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_ioctl.h"
35 #include "xfs_trace.h"
37 #include <linux/dcache.h>
38 #include <linux/falloc.h>
39 #include <linux/pagevec.h>
41 static const struct vm_operations_struct xfs_file_vm_ops;
44 * Locking primitives for read and write IO paths to ensure we consistently use
45 * and order the inode->i_mutex, ip->i_lock and ip->i_iolock.
52 if (type & XFS_IOLOCK_EXCL)
53 mutex_lock(&VFS_I(ip)->i_mutex);
62 xfs_iunlock(ip, type);
63 if (type & XFS_IOLOCK_EXCL)
64 mutex_unlock(&VFS_I(ip)->i_mutex);
72 xfs_ilock_demote(ip, type);
73 if (type & XFS_IOLOCK_EXCL)
74 mutex_unlock(&VFS_I(ip)->i_mutex);
80 * xfs_iozero clears the specified range of buffer supplied,
81 * and marks all the affected blocks as valid and modified. If
82 * an affected block is not allocated, it will be allocated. If
83 * an affected block is not completely overwritten, and is not
84 * valid before the operation, it will be read from disk before
85 * being partially zeroed.
89 struct xfs_inode *ip, /* inode */
90 loff_t pos, /* offset in file */
91 size_t count) /* size of data to zero */
94 struct address_space *mapping;
97 mapping = VFS_I(ip)->i_mapping;
99 unsigned offset, bytes;
102 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
103 bytes = PAGE_CACHE_SIZE - offset;
107 status = pagecache_write_begin(NULL, mapping, pos, bytes,
108 AOP_FLAG_UNINTERRUPTIBLE,
113 zero_user(page, offset, bytes);
115 status = pagecache_write_end(NULL, mapping, pos, bytes, bytes,
117 WARN_ON(status <= 0); /* can't return less than zero! */
127 * Fsync operations on directories are much simpler than on regular files,
128 * as there is no file data to flush, and thus also no need for explicit
129 * cache flush operations, and there are no non-transaction metadata updates
130 * on directories either.
139 struct xfs_inode *ip = XFS_I(file->f_mapping->host);
140 struct xfs_mount *mp = ip->i_mount;
143 trace_xfs_dir_fsync(ip);
145 xfs_ilock(ip, XFS_ILOCK_SHARED);
146 if (xfs_ipincount(ip))
147 lsn = ip->i_itemp->ili_last_lsn;
148 xfs_iunlock(ip, XFS_ILOCK_SHARED);
152 return _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, NULL);
162 struct inode *inode = file->f_mapping->host;
163 struct xfs_inode *ip = XFS_I(inode);
164 struct xfs_mount *mp = ip->i_mount;
169 trace_xfs_file_fsync(ip);
171 error = filemap_write_and_wait_range(inode->i_mapping, start, end);
175 if (XFS_FORCED_SHUTDOWN(mp))
176 return -XFS_ERROR(EIO);
178 xfs_iflags_clear(ip, XFS_ITRUNCATED);
180 if (mp->m_flags & XFS_MOUNT_BARRIER) {
182 * If we have an RT and/or log subvolume we need to make sure
183 * to flush the write cache the device used for file data
184 * first. This is to ensure newly written file data make
185 * it to disk before logging the new inode size in case of
186 * an extending write.
188 if (XFS_IS_REALTIME_INODE(ip))
189 xfs_blkdev_issue_flush(mp->m_rtdev_targp);
190 else if (mp->m_logdev_targp != mp->m_ddev_targp)
191 xfs_blkdev_issue_flush(mp->m_ddev_targp);
195 * All metadata updates are logged, which means that we just have
196 * to flush the log up to the latest LSN that touched the inode.
198 xfs_ilock(ip, XFS_ILOCK_SHARED);
199 if (xfs_ipincount(ip)) {
201 (ip->i_itemp->ili_fields & ~XFS_ILOG_TIMESTAMP))
202 lsn = ip->i_itemp->ili_last_lsn;
204 xfs_iunlock(ip, XFS_ILOCK_SHARED);
207 error = _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, &log_flushed);
210 * If we only have a single device, and the log force about was
211 * a no-op we might have to flush the data device cache here.
212 * This can only happen for fdatasync/O_DSYNC if we were overwriting
213 * an already allocated file and thus do not have any metadata to
216 if ((mp->m_flags & XFS_MOUNT_BARRIER) &&
217 mp->m_logdev_targp == mp->m_ddev_targp &&
218 !XFS_IS_REALTIME_INODE(ip) &&
220 xfs_blkdev_issue_flush(mp->m_ddev_targp);
228 const struct iovec *iovp,
229 unsigned long nr_segs,
232 struct file *file = iocb->ki_filp;
233 struct inode *inode = file->f_mapping->host;
234 struct xfs_inode *ip = XFS_I(inode);
235 struct xfs_mount *mp = ip->i_mount;
241 XFS_STATS_INC(xs_read_calls);
243 BUG_ON(iocb->ki_pos != pos);
245 if (unlikely(file->f_flags & O_DIRECT))
246 ioflags |= IO_ISDIRECT;
247 if (file->f_mode & FMODE_NOCMTIME)
250 ret = generic_segment_checks(iovp, &nr_segs, &size, VERIFY_WRITE);
254 if (unlikely(ioflags & IO_ISDIRECT)) {
255 xfs_buftarg_t *target =
256 XFS_IS_REALTIME_INODE(ip) ?
257 mp->m_rtdev_targp : mp->m_ddev_targp;
258 if ((iocb->ki_pos & target->bt_smask) ||
259 (size & target->bt_smask)) {
260 if (iocb->ki_pos == i_size_read(inode))
262 return -XFS_ERROR(EINVAL);
266 n = mp->m_super->s_maxbytes - iocb->ki_pos;
267 if (n <= 0 || size == 0)
273 if (XFS_FORCED_SHUTDOWN(mp))
277 * Locking is a bit tricky here. If we take an exclusive lock
278 * for direct IO, we effectively serialise all new concurrent
279 * read IO to this file and block it behind IO that is currently in
280 * progress because IO in progress holds the IO lock shared. We only
281 * need to hold the lock exclusive to blow away the page cache, so
282 * only take lock exclusively if the page cache needs invalidation.
283 * This allows the normal direct IO case of no page cache pages to
284 * proceeed concurrently without serialisation.
286 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
287 if ((ioflags & IO_ISDIRECT) && inode->i_mapping->nrpages) {
288 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
289 xfs_rw_ilock(ip, XFS_IOLOCK_EXCL);
291 if (inode->i_mapping->nrpages) {
292 ret = -xfs_flushinval_pages(ip,
293 (iocb->ki_pos & PAGE_CACHE_MASK),
294 -1, FI_REMAPF_LOCKED);
296 xfs_rw_iunlock(ip, XFS_IOLOCK_EXCL);
300 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
303 trace_xfs_file_read(ip, size, iocb->ki_pos, ioflags);
305 ret = generic_file_aio_read(iocb, iovp, nr_segs, iocb->ki_pos);
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 * xfs_file_splice_write() does not use xfs_rw_ilock() because
347 * generic_file_splice_write() takes the i_mutex itself. This, in theory,
348 * couuld cause lock inversions between the aio_write path and the splice path
349 * if someone is doing concurrent splice(2) based writes and write(2) based
350 * writes to the same inode. The only real way to fix this is to re-implement
351 * the generic code here with correct locking orders.
354 xfs_file_splice_write(
355 struct pipe_inode_info *pipe,
356 struct file *outfilp,
361 struct inode *inode = outfilp->f_mapping->host;
362 struct xfs_inode *ip = XFS_I(inode);
366 XFS_STATS_INC(xs_write_calls);
368 if (outfilp->f_mode & FMODE_NOCMTIME)
371 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
374 xfs_ilock(ip, XFS_IOLOCK_EXCL);
376 trace_xfs_file_splice_write(ip, count, *ppos, ioflags);
378 ret = generic_file_splice_write(pipe, outfilp, ppos, count, flags);
380 XFS_STATS_ADD(xs_write_bytes, ret);
382 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
387 * This routine is called to handle zeroing any space in the last block of the
388 * file that is beyond the EOF. We do this since the size is being increased
389 * without writing anything to that block and we don't want to read the
390 * garbage on the disk.
392 STATIC int /* error (positive) */
394 struct xfs_inode *ip,
398 struct xfs_mount *mp = ip->i_mount;
399 xfs_fileoff_t last_fsb = XFS_B_TO_FSBT(mp, isize);
400 int zero_offset = XFS_B_FSB_OFFSET(mp, isize);
404 struct xfs_bmbt_irec imap;
406 xfs_ilock(ip, XFS_ILOCK_EXCL);
407 error = xfs_bmapi_read(ip, last_fsb, 1, &imap, &nimaps, 0);
408 xfs_iunlock(ip, XFS_ILOCK_EXCL);
415 * If the block underlying isize is just a hole, then there
416 * is nothing to zero.
418 if (imap.br_startblock == HOLESTARTBLOCK)
421 zero_len = mp->m_sb.sb_blocksize - zero_offset;
422 if (isize + zero_len > offset)
423 zero_len = offset - isize;
424 return xfs_iozero(ip, isize, zero_len);
428 * Zero any on disk space between the current EOF and the new, larger EOF.
430 * This handles the normal case of zeroing the remainder of the last block in
431 * the file and the unusual case of zeroing blocks out beyond the size of the
432 * file. This second case only happens with fixed size extents and when the
433 * system crashes before the inode size was updated but after blocks were
436 * Expects the iolock to be held exclusive, and will take the ilock internally.
438 int /* error (positive) */
440 struct xfs_inode *ip,
441 xfs_off_t offset, /* starting I/O offset */
442 xfs_fsize_t isize) /* current inode size */
444 struct xfs_mount *mp = ip->i_mount;
445 xfs_fileoff_t start_zero_fsb;
446 xfs_fileoff_t end_zero_fsb;
447 xfs_fileoff_t zero_count_fsb;
448 xfs_fileoff_t last_fsb;
449 xfs_fileoff_t zero_off;
450 xfs_fsize_t zero_len;
453 struct xfs_bmbt_irec imap;
455 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
456 ASSERT(offset > isize);
459 * First handle zeroing the block on which isize resides.
461 * We only zero a part of that block so it is handled specially.
463 if (XFS_B_FSB_OFFSET(mp, isize) != 0) {
464 error = xfs_zero_last_block(ip, offset, isize);
470 * Calculate the range between the new size and the old where blocks
471 * needing to be zeroed may exist.
473 * To get the block where the last byte in the file currently resides,
474 * we need to subtract one from the size and truncate back to a block
475 * boundary. We subtract 1 in case the size is exactly on a block
478 last_fsb = isize ? XFS_B_TO_FSBT(mp, isize - 1) : (xfs_fileoff_t)-1;
479 start_zero_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
480 end_zero_fsb = XFS_B_TO_FSBT(mp, offset - 1);
481 ASSERT((xfs_sfiloff_t)last_fsb < (xfs_sfiloff_t)start_zero_fsb);
482 if (last_fsb == end_zero_fsb) {
484 * The size was only incremented on its last block.
485 * We took care of that above, so just return.
490 ASSERT(start_zero_fsb <= end_zero_fsb);
491 while (start_zero_fsb <= end_zero_fsb) {
493 zero_count_fsb = end_zero_fsb - start_zero_fsb + 1;
495 xfs_ilock(ip, XFS_ILOCK_EXCL);
496 error = xfs_bmapi_read(ip, start_zero_fsb, zero_count_fsb,
498 xfs_iunlock(ip, XFS_ILOCK_EXCL);
504 if (imap.br_state == XFS_EXT_UNWRITTEN ||
505 imap.br_startblock == HOLESTARTBLOCK) {
506 start_zero_fsb = imap.br_startoff + imap.br_blockcount;
507 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
512 * There are blocks we need to zero.
514 zero_off = XFS_FSB_TO_B(mp, start_zero_fsb);
515 zero_len = XFS_FSB_TO_B(mp, imap.br_blockcount);
517 if ((zero_off + zero_len) > offset)
518 zero_len = offset - zero_off;
520 error = xfs_iozero(ip, zero_off, zero_len);
524 start_zero_fsb = imap.br_startoff + imap.br_blockcount;
525 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
532 * Common pre-write limit and setup checks.
534 * Called with the iolocked held either shared and exclusive according to
535 * @iolock, and returns with it held. Might upgrade the iolock to exclusive
536 * if called for a direct write beyond i_size.
539 xfs_file_aio_write_checks(
545 struct inode *inode = file->f_mapping->host;
546 struct xfs_inode *ip = XFS_I(inode);
550 error = generic_write_checks(file, pos, count, S_ISBLK(inode->i_mode));
555 * If the offset is beyond the size of the file, we need to zero any
556 * blocks that fall between the existing EOF and the start of this
557 * write. If zeroing is needed and we are currently holding the
558 * iolock shared, we need to update it to exclusive which implies
559 * having to redo all checks before.
561 if (*pos > i_size_read(inode)) {
562 if (*iolock == XFS_IOLOCK_SHARED) {
563 xfs_rw_iunlock(ip, *iolock);
564 *iolock = XFS_IOLOCK_EXCL;
565 xfs_rw_ilock(ip, *iolock);
568 error = -xfs_zero_eof(ip, *pos, i_size_read(inode));
574 * Updating the timestamps will grab the ilock again from
575 * xfs_fs_dirty_inode, so we have to call it after dropping the
576 * lock above. Eventually we should look into a way to avoid
577 * the pointless lock roundtrip.
579 if (likely(!(file->f_mode & FMODE_NOCMTIME))) {
580 error = file_update_time(file);
586 * If we're writing the file then make sure to clear the setuid and
587 * setgid bits if the process is not being run by root. This keeps
588 * people from modifying setuid and setgid binaries.
590 return file_remove_suid(file);
594 * xfs_file_dio_aio_write - handle direct IO writes
596 * Lock the inode appropriately to prepare for and issue a direct IO write.
597 * By separating it from the buffered write path we remove all the tricky to
598 * follow locking changes and looping.
600 * If there are cached pages or we're extending the file, we need IOLOCK_EXCL
601 * until we're sure the bytes at the new EOF have been zeroed and/or the cached
602 * pages are flushed out.
604 * In most cases the direct IO writes will be done holding IOLOCK_SHARED
605 * allowing them to be done in parallel with reads and other direct IO writes.
606 * However, if the IO is not aligned to filesystem blocks, the direct IO layer
607 * needs to do sub-block zeroing and that requires serialisation against other
608 * direct IOs to the same block. In this case we need to serialise the
609 * submission of the unaligned IOs so that we don't get racing block zeroing in
610 * the dio layer. To avoid the problem with aio, we also need to wait for
611 * outstanding IOs to complete so that unwritten extent conversion is completed
612 * before we try to map the overlapping block. This is currently implemented by
613 * hitting it with a big hammer (i.e. inode_dio_wait()).
615 * Returns with locks held indicated by @iolock and errors indicated by
616 * negative return values.
619 xfs_file_dio_aio_write(
621 const struct iovec *iovp,
622 unsigned long nr_segs,
626 struct file *file = iocb->ki_filp;
627 struct address_space *mapping = file->f_mapping;
628 struct inode *inode = mapping->host;
629 struct xfs_inode *ip = XFS_I(inode);
630 struct xfs_mount *mp = ip->i_mount;
632 size_t count = ocount;
633 int unaligned_io = 0;
635 struct xfs_buftarg *target = XFS_IS_REALTIME_INODE(ip) ?
636 mp->m_rtdev_targp : mp->m_ddev_targp;
638 if ((pos & target->bt_smask) || (count & target->bt_smask))
639 return -XFS_ERROR(EINVAL);
641 if ((pos & mp->m_blockmask) || ((pos + count) & mp->m_blockmask))
645 * We don't need to take an exclusive lock unless there page cache needs
646 * to be invalidated or unaligned IO is being executed. We don't need to
647 * consider the EOF extension case here because
648 * xfs_file_aio_write_checks() will relock the inode as necessary for
649 * EOF zeroing cases and fill out the new inode size as appropriate.
651 if (unaligned_io || mapping->nrpages)
652 iolock = XFS_IOLOCK_EXCL;
654 iolock = XFS_IOLOCK_SHARED;
655 xfs_rw_ilock(ip, iolock);
658 * Recheck if there are cached pages that need invalidate after we got
659 * the iolock to protect against other threads adding new pages while
660 * we were waiting for the iolock.
662 if (mapping->nrpages && iolock == XFS_IOLOCK_SHARED) {
663 xfs_rw_iunlock(ip, iolock);
664 iolock = XFS_IOLOCK_EXCL;
665 xfs_rw_ilock(ip, iolock);
668 ret = xfs_file_aio_write_checks(file, &pos, &count, &iolock);
672 if (mapping->nrpages) {
673 ret = -xfs_flushinval_pages(ip, (pos & PAGE_CACHE_MASK), -1,
680 * If we are doing unaligned IO, wait for all other IO to drain,
681 * otherwise demote the lock if we had to flush cached pages
684 inode_dio_wait(inode);
685 else if (iolock == XFS_IOLOCK_EXCL) {
686 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
687 iolock = XFS_IOLOCK_SHARED;
690 trace_xfs_file_direct_write(ip, count, iocb->ki_pos, 0);
691 ret = generic_file_direct_write(iocb, iovp,
692 &nr_segs, pos, &iocb->ki_pos, count, ocount);
695 xfs_rw_iunlock(ip, iolock);
697 /* No fallback to buffered IO on errors for XFS. */
698 ASSERT(ret < 0 || ret == count);
703 xfs_file_buffered_aio_write(
705 const struct iovec *iovp,
706 unsigned long nr_segs,
710 struct file *file = iocb->ki_filp;
711 struct address_space *mapping = file->f_mapping;
712 struct inode *inode = mapping->host;
713 struct xfs_inode *ip = XFS_I(inode);
716 int iolock = XFS_IOLOCK_EXCL;
717 size_t count = ocount;
719 xfs_rw_ilock(ip, iolock);
721 ret = xfs_file_aio_write_checks(file, &pos, &count, &iolock);
725 /* We can write back this queue in page reclaim */
726 current->backing_dev_info = mapping->backing_dev_info;
729 trace_xfs_file_buffered_write(ip, count, iocb->ki_pos, 0);
730 ret = generic_file_buffered_write(iocb, iovp, nr_segs,
731 pos, &iocb->ki_pos, count, ret);
733 * if we just got an ENOSPC, flush the inode now we aren't holding any
734 * page locks and retry *once*
736 if (ret == -ENOSPC && !enospc) {
738 ret = -xfs_flush_pages(ip, 0, -1, 0, FI_NONE);
743 current->backing_dev_info = NULL;
745 xfs_rw_iunlock(ip, iolock);
752 const struct iovec *iovp,
753 unsigned long nr_segs,
756 struct file *file = iocb->ki_filp;
757 struct address_space *mapping = file->f_mapping;
758 struct inode *inode = mapping->host;
759 struct xfs_inode *ip = XFS_I(inode);
763 XFS_STATS_INC(xs_write_calls);
765 BUG_ON(iocb->ki_pos != pos);
767 ret = generic_segment_checks(iovp, &nr_segs, &ocount, VERIFY_READ);
774 sb_start_write(inode->i_sb);
776 if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
781 if (unlikely(file->f_flags & O_DIRECT))
782 ret = xfs_file_dio_aio_write(iocb, iovp, nr_segs, pos, ocount);
784 ret = xfs_file_buffered_aio_write(iocb, iovp, nr_segs, pos,
790 XFS_STATS_ADD(xs_write_bytes, ret);
792 /* Handle various SYNC-type writes */
793 err = generic_write_sync(file, pos, ret);
799 sb_end_write(inode->i_sb);
810 struct inode *inode = file->f_path.dentry->d_inode;
814 xfs_inode_t *ip = XFS_I(inode);
815 int cmd = XFS_IOC_RESVSP;
816 int attr_flags = XFS_ATTR_NOLOCK;
818 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
825 xfs_ilock(ip, XFS_IOLOCK_EXCL);
827 if (mode & FALLOC_FL_PUNCH_HOLE)
828 cmd = XFS_IOC_UNRESVSP;
830 /* check the new inode size is valid before allocating */
831 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
832 offset + len > i_size_read(inode)) {
833 new_size = offset + len;
834 error = inode_newsize_ok(inode, new_size);
839 if (file->f_flags & O_DSYNC)
840 attr_flags |= XFS_ATTR_SYNC;
842 error = -xfs_change_file_space(ip, cmd, &bf, 0, attr_flags);
846 /* Change file size if needed */
850 iattr.ia_valid = ATTR_SIZE;
851 iattr.ia_size = new_size;
852 error = -xfs_setattr_size(ip, &iattr, XFS_ATTR_NOLOCK);
856 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
866 if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
868 if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb)))
878 struct xfs_inode *ip = XFS_I(inode);
882 error = xfs_file_open(inode, file);
887 * If there are any blocks, read-ahead block 0 as we're almost
888 * certain to have the next operation be a read there.
890 mode = xfs_ilock_map_shared(ip);
891 if (ip->i_d.di_nextents > 0)
892 xfs_da_reada_buf(NULL, ip, 0, XFS_DATA_FORK);
893 xfs_iunlock(ip, mode);
902 return -xfs_release(XFS_I(inode));
911 struct inode *inode = filp->f_path.dentry->d_inode;
912 xfs_inode_t *ip = XFS_I(inode);
917 * The Linux API doesn't pass down the total size of the buffer
918 * we read into down to the filesystem. With the filldir concept
919 * it's not needed for correct information, but the XFS dir2 leaf
920 * code wants an estimate of the buffer size to calculate it's
921 * readahead window and size the buffers used for mapping to
924 * Try to give it an estimate that's good enough, maybe at some
925 * point we can change the ->readdir prototype to include the
926 * buffer size. For now we use the current glibc buffer size.
928 bufsize = (size_t)min_t(loff_t, 32768, ip->i_d.di_size);
930 error = xfs_readdir(ip, dirent, bufsize,
931 (xfs_off_t *)&filp->f_pos, filldir);
940 struct vm_area_struct *vma)
942 vma->vm_ops = &xfs_file_vm_ops;
943 vma->vm_flags |= VM_CAN_NONLINEAR;
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 if (offset != file->f_pos)
1273 file->f_pos = offset;
1276 xfs_iunlock_map_shared(ip, lock);
1288 struct inode *inode = file->f_mapping->host;
1289 struct xfs_inode *ip = XFS_I(inode);
1290 struct xfs_mount *mp = ip->i_mount;
1291 loff_t uninitialized_var(offset);
1293 xfs_fileoff_t fsbno;
1298 if (XFS_FORCED_SHUTDOWN(mp))
1299 return -XFS_ERROR(EIO);
1301 lock = xfs_ilock_map_shared(ip);
1303 isize = i_size_read(inode);
1304 if (start >= isize) {
1309 fsbno = XFS_B_TO_FSBT(mp, start);
1310 end = XFS_B_TO_FSB(mp, isize);
1313 struct xfs_bmbt_irec map[2];
1317 error = xfs_bmapi_read(ip, fsbno, end - fsbno, map, &nmap,
1322 /* No extents at given offset, must be beyond EOF */
1328 for (i = 0; i < nmap; i++) {
1329 offset = max_t(loff_t, start,
1330 XFS_FSB_TO_B(mp, map[i].br_startoff));
1332 /* Landed in a hole */
1333 if (map[i].br_startblock == HOLESTARTBLOCK)
1337 * Landed in an unwritten extent, try to search hole
1340 if (map[i].br_state == XFS_EXT_UNWRITTEN) {
1341 if (xfs_find_get_desired_pgoff(inode, &map[i],
1348 * map[0] contains data or its unwritten but contains
1349 * data in page cache, probably means that we are
1350 * reading after EOF. We should fix offset to point
1351 * to the end of the file(i.e., there is an implicit
1352 * hole at the end of any file).
1362 * Both mappings contains data, proceed to the next round of
1363 * search if the current reading offset not beyond or hit EOF.
1365 fsbno = map[i - 1].br_startoff + map[i - 1].br_blockcount;
1366 start = XFS_FSB_TO_B(mp, fsbno);
1367 if (start >= isize) {
1375 * At this point, we must have found a hole. However, the returned
1376 * offset may be bigger than the file size as it may be aligned to
1377 * page boundary for unwritten extents, we need to deal with this
1378 * situation in particular.
1380 offset = min_t(loff_t, offset, isize);
1381 if (offset != file->f_pos)
1382 file->f_pos = offset;
1385 xfs_iunlock_map_shared(ip, lock);
1402 return generic_file_llseek(file, offset, origin);
1404 return xfs_seek_data(file, offset);
1406 return xfs_seek_hole(file, offset);
1412 const struct file_operations xfs_file_operations = {
1413 .llseek = xfs_file_llseek,
1414 .read = do_sync_read,
1415 .write = do_sync_write,
1416 .aio_read = xfs_file_aio_read,
1417 .aio_write = xfs_file_aio_write,
1418 .splice_read = xfs_file_splice_read,
1419 .splice_write = xfs_file_splice_write,
1420 .unlocked_ioctl = xfs_file_ioctl,
1421 #ifdef CONFIG_COMPAT
1422 .compat_ioctl = xfs_file_compat_ioctl,
1424 .mmap = xfs_file_mmap,
1425 .open = xfs_file_open,
1426 .release = xfs_file_release,
1427 .fsync = xfs_file_fsync,
1428 .fallocate = xfs_file_fallocate,
1431 const struct file_operations xfs_dir_file_operations = {
1432 .open = xfs_dir_open,
1433 .read = generic_read_dir,
1434 .readdir = xfs_file_readdir,
1435 .llseek = generic_file_llseek,
1436 .unlocked_ioctl = xfs_file_ioctl,
1437 #ifdef CONFIG_COMPAT
1438 .compat_ioctl = xfs_file_compat_ioctl,
1440 .fsync = xfs_dir_fsync,
1443 static const struct vm_operations_struct xfs_file_vm_ops = {
1444 .fault = filemap_fault,
1445 .page_mkwrite = xfs_vm_page_mkwrite,