Loading...
1/*
2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
4 *
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.
8 *
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.
13 *
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
17 */
18#include "xfs.h"
19#include "xfs_fs.h"
20#include "xfs_bit.h"
21#include "xfs_log.h"
22#include "xfs_inum.h"
23#include "xfs_sb.h"
24#include "xfs_ag.h"
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"
32#include "xfs_bmap.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"
38
39#include <linux/dcache.h>
40#include <linux/falloc.h>
41
42static const struct vm_operations_struct xfs_file_vm_ops;
43
44/*
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.
47 */
48static inline void
49xfs_rw_ilock(
50 struct xfs_inode *ip,
51 int type)
52{
53 if (type & XFS_IOLOCK_EXCL)
54 mutex_lock(&VFS_I(ip)->i_mutex);
55 xfs_ilock(ip, type);
56}
57
58static inline void
59xfs_rw_iunlock(
60 struct xfs_inode *ip,
61 int type)
62{
63 xfs_iunlock(ip, type);
64 if (type & XFS_IOLOCK_EXCL)
65 mutex_unlock(&VFS_I(ip)->i_mutex);
66}
67
68static inline void
69xfs_rw_ilock_demote(
70 struct xfs_inode *ip,
71 int type)
72{
73 xfs_ilock_demote(ip, type);
74 if (type & XFS_IOLOCK_EXCL)
75 mutex_unlock(&VFS_I(ip)->i_mutex);
76}
77
78/*
79 * xfs_iozero
80 *
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.
87 */
88STATIC int
89xfs_iozero(
90 struct xfs_inode *ip, /* inode */
91 loff_t pos, /* offset in file */
92 size_t count) /* size of data to zero */
93{
94 struct page *page;
95 struct address_space *mapping;
96 int status;
97
98 mapping = VFS_I(ip)->i_mapping;
99 do {
100 unsigned offset, bytes;
101 void *fsdata;
102
103 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
104 bytes = PAGE_CACHE_SIZE - offset;
105 if (bytes > count)
106 bytes = count;
107
108 status = pagecache_write_begin(NULL, mapping, pos, bytes,
109 AOP_FLAG_UNINTERRUPTIBLE,
110 &page, &fsdata);
111 if (status)
112 break;
113
114 zero_user(page, offset, bytes);
115
116 status = pagecache_write_end(NULL, mapping, pos, bytes, bytes,
117 page, fsdata);
118 WARN_ON(status <= 0); /* can't return less than zero! */
119 pos += bytes;
120 count -= bytes;
121 status = 0;
122 } while (count);
123
124 return (-status);
125}
126
127STATIC int
128xfs_file_fsync(
129 struct file *file,
130 loff_t start,
131 loff_t end,
132 int datasync)
133{
134 struct inode *inode = file->f_mapping->host;
135 struct xfs_inode *ip = XFS_I(inode);
136 struct xfs_mount *mp = ip->i_mount;
137 struct xfs_trans *tp;
138 int error = 0;
139 int log_flushed = 0;
140
141 trace_xfs_file_fsync(ip);
142
143 error = filemap_write_and_wait_range(inode->i_mapping, start, end);
144 if (error)
145 return error;
146
147 if (XFS_FORCED_SHUTDOWN(mp))
148 return -XFS_ERROR(EIO);
149
150 xfs_iflags_clear(ip, XFS_ITRUNCATED);
151
152 xfs_ilock(ip, XFS_IOLOCK_SHARED);
153 xfs_ioend_wait(ip);
154 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
155
156 if (mp->m_flags & XFS_MOUNT_BARRIER) {
157 /*
158 * If we have an RT and/or log subvolume we need to make sure
159 * to flush the write cache the device used for file data
160 * first. This is to ensure newly written file data make
161 * it to disk before logging the new inode size in case of
162 * an extending write.
163 */
164 if (XFS_IS_REALTIME_INODE(ip))
165 xfs_blkdev_issue_flush(mp->m_rtdev_targp);
166 else if (mp->m_logdev_targp != mp->m_ddev_targp)
167 xfs_blkdev_issue_flush(mp->m_ddev_targp);
168 }
169
170 /*
171 * We always need to make sure that the required inode state is safe on
172 * disk. The inode might be clean but we still might need to force the
173 * log because of committed transactions that haven't hit the disk yet.
174 * Likewise, there could be unflushed non-transactional changes to the
175 * inode core that have to go to disk and this requires us to issue
176 * a synchronous transaction to capture these changes correctly.
177 *
178 * This code relies on the assumption that if the i_update_core field
179 * of the inode is clear and the inode is unpinned then it is clean
180 * and no action is required.
181 */
182 xfs_ilock(ip, XFS_ILOCK_SHARED);
183
184 /*
185 * First check if the VFS inode is marked dirty. All the dirtying
186 * of non-transactional updates no goes through mark_inode_dirty*,
187 * which allows us to distinguish beteeen pure timestamp updates
188 * and i_size updates which need to be caught for fdatasync.
189 * After that also theck for the dirty state in the XFS inode, which
190 * might gets cleared when the inode gets written out via the AIL
191 * or xfs_iflush_cluster.
192 */
193 if (((inode->i_state & I_DIRTY_DATASYNC) ||
194 ((inode->i_state & I_DIRTY_SYNC) && !datasync)) &&
195 ip->i_update_core) {
196 /*
197 * Kick off a transaction to log the inode core to get the
198 * updates. The sync transaction will also force the log.
199 */
200 xfs_iunlock(ip, XFS_ILOCK_SHARED);
201 tp = xfs_trans_alloc(mp, XFS_TRANS_FSYNC_TS);
202 error = xfs_trans_reserve(tp, 0,
203 XFS_FSYNC_TS_LOG_RES(mp), 0, 0, 0);
204 if (error) {
205 xfs_trans_cancel(tp, 0);
206 return -error;
207 }
208 xfs_ilock(ip, XFS_ILOCK_EXCL);
209
210 /*
211 * Note - it's possible that we might have pushed ourselves out
212 * of the way during trans_reserve which would flush the inode.
213 * But there's no guarantee that the inode buffer has actually
214 * gone out yet (it's delwri). Plus the buffer could be pinned
215 * anyway if it's part of an inode in another recent
216 * transaction. So we play it safe and fire off the
217 * transaction anyway.
218 */
219 xfs_trans_ijoin(tp, ip);
220 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
221 xfs_trans_set_sync(tp);
222 error = _xfs_trans_commit(tp, 0, &log_flushed);
223
224 xfs_iunlock(ip, XFS_ILOCK_EXCL);
225 } else {
226 /*
227 * Timestamps/size haven't changed since last inode flush or
228 * inode transaction commit. That means either nothing got
229 * written or a transaction committed which caught the updates.
230 * If the latter happened and the transaction hasn't hit the
231 * disk yet, the inode will be still be pinned. If it is,
232 * force the log.
233 */
234 if (xfs_ipincount(ip)) {
235 error = _xfs_log_force_lsn(mp,
236 ip->i_itemp->ili_last_lsn,
237 XFS_LOG_SYNC, &log_flushed);
238 }
239 xfs_iunlock(ip, XFS_ILOCK_SHARED);
240 }
241
242 /*
243 * If we only have a single device, and the log force about was
244 * a no-op we might have to flush the data device cache here.
245 * This can only happen for fdatasync/O_DSYNC if we were overwriting
246 * an already allocated file and thus do not have any metadata to
247 * commit.
248 */
249 if ((mp->m_flags & XFS_MOUNT_BARRIER) &&
250 mp->m_logdev_targp == mp->m_ddev_targp &&
251 !XFS_IS_REALTIME_INODE(ip) &&
252 !log_flushed)
253 xfs_blkdev_issue_flush(mp->m_ddev_targp);
254
255 return -error;
256}
257
258STATIC ssize_t
259xfs_file_aio_read(
260 struct kiocb *iocb,
261 const struct iovec *iovp,
262 unsigned long nr_segs,
263 loff_t pos)
264{
265 struct file *file = iocb->ki_filp;
266 struct inode *inode = file->f_mapping->host;
267 struct xfs_inode *ip = XFS_I(inode);
268 struct xfs_mount *mp = ip->i_mount;
269 size_t size = 0;
270 ssize_t ret = 0;
271 int ioflags = 0;
272 xfs_fsize_t n;
273 unsigned long seg;
274
275 XFS_STATS_INC(xs_read_calls);
276
277 BUG_ON(iocb->ki_pos != pos);
278
279 if (unlikely(file->f_flags & O_DIRECT))
280 ioflags |= IO_ISDIRECT;
281 if (file->f_mode & FMODE_NOCMTIME)
282 ioflags |= IO_INVIS;
283
284 /* START copy & waste from filemap.c */
285 for (seg = 0; seg < nr_segs; seg++) {
286 const struct iovec *iv = &iovp[seg];
287
288 /*
289 * If any segment has a negative length, or the cumulative
290 * length ever wraps negative then return -EINVAL.
291 */
292 size += iv->iov_len;
293 if (unlikely((ssize_t)(size|iv->iov_len) < 0))
294 return XFS_ERROR(-EINVAL);
295 }
296 /* END copy & waste from filemap.c */
297
298 if (unlikely(ioflags & IO_ISDIRECT)) {
299 xfs_buftarg_t *target =
300 XFS_IS_REALTIME_INODE(ip) ?
301 mp->m_rtdev_targp : mp->m_ddev_targp;
302 if ((iocb->ki_pos & target->bt_smask) ||
303 (size & target->bt_smask)) {
304 if (iocb->ki_pos == ip->i_size)
305 return 0;
306 return -XFS_ERROR(EINVAL);
307 }
308 }
309
310 n = XFS_MAXIOFFSET(mp) - iocb->ki_pos;
311 if (n <= 0 || size == 0)
312 return 0;
313
314 if (n < size)
315 size = n;
316
317 if (XFS_FORCED_SHUTDOWN(mp))
318 return -EIO;
319
320 if (unlikely(ioflags & IO_ISDIRECT)) {
321 xfs_rw_ilock(ip, XFS_IOLOCK_EXCL);
322
323 if (inode->i_mapping->nrpages) {
324 ret = -xfs_flushinval_pages(ip,
325 (iocb->ki_pos & PAGE_CACHE_MASK),
326 -1, FI_REMAPF_LOCKED);
327 if (ret) {
328 xfs_rw_iunlock(ip, XFS_IOLOCK_EXCL);
329 return ret;
330 }
331 }
332 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
333 } else
334 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
335
336 trace_xfs_file_read(ip, size, iocb->ki_pos, ioflags);
337
338 ret = generic_file_aio_read(iocb, iovp, nr_segs, iocb->ki_pos);
339 if (ret > 0)
340 XFS_STATS_ADD(xs_read_bytes, ret);
341
342 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
343 return ret;
344}
345
346STATIC ssize_t
347xfs_file_splice_read(
348 struct file *infilp,
349 loff_t *ppos,
350 struct pipe_inode_info *pipe,
351 size_t count,
352 unsigned int flags)
353{
354 struct xfs_inode *ip = XFS_I(infilp->f_mapping->host);
355 int ioflags = 0;
356 ssize_t ret;
357
358 XFS_STATS_INC(xs_read_calls);
359
360 if (infilp->f_mode & FMODE_NOCMTIME)
361 ioflags |= IO_INVIS;
362
363 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
364 return -EIO;
365
366 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
367
368 trace_xfs_file_splice_read(ip, count, *ppos, ioflags);
369
370 ret = generic_file_splice_read(infilp, ppos, pipe, count, flags);
371 if (ret > 0)
372 XFS_STATS_ADD(xs_read_bytes, ret);
373
374 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
375 return ret;
376}
377
378STATIC void
379xfs_aio_write_isize_update(
380 struct inode *inode,
381 loff_t *ppos,
382 ssize_t bytes_written)
383{
384 struct xfs_inode *ip = XFS_I(inode);
385 xfs_fsize_t isize = i_size_read(inode);
386
387 if (bytes_written > 0)
388 XFS_STATS_ADD(xs_write_bytes, bytes_written);
389
390 if (unlikely(bytes_written < 0 && bytes_written != -EFAULT &&
391 *ppos > isize))
392 *ppos = isize;
393
394 if (*ppos > ip->i_size) {
395 xfs_rw_ilock(ip, XFS_ILOCK_EXCL);
396 if (*ppos > ip->i_size)
397 ip->i_size = *ppos;
398 xfs_rw_iunlock(ip, XFS_ILOCK_EXCL);
399 }
400}
401
402/*
403 * If this was a direct or synchronous I/O that failed (such as ENOSPC) then
404 * part of the I/O may have been written to disk before the error occurred. In
405 * this case the on-disk file size may have been adjusted beyond the in-memory
406 * file size and now needs to be truncated back.
407 */
408STATIC void
409xfs_aio_write_newsize_update(
410 struct xfs_inode *ip)
411{
412 if (ip->i_new_size) {
413 xfs_rw_ilock(ip, XFS_ILOCK_EXCL);
414 ip->i_new_size = 0;
415 if (ip->i_d.di_size > ip->i_size)
416 ip->i_d.di_size = ip->i_size;
417 xfs_rw_iunlock(ip, XFS_ILOCK_EXCL);
418 }
419}
420
421/*
422 * xfs_file_splice_write() does not use xfs_rw_ilock() because
423 * generic_file_splice_write() takes the i_mutex itself. This, in theory,
424 * couuld cause lock inversions between the aio_write path and the splice path
425 * if someone is doing concurrent splice(2) based writes and write(2) based
426 * writes to the same inode. The only real way to fix this is to re-implement
427 * the generic code here with correct locking orders.
428 */
429STATIC ssize_t
430xfs_file_splice_write(
431 struct pipe_inode_info *pipe,
432 struct file *outfilp,
433 loff_t *ppos,
434 size_t count,
435 unsigned int flags)
436{
437 struct inode *inode = outfilp->f_mapping->host;
438 struct xfs_inode *ip = XFS_I(inode);
439 xfs_fsize_t new_size;
440 int ioflags = 0;
441 ssize_t ret;
442
443 XFS_STATS_INC(xs_write_calls);
444
445 if (outfilp->f_mode & FMODE_NOCMTIME)
446 ioflags |= IO_INVIS;
447
448 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
449 return -EIO;
450
451 xfs_ilock(ip, XFS_IOLOCK_EXCL);
452
453 new_size = *ppos + count;
454
455 xfs_ilock(ip, XFS_ILOCK_EXCL);
456 if (new_size > ip->i_size)
457 ip->i_new_size = new_size;
458 xfs_iunlock(ip, XFS_ILOCK_EXCL);
459
460 trace_xfs_file_splice_write(ip, count, *ppos, ioflags);
461
462 ret = generic_file_splice_write(pipe, outfilp, ppos, count, flags);
463
464 xfs_aio_write_isize_update(inode, ppos, ret);
465 xfs_aio_write_newsize_update(ip);
466 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
467 return ret;
468}
469
470/*
471 * This routine is called to handle zeroing any space in the last
472 * block of the file that is beyond the EOF. We do this since the
473 * size is being increased without writing anything to that block
474 * and we don't want anyone to read the garbage on the disk.
475 */
476STATIC int /* error (positive) */
477xfs_zero_last_block(
478 xfs_inode_t *ip,
479 xfs_fsize_t offset,
480 xfs_fsize_t isize)
481{
482 xfs_fileoff_t last_fsb;
483 xfs_mount_t *mp = ip->i_mount;
484 int nimaps;
485 int zero_offset;
486 int zero_len;
487 int error = 0;
488 xfs_bmbt_irec_t imap;
489
490 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
491
492 zero_offset = XFS_B_FSB_OFFSET(mp, isize);
493 if (zero_offset == 0) {
494 /*
495 * There are no extra bytes in the last block on disk to
496 * zero, so return.
497 */
498 return 0;
499 }
500
501 last_fsb = XFS_B_TO_FSBT(mp, isize);
502 nimaps = 1;
503 error = xfs_bmapi(NULL, ip, last_fsb, 1, 0, NULL, 0, &imap,
504 &nimaps, NULL);
505 if (error) {
506 return error;
507 }
508 ASSERT(nimaps > 0);
509 /*
510 * If the block underlying isize is just a hole, then there
511 * is nothing to zero.
512 */
513 if (imap.br_startblock == HOLESTARTBLOCK) {
514 return 0;
515 }
516 /*
517 * Zero the part of the last block beyond the EOF, and write it
518 * out sync. We need to drop the ilock while we do this so we
519 * don't deadlock when the buffer cache calls back to us.
520 */
521 xfs_iunlock(ip, XFS_ILOCK_EXCL);
522
523 zero_len = mp->m_sb.sb_blocksize - zero_offset;
524 if (isize + zero_len > offset)
525 zero_len = offset - isize;
526 error = xfs_iozero(ip, isize, zero_len);
527
528 xfs_ilock(ip, XFS_ILOCK_EXCL);
529 ASSERT(error >= 0);
530 return error;
531}
532
533/*
534 * Zero any on disk space between the current EOF and the new,
535 * larger EOF. This handles the normal case of zeroing the remainder
536 * of the last block in the file and the unusual case of zeroing blocks
537 * out beyond the size of the file. This second case only happens
538 * with fixed size extents and when the system crashes before the inode
539 * size was updated but after blocks were allocated. If fill is set,
540 * then any holes in the range are filled and zeroed. If not, the holes
541 * are left alone as holes.
542 */
543
544int /* error (positive) */
545xfs_zero_eof(
546 xfs_inode_t *ip,
547 xfs_off_t offset, /* starting I/O offset */
548 xfs_fsize_t isize) /* current inode size */
549{
550 xfs_mount_t *mp = ip->i_mount;
551 xfs_fileoff_t start_zero_fsb;
552 xfs_fileoff_t end_zero_fsb;
553 xfs_fileoff_t zero_count_fsb;
554 xfs_fileoff_t last_fsb;
555 xfs_fileoff_t zero_off;
556 xfs_fsize_t zero_len;
557 int nimaps;
558 int error = 0;
559 xfs_bmbt_irec_t imap;
560
561 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_IOLOCK_EXCL));
562 ASSERT(offset > isize);
563
564 /*
565 * First handle zeroing the block on which isize resides.
566 * We only zero a part of that block so it is handled specially.
567 */
568 error = xfs_zero_last_block(ip, offset, isize);
569 if (error) {
570 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_IOLOCK_EXCL));
571 return error;
572 }
573
574 /*
575 * Calculate the range between the new size and the old
576 * where blocks needing to be zeroed may exist. To get the
577 * block where the last byte in the file currently resides,
578 * we need to subtract one from the size and truncate back
579 * to a block boundary. We subtract 1 in case the size is
580 * exactly on a block boundary.
581 */
582 last_fsb = isize ? XFS_B_TO_FSBT(mp, isize - 1) : (xfs_fileoff_t)-1;
583 start_zero_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
584 end_zero_fsb = XFS_B_TO_FSBT(mp, offset - 1);
585 ASSERT((xfs_sfiloff_t)last_fsb < (xfs_sfiloff_t)start_zero_fsb);
586 if (last_fsb == end_zero_fsb) {
587 /*
588 * The size was only incremented on its last block.
589 * We took care of that above, so just return.
590 */
591 return 0;
592 }
593
594 ASSERT(start_zero_fsb <= end_zero_fsb);
595 while (start_zero_fsb <= end_zero_fsb) {
596 nimaps = 1;
597 zero_count_fsb = end_zero_fsb - start_zero_fsb + 1;
598 error = xfs_bmapi(NULL, ip, start_zero_fsb, zero_count_fsb,
599 0, NULL, 0, &imap, &nimaps, NULL);
600 if (error) {
601 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_IOLOCK_EXCL));
602 return error;
603 }
604 ASSERT(nimaps > 0);
605
606 if (imap.br_state == XFS_EXT_UNWRITTEN ||
607 imap.br_startblock == HOLESTARTBLOCK) {
608 /*
609 * This loop handles initializing pages that were
610 * partially initialized by the code below this
611 * loop. It basically zeroes the part of the page
612 * that sits on a hole and sets the page as P_HOLE
613 * and calls remapf if it is a mapped file.
614 */
615 start_zero_fsb = imap.br_startoff + imap.br_blockcount;
616 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
617 continue;
618 }
619
620 /*
621 * There are blocks we need to zero.
622 * Drop the inode lock while we're doing the I/O.
623 * We'll still have the iolock to protect us.
624 */
625 xfs_iunlock(ip, XFS_ILOCK_EXCL);
626
627 zero_off = XFS_FSB_TO_B(mp, start_zero_fsb);
628 zero_len = XFS_FSB_TO_B(mp, imap.br_blockcount);
629
630 if ((zero_off + zero_len) > offset)
631 zero_len = offset - zero_off;
632
633 error = xfs_iozero(ip, zero_off, zero_len);
634 if (error) {
635 goto out_lock;
636 }
637
638 start_zero_fsb = imap.br_startoff + imap.br_blockcount;
639 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
640
641 xfs_ilock(ip, XFS_ILOCK_EXCL);
642 }
643
644 return 0;
645
646out_lock:
647 xfs_ilock(ip, XFS_ILOCK_EXCL);
648 ASSERT(error >= 0);
649 return error;
650}
651
652/*
653 * Common pre-write limit and setup checks.
654 *
655 * Returns with iolock held according to @iolock.
656 */
657STATIC ssize_t
658xfs_file_aio_write_checks(
659 struct file *file,
660 loff_t *pos,
661 size_t *count,
662 int *iolock)
663{
664 struct inode *inode = file->f_mapping->host;
665 struct xfs_inode *ip = XFS_I(inode);
666 xfs_fsize_t new_size;
667 int error = 0;
668
669 error = generic_write_checks(file, pos, count, S_ISBLK(inode->i_mode));
670 if (error) {
671 xfs_rw_iunlock(ip, XFS_ILOCK_EXCL | *iolock);
672 *iolock = 0;
673 return error;
674 }
675
676 new_size = *pos + *count;
677 if (new_size > ip->i_size)
678 ip->i_new_size = new_size;
679
680 if (likely(!(file->f_mode & FMODE_NOCMTIME)))
681 file_update_time(file);
682
683 /*
684 * If the offset is beyond the size of the file, we need to zero any
685 * blocks that fall between the existing EOF and the start of this
686 * write.
687 */
688 if (*pos > ip->i_size)
689 error = -xfs_zero_eof(ip, *pos, ip->i_size);
690
691 xfs_rw_iunlock(ip, XFS_ILOCK_EXCL);
692 if (error)
693 return error;
694
695 /*
696 * If we're writing the file then make sure to clear the setuid and
697 * setgid bits if the process is not being run by root. This keeps
698 * people from modifying setuid and setgid binaries.
699 */
700 return file_remove_suid(file);
701
702}
703
704/*
705 * xfs_file_dio_aio_write - handle direct IO writes
706 *
707 * Lock the inode appropriately to prepare for and issue a direct IO write.
708 * By separating it from the buffered write path we remove all the tricky to
709 * follow locking changes and looping.
710 *
711 * If there are cached pages or we're extending the file, we need IOLOCK_EXCL
712 * until we're sure the bytes at the new EOF have been zeroed and/or the cached
713 * pages are flushed out.
714 *
715 * In most cases the direct IO writes will be done holding IOLOCK_SHARED
716 * allowing them to be done in parallel with reads and other direct IO writes.
717 * However, if the IO is not aligned to filesystem blocks, the direct IO layer
718 * needs to do sub-block zeroing and that requires serialisation against other
719 * direct IOs to the same block. In this case we need to serialise the
720 * submission of the unaligned IOs so that we don't get racing block zeroing in
721 * the dio layer. To avoid the problem with aio, we also need to wait for
722 * outstanding IOs to complete so that unwritten extent conversion is completed
723 * before we try to map the overlapping block. This is currently implemented by
724 * hitting it with a big hammer (i.e. xfs_ioend_wait()).
725 *
726 * Returns with locks held indicated by @iolock and errors indicated by
727 * negative return values.
728 */
729STATIC ssize_t
730xfs_file_dio_aio_write(
731 struct kiocb *iocb,
732 const struct iovec *iovp,
733 unsigned long nr_segs,
734 loff_t pos,
735 size_t ocount,
736 int *iolock)
737{
738 struct file *file = iocb->ki_filp;
739 struct address_space *mapping = file->f_mapping;
740 struct inode *inode = mapping->host;
741 struct xfs_inode *ip = XFS_I(inode);
742 struct xfs_mount *mp = ip->i_mount;
743 ssize_t ret = 0;
744 size_t count = ocount;
745 int unaligned_io = 0;
746 struct xfs_buftarg *target = XFS_IS_REALTIME_INODE(ip) ?
747 mp->m_rtdev_targp : mp->m_ddev_targp;
748
749 *iolock = 0;
750 if ((pos & target->bt_smask) || (count & target->bt_smask))
751 return -XFS_ERROR(EINVAL);
752
753 if ((pos & mp->m_blockmask) || ((pos + count) & mp->m_blockmask))
754 unaligned_io = 1;
755
756 if (unaligned_io || mapping->nrpages || pos > ip->i_size)
757 *iolock = XFS_IOLOCK_EXCL;
758 else
759 *iolock = XFS_IOLOCK_SHARED;
760 xfs_rw_ilock(ip, XFS_ILOCK_EXCL | *iolock);
761
762 ret = xfs_file_aio_write_checks(file, &pos, &count, iolock);
763 if (ret)
764 return ret;
765
766 if (mapping->nrpages) {
767 WARN_ON(*iolock != XFS_IOLOCK_EXCL);
768 ret = -xfs_flushinval_pages(ip, (pos & PAGE_CACHE_MASK), -1,
769 FI_REMAPF_LOCKED);
770 if (ret)
771 return ret;
772 }
773
774 /*
775 * If we are doing unaligned IO, wait for all other IO to drain,
776 * otherwise demote the lock if we had to flush cached pages
777 */
778 if (unaligned_io)
779 xfs_ioend_wait(ip);
780 else if (*iolock == XFS_IOLOCK_EXCL) {
781 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
782 *iolock = XFS_IOLOCK_SHARED;
783 }
784
785 trace_xfs_file_direct_write(ip, count, iocb->ki_pos, 0);
786 ret = generic_file_direct_write(iocb, iovp,
787 &nr_segs, pos, &iocb->ki_pos, count, ocount);
788
789 /* No fallback to buffered IO on errors for XFS. */
790 ASSERT(ret < 0 || ret == count);
791 return ret;
792}
793
794STATIC ssize_t
795xfs_file_buffered_aio_write(
796 struct kiocb *iocb,
797 const struct iovec *iovp,
798 unsigned long nr_segs,
799 loff_t pos,
800 size_t ocount,
801 int *iolock)
802{
803 struct file *file = iocb->ki_filp;
804 struct address_space *mapping = file->f_mapping;
805 struct inode *inode = mapping->host;
806 struct xfs_inode *ip = XFS_I(inode);
807 ssize_t ret;
808 int enospc = 0;
809 size_t count = ocount;
810
811 *iolock = XFS_IOLOCK_EXCL;
812 xfs_rw_ilock(ip, XFS_ILOCK_EXCL | *iolock);
813
814 ret = xfs_file_aio_write_checks(file, &pos, &count, iolock);
815 if (ret)
816 return ret;
817
818 /* We can write back this queue in page reclaim */
819 current->backing_dev_info = mapping->backing_dev_info;
820
821write_retry:
822 trace_xfs_file_buffered_write(ip, count, iocb->ki_pos, 0);
823 ret = generic_file_buffered_write(iocb, iovp, nr_segs,
824 pos, &iocb->ki_pos, count, ret);
825 /*
826 * if we just got an ENOSPC, flush the inode now we aren't holding any
827 * page locks and retry *once*
828 */
829 if (ret == -ENOSPC && !enospc) {
830 ret = -xfs_flush_pages(ip, 0, -1, 0, FI_NONE);
831 if (ret)
832 return ret;
833 enospc = 1;
834 goto write_retry;
835 }
836 current->backing_dev_info = NULL;
837 return ret;
838}
839
840STATIC ssize_t
841xfs_file_aio_write(
842 struct kiocb *iocb,
843 const struct iovec *iovp,
844 unsigned long nr_segs,
845 loff_t pos)
846{
847 struct file *file = iocb->ki_filp;
848 struct address_space *mapping = file->f_mapping;
849 struct inode *inode = mapping->host;
850 struct xfs_inode *ip = XFS_I(inode);
851 ssize_t ret;
852 int iolock;
853 size_t ocount = 0;
854
855 XFS_STATS_INC(xs_write_calls);
856
857 BUG_ON(iocb->ki_pos != pos);
858
859 ret = generic_segment_checks(iovp, &nr_segs, &ocount, VERIFY_READ);
860 if (ret)
861 return ret;
862
863 if (ocount == 0)
864 return 0;
865
866 xfs_wait_for_freeze(ip->i_mount, SB_FREEZE_WRITE);
867
868 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
869 return -EIO;
870
871 if (unlikely(file->f_flags & O_DIRECT))
872 ret = xfs_file_dio_aio_write(iocb, iovp, nr_segs, pos,
873 ocount, &iolock);
874 else
875 ret = xfs_file_buffered_aio_write(iocb, iovp, nr_segs, pos,
876 ocount, &iolock);
877
878 xfs_aio_write_isize_update(inode, &iocb->ki_pos, ret);
879
880 if (ret <= 0)
881 goto out_unlock;
882
883 /* Handle various SYNC-type writes */
884 if ((file->f_flags & O_DSYNC) || IS_SYNC(inode)) {
885 loff_t end = pos + ret - 1;
886 int error;
887
888 xfs_rw_iunlock(ip, iolock);
889 error = xfs_file_fsync(file, pos, end,
890 (file->f_flags & __O_SYNC) ? 0 : 1);
891 xfs_rw_ilock(ip, iolock);
892 if (error)
893 ret = error;
894 }
895
896out_unlock:
897 xfs_aio_write_newsize_update(ip);
898 xfs_rw_iunlock(ip, iolock);
899 return ret;
900}
901
902STATIC long
903xfs_file_fallocate(
904 struct file *file,
905 int mode,
906 loff_t offset,
907 loff_t len)
908{
909 struct inode *inode = file->f_path.dentry->d_inode;
910 long error;
911 loff_t new_size = 0;
912 xfs_flock64_t bf;
913 xfs_inode_t *ip = XFS_I(inode);
914 int cmd = XFS_IOC_RESVSP;
915 int attr_flags = XFS_ATTR_NOLOCK;
916
917 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
918 return -EOPNOTSUPP;
919
920 bf.l_whence = 0;
921 bf.l_start = offset;
922 bf.l_len = len;
923
924 xfs_ilock(ip, XFS_IOLOCK_EXCL);
925
926 if (mode & FALLOC_FL_PUNCH_HOLE)
927 cmd = XFS_IOC_UNRESVSP;
928
929 /* check the new inode size is valid before allocating */
930 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
931 offset + len > i_size_read(inode)) {
932 new_size = offset + len;
933 error = inode_newsize_ok(inode, new_size);
934 if (error)
935 goto out_unlock;
936 }
937
938 if (file->f_flags & O_DSYNC)
939 attr_flags |= XFS_ATTR_SYNC;
940
941 error = -xfs_change_file_space(ip, cmd, &bf, 0, attr_flags);
942 if (error)
943 goto out_unlock;
944
945 /* Change file size if needed */
946 if (new_size) {
947 struct iattr iattr;
948
949 iattr.ia_valid = ATTR_SIZE;
950 iattr.ia_size = new_size;
951 error = -xfs_setattr_size(ip, &iattr, XFS_ATTR_NOLOCK);
952 }
953
954out_unlock:
955 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
956 return error;
957}
958
959
960STATIC int
961xfs_file_open(
962 struct inode *inode,
963 struct file *file)
964{
965 if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
966 return -EFBIG;
967 if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb)))
968 return -EIO;
969 return 0;
970}
971
972STATIC int
973xfs_dir_open(
974 struct inode *inode,
975 struct file *file)
976{
977 struct xfs_inode *ip = XFS_I(inode);
978 int mode;
979 int error;
980
981 error = xfs_file_open(inode, file);
982 if (error)
983 return error;
984
985 /*
986 * If there are any blocks, read-ahead block 0 as we're almost
987 * certain to have the next operation be a read there.
988 */
989 mode = xfs_ilock_map_shared(ip);
990 if (ip->i_d.di_nextents > 0)
991 xfs_da_reada_buf(NULL, ip, 0, XFS_DATA_FORK);
992 xfs_iunlock(ip, mode);
993 return 0;
994}
995
996STATIC int
997xfs_file_release(
998 struct inode *inode,
999 struct file *filp)
1000{
1001 return -xfs_release(XFS_I(inode));
1002}
1003
1004STATIC int
1005xfs_file_readdir(
1006 struct file *filp,
1007 void *dirent,
1008 filldir_t filldir)
1009{
1010 struct inode *inode = filp->f_path.dentry->d_inode;
1011 xfs_inode_t *ip = XFS_I(inode);
1012 int error;
1013 size_t bufsize;
1014
1015 /*
1016 * The Linux API doesn't pass down the total size of the buffer
1017 * we read into down to the filesystem. With the filldir concept
1018 * it's not needed for correct information, but the XFS dir2 leaf
1019 * code wants an estimate of the buffer size to calculate it's
1020 * readahead window and size the buffers used for mapping to
1021 * physical blocks.
1022 *
1023 * Try to give it an estimate that's good enough, maybe at some
1024 * point we can change the ->readdir prototype to include the
1025 * buffer size. For now we use the current glibc buffer size.
1026 */
1027 bufsize = (size_t)min_t(loff_t, 32768, ip->i_d.di_size);
1028
1029 error = xfs_readdir(ip, dirent, bufsize,
1030 (xfs_off_t *)&filp->f_pos, filldir);
1031 if (error)
1032 return -error;
1033 return 0;
1034}
1035
1036STATIC int
1037xfs_file_mmap(
1038 struct file *filp,
1039 struct vm_area_struct *vma)
1040{
1041 vma->vm_ops = &xfs_file_vm_ops;
1042 vma->vm_flags |= VM_CAN_NONLINEAR;
1043
1044 file_accessed(filp);
1045 return 0;
1046}
1047
1048/*
1049 * mmap()d file has taken write protection fault and is being made
1050 * writable. We can set the page state up correctly for a writable
1051 * page, which means we can do correct delalloc accounting (ENOSPC
1052 * checking!) and unwritten extent mapping.
1053 */
1054STATIC int
1055xfs_vm_page_mkwrite(
1056 struct vm_area_struct *vma,
1057 struct vm_fault *vmf)
1058{
1059 return block_page_mkwrite(vma, vmf, xfs_get_blocks);
1060}
1061
1062const struct file_operations xfs_file_operations = {
1063 .llseek = generic_file_llseek,
1064 .read = do_sync_read,
1065 .write = do_sync_write,
1066 .aio_read = xfs_file_aio_read,
1067 .aio_write = xfs_file_aio_write,
1068 .splice_read = xfs_file_splice_read,
1069 .splice_write = xfs_file_splice_write,
1070 .unlocked_ioctl = xfs_file_ioctl,
1071#ifdef CONFIG_COMPAT
1072 .compat_ioctl = xfs_file_compat_ioctl,
1073#endif
1074 .mmap = xfs_file_mmap,
1075 .open = xfs_file_open,
1076 .release = xfs_file_release,
1077 .fsync = xfs_file_fsync,
1078 .fallocate = xfs_file_fallocate,
1079};
1080
1081const struct file_operations xfs_dir_file_operations = {
1082 .open = xfs_dir_open,
1083 .read = generic_read_dir,
1084 .readdir = xfs_file_readdir,
1085 .llseek = generic_file_llseek,
1086 .unlocked_ioctl = xfs_file_ioctl,
1087#ifdef CONFIG_COMPAT
1088 .compat_ioctl = xfs_file_compat_ioctl,
1089#endif
1090 .fsync = xfs_file_fsync,
1091};
1092
1093static const struct vm_operations_struct xfs_file_vm_ops = {
1094 .fault = filemap_fault,
1095 .page_mkwrite = xfs_vm_page_mkwrite,
1096};
1/*
2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
4 *
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.
8 *
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.
13 *
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
17 */
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_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"
30#include "xfs_bmap.h"
31#include "xfs_bmap_util.h"
32#include "xfs_error.h"
33#include "xfs_dir2.h"
34#include "xfs_dir2_priv.h"
35#include "xfs_ioctl.h"
36#include "xfs_trace.h"
37#include "xfs_log.h"
38#include "xfs_icache.h"
39#include "xfs_pnfs.h"
40
41#include <linux/dcache.h>
42#include <linux/falloc.h>
43#include <linux/pagevec.h>
44#include <linux/backing-dev.h>
45
46static const struct vm_operations_struct xfs_file_vm_ops;
47
48/*
49 * Locking primitives for read and write IO paths to ensure we consistently use
50 * and order the inode->i_mutex, ip->i_lock and ip->i_iolock.
51 */
52static inline void
53xfs_rw_ilock(
54 struct xfs_inode *ip,
55 int type)
56{
57 if (type & XFS_IOLOCK_EXCL)
58 inode_lock(VFS_I(ip));
59 xfs_ilock(ip, type);
60}
61
62static inline void
63xfs_rw_iunlock(
64 struct xfs_inode *ip,
65 int type)
66{
67 xfs_iunlock(ip, type);
68 if (type & XFS_IOLOCK_EXCL)
69 inode_unlock(VFS_I(ip));
70}
71
72static inline void
73xfs_rw_ilock_demote(
74 struct xfs_inode *ip,
75 int type)
76{
77 xfs_ilock_demote(ip, type);
78 if (type & XFS_IOLOCK_EXCL)
79 inode_unlock(VFS_I(ip));
80}
81
82/*
83 * xfs_iozero clears the specified range supplied via the page cache (except in
84 * the DAX case). Writes through the page cache will allocate blocks over holes,
85 * though the callers usually map the holes first and avoid them. If a block is
86 * not completely zeroed, then it will be read from disk before being partially
87 * zeroed.
88 *
89 * In the DAX case, we can just directly write to the underlying pages. This
90 * will not allocate blocks, but will avoid holes and unwritten extents and so
91 * not do unnecessary work.
92 */
93int
94xfs_iozero(
95 struct xfs_inode *ip, /* inode */
96 loff_t pos, /* offset in file */
97 size_t count) /* size of data to zero */
98{
99 struct page *page;
100 struct address_space *mapping;
101 int status = 0;
102
103
104 mapping = VFS_I(ip)->i_mapping;
105 do {
106 unsigned offset, bytes;
107 void *fsdata;
108
109 offset = (pos & (PAGE_SIZE -1)); /* Within page */
110 bytes = PAGE_SIZE - offset;
111 if (bytes > count)
112 bytes = count;
113
114 if (IS_DAX(VFS_I(ip))) {
115 status = dax_zero_page_range(VFS_I(ip), pos, bytes,
116 xfs_get_blocks_direct);
117 if (status)
118 break;
119 } else {
120 status = pagecache_write_begin(NULL, mapping, pos, bytes,
121 AOP_FLAG_UNINTERRUPTIBLE,
122 &page, &fsdata);
123 if (status)
124 break;
125
126 zero_user(page, offset, bytes);
127
128 status = pagecache_write_end(NULL, mapping, pos, bytes,
129 bytes, page, fsdata);
130 WARN_ON(status <= 0); /* can't return less than zero! */
131 status = 0;
132 }
133 pos += bytes;
134 count -= bytes;
135 } while (count);
136
137 return status;
138}
139
140int
141xfs_update_prealloc_flags(
142 struct xfs_inode *ip,
143 enum xfs_prealloc_flags flags)
144{
145 struct xfs_trans *tp;
146 int error;
147
148 tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_WRITEID);
149 error = xfs_trans_reserve(tp, &M_RES(ip->i_mount)->tr_writeid, 0, 0);
150 if (error) {
151 xfs_trans_cancel(tp);
152 return error;
153 }
154
155 xfs_ilock(ip, XFS_ILOCK_EXCL);
156 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
157
158 if (!(flags & XFS_PREALLOC_INVISIBLE)) {
159 VFS_I(ip)->i_mode &= ~S_ISUID;
160 if (VFS_I(ip)->i_mode & S_IXGRP)
161 VFS_I(ip)->i_mode &= ~S_ISGID;
162 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
163 }
164
165 if (flags & XFS_PREALLOC_SET)
166 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
167 if (flags & XFS_PREALLOC_CLEAR)
168 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
169
170 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
171 if (flags & XFS_PREALLOC_SYNC)
172 xfs_trans_set_sync(tp);
173 return xfs_trans_commit(tp);
174}
175
176/*
177 * Fsync operations on directories are much simpler than on regular files,
178 * as there is no file data to flush, and thus also no need for explicit
179 * cache flush operations, and there are no non-transaction metadata updates
180 * on directories either.
181 */
182STATIC int
183xfs_dir_fsync(
184 struct file *file,
185 loff_t start,
186 loff_t end,
187 int datasync)
188{
189 struct xfs_inode *ip = XFS_I(file->f_mapping->host);
190 struct xfs_mount *mp = ip->i_mount;
191 xfs_lsn_t lsn = 0;
192
193 trace_xfs_dir_fsync(ip);
194
195 xfs_ilock(ip, XFS_ILOCK_SHARED);
196 if (xfs_ipincount(ip))
197 lsn = ip->i_itemp->ili_last_lsn;
198 xfs_iunlock(ip, XFS_ILOCK_SHARED);
199
200 if (!lsn)
201 return 0;
202 return _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, NULL);
203}
204
205STATIC int
206xfs_file_fsync(
207 struct file *file,
208 loff_t start,
209 loff_t end,
210 int datasync)
211{
212 struct inode *inode = file->f_mapping->host;
213 struct xfs_inode *ip = XFS_I(inode);
214 struct xfs_mount *mp = ip->i_mount;
215 int error = 0;
216 int log_flushed = 0;
217 xfs_lsn_t lsn = 0;
218
219 trace_xfs_file_fsync(ip);
220
221 error = filemap_write_and_wait_range(inode->i_mapping, start, end);
222 if (error)
223 return error;
224
225 if (XFS_FORCED_SHUTDOWN(mp))
226 return -EIO;
227
228 xfs_iflags_clear(ip, XFS_ITRUNCATED);
229
230 if (mp->m_flags & XFS_MOUNT_BARRIER) {
231 /*
232 * If we have an RT and/or log subvolume we need to make sure
233 * to flush the write cache the device used for file data
234 * first. This is to ensure newly written file data make
235 * it to disk before logging the new inode size in case of
236 * an extending write.
237 */
238 if (XFS_IS_REALTIME_INODE(ip))
239 xfs_blkdev_issue_flush(mp->m_rtdev_targp);
240 else if (mp->m_logdev_targp != mp->m_ddev_targp)
241 xfs_blkdev_issue_flush(mp->m_ddev_targp);
242 }
243
244 /*
245 * All metadata updates are logged, which means that we just have to
246 * flush the log up to the latest LSN that touched the inode. If we have
247 * concurrent fsync/fdatasync() calls, we need them to all block on the
248 * log force before we clear the ili_fsync_fields field. This ensures
249 * that we don't get a racing sync operation that does not wait for the
250 * metadata to hit the journal before returning. If we race with
251 * clearing the ili_fsync_fields, then all that will happen is the log
252 * force will do nothing as the lsn will already be on disk. We can't
253 * race with setting ili_fsync_fields because that is done under
254 * XFS_ILOCK_EXCL, and that can't happen because we hold the lock shared
255 * until after the ili_fsync_fields is cleared.
256 */
257 xfs_ilock(ip, XFS_ILOCK_SHARED);
258 if (xfs_ipincount(ip)) {
259 if (!datasync ||
260 (ip->i_itemp->ili_fsync_fields & ~XFS_ILOG_TIMESTAMP))
261 lsn = ip->i_itemp->ili_last_lsn;
262 }
263
264 if (lsn) {
265 error = _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, &log_flushed);
266 ip->i_itemp->ili_fsync_fields = 0;
267 }
268 xfs_iunlock(ip, XFS_ILOCK_SHARED);
269
270 /*
271 * If we only have a single device, and the log force about was
272 * a no-op we might have to flush the data device cache here.
273 * This can only happen for fdatasync/O_DSYNC if we were overwriting
274 * an already allocated file and thus do not have any metadata to
275 * commit.
276 */
277 if ((mp->m_flags & XFS_MOUNT_BARRIER) &&
278 mp->m_logdev_targp == mp->m_ddev_targp &&
279 !XFS_IS_REALTIME_INODE(ip) &&
280 !log_flushed)
281 xfs_blkdev_issue_flush(mp->m_ddev_targp);
282
283 return error;
284}
285
286STATIC ssize_t
287xfs_file_read_iter(
288 struct kiocb *iocb,
289 struct iov_iter *to)
290{
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;
295 size_t size = iov_iter_count(to);
296 ssize_t ret = 0;
297 int ioflags = 0;
298 xfs_fsize_t n;
299 loff_t pos = iocb->ki_pos;
300
301 XFS_STATS_INC(mp, xs_read_calls);
302
303 if (unlikely(iocb->ki_flags & IOCB_DIRECT))
304 ioflags |= XFS_IO_ISDIRECT;
305 if (file->f_mode & FMODE_NOCMTIME)
306 ioflags |= XFS_IO_INVIS;
307
308 if ((ioflags & XFS_IO_ISDIRECT) && !IS_DAX(inode)) {
309 xfs_buftarg_t *target =
310 XFS_IS_REALTIME_INODE(ip) ?
311 mp->m_rtdev_targp : mp->m_ddev_targp;
312 /* DIO must be aligned to device logical sector size */
313 if ((pos | size) & target->bt_logical_sectormask) {
314 if (pos == i_size_read(inode))
315 return 0;
316 return -EINVAL;
317 }
318 }
319
320 n = mp->m_super->s_maxbytes - pos;
321 if (n <= 0 || size == 0)
322 return 0;
323
324 if (n < size)
325 size = n;
326
327 if (XFS_FORCED_SHUTDOWN(mp))
328 return -EIO;
329
330 /*
331 * Locking is a bit tricky here. If we take an exclusive lock for direct
332 * IO, we effectively serialise all new concurrent read IO to this file
333 * and block it behind IO that is currently in progress because IO in
334 * progress holds the IO lock shared. We only need to hold the lock
335 * exclusive to blow away the page cache, so only take lock exclusively
336 * if the page cache needs invalidation. This allows the normal direct
337 * IO case of no page cache pages to proceeed concurrently without
338 * serialisation.
339 */
340 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
341 if ((ioflags & XFS_IO_ISDIRECT) && inode->i_mapping->nrpages) {
342 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
343 xfs_rw_ilock(ip, XFS_IOLOCK_EXCL);
344
345 /*
346 * The generic dio code only flushes the range of the particular
347 * I/O. Because we take an exclusive lock here, this whole
348 * sequence is considerably more expensive for us. This has a
349 * noticeable performance impact for any file with cached pages,
350 * even when outside of the range of the particular I/O.
351 *
352 * Hence, amortize the cost of the lock against a full file
353 * flush and reduce the chances of repeated iolock cycles going
354 * forward.
355 */
356 if (inode->i_mapping->nrpages) {
357 ret = filemap_write_and_wait(VFS_I(ip)->i_mapping);
358 if (ret) {
359 xfs_rw_iunlock(ip, XFS_IOLOCK_EXCL);
360 return ret;
361 }
362
363 /*
364 * Invalidate whole pages. This can return an error if
365 * we fail to invalidate a page, but this should never
366 * happen on XFS. Warn if it does fail.
367 */
368 ret = invalidate_inode_pages2(VFS_I(ip)->i_mapping);
369 WARN_ON_ONCE(ret);
370 ret = 0;
371 }
372 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
373 }
374
375 trace_xfs_file_read(ip, size, pos, ioflags);
376
377 ret = generic_file_read_iter(iocb, to);
378 if (ret > 0)
379 XFS_STATS_ADD(mp, xs_read_bytes, ret);
380
381 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
382 return ret;
383}
384
385STATIC ssize_t
386xfs_file_splice_read(
387 struct file *infilp,
388 loff_t *ppos,
389 struct pipe_inode_info *pipe,
390 size_t count,
391 unsigned int flags)
392{
393 struct xfs_inode *ip = XFS_I(infilp->f_mapping->host);
394 int ioflags = 0;
395 ssize_t ret;
396
397 XFS_STATS_INC(ip->i_mount, xs_read_calls);
398
399 if (infilp->f_mode & FMODE_NOCMTIME)
400 ioflags |= XFS_IO_INVIS;
401
402 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
403 return -EIO;
404
405 trace_xfs_file_splice_read(ip, count, *ppos, ioflags);
406
407 /*
408 * DAX inodes cannot ues the page cache for splice, so we have to push
409 * them through the VFS IO path. This means it goes through
410 * ->read_iter, which for us takes the XFS_IOLOCK_SHARED. Hence we
411 * cannot lock the splice operation at this level for DAX inodes.
412 */
413 if (IS_DAX(VFS_I(ip))) {
414 ret = default_file_splice_read(infilp, ppos, pipe, count,
415 flags);
416 goto out;
417 }
418
419 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
420 ret = generic_file_splice_read(infilp, ppos, pipe, count, flags);
421 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
422out:
423 if (ret > 0)
424 XFS_STATS_ADD(ip->i_mount, xs_read_bytes, ret);
425 return ret;
426}
427
428/*
429 * This routine is called to handle zeroing any space in the last block of the
430 * file that is beyond the EOF. We do this since the size is being increased
431 * without writing anything to that block and we don't want to read the
432 * garbage on the disk.
433 */
434STATIC int /* error (positive) */
435xfs_zero_last_block(
436 struct xfs_inode *ip,
437 xfs_fsize_t offset,
438 xfs_fsize_t isize,
439 bool *did_zeroing)
440{
441 struct xfs_mount *mp = ip->i_mount;
442 xfs_fileoff_t last_fsb = XFS_B_TO_FSBT(mp, isize);
443 int zero_offset = XFS_B_FSB_OFFSET(mp, isize);
444 int zero_len;
445 int nimaps = 1;
446 int error = 0;
447 struct xfs_bmbt_irec imap;
448
449 xfs_ilock(ip, XFS_ILOCK_EXCL);
450 error = xfs_bmapi_read(ip, last_fsb, 1, &imap, &nimaps, 0);
451 xfs_iunlock(ip, XFS_ILOCK_EXCL);
452 if (error)
453 return error;
454
455 ASSERT(nimaps > 0);
456
457 /*
458 * If the block underlying isize is just a hole, then there
459 * is nothing to zero.
460 */
461 if (imap.br_startblock == HOLESTARTBLOCK)
462 return 0;
463
464 zero_len = mp->m_sb.sb_blocksize - zero_offset;
465 if (isize + zero_len > offset)
466 zero_len = offset - isize;
467 *did_zeroing = true;
468 return xfs_iozero(ip, isize, zero_len);
469}
470
471/*
472 * Zero any on disk space between the current EOF and the new, larger EOF.
473 *
474 * This handles the normal case of zeroing the remainder of the last block in
475 * the file and the unusual case of zeroing blocks out beyond the size of the
476 * file. This second case only happens with fixed size extents and when the
477 * system crashes before the inode size was updated but after blocks were
478 * allocated.
479 *
480 * Expects the iolock to be held exclusive, and will take the ilock internally.
481 */
482int /* error (positive) */
483xfs_zero_eof(
484 struct xfs_inode *ip,
485 xfs_off_t offset, /* starting I/O offset */
486 xfs_fsize_t isize, /* current inode size */
487 bool *did_zeroing)
488{
489 struct xfs_mount *mp = ip->i_mount;
490 xfs_fileoff_t start_zero_fsb;
491 xfs_fileoff_t end_zero_fsb;
492 xfs_fileoff_t zero_count_fsb;
493 xfs_fileoff_t last_fsb;
494 xfs_fileoff_t zero_off;
495 xfs_fsize_t zero_len;
496 int nimaps;
497 int error = 0;
498 struct xfs_bmbt_irec imap;
499
500 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
501 ASSERT(offset > isize);
502
503 trace_xfs_zero_eof(ip, isize, offset - isize);
504
505 /*
506 * First handle zeroing the block on which isize resides.
507 *
508 * We only zero a part of that block so it is handled specially.
509 */
510 if (XFS_B_FSB_OFFSET(mp, isize) != 0) {
511 error = xfs_zero_last_block(ip, offset, isize, did_zeroing);
512 if (error)
513 return error;
514 }
515
516 /*
517 * Calculate the range between the new size and the old where blocks
518 * needing to be zeroed may exist.
519 *
520 * To get the block where the last byte in the file currently resides,
521 * we need to subtract one from the size and truncate back to a block
522 * boundary. We subtract 1 in case the size is exactly on a block
523 * boundary.
524 */
525 last_fsb = isize ? XFS_B_TO_FSBT(mp, isize - 1) : (xfs_fileoff_t)-1;
526 start_zero_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
527 end_zero_fsb = XFS_B_TO_FSBT(mp, offset - 1);
528 ASSERT((xfs_sfiloff_t)last_fsb < (xfs_sfiloff_t)start_zero_fsb);
529 if (last_fsb == end_zero_fsb) {
530 /*
531 * The size was only incremented on its last block.
532 * We took care of that above, so just return.
533 */
534 return 0;
535 }
536
537 ASSERT(start_zero_fsb <= end_zero_fsb);
538 while (start_zero_fsb <= end_zero_fsb) {
539 nimaps = 1;
540 zero_count_fsb = end_zero_fsb - start_zero_fsb + 1;
541
542 xfs_ilock(ip, XFS_ILOCK_EXCL);
543 error = xfs_bmapi_read(ip, start_zero_fsb, zero_count_fsb,
544 &imap, &nimaps, 0);
545 xfs_iunlock(ip, XFS_ILOCK_EXCL);
546 if (error)
547 return error;
548
549 ASSERT(nimaps > 0);
550
551 if (imap.br_state == XFS_EXT_UNWRITTEN ||
552 imap.br_startblock == HOLESTARTBLOCK) {
553 start_zero_fsb = imap.br_startoff + imap.br_blockcount;
554 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
555 continue;
556 }
557
558 /*
559 * There are blocks we need to zero.
560 */
561 zero_off = XFS_FSB_TO_B(mp, start_zero_fsb);
562 zero_len = XFS_FSB_TO_B(mp, imap.br_blockcount);
563
564 if ((zero_off + zero_len) > offset)
565 zero_len = offset - zero_off;
566
567 error = xfs_iozero(ip, zero_off, zero_len);
568 if (error)
569 return error;
570
571 *did_zeroing = true;
572 start_zero_fsb = imap.br_startoff + imap.br_blockcount;
573 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
574 }
575
576 return 0;
577}
578
579/*
580 * Common pre-write limit and setup checks.
581 *
582 * Called with the iolocked held either shared and exclusive according to
583 * @iolock, and returns with it held. Might upgrade the iolock to exclusive
584 * if called for a direct write beyond i_size.
585 */
586STATIC ssize_t
587xfs_file_aio_write_checks(
588 struct kiocb *iocb,
589 struct iov_iter *from,
590 int *iolock)
591{
592 struct file *file = iocb->ki_filp;
593 struct inode *inode = file->f_mapping->host;
594 struct xfs_inode *ip = XFS_I(inode);
595 ssize_t error = 0;
596 size_t count = iov_iter_count(from);
597 bool drained_dio = false;
598
599restart:
600 error = generic_write_checks(iocb, from);
601 if (error <= 0)
602 return error;
603
604 error = xfs_break_layouts(inode, iolock, true);
605 if (error)
606 return error;
607
608 /* For changing security info in file_remove_privs() we need i_mutex */
609 if (*iolock == XFS_IOLOCK_SHARED && !IS_NOSEC(inode)) {
610 xfs_rw_iunlock(ip, *iolock);
611 *iolock = XFS_IOLOCK_EXCL;
612 xfs_rw_ilock(ip, *iolock);
613 goto restart;
614 }
615 /*
616 * If the offset is beyond the size of the file, we need to zero any
617 * blocks that fall between the existing EOF and the start of this
618 * write. If zeroing is needed and we are currently holding the
619 * iolock shared, we need to update it to exclusive which implies
620 * having to redo all checks before.
621 *
622 * We need to serialise against EOF updates that occur in IO
623 * completions here. We want to make sure that nobody is changing the
624 * size while we do this check until we have placed an IO barrier (i.e.
625 * hold the XFS_IOLOCK_EXCL) that prevents new IO from being dispatched.
626 * The spinlock effectively forms a memory barrier once we have the
627 * XFS_IOLOCK_EXCL so we are guaranteed to see the latest EOF value
628 * and hence be able to correctly determine if we need to run zeroing.
629 */
630 spin_lock(&ip->i_flags_lock);
631 if (iocb->ki_pos > i_size_read(inode)) {
632 bool zero = false;
633
634 spin_unlock(&ip->i_flags_lock);
635 if (!drained_dio) {
636 if (*iolock == XFS_IOLOCK_SHARED) {
637 xfs_rw_iunlock(ip, *iolock);
638 *iolock = XFS_IOLOCK_EXCL;
639 xfs_rw_ilock(ip, *iolock);
640 iov_iter_reexpand(from, count);
641 }
642 /*
643 * We now have an IO submission barrier in place, but
644 * AIO can do EOF updates during IO completion and hence
645 * we now need to wait for all of them to drain. Non-AIO
646 * DIO will have drained before we are given the
647 * XFS_IOLOCK_EXCL, and so for most cases this wait is a
648 * no-op.
649 */
650 inode_dio_wait(inode);
651 drained_dio = true;
652 goto restart;
653 }
654 error = xfs_zero_eof(ip, iocb->ki_pos, i_size_read(inode), &zero);
655 if (error)
656 return error;
657 } else
658 spin_unlock(&ip->i_flags_lock);
659
660 /*
661 * Updating the timestamps will grab the ilock again from
662 * xfs_fs_dirty_inode, so we have to call it after dropping the
663 * lock above. Eventually we should look into a way to avoid
664 * the pointless lock roundtrip.
665 */
666 if (likely(!(file->f_mode & FMODE_NOCMTIME))) {
667 error = file_update_time(file);
668 if (error)
669 return error;
670 }
671
672 /*
673 * If we're writing the file then make sure to clear the setuid and
674 * setgid bits if the process is not being run by root. This keeps
675 * people from modifying setuid and setgid binaries.
676 */
677 if (!IS_NOSEC(inode))
678 return file_remove_privs(file);
679 return 0;
680}
681
682/*
683 * xfs_file_dio_aio_write - handle direct IO writes
684 *
685 * Lock the inode appropriately to prepare for and issue a direct IO write.
686 * By separating it from the buffered write path we remove all the tricky to
687 * follow locking changes and looping.
688 *
689 * If there are cached pages or we're extending the file, we need IOLOCK_EXCL
690 * until we're sure the bytes at the new EOF have been zeroed and/or the cached
691 * pages are flushed out.
692 *
693 * In most cases the direct IO writes will be done holding IOLOCK_SHARED
694 * allowing them to be done in parallel with reads and other direct IO writes.
695 * However, if the IO is not aligned to filesystem blocks, the direct IO layer
696 * needs to do sub-block zeroing and that requires serialisation against other
697 * direct IOs to the same block. In this case we need to serialise the
698 * submission of the unaligned IOs so that we don't get racing block zeroing in
699 * the dio layer. To avoid the problem with aio, we also need to wait for
700 * outstanding IOs to complete so that unwritten extent conversion is completed
701 * before we try to map the overlapping block. This is currently implemented by
702 * hitting it with a big hammer (i.e. inode_dio_wait()).
703 *
704 * Returns with locks held indicated by @iolock and errors indicated by
705 * negative return values.
706 */
707STATIC ssize_t
708xfs_file_dio_aio_write(
709 struct kiocb *iocb,
710 struct iov_iter *from)
711{
712 struct file *file = iocb->ki_filp;
713 struct address_space *mapping = file->f_mapping;
714 struct inode *inode = mapping->host;
715 struct xfs_inode *ip = XFS_I(inode);
716 struct xfs_mount *mp = ip->i_mount;
717 ssize_t ret = 0;
718 int unaligned_io = 0;
719 int iolock;
720 size_t count = iov_iter_count(from);
721 loff_t pos = iocb->ki_pos;
722 loff_t end;
723 struct iov_iter data;
724 struct xfs_buftarg *target = XFS_IS_REALTIME_INODE(ip) ?
725 mp->m_rtdev_targp : mp->m_ddev_targp;
726
727 /* DIO must be aligned to device logical sector size */
728 if (!IS_DAX(inode) && ((pos | count) & target->bt_logical_sectormask))
729 return -EINVAL;
730
731 /* "unaligned" here means not aligned to a filesystem block */
732 if ((pos & mp->m_blockmask) || ((pos + count) & mp->m_blockmask))
733 unaligned_io = 1;
734
735 /*
736 * We don't need to take an exclusive lock unless there page cache needs
737 * to be invalidated or unaligned IO is being executed. We don't need to
738 * consider the EOF extension case here because
739 * xfs_file_aio_write_checks() will relock the inode as necessary for
740 * EOF zeroing cases and fill out the new inode size as appropriate.
741 */
742 if (unaligned_io || mapping->nrpages)
743 iolock = XFS_IOLOCK_EXCL;
744 else
745 iolock = XFS_IOLOCK_SHARED;
746 xfs_rw_ilock(ip, iolock);
747
748 /*
749 * Recheck if there are cached pages that need invalidate after we got
750 * the iolock to protect against other threads adding new pages while
751 * we were waiting for the iolock.
752 */
753 if (mapping->nrpages && iolock == XFS_IOLOCK_SHARED) {
754 xfs_rw_iunlock(ip, iolock);
755 iolock = XFS_IOLOCK_EXCL;
756 xfs_rw_ilock(ip, iolock);
757 }
758
759 ret = xfs_file_aio_write_checks(iocb, from, &iolock);
760 if (ret)
761 goto out;
762 count = iov_iter_count(from);
763 pos = iocb->ki_pos;
764 end = pos + count - 1;
765
766 /*
767 * See xfs_file_read_iter() for why we do a full-file flush here.
768 */
769 if (mapping->nrpages) {
770 ret = filemap_write_and_wait(VFS_I(ip)->i_mapping);
771 if (ret)
772 goto out;
773 /*
774 * Invalidate whole pages. This can return an error if we fail
775 * to invalidate a page, but this should never happen on XFS.
776 * Warn if it does fail.
777 */
778 ret = invalidate_inode_pages2(VFS_I(ip)->i_mapping);
779 WARN_ON_ONCE(ret);
780 ret = 0;
781 }
782
783 /*
784 * If we are doing unaligned IO, wait for all other IO to drain,
785 * otherwise demote the lock if we had to flush cached pages
786 */
787 if (unaligned_io)
788 inode_dio_wait(inode);
789 else if (iolock == XFS_IOLOCK_EXCL) {
790 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
791 iolock = XFS_IOLOCK_SHARED;
792 }
793
794 trace_xfs_file_direct_write(ip, count, iocb->ki_pos, 0);
795
796 data = *from;
797 ret = mapping->a_ops->direct_IO(iocb, &data, pos);
798
799 /* see generic_file_direct_write() for why this is necessary */
800 if (mapping->nrpages) {
801 invalidate_inode_pages2_range(mapping,
802 pos >> PAGE_SHIFT,
803 end >> PAGE_SHIFT);
804 }
805
806 if (ret > 0) {
807 pos += ret;
808 iov_iter_advance(from, ret);
809 iocb->ki_pos = pos;
810 }
811out:
812 xfs_rw_iunlock(ip, iolock);
813
814 /*
815 * No fallback to buffered IO on errors for XFS. DAX can result in
816 * partial writes, but direct IO will either complete fully or fail.
817 */
818 ASSERT(ret < 0 || ret == count || IS_DAX(VFS_I(ip)));
819 return ret;
820}
821
822STATIC ssize_t
823xfs_file_buffered_aio_write(
824 struct kiocb *iocb,
825 struct iov_iter *from)
826{
827 struct file *file = iocb->ki_filp;
828 struct address_space *mapping = file->f_mapping;
829 struct inode *inode = mapping->host;
830 struct xfs_inode *ip = XFS_I(inode);
831 ssize_t ret;
832 int enospc = 0;
833 int iolock = XFS_IOLOCK_EXCL;
834
835 xfs_rw_ilock(ip, iolock);
836
837 ret = xfs_file_aio_write_checks(iocb, from, &iolock);
838 if (ret)
839 goto out;
840
841 /* We can write back this queue in page reclaim */
842 current->backing_dev_info = inode_to_bdi(inode);
843
844write_retry:
845 trace_xfs_file_buffered_write(ip, iov_iter_count(from),
846 iocb->ki_pos, 0);
847 ret = generic_perform_write(file, from, iocb->ki_pos);
848 if (likely(ret >= 0))
849 iocb->ki_pos += ret;
850
851 /*
852 * If we hit a space limit, try to free up some lingering preallocated
853 * space before returning an error. In the case of ENOSPC, first try to
854 * write back all dirty inodes to free up some of the excess reserved
855 * metadata space. This reduces the chances that the eofblocks scan
856 * waits on dirty mappings. Since xfs_flush_inodes() is serialized, this
857 * also behaves as a filter to prevent too many eofblocks scans from
858 * running at the same time.
859 */
860 if (ret == -EDQUOT && !enospc) {
861 enospc = xfs_inode_free_quota_eofblocks(ip);
862 if (enospc)
863 goto write_retry;
864 } else if (ret == -ENOSPC && !enospc) {
865 struct xfs_eofblocks eofb = {0};
866
867 enospc = 1;
868 xfs_flush_inodes(ip->i_mount);
869 eofb.eof_scan_owner = ip->i_ino; /* for locking */
870 eofb.eof_flags = XFS_EOF_FLAGS_SYNC;
871 xfs_icache_free_eofblocks(ip->i_mount, &eofb);
872 goto write_retry;
873 }
874
875 current->backing_dev_info = NULL;
876out:
877 xfs_rw_iunlock(ip, iolock);
878 return ret;
879}
880
881STATIC ssize_t
882xfs_file_write_iter(
883 struct kiocb *iocb,
884 struct iov_iter *from)
885{
886 struct file *file = iocb->ki_filp;
887 struct address_space *mapping = file->f_mapping;
888 struct inode *inode = mapping->host;
889 struct xfs_inode *ip = XFS_I(inode);
890 ssize_t ret;
891 size_t ocount = iov_iter_count(from);
892
893 XFS_STATS_INC(ip->i_mount, xs_write_calls);
894
895 if (ocount == 0)
896 return 0;
897
898 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
899 return -EIO;
900
901 if ((iocb->ki_flags & IOCB_DIRECT) || IS_DAX(inode))
902 ret = xfs_file_dio_aio_write(iocb, from);
903 else
904 ret = xfs_file_buffered_aio_write(iocb, from);
905
906 if (ret > 0) {
907 ssize_t err;
908
909 XFS_STATS_ADD(ip->i_mount, xs_write_bytes, ret);
910
911 /* Handle various SYNC-type writes */
912 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
913 if (err < 0)
914 ret = err;
915 }
916 return ret;
917}
918
919#define XFS_FALLOC_FL_SUPPORTED \
920 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
921 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE | \
922 FALLOC_FL_INSERT_RANGE)
923
924STATIC long
925xfs_file_fallocate(
926 struct file *file,
927 int mode,
928 loff_t offset,
929 loff_t len)
930{
931 struct inode *inode = file_inode(file);
932 struct xfs_inode *ip = XFS_I(inode);
933 long error;
934 enum xfs_prealloc_flags flags = 0;
935 uint iolock = XFS_IOLOCK_EXCL;
936 loff_t new_size = 0;
937 bool do_file_insert = 0;
938
939 if (!S_ISREG(inode->i_mode))
940 return -EINVAL;
941 if (mode & ~XFS_FALLOC_FL_SUPPORTED)
942 return -EOPNOTSUPP;
943
944 xfs_ilock(ip, iolock);
945 error = xfs_break_layouts(inode, &iolock, false);
946 if (error)
947 goto out_unlock;
948
949 xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
950 iolock |= XFS_MMAPLOCK_EXCL;
951
952 if (mode & FALLOC_FL_PUNCH_HOLE) {
953 error = xfs_free_file_space(ip, offset, len);
954 if (error)
955 goto out_unlock;
956 } else if (mode & FALLOC_FL_COLLAPSE_RANGE) {
957 unsigned blksize_mask = (1 << inode->i_blkbits) - 1;
958
959 if (offset & blksize_mask || len & blksize_mask) {
960 error = -EINVAL;
961 goto out_unlock;
962 }
963
964 /*
965 * There is no need to overlap collapse range with EOF,
966 * in which case it is effectively a truncate operation
967 */
968 if (offset + len >= i_size_read(inode)) {
969 error = -EINVAL;
970 goto out_unlock;
971 }
972
973 new_size = i_size_read(inode) - len;
974
975 error = xfs_collapse_file_space(ip, offset, len);
976 if (error)
977 goto out_unlock;
978 } else if (mode & FALLOC_FL_INSERT_RANGE) {
979 unsigned blksize_mask = (1 << inode->i_blkbits) - 1;
980
981 new_size = i_size_read(inode) + len;
982 if (offset & blksize_mask || len & blksize_mask) {
983 error = -EINVAL;
984 goto out_unlock;
985 }
986
987 /* check the new inode size does not wrap through zero */
988 if (new_size > inode->i_sb->s_maxbytes) {
989 error = -EFBIG;
990 goto out_unlock;
991 }
992
993 /* Offset should be less than i_size */
994 if (offset >= i_size_read(inode)) {
995 error = -EINVAL;
996 goto out_unlock;
997 }
998 do_file_insert = 1;
999 } else {
1000 flags |= XFS_PREALLOC_SET;
1001
1002 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
1003 offset + len > i_size_read(inode)) {
1004 new_size = offset + len;
1005 error = inode_newsize_ok(inode, new_size);
1006 if (error)
1007 goto out_unlock;
1008 }
1009
1010 if (mode & FALLOC_FL_ZERO_RANGE)
1011 error = xfs_zero_file_space(ip, offset, len);
1012 else
1013 error = xfs_alloc_file_space(ip, offset, len,
1014 XFS_BMAPI_PREALLOC);
1015 if (error)
1016 goto out_unlock;
1017 }
1018
1019 if (file->f_flags & O_DSYNC)
1020 flags |= XFS_PREALLOC_SYNC;
1021
1022 error = xfs_update_prealloc_flags(ip, flags);
1023 if (error)
1024 goto out_unlock;
1025
1026 /* Change file size if needed */
1027 if (new_size) {
1028 struct iattr iattr;
1029
1030 iattr.ia_valid = ATTR_SIZE;
1031 iattr.ia_size = new_size;
1032 error = xfs_setattr_size(ip, &iattr);
1033 if (error)
1034 goto out_unlock;
1035 }
1036
1037 /*
1038 * Perform hole insertion now that the file size has been
1039 * updated so that if we crash during the operation we don't
1040 * leave shifted extents past EOF and hence losing access to
1041 * the data that is contained within them.
1042 */
1043 if (do_file_insert)
1044 error = xfs_insert_file_space(ip, offset, len);
1045
1046out_unlock:
1047 xfs_iunlock(ip, iolock);
1048 return error;
1049}
1050
1051
1052STATIC int
1053xfs_file_open(
1054 struct inode *inode,
1055 struct file *file)
1056{
1057 if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1058 return -EFBIG;
1059 if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb)))
1060 return -EIO;
1061 return 0;
1062}
1063
1064STATIC int
1065xfs_dir_open(
1066 struct inode *inode,
1067 struct file *file)
1068{
1069 struct xfs_inode *ip = XFS_I(inode);
1070 int mode;
1071 int error;
1072
1073 error = xfs_file_open(inode, file);
1074 if (error)
1075 return error;
1076
1077 /*
1078 * If there are any blocks, read-ahead block 0 as we're almost
1079 * certain to have the next operation be a read there.
1080 */
1081 mode = xfs_ilock_data_map_shared(ip);
1082 if (ip->i_d.di_nextents > 0)
1083 xfs_dir3_data_readahead(ip, 0, -1);
1084 xfs_iunlock(ip, mode);
1085 return 0;
1086}
1087
1088STATIC int
1089xfs_file_release(
1090 struct inode *inode,
1091 struct file *filp)
1092{
1093 return xfs_release(XFS_I(inode));
1094}
1095
1096STATIC int
1097xfs_file_readdir(
1098 struct file *file,
1099 struct dir_context *ctx)
1100{
1101 struct inode *inode = file_inode(file);
1102 xfs_inode_t *ip = XFS_I(inode);
1103 size_t bufsize;
1104
1105 /*
1106 * The Linux API doesn't pass down the total size of the buffer
1107 * we read into down to the filesystem. With the filldir concept
1108 * it's not needed for correct information, but the XFS dir2 leaf
1109 * code wants an estimate of the buffer size to calculate it's
1110 * readahead window and size the buffers used for mapping to
1111 * physical blocks.
1112 *
1113 * Try to give it an estimate that's good enough, maybe at some
1114 * point we can change the ->readdir prototype to include the
1115 * buffer size. For now we use the current glibc buffer size.
1116 */
1117 bufsize = (size_t)min_t(loff_t, 32768, ip->i_d.di_size);
1118
1119 return xfs_readdir(ip, ctx, bufsize);
1120}
1121
1122/*
1123 * This type is designed to indicate the type of offset we would like
1124 * to search from page cache for xfs_seek_hole_data().
1125 */
1126enum {
1127 HOLE_OFF = 0,
1128 DATA_OFF,
1129};
1130
1131/*
1132 * Lookup the desired type of offset from the given page.
1133 *
1134 * On success, return true and the offset argument will point to the
1135 * start of the region that was found. Otherwise this function will
1136 * return false and keep the offset argument unchanged.
1137 */
1138STATIC bool
1139xfs_lookup_buffer_offset(
1140 struct page *page,
1141 loff_t *offset,
1142 unsigned int type)
1143{
1144 loff_t lastoff = page_offset(page);
1145 bool found = false;
1146 struct buffer_head *bh, *head;
1147
1148 bh = head = page_buffers(page);
1149 do {
1150 /*
1151 * Unwritten extents that have data in the page
1152 * cache covering them can be identified by the
1153 * BH_Unwritten state flag. Pages with multiple
1154 * buffers might have a mix of holes, data and
1155 * unwritten extents - any buffer with valid
1156 * data in it should have BH_Uptodate flag set
1157 * on it.
1158 */
1159 if (buffer_unwritten(bh) ||
1160 buffer_uptodate(bh)) {
1161 if (type == DATA_OFF)
1162 found = true;
1163 } else {
1164 if (type == HOLE_OFF)
1165 found = true;
1166 }
1167
1168 if (found) {
1169 *offset = lastoff;
1170 break;
1171 }
1172 lastoff += bh->b_size;
1173 } while ((bh = bh->b_this_page) != head);
1174
1175 return found;
1176}
1177
1178/*
1179 * This routine is called to find out and return a data or hole offset
1180 * from the page cache for unwritten extents according to the desired
1181 * type for xfs_seek_hole_data().
1182 *
1183 * The argument offset is used to tell where we start to search from the
1184 * page cache. Map is used to figure out the end points of the range to
1185 * lookup pages.
1186 *
1187 * Return true if the desired type of offset was found, and the argument
1188 * offset is filled with that address. Otherwise, return false and keep
1189 * offset unchanged.
1190 */
1191STATIC bool
1192xfs_find_get_desired_pgoff(
1193 struct inode *inode,
1194 struct xfs_bmbt_irec *map,
1195 unsigned int type,
1196 loff_t *offset)
1197{
1198 struct xfs_inode *ip = XFS_I(inode);
1199 struct xfs_mount *mp = ip->i_mount;
1200 struct pagevec pvec;
1201 pgoff_t index;
1202 pgoff_t end;
1203 loff_t endoff;
1204 loff_t startoff = *offset;
1205 loff_t lastoff = startoff;
1206 bool found = false;
1207
1208 pagevec_init(&pvec, 0);
1209
1210 index = startoff >> PAGE_SHIFT;
1211 endoff = XFS_FSB_TO_B(mp, map->br_startoff + map->br_blockcount);
1212 end = endoff >> PAGE_SHIFT;
1213 do {
1214 int want;
1215 unsigned nr_pages;
1216 unsigned int i;
1217
1218 want = min_t(pgoff_t, end - index, PAGEVEC_SIZE);
1219 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index,
1220 want);
1221 /*
1222 * No page mapped into given range. If we are searching holes
1223 * and if this is the first time we got into the loop, it means
1224 * that the given offset is landed in a hole, return it.
1225 *
1226 * If we have already stepped through some block buffers to find
1227 * holes but they all contains data. In this case, the last
1228 * offset is already updated and pointed to the end of the last
1229 * mapped page, if it does not reach the endpoint to search,
1230 * that means there should be a hole between them.
1231 */
1232 if (nr_pages == 0) {
1233 /* Data search found nothing */
1234 if (type == DATA_OFF)
1235 break;
1236
1237 ASSERT(type == HOLE_OFF);
1238 if (lastoff == startoff || lastoff < endoff) {
1239 found = true;
1240 *offset = lastoff;
1241 }
1242 break;
1243 }
1244
1245 /*
1246 * At lease we found one page. If this is the first time we
1247 * step into the loop, and if the first page index offset is
1248 * greater than the given search offset, a hole was found.
1249 */
1250 if (type == HOLE_OFF && lastoff == startoff &&
1251 lastoff < page_offset(pvec.pages[0])) {
1252 found = true;
1253 break;
1254 }
1255
1256 for (i = 0; i < nr_pages; i++) {
1257 struct page *page = pvec.pages[i];
1258 loff_t b_offset;
1259
1260 /*
1261 * At this point, the page may be truncated or
1262 * invalidated (changing page->mapping to NULL),
1263 * or even swizzled back from swapper_space to tmpfs
1264 * file mapping. However, page->index will not change
1265 * because we have a reference on the page.
1266 *
1267 * Searching done if the page index is out of range.
1268 * If the current offset is not reaches the end of
1269 * the specified search range, there should be a hole
1270 * between them.
1271 */
1272 if (page->index > end) {
1273 if (type == HOLE_OFF && lastoff < endoff) {
1274 *offset = lastoff;
1275 found = true;
1276 }
1277 goto out;
1278 }
1279
1280 lock_page(page);
1281 /*
1282 * Page truncated or invalidated(page->mapping == NULL).
1283 * We can freely skip it and proceed to check the next
1284 * page.
1285 */
1286 if (unlikely(page->mapping != inode->i_mapping)) {
1287 unlock_page(page);
1288 continue;
1289 }
1290
1291 if (!page_has_buffers(page)) {
1292 unlock_page(page);
1293 continue;
1294 }
1295
1296 found = xfs_lookup_buffer_offset(page, &b_offset, type);
1297 if (found) {
1298 /*
1299 * The found offset may be less than the start
1300 * point to search if this is the first time to
1301 * come here.
1302 */
1303 *offset = max_t(loff_t, startoff, b_offset);
1304 unlock_page(page);
1305 goto out;
1306 }
1307
1308 /*
1309 * We either searching data but nothing was found, or
1310 * searching hole but found a data buffer. In either
1311 * case, probably the next page contains the desired
1312 * things, update the last offset to it so.
1313 */
1314 lastoff = page_offset(page) + PAGE_SIZE;
1315 unlock_page(page);
1316 }
1317
1318 /*
1319 * The number of returned pages less than our desired, search
1320 * done. In this case, nothing was found for searching data,
1321 * but we found a hole behind the last offset.
1322 */
1323 if (nr_pages < want) {
1324 if (type == HOLE_OFF) {
1325 *offset = lastoff;
1326 found = true;
1327 }
1328 break;
1329 }
1330
1331 index = pvec.pages[i - 1]->index + 1;
1332 pagevec_release(&pvec);
1333 } while (index <= end);
1334
1335out:
1336 pagevec_release(&pvec);
1337 return found;
1338}
1339
1340/*
1341 * caller must lock inode with xfs_ilock_data_map_shared,
1342 * can we craft an appropriate ASSERT?
1343 *
1344 * end is because the VFS-level lseek interface is defined such that any
1345 * offset past i_size shall return -ENXIO, but we use this for quota code
1346 * which does not maintain i_size, and we want to SEEK_DATA past i_size.
1347 */
1348loff_t
1349__xfs_seek_hole_data(
1350 struct inode *inode,
1351 loff_t start,
1352 loff_t end,
1353 int whence)
1354{
1355 struct xfs_inode *ip = XFS_I(inode);
1356 struct xfs_mount *mp = ip->i_mount;
1357 loff_t uninitialized_var(offset);
1358 xfs_fileoff_t fsbno;
1359 xfs_filblks_t lastbno;
1360 int error;
1361
1362 if (start >= end) {
1363 error = -ENXIO;
1364 goto out_error;
1365 }
1366
1367 /*
1368 * Try to read extents from the first block indicated
1369 * by fsbno to the end block of the file.
1370 */
1371 fsbno = XFS_B_TO_FSBT(mp, start);
1372 lastbno = XFS_B_TO_FSB(mp, end);
1373
1374 for (;;) {
1375 struct xfs_bmbt_irec map[2];
1376 int nmap = 2;
1377 unsigned int i;
1378
1379 error = xfs_bmapi_read(ip, fsbno, lastbno - fsbno, map, &nmap,
1380 XFS_BMAPI_ENTIRE);
1381 if (error)
1382 goto out_error;
1383
1384 /* No extents at given offset, must be beyond EOF */
1385 if (nmap == 0) {
1386 error = -ENXIO;
1387 goto out_error;
1388 }
1389
1390 for (i = 0; i < nmap; i++) {
1391 offset = max_t(loff_t, start,
1392 XFS_FSB_TO_B(mp, map[i].br_startoff));
1393
1394 /* Landed in the hole we wanted? */
1395 if (whence == SEEK_HOLE &&
1396 map[i].br_startblock == HOLESTARTBLOCK)
1397 goto out;
1398
1399 /* Landed in the data extent we wanted? */
1400 if (whence == SEEK_DATA &&
1401 (map[i].br_startblock == DELAYSTARTBLOCK ||
1402 (map[i].br_state == XFS_EXT_NORM &&
1403 !isnullstartblock(map[i].br_startblock))))
1404 goto out;
1405
1406 /*
1407 * Landed in an unwritten extent, try to search
1408 * for hole or data from page cache.
1409 */
1410 if (map[i].br_state == XFS_EXT_UNWRITTEN) {
1411 if (xfs_find_get_desired_pgoff(inode, &map[i],
1412 whence == SEEK_HOLE ? HOLE_OFF : DATA_OFF,
1413 &offset))
1414 goto out;
1415 }
1416 }
1417
1418 /*
1419 * We only received one extent out of the two requested. This
1420 * means we've hit EOF and didn't find what we are looking for.
1421 */
1422 if (nmap == 1) {
1423 /*
1424 * If we were looking for a hole, set offset to
1425 * the end of the file (i.e., there is an implicit
1426 * hole at the end of any file).
1427 */
1428 if (whence == SEEK_HOLE) {
1429 offset = end;
1430 break;
1431 }
1432 /*
1433 * If we were looking for data, it's nowhere to be found
1434 */
1435 ASSERT(whence == SEEK_DATA);
1436 error = -ENXIO;
1437 goto out_error;
1438 }
1439
1440 ASSERT(i > 1);
1441
1442 /*
1443 * Nothing was found, proceed to the next round of search
1444 * if the next reading offset is not at or beyond EOF.
1445 */
1446 fsbno = map[i - 1].br_startoff + map[i - 1].br_blockcount;
1447 start = XFS_FSB_TO_B(mp, fsbno);
1448 if (start >= end) {
1449 if (whence == SEEK_HOLE) {
1450 offset = end;
1451 break;
1452 }
1453 ASSERT(whence == SEEK_DATA);
1454 error = -ENXIO;
1455 goto out_error;
1456 }
1457 }
1458
1459out:
1460 /*
1461 * If at this point we have found the hole we wanted, the returned
1462 * offset may be bigger than the file size as it may be aligned to
1463 * page boundary for unwritten extents. We need to deal with this
1464 * situation in particular.
1465 */
1466 if (whence == SEEK_HOLE)
1467 offset = min_t(loff_t, offset, end);
1468
1469 return offset;
1470
1471out_error:
1472 return error;
1473}
1474
1475STATIC loff_t
1476xfs_seek_hole_data(
1477 struct file *file,
1478 loff_t start,
1479 int whence)
1480{
1481 struct inode *inode = file->f_mapping->host;
1482 struct xfs_inode *ip = XFS_I(inode);
1483 struct xfs_mount *mp = ip->i_mount;
1484 uint lock;
1485 loff_t offset, end;
1486 int error = 0;
1487
1488 if (XFS_FORCED_SHUTDOWN(mp))
1489 return -EIO;
1490
1491 lock = xfs_ilock_data_map_shared(ip);
1492
1493 end = i_size_read(inode);
1494 offset = __xfs_seek_hole_data(inode, start, end, whence);
1495 if (offset < 0) {
1496 error = offset;
1497 goto out_unlock;
1498 }
1499
1500 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
1501
1502out_unlock:
1503 xfs_iunlock(ip, lock);
1504
1505 if (error)
1506 return error;
1507 return offset;
1508}
1509
1510STATIC loff_t
1511xfs_file_llseek(
1512 struct file *file,
1513 loff_t offset,
1514 int whence)
1515{
1516 switch (whence) {
1517 case SEEK_END:
1518 case SEEK_CUR:
1519 case SEEK_SET:
1520 return generic_file_llseek(file, offset, whence);
1521 case SEEK_HOLE:
1522 case SEEK_DATA:
1523 return xfs_seek_hole_data(file, offset, whence);
1524 default:
1525 return -EINVAL;
1526 }
1527}
1528
1529/*
1530 * Locking for serialisation of IO during page faults. This results in a lock
1531 * ordering of:
1532 *
1533 * mmap_sem (MM)
1534 * sb_start_pagefault(vfs, freeze)
1535 * i_mmaplock (XFS - truncate serialisation)
1536 * page_lock (MM)
1537 * i_lock (XFS - extent map serialisation)
1538 */
1539
1540/*
1541 * mmap()d file has taken write protection fault and is being made writable. We
1542 * can set the page state up correctly for a writable page, which means we can
1543 * do correct delalloc accounting (ENOSPC checking!) and unwritten extent
1544 * mapping.
1545 */
1546STATIC int
1547xfs_filemap_page_mkwrite(
1548 struct vm_area_struct *vma,
1549 struct vm_fault *vmf)
1550{
1551 struct inode *inode = file_inode(vma->vm_file);
1552 int ret;
1553
1554 trace_xfs_filemap_page_mkwrite(XFS_I(inode));
1555
1556 sb_start_pagefault(inode->i_sb);
1557 file_update_time(vma->vm_file);
1558 xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1559
1560 if (IS_DAX(inode)) {
1561 ret = __dax_mkwrite(vma, vmf, xfs_get_blocks_dax_fault, NULL);
1562 } else {
1563 ret = block_page_mkwrite(vma, vmf, xfs_get_blocks);
1564 ret = block_page_mkwrite_return(ret);
1565 }
1566
1567 xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1568 sb_end_pagefault(inode->i_sb);
1569
1570 return ret;
1571}
1572
1573STATIC int
1574xfs_filemap_fault(
1575 struct vm_area_struct *vma,
1576 struct vm_fault *vmf)
1577{
1578 struct inode *inode = file_inode(vma->vm_file);
1579 int ret;
1580
1581 trace_xfs_filemap_fault(XFS_I(inode));
1582
1583 /* DAX can shortcut the normal fault path on write faults! */
1584 if ((vmf->flags & FAULT_FLAG_WRITE) && IS_DAX(inode))
1585 return xfs_filemap_page_mkwrite(vma, vmf);
1586
1587 xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1588 if (IS_DAX(inode)) {
1589 /*
1590 * we do not want to trigger unwritten extent conversion on read
1591 * faults - that is unnecessary overhead and would also require
1592 * changes to xfs_get_blocks_direct() to map unwritten extent
1593 * ioend for conversion on read-only mappings.
1594 */
1595 ret = __dax_fault(vma, vmf, xfs_get_blocks_dax_fault, NULL);
1596 } else
1597 ret = filemap_fault(vma, vmf);
1598 xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1599
1600 return ret;
1601}
1602
1603/*
1604 * Similar to xfs_filemap_fault(), the DAX fault path can call into here on
1605 * both read and write faults. Hence we need to handle both cases. There is no
1606 * ->pmd_mkwrite callout for huge pages, so we have a single function here to
1607 * handle both cases here. @flags carries the information on the type of fault
1608 * occuring.
1609 */
1610STATIC int
1611xfs_filemap_pmd_fault(
1612 struct vm_area_struct *vma,
1613 unsigned long addr,
1614 pmd_t *pmd,
1615 unsigned int flags)
1616{
1617 struct inode *inode = file_inode(vma->vm_file);
1618 struct xfs_inode *ip = XFS_I(inode);
1619 int ret;
1620
1621 if (!IS_DAX(inode))
1622 return VM_FAULT_FALLBACK;
1623
1624 trace_xfs_filemap_pmd_fault(ip);
1625
1626 if (flags & FAULT_FLAG_WRITE) {
1627 sb_start_pagefault(inode->i_sb);
1628 file_update_time(vma->vm_file);
1629 }
1630
1631 xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1632 ret = __dax_pmd_fault(vma, addr, pmd, flags, xfs_get_blocks_dax_fault,
1633 NULL);
1634 xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1635
1636 if (flags & FAULT_FLAG_WRITE)
1637 sb_end_pagefault(inode->i_sb);
1638
1639 return ret;
1640}
1641
1642/*
1643 * pfn_mkwrite was originally inteneded to ensure we capture time stamp
1644 * updates on write faults. In reality, it's need to serialise against
1645 * truncate similar to page_mkwrite. Hence we cycle the XFS_MMAPLOCK_SHARED
1646 * to ensure we serialise the fault barrier in place.
1647 */
1648static int
1649xfs_filemap_pfn_mkwrite(
1650 struct vm_area_struct *vma,
1651 struct vm_fault *vmf)
1652{
1653
1654 struct inode *inode = file_inode(vma->vm_file);
1655 struct xfs_inode *ip = XFS_I(inode);
1656 int ret = VM_FAULT_NOPAGE;
1657 loff_t size;
1658
1659 trace_xfs_filemap_pfn_mkwrite(ip);
1660
1661 sb_start_pagefault(inode->i_sb);
1662 file_update_time(vma->vm_file);
1663
1664 /* check if the faulting page hasn't raced with truncate */
1665 xfs_ilock(ip, XFS_MMAPLOCK_SHARED);
1666 size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT;
1667 if (vmf->pgoff >= size)
1668 ret = VM_FAULT_SIGBUS;
1669 else if (IS_DAX(inode))
1670 ret = dax_pfn_mkwrite(vma, vmf);
1671 xfs_iunlock(ip, XFS_MMAPLOCK_SHARED);
1672 sb_end_pagefault(inode->i_sb);
1673 return ret;
1674
1675}
1676
1677static const struct vm_operations_struct xfs_file_vm_ops = {
1678 .fault = xfs_filemap_fault,
1679 .pmd_fault = xfs_filemap_pmd_fault,
1680 .map_pages = filemap_map_pages,
1681 .page_mkwrite = xfs_filemap_page_mkwrite,
1682 .pfn_mkwrite = xfs_filemap_pfn_mkwrite,
1683};
1684
1685STATIC int
1686xfs_file_mmap(
1687 struct file *filp,
1688 struct vm_area_struct *vma)
1689{
1690 file_accessed(filp);
1691 vma->vm_ops = &xfs_file_vm_ops;
1692 if (IS_DAX(file_inode(filp)))
1693 vma->vm_flags |= VM_MIXEDMAP | VM_HUGEPAGE;
1694 return 0;
1695}
1696
1697const struct file_operations xfs_file_operations = {
1698 .llseek = xfs_file_llseek,
1699 .read_iter = xfs_file_read_iter,
1700 .write_iter = xfs_file_write_iter,
1701 .splice_read = xfs_file_splice_read,
1702 .splice_write = iter_file_splice_write,
1703 .unlocked_ioctl = xfs_file_ioctl,
1704#ifdef CONFIG_COMPAT
1705 .compat_ioctl = xfs_file_compat_ioctl,
1706#endif
1707 .mmap = xfs_file_mmap,
1708 .open = xfs_file_open,
1709 .release = xfs_file_release,
1710 .fsync = xfs_file_fsync,
1711 .fallocate = xfs_file_fallocate,
1712};
1713
1714const struct file_operations xfs_dir_file_operations = {
1715 .open = xfs_dir_open,
1716 .read = generic_read_dir,
1717 .iterate = xfs_file_readdir,
1718 .llseek = generic_file_llseek,
1719 .unlocked_ioctl = xfs_file_ioctl,
1720#ifdef CONFIG_COMPAT
1721 .compat_ioctl = xfs_file_compat_ioctl,
1722#endif
1723 .fsync = xfs_dir_fsync,
1724};