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
20 #include "xfs_shared.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
24 #include "xfs_mount.h"
25 #include "xfs_da_format.h"
26 #include "xfs_da_btree.h"
27 #include "xfs_inode.h"
28 #include "xfs_trans.h"
29 #include "xfs_inode_item.h"
31 #include "xfs_bmap_util.h"
32 #include "xfs_error.h"
34 #include "xfs_dir2_priv.h"
35 #include "xfs_ioctl.h"
36 #include "xfs_trace.h"
38 #include "xfs_icache.h"
41 #include <linux/dcache.h>
42 #include <linux/falloc.h>
43 #include <linux/pagevec.h>
45 static const struct vm_operations_struct xfs_file_vm_ops;
48 * Locking primitives for read and write IO paths to ensure we consistently use
49 * and order the inode->i_mutex, ip->i_lock and ip->i_iolock.
56 if (type & XFS_IOLOCK_EXCL)
57 mutex_lock(&VFS_I(ip)->i_mutex);
66 xfs_iunlock(ip, type);
67 if (type & XFS_IOLOCK_EXCL)
68 mutex_unlock(&VFS_I(ip)->i_mutex);
76 xfs_ilock_demote(ip, type);
77 if (type & XFS_IOLOCK_EXCL)
78 mutex_unlock(&VFS_I(ip)->i_mutex);
84 * xfs_iozero clears the specified range of buffer supplied,
85 * and marks all the affected blocks as valid and modified. If
86 * an affected block is not allocated, it will be allocated. If
87 * an affected block is not completely overwritten, and is not
88 * valid before the operation, it will be read from disk before
89 * being partially zeroed.
93 struct xfs_inode *ip, /* inode */
94 loff_t pos, /* offset in file */
95 size_t count) /* size of data to zero */
98 struct address_space *mapping;
101 mapping = VFS_I(ip)->i_mapping;
103 unsigned offset, bytes;
106 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
107 bytes = PAGE_CACHE_SIZE - offset;
111 status = pagecache_write_begin(NULL, mapping, pos, bytes,
112 AOP_FLAG_UNINTERRUPTIBLE,
117 zero_user(page, offset, bytes);
119 status = pagecache_write_end(NULL, mapping, pos, bytes, bytes,
121 WARN_ON(status <= 0); /* can't return less than zero! */
131 xfs_update_prealloc_flags(
132 struct xfs_inode *ip,
133 enum xfs_prealloc_flags flags)
135 struct xfs_trans *tp;
138 tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_WRITEID);
139 error = xfs_trans_reserve(tp, &M_RES(ip->i_mount)->tr_writeid, 0, 0);
141 xfs_trans_cancel(tp, 0);
145 xfs_ilock(ip, XFS_ILOCK_EXCL);
146 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
148 if (!(flags & XFS_PREALLOC_INVISIBLE)) {
149 ip->i_d.di_mode &= ~S_ISUID;
150 if (ip->i_d.di_mode & S_IXGRP)
151 ip->i_d.di_mode &= ~S_ISGID;
152 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
155 if (flags & XFS_PREALLOC_SET)
156 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
157 if (flags & XFS_PREALLOC_CLEAR)
158 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
160 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
161 if (flags & XFS_PREALLOC_SYNC)
162 xfs_trans_set_sync(tp);
163 return xfs_trans_commit(tp, 0);
167 * Fsync operations on directories are much simpler than on regular files,
168 * as there is no file data to flush, and thus also no need for explicit
169 * cache flush operations, and there are no non-transaction metadata updates
170 * on directories either.
179 struct xfs_inode *ip = XFS_I(file->f_mapping->host);
180 struct xfs_mount *mp = ip->i_mount;
183 trace_xfs_dir_fsync(ip);
185 xfs_ilock(ip, XFS_ILOCK_SHARED);
186 if (xfs_ipincount(ip))
187 lsn = ip->i_itemp->ili_last_lsn;
188 xfs_iunlock(ip, XFS_ILOCK_SHARED);
192 return _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, NULL);
202 struct inode *inode = file->f_mapping->host;
203 struct xfs_inode *ip = XFS_I(inode);
204 struct xfs_mount *mp = ip->i_mount;
209 trace_xfs_file_fsync(ip);
211 error = filemap_write_and_wait_range(inode->i_mapping, start, end);
215 if (XFS_FORCED_SHUTDOWN(mp))
218 xfs_iflags_clear(ip, XFS_ITRUNCATED);
220 if (mp->m_flags & XFS_MOUNT_BARRIER) {
222 * If we have an RT and/or log subvolume we need to make sure
223 * to flush the write cache the device used for file data
224 * first. This is to ensure newly written file data make
225 * it to disk before logging the new inode size in case of
226 * an extending write.
228 if (XFS_IS_REALTIME_INODE(ip))
229 xfs_blkdev_issue_flush(mp->m_rtdev_targp);
230 else if (mp->m_logdev_targp != mp->m_ddev_targp)
231 xfs_blkdev_issue_flush(mp->m_ddev_targp);
235 * All metadata updates are logged, which means that we just have
236 * to flush the log up to the latest LSN that touched the inode.
238 xfs_ilock(ip, XFS_ILOCK_SHARED);
239 if (xfs_ipincount(ip)) {
241 (ip->i_itemp->ili_fields & ~XFS_ILOG_TIMESTAMP))
242 lsn = ip->i_itemp->ili_last_lsn;
244 xfs_iunlock(ip, XFS_ILOCK_SHARED);
247 error = _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, &log_flushed);
250 * If we only have a single device, and the log force about was
251 * a no-op we might have to flush the data device cache here.
252 * This can only happen for fdatasync/O_DSYNC if we were overwriting
253 * an already allocated file and thus do not have any metadata to
256 if ((mp->m_flags & XFS_MOUNT_BARRIER) &&
257 mp->m_logdev_targp == mp->m_ddev_targp &&
258 !XFS_IS_REALTIME_INODE(ip) &&
260 xfs_blkdev_issue_flush(mp->m_ddev_targp);
270 struct file *file = iocb->ki_filp;
271 struct inode *inode = file->f_mapping->host;
272 struct xfs_inode *ip = XFS_I(inode);
273 struct xfs_mount *mp = ip->i_mount;
274 size_t size = iov_iter_count(to);
278 loff_t pos = iocb->ki_pos;
280 XFS_STATS_INC(xs_read_calls);
282 if (unlikely(iocb->ki_flags & IOCB_DIRECT))
283 ioflags |= XFS_IO_ISDIRECT;
284 if (file->f_mode & FMODE_NOCMTIME)
285 ioflags |= XFS_IO_INVIS;
287 if (unlikely(ioflags & XFS_IO_ISDIRECT)) {
288 xfs_buftarg_t *target =
289 XFS_IS_REALTIME_INODE(ip) ?
290 mp->m_rtdev_targp : mp->m_ddev_targp;
291 /* DIO must be aligned to device logical sector size */
292 if ((pos | size) & target->bt_logical_sectormask) {
293 if (pos == i_size_read(inode))
299 n = mp->m_super->s_maxbytes - pos;
300 if (n <= 0 || size == 0)
306 if (XFS_FORCED_SHUTDOWN(mp))
310 * Locking is a bit tricky here. If we take an exclusive lock
311 * for direct IO, we effectively serialise all new concurrent
312 * read IO to this file and block it behind IO that is currently in
313 * progress because IO in progress holds the IO lock shared. We only
314 * need to hold the lock exclusive to blow away the page cache, so
315 * only take lock exclusively if the page cache needs invalidation.
316 * This allows the normal direct IO case of no page cache pages to
317 * proceeed concurrently without serialisation.
319 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
320 if ((ioflags & XFS_IO_ISDIRECT) && inode->i_mapping->nrpages) {
321 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
322 xfs_rw_ilock(ip, XFS_IOLOCK_EXCL);
324 if (inode->i_mapping->nrpages) {
325 ret = filemap_write_and_wait_range(
326 VFS_I(ip)->i_mapping,
327 pos, pos + size - 1);
329 xfs_rw_iunlock(ip, XFS_IOLOCK_EXCL);
334 * Invalidate whole pages. This can return an error if
335 * we fail to invalidate a page, but this should never
336 * happen on XFS. Warn if it does fail.
338 ret = invalidate_inode_pages2_range(VFS_I(ip)->i_mapping,
339 pos >> PAGE_CACHE_SHIFT,
340 (pos + size - 1) >> PAGE_CACHE_SHIFT);
344 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
347 trace_xfs_file_read(ip, size, pos, ioflags);
349 ret = generic_file_read_iter(iocb, to);
351 XFS_STATS_ADD(xs_read_bytes, ret);
353 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
358 xfs_file_splice_read(
361 struct pipe_inode_info *pipe,
365 struct xfs_inode *ip = XFS_I(infilp->f_mapping->host);
369 XFS_STATS_INC(xs_read_calls);
371 if (infilp->f_mode & FMODE_NOCMTIME)
372 ioflags |= XFS_IO_INVIS;
374 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
377 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
379 trace_xfs_file_splice_read(ip, count, *ppos, ioflags);
381 ret = generic_file_splice_read(infilp, ppos, pipe, count, flags);
383 XFS_STATS_ADD(xs_read_bytes, ret);
385 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
390 * This routine is called to handle zeroing any space in the last block of the
391 * file that is beyond the EOF. We do this since the size is being increased
392 * without writing anything to that block and we don't want to read the
393 * garbage on the disk.
395 STATIC int /* error (positive) */
397 struct xfs_inode *ip,
402 struct xfs_mount *mp = ip->i_mount;
403 xfs_fileoff_t last_fsb = XFS_B_TO_FSBT(mp, isize);
404 int zero_offset = XFS_B_FSB_OFFSET(mp, isize);
408 struct xfs_bmbt_irec imap;
410 xfs_ilock(ip, XFS_ILOCK_EXCL);
411 error = xfs_bmapi_read(ip, last_fsb, 1, &imap, &nimaps, 0);
412 xfs_iunlock(ip, XFS_ILOCK_EXCL);
419 * If the block underlying isize is just a hole, then there
420 * is nothing to zero.
422 if (imap.br_startblock == HOLESTARTBLOCK)
425 zero_len = mp->m_sb.sb_blocksize - zero_offset;
426 if (isize + zero_len > offset)
427 zero_len = offset - isize;
429 return xfs_iozero(ip, isize, zero_len);
433 * Zero any on disk space between the current EOF and the new, larger EOF.
435 * This handles the normal case of zeroing the remainder of the last block in
436 * the file and the unusual case of zeroing blocks out beyond the size of the
437 * file. This second case only happens with fixed size extents and when the
438 * system crashes before the inode size was updated but after blocks were
441 * Expects the iolock to be held exclusive, and will take the ilock internally.
443 int /* error (positive) */
445 struct xfs_inode *ip,
446 xfs_off_t offset, /* starting I/O offset */
447 xfs_fsize_t isize, /* current inode size */
450 struct xfs_mount *mp = ip->i_mount;
451 xfs_fileoff_t start_zero_fsb;
452 xfs_fileoff_t end_zero_fsb;
453 xfs_fileoff_t zero_count_fsb;
454 xfs_fileoff_t last_fsb;
455 xfs_fileoff_t zero_off;
456 xfs_fsize_t zero_len;
459 struct xfs_bmbt_irec imap;
461 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
462 ASSERT(offset > isize);
465 * First handle zeroing the block on which isize resides.
467 * We only zero a part of that block so it is handled specially.
469 if (XFS_B_FSB_OFFSET(mp, isize) != 0) {
470 error = xfs_zero_last_block(ip, offset, isize, did_zeroing);
476 * Calculate the range between the new size and the old where blocks
477 * needing to be zeroed may exist.
479 * To get the block where the last byte in the file currently resides,
480 * we need to subtract one from the size and truncate back to a block
481 * boundary. We subtract 1 in case the size is exactly on a block
484 last_fsb = isize ? XFS_B_TO_FSBT(mp, isize - 1) : (xfs_fileoff_t)-1;
485 start_zero_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
486 end_zero_fsb = XFS_B_TO_FSBT(mp, offset - 1);
487 ASSERT((xfs_sfiloff_t)last_fsb < (xfs_sfiloff_t)start_zero_fsb);
488 if (last_fsb == end_zero_fsb) {
490 * The size was only incremented on its last block.
491 * We took care of that above, so just return.
496 ASSERT(start_zero_fsb <= end_zero_fsb);
497 while (start_zero_fsb <= end_zero_fsb) {
499 zero_count_fsb = end_zero_fsb - start_zero_fsb + 1;
501 xfs_ilock(ip, XFS_ILOCK_EXCL);
502 error = xfs_bmapi_read(ip, start_zero_fsb, zero_count_fsb,
504 xfs_iunlock(ip, XFS_ILOCK_EXCL);
510 if (imap.br_state == XFS_EXT_UNWRITTEN ||
511 imap.br_startblock == HOLESTARTBLOCK) {
512 start_zero_fsb = imap.br_startoff + imap.br_blockcount;
513 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
518 * There are blocks we need to zero.
520 zero_off = XFS_FSB_TO_B(mp, start_zero_fsb);
521 zero_len = XFS_FSB_TO_B(mp, imap.br_blockcount);
523 if ((zero_off + zero_len) > offset)
524 zero_len = offset - zero_off;
526 error = xfs_iozero(ip, zero_off, zero_len);
531 start_zero_fsb = imap.br_startoff + imap.br_blockcount;
532 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
539 * Common pre-write limit and setup checks.
541 * Called with the iolocked held either shared and exclusive according to
542 * @iolock, and returns with it held. Might upgrade the iolock to exclusive
543 * if called for a direct write beyond i_size.
546 xfs_file_aio_write_checks(
548 struct iov_iter *from,
551 struct file *file = iocb->ki_filp;
552 struct inode *inode = file->f_mapping->host;
553 struct xfs_inode *ip = XFS_I(inode);
555 size_t count = iov_iter_count(from);
558 error = generic_write_checks(iocb, from);
562 error = xfs_break_layouts(inode, iolock, true);
566 /* For changing security info in file_remove_privs() we need i_mutex */
567 if (*iolock == XFS_IOLOCK_SHARED && !IS_NOSEC(inode)) {
568 xfs_rw_iunlock(ip, *iolock);
569 *iolock = XFS_IOLOCK_EXCL;
570 xfs_rw_ilock(ip, *iolock);
574 * If the offset is beyond the size of the file, we need to zero any
575 * blocks that fall between the existing EOF and the start of this
576 * write. If zeroing is needed and we are currently holding the
577 * iolock shared, we need to update it to exclusive which implies
578 * having to redo all checks before.
580 * We need to serialise against EOF updates that occur in IO
581 * completions here. We want to make sure that nobody is changing the
582 * size while we do this check until we have placed an IO barrier (i.e.
583 * hold the XFS_IOLOCK_EXCL) that prevents new IO from being dispatched.
584 * The spinlock effectively forms a memory barrier once we have the
585 * XFS_IOLOCK_EXCL so we are guaranteed to see the latest EOF value
586 * and hence be able to correctly determine if we need to run zeroing.
588 spin_lock(&ip->i_flags_lock);
589 if (iocb->ki_pos > i_size_read(inode)) {
592 spin_unlock(&ip->i_flags_lock);
593 if (*iolock == XFS_IOLOCK_SHARED) {
594 xfs_rw_iunlock(ip, *iolock);
595 *iolock = XFS_IOLOCK_EXCL;
596 xfs_rw_ilock(ip, *iolock);
597 iov_iter_reexpand(from, count);
600 * We now have an IO submission barrier in place, but
601 * AIO can do EOF updates during IO completion and hence
602 * we now need to wait for all of them to drain. Non-AIO
603 * DIO will have drained before we are given the
604 * XFS_IOLOCK_EXCL, and so for most cases this wait is a
607 inode_dio_wait(inode);
610 error = xfs_zero_eof(ip, iocb->ki_pos, i_size_read(inode), &zero);
614 spin_unlock(&ip->i_flags_lock);
617 * Updating the timestamps will grab the ilock again from
618 * xfs_fs_dirty_inode, so we have to call it after dropping the
619 * lock above. Eventually we should look into a way to avoid
620 * the pointless lock roundtrip.
622 if (likely(!(file->f_mode & FMODE_NOCMTIME))) {
623 error = file_update_time(file);
629 * If we're writing the file then make sure to clear the setuid and
630 * setgid bits if the process is not being run by root. This keeps
631 * people from modifying setuid and setgid binaries.
633 if (!IS_NOSEC(inode))
634 return file_remove_privs(file);
639 * xfs_file_dio_aio_write - handle direct IO writes
641 * Lock the inode appropriately to prepare for and issue a direct IO write.
642 * By separating it from the buffered write path we remove all the tricky to
643 * follow locking changes and looping.
645 * If there are cached pages or we're extending the file, we need IOLOCK_EXCL
646 * until we're sure the bytes at the new EOF have been zeroed and/or the cached
647 * pages are flushed out.
649 * In most cases the direct IO writes will be done holding IOLOCK_SHARED
650 * allowing them to be done in parallel with reads and other direct IO writes.
651 * However, if the IO is not aligned to filesystem blocks, the direct IO layer
652 * needs to do sub-block zeroing and that requires serialisation against other
653 * direct IOs to the same block. In this case we need to serialise the
654 * submission of the unaligned IOs so that we don't get racing block zeroing in
655 * the dio layer. To avoid the problem with aio, we also need to wait for
656 * outstanding IOs to complete so that unwritten extent conversion is completed
657 * before we try to map the overlapping block. This is currently implemented by
658 * hitting it with a big hammer (i.e. inode_dio_wait()).
660 * Returns with locks held indicated by @iolock and errors indicated by
661 * negative return values.
664 xfs_file_dio_aio_write(
666 struct iov_iter *from)
668 struct file *file = iocb->ki_filp;
669 struct address_space *mapping = file->f_mapping;
670 struct inode *inode = mapping->host;
671 struct xfs_inode *ip = XFS_I(inode);
672 struct xfs_mount *mp = ip->i_mount;
674 int unaligned_io = 0;
676 size_t count = iov_iter_count(from);
677 loff_t pos = iocb->ki_pos;
679 struct iov_iter data;
680 struct xfs_buftarg *target = XFS_IS_REALTIME_INODE(ip) ?
681 mp->m_rtdev_targp : mp->m_ddev_targp;
683 /* DIO must be aligned to device logical sector size */
684 if ((pos | count) & target->bt_logical_sectormask)
687 /* "unaligned" here means not aligned to a filesystem block */
688 if ((pos & mp->m_blockmask) || ((pos + count) & mp->m_blockmask))
692 * We don't need to take an exclusive lock unless there page cache needs
693 * to be invalidated or unaligned IO is being executed. We don't need to
694 * consider the EOF extension case here because
695 * xfs_file_aio_write_checks() will relock the inode as necessary for
696 * EOF zeroing cases and fill out the new inode size as appropriate.
698 if (unaligned_io || mapping->nrpages)
699 iolock = XFS_IOLOCK_EXCL;
701 iolock = XFS_IOLOCK_SHARED;
702 xfs_rw_ilock(ip, iolock);
705 * Recheck if there are cached pages that need invalidate after we got
706 * the iolock to protect against other threads adding new pages while
707 * we were waiting for the iolock.
709 if (mapping->nrpages && iolock == XFS_IOLOCK_SHARED) {
710 xfs_rw_iunlock(ip, iolock);
711 iolock = XFS_IOLOCK_EXCL;
712 xfs_rw_ilock(ip, iolock);
715 ret = xfs_file_aio_write_checks(iocb, from, &iolock);
718 count = iov_iter_count(from);
720 end = pos + count - 1;
722 if (mapping->nrpages) {
723 ret = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
728 * Invalidate whole pages. This can return an error if
729 * we fail to invalidate a page, but this should never
730 * happen on XFS. Warn if it does fail.
732 ret = invalidate_inode_pages2_range(VFS_I(ip)->i_mapping,
733 pos >> PAGE_CACHE_SHIFT,
734 end >> PAGE_CACHE_SHIFT);
740 * If we are doing unaligned IO, wait for all other IO to drain,
741 * otherwise demote the lock if we had to flush cached pages
744 inode_dio_wait(inode);
745 else if (iolock == XFS_IOLOCK_EXCL) {
746 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
747 iolock = XFS_IOLOCK_SHARED;
750 trace_xfs_file_direct_write(ip, count, iocb->ki_pos, 0);
753 ret = mapping->a_ops->direct_IO(iocb, &data, pos);
755 /* see generic_file_direct_write() for why this is necessary */
756 if (mapping->nrpages) {
757 invalidate_inode_pages2_range(mapping,
758 pos >> PAGE_CACHE_SHIFT,
759 end >> PAGE_CACHE_SHIFT);
764 iov_iter_advance(from, ret);
768 xfs_rw_iunlock(ip, iolock);
770 /* No fallback to buffered IO on errors for XFS. */
771 ASSERT(ret < 0 || ret == count);
776 xfs_file_buffered_aio_write(
778 struct iov_iter *from)
780 struct file *file = iocb->ki_filp;
781 struct address_space *mapping = file->f_mapping;
782 struct inode *inode = mapping->host;
783 struct xfs_inode *ip = XFS_I(inode);
786 int iolock = XFS_IOLOCK_EXCL;
788 xfs_rw_ilock(ip, iolock);
790 ret = xfs_file_aio_write_checks(iocb, from, &iolock);
794 /* We can write back this queue in page reclaim */
795 current->backing_dev_info = inode_to_bdi(inode);
798 trace_xfs_file_buffered_write(ip, iov_iter_count(from),
800 ret = generic_perform_write(file, from, iocb->ki_pos);
801 if (likely(ret >= 0))
805 * If we hit a space limit, try to free up some lingering preallocated
806 * space before returning an error. In the case of ENOSPC, first try to
807 * write back all dirty inodes to free up some of the excess reserved
808 * metadata space. This reduces the chances that the eofblocks scan
809 * waits on dirty mappings. Since xfs_flush_inodes() is serialized, this
810 * also behaves as a filter to prevent too many eofblocks scans from
811 * running at the same time.
813 if (ret == -EDQUOT && !enospc) {
814 enospc = xfs_inode_free_quota_eofblocks(ip);
817 } else if (ret == -ENOSPC && !enospc) {
818 struct xfs_eofblocks eofb = {0};
821 xfs_flush_inodes(ip->i_mount);
822 eofb.eof_scan_owner = ip->i_ino; /* for locking */
823 eofb.eof_flags = XFS_EOF_FLAGS_SYNC;
824 xfs_icache_free_eofblocks(ip->i_mount, &eofb);
828 current->backing_dev_info = NULL;
830 xfs_rw_iunlock(ip, iolock);
837 struct iov_iter *from)
839 struct file *file = iocb->ki_filp;
840 struct address_space *mapping = file->f_mapping;
841 struct inode *inode = mapping->host;
842 struct xfs_inode *ip = XFS_I(inode);
844 size_t ocount = iov_iter_count(from);
846 XFS_STATS_INC(xs_write_calls);
851 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
854 if (unlikely(iocb->ki_flags & IOCB_DIRECT))
855 ret = xfs_file_dio_aio_write(iocb, from);
857 ret = xfs_file_buffered_aio_write(iocb, from);
862 XFS_STATS_ADD(xs_write_bytes, ret);
864 /* Handle various SYNC-type writes */
865 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
872 #define XFS_FALLOC_FL_SUPPORTED \
873 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
874 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE | \
875 FALLOC_FL_INSERT_RANGE)
884 struct inode *inode = file_inode(file);
885 struct xfs_inode *ip = XFS_I(inode);
887 enum xfs_prealloc_flags flags = 0;
888 uint iolock = XFS_IOLOCK_EXCL;
890 bool do_file_insert = 0;
892 if (!S_ISREG(inode->i_mode))
894 if (mode & ~XFS_FALLOC_FL_SUPPORTED)
897 xfs_ilock(ip, iolock);
898 error = xfs_break_layouts(inode, &iolock, false);
902 xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
903 iolock |= XFS_MMAPLOCK_EXCL;
905 if (mode & FALLOC_FL_PUNCH_HOLE) {
906 error = xfs_free_file_space(ip, offset, len);
909 } else if (mode & FALLOC_FL_COLLAPSE_RANGE) {
910 unsigned blksize_mask = (1 << inode->i_blkbits) - 1;
912 if (offset & blksize_mask || len & blksize_mask) {
918 * There is no need to overlap collapse range with EOF,
919 * in which case it is effectively a truncate operation
921 if (offset + len >= i_size_read(inode)) {
926 new_size = i_size_read(inode) - len;
928 error = xfs_collapse_file_space(ip, offset, len);
931 } else if (mode & FALLOC_FL_INSERT_RANGE) {
932 unsigned blksize_mask = (1 << inode->i_blkbits) - 1;
934 new_size = i_size_read(inode) + len;
935 if (offset & blksize_mask || len & blksize_mask) {
940 /* check the new inode size does not wrap through zero */
941 if (new_size > inode->i_sb->s_maxbytes) {
946 /* Offset should be less than i_size */
947 if (offset >= i_size_read(inode)) {
953 flags |= XFS_PREALLOC_SET;
955 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
956 offset + len > i_size_read(inode)) {
957 new_size = offset + len;
958 error = inode_newsize_ok(inode, new_size);
963 if (mode & FALLOC_FL_ZERO_RANGE)
964 error = xfs_zero_file_space(ip, offset, len);
966 error = xfs_alloc_file_space(ip, offset, len,
972 if (file->f_flags & O_DSYNC)
973 flags |= XFS_PREALLOC_SYNC;
975 error = xfs_update_prealloc_flags(ip, flags);
979 /* Change file size if needed */
983 iattr.ia_valid = ATTR_SIZE;
984 iattr.ia_size = new_size;
985 error = xfs_setattr_size(ip, &iattr);
991 * Perform hole insertion now that the file size has been
992 * updated so that if we crash during the operation we don't
993 * leave shifted extents past EOF and hence losing access to
994 * the data that is contained within them.
997 error = xfs_insert_file_space(ip, offset, len);
1000 xfs_iunlock(ip, iolock);
1007 struct inode *inode,
1010 if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1012 if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb)))
1019 struct inode *inode,
1022 struct xfs_inode *ip = XFS_I(inode);
1026 error = xfs_file_open(inode, file);
1031 * If there are any blocks, read-ahead block 0 as we're almost
1032 * certain to have the next operation be a read there.
1034 mode = xfs_ilock_data_map_shared(ip);
1035 if (ip->i_d.di_nextents > 0)
1036 xfs_dir3_data_readahead(ip, 0, -1);
1037 xfs_iunlock(ip, mode);
1043 struct inode *inode,
1046 return xfs_release(XFS_I(inode));
1052 struct dir_context *ctx)
1054 struct inode *inode = file_inode(file);
1055 xfs_inode_t *ip = XFS_I(inode);
1059 * The Linux API doesn't pass down the total size of the buffer
1060 * we read into down to the filesystem. With the filldir concept
1061 * it's not needed for correct information, but the XFS dir2 leaf
1062 * code wants an estimate of the buffer size to calculate it's
1063 * readahead window and size the buffers used for mapping to
1066 * Try to give it an estimate that's good enough, maybe at some
1067 * point we can change the ->readdir prototype to include the
1068 * buffer size. For now we use the current glibc buffer size.
1070 bufsize = (size_t)min_t(loff_t, 32768, ip->i_d.di_size);
1072 return xfs_readdir(ip, ctx, bufsize);
1078 struct vm_area_struct *vma)
1080 vma->vm_ops = &xfs_file_vm_ops;
1082 file_accessed(filp);
1087 * This type is designed to indicate the type of offset we would like
1088 * to search from page cache for xfs_seek_hole_data().
1096 * Lookup the desired type of offset from the given page.
1098 * On success, return true and the offset argument will point to the
1099 * start of the region that was found. Otherwise this function will
1100 * return false and keep the offset argument unchanged.
1103 xfs_lookup_buffer_offset(
1108 loff_t lastoff = page_offset(page);
1110 struct buffer_head *bh, *head;
1112 bh = head = page_buffers(page);
1115 * Unwritten extents that have data in the page
1116 * cache covering them can be identified by the
1117 * BH_Unwritten state flag. Pages with multiple
1118 * buffers might have a mix of holes, data and
1119 * unwritten extents - any buffer with valid
1120 * data in it should have BH_Uptodate flag set
1123 if (buffer_unwritten(bh) ||
1124 buffer_uptodate(bh)) {
1125 if (type == DATA_OFF)
1128 if (type == HOLE_OFF)
1136 lastoff += bh->b_size;
1137 } while ((bh = bh->b_this_page) != head);
1143 * This routine is called to find out and return a data or hole offset
1144 * from the page cache for unwritten extents according to the desired
1145 * type for xfs_seek_hole_data().
1147 * The argument offset is used to tell where we start to search from the
1148 * page cache. Map is used to figure out the end points of the range to
1151 * Return true if the desired type of offset was found, and the argument
1152 * offset is filled with that address. Otherwise, return false and keep
1156 xfs_find_get_desired_pgoff(
1157 struct inode *inode,
1158 struct xfs_bmbt_irec *map,
1162 struct xfs_inode *ip = XFS_I(inode);
1163 struct xfs_mount *mp = ip->i_mount;
1164 struct pagevec pvec;
1168 loff_t startoff = *offset;
1169 loff_t lastoff = startoff;
1172 pagevec_init(&pvec, 0);
1174 index = startoff >> PAGE_CACHE_SHIFT;
1175 endoff = XFS_FSB_TO_B(mp, map->br_startoff + map->br_blockcount);
1176 end = endoff >> PAGE_CACHE_SHIFT;
1182 want = min_t(pgoff_t, end - index, PAGEVEC_SIZE);
1183 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index,
1186 * No page mapped into given range. If we are searching holes
1187 * and if this is the first time we got into the loop, it means
1188 * that the given offset is landed in a hole, return it.
1190 * If we have already stepped through some block buffers to find
1191 * holes but they all contains data. In this case, the last
1192 * offset is already updated and pointed to the end of the last
1193 * mapped page, if it does not reach the endpoint to search,
1194 * that means there should be a hole between them.
1196 if (nr_pages == 0) {
1197 /* Data search found nothing */
1198 if (type == DATA_OFF)
1201 ASSERT(type == HOLE_OFF);
1202 if (lastoff == startoff || lastoff < endoff) {
1210 * At lease we found one page. If this is the first time we
1211 * step into the loop, and if the first page index offset is
1212 * greater than the given search offset, a hole was found.
1214 if (type == HOLE_OFF && lastoff == startoff &&
1215 lastoff < page_offset(pvec.pages[0])) {
1220 for (i = 0; i < nr_pages; i++) {
1221 struct page *page = pvec.pages[i];
1225 * At this point, the page may be truncated or
1226 * invalidated (changing page->mapping to NULL),
1227 * or even swizzled back from swapper_space to tmpfs
1228 * file mapping. However, page->index will not change
1229 * because we have a reference on the page.
1231 * Searching done if the page index is out of range.
1232 * If the current offset is not reaches the end of
1233 * the specified search range, there should be a hole
1236 if (page->index > end) {
1237 if (type == HOLE_OFF && lastoff < endoff) {
1246 * Page truncated or invalidated(page->mapping == NULL).
1247 * We can freely skip it and proceed to check the next
1250 if (unlikely(page->mapping != inode->i_mapping)) {
1255 if (!page_has_buffers(page)) {
1260 found = xfs_lookup_buffer_offset(page, &b_offset, type);
1263 * The found offset may be less than the start
1264 * point to search if this is the first time to
1267 *offset = max_t(loff_t, startoff, b_offset);
1273 * We either searching data but nothing was found, or
1274 * searching hole but found a data buffer. In either
1275 * case, probably the next page contains the desired
1276 * things, update the last offset to it so.
1278 lastoff = page_offset(page) + PAGE_SIZE;
1283 * The number of returned pages less than our desired, search
1284 * done. In this case, nothing was found for searching data,
1285 * but we found a hole behind the last offset.
1287 if (nr_pages < want) {
1288 if (type == HOLE_OFF) {
1295 index = pvec.pages[i - 1]->index + 1;
1296 pagevec_release(&pvec);
1297 } while (index <= end);
1300 pagevec_release(&pvec);
1310 struct inode *inode = file->f_mapping->host;
1311 struct xfs_inode *ip = XFS_I(inode);
1312 struct xfs_mount *mp = ip->i_mount;
1313 loff_t uninitialized_var(offset);
1315 xfs_fileoff_t fsbno;
1320 if (XFS_FORCED_SHUTDOWN(mp))
1323 lock = xfs_ilock_data_map_shared(ip);
1325 isize = i_size_read(inode);
1326 if (start >= isize) {
1332 * Try to read extents from the first block indicated
1333 * by fsbno to the end block of the file.
1335 fsbno = XFS_B_TO_FSBT(mp, start);
1336 end = XFS_B_TO_FSB(mp, isize);
1339 struct xfs_bmbt_irec map[2];
1343 error = xfs_bmapi_read(ip, fsbno, end - fsbno, map, &nmap,
1348 /* No extents at given offset, must be beyond EOF */
1354 for (i = 0; i < nmap; i++) {
1355 offset = max_t(loff_t, start,
1356 XFS_FSB_TO_B(mp, map[i].br_startoff));
1358 /* Landed in the hole we wanted? */
1359 if (whence == SEEK_HOLE &&
1360 map[i].br_startblock == HOLESTARTBLOCK)
1363 /* Landed in the data extent we wanted? */
1364 if (whence == SEEK_DATA &&
1365 (map[i].br_startblock == DELAYSTARTBLOCK ||
1366 (map[i].br_state == XFS_EXT_NORM &&
1367 !isnullstartblock(map[i].br_startblock))))
1371 * Landed in an unwritten extent, try to search
1372 * for hole or data from page cache.
1374 if (map[i].br_state == XFS_EXT_UNWRITTEN) {
1375 if (xfs_find_get_desired_pgoff(inode, &map[i],
1376 whence == SEEK_HOLE ? HOLE_OFF : DATA_OFF,
1383 * We only received one extent out of the two requested. This
1384 * means we've hit EOF and didn't find what we are looking for.
1388 * If we were looking for a hole, set offset to
1389 * the end of the file (i.e., there is an implicit
1390 * hole at the end of any file).
1392 if (whence == SEEK_HOLE) {
1397 * If we were looking for data, it's nowhere to be found
1399 ASSERT(whence == SEEK_DATA);
1407 * Nothing was found, proceed to the next round of search
1408 * if the next reading offset is not at or beyond EOF.
1410 fsbno = map[i - 1].br_startoff + map[i - 1].br_blockcount;
1411 start = XFS_FSB_TO_B(mp, fsbno);
1412 if (start >= isize) {
1413 if (whence == SEEK_HOLE) {
1417 ASSERT(whence == SEEK_DATA);
1425 * If at this point we have found the hole we wanted, the returned
1426 * offset may be bigger than the file size as it may be aligned to
1427 * page boundary for unwritten extents. We need to deal with this
1428 * situation in particular.
1430 if (whence == SEEK_HOLE)
1431 offset = min_t(loff_t, offset, isize);
1432 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
1435 xfs_iunlock(ip, lock);
1452 return generic_file_llseek(file, offset, whence);
1455 return xfs_seek_hole_data(file, offset, whence);
1462 * Locking for serialisation of IO during page faults. This results in a lock
1466 * i_mmap_lock (XFS - truncate serialisation)
1468 * i_lock (XFS - extent map serialisation)
1472 struct vm_area_struct *vma,
1473 struct vm_fault *vmf)
1475 struct xfs_inode *ip = XFS_I(vma->vm_file->f_mapping->host);
1478 trace_xfs_filemap_fault(ip);
1480 xfs_ilock(ip, XFS_MMAPLOCK_SHARED);
1481 error = filemap_fault(vma, vmf);
1482 xfs_iunlock(ip, XFS_MMAPLOCK_SHARED);
1488 * mmap()d file has taken write protection fault and is being made writable. We
1489 * can set the page state up correctly for a writable page, which means we can
1490 * do correct delalloc accounting (ENOSPC checking!) and unwritten extent
1494 xfs_filemap_page_mkwrite(
1495 struct vm_area_struct *vma,
1496 struct vm_fault *vmf)
1498 struct xfs_inode *ip = XFS_I(vma->vm_file->f_mapping->host);
1501 trace_xfs_filemap_page_mkwrite(ip);
1503 xfs_ilock(ip, XFS_MMAPLOCK_SHARED);
1504 error = block_page_mkwrite(vma, vmf, xfs_get_blocks);
1505 xfs_iunlock(ip, XFS_MMAPLOCK_SHARED);
1510 const struct file_operations xfs_file_operations = {
1511 .llseek = xfs_file_llseek,
1512 .read_iter = xfs_file_read_iter,
1513 .write_iter = xfs_file_write_iter,
1514 .splice_read = xfs_file_splice_read,
1515 .splice_write = iter_file_splice_write,
1516 .unlocked_ioctl = xfs_file_ioctl,
1517 #ifdef CONFIG_COMPAT
1518 .compat_ioctl = xfs_file_compat_ioctl,
1520 .mmap = xfs_file_mmap,
1521 .open = xfs_file_open,
1522 .release = xfs_file_release,
1523 .fsync = xfs_file_fsync,
1524 .fallocate = xfs_file_fallocate,
1527 const struct file_operations xfs_dir_file_operations = {
1528 .open = xfs_dir_open,
1529 .read = generic_read_dir,
1530 .iterate = xfs_file_readdir,
1531 .llseek = generic_file_llseek,
1532 .unlocked_ioctl = xfs_file_ioctl,
1533 #ifdef CONFIG_COMPAT
1534 .compat_ioctl = xfs_file_compat_ioctl,
1536 .fsync = xfs_dir_fsync,
1539 static const struct vm_operations_struct xfs_file_vm_ops = {
1540 .fault = xfs_filemap_fault,
1541 .map_pages = filemap_map_pages,
1542 .page_mkwrite = xfs_filemap_page_mkwrite,