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>
32 static const struct vm_operations_struct xfs_file_vm_ops;
35 xfs_update_prealloc_flags(
37 enum xfs_prealloc_flags flags)
42 error = xfs_trans_alloc(ip->i_mount, &M_RES(ip->i_mount)->tr_writeid,
47 xfs_ilock(ip, XFS_ILOCK_EXCL);
48 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
50 if (!(flags & XFS_PREALLOC_INVISIBLE)) {
51 VFS_I(ip)->i_mode &= ~S_ISUID;
52 if (VFS_I(ip)->i_mode & S_IXGRP)
53 VFS_I(ip)->i_mode &= ~S_ISGID;
54 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
57 if (flags & XFS_PREALLOC_SET)
58 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
59 if (flags & XFS_PREALLOC_CLEAR)
60 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
62 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
63 if (flags & XFS_PREALLOC_SYNC)
64 xfs_trans_set_sync(tp);
65 return xfs_trans_commit(tp);
69 * Fsync operations on directories are much simpler than on regular files,
70 * as there is no file data to flush, and thus also no need for explicit
71 * cache flush operations, and there are no non-transaction metadata updates
72 * on directories either.
81 struct xfs_inode *ip = XFS_I(file->f_mapping->host);
82 struct xfs_mount *mp = ip->i_mount;
85 trace_xfs_dir_fsync(ip);
87 xfs_ilock(ip, XFS_ILOCK_SHARED);
88 if (xfs_ipincount(ip))
89 lsn = ip->i_itemp->ili_last_lsn;
90 xfs_iunlock(ip, XFS_ILOCK_SHARED);
94 return xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, NULL);
104 struct inode *inode = file->f_mapping->host;
105 struct xfs_inode *ip = XFS_I(inode);
106 struct xfs_mount *mp = ip->i_mount;
111 trace_xfs_file_fsync(ip);
113 error = file_write_and_wait_range(file, start, end);
117 if (XFS_FORCED_SHUTDOWN(mp))
120 xfs_iflags_clear(ip, XFS_ITRUNCATED);
123 * If we have an RT and/or log subvolume we need to make sure to flush
124 * the write cache the device used for file data first. This is to
125 * ensure newly written file data make it to disk before logging the new
126 * inode size in case of an extending write.
128 if (XFS_IS_REALTIME_INODE(ip))
129 xfs_blkdev_issue_flush(mp->m_rtdev_targp);
130 else if (mp->m_logdev_targp != mp->m_ddev_targp)
131 xfs_blkdev_issue_flush(mp->m_ddev_targp);
134 * All metadata updates are logged, which means that we just have to
135 * flush the log up to the latest LSN that touched the inode. If we have
136 * concurrent fsync/fdatasync() calls, we need them to all block on the
137 * log force before we clear the ili_fsync_fields field. This ensures
138 * that we don't get a racing sync operation that does not wait for the
139 * metadata to hit the journal before returning. If we race with
140 * clearing the ili_fsync_fields, then all that will happen is the log
141 * force will do nothing as the lsn will already be on disk. We can't
142 * race with setting ili_fsync_fields because that is done under
143 * XFS_ILOCK_EXCL, and that can't happen because we hold the lock shared
144 * until after the ili_fsync_fields is cleared.
146 xfs_ilock(ip, XFS_ILOCK_SHARED);
147 if (xfs_ipincount(ip)) {
149 (ip->i_itemp->ili_fsync_fields & ~XFS_ILOG_TIMESTAMP))
150 lsn = ip->i_itemp->ili_last_lsn;
154 error = xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, &log_flushed);
155 ip->i_itemp->ili_fsync_fields = 0;
157 xfs_iunlock(ip, XFS_ILOCK_SHARED);
160 * If we only have a single device, and the log force about was
161 * a no-op we might have to flush the data device cache here.
162 * This can only happen for fdatasync/O_DSYNC if we were overwriting
163 * an already allocated file and thus do not have any metadata to
166 if (!log_flushed && !XFS_IS_REALTIME_INODE(ip) &&
167 mp->m_logdev_targp == mp->m_ddev_targp)
168 xfs_blkdev_issue_flush(mp->m_ddev_targp);
174 xfs_file_dio_aio_read(
178 struct xfs_inode *ip = XFS_I(file_inode(iocb->ki_filp));
179 size_t count = iov_iter_count(to);
182 trace_xfs_file_direct_read(ip, count, iocb->ki_pos);
185 return 0; /* skip atime */
187 file_accessed(iocb->ki_filp);
189 xfs_ilock(ip, XFS_IOLOCK_SHARED);
190 ret = iomap_dio_rw(iocb, to, &xfs_iomap_ops, NULL);
191 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
196 static noinline ssize_t
201 struct xfs_inode *ip = XFS_I(iocb->ki_filp->f_mapping->host);
202 size_t count = iov_iter_count(to);
205 trace_xfs_file_dax_read(ip, count, iocb->ki_pos);
208 return 0; /* skip atime */
210 if (iocb->ki_flags & IOCB_NOWAIT) {
211 if (!xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED))
214 xfs_ilock(ip, XFS_IOLOCK_SHARED);
217 ret = dax_iomap_rw(iocb, to, &xfs_iomap_ops);
218 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
220 file_accessed(iocb->ki_filp);
225 xfs_file_buffered_aio_read(
229 struct xfs_inode *ip = XFS_I(file_inode(iocb->ki_filp));
232 trace_xfs_file_buffered_read(ip, iov_iter_count(to), iocb->ki_pos);
234 if (iocb->ki_flags & IOCB_NOWAIT) {
235 if (!xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED))
238 xfs_ilock(ip, XFS_IOLOCK_SHARED);
240 ret = generic_file_read_iter(iocb, to);
241 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
251 struct inode *inode = file_inode(iocb->ki_filp);
252 struct xfs_mount *mp = XFS_I(inode)->i_mount;
255 XFS_STATS_INC(mp, xs_read_calls);
257 if (XFS_FORCED_SHUTDOWN(mp))
261 ret = xfs_file_dax_read(iocb, to);
262 else if (iocb->ki_flags & IOCB_DIRECT)
263 ret = xfs_file_dio_aio_read(iocb, to);
265 ret = xfs_file_buffered_aio_read(iocb, to);
268 XFS_STATS_ADD(mp, xs_read_bytes, ret);
273 * Common pre-write limit and setup checks.
275 * Called with the iolocked held either shared and exclusive according to
276 * @iolock, and returns with it held. Might upgrade the iolock to exclusive
277 * if called for a direct write beyond i_size.
280 xfs_file_aio_write_checks(
282 struct iov_iter *from,
285 struct file *file = iocb->ki_filp;
286 struct inode *inode = file->f_mapping->host;
287 struct xfs_inode *ip = XFS_I(inode);
289 size_t count = iov_iter_count(from);
290 bool drained_dio = false;
294 error = generic_write_checks(iocb, from);
298 error = xfs_break_layouts(inode, iolock, BREAK_WRITE);
303 * For changing security info in file_remove_privs() we need i_rwsem
306 if (*iolock == XFS_IOLOCK_SHARED && !IS_NOSEC(inode)) {
307 xfs_iunlock(ip, *iolock);
308 *iolock = XFS_IOLOCK_EXCL;
309 xfs_ilock(ip, *iolock);
313 * If the offset is beyond the size of the file, we need to zero any
314 * blocks that fall between the existing EOF and the start of this
315 * write. If zeroing is needed and we are currently holding the
316 * iolock shared, we need to update it to exclusive which implies
317 * having to redo all checks before.
319 * We need to serialise against EOF updates that occur in IO
320 * completions here. We want to make sure that nobody is changing the
321 * size while we do this check until we have placed an IO barrier (i.e.
322 * hold the XFS_IOLOCK_EXCL) that prevents new IO from being dispatched.
323 * The spinlock effectively forms a memory barrier once we have the
324 * XFS_IOLOCK_EXCL so we are guaranteed to see the latest EOF value
325 * and hence be able to correctly determine if we need to run zeroing.
327 spin_lock(&ip->i_flags_lock);
328 isize = i_size_read(inode);
329 if (iocb->ki_pos > isize) {
330 spin_unlock(&ip->i_flags_lock);
332 if (*iolock == XFS_IOLOCK_SHARED) {
333 xfs_iunlock(ip, *iolock);
334 *iolock = XFS_IOLOCK_EXCL;
335 xfs_ilock(ip, *iolock);
336 iov_iter_reexpand(from, count);
339 * We now have an IO submission barrier in place, but
340 * AIO can do EOF updates during IO completion and hence
341 * we now need to wait for all of them to drain. Non-AIO
342 * DIO will have drained before we are given the
343 * XFS_IOLOCK_EXCL, and so for most cases this wait is a
346 inode_dio_wait(inode);
351 trace_xfs_zero_eof(ip, isize, iocb->ki_pos - isize);
352 error = iomap_zero_range(inode, isize, iocb->ki_pos - isize,
353 NULL, &xfs_iomap_ops);
357 spin_unlock(&ip->i_flags_lock);
360 * Updating the timestamps will grab the ilock again from
361 * xfs_fs_dirty_inode, so we have to call it after dropping the
362 * lock above. Eventually we should look into a way to avoid
363 * the pointless lock roundtrip.
365 return file_modified(file);
369 xfs_dio_write_end_io(
374 struct inode *inode = file_inode(iocb->ki_filp);
375 struct xfs_inode *ip = XFS_I(inode);
376 loff_t offset = iocb->ki_pos;
377 unsigned int nofs_flag;
380 trace_xfs_end_io_direct_write(ip, offset, size);
382 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
389 * Capture amount written on completion as we can't reliably account
390 * for it on submission.
392 XFS_STATS_ADD(ip->i_mount, xs_write_bytes, size);
395 * We can allocate memory here while doing writeback on behalf of
396 * memory reclaim. To avoid memory allocation deadlocks set the
397 * task-wide nofs context for the following operations.
399 nofs_flag = memalloc_nofs_save();
401 if (flags & IOMAP_DIO_COW) {
402 error = xfs_reflink_end_cow(ip, offset, size);
408 * Unwritten conversion updates the in-core isize after extent
409 * conversion but before updating the on-disk size. Updating isize any
410 * earlier allows a racing dio read to find unwritten extents before
411 * they are converted.
413 if (flags & IOMAP_DIO_UNWRITTEN) {
414 error = xfs_iomap_write_unwritten(ip, offset, size, true);
419 * We need to update the in-core inode size here so that we don't end up
420 * with the on-disk inode size being outside the in-core inode size. We
421 * have no other method of updating EOF for AIO, so always do it here
424 * We need to lock the test/set EOF update as we can be racing with
425 * other IO completions here to update the EOF. Failing to serialise
426 * here can result in EOF moving backwards and Bad Things Happen when
429 spin_lock(&ip->i_flags_lock);
430 if (offset + size > i_size_read(inode)) {
431 i_size_write(inode, offset + size);
432 spin_unlock(&ip->i_flags_lock);
433 error = xfs_setfilesize(ip, offset, size);
435 spin_unlock(&ip->i_flags_lock);
439 memalloc_nofs_restore(nofs_flag);
444 * xfs_file_dio_aio_write - handle direct IO writes
446 * Lock the inode appropriately to prepare for and issue a direct IO write.
447 * By separating it from the buffered write path we remove all the tricky to
448 * follow locking changes and looping.
450 * If there are cached pages or we're extending the file, we need IOLOCK_EXCL
451 * until we're sure the bytes at the new EOF have been zeroed and/or the cached
452 * pages are flushed out.
454 * In most cases the direct IO writes will be done holding IOLOCK_SHARED
455 * allowing them to be done in parallel with reads and other direct IO writes.
456 * However, if the IO is not aligned to filesystem blocks, the direct IO layer
457 * needs to do sub-block zeroing and that requires serialisation against other
458 * direct IOs to the same block. In this case we need to serialise the
459 * submission of the unaligned IOs so that we don't get racing block zeroing in
460 * the dio layer. To avoid the problem with aio, we also need to wait for
461 * outstanding IOs to complete so that unwritten extent conversion is completed
462 * before we try to map the overlapping block. This is currently implemented by
463 * hitting it with a big hammer (i.e. inode_dio_wait()).
465 * Returns with locks held indicated by @iolock and errors indicated by
466 * negative return values.
469 xfs_file_dio_aio_write(
471 struct iov_iter *from)
473 struct file *file = iocb->ki_filp;
474 struct address_space *mapping = file->f_mapping;
475 struct inode *inode = mapping->host;
476 struct xfs_inode *ip = XFS_I(inode);
477 struct xfs_mount *mp = ip->i_mount;
479 int unaligned_io = 0;
481 size_t count = iov_iter_count(from);
482 struct xfs_buftarg *target = XFS_IS_REALTIME_INODE(ip) ?
483 mp->m_rtdev_targp : mp->m_ddev_targp;
485 /* DIO must be aligned to device logical sector size */
486 if ((iocb->ki_pos | count) & target->bt_logical_sectormask)
490 * Don't take the exclusive iolock here unless the I/O is unaligned to
491 * the file system block size. We don't need to consider the EOF
492 * extension case here because xfs_file_aio_write_checks() will relock
493 * the inode as necessary for EOF zeroing cases and fill out the new
494 * inode size as appropriate.
496 if ((iocb->ki_pos & mp->m_blockmask) ||
497 ((iocb->ki_pos + count) & mp->m_blockmask)) {
501 * We can't properly handle unaligned direct I/O to reflink
502 * files yet, as we can't unshare a partial block.
504 if (xfs_is_cow_inode(ip)) {
505 trace_xfs_reflink_bounce_dio_write(ip, iocb->ki_pos, count);
508 iolock = XFS_IOLOCK_EXCL;
510 iolock = XFS_IOLOCK_SHARED;
513 if (iocb->ki_flags & IOCB_NOWAIT) {
514 /* unaligned dio always waits, bail */
517 if (!xfs_ilock_nowait(ip, iolock))
520 xfs_ilock(ip, iolock);
523 ret = xfs_file_aio_write_checks(iocb, from, &iolock);
526 count = iov_iter_count(from);
529 * If we are doing unaligned IO, we can't allow any other overlapping IO
530 * in-flight at the same time or we risk data corruption. Wait for all
531 * other IO to drain before we submit. If the IO is aligned, demote the
532 * iolock if we had to take the exclusive lock in
533 * xfs_file_aio_write_checks() for other reasons.
536 inode_dio_wait(inode);
537 } else if (iolock == XFS_IOLOCK_EXCL) {
538 xfs_ilock_demote(ip, XFS_IOLOCK_EXCL);
539 iolock = XFS_IOLOCK_SHARED;
542 trace_xfs_file_direct_write(ip, count, iocb->ki_pos);
543 ret = iomap_dio_rw(iocb, from, &xfs_iomap_ops, xfs_dio_write_end_io);
546 * If unaligned, this is the only IO in-flight. If it has not yet
547 * completed, wait on it before we release the iolock to prevent
548 * subsequent overlapping IO.
550 if (ret == -EIOCBQUEUED && unaligned_io)
551 inode_dio_wait(inode);
553 xfs_iunlock(ip, iolock);
556 * No fallback to buffered IO on errors for XFS, direct IO will either
557 * complete fully or fail.
559 ASSERT(ret < 0 || ret == count);
563 static noinline ssize_t
566 struct iov_iter *from)
568 struct inode *inode = iocb->ki_filp->f_mapping->host;
569 struct xfs_inode *ip = XFS_I(inode);
570 int iolock = XFS_IOLOCK_EXCL;
571 ssize_t ret, error = 0;
575 if (iocb->ki_flags & IOCB_NOWAIT) {
576 if (!xfs_ilock_nowait(ip, iolock))
579 xfs_ilock(ip, iolock);
582 ret = xfs_file_aio_write_checks(iocb, from, &iolock);
587 count = iov_iter_count(from);
589 trace_xfs_file_dax_write(ip, count, pos);
590 ret = dax_iomap_rw(iocb, from, &xfs_iomap_ops);
591 if (ret > 0 && iocb->ki_pos > i_size_read(inode)) {
592 i_size_write(inode, iocb->ki_pos);
593 error = xfs_setfilesize(ip, pos, ret);
596 xfs_iunlock(ip, iolock);
601 XFS_STATS_ADD(ip->i_mount, xs_write_bytes, ret);
603 /* Handle various SYNC-type writes */
604 ret = generic_write_sync(iocb, ret);
610 xfs_file_buffered_aio_write(
612 struct iov_iter *from)
614 struct file *file = iocb->ki_filp;
615 struct address_space *mapping = file->f_mapping;
616 struct inode *inode = mapping->host;
617 struct xfs_inode *ip = XFS_I(inode);
622 if (iocb->ki_flags & IOCB_NOWAIT)
626 iolock = XFS_IOLOCK_EXCL;
627 xfs_ilock(ip, iolock);
629 ret = xfs_file_aio_write_checks(iocb, from, &iolock);
633 /* We can write back this queue in page reclaim */
634 current->backing_dev_info = inode_to_bdi(inode);
636 trace_xfs_file_buffered_write(ip, iov_iter_count(from), iocb->ki_pos);
637 ret = iomap_file_buffered_write(iocb, from, &xfs_iomap_ops);
638 if (likely(ret >= 0))
642 * If we hit a space limit, try to free up some lingering preallocated
643 * space before returning an error. In the case of ENOSPC, first try to
644 * write back all dirty inodes to free up some of the excess reserved
645 * metadata space. This reduces the chances that the eofblocks scan
646 * waits on dirty mappings. Since xfs_flush_inodes() is serialized, this
647 * also behaves as a filter to prevent too many eofblocks scans from
648 * running at the same time.
650 if (ret == -EDQUOT && !enospc) {
651 xfs_iunlock(ip, iolock);
652 enospc = xfs_inode_free_quota_eofblocks(ip);
655 enospc = xfs_inode_free_quota_cowblocks(ip);
659 } else if (ret == -ENOSPC && !enospc) {
660 struct xfs_eofblocks eofb = {0};
663 xfs_flush_inodes(ip->i_mount);
665 xfs_iunlock(ip, iolock);
666 eofb.eof_flags = XFS_EOF_FLAGS_SYNC;
667 xfs_icache_free_eofblocks(ip->i_mount, &eofb);
668 xfs_icache_free_cowblocks(ip->i_mount, &eofb);
672 current->backing_dev_info = NULL;
675 xfs_iunlock(ip, iolock);
678 XFS_STATS_ADD(ip->i_mount, xs_write_bytes, ret);
679 /* Handle various SYNC-type writes */
680 ret = generic_write_sync(iocb, ret);
688 struct iov_iter *from)
690 struct file *file = iocb->ki_filp;
691 struct address_space *mapping = file->f_mapping;
692 struct inode *inode = mapping->host;
693 struct xfs_inode *ip = XFS_I(inode);
695 size_t ocount = iov_iter_count(from);
697 XFS_STATS_INC(ip->i_mount, xs_write_calls);
702 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
706 return xfs_file_dax_write(iocb, from);
708 if (iocb->ki_flags & IOCB_DIRECT) {
710 * Allow a directio write to fall back to a buffered
711 * write *only* in the case that we're doing a reflink
712 * CoW. In all other directio scenarios we do not
713 * allow an operation to fall back to buffered mode.
715 ret = xfs_file_dio_aio_write(iocb, from);
720 return xfs_file_buffered_aio_write(iocb, from);
727 struct xfs_inode *ip = XFS_I(inode);
729 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
731 xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
735 xfs_break_dax_layouts(
741 ASSERT(xfs_isilocked(XFS_I(inode), XFS_MMAPLOCK_EXCL));
743 page = dax_layout_busy_page(inode->i_mapping);
748 return ___wait_var_event(&page->_refcount,
749 atomic_read(&page->_refcount) == 1, TASK_INTERRUPTIBLE,
750 0, 0, xfs_wait_dax_page(inode));
757 enum layout_break_reason reason)
762 ASSERT(xfs_isilocked(XFS_I(inode), XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL));
768 error = xfs_break_dax_layouts(inode, &retry);
773 error = xfs_break_leased_layouts(inode, iolock, &retry);
779 } while (error == 0 && retry);
784 #define XFS_FALLOC_FL_SUPPORTED \
785 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
786 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE | \
787 FALLOC_FL_INSERT_RANGE | FALLOC_FL_UNSHARE_RANGE)
796 struct inode *inode = file_inode(file);
797 struct xfs_inode *ip = XFS_I(inode);
799 enum xfs_prealloc_flags flags = 0;
800 uint iolock = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL;
802 bool do_file_insert = false;
804 if (!S_ISREG(inode->i_mode))
806 if (mode & ~XFS_FALLOC_FL_SUPPORTED)
809 xfs_ilock(ip, iolock);
810 error = xfs_break_layouts(inode, &iolock, BREAK_UNMAP);
814 if (mode & FALLOC_FL_PUNCH_HOLE) {
815 error = xfs_free_file_space(ip, offset, len);
818 } else if (mode & FALLOC_FL_COLLAPSE_RANGE) {
819 unsigned int blksize_mask = i_blocksize(inode) - 1;
821 if (offset & blksize_mask || len & blksize_mask) {
827 * There is no need to overlap collapse range with EOF,
828 * in which case it is effectively a truncate operation
830 if (offset + len >= i_size_read(inode)) {
835 new_size = i_size_read(inode) - len;
837 error = xfs_collapse_file_space(ip, offset, len);
840 } else if (mode & FALLOC_FL_INSERT_RANGE) {
841 unsigned int blksize_mask = i_blocksize(inode) - 1;
842 loff_t isize = i_size_read(inode);
844 if (offset & blksize_mask || len & blksize_mask) {
850 * New inode size must not exceed ->s_maxbytes, accounting for
851 * possible signed overflow.
853 if (inode->i_sb->s_maxbytes - isize < len) {
857 new_size = isize + len;
859 /* Offset should be less than i_size */
860 if (offset >= isize) {
864 do_file_insert = true;
866 flags |= XFS_PREALLOC_SET;
868 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
869 offset + len > i_size_read(inode)) {
870 new_size = offset + len;
871 error = inode_newsize_ok(inode, new_size);
876 if (mode & FALLOC_FL_ZERO_RANGE) {
877 error = xfs_zero_file_space(ip, offset, len);
878 } else if (mode & FALLOC_FL_UNSHARE_RANGE) {
879 error = xfs_reflink_unshare(ip, offset, len);
883 if (!xfs_is_always_cow_inode(ip)) {
884 error = xfs_alloc_file_space(ip, offset, len,
889 * If always_cow mode we can't use preallocations and
890 * thus should not create them.
892 if (xfs_is_always_cow_inode(ip)) {
897 error = xfs_alloc_file_space(ip, offset, len,
904 if (file->f_flags & O_DSYNC)
905 flags |= XFS_PREALLOC_SYNC;
907 error = xfs_update_prealloc_flags(ip, flags);
911 /* Change file size if needed */
915 iattr.ia_valid = ATTR_SIZE;
916 iattr.ia_size = new_size;
917 error = xfs_vn_setattr_size(file_dentry(file), &iattr);
923 * Perform hole insertion now that the file size has been
924 * updated so that if we crash during the operation we don't
925 * leave shifted extents past EOF and hence losing access to
926 * the data that is contained within them.
929 error = xfs_insert_file_space(ip, offset, len);
932 xfs_iunlock(ip, iolock);
938 xfs_file_remap_range(
939 struct file *file_in,
941 struct file *file_out,
944 unsigned int remap_flags)
946 struct inode *inode_in = file_inode(file_in);
947 struct xfs_inode *src = XFS_I(inode_in);
948 struct inode *inode_out = file_inode(file_out);
949 struct xfs_inode *dest = XFS_I(inode_out);
950 struct xfs_mount *mp = src->i_mount;
952 xfs_extlen_t cowextsize;
955 if (remap_flags & ~(REMAP_FILE_DEDUP | REMAP_FILE_ADVISORY))
958 if (!xfs_sb_version_hasreflink(&mp->m_sb))
961 if (XFS_FORCED_SHUTDOWN(mp))
964 /* Prepare and then clone file data. */
965 ret = xfs_reflink_remap_prep(file_in, pos_in, file_out, pos_out,
967 if (ret < 0 || len == 0)
970 trace_xfs_reflink_remap_range(src, pos_in, len, dest, pos_out);
972 ret = xfs_reflink_remap_blocks(src, pos_in, dest, pos_out, len,
978 * Carry the cowextsize hint from src to dest if we're sharing the
979 * entire source file to the entire destination file, the source file
980 * has a cowextsize hint, and the destination file does not.
983 if (pos_in == 0 && len == i_size_read(inode_in) &&
984 (src->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE) &&
985 pos_out == 0 && len >= i_size_read(inode_out) &&
986 !(dest->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE))
987 cowextsize = src->i_d.di_cowextsize;
989 ret = xfs_reflink_update_dest(dest, pos_out + len, cowextsize,
993 xfs_reflink_remap_unlock(file_in, file_out);
995 trace_xfs_reflink_remap_range_error(dest, ret, _RET_IP_);
996 return remapped > 0 ? remapped : ret;
1001 struct inode *inode,
1004 if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1006 if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb)))
1008 file->f_mode |= FMODE_NOWAIT;
1014 struct inode *inode,
1017 struct xfs_inode *ip = XFS_I(inode);
1021 error = xfs_file_open(inode, file);
1026 * If there are any blocks, read-ahead block 0 as we're almost
1027 * certain to have the next operation be a read there.
1029 mode = xfs_ilock_data_map_shared(ip);
1030 if (ip->i_d.di_nextents > 0)
1031 error = xfs_dir3_data_readahead(ip, 0, -1);
1032 xfs_iunlock(ip, mode);
1038 struct inode *inode,
1041 return xfs_release(XFS_I(inode));
1047 struct dir_context *ctx)
1049 struct inode *inode = file_inode(file);
1050 xfs_inode_t *ip = XFS_I(inode);
1054 * The Linux API doesn't pass down the total size of the buffer
1055 * we read into down to the filesystem. With the filldir concept
1056 * it's not needed for correct information, but the XFS dir2 leaf
1057 * code wants an estimate of the buffer size to calculate it's
1058 * readahead window and size the buffers used for mapping to
1061 * Try to give it an estimate that's good enough, maybe at some
1062 * point we can change the ->readdir prototype to include the
1063 * buffer size. For now we use the current glibc buffer size.
1065 bufsize = (size_t)min_t(loff_t, XFS_READDIR_BUFSIZE, ip->i_d.di_size);
1067 return xfs_readdir(NULL, ip, ctx, bufsize);
1076 struct inode *inode = file->f_mapping->host;
1078 if (XFS_FORCED_SHUTDOWN(XFS_I(inode)->i_mount))
1083 return generic_file_llseek(file, offset, whence);
1085 offset = iomap_seek_hole(inode, offset, &xfs_seek_iomap_ops);
1088 offset = iomap_seek_data(inode, offset, &xfs_seek_iomap_ops);
1094 return vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
1098 * Locking for serialisation of IO during page faults. This results in a lock
1102 * sb_start_pagefault(vfs, freeze)
1103 * i_mmaplock (XFS - truncate serialisation)
1105 * i_lock (XFS - extent map serialisation)
1108 __xfs_filemap_fault(
1109 struct vm_fault *vmf,
1110 enum page_entry_size pe_size,
1113 struct inode *inode = file_inode(vmf->vma->vm_file);
1114 struct xfs_inode *ip = XFS_I(inode);
1117 trace_xfs_filemap_fault(ip, pe_size, write_fault);
1120 sb_start_pagefault(inode->i_sb);
1121 file_update_time(vmf->vma->vm_file);
1124 xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1125 if (IS_DAX(inode)) {
1128 ret = dax_iomap_fault(vmf, pe_size, &pfn, NULL, &xfs_iomap_ops);
1129 if (ret & VM_FAULT_NEEDDSYNC)
1130 ret = dax_finish_sync_fault(vmf, pe_size, pfn);
1133 ret = iomap_page_mkwrite(vmf, &xfs_iomap_ops);
1135 ret = filemap_fault(vmf);
1137 xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1140 sb_end_pagefault(inode->i_sb);
1146 struct vm_fault *vmf)
1148 /* DAX can shortcut the normal fault path on write faults! */
1149 return __xfs_filemap_fault(vmf, PE_SIZE_PTE,
1150 IS_DAX(file_inode(vmf->vma->vm_file)) &&
1151 (vmf->flags & FAULT_FLAG_WRITE));
1155 xfs_filemap_huge_fault(
1156 struct vm_fault *vmf,
1157 enum page_entry_size pe_size)
1159 if (!IS_DAX(file_inode(vmf->vma->vm_file)))
1160 return VM_FAULT_FALLBACK;
1162 /* DAX can shortcut the normal fault path on write faults! */
1163 return __xfs_filemap_fault(vmf, pe_size,
1164 (vmf->flags & FAULT_FLAG_WRITE));
1168 xfs_filemap_page_mkwrite(
1169 struct vm_fault *vmf)
1171 return __xfs_filemap_fault(vmf, PE_SIZE_PTE, true);
1175 * pfn_mkwrite was originally intended to ensure we capture time stamp updates
1176 * on write faults. In reality, it needs to serialise against truncate and
1177 * prepare memory for writing so handle is as standard write fault.
1180 xfs_filemap_pfn_mkwrite(
1181 struct vm_fault *vmf)
1184 return __xfs_filemap_fault(vmf, PE_SIZE_PTE, true);
1187 static const struct vm_operations_struct xfs_file_vm_ops = {
1188 .fault = xfs_filemap_fault,
1189 .huge_fault = xfs_filemap_huge_fault,
1190 .map_pages = filemap_map_pages,
1191 .page_mkwrite = xfs_filemap_page_mkwrite,
1192 .pfn_mkwrite = xfs_filemap_pfn_mkwrite,
1198 struct vm_area_struct *vma)
1200 struct dax_device *dax_dev;
1202 dax_dev = xfs_find_daxdev_for_inode(file_inode(filp));
1204 * We don't support synchronous mappings for non-DAX files and
1205 * for DAX files if underneath dax_device is not synchronous.
1207 if (!daxdev_mapping_supported(vma, dax_dev))
1210 file_accessed(filp);
1211 vma->vm_ops = &xfs_file_vm_ops;
1212 if (IS_DAX(file_inode(filp)))
1213 vma->vm_flags |= VM_HUGEPAGE;
1217 const struct file_operations xfs_file_operations = {
1218 .llseek = xfs_file_llseek,
1219 .read_iter = xfs_file_read_iter,
1220 .write_iter = xfs_file_write_iter,
1221 .splice_read = generic_file_splice_read,
1222 .splice_write = iter_file_splice_write,
1223 .iopoll = iomap_dio_iopoll,
1224 .unlocked_ioctl = xfs_file_ioctl,
1225 #ifdef CONFIG_COMPAT
1226 .compat_ioctl = xfs_file_compat_ioctl,
1228 .mmap = xfs_file_mmap,
1229 .mmap_supported_flags = MAP_SYNC,
1230 .open = xfs_file_open,
1231 .release = xfs_file_release,
1232 .fsync = xfs_file_fsync,
1233 .get_unmapped_area = thp_get_unmapped_area,
1234 .fallocate = xfs_file_fallocate,
1235 .remap_file_range = xfs_file_remap_range,
1238 const struct file_operations xfs_dir_file_operations = {
1239 .open = xfs_dir_open,
1240 .read = generic_read_dir,
1241 .iterate_shared = xfs_file_readdir,
1242 .llseek = generic_file_llseek,
1243 .unlocked_ioctl = xfs_file_ioctl,
1244 #ifdef CONFIG_COMPAT
1245 .compat_ioctl = xfs_file_compat_ioctl,
1247 .fsync = xfs_dir_fsync,