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;
166 struct xfs_trans *tp;
171 trace_xfs_file_fsync(ip);
173 error = filemap_write_and_wait_range(inode->i_mapping, start, end);
177 if (XFS_FORCED_SHUTDOWN(mp))
178 return -XFS_ERROR(EIO);
180 xfs_iflags_clear(ip, XFS_ITRUNCATED);
182 if (mp->m_flags & XFS_MOUNT_BARRIER) {
184 * If we have an RT and/or log subvolume we need to make sure
185 * to flush the write cache the device used for file data
186 * first. This is to ensure newly written file data make
187 * it to disk before logging the new inode size in case of
188 * an extending write.
190 if (XFS_IS_REALTIME_INODE(ip))
191 xfs_blkdev_issue_flush(mp->m_rtdev_targp);
192 else if (mp->m_logdev_targp != mp->m_ddev_targp)
193 xfs_blkdev_issue_flush(mp->m_ddev_targp);
197 * We always need to make sure that the required inode state is safe on
198 * disk. The inode might be clean but we still might need to force the
199 * log because of committed transactions that haven't hit the disk yet.
200 * Likewise, there could be unflushed non-transactional changes to the
201 * inode core that have to go to disk and this requires us to issue
202 * a synchronous transaction to capture these changes correctly.
204 * This code relies on the assumption that if the i_update_core field
205 * of the inode is clear and the inode is unpinned then it is clean
206 * and no action is required.
208 xfs_ilock(ip, XFS_ILOCK_SHARED);
211 * First check if the VFS inode is marked dirty. All the dirtying
212 * of non-transactional updates do not go through mark_inode_dirty*,
213 * which allows us to distinguish between pure timestamp updates
214 * and i_size updates which need to be caught for fdatasync.
215 * After that also check for the dirty state in the XFS inode, which
216 * might gets cleared when the inode gets written out via the AIL
217 * or xfs_iflush_cluster.
219 if (((inode->i_state & I_DIRTY_DATASYNC) ||
220 ((inode->i_state & I_DIRTY_SYNC) && !datasync)) &&
223 * Kick off a transaction to log the inode core to get the
224 * updates. The sync transaction will also force the log.
226 xfs_iunlock(ip, XFS_ILOCK_SHARED);
227 tp = xfs_trans_alloc(mp, XFS_TRANS_FSYNC_TS);
228 error = xfs_trans_reserve(tp, 0,
229 XFS_FSYNC_TS_LOG_RES(mp), 0, 0, 0);
231 xfs_trans_cancel(tp, 0);
234 xfs_ilock(ip, XFS_ILOCK_EXCL);
237 * Note - it's possible that we might have pushed ourselves out
238 * of the way during trans_reserve which would flush the inode.
239 * But there's no guarantee that the inode buffer has actually
240 * gone out yet (it's delwri). Plus the buffer could be pinned
241 * anyway if it's part of an inode in another recent
242 * transaction. So we play it safe and fire off the
243 * transaction anyway.
245 xfs_trans_ijoin(tp, ip, 0);
246 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
247 error = xfs_trans_commit(tp, 0);
249 lsn = ip->i_itemp->ili_last_lsn;
250 xfs_iunlock(ip, XFS_ILOCK_EXCL);
253 * Timestamps/size haven't changed since last inode flush or
254 * inode transaction commit. That means either nothing got
255 * written or a transaction committed which caught the updates.
256 * If the latter happened and the transaction hasn't hit the
257 * disk yet, the inode will be still be pinned. If it is,
260 if (xfs_ipincount(ip))
261 lsn = ip->i_itemp->ili_last_lsn;
262 xfs_iunlock(ip, XFS_ILOCK_SHARED);
266 error = _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, &log_flushed);
269 * If we only have a single device, and the log force about was
270 * a no-op we might have to flush the data device cache here.
271 * This can only happen for fdatasync/O_DSYNC if we were overwriting
272 * an already allocated file and thus do not have any metadata to
275 if ((mp->m_flags & XFS_MOUNT_BARRIER) &&
276 mp->m_logdev_targp == mp->m_ddev_targp &&
277 !XFS_IS_REALTIME_INODE(ip) &&
279 xfs_blkdev_issue_flush(mp->m_ddev_targp);
287 const struct iovec *iovp,
288 unsigned long nr_segs,
291 struct file *file = iocb->ki_filp;
292 struct inode *inode = file->f_mapping->host;
293 struct xfs_inode *ip = XFS_I(inode);
294 struct xfs_mount *mp = ip->i_mount;
301 XFS_STATS_INC(xs_read_calls);
303 BUG_ON(iocb->ki_pos != pos);
305 if (unlikely(file->f_flags & O_DIRECT))
306 ioflags |= IO_ISDIRECT;
307 if (file->f_mode & FMODE_NOCMTIME)
310 /* START copy & waste from filemap.c */
311 for (seg = 0; seg < nr_segs; seg++) {
312 const struct iovec *iv = &iovp[seg];
315 * If any segment has a negative length, or the cumulative
316 * length ever wraps negative then return -EINVAL.
319 if (unlikely((ssize_t)(size|iv->iov_len) < 0))
320 return XFS_ERROR(-EINVAL);
322 /* END copy & waste from filemap.c */
324 if (unlikely(ioflags & IO_ISDIRECT)) {
325 xfs_buftarg_t *target =
326 XFS_IS_REALTIME_INODE(ip) ?
327 mp->m_rtdev_targp : mp->m_ddev_targp;
328 if ((iocb->ki_pos & target->bt_smask) ||
329 (size & target->bt_smask)) {
330 if (iocb->ki_pos == ip->i_size)
332 return -XFS_ERROR(EINVAL);
336 n = XFS_MAXIOFFSET(mp) - iocb->ki_pos;
337 if (n <= 0 || size == 0)
343 if (XFS_FORCED_SHUTDOWN(mp))
347 * Locking is a bit tricky here. If we take an exclusive lock
348 * for direct IO, we effectively serialise all new concurrent
349 * read IO to this file and block it behind IO that is currently in
350 * progress because IO in progress holds the IO lock shared. We only
351 * need to hold the lock exclusive to blow away the page cache, so
352 * only take lock exclusively if the page cache needs invalidation.
353 * This allows the normal direct IO case of no page cache pages to
354 * proceeed concurrently without serialisation.
356 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
357 if ((ioflags & IO_ISDIRECT) && inode->i_mapping->nrpages) {
358 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
359 xfs_rw_ilock(ip, XFS_IOLOCK_EXCL);
361 if (inode->i_mapping->nrpages) {
362 ret = -xfs_flushinval_pages(ip,
363 (iocb->ki_pos & PAGE_CACHE_MASK),
364 -1, FI_REMAPF_LOCKED);
366 xfs_rw_iunlock(ip, XFS_IOLOCK_EXCL);
370 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
373 trace_xfs_file_read(ip, size, iocb->ki_pos, ioflags);
375 ret = generic_file_aio_read(iocb, iovp, nr_segs, iocb->ki_pos);
377 XFS_STATS_ADD(xs_read_bytes, ret);
379 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
384 xfs_file_splice_read(
387 struct pipe_inode_info *pipe,
391 struct xfs_inode *ip = XFS_I(infilp->f_mapping->host);
395 XFS_STATS_INC(xs_read_calls);
397 if (infilp->f_mode & FMODE_NOCMTIME)
400 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
403 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
405 trace_xfs_file_splice_read(ip, count, *ppos, ioflags);
407 ret = generic_file_splice_read(infilp, ppos, pipe, count, flags);
409 XFS_STATS_ADD(xs_read_bytes, ret);
411 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
416 xfs_aio_write_isize_update(
419 ssize_t bytes_written)
421 struct xfs_inode *ip = XFS_I(inode);
422 xfs_fsize_t isize = i_size_read(inode);
424 if (bytes_written > 0)
425 XFS_STATS_ADD(xs_write_bytes, bytes_written);
427 if (unlikely(bytes_written < 0 && bytes_written != -EFAULT &&
431 if (*ppos > ip->i_size) {
432 xfs_rw_ilock(ip, XFS_ILOCK_EXCL);
433 if (*ppos > ip->i_size)
435 xfs_rw_iunlock(ip, XFS_ILOCK_EXCL);
440 * If this was a direct or synchronous I/O that failed (such as ENOSPC) then
441 * part of the I/O may have been written to disk before the error occurred. In
442 * this case the on-disk file size may have been adjusted beyond the in-memory
443 * file size and now needs to be truncated back.
446 xfs_aio_write_newsize_update(
447 struct xfs_inode *ip,
448 xfs_fsize_t new_size)
450 if (new_size == ip->i_new_size) {
451 xfs_rw_ilock(ip, XFS_ILOCK_EXCL);
452 if (new_size == ip->i_new_size)
454 if (ip->i_d.di_size > ip->i_size)
455 ip->i_d.di_size = ip->i_size;
456 xfs_rw_iunlock(ip, XFS_ILOCK_EXCL);
461 * xfs_file_splice_write() does not use xfs_rw_ilock() because
462 * generic_file_splice_write() takes the i_mutex itself. This, in theory,
463 * couuld cause lock inversions between the aio_write path and the splice path
464 * if someone is doing concurrent splice(2) based writes and write(2) based
465 * writes to the same inode. The only real way to fix this is to re-implement
466 * the generic code here with correct locking orders.
469 xfs_file_splice_write(
470 struct pipe_inode_info *pipe,
471 struct file *outfilp,
476 struct inode *inode = outfilp->f_mapping->host;
477 struct xfs_inode *ip = XFS_I(inode);
478 xfs_fsize_t new_size;
482 XFS_STATS_INC(xs_write_calls);
484 if (outfilp->f_mode & FMODE_NOCMTIME)
487 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
490 xfs_ilock(ip, XFS_IOLOCK_EXCL);
492 new_size = *ppos + count;
494 xfs_ilock(ip, XFS_ILOCK_EXCL);
495 if (new_size > ip->i_size)
496 ip->i_new_size = new_size;
497 xfs_iunlock(ip, XFS_ILOCK_EXCL);
499 trace_xfs_file_splice_write(ip, count, *ppos, ioflags);
501 ret = generic_file_splice_write(pipe, outfilp, ppos, count, flags);
503 xfs_aio_write_isize_update(inode, ppos, ret);
504 xfs_aio_write_newsize_update(ip, new_size);
505 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
510 * This routine is called to handle zeroing any space in the last
511 * block of the file that is beyond the EOF. We do this since the
512 * size is being increased without writing anything to that block
513 * and we don't want anyone to read the garbage on the disk.
515 STATIC int /* error (positive) */
521 xfs_fileoff_t last_fsb;
522 xfs_mount_t *mp = ip->i_mount;
527 xfs_bmbt_irec_t imap;
529 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
531 zero_offset = XFS_B_FSB_OFFSET(mp, isize);
532 if (zero_offset == 0) {
534 * There are no extra bytes in the last block on disk to
540 last_fsb = XFS_B_TO_FSBT(mp, isize);
542 error = xfs_bmapi_read(ip, last_fsb, 1, &imap, &nimaps, 0);
547 * If the block underlying isize is just a hole, then there
548 * is nothing to zero.
550 if (imap.br_startblock == HOLESTARTBLOCK) {
554 * Zero the part of the last block beyond the EOF, and write it
555 * out sync. We need to drop the ilock while we do this so we
556 * don't deadlock when the buffer cache calls back to us.
558 xfs_iunlock(ip, XFS_ILOCK_EXCL);
560 zero_len = mp->m_sb.sb_blocksize - zero_offset;
561 if (isize + zero_len > offset)
562 zero_len = offset - isize;
563 error = xfs_iozero(ip, isize, zero_len);
565 xfs_ilock(ip, XFS_ILOCK_EXCL);
571 * Zero any on disk space between the current EOF and the new,
572 * larger EOF. This handles the normal case of zeroing the remainder
573 * of the last block in the file and the unusual case of zeroing blocks
574 * out beyond the size of the file. This second case only happens
575 * with fixed size extents and when the system crashes before the inode
576 * size was updated but after blocks were allocated. If fill is set,
577 * then any holes in the range are filled and zeroed. If not, the holes
578 * are left alone as holes.
581 int /* error (positive) */
584 xfs_off_t offset, /* starting I/O offset */
585 xfs_fsize_t isize) /* current inode size */
587 xfs_mount_t *mp = ip->i_mount;
588 xfs_fileoff_t start_zero_fsb;
589 xfs_fileoff_t end_zero_fsb;
590 xfs_fileoff_t zero_count_fsb;
591 xfs_fileoff_t last_fsb;
592 xfs_fileoff_t zero_off;
593 xfs_fsize_t zero_len;
596 xfs_bmbt_irec_t imap;
598 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_IOLOCK_EXCL));
599 ASSERT(offset > isize);
602 * First handle zeroing the block on which isize resides.
603 * We only zero a part of that block so it is handled specially.
605 error = xfs_zero_last_block(ip, offset, isize);
607 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_IOLOCK_EXCL));
612 * Calculate the range between the new size and the old
613 * where blocks needing to be zeroed may exist. To get the
614 * block where the last byte in the file currently resides,
615 * we need to subtract one from the size and truncate back
616 * to a block boundary. We subtract 1 in case the size is
617 * exactly on a block boundary.
619 last_fsb = isize ? XFS_B_TO_FSBT(mp, isize - 1) : (xfs_fileoff_t)-1;
620 start_zero_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
621 end_zero_fsb = XFS_B_TO_FSBT(mp, offset - 1);
622 ASSERT((xfs_sfiloff_t)last_fsb < (xfs_sfiloff_t)start_zero_fsb);
623 if (last_fsb == end_zero_fsb) {
625 * The size was only incremented on its last block.
626 * We took care of that above, so just return.
631 ASSERT(start_zero_fsb <= end_zero_fsb);
632 while (start_zero_fsb <= end_zero_fsb) {
634 zero_count_fsb = end_zero_fsb - start_zero_fsb + 1;
635 error = xfs_bmapi_read(ip, start_zero_fsb, zero_count_fsb,
638 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_IOLOCK_EXCL));
643 if (imap.br_state == XFS_EXT_UNWRITTEN ||
644 imap.br_startblock == HOLESTARTBLOCK) {
646 * This loop handles initializing pages that were
647 * partially initialized by the code below this
648 * loop. It basically zeroes the part of the page
649 * that sits on a hole and sets the page as P_HOLE
650 * and calls remapf if it is a mapped file.
652 start_zero_fsb = imap.br_startoff + imap.br_blockcount;
653 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
658 * There are blocks we need to zero.
659 * Drop the inode lock while we're doing the I/O.
660 * We'll still have the iolock to protect us.
662 xfs_iunlock(ip, XFS_ILOCK_EXCL);
664 zero_off = XFS_FSB_TO_B(mp, start_zero_fsb);
665 zero_len = XFS_FSB_TO_B(mp, imap.br_blockcount);
667 if ((zero_off + zero_len) > offset)
668 zero_len = offset - zero_off;
670 error = xfs_iozero(ip, zero_off, zero_len);
675 start_zero_fsb = imap.br_startoff + imap.br_blockcount;
676 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
678 xfs_ilock(ip, XFS_ILOCK_EXCL);
684 xfs_ilock(ip, XFS_ILOCK_EXCL);
690 * Common pre-write limit and setup checks.
692 * Returns with iolock held according to @iolock.
695 xfs_file_aio_write_checks(
699 xfs_fsize_t *new_sizep,
702 struct inode *inode = file->f_mapping->host;
703 struct xfs_inode *ip = XFS_I(inode);
704 xfs_fsize_t new_size;
707 xfs_rw_ilock(ip, XFS_ILOCK_EXCL);
710 error = generic_write_checks(file, pos, count, S_ISBLK(inode->i_mode));
712 xfs_rw_iunlock(ip, XFS_ILOCK_EXCL | *iolock);
717 if (likely(!(file->f_mode & FMODE_NOCMTIME)))
718 file_update_time(file);
721 * If the offset is beyond the size of the file, we need to zero any
722 * blocks that fall between the existing EOF and the start of this
723 * write. There is no need to issue zeroing if another in-flght IO ends
724 * at or before this one If zeronig is needed and we are currently
725 * holding the iolock shared, we need to update it to exclusive which
726 * involves dropping all locks and relocking to maintain correct locking
727 * order. If we do this, restart the function to ensure all checks and
728 * values are still valid.
730 if ((ip->i_new_size && *pos > ip->i_new_size) ||
731 (!ip->i_new_size && *pos > ip->i_size)) {
732 if (*iolock == XFS_IOLOCK_SHARED) {
733 xfs_rw_iunlock(ip, XFS_ILOCK_EXCL | *iolock);
734 *iolock = XFS_IOLOCK_EXCL;
735 xfs_rw_ilock(ip, XFS_ILOCK_EXCL | *iolock);
738 error = -xfs_zero_eof(ip, *pos, ip->i_size);
742 * If this IO extends beyond EOF, we may need to update ip->i_new_size.
743 * We have already zeroed space beyond EOF (if necessary). Only update
744 * ip->i_new_size if this IO ends beyond any other in-flight writes.
746 new_size = *pos + *count;
747 if (new_size > ip->i_size) {
748 if (new_size > ip->i_new_size)
749 ip->i_new_size = new_size;
750 *new_sizep = new_size;
753 xfs_rw_iunlock(ip, XFS_ILOCK_EXCL);
758 * If we're writing the file then make sure to clear the setuid and
759 * setgid bits if the process is not being run by root. This keeps
760 * people from modifying setuid and setgid binaries.
762 return file_remove_suid(file);
767 * xfs_file_dio_aio_write - handle direct IO writes
769 * Lock the inode appropriately to prepare for and issue a direct IO write.
770 * By separating it from the buffered write path we remove all the tricky to
771 * follow locking changes and looping.
773 * If there are cached pages or we're extending the file, we need IOLOCK_EXCL
774 * until we're sure the bytes at the new EOF have been zeroed and/or the cached
775 * pages are flushed out.
777 * In most cases the direct IO writes will be done holding IOLOCK_SHARED
778 * allowing them to be done in parallel with reads and other direct IO writes.
779 * However, if the IO is not aligned to filesystem blocks, the direct IO layer
780 * needs to do sub-block zeroing and that requires serialisation against other
781 * direct IOs to the same block. In this case we need to serialise the
782 * submission of the unaligned IOs so that we don't get racing block zeroing in
783 * the dio layer. To avoid the problem with aio, we also need to wait for
784 * outstanding IOs to complete so that unwritten extent conversion is completed
785 * before we try to map the overlapping block. This is currently implemented by
786 * hitting it with a big hammer (i.e. inode_dio_wait()).
788 * Returns with locks held indicated by @iolock and errors indicated by
789 * negative return values.
792 xfs_file_dio_aio_write(
794 const struct iovec *iovp,
795 unsigned long nr_segs,
798 xfs_fsize_t *new_size,
801 struct file *file = iocb->ki_filp;
802 struct address_space *mapping = file->f_mapping;
803 struct inode *inode = mapping->host;
804 struct xfs_inode *ip = XFS_I(inode);
805 struct xfs_mount *mp = ip->i_mount;
807 size_t count = ocount;
808 int unaligned_io = 0;
809 struct xfs_buftarg *target = XFS_IS_REALTIME_INODE(ip) ?
810 mp->m_rtdev_targp : mp->m_ddev_targp;
813 if ((pos & target->bt_smask) || (count & target->bt_smask))
814 return -XFS_ERROR(EINVAL);
816 if ((pos & mp->m_blockmask) || ((pos + count) & mp->m_blockmask))
820 * We don't need to take an exclusive lock unless there page cache needs
821 * to be invalidated or unaligned IO is being executed. We don't need to
822 * consider the EOF extension case here because
823 * xfs_file_aio_write_checks() will relock the inode as necessary for
824 * EOF zeroing cases and fill out the new inode size as appropriate.
826 if (unaligned_io || mapping->nrpages)
827 *iolock = XFS_IOLOCK_EXCL;
829 *iolock = XFS_IOLOCK_SHARED;
830 xfs_rw_ilock(ip, *iolock);
833 * Recheck if there are cached pages that need invalidate after we got
834 * the iolock to protect against other threads adding new pages while
835 * we were waiting for the iolock.
837 if (mapping->nrpages && *iolock == XFS_IOLOCK_SHARED) {
838 xfs_rw_iunlock(ip, *iolock);
839 *iolock = XFS_IOLOCK_EXCL;
840 xfs_rw_ilock(ip, *iolock);
843 ret = xfs_file_aio_write_checks(file, &pos, &count, new_size, iolock);
847 if (mapping->nrpages) {
848 ret = -xfs_flushinval_pages(ip, (pos & PAGE_CACHE_MASK), -1,
855 * If we are doing unaligned IO, wait for all other IO to drain,
856 * otherwise demote the lock if we had to flush cached pages
859 inode_dio_wait(inode);
860 else if (*iolock == XFS_IOLOCK_EXCL) {
861 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
862 *iolock = XFS_IOLOCK_SHARED;
865 trace_xfs_file_direct_write(ip, count, iocb->ki_pos, 0);
866 ret = generic_file_direct_write(iocb, iovp,
867 &nr_segs, pos, &iocb->ki_pos, count, ocount);
869 /* No fallback to buffered IO on errors for XFS. */
870 ASSERT(ret < 0 || ret == count);
875 xfs_file_buffered_aio_write(
877 const struct iovec *iovp,
878 unsigned long nr_segs,
881 xfs_fsize_t *new_size,
884 struct file *file = iocb->ki_filp;
885 struct address_space *mapping = file->f_mapping;
886 struct inode *inode = mapping->host;
887 struct xfs_inode *ip = XFS_I(inode);
890 size_t count = ocount;
892 *iolock = XFS_IOLOCK_EXCL;
893 xfs_rw_ilock(ip, *iolock);
895 ret = xfs_file_aio_write_checks(file, &pos, &count, new_size, iolock);
899 /* We can write back this queue in page reclaim */
900 current->backing_dev_info = mapping->backing_dev_info;
903 trace_xfs_file_buffered_write(ip, count, iocb->ki_pos, 0);
904 ret = generic_file_buffered_write(iocb, iovp, nr_segs,
905 pos, &iocb->ki_pos, count, ret);
907 * if we just got an ENOSPC, flush the inode now we aren't holding any
908 * page locks and retry *once*
910 if (ret == -ENOSPC && !enospc) {
911 ret = -xfs_flush_pages(ip, 0, -1, 0, FI_NONE);
917 current->backing_dev_info = NULL;
924 const struct iovec *iovp,
925 unsigned long nr_segs,
928 struct file *file = iocb->ki_filp;
929 struct address_space *mapping = file->f_mapping;
930 struct inode *inode = mapping->host;
931 struct xfs_inode *ip = XFS_I(inode);
935 xfs_fsize_t new_size = 0;
937 XFS_STATS_INC(xs_write_calls);
939 BUG_ON(iocb->ki_pos != pos);
941 ret = generic_segment_checks(iovp, &nr_segs, &ocount, VERIFY_READ);
948 xfs_wait_for_freeze(ip->i_mount, SB_FREEZE_WRITE);
950 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
953 if (unlikely(file->f_flags & O_DIRECT))
954 ret = xfs_file_dio_aio_write(iocb, iovp, nr_segs, pos,
955 ocount, &new_size, &iolock);
957 ret = xfs_file_buffered_aio_write(iocb, iovp, nr_segs, pos,
958 ocount, &new_size, &iolock);
960 xfs_aio_write_isize_update(inode, &iocb->ki_pos, ret);
965 /* Handle various SYNC-type writes */
966 if ((file->f_flags & O_DSYNC) || IS_SYNC(inode)) {
967 loff_t end = pos + ret - 1;
970 xfs_rw_iunlock(ip, iolock);
971 error = xfs_file_fsync(file, pos, end,
972 (file->f_flags & __O_SYNC) ? 0 : 1);
973 xfs_rw_ilock(ip, iolock);
979 xfs_aio_write_newsize_update(ip, new_size);
980 xfs_rw_iunlock(ip, iolock);
991 struct inode *inode = file->f_path.dentry->d_inode;
995 xfs_inode_t *ip = XFS_I(inode);
996 int cmd = XFS_IOC_RESVSP;
997 int attr_flags = XFS_ATTR_NOLOCK;
999 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
1003 bf.l_start = offset;
1006 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1008 if (mode & FALLOC_FL_PUNCH_HOLE)
1009 cmd = XFS_IOC_UNRESVSP;
1011 /* check the new inode size is valid before allocating */
1012 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
1013 offset + len > i_size_read(inode)) {
1014 new_size = offset + len;
1015 error = inode_newsize_ok(inode, new_size);
1020 if (file->f_flags & O_DSYNC)
1021 attr_flags |= XFS_ATTR_SYNC;
1023 error = -xfs_change_file_space(ip, cmd, &bf, 0, attr_flags);
1027 /* Change file size if needed */
1031 iattr.ia_valid = ATTR_SIZE;
1032 iattr.ia_size = new_size;
1033 error = -xfs_setattr_size(ip, &iattr, XFS_ATTR_NOLOCK);
1037 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1044 struct inode *inode,
1047 if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1049 if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb)))
1056 struct inode *inode,
1059 struct xfs_inode *ip = XFS_I(inode);
1063 error = xfs_file_open(inode, file);
1068 * If there are any blocks, read-ahead block 0 as we're almost
1069 * certain to have the next operation be a read there.
1071 mode = xfs_ilock_map_shared(ip);
1072 if (ip->i_d.di_nextents > 0)
1073 xfs_da_reada_buf(NULL, ip, 0, XFS_DATA_FORK);
1074 xfs_iunlock(ip, mode);
1080 struct inode *inode,
1083 return -xfs_release(XFS_I(inode));
1092 struct inode *inode = filp->f_path.dentry->d_inode;
1093 xfs_inode_t *ip = XFS_I(inode);
1098 * The Linux API doesn't pass down the total size of the buffer
1099 * we read into down to the filesystem. With the filldir concept
1100 * it's not needed for correct information, but the XFS dir2 leaf
1101 * code wants an estimate of the buffer size to calculate it's
1102 * readahead window and size the buffers used for mapping to
1105 * Try to give it an estimate that's good enough, maybe at some
1106 * point we can change the ->readdir prototype to include the
1107 * buffer size. For now we use the current glibc buffer size.
1109 bufsize = (size_t)min_t(loff_t, 32768, ip->i_d.di_size);
1111 error = xfs_readdir(ip, dirent, bufsize,
1112 (xfs_off_t *)&filp->f_pos, filldir);
1121 struct vm_area_struct *vma)
1123 vma->vm_ops = &xfs_file_vm_ops;
1124 vma->vm_flags |= VM_CAN_NONLINEAR;
1126 file_accessed(filp);
1131 * mmap()d file has taken write protection fault and is being made
1132 * writable. We can set the page state up correctly for a writable
1133 * page, which means we can do correct delalloc accounting (ENOSPC
1134 * checking!) and unwritten extent mapping.
1137 xfs_vm_page_mkwrite(
1138 struct vm_area_struct *vma,
1139 struct vm_fault *vmf)
1141 return block_page_mkwrite(vma, vmf, xfs_get_blocks);
1144 const struct file_operations xfs_file_operations = {
1145 .llseek = generic_file_llseek,
1146 .read = do_sync_read,
1147 .write = do_sync_write,
1148 .aio_read = xfs_file_aio_read,
1149 .aio_write = xfs_file_aio_write,
1150 .splice_read = xfs_file_splice_read,
1151 .splice_write = xfs_file_splice_write,
1152 .unlocked_ioctl = xfs_file_ioctl,
1153 #ifdef CONFIG_COMPAT
1154 .compat_ioctl = xfs_file_compat_ioctl,
1156 .mmap = xfs_file_mmap,
1157 .open = xfs_file_open,
1158 .release = xfs_file_release,
1159 .fsync = xfs_file_fsync,
1160 .fallocate = xfs_file_fallocate,
1163 const struct file_operations xfs_dir_file_operations = {
1164 .open = xfs_dir_open,
1165 .read = generic_read_dir,
1166 .readdir = xfs_file_readdir,
1167 .llseek = generic_file_llseek,
1168 .unlocked_ioctl = xfs_file_ioctl,
1169 #ifdef CONFIG_COMPAT
1170 .compat_ioctl = xfs_file_compat_ioctl,
1172 .fsync = xfs_dir_fsync,
1175 static const struct vm_operations_struct xfs_file_vm_ops = {
1176 .fault = filemap_fault,
1177 .page_mkwrite = xfs_vm_page_mkwrite,