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 | |
2fe17c10 | 28 | #include <linux/falloc.h> |
66114cad | 29 | #include <linux/backing-dev.h> |
a39e596b | 30 | #include <linux/mman.h> |
40144e49 | 31 | #include <linux/fadvise.h> |
1da177e4 | 32 | |
f0f37e2f | 33 | static const struct vm_operations_struct xfs_file_vm_ops; |
1da177e4 | 34 | |
8add71ca CH |
35 | int |
36 | xfs_update_prealloc_flags( | |
37 | struct xfs_inode *ip, | |
38 | enum xfs_prealloc_flags flags) | |
39 | { | |
40 | struct xfs_trans *tp; | |
41 | int error; | |
42 | ||
253f4911 CH |
43 | error = xfs_trans_alloc(ip->i_mount, &M_RES(ip->i_mount)->tr_writeid, |
44 | 0, 0, 0, &tp); | |
45 | if (error) | |
8add71ca | 46 | return error; |
8add71ca CH |
47 | |
48 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
49 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
50 | ||
51 | if (!(flags & XFS_PREALLOC_INVISIBLE)) { | |
c19b3b05 DC |
52 | VFS_I(ip)->i_mode &= ~S_ISUID; |
53 | if (VFS_I(ip)->i_mode & S_IXGRP) | |
54 | VFS_I(ip)->i_mode &= ~S_ISGID; | |
8add71ca CH |
55 | xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); |
56 | } | |
57 | ||
58 | if (flags & XFS_PREALLOC_SET) | |
59 | ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC; | |
60 | if (flags & XFS_PREALLOC_CLEAR) | |
61 | ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC; | |
62 | ||
63 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
64 | if (flags & XFS_PREALLOC_SYNC) | |
65 | xfs_trans_set_sync(tp); | |
70393313 | 66 | return xfs_trans_commit(tp); |
8add71ca CH |
67 | } |
68 | ||
1da2f2db CH |
69 | /* |
70 | * Fsync operations on directories are much simpler than on regular files, | |
71 | * as there is no file data to flush, and thus also no need for explicit | |
72 | * cache flush operations, and there are no non-transaction metadata updates | |
73 | * on directories either. | |
74 | */ | |
75 | STATIC int | |
76 | xfs_dir_fsync( | |
77 | struct file *file, | |
78 | loff_t start, | |
79 | loff_t end, | |
80 | int datasync) | |
81 | { | |
82 | struct xfs_inode *ip = XFS_I(file->f_mapping->host); | |
83 | struct xfs_mount *mp = ip->i_mount; | |
84 | xfs_lsn_t lsn = 0; | |
85 | ||
86 | trace_xfs_dir_fsync(ip); | |
87 | ||
88 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
89 | if (xfs_ipincount(ip)) | |
90 | lsn = ip->i_itemp->ili_last_lsn; | |
91 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
92 | ||
93 | if (!lsn) | |
94 | return 0; | |
656de4ff | 95 | return xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, NULL); |
1da2f2db CH |
96 | } |
97 | ||
fd3200be CH |
98 | STATIC int |
99 | xfs_file_fsync( | |
100 | struct file *file, | |
02c24a82 JB |
101 | loff_t start, |
102 | loff_t end, | |
fd3200be CH |
103 | int datasync) |
104 | { | |
7ea80859 CH |
105 | struct inode *inode = file->f_mapping->host; |
106 | struct xfs_inode *ip = XFS_I(inode); | |
a27a263b | 107 | struct xfs_mount *mp = ip->i_mount; |
fd3200be CH |
108 | int error = 0; |
109 | int log_flushed = 0; | |
b1037058 | 110 | xfs_lsn_t lsn = 0; |
fd3200be | 111 | |
cca28fb8 | 112 | trace_xfs_file_fsync(ip); |
fd3200be | 113 | |
1b180274 | 114 | error = file_write_and_wait_range(file, start, end); |
02c24a82 JB |
115 | if (error) |
116 | return error; | |
117 | ||
a27a263b | 118 | if (XFS_FORCED_SHUTDOWN(mp)) |
b474c7ae | 119 | return -EIO; |
fd3200be CH |
120 | |
121 | xfs_iflags_clear(ip, XFS_ITRUNCATED); | |
122 | ||
2291dab2 DC |
123 | /* |
124 | * If we have an RT and/or log subvolume we need to make sure to flush | |
125 | * the write cache the device used for file data first. This is to | |
126 | * ensure newly written file data make it to disk before logging the new | |
127 | * inode size in case of an extending write. | |
128 | */ | |
129 | if (XFS_IS_REALTIME_INODE(ip)) | |
130 | xfs_blkdev_issue_flush(mp->m_rtdev_targp); | |
131 | else if (mp->m_logdev_targp != mp->m_ddev_targp) | |
132 | xfs_blkdev_issue_flush(mp->m_ddev_targp); | |
a27a263b | 133 | |
fd3200be | 134 | /* |
fc0561ce DC |
135 | * All metadata updates are logged, which means that we just have to |
136 | * flush the log up to the latest LSN that touched the inode. If we have | |
137 | * concurrent fsync/fdatasync() calls, we need them to all block on the | |
138 | * log force before we clear the ili_fsync_fields field. This ensures | |
139 | * that we don't get a racing sync operation that does not wait for the | |
140 | * metadata to hit the journal before returning. If we race with | |
141 | * clearing the ili_fsync_fields, then all that will happen is the log | |
142 | * force will do nothing as the lsn will already be on disk. We can't | |
143 | * race with setting ili_fsync_fields because that is done under | |
144 | * XFS_ILOCK_EXCL, and that can't happen because we hold the lock shared | |
145 | * until after the ili_fsync_fields is cleared. | |
fd3200be CH |
146 | */ |
147 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
8f639dde CH |
148 | if (xfs_ipincount(ip)) { |
149 | if (!datasync || | |
fc0561ce | 150 | (ip->i_itemp->ili_fsync_fields & ~XFS_ILOG_TIMESTAMP)) |
8f639dde CH |
151 | lsn = ip->i_itemp->ili_last_lsn; |
152 | } | |
fd3200be | 153 | |
fc0561ce | 154 | if (lsn) { |
656de4ff | 155 | error = xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, &log_flushed); |
fc0561ce DC |
156 | ip->i_itemp->ili_fsync_fields = 0; |
157 | } | |
158 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
b1037058 | 159 | |
a27a263b CH |
160 | /* |
161 | * If we only have a single device, and the log force about was | |
162 | * a no-op we might have to flush the data device cache here. | |
163 | * This can only happen for fdatasync/O_DSYNC if we were overwriting | |
164 | * an already allocated file and thus do not have any metadata to | |
165 | * commit. | |
166 | */ | |
2291dab2 DC |
167 | if (!log_flushed && !XFS_IS_REALTIME_INODE(ip) && |
168 | mp->m_logdev_targp == mp->m_ddev_targp) | |
a27a263b | 169 | xfs_blkdev_issue_flush(mp->m_ddev_targp); |
fd3200be | 170 | |
2451337d | 171 | return error; |
fd3200be CH |
172 | } |
173 | ||
00258e36 | 174 | STATIC ssize_t |
bbc5a740 | 175 | xfs_file_dio_aio_read( |
dda35b8f | 176 | struct kiocb *iocb, |
b4f5d2c6 | 177 | struct iov_iter *to) |
dda35b8f | 178 | { |
acdda3aa | 179 | struct xfs_inode *ip = XFS_I(file_inode(iocb->ki_filp)); |
bbc5a740 | 180 | size_t count = iov_iter_count(to); |
acdda3aa | 181 | ssize_t ret; |
dda35b8f | 182 | |
bbc5a740 | 183 | trace_xfs_file_direct_read(ip, count, iocb->ki_pos); |
dda35b8f | 184 | |
f1285ff0 CH |
185 | if (!count) |
186 | return 0; /* skip atime */ | |
dda35b8f | 187 | |
a447d7cd CH |
188 | file_accessed(iocb->ki_filp); |
189 | ||
65523218 | 190 | xfs_ilock(ip, XFS_IOLOCK_SHARED); |
acdda3aa | 191 | ret = iomap_dio_rw(iocb, to, &xfs_iomap_ops, NULL); |
65523218 | 192 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); |
acdda3aa | 193 | |
16d4d435 CH |
194 | return ret; |
195 | } | |
196 | ||
f021bd07 | 197 | static noinline ssize_t |
16d4d435 CH |
198 | xfs_file_dax_read( |
199 | struct kiocb *iocb, | |
200 | struct iov_iter *to) | |
201 | { | |
6c31f495 | 202 | struct xfs_inode *ip = XFS_I(iocb->ki_filp->f_mapping->host); |
16d4d435 CH |
203 | size_t count = iov_iter_count(to); |
204 | ssize_t ret = 0; | |
205 | ||
206 | trace_xfs_file_dax_read(ip, count, iocb->ki_pos); | |
207 | ||
208 | if (!count) | |
209 | return 0; /* skip atime */ | |
210 | ||
942491c9 CH |
211 | if (iocb->ki_flags & IOCB_NOWAIT) { |
212 | if (!xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED)) | |
29a5d29e | 213 | return -EAGAIN; |
942491c9 | 214 | } else { |
29a5d29e GR |
215 | xfs_ilock(ip, XFS_IOLOCK_SHARED); |
216 | } | |
942491c9 | 217 | |
11c59c92 | 218 | ret = dax_iomap_rw(iocb, to, &xfs_iomap_ops); |
65523218 | 219 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); |
bbc5a740 | 220 | |
f1285ff0 | 221 | file_accessed(iocb->ki_filp); |
bbc5a740 CH |
222 | return ret; |
223 | } | |
224 | ||
225 | STATIC ssize_t | |
226 | xfs_file_buffered_aio_read( | |
227 | struct kiocb *iocb, | |
228 | struct iov_iter *to) | |
229 | { | |
230 | struct xfs_inode *ip = XFS_I(file_inode(iocb->ki_filp)); | |
231 | ssize_t ret; | |
232 | ||
233 | trace_xfs_file_buffered_read(ip, iov_iter_count(to), iocb->ki_pos); | |
dda35b8f | 234 | |
942491c9 CH |
235 | if (iocb->ki_flags & IOCB_NOWAIT) { |
236 | if (!xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED)) | |
91f9943e | 237 | return -EAGAIN; |
942491c9 | 238 | } else { |
91f9943e CH |
239 | xfs_ilock(ip, XFS_IOLOCK_SHARED); |
240 | } | |
b4f5d2c6 | 241 | ret = generic_file_read_iter(iocb, to); |
65523218 | 242 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); |
bbc5a740 CH |
243 | |
244 | return ret; | |
245 | } | |
246 | ||
247 | STATIC ssize_t | |
248 | xfs_file_read_iter( | |
249 | struct kiocb *iocb, | |
250 | struct iov_iter *to) | |
251 | { | |
16d4d435 CH |
252 | struct inode *inode = file_inode(iocb->ki_filp); |
253 | struct xfs_mount *mp = XFS_I(inode)->i_mount; | |
bbc5a740 CH |
254 | ssize_t ret = 0; |
255 | ||
256 | XFS_STATS_INC(mp, xs_read_calls); | |
257 | ||
258 | if (XFS_FORCED_SHUTDOWN(mp)) | |
259 | return -EIO; | |
260 | ||
16d4d435 CH |
261 | if (IS_DAX(inode)) |
262 | ret = xfs_file_dax_read(iocb, to); | |
263 | else if (iocb->ki_flags & IOCB_DIRECT) | |
bbc5a740 | 264 | ret = xfs_file_dio_aio_read(iocb, to); |
3176c3e0 | 265 | else |
bbc5a740 | 266 | ret = xfs_file_buffered_aio_read(iocb, to); |
dda35b8f | 267 | |
dda35b8f | 268 | if (ret > 0) |
ff6d6af2 | 269 | XFS_STATS_ADD(mp, xs_read_bytes, ret); |
dda35b8f CH |
270 | return ret; |
271 | } | |
272 | ||
4d8d1581 DC |
273 | /* |
274 | * Common pre-write limit and setup checks. | |
275 | * | |
5bf1f262 CH |
276 | * Called with the iolocked held either shared and exclusive according to |
277 | * @iolock, and returns with it held. Might upgrade the iolock to exclusive | |
278 | * if called for a direct write beyond i_size. | |
4d8d1581 DC |
279 | */ |
280 | STATIC ssize_t | |
281 | xfs_file_aio_write_checks( | |
99733fa3 AV |
282 | struct kiocb *iocb, |
283 | struct iov_iter *from, | |
4d8d1581 DC |
284 | int *iolock) |
285 | { | |
99733fa3 | 286 | struct file *file = iocb->ki_filp; |
4d8d1581 DC |
287 | struct inode *inode = file->f_mapping->host; |
288 | struct xfs_inode *ip = XFS_I(inode); | |
3309dd04 | 289 | ssize_t error = 0; |
99733fa3 | 290 | size_t count = iov_iter_count(from); |
3136e8bb | 291 | bool drained_dio = false; |
f5c54717 | 292 | loff_t isize; |
4d8d1581 | 293 | |
7271d243 | 294 | restart: |
3309dd04 AV |
295 | error = generic_write_checks(iocb, from); |
296 | if (error <= 0) | |
4d8d1581 | 297 | return error; |
4d8d1581 | 298 | |
69eb5fa1 | 299 | error = xfs_break_layouts(inode, iolock, BREAK_WRITE); |
781355c6 CH |
300 | if (error) |
301 | return error; | |
302 | ||
65523218 CH |
303 | /* |
304 | * For changing security info in file_remove_privs() we need i_rwsem | |
305 | * exclusively. | |
306 | */ | |
a6de82ca | 307 | if (*iolock == XFS_IOLOCK_SHARED && !IS_NOSEC(inode)) { |
65523218 | 308 | xfs_iunlock(ip, *iolock); |
a6de82ca | 309 | *iolock = XFS_IOLOCK_EXCL; |
65523218 | 310 | xfs_ilock(ip, *iolock); |
a6de82ca JK |
311 | goto restart; |
312 | } | |
4d8d1581 DC |
313 | /* |
314 | * If the offset is beyond the size of the file, we need to zero any | |
315 | * blocks that fall between the existing EOF and the start of this | |
2813d682 | 316 | * write. If zeroing is needed and we are currently holding the |
467f7899 CH |
317 | * iolock shared, we need to update it to exclusive which implies |
318 | * having to redo all checks before. | |
b9d59846 DC |
319 | * |
320 | * We need to serialise against EOF updates that occur in IO | |
321 | * completions here. We want to make sure that nobody is changing the | |
322 | * size while we do this check until we have placed an IO barrier (i.e. | |
323 | * hold the XFS_IOLOCK_EXCL) that prevents new IO from being dispatched. | |
324 | * The spinlock effectively forms a memory barrier once we have the | |
325 | * XFS_IOLOCK_EXCL so we are guaranteed to see the latest EOF value | |
326 | * and hence be able to correctly determine if we need to run zeroing. | |
4d8d1581 | 327 | */ |
b9d59846 | 328 | spin_lock(&ip->i_flags_lock); |
f5c54717 CH |
329 | isize = i_size_read(inode); |
330 | if (iocb->ki_pos > isize) { | |
b9d59846 | 331 | spin_unlock(&ip->i_flags_lock); |
3136e8bb BF |
332 | if (!drained_dio) { |
333 | if (*iolock == XFS_IOLOCK_SHARED) { | |
65523218 | 334 | xfs_iunlock(ip, *iolock); |
3136e8bb | 335 | *iolock = XFS_IOLOCK_EXCL; |
65523218 | 336 | xfs_ilock(ip, *iolock); |
3136e8bb BF |
337 | iov_iter_reexpand(from, count); |
338 | } | |
40c63fbc DC |
339 | /* |
340 | * We now have an IO submission barrier in place, but | |
341 | * AIO can do EOF updates during IO completion and hence | |
342 | * we now need to wait for all of them to drain. Non-AIO | |
343 | * DIO will have drained before we are given the | |
344 | * XFS_IOLOCK_EXCL, and so for most cases this wait is a | |
345 | * no-op. | |
346 | */ | |
347 | inode_dio_wait(inode); | |
3136e8bb | 348 | drained_dio = true; |
7271d243 DC |
349 | goto restart; |
350 | } | |
f5c54717 CH |
351 | |
352 | trace_xfs_zero_eof(ip, isize, iocb->ki_pos - isize); | |
353 | error = iomap_zero_range(inode, isize, iocb->ki_pos - isize, | |
354 | NULL, &xfs_iomap_ops); | |
467f7899 CH |
355 | if (error) |
356 | return error; | |
b9d59846 DC |
357 | } else |
358 | spin_unlock(&ip->i_flags_lock); | |
4d8d1581 | 359 | |
8a9c9980 CH |
360 | /* |
361 | * Updating the timestamps will grab the ilock again from | |
362 | * xfs_fs_dirty_inode, so we have to call it after dropping the | |
363 | * lock above. Eventually we should look into a way to avoid | |
364 | * the pointless lock roundtrip. | |
365 | */ | |
8c3f406c | 366 | return file_modified(file); |
4d8d1581 DC |
367 | } |
368 | ||
acdda3aa CH |
369 | static int |
370 | xfs_dio_write_end_io( | |
371 | struct kiocb *iocb, | |
372 | ssize_t size, | |
6fe7b990 | 373 | int error, |
acdda3aa CH |
374 | unsigned flags) |
375 | { | |
376 | struct inode *inode = file_inode(iocb->ki_filp); | |
377 | struct xfs_inode *ip = XFS_I(inode); | |
378 | loff_t offset = iocb->ki_pos; | |
73d30d48 | 379 | unsigned int nofs_flag; |
acdda3aa CH |
380 | |
381 | trace_xfs_end_io_direct_write(ip, offset, size); | |
382 | ||
383 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) | |
384 | return -EIO; | |
385 | ||
6fe7b990 MB |
386 | if (error) |
387 | return error; | |
388 | if (!size) | |
389 | return 0; | |
acdda3aa | 390 | |
ed5c3e66 DC |
391 | /* |
392 | * Capture amount written on completion as we can't reliably account | |
393 | * for it on submission. | |
394 | */ | |
395 | XFS_STATS_ADD(ip->i_mount, xs_write_bytes, size); | |
396 | ||
73d30d48 CH |
397 | /* |
398 | * We can allocate memory here while doing writeback on behalf of | |
399 | * memory reclaim. To avoid memory allocation deadlocks set the | |
400 | * task-wide nofs context for the following operations. | |
401 | */ | |
402 | nofs_flag = memalloc_nofs_save(); | |
403 | ||
ee70daab EG |
404 | if (flags & IOMAP_DIO_COW) { |
405 | error = xfs_reflink_end_cow(ip, offset, size); | |
406 | if (error) | |
73d30d48 | 407 | goto out; |
ee70daab EG |
408 | } |
409 | ||
410 | /* | |
411 | * Unwritten conversion updates the in-core isize after extent | |
412 | * conversion but before updating the on-disk size. Updating isize any | |
413 | * earlier allows a racing dio read to find unwritten extents before | |
414 | * they are converted. | |
415 | */ | |
73d30d48 CH |
416 | if (flags & IOMAP_DIO_UNWRITTEN) { |
417 | error = xfs_iomap_write_unwritten(ip, offset, size, true); | |
418 | goto out; | |
419 | } | |
ee70daab | 420 | |
acdda3aa CH |
421 | /* |
422 | * We need to update the in-core inode size here so that we don't end up | |
423 | * with the on-disk inode size being outside the in-core inode size. We | |
424 | * have no other method of updating EOF for AIO, so always do it here | |
425 | * if necessary. | |
426 | * | |
427 | * We need to lock the test/set EOF update as we can be racing with | |
428 | * other IO completions here to update the EOF. Failing to serialise | |
429 | * here can result in EOF moving backwards and Bad Things Happen when | |
430 | * that occurs. | |
431 | */ | |
432 | spin_lock(&ip->i_flags_lock); | |
433 | if (offset + size > i_size_read(inode)) { | |
434 | i_size_write(inode, offset + size); | |
ee70daab | 435 | spin_unlock(&ip->i_flags_lock); |
acdda3aa | 436 | error = xfs_setfilesize(ip, offset, size); |
ee70daab EG |
437 | } else { |
438 | spin_unlock(&ip->i_flags_lock); | |
439 | } | |
acdda3aa | 440 | |
73d30d48 CH |
441 | out: |
442 | memalloc_nofs_restore(nofs_flag); | |
acdda3aa CH |
443 | return error; |
444 | } | |
445 | ||
838c4f3d CH |
446 | static const struct iomap_dio_ops xfs_dio_write_ops = { |
447 | .end_io = xfs_dio_write_end_io, | |
448 | }; | |
449 | ||
f0d26e86 DC |
450 | /* |
451 | * xfs_file_dio_aio_write - handle direct IO writes | |
452 | * | |
453 | * Lock the inode appropriately to prepare for and issue a direct IO write. | |
eda77982 | 454 | * By separating it from the buffered write path we remove all the tricky to |
f0d26e86 DC |
455 | * follow locking changes and looping. |
456 | * | |
eda77982 DC |
457 | * If there are cached pages or we're extending the file, we need IOLOCK_EXCL |
458 | * until we're sure the bytes at the new EOF have been zeroed and/or the cached | |
459 | * pages are flushed out. | |
460 | * | |
461 | * In most cases the direct IO writes will be done holding IOLOCK_SHARED | |
462 | * allowing them to be done in parallel with reads and other direct IO writes. | |
463 | * However, if the IO is not aligned to filesystem blocks, the direct IO layer | |
464 | * needs to do sub-block zeroing and that requires serialisation against other | |
465 | * direct IOs to the same block. In this case we need to serialise the | |
466 | * submission of the unaligned IOs so that we don't get racing block zeroing in | |
467 | * the dio layer. To avoid the problem with aio, we also need to wait for | |
468 | * outstanding IOs to complete so that unwritten extent conversion is completed | |
469 | * before we try to map the overlapping block. This is currently implemented by | |
4a06fd26 | 470 | * hitting it with a big hammer (i.e. inode_dio_wait()). |
eda77982 | 471 | * |
f0d26e86 DC |
472 | * Returns with locks held indicated by @iolock and errors indicated by |
473 | * negative return values. | |
474 | */ | |
475 | STATIC ssize_t | |
476 | xfs_file_dio_aio_write( | |
477 | struct kiocb *iocb, | |
b3188919 | 478 | struct iov_iter *from) |
f0d26e86 DC |
479 | { |
480 | struct file *file = iocb->ki_filp; | |
481 | struct address_space *mapping = file->f_mapping; | |
482 | struct inode *inode = mapping->host; | |
483 | struct xfs_inode *ip = XFS_I(inode); | |
484 | struct xfs_mount *mp = ip->i_mount; | |
485 | ssize_t ret = 0; | |
eda77982 | 486 | int unaligned_io = 0; |
d0606464 | 487 | int iolock; |
b3188919 | 488 | size_t count = iov_iter_count(from); |
acdda3aa | 489 | struct xfs_buftarg *target = XFS_IS_REALTIME_INODE(ip) ? |
f0d26e86 DC |
490 | mp->m_rtdev_targp : mp->m_ddev_targp; |
491 | ||
7c71ee78 | 492 | /* DIO must be aligned to device logical sector size */ |
16d4d435 | 493 | if ((iocb->ki_pos | count) & target->bt_logical_sectormask) |
b474c7ae | 494 | return -EINVAL; |
f0d26e86 | 495 | |
7271d243 | 496 | /* |
0ee7a3f6 CH |
497 | * Don't take the exclusive iolock here unless the I/O is unaligned to |
498 | * the file system block size. We don't need to consider the EOF | |
499 | * extension case here because xfs_file_aio_write_checks() will relock | |
500 | * the inode as necessary for EOF zeroing cases and fill out the new | |
501 | * inode size as appropriate. | |
7271d243 | 502 | */ |
0ee7a3f6 CH |
503 | if ((iocb->ki_pos & mp->m_blockmask) || |
504 | ((iocb->ki_pos + count) & mp->m_blockmask)) { | |
505 | unaligned_io = 1; | |
54a4ef8a CH |
506 | |
507 | /* | |
508 | * We can't properly handle unaligned direct I/O to reflink | |
509 | * files yet, as we can't unshare a partial block. | |
510 | */ | |
66ae56a5 | 511 | if (xfs_is_cow_inode(ip)) { |
54a4ef8a CH |
512 | trace_xfs_reflink_bounce_dio_write(ip, iocb->ki_pos, count); |
513 | return -EREMCHG; | |
514 | } | |
d0606464 | 515 | iolock = XFS_IOLOCK_EXCL; |
0ee7a3f6 | 516 | } else { |
d0606464 | 517 | iolock = XFS_IOLOCK_SHARED; |
c58cb165 | 518 | } |
f0d26e86 | 519 | |
942491c9 | 520 | if (iocb->ki_flags & IOCB_NOWAIT) { |
1fdeaea4 DW |
521 | /* unaligned dio always waits, bail */ |
522 | if (unaligned_io) | |
523 | return -EAGAIN; | |
942491c9 | 524 | if (!xfs_ilock_nowait(ip, iolock)) |
29a5d29e | 525 | return -EAGAIN; |
942491c9 | 526 | } else { |
29a5d29e GR |
527 | xfs_ilock(ip, iolock); |
528 | } | |
0ee7a3f6 | 529 | |
99733fa3 | 530 | ret = xfs_file_aio_write_checks(iocb, from, &iolock); |
4d8d1581 | 531 | if (ret) |
d0606464 | 532 | goto out; |
99733fa3 | 533 | count = iov_iter_count(from); |
f0d26e86 | 534 | |
eda77982 | 535 | /* |
2032a8a2 BF |
536 | * If we are doing unaligned IO, we can't allow any other overlapping IO |
537 | * in-flight at the same time or we risk data corruption. Wait for all | |
538 | * other IO to drain before we submit. If the IO is aligned, demote the | |
539 | * iolock if we had to take the exclusive lock in | |
540 | * xfs_file_aio_write_checks() for other reasons. | |
eda77982 | 541 | */ |
29a5d29e | 542 | if (unaligned_io) { |
2032a8a2 | 543 | inode_dio_wait(inode); |
29a5d29e | 544 | } else if (iolock == XFS_IOLOCK_EXCL) { |
65523218 | 545 | xfs_ilock_demote(ip, XFS_IOLOCK_EXCL); |
d0606464 | 546 | iolock = XFS_IOLOCK_SHARED; |
f0d26e86 DC |
547 | } |
548 | ||
3176c3e0 | 549 | trace_xfs_file_direct_write(ip, count, iocb->ki_pos); |
838c4f3d | 550 | ret = iomap_dio_rw(iocb, from, &xfs_iomap_ops, &xfs_dio_write_ops); |
2032a8a2 BF |
551 | |
552 | /* | |
553 | * If unaligned, this is the only IO in-flight. If it has not yet | |
554 | * completed, wait on it before we release the iolock to prevent | |
555 | * subsequent overlapping IO. | |
556 | */ | |
557 | if (ret == -EIOCBQUEUED && unaligned_io) | |
558 | inode_dio_wait(inode); | |
d0606464 | 559 | out: |
65523218 | 560 | xfs_iunlock(ip, iolock); |
d0606464 | 561 | |
6b698ede | 562 | /* |
16d4d435 CH |
563 | * No fallback to buffered IO on errors for XFS, direct IO will either |
564 | * complete fully or fail. | |
6b698ede | 565 | */ |
16d4d435 CH |
566 | ASSERT(ret < 0 || ret == count); |
567 | return ret; | |
568 | } | |
569 | ||
f021bd07 | 570 | static noinline ssize_t |
16d4d435 CH |
571 | xfs_file_dax_write( |
572 | struct kiocb *iocb, | |
573 | struct iov_iter *from) | |
574 | { | |
6c31f495 | 575 | struct inode *inode = iocb->ki_filp->f_mapping->host; |
16d4d435 | 576 | struct xfs_inode *ip = XFS_I(inode); |
17879e8f | 577 | int iolock = XFS_IOLOCK_EXCL; |
6c31f495 CH |
578 | ssize_t ret, error = 0; |
579 | size_t count; | |
580 | loff_t pos; | |
16d4d435 | 581 | |
942491c9 CH |
582 | if (iocb->ki_flags & IOCB_NOWAIT) { |
583 | if (!xfs_ilock_nowait(ip, iolock)) | |
29a5d29e | 584 | return -EAGAIN; |
942491c9 | 585 | } else { |
29a5d29e GR |
586 | xfs_ilock(ip, iolock); |
587 | } | |
588 | ||
16d4d435 CH |
589 | ret = xfs_file_aio_write_checks(iocb, from, &iolock); |
590 | if (ret) | |
591 | goto out; | |
592 | ||
6c31f495 CH |
593 | pos = iocb->ki_pos; |
594 | count = iov_iter_count(from); | |
8b2180b3 | 595 | |
6c31f495 | 596 | trace_xfs_file_dax_write(ip, count, pos); |
11c59c92 | 597 | ret = dax_iomap_rw(iocb, from, &xfs_iomap_ops); |
6c31f495 CH |
598 | if (ret > 0 && iocb->ki_pos > i_size_read(inode)) { |
599 | i_size_write(inode, iocb->ki_pos); | |
600 | error = xfs_setfilesize(ip, pos, ret); | |
16d4d435 | 601 | } |
16d4d435 | 602 | out: |
65523218 | 603 | xfs_iunlock(ip, iolock); |
ed5c3e66 DC |
604 | if (error) |
605 | return error; | |
606 | ||
607 | if (ret > 0) { | |
608 | XFS_STATS_ADD(ip->i_mount, xs_write_bytes, ret); | |
609 | ||
610 | /* Handle various SYNC-type writes */ | |
611 | ret = generic_write_sync(iocb, ret); | |
612 | } | |
613 | return ret; | |
f0d26e86 DC |
614 | } |
615 | ||
00258e36 | 616 | STATIC ssize_t |
637bbc75 | 617 | xfs_file_buffered_aio_write( |
dda35b8f | 618 | struct kiocb *iocb, |
b3188919 | 619 | struct iov_iter *from) |
dda35b8f CH |
620 | { |
621 | struct file *file = iocb->ki_filp; | |
622 | struct address_space *mapping = file->f_mapping; | |
623 | struct inode *inode = mapping->host; | |
00258e36 | 624 | struct xfs_inode *ip = XFS_I(inode); |
637bbc75 DC |
625 | ssize_t ret; |
626 | int enospc = 0; | |
c3155097 | 627 | int iolock; |
dda35b8f | 628 | |
91f9943e CH |
629 | if (iocb->ki_flags & IOCB_NOWAIT) |
630 | return -EOPNOTSUPP; | |
631 | ||
c3155097 BF |
632 | write_retry: |
633 | iolock = XFS_IOLOCK_EXCL; | |
65523218 | 634 | xfs_ilock(ip, iolock); |
dda35b8f | 635 | |
99733fa3 | 636 | ret = xfs_file_aio_write_checks(iocb, from, &iolock); |
4d8d1581 | 637 | if (ret) |
d0606464 | 638 | goto out; |
dda35b8f CH |
639 | |
640 | /* We can write back this queue in page reclaim */ | |
de1414a6 | 641 | current->backing_dev_info = inode_to_bdi(inode); |
dda35b8f | 642 | |
3176c3e0 | 643 | trace_xfs_file_buffered_write(ip, iov_iter_count(from), iocb->ki_pos); |
68a9f5e7 | 644 | ret = iomap_file_buffered_write(iocb, from, &xfs_iomap_ops); |
0a64bc2c | 645 | if (likely(ret >= 0)) |
99733fa3 | 646 | iocb->ki_pos += ret; |
dc06f398 | 647 | |
637bbc75 | 648 | /* |
dc06f398 BF |
649 | * If we hit a space limit, try to free up some lingering preallocated |
650 | * space before returning an error. In the case of ENOSPC, first try to | |
651 | * write back all dirty inodes to free up some of the excess reserved | |
652 | * metadata space. This reduces the chances that the eofblocks scan | |
653 | * waits on dirty mappings. Since xfs_flush_inodes() is serialized, this | |
654 | * also behaves as a filter to prevent too many eofblocks scans from | |
655 | * running at the same time. | |
637bbc75 | 656 | */ |
dc06f398 | 657 | if (ret == -EDQUOT && !enospc) { |
c3155097 | 658 | xfs_iunlock(ip, iolock); |
dc06f398 BF |
659 | enospc = xfs_inode_free_quota_eofblocks(ip); |
660 | if (enospc) | |
661 | goto write_retry; | |
83104d44 DW |
662 | enospc = xfs_inode_free_quota_cowblocks(ip); |
663 | if (enospc) | |
664 | goto write_retry; | |
c3155097 | 665 | iolock = 0; |
dc06f398 BF |
666 | } else if (ret == -ENOSPC && !enospc) { |
667 | struct xfs_eofblocks eofb = {0}; | |
668 | ||
637bbc75 | 669 | enospc = 1; |
9aa05000 | 670 | xfs_flush_inodes(ip->i_mount); |
c3155097 BF |
671 | |
672 | xfs_iunlock(ip, iolock); | |
dc06f398 BF |
673 | eofb.eof_flags = XFS_EOF_FLAGS_SYNC; |
674 | xfs_icache_free_eofblocks(ip->i_mount, &eofb); | |
cf2cb784 | 675 | xfs_icache_free_cowblocks(ip->i_mount, &eofb); |
9aa05000 | 676 | goto write_retry; |
dda35b8f | 677 | } |
d0606464 | 678 | |
dda35b8f | 679 | current->backing_dev_info = NULL; |
d0606464 | 680 | out: |
c3155097 BF |
681 | if (iolock) |
682 | xfs_iunlock(ip, iolock); | |
ed5c3e66 DC |
683 | |
684 | if (ret > 0) { | |
685 | XFS_STATS_ADD(ip->i_mount, xs_write_bytes, ret); | |
686 | /* Handle various SYNC-type writes */ | |
687 | ret = generic_write_sync(iocb, ret); | |
688 | } | |
637bbc75 DC |
689 | return ret; |
690 | } | |
691 | ||
692 | STATIC ssize_t | |
bf97f3bc | 693 | xfs_file_write_iter( |
637bbc75 | 694 | struct kiocb *iocb, |
bf97f3bc | 695 | struct iov_iter *from) |
637bbc75 DC |
696 | { |
697 | struct file *file = iocb->ki_filp; | |
698 | struct address_space *mapping = file->f_mapping; | |
699 | struct inode *inode = mapping->host; | |
700 | struct xfs_inode *ip = XFS_I(inode); | |
701 | ssize_t ret; | |
bf97f3bc | 702 | size_t ocount = iov_iter_count(from); |
637bbc75 | 703 | |
ff6d6af2 | 704 | XFS_STATS_INC(ip->i_mount, xs_write_calls); |
637bbc75 | 705 | |
637bbc75 DC |
706 | if (ocount == 0) |
707 | return 0; | |
708 | ||
bf97f3bc AV |
709 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
710 | return -EIO; | |
637bbc75 | 711 | |
16d4d435 | 712 | if (IS_DAX(inode)) |
ed5c3e66 DC |
713 | return xfs_file_dax_write(iocb, from); |
714 | ||
715 | if (iocb->ki_flags & IOCB_DIRECT) { | |
0613f16c DW |
716 | /* |
717 | * Allow a directio write to fall back to a buffered | |
718 | * write *only* in the case that we're doing a reflink | |
719 | * CoW. In all other directio scenarios we do not | |
720 | * allow an operation to fall back to buffered mode. | |
721 | */ | |
bf97f3bc | 722 | ret = xfs_file_dio_aio_write(iocb, from); |
ed5c3e66 DC |
723 | if (ret != -EREMCHG) |
724 | return ret; | |
0613f16c | 725 | } |
dda35b8f | 726 | |
ed5c3e66 | 727 | return xfs_file_buffered_aio_write(iocb, from); |
dda35b8f CH |
728 | } |
729 | ||
d6dc57e2 DW |
730 | static void |
731 | xfs_wait_dax_page( | |
e25ff835 | 732 | struct inode *inode) |
d6dc57e2 DW |
733 | { |
734 | struct xfs_inode *ip = XFS_I(inode); | |
735 | ||
d6dc57e2 DW |
736 | xfs_iunlock(ip, XFS_MMAPLOCK_EXCL); |
737 | schedule(); | |
738 | xfs_ilock(ip, XFS_MMAPLOCK_EXCL); | |
739 | } | |
740 | ||
741 | static int | |
742 | xfs_break_dax_layouts( | |
743 | struct inode *inode, | |
e25ff835 | 744 | bool *retry) |
d6dc57e2 DW |
745 | { |
746 | struct page *page; | |
747 | ||
748 | ASSERT(xfs_isilocked(XFS_I(inode), XFS_MMAPLOCK_EXCL)); | |
749 | ||
750 | page = dax_layout_busy_page(inode->i_mapping); | |
751 | if (!page) | |
752 | return 0; | |
753 | ||
e25ff835 | 754 | *retry = true; |
d6dc57e2 DW |
755 | return ___wait_var_event(&page->_refcount, |
756 | atomic_read(&page->_refcount) == 1, TASK_INTERRUPTIBLE, | |
e25ff835 | 757 | 0, 0, xfs_wait_dax_page(inode)); |
d6dc57e2 DW |
758 | } |
759 | ||
69eb5fa1 DW |
760 | int |
761 | xfs_break_layouts( | |
762 | struct inode *inode, | |
763 | uint *iolock, | |
764 | enum layout_break_reason reason) | |
765 | { | |
766 | bool retry; | |
d6dc57e2 | 767 | int error; |
69eb5fa1 DW |
768 | |
769 | ASSERT(xfs_isilocked(XFS_I(inode), XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)); | |
770 | ||
d6dc57e2 DW |
771 | do { |
772 | retry = false; | |
773 | switch (reason) { | |
774 | case BREAK_UNMAP: | |
a4722a64 | 775 | error = xfs_break_dax_layouts(inode, &retry); |
d6dc57e2 DW |
776 | if (error || retry) |
777 | break; | |
778 | /* fall through */ | |
779 | case BREAK_WRITE: | |
780 | error = xfs_break_leased_layouts(inode, iolock, &retry); | |
781 | break; | |
782 | default: | |
783 | WARN_ON_ONCE(1); | |
784 | error = -EINVAL; | |
785 | } | |
786 | } while (error == 0 && retry); | |
787 | ||
788 | return error; | |
69eb5fa1 DW |
789 | } |
790 | ||
a904b1ca NJ |
791 | #define XFS_FALLOC_FL_SUPPORTED \ |
792 | (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \ | |
793 | FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE | \ | |
98cc2db5 | 794 | FALLOC_FL_INSERT_RANGE | FALLOC_FL_UNSHARE_RANGE) |
a904b1ca | 795 | |
2fe17c10 CH |
796 | STATIC long |
797 | xfs_file_fallocate( | |
83aee9e4 CH |
798 | struct file *file, |
799 | int mode, | |
800 | loff_t offset, | |
801 | loff_t len) | |
2fe17c10 | 802 | { |
83aee9e4 CH |
803 | struct inode *inode = file_inode(file); |
804 | struct xfs_inode *ip = XFS_I(inode); | |
83aee9e4 | 805 | long error; |
8add71ca | 806 | enum xfs_prealloc_flags flags = 0; |
c63a8eae | 807 | uint iolock = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL; |
83aee9e4 | 808 | loff_t new_size = 0; |
749f24f3 | 809 | bool do_file_insert = false; |
2fe17c10 | 810 | |
83aee9e4 CH |
811 | if (!S_ISREG(inode->i_mode)) |
812 | return -EINVAL; | |
a904b1ca | 813 | if (mode & ~XFS_FALLOC_FL_SUPPORTED) |
2fe17c10 CH |
814 | return -EOPNOTSUPP; |
815 | ||
781355c6 | 816 | xfs_ilock(ip, iolock); |
69eb5fa1 | 817 | error = xfs_break_layouts(inode, &iolock, BREAK_UNMAP); |
781355c6 CH |
818 | if (error) |
819 | goto out_unlock; | |
820 | ||
83aee9e4 CH |
821 | if (mode & FALLOC_FL_PUNCH_HOLE) { |
822 | error = xfs_free_file_space(ip, offset, len); | |
823 | if (error) | |
824 | goto out_unlock; | |
e1d8fb88 | 825 | } else if (mode & FALLOC_FL_COLLAPSE_RANGE) { |
93407472 | 826 | unsigned int blksize_mask = i_blocksize(inode) - 1; |
e1d8fb88 NJ |
827 | |
828 | if (offset & blksize_mask || len & blksize_mask) { | |
2451337d | 829 | error = -EINVAL; |
e1d8fb88 NJ |
830 | goto out_unlock; |
831 | } | |
832 | ||
23fffa92 LC |
833 | /* |
834 | * There is no need to overlap collapse range with EOF, | |
835 | * in which case it is effectively a truncate operation | |
836 | */ | |
837 | if (offset + len >= i_size_read(inode)) { | |
2451337d | 838 | error = -EINVAL; |
23fffa92 LC |
839 | goto out_unlock; |
840 | } | |
841 | ||
e1d8fb88 NJ |
842 | new_size = i_size_read(inode) - len; |
843 | ||
844 | error = xfs_collapse_file_space(ip, offset, len); | |
845 | if (error) | |
846 | goto out_unlock; | |
a904b1ca | 847 | } else if (mode & FALLOC_FL_INSERT_RANGE) { |
7d83fb14 DW |
848 | unsigned int blksize_mask = i_blocksize(inode) - 1; |
849 | loff_t isize = i_size_read(inode); | |
a904b1ca | 850 | |
a904b1ca NJ |
851 | if (offset & blksize_mask || len & blksize_mask) { |
852 | error = -EINVAL; | |
853 | goto out_unlock; | |
854 | } | |
855 | ||
7d83fb14 DW |
856 | /* |
857 | * New inode size must not exceed ->s_maxbytes, accounting for | |
858 | * possible signed overflow. | |
859 | */ | |
860 | if (inode->i_sb->s_maxbytes - isize < len) { | |
a904b1ca NJ |
861 | error = -EFBIG; |
862 | goto out_unlock; | |
863 | } | |
7d83fb14 | 864 | new_size = isize + len; |
a904b1ca NJ |
865 | |
866 | /* Offset should be less than i_size */ | |
7d83fb14 | 867 | if (offset >= isize) { |
a904b1ca NJ |
868 | error = -EINVAL; |
869 | goto out_unlock; | |
870 | } | |
749f24f3 | 871 | do_file_insert = true; |
83aee9e4 | 872 | } else { |
8add71ca CH |
873 | flags |= XFS_PREALLOC_SET; |
874 | ||
83aee9e4 CH |
875 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
876 | offset + len > i_size_read(inode)) { | |
877 | new_size = offset + len; | |
2451337d | 878 | error = inode_newsize_ok(inode, new_size); |
83aee9e4 CH |
879 | if (error) |
880 | goto out_unlock; | |
881 | } | |
2fe17c10 | 882 | |
66ae56a5 | 883 | if (mode & FALLOC_FL_ZERO_RANGE) { |
376ba313 | 884 | error = xfs_zero_file_space(ip, offset, len); |
66ae56a5 CH |
885 | } else if (mode & FALLOC_FL_UNSHARE_RANGE) { |
886 | error = xfs_reflink_unshare(ip, offset, len); | |
887 | if (error) | |
888 | goto out_unlock; | |
889 | ||
890 | if (!xfs_is_always_cow_inode(ip)) { | |
891 | error = xfs_alloc_file_space(ip, offset, len, | |
892 | XFS_BMAPI_PREALLOC); | |
98cc2db5 | 893 | } |
66ae56a5 CH |
894 | } else { |
895 | /* | |
896 | * If always_cow mode we can't use preallocations and | |
897 | * thus should not create them. | |
898 | */ | |
899 | if (xfs_is_always_cow_inode(ip)) { | |
900 | error = -EOPNOTSUPP; | |
901 | goto out_unlock; | |
902 | } | |
903 | ||
376ba313 LC |
904 | error = xfs_alloc_file_space(ip, offset, len, |
905 | XFS_BMAPI_PREALLOC); | |
98cc2db5 | 906 | } |
2fe17c10 CH |
907 | if (error) |
908 | goto out_unlock; | |
909 | } | |
910 | ||
83aee9e4 | 911 | if (file->f_flags & O_DSYNC) |
8add71ca CH |
912 | flags |= XFS_PREALLOC_SYNC; |
913 | ||
914 | error = xfs_update_prealloc_flags(ip, flags); | |
2fe17c10 CH |
915 | if (error) |
916 | goto out_unlock; | |
917 | ||
918 | /* Change file size if needed */ | |
919 | if (new_size) { | |
920 | struct iattr iattr; | |
921 | ||
922 | iattr.ia_valid = ATTR_SIZE; | |
923 | iattr.ia_size = new_size; | |
69bca807 | 924 | error = xfs_vn_setattr_size(file_dentry(file), &iattr); |
a904b1ca NJ |
925 | if (error) |
926 | goto out_unlock; | |
2fe17c10 CH |
927 | } |
928 | ||
a904b1ca NJ |
929 | /* |
930 | * Perform hole insertion now that the file size has been | |
931 | * updated so that if we crash during the operation we don't | |
932 | * leave shifted extents past EOF and hence losing access to | |
933 | * the data that is contained within them. | |
934 | */ | |
935 | if (do_file_insert) | |
936 | error = xfs_insert_file_space(ip, offset, len); | |
937 | ||
2fe17c10 | 938 | out_unlock: |
781355c6 | 939 | xfs_iunlock(ip, iolock); |
2451337d | 940 | return error; |
2fe17c10 CH |
941 | } |
942 | ||
40144e49 JK |
943 | STATIC int |
944 | xfs_file_fadvise( | |
945 | struct file *file, | |
946 | loff_t start, | |
947 | loff_t end, | |
948 | int advice) | |
949 | { | |
950 | struct xfs_inode *ip = XFS_I(file_inode(file)); | |
951 | int ret; | |
952 | int lockflags = 0; | |
953 | ||
954 | /* | |
955 | * Operations creating pages in page cache need protection from hole | |
956 | * punching and similar ops | |
957 | */ | |
958 | if (advice == POSIX_FADV_WILLNEED) { | |
959 | lockflags = XFS_IOLOCK_SHARED; | |
960 | xfs_ilock(ip, lockflags); | |
961 | } | |
962 | ret = generic_fadvise(file, start, end, advice); | |
963 | if (lockflags) | |
964 | xfs_iunlock(ip, lockflags); | |
965 | return ret; | |
966 | } | |
3fc9f5e4 | 967 | |
da034bcc | 968 | STATIC loff_t |
2e5dfc99 | 969 | xfs_file_remap_range( |
3fc9f5e4 DW |
970 | struct file *file_in, |
971 | loff_t pos_in, | |
972 | struct file *file_out, | |
973 | loff_t pos_out, | |
974 | loff_t len, | |
975 | unsigned int remap_flags) | |
9fe26045 | 976 | { |
3fc9f5e4 DW |
977 | struct inode *inode_in = file_inode(file_in); |
978 | struct xfs_inode *src = XFS_I(inode_in); | |
979 | struct inode *inode_out = file_inode(file_out); | |
980 | struct xfs_inode *dest = XFS_I(inode_out); | |
981 | struct xfs_mount *mp = src->i_mount; | |
982 | loff_t remapped = 0; | |
983 | xfs_extlen_t cowextsize; | |
984 | int ret; | |
985 | ||
2e5dfc99 DW |
986 | if (remap_flags & ~(REMAP_FILE_DEDUP | REMAP_FILE_ADVISORY)) |
987 | return -EINVAL; | |
cc714660 | 988 | |
3fc9f5e4 DW |
989 | if (!xfs_sb_version_hasreflink(&mp->m_sb)) |
990 | return -EOPNOTSUPP; | |
991 | ||
992 | if (XFS_FORCED_SHUTDOWN(mp)) | |
993 | return -EIO; | |
994 | ||
995 | /* Prepare and then clone file data. */ | |
996 | ret = xfs_reflink_remap_prep(file_in, pos_in, file_out, pos_out, | |
997 | &len, remap_flags); | |
998 | if (ret < 0 || len == 0) | |
999 | return ret; | |
1000 | ||
1001 | trace_xfs_reflink_remap_range(src, pos_in, len, dest, pos_out); | |
1002 | ||
1003 | ret = xfs_reflink_remap_blocks(src, pos_in, dest, pos_out, len, | |
1004 | &remapped); | |
1005 | if (ret) | |
1006 | goto out_unlock; | |
1007 | ||
1008 | /* | |
1009 | * Carry the cowextsize hint from src to dest if we're sharing the | |
1010 | * entire source file to the entire destination file, the source file | |
1011 | * has a cowextsize hint, and the destination file does not. | |
1012 | */ | |
1013 | cowextsize = 0; | |
1014 | if (pos_in == 0 && len == i_size_read(inode_in) && | |
1015 | (src->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE) && | |
1016 | pos_out == 0 && len >= i_size_read(inode_out) && | |
1017 | !(dest->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE)) | |
1018 | cowextsize = src->i_d.di_cowextsize; | |
1019 | ||
1020 | ret = xfs_reflink_update_dest(dest, pos_out + len, cowextsize, | |
1021 | remap_flags); | |
1022 | ||
1023 | out_unlock: | |
1024 | xfs_reflink_remap_unlock(file_in, file_out); | |
1025 | if (ret) | |
1026 | trace_xfs_reflink_remap_range_error(dest, ret, _RET_IP_); | |
1027 | return remapped > 0 ? remapped : ret; | |
9fe26045 | 1028 | } |
2fe17c10 | 1029 | |
1da177e4 | 1030 | STATIC int |
3562fd45 | 1031 | xfs_file_open( |
1da177e4 | 1032 | struct inode *inode, |
f999a5bf | 1033 | struct file *file) |
1da177e4 | 1034 | { |
f999a5bf | 1035 | if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS) |
1da177e4 | 1036 | return -EFBIG; |
f999a5bf CH |
1037 | if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb))) |
1038 | return -EIO; | |
91f9943e | 1039 | file->f_mode |= FMODE_NOWAIT; |
f999a5bf CH |
1040 | return 0; |
1041 | } | |
1042 | ||
1043 | STATIC int | |
1044 | xfs_dir_open( | |
1045 | struct inode *inode, | |
1046 | struct file *file) | |
1047 | { | |
1048 | struct xfs_inode *ip = XFS_I(inode); | |
1049 | int mode; | |
1050 | int error; | |
1051 | ||
1052 | error = xfs_file_open(inode, file); | |
1053 | if (error) | |
1054 | return error; | |
1055 | ||
1056 | /* | |
1057 | * If there are any blocks, read-ahead block 0 as we're almost | |
1058 | * certain to have the next operation be a read there. | |
1059 | */ | |
309ecac8 | 1060 | mode = xfs_ilock_data_map_shared(ip); |
f999a5bf | 1061 | if (ip->i_d.di_nextents > 0) |
7a652bbe | 1062 | error = xfs_dir3_data_readahead(ip, 0, -1); |
f999a5bf | 1063 | xfs_iunlock(ip, mode); |
7a652bbe | 1064 | return error; |
1da177e4 LT |
1065 | } |
1066 | ||
1da177e4 | 1067 | STATIC int |
3562fd45 | 1068 | xfs_file_release( |
1da177e4 LT |
1069 | struct inode *inode, |
1070 | struct file *filp) | |
1071 | { | |
2451337d | 1072 | return xfs_release(XFS_I(inode)); |
1da177e4 LT |
1073 | } |
1074 | ||
1da177e4 | 1075 | STATIC int |
3562fd45 | 1076 | xfs_file_readdir( |
b8227554 AV |
1077 | struct file *file, |
1078 | struct dir_context *ctx) | |
1da177e4 | 1079 | { |
b8227554 | 1080 | struct inode *inode = file_inode(file); |
739bfb2a | 1081 | xfs_inode_t *ip = XFS_I(inode); |
051e7cd4 CH |
1082 | size_t bufsize; |
1083 | ||
1084 | /* | |
1085 | * The Linux API doesn't pass down the total size of the buffer | |
1086 | * we read into down to the filesystem. With the filldir concept | |
1087 | * it's not needed for correct information, but the XFS dir2 leaf | |
1088 | * code wants an estimate of the buffer size to calculate it's | |
1089 | * readahead window and size the buffers used for mapping to | |
1090 | * physical blocks. | |
1091 | * | |
1092 | * Try to give it an estimate that's good enough, maybe at some | |
1093 | * point we can change the ->readdir prototype to include the | |
a9cc799e | 1094 | * buffer size. For now we use the current glibc buffer size. |
051e7cd4 | 1095 | */ |
a5c46e5e | 1096 | bufsize = (size_t)min_t(loff_t, XFS_READDIR_BUFSIZE, ip->i_d.di_size); |
051e7cd4 | 1097 | |
acb9553c | 1098 | return xfs_readdir(NULL, ip, ctx, bufsize); |
3fe3e6b1 JL |
1099 | } |
1100 | ||
1101 | STATIC loff_t | |
1102 | xfs_file_llseek( | |
1103 | struct file *file, | |
1104 | loff_t offset, | |
59f9c004 | 1105 | int whence) |
3fe3e6b1 | 1106 | { |
9b2970aa CH |
1107 | struct inode *inode = file->f_mapping->host; |
1108 | ||
1109 | if (XFS_FORCED_SHUTDOWN(XFS_I(inode)->i_mount)) | |
1110 | return -EIO; | |
1111 | ||
59f9c004 | 1112 | switch (whence) { |
9b2970aa | 1113 | default: |
59f9c004 | 1114 | return generic_file_llseek(file, offset, whence); |
3fe3e6b1 | 1115 | case SEEK_HOLE: |
60271ab7 | 1116 | offset = iomap_seek_hole(inode, offset, &xfs_seek_iomap_ops); |
9b2970aa | 1117 | break; |
49c69591 | 1118 | case SEEK_DATA: |
60271ab7 | 1119 | offset = iomap_seek_data(inode, offset, &xfs_seek_iomap_ops); |
9b2970aa | 1120 | break; |
3fe3e6b1 | 1121 | } |
9b2970aa CH |
1122 | |
1123 | if (offset < 0) | |
1124 | return offset; | |
1125 | return vfs_setpos(file, offset, inode->i_sb->s_maxbytes); | |
3fe3e6b1 JL |
1126 | } |
1127 | ||
de0e8c20 DC |
1128 | /* |
1129 | * Locking for serialisation of IO during page faults. This results in a lock | |
1130 | * ordering of: | |
1131 | * | |
1132 | * mmap_sem (MM) | |
6b698ede | 1133 | * sb_start_pagefault(vfs, freeze) |
13ad4fe3 | 1134 | * i_mmaplock (XFS - truncate serialisation) |
6b698ede DC |
1135 | * page_lock (MM) |
1136 | * i_lock (XFS - extent map serialisation) | |
de0e8c20 | 1137 | */ |
05edd888 | 1138 | static vm_fault_t |
d522d569 CH |
1139 | __xfs_filemap_fault( |
1140 | struct vm_fault *vmf, | |
1141 | enum page_entry_size pe_size, | |
1142 | bool write_fault) | |
de0e8c20 | 1143 | { |
11bac800 | 1144 | struct inode *inode = file_inode(vmf->vma->vm_file); |
d522d569 | 1145 | struct xfs_inode *ip = XFS_I(inode); |
05edd888 | 1146 | vm_fault_t ret; |
de0e8c20 | 1147 | |
d522d569 | 1148 | trace_xfs_filemap_fault(ip, pe_size, write_fault); |
de0e8c20 | 1149 | |
d522d569 CH |
1150 | if (write_fault) { |
1151 | sb_start_pagefault(inode->i_sb); | |
1152 | file_update_time(vmf->vma->vm_file); | |
1153 | } | |
de0e8c20 | 1154 | |
d522d569 | 1155 | xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED); |
6b698ede | 1156 | if (IS_DAX(inode)) { |
a39e596b CH |
1157 | pfn_t pfn; |
1158 | ||
c0b24625 | 1159 | ret = dax_iomap_fault(vmf, pe_size, &pfn, NULL, &xfs_iomap_ops); |
a39e596b CH |
1160 | if (ret & VM_FAULT_NEEDDSYNC) |
1161 | ret = dax_finish_sync_fault(vmf, pe_size, pfn); | |
6b698ede | 1162 | } else { |
d522d569 CH |
1163 | if (write_fault) |
1164 | ret = iomap_page_mkwrite(vmf, &xfs_iomap_ops); | |
1165 | else | |
1166 | ret = filemap_fault(vmf); | |
6b698ede | 1167 | } |
6b698ede | 1168 | xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED); |
6b698ede | 1169 | |
d522d569 CH |
1170 | if (write_fault) |
1171 | sb_end_pagefault(inode->i_sb); | |
6b698ede | 1172 | return ret; |
de0e8c20 DC |
1173 | } |
1174 | ||
05edd888 | 1175 | static vm_fault_t |
6b698ede | 1176 | xfs_filemap_fault( |
075a924d DC |
1177 | struct vm_fault *vmf) |
1178 | { | |
6b698ede | 1179 | /* DAX can shortcut the normal fault path on write faults! */ |
d522d569 CH |
1180 | return __xfs_filemap_fault(vmf, PE_SIZE_PTE, |
1181 | IS_DAX(file_inode(vmf->vma->vm_file)) && | |
1182 | (vmf->flags & FAULT_FLAG_WRITE)); | |
6b698ede DC |
1183 | } |
1184 | ||
05edd888 | 1185 | static vm_fault_t |
a2d58167 | 1186 | xfs_filemap_huge_fault( |
c791ace1 DJ |
1187 | struct vm_fault *vmf, |
1188 | enum page_entry_size pe_size) | |
acd76e74 | 1189 | { |
d522d569 | 1190 | if (!IS_DAX(file_inode(vmf->vma->vm_file))) |
acd76e74 MW |
1191 | return VM_FAULT_FALLBACK; |
1192 | ||
d522d569 CH |
1193 | /* DAX can shortcut the normal fault path on write faults! */ |
1194 | return __xfs_filemap_fault(vmf, pe_size, | |
1195 | (vmf->flags & FAULT_FLAG_WRITE)); | |
1196 | } | |
acd76e74 | 1197 | |
05edd888 | 1198 | static vm_fault_t |
d522d569 CH |
1199 | xfs_filemap_page_mkwrite( |
1200 | struct vm_fault *vmf) | |
1201 | { | |
1202 | return __xfs_filemap_fault(vmf, PE_SIZE_PTE, true); | |
acd76e74 MW |
1203 | } |
1204 | ||
3af49285 | 1205 | /* |
7b565c9f JK |
1206 | * pfn_mkwrite was originally intended to ensure we capture time stamp updates |
1207 | * on write faults. In reality, it needs to serialise against truncate and | |
1208 | * prepare memory for writing so handle is as standard write fault. | |
3af49285 | 1209 | */ |
05edd888 | 1210 | static vm_fault_t |
3af49285 | 1211 | xfs_filemap_pfn_mkwrite( |
3af49285 DC |
1212 | struct vm_fault *vmf) |
1213 | { | |
1214 | ||
7b565c9f | 1215 | return __xfs_filemap_fault(vmf, PE_SIZE_PTE, true); |
acd76e74 MW |
1216 | } |
1217 | ||
6b698ede DC |
1218 | static const struct vm_operations_struct xfs_file_vm_ops = { |
1219 | .fault = xfs_filemap_fault, | |
a2d58167 | 1220 | .huge_fault = xfs_filemap_huge_fault, |
6b698ede DC |
1221 | .map_pages = filemap_map_pages, |
1222 | .page_mkwrite = xfs_filemap_page_mkwrite, | |
3af49285 | 1223 | .pfn_mkwrite = xfs_filemap_pfn_mkwrite, |
6b698ede DC |
1224 | }; |
1225 | ||
1226 | STATIC int | |
1227 | xfs_file_mmap( | |
1228 | struct file *filp, | |
1229 | struct vm_area_struct *vma) | |
1230 | { | |
b21fec41 PG |
1231 | struct dax_device *dax_dev; |
1232 | ||
1233 | dax_dev = xfs_find_daxdev_for_inode(file_inode(filp)); | |
a39e596b | 1234 | /* |
b21fec41 PG |
1235 | * We don't support synchronous mappings for non-DAX files and |
1236 | * for DAX files if underneath dax_device is not synchronous. | |
a39e596b | 1237 | */ |
b21fec41 | 1238 | if (!daxdev_mapping_supported(vma, dax_dev)) |
a39e596b CH |
1239 | return -EOPNOTSUPP; |
1240 | ||
6b698ede DC |
1241 | file_accessed(filp); |
1242 | vma->vm_ops = &xfs_file_vm_ops; | |
1243 | if (IS_DAX(file_inode(filp))) | |
e1fb4a08 | 1244 | vma->vm_flags |= VM_HUGEPAGE; |
6b698ede | 1245 | return 0; |
075a924d DC |
1246 | } |
1247 | ||
4b6f5d20 | 1248 | const struct file_operations xfs_file_operations = { |
3fe3e6b1 | 1249 | .llseek = xfs_file_llseek, |
b4f5d2c6 | 1250 | .read_iter = xfs_file_read_iter, |
bf97f3bc | 1251 | .write_iter = xfs_file_write_iter, |
82c156f8 | 1252 | .splice_read = generic_file_splice_read, |
8d020765 | 1253 | .splice_write = iter_file_splice_write, |
81214bab | 1254 | .iopoll = iomap_dio_iopoll, |
3562fd45 | 1255 | .unlocked_ioctl = xfs_file_ioctl, |
1da177e4 | 1256 | #ifdef CONFIG_COMPAT |
3562fd45 | 1257 | .compat_ioctl = xfs_file_compat_ioctl, |
1da177e4 | 1258 | #endif |
3562fd45 | 1259 | .mmap = xfs_file_mmap, |
a39e596b | 1260 | .mmap_supported_flags = MAP_SYNC, |
3562fd45 NS |
1261 | .open = xfs_file_open, |
1262 | .release = xfs_file_release, | |
1263 | .fsync = xfs_file_fsync, | |
dbe6ec81 | 1264 | .get_unmapped_area = thp_get_unmapped_area, |
2fe17c10 | 1265 | .fallocate = xfs_file_fallocate, |
40144e49 | 1266 | .fadvise = xfs_file_fadvise, |
2e5dfc99 | 1267 | .remap_file_range = xfs_file_remap_range, |
1da177e4 LT |
1268 | }; |
1269 | ||
4b6f5d20 | 1270 | const struct file_operations xfs_dir_file_operations = { |
f999a5bf | 1271 | .open = xfs_dir_open, |
1da177e4 | 1272 | .read = generic_read_dir, |
3b0a3c1a | 1273 | .iterate_shared = xfs_file_readdir, |
59af1584 | 1274 | .llseek = generic_file_llseek, |
3562fd45 | 1275 | .unlocked_ioctl = xfs_file_ioctl, |
d3870398 | 1276 | #ifdef CONFIG_COMPAT |
3562fd45 | 1277 | .compat_ioctl = xfs_file_compat_ioctl, |
d3870398 | 1278 | #endif |
1da2f2db | 1279 | .fsync = xfs_dir_fsync, |
1da177e4 | 1280 | }; |