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"
42 #include <linux/aio.h>
43 #include <linux/dcache.h>
44 #include <linux/falloc.h>
45 #include <linux/pagevec.h>
47 static const struct vm_operations_struct xfs_file_vm_ops;
50 * Locking primitives for read and write IO paths to ensure we consistently use
51 * and order the inode->i_mutex, ip->i_lock and ip->i_iolock.
58 if (type & XFS_IOLOCK_EXCL)
59 mutex_lock(&VFS_I(ip)->i_mutex);
68 xfs_iunlock(ip, type);
69 if (type & XFS_IOLOCK_EXCL)
70 mutex_unlock(&VFS_I(ip)->i_mutex);
78 xfs_ilock_demote(ip, type);
79 if (type & XFS_IOLOCK_EXCL)
80 mutex_unlock(&VFS_I(ip)->i_mutex);
86 * xfs_iozero clears the specified range of buffer supplied,
87 * and marks all the affected blocks as valid and modified. If
88 * an affected block is not allocated, it will be allocated. If
89 * an affected block is not completely overwritten, and is not
90 * valid before the operation, it will be read from disk before
91 * being partially zeroed.
95 struct xfs_inode *ip, /* inode */
96 loff_t pos, /* offset in file */
97 size_t count) /* size of data to zero */
100 struct address_space *mapping;
103 mapping = VFS_I(ip)->i_mapping;
105 unsigned offset, bytes;
108 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
109 bytes = PAGE_CACHE_SIZE - offset;
113 status = pagecache_write_begin(NULL, mapping, pos, bytes,
114 AOP_FLAG_UNINTERRUPTIBLE,
119 zero_user(page, offset, bytes);
121 status = pagecache_write_end(NULL, mapping, pos, bytes, bytes,
123 WARN_ON(status <= 0); /* can't return less than zero! */
133 * Fsync operations on directories are much simpler than on regular files,
134 * as there is no file data to flush, and thus also no need for explicit
135 * cache flush operations, and there are no non-transaction metadata updates
136 * on directories either.
145 struct xfs_inode *ip = XFS_I(file->f_mapping->host);
146 struct xfs_mount *mp = ip->i_mount;
149 trace_xfs_dir_fsync(ip);
151 xfs_ilock(ip, XFS_ILOCK_SHARED);
152 if (xfs_ipincount(ip))
153 lsn = ip->i_itemp->ili_last_lsn;
154 xfs_iunlock(ip, XFS_ILOCK_SHARED);
158 return -_xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, NULL);
168 struct inode *inode = file->f_mapping->host;
169 struct xfs_inode *ip = XFS_I(inode);
170 struct xfs_mount *mp = ip->i_mount;
175 trace_xfs_file_fsync(ip);
177 error = filemap_write_and_wait_range(inode->i_mapping, start, end);
181 if (XFS_FORCED_SHUTDOWN(mp))
182 return -XFS_ERROR(EIO);
184 xfs_iflags_clear(ip, XFS_ITRUNCATED);
186 if (mp->m_flags & XFS_MOUNT_BARRIER) {
188 * If we have an RT and/or log subvolume we need to make sure
189 * to flush the write cache the device used for file data
190 * first. This is to ensure newly written file data make
191 * it to disk before logging the new inode size in case of
192 * an extending write.
194 if (XFS_IS_REALTIME_INODE(ip))
195 xfs_blkdev_issue_flush(mp->m_rtdev_targp);
196 else if (mp->m_logdev_targp != mp->m_ddev_targp)
197 xfs_blkdev_issue_flush(mp->m_ddev_targp);
201 * All metadata updates are logged, which means that we just have
202 * to flush the log up to the latest LSN that touched the inode.
204 xfs_ilock(ip, XFS_ILOCK_SHARED);
205 if (xfs_ipincount(ip)) {
207 (ip->i_itemp->ili_fields & ~XFS_ILOG_TIMESTAMP))
208 lsn = ip->i_itemp->ili_last_lsn;
210 xfs_iunlock(ip, XFS_ILOCK_SHARED);
213 error = _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, &log_flushed);
216 * If we only have a single device, and the log force about was
217 * a no-op we might have to flush the data device cache here.
218 * This can only happen for fdatasync/O_DSYNC if we were overwriting
219 * an already allocated file and thus do not have any metadata to
222 if ((mp->m_flags & XFS_MOUNT_BARRIER) &&
223 mp->m_logdev_targp == mp->m_ddev_targp &&
224 !XFS_IS_REALTIME_INODE(ip) &&
226 xfs_blkdev_issue_flush(mp->m_ddev_targp);
236 struct file *file = iocb->ki_filp;
237 struct inode *inode = file->f_mapping->host;
238 struct xfs_inode *ip = XFS_I(inode);
239 struct xfs_mount *mp = ip->i_mount;
240 size_t size = iov_iter_count(to);
244 loff_t pos = iocb->ki_pos;
246 XFS_STATS_INC(xs_read_calls);
248 if (unlikely(file->f_flags & O_DIRECT))
249 ioflags |= IO_ISDIRECT;
250 if (file->f_mode & FMODE_NOCMTIME)
253 if (unlikely(ioflags & IO_ISDIRECT)) {
254 xfs_buftarg_t *target =
255 XFS_IS_REALTIME_INODE(ip) ?
256 mp->m_rtdev_targp : mp->m_ddev_targp;
257 /* DIO must be aligned to device logical sector size */
258 if ((pos | size) & target->bt_logical_sectormask) {
259 if (pos == i_size_read(inode))
261 return -XFS_ERROR(EINVAL);
265 n = mp->m_super->s_maxbytes - pos;
266 if (n <= 0 || size == 0)
272 if (XFS_FORCED_SHUTDOWN(mp))
276 * Locking is a bit tricky here. If we take an exclusive lock
277 * for direct IO, we effectively serialise all new concurrent
278 * read IO to this file and block it behind IO that is currently in
279 * progress because IO in progress holds the IO lock shared. We only
280 * need to hold the lock exclusive to blow away the page cache, so
281 * only take lock exclusively if the page cache needs invalidation.
282 * This allows the normal direct IO case of no page cache pages to
283 * proceeed concurrently without serialisation.
285 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
286 if ((ioflags & IO_ISDIRECT) && inode->i_mapping->nrpages) {
287 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
288 xfs_rw_ilock(ip, XFS_IOLOCK_EXCL);
290 if (inode->i_mapping->nrpages) {
291 ret = filemap_write_and_wait_range(
292 VFS_I(ip)->i_mapping,
295 xfs_rw_iunlock(ip, XFS_IOLOCK_EXCL);
298 truncate_pagecache_range(VFS_I(ip), pos, -1);
300 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
303 trace_xfs_file_read(ip, size, pos, ioflags);
305 ret = generic_file_read_iter(iocb, to);
307 XFS_STATS_ADD(xs_read_bytes, ret);
309 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
314 xfs_file_splice_read(
317 struct pipe_inode_info *pipe,
321 struct xfs_inode *ip = XFS_I(infilp->f_mapping->host);
325 XFS_STATS_INC(xs_read_calls);
327 if (infilp->f_mode & FMODE_NOCMTIME)
330 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
333 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
335 trace_xfs_file_splice_read(ip, count, *ppos, ioflags);
337 ret = generic_file_splice_read(infilp, ppos, pipe, count, flags);
339 XFS_STATS_ADD(xs_read_bytes, ret);
341 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
346 * This routine is called to handle zeroing any space in the last block of the
347 * file that is beyond the EOF. We do this since the size is being increased
348 * without writing anything to that block and we don't want to read the
349 * garbage on the disk.
351 STATIC int /* error (positive) */
353 struct xfs_inode *ip,
357 struct xfs_mount *mp = ip->i_mount;
358 xfs_fileoff_t last_fsb = XFS_B_TO_FSBT(mp, isize);
359 int zero_offset = XFS_B_FSB_OFFSET(mp, isize);
363 struct xfs_bmbt_irec imap;
365 xfs_ilock(ip, XFS_ILOCK_EXCL);
366 error = xfs_bmapi_read(ip, last_fsb, 1, &imap, &nimaps, 0);
367 xfs_iunlock(ip, XFS_ILOCK_EXCL);
374 * If the block underlying isize is just a hole, then there
375 * is nothing to zero.
377 if (imap.br_startblock == HOLESTARTBLOCK)
380 zero_len = mp->m_sb.sb_blocksize - zero_offset;
381 if (isize + zero_len > offset)
382 zero_len = offset - isize;
383 return xfs_iozero(ip, isize, zero_len);
387 * Zero any on disk space between the current EOF and the new, larger EOF.
389 * This handles the normal case of zeroing the remainder of the last block in
390 * the file and the unusual case of zeroing blocks out beyond the size of the
391 * file. This second case only happens with fixed size extents and when the
392 * system crashes before the inode size was updated but after blocks were
395 * Expects the iolock to be held exclusive, and will take the ilock internally.
397 int /* error (positive) */
399 struct xfs_inode *ip,
400 xfs_off_t offset, /* starting I/O offset */
401 xfs_fsize_t isize) /* current inode size */
403 struct xfs_mount *mp = ip->i_mount;
404 xfs_fileoff_t start_zero_fsb;
405 xfs_fileoff_t end_zero_fsb;
406 xfs_fileoff_t zero_count_fsb;
407 xfs_fileoff_t last_fsb;
408 xfs_fileoff_t zero_off;
409 xfs_fsize_t zero_len;
412 struct xfs_bmbt_irec imap;
414 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
415 ASSERT(offset > isize);
418 * First handle zeroing the block on which isize resides.
420 * We only zero a part of that block so it is handled specially.
422 if (XFS_B_FSB_OFFSET(mp, isize) != 0) {
423 error = xfs_zero_last_block(ip, offset, isize);
429 * Calculate the range between the new size and the old where blocks
430 * needing to be zeroed may exist.
432 * To get the block where the last byte in the file currently resides,
433 * we need to subtract one from the size and truncate back to a block
434 * boundary. We subtract 1 in case the size is exactly on a block
437 last_fsb = isize ? XFS_B_TO_FSBT(mp, isize - 1) : (xfs_fileoff_t)-1;
438 start_zero_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
439 end_zero_fsb = XFS_B_TO_FSBT(mp, offset - 1);
440 ASSERT((xfs_sfiloff_t)last_fsb < (xfs_sfiloff_t)start_zero_fsb);
441 if (last_fsb == end_zero_fsb) {
443 * The size was only incremented on its last block.
444 * We took care of that above, so just return.
449 ASSERT(start_zero_fsb <= end_zero_fsb);
450 while (start_zero_fsb <= end_zero_fsb) {
452 zero_count_fsb = end_zero_fsb - start_zero_fsb + 1;
454 xfs_ilock(ip, XFS_ILOCK_EXCL);
455 error = xfs_bmapi_read(ip, start_zero_fsb, zero_count_fsb,
457 xfs_iunlock(ip, XFS_ILOCK_EXCL);
463 if (imap.br_state == XFS_EXT_UNWRITTEN ||
464 imap.br_startblock == HOLESTARTBLOCK) {
465 start_zero_fsb = imap.br_startoff + imap.br_blockcount;
466 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
471 * There are blocks we need to zero.
473 zero_off = XFS_FSB_TO_B(mp, start_zero_fsb);
474 zero_len = XFS_FSB_TO_B(mp, imap.br_blockcount);
476 if ((zero_off + zero_len) > offset)
477 zero_len = offset - zero_off;
479 error = xfs_iozero(ip, zero_off, zero_len);
483 start_zero_fsb = imap.br_startoff + imap.br_blockcount;
484 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
491 * Common pre-write limit and setup checks.
493 * Called with the iolocked held either shared and exclusive according to
494 * @iolock, and returns with it held. Might upgrade the iolock to exclusive
495 * if called for a direct write beyond i_size.
498 xfs_file_aio_write_checks(
504 struct inode *inode = file->f_mapping->host;
505 struct xfs_inode *ip = XFS_I(inode);
509 error = generic_write_checks(file, pos, count, S_ISBLK(inode->i_mode));
514 * If the offset is beyond the size of the file, we need to zero any
515 * blocks that fall between the existing EOF and the start of this
516 * write. If zeroing is needed and we are currently holding the
517 * iolock shared, we need to update it to exclusive which implies
518 * having to redo all checks before.
520 if (*pos > i_size_read(inode)) {
521 if (*iolock == XFS_IOLOCK_SHARED) {
522 xfs_rw_iunlock(ip, *iolock);
523 *iolock = XFS_IOLOCK_EXCL;
524 xfs_rw_ilock(ip, *iolock);
527 error = -xfs_zero_eof(ip, *pos, i_size_read(inode));
533 * Updating the timestamps will grab the ilock again from
534 * xfs_fs_dirty_inode, so we have to call it after dropping the
535 * lock above. Eventually we should look into a way to avoid
536 * the pointless lock roundtrip.
538 if (likely(!(file->f_mode & FMODE_NOCMTIME))) {
539 error = file_update_time(file);
545 * If we're writing the file then make sure to clear the setuid and
546 * setgid bits if the process is not being run by root. This keeps
547 * people from modifying setuid and setgid binaries.
549 return file_remove_suid(file);
553 * xfs_file_dio_aio_write - handle direct IO writes
555 * Lock the inode appropriately to prepare for and issue a direct IO write.
556 * By separating it from the buffered write path we remove all the tricky to
557 * follow locking changes and looping.
559 * If there are cached pages or we're extending the file, we need IOLOCK_EXCL
560 * until we're sure the bytes at the new EOF have been zeroed and/or the cached
561 * pages are flushed out.
563 * In most cases the direct IO writes will be done holding IOLOCK_SHARED
564 * allowing them to be done in parallel with reads and other direct IO writes.
565 * However, if the IO is not aligned to filesystem blocks, the direct IO layer
566 * needs to do sub-block zeroing and that requires serialisation against other
567 * direct IOs to the same block. In this case we need to serialise the
568 * submission of the unaligned IOs so that we don't get racing block zeroing in
569 * the dio layer. To avoid the problem with aio, we also need to wait for
570 * outstanding IOs to complete so that unwritten extent conversion is completed
571 * before we try to map the overlapping block. This is currently implemented by
572 * hitting it with a big hammer (i.e. inode_dio_wait()).
574 * Returns with locks held indicated by @iolock and errors indicated by
575 * negative return values.
578 xfs_file_dio_aio_write(
580 struct iov_iter *from)
582 struct file *file = iocb->ki_filp;
583 struct address_space *mapping = file->f_mapping;
584 struct inode *inode = mapping->host;
585 struct xfs_inode *ip = XFS_I(inode);
586 struct xfs_mount *mp = ip->i_mount;
588 int unaligned_io = 0;
590 size_t count = iov_iter_count(from);
591 loff_t pos = iocb->ki_pos;
592 struct xfs_buftarg *target = XFS_IS_REALTIME_INODE(ip) ?
593 mp->m_rtdev_targp : mp->m_ddev_targp;
595 /* DIO must be aligned to device logical sector size */
596 if ((pos | count) & target->bt_logical_sectormask)
597 return -XFS_ERROR(EINVAL);
599 /* "unaligned" here means not aligned to a filesystem block */
600 if ((pos & mp->m_blockmask) || ((pos + count) & mp->m_blockmask))
604 * We don't need to take an exclusive lock unless there page cache needs
605 * to be invalidated or unaligned IO is being executed. We don't need to
606 * consider the EOF extension case here because
607 * xfs_file_aio_write_checks() will relock the inode as necessary for
608 * EOF zeroing cases and fill out the new inode size as appropriate.
610 if (unaligned_io || mapping->nrpages)
611 iolock = XFS_IOLOCK_EXCL;
613 iolock = XFS_IOLOCK_SHARED;
614 xfs_rw_ilock(ip, iolock);
617 * Recheck if there are cached pages that need invalidate after we got
618 * the iolock to protect against other threads adding new pages while
619 * we were waiting for the iolock.
621 if (mapping->nrpages && iolock == XFS_IOLOCK_SHARED) {
622 xfs_rw_iunlock(ip, iolock);
623 iolock = XFS_IOLOCK_EXCL;
624 xfs_rw_ilock(ip, iolock);
627 ret = xfs_file_aio_write_checks(file, &pos, &count, &iolock);
630 iov_iter_truncate(from, count);
632 if (mapping->nrpages) {
633 ret = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
637 truncate_pagecache_range(VFS_I(ip), pos, -1);
641 * If we are doing unaligned IO, wait for all other IO to drain,
642 * otherwise demote the lock if we had to flush cached pages
645 inode_dio_wait(inode);
646 else if (iolock == XFS_IOLOCK_EXCL) {
647 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
648 iolock = XFS_IOLOCK_SHARED;
651 trace_xfs_file_direct_write(ip, count, iocb->ki_pos, 0);
652 ret = generic_file_direct_write(iocb, from, pos);
655 xfs_rw_iunlock(ip, iolock);
657 /* No fallback to buffered IO on errors for XFS. */
658 ASSERT(ret < 0 || ret == count);
663 xfs_file_buffered_aio_write(
665 struct iov_iter *from)
667 struct file *file = iocb->ki_filp;
668 struct address_space *mapping = file->f_mapping;
669 struct inode *inode = mapping->host;
670 struct xfs_inode *ip = XFS_I(inode);
673 int iolock = XFS_IOLOCK_EXCL;
674 loff_t pos = iocb->ki_pos;
675 size_t count = iov_iter_count(from);
677 xfs_rw_ilock(ip, iolock);
679 ret = xfs_file_aio_write_checks(file, &pos, &count, &iolock);
683 iov_iter_truncate(from, count);
684 /* We can write back this queue in page reclaim */
685 current->backing_dev_info = mapping->backing_dev_info;
688 trace_xfs_file_buffered_write(ip, count, iocb->ki_pos, 0);
689 ret = generic_perform_write(file, from, pos);
690 if (likely(ret >= 0))
691 iocb->ki_pos = pos + ret;
693 * If we just got an ENOSPC, try to write back all dirty inodes to
694 * convert delalloc space to free up some of the excess reserved
697 if (ret == -ENOSPC && !enospc) {
699 xfs_flush_inodes(ip->i_mount);
703 current->backing_dev_info = NULL;
705 xfs_rw_iunlock(ip, iolock);
712 struct iov_iter *from)
714 struct file *file = iocb->ki_filp;
715 struct address_space *mapping = file->f_mapping;
716 struct inode *inode = mapping->host;
717 struct xfs_inode *ip = XFS_I(inode);
719 size_t ocount = iov_iter_count(from);
721 XFS_STATS_INC(xs_write_calls);
726 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
729 if (unlikely(file->f_flags & O_DIRECT))
730 ret = xfs_file_dio_aio_write(iocb, from);
732 ret = xfs_file_buffered_aio_write(iocb, from);
737 XFS_STATS_ADD(xs_write_bytes, ret);
739 /* Handle various SYNC-type writes */
740 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
754 struct inode *inode = file_inode(file);
755 struct xfs_inode *ip = XFS_I(inode);
756 struct xfs_trans *tp;
760 if (!S_ISREG(inode->i_mode))
762 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
763 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE))
766 xfs_ilock(ip, XFS_IOLOCK_EXCL);
767 if (mode & FALLOC_FL_PUNCH_HOLE) {
768 error = xfs_free_file_space(ip, offset, len);
771 } else if (mode & FALLOC_FL_COLLAPSE_RANGE) {
772 unsigned blksize_mask = (1 << inode->i_blkbits) - 1;
774 if (offset & blksize_mask || len & blksize_mask) {
780 * There is no need to overlap collapse range with EOF,
781 * in which case it is effectively a truncate operation
783 if (offset + len >= i_size_read(inode)) {
788 new_size = i_size_read(inode) - len;
790 error = xfs_collapse_file_space(ip, offset, len);
794 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
795 offset + len > i_size_read(inode)) {
796 new_size = offset + len;
797 error = -inode_newsize_ok(inode, new_size);
802 if (mode & FALLOC_FL_ZERO_RANGE)
803 error = xfs_zero_file_space(ip, offset, len);
805 error = xfs_alloc_file_space(ip, offset, len,
811 tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_WRITEID);
812 error = xfs_trans_reserve(tp, &M_RES(ip->i_mount)->tr_writeid, 0, 0);
814 xfs_trans_cancel(tp, 0);
818 xfs_ilock(ip, XFS_ILOCK_EXCL);
819 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
820 ip->i_d.di_mode &= ~S_ISUID;
821 if (ip->i_d.di_mode & S_IXGRP)
822 ip->i_d.di_mode &= ~S_ISGID;
824 if (!(mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_COLLAPSE_RANGE)))
825 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
827 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
828 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
830 if (file->f_flags & O_DSYNC)
831 xfs_trans_set_sync(tp);
832 error = xfs_trans_commit(tp, 0);
836 /* Change file size if needed */
840 iattr.ia_valid = ATTR_SIZE;
841 iattr.ia_size = new_size;
842 error = xfs_setattr_size(ip, &iattr);
846 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
856 if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
858 if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb)))
868 struct xfs_inode *ip = XFS_I(inode);
872 error = xfs_file_open(inode, file);
877 * If there are any blocks, read-ahead block 0 as we're almost
878 * certain to have the next operation be a read there.
880 mode = xfs_ilock_data_map_shared(ip);
881 if (ip->i_d.di_nextents > 0)
882 xfs_dir3_data_readahead(ip, 0, -1);
883 xfs_iunlock(ip, mode);
892 return -xfs_release(XFS_I(inode));
898 struct dir_context *ctx)
900 struct inode *inode = file_inode(file);
901 xfs_inode_t *ip = XFS_I(inode);
906 * The Linux API doesn't pass down the total size of the buffer
907 * we read into down to the filesystem. With the filldir concept
908 * it's not needed for correct information, but the XFS dir2 leaf
909 * code wants an estimate of the buffer size to calculate it's
910 * readahead window and size the buffers used for mapping to
913 * Try to give it an estimate that's good enough, maybe at some
914 * point we can change the ->readdir prototype to include the
915 * buffer size. For now we use the current glibc buffer size.
917 bufsize = (size_t)min_t(loff_t, 32768, ip->i_d.di_size);
919 error = xfs_readdir(ip, ctx, bufsize);
928 struct vm_area_struct *vma)
930 vma->vm_ops = &xfs_file_vm_ops;
937 * mmap()d file has taken write protection fault and is being made
938 * writable. We can set the page state up correctly for a writable
939 * page, which means we can do correct delalloc accounting (ENOSPC
940 * checking!) and unwritten extent mapping.
944 struct vm_area_struct *vma,
945 struct vm_fault *vmf)
947 return block_page_mkwrite(vma, vmf, xfs_get_blocks);
951 * This type is designed to indicate the type of offset we would like
952 * to search from page cache for either xfs_seek_data() or xfs_seek_hole().
960 * Lookup the desired type of offset from the given page.
962 * On success, return true and the offset argument will point to the
963 * start of the region that was found. Otherwise this function will
964 * return false and keep the offset argument unchanged.
967 xfs_lookup_buffer_offset(
972 loff_t lastoff = page_offset(page);
974 struct buffer_head *bh, *head;
976 bh = head = page_buffers(page);
979 * Unwritten extents that have data in the page
980 * cache covering them can be identified by the
981 * BH_Unwritten state flag. Pages with multiple
982 * buffers might have a mix of holes, data and
983 * unwritten extents - any buffer with valid
984 * data in it should have BH_Uptodate flag set
987 if (buffer_unwritten(bh) ||
988 buffer_uptodate(bh)) {
989 if (type == DATA_OFF)
992 if (type == HOLE_OFF)
1000 lastoff += bh->b_size;
1001 } while ((bh = bh->b_this_page) != head);
1007 * This routine is called to find out and return a data or hole offset
1008 * from the page cache for unwritten extents according to the desired
1009 * type for xfs_seek_data() or xfs_seek_hole().
1011 * The argument offset is used to tell where we start to search from the
1012 * page cache. Map is used to figure out the end points of the range to
1015 * Return true if the desired type of offset was found, and the argument
1016 * offset is filled with that address. Otherwise, return false and keep
1020 xfs_find_get_desired_pgoff(
1021 struct inode *inode,
1022 struct xfs_bmbt_irec *map,
1026 struct xfs_inode *ip = XFS_I(inode);
1027 struct xfs_mount *mp = ip->i_mount;
1028 struct pagevec pvec;
1032 loff_t startoff = *offset;
1033 loff_t lastoff = startoff;
1036 pagevec_init(&pvec, 0);
1038 index = startoff >> PAGE_CACHE_SHIFT;
1039 endoff = XFS_FSB_TO_B(mp, map->br_startoff + map->br_blockcount);
1040 end = endoff >> PAGE_CACHE_SHIFT;
1046 want = min_t(pgoff_t, end - index, PAGEVEC_SIZE);
1047 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index,
1050 * No page mapped into given range. If we are searching holes
1051 * and if this is the first time we got into the loop, it means
1052 * that the given offset is landed in a hole, return it.
1054 * If we have already stepped through some block buffers to find
1055 * holes but they all contains data. In this case, the last
1056 * offset is already updated and pointed to the end of the last
1057 * mapped page, if it does not reach the endpoint to search,
1058 * that means there should be a hole between them.
1060 if (nr_pages == 0) {
1061 /* Data search found nothing */
1062 if (type == DATA_OFF)
1065 ASSERT(type == HOLE_OFF);
1066 if (lastoff == startoff || lastoff < endoff) {
1074 * At lease we found one page. If this is the first time we
1075 * step into the loop, and if the first page index offset is
1076 * greater than the given search offset, a hole was found.
1078 if (type == HOLE_OFF && lastoff == startoff &&
1079 lastoff < page_offset(pvec.pages[0])) {
1084 for (i = 0; i < nr_pages; i++) {
1085 struct page *page = pvec.pages[i];
1089 * At this point, the page may be truncated or
1090 * invalidated (changing page->mapping to NULL),
1091 * or even swizzled back from swapper_space to tmpfs
1092 * file mapping. However, page->index will not change
1093 * because we have a reference on the page.
1095 * Searching done if the page index is out of range.
1096 * If the current offset is not reaches the end of
1097 * the specified search range, there should be a hole
1100 if (page->index > end) {
1101 if (type == HOLE_OFF && lastoff < endoff) {
1110 * Page truncated or invalidated(page->mapping == NULL).
1111 * We can freely skip it and proceed to check the next
1114 if (unlikely(page->mapping != inode->i_mapping)) {
1119 if (!page_has_buffers(page)) {
1124 found = xfs_lookup_buffer_offset(page, &b_offset, type);
1127 * The found offset may be less than the start
1128 * point to search if this is the first time to
1131 *offset = max_t(loff_t, startoff, b_offset);
1137 * We either searching data but nothing was found, or
1138 * searching hole but found a data buffer. In either
1139 * case, probably the next page contains the desired
1140 * things, update the last offset to it so.
1142 lastoff = page_offset(page) + PAGE_SIZE;
1147 * The number of returned pages less than our desired, search
1148 * done. In this case, nothing was found for searching data,
1149 * but we found a hole behind the last offset.
1151 if (nr_pages < want) {
1152 if (type == HOLE_OFF) {
1159 index = pvec.pages[i - 1]->index + 1;
1160 pagevec_release(&pvec);
1161 } while (index <= end);
1164 pagevec_release(&pvec);
1173 struct inode *inode = file->f_mapping->host;
1174 struct xfs_inode *ip = XFS_I(inode);
1175 struct xfs_mount *mp = ip->i_mount;
1176 loff_t uninitialized_var(offset);
1178 xfs_fileoff_t fsbno;
1183 lock = xfs_ilock_data_map_shared(ip);
1185 isize = i_size_read(inode);
1186 if (start >= isize) {
1192 * Try to read extents from the first block indicated
1193 * by fsbno to the end block of the file.
1195 fsbno = XFS_B_TO_FSBT(mp, start);
1196 end = XFS_B_TO_FSB(mp, isize);
1198 struct xfs_bmbt_irec map[2];
1202 error = xfs_bmapi_read(ip, fsbno, end - fsbno, map, &nmap,
1207 /* No extents at given offset, must be beyond EOF */
1213 for (i = 0; i < nmap; i++) {
1214 offset = max_t(loff_t, start,
1215 XFS_FSB_TO_B(mp, map[i].br_startoff));
1217 /* Landed in a data extent */
1218 if (map[i].br_startblock == DELAYSTARTBLOCK ||
1219 (map[i].br_state == XFS_EXT_NORM &&
1220 !isnullstartblock(map[i].br_startblock)))
1224 * Landed in an unwritten extent, try to search data
1227 if (map[i].br_state == XFS_EXT_UNWRITTEN) {
1228 if (xfs_find_get_desired_pgoff(inode, &map[i],
1235 * map[0] is hole or its an unwritten extent but
1236 * without data in page cache. Probably means that
1237 * we are reading after EOF if nothing in map[1].
1247 * Nothing was found, proceed to the next round of search
1248 * if reading offset not beyond or hit EOF.
1250 fsbno = map[i - 1].br_startoff + map[i - 1].br_blockcount;
1251 start = XFS_FSB_TO_B(mp, fsbno);
1252 if (start >= isize) {
1259 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
1262 xfs_iunlock(ip, lock);
1274 struct inode *inode = file->f_mapping->host;
1275 struct xfs_inode *ip = XFS_I(inode);
1276 struct xfs_mount *mp = ip->i_mount;
1277 loff_t uninitialized_var(offset);
1279 xfs_fileoff_t fsbno;
1284 if (XFS_FORCED_SHUTDOWN(mp))
1285 return -XFS_ERROR(EIO);
1287 lock = xfs_ilock_data_map_shared(ip);
1289 isize = i_size_read(inode);
1290 if (start >= isize) {
1295 fsbno = XFS_B_TO_FSBT(mp, start);
1296 end = XFS_B_TO_FSB(mp, isize);
1299 struct xfs_bmbt_irec map[2];
1303 error = xfs_bmapi_read(ip, fsbno, end - fsbno, map, &nmap,
1308 /* No extents at given offset, must be beyond EOF */
1314 for (i = 0; i < nmap; i++) {
1315 offset = max_t(loff_t, start,
1316 XFS_FSB_TO_B(mp, map[i].br_startoff));
1318 /* Landed in a hole */
1319 if (map[i].br_startblock == HOLESTARTBLOCK)
1323 * Landed in an unwritten extent, try to search hole
1326 if (map[i].br_state == XFS_EXT_UNWRITTEN) {
1327 if (xfs_find_get_desired_pgoff(inode, &map[i],
1334 * map[0] contains data or its unwritten but contains
1335 * data in page cache, probably means that we are
1336 * reading after EOF. We should fix offset to point
1337 * to the end of the file(i.e., there is an implicit
1338 * hole at the end of any file).
1348 * Both mappings contains data, proceed to the next round of
1349 * search if the current reading offset not beyond or hit EOF.
1351 fsbno = map[i - 1].br_startoff + map[i - 1].br_blockcount;
1352 start = XFS_FSB_TO_B(mp, fsbno);
1353 if (start >= isize) {
1361 * At this point, we must have found a hole. However, the returned
1362 * offset may be bigger than the file size as it may be aligned to
1363 * page boundary for unwritten extents, we need to deal with this
1364 * situation in particular.
1366 offset = min_t(loff_t, offset, isize);
1367 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
1370 xfs_iunlock(ip, lock);
1387 return generic_file_llseek(file, offset, origin);
1389 return xfs_seek_data(file, offset);
1391 return xfs_seek_hole(file, offset);
1397 const struct file_operations xfs_file_operations = {
1398 .llseek = xfs_file_llseek,
1399 .read = new_sync_read,
1400 .write = new_sync_write,
1401 .read_iter = xfs_file_read_iter,
1402 .write_iter = xfs_file_write_iter,
1403 .splice_read = xfs_file_splice_read,
1404 .splice_write = iter_file_splice_write,
1405 .unlocked_ioctl = xfs_file_ioctl,
1406 #ifdef CONFIG_COMPAT
1407 .compat_ioctl = xfs_file_compat_ioctl,
1409 .mmap = xfs_file_mmap,
1410 .open = xfs_file_open,
1411 .release = xfs_file_release,
1412 .fsync = xfs_file_fsync,
1413 .fallocate = xfs_file_fallocate,
1416 const struct file_operations xfs_dir_file_operations = {
1417 .open = xfs_dir_open,
1418 .read = generic_read_dir,
1419 .iterate = xfs_file_readdir,
1420 .llseek = generic_file_llseek,
1421 .unlocked_ioctl = xfs_file_ioctl,
1422 #ifdef CONFIG_COMPAT
1423 .compat_ioctl = xfs_file_compat_ioctl,
1425 .fsync = xfs_dir_fsync,
1428 static const struct vm_operations_struct xfs_file_vm_ops = {
1429 .fault = filemap_fault,
1430 .map_pages = filemap_map_pages,
1431 .page_mkwrite = xfs_vm_page_mkwrite,
1432 .remap_pages = generic_file_remap_pages,