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/dax.h>
29 #include <linux/falloc.h>
30 #include <linux/backing-dev.h>
31 #include <linux/mman.h>
32 #include <linux/fadvise.h>
33 #include <linux/mount.h>
35 static const struct vm_operations_struct xfs_file_vm_ops;
38 * Decide if the given file range is aligned to the size of the fundamental
39 * allocation unit for the file.
42 xfs_is_falloc_aligned(
47 struct xfs_mount *mp = ip->i_mount;
50 if (XFS_IS_REALTIME_INODE(ip)) {
51 if (!is_power_of_2(mp->m_sb.sb_rextsize)) {
55 rextbytes = XFS_FSB_TO_B(mp, mp->m_sb.sb_rextsize);
56 div_u64_rem(pos, rextbytes, &mod);
59 div_u64_rem(len, rextbytes, &mod);
62 mask = XFS_FSB_TO_B(mp, mp->m_sb.sb_rextsize) - 1;
64 mask = mp->m_sb.sb_blocksize - 1;
67 return !((pos | len) & mask);
71 * Fsync operations on directories are much simpler than on regular files,
72 * as there is no file data to flush, and thus also no need for explicit
73 * cache flush operations, and there are no non-transaction metadata updates
74 * on directories either.
83 struct xfs_inode *ip = XFS_I(file->f_mapping->host);
85 trace_xfs_dir_fsync(ip);
86 return xfs_log_force_inode(ip);
94 if (!xfs_ipincount(ip))
96 if (datasync && !(ip->i_itemp->ili_fsync_fields & ~XFS_ILOG_TIMESTAMP))
98 return ip->i_itemp->ili_commit_seq;
102 * All metadata updates are logged, which means that we just have to flush the
103 * log up to the latest LSN that touched the inode.
105 * If we have concurrent fsync/fdatasync() calls, we need them to all block on
106 * the log force before we clear the ili_fsync_fields field. This ensures that
107 * we don't get a racing sync operation that does not wait for the metadata to
108 * hit the journal before returning. If we race with clearing ili_fsync_fields,
109 * then all that will happen is the log force will do nothing as the lsn will
110 * already be on disk. We can't race with setting ili_fsync_fields because that
111 * is done under XFS_ILOCK_EXCL, and that can't happen because we hold the lock
112 * shared until after the ili_fsync_fields is cleared.
116 struct xfs_inode *ip,
123 xfs_ilock(ip, XFS_ILOCK_SHARED);
124 seq = xfs_fsync_seq(ip, datasync);
126 error = xfs_log_force_seq(ip->i_mount, seq, XFS_LOG_SYNC,
129 spin_lock(&ip->i_itemp->ili_lock);
130 ip->i_itemp->ili_fsync_fields = 0;
131 spin_unlock(&ip->i_itemp->ili_lock);
133 xfs_iunlock(ip, XFS_ILOCK_SHARED);
144 struct xfs_inode *ip = XFS_I(file->f_mapping->host);
145 struct xfs_mount *mp = ip->i_mount;
149 trace_xfs_file_fsync(ip);
151 error = file_write_and_wait_range(file, start, end);
155 if (xfs_is_shutdown(mp))
158 xfs_iflags_clear(ip, XFS_ITRUNCATED);
161 * If we have an RT and/or log subvolume we need to make sure to flush
162 * the write cache the device used for file data first. This is to
163 * ensure newly written file data make it to disk before logging the new
164 * inode size in case of an extending write.
166 if (XFS_IS_REALTIME_INODE(ip))
167 blkdev_issue_flush(mp->m_rtdev_targp->bt_bdev);
168 else if (mp->m_logdev_targp != mp->m_ddev_targp)
169 blkdev_issue_flush(mp->m_ddev_targp->bt_bdev);
172 * Any inode that has dirty modifications in the log is pinned. The
173 * racy check here for a pinned inode while not catch modifications
174 * that happen concurrently to the fsync call, but fsync semantics
175 * only require to sync previously completed I/O.
177 if (xfs_ipincount(ip))
178 error = xfs_fsync_flush_log(ip, datasync, &log_flushed);
181 * If we only have a single device, and the log force about was
182 * a no-op we might have to flush the data device cache here.
183 * This can only happen for fdatasync/O_DSYNC if we were overwriting
184 * an already allocated file and thus do not have any metadata to
187 if (!log_flushed && !XFS_IS_REALTIME_INODE(ip) &&
188 mp->m_logdev_targp == mp->m_ddev_targp)
189 blkdev_issue_flush(mp->m_ddev_targp->bt_bdev);
197 unsigned int lock_mode)
199 struct xfs_inode *ip = XFS_I(file_inode(iocb->ki_filp));
201 if (iocb->ki_flags & IOCB_NOWAIT) {
202 if (!xfs_ilock_nowait(ip, lock_mode))
205 xfs_ilock(ip, lock_mode);
216 struct xfs_inode *ip = XFS_I(file_inode(iocb->ki_filp));
219 trace_xfs_file_direct_read(iocb, to);
221 if (!iov_iter_count(to))
222 return 0; /* skip atime */
224 file_accessed(iocb->ki_filp);
226 ret = xfs_ilock_iocb(iocb, XFS_IOLOCK_SHARED);
229 ret = iomap_dio_rw(iocb, to, &xfs_read_iomap_ops, NULL, 0, NULL, 0);
230 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
235 static noinline ssize_t
240 struct xfs_inode *ip = XFS_I(iocb->ki_filp->f_mapping->host);
243 trace_xfs_file_dax_read(iocb, to);
245 if (!iov_iter_count(to))
246 return 0; /* skip atime */
248 ret = xfs_ilock_iocb(iocb, XFS_IOLOCK_SHARED);
251 ret = dax_iomap_rw(iocb, to, &xfs_read_iomap_ops);
252 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
254 file_accessed(iocb->ki_filp);
259 xfs_file_buffered_read(
263 struct xfs_inode *ip = XFS_I(file_inode(iocb->ki_filp));
266 trace_xfs_file_buffered_read(iocb, to);
268 ret = xfs_ilock_iocb(iocb, 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_is_shutdown(mp))
292 ret = xfs_file_dax_read(iocb, to);
293 else if (iocb->ki_flags & IOCB_DIRECT)
294 ret = xfs_file_dio_read(iocb, to);
296 ret = xfs_file_buffered_read(iocb, to);
299 XFS_STATS_ADD(mp, xs_read_bytes, ret);
304 * Common pre-write limit and setup checks.
306 * Called with the iolocked held either shared and exclusive according to
307 * @iolock, and returns with it held. Might upgrade the iolock to exclusive
308 * if called for a direct write beyond i_size.
311 xfs_file_write_checks(
313 struct iov_iter *from,
314 unsigned int *iolock)
316 struct file *file = iocb->ki_filp;
317 struct inode *inode = file->f_mapping->host;
318 struct xfs_inode *ip = XFS_I(inode);
320 size_t count = iov_iter_count(from);
321 bool drained_dio = false;
325 error = generic_write_checks(iocb, from);
329 if (iocb->ki_flags & IOCB_NOWAIT) {
330 error = break_layout(inode, false);
331 if (error == -EWOULDBLOCK)
334 error = xfs_break_layouts(inode, iolock, BREAK_WRITE);
341 * For changing security info in file_remove_privs() we need i_rwsem
344 if (*iolock == XFS_IOLOCK_SHARED && !IS_NOSEC(inode)) {
345 xfs_iunlock(ip, *iolock);
346 *iolock = XFS_IOLOCK_EXCL;
347 error = xfs_ilock_iocb(iocb, *iolock);
356 * If the offset is beyond the size of the file, we need to zero any
357 * blocks that fall between the existing EOF and the start of this
358 * write. If zeroing is needed and we are currently holding the iolock
359 * shared, we need to update it to exclusive which implies having to
360 * redo all checks before.
362 * We need to serialise against EOF updates that occur in IO completions
363 * here. We want to make sure that nobody is changing the size while we
364 * do this check until we have placed an IO barrier (i.e. hold the
365 * XFS_IOLOCK_EXCL) that prevents new IO from being dispatched. The
366 * spinlock effectively forms a memory barrier once we have the
367 * XFS_IOLOCK_EXCL so we are guaranteed to see the latest EOF value and
368 * hence be able to correctly determine if we need to run zeroing.
370 * We can do an unlocked check here safely as IO completion can only
371 * extend EOF. Truncate is locked out at this point, so the EOF can
372 * not move backwards, only forwards. Hence we only need to take the
373 * slow path and spin locks when we are at or beyond the current EOF.
375 if (iocb->ki_pos <= i_size_read(inode))
378 spin_lock(&ip->i_flags_lock);
379 isize = i_size_read(inode);
380 if (iocb->ki_pos > isize) {
381 spin_unlock(&ip->i_flags_lock);
383 if (iocb->ki_flags & IOCB_NOWAIT)
387 if (*iolock == XFS_IOLOCK_SHARED) {
388 xfs_iunlock(ip, *iolock);
389 *iolock = XFS_IOLOCK_EXCL;
390 xfs_ilock(ip, *iolock);
391 iov_iter_reexpand(from, count);
394 * We now have an IO submission barrier in place, but
395 * AIO can do EOF updates during IO completion and hence
396 * we now need to wait for all of them to drain. Non-AIO
397 * DIO will have drained before we are given the
398 * XFS_IOLOCK_EXCL, and so for most cases this wait is a
401 inode_dio_wait(inode);
406 trace_xfs_zero_eof(ip, isize, iocb->ki_pos - isize);
407 error = xfs_zero_range(ip, isize, iocb->ki_pos - isize, NULL);
411 spin_unlock(&ip->i_flags_lock);
414 return kiocb_modified(iocb);
418 xfs_dio_write_end_io(
424 struct inode *inode = file_inode(iocb->ki_filp);
425 struct xfs_inode *ip = XFS_I(inode);
426 loff_t offset = iocb->ki_pos;
427 unsigned int nofs_flag;
429 trace_xfs_end_io_direct_write(ip, offset, size);
431 if (xfs_is_shutdown(ip->i_mount))
440 * Capture amount written on completion as we can't reliably account
441 * for it on submission.
443 XFS_STATS_ADD(ip->i_mount, xs_write_bytes, size);
446 * We can allocate memory here while doing writeback on behalf of
447 * memory reclaim. To avoid memory allocation deadlocks set the
448 * task-wide nofs context for the following operations.
450 nofs_flag = memalloc_nofs_save();
452 if (flags & IOMAP_DIO_COW) {
453 error = xfs_reflink_end_cow(ip, offset, size);
459 * Unwritten conversion updates the in-core isize after extent
460 * conversion but before updating the on-disk size. Updating isize any
461 * earlier allows a racing dio read to find unwritten extents before
462 * they are converted.
464 if (flags & IOMAP_DIO_UNWRITTEN) {
465 error = xfs_iomap_write_unwritten(ip, offset, size, true);
470 * We need to update the in-core inode size here so that we don't end up
471 * with the on-disk inode size being outside the in-core inode size. We
472 * have no other method of updating EOF for AIO, so always do it here
475 * We need to lock the test/set EOF update as we can be racing with
476 * other IO completions here to update the EOF. Failing to serialise
477 * here can result in EOF moving backwards and Bad Things Happen when
480 * As IO completion only ever extends EOF, we can do an unlocked check
481 * here to avoid taking the spinlock. If we land within the current EOF,
482 * then we do not need to do an extending update at all, and we don't
483 * need to take the lock to check this. If we race with an update moving
484 * EOF, then we'll either still be beyond EOF and need to take the lock,
485 * or we'll be within EOF and we don't need to take it at all.
487 if (offset + size <= i_size_read(inode))
490 spin_lock(&ip->i_flags_lock);
491 if (offset + size > i_size_read(inode)) {
492 i_size_write(inode, offset + size);
493 spin_unlock(&ip->i_flags_lock);
494 error = xfs_setfilesize(ip, offset, size);
496 spin_unlock(&ip->i_flags_lock);
500 memalloc_nofs_restore(nofs_flag);
504 static const struct iomap_dio_ops xfs_dio_write_ops = {
505 .end_io = xfs_dio_write_end_io,
509 * Handle block aligned direct I/O writes
511 static noinline ssize_t
512 xfs_file_dio_write_aligned(
513 struct xfs_inode *ip,
515 struct iov_iter *from)
517 unsigned int iolock = XFS_IOLOCK_SHARED;
520 ret = xfs_ilock_iocb(iocb, iolock);
523 ret = xfs_file_write_checks(iocb, from, &iolock);
528 * We don't need to hold the IOLOCK exclusively across the IO, so demote
529 * the iolock back to shared if we had to take the exclusive lock in
530 * xfs_file_write_checks() for other reasons.
532 if (iolock == XFS_IOLOCK_EXCL) {
533 xfs_ilock_demote(ip, XFS_IOLOCK_EXCL);
534 iolock = XFS_IOLOCK_SHARED;
536 trace_xfs_file_direct_write(iocb, from);
537 ret = iomap_dio_rw(iocb, from, &xfs_direct_write_iomap_ops,
538 &xfs_dio_write_ops, 0, NULL, 0);
541 xfs_iunlock(ip, iolock);
546 * Handle block unaligned direct I/O writes
548 * In most cases direct I/O writes will be done holding IOLOCK_SHARED, allowing
549 * them to be done in parallel with reads and other direct I/O writes. However,
550 * if the I/O is not aligned to filesystem blocks, the direct I/O layer may need
551 * to do sub-block zeroing and that requires serialisation against other direct
552 * I/O to the same block. In this case we need to serialise the submission of
553 * the unaligned I/O so that we don't get racing block zeroing in the dio layer.
554 * In the case where sub-block zeroing is not required, we can do concurrent
555 * sub-block dios to the same block successfully.
557 * Optimistically submit the I/O using the shared lock first, but use the
558 * IOMAP_DIO_OVERWRITE_ONLY flag to tell the lower layers to return -EAGAIN
559 * if block allocation or partial block zeroing would be required. In that case
560 * we try again with the exclusive lock.
562 static noinline ssize_t
563 xfs_file_dio_write_unaligned(
564 struct xfs_inode *ip,
566 struct iov_iter *from)
568 size_t isize = i_size_read(VFS_I(ip));
569 size_t count = iov_iter_count(from);
570 unsigned int iolock = XFS_IOLOCK_SHARED;
571 unsigned int flags = IOMAP_DIO_OVERWRITE_ONLY;
575 * Extending writes need exclusivity because of the sub-block zeroing
576 * that the DIO code always does for partial tail blocks beyond EOF, so
577 * don't even bother trying the fast path in this case.
579 if (iocb->ki_pos > isize || iocb->ki_pos + count >= isize) {
580 if (iocb->ki_flags & IOCB_NOWAIT)
583 iolock = XFS_IOLOCK_EXCL;
584 flags = IOMAP_DIO_FORCE_WAIT;
587 ret = xfs_ilock_iocb(iocb, iolock);
592 * We can't properly handle unaligned direct I/O to reflink files yet,
593 * as we can't unshare a partial block.
595 if (xfs_is_cow_inode(ip)) {
596 trace_xfs_reflink_bounce_dio_write(iocb, from);
601 ret = xfs_file_write_checks(iocb, from, &iolock);
606 * If we are doing exclusive unaligned I/O, this must be the only I/O
607 * in-flight. Otherwise we risk data corruption due to unwritten extent
608 * conversions from the AIO end_io handler. Wait for all other I/O to
611 if (flags & IOMAP_DIO_FORCE_WAIT)
612 inode_dio_wait(VFS_I(ip));
614 trace_xfs_file_direct_write(iocb, from);
615 ret = iomap_dio_rw(iocb, from, &xfs_direct_write_iomap_ops,
616 &xfs_dio_write_ops, flags, NULL, 0);
619 * Retry unaligned I/O with exclusive blocking semantics if the DIO
620 * layer rejected it for mapping or locking reasons. If we are doing
621 * nonblocking user I/O, propagate the error.
623 if (ret == -EAGAIN && !(iocb->ki_flags & IOCB_NOWAIT)) {
624 ASSERT(flags & IOMAP_DIO_OVERWRITE_ONLY);
625 xfs_iunlock(ip, iolock);
626 goto retry_exclusive;
631 xfs_iunlock(ip, iolock);
638 struct iov_iter *from)
640 struct xfs_inode *ip = XFS_I(file_inode(iocb->ki_filp));
641 struct xfs_buftarg *target = xfs_inode_buftarg(ip);
642 size_t count = iov_iter_count(from);
644 /* direct I/O must be aligned to device logical sector size */
645 if ((iocb->ki_pos | count) & target->bt_logical_sectormask)
647 if ((iocb->ki_pos | count) & ip->i_mount->m_blockmask)
648 return xfs_file_dio_write_unaligned(ip, iocb, from);
649 return xfs_file_dio_write_aligned(ip, iocb, from);
652 static noinline ssize_t
655 struct iov_iter *from)
657 struct inode *inode = iocb->ki_filp->f_mapping->host;
658 struct xfs_inode *ip = XFS_I(inode);
659 unsigned int iolock = XFS_IOLOCK_EXCL;
660 ssize_t ret, error = 0;
663 ret = xfs_ilock_iocb(iocb, iolock);
666 ret = xfs_file_write_checks(iocb, from, &iolock);
672 trace_xfs_file_dax_write(iocb, from);
673 ret = dax_iomap_rw(iocb, from, &xfs_dax_write_iomap_ops);
674 if (ret > 0 && iocb->ki_pos > i_size_read(inode)) {
675 i_size_write(inode, iocb->ki_pos);
676 error = xfs_setfilesize(ip, pos, ret);
680 xfs_iunlock(ip, iolock);
685 XFS_STATS_ADD(ip->i_mount, xs_write_bytes, ret);
687 /* Handle various SYNC-type writes */
688 ret = generic_write_sync(iocb, ret);
694 xfs_file_buffered_write(
696 struct iov_iter *from)
698 struct inode *inode = iocb->ki_filp->f_mapping->host;
699 struct xfs_inode *ip = XFS_I(inode);
701 bool cleared_space = false;
705 iolock = XFS_IOLOCK_EXCL;
706 ret = xfs_ilock_iocb(iocb, iolock);
710 ret = xfs_file_write_checks(iocb, from, &iolock);
714 /* We can write back this queue in page reclaim */
715 current->backing_dev_info = inode_to_bdi(inode);
717 trace_xfs_file_buffered_write(iocb, from);
718 ret = iomap_file_buffered_write(iocb, from,
719 &xfs_buffered_write_iomap_ops);
720 if (likely(ret >= 0))
724 * If we hit a space limit, try to free up some lingering preallocated
725 * space before returning an error. In the case of ENOSPC, first try to
726 * write back all dirty inodes to free up some of the excess reserved
727 * metadata space. This reduces the chances that the eofblocks scan
728 * waits on dirty mappings. Since xfs_flush_inodes() is serialized, this
729 * also behaves as a filter to prevent too many eofblocks scans from
730 * running at the same time. Use a synchronous scan to increase the
731 * effectiveness of the scan.
733 if (ret == -EDQUOT && !cleared_space) {
734 xfs_iunlock(ip, iolock);
735 xfs_blockgc_free_quota(ip, XFS_ICWALK_FLAG_SYNC);
736 cleared_space = true;
738 } else if (ret == -ENOSPC && !cleared_space) {
739 struct xfs_icwalk icw = {0};
741 cleared_space = true;
742 xfs_flush_inodes(ip->i_mount);
744 xfs_iunlock(ip, iolock);
745 icw.icw_flags = XFS_ICWALK_FLAG_SYNC;
746 xfs_blockgc_free_space(ip->i_mount, &icw);
750 current->backing_dev_info = NULL;
753 xfs_iunlock(ip, iolock);
756 XFS_STATS_ADD(ip->i_mount, xs_write_bytes, ret);
757 /* Handle various SYNC-type writes */
758 ret = generic_write_sync(iocb, ret);
766 struct iov_iter *from)
768 struct inode *inode = iocb->ki_filp->f_mapping->host;
769 struct xfs_inode *ip = XFS_I(inode);
771 size_t ocount = iov_iter_count(from);
773 XFS_STATS_INC(ip->i_mount, xs_write_calls);
778 if (xfs_is_shutdown(ip->i_mount))
782 return xfs_file_dax_write(iocb, from);
784 if (iocb->ki_flags & IOCB_DIRECT) {
786 * Allow a directio write to fall back to a buffered
787 * write *only* in the case that we're doing a reflink
788 * CoW. In all other directio scenarios we do not
789 * allow an operation to fall back to buffered mode.
791 ret = xfs_file_dio_write(iocb, from);
796 return xfs_file_buffered_write(iocb, from);
803 struct xfs_inode *ip = XFS_I(inode);
805 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
807 xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
811 xfs_break_dax_layouts(
817 ASSERT(xfs_isilocked(XFS_I(inode), XFS_MMAPLOCK_EXCL));
819 page = dax_layout_busy_page(inode->i_mapping);
824 return ___wait_var_event(&page->_refcount,
825 atomic_read(&page->_refcount) == 1, TASK_INTERRUPTIBLE,
826 0, 0, xfs_wait_dax_page(inode));
833 enum layout_break_reason reason)
838 ASSERT(xfs_isilocked(XFS_I(inode), XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL));
844 error = xfs_break_dax_layouts(inode, &retry);
849 error = xfs_break_leased_layouts(inode, iolock, &retry);
855 } while (error == 0 && retry);
860 /* Does this file, inode, or mount want synchronous writes? */
861 static inline bool xfs_file_sync_writes(struct file *filp)
863 struct xfs_inode *ip = XFS_I(file_inode(filp));
865 if (xfs_has_wsync(ip->i_mount))
867 if (filp->f_flags & (__O_SYNC | O_DSYNC))
869 if (IS_SYNC(file_inode(filp)))
875 #define XFS_FALLOC_FL_SUPPORTED \
876 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
877 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE | \
878 FALLOC_FL_INSERT_RANGE | FALLOC_FL_UNSHARE_RANGE)
887 struct inode *inode = file_inode(file);
888 struct xfs_inode *ip = XFS_I(inode);
890 uint iolock = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL;
892 bool do_file_insert = false;
894 if (!S_ISREG(inode->i_mode))
896 if (mode & ~XFS_FALLOC_FL_SUPPORTED)
899 xfs_ilock(ip, iolock);
900 error = xfs_break_layouts(inode, &iolock, BREAK_UNMAP);
905 * Must wait for all AIO to complete before we continue as AIO can
906 * change the file size on completion without holding any locks we
907 * currently hold. We must do this first because AIO can update both
908 * the on disk and in memory inode sizes, and the operations that follow
909 * require the in-memory size to be fully up-to-date.
911 inode_dio_wait(inode);
914 * Now AIO and DIO has drained we flush and (if necessary) invalidate
915 * the cached range over the first operation we are about to run.
917 * We care about zero and collapse here because they both run a hole
918 * punch over the range first. Because that can zero data, and the range
919 * of invalidation for the shift operations is much larger, we still do
920 * the required flush for collapse in xfs_prepare_shift().
922 * Insert has the same range requirements as collapse, and we extend the
923 * file first which can zero data. Hence insert has the same
924 * flush/invalidate requirements as collapse and so they are both
925 * handled at the right time by xfs_prepare_shift().
927 if (mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE |
928 FALLOC_FL_COLLAPSE_RANGE)) {
929 error = xfs_flush_unmap_range(ip, offset, len);
934 error = file_modified(file);
938 if (mode & FALLOC_FL_PUNCH_HOLE) {
939 error = xfs_free_file_space(ip, offset, len);
942 } else if (mode & FALLOC_FL_COLLAPSE_RANGE) {
943 if (!xfs_is_falloc_aligned(ip, offset, len)) {
949 * There is no need to overlap collapse range with EOF,
950 * in which case it is effectively a truncate operation
952 if (offset + len >= i_size_read(inode)) {
957 new_size = i_size_read(inode) - len;
959 error = xfs_collapse_file_space(ip, offset, len);
962 } else if (mode & FALLOC_FL_INSERT_RANGE) {
963 loff_t isize = i_size_read(inode);
965 if (!xfs_is_falloc_aligned(ip, offset, len)) {
971 * New inode size must not exceed ->s_maxbytes, accounting for
972 * possible signed overflow.
974 if (inode->i_sb->s_maxbytes - isize < len) {
978 new_size = isize + len;
980 /* Offset should be less than i_size */
981 if (offset >= isize) {
985 do_file_insert = true;
987 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
988 offset + len > i_size_read(inode)) {
989 new_size = offset + len;
990 error = inode_newsize_ok(inode, new_size);
995 if (mode & FALLOC_FL_ZERO_RANGE) {
997 * Punch a hole and prealloc the range. We use a hole
998 * punch rather than unwritten extent conversion for two
1001 * 1.) Hole punch handles partial block zeroing for us.
1002 * 2.) If prealloc returns ENOSPC, the file range is
1003 * still zero-valued by virtue of the hole punch.
1005 unsigned int blksize = i_blocksize(inode);
1007 trace_xfs_zero_file_space(ip);
1009 error = xfs_free_file_space(ip, offset, len);
1013 len = round_up(offset + len, blksize) -
1014 round_down(offset, blksize);
1015 offset = round_down(offset, blksize);
1016 } else if (mode & FALLOC_FL_UNSHARE_RANGE) {
1017 error = xfs_reflink_unshare(ip, offset, len);
1022 * If always_cow mode we can't use preallocations and
1023 * thus should not create them.
1025 if (xfs_is_always_cow_inode(ip)) {
1026 error = -EOPNOTSUPP;
1031 if (!xfs_is_always_cow_inode(ip)) {
1032 error = xfs_alloc_file_space(ip, offset, len);
1038 /* Change file size if needed */
1042 iattr.ia_valid = ATTR_SIZE;
1043 iattr.ia_size = new_size;
1044 error = xfs_vn_setattr_size(file_mnt_user_ns(file),
1045 file_dentry(file), &iattr);
1051 * Perform hole insertion now that the file size has been
1052 * updated so that if we crash during the operation we don't
1053 * leave shifted extents past EOF and hence losing access to
1054 * the data that is contained within them.
1056 if (do_file_insert) {
1057 error = xfs_insert_file_space(ip, offset, len);
1062 if (xfs_file_sync_writes(file))
1063 error = xfs_log_force_inode(ip);
1066 xfs_iunlock(ip, iolock);
1077 struct xfs_inode *ip = XFS_I(file_inode(file));
1082 * Operations creating pages in page cache need protection from hole
1083 * punching and similar ops
1085 if (advice == POSIX_FADV_WILLNEED) {
1086 lockflags = XFS_IOLOCK_SHARED;
1087 xfs_ilock(ip, lockflags);
1089 ret = generic_fadvise(file, start, end, advice);
1091 xfs_iunlock(ip, lockflags);
1096 xfs_file_remap_range(
1097 struct file *file_in,
1099 struct file *file_out,
1102 unsigned int remap_flags)
1104 struct inode *inode_in = file_inode(file_in);
1105 struct xfs_inode *src = XFS_I(inode_in);
1106 struct inode *inode_out = file_inode(file_out);
1107 struct xfs_inode *dest = XFS_I(inode_out);
1108 struct xfs_mount *mp = src->i_mount;
1109 loff_t remapped = 0;
1110 xfs_extlen_t cowextsize;
1113 if (remap_flags & ~(REMAP_FILE_DEDUP | REMAP_FILE_ADVISORY))
1116 if (!xfs_has_reflink(mp))
1119 if (xfs_is_shutdown(mp))
1122 /* Prepare and then clone file data. */
1123 ret = xfs_reflink_remap_prep(file_in, pos_in, file_out, pos_out,
1125 if (ret || len == 0)
1128 trace_xfs_reflink_remap_range(src, pos_in, len, dest, pos_out);
1130 ret = xfs_reflink_remap_blocks(src, pos_in, dest, pos_out, len,
1136 * Carry the cowextsize hint from src to dest if we're sharing the
1137 * entire source file to the entire destination file, the source file
1138 * has a cowextsize hint, and the destination file does not.
1141 if (pos_in == 0 && len == i_size_read(inode_in) &&
1142 (src->i_diflags2 & XFS_DIFLAG2_COWEXTSIZE) &&
1143 pos_out == 0 && len >= i_size_read(inode_out) &&
1144 !(dest->i_diflags2 & XFS_DIFLAG2_COWEXTSIZE))
1145 cowextsize = src->i_cowextsize;
1147 ret = xfs_reflink_update_dest(dest, pos_out + len, cowextsize,
1152 if (xfs_file_sync_writes(file_in) || xfs_file_sync_writes(file_out))
1153 xfs_log_force_inode(dest);
1155 xfs_iunlock2_io_mmap(src, dest);
1157 trace_xfs_reflink_remap_range_error(dest, ret, _RET_IP_);
1158 return remapped > 0 ? remapped : ret;
1163 struct inode *inode,
1166 if (xfs_is_shutdown(XFS_M(inode->i_sb)))
1168 file->f_mode |= FMODE_NOWAIT | FMODE_BUF_RASYNC | FMODE_BUF_WASYNC;
1169 return generic_file_open(inode, file);
1174 struct inode *inode,
1177 struct xfs_inode *ip = XFS_I(inode);
1181 error = xfs_file_open(inode, file);
1186 * If there are any blocks, read-ahead block 0 as we're almost
1187 * certain to have the next operation be a read there.
1189 mode = xfs_ilock_data_map_shared(ip);
1190 if (ip->i_df.if_nextents > 0)
1191 error = xfs_dir3_data_readahead(ip, 0, 0);
1192 xfs_iunlock(ip, mode);
1198 struct inode *inode,
1201 return xfs_release(XFS_I(inode));
1207 struct dir_context *ctx)
1209 struct inode *inode = file_inode(file);
1210 xfs_inode_t *ip = XFS_I(inode);
1214 * The Linux API doesn't pass down the total size of the buffer
1215 * we read into down to the filesystem. With the filldir concept
1216 * it's not needed for correct information, but the XFS dir2 leaf
1217 * code wants an estimate of the buffer size to calculate it's
1218 * readahead window and size the buffers used for mapping to
1221 * Try to give it an estimate that's good enough, maybe at some
1222 * point we can change the ->readdir prototype to include the
1223 * buffer size. For now we use the current glibc buffer size.
1225 bufsize = (size_t)min_t(loff_t, XFS_READDIR_BUFSIZE, ip->i_disk_size);
1227 return xfs_readdir(NULL, ip, ctx, bufsize);
1236 struct inode *inode = file->f_mapping->host;
1238 if (xfs_is_shutdown(XFS_I(inode)->i_mount))
1243 return generic_file_llseek(file, offset, whence);
1245 offset = iomap_seek_hole(inode, offset, &xfs_seek_iomap_ops);
1248 offset = iomap_seek_data(inode, offset, &xfs_seek_iomap_ops);
1254 return vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
1257 #ifdef CONFIG_FS_DAX
1260 struct vm_fault *vmf,
1261 enum page_entry_size pe_size,
1265 return dax_iomap_fault(vmf, pe_size, pfn, NULL,
1266 (write_fault && !vmf->cow_page) ?
1267 &xfs_dax_write_iomap_ops :
1268 &xfs_read_iomap_ops);
1273 struct vm_fault *vmf,
1274 enum page_entry_size pe_size,
1283 * Locking for serialisation of IO during page faults. This results in a lock
1287 * sb_start_pagefault(vfs, freeze)
1288 * invalidate_lock (vfs/XFS_MMAPLOCK - truncate serialisation)
1290 * i_lock (XFS - extent map serialisation)
1293 __xfs_filemap_fault(
1294 struct vm_fault *vmf,
1295 enum page_entry_size pe_size,
1298 struct inode *inode = file_inode(vmf->vma->vm_file);
1299 struct xfs_inode *ip = XFS_I(inode);
1302 trace_xfs_filemap_fault(ip, pe_size, write_fault);
1305 sb_start_pagefault(inode->i_sb);
1306 file_update_time(vmf->vma->vm_file);
1309 if (IS_DAX(inode)) {
1312 xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1313 ret = xfs_dax_fault(vmf, pe_size, write_fault, &pfn);
1314 if (ret & VM_FAULT_NEEDDSYNC)
1315 ret = dax_finish_sync_fault(vmf, pe_size, pfn);
1316 xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1319 xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1320 ret = iomap_page_mkwrite(vmf,
1321 &xfs_buffered_write_iomap_ops);
1322 xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1324 ret = filemap_fault(vmf);
1329 sb_end_pagefault(inode->i_sb);
1335 struct vm_fault *vmf)
1337 return (vmf->flags & FAULT_FLAG_WRITE) &&
1338 (vmf->vma->vm_flags & VM_SHARED);
1343 struct vm_fault *vmf)
1345 /* DAX can shortcut the normal fault path on write faults! */
1346 return __xfs_filemap_fault(vmf, PE_SIZE_PTE,
1347 IS_DAX(file_inode(vmf->vma->vm_file)) &&
1348 xfs_is_write_fault(vmf));
1352 xfs_filemap_huge_fault(
1353 struct vm_fault *vmf,
1354 enum page_entry_size pe_size)
1356 if (!IS_DAX(file_inode(vmf->vma->vm_file)))
1357 return VM_FAULT_FALLBACK;
1359 /* DAX can shortcut the normal fault path on write faults! */
1360 return __xfs_filemap_fault(vmf, pe_size,
1361 xfs_is_write_fault(vmf));
1365 xfs_filemap_page_mkwrite(
1366 struct vm_fault *vmf)
1368 return __xfs_filemap_fault(vmf, PE_SIZE_PTE, true);
1372 * pfn_mkwrite was originally intended to ensure we capture time stamp updates
1373 * on write faults. In reality, it needs to serialise against truncate and
1374 * prepare memory for writing so handle is as standard write fault.
1377 xfs_filemap_pfn_mkwrite(
1378 struct vm_fault *vmf)
1381 return __xfs_filemap_fault(vmf, PE_SIZE_PTE, true);
1385 xfs_filemap_map_pages(
1386 struct vm_fault *vmf,
1387 pgoff_t start_pgoff,
1390 struct inode *inode = file_inode(vmf->vma->vm_file);
1393 xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1394 ret = filemap_map_pages(vmf, start_pgoff, end_pgoff);
1395 xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1399 static const struct vm_operations_struct xfs_file_vm_ops = {
1400 .fault = xfs_filemap_fault,
1401 .huge_fault = xfs_filemap_huge_fault,
1402 .map_pages = xfs_filemap_map_pages,
1403 .page_mkwrite = xfs_filemap_page_mkwrite,
1404 .pfn_mkwrite = xfs_filemap_pfn_mkwrite,
1410 struct vm_area_struct *vma)
1412 struct inode *inode = file_inode(file);
1413 struct xfs_buftarg *target = xfs_inode_buftarg(XFS_I(inode));
1416 * We don't support synchronous mappings for non-DAX files and
1417 * for DAX files if underneath dax_device is not synchronous.
1419 if (!daxdev_mapping_supported(vma, target->bt_daxdev))
1422 file_accessed(file);
1423 vma->vm_ops = &xfs_file_vm_ops;
1425 vma->vm_flags |= VM_HUGEPAGE;
1429 const struct file_operations xfs_file_operations = {
1430 .llseek = xfs_file_llseek,
1431 .read_iter = xfs_file_read_iter,
1432 .write_iter = xfs_file_write_iter,
1433 .splice_read = generic_file_splice_read,
1434 .splice_write = iter_file_splice_write,
1435 .iopoll = iocb_bio_iopoll,
1436 .unlocked_ioctl = xfs_file_ioctl,
1437 #ifdef CONFIG_COMPAT
1438 .compat_ioctl = xfs_file_compat_ioctl,
1440 .mmap = xfs_file_mmap,
1441 .mmap_supported_flags = MAP_SYNC,
1442 .open = xfs_file_open,
1443 .release = xfs_file_release,
1444 .fsync = xfs_file_fsync,
1445 .get_unmapped_area = thp_get_unmapped_area,
1446 .fallocate = xfs_file_fallocate,
1447 .fadvise = xfs_file_fadvise,
1448 .remap_file_range = xfs_file_remap_range,
1451 const struct file_operations xfs_dir_file_operations = {
1452 .open = xfs_dir_open,
1453 .read = generic_read_dir,
1454 .iterate_shared = xfs_file_readdir,
1455 .llseek = generic_file_llseek,
1456 .unlocked_ioctl = xfs_file_ioctl,
1457 #ifdef CONFIG_COMPAT
1458 .compat_ioctl = xfs_file_compat_ioctl,
1460 .fsync = xfs_dir_fsync,