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"
41 #include <linux/dcache.h>
42 #include <linux/falloc.h>
43 #include <linux/pagevec.h>
45 static const struct vm_operations_struct xfs_file_vm_ops;
48 * Locking primitives for read and write IO paths to ensure we consistently use
49 * and order the inode->i_mutex, ip->i_lock and ip->i_iolock.
56 if (type & XFS_IOLOCK_EXCL)
57 mutex_lock(&VFS_I(ip)->i_mutex);
66 xfs_iunlock(ip, type);
67 if (type & XFS_IOLOCK_EXCL)
68 mutex_unlock(&VFS_I(ip)->i_mutex);
76 xfs_ilock_demote(ip, type);
77 if (type & XFS_IOLOCK_EXCL)
78 mutex_unlock(&VFS_I(ip)->i_mutex);
82 * xfs_iozero clears the specified range supplied via the page cache (except in
83 * the DAX case). Writes through the page cache will allocate blocks over holes,
84 * though the callers usually map the holes first and avoid them. If a block is
85 * not completely zeroed, then it will be read from disk before being partially
88 * In the DAX case, we can just directly write to the underlying pages. This
89 * will not allocate blocks, but will avoid holes and unwritten extents and so
90 * not do unnecessary work.
94 struct xfs_inode *ip, /* inode */
95 loff_t pos, /* offset in file */
96 size_t count) /* size of data to zero */
99 struct address_space *mapping;
103 mapping = VFS_I(ip)->i_mapping;
105 unsigned offset, bytes;
108 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
109 bytes = PAGE_CACHE_SIZE - offset;
113 if (IS_DAX(VFS_I(ip))) {
114 status = dax_zero_page_range(VFS_I(ip), pos, bytes,
115 xfs_get_blocks_direct);
119 status = pagecache_write_begin(NULL, mapping, pos, bytes,
120 AOP_FLAG_UNINTERRUPTIBLE,
125 zero_user(page, offset, bytes);
127 status = pagecache_write_end(NULL, mapping, pos, bytes,
128 bytes, page, fsdata);
129 WARN_ON(status <= 0); /* can't return less than zero! */
140 xfs_update_prealloc_flags(
141 struct xfs_inode *ip,
142 enum xfs_prealloc_flags flags)
144 struct xfs_trans *tp;
147 tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_WRITEID);
148 error = xfs_trans_reserve(tp, &M_RES(ip->i_mount)->tr_writeid, 0, 0);
150 xfs_trans_cancel(tp, 0);
154 xfs_ilock(ip, XFS_ILOCK_EXCL);
155 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
157 if (!(flags & XFS_PREALLOC_INVISIBLE)) {
158 ip->i_d.di_mode &= ~S_ISUID;
159 if (ip->i_d.di_mode & S_IXGRP)
160 ip->i_d.di_mode &= ~S_ISGID;
161 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
164 if (flags & XFS_PREALLOC_SET)
165 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
166 if (flags & XFS_PREALLOC_CLEAR)
167 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
169 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
170 if (flags & XFS_PREALLOC_SYNC)
171 xfs_trans_set_sync(tp);
172 return xfs_trans_commit(tp, 0);
176 * Fsync operations on directories are much simpler than on regular files,
177 * as there is no file data to flush, and thus also no need for explicit
178 * cache flush operations, and there are no non-transaction metadata updates
179 * on directories either.
188 struct xfs_inode *ip = XFS_I(file->f_mapping->host);
189 struct xfs_mount *mp = ip->i_mount;
192 trace_xfs_dir_fsync(ip);
194 xfs_ilock(ip, XFS_ILOCK_SHARED);
195 if (xfs_ipincount(ip))
196 lsn = ip->i_itemp->ili_last_lsn;
197 xfs_iunlock(ip, XFS_ILOCK_SHARED);
201 return _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, NULL);
211 struct inode *inode = file->f_mapping->host;
212 struct xfs_inode *ip = XFS_I(inode);
213 struct xfs_mount *mp = ip->i_mount;
218 trace_xfs_file_fsync(ip);
220 error = filemap_write_and_wait_range(inode->i_mapping, start, end);
224 if (XFS_FORCED_SHUTDOWN(mp))
227 xfs_iflags_clear(ip, XFS_ITRUNCATED);
229 if (mp->m_flags & XFS_MOUNT_BARRIER) {
231 * If we have an RT and/or log subvolume we need to make sure
232 * to flush the write cache the device used for file data
233 * first. This is to ensure newly written file data make
234 * it to disk before logging the new inode size in case of
235 * an extending write.
237 if (XFS_IS_REALTIME_INODE(ip))
238 xfs_blkdev_issue_flush(mp->m_rtdev_targp);
239 else if (mp->m_logdev_targp != mp->m_ddev_targp)
240 xfs_blkdev_issue_flush(mp->m_ddev_targp);
244 * All metadata updates are logged, which means that we just have
245 * to flush the log up to the latest LSN that touched the inode.
247 xfs_ilock(ip, XFS_ILOCK_SHARED);
248 if (xfs_ipincount(ip)) {
250 (ip->i_itemp->ili_fields & ~XFS_ILOG_TIMESTAMP))
251 lsn = ip->i_itemp->ili_last_lsn;
253 xfs_iunlock(ip, XFS_ILOCK_SHARED);
256 error = _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, &log_flushed);
259 * If we only have a single device, and the log force about was
260 * a no-op we might have to flush the data device cache here.
261 * This can only happen for fdatasync/O_DSYNC if we were overwriting
262 * an already allocated file and thus do not have any metadata to
265 if ((mp->m_flags & XFS_MOUNT_BARRIER) &&
266 mp->m_logdev_targp == mp->m_ddev_targp &&
267 !XFS_IS_REALTIME_INODE(ip) &&
269 xfs_blkdev_issue_flush(mp->m_ddev_targp);
279 struct file *file = iocb->ki_filp;
280 struct inode *inode = file->f_mapping->host;
281 struct xfs_inode *ip = XFS_I(inode);
282 struct xfs_mount *mp = ip->i_mount;
283 size_t size = iov_iter_count(to);
287 loff_t pos = iocb->ki_pos;
289 XFS_STATS_INC(xs_read_calls);
291 if (unlikely(iocb->ki_flags & IOCB_DIRECT))
292 ioflags |= XFS_IO_ISDIRECT;
293 if (file->f_mode & FMODE_NOCMTIME)
294 ioflags |= XFS_IO_INVIS;
296 if ((ioflags & XFS_IO_ISDIRECT) && !IS_DAX(inode)) {
297 xfs_buftarg_t *target =
298 XFS_IS_REALTIME_INODE(ip) ?
299 mp->m_rtdev_targp : mp->m_ddev_targp;
300 /* DIO must be aligned to device logical sector size */
301 if ((pos | size) & target->bt_logical_sectormask) {
302 if (pos == i_size_read(inode))
308 n = mp->m_super->s_maxbytes - pos;
309 if (n <= 0 || size == 0)
315 if (XFS_FORCED_SHUTDOWN(mp))
319 * Locking is a bit tricky here. If we take an exclusive lock
320 * for direct IO, we effectively serialise all new concurrent
321 * read IO to this file and block it behind IO that is currently in
322 * progress because IO in progress holds the IO lock shared. We only
323 * need to hold the lock exclusive to blow away the page cache, so
324 * only take lock exclusively if the page cache needs invalidation.
325 * This allows the normal direct IO case of no page cache pages to
326 * proceeed concurrently without serialisation.
328 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
329 if ((ioflags & XFS_IO_ISDIRECT) && inode->i_mapping->nrpages) {
330 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
331 xfs_rw_ilock(ip, XFS_IOLOCK_EXCL);
333 if (inode->i_mapping->nrpages) {
334 ret = filemap_write_and_wait_range(
335 VFS_I(ip)->i_mapping,
336 pos, pos + size - 1);
338 xfs_rw_iunlock(ip, XFS_IOLOCK_EXCL);
343 * Invalidate whole pages. This can return an error if
344 * we fail to invalidate a page, but this should never
345 * happen on XFS. Warn if it does fail.
347 ret = invalidate_inode_pages2_range(VFS_I(ip)->i_mapping,
348 pos >> PAGE_CACHE_SHIFT,
349 (pos + size - 1) >> PAGE_CACHE_SHIFT);
353 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
356 trace_xfs_file_read(ip, size, pos, ioflags);
358 ret = generic_file_read_iter(iocb, to);
360 XFS_STATS_ADD(xs_read_bytes, ret);
362 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
367 xfs_file_splice_read(
370 struct pipe_inode_info *pipe,
374 struct xfs_inode *ip = XFS_I(infilp->f_mapping->host);
378 XFS_STATS_INC(xs_read_calls);
380 if (infilp->f_mode & FMODE_NOCMTIME)
381 ioflags |= XFS_IO_INVIS;
383 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
386 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
388 trace_xfs_file_splice_read(ip, count, *ppos, ioflags);
390 /* for dax, we need to avoid the page cache */
391 if (IS_DAX(VFS_I(ip)))
392 ret = default_file_splice_read(infilp, ppos, pipe, count, flags);
394 ret = generic_file_splice_read(infilp, ppos, pipe, count, flags);
396 XFS_STATS_ADD(xs_read_bytes, ret);
398 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
403 * This routine is called to handle zeroing any space in the last block of the
404 * file that is beyond the EOF. We do this since the size is being increased
405 * without writing anything to that block and we don't want to read the
406 * garbage on the disk.
408 STATIC int /* error (positive) */
410 struct xfs_inode *ip,
415 struct xfs_mount *mp = ip->i_mount;
416 xfs_fileoff_t last_fsb = XFS_B_TO_FSBT(mp, isize);
417 int zero_offset = XFS_B_FSB_OFFSET(mp, isize);
421 struct xfs_bmbt_irec imap;
423 xfs_ilock(ip, XFS_ILOCK_EXCL);
424 error = xfs_bmapi_read(ip, last_fsb, 1, &imap, &nimaps, 0);
425 xfs_iunlock(ip, XFS_ILOCK_EXCL);
432 * If the block underlying isize is just a hole, then there
433 * is nothing to zero.
435 if (imap.br_startblock == HOLESTARTBLOCK)
438 zero_len = mp->m_sb.sb_blocksize - zero_offset;
439 if (isize + zero_len > offset)
440 zero_len = offset - isize;
442 return xfs_iozero(ip, isize, zero_len);
446 * Zero any on disk space between the current EOF and the new, larger EOF.
448 * This handles the normal case of zeroing the remainder of the last block in
449 * the file and the unusual case of zeroing blocks out beyond the size of the
450 * file. This second case only happens with fixed size extents and when the
451 * system crashes before the inode size was updated but after blocks were
454 * Expects the iolock to be held exclusive, and will take the ilock internally.
456 int /* error (positive) */
458 struct xfs_inode *ip,
459 xfs_off_t offset, /* starting I/O offset */
460 xfs_fsize_t isize, /* current inode size */
463 struct xfs_mount *mp = ip->i_mount;
464 xfs_fileoff_t start_zero_fsb;
465 xfs_fileoff_t end_zero_fsb;
466 xfs_fileoff_t zero_count_fsb;
467 xfs_fileoff_t last_fsb;
468 xfs_fileoff_t zero_off;
469 xfs_fsize_t zero_len;
472 struct xfs_bmbt_irec imap;
474 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
475 ASSERT(offset > isize);
478 * First handle zeroing the block on which isize resides.
480 * We only zero a part of that block so it is handled specially.
482 if (XFS_B_FSB_OFFSET(mp, isize) != 0) {
483 error = xfs_zero_last_block(ip, offset, isize, did_zeroing);
489 * Calculate the range between the new size and the old where blocks
490 * needing to be zeroed may exist.
492 * To get the block where the last byte in the file currently resides,
493 * we need to subtract one from the size and truncate back to a block
494 * boundary. We subtract 1 in case the size is exactly on a block
497 last_fsb = isize ? XFS_B_TO_FSBT(mp, isize - 1) : (xfs_fileoff_t)-1;
498 start_zero_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
499 end_zero_fsb = XFS_B_TO_FSBT(mp, offset - 1);
500 ASSERT((xfs_sfiloff_t)last_fsb < (xfs_sfiloff_t)start_zero_fsb);
501 if (last_fsb == end_zero_fsb) {
503 * The size was only incremented on its last block.
504 * We took care of that above, so just return.
509 ASSERT(start_zero_fsb <= end_zero_fsb);
510 while (start_zero_fsb <= end_zero_fsb) {
512 zero_count_fsb = end_zero_fsb - start_zero_fsb + 1;
514 xfs_ilock(ip, XFS_ILOCK_EXCL);
515 error = xfs_bmapi_read(ip, start_zero_fsb, zero_count_fsb,
517 xfs_iunlock(ip, XFS_ILOCK_EXCL);
523 if (imap.br_state == XFS_EXT_UNWRITTEN ||
524 imap.br_startblock == HOLESTARTBLOCK) {
525 start_zero_fsb = imap.br_startoff + imap.br_blockcount;
526 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
531 * There are blocks we need to zero.
533 zero_off = XFS_FSB_TO_B(mp, start_zero_fsb);
534 zero_len = XFS_FSB_TO_B(mp, imap.br_blockcount);
536 if ((zero_off + zero_len) > offset)
537 zero_len = offset - zero_off;
539 error = xfs_iozero(ip, zero_off, zero_len);
544 start_zero_fsb = imap.br_startoff + imap.br_blockcount;
545 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
552 * Common pre-write limit and setup checks.
554 * Called with the iolocked held either shared and exclusive according to
555 * @iolock, and returns with it held. Might upgrade the iolock to exclusive
556 * if called for a direct write beyond i_size.
559 xfs_file_aio_write_checks(
561 struct iov_iter *from,
564 struct file *file = iocb->ki_filp;
565 struct inode *inode = file->f_mapping->host;
566 struct xfs_inode *ip = XFS_I(inode);
568 size_t count = iov_iter_count(from);
571 error = generic_write_checks(iocb, from);
575 error = xfs_break_layouts(inode, iolock, true);
580 * If the offset is beyond the size of the file, we need to zero any
581 * blocks that fall between the existing EOF and the start of this
582 * write. If zeroing is needed and we are currently holding the
583 * iolock shared, we need to update it to exclusive which implies
584 * having to redo all checks before.
586 * We need to serialise against EOF updates that occur in IO
587 * completions here. We want to make sure that nobody is changing the
588 * size while we do this check until we have placed an IO barrier (i.e.
589 * hold the XFS_IOLOCK_EXCL) that prevents new IO from being dispatched.
590 * The spinlock effectively forms a memory barrier once we have the
591 * XFS_IOLOCK_EXCL so we are guaranteed to see the latest EOF value
592 * and hence be able to correctly determine if we need to run zeroing.
594 spin_lock(&ip->i_flags_lock);
595 if (iocb->ki_pos > i_size_read(inode)) {
598 spin_unlock(&ip->i_flags_lock);
599 if (*iolock == XFS_IOLOCK_SHARED) {
600 xfs_rw_iunlock(ip, *iolock);
601 *iolock = XFS_IOLOCK_EXCL;
602 xfs_rw_ilock(ip, *iolock);
603 iov_iter_reexpand(from, count);
606 * We now have an IO submission barrier in place, but
607 * AIO can do EOF updates during IO completion and hence
608 * we now need to wait for all of them to drain. Non-AIO
609 * DIO will have drained before we are given the
610 * XFS_IOLOCK_EXCL, and so for most cases this wait is a
613 inode_dio_wait(inode);
616 error = xfs_zero_eof(ip, iocb->ki_pos, i_size_read(inode), &zero);
620 spin_unlock(&ip->i_flags_lock);
623 * Updating the timestamps will grab the ilock again from
624 * xfs_fs_dirty_inode, so we have to call it after dropping the
625 * lock above. Eventually we should look into a way to avoid
626 * the pointless lock roundtrip.
628 if (likely(!(file->f_mode & FMODE_NOCMTIME))) {
629 error = file_update_time(file);
635 * If we're writing the file then make sure to clear the setuid and
636 * setgid bits if the process is not being run by root. This keeps
637 * people from modifying setuid and setgid binaries.
639 return file_remove_suid(file);
643 * xfs_file_dio_aio_write - handle direct IO writes
645 * Lock the inode appropriately to prepare for and issue a direct IO write.
646 * By separating it from the buffered write path we remove all the tricky to
647 * follow locking changes and looping.
649 * If there are cached pages or we're extending the file, we need IOLOCK_EXCL
650 * until we're sure the bytes at the new EOF have been zeroed and/or the cached
651 * pages are flushed out.
653 * In most cases the direct IO writes will be done holding IOLOCK_SHARED
654 * allowing them to be done in parallel with reads and other direct IO writes.
655 * However, if the IO is not aligned to filesystem blocks, the direct IO layer
656 * needs to do sub-block zeroing and that requires serialisation against other
657 * direct IOs to the same block. In this case we need to serialise the
658 * submission of the unaligned IOs so that we don't get racing block zeroing in
659 * the dio layer. To avoid the problem with aio, we also need to wait for
660 * outstanding IOs to complete so that unwritten extent conversion is completed
661 * before we try to map the overlapping block. This is currently implemented by
662 * hitting it with a big hammer (i.e. inode_dio_wait()).
664 * Returns with locks held indicated by @iolock and errors indicated by
665 * negative return values.
668 xfs_file_dio_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);
676 struct xfs_mount *mp = ip->i_mount;
678 int unaligned_io = 0;
680 size_t count = iov_iter_count(from);
681 loff_t pos = iocb->ki_pos;
683 struct iov_iter data;
684 struct xfs_buftarg *target = XFS_IS_REALTIME_INODE(ip) ?
685 mp->m_rtdev_targp : mp->m_ddev_targp;
687 /* DIO must be aligned to device logical sector size */
688 if (!IS_DAX(inode) && ((pos | count) & target->bt_logical_sectormask))
691 /* "unaligned" here means not aligned to a filesystem block */
692 if ((pos & mp->m_blockmask) || ((pos + count) & mp->m_blockmask))
696 * We don't need to take an exclusive lock unless there page cache needs
697 * to be invalidated or unaligned IO is being executed. We don't need to
698 * consider the EOF extension case here because
699 * xfs_file_aio_write_checks() will relock the inode as necessary for
700 * EOF zeroing cases and fill out the new inode size as appropriate.
702 if (unaligned_io || mapping->nrpages)
703 iolock = XFS_IOLOCK_EXCL;
705 iolock = XFS_IOLOCK_SHARED;
706 xfs_rw_ilock(ip, iolock);
709 * Recheck if there are cached pages that need invalidate after we got
710 * the iolock to protect against other threads adding new pages while
711 * we were waiting for the iolock.
713 if (mapping->nrpages && iolock == XFS_IOLOCK_SHARED) {
714 xfs_rw_iunlock(ip, iolock);
715 iolock = XFS_IOLOCK_EXCL;
716 xfs_rw_ilock(ip, iolock);
719 ret = xfs_file_aio_write_checks(iocb, from, &iolock);
722 count = iov_iter_count(from);
724 end = pos + count - 1;
726 if (mapping->nrpages) {
727 ret = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
732 * Invalidate whole pages. This can return an error if
733 * we fail to invalidate a page, but this should never
734 * happen on XFS. Warn if it does fail.
736 ret = invalidate_inode_pages2_range(VFS_I(ip)->i_mapping,
737 pos >> PAGE_CACHE_SHIFT,
738 end >> PAGE_CACHE_SHIFT);
744 * If we are doing unaligned IO, wait for all other IO to drain,
745 * otherwise demote the lock if we had to flush cached pages
748 inode_dio_wait(inode);
749 else if (iolock == XFS_IOLOCK_EXCL) {
750 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
751 iolock = XFS_IOLOCK_SHARED;
754 trace_xfs_file_direct_write(ip, count, iocb->ki_pos, 0);
757 ret = mapping->a_ops->direct_IO(iocb, &data, pos);
759 /* see generic_file_direct_write() for why this is necessary */
760 if (mapping->nrpages) {
761 invalidate_inode_pages2_range(mapping,
762 pos >> PAGE_CACHE_SHIFT,
763 end >> PAGE_CACHE_SHIFT);
768 iov_iter_advance(from, ret);
772 xfs_rw_iunlock(ip, iolock);
775 * No fallback to buffered IO on errors for XFS. DAX can result in
776 * partial writes, but direct IO will either complete fully or fail.
778 ASSERT(ret < 0 || ret == count || IS_DAX(VFS_I(ip)));
783 xfs_file_buffered_aio_write(
785 struct iov_iter *from)
787 struct file *file = iocb->ki_filp;
788 struct address_space *mapping = file->f_mapping;
789 struct inode *inode = mapping->host;
790 struct xfs_inode *ip = XFS_I(inode);
793 int iolock = XFS_IOLOCK_EXCL;
795 xfs_rw_ilock(ip, iolock);
797 ret = xfs_file_aio_write_checks(iocb, from, &iolock);
801 /* We can write back this queue in page reclaim */
802 current->backing_dev_info = inode_to_bdi(inode);
805 trace_xfs_file_buffered_write(ip, iov_iter_count(from),
807 ret = generic_perform_write(file, from, iocb->ki_pos);
808 if (likely(ret >= 0))
812 * If we hit a space limit, try to free up some lingering preallocated
813 * space before returning an error. In the case of ENOSPC, first try to
814 * write back all dirty inodes to free up some of the excess reserved
815 * metadata space. This reduces the chances that the eofblocks scan
816 * waits on dirty mappings. Since xfs_flush_inodes() is serialized, this
817 * also behaves as a filter to prevent too many eofblocks scans from
818 * running at the same time.
820 if (ret == -EDQUOT && !enospc) {
821 enospc = xfs_inode_free_quota_eofblocks(ip);
824 } else if (ret == -ENOSPC && !enospc) {
825 struct xfs_eofblocks eofb = {0};
828 xfs_flush_inodes(ip->i_mount);
829 eofb.eof_scan_owner = ip->i_ino; /* for locking */
830 eofb.eof_flags = XFS_EOF_FLAGS_SYNC;
831 xfs_icache_free_eofblocks(ip->i_mount, &eofb);
835 current->backing_dev_info = NULL;
837 xfs_rw_iunlock(ip, iolock);
844 struct iov_iter *from)
846 struct file *file = iocb->ki_filp;
847 struct address_space *mapping = file->f_mapping;
848 struct inode *inode = mapping->host;
849 struct xfs_inode *ip = XFS_I(inode);
851 size_t ocount = iov_iter_count(from);
853 XFS_STATS_INC(xs_write_calls);
858 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
861 if ((iocb->ki_flags & IOCB_DIRECT) || IS_DAX(inode))
862 ret = xfs_file_dio_aio_write(iocb, from);
864 ret = xfs_file_buffered_aio_write(iocb, from);
869 XFS_STATS_ADD(xs_write_bytes, ret);
871 /* Handle various SYNC-type writes */
872 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
879 #define XFS_FALLOC_FL_SUPPORTED \
880 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
881 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE | \
882 FALLOC_FL_INSERT_RANGE)
891 struct inode *inode = file_inode(file);
892 struct xfs_inode *ip = XFS_I(inode);
894 enum xfs_prealloc_flags flags = 0;
895 uint iolock = XFS_IOLOCK_EXCL;
897 bool do_file_insert = 0;
899 if (!S_ISREG(inode->i_mode))
901 if (mode & ~XFS_FALLOC_FL_SUPPORTED)
904 xfs_ilock(ip, iolock);
905 error = xfs_break_layouts(inode, &iolock, false);
909 xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
910 iolock |= XFS_MMAPLOCK_EXCL;
912 if (mode & FALLOC_FL_PUNCH_HOLE) {
913 error = xfs_free_file_space(ip, offset, len);
916 } else if (mode & FALLOC_FL_COLLAPSE_RANGE) {
917 unsigned blksize_mask = (1 << inode->i_blkbits) - 1;
919 if (offset & blksize_mask || len & blksize_mask) {
925 * There is no need to overlap collapse range with EOF,
926 * in which case it is effectively a truncate operation
928 if (offset + len >= i_size_read(inode)) {
933 new_size = i_size_read(inode) - len;
935 error = xfs_collapse_file_space(ip, offset, len);
938 } else if (mode & FALLOC_FL_INSERT_RANGE) {
939 unsigned blksize_mask = (1 << inode->i_blkbits) - 1;
941 new_size = i_size_read(inode) + len;
942 if (offset & blksize_mask || len & blksize_mask) {
947 /* check the new inode size does not wrap through zero */
948 if (new_size > inode->i_sb->s_maxbytes) {
953 /* Offset should be less than i_size */
954 if (offset >= i_size_read(inode)) {
960 flags |= XFS_PREALLOC_SET;
962 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
963 offset + len > i_size_read(inode)) {
964 new_size = offset + len;
965 error = inode_newsize_ok(inode, new_size);
970 if (mode & FALLOC_FL_ZERO_RANGE)
971 error = xfs_zero_file_space(ip, offset, len);
973 error = xfs_alloc_file_space(ip, offset, len,
979 if (file->f_flags & O_DSYNC)
980 flags |= XFS_PREALLOC_SYNC;
982 error = xfs_update_prealloc_flags(ip, flags);
986 /* Change file size if needed */
990 iattr.ia_valid = ATTR_SIZE;
991 iattr.ia_size = new_size;
992 error = xfs_setattr_size(ip, &iattr);
998 * Perform hole insertion now that the file size has been
999 * updated so that if we crash during the operation we don't
1000 * leave shifted extents past EOF and hence losing access to
1001 * the data that is contained within them.
1004 error = xfs_insert_file_space(ip, offset, len);
1007 xfs_iunlock(ip, iolock);
1014 struct inode *inode,
1017 if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1019 if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb)))
1026 struct inode *inode,
1029 struct xfs_inode *ip = XFS_I(inode);
1033 error = xfs_file_open(inode, file);
1038 * If there are any blocks, read-ahead block 0 as we're almost
1039 * certain to have the next operation be a read there.
1041 mode = xfs_ilock_data_map_shared(ip);
1042 if (ip->i_d.di_nextents > 0)
1043 xfs_dir3_data_readahead(ip, 0, -1);
1044 xfs_iunlock(ip, mode);
1050 struct inode *inode,
1053 return xfs_release(XFS_I(inode));
1059 struct dir_context *ctx)
1061 struct inode *inode = file_inode(file);
1062 xfs_inode_t *ip = XFS_I(inode);
1066 * The Linux API doesn't pass down the total size of the buffer
1067 * we read into down to the filesystem. With the filldir concept
1068 * it's not needed for correct information, but the XFS dir2 leaf
1069 * code wants an estimate of the buffer size to calculate it's
1070 * readahead window and size the buffers used for mapping to
1073 * Try to give it an estimate that's good enough, maybe at some
1074 * point we can change the ->readdir prototype to include the
1075 * buffer size. For now we use the current glibc buffer size.
1077 bufsize = (size_t)min_t(loff_t, 32768, ip->i_d.di_size);
1079 return xfs_readdir(ip, ctx, bufsize);
1083 * This type is designed to indicate the type of offset we would like
1084 * to search from page cache for xfs_seek_hole_data().
1092 * Lookup the desired type of offset from the given page.
1094 * On success, return true and the offset argument will point to the
1095 * start of the region that was found. Otherwise this function will
1096 * return false and keep the offset argument unchanged.
1099 xfs_lookup_buffer_offset(
1104 loff_t lastoff = page_offset(page);
1106 struct buffer_head *bh, *head;
1108 bh = head = page_buffers(page);
1111 * Unwritten extents that have data in the page
1112 * cache covering them can be identified by the
1113 * BH_Unwritten state flag. Pages with multiple
1114 * buffers might have a mix of holes, data and
1115 * unwritten extents - any buffer with valid
1116 * data in it should have BH_Uptodate flag set
1119 if (buffer_unwritten(bh) ||
1120 buffer_uptodate(bh)) {
1121 if (type == DATA_OFF)
1124 if (type == HOLE_OFF)
1132 lastoff += bh->b_size;
1133 } while ((bh = bh->b_this_page) != head);
1139 * This routine is called to find out and return a data or hole offset
1140 * from the page cache for unwritten extents according to the desired
1141 * type for xfs_seek_hole_data().
1143 * The argument offset is used to tell where we start to search from the
1144 * page cache. Map is used to figure out the end points of the range to
1147 * Return true if the desired type of offset was found, and the argument
1148 * offset is filled with that address. Otherwise, return false and keep
1152 xfs_find_get_desired_pgoff(
1153 struct inode *inode,
1154 struct xfs_bmbt_irec *map,
1158 struct xfs_inode *ip = XFS_I(inode);
1159 struct xfs_mount *mp = ip->i_mount;
1160 struct pagevec pvec;
1164 loff_t startoff = *offset;
1165 loff_t lastoff = startoff;
1168 pagevec_init(&pvec, 0);
1170 index = startoff >> PAGE_CACHE_SHIFT;
1171 endoff = XFS_FSB_TO_B(mp, map->br_startoff + map->br_blockcount);
1172 end = endoff >> PAGE_CACHE_SHIFT;
1178 want = min_t(pgoff_t, end - index, PAGEVEC_SIZE);
1179 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index,
1182 * No page mapped into given range. If we are searching holes
1183 * and if this is the first time we got into the loop, it means
1184 * that the given offset is landed in a hole, return it.
1186 * If we have already stepped through some block buffers to find
1187 * holes but they all contains data. In this case, the last
1188 * offset is already updated and pointed to the end of the last
1189 * mapped page, if it does not reach the endpoint to search,
1190 * that means there should be a hole between them.
1192 if (nr_pages == 0) {
1193 /* Data search found nothing */
1194 if (type == DATA_OFF)
1197 ASSERT(type == HOLE_OFF);
1198 if (lastoff == startoff || lastoff < endoff) {
1206 * At lease we found one page. If this is the first time we
1207 * step into the loop, and if the first page index offset is
1208 * greater than the given search offset, a hole was found.
1210 if (type == HOLE_OFF && lastoff == startoff &&
1211 lastoff < page_offset(pvec.pages[0])) {
1216 for (i = 0; i < nr_pages; i++) {
1217 struct page *page = pvec.pages[i];
1221 * At this point, the page may be truncated or
1222 * invalidated (changing page->mapping to NULL),
1223 * or even swizzled back from swapper_space to tmpfs
1224 * file mapping. However, page->index will not change
1225 * because we have a reference on the page.
1227 * Searching done if the page index is out of range.
1228 * If the current offset is not reaches the end of
1229 * the specified search range, there should be a hole
1232 if (page->index > end) {
1233 if (type == HOLE_OFF && lastoff < endoff) {
1242 * Page truncated or invalidated(page->mapping == NULL).
1243 * We can freely skip it and proceed to check the next
1246 if (unlikely(page->mapping != inode->i_mapping)) {
1251 if (!page_has_buffers(page)) {
1256 found = xfs_lookup_buffer_offset(page, &b_offset, type);
1259 * The found offset may be less than the start
1260 * point to search if this is the first time to
1263 *offset = max_t(loff_t, startoff, b_offset);
1269 * We either searching data but nothing was found, or
1270 * searching hole but found a data buffer. In either
1271 * case, probably the next page contains the desired
1272 * things, update the last offset to it so.
1274 lastoff = page_offset(page) + PAGE_SIZE;
1279 * The number of returned pages less than our desired, search
1280 * done. In this case, nothing was found for searching data,
1281 * but we found a hole behind the last offset.
1283 if (nr_pages < want) {
1284 if (type == HOLE_OFF) {
1291 index = pvec.pages[i - 1]->index + 1;
1292 pagevec_release(&pvec);
1293 } while (index <= end);
1296 pagevec_release(&pvec);
1306 struct inode *inode = file->f_mapping->host;
1307 struct xfs_inode *ip = XFS_I(inode);
1308 struct xfs_mount *mp = ip->i_mount;
1309 loff_t uninitialized_var(offset);
1311 xfs_fileoff_t fsbno;
1316 if (XFS_FORCED_SHUTDOWN(mp))
1319 lock = xfs_ilock_data_map_shared(ip);
1321 isize = i_size_read(inode);
1322 if (start >= isize) {
1328 * Try to read extents from the first block indicated
1329 * by fsbno to the end block of the file.
1331 fsbno = XFS_B_TO_FSBT(mp, start);
1332 end = XFS_B_TO_FSB(mp, isize);
1335 struct xfs_bmbt_irec map[2];
1339 error = xfs_bmapi_read(ip, fsbno, end - fsbno, map, &nmap,
1344 /* No extents at given offset, must be beyond EOF */
1350 for (i = 0; i < nmap; i++) {
1351 offset = max_t(loff_t, start,
1352 XFS_FSB_TO_B(mp, map[i].br_startoff));
1354 /* Landed in the hole we wanted? */
1355 if (whence == SEEK_HOLE &&
1356 map[i].br_startblock == HOLESTARTBLOCK)
1359 /* Landed in the data extent we wanted? */
1360 if (whence == SEEK_DATA &&
1361 (map[i].br_startblock == DELAYSTARTBLOCK ||
1362 (map[i].br_state == XFS_EXT_NORM &&
1363 !isnullstartblock(map[i].br_startblock))))
1367 * Landed in an unwritten extent, try to search
1368 * for hole or data from page cache.
1370 if (map[i].br_state == XFS_EXT_UNWRITTEN) {
1371 if (xfs_find_get_desired_pgoff(inode, &map[i],
1372 whence == SEEK_HOLE ? HOLE_OFF : DATA_OFF,
1379 * We only received one extent out of the two requested. This
1380 * means we've hit EOF and didn't find what we are looking for.
1384 * If we were looking for a hole, set offset to
1385 * the end of the file (i.e., there is an implicit
1386 * hole at the end of any file).
1388 if (whence == SEEK_HOLE) {
1393 * If we were looking for data, it's nowhere to be found
1395 ASSERT(whence == SEEK_DATA);
1403 * Nothing was found, proceed to the next round of search
1404 * if the next reading offset is not at or beyond EOF.
1406 fsbno = map[i - 1].br_startoff + map[i - 1].br_blockcount;
1407 start = XFS_FSB_TO_B(mp, fsbno);
1408 if (start >= isize) {
1409 if (whence == SEEK_HOLE) {
1413 ASSERT(whence == SEEK_DATA);
1421 * If at this point we have found the hole we wanted, the returned
1422 * offset may be bigger than the file size as it may be aligned to
1423 * page boundary for unwritten extents. We need to deal with this
1424 * situation in particular.
1426 if (whence == SEEK_HOLE)
1427 offset = min_t(loff_t, offset, isize);
1428 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
1431 xfs_iunlock(ip, lock);
1448 return generic_file_llseek(file, offset, whence);
1451 return xfs_seek_hole_data(file, offset, whence);
1458 * Locking for serialisation of IO during page faults. This results in a lock
1462 * sb_start_pagefault(vfs, freeze)
1463 * i_mmap_lock (XFS - truncate serialisation)
1465 * i_lock (XFS - extent map serialisation)
1469 * mmap()d file has taken write protection fault and is being made writable. We
1470 * can set the page state up correctly for a writable page, which means we can
1471 * do correct delalloc accounting (ENOSPC checking!) and unwritten extent
1475 xfs_filemap_page_mkwrite(
1476 struct vm_area_struct *vma,
1477 struct vm_fault *vmf)
1479 struct inode *inode = file_inode(vma->vm_file);
1482 trace_xfs_filemap_page_mkwrite(XFS_I(inode));
1484 sb_start_pagefault(inode->i_sb);
1485 file_update_time(vma->vm_file);
1486 xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1488 if (IS_DAX(inode)) {
1489 ret = __dax_mkwrite(vma, vmf, xfs_get_blocks_direct,
1490 xfs_end_io_dax_write);
1492 ret = __block_page_mkwrite(vma, vmf, xfs_get_blocks);
1493 ret = block_page_mkwrite_return(ret);
1496 xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1497 sb_end_pagefault(inode->i_sb);
1504 struct vm_area_struct *vma,
1505 struct vm_fault *vmf)
1507 struct xfs_inode *ip = XFS_I(file_inode(vma->vm_file));
1510 trace_xfs_filemap_fault(ip);
1512 /* DAX can shortcut the normal fault path on write faults! */
1513 if ((vmf->flags & FAULT_FLAG_WRITE) && IS_DAX(VFS_I(ip)))
1514 return xfs_filemap_page_mkwrite(vma, vmf);
1516 xfs_ilock(ip, XFS_MMAPLOCK_SHARED);
1517 ret = filemap_fault(vma, vmf);
1518 xfs_iunlock(ip, XFS_MMAPLOCK_SHARED);
1523 static const struct vm_operations_struct xfs_file_vm_ops = {
1524 .fault = xfs_filemap_fault,
1525 .map_pages = filemap_map_pages,
1526 .page_mkwrite = xfs_filemap_page_mkwrite,
1532 struct vm_area_struct *vma)
1534 file_accessed(filp);
1535 vma->vm_ops = &xfs_file_vm_ops;
1536 if (IS_DAX(file_inode(filp)))
1537 vma->vm_flags |= VM_MIXEDMAP;
1541 const struct file_operations xfs_file_operations = {
1542 .llseek = xfs_file_llseek,
1543 .read_iter = xfs_file_read_iter,
1544 .write_iter = xfs_file_write_iter,
1545 .splice_read = xfs_file_splice_read,
1546 .splice_write = iter_file_splice_write,
1547 .unlocked_ioctl = xfs_file_ioctl,
1548 #ifdef CONFIG_COMPAT
1549 .compat_ioctl = xfs_file_compat_ioctl,
1551 .mmap = xfs_file_mmap,
1552 .open = xfs_file_open,
1553 .release = xfs_file_release,
1554 .fsync = xfs_file_fsync,
1555 .fallocate = xfs_file_fallocate,
1558 const struct file_operations xfs_dir_file_operations = {
1559 .open = xfs_dir_open,
1560 .read = generic_read_dir,
1561 .iterate = xfs_file_readdir,
1562 .llseek = generic_file_llseek,
1563 .unlocked_ioctl = xfs_file_ioctl,
1564 #ifdef CONFIG_COMPAT
1565 .compat_ioctl = xfs_file_compat_ioctl,
1567 .fsync = xfs_dir_fsync,