1240 struct inode *inode,
1243 return xfs_release(XFS_I(inode));
1249 struct dir_context *ctx)
1251 struct inode *inode = file_inode(file);
1252 xfs_inode_t *ip = XFS_I(inode);
1256 * The Linux API doesn't pass down the total size of the buffer
1257 * we read into down to the filesystem. With the filldir concept
1258 * it's not needed for correct information, but the XFS dir2 leaf
1259 * code wants an estimate of the buffer size to calculate it's
1260 * readahead window and size the buffers used for mapping to
1263 * Try to give it an estimate that's good enough, maybe at some
1264 * point we can change the ->readdir prototype to include the
1265 * buffer size. For now we use the current glibc buffer size.
1267 bufsize = (size_t)min_t(loff_t, XFS_READDIR_BUFSIZE, ip->i_disk_size);
1269 return xfs_readdir(NULL, ip, ctx, bufsize);
1278 struct inode *inode = file->f_mapping->host;
1280 if (xfs_is_shutdown(XFS_I(inode)->i_mount))
1285 return generic_file_llseek(file, offset, whence);
1287 offset = iomap_seek_hole(inode, offset, &xfs_seek_iomap_ops);
1290 offset = iomap_seek_data(inode, offset, &xfs_seek_iomap_ops);
1296 return vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
1300 * Locking for serialisation of IO during page faults. This results in a lock
1304 * sb_start_pagefault(vfs, freeze)
1305 * invalidate_lock (vfs/XFS_MMAPLOCK - truncate serialisation)
1307 * i_lock (XFS - extent map serialisation)
1310 __xfs_filemap_fault(
1311 struct vm_fault *vmf,
1312 enum page_entry_size pe_size,
1315 struct inode *inode = file_inode(vmf->vma->vm_file);
1316 struct xfs_inode *ip = XFS_I(inode);
1319 trace_xfs_filemap_fault(ip, pe_size, write_fault);
1322 sb_start_pagefault(inode->i_sb);
1323 file_update_time(vmf->vma->vm_file);
1326 if (IS_DAX(inode)) {
1329 xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1330 ret = dax_iomap_fault(vmf, pe_size, &pfn, NULL,
1331 (write_fault && !vmf->cow_page) ?
1332 &xfs_direct_write_iomap_ops :
1333 &xfs_read_iomap_ops);
1334 if (ret & VM_FAULT_NEEDDSYNC)
1335 ret = dax_finish_sync_fault(vmf, pe_size, pfn);
1336 xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1339 xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1340 ret = iomap_page_mkwrite(vmf,
1341 &xfs_buffered_write_iomap_ops);
1342 xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1344 ret = filemap_fault(vmf);
1349 sb_end_pagefault(inode->i_sb);
1355 struct vm_fault *vmf)
1357 return (vmf->flags & FAULT_FLAG_WRITE) &&
1358 (vmf->vma->vm_flags & VM_SHARED);
1363 struct vm_fault *vmf)
1365 /* DAX can shortcut the normal fault path on write faults! */
1366 return __xfs_filemap_fault(vmf, PE_SIZE_PTE,
1367 IS_DAX(file_inode(vmf->vma->vm_file)) &&
1368 xfs_is_write_fault(vmf));
1372 xfs_filemap_huge_fault(
1373 struct vm_fault *vmf,
1374 enum page_entry_size pe_size)
1376 if (!IS_DAX(file_inode(vmf->vma->vm_file)))
1377 return VM_FAULT_FALLBACK;
1379 /* DAX can shortcut the normal fault path on write faults! */
1380 return __xfs_filemap_fault(vmf, pe_size,
1381 xfs_is_write_fault(vmf));
1385 xfs_filemap_page_mkwrite(
1386 struct vm_fault *vmf)
1388 return __xfs_filemap_fault(vmf, PE_SIZE_PTE, true);
1392 * pfn_mkwrite was originally intended to ensure we capture time stamp updates
1393 * on write faults. In reality, it needs to serialise against truncate and
1394 * prepare memory for writing so handle is as standard write fault.
1397 xfs_filemap_pfn_mkwrite(
1398 struct vm_fault *vmf)
1401 return __xfs_filemap_fault(vmf, PE_SIZE_PTE, true);
1405 xfs_filemap_map_pages(
1406 struct vm_fault *vmf,
1407 pgoff_t start_pgoff,
1410 struct inode *inode = file_inode(vmf->vma->vm_file);
1413 xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1414 ret = filemap_map_pages(vmf, start_pgoff, end_pgoff);
1415 xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1419 static const struct vm_operations_struct xfs_file_vm_ops = {
1420 .fault = xfs_filemap_fault,
1421 .huge_fault = xfs_filemap_huge_fault,
1422 .map_pages = xfs_filemap_map_pages,
1423 .page_mkwrite = xfs_filemap_page_mkwrite,
1424 .pfn_mkwrite = xfs_filemap_pfn_mkwrite,
1430 struct vm_area_struct *vma)
1432 struct inode *inode = file_inode(file);
1433 struct xfs_buftarg *target = xfs_inode_buftarg(XFS_I(inode));
1436 * We don't support synchronous mappings for non-DAX files and
1437 * for DAX files if underneath dax_device is not synchronous.
1439 if (!daxdev_mapping_supported(vma, target->bt_daxdev))
1442 file_accessed(file);
1443 vma->vm_ops = &xfs_file_vm_ops;
1445 vma->vm_flags |= VM_HUGEPAGE;
1449 const struct file_operations xfs_file_operations = {
1450 .llseek = xfs_file_llseek,
1451 .read_iter = xfs_file_read_iter,
1452 .write_iter = xfs_file_write_iter,
1453 .splice_read = generic_file_splice_read,
1454 .splice_write = iter_file_splice_write,
1455 .iopoll = iomap_dio_iopoll,
1456 .unlocked_ioctl = xfs_file_ioctl,
1457 #ifdef CONFIG_COMPAT
1458 .compat_ioctl = xfs_file_compat_ioctl,
1460 .mmap = xfs_file_mmap,
1461 .mmap_supported_flags = MAP_SYNC,
1462 .open = xfs_file_open,
1463 .release = xfs_file_release,
1464 .fsync = xfs_file_fsync,
1465 .get_unmapped_area = thp_get_unmapped_area,
1466 .fallocate = xfs_file_fallocate,
1467 .fadvise = xfs_file_fadvise,
1468 .remap_file_range = xfs_file_remap_range,
1471 const struct file_operations xfs_dir_file_operations = {
1472 .open = xfs_dir_open,
1473 .read = generic_read_dir,
1474 .iterate_shared = xfs_file_readdir,
1475 .llseek = generic_file_llseek,
1476 .unlocked_ioctl = xfs_file_ioctl,
1477 #ifdef CONFIG_COMPAT
1478 .compat_ioctl = xfs_file_compat_ioctl,
1480 .fsync = xfs_dir_fsync,