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>
40 static const struct vm_operations_struct xfs_file_vm_ops;
43 * Locking primitives for read and write IO paths to ensure we consistently use
44 * and order the inode->i_mutex, ip->i_lock and ip->i_iolock.
51 if (type & XFS_IOLOCK_EXCL)
52 mutex_lock(&VFS_I(ip)->i_mutex);
61 xfs_iunlock(ip, type);
62 if (type & XFS_IOLOCK_EXCL)
63 mutex_unlock(&VFS_I(ip)->i_mutex);
71 xfs_ilock_demote(ip, type);
72 if (type & XFS_IOLOCK_EXCL)
73 mutex_unlock(&VFS_I(ip)->i_mutex);
79 * xfs_iozero clears the specified range of buffer supplied,
80 * and marks all the affected blocks as valid and modified. If
81 * an affected block is not allocated, it will be allocated. If
82 * an affected block is not completely overwritten, and is not
83 * valid before the operation, it will be read from disk before
84 * being partially zeroed.
88 struct xfs_inode *ip, /* inode */
89 loff_t pos, /* offset in file */
90 size_t count) /* size of data to zero */
93 struct address_space *mapping;
96 mapping = VFS_I(ip)->i_mapping;
98 unsigned offset, bytes;
101 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
102 bytes = PAGE_CACHE_SIZE - offset;
106 status = pagecache_write_begin(NULL, mapping, pos, bytes,
107 AOP_FLAG_UNINTERRUPTIBLE,
112 zero_user(page, offset, bytes);
114 status = pagecache_write_end(NULL, mapping, pos, bytes, bytes,
116 WARN_ON(status <= 0); /* can't return less than zero! */
126 * Fsync operations on directories are much simpler than on regular files,
127 * as there is no file data to flush, and thus also no need for explicit
128 * cache flush operations, and there are no non-transaction metadata updates
129 * on directories either.
138 struct xfs_inode *ip = XFS_I(file->f_mapping->host);
139 struct xfs_mount *mp = ip->i_mount;
142 trace_xfs_dir_fsync(ip);
144 xfs_ilock(ip, XFS_ILOCK_SHARED);
145 if (xfs_ipincount(ip))
146 lsn = ip->i_itemp->ili_last_lsn;
147 xfs_iunlock(ip, XFS_ILOCK_SHARED);
151 return _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, NULL);
161 struct inode *inode = file->f_mapping->host;
162 struct xfs_inode *ip = XFS_I(inode);
163 struct xfs_mount *mp = ip->i_mount;
168 trace_xfs_file_fsync(ip);
170 error = filemap_write_and_wait_range(inode->i_mapping, start, end);
174 if (XFS_FORCED_SHUTDOWN(mp))
175 return -XFS_ERROR(EIO);
177 xfs_iflags_clear(ip, XFS_ITRUNCATED);
179 if (mp->m_flags & XFS_MOUNT_BARRIER) {
181 * If we have an RT and/or log subvolume we need to make sure
182 * to flush the write cache the device used for file data
183 * first. This is to ensure newly written file data make
184 * it to disk before logging the new inode size in case of
185 * an extending write.
187 if (XFS_IS_REALTIME_INODE(ip))
188 xfs_blkdev_issue_flush(mp->m_rtdev_targp);
189 else if (mp->m_logdev_targp != mp->m_ddev_targp)
190 xfs_blkdev_issue_flush(mp->m_ddev_targp);
194 * All metadata updates are logged, which means that we just have
195 * to flush the log up to the latest LSN that touched the inode.
197 xfs_ilock(ip, XFS_ILOCK_SHARED);
198 if (xfs_ipincount(ip)) {
200 (ip->i_itemp->ili_fields & ~XFS_ILOG_TIMESTAMP))
201 lsn = ip->i_itemp->ili_last_lsn;
203 xfs_iunlock(ip, XFS_ILOCK_SHARED);
206 error = _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, &log_flushed);
209 * If we only have a single device, and the log force about was
210 * a no-op we might have to flush the data device cache here.
211 * This can only happen for fdatasync/O_DSYNC if we were overwriting
212 * an already allocated file and thus do not have any metadata to
215 if ((mp->m_flags & XFS_MOUNT_BARRIER) &&
216 mp->m_logdev_targp == mp->m_ddev_targp &&
217 !XFS_IS_REALTIME_INODE(ip) &&
219 xfs_blkdev_issue_flush(mp->m_ddev_targp);
227 const struct iovec *iovp,
228 unsigned long nr_segs,
231 struct file *file = iocb->ki_filp;
232 struct inode *inode = file->f_mapping->host;
233 struct xfs_inode *ip = XFS_I(inode);
234 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 /* START copy & waste from filemap.c */
251 for (seg = 0; seg < nr_segs; seg++) {
252 const struct iovec *iv = &iovp[seg];
255 * If any segment has a negative length, or the cumulative
256 * length ever wraps negative then return -EINVAL.
259 if (unlikely((ssize_t)(size|iv->iov_len) < 0))
260 return XFS_ERROR(-EINVAL);
262 /* END copy & waste from filemap.c */
264 if (unlikely(ioflags & IO_ISDIRECT)) {
265 xfs_buftarg_t *target =
266 XFS_IS_REALTIME_INODE(ip) ?
267 mp->m_rtdev_targp : mp->m_ddev_targp;
268 if ((iocb->ki_pos & target->bt_smask) ||
269 (size & target->bt_smask)) {
270 if (iocb->ki_pos == i_size_read(inode))
272 return -XFS_ERROR(EINVAL);
276 n = XFS_MAXIOFFSET(mp) - iocb->ki_pos;
277 if (n <= 0 || size == 0)
283 if (XFS_FORCED_SHUTDOWN(mp))
287 * Locking is a bit tricky here. If we take an exclusive lock
288 * for direct IO, we effectively serialise all new concurrent
289 * read IO to this file and block it behind IO that is currently in
290 * progress because IO in progress holds the IO lock shared. We only
291 * need to hold the lock exclusive to blow away the page cache, so
292 * only take lock exclusively if the page cache needs invalidation.
293 * This allows the normal direct IO case of no page cache pages to
294 * proceeed concurrently without serialisation.
296 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
297 if ((ioflags & IO_ISDIRECT) && inode->i_mapping->nrpages) {
298 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
299 xfs_rw_ilock(ip, XFS_IOLOCK_EXCL);
301 if (inode->i_mapping->nrpages) {
302 ret = -xfs_flushinval_pages(ip,
303 (iocb->ki_pos & PAGE_CACHE_MASK),
304 -1, FI_REMAPF_LOCKED);
306 xfs_rw_iunlock(ip, XFS_IOLOCK_EXCL);
310 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
313 trace_xfs_file_read(ip, size, iocb->ki_pos, ioflags);
315 ret = generic_file_aio_read(iocb, iovp, nr_segs, iocb->ki_pos);
317 XFS_STATS_ADD(xs_read_bytes, ret);
319 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
324 xfs_file_splice_read(
327 struct pipe_inode_info *pipe,
331 struct xfs_inode *ip = XFS_I(infilp->f_mapping->host);
335 XFS_STATS_INC(xs_read_calls);
337 if (infilp->f_mode & FMODE_NOCMTIME)
340 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
343 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
345 trace_xfs_file_splice_read(ip, count, *ppos, ioflags);
347 ret = generic_file_splice_read(infilp, ppos, pipe, count, flags);
349 XFS_STATS_ADD(xs_read_bytes, ret);
351 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
356 * xfs_file_splice_write() does not use xfs_rw_ilock() because
357 * generic_file_splice_write() takes the i_mutex itself. This, in theory,
358 * couuld cause lock inversions between the aio_write path and the splice path
359 * if someone is doing concurrent splice(2) based writes and write(2) based
360 * writes to the same inode. The only real way to fix this is to re-implement
361 * the generic code here with correct locking orders.
364 xfs_file_splice_write(
365 struct pipe_inode_info *pipe,
366 struct file *outfilp,
371 struct inode *inode = outfilp->f_mapping->host;
372 struct xfs_inode *ip = XFS_I(inode);
376 XFS_STATS_INC(xs_write_calls);
378 if (outfilp->f_mode & FMODE_NOCMTIME)
381 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
384 xfs_ilock(ip, XFS_IOLOCK_EXCL);
386 trace_xfs_file_splice_write(ip, count, *ppos, ioflags);
388 ret = generic_file_splice_write(pipe, outfilp, ppos, count, flags);
390 XFS_STATS_ADD(xs_write_bytes, ret);
392 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
397 * This routine is called to handle zeroing any space in the last block of the
398 * file that is beyond the EOF. We do this since the size is being increased
399 * without writing anything to that block and we don't want to read the
400 * garbage on the disk.
402 STATIC int /* error (positive) */
404 struct xfs_inode *ip,
408 struct xfs_mount *mp = ip->i_mount;
409 xfs_fileoff_t last_fsb = XFS_B_TO_FSBT(mp, isize);
410 int zero_offset = XFS_B_FSB_OFFSET(mp, isize);
414 struct xfs_bmbt_irec imap;
416 xfs_ilock(ip, XFS_ILOCK_EXCL);
417 error = xfs_bmapi_read(ip, last_fsb, 1, &imap, &nimaps, 0);
418 xfs_iunlock(ip, XFS_ILOCK_EXCL);
425 * If the block underlying isize is just a hole, then there
426 * is nothing to zero.
428 if (imap.br_startblock == HOLESTARTBLOCK)
431 zero_len = mp->m_sb.sb_blocksize - zero_offset;
432 if (isize + zero_len > offset)
433 zero_len = offset - isize;
434 return xfs_iozero(ip, isize, zero_len);
438 * Zero any on disk space between the current EOF and the new, larger EOF.
440 * This handles the normal case of zeroing the remainder of the last block in
441 * the file and the unusual case of zeroing blocks out beyond the size of the
442 * file. This second case only happens with fixed size extents and when the
443 * system crashes before the inode size was updated but after blocks were
446 * Expects the iolock to be held exclusive, and will take the ilock internally.
448 int /* error (positive) */
450 struct xfs_inode *ip,
451 xfs_off_t offset, /* starting I/O offset */
452 xfs_fsize_t isize) /* current inode size */
454 struct xfs_mount *mp = ip->i_mount;
455 xfs_fileoff_t start_zero_fsb;
456 xfs_fileoff_t end_zero_fsb;
457 xfs_fileoff_t zero_count_fsb;
458 xfs_fileoff_t last_fsb;
459 xfs_fileoff_t zero_off;
460 xfs_fsize_t zero_len;
463 struct xfs_bmbt_irec imap;
465 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
466 ASSERT(offset > isize);
469 * First handle zeroing the block on which isize resides.
471 * We only zero a part of that block so it is handled specially.
473 if (XFS_B_FSB_OFFSET(mp, isize) != 0) {
474 error = xfs_zero_last_block(ip, offset, isize);
480 * Calculate the range between the new size and the old where blocks
481 * needing to be zeroed may exist.
483 * To get the block where the last byte in the file currently resides,
484 * we need to subtract one from the size and truncate back to a block
485 * boundary. We subtract 1 in case the size is exactly on a block
488 last_fsb = isize ? XFS_B_TO_FSBT(mp, isize - 1) : (xfs_fileoff_t)-1;
489 start_zero_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
490 end_zero_fsb = XFS_B_TO_FSBT(mp, offset - 1);
491 ASSERT((xfs_sfiloff_t)last_fsb < (xfs_sfiloff_t)start_zero_fsb);
492 if (last_fsb == end_zero_fsb) {
494 * The size was only incremented on its last block.
495 * We took care of that above, so just return.
500 ASSERT(start_zero_fsb <= end_zero_fsb);
501 while (start_zero_fsb <= end_zero_fsb) {
503 zero_count_fsb = end_zero_fsb - start_zero_fsb + 1;
505 xfs_ilock(ip, XFS_ILOCK_EXCL);
506 error = xfs_bmapi_read(ip, start_zero_fsb, zero_count_fsb,
508 xfs_iunlock(ip, XFS_ILOCK_EXCL);
514 if (imap.br_state == XFS_EXT_UNWRITTEN ||
515 imap.br_startblock == HOLESTARTBLOCK) {
516 start_zero_fsb = imap.br_startoff + imap.br_blockcount;
517 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
522 * There are blocks we need to zero.
524 zero_off = XFS_FSB_TO_B(mp, start_zero_fsb);
525 zero_len = XFS_FSB_TO_B(mp, imap.br_blockcount);
527 if ((zero_off + zero_len) > offset)
528 zero_len = offset - zero_off;
530 error = xfs_iozero(ip, zero_off, zero_len);
534 start_zero_fsb = imap.br_startoff + imap.br_blockcount;
535 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
542 * Common pre-write limit and setup checks.
544 * Called with the iolocked held either shared and exclusive according to
545 * @iolock, and returns with it held. Might upgrade the iolock to exclusive
546 * if called for a direct write beyond i_size.
549 xfs_file_aio_write_checks(
555 struct inode *inode = file->f_mapping->host;
556 struct xfs_inode *ip = XFS_I(inode);
560 error = generic_write_checks(file, pos, count, S_ISBLK(inode->i_mode));
565 * If the offset is beyond the size of the file, we need to zero any
566 * blocks that fall between the existing EOF and the start of this
567 * write. If zeroing is needed and we are currently holding the
568 * iolock shared, we need to update it to exclusive which implies
569 * having to redo all checks before.
571 if (*pos > i_size_read(inode)) {
572 if (*iolock == XFS_IOLOCK_SHARED) {
573 xfs_rw_iunlock(ip, *iolock);
574 *iolock = XFS_IOLOCK_EXCL;
575 xfs_rw_ilock(ip, *iolock);
578 error = -xfs_zero_eof(ip, *pos, i_size_read(inode));
584 * Updating the timestamps will grab the ilock again from
585 * xfs_fs_dirty_inode, so we have to call it after dropping the
586 * lock above. Eventually we should look into a way to avoid
587 * the pointless lock roundtrip.
589 if (likely(!(file->f_mode & FMODE_NOCMTIME)))
590 file_update_time(file);
593 * If we're writing the file then make sure to clear the setuid and
594 * setgid bits if the process is not being run by root. This keeps
595 * people from modifying setuid and setgid binaries.
597 return file_remove_suid(file);
601 * xfs_file_dio_aio_write - handle direct IO writes
603 * Lock the inode appropriately to prepare for and issue a direct IO write.
604 * By separating it from the buffered write path we remove all the tricky to
605 * follow locking changes and looping.
607 * If there are cached pages or we're extending the file, we need IOLOCK_EXCL
608 * until we're sure the bytes at the new EOF have been zeroed and/or the cached
609 * pages are flushed out.
611 * In most cases the direct IO writes will be done holding IOLOCK_SHARED
612 * allowing them to be done in parallel with reads and other direct IO writes.
613 * However, if the IO is not aligned to filesystem blocks, the direct IO layer
614 * needs to do sub-block zeroing and that requires serialisation against other
615 * direct IOs to the same block. In this case we need to serialise the
616 * submission of the unaligned IOs so that we don't get racing block zeroing in
617 * the dio layer. To avoid the problem with aio, we also need to wait for
618 * outstanding IOs to complete so that unwritten extent conversion is completed
619 * before we try to map the overlapping block. This is currently implemented by
620 * hitting it with a big hammer (i.e. inode_dio_wait()).
622 * Returns with locks held indicated by @iolock and errors indicated by
623 * negative return values.
626 xfs_file_dio_aio_write(
628 const struct iovec *iovp,
629 unsigned long nr_segs,
633 struct file *file = iocb->ki_filp;
634 struct address_space *mapping = file->f_mapping;
635 struct inode *inode = mapping->host;
636 struct xfs_inode *ip = XFS_I(inode);
637 struct xfs_mount *mp = ip->i_mount;
639 size_t count = ocount;
640 int unaligned_io = 0;
642 struct xfs_buftarg *target = XFS_IS_REALTIME_INODE(ip) ?
643 mp->m_rtdev_targp : mp->m_ddev_targp;
645 if ((pos & target->bt_smask) || (count & target->bt_smask))
646 return -XFS_ERROR(EINVAL);
648 if ((pos & mp->m_blockmask) || ((pos + count) & mp->m_blockmask))
652 * We don't need to take an exclusive lock unless there page cache needs
653 * to be invalidated or unaligned IO is being executed. We don't need to
654 * consider the EOF extension case here because
655 * xfs_file_aio_write_checks() will relock the inode as necessary for
656 * EOF zeroing cases and fill out the new inode size as appropriate.
658 if (unaligned_io || mapping->nrpages)
659 iolock = XFS_IOLOCK_EXCL;
661 iolock = XFS_IOLOCK_SHARED;
662 xfs_rw_ilock(ip, iolock);
665 * Recheck if there are cached pages that need invalidate after we got
666 * the iolock to protect against other threads adding new pages while
667 * we were waiting for the iolock.
669 if (mapping->nrpages && iolock == XFS_IOLOCK_SHARED) {
670 xfs_rw_iunlock(ip, iolock);
671 iolock = XFS_IOLOCK_EXCL;
672 xfs_rw_ilock(ip, iolock);
675 ret = xfs_file_aio_write_checks(file, &pos, &count, &iolock);
679 if (mapping->nrpages) {
680 ret = -xfs_flushinval_pages(ip, (pos & PAGE_CACHE_MASK), -1,
687 * If we are doing unaligned IO, wait for all other IO to drain,
688 * otherwise demote the lock if we had to flush cached pages
691 inode_dio_wait(inode);
692 else if (iolock == XFS_IOLOCK_EXCL) {
693 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
694 iolock = XFS_IOLOCK_SHARED;
697 trace_xfs_file_direct_write(ip, count, iocb->ki_pos, 0);
698 ret = generic_file_direct_write(iocb, iovp,
699 &nr_segs, pos, &iocb->ki_pos, count, ocount);
702 xfs_rw_iunlock(ip, iolock);
704 /* No fallback to buffered IO on errors for XFS. */
705 ASSERT(ret < 0 || ret == count);
710 xfs_file_buffered_aio_write(
712 const struct iovec *iovp,
713 unsigned long nr_segs,
717 struct file *file = iocb->ki_filp;
718 struct address_space *mapping = file->f_mapping;
719 struct inode *inode = mapping->host;
720 struct xfs_inode *ip = XFS_I(inode);
723 int iolock = XFS_IOLOCK_EXCL;
724 size_t count = ocount;
726 xfs_rw_ilock(ip, iolock);
728 ret = xfs_file_aio_write_checks(file, &pos, &count, &iolock);
732 /* We can write back this queue in page reclaim */
733 current->backing_dev_info = mapping->backing_dev_info;
736 trace_xfs_file_buffered_write(ip, count, iocb->ki_pos, 0);
737 ret = generic_file_buffered_write(iocb, iovp, nr_segs,
738 pos, &iocb->ki_pos, count, ret);
740 * if we just got an ENOSPC, flush the inode now we aren't holding any
741 * page locks and retry *once*
743 if (ret == -ENOSPC && !enospc) {
745 ret = -xfs_flush_pages(ip, 0, -1, 0, FI_NONE);
750 current->backing_dev_info = NULL;
752 xfs_rw_iunlock(ip, iolock);
759 const struct iovec *iovp,
760 unsigned long nr_segs,
763 struct file *file = iocb->ki_filp;
764 struct address_space *mapping = file->f_mapping;
765 struct inode *inode = mapping->host;
766 struct xfs_inode *ip = XFS_I(inode);
770 XFS_STATS_INC(xs_write_calls);
772 BUG_ON(iocb->ki_pos != pos);
774 ret = generic_segment_checks(iovp, &nr_segs, &ocount, VERIFY_READ);
781 xfs_wait_for_freeze(ip->i_mount, SB_FREEZE_WRITE);
783 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
786 if (unlikely(file->f_flags & O_DIRECT))
787 ret = xfs_file_dio_aio_write(iocb, iovp, nr_segs, pos, ocount);
789 ret = xfs_file_buffered_aio_write(iocb, iovp, nr_segs, pos,
795 XFS_STATS_ADD(xs_write_bytes, ret);
797 /* Handle various SYNC-type writes */
798 err = generic_write_sync(file, pos, ret);
813 struct inode *inode = file->f_path.dentry->d_inode;
817 xfs_inode_t *ip = XFS_I(inode);
818 int cmd = XFS_IOC_RESVSP;
819 int attr_flags = XFS_ATTR_NOLOCK;
821 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
828 xfs_ilock(ip, XFS_IOLOCK_EXCL);
830 if (mode & FALLOC_FL_PUNCH_HOLE)
831 cmd = XFS_IOC_UNRESVSP;
833 /* check the new inode size is valid before allocating */
834 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
835 offset + len > i_size_read(inode)) {
836 new_size = offset + len;
837 error = inode_newsize_ok(inode, new_size);
842 if (file->f_flags & O_DSYNC)
843 attr_flags |= XFS_ATTR_SYNC;
845 error = -xfs_change_file_space(ip, cmd, &bf, 0, attr_flags);
849 /* Change file size if needed */
853 iattr.ia_valid = ATTR_SIZE;
854 iattr.ia_size = new_size;
855 error = -xfs_setattr_size(ip, &iattr, XFS_ATTR_NOLOCK);
859 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
869 if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
871 if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb)))
881 struct xfs_inode *ip = XFS_I(inode);
885 error = xfs_file_open(inode, file);
890 * If there are any blocks, read-ahead block 0 as we're almost
891 * certain to have the next operation be a read there.
893 mode = xfs_ilock_map_shared(ip);
894 if (ip->i_d.di_nextents > 0)
895 xfs_da_reada_buf(NULL, ip, 0, XFS_DATA_FORK);
896 xfs_iunlock(ip, mode);
905 return -xfs_release(XFS_I(inode));
914 struct inode *inode = filp->f_path.dentry->d_inode;
915 xfs_inode_t *ip = XFS_I(inode);
920 * The Linux API doesn't pass down the total size of the buffer
921 * we read into down to the filesystem. With the filldir concept
922 * it's not needed for correct information, but the XFS dir2 leaf
923 * code wants an estimate of the buffer size to calculate it's
924 * readahead window and size the buffers used for mapping to
927 * Try to give it an estimate that's good enough, maybe at some
928 * point we can change the ->readdir prototype to include the
929 * buffer size. For now we use the current glibc buffer size.
931 bufsize = (size_t)min_t(loff_t, 32768, ip->i_d.di_size);
933 error = xfs_readdir(ip, dirent, bufsize,
934 (xfs_off_t *)&filp->f_pos, filldir);
943 struct vm_area_struct *vma)
945 vma->vm_ops = &xfs_file_vm_ops;
946 vma->vm_flags |= VM_CAN_NONLINEAR;
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);
972 struct inode *inode = file->f_mapping->host;
973 struct xfs_inode *ip = XFS_I(inode);
974 struct xfs_mount *mp = ip->i_mount;
975 struct xfs_bmbt_irec map[2];
977 loff_t uninitialized_var(offset);
984 lock = xfs_ilock_map_shared(ip);
986 isize = i_size_read(inode);
987 if (start >= isize) {
992 fsbno = XFS_B_TO_FSBT(mp, start);
995 * Try to read extents from the first block indicated
996 * by fsbno to the end block of the file.
998 end = XFS_B_TO_FSB(mp, isize);
1000 error = xfs_bmapi_read(ip, fsbno, end - fsbno, map, &nmap,
1006 * Treat unwritten extent as data extent since it might
1007 * contains dirty data in page cache.
1009 if (map[0].br_startblock != HOLESTARTBLOCK) {
1010 offset = max_t(loff_t, start,
1011 XFS_FSB_TO_B(mp, map[0].br_startoff));
1018 offset = max_t(loff_t, start,
1019 XFS_FSB_TO_B(mp, map[1].br_startoff));
1022 if (offset != file->f_pos)
1023 file->f_pos = offset;
1026 xfs_iunlock_map_shared(ip, lock);
1039 struct inode *inode = file->f_mapping->host;
1040 struct xfs_inode *ip = XFS_I(inode);
1041 struct xfs_mount *mp = ip->i_mount;
1042 loff_t uninitialized_var(offset);
1045 xfs_fileoff_t fsbno;
1049 if (XFS_FORCED_SHUTDOWN(mp))
1050 return -XFS_ERROR(EIO);
1052 lock = xfs_ilock_map_shared(ip);
1054 isize = i_size_read(inode);
1055 if (start >= isize) {
1060 fsbno = XFS_B_TO_FSBT(mp, start);
1061 error = xfs_bmap_first_unused(NULL, ip, 1, &fsbno, XFS_DATA_FORK);
1065 holeoff = XFS_FSB_TO_B(mp, fsbno);
1066 if (holeoff <= start)
1070 * xfs_bmap_first_unused() could return a value bigger than
1071 * isize if there are no more holes past the supplied offset.
1073 offset = min_t(loff_t, holeoff, isize);
1076 if (offset != file->f_pos)
1077 file->f_pos = offset;
1080 xfs_iunlock_map_shared(ip, lock);
1097 return generic_file_llseek(file, offset, origin);
1099 return xfs_seek_data(file, offset, origin);
1101 return xfs_seek_hole(file, offset, origin);
1107 const struct file_operations xfs_file_operations = {
1108 .llseek = xfs_file_llseek,
1109 .read = do_sync_read,
1110 .write = do_sync_write,
1111 .aio_read = xfs_file_aio_read,
1112 .aio_write = xfs_file_aio_write,
1113 .splice_read = xfs_file_splice_read,
1114 .splice_write = xfs_file_splice_write,
1115 .unlocked_ioctl = xfs_file_ioctl,
1116 #ifdef CONFIG_COMPAT
1117 .compat_ioctl = xfs_file_compat_ioctl,
1119 .mmap = xfs_file_mmap,
1120 .open = xfs_file_open,
1121 .release = xfs_file_release,
1122 .fsync = xfs_file_fsync,
1123 .fallocate = xfs_file_fallocate,
1126 const struct file_operations xfs_dir_file_operations = {
1127 .open = xfs_dir_open,
1128 .read = generic_read_dir,
1129 .readdir = xfs_file_readdir,
1130 .llseek = generic_file_llseek,
1131 .unlocked_ioctl = xfs_file_ioctl,
1132 #ifdef CONFIG_COMPAT
1133 .compat_ioctl = xfs_file_compat_ioctl,
1135 .fsync = xfs_dir_fsync,
1138 static const struct vm_operations_struct xfs_file_vm_ops = {
1139 .fault = filemap_fault,
1140 .page_mkwrite = xfs_vm_page_mkwrite,