Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
7b718769 NS |
2 | * Copyright (c) 2000-2005 Silicon Graphics, Inc. |
3 | * All Rights Reserved. | |
1da177e4 | 4 | * |
7b718769 NS |
5 | * This program is free software; you can redistribute it and/or |
6 | * modify it under the terms of the GNU General Public License as | |
1da177e4 LT |
7 | * published by the Free Software Foundation. |
8 | * | |
7b718769 NS |
9 | * This program is distributed in the hope that it would be useful, |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
1da177e4 | 13 | * |
7b718769 NS |
14 | * You should have received a copy of the GNU General Public License |
15 | * along with this program; if not, write the Free Software Foundation, | |
16 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
1da177e4 | 17 | */ |
1da177e4 | 18 | #include "xfs.h" |
dda35b8f | 19 | #include "xfs_fs.h" |
70a9883c | 20 | #include "xfs_shared.h" |
a4fbe6ab | 21 | #include "xfs_format.h" |
239880ef DC |
22 | #include "xfs_log_format.h" |
23 | #include "xfs_trans_resv.h" | |
1da177e4 | 24 | #include "xfs_mount.h" |
57062787 DC |
25 | #include "xfs_da_format.h" |
26 | #include "xfs_da_btree.h" | |
1da177e4 | 27 | #include "xfs_inode.h" |
239880ef | 28 | #include "xfs_trans.h" |
fd3200be | 29 | #include "xfs_inode_item.h" |
dda35b8f | 30 | #include "xfs_bmap.h" |
c24b5dfa | 31 | #include "xfs_bmap_util.h" |
1da177e4 | 32 | #include "xfs_error.h" |
2b9ab5ab | 33 | #include "xfs_dir2.h" |
c24b5dfa | 34 | #include "xfs_dir2_priv.h" |
ddcd856d | 35 | #include "xfs_ioctl.h" |
dda35b8f | 36 | #include "xfs_trace.h" |
239880ef | 37 | #include "xfs_log.h" |
dc06f398 | 38 | #include "xfs_icache.h" |
781355c6 | 39 | #include "xfs_pnfs.h" |
68a9f5e7 | 40 | #include "xfs_iomap.h" |
0613f16c | 41 | #include "xfs_reflink.h" |
1da177e4 LT |
42 | |
43 | #include <linux/dcache.h> | |
2fe17c10 | 44 | #include <linux/falloc.h> |
d126d43f | 45 | #include <linux/pagevec.h> |
66114cad | 46 | #include <linux/backing-dev.h> |
1da177e4 | 47 | |
f0f37e2f | 48 | static const struct vm_operations_struct xfs_file_vm_ops; |
1da177e4 | 49 | |
487f84f3 DC |
50 | /* |
51 | * Locking primitives for read and write IO paths to ensure we consistently use | |
52 | * and order the inode->i_mutex, ip->i_lock and ip->i_iolock. | |
53 | */ | |
54 | static inline void | |
55 | xfs_rw_ilock( | |
56 | struct xfs_inode *ip, | |
57 | int type) | |
58 | { | |
59 | if (type & XFS_IOLOCK_EXCL) | |
5955102c | 60 | inode_lock(VFS_I(ip)); |
487f84f3 DC |
61 | xfs_ilock(ip, type); |
62 | } | |
63 | ||
64 | static inline void | |
65 | xfs_rw_iunlock( | |
66 | struct xfs_inode *ip, | |
67 | int type) | |
68 | { | |
69 | xfs_iunlock(ip, type); | |
70 | if (type & XFS_IOLOCK_EXCL) | |
5955102c | 71 | inode_unlock(VFS_I(ip)); |
487f84f3 DC |
72 | } |
73 | ||
74 | static inline void | |
75 | xfs_rw_ilock_demote( | |
76 | struct xfs_inode *ip, | |
77 | int type) | |
78 | { | |
79 | xfs_ilock_demote(ip, type); | |
80 | if (type & XFS_IOLOCK_EXCL) | |
5955102c | 81 | inode_unlock(VFS_I(ip)); |
487f84f3 DC |
82 | } |
83 | ||
dda35b8f | 84 | /* |
68a9f5e7 CH |
85 | * Clear the specified ranges to zero through either the pagecache or DAX. |
86 | * Holes and unwritten extents will be left as-is as they already are zeroed. | |
dda35b8f | 87 | */ |
ef9d8733 | 88 | int |
7bb41db3 | 89 | xfs_zero_range( |
68a9f5e7 | 90 | struct xfs_inode *ip, |
7bb41db3 CH |
91 | xfs_off_t pos, |
92 | xfs_off_t count, | |
93 | bool *did_zero) | |
dda35b8f | 94 | { |
459f0fbc | 95 | return iomap_zero_range(VFS_I(ip), pos, count, NULL, &xfs_iomap_ops); |
dda35b8f CH |
96 | } |
97 | ||
8add71ca CH |
98 | int |
99 | xfs_update_prealloc_flags( | |
100 | struct xfs_inode *ip, | |
101 | enum xfs_prealloc_flags flags) | |
102 | { | |
103 | struct xfs_trans *tp; | |
104 | int error; | |
105 | ||
253f4911 CH |
106 | error = xfs_trans_alloc(ip->i_mount, &M_RES(ip->i_mount)->tr_writeid, |
107 | 0, 0, 0, &tp); | |
108 | if (error) | |
8add71ca | 109 | return error; |
8add71ca CH |
110 | |
111 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
112 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
113 | ||
114 | if (!(flags & XFS_PREALLOC_INVISIBLE)) { | |
c19b3b05 DC |
115 | VFS_I(ip)->i_mode &= ~S_ISUID; |
116 | if (VFS_I(ip)->i_mode & S_IXGRP) | |
117 | VFS_I(ip)->i_mode &= ~S_ISGID; | |
8add71ca CH |
118 | xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); |
119 | } | |
120 | ||
121 | if (flags & XFS_PREALLOC_SET) | |
122 | ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC; | |
123 | if (flags & XFS_PREALLOC_CLEAR) | |
124 | ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC; | |
125 | ||
126 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
127 | if (flags & XFS_PREALLOC_SYNC) | |
128 | xfs_trans_set_sync(tp); | |
70393313 | 129 | return xfs_trans_commit(tp); |
8add71ca CH |
130 | } |
131 | ||
1da2f2db CH |
132 | /* |
133 | * Fsync operations on directories are much simpler than on regular files, | |
134 | * as there is no file data to flush, and thus also no need for explicit | |
135 | * cache flush operations, and there are no non-transaction metadata updates | |
136 | * on directories either. | |
137 | */ | |
138 | STATIC int | |
139 | xfs_dir_fsync( | |
140 | struct file *file, | |
141 | loff_t start, | |
142 | loff_t end, | |
143 | int datasync) | |
144 | { | |
145 | struct xfs_inode *ip = XFS_I(file->f_mapping->host); | |
146 | struct xfs_mount *mp = ip->i_mount; | |
147 | xfs_lsn_t lsn = 0; | |
148 | ||
149 | trace_xfs_dir_fsync(ip); | |
150 | ||
151 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
152 | if (xfs_ipincount(ip)) | |
153 | lsn = ip->i_itemp->ili_last_lsn; | |
154 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
155 | ||
156 | if (!lsn) | |
157 | return 0; | |
2451337d | 158 | return _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, NULL); |
1da2f2db CH |
159 | } |
160 | ||
fd3200be CH |
161 | STATIC int |
162 | xfs_file_fsync( | |
163 | struct file *file, | |
02c24a82 JB |
164 | loff_t start, |
165 | loff_t end, | |
fd3200be CH |
166 | int datasync) |
167 | { | |
7ea80859 CH |
168 | struct inode *inode = file->f_mapping->host; |
169 | struct xfs_inode *ip = XFS_I(inode); | |
a27a263b | 170 | struct xfs_mount *mp = ip->i_mount; |
fd3200be CH |
171 | int error = 0; |
172 | int log_flushed = 0; | |
b1037058 | 173 | xfs_lsn_t lsn = 0; |
fd3200be | 174 | |
cca28fb8 | 175 | trace_xfs_file_fsync(ip); |
fd3200be | 176 | |
02c24a82 JB |
177 | error = filemap_write_and_wait_range(inode->i_mapping, start, end); |
178 | if (error) | |
179 | return error; | |
180 | ||
a27a263b | 181 | if (XFS_FORCED_SHUTDOWN(mp)) |
b474c7ae | 182 | return -EIO; |
fd3200be CH |
183 | |
184 | xfs_iflags_clear(ip, XFS_ITRUNCATED); | |
185 | ||
a27a263b CH |
186 | if (mp->m_flags & XFS_MOUNT_BARRIER) { |
187 | /* | |
188 | * If we have an RT and/or log subvolume we need to make sure | |
189 | * to flush the write cache the device used for file data | |
190 | * first. This is to ensure newly written file data make | |
191 | * it to disk before logging the new inode size in case of | |
192 | * an extending write. | |
193 | */ | |
194 | if (XFS_IS_REALTIME_INODE(ip)) | |
195 | xfs_blkdev_issue_flush(mp->m_rtdev_targp); | |
196 | else if (mp->m_logdev_targp != mp->m_ddev_targp) | |
197 | xfs_blkdev_issue_flush(mp->m_ddev_targp); | |
198 | } | |
199 | ||
fd3200be | 200 | /* |
fc0561ce DC |
201 | * All metadata updates are logged, which means that we just have to |
202 | * flush the log up to the latest LSN that touched the inode. If we have | |
203 | * concurrent fsync/fdatasync() calls, we need them to all block on the | |
204 | * log force before we clear the ili_fsync_fields field. This ensures | |
205 | * that we don't get a racing sync operation that does not wait for the | |
206 | * metadata to hit the journal before returning. If we race with | |
207 | * clearing the ili_fsync_fields, then all that will happen is the log | |
208 | * force will do nothing as the lsn will already be on disk. We can't | |
209 | * race with setting ili_fsync_fields because that is done under | |
210 | * XFS_ILOCK_EXCL, and that can't happen because we hold the lock shared | |
211 | * until after the ili_fsync_fields is cleared. | |
fd3200be CH |
212 | */ |
213 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
8f639dde CH |
214 | if (xfs_ipincount(ip)) { |
215 | if (!datasync || | |
fc0561ce | 216 | (ip->i_itemp->ili_fsync_fields & ~XFS_ILOG_TIMESTAMP)) |
8f639dde CH |
217 | lsn = ip->i_itemp->ili_last_lsn; |
218 | } | |
fd3200be | 219 | |
fc0561ce | 220 | if (lsn) { |
b1037058 | 221 | error = _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, &log_flushed); |
fc0561ce DC |
222 | ip->i_itemp->ili_fsync_fields = 0; |
223 | } | |
224 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
b1037058 | 225 | |
a27a263b CH |
226 | /* |
227 | * If we only have a single device, and the log force about was | |
228 | * a no-op we might have to flush the data device cache here. | |
229 | * This can only happen for fdatasync/O_DSYNC if we were overwriting | |
230 | * an already allocated file and thus do not have any metadata to | |
231 | * commit. | |
232 | */ | |
233 | if ((mp->m_flags & XFS_MOUNT_BARRIER) && | |
234 | mp->m_logdev_targp == mp->m_ddev_targp && | |
235 | !XFS_IS_REALTIME_INODE(ip) && | |
236 | !log_flushed) | |
237 | xfs_blkdev_issue_flush(mp->m_ddev_targp); | |
fd3200be | 238 | |
2451337d | 239 | return error; |
fd3200be CH |
240 | } |
241 | ||
00258e36 | 242 | STATIC ssize_t |
bbc5a740 | 243 | xfs_file_dio_aio_read( |
dda35b8f | 244 | struct kiocb *iocb, |
b4f5d2c6 | 245 | struct iov_iter *to) |
dda35b8f | 246 | { |
bbc5a740 CH |
247 | struct address_space *mapping = iocb->ki_filp->f_mapping; |
248 | struct inode *inode = mapping->host; | |
00258e36 | 249 | struct xfs_inode *ip = XFS_I(inode); |
f1285ff0 | 250 | loff_t isize = i_size_read(inode); |
bbc5a740 | 251 | size_t count = iov_iter_count(to); |
f1285ff0 | 252 | struct iov_iter data; |
bbc5a740 | 253 | struct xfs_buftarg *target; |
dda35b8f | 254 | ssize_t ret = 0; |
dda35b8f | 255 | |
bbc5a740 | 256 | trace_xfs_file_direct_read(ip, count, iocb->ki_pos); |
dda35b8f | 257 | |
f1285ff0 CH |
258 | if (!count) |
259 | return 0; /* skip atime */ | |
dda35b8f | 260 | |
bbc5a740 CH |
261 | if (XFS_IS_REALTIME_INODE(ip)) |
262 | target = ip->i_mount->m_rtdev_targp; | |
263 | else | |
264 | target = ip->i_mount->m_ddev_targp; | |
dda35b8f | 265 | |
16d4d435 CH |
266 | /* DIO must be aligned to device logical sector size */ |
267 | if ((iocb->ki_pos | count) & target->bt_logical_sectormask) { | |
268 | if (iocb->ki_pos == isize) | |
269 | return 0; | |
270 | return -EINVAL; | |
dda35b8f | 271 | } |
dda35b8f | 272 | |
a447d7cd CH |
273 | file_accessed(iocb->ki_filp); |
274 | ||
0c38a251 | 275 | /* |
3d751af2 BF |
276 | * Locking is a bit tricky here. If we take an exclusive lock for direct |
277 | * IO, we effectively serialise all new concurrent read IO to this file | |
278 | * and block it behind IO that is currently in progress because IO in | |
279 | * progress holds the IO lock shared. We only need to hold the lock | |
280 | * exclusive to blow away the page cache, so only take lock exclusively | |
281 | * if the page cache needs invalidation. This allows the normal direct | |
282 | * IO case of no page cache pages to proceeed concurrently without | |
283 | * serialisation. | |
0c38a251 DC |
284 | */ |
285 | xfs_rw_ilock(ip, XFS_IOLOCK_SHARED); | |
bbc5a740 | 286 | if (mapping->nrpages) { |
0c38a251 | 287 | xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED); |
487f84f3 DC |
288 | xfs_rw_ilock(ip, XFS_IOLOCK_EXCL); |
289 | ||
3d751af2 BF |
290 | /* |
291 | * The generic dio code only flushes the range of the particular | |
292 | * I/O. Because we take an exclusive lock here, this whole | |
293 | * sequence is considerably more expensive for us. This has a | |
294 | * noticeable performance impact for any file with cached pages, | |
295 | * even when outside of the range of the particular I/O. | |
296 | * | |
297 | * Hence, amortize the cost of the lock against a full file | |
298 | * flush and reduce the chances of repeated iolock cycles going | |
299 | * forward. | |
300 | */ | |
bbc5a740 CH |
301 | if (mapping->nrpages) { |
302 | ret = filemap_write_and_wait(mapping); | |
487f84f3 DC |
303 | if (ret) { |
304 | xfs_rw_iunlock(ip, XFS_IOLOCK_EXCL); | |
305 | return ret; | |
306 | } | |
85e584da CM |
307 | |
308 | /* | |
309 | * Invalidate whole pages. This can return an error if | |
310 | * we fail to invalidate a page, but this should never | |
311 | * happen on XFS. Warn if it does fail. | |
312 | */ | |
bbc5a740 | 313 | ret = invalidate_inode_pages2(mapping); |
85e584da CM |
314 | WARN_ON_ONCE(ret); |
315 | ret = 0; | |
00258e36 | 316 | } |
487f84f3 | 317 | xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL); |
0c38a251 | 318 | } |
dda35b8f | 319 | |
f1285ff0 | 320 | data = *to; |
16d4d435 CH |
321 | ret = __blockdev_direct_IO(iocb, inode, target->bt_bdev, &data, |
322 | xfs_get_blocks_direct, NULL, NULL, 0); | |
c3a69024 | 323 | if (ret >= 0) { |
16d4d435 CH |
324 | iocb->ki_pos += ret; |
325 | iov_iter_advance(to, ret); | |
fa8d972d | 326 | } |
16d4d435 | 327 | xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED); |
dda35b8f | 328 | |
16d4d435 CH |
329 | return ret; |
330 | } | |
331 | ||
f021bd07 | 332 | static noinline ssize_t |
16d4d435 CH |
333 | xfs_file_dax_read( |
334 | struct kiocb *iocb, | |
335 | struct iov_iter *to) | |
336 | { | |
6c31f495 | 337 | struct xfs_inode *ip = XFS_I(iocb->ki_filp->f_mapping->host); |
16d4d435 CH |
338 | size_t count = iov_iter_count(to); |
339 | ssize_t ret = 0; | |
340 | ||
341 | trace_xfs_file_dax_read(ip, count, iocb->ki_pos); | |
342 | ||
343 | if (!count) | |
344 | return 0; /* skip atime */ | |
345 | ||
346 | xfs_rw_ilock(ip, XFS_IOLOCK_SHARED); | |
6c31f495 | 347 | ret = iomap_dax_rw(iocb, to, &xfs_iomap_ops); |
bbc5a740 CH |
348 | xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED); |
349 | ||
f1285ff0 | 350 | file_accessed(iocb->ki_filp); |
bbc5a740 CH |
351 | return ret; |
352 | } | |
353 | ||
354 | STATIC ssize_t | |
355 | xfs_file_buffered_aio_read( | |
356 | struct kiocb *iocb, | |
357 | struct iov_iter *to) | |
358 | { | |
359 | struct xfs_inode *ip = XFS_I(file_inode(iocb->ki_filp)); | |
360 | ssize_t ret; | |
361 | ||
362 | trace_xfs_file_buffered_read(ip, iov_iter_count(to), iocb->ki_pos); | |
dda35b8f | 363 | |
bbc5a740 | 364 | xfs_rw_ilock(ip, XFS_IOLOCK_SHARED); |
b4f5d2c6 | 365 | ret = generic_file_read_iter(iocb, to); |
bbc5a740 CH |
366 | xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED); |
367 | ||
368 | return ret; | |
369 | } | |
370 | ||
371 | STATIC ssize_t | |
372 | xfs_file_read_iter( | |
373 | struct kiocb *iocb, | |
374 | struct iov_iter *to) | |
375 | { | |
16d4d435 CH |
376 | struct inode *inode = file_inode(iocb->ki_filp); |
377 | struct xfs_mount *mp = XFS_I(inode)->i_mount; | |
bbc5a740 CH |
378 | ssize_t ret = 0; |
379 | ||
380 | XFS_STATS_INC(mp, xs_read_calls); | |
381 | ||
382 | if (XFS_FORCED_SHUTDOWN(mp)) | |
383 | return -EIO; | |
384 | ||
16d4d435 CH |
385 | if (IS_DAX(inode)) |
386 | ret = xfs_file_dax_read(iocb, to); | |
387 | else if (iocb->ki_flags & IOCB_DIRECT) | |
bbc5a740 | 388 | ret = xfs_file_dio_aio_read(iocb, to); |
3176c3e0 | 389 | else |
bbc5a740 | 390 | ret = xfs_file_buffered_aio_read(iocb, to); |
dda35b8f | 391 | |
dda35b8f | 392 | if (ret > 0) |
ff6d6af2 | 393 | XFS_STATS_ADD(mp, xs_read_bytes, ret); |
dda35b8f CH |
394 | return ret; |
395 | } | |
396 | ||
dda35b8f | 397 | /* |
193aec10 CH |
398 | * Zero any on disk space between the current EOF and the new, larger EOF. |
399 | * | |
400 | * This handles the normal case of zeroing the remainder of the last block in | |
401 | * the file and the unusual case of zeroing blocks out beyond the size of the | |
402 | * file. This second case only happens with fixed size extents and when the | |
403 | * system crashes before the inode size was updated but after blocks were | |
404 | * allocated. | |
405 | * | |
406 | * Expects the iolock to be held exclusive, and will take the ilock internally. | |
dda35b8f | 407 | */ |
dda35b8f CH |
408 | int /* error (positive) */ |
409 | xfs_zero_eof( | |
193aec10 CH |
410 | struct xfs_inode *ip, |
411 | xfs_off_t offset, /* starting I/O offset */ | |
5885ebda DC |
412 | xfs_fsize_t isize, /* current inode size */ |
413 | bool *did_zeroing) | |
dda35b8f | 414 | { |
193aec10 | 415 | ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL)); |
dda35b8f CH |
416 | ASSERT(offset > isize); |
417 | ||
0a50f162 | 418 | trace_xfs_zero_eof(ip, isize, offset - isize); |
570b6211 | 419 | return xfs_zero_range(ip, isize, offset - isize, did_zeroing); |
dda35b8f CH |
420 | } |
421 | ||
4d8d1581 DC |
422 | /* |
423 | * Common pre-write limit and setup checks. | |
424 | * | |
5bf1f262 CH |
425 | * Called with the iolocked held either shared and exclusive according to |
426 | * @iolock, and returns with it held. Might upgrade the iolock to exclusive | |
427 | * if called for a direct write beyond i_size. | |
4d8d1581 DC |
428 | */ |
429 | STATIC ssize_t | |
430 | xfs_file_aio_write_checks( | |
99733fa3 AV |
431 | struct kiocb *iocb, |
432 | struct iov_iter *from, | |
4d8d1581 DC |
433 | int *iolock) |
434 | { | |
99733fa3 | 435 | struct file *file = iocb->ki_filp; |
4d8d1581 DC |
436 | struct inode *inode = file->f_mapping->host; |
437 | struct xfs_inode *ip = XFS_I(inode); | |
3309dd04 | 438 | ssize_t error = 0; |
99733fa3 | 439 | size_t count = iov_iter_count(from); |
3136e8bb | 440 | bool drained_dio = false; |
4d8d1581 | 441 | |
7271d243 | 442 | restart: |
3309dd04 AV |
443 | error = generic_write_checks(iocb, from); |
444 | if (error <= 0) | |
4d8d1581 | 445 | return error; |
4d8d1581 | 446 | |
21c3ea18 | 447 | error = xfs_break_layouts(inode, iolock, true); |
781355c6 CH |
448 | if (error) |
449 | return error; | |
450 | ||
a6de82ca JK |
451 | /* For changing security info in file_remove_privs() we need i_mutex */ |
452 | if (*iolock == XFS_IOLOCK_SHARED && !IS_NOSEC(inode)) { | |
453 | xfs_rw_iunlock(ip, *iolock); | |
454 | *iolock = XFS_IOLOCK_EXCL; | |
455 | xfs_rw_ilock(ip, *iolock); | |
456 | goto restart; | |
457 | } | |
4d8d1581 DC |
458 | /* |
459 | * If the offset is beyond the size of the file, we need to zero any | |
460 | * blocks that fall between the existing EOF and the start of this | |
2813d682 | 461 | * write. If zeroing is needed and we are currently holding the |
467f7899 CH |
462 | * iolock shared, we need to update it to exclusive which implies |
463 | * having to redo all checks before. | |
b9d59846 DC |
464 | * |
465 | * We need to serialise against EOF updates that occur in IO | |
466 | * completions here. We want to make sure that nobody is changing the | |
467 | * size while we do this check until we have placed an IO barrier (i.e. | |
468 | * hold the XFS_IOLOCK_EXCL) that prevents new IO from being dispatched. | |
469 | * The spinlock effectively forms a memory barrier once we have the | |
470 | * XFS_IOLOCK_EXCL so we are guaranteed to see the latest EOF value | |
471 | * and hence be able to correctly determine if we need to run zeroing. | |
4d8d1581 | 472 | */ |
b9d59846 | 473 | spin_lock(&ip->i_flags_lock); |
99733fa3 | 474 | if (iocb->ki_pos > i_size_read(inode)) { |
5885ebda DC |
475 | bool zero = false; |
476 | ||
b9d59846 | 477 | spin_unlock(&ip->i_flags_lock); |
3136e8bb BF |
478 | if (!drained_dio) { |
479 | if (*iolock == XFS_IOLOCK_SHARED) { | |
480 | xfs_rw_iunlock(ip, *iolock); | |
481 | *iolock = XFS_IOLOCK_EXCL; | |
482 | xfs_rw_ilock(ip, *iolock); | |
483 | iov_iter_reexpand(from, count); | |
484 | } | |
40c63fbc DC |
485 | /* |
486 | * We now have an IO submission barrier in place, but | |
487 | * AIO can do EOF updates during IO completion and hence | |
488 | * we now need to wait for all of them to drain. Non-AIO | |
489 | * DIO will have drained before we are given the | |
490 | * XFS_IOLOCK_EXCL, and so for most cases this wait is a | |
491 | * no-op. | |
492 | */ | |
493 | inode_dio_wait(inode); | |
3136e8bb | 494 | drained_dio = true; |
7271d243 DC |
495 | goto restart; |
496 | } | |
99733fa3 | 497 | error = xfs_zero_eof(ip, iocb->ki_pos, i_size_read(inode), &zero); |
467f7899 CH |
498 | if (error) |
499 | return error; | |
b9d59846 DC |
500 | } else |
501 | spin_unlock(&ip->i_flags_lock); | |
4d8d1581 | 502 | |
8a9c9980 CH |
503 | /* |
504 | * Updating the timestamps will grab the ilock again from | |
505 | * xfs_fs_dirty_inode, so we have to call it after dropping the | |
506 | * lock above. Eventually we should look into a way to avoid | |
507 | * the pointless lock roundtrip. | |
508 | */ | |
c3b2da31 JB |
509 | if (likely(!(file->f_mode & FMODE_NOCMTIME))) { |
510 | error = file_update_time(file); | |
511 | if (error) | |
512 | return error; | |
513 | } | |
8a9c9980 | 514 | |
4d8d1581 DC |
515 | /* |
516 | * If we're writing the file then make sure to clear the setuid and | |
517 | * setgid bits if the process is not being run by root. This keeps | |
518 | * people from modifying setuid and setgid binaries. | |
519 | */ | |
a6de82ca JK |
520 | if (!IS_NOSEC(inode)) |
521 | return file_remove_privs(file); | |
522 | return 0; | |
4d8d1581 DC |
523 | } |
524 | ||
f0d26e86 DC |
525 | /* |
526 | * xfs_file_dio_aio_write - handle direct IO writes | |
527 | * | |
528 | * Lock the inode appropriately to prepare for and issue a direct IO write. | |
eda77982 | 529 | * By separating it from the buffered write path we remove all the tricky to |
f0d26e86 DC |
530 | * follow locking changes and looping. |
531 | * | |
eda77982 DC |
532 | * If there are cached pages or we're extending the file, we need IOLOCK_EXCL |
533 | * until we're sure the bytes at the new EOF have been zeroed and/or the cached | |
534 | * pages are flushed out. | |
535 | * | |
536 | * In most cases the direct IO writes will be done holding IOLOCK_SHARED | |
537 | * allowing them to be done in parallel with reads and other direct IO writes. | |
538 | * However, if the IO is not aligned to filesystem blocks, the direct IO layer | |
539 | * needs to do sub-block zeroing and that requires serialisation against other | |
540 | * direct IOs to the same block. In this case we need to serialise the | |
541 | * submission of the unaligned IOs so that we don't get racing block zeroing in | |
542 | * the dio layer. To avoid the problem with aio, we also need to wait for | |
543 | * outstanding IOs to complete so that unwritten extent conversion is completed | |
544 | * before we try to map the overlapping block. This is currently implemented by | |
4a06fd26 | 545 | * hitting it with a big hammer (i.e. inode_dio_wait()). |
eda77982 | 546 | * |
f0d26e86 DC |
547 | * Returns with locks held indicated by @iolock and errors indicated by |
548 | * negative return values. | |
549 | */ | |
550 | STATIC ssize_t | |
551 | xfs_file_dio_aio_write( | |
552 | struct kiocb *iocb, | |
b3188919 | 553 | struct iov_iter *from) |
f0d26e86 DC |
554 | { |
555 | struct file *file = iocb->ki_filp; | |
556 | struct address_space *mapping = file->f_mapping; | |
557 | struct inode *inode = mapping->host; | |
558 | struct xfs_inode *ip = XFS_I(inode); | |
559 | struct xfs_mount *mp = ip->i_mount; | |
560 | ssize_t ret = 0; | |
eda77982 | 561 | int unaligned_io = 0; |
d0606464 | 562 | int iolock; |
b3188919 | 563 | size_t count = iov_iter_count(from); |
0cefb29e DC |
564 | loff_t end; |
565 | struct iov_iter data; | |
f0d26e86 DC |
566 | struct xfs_buftarg *target = XFS_IS_REALTIME_INODE(ip) ? |
567 | mp->m_rtdev_targp : mp->m_ddev_targp; | |
568 | ||
7c71ee78 | 569 | /* DIO must be aligned to device logical sector size */ |
16d4d435 | 570 | if ((iocb->ki_pos | count) & target->bt_logical_sectormask) |
b474c7ae | 571 | return -EINVAL; |
f0d26e86 | 572 | |
7c71ee78 | 573 | /* "unaligned" here means not aligned to a filesystem block */ |
13712713 CH |
574 | if ((iocb->ki_pos & mp->m_blockmask) || |
575 | ((iocb->ki_pos + count) & mp->m_blockmask)) | |
eda77982 DC |
576 | unaligned_io = 1; |
577 | ||
7271d243 DC |
578 | /* |
579 | * We don't need to take an exclusive lock unless there page cache needs | |
580 | * to be invalidated or unaligned IO is being executed. We don't need to | |
581 | * consider the EOF extension case here because | |
582 | * xfs_file_aio_write_checks() will relock the inode as necessary for | |
583 | * EOF zeroing cases and fill out the new inode size as appropriate. | |
584 | */ | |
585 | if (unaligned_io || mapping->nrpages) | |
d0606464 | 586 | iolock = XFS_IOLOCK_EXCL; |
f0d26e86 | 587 | else |
d0606464 CH |
588 | iolock = XFS_IOLOCK_SHARED; |
589 | xfs_rw_ilock(ip, iolock); | |
c58cb165 CH |
590 | |
591 | /* | |
592 | * Recheck if there are cached pages that need invalidate after we got | |
593 | * the iolock to protect against other threads adding new pages while | |
594 | * we were waiting for the iolock. | |
595 | */ | |
d0606464 CH |
596 | if (mapping->nrpages && iolock == XFS_IOLOCK_SHARED) { |
597 | xfs_rw_iunlock(ip, iolock); | |
598 | iolock = XFS_IOLOCK_EXCL; | |
599 | xfs_rw_ilock(ip, iolock); | |
c58cb165 | 600 | } |
f0d26e86 | 601 | |
99733fa3 | 602 | ret = xfs_file_aio_write_checks(iocb, from, &iolock); |
4d8d1581 | 603 | if (ret) |
d0606464 | 604 | goto out; |
99733fa3 | 605 | count = iov_iter_count(from); |
13712713 | 606 | end = iocb->ki_pos + count - 1; |
f0d26e86 | 607 | |
3d751af2 | 608 | /* |
bbc5a740 | 609 | * See xfs_file_dio_aio_read() for why we do a full-file flush here. |
3d751af2 | 610 | */ |
f0d26e86 | 611 | if (mapping->nrpages) { |
3d751af2 | 612 | ret = filemap_write_and_wait(VFS_I(ip)->i_mapping); |
f0d26e86 | 613 | if (ret) |
d0606464 | 614 | goto out; |
834ffca6 | 615 | /* |
3d751af2 BF |
616 | * Invalidate whole pages. This can return an error if we fail |
617 | * to invalidate a page, but this should never happen on XFS. | |
618 | * Warn if it does fail. | |
834ffca6 | 619 | */ |
3d751af2 | 620 | ret = invalidate_inode_pages2(VFS_I(ip)->i_mapping); |
834ffca6 DC |
621 | WARN_ON_ONCE(ret); |
622 | ret = 0; | |
f0d26e86 DC |
623 | } |
624 | ||
eda77982 DC |
625 | /* |
626 | * If we are doing unaligned IO, wait for all other IO to drain, | |
627 | * otherwise demote the lock if we had to flush cached pages | |
628 | */ | |
629 | if (unaligned_io) | |
4a06fd26 | 630 | inode_dio_wait(inode); |
d0606464 | 631 | else if (iolock == XFS_IOLOCK_EXCL) { |
f0d26e86 | 632 | xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL); |
d0606464 | 633 | iolock = XFS_IOLOCK_SHARED; |
f0d26e86 DC |
634 | } |
635 | ||
3176c3e0 | 636 | trace_xfs_file_direct_write(ip, count, iocb->ki_pos); |
f0d26e86 | 637 | |
0613f16c DW |
638 | /* If this is a block-aligned directio CoW, remap immediately. */ |
639 | if (xfs_is_reflink_inode(ip) && !unaligned_io) { | |
640 | ret = xfs_reflink_allocate_cow_range(ip, iocb->ki_pos, count); | |
641 | if (ret) | |
642 | goto out; | |
643 | } | |
644 | ||
0cefb29e | 645 | data = *from; |
16d4d435 CH |
646 | ret = __blockdev_direct_IO(iocb, inode, target->bt_bdev, &data, |
647 | xfs_get_blocks_direct, xfs_end_io_direct_write, | |
648 | NULL, DIO_ASYNC_EXTEND); | |
0cefb29e DC |
649 | |
650 | /* see generic_file_direct_write() for why this is necessary */ | |
651 | if (mapping->nrpages) { | |
652 | invalidate_inode_pages2_range(mapping, | |
13712713 | 653 | iocb->ki_pos >> PAGE_SHIFT, |
09cbfeaf | 654 | end >> PAGE_SHIFT); |
0cefb29e DC |
655 | } |
656 | ||
657 | if (ret > 0) { | |
13712713 | 658 | iocb->ki_pos += ret; |
0cefb29e | 659 | iov_iter_advance(from, ret); |
0cefb29e | 660 | } |
d0606464 CH |
661 | out: |
662 | xfs_rw_iunlock(ip, iolock); | |
663 | ||
6b698ede | 664 | /* |
16d4d435 CH |
665 | * No fallback to buffered IO on errors for XFS, direct IO will either |
666 | * complete fully or fail. | |
6b698ede | 667 | */ |
16d4d435 CH |
668 | ASSERT(ret < 0 || ret == count); |
669 | return ret; | |
670 | } | |
671 | ||
f021bd07 | 672 | static noinline ssize_t |
16d4d435 CH |
673 | xfs_file_dax_write( |
674 | struct kiocb *iocb, | |
675 | struct iov_iter *from) | |
676 | { | |
6c31f495 | 677 | struct inode *inode = iocb->ki_filp->f_mapping->host; |
16d4d435 | 678 | struct xfs_inode *ip = XFS_I(inode); |
17879e8f | 679 | int iolock = XFS_IOLOCK_EXCL; |
6c31f495 CH |
680 | ssize_t ret, error = 0; |
681 | size_t count; | |
682 | loff_t pos; | |
16d4d435 | 683 | |
16d4d435 | 684 | xfs_rw_ilock(ip, iolock); |
16d4d435 CH |
685 | ret = xfs_file_aio_write_checks(iocb, from, &iolock); |
686 | if (ret) | |
687 | goto out; | |
688 | ||
6c31f495 CH |
689 | pos = iocb->ki_pos; |
690 | count = iov_iter_count(from); | |
8b2180b3 | 691 | |
6c31f495 | 692 | trace_xfs_file_dax_write(ip, count, pos); |
16d4d435 | 693 | |
6c31f495 CH |
694 | ret = iomap_dax_rw(iocb, from, &xfs_iomap_ops); |
695 | if (ret > 0 && iocb->ki_pos > i_size_read(inode)) { | |
696 | i_size_write(inode, iocb->ki_pos); | |
697 | error = xfs_setfilesize(ip, pos, ret); | |
16d4d435 CH |
698 | } |
699 | ||
16d4d435 CH |
700 | out: |
701 | xfs_rw_iunlock(ip, iolock); | |
6c31f495 | 702 | return error ? error : ret; |
f0d26e86 DC |
703 | } |
704 | ||
00258e36 | 705 | STATIC ssize_t |
637bbc75 | 706 | xfs_file_buffered_aio_write( |
dda35b8f | 707 | struct kiocb *iocb, |
b3188919 | 708 | struct iov_iter *from) |
dda35b8f CH |
709 | { |
710 | struct file *file = iocb->ki_filp; | |
711 | struct address_space *mapping = file->f_mapping; | |
712 | struct inode *inode = mapping->host; | |
00258e36 | 713 | struct xfs_inode *ip = XFS_I(inode); |
637bbc75 DC |
714 | ssize_t ret; |
715 | int enospc = 0; | |
d0606464 | 716 | int iolock = XFS_IOLOCK_EXCL; |
dda35b8f | 717 | |
d0606464 | 718 | xfs_rw_ilock(ip, iolock); |
dda35b8f | 719 | |
99733fa3 | 720 | ret = xfs_file_aio_write_checks(iocb, from, &iolock); |
4d8d1581 | 721 | if (ret) |
d0606464 | 722 | goto out; |
dda35b8f CH |
723 | |
724 | /* We can write back this queue in page reclaim */ | |
de1414a6 | 725 | current->backing_dev_info = inode_to_bdi(inode); |
dda35b8f | 726 | |
dda35b8f | 727 | write_retry: |
3176c3e0 | 728 | trace_xfs_file_buffered_write(ip, iov_iter_count(from), iocb->ki_pos); |
68a9f5e7 | 729 | ret = iomap_file_buffered_write(iocb, from, &xfs_iomap_ops); |
0a64bc2c | 730 | if (likely(ret >= 0)) |
99733fa3 | 731 | iocb->ki_pos += ret; |
dc06f398 | 732 | |
637bbc75 | 733 | /* |
dc06f398 BF |
734 | * If we hit a space limit, try to free up some lingering preallocated |
735 | * space before returning an error. In the case of ENOSPC, first try to | |
736 | * write back all dirty inodes to free up some of the excess reserved | |
737 | * metadata space. This reduces the chances that the eofblocks scan | |
738 | * waits on dirty mappings. Since xfs_flush_inodes() is serialized, this | |
739 | * also behaves as a filter to prevent too many eofblocks scans from | |
740 | * running at the same time. | |
637bbc75 | 741 | */ |
dc06f398 BF |
742 | if (ret == -EDQUOT && !enospc) { |
743 | enospc = xfs_inode_free_quota_eofblocks(ip); | |
744 | if (enospc) | |
745 | goto write_retry; | |
83104d44 DW |
746 | enospc = xfs_inode_free_quota_cowblocks(ip); |
747 | if (enospc) | |
748 | goto write_retry; | |
dc06f398 BF |
749 | } else if (ret == -ENOSPC && !enospc) { |
750 | struct xfs_eofblocks eofb = {0}; | |
751 | ||
637bbc75 | 752 | enospc = 1; |
9aa05000 | 753 | xfs_flush_inodes(ip->i_mount); |
dc06f398 BF |
754 | eofb.eof_scan_owner = ip->i_ino; /* for locking */ |
755 | eofb.eof_flags = XFS_EOF_FLAGS_SYNC; | |
756 | xfs_icache_free_eofblocks(ip->i_mount, &eofb); | |
9aa05000 | 757 | goto write_retry; |
dda35b8f | 758 | } |
d0606464 | 759 | |
dda35b8f | 760 | current->backing_dev_info = NULL; |
d0606464 CH |
761 | out: |
762 | xfs_rw_iunlock(ip, iolock); | |
637bbc75 DC |
763 | return ret; |
764 | } | |
765 | ||
766 | STATIC ssize_t | |
bf97f3bc | 767 | xfs_file_write_iter( |
637bbc75 | 768 | struct kiocb *iocb, |
bf97f3bc | 769 | struct iov_iter *from) |
637bbc75 DC |
770 | { |
771 | struct file *file = iocb->ki_filp; | |
772 | struct address_space *mapping = file->f_mapping; | |
773 | struct inode *inode = mapping->host; | |
774 | struct xfs_inode *ip = XFS_I(inode); | |
775 | ssize_t ret; | |
bf97f3bc | 776 | size_t ocount = iov_iter_count(from); |
637bbc75 | 777 | |
ff6d6af2 | 778 | XFS_STATS_INC(ip->i_mount, xs_write_calls); |
637bbc75 | 779 | |
637bbc75 DC |
780 | if (ocount == 0) |
781 | return 0; | |
782 | ||
bf97f3bc AV |
783 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
784 | return -EIO; | |
637bbc75 | 785 | |
16d4d435 CH |
786 | if (IS_DAX(inode)) |
787 | ret = xfs_file_dax_write(iocb, from); | |
0613f16c DW |
788 | else if (iocb->ki_flags & IOCB_DIRECT) { |
789 | /* | |
790 | * Allow a directio write to fall back to a buffered | |
791 | * write *only* in the case that we're doing a reflink | |
792 | * CoW. In all other directio scenarios we do not | |
793 | * allow an operation to fall back to buffered mode. | |
794 | */ | |
bf97f3bc | 795 | ret = xfs_file_dio_aio_write(iocb, from); |
0613f16c DW |
796 | if (ret == -EREMCHG) |
797 | goto buffered; | |
798 | } else { | |
799 | buffered: | |
bf97f3bc | 800 | ret = xfs_file_buffered_aio_write(iocb, from); |
0613f16c | 801 | } |
dda35b8f | 802 | |
d0606464 | 803 | if (ret > 0) { |
ff6d6af2 | 804 | XFS_STATS_ADD(ip->i_mount, xs_write_bytes, ret); |
dda35b8f | 805 | |
d0606464 | 806 | /* Handle various SYNC-type writes */ |
e2592217 | 807 | ret = generic_write_sync(iocb, ret); |
dda35b8f | 808 | } |
a363f0c2 | 809 | return ret; |
dda35b8f CH |
810 | } |
811 | ||
a904b1ca NJ |
812 | #define XFS_FALLOC_FL_SUPPORTED \ |
813 | (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \ | |
814 | FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE | \ | |
98cc2db5 | 815 | FALLOC_FL_INSERT_RANGE | FALLOC_FL_UNSHARE_RANGE) |
a904b1ca | 816 | |
2fe17c10 CH |
817 | STATIC long |
818 | xfs_file_fallocate( | |
83aee9e4 CH |
819 | struct file *file, |
820 | int mode, | |
821 | loff_t offset, | |
822 | loff_t len) | |
2fe17c10 | 823 | { |
83aee9e4 CH |
824 | struct inode *inode = file_inode(file); |
825 | struct xfs_inode *ip = XFS_I(inode); | |
83aee9e4 | 826 | long error; |
8add71ca | 827 | enum xfs_prealloc_flags flags = 0; |
781355c6 | 828 | uint iolock = XFS_IOLOCK_EXCL; |
83aee9e4 | 829 | loff_t new_size = 0; |
a904b1ca | 830 | bool do_file_insert = 0; |
2fe17c10 | 831 | |
83aee9e4 CH |
832 | if (!S_ISREG(inode->i_mode)) |
833 | return -EINVAL; | |
a904b1ca | 834 | if (mode & ~XFS_FALLOC_FL_SUPPORTED) |
2fe17c10 CH |
835 | return -EOPNOTSUPP; |
836 | ||
781355c6 | 837 | xfs_ilock(ip, iolock); |
21c3ea18 | 838 | error = xfs_break_layouts(inode, &iolock, false); |
781355c6 CH |
839 | if (error) |
840 | goto out_unlock; | |
841 | ||
e8e9ad42 DC |
842 | xfs_ilock(ip, XFS_MMAPLOCK_EXCL); |
843 | iolock |= XFS_MMAPLOCK_EXCL; | |
844 | ||
83aee9e4 CH |
845 | if (mode & FALLOC_FL_PUNCH_HOLE) { |
846 | error = xfs_free_file_space(ip, offset, len); | |
847 | if (error) | |
848 | goto out_unlock; | |
e1d8fb88 NJ |
849 | } else if (mode & FALLOC_FL_COLLAPSE_RANGE) { |
850 | unsigned blksize_mask = (1 << inode->i_blkbits) - 1; | |
851 | ||
852 | if (offset & blksize_mask || len & blksize_mask) { | |
2451337d | 853 | error = -EINVAL; |
e1d8fb88 NJ |
854 | goto out_unlock; |
855 | } | |
856 | ||
23fffa92 LC |
857 | /* |
858 | * There is no need to overlap collapse range with EOF, | |
859 | * in which case it is effectively a truncate operation | |
860 | */ | |
861 | if (offset + len >= i_size_read(inode)) { | |
2451337d | 862 | error = -EINVAL; |
23fffa92 LC |
863 | goto out_unlock; |
864 | } | |
865 | ||
e1d8fb88 NJ |
866 | new_size = i_size_read(inode) - len; |
867 | ||
868 | error = xfs_collapse_file_space(ip, offset, len); | |
869 | if (error) | |
870 | goto out_unlock; | |
a904b1ca NJ |
871 | } else if (mode & FALLOC_FL_INSERT_RANGE) { |
872 | unsigned blksize_mask = (1 << inode->i_blkbits) - 1; | |
873 | ||
874 | new_size = i_size_read(inode) + len; | |
875 | if (offset & blksize_mask || len & blksize_mask) { | |
876 | error = -EINVAL; | |
877 | goto out_unlock; | |
878 | } | |
879 | ||
880 | /* check the new inode size does not wrap through zero */ | |
881 | if (new_size > inode->i_sb->s_maxbytes) { | |
882 | error = -EFBIG; | |
883 | goto out_unlock; | |
884 | } | |
885 | ||
886 | /* Offset should be less than i_size */ | |
887 | if (offset >= i_size_read(inode)) { | |
888 | error = -EINVAL; | |
889 | goto out_unlock; | |
890 | } | |
891 | do_file_insert = 1; | |
83aee9e4 | 892 | } else { |
8add71ca CH |
893 | flags |= XFS_PREALLOC_SET; |
894 | ||
83aee9e4 CH |
895 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
896 | offset + len > i_size_read(inode)) { | |
897 | new_size = offset + len; | |
2451337d | 898 | error = inode_newsize_ok(inode, new_size); |
83aee9e4 CH |
899 | if (error) |
900 | goto out_unlock; | |
901 | } | |
2fe17c10 | 902 | |
376ba313 LC |
903 | if (mode & FALLOC_FL_ZERO_RANGE) |
904 | error = xfs_zero_file_space(ip, offset, len); | |
98cc2db5 DW |
905 | else { |
906 | if (mode & FALLOC_FL_UNSHARE_RANGE) { | |
907 | error = xfs_reflink_unshare(ip, offset, len); | |
908 | if (error) | |
909 | goto out_unlock; | |
910 | } | |
376ba313 LC |
911 | error = xfs_alloc_file_space(ip, offset, len, |
912 | XFS_BMAPI_PREALLOC); | |
98cc2db5 | 913 | } |
2fe17c10 CH |
914 | if (error) |
915 | goto out_unlock; | |
916 | } | |
917 | ||
83aee9e4 | 918 | if (file->f_flags & O_DSYNC) |
8add71ca CH |
919 | flags |= XFS_PREALLOC_SYNC; |
920 | ||
921 | error = xfs_update_prealloc_flags(ip, flags); | |
2fe17c10 CH |
922 | if (error) |
923 | goto out_unlock; | |
924 | ||
925 | /* Change file size if needed */ | |
926 | if (new_size) { | |
927 | struct iattr iattr; | |
928 | ||
929 | iattr.ia_valid = ATTR_SIZE; | |
930 | iattr.ia_size = new_size; | |
69bca807 | 931 | error = xfs_vn_setattr_size(file_dentry(file), &iattr); |
a904b1ca NJ |
932 | if (error) |
933 | goto out_unlock; | |
2fe17c10 CH |
934 | } |
935 | ||
a904b1ca NJ |
936 | /* |
937 | * Perform hole insertion now that the file size has been | |
938 | * updated so that if we crash during the operation we don't | |
939 | * leave shifted extents past EOF and hence losing access to | |
940 | * the data that is contained within them. | |
941 | */ | |
942 | if (do_file_insert) | |
943 | error = xfs_insert_file_space(ip, offset, len); | |
944 | ||
2fe17c10 | 945 | out_unlock: |
781355c6 | 946 | xfs_iunlock(ip, iolock); |
2451337d | 947 | return error; |
2fe17c10 CH |
948 | } |
949 | ||
9fe26045 DW |
950 | /* |
951 | * Flush all file writes out to disk. | |
952 | */ | |
953 | static int | |
954 | xfs_file_wait_for_io( | |
955 | struct inode *inode, | |
956 | loff_t offset, | |
957 | size_t len) | |
958 | { | |
959 | loff_t rounding; | |
960 | loff_t ioffset; | |
961 | loff_t iendoffset; | |
962 | loff_t bs; | |
963 | int ret; | |
964 | ||
965 | bs = inode->i_sb->s_blocksize; | |
966 | inode_dio_wait(inode); | |
967 | ||
968 | rounding = max_t(xfs_off_t, bs, PAGE_SIZE); | |
969 | ioffset = round_down(offset, rounding); | |
970 | iendoffset = round_up(offset + len, rounding) - 1; | |
971 | ret = filemap_write_and_wait_range(inode->i_mapping, ioffset, | |
972 | iendoffset); | |
973 | return ret; | |
974 | } | |
975 | ||
976 | /* Hook up to the VFS reflink function */ | |
977 | STATIC int | |
978 | xfs_file_share_range( | |
979 | struct file *file_in, | |
980 | loff_t pos_in, | |
981 | struct file *file_out, | |
982 | loff_t pos_out, | |
cc714660 DW |
983 | u64 len, |
984 | bool is_dedupe) | |
9fe26045 DW |
985 | { |
986 | struct inode *inode_in; | |
987 | struct inode *inode_out; | |
988 | ssize_t ret; | |
989 | loff_t bs; | |
990 | loff_t isize; | |
991 | int same_inode; | |
992 | loff_t blen; | |
cc714660 | 993 | unsigned int flags = 0; |
9fe26045 DW |
994 | |
995 | inode_in = file_inode(file_in); | |
996 | inode_out = file_inode(file_out); | |
997 | bs = inode_out->i_sb->s_blocksize; | |
998 | ||
999 | /* Don't touch certain kinds of inodes */ | |
1000 | if (IS_IMMUTABLE(inode_out)) | |
1001 | return -EPERM; | |
1002 | if (IS_SWAPFILE(inode_in) || | |
1003 | IS_SWAPFILE(inode_out)) | |
1004 | return -ETXTBSY; | |
1005 | ||
1006 | /* Reflink only works within this filesystem. */ | |
1007 | if (inode_in->i_sb != inode_out->i_sb) | |
1008 | return -EXDEV; | |
1009 | same_inode = (inode_in->i_ino == inode_out->i_ino); | |
1010 | ||
1011 | /* Don't reflink dirs, pipes, sockets... */ | |
1012 | if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode)) | |
1013 | return -EISDIR; | |
1014 | if (S_ISFIFO(inode_in->i_mode) || S_ISFIFO(inode_out->i_mode)) | |
1015 | return -EINVAL; | |
1016 | if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode)) | |
1017 | return -EINVAL; | |
1018 | ||
4f435ebe DW |
1019 | /* Don't share DAX file data for now. */ |
1020 | if (IS_DAX(inode_in) || IS_DAX(inode_out)) | |
1021 | return -EINVAL; | |
1022 | ||
9fe26045 DW |
1023 | /* Are we going all the way to the end? */ |
1024 | isize = i_size_read(inode_in); | |
1025 | if (isize == 0) | |
1026 | return 0; | |
1027 | if (len == 0) | |
1028 | len = isize - pos_in; | |
1029 | ||
1030 | /* Ensure offsets don't wrap and the input is inside i_size */ | |
1031 | if (pos_in + len < pos_in || pos_out + len < pos_out || | |
1032 | pos_in + len > isize) | |
1033 | return -EINVAL; | |
1034 | ||
cc714660 DW |
1035 | /* Don't allow dedupe past EOF in the dest file */ |
1036 | if (is_dedupe) { | |
1037 | loff_t disize; | |
1038 | ||
1039 | disize = i_size_read(inode_out); | |
1040 | if (pos_out >= disize || pos_out + len > disize) | |
1041 | return -EINVAL; | |
1042 | } | |
1043 | ||
9fe26045 DW |
1044 | /* If we're linking to EOF, continue to the block boundary. */ |
1045 | if (pos_in + len == isize) | |
1046 | blen = ALIGN(isize, bs) - pos_in; | |
1047 | else | |
1048 | blen = len; | |
1049 | ||
1050 | /* Only reflink if we're aligned to block boundaries */ | |
1051 | if (!IS_ALIGNED(pos_in, bs) || !IS_ALIGNED(pos_in + blen, bs) || | |
1052 | !IS_ALIGNED(pos_out, bs) || !IS_ALIGNED(pos_out + blen, bs)) | |
1053 | return -EINVAL; | |
1054 | ||
1055 | /* Don't allow overlapped reflink within the same file */ | |
1056 | if (same_inode && pos_out + blen > pos_in && pos_out < pos_in + blen) | |
1057 | return -EINVAL; | |
1058 | ||
1059 | /* Wait for the completion of any pending IOs on srcfile */ | |
1060 | ret = xfs_file_wait_for_io(inode_in, pos_in, len); | |
1061 | if (ret) | |
93fed470 | 1062 | goto out; |
9fe26045 DW |
1063 | ret = xfs_file_wait_for_io(inode_out, pos_out, len); |
1064 | if (ret) | |
93fed470 | 1065 | goto out; |
9fe26045 | 1066 | |
cc714660 DW |
1067 | if (is_dedupe) |
1068 | flags |= XFS_REFLINK_DEDUPE; | |
9fe26045 | 1069 | ret = xfs_reflink_remap_range(XFS_I(inode_in), pos_in, XFS_I(inode_out), |
cc714660 | 1070 | pos_out, len, flags); |
9fe26045 | 1071 | if (ret < 0) |
93fed470 | 1072 | goto out; |
9fe26045 | 1073 | |
93fed470 | 1074 | out: |
9fe26045 DW |
1075 | return ret; |
1076 | } | |
1077 | ||
1078 | STATIC ssize_t | |
1079 | xfs_file_copy_range( | |
1080 | struct file *file_in, | |
1081 | loff_t pos_in, | |
1082 | struct file *file_out, | |
1083 | loff_t pos_out, | |
1084 | size_t len, | |
1085 | unsigned int flags) | |
1086 | { | |
1087 | int error; | |
1088 | ||
1089 | error = xfs_file_share_range(file_in, pos_in, file_out, pos_out, | |
cc714660 | 1090 | len, false); |
9fe26045 DW |
1091 | if (error) |
1092 | return error; | |
1093 | return len; | |
1094 | } | |
1095 | ||
1096 | STATIC int | |
1097 | xfs_file_clone_range( | |
1098 | struct file *file_in, | |
1099 | loff_t pos_in, | |
1100 | struct file *file_out, | |
1101 | loff_t pos_out, | |
1102 | u64 len) | |
1103 | { | |
1104 | return xfs_file_share_range(file_in, pos_in, file_out, pos_out, | |
cc714660 DW |
1105 | len, false); |
1106 | } | |
1107 | ||
1108 | #define XFS_MAX_DEDUPE_LEN (16 * 1024 * 1024) | |
1109 | STATIC ssize_t | |
1110 | xfs_file_dedupe_range( | |
1111 | struct file *src_file, | |
1112 | u64 loff, | |
1113 | u64 len, | |
1114 | struct file *dst_file, | |
1115 | u64 dst_loff) | |
1116 | { | |
1117 | int error; | |
1118 | ||
1119 | /* | |
1120 | * Limit the total length we will dedupe for each operation. | |
1121 | * This is intended to bound the total time spent in this | |
1122 | * ioctl to something sane. | |
1123 | */ | |
1124 | if (len > XFS_MAX_DEDUPE_LEN) | |
1125 | len = XFS_MAX_DEDUPE_LEN; | |
1126 | ||
1127 | error = xfs_file_share_range(src_file, loff, dst_file, dst_loff, | |
1128 | len, true); | |
1129 | if (error) | |
1130 | return error; | |
1131 | return len; | |
9fe26045 | 1132 | } |
2fe17c10 | 1133 | |
1da177e4 | 1134 | STATIC int |
3562fd45 | 1135 | xfs_file_open( |
1da177e4 | 1136 | struct inode *inode, |
f999a5bf | 1137 | struct file *file) |
1da177e4 | 1138 | { |
f999a5bf | 1139 | if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS) |
1da177e4 | 1140 | return -EFBIG; |
f999a5bf CH |
1141 | if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb))) |
1142 | return -EIO; | |
1143 | return 0; | |
1144 | } | |
1145 | ||
1146 | STATIC int | |
1147 | xfs_dir_open( | |
1148 | struct inode *inode, | |
1149 | struct file *file) | |
1150 | { | |
1151 | struct xfs_inode *ip = XFS_I(inode); | |
1152 | int mode; | |
1153 | int error; | |
1154 | ||
1155 | error = xfs_file_open(inode, file); | |
1156 | if (error) | |
1157 | return error; | |
1158 | ||
1159 | /* | |
1160 | * If there are any blocks, read-ahead block 0 as we're almost | |
1161 | * certain to have the next operation be a read there. | |
1162 | */ | |
309ecac8 | 1163 | mode = xfs_ilock_data_map_shared(ip); |
f999a5bf | 1164 | if (ip->i_d.di_nextents > 0) |
9df2dd0b | 1165 | xfs_dir3_data_readahead(ip, 0, -1); |
f999a5bf CH |
1166 | xfs_iunlock(ip, mode); |
1167 | return 0; | |
1da177e4 LT |
1168 | } |
1169 | ||
1da177e4 | 1170 | STATIC int |
3562fd45 | 1171 | xfs_file_release( |
1da177e4 LT |
1172 | struct inode *inode, |
1173 | struct file *filp) | |
1174 | { | |
2451337d | 1175 | return xfs_release(XFS_I(inode)); |
1da177e4 LT |
1176 | } |
1177 | ||
1da177e4 | 1178 | STATIC int |
3562fd45 | 1179 | xfs_file_readdir( |
b8227554 AV |
1180 | struct file *file, |
1181 | struct dir_context *ctx) | |
1da177e4 | 1182 | { |
b8227554 | 1183 | struct inode *inode = file_inode(file); |
739bfb2a | 1184 | xfs_inode_t *ip = XFS_I(inode); |
051e7cd4 CH |
1185 | size_t bufsize; |
1186 | ||
1187 | /* | |
1188 | * The Linux API doesn't pass down the total size of the buffer | |
1189 | * we read into down to the filesystem. With the filldir concept | |
1190 | * it's not needed for correct information, but the XFS dir2 leaf | |
1191 | * code wants an estimate of the buffer size to calculate it's | |
1192 | * readahead window and size the buffers used for mapping to | |
1193 | * physical blocks. | |
1194 | * | |
1195 | * Try to give it an estimate that's good enough, maybe at some | |
1196 | * point we can change the ->readdir prototype to include the | |
a9cc799e | 1197 | * buffer size. For now we use the current glibc buffer size. |
051e7cd4 | 1198 | */ |
a9cc799e | 1199 | bufsize = (size_t)min_t(loff_t, 32768, ip->i_d.di_size); |
051e7cd4 | 1200 | |
8300475e | 1201 | return xfs_readdir(ip, ctx, bufsize); |
1da177e4 LT |
1202 | } |
1203 | ||
d126d43f JL |
1204 | /* |
1205 | * This type is designed to indicate the type of offset we would like | |
49c69591 | 1206 | * to search from page cache for xfs_seek_hole_data(). |
d126d43f JL |
1207 | */ |
1208 | enum { | |
1209 | HOLE_OFF = 0, | |
1210 | DATA_OFF, | |
1211 | }; | |
1212 | ||
1213 | /* | |
1214 | * Lookup the desired type of offset from the given page. | |
1215 | * | |
1216 | * On success, return true and the offset argument will point to the | |
1217 | * start of the region that was found. Otherwise this function will | |
1218 | * return false and keep the offset argument unchanged. | |
1219 | */ | |
1220 | STATIC bool | |
1221 | xfs_lookup_buffer_offset( | |
1222 | struct page *page, | |
1223 | loff_t *offset, | |
1224 | unsigned int type) | |
1225 | { | |
1226 | loff_t lastoff = page_offset(page); | |
1227 | bool found = false; | |
1228 | struct buffer_head *bh, *head; | |
1229 | ||
1230 | bh = head = page_buffers(page); | |
1231 | do { | |
1232 | /* | |
1233 | * Unwritten extents that have data in the page | |
1234 | * cache covering them can be identified by the | |
1235 | * BH_Unwritten state flag. Pages with multiple | |
1236 | * buffers might have a mix of holes, data and | |
1237 | * unwritten extents - any buffer with valid | |
1238 | * data in it should have BH_Uptodate flag set | |
1239 | * on it. | |
1240 | */ | |
1241 | if (buffer_unwritten(bh) || | |
1242 | buffer_uptodate(bh)) { | |
1243 | if (type == DATA_OFF) | |
1244 | found = true; | |
1245 | } else { | |
1246 | if (type == HOLE_OFF) | |
1247 | found = true; | |
1248 | } | |
1249 | ||
1250 | if (found) { | |
1251 | *offset = lastoff; | |
1252 | break; | |
1253 | } | |
1254 | lastoff += bh->b_size; | |
1255 | } while ((bh = bh->b_this_page) != head); | |
1256 | ||
1257 | return found; | |
1258 | } | |
1259 | ||
1260 | /* | |
1261 | * This routine is called to find out and return a data or hole offset | |
1262 | * from the page cache for unwritten extents according to the desired | |
49c69591 | 1263 | * type for xfs_seek_hole_data(). |
d126d43f JL |
1264 | * |
1265 | * The argument offset is used to tell where we start to search from the | |
1266 | * page cache. Map is used to figure out the end points of the range to | |
1267 | * lookup pages. | |
1268 | * | |
1269 | * Return true if the desired type of offset was found, and the argument | |
1270 | * offset is filled with that address. Otherwise, return false and keep | |
1271 | * offset unchanged. | |
1272 | */ | |
1273 | STATIC bool | |
1274 | xfs_find_get_desired_pgoff( | |
1275 | struct inode *inode, | |
1276 | struct xfs_bmbt_irec *map, | |
1277 | unsigned int type, | |
1278 | loff_t *offset) | |
1279 | { | |
1280 | struct xfs_inode *ip = XFS_I(inode); | |
1281 | struct xfs_mount *mp = ip->i_mount; | |
1282 | struct pagevec pvec; | |
1283 | pgoff_t index; | |
1284 | pgoff_t end; | |
1285 | loff_t endoff; | |
1286 | loff_t startoff = *offset; | |
1287 | loff_t lastoff = startoff; | |
1288 | bool found = false; | |
1289 | ||
1290 | pagevec_init(&pvec, 0); | |
1291 | ||
09cbfeaf | 1292 | index = startoff >> PAGE_SHIFT; |
d126d43f | 1293 | endoff = XFS_FSB_TO_B(mp, map->br_startoff + map->br_blockcount); |
09cbfeaf | 1294 | end = endoff >> PAGE_SHIFT; |
d126d43f JL |
1295 | do { |
1296 | int want; | |
1297 | unsigned nr_pages; | |
1298 | unsigned int i; | |
1299 | ||
1300 | want = min_t(pgoff_t, end - index, PAGEVEC_SIZE); | |
1301 | nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index, | |
1302 | want); | |
1303 | /* | |
1304 | * No page mapped into given range. If we are searching holes | |
1305 | * and if this is the first time we got into the loop, it means | |
1306 | * that the given offset is landed in a hole, return it. | |
1307 | * | |
1308 | * If we have already stepped through some block buffers to find | |
1309 | * holes but they all contains data. In this case, the last | |
1310 | * offset is already updated and pointed to the end of the last | |
1311 | * mapped page, if it does not reach the endpoint to search, | |
1312 | * that means there should be a hole between them. | |
1313 | */ | |
1314 | if (nr_pages == 0) { | |
1315 | /* Data search found nothing */ | |
1316 | if (type == DATA_OFF) | |
1317 | break; | |
1318 | ||
1319 | ASSERT(type == HOLE_OFF); | |
1320 | if (lastoff == startoff || lastoff < endoff) { | |
1321 | found = true; | |
1322 | *offset = lastoff; | |
1323 | } | |
1324 | break; | |
1325 | } | |
1326 | ||
1327 | /* | |
1328 | * At lease we found one page. If this is the first time we | |
1329 | * step into the loop, and if the first page index offset is | |
1330 | * greater than the given search offset, a hole was found. | |
1331 | */ | |
1332 | if (type == HOLE_OFF && lastoff == startoff && | |
1333 | lastoff < page_offset(pvec.pages[0])) { | |
1334 | found = true; | |
1335 | break; | |
1336 | } | |
1337 | ||
1338 | for (i = 0; i < nr_pages; i++) { | |
1339 | struct page *page = pvec.pages[i]; | |
1340 | loff_t b_offset; | |
1341 | ||
1342 | /* | |
1343 | * At this point, the page may be truncated or | |
1344 | * invalidated (changing page->mapping to NULL), | |
1345 | * or even swizzled back from swapper_space to tmpfs | |
1346 | * file mapping. However, page->index will not change | |
1347 | * because we have a reference on the page. | |
1348 | * | |
1349 | * Searching done if the page index is out of range. | |
1350 | * If the current offset is not reaches the end of | |
1351 | * the specified search range, there should be a hole | |
1352 | * between them. | |
1353 | */ | |
1354 | if (page->index > end) { | |
1355 | if (type == HOLE_OFF && lastoff < endoff) { | |
1356 | *offset = lastoff; | |
1357 | found = true; | |
1358 | } | |
1359 | goto out; | |
1360 | } | |
1361 | ||
1362 | lock_page(page); | |
1363 | /* | |
1364 | * Page truncated or invalidated(page->mapping == NULL). | |
1365 | * We can freely skip it and proceed to check the next | |
1366 | * page. | |
1367 | */ | |
1368 | if (unlikely(page->mapping != inode->i_mapping)) { | |
1369 | unlock_page(page); | |
1370 | continue; | |
1371 | } | |
1372 | ||
1373 | if (!page_has_buffers(page)) { | |
1374 | unlock_page(page); | |
1375 | continue; | |
1376 | } | |
1377 | ||
1378 | found = xfs_lookup_buffer_offset(page, &b_offset, type); | |
1379 | if (found) { | |
1380 | /* | |
1381 | * The found offset may be less than the start | |
1382 | * point to search if this is the first time to | |
1383 | * come here. | |
1384 | */ | |
1385 | *offset = max_t(loff_t, startoff, b_offset); | |
1386 | unlock_page(page); | |
1387 | goto out; | |
1388 | } | |
1389 | ||
1390 | /* | |
1391 | * We either searching data but nothing was found, or | |
1392 | * searching hole but found a data buffer. In either | |
1393 | * case, probably the next page contains the desired | |
1394 | * things, update the last offset to it so. | |
1395 | */ | |
1396 | lastoff = page_offset(page) + PAGE_SIZE; | |
1397 | unlock_page(page); | |
1398 | } | |
1399 | ||
1400 | /* | |
1401 | * The number of returned pages less than our desired, search | |
1402 | * done. In this case, nothing was found for searching data, | |
1403 | * but we found a hole behind the last offset. | |
1404 | */ | |
1405 | if (nr_pages < want) { | |
1406 | if (type == HOLE_OFF) { | |
1407 | *offset = lastoff; | |
1408 | found = true; | |
1409 | } | |
1410 | break; | |
1411 | } | |
1412 | ||
1413 | index = pvec.pages[i - 1]->index + 1; | |
1414 | pagevec_release(&pvec); | |
1415 | } while (index <= end); | |
1416 | ||
1417 | out: | |
1418 | pagevec_release(&pvec); | |
1419 | return found; | |
1420 | } | |
1421 | ||
8aa7d37e ES |
1422 | /* |
1423 | * caller must lock inode with xfs_ilock_data_map_shared, | |
1424 | * can we craft an appropriate ASSERT? | |
1425 | * | |
1426 | * end is because the VFS-level lseek interface is defined such that any | |
1427 | * offset past i_size shall return -ENXIO, but we use this for quota code | |
1428 | * which does not maintain i_size, and we want to SEEK_DATA past i_size. | |
1429 | */ | |
1430 | loff_t | |
1431 | __xfs_seek_hole_data( | |
1432 | struct inode *inode, | |
49c69591 | 1433 | loff_t start, |
8aa7d37e | 1434 | loff_t end, |
49c69591 | 1435 | int whence) |
3fe3e6b1 | 1436 | { |
3fe3e6b1 JL |
1437 | struct xfs_inode *ip = XFS_I(inode); |
1438 | struct xfs_mount *mp = ip->i_mount; | |
3fe3e6b1 | 1439 | loff_t uninitialized_var(offset); |
3fe3e6b1 | 1440 | xfs_fileoff_t fsbno; |
8aa7d37e | 1441 | xfs_filblks_t lastbno; |
3fe3e6b1 JL |
1442 | int error; |
1443 | ||
8aa7d37e | 1444 | if (start >= end) { |
2451337d | 1445 | error = -ENXIO; |
8aa7d37e | 1446 | goto out_error; |
3fe3e6b1 JL |
1447 | } |
1448 | ||
3fe3e6b1 JL |
1449 | /* |
1450 | * Try to read extents from the first block indicated | |
1451 | * by fsbno to the end block of the file. | |
1452 | */ | |
52f1acc8 | 1453 | fsbno = XFS_B_TO_FSBT(mp, start); |
8aa7d37e | 1454 | lastbno = XFS_B_TO_FSB(mp, end); |
49c69591 | 1455 | |
52f1acc8 JL |
1456 | for (;;) { |
1457 | struct xfs_bmbt_irec map[2]; | |
1458 | int nmap = 2; | |
1459 | unsigned int i; | |
3fe3e6b1 | 1460 | |
8aa7d37e | 1461 | error = xfs_bmapi_read(ip, fsbno, lastbno - fsbno, map, &nmap, |
52f1acc8 JL |
1462 | XFS_BMAPI_ENTIRE); |
1463 | if (error) | |
8aa7d37e | 1464 | goto out_error; |
3fe3e6b1 | 1465 | |
52f1acc8 JL |
1466 | /* No extents at given offset, must be beyond EOF */ |
1467 | if (nmap == 0) { | |
2451337d | 1468 | error = -ENXIO; |
8aa7d37e | 1469 | goto out_error; |
52f1acc8 JL |
1470 | } |
1471 | ||
1472 | for (i = 0; i < nmap; i++) { | |
1473 | offset = max_t(loff_t, start, | |
1474 | XFS_FSB_TO_B(mp, map[i].br_startoff)); | |
1475 | ||
49c69591 ES |
1476 | /* Landed in the hole we wanted? */ |
1477 | if (whence == SEEK_HOLE && | |
1478 | map[i].br_startblock == HOLESTARTBLOCK) | |
1479 | goto out; | |
1480 | ||
1481 | /* Landed in the data extent we wanted? */ | |
1482 | if (whence == SEEK_DATA && | |
1483 | (map[i].br_startblock == DELAYSTARTBLOCK || | |
1484 | (map[i].br_state == XFS_EXT_NORM && | |
1485 | !isnullstartblock(map[i].br_startblock)))) | |
52f1acc8 JL |
1486 | goto out; |
1487 | ||
1488 | /* | |
49c69591 ES |
1489 | * Landed in an unwritten extent, try to search |
1490 | * for hole or data from page cache. | |
52f1acc8 JL |
1491 | */ |
1492 | if (map[i].br_state == XFS_EXT_UNWRITTEN) { | |
1493 | if (xfs_find_get_desired_pgoff(inode, &map[i], | |
49c69591 ES |
1494 | whence == SEEK_HOLE ? HOLE_OFF : DATA_OFF, |
1495 | &offset)) | |
52f1acc8 JL |
1496 | goto out; |
1497 | } | |
1498 | } | |
1499 | ||
1500 | /* | |
49c69591 ES |
1501 | * We only received one extent out of the two requested. This |
1502 | * means we've hit EOF and didn't find what we are looking for. | |
52f1acc8 | 1503 | */ |
3fe3e6b1 | 1504 | if (nmap == 1) { |
49c69591 ES |
1505 | /* |
1506 | * If we were looking for a hole, set offset to | |
1507 | * the end of the file (i.e., there is an implicit | |
1508 | * hole at the end of any file). | |
1509 | */ | |
1510 | if (whence == SEEK_HOLE) { | |
8aa7d37e | 1511 | offset = end; |
49c69591 ES |
1512 | break; |
1513 | } | |
1514 | /* | |
1515 | * If we were looking for data, it's nowhere to be found | |
1516 | */ | |
1517 | ASSERT(whence == SEEK_DATA); | |
2451337d | 1518 | error = -ENXIO; |
8aa7d37e | 1519 | goto out_error; |
3fe3e6b1 JL |
1520 | } |
1521 | ||
52f1acc8 JL |
1522 | ASSERT(i > 1); |
1523 | ||
1524 | /* | |
1525 | * Nothing was found, proceed to the next round of search | |
49c69591 | 1526 | * if the next reading offset is not at or beyond EOF. |
52f1acc8 JL |
1527 | */ |
1528 | fsbno = map[i - 1].br_startoff + map[i - 1].br_blockcount; | |
1529 | start = XFS_FSB_TO_B(mp, fsbno); | |
8aa7d37e | 1530 | if (start >= end) { |
49c69591 | 1531 | if (whence == SEEK_HOLE) { |
8aa7d37e | 1532 | offset = end; |
49c69591 ES |
1533 | break; |
1534 | } | |
1535 | ASSERT(whence == SEEK_DATA); | |
2451337d | 1536 | error = -ENXIO; |
8aa7d37e | 1537 | goto out_error; |
52f1acc8 | 1538 | } |
3fe3e6b1 JL |
1539 | } |
1540 | ||
b686d1f7 JL |
1541 | out: |
1542 | /* | |
49c69591 | 1543 | * If at this point we have found the hole we wanted, the returned |
b686d1f7 | 1544 | * offset may be bigger than the file size as it may be aligned to |
49c69591 | 1545 | * page boundary for unwritten extents. We need to deal with this |
b686d1f7 JL |
1546 | * situation in particular. |
1547 | */ | |
49c69591 | 1548 | if (whence == SEEK_HOLE) |
8aa7d37e ES |
1549 | offset = min_t(loff_t, offset, end); |
1550 | ||
1551 | return offset; | |
1552 | ||
1553 | out_error: | |
1554 | return error; | |
1555 | } | |
1556 | ||
1557 | STATIC loff_t | |
1558 | xfs_seek_hole_data( | |
1559 | struct file *file, | |
1560 | loff_t start, | |
1561 | int whence) | |
1562 | { | |
1563 | struct inode *inode = file->f_mapping->host; | |
1564 | struct xfs_inode *ip = XFS_I(inode); | |
1565 | struct xfs_mount *mp = ip->i_mount; | |
1566 | uint lock; | |
1567 | loff_t offset, end; | |
1568 | int error = 0; | |
1569 | ||
1570 | if (XFS_FORCED_SHUTDOWN(mp)) | |
1571 | return -EIO; | |
1572 | ||
1573 | lock = xfs_ilock_data_map_shared(ip); | |
1574 | ||
1575 | end = i_size_read(inode); | |
1576 | offset = __xfs_seek_hole_data(inode, start, end, whence); | |
1577 | if (offset < 0) { | |
1578 | error = offset; | |
1579 | goto out_unlock; | |
1580 | } | |
1581 | ||
46a1c2c7 | 1582 | offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes); |
3fe3e6b1 JL |
1583 | |
1584 | out_unlock: | |
01f4f327 | 1585 | xfs_iunlock(ip, lock); |
3fe3e6b1 JL |
1586 | |
1587 | if (error) | |
2451337d | 1588 | return error; |
3fe3e6b1 JL |
1589 | return offset; |
1590 | } | |
1591 | ||
1592 | STATIC loff_t | |
1593 | xfs_file_llseek( | |
1594 | struct file *file, | |
1595 | loff_t offset, | |
59f9c004 | 1596 | int whence) |
3fe3e6b1 | 1597 | { |
59f9c004 | 1598 | switch (whence) { |
3fe3e6b1 JL |
1599 | case SEEK_END: |
1600 | case SEEK_CUR: | |
1601 | case SEEK_SET: | |
59f9c004 | 1602 | return generic_file_llseek(file, offset, whence); |
3fe3e6b1 | 1603 | case SEEK_HOLE: |
49c69591 | 1604 | case SEEK_DATA: |
59f9c004 | 1605 | return xfs_seek_hole_data(file, offset, whence); |
3fe3e6b1 JL |
1606 | default: |
1607 | return -EINVAL; | |
1608 | } | |
1609 | } | |
1610 | ||
de0e8c20 DC |
1611 | /* |
1612 | * Locking for serialisation of IO during page faults. This results in a lock | |
1613 | * ordering of: | |
1614 | * | |
1615 | * mmap_sem (MM) | |
6b698ede | 1616 | * sb_start_pagefault(vfs, freeze) |
13ad4fe3 | 1617 | * i_mmaplock (XFS - truncate serialisation) |
6b698ede DC |
1618 | * page_lock (MM) |
1619 | * i_lock (XFS - extent map serialisation) | |
de0e8c20 | 1620 | */ |
de0e8c20 | 1621 | |
075a924d DC |
1622 | /* |
1623 | * mmap()d file has taken write protection fault and is being made writable. We | |
1624 | * can set the page state up correctly for a writable page, which means we can | |
1625 | * do correct delalloc accounting (ENOSPC checking!) and unwritten extent | |
1626 | * mapping. | |
de0e8c20 DC |
1627 | */ |
1628 | STATIC int | |
075a924d | 1629 | xfs_filemap_page_mkwrite( |
de0e8c20 DC |
1630 | struct vm_area_struct *vma, |
1631 | struct vm_fault *vmf) | |
1632 | { | |
6b698ede | 1633 | struct inode *inode = file_inode(vma->vm_file); |
ec56b1f1 | 1634 | int ret; |
de0e8c20 | 1635 | |
6b698ede | 1636 | trace_xfs_filemap_page_mkwrite(XFS_I(inode)); |
de0e8c20 | 1637 | |
6b698ede | 1638 | sb_start_pagefault(inode->i_sb); |
ec56b1f1 | 1639 | file_update_time(vma->vm_file); |
6b698ede | 1640 | xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED); |
de0e8c20 | 1641 | |
6b698ede | 1642 | if (IS_DAX(inode)) { |
6c31f495 | 1643 | ret = iomap_dax_fault(vma, vmf, &xfs_iomap_ops); |
6b698ede | 1644 | } else { |
68a9f5e7 | 1645 | ret = iomap_page_mkwrite(vma, vmf, &xfs_iomap_ops); |
6b698ede DC |
1646 | ret = block_page_mkwrite_return(ret); |
1647 | } | |
1648 | ||
1649 | xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED); | |
1650 | sb_end_pagefault(inode->i_sb); | |
1651 | ||
1652 | return ret; | |
de0e8c20 DC |
1653 | } |
1654 | ||
075a924d | 1655 | STATIC int |
6b698ede | 1656 | xfs_filemap_fault( |
075a924d DC |
1657 | struct vm_area_struct *vma, |
1658 | struct vm_fault *vmf) | |
1659 | { | |
b2442c5a | 1660 | struct inode *inode = file_inode(vma->vm_file); |
6b698ede | 1661 | int ret; |
ec56b1f1 | 1662 | |
b2442c5a | 1663 | trace_xfs_filemap_fault(XFS_I(inode)); |
075a924d | 1664 | |
6b698ede | 1665 | /* DAX can shortcut the normal fault path on write faults! */ |
b2442c5a | 1666 | if ((vmf->flags & FAULT_FLAG_WRITE) && IS_DAX(inode)) |
6b698ede | 1667 | return xfs_filemap_page_mkwrite(vma, vmf); |
075a924d | 1668 | |
b2442c5a DC |
1669 | xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED); |
1670 | if (IS_DAX(inode)) { | |
1671 | /* | |
1672 | * we do not want to trigger unwritten extent conversion on read | |
1673 | * faults - that is unnecessary overhead and would also require | |
1674 | * changes to xfs_get_blocks_direct() to map unwritten extent | |
1675 | * ioend for conversion on read-only mappings. | |
1676 | */ | |
6c31f495 | 1677 | ret = iomap_dax_fault(vma, vmf, &xfs_iomap_ops); |
b2442c5a DC |
1678 | } else |
1679 | ret = filemap_fault(vma, vmf); | |
1680 | xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED); | |
075a924d | 1681 | |
6b698ede DC |
1682 | return ret; |
1683 | } | |
1684 | ||
13ad4fe3 DC |
1685 | /* |
1686 | * Similar to xfs_filemap_fault(), the DAX fault path can call into here on | |
1687 | * both read and write faults. Hence we need to handle both cases. There is no | |
1688 | * ->pmd_mkwrite callout for huge pages, so we have a single function here to | |
1689 | * handle both cases here. @flags carries the information on the type of fault | |
1690 | * occuring. | |
1691 | */ | |
acd76e74 MW |
1692 | STATIC int |
1693 | xfs_filemap_pmd_fault( | |
1694 | struct vm_area_struct *vma, | |
1695 | unsigned long addr, | |
1696 | pmd_t *pmd, | |
1697 | unsigned int flags) | |
1698 | { | |
1699 | struct inode *inode = file_inode(vma->vm_file); | |
1700 | struct xfs_inode *ip = XFS_I(inode); | |
1701 | int ret; | |
1702 | ||
1703 | if (!IS_DAX(inode)) | |
1704 | return VM_FAULT_FALLBACK; | |
1705 | ||
1706 | trace_xfs_filemap_pmd_fault(ip); | |
1707 | ||
13ad4fe3 DC |
1708 | if (flags & FAULT_FLAG_WRITE) { |
1709 | sb_start_pagefault(inode->i_sb); | |
1710 | file_update_time(vma->vm_file); | |
1711 | } | |
1712 | ||
acd76e74 | 1713 | xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED); |
6b524995 | 1714 | ret = dax_pmd_fault(vma, addr, pmd, flags, xfs_get_blocks_dax_fault); |
acd76e74 | 1715 | xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED); |
acd76e74 | 1716 | |
13ad4fe3 DC |
1717 | if (flags & FAULT_FLAG_WRITE) |
1718 | sb_end_pagefault(inode->i_sb); | |
acd76e74 MW |
1719 | |
1720 | return ret; | |
1721 | } | |
1722 | ||
3af49285 DC |
1723 | /* |
1724 | * pfn_mkwrite was originally inteneded to ensure we capture time stamp | |
1725 | * updates on write faults. In reality, it's need to serialise against | |
5eb88dca RZ |
1726 | * truncate similar to page_mkwrite. Hence we cycle the XFS_MMAPLOCK_SHARED |
1727 | * to ensure we serialise the fault barrier in place. | |
3af49285 DC |
1728 | */ |
1729 | static int | |
1730 | xfs_filemap_pfn_mkwrite( | |
1731 | struct vm_area_struct *vma, | |
1732 | struct vm_fault *vmf) | |
1733 | { | |
1734 | ||
1735 | struct inode *inode = file_inode(vma->vm_file); | |
1736 | struct xfs_inode *ip = XFS_I(inode); | |
1737 | int ret = VM_FAULT_NOPAGE; | |
1738 | loff_t size; | |
1739 | ||
1740 | trace_xfs_filemap_pfn_mkwrite(ip); | |
1741 | ||
1742 | sb_start_pagefault(inode->i_sb); | |
1743 | file_update_time(vma->vm_file); | |
1744 | ||
1745 | /* check if the faulting page hasn't raced with truncate */ | |
1746 | xfs_ilock(ip, XFS_MMAPLOCK_SHARED); | |
1747 | size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
1748 | if (vmf->pgoff >= size) | |
1749 | ret = VM_FAULT_SIGBUS; | |
5eb88dca RZ |
1750 | else if (IS_DAX(inode)) |
1751 | ret = dax_pfn_mkwrite(vma, vmf); | |
3af49285 DC |
1752 | xfs_iunlock(ip, XFS_MMAPLOCK_SHARED); |
1753 | sb_end_pagefault(inode->i_sb); | |
acd76e74 | 1754 | return ret; |
3af49285 | 1755 | |
acd76e74 MW |
1756 | } |
1757 | ||
6b698ede DC |
1758 | static const struct vm_operations_struct xfs_file_vm_ops = { |
1759 | .fault = xfs_filemap_fault, | |
acd76e74 | 1760 | .pmd_fault = xfs_filemap_pmd_fault, |
6b698ede DC |
1761 | .map_pages = filemap_map_pages, |
1762 | .page_mkwrite = xfs_filemap_page_mkwrite, | |
3af49285 | 1763 | .pfn_mkwrite = xfs_filemap_pfn_mkwrite, |
6b698ede DC |
1764 | }; |
1765 | ||
1766 | STATIC int | |
1767 | xfs_file_mmap( | |
1768 | struct file *filp, | |
1769 | struct vm_area_struct *vma) | |
1770 | { | |
1771 | file_accessed(filp); | |
1772 | vma->vm_ops = &xfs_file_vm_ops; | |
1773 | if (IS_DAX(file_inode(filp))) | |
acd76e74 | 1774 | vma->vm_flags |= VM_MIXEDMAP | VM_HUGEPAGE; |
6b698ede | 1775 | return 0; |
075a924d DC |
1776 | } |
1777 | ||
4b6f5d20 | 1778 | const struct file_operations xfs_file_operations = { |
3fe3e6b1 | 1779 | .llseek = xfs_file_llseek, |
b4f5d2c6 | 1780 | .read_iter = xfs_file_read_iter, |
bf97f3bc | 1781 | .write_iter = xfs_file_write_iter, |
82c156f8 | 1782 | .splice_read = generic_file_splice_read, |
8d020765 | 1783 | .splice_write = iter_file_splice_write, |
3562fd45 | 1784 | .unlocked_ioctl = xfs_file_ioctl, |
1da177e4 | 1785 | #ifdef CONFIG_COMPAT |
3562fd45 | 1786 | .compat_ioctl = xfs_file_compat_ioctl, |
1da177e4 | 1787 | #endif |
3562fd45 NS |
1788 | .mmap = xfs_file_mmap, |
1789 | .open = xfs_file_open, | |
1790 | .release = xfs_file_release, | |
1791 | .fsync = xfs_file_fsync, | |
dbe6ec81 | 1792 | .get_unmapped_area = thp_get_unmapped_area, |
2fe17c10 | 1793 | .fallocate = xfs_file_fallocate, |
9fe26045 DW |
1794 | .copy_file_range = xfs_file_copy_range, |
1795 | .clone_file_range = xfs_file_clone_range, | |
cc714660 | 1796 | .dedupe_file_range = xfs_file_dedupe_range, |
1da177e4 LT |
1797 | }; |
1798 | ||
4b6f5d20 | 1799 | const struct file_operations xfs_dir_file_operations = { |
f999a5bf | 1800 | .open = xfs_dir_open, |
1da177e4 | 1801 | .read = generic_read_dir, |
3b0a3c1a | 1802 | .iterate_shared = xfs_file_readdir, |
59af1584 | 1803 | .llseek = generic_file_llseek, |
3562fd45 | 1804 | .unlocked_ioctl = xfs_file_ioctl, |
d3870398 | 1805 | #ifdef CONFIG_COMPAT |
3562fd45 | 1806 | .compat_ioctl = xfs_file_compat_ioctl, |
d3870398 | 1807 | #endif |
1da2f2db | 1808 | .fsync = xfs_dir_fsync, |
1da177e4 | 1809 | }; |