2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_shared.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
24 #include "xfs_mount.h"
25 #include "xfs_da_format.h"
26 #include "xfs_da_btree.h"
27 #include "xfs_inode.h"
28 #include "xfs_trans.h"
29 #include "xfs_inode_item.h"
31 #include "xfs_bmap_util.h"
32 #include "xfs_error.h"
34 #include "xfs_dir2_priv.h"
35 #include "xfs_ioctl.h"
36 #include "xfs_trace.h"
38 #include "xfs_icache.h"
40 #include "xfs_iomap.h"
41 #include "xfs_reflink.h"
43 #include <linux/dcache.h>
44 #include <linux/falloc.h>
45 #include <linux/pagevec.h>
46 #include <linux/backing-dev.h>
48 static const struct vm_operations_struct xfs_file_vm_ops;
51 * Clear the specified ranges to zero through either the pagecache or DAX.
52 * Holes and unwritten extents will be left as-is as they already are zeroed.
61 return iomap_zero_range(VFS_I(ip), pos, count, did_zero, &xfs_iomap_ops);
65 xfs_update_prealloc_flags(
67 enum xfs_prealloc_flags flags)
72 error = xfs_trans_alloc(ip->i_mount, &M_RES(ip->i_mount)->tr_writeid,
77 xfs_ilock(ip, XFS_ILOCK_EXCL);
78 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
80 if (!(flags & XFS_PREALLOC_INVISIBLE)) {
81 VFS_I(ip)->i_mode &= ~S_ISUID;
82 if (VFS_I(ip)->i_mode & S_IXGRP)
83 VFS_I(ip)->i_mode &= ~S_ISGID;
84 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
87 if (flags & XFS_PREALLOC_SET)
88 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
89 if (flags & XFS_PREALLOC_CLEAR)
90 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
92 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
93 if (flags & XFS_PREALLOC_SYNC)
94 xfs_trans_set_sync(tp);
95 return xfs_trans_commit(tp);
99 * Fsync operations on directories are much simpler than on regular files,
100 * as there is no file data to flush, and thus also no need for explicit
101 * cache flush operations, and there are no non-transaction metadata updates
102 * on directories either.
111 struct xfs_inode *ip = XFS_I(file->f_mapping->host);
112 struct xfs_mount *mp = ip->i_mount;
115 trace_xfs_dir_fsync(ip);
117 xfs_ilock(ip, XFS_ILOCK_SHARED);
118 if (xfs_ipincount(ip))
119 lsn = ip->i_itemp->ili_last_lsn;
120 xfs_iunlock(ip, XFS_ILOCK_SHARED);
124 return _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, NULL);
134 struct inode *inode = file->f_mapping->host;
135 struct xfs_inode *ip = XFS_I(inode);
136 struct xfs_mount *mp = ip->i_mount;
141 trace_xfs_file_fsync(ip);
143 error = file_write_and_wait_range(file, start, end);
147 if (XFS_FORCED_SHUTDOWN(mp))
150 xfs_iflags_clear(ip, XFS_ITRUNCATED);
153 * If we have an RT and/or log subvolume we need to make sure to flush
154 * the write cache the device used for file data first. This is to
155 * ensure newly written file data make it to disk before logging the new
156 * inode size in case of an extending write.
158 if (XFS_IS_REALTIME_INODE(ip))
159 xfs_blkdev_issue_flush(mp->m_rtdev_targp);
160 else if (mp->m_logdev_targp != mp->m_ddev_targp)
161 xfs_blkdev_issue_flush(mp->m_ddev_targp);
164 * All metadata updates are logged, which means that we just have to
165 * flush the log up to the latest LSN that touched the inode. If we have
166 * concurrent fsync/fdatasync() calls, we need them to all block on the
167 * log force before we clear the ili_fsync_fields field. This ensures
168 * that we don't get a racing sync operation that does not wait for the
169 * metadata to hit the journal before returning. If we race with
170 * clearing the ili_fsync_fields, then all that will happen is the log
171 * force will do nothing as the lsn will already be on disk. We can't
172 * race with setting ili_fsync_fields because that is done under
173 * XFS_ILOCK_EXCL, and that can't happen because we hold the lock shared
174 * until after the ili_fsync_fields is cleared.
176 xfs_ilock(ip, XFS_ILOCK_SHARED);
177 if (xfs_ipincount(ip)) {
179 (ip->i_itemp->ili_fsync_fields & ~XFS_ILOG_TIMESTAMP))
180 lsn = ip->i_itemp->ili_last_lsn;
184 error = _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, &log_flushed);
185 ip->i_itemp->ili_fsync_fields = 0;
187 xfs_iunlock(ip, XFS_ILOCK_SHARED);
190 * If we only have a single device, and the log force about was
191 * a no-op we might have to flush the data device cache here.
192 * This can only happen for fdatasync/O_DSYNC if we were overwriting
193 * an already allocated file and thus do not have any metadata to
196 if (!log_flushed && !XFS_IS_REALTIME_INODE(ip) &&
197 mp->m_logdev_targp == mp->m_ddev_targp)
198 xfs_blkdev_issue_flush(mp->m_ddev_targp);
204 xfs_file_dio_aio_read(
208 struct xfs_inode *ip = XFS_I(file_inode(iocb->ki_filp));
209 size_t count = iov_iter_count(to);
212 trace_xfs_file_direct_read(ip, count, iocb->ki_pos);
215 return 0; /* skip atime */
217 file_accessed(iocb->ki_filp);
219 xfs_ilock(ip, XFS_IOLOCK_SHARED);
220 ret = iomap_dio_rw(iocb, to, &xfs_iomap_ops, NULL);
221 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
226 static noinline ssize_t
231 struct xfs_inode *ip = XFS_I(iocb->ki_filp->f_mapping->host);
232 size_t count = iov_iter_count(to);
235 trace_xfs_file_dax_read(ip, count, iocb->ki_pos);
238 return 0; /* skip atime */
240 if (iocb->ki_flags & IOCB_NOWAIT) {
241 if (!xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED))
244 xfs_ilock(ip, XFS_IOLOCK_SHARED);
247 ret = dax_iomap_rw(iocb, to, &xfs_iomap_ops);
248 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
250 file_accessed(iocb->ki_filp);
255 xfs_file_buffered_aio_read(
259 struct xfs_inode *ip = XFS_I(file_inode(iocb->ki_filp));
262 trace_xfs_file_buffered_read(ip, iov_iter_count(to), iocb->ki_pos);
264 if (iocb->ki_flags & IOCB_NOWAIT) {
265 if (!xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED))
268 xfs_ilock(ip, 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_FORCED_SHUTDOWN(mp))
291 ret = xfs_file_dax_read(iocb, to);
292 else if (iocb->ki_flags & IOCB_DIRECT)
293 ret = xfs_file_dio_aio_read(iocb, to);
295 ret = xfs_file_buffered_aio_read(iocb, to);
298 XFS_STATS_ADD(mp, xs_read_bytes, ret);
303 * Zero any on disk space between the current EOF and the new, larger EOF.
305 * This handles the normal case of zeroing the remainder of the last block in
306 * the file and the unusual case of zeroing blocks out beyond the size of the
307 * file. This second case only happens with fixed size extents and when the
308 * system crashes before the inode size was updated but after blocks were
311 * Expects the iolock to be held exclusive, and will take the ilock internally.
313 int /* error (positive) */
315 struct xfs_inode *ip,
316 xfs_off_t offset, /* starting I/O offset */
317 xfs_fsize_t isize, /* current inode size */
320 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
321 ASSERT(offset > isize);
323 trace_xfs_zero_eof(ip, isize, offset - isize);
324 return xfs_zero_range(ip, isize, offset - isize, did_zeroing);
328 * Common pre-write limit and setup checks.
330 * Called with the iolocked held either shared and exclusive according to
331 * @iolock, and returns with it held. Might upgrade the iolock to exclusive
332 * if called for a direct write beyond i_size.
335 xfs_file_aio_write_checks(
337 struct iov_iter *from,
340 struct file *file = iocb->ki_filp;
341 struct inode *inode = file->f_mapping->host;
342 struct xfs_inode *ip = XFS_I(inode);
344 size_t count = iov_iter_count(from);
345 bool drained_dio = false;
348 error = generic_write_checks(iocb, from);
352 error = xfs_break_layouts(inode, iolock);
357 * For changing security info in file_remove_privs() we need i_rwsem
360 if (*iolock == XFS_IOLOCK_SHARED && !IS_NOSEC(inode)) {
361 xfs_iunlock(ip, *iolock);
362 *iolock = XFS_IOLOCK_EXCL;
363 xfs_ilock(ip, *iolock);
367 * If the offset is beyond the size of the file, we need to zero any
368 * blocks that fall between the existing EOF and the start of this
369 * write. If zeroing is needed and we are currently holding the
370 * iolock shared, we need to update it to exclusive which implies
371 * having to redo all checks before.
373 * We need to serialise against EOF updates that occur in IO
374 * completions here. We want to make sure that nobody is changing the
375 * size while we do this check until we have placed an IO barrier (i.e.
376 * hold the XFS_IOLOCK_EXCL) that prevents new IO from being dispatched.
377 * The spinlock effectively forms a memory barrier once we have the
378 * XFS_IOLOCK_EXCL so we are guaranteed to see the latest EOF value
379 * and hence be able to correctly determine if we need to run zeroing.
381 spin_lock(&ip->i_flags_lock);
382 if (iocb->ki_pos > i_size_read(inode)) {
383 spin_unlock(&ip->i_flags_lock);
385 if (*iolock == XFS_IOLOCK_SHARED) {
386 xfs_iunlock(ip, *iolock);
387 *iolock = XFS_IOLOCK_EXCL;
388 xfs_ilock(ip, *iolock);
389 iov_iter_reexpand(from, count);
392 * We now have an IO submission barrier in place, but
393 * AIO can do EOF updates during IO completion and hence
394 * we now need to wait for all of them to drain. Non-AIO
395 * DIO will have drained before we are given the
396 * XFS_IOLOCK_EXCL, and so for most cases this wait is a
399 inode_dio_wait(inode);
403 error = xfs_zero_eof(ip, iocb->ki_pos, i_size_read(inode), NULL);
407 spin_unlock(&ip->i_flags_lock);
410 * Updating the timestamps will grab the ilock again from
411 * xfs_fs_dirty_inode, so we have to call it after dropping the
412 * lock above. Eventually we should look into a way to avoid
413 * the pointless lock roundtrip.
415 if (likely(!(file->f_mode & FMODE_NOCMTIME))) {
416 error = file_update_time(file);
422 * If we're writing the file then make sure to clear the setuid and
423 * setgid bits if the process is not being run by root. This keeps
424 * people from modifying setuid and setgid binaries.
426 if (!IS_NOSEC(inode))
427 return file_remove_privs(file);
432 xfs_dio_write_end_io(
437 struct inode *inode = file_inode(iocb->ki_filp);
438 struct xfs_inode *ip = XFS_I(inode);
439 loff_t offset = iocb->ki_pos;
442 trace_xfs_end_io_direct_write(ip, offset, size);
444 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
450 if (flags & IOMAP_DIO_COW) {
451 error = xfs_reflink_end_cow(ip, offset, size);
457 * Unwritten conversion updates the in-core isize after extent
458 * conversion but before updating the on-disk size. Updating isize any
459 * earlier allows a racing dio read to find unwritten extents before
460 * they are converted.
462 if (flags & IOMAP_DIO_UNWRITTEN)
463 return xfs_iomap_write_unwritten(ip, offset, size, true);
466 * We need to update the in-core inode size here so that we don't end up
467 * with the on-disk inode size being outside the in-core inode size. We
468 * have no other method of updating EOF for AIO, so always do it here
471 * We need to lock the test/set EOF update as we can be racing with
472 * other IO completions here to update the EOF. Failing to serialise
473 * here can result in EOF moving backwards and Bad Things Happen when
476 spin_lock(&ip->i_flags_lock);
477 if (offset + size > i_size_read(inode)) {
478 i_size_write(inode, offset + size);
479 spin_unlock(&ip->i_flags_lock);
480 error = xfs_setfilesize(ip, offset, size);
482 spin_unlock(&ip->i_flags_lock);
489 * xfs_file_dio_aio_write - handle direct IO writes
491 * Lock the inode appropriately to prepare for and issue a direct IO write.
492 * By separating it from the buffered write path we remove all the tricky to
493 * follow locking changes and looping.
495 * If there are cached pages or we're extending the file, we need IOLOCK_EXCL
496 * until we're sure the bytes at the new EOF have been zeroed and/or the cached
497 * pages are flushed out.
499 * In most cases the direct IO writes will be done holding IOLOCK_SHARED
500 * allowing them to be done in parallel with reads and other direct IO writes.
501 * However, if the IO is not aligned to filesystem blocks, the direct IO layer
502 * needs to do sub-block zeroing and that requires serialisation against other
503 * direct IOs to the same block. In this case we need to serialise the
504 * submission of the unaligned IOs so that we don't get racing block zeroing in
505 * the dio layer. To avoid the problem with aio, we also need to wait for
506 * outstanding IOs to complete so that unwritten extent conversion is completed
507 * before we try to map the overlapping block. This is currently implemented by
508 * hitting it with a big hammer (i.e. inode_dio_wait()).
510 * Returns with locks held indicated by @iolock and errors indicated by
511 * negative return values.
514 xfs_file_dio_aio_write(
516 struct iov_iter *from)
518 struct file *file = iocb->ki_filp;
519 struct address_space *mapping = file->f_mapping;
520 struct inode *inode = mapping->host;
521 struct xfs_inode *ip = XFS_I(inode);
522 struct xfs_mount *mp = ip->i_mount;
524 int unaligned_io = 0;
526 size_t count = iov_iter_count(from);
527 struct xfs_buftarg *target = XFS_IS_REALTIME_INODE(ip) ?
528 mp->m_rtdev_targp : mp->m_ddev_targp;
530 /* DIO must be aligned to device logical sector size */
531 if ((iocb->ki_pos | count) & target->bt_logical_sectormask)
535 * Don't take the exclusive iolock here unless the I/O is unaligned to
536 * the file system block size. We don't need to consider the EOF
537 * extension case here because xfs_file_aio_write_checks() will relock
538 * the inode as necessary for EOF zeroing cases and fill out the new
539 * inode size as appropriate.
541 if ((iocb->ki_pos & mp->m_blockmask) ||
542 ((iocb->ki_pos + count) & mp->m_blockmask)) {
546 * We can't properly handle unaligned direct I/O to reflink
547 * files yet, as we can't unshare a partial block.
549 if (xfs_is_reflink_inode(ip)) {
550 trace_xfs_reflink_bounce_dio_write(ip, iocb->ki_pos, count);
553 iolock = XFS_IOLOCK_EXCL;
555 iolock = XFS_IOLOCK_SHARED;
558 if (iocb->ki_flags & IOCB_NOWAIT) {
559 if (!xfs_ilock_nowait(ip, iolock))
562 xfs_ilock(ip, iolock);
565 ret = xfs_file_aio_write_checks(iocb, from, &iolock);
568 count = iov_iter_count(from);
571 * If we are doing unaligned IO, wait for all other IO to drain,
572 * otherwise demote the lock if we had to take the exclusive lock
573 * for other reasons in xfs_file_aio_write_checks.
576 /* If we are going to wait for other DIO to finish, bail */
577 if (iocb->ki_flags & IOCB_NOWAIT) {
578 if (atomic_read(&inode->i_dio_count))
581 inode_dio_wait(inode);
583 } else if (iolock == XFS_IOLOCK_EXCL) {
584 xfs_ilock_demote(ip, XFS_IOLOCK_EXCL);
585 iolock = XFS_IOLOCK_SHARED;
588 trace_xfs_file_direct_write(ip, count, iocb->ki_pos);
589 ret = iomap_dio_rw(iocb, from, &xfs_iomap_ops, xfs_dio_write_end_io);
591 xfs_iunlock(ip, iolock);
594 * No fallback to buffered IO on errors for XFS, direct IO will either
595 * complete fully or fail.
597 ASSERT(ret < 0 || ret == count);
601 static noinline ssize_t
604 struct iov_iter *from)
606 struct inode *inode = iocb->ki_filp->f_mapping->host;
607 struct xfs_inode *ip = XFS_I(inode);
608 int iolock = XFS_IOLOCK_EXCL;
609 ssize_t ret, error = 0;
613 if (iocb->ki_flags & IOCB_NOWAIT) {
614 if (!xfs_ilock_nowait(ip, iolock))
617 xfs_ilock(ip, iolock);
620 ret = xfs_file_aio_write_checks(iocb, from, &iolock);
625 count = iov_iter_count(from);
627 trace_xfs_file_dax_write(ip, count, pos);
628 ret = dax_iomap_rw(iocb, from, &xfs_iomap_ops);
629 if (ret > 0 && iocb->ki_pos > i_size_read(inode)) {
630 i_size_write(inode, iocb->ki_pos);
631 error = xfs_setfilesize(ip, pos, ret);
634 xfs_iunlock(ip, iolock);
635 return error ? error : ret;
639 xfs_file_buffered_aio_write(
641 struct iov_iter *from)
643 struct file *file = iocb->ki_filp;
644 struct address_space *mapping = file->f_mapping;
645 struct inode *inode = mapping->host;
646 struct xfs_inode *ip = XFS_I(inode);
651 if (iocb->ki_flags & IOCB_NOWAIT)
655 iolock = XFS_IOLOCK_EXCL;
656 xfs_ilock(ip, iolock);
658 ret = xfs_file_aio_write_checks(iocb, from, &iolock);
662 /* We can write back this queue in page reclaim */
663 current->backing_dev_info = inode_to_bdi(inode);
665 trace_xfs_file_buffered_write(ip, iov_iter_count(from), iocb->ki_pos);
666 ret = iomap_file_buffered_write(iocb, from, &xfs_iomap_ops);
667 if (likely(ret >= 0))
671 * If we hit a space limit, try to free up some lingering preallocated
672 * space before returning an error. In the case of ENOSPC, first try to
673 * write back all dirty inodes to free up some of the excess reserved
674 * metadata space. This reduces the chances that the eofblocks scan
675 * waits on dirty mappings. Since xfs_flush_inodes() is serialized, this
676 * also behaves as a filter to prevent too many eofblocks scans from
677 * running at the same time.
679 if (ret == -EDQUOT && !enospc) {
680 xfs_iunlock(ip, iolock);
681 enospc = xfs_inode_free_quota_eofblocks(ip);
684 enospc = xfs_inode_free_quota_cowblocks(ip);
688 } else if (ret == -ENOSPC && !enospc) {
689 struct xfs_eofblocks eofb = {0};
692 xfs_flush_inodes(ip->i_mount);
694 xfs_iunlock(ip, iolock);
695 eofb.eof_flags = XFS_EOF_FLAGS_SYNC;
696 xfs_icache_free_eofblocks(ip->i_mount, &eofb);
697 xfs_icache_free_cowblocks(ip->i_mount, &eofb);
701 current->backing_dev_info = NULL;
704 xfs_iunlock(ip, iolock);
711 struct iov_iter *from)
713 struct file *file = iocb->ki_filp;
714 struct address_space *mapping = file->f_mapping;
715 struct inode *inode = mapping->host;
716 struct xfs_inode *ip = XFS_I(inode);
718 size_t ocount = iov_iter_count(from);
720 XFS_STATS_INC(ip->i_mount, xs_write_calls);
725 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
729 ret = xfs_file_dax_write(iocb, from);
730 else if (iocb->ki_flags & IOCB_DIRECT) {
732 * Allow a directio write to fall back to a buffered
733 * write *only* in the case that we're doing a reflink
734 * CoW. In all other directio scenarios we do not
735 * allow an operation to fall back to buffered mode.
737 ret = xfs_file_dio_aio_write(iocb, from);
742 ret = xfs_file_buffered_aio_write(iocb, from);
746 XFS_STATS_ADD(ip->i_mount, xs_write_bytes, ret);
748 /* Handle various SYNC-type writes */
749 ret = generic_write_sync(iocb, ret);
754 #define XFS_FALLOC_FL_SUPPORTED \
755 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
756 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE | \
757 FALLOC_FL_INSERT_RANGE | FALLOC_FL_UNSHARE_RANGE)
766 struct inode *inode = file_inode(file);
767 struct xfs_inode *ip = XFS_I(inode);
769 enum xfs_prealloc_flags flags = 0;
770 uint iolock = XFS_IOLOCK_EXCL;
772 bool do_file_insert = false;
774 if (!S_ISREG(inode->i_mode))
776 if (mode & ~XFS_FALLOC_FL_SUPPORTED)
779 xfs_ilock(ip, iolock);
780 error = xfs_break_layouts(inode, &iolock);
784 xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
785 iolock |= XFS_MMAPLOCK_EXCL;
787 if (mode & FALLOC_FL_PUNCH_HOLE) {
788 error = xfs_free_file_space(ip, offset, len);
791 } else if (mode & FALLOC_FL_COLLAPSE_RANGE) {
792 unsigned int blksize_mask = i_blocksize(inode) - 1;
794 if (offset & blksize_mask || len & blksize_mask) {
800 * There is no need to overlap collapse range with EOF,
801 * in which case it is effectively a truncate operation
803 if (offset + len >= i_size_read(inode)) {
808 new_size = i_size_read(inode) - len;
810 error = xfs_collapse_file_space(ip, offset, len);
813 } else if (mode & FALLOC_FL_INSERT_RANGE) {
814 unsigned int blksize_mask = i_blocksize(inode) - 1;
816 new_size = i_size_read(inode) + len;
817 if (offset & blksize_mask || len & blksize_mask) {
822 /* check the new inode size does not wrap through zero */
823 if (new_size > inode->i_sb->s_maxbytes) {
828 /* Offset should be less than i_size */
829 if (offset >= i_size_read(inode)) {
833 do_file_insert = true;
835 flags |= XFS_PREALLOC_SET;
837 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
838 offset + len > i_size_read(inode)) {
839 new_size = offset + len;
840 error = inode_newsize_ok(inode, new_size);
845 if (mode & FALLOC_FL_ZERO_RANGE)
846 error = xfs_zero_file_space(ip, offset, len);
848 if (mode & FALLOC_FL_UNSHARE_RANGE) {
849 error = xfs_reflink_unshare(ip, offset, len);
853 error = xfs_alloc_file_space(ip, offset, len,
860 if (file->f_flags & O_DSYNC)
861 flags |= XFS_PREALLOC_SYNC;
863 error = xfs_update_prealloc_flags(ip, flags);
867 /* Change file size if needed */
871 iattr.ia_valid = ATTR_SIZE;
872 iattr.ia_size = new_size;
873 error = xfs_vn_setattr_size(file_dentry(file), &iattr);
879 * Perform hole insertion now that the file size has been
880 * updated so that if we crash during the operation we don't
881 * leave shifted extents past EOF and hence losing access to
882 * the data that is contained within them.
885 error = xfs_insert_file_space(ip, offset, len);
888 xfs_iunlock(ip, iolock);
893 xfs_file_clone_range(
894 struct file *file_in,
896 struct file *file_out,
900 return xfs_reflink_remap_range(file_in, pos_in, file_out, pos_out,
905 xfs_file_dedupe_range(
906 struct file *src_file,
909 struct file *dst_file,
914 error = xfs_reflink_remap_range(src_file, loff, dst_file, dst_loff,
926 if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
928 if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb)))
930 file->f_mode |= FMODE_NOWAIT;
939 struct xfs_inode *ip = XFS_I(inode);
943 error = xfs_file_open(inode, file);
948 * If there are any blocks, read-ahead block 0 as we're almost
949 * certain to have the next operation be a read there.
951 mode = xfs_ilock_data_map_shared(ip);
952 if (ip->i_d.di_nextents > 0)
953 error = xfs_dir3_data_readahead(ip, 0, -1);
954 xfs_iunlock(ip, mode);
963 return xfs_release(XFS_I(inode));
969 struct dir_context *ctx)
971 struct inode *inode = file_inode(file);
972 xfs_inode_t *ip = XFS_I(inode);
976 * The Linux API doesn't pass down the total size of the buffer
977 * we read into down to the filesystem. With the filldir concept
978 * it's not needed for correct information, but the XFS dir2 leaf
979 * code wants an estimate of the buffer size to calculate it's
980 * readahead window and size the buffers used for mapping to
983 * Try to give it an estimate that's good enough, maybe at some
984 * point we can change the ->readdir prototype to include the
985 * buffer size. For now we use the current glibc buffer size.
987 bufsize = (size_t)min_t(loff_t, XFS_READDIR_BUFSIZE, ip->i_d.di_size);
989 return xfs_readdir(NULL, ip, ctx, bufsize);
998 struct inode *inode = file->f_mapping->host;
1000 if (XFS_FORCED_SHUTDOWN(XFS_I(inode)->i_mount))
1005 return generic_file_llseek(file, offset, whence);
1007 offset = iomap_seek_hole(inode, offset, &xfs_iomap_ops);
1010 offset = iomap_seek_data(inode, offset, &xfs_iomap_ops);
1016 return vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
1020 * Locking for serialisation of IO during page faults. This results in a lock
1024 * sb_start_pagefault(vfs, freeze)
1025 * i_mmaplock (XFS - truncate serialisation)
1027 * i_lock (XFS - extent map serialisation)
1030 __xfs_filemap_fault(
1031 struct vm_fault *vmf,
1032 enum page_entry_size pe_size,
1035 struct inode *inode = file_inode(vmf->vma->vm_file);
1036 struct xfs_inode *ip = XFS_I(inode);
1039 trace_xfs_filemap_fault(ip, pe_size, write_fault);
1042 sb_start_pagefault(inode->i_sb);
1043 file_update_time(vmf->vma->vm_file);
1046 xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1047 if (IS_DAX(inode)) {
1048 ret = dax_iomap_fault(vmf, pe_size, &xfs_iomap_ops);
1051 ret = iomap_page_mkwrite(vmf, &xfs_iomap_ops);
1053 ret = filemap_fault(vmf);
1055 xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1058 sb_end_pagefault(inode->i_sb);
1064 struct vm_fault *vmf)
1066 /* DAX can shortcut the normal fault path on write faults! */
1067 return __xfs_filemap_fault(vmf, PE_SIZE_PTE,
1068 IS_DAX(file_inode(vmf->vma->vm_file)) &&
1069 (vmf->flags & FAULT_FLAG_WRITE));
1073 xfs_filemap_huge_fault(
1074 struct vm_fault *vmf,
1075 enum page_entry_size pe_size)
1077 if (!IS_DAX(file_inode(vmf->vma->vm_file)))
1078 return VM_FAULT_FALLBACK;
1080 /* DAX can shortcut the normal fault path on write faults! */
1081 return __xfs_filemap_fault(vmf, pe_size,
1082 (vmf->flags & FAULT_FLAG_WRITE));
1086 xfs_filemap_page_mkwrite(
1087 struct vm_fault *vmf)
1089 return __xfs_filemap_fault(vmf, PE_SIZE_PTE, true);
1093 * pfn_mkwrite was originally inteneded to ensure we capture time stamp
1094 * updates on write faults. In reality, it's need to serialise against
1095 * truncate similar to page_mkwrite. Hence we cycle the XFS_MMAPLOCK_SHARED
1096 * to ensure we serialise the fault barrier in place.
1099 xfs_filemap_pfn_mkwrite(
1100 struct vm_fault *vmf)
1103 struct inode *inode = file_inode(vmf->vma->vm_file);
1104 struct xfs_inode *ip = XFS_I(inode);
1105 int ret = VM_FAULT_NOPAGE;
1108 trace_xfs_filemap_pfn_mkwrite(ip);
1110 sb_start_pagefault(inode->i_sb);
1111 file_update_time(vmf->vma->vm_file);
1113 /* check if the faulting page hasn't raced with truncate */
1114 xfs_ilock(ip, XFS_MMAPLOCK_SHARED);
1115 size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT;
1116 if (vmf->pgoff >= size)
1117 ret = VM_FAULT_SIGBUS;
1118 else if (IS_DAX(inode))
1119 ret = dax_iomap_fault(vmf, PE_SIZE_PTE, &xfs_iomap_ops);
1120 xfs_iunlock(ip, XFS_MMAPLOCK_SHARED);
1121 sb_end_pagefault(inode->i_sb);
1126 static const struct vm_operations_struct xfs_file_vm_ops = {
1127 .fault = xfs_filemap_fault,
1128 .huge_fault = xfs_filemap_huge_fault,
1129 .map_pages = filemap_map_pages,
1130 .page_mkwrite = xfs_filemap_page_mkwrite,
1131 .pfn_mkwrite = xfs_filemap_pfn_mkwrite,
1137 struct vm_area_struct *vma)
1139 file_accessed(filp);
1140 vma->vm_ops = &xfs_file_vm_ops;
1141 if (IS_DAX(file_inode(filp)))
1142 vma->vm_flags |= VM_MIXEDMAP | VM_HUGEPAGE;
1146 const struct file_operations xfs_file_operations = {
1147 .llseek = xfs_file_llseek,
1148 .read_iter = xfs_file_read_iter,
1149 .write_iter = xfs_file_write_iter,
1150 .splice_read = generic_file_splice_read,
1151 .splice_write = iter_file_splice_write,
1152 .unlocked_ioctl = xfs_file_ioctl,
1153 #ifdef CONFIG_COMPAT
1154 .compat_ioctl = xfs_file_compat_ioctl,
1156 .mmap = xfs_file_mmap,
1157 .open = xfs_file_open,
1158 .release = xfs_file_release,
1159 .fsync = xfs_file_fsync,
1160 .get_unmapped_area = thp_get_unmapped_area,
1161 .fallocate = xfs_file_fallocate,
1162 .clone_file_range = xfs_file_clone_range,
1163 .dedupe_file_range = xfs_file_dedupe_range,
1166 const struct file_operations xfs_dir_file_operations = {
1167 .open = xfs_dir_open,
1168 .read = generic_read_dir,
1169 .iterate_shared = xfs_file_readdir,
1170 .llseek = generic_file_llseek,
1171 .unlocked_ioctl = xfs_file_ioctl,
1172 #ifdef CONFIG_COMPAT
1173 .compat_ioctl = xfs_file_compat_ioctl,
1175 .fsync = xfs_dir_fsync,