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);
84 * xfs_iozero clears the specified range of buffer supplied,
85 * and marks all the affected blocks as valid and modified. If
86 * an affected block is not allocated, it will be allocated. If
87 * an affected block is not completely overwritten, and is not
88 * valid before the operation, it will be read from disk before
89 * being partially zeroed.
93 struct xfs_inode *ip, /* inode */
94 loff_t pos, /* offset in file */
95 size_t count) /* size of data to zero */
98 struct address_space *mapping;
101 mapping = VFS_I(ip)->i_mapping;
103 unsigned offset, bytes;
106 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
107 bytes = PAGE_CACHE_SIZE - offset;
111 status = pagecache_write_begin(NULL, mapping, pos, bytes,
112 AOP_FLAG_UNINTERRUPTIBLE,
117 zero_user(page, offset, bytes);
119 status = pagecache_write_end(NULL, mapping, pos, bytes, bytes,
121 WARN_ON(status <= 0); /* can't return less than zero! */
131 xfs_update_prealloc_flags(
132 struct xfs_inode *ip,
133 enum xfs_prealloc_flags flags)
135 struct xfs_trans *tp;
138 tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_WRITEID);
139 error = xfs_trans_reserve(tp, &M_RES(ip->i_mount)->tr_writeid, 0, 0);
141 xfs_trans_cancel(tp, 0);
145 xfs_ilock(ip, XFS_ILOCK_EXCL);
146 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
148 if (!(flags & XFS_PREALLOC_INVISIBLE)) {
149 ip->i_d.di_mode &= ~S_ISUID;
150 if (ip->i_d.di_mode & S_IXGRP)
151 ip->i_d.di_mode &= ~S_ISGID;
152 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
155 if (flags & XFS_PREALLOC_SET)
156 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
157 if (flags & XFS_PREALLOC_CLEAR)
158 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
160 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
161 if (flags & XFS_PREALLOC_SYNC)
162 xfs_trans_set_sync(tp);
163 return xfs_trans_commit(tp, 0);
167 * Fsync operations on directories are much simpler than on regular files,
168 * as there is no file data to flush, and thus also no need for explicit
169 * cache flush operations, and there are no non-transaction metadata updates
170 * on directories either.
179 struct xfs_inode *ip = XFS_I(file->f_mapping->host);
180 struct xfs_mount *mp = ip->i_mount;
183 trace_xfs_dir_fsync(ip);
185 xfs_ilock(ip, XFS_ILOCK_SHARED);
186 if (xfs_ipincount(ip))
187 lsn = ip->i_itemp->ili_last_lsn;
188 xfs_iunlock(ip, XFS_ILOCK_SHARED);
192 return _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, NULL);
202 struct inode *inode = file->f_mapping->host;
203 struct xfs_inode *ip = XFS_I(inode);
204 struct xfs_mount *mp = ip->i_mount;
209 trace_xfs_file_fsync(ip);
211 error = filemap_write_and_wait_range(inode->i_mapping, start, end);
215 if (XFS_FORCED_SHUTDOWN(mp))
218 xfs_iflags_clear(ip, XFS_ITRUNCATED);
220 if (mp->m_flags & XFS_MOUNT_BARRIER) {
222 * If we have an RT and/or log subvolume we need to make sure
223 * to flush the write cache the device used for file data
224 * first. This is to ensure newly written file data make
225 * it to disk before logging the new inode size in case of
226 * an extending write.
228 if (XFS_IS_REALTIME_INODE(ip))
229 xfs_blkdev_issue_flush(mp->m_rtdev_targp);
230 else if (mp->m_logdev_targp != mp->m_ddev_targp)
231 xfs_blkdev_issue_flush(mp->m_ddev_targp);
235 * All metadata updates are logged, which means that we just have
236 * to flush the log up to the latest LSN that touched the inode.
238 xfs_ilock(ip, XFS_ILOCK_SHARED);
239 if (xfs_ipincount(ip)) {
241 (ip->i_itemp->ili_fields & ~XFS_ILOG_TIMESTAMP))
242 lsn = ip->i_itemp->ili_last_lsn;
244 xfs_iunlock(ip, XFS_ILOCK_SHARED);
247 error = _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, &log_flushed);
250 * If we only have a single device, and the log force about was
251 * a no-op we might have to flush the data device cache here.
252 * This can only happen for fdatasync/O_DSYNC if we were overwriting
253 * an already allocated file and thus do not have any metadata to
256 if ((mp->m_flags & XFS_MOUNT_BARRIER) &&
257 mp->m_logdev_targp == mp->m_ddev_targp &&
258 !XFS_IS_REALTIME_INODE(ip) &&
260 xfs_blkdev_issue_flush(mp->m_ddev_targp);
270 struct file *file = iocb->ki_filp;
271 struct inode *inode = file->f_mapping->host;
272 struct xfs_inode *ip = XFS_I(inode);
273 struct xfs_mount *mp = ip->i_mount;
274 size_t size = iov_iter_count(to);
278 loff_t pos = iocb->ki_pos;
280 XFS_STATS_INC(xs_read_calls);
282 if (unlikely(iocb->ki_flags & IOCB_DIRECT))
283 ioflags |= XFS_IO_ISDIRECT;
284 if (file->f_mode & FMODE_NOCMTIME)
285 ioflags |= XFS_IO_INVIS;
287 if (unlikely(ioflags & XFS_IO_ISDIRECT)) {
288 xfs_buftarg_t *target =
289 XFS_IS_REALTIME_INODE(ip) ?
290 mp->m_rtdev_targp : mp->m_ddev_targp;
291 /* DIO must be aligned to device logical sector size */
292 if ((pos | size) & target->bt_logical_sectormask) {
293 if (pos == i_size_read(inode))
299 n = mp->m_super->s_maxbytes - pos;
300 if (n <= 0 || size == 0)
306 if (XFS_FORCED_SHUTDOWN(mp))
310 * Locking is a bit tricky here. If we take an exclusive lock
311 * for direct IO, we effectively serialise all new concurrent
312 * read IO to this file and block it behind IO that is currently in
313 * progress because IO in progress holds the IO lock shared. We only
314 * need to hold the lock exclusive to blow away the page cache, so
315 * only take lock exclusively if the page cache needs invalidation.
316 * This allows the normal direct IO case of no page cache pages to
317 * proceeed concurrently without serialisation.
319 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
320 if ((ioflags & XFS_IO_ISDIRECT) && inode->i_mapping->nrpages) {
321 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
322 xfs_rw_ilock(ip, XFS_IOLOCK_EXCL);
324 if (inode->i_mapping->nrpages) {
325 ret = filemap_write_and_wait_range(
326 VFS_I(ip)->i_mapping,
327 pos, pos + size - 1);
329 xfs_rw_iunlock(ip, XFS_IOLOCK_EXCL);
334 * Invalidate whole pages. This can return an error if
335 * we fail to invalidate a page, but this should never
336 * happen on XFS. Warn if it does fail.
338 ret = invalidate_inode_pages2_range(VFS_I(ip)->i_mapping,
339 pos >> PAGE_CACHE_SHIFT,
340 (pos + size - 1) >> PAGE_CACHE_SHIFT);
344 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
347 trace_xfs_file_read(ip, size, pos, ioflags);
349 ret = generic_file_read_iter(iocb, to);
351 XFS_STATS_ADD(xs_read_bytes, ret);
353 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
358 xfs_file_splice_read(
361 struct pipe_inode_info *pipe,
365 struct xfs_inode *ip = XFS_I(infilp->f_mapping->host);
369 XFS_STATS_INC(xs_read_calls);
371 if (infilp->f_mode & FMODE_NOCMTIME)
372 ioflags |= XFS_IO_INVIS;
374 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
377 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
379 trace_xfs_file_splice_read(ip, count, *ppos, ioflags);
381 ret = generic_file_splice_read(infilp, ppos, pipe, count, flags);
383 XFS_STATS_ADD(xs_read_bytes, ret);
385 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
390 * This routine is called to handle zeroing any space in the last block of the
391 * file that is beyond the EOF. We do this since the size is being increased
392 * without writing anything to that block and we don't want to read the
393 * garbage on the disk.
395 STATIC int /* error (positive) */
397 struct xfs_inode *ip,
402 struct xfs_mount *mp = ip->i_mount;
403 xfs_fileoff_t last_fsb = XFS_B_TO_FSBT(mp, isize);
404 int zero_offset = XFS_B_FSB_OFFSET(mp, isize);
408 struct xfs_bmbt_irec imap;
410 xfs_ilock(ip, XFS_ILOCK_EXCL);
411 error = xfs_bmapi_read(ip, last_fsb, 1, &imap, &nimaps, 0);
412 xfs_iunlock(ip, XFS_ILOCK_EXCL);
419 * If the block underlying isize is just a hole, then there
420 * is nothing to zero.
422 if (imap.br_startblock == HOLESTARTBLOCK)
425 zero_len = mp->m_sb.sb_blocksize - zero_offset;
426 if (isize + zero_len > offset)
427 zero_len = offset - isize;
429 return xfs_iozero(ip, isize, zero_len);
433 * Zero any on disk space between the current EOF and the new, larger EOF.
435 * This handles the normal case of zeroing the remainder of the last block in
436 * the file and the unusual case of zeroing blocks out beyond the size of the
437 * file. This second case only happens with fixed size extents and when the
438 * system crashes before the inode size was updated but after blocks were
441 * Expects the iolock to be held exclusive, and will take the ilock internally.
443 int /* error (positive) */
445 struct xfs_inode *ip,
446 xfs_off_t offset, /* starting I/O offset */
447 xfs_fsize_t isize, /* current inode size */
450 struct xfs_mount *mp = ip->i_mount;
451 xfs_fileoff_t start_zero_fsb;
452 xfs_fileoff_t end_zero_fsb;
453 xfs_fileoff_t zero_count_fsb;
454 xfs_fileoff_t last_fsb;
455 xfs_fileoff_t zero_off;
456 xfs_fsize_t zero_len;
459 struct xfs_bmbt_irec imap;
461 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
462 ASSERT(offset > isize);
465 * First handle zeroing the block on which isize resides.
467 * We only zero a part of that block so it is handled specially.
469 if (XFS_B_FSB_OFFSET(mp, isize) != 0) {
470 error = xfs_zero_last_block(ip, offset, isize, did_zeroing);
476 * Calculate the range between the new size and the old where blocks
477 * needing to be zeroed may exist.
479 * To get the block where the last byte in the file currently resides,
480 * we need to subtract one from the size and truncate back to a block
481 * boundary. We subtract 1 in case the size is exactly on a block
484 last_fsb = isize ? XFS_B_TO_FSBT(mp, isize - 1) : (xfs_fileoff_t)-1;
485 start_zero_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
486 end_zero_fsb = XFS_B_TO_FSBT(mp, offset - 1);
487 ASSERT((xfs_sfiloff_t)last_fsb < (xfs_sfiloff_t)start_zero_fsb);
488 if (last_fsb == end_zero_fsb) {
490 * The size was only incremented on its last block.
491 * We took care of that above, so just return.
496 ASSERT(start_zero_fsb <= end_zero_fsb);
497 while (start_zero_fsb <= end_zero_fsb) {
499 zero_count_fsb = end_zero_fsb - start_zero_fsb + 1;
501 xfs_ilock(ip, XFS_ILOCK_EXCL);
502 error = xfs_bmapi_read(ip, start_zero_fsb, zero_count_fsb,
504 xfs_iunlock(ip, XFS_ILOCK_EXCL);
510 if (imap.br_state == XFS_EXT_UNWRITTEN ||
511 imap.br_startblock == HOLESTARTBLOCK) {
512 start_zero_fsb = imap.br_startoff + imap.br_blockcount;
513 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
518 * There are blocks we need to zero.
520 zero_off = XFS_FSB_TO_B(mp, start_zero_fsb);
521 zero_len = XFS_FSB_TO_B(mp, imap.br_blockcount);
523 if ((zero_off + zero_len) > offset)
524 zero_len = offset - zero_off;
526 error = xfs_iozero(ip, zero_off, zero_len);
531 start_zero_fsb = imap.br_startoff + imap.br_blockcount;
532 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
539 * Common pre-write limit and setup checks.
541 * Called with the iolocked held either shared and exclusive according to
542 * @iolock, and returns with it held. Might upgrade the iolock to exclusive
543 * if called for a direct write beyond i_size.
546 xfs_file_aio_write_checks(
548 struct iov_iter *from,
551 struct file *file = iocb->ki_filp;
552 struct inode *inode = file->f_mapping->host;
553 struct xfs_inode *ip = XFS_I(inode);
555 size_t count = iov_iter_count(from);
558 error = generic_write_checks(iocb, from);
562 error = xfs_break_layouts(inode, iolock, true);
567 * If the offset is beyond the size of the file, we need to zero any
568 * blocks that fall between the existing EOF and the start of this
569 * write. If zeroing is needed and we are currently holding the
570 * iolock shared, we need to update it to exclusive which implies
571 * having to redo all checks before.
573 * We need to serialise against EOF updates that occur in IO
574 * completions here. We want to make sure that nobody is changing the
575 * size while we do this check until we have placed an IO barrier (i.e.
576 * hold the XFS_IOLOCK_EXCL) that prevents new IO from being dispatched.
577 * The spinlock effectively forms a memory barrier once we have the
578 * XFS_IOLOCK_EXCL so we are guaranteed to see the latest EOF value
579 * and hence be able to correctly determine if we need to run zeroing.
581 spin_lock(&ip->i_flags_lock);
582 if (iocb->ki_pos > i_size_read(inode)) {
585 spin_unlock(&ip->i_flags_lock);
586 if (*iolock == XFS_IOLOCK_SHARED) {
587 xfs_rw_iunlock(ip, *iolock);
588 *iolock = XFS_IOLOCK_EXCL;
589 xfs_rw_ilock(ip, *iolock);
590 iov_iter_reexpand(from, count);
593 * We now have an IO submission barrier in place, but
594 * AIO can do EOF updates during IO completion and hence
595 * we now need to wait for all of them to drain. Non-AIO
596 * DIO will have drained before we are given the
597 * XFS_IOLOCK_EXCL, and so for most cases this wait is a
600 inode_dio_wait(inode);
603 error = xfs_zero_eof(ip, iocb->ki_pos, i_size_read(inode), &zero);
607 spin_unlock(&ip->i_flags_lock);
610 * Updating the timestamps will grab the ilock again from
611 * xfs_fs_dirty_inode, so we have to call it after dropping the
612 * lock above. Eventually we should look into a way to avoid
613 * the pointless lock roundtrip.
615 if (likely(!(file->f_mode & FMODE_NOCMTIME))) {
616 error = file_update_time(file);
622 * If we're writing the file then make sure to clear the setuid and
623 * setgid bits if the process is not being run by root. This keeps
624 * people from modifying setuid and setgid binaries.
626 return file_remove_suid(file);
630 * xfs_file_dio_aio_write - handle direct IO writes
632 * Lock the inode appropriately to prepare for and issue a direct IO write.
633 * By separating it from the buffered write path we remove all the tricky to
634 * follow locking changes and looping.
636 * If there are cached pages or we're extending the file, we need IOLOCK_EXCL
637 * until we're sure the bytes at the new EOF have been zeroed and/or the cached
638 * pages are flushed out.
640 * In most cases the direct IO writes will be done holding IOLOCK_SHARED
641 * allowing them to be done in parallel with reads and other direct IO writes.
642 * However, if the IO is not aligned to filesystem blocks, the direct IO layer
643 * needs to do sub-block zeroing and that requires serialisation against other
644 * direct IOs to the same block. In this case we need to serialise the
645 * submission of the unaligned IOs so that we don't get racing block zeroing in
646 * the dio layer. To avoid the problem with aio, we also need to wait for
647 * outstanding IOs to complete so that unwritten extent conversion is completed
648 * before we try to map the overlapping block. This is currently implemented by
649 * hitting it with a big hammer (i.e. inode_dio_wait()).
651 * Returns with locks held indicated by @iolock and errors indicated by
652 * negative return values.
655 xfs_file_dio_aio_write(
657 struct iov_iter *from)
659 struct file *file = iocb->ki_filp;
660 struct address_space *mapping = file->f_mapping;
661 struct inode *inode = mapping->host;
662 struct xfs_inode *ip = XFS_I(inode);
663 struct xfs_mount *mp = ip->i_mount;
665 int unaligned_io = 0;
667 size_t count = iov_iter_count(from);
668 loff_t pos = iocb->ki_pos;
670 struct iov_iter data;
671 struct xfs_buftarg *target = XFS_IS_REALTIME_INODE(ip) ?
672 mp->m_rtdev_targp : mp->m_ddev_targp;
674 /* DIO must be aligned to device logical sector size */
675 if ((pos | count) & target->bt_logical_sectormask)
678 /* "unaligned" here means not aligned to a filesystem block */
679 if ((pos & mp->m_blockmask) || ((pos + count) & mp->m_blockmask))
683 * We don't need to take an exclusive lock unless there page cache needs
684 * to be invalidated or unaligned IO is being executed. We don't need to
685 * consider the EOF extension case here because
686 * xfs_file_aio_write_checks() will relock the inode as necessary for
687 * EOF zeroing cases and fill out the new inode size as appropriate.
689 if (unaligned_io || mapping->nrpages)
690 iolock = XFS_IOLOCK_EXCL;
692 iolock = XFS_IOLOCK_SHARED;
693 xfs_rw_ilock(ip, iolock);
696 * Recheck if there are cached pages that need invalidate after we got
697 * the iolock to protect against other threads adding new pages while
698 * we were waiting for the iolock.
700 if (mapping->nrpages && iolock == XFS_IOLOCK_SHARED) {
701 xfs_rw_iunlock(ip, iolock);
702 iolock = XFS_IOLOCK_EXCL;
703 xfs_rw_ilock(ip, iolock);
706 ret = xfs_file_aio_write_checks(iocb, from, &iolock);
709 count = iov_iter_count(from);
711 end = pos + count - 1;
713 if (mapping->nrpages) {
714 ret = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
719 * Invalidate whole pages. This can return an error if
720 * we fail to invalidate a page, but this should never
721 * happen on XFS. Warn if it does fail.
723 ret = invalidate_inode_pages2_range(VFS_I(ip)->i_mapping,
724 pos >> PAGE_CACHE_SHIFT,
725 end >> PAGE_CACHE_SHIFT);
731 * If we are doing unaligned IO, wait for all other IO to drain,
732 * otherwise demote the lock if we had to flush cached pages
735 inode_dio_wait(inode);
736 else if (iolock == XFS_IOLOCK_EXCL) {
737 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
738 iolock = XFS_IOLOCK_SHARED;
741 trace_xfs_file_direct_write(ip, count, iocb->ki_pos, 0);
744 ret = mapping->a_ops->direct_IO(iocb, &data, pos);
746 /* see generic_file_direct_write() for why this is necessary */
747 if (mapping->nrpages) {
748 invalidate_inode_pages2_range(mapping,
749 pos >> PAGE_CACHE_SHIFT,
750 end >> PAGE_CACHE_SHIFT);
755 iov_iter_advance(from, ret);
759 xfs_rw_iunlock(ip, iolock);
761 /* No fallback to buffered IO on errors for XFS. */
762 ASSERT(ret < 0 || ret == count);
767 xfs_file_buffered_aio_write(
769 struct iov_iter *from)
771 struct file *file = iocb->ki_filp;
772 struct address_space *mapping = file->f_mapping;
773 struct inode *inode = mapping->host;
774 struct xfs_inode *ip = XFS_I(inode);
777 int iolock = XFS_IOLOCK_EXCL;
779 xfs_rw_ilock(ip, iolock);
781 ret = xfs_file_aio_write_checks(iocb, from, &iolock);
785 /* We can write back this queue in page reclaim */
786 current->backing_dev_info = inode_to_bdi(inode);
789 trace_xfs_file_buffered_write(ip, iov_iter_count(from),
791 ret = generic_perform_write(file, from, iocb->ki_pos);
792 if (likely(ret >= 0))
796 * If we hit a space limit, try to free up some lingering preallocated
797 * space before returning an error. In the case of ENOSPC, first try to
798 * write back all dirty inodes to free up some of the excess reserved
799 * metadata space. This reduces the chances that the eofblocks scan
800 * waits on dirty mappings. Since xfs_flush_inodes() is serialized, this
801 * also behaves as a filter to prevent too many eofblocks scans from
802 * running at the same time.
804 if (ret == -EDQUOT && !enospc) {
805 enospc = xfs_inode_free_quota_eofblocks(ip);
808 } else if (ret == -ENOSPC && !enospc) {
809 struct xfs_eofblocks eofb = {0};
812 xfs_flush_inodes(ip->i_mount);
813 eofb.eof_scan_owner = ip->i_ino; /* for locking */
814 eofb.eof_flags = XFS_EOF_FLAGS_SYNC;
815 xfs_icache_free_eofblocks(ip->i_mount, &eofb);
819 current->backing_dev_info = NULL;
821 xfs_rw_iunlock(ip, iolock);
828 struct iov_iter *from)
830 struct file *file = iocb->ki_filp;
831 struct address_space *mapping = file->f_mapping;
832 struct inode *inode = mapping->host;
833 struct xfs_inode *ip = XFS_I(inode);
835 size_t ocount = iov_iter_count(from);
837 XFS_STATS_INC(xs_write_calls);
842 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
845 if (unlikely(iocb->ki_flags & IOCB_DIRECT))
846 ret = xfs_file_dio_aio_write(iocb, from);
848 ret = xfs_file_buffered_aio_write(iocb, from);
853 XFS_STATS_ADD(xs_write_bytes, ret);
855 /* Handle various SYNC-type writes */
856 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
863 #define XFS_FALLOC_FL_SUPPORTED \
864 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
865 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE | \
866 FALLOC_FL_INSERT_RANGE)
875 struct inode *inode = file_inode(file);
876 struct xfs_inode *ip = XFS_I(inode);
878 enum xfs_prealloc_flags flags = 0;
879 uint iolock = XFS_IOLOCK_EXCL;
881 bool do_file_insert = 0;
883 if (!S_ISREG(inode->i_mode))
885 if (mode & ~XFS_FALLOC_FL_SUPPORTED)
888 xfs_ilock(ip, iolock);
889 error = xfs_break_layouts(inode, &iolock, false);
893 xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
894 iolock |= XFS_MMAPLOCK_EXCL;
896 if (mode & FALLOC_FL_PUNCH_HOLE) {
897 error = xfs_free_file_space(ip, offset, len);
900 } else if (mode & FALLOC_FL_COLLAPSE_RANGE) {
901 unsigned blksize_mask = (1 << inode->i_blkbits) - 1;
903 if (offset & blksize_mask || len & blksize_mask) {
909 * There is no need to overlap collapse range with EOF,
910 * in which case it is effectively a truncate operation
912 if (offset + len >= i_size_read(inode)) {
917 new_size = i_size_read(inode) - len;
919 error = xfs_collapse_file_space(ip, offset, len);
922 } else if (mode & FALLOC_FL_INSERT_RANGE) {
923 unsigned blksize_mask = (1 << inode->i_blkbits) - 1;
925 new_size = i_size_read(inode) + len;
926 if (offset & blksize_mask || len & blksize_mask) {
931 /* check the new inode size does not wrap through zero */
932 if (new_size > inode->i_sb->s_maxbytes) {
937 /* Offset should be less than i_size */
938 if (offset >= i_size_read(inode)) {
944 flags |= XFS_PREALLOC_SET;
946 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
947 offset + len > i_size_read(inode)) {
948 new_size = offset + len;
949 error = inode_newsize_ok(inode, new_size);
954 if (mode & FALLOC_FL_ZERO_RANGE)
955 error = xfs_zero_file_space(ip, offset, len);
957 error = xfs_alloc_file_space(ip, offset, len,
963 if (file->f_flags & O_DSYNC)
964 flags |= XFS_PREALLOC_SYNC;
966 error = xfs_update_prealloc_flags(ip, flags);
970 /* Change file size if needed */
974 iattr.ia_valid = ATTR_SIZE;
975 iattr.ia_size = new_size;
976 error = xfs_setattr_size(ip, &iattr);
982 * Perform hole insertion now that the file size has been
983 * updated so that if we crash during the operation we don't
984 * leave shifted extents past EOF and hence losing access to
985 * the data that is contained within them.
988 error = xfs_insert_file_space(ip, offset, len);
991 xfs_iunlock(ip, iolock);
1001 if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1003 if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb)))
1010 struct inode *inode,
1013 struct xfs_inode *ip = XFS_I(inode);
1017 error = xfs_file_open(inode, file);
1022 * If there are any blocks, read-ahead block 0 as we're almost
1023 * certain to have the next operation be a read there.
1025 mode = xfs_ilock_data_map_shared(ip);
1026 if (ip->i_d.di_nextents > 0)
1027 xfs_dir3_data_readahead(ip, 0, -1);
1028 xfs_iunlock(ip, mode);
1034 struct inode *inode,
1037 return xfs_release(XFS_I(inode));
1043 struct dir_context *ctx)
1045 struct inode *inode = file_inode(file);
1046 xfs_inode_t *ip = XFS_I(inode);
1050 * The Linux API doesn't pass down the total size of the buffer
1051 * we read into down to the filesystem. With the filldir concept
1052 * it's not needed for correct information, but the XFS dir2 leaf
1053 * code wants an estimate of the buffer size to calculate it's
1054 * readahead window and size the buffers used for mapping to
1057 * Try to give it an estimate that's good enough, maybe at some
1058 * point we can change the ->readdir prototype to include the
1059 * buffer size. For now we use the current glibc buffer size.
1061 bufsize = (size_t)min_t(loff_t, 32768, ip->i_d.di_size);
1063 return xfs_readdir(ip, ctx, bufsize);
1069 struct vm_area_struct *vma)
1071 vma->vm_ops = &xfs_file_vm_ops;
1073 file_accessed(filp);
1078 * This type is designed to indicate the type of offset we would like
1079 * to search from page cache for xfs_seek_hole_data().
1087 * Lookup the desired type of offset from the given page.
1089 * On success, return true and the offset argument will point to the
1090 * start of the region that was found. Otherwise this function will
1091 * return false and keep the offset argument unchanged.
1094 xfs_lookup_buffer_offset(
1099 loff_t lastoff = page_offset(page);
1101 struct buffer_head *bh, *head;
1103 bh = head = page_buffers(page);
1106 * Unwritten extents that have data in the page
1107 * cache covering them can be identified by the
1108 * BH_Unwritten state flag. Pages with multiple
1109 * buffers might have a mix of holes, data and
1110 * unwritten extents - any buffer with valid
1111 * data in it should have BH_Uptodate flag set
1114 if (buffer_unwritten(bh) ||
1115 buffer_uptodate(bh)) {
1116 if (type == DATA_OFF)
1119 if (type == HOLE_OFF)
1127 lastoff += bh->b_size;
1128 } while ((bh = bh->b_this_page) != head);
1134 * This routine is called to find out and return a data or hole offset
1135 * from the page cache for unwritten extents according to the desired
1136 * type for xfs_seek_hole_data().
1138 * The argument offset is used to tell where we start to search from the
1139 * page cache. Map is used to figure out the end points of the range to
1142 * Return true if the desired type of offset was found, and the argument
1143 * offset is filled with that address. Otherwise, return false and keep
1147 xfs_find_get_desired_pgoff(
1148 struct inode *inode,
1149 struct xfs_bmbt_irec *map,
1153 struct xfs_inode *ip = XFS_I(inode);
1154 struct xfs_mount *mp = ip->i_mount;
1155 struct pagevec pvec;
1159 loff_t startoff = *offset;
1160 loff_t lastoff = startoff;
1163 pagevec_init(&pvec, 0);
1165 index = startoff >> PAGE_CACHE_SHIFT;
1166 endoff = XFS_FSB_TO_B(mp, map->br_startoff + map->br_blockcount);
1167 end = endoff >> PAGE_CACHE_SHIFT;
1173 want = min_t(pgoff_t, end - index, PAGEVEC_SIZE);
1174 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index,
1177 * No page mapped into given range. If we are searching holes
1178 * and if this is the first time we got into the loop, it means
1179 * that the given offset is landed in a hole, return it.
1181 * If we have already stepped through some block buffers to find
1182 * holes but they all contains data. In this case, the last
1183 * offset is already updated and pointed to the end of the last
1184 * mapped page, if it does not reach the endpoint to search,
1185 * that means there should be a hole between them.
1187 if (nr_pages == 0) {
1188 /* Data search found nothing */
1189 if (type == DATA_OFF)
1192 ASSERT(type == HOLE_OFF);
1193 if (lastoff == startoff || lastoff < endoff) {
1201 * At lease we found one page. If this is the first time we
1202 * step into the loop, and if the first page index offset is
1203 * greater than the given search offset, a hole was found.
1205 if (type == HOLE_OFF && lastoff == startoff &&
1206 lastoff < page_offset(pvec.pages[0])) {
1211 for (i = 0; i < nr_pages; i++) {
1212 struct page *page = pvec.pages[i];
1216 * At this point, the page may be truncated or
1217 * invalidated (changing page->mapping to NULL),
1218 * or even swizzled back from swapper_space to tmpfs
1219 * file mapping. However, page->index will not change
1220 * because we have a reference on the page.
1222 * Searching done if the page index is out of range.
1223 * If the current offset is not reaches the end of
1224 * the specified search range, there should be a hole
1227 if (page->index > end) {
1228 if (type == HOLE_OFF && lastoff < endoff) {
1237 * Page truncated or invalidated(page->mapping == NULL).
1238 * We can freely skip it and proceed to check the next
1241 if (unlikely(page->mapping != inode->i_mapping)) {
1246 if (!page_has_buffers(page)) {
1251 found = xfs_lookup_buffer_offset(page, &b_offset, type);
1254 * The found offset may be less than the start
1255 * point to search if this is the first time to
1258 *offset = max_t(loff_t, startoff, b_offset);
1264 * We either searching data but nothing was found, or
1265 * searching hole but found a data buffer. In either
1266 * case, probably the next page contains the desired
1267 * things, update the last offset to it so.
1269 lastoff = page_offset(page) + PAGE_SIZE;
1274 * The number of returned pages less than our desired, search
1275 * done. In this case, nothing was found for searching data,
1276 * but we found a hole behind the last offset.
1278 if (nr_pages < want) {
1279 if (type == HOLE_OFF) {
1286 index = pvec.pages[i - 1]->index + 1;
1287 pagevec_release(&pvec);
1288 } while (index <= end);
1291 pagevec_release(&pvec);
1301 struct inode *inode = file->f_mapping->host;
1302 struct xfs_inode *ip = XFS_I(inode);
1303 struct xfs_mount *mp = ip->i_mount;
1304 loff_t uninitialized_var(offset);
1306 xfs_fileoff_t fsbno;
1311 if (XFS_FORCED_SHUTDOWN(mp))
1314 lock = xfs_ilock_data_map_shared(ip);
1316 isize = i_size_read(inode);
1317 if (start >= isize) {
1323 * Try to read extents from the first block indicated
1324 * by fsbno to the end block of the file.
1326 fsbno = XFS_B_TO_FSBT(mp, start);
1327 end = XFS_B_TO_FSB(mp, isize);
1330 struct xfs_bmbt_irec map[2];
1334 error = xfs_bmapi_read(ip, fsbno, end - fsbno, map, &nmap,
1339 /* No extents at given offset, must be beyond EOF */
1345 for (i = 0; i < nmap; i++) {
1346 offset = max_t(loff_t, start,
1347 XFS_FSB_TO_B(mp, map[i].br_startoff));
1349 /* Landed in the hole we wanted? */
1350 if (whence == SEEK_HOLE &&
1351 map[i].br_startblock == HOLESTARTBLOCK)
1354 /* Landed in the data extent we wanted? */
1355 if (whence == SEEK_DATA &&
1356 (map[i].br_startblock == DELAYSTARTBLOCK ||
1357 (map[i].br_state == XFS_EXT_NORM &&
1358 !isnullstartblock(map[i].br_startblock))))
1362 * Landed in an unwritten extent, try to search
1363 * for hole or data from page cache.
1365 if (map[i].br_state == XFS_EXT_UNWRITTEN) {
1366 if (xfs_find_get_desired_pgoff(inode, &map[i],
1367 whence == SEEK_HOLE ? HOLE_OFF : DATA_OFF,
1374 * We only received one extent out of the two requested. This
1375 * means we've hit EOF and didn't find what we are looking for.
1379 * If we were looking for a hole, set offset to
1380 * the end of the file (i.e., there is an implicit
1381 * hole at the end of any file).
1383 if (whence == SEEK_HOLE) {
1388 * If we were looking for data, it's nowhere to be found
1390 ASSERT(whence == SEEK_DATA);
1398 * Nothing was found, proceed to the next round of search
1399 * if the next reading offset is not at or beyond EOF.
1401 fsbno = map[i - 1].br_startoff + map[i - 1].br_blockcount;
1402 start = XFS_FSB_TO_B(mp, fsbno);
1403 if (start >= isize) {
1404 if (whence == SEEK_HOLE) {
1408 ASSERT(whence == SEEK_DATA);
1416 * If at this point we have found the hole we wanted, the returned
1417 * offset may be bigger than the file size as it may be aligned to
1418 * page boundary for unwritten extents. We need to deal with this
1419 * situation in particular.
1421 if (whence == SEEK_HOLE)
1422 offset = min_t(loff_t, offset, isize);
1423 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
1426 xfs_iunlock(ip, lock);
1443 return generic_file_llseek(file, offset, whence);
1446 return xfs_seek_hole_data(file, offset, whence);
1453 * Locking for serialisation of IO during page faults. This results in a lock
1457 * i_mmap_lock (XFS - truncate serialisation)
1459 * i_lock (XFS - extent map serialisation)
1463 struct vm_area_struct *vma,
1464 struct vm_fault *vmf)
1466 struct xfs_inode *ip = XFS_I(vma->vm_file->f_mapping->host);
1469 trace_xfs_filemap_fault(ip);
1471 xfs_ilock(ip, XFS_MMAPLOCK_SHARED);
1472 error = filemap_fault(vma, vmf);
1473 xfs_iunlock(ip, XFS_MMAPLOCK_SHARED);
1479 * mmap()d file has taken write protection fault and is being made writable. We
1480 * can set the page state up correctly for a writable page, which means we can
1481 * do correct delalloc accounting (ENOSPC checking!) and unwritten extent
1485 xfs_filemap_page_mkwrite(
1486 struct vm_area_struct *vma,
1487 struct vm_fault *vmf)
1489 struct xfs_inode *ip = XFS_I(vma->vm_file->f_mapping->host);
1492 trace_xfs_filemap_page_mkwrite(ip);
1494 xfs_ilock(ip, XFS_MMAPLOCK_SHARED);
1495 error = block_page_mkwrite(vma, vmf, xfs_get_blocks);
1496 xfs_iunlock(ip, XFS_MMAPLOCK_SHARED);
1501 const struct file_operations xfs_file_operations = {
1502 .llseek = xfs_file_llseek,
1503 .read_iter = xfs_file_read_iter,
1504 .write_iter = xfs_file_write_iter,
1505 .splice_read = xfs_file_splice_read,
1506 .splice_write = iter_file_splice_write,
1507 .unlocked_ioctl = xfs_file_ioctl,
1508 #ifdef CONFIG_COMPAT
1509 .compat_ioctl = xfs_file_compat_ioctl,
1511 .mmap = xfs_file_mmap,
1512 .open = xfs_file_open,
1513 .release = xfs_file_release,
1514 .fsync = xfs_file_fsync,
1515 .fallocate = xfs_file_fallocate,
1518 const struct file_operations xfs_dir_file_operations = {
1519 .open = xfs_dir_open,
1520 .read = generic_read_dir,
1521 .iterate = xfs_file_readdir,
1522 .llseek = generic_file_llseek,
1523 .unlocked_ioctl = xfs_file_ioctl,
1524 #ifdef CONFIG_COMPAT
1525 .compat_ioctl = xfs_file_compat_ioctl,
1527 .fsync = xfs_dir_fsync,
1530 static const struct vm_operations_struct xfs_file_vm_ops = {
1531 .fault = xfs_filemap_fault,
1532 .map_pages = filemap_map_pages,
1533 .page_mkwrite = xfs_filemap_page_mkwrite,