]>
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" |
1da177e4 LT |
41 | |
42 | #include <linux/dcache.h> | |
2fe17c10 | 43 | #include <linux/falloc.h> |
d126d43f | 44 | #include <linux/pagevec.h> |
66114cad | 45 | #include <linux/backing-dev.h> |
1da177e4 | 46 | |
f0f37e2f | 47 | static const struct vm_operations_struct xfs_file_vm_ops; |
1da177e4 | 48 | |
487f84f3 DC |
49 | /* |
50 | * Locking primitives for read and write IO paths to ensure we consistently use | |
51 | * and order the inode->i_mutex, ip->i_lock and ip->i_iolock. | |
52 | */ | |
53 | static inline void | |
54 | xfs_rw_ilock( | |
55 | struct xfs_inode *ip, | |
56 | int type) | |
57 | { | |
58 | if (type & XFS_IOLOCK_EXCL) | |
5955102c | 59 | inode_lock(VFS_I(ip)); |
487f84f3 DC |
60 | xfs_ilock(ip, type); |
61 | } | |
62 | ||
63 | static inline void | |
64 | xfs_rw_iunlock( | |
65 | struct xfs_inode *ip, | |
66 | int type) | |
67 | { | |
68 | xfs_iunlock(ip, type); | |
69 | if (type & XFS_IOLOCK_EXCL) | |
5955102c | 70 | inode_unlock(VFS_I(ip)); |
487f84f3 DC |
71 | } |
72 | ||
73 | static inline void | |
74 | xfs_rw_ilock_demote( | |
75 | struct xfs_inode *ip, | |
76 | int type) | |
77 | { | |
78 | xfs_ilock_demote(ip, type); | |
79 | if (type & XFS_IOLOCK_EXCL) | |
5955102c | 80 | inode_unlock(VFS_I(ip)); |
487f84f3 DC |
81 | } |
82 | ||
dda35b8f | 83 | /* |
68a9f5e7 CH |
84 | * Clear the specified ranges to zero through either the pagecache or DAX. |
85 | * Holes and unwritten extents will be left as-is as they already are zeroed. | |
dda35b8f | 86 | */ |
ef9d8733 | 87 | int |
7bb41db3 | 88 | xfs_zero_range( |
68a9f5e7 | 89 | struct xfs_inode *ip, |
7bb41db3 CH |
90 | xfs_off_t pos, |
91 | xfs_off_t count, | |
92 | bool *did_zero) | |
dda35b8f | 93 | { |
459f0fbc | 94 | return iomap_zero_range(VFS_I(ip), pos, count, NULL, &xfs_iomap_ops); |
dda35b8f CH |
95 | } |
96 | ||
8add71ca CH |
97 | int |
98 | xfs_update_prealloc_flags( | |
99 | struct xfs_inode *ip, | |
100 | enum xfs_prealloc_flags flags) | |
101 | { | |
102 | struct xfs_trans *tp; | |
103 | int error; | |
104 | ||
253f4911 CH |
105 | error = xfs_trans_alloc(ip->i_mount, &M_RES(ip->i_mount)->tr_writeid, |
106 | 0, 0, 0, &tp); | |
107 | if (error) | |
8add71ca | 108 | return error; |
8add71ca CH |
109 | |
110 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
111 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
112 | ||
113 | if (!(flags & XFS_PREALLOC_INVISIBLE)) { | |
c19b3b05 DC |
114 | VFS_I(ip)->i_mode &= ~S_ISUID; |
115 | if (VFS_I(ip)->i_mode & S_IXGRP) | |
116 | VFS_I(ip)->i_mode &= ~S_ISGID; | |
8add71ca CH |
117 | xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); |
118 | } | |
119 | ||
120 | if (flags & XFS_PREALLOC_SET) | |
121 | ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC; | |
122 | if (flags & XFS_PREALLOC_CLEAR) | |
123 | ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC; | |
124 | ||
125 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
126 | if (flags & XFS_PREALLOC_SYNC) | |
127 | xfs_trans_set_sync(tp); | |
70393313 | 128 | return xfs_trans_commit(tp); |
8add71ca CH |
129 | } |
130 | ||
1da2f2db CH |
131 | /* |
132 | * Fsync operations on directories are much simpler than on regular files, | |
133 | * as there is no file data to flush, and thus also no need for explicit | |
134 | * cache flush operations, and there are no non-transaction metadata updates | |
135 | * on directories either. | |
136 | */ | |
137 | STATIC int | |
138 | xfs_dir_fsync( | |
139 | struct file *file, | |
140 | loff_t start, | |
141 | loff_t end, | |
142 | int datasync) | |
143 | { | |
144 | struct xfs_inode *ip = XFS_I(file->f_mapping->host); | |
145 | struct xfs_mount *mp = ip->i_mount; | |
146 | xfs_lsn_t lsn = 0; | |
147 | ||
148 | trace_xfs_dir_fsync(ip); | |
149 | ||
150 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
151 | if (xfs_ipincount(ip)) | |
152 | lsn = ip->i_itemp->ili_last_lsn; | |
153 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
154 | ||
155 | if (!lsn) | |
156 | return 0; | |
2451337d | 157 | return _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, NULL); |
1da2f2db CH |
158 | } |
159 | ||
fd3200be CH |
160 | STATIC int |
161 | xfs_file_fsync( | |
162 | struct file *file, | |
02c24a82 JB |
163 | loff_t start, |
164 | loff_t end, | |
fd3200be CH |
165 | int datasync) |
166 | { | |
7ea80859 CH |
167 | struct inode *inode = file->f_mapping->host; |
168 | struct xfs_inode *ip = XFS_I(inode); | |
a27a263b | 169 | struct xfs_mount *mp = ip->i_mount; |
fd3200be CH |
170 | int error = 0; |
171 | int log_flushed = 0; | |
b1037058 | 172 | xfs_lsn_t lsn = 0; |
fd3200be | 173 | |
cca28fb8 | 174 | trace_xfs_file_fsync(ip); |
fd3200be | 175 | |
02c24a82 JB |
176 | error = filemap_write_and_wait_range(inode->i_mapping, start, end); |
177 | if (error) | |
178 | return error; | |
179 | ||
a27a263b | 180 | if (XFS_FORCED_SHUTDOWN(mp)) |
b474c7ae | 181 | return -EIO; |
fd3200be CH |
182 | |
183 | xfs_iflags_clear(ip, XFS_ITRUNCATED); | |
184 | ||
a27a263b CH |
185 | if (mp->m_flags & XFS_MOUNT_BARRIER) { |
186 | /* | |
187 | * If we have an RT and/or log subvolume we need to make sure | |
188 | * to flush the write cache the device used for file data | |
189 | * first. This is to ensure newly written file data make | |
190 | * it to disk before logging the new inode size in case of | |
191 | * an extending write. | |
192 | */ | |
193 | if (XFS_IS_REALTIME_INODE(ip)) | |
194 | xfs_blkdev_issue_flush(mp->m_rtdev_targp); | |
195 | else if (mp->m_logdev_targp != mp->m_ddev_targp) | |
196 | xfs_blkdev_issue_flush(mp->m_ddev_targp); | |
197 | } | |
198 | ||
fd3200be | 199 | /* |
fc0561ce DC |
200 | * All metadata updates are logged, which means that we just have to |
201 | * flush the log up to the latest LSN that touched the inode. If we have | |
202 | * concurrent fsync/fdatasync() calls, we need them to all block on the | |
203 | * log force before we clear the ili_fsync_fields field. This ensures | |
204 | * that we don't get a racing sync operation that does not wait for the | |
205 | * metadata to hit the journal before returning. If we race with | |
206 | * clearing the ili_fsync_fields, then all that will happen is the log | |
207 | * force will do nothing as the lsn will already be on disk. We can't | |
208 | * race with setting ili_fsync_fields because that is done under | |
209 | * XFS_ILOCK_EXCL, and that can't happen because we hold the lock shared | |
210 | * until after the ili_fsync_fields is cleared. | |
fd3200be CH |
211 | */ |
212 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
8f639dde CH |
213 | if (xfs_ipincount(ip)) { |
214 | if (!datasync || | |
fc0561ce | 215 | (ip->i_itemp->ili_fsync_fields & ~XFS_ILOG_TIMESTAMP)) |
8f639dde CH |
216 | lsn = ip->i_itemp->ili_last_lsn; |
217 | } | |
fd3200be | 218 | |
fc0561ce | 219 | if (lsn) { |
b1037058 | 220 | error = _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, &log_flushed); |
fc0561ce DC |
221 | ip->i_itemp->ili_fsync_fields = 0; |
222 | } | |
223 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
b1037058 | 224 | |
a27a263b CH |
225 | /* |
226 | * If we only have a single device, and the log force about was | |
227 | * a no-op we might have to flush the data device cache here. | |
228 | * This can only happen for fdatasync/O_DSYNC if we were overwriting | |
229 | * an already allocated file and thus do not have any metadata to | |
230 | * commit. | |
231 | */ | |
232 | if ((mp->m_flags & XFS_MOUNT_BARRIER) && | |
233 | mp->m_logdev_targp == mp->m_ddev_targp && | |
234 | !XFS_IS_REALTIME_INODE(ip) && | |
235 | !log_flushed) | |
236 | xfs_blkdev_issue_flush(mp->m_ddev_targp); | |
fd3200be | 237 | |
2451337d | 238 | return error; |
fd3200be CH |
239 | } |
240 | ||
00258e36 | 241 | STATIC ssize_t |
bbc5a740 | 242 | xfs_file_dio_aio_read( |
dda35b8f | 243 | struct kiocb *iocb, |
b4f5d2c6 | 244 | struct iov_iter *to) |
dda35b8f | 245 | { |
bbc5a740 CH |
246 | struct address_space *mapping = iocb->ki_filp->f_mapping; |
247 | struct inode *inode = mapping->host; | |
00258e36 | 248 | struct xfs_inode *ip = XFS_I(inode); |
f1285ff0 | 249 | loff_t isize = i_size_read(inode); |
bbc5a740 | 250 | size_t count = iov_iter_count(to); |
f1285ff0 | 251 | struct iov_iter data; |
bbc5a740 | 252 | struct xfs_buftarg *target; |
dda35b8f | 253 | ssize_t ret = 0; |
dda35b8f | 254 | |
bbc5a740 | 255 | trace_xfs_file_direct_read(ip, count, iocb->ki_pos); |
dda35b8f | 256 | |
f1285ff0 CH |
257 | if (!count) |
258 | return 0; /* skip atime */ | |
dda35b8f | 259 | |
bbc5a740 CH |
260 | if (XFS_IS_REALTIME_INODE(ip)) |
261 | target = ip->i_mount->m_rtdev_targp; | |
262 | else | |
263 | target = ip->i_mount->m_ddev_targp; | |
dda35b8f | 264 | |
16d4d435 CH |
265 | /* DIO must be aligned to device logical sector size */ |
266 | if ((iocb->ki_pos | count) & target->bt_logical_sectormask) { | |
267 | if (iocb->ki_pos == isize) | |
268 | return 0; | |
269 | return -EINVAL; | |
dda35b8f | 270 | } |
dda35b8f | 271 | |
0c38a251 | 272 | /* |
3d751af2 BF |
273 | * Locking is a bit tricky here. If we take an exclusive lock for direct |
274 | * IO, we effectively serialise all new concurrent read IO to this file | |
275 | * and block it behind IO that is currently in progress because IO in | |
276 | * progress holds the IO lock shared. We only need to hold the lock | |
277 | * exclusive to blow away the page cache, so only take lock exclusively | |
278 | * if the page cache needs invalidation. This allows the normal direct | |
279 | * IO case of no page cache pages to proceeed concurrently without | |
280 | * serialisation. | |
0c38a251 DC |
281 | */ |
282 | xfs_rw_ilock(ip, XFS_IOLOCK_SHARED); | |
bbc5a740 | 283 | if (mapping->nrpages) { |
0c38a251 | 284 | xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED); |
487f84f3 DC |
285 | xfs_rw_ilock(ip, XFS_IOLOCK_EXCL); |
286 | ||
3d751af2 BF |
287 | /* |
288 | * The generic dio code only flushes the range of the particular | |
289 | * I/O. Because we take an exclusive lock here, this whole | |
290 | * sequence is considerably more expensive for us. This has a | |
291 | * noticeable performance impact for any file with cached pages, | |
292 | * even when outside of the range of the particular I/O. | |
293 | * | |
294 | * Hence, amortize the cost of the lock against a full file | |
295 | * flush and reduce the chances of repeated iolock cycles going | |
296 | * forward. | |
297 | */ | |
bbc5a740 CH |
298 | if (mapping->nrpages) { |
299 | ret = filemap_write_and_wait(mapping); | |
487f84f3 DC |
300 | if (ret) { |
301 | xfs_rw_iunlock(ip, XFS_IOLOCK_EXCL); | |
302 | return ret; | |
303 | } | |
85e584da CM |
304 | |
305 | /* | |
306 | * Invalidate whole pages. This can return an error if | |
307 | * we fail to invalidate a page, but this should never | |
308 | * happen on XFS. Warn if it does fail. | |
309 | */ | |
bbc5a740 | 310 | ret = invalidate_inode_pages2(mapping); |
85e584da CM |
311 | WARN_ON_ONCE(ret); |
312 | ret = 0; | |
00258e36 | 313 | } |
487f84f3 | 314 | xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL); |
0c38a251 | 315 | } |
dda35b8f | 316 | |
f1285ff0 | 317 | data = *to; |
16d4d435 CH |
318 | ret = __blockdev_direct_IO(iocb, inode, target->bt_bdev, &data, |
319 | xfs_get_blocks_direct, NULL, NULL, 0); | |
320 | if (ret > 0) { | |
321 | iocb->ki_pos += ret; | |
322 | iov_iter_advance(to, ret); | |
fa8d972d | 323 | } |
16d4d435 | 324 | xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED); |
dda35b8f | 325 | |
16d4d435 CH |
326 | file_accessed(iocb->ki_filp); |
327 | return ret; | |
328 | } | |
329 | ||
f021bd07 | 330 | static noinline ssize_t |
16d4d435 CH |
331 | xfs_file_dax_read( |
332 | struct kiocb *iocb, | |
333 | struct iov_iter *to) | |
334 | { | |
335 | struct address_space *mapping = iocb->ki_filp->f_mapping; | |
336 | struct inode *inode = mapping->host; | |
337 | struct xfs_inode *ip = XFS_I(inode); | |
338 | struct iov_iter data = *to; | |
339 | size_t count = iov_iter_count(to); | |
340 | ssize_t ret = 0; | |
341 | ||
342 | trace_xfs_file_dax_read(ip, count, iocb->ki_pos); | |
343 | ||
344 | if (!count) | |
345 | return 0; /* skip atime */ | |
346 | ||
347 | xfs_rw_ilock(ip, XFS_IOLOCK_SHARED); | |
348 | ret = dax_do_io(iocb, inode, &data, xfs_get_blocks_direct, NULL, 0); | |
f1285ff0 CH |
349 | if (ret > 0) { |
350 | iocb->ki_pos += ret; | |
351 | iov_iter_advance(to, ret); | |
352 | } | |
bbc5a740 CH |
353 | xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED); |
354 | ||
f1285ff0 | 355 | file_accessed(iocb->ki_filp); |
bbc5a740 CH |
356 | return ret; |
357 | } | |
358 | ||
359 | STATIC ssize_t | |
360 | xfs_file_buffered_aio_read( | |
361 | struct kiocb *iocb, | |
362 | struct iov_iter *to) | |
363 | { | |
364 | struct xfs_inode *ip = XFS_I(file_inode(iocb->ki_filp)); | |
365 | ssize_t ret; | |
366 | ||
367 | trace_xfs_file_buffered_read(ip, iov_iter_count(to), iocb->ki_pos); | |
dda35b8f | 368 | |
bbc5a740 | 369 | xfs_rw_ilock(ip, XFS_IOLOCK_SHARED); |
b4f5d2c6 | 370 | ret = generic_file_read_iter(iocb, to); |
bbc5a740 CH |
371 | xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED); |
372 | ||
373 | return ret; | |
374 | } | |
375 | ||
376 | STATIC ssize_t | |
377 | xfs_file_read_iter( | |
378 | struct kiocb *iocb, | |
379 | struct iov_iter *to) | |
380 | { | |
16d4d435 CH |
381 | struct inode *inode = file_inode(iocb->ki_filp); |
382 | struct xfs_mount *mp = XFS_I(inode)->i_mount; | |
bbc5a740 CH |
383 | ssize_t ret = 0; |
384 | ||
385 | XFS_STATS_INC(mp, xs_read_calls); | |
386 | ||
387 | if (XFS_FORCED_SHUTDOWN(mp)) | |
388 | return -EIO; | |
389 | ||
16d4d435 CH |
390 | if (IS_DAX(inode)) |
391 | ret = xfs_file_dax_read(iocb, to); | |
392 | else if (iocb->ki_flags & IOCB_DIRECT) | |
bbc5a740 | 393 | ret = xfs_file_dio_aio_read(iocb, to); |
3176c3e0 | 394 | else |
bbc5a740 | 395 | ret = xfs_file_buffered_aio_read(iocb, to); |
dda35b8f | 396 | |
dda35b8f | 397 | if (ret > 0) |
ff6d6af2 | 398 | XFS_STATS_ADD(mp, xs_read_bytes, ret); |
dda35b8f CH |
399 | return ret; |
400 | } | |
401 | ||
00258e36 CH |
402 | STATIC ssize_t |
403 | xfs_file_splice_read( | |
dda35b8f CH |
404 | struct file *infilp, |
405 | loff_t *ppos, | |
406 | struct pipe_inode_info *pipe, | |
407 | size_t count, | |
00258e36 | 408 | unsigned int flags) |
dda35b8f | 409 | { |
00258e36 | 410 | struct xfs_inode *ip = XFS_I(infilp->f_mapping->host); |
dda35b8f CH |
411 | ssize_t ret; |
412 | ||
ff6d6af2 | 413 | XFS_STATS_INC(ip->i_mount, xs_read_calls); |
00258e36 | 414 | |
dda35b8f CH |
415 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
416 | return -EIO; | |
417 | ||
3176c3e0 | 418 | trace_xfs_file_splice_read(ip, count, *ppos); |
dda35b8f | 419 | |
a6d7636e DC |
420 | /* |
421 | * DAX inodes cannot ues the page cache for splice, so we have to push | |
422 | * them through the VFS IO path. This means it goes through | |
423 | * ->read_iter, which for us takes the XFS_IOLOCK_SHARED. Hence we | |
424 | * cannot lock the splice operation at this level for DAX inodes. | |
425 | */ | |
426 | if (IS_DAX(VFS_I(ip))) { | |
427 | ret = default_file_splice_read(infilp, ppos, pipe, count, | |
428 | flags); | |
429 | goto out; | |
430 | } | |
dda35b8f | 431 | |
a6d7636e DC |
432 | xfs_rw_ilock(ip, XFS_IOLOCK_SHARED); |
433 | ret = generic_file_splice_read(infilp, ppos, pipe, count, flags); | |
487f84f3 | 434 | xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED); |
a6d7636e DC |
435 | out: |
436 | if (ret > 0) | |
437 | XFS_STATS_ADD(ip->i_mount, xs_read_bytes, ret); | |
dda35b8f CH |
438 | return ret; |
439 | } | |
440 | ||
dda35b8f | 441 | /* |
193aec10 CH |
442 | * Zero any on disk space between the current EOF and the new, larger EOF. |
443 | * | |
444 | * This handles the normal case of zeroing the remainder of the last block in | |
445 | * the file and the unusual case of zeroing blocks out beyond the size of the | |
446 | * file. This second case only happens with fixed size extents and when the | |
447 | * system crashes before the inode size was updated but after blocks were | |
448 | * allocated. | |
449 | * | |
450 | * Expects the iolock to be held exclusive, and will take the ilock internally. | |
dda35b8f | 451 | */ |
dda35b8f CH |
452 | int /* error (positive) */ |
453 | xfs_zero_eof( | |
193aec10 CH |
454 | struct xfs_inode *ip, |
455 | xfs_off_t offset, /* starting I/O offset */ | |
5885ebda DC |
456 | xfs_fsize_t isize, /* current inode size */ |
457 | bool *did_zeroing) | |
dda35b8f | 458 | { |
193aec10 | 459 | ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL)); |
dda35b8f CH |
460 | ASSERT(offset > isize); |
461 | ||
0a50f162 | 462 | trace_xfs_zero_eof(ip, isize, offset - isize); |
570b6211 | 463 | return xfs_zero_range(ip, isize, offset - isize, did_zeroing); |
dda35b8f CH |
464 | } |
465 | ||
4d8d1581 DC |
466 | /* |
467 | * Common pre-write limit and setup checks. | |
468 | * | |
5bf1f262 CH |
469 | * Called with the iolocked held either shared and exclusive according to |
470 | * @iolock, and returns with it held. Might upgrade the iolock to exclusive | |
471 | * if called for a direct write beyond i_size. | |
4d8d1581 DC |
472 | */ |
473 | STATIC ssize_t | |
474 | xfs_file_aio_write_checks( | |
99733fa3 AV |
475 | struct kiocb *iocb, |
476 | struct iov_iter *from, | |
4d8d1581 DC |
477 | int *iolock) |
478 | { | |
99733fa3 | 479 | struct file *file = iocb->ki_filp; |
4d8d1581 DC |
480 | struct inode *inode = file->f_mapping->host; |
481 | struct xfs_inode *ip = XFS_I(inode); | |
3309dd04 | 482 | ssize_t error = 0; |
99733fa3 | 483 | size_t count = iov_iter_count(from); |
3136e8bb | 484 | bool drained_dio = false; |
4d8d1581 | 485 | |
7271d243 | 486 | restart: |
3309dd04 AV |
487 | error = generic_write_checks(iocb, from); |
488 | if (error <= 0) | |
4d8d1581 | 489 | return error; |
4d8d1581 | 490 | |
21c3ea18 | 491 | error = xfs_break_layouts(inode, iolock, true); |
781355c6 CH |
492 | if (error) |
493 | return error; | |
494 | ||
a6de82ca JK |
495 | /* For changing security info in file_remove_privs() we need i_mutex */ |
496 | if (*iolock == XFS_IOLOCK_SHARED && !IS_NOSEC(inode)) { | |
497 | xfs_rw_iunlock(ip, *iolock); | |
498 | *iolock = XFS_IOLOCK_EXCL; | |
499 | xfs_rw_ilock(ip, *iolock); | |
500 | goto restart; | |
501 | } | |
4d8d1581 DC |
502 | /* |
503 | * If the offset is beyond the size of the file, we need to zero any | |
504 | * blocks that fall between the existing EOF and the start of this | |
2813d682 | 505 | * write. If zeroing is needed and we are currently holding the |
467f7899 CH |
506 | * iolock shared, we need to update it to exclusive which implies |
507 | * having to redo all checks before. | |
b9d59846 DC |
508 | * |
509 | * We need to serialise against EOF updates that occur in IO | |
510 | * completions here. We want to make sure that nobody is changing the | |
511 | * size while we do this check until we have placed an IO barrier (i.e. | |
512 | * hold the XFS_IOLOCK_EXCL) that prevents new IO from being dispatched. | |
513 | * The spinlock effectively forms a memory barrier once we have the | |
514 | * XFS_IOLOCK_EXCL so we are guaranteed to see the latest EOF value | |
515 | * and hence be able to correctly determine if we need to run zeroing. | |
4d8d1581 | 516 | */ |
b9d59846 | 517 | spin_lock(&ip->i_flags_lock); |
99733fa3 | 518 | if (iocb->ki_pos > i_size_read(inode)) { |
5885ebda DC |
519 | bool zero = false; |
520 | ||
b9d59846 | 521 | spin_unlock(&ip->i_flags_lock); |
3136e8bb BF |
522 | if (!drained_dio) { |
523 | if (*iolock == XFS_IOLOCK_SHARED) { | |
524 | xfs_rw_iunlock(ip, *iolock); | |
525 | *iolock = XFS_IOLOCK_EXCL; | |
526 | xfs_rw_ilock(ip, *iolock); | |
527 | iov_iter_reexpand(from, count); | |
528 | } | |
40c63fbc DC |
529 | /* |
530 | * We now have an IO submission barrier in place, but | |
531 | * AIO can do EOF updates during IO completion and hence | |
532 | * we now need to wait for all of them to drain. Non-AIO | |
533 | * DIO will have drained before we are given the | |
534 | * XFS_IOLOCK_EXCL, and so for most cases this wait is a | |
535 | * no-op. | |
536 | */ | |
537 | inode_dio_wait(inode); | |
3136e8bb | 538 | drained_dio = true; |
7271d243 DC |
539 | goto restart; |
540 | } | |
99733fa3 | 541 | error = xfs_zero_eof(ip, iocb->ki_pos, i_size_read(inode), &zero); |
467f7899 CH |
542 | if (error) |
543 | return error; | |
b9d59846 DC |
544 | } else |
545 | spin_unlock(&ip->i_flags_lock); | |
4d8d1581 | 546 | |
8a9c9980 CH |
547 | /* |
548 | * Updating the timestamps will grab the ilock again from | |
549 | * xfs_fs_dirty_inode, so we have to call it after dropping the | |
550 | * lock above. Eventually we should look into a way to avoid | |
551 | * the pointless lock roundtrip. | |
552 | */ | |
c3b2da31 JB |
553 | if (likely(!(file->f_mode & FMODE_NOCMTIME))) { |
554 | error = file_update_time(file); | |
555 | if (error) | |
556 | return error; | |
557 | } | |
8a9c9980 | 558 | |
4d8d1581 DC |
559 | /* |
560 | * If we're writing the file then make sure to clear the setuid and | |
561 | * setgid bits if the process is not being run by root. This keeps | |
562 | * people from modifying setuid and setgid binaries. | |
563 | */ | |
a6de82ca JK |
564 | if (!IS_NOSEC(inode)) |
565 | return file_remove_privs(file); | |
566 | return 0; | |
4d8d1581 DC |
567 | } |
568 | ||
f0d26e86 DC |
569 | /* |
570 | * xfs_file_dio_aio_write - handle direct IO writes | |
571 | * | |
572 | * Lock the inode appropriately to prepare for and issue a direct IO write. | |
eda77982 | 573 | * By separating it from the buffered write path we remove all the tricky to |
f0d26e86 DC |
574 | * follow locking changes and looping. |
575 | * | |
eda77982 DC |
576 | * If there are cached pages or we're extending the file, we need IOLOCK_EXCL |
577 | * until we're sure the bytes at the new EOF have been zeroed and/or the cached | |
578 | * pages are flushed out. | |
579 | * | |
580 | * In most cases the direct IO writes will be done holding IOLOCK_SHARED | |
581 | * allowing them to be done in parallel with reads and other direct IO writes. | |
582 | * However, if the IO is not aligned to filesystem blocks, the direct IO layer | |
583 | * needs to do sub-block zeroing and that requires serialisation against other | |
584 | * direct IOs to the same block. In this case we need to serialise the | |
585 | * submission of the unaligned IOs so that we don't get racing block zeroing in | |
586 | * the dio layer. To avoid the problem with aio, we also need to wait for | |
587 | * outstanding IOs to complete so that unwritten extent conversion is completed | |
588 | * before we try to map the overlapping block. This is currently implemented by | |
4a06fd26 | 589 | * hitting it with a big hammer (i.e. inode_dio_wait()). |
eda77982 | 590 | * |
f0d26e86 DC |
591 | * Returns with locks held indicated by @iolock and errors indicated by |
592 | * negative return values. | |
593 | */ | |
594 | STATIC ssize_t | |
595 | xfs_file_dio_aio_write( | |
596 | struct kiocb *iocb, | |
b3188919 | 597 | struct iov_iter *from) |
f0d26e86 DC |
598 | { |
599 | struct file *file = iocb->ki_filp; | |
600 | struct address_space *mapping = file->f_mapping; | |
601 | struct inode *inode = mapping->host; | |
602 | struct xfs_inode *ip = XFS_I(inode); | |
603 | struct xfs_mount *mp = ip->i_mount; | |
604 | ssize_t ret = 0; | |
eda77982 | 605 | int unaligned_io = 0; |
d0606464 | 606 | int iolock; |
b3188919 | 607 | size_t count = iov_iter_count(from); |
0cefb29e DC |
608 | loff_t end; |
609 | struct iov_iter data; | |
f0d26e86 DC |
610 | struct xfs_buftarg *target = XFS_IS_REALTIME_INODE(ip) ? |
611 | mp->m_rtdev_targp : mp->m_ddev_targp; | |
612 | ||
7c71ee78 | 613 | /* DIO must be aligned to device logical sector size */ |
16d4d435 | 614 | if ((iocb->ki_pos | count) & target->bt_logical_sectormask) |
b474c7ae | 615 | return -EINVAL; |
f0d26e86 | 616 | |
7c71ee78 | 617 | /* "unaligned" here means not aligned to a filesystem block */ |
13712713 CH |
618 | if ((iocb->ki_pos & mp->m_blockmask) || |
619 | ((iocb->ki_pos + count) & mp->m_blockmask)) | |
eda77982 DC |
620 | unaligned_io = 1; |
621 | ||
7271d243 DC |
622 | /* |
623 | * We don't need to take an exclusive lock unless there page cache needs | |
624 | * to be invalidated or unaligned IO is being executed. We don't need to | |
625 | * consider the EOF extension case here because | |
626 | * xfs_file_aio_write_checks() will relock the inode as necessary for | |
627 | * EOF zeroing cases and fill out the new inode size as appropriate. | |
628 | */ | |
629 | if (unaligned_io || mapping->nrpages) | |
d0606464 | 630 | iolock = XFS_IOLOCK_EXCL; |
f0d26e86 | 631 | else |
d0606464 CH |
632 | iolock = XFS_IOLOCK_SHARED; |
633 | xfs_rw_ilock(ip, iolock); | |
c58cb165 CH |
634 | |
635 | /* | |
636 | * Recheck if there are cached pages that need invalidate after we got | |
637 | * the iolock to protect against other threads adding new pages while | |
638 | * we were waiting for the iolock. | |
639 | */ | |
d0606464 CH |
640 | if (mapping->nrpages && iolock == XFS_IOLOCK_SHARED) { |
641 | xfs_rw_iunlock(ip, iolock); | |
642 | iolock = XFS_IOLOCK_EXCL; | |
643 | xfs_rw_ilock(ip, iolock); | |
c58cb165 | 644 | } |
f0d26e86 | 645 | |
99733fa3 | 646 | ret = xfs_file_aio_write_checks(iocb, from, &iolock); |
4d8d1581 | 647 | if (ret) |
d0606464 | 648 | goto out; |
99733fa3 | 649 | count = iov_iter_count(from); |
13712713 | 650 | end = iocb->ki_pos + count - 1; |
f0d26e86 | 651 | |
3d751af2 | 652 | /* |
bbc5a740 | 653 | * See xfs_file_dio_aio_read() for why we do a full-file flush here. |
3d751af2 | 654 | */ |
f0d26e86 | 655 | if (mapping->nrpages) { |
3d751af2 | 656 | ret = filemap_write_and_wait(VFS_I(ip)->i_mapping); |
f0d26e86 | 657 | if (ret) |
d0606464 | 658 | goto out; |
834ffca6 | 659 | /* |
3d751af2 BF |
660 | * Invalidate whole pages. This can return an error if we fail |
661 | * to invalidate a page, but this should never happen on XFS. | |
662 | * Warn if it does fail. | |
834ffca6 | 663 | */ |
3d751af2 | 664 | ret = invalidate_inode_pages2(VFS_I(ip)->i_mapping); |
834ffca6 DC |
665 | WARN_ON_ONCE(ret); |
666 | ret = 0; | |
f0d26e86 DC |
667 | } |
668 | ||
eda77982 DC |
669 | /* |
670 | * If we are doing unaligned IO, wait for all other IO to drain, | |
671 | * otherwise demote the lock if we had to flush cached pages | |
672 | */ | |
673 | if (unaligned_io) | |
4a06fd26 | 674 | inode_dio_wait(inode); |
d0606464 | 675 | else if (iolock == XFS_IOLOCK_EXCL) { |
f0d26e86 | 676 | xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL); |
d0606464 | 677 | iolock = XFS_IOLOCK_SHARED; |
f0d26e86 DC |
678 | } |
679 | ||
3176c3e0 | 680 | trace_xfs_file_direct_write(ip, count, iocb->ki_pos); |
f0d26e86 | 681 | |
0cefb29e | 682 | data = *from; |
16d4d435 CH |
683 | ret = __blockdev_direct_IO(iocb, inode, target->bt_bdev, &data, |
684 | xfs_get_blocks_direct, xfs_end_io_direct_write, | |
685 | NULL, DIO_ASYNC_EXTEND); | |
0cefb29e DC |
686 | |
687 | /* see generic_file_direct_write() for why this is necessary */ | |
688 | if (mapping->nrpages) { | |
689 | invalidate_inode_pages2_range(mapping, | |
13712713 | 690 | iocb->ki_pos >> PAGE_SHIFT, |
09cbfeaf | 691 | end >> PAGE_SHIFT); |
0cefb29e DC |
692 | } |
693 | ||
694 | if (ret > 0) { | |
13712713 | 695 | iocb->ki_pos += ret; |
0cefb29e | 696 | iov_iter_advance(from, ret); |
0cefb29e | 697 | } |
d0606464 CH |
698 | out: |
699 | xfs_rw_iunlock(ip, iolock); | |
700 | ||
6b698ede | 701 | /* |
16d4d435 CH |
702 | * No fallback to buffered IO on errors for XFS, direct IO will either |
703 | * complete fully or fail. | |
6b698ede | 704 | */ |
16d4d435 CH |
705 | ASSERT(ret < 0 || ret == count); |
706 | return ret; | |
707 | } | |
708 | ||
f021bd07 | 709 | static noinline ssize_t |
16d4d435 CH |
710 | xfs_file_dax_write( |
711 | struct kiocb *iocb, | |
712 | struct iov_iter *from) | |
713 | { | |
714 | struct address_space *mapping = iocb->ki_filp->f_mapping; | |
715 | struct inode *inode = mapping->host; | |
716 | struct xfs_inode *ip = XFS_I(inode); | |
717 | struct xfs_mount *mp = ip->i_mount; | |
718 | ssize_t ret = 0; | |
719 | int unaligned_io = 0; | |
720 | int iolock; | |
721 | struct iov_iter data; | |
722 | ||
723 | /* "unaligned" here means not aligned to a filesystem block */ | |
724 | if ((iocb->ki_pos & mp->m_blockmask) || | |
725 | ((iocb->ki_pos + iov_iter_count(from)) & mp->m_blockmask)) { | |
726 | unaligned_io = 1; | |
727 | iolock = XFS_IOLOCK_EXCL; | |
728 | } else if (mapping->nrpages) { | |
729 | iolock = XFS_IOLOCK_EXCL; | |
730 | } else { | |
731 | iolock = XFS_IOLOCK_SHARED; | |
732 | } | |
733 | xfs_rw_ilock(ip, iolock); | |
734 | ||
735 | ret = xfs_file_aio_write_checks(iocb, from, &iolock); | |
736 | if (ret) | |
737 | goto out; | |
738 | ||
739 | /* | |
740 | * Yes, even DAX files can have page cache attached to them: A zeroed | |
741 | * page is inserted into the pagecache when we have to serve a write | |
742 | * fault on a hole. It should never be dirtied and can simply be | |
743 | * dropped from the pagecache once we get real data for the page. | |
8b2180b3 DC |
744 | * |
745 | * XXX: This is racy against mmap, and there's nothing we can do about | |
746 | * it. dax_do_io() should really do this invalidation internally as | |
747 | * it will know if we've allocated over a holei for this specific IO and | |
748 | * if so it needs to update the mapping tree and invalidate existing | |
749 | * PTEs over the newly allocated range. Remove this invalidation when | |
750 | * dax_do_io() is fixed up. | |
6b698ede | 751 | */ |
16d4d435 | 752 | if (mapping->nrpages) { |
8b2180b3 DC |
753 | loff_t end = iocb->ki_pos + iov_iter_count(from) - 1; |
754 | ||
755 | ret = invalidate_inode_pages2_range(mapping, | |
756 | iocb->ki_pos >> PAGE_SHIFT, | |
757 | end >> PAGE_SHIFT); | |
16d4d435 CH |
758 | WARN_ON_ONCE(ret); |
759 | } | |
760 | ||
761 | if (iolock == XFS_IOLOCK_EXCL && !unaligned_io) { | |
762 | xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL); | |
763 | iolock = XFS_IOLOCK_SHARED; | |
764 | } | |
765 | ||
766 | trace_xfs_file_dax_write(ip, iov_iter_count(from), iocb->ki_pos); | |
767 | ||
768 | data = *from; | |
769 | ret = dax_do_io(iocb, inode, &data, xfs_get_blocks_direct, | |
770 | xfs_end_io_direct_write, 0); | |
771 | if (ret > 0) { | |
772 | iocb->ki_pos += ret; | |
773 | iov_iter_advance(from, ret); | |
774 | } | |
775 | out: | |
776 | xfs_rw_iunlock(ip, iolock); | |
f0d26e86 DC |
777 | return ret; |
778 | } | |
779 | ||
00258e36 | 780 | STATIC ssize_t |
637bbc75 | 781 | xfs_file_buffered_aio_write( |
dda35b8f | 782 | struct kiocb *iocb, |
b3188919 | 783 | struct iov_iter *from) |
dda35b8f CH |
784 | { |
785 | struct file *file = iocb->ki_filp; | |
786 | struct address_space *mapping = file->f_mapping; | |
787 | struct inode *inode = mapping->host; | |
00258e36 | 788 | struct xfs_inode *ip = XFS_I(inode); |
637bbc75 DC |
789 | ssize_t ret; |
790 | int enospc = 0; | |
d0606464 | 791 | int iolock = XFS_IOLOCK_EXCL; |
dda35b8f | 792 | |
d0606464 | 793 | xfs_rw_ilock(ip, iolock); |
dda35b8f | 794 | |
99733fa3 | 795 | ret = xfs_file_aio_write_checks(iocb, from, &iolock); |
4d8d1581 | 796 | if (ret) |
d0606464 | 797 | goto out; |
dda35b8f CH |
798 | |
799 | /* We can write back this queue in page reclaim */ | |
de1414a6 | 800 | current->backing_dev_info = inode_to_bdi(inode); |
dda35b8f | 801 | |
dda35b8f | 802 | write_retry: |
3176c3e0 | 803 | trace_xfs_file_buffered_write(ip, iov_iter_count(from), iocb->ki_pos); |
68a9f5e7 | 804 | ret = iomap_file_buffered_write(iocb, from, &xfs_iomap_ops); |
0a64bc2c | 805 | if (likely(ret >= 0)) |
99733fa3 | 806 | iocb->ki_pos += ret; |
dc06f398 | 807 | |
637bbc75 | 808 | /* |
dc06f398 BF |
809 | * If we hit a space limit, try to free up some lingering preallocated |
810 | * space before returning an error. In the case of ENOSPC, first try to | |
811 | * write back all dirty inodes to free up some of the excess reserved | |
812 | * metadata space. This reduces the chances that the eofblocks scan | |
813 | * waits on dirty mappings. Since xfs_flush_inodes() is serialized, this | |
814 | * also behaves as a filter to prevent too many eofblocks scans from | |
815 | * running at the same time. | |
637bbc75 | 816 | */ |
dc06f398 BF |
817 | if (ret == -EDQUOT && !enospc) { |
818 | enospc = xfs_inode_free_quota_eofblocks(ip); | |
819 | if (enospc) | |
820 | goto write_retry; | |
821 | } else if (ret == -ENOSPC && !enospc) { | |
822 | struct xfs_eofblocks eofb = {0}; | |
823 | ||
637bbc75 | 824 | enospc = 1; |
9aa05000 | 825 | xfs_flush_inodes(ip->i_mount); |
dc06f398 BF |
826 | eofb.eof_scan_owner = ip->i_ino; /* for locking */ |
827 | eofb.eof_flags = XFS_EOF_FLAGS_SYNC; | |
828 | xfs_icache_free_eofblocks(ip->i_mount, &eofb); | |
9aa05000 | 829 | goto write_retry; |
dda35b8f | 830 | } |
d0606464 | 831 | |
dda35b8f | 832 | current->backing_dev_info = NULL; |
d0606464 CH |
833 | out: |
834 | xfs_rw_iunlock(ip, iolock); | |
637bbc75 DC |
835 | return ret; |
836 | } | |
837 | ||
838 | STATIC ssize_t | |
bf97f3bc | 839 | xfs_file_write_iter( |
637bbc75 | 840 | struct kiocb *iocb, |
bf97f3bc | 841 | struct iov_iter *from) |
637bbc75 DC |
842 | { |
843 | struct file *file = iocb->ki_filp; | |
844 | struct address_space *mapping = file->f_mapping; | |
845 | struct inode *inode = mapping->host; | |
846 | struct xfs_inode *ip = XFS_I(inode); | |
847 | ssize_t ret; | |
bf97f3bc | 848 | size_t ocount = iov_iter_count(from); |
637bbc75 | 849 | |
ff6d6af2 | 850 | XFS_STATS_INC(ip->i_mount, xs_write_calls); |
637bbc75 | 851 | |
637bbc75 DC |
852 | if (ocount == 0) |
853 | return 0; | |
854 | ||
bf97f3bc AV |
855 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
856 | return -EIO; | |
637bbc75 | 857 | |
16d4d435 CH |
858 | if (IS_DAX(inode)) |
859 | ret = xfs_file_dax_write(iocb, from); | |
860 | else if (iocb->ki_flags & IOCB_DIRECT) | |
bf97f3bc | 861 | ret = xfs_file_dio_aio_write(iocb, from); |
637bbc75 | 862 | else |
bf97f3bc | 863 | ret = xfs_file_buffered_aio_write(iocb, from); |
dda35b8f | 864 | |
d0606464 | 865 | if (ret > 0) { |
ff6d6af2 | 866 | XFS_STATS_ADD(ip->i_mount, xs_write_bytes, ret); |
dda35b8f | 867 | |
d0606464 | 868 | /* Handle various SYNC-type writes */ |
e2592217 | 869 | ret = generic_write_sync(iocb, ret); |
dda35b8f | 870 | } |
a363f0c2 | 871 | return ret; |
dda35b8f CH |
872 | } |
873 | ||
a904b1ca NJ |
874 | #define XFS_FALLOC_FL_SUPPORTED \ |
875 | (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \ | |
876 | FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE | \ | |
877 | FALLOC_FL_INSERT_RANGE) | |
878 | ||
2fe17c10 CH |
879 | STATIC long |
880 | xfs_file_fallocate( | |
83aee9e4 CH |
881 | struct file *file, |
882 | int mode, | |
883 | loff_t offset, | |
884 | loff_t len) | |
2fe17c10 | 885 | { |
83aee9e4 CH |
886 | struct inode *inode = file_inode(file); |
887 | struct xfs_inode *ip = XFS_I(inode); | |
83aee9e4 | 888 | long error; |
8add71ca | 889 | enum xfs_prealloc_flags flags = 0; |
781355c6 | 890 | uint iolock = XFS_IOLOCK_EXCL; |
83aee9e4 | 891 | loff_t new_size = 0; |
a904b1ca | 892 | bool do_file_insert = 0; |
2fe17c10 | 893 | |
83aee9e4 CH |
894 | if (!S_ISREG(inode->i_mode)) |
895 | return -EINVAL; | |
a904b1ca | 896 | if (mode & ~XFS_FALLOC_FL_SUPPORTED) |
2fe17c10 CH |
897 | return -EOPNOTSUPP; |
898 | ||
781355c6 | 899 | xfs_ilock(ip, iolock); |
21c3ea18 | 900 | error = xfs_break_layouts(inode, &iolock, false); |
781355c6 CH |
901 | if (error) |
902 | goto out_unlock; | |
903 | ||
e8e9ad42 DC |
904 | xfs_ilock(ip, XFS_MMAPLOCK_EXCL); |
905 | iolock |= XFS_MMAPLOCK_EXCL; | |
906 | ||
83aee9e4 CH |
907 | if (mode & FALLOC_FL_PUNCH_HOLE) { |
908 | error = xfs_free_file_space(ip, offset, len); | |
909 | if (error) | |
910 | goto out_unlock; | |
e1d8fb88 NJ |
911 | } else if (mode & FALLOC_FL_COLLAPSE_RANGE) { |
912 | unsigned blksize_mask = (1 << inode->i_blkbits) - 1; | |
913 | ||
914 | if (offset & blksize_mask || len & blksize_mask) { | |
2451337d | 915 | error = -EINVAL; |
e1d8fb88 NJ |
916 | goto out_unlock; |
917 | } | |
918 | ||
23fffa92 LC |
919 | /* |
920 | * There is no need to overlap collapse range with EOF, | |
921 | * in which case it is effectively a truncate operation | |
922 | */ | |
923 | if (offset + len >= i_size_read(inode)) { | |
2451337d | 924 | error = -EINVAL; |
23fffa92 LC |
925 | goto out_unlock; |
926 | } | |
927 | ||
e1d8fb88 NJ |
928 | new_size = i_size_read(inode) - len; |
929 | ||
930 | error = xfs_collapse_file_space(ip, offset, len); | |
931 | if (error) | |
932 | goto out_unlock; | |
a904b1ca NJ |
933 | } else if (mode & FALLOC_FL_INSERT_RANGE) { |
934 | unsigned blksize_mask = (1 << inode->i_blkbits) - 1; | |
935 | ||
936 | new_size = i_size_read(inode) + len; | |
937 | if (offset & blksize_mask || len & blksize_mask) { | |
938 | error = -EINVAL; | |
939 | goto out_unlock; | |
940 | } | |
941 | ||
942 | /* check the new inode size does not wrap through zero */ | |
943 | if (new_size > inode->i_sb->s_maxbytes) { | |
944 | error = -EFBIG; | |
945 | goto out_unlock; | |
946 | } | |
947 | ||
948 | /* Offset should be less than i_size */ | |
949 | if (offset >= i_size_read(inode)) { | |
950 | error = -EINVAL; | |
951 | goto out_unlock; | |
952 | } | |
953 | do_file_insert = 1; | |
83aee9e4 | 954 | } else { |
8add71ca CH |
955 | flags |= XFS_PREALLOC_SET; |
956 | ||
83aee9e4 CH |
957 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
958 | offset + len > i_size_read(inode)) { | |
959 | new_size = offset + len; | |
2451337d | 960 | error = inode_newsize_ok(inode, new_size); |
83aee9e4 CH |
961 | if (error) |
962 | goto out_unlock; | |
963 | } | |
2fe17c10 | 964 | |
376ba313 LC |
965 | if (mode & FALLOC_FL_ZERO_RANGE) |
966 | error = xfs_zero_file_space(ip, offset, len); | |
967 | else | |
968 | error = xfs_alloc_file_space(ip, offset, len, | |
969 | XFS_BMAPI_PREALLOC); | |
2fe17c10 CH |
970 | if (error) |
971 | goto out_unlock; | |
972 | } | |
973 | ||
83aee9e4 | 974 | if (file->f_flags & O_DSYNC) |
8add71ca CH |
975 | flags |= XFS_PREALLOC_SYNC; |
976 | ||
977 | error = xfs_update_prealloc_flags(ip, flags); | |
2fe17c10 CH |
978 | if (error) |
979 | goto out_unlock; | |
980 | ||
981 | /* Change file size if needed */ | |
982 | if (new_size) { | |
983 | struct iattr iattr; | |
984 | ||
985 | iattr.ia_valid = ATTR_SIZE; | |
986 | iattr.ia_size = new_size; | |
83aee9e4 | 987 | error = xfs_setattr_size(ip, &iattr); |
a904b1ca NJ |
988 | if (error) |
989 | goto out_unlock; | |
2fe17c10 CH |
990 | } |
991 | ||
a904b1ca NJ |
992 | /* |
993 | * Perform hole insertion now that the file size has been | |
994 | * updated so that if we crash during the operation we don't | |
995 | * leave shifted extents past EOF and hence losing access to | |
996 | * the data that is contained within them. | |
997 | */ | |
998 | if (do_file_insert) | |
999 | error = xfs_insert_file_space(ip, offset, len); | |
1000 | ||
2fe17c10 | 1001 | out_unlock: |
781355c6 | 1002 | xfs_iunlock(ip, iolock); |
2451337d | 1003 | return error; |
2fe17c10 CH |
1004 | } |
1005 | ||
1006 | ||
1da177e4 | 1007 | STATIC int |
3562fd45 | 1008 | xfs_file_open( |
1da177e4 | 1009 | struct inode *inode, |
f999a5bf | 1010 | struct file *file) |
1da177e4 | 1011 | { |
f999a5bf | 1012 | if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS) |
1da177e4 | 1013 | return -EFBIG; |
f999a5bf CH |
1014 | if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb))) |
1015 | return -EIO; | |
1016 | return 0; | |
1017 | } | |
1018 | ||
1019 | STATIC int | |
1020 | xfs_dir_open( | |
1021 | struct inode *inode, | |
1022 | struct file *file) | |
1023 | { | |
1024 | struct xfs_inode *ip = XFS_I(inode); | |
1025 | int mode; | |
1026 | int error; | |
1027 | ||
1028 | error = xfs_file_open(inode, file); | |
1029 | if (error) | |
1030 | return error; | |
1031 | ||
1032 | /* | |
1033 | * If there are any blocks, read-ahead block 0 as we're almost | |
1034 | * certain to have the next operation be a read there. | |
1035 | */ | |
309ecac8 | 1036 | mode = xfs_ilock_data_map_shared(ip); |
f999a5bf | 1037 | if (ip->i_d.di_nextents > 0) |
9df2dd0b | 1038 | xfs_dir3_data_readahead(ip, 0, -1); |
f999a5bf CH |
1039 | xfs_iunlock(ip, mode); |
1040 | return 0; | |
1da177e4 LT |
1041 | } |
1042 | ||
1da177e4 | 1043 | STATIC int |
3562fd45 | 1044 | xfs_file_release( |
1da177e4 LT |
1045 | struct inode *inode, |
1046 | struct file *filp) | |
1047 | { | |
2451337d | 1048 | return xfs_release(XFS_I(inode)); |
1da177e4 LT |
1049 | } |
1050 | ||
1da177e4 | 1051 | STATIC int |
3562fd45 | 1052 | xfs_file_readdir( |
b8227554 AV |
1053 | struct file *file, |
1054 | struct dir_context *ctx) | |
1da177e4 | 1055 | { |
b8227554 | 1056 | struct inode *inode = file_inode(file); |
739bfb2a | 1057 | xfs_inode_t *ip = XFS_I(inode); |
051e7cd4 CH |
1058 | size_t bufsize; |
1059 | ||
1060 | /* | |
1061 | * The Linux API doesn't pass down the total size of the buffer | |
1062 | * we read into down to the filesystem. With the filldir concept | |
1063 | * it's not needed for correct information, but the XFS dir2 leaf | |
1064 | * code wants an estimate of the buffer size to calculate it's | |
1065 | * readahead window and size the buffers used for mapping to | |
1066 | * physical blocks. | |
1067 | * | |
1068 | * Try to give it an estimate that's good enough, maybe at some | |
1069 | * point we can change the ->readdir prototype to include the | |
a9cc799e | 1070 | * buffer size. For now we use the current glibc buffer size. |
051e7cd4 | 1071 | */ |
a9cc799e | 1072 | bufsize = (size_t)min_t(loff_t, 32768, ip->i_d.di_size); |
051e7cd4 | 1073 | |
8300475e | 1074 | return xfs_readdir(ip, ctx, bufsize); |
1da177e4 LT |
1075 | } |
1076 | ||
d126d43f JL |
1077 | /* |
1078 | * This type is designed to indicate the type of offset we would like | |
49c69591 | 1079 | * to search from page cache for xfs_seek_hole_data(). |
d126d43f JL |
1080 | */ |
1081 | enum { | |
1082 | HOLE_OFF = 0, | |
1083 | DATA_OFF, | |
1084 | }; | |
1085 | ||
1086 | /* | |
1087 | * Lookup the desired type of offset from the given page. | |
1088 | * | |
1089 | * On success, return true and the offset argument will point to the | |
1090 | * start of the region that was found. Otherwise this function will | |
1091 | * return false and keep the offset argument unchanged. | |
1092 | */ | |
1093 | STATIC bool | |
1094 | xfs_lookup_buffer_offset( | |
1095 | struct page *page, | |
1096 | loff_t *offset, | |
1097 | unsigned int type) | |
1098 | { | |
1099 | loff_t lastoff = page_offset(page); | |
1100 | bool found = false; | |
1101 | struct buffer_head *bh, *head; | |
1102 | ||
1103 | bh = head = page_buffers(page); | |
1104 | do { | |
1105 | /* | |
1106 | * Unwritten extents that have data in the page | |
1107 | * cache covering them can be identified by the | |
1108 | * BH_Unwritten state flag. Pages with multiple | |
1109 | * buffers might have a mix of holes, data and | |
1110 | * unwritten extents - any buffer with valid | |
1111 | * data in it should have BH_Uptodate flag set | |
1112 | * on it. | |
1113 | */ | |
1114 | if (buffer_unwritten(bh) || | |
1115 | buffer_uptodate(bh)) { | |
1116 | if (type == DATA_OFF) | |
1117 | found = true; | |
1118 | } else { | |
1119 | if (type == HOLE_OFF) | |
1120 | found = true; | |
1121 | } | |
1122 | ||
1123 | if (found) { | |
1124 | *offset = lastoff; | |
1125 | break; | |
1126 | } | |
1127 | lastoff += bh->b_size; | |
1128 | } while ((bh = bh->b_this_page) != head); | |
1129 | ||
1130 | return found; | |
1131 | } | |
1132 | ||
1133 | /* | |
1134 | * This routine is called to find out and return a data or hole offset | |
1135 | * from the page cache for unwritten extents according to the desired | |
49c69591 | 1136 | * type for xfs_seek_hole_data(). |
d126d43f JL |
1137 | * |
1138 | * The argument offset is used to tell where we start to search from the | |
1139 | * page cache. Map is used to figure out the end points of the range to | |
1140 | * lookup pages. | |
1141 | * | |
1142 | * Return true if the desired type of offset was found, and the argument | |
1143 | * offset is filled with that address. Otherwise, return false and keep | |
1144 | * offset unchanged. | |
1145 | */ | |
1146 | STATIC bool | |
1147 | xfs_find_get_desired_pgoff( | |
1148 | struct inode *inode, | |
1149 | struct xfs_bmbt_irec *map, | |
1150 | unsigned int type, | |
1151 | loff_t *offset) | |
1152 | { | |
1153 | struct xfs_inode *ip = XFS_I(inode); | |
1154 | struct xfs_mount *mp = ip->i_mount; | |
1155 | struct pagevec pvec; | |
1156 | pgoff_t index; | |
1157 | pgoff_t end; | |
1158 | loff_t endoff; | |
1159 | loff_t startoff = *offset; | |
1160 | loff_t lastoff = startoff; | |
1161 | bool found = false; | |
1162 | ||
1163 | pagevec_init(&pvec, 0); | |
1164 | ||
09cbfeaf | 1165 | index = startoff >> PAGE_SHIFT; |
d126d43f | 1166 | endoff = XFS_FSB_TO_B(mp, map->br_startoff + map->br_blockcount); |
09cbfeaf | 1167 | end = endoff >> PAGE_SHIFT; |
d126d43f JL |
1168 | do { |
1169 | int want; | |
1170 | unsigned nr_pages; | |
1171 | unsigned int i; | |
1172 | ||
1173 | want = min_t(pgoff_t, end - index, PAGEVEC_SIZE); | |
1174 | nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index, | |
1175 | want); | |
1176 | /* | |
1177 | * No page mapped into given range. If we are searching holes | |
1178 | * and if this is the first time we got into the loop, it means | |
1179 | * that the given offset is landed in a hole, return it. | |
1180 | * | |
1181 | * If we have already stepped through some block buffers to find | |
1182 | * holes but they all contains data. In this case, the last | |
1183 | * offset is already updated and pointed to the end of the last | |
1184 | * mapped page, if it does not reach the endpoint to search, | |
1185 | * that means there should be a hole between them. | |
1186 | */ | |
1187 | if (nr_pages == 0) { | |
1188 | /* Data search found nothing */ | |
1189 | if (type == DATA_OFF) | |
1190 | break; | |
1191 | ||
1192 | ASSERT(type == HOLE_OFF); | |
1193 | if (lastoff == startoff || lastoff < endoff) { | |
1194 | found = true; | |
1195 | *offset = lastoff; | |
1196 | } | |
1197 | break; | |
1198 | } | |
1199 | ||
1200 | /* | |
1201 | * At lease we found one page. If this is the first time we | |
1202 | * step into the loop, and if the first page index offset is | |
1203 | * greater than the given search offset, a hole was found. | |
1204 | */ | |
1205 | if (type == HOLE_OFF && lastoff == startoff && | |
1206 | lastoff < page_offset(pvec.pages[0])) { | |
1207 | found = true; | |
1208 | break; | |
1209 | } | |
1210 | ||
1211 | for (i = 0; i < nr_pages; i++) { | |
1212 | struct page *page = pvec.pages[i]; | |
1213 | loff_t b_offset; | |
1214 | ||
1215 | /* | |
1216 | * At this point, the page may be truncated or | |
1217 | * invalidated (changing page->mapping to NULL), | |
1218 | * or even swizzled back from swapper_space to tmpfs | |
1219 | * file mapping. However, page->index will not change | |
1220 | * because we have a reference on the page. | |
1221 | * | |
1222 | * Searching done if the page index is out of range. | |
1223 | * If the current offset is not reaches the end of | |
1224 | * the specified search range, there should be a hole | |
1225 | * between them. | |
1226 | */ | |
1227 | if (page->index > end) { | |
1228 | if (type == HOLE_OFF && lastoff < endoff) { | |
1229 | *offset = lastoff; | |
1230 | found = true; | |
1231 | } | |
1232 | goto out; | |
1233 | } | |
1234 | ||
1235 | lock_page(page); | |
1236 | /* | |
1237 | * Page truncated or invalidated(page->mapping == NULL). | |
1238 | * We can freely skip it and proceed to check the next | |
1239 | * page. | |
1240 | */ | |
1241 | if (unlikely(page->mapping != inode->i_mapping)) { | |
1242 | unlock_page(page); | |
1243 | continue; | |
1244 | } | |
1245 | ||
1246 | if (!page_has_buffers(page)) { | |
1247 | unlock_page(page); | |
1248 | continue; | |
1249 | } | |
1250 | ||
1251 | found = xfs_lookup_buffer_offset(page, &b_offset, type); | |
1252 | if (found) { | |
1253 | /* | |
1254 | * The found offset may be less than the start | |
1255 | * point to search if this is the first time to | |
1256 | * come here. | |
1257 | */ | |
1258 | *offset = max_t(loff_t, startoff, b_offset); | |
1259 | unlock_page(page); | |
1260 | goto out; | |
1261 | } | |
1262 | ||
1263 | /* | |
1264 | * We either searching data but nothing was found, or | |
1265 | * searching hole but found a data buffer. In either | |
1266 | * case, probably the next page contains the desired | |
1267 | * things, update the last offset to it so. | |
1268 | */ | |
1269 | lastoff = page_offset(page) + PAGE_SIZE; | |
1270 | unlock_page(page); | |
1271 | } | |
1272 | ||
1273 | /* | |
1274 | * The number of returned pages less than our desired, search | |
1275 | * done. In this case, nothing was found for searching data, | |
1276 | * but we found a hole behind the last offset. | |
1277 | */ | |
1278 | if (nr_pages < want) { | |
1279 | if (type == HOLE_OFF) { | |
1280 | *offset = lastoff; | |
1281 | found = true; | |
1282 | } | |
1283 | break; | |
1284 | } | |
1285 | ||
1286 | index = pvec.pages[i - 1]->index + 1; | |
1287 | pagevec_release(&pvec); | |
1288 | } while (index <= end); | |
1289 | ||
1290 | out: | |
1291 | pagevec_release(&pvec); | |
1292 | return found; | |
1293 | } | |
1294 | ||
8aa7d37e ES |
1295 | /* |
1296 | * caller must lock inode with xfs_ilock_data_map_shared, | |
1297 | * can we craft an appropriate ASSERT? | |
1298 | * | |
1299 | * end is because the VFS-level lseek interface is defined such that any | |
1300 | * offset past i_size shall return -ENXIO, but we use this for quota code | |
1301 | * which does not maintain i_size, and we want to SEEK_DATA past i_size. | |
1302 | */ | |
1303 | loff_t | |
1304 | __xfs_seek_hole_data( | |
1305 | struct inode *inode, | |
49c69591 | 1306 | loff_t start, |
8aa7d37e | 1307 | loff_t end, |
49c69591 | 1308 | int whence) |
3fe3e6b1 | 1309 | { |
3fe3e6b1 JL |
1310 | struct xfs_inode *ip = XFS_I(inode); |
1311 | struct xfs_mount *mp = ip->i_mount; | |
3fe3e6b1 | 1312 | loff_t uninitialized_var(offset); |
3fe3e6b1 | 1313 | xfs_fileoff_t fsbno; |
8aa7d37e | 1314 | xfs_filblks_t lastbno; |
3fe3e6b1 JL |
1315 | int error; |
1316 | ||
8aa7d37e | 1317 | if (start >= end) { |
2451337d | 1318 | error = -ENXIO; |
8aa7d37e | 1319 | goto out_error; |
3fe3e6b1 JL |
1320 | } |
1321 | ||
3fe3e6b1 JL |
1322 | /* |
1323 | * Try to read extents from the first block indicated | |
1324 | * by fsbno to the end block of the file. | |
1325 | */ | |
52f1acc8 | 1326 | fsbno = XFS_B_TO_FSBT(mp, start); |
8aa7d37e | 1327 | lastbno = XFS_B_TO_FSB(mp, end); |
49c69591 | 1328 | |
52f1acc8 JL |
1329 | for (;;) { |
1330 | struct xfs_bmbt_irec map[2]; | |
1331 | int nmap = 2; | |
1332 | unsigned int i; | |
3fe3e6b1 | 1333 | |
8aa7d37e | 1334 | error = xfs_bmapi_read(ip, fsbno, lastbno - fsbno, map, &nmap, |
52f1acc8 JL |
1335 | XFS_BMAPI_ENTIRE); |
1336 | if (error) | |
8aa7d37e | 1337 | goto out_error; |
3fe3e6b1 | 1338 | |
52f1acc8 JL |
1339 | /* No extents at given offset, must be beyond EOF */ |
1340 | if (nmap == 0) { | |
2451337d | 1341 | error = -ENXIO; |
8aa7d37e | 1342 | goto out_error; |
52f1acc8 JL |
1343 | } |
1344 | ||
1345 | for (i = 0; i < nmap; i++) { | |
1346 | offset = max_t(loff_t, start, | |
1347 | XFS_FSB_TO_B(mp, map[i].br_startoff)); | |
1348 | ||
49c69591 ES |
1349 | /* Landed in the hole we wanted? */ |
1350 | if (whence == SEEK_HOLE && | |
1351 | map[i].br_startblock == HOLESTARTBLOCK) | |
1352 | goto out; | |
1353 | ||
1354 | /* Landed in the data extent we wanted? */ | |
1355 | if (whence == SEEK_DATA && | |
1356 | (map[i].br_startblock == DELAYSTARTBLOCK || | |
1357 | (map[i].br_state == XFS_EXT_NORM && | |
1358 | !isnullstartblock(map[i].br_startblock)))) | |
52f1acc8 JL |
1359 | goto out; |
1360 | ||
1361 | /* | |
49c69591 ES |
1362 | * Landed in an unwritten extent, try to search |
1363 | * for hole or data from page cache. | |
52f1acc8 JL |
1364 | */ |
1365 | if (map[i].br_state == XFS_EXT_UNWRITTEN) { | |
1366 | if (xfs_find_get_desired_pgoff(inode, &map[i], | |
49c69591 ES |
1367 | whence == SEEK_HOLE ? HOLE_OFF : DATA_OFF, |
1368 | &offset)) | |
52f1acc8 JL |
1369 | goto out; |
1370 | } | |
1371 | } | |
1372 | ||
1373 | /* | |
49c69591 ES |
1374 | * We only received one extent out of the two requested. This |
1375 | * means we've hit EOF and didn't find what we are looking for. | |
52f1acc8 | 1376 | */ |
3fe3e6b1 | 1377 | if (nmap == 1) { |
49c69591 ES |
1378 | /* |
1379 | * If we were looking for a hole, set offset to | |
1380 | * the end of the file (i.e., there is an implicit | |
1381 | * hole at the end of any file). | |
1382 | */ | |
1383 | if (whence == SEEK_HOLE) { | |
8aa7d37e | 1384 | offset = end; |
49c69591 ES |
1385 | break; |
1386 | } | |
1387 | /* | |
1388 | * If we were looking for data, it's nowhere to be found | |
1389 | */ | |
1390 | ASSERT(whence == SEEK_DATA); | |
2451337d | 1391 | error = -ENXIO; |
8aa7d37e | 1392 | goto out_error; |
3fe3e6b1 JL |
1393 | } |
1394 | ||
52f1acc8 JL |
1395 | ASSERT(i > 1); |
1396 | ||
1397 | /* | |
1398 | * Nothing was found, proceed to the next round of search | |
49c69591 | 1399 | * if the next reading offset is not at or beyond EOF. |
52f1acc8 JL |
1400 | */ |
1401 | fsbno = map[i - 1].br_startoff + map[i - 1].br_blockcount; | |
1402 | start = XFS_FSB_TO_B(mp, fsbno); | |
8aa7d37e | 1403 | if (start >= end) { |
49c69591 | 1404 | if (whence == SEEK_HOLE) { |
8aa7d37e | 1405 | offset = end; |
49c69591 ES |
1406 | break; |
1407 | } | |
1408 | ASSERT(whence == SEEK_DATA); | |
2451337d | 1409 | error = -ENXIO; |
8aa7d37e | 1410 | goto out_error; |
52f1acc8 | 1411 | } |
3fe3e6b1 JL |
1412 | } |
1413 | ||
b686d1f7 JL |
1414 | out: |
1415 | /* | |
49c69591 | 1416 | * If at this point we have found the hole we wanted, the returned |
b686d1f7 | 1417 | * offset may be bigger than the file size as it may be aligned to |
49c69591 | 1418 | * page boundary for unwritten extents. We need to deal with this |
b686d1f7 JL |
1419 | * situation in particular. |
1420 | */ | |
49c69591 | 1421 | if (whence == SEEK_HOLE) |
8aa7d37e ES |
1422 | offset = min_t(loff_t, offset, end); |
1423 | ||
1424 | return offset; | |
1425 | ||
1426 | out_error: | |
1427 | return error; | |
1428 | } | |
1429 | ||
1430 | STATIC loff_t | |
1431 | xfs_seek_hole_data( | |
1432 | struct file *file, | |
1433 | loff_t start, | |
1434 | int whence) | |
1435 | { | |
1436 | struct inode *inode = file->f_mapping->host; | |
1437 | struct xfs_inode *ip = XFS_I(inode); | |
1438 | struct xfs_mount *mp = ip->i_mount; | |
1439 | uint lock; | |
1440 | loff_t offset, end; | |
1441 | int error = 0; | |
1442 | ||
1443 | if (XFS_FORCED_SHUTDOWN(mp)) | |
1444 | return -EIO; | |
1445 | ||
1446 | lock = xfs_ilock_data_map_shared(ip); | |
1447 | ||
1448 | end = i_size_read(inode); | |
1449 | offset = __xfs_seek_hole_data(inode, start, end, whence); | |
1450 | if (offset < 0) { | |
1451 | error = offset; | |
1452 | goto out_unlock; | |
1453 | } | |
1454 | ||
46a1c2c7 | 1455 | offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes); |
3fe3e6b1 JL |
1456 | |
1457 | out_unlock: | |
01f4f327 | 1458 | xfs_iunlock(ip, lock); |
3fe3e6b1 JL |
1459 | |
1460 | if (error) | |
2451337d | 1461 | return error; |
3fe3e6b1 JL |
1462 | return offset; |
1463 | } | |
1464 | ||
1465 | STATIC loff_t | |
1466 | xfs_file_llseek( | |
1467 | struct file *file, | |
1468 | loff_t offset, | |
59f9c004 | 1469 | int whence) |
3fe3e6b1 | 1470 | { |
59f9c004 | 1471 | switch (whence) { |
3fe3e6b1 JL |
1472 | case SEEK_END: |
1473 | case SEEK_CUR: | |
1474 | case SEEK_SET: | |
59f9c004 | 1475 | return generic_file_llseek(file, offset, whence); |
3fe3e6b1 | 1476 | case SEEK_HOLE: |
49c69591 | 1477 | case SEEK_DATA: |
59f9c004 | 1478 | return xfs_seek_hole_data(file, offset, whence); |
3fe3e6b1 JL |
1479 | default: |
1480 | return -EINVAL; | |
1481 | } | |
1482 | } | |
1483 | ||
de0e8c20 DC |
1484 | /* |
1485 | * Locking for serialisation of IO during page faults. This results in a lock | |
1486 | * ordering of: | |
1487 | * | |
1488 | * mmap_sem (MM) | |
6b698ede | 1489 | * sb_start_pagefault(vfs, freeze) |
13ad4fe3 | 1490 | * i_mmaplock (XFS - truncate serialisation) |
6b698ede DC |
1491 | * page_lock (MM) |
1492 | * i_lock (XFS - extent map serialisation) | |
de0e8c20 | 1493 | */ |
de0e8c20 | 1494 | |
075a924d DC |
1495 | /* |
1496 | * mmap()d file has taken write protection fault and is being made writable. We | |
1497 | * can set the page state up correctly for a writable page, which means we can | |
1498 | * do correct delalloc accounting (ENOSPC checking!) and unwritten extent | |
1499 | * mapping. | |
de0e8c20 DC |
1500 | */ |
1501 | STATIC int | |
075a924d | 1502 | xfs_filemap_page_mkwrite( |
de0e8c20 DC |
1503 | struct vm_area_struct *vma, |
1504 | struct vm_fault *vmf) | |
1505 | { | |
6b698ede | 1506 | struct inode *inode = file_inode(vma->vm_file); |
ec56b1f1 | 1507 | int ret; |
de0e8c20 | 1508 | |
6b698ede | 1509 | trace_xfs_filemap_page_mkwrite(XFS_I(inode)); |
de0e8c20 | 1510 | |
6b698ede | 1511 | sb_start_pagefault(inode->i_sb); |
ec56b1f1 | 1512 | file_update_time(vma->vm_file); |
6b698ede | 1513 | xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED); |
de0e8c20 | 1514 | |
6b698ede | 1515 | if (IS_DAX(inode)) { |
6b524995 | 1516 | ret = dax_mkwrite(vma, vmf, xfs_get_blocks_dax_fault); |
6b698ede | 1517 | } else { |
68a9f5e7 | 1518 | ret = iomap_page_mkwrite(vma, vmf, &xfs_iomap_ops); |
6b698ede DC |
1519 | ret = block_page_mkwrite_return(ret); |
1520 | } | |
1521 | ||
1522 | xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED); | |
1523 | sb_end_pagefault(inode->i_sb); | |
1524 | ||
1525 | return ret; | |
de0e8c20 DC |
1526 | } |
1527 | ||
075a924d | 1528 | STATIC int |
6b698ede | 1529 | xfs_filemap_fault( |
075a924d DC |
1530 | struct vm_area_struct *vma, |
1531 | struct vm_fault *vmf) | |
1532 | { | |
b2442c5a | 1533 | struct inode *inode = file_inode(vma->vm_file); |
6b698ede | 1534 | int ret; |
ec56b1f1 | 1535 | |
b2442c5a | 1536 | trace_xfs_filemap_fault(XFS_I(inode)); |
075a924d | 1537 | |
6b698ede | 1538 | /* DAX can shortcut the normal fault path on write faults! */ |
b2442c5a | 1539 | if ((vmf->flags & FAULT_FLAG_WRITE) && IS_DAX(inode)) |
6b698ede | 1540 | return xfs_filemap_page_mkwrite(vma, vmf); |
075a924d | 1541 | |
b2442c5a DC |
1542 | xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED); |
1543 | if (IS_DAX(inode)) { | |
1544 | /* | |
1545 | * we do not want to trigger unwritten extent conversion on read | |
1546 | * faults - that is unnecessary overhead and would also require | |
1547 | * changes to xfs_get_blocks_direct() to map unwritten extent | |
1548 | * ioend for conversion on read-only mappings. | |
1549 | */ | |
6b524995 | 1550 | ret = dax_fault(vma, vmf, xfs_get_blocks_dax_fault); |
b2442c5a DC |
1551 | } else |
1552 | ret = filemap_fault(vma, vmf); | |
1553 | xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED); | |
075a924d | 1554 | |
6b698ede DC |
1555 | return ret; |
1556 | } | |
1557 | ||
13ad4fe3 DC |
1558 | /* |
1559 | * Similar to xfs_filemap_fault(), the DAX fault path can call into here on | |
1560 | * both read and write faults. Hence we need to handle both cases. There is no | |
1561 | * ->pmd_mkwrite callout for huge pages, so we have a single function here to | |
1562 | * handle both cases here. @flags carries the information on the type of fault | |
1563 | * occuring. | |
1564 | */ | |
acd76e74 MW |
1565 | STATIC int |
1566 | xfs_filemap_pmd_fault( | |
1567 | struct vm_area_struct *vma, | |
1568 | unsigned long addr, | |
1569 | pmd_t *pmd, | |
1570 | unsigned int flags) | |
1571 | { | |
1572 | struct inode *inode = file_inode(vma->vm_file); | |
1573 | struct xfs_inode *ip = XFS_I(inode); | |
1574 | int ret; | |
1575 | ||
1576 | if (!IS_DAX(inode)) | |
1577 | return VM_FAULT_FALLBACK; | |
1578 | ||
1579 | trace_xfs_filemap_pmd_fault(ip); | |
1580 | ||
13ad4fe3 DC |
1581 | if (flags & FAULT_FLAG_WRITE) { |
1582 | sb_start_pagefault(inode->i_sb); | |
1583 | file_update_time(vma->vm_file); | |
1584 | } | |
1585 | ||
acd76e74 | 1586 | xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED); |
6b524995 | 1587 | ret = dax_pmd_fault(vma, addr, pmd, flags, xfs_get_blocks_dax_fault); |
acd76e74 | 1588 | xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED); |
acd76e74 | 1589 | |
13ad4fe3 DC |
1590 | if (flags & FAULT_FLAG_WRITE) |
1591 | sb_end_pagefault(inode->i_sb); | |
acd76e74 MW |
1592 | |
1593 | return ret; | |
1594 | } | |
1595 | ||
3af49285 DC |
1596 | /* |
1597 | * pfn_mkwrite was originally inteneded to ensure we capture time stamp | |
1598 | * updates on write faults. In reality, it's need to serialise against | |
5eb88dca RZ |
1599 | * truncate similar to page_mkwrite. Hence we cycle the XFS_MMAPLOCK_SHARED |
1600 | * to ensure we serialise the fault barrier in place. | |
3af49285 DC |
1601 | */ |
1602 | static int | |
1603 | xfs_filemap_pfn_mkwrite( | |
1604 | struct vm_area_struct *vma, | |
1605 | struct vm_fault *vmf) | |
1606 | { | |
1607 | ||
1608 | struct inode *inode = file_inode(vma->vm_file); | |
1609 | struct xfs_inode *ip = XFS_I(inode); | |
1610 | int ret = VM_FAULT_NOPAGE; | |
1611 | loff_t size; | |
1612 | ||
1613 | trace_xfs_filemap_pfn_mkwrite(ip); | |
1614 | ||
1615 | sb_start_pagefault(inode->i_sb); | |
1616 | file_update_time(vma->vm_file); | |
1617 | ||
1618 | /* check if the faulting page hasn't raced with truncate */ | |
1619 | xfs_ilock(ip, XFS_MMAPLOCK_SHARED); | |
1620 | size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
1621 | if (vmf->pgoff >= size) | |
1622 | ret = VM_FAULT_SIGBUS; | |
5eb88dca RZ |
1623 | else if (IS_DAX(inode)) |
1624 | ret = dax_pfn_mkwrite(vma, vmf); | |
3af49285 DC |
1625 | xfs_iunlock(ip, XFS_MMAPLOCK_SHARED); |
1626 | sb_end_pagefault(inode->i_sb); | |
acd76e74 | 1627 | return ret; |
3af49285 | 1628 | |
acd76e74 MW |
1629 | } |
1630 | ||
6b698ede DC |
1631 | static const struct vm_operations_struct xfs_file_vm_ops = { |
1632 | .fault = xfs_filemap_fault, | |
acd76e74 | 1633 | .pmd_fault = xfs_filemap_pmd_fault, |
6b698ede DC |
1634 | .map_pages = filemap_map_pages, |
1635 | .page_mkwrite = xfs_filemap_page_mkwrite, | |
3af49285 | 1636 | .pfn_mkwrite = xfs_filemap_pfn_mkwrite, |
6b698ede DC |
1637 | }; |
1638 | ||
1639 | STATIC int | |
1640 | xfs_file_mmap( | |
1641 | struct file *filp, | |
1642 | struct vm_area_struct *vma) | |
1643 | { | |
1644 | file_accessed(filp); | |
1645 | vma->vm_ops = &xfs_file_vm_ops; | |
1646 | if (IS_DAX(file_inode(filp))) | |
acd76e74 | 1647 | vma->vm_flags |= VM_MIXEDMAP | VM_HUGEPAGE; |
6b698ede | 1648 | return 0; |
075a924d DC |
1649 | } |
1650 | ||
4b6f5d20 | 1651 | const struct file_operations xfs_file_operations = { |
3fe3e6b1 | 1652 | .llseek = xfs_file_llseek, |
b4f5d2c6 | 1653 | .read_iter = xfs_file_read_iter, |
bf97f3bc | 1654 | .write_iter = xfs_file_write_iter, |
1b895840 | 1655 | .splice_read = xfs_file_splice_read, |
8d020765 | 1656 | .splice_write = iter_file_splice_write, |
3562fd45 | 1657 | .unlocked_ioctl = xfs_file_ioctl, |
1da177e4 | 1658 | #ifdef CONFIG_COMPAT |
3562fd45 | 1659 | .compat_ioctl = xfs_file_compat_ioctl, |
1da177e4 | 1660 | #endif |
3562fd45 NS |
1661 | .mmap = xfs_file_mmap, |
1662 | .open = xfs_file_open, | |
1663 | .release = xfs_file_release, | |
1664 | .fsync = xfs_file_fsync, | |
2fe17c10 | 1665 | .fallocate = xfs_file_fallocate, |
1da177e4 LT |
1666 | }; |
1667 | ||
4b6f5d20 | 1668 | const struct file_operations xfs_dir_file_operations = { |
f999a5bf | 1669 | .open = xfs_dir_open, |
1da177e4 | 1670 | .read = generic_read_dir, |
3b0a3c1a | 1671 | .iterate_shared = xfs_file_readdir, |
59af1584 | 1672 | .llseek = generic_file_llseek, |
3562fd45 | 1673 | .unlocked_ioctl = xfs_file_ioctl, |
d3870398 | 1674 | #ifdef CONFIG_COMPAT |
3562fd45 | 1675 | .compat_ioctl = xfs_file_compat_ioctl, |
d3870398 | 1676 | #endif |
1da2f2db | 1677 | .fsync = xfs_dir_fsync, |
1da177e4 | 1678 | }; |