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_inode_log_item *iip = ip->i_itemp;
98 struct xfs_mount *mp = ip->i_mount;
103 trace_xfs_file_fsync(ip);
105 error = file_write_and_wait_range(file, start, end);
109 if (XFS_FORCED_SHUTDOWN(mp))
112 xfs_iflags_clear(ip, XFS_ITRUNCATED);
115 * If we have an RT and/or log subvolume we need to make sure to flush
116 * the write cache the device used for file data first. This is to
117 * ensure newly written file data make it to disk before logging the new
118 * inode size in case of an extending write.
120 if (XFS_IS_REALTIME_INODE(ip))
121 xfs_blkdev_issue_flush(mp->m_rtdev_targp);
122 else if (mp->m_logdev_targp != mp->m_ddev_targp)
123 xfs_blkdev_issue_flush(mp->m_ddev_targp);
126 * All metadata updates are logged, which means that we just have to
127 * flush the log up to the latest LSN that touched the inode. If we have
128 * concurrent fsync/fdatasync() calls, we need them to all block on the
129 * log force before we clear the ili_fsync_fields field. This ensures
130 * that we don't get a racing sync operation that does not wait for the
131 * metadata to hit the journal before returning. If we race with
132 * clearing the ili_fsync_fields, then all that will happen is the log
133 * force will do nothing as the lsn will already be on disk. We can't
134 * race with setting ili_fsync_fields because that is done under
135 * XFS_ILOCK_EXCL, and that can't happen because we hold the lock shared
136 * until after the ili_fsync_fields is cleared.
138 xfs_ilock(ip, XFS_ILOCK_SHARED);
139 if (xfs_ipincount(ip)) {
141 (iip->ili_fsync_fields & ~XFS_ILOG_TIMESTAMP))
142 lsn = iip->ili_last_lsn;
146 error = xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, &log_flushed);
147 spin_lock(&iip->ili_lock);
148 iip->ili_fsync_fields = 0;
149 spin_unlock(&iip->ili_lock);
151 xfs_iunlock(ip, XFS_ILOCK_SHARED);
154 * If we only have a single device, and the log force about was
155 * a no-op we might have to flush the data device cache here.
156 * This can only happen for fdatasync/O_DSYNC if we were overwriting
157 * an already allocated file and thus do not have any metadata to
160 if (!log_flushed && !XFS_IS_REALTIME_INODE(ip) &&
161 mp->m_logdev_targp == mp->m_ddev_targp)
162 xfs_blkdev_issue_flush(mp->m_ddev_targp);
168 xfs_file_dio_aio_read(
172 struct xfs_inode *ip = XFS_I(file_inode(iocb->ki_filp));
173 size_t count = iov_iter_count(to);
176 trace_xfs_file_direct_read(ip, count, iocb->ki_pos);
179 return 0; /* skip atime */
181 file_accessed(iocb->ki_filp);
183 if (iocb->ki_flags & IOCB_NOWAIT) {
184 if (!xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED))
187 xfs_ilock(ip, XFS_IOLOCK_SHARED);
189 ret = iomap_dio_rw(iocb, to, &xfs_read_iomap_ops, NULL,
190 is_sync_kiocb(iocb));
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_read_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_buffered_write_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(
375 struct inode *inode = file_inode(iocb->ki_filp);
376 struct xfs_inode *ip = XFS_I(inode);
377 loff_t offset = iocb->ki_pos;
378 unsigned int nofs_flag;
380 trace_xfs_end_io_direct_write(ip, offset, size);
382 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
391 * Capture amount written on completion as we can't reliably account
392 * for it on submission.
394 XFS_STATS_ADD(ip->i_mount, xs_write_bytes, size);
397 * We can allocate memory here while doing writeback on behalf of
398 * memory reclaim. To avoid memory allocation deadlocks set the
399 * task-wide nofs context for the following operations.
401 nofs_flag = memalloc_nofs_save();
403 if (flags & IOMAP_DIO_COW) {
404 error = xfs_reflink_end_cow(ip, offset, size);
410 * Unwritten conversion updates the in-core isize after extent
411 * conversion but before updating the on-disk size. Updating isize any
412 * earlier allows a racing dio read to find unwritten extents before
413 * they are converted.
415 if (flags & IOMAP_DIO_UNWRITTEN) {
416 error = xfs_iomap_write_unwritten(ip, offset, size, true);
421 * We need to update the in-core inode size here so that we don't end up
422 * with the on-disk inode size being outside the in-core inode size. We
423 * have no other method of updating EOF for AIO, so always do it here
426 * We need to lock the test/set EOF update as we can be racing with
427 * other IO completions here to update the EOF. Failing to serialise
428 * here can result in EOF moving backwards and Bad Things Happen when
431 spin_lock(&ip->i_flags_lock);
432 if (offset + size > i_size_read(inode)) {
433 i_size_write(inode, offset + size);
434 spin_unlock(&ip->i_flags_lock);
435 error = xfs_setfilesize(ip, offset, size);
437 spin_unlock(&ip->i_flags_lock);
441 memalloc_nofs_restore(nofs_flag);
445 static const struct iomap_dio_ops xfs_dio_write_ops = {
446 .end_io = xfs_dio_write_end_io,
450 * xfs_file_dio_aio_write - handle direct IO writes
452 * Lock the inode appropriately to prepare for and issue a direct IO write.
453 * By separating it from the buffered write path we remove all the tricky to
454 * follow locking changes and looping.
456 * If there are cached pages or we're extending the file, we need IOLOCK_EXCL
457 * until we're sure the bytes at the new EOF have been zeroed and/or the cached
458 * pages are flushed out.
460 * In most cases the direct IO writes will be done holding IOLOCK_SHARED
461 * allowing them to be done in parallel with reads and other direct IO writes.
462 * However, if the IO is not aligned to filesystem blocks, the direct IO layer
463 * needs to do sub-block zeroing and that requires serialisation against other
464 * direct IOs to the same block. In this case we need to serialise the
465 * submission of the unaligned IOs so that we don't get racing block zeroing in
466 * the dio layer. To avoid the problem with aio, we also need to wait for
467 * outstanding IOs to complete so that unwritten extent conversion is completed
468 * before we try to map the overlapping block. This is currently implemented by
469 * hitting it with a big hammer (i.e. inode_dio_wait()).
471 * Returns with locks held indicated by @iolock and errors indicated by
472 * negative return values.
475 xfs_file_dio_aio_write(
477 struct iov_iter *from)
479 struct file *file = iocb->ki_filp;
480 struct address_space *mapping = file->f_mapping;
481 struct inode *inode = mapping->host;
482 struct xfs_inode *ip = XFS_I(inode);
483 struct xfs_mount *mp = ip->i_mount;
485 int unaligned_io = 0;
487 size_t count = iov_iter_count(from);
488 struct xfs_buftarg *target = xfs_inode_buftarg(ip);
490 /* DIO must be aligned to device logical sector size */
491 if ((iocb->ki_pos | count) & target->bt_logical_sectormask)
495 * Don't take the exclusive iolock here unless the I/O is unaligned to
496 * the file system block size. We don't need to consider the EOF
497 * extension case here because xfs_file_aio_write_checks() will relock
498 * the inode as necessary for EOF zeroing cases and fill out the new
499 * inode size as appropriate.
501 if ((iocb->ki_pos & mp->m_blockmask) ||
502 ((iocb->ki_pos + count) & mp->m_blockmask)) {
506 * We can't properly handle unaligned direct I/O to reflink
507 * files yet, as we can't unshare a partial block.
509 if (xfs_is_cow_inode(ip)) {
510 trace_xfs_reflink_bounce_dio_write(ip, iocb->ki_pos, count);
513 iolock = XFS_IOLOCK_EXCL;
515 iolock = XFS_IOLOCK_SHARED;
518 if (iocb->ki_flags & IOCB_NOWAIT) {
519 /* unaligned dio always waits, bail */
522 if (!xfs_ilock_nowait(ip, iolock))
525 xfs_ilock(ip, iolock);
528 ret = xfs_file_aio_write_checks(iocb, from, &iolock);
531 count = iov_iter_count(from);
534 * If we are doing unaligned IO, we can't allow any other overlapping IO
535 * in-flight at the same time or we risk data corruption. Wait for all
536 * other IO to drain before we submit. If the IO is aligned, demote the
537 * iolock if we had to take the exclusive lock in
538 * xfs_file_aio_write_checks() for other reasons.
541 inode_dio_wait(inode);
542 } else if (iolock == XFS_IOLOCK_EXCL) {
543 xfs_ilock_demote(ip, XFS_IOLOCK_EXCL);
544 iolock = XFS_IOLOCK_SHARED;
547 trace_xfs_file_direct_write(ip, count, iocb->ki_pos);
549 * If unaligned, this is the only IO in-flight. Wait on it before we
550 * release the iolock to prevent subsequent overlapping IO.
552 ret = iomap_dio_rw(iocb, from, &xfs_direct_write_iomap_ops,
554 is_sync_kiocb(iocb) || unaligned_io);
556 xfs_iunlock(ip, iolock);
559 * No fallback to buffered IO after short writes for XFS, direct I/O
560 * will either complete fully or return an error.
562 ASSERT(ret < 0 || ret == count);
566 static noinline ssize_t
569 struct iov_iter *from)
571 struct inode *inode = iocb->ki_filp->f_mapping->host;
572 struct xfs_inode *ip = XFS_I(inode);
573 int iolock = XFS_IOLOCK_EXCL;
574 ssize_t ret, error = 0;
578 if (iocb->ki_flags & IOCB_NOWAIT) {
579 if (!xfs_ilock_nowait(ip, iolock))
582 xfs_ilock(ip, iolock);
585 ret = xfs_file_aio_write_checks(iocb, from, &iolock);
590 count = iov_iter_count(from);
592 trace_xfs_file_dax_write(ip, count, pos);
593 ret = dax_iomap_rw(iocb, from, &xfs_direct_write_iomap_ops);
594 if (ret > 0 && iocb->ki_pos > i_size_read(inode)) {
595 i_size_write(inode, iocb->ki_pos);
596 error = xfs_setfilesize(ip, pos, ret);
599 xfs_iunlock(ip, iolock);
604 XFS_STATS_ADD(ip->i_mount, xs_write_bytes, ret);
606 /* Handle various SYNC-type writes */
607 ret = generic_write_sync(iocb, ret);
613 xfs_file_buffered_aio_write(
615 struct iov_iter *from)
617 struct file *file = iocb->ki_filp;
618 struct address_space *mapping = file->f_mapping;
619 struct inode *inode = mapping->host;
620 struct xfs_inode *ip = XFS_I(inode);
625 if (iocb->ki_flags & IOCB_NOWAIT)
629 iolock = XFS_IOLOCK_EXCL;
630 xfs_ilock(ip, iolock);
632 ret = xfs_file_aio_write_checks(iocb, from, &iolock);
636 /* We can write back this queue in page reclaim */
637 current->backing_dev_info = inode_to_bdi(inode);
639 trace_xfs_file_buffered_write(ip, iov_iter_count(from), iocb->ki_pos);
640 ret = iomap_file_buffered_write(iocb, from,
641 &xfs_buffered_write_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);
819 * Must wait for all AIO to complete before we continue as AIO can
820 * change the file size on completion without holding any locks we
821 * currently hold. We must do this first because AIO can update both
822 * the on disk and in memory inode sizes, and the operations that follow
823 * require the in-memory size to be fully up-to-date.
825 inode_dio_wait(inode);
828 * Now AIO and DIO has drained we flush and (if necessary) invalidate
829 * the cached range over the first operation we are about to run.
831 * We care about zero and collapse here because they both run a hole
832 * punch over the range first. Because that can zero data, and the range
833 * of invalidation for the shift operations is much larger, we still do
834 * the required flush for collapse in xfs_prepare_shift().
836 * Insert has the same range requirements as collapse, and we extend the
837 * file first which can zero data. Hence insert has the same
838 * flush/invalidate requirements as collapse and so they are both
839 * handled at the right time by xfs_prepare_shift().
841 if (mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE |
842 FALLOC_FL_COLLAPSE_RANGE)) {
843 error = xfs_flush_unmap_range(ip, offset, len);
848 if (mode & FALLOC_FL_PUNCH_HOLE) {
849 error = xfs_free_file_space(ip, offset, len);
852 } else if (mode & FALLOC_FL_COLLAPSE_RANGE) {
853 unsigned int blksize_mask = i_blocksize(inode) - 1;
855 if (offset & blksize_mask || len & blksize_mask) {
861 * There is no need to overlap collapse range with EOF,
862 * in which case it is effectively a truncate operation
864 if (offset + len >= i_size_read(inode)) {
869 new_size = i_size_read(inode) - len;
871 error = xfs_collapse_file_space(ip, offset, len);
874 } else if (mode & FALLOC_FL_INSERT_RANGE) {
875 unsigned int blksize_mask = i_blocksize(inode) - 1;
876 loff_t isize = i_size_read(inode);
878 if (offset & blksize_mask || len & blksize_mask) {
884 * New inode size must not exceed ->s_maxbytes, accounting for
885 * possible signed overflow.
887 if (inode->i_sb->s_maxbytes - isize < len) {
891 new_size = isize + len;
893 /* Offset should be less than i_size */
894 if (offset >= isize) {
898 do_file_insert = true;
900 flags |= XFS_PREALLOC_SET;
902 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
903 offset + len > i_size_read(inode)) {
904 new_size = offset + len;
905 error = inode_newsize_ok(inode, new_size);
910 if (mode & FALLOC_FL_ZERO_RANGE) {
912 * Punch a hole and prealloc the range. We use a hole
913 * punch rather than unwritten extent conversion for two
916 * 1.) Hole punch handles partial block zeroing for us.
917 * 2.) If prealloc returns ENOSPC, the file range is
918 * still zero-valued by virtue of the hole punch.
920 unsigned int blksize = i_blocksize(inode);
922 trace_xfs_zero_file_space(ip);
924 error = xfs_free_file_space(ip, offset, len);
928 len = round_up(offset + len, blksize) -
929 round_down(offset, blksize);
930 offset = round_down(offset, blksize);
931 } else if (mode & FALLOC_FL_UNSHARE_RANGE) {
932 error = xfs_reflink_unshare(ip, offset, len);
937 * If always_cow mode we can't use preallocations and
938 * thus should not create them.
940 if (xfs_is_always_cow_inode(ip)) {
946 if (!xfs_is_always_cow_inode(ip)) {
947 error = xfs_alloc_file_space(ip, offset, len,
954 if (file->f_flags & O_DSYNC)
955 flags |= XFS_PREALLOC_SYNC;
957 error = xfs_update_prealloc_flags(ip, flags);
961 /* Change file size if needed */
965 iattr.ia_valid = ATTR_SIZE;
966 iattr.ia_size = new_size;
967 error = xfs_vn_setattr_size(file_dentry(file), &iattr);
973 * Perform hole insertion now that the file size has been
974 * updated so that if we crash during the operation we don't
975 * leave shifted extents past EOF and hence losing access to
976 * the data that is contained within them.
979 error = xfs_insert_file_space(ip, offset, len);
982 xfs_iunlock(ip, iolock);
993 struct xfs_inode *ip = XFS_I(file_inode(file));
998 * Operations creating pages in page cache need protection from hole
999 * punching and similar ops
1001 if (advice == POSIX_FADV_WILLNEED) {
1002 lockflags = XFS_IOLOCK_SHARED;
1003 xfs_ilock(ip, lockflags);
1005 ret = generic_fadvise(file, start, end, advice);
1007 xfs_iunlock(ip, lockflags);
1012 xfs_file_remap_range(
1013 struct file *file_in,
1015 struct file *file_out,
1018 unsigned int remap_flags)
1020 struct inode *inode_in = file_inode(file_in);
1021 struct xfs_inode *src = XFS_I(inode_in);
1022 struct inode *inode_out = file_inode(file_out);
1023 struct xfs_inode *dest = XFS_I(inode_out);
1024 struct xfs_mount *mp = src->i_mount;
1025 loff_t remapped = 0;
1026 xfs_extlen_t cowextsize;
1029 if (remap_flags & ~(REMAP_FILE_DEDUP | REMAP_FILE_ADVISORY))
1032 if (!xfs_sb_version_hasreflink(&mp->m_sb))
1035 if (XFS_FORCED_SHUTDOWN(mp))
1038 /* Prepare and then clone file data. */
1039 ret = xfs_reflink_remap_prep(file_in, pos_in, file_out, pos_out,
1041 if (ret || len == 0)
1044 trace_xfs_reflink_remap_range(src, pos_in, len, dest, pos_out);
1046 ret = xfs_reflink_remap_blocks(src, pos_in, dest, pos_out, len,
1052 * Carry the cowextsize hint from src to dest if we're sharing the
1053 * entire source file to the entire destination file, the source file
1054 * has a cowextsize hint, and the destination file does not.
1057 if (pos_in == 0 && len == i_size_read(inode_in) &&
1058 (src->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE) &&
1059 pos_out == 0 && len >= i_size_read(inode_out) &&
1060 !(dest->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE))
1061 cowextsize = src->i_d.di_cowextsize;
1063 ret = xfs_reflink_update_dest(dest, pos_out + len, cowextsize,
1068 if (mp->m_flags & XFS_MOUNT_WSYNC)
1069 xfs_log_force_inode(dest);
1071 xfs_iunlock2_io_mmap(src, dest);
1073 trace_xfs_reflink_remap_range_error(dest, ret, _RET_IP_);
1074 return remapped > 0 ? remapped : ret;
1079 struct inode *inode,
1082 if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1084 if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb)))
1086 file->f_mode |= FMODE_NOWAIT | FMODE_BUF_RASYNC;
1092 struct inode *inode,
1095 struct xfs_inode *ip = XFS_I(inode);
1099 error = xfs_file_open(inode, file);
1104 * If there are any blocks, read-ahead block 0 as we're almost
1105 * certain to have the next operation be a read there.
1107 mode = xfs_ilock_data_map_shared(ip);
1108 if (ip->i_df.if_nextents > 0)
1109 error = xfs_dir3_data_readahead(ip, 0, 0);
1110 xfs_iunlock(ip, mode);
1116 struct inode *inode,
1119 return xfs_release(XFS_I(inode));
1125 struct dir_context *ctx)
1127 struct inode *inode = file_inode(file);
1128 xfs_inode_t *ip = XFS_I(inode);
1132 * The Linux API doesn't pass down the total size of the buffer
1133 * we read into down to the filesystem. With the filldir concept
1134 * it's not needed for correct information, but the XFS dir2 leaf
1135 * code wants an estimate of the buffer size to calculate it's
1136 * readahead window and size the buffers used for mapping to
1139 * Try to give it an estimate that's good enough, maybe at some
1140 * point we can change the ->readdir prototype to include the
1141 * buffer size. For now we use the current glibc buffer size.
1143 bufsize = (size_t)min_t(loff_t, XFS_READDIR_BUFSIZE, ip->i_d.di_size);
1145 return xfs_readdir(NULL, ip, ctx, bufsize);
1154 struct inode *inode = file->f_mapping->host;
1156 if (XFS_FORCED_SHUTDOWN(XFS_I(inode)->i_mount))
1161 return generic_file_llseek(file, offset, whence);
1163 offset = iomap_seek_hole(inode, offset, &xfs_seek_iomap_ops);
1166 offset = iomap_seek_data(inode, offset, &xfs_seek_iomap_ops);
1172 return vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
1176 * Locking for serialisation of IO during page faults. This results in a lock
1180 * sb_start_pagefault(vfs, freeze)
1181 * i_mmaplock (XFS - truncate serialisation)
1183 * i_lock (XFS - extent map serialisation)
1186 __xfs_filemap_fault(
1187 struct vm_fault *vmf,
1188 enum page_entry_size pe_size,
1191 struct inode *inode = file_inode(vmf->vma->vm_file);
1192 struct xfs_inode *ip = XFS_I(inode);
1195 trace_xfs_filemap_fault(ip, pe_size, write_fault);
1198 sb_start_pagefault(inode->i_sb);
1199 file_update_time(vmf->vma->vm_file);
1202 xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1203 if (IS_DAX(inode)) {
1206 ret = dax_iomap_fault(vmf, pe_size, &pfn, NULL,
1207 (write_fault && !vmf->cow_page) ?
1208 &xfs_direct_write_iomap_ops :
1209 &xfs_read_iomap_ops);
1210 if (ret & VM_FAULT_NEEDDSYNC)
1211 ret = dax_finish_sync_fault(vmf, pe_size, pfn);
1214 ret = iomap_page_mkwrite(vmf,
1215 &xfs_buffered_write_iomap_ops);
1217 ret = filemap_fault(vmf);
1219 xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1222 sb_end_pagefault(inode->i_sb);
1228 struct vm_fault *vmf)
1230 return (vmf->flags & FAULT_FLAG_WRITE) &&
1231 (vmf->vma->vm_flags & VM_SHARED);
1236 struct vm_fault *vmf)
1238 /* DAX can shortcut the normal fault path on write faults! */
1239 return __xfs_filemap_fault(vmf, PE_SIZE_PTE,
1240 IS_DAX(file_inode(vmf->vma->vm_file)) &&
1241 xfs_is_write_fault(vmf));
1245 xfs_filemap_huge_fault(
1246 struct vm_fault *vmf,
1247 enum page_entry_size pe_size)
1249 if (!IS_DAX(file_inode(vmf->vma->vm_file)))
1250 return VM_FAULT_FALLBACK;
1252 /* DAX can shortcut the normal fault path on write faults! */
1253 return __xfs_filemap_fault(vmf, pe_size,
1254 xfs_is_write_fault(vmf));
1258 xfs_filemap_page_mkwrite(
1259 struct vm_fault *vmf)
1261 return __xfs_filemap_fault(vmf, PE_SIZE_PTE, true);
1265 * pfn_mkwrite was originally intended to ensure we capture time stamp updates
1266 * on write faults. In reality, it needs to serialise against truncate and
1267 * prepare memory for writing so handle is as standard write fault.
1270 xfs_filemap_pfn_mkwrite(
1271 struct vm_fault *vmf)
1274 return __xfs_filemap_fault(vmf, PE_SIZE_PTE, true);
1278 xfs_filemap_map_pages(
1279 struct vm_fault *vmf,
1280 pgoff_t start_pgoff,
1283 struct inode *inode = file_inode(vmf->vma->vm_file);
1285 xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1286 filemap_map_pages(vmf, start_pgoff, end_pgoff);
1287 xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1290 static const struct vm_operations_struct xfs_file_vm_ops = {
1291 .fault = xfs_filemap_fault,
1292 .huge_fault = xfs_filemap_huge_fault,
1293 .map_pages = xfs_filemap_map_pages,
1294 .page_mkwrite = xfs_filemap_page_mkwrite,
1295 .pfn_mkwrite = xfs_filemap_pfn_mkwrite,
1301 struct vm_area_struct *vma)
1303 struct inode *inode = file_inode(file);
1304 struct xfs_buftarg *target = xfs_inode_buftarg(XFS_I(inode));
1307 * We don't support synchronous mappings for non-DAX files and
1308 * for DAX files if underneath dax_device is not synchronous.
1310 if (!daxdev_mapping_supported(vma, target->bt_daxdev))
1313 file_accessed(file);
1314 vma->vm_ops = &xfs_file_vm_ops;
1316 vma->vm_flags |= VM_HUGEPAGE;
1320 const struct file_operations xfs_file_operations = {
1321 .llseek = xfs_file_llseek,
1322 .read_iter = xfs_file_read_iter,
1323 .write_iter = xfs_file_write_iter,
1324 .splice_read = generic_file_splice_read,
1325 .splice_write = iter_file_splice_write,
1326 .iopoll = iomap_dio_iopoll,
1327 .unlocked_ioctl = xfs_file_ioctl,
1328 #ifdef CONFIG_COMPAT
1329 .compat_ioctl = xfs_file_compat_ioctl,
1331 .mmap = xfs_file_mmap,
1332 .mmap_supported_flags = MAP_SYNC,
1333 .open = xfs_file_open,
1334 .release = xfs_file_release,
1335 .fsync = xfs_file_fsync,
1336 .get_unmapped_area = thp_get_unmapped_area,
1337 .fallocate = xfs_file_fallocate,
1338 .fadvise = xfs_file_fadvise,
1339 .remap_file_range = xfs_file_remap_range,
1342 const struct file_operations xfs_dir_file_operations = {
1343 .open = xfs_dir_open,
1344 .read = generic_read_dir,
1345 .iterate_shared = xfs_file_readdir,
1346 .llseek = generic_file_llseek,
1347 .unlocked_ioctl = xfs_file_ioctl,
1348 #ifdef CONFIG_COMPAT
1349 .compat_ioctl = xfs_file_compat_ioctl,
1351 .fsync = xfs_dir_fsync,