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_iomap_ops, NULL);
187 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
192 static noinline ssize_t
197 struct xfs_inode *ip = XFS_I(iocb->ki_filp->f_mapping->host);
198 size_t count = iov_iter_count(to);
201 trace_xfs_file_dax_read(ip, count, iocb->ki_pos);
204 return 0; /* skip atime */
206 if (iocb->ki_flags & IOCB_NOWAIT) {
207 if (!xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED))
210 xfs_ilock(ip, XFS_IOLOCK_SHARED);
213 ret = dax_iomap_rw(iocb, to, &xfs_iomap_ops);
214 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
216 file_accessed(iocb->ki_filp);
221 xfs_file_buffered_aio_read(
225 struct xfs_inode *ip = XFS_I(file_inode(iocb->ki_filp));
228 trace_xfs_file_buffered_read(ip, iov_iter_count(to), iocb->ki_pos);
230 if (iocb->ki_flags & IOCB_NOWAIT) {
231 if (!xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED))
234 xfs_ilock(ip, XFS_IOLOCK_SHARED);
236 ret = generic_file_read_iter(iocb, to);
237 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
247 struct inode *inode = file_inode(iocb->ki_filp);
248 struct xfs_mount *mp = XFS_I(inode)->i_mount;
251 XFS_STATS_INC(mp, xs_read_calls);
253 if (XFS_FORCED_SHUTDOWN(mp))
257 ret = xfs_file_dax_read(iocb, to);
258 else if (iocb->ki_flags & IOCB_DIRECT)
259 ret = xfs_file_dio_aio_read(iocb, to);
261 ret = xfs_file_buffered_aio_read(iocb, to);
264 XFS_STATS_ADD(mp, xs_read_bytes, ret);
269 * Common pre-write limit and setup checks.
271 * Called with the iolocked held either shared and exclusive according to
272 * @iolock, and returns with it held. Might upgrade the iolock to exclusive
273 * if called for a direct write beyond i_size.
276 xfs_file_aio_write_checks(
278 struct iov_iter *from,
281 struct file *file = iocb->ki_filp;
282 struct inode *inode = file->f_mapping->host;
283 struct xfs_inode *ip = XFS_I(inode);
285 size_t count = iov_iter_count(from);
286 bool drained_dio = false;
290 error = generic_write_checks(iocb, from);
294 error = xfs_break_layouts(inode, iolock, BREAK_WRITE);
299 * For changing security info in file_remove_privs() we need i_rwsem
302 if (*iolock == XFS_IOLOCK_SHARED && !IS_NOSEC(inode)) {
303 xfs_iunlock(ip, *iolock);
304 *iolock = XFS_IOLOCK_EXCL;
305 xfs_ilock(ip, *iolock);
309 * If the offset is beyond the size of the file, we need to zero any
310 * blocks that fall between the existing EOF and the start of this
311 * write. If zeroing is needed and we are currently holding the
312 * iolock shared, we need to update it to exclusive which implies
313 * having to redo all checks before.
315 * We need to serialise against EOF updates that occur in IO
316 * completions here. We want to make sure that nobody is changing the
317 * size while we do this check until we have placed an IO barrier (i.e.
318 * hold the XFS_IOLOCK_EXCL) that prevents new IO from being dispatched.
319 * The spinlock effectively forms a memory barrier once we have the
320 * XFS_IOLOCK_EXCL so we are guaranteed to see the latest EOF value
321 * and hence be able to correctly determine if we need to run zeroing.
323 spin_lock(&ip->i_flags_lock);
324 isize = i_size_read(inode);
325 if (iocb->ki_pos > isize) {
326 spin_unlock(&ip->i_flags_lock);
328 if (*iolock == XFS_IOLOCK_SHARED) {
329 xfs_iunlock(ip, *iolock);
330 *iolock = XFS_IOLOCK_EXCL;
331 xfs_ilock(ip, *iolock);
332 iov_iter_reexpand(from, count);
335 * We now have an IO submission barrier in place, but
336 * AIO can do EOF updates during IO completion and hence
337 * we now need to wait for all of them to drain. Non-AIO
338 * DIO will have drained before we are given the
339 * XFS_IOLOCK_EXCL, and so for most cases this wait is a
342 inode_dio_wait(inode);
347 trace_xfs_zero_eof(ip, isize, iocb->ki_pos - isize);
348 error = iomap_zero_range(inode, isize, iocb->ki_pos - isize,
349 NULL, &xfs_iomap_ops);
353 spin_unlock(&ip->i_flags_lock);
356 * Updating the timestamps will grab the ilock again from
357 * xfs_fs_dirty_inode, so we have to call it after dropping the
358 * lock above. Eventually we should look into a way to avoid
359 * the pointless lock roundtrip.
361 return file_modified(file);
365 xfs_dio_write_end_io(
371 struct inode *inode = file_inode(iocb->ki_filp);
372 struct xfs_inode *ip = XFS_I(inode);
373 loff_t offset = iocb->ki_pos;
374 unsigned int nofs_flag;
376 trace_xfs_end_io_direct_write(ip, offset, size);
378 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
387 * Capture amount written on completion as we can't reliably account
388 * for it on submission.
390 XFS_STATS_ADD(ip->i_mount, xs_write_bytes, size);
393 * We can allocate memory here while doing writeback on behalf of
394 * memory reclaim. To avoid memory allocation deadlocks set the
395 * task-wide nofs context for the following operations.
397 nofs_flag = memalloc_nofs_save();
399 if (flags & IOMAP_DIO_COW) {
400 error = xfs_reflink_end_cow(ip, offset, size);
406 * Unwritten conversion updates the in-core isize after extent
407 * conversion but before updating the on-disk size. Updating isize any
408 * earlier allows a racing dio read to find unwritten extents before
409 * they are converted.
411 if (flags & IOMAP_DIO_UNWRITTEN) {
412 error = xfs_iomap_write_unwritten(ip, offset, size, true);
417 * We need to update the in-core inode size here so that we don't end up
418 * with the on-disk inode size being outside the in-core inode size. We
419 * have no other method of updating EOF for AIO, so always do it here
422 * We need to lock the test/set EOF update as we can be racing with
423 * other IO completions here to update the EOF. Failing to serialise
424 * here can result in EOF moving backwards and Bad Things Happen when
427 spin_lock(&ip->i_flags_lock);
428 if (offset + size > i_size_read(inode)) {
429 i_size_write(inode, offset + size);
430 spin_unlock(&ip->i_flags_lock);
431 error = xfs_setfilesize(ip, offset, size);
433 spin_unlock(&ip->i_flags_lock);
437 memalloc_nofs_restore(nofs_flag);
441 static const struct iomap_dio_ops xfs_dio_write_ops = {
442 .end_io = xfs_dio_write_end_io,
446 * xfs_file_dio_aio_write - handle direct IO writes
448 * Lock the inode appropriately to prepare for and issue a direct IO write.
449 * By separating it from the buffered write path we remove all the tricky to
450 * follow locking changes and looping.
452 * If there are cached pages or we're extending the file, we need IOLOCK_EXCL
453 * until we're sure the bytes at the new EOF have been zeroed and/or the cached
454 * pages are flushed out.
456 * In most cases the direct IO writes will be done holding IOLOCK_SHARED
457 * allowing them to be done in parallel with reads and other direct IO writes.
458 * However, if the IO is not aligned to filesystem blocks, the direct IO layer
459 * needs to do sub-block zeroing and that requires serialisation against other
460 * direct IOs to the same block. In this case we need to serialise the
461 * submission of the unaligned IOs so that we don't get racing block zeroing in
462 * the dio layer. To avoid the problem with aio, we also need to wait for
463 * outstanding IOs to complete so that unwritten extent conversion is completed
464 * before we try to map the overlapping block. This is currently implemented by
465 * hitting it with a big hammer (i.e. inode_dio_wait()).
467 * Returns with locks held indicated by @iolock and errors indicated by
468 * negative return values.
471 xfs_file_dio_aio_write(
473 struct iov_iter *from)
475 struct file *file = iocb->ki_filp;
476 struct address_space *mapping = file->f_mapping;
477 struct inode *inode = mapping->host;
478 struct xfs_inode *ip = XFS_I(inode);
479 struct xfs_mount *mp = ip->i_mount;
481 int unaligned_io = 0;
483 size_t count = iov_iter_count(from);
484 struct xfs_buftarg *target = XFS_IS_REALTIME_INODE(ip) ?
485 mp->m_rtdev_targp : mp->m_ddev_targp;
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);
545 ret = iomap_dio_rw(iocb, from, &xfs_iomap_ops, &xfs_dio_write_ops);
548 * If unaligned, this is the only IO in-flight. If it has not yet
549 * completed, wait on it before we release the iolock to prevent
550 * subsequent overlapping IO.
552 if (ret == -EIOCBQUEUED && unaligned_io)
553 inode_dio_wait(inode);
555 xfs_iunlock(ip, iolock);
558 * No fallback to buffered IO on errors for XFS, direct IO will either
559 * complete fully or fail.
561 ASSERT(ret < 0 || ret == count);
565 static noinline ssize_t
568 struct iov_iter *from)
570 struct inode *inode = iocb->ki_filp->f_mapping->host;
571 struct xfs_inode *ip = XFS_I(inode);
572 int iolock = XFS_IOLOCK_EXCL;
573 ssize_t ret, error = 0;
577 if (iocb->ki_flags & IOCB_NOWAIT) {
578 if (!xfs_ilock_nowait(ip, iolock))
581 xfs_ilock(ip, iolock);
584 ret = xfs_file_aio_write_checks(iocb, from, &iolock);
589 count = iov_iter_count(from);
591 trace_xfs_file_dax_write(ip, count, pos);
592 ret = dax_iomap_rw(iocb, from, &xfs_iomap_ops);
593 if (ret > 0 && iocb->ki_pos > i_size_read(inode)) {
594 i_size_write(inode, iocb->ki_pos);
595 error = xfs_setfilesize(ip, pos, ret);
598 xfs_iunlock(ip, iolock);
603 XFS_STATS_ADD(ip->i_mount, xs_write_bytes, ret);
605 /* Handle various SYNC-type writes */
606 ret = generic_write_sync(iocb, ret);
612 xfs_file_buffered_aio_write(
614 struct iov_iter *from)
616 struct file *file = iocb->ki_filp;
617 struct address_space *mapping = file->f_mapping;
618 struct inode *inode = mapping->host;
619 struct xfs_inode *ip = XFS_I(inode);
624 if (iocb->ki_flags & IOCB_NOWAIT)
628 iolock = XFS_IOLOCK_EXCL;
629 xfs_ilock(ip, iolock);
631 ret = xfs_file_aio_write_checks(iocb, from, &iolock);
635 /* We can write back this queue in page reclaim */
636 current->backing_dev_info = inode_to_bdi(inode);
638 trace_xfs_file_buffered_write(ip, iov_iter_count(from), iocb->ki_pos);
639 ret = iomap_file_buffered_write(iocb, from, &xfs_iomap_ops);
640 if (likely(ret >= 0))
644 * If we hit a space limit, try to free up some lingering preallocated
645 * space before returning an error. In the case of ENOSPC, first try to
646 * write back all dirty inodes to free up some of the excess reserved
647 * metadata space. This reduces the chances that the eofblocks scan
648 * waits on dirty mappings. Since xfs_flush_inodes() is serialized, this
649 * also behaves as a filter to prevent too many eofblocks scans from
650 * running at the same time.
652 if (ret == -EDQUOT && !enospc) {
653 xfs_iunlock(ip, iolock);
654 enospc = xfs_inode_free_quota_eofblocks(ip);
657 enospc = xfs_inode_free_quota_cowblocks(ip);
661 } else if (ret == -ENOSPC && !enospc) {
662 struct xfs_eofblocks eofb = {0};
665 xfs_flush_inodes(ip->i_mount);
667 xfs_iunlock(ip, iolock);
668 eofb.eof_flags = XFS_EOF_FLAGS_SYNC;
669 xfs_icache_free_eofblocks(ip->i_mount, &eofb);
670 xfs_icache_free_cowblocks(ip->i_mount, &eofb);
674 current->backing_dev_info = NULL;
677 xfs_iunlock(ip, iolock);
680 XFS_STATS_ADD(ip->i_mount, xs_write_bytes, ret);
681 /* Handle various SYNC-type writes */
682 ret = generic_write_sync(iocb, ret);
690 struct iov_iter *from)
692 struct file *file = iocb->ki_filp;
693 struct address_space *mapping = file->f_mapping;
694 struct inode *inode = mapping->host;
695 struct xfs_inode *ip = XFS_I(inode);
697 size_t ocount = iov_iter_count(from);
699 XFS_STATS_INC(ip->i_mount, xs_write_calls);
704 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
708 return xfs_file_dax_write(iocb, from);
710 if (iocb->ki_flags & IOCB_DIRECT) {
712 * Allow a directio write to fall back to a buffered
713 * write *only* in the case that we're doing a reflink
714 * CoW. In all other directio scenarios we do not
715 * allow an operation to fall back to buffered mode.
717 ret = xfs_file_dio_aio_write(iocb, from);
722 return xfs_file_buffered_aio_write(iocb, from);
729 struct xfs_inode *ip = XFS_I(inode);
731 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
733 xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
737 xfs_break_dax_layouts(
743 ASSERT(xfs_isilocked(XFS_I(inode), XFS_MMAPLOCK_EXCL));
745 page = dax_layout_busy_page(inode->i_mapping);
750 return ___wait_var_event(&page->_refcount,
751 atomic_read(&page->_refcount) == 1, TASK_INTERRUPTIBLE,
752 0, 0, xfs_wait_dax_page(inode));
759 enum layout_break_reason reason)
764 ASSERT(xfs_isilocked(XFS_I(inode), XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL));
770 error = xfs_break_dax_layouts(inode, &retry);
775 error = xfs_break_leased_layouts(inode, iolock, &retry);
781 } while (error == 0 && retry);
786 #define XFS_FALLOC_FL_SUPPORTED \
787 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
788 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE | \
789 FALLOC_FL_INSERT_RANGE | FALLOC_FL_UNSHARE_RANGE)
798 struct inode *inode = file_inode(file);
799 struct xfs_inode *ip = XFS_I(inode);
801 enum xfs_prealloc_flags flags = 0;
802 uint iolock = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL;
804 bool do_file_insert = false;
806 if (!S_ISREG(inode->i_mode))
808 if (mode & ~XFS_FALLOC_FL_SUPPORTED)
811 xfs_ilock(ip, iolock);
812 error = xfs_break_layouts(inode, &iolock, BREAK_UNMAP);
817 * Must wait for all AIO to complete before we continue as AIO can
818 * change the file size on completion without holding any locks we
819 * currently hold. We must do this first because AIO can update both
820 * the on disk and in memory inode sizes, and the operations that follow
821 * require the in-memory size to be fully up-to-date.
823 inode_dio_wait(inode);
826 * Now AIO and DIO has drained we flush and (if necessary) invalidate
827 * the cached range over the first operation we are about to run.
829 * We care about zero and collapse here because they both run a hole
830 * punch over the range first. Because that can zero data, and the range
831 * of invalidation for the shift operations is much larger, we still do
832 * the required flush for collapse in xfs_prepare_shift().
834 * Insert has the same range requirements as collapse, and we extend the
835 * file first which can zero data. Hence insert has the same
836 * flush/invalidate requirements as collapse and so they are both
837 * handled at the right time by xfs_prepare_shift().
839 if (mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE |
840 FALLOC_FL_COLLAPSE_RANGE)) {
841 error = xfs_flush_unmap_range(ip, offset, len);
846 if (mode & FALLOC_FL_PUNCH_HOLE) {
847 error = xfs_free_file_space(ip, offset, len);
850 } else if (mode & FALLOC_FL_COLLAPSE_RANGE) {
851 unsigned int blksize_mask = i_blocksize(inode) - 1;
853 if (offset & blksize_mask || len & blksize_mask) {
859 * There is no need to overlap collapse range with EOF,
860 * in which case it is effectively a truncate operation
862 if (offset + len >= i_size_read(inode)) {
867 new_size = i_size_read(inode) - len;
869 error = xfs_collapse_file_space(ip, offset, len);
872 } else if (mode & FALLOC_FL_INSERT_RANGE) {
873 unsigned int blksize_mask = i_blocksize(inode) - 1;
874 loff_t isize = i_size_read(inode);
876 if (offset & blksize_mask || len & blksize_mask) {
882 * New inode size must not exceed ->s_maxbytes, accounting for
883 * possible signed overflow.
885 if (inode->i_sb->s_maxbytes - isize < len) {
889 new_size = isize + len;
891 /* Offset should be less than i_size */
892 if (offset >= isize) {
896 do_file_insert = true;
898 flags |= XFS_PREALLOC_SET;
900 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
901 offset + len > i_size_read(inode)) {
902 new_size = offset + len;
903 error = inode_newsize_ok(inode, new_size);
908 if (mode & FALLOC_FL_ZERO_RANGE) {
909 error = xfs_zero_file_space(ip, offset, len);
910 } else if (mode & FALLOC_FL_UNSHARE_RANGE) {
911 error = xfs_reflink_unshare(ip, offset, len);
915 if (!xfs_is_always_cow_inode(ip)) {
916 error = xfs_alloc_file_space(ip, offset, len,
921 * If always_cow mode we can't use preallocations and
922 * thus should not create them.
924 if (xfs_is_always_cow_inode(ip)) {
929 error = xfs_alloc_file_space(ip, offset, len,
936 if (file->f_flags & O_DSYNC)
937 flags |= XFS_PREALLOC_SYNC;
939 error = xfs_update_prealloc_flags(ip, flags);
943 /* Change file size if needed */
947 iattr.ia_valid = ATTR_SIZE;
948 iattr.ia_size = new_size;
949 error = xfs_vn_setattr_size(file_dentry(file), &iattr);
955 * Perform hole insertion now that the file size has been
956 * updated so that if we crash during the operation we don't
957 * leave shifted extents past EOF and hence losing access to
958 * the data that is contained within them.
961 error = xfs_insert_file_space(ip, offset, len);
964 xfs_iunlock(ip, iolock);
975 struct xfs_inode *ip = XFS_I(file_inode(file));
980 * Operations creating pages in page cache need protection from hole
981 * punching and similar ops
983 if (advice == POSIX_FADV_WILLNEED) {
984 lockflags = XFS_IOLOCK_SHARED;
985 xfs_ilock(ip, lockflags);
987 ret = generic_fadvise(file, start, end, advice);
989 xfs_iunlock(ip, lockflags);
993 /* Does this file, inode, or mount want synchronous writes? */
994 static inline bool xfs_file_sync_writes(struct file *filp)
996 struct xfs_inode *ip = XFS_I(file_inode(filp));
998 if (ip->i_mount->m_flags & XFS_MOUNT_WSYNC)
1000 if (filp->f_flags & (__O_SYNC | O_DSYNC))
1002 if (IS_SYNC(file_inode(filp)))
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 (xfs_file_sync_writes(file_in) || xfs_file_sync_writes(file_out))
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_d.di_nextents > 0)
1106 error = xfs_dir3_data_readahead(ip, 0, -1);
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, &xfs_iomap_ops);
1204 if (ret & VM_FAULT_NEEDDSYNC)
1205 ret = dax_finish_sync_fault(vmf, pe_size, pfn);
1208 ret = iomap_page_mkwrite(vmf, &xfs_iomap_ops);
1210 ret = filemap_fault(vmf);
1212 xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1215 sb_end_pagefault(inode->i_sb);
1221 struct vm_fault *vmf)
1223 return (vmf->flags & FAULT_FLAG_WRITE) &&
1224 (vmf->vma->vm_flags & VM_SHARED);
1229 struct vm_fault *vmf)
1231 /* DAX can shortcut the normal fault path on write faults! */
1232 return __xfs_filemap_fault(vmf, PE_SIZE_PTE,
1233 IS_DAX(file_inode(vmf->vma->vm_file)) &&
1234 xfs_is_write_fault(vmf));
1238 xfs_filemap_huge_fault(
1239 struct vm_fault *vmf,
1240 enum page_entry_size pe_size)
1242 if (!IS_DAX(file_inode(vmf->vma->vm_file)))
1243 return VM_FAULT_FALLBACK;
1245 /* DAX can shortcut the normal fault path on write faults! */
1246 return __xfs_filemap_fault(vmf, pe_size,
1247 xfs_is_write_fault(vmf));
1251 xfs_filemap_page_mkwrite(
1252 struct vm_fault *vmf)
1254 return __xfs_filemap_fault(vmf, PE_SIZE_PTE, true);
1258 * pfn_mkwrite was originally intended to ensure we capture time stamp updates
1259 * on write faults. In reality, it needs to serialise against truncate and
1260 * prepare memory for writing so handle is as standard write fault.
1263 xfs_filemap_pfn_mkwrite(
1264 struct vm_fault *vmf)
1267 return __xfs_filemap_fault(vmf, PE_SIZE_PTE, true);
1271 xfs_filemap_map_pages(
1272 struct vm_fault *vmf,
1273 pgoff_t start_pgoff,
1276 struct inode *inode = file_inode(vmf->vma->vm_file);
1278 xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1279 filemap_map_pages(vmf, start_pgoff, end_pgoff);
1280 xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1283 static const struct vm_operations_struct xfs_file_vm_ops = {
1284 .fault = xfs_filemap_fault,
1285 .huge_fault = xfs_filemap_huge_fault,
1286 .map_pages = xfs_filemap_map_pages,
1287 .page_mkwrite = xfs_filemap_page_mkwrite,
1288 .pfn_mkwrite = xfs_filemap_pfn_mkwrite,
1294 struct vm_area_struct *vma)
1296 struct dax_device *dax_dev;
1298 dax_dev = xfs_find_daxdev_for_inode(file_inode(filp));
1300 * We don't support synchronous mappings for non-DAX files and
1301 * for DAX files if underneath dax_device is not synchronous.
1303 if (!daxdev_mapping_supported(vma, dax_dev))
1306 file_accessed(filp);
1307 vma->vm_ops = &xfs_file_vm_ops;
1308 if (IS_DAX(file_inode(filp)))
1309 vma->vm_flags |= VM_HUGEPAGE;
1313 const struct file_operations xfs_file_operations = {
1314 .llseek = xfs_file_llseek,
1315 .read_iter = xfs_file_read_iter,
1316 .write_iter = xfs_file_write_iter,
1317 .splice_read = generic_file_splice_read,
1318 .splice_write = iter_file_splice_write,
1319 .iopoll = iomap_dio_iopoll,
1320 .unlocked_ioctl = xfs_file_ioctl,
1321 #ifdef CONFIG_COMPAT
1322 .compat_ioctl = xfs_file_compat_ioctl,
1324 .mmap = xfs_file_mmap,
1325 .mmap_supported_flags = MAP_SYNC,
1326 .open = xfs_file_open,
1327 .release = xfs_file_release,
1328 .fsync = xfs_file_fsync,
1329 .get_unmapped_area = thp_get_unmapped_area,
1330 .fallocate = xfs_file_fallocate,
1331 .fadvise = xfs_file_fadvise,
1332 .remap_file_range = xfs_file_remap_range,
1335 const struct file_operations xfs_dir_file_operations = {
1336 .open = xfs_dir_open,
1337 .read = generic_read_dir,
1338 .iterate_shared = xfs_file_readdir,
1339 .llseek = generic_file_llseek,
1340 .unlocked_ioctl = xfs_file_ioctl,
1341 #ifdef CONFIG_COMPAT
1342 .compat_ioctl = xfs_file_compat_ioctl,
1344 .fsync = xfs_dir_fsync,