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_error.h"
32 #include "xfs_vnodeops.h"
33 #include "xfs_da_btree.h"
34 #include "xfs_dir2_format.h"
36 #include "xfs_ioctl.h"
37 #include "xfs_trace.h"
39 #include <linux/aio.h>
40 #include <linux/dcache.h>
41 #include <linux/falloc.h>
42 #include <linux/pagevec.h>
44 static const struct vm_operations_struct xfs_file_vm_ops;
47 * Locking primitives for read and write IO paths to ensure we consistently use
48 * and order the inode->i_mutex, ip->i_lock and ip->i_iolock.
55 if (type & XFS_IOLOCK_EXCL)
56 mutex_lock(&VFS_I(ip)->i_mutex);
65 xfs_iunlock(ip, type);
66 if (type & XFS_IOLOCK_EXCL)
67 mutex_unlock(&VFS_I(ip)->i_mutex);
75 xfs_ilock_demote(ip, type);
76 if (type & XFS_IOLOCK_EXCL)
77 mutex_unlock(&VFS_I(ip)->i_mutex);
83 * xfs_iozero clears the specified range of buffer supplied,
84 * and marks all the affected blocks as valid and modified. If
85 * an affected block is not allocated, it will be allocated. If
86 * an affected block is not completely overwritten, and is not
87 * valid before the operation, it will be read from disk before
88 * being partially zeroed.
92 struct xfs_inode *ip, /* inode */
93 loff_t pos, /* offset in file */
94 size_t count) /* size of data to zero */
97 struct address_space *mapping;
100 mapping = VFS_I(ip)->i_mapping;
102 unsigned offset, bytes;
105 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
106 bytes = PAGE_CACHE_SIZE - offset;
110 status = pagecache_write_begin(NULL, mapping, pos, bytes,
111 AOP_FLAG_UNINTERRUPTIBLE,
116 zero_user(page, offset, bytes);
118 status = pagecache_write_end(NULL, mapping, pos, bytes, bytes,
120 WARN_ON(status <= 0); /* can't return less than zero! */
130 * Fsync operations on directories are much simpler than on regular files,
131 * as there is no file data to flush, and thus also no need for explicit
132 * cache flush operations, and there are no non-transaction metadata updates
133 * on directories either.
142 struct xfs_inode *ip = XFS_I(file->f_mapping->host);
143 struct xfs_mount *mp = ip->i_mount;
146 trace_xfs_dir_fsync(ip);
148 xfs_ilock(ip, XFS_ILOCK_SHARED);
149 if (xfs_ipincount(ip))
150 lsn = ip->i_itemp->ili_last_lsn;
151 xfs_iunlock(ip, XFS_ILOCK_SHARED);
155 return _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, NULL);
165 struct inode *inode = file->f_mapping->host;
166 struct xfs_inode *ip = XFS_I(inode);
167 struct xfs_mount *mp = ip->i_mount;
172 trace_xfs_file_fsync(ip);
174 error = filemap_write_and_wait_range(inode->i_mapping, start, end);
178 if (XFS_FORCED_SHUTDOWN(mp))
179 return -XFS_ERROR(EIO);
181 xfs_iflags_clear(ip, XFS_ITRUNCATED);
183 if (mp->m_flags & XFS_MOUNT_BARRIER) {
185 * If we have an RT and/or log subvolume we need to make sure
186 * to flush the write cache the device used for file data
187 * first. This is to ensure newly written file data make
188 * it to disk before logging the new inode size in case of
189 * an extending write.
191 if (XFS_IS_REALTIME_INODE(ip))
192 xfs_blkdev_issue_flush(mp->m_rtdev_targp);
193 else if (mp->m_logdev_targp != mp->m_ddev_targp)
194 xfs_blkdev_issue_flush(mp->m_ddev_targp);
198 * All metadata updates are logged, which means that we just have
199 * to flush the log up to the latest LSN that touched the inode.
201 xfs_ilock(ip, XFS_ILOCK_SHARED);
202 if (xfs_ipincount(ip)) {
204 (ip->i_itemp->ili_fields & ~XFS_ILOG_TIMESTAMP))
205 lsn = ip->i_itemp->ili_last_lsn;
207 xfs_iunlock(ip, XFS_ILOCK_SHARED);
210 error = _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, &log_flushed);
213 * If we only have a single device, and the log force about was
214 * a no-op we might have to flush the data device cache here.
215 * This can only happen for fdatasync/O_DSYNC if we were overwriting
216 * an already allocated file and thus do not have any metadata to
219 if ((mp->m_flags & XFS_MOUNT_BARRIER) &&
220 mp->m_logdev_targp == mp->m_ddev_targp &&
221 !XFS_IS_REALTIME_INODE(ip) &&
223 xfs_blkdev_issue_flush(mp->m_ddev_targp);
231 const struct iovec *iovp,
232 unsigned long nr_segs,
235 struct file *file = iocb->ki_filp;
236 struct inode *inode = file->f_mapping->host;
237 struct xfs_inode *ip = XFS_I(inode);
238 struct xfs_mount *mp = ip->i_mount;
244 XFS_STATS_INC(xs_read_calls);
246 BUG_ON(iocb->ki_pos != pos);
248 if (unlikely(file->f_flags & O_DIRECT))
249 ioflags |= IO_ISDIRECT;
250 if (file->f_mode & FMODE_NOCMTIME)
253 ret = generic_segment_checks(iovp, &nr_segs, &size, VERIFY_WRITE);
257 if (unlikely(ioflags & IO_ISDIRECT)) {
258 xfs_buftarg_t *target =
259 XFS_IS_REALTIME_INODE(ip) ?
260 mp->m_rtdev_targp : mp->m_ddev_targp;
261 if ((pos & target->bt_smask) || (size & target->bt_smask)) {
262 if (pos == i_size_read(inode))
264 return -XFS_ERROR(EINVAL);
268 n = mp->m_super->s_maxbytes - pos;
269 if (n <= 0 || size == 0)
275 if (XFS_FORCED_SHUTDOWN(mp))
279 * Locking is a bit tricky here. If we take an exclusive lock
280 * for direct IO, we effectively serialise all new concurrent
281 * read IO to this file and block it behind IO that is currently in
282 * progress because IO in progress holds the IO lock shared. We only
283 * need to hold the lock exclusive to blow away the page cache, so
284 * only take lock exclusively if the page cache needs invalidation.
285 * This allows the normal direct IO case of no page cache pages to
286 * proceeed concurrently without serialisation.
288 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
289 if ((ioflags & IO_ISDIRECT) && inode->i_mapping->nrpages) {
290 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
291 xfs_rw_ilock(ip, XFS_IOLOCK_EXCL);
293 if (inode->i_mapping->nrpages) {
294 ret = -filemap_write_and_wait_range(
295 VFS_I(ip)->i_mapping,
298 xfs_rw_iunlock(ip, XFS_IOLOCK_EXCL);
301 truncate_pagecache_range(VFS_I(ip), pos, -1);
303 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
306 trace_xfs_file_read(ip, size, pos, ioflags);
308 ret = generic_file_aio_read(iocb, iovp, nr_segs, pos);
310 XFS_STATS_ADD(xs_read_bytes, ret);
312 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
317 xfs_file_splice_read(
320 struct pipe_inode_info *pipe,
324 struct xfs_inode *ip = XFS_I(infilp->f_mapping->host);
328 XFS_STATS_INC(xs_read_calls);
330 if (infilp->f_mode & FMODE_NOCMTIME)
333 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
336 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
338 trace_xfs_file_splice_read(ip, count, *ppos, ioflags);
340 ret = generic_file_splice_read(infilp, ppos, pipe, count, flags);
342 XFS_STATS_ADD(xs_read_bytes, ret);
344 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
349 * xfs_file_splice_write() does not use xfs_rw_ilock() because
350 * generic_file_splice_write() takes the i_mutex itself. This, in theory,
351 * couuld cause lock inversions between the aio_write path and the splice path
352 * if someone is doing concurrent splice(2) based writes and write(2) based
353 * writes to the same inode. The only real way to fix this is to re-implement
354 * the generic code here with correct locking orders.
357 xfs_file_splice_write(
358 struct pipe_inode_info *pipe,
359 struct file *outfilp,
364 struct inode *inode = outfilp->f_mapping->host;
365 struct xfs_inode *ip = XFS_I(inode);
369 XFS_STATS_INC(xs_write_calls);
371 if (outfilp->f_mode & FMODE_NOCMTIME)
374 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
377 xfs_ilock(ip, XFS_IOLOCK_EXCL);
379 trace_xfs_file_splice_write(ip, count, *ppos, ioflags);
381 ret = generic_file_splice_write(pipe, outfilp, ppos, count, flags);
383 XFS_STATS_ADD(xs_write_bytes, ret);
385 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
390 * This routine is called to handle zeroing any space in the last block of the
391 * file that is beyond the EOF. We do this since the size is being increased
392 * without writing anything to that block and we don't want to read the
393 * garbage on the disk.
395 STATIC int /* error (positive) */
397 struct xfs_inode *ip,
401 struct xfs_mount *mp = ip->i_mount;
402 xfs_fileoff_t last_fsb = XFS_B_TO_FSBT(mp, isize);
403 int zero_offset = XFS_B_FSB_OFFSET(mp, isize);
407 struct xfs_bmbt_irec imap;
409 xfs_ilock(ip, XFS_ILOCK_EXCL);
410 error = xfs_bmapi_read(ip, last_fsb, 1, &imap, &nimaps, 0);
411 xfs_iunlock(ip, XFS_ILOCK_EXCL);
418 * If the block underlying isize is just a hole, then there
419 * is nothing to zero.
421 if (imap.br_startblock == HOLESTARTBLOCK)
424 zero_len = mp->m_sb.sb_blocksize - zero_offset;
425 if (isize + zero_len > offset)
426 zero_len = offset - isize;
427 return xfs_iozero(ip, isize, zero_len);
431 * Zero any on disk space between the current EOF and the new, larger EOF.
433 * This handles the normal case of zeroing the remainder of the last block in
434 * the file and the unusual case of zeroing blocks out beyond the size of the
435 * file. This second case only happens with fixed size extents and when the
436 * system crashes before the inode size was updated but after blocks were
439 * Expects the iolock to be held exclusive, and will take the ilock internally.
441 int /* error (positive) */
443 struct xfs_inode *ip,
444 xfs_off_t offset, /* starting I/O offset */
445 xfs_fsize_t isize) /* current inode size */
447 struct xfs_mount *mp = ip->i_mount;
448 xfs_fileoff_t start_zero_fsb;
449 xfs_fileoff_t end_zero_fsb;
450 xfs_fileoff_t zero_count_fsb;
451 xfs_fileoff_t last_fsb;
452 xfs_fileoff_t zero_off;
453 xfs_fsize_t zero_len;
456 struct xfs_bmbt_irec imap;
458 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
459 ASSERT(offset > isize);
462 * First handle zeroing the block on which isize resides.
464 * We only zero a part of that block so it is handled specially.
466 if (XFS_B_FSB_OFFSET(mp, isize) != 0) {
467 error = xfs_zero_last_block(ip, offset, isize);
473 * Calculate the range between the new size and the old where blocks
474 * needing to be zeroed may exist.
476 * To get the block where the last byte in the file currently resides,
477 * we need to subtract one from the size and truncate back to a block
478 * boundary. We subtract 1 in case the size is exactly on a block
481 last_fsb = isize ? XFS_B_TO_FSBT(mp, isize - 1) : (xfs_fileoff_t)-1;
482 start_zero_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
483 end_zero_fsb = XFS_B_TO_FSBT(mp, offset - 1);
484 ASSERT((xfs_sfiloff_t)last_fsb < (xfs_sfiloff_t)start_zero_fsb);
485 if (last_fsb == end_zero_fsb) {
487 * The size was only incremented on its last block.
488 * We took care of that above, so just return.
493 ASSERT(start_zero_fsb <= end_zero_fsb);
494 while (start_zero_fsb <= end_zero_fsb) {
496 zero_count_fsb = end_zero_fsb - start_zero_fsb + 1;
498 xfs_ilock(ip, XFS_ILOCK_EXCL);
499 error = xfs_bmapi_read(ip, start_zero_fsb, zero_count_fsb,
501 xfs_iunlock(ip, XFS_ILOCK_EXCL);
507 if (imap.br_state == XFS_EXT_UNWRITTEN ||
508 imap.br_startblock == HOLESTARTBLOCK) {
509 start_zero_fsb = imap.br_startoff + imap.br_blockcount;
510 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
515 * There are blocks we need to zero.
517 zero_off = XFS_FSB_TO_B(mp, start_zero_fsb);
518 zero_len = XFS_FSB_TO_B(mp, imap.br_blockcount);
520 if ((zero_off + zero_len) > offset)
521 zero_len = offset - zero_off;
523 error = xfs_iozero(ip, zero_off, zero_len);
527 start_zero_fsb = imap.br_startoff + imap.br_blockcount;
528 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
535 * Common pre-write limit and setup checks.
537 * Called with the iolocked held either shared and exclusive according to
538 * @iolock, and returns with it held. Might upgrade the iolock to exclusive
539 * if called for a direct write beyond i_size.
542 xfs_file_aio_write_checks(
548 struct inode *inode = file->f_mapping->host;
549 struct xfs_inode *ip = XFS_I(inode);
553 error = generic_write_checks(file, pos, count, S_ISBLK(inode->i_mode));
558 * If the offset is beyond the size of the file, we need to zero any
559 * blocks that fall between the existing EOF and the start of this
560 * write. If zeroing is needed and we are currently holding the
561 * iolock shared, we need to update it to exclusive which implies
562 * having to redo all checks before.
564 if (*pos > i_size_read(inode)) {
565 if (*iolock == XFS_IOLOCK_SHARED) {
566 xfs_rw_iunlock(ip, *iolock);
567 *iolock = XFS_IOLOCK_EXCL;
568 xfs_rw_ilock(ip, *iolock);
571 error = -xfs_zero_eof(ip, *pos, i_size_read(inode));
577 * Updating the timestamps will grab the ilock again from
578 * xfs_fs_dirty_inode, so we have to call it after dropping the
579 * lock above. Eventually we should look into a way to avoid
580 * the pointless lock roundtrip.
582 if (likely(!(file->f_mode & FMODE_NOCMTIME))) {
583 error = file_update_time(file);
589 * If we're writing the file then make sure to clear the setuid and
590 * setgid bits if the process is not being run by root. This keeps
591 * people from modifying setuid and setgid binaries.
593 return file_remove_suid(file);
597 * xfs_file_dio_aio_write - handle direct IO writes
599 * Lock the inode appropriately to prepare for and issue a direct IO write.
600 * By separating it from the buffered write path we remove all the tricky to
601 * follow locking changes and looping.
603 * If there are cached pages or we're extending the file, we need IOLOCK_EXCL
604 * until we're sure the bytes at the new EOF have been zeroed and/or the cached
605 * pages are flushed out.
607 * In most cases the direct IO writes will be done holding IOLOCK_SHARED
608 * allowing them to be done in parallel with reads and other direct IO writes.
609 * However, if the IO is not aligned to filesystem blocks, the direct IO layer
610 * needs to do sub-block zeroing and that requires serialisation against other
611 * direct IOs to the same block. In this case we need to serialise the
612 * submission of the unaligned IOs so that we don't get racing block zeroing in
613 * the dio layer. To avoid the problem with aio, we also need to wait for
614 * outstanding IOs to complete so that unwritten extent conversion is completed
615 * before we try to map the overlapping block. This is currently implemented by
616 * hitting it with a big hammer (i.e. inode_dio_wait()).
618 * Returns with locks held indicated by @iolock and errors indicated by
619 * negative return values.
622 xfs_file_dio_aio_write(
624 const struct iovec *iovp,
625 unsigned long nr_segs,
629 struct file *file = iocb->ki_filp;
630 struct address_space *mapping = file->f_mapping;
631 struct inode *inode = mapping->host;
632 struct xfs_inode *ip = XFS_I(inode);
633 struct xfs_mount *mp = ip->i_mount;
635 size_t count = ocount;
636 int unaligned_io = 0;
638 struct xfs_buftarg *target = XFS_IS_REALTIME_INODE(ip) ?
639 mp->m_rtdev_targp : mp->m_ddev_targp;
641 if ((pos & target->bt_smask) || (count & target->bt_smask))
642 return -XFS_ERROR(EINVAL);
644 if ((pos & mp->m_blockmask) || ((pos + count) & mp->m_blockmask))
648 * We don't need to take an exclusive lock unless there page cache needs
649 * to be invalidated or unaligned IO is being executed. We don't need to
650 * consider the EOF extension case here because
651 * xfs_file_aio_write_checks() will relock the inode as necessary for
652 * EOF zeroing cases and fill out the new inode size as appropriate.
654 if (unaligned_io || mapping->nrpages)
655 iolock = XFS_IOLOCK_EXCL;
657 iolock = XFS_IOLOCK_SHARED;
658 xfs_rw_ilock(ip, iolock);
661 * Recheck if there are cached pages that need invalidate after we got
662 * the iolock to protect against other threads adding new pages while
663 * we were waiting for the iolock.
665 if (mapping->nrpages && iolock == XFS_IOLOCK_SHARED) {
666 xfs_rw_iunlock(ip, iolock);
667 iolock = XFS_IOLOCK_EXCL;
668 xfs_rw_ilock(ip, iolock);
671 ret = xfs_file_aio_write_checks(file, &pos, &count, &iolock);
675 if (mapping->nrpages) {
676 ret = -filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
680 truncate_pagecache_range(VFS_I(ip), pos, -1);
684 * If we are doing unaligned IO, wait for all other IO to drain,
685 * otherwise demote the lock if we had to flush cached pages
688 inode_dio_wait(inode);
689 else if (iolock == XFS_IOLOCK_EXCL) {
690 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
691 iolock = XFS_IOLOCK_SHARED;
694 trace_xfs_file_direct_write(ip, count, iocb->ki_pos, 0);
695 ret = generic_file_direct_write(iocb, iovp,
696 &nr_segs, pos, &iocb->ki_pos, count, ocount);
699 xfs_rw_iunlock(ip, iolock);
701 /* No fallback to buffered IO on errors for XFS. */
702 ASSERT(ret < 0 || ret == count);
707 xfs_file_buffered_aio_write(
709 const struct iovec *iovp,
710 unsigned long nr_segs,
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);
720 int iolock = XFS_IOLOCK_EXCL;
721 size_t count = ocount;
723 xfs_rw_ilock(ip, iolock);
725 ret = xfs_file_aio_write_checks(file, &pos, &count, &iolock);
729 /* We can write back this queue in page reclaim */
730 current->backing_dev_info = mapping->backing_dev_info;
733 trace_xfs_file_buffered_write(ip, count, iocb->ki_pos, 0);
734 ret = generic_file_buffered_write(iocb, iovp, nr_segs,
735 pos, &iocb->ki_pos, count, 0);
738 * If we just got an ENOSPC, try to write back all dirty inodes to
739 * convert delalloc space to free up some of the excess reserved
742 if (ret == -ENOSPC && !enospc) {
744 xfs_flush_inodes(ip->i_mount);
748 current->backing_dev_info = NULL;
750 xfs_rw_iunlock(ip, iolock);
757 const struct iovec *iovp,
758 unsigned long nr_segs,
761 struct file *file = iocb->ki_filp;
762 struct address_space *mapping = file->f_mapping;
763 struct inode *inode = mapping->host;
764 struct xfs_inode *ip = XFS_I(inode);
768 XFS_STATS_INC(xs_write_calls);
770 BUG_ON(iocb->ki_pos != pos);
772 ret = generic_segment_checks(iovp, &nr_segs, &ocount, VERIFY_READ);
779 if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
784 if (unlikely(file->f_flags & O_DIRECT))
785 ret = xfs_file_dio_aio_write(iocb, iovp, nr_segs, pos, ocount);
787 ret = xfs_file_buffered_aio_write(iocb, iovp, nr_segs, pos,
793 XFS_STATS_ADD(xs_write_bytes, ret);
795 /* Handle various SYNC-type writes */
796 err = generic_write_sync(file, pos, ret);
812 struct inode *inode = file_inode(file);
816 xfs_inode_t *ip = XFS_I(inode);
817 int cmd = XFS_IOC_RESVSP;
818 int attr_flags = XFS_ATTR_NOLOCK;
820 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
827 xfs_ilock(ip, XFS_IOLOCK_EXCL);
829 if (mode & FALLOC_FL_PUNCH_HOLE)
830 cmd = XFS_IOC_UNRESVSP;
832 /* check the new inode size is valid before allocating */
833 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
834 offset + len > i_size_read(inode)) {
835 new_size = offset + len;
836 error = inode_newsize_ok(inode, new_size);
841 if (file->f_flags & O_DSYNC)
842 attr_flags |= XFS_ATTR_SYNC;
844 error = -xfs_change_file_space(ip, cmd, &bf, 0, attr_flags);
848 /* Change file size if needed */
852 iattr.ia_valid = ATTR_SIZE;
853 iattr.ia_size = new_size;
854 error = -xfs_setattr_size(ip, &iattr, XFS_ATTR_NOLOCK);
858 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
868 if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
870 if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb)))
880 struct xfs_inode *ip = XFS_I(inode);
884 error = xfs_file_open(inode, file);
889 * If there are any blocks, read-ahead block 0 as we're almost
890 * certain to have the next operation be a read there.
892 mode = xfs_ilock_map_shared(ip);
893 if (ip->i_d.di_nextents > 0)
894 xfs_dir3_data_readahead(NULL, ip, 0, -1);
895 xfs_iunlock(ip, mode);
904 return -xfs_release(XFS_I(inode));
910 struct dir_context *ctx)
912 struct inode *inode = file_inode(file);
913 xfs_inode_t *ip = XFS_I(inode);
918 * The Linux API doesn't pass down the total size of the buffer
919 * we read into down to the filesystem. With the filldir concept
920 * it's not needed for correct information, but the XFS dir2 leaf
921 * code wants an estimate of the buffer size to calculate it's
922 * readahead window and size the buffers used for mapping to
925 * Try to give it an estimate that's good enough, maybe at some
926 * point we can change the ->readdir prototype to include the
927 * buffer size. For now we use the current glibc buffer size.
929 bufsize = (size_t)min_t(loff_t, 32768, ip->i_d.di_size);
931 error = xfs_readdir(ip, ctx, bufsize);
940 struct vm_area_struct *vma)
942 vma->vm_ops = &xfs_file_vm_ops;
949 * mmap()d file has taken write protection fault and is being made
950 * writable. We can set the page state up correctly for a writable
951 * page, which means we can do correct delalloc accounting (ENOSPC
952 * checking!) and unwritten extent mapping.
956 struct vm_area_struct *vma,
957 struct vm_fault *vmf)
959 return block_page_mkwrite(vma, vmf, xfs_get_blocks);
963 * This type is designed to indicate the type of offset we would like
964 * to search from page cache for either xfs_seek_data() or xfs_seek_hole().
972 * Lookup the desired type of offset from the given page.
974 * On success, return true and the offset argument will point to the
975 * start of the region that was found. Otherwise this function will
976 * return false and keep the offset argument unchanged.
979 xfs_lookup_buffer_offset(
984 loff_t lastoff = page_offset(page);
986 struct buffer_head *bh, *head;
988 bh = head = page_buffers(page);
991 * Unwritten extents that have data in the page
992 * cache covering them can be identified by the
993 * BH_Unwritten state flag. Pages with multiple
994 * buffers might have a mix of holes, data and
995 * unwritten extents - any buffer with valid
996 * data in it should have BH_Uptodate flag set
999 if (buffer_unwritten(bh) ||
1000 buffer_uptodate(bh)) {
1001 if (type == DATA_OFF)
1004 if (type == HOLE_OFF)
1012 lastoff += bh->b_size;
1013 } while ((bh = bh->b_this_page) != head);
1019 * This routine is called to find out and return a data or hole offset
1020 * from the page cache for unwritten extents according to the desired
1021 * type for xfs_seek_data() or xfs_seek_hole().
1023 * The argument offset is used to tell where we start to search from the
1024 * page cache. Map is used to figure out the end points of the range to
1027 * Return true if the desired type of offset was found, and the argument
1028 * offset is filled with that address. Otherwise, return false and keep
1032 xfs_find_get_desired_pgoff(
1033 struct inode *inode,
1034 struct xfs_bmbt_irec *map,
1038 struct xfs_inode *ip = XFS_I(inode);
1039 struct xfs_mount *mp = ip->i_mount;
1040 struct pagevec pvec;
1044 loff_t startoff = *offset;
1045 loff_t lastoff = startoff;
1048 pagevec_init(&pvec, 0);
1050 index = startoff >> PAGE_CACHE_SHIFT;
1051 endoff = XFS_FSB_TO_B(mp, map->br_startoff + map->br_blockcount);
1052 end = endoff >> PAGE_CACHE_SHIFT;
1058 want = min_t(pgoff_t, end - index, PAGEVEC_SIZE);
1059 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index,
1062 * No page mapped into given range. If we are searching holes
1063 * and if this is the first time we got into the loop, it means
1064 * that the given offset is landed in a hole, return it.
1066 * If we have already stepped through some block buffers to find
1067 * holes but they all contains data. In this case, the last
1068 * offset is already updated and pointed to the end of the last
1069 * mapped page, if it does not reach the endpoint to search,
1070 * that means there should be a hole between them.
1072 if (nr_pages == 0) {
1073 /* Data search found nothing */
1074 if (type == DATA_OFF)
1077 ASSERT(type == HOLE_OFF);
1078 if (lastoff == startoff || lastoff < endoff) {
1086 * At lease we found one page. If this is the first time we
1087 * step into the loop, and if the first page index offset is
1088 * greater than the given search offset, a hole was found.
1090 if (type == HOLE_OFF && lastoff == startoff &&
1091 lastoff < page_offset(pvec.pages[0])) {
1096 for (i = 0; i < nr_pages; i++) {
1097 struct page *page = pvec.pages[i];
1101 * At this point, the page may be truncated or
1102 * invalidated (changing page->mapping to NULL),
1103 * or even swizzled back from swapper_space to tmpfs
1104 * file mapping. However, page->index will not change
1105 * because we have a reference on the page.
1107 * Searching done if the page index is out of range.
1108 * If the current offset is not reaches the end of
1109 * the specified search range, there should be a hole
1112 if (page->index > end) {
1113 if (type == HOLE_OFF && lastoff < endoff) {
1122 * Page truncated or invalidated(page->mapping == NULL).
1123 * We can freely skip it and proceed to check the next
1126 if (unlikely(page->mapping != inode->i_mapping)) {
1131 if (!page_has_buffers(page)) {
1136 found = xfs_lookup_buffer_offset(page, &b_offset, type);
1139 * The found offset may be less than the start
1140 * point to search if this is the first time to
1143 *offset = max_t(loff_t, startoff, b_offset);
1149 * We either searching data but nothing was found, or
1150 * searching hole but found a data buffer. In either
1151 * case, probably the next page contains the desired
1152 * things, update the last offset to it so.
1154 lastoff = page_offset(page) + PAGE_SIZE;
1159 * The number of returned pages less than our desired, search
1160 * done. In this case, nothing was found for searching data,
1161 * but we found a hole behind the last offset.
1163 if (nr_pages < want) {
1164 if (type == HOLE_OFF) {
1171 index = pvec.pages[i - 1]->index + 1;
1172 pagevec_release(&pvec);
1173 } while (index <= end);
1176 pagevec_release(&pvec);
1185 struct inode *inode = file->f_mapping->host;
1186 struct xfs_inode *ip = XFS_I(inode);
1187 struct xfs_mount *mp = ip->i_mount;
1188 loff_t uninitialized_var(offset);
1190 xfs_fileoff_t fsbno;
1195 lock = xfs_ilock_map_shared(ip);
1197 isize = i_size_read(inode);
1198 if (start >= isize) {
1204 * Try to read extents from the first block indicated
1205 * by fsbno to the end block of the file.
1207 fsbno = XFS_B_TO_FSBT(mp, start);
1208 end = XFS_B_TO_FSB(mp, isize);
1210 struct xfs_bmbt_irec map[2];
1214 error = xfs_bmapi_read(ip, fsbno, end - fsbno, map, &nmap,
1219 /* No extents at given offset, must be beyond EOF */
1225 for (i = 0; i < nmap; i++) {
1226 offset = max_t(loff_t, start,
1227 XFS_FSB_TO_B(mp, map[i].br_startoff));
1229 /* Landed in a data extent */
1230 if (map[i].br_startblock == DELAYSTARTBLOCK ||
1231 (map[i].br_state == XFS_EXT_NORM &&
1232 !isnullstartblock(map[i].br_startblock)))
1236 * Landed in an unwritten extent, try to search data
1239 if (map[i].br_state == XFS_EXT_UNWRITTEN) {
1240 if (xfs_find_get_desired_pgoff(inode, &map[i],
1247 * map[0] is hole or its an unwritten extent but
1248 * without data in page cache. Probably means that
1249 * we are reading after EOF if nothing in map[1].
1259 * Nothing was found, proceed to the next round of search
1260 * if reading offset not beyond or hit EOF.
1262 fsbno = map[i - 1].br_startoff + map[i - 1].br_blockcount;
1263 start = XFS_FSB_TO_B(mp, fsbno);
1264 if (start >= isize) {
1271 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
1274 xfs_iunlock_map_shared(ip, lock);
1286 struct inode *inode = file->f_mapping->host;
1287 struct xfs_inode *ip = XFS_I(inode);
1288 struct xfs_mount *mp = ip->i_mount;
1289 loff_t uninitialized_var(offset);
1291 xfs_fileoff_t fsbno;
1296 if (XFS_FORCED_SHUTDOWN(mp))
1297 return -XFS_ERROR(EIO);
1299 lock = xfs_ilock_map_shared(ip);
1301 isize = i_size_read(inode);
1302 if (start >= isize) {
1307 fsbno = XFS_B_TO_FSBT(mp, start);
1308 end = XFS_B_TO_FSB(mp, isize);
1311 struct xfs_bmbt_irec map[2];
1315 error = xfs_bmapi_read(ip, fsbno, end - fsbno, map, &nmap,
1320 /* No extents at given offset, must be beyond EOF */
1326 for (i = 0; i < nmap; i++) {
1327 offset = max_t(loff_t, start,
1328 XFS_FSB_TO_B(mp, map[i].br_startoff));
1330 /* Landed in a hole */
1331 if (map[i].br_startblock == HOLESTARTBLOCK)
1335 * Landed in an unwritten extent, try to search hole
1338 if (map[i].br_state == XFS_EXT_UNWRITTEN) {
1339 if (xfs_find_get_desired_pgoff(inode, &map[i],
1346 * map[0] contains data or its unwritten but contains
1347 * data in page cache, probably means that we are
1348 * reading after EOF. We should fix offset to point
1349 * to the end of the file(i.e., there is an implicit
1350 * hole at the end of any file).
1360 * Both mappings contains data, proceed to the next round of
1361 * search if the current reading offset not beyond or hit EOF.
1363 fsbno = map[i - 1].br_startoff + map[i - 1].br_blockcount;
1364 start = XFS_FSB_TO_B(mp, fsbno);
1365 if (start >= isize) {
1373 * At this point, we must have found a hole. However, the returned
1374 * offset may be bigger than the file size as it may be aligned to
1375 * page boundary for unwritten extents, we need to deal with this
1376 * situation in particular.
1378 offset = min_t(loff_t, offset, isize);
1379 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
1382 xfs_iunlock_map_shared(ip, lock);
1399 return generic_file_llseek(file, offset, origin);
1401 return xfs_seek_data(file, offset);
1403 return xfs_seek_hole(file, offset);
1409 const struct file_operations xfs_file_operations = {
1410 .llseek = xfs_file_llseek,
1411 .read = do_sync_read,
1412 .write = do_sync_write,
1413 .aio_read = xfs_file_aio_read,
1414 .aio_write = xfs_file_aio_write,
1415 .splice_read = xfs_file_splice_read,
1416 .splice_write = xfs_file_splice_write,
1417 .unlocked_ioctl = xfs_file_ioctl,
1418 #ifdef CONFIG_COMPAT
1419 .compat_ioctl = xfs_file_compat_ioctl,
1421 .mmap = xfs_file_mmap,
1422 .open = xfs_file_open,
1423 .release = xfs_file_release,
1424 .fsync = xfs_file_fsync,
1425 .fallocate = xfs_file_fallocate,
1428 const struct file_operations xfs_dir_file_operations = {
1429 .open = xfs_dir_open,
1430 .read = generic_read_dir,
1431 .iterate = xfs_file_readdir,
1432 .llseek = generic_file_llseek,
1433 .unlocked_ioctl = xfs_file_ioctl,
1434 #ifdef CONFIG_COMPAT
1435 .compat_ioctl = xfs_file_compat_ioctl,
1437 .fsync = xfs_dir_fsync,
1440 static const struct vm_operations_struct xfs_file_vm_ops = {
1441 .fault = filemap_fault,
1442 .page_mkwrite = xfs_vm_page_mkwrite,
1443 .remap_pages = generic_file_remap_pages,