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 * Clear the specified ranges to zero through either the pagecache or DAX.
52 * Holes and unwritten extents will be left as-is as they already are zeroed.
61 return iomap_zero_range(VFS_I(ip), pos, count, NULL, &xfs_iomap_ops);
65 xfs_update_prealloc_flags(
67 enum xfs_prealloc_flags flags)
72 error = xfs_trans_alloc(ip->i_mount, &M_RES(ip->i_mount)->tr_writeid,
77 xfs_ilock(ip, XFS_ILOCK_EXCL);
78 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
80 if (!(flags & XFS_PREALLOC_INVISIBLE)) {
81 VFS_I(ip)->i_mode &= ~S_ISUID;
82 if (VFS_I(ip)->i_mode & S_IXGRP)
83 VFS_I(ip)->i_mode &= ~S_ISGID;
84 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
87 if (flags & XFS_PREALLOC_SET)
88 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
89 if (flags & XFS_PREALLOC_CLEAR)
90 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
92 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
93 if (flags & XFS_PREALLOC_SYNC)
94 xfs_trans_set_sync(tp);
95 return xfs_trans_commit(tp);
99 * Fsync operations on directories are much simpler than on regular files,
100 * as there is no file data to flush, and thus also no need for explicit
101 * cache flush operations, and there are no non-transaction metadata updates
102 * on directories either.
111 struct xfs_inode *ip = XFS_I(file->f_mapping->host);
112 struct xfs_mount *mp = ip->i_mount;
115 trace_xfs_dir_fsync(ip);
117 xfs_ilock(ip, XFS_ILOCK_SHARED);
118 if (xfs_ipincount(ip))
119 lsn = ip->i_itemp->ili_last_lsn;
120 xfs_iunlock(ip, XFS_ILOCK_SHARED);
124 return _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, NULL);
134 struct inode *inode = file->f_mapping->host;
135 struct xfs_inode *ip = XFS_I(inode);
136 struct xfs_mount *mp = ip->i_mount;
141 trace_xfs_file_fsync(ip);
143 error = filemap_write_and_wait_range(inode->i_mapping, start, end);
147 if (XFS_FORCED_SHUTDOWN(mp))
150 xfs_iflags_clear(ip, XFS_ITRUNCATED);
153 * If we have an RT and/or log subvolume we need to make sure to flush
154 * the write cache the device used for file data first. This is to
155 * ensure newly written file data make it to disk before logging the new
156 * inode size in case of an extending write.
158 if (XFS_IS_REALTIME_INODE(ip))
159 xfs_blkdev_issue_flush(mp->m_rtdev_targp);
160 else if (mp->m_logdev_targp != mp->m_ddev_targp)
161 xfs_blkdev_issue_flush(mp->m_ddev_targp);
164 * All metadata updates are logged, which means that we just have to
165 * flush the log up to the latest LSN that touched the inode. If we have
166 * concurrent fsync/fdatasync() calls, we need them to all block on the
167 * log force before we clear the ili_fsync_fields field. This ensures
168 * that we don't get a racing sync operation that does not wait for the
169 * metadata to hit the journal before returning. If we race with
170 * clearing the ili_fsync_fields, then all that will happen is the log
171 * force will do nothing as the lsn will already be on disk. We can't
172 * race with setting ili_fsync_fields because that is done under
173 * XFS_ILOCK_EXCL, and that can't happen because we hold the lock shared
174 * until after the ili_fsync_fields is cleared.
176 xfs_ilock(ip, XFS_ILOCK_SHARED);
177 if (xfs_ipincount(ip)) {
179 (ip->i_itemp->ili_fsync_fields & ~XFS_ILOG_TIMESTAMP))
180 lsn = ip->i_itemp->ili_last_lsn;
184 error = _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, &log_flushed);
185 ip->i_itemp->ili_fsync_fields = 0;
187 xfs_iunlock(ip, XFS_ILOCK_SHARED);
190 * If we only have a single device, and the log force about was
191 * a no-op we might have to flush the data device cache here.
192 * This can only happen for fdatasync/O_DSYNC if we were overwriting
193 * an already allocated file and thus do not have any metadata to
196 if (!log_flushed && !XFS_IS_REALTIME_INODE(ip) &&
197 mp->m_logdev_targp == mp->m_ddev_targp)
198 xfs_blkdev_issue_flush(mp->m_ddev_targp);
204 xfs_file_dio_aio_read(
208 struct xfs_inode *ip = XFS_I(file_inode(iocb->ki_filp));
209 size_t count = iov_iter_count(to);
212 trace_xfs_file_direct_read(ip, count, iocb->ki_pos);
215 return 0; /* skip atime */
217 file_accessed(iocb->ki_filp);
219 xfs_ilock(ip, XFS_IOLOCK_SHARED);
220 ret = iomap_dio_rw(iocb, to, &xfs_iomap_ops, NULL);
221 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
226 static noinline ssize_t
231 struct xfs_inode *ip = XFS_I(iocb->ki_filp->f_mapping->host);
232 size_t count = iov_iter_count(to);
235 trace_xfs_file_dax_read(ip, count, iocb->ki_pos);
238 return 0; /* skip atime */
240 xfs_ilock(ip, XFS_IOLOCK_SHARED);
241 ret = dax_iomap_rw(iocb, to, &xfs_iomap_ops);
242 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
244 file_accessed(iocb->ki_filp);
249 xfs_file_buffered_aio_read(
253 struct xfs_inode *ip = XFS_I(file_inode(iocb->ki_filp));
256 trace_xfs_file_buffered_read(ip, iov_iter_count(to), iocb->ki_pos);
258 xfs_ilock(ip, XFS_IOLOCK_SHARED);
259 ret = generic_file_read_iter(iocb, to);
260 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
270 struct inode *inode = file_inode(iocb->ki_filp);
271 struct xfs_mount *mp = XFS_I(inode)->i_mount;
274 XFS_STATS_INC(mp, xs_read_calls);
276 if (XFS_FORCED_SHUTDOWN(mp))
280 ret = xfs_file_dax_read(iocb, to);
281 else if (iocb->ki_flags & IOCB_DIRECT)
282 ret = xfs_file_dio_aio_read(iocb, to);
284 ret = xfs_file_buffered_aio_read(iocb, to);
287 XFS_STATS_ADD(mp, xs_read_bytes, ret);
292 * Zero any on disk space between the current EOF and the new, larger EOF.
294 * This handles the normal case of zeroing the remainder of the last block in
295 * the file and the unusual case of zeroing blocks out beyond the size of the
296 * file. This second case only happens with fixed size extents and when the
297 * system crashes before the inode size was updated but after blocks were
300 * Expects the iolock to be held exclusive, and will take the ilock internally.
302 int /* error (positive) */
304 struct xfs_inode *ip,
305 xfs_off_t offset, /* starting I/O offset */
306 xfs_fsize_t isize, /* current inode size */
309 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
310 ASSERT(offset > isize);
312 trace_xfs_zero_eof(ip, isize, offset - isize);
313 return xfs_zero_range(ip, isize, offset - isize, did_zeroing);
317 * Common pre-write limit and setup checks.
319 * Called with the iolocked held either shared and exclusive according to
320 * @iolock, and returns with it held. Might upgrade the iolock to exclusive
321 * if called for a direct write beyond i_size.
324 xfs_file_aio_write_checks(
326 struct iov_iter *from,
329 struct file *file = iocb->ki_filp;
330 struct inode *inode = file->f_mapping->host;
331 struct xfs_inode *ip = XFS_I(inode);
333 size_t count = iov_iter_count(from);
334 bool drained_dio = false;
337 error = generic_write_checks(iocb, from);
341 error = xfs_break_layouts(inode, iolock);
346 * For changing security info in file_remove_privs() we need i_rwsem
349 if (*iolock == XFS_IOLOCK_SHARED && !IS_NOSEC(inode)) {
350 xfs_iunlock(ip, *iolock);
351 *iolock = XFS_IOLOCK_EXCL;
352 xfs_ilock(ip, *iolock);
356 * If the offset is beyond the size of the file, we need to zero any
357 * blocks that fall between the existing EOF and the start of this
358 * write. If zeroing is needed and we are currently holding the
359 * iolock shared, we need to update it to exclusive which implies
360 * having to redo all checks before.
362 * We need to serialise against EOF updates that occur in IO
363 * completions here. We want to make sure that nobody is changing the
364 * size while we do this check until we have placed an IO barrier (i.e.
365 * hold the XFS_IOLOCK_EXCL) that prevents new IO from being dispatched.
366 * The spinlock effectively forms a memory barrier once we have the
367 * XFS_IOLOCK_EXCL so we are guaranteed to see the latest EOF value
368 * and hence be able to correctly determine if we need to run zeroing.
370 spin_lock(&ip->i_flags_lock);
371 if (iocb->ki_pos > i_size_read(inode)) {
374 spin_unlock(&ip->i_flags_lock);
376 if (*iolock == XFS_IOLOCK_SHARED) {
377 xfs_iunlock(ip, *iolock);
378 *iolock = XFS_IOLOCK_EXCL;
379 xfs_ilock(ip, *iolock);
380 iov_iter_reexpand(from, count);
383 * We now have an IO submission barrier in place, but
384 * AIO can do EOF updates during IO completion and hence
385 * we now need to wait for all of them to drain. Non-AIO
386 * DIO will have drained before we are given the
387 * XFS_IOLOCK_EXCL, and so for most cases this wait is a
390 inode_dio_wait(inode);
394 error = xfs_zero_eof(ip, iocb->ki_pos, i_size_read(inode), &zero);
398 spin_unlock(&ip->i_flags_lock);
401 * Updating the timestamps will grab the ilock again from
402 * xfs_fs_dirty_inode, so we have to call it after dropping the
403 * lock above. Eventually we should look into a way to avoid
404 * the pointless lock roundtrip.
406 if (likely(!(file->f_mode & FMODE_NOCMTIME))) {
407 error = file_update_time(file);
413 * If we're writing the file then make sure to clear the setuid and
414 * setgid bits if the process is not being run by root. This keeps
415 * people from modifying setuid and setgid binaries.
417 if (!IS_NOSEC(inode))
418 return file_remove_privs(file);
423 xfs_dio_write_end_io(
428 struct inode *inode = file_inode(iocb->ki_filp);
429 struct xfs_inode *ip = XFS_I(inode);
430 loff_t offset = iocb->ki_pos;
431 bool update_size = false;
434 trace_xfs_end_io_direct_write(ip, offset, size);
436 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
443 * We need to update the in-core inode size here so that we don't end up
444 * with the on-disk inode size being outside the in-core inode size. We
445 * have no other method of updating EOF for AIO, so always do it here
448 * We need to lock the test/set EOF update as we can be racing with
449 * other IO completions here to update the EOF. Failing to serialise
450 * here can result in EOF moving backwards and Bad Things Happen when
453 spin_lock(&ip->i_flags_lock);
454 if (offset + size > i_size_read(inode)) {
455 i_size_write(inode, offset + size);
458 spin_unlock(&ip->i_flags_lock);
460 if (flags & IOMAP_DIO_COW) {
461 error = xfs_reflink_end_cow(ip, offset, size);
466 if (flags & IOMAP_DIO_UNWRITTEN)
467 error = xfs_iomap_write_unwritten(ip, offset, size);
468 else if (update_size)
469 error = xfs_setfilesize(ip, offset, size);
475 * xfs_file_dio_aio_write - handle direct IO writes
477 * Lock the inode appropriately to prepare for and issue a direct IO write.
478 * By separating it from the buffered write path we remove all the tricky to
479 * follow locking changes and looping.
481 * If there are cached pages or we're extending the file, we need IOLOCK_EXCL
482 * until we're sure the bytes at the new EOF have been zeroed and/or the cached
483 * pages are flushed out.
485 * In most cases the direct IO writes will be done holding IOLOCK_SHARED
486 * allowing them to be done in parallel with reads and other direct IO writes.
487 * However, if the IO is not aligned to filesystem blocks, the direct IO layer
488 * needs to do sub-block zeroing and that requires serialisation against other
489 * direct IOs to the same block. In this case we need to serialise the
490 * submission of the unaligned IOs so that we don't get racing block zeroing in
491 * the dio layer. To avoid the problem with aio, we also need to wait for
492 * outstanding IOs to complete so that unwritten extent conversion is completed
493 * before we try to map the overlapping block. This is currently implemented by
494 * hitting it with a big hammer (i.e. inode_dio_wait()).
496 * Returns with locks held indicated by @iolock and errors indicated by
497 * negative return values.
500 xfs_file_dio_aio_write(
502 struct iov_iter *from)
504 struct file *file = iocb->ki_filp;
505 struct address_space *mapping = file->f_mapping;
506 struct inode *inode = mapping->host;
507 struct xfs_inode *ip = XFS_I(inode);
508 struct xfs_mount *mp = ip->i_mount;
510 int unaligned_io = 0;
512 size_t count = iov_iter_count(from);
513 struct xfs_buftarg *target = XFS_IS_REALTIME_INODE(ip) ?
514 mp->m_rtdev_targp : mp->m_ddev_targp;
516 /* DIO must be aligned to device logical sector size */
517 if ((iocb->ki_pos | count) & target->bt_logical_sectormask)
521 * Don't take the exclusive iolock here unless the I/O is unaligned to
522 * the file system block size. We don't need to consider the EOF
523 * extension case here because xfs_file_aio_write_checks() will relock
524 * the inode as necessary for EOF zeroing cases and fill out the new
525 * inode size as appropriate.
527 if ((iocb->ki_pos & mp->m_blockmask) ||
528 ((iocb->ki_pos + count) & mp->m_blockmask)) {
530 iolock = XFS_IOLOCK_EXCL;
532 iolock = XFS_IOLOCK_SHARED;
535 xfs_ilock(ip, iolock);
537 ret = xfs_file_aio_write_checks(iocb, from, &iolock);
540 count = iov_iter_count(from);
543 * If we are doing unaligned IO, wait for all other IO to drain,
544 * otherwise demote the lock if we had to take the exclusive lock
545 * for other reasons in xfs_file_aio_write_checks.
548 inode_dio_wait(inode);
549 else if (iolock == XFS_IOLOCK_EXCL) {
550 xfs_ilock_demote(ip, XFS_IOLOCK_EXCL);
551 iolock = XFS_IOLOCK_SHARED;
554 trace_xfs_file_direct_write(ip, count, iocb->ki_pos);
556 /* If this is a block-aligned directio CoW, remap immediately. */
557 if (xfs_is_reflink_inode(ip) && !unaligned_io) {
558 ret = xfs_reflink_allocate_cow_range(ip, iocb->ki_pos, count);
563 ret = iomap_dio_rw(iocb, from, &xfs_iomap_ops, xfs_dio_write_end_io);
565 xfs_iunlock(ip, iolock);
568 * No fallback to buffered IO on errors for XFS, direct IO will either
569 * complete fully or fail.
571 ASSERT(ret < 0 || ret == count);
575 static noinline ssize_t
578 struct iov_iter *from)
580 struct inode *inode = iocb->ki_filp->f_mapping->host;
581 struct xfs_inode *ip = XFS_I(inode);
582 int iolock = XFS_IOLOCK_EXCL;
583 ssize_t ret, error = 0;
587 xfs_ilock(ip, iolock);
588 ret = xfs_file_aio_write_checks(iocb, from, &iolock);
593 count = iov_iter_count(from);
595 trace_xfs_file_dax_write(ip, count, pos);
596 ret = dax_iomap_rw(iocb, from, &xfs_iomap_ops);
597 if (ret > 0 && iocb->ki_pos > i_size_read(inode)) {
598 i_size_write(inode, iocb->ki_pos);
599 error = xfs_setfilesize(ip, pos, ret);
602 xfs_iunlock(ip, iolock);
603 return error ? error : ret;
607 xfs_file_buffered_aio_write(
609 struct iov_iter *from)
611 struct file *file = iocb->ki_filp;
612 struct address_space *mapping = file->f_mapping;
613 struct inode *inode = mapping->host;
614 struct xfs_inode *ip = XFS_I(inode);
617 int iolock = XFS_IOLOCK_EXCL;
619 xfs_ilock(ip, iolock);
621 ret = xfs_file_aio_write_checks(iocb, from, &iolock);
625 /* We can write back this queue in page reclaim */
626 current->backing_dev_info = inode_to_bdi(inode);
629 trace_xfs_file_buffered_write(ip, iov_iter_count(from), iocb->ki_pos);
630 ret = iomap_file_buffered_write(iocb, from, &xfs_iomap_ops);
631 if (likely(ret >= 0))
635 * If we hit a space limit, try to free up some lingering preallocated
636 * space before returning an error. In the case of ENOSPC, first try to
637 * write back all dirty inodes to free up some of the excess reserved
638 * metadata space. This reduces the chances that the eofblocks scan
639 * waits on dirty mappings. Since xfs_flush_inodes() is serialized, this
640 * also behaves as a filter to prevent too many eofblocks scans from
641 * running at the same time.
643 if (ret == -EDQUOT && !enospc) {
644 enospc = xfs_inode_free_quota_eofblocks(ip);
647 enospc = xfs_inode_free_quota_cowblocks(ip);
650 } else if (ret == -ENOSPC && !enospc) {
651 struct xfs_eofblocks eofb = {0};
654 xfs_flush_inodes(ip->i_mount);
655 eofb.eof_scan_owner = ip->i_ino; /* for locking */
656 eofb.eof_flags = XFS_EOF_FLAGS_SYNC;
657 xfs_icache_free_eofblocks(ip->i_mount, &eofb);
661 current->backing_dev_info = NULL;
663 xfs_iunlock(ip, iolock);
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);
677 size_t ocount = iov_iter_count(from);
679 XFS_STATS_INC(ip->i_mount, xs_write_calls);
684 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
688 ret = xfs_file_dax_write(iocb, from);
689 else if (iocb->ki_flags & IOCB_DIRECT) {
691 * Allow a directio write to fall back to a buffered
692 * write *only* in the case that we're doing a reflink
693 * CoW. In all other directio scenarios we do not
694 * allow an operation to fall back to buffered mode.
696 ret = xfs_file_dio_aio_write(iocb, from);
701 ret = xfs_file_buffered_aio_write(iocb, from);
705 XFS_STATS_ADD(ip->i_mount, xs_write_bytes, ret);
707 /* Handle various SYNC-type writes */
708 ret = generic_write_sync(iocb, ret);
713 #define XFS_FALLOC_FL_SUPPORTED \
714 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
715 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE | \
716 FALLOC_FL_INSERT_RANGE | FALLOC_FL_UNSHARE_RANGE)
725 struct inode *inode = file_inode(file);
726 struct xfs_inode *ip = XFS_I(inode);
728 enum xfs_prealloc_flags flags = 0;
729 uint iolock = XFS_IOLOCK_EXCL;
731 bool do_file_insert = 0;
733 if (!S_ISREG(inode->i_mode))
735 if (mode & ~XFS_FALLOC_FL_SUPPORTED)
738 xfs_ilock(ip, iolock);
739 error = xfs_break_layouts(inode, &iolock);
743 xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
744 iolock |= XFS_MMAPLOCK_EXCL;
746 if (mode & FALLOC_FL_PUNCH_HOLE) {
747 error = xfs_free_file_space(ip, offset, len);
750 } else if (mode & FALLOC_FL_COLLAPSE_RANGE) {
751 unsigned blksize_mask = (1 << inode->i_blkbits) - 1;
753 if (offset & blksize_mask || len & blksize_mask) {
759 * There is no need to overlap collapse range with EOF,
760 * in which case it is effectively a truncate operation
762 if (offset + len >= i_size_read(inode)) {
767 new_size = i_size_read(inode) - len;
769 error = xfs_collapse_file_space(ip, offset, len);
772 } else if (mode & FALLOC_FL_INSERT_RANGE) {
773 unsigned blksize_mask = (1 << inode->i_blkbits) - 1;
775 new_size = i_size_read(inode) + len;
776 if (offset & blksize_mask || len & blksize_mask) {
781 /* check the new inode size does not wrap through zero */
782 if (new_size > inode->i_sb->s_maxbytes) {
787 /* Offset should be less than i_size */
788 if (offset >= i_size_read(inode)) {
794 flags |= XFS_PREALLOC_SET;
796 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
797 offset + len > i_size_read(inode)) {
798 new_size = offset + len;
799 error = inode_newsize_ok(inode, new_size);
804 if (mode & FALLOC_FL_ZERO_RANGE)
805 error = xfs_zero_file_space(ip, offset, len);
807 if (mode & FALLOC_FL_UNSHARE_RANGE) {
808 error = xfs_reflink_unshare(ip, offset, len);
812 error = xfs_alloc_file_space(ip, offset, len,
819 if (file->f_flags & O_DSYNC)
820 flags |= XFS_PREALLOC_SYNC;
822 error = xfs_update_prealloc_flags(ip, flags);
826 /* Change file size if needed */
830 iattr.ia_valid = ATTR_SIZE;
831 iattr.ia_size = new_size;
832 error = xfs_vn_setattr_size(file_dentry(file), &iattr);
838 * Perform hole insertion now that the file size has been
839 * updated so that if we crash during the operation we don't
840 * leave shifted extents past EOF and hence losing access to
841 * the data that is contained within them.
844 error = xfs_insert_file_space(ip, offset, len);
847 xfs_iunlock(ip, iolock);
853 struct file *file_in,
855 struct file *file_out,
862 error = xfs_reflink_remap_range(file_in, pos_in, file_out, pos_out,
870 xfs_file_clone_range(
871 struct file *file_in,
873 struct file *file_out,
877 return xfs_reflink_remap_range(file_in, pos_in, file_out, pos_out,
882 xfs_file_dedupe_range(
883 struct file *src_file,
886 struct file *dst_file,
891 error = xfs_reflink_remap_range(src_file, loff, dst_file, dst_loff,
903 if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
905 if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb)))
915 struct xfs_inode *ip = XFS_I(inode);
919 error = xfs_file_open(inode, file);
924 * If there are any blocks, read-ahead block 0 as we're almost
925 * certain to have the next operation be a read there.
927 mode = xfs_ilock_data_map_shared(ip);
928 if (ip->i_d.di_nextents > 0)
929 xfs_dir3_data_readahead(ip, 0, -1);
930 xfs_iunlock(ip, mode);
939 return xfs_release(XFS_I(inode));
945 struct dir_context *ctx)
947 struct inode *inode = file_inode(file);
948 xfs_inode_t *ip = XFS_I(inode);
952 * The Linux API doesn't pass down the total size of the buffer
953 * we read into down to the filesystem. With the filldir concept
954 * it's not needed for correct information, but the XFS dir2 leaf
955 * code wants an estimate of the buffer size to calculate it's
956 * readahead window and size the buffers used for mapping to
959 * Try to give it an estimate that's good enough, maybe at some
960 * point we can change the ->readdir prototype to include the
961 * buffer size. For now we use the current glibc buffer size.
963 bufsize = (size_t)min_t(loff_t, 32768, ip->i_d.di_size);
965 return xfs_readdir(ip, ctx, bufsize);
969 * This type is designed to indicate the type of offset we would like
970 * to search from page cache for xfs_seek_hole_data().
978 * Lookup the desired type of offset from the given page.
980 * On success, return true and the offset argument will point to the
981 * start of the region that was found. Otherwise this function will
982 * return false and keep the offset argument unchanged.
985 xfs_lookup_buffer_offset(
990 loff_t lastoff = page_offset(page);
992 struct buffer_head *bh, *head;
994 bh = head = page_buffers(page);
997 * Unwritten extents that have data in the page
998 * cache covering them can be identified by the
999 * BH_Unwritten state flag. Pages with multiple
1000 * buffers might have a mix of holes, data and
1001 * unwritten extents - any buffer with valid
1002 * data in it should have BH_Uptodate flag set
1005 if (buffer_unwritten(bh) ||
1006 buffer_uptodate(bh)) {
1007 if (type == DATA_OFF)
1010 if (type == HOLE_OFF)
1018 lastoff += bh->b_size;
1019 } while ((bh = bh->b_this_page) != head);
1025 * This routine is called to find out and return a data or hole offset
1026 * from the page cache for unwritten extents according to the desired
1027 * type for xfs_seek_hole_data().
1029 * The argument offset is used to tell where we start to search from the
1030 * page cache. Map is used to figure out the end points of the range to
1033 * Return true if the desired type of offset was found, and the argument
1034 * offset is filled with that address. Otherwise, return false and keep
1038 xfs_find_get_desired_pgoff(
1039 struct inode *inode,
1040 struct xfs_bmbt_irec *map,
1044 struct xfs_inode *ip = XFS_I(inode);
1045 struct xfs_mount *mp = ip->i_mount;
1046 struct pagevec pvec;
1050 loff_t startoff = *offset;
1051 loff_t lastoff = startoff;
1054 pagevec_init(&pvec, 0);
1056 index = startoff >> PAGE_SHIFT;
1057 endoff = XFS_FSB_TO_B(mp, map->br_startoff + map->br_blockcount);
1058 end = endoff >> PAGE_SHIFT;
1064 want = min_t(pgoff_t, end - index, PAGEVEC_SIZE);
1065 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index,
1068 * No page mapped into given range. If we are searching holes
1069 * and if this is the first time we got into the loop, it means
1070 * that the given offset is landed in a hole, return it.
1072 * If we have already stepped through some block buffers to find
1073 * holes but they all contains data. In this case, the last
1074 * offset is already updated and pointed to the end of the last
1075 * mapped page, if it does not reach the endpoint to search,
1076 * that means there should be a hole between them.
1078 if (nr_pages == 0) {
1079 /* Data search found nothing */
1080 if (type == DATA_OFF)
1083 ASSERT(type == HOLE_OFF);
1084 if (lastoff == startoff || lastoff < endoff) {
1092 * At lease we found one page. If this is the first time we
1093 * step into the loop, and if the first page index offset is
1094 * greater than the given search offset, a hole was found.
1096 if (type == HOLE_OFF && lastoff == startoff &&
1097 lastoff < page_offset(pvec.pages[0])) {
1102 for (i = 0; i < nr_pages; i++) {
1103 struct page *page = pvec.pages[i];
1107 * At this point, the page may be truncated or
1108 * invalidated (changing page->mapping to NULL),
1109 * or even swizzled back from swapper_space to tmpfs
1110 * file mapping. However, page->index will not change
1111 * because we have a reference on the page.
1113 * Searching done if the page index is out of range.
1114 * If the current offset is not reaches the end of
1115 * the specified search range, there should be a hole
1118 if (page->index > end) {
1119 if (type == HOLE_OFF && lastoff < endoff) {
1128 * Page truncated or invalidated(page->mapping == NULL).
1129 * We can freely skip it and proceed to check the next
1132 if (unlikely(page->mapping != inode->i_mapping)) {
1137 if (!page_has_buffers(page)) {
1142 found = xfs_lookup_buffer_offset(page, &b_offset, type);
1145 * The found offset may be less than the start
1146 * point to search if this is the first time to
1149 *offset = max_t(loff_t, startoff, b_offset);
1155 * We either searching data but nothing was found, or
1156 * searching hole but found a data buffer. In either
1157 * case, probably the next page contains the desired
1158 * things, update the last offset to it so.
1160 lastoff = page_offset(page) + PAGE_SIZE;
1165 * The number of returned pages less than our desired, search
1166 * done. In this case, nothing was found for searching data,
1167 * but we found a hole behind the last offset.
1169 if (nr_pages < want) {
1170 if (type == HOLE_OFF) {
1177 index = pvec.pages[i - 1]->index + 1;
1178 pagevec_release(&pvec);
1179 } while (index <= end);
1182 pagevec_release(&pvec);
1187 * caller must lock inode with xfs_ilock_data_map_shared,
1188 * can we craft an appropriate ASSERT?
1190 * end is because the VFS-level lseek interface is defined such that any
1191 * offset past i_size shall return -ENXIO, but we use this for quota code
1192 * which does not maintain i_size, and we want to SEEK_DATA past i_size.
1195 __xfs_seek_hole_data(
1196 struct inode *inode,
1201 struct xfs_inode *ip = XFS_I(inode);
1202 struct xfs_mount *mp = ip->i_mount;
1203 loff_t uninitialized_var(offset);
1204 xfs_fileoff_t fsbno;
1205 xfs_filblks_t lastbno;
1214 * Try to read extents from the first block indicated
1215 * by fsbno to the end block of the file.
1217 fsbno = XFS_B_TO_FSBT(mp, start);
1218 lastbno = XFS_B_TO_FSB(mp, end);
1221 struct xfs_bmbt_irec map[2];
1225 error = xfs_bmapi_read(ip, fsbno, lastbno - fsbno, map, &nmap,
1230 /* No extents at given offset, must be beyond EOF */
1236 for (i = 0; i < nmap; i++) {
1237 offset = max_t(loff_t, start,
1238 XFS_FSB_TO_B(mp, map[i].br_startoff));
1240 /* Landed in the hole we wanted? */
1241 if (whence == SEEK_HOLE &&
1242 map[i].br_startblock == HOLESTARTBLOCK)
1245 /* Landed in the data extent we wanted? */
1246 if (whence == SEEK_DATA &&
1247 (map[i].br_startblock == DELAYSTARTBLOCK ||
1248 (map[i].br_state == XFS_EXT_NORM &&
1249 !isnullstartblock(map[i].br_startblock))))
1253 * Landed in an unwritten extent, try to search
1254 * for hole or data from page cache.
1256 if (map[i].br_state == XFS_EXT_UNWRITTEN) {
1257 if (xfs_find_get_desired_pgoff(inode, &map[i],
1258 whence == SEEK_HOLE ? HOLE_OFF : DATA_OFF,
1265 * We only received one extent out of the two requested. This
1266 * means we've hit EOF and didn't find what we are looking for.
1270 * If we were looking for a hole, set offset to
1271 * the end of the file (i.e., there is an implicit
1272 * hole at the end of any file).
1274 if (whence == SEEK_HOLE) {
1279 * If we were looking for data, it's nowhere to be found
1281 ASSERT(whence == SEEK_DATA);
1289 * Nothing was found, proceed to the next round of search
1290 * if the next reading offset is not at or beyond EOF.
1292 fsbno = map[i - 1].br_startoff + map[i - 1].br_blockcount;
1293 start = XFS_FSB_TO_B(mp, fsbno);
1295 if (whence == SEEK_HOLE) {
1299 ASSERT(whence == SEEK_DATA);
1307 * If at this point we have found the hole we wanted, the returned
1308 * offset may be bigger than the file size as it may be aligned to
1309 * page boundary for unwritten extents. We need to deal with this
1310 * situation in particular.
1312 if (whence == SEEK_HOLE)
1313 offset = min_t(loff_t, offset, end);
1327 struct inode *inode = file->f_mapping->host;
1328 struct xfs_inode *ip = XFS_I(inode);
1329 struct xfs_mount *mp = ip->i_mount;
1334 if (XFS_FORCED_SHUTDOWN(mp))
1337 lock = xfs_ilock_data_map_shared(ip);
1339 end = i_size_read(inode);
1340 offset = __xfs_seek_hole_data(inode, start, end, whence);
1346 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
1349 xfs_iunlock(ip, lock);
1366 return generic_file_llseek(file, offset, whence);
1369 return xfs_seek_hole_data(file, offset, whence);
1376 * Locking for serialisation of IO during page faults. This results in a lock
1380 * sb_start_pagefault(vfs, freeze)
1381 * i_mmaplock (XFS - truncate serialisation)
1383 * i_lock (XFS - extent map serialisation)
1387 * mmap()d file has taken write protection fault and is being made writable. We
1388 * can set the page state up correctly for a writable page, which means we can
1389 * do correct delalloc accounting (ENOSPC checking!) and unwritten extent
1393 xfs_filemap_page_mkwrite(
1394 struct vm_area_struct *vma,
1395 struct vm_fault *vmf)
1397 struct inode *inode = file_inode(vma->vm_file);
1400 trace_xfs_filemap_page_mkwrite(XFS_I(inode));
1402 sb_start_pagefault(inode->i_sb);
1403 file_update_time(vma->vm_file);
1404 xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1406 if (IS_DAX(inode)) {
1407 ret = dax_iomap_fault(vma, vmf, &xfs_iomap_ops);
1409 ret = iomap_page_mkwrite(vma, vmf, &xfs_iomap_ops);
1410 ret = block_page_mkwrite_return(ret);
1413 xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1414 sb_end_pagefault(inode->i_sb);
1421 struct vm_area_struct *vma,
1422 struct vm_fault *vmf)
1424 struct inode *inode = file_inode(vma->vm_file);
1427 trace_xfs_filemap_fault(XFS_I(inode));
1429 /* DAX can shortcut the normal fault path on write faults! */
1430 if ((vmf->flags & FAULT_FLAG_WRITE) && IS_DAX(inode))
1431 return xfs_filemap_page_mkwrite(vma, vmf);
1433 xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1435 ret = dax_iomap_fault(vma, vmf, &xfs_iomap_ops);
1437 ret = filemap_fault(vma, vmf);
1438 xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1444 * Similar to xfs_filemap_fault(), the DAX fault path can call into here on
1445 * both read and write faults. Hence we need to handle both cases. There is no
1446 * ->pmd_mkwrite callout for huge pages, so we have a single function here to
1447 * handle both cases here. @flags carries the information on the type of fault
1451 xfs_filemap_pmd_fault(
1452 struct vm_area_struct *vma,
1457 struct inode *inode = file_inode(vma->vm_file);
1458 struct xfs_inode *ip = XFS_I(inode);
1462 return VM_FAULT_FALLBACK;
1464 trace_xfs_filemap_pmd_fault(ip);
1466 if (flags & FAULT_FLAG_WRITE) {
1467 sb_start_pagefault(inode->i_sb);
1468 file_update_time(vma->vm_file);
1471 xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1472 ret = dax_iomap_pmd_fault(vma, addr, pmd, flags, &xfs_iomap_ops);
1473 xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1475 if (flags & FAULT_FLAG_WRITE)
1476 sb_end_pagefault(inode->i_sb);
1482 * pfn_mkwrite was originally inteneded to ensure we capture time stamp
1483 * updates on write faults. In reality, it's need to serialise against
1484 * truncate similar to page_mkwrite. Hence we cycle the XFS_MMAPLOCK_SHARED
1485 * to ensure we serialise the fault barrier in place.
1488 xfs_filemap_pfn_mkwrite(
1489 struct vm_area_struct *vma,
1490 struct vm_fault *vmf)
1493 struct inode *inode = file_inode(vma->vm_file);
1494 struct xfs_inode *ip = XFS_I(inode);
1495 int ret = VM_FAULT_NOPAGE;
1498 trace_xfs_filemap_pfn_mkwrite(ip);
1500 sb_start_pagefault(inode->i_sb);
1501 file_update_time(vma->vm_file);
1503 /* check if the faulting page hasn't raced with truncate */
1504 xfs_ilock(ip, XFS_MMAPLOCK_SHARED);
1505 size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT;
1506 if (vmf->pgoff >= size)
1507 ret = VM_FAULT_SIGBUS;
1508 else if (IS_DAX(inode))
1509 ret = dax_pfn_mkwrite(vma, vmf);
1510 xfs_iunlock(ip, XFS_MMAPLOCK_SHARED);
1511 sb_end_pagefault(inode->i_sb);
1516 static const struct vm_operations_struct xfs_file_vm_ops = {
1517 .fault = xfs_filemap_fault,
1518 .pmd_fault = xfs_filemap_pmd_fault,
1519 .map_pages = filemap_map_pages,
1520 .page_mkwrite = xfs_filemap_page_mkwrite,
1521 .pfn_mkwrite = xfs_filemap_pfn_mkwrite,
1527 struct vm_area_struct *vma)
1529 file_accessed(filp);
1530 vma->vm_ops = &xfs_file_vm_ops;
1531 if (IS_DAX(file_inode(filp)))
1532 vma->vm_flags |= VM_MIXEDMAP | VM_HUGEPAGE;
1536 const struct file_operations xfs_file_operations = {
1537 .llseek = xfs_file_llseek,
1538 .read_iter = xfs_file_read_iter,
1539 .write_iter = xfs_file_write_iter,
1540 .splice_read = generic_file_splice_read,
1541 .splice_write = iter_file_splice_write,
1542 .unlocked_ioctl = xfs_file_ioctl,
1543 #ifdef CONFIG_COMPAT
1544 .compat_ioctl = xfs_file_compat_ioctl,
1546 .mmap = xfs_file_mmap,
1547 .open = xfs_file_open,
1548 .release = xfs_file_release,
1549 .fsync = xfs_file_fsync,
1550 .get_unmapped_area = thp_get_unmapped_area,
1551 .fallocate = xfs_file_fallocate,
1552 .copy_file_range = xfs_file_copy_range,
1553 .clone_file_range = xfs_file_clone_range,
1554 .dedupe_file_range = xfs_file_dedupe_range,
1557 const struct file_operations xfs_dir_file_operations = {
1558 .open = xfs_dir_open,
1559 .read = generic_read_dir,
1560 .iterate_shared = xfs_file_readdir,
1561 .llseek = generic_file_llseek,
1562 .unlocked_ioctl = xfs_file_ioctl,
1563 #ifdef CONFIG_COMPAT
1564 .compat_ioctl = xfs_file_compat_ioctl,
1566 .fsync = xfs_dir_fsync,