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 loff_t end = iocb->ki_pos + count - 1;
253 struct iov_iter data;
254 struct xfs_buftarg *target;
257 trace_xfs_file_direct_read(ip, count, iocb->ki_pos);
260 return 0; /* skip atime */
262 if (XFS_IS_REALTIME_INODE(ip))
263 target = ip->i_mount->m_rtdev_targp;
265 target = ip->i_mount->m_ddev_targp;
267 /* DIO must be aligned to device logical sector size */
268 if ((iocb->ki_pos | count) & target->bt_logical_sectormask) {
269 if (iocb->ki_pos == isize)
274 file_accessed(iocb->ki_filp);
276 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
277 if (mapping->nrpages) {
278 ret = filemap_write_and_wait_range(mapping, iocb->ki_pos, end);
283 * Invalidate whole pages. This can return an error if we fail
284 * to invalidate a page, but this should never happen on XFS.
285 * Warn if it does fail.
287 ret = invalidate_inode_pages2_range(mapping,
288 iocb->ki_pos >> PAGE_SHIFT, end >> PAGE_SHIFT);
294 ret = __blockdev_direct_IO(iocb, inode, target->bt_bdev, &data,
295 xfs_get_blocks_direct, NULL, NULL, 0);
298 iov_iter_advance(to, ret);
302 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
306 static noinline ssize_t
311 struct xfs_inode *ip = XFS_I(iocb->ki_filp->f_mapping->host);
312 size_t count = iov_iter_count(to);
315 trace_xfs_file_dax_read(ip, count, iocb->ki_pos);
318 return 0; /* skip atime */
320 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
321 ret = iomap_dax_rw(iocb, to, &xfs_iomap_ops);
322 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
324 file_accessed(iocb->ki_filp);
329 xfs_file_buffered_aio_read(
333 struct xfs_inode *ip = XFS_I(file_inode(iocb->ki_filp));
336 trace_xfs_file_buffered_read(ip, iov_iter_count(to), iocb->ki_pos);
338 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
339 ret = generic_file_read_iter(iocb, to);
340 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
350 struct inode *inode = file_inode(iocb->ki_filp);
351 struct xfs_mount *mp = XFS_I(inode)->i_mount;
354 XFS_STATS_INC(mp, xs_read_calls);
356 if (XFS_FORCED_SHUTDOWN(mp))
360 ret = xfs_file_dax_read(iocb, to);
361 else if (iocb->ki_flags & IOCB_DIRECT)
362 ret = xfs_file_dio_aio_read(iocb, to);
364 ret = xfs_file_buffered_aio_read(iocb, to);
367 XFS_STATS_ADD(mp, xs_read_bytes, ret);
372 * Zero any on disk space between the current EOF and the new, larger EOF.
374 * This handles the normal case of zeroing the remainder of the last block in
375 * the file and the unusual case of zeroing blocks out beyond the size of the
376 * file. This second case only happens with fixed size extents and when the
377 * system crashes before the inode size was updated but after blocks were
380 * Expects the iolock to be held exclusive, and will take the ilock internally.
382 int /* error (positive) */
384 struct xfs_inode *ip,
385 xfs_off_t offset, /* starting I/O offset */
386 xfs_fsize_t isize, /* current inode size */
389 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
390 ASSERT(offset > isize);
392 trace_xfs_zero_eof(ip, isize, offset - isize);
393 return xfs_zero_range(ip, isize, offset - isize, did_zeroing);
397 * Common pre-write limit and setup checks.
399 * Called with the iolocked held either shared and exclusive according to
400 * @iolock, and returns with it held. Might upgrade the iolock to exclusive
401 * if called for a direct write beyond i_size.
404 xfs_file_aio_write_checks(
406 struct iov_iter *from,
409 struct file *file = iocb->ki_filp;
410 struct inode *inode = file->f_mapping->host;
411 struct xfs_inode *ip = XFS_I(inode);
413 size_t count = iov_iter_count(from);
414 bool drained_dio = false;
417 error = generic_write_checks(iocb, from);
421 error = xfs_break_layouts(inode, iolock, true);
425 /* For changing security info in file_remove_privs() we need i_mutex */
426 if (*iolock == XFS_IOLOCK_SHARED && !IS_NOSEC(inode)) {
427 xfs_rw_iunlock(ip, *iolock);
428 *iolock = XFS_IOLOCK_EXCL;
429 xfs_rw_ilock(ip, *iolock);
433 * If the offset is beyond the size of the file, we need to zero any
434 * blocks that fall between the existing EOF and the start of this
435 * write. If zeroing is needed and we are currently holding the
436 * iolock shared, we need to update it to exclusive which implies
437 * having to redo all checks before.
439 * We need to serialise against EOF updates that occur in IO
440 * completions here. We want to make sure that nobody is changing the
441 * size while we do this check until we have placed an IO barrier (i.e.
442 * hold the XFS_IOLOCK_EXCL) that prevents new IO from being dispatched.
443 * The spinlock effectively forms a memory barrier once we have the
444 * XFS_IOLOCK_EXCL so we are guaranteed to see the latest EOF value
445 * and hence be able to correctly determine if we need to run zeroing.
447 spin_lock(&ip->i_flags_lock);
448 if (iocb->ki_pos > i_size_read(inode)) {
451 spin_unlock(&ip->i_flags_lock);
453 if (*iolock == XFS_IOLOCK_SHARED) {
454 xfs_rw_iunlock(ip, *iolock);
455 *iolock = XFS_IOLOCK_EXCL;
456 xfs_rw_ilock(ip, *iolock);
457 iov_iter_reexpand(from, count);
460 * We now have an IO submission barrier in place, but
461 * AIO can do EOF updates during IO completion and hence
462 * we now need to wait for all of them to drain. Non-AIO
463 * DIO will have drained before we are given the
464 * XFS_IOLOCK_EXCL, and so for most cases this wait is a
467 inode_dio_wait(inode);
471 error = xfs_zero_eof(ip, iocb->ki_pos, i_size_read(inode), &zero);
475 spin_unlock(&ip->i_flags_lock);
478 * Updating the timestamps will grab the ilock again from
479 * xfs_fs_dirty_inode, so we have to call it after dropping the
480 * lock above. Eventually we should look into a way to avoid
481 * the pointless lock roundtrip.
483 if (likely(!(file->f_mode & FMODE_NOCMTIME))) {
484 error = file_update_time(file);
490 * If we're writing the file then make sure to clear the setuid and
491 * setgid bits if the process is not being run by root. This keeps
492 * people from modifying setuid and setgid binaries.
494 if (!IS_NOSEC(inode))
495 return file_remove_privs(file);
500 * xfs_file_dio_aio_write - handle direct IO writes
502 * Lock the inode appropriately to prepare for and issue a direct IO write.
503 * By separating it from the buffered write path we remove all the tricky to
504 * follow locking changes and looping.
506 * If there are cached pages or we're extending the file, we need IOLOCK_EXCL
507 * until we're sure the bytes at the new EOF have been zeroed and/or the cached
508 * pages are flushed out.
510 * In most cases the direct IO writes will be done holding IOLOCK_SHARED
511 * allowing them to be done in parallel with reads and other direct IO writes.
512 * However, if the IO is not aligned to filesystem blocks, the direct IO layer
513 * needs to do sub-block zeroing and that requires serialisation against other
514 * direct IOs to the same block. In this case we need to serialise the
515 * submission of the unaligned IOs so that we don't get racing block zeroing in
516 * the dio layer. To avoid the problem with aio, we also need to wait for
517 * outstanding IOs to complete so that unwritten extent conversion is completed
518 * before we try to map the overlapping block. This is currently implemented by
519 * hitting it with a big hammer (i.e. inode_dio_wait()).
521 * Returns with locks held indicated by @iolock and errors indicated by
522 * negative return values.
525 xfs_file_dio_aio_write(
527 struct iov_iter *from)
529 struct file *file = iocb->ki_filp;
530 struct address_space *mapping = file->f_mapping;
531 struct inode *inode = mapping->host;
532 struct xfs_inode *ip = XFS_I(inode);
533 struct xfs_mount *mp = ip->i_mount;
535 int unaligned_io = 0;
537 size_t count = iov_iter_count(from);
539 struct iov_iter data;
540 struct xfs_buftarg *target = XFS_IS_REALTIME_INODE(ip) ?
541 mp->m_rtdev_targp : mp->m_ddev_targp;
543 /* DIO must be aligned to device logical sector size */
544 if ((iocb->ki_pos | count) & target->bt_logical_sectormask)
548 * Don't take the exclusive iolock here unless the I/O is unaligned to
549 * the file system block size. We don't need to consider the EOF
550 * extension case here because xfs_file_aio_write_checks() will relock
551 * the inode as necessary for EOF zeroing cases and fill out the new
552 * inode size as appropriate.
554 if ((iocb->ki_pos & mp->m_blockmask) ||
555 ((iocb->ki_pos + count) & mp->m_blockmask)) {
557 iolock = XFS_IOLOCK_EXCL;
559 iolock = XFS_IOLOCK_SHARED;
562 xfs_rw_ilock(ip, iolock);
564 ret = xfs_file_aio_write_checks(iocb, from, &iolock);
567 count = iov_iter_count(from);
568 end = iocb->ki_pos + count - 1;
570 if (mapping->nrpages) {
571 ret = filemap_write_and_wait_range(mapping, iocb->ki_pos, end);
576 * Invalidate whole pages. This can return an error if we fail
577 * to invalidate a page, but this should never happen on XFS.
578 * Warn if it does fail.
580 ret = invalidate_inode_pages2_range(mapping,
581 iocb->ki_pos >> PAGE_SHIFT, end >> PAGE_SHIFT);
587 * If we are doing unaligned IO, wait for all other IO to drain,
588 * otherwise demote the lock if we had to take the exclusive lock
589 * for other reasons in xfs_file_aio_write_checks.
592 inode_dio_wait(inode);
593 else if (iolock == XFS_IOLOCK_EXCL) {
594 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
595 iolock = XFS_IOLOCK_SHARED;
598 trace_xfs_file_direct_write(ip, count, iocb->ki_pos);
600 /* If this is a block-aligned directio CoW, remap immediately. */
601 if (xfs_is_reflink_inode(ip) && !unaligned_io) {
602 ret = xfs_reflink_allocate_cow_range(ip, iocb->ki_pos, count);
608 ret = __blockdev_direct_IO(iocb, inode, target->bt_bdev, &data,
609 xfs_get_blocks_direct, xfs_end_io_direct_write,
610 NULL, DIO_ASYNC_EXTEND);
612 /* see generic_file_direct_write() for why this is necessary */
613 if (mapping->nrpages) {
614 invalidate_inode_pages2_range(mapping,
615 iocb->ki_pos >> PAGE_SHIFT,
621 iov_iter_advance(from, ret);
624 xfs_rw_iunlock(ip, iolock);
627 * No fallback to buffered IO on errors for XFS, direct IO will either
628 * complete fully or fail.
630 ASSERT(ret < 0 || ret == count);
634 static noinline ssize_t
637 struct iov_iter *from)
639 struct inode *inode = iocb->ki_filp->f_mapping->host;
640 struct xfs_inode *ip = XFS_I(inode);
641 int iolock = XFS_IOLOCK_EXCL;
642 ssize_t ret, error = 0;
646 xfs_rw_ilock(ip, iolock);
647 ret = xfs_file_aio_write_checks(iocb, from, &iolock);
652 count = iov_iter_count(from);
654 trace_xfs_file_dax_write(ip, count, pos);
656 ret = iomap_dax_rw(iocb, from, &xfs_iomap_ops);
657 if (ret > 0 && iocb->ki_pos > i_size_read(inode)) {
658 i_size_write(inode, iocb->ki_pos);
659 error = xfs_setfilesize(ip, pos, ret);
663 xfs_rw_iunlock(ip, iolock);
664 return error ? error : ret;
668 xfs_file_buffered_aio_write(
670 struct iov_iter *from)
672 struct file *file = iocb->ki_filp;
673 struct address_space *mapping = file->f_mapping;
674 struct inode *inode = mapping->host;
675 struct xfs_inode *ip = XFS_I(inode);
678 int iolock = XFS_IOLOCK_EXCL;
680 xfs_rw_ilock(ip, iolock);
682 ret = xfs_file_aio_write_checks(iocb, from, &iolock);
686 /* We can write back this queue in page reclaim */
687 current->backing_dev_info = inode_to_bdi(inode);
690 trace_xfs_file_buffered_write(ip, iov_iter_count(from), iocb->ki_pos);
691 ret = iomap_file_buffered_write(iocb, from, &xfs_iomap_ops);
692 if (likely(ret >= 0))
696 * If we hit a space limit, try to free up some lingering preallocated
697 * space before returning an error. In the case of ENOSPC, first try to
698 * write back all dirty inodes to free up some of the excess reserved
699 * metadata space. This reduces the chances that the eofblocks scan
700 * waits on dirty mappings. Since xfs_flush_inodes() is serialized, this
701 * also behaves as a filter to prevent too many eofblocks scans from
702 * running at the same time.
704 if (ret == -EDQUOT && !enospc) {
705 enospc = xfs_inode_free_quota_eofblocks(ip);
708 enospc = xfs_inode_free_quota_cowblocks(ip);
711 } else if (ret == -ENOSPC && !enospc) {
712 struct xfs_eofblocks eofb = {0};
715 xfs_flush_inodes(ip->i_mount);
716 eofb.eof_scan_owner = ip->i_ino; /* for locking */
717 eofb.eof_flags = XFS_EOF_FLAGS_SYNC;
718 xfs_icache_free_eofblocks(ip->i_mount, &eofb);
722 current->backing_dev_info = NULL;
724 xfs_rw_iunlock(ip, iolock);
731 struct iov_iter *from)
733 struct file *file = iocb->ki_filp;
734 struct address_space *mapping = file->f_mapping;
735 struct inode *inode = mapping->host;
736 struct xfs_inode *ip = XFS_I(inode);
738 size_t ocount = iov_iter_count(from);
740 XFS_STATS_INC(ip->i_mount, xs_write_calls);
745 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
749 ret = xfs_file_dax_write(iocb, from);
750 else if (iocb->ki_flags & IOCB_DIRECT) {
752 * Allow a directio write to fall back to a buffered
753 * write *only* in the case that we're doing a reflink
754 * CoW. In all other directio scenarios we do not
755 * allow an operation to fall back to buffered mode.
757 ret = xfs_file_dio_aio_write(iocb, from);
762 ret = xfs_file_buffered_aio_write(iocb, from);
766 XFS_STATS_ADD(ip->i_mount, xs_write_bytes, ret);
768 /* Handle various SYNC-type writes */
769 ret = generic_write_sync(iocb, ret);
774 #define XFS_FALLOC_FL_SUPPORTED \
775 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
776 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE | \
777 FALLOC_FL_INSERT_RANGE | FALLOC_FL_UNSHARE_RANGE)
786 struct inode *inode = file_inode(file);
787 struct xfs_inode *ip = XFS_I(inode);
789 enum xfs_prealloc_flags flags = 0;
790 uint iolock = XFS_IOLOCK_EXCL;
792 bool do_file_insert = 0;
794 if (!S_ISREG(inode->i_mode))
796 if (mode & ~XFS_FALLOC_FL_SUPPORTED)
799 xfs_ilock(ip, iolock);
800 error = xfs_break_layouts(inode, &iolock, false);
804 xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
805 iolock |= XFS_MMAPLOCK_EXCL;
807 if (mode & FALLOC_FL_PUNCH_HOLE) {
808 error = xfs_free_file_space(ip, offset, len);
811 } else if (mode & FALLOC_FL_COLLAPSE_RANGE) {
812 unsigned blksize_mask = (1 << inode->i_blkbits) - 1;
814 if (offset & blksize_mask || len & blksize_mask) {
820 * There is no need to overlap collapse range with EOF,
821 * in which case it is effectively a truncate operation
823 if (offset + len >= i_size_read(inode)) {
828 new_size = i_size_read(inode) - len;
830 error = xfs_collapse_file_space(ip, offset, len);
833 } else if (mode & FALLOC_FL_INSERT_RANGE) {
834 unsigned blksize_mask = (1 << inode->i_blkbits) - 1;
836 new_size = i_size_read(inode) + len;
837 if (offset & blksize_mask || len & blksize_mask) {
842 /* check the new inode size does not wrap through zero */
843 if (new_size > inode->i_sb->s_maxbytes) {
848 /* Offset should be less than i_size */
849 if (offset >= i_size_read(inode)) {
855 flags |= XFS_PREALLOC_SET;
857 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
858 offset + len > i_size_read(inode)) {
859 new_size = offset + len;
860 error = inode_newsize_ok(inode, new_size);
865 if (mode & FALLOC_FL_ZERO_RANGE)
866 error = xfs_zero_file_space(ip, offset, len);
868 if (mode & FALLOC_FL_UNSHARE_RANGE) {
869 error = xfs_reflink_unshare(ip, offset, len);
873 error = xfs_alloc_file_space(ip, offset, len,
880 if (file->f_flags & O_DSYNC)
881 flags |= XFS_PREALLOC_SYNC;
883 error = xfs_update_prealloc_flags(ip, flags);
887 /* Change file size if needed */
891 iattr.ia_valid = ATTR_SIZE;
892 iattr.ia_size = new_size;
893 error = xfs_vn_setattr_size(file_dentry(file), &iattr);
899 * Perform hole insertion now that the file size has been
900 * updated so that if we crash during the operation we don't
901 * leave shifted extents past EOF and hence losing access to
902 * the data that is contained within them.
905 error = xfs_insert_file_space(ip, offset, len);
908 xfs_iunlock(ip, iolock);
912 /* Hook up to the VFS reflink function */
914 xfs_file_share_range(
915 struct file *file_in,
917 struct file *file_out,
922 struct inode *inode_in;
923 struct inode *inode_out;
929 unsigned int flags = 0;
931 inode_in = file_inode(file_in);
932 inode_out = file_inode(file_out);
933 bs = inode_out->i_sb->s_blocksize;
935 /* Lock both files against IO */
936 same_inode = (inode_in == inode_out);
938 xfs_ilock(XFS_I(inode_in), XFS_IOLOCK_EXCL);
939 xfs_ilock(XFS_I(inode_in), XFS_MMAPLOCK_EXCL);
941 xfs_lock_two_inodes(XFS_I(inode_in), XFS_I(inode_out),
943 xfs_lock_two_inodes(XFS_I(inode_in), XFS_I(inode_out),
947 /* Don't touch certain kinds of inodes */
949 if (IS_IMMUTABLE(inode_out))
952 if (IS_SWAPFILE(inode_in) || IS_SWAPFILE(inode_out))
955 /* Don't reflink dirs, pipes, sockets... */
957 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
960 if (S_ISFIFO(inode_in->i_mode) || S_ISFIFO(inode_out->i_mode))
962 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
965 /* Don't share DAX file data for now. */
966 if (IS_DAX(inode_in) || IS_DAX(inode_out))
969 /* Are we going all the way to the end? */
970 isize = i_size_read(inode_in);
977 len = isize - pos_in;
979 /* Ensure offsets don't wrap and the input is inside i_size */
980 if (pos_in + len < pos_in || pos_out + len < pos_out ||
981 pos_in + len > isize)
984 /* Don't allow dedupe past EOF in the dest file */
988 disize = i_size_read(inode_out);
989 if (pos_out >= disize || pos_out + len > disize)
993 /* If we're linking to EOF, continue to the block boundary. */
994 if (pos_in + len == isize)
995 blen = ALIGN(isize, bs) - pos_in;
999 /* Only reflink if we're aligned to block boundaries */
1000 if (!IS_ALIGNED(pos_in, bs) || !IS_ALIGNED(pos_in + blen, bs) ||
1001 !IS_ALIGNED(pos_out, bs) || !IS_ALIGNED(pos_out + blen, bs))
1004 /* Don't allow overlapped reflink within the same file */
1005 if (same_inode && pos_out + blen > pos_in && pos_out < pos_in + blen)
1008 /* Wait for the completion of any pending IOs on both files */
1009 inode_dio_wait(inode_in);
1011 inode_dio_wait(inode_out);
1013 ret = filemap_write_and_wait_range(inode_in->i_mapping,
1014 pos_in, pos_in + len - 1);
1018 ret = filemap_write_and_wait_range(inode_out->i_mapping,
1019 pos_out, pos_out + len - 1);
1024 flags |= XFS_REFLINK_DEDUPE;
1025 ret = xfs_reflink_remap_range(XFS_I(inode_in), pos_in, XFS_I(inode_out),
1026 pos_out, len, flags);
1029 xfs_iunlock(XFS_I(inode_in), XFS_MMAPLOCK_EXCL);
1030 xfs_iunlock(XFS_I(inode_in), XFS_IOLOCK_EXCL);
1032 xfs_iunlock(XFS_I(inode_out), XFS_MMAPLOCK_EXCL);
1033 xfs_iunlock(XFS_I(inode_out), XFS_IOLOCK_EXCL);
1039 xfs_file_copy_range(
1040 struct file *file_in,
1042 struct file *file_out,
1049 error = xfs_file_share_range(file_in, pos_in, file_out, pos_out,
1057 xfs_file_clone_range(
1058 struct file *file_in,
1060 struct file *file_out,
1064 return xfs_file_share_range(file_in, pos_in, file_out, pos_out,
1068 #define XFS_MAX_DEDUPE_LEN (16 * 1024 * 1024)
1070 xfs_file_dedupe_range(
1071 struct file *src_file,
1074 struct file *dst_file,
1080 * Limit the total length we will dedupe for each operation.
1081 * This is intended to bound the total time spent in this
1082 * ioctl to something sane.
1084 if (len > XFS_MAX_DEDUPE_LEN)
1085 len = XFS_MAX_DEDUPE_LEN;
1087 error = xfs_file_share_range(src_file, loff, dst_file, dst_loff,
1096 struct inode *inode,
1099 if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1101 if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb)))
1108 struct inode *inode,
1111 struct xfs_inode *ip = XFS_I(inode);
1115 error = xfs_file_open(inode, file);
1120 * If there are any blocks, read-ahead block 0 as we're almost
1121 * certain to have the next operation be a read there.
1123 mode = xfs_ilock_data_map_shared(ip);
1124 if (ip->i_d.di_nextents > 0)
1125 xfs_dir3_data_readahead(ip, 0, -1);
1126 xfs_iunlock(ip, mode);
1132 struct inode *inode,
1135 return xfs_release(XFS_I(inode));
1141 struct dir_context *ctx)
1143 struct inode *inode = file_inode(file);
1144 xfs_inode_t *ip = XFS_I(inode);
1148 * The Linux API doesn't pass down the total size of the buffer
1149 * we read into down to the filesystem. With the filldir concept
1150 * it's not needed for correct information, but the XFS dir2 leaf
1151 * code wants an estimate of the buffer size to calculate it's
1152 * readahead window and size the buffers used for mapping to
1155 * Try to give it an estimate that's good enough, maybe at some
1156 * point we can change the ->readdir prototype to include the
1157 * buffer size. For now we use the current glibc buffer size.
1159 bufsize = (size_t)min_t(loff_t, 32768, ip->i_d.di_size);
1161 return xfs_readdir(ip, ctx, bufsize);
1165 * This type is designed to indicate the type of offset we would like
1166 * to search from page cache for xfs_seek_hole_data().
1174 * Lookup the desired type of offset from the given page.
1176 * On success, return true and the offset argument will point to the
1177 * start of the region that was found. Otherwise this function will
1178 * return false and keep the offset argument unchanged.
1181 xfs_lookup_buffer_offset(
1186 loff_t lastoff = page_offset(page);
1188 struct buffer_head *bh, *head;
1190 bh = head = page_buffers(page);
1193 * Unwritten extents that have data in the page
1194 * cache covering them can be identified by the
1195 * BH_Unwritten state flag. Pages with multiple
1196 * buffers might have a mix of holes, data and
1197 * unwritten extents - any buffer with valid
1198 * data in it should have BH_Uptodate flag set
1201 if (buffer_unwritten(bh) ||
1202 buffer_uptodate(bh)) {
1203 if (type == DATA_OFF)
1206 if (type == HOLE_OFF)
1214 lastoff += bh->b_size;
1215 } while ((bh = bh->b_this_page) != head);
1221 * This routine is called to find out and return a data or hole offset
1222 * from the page cache for unwritten extents according to the desired
1223 * type for xfs_seek_hole_data().
1225 * The argument offset is used to tell where we start to search from the
1226 * page cache. Map is used to figure out the end points of the range to
1229 * Return true if the desired type of offset was found, and the argument
1230 * offset is filled with that address. Otherwise, return false and keep
1234 xfs_find_get_desired_pgoff(
1235 struct inode *inode,
1236 struct xfs_bmbt_irec *map,
1240 struct xfs_inode *ip = XFS_I(inode);
1241 struct xfs_mount *mp = ip->i_mount;
1242 struct pagevec pvec;
1246 loff_t startoff = *offset;
1247 loff_t lastoff = startoff;
1250 pagevec_init(&pvec, 0);
1252 index = startoff >> PAGE_SHIFT;
1253 endoff = XFS_FSB_TO_B(mp, map->br_startoff + map->br_blockcount);
1254 end = endoff >> PAGE_SHIFT;
1260 want = min_t(pgoff_t, end - index, PAGEVEC_SIZE);
1261 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index,
1264 * No page mapped into given range. If we are searching holes
1265 * and if this is the first time we got into the loop, it means
1266 * that the given offset is landed in a hole, return it.
1268 * If we have already stepped through some block buffers to find
1269 * holes but they all contains data. In this case, the last
1270 * offset is already updated and pointed to the end of the last
1271 * mapped page, if it does not reach the endpoint to search,
1272 * that means there should be a hole between them.
1274 if (nr_pages == 0) {
1275 /* Data search found nothing */
1276 if (type == DATA_OFF)
1279 ASSERT(type == HOLE_OFF);
1280 if (lastoff == startoff || lastoff < endoff) {
1288 * At lease we found one page. If this is the first time we
1289 * step into the loop, and if the first page index offset is
1290 * greater than the given search offset, a hole was found.
1292 if (type == HOLE_OFF && lastoff == startoff &&
1293 lastoff < page_offset(pvec.pages[0])) {
1298 for (i = 0; i < nr_pages; i++) {
1299 struct page *page = pvec.pages[i];
1303 * At this point, the page may be truncated or
1304 * invalidated (changing page->mapping to NULL),
1305 * or even swizzled back from swapper_space to tmpfs
1306 * file mapping. However, page->index will not change
1307 * because we have a reference on the page.
1309 * Searching done if the page index is out of range.
1310 * If the current offset is not reaches the end of
1311 * the specified search range, there should be a hole
1314 if (page->index > end) {
1315 if (type == HOLE_OFF && lastoff < endoff) {
1324 * Page truncated or invalidated(page->mapping == NULL).
1325 * We can freely skip it and proceed to check the next
1328 if (unlikely(page->mapping != inode->i_mapping)) {
1333 if (!page_has_buffers(page)) {
1338 found = xfs_lookup_buffer_offset(page, &b_offset, type);
1341 * The found offset may be less than the start
1342 * point to search if this is the first time to
1345 *offset = max_t(loff_t, startoff, b_offset);
1351 * We either searching data but nothing was found, or
1352 * searching hole but found a data buffer. In either
1353 * case, probably the next page contains the desired
1354 * things, update the last offset to it so.
1356 lastoff = page_offset(page) + PAGE_SIZE;
1361 * The number of returned pages less than our desired, search
1362 * done. In this case, nothing was found for searching data,
1363 * but we found a hole behind the last offset.
1365 if (nr_pages < want) {
1366 if (type == HOLE_OFF) {
1373 index = pvec.pages[i - 1]->index + 1;
1374 pagevec_release(&pvec);
1375 } while (index <= end);
1378 pagevec_release(&pvec);
1383 * caller must lock inode with xfs_ilock_data_map_shared,
1384 * can we craft an appropriate ASSERT?
1386 * end is because the VFS-level lseek interface is defined such that any
1387 * offset past i_size shall return -ENXIO, but we use this for quota code
1388 * which does not maintain i_size, and we want to SEEK_DATA past i_size.
1391 __xfs_seek_hole_data(
1392 struct inode *inode,
1397 struct xfs_inode *ip = XFS_I(inode);
1398 struct xfs_mount *mp = ip->i_mount;
1399 loff_t uninitialized_var(offset);
1400 xfs_fileoff_t fsbno;
1401 xfs_filblks_t lastbno;
1410 * Try to read extents from the first block indicated
1411 * by fsbno to the end block of the file.
1413 fsbno = XFS_B_TO_FSBT(mp, start);
1414 lastbno = XFS_B_TO_FSB(mp, end);
1417 struct xfs_bmbt_irec map[2];
1421 error = xfs_bmapi_read(ip, fsbno, lastbno - fsbno, map, &nmap,
1426 /* No extents at given offset, must be beyond EOF */
1432 for (i = 0; i < nmap; i++) {
1433 offset = max_t(loff_t, start,
1434 XFS_FSB_TO_B(mp, map[i].br_startoff));
1436 /* Landed in the hole we wanted? */
1437 if (whence == SEEK_HOLE &&
1438 map[i].br_startblock == HOLESTARTBLOCK)
1441 /* Landed in the data extent we wanted? */
1442 if (whence == SEEK_DATA &&
1443 (map[i].br_startblock == DELAYSTARTBLOCK ||
1444 (map[i].br_state == XFS_EXT_NORM &&
1445 !isnullstartblock(map[i].br_startblock))))
1449 * Landed in an unwritten extent, try to search
1450 * for hole or data from page cache.
1452 if (map[i].br_state == XFS_EXT_UNWRITTEN) {
1453 if (xfs_find_get_desired_pgoff(inode, &map[i],
1454 whence == SEEK_HOLE ? HOLE_OFF : DATA_OFF,
1461 * We only received one extent out of the two requested. This
1462 * means we've hit EOF and didn't find what we are looking for.
1466 * If we were looking for a hole, set offset to
1467 * the end of the file (i.e., there is an implicit
1468 * hole at the end of any file).
1470 if (whence == SEEK_HOLE) {
1475 * If we were looking for data, it's nowhere to be found
1477 ASSERT(whence == SEEK_DATA);
1485 * Nothing was found, proceed to the next round of search
1486 * if the next reading offset is not at or beyond EOF.
1488 fsbno = map[i - 1].br_startoff + map[i - 1].br_blockcount;
1489 start = XFS_FSB_TO_B(mp, fsbno);
1491 if (whence == SEEK_HOLE) {
1495 ASSERT(whence == SEEK_DATA);
1503 * If at this point we have found the hole we wanted, the returned
1504 * offset may be bigger than the file size as it may be aligned to
1505 * page boundary for unwritten extents. We need to deal with this
1506 * situation in particular.
1508 if (whence == SEEK_HOLE)
1509 offset = min_t(loff_t, offset, end);
1523 struct inode *inode = file->f_mapping->host;
1524 struct xfs_inode *ip = XFS_I(inode);
1525 struct xfs_mount *mp = ip->i_mount;
1530 if (XFS_FORCED_SHUTDOWN(mp))
1533 lock = xfs_ilock_data_map_shared(ip);
1535 end = i_size_read(inode);
1536 offset = __xfs_seek_hole_data(inode, start, end, whence);
1542 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
1545 xfs_iunlock(ip, lock);
1562 return generic_file_llseek(file, offset, whence);
1565 return xfs_seek_hole_data(file, offset, whence);
1572 * Locking for serialisation of IO during page faults. This results in a lock
1576 * sb_start_pagefault(vfs, freeze)
1577 * i_mmaplock (XFS - truncate serialisation)
1579 * i_lock (XFS - extent map serialisation)
1583 * mmap()d file has taken write protection fault and is being made writable. We
1584 * can set the page state up correctly for a writable page, which means we can
1585 * do correct delalloc accounting (ENOSPC checking!) and unwritten extent
1589 xfs_filemap_page_mkwrite(
1590 struct vm_area_struct *vma,
1591 struct vm_fault *vmf)
1593 struct inode *inode = file_inode(vma->vm_file);
1596 trace_xfs_filemap_page_mkwrite(XFS_I(inode));
1598 sb_start_pagefault(inode->i_sb);
1599 file_update_time(vma->vm_file);
1600 xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1602 if (IS_DAX(inode)) {
1603 ret = iomap_dax_fault(vma, vmf, &xfs_iomap_ops);
1605 ret = iomap_page_mkwrite(vma, vmf, &xfs_iomap_ops);
1606 ret = block_page_mkwrite_return(ret);
1609 xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1610 sb_end_pagefault(inode->i_sb);
1617 struct vm_area_struct *vma,
1618 struct vm_fault *vmf)
1620 struct inode *inode = file_inode(vma->vm_file);
1623 trace_xfs_filemap_fault(XFS_I(inode));
1625 /* DAX can shortcut the normal fault path on write faults! */
1626 if ((vmf->flags & FAULT_FLAG_WRITE) && IS_DAX(inode))
1627 return xfs_filemap_page_mkwrite(vma, vmf);
1629 xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1630 if (IS_DAX(inode)) {
1632 * we do not want to trigger unwritten extent conversion on read
1633 * faults - that is unnecessary overhead and would also require
1634 * changes to xfs_get_blocks_direct() to map unwritten extent
1635 * ioend for conversion on read-only mappings.
1637 ret = iomap_dax_fault(vma, vmf, &xfs_iomap_ops);
1639 ret = filemap_fault(vma, vmf);
1640 xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1646 * Similar to xfs_filemap_fault(), the DAX fault path can call into here on
1647 * both read and write faults. Hence we need to handle both cases. There is no
1648 * ->pmd_mkwrite callout for huge pages, so we have a single function here to
1649 * handle both cases here. @flags carries the information on the type of fault
1653 xfs_filemap_pmd_fault(
1654 struct vm_area_struct *vma,
1659 struct inode *inode = file_inode(vma->vm_file);
1660 struct xfs_inode *ip = XFS_I(inode);
1664 return VM_FAULT_FALLBACK;
1666 trace_xfs_filemap_pmd_fault(ip);
1668 if (flags & FAULT_FLAG_WRITE) {
1669 sb_start_pagefault(inode->i_sb);
1670 file_update_time(vma->vm_file);
1673 xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1674 ret = dax_pmd_fault(vma, addr, pmd, flags, xfs_get_blocks_dax_fault);
1675 xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1677 if (flags & FAULT_FLAG_WRITE)
1678 sb_end_pagefault(inode->i_sb);
1684 * pfn_mkwrite was originally inteneded to ensure we capture time stamp
1685 * updates on write faults. In reality, it's need to serialise against
1686 * truncate similar to page_mkwrite. Hence we cycle the XFS_MMAPLOCK_SHARED
1687 * to ensure we serialise the fault barrier in place.
1690 xfs_filemap_pfn_mkwrite(
1691 struct vm_area_struct *vma,
1692 struct vm_fault *vmf)
1695 struct inode *inode = file_inode(vma->vm_file);
1696 struct xfs_inode *ip = XFS_I(inode);
1697 int ret = VM_FAULT_NOPAGE;
1700 trace_xfs_filemap_pfn_mkwrite(ip);
1702 sb_start_pagefault(inode->i_sb);
1703 file_update_time(vma->vm_file);
1705 /* check if the faulting page hasn't raced with truncate */
1706 xfs_ilock(ip, XFS_MMAPLOCK_SHARED);
1707 size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT;
1708 if (vmf->pgoff >= size)
1709 ret = VM_FAULT_SIGBUS;
1710 else if (IS_DAX(inode))
1711 ret = dax_pfn_mkwrite(vma, vmf);
1712 xfs_iunlock(ip, XFS_MMAPLOCK_SHARED);
1713 sb_end_pagefault(inode->i_sb);
1718 static const struct vm_operations_struct xfs_file_vm_ops = {
1719 .fault = xfs_filemap_fault,
1720 .pmd_fault = xfs_filemap_pmd_fault,
1721 .map_pages = filemap_map_pages,
1722 .page_mkwrite = xfs_filemap_page_mkwrite,
1723 .pfn_mkwrite = xfs_filemap_pfn_mkwrite,
1729 struct vm_area_struct *vma)
1731 file_accessed(filp);
1732 vma->vm_ops = &xfs_file_vm_ops;
1733 if (IS_DAX(file_inode(filp)))
1734 vma->vm_flags |= VM_MIXEDMAP | VM_HUGEPAGE;
1738 const struct file_operations xfs_file_operations = {
1739 .llseek = xfs_file_llseek,
1740 .read_iter = xfs_file_read_iter,
1741 .write_iter = xfs_file_write_iter,
1742 .splice_read = generic_file_splice_read,
1743 .splice_write = iter_file_splice_write,
1744 .unlocked_ioctl = xfs_file_ioctl,
1745 #ifdef CONFIG_COMPAT
1746 .compat_ioctl = xfs_file_compat_ioctl,
1748 .mmap = xfs_file_mmap,
1749 .open = xfs_file_open,
1750 .release = xfs_file_release,
1751 .fsync = xfs_file_fsync,
1752 .get_unmapped_area = thp_get_unmapped_area,
1753 .fallocate = xfs_file_fallocate,
1754 .copy_file_range = xfs_file_copy_range,
1755 .clone_file_range = xfs_file_clone_range,
1756 .dedupe_file_range = xfs_file_dedupe_range,
1759 const struct file_operations xfs_dir_file_operations = {
1760 .open = xfs_dir_open,
1761 .read = generic_read_dir,
1762 .iterate_shared = xfs_file_readdir,
1763 .llseek = generic_file_llseek,
1764 .unlocked_ioctl = xfs_file_ioctl,
1765 #ifdef CONFIG_COMPAT
1766 .compat_ioctl = xfs_file_compat_ioctl,
1768 .fsync = xfs_dir_fsync,