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 <linux/dcache.h>
41 #include <linux/falloc.h>
42 #include <linux/pagevec.h>
44 static const struct vm_operations_struct xfs_file_vm_ops;
47 * Locking primitives for read and write IO paths to ensure we consistently use
48 * and order the inode->i_mutex, ip->i_lock and ip->i_iolock.
55 if (type & XFS_IOLOCK_EXCL)
56 mutex_lock(&VFS_I(ip)->i_mutex);
65 xfs_iunlock(ip, type);
66 if (type & XFS_IOLOCK_EXCL)
67 mutex_unlock(&VFS_I(ip)->i_mutex);
75 xfs_ilock_demote(ip, type);
76 if (type & XFS_IOLOCK_EXCL)
77 mutex_unlock(&VFS_I(ip)->i_mutex);
83 * xfs_iozero clears the specified range of buffer supplied,
84 * and marks all the affected blocks as valid and modified. If
85 * an affected block is not allocated, it will be allocated. If
86 * an affected block is not completely overwritten, and is not
87 * valid before the operation, it will be read from disk before
88 * being partially zeroed.
92 struct xfs_inode *ip, /* inode */
93 loff_t pos, /* offset in file */
94 size_t count) /* size of data to zero */
97 struct address_space *mapping;
100 mapping = VFS_I(ip)->i_mapping;
102 unsigned offset, bytes;
105 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
106 bytes = PAGE_CACHE_SIZE - offset;
110 status = pagecache_write_begin(NULL, mapping, pos, bytes,
111 AOP_FLAG_UNINTERRUPTIBLE,
116 zero_user(page, offset, bytes);
118 status = pagecache_write_end(NULL, mapping, pos, bytes, bytes,
120 WARN_ON(status <= 0); /* can't return less than zero! */
130 xfs_update_prealloc_flags(
131 struct xfs_inode *ip,
132 enum xfs_prealloc_flags flags)
134 struct xfs_trans *tp;
137 tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_WRITEID);
138 error = xfs_trans_reserve(tp, &M_RES(ip->i_mount)->tr_writeid, 0, 0);
140 xfs_trans_cancel(tp, 0);
144 xfs_ilock(ip, XFS_ILOCK_EXCL);
145 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
147 if (!(flags & XFS_PREALLOC_INVISIBLE)) {
148 ip->i_d.di_mode &= ~S_ISUID;
149 if (ip->i_d.di_mode & S_IXGRP)
150 ip->i_d.di_mode &= ~S_ISGID;
151 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
154 if (flags & XFS_PREALLOC_SET)
155 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
156 if (flags & XFS_PREALLOC_CLEAR)
157 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
159 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
160 if (flags & XFS_PREALLOC_SYNC)
161 xfs_trans_set_sync(tp);
162 return xfs_trans_commit(tp, 0);
166 * Fsync operations on directories are much simpler than on regular files,
167 * as there is no file data to flush, and thus also no need for explicit
168 * cache flush operations, and there are no non-transaction metadata updates
169 * on directories either.
178 struct xfs_inode *ip = XFS_I(file->f_mapping->host);
179 struct xfs_mount *mp = ip->i_mount;
182 trace_xfs_dir_fsync(ip);
184 xfs_ilock(ip, XFS_ILOCK_SHARED);
185 if (xfs_ipincount(ip))
186 lsn = ip->i_itemp->ili_last_lsn;
187 xfs_iunlock(ip, XFS_ILOCK_SHARED);
191 return _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, NULL);
201 struct inode *inode = file->f_mapping->host;
202 struct xfs_inode *ip = XFS_I(inode);
203 struct xfs_mount *mp = ip->i_mount;
208 trace_xfs_file_fsync(ip);
210 error = filemap_write_and_wait_range(inode->i_mapping, start, end);
214 if (XFS_FORCED_SHUTDOWN(mp))
217 xfs_iflags_clear(ip, XFS_ITRUNCATED);
219 if (mp->m_flags & XFS_MOUNT_BARRIER) {
221 * If we have an RT and/or log subvolume we need to make sure
222 * to flush the write cache the device used for file data
223 * first. This is to ensure newly written file data make
224 * it to disk before logging the new inode size in case of
225 * an extending write.
227 if (XFS_IS_REALTIME_INODE(ip))
228 xfs_blkdev_issue_flush(mp->m_rtdev_targp);
229 else if (mp->m_logdev_targp != mp->m_ddev_targp)
230 xfs_blkdev_issue_flush(mp->m_ddev_targp);
234 * All metadata updates are logged, which means that we just have
235 * to flush the log up to the latest LSN that touched the inode.
237 xfs_ilock(ip, XFS_ILOCK_SHARED);
238 if (xfs_ipincount(ip)) {
240 (ip->i_itemp->ili_fields & ~XFS_ILOG_TIMESTAMP))
241 lsn = ip->i_itemp->ili_last_lsn;
243 xfs_iunlock(ip, XFS_ILOCK_SHARED);
246 error = _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, &log_flushed);
249 * If we only have a single device, and the log force about was
250 * a no-op we might have to flush the data device cache here.
251 * This can only happen for fdatasync/O_DSYNC if we were overwriting
252 * an already allocated file and thus do not have any metadata to
255 if ((mp->m_flags & XFS_MOUNT_BARRIER) &&
256 mp->m_logdev_targp == mp->m_ddev_targp &&
257 !XFS_IS_REALTIME_INODE(ip) &&
259 xfs_blkdev_issue_flush(mp->m_ddev_targp);
269 struct file *file = iocb->ki_filp;
270 struct inode *inode = file->f_mapping->host;
271 struct xfs_inode *ip = XFS_I(inode);
272 struct xfs_mount *mp = ip->i_mount;
273 size_t size = iov_iter_count(to);
277 loff_t pos = iocb->ki_pos;
279 XFS_STATS_INC(xs_read_calls);
281 if (unlikely(file->f_flags & O_DIRECT))
282 ioflags |= XFS_IO_ISDIRECT;
283 if (file->f_mode & FMODE_NOCMTIME)
284 ioflags |= XFS_IO_INVIS;
286 if (unlikely(ioflags & XFS_IO_ISDIRECT)) {
287 xfs_buftarg_t *target =
288 XFS_IS_REALTIME_INODE(ip) ?
289 mp->m_rtdev_targp : mp->m_ddev_targp;
290 /* DIO must be aligned to device logical sector size */
291 if ((pos | size) & target->bt_logical_sectormask) {
292 if (pos == i_size_read(inode))
298 n = mp->m_super->s_maxbytes - pos;
299 if (n <= 0 || size == 0)
305 if (XFS_FORCED_SHUTDOWN(mp))
309 * Locking is a bit tricky here. If we take an exclusive lock
310 * for direct IO, we effectively serialise all new concurrent
311 * read IO to this file and block it behind IO that is currently in
312 * progress because IO in progress holds the IO lock shared. We only
313 * need to hold the lock exclusive to blow away the page cache, so
314 * only take lock exclusively if the page cache needs invalidation.
315 * This allows the normal direct IO case of no page cache pages to
316 * proceeed concurrently without serialisation.
318 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
319 if ((ioflags & XFS_IO_ISDIRECT) && inode->i_mapping->nrpages) {
320 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
321 xfs_rw_ilock(ip, XFS_IOLOCK_EXCL);
323 if (inode->i_mapping->nrpages) {
324 ret = filemap_write_and_wait_range(
325 VFS_I(ip)->i_mapping,
326 pos, pos + size - 1);
328 xfs_rw_iunlock(ip, XFS_IOLOCK_EXCL);
333 * Invalidate whole pages. This can return an error if
334 * we fail to invalidate a page, but this should never
335 * happen on XFS. Warn if it does fail.
337 ret = invalidate_inode_pages2_range(VFS_I(ip)->i_mapping,
338 pos >> PAGE_CACHE_SHIFT,
339 (pos + size - 1) >> PAGE_CACHE_SHIFT);
343 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
346 trace_xfs_file_read(ip, size, pos, ioflags);
348 ret = generic_file_read_iter(iocb, to);
350 XFS_STATS_ADD(xs_read_bytes, ret);
352 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
357 xfs_file_splice_read(
360 struct pipe_inode_info *pipe,
364 struct xfs_inode *ip = XFS_I(infilp->f_mapping->host);
368 XFS_STATS_INC(xs_read_calls);
370 if (infilp->f_mode & FMODE_NOCMTIME)
371 ioflags |= XFS_IO_INVIS;
373 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
376 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
378 trace_xfs_file_splice_read(ip, count, *ppos, ioflags);
380 ret = generic_file_splice_read(infilp, ppos, pipe, count, flags);
382 XFS_STATS_ADD(xs_read_bytes, ret);
384 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
389 * This routine is called to handle zeroing any space in the last block of the
390 * file that is beyond the EOF. We do this since the size is being increased
391 * without writing anything to that block and we don't want to read the
392 * garbage on the disk.
394 STATIC int /* error (positive) */
396 struct xfs_inode *ip,
400 struct xfs_mount *mp = ip->i_mount;
401 xfs_fileoff_t last_fsb = XFS_B_TO_FSBT(mp, isize);
402 int zero_offset = XFS_B_FSB_OFFSET(mp, isize);
406 struct xfs_bmbt_irec imap;
408 xfs_ilock(ip, XFS_ILOCK_EXCL);
409 error = xfs_bmapi_read(ip, last_fsb, 1, &imap, &nimaps, 0);
410 xfs_iunlock(ip, XFS_ILOCK_EXCL);
417 * If the block underlying isize is just a hole, then there
418 * is nothing to zero.
420 if (imap.br_startblock == HOLESTARTBLOCK)
423 zero_len = mp->m_sb.sb_blocksize - zero_offset;
424 if (isize + zero_len > offset)
425 zero_len = offset - isize;
426 return xfs_iozero(ip, isize, zero_len);
430 * Zero any on disk space between the current EOF and the new, larger EOF.
432 * This handles the normal case of zeroing the remainder of the last block in
433 * the file and the unusual case of zeroing blocks out beyond the size of the
434 * file. This second case only happens with fixed size extents and when the
435 * system crashes before the inode size was updated but after blocks were
438 * Expects the iolock to be held exclusive, and will take the ilock internally.
440 int /* error (positive) */
442 struct xfs_inode *ip,
443 xfs_off_t offset, /* starting I/O offset */
444 xfs_fsize_t isize) /* current inode size */
446 struct xfs_mount *mp = ip->i_mount;
447 xfs_fileoff_t start_zero_fsb;
448 xfs_fileoff_t end_zero_fsb;
449 xfs_fileoff_t zero_count_fsb;
450 xfs_fileoff_t last_fsb;
451 xfs_fileoff_t zero_off;
452 xfs_fsize_t zero_len;
455 struct xfs_bmbt_irec imap;
457 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
458 ASSERT(offset > isize);
461 * First handle zeroing the block on which isize resides.
463 * We only zero a part of that block so it is handled specially.
465 if (XFS_B_FSB_OFFSET(mp, isize) != 0) {
466 error = xfs_zero_last_block(ip, offset, isize);
472 * Calculate the range between the new size and the old where blocks
473 * needing to be zeroed may exist.
475 * To get the block where the last byte in the file currently resides,
476 * we need to subtract one from the size and truncate back to a block
477 * boundary. We subtract 1 in case the size is exactly on a block
480 last_fsb = isize ? XFS_B_TO_FSBT(mp, isize - 1) : (xfs_fileoff_t)-1;
481 start_zero_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
482 end_zero_fsb = XFS_B_TO_FSBT(mp, offset - 1);
483 ASSERT((xfs_sfiloff_t)last_fsb < (xfs_sfiloff_t)start_zero_fsb);
484 if (last_fsb == end_zero_fsb) {
486 * The size was only incremented on its last block.
487 * We took care of that above, so just return.
492 ASSERT(start_zero_fsb <= end_zero_fsb);
493 while (start_zero_fsb <= end_zero_fsb) {
495 zero_count_fsb = end_zero_fsb - start_zero_fsb + 1;
497 xfs_ilock(ip, XFS_ILOCK_EXCL);
498 error = xfs_bmapi_read(ip, start_zero_fsb, zero_count_fsb,
500 xfs_iunlock(ip, XFS_ILOCK_EXCL);
506 if (imap.br_state == XFS_EXT_UNWRITTEN ||
507 imap.br_startblock == HOLESTARTBLOCK) {
508 start_zero_fsb = imap.br_startoff + imap.br_blockcount;
509 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
514 * There are blocks we need to zero.
516 zero_off = XFS_FSB_TO_B(mp, start_zero_fsb);
517 zero_len = XFS_FSB_TO_B(mp, imap.br_blockcount);
519 if ((zero_off + zero_len) > offset)
520 zero_len = offset - zero_off;
522 error = xfs_iozero(ip, zero_off, zero_len);
526 start_zero_fsb = imap.br_startoff + imap.br_blockcount;
527 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
534 * Common pre-write limit and setup checks.
536 * Called with the iolocked held either shared and exclusive according to
537 * @iolock, and returns with it held. Might upgrade the iolock to exclusive
538 * if called for a direct write beyond i_size.
541 xfs_file_aio_write_checks(
547 struct inode *inode = file->f_mapping->host;
548 struct xfs_inode *ip = XFS_I(inode);
552 error = generic_write_checks(file, pos, count, S_ISBLK(inode->i_mode));
557 * If the offset is beyond the size of the file, we need to zero any
558 * blocks that fall between the existing EOF and the start of this
559 * write. If zeroing is needed and we are currently holding the
560 * iolock shared, we need to update it to exclusive which implies
561 * having to redo all checks before.
563 if (*pos > i_size_read(inode)) {
564 if (*iolock == XFS_IOLOCK_SHARED) {
565 xfs_rw_iunlock(ip, *iolock);
566 *iolock = XFS_IOLOCK_EXCL;
567 xfs_rw_ilock(ip, *iolock);
570 error = xfs_zero_eof(ip, *pos, i_size_read(inode));
576 * Updating the timestamps will grab the ilock again from
577 * xfs_fs_dirty_inode, so we have to call it after dropping the
578 * lock above. Eventually we should look into a way to avoid
579 * the pointless lock roundtrip.
581 if (likely(!(file->f_mode & FMODE_NOCMTIME))) {
582 error = file_update_time(file);
588 * If we're writing the file then make sure to clear the setuid and
589 * setgid bits if the process is not being run by root. This keeps
590 * people from modifying setuid and setgid binaries.
592 return file_remove_suid(file);
596 * xfs_file_dio_aio_write - handle direct IO writes
598 * Lock the inode appropriately to prepare for and issue a direct IO write.
599 * By separating it from the buffered write path we remove all the tricky to
600 * follow locking changes and looping.
602 * If there are cached pages or we're extending the file, we need IOLOCK_EXCL
603 * until we're sure the bytes at the new EOF have been zeroed and/or the cached
604 * pages are flushed out.
606 * In most cases the direct IO writes will be done holding IOLOCK_SHARED
607 * allowing them to be done in parallel with reads and other direct IO writes.
608 * However, if the IO is not aligned to filesystem blocks, the direct IO layer
609 * needs to do sub-block zeroing and that requires serialisation against other
610 * direct IOs to the same block. In this case we need to serialise the
611 * submission of the unaligned IOs so that we don't get racing block zeroing in
612 * the dio layer. To avoid the problem with aio, we also need to wait for
613 * outstanding IOs to complete so that unwritten extent conversion is completed
614 * before we try to map the overlapping block. This is currently implemented by
615 * hitting it with a big hammer (i.e. inode_dio_wait()).
617 * Returns with locks held indicated by @iolock and errors indicated by
618 * negative return values.
621 xfs_file_dio_aio_write(
623 struct iov_iter *from)
625 struct file *file = iocb->ki_filp;
626 struct address_space *mapping = file->f_mapping;
627 struct inode *inode = mapping->host;
628 struct xfs_inode *ip = XFS_I(inode);
629 struct xfs_mount *mp = ip->i_mount;
631 int unaligned_io = 0;
633 size_t count = iov_iter_count(from);
634 loff_t pos = iocb->ki_pos;
635 struct xfs_buftarg *target = XFS_IS_REALTIME_INODE(ip) ?
636 mp->m_rtdev_targp : mp->m_ddev_targp;
638 /* DIO must be aligned to device logical sector size */
639 if ((pos | count) & target->bt_logical_sectormask)
642 /* "unaligned" here means not aligned to a filesystem block */
643 if ((pos & mp->m_blockmask) || ((pos + count) & mp->m_blockmask))
647 * We don't need to take an exclusive lock unless there page cache needs
648 * to be invalidated or unaligned IO is being executed. We don't need to
649 * consider the EOF extension case here because
650 * xfs_file_aio_write_checks() will relock the inode as necessary for
651 * EOF zeroing cases and fill out the new inode size as appropriate.
653 if (unaligned_io || mapping->nrpages)
654 iolock = XFS_IOLOCK_EXCL;
656 iolock = XFS_IOLOCK_SHARED;
657 xfs_rw_ilock(ip, iolock);
660 * Recheck if there are cached pages that need invalidate after we got
661 * the iolock to protect against other threads adding new pages while
662 * we were waiting for the iolock.
664 if (mapping->nrpages && iolock == XFS_IOLOCK_SHARED) {
665 xfs_rw_iunlock(ip, iolock);
666 iolock = XFS_IOLOCK_EXCL;
667 xfs_rw_ilock(ip, iolock);
670 ret = xfs_file_aio_write_checks(file, &pos, &count, &iolock);
673 iov_iter_truncate(from, count);
675 if (mapping->nrpages) {
676 ret = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
677 pos, pos + count - 1);
681 * Invalidate whole pages. This can return an error if
682 * we fail to invalidate a page, but this should never
683 * happen on XFS. Warn if it does fail.
685 ret = invalidate_inode_pages2_range(VFS_I(ip)->i_mapping,
686 pos >> PAGE_CACHE_SHIFT,
687 (pos + count - 1) >> PAGE_CACHE_SHIFT);
693 * If we are doing unaligned IO, wait for all other IO to drain,
694 * otherwise demote the lock if we had to flush cached pages
697 inode_dio_wait(inode);
698 else if (iolock == XFS_IOLOCK_EXCL) {
699 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
700 iolock = XFS_IOLOCK_SHARED;
703 trace_xfs_file_direct_write(ip, count, iocb->ki_pos, 0);
704 ret = generic_file_direct_write(iocb, from, pos);
707 xfs_rw_iunlock(ip, iolock);
709 /* No fallback to buffered IO on errors for XFS. */
710 ASSERT(ret < 0 || ret == count);
715 xfs_file_buffered_aio_write(
717 struct iov_iter *from)
719 struct file *file = iocb->ki_filp;
720 struct address_space *mapping = file->f_mapping;
721 struct inode *inode = mapping->host;
722 struct xfs_inode *ip = XFS_I(inode);
725 int iolock = XFS_IOLOCK_EXCL;
726 loff_t pos = iocb->ki_pos;
727 size_t count = iov_iter_count(from);
729 xfs_rw_ilock(ip, iolock);
731 ret = xfs_file_aio_write_checks(file, &pos, &count, &iolock);
735 iov_iter_truncate(from, count);
736 /* We can write back this queue in page reclaim */
737 current->backing_dev_info = inode_to_bdi(inode);
740 trace_xfs_file_buffered_write(ip, count, iocb->ki_pos, 0);
741 ret = generic_perform_write(file, from, pos);
742 if (likely(ret >= 0))
743 iocb->ki_pos = pos + ret;
746 * If we hit a space limit, try to free up some lingering preallocated
747 * space before returning an error. In the case of ENOSPC, first try to
748 * write back all dirty inodes to free up some of the excess reserved
749 * metadata space. This reduces the chances that the eofblocks scan
750 * waits on dirty mappings. Since xfs_flush_inodes() is serialized, this
751 * also behaves as a filter to prevent too many eofblocks scans from
752 * running at the same time.
754 if (ret == -EDQUOT && !enospc) {
755 enospc = xfs_inode_free_quota_eofblocks(ip);
758 } else if (ret == -ENOSPC && !enospc) {
759 struct xfs_eofblocks eofb = {0};
762 xfs_flush_inodes(ip->i_mount);
763 eofb.eof_scan_owner = ip->i_ino; /* for locking */
764 eofb.eof_flags = XFS_EOF_FLAGS_SYNC;
765 xfs_icache_free_eofblocks(ip->i_mount, &eofb);
769 current->backing_dev_info = NULL;
771 xfs_rw_iunlock(ip, iolock);
778 struct iov_iter *from)
780 struct file *file = iocb->ki_filp;
781 struct address_space *mapping = file->f_mapping;
782 struct inode *inode = mapping->host;
783 struct xfs_inode *ip = XFS_I(inode);
785 size_t ocount = iov_iter_count(from);
787 XFS_STATS_INC(xs_write_calls);
792 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
795 if (unlikely(file->f_flags & O_DIRECT))
796 ret = xfs_file_dio_aio_write(iocb, from);
798 ret = xfs_file_buffered_aio_write(iocb, from);
803 XFS_STATS_ADD(xs_write_bytes, ret);
805 /* Handle various SYNC-type writes */
806 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
820 struct inode *inode = file_inode(file);
821 struct xfs_inode *ip = XFS_I(inode);
823 enum xfs_prealloc_flags flags = 0;
826 if (!S_ISREG(inode->i_mode))
828 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
829 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE))
832 xfs_ilock(ip, XFS_IOLOCK_EXCL);
833 if (mode & FALLOC_FL_PUNCH_HOLE) {
834 error = xfs_free_file_space(ip, offset, len);
837 } else if (mode & FALLOC_FL_COLLAPSE_RANGE) {
838 unsigned blksize_mask = (1 << inode->i_blkbits) - 1;
840 if (offset & blksize_mask || len & blksize_mask) {
846 * There is no need to overlap collapse range with EOF,
847 * in which case it is effectively a truncate operation
849 if (offset + len >= i_size_read(inode)) {
854 new_size = i_size_read(inode) - len;
856 error = xfs_collapse_file_space(ip, offset, len);
860 flags |= XFS_PREALLOC_SET;
862 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
863 offset + len > i_size_read(inode)) {
864 new_size = offset + len;
865 error = inode_newsize_ok(inode, new_size);
870 if (mode & FALLOC_FL_ZERO_RANGE)
871 error = xfs_zero_file_space(ip, offset, len);
873 error = xfs_alloc_file_space(ip, offset, len,
879 if (file->f_flags & O_DSYNC)
880 flags |= XFS_PREALLOC_SYNC;
882 error = xfs_update_prealloc_flags(ip, flags);
886 /* Change file size if needed */
890 iattr.ia_valid = ATTR_SIZE;
891 iattr.ia_size = new_size;
892 error = xfs_setattr_size(ip, &iattr);
896 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
906 if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
908 if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb)))
918 struct xfs_inode *ip = XFS_I(inode);
922 error = xfs_file_open(inode, file);
927 * If there are any blocks, read-ahead block 0 as we're almost
928 * certain to have the next operation be a read there.
930 mode = xfs_ilock_data_map_shared(ip);
931 if (ip->i_d.di_nextents > 0)
932 xfs_dir3_data_readahead(ip, 0, -1);
933 xfs_iunlock(ip, mode);
942 return xfs_release(XFS_I(inode));
948 struct dir_context *ctx)
950 struct inode *inode = file_inode(file);
951 xfs_inode_t *ip = XFS_I(inode);
955 * The Linux API doesn't pass down the total size of the buffer
956 * we read into down to the filesystem. With the filldir concept
957 * it's not needed for correct information, but the XFS dir2 leaf
958 * code wants an estimate of the buffer size to calculate it's
959 * readahead window and size the buffers used for mapping to
962 * Try to give it an estimate that's good enough, maybe at some
963 * point we can change the ->readdir prototype to include the
964 * buffer size. For now we use the current glibc buffer size.
966 bufsize = (size_t)min_t(loff_t, 32768, ip->i_d.di_size);
968 return xfs_readdir(ip, ctx, bufsize);
974 struct vm_area_struct *vma)
976 vma->vm_ops = &xfs_file_vm_ops;
983 * mmap()d file has taken write protection fault and is being made
984 * writable. We can set the page state up correctly for a writable
985 * page, which means we can do correct delalloc accounting (ENOSPC
986 * checking!) and unwritten extent mapping.
990 struct vm_area_struct *vma,
991 struct vm_fault *vmf)
993 return block_page_mkwrite(vma, vmf, xfs_get_blocks);
997 * This type is designed to indicate the type of offset we would like
998 * to search from page cache for xfs_seek_hole_data().
1006 * Lookup the desired type of offset from the given page.
1008 * On success, return true and the offset argument will point to the
1009 * start of the region that was found. Otherwise this function will
1010 * return false and keep the offset argument unchanged.
1013 xfs_lookup_buffer_offset(
1018 loff_t lastoff = page_offset(page);
1020 struct buffer_head *bh, *head;
1022 bh = head = page_buffers(page);
1025 * Unwritten extents that have data in the page
1026 * cache covering them can be identified by the
1027 * BH_Unwritten state flag. Pages with multiple
1028 * buffers might have a mix of holes, data and
1029 * unwritten extents - any buffer with valid
1030 * data in it should have BH_Uptodate flag set
1033 if (buffer_unwritten(bh) ||
1034 buffer_uptodate(bh)) {
1035 if (type == DATA_OFF)
1038 if (type == HOLE_OFF)
1046 lastoff += bh->b_size;
1047 } while ((bh = bh->b_this_page) != head);
1053 * This routine is called to find out and return a data or hole offset
1054 * from the page cache for unwritten extents according to the desired
1055 * type for xfs_seek_hole_data().
1057 * The argument offset is used to tell where we start to search from the
1058 * page cache. Map is used to figure out the end points of the range to
1061 * Return true if the desired type of offset was found, and the argument
1062 * offset is filled with that address. Otherwise, return false and keep
1066 xfs_find_get_desired_pgoff(
1067 struct inode *inode,
1068 struct xfs_bmbt_irec *map,
1072 struct xfs_inode *ip = XFS_I(inode);
1073 struct xfs_mount *mp = ip->i_mount;
1074 struct pagevec pvec;
1078 loff_t startoff = *offset;
1079 loff_t lastoff = startoff;
1082 pagevec_init(&pvec, 0);
1084 index = startoff >> PAGE_CACHE_SHIFT;
1085 endoff = XFS_FSB_TO_B(mp, map->br_startoff + map->br_blockcount);
1086 end = endoff >> PAGE_CACHE_SHIFT;
1092 want = min_t(pgoff_t, end - index, PAGEVEC_SIZE);
1093 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index,
1096 * No page mapped into given range. If we are searching holes
1097 * and if this is the first time we got into the loop, it means
1098 * that the given offset is landed in a hole, return it.
1100 * If we have already stepped through some block buffers to find
1101 * holes but they all contains data. In this case, the last
1102 * offset is already updated and pointed to the end of the last
1103 * mapped page, if it does not reach the endpoint to search,
1104 * that means there should be a hole between them.
1106 if (nr_pages == 0) {
1107 /* Data search found nothing */
1108 if (type == DATA_OFF)
1111 ASSERT(type == HOLE_OFF);
1112 if (lastoff == startoff || lastoff < endoff) {
1120 * At lease we found one page. If this is the first time we
1121 * step into the loop, and if the first page index offset is
1122 * greater than the given search offset, a hole was found.
1124 if (type == HOLE_OFF && lastoff == startoff &&
1125 lastoff < page_offset(pvec.pages[0])) {
1130 for (i = 0; i < nr_pages; i++) {
1131 struct page *page = pvec.pages[i];
1135 * At this point, the page may be truncated or
1136 * invalidated (changing page->mapping to NULL),
1137 * or even swizzled back from swapper_space to tmpfs
1138 * file mapping. However, page->index will not change
1139 * because we have a reference on the page.
1141 * Searching done if the page index is out of range.
1142 * If the current offset is not reaches the end of
1143 * the specified search range, there should be a hole
1146 if (page->index > end) {
1147 if (type == HOLE_OFF && lastoff < endoff) {
1156 * Page truncated or invalidated(page->mapping == NULL).
1157 * We can freely skip it and proceed to check the next
1160 if (unlikely(page->mapping != inode->i_mapping)) {
1165 if (!page_has_buffers(page)) {
1170 found = xfs_lookup_buffer_offset(page, &b_offset, type);
1173 * The found offset may be less than the start
1174 * point to search if this is the first time to
1177 *offset = max_t(loff_t, startoff, b_offset);
1183 * We either searching data but nothing was found, or
1184 * searching hole but found a data buffer. In either
1185 * case, probably the next page contains the desired
1186 * things, update the last offset to it so.
1188 lastoff = page_offset(page) + PAGE_SIZE;
1193 * The number of returned pages less than our desired, search
1194 * done. In this case, nothing was found for searching data,
1195 * but we found a hole behind the last offset.
1197 if (nr_pages < want) {
1198 if (type == HOLE_OFF) {
1205 index = pvec.pages[i - 1]->index + 1;
1206 pagevec_release(&pvec);
1207 } while (index <= end);
1210 pagevec_release(&pvec);
1220 struct inode *inode = file->f_mapping->host;
1221 struct xfs_inode *ip = XFS_I(inode);
1222 struct xfs_mount *mp = ip->i_mount;
1223 loff_t uninitialized_var(offset);
1225 xfs_fileoff_t fsbno;
1230 if (XFS_FORCED_SHUTDOWN(mp))
1233 lock = xfs_ilock_data_map_shared(ip);
1235 isize = i_size_read(inode);
1236 if (start >= isize) {
1242 * Try to read extents from the first block indicated
1243 * by fsbno to the end block of the file.
1245 fsbno = XFS_B_TO_FSBT(mp, start);
1246 end = XFS_B_TO_FSB(mp, isize);
1249 struct xfs_bmbt_irec map[2];
1253 error = xfs_bmapi_read(ip, fsbno, end - fsbno, map, &nmap,
1258 /* No extents at given offset, must be beyond EOF */
1264 for (i = 0; i < nmap; i++) {
1265 offset = max_t(loff_t, start,
1266 XFS_FSB_TO_B(mp, map[i].br_startoff));
1268 /* Landed in the hole we wanted? */
1269 if (whence == SEEK_HOLE &&
1270 map[i].br_startblock == HOLESTARTBLOCK)
1273 /* Landed in the data extent we wanted? */
1274 if (whence == SEEK_DATA &&
1275 (map[i].br_startblock == DELAYSTARTBLOCK ||
1276 (map[i].br_state == XFS_EXT_NORM &&
1277 !isnullstartblock(map[i].br_startblock))))
1281 * Landed in an unwritten extent, try to search
1282 * for hole or data from page cache.
1284 if (map[i].br_state == XFS_EXT_UNWRITTEN) {
1285 if (xfs_find_get_desired_pgoff(inode, &map[i],
1286 whence == SEEK_HOLE ? HOLE_OFF : DATA_OFF,
1293 * We only received one extent out of the two requested. This
1294 * means we've hit EOF and didn't find what we are looking for.
1298 * If we were looking for a hole, set offset to
1299 * the end of the file (i.e., there is an implicit
1300 * hole at the end of any file).
1302 if (whence == SEEK_HOLE) {
1307 * If we were looking for data, it's nowhere to be found
1309 ASSERT(whence == SEEK_DATA);
1317 * Nothing was found, proceed to the next round of search
1318 * if the next reading offset is not at or beyond EOF.
1320 fsbno = map[i - 1].br_startoff + map[i - 1].br_blockcount;
1321 start = XFS_FSB_TO_B(mp, fsbno);
1322 if (start >= isize) {
1323 if (whence == SEEK_HOLE) {
1327 ASSERT(whence == SEEK_DATA);
1335 * If at this point we have found the hole we wanted, the returned
1336 * offset may be bigger than the file size as it may be aligned to
1337 * page boundary for unwritten extents. We need to deal with this
1338 * situation in particular.
1340 if (whence == SEEK_HOLE)
1341 offset = min_t(loff_t, offset, isize);
1342 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
1345 xfs_iunlock(ip, lock);
1362 return generic_file_llseek(file, offset, whence);
1365 return xfs_seek_hole_data(file, offset, whence);
1371 const struct file_operations xfs_file_operations = {
1372 .llseek = xfs_file_llseek,
1373 .read = new_sync_read,
1374 .write = new_sync_write,
1375 .read_iter = xfs_file_read_iter,
1376 .write_iter = xfs_file_write_iter,
1377 .splice_read = xfs_file_splice_read,
1378 .splice_write = iter_file_splice_write,
1379 .unlocked_ioctl = xfs_file_ioctl,
1380 #ifdef CONFIG_COMPAT
1381 .compat_ioctl = xfs_file_compat_ioctl,
1383 .mmap = xfs_file_mmap,
1384 .open = xfs_file_open,
1385 .release = xfs_file_release,
1386 .fsync = xfs_file_fsync,
1387 .fallocate = xfs_file_fallocate,
1390 const struct file_operations xfs_dir_file_operations = {
1391 .open = xfs_dir_open,
1392 .read = generic_read_dir,
1393 .iterate = xfs_file_readdir,
1394 .llseek = generic_file_llseek,
1395 .unlocked_ioctl = xfs_file_ioctl,
1396 #ifdef CONFIG_COMPAT
1397 .compat_ioctl = xfs_file_compat_ioctl,
1399 .fsync = xfs_dir_fsync,
1402 static const struct vm_operations_struct xfs_file_vm_ops = {
1403 .fault = filemap_fault,
1404 .map_pages = filemap_map_pages,
1405 .page_mkwrite = xfs_vm_page_mkwrite,