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>
47 #include <linux/mman.h>
49 static const struct vm_operations_struct xfs_file_vm_ops;
52 * Clear the specified ranges to zero through either the pagecache or DAX.
53 * Holes and unwritten extents will be left as-is as they already are zeroed.
62 return iomap_zero_range(VFS_I(ip), pos, count, did_zero, &xfs_iomap_ops);
66 xfs_update_prealloc_flags(
68 enum xfs_prealloc_flags flags)
73 error = xfs_trans_alloc(ip->i_mount, &M_RES(ip->i_mount)->tr_writeid,
78 xfs_ilock(ip, XFS_ILOCK_EXCL);
79 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
81 if (!(flags & XFS_PREALLOC_INVISIBLE)) {
82 VFS_I(ip)->i_mode &= ~S_ISUID;
83 if (VFS_I(ip)->i_mode & S_IXGRP)
84 VFS_I(ip)->i_mode &= ~S_ISGID;
85 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
88 if (flags & XFS_PREALLOC_SET)
89 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
90 if (flags & XFS_PREALLOC_CLEAR)
91 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
93 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
94 if (flags & XFS_PREALLOC_SYNC)
95 xfs_trans_set_sync(tp);
96 return xfs_trans_commit(tp);
100 * Fsync operations on directories are much simpler than on regular files,
101 * as there is no file data to flush, and thus also no need for explicit
102 * cache flush operations, and there are no non-transaction metadata updates
103 * on directories either.
112 struct xfs_inode *ip = XFS_I(file->f_mapping->host);
113 struct xfs_mount *mp = ip->i_mount;
116 trace_xfs_dir_fsync(ip);
118 xfs_ilock(ip, XFS_ILOCK_SHARED);
119 if (xfs_ipincount(ip))
120 lsn = ip->i_itemp->ili_last_lsn;
121 xfs_iunlock(ip, XFS_ILOCK_SHARED);
125 return _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, NULL);
135 struct inode *inode = file->f_mapping->host;
136 struct xfs_inode *ip = XFS_I(inode);
137 struct xfs_mount *mp = ip->i_mount;
142 trace_xfs_file_fsync(ip);
144 error = file_write_and_wait_range(file, start, end);
148 if (XFS_FORCED_SHUTDOWN(mp))
151 xfs_iflags_clear(ip, XFS_ITRUNCATED);
154 * If we have an RT and/or log subvolume we need to make sure to flush
155 * the write cache the device used for file data first. This is to
156 * ensure newly written file data make it to disk before logging the new
157 * inode size in case of an extending write.
159 if (XFS_IS_REALTIME_INODE(ip))
160 xfs_blkdev_issue_flush(mp->m_rtdev_targp);
161 else if (mp->m_logdev_targp != mp->m_ddev_targp)
162 xfs_blkdev_issue_flush(mp->m_ddev_targp);
165 * All metadata updates are logged, which means that we just have to
166 * flush the log up to the latest LSN that touched the inode. If we have
167 * concurrent fsync/fdatasync() calls, we need them to all block on the
168 * log force before we clear the ili_fsync_fields field. This ensures
169 * that we don't get a racing sync operation that does not wait for the
170 * metadata to hit the journal before returning. If we race with
171 * clearing the ili_fsync_fields, then all that will happen is the log
172 * force will do nothing as the lsn will already be on disk. We can't
173 * race with setting ili_fsync_fields because that is done under
174 * XFS_ILOCK_EXCL, and that can't happen because we hold the lock shared
175 * until after the ili_fsync_fields is cleared.
177 xfs_ilock(ip, XFS_ILOCK_SHARED);
178 if (xfs_ipincount(ip)) {
180 (ip->i_itemp->ili_fsync_fields & ~XFS_ILOG_TIMESTAMP))
181 lsn = ip->i_itemp->ili_last_lsn;
185 error = _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, &log_flushed);
186 ip->i_itemp->ili_fsync_fields = 0;
188 xfs_iunlock(ip, XFS_ILOCK_SHARED);
191 * If we only have a single device, and the log force about was
192 * a no-op we might have to flush the data device cache here.
193 * This can only happen for fdatasync/O_DSYNC if we were overwriting
194 * an already allocated file and thus do not have any metadata to
197 if (!log_flushed && !XFS_IS_REALTIME_INODE(ip) &&
198 mp->m_logdev_targp == mp->m_ddev_targp)
199 xfs_blkdev_issue_flush(mp->m_ddev_targp);
205 xfs_file_dio_aio_read(
209 struct xfs_inode *ip = XFS_I(file_inode(iocb->ki_filp));
210 size_t count = iov_iter_count(to);
213 trace_xfs_file_direct_read(ip, count, iocb->ki_pos);
216 return 0; /* skip atime */
218 file_accessed(iocb->ki_filp);
220 xfs_ilock(ip, XFS_IOLOCK_SHARED);
221 ret = iomap_dio_rw(iocb, to, &xfs_iomap_ops, NULL);
222 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
227 static noinline ssize_t
232 struct xfs_inode *ip = XFS_I(iocb->ki_filp->f_mapping->host);
233 size_t count = iov_iter_count(to);
236 trace_xfs_file_dax_read(ip, count, iocb->ki_pos);
239 return 0; /* skip atime */
241 if (iocb->ki_flags & IOCB_NOWAIT) {
242 if (!xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED))
245 xfs_ilock(ip, XFS_IOLOCK_SHARED);
248 ret = dax_iomap_rw(iocb, to, &xfs_iomap_ops);
249 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
251 file_accessed(iocb->ki_filp);
256 xfs_file_buffered_aio_read(
260 struct xfs_inode *ip = XFS_I(file_inode(iocb->ki_filp));
263 trace_xfs_file_buffered_read(ip, iov_iter_count(to), iocb->ki_pos);
265 if (iocb->ki_flags & IOCB_NOWAIT) {
266 if (!xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED))
269 xfs_ilock(ip, XFS_IOLOCK_SHARED);
271 ret = generic_file_read_iter(iocb, to);
272 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
282 struct inode *inode = file_inode(iocb->ki_filp);
283 struct xfs_mount *mp = XFS_I(inode)->i_mount;
286 XFS_STATS_INC(mp, xs_read_calls);
288 if (XFS_FORCED_SHUTDOWN(mp))
292 ret = xfs_file_dax_read(iocb, to);
293 else if (iocb->ki_flags & IOCB_DIRECT)
294 ret = xfs_file_dio_aio_read(iocb, to);
296 ret = xfs_file_buffered_aio_read(iocb, to);
299 XFS_STATS_ADD(mp, xs_read_bytes, ret);
304 * Zero any on disk space between the current EOF and the new, larger EOF.
306 * This handles the normal case of zeroing the remainder of the last block in
307 * the file and the unusual case of zeroing blocks out beyond the size of the
308 * file. This second case only happens with fixed size extents and when the
309 * system crashes before the inode size was updated but after blocks were
312 * Expects the iolock to be held exclusive, and will take the ilock internally.
314 int /* error (positive) */
316 struct xfs_inode *ip,
317 xfs_off_t offset, /* starting I/O offset */
318 xfs_fsize_t isize, /* current inode size */
321 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
322 ASSERT(offset > isize);
324 trace_xfs_zero_eof(ip, isize, offset - isize);
325 return xfs_zero_range(ip, isize, offset - isize, did_zeroing);
329 * Common pre-write limit and setup checks.
331 * Called with the iolocked held either shared and exclusive according to
332 * @iolock, and returns with it held. Might upgrade the iolock to exclusive
333 * if called for a direct write beyond i_size.
336 xfs_file_aio_write_checks(
338 struct iov_iter *from,
341 struct file *file = iocb->ki_filp;
342 struct inode *inode = file->f_mapping->host;
343 struct xfs_inode *ip = XFS_I(inode);
345 size_t count = iov_iter_count(from);
346 bool drained_dio = false;
349 error = generic_write_checks(iocb, from);
353 error = xfs_break_layouts(inode, iolock);
358 * For changing security info in file_remove_privs() we need i_rwsem
361 if (*iolock == XFS_IOLOCK_SHARED && !IS_NOSEC(inode)) {
362 xfs_iunlock(ip, *iolock);
363 *iolock = XFS_IOLOCK_EXCL;
364 xfs_ilock(ip, *iolock);
368 * If the offset is beyond the size of the file, we need to zero any
369 * blocks that fall between the existing EOF and the start of this
370 * write. If zeroing is needed and we are currently holding the
371 * iolock shared, we need to update it to exclusive which implies
372 * having to redo all checks before.
374 * We need to serialise against EOF updates that occur in IO
375 * completions here. We want to make sure that nobody is changing the
376 * size while we do this check until we have placed an IO barrier (i.e.
377 * hold the XFS_IOLOCK_EXCL) that prevents new IO from being dispatched.
378 * The spinlock effectively forms a memory barrier once we have the
379 * XFS_IOLOCK_EXCL so we are guaranteed to see the latest EOF value
380 * and hence be able to correctly determine if we need to run zeroing.
382 spin_lock(&ip->i_flags_lock);
383 if (iocb->ki_pos > i_size_read(inode)) {
384 spin_unlock(&ip->i_flags_lock);
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);
404 error = xfs_zero_eof(ip, iocb->ki_pos, i_size_read(inode), NULL);
408 spin_unlock(&ip->i_flags_lock);
411 * Updating the timestamps will grab the ilock again from
412 * xfs_fs_dirty_inode, so we have to call it after dropping the
413 * lock above. Eventually we should look into a way to avoid
414 * the pointless lock roundtrip.
416 if (likely(!(file->f_mode & FMODE_NOCMTIME))) {
417 error = file_update_time(file);
423 * If we're writing the file then make sure to clear the setuid and
424 * setgid bits if the process is not being run by root. This keeps
425 * people from modifying setuid and setgid binaries.
427 if (!IS_NOSEC(inode))
428 return file_remove_privs(file);
433 xfs_dio_write_end_io(
438 struct inode *inode = file_inode(iocb->ki_filp);
439 struct xfs_inode *ip = XFS_I(inode);
440 loff_t offset = iocb->ki_pos;
443 trace_xfs_end_io_direct_write(ip, offset, size);
445 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
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 return xfs_iomap_write_unwritten(ip, offset, size, true);
467 * We need to update the in-core inode size here so that we don't end up
468 * with the on-disk inode size being outside the in-core inode size. We
469 * have no other method of updating EOF for AIO, so always do it here
472 * We need to lock the test/set EOF update as we can be racing with
473 * other IO completions here to update the EOF. Failing to serialise
474 * here can result in EOF moving backwards and Bad Things Happen when
477 spin_lock(&ip->i_flags_lock);
478 if (offset + size > i_size_read(inode)) {
479 i_size_write(inode, offset + size);
480 spin_unlock(&ip->i_flags_lock);
481 error = xfs_setfilesize(ip, offset, size);
483 spin_unlock(&ip->i_flags_lock);
490 * xfs_file_dio_aio_write - handle direct IO writes
492 * Lock the inode appropriately to prepare for and issue a direct IO write.
493 * By separating it from the buffered write path we remove all the tricky to
494 * follow locking changes and looping.
496 * If there are cached pages or we're extending the file, we need IOLOCK_EXCL
497 * until we're sure the bytes at the new EOF have been zeroed and/or the cached
498 * pages are flushed out.
500 * In most cases the direct IO writes will be done holding IOLOCK_SHARED
501 * allowing them to be done in parallel with reads and other direct IO writes.
502 * However, if the IO is not aligned to filesystem blocks, the direct IO layer
503 * needs to do sub-block zeroing and that requires serialisation against other
504 * direct IOs to the same block. In this case we need to serialise the
505 * submission of the unaligned IOs so that we don't get racing block zeroing in
506 * the dio layer. To avoid the problem with aio, we also need to wait for
507 * outstanding IOs to complete so that unwritten extent conversion is completed
508 * before we try to map the overlapping block. This is currently implemented by
509 * hitting it with a big hammer (i.e. inode_dio_wait()).
511 * Returns with locks held indicated by @iolock and errors indicated by
512 * negative return values.
515 xfs_file_dio_aio_write(
517 struct iov_iter *from)
519 struct file *file = iocb->ki_filp;
520 struct address_space *mapping = file->f_mapping;
521 struct inode *inode = mapping->host;
522 struct xfs_inode *ip = XFS_I(inode);
523 struct xfs_mount *mp = ip->i_mount;
525 int unaligned_io = 0;
527 size_t count = iov_iter_count(from);
528 struct xfs_buftarg *target = XFS_IS_REALTIME_INODE(ip) ?
529 mp->m_rtdev_targp : mp->m_ddev_targp;
531 /* DIO must be aligned to device logical sector size */
532 if ((iocb->ki_pos | count) & target->bt_logical_sectormask)
536 * Don't take the exclusive iolock here unless the I/O is unaligned to
537 * the file system block size. We don't need to consider the EOF
538 * extension case here because xfs_file_aio_write_checks() will relock
539 * the inode as necessary for EOF zeroing cases and fill out the new
540 * inode size as appropriate.
542 if ((iocb->ki_pos & mp->m_blockmask) ||
543 ((iocb->ki_pos + count) & mp->m_blockmask)) {
547 * We can't properly handle unaligned direct I/O to reflink
548 * files yet, as we can't unshare a partial block.
550 if (xfs_is_reflink_inode(ip)) {
551 trace_xfs_reflink_bounce_dio_write(ip, iocb->ki_pos, count);
554 iolock = XFS_IOLOCK_EXCL;
556 iolock = XFS_IOLOCK_SHARED;
559 if (iocb->ki_flags & IOCB_NOWAIT) {
560 if (!xfs_ilock_nowait(ip, iolock))
563 xfs_ilock(ip, iolock);
566 ret = xfs_file_aio_write_checks(iocb, from, &iolock);
569 count = iov_iter_count(from);
572 * If we are doing unaligned IO, wait for all other IO to drain,
573 * otherwise demote the lock if we had to take the exclusive lock
574 * for other reasons in xfs_file_aio_write_checks.
577 /* If we are going to wait for other DIO to finish, bail */
578 if (iocb->ki_flags & IOCB_NOWAIT) {
579 if (atomic_read(&inode->i_dio_count))
582 inode_dio_wait(inode);
584 } else if (iolock == XFS_IOLOCK_EXCL) {
585 xfs_ilock_demote(ip, XFS_IOLOCK_EXCL);
586 iolock = XFS_IOLOCK_SHARED;
589 trace_xfs_file_direct_write(ip, count, iocb->ki_pos);
590 ret = iomap_dio_rw(iocb, from, &xfs_iomap_ops, xfs_dio_write_end_io);
592 xfs_iunlock(ip, iolock);
595 * No fallback to buffered IO on errors for XFS, direct IO will either
596 * complete fully or fail.
598 ASSERT(ret < 0 || ret == count);
602 static noinline ssize_t
605 struct iov_iter *from)
607 struct inode *inode = iocb->ki_filp->f_mapping->host;
608 struct xfs_inode *ip = XFS_I(inode);
609 int iolock = XFS_IOLOCK_EXCL;
610 ssize_t ret, error = 0;
614 if (iocb->ki_flags & IOCB_NOWAIT) {
615 if (!xfs_ilock_nowait(ip, iolock))
618 xfs_ilock(ip, iolock);
621 ret = xfs_file_aio_write_checks(iocb, from, &iolock);
626 count = iov_iter_count(from);
628 trace_xfs_file_dax_write(ip, count, pos);
629 ret = dax_iomap_rw(iocb, from, &xfs_iomap_ops);
630 if (ret > 0 && iocb->ki_pos > i_size_read(inode)) {
631 i_size_write(inode, iocb->ki_pos);
632 error = xfs_setfilesize(ip, pos, ret);
635 xfs_iunlock(ip, iolock);
636 return error ? error : ret;
640 xfs_file_buffered_aio_write(
642 struct iov_iter *from)
644 struct file *file = iocb->ki_filp;
645 struct address_space *mapping = file->f_mapping;
646 struct inode *inode = mapping->host;
647 struct xfs_inode *ip = XFS_I(inode);
652 if (iocb->ki_flags & IOCB_NOWAIT)
656 iolock = XFS_IOLOCK_EXCL;
657 xfs_ilock(ip, iolock);
659 ret = xfs_file_aio_write_checks(iocb, from, &iolock);
663 /* We can write back this queue in page reclaim */
664 current->backing_dev_info = inode_to_bdi(inode);
666 trace_xfs_file_buffered_write(ip, iov_iter_count(from), iocb->ki_pos);
667 ret = iomap_file_buffered_write(iocb, from, &xfs_iomap_ops);
668 if (likely(ret >= 0))
672 * If we hit a space limit, try to free up some lingering preallocated
673 * space before returning an error. In the case of ENOSPC, first try to
674 * write back all dirty inodes to free up some of the excess reserved
675 * metadata space. This reduces the chances that the eofblocks scan
676 * waits on dirty mappings. Since xfs_flush_inodes() is serialized, this
677 * also behaves as a filter to prevent too many eofblocks scans from
678 * running at the same time.
680 if (ret == -EDQUOT && !enospc) {
681 xfs_iunlock(ip, iolock);
682 enospc = xfs_inode_free_quota_eofblocks(ip);
685 enospc = xfs_inode_free_quota_cowblocks(ip);
689 } else if (ret == -ENOSPC && !enospc) {
690 struct xfs_eofblocks eofb = {0};
693 xfs_flush_inodes(ip->i_mount);
695 xfs_iunlock(ip, iolock);
696 eofb.eof_flags = XFS_EOF_FLAGS_SYNC;
697 xfs_icache_free_eofblocks(ip->i_mount, &eofb);
698 xfs_icache_free_cowblocks(ip->i_mount, &eofb);
702 current->backing_dev_info = NULL;
705 xfs_iunlock(ip, iolock);
712 struct iov_iter *from)
714 struct file *file = iocb->ki_filp;
715 struct address_space *mapping = file->f_mapping;
716 struct inode *inode = mapping->host;
717 struct xfs_inode *ip = XFS_I(inode);
719 size_t ocount = iov_iter_count(from);
721 XFS_STATS_INC(ip->i_mount, xs_write_calls);
726 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
730 ret = xfs_file_dax_write(iocb, from);
731 else if (iocb->ki_flags & IOCB_DIRECT) {
733 * Allow a directio write to fall back to a buffered
734 * write *only* in the case that we're doing a reflink
735 * CoW. In all other directio scenarios we do not
736 * allow an operation to fall back to buffered mode.
738 ret = xfs_file_dio_aio_write(iocb, from);
743 ret = xfs_file_buffered_aio_write(iocb, from);
747 XFS_STATS_ADD(ip->i_mount, xs_write_bytes, ret);
749 /* Handle various SYNC-type writes */
750 ret = generic_write_sync(iocb, ret);
755 #define XFS_FALLOC_FL_SUPPORTED \
756 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
757 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE | \
758 FALLOC_FL_INSERT_RANGE | FALLOC_FL_UNSHARE_RANGE)
767 struct inode *inode = file_inode(file);
768 struct xfs_inode *ip = XFS_I(inode);
770 enum xfs_prealloc_flags flags = 0;
771 uint iolock = XFS_IOLOCK_EXCL;
773 bool do_file_insert = false;
775 if (!S_ISREG(inode->i_mode))
777 if (mode & ~XFS_FALLOC_FL_SUPPORTED)
780 xfs_ilock(ip, iolock);
781 error = xfs_break_layouts(inode, &iolock);
785 xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
786 iolock |= XFS_MMAPLOCK_EXCL;
788 if (mode & FALLOC_FL_PUNCH_HOLE) {
789 error = xfs_free_file_space(ip, offset, len);
792 } else if (mode & FALLOC_FL_COLLAPSE_RANGE) {
793 unsigned int blksize_mask = i_blocksize(inode) - 1;
795 if (offset & blksize_mask || len & blksize_mask) {
801 * There is no need to overlap collapse range with EOF,
802 * in which case it is effectively a truncate operation
804 if (offset + len >= i_size_read(inode)) {
809 new_size = i_size_read(inode) - len;
811 error = xfs_collapse_file_space(ip, offset, len);
814 } else if (mode & FALLOC_FL_INSERT_RANGE) {
815 unsigned int blksize_mask = i_blocksize(inode) - 1;
817 new_size = i_size_read(inode) + len;
818 if (offset & blksize_mask || len & blksize_mask) {
823 /* check the new inode size does not wrap through zero */
824 if (new_size > inode->i_sb->s_maxbytes) {
829 /* Offset should be less than i_size */
830 if (offset >= i_size_read(inode)) {
834 do_file_insert = true;
836 flags |= XFS_PREALLOC_SET;
838 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
839 offset + len > i_size_read(inode)) {
840 new_size = offset + len;
841 error = inode_newsize_ok(inode, new_size);
846 if (mode & FALLOC_FL_ZERO_RANGE)
847 error = xfs_zero_file_space(ip, offset, len);
849 if (mode & FALLOC_FL_UNSHARE_RANGE) {
850 error = xfs_reflink_unshare(ip, offset, len);
854 error = xfs_alloc_file_space(ip, offset, len,
861 if (file->f_flags & O_DSYNC)
862 flags |= XFS_PREALLOC_SYNC;
864 error = xfs_update_prealloc_flags(ip, flags);
868 /* Change file size if needed */
872 iattr.ia_valid = ATTR_SIZE;
873 iattr.ia_size = new_size;
874 error = xfs_vn_setattr_size(file_dentry(file), &iattr);
880 * Perform hole insertion now that the file size has been
881 * updated so that if we crash during the operation we don't
882 * leave shifted extents past EOF and hence losing access to
883 * the data that is contained within them.
886 error = xfs_insert_file_space(ip, offset, len);
889 xfs_iunlock(ip, iolock);
894 xfs_file_clone_range(
895 struct file *file_in,
897 struct file *file_out,
901 return xfs_reflink_remap_range(file_in, pos_in, file_out, pos_out,
906 xfs_file_dedupe_range(
907 struct file *src_file,
910 struct file *dst_file,
915 error = xfs_reflink_remap_range(src_file, loff, dst_file, dst_loff,
927 if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
929 if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb)))
931 file->f_mode |= FMODE_NOWAIT;
940 struct xfs_inode *ip = XFS_I(inode);
944 error = xfs_file_open(inode, file);
949 * If there are any blocks, read-ahead block 0 as we're almost
950 * certain to have the next operation be a read there.
952 mode = xfs_ilock_data_map_shared(ip);
953 if (ip->i_d.di_nextents > 0)
954 error = xfs_dir3_data_readahead(ip, 0, -1);
955 xfs_iunlock(ip, mode);
964 return xfs_release(XFS_I(inode));
970 struct dir_context *ctx)
972 struct inode *inode = file_inode(file);
973 xfs_inode_t *ip = XFS_I(inode);
977 * The Linux API doesn't pass down the total size of the buffer
978 * we read into down to the filesystem. With the filldir concept
979 * it's not needed for correct information, but the XFS dir2 leaf
980 * code wants an estimate of the buffer size to calculate it's
981 * readahead window and size the buffers used for mapping to
984 * Try to give it an estimate that's good enough, maybe at some
985 * point we can change the ->readdir prototype to include the
986 * buffer size. For now we use the current glibc buffer size.
988 bufsize = (size_t)min_t(loff_t, XFS_READDIR_BUFSIZE, ip->i_d.di_size);
990 return xfs_readdir(NULL, ip, ctx, bufsize);
999 struct inode *inode = file->f_mapping->host;
1001 if (XFS_FORCED_SHUTDOWN(XFS_I(inode)->i_mount))
1006 return generic_file_llseek(file, offset, whence);
1008 offset = iomap_seek_hole(inode, offset, &xfs_iomap_ops);
1011 offset = iomap_seek_data(inode, offset, &xfs_iomap_ops);
1017 return vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
1021 * Locking for serialisation of IO during page faults. This results in a lock
1025 * sb_start_pagefault(vfs, freeze)
1026 * i_mmaplock (XFS - truncate serialisation)
1028 * i_lock (XFS - extent map serialisation)
1031 __xfs_filemap_fault(
1032 struct vm_fault *vmf,
1033 enum page_entry_size pe_size,
1036 struct inode *inode = file_inode(vmf->vma->vm_file);
1037 struct xfs_inode *ip = XFS_I(inode);
1040 trace_xfs_filemap_fault(ip, pe_size, write_fault);
1043 sb_start_pagefault(inode->i_sb);
1044 file_update_time(vmf->vma->vm_file);
1047 xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1048 if (IS_DAX(inode)) {
1051 ret = dax_iomap_fault(vmf, pe_size, &pfn, &xfs_iomap_ops);
1052 if (ret & VM_FAULT_NEEDDSYNC)
1053 ret = dax_finish_sync_fault(vmf, pe_size, pfn);
1056 ret = iomap_page_mkwrite(vmf, &xfs_iomap_ops);
1058 ret = filemap_fault(vmf);
1060 xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1063 sb_end_pagefault(inode->i_sb);
1069 struct vm_fault *vmf)
1071 /* DAX can shortcut the normal fault path on write faults! */
1072 return __xfs_filemap_fault(vmf, PE_SIZE_PTE,
1073 IS_DAX(file_inode(vmf->vma->vm_file)) &&
1074 (vmf->flags & FAULT_FLAG_WRITE));
1078 xfs_filemap_huge_fault(
1079 struct vm_fault *vmf,
1080 enum page_entry_size pe_size)
1082 if (!IS_DAX(file_inode(vmf->vma->vm_file)))
1083 return VM_FAULT_FALLBACK;
1085 /* DAX can shortcut the normal fault path on write faults! */
1086 return __xfs_filemap_fault(vmf, pe_size,
1087 (vmf->flags & FAULT_FLAG_WRITE));
1091 xfs_filemap_page_mkwrite(
1092 struct vm_fault *vmf)
1094 return __xfs_filemap_fault(vmf, PE_SIZE_PTE, true);
1098 * pfn_mkwrite was originally intended to ensure we capture time stamp updates
1099 * on write faults. In reality, it needs to serialise against truncate and
1100 * prepare memory for writing so handle is as standard write fault.
1103 xfs_filemap_pfn_mkwrite(
1104 struct vm_fault *vmf)
1107 return __xfs_filemap_fault(vmf, PE_SIZE_PTE, true);
1110 static const struct vm_operations_struct xfs_file_vm_ops = {
1111 .fault = xfs_filemap_fault,
1112 .huge_fault = xfs_filemap_huge_fault,
1113 .map_pages = filemap_map_pages,
1114 .page_mkwrite = xfs_filemap_page_mkwrite,
1115 .pfn_mkwrite = xfs_filemap_pfn_mkwrite,
1121 struct vm_area_struct *vma)
1124 * We don't support synchronous mappings for non-DAX files. At least
1125 * until someone comes with a sensible use case.
1127 if (!IS_DAX(file_inode(filp)) && (vma->vm_flags & VM_SYNC))
1130 file_accessed(filp);
1131 vma->vm_ops = &xfs_file_vm_ops;
1132 if (IS_DAX(file_inode(filp)))
1133 vma->vm_flags |= VM_MIXEDMAP | VM_HUGEPAGE;
1137 const struct file_operations xfs_file_operations = {
1138 .llseek = xfs_file_llseek,
1139 .read_iter = xfs_file_read_iter,
1140 .write_iter = xfs_file_write_iter,
1141 .splice_read = generic_file_splice_read,
1142 .splice_write = iter_file_splice_write,
1143 .unlocked_ioctl = xfs_file_ioctl,
1144 #ifdef CONFIG_COMPAT
1145 .compat_ioctl = xfs_file_compat_ioctl,
1147 .mmap = xfs_file_mmap,
1148 .mmap_supported_flags = MAP_SYNC,
1149 .open = xfs_file_open,
1150 .release = xfs_file_release,
1151 .fsync = xfs_file_fsync,
1152 .get_unmapped_area = thp_get_unmapped_area,
1153 .fallocate = xfs_file_fallocate,
1154 .clone_file_range = xfs_file_clone_range,
1155 .dedupe_file_range = xfs_file_dedupe_range,
1158 const struct file_operations xfs_dir_file_operations = {
1159 .open = xfs_dir_open,
1160 .read = generic_read_dir,
1161 .iterate_shared = xfs_file_readdir,
1162 .llseek = generic_file_llseek,
1163 .unlocked_ioctl = xfs_file_ioctl,
1164 #ifdef CONFIG_COMPAT
1165 .compat_ioctl = xfs_file_compat_ioctl,
1167 .fsync = xfs_dir_fsync,