hyperv: Add processing of MTU reduced by the host
[linux/fpc-iii.git] / fs / xfs / xfs_file.c
blobeb596b4199420fa8378dcd49146640c9783037b4
1 /*
2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
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
18 #include "xfs.h"
19 #include "xfs_fs.h"
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_sb.h"
25 #include "xfs_ag.h"
26 #include "xfs_mount.h"
27 #include "xfs_da_format.h"
28 #include "xfs_da_btree.h"
29 #include "xfs_inode.h"
30 #include "xfs_trans.h"
31 #include "xfs_inode_item.h"
32 #include "xfs_bmap.h"
33 #include "xfs_bmap_util.h"
34 #include "xfs_error.h"
35 #include "xfs_dir2.h"
36 #include "xfs_dir2_priv.h"
37 #include "xfs_ioctl.h"
38 #include "xfs_trace.h"
39 #include "xfs_log.h"
40 #include "xfs_dinode.h"
41 #include "xfs_icache.h"
43 #include <linux/aio.h>
44 #include <linux/dcache.h>
45 #include <linux/falloc.h>
46 #include <linux/pagevec.h>
48 static const struct vm_operations_struct xfs_file_vm_ops;
51 * Locking primitives for read and write IO paths to ensure we consistently use
52 * and order the inode->i_mutex, ip->i_lock and ip->i_iolock.
54 static inline void
55 xfs_rw_ilock(
56 struct xfs_inode *ip,
57 int type)
59 if (type & XFS_IOLOCK_EXCL)
60 mutex_lock(&VFS_I(ip)->i_mutex);
61 xfs_ilock(ip, type);
64 static inline void
65 xfs_rw_iunlock(
66 struct xfs_inode *ip,
67 int type)
69 xfs_iunlock(ip, type);
70 if (type & XFS_IOLOCK_EXCL)
71 mutex_unlock(&VFS_I(ip)->i_mutex);
74 static inline void
75 xfs_rw_ilock_demote(
76 struct xfs_inode *ip,
77 int type)
79 xfs_ilock_demote(ip, type);
80 if (type & XFS_IOLOCK_EXCL)
81 mutex_unlock(&VFS_I(ip)->i_mutex);
85 * xfs_iozero
87 * xfs_iozero clears the specified range of buffer supplied,
88 * and marks all the affected blocks as valid and modified. If
89 * an affected block is not allocated, it will be allocated. If
90 * an affected block is not completely overwritten, and is not
91 * valid before the operation, it will be read from disk before
92 * being partially zeroed.
94 int
95 xfs_iozero(
96 struct xfs_inode *ip, /* inode */
97 loff_t pos, /* offset in file */
98 size_t count) /* size of data to zero */
100 struct page *page;
101 struct address_space *mapping;
102 int status;
104 mapping = VFS_I(ip)->i_mapping;
105 do {
106 unsigned offset, bytes;
107 void *fsdata;
109 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
110 bytes = PAGE_CACHE_SIZE - offset;
111 if (bytes > count)
112 bytes = count;
114 status = pagecache_write_begin(NULL, mapping, pos, bytes,
115 AOP_FLAG_UNINTERRUPTIBLE,
116 &page, &fsdata);
117 if (status)
118 break;
120 zero_user(page, offset, bytes);
122 status = pagecache_write_end(NULL, mapping, pos, bytes, bytes,
123 page, fsdata);
124 WARN_ON(status <= 0); /* can't return less than zero! */
125 pos += bytes;
126 count -= bytes;
127 status = 0;
128 } while (count);
130 return (-status);
134 * Fsync operations on directories are much simpler than on regular files,
135 * as there is no file data to flush, and thus also no need for explicit
136 * cache flush operations, and there are no non-transaction metadata updates
137 * on directories either.
139 STATIC int
140 xfs_dir_fsync(
141 struct file *file,
142 loff_t start,
143 loff_t end,
144 int datasync)
146 struct xfs_inode *ip = XFS_I(file->f_mapping->host);
147 struct xfs_mount *mp = ip->i_mount;
148 xfs_lsn_t lsn = 0;
150 trace_xfs_dir_fsync(ip);
152 xfs_ilock(ip, XFS_ILOCK_SHARED);
153 if (xfs_ipincount(ip))
154 lsn = ip->i_itemp->ili_last_lsn;
155 xfs_iunlock(ip, XFS_ILOCK_SHARED);
157 if (!lsn)
158 return 0;
159 return _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, NULL);
162 STATIC int
163 xfs_file_fsync(
164 struct file *file,
165 loff_t start,
166 loff_t end,
167 int datasync)
169 struct inode *inode = file->f_mapping->host;
170 struct xfs_inode *ip = XFS_I(inode);
171 struct xfs_mount *mp = ip->i_mount;
172 int error = 0;
173 int log_flushed = 0;
174 xfs_lsn_t lsn = 0;
176 trace_xfs_file_fsync(ip);
178 error = filemap_write_and_wait_range(inode->i_mapping, start, end);
179 if (error)
180 return error;
182 if (XFS_FORCED_SHUTDOWN(mp))
183 return -EIO;
185 xfs_iflags_clear(ip, XFS_ITRUNCATED);
187 if (mp->m_flags & XFS_MOUNT_BARRIER) {
189 * If we have an RT and/or log subvolume we need to make sure
190 * to flush the write cache the device used for file data
191 * first. This is to ensure newly written file data make
192 * it to disk before logging the new inode size in case of
193 * an extending write.
195 if (XFS_IS_REALTIME_INODE(ip))
196 xfs_blkdev_issue_flush(mp->m_rtdev_targp);
197 else if (mp->m_logdev_targp != mp->m_ddev_targp)
198 xfs_blkdev_issue_flush(mp->m_ddev_targp);
202 * All metadata updates are logged, which means that we just have
203 * to flush the log up to the latest LSN that touched the inode.
205 xfs_ilock(ip, XFS_ILOCK_SHARED);
206 if (xfs_ipincount(ip)) {
207 if (!datasync ||
208 (ip->i_itemp->ili_fields & ~XFS_ILOG_TIMESTAMP))
209 lsn = ip->i_itemp->ili_last_lsn;
211 xfs_iunlock(ip, XFS_ILOCK_SHARED);
213 if (lsn)
214 error = _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, &log_flushed);
217 * If we only have a single device, and the log force about was
218 * a no-op we might have to flush the data device cache here.
219 * This can only happen for fdatasync/O_DSYNC if we were overwriting
220 * an already allocated file and thus do not have any metadata to
221 * commit.
223 if ((mp->m_flags & XFS_MOUNT_BARRIER) &&
224 mp->m_logdev_targp == mp->m_ddev_targp &&
225 !XFS_IS_REALTIME_INODE(ip) &&
226 !log_flushed)
227 xfs_blkdev_issue_flush(mp->m_ddev_targp);
229 return error;
232 STATIC ssize_t
233 xfs_file_read_iter(
234 struct kiocb *iocb,
235 struct iov_iter *to)
237 struct file *file = iocb->ki_filp;
238 struct inode *inode = file->f_mapping->host;
239 struct xfs_inode *ip = XFS_I(inode);
240 struct xfs_mount *mp = ip->i_mount;
241 size_t size = iov_iter_count(to);
242 ssize_t ret = 0;
243 int ioflags = 0;
244 xfs_fsize_t n;
245 loff_t pos = iocb->ki_pos;
247 XFS_STATS_INC(xs_read_calls);
249 if (unlikely(file->f_flags & O_DIRECT))
250 ioflags |= XFS_IO_ISDIRECT;
251 if (file->f_mode & FMODE_NOCMTIME)
252 ioflags |= XFS_IO_INVIS;
254 if (unlikely(ioflags & XFS_IO_ISDIRECT)) {
255 xfs_buftarg_t *target =
256 XFS_IS_REALTIME_INODE(ip) ?
257 mp->m_rtdev_targp : mp->m_ddev_targp;
258 /* DIO must be aligned to device logical sector size */
259 if ((pos | size) & target->bt_logical_sectormask) {
260 if (pos == i_size_read(inode))
261 return 0;
262 return -EINVAL;
266 n = mp->m_super->s_maxbytes - pos;
267 if (n <= 0 || size == 0)
268 return 0;
270 if (n < size)
271 size = n;
273 if (XFS_FORCED_SHUTDOWN(mp))
274 return -EIO;
277 * Locking is a bit tricky here. If we take an exclusive lock
278 * for direct IO, we effectively serialise all new concurrent
279 * read IO to this file and block it behind IO that is currently in
280 * progress because IO in progress holds the IO lock shared. We only
281 * need to hold the lock exclusive to blow away the page cache, so
282 * only take lock exclusively if the page cache needs invalidation.
283 * This allows the normal direct IO case of no page cache pages to
284 * proceeed concurrently without serialisation.
286 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
287 if ((ioflags & XFS_IO_ISDIRECT) && inode->i_mapping->nrpages) {
288 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
289 xfs_rw_ilock(ip, XFS_IOLOCK_EXCL);
291 if (inode->i_mapping->nrpages) {
292 ret = filemap_write_and_wait_range(
293 VFS_I(ip)->i_mapping,
294 pos, pos + size - 1);
295 if (ret) {
296 xfs_rw_iunlock(ip, XFS_IOLOCK_EXCL);
297 return ret;
301 * Invalidate whole pages. This can return an error if
302 * we fail to invalidate a page, but this should never
303 * happen on XFS. Warn if it does fail.
305 ret = invalidate_inode_pages2_range(VFS_I(ip)->i_mapping,
306 pos >> PAGE_CACHE_SHIFT,
307 (pos + size - 1) >> PAGE_CACHE_SHIFT);
308 WARN_ON_ONCE(ret);
309 ret = 0;
311 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
314 trace_xfs_file_read(ip, size, pos, ioflags);
316 ret = generic_file_read_iter(iocb, to);
317 if (ret > 0)
318 XFS_STATS_ADD(xs_read_bytes, ret);
320 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
321 return ret;
324 STATIC ssize_t
325 xfs_file_splice_read(
326 struct file *infilp,
327 loff_t *ppos,
328 struct pipe_inode_info *pipe,
329 size_t count,
330 unsigned int flags)
332 struct xfs_inode *ip = XFS_I(infilp->f_mapping->host);
333 int ioflags = 0;
334 ssize_t ret;
336 XFS_STATS_INC(xs_read_calls);
338 if (infilp->f_mode & FMODE_NOCMTIME)
339 ioflags |= XFS_IO_INVIS;
341 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
342 return -EIO;
344 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
346 trace_xfs_file_splice_read(ip, count, *ppos, ioflags);
348 ret = generic_file_splice_read(infilp, ppos, pipe, count, flags);
349 if (ret > 0)
350 XFS_STATS_ADD(xs_read_bytes, ret);
352 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
353 return ret;
357 * This routine is called to handle zeroing any space in the last block of the
358 * file that is beyond the EOF. We do this since the size is being increased
359 * without writing anything to that block and we don't want to read the
360 * garbage on the disk.
362 STATIC int /* error (positive) */
363 xfs_zero_last_block(
364 struct xfs_inode *ip,
365 xfs_fsize_t offset,
366 xfs_fsize_t isize)
368 struct xfs_mount *mp = ip->i_mount;
369 xfs_fileoff_t last_fsb = XFS_B_TO_FSBT(mp, isize);
370 int zero_offset = XFS_B_FSB_OFFSET(mp, isize);
371 int zero_len;
372 int nimaps = 1;
373 int error = 0;
374 struct xfs_bmbt_irec imap;
376 xfs_ilock(ip, XFS_ILOCK_EXCL);
377 error = xfs_bmapi_read(ip, last_fsb, 1, &imap, &nimaps, 0);
378 xfs_iunlock(ip, XFS_ILOCK_EXCL);
379 if (error)
380 return error;
382 ASSERT(nimaps > 0);
385 * If the block underlying isize is just a hole, then there
386 * is nothing to zero.
388 if (imap.br_startblock == HOLESTARTBLOCK)
389 return 0;
391 zero_len = mp->m_sb.sb_blocksize - zero_offset;
392 if (isize + zero_len > offset)
393 zero_len = offset - isize;
394 return xfs_iozero(ip, isize, zero_len);
398 * Zero any on disk space between the current EOF and the new, larger EOF.
400 * This handles the normal case of zeroing the remainder of the last block in
401 * the file and the unusual case of zeroing blocks out beyond the size of the
402 * file. This second case only happens with fixed size extents and when the
403 * system crashes before the inode size was updated but after blocks were
404 * allocated.
406 * Expects the iolock to be held exclusive, and will take the ilock internally.
408 int /* error (positive) */
409 xfs_zero_eof(
410 struct xfs_inode *ip,
411 xfs_off_t offset, /* starting I/O offset */
412 xfs_fsize_t isize) /* current inode size */
414 struct xfs_mount *mp = ip->i_mount;
415 xfs_fileoff_t start_zero_fsb;
416 xfs_fileoff_t end_zero_fsb;
417 xfs_fileoff_t zero_count_fsb;
418 xfs_fileoff_t last_fsb;
419 xfs_fileoff_t zero_off;
420 xfs_fsize_t zero_len;
421 int nimaps;
422 int error = 0;
423 struct xfs_bmbt_irec imap;
425 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
426 ASSERT(offset > isize);
429 * First handle zeroing the block on which isize resides.
431 * We only zero a part of that block so it is handled specially.
433 if (XFS_B_FSB_OFFSET(mp, isize) != 0) {
434 error = xfs_zero_last_block(ip, offset, isize);
435 if (error)
436 return error;
440 * Calculate the range between the new size and the old where blocks
441 * needing to be zeroed may exist.
443 * To get the block where the last byte in the file currently resides,
444 * we need to subtract one from the size and truncate back to a block
445 * boundary. We subtract 1 in case the size is exactly on a block
446 * boundary.
448 last_fsb = isize ? XFS_B_TO_FSBT(mp, isize - 1) : (xfs_fileoff_t)-1;
449 start_zero_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
450 end_zero_fsb = XFS_B_TO_FSBT(mp, offset - 1);
451 ASSERT((xfs_sfiloff_t)last_fsb < (xfs_sfiloff_t)start_zero_fsb);
452 if (last_fsb == end_zero_fsb) {
454 * The size was only incremented on its last block.
455 * We took care of that above, so just return.
457 return 0;
460 ASSERT(start_zero_fsb <= end_zero_fsb);
461 while (start_zero_fsb <= end_zero_fsb) {
462 nimaps = 1;
463 zero_count_fsb = end_zero_fsb - start_zero_fsb + 1;
465 xfs_ilock(ip, XFS_ILOCK_EXCL);
466 error = xfs_bmapi_read(ip, start_zero_fsb, zero_count_fsb,
467 &imap, &nimaps, 0);
468 xfs_iunlock(ip, XFS_ILOCK_EXCL);
469 if (error)
470 return error;
472 ASSERT(nimaps > 0);
474 if (imap.br_state == XFS_EXT_UNWRITTEN ||
475 imap.br_startblock == HOLESTARTBLOCK) {
476 start_zero_fsb = imap.br_startoff + imap.br_blockcount;
477 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
478 continue;
482 * There are blocks we need to zero.
484 zero_off = XFS_FSB_TO_B(mp, start_zero_fsb);
485 zero_len = XFS_FSB_TO_B(mp, imap.br_blockcount);
487 if ((zero_off + zero_len) > offset)
488 zero_len = offset - zero_off;
490 error = xfs_iozero(ip, zero_off, zero_len);
491 if (error)
492 return error;
494 start_zero_fsb = imap.br_startoff + imap.br_blockcount;
495 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
498 return 0;
502 * Common pre-write limit and setup checks.
504 * Called with the iolocked held either shared and exclusive according to
505 * @iolock, and returns with it held. Might upgrade the iolock to exclusive
506 * if called for a direct write beyond i_size.
508 STATIC ssize_t
509 xfs_file_aio_write_checks(
510 struct file *file,
511 loff_t *pos,
512 size_t *count,
513 int *iolock)
515 struct inode *inode = file->f_mapping->host;
516 struct xfs_inode *ip = XFS_I(inode);
517 int error = 0;
519 restart:
520 error = generic_write_checks(file, pos, count, S_ISBLK(inode->i_mode));
521 if (error)
522 return error;
525 * If the offset is beyond the size of the file, we need to zero any
526 * blocks that fall between the existing EOF and the start of this
527 * write. If zeroing is needed and we are currently holding the
528 * iolock shared, we need to update it to exclusive which implies
529 * having to redo all checks before.
531 if (*pos > i_size_read(inode)) {
532 if (*iolock == XFS_IOLOCK_SHARED) {
533 xfs_rw_iunlock(ip, *iolock);
534 *iolock = XFS_IOLOCK_EXCL;
535 xfs_rw_ilock(ip, *iolock);
536 goto restart;
538 error = xfs_zero_eof(ip, *pos, i_size_read(inode));
539 if (error)
540 return error;
544 * Updating the timestamps will grab the ilock again from
545 * xfs_fs_dirty_inode, so we have to call it after dropping the
546 * lock above. Eventually we should look into a way to avoid
547 * the pointless lock roundtrip.
549 if (likely(!(file->f_mode & FMODE_NOCMTIME))) {
550 error = file_update_time(file);
551 if (error)
552 return error;
556 * If we're writing the file then make sure to clear the setuid and
557 * setgid bits if the process is not being run by root. This keeps
558 * people from modifying setuid and setgid binaries.
560 return file_remove_suid(file);
564 * xfs_file_dio_aio_write - handle direct IO writes
566 * Lock the inode appropriately to prepare for and issue a direct IO write.
567 * By separating it from the buffered write path we remove all the tricky to
568 * follow locking changes and looping.
570 * If there are cached pages or we're extending the file, we need IOLOCK_EXCL
571 * until we're sure the bytes at the new EOF have been zeroed and/or the cached
572 * pages are flushed out.
574 * In most cases the direct IO writes will be done holding IOLOCK_SHARED
575 * allowing them to be done in parallel with reads and other direct IO writes.
576 * However, if the IO is not aligned to filesystem blocks, the direct IO layer
577 * needs to do sub-block zeroing and that requires serialisation against other
578 * direct IOs to the same block. In this case we need to serialise the
579 * submission of the unaligned IOs so that we don't get racing block zeroing in
580 * the dio layer. To avoid the problem with aio, we also need to wait for
581 * outstanding IOs to complete so that unwritten extent conversion is completed
582 * before we try to map the overlapping block. This is currently implemented by
583 * hitting it with a big hammer (i.e. inode_dio_wait()).
585 * Returns with locks held indicated by @iolock and errors indicated by
586 * negative return values.
588 STATIC ssize_t
589 xfs_file_dio_aio_write(
590 struct kiocb *iocb,
591 struct iov_iter *from)
593 struct file *file = iocb->ki_filp;
594 struct address_space *mapping = file->f_mapping;
595 struct inode *inode = mapping->host;
596 struct xfs_inode *ip = XFS_I(inode);
597 struct xfs_mount *mp = ip->i_mount;
598 ssize_t ret = 0;
599 int unaligned_io = 0;
600 int iolock;
601 size_t count = iov_iter_count(from);
602 loff_t pos = iocb->ki_pos;
603 struct xfs_buftarg *target = XFS_IS_REALTIME_INODE(ip) ?
604 mp->m_rtdev_targp : mp->m_ddev_targp;
606 /* DIO must be aligned to device logical sector size */
607 if ((pos | count) & target->bt_logical_sectormask)
608 return -EINVAL;
610 /* "unaligned" here means not aligned to a filesystem block */
611 if ((pos & mp->m_blockmask) || ((pos + count) & mp->m_blockmask))
612 unaligned_io = 1;
615 * We don't need to take an exclusive lock unless there page cache needs
616 * to be invalidated or unaligned IO is being executed. We don't need to
617 * consider the EOF extension case here because
618 * xfs_file_aio_write_checks() will relock the inode as necessary for
619 * EOF zeroing cases and fill out the new inode size as appropriate.
621 if (unaligned_io || mapping->nrpages)
622 iolock = XFS_IOLOCK_EXCL;
623 else
624 iolock = XFS_IOLOCK_SHARED;
625 xfs_rw_ilock(ip, iolock);
628 * Recheck if there are cached pages that need invalidate after we got
629 * the iolock to protect against other threads adding new pages while
630 * we were waiting for the iolock.
632 if (mapping->nrpages && iolock == XFS_IOLOCK_SHARED) {
633 xfs_rw_iunlock(ip, iolock);
634 iolock = XFS_IOLOCK_EXCL;
635 xfs_rw_ilock(ip, iolock);
638 ret = xfs_file_aio_write_checks(file, &pos, &count, &iolock);
639 if (ret)
640 goto out;
641 iov_iter_truncate(from, count);
643 if (mapping->nrpages) {
644 ret = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
645 pos, pos + count - 1);
646 if (ret)
647 goto out;
649 * Invalidate whole pages. This can return an error if
650 * we fail to invalidate a page, but this should never
651 * happen on XFS. Warn if it does fail.
653 ret = invalidate_inode_pages2_range(VFS_I(ip)->i_mapping,
654 pos >> PAGE_CACHE_SHIFT,
655 (pos + count - 1) >> PAGE_CACHE_SHIFT);
656 WARN_ON_ONCE(ret);
657 ret = 0;
661 * If we are doing unaligned IO, wait for all other IO to drain,
662 * otherwise demote the lock if we had to flush cached pages
664 if (unaligned_io)
665 inode_dio_wait(inode);
666 else if (iolock == XFS_IOLOCK_EXCL) {
667 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
668 iolock = XFS_IOLOCK_SHARED;
671 trace_xfs_file_direct_write(ip, count, iocb->ki_pos, 0);
672 ret = generic_file_direct_write(iocb, from, pos);
674 out:
675 xfs_rw_iunlock(ip, iolock);
677 /* No fallback to buffered IO on errors for XFS. */
678 ASSERT(ret < 0 || ret == count);
679 return ret;
682 STATIC ssize_t
683 xfs_file_buffered_aio_write(
684 struct kiocb *iocb,
685 struct iov_iter *from)
687 struct file *file = iocb->ki_filp;
688 struct address_space *mapping = file->f_mapping;
689 struct inode *inode = mapping->host;
690 struct xfs_inode *ip = XFS_I(inode);
691 ssize_t ret;
692 int enospc = 0;
693 int iolock = XFS_IOLOCK_EXCL;
694 loff_t pos = iocb->ki_pos;
695 size_t count = iov_iter_count(from);
697 xfs_rw_ilock(ip, iolock);
699 ret = xfs_file_aio_write_checks(file, &pos, &count, &iolock);
700 if (ret)
701 goto out;
703 iov_iter_truncate(from, count);
704 /* We can write back this queue in page reclaim */
705 current->backing_dev_info = mapping->backing_dev_info;
707 write_retry:
708 trace_xfs_file_buffered_write(ip, count, iocb->ki_pos, 0);
709 ret = generic_perform_write(file, from, pos);
710 if (likely(ret >= 0))
711 iocb->ki_pos = pos + ret;
714 * If we hit a space limit, try to free up some lingering preallocated
715 * space before returning an error. In the case of ENOSPC, first try to
716 * write back all dirty inodes to free up some of the excess reserved
717 * metadata space. This reduces the chances that the eofblocks scan
718 * waits on dirty mappings. Since xfs_flush_inodes() is serialized, this
719 * also behaves as a filter to prevent too many eofblocks scans from
720 * running at the same time.
722 if (ret == -EDQUOT && !enospc) {
723 enospc = xfs_inode_free_quota_eofblocks(ip);
724 if (enospc)
725 goto write_retry;
726 } else if (ret == -ENOSPC && !enospc) {
727 struct xfs_eofblocks eofb = {0};
729 enospc = 1;
730 xfs_flush_inodes(ip->i_mount);
731 eofb.eof_scan_owner = ip->i_ino; /* for locking */
732 eofb.eof_flags = XFS_EOF_FLAGS_SYNC;
733 xfs_icache_free_eofblocks(ip->i_mount, &eofb);
734 goto write_retry;
737 current->backing_dev_info = NULL;
738 out:
739 xfs_rw_iunlock(ip, iolock);
740 return ret;
743 STATIC ssize_t
744 xfs_file_write_iter(
745 struct kiocb *iocb,
746 struct iov_iter *from)
748 struct file *file = iocb->ki_filp;
749 struct address_space *mapping = file->f_mapping;
750 struct inode *inode = mapping->host;
751 struct xfs_inode *ip = XFS_I(inode);
752 ssize_t ret;
753 size_t ocount = iov_iter_count(from);
755 XFS_STATS_INC(xs_write_calls);
757 if (ocount == 0)
758 return 0;
760 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
761 return -EIO;
763 if (unlikely(file->f_flags & O_DIRECT))
764 ret = xfs_file_dio_aio_write(iocb, from);
765 else
766 ret = xfs_file_buffered_aio_write(iocb, from);
768 if (ret > 0) {
769 ssize_t err;
771 XFS_STATS_ADD(xs_write_bytes, ret);
773 /* Handle various SYNC-type writes */
774 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
775 if (err < 0)
776 ret = err;
778 return ret;
781 STATIC long
782 xfs_file_fallocate(
783 struct file *file,
784 int mode,
785 loff_t offset,
786 loff_t len)
788 struct inode *inode = file_inode(file);
789 struct xfs_inode *ip = XFS_I(inode);
790 struct xfs_trans *tp;
791 long error;
792 loff_t new_size = 0;
794 if (!S_ISREG(inode->i_mode))
795 return -EINVAL;
796 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
797 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE))
798 return -EOPNOTSUPP;
800 xfs_ilock(ip, XFS_IOLOCK_EXCL);
801 if (mode & FALLOC_FL_PUNCH_HOLE) {
802 error = xfs_free_file_space(ip, offset, len);
803 if (error)
804 goto out_unlock;
805 } else if (mode & FALLOC_FL_COLLAPSE_RANGE) {
806 unsigned blksize_mask = (1 << inode->i_blkbits) - 1;
808 if (offset & blksize_mask || len & blksize_mask) {
809 error = -EINVAL;
810 goto out_unlock;
814 * There is no need to overlap collapse range with EOF,
815 * in which case it is effectively a truncate operation
817 if (offset + len >= i_size_read(inode)) {
818 error = -EINVAL;
819 goto out_unlock;
822 new_size = i_size_read(inode) - len;
824 error = xfs_collapse_file_space(ip, offset, len);
825 if (error)
826 goto out_unlock;
827 } else {
828 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
829 offset + len > i_size_read(inode)) {
830 new_size = offset + len;
831 error = inode_newsize_ok(inode, new_size);
832 if (error)
833 goto out_unlock;
836 if (mode & FALLOC_FL_ZERO_RANGE)
837 error = xfs_zero_file_space(ip, offset, len);
838 else
839 error = xfs_alloc_file_space(ip, offset, len,
840 XFS_BMAPI_PREALLOC);
841 if (error)
842 goto out_unlock;
845 tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_WRITEID);
846 error = xfs_trans_reserve(tp, &M_RES(ip->i_mount)->tr_writeid, 0, 0);
847 if (error) {
848 xfs_trans_cancel(tp, 0);
849 goto out_unlock;
852 xfs_ilock(ip, XFS_ILOCK_EXCL);
853 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
854 ip->i_d.di_mode &= ~S_ISUID;
855 if (ip->i_d.di_mode & S_IXGRP)
856 ip->i_d.di_mode &= ~S_ISGID;
858 if (!(mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_COLLAPSE_RANGE)))
859 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
861 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
862 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
864 if (file->f_flags & O_DSYNC)
865 xfs_trans_set_sync(tp);
866 error = xfs_trans_commit(tp, 0);
867 if (error)
868 goto out_unlock;
870 /* Change file size if needed */
871 if (new_size) {
872 struct iattr iattr;
874 iattr.ia_valid = ATTR_SIZE;
875 iattr.ia_size = new_size;
876 error = xfs_setattr_size(ip, &iattr);
879 out_unlock:
880 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
881 return error;
885 STATIC int
886 xfs_file_open(
887 struct inode *inode,
888 struct file *file)
890 if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
891 return -EFBIG;
892 if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb)))
893 return -EIO;
894 return 0;
897 STATIC int
898 xfs_dir_open(
899 struct inode *inode,
900 struct file *file)
902 struct xfs_inode *ip = XFS_I(inode);
903 int mode;
904 int error;
906 error = xfs_file_open(inode, file);
907 if (error)
908 return error;
911 * If there are any blocks, read-ahead block 0 as we're almost
912 * certain to have the next operation be a read there.
914 mode = xfs_ilock_data_map_shared(ip);
915 if (ip->i_d.di_nextents > 0)
916 xfs_dir3_data_readahead(ip, 0, -1);
917 xfs_iunlock(ip, mode);
918 return 0;
921 STATIC int
922 xfs_file_release(
923 struct inode *inode,
924 struct file *filp)
926 return xfs_release(XFS_I(inode));
929 STATIC int
930 xfs_file_readdir(
931 struct file *file,
932 struct dir_context *ctx)
934 struct inode *inode = file_inode(file);
935 xfs_inode_t *ip = XFS_I(inode);
936 int error;
937 size_t bufsize;
940 * The Linux API doesn't pass down the total size of the buffer
941 * we read into down to the filesystem. With the filldir concept
942 * it's not needed for correct information, but the XFS dir2 leaf
943 * code wants an estimate of the buffer size to calculate it's
944 * readahead window and size the buffers used for mapping to
945 * physical blocks.
947 * Try to give it an estimate that's good enough, maybe at some
948 * point we can change the ->readdir prototype to include the
949 * buffer size. For now we use the current glibc buffer size.
951 bufsize = (size_t)min_t(loff_t, 32768, ip->i_d.di_size);
953 error = xfs_readdir(ip, ctx, bufsize);
954 if (error)
955 return error;
956 return 0;
959 STATIC int
960 xfs_file_mmap(
961 struct file *filp,
962 struct vm_area_struct *vma)
964 vma->vm_ops = &xfs_file_vm_ops;
966 file_accessed(filp);
967 return 0;
971 * mmap()d file has taken write protection fault and is being made
972 * writable. We can set the page state up correctly for a writable
973 * page, which means we can do correct delalloc accounting (ENOSPC
974 * checking!) and unwritten extent mapping.
976 STATIC int
977 xfs_vm_page_mkwrite(
978 struct vm_area_struct *vma,
979 struct vm_fault *vmf)
981 return block_page_mkwrite(vma, vmf, xfs_get_blocks);
985 * This type is designed to indicate the type of offset we would like
986 * to search from page cache for xfs_seek_hole_data().
988 enum {
989 HOLE_OFF = 0,
990 DATA_OFF,
994 * Lookup the desired type of offset from the given page.
996 * On success, return true and the offset argument will point to the
997 * start of the region that was found. Otherwise this function will
998 * return false and keep the offset argument unchanged.
1000 STATIC bool
1001 xfs_lookup_buffer_offset(
1002 struct page *page,
1003 loff_t *offset,
1004 unsigned int type)
1006 loff_t lastoff = page_offset(page);
1007 bool found = false;
1008 struct buffer_head *bh, *head;
1010 bh = head = page_buffers(page);
1011 do {
1013 * Unwritten extents that have data in the page
1014 * cache covering them can be identified by the
1015 * BH_Unwritten state flag. Pages with multiple
1016 * buffers might have a mix of holes, data and
1017 * unwritten extents - any buffer with valid
1018 * data in it should have BH_Uptodate flag set
1019 * on it.
1021 if (buffer_unwritten(bh) ||
1022 buffer_uptodate(bh)) {
1023 if (type == DATA_OFF)
1024 found = true;
1025 } else {
1026 if (type == HOLE_OFF)
1027 found = true;
1030 if (found) {
1031 *offset = lastoff;
1032 break;
1034 lastoff += bh->b_size;
1035 } while ((bh = bh->b_this_page) != head);
1037 return found;
1041 * This routine is called to find out and return a data or hole offset
1042 * from the page cache for unwritten extents according to the desired
1043 * type for xfs_seek_hole_data().
1045 * The argument offset is used to tell where we start to search from the
1046 * page cache. Map is used to figure out the end points of the range to
1047 * lookup pages.
1049 * Return true if the desired type of offset was found, and the argument
1050 * offset is filled with that address. Otherwise, return false and keep
1051 * offset unchanged.
1053 STATIC bool
1054 xfs_find_get_desired_pgoff(
1055 struct inode *inode,
1056 struct xfs_bmbt_irec *map,
1057 unsigned int type,
1058 loff_t *offset)
1060 struct xfs_inode *ip = XFS_I(inode);
1061 struct xfs_mount *mp = ip->i_mount;
1062 struct pagevec pvec;
1063 pgoff_t index;
1064 pgoff_t end;
1065 loff_t endoff;
1066 loff_t startoff = *offset;
1067 loff_t lastoff = startoff;
1068 bool found = false;
1070 pagevec_init(&pvec, 0);
1072 index = startoff >> PAGE_CACHE_SHIFT;
1073 endoff = XFS_FSB_TO_B(mp, map->br_startoff + map->br_blockcount);
1074 end = endoff >> PAGE_CACHE_SHIFT;
1075 do {
1076 int want;
1077 unsigned nr_pages;
1078 unsigned int i;
1080 want = min_t(pgoff_t, end - index, PAGEVEC_SIZE);
1081 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index,
1082 want);
1084 * No page mapped into given range. If we are searching holes
1085 * and if this is the first time we got into the loop, it means
1086 * that the given offset is landed in a hole, return it.
1088 * If we have already stepped through some block buffers to find
1089 * holes but they all contains data. In this case, the last
1090 * offset is already updated and pointed to the end of the last
1091 * mapped page, if it does not reach the endpoint to search,
1092 * that means there should be a hole between them.
1094 if (nr_pages == 0) {
1095 /* Data search found nothing */
1096 if (type == DATA_OFF)
1097 break;
1099 ASSERT(type == HOLE_OFF);
1100 if (lastoff == startoff || lastoff < endoff) {
1101 found = true;
1102 *offset = lastoff;
1104 break;
1108 * At lease we found one page. If this is the first time we
1109 * step into the loop, and if the first page index offset is
1110 * greater than the given search offset, a hole was found.
1112 if (type == HOLE_OFF && lastoff == startoff &&
1113 lastoff < page_offset(pvec.pages[0])) {
1114 found = true;
1115 break;
1118 for (i = 0; i < nr_pages; i++) {
1119 struct page *page = pvec.pages[i];
1120 loff_t b_offset;
1123 * At this point, the page may be truncated or
1124 * invalidated (changing page->mapping to NULL),
1125 * or even swizzled back from swapper_space to tmpfs
1126 * file mapping. However, page->index will not change
1127 * because we have a reference on the page.
1129 * Searching done if the page index is out of range.
1130 * If the current offset is not reaches the end of
1131 * the specified search range, there should be a hole
1132 * between them.
1134 if (page->index > end) {
1135 if (type == HOLE_OFF && lastoff < endoff) {
1136 *offset = lastoff;
1137 found = true;
1139 goto out;
1142 lock_page(page);
1144 * Page truncated or invalidated(page->mapping == NULL).
1145 * We can freely skip it and proceed to check the next
1146 * page.
1148 if (unlikely(page->mapping != inode->i_mapping)) {
1149 unlock_page(page);
1150 continue;
1153 if (!page_has_buffers(page)) {
1154 unlock_page(page);
1155 continue;
1158 found = xfs_lookup_buffer_offset(page, &b_offset, type);
1159 if (found) {
1161 * The found offset may be less than the start
1162 * point to search if this is the first time to
1163 * come here.
1165 *offset = max_t(loff_t, startoff, b_offset);
1166 unlock_page(page);
1167 goto out;
1171 * We either searching data but nothing was found, or
1172 * searching hole but found a data buffer. In either
1173 * case, probably the next page contains the desired
1174 * things, update the last offset to it so.
1176 lastoff = page_offset(page) + PAGE_SIZE;
1177 unlock_page(page);
1181 * The number of returned pages less than our desired, search
1182 * done. In this case, nothing was found for searching data,
1183 * but we found a hole behind the last offset.
1185 if (nr_pages < want) {
1186 if (type == HOLE_OFF) {
1187 *offset = lastoff;
1188 found = true;
1190 break;
1193 index = pvec.pages[i - 1]->index + 1;
1194 pagevec_release(&pvec);
1195 } while (index <= end);
1197 out:
1198 pagevec_release(&pvec);
1199 return found;
1202 STATIC loff_t
1203 xfs_seek_hole_data(
1204 struct file *file,
1205 loff_t start,
1206 int whence)
1208 struct inode *inode = file->f_mapping->host;
1209 struct xfs_inode *ip = XFS_I(inode);
1210 struct xfs_mount *mp = ip->i_mount;
1211 loff_t uninitialized_var(offset);
1212 xfs_fsize_t isize;
1213 xfs_fileoff_t fsbno;
1214 xfs_filblks_t end;
1215 uint lock;
1216 int error;
1218 if (XFS_FORCED_SHUTDOWN(mp))
1219 return -EIO;
1221 lock = xfs_ilock_data_map_shared(ip);
1223 isize = i_size_read(inode);
1224 if (start >= isize) {
1225 error = -ENXIO;
1226 goto out_unlock;
1230 * Try to read extents from the first block indicated
1231 * by fsbno to the end block of the file.
1233 fsbno = XFS_B_TO_FSBT(mp, start);
1234 end = XFS_B_TO_FSB(mp, isize);
1236 for (;;) {
1237 struct xfs_bmbt_irec map[2];
1238 int nmap = 2;
1239 unsigned int i;
1241 error = xfs_bmapi_read(ip, fsbno, end - fsbno, map, &nmap,
1242 XFS_BMAPI_ENTIRE);
1243 if (error)
1244 goto out_unlock;
1246 /* No extents at given offset, must be beyond EOF */
1247 if (nmap == 0) {
1248 error = -ENXIO;
1249 goto out_unlock;
1252 for (i = 0; i < nmap; i++) {
1253 offset = max_t(loff_t, start,
1254 XFS_FSB_TO_B(mp, map[i].br_startoff));
1256 /* Landed in the hole we wanted? */
1257 if (whence == SEEK_HOLE &&
1258 map[i].br_startblock == HOLESTARTBLOCK)
1259 goto out;
1261 /* Landed in the data extent we wanted? */
1262 if (whence == SEEK_DATA &&
1263 (map[i].br_startblock == DELAYSTARTBLOCK ||
1264 (map[i].br_state == XFS_EXT_NORM &&
1265 !isnullstartblock(map[i].br_startblock))))
1266 goto out;
1269 * Landed in an unwritten extent, try to search
1270 * for hole or data from page cache.
1272 if (map[i].br_state == XFS_EXT_UNWRITTEN) {
1273 if (xfs_find_get_desired_pgoff(inode, &map[i],
1274 whence == SEEK_HOLE ? HOLE_OFF : DATA_OFF,
1275 &offset))
1276 goto out;
1281 * We only received one extent out of the two requested. This
1282 * means we've hit EOF and didn't find what we are looking for.
1284 if (nmap == 1) {
1286 * If we were looking for a hole, set offset to
1287 * the end of the file (i.e., there is an implicit
1288 * hole at the end of any file).
1290 if (whence == SEEK_HOLE) {
1291 offset = isize;
1292 break;
1295 * If we were looking for data, it's nowhere to be found
1297 ASSERT(whence == SEEK_DATA);
1298 error = -ENXIO;
1299 goto out_unlock;
1302 ASSERT(i > 1);
1305 * Nothing was found, proceed to the next round of search
1306 * if the next reading offset is not at or beyond EOF.
1308 fsbno = map[i - 1].br_startoff + map[i - 1].br_blockcount;
1309 start = XFS_FSB_TO_B(mp, fsbno);
1310 if (start >= isize) {
1311 if (whence == SEEK_HOLE) {
1312 offset = isize;
1313 break;
1315 ASSERT(whence == SEEK_DATA);
1316 error = -ENXIO;
1317 goto out_unlock;
1321 out:
1323 * If at this point we have found the hole we wanted, the returned
1324 * offset may be bigger than the file size as it may be aligned to
1325 * page boundary for unwritten extents. We need to deal with this
1326 * situation in particular.
1328 if (whence == SEEK_HOLE)
1329 offset = min_t(loff_t, offset, isize);
1330 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
1332 out_unlock:
1333 xfs_iunlock(ip, lock);
1335 if (error)
1336 return error;
1337 return offset;
1340 STATIC loff_t
1341 xfs_file_llseek(
1342 struct file *file,
1343 loff_t offset,
1344 int whence)
1346 switch (whence) {
1347 case SEEK_END:
1348 case SEEK_CUR:
1349 case SEEK_SET:
1350 return generic_file_llseek(file, offset, whence);
1351 case SEEK_HOLE:
1352 case SEEK_DATA:
1353 return xfs_seek_hole_data(file, offset, whence);
1354 default:
1355 return -EINVAL;
1359 const struct file_operations xfs_file_operations = {
1360 .llseek = xfs_file_llseek,
1361 .read = new_sync_read,
1362 .write = new_sync_write,
1363 .read_iter = xfs_file_read_iter,
1364 .write_iter = xfs_file_write_iter,
1365 .splice_read = xfs_file_splice_read,
1366 .splice_write = iter_file_splice_write,
1367 .unlocked_ioctl = xfs_file_ioctl,
1368 #ifdef CONFIG_COMPAT
1369 .compat_ioctl = xfs_file_compat_ioctl,
1370 #endif
1371 .mmap = xfs_file_mmap,
1372 .open = xfs_file_open,
1373 .release = xfs_file_release,
1374 .fsync = xfs_file_fsync,
1375 .fallocate = xfs_file_fallocate,
1378 const struct file_operations xfs_dir_file_operations = {
1379 .open = xfs_dir_open,
1380 .read = generic_read_dir,
1381 .iterate = xfs_file_readdir,
1382 .llseek = generic_file_llseek,
1383 .unlocked_ioctl = xfs_file_ioctl,
1384 #ifdef CONFIG_COMPAT
1385 .compat_ioctl = xfs_file_compat_ioctl,
1386 #endif
1387 .fsync = xfs_dir_fsync,
1390 static const struct vm_operations_struct xfs_file_vm_ops = {
1391 .fault = filemap_fault,
1392 .map_pages = filemap_map_pages,
1393 .page_mkwrite = xfs_vm_page_mkwrite,
1394 .remap_pages = generic_file_remap_pages,