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
23 #include "xfs_trans.h"
24 #include "xfs_mount.h"
25 #include "xfs_bmap_btree.h"
26 #include "xfs_alloc.h"
27 #include "xfs_dinode.h"
28 #include "xfs_inode.h"
29 #include "xfs_inode_item.h"
31 #include "xfs_bmap_util.h"
32 #include "xfs_error.h"
33 #include "xfs_da_btree.h"
34 #include "xfs_dir2_format.h"
36 #include "xfs_dir2_priv.h"
37 #include "xfs_ioctl.h"
38 #include "xfs_trace.h"
40 #include <linux/aio.h>
41 #include <linux/dcache.h>
42 #include <linux/falloc.h>
43 #include <linux/pagevec.h>
45 static const struct vm_operations_struct xfs_file_vm_ops;
48 * Locking primitives for read and write IO paths to ensure we consistently use
49 * and order the inode->i_mutex, ip->i_lock and ip->i_iolock.
56 if (type & XFS_IOLOCK_EXCL)
57 mutex_lock(&VFS_I(ip)->i_mutex);
66 xfs_iunlock(ip, type);
67 if (type & XFS_IOLOCK_EXCL)
68 mutex_unlock(&VFS_I(ip)->i_mutex);
76 xfs_ilock_demote(ip, type);
77 if (type & XFS_IOLOCK_EXCL)
78 mutex_unlock(&VFS_I(ip)->i_mutex);
84 * xfs_iozero clears the specified range of buffer supplied,
85 * and marks all the affected blocks as valid and modified. If
86 * an affected block is not allocated, it will be allocated. If
87 * an affected block is not completely overwritten, and is not
88 * valid before the operation, it will be read from disk before
89 * being partially zeroed.
93 struct xfs_inode *ip, /* inode */
94 loff_t pos, /* offset in file */
95 size_t count) /* size of data to zero */
98 struct address_space *mapping;
101 mapping = VFS_I(ip)->i_mapping;
103 unsigned offset, bytes;
106 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
107 bytes = PAGE_CACHE_SIZE - offset;
111 status = pagecache_write_begin(NULL, mapping, pos, bytes,
112 AOP_FLAG_UNINTERRUPTIBLE,
117 zero_user(page, offset, bytes);
119 status = pagecache_write_end(NULL, mapping, pos, bytes, bytes,
121 WARN_ON(status <= 0); /* can't return less than zero! */
131 * Fsync operations on directories are much simpler than on regular files,
132 * as there is no file data to flush, and thus also no need for explicit
133 * cache flush operations, and there are no non-transaction metadata updates
134 * on directories either.
143 struct xfs_inode *ip = XFS_I(file->f_mapping->host);
144 struct xfs_mount *mp = ip->i_mount;
147 trace_xfs_dir_fsync(ip);
149 xfs_ilock(ip, XFS_ILOCK_SHARED);
150 if (xfs_ipincount(ip))
151 lsn = ip->i_itemp->ili_last_lsn;
152 xfs_iunlock(ip, XFS_ILOCK_SHARED);
156 return _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, NULL);
166 struct inode *inode = file->f_mapping->host;
167 struct xfs_inode *ip = XFS_I(inode);
168 struct xfs_mount *mp = ip->i_mount;
173 trace_xfs_file_fsync(ip);
175 error = filemap_write_and_wait_range(inode->i_mapping, start, end);
179 if (XFS_FORCED_SHUTDOWN(mp))
180 return -XFS_ERROR(EIO);
182 xfs_iflags_clear(ip, XFS_ITRUNCATED);
184 if (mp->m_flags & XFS_MOUNT_BARRIER) {
186 * If we have an RT and/or log subvolume we need to make sure
187 * to flush the write cache the device used for file data
188 * first. This is to ensure newly written file data make
189 * it to disk before logging the new inode size in case of
190 * an extending write.
192 if (XFS_IS_REALTIME_INODE(ip))
193 xfs_blkdev_issue_flush(mp->m_rtdev_targp);
194 else if (mp->m_logdev_targp != mp->m_ddev_targp)
195 xfs_blkdev_issue_flush(mp->m_ddev_targp);
199 * All metadata updates are logged, which means that we just have
200 * to flush the log up to the latest LSN that touched the inode.
202 xfs_ilock(ip, XFS_ILOCK_SHARED);
203 if (xfs_ipincount(ip)) {
205 (ip->i_itemp->ili_fields & ~XFS_ILOG_TIMESTAMP))
206 lsn = ip->i_itemp->ili_last_lsn;
208 xfs_iunlock(ip, XFS_ILOCK_SHARED);
211 error = _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, &log_flushed);
214 * If we only have a single device, and the log force about was
215 * a no-op we might have to flush the data device cache here.
216 * This can only happen for fdatasync/O_DSYNC if we were overwriting
217 * an already allocated file and thus do not have any metadata to
220 if ((mp->m_flags & XFS_MOUNT_BARRIER) &&
221 mp->m_logdev_targp == mp->m_ddev_targp &&
222 !XFS_IS_REALTIME_INODE(ip) &&
224 xfs_blkdev_issue_flush(mp->m_ddev_targp);
232 const struct iovec *iovp,
233 unsigned long nr_segs,
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;
245 XFS_STATS_INC(xs_read_calls);
247 BUG_ON(iocb->ki_pos != pos);
249 if (unlikely(file->f_flags & O_DIRECT))
250 ioflags |= IO_ISDIRECT;
251 if (file->f_mode & FMODE_NOCMTIME)
254 ret = generic_segment_checks(iovp, &nr_segs, &size, VERIFY_WRITE);
258 if (unlikely(ioflags & IO_ISDIRECT)) {
259 xfs_buftarg_t *target =
260 XFS_IS_REALTIME_INODE(ip) ?
261 mp->m_rtdev_targp : mp->m_ddev_targp;
262 if ((pos & target->bt_smask) || (size & target->bt_smask)) {
263 if (pos == i_size_read(inode))
265 return -XFS_ERROR(EINVAL);
269 n = mp->m_super->s_maxbytes - pos;
270 if (n <= 0 || size == 0)
276 if (XFS_FORCED_SHUTDOWN(mp))
280 * Locking is a bit tricky here. If we take an exclusive lock
281 * for direct IO, we effectively serialise all new concurrent
282 * read IO to this file and block it behind IO that is currently in
283 * progress because IO in progress holds the IO lock shared. We only
284 * need to hold the lock exclusive to blow away the page cache, so
285 * only take lock exclusively if the page cache needs invalidation.
286 * This allows the normal direct IO case of no page cache pages to
287 * proceeed concurrently without serialisation.
289 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
290 if ((ioflags & IO_ISDIRECT) && inode->i_mapping->nrpages) {
291 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
292 xfs_rw_ilock(ip, XFS_IOLOCK_EXCL);
294 if (inode->i_mapping->nrpages) {
295 ret = -filemap_write_and_wait_range(
296 VFS_I(ip)->i_mapping,
299 xfs_rw_iunlock(ip, XFS_IOLOCK_EXCL);
302 truncate_pagecache_range(VFS_I(ip), pos, -1);
304 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
307 trace_xfs_file_read(ip, size, pos, ioflags);
309 ret = generic_file_aio_read(iocb, iovp, nr_segs, pos);
311 XFS_STATS_ADD(xs_read_bytes, ret);
313 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
318 xfs_file_splice_read(
321 struct pipe_inode_info *pipe,
325 struct xfs_inode *ip = XFS_I(infilp->f_mapping->host);
329 XFS_STATS_INC(xs_read_calls);
331 if (infilp->f_mode & FMODE_NOCMTIME)
334 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
337 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
339 trace_xfs_file_splice_read(ip, count, *ppos, ioflags);
341 ret = generic_file_splice_read(infilp, ppos, pipe, count, flags);
343 XFS_STATS_ADD(xs_read_bytes, ret);
345 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
350 * xfs_file_splice_write() does not use xfs_rw_ilock() because
351 * generic_file_splice_write() takes the i_mutex itself. This, in theory,
352 * couuld cause lock inversions between the aio_write path and the splice path
353 * if someone is doing concurrent splice(2) based writes and write(2) based
354 * writes to the same inode. The only real way to fix this is to re-implement
355 * the generic code here with correct locking orders.
358 xfs_file_splice_write(
359 struct pipe_inode_info *pipe,
360 struct file *outfilp,
365 struct inode *inode = outfilp->f_mapping->host;
366 struct xfs_inode *ip = XFS_I(inode);
370 XFS_STATS_INC(xs_write_calls);
372 if (outfilp->f_mode & FMODE_NOCMTIME)
375 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
378 xfs_ilock(ip, XFS_IOLOCK_EXCL);
380 trace_xfs_file_splice_write(ip, count, *ppos, ioflags);
382 ret = generic_file_splice_write(pipe, outfilp, ppos, count, flags);
384 XFS_STATS_ADD(xs_write_bytes, ret);
386 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
391 * This routine is called to handle zeroing any space in the last block of the
392 * file that is beyond the EOF. We do this since the size is being increased
393 * without writing anything to that block and we don't want to read the
394 * garbage on the disk.
396 STATIC int /* error (positive) */
398 struct xfs_inode *ip,
402 struct xfs_mount *mp = ip->i_mount;
403 xfs_fileoff_t last_fsb = XFS_B_TO_FSBT(mp, isize);
404 int zero_offset = XFS_B_FSB_OFFSET(mp, isize);
408 struct xfs_bmbt_irec imap;
410 xfs_ilock(ip, XFS_ILOCK_EXCL);
411 error = xfs_bmapi_read(ip, last_fsb, 1, &imap, &nimaps, 0);
412 xfs_iunlock(ip, XFS_ILOCK_EXCL);
419 * If the block underlying isize is just a hole, then there
420 * is nothing to zero.
422 if (imap.br_startblock == HOLESTARTBLOCK)
425 zero_len = mp->m_sb.sb_blocksize - zero_offset;
426 if (isize + zero_len > offset)
427 zero_len = offset - isize;
428 return xfs_iozero(ip, isize, zero_len);
432 * Zero any on disk space between the current EOF and the new, larger EOF.
434 * This handles the normal case of zeroing the remainder of the last block in
435 * the file and the unusual case of zeroing blocks out beyond the size of the
436 * file. This second case only happens with fixed size extents and when the
437 * system crashes before the inode size was updated but after blocks were
440 * Expects the iolock to be held exclusive, and will take the ilock internally.
442 int /* error (positive) */
444 struct xfs_inode *ip,
445 xfs_off_t offset, /* starting I/O offset */
446 xfs_fsize_t isize) /* current inode size */
448 struct xfs_mount *mp = ip->i_mount;
449 xfs_fileoff_t start_zero_fsb;
450 xfs_fileoff_t end_zero_fsb;
451 xfs_fileoff_t zero_count_fsb;
452 xfs_fileoff_t last_fsb;
453 xfs_fileoff_t zero_off;
454 xfs_fsize_t zero_len;
457 struct xfs_bmbt_irec imap;
459 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
460 ASSERT(offset > isize);
463 * First handle zeroing the block on which isize resides.
465 * We only zero a part of that block so it is handled specially.
467 if (XFS_B_FSB_OFFSET(mp, isize) != 0) {
468 error = xfs_zero_last_block(ip, offset, isize);
474 * Calculate the range between the new size and the old where blocks
475 * needing to be zeroed may exist.
477 * To get the block where the last byte in the file currently resides,
478 * we need to subtract one from the size and truncate back to a block
479 * boundary. We subtract 1 in case the size is exactly on a block
482 last_fsb = isize ? XFS_B_TO_FSBT(mp, isize - 1) : (xfs_fileoff_t)-1;
483 start_zero_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
484 end_zero_fsb = XFS_B_TO_FSBT(mp, offset - 1);
485 ASSERT((xfs_sfiloff_t)last_fsb < (xfs_sfiloff_t)start_zero_fsb);
486 if (last_fsb == end_zero_fsb) {
488 * The size was only incremented on its last block.
489 * We took care of that above, so just return.
494 ASSERT(start_zero_fsb <= end_zero_fsb);
495 while (start_zero_fsb <= end_zero_fsb) {
497 zero_count_fsb = end_zero_fsb - start_zero_fsb + 1;
499 xfs_ilock(ip, XFS_ILOCK_EXCL);
500 error = xfs_bmapi_read(ip, start_zero_fsb, zero_count_fsb,
502 xfs_iunlock(ip, XFS_ILOCK_EXCL);
508 if (imap.br_state == XFS_EXT_UNWRITTEN ||
509 imap.br_startblock == HOLESTARTBLOCK) {
510 start_zero_fsb = imap.br_startoff + imap.br_blockcount;
511 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
516 * There are blocks we need to zero.
518 zero_off = XFS_FSB_TO_B(mp, start_zero_fsb);
519 zero_len = XFS_FSB_TO_B(mp, imap.br_blockcount);
521 if ((zero_off + zero_len) > offset)
522 zero_len = offset - zero_off;
524 error = xfs_iozero(ip, zero_off, zero_len);
528 start_zero_fsb = imap.br_startoff + imap.br_blockcount;
529 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
536 * Common pre-write limit and setup checks.
538 * Called with the iolocked held either shared and exclusive according to
539 * @iolock, and returns with it held. Might upgrade the iolock to exclusive
540 * if called for a direct write beyond i_size.
543 xfs_file_aio_write_checks(
549 struct inode *inode = file->f_mapping->host;
550 struct xfs_inode *ip = XFS_I(inode);
554 error = generic_write_checks(file, pos, count, S_ISBLK(inode->i_mode));
559 * If the offset is beyond the size of the file, we need to zero any
560 * blocks that fall between the existing EOF and the start of this
561 * write. If zeroing is needed and we are currently holding the
562 * iolock shared, we need to update it to exclusive which implies
563 * having to redo all checks before.
565 if (*pos > i_size_read(inode)) {
566 if (*iolock == XFS_IOLOCK_SHARED) {
567 xfs_rw_iunlock(ip, *iolock);
568 *iolock = XFS_IOLOCK_EXCL;
569 xfs_rw_ilock(ip, *iolock);
572 error = -xfs_zero_eof(ip, *pos, i_size_read(inode));
578 * Updating the timestamps will grab the ilock again from
579 * xfs_fs_dirty_inode, so we have to call it after dropping the
580 * lock above. Eventually we should look into a way to avoid
581 * the pointless lock roundtrip.
583 if (likely(!(file->f_mode & FMODE_NOCMTIME))) {
584 error = file_update_time(file);
590 * If we're writing the file then make sure to clear the setuid and
591 * setgid bits if the process is not being run by root. This keeps
592 * people from modifying setuid and setgid binaries.
594 return file_remove_suid(file);
598 * xfs_file_dio_aio_write - handle direct IO writes
600 * Lock the inode appropriately to prepare for and issue a direct IO write.
601 * By separating it from the buffered write path we remove all the tricky to
602 * follow locking changes and looping.
604 * If there are cached pages or we're extending the file, we need IOLOCK_EXCL
605 * until we're sure the bytes at the new EOF have been zeroed and/or the cached
606 * pages are flushed out.
608 * In most cases the direct IO writes will be done holding IOLOCK_SHARED
609 * allowing them to be done in parallel with reads and other direct IO writes.
610 * However, if the IO is not aligned to filesystem blocks, the direct IO layer
611 * needs to do sub-block zeroing and that requires serialisation against other
612 * direct IOs to the same block. In this case we need to serialise the
613 * submission of the unaligned IOs so that we don't get racing block zeroing in
614 * the dio layer. To avoid the problem with aio, we also need to wait for
615 * outstanding IOs to complete so that unwritten extent conversion is completed
616 * before we try to map the overlapping block. This is currently implemented by
617 * hitting it with a big hammer (i.e. inode_dio_wait()).
619 * Returns with locks held indicated by @iolock and errors indicated by
620 * negative return values.
623 xfs_file_dio_aio_write(
625 const struct iovec *iovp,
626 unsigned long nr_segs,
630 struct file *file = iocb->ki_filp;
631 struct address_space *mapping = file->f_mapping;
632 struct inode *inode = mapping->host;
633 struct xfs_inode *ip = XFS_I(inode);
634 struct xfs_mount *mp = ip->i_mount;
636 size_t count = ocount;
637 int unaligned_io = 0;
639 struct xfs_buftarg *target = XFS_IS_REALTIME_INODE(ip) ?
640 mp->m_rtdev_targp : mp->m_ddev_targp;
642 if ((pos & target->bt_smask) || (count & target->bt_smask))
643 return -XFS_ERROR(EINVAL);
645 if ((pos & mp->m_blockmask) || ((pos + count) & mp->m_blockmask))
649 * We don't need to take an exclusive lock unless there page cache needs
650 * to be invalidated or unaligned IO is being executed. We don't need to
651 * consider the EOF extension case here because
652 * xfs_file_aio_write_checks() will relock the inode as necessary for
653 * EOF zeroing cases and fill out the new inode size as appropriate.
655 if (unaligned_io || mapping->nrpages)
656 iolock = XFS_IOLOCK_EXCL;
658 iolock = XFS_IOLOCK_SHARED;
659 xfs_rw_ilock(ip, iolock);
662 * Recheck if there are cached pages that need invalidate after we got
663 * the iolock to protect against other threads adding new pages while
664 * we were waiting for the iolock.
666 if (mapping->nrpages && iolock == XFS_IOLOCK_SHARED) {
667 xfs_rw_iunlock(ip, iolock);
668 iolock = XFS_IOLOCK_EXCL;
669 xfs_rw_ilock(ip, iolock);
672 ret = xfs_file_aio_write_checks(file, &pos, &count, &iolock);
676 if (mapping->nrpages) {
677 ret = -filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
681 truncate_pagecache_range(VFS_I(ip), pos, -1);
685 * If we are doing unaligned IO, wait for all other IO to drain,
686 * otherwise demote the lock if we had to flush cached pages
689 inode_dio_wait(inode);
690 else if (iolock == XFS_IOLOCK_EXCL) {
691 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
692 iolock = XFS_IOLOCK_SHARED;
695 trace_xfs_file_direct_write(ip, count, iocb->ki_pos, 0);
696 ret = generic_file_direct_write(iocb, iovp,
697 &nr_segs, pos, &iocb->ki_pos, count, ocount);
700 xfs_rw_iunlock(ip, iolock);
702 /* No fallback to buffered IO on errors for XFS. */
703 ASSERT(ret < 0 || ret == count);
708 xfs_file_buffered_aio_write(
710 const struct iovec *iovp,
711 unsigned long nr_segs,
715 struct file *file = iocb->ki_filp;
716 struct address_space *mapping = file->f_mapping;
717 struct inode *inode = mapping->host;
718 struct xfs_inode *ip = XFS_I(inode);
721 int iolock = XFS_IOLOCK_EXCL;
722 size_t count = ocount;
724 xfs_rw_ilock(ip, iolock);
726 ret = xfs_file_aio_write_checks(file, &pos, &count, &iolock);
730 /* We can write back this queue in page reclaim */
731 current->backing_dev_info = mapping->backing_dev_info;
734 trace_xfs_file_buffered_write(ip, count, iocb->ki_pos, 0);
735 ret = generic_file_buffered_write(iocb, iovp, nr_segs,
736 pos, &iocb->ki_pos, count, 0);
739 * If we just got an ENOSPC, try to write back all dirty inodes to
740 * convert delalloc space to free up some of the excess reserved
743 if (ret == -ENOSPC && !enospc) {
745 xfs_flush_inodes(ip->i_mount);
749 current->backing_dev_info = NULL;
751 xfs_rw_iunlock(ip, iolock);
758 const struct iovec *iovp,
759 unsigned long nr_segs,
762 struct file *file = iocb->ki_filp;
763 struct address_space *mapping = file->f_mapping;
764 struct inode *inode = mapping->host;
765 struct xfs_inode *ip = XFS_I(inode);
769 XFS_STATS_INC(xs_write_calls);
771 BUG_ON(iocb->ki_pos != pos);
773 ret = generic_segment_checks(iovp, &nr_segs, &ocount, VERIFY_READ);
780 if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
785 if (unlikely(file->f_flags & O_DIRECT))
786 ret = xfs_file_dio_aio_write(iocb, iovp, nr_segs, pos, ocount);
788 ret = xfs_file_buffered_aio_write(iocb, iovp, nr_segs, pos,
794 XFS_STATS_ADD(xs_write_bytes, ret);
796 /* Handle various SYNC-type writes */
797 err = generic_write_sync(file, pos, ret);
813 struct inode *inode = file_inode(file);
817 xfs_inode_t *ip = XFS_I(inode);
818 int cmd = XFS_IOC_RESVSP;
819 int attr_flags = XFS_ATTR_NOLOCK;
821 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
828 xfs_ilock(ip, XFS_IOLOCK_EXCL);
830 if (mode & FALLOC_FL_PUNCH_HOLE)
831 cmd = XFS_IOC_UNRESVSP;
833 /* check the new inode size is valid before allocating */
834 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
835 offset + len > i_size_read(inode)) {
836 new_size = offset + len;
837 error = inode_newsize_ok(inode, new_size);
842 if (file->f_flags & O_DSYNC)
843 attr_flags |= XFS_ATTR_SYNC;
845 error = -xfs_change_file_space(ip, cmd, &bf, 0, attr_flags);
849 /* Change file size if needed */
853 iattr.ia_valid = ATTR_SIZE;
854 iattr.ia_size = new_size;
855 error = -xfs_setattr_size(ip, &iattr, XFS_ATTR_NOLOCK);
859 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
869 if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
871 if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb)))
881 struct xfs_inode *ip = XFS_I(inode);
885 error = xfs_file_open(inode, file);
890 * If there are any blocks, read-ahead block 0 as we're almost
891 * certain to have the next operation be a read there.
893 mode = xfs_ilock_map_shared(ip);
894 if (ip->i_d.di_nextents > 0)
895 xfs_dir3_data_readahead(NULL, ip, 0, -1);
896 xfs_iunlock(ip, mode);
905 return -xfs_release(XFS_I(inode));
911 struct dir_context *ctx)
913 struct inode *inode = file_inode(file);
914 xfs_inode_t *ip = XFS_I(inode);
919 * The Linux API doesn't pass down the total size of the buffer
920 * we read into down to the filesystem. With the filldir concept
921 * it's not needed for correct information, but the XFS dir2 leaf
922 * code wants an estimate of the buffer size to calculate it's
923 * readahead window and size the buffers used for mapping to
926 * Try to give it an estimate that's good enough, maybe at some
927 * point we can change the ->readdir prototype to include the
928 * buffer size. For now we use the current glibc buffer size.
930 bufsize = (size_t)min_t(loff_t, 32768, ip->i_d.di_size);
932 error = xfs_readdir(ip, ctx, bufsize);
941 struct vm_area_struct *vma)
943 vma->vm_ops = &xfs_file_vm_ops;
950 * mmap()d file has taken write protection fault and is being made
951 * writable. We can set the page state up correctly for a writable
952 * page, which means we can do correct delalloc accounting (ENOSPC
953 * checking!) and unwritten extent mapping.
957 struct vm_area_struct *vma,
958 struct vm_fault *vmf)
960 return block_page_mkwrite(vma, vmf, xfs_get_blocks);
964 * This type is designed to indicate the type of offset we would like
965 * to search from page cache for either xfs_seek_data() or xfs_seek_hole().
973 * Lookup the desired type of offset from the given page.
975 * On success, return true and the offset argument will point to the
976 * start of the region that was found. Otherwise this function will
977 * return false and keep the offset argument unchanged.
980 xfs_lookup_buffer_offset(
985 loff_t lastoff = page_offset(page);
987 struct buffer_head *bh, *head;
989 bh = head = page_buffers(page);
992 * Unwritten extents that have data in the page
993 * cache covering them can be identified by the
994 * BH_Unwritten state flag. Pages with multiple
995 * buffers might have a mix of holes, data and
996 * unwritten extents - any buffer with valid
997 * data in it should have BH_Uptodate flag set
1000 if (buffer_unwritten(bh) ||
1001 buffer_uptodate(bh)) {
1002 if (type == DATA_OFF)
1005 if (type == HOLE_OFF)
1013 lastoff += bh->b_size;
1014 } while ((bh = bh->b_this_page) != head);
1020 * This routine is called to find out and return a data or hole offset
1021 * from the page cache for unwritten extents according to the desired
1022 * type for xfs_seek_data() or xfs_seek_hole().
1024 * The argument offset is used to tell where we start to search from the
1025 * page cache. Map is used to figure out the end points of the range to
1028 * Return true if the desired type of offset was found, and the argument
1029 * offset is filled with that address. Otherwise, return false and keep
1033 xfs_find_get_desired_pgoff(
1034 struct inode *inode,
1035 struct xfs_bmbt_irec *map,
1039 struct xfs_inode *ip = XFS_I(inode);
1040 struct xfs_mount *mp = ip->i_mount;
1041 struct pagevec pvec;
1045 loff_t startoff = *offset;
1046 loff_t lastoff = startoff;
1049 pagevec_init(&pvec, 0);
1051 index = startoff >> PAGE_CACHE_SHIFT;
1052 endoff = XFS_FSB_TO_B(mp, map->br_startoff + map->br_blockcount);
1053 end = endoff >> PAGE_CACHE_SHIFT;
1059 want = min_t(pgoff_t, end - index, PAGEVEC_SIZE);
1060 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index,
1063 * No page mapped into given range. If we are searching holes
1064 * and if this is the first time we got into the loop, it means
1065 * that the given offset is landed in a hole, return it.
1067 * If we have already stepped through some block buffers to find
1068 * holes but they all contains data. In this case, the last
1069 * offset is already updated and pointed to the end of the last
1070 * mapped page, if it does not reach the endpoint to search,
1071 * that means there should be a hole between them.
1073 if (nr_pages == 0) {
1074 /* Data search found nothing */
1075 if (type == DATA_OFF)
1078 ASSERT(type == HOLE_OFF);
1079 if (lastoff == startoff || lastoff < endoff) {
1087 * At lease we found one page. If this is the first time we
1088 * step into the loop, and if the first page index offset is
1089 * greater than the given search offset, a hole was found.
1091 if (type == HOLE_OFF && lastoff == startoff &&
1092 lastoff < page_offset(pvec.pages[0])) {
1097 for (i = 0; i < nr_pages; i++) {
1098 struct page *page = pvec.pages[i];
1102 * At this point, the page may be truncated or
1103 * invalidated (changing page->mapping to NULL),
1104 * or even swizzled back from swapper_space to tmpfs
1105 * file mapping. However, page->index will not change
1106 * because we have a reference on the page.
1108 * Searching done if the page index is out of range.
1109 * If the current offset is not reaches the end of
1110 * the specified search range, there should be a hole
1113 if (page->index > end) {
1114 if (type == HOLE_OFF && lastoff < endoff) {
1123 * Page truncated or invalidated(page->mapping == NULL).
1124 * We can freely skip it and proceed to check the next
1127 if (unlikely(page->mapping != inode->i_mapping)) {
1132 if (!page_has_buffers(page)) {
1137 found = xfs_lookup_buffer_offset(page, &b_offset, type);
1140 * The found offset may be less than the start
1141 * point to search if this is the first time to
1144 *offset = max_t(loff_t, startoff, b_offset);
1150 * We either searching data but nothing was found, or
1151 * searching hole but found a data buffer. In either
1152 * case, probably the next page contains the desired
1153 * things, update the last offset to it so.
1155 lastoff = page_offset(page) + PAGE_SIZE;
1160 * The number of returned pages less than our desired, search
1161 * done. In this case, nothing was found for searching data,
1162 * but we found a hole behind the last offset.
1164 if (nr_pages < want) {
1165 if (type == HOLE_OFF) {
1172 index = pvec.pages[i - 1]->index + 1;
1173 pagevec_release(&pvec);
1174 } while (index <= end);
1177 pagevec_release(&pvec);
1186 struct inode *inode = file->f_mapping->host;
1187 struct xfs_inode *ip = XFS_I(inode);
1188 struct xfs_mount *mp = ip->i_mount;
1189 loff_t uninitialized_var(offset);
1191 xfs_fileoff_t fsbno;
1196 lock = xfs_ilock_map_shared(ip);
1198 isize = i_size_read(inode);
1199 if (start >= isize) {
1205 * Try to read extents from the first block indicated
1206 * by fsbno to the end block of the file.
1208 fsbno = XFS_B_TO_FSBT(mp, start);
1209 end = XFS_B_TO_FSB(mp, isize);
1211 struct xfs_bmbt_irec map[2];
1215 error = xfs_bmapi_read(ip, fsbno, end - fsbno, map, &nmap,
1220 /* No extents at given offset, must be beyond EOF */
1226 for (i = 0; i < nmap; i++) {
1227 offset = max_t(loff_t, start,
1228 XFS_FSB_TO_B(mp, map[i].br_startoff));
1230 /* Landed in a data extent */
1231 if (map[i].br_startblock == DELAYSTARTBLOCK ||
1232 (map[i].br_state == XFS_EXT_NORM &&
1233 !isnullstartblock(map[i].br_startblock)))
1237 * Landed in an unwritten extent, try to search data
1240 if (map[i].br_state == XFS_EXT_UNWRITTEN) {
1241 if (xfs_find_get_desired_pgoff(inode, &map[i],
1248 * map[0] is hole or its an unwritten extent but
1249 * without data in page cache. Probably means that
1250 * we are reading after EOF if nothing in map[1].
1260 * Nothing was found, proceed to the next round of search
1261 * if reading offset not beyond or hit EOF.
1263 fsbno = map[i - 1].br_startoff + map[i - 1].br_blockcount;
1264 start = XFS_FSB_TO_B(mp, fsbno);
1265 if (start >= isize) {
1272 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
1275 xfs_iunlock_map_shared(ip, lock);
1287 struct inode *inode = file->f_mapping->host;
1288 struct xfs_inode *ip = XFS_I(inode);
1289 struct xfs_mount *mp = ip->i_mount;
1290 loff_t uninitialized_var(offset);
1292 xfs_fileoff_t fsbno;
1297 if (XFS_FORCED_SHUTDOWN(mp))
1298 return -XFS_ERROR(EIO);
1300 lock = xfs_ilock_map_shared(ip);
1302 isize = i_size_read(inode);
1303 if (start >= isize) {
1308 fsbno = XFS_B_TO_FSBT(mp, start);
1309 end = XFS_B_TO_FSB(mp, isize);
1312 struct xfs_bmbt_irec map[2];
1316 error = xfs_bmapi_read(ip, fsbno, end - fsbno, map, &nmap,
1321 /* No extents at given offset, must be beyond EOF */
1327 for (i = 0; i < nmap; i++) {
1328 offset = max_t(loff_t, start,
1329 XFS_FSB_TO_B(mp, map[i].br_startoff));
1331 /* Landed in a hole */
1332 if (map[i].br_startblock == HOLESTARTBLOCK)
1336 * Landed in an unwritten extent, try to search hole
1339 if (map[i].br_state == XFS_EXT_UNWRITTEN) {
1340 if (xfs_find_get_desired_pgoff(inode, &map[i],
1347 * map[0] contains data or its unwritten but contains
1348 * data in page cache, probably means that we are
1349 * reading after EOF. We should fix offset to point
1350 * to the end of the file(i.e., there is an implicit
1351 * hole at the end of any file).
1361 * Both mappings contains data, proceed to the next round of
1362 * search if the current reading offset not beyond or hit EOF.
1364 fsbno = map[i - 1].br_startoff + map[i - 1].br_blockcount;
1365 start = XFS_FSB_TO_B(mp, fsbno);
1366 if (start >= isize) {
1374 * At this point, we must have found a hole. However, the returned
1375 * offset may be bigger than the file size as it may be aligned to
1376 * page boundary for unwritten extents, we need to deal with this
1377 * situation in particular.
1379 offset = min_t(loff_t, offset, isize);
1380 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
1383 xfs_iunlock_map_shared(ip, lock);
1400 return generic_file_llseek(file, offset, origin);
1402 return xfs_seek_data(file, offset);
1404 return xfs_seek_hole(file, offset);
1410 const struct file_operations xfs_file_operations = {
1411 .llseek = xfs_file_llseek,
1412 .read = do_sync_read,
1413 .write = do_sync_write,
1414 .aio_read = xfs_file_aio_read,
1415 .aio_write = xfs_file_aio_write,
1416 .splice_read = xfs_file_splice_read,
1417 .splice_write = xfs_file_splice_write,
1418 .unlocked_ioctl = xfs_file_ioctl,
1419 #ifdef CONFIG_COMPAT
1420 .compat_ioctl = xfs_file_compat_ioctl,
1422 .mmap = xfs_file_mmap,
1423 .open = xfs_file_open,
1424 .release = xfs_file_release,
1425 .fsync = xfs_file_fsync,
1426 .fallocate = xfs_file_fallocate,
1429 const struct file_operations xfs_dir_file_operations = {
1430 .open = xfs_dir_open,
1431 .read = generic_read_dir,
1432 .iterate = xfs_file_readdir,
1433 .llseek = generic_file_llseek,
1434 .unlocked_ioctl = xfs_file_ioctl,
1435 #ifdef CONFIG_COMPAT
1436 .compat_ioctl = xfs_file_compat_ioctl,
1438 .fsync = xfs_dir_fsync,
1441 static const struct vm_operations_struct xfs_file_vm_ops = {
1442 .fault = filemap_fault,
1443 .page_mkwrite = xfs_vm_page_mkwrite,
1444 .remap_pages = generic_file_remap_pages,