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>
33 static const struct vm_operations_struct xfs_file_vm_ops;
36 xfs_update_prealloc_flags(
38 enum xfs_prealloc_flags flags)
43 error = xfs_trans_alloc(ip->i_mount, &M_RES(ip->i_mount)->tr_writeid,
48 xfs_ilock(ip, XFS_ILOCK_EXCL);
49 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
51 if (!(flags & XFS_PREALLOC_INVISIBLE)) {
52 VFS_I(ip)->i_mode &= ~S_ISUID;
53 if (VFS_I(ip)->i_mode & S_IXGRP)
54 VFS_I(ip)->i_mode &= ~S_ISGID;
55 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
58 if (flags & XFS_PREALLOC_SET)
59 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
60 if (flags & XFS_PREALLOC_CLEAR)
61 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
63 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
64 if (flags & XFS_PREALLOC_SYNC)
65 xfs_trans_set_sync(tp);
66 return xfs_trans_commit(tp);
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);
83 struct xfs_mount *mp = ip->i_mount;
86 trace_xfs_dir_fsync(ip);
88 xfs_ilock(ip, XFS_ILOCK_SHARED);
89 if (xfs_ipincount(ip))
90 lsn = ip->i_itemp->ili_last_lsn;
91 xfs_iunlock(ip, XFS_ILOCK_SHARED);
95 return xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, NULL);
105 struct inode *inode = file->f_mapping->host;
106 struct xfs_inode *ip = XFS_I(inode);
107 struct xfs_mount *mp = ip->i_mount;
112 trace_xfs_file_fsync(ip);
114 error = file_write_and_wait_range(file, start, end);
118 if (XFS_FORCED_SHUTDOWN(mp))
121 xfs_iflags_clear(ip, XFS_ITRUNCATED);
124 * If we have an RT and/or log subvolume we need to make sure to flush
125 * the write cache the device used for file data first. This is to
126 * ensure newly written file data make it to disk before logging the new
127 * inode size in case of an extending write.
129 if (XFS_IS_REALTIME_INODE(ip))
130 xfs_blkdev_issue_flush(mp->m_rtdev_targp);
131 else if (mp->m_logdev_targp != mp->m_ddev_targp)
132 xfs_blkdev_issue_flush(mp->m_ddev_targp);
135 * All metadata updates are logged, which means that we just have to
136 * flush the log up to the latest LSN that touched the inode. If we have
137 * concurrent fsync/fdatasync() calls, we need them to all block on the
138 * log force before we clear the ili_fsync_fields field. This ensures
139 * that we don't get a racing sync operation that does not wait for the
140 * metadata to hit the journal before returning. If we race with
141 * clearing the ili_fsync_fields, then all that will happen is the log
142 * force will do nothing as the lsn will already be on disk. We can't
143 * race with setting ili_fsync_fields because that is done under
144 * XFS_ILOCK_EXCL, and that can't happen because we hold the lock shared
145 * until after the ili_fsync_fields is cleared.
147 xfs_ilock(ip, XFS_ILOCK_SHARED);
148 if (xfs_ipincount(ip)) {
150 (ip->i_itemp->ili_fsync_fields & ~XFS_ILOG_TIMESTAMP))
151 lsn = ip->i_itemp->ili_last_lsn;
155 error = xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, &log_flushed);
156 ip->i_itemp->ili_fsync_fields = 0;
158 xfs_iunlock(ip, XFS_ILOCK_SHARED);
161 * If we only have a single device, and the log force about was
162 * a no-op we might have to flush the data device cache here.
163 * This can only happen for fdatasync/O_DSYNC if we were overwriting
164 * an already allocated file and thus do not have any metadata to
167 if (!log_flushed && !XFS_IS_REALTIME_INODE(ip) &&
168 mp->m_logdev_targp == mp->m_ddev_targp)
169 xfs_blkdev_issue_flush(mp->m_ddev_targp);
175 xfs_file_dio_aio_read(
179 struct xfs_inode *ip = XFS_I(file_inode(iocb->ki_filp));
180 size_t count = iov_iter_count(to);
183 trace_xfs_file_direct_read(ip, count, iocb->ki_pos);
186 return 0; /* skip atime */
188 file_accessed(iocb->ki_filp);
190 xfs_ilock(ip, XFS_IOLOCK_SHARED);
191 ret = iomap_dio_rw(iocb, to, &xfs_iomap_ops, NULL, is_sync_kiocb(iocb));
192 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
197 static noinline ssize_t
202 struct xfs_inode *ip = XFS_I(iocb->ki_filp->f_mapping->host);
203 size_t count = iov_iter_count(to);
206 trace_xfs_file_dax_read(ip, count, iocb->ki_pos);
209 return 0; /* skip atime */
211 if (iocb->ki_flags & IOCB_NOWAIT) {
212 if (!xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED))
215 xfs_ilock(ip, XFS_IOLOCK_SHARED);
218 ret = dax_iomap_rw(iocb, to, &xfs_iomap_ops);
219 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
221 file_accessed(iocb->ki_filp);
226 xfs_file_buffered_aio_read(
230 struct xfs_inode *ip = XFS_I(file_inode(iocb->ki_filp));
233 trace_xfs_file_buffered_read(ip, iov_iter_count(to), iocb->ki_pos);
235 if (iocb->ki_flags & IOCB_NOWAIT) {
236 if (!xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED))
239 xfs_ilock(ip, XFS_IOLOCK_SHARED);
241 ret = generic_file_read_iter(iocb, to);
242 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
252 struct inode *inode = file_inode(iocb->ki_filp);
253 struct xfs_mount *mp = XFS_I(inode)->i_mount;
256 XFS_STATS_INC(mp, xs_read_calls);
258 if (XFS_FORCED_SHUTDOWN(mp))
262 ret = xfs_file_dax_read(iocb, to);
263 else if (iocb->ki_flags & IOCB_DIRECT)
264 ret = xfs_file_dio_aio_read(iocb, to);
266 ret = xfs_file_buffered_aio_read(iocb, to);
269 XFS_STATS_ADD(mp, xs_read_bytes, ret);
274 * Common pre-write limit and setup checks.
276 * Called with the iolocked held either shared and exclusive according to
277 * @iolock, and returns with it held. Might upgrade the iolock to exclusive
278 * if called for a direct write beyond i_size.
281 xfs_file_aio_write_checks(
283 struct iov_iter *from,
286 struct file *file = iocb->ki_filp;
287 struct inode *inode = file->f_mapping->host;
288 struct xfs_inode *ip = XFS_I(inode);
290 size_t count = iov_iter_count(from);
291 bool drained_dio = false;
295 error = generic_write_checks(iocb, from);
299 error = xfs_break_layouts(inode, iolock, BREAK_WRITE);
304 * For changing security info in file_remove_privs() we need i_rwsem
307 if (*iolock == XFS_IOLOCK_SHARED && !IS_NOSEC(inode)) {
308 xfs_iunlock(ip, *iolock);
309 *iolock = XFS_IOLOCK_EXCL;
310 xfs_ilock(ip, *iolock);
314 * If the offset is beyond the size of the file, we need to zero any
315 * blocks that fall between the existing EOF and the start of this
316 * write. If zeroing is needed and we are currently holding the
317 * iolock shared, we need to update it to exclusive which implies
318 * having to redo all checks before.
320 * We need to serialise against EOF updates that occur in IO
321 * completions here. We want to make sure that nobody is changing the
322 * size while we do this check until we have placed an IO barrier (i.e.
323 * hold the XFS_IOLOCK_EXCL) that prevents new IO from being dispatched.
324 * The spinlock effectively forms a memory barrier once we have the
325 * XFS_IOLOCK_EXCL so we are guaranteed to see the latest EOF value
326 * and hence be able to correctly determine if we need to run zeroing.
328 spin_lock(&ip->i_flags_lock);
329 isize = i_size_read(inode);
330 if (iocb->ki_pos > isize) {
331 spin_unlock(&ip->i_flags_lock);
333 if (*iolock == XFS_IOLOCK_SHARED) {
334 xfs_iunlock(ip, *iolock);
335 *iolock = XFS_IOLOCK_EXCL;
336 xfs_ilock(ip, *iolock);
337 iov_iter_reexpand(from, count);
340 * We now have an IO submission barrier in place, but
341 * AIO can do EOF updates during IO completion and hence
342 * we now need to wait for all of them to drain. Non-AIO
343 * DIO will have drained before we are given the
344 * XFS_IOLOCK_EXCL, and so for most cases this wait is a
347 inode_dio_wait(inode);
352 trace_xfs_zero_eof(ip, isize, iocb->ki_pos - isize);
353 error = iomap_zero_range(inode, isize, iocb->ki_pos - isize,
354 NULL, &xfs_iomap_ops);
358 spin_unlock(&ip->i_flags_lock);
361 * Updating the timestamps will grab the ilock again from
362 * xfs_fs_dirty_inode, so we have to call it after dropping the
363 * lock above. Eventually we should look into a way to avoid
364 * the pointless lock roundtrip.
366 return file_modified(file);
370 xfs_dio_write_end_io(
376 struct inode *inode = file_inode(iocb->ki_filp);
377 struct xfs_inode *ip = XFS_I(inode);
378 loff_t offset = iocb->ki_pos;
379 unsigned int nofs_flag;
381 trace_xfs_end_io_direct_write(ip, offset, size);
383 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
392 * Capture amount written on completion as we can't reliably account
393 * for it on submission.
395 XFS_STATS_ADD(ip->i_mount, xs_write_bytes, size);
398 * We can allocate memory here while doing writeback on behalf of
399 * memory reclaim. To avoid memory allocation deadlocks set the
400 * task-wide nofs context for the following operations.
402 nofs_flag = memalloc_nofs_save();
404 if (flags & IOMAP_DIO_COW) {
405 error = xfs_reflink_end_cow(ip, offset, size);
411 * Unwritten conversion updates the in-core isize after extent
412 * conversion but before updating the on-disk size. Updating isize any
413 * earlier allows a racing dio read to find unwritten extents before
414 * they are converted.
416 if (flags & IOMAP_DIO_UNWRITTEN) {
417 error = xfs_iomap_write_unwritten(ip, offset, size, true);
422 * We need to update the in-core inode size here so that we don't end up
423 * with the on-disk inode size being outside the in-core inode size. We
424 * have no other method of updating EOF for AIO, so always do it here
427 * We need to lock the test/set EOF update as we can be racing with
428 * other IO completions here to update the EOF. Failing to serialise
429 * here can result in EOF moving backwards and Bad Things Happen when
432 spin_lock(&ip->i_flags_lock);
433 if (offset + size > i_size_read(inode)) {
434 i_size_write(inode, offset + size);
435 spin_unlock(&ip->i_flags_lock);
436 error = xfs_setfilesize(ip, offset, size);
438 spin_unlock(&ip->i_flags_lock);
442 memalloc_nofs_restore(nofs_flag);
446 static const struct iomap_dio_ops xfs_dio_write_ops = {
447 .end_io = xfs_dio_write_end_io,
451 * xfs_file_dio_aio_write - handle direct IO writes
453 * Lock the inode appropriately to prepare for and issue a direct IO write.
454 * By separating it from the buffered write path we remove all the tricky to
455 * follow locking changes and looping.
457 * If there are cached pages or we're extending the file, we need IOLOCK_EXCL
458 * until we're sure the bytes at the new EOF have been zeroed and/or the cached
459 * pages are flushed out.
461 * In most cases the direct IO writes will be done holding IOLOCK_SHARED
462 * allowing them to be done in parallel with reads and other direct IO writes.
463 * However, if the IO is not aligned to filesystem blocks, the direct IO layer
464 * needs to do sub-block zeroing and that requires serialisation against other
465 * direct IOs to the same block. In this case we need to serialise the
466 * submission of the unaligned IOs so that we don't get racing block zeroing in
467 * the dio layer. To avoid the problem with aio, we also need to wait for
468 * outstanding IOs to complete so that unwritten extent conversion is completed
469 * before we try to map the overlapping block. This is currently implemented by
470 * hitting it with a big hammer (i.e. inode_dio_wait()).
472 * Returns with locks held indicated by @iolock and errors indicated by
473 * negative return values.
476 xfs_file_dio_aio_write(
478 struct iov_iter *from)
480 struct file *file = iocb->ki_filp;
481 struct address_space *mapping = file->f_mapping;
482 struct inode *inode = mapping->host;
483 struct xfs_inode *ip = XFS_I(inode);
484 struct xfs_mount *mp = ip->i_mount;
486 int unaligned_io = 0;
488 size_t count = iov_iter_count(from);
489 struct xfs_buftarg *target = XFS_IS_REALTIME_INODE(ip) ?
490 mp->m_rtdev_targp : mp->m_ddev_targp;
492 /* DIO must be aligned to device logical sector size */
493 if ((iocb->ki_pos | count) & target->bt_logical_sectormask)
497 * Don't take the exclusive iolock here unless the I/O is unaligned to
498 * the file system block size. We don't need to consider the EOF
499 * extension case here because xfs_file_aio_write_checks() will relock
500 * the inode as necessary for EOF zeroing cases and fill out the new
501 * inode size as appropriate.
503 if ((iocb->ki_pos & mp->m_blockmask) ||
504 ((iocb->ki_pos + count) & mp->m_blockmask)) {
508 * We can't properly handle unaligned direct I/O to reflink
509 * files yet, as we can't unshare a partial block.
511 if (xfs_is_cow_inode(ip)) {
512 trace_xfs_reflink_bounce_dio_write(ip, iocb->ki_pos, count);
515 iolock = XFS_IOLOCK_EXCL;
517 iolock = XFS_IOLOCK_SHARED;
520 if (iocb->ki_flags & IOCB_NOWAIT) {
521 /* unaligned dio always waits, bail */
524 if (!xfs_ilock_nowait(ip, iolock))
527 xfs_ilock(ip, iolock);
530 ret = xfs_file_aio_write_checks(iocb, from, &iolock);
533 count = iov_iter_count(from);
536 * If we are doing unaligned IO, we can't allow any other overlapping IO
537 * in-flight at the same time or we risk data corruption. Wait for all
538 * other IO to drain before we submit. If the IO is aligned, demote the
539 * iolock if we had to take the exclusive lock in
540 * xfs_file_aio_write_checks() for other reasons.
543 inode_dio_wait(inode);
544 } else if (iolock == XFS_IOLOCK_EXCL) {
545 xfs_ilock_demote(ip, XFS_IOLOCK_EXCL);
546 iolock = XFS_IOLOCK_SHARED;
549 trace_xfs_file_direct_write(ip, count, iocb->ki_pos);
551 * If unaligned, this is the only IO in-flight. Wait on it before we
552 * release the iolock to prevent subsequent overlapping IO.
554 ret = iomap_dio_rw(iocb, from, &xfs_iomap_ops, &xfs_dio_write_ops,
555 is_sync_kiocb(iocb) || unaligned_io);
557 xfs_iunlock(ip, iolock);
560 * No fallback to buffered IO on errors for XFS, direct IO will either
561 * complete fully or fail.
563 ASSERT(ret < 0 || ret == count);
567 static noinline ssize_t
570 struct iov_iter *from)
572 struct inode *inode = iocb->ki_filp->f_mapping->host;
573 struct xfs_inode *ip = XFS_I(inode);
574 int iolock = XFS_IOLOCK_EXCL;
575 ssize_t ret, error = 0;
579 if (iocb->ki_flags & IOCB_NOWAIT) {
580 if (!xfs_ilock_nowait(ip, iolock))
583 xfs_ilock(ip, iolock);
586 ret = xfs_file_aio_write_checks(iocb, from, &iolock);
591 count = iov_iter_count(from);
593 trace_xfs_file_dax_write(ip, count, pos);
594 ret = dax_iomap_rw(iocb, from, &xfs_iomap_ops);
595 if (ret > 0 && iocb->ki_pos > i_size_read(inode)) {
596 i_size_write(inode, iocb->ki_pos);
597 error = xfs_setfilesize(ip, pos, ret);
600 xfs_iunlock(ip, iolock);
605 XFS_STATS_ADD(ip->i_mount, xs_write_bytes, ret);
607 /* Handle various SYNC-type writes */
608 ret = generic_write_sync(iocb, ret);
614 xfs_file_buffered_aio_write(
616 struct iov_iter *from)
618 struct file *file = iocb->ki_filp;
619 struct address_space *mapping = file->f_mapping;
620 struct inode *inode = mapping->host;
621 struct xfs_inode *ip = XFS_I(inode);
626 if (iocb->ki_flags & IOCB_NOWAIT)
630 iolock = XFS_IOLOCK_EXCL;
631 xfs_ilock(ip, iolock);
633 ret = xfs_file_aio_write_checks(iocb, from, &iolock);
637 /* We can write back this queue in page reclaim */
638 current->backing_dev_info = inode_to_bdi(inode);
640 trace_xfs_file_buffered_write(ip, iov_iter_count(from), iocb->ki_pos);
641 ret = iomap_file_buffered_write(iocb, from, &xfs_iomap_ops);
642 if (likely(ret >= 0))
646 * If we hit a space limit, try to free up some lingering preallocated
647 * space before returning an error. In the case of ENOSPC, first try to
648 * write back all dirty inodes to free up some of the excess reserved
649 * metadata space. This reduces the chances that the eofblocks scan
650 * waits on dirty mappings. Since xfs_flush_inodes() is serialized, this
651 * also behaves as a filter to prevent too many eofblocks scans from
652 * running at the same time.
654 if (ret == -EDQUOT && !enospc) {
655 xfs_iunlock(ip, iolock);
656 enospc = xfs_inode_free_quota_eofblocks(ip);
659 enospc = xfs_inode_free_quota_cowblocks(ip);
663 } else if (ret == -ENOSPC && !enospc) {
664 struct xfs_eofblocks eofb = {0};
667 xfs_flush_inodes(ip->i_mount);
669 xfs_iunlock(ip, iolock);
670 eofb.eof_flags = XFS_EOF_FLAGS_SYNC;
671 xfs_icache_free_eofblocks(ip->i_mount, &eofb);
672 xfs_icache_free_cowblocks(ip->i_mount, &eofb);
676 current->backing_dev_info = NULL;
679 xfs_iunlock(ip, iolock);
682 XFS_STATS_ADD(ip->i_mount, xs_write_bytes, ret);
683 /* Handle various SYNC-type writes */
684 ret = generic_write_sync(iocb, ret);
692 struct iov_iter *from)
694 struct file *file = iocb->ki_filp;
695 struct address_space *mapping = file->f_mapping;
696 struct inode *inode = mapping->host;
697 struct xfs_inode *ip = XFS_I(inode);
699 size_t ocount = iov_iter_count(from);
701 XFS_STATS_INC(ip->i_mount, xs_write_calls);
706 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
710 return xfs_file_dax_write(iocb, from);
712 if (iocb->ki_flags & IOCB_DIRECT) {
714 * Allow a directio write to fall back to a buffered
715 * write *only* in the case that we're doing a reflink
716 * CoW. In all other directio scenarios we do not
717 * allow an operation to fall back to buffered mode.
719 ret = xfs_file_dio_aio_write(iocb, from);
724 return xfs_file_buffered_aio_write(iocb, from);
731 struct xfs_inode *ip = XFS_I(inode);
733 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
735 xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
739 xfs_break_dax_layouts(
745 ASSERT(xfs_isilocked(XFS_I(inode), XFS_MMAPLOCK_EXCL));
747 page = dax_layout_busy_page(inode->i_mapping);
752 return ___wait_var_event(&page->_refcount,
753 atomic_read(&page->_refcount) == 1, TASK_INTERRUPTIBLE,
754 0, 0, xfs_wait_dax_page(inode));
761 enum layout_break_reason reason)
766 ASSERT(xfs_isilocked(XFS_I(inode), XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL));
772 error = xfs_break_dax_layouts(inode, &retry);
777 error = xfs_break_leased_layouts(inode, iolock, &retry);
783 } while (error == 0 && retry);
788 #define XFS_FALLOC_FL_SUPPORTED \
789 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
790 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE | \
791 FALLOC_FL_INSERT_RANGE | FALLOC_FL_UNSHARE_RANGE)
800 struct inode *inode = file_inode(file);
801 struct xfs_inode *ip = XFS_I(inode);
803 enum xfs_prealloc_flags flags = 0;
804 uint iolock = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL;
806 bool do_file_insert = false;
808 if (!S_ISREG(inode->i_mode))
810 if (mode & ~XFS_FALLOC_FL_SUPPORTED)
813 xfs_ilock(ip, iolock);
814 error = xfs_break_layouts(inode, &iolock, BREAK_UNMAP);
818 if (mode & FALLOC_FL_PUNCH_HOLE) {
819 error = xfs_free_file_space(ip, offset, len);
822 } else if (mode & FALLOC_FL_COLLAPSE_RANGE) {
823 unsigned int blksize_mask = i_blocksize(inode) - 1;
825 if (offset & blksize_mask || len & blksize_mask) {
831 * There is no need to overlap collapse range with EOF,
832 * in which case it is effectively a truncate operation
834 if (offset + len >= i_size_read(inode)) {
839 new_size = i_size_read(inode) - len;
841 error = xfs_collapse_file_space(ip, offset, len);
844 } else if (mode & FALLOC_FL_INSERT_RANGE) {
845 unsigned int blksize_mask = i_blocksize(inode) - 1;
846 loff_t isize = i_size_read(inode);
848 if (offset & blksize_mask || len & blksize_mask) {
854 * New inode size must not exceed ->s_maxbytes, accounting for
855 * possible signed overflow.
857 if (inode->i_sb->s_maxbytes - isize < len) {
861 new_size = isize + len;
863 /* Offset should be less than i_size */
864 if (offset >= isize) {
868 do_file_insert = true;
870 flags |= XFS_PREALLOC_SET;
872 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
873 offset + len > i_size_read(inode)) {
874 new_size = offset + len;
875 error = inode_newsize_ok(inode, new_size);
880 if (mode & FALLOC_FL_ZERO_RANGE) {
881 error = xfs_zero_file_space(ip, offset, len);
882 } else if (mode & FALLOC_FL_UNSHARE_RANGE) {
883 error = xfs_reflink_unshare(ip, offset, len);
887 if (!xfs_is_always_cow_inode(ip)) {
888 error = xfs_alloc_file_space(ip, offset, len,
893 * If always_cow mode we can't use preallocations and
894 * thus should not create them.
896 if (xfs_is_always_cow_inode(ip)) {
901 error = xfs_alloc_file_space(ip, offset, len,
908 if (file->f_flags & O_DSYNC)
909 flags |= XFS_PREALLOC_SYNC;
911 error = xfs_update_prealloc_flags(ip, flags);
915 /* Change file size if needed */
919 iattr.ia_valid = ATTR_SIZE;
920 iattr.ia_size = new_size;
921 error = xfs_vn_setattr_size(file_dentry(file), &iattr);
927 * Perform hole insertion now that the file size has been
928 * updated so that if we crash during the operation we don't
929 * leave shifted extents past EOF and hence losing access to
930 * the data that is contained within them.
933 error = xfs_insert_file_space(ip, offset, len);
936 xfs_iunlock(ip, iolock);
947 struct xfs_inode *ip = XFS_I(file_inode(file));
952 * Operations creating pages in page cache need protection from hole
953 * punching and similar ops
955 if (advice == POSIX_FADV_WILLNEED) {
956 lockflags = XFS_IOLOCK_SHARED;
957 xfs_ilock(ip, lockflags);
959 ret = generic_fadvise(file, start, end, advice);
961 xfs_iunlock(ip, lockflags);
966 xfs_file_remap_range(
967 struct file *file_in,
969 struct file *file_out,
972 unsigned int remap_flags)
974 struct inode *inode_in = file_inode(file_in);
975 struct xfs_inode *src = XFS_I(inode_in);
976 struct inode *inode_out = file_inode(file_out);
977 struct xfs_inode *dest = XFS_I(inode_out);
978 struct xfs_mount *mp = src->i_mount;
980 xfs_extlen_t cowextsize;
983 if (remap_flags & ~(REMAP_FILE_DEDUP | REMAP_FILE_ADVISORY))
986 if (!xfs_sb_version_hasreflink(&mp->m_sb))
989 if (XFS_FORCED_SHUTDOWN(mp))
992 /* Prepare and then clone file data. */
993 ret = xfs_reflink_remap_prep(file_in, pos_in, file_out, pos_out,
995 if (ret < 0 || len == 0)
998 trace_xfs_reflink_remap_range(src, pos_in, len, dest, pos_out);
1000 ret = xfs_reflink_remap_blocks(src, pos_in, dest, pos_out, len,
1006 * Carry the cowextsize hint from src to dest if we're sharing the
1007 * entire source file to the entire destination file, the source file
1008 * has a cowextsize hint, and the destination file does not.
1011 if (pos_in == 0 && len == i_size_read(inode_in) &&
1012 (src->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE) &&
1013 pos_out == 0 && len >= i_size_read(inode_out) &&
1014 !(dest->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE))
1015 cowextsize = src->i_d.di_cowextsize;
1017 ret = xfs_reflink_update_dest(dest, pos_out + len, cowextsize,
1021 xfs_reflink_remap_unlock(file_in, file_out);
1023 trace_xfs_reflink_remap_range_error(dest, ret, _RET_IP_);
1024 return remapped > 0 ? remapped : ret;
1029 struct inode *inode,
1032 if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1034 if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb)))
1036 file->f_mode |= FMODE_NOWAIT;
1042 struct inode *inode,
1045 struct xfs_inode *ip = XFS_I(inode);
1049 error = xfs_file_open(inode, file);
1054 * If there are any blocks, read-ahead block 0 as we're almost
1055 * certain to have the next operation be a read there.
1057 mode = xfs_ilock_data_map_shared(ip);
1058 if (ip->i_d.di_nextents > 0)
1059 error = xfs_dir3_data_readahead(ip, 0, -1);
1060 xfs_iunlock(ip, mode);
1066 struct inode *inode,
1069 return xfs_release(XFS_I(inode));
1075 struct dir_context *ctx)
1077 struct inode *inode = file_inode(file);
1078 xfs_inode_t *ip = XFS_I(inode);
1082 * The Linux API doesn't pass down the total size of the buffer
1083 * we read into down to the filesystem. With the filldir concept
1084 * it's not needed for correct information, but the XFS dir2 leaf
1085 * code wants an estimate of the buffer size to calculate it's
1086 * readahead window and size the buffers used for mapping to
1089 * Try to give it an estimate that's good enough, maybe at some
1090 * point we can change the ->readdir prototype to include the
1091 * buffer size. For now we use the current glibc buffer size.
1093 bufsize = (size_t)min_t(loff_t, XFS_READDIR_BUFSIZE, ip->i_d.di_size);
1095 return xfs_readdir(NULL, ip, ctx, bufsize);
1104 struct inode *inode = file->f_mapping->host;
1106 if (XFS_FORCED_SHUTDOWN(XFS_I(inode)->i_mount))
1111 return generic_file_llseek(file, offset, whence);
1113 offset = iomap_seek_hole(inode, offset, &xfs_seek_iomap_ops);
1116 offset = iomap_seek_data(inode, offset, &xfs_seek_iomap_ops);
1122 return vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
1126 * Locking for serialisation of IO during page faults. This results in a lock
1130 * sb_start_pagefault(vfs, freeze)
1131 * i_mmaplock (XFS - truncate serialisation)
1133 * i_lock (XFS - extent map serialisation)
1136 __xfs_filemap_fault(
1137 struct vm_fault *vmf,
1138 enum page_entry_size pe_size,
1141 struct inode *inode = file_inode(vmf->vma->vm_file);
1142 struct xfs_inode *ip = XFS_I(inode);
1145 trace_xfs_filemap_fault(ip, pe_size, write_fault);
1148 sb_start_pagefault(inode->i_sb);
1149 file_update_time(vmf->vma->vm_file);
1152 xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1153 if (IS_DAX(inode)) {
1156 ret = dax_iomap_fault(vmf, pe_size, &pfn, NULL, &xfs_iomap_ops);
1157 if (ret & VM_FAULT_NEEDDSYNC)
1158 ret = dax_finish_sync_fault(vmf, pe_size, pfn);
1161 ret = iomap_page_mkwrite(vmf, &xfs_iomap_ops);
1163 ret = filemap_fault(vmf);
1165 xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1168 sb_end_pagefault(inode->i_sb);
1174 struct vm_fault *vmf)
1176 /* DAX can shortcut the normal fault path on write faults! */
1177 return __xfs_filemap_fault(vmf, PE_SIZE_PTE,
1178 IS_DAX(file_inode(vmf->vma->vm_file)) &&
1179 (vmf->flags & FAULT_FLAG_WRITE));
1183 xfs_filemap_huge_fault(
1184 struct vm_fault *vmf,
1185 enum page_entry_size pe_size)
1187 if (!IS_DAX(file_inode(vmf->vma->vm_file)))
1188 return VM_FAULT_FALLBACK;
1190 /* DAX can shortcut the normal fault path on write faults! */
1191 return __xfs_filemap_fault(vmf, pe_size,
1192 (vmf->flags & FAULT_FLAG_WRITE));
1196 xfs_filemap_page_mkwrite(
1197 struct vm_fault *vmf)
1199 return __xfs_filemap_fault(vmf, PE_SIZE_PTE, true);
1203 * pfn_mkwrite was originally intended to ensure we capture time stamp updates
1204 * on write faults. In reality, it needs to serialise against truncate and
1205 * prepare memory for writing so handle is as standard write fault.
1208 xfs_filemap_pfn_mkwrite(
1209 struct vm_fault *vmf)
1212 return __xfs_filemap_fault(vmf, PE_SIZE_PTE, true);
1215 static const struct vm_operations_struct xfs_file_vm_ops = {
1216 .fault = xfs_filemap_fault,
1217 .huge_fault = xfs_filemap_huge_fault,
1218 .map_pages = filemap_map_pages,
1219 .page_mkwrite = xfs_filemap_page_mkwrite,
1220 .pfn_mkwrite = xfs_filemap_pfn_mkwrite,
1226 struct vm_area_struct *vma)
1228 struct dax_device *dax_dev;
1230 dax_dev = xfs_find_daxdev_for_inode(file_inode(filp));
1232 * We don't support synchronous mappings for non-DAX files and
1233 * for DAX files if underneath dax_device is not synchronous.
1235 if (!daxdev_mapping_supported(vma, dax_dev))
1238 file_accessed(filp);
1239 vma->vm_ops = &xfs_file_vm_ops;
1240 if (IS_DAX(file_inode(filp)))
1241 vma->vm_flags |= VM_HUGEPAGE;
1245 const struct file_operations xfs_file_operations = {
1246 .llseek = xfs_file_llseek,
1247 .read_iter = xfs_file_read_iter,
1248 .write_iter = xfs_file_write_iter,
1249 .splice_read = generic_file_splice_read,
1250 .splice_write = iter_file_splice_write,
1251 .iopoll = iomap_dio_iopoll,
1252 .unlocked_ioctl = xfs_file_ioctl,
1253 #ifdef CONFIG_COMPAT
1254 .compat_ioctl = xfs_file_compat_ioctl,
1256 .mmap = xfs_file_mmap,
1257 .mmap_supported_flags = MAP_SYNC,
1258 .open = xfs_file_open,
1259 .release = xfs_file_release,
1260 .fsync = xfs_file_fsync,
1261 .get_unmapped_area = thp_get_unmapped_area,
1262 .fallocate = xfs_file_fallocate,
1263 .fadvise = xfs_file_fadvise,
1264 .remap_file_range = xfs_file_remap_range,
1267 const struct file_operations xfs_dir_file_operations = {
1268 .open = xfs_dir_open,
1269 .read = generic_read_dir,
1270 .iterate_shared = xfs_file_readdir,
1271 .llseek = generic_file_llseek,
1272 .unlocked_ioctl = xfs_file_ioctl,
1273 #ifdef CONFIG_COMPAT
1274 .compat_ioctl = xfs_file_compat_ioctl,
1276 .fsync = xfs_dir_fsync,