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" |
57062787 DC |
13 | #include "xfs_da_format.h" |
14 | #include "xfs_da_btree.h" | |
1da177e4 | 15 | #include "xfs_inode.h" |
239880ef | 16 | #include "xfs_trans.h" |
fd3200be | 17 | #include "xfs_inode_item.h" |
dda35b8f | 18 | #include "xfs_bmap.h" |
c24b5dfa | 19 | #include "xfs_bmap_util.h" |
1da177e4 | 20 | #include "xfs_error.h" |
2b9ab5ab | 21 | #include "xfs_dir2.h" |
c24b5dfa | 22 | #include "xfs_dir2_priv.h" |
ddcd856d | 23 | #include "xfs_ioctl.h" |
dda35b8f | 24 | #include "xfs_trace.h" |
239880ef | 25 | #include "xfs_log.h" |
dc06f398 | 26 | #include "xfs_icache.h" |
781355c6 | 27 | #include "xfs_pnfs.h" |
68a9f5e7 | 28 | #include "xfs_iomap.h" |
0613f16c | 29 | #include "xfs_reflink.h" |
1da177e4 LT |
30 | |
31 | #include <linux/dcache.h> | |
2fe17c10 | 32 | #include <linux/falloc.h> |
d126d43f | 33 | #include <linux/pagevec.h> |
66114cad | 34 | #include <linux/backing-dev.h> |
a39e596b | 35 | #include <linux/mman.h> |
1da177e4 | 36 | |
f0f37e2f | 37 | static const struct vm_operations_struct xfs_file_vm_ops; |
1da177e4 | 38 | |
8add71ca CH |
39 | int |
40 | xfs_update_prealloc_flags( | |
41 | struct xfs_inode *ip, | |
42 | enum xfs_prealloc_flags flags) | |
43 | { | |
44 | struct xfs_trans *tp; | |
45 | int error; | |
46 | ||
253f4911 CH |
47 | error = xfs_trans_alloc(ip->i_mount, &M_RES(ip->i_mount)->tr_writeid, |
48 | 0, 0, 0, &tp); | |
49 | if (error) | |
8add71ca | 50 | return error; |
8add71ca CH |
51 | |
52 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
53 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
54 | ||
55 | if (!(flags & XFS_PREALLOC_INVISIBLE)) { | |
c19b3b05 DC |
56 | VFS_I(ip)->i_mode &= ~S_ISUID; |
57 | if (VFS_I(ip)->i_mode & S_IXGRP) | |
58 | VFS_I(ip)->i_mode &= ~S_ISGID; | |
8add71ca CH |
59 | xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); |
60 | } | |
61 | ||
62 | if (flags & XFS_PREALLOC_SET) | |
63 | ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC; | |
64 | if (flags & XFS_PREALLOC_CLEAR) | |
65 | ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC; | |
66 | ||
67 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
68 | if (flags & XFS_PREALLOC_SYNC) | |
69 | xfs_trans_set_sync(tp); | |
70393313 | 70 | return xfs_trans_commit(tp); |
8add71ca CH |
71 | } |
72 | ||
1da2f2db CH |
73 | /* |
74 | * Fsync operations on directories are much simpler than on regular files, | |
75 | * as there is no file data to flush, and thus also no need for explicit | |
76 | * cache flush operations, and there are no non-transaction metadata updates | |
77 | * on directories either. | |
78 | */ | |
79 | STATIC int | |
80 | xfs_dir_fsync( | |
81 | struct file *file, | |
82 | loff_t start, | |
83 | loff_t end, | |
84 | int datasync) | |
85 | { | |
86 | struct xfs_inode *ip = XFS_I(file->f_mapping->host); | |
87 | struct xfs_mount *mp = ip->i_mount; | |
88 | xfs_lsn_t lsn = 0; | |
89 | ||
90 | trace_xfs_dir_fsync(ip); | |
91 | ||
92 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
93 | if (xfs_ipincount(ip)) | |
94 | lsn = ip->i_itemp->ili_last_lsn; | |
95 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
96 | ||
97 | if (!lsn) | |
98 | return 0; | |
656de4ff | 99 | return xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, NULL); |
1da2f2db CH |
100 | } |
101 | ||
fd3200be CH |
102 | STATIC int |
103 | xfs_file_fsync( | |
104 | struct file *file, | |
02c24a82 JB |
105 | loff_t start, |
106 | loff_t end, | |
fd3200be CH |
107 | int datasync) |
108 | { | |
7ea80859 CH |
109 | struct inode *inode = file->f_mapping->host; |
110 | struct xfs_inode *ip = XFS_I(inode); | |
a27a263b | 111 | struct xfs_mount *mp = ip->i_mount; |
fd3200be CH |
112 | int error = 0; |
113 | int log_flushed = 0; | |
b1037058 | 114 | xfs_lsn_t lsn = 0; |
fd3200be | 115 | |
cca28fb8 | 116 | trace_xfs_file_fsync(ip); |
fd3200be | 117 | |
1b180274 | 118 | error = file_write_and_wait_range(file, start, end); |
02c24a82 JB |
119 | if (error) |
120 | return error; | |
121 | ||
a27a263b | 122 | if (XFS_FORCED_SHUTDOWN(mp)) |
b474c7ae | 123 | return -EIO; |
fd3200be CH |
124 | |
125 | xfs_iflags_clear(ip, XFS_ITRUNCATED); | |
126 | ||
2291dab2 DC |
127 | /* |
128 | * If we have an RT and/or log subvolume we need to make sure to flush | |
129 | * the write cache the device used for file data first. This is to | |
130 | * ensure newly written file data make it to disk before logging the new | |
131 | * inode size in case of an extending write. | |
132 | */ | |
133 | if (XFS_IS_REALTIME_INODE(ip)) | |
134 | xfs_blkdev_issue_flush(mp->m_rtdev_targp); | |
135 | else if (mp->m_logdev_targp != mp->m_ddev_targp) | |
136 | xfs_blkdev_issue_flush(mp->m_ddev_targp); | |
a27a263b | 137 | |
fd3200be | 138 | /* |
fc0561ce DC |
139 | * All metadata updates are logged, which means that we just have to |
140 | * flush the log up to the latest LSN that touched the inode. If we have | |
141 | * concurrent fsync/fdatasync() calls, we need them to all block on the | |
142 | * log force before we clear the ili_fsync_fields field. This ensures | |
143 | * that we don't get a racing sync operation that does not wait for the | |
144 | * metadata to hit the journal before returning. If we race with | |
145 | * clearing the ili_fsync_fields, then all that will happen is the log | |
146 | * force will do nothing as the lsn will already be on disk. We can't | |
147 | * race with setting ili_fsync_fields because that is done under | |
148 | * XFS_ILOCK_EXCL, and that can't happen because we hold the lock shared | |
149 | * until after the ili_fsync_fields is cleared. | |
fd3200be CH |
150 | */ |
151 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
8f639dde CH |
152 | if (xfs_ipincount(ip)) { |
153 | if (!datasync || | |
fc0561ce | 154 | (ip->i_itemp->ili_fsync_fields & ~XFS_ILOG_TIMESTAMP)) |
8f639dde CH |
155 | lsn = ip->i_itemp->ili_last_lsn; |
156 | } | |
fd3200be | 157 | |
fc0561ce | 158 | if (lsn) { |
656de4ff | 159 | error = xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, &log_flushed); |
fc0561ce DC |
160 | ip->i_itemp->ili_fsync_fields = 0; |
161 | } | |
162 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
b1037058 | 163 | |
a27a263b CH |
164 | /* |
165 | * If we only have a single device, and the log force about was | |
166 | * a no-op we might have to flush the data device cache here. | |
167 | * This can only happen for fdatasync/O_DSYNC if we were overwriting | |
168 | * an already allocated file and thus do not have any metadata to | |
169 | * commit. | |
170 | */ | |
2291dab2 DC |
171 | if (!log_flushed && !XFS_IS_REALTIME_INODE(ip) && |
172 | mp->m_logdev_targp == mp->m_ddev_targp) | |
a27a263b | 173 | xfs_blkdev_issue_flush(mp->m_ddev_targp); |
fd3200be | 174 | |
2451337d | 175 | return error; |
fd3200be CH |
176 | } |
177 | ||
00258e36 | 178 | STATIC ssize_t |
bbc5a740 | 179 | xfs_file_dio_aio_read( |
dda35b8f | 180 | struct kiocb *iocb, |
b4f5d2c6 | 181 | struct iov_iter *to) |
dda35b8f | 182 | { |
acdda3aa | 183 | struct xfs_inode *ip = XFS_I(file_inode(iocb->ki_filp)); |
bbc5a740 | 184 | size_t count = iov_iter_count(to); |
acdda3aa | 185 | ssize_t ret; |
dda35b8f | 186 | |
bbc5a740 | 187 | trace_xfs_file_direct_read(ip, count, iocb->ki_pos); |
dda35b8f | 188 | |
f1285ff0 CH |
189 | if (!count) |
190 | return 0; /* skip atime */ | |
dda35b8f | 191 | |
a447d7cd CH |
192 | file_accessed(iocb->ki_filp); |
193 | ||
65523218 | 194 | xfs_ilock(ip, XFS_IOLOCK_SHARED); |
acdda3aa | 195 | ret = iomap_dio_rw(iocb, to, &xfs_iomap_ops, NULL); |
65523218 | 196 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); |
acdda3aa | 197 | |
16d4d435 CH |
198 | return ret; |
199 | } | |
200 | ||
f021bd07 | 201 | static noinline ssize_t |
16d4d435 CH |
202 | xfs_file_dax_read( |
203 | struct kiocb *iocb, | |
204 | struct iov_iter *to) | |
205 | { | |
6c31f495 | 206 | struct xfs_inode *ip = XFS_I(iocb->ki_filp->f_mapping->host); |
16d4d435 CH |
207 | size_t count = iov_iter_count(to); |
208 | ssize_t ret = 0; | |
209 | ||
210 | trace_xfs_file_dax_read(ip, count, iocb->ki_pos); | |
211 | ||
212 | if (!count) | |
213 | return 0; /* skip atime */ | |
214 | ||
942491c9 CH |
215 | if (iocb->ki_flags & IOCB_NOWAIT) { |
216 | if (!xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED)) | |
29a5d29e | 217 | return -EAGAIN; |
942491c9 | 218 | } else { |
29a5d29e GR |
219 | xfs_ilock(ip, XFS_IOLOCK_SHARED); |
220 | } | |
942491c9 | 221 | |
11c59c92 | 222 | ret = dax_iomap_rw(iocb, to, &xfs_iomap_ops); |
65523218 | 223 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); |
bbc5a740 | 224 | |
f1285ff0 | 225 | file_accessed(iocb->ki_filp); |
bbc5a740 CH |
226 | return ret; |
227 | } | |
228 | ||
229 | STATIC ssize_t | |
230 | xfs_file_buffered_aio_read( | |
231 | struct kiocb *iocb, | |
232 | struct iov_iter *to) | |
233 | { | |
234 | struct xfs_inode *ip = XFS_I(file_inode(iocb->ki_filp)); | |
235 | ssize_t ret; | |
236 | ||
237 | trace_xfs_file_buffered_read(ip, iov_iter_count(to), iocb->ki_pos); | |
dda35b8f | 238 | |
942491c9 CH |
239 | if (iocb->ki_flags & IOCB_NOWAIT) { |
240 | if (!xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED)) | |
91f9943e | 241 | return -EAGAIN; |
942491c9 | 242 | } else { |
91f9943e CH |
243 | xfs_ilock(ip, XFS_IOLOCK_SHARED); |
244 | } | |
b4f5d2c6 | 245 | ret = generic_file_read_iter(iocb, to); |
65523218 | 246 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); |
bbc5a740 CH |
247 | |
248 | return ret; | |
249 | } | |
250 | ||
251 | STATIC ssize_t | |
252 | xfs_file_read_iter( | |
253 | struct kiocb *iocb, | |
254 | struct iov_iter *to) | |
255 | { | |
16d4d435 CH |
256 | struct inode *inode = file_inode(iocb->ki_filp); |
257 | struct xfs_mount *mp = XFS_I(inode)->i_mount; | |
bbc5a740 CH |
258 | ssize_t ret = 0; |
259 | ||
260 | XFS_STATS_INC(mp, xs_read_calls); | |
261 | ||
262 | if (XFS_FORCED_SHUTDOWN(mp)) | |
263 | return -EIO; | |
264 | ||
16d4d435 CH |
265 | if (IS_DAX(inode)) |
266 | ret = xfs_file_dax_read(iocb, to); | |
267 | else if (iocb->ki_flags & IOCB_DIRECT) | |
bbc5a740 | 268 | ret = xfs_file_dio_aio_read(iocb, to); |
3176c3e0 | 269 | else |
bbc5a740 | 270 | ret = xfs_file_buffered_aio_read(iocb, to); |
dda35b8f | 271 | |
dda35b8f | 272 | if (ret > 0) |
ff6d6af2 | 273 | XFS_STATS_ADD(mp, xs_read_bytes, ret); |
dda35b8f CH |
274 | return ret; |
275 | } | |
276 | ||
4d8d1581 DC |
277 | /* |
278 | * Common pre-write limit and setup checks. | |
279 | * | |
5bf1f262 CH |
280 | * Called with the iolocked held either shared and exclusive according to |
281 | * @iolock, and returns with it held. Might upgrade the iolock to exclusive | |
282 | * if called for a direct write beyond i_size. | |
4d8d1581 DC |
283 | */ |
284 | STATIC ssize_t | |
285 | xfs_file_aio_write_checks( | |
99733fa3 AV |
286 | struct kiocb *iocb, |
287 | struct iov_iter *from, | |
4d8d1581 DC |
288 | int *iolock) |
289 | { | |
99733fa3 | 290 | struct file *file = iocb->ki_filp; |
4d8d1581 DC |
291 | struct inode *inode = file->f_mapping->host; |
292 | struct xfs_inode *ip = XFS_I(inode); | |
3309dd04 | 293 | ssize_t error = 0; |
99733fa3 | 294 | size_t count = iov_iter_count(from); |
3136e8bb | 295 | bool drained_dio = false; |
f5c54717 | 296 | loff_t isize; |
4d8d1581 | 297 | |
7271d243 | 298 | restart: |
3309dd04 AV |
299 | error = generic_write_checks(iocb, from); |
300 | if (error <= 0) | |
4d8d1581 | 301 | return error; |
4d8d1581 | 302 | |
69eb5fa1 | 303 | error = xfs_break_layouts(inode, iolock, BREAK_WRITE); |
781355c6 CH |
304 | if (error) |
305 | return error; | |
306 | ||
65523218 CH |
307 | /* |
308 | * For changing security info in file_remove_privs() we need i_rwsem | |
309 | * exclusively. | |
310 | */ | |
a6de82ca | 311 | if (*iolock == XFS_IOLOCK_SHARED && !IS_NOSEC(inode)) { |
65523218 | 312 | xfs_iunlock(ip, *iolock); |
a6de82ca | 313 | *iolock = XFS_IOLOCK_EXCL; |
65523218 | 314 | xfs_ilock(ip, *iolock); |
a6de82ca JK |
315 | goto restart; |
316 | } | |
4d8d1581 DC |
317 | /* |
318 | * If the offset is beyond the size of the file, we need to zero any | |
319 | * blocks that fall between the existing EOF and the start of this | |
2813d682 | 320 | * write. If zeroing is needed and we are currently holding the |
467f7899 CH |
321 | * iolock shared, we need to update it to exclusive which implies |
322 | * having to redo all checks before. | |
b9d59846 DC |
323 | * |
324 | * We need to serialise against EOF updates that occur in IO | |
325 | * completions here. We want to make sure that nobody is changing the | |
326 | * size while we do this check until we have placed an IO barrier (i.e. | |
327 | * hold the XFS_IOLOCK_EXCL) that prevents new IO from being dispatched. | |
328 | * The spinlock effectively forms a memory barrier once we have the | |
329 | * XFS_IOLOCK_EXCL so we are guaranteed to see the latest EOF value | |
330 | * and hence be able to correctly determine if we need to run zeroing. | |
4d8d1581 | 331 | */ |
b9d59846 | 332 | spin_lock(&ip->i_flags_lock); |
f5c54717 CH |
333 | isize = i_size_read(inode); |
334 | if (iocb->ki_pos > isize) { | |
b9d59846 | 335 | spin_unlock(&ip->i_flags_lock); |
3136e8bb BF |
336 | if (!drained_dio) { |
337 | if (*iolock == XFS_IOLOCK_SHARED) { | |
65523218 | 338 | xfs_iunlock(ip, *iolock); |
3136e8bb | 339 | *iolock = XFS_IOLOCK_EXCL; |
65523218 | 340 | xfs_ilock(ip, *iolock); |
3136e8bb BF |
341 | iov_iter_reexpand(from, count); |
342 | } | |
40c63fbc DC |
343 | /* |
344 | * We now have an IO submission barrier in place, but | |
345 | * AIO can do EOF updates during IO completion and hence | |
346 | * we now need to wait for all of them to drain. Non-AIO | |
347 | * DIO will have drained before we are given the | |
348 | * XFS_IOLOCK_EXCL, and so for most cases this wait is a | |
349 | * no-op. | |
350 | */ | |
351 | inode_dio_wait(inode); | |
3136e8bb | 352 | drained_dio = true; |
7271d243 DC |
353 | goto restart; |
354 | } | |
f5c54717 CH |
355 | |
356 | trace_xfs_zero_eof(ip, isize, iocb->ki_pos - isize); | |
357 | error = iomap_zero_range(inode, isize, iocb->ki_pos - isize, | |
358 | NULL, &xfs_iomap_ops); | |
467f7899 CH |
359 | if (error) |
360 | return error; | |
b9d59846 DC |
361 | } else |
362 | spin_unlock(&ip->i_flags_lock); | |
4d8d1581 | 363 | |
8a9c9980 CH |
364 | /* |
365 | * Updating the timestamps will grab the ilock again from | |
366 | * xfs_fs_dirty_inode, so we have to call it after dropping the | |
367 | * lock above. Eventually we should look into a way to avoid | |
368 | * the pointless lock roundtrip. | |
369 | */ | |
c3b2da31 JB |
370 | if (likely(!(file->f_mode & FMODE_NOCMTIME))) { |
371 | error = file_update_time(file); | |
372 | if (error) | |
373 | return error; | |
374 | } | |
8a9c9980 | 375 | |
4d8d1581 DC |
376 | /* |
377 | * If we're writing the file then make sure to clear the setuid and | |
378 | * setgid bits if the process is not being run by root. This keeps | |
379 | * people from modifying setuid and setgid binaries. | |
380 | */ | |
a6de82ca JK |
381 | if (!IS_NOSEC(inode)) |
382 | return file_remove_privs(file); | |
383 | return 0; | |
4d8d1581 DC |
384 | } |
385 | ||
acdda3aa CH |
386 | static int |
387 | xfs_dio_write_end_io( | |
388 | struct kiocb *iocb, | |
389 | ssize_t size, | |
390 | unsigned flags) | |
391 | { | |
392 | struct inode *inode = file_inode(iocb->ki_filp); | |
393 | struct xfs_inode *ip = XFS_I(inode); | |
394 | loff_t offset = iocb->ki_pos; | |
acdda3aa CH |
395 | int error = 0; |
396 | ||
397 | trace_xfs_end_io_direct_write(ip, offset, size); | |
398 | ||
399 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) | |
400 | return -EIO; | |
401 | ||
402 | if (size <= 0) | |
403 | return size; | |
404 | ||
ed5c3e66 DC |
405 | /* |
406 | * Capture amount written on completion as we can't reliably account | |
407 | * for it on submission. | |
408 | */ | |
409 | XFS_STATS_ADD(ip->i_mount, xs_write_bytes, size); | |
410 | ||
ee70daab EG |
411 | if (flags & IOMAP_DIO_COW) { |
412 | error = xfs_reflink_end_cow(ip, offset, size); | |
413 | if (error) | |
414 | return error; | |
415 | } | |
416 | ||
417 | /* | |
418 | * Unwritten conversion updates the in-core isize after extent | |
419 | * conversion but before updating the on-disk size. Updating isize any | |
420 | * earlier allows a racing dio read to find unwritten extents before | |
421 | * they are converted. | |
422 | */ | |
423 | if (flags & IOMAP_DIO_UNWRITTEN) | |
424 | return xfs_iomap_write_unwritten(ip, offset, size, true); | |
425 | ||
acdda3aa CH |
426 | /* |
427 | * We need to update the in-core inode size here so that we don't end up | |
428 | * with the on-disk inode size being outside the in-core inode size. We | |
429 | * have no other method of updating EOF for AIO, so always do it here | |
430 | * if necessary. | |
431 | * | |
432 | * We need to lock the test/set EOF update as we can be racing with | |
433 | * other IO completions here to update the EOF. Failing to serialise | |
434 | * here can result in EOF moving backwards and Bad Things Happen when | |
435 | * that occurs. | |
436 | */ | |
437 | spin_lock(&ip->i_flags_lock); | |
438 | if (offset + size > i_size_read(inode)) { | |
439 | i_size_write(inode, offset + size); | |
ee70daab | 440 | spin_unlock(&ip->i_flags_lock); |
acdda3aa | 441 | error = xfs_setfilesize(ip, offset, size); |
ee70daab EG |
442 | } else { |
443 | spin_unlock(&ip->i_flags_lock); | |
444 | } | |
acdda3aa CH |
445 | |
446 | return error; | |
447 | } | |
448 | ||
f0d26e86 DC |
449 | /* |
450 | * xfs_file_dio_aio_write - handle direct IO writes | |
451 | * | |
452 | * Lock the inode appropriately to prepare for and issue a direct IO write. | |
eda77982 | 453 | * By separating it from the buffered write path we remove all the tricky to |
f0d26e86 DC |
454 | * follow locking changes and looping. |
455 | * | |
eda77982 DC |
456 | * If there are cached pages or we're extending the file, we need IOLOCK_EXCL |
457 | * until we're sure the bytes at the new EOF have been zeroed and/or the cached | |
458 | * pages are flushed out. | |
459 | * | |
460 | * In most cases the direct IO writes will be done holding IOLOCK_SHARED | |
461 | * allowing them to be done in parallel with reads and other direct IO writes. | |
462 | * However, if the IO is not aligned to filesystem blocks, the direct IO layer | |
463 | * needs to do sub-block zeroing and that requires serialisation against other | |
464 | * direct IOs to the same block. In this case we need to serialise the | |
465 | * submission of the unaligned IOs so that we don't get racing block zeroing in | |
466 | * the dio layer. To avoid the problem with aio, we also need to wait for | |
467 | * outstanding IOs to complete so that unwritten extent conversion is completed | |
468 | * before we try to map the overlapping block. This is currently implemented by | |
4a06fd26 | 469 | * hitting it with a big hammer (i.e. inode_dio_wait()). |
eda77982 | 470 | * |
f0d26e86 DC |
471 | * Returns with locks held indicated by @iolock and errors indicated by |
472 | * negative return values. | |
473 | */ | |
474 | STATIC ssize_t | |
475 | xfs_file_dio_aio_write( | |
476 | struct kiocb *iocb, | |
b3188919 | 477 | struct iov_iter *from) |
f0d26e86 DC |
478 | { |
479 | struct file *file = iocb->ki_filp; | |
480 | struct address_space *mapping = file->f_mapping; | |
481 | struct inode *inode = mapping->host; | |
482 | struct xfs_inode *ip = XFS_I(inode); | |
483 | struct xfs_mount *mp = ip->i_mount; | |
484 | ssize_t ret = 0; | |
eda77982 | 485 | int unaligned_io = 0; |
d0606464 | 486 | int iolock; |
b3188919 | 487 | size_t count = iov_iter_count(from); |
acdda3aa | 488 | struct xfs_buftarg *target = XFS_IS_REALTIME_INODE(ip) ? |
f0d26e86 DC |
489 | mp->m_rtdev_targp : mp->m_ddev_targp; |
490 | ||
7c71ee78 | 491 | /* DIO must be aligned to device logical sector size */ |
16d4d435 | 492 | if ((iocb->ki_pos | count) & target->bt_logical_sectormask) |
b474c7ae | 493 | return -EINVAL; |
f0d26e86 | 494 | |
7271d243 | 495 | /* |
0ee7a3f6 CH |
496 | * Don't take the exclusive iolock here unless the I/O is unaligned to |
497 | * the file system block size. We don't need to consider the EOF | |
498 | * extension case here because xfs_file_aio_write_checks() will relock | |
499 | * the inode as necessary for EOF zeroing cases and fill out the new | |
500 | * inode size as appropriate. | |
7271d243 | 501 | */ |
0ee7a3f6 CH |
502 | if ((iocb->ki_pos & mp->m_blockmask) || |
503 | ((iocb->ki_pos + count) & mp->m_blockmask)) { | |
504 | unaligned_io = 1; | |
54a4ef8a CH |
505 | |
506 | /* | |
507 | * We can't properly handle unaligned direct I/O to reflink | |
508 | * files yet, as we can't unshare a partial block. | |
509 | */ | |
66ae56a5 | 510 | if (xfs_is_cow_inode(ip)) { |
54a4ef8a CH |
511 | trace_xfs_reflink_bounce_dio_write(ip, iocb->ki_pos, count); |
512 | return -EREMCHG; | |
513 | } | |
d0606464 | 514 | iolock = XFS_IOLOCK_EXCL; |
0ee7a3f6 | 515 | } else { |
d0606464 | 516 | iolock = XFS_IOLOCK_SHARED; |
c58cb165 | 517 | } |
f0d26e86 | 518 | |
942491c9 CH |
519 | if (iocb->ki_flags & IOCB_NOWAIT) { |
520 | if (!xfs_ilock_nowait(ip, iolock)) | |
29a5d29e | 521 | return -EAGAIN; |
942491c9 | 522 | } else { |
29a5d29e GR |
523 | xfs_ilock(ip, iolock); |
524 | } | |
0ee7a3f6 | 525 | |
99733fa3 | 526 | ret = xfs_file_aio_write_checks(iocb, from, &iolock); |
4d8d1581 | 527 | if (ret) |
d0606464 | 528 | goto out; |
99733fa3 | 529 | count = iov_iter_count(from); |
f0d26e86 | 530 | |
eda77982 | 531 | /* |
2032a8a2 BF |
532 | * If we are doing unaligned IO, we can't allow any other overlapping IO |
533 | * in-flight at the same time or we risk data corruption. Wait for all | |
534 | * other IO to drain before we submit. If the IO is aligned, demote the | |
535 | * iolock if we had to take the exclusive lock in | |
536 | * xfs_file_aio_write_checks() for other reasons. | |
eda77982 | 537 | */ |
29a5d29e | 538 | if (unaligned_io) { |
2032a8a2 BF |
539 | /* unaligned dio always waits, bail */ |
540 | if (iocb->ki_flags & IOCB_NOWAIT) | |
541 | return -EAGAIN; | |
542 | inode_dio_wait(inode); | |
29a5d29e | 543 | } else if (iolock == XFS_IOLOCK_EXCL) { |
65523218 | 544 | xfs_ilock_demote(ip, XFS_IOLOCK_EXCL); |
d0606464 | 545 | iolock = XFS_IOLOCK_SHARED; |
f0d26e86 DC |
546 | } |
547 | ||
3176c3e0 | 548 | trace_xfs_file_direct_write(ip, count, iocb->ki_pos); |
acdda3aa | 549 | ret = iomap_dio_rw(iocb, from, &xfs_iomap_ops, xfs_dio_write_end_io); |
2032a8a2 BF |
550 | |
551 | /* | |
552 | * If unaligned, this is the only IO in-flight. If it has not yet | |
553 | * completed, wait on it before we release the iolock to prevent | |
554 | * subsequent overlapping IO. | |
555 | */ | |
556 | if (ret == -EIOCBQUEUED && unaligned_io) | |
557 | inode_dio_wait(inode); | |
d0606464 | 558 | out: |
65523218 | 559 | xfs_iunlock(ip, iolock); |
d0606464 | 560 | |
6b698ede | 561 | /* |
16d4d435 CH |
562 | * No fallback to buffered IO on errors for XFS, direct IO will either |
563 | * complete fully or fail. | |
6b698ede | 564 | */ |
16d4d435 CH |
565 | ASSERT(ret < 0 || ret == count); |
566 | return ret; | |
567 | } | |
568 | ||
f021bd07 | 569 | static noinline ssize_t |
16d4d435 CH |
570 | xfs_file_dax_write( |
571 | struct kiocb *iocb, | |
572 | struct iov_iter *from) | |
573 | { | |
6c31f495 | 574 | struct inode *inode = iocb->ki_filp->f_mapping->host; |
16d4d435 | 575 | struct xfs_inode *ip = XFS_I(inode); |
17879e8f | 576 | int iolock = XFS_IOLOCK_EXCL; |
6c31f495 CH |
577 | ssize_t ret, error = 0; |
578 | size_t count; | |
579 | loff_t pos; | |
16d4d435 | 580 | |
942491c9 CH |
581 | if (iocb->ki_flags & IOCB_NOWAIT) { |
582 | if (!xfs_ilock_nowait(ip, iolock)) | |
29a5d29e | 583 | return -EAGAIN; |
942491c9 | 584 | } else { |
29a5d29e GR |
585 | xfs_ilock(ip, iolock); |
586 | } | |
587 | ||
16d4d435 CH |
588 | ret = xfs_file_aio_write_checks(iocb, from, &iolock); |
589 | if (ret) | |
590 | goto out; | |
591 | ||
6c31f495 CH |
592 | pos = iocb->ki_pos; |
593 | count = iov_iter_count(from); | |
8b2180b3 | 594 | |
6c31f495 | 595 | trace_xfs_file_dax_write(ip, count, pos); |
11c59c92 | 596 | ret = dax_iomap_rw(iocb, from, &xfs_iomap_ops); |
6c31f495 CH |
597 | if (ret > 0 && iocb->ki_pos > i_size_read(inode)) { |
598 | i_size_write(inode, iocb->ki_pos); | |
599 | error = xfs_setfilesize(ip, pos, ret); | |
16d4d435 | 600 | } |
16d4d435 | 601 | out: |
65523218 | 602 | xfs_iunlock(ip, iolock); |
ed5c3e66 DC |
603 | if (error) |
604 | return error; | |
605 | ||
606 | if (ret > 0) { | |
607 | XFS_STATS_ADD(ip->i_mount, xs_write_bytes, ret); | |
608 | ||
609 | /* Handle various SYNC-type writes */ | |
610 | ret = generic_write_sync(iocb, ret); | |
611 | } | |
612 | return ret; | |
f0d26e86 DC |
613 | } |
614 | ||
00258e36 | 615 | STATIC ssize_t |
637bbc75 | 616 | xfs_file_buffered_aio_write( |
dda35b8f | 617 | struct kiocb *iocb, |
b3188919 | 618 | struct iov_iter *from) |
dda35b8f CH |
619 | { |
620 | struct file *file = iocb->ki_filp; | |
621 | struct address_space *mapping = file->f_mapping; | |
622 | struct inode *inode = mapping->host; | |
00258e36 | 623 | struct xfs_inode *ip = XFS_I(inode); |
637bbc75 DC |
624 | ssize_t ret; |
625 | int enospc = 0; | |
c3155097 | 626 | int iolock; |
dda35b8f | 627 | |
91f9943e CH |
628 | if (iocb->ki_flags & IOCB_NOWAIT) |
629 | return -EOPNOTSUPP; | |
630 | ||
c3155097 BF |
631 | write_retry: |
632 | iolock = XFS_IOLOCK_EXCL; | |
65523218 | 633 | xfs_ilock(ip, iolock); |
dda35b8f | 634 | |
99733fa3 | 635 | ret = xfs_file_aio_write_checks(iocb, from, &iolock); |
4d8d1581 | 636 | if (ret) |
d0606464 | 637 | goto out; |
dda35b8f CH |
638 | |
639 | /* We can write back this queue in page reclaim */ | |
de1414a6 | 640 | current->backing_dev_info = inode_to_bdi(inode); |
dda35b8f | 641 | |
3176c3e0 | 642 | trace_xfs_file_buffered_write(ip, iov_iter_count(from), iocb->ki_pos); |
68a9f5e7 | 643 | ret = iomap_file_buffered_write(iocb, from, &xfs_iomap_ops); |
0a64bc2c | 644 | if (likely(ret >= 0)) |
99733fa3 | 645 | iocb->ki_pos += ret; |
dc06f398 | 646 | |
637bbc75 | 647 | /* |
dc06f398 BF |
648 | * If we hit a space limit, try to free up some lingering preallocated |
649 | * space before returning an error. In the case of ENOSPC, first try to | |
650 | * write back all dirty inodes to free up some of the excess reserved | |
651 | * metadata space. This reduces the chances that the eofblocks scan | |
652 | * waits on dirty mappings. Since xfs_flush_inodes() is serialized, this | |
653 | * also behaves as a filter to prevent too many eofblocks scans from | |
654 | * running at the same time. | |
637bbc75 | 655 | */ |
dc06f398 | 656 | if (ret == -EDQUOT && !enospc) { |
c3155097 | 657 | xfs_iunlock(ip, iolock); |
dc06f398 BF |
658 | enospc = xfs_inode_free_quota_eofblocks(ip); |
659 | if (enospc) | |
660 | goto write_retry; | |
83104d44 DW |
661 | enospc = xfs_inode_free_quota_cowblocks(ip); |
662 | if (enospc) | |
663 | goto write_retry; | |
c3155097 | 664 | iolock = 0; |
dc06f398 BF |
665 | } else if (ret == -ENOSPC && !enospc) { |
666 | struct xfs_eofblocks eofb = {0}; | |
667 | ||
637bbc75 | 668 | enospc = 1; |
9aa05000 | 669 | xfs_flush_inodes(ip->i_mount); |
c3155097 BF |
670 | |
671 | xfs_iunlock(ip, iolock); | |
dc06f398 BF |
672 | eofb.eof_flags = XFS_EOF_FLAGS_SYNC; |
673 | xfs_icache_free_eofblocks(ip->i_mount, &eofb); | |
cf2cb784 | 674 | xfs_icache_free_cowblocks(ip->i_mount, &eofb); |
9aa05000 | 675 | goto write_retry; |
dda35b8f | 676 | } |
d0606464 | 677 | |
dda35b8f | 678 | current->backing_dev_info = NULL; |
d0606464 | 679 | out: |
c3155097 BF |
680 | if (iolock) |
681 | xfs_iunlock(ip, iolock); | |
ed5c3e66 DC |
682 | |
683 | if (ret > 0) { | |
684 | XFS_STATS_ADD(ip->i_mount, xs_write_bytes, ret); | |
685 | /* Handle various SYNC-type writes */ | |
686 | ret = generic_write_sync(iocb, ret); | |
687 | } | |
637bbc75 DC |
688 | return ret; |
689 | } | |
690 | ||
691 | STATIC ssize_t | |
bf97f3bc | 692 | xfs_file_write_iter( |
637bbc75 | 693 | struct kiocb *iocb, |
bf97f3bc | 694 | struct iov_iter *from) |
637bbc75 DC |
695 | { |
696 | struct file *file = iocb->ki_filp; | |
697 | struct address_space *mapping = file->f_mapping; | |
698 | struct inode *inode = mapping->host; | |
699 | struct xfs_inode *ip = XFS_I(inode); | |
700 | ssize_t ret; | |
bf97f3bc | 701 | size_t ocount = iov_iter_count(from); |
637bbc75 | 702 | |
ff6d6af2 | 703 | XFS_STATS_INC(ip->i_mount, xs_write_calls); |
637bbc75 | 704 | |
637bbc75 DC |
705 | if (ocount == 0) |
706 | return 0; | |
707 | ||
bf97f3bc AV |
708 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
709 | return -EIO; | |
637bbc75 | 710 | |
16d4d435 | 711 | if (IS_DAX(inode)) |
ed5c3e66 DC |
712 | return xfs_file_dax_write(iocb, from); |
713 | ||
714 | if (iocb->ki_flags & IOCB_DIRECT) { | |
0613f16c DW |
715 | /* |
716 | * Allow a directio write to fall back to a buffered | |
717 | * write *only* in the case that we're doing a reflink | |
718 | * CoW. In all other directio scenarios we do not | |
719 | * allow an operation to fall back to buffered mode. | |
720 | */ | |
bf97f3bc | 721 | ret = xfs_file_dio_aio_write(iocb, from); |
ed5c3e66 DC |
722 | if (ret != -EREMCHG) |
723 | return ret; | |
0613f16c | 724 | } |
dda35b8f | 725 | |
ed5c3e66 | 726 | return xfs_file_buffered_aio_write(iocb, from); |
dda35b8f CH |
727 | } |
728 | ||
d6dc57e2 DW |
729 | static void |
730 | xfs_wait_dax_page( | |
e25ff835 | 731 | struct inode *inode) |
d6dc57e2 DW |
732 | { |
733 | struct xfs_inode *ip = XFS_I(inode); | |
734 | ||
d6dc57e2 DW |
735 | xfs_iunlock(ip, XFS_MMAPLOCK_EXCL); |
736 | schedule(); | |
737 | xfs_ilock(ip, XFS_MMAPLOCK_EXCL); | |
738 | } | |
739 | ||
740 | static int | |
741 | xfs_break_dax_layouts( | |
742 | struct inode *inode, | |
e25ff835 | 743 | bool *retry) |
d6dc57e2 DW |
744 | { |
745 | struct page *page; | |
746 | ||
747 | ASSERT(xfs_isilocked(XFS_I(inode), XFS_MMAPLOCK_EXCL)); | |
748 | ||
749 | page = dax_layout_busy_page(inode->i_mapping); | |
750 | if (!page) | |
751 | return 0; | |
752 | ||
e25ff835 | 753 | *retry = true; |
d6dc57e2 DW |
754 | return ___wait_var_event(&page->_refcount, |
755 | atomic_read(&page->_refcount) == 1, TASK_INTERRUPTIBLE, | |
e25ff835 | 756 | 0, 0, xfs_wait_dax_page(inode)); |
d6dc57e2 DW |
757 | } |
758 | ||
69eb5fa1 DW |
759 | int |
760 | xfs_break_layouts( | |
761 | struct inode *inode, | |
762 | uint *iolock, | |
763 | enum layout_break_reason reason) | |
764 | { | |
765 | bool retry; | |
d6dc57e2 | 766 | int error; |
69eb5fa1 DW |
767 | |
768 | ASSERT(xfs_isilocked(XFS_I(inode), XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)); | |
769 | ||
d6dc57e2 DW |
770 | do { |
771 | retry = false; | |
772 | switch (reason) { | |
773 | case BREAK_UNMAP: | |
a4722a64 | 774 | error = xfs_break_dax_layouts(inode, &retry); |
d6dc57e2 DW |
775 | if (error || retry) |
776 | break; | |
777 | /* fall through */ | |
778 | case BREAK_WRITE: | |
779 | error = xfs_break_leased_layouts(inode, iolock, &retry); | |
780 | break; | |
781 | default: | |
782 | WARN_ON_ONCE(1); | |
783 | error = -EINVAL; | |
784 | } | |
785 | } while (error == 0 && retry); | |
786 | ||
787 | return error; | |
69eb5fa1 DW |
788 | } |
789 | ||
a904b1ca NJ |
790 | #define XFS_FALLOC_FL_SUPPORTED \ |
791 | (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \ | |
792 | FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE | \ | |
98cc2db5 | 793 | FALLOC_FL_INSERT_RANGE | FALLOC_FL_UNSHARE_RANGE) |
a904b1ca | 794 | |
2fe17c10 CH |
795 | STATIC long |
796 | xfs_file_fallocate( | |
83aee9e4 CH |
797 | struct file *file, |
798 | int mode, | |
799 | loff_t offset, | |
800 | loff_t len) | |
2fe17c10 | 801 | { |
83aee9e4 CH |
802 | struct inode *inode = file_inode(file); |
803 | struct xfs_inode *ip = XFS_I(inode); | |
83aee9e4 | 804 | long error; |
8add71ca | 805 | enum xfs_prealloc_flags flags = 0; |
c63a8eae | 806 | uint iolock = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL; |
83aee9e4 | 807 | loff_t new_size = 0; |
749f24f3 | 808 | bool do_file_insert = false; |
2fe17c10 | 809 | |
83aee9e4 CH |
810 | if (!S_ISREG(inode->i_mode)) |
811 | return -EINVAL; | |
a904b1ca | 812 | if (mode & ~XFS_FALLOC_FL_SUPPORTED) |
2fe17c10 CH |
813 | return -EOPNOTSUPP; |
814 | ||
781355c6 | 815 | xfs_ilock(ip, iolock); |
69eb5fa1 | 816 | error = xfs_break_layouts(inode, &iolock, BREAK_UNMAP); |
781355c6 CH |
817 | if (error) |
818 | goto out_unlock; | |
819 | ||
83aee9e4 CH |
820 | if (mode & FALLOC_FL_PUNCH_HOLE) { |
821 | error = xfs_free_file_space(ip, offset, len); | |
822 | if (error) | |
823 | goto out_unlock; | |
e1d8fb88 | 824 | } else if (mode & FALLOC_FL_COLLAPSE_RANGE) { |
93407472 | 825 | unsigned int blksize_mask = i_blocksize(inode) - 1; |
e1d8fb88 NJ |
826 | |
827 | if (offset & blksize_mask || len & blksize_mask) { | |
2451337d | 828 | error = -EINVAL; |
e1d8fb88 NJ |
829 | goto out_unlock; |
830 | } | |
831 | ||
23fffa92 LC |
832 | /* |
833 | * There is no need to overlap collapse range with EOF, | |
834 | * in which case it is effectively a truncate operation | |
835 | */ | |
836 | if (offset + len >= i_size_read(inode)) { | |
2451337d | 837 | error = -EINVAL; |
23fffa92 LC |
838 | goto out_unlock; |
839 | } | |
840 | ||
e1d8fb88 NJ |
841 | new_size = i_size_read(inode) - len; |
842 | ||
843 | error = xfs_collapse_file_space(ip, offset, len); | |
844 | if (error) | |
845 | goto out_unlock; | |
a904b1ca | 846 | } else if (mode & FALLOC_FL_INSERT_RANGE) { |
7d83fb14 DW |
847 | unsigned int blksize_mask = i_blocksize(inode) - 1; |
848 | loff_t isize = i_size_read(inode); | |
a904b1ca | 849 | |
a904b1ca NJ |
850 | if (offset & blksize_mask || len & blksize_mask) { |
851 | error = -EINVAL; | |
852 | goto out_unlock; | |
853 | } | |
854 | ||
7d83fb14 DW |
855 | /* |
856 | * New inode size must not exceed ->s_maxbytes, accounting for | |
857 | * possible signed overflow. | |
858 | */ | |
859 | if (inode->i_sb->s_maxbytes - isize < len) { | |
a904b1ca NJ |
860 | error = -EFBIG; |
861 | goto out_unlock; | |
862 | } | |
7d83fb14 | 863 | new_size = isize + len; |
a904b1ca NJ |
864 | |
865 | /* Offset should be less than i_size */ | |
7d83fb14 | 866 | if (offset >= isize) { |
a904b1ca NJ |
867 | error = -EINVAL; |
868 | goto out_unlock; | |
869 | } | |
749f24f3 | 870 | do_file_insert = true; |
83aee9e4 | 871 | } else { |
8add71ca CH |
872 | flags |= XFS_PREALLOC_SET; |
873 | ||
83aee9e4 CH |
874 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
875 | offset + len > i_size_read(inode)) { | |
876 | new_size = offset + len; | |
2451337d | 877 | error = inode_newsize_ok(inode, new_size); |
83aee9e4 CH |
878 | if (error) |
879 | goto out_unlock; | |
880 | } | |
2fe17c10 | 881 | |
66ae56a5 | 882 | if (mode & FALLOC_FL_ZERO_RANGE) { |
376ba313 | 883 | error = xfs_zero_file_space(ip, offset, len); |
66ae56a5 CH |
884 | } else if (mode & FALLOC_FL_UNSHARE_RANGE) { |
885 | error = xfs_reflink_unshare(ip, offset, len); | |
886 | if (error) | |
887 | goto out_unlock; | |
888 | ||
889 | if (!xfs_is_always_cow_inode(ip)) { | |
890 | error = xfs_alloc_file_space(ip, offset, len, | |
891 | XFS_BMAPI_PREALLOC); | |
98cc2db5 | 892 | } |
66ae56a5 CH |
893 | } else { |
894 | /* | |
895 | * If always_cow mode we can't use preallocations and | |
896 | * thus should not create them. | |
897 | */ | |
898 | if (xfs_is_always_cow_inode(ip)) { | |
899 | error = -EOPNOTSUPP; | |
900 | goto out_unlock; | |
901 | } | |
902 | ||
376ba313 LC |
903 | error = xfs_alloc_file_space(ip, offset, len, |
904 | XFS_BMAPI_PREALLOC); | |
98cc2db5 | 905 | } |
2fe17c10 CH |
906 | if (error) |
907 | goto out_unlock; | |
908 | } | |
909 | ||
83aee9e4 | 910 | if (file->f_flags & O_DSYNC) |
8add71ca CH |
911 | flags |= XFS_PREALLOC_SYNC; |
912 | ||
913 | error = xfs_update_prealloc_flags(ip, flags); | |
2fe17c10 CH |
914 | if (error) |
915 | goto out_unlock; | |
916 | ||
917 | /* Change file size if needed */ | |
918 | if (new_size) { | |
919 | struct iattr iattr; | |
920 | ||
921 | iattr.ia_valid = ATTR_SIZE; | |
922 | iattr.ia_size = new_size; | |
69bca807 | 923 | error = xfs_vn_setattr_size(file_dentry(file), &iattr); |
a904b1ca NJ |
924 | if (error) |
925 | goto out_unlock; | |
2fe17c10 CH |
926 | } |
927 | ||
a904b1ca NJ |
928 | /* |
929 | * Perform hole insertion now that the file size has been | |
930 | * updated so that if we crash during the operation we don't | |
931 | * leave shifted extents past EOF and hence losing access to | |
932 | * the data that is contained within them. | |
933 | */ | |
934 | if (do_file_insert) | |
935 | error = xfs_insert_file_space(ip, offset, len); | |
936 | ||
2fe17c10 | 937 | out_unlock: |
781355c6 | 938 | xfs_iunlock(ip, iolock); |
2451337d | 939 | return error; |
2fe17c10 CH |
940 | } |
941 | ||
3fc9f5e4 | 942 | |
da034bcc | 943 | STATIC loff_t |
2e5dfc99 | 944 | xfs_file_remap_range( |
3fc9f5e4 DW |
945 | struct file *file_in, |
946 | loff_t pos_in, | |
947 | struct file *file_out, | |
948 | loff_t pos_out, | |
949 | loff_t len, | |
950 | unsigned int remap_flags) | |
9fe26045 | 951 | { |
3fc9f5e4 DW |
952 | struct inode *inode_in = file_inode(file_in); |
953 | struct xfs_inode *src = XFS_I(inode_in); | |
954 | struct inode *inode_out = file_inode(file_out); | |
955 | struct xfs_inode *dest = XFS_I(inode_out); | |
956 | struct xfs_mount *mp = src->i_mount; | |
957 | loff_t remapped = 0; | |
958 | xfs_extlen_t cowextsize; | |
959 | int ret; | |
960 | ||
2e5dfc99 DW |
961 | if (remap_flags & ~(REMAP_FILE_DEDUP | REMAP_FILE_ADVISORY)) |
962 | return -EINVAL; | |
cc714660 | 963 | |
3fc9f5e4 DW |
964 | if (!xfs_sb_version_hasreflink(&mp->m_sb)) |
965 | return -EOPNOTSUPP; | |
966 | ||
967 | if (XFS_FORCED_SHUTDOWN(mp)) | |
968 | return -EIO; | |
969 | ||
970 | /* Prepare and then clone file data. */ | |
971 | ret = xfs_reflink_remap_prep(file_in, pos_in, file_out, pos_out, | |
972 | &len, remap_flags); | |
973 | if (ret < 0 || len == 0) | |
974 | return ret; | |
975 | ||
976 | trace_xfs_reflink_remap_range(src, pos_in, len, dest, pos_out); | |
977 | ||
978 | ret = xfs_reflink_remap_blocks(src, pos_in, dest, pos_out, len, | |
979 | &remapped); | |
980 | if (ret) | |
981 | goto out_unlock; | |
982 | ||
983 | /* | |
984 | * Carry the cowextsize hint from src to dest if we're sharing the | |
985 | * entire source file to the entire destination file, the source file | |
986 | * has a cowextsize hint, and the destination file does not. | |
987 | */ | |
988 | cowextsize = 0; | |
989 | if (pos_in == 0 && len == i_size_read(inode_in) && | |
990 | (src->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE) && | |
991 | pos_out == 0 && len >= i_size_read(inode_out) && | |
992 | !(dest->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE)) | |
993 | cowextsize = src->i_d.di_cowextsize; | |
994 | ||
995 | ret = xfs_reflink_update_dest(dest, pos_out + len, cowextsize, | |
996 | remap_flags); | |
997 | ||
998 | out_unlock: | |
999 | xfs_reflink_remap_unlock(file_in, file_out); | |
1000 | if (ret) | |
1001 | trace_xfs_reflink_remap_range_error(dest, ret, _RET_IP_); | |
1002 | return remapped > 0 ? remapped : ret; | |
9fe26045 | 1003 | } |
2fe17c10 | 1004 | |
1da177e4 | 1005 | STATIC int |
3562fd45 | 1006 | xfs_file_open( |
1da177e4 | 1007 | struct inode *inode, |
f999a5bf | 1008 | struct file *file) |
1da177e4 | 1009 | { |
f999a5bf | 1010 | if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS) |
1da177e4 | 1011 | return -EFBIG; |
f999a5bf CH |
1012 | if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb))) |
1013 | return -EIO; | |
91f9943e | 1014 | file->f_mode |= FMODE_NOWAIT; |
f999a5bf CH |
1015 | return 0; |
1016 | } | |
1017 | ||
1018 | STATIC int | |
1019 | xfs_dir_open( | |
1020 | struct inode *inode, | |
1021 | struct file *file) | |
1022 | { | |
1023 | struct xfs_inode *ip = XFS_I(inode); | |
1024 | int mode; | |
1025 | int error; | |
1026 | ||
1027 | error = xfs_file_open(inode, file); | |
1028 | if (error) | |
1029 | return error; | |
1030 | ||
1031 | /* | |
1032 | * If there are any blocks, read-ahead block 0 as we're almost | |
1033 | * certain to have the next operation be a read there. | |
1034 | */ | |
309ecac8 | 1035 | mode = xfs_ilock_data_map_shared(ip); |
f999a5bf | 1036 | if (ip->i_d.di_nextents > 0) |
7a652bbe | 1037 | error = xfs_dir3_data_readahead(ip, 0, -1); |
f999a5bf | 1038 | xfs_iunlock(ip, mode); |
7a652bbe | 1039 | return error; |
1da177e4 LT |
1040 | } |
1041 | ||
1da177e4 | 1042 | STATIC int |
3562fd45 | 1043 | xfs_file_release( |
1da177e4 LT |
1044 | struct inode *inode, |
1045 | struct file *filp) | |
1046 | { | |
2451337d | 1047 | return xfs_release(XFS_I(inode)); |
1da177e4 LT |
1048 | } |
1049 | ||
1da177e4 | 1050 | STATIC int |
3562fd45 | 1051 | xfs_file_readdir( |
b8227554 AV |
1052 | struct file *file, |
1053 | struct dir_context *ctx) | |
1da177e4 | 1054 | { |
b8227554 | 1055 | struct inode *inode = file_inode(file); |
739bfb2a | 1056 | xfs_inode_t *ip = XFS_I(inode); |
051e7cd4 CH |
1057 | size_t bufsize; |
1058 | ||
1059 | /* | |
1060 | * The Linux API doesn't pass down the total size of the buffer | |
1061 | * we read into down to the filesystem. With the filldir concept | |
1062 | * it's not needed for correct information, but the XFS dir2 leaf | |
1063 | * code wants an estimate of the buffer size to calculate it's | |
1064 | * readahead window and size the buffers used for mapping to | |
1065 | * physical blocks. | |
1066 | * | |
1067 | * Try to give it an estimate that's good enough, maybe at some | |
1068 | * point we can change the ->readdir prototype to include the | |
a9cc799e | 1069 | * buffer size. For now we use the current glibc buffer size. |
051e7cd4 | 1070 | */ |
a5c46e5e | 1071 | bufsize = (size_t)min_t(loff_t, XFS_READDIR_BUFSIZE, ip->i_d.di_size); |
051e7cd4 | 1072 | |
acb9553c | 1073 | return xfs_readdir(NULL, ip, ctx, bufsize); |
3fe3e6b1 JL |
1074 | } |
1075 | ||
1076 | STATIC loff_t | |
1077 | xfs_file_llseek( | |
1078 | struct file *file, | |
1079 | loff_t offset, | |
59f9c004 | 1080 | int whence) |
3fe3e6b1 | 1081 | { |
9b2970aa CH |
1082 | struct inode *inode = file->f_mapping->host; |
1083 | ||
1084 | if (XFS_FORCED_SHUTDOWN(XFS_I(inode)->i_mount)) | |
1085 | return -EIO; | |
1086 | ||
59f9c004 | 1087 | switch (whence) { |
9b2970aa | 1088 | default: |
59f9c004 | 1089 | return generic_file_llseek(file, offset, whence); |
3fe3e6b1 | 1090 | case SEEK_HOLE: |
60271ab7 | 1091 | offset = iomap_seek_hole(inode, offset, &xfs_seek_iomap_ops); |
9b2970aa | 1092 | break; |
49c69591 | 1093 | case SEEK_DATA: |
60271ab7 | 1094 | offset = iomap_seek_data(inode, offset, &xfs_seek_iomap_ops); |
9b2970aa | 1095 | break; |
3fe3e6b1 | 1096 | } |
9b2970aa CH |
1097 | |
1098 | if (offset < 0) | |
1099 | return offset; | |
1100 | return vfs_setpos(file, offset, inode->i_sb->s_maxbytes); | |
3fe3e6b1 JL |
1101 | } |
1102 | ||
de0e8c20 DC |
1103 | /* |
1104 | * Locking for serialisation of IO during page faults. This results in a lock | |
1105 | * ordering of: | |
1106 | * | |
1107 | * mmap_sem (MM) | |
6b698ede | 1108 | * sb_start_pagefault(vfs, freeze) |
13ad4fe3 | 1109 | * i_mmaplock (XFS - truncate serialisation) |
6b698ede DC |
1110 | * page_lock (MM) |
1111 | * i_lock (XFS - extent map serialisation) | |
de0e8c20 | 1112 | */ |
05edd888 | 1113 | static vm_fault_t |
d522d569 CH |
1114 | __xfs_filemap_fault( |
1115 | struct vm_fault *vmf, | |
1116 | enum page_entry_size pe_size, | |
1117 | bool write_fault) | |
de0e8c20 | 1118 | { |
11bac800 | 1119 | struct inode *inode = file_inode(vmf->vma->vm_file); |
d522d569 | 1120 | struct xfs_inode *ip = XFS_I(inode); |
05edd888 | 1121 | vm_fault_t ret; |
de0e8c20 | 1122 | |
d522d569 | 1123 | trace_xfs_filemap_fault(ip, pe_size, write_fault); |
de0e8c20 | 1124 | |
d522d569 CH |
1125 | if (write_fault) { |
1126 | sb_start_pagefault(inode->i_sb); | |
1127 | file_update_time(vmf->vma->vm_file); | |
1128 | } | |
de0e8c20 | 1129 | |
d522d569 | 1130 | xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED); |
6b698ede | 1131 | if (IS_DAX(inode)) { |
a39e596b CH |
1132 | pfn_t pfn; |
1133 | ||
c0b24625 | 1134 | ret = dax_iomap_fault(vmf, pe_size, &pfn, NULL, &xfs_iomap_ops); |
a39e596b CH |
1135 | if (ret & VM_FAULT_NEEDDSYNC) |
1136 | ret = dax_finish_sync_fault(vmf, pe_size, pfn); | |
6b698ede | 1137 | } else { |
d522d569 CH |
1138 | if (write_fault) |
1139 | ret = iomap_page_mkwrite(vmf, &xfs_iomap_ops); | |
1140 | else | |
1141 | ret = filemap_fault(vmf); | |
6b698ede | 1142 | } |
6b698ede | 1143 | xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED); |
6b698ede | 1144 | |
d522d569 CH |
1145 | if (write_fault) |
1146 | sb_end_pagefault(inode->i_sb); | |
6b698ede | 1147 | return ret; |
de0e8c20 DC |
1148 | } |
1149 | ||
05edd888 | 1150 | static vm_fault_t |
6b698ede | 1151 | xfs_filemap_fault( |
075a924d DC |
1152 | struct vm_fault *vmf) |
1153 | { | |
6b698ede | 1154 | /* DAX can shortcut the normal fault path on write faults! */ |
d522d569 CH |
1155 | return __xfs_filemap_fault(vmf, PE_SIZE_PTE, |
1156 | IS_DAX(file_inode(vmf->vma->vm_file)) && | |
1157 | (vmf->flags & FAULT_FLAG_WRITE)); | |
6b698ede DC |
1158 | } |
1159 | ||
05edd888 | 1160 | static vm_fault_t |
a2d58167 | 1161 | xfs_filemap_huge_fault( |
c791ace1 DJ |
1162 | struct vm_fault *vmf, |
1163 | enum page_entry_size pe_size) | |
acd76e74 | 1164 | { |
d522d569 | 1165 | if (!IS_DAX(file_inode(vmf->vma->vm_file))) |
acd76e74 MW |
1166 | return VM_FAULT_FALLBACK; |
1167 | ||
d522d569 CH |
1168 | /* DAX can shortcut the normal fault path on write faults! */ |
1169 | return __xfs_filemap_fault(vmf, pe_size, | |
1170 | (vmf->flags & FAULT_FLAG_WRITE)); | |
1171 | } | |
acd76e74 | 1172 | |
05edd888 | 1173 | static vm_fault_t |
d522d569 CH |
1174 | xfs_filemap_page_mkwrite( |
1175 | struct vm_fault *vmf) | |
1176 | { | |
1177 | return __xfs_filemap_fault(vmf, PE_SIZE_PTE, true); | |
acd76e74 MW |
1178 | } |
1179 | ||
3af49285 | 1180 | /* |
7b565c9f JK |
1181 | * pfn_mkwrite was originally intended to ensure we capture time stamp updates |
1182 | * on write faults. In reality, it needs to serialise against truncate and | |
1183 | * prepare memory for writing so handle is as standard write fault. | |
3af49285 | 1184 | */ |
05edd888 | 1185 | static vm_fault_t |
3af49285 | 1186 | xfs_filemap_pfn_mkwrite( |
3af49285 DC |
1187 | struct vm_fault *vmf) |
1188 | { | |
1189 | ||
7b565c9f | 1190 | return __xfs_filemap_fault(vmf, PE_SIZE_PTE, true); |
acd76e74 MW |
1191 | } |
1192 | ||
6b698ede DC |
1193 | static const struct vm_operations_struct xfs_file_vm_ops = { |
1194 | .fault = xfs_filemap_fault, | |
a2d58167 | 1195 | .huge_fault = xfs_filemap_huge_fault, |
6b698ede DC |
1196 | .map_pages = filemap_map_pages, |
1197 | .page_mkwrite = xfs_filemap_page_mkwrite, | |
3af49285 | 1198 | .pfn_mkwrite = xfs_filemap_pfn_mkwrite, |
6b698ede DC |
1199 | }; |
1200 | ||
1201 | STATIC int | |
1202 | xfs_file_mmap( | |
1203 | struct file *filp, | |
1204 | struct vm_area_struct *vma) | |
1205 | { | |
a39e596b CH |
1206 | /* |
1207 | * We don't support synchronous mappings for non-DAX files. At least | |
1208 | * until someone comes with a sensible use case. | |
1209 | */ | |
1210 | if (!IS_DAX(file_inode(filp)) && (vma->vm_flags & VM_SYNC)) | |
1211 | return -EOPNOTSUPP; | |
1212 | ||
6b698ede DC |
1213 | file_accessed(filp); |
1214 | vma->vm_ops = &xfs_file_vm_ops; | |
1215 | if (IS_DAX(file_inode(filp))) | |
e1fb4a08 | 1216 | vma->vm_flags |= VM_HUGEPAGE; |
6b698ede | 1217 | return 0; |
075a924d DC |
1218 | } |
1219 | ||
4b6f5d20 | 1220 | const struct file_operations xfs_file_operations = { |
3fe3e6b1 | 1221 | .llseek = xfs_file_llseek, |
b4f5d2c6 | 1222 | .read_iter = xfs_file_read_iter, |
bf97f3bc | 1223 | .write_iter = xfs_file_write_iter, |
82c156f8 | 1224 | .splice_read = generic_file_splice_read, |
8d020765 | 1225 | .splice_write = iter_file_splice_write, |
81214bab | 1226 | .iopoll = iomap_dio_iopoll, |
3562fd45 | 1227 | .unlocked_ioctl = xfs_file_ioctl, |
1da177e4 | 1228 | #ifdef CONFIG_COMPAT |
3562fd45 | 1229 | .compat_ioctl = xfs_file_compat_ioctl, |
1da177e4 | 1230 | #endif |
3562fd45 | 1231 | .mmap = xfs_file_mmap, |
a39e596b | 1232 | .mmap_supported_flags = MAP_SYNC, |
3562fd45 NS |
1233 | .open = xfs_file_open, |
1234 | .release = xfs_file_release, | |
1235 | .fsync = xfs_file_fsync, | |
dbe6ec81 | 1236 | .get_unmapped_area = thp_get_unmapped_area, |
2fe17c10 | 1237 | .fallocate = xfs_file_fallocate, |
2e5dfc99 | 1238 | .remap_file_range = xfs_file_remap_range, |
1da177e4 LT |
1239 | }; |
1240 | ||
4b6f5d20 | 1241 | const struct file_operations xfs_dir_file_operations = { |
f999a5bf | 1242 | .open = xfs_dir_open, |
1da177e4 | 1243 | .read = generic_read_dir, |
3b0a3c1a | 1244 | .iterate_shared = xfs_file_readdir, |
59af1584 | 1245 | .llseek = generic_file_llseek, |
3562fd45 | 1246 | .unlocked_ioctl = xfs_file_ioctl, |
d3870398 | 1247 | #ifdef CONFIG_COMPAT |
3562fd45 | 1248 | .compat_ioctl = xfs_file_compat_ioctl, |
d3870398 | 1249 | #endif |
1da2f2db | 1250 | .fsync = xfs_dir_fsync, |
1da177e4 | 1251 | }; |