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);
84 trace_xfs_dir_fsync(ip);
85 return xfs_log_force_inode(ip);
95 struct inode *inode = file->f_mapping->host;
96 struct xfs_inode *ip = XFS_I(inode);
97 struct xfs_mount *mp = ip->i_mount;
102 trace_xfs_file_fsync(ip);
104 error = file_write_and_wait_range(file, start, end);
108 if (XFS_FORCED_SHUTDOWN(mp))
111 xfs_iflags_clear(ip, XFS_ITRUNCATED);
114 * If we have an RT and/or log subvolume we need to make sure to flush
115 * the write cache the device used for file data first. This is to
116 * ensure newly written file data make it to disk before logging the new
117 * inode size in case of an extending write.
119 if (XFS_IS_REALTIME_INODE(ip))
120 xfs_blkdev_issue_flush(mp->m_rtdev_targp);
121 else if (mp->m_logdev_targp != mp->m_ddev_targp)
122 xfs_blkdev_issue_flush(mp->m_ddev_targp);
125 * All metadata updates are logged, which means that we just have to
126 * flush the log up to the latest LSN that touched the inode. If we have
127 * concurrent fsync/fdatasync() calls, we need them to all block on the
128 * log force before we clear the ili_fsync_fields field. This ensures
129 * that we don't get a racing sync operation that does not wait for the
130 * metadata to hit the journal before returning. If we race with
131 * clearing the ili_fsync_fields, then all that will happen is the log
132 * force will do nothing as the lsn will already be on disk. We can't
133 * race with setting ili_fsync_fields because that is done under
134 * XFS_ILOCK_EXCL, and that can't happen because we hold the lock shared
135 * until after the ili_fsync_fields is cleared.
137 xfs_ilock(ip, XFS_ILOCK_SHARED);
138 if (xfs_ipincount(ip)) {
140 (ip->i_itemp->ili_fsync_fields & ~XFS_ILOG_TIMESTAMP))
141 lsn = ip->i_itemp->ili_last_lsn;
145 error = xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, &log_flushed);
146 ip->i_itemp->ili_fsync_fields = 0;
148 xfs_iunlock(ip, XFS_ILOCK_SHARED);
151 * If we only have a single device, and the log force about was
152 * a no-op we might have to flush the data device cache here.
153 * This can only happen for fdatasync/O_DSYNC if we were overwriting
154 * an already allocated file and thus do not have any metadata to
157 if (!log_flushed && !XFS_IS_REALTIME_INODE(ip) &&
158 mp->m_logdev_targp == mp->m_ddev_targp)
159 xfs_blkdev_issue_flush(mp->m_ddev_targp);
165 xfs_file_dio_aio_read(
169 struct xfs_inode *ip = XFS_I(file_inode(iocb->ki_filp));
170 size_t count = iov_iter_count(to);
173 trace_xfs_file_direct_read(ip, count, iocb->ki_pos);
176 return 0; /* skip atime */
178 file_accessed(iocb->ki_filp);
180 if (iocb->ki_flags & IOCB_NOWAIT) {
181 if (!xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED))
184 xfs_ilock(ip, XFS_IOLOCK_SHARED);
186 ret = iomap_dio_rw(iocb, to, &xfs_read_iomap_ops, NULL,
187 is_sync_kiocb(iocb));
188 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
193 static noinline ssize_t
198 struct xfs_inode *ip = XFS_I(iocb->ki_filp->f_mapping->host);
199 size_t count = iov_iter_count(to);
202 trace_xfs_file_dax_read(ip, count, iocb->ki_pos);
205 return 0; /* skip atime */
207 if (iocb->ki_flags & IOCB_NOWAIT) {
208 if (!xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED))
211 xfs_ilock(ip, XFS_IOLOCK_SHARED);
214 ret = dax_iomap_rw(iocb, to, &xfs_read_iomap_ops);
215 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
217 file_accessed(iocb->ki_filp);
222 xfs_file_buffered_aio_read(
226 struct xfs_inode *ip = XFS_I(file_inode(iocb->ki_filp));
229 trace_xfs_file_buffered_read(ip, iov_iter_count(to), iocb->ki_pos);
231 if (iocb->ki_flags & IOCB_NOWAIT) {
232 if (!xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED))
235 xfs_ilock(ip, XFS_IOLOCK_SHARED);
237 ret = generic_file_read_iter(iocb, to);
238 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
248 struct inode *inode = file_inode(iocb->ki_filp);
249 struct xfs_mount *mp = XFS_I(inode)->i_mount;
252 XFS_STATS_INC(mp, xs_read_calls);
254 if (XFS_FORCED_SHUTDOWN(mp))
258 ret = xfs_file_dax_read(iocb, to);
259 else if (iocb->ki_flags & IOCB_DIRECT)
260 ret = xfs_file_dio_aio_read(iocb, to);
262 ret = xfs_file_buffered_aio_read(iocb, to);
265 XFS_STATS_ADD(mp, xs_read_bytes, ret);
270 * Common pre-write limit and setup checks.
272 * Called with the iolocked held either shared and exclusive according to
273 * @iolock, and returns with it held. Might upgrade the iolock to exclusive
274 * if called for a direct write beyond i_size.
277 xfs_file_aio_write_checks(
279 struct iov_iter *from,
282 struct file *file = iocb->ki_filp;
283 struct inode *inode = file->f_mapping->host;
284 struct xfs_inode *ip = XFS_I(inode);
286 size_t count = iov_iter_count(from);
287 bool drained_dio = false;
291 error = generic_write_checks(iocb, from);
295 error = xfs_break_layouts(inode, iolock, BREAK_WRITE);
300 * For changing security info in file_remove_privs() we need i_rwsem
303 if (*iolock == XFS_IOLOCK_SHARED && !IS_NOSEC(inode)) {
304 xfs_iunlock(ip, *iolock);
305 *iolock = XFS_IOLOCK_EXCL;
306 xfs_ilock(ip, *iolock);
310 * If the offset is beyond the size of the file, we need to zero any
311 * blocks that fall between the existing EOF and the start of this
312 * write. If zeroing is needed and we are currently holding the
313 * iolock shared, we need to update it to exclusive which implies
314 * having to redo all checks before.
316 * We need to serialise against EOF updates that occur in IO
317 * completions here. We want to make sure that nobody is changing the
318 * size while we do this check until we have placed an IO barrier (i.e.
319 * hold the XFS_IOLOCK_EXCL) that prevents new IO from being dispatched.
320 * The spinlock effectively forms a memory barrier once we have the
321 * XFS_IOLOCK_EXCL so we are guaranteed to see the latest EOF value
322 * and hence be able to correctly determine if we need to run zeroing.
324 spin_lock(&ip->i_flags_lock);
325 isize = i_size_read(inode);
326 if (iocb->ki_pos > isize) {
327 spin_unlock(&ip->i_flags_lock);
329 if (*iolock == XFS_IOLOCK_SHARED) {
330 xfs_iunlock(ip, *iolock);
331 *iolock = XFS_IOLOCK_EXCL;
332 xfs_ilock(ip, *iolock);
333 iov_iter_reexpand(from, count);
336 * We now have an IO submission barrier in place, but
337 * AIO can do EOF updates during IO completion and hence
338 * we now need to wait for all of them to drain. Non-AIO
339 * DIO will have drained before we are given the
340 * XFS_IOLOCK_EXCL, and so for most cases this wait is a
343 inode_dio_wait(inode);
348 trace_xfs_zero_eof(ip, isize, iocb->ki_pos - isize);
349 error = iomap_zero_range(inode, isize, iocb->ki_pos - isize,
350 NULL, &xfs_buffered_write_iomap_ops);
354 spin_unlock(&ip->i_flags_lock);
357 * Updating the timestamps will grab the ilock again from
358 * xfs_fs_dirty_inode, so we have to call it after dropping the
359 * lock above. Eventually we should look into a way to avoid
360 * the pointless lock roundtrip.
362 return file_modified(file);
366 xfs_dio_write_end_io(
372 struct inode *inode = file_inode(iocb->ki_filp);
373 struct xfs_inode *ip = XFS_I(inode);
374 loff_t offset = iocb->ki_pos;
375 unsigned int nofs_flag;
377 trace_xfs_end_io_direct_write(ip, offset, size);
379 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
388 * Capture amount written on completion as we can't reliably account
389 * for it on submission.
391 XFS_STATS_ADD(ip->i_mount, xs_write_bytes, size);
394 * We can allocate memory here while doing writeback on behalf of
395 * memory reclaim. To avoid memory allocation deadlocks set the
396 * task-wide nofs context for the following operations.
398 nofs_flag = memalloc_nofs_save();
400 if (flags & IOMAP_DIO_COW) {
401 error = xfs_reflink_end_cow(ip, offset, size);
407 * Unwritten conversion updates the in-core isize after extent
408 * conversion but before updating the on-disk size. Updating isize any
409 * earlier allows a racing dio read to find unwritten extents before
410 * they are converted.
412 if (flags & IOMAP_DIO_UNWRITTEN) {
413 error = xfs_iomap_write_unwritten(ip, offset, size, true);
418 * We need to update the in-core inode size here so that we don't end up
419 * with the on-disk inode size being outside the in-core inode size. We
420 * have no other method of updating EOF for AIO, so always do it here
423 * We need to lock the test/set EOF update as we can be racing with
424 * other IO completions here to update the EOF. Failing to serialise
425 * here can result in EOF moving backwards and Bad Things Happen when
428 spin_lock(&ip->i_flags_lock);
429 if (offset + size > i_size_read(inode)) {
430 i_size_write(inode, offset + size);
431 spin_unlock(&ip->i_flags_lock);
432 error = xfs_setfilesize(ip, offset, size);
434 spin_unlock(&ip->i_flags_lock);
438 memalloc_nofs_restore(nofs_flag);
442 static const struct iomap_dio_ops xfs_dio_write_ops = {
443 .end_io = xfs_dio_write_end_io,
447 * xfs_file_dio_aio_write - handle direct IO writes
449 * Lock the inode appropriately to prepare for and issue a direct IO write.
450 * By separating it from the buffered write path we remove all the tricky to
451 * follow locking changes and looping.
453 * If there are cached pages or we're extending the file, we need IOLOCK_EXCL
454 * until we're sure the bytes at the new EOF have been zeroed and/or the cached
455 * pages are flushed out.
457 * In most cases the direct IO writes will be done holding IOLOCK_SHARED
458 * allowing them to be done in parallel with reads and other direct IO writes.
459 * However, if the IO is not aligned to filesystem blocks, the direct IO layer
460 * needs to do sub-block zeroing and that requires serialisation against other
461 * direct IOs to the same block. In this case we need to serialise the
462 * submission of the unaligned IOs so that we don't get racing block zeroing in
463 * the dio layer. To avoid the problem with aio, we also need to wait for
464 * outstanding IOs to complete so that unwritten extent conversion is completed
465 * before we try to map the overlapping block. This is currently implemented by
466 * hitting it with a big hammer (i.e. inode_dio_wait()).
468 * Returns with locks held indicated by @iolock and errors indicated by
469 * negative return values.
472 xfs_file_dio_aio_write(
474 struct iov_iter *from)
476 struct file *file = iocb->ki_filp;
477 struct address_space *mapping = file->f_mapping;
478 struct inode *inode = mapping->host;
479 struct xfs_inode *ip = XFS_I(inode);
480 struct xfs_mount *mp = ip->i_mount;
482 int unaligned_io = 0;
484 size_t count = iov_iter_count(from);
485 struct xfs_buftarg *target = xfs_inode_buftarg(ip);
487 /* DIO must be aligned to device logical sector size */
488 if ((iocb->ki_pos | count) & target->bt_logical_sectormask)
492 * Don't take the exclusive iolock here unless the I/O is unaligned to
493 * the file system block size. We don't need to consider the EOF
494 * extension case here because xfs_file_aio_write_checks() will relock
495 * the inode as necessary for EOF zeroing cases and fill out the new
496 * inode size as appropriate.
498 if ((iocb->ki_pos & mp->m_blockmask) ||
499 ((iocb->ki_pos + count) & mp->m_blockmask)) {
503 * We can't properly handle unaligned direct I/O to reflink
504 * files yet, as we can't unshare a partial block.
506 if (xfs_is_cow_inode(ip)) {
507 trace_xfs_reflink_bounce_dio_write(ip, iocb->ki_pos, count);
510 iolock = XFS_IOLOCK_EXCL;
512 iolock = XFS_IOLOCK_SHARED;
515 if (iocb->ki_flags & IOCB_NOWAIT) {
516 /* unaligned dio always waits, bail */
519 if (!xfs_ilock_nowait(ip, iolock))
522 xfs_ilock(ip, iolock);
525 ret = xfs_file_aio_write_checks(iocb, from, &iolock);
528 count = iov_iter_count(from);
531 * If we are doing unaligned IO, we can't allow any other overlapping IO
532 * in-flight at the same time or we risk data corruption. Wait for all
533 * other IO to drain before we submit. If the IO is aligned, demote the
534 * iolock if we had to take the exclusive lock in
535 * xfs_file_aio_write_checks() for other reasons.
538 inode_dio_wait(inode);
539 } else if (iolock == XFS_IOLOCK_EXCL) {
540 xfs_ilock_demote(ip, XFS_IOLOCK_EXCL);
541 iolock = XFS_IOLOCK_SHARED;
544 trace_xfs_file_direct_write(ip, count, iocb->ki_pos);
546 * If unaligned, this is the only IO in-flight. Wait on it before we
547 * release the iolock to prevent subsequent overlapping IO.
549 ret = iomap_dio_rw(iocb, from, &xfs_direct_write_iomap_ops,
551 is_sync_kiocb(iocb) || unaligned_io);
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_direct_write_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,
638 &xfs_buffered_write_iomap_ops);
639 if (likely(ret >= 0))
643 * If we hit a space limit, try to free up some lingering preallocated
644 * space before returning an error. In the case of ENOSPC, first try to
645 * write back all dirty inodes to free up some of the excess reserved
646 * metadata space. This reduces the chances that the eofblocks scan
647 * waits on dirty mappings. Since xfs_flush_inodes() is serialized, this
648 * also behaves as a filter to prevent too many eofblocks scans from
649 * running at the same time.
651 if (ret == -EDQUOT && !enospc) {
652 xfs_iunlock(ip, iolock);
653 enospc = xfs_inode_free_quota_eofblocks(ip);
656 enospc = xfs_inode_free_quota_cowblocks(ip);
660 } else if (ret == -ENOSPC && !enospc) {
661 struct xfs_eofblocks eofb = {0};
664 xfs_flush_inodes(ip->i_mount);
666 xfs_iunlock(ip, iolock);
667 eofb.eof_flags = XFS_EOF_FLAGS_SYNC;
668 xfs_icache_free_eofblocks(ip->i_mount, &eofb);
669 xfs_icache_free_cowblocks(ip->i_mount, &eofb);
673 current->backing_dev_info = NULL;
676 xfs_iunlock(ip, iolock);
679 XFS_STATS_ADD(ip->i_mount, xs_write_bytes, ret);
680 /* Handle various SYNC-type writes */
681 ret = generic_write_sync(iocb, ret);
689 struct iov_iter *from)
691 struct file *file = iocb->ki_filp;
692 struct address_space *mapping = file->f_mapping;
693 struct inode *inode = mapping->host;
694 struct xfs_inode *ip = XFS_I(inode);
696 size_t ocount = iov_iter_count(from);
698 XFS_STATS_INC(ip->i_mount, xs_write_calls);
703 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
707 return xfs_file_dax_write(iocb, from);
709 if (iocb->ki_flags & IOCB_DIRECT) {
711 * Allow a directio write to fall back to a buffered
712 * write *only* in the case that we're doing a reflink
713 * CoW. In all other directio scenarios we do not
714 * allow an operation to fall back to buffered mode.
716 ret = xfs_file_dio_aio_write(iocb, from);
721 return xfs_file_buffered_aio_write(iocb, from);
728 struct xfs_inode *ip = XFS_I(inode);
730 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
732 xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
736 xfs_break_dax_layouts(
742 ASSERT(xfs_isilocked(XFS_I(inode), XFS_MMAPLOCK_EXCL));
744 page = dax_layout_busy_page(inode->i_mapping);
749 return ___wait_var_event(&page->_refcount,
750 atomic_read(&page->_refcount) == 1, TASK_INTERRUPTIBLE,
751 0, 0, xfs_wait_dax_page(inode));
758 enum layout_break_reason reason)
763 ASSERT(xfs_isilocked(XFS_I(inode), XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL));
769 error = xfs_break_dax_layouts(inode, &retry);
774 error = xfs_break_leased_layouts(inode, iolock, &retry);
780 } while (error == 0 && retry);
785 #define XFS_FALLOC_FL_SUPPORTED \
786 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
787 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE | \
788 FALLOC_FL_INSERT_RANGE | FALLOC_FL_UNSHARE_RANGE)
797 struct inode *inode = file_inode(file);
798 struct xfs_inode *ip = XFS_I(inode);
800 enum xfs_prealloc_flags flags = 0;
801 uint iolock = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL;
803 bool do_file_insert = false;
805 if (!S_ISREG(inode->i_mode))
807 if (mode & ~XFS_FALLOC_FL_SUPPORTED)
810 xfs_ilock(ip, iolock);
811 error = xfs_break_layouts(inode, &iolock, BREAK_UNMAP);
816 * Must wait for all AIO to complete before we continue as AIO can
817 * change the file size on completion without holding any locks we
818 * currently hold. We must do this first because AIO can update both
819 * the on disk and in memory inode sizes, and the operations that follow
820 * require the in-memory size to be fully up-to-date.
822 inode_dio_wait(inode);
825 * Now AIO and DIO has drained we flush and (if necessary) invalidate
826 * the cached range over the first operation we are about to run.
828 * We care about zero and collapse here because they both run a hole
829 * punch over the range first. Because that can zero data, and the range
830 * of invalidation for the shift operations is much larger, we still do
831 * the required flush for collapse in xfs_prepare_shift().
833 * Insert has the same range requirements as collapse, and we extend the
834 * file first which can zero data. Hence insert has the same
835 * flush/invalidate requirements as collapse and so they are both
836 * handled at the right time by xfs_prepare_shift().
838 if (mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE |
839 FALLOC_FL_COLLAPSE_RANGE)) {
840 error = xfs_flush_unmap_range(ip, offset, len);
845 if (mode & FALLOC_FL_PUNCH_HOLE) {
846 error = xfs_free_file_space(ip, offset, len);
849 } else if (mode & FALLOC_FL_COLLAPSE_RANGE) {
850 unsigned int blksize_mask = i_blocksize(inode) - 1;
852 if (offset & blksize_mask || len & blksize_mask) {
858 * There is no need to overlap collapse range with EOF,
859 * in which case it is effectively a truncate operation
861 if (offset + len >= i_size_read(inode)) {
866 new_size = i_size_read(inode) - len;
868 error = xfs_collapse_file_space(ip, offset, len);
871 } else if (mode & FALLOC_FL_INSERT_RANGE) {
872 unsigned int blksize_mask = i_blocksize(inode) - 1;
873 loff_t isize = i_size_read(inode);
875 if (offset & blksize_mask || len & blksize_mask) {
881 * New inode size must not exceed ->s_maxbytes, accounting for
882 * possible signed overflow.
884 if (inode->i_sb->s_maxbytes - isize < len) {
888 new_size = isize + len;
890 /* Offset should be less than i_size */
891 if (offset >= isize) {
895 do_file_insert = true;
897 flags |= XFS_PREALLOC_SET;
899 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
900 offset + len > i_size_read(inode)) {
901 new_size = offset + len;
902 error = inode_newsize_ok(inode, new_size);
907 if (mode & FALLOC_FL_ZERO_RANGE) {
909 * Punch a hole and prealloc the range. We use a hole
910 * punch rather than unwritten extent conversion for two
913 * 1.) Hole punch handles partial block zeroing for us.
914 * 2.) If prealloc returns ENOSPC, the file range is
915 * still zero-valued by virtue of the hole punch.
917 unsigned int blksize = i_blocksize(inode);
919 trace_xfs_zero_file_space(ip);
921 error = xfs_free_file_space(ip, offset, len);
925 len = round_up(offset + len, blksize) -
926 round_down(offset, blksize);
927 offset = round_down(offset, blksize);
928 } else if (mode & FALLOC_FL_UNSHARE_RANGE) {
929 error = xfs_reflink_unshare(ip, offset, len);
934 * If always_cow mode we can't use preallocations and
935 * thus should not create them.
937 if (xfs_is_always_cow_inode(ip)) {
943 if (!xfs_is_always_cow_inode(ip)) {
944 error = xfs_alloc_file_space(ip, offset, len,
951 if (file->f_flags & O_DSYNC)
952 flags |= XFS_PREALLOC_SYNC;
954 error = xfs_update_prealloc_flags(ip, flags);
958 /* Change file size if needed */
962 iattr.ia_valid = ATTR_SIZE;
963 iattr.ia_size = new_size;
964 error = xfs_vn_setattr_size(file_dentry(file), &iattr);
970 * Perform hole insertion now that the file size has been
971 * updated so that if we crash during the operation we don't
972 * leave shifted extents past EOF and hence losing access to
973 * the data that is contained within them.
976 error = xfs_insert_file_space(ip, offset, len);
979 xfs_iunlock(ip, iolock);
990 struct xfs_inode *ip = XFS_I(file_inode(file));
995 * Operations creating pages in page cache need protection from hole
996 * punching and similar ops
998 if (advice == POSIX_FADV_WILLNEED) {
999 lockflags = XFS_IOLOCK_SHARED;
1000 xfs_ilock(ip, lockflags);
1002 ret = generic_fadvise(file, start, end, advice);
1004 xfs_iunlock(ip, lockflags);
1009 xfs_file_remap_range(
1010 struct file *file_in,
1012 struct file *file_out,
1015 unsigned int remap_flags)
1017 struct inode *inode_in = file_inode(file_in);
1018 struct xfs_inode *src = XFS_I(inode_in);
1019 struct inode *inode_out = file_inode(file_out);
1020 struct xfs_inode *dest = XFS_I(inode_out);
1021 struct xfs_mount *mp = src->i_mount;
1022 loff_t remapped = 0;
1023 xfs_extlen_t cowextsize;
1026 if (remap_flags & ~(REMAP_FILE_DEDUP | REMAP_FILE_ADVISORY))
1029 if (!xfs_sb_version_hasreflink(&mp->m_sb))
1032 if (XFS_FORCED_SHUTDOWN(mp))
1035 /* Prepare and then clone file data. */
1036 ret = xfs_reflink_remap_prep(file_in, pos_in, file_out, pos_out,
1038 if (ret < 0 || len == 0)
1041 trace_xfs_reflink_remap_range(src, pos_in, len, dest, pos_out);
1043 ret = xfs_reflink_remap_blocks(src, pos_in, dest, pos_out, len,
1049 * Carry the cowextsize hint from src to dest if we're sharing the
1050 * entire source file to the entire destination file, the source file
1051 * has a cowextsize hint, and the destination file does not.
1054 if (pos_in == 0 && len == i_size_read(inode_in) &&
1055 (src->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE) &&
1056 pos_out == 0 && len >= i_size_read(inode_out) &&
1057 !(dest->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE))
1058 cowextsize = src->i_d.di_cowextsize;
1060 ret = xfs_reflink_update_dest(dest, pos_out + len, cowextsize,
1065 if (mp->m_flags & XFS_MOUNT_WSYNC)
1066 xfs_log_force_inode(dest);
1068 xfs_reflink_remap_unlock(file_in, file_out);
1070 trace_xfs_reflink_remap_range_error(dest, ret, _RET_IP_);
1071 return remapped > 0 ? remapped : ret;
1076 struct inode *inode,
1079 if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1081 if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb)))
1083 file->f_mode |= FMODE_NOWAIT;
1089 struct inode *inode,
1092 struct xfs_inode *ip = XFS_I(inode);
1096 error = xfs_file_open(inode, file);
1101 * If there are any blocks, read-ahead block 0 as we're almost
1102 * certain to have the next operation be a read there.
1104 mode = xfs_ilock_data_map_shared(ip);
1105 if (ip->i_df.if_nextents > 0)
1106 error = xfs_dir3_data_readahead(ip, 0, 0);
1107 xfs_iunlock(ip, mode);
1113 struct inode *inode,
1116 return xfs_release(XFS_I(inode));
1122 struct dir_context *ctx)
1124 struct inode *inode = file_inode(file);
1125 xfs_inode_t *ip = XFS_I(inode);
1129 * The Linux API doesn't pass down the total size of the buffer
1130 * we read into down to the filesystem. With the filldir concept
1131 * it's not needed for correct information, but the XFS dir2 leaf
1132 * code wants an estimate of the buffer size to calculate it's
1133 * readahead window and size the buffers used for mapping to
1136 * Try to give it an estimate that's good enough, maybe at some
1137 * point we can change the ->readdir prototype to include the
1138 * buffer size. For now we use the current glibc buffer size.
1140 bufsize = (size_t)min_t(loff_t, XFS_READDIR_BUFSIZE, ip->i_d.di_size);
1142 return xfs_readdir(NULL, ip, ctx, bufsize);
1151 struct inode *inode = file->f_mapping->host;
1153 if (XFS_FORCED_SHUTDOWN(XFS_I(inode)->i_mount))
1158 return generic_file_llseek(file, offset, whence);
1160 offset = iomap_seek_hole(inode, offset, &xfs_seek_iomap_ops);
1163 offset = iomap_seek_data(inode, offset, &xfs_seek_iomap_ops);
1169 return vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
1173 * Locking for serialisation of IO during page faults. This results in a lock
1177 * sb_start_pagefault(vfs, freeze)
1178 * i_mmaplock (XFS - truncate serialisation)
1180 * i_lock (XFS - extent map serialisation)
1183 __xfs_filemap_fault(
1184 struct vm_fault *vmf,
1185 enum page_entry_size pe_size,
1188 struct inode *inode = file_inode(vmf->vma->vm_file);
1189 struct xfs_inode *ip = XFS_I(inode);
1192 trace_xfs_filemap_fault(ip, pe_size, write_fault);
1195 sb_start_pagefault(inode->i_sb);
1196 file_update_time(vmf->vma->vm_file);
1199 xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1200 if (IS_DAX(inode)) {
1203 ret = dax_iomap_fault(vmf, pe_size, &pfn, NULL,
1204 (write_fault && !vmf->cow_page) ?
1205 &xfs_direct_write_iomap_ops :
1206 &xfs_read_iomap_ops);
1207 if (ret & VM_FAULT_NEEDDSYNC)
1208 ret = dax_finish_sync_fault(vmf, pe_size, pfn);
1211 ret = iomap_page_mkwrite(vmf,
1212 &xfs_buffered_write_iomap_ops);
1214 ret = filemap_fault(vmf);
1216 xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1219 sb_end_pagefault(inode->i_sb);
1225 struct vm_fault *vmf)
1227 /* DAX can shortcut the normal fault path on write faults! */
1228 return __xfs_filemap_fault(vmf, PE_SIZE_PTE,
1229 IS_DAX(file_inode(vmf->vma->vm_file)) &&
1230 (vmf->flags & FAULT_FLAG_WRITE));
1234 xfs_filemap_huge_fault(
1235 struct vm_fault *vmf,
1236 enum page_entry_size pe_size)
1238 if (!IS_DAX(file_inode(vmf->vma->vm_file)))
1239 return VM_FAULT_FALLBACK;
1241 /* DAX can shortcut the normal fault path on write faults! */
1242 return __xfs_filemap_fault(vmf, pe_size,
1243 (vmf->flags & FAULT_FLAG_WRITE));
1247 xfs_filemap_page_mkwrite(
1248 struct vm_fault *vmf)
1250 return __xfs_filemap_fault(vmf, PE_SIZE_PTE, true);
1254 * pfn_mkwrite was originally intended to ensure we capture time stamp updates
1255 * on write faults. In reality, it needs to serialise against truncate and
1256 * prepare memory for writing so handle is as standard write fault.
1259 xfs_filemap_pfn_mkwrite(
1260 struct vm_fault *vmf)
1263 return __xfs_filemap_fault(vmf, PE_SIZE_PTE, true);
1266 static const struct vm_operations_struct xfs_file_vm_ops = {
1267 .fault = xfs_filemap_fault,
1268 .huge_fault = xfs_filemap_huge_fault,
1269 .map_pages = filemap_map_pages,
1270 .page_mkwrite = xfs_filemap_page_mkwrite,
1271 .pfn_mkwrite = xfs_filemap_pfn_mkwrite,
1277 struct vm_area_struct *vma)
1279 struct inode *inode = file_inode(file);
1280 struct xfs_buftarg *target = xfs_inode_buftarg(XFS_I(inode));
1283 * We don't support synchronous mappings for non-DAX files and
1284 * for DAX files if underneath dax_device is not synchronous.
1286 if (!daxdev_mapping_supported(vma, target->bt_daxdev))
1289 file_accessed(file);
1290 vma->vm_ops = &xfs_file_vm_ops;
1292 vma->vm_flags |= VM_HUGEPAGE;
1296 const struct file_operations xfs_file_operations = {
1297 .llseek = xfs_file_llseek,
1298 .read_iter = xfs_file_read_iter,
1299 .write_iter = xfs_file_write_iter,
1300 .splice_read = generic_file_splice_read,
1301 .splice_write = iter_file_splice_write,
1302 .iopoll = iomap_dio_iopoll,
1303 .unlocked_ioctl = xfs_file_ioctl,
1304 #ifdef CONFIG_COMPAT
1305 .compat_ioctl = xfs_file_compat_ioctl,
1307 .mmap = xfs_file_mmap,
1308 .mmap_supported_flags = MAP_SYNC,
1309 .open = xfs_file_open,
1310 .release = xfs_file_release,
1311 .fsync = xfs_file_fsync,
1312 .get_unmapped_area = thp_get_unmapped_area,
1313 .fallocate = xfs_file_fallocate,
1314 .fadvise = xfs_file_fadvise,
1315 .remap_file_range = xfs_file_remap_range,
1318 const struct file_operations xfs_dir_file_operations = {
1319 .open = xfs_dir_open,
1320 .read = generic_read_dir,
1321 .iterate_shared = xfs_file_readdir,
1322 .llseek = generic_file_llseek,
1323 .unlocked_ioctl = xfs_file_ioctl,
1324 #ifdef CONFIG_COMPAT
1325 .compat_ioctl = xfs_file_compat_ioctl,
1327 .fsync = xfs_dir_fsync,