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/aio.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(file->f_flags & O_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,
401 struct xfs_mount *mp = ip->i_mount;
402 xfs_fileoff_t last_fsb = XFS_B_TO_FSBT(mp, isize);
403 int zero_offset = XFS_B_FSB_OFFSET(mp, isize);
407 struct xfs_bmbt_irec imap;
409 xfs_ilock(ip, XFS_ILOCK_EXCL);
410 error = xfs_bmapi_read(ip, last_fsb, 1, &imap, &nimaps, 0);
411 xfs_iunlock(ip, XFS_ILOCK_EXCL);
418 * If the block underlying isize is just a hole, then there
419 * is nothing to zero.
421 if (imap.br_startblock == HOLESTARTBLOCK)
424 zero_len = mp->m_sb.sb_blocksize - zero_offset;
425 if (isize + zero_len > offset)
426 zero_len = offset - isize;
427 return xfs_iozero(ip, isize, zero_len);
431 * Zero any on disk space between the current EOF and the new, larger EOF.
433 * This handles the normal case of zeroing the remainder of the last block in
434 * the file and the unusual case of zeroing blocks out beyond the size of the
435 * file. This second case only happens with fixed size extents and when the
436 * system crashes before the inode size was updated but after blocks were
439 * Expects the iolock to be held exclusive, and will take the ilock internally.
441 int /* error (positive) */
443 struct xfs_inode *ip,
444 xfs_off_t offset, /* starting I/O offset */
445 xfs_fsize_t isize) /* current inode size */
447 struct xfs_mount *mp = ip->i_mount;
448 xfs_fileoff_t start_zero_fsb;
449 xfs_fileoff_t end_zero_fsb;
450 xfs_fileoff_t zero_count_fsb;
451 xfs_fileoff_t last_fsb;
452 xfs_fileoff_t zero_off;
453 xfs_fsize_t zero_len;
456 struct xfs_bmbt_irec imap;
458 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
459 ASSERT(offset > isize);
462 * First handle zeroing the block on which isize resides.
464 * We only zero a part of that block so it is handled specially.
466 if (XFS_B_FSB_OFFSET(mp, isize) != 0) {
467 error = xfs_zero_last_block(ip, offset, isize);
473 * Calculate the range between the new size and the old where blocks
474 * needing to be zeroed may exist.
476 * To get the block where the last byte in the file currently resides,
477 * we need to subtract one from the size and truncate back to a block
478 * boundary. We subtract 1 in case the size is exactly on a block
481 last_fsb = isize ? XFS_B_TO_FSBT(mp, isize - 1) : (xfs_fileoff_t)-1;
482 start_zero_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
483 end_zero_fsb = XFS_B_TO_FSBT(mp, offset - 1);
484 ASSERT((xfs_sfiloff_t)last_fsb < (xfs_sfiloff_t)start_zero_fsb);
485 if (last_fsb == end_zero_fsb) {
487 * The size was only incremented on its last block.
488 * We took care of that above, so just return.
493 ASSERT(start_zero_fsb <= end_zero_fsb);
494 while (start_zero_fsb <= end_zero_fsb) {
496 zero_count_fsb = end_zero_fsb - start_zero_fsb + 1;
498 xfs_ilock(ip, XFS_ILOCK_EXCL);
499 error = xfs_bmapi_read(ip, start_zero_fsb, zero_count_fsb,
501 xfs_iunlock(ip, XFS_ILOCK_EXCL);
507 if (imap.br_state == XFS_EXT_UNWRITTEN ||
508 imap.br_startblock == HOLESTARTBLOCK) {
509 start_zero_fsb = imap.br_startoff + imap.br_blockcount;
510 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
515 * There are blocks we need to zero.
517 zero_off = XFS_FSB_TO_B(mp, start_zero_fsb);
518 zero_len = XFS_FSB_TO_B(mp, imap.br_blockcount);
520 if ((zero_off + zero_len) > offset)
521 zero_len = offset - zero_off;
523 error = xfs_iozero(ip, zero_off, zero_len);
527 start_zero_fsb = imap.br_startoff + imap.br_blockcount;
528 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
535 * Common pre-write limit and setup checks.
537 * Called with the iolocked held either shared and exclusive according to
538 * @iolock, and returns with it held. Might upgrade the iolock to exclusive
539 * if called for a direct write beyond i_size.
542 xfs_file_aio_write_checks(
548 struct inode *inode = file->f_mapping->host;
549 struct xfs_inode *ip = XFS_I(inode);
553 error = generic_write_checks(file, pos, count, S_ISBLK(inode->i_mode));
558 * If the offset is beyond the size of the file, we need to zero any
559 * blocks that fall between the existing EOF and the start of this
560 * write. If zeroing is needed and we are currently holding the
561 * iolock shared, we need to update it to exclusive which implies
562 * having to redo all checks before.
564 if (*pos > i_size_read(inode)) {
565 if (*iolock == XFS_IOLOCK_SHARED) {
566 xfs_rw_iunlock(ip, *iolock);
567 *iolock = XFS_IOLOCK_EXCL;
568 xfs_rw_ilock(ip, *iolock);
571 error = xfs_zero_eof(ip, *pos, i_size_read(inode));
577 * Updating the timestamps will grab the ilock again from
578 * xfs_fs_dirty_inode, so we have to call it after dropping the
579 * lock above. Eventually we should look into a way to avoid
580 * the pointless lock roundtrip.
582 if (likely(!(file->f_mode & FMODE_NOCMTIME))) {
583 error = file_update_time(file);
589 * If we're writing the file then make sure to clear the setuid and
590 * setgid bits if the process is not being run by root. This keeps
591 * people from modifying setuid and setgid binaries.
593 return file_remove_suid(file);
597 * xfs_file_dio_aio_write - handle direct IO writes
599 * Lock the inode appropriately to prepare for and issue a direct IO write.
600 * By separating it from the buffered write path we remove all the tricky to
601 * follow locking changes and looping.
603 * If there are cached pages or we're extending the file, we need IOLOCK_EXCL
604 * until we're sure the bytes at the new EOF have been zeroed and/or the cached
605 * pages are flushed out.
607 * In most cases the direct IO writes will be done holding IOLOCK_SHARED
608 * allowing them to be done in parallel with reads and other direct IO writes.
609 * However, if the IO is not aligned to filesystem blocks, the direct IO layer
610 * needs to do sub-block zeroing and that requires serialisation against other
611 * direct IOs to the same block. In this case we need to serialise the
612 * submission of the unaligned IOs so that we don't get racing block zeroing in
613 * the dio layer. To avoid the problem with aio, we also need to wait for
614 * outstanding IOs to complete so that unwritten extent conversion is completed
615 * before we try to map the overlapping block. This is currently implemented by
616 * hitting it with a big hammer (i.e. inode_dio_wait()).
618 * Returns with locks held indicated by @iolock and errors indicated by
619 * negative return values.
622 xfs_file_dio_aio_write(
624 struct iov_iter *from)
626 struct file *file = iocb->ki_filp;
627 struct address_space *mapping = file->f_mapping;
628 struct inode *inode = mapping->host;
629 struct xfs_inode *ip = XFS_I(inode);
630 struct xfs_mount *mp = ip->i_mount;
632 int unaligned_io = 0;
634 size_t count = iov_iter_count(from);
635 loff_t pos = iocb->ki_pos;
636 struct xfs_buftarg *target = XFS_IS_REALTIME_INODE(ip) ?
637 mp->m_rtdev_targp : mp->m_ddev_targp;
639 /* DIO must be aligned to device logical sector size */
640 if ((pos | count) & target->bt_logical_sectormask)
643 /* "unaligned" here means not aligned to a filesystem block */
644 if ((pos & mp->m_blockmask) || ((pos + count) & mp->m_blockmask))
648 * We don't need to take an exclusive lock unless there page cache needs
649 * to be invalidated or unaligned IO is being executed. We don't need to
650 * consider the EOF extension case here because
651 * xfs_file_aio_write_checks() will relock the inode as necessary for
652 * EOF zeroing cases and fill out the new inode size as appropriate.
654 if (unaligned_io || mapping->nrpages)
655 iolock = XFS_IOLOCK_EXCL;
657 iolock = XFS_IOLOCK_SHARED;
658 xfs_rw_ilock(ip, iolock);
661 * Recheck if there are cached pages that need invalidate after we got
662 * the iolock to protect against other threads adding new pages while
663 * we were waiting for the iolock.
665 if (mapping->nrpages && iolock == XFS_IOLOCK_SHARED) {
666 xfs_rw_iunlock(ip, iolock);
667 iolock = XFS_IOLOCK_EXCL;
668 xfs_rw_ilock(ip, iolock);
671 ret = xfs_file_aio_write_checks(file, &pos, &count, &iolock);
674 iov_iter_truncate(from, count);
676 if (mapping->nrpages) {
677 ret = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
678 pos, pos + count - 1);
682 * Invalidate whole pages. This can return an error if
683 * we fail to invalidate a page, but this should never
684 * happen on XFS. Warn if it does fail.
686 ret = invalidate_inode_pages2_range(VFS_I(ip)->i_mapping,
687 pos >> PAGE_CACHE_SHIFT,
688 (pos + count - 1) >> PAGE_CACHE_SHIFT);
694 * If we are doing unaligned IO, wait for all other IO to drain,
695 * otherwise demote the lock if we had to flush cached pages
698 inode_dio_wait(inode);
699 else if (iolock == XFS_IOLOCK_EXCL) {
700 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
701 iolock = XFS_IOLOCK_SHARED;
704 trace_xfs_file_direct_write(ip, count, iocb->ki_pos, 0);
705 ret = generic_file_direct_write(iocb, from, pos);
708 xfs_rw_iunlock(ip, iolock);
710 /* No fallback to buffered IO on errors for XFS. */
711 ASSERT(ret < 0 || ret == count);
716 xfs_file_buffered_aio_write(
718 struct iov_iter *from)
720 struct file *file = iocb->ki_filp;
721 struct address_space *mapping = file->f_mapping;
722 struct inode *inode = mapping->host;
723 struct xfs_inode *ip = XFS_I(inode);
726 int iolock = XFS_IOLOCK_EXCL;
727 loff_t pos = iocb->ki_pos;
728 size_t count = iov_iter_count(from);
730 xfs_rw_ilock(ip, iolock);
732 ret = xfs_file_aio_write_checks(file, &pos, &count, &iolock);
736 iov_iter_truncate(from, count);
737 /* We can write back this queue in page reclaim */
738 current->backing_dev_info = inode_to_bdi(inode);
741 trace_xfs_file_buffered_write(ip, count, iocb->ki_pos, 0);
742 ret = generic_perform_write(file, from, pos);
743 if (likely(ret >= 0))
744 iocb->ki_pos = pos + ret;
747 * If we hit a space limit, try to free up some lingering preallocated
748 * space before returning an error. In the case of ENOSPC, first try to
749 * write back all dirty inodes to free up some of the excess reserved
750 * metadata space. This reduces the chances that the eofblocks scan
751 * waits on dirty mappings. Since xfs_flush_inodes() is serialized, this
752 * also behaves as a filter to prevent too many eofblocks scans from
753 * running at the same time.
755 if (ret == -EDQUOT && !enospc) {
756 enospc = xfs_inode_free_quota_eofblocks(ip);
759 } else if (ret == -ENOSPC && !enospc) {
760 struct xfs_eofblocks eofb = {0};
763 xfs_flush_inodes(ip->i_mount);
764 eofb.eof_scan_owner = ip->i_ino; /* for locking */
765 eofb.eof_flags = XFS_EOF_FLAGS_SYNC;
766 xfs_icache_free_eofblocks(ip->i_mount, &eofb);
770 current->backing_dev_info = NULL;
772 xfs_rw_iunlock(ip, iolock);
779 struct iov_iter *from)
781 struct file *file = iocb->ki_filp;
782 struct address_space *mapping = file->f_mapping;
783 struct inode *inode = mapping->host;
784 struct xfs_inode *ip = XFS_I(inode);
786 size_t ocount = iov_iter_count(from);
788 XFS_STATS_INC(xs_write_calls);
793 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
796 if (unlikely(file->f_flags & O_DIRECT))
797 ret = xfs_file_dio_aio_write(iocb, from);
799 ret = xfs_file_buffered_aio_write(iocb, from);
804 XFS_STATS_ADD(xs_write_bytes, ret);
806 /* Handle various SYNC-type writes */
807 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
821 struct inode *inode = file_inode(file);
822 struct xfs_inode *ip = XFS_I(inode);
824 enum xfs_prealloc_flags flags = 0;
827 if (!S_ISREG(inode->i_mode))
829 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
830 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE))
833 xfs_ilock(ip, XFS_IOLOCK_EXCL);
834 if (mode & FALLOC_FL_PUNCH_HOLE) {
835 error = xfs_free_file_space(ip, offset, len);
838 } else if (mode & FALLOC_FL_COLLAPSE_RANGE) {
839 unsigned blksize_mask = (1 << inode->i_blkbits) - 1;
841 if (offset & blksize_mask || len & blksize_mask) {
847 * There is no need to overlap collapse range with EOF,
848 * in which case it is effectively a truncate operation
850 if (offset + len >= i_size_read(inode)) {
855 new_size = i_size_read(inode) - len;
857 error = xfs_collapse_file_space(ip, offset, len);
861 flags |= XFS_PREALLOC_SET;
863 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
864 offset + len > i_size_read(inode)) {
865 new_size = offset + len;
866 error = inode_newsize_ok(inode, new_size);
871 if (mode & FALLOC_FL_ZERO_RANGE)
872 error = xfs_zero_file_space(ip, offset, len);
874 error = xfs_alloc_file_space(ip, offset, len,
880 if (file->f_flags & O_DSYNC)
881 flags |= XFS_PREALLOC_SYNC;
883 error = xfs_update_prealloc_flags(ip, flags);
887 /* Change file size if needed */
891 iattr.ia_valid = ATTR_SIZE;
892 iattr.ia_size = new_size;
893 error = xfs_setattr_size(ip, &iattr);
897 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
907 if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
909 if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb)))
919 struct xfs_inode *ip = XFS_I(inode);
923 error = xfs_file_open(inode, file);
928 * If there are any blocks, read-ahead block 0 as we're almost
929 * certain to have the next operation be a read there.
931 mode = xfs_ilock_data_map_shared(ip);
932 if (ip->i_d.di_nextents > 0)
933 xfs_dir3_data_readahead(ip, 0, -1);
934 xfs_iunlock(ip, mode);
943 return xfs_release(XFS_I(inode));
949 struct dir_context *ctx)
951 struct inode *inode = file_inode(file);
952 xfs_inode_t *ip = XFS_I(inode);
956 * The Linux API doesn't pass down the total size of the buffer
957 * we read into down to the filesystem. With the filldir concept
958 * it's not needed for correct information, but the XFS dir2 leaf
959 * code wants an estimate of the buffer size to calculate it's
960 * readahead window and size the buffers used for mapping to
963 * Try to give it an estimate that's good enough, maybe at some
964 * point we can change the ->readdir prototype to include the
965 * buffer size. For now we use the current glibc buffer size.
967 bufsize = (size_t)min_t(loff_t, 32768, ip->i_d.di_size);
969 return xfs_readdir(ip, ctx, bufsize);
975 struct vm_area_struct *vma)
977 vma->vm_ops = &xfs_file_vm_ops;
984 * mmap()d file has taken write protection fault and is being made
985 * writable. We can set the page state up correctly for a writable
986 * page, which means we can do correct delalloc accounting (ENOSPC
987 * checking!) and unwritten extent mapping.
991 struct vm_area_struct *vma,
992 struct vm_fault *vmf)
994 return block_page_mkwrite(vma, vmf, xfs_get_blocks);
998 * This type is designed to indicate the type of offset we would like
999 * to search from page cache for xfs_seek_hole_data().
1007 * Lookup the desired type of offset from the given page.
1009 * On success, return true and the offset argument will point to the
1010 * start of the region that was found. Otherwise this function will
1011 * return false and keep the offset argument unchanged.
1014 xfs_lookup_buffer_offset(
1019 loff_t lastoff = page_offset(page);
1021 struct buffer_head *bh, *head;
1023 bh = head = page_buffers(page);
1026 * Unwritten extents that have data in the page
1027 * cache covering them can be identified by the
1028 * BH_Unwritten state flag. Pages with multiple
1029 * buffers might have a mix of holes, data and
1030 * unwritten extents - any buffer with valid
1031 * data in it should have BH_Uptodate flag set
1034 if (buffer_unwritten(bh) ||
1035 buffer_uptodate(bh)) {
1036 if (type == DATA_OFF)
1039 if (type == HOLE_OFF)
1047 lastoff += bh->b_size;
1048 } while ((bh = bh->b_this_page) != head);
1054 * This routine is called to find out and return a data or hole offset
1055 * from the page cache for unwritten extents according to the desired
1056 * type for xfs_seek_hole_data().
1058 * The argument offset is used to tell where we start to search from the
1059 * page cache. Map is used to figure out the end points of the range to
1062 * Return true if the desired type of offset was found, and the argument
1063 * offset is filled with that address. Otherwise, return false and keep
1067 xfs_find_get_desired_pgoff(
1068 struct inode *inode,
1069 struct xfs_bmbt_irec *map,
1073 struct xfs_inode *ip = XFS_I(inode);
1074 struct xfs_mount *mp = ip->i_mount;
1075 struct pagevec pvec;
1079 loff_t startoff = *offset;
1080 loff_t lastoff = startoff;
1083 pagevec_init(&pvec, 0);
1085 index = startoff >> PAGE_CACHE_SHIFT;
1086 endoff = XFS_FSB_TO_B(mp, map->br_startoff + map->br_blockcount);
1087 end = endoff >> PAGE_CACHE_SHIFT;
1093 want = min_t(pgoff_t, end - index, PAGEVEC_SIZE);
1094 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index,
1097 * No page mapped into given range. If we are searching holes
1098 * and if this is the first time we got into the loop, it means
1099 * that the given offset is landed in a hole, return it.
1101 * If we have already stepped through some block buffers to find
1102 * holes but they all contains data. In this case, the last
1103 * offset is already updated and pointed to the end of the last
1104 * mapped page, if it does not reach the endpoint to search,
1105 * that means there should be a hole between them.
1107 if (nr_pages == 0) {
1108 /* Data search found nothing */
1109 if (type == DATA_OFF)
1112 ASSERT(type == HOLE_OFF);
1113 if (lastoff == startoff || lastoff < endoff) {
1121 * At lease we found one page. If this is the first time we
1122 * step into the loop, and if the first page index offset is
1123 * greater than the given search offset, a hole was found.
1125 if (type == HOLE_OFF && lastoff == startoff &&
1126 lastoff < page_offset(pvec.pages[0])) {
1131 for (i = 0; i < nr_pages; i++) {
1132 struct page *page = pvec.pages[i];
1136 * At this point, the page may be truncated or
1137 * invalidated (changing page->mapping to NULL),
1138 * or even swizzled back from swapper_space to tmpfs
1139 * file mapping. However, page->index will not change
1140 * because we have a reference on the page.
1142 * Searching done if the page index is out of range.
1143 * If the current offset is not reaches the end of
1144 * the specified search range, there should be a hole
1147 if (page->index > end) {
1148 if (type == HOLE_OFF && lastoff < endoff) {
1157 * Page truncated or invalidated(page->mapping == NULL).
1158 * We can freely skip it and proceed to check the next
1161 if (unlikely(page->mapping != inode->i_mapping)) {
1166 if (!page_has_buffers(page)) {
1171 found = xfs_lookup_buffer_offset(page, &b_offset, type);
1174 * The found offset may be less than the start
1175 * point to search if this is the first time to
1178 *offset = max_t(loff_t, startoff, b_offset);
1184 * We either searching data but nothing was found, or
1185 * searching hole but found a data buffer. In either
1186 * case, probably the next page contains the desired
1187 * things, update the last offset to it so.
1189 lastoff = page_offset(page) + PAGE_SIZE;
1194 * The number of returned pages less than our desired, search
1195 * done. In this case, nothing was found for searching data,
1196 * but we found a hole behind the last offset.
1198 if (nr_pages < want) {
1199 if (type == HOLE_OFF) {
1206 index = pvec.pages[i - 1]->index + 1;
1207 pagevec_release(&pvec);
1208 } while (index <= end);
1211 pagevec_release(&pvec);
1221 struct inode *inode = file->f_mapping->host;
1222 struct xfs_inode *ip = XFS_I(inode);
1223 struct xfs_mount *mp = ip->i_mount;
1224 loff_t uninitialized_var(offset);
1226 xfs_fileoff_t fsbno;
1231 if (XFS_FORCED_SHUTDOWN(mp))
1234 lock = xfs_ilock_data_map_shared(ip);
1236 isize = i_size_read(inode);
1237 if (start >= isize) {
1243 * Try to read extents from the first block indicated
1244 * by fsbno to the end block of the file.
1246 fsbno = XFS_B_TO_FSBT(mp, start);
1247 end = XFS_B_TO_FSB(mp, isize);
1250 struct xfs_bmbt_irec map[2];
1254 error = xfs_bmapi_read(ip, fsbno, end - fsbno, map, &nmap,
1259 /* No extents at given offset, must be beyond EOF */
1265 for (i = 0; i < nmap; i++) {
1266 offset = max_t(loff_t, start,
1267 XFS_FSB_TO_B(mp, map[i].br_startoff));
1269 /* Landed in the hole we wanted? */
1270 if (whence == SEEK_HOLE &&
1271 map[i].br_startblock == HOLESTARTBLOCK)
1274 /* Landed in the data extent we wanted? */
1275 if (whence == SEEK_DATA &&
1276 (map[i].br_startblock == DELAYSTARTBLOCK ||
1277 (map[i].br_state == XFS_EXT_NORM &&
1278 !isnullstartblock(map[i].br_startblock))))
1282 * Landed in an unwritten extent, try to search
1283 * for hole or data from page cache.
1285 if (map[i].br_state == XFS_EXT_UNWRITTEN) {
1286 if (xfs_find_get_desired_pgoff(inode, &map[i],
1287 whence == SEEK_HOLE ? HOLE_OFF : DATA_OFF,
1294 * We only received one extent out of the two requested. This
1295 * means we've hit EOF and didn't find what we are looking for.
1299 * If we were looking for a hole, set offset to
1300 * the end of the file (i.e., there is an implicit
1301 * hole at the end of any file).
1303 if (whence == SEEK_HOLE) {
1308 * If we were looking for data, it's nowhere to be found
1310 ASSERT(whence == SEEK_DATA);
1318 * Nothing was found, proceed to the next round of search
1319 * if the next reading offset is not at or beyond EOF.
1321 fsbno = map[i - 1].br_startoff + map[i - 1].br_blockcount;
1322 start = XFS_FSB_TO_B(mp, fsbno);
1323 if (start >= isize) {
1324 if (whence == SEEK_HOLE) {
1328 ASSERT(whence == SEEK_DATA);
1336 * If at this point we have found the hole we wanted, the returned
1337 * offset may be bigger than the file size as it may be aligned to
1338 * page boundary for unwritten extents. We need to deal with this
1339 * situation in particular.
1341 if (whence == SEEK_HOLE)
1342 offset = min_t(loff_t, offset, isize);
1343 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
1346 xfs_iunlock(ip, lock);
1363 return generic_file_llseek(file, offset, whence);
1366 return xfs_seek_hole_data(file, offset, whence);
1372 const struct file_operations xfs_file_operations = {
1373 .llseek = xfs_file_llseek,
1374 .read = new_sync_read,
1375 .write = new_sync_write,
1376 .read_iter = xfs_file_read_iter,
1377 .write_iter = xfs_file_write_iter,
1378 .splice_read = xfs_file_splice_read,
1379 .splice_write = iter_file_splice_write,
1380 .unlocked_ioctl = xfs_file_ioctl,
1381 #ifdef CONFIG_COMPAT
1382 .compat_ioctl = xfs_file_compat_ioctl,
1384 .mmap = xfs_file_mmap,
1385 .open = xfs_file_open,
1386 .release = xfs_file_release,
1387 .fsync = xfs_file_fsync,
1388 .fallocate = xfs_file_fallocate,
1391 const struct file_operations xfs_dir_file_operations = {
1392 .open = xfs_dir_open,
1393 .read = generic_read_dir,
1394 .iterate = xfs_file_readdir,
1395 .llseek = generic_file_llseek,
1396 .unlocked_ioctl = xfs_file_ioctl,
1397 #ifdef CONFIG_COMPAT
1398 .compat_ioctl = xfs_file_compat_ioctl,
1400 .fsync = xfs_dir_fsync,
1403 static const struct vm_operations_struct xfs_file_vm_ops = {
1404 .fault = filemap_fault,
1405 .map_pages = filemap_map_pages,
1406 .page_mkwrite = xfs_vm_page_mkwrite,