1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
4 * All Rights Reserved.
5 */
6 #include "xfs.h"
7 #include "xfs_fs.h"
8 #include "xfs_shared.h"
9 #include "xfs_format.h"
10 #include "xfs_log_format.h"
11 #include "xfs_trans_resv.h"
12 #include "xfs_mount.h"
13 #include "xfs_inode.h"
14 #include "xfs_trans.h"
15 #include "xfs_inode_item.h"
16 #include "xfs_bmap.h"
17 #include "xfs_bmap_util.h"
18 #include "xfs_dir2.h"
19 #include "xfs_dir2_priv.h"
20 #include "xfs_ioctl.h"
21 #include "xfs_trace.h"
22 #include "xfs_log.h"
23 #include "xfs_icache.h"
24 #include "xfs_pnfs.h"
25 #include "xfs_iomap.h"
26 #include "xfs_reflink.h"
27
28 #include <linux/dax.h>
29 #include <linux/falloc.h>
30 #include <linux/backing-dev.h>
31 #include <linux/mman.h>
32 #include <linux/fadvise.h>
33 #include <linux/mount.h>
34
35 static const struct vm_operations_struct xfs_file_vm_ops;
36
37 /*
38 * Decide if the given file range is aligned to the size of the fundamental
39 * allocation unit for the file.
40 */
41 static bool
xfs_is_falloc_aligned(struct xfs_inode * ip,loff_t pos,long long int len)42 xfs_is_falloc_aligned(
43 struct xfs_inode *ip,
44 loff_t pos,
45 long long int len)
46 {
47 struct xfs_mount *mp = ip->i_mount;
48 uint64_t mask;
49
50 if (XFS_IS_REALTIME_INODE(ip)) {
51 if (!is_power_of_2(mp->m_sb.sb_rextsize)) {
52 u64 rextbytes;
53 u32 mod;
54
55 rextbytes = XFS_FSB_TO_B(mp, mp->m_sb.sb_rextsize);
56 div_u64_rem(pos, rextbytes, &mod);
57 if (mod)
58 return false;
59 div_u64_rem(len, rextbytes, &mod);
60 return mod == 0;
61 }
62 mask = XFS_FSB_TO_B(mp, mp->m_sb.sb_rextsize) - 1;
63 } else {
64 mask = mp->m_sb.sb_blocksize - 1;
65 }
66
67 return !((pos | len) & mask);
68 }
69
70 /*
71 * Fsync operations on directories are much simpler than on regular files,
72 * as there is no file data to flush, and thus also no need for explicit
73 * cache flush operations, and there are no non-transaction metadata updates
74 * on directories either.
75 */
76 STATIC int
xfs_dir_fsync(struct file * file,loff_t start,loff_t end,int datasync)77 xfs_dir_fsync(
78 struct file *file,
79 loff_t start,
80 loff_t end,
81 int datasync)
82 {
83 struct xfs_inode *ip = XFS_I(file->f_mapping->host);
84
85 trace_xfs_dir_fsync(ip);
86 return xfs_log_force_inode(ip);
87 }
88
89 static xfs_csn_t
xfs_fsync_seq(struct xfs_inode * ip,bool datasync)90 xfs_fsync_seq(
91 struct xfs_inode *ip,
92 bool datasync)
93 {
94 if (!xfs_ipincount(ip))
95 return 0;
96 if (datasync && !(ip->i_itemp->ili_fsync_fields & ~XFS_ILOG_TIMESTAMP))
97 return 0;
98 return ip->i_itemp->ili_commit_seq;
99 }
100
101 /*
102 * All metadata updates are logged, which means that we just have to flush the
103 * log up to the latest LSN that touched the inode.
104 *
105 * If we have concurrent fsync/fdatasync() calls, we need them to all block on
106 * the log force before we clear the ili_fsync_fields field. This ensures that
107 * we don't get a racing sync operation that does not wait for the metadata to
108 * hit the journal before returning. If we race with clearing ili_fsync_fields,
109 * then all that will happen is the log force will do nothing as the lsn will
110 * already be on disk. We can't race with setting ili_fsync_fields because that
111 * is done under XFS_ILOCK_EXCL, and that can't happen because we hold the lock
112 * shared until after the ili_fsync_fields is cleared.
113 */
114 static int
xfs_fsync_flush_log(struct xfs_inode * ip,bool datasync,int * log_flushed)115 xfs_fsync_flush_log(
116 struct xfs_inode *ip,
117 bool datasync,
118 int *log_flushed)
119 {
120 int error = 0;
121 xfs_csn_t seq;
122
123 xfs_ilock(ip, XFS_ILOCK_SHARED);
124 seq = xfs_fsync_seq(ip, datasync);
125 if (seq) {
126 error = xfs_log_force_seq(ip->i_mount, seq, XFS_LOG_SYNC,
127 log_flushed);
128
129 spin_lock(&ip->i_itemp->ili_lock);
130 ip->i_itemp->ili_fsync_fields = 0;
131 spin_unlock(&ip->i_itemp->ili_lock);
132 }
133 xfs_iunlock(ip, XFS_ILOCK_SHARED);
134 return error;
135 }
136
137 STATIC int
xfs_file_fsync(struct file * file,loff_t start,loff_t end,int datasync)138 xfs_file_fsync(
139 struct file *file,
140 loff_t start,
141 loff_t end,
142 int datasync)
143 {
144 struct xfs_inode *ip = XFS_I(file->f_mapping->host);
145 struct xfs_mount *mp = ip->i_mount;
146 int error, err2;
147 int log_flushed = 0;
148
149 trace_xfs_file_fsync(ip);
150
151 error = file_write_and_wait_range(file, start, end);
152 if (error)
153 return error;
154
155 if (xfs_is_shutdown(mp))
156 return -EIO;
157
158 xfs_iflags_clear(ip, XFS_ITRUNCATED);
159
160 /*
161 * If we have an RT and/or log subvolume we need to make sure to flush
162 * the write cache the device used for file data first. This is to
163 * ensure newly written file data make it to disk before logging the new
164 * inode size in case of an extending write.
165 */
166 if (XFS_IS_REALTIME_INODE(ip))
167 error = blkdev_issue_flush(mp->m_rtdev_targp->bt_bdev);
168 else if (mp->m_logdev_targp != mp->m_ddev_targp)
169 error = blkdev_issue_flush(mp->m_ddev_targp->bt_bdev);
170
171 /*
172 * Any inode that has dirty modifications in the log is pinned. The
173 * racy check here for a pinned inode will not catch modifications
174 * that happen concurrently to the fsync call, but fsync semantics
175 * only require to sync previously completed I/O.
176 */
177 if (xfs_ipincount(ip)) {
178 err2 = xfs_fsync_flush_log(ip, datasync, &log_flushed);
179 if (err2 && !error)
180 error = err2;
181 }
182
183 /*
184 * If we only have a single device, and the log force about was
185 * a no-op we might have to flush the data device cache here.
186 * This can only happen for fdatasync/O_DSYNC if we were overwriting
187 * an already allocated file and thus do not have any metadata to
188 * commit.
189 */
190 if (!log_flushed && !XFS_IS_REALTIME_INODE(ip) &&
191 mp->m_logdev_targp == mp->m_ddev_targp) {
192 err2 = blkdev_issue_flush(mp->m_ddev_targp->bt_bdev);
193 if (err2 && !error)
194 error = err2;
195 }
196
197 return error;
198 }
199
200 static int
xfs_ilock_iocb(struct kiocb * iocb,unsigned int lock_mode)201 xfs_ilock_iocb(
202 struct kiocb *iocb,
203 unsigned int lock_mode)
204 {
205 struct xfs_inode *ip = XFS_I(file_inode(iocb->ki_filp));
206
207 if (iocb->ki_flags & IOCB_NOWAIT) {
208 if (!xfs_ilock_nowait(ip, lock_mode))
209 return -EAGAIN;
210 } else {
211 xfs_ilock(ip, lock_mode);
212 }
213
214 return 0;
215 }
216
217 STATIC ssize_t
xfs_file_dio_read(struct kiocb * iocb,struct iov_iter * to)218 xfs_file_dio_read(
219 struct kiocb *iocb,
220 struct iov_iter *to)
221 {
222 struct xfs_inode *ip = XFS_I(file_inode(iocb->ki_filp));
223 ssize_t ret;
224
225 trace_xfs_file_direct_read(iocb, to);
226
227 if (!iov_iter_count(to))
228 return 0; /* skip atime */
229
230 file_accessed(iocb->ki_filp);
231
232 ret = xfs_ilock_iocb(iocb, XFS_IOLOCK_SHARED);
233 if (ret)
234 return ret;
235 ret = iomap_dio_rw(iocb, to, &xfs_read_iomap_ops, NULL, 0, NULL, 0);
236 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
237
238 return ret;
239 }
240
241 static noinline ssize_t
xfs_file_dax_read(struct kiocb * iocb,struct iov_iter * to)242 xfs_file_dax_read(
243 struct kiocb *iocb,
244 struct iov_iter *to)
245 {
246 struct xfs_inode *ip = XFS_I(iocb->ki_filp->f_mapping->host);
247 ssize_t ret = 0;
248
249 trace_xfs_file_dax_read(iocb, to);
250
251 if (!iov_iter_count(to))
252 return 0; /* skip atime */
253
254 ret = xfs_ilock_iocb(iocb, XFS_IOLOCK_SHARED);
255 if (ret)
256 return ret;
257 ret = dax_iomap_rw(iocb, to, &xfs_read_iomap_ops);
258 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
259
260 file_accessed(iocb->ki_filp);
261 return ret;
262 }
263
264 STATIC ssize_t
xfs_file_buffered_read(struct kiocb * iocb,struct iov_iter * to)265 xfs_file_buffered_read(
266 struct kiocb *iocb,
267 struct iov_iter *to)
268 {
269 struct xfs_inode *ip = XFS_I(file_inode(iocb->ki_filp));
270 ssize_t ret;
271
272 trace_xfs_file_buffered_read(iocb, to);
273
274 ret = xfs_ilock_iocb(iocb, XFS_IOLOCK_SHARED);
275 if (ret)
276 return ret;
277 ret = generic_file_read_iter(iocb, to);
278 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
279
280 return ret;
281 }
282
283 STATIC ssize_t
xfs_file_read_iter(struct kiocb * iocb,struct iov_iter * to)284 xfs_file_read_iter(
285 struct kiocb *iocb,
286 struct iov_iter *to)
287 {
288 struct inode *inode = file_inode(iocb->ki_filp);
289 struct xfs_mount *mp = XFS_I(inode)->i_mount;
290 ssize_t ret = 0;
291
292 XFS_STATS_INC(mp, xs_read_calls);
293
294 if (xfs_is_shutdown(mp))
295 return -EIO;
296
297 if (IS_DAX(inode))
298 ret = xfs_file_dax_read(iocb, to);
299 else if (iocb->ki_flags & IOCB_DIRECT)
300 ret = xfs_file_dio_read(iocb, to);
301 else
302 ret = xfs_file_buffered_read(iocb, to);
303
304 if (ret > 0)
305 XFS_STATS_ADD(mp, xs_read_bytes, ret);
306 return ret;
307 }
308
309 STATIC ssize_t
xfs_file_splice_read(struct file * in,loff_t * ppos,struct pipe_inode_info * pipe,size_t len,unsigned int flags)310 xfs_file_splice_read(
311 struct file *in,
312 loff_t *ppos,
313 struct pipe_inode_info *pipe,
314 size_t len,
315 unsigned int flags)
316 {
317 struct inode *inode = file_inode(in);
318 struct xfs_inode *ip = XFS_I(inode);
319 struct xfs_mount *mp = ip->i_mount;
320 ssize_t ret = 0;
321
322 XFS_STATS_INC(mp, xs_read_calls);
323
324 if (xfs_is_shutdown(mp))
325 return -EIO;
326
327 trace_xfs_file_splice_read(ip, *ppos, len);
328
329 xfs_ilock(ip, XFS_IOLOCK_SHARED);
330 ret = filemap_splice_read(in, ppos, pipe, len, flags);
331 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
332 if (ret > 0)
333 XFS_STATS_ADD(mp, xs_read_bytes, ret);
334 return ret;
335 }
336
337 /*
338 * Common pre-write limit and setup checks.
339 *
340 * Called with the iolocked held either shared and exclusive according to
341 * @iolock, and returns with it held. Might upgrade the iolock to exclusive
342 * if called for a direct write beyond i_size.
343 */
344 STATIC ssize_t
xfs_file_write_checks(struct kiocb * iocb,struct iov_iter * from,unsigned int * iolock)345 xfs_file_write_checks(
346 struct kiocb *iocb,
347 struct iov_iter *from,
348 unsigned int *iolock)
349 {
350 struct file *file = iocb->ki_filp;
351 struct inode *inode = file->f_mapping->host;
352 struct xfs_inode *ip = XFS_I(inode);
353 ssize_t error = 0;
354 size_t count = iov_iter_count(from);
355 bool drained_dio = false;
356 loff_t isize;
357
358 restart:
359 error = generic_write_checks(iocb, from);
360 if (error <= 0)
361 return error;
362
363 if (iocb->ki_flags & IOCB_NOWAIT) {
364 error = break_layout(inode, false);
365 if (error == -EWOULDBLOCK)
366 error = -EAGAIN;
367 } else {
368 error = xfs_break_layouts(inode, iolock, BREAK_WRITE);
369 }
370
371 if (error)
372 return error;
373
374 /*
375 * For changing security info in file_remove_privs() we need i_rwsem
376 * exclusively.
377 */
378 if (*iolock == XFS_IOLOCK_SHARED && !IS_NOSEC(inode)) {
379 xfs_iunlock(ip, *iolock);
380 *iolock = XFS_IOLOCK_EXCL;
381 error = xfs_ilock_iocb(iocb, *iolock);
382 if (error) {
383 *iolock = 0;
384 return error;
385 }
386 goto restart;
387 }
388
389 /*
390 * If the offset is beyond the size of the file, we need to zero any
391 * blocks that fall between the existing EOF and the start of this
392 * write. If zeroing is needed and we are currently holding the iolock
393 * shared, we need to update it to exclusive which implies having to
394 * redo all checks before.
395 *
396 * We need to serialise against EOF updates that occur in IO completions
397 * here. We want to make sure that nobody is changing the size while we
398 * do this check until we have placed an IO barrier (i.e. hold the
399 * XFS_IOLOCK_EXCL) that prevents new IO from being dispatched. The
400 * spinlock effectively forms a memory barrier once we have the
401 * XFS_IOLOCK_EXCL so we are guaranteed to see the latest EOF value and
402 * hence be able to correctly determine if we need to run zeroing.
403 *
404 * We can do an unlocked check here safely as IO completion can only
405 * extend EOF. Truncate is locked out at this point, so the EOF can
406 * not move backwards, only forwards. Hence we only need to take the
407 * slow path and spin locks when we are at or beyond the current EOF.
408 */
409 if (iocb->ki_pos <= i_size_read(inode))
410 goto out;
411
412 spin_lock(&ip->i_flags_lock);
413 isize = i_size_read(inode);
414 if (iocb->ki_pos > isize) {
415 spin_unlock(&ip->i_flags_lock);
416
417 if (iocb->ki_flags & IOCB_NOWAIT)
418 return -EAGAIN;
419
420 if (!drained_dio) {
421 if (*iolock == XFS_IOLOCK_SHARED) {
422 xfs_iunlock(ip, *iolock);
423 *iolock = XFS_IOLOCK_EXCL;
424 xfs_ilock(ip, *iolock);
425 iov_iter_reexpand(from, count);
426 }
427 /*
428 * We now have an IO submission barrier in place, but
429 * AIO can do EOF updates during IO completion and hence
430 * we now need to wait for all of them to drain. Non-AIO
431 * DIO will have drained before we are given the
432 * XFS_IOLOCK_EXCL, and so for most cases this wait is a
433 * no-op.
434 */
435 inode_dio_wait(inode);
436 drained_dio = true;
437 goto restart;
438 }
439
440 trace_xfs_zero_eof(ip, isize, iocb->ki_pos - isize);
441 error = xfs_zero_range(ip, isize, iocb->ki_pos - isize, NULL);
442 if (error)
443 return error;
444 } else
445 spin_unlock(&ip->i_flags_lock);
446
447 out:
448 return kiocb_modified(iocb);
449 }
450
451 static int
xfs_dio_write_end_io(struct kiocb * iocb,ssize_t size,int error,unsigned flags)452 xfs_dio_write_end_io(
453 struct kiocb *iocb,
454 ssize_t size,
455 int error,
456 unsigned flags)
457 {
458 struct inode *inode = file_inode(iocb->ki_filp);
459 struct xfs_inode *ip = XFS_I(inode);
460 loff_t offset = iocb->ki_pos;
461 unsigned int nofs_flag;
462
463 trace_xfs_end_io_direct_write(ip, offset, size);
464
465 if (xfs_is_shutdown(ip->i_mount))
466 return -EIO;
467
468 if (error)
469 return error;
470 if (!size)
471 return 0;
472
473 /*
474 * Capture amount written on completion as we can't reliably account
475 * for it on submission.
476 */
477 XFS_STATS_ADD(ip->i_mount, xs_write_bytes, size);
478
479 /*
480 * We can allocate memory here while doing writeback on behalf of
481 * memory reclaim. To avoid memory allocation deadlocks set the
482 * task-wide nofs context for the following operations.
483 */
484 nofs_flag = memalloc_nofs_save();
485
486 if (flags & IOMAP_DIO_COW) {
487 error = xfs_reflink_end_cow(ip, offset, size);
488 if (error)
489 goto out;
490 }
491
492 /*
493 * Unwritten conversion updates the in-core isize after extent
494 * conversion but before updating the on-disk size. Updating isize any
495 * earlier allows a racing dio read to find unwritten extents before
496 * they are converted.
497 */
498 if (flags & IOMAP_DIO_UNWRITTEN) {
499 error = xfs_iomap_write_unwritten(ip, offset, size, true);
500 goto out;
501 }
502
503 /*
504 * We need to update the in-core inode size here so that we don't end up
505 * with the on-disk inode size being outside the in-core inode size. We
506 * have no other method of updating EOF for AIO, so always do it here
507 * if necessary.
508 *
509 * We need to lock the test/set EOF update as we can be racing with
510 * other IO completions here to update the EOF. Failing to serialise
511 * here can result in EOF moving backwards and Bad Things Happen when
512 * that occurs.
513 *
514 * As IO completion only ever extends EOF, we can do an unlocked check
515 * here to avoid taking the spinlock. If we land within the current EOF,
516 * then we do not need to do an extending update at all, and we don't
517 * need to take the lock to check this. If we race with an update moving
518 * EOF, then we'll either still be beyond EOF and need to take the lock,
519 * or we'll be within EOF and we don't need to take it at all.
520 */
521 if (offset + size <= i_size_read(inode))
522 goto out;
523
524 spin_lock(&ip->i_flags_lock);
525 if (offset + size > i_size_read(inode)) {
526 i_size_write(inode, offset + size);
527 spin_unlock(&ip->i_flags_lock);
528 error = xfs_setfilesize(ip, offset, size);
529 } else {
530 spin_unlock(&ip->i_flags_lock);
531 }
532
533 out:
534 memalloc_nofs_restore(nofs_flag);
535 return error;
536 }
537
538 static const struct iomap_dio_ops xfs_dio_write_ops = {
539 .end_io = xfs_dio_write_end_io,
540 };
541
542 /*
543 * Handle block aligned direct I/O writes
544 */
545 static noinline ssize_t
xfs_file_dio_write_aligned(struct xfs_inode * ip,struct kiocb * iocb,struct iov_iter * from)546 xfs_file_dio_write_aligned(
547 struct xfs_inode *ip,
548 struct kiocb *iocb,
549 struct iov_iter *from)
550 {
551 unsigned int iolock = XFS_IOLOCK_SHARED;
552 ssize_t ret;
553
554 ret = xfs_ilock_iocb(iocb, iolock);
555 if (ret)
556 return ret;
557 ret = xfs_file_write_checks(iocb, from, &iolock);
558 if (ret)
559 goto out_unlock;
560
561 /*
562 * We don't need to hold the IOLOCK exclusively across the IO, so demote
563 * the iolock back to shared if we had to take the exclusive lock in
564 * xfs_file_write_checks() for other reasons.
565 */
566 if (iolock == XFS_IOLOCK_EXCL) {
567 xfs_ilock_demote(ip, XFS_IOLOCK_EXCL);
568 iolock = XFS_IOLOCK_SHARED;
569 }
570 trace_xfs_file_direct_write(iocb, from);
571 ret = iomap_dio_rw(iocb, from, &xfs_direct_write_iomap_ops,
572 &xfs_dio_write_ops, 0, NULL, 0);
573 out_unlock:
574 if (iolock)
575 xfs_iunlock(ip, iolock);
576 return ret;
577 }
578
579 /*
580 * Handle block unaligned direct I/O writes
581 *
582 * In most cases direct I/O writes will be done holding IOLOCK_SHARED, allowing
583 * them to be done in parallel with reads and other direct I/O writes. However,
584 * if the I/O is not aligned to filesystem blocks, the direct I/O layer may need
585 * to do sub-block zeroing and that requires serialisation against other direct
586 * I/O to the same block. In this case we need to serialise the submission of
587 * the unaligned I/O so that we don't get racing block zeroing in the dio layer.
588 * In the case where sub-block zeroing is not required, we can do concurrent
589 * sub-block dios to the same block successfully.
590 *
591 * Optimistically submit the I/O using the shared lock first, but use the
592 * IOMAP_DIO_OVERWRITE_ONLY flag to tell the lower layers to return -EAGAIN
593 * if block allocation or partial block zeroing would be required. In that case
594 * we try again with the exclusive lock.
595 */
596 static noinline ssize_t
xfs_file_dio_write_unaligned(struct xfs_inode * ip,struct kiocb * iocb,struct iov_iter * from)597 xfs_file_dio_write_unaligned(
598 struct xfs_inode *ip,
599 struct kiocb *iocb,
600 struct iov_iter *from)
601 {
602 size_t isize = i_size_read(VFS_I(ip));
603 size_t count = iov_iter_count(from);
604 unsigned int iolock = XFS_IOLOCK_SHARED;
605 unsigned int flags = IOMAP_DIO_OVERWRITE_ONLY;
606 ssize_t ret;
607
608 /*
609 * Extending writes need exclusivity because of the sub-block zeroing
610 * that the DIO code always does for partial tail blocks beyond EOF, so
611 * don't even bother trying the fast path in this case.
612 */
613 if (iocb->ki_pos > isize || iocb->ki_pos + count >= isize) {
614 if (iocb->ki_flags & IOCB_NOWAIT)
615 return -EAGAIN;
616 retry_exclusive:
617 iolock = XFS_IOLOCK_EXCL;
618 flags = IOMAP_DIO_FORCE_WAIT;
619 }
620
621 ret = xfs_ilock_iocb(iocb, iolock);
622 if (ret)
623 return ret;
624
625 /*
626 * We can't properly handle unaligned direct I/O to reflink files yet,
627 * as we can't unshare a partial block.
628 */
629 if (xfs_is_cow_inode(ip)) {
630 trace_xfs_reflink_bounce_dio_write(iocb, from);
631 ret = -ENOTBLK;
632 goto out_unlock;
633 }
634
635 ret = xfs_file_write_checks(iocb, from, &iolock);
636 if (ret)
637 goto out_unlock;
638
639 /*
640 * If we are doing exclusive unaligned I/O, this must be the only I/O
641 * in-flight. Otherwise we risk data corruption due to unwritten extent
642 * conversions from the AIO end_io handler. Wait for all other I/O to
643 * drain first.
644 */
645 if (flags & IOMAP_DIO_FORCE_WAIT)
646 inode_dio_wait(VFS_I(ip));
647
648 trace_xfs_file_direct_write(iocb, from);
649 ret = iomap_dio_rw(iocb, from, &xfs_direct_write_iomap_ops,
650 &xfs_dio_write_ops, flags, NULL, 0);
651
652 /*
653 * Retry unaligned I/O with exclusive blocking semantics if the DIO
654 * layer rejected it for mapping or locking reasons. If we are doing
655 * nonblocking user I/O, propagate the error.
656 */
657 if (ret == -EAGAIN && !(iocb->ki_flags & IOCB_NOWAIT)) {
658 ASSERT(flags & IOMAP_DIO_OVERWRITE_ONLY);
659 xfs_iunlock(ip, iolock);
660 goto retry_exclusive;
661 }
662
663 out_unlock:
664 if (iolock)
665 xfs_iunlock(ip, iolock);
666 return ret;
667 }
668
669 static ssize_t
xfs_file_dio_write(struct kiocb * iocb,struct iov_iter * from)670 xfs_file_dio_write(
671 struct kiocb *iocb,
672 struct iov_iter *from)
673 {
674 struct xfs_inode *ip = XFS_I(file_inode(iocb->ki_filp));
675 struct xfs_buftarg *target = xfs_inode_buftarg(ip);
676 size_t count = iov_iter_count(from);
677
678 /* direct I/O must be aligned to device logical sector size */
679 if ((iocb->ki_pos | count) & target->bt_logical_sectormask)
680 return -EINVAL;
681 if ((iocb->ki_pos | count) & ip->i_mount->m_blockmask)
682 return xfs_file_dio_write_unaligned(ip, iocb, from);
683 return xfs_file_dio_write_aligned(ip, iocb, from);
684 }
685
686 static noinline ssize_t
xfs_file_dax_write(struct kiocb * iocb,struct iov_iter * from)687 xfs_file_dax_write(
688 struct kiocb *iocb,
689 struct iov_iter *from)
690 {
691 struct inode *inode = iocb->ki_filp->f_mapping->host;
692 struct xfs_inode *ip = XFS_I(inode);
693 unsigned int iolock = XFS_IOLOCK_EXCL;
694 ssize_t ret, error = 0;
695 loff_t pos;
696
697 ret = xfs_ilock_iocb(iocb, iolock);
698 if (ret)
699 return ret;
700 ret = xfs_file_write_checks(iocb, from, &iolock);
701 if (ret)
702 goto out;
703
704 pos = iocb->ki_pos;
705
706 trace_xfs_file_dax_write(iocb, from);
707 ret = dax_iomap_rw(iocb, from, &xfs_dax_write_iomap_ops);
708 if (ret > 0 && iocb->ki_pos > i_size_read(inode)) {
709 i_size_write(inode, iocb->ki_pos);
710 error = xfs_setfilesize(ip, pos, ret);
711 }
712 out:
713 if (iolock)
714 xfs_iunlock(ip, iolock);
715 if (error)
716 return error;
717
718 if (ret > 0) {
719 XFS_STATS_ADD(ip->i_mount, xs_write_bytes, ret);
720
721 /* Handle various SYNC-type writes */
722 ret = generic_write_sync(iocb, ret);
723 }
724 return ret;
725 }
726
727 STATIC ssize_t
xfs_file_buffered_write(struct kiocb * iocb,struct iov_iter * from)728 xfs_file_buffered_write(
729 struct kiocb *iocb,
730 struct iov_iter *from)
731 {
732 struct inode *inode = iocb->ki_filp->f_mapping->host;
733 struct xfs_inode *ip = XFS_I(inode);
734 ssize_t ret;
735 bool cleared_space = false;
736 unsigned int iolock;
737
738 write_retry:
739 iolock = XFS_IOLOCK_EXCL;
740 ret = xfs_ilock_iocb(iocb, iolock);
741 if (ret)
742 return ret;
743
744 ret = xfs_file_write_checks(iocb, from, &iolock);
745 if (ret)
746 goto out;
747
748 trace_xfs_file_buffered_write(iocb, from);
749 ret = iomap_file_buffered_write(iocb, from,
750 &xfs_buffered_write_iomap_ops);
751
752 /*
753 * If we hit a space limit, try to free up some lingering preallocated
754 * space before returning an error. In the case of ENOSPC, first try to
755 * write back all dirty inodes to free up some of the excess reserved
756 * metadata space. This reduces the chances that the eofblocks scan
757 * waits on dirty mappings. Since xfs_flush_inodes() is serialized, this
758 * also behaves as a filter to prevent too many eofblocks scans from
759 * running at the same time. Use a synchronous scan to increase the
760 * effectiveness of the scan.
761 */
762 if (ret == -EDQUOT && !cleared_space) {
763 xfs_iunlock(ip, iolock);
764 xfs_blockgc_free_quota(ip, XFS_ICWALK_FLAG_SYNC);
765 cleared_space = true;
766 goto write_retry;
767 } else if (ret == -ENOSPC && !cleared_space) {
768 struct xfs_icwalk icw = {0};
769
770 cleared_space = true;
771 xfs_flush_inodes(ip->i_mount);
772
773 xfs_iunlock(ip, iolock);
774 icw.icw_flags = XFS_ICWALK_FLAG_SYNC;
775 xfs_blockgc_free_space(ip->i_mount, &icw);
776 goto write_retry;
777 }
778
779 out:
780 if (iolock)
781 xfs_iunlock(ip, iolock);
782
783 if (ret > 0) {
784 XFS_STATS_ADD(ip->i_mount, xs_write_bytes, ret);
785 /* Handle various SYNC-type writes */
786 ret = generic_write_sync(iocb, ret);
787 }
788 return ret;
789 }
790
791 STATIC ssize_t
xfs_file_write_iter(struct kiocb * iocb,struct iov_iter * from)792 xfs_file_write_iter(
793 struct kiocb *iocb,
794 struct iov_iter *from)
795 {
796 struct inode *inode = iocb->ki_filp->f_mapping->host;
797 struct xfs_inode *ip = XFS_I(inode);
798 ssize_t ret;
799 size_t ocount = iov_iter_count(from);
800
801 XFS_STATS_INC(ip->i_mount, xs_write_calls);
802
803 if (ocount == 0)
804 return 0;
805
806 if (xfs_is_shutdown(ip->i_mount))
807 return -EIO;
808
809 if (IS_DAX(inode))
810 return xfs_file_dax_write(iocb, from);
811
812 if (iocb->ki_flags & IOCB_DIRECT) {
813 /*
814 * Allow a directio write to fall back to a buffered
815 * write *only* in the case that we're doing a reflink
816 * CoW. In all other directio scenarios we do not
817 * allow an operation to fall back to buffered mode.
818 */
819 ret = xfs_file_dio_write(iocb, from);
820 if (ret != -ENOTBLK)
821 return ret;
822 }
823
824 return xfs_file_buffered_write(iocb, from);
825 }
826
827 static void
xfs_wait_dax_page(struct inode * inode)828 xfs_wait_dax_page(
829 struct inode *inode)
830 {
831 struct xfs_inode *ip = XFS_I(inode);
832
833 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
834 schedule();
835 xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
836 }
837
838 int
xfs_break_dax_layouts(struct inode * inode,bool * retry)839 xfs_break_dax_layouts(
840 struct inode *inode,
841 bool *retry)
842 {
843 struct page *page;
844
845 ASSERT(xfs_isilocked(XFS_I(inode), XFS_MMAPLOCK_EXCL));
846
847 page = dax_layout_busy_page(inode->i_mapping);
848 if (!page)
849 return 0;
850
851 *retry = true;
852 return ___wait_var_event(&page->_refcount,
853 atomic_read(&page->_refcount) == 1, TASK_INTERRUPTIBLE,
854 0, 0, xfs_wait_dax_page(inode));
855 }
856
857 int
xfs_break_layouts(struct inode * inode,uint * iolock,enum layout_break_reason reason)858 xfs_break_layouts(
859 struct inode *inode,
860 uint *iolock,
861 enum layout_break_reason reason)
862 {
863 bool retry;
864 int error;
865
866 ASSERT(xfs_isilocked(XFS_I(inode), XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL));
867
868 do {
869 retry = false;
870 switch (reason) {
871 case BREAK_UNMAP:
872 error = xfs_break_dax_layouts(inode, &retry);
873 if (error || retry)
874 break;
875 fallthrough;
876 case BREAK_WRITE:
877 error = xfs_break_leased_layouts(inode, iolock, &retry);
878 break;
879 default:
880 WARN_ON_ONCE(1);
881 error = -EINVAL;
882 }
883 } while (error == 0 && retry);
884
885 return error;
886 }
887
888 /* Does this file, inode, or mount want synchronous writes? */
xfs_file_sync_writes(struct file * filp)889 static inline bool xfs_file_sync_writes(struct file *filp)
890 {
891 struct xfs_inode *ip = XFS_I(file_inode(filp));
892
893 if (xfs_has_wsync(ip->i_mount))
894 return true;
895 if (filp->f_flags & (__O_SYNC | O_DSYNC))
896 return true;
897 if (IS_SYNC(file_inode(filp)))
898 return true;
899
900 return false;
901 }
902
903 #define XFS_FALLOC_FL_SUPPORTED \
904 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
905 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE | \
906 FALLOC_FL_INSERT_RANGE | FALLOC_FL_UNSHARE_RANGE)
907
908 STATIC long
xfs_file_fallocate(struct file * file,int mode,loff_t offset,loff_t len)909 xfs_file_fallocate(
910 struct file *file,
911 int mode,
912 loff_t offset,
913 loff_t len)
914 {
915 struct inode *inode = file_inode(file);
916 struct xfs_inode *ip = XFS_I(inode);
917 long error;
918 uint iolock = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL;
919 loff_t new_size = 0;
920 bool do_file_insert = false;
921
922 if (!S_ISREG(inode->i_mode))
923 return -EINVAL;
924 if (mode & ~XFS_FALLOC_FL_SUPPORTED)
925 return -EOPNOTSUPP;
926
927 xfs_ilock(ip, iolock);
928 error = xfs_break_layouts(inode, &iolock, BREAK_UNMAP);
929 if (error)
930 goto out_unlock;
931
932 /*
933 * Must wait for all AIO to complete before we continue as AIO can
934 * change the file size on completion without holding any locks we
935 * currently hold. We must do this first because AIO can update both
936 * the on disk and in memory inode sizes, and the operations that follow
937 * require the in-memory size to be fully up-to-date.
938 */
939 inode_dio_wait(inode);
940
941 /*
942 * Now AIO and DIO has drained we flush and (if necessary) invalidate
943 * the cached range over the first operation we are about to run.
944 *
945 * We care about zero and collapse here because they both run a hole
946 * punch over the range first. Because that can zero data, and the range
947 * of invalidation for the shift operations is much larger, we still do
948 * the required flush for collapse in xfs_prepare_shift().
949 *
950 * Insert has the same range requirements as collapse, and we extend the
951 * file first which can zero data. Hence insert has the same
952 * flush/invalidate requirements as collapse and so they are both
953 * handled at the right time by xfs_prepare_shift().
954 */
955 if (mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE |
956 FALLOC_FL_COLLAPSE_RANGE)) {
957 error = xfs_flush_unmap_range(ip, offset, len);
958 if (error)
959 goto out_unlock;
960 }
961
962 error = file_modified(file);
963 if (error)
964 goto out_unlock;
965
966 if (mode & FALLOC_FL_PUNCH_HOLE) {
967 error = xfs_free_file_space(ip, offset, len);
968 if (error)
969 goto out_unlock;
970 } else if (mode & FALLOC_FL_COLLAPSE_RANGE) {
971 if (!xfs_is_falloc_aligned(ip, offset, len)) {
972 error = -EINVAL;
973 goto out_unlock;
974 }
975
976 /*
977 * There is no need to overlap collapse range with EOF,
978 * in which case it is effectively a truncate operation
979 */
980 if (offset + len >= i_size_read(inode)) {
981 error = -EINVAL;
982 goto out_unlock;
983 }
984
985 new_size = i_size_read(inode) - len;
986
987 error = xfs_collapse_file_space(ip, offset, len);
988 if (error)
989 goto out_unlock;
990 } else if (mode & FALLOC_FL_INSERT_RANGE) {
991 loff_t isize = i_size_read(inode);
992
993 if (!xfs_is_falloc_aligned(ip, offset, len)) {
994 error = -EINVAL;
995 goto out_unlock;
996 }
997
998 /*
999 * New inode size must not exceed ->s_maxbytes, accounting for
1000 * possible signed overflow.
1001 */
1002 if (inode->i_sb->s_maxbytes - isize < len) {
1003 error = -EFBIG;
1004 goto out_unlock;
1005 }
1006 new_size = isize + len;
1007
1008 /* Offset should be less than i_size */
1009 if (offset >= isize) {
1010 error = -EINVAL;
1011 goto out_unlock;
1012 }
1013 do_file_insert = true;
1014 } else {
1015 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
1016 offset + len > i_size_read(inode)) {
1017 new_size = offset + len;
1018 error = inode_newsize_ok(inode, new_size);
1019 if (error)
1020 goto out_unlock;
1021 }
1022
1023 if (mode & FALLOC_FL_ZERO_RANGE) {
1024 /*
1025 * Punch a hole and prealloc the range. We use a hole
1026 * punch rather than unwritten extent conversion for two
1027 * reasons:
1028 *
1029 * 1.) Hole punch handles partial block zeroing for us.
1030 * 2.) If prealloc returns ENOSPC, the file range is
1031 * still zero-valued by virtue of the hole punch.
1032 */
1033 unsigned int blksize = i_blocksize(inode);
1034
1035 trace_xfs_zero_file_space(ip);
1036
1037 error = xfs_free_file_space(ip, offset, len);
1038 if (error)
1039 goto out_unlock;
1040
1041 len = round_up(offset + len, blksize) -
1042 round_down(offset, blksize);
1043 offset = round_down(offset, blksize);
1044 } else if (mode & FALLOC_FL_UNSHARE_RANGE) {
1045 error = xfs_reflink_unshare(ip, offset, len);
1046 if (error)
1047 goto out_unlock;
1048 } else {
1049 /*
1050 * If always_cow mode we can't use preallocations and
1051 * thus should not create them.
1052 */
1053 if (xfs_is_always_cow_inode(ip)) {
1054 error = -EOPNOTSUPP;
1055 goto out_unlock;
1056 }
1057 }
1058
1059 if (!xfs_is_always_cow_inode(ip)) {
1060 error = xfs_alloc_file_space(ip, offset, len);
1061 if (error)
1062 goto out_unlock;
1063 }
1064 }
1065
1066 /* Change file size if needed */
1067 if (new_size) {
1068 struct iattr iattr;
1069
1070 iattr.ia_valid = ATTR_SIZE;
1071 iattr.ia_size = new_size;
1072 error = xfs_vn_setattr_size(file_mnt_idmap(file),
1073 file_dentry(file), &iattr);
1074 if (error)
1075 goto out_unlock;
1076 }
1077
1078 /*
1079 * Perform hole insertion now that the file size has been
1080 * updated so that if we crash during the operation we don't
1081 * leave shifted extents past EOF and hence losing access to
1082 * the data that is contained within them.
1083 */
1084 if (do_file_insert) {
1085 error = xfs_insert_file_space(ip, offset, len);
1086 if (error)
1087 goto out_unlock;
1088 }
1089
1090 if (xfs_file_sync_writes(file))
1091 error = xfs_log_force_inode(ip);
1092
1093 out_unlock:
1094 xfs_iunlock(ip, iolock);
1095 return error;
1096 }
1097
1098 STATIC int
xfs_file_fadvise(struct file * file,loff_t start,loff_t end,int advice)1099 xfs_file_fadvise(
1100 struct file *file,
1101 loff_t start,
1102 loff_t end,
1103 int advice)
1104 {
1105 struct xfs_inode *ip = XFS_I(file_inode(file));
1106 int ret;
1107 int lockflags = 0;
1108
1109 /*
1110 * Operations creating pages in page cache need protection from hole
1111 * punching and similar ops
1112 */
1113 if (advice == POSIX_FADV_WILLNEED) {
1114 lockflags = XFS_IOLOCK_SHARED;
1115 xfs_ilock(ip, lockflags);
1116 }
1117 ret = generic_fadvise(file, start, end, advice);
1118 if (lockflags)
1119 xfs_iunlock(ip, lockflags);
1120 return ret;
1121 }
1122
1123 STATIC loff_t
xfs_file_remap_range(struct file * file_in,loff_t pos_in,struct file * file_out,loff_t pos_out,loff_t len,unsigned int remap_flags)1124 xfs_file_remap_range(
1125 struct file *file_in,
1126 loff_t pos_in,
1127 struct file *file_out,
1128 loff_t pos_out,
1129 loff_t len,
1130 unsigned int remap_flags)
1131 {
1132 struct inode *inode_in = file_inode(file_in);
1133 struct xfs_inode *src = XFS_I(inode_in);
1134 struct inode *inode_out = file_inode(file_out);
1135 struct xfs_inode *dest = XFS_I(inode_out);
1136 struct xfs_mount *mp = src->i_mount;
1137 loff_t remapped = 0;
1138 xfs_extlen_t cowextsize;
1139 int ret;
1140
1141 if (remap_flags & ~(REMAP_FILE_DEDUP | REMAP_FILE_ADVISORY))
1142 return -EINVAL;
1143
1144 if (!xfs_has_reflink(mp))
1145 return -EOPNOTSUPP;
1146
1147 if (xfs_is_shutdown(mp))
1148 return -EIO;
1149
1150 /* Prepare and then clone file data. */
1151 ret = xfs_reflink_remap_prep(file_in, pos_in, file_out, pos_out,
1152 &len, remap_flags);
1153 if (ret || len == 0)
1154 return ret;
1155
1156 trace_xfs_reflink_remap_range(src, pos_in, len, dest, pos_out);
1157
1158 ret = xfs_reflink_remap_blocks(src, pos_in, dest, pos_out, len,
1159 &remapped);
1160 if (ret)
1161 goto out_unlock;
1162
1163 /*
1164 * Carry the cowextsize hint from src to dest if we're sharing the
1165 * entire source file to the entire destination file, the source file
1166 * has a cowextsize hint, and the destination file does not.
1167 */
1168 cowextsize = 0;
1169 if (pos_in == 0 && len == i_size_read(inode_in) &&
1170 (src->i_diflags2 & XFS_DIFLAG2_COWEXTSIZE) &&
1171 pos_out == 0 && len >= i_size_read(inode_out) &&
1172 !(dest->i_diflags2 & XFS_DIFLAG2_COWEXTSIZE))
1173 cowextsize = src->i_cowextsize;
1174
1175 ret = xfs_reflink_update_dest(dest, pos_out + len, cowextsize,
1176 remap_flags);
1177 if (ret)
1178 goto out_unlock;
1179
1180 if (xfs_file_sync_writes(file_in) || xfs_file_sync_writes(file_out))
1181 xfs_log_force_inode(dest);
1182 out_unlock:
1183 xfs_iunlock2_io_mmap(src, dest);
1184 if (ret)
1185 trace_xfs_reflink_remap_range_error(dest, ret, _RET_IP_);
1186 return remapped > 0 ? remapped : ret;
1187 }
1188
1189 STATIC int
xfs_file_open(struct inode * inode,struct file * file)1190 xfs_file_open(
1191 struct inode *inode,
1192 struct file *file)
1193 {
1194 if (xfs_is_shutdown(XFS_M(inode->i_sb)))
1195 return -EIO;
1196 file->f_mode |= FMODE_NOWAIT | FMODE_BUF_RASYNC | FMODE_BUF_WASYNC |
1197 FMODE_DIO_PARALLEL_WRITE | FMODE_CAN_ODIRECT;
1198 return generic_file_open(inode, file);
1199 }
1200
1201 STATIC int
xfs_dir_open(struct inode * inode,struct file * file)1202 xfs_dir_open(
1203 struct inode *inode,
1204 struct file *file)
1205 {
1206 struct xfs_inode *ip = XFS_I(inode);
1207 unsigned int mode;
1208 int error;
1209
1210 error = xfs_file_open(inode, file);
1211 if (error)
1212 return error;
1213
1214 /*
1215 * If there are any blocks, read-ahead block 0 as we're almost
1216 * certain to have the next operation be a read there.
1217 */
1218 mode = xfs_ilock_data_map_shared(ip);
1219 if (ip->i_df.if_nextents > 0)
1220 error = xfs_dir3_data_readahead(ip, 0, 0);
1221 xfs_iunlock(ip, mode);
1222 return error;
1223 }
1224
1225 STATIC int
xfs_file_release(struct inode * inode,struct file * filp)1226 xfs_file_release(
1227 struct inode *inode,
1228 struct file *filp)
1229 {
1230 return xfs_release(XFS_I(inode));
1231 }
1232
1233 STATIC int
xfs_file_readdir(struct file * file,struct dir_context * ctx)1234 xfs_file_readdir(
1235 struct file *file,
1236 struct dir_context *ctx)
1237 {
1238 struct inode *inode = file_inode(file);
1239 xfs_inode_t *ip = XFS_I(inode);
1240 size_t bufsize;
1241
1242 /*
1243 * The Linux API doesn't pass down the total size of the buffer
1244 * we read into down to the filesystem. With the filldir concept
1245 * it's not needed for correct information, but the XFS dir2 leaf
1246 * code wants an estimate of the buffer size to calculate it's
1247 * readahead window and size the buffers used for mapping to
1248 * physical blocks.
1249 *
1250 * Try to give it an estimate that's good enough, maybe at some
1251 * point we can change the ->readdir prototype to include the
1252 * buffer size. For now we use the current glibc buffer size.
1253 */
1254 bufsize = (size_t)min_t(loff_t, XFS_READDIR_BUFSIZE, ip->i_disk_size);
1255
1256 return xfs_readdir(NULL, ip, ctx, bufsize);
1257 }
1258
1259 STATIC loff_t
xfs_file_llseek(struct file * file,loff_t offset,int whence)1260 xfs_file_llseek(
1261 struct file *file,
1262 loff_t offset,
1263 int whence)
1264 {
1265 struct inode *inode = file->f_mapping->host;
1266
1267 if (xfs_is_shutdown(XFS_I(inode)->i_mount))
1268 return -EIO;
1269
1270 switch (whence) {
1271 default:
1272 return generic_file_llseek(file, offset, whence);
1273 case SEEK_HOLE:
1274 offset = iomap_seek_hole(inode, offset, &xfs_seek_iomap_ops);
1275 break;
1276 case SEEK_DATA:
1277 offset = iomap_seek_data(inode, offset, &xfs_seek_iomap_ops);
1278 break;
1279 }
1280
1281 if (offset < 0)
1282 return offset;
1283 return vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
1284 }
1285
1286 #ifdef CONFIG_FS_DAX
1287 static inline vm_fault_t
xfs_dax_fault(struct vm_fault * vmf,unsigned int order,bool write_fault,pfn_t * pfn)1288 xfs_dax_fault(
1289 struct vm_fault *vmf,
1290 unsigned int order,
1291 bool write_fault,
1292 pfn_t *pfn)
1293 {
1294 return dax_iomap_fault(vmf, order, pfn, NULL,
1295 (write_fault && !vmf->cow_page) ?
1296 &xfs_dax_write_iomap_ops :
1297 &xfs_read_iomap_ops);
1298 }
1299 #else
1300 static inline vm_fault_t
xfs_dax_fault(struct vm_fault * vmf,unsigned int order,bool write_fault,pfn_t * pfn)1301 xfs_dax_fault(
1302 struct vm_fault *vmf,
1303 unsigned int order,
1304 bool write_fault,
1305 pfn_t *pfn)
1306 {
1307 ASSERT(0);
1308 return VM_FAULT_SIGBUS;
1309 }
1310 #endif
1311
1312 /*
1313 * Locking for serialisation of IO during page faults. This results in a lock
1314 * ordering of:
1315 *
1316 * mmap_lock (MM)
1317 * sb_start_pagefault(vfs, freeze)
1318 * invalidate_lock (vfs/XFS_MMAPLOCK - truncate serialisation)
1319 * page_lock (MM)
1320 * i_lock (XFS - extent map serialisation)
1321 */
1322 static vm_fault_t
__xfs_filemap_fault(struct vm_fault * vmf,unsigned int order,bool write_fault)1323 __xfs_filemap_fault(
1324 struct vm_fault *vmf,
1325 unsigned int order,
1326 bool write_fault)
1327 {
1328 struct inode *inode = file_inode(vmf->vma->vm_file);
1329 struct xfs_inode *ip = XFS_I(inode);
1330 vm_fault_t ret;
1331
1332 trace_xfs_filemap_fault(ip, order, write_fault);
1333
1334 if (write_fault) {
1335 sb_start_pagefault(inode->i_sb);
1336 file_update_time(vmf->vma->vm_file);
1337 }
1338
1339 if (IS_DAX(inode)) {
1340 pfn_t pfn;
1341
1342 xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1343 ret = xfs_dax_fault(vmf, order, write_fault, &pfn);
1344 if (ret & VM_FAULT_NEEDDSYNC)
1345 ret = dax_finish_sync_fault(vmf, order, pfn);
1346 xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1347 } else {
1348 if (write_fault) {
1349 xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1350 ret = iomap_page_mkwrite(vmf,
1351 &xfs_page_mkwrite_iomap_ops);
1352 xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1353 } else {
1354 ret = filemap_fault(vmf);
1355 }
1356 }
1357
1358 if (write_fault)
1359 sb_end_pagefault(inode->i_sb);
1360 return ret;
1361 }
1362
1363 static inline bool
xfs_is_write_fault(struct vm_fault * vmf)1364 xfs_is_write_fault(
1365 struct vm_fault *vmf)
1366 {
1367 return (vmf->flags & FAULT_FLAG_WRITE) &&
1368 (vmf->vma->vm_flags & VM_SHARED);
1369 }
1370
1371 static vm_fault_t
xfs_filemap_fault(struct vm_fault * vmf)1372 xfs_filemap_fault(
1373 struct vm_fault *vmf)
1374 {
1375 /* DAX can shortcut the normal fault path on write faults! */
1376 return __xfs_filemap_fault(vmf, 0,
1377 IS_DAX(file_inode(vmf->vma->vm_file)) &&
1378 xfs_is_write_fault(vmf));
1379 }
1380
1381 static vm_fault_t
xfs_filemap_huge_fault(struct vm_fault * vmf,unsigned int order)1382 xfs_filemap_huge_fault(
1383 struct vm_fault *vmf,
1384 unsigned int order)
1385 {
1386 if (!IS_DAX(file_inode(vmf->vma->vm_file)))
1387 return VM_FAULT_FALLBACK;
1388
1389 /* DAX can shortcut the normal fault path on write faults! */
1390 return __xfs_filemap_fault(vmf, order,
1391 xfs_is_write_fault(vmf));
1392 }
1393
1394 static vm_fault_t
xfs_filemap_page_mkwrite(struct vm_fault * vmf)1395 xfs_filemap_page_mkwrite(
1396 struct vm_fault *vmf)
1397 {
1398 return __xfs_filemap_fault(vmf, 0, true);
1399 }
1400
1401 /*
1402 * pfn_mkwrite was originally intended to ensure we capture time stamp updates
1403 * on write faults. In reality, it needs to serialise against truncate and
1404 * prepare memory for writing so handle is as standard write fault.
1405 */
1406 static vm_fault_t
xfs_filemap_pfn_mkwrite(struct vm_fault * vmf)1407 xfs_filemap_pfn_mkwrite(
1408 struct vm_fault *vmf)
1409 {
1410
1411 return __xfs_filemap_fault(vmf, 0, true);
1412 }
1413
1414 static const struct vm_operations_struct xfs_file_vm_ops = {
1415 .fault = xfs_filemap_fault,
1416 .huge_fault = xfs_filemap_huge_fault,
1417 .map_pages = filemap_map_pages,
1418 .page_mkwrite = xfs_filemap_page_mkwrite,
1419 .pfn_mkwrite = xfs_filemap_pfn_mkwrite,
1420 };
1421
1422 STATIC int
xfs_file_mmap(struct file * file,struct vm_area_struct * vma)1423 xfs_file_mmap(
1424 struct file *file,
1425 struct vm_area_struct *vma)
1426 {
1427 struct inode *inode = file_inode(file);
1428 struct xfs_buftarg *target = xfs_inode_buftarg(XFS_I(inode));
1429
1430 /*
1431 * We don't support synchronous mappings for non-DAX files and
1432 * for DAX files if underneath dax_device is not synchronous.
1433 */
1434 if (!daxdev_mapping_supported(vma, target->bt_daxdev))
1435 return -EOPNOTSUPP;
1436
1437 file_accessed(file);
1438 vma->vm_ops = &xfs_file_vm_ops;
1439 if (IS_DAX(inode))
1440 vm_flags_set(vma, VM_HUGEPAGE);
1441 return 0;
1442 }
1443
1444 const struct file_operations xfs_file_operations = {
1445 .llseek = xfs_file_llseek,
1446 .read_iter = xfs_file_read_iter,
1447 .write_iter = xfs_file_write_iter,
1448 .splice_read = xfs_file_splice_read,
1449 .splice_write = iter_file_splice_write,
1450 .iopoll = iocb_bio_iopoll,
1451 .unlocked_ioctl = xfs_file_ioctl,
1452 #ifdef CONFIG_COMPAT
1453 .compat_ioctl = xfs_file_compat_ioctl,
1454 #endif
1455 .mmap = xfs_file_mmap,
1456 .mmap_supported_flags = MAP_SYNC,
1457 .open = xfs_file_open,
1458 .release = xfs_file_release,
1459 .fsync = xfs_file_fsync,
1460 .get_unmapped_area = thp_get_unmapped_area,
1461 .fallocate = xfs_file_fallocate,
1462 .fadvise = xfs_file_fadvise,
1463 .remap_file_range = xfs_file_remap_range,
1464 };
1465
1466 const struct file_operations xfs_dir_file_operations = {
1467 .open = xfs_dir_open,
1468 .read = generic_read_dir,
1469 .iterate_shared = xfs_file_readdir,
1470 .llseek = generic_file_llseek,
1471 .unlocked_ioctl = xfs_file_ioctl,
1472 #ifdef CONFIG_COMPAT
1473 .compat_ioctl = xfs_file_compat_ioctl,
1474 #endif
1475 .fsync = xfs_dir_fsync,
1476 };
1477