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! */
134 struct inode *inode = file->f_mapping->host;
135 struct xfs_inode *ip = XFS_I(inode);
136 struct xfs_mount *mp = ip->i_mount;
137 struct xfs_trans *tp;
141 trace_xfs_file_fsync(ip);
143 error = filemap_write_and_wait_range(inode->i_mapping, start, end);
147 if (XFS_FORCED_SHUTDOWN(mp))
148 return -XFS_ERROR(EIO);
150 xfs_iflags_clear(ip, XFS_ITRUNCATED);
152 if (mp->m_flags & XFS_MOUNT_BARRIER) {
154 * If we have an RT and/or log subvolume we need to make sure
155 * to flush the write cache the device used for file data
156 * first. This is to ensure newly written file data make
157 * it to disk before logging the new inode size in case of
158 * an extending write.
160 if (XFS_IS_REALTIME_INODE(ip))
161 xfs_blkdev_issue_flush(mp->m_rtdev_targp);
162 else if (mp->m_logdev_targp != mp->m_ddev_targp)
163 xfs_blkdev_issue_flush(mp->m_ddev_targp);
167 * We always need to make sure that the required inode state is safe on
168 * disk. The inode might be clean but we still might need to force the
169 * log because of committed transactions that haven't hit the disk yet.
170 * Likewise, there could be unflushed non-transactional changes to the
171 * inode core that have to go to disk and this requires us to issue
172 * a synchronous transaction to capture these changes correctly.
174 * This code relies on the assumption that if the i_update_core field
175 * of the inode is clear and the inode is unpinned then it is clean
176 * and no action is required.
178 xfs_ilock(ip, XFS_ILOCK_SHARED);
181 * First check if the VFS inode is marked dirty. All the dirtying
182 * of non-transactional updates no goes through mark_inode_dirty*,
183 * which allows us to distinguish beteeen pure timestamp updates
184 * and i_size updates which need to be caught for fdatasync.
185 * After that also theck for the dirty state in the XFS inode, which
186 * might gets cleared when the inode gets written out via the AIL
187 * or xfs_iflush_cluster.
189 if (((inode->i_state & I_DIRTY_DATASYNC) ||
190 ((inode->i_state & I_DIRTY_SYNC) && !datasync)) &&
193 * Kick off a transaction to log the inode core to get the
194 * updates. The sync transaction will also force the log.
196 xfs_iunlock(ip, XFS_ILOCK_SHARED);
197 tp = xfs_trans_alloc(mp, XFS_TRANS_FSYNC_TS);
198 error = xfs_trans_reserve(tp, 0,
199 XFS_FSYNC_TS_LOG_RES(mp), 0, 0, 0);
201 xfs_trans_cancel(tp, 0);
204 xfs_ilock(ip, XFS_ILOCK_EXCL);
207 * Note - it's possible that we might have pushed ourselves out
208 * of the way during trans_reserve which would flush the inode.
209 * But there's no guarantee that the inode buffer has actually
210 * gone out yet (it's delwri). Plus the buffer could be pinned
211 * anyway if it's part of an inode in another recent
212 * transaction. So we play it safe and fire off the
213 * transaction anyway.
215 xfs_trans_ijoin(tp, ip);
216 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
217 xfs_trans_set_sync(tp);
218 error = _xfs_trans_commit(tp, 0, &log_flushed);
220 xfs_iunlock(ip, XFS_ILOCK_EXCL);
223 * Timestamps/size haven't changed since last inode flush or
224 * inode transaction commit. That means either nothing got
225 * written or a transaction committed which caught the updates.
226 * If the latter happened and the transaction hasn't hit the
227 * disk yet, the inode will be still be pinned. If it is,
230 if (xfs_ipincount(ip)) {
231 error = _xfs_log_force_lsn(mp,
232 ip->i_itemp->ili_last_lsn,
233 XFS_LOG_SYNC, &log_flushed);
235 xfs_iunlock(ip, XFS_ILOCK_SHARED);
239 * If we only have a single device, and the log force about was
240 * a no-op we might have to flush the data device cache here.
241 * This can only happen for fdatasync/O_DSYNC if we were overwriting
242 * an already allocated file and thus do not have any metadata to
245 if ((mp->m_flags & XFS_MOUNT_BARRIER) &&
246 mp->m_logdev_targp == mp->m_ddev_targp &&
247 !XFS_IS_REALTIME_INODE(ip) &&
249 xfs_blkdev_issue_flush(mp->m_ddev_targp);
257 const struct iovec *iovp,
258 unsigned long nr_segs,
261 struct file *file = iocb->ki_filp;
262 struct inode *inode = file->f_mapping->host;
263 struct xfs_inode *ip = XFS_I(inode);
264 struct xfs_mount *mp = ip->i_mount;
271 XFS_STATS_INC(xs_read_calls);
273 BUG_ON(iocb->ki_pos != pos);
275 if (unlikely(file->f_flags & O_DIRECT))
276 ioflags |= IO_ISDIRECT;
277 if (file->f_mode & FMODE_NOCMTIME)
280 /* START copy & waste from filemap.c */
281 for (seg = 0; seg < nr_segs; seg++) {
282 const struct iovec *iv = &iovp[seg];
285 * If any segment has a negative length, or the cumulative
286 * length ever wraps negative then return -EINVAL.
289 if (unlikely((ssize_t)(size|iv->iov_len) < 0))
290 return XFS_ERROR(-EINVAL);
292 /* END copy & waste from filemap.c */
294 if (unlikely(ioflags & IO_ISDIRECT)) {
295 xfs_buftarg_t *target =
296 XFS_IS_REALTIME_INODE(ip) ?
297 mp->m_rtdev_targp : mp->m_ddev_targp;
298 if ((iocb->ki_pos & target->bt_smask) ||
299 (size & target->bt_smask)) {
300 if (iocb->ki_pos == ip->i_size)
302 return -XFS_ERROR(EINVAL);
306 n = XFS_MAXIOFFSET(mp) - iocb->ki_pos;
307 if (n <= 0 || size == 0)
313 if (XFS_FORCED_SHUTDOWN(mp))
317 * Locking is a bit tricky here. If we take an exclusive lock
318 * for direct IO, we effectively serialise all new concurrent
319 * read IO to this file and block it behind IO that is currently in
320 * progress because IO in progress holds the IO lock shared. We only
321 * need to hold the lock exclusive to blow away the page cache, so
322 * only take lock exclusively if the page cache needs invalidation.
323 * This allows the normal direct IO case of no page cache pages to
324 * proceeed concurrently without serialisation.
326 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
327 if ((ioflags & IO_ISDIRECT) && inode->i_mapping->nrpages) {
328 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
329 xfs_rw_ilock(ip, XFS_IOLOCK_EXCL);
331 if (inode->i_mapping->nrpages) {
332 ret = -xfs_flushinval_pages(ip,
333 (iocb->ki_pos & PAGE_CACHE_MASK),
334 -1, FI_REMAPF_LOCKED);
336 xfs_rw_iunlock(ip, XFS_IOLOCK_EXCL);
340 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
343 trace_xfs_file_read(ip, size, iocb->ki_pos, ioflags);
345 ret = generic_file_aio_read(iocb, iovp, nr_segs, iocb->ki_pos);
347 XFS_STATS_ADD(xs_read_bytes, ret);
349 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
354 xfs_file_splice_read(
357 struct pipe_inode_info *pipe,
361 struct xfs_inode *ip = XFS_I(infilp->f_mapping->host);
365 XFS_STATS_INC(xs_read_calls);
367 if (infilp->f_mode & FMODE_NOCMTIME)
370 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
373 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
375 trace_xfs_file_splice_read(ip, count, *ppos, ioflags);
377 ret = generic_file_splice_read(infilp, ppos, pipe, count, flags);
379 XFS_STATS_ADD(xs_read_bytes, ret);
381 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
386 xfs_aio_write_isize_update(
389 ssize_t bytes_written)
391 struct xfs_inode *ip = XFS_I(inode);
392 xfs_fsize_t isize = i_size_read(inode);
394 if (bytes_written > 0)
395 XFS_STATS_ADD(xs_write_bytes, bytes_written);
397 if (unlikely(bytes_written < 0 && bytes_written != -EFAULT &&
401 if (*ppos > ip->i_size) {
402 xfs_rw_ilock(ip, XFS_ILOCK_EXCL);
403 if (*ppos > ip->i_size)
405 xfs_rw_iunlock(ip, XFS_ILOCK_EXCL);
410 * If this was a direct or synchronous I/O that failed (such as ENOSPC) then
411 * part of the I/O may have been written to disk before the error occurred. In
412 * this case the on-disk file size may have been adjusted beyond the in-memory
413 * file size and now needs to be truncated back.
416 xfs_aio_write_newsize_update(
417 struct xfs_inode *ip,
418 xfs_fsize_t new_size)
420 if (new_size == ip->i_new_size) {
421 xfs_rw_ilock(ip, XFS_ILOCK_EXCL);
422 if (new_size == ip->i_new_size)
424 if (ip->i_d.di_size > ip->i_size)
425 ip->i_d.di_size = ip->i_size;
426 xfs_rw_iunlock(ip, XFS_ILOCK_EXCL);
431 * xfs_file_splice_write() does not use xfs_rw_ilock() because
432 * generic_file_splice_write() takes the i_mutex itself. This, in theory,
433 * couuld cause lock inversions between the aio_write path and the splice path
434 * if someone is doing concurrent splice(2) based writes and write(2) based
435 * writes to the same inode. The only real way to fix this is to re-implement
436 * the generic code here with correct locking orders.
439 xfs_file_splice_write(
440 struct pipe_inode_info *pipe,
441 struct file *outfilp,
446 struct inode *inode = outfilp->f_mapping->host;
447 struct xfs_inode *ip = XFS_I(inode);
448 xfs_fsize_t new_size;
452 XFS_STATS_INC(xs_write_calls);
454 if (outfilp->f_mode & FMODE_NOCMTIME)
457 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
460 xfs_ilock(ip, XFS_IOLOCK_EXCL);
462 new_size = *ppos + count;
464 xfs_ilock(ip, XFS_ILOCK_EXCL);
465 if (new_size > ip->i_size)
466 ip->i_new_size = new_size;
467 xfs_iunlock(ip, XFS_ILOCK_EXCL);
469 trace_xfs_file_splice_write(ip, count, *ppos, ioflags);
471 ret = generic_file_splice_write(pipe, outfilp, ppos, count, flags);
473 xfs_aio_write_isize_update(inode, ppos, ret);
474 xfs_aio_write_newsize_update(ip, new_size);
475 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
480 * This routine is called to handle zeroing any space in the last
481 * block of the file that is beyond the EOF. We do this since the
482 * size is being increased without writing anything to that block
483 * and we don't want anyone to read the garbage on the disk.
485 STATIC int /* error (positive) */
491 xfs_fileoff_t last_fsb;
492 xfs_mount_t *mp = ip->i_mount;
497 xfs_bmbt_irec_t imap;
499 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
501 zero_offset = XFS_B_FSB_OFFSET(mp, isize);
502 if (zero_offset == 0) {
504 * There are no extra bytes in the last block on disk to
510 last_fsb = XFS_B_TO_FSBT(mp, isize);
512 error = xfs_bmapi(NULL, ip, last_fsb, 1, 0, NULL, 0, &imap,
519 * If the block underlying isize is just a hole, then there
520 * is nothing to zero.
522 if (imap.br_startblock == HOLESTARTBLOCK) {
526 * Zero the part of the last block beyond the EOF, and write it
527 * out sync. We need to drop the ilock while we do this so we
528 * don't deadlock when the buffer cache calls back to us.
530 xfs_iunlock(ip, XFS_ILOCK_EXCL);
532 zero_len = mp->m_sb.sb_blocksize - zero_offset;
533 if (isize + zero_len > offset)
534 zero_len = offset - isize;
535 error = xfs_iozero(ip, isize, zero_len);
537 xfs_ilock(ip, XFS_ILOCK_EXCL);
543 * Zero any on disk space between the current EOF and the new,
544 * larger EOF. This handles the normal case of zeroing the remainder
545 * of the last block in the file and the unusual case of zeroing blocks
546 * out beyond the size of the file. This second case only happens
547 * with fixed size extents and when the system crashes before the inode
548 * size was updated but after blocks were allocated. If fill is set,
549 * then any holes in the range are filled and zeroed. If not, the holes
550 * are left alone as holes.
553 int /* error (positive) */
556 xfs_off_t offset, /* starting I/O offset */
557 xfs_fsize_t isize) /* current inode size */
559 xfs_mount_t *mp = ip->i_mount;
560 xfs_fileoff_t start_zero_fsb;
561 xfs_fileoff_t end_zero_fsb;
562 xfs_fileoff_t zero_count_fsb;
563 xfs_fileoff_t last_fsb;
564 xfs_fileoff_t zero_off;
565 xfs_fsize_t zero_len;
568 xfs_bmbt_irec_t imap;
570 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_IOLOCK_EXCL));
571 ASSERT(offset > isize);
574 * First handle zeroing the block on which isize resides.
575 * We only zero a part of that block so it is handled specially.
577 error = xfs_zero_last_block(ip, offset, isize);
579 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_IOLOCK_EXCL));
584 * Calculate the range between the new size and the old
585 * where blocks needing to be zeroed may exist. To get the
586 * block where the last byte in the file currently resides,
587 * we need to subtract one from the size and truncate back
588 * to a block boundary. We subtract 1 in case the size is
589 * exactly on a block boundary.
591 last_fsb = isize ? XFS_B_TO_FSBT(mp, isize - 1) : (xfs_fileoff_t)-1;
592 start_zero_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
593 end_zero_fsb = XFS_B_TO_FSBT(mp, offset - 1);
594 ASSERT((xfs_sfiloff_t)last_fsb < (xfs_sfiloff_t)start_zero_fsb);
595 if (last_fsb == end_zero_fsb) {
597 * The size was only incremented on its last block.
598 * We took care of that above, so just return.
603 ASSERT(start_zero_fsb <= end_zero_fsb);
604 while (start_zero_fsb <= end_zero_fsb) {
606 zero_count_fsb = end_zero_fsb - start_zero_fsb + 1;
607 error = xfs_bmapi(NULL, ip, start_zero_fsb, zero_count_fsb,
608 0, NULL, 0, &imap, &nimaps, NULL);
610 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_IOLOCK_EXCL));
615 if (imap.br_state == XFS_EXT_UNWRITTEN ||
616 imap.br_startblock == HOLESTARTBLOCK) {
618 * This loop handles initializing pages that were
619 * partially initialized by the code below this
620 * loop. It basically zeroes the part of the page
621 * that sits on a hole and sets the page as P_HOLE
622 * and calls remapf if it is a mapped file.
624 start_zero_fsb = imap.br_startoff + imap.br_blockcount;
625 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
630 * There are blocks we need to zero.
631 * Drop the inode lock while we're doing the I/O.
632 * We'll still have the iolock to protect us.
634 xfs_iunlock(ip, XFS_ILOCK_EXCL);
636 zero_off = XFS_FSB_TO_B(mp, start_zero_fsb);
637 zero_len = XFS_FSB_TO_B(mp, imap.br_blockcount);
639 if ((zero_off + zero_len) > offset)
640 zero_len = offset - zero_off;
642 error = xfs_iozero(ip, zero_off, zero_len);
647 start_zero_fsb = imap.br_startoff + imap.br_blockcount;
648 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
650 xfs_ilock(ip, XFS_ILOCK_EXCL);
656 xfs_ilock(ip, XFS_ILOCK_EXCL);
662 * Common pre-write limit and setup checks.
664 * Returns with iolock held according to @iolock.
667 xfs_file_aio_write_checks(
671 xfs_fsize_t *new_sizep,
674 struct inode *inode = file->f_mapping->host;
675 struct xfs_inode *ip = XFS_I(inode);
676 xfs_fsize_t new_size;
679 xfs_rw_ilock(ip, XFS_ILOCK_EXCL);
682 error = generic_write_checks(file, pos, count, S_ISBLK(inode->i_mode));
684 xfs_rw_iunlock(ip, XFS_ILOCK_EXCL | *iolock);
689 if (likely(!(file->f_mode & FMODE_NOCMTIME)))
690 file_update_time(file);
693 * If the offset is beyond the size of the file, we need to zero any
694 * blocks that fall between the existing EOF and the start of this
695 * write. There is no need to issue zeroing if another in-flght IO ends
696 * at or before this one If zeronig is needed and we are currently
697 * holding the iolock shared, we need to update it to exclusive which
698 * involves dropping all locks and relocking to maintain correct locking
699 * order. If we do this, restart the function to ensure all checks and
700 * values are still valid.
702 if ((ip->i_new_size && *pos > ip->i_new_size) ||
703 (!ip->i_new_size && *pos > ip->i_size)) {
704 if (*iolock == XFS_IOLOCK_SHARED) {
705 xfs_rw_iunlock(ip, XFS_ILOCK_EXCL | *iolock);
706 *iolock = XFS_IOLOCK_EXCL;
707 xfs_rw_ilock(ip, XFS_ILOCK_EXCL | *iolock);
710 error = -xfs_zero_eof(ip, *pos, ip->i_size);
714 * If this IO extends beyond EOF, we may need to update ip->i_new_size.
715 * We have already zeroed space beyond EOF (if necessary). Only update
716 * ip->i_new_size if this IO ends beyond any other in-flight writes.
718 new_size = *pos + *count;
719 if (new_size > ip->i_size) {
720 if (new_size > ip->i_new_size)
721 ip->i_new_size = new_size;
722 *new_sizep = new_size;
725 xfs_rw_iunlock(ip, XFS_ILOCK_EXCL);
730 * If we're writing the file then make sure to clear the setuid and
731 * setgid bits if the process is not being run by root. This keeps
732 * people from modifying setuid and setgid binaries.
734 return file_remove_suid(file);
739 * xfs_file_dio_aio_write - handle direct IO writes
741 * Lock the inode appropriately to prepare for and issue a direct IO write.
742 * By separating it from the buffered write path we remove all the tricky to
743 * follow locking changes and looping.
745 * If there are cached pages or we're extending the file, we need IOLOCK_EXCL
746 * until we're sure the bytes at the new EOF have been zeroed and/or the cached
747 * pages are flushed out.
749 * In most cases the direct IO writes will be done holding IOLOCK_SHARED
750 * allowing them to be done in parallel with reads and other direct IO writes.
751 * However, if the IO is not aligned to filesystem blocks, the direct IO layer
752 * needs to do sub-block zeroing and that requires serialisation against other
753 * direct IOs to the same block. In this case we need to serialise the
754 * submission of the unaligned IOs so that we don't get racing block zeroing in
755 * the dio layer. To avoid the problem with aio, we also need to wait for
756 * outstanding IOs to complete so that unwritten extent conversion is completed
757 * before we try to map the overlapping block. This is currently implemented by
758 * hitting it with a big hammer (i.e. inode_dio_wait()).
760 * Returns with locks held indicated by @iolock and errors indicated by
761 * negative return values.
764 xfs_file_dio_aio_write(
766 const struct iovec *iovp,
767 unsigned long nr_segs,
770 xfs_fsize_t *new_size,
773 struct file *file = iocb->ki_filp;
774 struct address_space *mapping = file->f_mapping;
775 struct inode *inode = mapping->host;
776 struct xfs_inode *ip = XFS_I(inode);
777 struct xfs_mount *mp = ip->i_mount;
779 size_t count = ocount;
780 int unaligned_io = 0;
781 struct xfs_buftarg *target = XFS_IS_REALTIME_INODE(ip) ?
782 mp->m_rtdev_targp : mp->m_ddev_targp;
785 if ((pos & target->bt_smask) || (count & target->bt_smask))
786 return -XFS_ERROR(EINVAL);
788 if ((pos & mp->m_blockmask) || ((pos + count) & mp->m_blockmask))
792 * We don't need to take an exclusive lock unless there page cache needs
793 * to be invalidated or unaligned IO is being executed. We don't need to
794 * consider the EOF extension case here because
795 * xfs_file_aio_write_checks() will relock the inode as necessary for
796 * EOF zeroing cases and fill out the new inode size as appropriate.
798 if (unaligned_io || mapping->nrpages)
799 *iolock = XFS_IOLOCK_EXCL;
801 *iolock = XFS_IOLOCK_SHARED;
802 xfs_rw_ilock(ip, *iolock);
805 * Recheck if there are cached pages that need invalidate after we got
806 * the iolock to protect against other threads adding new pages while
807 * we were waiting for the iolock.
809 if (mapping->nrpages && *iolock == XFS_IOLOCK_SHARED) {
810 xfs_rw_iunlock(ip, *iolock);
811 *iolock = XFS_IOLOCK_EXCL;
812 xfs_rw_ilock(ip, *iolock);
815 ret = xfs_file_aio_write_checks(file, &pos, &count, new_size, iolock);
819 if (mapping->nrpages) {
820 ret = -xfs_flushinval_pages(ip, (pos & PAGE_CACHE_MASK), -1,
827 * If we are doing unaligned IO, wait for all other IO to drain,
828 * otherwise demote the lock if we had to flush cached pages
831 inode_dio_wait(inode);
832 else if (*iolock == XFS_IOLOCK_EXCL) {
833 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
834 *iolock = XFS_IOLOCK_SHARED;
837 trace_xfs_file_direct_write(ip, count, iocb->ki_pos, 0);
838 ret = generic_file_direct_write(iocb, iovp,
839 &nr_segs, pos, &iocb->ki_pos, count, ocount);
841 /* No fallback to buffered IO on errors for XFS. */
842 ASSERT(ret < 0 || ret == count);
847 xfs_file_buffered_aio_write(
849 const struct iovec *iovp,
850 unsigned long nr_segs,
853 xfs_fsize_t *new_size,
856 struct file *file = iocb->ki_filp;
857 struct address_space *mapping = file->f_mapping;
858 struct inode *inode = mapping->host;
859 struct xfs_inode *ip = XFS_I(inode);
862 size_t count = ocount;
864 *iolock = XFS_IOLOCK_EXCL;
865 xfs_rw_ilock(ip, *iolock);
867 ret = xfs_file_aio_write_checks(file, &pos, &count, new_size, iolock);
871 /* We can write back this queue in page reclaim */
872 current->backing_dev_info = mapping->backing_dev_info;
875 trace_xfs_file_buffered_write(ip, count, iocb->ki_pos, 0);
876 ret = generic_file_buffered_write(iocb, iovp, nr_segs,
877 pos, &iocb->ki_pos, count, ret);
879 * if we just got an ENOSPC, flush the inode now we aren't holding any
880 * page locks and retry *once*
882 if (ret == -ENOSPC && !enospc) {
883 ret = -xfs_flush_pages(ip, 0, -1, 0, FI_NONE);
889 current->backing_dev_info = NULL;
896 const struct iovec *iovp,
897 unsigned long nr_segs,
900 struct file *file = iocb->ki_filp;
901 struct address_space *mapping = file->f_mapping;
902 struct inode *inode = mapping->host;
903 struct xfs_inode *ip = XFS_I(inode);
907 xfs_fsize_t new_size = 0;
909 XFS_STATS_INC(xs_write_calls);
911 BUG_ON(iocb->ki_pos != pos);
913 ret = generic_segment_checks(iovp, &nr_segs, &ocount, VERIFY_READ);
920 xfs_wait_for_freeze(ip->i_mount, SB_FREEZE_WRITE);
922 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
925 if (unlikely(file->f_flags & O_DIRECT))
926 ret = xfs_file_dio_aio_write(iocb, iovp, nr_segs, pos,
927 ocount, &new_size, &iolock);
929 ret = xfs_file_buffered_aio_write(iocb, iovp, nr_segs, pos,
930 ocount, &new_size, &iolock);
932 xfs_aio_write_isize_update(inode, &iocb->ki_pos, ret);
937 /* Handle various SYNC-type writes */
938 if ((file->f_flags & O_DSYNC) || IS_SYNC(inode)) {
939 loff_t end = pos + ret - 1;
942 xfs_rw_iunlock(ip, iolock);
943 error = xfs_file_fsync(file, pos, end,
944 (file->f_flags & __O_SYNC) ? 0 : 1);
945 xfs_rw_ilock(ip, iolock);
951 xfs_aio_write_newsize_update(ip, new_size);
952 xfs_rw_iunlock(ip, iolock);
963 struct inode *inode = file->f_path.dentry->d_inode;
967 xfs_inode_t *ip = XFS_I(inode);
968 int cmd = XFS_IOC_RESVSP;
969 int attr_flags = XFS_ATTR_NOLOCK;
971 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
978 xfs_ilock(ip, XFS_IOLOCK_EXCL);
980 if (mode & FALLOC_FL_PUNCH_HOLE)
981 cmd = XFS_IOC_UNRESVSP;
983 /* check the new inode size is valid before allocating */
984 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
985 offset + len > i_size_read(inode)) {
986 new_size = offset + len;
987 error = inode_newsize_ok(inode, new_size);
992 if (file->f_flags & O_DSYNC)
993 attr_flags |= XFS_ATTR_SYNC;
995 error = -xfs_change_file_space(ip, cmd, &bf, 0, attr_flags);
999 /* Change file size if needed */
1003 iattr.ia_valid = ATTR_SIZE;
1004 iattr.ia_size = new_size;
1005 error = -xfs_setattr_size(ip, &iattr, XFS_ATTR_NOLOCK);
1009 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1016 struct inode *inode,
1019 if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1021 if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb)))
1028 struct inode *inode,
1031 struct xfs_inode *ip = XFS_I(inode);
1035 error = xfs_file_open(inode, file);
1040 * If there are any blocks, read-ahead block 0 as we're almost
1041 * certain to have the next operation be a read there.
1043 mode = xfs_ilock_map_shared(ip);
1044 if (ip->i_d.di_nextents > 0)
1045 xfs_da_reada_buf(NULL, ip, 0, XFS_DATA_FORK);
1046 xfs_iunlock(ip, mode);
1052 struct inode *inode,
1055 return -xfs_release(XFS_I(inode));
1064 struct inode *inode = filp->f_path.dentry->d_inode;
1065 xfs_inode_t *ip = XFS_I(inode);
1070 * The Linux API doesn't pass down the total size of the buffer
1071 * we read into down to the filesystem. With the filldir concept
1072 * it's not needed for correct information, but the XFS dir2 leaf
1073 * code wants an estimate of the buffer size to calculate it's
1074 * readahead window and size the buffers used for mapping to
1077 * Try to give it an estimate that's good enough, maybe at some
1078 * point we can change the ->readdir prototype to include the
1079 * buffer size. For now we use the current glibc buffer size.
1081 bufsize = (size_t)min_t(loff_t, 32768, ip->i_d.di_size);
1083 error = xfs_readdir(ip, dirent, bufsize,
1084 (xfs_off_t *)&filp->f_pos, filldir);
1093 struct vm_area_struct *vma)
1095 vma->vm_ops = &xfs_file_vm_ops;
1096 vma->vm_flags |= VM_CAN_NONLINEAR;
1098 file_accessed(filp);
1103 * mmap()d file has taken write protection fault and is being made
1104 * writable. We can set the page state up correctly for a writable
1105 * page, which means we can do correct delalloc accounting (ENOSPC
1106 * checking!) and unwritten extent mapping.
1109 xfs_vm_page_mkwrite(
1110 struct vm_area_struct *vma,
1111 struct vm_fault *vmf)
1113 return block_page_mkwrite(vma, vmf, xfs_get_blocks);
1116 const struct file_operations xfs_file_operations = {
1117 .llseek = generic_file_llseek,
1118 .read = do_sync_read,
1119 .write = do_sync_write,
1120 .aio_read = xfs_file_aio_read,
1121 .aio_write = xfs_file_aio_write,
1122 .splice_read = xfs_file_splice_read,
1123 .splice_write = xfs_file_splice_write,
1124 .unlocked_ioctl = xfs_file_ioctl,
1125 #ifdef CONFIG_COMPAT
1126 .compat_ioctl = xfs_file_compat_ioctl,
1128 .mmap = xfs_file_mmap,
1129 .open = xfs_file_open,
1130 .release = xfs_file_release,
1131 .fsync = xfs_file_fsync,
1132 .fallocate = xfs_file_fallocate,
1135 const struct file_operations xfs_dir_file_operations = {
1136 .open = xfs_dir_open,
1137 .read = generic_read_dir,
1138 .readdir = xfs_file_readdir,
1139 .llseek = generic_file_llseek,
1140 .unlocked_ioctl = xfs_file_ioctl,
1141 #ifdef CONFIG_COMPAT
1142 .compat_ioctl = xfs_file_compat_ioctl,
1144 .fsync = xfs_file_fsync,
1147 static const struct vm_operations_struct xfs_file_vm_ops = {
1148 .fault = filemap_fault,
1149 .page_mkwrite = xfs_vm_page_mkwrite,