2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
23 #include "xfs_trans.h"
24 #include "xfs_mount.h"
25 #include "xfs_bmap_btree.h"
26 #include "xfs_alloc.h"
27 #include "xfs_dinode.h"
28 #include "xfs_inode.h"
29 #include "xfs_inode_item.h"
31 #include "xfs_bmap_util.h"
32 #include "xfs_error.h"
33 #include "xfs_da_btree.h"
34 #include "xfs_dir2_format.h"
36 #include "xfs_dir2_priv.h"
37 #include "xfs_ioctl.h"
38 #include "xfs_trace.h"
40 #include <linux/aio.h>
41 #include <linux/dcache.h>
42 #include <linux/falloc.h>
43 #include <linux/pagevec.h>
45 static const struct vm_operations_struct xfs_file_vm_ops;
48 * Locking primitives for read and write IO paths to ensure we consistently use
49 * and order the inode->i_mutex, ip->i_lock and ip->i_iolock.
56 if (type & XFS_IOLOCK_EXCL)
57 mutex_lock(&VFS_I(ip)->i_mutex);
66 xfs_iunlock(ip, type);
67 if (type & XFS_IOLOCK_EXCL)
68 mutex_unlock(&VFS_I(ip)->i_mutex);
76 xfs_ilock_demote(ip, type);
77 if (type & XFS_IOLOCK_EXCL)
78 mutex_unlock(&VFS_I(ip)->i_mutex);
84 * xfs_iozero clears the specified range of buffer supplied,
85 * and marks all the affected blocks as valid and modified. If
86 * an affected block is not allocated, it will be allocated. If
87 * an affected block is not completely overwritten, and is not
88 * valid before the operation, it will be read from disk before
89 * being partially zeroed.
93 struct xfs_inode *ip, /* inode */
94 loff_t pos, /* offset in file */
95 size_t count) /* size of data to zero */
98 struct address_space *mapping;
101 mapping = VFS_I(ip)->i_mapping;
103 unsigned offset, bytes;
106 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
107 bytes = PAGE_CACHE_SIZE - offset;
111 status = pagecache_write_begin(NULL, mapping, pos, bytes,
112 AOP_FLAG_UNINTERRUPTIBLE,
117 zero_user(page, offset, bytes);
119 status = pagecache_write_end(NULL, mapping, pos, bytes, bytes,
121 WARN_ON(status <= 0); /* can't return less than zero! */
131 * Fsync operations on directories are much simpler than on regular files,
132 * as there is no file data to flush, and thus also no need for explicit
133 * cache flush operations, and there are no non-transaction metadata updates
134 * on directories either.
143 struct xfs_inode *ip = XFS_I(file->f_mapping->host);
144 struct xfs_mount *mp = ip->i_mount;
147 trace_xfs_dir_fsync(ip);
149 xfs_ilock(ip, XFS_ILOCK_SHARED);
150 if (xfs_ipincount(ip))
151 lsn = ip->i_itemp->ili_last_lsn;
152 xfs_iunlock(ip, XFS_ILOCK_SHARED);
156 return _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, NULL);
166 struct inode *inode = file->f_mapping->host;
167 struct xfs_inode *ip = XFS_I(inode);
168 struct xfs_mount *mp = ip->i_mount;
173 trace_xfs_file_fsync(ip);
175 error = filemap_write_and_wait_range(inode->i_mapping, start, end);
179 if (XFS_FORCED_SHUTDOWN(mp))
180 return -XFS_ERROR(EIO);
182 xfs_iflags_clear(ip, XFS_ITRUNCATED);
184 if (mp->m_flags & XFS_MOUNT_BARRIER) {
186 * If we have an RT and/or log subvolume we need to make sure
187 * to flush the write cache the device used for file data
188 * first. This is to ensure newly written file data make
189 * it to disk before logging the new inode size in case of
190 * an extending write.
192 if (XFS_IS_REALTIME_INODE(ip))
193 xfs_blkdev_issue_flush(mp->m_rtdev_targp);
194 else if (mp->m_logdev_targp != mp->m_ddev_targp)
195 xfs_blkdev_issue_flush(mp->m_ddev_targp);
199 * All metadata updates are logged, which means that we just have
200 * to flush the log up to the latest LSN that touched the inode.
202 xfs_ilock(ip, XFS_ILOCK_SHARED);
203 if (xfs_ipincount(ip)) {
205 (ip->i_itemp->ili_fields & ~XFS_ILOG_TIMESTAMP))
206 lsn = ip->i_itemp->ili_last_lsn;
208 xfs_iunlock(ip, XFS_ILOCK_SHARED);
211 error = _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, &log_flushed);
214 * If we only have a single device, and the log force about was
215 * a no-op we might have to flush the data device cache here.
216 * This can only happen for fdatasync/O_DSYNC if we were overwriting
217 * an already allocated file and thus do not have any metadata to
220 if ((mp->m_flags & XFS_MOUNT_BARRIER) &&
221 mp->m_logdev_targp == mp->m_ddev_targp &&
222 !XFS_IS_REALTIME_INODE(ip) &&
224 xfs_blkdev_issue_flush(mp->m_ddev_targp);
232 struct iov_iter *iter,
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 size = iov_iter_count(iter);
255 if (unlikely(ioflags & IO_ISDIRECT)) {
256 xfs_buftarg_t *target =
257 XFS_IS_REALTIME_INODE(ip) ?
258 mp->m_rtdev_targp : mp->m_ddev_targp;
259 if ((pos & target->bt_smask) || (size & target->bt_smask)) {
260 if (pos == i_size_read(inode))
262 return -XFS_ERROR(EINVAL);
266 n = mp->m_super->s_maxbytes - 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 = -filemap_write_and_wait_range(
293 VFS_I(ip)->i_mapping,
296 xfs_rw_iunlock(ip, XFS_IOLOCK_EXCL);
299 truncate_pagecache_range(VFS_I(ip), pos, -1);
301 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
304 trace_xfs_file_read(ip, size, pos, ioflags);
306 ret = generic_file_read_iter(iocb, iter, pos);
308 XFS_STATS_ADD(xs_read_bytes, ret);
310 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
315 xfs_file_splice_read(
318 struct pipe_inode_info *pipe,
322 struct xfs_inode *ip = XFS_I(infilp->f_mapping->host);
326 XFS_STATS_INC(xs_read_calls);
328 if (infilp->f_mode & FMODE_NOCMTIME)
331 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
334 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
336 trace_xfs_file_splice_read(ip, count, *ppos, ioflags);
338 ret = generic_file_splice_read(infilp, ppos, pipe, count, flags);
340 XFS_STATS_ADD(xs_read_bytes, ret);
342 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
347 * xfs_file_splice_write() does not use xfs_rw_ilock() because
348 * generic_file_splice_write() takes the i_mutex itself. This, in theory,
349 * couuld cause lock inversions between the aio_write path and the splice path
350 * if someone is doing concurrent splice(2) based writes and write(2) based
351 * writes to the same inode. The only real way to fix this is to re-implement
352 * the generic code here with correct locking orders.
355 xfs_file_splice_write(
356 struct pipe_inode_info *pipe,
357 struct file *outfilp,
362 struct inode *inode = outfilp->f_mapping->host;
363 struct xfs_inode *ip = XFS_I(inode);
367 XFS_STATS_INC(xs_write_calls);
369 if (outfilp->f_mode & FMODE_NOCMTIME)
372 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
375 xfs_ilock(ip, XFS_IOLOCK_EXCL);
377 trace_xfs_file_splice_write(ip, count, *ppos, ioflags);
379 ret = generic_file_splice_write(pipe, outfilp, ppos, count, flags);
381 XFS_STATS_ADD(xs_write_bytes, ret);
383 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
388 * This routine is called to handle zeroing any space in the last block of the
389 * file that is beyond the EOF. We do this since the size is being increased
390 * without writing anything to that block and we don't want to read the
391 * garbage on the disk.
393 STATIC int /* error (positive) */
395 struct xfs_inode *ip,
399 struct xfs_mount *mp = ip->i_mount;
400 xfs_fileoff_t last_fsb = XFS_B_TO_FSBT(mp, isize);
401 int zero_offset = XFS_B_FSB_OFFSET(mp, isize);
405 struct xfs_bmbt_irec imap;
407 xfs_ilock(ip, XFS_ILOCK_EXCL);
408 error = xfs_bmapi_read(ip, last_fsb, 1, &imap, &nimaps, 0);
409 xfs_iunlock(ip, XFS_ILOCK_EXCL);
416 * If the block underlying isize is just a hole, then there
417 * is nothing to zero.
419 if (imap.br_startblock == HOLESTARTBLOCK)
422 zero_len = mp->m_sb.sb_blocksize - zero_offset;
423 if (isize + zero_len > offset)
424 zero_len = offset - isize;
425 return xfs_iozero(ip, isize, zero_len);
429 * Zero any on disk space between the current EOF and the new, larger EOF.
431 * This handles the normal case of zeroing the remainder of the last block in
432 * the file and the unusual case of zeroing blocks out beyond the size of the
433 * file. This second case only happens with fixed size extents and when the
434 * system crashes before the inode size was updated but after blocks were
437 * Expects the iolock to be held exclusive, and will take the ilock internally.
439 int /* error (positive) */
441 struct xfs_inode *ip,
442 xfs_off_t offset, /* starting I/O offset */
443 xfs_fsize_t isize) /* current inode size */
445 struct xfs_mount *mp = ip->i_mount;
446 xfs_fileoff_t start_zero_fsb;
447 xfs_fileoff_t end_zero_fsb;
448 xfs_fileoff_t zero_count_fsb;
449 xfs_fileoff_t last_fsb;
450 xfs_fileoff_t zero_off;
451 xfs_fsize_t zero_len;
454 struct xfs_bmbt_irec imap;
456 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
457 ASSERT(offset > isize);
460 * First handle zeroing the block on which isize resides.
462 * We only zero a part of that block so it is handled specially.
464 if (XFS_B_FSB_OFFSET(mp, isize) != 0) {
465 error = xfs_zero_last_block(ip, offset, isize);
471 * Calculate the range between the new size and the old where blocks
472 * needing to be zeroed may exist.
474 * To get the block where the last byte in the file currently resides,
475 * we need to subtract one from the size and truncate back to a block
476 * boundary. We subtract 1 in case the size is exactly on a block
479 last_fsb = isize ? XFS_B_TO_FSBT(mp, isize - 1) : (xfs_fileoff_t)-1;
480 start_zero_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
481 end_zero_fsb = XFS_B_TO_FSBT(mp, offset - 1);
482 ASSERT((xfs_sfiloff_t)last_fsb < (xfs_sfiloff_t)start_zero_fsb);
483 if (last_fsb == end_zero_fsb) {
485 * The size was only incremented on its last block.
486 * We took care of that above, so just return.
491 ASSERT(start_zero_fsb <= end_zero_fsb);
492 while (start_zero_fsb <= end_zero_fsb) {
494 zero_count_fsb = end_zero_fsb - start_zero_fsb + 1;
496 xfs_ilock(ip, XFS_ILOCK_EXCL);
497 error = xfs_bmapi_read(ip, start_zero_fsb, zero_count_fsb,
499 xfs_iunlock(ip, XFS_ILOCK_EXCL);
505 if (imap.br_state == XFS_EXT_UNWRITTEN ||
506 imap.br_startblock == HOLESTARTBLOCK) {
507 start_zero_fsb = imap.br_startoff + imap.br_blockcount;
508 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
513 * There are blocks we need to zero.
515 zero_off = XFS_FSB_TO_B(mp, start_zero_fsb);
516 zero_len = XFS_FSB_TO_B(mp, imap.br_blockcount);
518 if ((zero_off + zero_len) > offset)
519 zero_len = offset - zero_off;
521 error = xfs_iozero(ip, zero_off, zero_len);
525 start_zero_fsb = imap.br_startoff + imap.br_blockcount;
526 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
533 * Common pre-write limit and setup checks.
535 * Called with the iolocked held either shared and exclusive according to
536 * @iolock, and returns with it held. Might upgrade the iolock to exclusive
537 * if called for a direct write beyond i_size.
540 xfs_file_aio_write_checks(
546 struct inode *inode = file->f_mapping->host;
547 struct xfs_inode *ip = XFS_I(inode);
551 error = generic_write_checks(file, pos, count, S_ISBLK(inode->i_mode));
556 * If the offset is beyond the size of the file, we need to zero any
557 * blocks that fall between the existing EOF and the start of this
558 * write. If zeroing is needed and we are currently holding the
559 * iolock shared, we need to update it to exclusive which implies
560 * having to redo all checks before.
562 if (*pos > i_size_read(inode)) {
563 if (*iolock == XFS_IOLOCK_SHARED) {
564 xfs_rw_iunlock(ip, *iolock);
565 *iolock = XFS_IOLOCK_EXCL;
566 xfs_rw_ilock(ip, *iolock);
569 error = -xfs_zero_eof(ip, *pos, i_size_read(inode));
575 * Updating the timestamps will grab the ilock again from
576 * xfs_fs_dirty_inode, so we have to call it after dropping the
577 * lock above. Eventually we should look into a way to avoid
578 * the pointless lock roundtrip.
580 if (likely(!(file->f_mode & FMODE_NOCMTIME))) {
581 error = file_update_time(file);
587 * If we're writing the file then make sure to clear the setuid and
588 * setgid bits if the process is not being run by root. This keeps
589 * people from modifying setuid and setgid binaries.
591 return file_remove_suid(file);
595 * xfs_file_dio_aio_write - handle direct IO writes
597 * Lock the inode appropriately to prepare for and issue a direct IO write.
598 * By separating it from the buffered write path we remove all the tricky to
599 * follow locking changes and looping.
601 * If there are cached pages or we're extending the file, we need IOLOCK_EXCL
602 * until we're sure the bytes at the new EOF have been zeroed and/or the cached
603 * pages are flushed out.
605 * In most cases the direct IO writes will be done holding IOLOCK_SHARED
606 * allowing them to be done in parallel with reads and other direct IO writes.
607 * However, if the IO is not aligned to filesystem blocks, the direct IO layer
608 * needs to do sub-block zeroing and that requires serialisation against other
609 * direct IOs to the same block. In this case we need to serialise the
610 * submission of the unaligned IOs so that we don't get racing block zeroing in
611 * the dio layer. To avoid the problem with aio, we also need to wait for
612 * outstanding IOs to complete so that unwritten extent conversion is completed
613 * before we try to map the overlapping block. This is currently implemented by
614 * hitting it with a big hammer (i.e. inode_dio_wait()).
616 * Returns with locks held indicated by @iolock and errors indicated by
617 * negative return values.
620 xfs_file_dio_aio_write(
622 struct iov_iter *iter,
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 int unaligned_io = 0;
634 struct xfs_buftarg *target = XFS_IS_REALTIME_INODE(ip) ?
635 mp->m_rtdev_targp : mp->m_ddev_targp;
637 if ((pos & target->bt_smask) || (count & target->bt_smask))
638 return -XFS_ERROR(EINVAL);
640 if ((pos & mp->m_blockmask) || ((pos + count) & mp->m_blockmask))
644 * We don't need to take an exclusive lock unless there page cache needs
645 * to be invalidated or unaligned IO is being executed. We don't need to
646 * consider the EOF extension case here because
647 * xfs_file_aio_write_checks() will relock the inode as necessary for
648 * EOF zeroing cases and fill out the new inode size as appropriate.
650 if (unaligned_io || mapping->nrpages)
651 iolock = XFS_IOLOCK_EXCL;
653 iolock = XFS_IOLOCK_SHARED;
654 xfs_rw_ilock(ip, iolock);
657 * Recheck if there are cached pages that need invalidate after we got
658 * the iolock to protect against other threads adding new pages while
659 * we were waiting for the iolock.
661 if (mapping->nrpages && iolock == XFS_IOLOCK_SHARED) {
662 xfs_rw_iunlock(ip, iolock);
663 iolock = XFS_IOLOCK_EXCL;
664 xfs_rw_ilock(ip, iolock);
667 ret = xfs_file_aio_write_checks(file, &pos, &count, &iolock);
671 if (mapping->nrpages) {
672 ret = -filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
676 truncate_pagecache_range(VFS_I(ip), pos, -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_iter(iocb, iter,
692 pos, &iocb->ki_pos, count);
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 struct iov_iter *iter,
709 struct file *file = iocb->ki_filp;
710 struct address_space *mapping = file->f_mapping;
711 struct inode *inode = mapping->host;
712 struct xfs_inode *ip = XFS_I(inode);
715 int iolock = XFS_IOLOCK_EXCL;
717 xfs_rw_ilock(ip, iolock);
719 ret = xfs_file_aio_write_checks(file, &pos, &count, &iolock);
723 /* We can write back this queue in page reclaim */
724 current->backing_dev_info = mapping->backing_dev_info;
727 trace_xfs_file_buffered_write(ip, count, iocb->ki_pos, 0);
728 ret = generic_file_buffered_write_iter(iocb, iter,
729 pos, &iocb->ki_pos, count, 0);
732 * If we just got an ENOSPC, try to write back all dirty inodes to
733 * convert delalloc space to free up some of the excess reserved
736 if (ret == -ENOSPC && !enospc) {
738 xfs_flush_inodes(ip->i_mount);
742 current->backing_dev_info = NULL;
744 xfs_rw_iunlock(ip, iolock);
751 struct iov_iter *iter,
754 struct file *file = iocb->ki_filp;
755 struct address_space *mapping = file->f_mapping;
756 struct inode *inode = mapping->host;
757 struct xfs_inode *ip = XFS_I(inode);
761 XFS_STATS_INC(xs_write_calls);
763 BUG_ON(iocb->ki_pos != pos);
765 count = iov_iter_count(iter);
770 if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
775 if (unlikely(file->f_flags & O_DIRECT))
776 ret = xfs_file_dio_aio_write(iocb, iter, pos, count);
778 ret = xfs_file_buffered_aio_write(iocb, iter, pos, count);
783 XFS_STATS_ADD(xs_write_bytes, ret);
785 /* Handle various SYNC-type writes */
786 err = generic_write_sync(file, pos, ret);
802 struct inode *inode = file_inode(file);
806 xfs_inode_t *ip = XFS_I(inode);
807 int cmd = XFS_IOC_RESVSP;
808 int attr_flags = XFS_ATTR_NOLOCK;
810 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
817 xfs_ilock(ip, XFS_IOLOCK_EXCL);
819 if (mode & FALLOC_FL_PUNCH_HOLE)
820 cmd = XFS_IOC_UNRESVSP;
822 /* check the new inode size is valid before allocating */
823 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
824 offset + len > i_size_read(inode)) {
825 new_size = offset + len;
826 error = inode_newsize_ok(inode, new_size);
831 if (file->f_flags & O_DSYNC)
832 attr_flags |= XFS_ATTR_SYNC;
834 error = -xfs_change_file_space(ip, cmd, &bf, 0, attr_flags);
838 /* Change file size if needed */
842 iattr.ia_valid = ATTR_SIZE;
843 iattr.ia_size = new_size;
844 error = -xfs_setattr_size(ip, &iattr, XFS_ATTR_NOLOCK);
848 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
858 if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
860 if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb)))
870 struct xfs_inode *ip = XFS_I(inode);
874 error = xfs_file_open(inode, file);
879 * If there are any blocks, read-ahead block 0 as we're almost
880 * certain to have the next operation be a read there.
882 mode = xfs_ilock_map_shared(ip);
883 if (ip->i_d.di_nextents > 0)
884 xfs_dir3_data_readahead(NULL, ip, 0, -1);
885 xfs_iunlock(ip, mode);
894 return -xfs_release(XFS_I(inode));
900 struct dir_context *ctx)
902 struct inode *inode = file_inode(file);
903 xfs_inode_t *ip = XFS_I(inode);
908 * The Linux API doesn't pass down the total size of the buffer
909 * we read into down to the filesystem. With the filldir concept
910 * it's not needed for correct information, but the XFS dir2 leaf
911 * code wants an estimate of the buffer size to calculate it's
912 * readahead window and size the buffers used for mapping to
915 * Try to give it an estimate that's good enough, maybe at some
916 * point we can change the ->readdir prototype to include the
917 * buffer size. For now we use the current glibc buffer size.
919 bufsize = (size_t)min_t(loff_t, 32768, ip->i_d.di_size);
921 error = xfs_readdir(ip, ctx, bufsize);
930 struct vm_area_struct *vma)
932 vma->vm_ops = &xfs_file_vm_ops;
939 * mmap()d file has taken write protection fault and is being made
940 * writable. We can set the page state up correctly for a writable
941 * page, which means we can do correct delalloc accounting (ENOSPC
942 * checking!) and unwritten extent mapping.
946 struct vm_area_struct *vma,
947 struct vm_fault *vmf)
949 return block_page_mkwrite(vma, vmf, xfs_get_blocks);
953 * This type is designed to indicate the type of offset we would like
954 * to search from page cache for either xfs_seek_data() or xfs_seek_hole().
962 * Lookup the desired type of offset from the given page.
964 * On success, return true and the offset argument will point to the
965 * start of the region that was found. Otherwise this function will
966 * return false and keep the offset argument unchanged.
969 xfs_lookup_buffer_offset(
974 loff_t lastoff = page_offset(page);
976 struct buffer_head *bh, *head;
978 bh = head = page_buffers(page);
981 * Unwritten extents that have data in the page
982 * cache covering them can be identified by the
983 * BH_Unwritten state flag. Pages with multiple
984 * buffers might have a mix of holes, data and
985 * unwritten extents - any buffer with valid
986 * data in it should have BH_Uptodate flag set
989 if (buffer_unwritten(bh) ||
990 buffer_uptodate(bh)) {
991 if (type == DATA_OFF)
994 if (type == HOLE_OFF)
1002 lastoff += bh->b_size;
1003 } while ((bh = bh->b_this_page) != head);
1009 * This routine is called to find out and return a data or hole offset
1010 * from the page cache for unwritten extents according to the desired
1011 * type for xfs_seek_data() or xfs_seek_hole().
1013 * The argument offset is used to tell where we start to search from the
1014 * page cache. Map is used to figure out the end points of the range to
1017 * Return true if the desired type of offset was found, and the argument
1018 * offset is filled with that address. Otherwise, return false and keep
1022 xfs_find_get_desired_pgoff(
1023 struct inode *inode,
1024 struct xfs_bmbt_irec *map,
1028 struct xfs_inode *ip = XFS_I(inode);
1029 struct xfs_mount *mp = ip->i_mount;
1030 struct pagevec pvec;
1034 loff_t startoff = *offset;
1035 loff_t lastoff = startoff;
1038 pagevec_init(&pvec, 0);
1040 index = startoff >> PAGE_CACHE_SHIFT;
1041 endoff = XFS_FSB_TO_B(mp, map->br_startoff + map->br_blockcount);
1042 end = endoff >> PAGE_CACHE_SHIFT;
1048 want = min_t(pgoff_t, end - index, PAGEVEC_SIZE);
1049 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index,
1052 * No page mapped into given range. If we are searching holes
1053 * and if this is the first time we got into the loop, it means
1054 * that the given offset is landed in a hole, return it.
1056 * If we have already stepped through some block buffers to find
1057 * holes but they all contains data. In this case, the last
1058 * offset is already updated and pointed to the end of the last
1059 * mapped page, if it does not reach the endpoint to search,
1060 * that means there should be a hole between them.
1062 if (nr_pages == 0) {
1063 /* Data search found nothing */
1064 if (type == DATA_OFF)
1067 ASSERT(type == HOLE_OFF);
1068 if (lastoff == startoff || lastoff < endoff) {
1076 * At lease we found one page. If this is the first time we
1077 * step into the loop, and if the first page index offset is
1078 * greater than the given search offset, a hole was found.
1080 if (type == HOLE_OFF && lastoff == startoff &&
1081 lastoff < page_offset(pvec.pages[0])) {
1086 for (i = 0; i < nr_pages; i++) {
1087 struct page *page = pvec.pages[i];
1091 * At this point, the page may be truncated or
1092 * invalidated (changing page->mapping to NULL),
1093 * or even swizzled back from swapper_space to tmpfs
1094 * file mapping. However, page->index will not change
1095 * because we have a reference on the page.
1097 * Searching done if the page index is out of range.
1098 * If the current offset is not reaches the end of
1099 * the specified search range, there should be a hole
1102 if (page->index > end) {
1103 if (type == HOLE_OFF && lastoff < endoff) {
1112 * Page truncated or invalidated(page->mapping == NULL).
1113 * We can freely skip it and proceed to check the next
1116 if (unlikely(page->mapping != inode->i_mapping)) {
1121 if (!page_has_buffers(page)) {
1126 found = xfs_lookup_buffer_offset(page, &b_offset, type);
1129 * The found offset may be less than the start
1130 * point to search if this is the first time to
1133 *offset = max_t(loff_t, startoff, b_offset);
1139 * We either searching data but nothing was found, or
1140 * searching hole but found a data buffer. In either
1141 * case, probably the next page contains the desired
1142 * things, update the last offset to it so.
1144 lastoff = page_offset(page) + PAGE_SIZE;
1149 * The number of returned pages less than our desired, search
1150 * done. In this case, nothing was found for searching data,
1151 * but we found a hole behind the last offset.
1153 if (nr_pages < want) {
1154 if (type == HOLE_OFF) {
1161 index = pvec.pages[i - 1]->index + 1;
1162 pagevec_release(&pvec);
1163 } while (index <= end);
1166 pagevec_release(&pvec);
1175 struct inode *inode = file->f_mapping->host;
1176 struct xfs_inode *ip = XFS_I(inode);
1177 struct xfs_mount *mp = ip->i_mount;
1178 loff_t uninitialized_var(offset);
1180 xfs_fileoff_t fsbno;
1185 lock = xfs_ilock_map_shared(ip);
1187 isize = i_size_read(inode);
1188 if (start >= isize) {
1194 * Try to read extents from the first block indicated
1195 * by fsbno to the end block of the file.
1197 fsbno = XFS_B_TO_FSBT(mp, start);
1198 end = XFS_B_TO_FSB(mp, isize);
1200 struct xfs_bmbt_irec map[2];
1204 error = xfs_bmapi_read(ip, fsbno, end - fsbno, map, &nmap,
1209 /* No extents at given offset, must be beyond EOF */
1215 for (i = 0; i < nmap; i++) {
1216 offset = max_t(loff_t, start,
1217 XFS_FSB_TO_B(mp, map[i].br_startoff));
1219 /* Landed in a data extent */
1220 if (map[i].br_startblock == DELAYSTARTBLOCK ||
1221 (map[i].br_state == XFS_EXT_NORM &&
1222 !isnullstartblock(map[i].br_startblock)))
1226 * Landed in an unwritten extent, try to search data
1229 if (map[i].br_state == XFS_EXT_UNWRITTEN) {
1230 if (xfs_find_get_desired_pgoff(inode, &map[i],
1237 * map[0] is hole or its an unwritten extent but
1238 * without data in page cache. Probably means that
1239 * we are reading after EOF if nothing in map[1].
1249 * Nothing was found, proceed to the next round of search
1250 * if reading offset not beyond or hit EOF.
1252 fsbno = map[i - 1].br_startoff + map[i - 1].br_blockcount;
1253 start = XFS_FSB_TO_B(mp, fsbno);
1254 if (start >= isize) {
1261 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
1264 xfs_iunlock_map_shared(ip, lock);
1276 struct inode *inode = file->f_mapping->host;
1277 struct xfs_inode *ip = XFS_I(inode);
1278 struct xfs_mount *mp = ip->i_mount;
1279 loff_t uninitialized_var(offset);
1281 xfs_fileoff_t fsbno;
1286 if (XFS_FORCED_SHUTDOWN(mp))
1287 return -XFS_ERROR(EIO);
1289 lock = xfs_ilock_map_shared(ip);
1291 isize = i_size_read(inode);
1292 if (start >= isize) {
1297 fsbno = XFS_B_TO_FSBT(mp, start);
1298 end = XFS_B_TO_FSB(mp, isize);
1301 struct xfs_bmbt_irec map[2];
1305 error = xfs_bmapi_read(ip, fsbno, end - fsbno, map, &nmap,
1310 /* No extents at given offset, must be beyond EOF */
1316 for (i = 0; i < nmap; i++) {
1317 offset = max_t(loff_t, start,
1318 XFS_FSB_TO_B(mp, map[i].br_startoff));
1320 /* Landed in a hole */
1321 if (map[i].br_startblock == HOLESTARTBLOCK)
1325 * Landed in an unwritten extent, try to search hole
1328 if (map[i].br_state == XFS_EXT_UNWRITTEN) {
1329 if (xfs_find_get_desired_pgoff(inode, &map[i],
1336 * map[0] contains data or its unwritten but contains
1337 * data in page cache, probably means that we are
1338 * reading after EOF. We should fix offset to point
1339 * to the end of the file(i.e., there is an implicit
1340 * hole at the end of any file).
1350 * Both mappings contains data, proceed to the next round of
1351 * search if the current reading offset not beyond or hit EOF.
1353 fsbno = map[i - 1].br_startoff + map[i - 1].br_blockcount;
1354 start = XFS_FSB_TO_B(mp, fsbno);
1355 if (start >= isize) {
1363 * At this point, we must have found a hole. However, the returned
1364 * offset may be bigger than the file size as it may be aligned to
1365 * page boundary for unwritten extents, we need to deal with this
1366 * situation in particular.
1368 offset = min_t(loff_t, offset, isize);
1369 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
1372 xfs_iunlock_map_shared(ip, lock);
1389 return generic_file_llseek(file, offset, origin);
1391 return xfs_seek_data(file, offset);
1393 return xfs_seek_hole(file, offset);
1399 const struct file_operations xfs_file_operations = {
1400 .llseek = xfs_file_llseek,
1401 .read = do_sync_read,
1402 .write = do_sync_write,
1403 .read_iter = xfs_file_read_iter,
1404 .write_iter = xfs_file_write_iter,
1405 .splice_read = xfs_file_splice_read,
1406 .splice_write = xfs_file_splice_write,
1407 .unlocked_ioctl = xfs_file_ioctl,
1408 #ifdef CONFIG_COMPAT
1409 .compat_ioctl = xfs_file_compat_ioctl,
1411 .mmap = xfs_file_mmap,
1412 .open = xfs_file_open,
1413 .release = xfs_file_release,
1414 .fsync = xfs_file_fsync,
1415 .fallocate = xfs_file_fallocate,
1418 const struct file_operations xfs_dir_file_operations = {
1419 .open = xfs_dir_open,
1420 .read = generic_read_dir,
1421 .iterate = xfs_file_readdir,
1422 .llseek = generic_file_llseek,
1423 .unlocked_ioctl = xfs_file_ioctl,
1424 #ifdef CONFIG_COMPAT
1425 .compat_ioctl = xfs_file_compat_ioctl,
1427 .fsync = xfs_dir_fsync,
1430 static const struct vm_operations_struct xfs_file_vm_ops = {
1431 .fault = filemap_fault,
1432 .page_mkwrite = xfs_vm_page_mkwrite,
1433 .remap_pages = generic_file_remap_pages,