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/aio.h>
42 #include <linux/dcache.h>
43 #include <linux/falloc.h>
44 #include <linux/pagevec.h>
46 static const struct vm_operations_struct xfs_file_vm_ops;
49 * Locking primitives for read and write IO paths to ensure we consistently use
50 * and order the inode->i_mutex, ip->i_lock and ip->i_iolock.
57 if (type & XFS_IOLOCK_EXCL)
58 mutex_lock(&VFS_I(ip)->i_mutex);
67 xfs_iunlock(ip, type);
68 if (type & XFS_IOLOCK_EXCL)
69 mutex_unlock(&VFS_I(ip)->i_mutex);
77 xfs_ilock_demote(ip, type);
78 if (type & XFS_IOLOCK_EXCL)
79 mutex_unlock(&VFS_I(ip)->i_mutex);
85 * xfs_iozero clears the specified range of buffer supplied,
86 * and marks all the affected blocks as valid and modified. If
87 * an affected block is not allocated, it will be allocated. If
88 * an affected block is not completely overwritten, and is not
89 * valid before the operation, it will be read from disk before
90 * being partially zeroed.
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;
102 mapping = VFS_I(ip)->i_mapping;
104 unsigned offset, bytes;
107 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
108 bytes = PAGE_CACHE_SIZE - offset;
112 status = pagecache_write_begin(NULL, mapping, pos, bytes,
113 AOP_FLAG_UNINTERRUPTIBLE,
118 zero_user(page, offset, bytes);
120 status = pagecache_write_end(NULL, mapping, pos, bytes, bytes,
122 WARN_ON(status <= 0); /* can't return less than zero! */
132 xfs_update_prealloc_flags(
133 struct xfs_inode *ip,
134 enum xfs_prealloc_flags flags)
136 struct xfs_trans *tp;
139 tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_WRITEID);
140 error = xfs_trans_reserve(tp, &M_RES(ip->i_mount)->tr_writeid, 0, 0);
142 xfs_trans_cancel(tp, 0);
146 xfs_ilock(ip, XFS_ILOCK_EXCL);
147 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
149 if (!(flags & XFS_PREALLOC_INVISIBLE)) {
150 ip->i_d.di_mode &= ~S_ISUID;
151 if (ip->i_d.di_mode & S_IXGRP)
152 ip->i_d.di_mode &= ~S_ISGID;
153 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
156 if (flags & XFS_PREALLOC_SET)
157 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
158 if (flags & XFS_PREALLOC_CLEAR)
159 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
161 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
162 if (flags & XFS_PREALLOC_SYNC)
163 xfs_trans_set_sync(tp);
164 return xfs_trans_commit(tp, 0);
168 * Fsync operations on directories are much simpler than on regular files,
169 * as there is no file data to flush, and thus also no need for explicit
170 * cache flush operations, and there are no non-transaction metadata updates
171 * on directories either.
180 struct xfs_inode *ip = XFS_I(file->f_mapping->host);
181 struct xfs_mount *mp = ip->i_mount;
184 trace_xfs_dir_fsync(ip);
186 xfs_ilock(ip, XFS_ILOCK_SHARED);
187 if (xfs_ipincount(ip))
188 lsn = ip->i_itemp->ili_last_lsn;
189 xfs_iunlock(ip, XFS_ILOCK_SHARED);
193 return _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, NULL);
203 struct inode *inode = file->f_mapping->host;
204 struct xfs_inode *ip = XFS_I(inode);
205 struct xfs_mount *mp = ip->i_mount;
210 trace_xfs_file_fsync(ip);
212 error = filemap_write_and_wait_range(inode->i_mapping, start, end);
216 if (XFS_FORCED_SHUTDOWN(mp))
219 xfs_iflags_clear(ip, XFS_ITRUNCATED);
221 if (mp->m_flags & XFS_MOUNT_BARRIER) {
223 * If we have an RT and/or log subvolume we need to make sure
224 * to flush the write cache the device used for file data
225 * first. This is to ensure newly written file data make
226 * it to disk before logging the new inode size in case of
227 * an extending write.
229 if (XFS_IS_REALTIME_INODE(ip))
230 xfs_blkdev_issue_flush(mp->m_rtdev_targp);
231 else if (mp->m_logdev_targp != mp->m_ddev_targp)
232 xfs_blkdev_issue_flush(mp->m_ddev_targp);
236 * All metadata updates are logged, which means that we just have
237 * to flush the log up to the latest LSN that touched the inode.
239 xfs_ilock(ip, XFS_ILOCK_SHARED);
240 if (xfs_ipincount(ip)) {
242 (ip->i_itemp->ili_fields & ~XFS_ILOG_TIMESTAMP))
243 lsn = ip->i_itemp->ili_last_lsn;
245 xfs_iunlock(ip, XFS_ILOCK_SHARED);
248 error = _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, &log_flushed);
251 * If we only have a single device, and the log force about was
252 * a no-op we might have to flush the data device cache here.
253 * This can only happen for fdatasync/O_DSYNC if we were overwriting
254 * an already allocated file and thus do not have any metadata to
257 if ((mp->m_flags & XFS_MOUNT_BARRIER) &&
258 mp->m_logdev_targp == mp->m_ddev_targp &&
259 !XFS_IS_REALTIME_INODE(ip) &&
261 xfs_blkdev_issue_flush(mp->m_ddev_targp);
271 struct file *file = iocb->ki_filp;
272 struct inode *inode = file->f_mapping->host;
273 struct xfs_inode *ip = XFS_I(inode);
274 struct xfs_mount *mp = ip->i_mount;
275 size_t size = iov_iter_count(to);
279 loff_t pos = iocb->ki_pos;
281 XFS_STATS_INC(xs_read_calls);
283 if (unlikely(file->f_flags & O_DIRECT))
284 ioflags |= XFS_IO_ISDIRECT;
285 if (file->f_mode & FMODE_NOCMTIME)
286 ioflags |= XFS_IO_INVIS;
288 if (unlikely(ioflags & XFS_IO_ISDIRECT)) {
289 xfs_buftarg_t *target =
290 XFS_IS_REALTIME_INODE(ip) ?
291 mp->m_rtdev_targp : mp->m_ddev_targp;
292 /* DIO must be aligned to device logical sector size */
293 if ((pos | size) & target->bt_logical_sectormask) {
294 if (pos == i_size_read(inode))
300 n = mp->m_super->s_maxbytes - pos;
301 if (n <= 0 || size == 0)
307 if (XFS_FORCED_SHUTDOWN(mp))
311 * Locking is a bit tricky here. If we take an exclusive lock
312 * for direct IO, we effectively serialise all new concurrent
313 * read IO to this file and block it behind IO that is currently in
314 * progress because IO in progress holds the IO lock shared. We only
315 * need to hold the lock exclusive to blow away the page cache, so
316 * only take lock exclusively if the page cache needs invalidation.
317 * This allows the normal direct IO case of no page cache pages to
318 * proceeed concurrently without serialisation.
320 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
321 if ((ioflags & XFS_IO_ISDIRECT) && inode->i_mapping->nrpages) {
322 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
323 xfs_rw_ilock(ip, XFS_IOLOCK_EXCL);
325 if (inode->i_mapping->nrpages) {
326 ret = filemap_write_and_wait_range(
327 VFS_I(ip)->i_mapping,
328 pos, pos + size - 1);
330 xfs_rw_iunlock(ip, XFS_IOLOCK_EXCL);
335 * Invalidate whole pages. This can return an error if
336 * we fail to invalidate a page, but this should never
337 * happen on XFS. Warn if it does fail.
339 ret = invalidate_inode_pages2_range(VFS_I(ip)->i_mapping,
340 pos >> PAGE_CACHE_SHIFT,
341 (pos + size - 1) >> PAGE_CACHE_SHIFT);
345 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
348 trace_xfs_file_read(ip, size, pos, ioflags);
350 ret = generic_file_read_iter(iocb, to);
352 XFS_STATS_ADD(xs_read_bytes, ret);
354 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
359 xfs_file_splice_read(
362 struct pipe_inode_info *pipe,
366 struct xfs_inode *ip = XFS_I(infilp->f_mapping->host);
370 XFS_STATS_INC(xs_read_calls);
372 if (infilp->f_mode & FMODE_NOCMTIME)
373 ioflags |= XFS_IO_INVIS;
375 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
378 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
380 trace_xfs_file_splice_read(ip, count, *ppos, ioflags);
382 ret = generic_file_splice_read(infilp, ppos, pipe, count, flags);
384 XFS_STATS_ADD(xs_read_bytes, ret);
386 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
391 * This routine is called to handle zeroing any space in the last block of the
392 * file that is beyond the EOF. We do this since the size is being increased
393 * without writing anything to that block and we don't want to read the
394 * garbage on the disk.
396 STATIC int /* error (positive) */
398 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;
428 return xfs_iozero(ip, isize, zero_len);
432 * Zero any on disk space between the current EOF and the new, larger EOF.
434 * This handles the normal case of zeroing the remainder of the last block in
435 * the file and the unusual case of zeroing blocks out beyond the size of the
436 * file. This second case only happens with fixed size extents and when the
437 * system crashes before the inode size was updated but after blocks were
440 * Expects the iolock to be held exclusive, and will take the ilock internally.
442 int /* error (positive) */
444 struct xfs_inode *ip,
445 xfs_off_t offset, /* starting I/O offset */
446 xfs_fsize_t isize) /* current inode size */
448 struct xfs_mount *mp = ip->i_mount;
449 xfs_fileoff_t start_zero_fsb;
450 xfs_fileoff_t end_zero_fsb;
451 xfs_fileoff_t zero_count_fsb;
452 xfs_fileoff_t last_fsb;
453 xfs_fileoff_t zero_off;
454 xfs_fsize_t zero_len;
457 struct xfs_bmbt_irec imap;
459 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
460 ASSERT(offset > isize);
463 * First handle zeroing the block on which isize resides.
465 * We only zero a part of that block so it is handled specially.
467 if (XFS_B_FSB_OFFSET(mp, isize) != 0) {
468 error = xfs_zero_last_block(ip, offset, isize);
474 * Calculate the range between the new size and the old where blocks
475 * needing to be zeroed may exist.
477 * To get the block where the last byte in the file currently resides,
478 * we need to subtract one from the size and truncate back to a block
479 * boundary. We subtract 1 in case the size is exactly on a block
482 last_fsb = isize ? XFS_B_TO_FSBT(mp, isize - 1) : (xfs_fileoff_t)-1;
483 start_zero_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
484 end_zero_fsb = XFS_B_TO_FSBT(mp, offset - 1);
485 ASSERT((xfs_sfiloff_t)last_fsb < (xfs_sfiloff_t)start_zero_fsb);
486 if (last_fsb == end_zero_fsb) {
488 * The size was only incremented on its last block.
489 * We took care of that above, so just return.
494 ASSERT(start_zero_fsb <= end_zero_fsb);
495 while (start_zero_fsb <= end_zero_fsb) {
497 zero_count_fsb = end_zero_fsb - start_zero_fsb + 1;
499 xfs_ilock(ip, XFS_ILOCK_EXCL);
500 error = xfs_bmapi_read(ip, start_zero_fsb, zero_count_fsb,
502 xfs_iunlock(ip, XFS_ILOCK_EXCL);
508 if (imap.br_state == XFS_EXT_UNWRITTEN ||
509 imap.br_startblock == HOLESTARTBLOCK) {
510 start_zero_fsb = imap.br_startoff + imap.br_blockcount;
511 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
516 * There are blocks we need to zero.
518 zero_off = XFS_FSB_TO_B(mp, start_zero_fsb);
519 zero_len = XFS_FSB_TO_B(mp, imap.br_blockcount);
521 if ((zero_off + zero_len) > offset)
522 zero_len = offset - zero_off;
524 error = xfs_iozero(ip, zero_off, zero_len);
528 start_zero_fsb = imap.br_startoff + imap.br_blockcount;
529 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
536 * Common pre-write limit and setup checks.
538 * Called with the iolocked held either shared and exclusive according to
539 * @iolock, and returns with it held. Might upgrade the iolock to exclusive
540 * if called for a direct write beyond i_size.
543 xfs_file_aio_write_checks(
549 struct inode *inode = file->f_mapping->host;
550 struct xfs_inode *ip = XFS_I(inode);
554 error = generic_write_checks(file, pos, count, S_ISBLK(inode->i_mode));
558 error = xfs_break_layouts(inode, iolock);
563 * If the offset is beyond the size of the file, we need to zero any
564 * blocks that fall between the existing EOF and the start of this
565 * write. If zeroing is needed and we are currently holding the
566 * iolock shared, we need to update it to exclusive which implies
567 * having to redo all checks before.
569 if (*pos > i_size_read(inode)) {
570 if (*iolock == XFS_IOLOCK_SHARED) {
571 xfs_rw_iunlock(ip, *iolock);
572 *iolock = XFS_IOLOCK_EXCL;
573 xfs_rw_ilock(ip, *iolock);
576 error = xfs_zero_eof(ip, *pos, i_size_read(inode));
582 * Updating the timestamps will grab the ilock again from
583 * xfs_fs_dirty_inode, so we have to call it after dropping the
584 * lock above. Eventually we should look into a way to avoid
585 * the pointless lock roundtrip.
587 if (likely(!(file->f_mode & FMODE_NOCMTIME))) {
588 error = file_update_time(file);
594 * If we're writing the file then make sure to clear the setuid and
595 * setgid bits if the process is not being run by root. This keeps
596 * people from modifying setuid and setgid binaries.
598 return file_remove_suid(file);
602 * xfs_file_dio_aio_write - handle direct IO writes
604 * Lock the inode appropriately to prepare for and issue a direct IO write.
605 * By separating it from the buffered write path we remove all the tricky to
606 * follow locking changes and looping.
608 * If there are cached pages or we're extending the file, we need IOLOCK_EXCL
609 * until we're sure the bytes at the new EOF have been zeroed and/or the cached
610 * pages are flushed out.
612 * In most cases the direct IO writes will be done holding IOLOCK_SHARED
613 * allowing them to be done in parallel with reads and other direct IO writes.
614 * However, if the IO is not aligned to filesystem blocks, the direct IO layer
615 * needs to do sub-block zeroing and that requires serialisation against other
616 * direct IOs to the same block. In this case we need to serialise the
617 * submission of the unaligned IOs so that we don't get racing block zeroing in
618 * the dio layer. To avoid the problem with aio, we also need to wait for
619 * outstanding IOs to complete so that unwritten extent conversion is completed
620 * before we try to map the overlapping block. This is currently implemented by
621 * hitting it with a big hammer (i.e. inode_dio_wait()).
623 * Returns with locks held indicated by @iolock and errors indicated by
624 * negative return values.
627 xfs_file_dio_aio_write(
629 struct iov_iter *from)
631 struct file *file = iocb->ki_filp;
632 struct address_space *mapping = file->f_mapping;
633 struct inode *inode = mapping->host;
634 struct xfs_inode *ip = XFS_I(inode);
635 struct xfs_mount *mp = ip->i_mount;
637 int unaligned_io = 0;
639 size_t count = iov_iter_count(from);
640 loff_t pos = iocb->ki_pos;
641 struct xfs_buftarg *target = XFS_IS_REALTIME_INODE(ip) ?
642 mp->m_rtdev_targp : mp->m_ddev_targp;
644 /* DIO must be aligned to device logical sector size */
645 if ((pos | count) & target->bt_logical_sectormask)
648 /* "unaligned" here means not aligned to a filesystem block */
649 if ((pos & mp->m_blockmask) || ((pos + count) & mp->m_blockmask))
653 * We don't need to take an exclusive lock unless there page cache needs
654 * to be invalidated or unaligned IO is being executed. We don't need to
655 * consider the EOF extension case here because
656 * xfs_file_aio_write_checks() will relock the inode as necessary for
657 * EOF zeroing cases and fill out the new inode size as appropriate.
659 if (unaligned_io || mapping->nrpages)
660 iolock = XFS_IOLOCK_EXCL;
662 iolock = XFS_IOLOCK_SHARED;
663 xfs_rw_ilock(ip, iolock);
666 * Recheck if there are cached pages that need invalidate after we got
667 * the iolock to protect against other threads adding new pages while
668 * we were waiting for the iolock.
670 if (mapping->nrpages && iolock == XFS_IOLOCK_SHARED) {
671 xfs_rw_iunlock(ip, iolock);
672 iolock = XFS_IOLOCK_EXCL;
673 xfs_rw_ilock(ip, iolock);
676 ret = xfs_file_aio_write_checks(file, &pos, &count, &iolock);
679 iov_iter_truncate(from, count);
681 if (mapping->nrpages) {
682 ret = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
683 pos, pos + count - 1);
687 * Invalidate whole pages. This can return an error if
688 * we fail to invalidate a page, but this should never
689 * happen on XFS. Warn if it does fail.
691 ret = invalidate_inode_pages2_range(VFS_I(ip)->i_mapping,
692 pos >> PAGE_CACHE_SHIFT,
693 (pos + count - 1) >> PAGE_CACHE_SHIFT);
699 * If we are doing unaligned IO, wait for all other IO to drain,
700 * otherwise demote the lock if we had to flush cached pages
703 inode_dio_wait(inode);
704 else if (iolock == XFS_IOLOCK_EXCL) {
705 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
706 iolock = XFS_IOLOCK_SHARED;
709 trace_xfs_file_direct_write(ip, count, iocb->ki_pos, 0);
710 ret = generic_file_direct_write(iocb, from, pos);
713 xfs_rw_iunlock(ip, iolock);
715 /* No fallback to buffered IO on errors for XFS. */
716 ASSERT(ret < 0 || ret == count);
721 xfs_file_buffered_aio_write(
723 struct iov_iter *from)
725 struct file *file = iocb->ki_filp;
726 struct address_space *mapping = file->f_mapping;
727 struct inode *inode = mapping->host;
728 struct xfs_inode *ip = XFS_I(inode);
731 int iolock = XFS_IOLOCK_EXCL;
732 loff_t pos = iocb->ki_pos;
733 size_t count = iov_iter_count(from);
735 xfs_rw_ilock(ip, iolock);
737 ret = xfs_file_aio_write_checks(file, &pos, &count, &iolock);
741 iov_iter_truncate(from, count);
742 /* We can write back this queue in page reclaim */
743 current->backing_dev_info = inode_to_bdi(inode);
746 trace_xfs_file_buffered_write(ip, count, iocb->ki_pos, 0);
747 ret = generic_perform_write(file, from, pos);
748 if (likely(ret >= 0))
749 iocb->ki_pos = pos + ret;
752 * If we hit a space limit, try to free up some lingering preallocated
753 * space before returning an error. In the case of ENOSPC, first try to
754 * write back all dirty inodes to free up some of the excess reserved
755 * metadata space. This reduces the chances that the eofblocks scan
756 * waits on dirty mappings. Since xfs_flush_inodes() is serialized, this
757 * also behaves as a filter to prevent too many eofblocks scans from
758 * running at the same time.
760 if (ret == -EDQUOT && !enospc) {
761 enospc = xfs_inode_free_quota_eofblocks(ip);
764 } else if (ret == -ENOSPC && !enospc) {
765 struct xfs_eofblocks eofb = {0};
768 xfs_flush_inodes(ip->i_mount);
769 eofb.eof_scan_owner = ip->i_ino; /* for locking */
770 eofb.eof_flags = XFS_EOF_FLAGS_SYNC;
771 xfs_icache_free_eofblocks(ip->i_mount, &eofb);
775 current->backing_dev_info = NULL;
777 xfs_rw_iunlock(ip, iolock);
784 struct iov_iter *from)
786 struct file *file = iocb->ki_filp;
787 struct address_space *mapping = file->f_mapping;
788 struct inode *inode = mapping->host;
789 struct xfs_inode *ip = XFS_I(inode);
791 size_t ocount = iov_iter_count(from);
793 XFS_STATS_INC(xs_write_calls);
798 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
801 if (unlikely(file->f_flags & O_DIRECT))
802 ret = xfs_file_dio_aio_write(iocb, from);
804 ret = xfs_file_buffered_aio_write(iocb, from);
809 XFS_STATS_ADD(xs_write_bytes, ret);
811 /* Handle various SYNC-type writes */
812 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
826 struct inode *inode = file_inode(file);
827 struct xfs_inode *ip = XFS_I(inode);
829 enum xfs_prealloc_flags flags = 0;
830 uint iolock = XFS_IOLOCK_EXCL;
833 if (!S_ISREG(inode->i_mode))
835 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
836 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE))
839 xfs_ilock(ip, iolock);
840 error = xfs_break_layouts(inode, &iolock);
844 if (mode & FALLOC_FL_PUNCH_HOLE) {
845 error = xfs_free_file_space(ip, offset, len);
848 } else if (mode & FALLOC_FL_COLLAPSE_RANGE) {
849 unsigned blksize_mask = (1 << inode->i_blkbits) - 1;
851 if (offset & blksize_mask || len & blksize_mask) {
857 * There is no need to overlap collapse range with EOF,
858 * in which case it is effectively a truncate operation
860 if (offset + len >= i_size_read(inode)) {
865 new_size = i_size_read(inode) - len;
867 error = xfs_collapse_file_space(ip, offset, len);
871 flags |= XFS_PREALLOC_SET;
873 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
874 offset + len > i_size_read(inode)) {
875 new_size = offset + len;
876 error = inode_newsize_ok(inode, new_size);
881 if (mode & FALLOC_FL_ZERO_RANGE)
882 error = xfs_zero_file_space(ip, offset, len);
884 error = xfs_alloc_file_space(ip, offset, len,
890 if (file->f_flags & O_DSYNC)
891 flags |= XFS_PREALLOC_SYNC;
893 error = xfs_update_prealloc_flags(ip, flags);
897 /* Change file size if needed */
901 iattr.ia_valid = ATTR_SIZE;
902 iattr.ia_size = new_size;
903 error = xfs_setattr_size(ip, &iattr);
907 xfs_iunlock(ip, iolock);
917 if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
919 if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb)))
929 struct xfs_inode *ip = XFS_I(inode);
933 error = xfs_file_open(inode, file);
938 * If there are any blocks, read-ahead block 0 as we're almost
939 * certain to have the next operation be a read there.
941 mode = xfs_ilock_data_map_shared(ip);
942 if (ip->i_d.di_nextents > 0)
943 xfs_dir3_data_readahead(ip, 0, -1);
944 xfs_iunlock(ip, mode);
953 return xfs_release(XFS_I(inode));
959 struct dir_context *ctx)
961 struct inode *inode = file_inode(file);
962 xfs_inode_t *ip = XFS_I(inode);
966 * The Linux API doesn't pass down the total size of the buffer
967 * we read into down to the filesystem. With the filldir concept
968 * it's not needed for correct information, but the XFS dir2 leaf
969 * code wants an estimate of the buffer size to calculate it's
970 * readahead window and size the buffers used for mapping to
973 * Try to give it an estimate that's good enough, maybe at some
974 * point we can change the ->readdir prototype to include the
975 * buffer size. For now we use the current glibc buffer size.
977 bufsize = (size_t)min_t(loff_t, 32768, ip->i_d.di_size);
979 return xfs_readdir(ip, ctx, bufsize);
985 struct vm_area_struct *vma)
987 vma->vm_ops = &xfs_file_vm_ops;
994 * mmap()d file has taken write protection fault and is being made
995 * writable. We can set the page state up correctly for a writable
996 * page, which means we can do correct delalloc accounting (ENOSPC
997 * checking!) and unwritten extent mapping.
1000 xfs_vm_page_mkwrite(
1001 struct vm_area_struct *vma,
1002 struct vm_fault *vmf)
1004 return block_page_mkwrite(vma, vmf, xfs_get_blocks);
1008 * This type is designed to indicate the type of offset we would like
1009 * to search from page cache for xfs_seek_hole_data().
1017 * Lookup the desired type of offset from the given page.
1019 * On success, return true and the offset argument will point to the
1020 * start of the region that was found. Otherwise this function will
1021 * return false and keep the offset argument unchanged.
1024 xfs_lookup_buffer_offset(
1029 loff_t lastoff = page_offset(page);
1031 struct buffer_head *bh, *head;
1033 bh = head = page_buffers(page);
1036 * Unwritten extents that have data in the page
1037 * cache covering them can be identified by the
1038 * BH_Unwritten state flag. Pages with multiple
1039 * buffers might have a mix of holes, data and
1040 * unwritten extents - any buffer with valid
1041 * data in it should have BH_Uptodate flag set
1044 if (buffer_unwritten(bh) ||
1045 buffer_uptodate(bh)) {
1046 if (type == DATA_OFF)
1049 if (type == HOLE_OFF)
1057 lastoff += bh->b_size;
1058 } while ((bh = bh->b_this_page) != head);
1064 * This routine is called to find out and return a data or hole offset
1065 * from the page cache for unwritten extents according to the desired
1066 * type for xfs_seek_hole_data().
1068 * The argument offset is used to tell where we start to search from the
1069 * page cache. Map is used to figure out the end points of the range to
1072 * Return true if the desired type of offset was found, and the argument
1073 * offset is filled with that address. Otherwise, return false and keep
1077 xfs_find_get_desired_pgoff(
1078 struct inode *inode,
1079 struct xfs_bmbt_irec *map,
1083 struct xfs_inode *ip = XFS_I(inode);
1084 struct xfs_mount *mp = ip->i_mount;
1085 struct pagevec pvec;
1089 loff_t startoff = *offset;
1090 loff_t lastoff = startoff;
1093 pagevec_init(&pvec, 0);
1095 index = startoff >> PAGE_CACHE_SHIFT;
1096 endoff = XFS_FSB_TO_B(mp, map->br_startoff + map->br_blockcount);
1097 end = endoff >> PAGE_CACHE_SHIFT;
1103 want = min_t(pgoff_t, end - index, PAGEVEC_SIZE);
1104 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index,
1107 * No page mapped into given range. If we are searching holes
1108 * and if this is the first time we got into the loop, it means
1109 * that the given offset is landed in a hole, return it.
1111 * If we have already stepped through some block buffers to find
1112 * holes but they all contains data. In this case, the last
1113 * offset is already updated and pointed to the end of the last
1114 * mapped page, if it does not reach the endpoint to search,
1115 * that means there should be a hole between them.
1117 if (nr_pages == 0) {
1118 /* Data search found nothing */
1119 if (type == DATA_OFF)
1122 ASSERT(type == HOLE_OFF);
1123 if (lastoff == startoff || lastoff < endoff) {
1131 * At lease we found one page. If this is the first time we
1132 * step into the loop, and if the first page index offset is
1133 * greater than the given search offset, a hole was found.
1135 if (type == HOLE_OFF && lastoff == startoff &&
1136 lastoff < page_offset(pvec.pages[0])) {
1141 for (i = 0; i < nr_pages; i++) {
1142 struct page *page = pvec.pages[i];
1146 * At this point, the page may be truncated or
1147 * invalidated (changing page->mapping to NULL),
1148 * or even swizzled back from swapper_space to tmpfs
1149 * file mapping. However, page->index will not change
1150 * because we have a reference on the page.
1152 * Searching done if the page index is out of range.
1153 * If the current offset is not reaches the end of
1154 * the specified search range, there should be a hole
1157 if (page->index > end) {
1158 if (type == HOLE_OFF && lastoff < endoff) {
1167 * Page truncated or invalidated(page->mapping == NULL).
1168 * We can freely skip it and proceed to check the next
1171 if (unlikely(page->mapping != inode->i_mapping)) {
1176 if (!page_has_buffers(page)) {
1181 found = xfs_lookup_buffer_offset(page, &b_offset, type);
1184 * The found offset may be less than the start
1185 * point to search if this is the first time to
1188 *offset = max_t(loff_t, startoff, b_offset);
1194 * We either searching data but nothing was found, or
1195 * searching hole but found a data buffer. In either
1196 * case, probably the next page contains the desired
1197 * things, update the last offset to it so.
1199 lastoff = page_offset(page) + PAGE_SIZE;
1204 * The number of returned pages less than our desired, search
1205 * done. In this case, nothing was found for searching data,
1206 * but we found a hole behind the last offset.
1208 if (nr_pages < want) {
1209 if (type == HOLE_OFF) {
1216 index = pvec.pages[i - 1]->index + 1;
1217 pagevec_release(&pvec);
1218 } while (index <= end);
1221 pagevec_release(&pvec);
1231 struct inode *inode = file->f_mapping->host;
1232 struct xfs_inode *ip = XFS_I(inode);
1233 struct xfs_mount *mp = ip->i_mount;
1234 loff_t uninitialized_var(offset);
1236 xfs_fileoff_t fsbno;
1241 if (XFS_FORCED_SHUTDOWN(mp))
1244 lock = xfs_ilock_data_map_shared(ip);
1246 isize = i_size_read(inode);
1247 if (start >= isize) {
1253 * Try to read extents from the first block indicated
1254 * by fsbno to the end block of the file.
1256 fsbno = XFS_B_TO_FSBT(mp, start);
1257 end = XFS_B_TO_FSB(mp, isize);
1260 struct xfs_bmbt_irec map[2];
1264 error = xfs_bmapi_read(ip, fsbno, end - fsbno, map, &nmap,
1269 /* No extents at given offset, must be beyond EOF */
1275 for (i = 0; i < nmap; i++) {
1276 offset = max_t(loff_t, start,
1277 XFS_FSB_TO_B(mp, map[i].br_startoff));
1279 /* Landed in the hole we wanted? */
1280 if (whence == SEEK_HOLE &&
1281 map[i].br_startblock == HOLESTARTBLOCK)
1284 /* Landed in the data extent we wanted? */
1285 if (whence == SEEK_DATA &&
1286 (map[i].br_startblock == DELAYSTARTBLOCK ||
1287 (map[i].br_state == XFS_EXT_NORM &&
1288 !isnullstartblock(map[i].br_startblock))))
1292 * Landed in an unwritten extent, try to search
1293 * for hole or data from page cache.
1295 if (map[i].br_state == XFS_EXT_UNWRITTEN) {
1296 if (xfs_find_get_desired_pgoff(inode, &map[i],
1297 whence == SEEK_HOLE ? HOLE_OFF : DATA_OFF,
1304 * We only received one extent out of the two requested. This
1305 * means we've hit EOF and didn't find what we are looking for.
1309 * If we were looking for a hole, set offset to
1310 * the end of the file (i.e., there is an implicit
1311 * hole at the end of any file).
1313 if (whence == SEEK_HOLE) {
1318 * If we were looking for data, it's nowhere to be found
1320 ASSERT(whence == SEEK_DATA);
1328 * Nothing was found, proceed to the next round of search
1329 * if the next reading offset is not at or beyond EOF.
1331 fsbno = map[i - 1].br_startoff + map[i - 1].br_blockcount;
1332 start = XFS_FSB_TO_B(mp, fsbno);
1333 if (start >= isize) {
1334 if (whence == SEEK_HOLE) {
1338 ASSERT(whence == SEEK_DATA);
1346 * If at this point we have found the hole we wanted, the returned
1347 * offset may be bigger than the file size as it may be aligned to
1348 * page boundary for unwritten extents. We need to deal with this
1349 * situation in particular.
1351 if (whence == SEEK_HOLE)
1352 offset = min_t(loff_t, offset, isize);
1353 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
1356 xfs_iunlock(ip, lock);
1373 return generic_file_llseek(file, offset, whence);
1376 return xfs_seek_hole_data(file, offset, whence);
1382 const struct file_operations xfs_file_operations = {
1383 .llseek = xfs_file_llseek,
1384 .read = new_sync_read,
1385 .write = new_sync_write,
1386 .read_iter = xfs_file_read_iter,
1387 .write_iter = xfs_file_write_iter,
1388 .splice_read = xfs_file_splice_read,
1389 .splice_write = iter_file_splice_write,
1390 .unlocked_ioctl = xfs_file_ioctl,
1391 #ifdef CONFIG_COMPAT
1392 .compat_ioctl = xfs_file_compat_ioctl,
1394 .mmap = xfs_file_mmap,
1395 .open = xfs_file_open,
1396 .release = xfs_file_release,
1397 .fsync = xfs_file_fsync,
1398 .fallocate = xfs_file_fallocate,
1401 const struct file_operations xfs_dir_file_operations = {
1402 .open = xfs_dir_open,
1403 .read = generic_read_dir,
1404 .iterate = xfs_file_readdir,
1405 .llseek = generic_file_llseek,
1406 .unlocked_ioctl = xfs_file_ioctl,
1407 #ifdef CONFIG_COMPAT
1408 .compat_ioctl = xfs_file_compat_ioctl,
1410 .fsync = xfs_dir_fsync,
1413 static const struct vm_operations_struct xfs_file_vm_ops = {
1414 .fault = filemap_fault,
1415 .map_pages = filemap_map_pages,
1416 .page_mkwrite = xfs_vm_page_mkwrite,