Commit | Line | Data |
---|---|---|
0b61f8a4 | 1 | // SPDX-License-Identifier: GPL-2.0 |
1da177e4 | 2 | /* |
7b718769 NS |
3 | * Copyright (c) 2000-2005 Silicon Graphics, Inc. |
4 | * All Rights Reserved. | |
1da177e4 | 5 | */ |
1da177e4 | 6 | #include "xfs.h" |
dda35b8f | 7 | #include "xfs_fs.h" |
70a9883c | 8 | #include "xfs_shared.h" |
a4fbe6ab | 9 | #include "xfs_format.h" |
239880ef DC |
10 | #include "xfs_log_format.h" |
11 | #include "xfs_trans_resv.h" | |
1da177e4 | 12 | #include "xfs_mount.h" |
1da177e4 | 13 | #include "xfs_inode.h" |
239880ef | 14 | #include "xfs_trans.h" |
fd3200be | 15 | #include "xfs_inode_item.h" |
dda35b8f | 16 | #include "xfs_bmap.h" |
c24b5dfa | 17 | #include "xfs_bmap_util.h" |
2b9ab5ab | 18 | #include "xfs_dir2.h" |
c24b5dfa | 19 | #include "xfs_dir2_priv.h" |
ddcd856d | 20 | #include "xfs_ioctl.h" |
dda35b8f | 21 | #include "xfs_trace.h" |
239880ef | 22 | #include "xfs_log.h" |
dc06f398 | 23 | #include "xfs_icache.h" |
781355c6 | 24 | #include "xfs_pnfs.h" |
68a9f5e7 | 25 | #include "xfs_iomap.h" |
0613f16c | 26 | #include "xfs_reflink.h" |
1da177e4 | 27 | |
ea6c49b7 | 28 | #include <linux/dax.h> |
2fe17c10 | 29 | #include <linux/falloc.h> |
66114cad | 30 | #include <linux/backing-dev.h> |
a39e596b | 31 | #include <linux/mman.h> |
40144e49 | 32 | #include <linux/fadvise.h> |
f736d93d | 33 | #include <linux/mount.h> |
1da177e4 | 34 | |
f0f37e2f | 35 | static const struct vm_operations_struct xfs_file_vm_ops; |
1da177e4 | 36 | |
25219dbf DW |
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 | |
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 | ||
1da2f2db CH |
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 | |
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); | |
1da2f2db CH |
84 | |
85 | trace_xfs_dir_fsync(ip); | |
54fbdd10 | 86 | return xfs_log_force_inode(ip); |
1da2f2db CH |
87 | } |
88 | ||
5f9b4b0d DC |
89 | static xfs_csn_t |
90 | xfs_fsync_seq( | |
f22c7f87 CH |
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; | |
5f9b4b0d | 98 | return ip->i_itemp->ili_commit_seq; |
f22c7f87 CH |
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 | |
115 | xfs_fsync_flush_log( | |
116 | struct xfs_inode *ip, | |
117 | bool datasync, | |
118 | int *log_flushed) | |
119 | { | |
120 | int error = 0; | |
5f9b4b0d | 121 | xfs_csn_t seq; |
f22c7f87 CH |
122 | |
123 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
5f9b4b0d DC |
124 | seq = xfs_fsync_seq(ip, datasync); |
125 | if (seq) { | |
126 | error = xfs_log_force_seq(ip->i_mount, seq, XFS_LOG_SYNC, | |
f22c7f87 CH |
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 | ||
fd3200be CH |
137 | STATIC int |
138 | xfs_file_fsync( | |
139 | struct file *file, | |
02c24a82 JB |
140 | loff_t start, |
141 | loff_t end, | |
fd3200be CH |
142 | int datasync) |
143 | { | |
f22c7f87 | 144 | struct xfs_inode *ip = XFS_I(file->f_mapping->host); |
a27a263b | 145 | struct xfs_mount *mp = ip->i_mount; |
7d839e32 | 146 | int error, err2; |
fd3200be CH |
147 | int log_flushed = 0; |
148 | ||
cca28fb8 | 149 | trace_xfs_file_fsync(ip); |
fd3200be | 150 | |
1b180274 | 151 | error = file_write_and_wait_range(file, start, end); |
02c24a82 JB |
152 | if (error) |
153 | return error; | |
154 | ||
75c8c50f | 155 | if (xfs_is_shutdown(mp)) |
b474c7ae | 156 | return -EIO; |
fd3200be CH |
157 | |
158 | xfs_iflags_clear(ip, XFS_ITRUNCATED); | |
159 | ||
2291dab2 DC |
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)) | |
7d839e32 | 167 | error = blkdev_issue_flush(mp->m_rtdev_targp->bt_bdev); |
2291dab2 | 168 | else if (mp->m_logdev_targp != mp->m_ddev_targp) |
7d839e32 | 169 | error = blkdev_issue_flush(mp->m_ddev_targp->bt_bdev); |
a27a263b | 170 | |
fd3200be | 171 | /* |
ae29e422 | 172 | * Any inode that has dirty modifications in the log is pinned. The |
7d839e32 | 173 | * racy check here for a pinned inode will not catch modifications |
ae29e422 CH |
174 | * that happen concurrently to the fsync call, but fsync semantics |
175 | * only require to sync previously completed I/O. | |
fd3200be | 176 | */ |
7d839e32 DW |
177 | if (xfs_ipincount(ip)) { |
178 | err2 = xfs_fsync_flush_log(ip, datasync, &log_flushed); | |
179 | if (err2 && !error) | |
180 | error = err2; | |
181 | } | |
b1037058 | 182 | |
a27a263b CH |
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 | */ | |
2291dab2 | 190 | if (!log_flushed && !XFS_IS_REALTIME_INODE(ip) && |
7d839e32 DW |
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 | } | |
fd3200be | 196 | |
2451337d | 197 | return error; |
fd3200be CH |
198 | } |
199 | ||
f50b8f47 CH |
200 | static int |
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 | ||
00258e36 | 217 | STATIC ssize_t |
ee1b218b | 218 | xfs_file_dio_read( |
dda35b8f | 219 | struct kiocb *iocb, |
b4f5d2c6 | 220 | struct iov_iter *to) |
dda35b8f | 221 | { |
acdda3aa | 222 | struct xfs_inode *ip = XFS_I(file_inode(iocb->ki_filp)); |
acdda3aa | 223 | ssize_t ret; |
dda35b8f | 224 | |
3e40b13c | 225 | trace_xfs_file_direct_read(iocb, to); |
dda35b8f | 226 | |
3e40b13c | 227 | if (!iov_iter_count(to)) |
f1285ff0 | 228 | return 0; /* skip atime */ |
dda35b8f | 229 | |
a447d7cd CH |
230 | file_accessed(iocb->ki_filp); |
231 | ||
f50b8f47 CH |
232 | ret = xfs_ilock_iocb(iocb, XFS_IOLOCK_SHARED); |
233 | if (ret) | |
234 | return ret; | |
786f847f | 235 | ret = iomap_dio_rw(iocb, to, &xfs_read_iomap_ops, NULL, 0, NULL, 0); |
65523218 | 236 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); |
acdda3aa | 237 | |
16d4d435 CH |
238 | return ret; |
239 | } | |
240 | ||
f021bd07 | 241 | static noinline ssize_t |
16d4d435 CH |
242 | xfs_file_dax_read( |
243 | struct kiocb *iocb, | |
244 | struct iov_iter *to) | |
245 | { | |
6c31f495 | 246 | struct xfs_inode *ip = XFS_I(iocb->ki_filp->f_mapping->host); |
16d4d435 CH |
247 | ssize_t ret = 0; |
248 | ||
3e40b13c | 249 | trace_xfs_file_dax_read(iocb, to); |
16d4d435 | 250 | |
3e40b13c | 251 | if (!iov_iter_count(to)) |
16d4d435 CH |
252 | return 0; /* skip atime */ |
253 | ||
f50b8f47 CH |
254 | ret = xfs_ilock_iocb(iocb, XFS_IOLOCK_SHARED); |
255 | if (ret) | |
256 | return ret; | |
690c2a38 | 257 | ret = dax_iomap_rw(iocb, to, &xfs_read_iomap_ops); |
65523218 | 258 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); |
bbc5a740 | 259 | |
f1285ff0 | 260 | file_accessed(iocb->ki_filp); |
bbc5a740 CH |
261 | return ret; |
262 | } | |
263 | ||
264 | STATIC ssize_t | |
ee1b218b | 265 | xfs_file_buffered_read( |
bbc5a740 CH |
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 | ||
3e40b13c | 272 | trace_xfs_file_buffered_read(iocb, to); |
dda35b8f | 273 | |
f50b8f47 CH |
274 | ret = xfs_ilock_iocb(iocb, XFS_IOLOCK_SHARED); |
275 | if (ret) | |
276 | return ret; | |
b4f5d2c6 | 277 | ret = generic_file_read_iter(iocb, to); |
65523218 | 278 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); |
bbc5a740 CH |
279 | |
280 | return ret; | |
281 | } | |
282 | ||
283 | STATIC ssize_t | |
284 | xfs_file_read_iter( | |
285 | struct kiocb *iocb, | |
286 | struct iov_iter *to) | |
287 | { | |
16d4d435 CH |
288 | struct inode *inode = file_inode(iocb->ki_filp); |
289 | struct xfs_mount *mp = XFS_I(inode)->i_mount; | |
bbc5a740 CH |
290 | ssize_t ret = 0; |
291 | ||
292 | XFS_STATS_INC(mp, xs_read_calls); | |
293 | ||
75c8c50f | 294 | if (xfs_is_shutdown(mp)) |
bbc5a740 CH |
295 | return -EIO; |
296 | ||
16d4d435 CH |
297 | if (IS_DAX(inode)) |
298 | ret = xfs_file_dax_read(iocb, to); | |
299 | else if (iocb->ki_flags & IOCB_DIRECT) | |
ee1b218b | 300 | ret = xfs_file_dio_read(iocb, to); |
3176c3e0 | 301 | else |
ee1b218b | 302 | ret = xfs_file_buffered_read(iocb, to); |
dda35b8f | 303 | |
dda35b8f | 304 | if (ret > 0) |
ff6d6af2 | 305 | XFS_STATS_ADD(mp, xs_read_bytes, ret); |
dda35b8f CH |
306 | return ret; |
307 | } | |
308 | ||
4d8d1581 DC |
309 | /* |
310 | * Common pre-write limit and setup checks. | |
311 | * | |
5bf1f262 CH |
312 | * Called with the iolocked held either shared and exclusive according to |
313 | * @iolock, and returns with it held. Might upgrade the iolock to exclusive | |
314 | * if called for a direct write beyond i_size. | |
4d8d1581 DC |
315 | */ |
316 | STATIC ssize_t | |
ee1b218b | 317 | xfs_file_write_checks( |
99733fa3 AV |
318 | struct kiocb *iocb, |
319 | struct iov_iter *from, | |
a1033753 | 320 | unsigned int *iolock) |
4d8d1581 | 321 | { |
99733fa3 | 322 | struct file *file = iocb->ki_filp; |
4d8d1581 DC |
323 | struct inode *inode = file->f_mapping->host; |
324 | struct xfs_inode *ip = XFS_I(inode); | |
3309dd04 | 325 | ssize_t error = 0; |
99733fa3 | 326 | size_t count = iov_iter_count(from); |
3136e8bb | 327 | bool drained_dio = false; |
f5c54717 | 328 | loff_t isize; |
4d8d1581 | 329 | |
7271d243 | 330 | restart: |
3309dd04 AV |
331 | error = generic_write_checks(iocb, from); |
332 | if (error <= 0) | |
4d8d1581 | 333 | return error; |
4d8d1581 | 334 | |
354be7e3 CH |
335 | if (iocb->ki_flags & IOCB_NOWAIT) { |
336 | error = break_layout(inode, false); | |
337 | if (error == -EWOULDBLOCK) | |
338 | error = -EAGAIN; | |
339 | } else { | |
340 | error = xfs_break_layouts(inode, iolock, BREAK_WRITE); | |
341 | } | |
342 | ||
781355c6 CH |
343 | if (error) |
344 | return error; | |
345 | ||
65523218 CH |
346 | /* |
347 | * For changing security info in file_remove_privs() we need i_rwsem | |
348 | * exclusively. | |
349 | */ | |
a6de82ca | 350 | if (*iolock == XFS_IOLOCK_SHARED && !IS_NOSEC(inode)) { |
65523218 | 351 | xfs_iunlock(ip, *iolock); |
a6de82ca | 352 | *iolock = XFS_IOLOCK_EXCL; |
354be7e3 CH |
353 | error = xfs_ilock_iocb(iocb, *iolock); |
354 | if (error) { | |
355 | *iolock = 0; | |
356 | return error; | |
357 | } | |
a6de82ca JK |
358 | goto restart; |
359 | } | |
977ec4dd | 360 | |
4d8d1581 DC |
361 | /* |
362 | * If the offset is beyond the size of the file, we need to zero any | |
363 | * blocks that fall between the existing EOF and the start of this | |
977ec4dd DC |
364 | * write. If zeroing is needed and we are currently holding the iolock |
365 | * shared, we need to update it to exclusive which implies having to | |
366 | * redo all checks before. | |
367 | * | |
368 | * We need to serialise against EOF updates that occur in IO completions | |
369 | * here. We want to make sure that nobody is changing the size while we | |
370 | * do this check until we have placed an IO barrier (i.e. hold the | |
371 | * XFS_IOLOCK_EXCL) that prevents new IO from being dispatched. The | |
372 | * spinlock effectively forms a memory barrier once we have the | |
373 | * XFS_IOLOCK_EXCL so we are guaranteed to see the latest EOF value and | |
374 | * hence be able to correctly determine if we need to run zeroing. | |
b9d59846 | 375 | * |
977ec4dd DC |
376 | * We can do an unlocked check here safely as IO completion can only |
377 | * extend EOF. Truncate is locked out at this point, so the EOF can | |
378 | * not move backwards, only forwards. Hence we only need to take the | |
379 | * slow path and spin locks when we are at or beyond the current EOF. | |
4d8d1581 | 380 | */ |
977ec4dd DC |
381 | if (iocb->ki_pos <= i_size_read(inode)) |
382 | goto out; | |
383 | ||
b9d59846 | 384 | spin_lock(&ip->i_flags_lock); |
f5c54717 CH |
385 | isize = i_size_read(inode); |
386 | if (iocb->ki_pos > isize) { | |
b9d59846 | 387 | spin_unlock(&ip->i_flags_lock); |
354be7e3 CH |
388 | |
389 | if (iocb->ki_flags & IOCB_NOWAIT) | |
390 | return -EAGAIN; | |
391 | ||
3136e8bb BF |
392 | if (!drained_dio) { |
393 | if (*iolock == XFS_IOLOCK_SHARED) { | |
65523218 | 394 | xfs_iunlock(ip, *iolock); |
3136e8bb | 395 | *iolock = XFS_IOLOCK_EXCL; |
65523218 | 396 | xfs_ilock(ip, *iolock); |
3136e8bb BF |
397 | iov_iter_reexpand(from, count); |
398 | } | |
40c63fbc DC |
399 | /* |
400 | * We now have an IO submission barrier in place, but | |
401 | * AIO can do EOF updates during IO completion and hence | |
402 | * we now need to wait for all of them to drain. Non-AIO | |
403 | * DIO will have drained before we are given the | |
404 | * XFS_IOLOCK_EXCL, and so for most cases this wait is a | |
405 | * no-op. | |
406 | */ | |
407 | inode_dio_wait(inode); | |
3136e8bb | 408 | drained_dio = true; |
7271d243 DC |
409 | goto restart; |
410 | } | |
977ec4dd | 411 | |
f5c54717 | 412 | trace_xfs_zero_eof(ip, isize, iocb->ki_pos - isize); |
f1ba5faf | 413 | error = xfs_zero_range(ip, isize, iocb->ki_pos - isize, NULL); |
467f7899 CH |
414 | if (error) |
415 | return error; | |
b9d59846 DC |
416 | } else |
417 | spin_unlock(&ip->i_flags_lock); | |
4d8d1581 | 418 | |
977ec4dd | 419 | out: |
1aa91d9c | 420 | return kiocb_modified(iocb); |
4d8d1581 DC |
421 | } |
422 | ||
acdda3aa CH |
423 | static int |
424 | xfs_dio_write_end_io( | |
425 | struct kiocb *iocb, | |
426 | ssize_t size, | |
6fe7b990 | 427 | int error, |
acdda3aa CH |
428 | unsigned flags) |
429 | { | |
430 | struct inode *inode = file_inode(iocb->ki_filp); | |
431 | struct xfs_inode *ip = XFS_I(inode); | |
432 | loff_t offset = iocb->ki_pos; | |
73d30d48 | 433 | unsigned int nofs_flag; |
acdda3aa CH |
434 | |
435 | trace_xfs_end_io_direct_write(ip, offset, size); | |
436 | ||
75c8c50f | 437 | if (xfs_is_shutdown(ip->i_mount)) |
acdda3aa CH |
438 | return -EIO; |
439 | ||
6fe7b990 MB |
440 | if (error) |
441 | return error; | |
442 | if (!size) | |
443 | return 0; | |
acdda3aa | 444 | |
ed5c3e66 DC |
445 | /* |
446 | * Capture amount written on completion as we can't reliably account | |
447 | * for it on submission. | |
448 | */ | |
449 | XFS_STATS_ADD(ip->i_mount, xs_write_bytes, size); | |
450 | ||
73d30d48 CH |
451 | /* |
452 | * We can allocate memory here while doing writeback on behalf of | |
453 | * memory reclaim. To avoid memory allocation deadlocks set the | |
454 | * task-wide nofs context for the following operations. | |
455 | */ | |
456 | nofs_flag = memalloc_nofs_save(); | |
457 | ||
ee70daab EG |
458 | if (flags & IOMAP_DIO_COW) { |
459 | error = xfs_reflink_end_cow(ip, offset, size); | |
460 | if (error) | |
73d30d48 | 461 | goto out; |
ee70daab EG |
462 | } |
463 | ||
464 | /* | |
465 | * Unwritten conversion updates the in-core isize after extent | |
466 | * conversion but before updating the on-disk size. Updating isize any | |
467 | * earlier allows a racing dio read to find unwritten extents before | |
468 | * they are converted. | |
469 | */ | |
73d30d48 CH |
470 | if (flags & IOMAP_DIO_UNWRITTEN) { |
471 | error = xfs_iomap_write_unwritten(ip, offset, size, true); | |
472 | goto out; | |
473 | } | |
ee70daab | 474 | |
acdda3aa CH |
475 | /* |
476 | * We need to update the in-core inode size here so that we don't end up | |
477 | * with the on-disk inode size being outside the in-core inode size. We | |
478 | * have no other method of updating EOF for AIO, so always do it here | |
479 | * if necessary. | |
480 | * | |
481 | * We need to lock the test/set EOF update as we can be racing with | |
482 | * other IO completions here to update the EOF. Failing to serialise | |
483 | * here can result in EOF moving backwards and Bad Things Happen when | |
484 | * that occurs. | |
977ec4dd DC |
485 | * |
486 | * As IO completion only ever extends EOF, we can do an unlocked check | |
487 | * here to avoid taking the spinlock. If we land within the current EOF, | |
488 | * then we do not need to do an extending update at all, and we don't | |
489 | * need to take the lock to check this. If we race with an update moving | |
490 | * EOF, then we'll either still be beyond EOF and need to take the lock, | |
491 | * or we'll be within EOF and we don't need to take it at all. | |
acdda3aa | 492 | */ |
977ec4dd DC |
493 | if (offset + size <= i_size_read(inode)) |
494 | goto out; | |
495 | ||
acdda3aa CH |
496 | spin_lock(&ip->i_flags_lock); |
497 | if (offset + size > i_size_read(inode)) { | |
498 | i_size_write(inode, offset + size); | |
ee70daab | 499 | spin_unlock(&ip->i_flags_lock); |
acdda3aa | 500 | error = xfs_setfilesize(ip, offset, size); |
ee70daab EG |
501 | } else { |
502 | spin_unlock(&ip->i_flags_lock); | |
503 | } | |
acdda3aa | 504 | |
73d30d48 CH |
505 | out: |
506 | memalloc_nofs_restore(nofs_flag); | |
acdda3aa CH |
507 | return error; |
508 | } | |
509 | ||
838c4f3d CH |
510 | static const struct iomap_dio_ops xfs_dio_write_ops = { |
511 | .end_io = xfs_dio_write_end_io, | |
512 | }; | |
513 | ||
f0d26e86 | 514 | /* |
caa89dbc | 515 | * Handle block aligned direct I/O writes |
f0d26e86 | 516 | */ |
caa89dbc DC |
517 | static noinline ssize_t |
518 | xfs_file_dio_write_aligned( | |
519 | struct xfs_inode *ip, | |
f0d26e86 | 520 | struct kiocb *iocb, |
b3188919 | 521 | struct iov_iter *from) |
f0d26e86 | 522 | { |
a1033753 | 523 | unsigned int iolock = XFS_IOLOCK_SHARED; |
caa89dbc | 524 | ssize_t ret; |
f0d26e86 | 525 | |
caa89dbc DC |
526 | ret = xfs_ilock_iocb(iocb, iolock); |
527 | if (ret) | |
528 | return ret; | |
529 | ret = xfs_file_write_checks(iocb, from, &iolock); | |
530 | if (ret) | |
531 | goto out_unlock; | |
f0d26e86 | 532 | |
7271d243 | 533 | /* |
caa89dbc DC |
534 | * We don't need to hold the IOLOCK exclusively across the IO, so demote |
535 | * the iolock back to shared if we had to take the exclusive lock in | |
536 | * xfs_file_write_checks() for other reasons. | |
7271d243 | 537 | */ |
caa89dbc DC |
538 | if (iolock == XFS_IOLOCK_EXCL) { |
539 | xfs_ilock_demote(ip, XFS_IOLOCK_EXCL); | |
d0606464 | 540 | iolock = XFS_IOLOCK_SHARED; |
c58cb165 | 541 | } |
caa89dbc DC |
542 | trace_xfs_file_direct_write(iocb, from); |
543 | ret = iomap_dio_rw(iocb, from, &xfs_direct_write_iomap_ops, | |
786f847f | 544 | &xfs_dio_write_ops, 0, NULL, 0); |
caa89dbc DC |
545 | out_unlock: |
546 | if (iolock) | |
547 | xfs_iunlock(ip, iolock); | |
548 | return ret; | |
549 | } | |
f0d26e86 | 550 | |
caa89dbc DC |
551 | /* |
552 | * Handle block unaligned direct I/O writes | |
553 | * | |
554 | * In most cases direct I/O writes will be done holding IOLOCK_SHARED, allowing | |
555 | * them to be done in parallel with reads and other direct I/O writes. However, | |
556 | * if the I/O is not aligned to filesystem blocks, the direct I/O layer may need | |
557 | * to do sub-block zeroing and that requires serialisation against other direct | |
558 | * I/O to the same block. In this case we need to serialise the submission of | |
559 | * the unaligned I/O so that we don't get racing block zeroing in the dio layer. | |
ed1128c2 DC |
560 | * In the case where sub-block zeroing is not required, we can do concurrent |
561 | * sub-block dios to the same block successfully. | |
caa89dbc | 562 | * |
ed1128c2 DC |
563 | * Optimistically submit the I/O using the shared lock first, but use the |
564 | * IOMAP_DIO_OVERWRITE_ONLY flag to tell the lower layers to return -EAGAIN | |
565 | * if block allocation or partial block zeroing would be required. In that case | |
566 | * we try again with the exclusive lock. | |
caa89dbc DC |
567 | */ |
568 | static noinline ssize_t | |
569 | xfs_file_dio_write_unaligned( | |
570 | struct xfs_inode *ip, | |
571 | struct kiocb *iocb, | |
572 | struct iov_iter *from) | |
573 | { | |
ed1128c2 DC |
574 | size_t isize = i_size_read(VFS_I(ip)); |
575 | size_t count = iov_iter_count(from); | |
a1033753 | 576 | unsigned int iolock = XFS_IOLOCK_SHARED; |
ed1128c2 | 577 | unsigned int flags = IOMAP_DIO_OVERWRITE_ONLY; |
caa89dbc DC |
578 | ssize_t ret; |
579 | ||
ed1128c2 DC |
580 | /* |
581 | * Extending writes need exclusivity because of the sub-block zeroing | |
582 | * that the DIO code always does for partial tail blocks beyond EOF, so | |
583 | * don't even bother trying the fast path in this case. | |
584 | */ | |
585 | if (iocb->ki_pos > isize || iocb->ki_pos + count >= isize) { | |
ed1128c2 DC |
586 | if (iocb->ki_flags & IOCB_NOWAIT) |
587 | return -EAGAIN; | |
93e6aa43 | 588 | retry_exclusive: |
ed1128c2 DC |
589 | iolock = XFS_IOLOCK_EXCL; |
590 | flags = IOMAP_DIO_FORCE_WAIT; | |
591 | } | |
592 | ||
593 | ret = xfs_ilock_iocb(iocb, iolock); | |
594 | if (ret) | |
595 | return ret; | |
caa89dbc DC |
596 | |
597 | /* | |
598 | * We can't properly handle unaligned direct I/O to reflink files yet, | |
599 | * as we can't unshare a partial block. | |
600 | */ | |
601 | if (xfs_is_cow_inode(ip)) { | |
602 | trace_xfs_reflink_bounce_dio_write(iocb, from); | |
603 | ret = -ENOTBLK; | |
604 | goto out_unlock; | |
29a5d29e | 605 | } |
0ee7a3f6 | 606 | |
ee1b218b | 607 | ret = xfs_file_write_checks(iocb, from, &iolock); |
4d8d1581 | 608 | if (ret) |
caa89dbc | 609 | goto out_unlock; |
f0d26e86 | 610 | |
eda77982 | 611 | /* |
ed1128c2 DC |
612 | * If we are doing exclusive unaligned I/O, this must be the only I/O |
613 | * in-flight. Otherwise we risk data corruption due to unwritten extent | |
614 | * conversions from the AIO end_io handler. Wait for all other I/O to | |
615 | * drain first. | |
eda77982 | 616 | */ |
ed1128c2 DC |
617 | if (flags & IOMAP_DIO_FORCE_WAIT) |
618 | inode_dio_wait(VFS_I(ip)); | |
f0d26e86 | 619 | |
3e40b13c | 620 | trace_xfs_file_direct_write(iocb, from); |
f150b423 | 621 | ret = iomap_dio_rw(iocb, from, &xfs_direct_write_iomap_ops, |
786f847f | 622 | &xfs_dio_write_ops, flags, NULL, 0); |
ed1128c2 DC |
623 | |
624 | /* | |
625 | * Retry unaligned I/O with exclusive blocking semantics if the DIO | |
626 | * layer rejected it for mapping or locking reasons. If we are doing | |
627 | * nonblocking user I/O, propagate the error. | |
628 | */ | |
629 | if (ret == -EAGAIN && !(iocb->ki_flags & IOCB_NOWAIT)) { | |
630 | ASSERT(flags & IOMAP_DIO_OVERWRITE_ONLY); | |
631 | xfs_iunlock(ip, iolock); | |
632 | goto retry_exclusive; | |
633 | } | |
634 | ||
caa89dbc | 635 | out_unlock: |
354be7e3 CH |
636 | if (iolock) |
637 | xfs_iunlock(ip, iolock); | |
16d4d435 CH |
638 | return ret; |
639 | } | |
640 | ||
caa89dbc DC |
641 | static ssize_t |
642 | xfs_file_dio_write( | |
643 | struct kiocb *iocb, | |
644 | struct iov_iter *from) | |
645 | { | |
646 | struct xfs_inode *ip = XFS_I(file_inode(iocb->ki_filp)); | |
647 | struct xfs_buftarg *target = xfs_inode_buftarg(ip); | |
648 | size_t count = iov_iter_count(from); | |
649 | ||
650 | /* direct I/O must be aligned to device logical sector size */ | |
651 | if ((iocb->ki_pos | count) & target->bt_logical_sectormask) | |
652 | return -EINVAL; | |
653 | if ((iocb->ki_pos | count) & ip->i_mount->m_blockmask) | |
654 | return xfs_file_dio_write_unaligned(ip, iocb, from); | |
655 | return xfs_file_dio_write_aligned(ip, iocb, from); | |
656 | } | |
657 | ||
f021bd07 | 658 | static noinline ssize_t |
16d4d435 CH |
659 | xfs_file_dax_write( |
660 | struct kiocb *iocb, | |
661 | struct iov_iter *from) | |
662 | { | |
6c31f495 | 663 | struct inode *inode = iocb->ki_filp->f_mapping->host; |
16d4d435 | 664 | struct xfs_inode *ip = XFS_I(inode); |
a1033753 | 665 | unsigned int iolock = XFS_IOLOCK_EXCL; |
6c31f495 | 666 | ssize_t ret, error = 0; |
6c31f495 | 667 | loff_t pos; |
16d4d435 | 668 | |
f50b8f47 CH |
669 | ret = xfs_ilock_iocb(iocb, iolock); |
670 | if (ret) | |
671 | return ret; | |
ee1b218b | 672 | ret = xfs_file_write_checks(iocb, from, &iolock); |
16d4d435 CH |
673 | if (ret) |
674 | goto out; | |
675 | ||
6c31f495 | 676 | pos = iocb->ki_pos; |
8b2180b3 | 677 | |
3e40b13c | 678 | trace_xfs_file_dax_write(iocb, from); |
ea6c49b7 | 679 | ret = dax_iomap_rw(iocb, from, &xfs_dax_write_iomap_ops); |
6c31f495 CH |
680 | if (ret > 0 && iocb->ki_pos > i_size_read(inode)) { |
681 | i_size_write(inode, iocb->ki_pos); | |
682 | error = xfs_setfilesize(ip, pos, ret); | |
16d4d435 | 683 | } |
16d4d435 | 684 | out: |
354be7e3 CH |
685 | if (iolock) |
686 | xfs_iunlock(ip, iolock); | |
ed5c3e66 DC |
687 | if (error) |
688 | return error; | |
689 | ||
690 | if (ret > 0) { | |
691 | XFS_STATS_ADD(ip->i_mount, xs_write_bytes, ret); | |
692 | ||
693 | /* Handle various SYNC-type writes */ | |
694 | ret = generic_write_sync(iocb, ret); | |
695 | } | |
696 | return ret; | |
f0d26e86 DC |
697 | } |
698 | ||
00258e36 | 699 | STATIC ssize_t |
ee1b218b | 700 | xfs_file_buffered_write( |
dda35b8f | 701 | struct kiocb *iocb, |
b3188919 | 702 | struct iov_iter *from) |
dda35b8f | 703 | { |
2d9ac431 | 704 | struct inode *inode = iocb->ki_filp->f_mapping->host; |
00258e36 | 705 | struct xfs_inode *ip = XFS_I(inode); |
637bbc75 | 706 | ssize_t ret; |
a636b1d1 | 707 | bool cleared_space = false; |
a1033753 | 708 | unsigned int iolock; |
dda35b8f | 709 | |
c3155097 BF |
710 | write_retry: |
711 | iolock = XFS_IOLOCK_EXCL; | |
1aa91d9c SR |
712 | ret = xfs_ilock_iocb(iocb, iolock); |
713 | if (ret) | |
714 | return ret; | |
dda35b8f | 715 | |
ee1b218b | 716 | ret = xfs_file_write_checks(iocb, from, &iolock); |
4d8d1581 | 717 | if (ret) |
d0606464 | 718 | goto out; |
dda35b8f CH |
719 | |
720 | /* We can write back this queue in page reclaim */ | |
de1414a6 | 721 | current->backing_dev_info = inode_to_bdi(inode); |
dda35b8f | 722 | |
3e40b13c | 723 | trace_xfs_file_buffered_write(iocb, from); |
f150b423 CH |
724 | ret = iomap_file_buffered_write(iocb, from, |
725 | &xfs_buffered_write_iomap_ops); | |
0a64bc2c | 726 | if (likely(ret >= 0)) |
99733fa3 | 727 | iocb->ki_pos += ret; |
dc06f398 | 728 | |
637bbc75 | 729 | /* |
dc06f398 BF |
730 | * If we hit a space limit, try to free up some lingering preallocated |
731 | * space before returning an error. In the case of ENOSPC, first try to | |
732 | * write back all dirty inodes to free up some of the excess reserved | |
733 | * metadata space. This reduces the chances that the eofblocks scan | |
734 | * waits on dirty mappings. Since xfs_flush_inodes() is serialized, this | |
735 | * also behaves as a filter to prevent too many eofblocks scans from | |
111068f8 DW |
736 | * running at the same time. Use a synchronous scan to increase the |
737 | * effectiveness of the scan. | |
637bbc75 | 738 | */ |
a636b1d1 | 739 | if (ret == -EDQUOT && !cleared_space) { |
c3155097 | 740 | xfs_iunlock(ip, iolock); |
2d53f66b | 741 | xfs_blockgc_free_quota(ip, XFS_ICWALK_FLAG_SYNC); |
111068f8 DW |
742 | cleared_space = true; |
743 | goto write_retry; | |
a636b1d1 | 744 | } else if (ret == -ENOSPC && !cleared_space) { |
b26b2bf1 | 745 | struct xfs_icwalk icw = {0}; |
dc06f398 | 746 | |
a636b1d1 | 747 | cleared_space = true; |
9aa05000 | 748 | xfs_flush_inodes(ip->i_mount); |
c3155097 BF |
749 | |
750 | xfs_iunlock(ip, iolock); | |
b26b2bf1 DW |
751 | icw.icw_flags = XFS_ICWALK_FLAG_SYNC; |
752 | xfs_blockgc_free_space(ip->i_mount, &icw); | |
9aa05000 | 753 | goto write_retry; |
dda35b8f | 754 | } |
d0606464 | 755 | |
dda35b8f | 756 | current->backing_dev_info = NULL; |
d0606464 | 757 | out: |
c3155097 BF |
758 | if (iolock) |
759 | xfs_iunlock(ip, iolock); | |
ed5c3e66 DC |
760 | |
761 | if (ret > 0) { | |
762 | XFS_STATS_ADD(ip->i_mount, xs_write_bytes, ret); | |
763 | /* Handle various SYNC-type writes */ | |
764 | ret = generic_write_sync(iocb, ret); | |
765 | } | |
637bbc75 DC |
766 | return ret; |
767 | } | |
768 | ||
769 | STATIC ssize_t | |
bf97f3bc | 770 | xfs_file_write_iter( |
637bbc75 | 771 | struct kiocb *iocb, |
bf97f3bc | 772 | struct iov_iter *from) |
637bbc75 | 773 | { |
2d9ac431 | 774 | struct inode *inode = iocb->ki_filp->f_mapping->host; |
637bbc75 DC |
775 | struct xfs_inode *ip = XFS_I(inode); |
776 | ssize_t ret; | |
bf97f3bc | 777 | size_t ocount = iov_iter_count(from); |
637bbc75 | 778 | |
ff6d6af2 | 779 | XFS_STATS_INC(ip->i_mount, xs_write_calls); |
637bbc75 | 780 | |
637bbc75 DC |
781 | if (ocount == 0) |
782 | return 0; | |
783 | ||
75c8c50f | 784 | if (xfs_is_shutdown(ip->i_mount)) |
bf97f3bc | 785 | return -EIO; |
637bbc75 | 786 | |
16d4d435 | 787 | if (IS_DAX(inode)) |
ed5c3e66 DC |
788 | return xfs_file_dax_write(iocb, from); |
789 | ||
790 | if (iocb->ki_flags & IOCB_DIRECT) { | |
0613f16c DW |
791 | /* |
792 | * Allow a directio write to fall back to a buffered | |
793 | * write *only* in the case that we're doing a reflink | |
794 | * CoW. In all other directio scenarios we do not | |
795 | * allow an operation to fall back to buffered mode. | |
796 | */ | |
ee1b218b | 797 | ret = xfs_file_dio_write(iocb, from); |
80e543ae | 798 | if (ret != -ENOTBLK) |
ed5c3e66 | 799 | return ret; |
0613f16c | 800 | } |
dda35b8f | 801 | |
ee1b218b | 802 | return xfs_file_buffered_write(iocb, from); |
dda35b8f CH |
803 | } |
804 | ||
d6dc57e2 DW |
805 | static void |
806 | xfs_wait_dax_page( | |
e25ff835 | 807 | struct inode *inode) |
d6dc57e2 DW |
808 | { |
809 | struct xfs_inode *ip = XFS_I(inode); | |
810 | ||
d6dc57e2 DW |
811 | xfs_iunlock(ip, XFS_MMAPLOCK_EXCL); |
812 | schedule(); | |
813 | xfs_ilock(ip, XFS_MMAPLOCK_EXCL); | |
814 | } | |
815 | ||
13f9e267 | 816 | int |
d6dc57e2 DW |
817 | xfs_break_dax_layouts( |
818 | struct inode *inode, | |
e25ff835 | 819 | bool *retry) |
d6dc57e2 DW |
820 | { |
821 | struct page *page; | |
822 | ||
823 | ASSERT(xfs_isilocked(XFS_I(inode), XFS_MMAPLOCK_EXCL)); | |
824 | ||
825 | page = dax_layout_busy_page(inode->i_mapping); | |
826 | if (!page) | |
827 | return 0; | |
828 | ||
e25ff835 | 829 | *retry = true; |
d6dc57e2 DW |
830 | return ___wait_var_event(&page->_refcount, |
831 | atomic_read(&page->_refcount) == 1, TASK_INTERRUPTIBLE, | |
e25ff835 | 832 | 0, 0, xfs_wait_dax_page(inode)); |
d6dc57e2 DW |
833 | } |
834 | ||
69eb5fa1 DW |
835 | int |
836 | xfs_break_layouts( | |
837 | struct inode *inode, | |
838 | uint *iolock, | |
839 | enum layout_break_reason reason) | |
840 | { | |
841 | bool retry; | |
d6dc57e2 | 842 | int error; |
69eb5fa1 DW |
843 | |
844 | ASSERT(xfs_isilocked(XFS_I(inode), XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)); | |
845 | ||
d6dc57e2 DW |
846 | do { |
847 | retry = false; | |
848 | switch (reason) { | |
849 | case BREAK_UNMAP: | |
a4722a64 | 850 | error = xfs_break_dax_layouts(inode, &retry); |
d6dc57e2 DW |
851 | if (error || retry) |
852 | break; | |
53004ee7 | 853 | fallthrough; |
d6dc57e2 DW |
854 | case BREAK_WRITE: |
855 | error = xfs_break_leased_layouts(inode, iolock, &retry); | |
856 | break; | |
857 | default: | |
858 | WARN_ON_ONCE(1); | |
859 | error = -EINVAL; | |
860 | } | |
861 | } while (error == 0 && retry); | |
862 | ||
863 | return error; | |
69eb5fa1 DW |
864 | } |
865 | ||
cea267c2 DC |
866 | /* Does this file, inode, or mount want synchronous writes? */ |
867 | static inline bool xfs_file_sync_writes(struct file *filp) | |
868 | { | |
869 | struct xfs_inode *ip = XFS_I(file_inode(filp)); | |
870 | ||
871 | if (xfs_has_wsync(ip->i_mount)) | |
872 | return true; | |
873 | if (filp->f_flags & (__O_SYNC | O_DSYNC)) | |
874 | return true; | |
875 | if (IS_SYNC(file_inode(filp))) | |
876 | return true; | |
877 | ||
878 | return false; | |
879 | } | |
880 | ||
a904b1ca NJ |
881 | #define XFS_FALLOC_FL_SUPPORTED \ |
882 | (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \ | |
883 | FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE | \ | |
98cc2db5 | 884 | FALLOC_FL_INSERT_RANGE | FALLOC_FL_UNSHARE_RANGE) |
a904b1ca | 885 | |
2fe17c10 CH |
886 | STATIC long |
887 | xfs_file_fallocate( | |
83aee9e4 CH |
888 | struct file *file, |
889 | int mode, | |
890 | loff_t offset, | |
891 | loff_t len) | |
2fe17c10 | 892 | { |
83aee9e4 CH |
893 | struct inode *inode = file_inode(file); |
894 | struct xfs_inode *ip = XFS_I(inode); | |
83aee9e4 | 895 | long error; |
c63a8eae | 896 | uint iolock = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL; |
83aee9e4 | 897 | loff_t new_size = 0; |
749f24f3 | 898 | bool do_file_insert = false; |
2fe17c10 | 899 | |
83aee9e4 CH |
900 | if (!S_ISREG(inode->i_mode)) |
901 | return -EINVAL; | |
a904b1ca | 902 | if (mode & ~XFS_FALLOC_FL_SUPPORTED) |
2fe17c10 CH |
903 | return -EOPNOTSUPP; |
904 | ||
781355c6 | 905 | xfs_ilock(ip, iolock); |
69eb5fa1 | 906 | error = xfs_break_layouts(inode, &iolock, BREAK_UNMAP); |
781355c6 CH |
907 | if (error) |
908 | goto out_unlock; | |
909 | ||
249bd908 DC |
910 | /* |
911 | * Must wait for all AIO to complete before we continue as AIO can | |
912 | * change the file size on completion without holding any locks we | |
913 | * currently hold. We must do this first because AIO can update both | |
914 | * the on disk and in memory inode sizes, and the operations that follow | |
915 | * require the in-memory size to be fully up-to-date. | |
916 | */ | |
917 | inode_dio_wait(inode); | |
918 | ||
919 | /* | |
920 | * Now AIO and DIO has drained we flush and (if necessary) invalidate | |
921 | * the cached range over the first operation we are about to run. | |
922 | * | |
923 | * We care about zero and collapse here because they both run a hole | |
924 | * punch over the range first. Because that can zero data, and the range | |
925 | * of invalidation for the shift operations is much larger, we still do | |
926 | * the required flush for collapse in xfs_prepare_shift(). | |
927 | * | |
928 | * Insert has the same range requirements as collapse, and we extend the | |
929 | * file first which can zero data. Hence insert has the same | |
930 | * flush/invalidate requirements as collapse and so they are both | |
931 | * handled at the right time by xfs_prepare_shift(). | |
932 | */ | |
933 | if (mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE | | |
934 | FALLOC_FL_COLLAPSE_RANGE)) { | |
935 | error = xfs_flush_unmap_range(ip, offset, len); | |
936 | if (error) | |
937 | goto out_unlock; | |
938 | } | |
939 | ||
fbe7e520 DC |
940 | error = file_modified(file); |
941 | if (error) | |
942 | goto out_unlock; | |
943 | ||
83aee9e4 CH |
944 | if (mode & FALLOC_FL_PUNCH_HOLE) { |
945 | error = xfs_free_file_space(ip, offset, len); | |
946 | if (error) | |
947 | goto out_unlock; | |
e1d8fb88 | 948 | } else if (mode & FALLOC_FL_COLLAPSE_RANGE) { |
25219dbf | 949 | if (!xfs_is_falloc_aligned(ip, offset, len)) { |
2451337d | 950 | error = -EINVAL; |
e1d8fb88 NJ |
951 | goto out_unlock; |
952 | } | |
953 | ||
23fffa92 LC |
954 | /* |
955 | * There is no need to overlap collapse range with EOF, | |
956 | * in which case it is effectively a truncate operation | |
957 | */ | |
958 | if (offset + len >= i_size_read(inode)) { | |
2451337d | 959 | error = -EINVAL; |
23fffa92 LC |
960 | goto out_unlock; |
961 | } | |
962 | ||
e1d8fb88 NJ |
963 | new_size = i_size_read(inode) - len; |
964 | ||
965 | error = xfs_collapse_file_space(ip, offset, len); | |
966 | if (error) | |
967 | goto out_unlock; | |
a904b1ca | 968 | } else if (mode & FALLOC_FL_INSERT_RANGE) { |
7d83fb14 | 969 | loff_t isize = i_size_read(inode); |
a904b1ca | 970 | |
25219dbf | 971 | if (!xfs_is_falloc_aligned(ip, offset, len)) { |
a904b1ca NJ |
972 | error = -EINVAL; |
973 | goto out_unlock; | |
974 | } | |
975 | ||
7d83fb14 DW |
976 | /* |
977 | * New inode size must not exceed ->s_maxbytes, accounting for | |
978 | * possible signed overflow. | |
979 | */ | |
980 | if (inode->i_sb->s_maxbytes - isize < len) { | |
a904b1ca NJ |
981 | error = -EFBIG; |
982 | goto out_unlock; | |
983 | } | |
7d83fb14 | 984 | new_size = isize + len; |
a904b1ca NJ |
985 | |
986 | /* Offset should be less than i_size */ | |
7d83fb14 | 987 | if (offset >= isize) { |
a904b1ca NJ |
988 | error = -EINVAL; |
989 | goto out_unlock; | |
990 | } | |
749f24f3 | 991 | do_file_insert = true; |
83aee9e4 CH |
992 | } else { |
993 | if (!(mode & FALLOC_FL_KEEP_SIZE) && | |
994 | offset + len > i_size_read(inode)) { | |
995 | new_size = offset + len; | |
2451337d | 996 | error = inode_newsize_ok(inode, new_size); |
83aee9e4 CH |
997 | if (error) |
998 | goto out_unlock; | |
999 | } | |
2fe17c10 | 1000 | |
66ae56a5 | 1001 | if (mode & FALLOC_FL_ZERO_RANGE) { |
360c09c0 CH |
1002 | /* |
1003 | * Punch a hole and prealloc the range. We use a hole | |
1004 | * punch rather than unwritten extent conversion for two | |
1005 | * reasons: | |
1006 | * | |
1007 | * 1.) Hole punch handles partial block zeroing for us. | |
1008 | * 2.) If prealloc returns ENOSPC, the file range is | |
1009 | * still zero-valued by virtue of the hole punch. | |
1010 | */ | |
1011 | unsigned int blksize = i_blocksize(inode); | |
1012 | ||
1013 | trace_xfs_zero_file_space(ip); | |
1014 | ||
1015 | error = xfs_free_file_space(ip, offset, len); | |
1016 | if (error) | |
1017 | goto out_unlock; | |
1018 | ||
1019 | len = round_up(offset + len, blksize) - | |
1020 | round_down(offset, blksize); | |
1021 | offset = round_down(offset, blksize); | |
66ae56a5 CH |
1022 | } else if (mode & FALLOC_FL_UNSHARE_RANGE) { |
1023 | error = xfs_reflink_unshare(ip, offset, len); | |
1024 | if (error) | |
1025 | goto out_unlock; | |
66ae56a5 CH |
1026 | } else { |
1027 | /* | |
1028 | * If always_cow mode we can't use preallocations and | |
1029 | * thus should not create them. | |
1030 | */ | |
1031 | if (xfs_is_always_cow_inode(ip)) { | |
1032 | error = -EOPNOTSUPP; | |
1033 | goto out_unlock; | |
1034 | } | |
360c09c0 | 1035 | } |
66ae56a5 | 1036 | |
360c09c0 | 1037 | if (!xfs_is_always_cow_inode(ip)) { |
4d1b97f9 | 1038 | error = xfs_alloc_file_space(ip, offset, len); |
360c09c0 CH |
1039 | if (error) |
1040 | goto out_unlock; | |
98cc2db5 | 1041 | } |
fbe7e520 | 1042 | } |
2fe17c10 CH |
1043 | |
1044 | /* Change file size if needed */ | |
1045 | if (new_size) { | |
1046 | struct iattr iattr; | |
1047 | ||
1048 | iattr.ia_valid = ATTR_SIZE; | |
1049 | iattr.ia_size = new_size; | |
c1632a0f | 1050 | error = xfs_vn_setattr_size(file_mnt_idmap(file), |
f736d93d | 1051 | file_dentry(file), &iattr); |
a904b1ca NJ |
1052 | if (error) |
1053 | goto out_unlock; | |
2fe17c10 CH |
1054 | } |
1055 | ||
a904b1ca NJ |
1056 | /* |
1057 | * Perform hole insertion now that the file size has been | |
1058 | * updated so that if we crash during the operation we don't | |
1059 | * leave shifted extents past EOF and hence losing access to | |
1060 | * the data that is contained within them. | |
1061 | */ | |
472c6e46 | 1062 | if (do_file_insert) { |
a904b1ca | 1063 | error = xfs_insert_file_space(ip, offset, len); |
472c6e46 DC |
1064 | if (error) |
1065 | goto out_unlock; | |
1066 | } | |
1067 | ||
cea267c2 | 1068 | if (xfs_file_sync_writes(file)) |
472c6e46 | 1069 | error = xfs_log_force_inode(ip); |
a904b1ca | 1070 | |
2fe17c10 | 1071 | out_unlock: |
781355c6 | 1072 | xfs_iunlock(ip, iolock); |
2451337d | 1073 | return error; |
2fe17c10 CH |
1074 | } |
1075 | ||
40144e49 JK |
1076 | STATIC int |
1077 | xfs_file_fadvise( | |
1078 | struct file *file, | |
1079 | loff_t start, | |
1080 | loff_t end, | |
1081 | int advice) | |
1082 | { | |
1083 | struct xfs_inode *ip = XFS_I(file_inode(file)); | |
1084 | int ret; | |
1085 | int lockflags = 0; | |
1086 | ||
1087 | /* | |
1088 | * Operations creating pages in page cache need protection from hole | |
1089 | * punching and similar ops | |
1090 | */ | |
1091 | if (advice == POSIX_FADV_WILLNEED) { | |
1092 | lockflags = XFS_IOLOCK_SHARED; | |
1093 | xfs_ilock(ip, lockflags); | |
1094 | } | |
1095 | ret = generic_fadvise(file, start, end, advice); | |
1096 | if (lockflags) | |
1097 | xfs_iunlock(ip, lockflags); | |
1098 | return ret; | |
1099 | } | |
3fc9f5e4 | 1100 | |
da034bcc | 1101 | STATIC loff_t |
2e5dfc99 | 1102 | xfs_file_remap_range( |
3fc9f5e4 DW |
1103 | struct file *file_in, |
1104 | loff_t pos_in, | |
1105 | struct file *file_out, | |
1106 | loff_t pos_out, | |
1107 | loff_t len, | |
1108 | unsigned int remap_flags) | |
9fe26045 | 1109 | { |
3fc9f5e4 DW |
1110 | struct inode *inode_in = file_inode(file_in); |
1111 | struct xfs_inode *src = XFS_I(inode_in); | |
1112 | struct inode *inode_out = file_inode(file_out); | |
1113 | struct xfs_inode *dest = XFS_I(inode_out); | |
1114 | struct xfs_mount *mp = src->i_mount; | |
1115 | loff_t remapped = 0; | |
1116 | xfs_extlen_t cowextsize; | |
1117 | int ret; | |
1118 | ||
2e5dfc99 DW |
1119 | if (remap_flags & ~(REMAP_FILE_DEDUP | REMAP_FILE_ADVISORY)) |
1120 | return -EINVAL; | |
cc714660 | 1121 | |
38c26bfd | 1122 | if (!xfs_has_reflink(mp)) |
3fc9f5e4 DW |
1123 | return -EOPNOTSUPP; |
1124 | ||
75c8c50f | 1125 | if (xfs_is_shutdown(mp)) |
3fc9f5e4 DW |
1126 | return -EIO; |
1127 | ||
1128 | /* Prepare and then clone file data. */ | |
1129 | ret = xfs_reflink_remap_prep(file_in, pos_in, file_out, pos_out, | |
1130 | &len, remap_flags); | |
451d34ee | 1131 | if (ret || len == 0) |
3fc9f5e4 DW |
1132 | return ret; |
1133 | ||
1134 | trace_xfs_reflink_remap_range(src, pos_in, len, dest, pos_out); | |
1135 | ||
1136 | ret = xfs_reflink_remap_blocks(src, pos_in, dest, pos_out, len, | |
1137 | &remapped); | |
1138 | if (ret) | |
1139 | goto out_unlock; | |
1140 | ||
1141 | /* | |
1142 | * Carry the cowextsize hint from src to dest if we're sharing the | |
1143 | * entire source file to the entire destination file, the source file | |
1144 | * has a cowextsize hint, and the destination file does not. | |
1145 | */ | |
1146 | cowextsize = 0; | |
1147 | if (pos_in == 0 && len == i_size_read(inode_in) && | |
3e09ab8f | 1148 | (src->i_diflags2 & XFS_DIFLAG2_COWEXTSIZE) && |
3fc9f5e4 | 1149 | pos_out == 0 && len >= i_size_read(inode_out) && |
3e09ab8f | 1150 | !(dest->i_diflags2 & XFS_DIFLAG2_COWEXTSIZE)) |
b33ce57d | 1151 | cowextsize = src->i_cowextsize; |
3fc9f5e4 DW |
1152 | |
1153 | ret = xfs_reflink_update_dest(dest, pos_out + len, cowextsize, | |
1154 | remap_flags); | |
5833112d CH |
1155 | if (ret) |
1156 | goto out_unlock; | |
3fc9f5e4 | 1157 | |
5ffce3cc | 1158 | if (xfs_file_sync_writes(file_in) || xfs_file_sync_writes(file_out)) |
5833112d | 1159 | xfs_log_force_inode(dest); |
3fc9f5e4 | 1160 | out_unlock: |
e2aaee9c | 1161 | xfs_iunlock2_io_mmap(src, dest); |
3fc9f5e4 DW |
1162 | if (ret) |
1163 | trace_xfs_reflink_remap_range_error(dest, ret, _RET_IP_); | |
1164 | return remapped > 0 ? remapped : ret; | |
9fe26045 | 1165 | } |
2fe17c10 | 1166 | |
1da177e4 | 1167 | STATIC int |
3562fd45 | 1168 | xfs_file_open( |
1da177e4 | 1169 | struct inode *inode, |
f999a5bf | 1170 | struct file *file) |
1da177e4 | 1171 | { |
75c8c50f | 1172 | if (xfs_is_shutdown(XFS_M(inode->i_sb))) |
f999a5bf | 1173 | return -EIO; |
d8aeb44a JA |
1174 | file->f_mode |= FMODE_NOWAIT | FMODE_BUF_RASYNC | FMODE_BUF_WASYNC | |
1175 | FMODE_DIO_PARALLEL_WRITE; | |
f3bf67c6 | 1176 | return generic_file_open(inode, file); |
f999a5bf CH |
1177 | } |
1178 | ||
1179 | STATIC int | |
1180 | xfs_dir_open( | |
1181 | struct inode *inode, | |
1182 | struct file *file) | |
1183 | { | |
1184 | struct xfs_inode *ip = XFS_I(inode); | |
a1033753 | 1185 | unsigned int mode; |
f999a5bf CH |
1186 | int error; |
1187 | ||
1188 | error = xfs_file_open(inode, file); | |
1189 | if (error) | |
1190 | return error; | |
1191 | ||
1192 | /* | |
1193 | * If there are any blocks, read-ahead block 0 as we're almost | |
1194 | * certain to have the next operation be a read there. | |
1195 | */ | |
309ecac8 | 1196 | mode = xfs_ilock_data_map_shared(ip); |
daf83964 | 1197 | if (ip->i_df.if_nextents > 0) |
06566fda | 1198 | error = xfs_dir3_data_readahead(ip, 0, 0); |
f999a5bf | 1199 | xfs_iunlock(ip, mode); |
7a652bbe | 1200 | return error; |
1da177e4 LT |
1201 | } |
1202 | ||
1da177e4 | 1203 | STATIC int |
3562fd45 | 1204 | xfs_file_release( |
1da177e4 LT |
1205 | struct inode *inode, |
1206 | struct file *filp) | |
1207 | { | |
2451337d | 1208 | return xfs_release(XFS_I(inode)); |
1da177e4 LT |
1209 | } |
1210 | ||
1da177e4 | 1211 | STATIC int |
3562fd45 | 1212 | xfs_file_readdir( |
b8227554 AV |
1213 | struct file *file, |
1214 | struct dir_context *ctx) | |
1da177e4 | 1215 | { |
b8227554 | 1216 | struct inode *inode = file_inode(file); |
739bfb2a | 1217 | xfs_inode_t *ip = XFS_I(inode); |
051e7cd4 CH |
1218 | size_t bufsize; |
1219 | ||
1220 | /* | |
1221 | * The Linux API doesn't pass down the total size of the buffer | |
1222 | * we read into down to the filesystem. With the filldir concept | |
1223 | * it's not needed for correct information, but the XFS dir2 leaf | |
1224 | * code wants an estimate of the buffer size to calculate it's | |
1225 | * readahead window and size the buffers used for mapping to | |
1226 | * physical blocks. | |
1227 | * | |
1228 | * Try to give it an estimate that's good enough, maybe at some | |
1229 | * point we can change the ->readdir prototype to include the | |
a9cc799e | 1230 | * buffer size. For now we use the current glibc buffer size. |
051e7cd4 | 1231 | */ |
13d2c10b | 1232 | bufsize = (size_t)min_t(loff_t, XFS_READDIR_BUFSIZE, ip->i_disk_size); |
051e7cd4 | 1233 | |
acb9553c | 1234 | return xfs_readdir(NULL, ip, ctx, bufsize); |
3fe3e6b1 JL |
1235 | } |
1236 | ||
1237 | STATIC loff_t | |
1238 | xfs_file_llseek( | |
1239 | struct file *file, | |
1240 | loff_t offset, | |
59f9c004 | 1241 | int whence) |
3fe3e6b1 | 1242 | { |
9b2970aa CH |
1243 | struct inode *inode = file->f_mapping->host; |
1244 | ||
75c8c50f | 1245 | if (xfs_is_shutdown(XFS_I(inode)->i_mount)) |
9b2970aa CH |
1246 | return -EIO; |
1247 | ||
59f9c004 | 1248 | switch (whence) { |
9b2970aa | 1249 | default: |
59f9c004 | 1250 | return generic_file_llseek(file, offset, whence); |
3fe3e6b1 | 1251 | case SEEK_HOLE: |
60271ab7 | 1252 | offset = iomap_seek_hole(inode, offset, &xfs_seek_iomap_ops); |
9b2970aa | 1253 | break; |
49c69591 | 1254 | case SEEK_DATA: |
60271ab7 | 1255 | offset = iomap_seek_data(inode, offset, &xfs_seek_iomap_ops); |
9b2970aa | 1256 | break; |
3fe3e6b1 | 1257 | } |
9b2970aa CH |
1258 | |
1259 | if (offset < 0) | |
1260 | return offset; | |
1261 | return vfs_setpos(file, offset, inode->i_sb->s_maxbytes); | |
3fe3e6b1 JL |
1262 | } |
1263 | ||
ea6c49b7 | 1264 | #ifdef CONFIG_FS_DAX |
47ba8cc7 | 1265 | static inline vm_fault_t |
ea6c49b7 SR |
1266 | xfs_dax_fault( |
1267 | struct vm_fault *vmf, | |
1268 | enum page_entry_size pe_size, | |
1269 | bool write_fault, | |
1270 | pfn_t *pfn) | |
1271 | { | |
1272 | return dax_iomap_fault(vmf, pe_size, pfn, NULL, | |
1273 | (write_fault && !vmf->cow_page) ? | |
1274 | &xfs_dax_write_iomap_ops : | |
1275 | &xfs_read_iomap_ops); | |
1276 | } | |
1277 | #else | |
47ba8cc7 | 1278 | static inline vm_fault_t |
ea6c49b7 SR |
1279 | xfs_dax_fault( |
1280 | struct vm_fault *vmf, | |
1281 | enum page_entry_size pe_size, | |
1282 | bool write_fault, | |
1283 | pfn_t *pfn) | |
1284 | { | |
47ba8cc7 DW |
1285 | ASSERT(0); |
1286 | return VM_FAULT_SIGBUS; | |
ea6c49b7 SR |
1287 | } |
1288 | #endif | |
1289 | ||
de0e8c20 DC |
1290 | /* |
1291 | * Locking for serialisation of IO during page faults. This results in a lock | |
1292 | * ordering of: | |
1293 | * | |
c1e8d7c6 | 1294 | * mmap_lock (MM) |
6b698ede | 1295 | * sb_start_pagefault(vfs, freeze) |
2433480a | 1296 | * invalidate_lock (vfs/XFS_MMAPLOCK - truncate serialisation) |
6b698ede DC |
1297 | * page_lock (MM) |
1298 | * i_lock (XFS - extent map serialisation) | |
de0e8c20 | 1299 | */ |
05edd888 | 1300 | static vm_fault_t |
d522d569 CH |
1301 | __xfs_filemap_fault( |
1302 | struct vm_fault *vmf, | |
1303 | enum page_entry_size pe_size, | |
1304 | bool write_fault) | |
de0e8c20 | 1305 | { |
11bac800 | 1306 | struct inode *inode = file_inode(vmf->vma->vm_file); |
d522d569 | 1307 | struct xfs_inode *ip = XFS_I(inode); |
05edd888 | 1308 | vm_fault_t ret; |
de0e8c20 | 1309 | |
d522d569 | 1310 | trace_xfs_filemap_fault(ip, pe_size, write_fault); |
de0e8c20 | 1311 | |
d522d569 CH |
1312 | if (write_fault) { |
1313 | sb_start_pagefault(inode->i_sb); | |
1314 | file_update_time(vmf->vma->vm_file); | |
1315 | } | |
de0e8c20 | 1316 | |
6b698ede | 1317 | if (IS_DAX(inode)) { |
a39e596b CH |
1318 | pfn_t pfn; |
1319 | ||
2433480a | 1320 | xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED); |
ea6c49b7 | 1321 | ret = xfs_dax_fault(vmf, pe_size, write_fault, &pfn); |
a39e596b CH |
1322 | if (ret & VM_FAULT_NEEDDSYNC) |
1323 | ret = dax_finish_sync_fault(vmf, pe_size, pfn); | |
2433480a | 1324 | xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED); |
6b698ede | 1325 | } else { |
2433480a JK |
1326 | if (write_fault) { |
1327 | xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED); | |
f150b423 | 1328 | ret = iomap_page_mkwrite(vmf, |
118e021b | 1329 | &xfs_page_mkwrite_iomap_ops); |
2433480a JK |
1330 | xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED); |
1331 | } else { | |
d522d569 | 1332 | ret = filemap_fault(vmf); |
2433480a | 1333 | } |
6b698ede | 1334 | } |
6b698ede | 1335 | |
d522d569 CH |
1336 | if (write_fault) |
1337 | sb_end_pagefault(inode->i_sb); | |
6b698ede | 1338 | return ret; |
de0e8c20 DC |
1339 | } |
1340 | ||
b17164e2 MP |
1341 | static inline bool |
1342 | xfs_is_write_fault( | |
1343 | struct vm_fault *vmf) | |
1344 | { | |
1345 | return (vmf->flags & FAULT_FLAG_WRITE) && | |
1346 | (vmf->vma->vm_flags & VM_SHARED); | |
1347 | } | |
1348 | ||
05edd888 | 1349 | static vm_fault_t |
6b698ede | 1350 | xfs_filemap_fault( |
075a924d DC |
1351 | struct vm_fault *vmf) |
1352 | { | |
6b698ede | 1353 | /* DAX can shortcut the normal fault path on write faults! */ |
d522d569 CH |
1354 | return __xfs_filemap_fault(vmf, PE_SIZE_PTE, |
1355 | IS_DAX(file_inode(vmf->vma->vm_file)) && | |
b17164e2 | 1356 | xfs_is_write_fault(vmf)); |
6b698ede DC |
1357 | } |
1358 | ||
05edd888 | 1359 | static vm_fault_t |
a2d58167 | 1360 | xfs_filemap_huge_fault( |
c791ace1 DJ |
1361 | struct vm_fault *vmf, |
1362 | enum page_entry_size pe_size) | |
acd76e74 | 1363 | { |
d522d569 | 1364 | if (!IS_DAX(file_inode(vmf->vma->vm_file))) |
acd76e74 MW |
1365 | return VM_FAULT_FALLBACK; |
1366 | ||
d522d569 CH |
1367 | /* DAX can shortcut the normal fault path on write faults! */ |
1368 | return __xfs_filemap_fault(vmf, pe_size, | |
b17164e2 | 1369 | xfs_is_write_fault(vmf)); |
d522d569 | 1370 | } |
acd76e74 | 1371 | |
05edd888 | 1372 | static vm_fault_t |
d522d569 CH |
1373 | xfs_filemap_page_mkwrite( |
1374 | struct vm_fault *vmf) | |
1375 | { | |
1376 | return __xfs_filemap_fault(vmf, PE_SIZE_PTE, true); | |
acd76e74 MW |
1377 | } |
1378 | ||
3af49285 | 1379 | /* |
7b565c9f JK |
1380 | * pfn_mkwrite was originally intended to ensure we capture time stamp updates |
1381 | * on write faults. In reality, it needs to serialise against truncate and | |
1382 | * prepare memory for writing so handle is as standard write fault. | |
3af49285 | 1383 | */ |
05edd888 | 1384 | static vm_fault_t |
3af49285 | 1385 | xfs_filemap_pfn_mkwrite( |
3af49285 DC |
1386 | struct vm_fault *vmf) |
1387 | { | |
1388 | ||
7b565c9f | 1389 | return __xfs_filemap_fault(vmf, PE_SIZE_PTE, true); |
acd76e74 MW |
1390 | } |
1391 | ||
6b698ede DC |
1392 | static const struct vm_operations_struct xfs_file_vm_ops = { |
1393 | .fault = xfs_filemap_fault, | |
a2d58167 | 1394 | .huge_fault = xfs_filemap_huge_fault, |
945ea457 | 1395 | .map_pages = filemap_map_pages, |
6b698ede | 1396 | .page_mkwrite = xfs_filemap_page_mkwrite, |
3af49285 | 1397 | .pfn_mkwrite = xfs_filemap_pfn_mkwrite, |
6b698ede DC |
1398 | }; |
1399 | ||
1400 | STATIC int | |
1401 | xfs_file_mmap( | |
30fa529e CH |
1402 | struct file *file, |
1403 | struct vm_area_struct *vma) | |
6b698ede | 1404 | { |
30fa529e CH |
1405 | struct inode *inode = file_inode(file); |
1406 | struct xfs_buftarg *target = xfs_inode_buftarg(XFS_I(inode)); | |
b21fec41 | 1407 | |
a39e596b | 1408 | /* |
b21fec41 PG |
1409 | * We don't support synchronous mappings for non-DAX files and |
1410 | * for DAX files if underneath dax_device is not synchronous. | |
a39e596b | 1411 | */ |
30fa529e | 1412 | if (!daxdev_mapping_supported(vma, target->bt_daxdev)) |
a39e596b CH |
1413 | return -EOPNOTSUPP; |
1414 | ||
30fa529e | 1415 | file_accessed(file); |
6b698ede | 1416 | vma->vm_ops = &xfs_file_vm_ops; |
30fa529e | 1417 | if (IS_DAX(inode)) |
1c71222e | 1418 | vm_flags_set(vma, VM_HUGEPAGE); |
6b698ede | 1419 | return 0; |
075a924d DC |
1420 | } |
1421 | ||
4b6f5d20 | 1422 | const struct file_operations xfs_file_operations = { |
3fe3e6b1 | 1423 | .llseek = xfs_file_llseek, |
b4f5d2c6 | 1424 | .read_iter = xfs_file_read_iter, |
bf97f3bc | 1425 | .write_iter = xfs_file_write_iter, |
82c156f8 | 1426 | .splice_read = generic_file_splice_read, |
8d020765 | 1427 | .splice_write = iter_file_splice_write, |
3e08773c | 1428 | .iopoll = iocb_bio_iopoll, |
3562fd45 | 1429 | .unlocked_ioctl = xfs_file_ioctl, |
1da177e4 | 1430 | #ifdef CONFIG_COMPAT |
3562fd45 | 1431 | .compat_ioctl = xfs_file_compat_ioctl, |
1da177e4 | 1432 | #endif |
3562fd45 | 1433 | .mmap = xfs_file_mmap, |
a39e596b | 1434 | .mmap_supported_flags = MAP_SYNC, |
3562fd45 NS |
1435 | .open = xfs_file_open, |
1436 | .release = xfs_file_release, | |
1437 | .fsync = xfs_file_fsync, | |
dbe6ec81 | 1438 | .get_unmapped_area = thp_get_unmapped_area, |
2fe17c10 | 1439 | .fallocate = xfs_file_fallocate, |
40144e49 | 1440 | .fadvise = xfs_file_fadvise, |
2e5dfc99 | 1441 | .remap_file_range = xfs_file_remap_range, |
1da177e4 LT |
1442 | }; |
1443 | ||
4b6f5d20 | 1444 | const struct file_operations xfs_dir_file_operations = { |
f999a5bf | 1445 | .open = xfs_dir_open, |
1da177e4 | 1446 | .read = generic_read_dir, |
3b0a3c1a | 1447 | .iterate_shared = xfs_file_readdir, |
59af1584 | 1448 | .llseek = generic_file_llseek, |
3562fd45 | 1449 | .unlocked_ioctl = xfs_file_ioctl, |
d3870398 | 1450 | #ifdef CONFIG_COMPAT |
3562fd45 | 1451 | .compat_ioctl = xfs_file_compat_ioctl, |
d3870398 | 1452 | #endif |
1da2f2db | 1453 | .fsync = xfs_dir_fsync, |
1da177e4 | 1454 | }; |