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"
26 #include "xfs_mount.h"
27 #include "xfs_da_format.h"
28 #include "xfs_da_btree.h"
29 #include "xfs_inode.h"
30 #include "xfs_trans.h"
31 #include "xfs_inode_item.h"
33 #include "xfs_bmap_util.h"
34 #include "xfs_error.h"
36 #include "xfs_dir2_priv.h"
37 #include "xfs_ioctl.h"
38 #include "xfs_trace.h"
40 #include "xfs_dinode.h"
41 #include "xfs_icache.h"
43 #include <linux/aio.h>
44 #include <linux/dcache.h>
45 #include <linux/falloc.h>
46 #include <linux/pagevec.h>
48 static const struct vm_operations_struct xfs_file_vm_ops;
51 * Locking primitives for read and write IO paths to ensure we consistently use
52 * and order the inode->i_mutex, ip->i_lock and ip->i_iolock.
59 if (type & XFS_IOLOCK_EXCL)
60 mutex_lock(&VFS_I(ip)->i_mutex);
69 xfs_iunlock(ip, type);
70 if (type & XFS_IOLOCK_EXCL)
71 mutex_unlock(&VFS_I(ip)->i_mutex);
79 xfs_ilock_demote(ip, type);
80 if (type & XFS_IOLOCK_EXCL)
81 mutex_unlock(&VFS_I(ip)->i_mutex);
87 * xfs_iozero clears the specified range of buffer supplied,
88 * and marks all the affected blocks as valid and modified. If
89 * an affected block is not allocated, it will be allocated. If
90 * an affected block is not completely overwritten, and is not
91 * valid before the operation, it will be read from disk before
92 * being partially zeroed.
96 struct xfs_inode *ip, /* inode */
97 loff_t pos, /* offset in file */
98 size_t count) /* size of data to zero */
101 struct address_space *mapping;
104 mapping = VFS_I(ip)->i_mapping;
106 unsigned offset, bytes;
109 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
110 bytes = PAGE_CACHE_SIZE - offset;
114 status = pagecache_write_begin(NULL, mapping, pos, bytes,
115 AOP_FLAG_UNINTERRUPTIBLE,
120 zero_user(page, offset, bytes);
122 status = pagecache_write_end(NULL, mapping, pos, bytes, bytes,
124 WARN_ON(status <= 0); /* can't return less than zero! */
134 * Fsync operations on directories are much simpler than on regular files,
135 * as there is no file data to flush, and thus also no need for explicit
136 * cache flush operations, and there are no non-transaction metadata updates
137 * on directories either.
146 struct xfs_inode *ip = XFS_I(file->f_mapping->host);
147 struct xfs_mount *mp = ip->i_mount;
150 trace_xfs_dir_fsync(ip);
152 xfs_ilock(ip, XFS_ILOCK_SHARED);
153 if (xfs_ipincount(ip))
154 lsn = ip->i_itemp->ili_last_lsn;
155 xfs_iunlock(ip, XFS_ILOCK_SHARED);
159 return _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, NULL);
169 struct inode *inode = file->f_mapping->host;
170 struct xfs_inode *ip = XFS_I(inode);
171 struct xfs_mount *mp = ip->i_mount;
176 trace_xfs_file_fsync(ip);
178 error = filemap_write_and_wait_range(inode->i_mapping, start, end);
182 if (XFS_FORCED_SHUTDOWN(mp))
185 xfs_iflags_clear(ip, XFS_ITRUNCATED);
187 if (mp->m_flags & XFS_MOUNT_BARRIER) {
189 * If we have an RT and/or log subvolume we need to make sure
190 * to flush the write cache the device used for file data
191 * first. This is to ensure newly written file data make
192 * it to disk before logging the new inode size in case of
193 * an extending write.
195 if (XFS_IS_REALTIME_INODE(ip))
196 xfs_blkdev_issue_flush(mp->m_rtdev_targp);
197 else if (mp->m_logdev_targp != mp->m_ddev_targp)
198 xfs_blkdev_issue_flush(mp->m_ddev_targp);
202 * All metadata updates are logged, which means that we just have
203 * to flush the log up to the latest LSN that touched the inode.
205 xfs_ilock(ip, XFS_ILOCK_SHARED);
206 if (xfs_ipincount(ip)) {
208 (ip->i_itemp->ili_fields & ~XFS_ILOG_TIMESTAMP))
209 lsn = ip->i_itemp->ili_last_lsn;
211 xfs_iunlock(ip, XFS_ILOCK_SHARED);
214 error = _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, &log_flushed);
217 * If we only have a single device, and the log force about was
218 * a no-op we might have to flush the data device cache here.
219 * This can only happen for fdatasync/O_DSYNC if we were overwriting
220 * an already allocated file and thus do not have any metadata to
223 if ((mp->m_flags & XFS_MOUNT_BARRIER) &&
224 mp->m_logdev_targp == mp->m_ddev_targp &&
225 !XFS_IS_REALTIME_INODE(ip) &&
227 xfs_blkdev_issue_flush(mp->m_ddev_targp);
237 struct file *file = iocb->ki_filp;
238 struct inode *inode = file->f_mapping->host;
239 struct xfs_inode *ip = XFS_I(inode);
240 struct xfs_mount *mp = ip->i_mount;
241 size_t size = iov_iter_count(to);
245 loff_t pos = iocb->ki_pos;
247 XFS_STATS_INC(xs_read_calls);
249 if (unlikely(file->f_flags & O_DIRECT))
250 ioflags |= XFS_IO_ISDIRECT;
251 if (file->f_mode & FMODE_NOCMTIME)
252 ioflags |= XFS_IO_INVIS;
254 if (unlikely(ioflags & XFS_IO_ISDIRECT)) {
255 xfs_buftarg_t *target =
256 XFS_IS_REALTIME_INODE(ip) ?
257 mp->m_rtdev_targp : mp->m_ddev_targp;
258 /* DIO must be aligned to device logical sector size */
259 if ((pos | size) & target->bt_logical_sectormask) {
260 if (pos == i_size_read(inode))
266 n = mp->m_super->s_maxbytes - pos;
267 if (n <= 0 || size == 0)
273 if (XFS_FORCED_SHUTDOWN(mp))
277 * Locking is a bit tricky here. If we take an exclusive lock
278 * for direct IO, we effectively serialise all new concurrent
279 * read IO to this file and block it behind IO that is currently in
280 * progress because IO in progress holds the IO lock shared. We only
281 * need to hold the lock exclusive to blow away the page cache, so
282 * only take lock exclusively if the page cache needs invalidation.
283 * This allows the normal direct IO case of no page cache pages to
284 * proceeed concurrently without serialisation.
286 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
287 if ((ioflags & XFS_IO_ISDIRECT) && inode->i_mapping->nrpages) {
288 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
289 xfs_rw_ilock(ip, XFS_IOLOCK_EXCL);
291 if (inode->i_mapping->nrpages) {
292 ret = filemap_write_and_wait_range(
293 VFS_I(ip)->i_mapping,
296 xfs_rw_iunlock(ip, XFS_IOLOCK_EXCL);
301 * Invalidate whole pages. This can return an error if
302 * we fail to invalidate a page, but this should never
303 * happen on XFS. Warn if it does fail.
305 ret = invalidate_inode_pages2_range(VFS_I(ip)->i_mapping,
306 pos >> PAGE_CACHE_SHIFT, -1);
310 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
313 trace_xfs_file_read(ip, size, pos, ioflags);
315 ret = generic_file_read_iter(iocb, to);
317 XFS_STATS_ADD(xs_read_bytes, ret);
319 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
324 xfs_file_splice_read(
327 struct pipe_inode_info *pipe,
331 struct xfs_inode *ip = XFS_I(infilp->f_mapping->host);
335 XFS_STATS_INC(xs_read_calls);
337 if (infilp->f_mode & FMODE_NOCMTIME)
338 ioflags |= XFS_IO_INVIS;
340 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
343 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
345 trace_xfs_file_splice_read(ip, count, *ppos, ioflags);
347 ret = generic_file_splice_read(infilp, ppos, pipe, count, flags);
349 XFS_STATS_ADD(xs_read_bytes, ret);
351 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
356 * This routine is called to handle zeroing any space in the last block of the
357 * file that is beyond the EOF. We do this since the size is being increased
358 * without writing anything to that block and we don't want to read the
359 * garbage on the disk.
361 STATIC int /* error (positive) */
363 struct xfs_inode *ip,
367 struct xfs_mount *mp = ip->i_mount;
368 xfs_fileoff_t last_fsb = XFS_B_TO_FSBT(mp, isize);
369 int zero_offset = XFS_B_FSB_OFFSET(mp, isize);
373 struct xfs_bmbt_irec imap;
375 xfs_ilock(ip, XFS_ILOCK_EXCL);
376 error = xfs_bmapi_read(ip, last_fsb, 1, &imap, &nimaps, 0);
377 xfs_iunlock(ip, XFS_ILOCK_EXCL);
384 * If the block underlying isize is just a hole, then there
385 * is nothing to zero.
387 if (imap.br_startblock == HOLESTARTBLOCK)
390 zero_len = mp->m_sb.sb_blocksize - zero_offset;
391 if (isize + zero_len > offset)
392 zero_len = offset - isize;
393 return xfs_iozero(ip, isize, zero_len);
397 * Zero any on disk space between the current EOF and the new, larger EOF.
399 * This handles the normal case of zeroing the remainder of the last block in
400 * the file and the unusual case of zeroing blocks out beyond the size of the
401 * file. This second case only happens with fixed size extents and when the
402 * system crashes before the inode size was updated but after blocks were
405 * Expects the iolock to be held exclusive, and will take the ilock internally.
407 int /* error (positive) */
409 struct xfs_inode *ip,
410 xfs_off_t offset, /* starting I/O offset */
411 xfs_fsize_t isize) /* current inode size */
413 struct xfs_mount *mp = ip->i_mount;
414 xfs_fileoff_t start_zero_fsb;
415 xfs_fileoff_t end_zero_fsb;
416 xfs_fileoff_t zero_count_fsb;
417 xfs_fileoff_t last_fsb;
418 xfs_fileoff_t zero_off;
419 xfs_fsize_t zero_len;
422 struct xfs_bmbt_irec imap;
424 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
425 ASSERT(offset > isize);
428 * First handle zeroing the block on which isize resides.
430 * We only zero a part of that block so it is handled specially.
432 if (XFS_B_FSB_OFFSET(mp, isize) != 0) {
433 error = xfs_zero_last_block(ip, offset, isize);
439 * Calculate the range between the new size and the old where blocks
440 * needing to be zeroed may exist.
442 * To get the block where the last byte in the file currently resides,
443 * we need to subtract one from the size and truncate back to a block
444 * boundary. We subtract 1 in case the size is exactly on a block
447 last_fsb = isize ? XFS_B_TO_FSBT(mp, isize - 1) : (xfs_fileoff_t)-1;
448 start_zero_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
449 end_zero_fsb = XFS_B_TO_FSBT(mp, offset - 1);
450 ASSERT((xfs_sfiloff_t)last_fsb < (xfs_sfiloff_t)start_zero_fsb);
451 if (last_fsb == end_zero_fsb) {
453 * The size was only incremented on its last block.
454 * We took care of that above, so just return.
459 ASSERT(start_zero_fsb <= end_zero_fsb);
460 while (start_zero_fsb <= end_zero_fsb) {
462 zero_count_fsb = end_zero_fsb - start_zero_fsb + 1;
464 xfs_ilock(ip, XFS_ILOCK_EXCL);
465 error = xfs_bmapi_read(ip, start_zero_fsb, zero_count_fsb,
467 xfs_iunlock(ip, XFS_ILOCK_EXCL);
473 if (imap.br_state == XFS_EXT_UNWRITTEN ||
474 imap.br_startblock == HOLESTARTBLOCK) {
475 start_zero_fsb = imap.br_startoff + imap.br_blockcount;
476 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
481 * There are blocks we need to zero.
483 zero_off = XFS_FSB_TO_B(mp, start_zero_fsb);
484 zero_len = XFS_FSB_TO_B(mp, imap.br_blockcount);
486 if ((zero_off + zero_len) > offset)
487 zero_len = offset - zero_off;
489 error = xfs_iozero(ip, zero_off, zero_len);
493 start_zero_fsb = imap.br_startoff + imap.br_blockcount;
494 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
501 * Common pre-write limit and setup checks.
503 * Called with the iolocked held either shared and exclusive according to
504 * @iolock, and returns with it held. Might upgrade the iolock to exclusive
505 * if called for a direct write beyond i_size.
508 xfs_file_aio_write_checks(
514 struct inode *inode = file->f_mapping->host;
515 struct xfs_inode *ip = XFS_I(inode);
519 error = generic_write_checks(file, pos, count, S_ISBLK(inode->i_mode));
524 * If the offset is beyond the size of the file, we need to zero any
525 * blocks that fall between the existing EOF and the start of this
526 * write. If zeroing is needed and we are currently holding the
527 * iolock shared, we need to update it to exclusive which implies
528 * having to redo all checks before.
530 if (*pos > i_size_read(inode)) {
531 if (*iolock == XFS_IOLOCK_SHARED) {
532 xfs_rw_iunlock(ip, *iolock);
533 *iolock = XFS_IOLOCK_EXCL;
534 xfs_rw_ilock(ip, *iolock);
537 error = xfs_zero_eof(ip, *pos, i_size_read(inode));
543 * Updating the timestamps will grab the ilock again from
544 * xfs_fs_dirty_inode, so we have to call it after dropping the
545 * lock above. Eventually we should look into a way to avoid
546 * the pointless lock roundtrip.
548 if (likely(!(file->f_mode & FMODE_NOCMTIME))) {
549 error = file_update_time(file);
555 * If we're writing the file then make sure to clear the setuid and
556 * setgid bits if the process is not being run by root. This keeps
557 * people from modifying setuid and setgid binaries.
559 return file_remove_suid(file);
563 * xfs_file_dio_aio_write - handle direct IO writes
565 * Lock the inode appropriately to prepare for and issue a direct IO write.
566 * By separating it from the buffered write path we remove all the tricky to
567 * follow locking changes and looping.
569 * If there are cached pages or we're extending the file, we need IOLOCK_EXCL
570 * until we're sure the bytes at the new EOF have been zeroed and/or the cached
571 * pages are flushed out.
573 * In most cases the direct IO writes will be done holding IOLOCK_SHARED
574 * allowing them to be done in parallel with reads and other direct IO writes.
575 * However, if the IO is not aligned to filesystem blocks, the direct IO layer
576 * needs to do sub-block zeroing and that requires serialisation against other
577 * direct IOs to the same block. In this case we need to serialise the
578 * submission of the unaligned IOs so that we don't get racing block zeroing in
579 * the dio layer. To avoid the problem with aio, we also need to wait for
580 * outstanding IOs to complete so that unwritten extent conversion is completed
581 * before we try to map the overlapping block. This is currently implemented by
582 * hitting it with a big hammer (i.e. inode_dio_wait()).
584 * Returns with locks held indicated by @iolock and errors indicated by
585 * negative return values.
588 xfs_file_dio_aio_write(
590 struct iov_iter *from)
592 struct file *file = iocb->ki_filp;
593 struct address_space *mapping = file->f_mapping;
594 struct inode *inode = mapping->host;
595 struct xfs_inode *ip = XFS_I(inode);
596 struct xfs_mount *mp = ip->i_mount;
598 int unaligned_io = 0;
600 size_t count = iov_iter_count(from);
601 loff_t pos = iocb->ki_pos;
602 struct xfs_buftarg *target = XFS_IS_REALTIME_INODE(ip) ?
603 mp->m_rtdev_targp : mp->m_ddev_targp;
605 /* DIO must be aligned to device logical sector size */
606 if ((pos | count) & target->bt_logical_sectormask)
609 /* "unaligned" here means not aligned to a filesystem block */
610 if ((pos & mp->m_blockmask) || ((pos + count) & mp->m_blockmask))
614 * We don't need to take an exclusive lock unless there page cache needs
615 * to be invalidated or unaligned IO is being executed. We don't need to
616 * consider the EOF extension case here because
617 * xfs_file_aio_write_checks() will relock the inode as necessary for
618 * EOF zeroing cases and fill out the new inode size as appropriate.
620 if (unaligned_io || mapping->nrpages)
621 iolock = XFS_IOLOCK_EXCL;
623 iolock = XFS_IOLOCK_SHARED;
624 xfs_rw_ilock(ip, iolock);
627 * Recheck if there are cached pages that need invalidate after we got
628 * the iolock to protect against other threads adding new pages while
629 * we were waiting for the iolock.
631 if (mapping->nrpages && iolock == XFS_IOLOCK_SHARED) {
632 xfs_rw_iunlock(ip, iolock);
633 iolock = XFS_IOLOCK_EXCL;
634 xfs_rw_ilock(ip, iolock);
637 ret = xfs_file_aio_write_checks(file, &pos, &count, &iolock);
640 iov_iter_truncate(from, count);
642 if (mapping->nrpages) {
643 ret = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
647 truncate_pagecache_range(VFS_I(ip), pos, -1);
651 * If we are doing unaligned IO, wait for all other IO to drain,
652 * otherwise demote the lock if we had to flush cached pages
655 inode_dio_wait(inode);
656 else if (iolock == XFS_IOLOCK_EXCL) {
657 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
658 iolock = XFS_IOLOCK_SHARED;
661 trace_xfs_file_direct_write(ip, count, iocb->ki_pos, 0);
662 ret = generic_file_direct_write(iocb, from, pos);
665 xfs_rw_iunlock(ip, iolock);
667 /* No fallback to buffered IO on errors for XFS. */
668 ASSERT(ret < 0 || ret == count);
673 xfs_file_buffered_aio_write(
675 struct iov_iter *from)
677 struct file *file = iocb->ki_filp;
678 struct address_space *mapping = file->f_mapping;
679 struct inode *inode = mapping->host;
680 struct xfs_inode *ip = XFS_I(inode);
683 int iolock = XFS_IOLOCK_EXCL;
684 loff_t pos = iocb->ki_pos;
685 size_t count = iov_iter_count(from);
687 xfs_rw_ilock(ip, iolock);
689 ret = xfs_file_aio_write_checks(file, &pos, &count, &iolock);
693 iov_iter_truncate(from, count);
694 /* We can write back this queue in page reclaim */
695 current->backing_dev_info = mapping->backing_dev_info;
698 trace_xfs_file_buffered_write(ip, count, iocb->ki_pos, 0);
699 ret = generic_perform_write(file, from, pos);
700 if (likely(ret >= 0))
701 iocb->ki_pos = pos + ret;
704 * If we hit a space limit, try to free up some lingering preallocated
705 * space before returning an error. In the case of ENOSPC, first try to
706 * write back all dirty inodes to free up some of the excess reserved
707 * metadata space. This reduces the chances that the eofblocks scan
708 * waits on dirty mappings. Since xfs_flush_inodes() is serialized, this
709 * also behaves as a filter to prevent too many eofblocks scans from
710 * running at the same time.
712 if (ret == -EDQUOT && !enospc) {
713 enospc = xfs_inode_free_quota_eofblocks(ip);
716 } else if (ret == -ENOSPC && !enospc) {
717 struct xfs_eofblocks eofb = {0};
720 xfs_flush_inodes(ip->i_mount);
721 eofb.eof_scan_owner = ip->i_ino; /* for locking */
722 eofb.eof_flags = XFS_EOF_FLAGS_SYNC;
723 xfs_icache_free_eofblocks(ip->i_mount, &eofb);
727 current->backing_dev_info = NULL;
729 xfs_rw_iunlock(ip, iolock);
736 struct iov_iter *from)
738 struct file *file = iocb->ki_filp;
739 struct address_space *mapping = file->f_mapping;
740 struct inode *inode = mapping->host;
741 struct xfs_inode *ip = XFS_I(inode);
743 size_t ocount = iov_iter_count(from);
745 XFS_STATS_INC(xs_write_calls);
750 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
753 if (unlikely(file->f_flags & O_DIRECT))
754 ret = xfs_file_dio_aio_write(iocb, from);
756 ret = xfs_file_buffered_aio_write(iocb, from);
761 XFS_STATS_ADD(xs_write_bytes, ret);
763 /* Handle various SYNC-type writes */
764 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
778 struct inode *inode = file_inode(file);
779 struct xfs_inode *ip = XFS_I(inode);
780 struct xfs_trans *tp;
784 if (!S_ISREG(inode->i_mode))
786 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
787 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE))
790 xfs_ilock(ip, XFS_IOLOCK_EXCL);
791 if (mode & FALLOC_FL_PUNCH_HOLE) {
792 error = xfs_free_file_space(ip, offset, len);
795 } else if (mode & FALLOC_FL_COLLAPSE_RANGE) {
796 unsigned blksize_mask = (1 << inode->i_blkbits) - 1;
798 if (offset & blksize_mask || len & blksize_mask) {
804 * There is no need to overlap collapse range with EOF,
805 * in which case it is effectively a truncate operation
807 if (offset + len >= i_size_read(inode)) {
812 new_size = i_size_read(inode) - len;
814 error = xfs_collapse_file_space(ip, offset, len);
818 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
819 offset + len > i_size_read(inode)) {
820 new_size = offset + len;
821 error = inode_newsize_ok(inode, new_size);
826 if (mode & FALLOC_FL_ZERO_RANGE)
827 error = xfs_zero_file_space(ip, offset, len);
829 error = xfs_alloc_file_space(ip, offset, len,
835 tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_WRITEID);
836 error = xfs_trans_reserve(tp, &M_RES(ip->i_mount)->tr_writeid, 0, 0);
838 xfs_trans_cancel(tp, 0);
842 xfs_ilock(ip, XFS_ILOCK_EXCL);
843 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
844 ip->i_d.di_mode &= ~S_ISUID;
845 if (ip->i_d.di_mode & S_IXGRP)
846 ip->i_d.di_mode &= ~S_ISGID;
848 if (!(mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_COLLAPSE_RANGE)))
849 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
851 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
852 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
854 if (file->f_flags & O_DSYNC)
855 xfs_trans_set_sync(tp);
856 error = xfs_trans_commit(tp, 0);
860 /* Change file size if needed */
864 iattr.ia_valid = ATTR_SIZE;
865 iattr.ia_size = new_size;
866 error = xfs_setattr_size(ip, &iattr);
870 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
880 if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
882 if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb)))
892 struct xfs_inode *ip = XFS_I(inode);
896 error = xfs_file_open(inode, file);
901 * If there are any blocks, read-ahead block 0 as we're almost
902 * certain to have the next operation be a read there.
904 mode = xfs_ilock_data_map_shared(ip);
905 if (ip->i_d.di_nextents > 0)
906 xfs_dir3_data_readahead(ip, 0, -1);
907 xfs_iunlock(ip, mode);
916 return xfs_release(XFS_I(inode));
922 struct dir_context *ctx)
924 struct inode *inode = file_inode(file);
925 xfs_inode_t *ip = XFS_I(inode);
930 * The Linux API doesn't pass down the total size of the buffer
931 * we read into down to the filesystem. With the filldir concept
932 * it's not needed for correct information, but the XFS dir2 leaf
933 * code wants an estimate of the buffer size to calculate it's
934 * readahead window and size the buffers used for mapping to
937 * Try to give it an estimate that's good enough, maybe at some
938 * point we can change the ->readdir prototype to include the
939 * buffer size. For now we use the current glibc buffer size.
941 bufsize = (size_t)min_t(loff_t, 32768, ip->i_d.di_size);
943 error = xfs_readdir(ip, ctx, bufsize);
952 struct vm_area_struct *vma)
954 vma->vm_ops = &xfs_file_vm_ops;
961 * mmap()d file has taken write protection fault and is being made
962 * writable. We can set the page state up correctly for a writable
963 * page, which means we can do correct delalloc accounting (ENOSPC
964 * checking!) and unwritten extent mapping.
968 struct vm_area_struct *vma,
969 struct vm_fault *vmf)
971 return block_page_mkwrite(vma, vmf, xfs_get_blocks);
975 * This type is designed to indicate the type of offset we would like
976 * to search from page cache for either xfs_seek_data() or xfs_seek_hole().
984 * Lookup the desired type of offset from the given page.
986 * On success, return true and the offset argument will point to the
987 * start of the region that was found. Otherwise this function will
988 * return false and keep the offset argument unchanged.
991 xfs_lookup_buffer_offset(
996 loff_t lastoff = page_offset(page);
998 struct buffer_head *bh, *head;
1000 bh = head = page_buffers(page);
1003 * Unwritten extents that have data in the page
1004 * cache covering them can be identified by the
1005 * BH_Unwritten state flag. Pages with multiple
1006 * buffers might have a mix of holes, data and
1007 * unwritten extents - any buffer with valid
1008 * data in it should have BH_Uptodate flag set
1011 if (buffer_unwritten(bh) ||
1012 buffer_uptodate(bh)) {
1013 if (type == DATA_OFF)
1016 if (type == HOLE_OFF)
1024 lastoff += bh->b_size;
1025 } while ((bh = bh->b_this_page) != head);
1031 * This routine is called to find out and return a data or hole offset
1032 * from the page cache for unwritten extents according to the desired
1033 * type for xfs_seek_data() or xfs_seek_hole().
1035 * The argument offset is used to tell where we start to search from the
1036 * page cache. Map is used to figure out the end points of the range to
1039 * Return true if the desired type of offset was found, and the argument
1040 * offset is filled with that address. Otherwise, return false and keep
1044 xfs_find_get_desired_pgoff(
1045 struct inode *inode,
1046 struct xfs_bmbt_irec *map,
1050 struct xfs_inode *ip = XFS_I(inode);
1051 struct xfs_mount *mp = ip->i_mount;
1052 struct pagevec pvec;
1056 loff_t startoff = *offset;
1057 loff_t lastoff = startoff;
1060 pagevec_init(&pvec, 0);
1062 index = startoff >> PAGE_CACHE_SHIFT;
1063 endoff = XFS_FSB_TO_B(mp, map->br_startoff + map->br_blockcount);
1064 end = endoff >> PAGE_CACHE_SHIFT;
1070 want = min_t(pgoff_t, end - index, PAGEVEC_SIZE);
1071 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index,
1074 * No page mapped into given range. If we are searching holes
1075 * and if this is the first time we got into the loop, it means
1076 * that the given offset is landed in a hole, return it.
1078 * If we have already stepped through some block buffers to find
1079 * holes but they all contains data. In this case, the last
1080 * offset is already updated and pointed to the end of the last
1081 * mapped page, if it does not reach the endpoint to search,
1082 * that means there should be a hole between them.
1084 if (nr_pages == 0) {
1085 /* Data search found nothing */
1086 if (type == DATA_OFF)
1089 ASSERT(type == HOLE_OFF);
1090 if (lastoff == startoff || lastoff < endoff) {
1098 * At lease we found one page. If this is the first time we
1099 * step into the loop, and if the first page index offset is
1100 * greater than the given search offset, a hole was found.
1102 if (type == HOLE_OFF && lastoff == startoff &&
1103 lastoff < page_offset(pvec.pages[0])) {
1108 for (i = 0; i < nr_pages; i++) {
1109 struct page *page = pvec.pages[i];
1113 * At this point, the page may be truncated or
1114 * invalidated (changing page->mapping to NULL),
1115 * or even swizzled back from swapper_space to tmpfs
1116 * file mapping. However, page->index will not change
1117 * because we have a reference on the page.
1119 * Searching done if the page index is out of range.
1120 * If the current offset is not reaches the end of
1121 * the specified search range, there should be a hole
1124 if (page->index > end) {
1125 if (type == HOLE_OFF && lastoff < endoff) {
1134 * Page truncated or invalidated(page->mapping == NULL).
1135 * We can freely skip it and proceed to check the next
1138 if (unlikely(page->mapping != inode->i_mapping)) {
1143 if (!page_has_buffers(page)) {
1148 found = xfs_lookup_buffer_offset(page, &b_offset, type);
1151 * The found offset may be less than the start
1152 * point to search if this is the first time to
1155 *offset = max_t(loff_t, startoff, b_offset);
1161 * We either searching data but nothing was found, or
1162 * searching hole but found a data buffer. In either
1163 * case, probably the next page contains the desired
1164 * things, update the last offset to it so.
1166 lastoff = page_offset(page) + PAGE_SIZE;
1171 * The number of returned pages less than our desired, search
1172 * done. In this case, nothing was found for searching data,
1173 * but we found a hole behind the last offset.
1175 if (nr_pages < want) {
1176 if (type == HOLE_OFF) {
1183 index = pvec.pages[i - 1]->index + 1;
1184 pagevec_release(&pvec);
1185 } while (index <= end);
1188 pagevec_release(&pvec);
1197 struct inode *inode = file->f_mapping->host;
1198 struct xfs_inode *ip = XFS_I(inode);
1199 struct xfs_mount *mp = ip->i_mount;
1200 loff_t uninitialized_var(offset);
1202 xfs_fileoff_t fsbno;
1207 lock = xfs_ilock_data_map_shared(ip);
1209 isize = i_size_read(inode);
1210 if (start >= isize) {
1216 * Try to read extents from the first block indicated
1217 * by fsbno to the end block of the file.
1219 fsbno = XFS_B_TO_FSBT(mp, start);
1220 end = XFS_B_TO_FSB(mp, isize);
1222 struct xfs_bmbt_irec map[2];
1226 error = xfs_bmapi_read(ip, fsbno, end - fsbno, map, &nmap,
1231 /* No extents at given offset, must be beyond EOF */
1237 for (i = 0; i < nmap; i++) {
1238 offset = max_t(loff_t, start,
1239 XFS_FSB_TO_B(mp, map[i].br_startoff));
1241 /* Landed in a data extent */
1242 if (map[i].br_startblock == DELAYSTARTBLOCK ||
1243 (map[i].br_state == XFS_EXT_NORM &&
1244 !isnullstartblock(map[i].br_startblock)))
1248 * Landed in an unwritten extent, try to search data
1251 if (map[i].br_state == XFS_EXT_UNWRITTEN) {
1252 if (xfs_find_get_desired_pgoff(inode, &map[i],
1259 * map[0] is hole or its an unwritten extent but
1260 * without data in page cache. Probably means that
1261 * we are reading after EOF if nothing in map[1].
1271 * Nothing was found, proceed to the next round of search
1272 * if reading offset not beyond or hit EOF.
1274 fsbno = map[i - 1].br_startoff + map[i - 1].br_blockcount;
1275 start = XFS_FSB_TO_B(mp, fsbno);
1276 if (start >= isize) {
1283 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
1286 xfs_iunlock(ip, lock);
1298 struct inode *inode = file->f_mapping->host;
1299 struct xfs_inode *ip = XFS_I(inode);
1300 struct xfs_mount *mp = ip->i_mount;
1301 loff_t uninitialized_var(offset);
1303 xfs_fileoff_t fsbno;
1308 if (XFS_FORCED_SHUTDOWN(mp))
1311 lock = xfs_ilock_data_map_shared(ip);
1313 isize = i_size_read(inode);
1314 if (start >= isize) {
1319 fsbno = XFS_B_TO_FSBT(mp, start);
1320 end = XFS_B_TO_FSB(mp, isize);
1323 struct xfs_bmbt_irec map[2];
1327 error = xfs_bmapi_read(ip, fsbno, end - fsbno, map, &nmap,
1332 /* No extents at given offset, must be beyond EOF */
1338 for (i = 0; i < nmap; i++) {
1339 offset = max_t(loff_t, start,
1340 XFS_FSB_TO_B(mp, map[i].br_startoff));
1342 /* Landed in a hole */
1343 if (map[i].br_startblock == HOLESTARTBLOCK)
1347 * Landed in an unwritten extent, try to search hole
1350 if (map[i].br_state == XFS_EXT_UNWRITTEN) {
1351 if (xfs_find_get_desired_pgoff(inode, &map[i],
1358 * map[0] contains data or its unwritten but contains
1359 * data in page cache, probably means that we are
1360 * reading after EOF. We should fix offset to point
1361 * to the end of the file(i.e., there is an implicit
1362 * hole at the end of any file).
1372 * Both mappings contains data, proceed to the next round of
1373 * search if the current reading offset not beyond or hit EOF.
1375 fsbno = map[i - 1].br_startoff + map[i - 1].br_blockcount;
1376 start = XFS_FSB_TO_B(mp, fsbno);
1377 if (start >= isize) {
1385 * At this point, we must have found a hole. However, the returned
1386 * offset may be bigger than the file size as it may be aligned to
1387 * page boundary for unwritten extents, we need to deal with this
1388 * situation in particular.
1390 offset = min_t(loff_t, offset, isize);
1391 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
1394 xfs_iunlock(ip, lock);
1411 return generic_file_llseek(file, offset, origin);
1413 return xfs_seek_data(file, offset);
1415 return xfs_seek_hole(file, offset);
1421 const struct file_operations xfs_file_operations = {
1422 .llseek = xfs_file_llseek,
1423 .read = new_sync_read,
1424 .write = new_sync_write,
1425 .read_iter = xfs_file_read_iter,
1426 .write_iter = xfs_file_write_iter,
1427 .splice_read = xfs_file_splice_read,
1428 .splice_write = iter_file_splice_write,
1429 .unlocked_ioctl = xfs_file_ioctl,
1430 #ifdef CONFIG_COMPAT
1431 .compat_ioctl = xfs_file_compat_ioctl,
1433 .mmap = xfs_file_mmap,
1434 .open = xfs_file_open,
1435 .release = xfs_file_release,
1436 .fsync = xfs_file_fsync,
1437 .fallocate = xfs_file_fallocate,
1440 const struct file_operations xfs_dir_file_operations = {
1441 .open = xfs_dir_open,
1442 .read = generic_read_dir,
1443 .iterate = xfs_file_readdir,
1444 .llseek = generic_file_llseek,
1445 .unlocked_ioctl = xfs_file_ioctl,
1446 #ifdef CONFIG_COMPAT
1447 .compat_ioctl = xfs_file_compat_ioctl,
1449 .fsync = xfs_dir_fsync,
1452 static const struct vm_operations_struct xfs_file_vm_ops = {
1453 .fault = filemap_fault,
1454 .map_pages = filemap_map_pages,
1455 .page_mkwrite = xfs_vm_page_mkwrite,
1456 .remap_pages = generic_file_remap_pages,