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"
35 #include "xfs_dir2_priv.h"
36 #include "xfs_ioctl.h"
37 #include "xfs_trace.h"
39 #include <linux/dcache.h>
40 #include <linux/falloc.h>
41 #include <linux/pagevec.h>
43 static const struct vm_operations_struct xfs_file_vm_ops;
46 * Locking primitives for read and write IO paths to ensure we consistently use
47 * and order the inode->i_mutex, ip->i_lock and ip->i_iolock.
54 if (type & XFS_IOLOCK_EXCL)
55 mutex_lock(&VFS_I(ip)->i_mutex);
64 xfs_iunlock(ip, type);
65 if (type & XFS_IOLOCK_EXCL)
66 mutex_unlock(&VFS_I(ip)->i_mutex);
74 xfs_ilock_demote(ip, type);
75 if (type & XFS_IOLOCK_EXCL)
76 mutex_unlock(&VFS_I(ip)->i_mutex);
82 * xfs_iozero clears the specified range of buffer supplied,
83 * and marks all the affected blocks as valid and modified. If
84 * an affected block is not allocated, it will be allocated. If
85 * an affected block is not completely overwritten, and is not
86 * valid before the operation, it will be read from disk before
87 * being partially zeroed.
91 struct xfs_inode *ip, /* inode */
92 loff_t pos, /* offset in file */
93 size_t count) /* size of data to zero */
96 struct address_space *mapping;
99 mapping = VFS_I(ip)->i_mapping;
101 unsigned offset, bytes;
104 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
105 bytes = PAGE_CACHE_SIZE - offset;
109 status = pagecache_write_begin(NULL, mapping, pos, bytes,
110 AOP_FLAG_UNINTERRUPTIBLE,
115 zero_user(page, offset, bytes);
117 status = pagecache_write_end(NULL, mapping, pos, bytes, bytes,
119 WARN_ON(status <= 0); /* can't return less than zero! */
129 * Fsync operations on directories are much simpler than on regular files,
130 * as there is no file data to flush, and thus also no need for explicit
131 * cache flush operations, and there are no non-transaction metadata updates
132 * on directories either.
141 struct xfs_inode *ip = XFS_I(file->f_mapping->host);
142 struct xfs_mount *mp = ip->i_mount;
145 trace_xfs_dir_fsync(ip);
147 xfs_ilock(ip, XFS_ILOCK_SHARED);
148 if (xfs_ipincount(ip))
149 lsn = ip->i_itemp->ili_last_lsn;
150 xfs_iunlock(ip, XFS_ILOCK_SHARED);
154 return _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, NULL);
164 struct inode *inode = file->f_mapping->host;
165 struct xfs_inode *ip = XFS_I(inode);
166 struct xfs_mount *mp = ip->i_mount;
171 trace_xfs_file_fsync(ip);
173 error = filemap_write_and_wait_range(inode->i_mapping, start, end);
177 if (XFS_FORCED_SHUTDOWN(mp))
178 return -XFS_ERROR(EIO);
180 xfs_iflags_clear(ip, XFS_ITRUNCATED);
182 if (mp->m_flags & XFS_MOUNT_BARRIER) {
184 * If we have an RT and/or log subvolume we need to make sure
185 * to flush the write cache the device used for file data
186 * first. This is to ensure newly written file data make
187 * it to disk before logging the new inode size in case of
188 * an extending write.
190 if (XFS_IS_REALTIME_INODE(ip))
191 xfs_blkdev_issue_flush(mp->m_rtdev_targp);
192 else if (mp->m_logdev_targp != mp->m_ddev_targp)
193 xfs_blkdev_issue_flush(mp->m_ddev_targp);
197 * All metadata updates are logged, which means that we just have
198 * to flush the log up to the latest LSN that touched the inode.
200 xfs_ilock(ip, XFS_ILOCK_SHARED);
201 if (xfs_ipincount(ip)) {
203 (ip->i_itemp->ili_fields & ~XFS_ILOG_TIMESTAMP))
204 lsn = ip->i_itemp->ili_last_lsn;
206 xfs_iunlock(ip, XFS_ILOCK_SHARED);
209 error = _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, &log_flushed);
212 * If we only have a single device, and the log force about was
213 * a no-op we might have to flush the data device cache here.
214 * This can only happen for fdatasync/O_DSYNC if we were overwriting
215 * an already allocated file and thus do not have any metadata to
218 if ((mp->m_flags & XFS_MOUNT_BARRIER) &&
219 mp->m_logdev_targp == mp->m_ddev_targp &&
220 !XFS_IS_REALTIME_INODE(ip) &&
222 xfs_blkdev_issue_flush(mp->m_ddev_targp);
230 const struct iovec *iovp,
231 unsigned long nr_segs,
234 struct file *file = iocb->ki_filp;
235 struct inode *inode = file->f_mapping->host;
236 struct xfs_inode *ip = XFS_I(inode);
237 struct xfs_mount *mp = ip->i_mount;
243 XFS_STATS_INC(xs_read_calls);
245 BUG_ON(iocb->ki_pos != pos);
247 if (unlikely(file->f_flags & O_DIRECT))
248 ioflags |= IO_ISDIRECT;
249 if (file->f_mode & FMODE_NOCMTIME)
252 ret = generic_segment_checks(iovp, &nr_segs, &size, VERIFY_WRITE);
256 if (unlikely(ioflags & IO_ISDIRECT)) {
257 xfs_buftarg_t *target =
258 XFS_IS_REALTIME_INODE(ip) ?
259 mp->m_rtdev_targp : mp->m_ddev_targp;
260 if ((pos & target->bt_smask) || (size & target->bt_smask)) {
261 if (pos == i_size_read(inode))
263 return -XFS_ERROR(EINVAL);
267 n = mp->m_super->s_maxbytes - pos;
268 if (n <= 0 || size == 0)
274 if (XFS_FORCED_SHUTDOWN(mp))
278 * Locking is a bit tricky here. If we take an exclusive lock
279 * for direct IO, we effectively serialise all new concurrent
280 * read IO to this file and block it behind IO that is currently in
281 * progress because IO in progress holds the IO lock shared. We only
282 * need to hold the lock exclusive to blow away the page cache, so
283 * only take lock exclusively if the page cache needs invalidation.
284 * This allows the normal direct IO case of no page cache pages to
285 * proceeed concurrently without serialisation.
287 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
288 if ((ioflags & IO_ISDIRECT) && inode->i_mapping->nrpages) {
289 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
290 xfs_rw_ilock(ip, XFS_IOLOCK_EXCL);
292 if (inode->i_mapping->nrpages) {
293 ret = -filemap_write_and_wait_range(
294 VFS_I(ip)->i_mapping,
297 xfs_rw_iunlock(ip, XFS_IOLOCK_EXCL);
300 truncate_pagecache_range(VFS_I(ip), pos, -1);
302 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
305 trace_xfs_file_read(ip, size, pos, ioflags);
307 ret = generic_file_aio_read(iocb, iovp, nr_segs, pos);
309 XFS_STATS_ADD(xs_read_bytes, ret);
311 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
316 xfs_file_splice_read(
319 struct pipe_inode_info *pipe,
323 struct xfs_inode *ip = XFS_I(infilp->f_mapping->host);
327 XFS_STATS_INC(xs_read_calls);
329 if (infilp->f_mode & FMODE_NOCMTIME)
332 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
335 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
337 trace_xfs_file_splice_read(ip, count, *ppos, ioflags);
339 ret = generic_file_splice_read(infilp, ppos, pipe, count, flags);
341 XFS_STATS_ADD(xs_read_bytes, ret);
343 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
348 * xfs_file_splice_write() does not use xfs_rw_ilock() because
349 * generic_file_splice_write() takes the i_mutex itself. This, in theory,
350 * couuld cause lock inversions between the aio_write path and the splice path
351 * if someone is doing concurrent splice(2) based writes and write(2) based
352 * writes to the same inode. The only real way to fix this is to re-implement
353 * the generic code here with correct locking orders.
356 xfs_file_splice_write(
357 struct pipe_inode_info *pipe,
358 struct file *outfilp,
363 struct inode *inode = outfilp->f_mapping->host;
364 struct xfs_inode *ip = XFS_I(inode);
368 XFS_STATS_INC(xs_write_calls);
370 if (outfilp->f_mode & FMODE_NOCMTIME)
373 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
376 xfs_ilock(ip, XFS_IOLOCK_EXCL);
378 trace_xfs_file_splice_write(ip, count, *ppos, ioflags);
380 ret = generic_file_splice_write(pipe, outfilp, ppos, count, flags);
382 XFS_STATS_ADD(xs_write_bytes, ret);
384 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
389 * This routine is called to handle zeroing any space in the last block of the
390 * file that is beyond the EOF. We do this since the size is being increased
391 * without writing anything to that block and we don't want to read the
392 * garbage on the disk.
394 STATIC int /* error (positive) */
396 struct xfs_inode *ip,
400 struct xfs_mount *mp = ip->i_mount;
401 xfs_fileoff_t last_fsb = XFS_B_TO_FSBT(mp, isize);
402 int zero_offset = XFS_B_FSB_OFFSET(mp, isize);
406 struct xfs_bmbt_irec imap;
408 xfs_ilock(ip, XFS_ILOCK_EXCL);
409 error = xfs_bmapi_read(ip, last_fsb, 1, &imap, &nimaps, 0);
410 xfs_iunlock(ip, XFS_ILOCK_EXCL);
417 * If the block underlying isize is just a hole, then there
418 * is nothing to zero.
420 if (imap.br_startblock == HOLESTARTBLOCK)
423 zero_len = mp->m_sb.sb_blocksize - zero_offset;
424 if (isize + zero_len > offset)
425 zero_len = offset - isize;
426 return xfs_iozero(ip, isize, zero_len);
430 * Zero any on disk space between the current EOF and the new, larger EOF.
432 * This handles the normal case of zeroing the remainder of the last block in
433 * the file and the unusual case of zeroing blocks out beyond the size of the
434 * file. This second case only happens with fixed size extents and when the
435 * system crashes before the inode size was updated but after blocks were
438 * Expects the iolock to be held exclusive, and will take the ilock internally.
440 int /* error (positive) */
442 struct xfs_inode *ip,
443 xfs_off_t offset, /* starting I/O offset */
444 xfs_fsize_t isize) /* current inode size */
446 struct xfs_mount *mp = ip->i_mount;
447 xfs_fileoff_t start_zero_fsb;
448 xfs_fileoff_t end_zero_fsb;
449 xfs_fileoff_t zero_count_fsb;
450 xfs_fileoff_t last_fsb;
451 xfs_fileoff_t zero_off;
452 xfs_fsize_t zero_len;
455 struct xfs_bmbt_irec imap;
457 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
458 ASSERT(offset > isize);
461 * First handle zeroing the block on which isize resides.
463 * We only zero a part of that block so it is handled specially.
465 if (XFS_B_FSB_OFFSET(mp, isize) != 0) {
466 error = xfs_zero_last_block(ip, offset, isize);
472 * Calculate the range between the new size and the old where blocks
473 * needing to be zeroed may exist.
475 * To get the block where the last byte in the file currently resides,
476 * we need to subtract one from the size and truncate back to a block
477 * boundary. We subtract 1 in case the size is exactly on a block
480 last_fsb = isize ? XFS_B_TO_FSBT(mp, isize - 1) : (xfs_fileoff_t)-1;
481 start_zero_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
482 end_zero_fsb = XFS_B_TO_FSBT(mp, offset - 1);
483 ASSERT((xfs_sfiloff_t)last_fsb < (xfs_sfiloff_t)start_zero_fsb);
484 if (last_fsb == end_zero_fsb) {
486 * The size was only incremented on its last block.
487 * We took care of that above, so just return.
492 ASSERT(start_zero_fsb <= end_zero_fsb);
493 while (start_zero_fsb <= end_zero_fsb) {
495 zero_count_fsb = end_zero_fsb - start_zero_fsb + 1;
497 xfs_ilock(ip, XFS_ILOCK_EXCL);
498 error = xfs_bmapi_read(ip, start_zero_fsb, zero_count_fsb,
500 xfs_iunlock(ip, XFS_ILOCK_EXCL);
506 if (imap.br_state == XFS_EXT_UNWRITTEN ||
507 imap.br_startblock == HOLESTARTBLOCK) {
508 start_zero_fsb = imap.br_startoff + imap.br_blockcount;
509 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
514 * There are blocks we need to zero.
516 zero_off = XFS_FSB_TO_B(mp, start_zero_fsb);
517 zero_len = XFS_FSB_TO_B(mp, imap.br_blockcount);
519 if ((zero_off + zero_len) > offset)
520 zero_len = offset - zero_off;
522 error = xfs_iozero(ip, zero_off, zero_len);
526 start_zero_fsb = imap.br_startoff + imap.br_blockcount;
527 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
534 * Common pre-write limit and setup checks.
536 * Called with the iolocked held either shared and exclusive according to
537 * @iolock, and returns with it held. Might upgrade the iolock to exclusive
538 * if called for a direct write beyond i_size.
541 xfs_file_aio_write_checks(
547 struct inode *inode = file->f_mapping->host;
548 struct xfs_inode *ip = XFS_I(inode);
552 error = generic_write_checks(file, pos, count, S_ISBLK(inode->i_mode));
557 * If the offset is beyond the size of the file, we need to zero any
558 * blocks that fall between the existing EOF and the start of this
559 * write. If zeroing is needed and we are currently holding the
560 * iolock shared, we need to update it to exclusive which implies
561 * having to redo all checks before.
563 if (*pos > i_size_read(inode)) {
564 if (*iolock == XFS_IOLOCK_SHARED) {
565 xfs_rw_iunlock(ip, *iolock);
566 *iolock = XFS_IOLOCK_EXCL;
567 xfs_rw_ilock(ip, *iolock);
570 error = -xfs_zero_eof(ip, *pos, i_size_read(inode));
576 * Updating the timestamps will grab the ilock again from
577 * xfs_fs_dirty_inode, so we have to call it after dropping the
578 * lock above. Eventually we should look into a way to avoid
579 * the pointless lock roundtrip.
581 if (likely(!(file->f_mode & FMODE_NOCMTIME))) {
582 error = file_update_time(file);
588 * If we're writing the file then make sure to clear the setuid and
589 * setgid bits if the process is not being run by root. This keeps
590 * people from modifying setuid and setgid binaries.
592 return file_remove_suid(file);
596 * xfs_file_dio_aio_write - handle direct IO writes
598 * Lock the inode appropriately to prepare for and issue a direct IO write.
599 * By separating it from the buffered write path we remove all the tricky to
600 * follow locking changes and looping.
602 * If there are cached pages or we're extending the file, we need IOLOCK_EXCL
603 * until we're sure the bytes at the new EOF have been zeroed and/or the cached
604 * pages are flushed out.
606 * In most cases the direct IO writes will be done holding IOLOCK_SHARED
607 * allowing them to be done in parallel with reads and other direct IO writes.
608 * However, if the IO is not aligned to filesystem blocks, the direct IO layer
609 * needs to do sub-block zeroing and that requires serialisation against other
610 * direct IOs to the same block. In this case we need to serialise the
611 * submission of the unaligned IOs so that we don't get racing block zeroing in
612 * the dio layer. To avoid the problem with aio, we also need to wait for
613 * outstanding IOs to complete so that unwritten extent conversion is completed
614 * before we try to map the overlapping block. This is currently implemented by
615 * hitting it with a big hammer (i.e. inode_dio_wait()).
617 * Returns with locks held indicated by @iolock and errors indicated by
618 * negative return values.
621 xfs_file_dio_aio_write(
623 const struct iovec *iovp,
624 unsigned long nr_segs,
628 struct file *file = iocb->ki_filp;
629 struct address_space *mapping = file->f_mapping;
630 struct inode *inode = mapping->host;
631 struct xfs_inode *ip = XFS_I(inode);
632 struct xfs_mount *mp = ip->i_mount;
634 size_t count = ocount;
635 int unaligned_io = 0;
637 struct xfs_buftarg *target = XFS_IS_REALTIME_INODE(ip) ?
638 mp->m_rtdev_targp : mp->m_ddev_targp;
640 if ((pos & target->bt_smask) || (count & target->bt_smask))
641 return -XFS_ERROR(EINVAL);
643 if ((pos & mp->m_blockmask) || ((pos + count) & mp->m_blockmask))
647 * We don't need to take an exclusive lock unless there page cache needs
648 * to be invalidated or unaligned IO is being executed. We don't need to
649 * consider the EOF extension case here because
650 * xfs_file_aio_write_checks() will relock the inode as necessary for
651 * EOF zeroing cases and fill out the new inode size as appropriate.
653 if (unaligned_io || mapping->nrpages)
654 iolock = XFS_IOLOCK_EXCL;
656 iolock = XFS_IOLOCK_SHARED;
657 xfs_rw_ilock(ip, iolock);
660 * Recheck if there are cached pages that need invalidate after we got
661 * the iolock to protect against other threads adding new pages while
662 * we were waiting for the iolock.
664 if (mapping->nrpages && iolock == XFS_IOLOCK_SHARED) {
665 xfs_rw_iunlock(ip, iolock);
666 iolock = XFS_IOLOCK_EXCL;
667 xfs_rw_ilock(ip, iolock);
670 ret = xfs_file_aio_write_checks(file, &pos, &count, &iolock);
674 if (mapping->nrpages) {
675 ret = -filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
679 truncate_pagecache_range(VFS_I(ip), pos, -1);
683 * If we are doing unaligned IO, wait for all other IO to drain,
684 * otherwise demote the lock if we had to flush cached pages
687 inode_dio_wait(inode);
688 else if (iolock == XFS_IOLOCK_EXCL) {
689 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
690 iolock = XFS_IOLOCK_SHARED;
693 trace_xfs_file_direct_write(ip, count, iocb->ki_pos, 0);
694 ret = generic_file_direct_write(iocb, iovp,
695 &nr_segs, pos, &iocb->ki_pos, count, ocount);
698 xfs_rw_iunlock(ip, iolock);
700 /* No fallback to buffered IO on errors for XFS. */
701 ASSERT(ret < 0 || ret == count);
706 xfs_file_buffered_aio_write(
708 const struct iovec *iovp,
709 unsigned long nr_segs,
713 struct file *file = iocb->ki_filp;
714 struct address_space *mapping = file->f_mapping;
715 struct inode *inode = mapping->host;
716 struct xfs_inode *ip = XFS_I(inode);
719 int iolock = XFS_IOLOCK_EXCL;
720 size_t count = ocount;
722 xfs_rw_ilock(ip, iolock);
724 ret = xfs_file_aio_write_checks(file, &pos, &count, &iolock);
728 /* We can write back this queue in page reclaim */
729 current->backing_dev_info = mapping->backing_dev_info;
732 trace_xfs_file_buffered_write(ip, count, iocb->ki_pos, 0);
733 ret = generic_file_buffered_write(iocb, iovp, nr_segs,
734 pos, &iocb->ki_pos, count, 0);
737 * If we just got an ENOSPC, try to write back all dirty inodes to
738 * convert delalloc space to free up some of the excess reserved
741 if (ret == -ENOSPC && !enospc) {
743 xfs_flush_inodes(ip->i_mount);
747 current->backing_dev_info = NULL;
749 xfs_rw_iunlock(ip, iolock);
756 const struct iovec *iovp,
757 unsigned long nr_segs,
760 struct file *file = iocb->ki_filp;
761 struct address_space *mapping = file->f_mapping;
762 struct inode *inode = mapping->host;
763 struct xfs_inode *ip = XFS_I(inode);
767 XFS_STATS_INC(xs_write_calls);
769 BUG_ON(iocb->ki_pos != pos);
771 ret = generic_segment_checks(iovp, &nr_segs, &ocount, VERIFY_READ);
778 sb_start_write(inode->i_sb);
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);
803 sb_end_write(inode->i_sb);
814 struct inode *inode = file->f_path.dentry->d_inode;
818 xfs_inode_t *ip = XFS_I(inode);
819 int cmd = XFS_IOC_RESVSP;
820 int attr_flags = XFS_ATTR_NOLOCK;
822 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
829 xfs_ilock(ip, XFS_IOLOCK_EXCL);
831 if (mode & FALLOC_FL_PUNCH_HOLE)
832 cmd = XFS_IOC_UNRESVSP;
834 /* check the new inode size is valid before allocating */
835 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
836 offset + len > i_size_read(inode)) {
837 new_size = offset + len;
838 error = inode_newsize_ok(inode, new_size);
843 if (file->f_flags & O_DSYNC)
844 attr_flags |= XFS_ATTR_SYNC;
846 error = -xfs_change_file_space(ip, cmd, &bf, 0, attr_flags);
850 /* Change file size if needed */
854 iattr.ia_valid = ATTR_SIZE;
855 iattr.ia_size = new_size;
856 error = -xfs_setattr_size(ip, &iattr, XFS_ATTR_NOLOCK);
860 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
870 if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
872 if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb)))
882 struct xfs_inode *ip = XFS_I(inode);
886 error = xfs_file_open(inode, file);
891 * If there are any blocks, read-ahead block 0 as we're almost
892 * certain to have the next operation be a read there.
894 mode = xfs_ilock_map_shared(ip);
895 if (ip->i_d.di_nextents > 0)
896 xfs_dir2_data_readahead(NULL, ip, 0, -1);
897 xfs_iunlock(ip, mode);
906 return -xfs_release(XFS_I(inode));
915 struct inode *inode = filp->f_path.dentry->d_inode;
916 xfs_inode_t *ip = XFS_I(inode);
921 * The Linux API doesn't pass down the total size of the buffer
922 * we read into down to the filesystem. With the filldir concept
923 * it's not needed for correct information, but the XFS dir2 leaf
924 * code wants an estimate of the buffer size to calculate it's
925 * readahead window and size the buffers used for mapping to
928 * Try to give it an estimate that's good enough, maybe at some
929 * point we can change the ->readdir prototype to include the
930 * buffer size. For now we use the current glibc buffer size.
932 bufsize = (size_t)min_t(loff_t, 32768, ip->i_d.di_size);
934 error = xfs_readdir(ip, dirent, bufsize,
935 (xfs_off_t *)&filp->f_pos, filldir);
944 struct vm_area_struct *vma)
946 vma->vm_ops = &xfs_file_vm_ops;
953 * mmap()d file has taken write protection fault and is being made
954 * writable. We can set the page state up correctly for a writable
955 * page, which means we can do correct delalloc accounting (ENOSPC
956 * checking!) and unwritten extent mapping.
960 struct vm_area_struct *vma,
961 struct vm_fault *vmf)
963 return block_page_mkwrite(vma, vmf, xfs_get_blocks);
967 * This type is designed to indicate the type of offset we would like
968 * to search from page cache for either xfs_seek_data() or xfs_seek_hole().
976 * Lookup the desired type of offset from the given page.
978 * On success, return true and the offset argument will point to the
979 * start of the region that was found. Otherwise this function will
980 * return false and keep the offset argument unchanged.
983 xfs_lookup_buffer_offset(
988 loff_t lastoff = page_offset(page);
990 struct buffer_head *bh, *head;
992 bh = head = page_buffers(page);
995 * Unwritten extents that have data in the page
996 * cache covering them can be identified by the
997 * BH_Unwritten state flag. Pages with multiple
998 * buffers might have a mix of holes, data and
999 * unwritten extents - any buffer with valid
1000 * data in it should have BH_Uptodate flag set
1003 if (buffer_unwritten(bh) ||
1004 buffer_uptodate(bh)) {
1005 if (type == DATA_OFF)
1008 if (type == HOLE_OFF)
1016 lastoff += bh->b_size;
1017 } while ((bh = bh->b_this_page) != head);
1023 * This routine is called to find out and return a data or hole offset
1024 * from the page cache for unwritten extents according to the desired
1025 * type for xfs_seek_data() or xfs_seek_hole().
1027 * The argument offset is used to tell where we start to search from the
1028 * page cache. Map is used to figure out the end points of the range to
1031 * Return true if the desired type of offset was found, and the argument
1032 * offset is filled with that address. Otherwise, return false and keep
1036 xfs_find_get_desired_pgoff(
1037 struct inode *inode,
1038 struct xfs_bmbt_irec *map,
1042 struct xfs_inode *ip = XFS_I(inode);
1043 struct xfs_mount *mp = ip->i_mount;
1044 struct pagevec pvec;
1048 loff_t startoff = *offset;
1049 loff_t lastoff = startoff;
1052 pagevec_init(&pvec, 0);
1054 index = startoff >> PAGE_CACHE_SHIFT;
1055 endoff = XFS_FSB_TO_B(mp, map->br_startoff + map->br_blockcount);
1056 end = endoff >> PAGE_CACHE_SHIFT;
1062 want = min_t(pgoff_t, end - index, PAGEVEC_SIZE);
1063 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index,
1066 * No page mapped into given range. If we are searching holes
1067 * and if this is the first time we got into the loop, it means
1068 * that the given offset is landed in a hole, return it.
1070 * If we have already stepped through some block buffers to find
1071 * holes but they all contains data. In this case, the last
1072 * offset is already updated and pointed to the end of the last
1073 * mapped page, if it does not reach the endpoint to search,
1074 * that means there should be a hole between them.
1076 if (nr_pages == 0) {
1077 /* Data search found nothing */
1078 if (type == DATA_OFF)
1081 ASSERT(type == HOLE_OFF);
1082 if (lastoff == startoff || lastoff < endoff) {
1090 * At lease we found one page. If this is the first time we
1091 * step into the loop, and if the first page index offset is
1092 * greater than the given search offset, a hole was found.
1094 if (type == HOLE_OFF && lastoff == startoff &&
1095 lastoff < page_offset(pvec.pages[0])) {
1100 for (i = 0; i < nr_pages; i++) {
1101 struct page *page = pvec.pages[i];
1105 * At this point, the page may be truncated or
1106 * invalidated (changing page->mapping to NULL),
1107 * or even swizzled back from swapper_space to tmpfs
1108 * file mapping. However, page->index will not change
1109 * because we have a reference on the page.
1111 * Searching done if the page index is out of range.
1112 * If the current offset is not reaches the end of
1113 * the specified search range, there should be a hole
1116 if (page->index > end) {
1117 if (type == HOLE_OFF && lastoff < endoff) {
1126 * Page truncated or invalidated(page->mapping == NULL).
1127 * We can freely skip it and proceed to check the next
1130 if (unlikely(page->mapping != inode->i_mapping)) {
1135 if (!page_has_buffers(page)) {
1140 found = xfs_lookup_buffer_offset(page, &b_offset, type);
1143 * The found offset may be less than the start
1144 * point to search if this is the first time to
1147 *offset = max_t(loff_t, startoff, b_offset);
1153 * We either searching data but nothing was found, or
1154 * searching hole but found a data buffer. In either
1155 * case, probably the next page contains the desired
1156 * things, update the last offset to it so.
1158 lastoff = page_offset(page) + PAGE_SIZE;
1163 * The number of returned pages less than our desired, search
1164 * done. In this case, nothing was found for searching data,
1165 * but we found a hole behind the last offset.
1167 if (nr_pages < want) {
1168 if (type == HOLE_OFF) {
1175 index = pvec.pages[i - 1]->index + 1;
1176 pagevec_release(&pvec);
1177 } while (index <= end);
1180 pagevec_release(&pvec);
1189 struct inode *inode = file->f_mapping->host;
1190 struct xfs_inode *ip = XFS_I(inode);
1191 struct xfs_mount *mp = ip->i_mount;
1192 loff_t uninitialized_var(offset);
1194 xfs_fileoff_t fsbno;
1199 lock = xfs_ilock_map_shared(ip);
1201 isize = i_size_read(inode);
1202 if (start >= isize) {
1208 * Try to read extents from the first block indicated
1209 * by fsbno to the end block of the file.
1211 fsbno = XFS_B_TO_FSBT(mp, start);
1212 end = XFS_B_TO_FSB(mp, isize);
1214 struct xfs_bmbt_irec map[2];
1218 error = xfs_bmapi_read(ip, fsbno, end - fsbno, map, &nmap,
1223 /* No extents at given offset, must be beyond EOF */
1229 for (i = 0; i < nmap; i++) {
1230 offset = max_t(loff_t, start,
1231 XFS_FSB_TO_B(mp, map[i].br_startoff));
1233 /* Landed in a data extent */
1234 if (map[i].br_startblock == DELAYSTARTBLOCK ||
1235 (map[i].br_state == XFS_EXT_NORM &&
1236 !isnullstartblock(map[i].br_startblock)))
1240 * Landed in an unwritten extent, try to search data
1243 if (map[i].br_state == XFS_EXT_UNWRITTEN) {
1244 if (xfs_find_get_desired_pgoff(inode, &map[i],
1251 * map[0] is hole or its an unwritten extent but
1252 * without data in page cache. Probably means that
1253 * we are reading after EOF if nothing in map[1].
1263 * Nothing was found, proceed to the next round of search
1264 * if reading offset not beyond or hit EOF.
1266 fsbno = map[i - 1].br_startoff + map[i - 1].br_blockcount;
1267 start = XFS_FSB_TO_B(mp, fsbno);
1268 if (start >= isize) {
1275 if (offset != file->f_pos)
1276 file->f_pos = offset;
1279 xfs_iunlock_map_shared(ip, lock);
1291 struct inode *inode = file->f_mapping->host;
1292 struct xfs_inode *ip = XFS_I(inode);
1293 struct xfs_mount *mp = ip->i_mount;
1294 loff_t uninitialized_var(offset);
1296 xfs_fileoff_t fsbno;
1301 if (XFS_FORCED_SHUTDOWN(mp))
1302 return -XFS_ERROR(EIO);
1304 lock = xfs_ilock_map_shared(ip);
1306 isize = i_size_read(inode);
1307 if (start >= isize) {
1312 fsbno = XFS_B_TO_FSBT(mp, start);
1313 end = XFS_B_TO_FSB(mp, isize);
1316 struct xfs_bmbt_irec map[2];
1320 error = xfs_bmapi_read(ip, fsbno, end - fsbno, map, &nmap,
1325 /* No extents at given offset, must be beyond EOF */
1331 for (i = 0; i < nmap; i++) {
1332 offset = max_t(loff_t, start,
1333 XFS_FSB_TO_B(mp, map[i].br_startoff));
1335 /* Landed in a hole */
1336 if (map[i].br_startblock == HOLESTARTBLOCK)
1340 * Landed in an unwritten extent, try to search hole
1343 if (map[i].br_state == XFS_EXT_UNWRITTEN) {
1344 if (xfs_find_get_desired_pgoff(inode, &map[i],
1351 * map[0] contains data or its unwritten but contains
1352 * data in page cache, probably means that we are
1353 * reading after EOF. We should fix offset to point
1354 * to the end of the file(i.e., there is an implicit
1355 * hole at the end of any file).
1365 * Both mappings contains data, proceed to the next round of
1366 * search if the current reading offset not beyond or hit EOF.
1368 fsbno = map[i - 1].br_startoff + map[i - 1].br_blockcount;
1369 start = XFS_FSB_TO_B(mp, fsbno);
1370 if (start >= isize) {
1378 * At this point, we must have found a hole. However, the returned
1379 * offset may be bigger than the file size as it may be aligned to
1380 * page boundary for unwritten extents, we need to deal with this
1381 * situation in particular.
1383 offset = min_t(loff_t, offset, isize);
1384 if (offset != file->f_pos)
1385 file->f_pos = offset;
1388 xfs_iunlock_map_shared(ip, lock);
1405 return generic_file_llseek(file, offset, origin);
1407 return xfs_seek_data(file, offset);
1409 return xfs_seek_hole(file, offset);
1415 const struct file_operations xfs_file_operations = {
1416 .llseek = xfs_file_llseek,
1417 .read = do_sync_read,
1418 .write = do_sync_write,
1419 .aio_read = xfs_file_aio_read,
1420 .aio_write = xfs_file_aio_write,
1421 .splice_read = xfs_file_splice_read,
1422 .splice_write = xfs_file_splice_write,
1423 .unlocked_ioctl = xfs_file_ioctl,
1424 #ifdef CONFIG_COMPAT
1425 .compat_ioctl = xfs_file_compat_ioctl,
1427 .mmap = xfs_file_mmap,
1428 .open = xfs_file_open,
1429 .release = xfs_file_release,
1430 .fsync = xfs_file_fsync,
1431 .fallocate = xfs_file_fallocate,
1434 const struct file_operations xfs_dir_file_operations = {
1435 .open = xfs_dir_open,
1436 .read = generic_read_dir,
1437 .readdir = xfs_file_readdir,
1438 .llseek = generic_file_llseek,
1439 .unlocked_ioctl = xfs_file_ioctl,
1440 #ifdef CONFIG_COMPAT
1441 .compat_ioctl = xfs_file_compat_ioctl,
1443 .fsync = xfs_dir_fsync,
1446 static const struct vm_operations_struct xfs_file_vm_ops = {
1447 .fault = filemap_fault,
1448 .page_mkwrite = xfs_vm_page_mkwrite,
1449 .remap_pages = generic_file_remap_pages,