2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_shared.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
24 #include "xfs_mount.h"
25 #include "xfs_da_format.h"
26 #include "xfs_da_btree.h"
27 #include "xfs_inode.h"
28 #include "xfs_trans.h"
29 #include "xfs_inode_item.h"
31 #include "xfs_bmap_util.h"
32 #include "xfs_error.h"
34 #include "xfs_dir2_priv.h"
35 #include "xfs_ioctl.h"
36 #include "xfs_trace.h"
38 #include "xfs_icache.h"
40 #include "xfs_iomap.h"
41 #include "xfs_reflink.h"
43 #include <linux/dcache.h>
44 #include <linux/falloc.h>
45 #include <linux/pagevec.h>
46 #include <linux/backing-dev.h>
48 static const struct vm_operations_struct xfs_file_vm_ops;
51 * Locking primitives for read and write IO paths to ensure we consistently use
52 * and order the inode->i_mutex, ip->i_lock and ip->i_iolock.
59 if (type & XFS_IOLOCK_EXCL)
60 inode_lock(VFS_I(ip));
69 xfs_iunlock(ip, type);
70 if (type & XFS_IOLOCK_EXCL)
71 inode_unlock(VFS_I(ip));
79 xfs_ilock_demote(ip, type);
80 if (type & XFS_IOLOCK_EXCL)
81 inode_unlock(VFS_I(ip));
85 * Clear the specified ranges to zero through either the pagecache or DAX.
86 * Holes and unwritten extents will be left as-is as they already are zeroed.
95 return iomap_zero_range(VFS_I(ip), pos, count, NULL, &xfs_iomap_ops);
99 xfs_update_prealloc_flags(
100 struct xfs_inode *ip,
101 enum xfs_prealloc_flags flags)
103 struct xfs_trans *tp;
106 error = xfs_trans_alloc(ip->i_mount, &M_RES(ip->i_mount)->tr_writeid,
111 xfs_ilock(ip, XFS_ILOCK_EXCL);
112 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
114 if (!(flags & XFS_PREALLOC_INVISIBLE)) {
115 VFS_I(ip)->i_mode &= ~S_ISUID;
116 if (VFS_I(ip)->i_mode & S_IXGRP)
117 VFS_I(ip)->i_mode &= ~S_ISGID;
118 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
121 if (flags & XFS_PREALLOC_SET)
122 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
123 if (flags & XFS_PREALLOC_CLEAR)
124 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
126 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
127 if (flags & XFS_PREALLOC_SYNC)
128 xfs_trans_set_sync(tp);
129 return xfs_trans_commit(tp);
133 * Fsync operations on directories are much simpler than on regular files,
134 * as there is no file data to flush, and thus also no need for explicit
135 * cache flush operations, and there are no non-transaction metadata updates
136 * on directories either.
145 struct xfs_inode *ip = XFS_I(file->f_mapping->host);
146 struct xfs_mount *mp = ip->i_mount;
149 trace_xfs_dir_fsync(ip);
151 xfs_ilock(ip, XFS_ILOCK_SHARED);
152 if (xfs_ipincount(ip))
153 lsn = ip->i_itemp->ili_last_lsn;
154 xfs_iunlock(ip, XFS_ILOCK_SHARED);
158 return _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, NULL);
168 struct inode *inode = file->f_mapping->host;
169 struct xfs_inode *ip = XFS_I(inode);
170 struct xfs_mount *mp = ip->i_mount;
175 trace_xfs_file_fsync(ip);
177 error = filemap_write_and_wait_range(inode->i_mapping, start, end);
181 if (XFS_FORCED_SHUTDOWN(mp))
184 xfs_iflags_clear(ip, XFS_ITRUNCATED);
186 if (mp->m_flags & XFS_MOUNT_BARRIER) {
188 * If we have an RT and/or log subvolume we need to make sure
189 * to flush the write cache the device used for file data
190 * first. This is to ensure newly written file data make
191 * it to disk before logging the new inode size in case of
192 * an extending write.
194 if (XFS_IS_REALTIME_INODE(ip))
195 xfs_blkdev_issue_flush(mp->m_rtdev_targp);
196 else if (mp->m_logdev_targp != mp->m_ddev_targp)
197 xfs_blkdev_issue_flush(mp->m_ddev_targp);
201 * All metadata updates are logged, which means that we just have to
202 * flush the log up to the latest LSN that touched the inode. If we have
203 * concurrent fsync/fdatasync() calls, we need them to all block on the
204 * log force before we clear the ili_fsync_fields field. This ensures
205 * that we don't get a racing sync operation that does not wait for the
206 * metadata to hit the journal before returning. If we race with
207 * clearing the ili_fsync_fields, then all that will happen is the log
208 * force will do nothing as the lsn will already be on disk. We can't
209 * race with setting ili_fsync_fields because that is done under
210 * XFS_ILOCK_EXCL, and that can't happen because we hold the lock shared
211 * until after the ili_fsync_fields is cleared.
213 xfs_ilock(ip, XFS_ILOCK_SHARED);
214 if (xfs_ipincount(ip)) {
216 (ip->i_itemp->ili_fsync_fields & ~XFS_ILOG_TIMESTAMP))
217 lsn = ip->i_itemp->ili_last_lsn;
221 error = _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, &log_flushed);
222 ip->i_itemp->ili_fsync_fields = 0;
224 xfs_iunlock(ip, XFS_ILOCK_SHARED);
227 * If we only have a single device, and the log force about was
228 * a no-op we might have to flush the data device cache here.
229 * This can only happen for fdatasync/O_DSYNC if we were overwriting
230 * an already allocated file and thus do not have any metadata to
233 if ((mp->m_flags & XFS_MOUNT_BARRIER) &&
234 mp->m_logdev_targp == mp->m_ddev_targp &&
235 !XFS_IS_REALTIME_INODE(ip) &&
237 xfs_blkdev_issue_flush(mp->m_ddev_targp);
243 xfs_file_dio_aio_read(
247 struct address_space *mapping = iocb->ki_filp->f_mapping;
248 struct inode *inode = mapping->host;
249 struct xfs_inode *ip = XFS_I(inode);
250 loff_t isize = i_size_read(inode);
251 size_t count = iov_iter_count(to);
252 struct iov_iter data;
253 struct xfs_buftarg *target;
256 trace_xfs_file_direct_read(ip, count, iocb->ki_pos);
259 return 0; /* skip atime */
261 if (XFS_IS_REALTIME_INODE(ip))
262 target = ip->i_mount->m_rtdev_targp;
264 target = ip->i_mount->m_ddev_targp;
266 /* DIO must be aligned to device logical sector size */
267 if ((iocb->ki_pos | count) & target->bt_logical_sectormask) {
268 if (iocb->ki_pos == isize)
273 file_accessed(iocb->ki_filp);
276 * Locking is a bit tricky here. If we take an exclusive lock for direct
277 * IO, we effectively serialise all new concurrent read IO to this file
278 * and block it behind IO that is currently in progress because IO in
279 * progress holds the IO lock shared. We only need to hold the lock
280 * exclusive to blow away the page cache, so only take lock exclusively
281 * if the page cache needs invalidation. This allows the normal direct
282 * IO case of no page cache pages to proceeed concurrently without
285 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
286 if (mapping->nrpages) {
287 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
288 xfs_rw_ilock(ip, XFS_IOLOCK_EXCL);
291 * The generic dio code only flushes the range of the particular
292 * I/O. Because we take an exclusive lock here, this whole
293 * sequence is considerably more expensive for us. This has a
294 * noticeable performance impact for any file with cached pages,
295 * even when outside of the range of the particular I/O.
297 * Hence, amortize the cost of the lock against a full file
298 * flush and reduce the chances of repeated iolock cycles going
301 if (mapping->nrpages) {
302 ret = filemap_write_and_wait(mapping);
304 xfs_rw_iunlock(ip, XFS_IOLOCK_EXCL);
309 * Invalidate whole pages. This can return an error if
310 * we fail to invalidate a page, but this should never
311 * happen on XFS. Warn if it does fail.
313 ret = invalidate_inode_pages2(mapping);
317 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
321 ret = __blockdev_direct_IO(iocb, inode, target->bt_bdev, &data,
322 xfs_get_blocks_direct, NULL, NULL, 0);
325 iov_iter_advance(to, ret);
327 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
332 static noinline ssize_t
337 struct xfs_inode *ip = XFS_I(iocb->ki_filp->f_mapping->host);
338 size_t count = iov_iter_count(to);
341 trace_xfs_file_dax_read(ip, count, iocb->ki_pos);
344 return 0; /* skip atime */
346 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
347 ret = iomap_dax_rw(iocb, to, &xfs_iomap_ops);
348 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
350 file_accessed(iocb->ki_filp);
355 xfs_file_buffered_aio_read(
359 struct xfs_inode *ip = XFS_I(file_inode(iocb->ki_filp));
362 trace_xfs_file_buffered_read(ip, iov_iter_count(to), iocb->ki_pos);
364 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
365 ret = generic_file_read_iter(iocb, to);
366 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
376 struct inode *inode = file_inode(iocb->ki_filp);
377 struct xfs_mount *mp = XFS_I(inode)->i_mount;
380 XFS_STATS_INC(mp, xs_read_calls);
382 if (XFS_FORCED_SHUTDOWN(mp))
386 ret = xfs_file_dax_read(iocb, to);
387 else if (iocb->ki_flags & IOCB_DIRECT)
388 ret = xfs_file_dio_aio_read(iocb, to);
390 ret = xfs_file_buffered_aio_read(iocb, to);
393 XFS_STATS_ADD(mp, xs_read_bytes, ret);
398 * Zero any on disk space between the current EOF and the new, larger EOF.
400 * This handles the normal case of zeroing the remainder of the last block in
401 * the file and the unusual case of zeroing blocks out beyond the size of the
402 * file. This second case only happens with fixed size extents and when the
403 * system crashes before the inode size was updated but after blocks were
406 * Expects the iolock to be held exclusive, and will take the ilock internally.
408 int /* error (positive) */
410 struct xfs_inode *ip,
411 xfs_off_t offset, /* starting I/O offset */
412 xfs_fsize_t isize, /* current inode size */
415 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
416 ASSERT(offset > isize);
418 trace_xfs_zero_eof(ip, isize, offset - isize);
419 return xfs_zero_range(ip, isize, offset - isize, did_zeroing);
423 * Common pre-write limit and setup checks.
425 * Called with the iolocked held either shared and exclusive according to
426 * @iolock, and returns with it held. Might upgrade the iolock to exclusive
427 * if called for a direct write beyond i_size.
430 xfs_file_aio_write_checks(
432 struct iov_iter *from,
435 struct file *file = iocb->ki_filp;
436 struct inode *inode = file->f_mapping->host;
437 struct xfs_inode *ip = XFS_I(inode);
439 size_t count = iov_iter_count(from);
440 bool drained_dio = false;
443 error = generic_write_checks(iocb, from);
447 error = xfs_break_layouts(inode, iolock, true);
451 /* For changing security info in file_remove_privs() we need i_mutex */
452 if (*iolock == XFS_IOLOCK_SHARED && !IS_NOSEC(inode)) {
453 xfs_rw_iunlock(ip, *iolock);
454 *iolock = XFS_IOLOCK_EXCL;
455 xfs_rw_ilock(ip, *iolock);
459 * If the offset is beyond the size of the file, we need to zero any
460 * blocks that fall between the existing EOF and the start of this
461 * write. If zeroing is needed and we are currently holding the
462 * iolock shared, we need to update it to exclusive which implies
463 * having to redo all checks before.
465 * We need to serialise against EOF updates that occur in IO
466 * completions here. We want to make sure that nobody is changing the
467 * size while we do this check until we have placed an IO barrier (i.e.
468 * hold the XFS_IOLOCK_EXCL) that prevents new IO from being dispatched.
469 * The spinlock effectively forms a memory barrier once we have the
470 * XFS_IOLOCK_EXCL so we are guaranteed to see the latest EOF value
471 * and hence be able to correctly determine if we need to run zeroing.
473 spin_lock(&ip->i_flags_lock);
474 if (iocb->ki_pos > i_size_read(inode)) {
477 spin_unlock(&ip->i_flags_lock);
479 if (*iolock == XFS_IOLOCK_SHARED) {
480 xfs_rw_iunlock(ip, *iolock);
481 *iolock = XFS_IOLOCK_EXCL;
482 xfs_rw_ilock(ip, *iolock);
483 iov_iter_reexpand(from, count);
486 * We now have an IO submission barrier in place, but
487 * AIO can do EOF updates during IO completion and hence
488 * we now need to wait for all of them to drain. Non-AIO
489 * DIO will have drained before we are given the
490 * XFS_IOLOCK_EXCL, and so for most cases this wait is a
493 inode_dio_wait(inode);
497 error = xfs_zero_eof(ip, iocb->ki_pos, i_size_read(inode), &zero);
501 spin_unlock(&ip->i_flags_lock);
504 * Updating the timestamps will grab the ilock again from
505 * xfs_fs_dirty_inode, so we have to call it after dropping the
506 * lock above. Eventually we should look into a way to avoid
507 * the pointless lock roundtrip.
509 if (likely(!(file->f_mode & FMODE_NOCMTIME))) {
510 error = file_update_time(file);
516 * If we're writing the file then make sure to clear the setuid and
517 * setgid bits if the process is not being run by root. This keeps
518 * people from modifying setuid and setgid binaries.
520 if (!IS_NOSEC(inode))
521 return file_remove_privs(file);
526 * xfs_file_dio_aio_write - handle direct IO writes
528 * Lock the inode appropriately to prepare for and issue a direct IO write.
529 * By separating it from the buffered write path we remove all the tricky to
530 * follow locking changes and looping.
532 * If there are cached pages or we're extending the file, we need IOLOCK_EXCL
533 * until we're sure the bytes at the new EOF have been zeroed and/or the cached
534 * pages are flushed out.
536 * In most cases the direct IO writes will be done holding IOLOCK_SHARED
537 * allowing them to be done in parallel with reads and other direct IO writes.
538 * However, if the IO is not aligned to filesystem blocks, the direct IO layer
539 * needs to do sub-block zeroing and that requires serialisation against other
540 * direct IOs to the same block. In this case we need to serialise the
541 * submission of the unaligned IOs so that we don't get racing block zeroing in
542 * the dio layer. To avoid the problem with aio, we also need to wait for
543 * outstanding IOs to complete so that unwritten extent conversion is completed
544 * before we try to map the overlapping block. This is currently implemented by
545 * hitting it with a big hammer (i.e. inode_dio_wait()).
547 * Returns with locks held indicated by @iolock and errors indicated by
548 * negative return values.
551 xfs_file_dio_aio_write(
553 struct iov_iter *from)
555 struct file *file = iocb->ki_filp;
556 struct address_space *mapping = file->f_mapping;
557 struct inode *inode = mapping->host;
558 struct xfs_inode *ip = XFS_I(inode);
559 struct xfs_mount *mp = ip->i_mount;
561 int unaligned_io = 0;
563 size_t count = iov_iter_count(from);
565 struct iov_iter data;
566 struct xfs_buftarg *target = XFS_IS_REALTIME_INODE(ip) ?
567 mp->m_rtdev_targp : mp->m_ddev_targp;
569 /* DIO must be aligned to device logical sector size */
570 if ((iocb->ki_pos | count) & target->bt_logical_sectormask)
573 /* "unaligned" here means not aligned to a filesystem block */
574 if ((iocb->ki_pos & mp->m_blockmask) ||
575 ((iocb->ki_pos + count) & mp->m_blockmask))
579 * We don't need to take an exclusive lock unless there page cache needs
580 * to be invalidated or unaligned IO is being executed. We don't need to
581 * consider the EOF extension case here because
582 * xfs_file_aio_write_checks() will relock the inode as necessary for
583 * EOF zeroing cases and fill out the new inode size as appropriate.
585 if (unaligned_io || mapping->nrpages)
586 iolock = XFS_IOLOCK_EXCL;
588 iolock = XFS_IOLOCK_SHARED;
589 xfs_rw_ilock(ip, iolock);
592 * Recheck if there are cached pages that need invalidate after we got
593 * the iolock to protect against other threads adding new pages while
594 * we were waiting for the iolock.
596 if (mapping->nrpages && iolock == XFS_IOLOCK_SHARED) {
597 xfs_rw_iunlock(ip, iolock);
598 iolock = XFS_IOLOCK_EXCL;
599 xfs_rw_ilock(ip, iolock);
602 ret = xfs_file_aio_write_checks(iocb, from, &iolock);
605 count = iov_iter_count(from);
606 end = iocb->ki_pos + count - 1;
609 * See xfs_file_dio_aio_read() for why we do a full-file flush here.
611 if (mapping->nrpages) {
612 ret = filemap_write_and_wait(VFS_I(ip)->i_mapping);
616 * Invalidate whole pages. This can return an error if we fail
617 * to invalidate a page, but this should never happen on XFS.
618 * Warn if it does fail.
620 ret = invalidate_inode_pages2(VFS_I(ip)->i_mapping);
626 * If we are doing unaligned IO, wait for all other IO to drain,
627 * otherwise demote the lock if we had to flush cached pages
630 inode_dio_wait(inode);
631 else if (iolock == XFS_IOLOCK_EXCL) {
632 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
633 iolock = XFS_IOLOCK_SHARED;
636 trace_xfs_file_direct_write(ip, count, iocb->ki_pos);
638 /* If this is a block-aligned directio CoW, remap immediately. */
639 if (xfs_is_reflink_inode(ip) && !unaligned_io) {
640 ret = xfs_reflink_allocate_cow_range(ip, iocb->ki_pos, count);
646 ret = __blockdev_direct_IO(iocb, inode, target->bt_bdev, &data,
647 xfs_get_blocks_direct, xfs_end_io_direct_write,
648 NULL, DIO_ASYNC_EXTEND);
650 /* see generic_file_direct_write() for why this is necessary */
651 if (mapping->nrpages) {
652 invalidate_inode_pages2_range(mapping,
653 iocb->ki_pos >> PAGE_SHIFT,
659 iov_iter_advance(from, ret);
662 xfs_rw_iunlock(ip, iolock);
665 * No fallback to buffered IO on errors for XFS, direct IO will either
666 * complete fully or fail.
668 ASSERT(ret < 0 || ret == count);
672 static noinline ssize_t
675 struct iov_iter *from)
677 struct inode *inode = iocb->ki_filp->f_mapping->host;
678 struct xfs_inode *ip = XFS_I(inode);
679 int iolock = XFS_IOLOCK_EXCL;
680 ssize_t ret, error = 0;
684 xfs_rw_ilock(ip, iolock);
685 ret = xfs_file_aio_write_checks(iocb, from, &iolock);
690 count = iov_iter_count(from);
692 trace_xfs_file_dax_write(ip, count, pos);
694 ret = iomap_dax_rw(iocb, from, &xfs_iomap_ops);
695 if (ret > 0 && iocb->ki_pos > i_size_read(inode)) {
696 i_size_write(inode, iocb->ki_pos);
697 error = xfs_setfilesize(ip, pos, ret);
701 xfs_rw_iunlock(ip, iolock);
702 return error ? error : ret;
706 xfs_file_buffered_aio_write(
708 struct iov_iter *from)
710 struct file *file = iocb->ki_filp;
711 struct address_space *mapping = file->f_mapping;
712 struct inode *inode = mapping->host;
713 struct xfs_inode *ip = XFS_I(inode);
716 int iolock = XFS_IOLOCK_EXCL;
718 xfs_rw_ilock(ip, iolock);
720 ret = xfs_file_aio_write_checks(iocb, from, &iolock);
724 /* We can write back this queue in page reclaim */
725 current->backing_dev_info = inode_to_bdi(inode);
728 trace_xfs_file_buffered_write(ip, iov_iter_count(from), iocb->ki_pos);
729 ret = iomap_file_buffered_write(iocb, from, &xfs_iomap_ops);
730 if (likely(ret >= 0))
734 * If we hit a space limit, try to free up some lingering preallocated
735 * space before returning an error. In the case of ENOSPC, first try to
736 * write back all dirty inodes to free up some of the excess reserved
737 * metadata space. This reduces the chances that the eofblocks scan
738 * waits on dirty mappings. Since xfs_flush_inodes() is serialized, this
739 * also behaves as a filter to prevent too many eofblocks scans from
740 * running at the same time.
742 if (ret == -EDQUOT && !enospc) {
743 enospc = xfs_inode_free_quota_eofblocks(ip);
746 enospc = xfs_inode_free_quota_cowblocks(ip);
749 } else if (ret == -ENOSPC && !enospc) {
750 struct xfs_eofblocks eofb = {0};
753 xfs_flush_inodes(ip->i_mount);
754 eofb.eof_scan_owner = ip->i_ino; /* for locking */
755 eofb.eof_flags = XFS_EOF_FLAGS_SYNC;
756 xfs_icache_free_eofblocks(ip->i_mount, &eofb);
760 current->backing_dev_info = NULL;
762 xfs_rw_iunlock(ip, iolock);
769 struct iov_iter *from)
771 struct file *file = iocb->ki_filp;
772 struct address_space *mapping = file->f_mapping;
773 struct inode *inode = mapping->host;
774 struct xfs_inode *ip = XFS_I(inode);
776 size_t ocount = iov_iter_count(from);
778 XFS_STATS_INC(ip->i_mount, xs_write_calls);
783 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
787 ret = xfs_file_dax_write(iocb, from);
788 else if (iocb->ki_flags & IOCB_DIRECT) {
790 * Allow a directio write to fall back to a buffered
791 * write *only* in the case that we're doing a reflink
792 * CoW. In all other directio scenarios we do not
793 * allow an operation to fall back to buffered mode.
795 ret = xfs_file_dio_aio_write(iocb, from);
800 ret = xfs_file_buffered_aio_write(iocb, from);
804 XFS_STATS_ADD(ip->i_mount, xs_write_bytes, ret);
806 /* Handle various SYNC-type writes */
807 ret = generic_write_sync(iocb, ret);
812 #define XFS_FALLOC_FL_SUPPORTED \
813 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
814 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE | \
815 FALLOC_FL_INSERT_RANGE | FALLOC_FL_UNSHARE_RANGE)
824 struct inode *inode = file_inode(file);
825 struct xfs_inode *ip = XFS_I(inode);
827 enum xfs_prealloc_flags flags = 0;
828 uint iolock = XFS_IOLOCK_EXCL;
830 bool do_file_insert = 0;
832 if (!S_ISREG(inode->i_mode))
834 if (mode & ~XFS_FALLOC_FL_SUPPORTED)
837 xfs_ilock(ip, iolock);
838 error = xfs_break_layouts(inode, &iolock, false);
842 xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
843 iolock |= XFS_MMAPLOCK_EXCL;
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 blksize_mask = (1 << inode->i_blkbits) - 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 blksize_mask = (1 << inode->i_blkbits) - 1;
874 new_size = i_size_read(inode) + len;
875 if (offset & blksize_mask || len & blksize_mask) {
880 /* check the new inode size does not wrap through zero */
881 if (new_size > inode->i_sb->s_maxbytes) {
886 /* Offset should be less than i_size */
887 if (offset >= i_size_read(inode)) {
893 flags |= XFS_PREALLOC_SET;
895 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
896 offset + len > i_size_read(inode)) {
897 new_size = offset + len;
898 error = inode_newsize_ok(inode, new_size);
903 if (mode & FALLOC_FL_ZERO_RANGE)
904 error = xfs_zero_file_space(ip, offset, len);
906 if (mode & FALLOC_FL_UNSHARE_RANGE) {
907 error = xfs_reflink_unshare(ip, offset, len);
911 error = xfs_alloc_file_space(ip, offset, len,
918 if (file->f_flags & O_DSYNC)
919 flags |= XFS_PREALLOC_SYNC;
921 error = xfs_update_prealloc_flags(ip, flags);
925 /* Change file size if needed */
929 iattr.ia_valid = ATTR_SIZE;
930 iattr.ia_size = new_size;
931 error = xfs_vn_setattr_size(file_dentry(file), &iattr);
937 * Perform hole insertion now that the file size has been
938 * updated so that if we crash during the operation we don't
939 * leave shifted extents past EOF and hence losing access to
940 * the data that is contained within them.
943 error = xfs_insert_file_space(ip, offset, len);
946 xfs_iunlock(ip, iolock);
951 * Flush all file writes out to disk.
954 xfs_file_wait_for_io(
965 bs = inode->i_sb->s_blocksize;
966 inode_dio_wait(inode);
968 rounding = max_t(xfs_off_t, bs, PAGE_SIZE);
969 ioffset = round_down(offset, rounding);
970 iendoffset = round_up(offset + len, rounding) - 1;
971 ret = filemap_write_and_wait_range(inode->i_mapping, ioffset,
976 /* Hook up to the VFS reflink function */
978 xfs_file_share_range(
979 struct file *file_in,
981 struct file *file_out,
986 struct inode *inode_in;
987 struct inode *inode_out;
993 unsigned int flags = 0;
995 inode_in = file_inode(file_in);
996 inode_out = file_inode(file_out);
997 bs = inode_out->i_sb->s_blocksize;
999 /* Don't touch certain kinds of inodes */
1000 if (IS_IMMUTABLE(inode_out))
1002 if (IS_SWAPFILE(inode_in) ||
1003 IS_SWAPFILE(inode_out))
1006 /* Reflink only works within this filesystem. */
1007 if (inode_in->i_sb != inode_out->i_sb)
1009 same_inode = (inode_in->i_ino == inode_out->i_ino);
1011 /* Don't reflink dirs, pipes, sockets... */
1012 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
1014 if (S_ISFIFO(inode_in->i_mode) || S_ISFIFO(inode_out->i_mode))
1016 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
1019 /* Don't share DAX file data for now. */
1020 if (IS_DAX(inode_in) || IS_DAX(inode_out))
1023 /* Are we going all the way to the end? */
1024 isize = i_size_read(inode_in);
1028 len = isize - pos_in;
1030 /* Ensure offsets don't wrap and the input is inside i_size */
1031 if (pos_in + len < pos_in || pos_out + len < pos_out ||
1032 pos_in + len > isize)
1035 /* Don't allow dedupe past EOF in the dest file */
1039 disize = i_size_read(inode_out);
1040 if (pos_out >= disize || pos_out + len > disize)
1044 /* If we're linking to EOF, continue to the block boundary. */
1045 if (pos_in + len == isize)
1046 blen = ALIGN(isize, bs) - pos_in;
1050 /* Only reflink if we're aligned to block boundaries */
1051 if (!IS_ALIGNED(pos_in, bs) || !IS_ALIGNED(pos_in + blen, bs) ||
1052 !IS_ALIGNED(pos_out, bs) || !IS_ALIGNED(pos_out + blen, bs))
1055 /* Don't allow overlapped reflink within the same file */
1056 if (same_inode && pos_out + blen > pos_in && pos_out < pos_in + blen)
1059 /* Wait for the completion of any pending IOs on srcfile */
1060 ret = xfs_file_wait_for_io(inode_in, pos_in, len);
1063 ret = xfs_file_wait_for_io(inode_out, pos_out, len);
1068 flags |= XFS_REFLINK_DEDUPE;
1069 ret = xfs_reflink_remap_range(XFS_I(inode_in), pos_in, XFS_I(inode_out),
1070 pos_out, len, flags);
1079 xfs_file_copy_range(
1080 struct file *file_in,
1082 struct file *file_out,
1089 error = xfs_file_share_range(file_in, pos_in, file_out, pos_out,
1097 xfs_file_clone_range(
1098 struct file *file_in,
1100 struct file *file_out,
1104 return xfs_file_share_range(file_in, pos_in, file_out, pos_out,
1108 #define XFS_MAX_DEDUPE_LEN (16 * 1024 * 1024)
1110 xfs_file_dedupe_range(
1111 struct file *src_file,
1114 struct file *dst_file,
1120 * Limit the total length we will dedupe for each operation.
1121 * This is intended to bound the total time spent in this
1122 * ioctl to something sane.
1124 if (len > XFS_MAX_DEDUPE_LEN)
1125 len = XFS_MAX_DEDUPE_LEN;
1127 error = xfs_file_share_range(src_file, loff, dst_file, dst_loff,
1136 struct inode *inode,
1139 if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1141 if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb)))
1148 struct inode *inode,
1151 struct xfs_inode *ip = XFS_I(inode);
1155 error = xfs_file_open(inode, file);
1160 * If there are any blocks, read-ahead block 0 as we're almost
1161 * certain to have the next operation be a read there.
1163 mode = xfs_ilock_data_map_shared(ip);
1164 if (ip->i_d.di_nextents > 0)
1165 xfs_dir3_data_readahead(ip, 0, -1);
1166 xfs_iunlock(ip, mode);
1172 struct inode *inode,
1175 return xfs_release(XFS_I(inode));
1181 struct dir_context *ctx)
1183 struct inode *inode = file_inode(file);
1184 xfs_inode_t *ip = XFS_I(inode);
1188 * The Linux API doesn't pass down the total size of the buffer
1189 * we read into down to the filesystem. With the filldir concept
1190 * it's not needed for correct information, but the XFS dir2 leaf
1191 * code wants an estimate of the buffer size to calculate it's
1192 * readahead window and size the buffers used for mapping to
1195 * Try to give it an estimate that's good enough, maybe at some
1196 * point we can change the ->readdir prototype to include the
1197 * buffer size. For now we use the current glibc buffer size.
1199 bufsize = (size_t)min_t(loff_t, 32768, ip->i_d.di_size);
1201 return xfs_readdir(ip, ctx, bufsize);
1205 * This type is designed to indicate the type of offset we would like
1206 * to search from page cache for xfs_seek_hole_data().
1214 * Lookup the desired type of offset from the given page.
1216 * On success, return true and the offset argument will point to the
1217 * start of the region that was found. Otherwise this function will
1218 * return false and keep the offset argument unchanged.
1221 xfs_lookup_buffer_offset(
1226 loff_t lastoff = page_offset(page);
1228 struct buffer_head *bh, *head;
1230 bh = head = page_buffers(page);
1233 * Unwritten extents that have data in the page
1234 * cache covering them can be identified by the
1235 * BH_Unwritten state flag. Pages with multiple
1236 * buffers might have a mix of holes, data and
1237 * unwritten extents - any buffer with valid
1238 * data in it should have BH_Uptodate flag set
1241 if (buffer_unwritten(bh) ||
1242 buffer_uptodate(bh)) {
1243 if (type == DATA_OFF)
1246 if (type == HOLE_OFF)
1254 lastoff += bh->b_size;
1255 } while ((bh = bh->b_this_page) != head);
1261 * This routine is called to find out and return a data or hole offset
1262 * from the page cache for unwritten extents according to the desired
1263 * type for xfs_seek_hole_data().
1265 * The argument offset is used to tell where we start to search from the
1266 * page cache. Map is used to figure out the end points of the range to
1269 * Return true if the desired type of offset was found, and the argument
1270 * offset is filled with that address. Otherwise, return false and keep
1274 xfs_find_get_desired_pgoff(
1275 struct inode *inode,
1276 struct xfs_bmbt_irec *map,
1280 struct xfs_inode *ip = XFS_I(inode);
1281 struct xfs_mount *mp = ip->i_mount;
1282 struct pagevec pvec;
1286 loff_t startoff = *offset;
1287 loff_t lastoff = startoff;
1290 pagevec_init(&pvec, 0);
1292 index = startoff >> PAGE_SHIFT;
1293 endoff = XFS_FSB_TO_B(mp, map->br_startoff + map->br_blockcount);
1294 end = endoff >> PAGE_SHIFT;
1300 want = min_t(pgoff_t, end - index, PAGEVEC_SIZE);
1301 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index,
1304 * No page mapped into given range. If we are searching holes
1305 * and if this is the first time we got into the loop, it means
1306 * that the given offset is landed in a hole, return it.
1308 * If we have already stepped through some block buffers to find
1309 * holes but they all contains data. In this case, the last
1310 * offset is already updated and pointed to the end of the last
1311 * mapped page, if it does not reach the endpoint to search,
1312 * that means there should be a hole between them.
1314 if (nr_pages == 0) {
1315 /* Data search found nothing */
1316 if (type == DATA_OFF)
1319 ASSERT(type == HOLE_OFF);
1320 if (lastoff == startoff || lastoff < endoff) {
1328 * At lease we found one page. If this is the first time we
1329 * step into the loop, and if the first page index offset is
1330 * greater than the given search offset, a hole was found.
1332 if (type == HOLE_OFF && lastoff == startoff &&
1333 lastoff < page_offset(pvec.pages[0])) {
1338 for (i = 0; i < nr_pages; i++) {
1339 struct page *page = pvec.pages[i];
1343 * At this point, the page may be truncated or
1344 * invalidated (changing page->mapping to NULL),
1345 * or even swizzled back from swapper_space to tmpfs
1346 * file mapping. However, page->index will not change
1347 * because we have a reference on the page.
1349 * Searching done if the page index is out of range.
1350 * If the current offset is not reaches the end of
1351 * the specified search range, there should be a hole
1354 if (page->index > end) {
1355 if (type == HOLE_OFF && lastoff < endoff) {
1364 * Page truncated or invalidated(page->mapping == NULL).
1365 * We can freely skip it and proceed to check the next
1368 if (unlikely(page->mapping != inode->i_mapping)) {
1373 if (!page_has_buffers(page)) {
1378 found = xfs_lookup_buffer_offset(page, &b_offset, type);
1381 * The found offset may be less than the start
1382 * point to search if this is the first time to
1385 *offset = max_t(loff_t, startoff, b_offset);
1391 * We either searching data but nothing was found, or
1392 * searching hole but found a data buffer. In either
1393 * case, probably the next page contains the desired
1394 * things, update the last offset to it so.
1396 lastoff = page_offset(page) + PAGE_SIZE;
1401 * The number of returned pages less than our desired, search
1402 * done. In this case, nothing was found for searching data,
1403 * but we found a hole behind the last offset.
1405 if (nr_pages < want) {
1406 if (type == HOLE_OFF) {
1413 index = pvec.pages[i - 1]->index + 1;
1414 pagevec_release(&pvec);
1415 } while (index <= end);
1418 pagevec_release(&pvec);
1423 * caller must lock inode with xfs_ilock_data_map_shared,
1424 * can we craft an appropriate ASSERT?
1426 * end is because the VFS-level lseek interface is defined such that any
1427 * offset past i_size shall return -ENXIO, but we use this for quota code
1428 * which does not maintain i_size, and we want to SEEK_DATA past i_size.
1431 __xfs_seek_hole_data(
1432 struct inode *inode,
1437 struct xfs_inode *ip = XFS_I(inode);
1438 struct xfs_mount *mp = ip->i_mount;
1439 loff_t uninitialized_var(offset);
1440 xfs_fileoff_t fsbno;
1441 xfs_filblks_t lastbno;
1450 * Try to read extents from the first block indicated
1451 * by fsbno to the end block of the file.
1453 fsbno = XFS_B_TO_FSBT(mp, start);
1454 lastbno = XFS_B_TO_FSB(mp, end);
1457 struct xfs_bmbt_irec map[2];
1461 error = xfs_bmapi_read(ip, fsbno, lastbno - fsbno, map, &nmap,
1466 /* No extents at given offset, must be beyond EOF */
1472 for (i = 0; i < nmap; i++) {
1473 offset = max_t(loff_t, start,
1474 XFS_FSB_TO_B(mp, map[i].br_startoff));
1476 /* Landed in the hole we wanted? */
1477 if (whence == SEEK_HOLE &&
1478 map[i].br_startblock == HOLESTARTBLOCK)
1481 /* Landed in the data extent we wanted? */
1482 if (whence == SEEK_DATA &&
1483 (map[i].br_startblock == DELAYSTARTBLOCK ||
1484 (map[i].br_state == XFS_EXT_NORM &&
1485 !isnullstartblock(map[i].br_startblock))))
1489 * Landed in an unwritten extent, try to search
1490 * for hole or data from page cache.
1492 if (map[i].br_state == XFS_EXT_UNWRITTEN) {
1493 if (xfs_find_get_desired_pgoff(inode, &map[i],
1494 whence == SEEK_HOLE ? HOLE_OFF : DATA_OFF,
1501 * We only received one extent out of the two requested. This
1502 * means we've hit EOF and didn't find what we are looking for.
1506 * If we were looking for a hole, set offset to
1507 * the end of the file (i.e., there is an implicit
1508 * hole at the end of any file).
1510 if (whence == SEEK_HOLE) {
1515 * If we were looking for data, it's nowhere to be found
1517 ASSERT(whence == SEEK_DATA);
1525 * Nothing was found, proceed to the next round of search
1526 * if the next reading offset is not at or beyond EOF.
1528 fsbno = map[i - 1].br_startoff + map[i - 1].br_blockcount;
1529 start = XFS_FSB_TO_B(mp, fsbno);
1531 if (whence == SEEK_HOLE) {
1535 ASSERT(whence == SEEK_DATA);
1543 * If at this point we have found the hole we wanted, the returned
1544 * offset may be bigger than the file size as it may be aligned to
1545 * page boundary for unwritten extents. We need to deal with this
1546 * situation in particular.
1548 if (whence == SEEK_HOLE)
1549 offset = min_t(loff_t, offset, end);
1563 struct inode *inode = file->f_mapping->host;
1564 struct xfs_inode *ip = XFS_I(inode);
1565 struct xfs_mount *mp = ip->i_mount;
1570 if (XFS_FORCED_SHUTDOWN(mp))
1573 lock = xfs_ilock_data_map_shared(ip);
1575 end = i_size_read(inode);
1576 offset = __xfs_seek_hole_data(inode, start, end, whence);
1582 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
1585 xfs_iunlock(ip, lock);
1602 return generic_file_llseek(file, offset, whence);
1605 return xfs_seek_hole_data(file, offset, whence);
1612 * Locking for serialisation of IO during page faults. This results in a lock
1616 * sb_start_pagefault(vfs, freeze)
1617 * i_mmaplock (XFS - truncate serialisation)
1619 * i_lock (XFS - extent map serialisation)
1623 * mmap()d file has taken write protection fault and is being made writable. We
1624 * can set the page state up correctly for a writable page, which means we can
1625 * do correct delalloc accounting (ENOSPC checking!) and unwritten extent
1629 xfs_filemap_page_mkwrite(
1630 struct vm_area_struct *vma,
1631 struct vm_fault *vmf)
1633 struct inode *inode = file_inode(vma->vm_file);
1636 trace_xfs_filemap_page_mkwrite(XFS_I(inode));
1638 sb_start_pagefault(inode->i_sb);
1639 file_update_time(vma->vm_file);
1640 xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1642 if (IS_DAX(inode)) {
1643 ret = iomap_dax_fault(vma, vmf, &xfs_iomap_ops);
1645 ret = iomap_page_mkwrite(vma, vmf, &xfs_iomap_ops);
1646 ret = block_page_mkwrite_return(ret);
1649 xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1650 sb_end_pagefault(inode->i_sb);
1657 struct vm_area_struct *vma,
1658 struct vm_fault *vmf)
1660 struct inode *inode = file_inode(vma->vm_file);
1663 trace_xfs_filemap_fault(XFS_I(inode));
1665 /* DAX can shortcut the normal fault path on write faults! */
1666 if ((vmf->flags & FAULT_FLAG_WRITE) && IS_DAX(inode))
1667 return xfs_filemap_page_mkwrite(vma, vmf);
1669 xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1670 if (IS_DAX(inode)) {
1672 * we do not want to trigger unwritten extent conversion on read
1673 * faults - that is unnecessary overhead and would also require
1674 * changes to xfs_get_blocks_direct() to map unwritten extent
1675 * ioend for conversion on read-only mappings.
1677 ret = iomap_dax_fault(vma, vmf, &xfs_iomap_ops);
1679 ret = filemap_fault(vma, vmf);
1680 xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1686 * Similar to xfs_filemap_fault(), the DAX fault path can call into here on
1687 * both read and write faults. Hence we need to handle both cases. There is no
1688 * ->pmd_mkwrite callout for huge pages, so we have a single function here to
1689 * handle both cases here. @flags carries the information on the type of fault
1693 xfs_filemap_pmd_fault(
1694 struct vm_area_struct *vma,
1699 struct inode *inode = file_inode(vma->vm_file);
1700 struct xfs_inode *ip = XFS_I(inode);
1704 return VM_FAULT_FALLBACK;
1706 trace_xfs_filemap_pmd_fault(ip);
1708 if (flags & FAULT_FLAG_WRITE) {
1709 sb_start_pagefault(inode->i_sb);
1710 file_update_time(vma->vm_file);
1713 xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1714 ret = dax_pmd_fault(vma, addr, pmd, flags, xfs_get_blocks_dax_fault);
1715 xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1717 if (flags & FAULT_FLAG_WRITE)
1718 sb_end_pagefault(inode->i_sb);
1724 * pfn_mkwrite was originally inteneded to ensure we capture time stamp
1725 * updates on write faults. In reality, it's need to serialise against
1726 * truncate similar to page_mkwrite. Hence we cycle the XFS_MMAPLOCK_SHARED
1727 * to ensure we serialise the fault barrier in place.
1730 xfs_filemap_pfn_mkwrite(
1731 struct vm_area_struct *vma,
1732 struct vm_fault *vmf)
1735 struct inode *inode = file_inode(vma->vm_file);
1736 struct xfs_inode *ip = XFS_I(inode);
1737 int ret = VM_FAULT_NOPAGE;
1740 trace_xfs_filemap_pfn_mkwrite(ip);
1742 sb_start_pagefault(inode->i_sb);
1743 file_update_time(vma->vm_file);
1745 /* check if the faulting page hasn't raced with truncate */
1746 xfs_ilock(ip, XFS_MMAPLOCK_SHARED);
1747 size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT;
1748 if (vmf->pgoff >= size)
1749 ret = VM_FAULT_SIGBUS;
1750 else if (IS_DAX(inode))
1751 ret = dax_pfn_mkwrite(vma, vmf);
1752 xfs_iunlock(ip, XFS_MMAPLOCK_SHARED);
1753 sb_end_pagefault(inode->i_sb);
1758 static const struct vm_operations_struct xfs_file_vm_ops = {
1759 .fault = xfs_filemap_fault,
1760 .pmd_fault = xfs_filemap_pmd_fault,
1761 .map_pages = filemap_map_pages,
1762 .page_mkwrite = xfs_filemap_page_mkwrite,
1763 .pfn_mkwrite = xfs_filemap_pfn_mkwrite,
1769 struct vm_area_struct *vma)
1771 file_accessed(filp);
1772 vma->vm_ops = &xfs_file_vm_ops;
1773 if (IS_DAX(file_inode(filp)))
1774 vma->vm_flags |= VM_MIXEDMAP | VM_HUGEPAGE;
1778 const struct file_operations xfs_file_operations = {
1779 .llseek = xfs_file_llseek,
1780 .read_iter = xfs_file_read_iter,
1781 .write_iter = xfs_file_write_iter,
1782 .splice_read = generic_file_splice_read,
1783 .splice_write = iter_file_splice_write,
1784 .unlocked_ioctl = xfs_file_ioctl,
1785 #ifdef CONFIG_COMPAT
1786 .compat_ioctl = xfs_file_compat_ioctl,
1788 .mmap = xfs_file_mmap,
1789 .open = xfs_file_open,
1790 .release = xfs_file_release,
1791 .fsync = xfs_file_fsync,
1792 .get_unmapped_area = thp_get_unmapped_area,
1793 .fallocate = xfs_file_fallocate,
1794 .copy_file_range = xfs_file_copy_range,
1795 .clone_file_range = xfs_file_clone_range,
1796 .dedupe_file_range = xfs_file_dedupe_range,
1799 const struct file_operations xfs_dir_file_operations = {
1800 .open = xfs_dir_open,
1801 .read = generic_read_dir,
1802 .iterate_shared = xfs_file_readdir,
1803 .llseek = generic_file_llseek,
1804 .unlocked_ioctl = xfs_file_ioctl,
1805 #ifdef CONFIG_COMPAT
1806 .compat_ioctl = xfs_file_compat_ioctl,
1808 .fsync = xfs_dir_fsync,