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
25 #include "xfs_trans.h"
26 #include "xfs_mount.h"
27 #include "xfs_bmap_btree.h"
28 #include "xfs_alloc.h"
29 #include "xfs_dinode.h"
30 #include "xfs_inode.h"
31 #include "xfs_inode_item.h"
33 #include "xfs_error.h"
34 #include "xfs_vnodeops.h"
35 #include "xfs_da_btree.h"
36 #include "xfs_ioctl.h"
37 #include "xfs_trace.h"
39 #include <linux/dcache.h>
40 #include <linux/falloc.h>
42 static const struct vm_operations_struct xfs_file_vm_ops;
45 * Locking primitives for read and write IO paths to ensure we consistently use
46 * and order the inode->i_mutex, ip->i_lock and ip->i_iolock.
53 if (type & XFS_IOLOCK_EXCL)
54 mutex_lock(&VFS_I(ip)->i_mutex);
63 xfs_iunlock(ip, type);
64 if (type & XFS_IOLOCK_EXCL)
65 mutex_unlock(&VFS_I(ip)->i_mutex);
73 xfs_ilock_demote(ip, type);
74 if (type & XFS_IOLOCK_EXCL)
75 mutex_unlock(&VFS_I(ip)->i_mutex);
81 * xfs_iozero clears the specified range of buffer supplied,
82 * and marks all the affected blocks as valid and modified. If
83 * an affected block is not allocated, it will be allocated. If
84 * an affected block is not completely overwritten, and is not
85 * valid before the operation, it will be read from disk before
86 * being partially zeroed.
90 struct xfs_inode *ip, /* inode */
91 loff_t pos, /* offset in file */
92 size_t count) /* size of data to zero */
95 struct address_space *mapping;
98 mapping = VFS_I(ip)->i_mapping;
100 unsigned offset, bytes;
103 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
104 bytes = PAGE_CACHE_SIZE - offset;
108 status = pagecache_write_begin(NULL, mapping, pos, bytes,
109 AOP_FLAG_UNINTERRUPTIBLE,
114 zero_user(page, offset, bytes);
116 status = pagecache_write_end(NULL, mapping, pos, bytes, bytes,
118 WARN_ON(status <= 0); /* can't return less than zero! */
128 * Fsync operations on directories are much simpler than on regular files,
129 * as there is no file data to flush, and thus also no need for explicit
130 * cache flush operations, and there are no non-transaction metadata updates
131 * on directories either.
140 struct xfs_inode *ip = XFS_I(file->f_mapping->host);
141 struct xfs_mount *mp = ip->i_mount;
144 trace_xfs_dir_fsync(ip);
146 xfs_ilock(ip, XFS_ILOCK_SHARED);
147 if (xfs_ipincount(ip))
148 lsn = ip->i_itemp->ili_last_lsn;
149 xfs_iunlock(ip, XFS_ILOCK_SHARED);
153 return _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, NULL);
163 struct inode *inode = file->f_mapping->host;
164 struct xfs_inode *ip = XFS_I(inode);
165 struct xfs_mount *mp = ip->i_mount;
170 trace_xfs_file_fsync(ip);
172 error = filemap_write_and_wait_range(inode->i_mapping, start, end);
176 if (XFS_FORCED_SHUTDOWN(mp))
177 return -XFS_ERROR(EIO);
179 xfs_iflags_clear(ip, XFS_ITRUNCATED);
181 if (mp->m_flags & XFS_MOUNT_BARRIER) {
183 * If we have an RT and/or log subvolume we need to make sure
184 * to flush the write cache the device used for file data
185 * first. This is to ensure newly written file data make
186 * it to disk before logging the new inode size in case of
187 * an extending write.
189 if (XFS_IS_REALTIME_INODE(ip))
190 xfs_blkdev_issue_flush(mp->m_rtdev_targp);
191 else if (mp->m_logdev_targp != mp->m_ddev_targp)
192 xfs_blkdev_issue_flush(mp->m_ddev_targp);
196 * All metadata updates are logged, which means that we just have
197 * to flush the log up to the latest LSN that touched the inode.
199 xfs_ilock(ip, XFS_ILOCK_SHARED);
200 if (xfs_ipincount(ip)) {
202 (ip->i_itemp->ili_fields & ~XFS_ILOG_TIMESTAMP))
203 lsn = ip->i_itemp->ili_last_lsn;
205 xfs_iunlock(ip, XFS_ILOCK_SHARED);
208 error = _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, &log_flushed);
211 * If we only have a single device, and the log force about was
212 * a no-op we might have to flush the data device cache here.
213 * This can only happen for fdatasync/O_DSYNC if we were overwriting
214 * an already allocated file and thus do not have any metadata to
217 if ((mp->m_flags & XFS_MOUNT_BARRIER) &&
218 mp->m_logdev_targp == mp->m_ddev_targp &&
219 !XFS_IS_REALTIME_INODE(ip) &&
221 xfs_blkdev_issue_flush(mp->m_ddev_targp);
229 const struct iovec *iovp,
230 unsigned long nr_segs,
233 struct file *file = iocb->ki_filp;
234 struct inode *inode = file->f_mapping->host;
235 struct xfs_inode *ip = XFS_I(inode);
236 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 /* START copy & waste from filemap.c */
253 for (seg = 0; seg < nr_segs; seg++) {
254 const struct iovec *iv = &iovp[seg];
257 * If any segment has a negative length, or the cumulative
258 * length ever wraps negative then return -EINVAL.
261 if (unlikely((ssize_t)(size|iv->iov_len) < 0))
262 return XFS_ERROR(-EINVAL);
264 /* END copy & waste from filemap.c */
266 if (unlikely(ioflags & IO_ISDIRECT)) {
267 xfs_buftarg_t *target =
268 XFS_IS_REALTIME_INODE(ip) ?
269 mp->m_rtdev_targp : mp->m_ddev_targp;
270 if ((iocb->ki_pos & target->bt_smask) ||
271 (size & target->bt_smask)) {
272 if (iocb->ki_pos == i_size_read(inode))
274 return -XFS_ERROR(EINVAL);
278 n = XFS_MAXIOFFSET(mp) - iocb->ki_pos;
279 if (n <= 0 || size == 0)
285 if (XFS_FORCED_SHUTDOWN(mp))
289 * Locking is a bit tricky here. If we take an exclusive lock
290 * for direct IO, we effectively serialise all new concurrent
291 * read IO to this file and block it behind IO that is currently in
292 * progress because IO in progress holds the IO lock shared. We only
293 * need to hold the lock exclusive to blow away the page cache, so
294 * only take lock exclusively if the page cache needs invalidation.
295 * This allows the normal direct IO case of no page cache pages to
296 * proceeed concurrently without serialisation.
298 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
299 if ((ioflags & IO_ISDIRECT) && inode->i_mapping->nrpages) {
300 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
301 xfs_rw_ilock(ip, XFS_IOLOCK_EXCL);
303 if (inode->i_mapping->nrpages) {
304 ret = -xfs_flushinval_pages(ip,
305 (iocb->ki_pos & PAGE_CACHE_MASK),
306 -1, FI_REMAPF_LOCKED);
308 xfs_rw_iunlock(ip, XFS_IOLOCK_EXCL);
312 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
315 trace_xfs_file_read(ip, size, iocb->ki_pos, ioflags);
317 ret = generic_file_aio_read(iocb, iovp, nr_segs, iocb->ki_pos);
319 XFS_STATS_ADD(xs_read_bytes, ret);
321 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
326 xfs_file_splice_read(
329 struct pipe_inode_info *pipe,
333 struct xfs_inode *ip = XFS_I(infilp->f_mapping->host);
337 XFS_STATS_INC(xs_read_calls);
339 if (infilp->f_mode & FMODE_NOCMTIME)
342 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
345 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
347 trace_xfs_file_splice_read(ip, count, *ppos, ioflags);
349 ret = generic_file_splice_read(infilp, ppos, pipe, count, flags);
351 XFS_STATS_ADD(xs_read_bytes, ret);
353 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
358 * xfs_file_splice_write() does not use xfs_rw_ilock() because
359 * generic_file_splice_write() takes the i_mutex itself. This, in theory,
360 * couuld cause lock inversions between the aio_write path and the splice path
361 * if someone is doing concurrent splice(2) based writes and write(2) based
362 * writes to the same inode. The only real way to fix this is to re-implement
363 * the generic code here with correct locking orders.
366 xfs_file_splice_write(
367 struct pipe_inode_info *pipe,
368 struct file *outfilp,
373 struct inode *inode = outfilp->f_mapping->host;
374 struct xfs_inode *ip = XFS_I(inode);
378 XFS_STATS_INC(xs_write_calls);
380 if (outfilp->f_mode & FMODE_NOCMTIME)
383 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
386 xfs_ilock(ip, XFS_IOLOCK_EXCL);
388 trace_xfs_file_splice_write(ip, count, *ppos, ioflags);
390 ret = generic_file_splice_write(pipe, outfilp, ppos, count, flags);
392 XFS_STATS_ADD(xs_write_bytes, ret);
394 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
399 * This routine is called to handle zeroing any space in the last block of the
400 * file that is beyond the EOF. We do this since the size is being increased
401 * without writing anything to that block and we don't want to read the
402 * garbage on the disk.
404 STATIC int /* error (positive) */
406 struct xfs_inode *ip,
410 struct xfs_mount *mp = ip->i_mount;
411 xfs_fileoff_t last_fsb = XFS_B_TO_FSBT(mp, isize);
412 int zero_offset = XFS_B_FSB_OFFSET(mp, isize);
416 struct xfs_bmbt_irec imap;
418 xfs_ilock(ip, XFS_ILOCK_EXCL);
419 error = xfs_bmapi_read(ip, last_fsb, 1, &imap, &nimaps, 0);
420 xfs_iunlock(ip, XFS_ILOCK_EXCL);
427 * If the block underlying isize is just a hole, then there
428 * is nothing to zero.
430 if (imap.br_startblock == HOLESTARTBLOCK)
433 zero_len = mp->m_sb.sb_blocksize - zero_offset;
434 if (isize + zero_len > offset)
435 zero_len = offset - isize;
436 return xfs_iozero(ip, isize, zero_len);
440 * Zero any on disk space between the current EOF and the new, larger EOF.
442 * This handles the normal case of zeroing the remainder of the last block in
443 * the file and the unusual case of zeroing blocks out beyond the size of the
444 * file. This second case only happens with fixed size extents and when the
445 * system crashes before the inode size was updated but after blocks were
448 * Expects the iolock to be held exclusive, and will take the ilock internally.
450 int /* error (positive) */
452 struct xfs_inode *ip,
453 xfs_off_t offset, /* starting I/O offset */
454 xfs_fsize_t isize) /* current inode size */
456 struct xfs_mount *mp = ip->i_mount;
457 xfs_fileoff_t start_zero_fsb;
458 xfs_fileoff_t end_zero_fsb;
459 xfs_fileoff_t zero_count_fsb;
460 xfs_fileoff_t last_fsb;
461 xfs_fileoff_t zero_off;
462 xfs_fsize_t zero_len;
465 struct xfs_bmbt_irec imap;
467 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
468 ASSERT(offset > isize);
471 * First handle zeroing the block on which isize resides.
473 * We only zero a part of that block so it is handled specially.
475 if (XFS_B_FSB_OFFSET(mp, isize) != 0) {
476 error = xfs_zero_last_block(ip, offset, isize);
482 * Calculate the range between the new size and the old where blocks
483 * needing to be zeroed may exist.
485 * To get the block where the last byte in the file currently resides,
486 * we need to subtract one from the size and truncate back to a block
487 * boundary. We subtract 1 in case the size is exactly on a block
490 last_fsb = isize ? XFS_B_TO_FSBT(mp, isize - 1) : (xfs_fileoff_t)-1;
491 start_zero_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
492 end_zero_fsb = XFS_B_TO_FSBT(mp, offset - 1);
493 ASSERT((xfs_sfiloff_t)last_fsb < (xfs_sfiloff_t)start_zero_fsb);
494 if (last_fsb == end_zero_fsb) {
496 * The size was only incremented on its last block.
497 * We took care of that above, so just return.
502 ASSERT(start_zero_fsb <= end_zero_fsb);
503 while (start_zero_fsb <= end_zero_fsb) {
505 zero_count_fsb = end_zero_fsb - start_zero_fsb + 1;
507 xfs_ilock(ip, XFS_ILOCK_EXCL);
508 error = xfs_bmapi_read(ip, start_zero_fsb, zero_count_fsb,
510 xfs_iunlock(ip, XFS_ILOCK_EXCL);
516 if (imap.br_state == XFS_EXT_UNWRITTEN ||
517 imap.br_startblock == HOLESTARTBLOCK) {
518 start_zero_fsb = imap.br_startoff + imap.br_blockcount;
519 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
524 * There are blocks we need to zero.
526 zero_off = XFS_FSB_TO_B(mp, start_zero_fsb);
527 zero_len = XFS_FSB_TO_B(mp, imap.br_blockcount);
529 if ((zero_off + zero_len) > offset)
530 zero_len = offset - zero_off;
532 error = xfs_iozero(ip, zero_off, zero_len);
536 start_zero_fsb = imap.br_startoff + imap.br_blockcount;
537 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
544 * Common pre-write limit and setup checks.
546 * Called with the iolocked held either shared and exclusive according to
547 * @iolock, and returns with it held. Might upgrade the iolock to exclusive
548 * if called for a direct write beyond i_size.
551 xfs_file_aio_write_checks(
557 struct inode *inode = file->f_mapping->host;
558 struct xfs_inode *ip = XFS_I(inode);
562 error = generic_write_checks(file, pos, count, S_ISBLK(inode->i_mode));
567 * If the offset is beyond the size of the file, we need to zero any
568 * blocks that fall between the existing EOF and the start of this
569 * write. If zeroing is needed and we are currently holding the
570 * iolock shared, we need to update it to exclusive which implies
571 * having to redo all checks before.
573 if (*pos > i_size_read(inode)) {
574 if (*iolock == XFS_IOLOCK_SHARED) {
575 xfs_rw_iunlock(ip, *iolock);
576 *iolock = XFS_IOLOCK_EXCL;
577 xfs_rw_ilock(ip, *iolock);
580 error = -xfs_zero_eof(ip, *pos, i_size_read(inode));
586 * Updating the timestamps will grab the ilock again from
587 * xfs_fs_dirty_inode, so we have to call it after dropping the
588 * lock above. Eventually we should look into a way to avoid
589 * the pointless lock roundtrip.
591 if (likely(!(file->f_mode & FMODE_NOCMTIME)))
592 file_update_time(file);
595 * If we're writing the file then make sure to clear the setuid and
596 * setgid bits if the process is not being run by root. This keeps
597 * people from modifying setuid and setgid binaries.
599 return file_remove_suid(file);
603 * xfs_file_dio_aio_write - handle direct IO writes
605 * Lock the inode appropriately to prepare for and issue a direct IO write.
606 * By separating it from the buffered write path we remove all the tricky to
607 * follow locking changes and looping.
609 * If there are cached pages or we're extending the file, we need IOLOCK_EXCL
610 * until we're sure the bytes at the new EOF have been zeroed and/or the cached
611 * pages are flushed out.
613 * In most cases the direct IO writes will be done holding IOLOCK_SHARED
614 * allowing them to be done in parallel with reads and other direct IO writes.
615 * However, if the IO is not aligned to filesystem blocks, the direct IO layer
616 * needs to do sub-block zeroing and that requires serialisation against other
617 * direct IOs to the same block. In this case we need to serialise the
618 * submission of the unaligned IOs so that we don't get racing block zeroing in
619 * the dio layer. To avoid the problem with aio, we also need to wait for
620 * outstanding IOs to complete so that unwritten extent conversion is completed
621 * before we try to map the overlapping block. This is currently implemented by
622 * hitting it with a big hammer (i.e. inode_dio_wait()).
624 * Returns with locks held indicated by @iolock and errors indicated by
625 * negative return values.
628 xfs_file_dio_aio_write(
630 const struct iovec *iovp,
631 unsigned long nr_segs,
635 struct file *file = iocb->ki_filp;
636 struct address_space *mapping = file->f_mapping;
637 struct inode *inode = mapping->host;
638 struct xfs_inode *ip = XFS_I(inode);
639 struct xfs_mount *mp = ip->i_mount;
641 size_t count = ocount;
642 int unaligned_io = 0;
644 struct xfs_buftarg *target = XFS_IS_REALTIME_INODE(ip) ?
645 mp->m_rtdev_targp : mp->m_ddev_targp;
647 if ((pos & target->bt_smask) || (count & target->bt_smask))
648 return -XFS_ERROR(EINVAL);
650 if ((pos & mp->m_blockmask) || ((pos + count) & mp->m_blockmask))
654 * We don't need to take an exclusive lock unless there page cache needs
655 * to be invalidated or unaligned IO is being executed. We don't need to
656 * consider the EOF extension case here because
657 * xfs_file_aio_write_checks() will relock the inode as necessary for
658 * EOF zeroing cases and fill out the new inode size as appropriate.
660 if (unaligned_io || mapping->nrpages)
661 iolock = XFS_IOLOCK_EXCL;
663 iolock = XFS_IOLOCK_SHARED;
664 xfs_rw_ilock(ip, iolock);
667 * Recheck if there are cached pages that need invalidate after we got
668 * the iolock to protect against other threads adding new pages while
669 * we were waiting for the iolock.
671 if (mapping->nrpages && iolock == XFS_IOLOCK_SHARED) {
672 xfs_rw_iunlock(ip, iolock);
673 iolock = XFS_IOLOCK_EXCL;
674 xfs_rw_ilock(ip, iolock);
677 ret = xfs_file_aio_write_checks(file, &pos, &count, &iolock);
681 if (mapping->nrpages) {
682 ret = -xfs_flushinval_pages(ip, (pos & PAGE_CACHE_MASK), -1,
689 * If we are doing unaligned IO, wait for all other IO to drain,
690 * otherwise demote the lock if we had to flush cached pages
693 inode_dio_wait(inode);
694 else if (iolock == XFS_IOLOCK_EXCL) {
695 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
696 iolock = XFS_IOLOCK_SHARED;
699 trace_xfs_file_direct_write(ip, count, iocb->ki_pos, 0);
700 ret = generic_file_direct_write(iocb, iovp,
701 &nr_segs, pos, &iocb->ki_pos, count, ocount);
704 xfs_rw_iunlock(ip, iolock);
706 /* No fallback to buffered IO on errors for XFS. */
707 ASSERT(ret < 0 || ret == count);
712 xfs_file_buffered_aio_write(
714 const struct iovec *iovp,
715 unsigned long nr_segs,
719 struct file *file = iocb->ki_filp;
720 struct address_space *mapping = file->f_mapping;
721 struct inode *inode = mapping->host;
722 struct xfs_inode *ip = XFS_I(inode);
725 int iolock = XFS_IOLOCK_EXCL;
726 size_t count = ocount;
728 xfs_rw_ilock(ip, iolock);
730 ret = xfs_file_aio_write_checks(file, &pos, &count, &iolock);
734 /* We can write back this queue in page reclaim */
735 current->backing_dev_info = mapping->backing_dev_info;
738 trace_xfs_file_buffered_write(ip, count, iocb->ki_pos, 0);
739 ret = generic_file_buffered_write(iocb, iovp, nr_segs,
740 pos, &iocb->ki_pos, count, ret);
742 * if we just got an ENOSPC, flush the inode now we aren't holding any
743 * page locks and retry *once*
745 if (ret == -ENOSPC && !enospc) {
747 ret = -xfs_flush_pages(ip, 0, -1, 0, FI_NONE);
752 current->backing_dev_info = NULL;
754 xfs_rw_iunlock(ip, iolock);
761 const struct iovec *iovp,
762 unsigned long nr_segs,
765 struct file *file = iocb->ki_filp;
766 struct address_space *mapping = file->f_mapping;
767 struct inode *inode = mapping->host;
768 struct xfs_inode *ip = XFS_I(inode);
772 XFS_STATS_INC(xs_write_calls);
774 BUG_ON(iocb->ki_pos != pos);
776 ret = generic_segment_checks(iovp, &nr_segs, &ocount, VERIFY_READ);
783 xfs_wait_for_freeze(ip->i_mount, SB_FREEZE_WRITE);
785 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
788 if (unlikely(file->f_flags & O_DIRECT))
789 ret = xfs_file_dio_aio_write(iocb, iovp, nr_segs, pos, ocount);
791 ret = xfs_file_buffered_aio_write(iocb, iovp, nr_segs, pos,
797 XFS_STATS_ADD(xs_write_bytes, ret);
799 /* Handle various SYNC-type writes */
800 err = generic_write_sync(file, pos, ret);
815 struct inode *inode = file->f_path.dentry->d_inode;
819 xfs_inode_t *ip = XFS_I(inode);
820 int cmd = XFS_IOC_RESVSP;
821 int attr_flags = XFS_ATTR_NOLOCK;
823 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
830 xfs_ilock(ip, XFS_IOLOCK_EXCL);
832 if (mode & FALLOC_FL_PUNCH_HOLE)
833 cmd = XFS_IOC_UNRESVSP;
835 /* check the new inode size is valid before allocating */
836 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
837 offset + len > i_size_read(inode)) {
838 new_size = offset + len;
839 error = inode_newsize_ok(inode, new_size);
844 if (file->f_flags & O_DSYNC)
845 attr_flags |= XFS_ATTR_SYNC;
847 error = -xfs_change_file_space(ip, cmd, &bf, 0, attr_flags);
851 /* Change file size if needed */
855 iattr.ia_valid = ATTR_SIZE;
856 iattr.ia_size = new_size;
857 error = -xfs_setattr_size(ip, &iattr, XFS_ATTR_NOLOCK);
861 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
871 if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
873 if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb)))
883 struct xfs_inode *ip = XFS_I(inode);
887 error = xfs_file_open(inode, file);
892 * If there are any blocks, read-ahead block 0 as we're almost
893 * certain to have the next operation be a read there.
895 mode = xfs_ilock_map_shared(ip);
896 if (ip->i_d.di_nextents > 0)
897 xfs_da_reada_buf(NULL, ip, 0, XFS_DATA_FORK);
898 xfs_iunlock(ip, mode);
907 return -xfs_release(XFS_I(inode));
916 struct inode *inode = filp->f_path.dentry->d_inode;
917 xfs_inode_t *ip = XFS_I(inode);
922 * The Linux API doesn't pass down the total size of the buffer
923 * we read into down to the filesystem. With the filldir concept
924 * it's not needed for correct information, but the XFS dir2 leaf
925 * code wants an estimate of the buffer size to calculate it's
926 * readahead window and size the buffers used for mapping to
929 * Try to give it an estimate that's good enough, maybe at some
930 * point we can change the ->readdir prototype to include the
931 * buffer size. For now we use the current glibc buffer size.
933 bufsize = (size_t)min_t(loff_t, 32768, ip->i_d.di_size);
935 error = xfs_readdir(ip, dirent, bufsize,
936 (xfs_off_t *)&filp->f_pos, filldir);
945 struct vm_area_struct *vma)
947 vma->vm_ops = &xfs_file_vm_ops;
948 vma->vm_flags |= VM_CAN_NONLINEAR;
955 * mmap()d file has taken write protection fault and is being made
956 * writable. We can set the page state up correctly for a writable
957 * page, which means we can do correct delalloc accounting (ENOSPC
958 * checking!) and unwritten extent mapping.
962 struct vm_area_struct *vma,
963 struct vm_fault *vmf)
965 return block_page_mkwrite(vma, vmf, xfs_get_blocks);
968 const struct file_operations xfs_file_operations = {
969 .llseek = generic_file_llseek,
970 .read = do_sync_read,
971 .write = do_sync_write,
972 .aio_read = xfs_file_aio_read,
973 .aio_write = xfs_file_aio_write,
974 .splice_read = xfs_file_splice_read,
975 .splice_write = xfs_file_splice_write,
976 .unlocked_ioctl = xfs_file_ioctl,
978 .compat_ioctl = xfs_file_compat_ioctl,
980 .mmap = xfs_file_mmap,
981 .open = xfs_file_open,
982 .release = xfs_file_release,
983 .fsync = xfs_file_fsync,
984 .fallocate = xfs_file_fallocate,
987 const struct file_operations xfs_dir_file_operations = {
988 .open = xfs_dir_open,
989 .read = generic_read_dir,
990 .readdir = xfs_file_readdir,
991 .llseek = generic_file_llseek,
992 .unlocked_ioctl = xfs_file_ioctl,
994 .compat_ioctl = xfs_file_compat_ioctl,
996 .fsync = xfs_dir_fsync,
999 static const struct vm_operations_struct xfs_file_vm_ops = {
1000 .fault = filemap_fault,
1001 .page_mkwrite = xfs_vm_page_mkwrite,