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);
299 truncate_pagecache_range(VFS_I(ip), pos, -1);
301 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
304 trace_xfs_file_read(ip, size, pos, ioflags);
306 ret = generic_file_read_iter(iocb, to);
308 XFS_STATS_ADD(xs_read_bytes, ret);
310 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
315 xfs_file_splice_read(
318 struct pipe_inode_info *pipe,
322 struct xfs_inode *ip = XFS_I(infilp->f_mapping->host);
326 XFS_STATS_INC(xs_read_calls);
328 if (infilp->f_mode & FMODE_NOCMTIME)
329 ioflags |= XFS_IO_INVIS;
331 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
334 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
336 trace_xfs_file_splice_read(ip, count, *ppos, ioflags);
338 ret = generic_file_splice_read(infilp, ppos, pipe, count, flags);
340 XFS_STATS_ADD(xs_read_bytes, ret);
342 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
347 * This routine is called to handle zeroing any space in the last block of the
348 * file that is beyond the EOF. We do this since the size is being increased
349 * without writing anything to that block and we don't want to read the
350 * garbage on the disk.
352 STATIC int /* error (positive) */
354 struct xfs_inode *ip,
358 struct xfs_mount *mp = ip->i_mount;
359 xfs_fileoff_t last_fsb = XFS_B_TO_FSBT(mp, isize);
360 int zero_offset = XFS_B_FSB_OFFSET(mp, isize);
364 struct xfs_bmbt_irec imap;
366 xfs_ilock(ip, XFS_ILOCK_EXCL);
367 error = xfs_bmapi_read(ip, last_fsb, 1, &imap, &nimaps, 0);
368 xfs_iunlock(ip, XFS_ILOCK_EXCL);
375 * If the block underlying isize is just a hole, then there
376 * is nothing to zero.
378 if (imap.br_startblock == HOLESTARTBLOCK)
381 zero_len = mp->m_sb.sb_blocksize - zero_offset;
382 if (isize + zero_len > offset)
383 zero_len = offset - isize;
384 return xfs_iozero(ip, isize, zero_len);
388 * Zero any on disk space between the current EOF and the new, larger EOF.
390 * This handles the normal case of zeroing the remainder of the last block in
391 * the file and the unusual case of zeroing blocks out beyond the size of the
392 * file. This second case only happens with fixed size extents and when the
393 * system crashes before the inode size was updated but after blocks were
396 * Expects the iolock to be held exclusive, and will take the ilock internally.
398 int /* error (positive) */
400 struct xfs_inode *ip,
401 xfs_off_t offset, /* starting I/O offset */
402 xfs_fsize_t isize) /* current inode size */
404 struct xfs_mount *mp = ip->i_mount;
405 xfs_fileoff_t start_zero_fsb;
406 xfs_fileoff_t end_zero_fsb;
407 xfs_fileoff_t zero_count_fsb;
408 xfs_fileoff_t last_fsb;
409 xfs_fileoff_t zero_off;
410 xfs_fsize_t zero_len;
413 struct xfs_bmbt_irec imap;
415 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
416 ASSERT(offset > isize);
419 * First handle zeroing the block on which isize resides.
421 * We only zero a part of that block so it is handled specially.
423 if (XFS_B_FSB_OFFSET(mp, isize) != 0) {
424 error = xfs_zero_last_block(ip, offset, isize);
430 * Calculate the range between the new size and the old where blocks
431 * needing to be zeroed may exist.
433 * To get the block where the last byte in the file currently resides,
434 * we need to subtract one from the size and truncate back to a block
435 * boundary. We subtract 1 in case the size is exactly on a block
438 last_fsb = isize ? XFS_B_TO_FSBT(mp, isize - 1) : (xfs_fileoff_t)-1;
439 start_zero_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
440 end_zero_fsb = XFS_B_TO_FSBT(mp, offset - 1);
441 ASSERT((xfs_sfiloff_t)last_fsb < (xfs_sfiloff_t)start_zero_fsb);
442 if (last_fsb == end_zero_fsb) {
444 * The size was only incremented on its last block.
445 * We took care of that above, so just return.
450 ASSERT(start_zero_fsb <= end_zero_fsb);
451 while (start_zero_fsb <= end_zero_fsb) {
453 zero_count_fsb = end_zero_fsb - start_zero_fsb + 1;
455 xfs_ilock(ip, XFS_ILOCK_EXCL);
456 error = xfs_bmapi_read(ip, start_zero_fsb, zero_count_fsb,
458 xfs_iunlock(ip, XFS_ILOCK_EXCL);
464 if (imap.br_state == XFS_EXT_UNWRITTEN ||
465 imap.br_startblock == HOLESTARTBLOCK) {
466 start_zero_fsb = imap.br_startoff + imap.br_blockcount;
467 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
472 * There are blocks we need to zero.
474 zero_off = XFS_FSB_TO_B(mp, start_zero_fsb);
475 zero_len = XFS_FSB_TO_B(mp, imap.br_blockcount);
477 if ((zero_off + zero_len) > offset)
478 zero_len = offset - zero_off;
480 error = xfs_iozero(ip, zero_off, zero_len);
484 start_zero_fsb = imap.br_startoff + imap.br_blockcount;
485 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
492 * Common pre-write limit and setup checks.
494 * Called with the iolocked held either shared and exclusive according to
495 * @iolock, and returns with it held. Might upgrade the iolock to exclusive
496 * if called for a direct write beyond i_size.
499 xfs_file_aio_write_checks(
505 struct inode *inode = file->f_mapping->host;
506 struct xfs_inode *ip = XFS_I(inode);
510 error = generic_write_checks(file, pos, count, S_ISBLK(inode->i_mode));
515 * If the offset is beyond the size of the file, we need to zero any
516 * blocks that fall between the existing EOF and the start of this
517 * write. If zeroing is needed and we are currently holding the
518 * iolock shared, we need to update it to exclusive which implies
519 * having to redo all checks before.
521 if (*pos > i_size_read(inode)) {
522 if (*iolock == XFS_IOLOCK_SHARED) {
523 xfs_rw_iunlock(ip, *iolock);
524 *iolock = XFS_IOLOCK_EXCL;
525 xfs_rw_ilock(ip, *iolock);
528 error = xfs_zero_eof(ip, *pos, i_size_read(inode));
534 * Updating the timestamps will grab the ilock again from
535 * xfs_fs_dirty_inode, so we have to call it after dropping the
536 * lock above. Eventually we should look into a way to avoid
537 * the pointless lock roundtrip.
539 if (likely(!(file->f_mode & FMODE_NOCMTIME))) {
540 error = file_update_time(file);
546 * If we're writing the file then make sure to clear the setuid and
547 * setgid bits if the process is not being run by root. This keeps
548 * people from modifying setuid and setgid binaries.
550 return file_remove_suid(file);
554 * xfs_file_dio_aio_write - handle direct IO writes
556 * Lock the inode appropriately to prepare for and issue a direct IO write.
557 * By separating it from the buffered write path we remove all the tricky to
558 * follow locking changes and looping.
560 * If there are cached pages or we're extending the file, we need IOLOCK_EXCL
561 * until we're sure the bytes at the new EOF have been zeroed and/or the cached
562 * pages are flushed out.
564 * In most cases the direct IO writes will be done holding IOLOCK_SHARED
565 * allowing them to be done in parallel with reads and other direct IO writes.
566 * However, if the IO is not aligned to filesystem blocks, the direct IO layer
567 * needs to do sub-block zeroing and that requires serialisation against other
568 * direct IOs to the same block. In this case we need to serialise the
569 * submission of the unaligned IOs so that we don't get racing block zeroing in
570 * the dio layer. To avoid the problem with aio, we also need to wait for
571 * outstanding IOs to complete so that unwritten extent conversion is completed
572 * before we try to map the overlapping block. This is currently implemented by
573 * hitting it with a big hammer (i.e. inode_dio_wait()).
575 * Returns with locks held indicated by @iolock and errors indicated by
576 * negative return values.
579 xfs_file_dio_aio_write(
581 struct iov_iter *from)
583 struct file *file = iocb->ki_filp;
584 struct address_space *mapping = file->f_mapping;
585 struct inode *inode = mapping->host;
586 struct xfs_inode *ip = XFS_I(inode);
587 struct xfs_mount *mp = ip->i_mount;
589 int unaligned_io = 0;
591 size_t count = iov_iter_count(from);
592 loff_t pos = iocb->ki_pos;
593 struct xfs_buftarg *target = XFS_IS_REALTIME_INODE(ip) ?
594 mp->m_rtdev_targp : mp->m_ddev_targp;
596 /* DIO must be aligned to device logical sector size */
597 if ((pos | count) & target->bt_logical_sectormask)
600 /* "unaligned" here means not aligned to a filesystem block */
601 if ((pos & mp->m_blockmask) || ((pos + count) & mp->m_blockmask))
605 * We don't need to take an exclusive lock unless there page cache needs
606 * to be invalidated or unaligned IO is being executed. We don't need to
607 * consider the EOF extension case here because
608 * xfs_file_aio_write_checks() will relock the inode as necessary for
609 * EOF zeroing cases and fill out the new inode size as appropriate.
611 if (unaligned_io || mapping->nrpages)
612 iolock = XFS_IOLOCK_EXCL;
614 iolock = XFS_IOLOCK_SHARED;
615 xfs_rw_ilock(ip, iolock);
618 * Recheck if there are cached pages that need invalidate after we got
619 * the iolock to protect against other threads adding new pages while
620 * we were waiting for the iolock.
622 if (mapping->nrpages && iolock == XFS_IOLOCK_SHARED) {
623 xfs_rw_iunlock(ip, iolock);
624 iolock = XFS_IOLOCK_EXCL;
625 xfs_rw_ilock(ip, iolock);
628 ret = xfs_file_aio_write_checks(file, &pos, &count, &iolock);
631 iov_iter_truncate(from, count);
633 if (mapping->nrpages) {
634 ret = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
638 truncate_pagecache_range(VFS_I(ip), pos, -1);
642 * If we are doing unaligned IO, wait for all other IO to drain,
643 * otherwise demote the lock if we had to flush cached pages
646 inode_dio_wait(inode);
647 else if (iolock == XFS_IOLOCK_EXCL) {
648 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
649 iolock = XFS_IOLOCK_SHARED;
652 trace_xfs_file_direct_write(ip, count, iocb->ki_pos, 0);
653 ret = generic_file_direct_write(iocb, from, pos);
656 xfs_rw_iunlock(ip, iolock);
658 /* No fallback to buffered IO on errors for XFS. */
659 ASSERT(ret < 0 || ret == count);
664 xfs_file_buffered_aio_write(
666 struct iov_iter *from)
668 struct file *file = iocb->ki_filp;
669 struct address_space *mapping = file->f_mapping;
670 struct inode *inode = mapping->host;
671 struct xfs_inode *ip = XFS_I(inode);
674 int iolock = XFS_IOLOCK_EXCL;
675 loff_t pos = iocb->ki_pos;
676 size_t count = iov_iter_count(from);
678 xfs_rw_ilock(ip, iolock);
680 ret = xfs_file_aio_write_checks(file, &pos, &count, &iolock);
684 iov_iter_truncate(from, count);
685 /* We can write back this queue in page reclaim */
686 current->backing_dev_info = mapping->backing_dev_info;
689 trace_xfs_file_buffered_write(ip, count, iocb->ki_pos, 0);
690 ret = generic_perform_write(file, from, pos);
691 if (likely(ret >= 0))
692 iocb->ki_pos = pos + ret;
695 * If we hit a space limit, try to free up some lingering preallocated
696 * space before returning an error. In the case of ENOSPC, first try to
697 * write back all dirty inodes to free up some of the excess reserved
698 * metadata space. This reduces the chances that the eofblocks scan
699 * waits on dirty mappings. Since xfs_flush_inodes() is serialized, this
700 * also behaves as a filter to prevent too many eofblocks scans from
701 * running at the same time.
703 if (ret == -EDQUOT && !enospc) {
704 enospc = xfs_inode_free_quota_eofblocks(ip);
707 } else if (ret == -ENOSPC && !enospc) {
708 struct xfs_eofblocks eofb = {0};
711 xfs_flush_inodes(ip->i_mount);
712 eofb.eof_scan_owner = ip->i_ino; /* for locking */
713 eofb.eof_flags = XFS_EOF_FLAGS_SYNC;
714 xfs_icache_free_eofblocks(ip->i_mount, &eofb);
718 current->backing_dev_info = NULL;
720 xfs_rw_iunlock(ip, iolock);
727 struct iov_iter *from)
729 struct file *file = iocb->ki_filp;
730 struct address_space *mapping = file->f_mapping;
731 struct inode *inode = mapping->host;
732 struct xfs_inode *ip = XFS_I(inode);
734 size_t ocount = iov_iter_count(from);
736 XFS_STATS_INC(xs_write_calls);
741 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
744 if (unlikely(file->f_flags & O_DIRECT))
745 ret = xfs_file_dio_aio_write(iocb, from);
747 ret = xfs_file_buffered_aio_write(iocb, from);
752 XFS_STATS_ADD(xs_write_bytes, ret);
754 /* Handle various SYNC-type writes */
755 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
769 struct inode *inode = file_inode(file);
770 struct xfs_inode *ip = XFS_I(inode);
771 struct xfs_trans *tp;
775 if (!S_ISREG(inode->i_mode))
777 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
778 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE))
781 xfs_ilock(ip, XFS_IOLOCK_EXCL);
782 if (mode & FALLOC_FL_PUNCH_HOLE) {
783 error = xfs_free_file_space(ip, offset, len);
786 } else if (mode & FALLOC_FL_COLLAPSE_RANGE) {
787 unsigned blksize_mask = (1 << inode->i_blkbits) - 1;
789 if (offset & blksize_mask || len & blksize_mask) {
795 * There is no need to overlap collapse range with EOF,
796 * in which case it is effectively a truncate operation
798 if (offset + len >= i_size_read(inode)) {
803 new_size = i_size_read(inode) - len;
805 error = xfs_collapse_file_space(ip, offset, len);
809 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
810 offset + len > i_size_read(inode)) {
811 new_size = offset + len;
812 error = inode_newsize_ok(inode, new_size);
817 if (mode & FALLOC_FL_ZERO_RANGE)
818 error = xfs_zero_file_space(ip, offset, len);
820 error = xfs_alloc_file_space(ip, offset, len,
826 tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_WRITEID);
827 error = xfs_trans_reserve(tp, &M_RES(ip->i_mount)->tr_writeid, 0, 0);
829 xfs_trans_cancel(tp, 0);
833 xfs_ilock(ip, XFS_ILOCK_EXCL);
834 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
835 ip->i_d.di_mode &= ~S_ISUID;
836 if (ip->i_d.di_mode & S_IXGRP)
837 ip->i_d.di_mode &= ~S_ISGID;
839 if (!(mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_COLLAPSE_RANGE)))
840 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
842 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
843 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
845 if (file->f_flags & O_DSYNC)
846 xfs_trans_set_sync(tp);
847 error = xfs_trans_commit(tp, 0);
851 /* Change file size if needed */
855 iattr.ia_valid = ATTR_SIZE;
856 iattr.ia_size = new_size;
857 error = xfs_setattr_size(ip, &iattr);
861 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
871 if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
873 if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb)))
883 struct xfs_inode *ip = XFS_I(inode);
887 error = xfs_file_open(inode, file);
892 * If there are any blocks, read-ahead block 0 as we're almost
893 * certain to have the next operation be a read there.
895 mode = xfs_ilock_data_map_shared(ip);
896 if (ip->i_d.di_nextents > 0)
897 xfs_dir3_data_readahead(ip, 0, -1);
898 xfs_iunlock(ip, mode);
907 return xfs_release(XFS_I(inode));
913 struct dir_context *ctx)
915 struct inode *inode = file_inode(file);
916 xfs_inode_t *ip = XFS_I(inode);
921 * The Linux API doesn't pass down the total size of the buffer
922 * we read into down to the filesystem. With the filldir concept
923 * it's not needed for correct information, but the XFS dir2 leaf
924 * code wants an estimate of the buffer size to calculate it's
925 * readahead window and size the buffers used for mapping to
928 * Try to give it an estimate that's good enough, maybe at some
929 * point we can change the ->readdir prototype to include the
930 * buffer size. For now we use the current glibc buffer size.
932 bufsize = (size_t)min_t(loff_t, 32768, ip->i_d.di_size);
934 error = xfs_readdir(ip, ctx, bufsize);
943 struct vm_area_struct *vma)
945 vma->vm_ops = &xfs_file_vm_ops;
952 * mmap()d file has taken write protection fault and is being made
953 * writable. We can set the page state up correctly for a writable
954 * page, which means we can do correct delalloc accounting (ENOSPC
955 * checking!) and unwritten extent mapping.
959 struct vm_area_struct *vma,
960 struct vm_fault *vmf)
962 return block_page_mkwrite(vma, vmf, xfs_get_blocks);
966 * This type is designed to indicate the type of offset we would like
967 * to search from page cache for either xfs_seek_data() or xfs_seek_hole().
975 * Lookup the desired type of offset from the given page.
977 * On success, return true and the offset argument will point to the
978 * start of the region that was found. Otherwise this function will
979 * return false and keep the offset argument unchanged.
982 xfs_lookup_buffer_offset(
987 loff_t lastoff = page_offset(page);
989 struct buffer_head *bh, *head;
991 bh = head = page_buffers(page);
994 * Unwritten extents that have data in the page
995 * cache covering them can be identified by the
996 * BH_Unwritten state flag. Pages with multiple
997 * buffers might have a mix of holes, data and
998 * unwritten extents - any buffer with valid
999 * data in it should have BH_Uptodate flag set
1002 if (buffer_unwritten(bh) ||
1003 buffer_uptodate(bh)) {
1004 if (type == DATA_OFF)
1007 if (type == HOLE_OFF)
1015 lastoff += bh->b_size;
1016 } while ((bh = bh->b_this_page) != head);
1022 * This routine is called to find out and return a data or hole offset
1023 * from the page cache for unwritten extents according to the desired
1024 * type for xfs_seek_data() or xfs_seek_hole().
1026 * The argument offset is used to tell where we start to search from the
1027 * page cache. Map is used to figure out the end points of the range to
1030 * Return true if the desired type of offset was found, and the argument
1031 * offset is filled with that address. Otherwise, return false and keep
1035 xfs_find_get_desired_pgoff(
1036 struct inode *inode,
1037 struct xfs_bmbt_irec *map,
1041 struct xfs_inode *ip = XFS_I(inode);
1042 struct xfs_mount *mp = ip->i_mount;
1043 struct pagevec pvec;
1047 loff_t startoff = *offset;
1048 loff_t lastoff = startoff;
1051 pagevec_init(&pvec, 0);
1053 index = startoff >> PAGE_CACHE_SHIFT;
1054 endoff = XFS_FSB_TO_B(mp, map->br_startoff + map->br_blockcount);
1055 end = endoff >> PAGE_CACHE_SHIFT;
1061 want = min_t(pgoff_t, end - index, PAGEVEC_SIZE);
1062 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index,
1065 * No page mapped into given range. If we are searching holes
1066 * and if this is the first time we got into the loop, it means
1067 * that the given offset is landed in a hole, return it.
1069 * If we have already stepped through some block buffers to find
1070 * holes but they all contains data. In this case, the last
1071 * offset is already updated and pointed to the end of the last
1072 * mapped page, if it does not reach the endpoint to search,
1073 * that means there should be a hole between them.
1075 if (nr_pages == 0) {
1076 /* Data search found nothing */
1077 if (type == DATA_OFF)
1080 ASSERT(type == HOLE_OFF);
1081 if (lastoff == startoff || lastoff < endoff) {
1089 * At lease we found one page. If this is the first time we
1090 * step into the loop, and if the first page index offset is
1091 * greater than the given search offset, a hole was found.
1093 if (type == HOLE_OFF && lastoff == startoff &&
1094 lastoff < page_offset(pvec.pages[0])) {
1099 for (i = 0; i < nr_pages; i++) {
1100 struct page *page = pvec.pages[i];
1104 * At this point, the page may be truncated or
1105 * invalidated (changing page->mapping to NULL),
1106 * or even swizzled back from swapper_space to tmpfs
1107 * file mapping. However, page->index will not change
1108 * because we have a reference on the page.
1110 * Searching done if the page index is out of range.
1111 * If the current offset is not reaches the end of
1112 * the specified search range, there should be a hole
1115 if (page->index > end) {
1116 if (type == HOLE_OFF && lastoff < endoff) {
1125 * Page truncated or invalidated(page->mapping == NULL).
1126 * We can freely skip it and proceed to check the next
1129 if (unlikely(page->mapping != inode->i_mapping)) {
1134 if (!page_has_buffers(page)) {
1139 found = xfs_lookup_buffer_offset(page, &b_offset, type);
1142 * The found offset may be less than the start
1143 * point to search if this is the first time to
1146 *offset = max_t(loff_t, startoff, b_offset);
1152 * We either searching data but nothing was found, or
1153 * searching hole but found a data buffer. In either
1154 * case, probably the next page contains the desired
1155 * things, update the last offset to it so.
1157 lastoff = page_offset(page) + PAGE_SIZE;
1162 * The number of returned pages less than our desired, search
1163 * done. In this case, nothing was found for searching data,
1164 * but we found a hole behind the last offset.
1166 if (nr_pages < want) {
1167 if (type == HOLE_OFF) {
1174 index = pvec.pages[i - 1]->index + 1;
1175 pagevec_release(&pvec);
1176 } while (index <= end);
1179 pagevec_release(&pvec);
1188 struct inode *inode = file->f_mapping->host;
1189 struct xfs_inode *ip = XFS_I(inode);
1190 struct xfs_mount *mp = ip->i_mount;
1191 loff_t uninitialized_var(offset);
1193 xfs_fileoff_t fsbno;
1198 lock = xfs_ilock_data_map_shared(ip);
1200 isize = i_size_read(inode);
1201 if (start >= isize) {
1207 * Try to read extents from the first block indicated
1208 * by fsbno to the end block of the file.
1210 fsbno = XFS_B_TO_FSBT(mp, start);
1211 end = XFS_B_TO_FSB(mp, isize);
1213 struct xfs_bmbt_irec map[2];
1217 error = xfs_bmapi_read(ip, fsbno, end - fsbno, map, &nmap,
1222 /* No extents at given offset, must be beyond EOF */
1228 for (i = 0; i < nmap; i++) {
1229 offset = max_t(loff_t, start,
1230 XFS_FSB_TO_B(mp, map[i].br_startoff));
1232 /* Landed in a data extent */
1233 if (map[i].br_startblock == DELAYSTARTBLOCK ||
1234 (map[i].br_state == XFS_EXT_NORM &&
1235 !isnullstartblock(map[i].br_startblock)))
1239 * Landed in an unwritten extent, try to search data
1242 if (map[i].br_state == XFS_EXT_UNWRITTEN) {
1243 if (xfs_find_get_desired_pgoff(inode, &map[i],
1250 * map[0] is hole or its an unwritten extent but
1251 * without data in page cache. Probably means that
1252 * we are reading after EOF if nothing in map[1].
1262 * Nothing was found, proceed to the next round of search
1263 * if reading offset not beyond or hit EOF.
1265 fsbno = map[i - 1].br_startoff + map[i - 1].br_blockcount;
1266 start = XFS_FSB_TO_B(mp, fsbno);
1267 if (start >= isize) {
1274 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
1277 xfs_iunlock(ip, lock);
1289 struct inode *inode = file->f_mapping->host;
1290 struct xfs_inode *ip = XFS_I(inode);
1291 struct xfs_mount *mp = ip->i_mount;
1292 loff_t uninitialized_var(offset);
1294 xfs_fileoff_t fsbno;
1299 if (XFS_FORCED_SHUTDOWN(mp))
1302 lock = xfs_ilock_data_map_shared(ip);
1304 isize = i_size_read(inode);
1305 if (start >= isize) {
1310 fsbno = XFS_B_TO_FSBT(mp, start);
1311 end = XFS_B_TO_FSB(mp, isize);
1314 struct xfs_bmbt_irec map[2];
1318 error = xfs_bmapi_read(ip, fsbno, end - fsbno, map, &nmap,
1323 /* No extents at given offset, must be beyond EOF */
1329 for (i = 0; i < nmap; i++) {
1330 offset = max_t(loff_t, start,
1331 XFS_FSB_TO_B(mp, map[i].br_startoff));
1333 /* Landed in a hole */
1334 if (map[i].br_startblock == HOLESTARTBLOCK)
1338 * Landed in an unwritten extent, try to search hole
1341 if (map[i].br_state == XFS_EXT_UNWRITTEN) {
1342 if (xfs_find_get_desired_pgoff(inode, &map[i],
1349 * map[0] contains data or its unwritten but contains
1350 * data in page cache, probably means that we are
1351 * reading after EOF. We should fix offset to point
1352 * to the end of the file(i.e., there is an implicit
1353 * hole at the end of any file).
1363 * Both mappings contains data, proceed to the next round of
1364 * search if the current reading offset not beyond or hit EOF.
1366 fsbno = map[i - 1].br_startoff + map[i - 1].br_blockcount;
1367 start = XFS_FSB_TO_B(mp, fsbno);
1368 if (start >= isize) {
1376 * At this point, we must have found a hole. However, the returned
1377 * offset may be bigger than the file size as it may be aligned to
1378 * page boundary for unwritten extents, we need to deal with this
1379 * situation in particular.
1381 offset = min_t(loff_t, offset, isize);
1382 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
1385 xfs_iunlock(ip, lock);
1402 return generic_file_llseek(file, offset, origin);
1404 return xfs_seek_data(file, offset);
1406 return xfs_seek_hole(file, offset);
1412 const struct file_operations xfs_file_operations = {
1413 .llseek = xfs_file_llseek,
1414 .read = new_sync_read,
1415 .write = new_sync_write,
1416 .read_iter = xfs_file_read_iter,
1417 .write_iter = xfs_file_write_iter,
1418 .splice_read = xfs_file_splice_read,
1419 .splice_write = iter_file_splice_write,
1420 .unlocked_ioctl = xfs_file_ioctl,
1421 #ifdef CONFIG_COMPAT
1422 .compat_ioctl = xfs_file_compat_ioctl,
1424 .mmap = xfs_file_mmap,
1425 .open = xfs_file_open,
1426 .release = xfs_file_release,
1427 .fsync = xfs_file_fsync,
1428 .fallocate = xfs_file_fallocate,
1431 const struct file_operations xfs_dir_file_operations = {
1432 .open = xfs_dir_open,
1433 .read = generic_read_dir,
1434 .iterate = xfs_file_readdir,
1435 .llseek = generic_file_llseek,
1436 .unlocked_ioctl = xfs_file_ioctl,
1437 #ifdef CONFIG_COMPAT
1438 .compat_ioctl = xfs_file_compat_ioctl,
1440 .fsync = xfs_dir_fsync,
1443 static const struct vm_operations_struct xfs_file_vm_ops = {
1444 .fault = filemap_fault,
1445 .map_pages = filemap_map_pages,
1446 .page_mkwrite = xfs_vm_page_mkwrite,
1447 .remap_pages = generic_file_remap_pages,