]>
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 | ||
dda35b8f | 402 | /* |
193aec10 CH |
403 | * Zero any on disk space between the current EOF and the new, larger EOF. |
404 | * | |
405 | * This handles the normal case of zeroing the remainder of the last block in | |
406 | * the file and the unusual case of zeroing blocks out beyond the size of the | |
407 | * file. This second case only happens with fixed size extents and when the | |
408 | * system crashes before the inode size was updated but after blocks were | |
409 | * allocated. | |
410 | * | |
411 | * Expects the iolock to be held exclusive, and will take the ilock internally. | |
dda35b8f | 412 | */ |
dda35b8f CH |
413 | int /* error (positive) */ |
414 | xfs_zero_eof( | |
193aec10 CH |
415 | struct xfs_inode *ip, |
416 | xfs_off_t offset, /* starting I/O offset */ | |
5885ebda DC |
417 | xfs_fsize_t isize, /* current inode size */ |
418 | bool *did_zeroing) | |
dda35b8f | 419 | { |
193aec10 | 420 | ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL)); |
dda35b8f CH |
421 | ASSERT(offset > isize); |
422 | ||
0a50f162 | 423 | trace_xfs_zero_eof(ip, isize, offset - isize); |
570b6211 | 424 | return xfs_zero_range(ip, isize, offset - isize, did_zeroing); |
dda35b8f CH |
425 | } |
426 | ||
4d8d1581 DC |
427 | /* |
428 | * Common pre-write limit and setup checks. | |
429 | * | |
5bf1f262 CH |
430 | * Called with the iolocked held either shared and exclusive according to |
431 | * @iolock, and returns with it held. Might upgrade the iolock to exclusive | |
432 | * if called for a direct write beyond i_size. | |
4d8d1581 DC |
433 | */ |
434 | STATIC ssize_t | |
435 | xfs_file_aio_write_checks( | |
99733fa3 AV |
436 | struct kiocb *iocb, |
437 | struct iov_iter *from, | |
4d8d1581 DC |
438 | int *iolock) |
439 | { | |
99733fa3 | 440 | struct file *file = iocb->ki_filp; |
4d8d1581 DC |
441 | struct inode *inode = file->f_mapping->host; |
442 | struct xfs_inode *ip = XFS_I(inode); | |
3309dd04 | 443 | ssize_t error = 0; |
99733fa3 | 444 | size_t count = iov_iter_count(from); |
3136e8bb | 445 | bool drained_dio = false; |
4d8d1581 | 446 | |
7271d243 | 447 | restart: |
3309dd04 AV |
448 | error = generic_write_checks(iocb, from); |
449 | if (error <= 0) | |
4d8d1581 | 450 | return error; |
4d8d1581 | 451 | |
21c3ea18 | 452 | error = xfs_break_layouts(inode, iolock, true); |
781355c6 CH |
453 | if (error) |
454 | return error; | |
455 | ||
a6de82ca JK |
456 | /* For changing security info in file_remove_privs() we need i_mutex */ |
457 | if (*iolock == XFS_IOLOCK_SHARED && !IS_NOSEC(inode)) { | |
458 | xfs_rw_iunlock(ip, *iolock); | |
459 | *iolock = XFS_IOLOCK_EXCL; | |
460 | xfs_rw_ilock(ip, *iolock); | |
461 | goto restart; | |
462 | } | |
4d8d1581 DC |
463 | /* |
464 | * If the offset is beyond the size of the file, we need to zero any | |
465 | * blocks that fall between the existing EOF and the start of this | |
2813d682 | 466 | * write. If zeroing is needed and we are currently holding the |
467f7899 CH |
467 | * iolock shared, we need to update it to exclusive which implies |
468 | * having to redo all checks before. | |
b9d59846 DC |
469 | * |
470 | * We need to serialise against EOF updates that occur in IO | |
471 | * completions here. We want to make sure that nobody is changing the | |
472 | * size while we do this check until we have placed an IO barrier (i.e. | |
473 | * hold the XFS_IOLOCK_EXCL) that prevents new IO from being dispatched. | |
474 | * The spinlock effectively forms a memory barrier once we have the | |
475 | * XFS_IOLOCK_EXCL so we are guaranteed to see the latest EOF value | |
476 | * and hence be able to correctly determine if we need to run zeroing. | |
4d8d1581 | 477 | */ |
b9d59846 | 478 | spin_lock(&ip->i_flags_lock); |
99733fa3 | 479 | if (iocb->ki_pos > i_size_read(inode)) { |
5885ebda DC |
480 | bool zero = false; |
481 | ||
b9d59846 | 482 | spin_unlock(&ip->i_flags_lock); |
3136e8bb BF |
483 | if (!drained_dio) { |
484 | if (*iolock == XFS_IOLOCK_SHARED) { | |
485 | xfs_rw_iunlock(ip, *iolock); | |
486 | *iolock = XFS_IOLOCK_EXCL; | |
487 | xfs_rw_ilock(ip, *iolock); | |
488 | iov_iter_reexpand(from, count); | |
489 | } | |
40c63fbc DC |
490 | /* |
491 | * We now have an IO submission barrier in place, but | |
492 | * AIO can do EOF updates during IO completion and hence | |
493 | * we now need to wait for all of them to drain. Non-AIO | |
494 | * DIO will have drained before we are given the | |
495 | * XFS_IOLOCK_EXCL, and so for most cases this wait is a | |
496 | * no-op. | |
497 | */ | |
498 | inode_dio_wait(inode); | |
3136e8bb | 499 | drained_dio = true; |
7271d243 DC |
500 | goto restart; |
501 | } | |
99733fa3 | 502 | error = xfs_zero_eof(ip, iocb->ki_pos, i_size_read(inode), &zero); |
467f7899 CH |
503 | if (error) |
504 | return error; | |
b9d59846 DC |
505 | } else |
506 | spin_unlock(&ip->i_flags_lock); | |
4d8d1581 | 507 | |
8a9c9980 CH |
508 | /* |
509 | * Updating the timestamps will grab the ilock again from | |
510 | * xfs_fs_dirty_inode, so we have to call it after dropping the | |
511 | * lock above. Eventually we should look into a way to avoid | |
512 | * the pointless lock roundtrip. | |
513 | */ | |
c3b2da31 JB |
514 | if (likely(!(file->f_mode & FMODE_NOCMTIME))) { |
515 | error = file_update_time(file); | |
516 | if (error) | |
517 | return error; | |
518 | } | |
8a9c9980 | 519 | |
4d8d1581 DC |
520 | /* |
521 | * If we're writing the file then make sure to clear the setuid and | |
522 | * setgid bits if the process is not being run by root. This keeps | |
523 | * people from modifying setuid and setgid binaries. | |
524 | */ | |
a6de82ca JK |
525 | if (!IS_NOSEC(inode)) |
526 | return file_remove_privs(file); | |
527 | return 0; | |
4d8d1581 DC |
528 | } |
529 | ||
f0d26e86 DC |
530 | /* |
531 | * xfs_file_dio_aio_write - handle direct IO writes | |
532 | * | |
533 | * Lock the inode appropriately to prepare for and issue a direct IO write. | |
eda77982 | 534 | * By separating it from the buffered write path we remove all the tricky to |
f0d26e86 DC |
535 | * follow locking changes and looping. |
536 | * | |
eda77982 DC |
537 | * If there are cached pages or we're extending the file, we need IOLOCK_EXCL |
538 | * until we're sure the bytes at the new EOF have been zeroed and/or the cached | |
539 | * pages are flushed out. | |
540 | * | |
541 | * In most cases the direct IO writes will be done holding IOLOCK_SHARED | |
542 | * allowing them to be done in parallel with reads and other direct IO writes. | |
543 | * However, if the IO is not aligned to filesystem blocks, the direct IO layer | |
544 | * needs to do sub-block zeroing and that requires serialisation against other | |
545 | * direct IOs to the same block. In this case we need to serialise the | |
546 | * submission of the unaligned IOs so that we don't get racing block zeroing in | |
547 | * the dio layer. To avoid the problem with aio, we also need to wait for | |
548 | * outstanding IOs to complete so that unwritten extent conversion is completed | |
549 | * before we try to map the overlapping block. This is currently implemented by | |
4a06fd26 | 550 | * hitting it with a big hammer (i.e. inode_dio_wait()). |
eda77982 | 551 | * |
f0d26e86 DC |
552 | * Returns with locks held indicated by @iolock and errors indicated by |
553 | * negative return values. | |
554 | */ | |
555 | STATIC ssize_t | |
556 | xfs_file_dio_aio_write( | |
557 | struct kiocb *iocb, | |
b3188919 | 558 | struct iov_iter *from) |
f0d26e86 DC |
559 | { |
560 | struct file *file = iocb->ki_filp; | |
561 | struct address_space *mapping = file->f_mapping; | |
562 | struct inode *inode = mapping->host; | |
563 | struct xfs_inode *ip = XFS_I(inode); | |
564 | struct xfs_mount *mp = ip->i_mount; | |
565 | ssize_t ret = 0; | |
eda77982 | 566 | int unaligned_io = 0; |
d0606464 | 567 | int iolock; |
b3188919 | 568 | size_t count = iov_iter_count(from); |
0cefb29e DC |
569 | loff_t end; |
570 | struct iov_iter data; | |
f0d26e86 DC |
571 | struct xfs_buftarg *target = XFS_IS_REALTIME_INODE(ip) ? |
572 | mp->m_rtdev_targp : mp->m_ddev_targp; | |
573 | ||
7c71ee78 | 574 | /* DIO must be aligned to device logical sector size */ |
16d4d435 | 575 | if ((iocb->ki_pos | count) & target->bt_logical_sectormask) |
b474c7ae | 576 | return -EINVAL; |
f0d26e86 | 577 | |
7c71ee78 | 578 | /* "unaligned" here means not aligned to a filesystem block */ |
13712713 CH |
579 | if ((iocb->ki_pos & mp->m_blockmask) || |
580 | ((iocb->ki_pos + count) & mp->m_blockmask)) | |
eda77982 DC |
581 | unaligned_io = 1; |
582 | ||
7271d243 DC |
583 | /* |
584 | * We don't need to take an exclusive lock unless there page cache needs | |
585 | * to be invalidated or unaligned IO is being executed. We don't need to | |
586 | * consider the EOF extension case here because | |
587 | * xfs_file_aio_write_checks() will relock the inode as necessary for | |
588 | * EOF zeroing cases and fill out the new inode size as appropriate. | |
589 | */ | |
590 | if (unaligned_io || mapping->nrpages) | |
d0606464 | 591 | iolock = XFS_IOLOCK_EXCL; |
f0d26e86 | 592 | else |
d0606464 CH |
593 | iolock = XFS_IOLOCK_SHARED; |
594 | xfs_rw_ilock(ip, iolock); | |
c58cb165 CH |
595 | |
596 | /* | |
597 | * Recheck if there are cached pages that need invalidate after we got | |
598 | * the iolock to protect against other threads adding new pages while | |
599 | * we were waiting for the iolock. | |
600 | */ | |
d0606464 CH |
601 | if (mapping->nrpages && iolock == XFS_IOLOCK_SHARED) { |
602 | xfs_rw_iunlock(ip, iolock); | |
603 | iolock = XFS_IOLOCK_EXCL; | |
604 | xfs_rw_ilock(ip, iolock); | |
c58cb165 | 605 | } |
f0d26e86 | 606 | |
99733fa3 | 607 | ret = xfs_file_aio_write_checks(iocb, from, &iolock); |
4d8d1581 | 608 | if (ret) |
d0606464 | 609 | goto out; |
99733fa3 | 610 | count = iov_iter_count(from); |
13712713 | 611 | end = iocb->ki_pos + count - 1; |
f0d26e86 | 612 | |
3d751af2 | 613 | /* |
bbc5a740 | 614 | * See xfs_file_dio_aio_read() for why we do a full-file flush here. |
3d751af2 | 615 | */ |
f0d26e86 | 616 | if (mapping->nrpages) { |
3d751af2 | 617 | ret = filemap_write_and_wait(VFS_I(ip)->i_mapping); |
f0d26e86 | 618 | if (ret) |
d0606464 | 619 | goto out; |
834ffca6 | 620 | /* |
3d751af2 BF |
621 | * Invalidate whole pages. This can return an error if we fail |
622 | * to invalidate a page, but this should never happen on XFS. | |
623 | * Warn if it does fail. | |
834ffca6 | 624 | */ |
3d751af2 | 625 | ret = invalidate_inode_pages2(VFS_I(ip)->i_mapping); |
834ffca6 DC |
626 | WARN_ON_ONCE(ret); |
627 | ret = 0; | |
f0d26e86 DC |
628 | } |
629 | ||
eda77982 DC |
630 | /* |
631 | * If we are doing unaligned IO, wait for all other IO to drain, | |
632 | * otherwise demote the lock if we had to flush cached pages | |
633 | */ | |
634 | if (unaligned_io) | |
4a06fd26 | 635 | inode_dio_wait(inode); |
d0606464 | 636 | else if (iolock == XFS_IOLOCK_EXCL) { |
f0d26e86 | 637 | xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL); |
d0606464 | 638 | iolock = XFS_IOLOCK_SHARED; |
f0d26e86 DC |
639 | } |
640 | ||
3176c3e0 | 641 | trace_xfs_file_direct_write(ip, count, iocb->ki_pos); |
f0d26e86 | 642 | |
0cefb29e | 643 | data = *from; |
16d4d435 CH |
644 | ret = __blockdev_direct_IO(iocb, inode, target->bt_bdev, &data, |
645 | xfs_get_blocks_direct, xfs_end_io_direct_write, | |
646 | NULL, DIO_ASYNC_EXTEND); | |
0cefb29e DC |
647 | |
648 | /* see generic_file_direct_write() for why this is necessary */ | |
649 | if (mapping->nrpages) { | |
650 | invalidate_inode_pages2_range(mapping, | |
13712713 | 651 | iocb->ki_pos >> PAGE_SHIFT, |
09cbfeaf | 652 | end >> PAGE_SHIFT); |
0cefb29e DC |
653 | } |
654 | ||
655 | if (ret > 0) { | |
13712713 | 656 | iocb->ki_pos += ret; |
0cefb29e | 657 | iov_iter_advance(from, ret); |
0cefb29e | 658 | } |
d0606464 CH |
659 | out: |
660 | xfs_rw_iunlock(ip, iolock); | |
661 | ||
6b698ede | 662 | /* |
16d4d435 CH |
663 | * No fallback to buffered IO on errors for XFS, direct IO will either |
664 | * complete fully or fail. | |
6b698ede | 665 | */ |
16d4d435 CH |
666 | ASSERT(ret < 0 || ret == count); |
667 | return ret; | |
668 | } | |
669 | ||
f021bd07 | 670 | static noinline ssize_t |
16d4d435 CH |
671 | xfs_file_dax_write( |
672 | struct kiocb *iocb, | |
673 | struct iov_iter *from) | |
674 | { | |
675 | struct address_space *mapping = iocb->ki_filp->f_mapping; | |
676 | struct inode *inode = mapping->host; | |
677 | struct xfs_inode *ip = XFS_I(inode); | |
678 | struct xfs_mount *mp = ip->i_mount; | |
679 | ssize_t ret = 0; | |
680 | int unaligned_io = 0; | |
681 | int iolock; | |
682 | struct iov_iter data; | |
683 | ||
684 | /* "unaligned" here means not aligned to a filesystem block */ | |
685 | if ((iocb->ki_pos & mp->m_blockmask) || | |
686 | ((iocb->ki_pos + iov_iter_count(from)) & mp->m_blockmask)) { | |
687 | unaligned_io = 1; | |
688 | iolock = XFS_IOLOCK_EXCL; | |
689 | } else if (mapping->nrpages) { | |
690 | iolock = XFS_IOLOCK_EXCL; | |
691 | } else { | |
692 | iolock = XFS_IOLOCK_SHARED; | |
693 | } | |
694 | xfs_rw_ilock(ip, iolock); | |
695 | ||
696 | ret = xfs_file_aio_write_checks(iocb, from, &iolock); | |
697 | if (ret) | |
698 | goto out; | |
699 | ||
700 | /* | |
701 | * Yes, even DAX files can have page cache attached to them: A zeroed | |
702 | * page is inserted into the pagecache when we have to serve a write | |
703 | * fault on a hole. It should never be dirtied and can simply be | |
704 | * dropped from the pagecache once we get real data for the page. | |
8b2180b3 DC |
705 | * |
706 | * XXX: This is racy against mmap, and there's nothing we can do about | |
707 | * it. dax_do_io() should really do this invalidation internally as | |
708 | * it will know if we've allocated over a holei for this specific IO and | |
709 | * if so it needs to update the mapping tree and invalidate existing | |
710 | * PTEs over the newly allocated range. Remove this invalidation when | |
711 | * dax_do_io() is fixed up. | |
6b698ede | 712 | */ |
16d4d435 | 713 | if (mapping->nrpages) { |
8b2180b3 DC |
714 | loff_t end = iocb->ki_pos + iov_iter_count(from) - 1; |
715 | ||
716 | ret = invalidate_inode_pages2_range(mapping, | |
717 | iocb->ki_pos >> PAGE_SHIFT, | |
718 | end >> PAGE_SHIFT); | |
16d4d435 CH |
719 | WARN_ON_ONCE(ret); |
720 | } | |
721 | ||
722 | if (iolock == XFS_IOLOCK_EXCL && !unaligned_io) { | |
723 | xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL); | |
724 | iolock = XFS_IOLOCK_SHARED; | |
725 | } | |
726 | ||
727 | trace_xfs_file_dax_write(ip, iov_iter_count(from), iocb->ki_pos); | |
728 | ||
729 | data = *from; | |
730 | ret = dax_do_io(iocb, inode, &data, xfs_get_blocks_direct, | |
731 | xfs_end_io_direct_write, 0); | |
732 | if (ret > 0) { | |
733 | iocb->ki_pos += ret; | |
734 | iov_iter_advance(from, ret); | |
735 | } | |
736 | out: | |
737 | xfs_rw_iunlock(ip, iolock); | |
f0d26e86 DC |
738 | return ret; |
739 | } | |
740 | ||
00258e36 | 741 | STATIC ssize_t |
637bbc75 | 742 | xfs_file_buffered_aio_write( |
dda35b8f | 743 | struct kiocb *iocb, |
b3188919 | 744 | struct iov_iter *from) |
dda35b8f CH |
745 | { |
746 | struct file *file = iocb->ki_filp; | |
747 | struct address_space *mapping = file->f_mapping; | |
748 | struct inode *inode = mapping->host; | |
00258e36 | 749 | struct xfs_inode *ip = XFS_I(inode); |
637bbc75 DC |
750 | ssize_t ret; |
751 | int enospc = 0; | |
d0606464 | 752 | int iolock = XFS_IOLOCK_EXCL; |
dda35b8f | 753 | |
d0606464 | 754 | xfs_rw_ilock(ip, iolock); |
dda35b8f | 755 | |
99733fa3 | 756 | ret = xfs_file_aio_write_checks(iocb, from, &iolock); |
4d8d1581 | 757 | if (ret) |
d0606464 | 758 | goto out; |
dda35b8f CH |
759 | |
760 | /* We can write back this queue in page reclaim */ | |
de1414a6 | 761 | current->backing_dev_info = inode_to_bdi(inode); |
dda35b8f | 762 | |
dda35b8f | 763 | write_retry: |
3176c3e0 | 764 | trace_xfs_file_buffered_write(ip, iov_iter_count(from), iocb->ki_pos); |
68a9f5e7 | 765 | ret = iomap_file_buffered_write(iocb, from, &xfs_iomap_ops); |
0a64bc2c | 766 | if (likely(ret >= 0)) |
99733fa3 | 767 | iocb->ki_pos += ret; |
dc06f398 | 768 | |
637bbc75 | 769 | /* |
dc06f398 BF |
770 | * If we hit a space limit, try to free up some lingering preallocated |
771 | * space before returning an error. In the case of ENOSPC, first try to | |
772 | * write back all dirty inodes to free up some of the excess reserved | |
773 | * metadata space. This reduces the chances that the eofblocks scan | |
774 | * waits on dirty mappings. Since xfs_flush_inodes() is serialized, this | |
775 | * also behaves as a filter to prevent too many eofblocks scans from | |
776 | * running at the same time. | |
637bbc75 | 777 | */ |
dc06f398 BF |
778 | if (ret == -EDQUOT && !enospc) { |
779 | enospc = xfs_inode_free_quota_eofblocks(ip); | |
780 | if (enospc) | |
781 | goto write_retry; | |
782 | } else if (ret == -ENOSPC && !enospc) { | |
783 | struct xfs_eofblocks eofb = {0}; | |
784 | ||
637bbc75 | 785 | enospc = 1; |
9aa05000 | 786 | xfs_flush_inodes(ip->i_mount); |
dc06f398 BF |
787 | eofb.eof_scan_owner = ip->i_ino; /* for locking */ |
788 | eofb.eof_flags = XFS_EOF_FLAGS_SYNC; | |
789 | xfs_icache_free_eofblocks(ip->i_mount, &eofb); | |
9aa05000 | 790 | goto write_retry; |
dda35b8f | 791 | } |
d0606464 | 792 | |
dda35b8f | 793 | current->backing_dev_info = NULL; |
d0606464 CH |
794 | out: |
795 | xfs_rw_iunlock(ip, iolock); | |
637bbc75 DC |
796 | return ret; |
797 | } | |
798 | ||
799 | STATIC ssize_t | |
bf97f3bc | 800 | xfs_file_write_iter( |
637bbc75 | 801 | struct kiocb *iocb, |
bf97f3bc | 802 | struct iov_iter *from) |
637bbc75 DC |
803 | { |
804 | struct file *file = iocb->ki_filp; | |
805 | struct address_space *mapping = file->f_mapping; | |
806 | struct inode *inode = mapping->host; | |
807 | struct xfs_inode *ip = XFS_I(inode); | |
808 | ssize_t ret; | |
bf97f3bc | 809 | size_t ocount = iov_iter_count(from); |
637bbc75 | 810 | |
ff6d6af2 | 811 | XFS_STATS_INC(ip->i_mount, xs_write_calls); |
637bbc75 | 812 | |
637bbc75 DC |
813 | if (ocount == 0) |
814 | return 0; | |
815 | ||
bf97f3bc AV |
816 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
817 | return -EIO; | |
637bbc75 | 818 | |
16d4d435 CH |
819 | if (IS_DAX(inode)) |
820 | ret = xfs_file_dax_write(iocb, from); | |
821 | else if (iocb->ki_flags & IOCB_DIRECT) | |
bf97f3bc | 822 | ret = xfs_file_dio_aio_write(iocb, from); |
637bbc75 | 823 | else |
bf97f3bc | 824 | ret = xfs_file_buffered_aio_write(iocb, from); |
dda35b8f | 825 | |
d0606464 | 826 | if (ret > 0) { |
ff6d6af2 | 827 | XFS_STATS_ADD(ip->i_mount, xs_write_bytes, ret); |
dda35b8f | 828 | |
d0606464 | 829 | /* Handle various SYNC-type writes */ |
e2592217 | 830 | ret = generic_write_sync(iocb, ret); |
dda35b8f | 831 | } |
a363f0c2 | 832 | return ret; |
dda35b8f CH |
833 | } |
834 | ||
a904b1ca NJ |
835 | #define XFS_FALLOC_FL_SUPPORTED \ |
836 | (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \ | |
837 | FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE | \ | |
838 | FALLOC_FL_INSERT_RANGE) | |
839 | ||
2fe17c10 CH |
840 | STATIC long |
841 | xfs_file_fallocate( | |
83aee9e4 CH |
842 | struct file *file, |
843 | int mode, | |
844 | loff_t offset, | |
845 | loff_t len) | |
2fe17c10 | 846 | { |
83aee9e4 CH |
847 | struct inode *inode = file_inode(file); |
848 | struct xfs_inode *ip = XFS_I(inode); | |
83aee9e4 | 849 | long error; |
8add71ca | 850 | enum xfs_prealloc_flags flags = 0; |
781355c6 | 851 | uint iolock = XFS_IOLOCK_EXCL; |
83aee9e4 | 852 | loff_t new_size = 0; |
a904b1ca | 853 | bool do_file_insert = 0; |
2fe17c10 | 854 | |
83aee9e4 CH |
855 | if (!S_ISREG(inode->i_mode)) |
856 | return -EINVAL; | |
a904b1ca | 857 | if (mode & ~XFS_FALLOC_FL_SUPPORTED) |
2fe17c10 CH |
858 | return -EOPNOTSUPP; |
859 | ||
781355c6 | 860 | xfs_ilock(ip, iolock); |
21c3ea18 | 861 | error = xfs_break_layouts(inode, &iolock, false); |
781355c6 CH |
862 | if (error) |
863 | goto out_unlock; | |
864 | ||
e8e9ad42 DC |
865 | xfs_ilock(ip, XFS_MMAPLOCK_EXCL); |
866 | iolock |= XFS_MMAPLOCK_EXCL; | |
867 | ||
83aee9e4 CH |
868 | if (mode & FALLOC_FL_PUNCH_HOLE) { |
869 | error = xfs_free_file_space(ip, offset, len); | |
870 | if (error) | |
871 | goto out_unlock; | |
e1d8fb88 NJ |
872 | } else if (mode & FALLOC_FL_COLLAPSE_RANGE) { |
873 | unsigned blksize_mask = (1 << inode->i_blkbits) - 1; | |
874 | ||
875 | if (offset & blksize_mask || len & blksize_mask) { | |
2451337d | 876 | error = -EINVAL; |
e1d8fb88 NJ |
877 | goto out_unlock; |
878 | } | |
879 | ||
23fffa92 LC |
880 | /* |
881 | * There is no need to overlap collapse range with EOF, | |
882 | * in which case it is effectively a truncate operation | |
883 | */ | |
884 | if (offset + len >= i_size_read(inode)) { | |
2451337d | 885 | error = -EINVAL; |
23fffa92 LC |
886 | goto out_unlock; |
887 | } | |
888 | ||
e1d8fb88 NJ |
889 | new_size = i_size_read(inode) - len; |
890 | ||
891 | error = xfs_collapse_file_space(ip, offset, len); | |
892 | if (error) | |
893 | goto out_unlock; | |
a904b1ca NJ |
894 | } else if (mode & FALLOC_FL_INSERT_RANGE) { |
895 | unsigned blksize_mask = (1 << inode->i_blkbits) - 1; | |
896 | ||
897 | new_size = i_size_read(inode) + len; | |
898 | if (offset & blksize_mask || len & blksize_mask) { | |
899 | error = -EINVAL; | |
900 | goto out_unlock; | |
901 | } | |
902 | ||
903 | /* check the new inode size does not wrap through zero */ | |
904 | if (new_size > inode->i_sb->s_maxbytes) { | |
905 | error = -EFBIG; | |
906 | goto out_unlock; | |
907 | } | |
908 | ||
909 | /* Offset should be less than i_size */ | |
910 | if (offset >= i_size_read(inode)) { | |
911 | error = -EINVAL; | |
912 | goto out_unlock; | |
913 | } | |
914 | do_file_insert = 1; | |
83aee9e4 | 915 | } else { |
8add71ca CH |
916 | flags |= XFS_PREALLOC_SET; |
917 | ||
83aee9e4 CH |
918 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
919 | offset + len > i_size_read(inode)) { | |
920 | new_size = offset + len; | |
2451337d | 921 | error = inode_newsize_ok(inode, new_size); |
83aee9e4 CH |
922 | if (error) |
923 | goto out_unlock; | |
924 | } | |
2fe17c10 | 925 | |
376ba313 LC |
926 | if (mode & FALLOC_FL_ZERO_RANGE) |
927 | error = xfs_zero_file_space(ip, offset, len); | |
928 | else | |
929 | error = xfs_alloc_file_space(ip, offset, len, | |
930 | XFS_BMAPI_PREALLOC); | |
2fe17c10 CH |
931 | if (error) |
932 | goto out_unlock; | |
933 | } | |
934 | ||
83aee9e4 | 935 | if (file->f_flags & O_DSYNC) |
8add71ca CH |
936 | flags |= XFS_PREALLOC_SYNC; |
937 | ||
938 | error = xfs_update_prealloc_flags(ip, flags); | |
2fe17c10 CH |
939 | if (error) |
940 | goto out_unlock; | |
941 | ||
942 | /* Change file size if needed */ | |
943 | if (new_size) { | |
944 | struct iattr iattr; | |
945 | ||
946 | iattr.ia_valid = ATTR_SIZE; | |
947 | iattr.ia_size = new_size; | |
83aee9e4 | 948 | error = xfs_setattr_size(ip, &iattr); |
a904b1ca NJ |
949 | if (error) |
950 | goto out_unlock; | |
2fe17c10 CH |
951 | } |
952 | ||
a904b1ca NJ |
953 | /* |
954 | * Perform hole insertion now that the file size has been | |
955 | * updated so that if we crash during the operation we don't | |
956 | * leave shifted extents past EOF and hence losing access to | |
957 | * the data that is contained within them. | |
958 | */ | |
959 | if (do_file_insert) | |
960 | error = xfs_insert_file_space(ip, offset, len); | |
961 | ||
2fe17c10 | 962 | out_unlock: |
781355c6 | 963 | xfs_iunlock(ip, iolock); |
2451337d | 964 | return error; |
2fe17c10 CH |
965 | } |
966 | ||
967 | ||
1da177e4 | 968 | STATIC int |
3562fd45 | 969 | xfs_file_open( |
1da177e4 | 970 | struct inode *inode, |
f999a5bf | 971 | struct file *file) |
1da177e4 | 972 | { |
f999a5bf | 973 | if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS) |
1da177e4 | 974 | return -EFBIG; |
f999a5bf CH |
975 | if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb))) |
976 | return -EIO; | |
977 | return 0; | |
978 | } | |
979 | ||
980 | STATIC int | |
981 | xfs_dir_open( | |
982 | struct inode *inode, | |
983 | struct file *file) | |
984 | { | |
985 | struct xfs_inode *ip = XFS_I(inode); | |
986 | int mode; | |
987 | int error; | |
988 | ||
989 | error = xfs_file_open(inode, file); | |
990 | if (error) | |
991 | return error; | |
992 | ||
993 | /* | |
994 | * If there are any blocks, read-ahead block 0 as we're almost | |
995 | * certain to have the next operation be a read there. | |
996 | */ | |
309ecac8 | 997 | mode = xfs_ilock_data_map_shared(ip); |
f999a5bf | 998 | if (ip->i_d.di_nextents > 0) |
9df2dd0b | 999 | xfs_dir3_data_readahead(ip, 0, -1); |
f999a5bf CH |
1000 | xfs_iunlock(ip, mode); |
1001 | return 0; | |
1da177e4 LT |
1002 | } |
1003 | ||
1da177e4 | 1004 | STATIC int |
3562fd45 | 1005 | xfs_file_release( |
1da177e4 LT |
1006 | struct inode *inode, |
1007 | struct file *filp) | |
1008 | { | |
2451337d | 1009 | return xfs_release(XFS_I(inode)); |
1da177e4 LT |
1010 | } |
1011 | ||
1da177e4 | 1012 | STATIC int |
3562fd45 | 1013 | xfs_file_readdir( |
b8227554 AV |
1014 | struct file *file, |
1015 | struct dir_context *ctx) | |
1da177e4 | 1016 | { |
b8227554 | 1017 | struct inode *inode = file_inode(file); |
739bfb2a | 1018 | xfs_inode_t *ip = XFS_I(inode); |
051e7cd4 CH |
1019 | size_t bufsize; |
1020 | ||
1021 | /* | |
1022 | * The Linux API doesn't pass down the total size of the buffer | |
1023 | * we read into down to the filesystem. With the filldir concept | |
1024 | * it's not needed for correct information, but the XFS dir2 leaf | |
1025 | * code wants an estimate of the buffer size to calculate it's | |
1026 | * readahead window and size the buffers used for mapping to | |
1027 | * physical blocks. | |
1028 | * | |
1029 | * Try to give it an estimate that's good enough, maybe at some | |
1030 | * point we can change the ->readdir prototype to include the | |
a9cc799e | 1031 | * buffer size. For now we use the current glibc buffer size. |
051e7cd4 | 1032 | */ |
a9cc799e | 1033 | bufsize = (size_t)min_t(loff_t, 32768, ip->i_d.di_size); |
051e7cd4 | 1034 | |
8300475e | 1035 | return xfs_readdir(ip, ctx, bufsize); |
1da177e4 LT |
1036 | } |
1037 | ||
d126d43f JL |
1038 | /* |
1039 | * This type is designed to indicate the type of offset we would like | |
49c69591 | 1040 | * to search from page cache for xfs_seek_hole_data(). |
d126d43f JL |
1041 | */ |
1042 | enum { | |
1043 | HOLE_OFF = 0, | |
1044 | DATA_OFF, | |
1045 | }; | |
1046 | ||
1047 | /* | |
1048 | * Lookup the desired type of offset from the given page. | |
1049 | * | |
1050 | * On success, return true and the offset argument will point to the | |
1051 | * start of the region that was found. Otherwise this function will | |
1052 | * return false and keep the offset argument unchanged. | |
1053 | */ | |
1054 | STATIC bool | |
1055 | xfs_lookup_buffer_offset( | |
1056 | struct page *page, | |
1057 | loff_t *offset, | |
1058 | unsigned int type) | |
1059 | { | |
1060 | loff_t lastoff = page_offset(page); | |
1061 | bool found = false; | |
1062 | struct buffer_head *bh, *head; | |
1063 | ||
1064 | bh = head = page_buffers(page); | |
1065 | do { | |
1066 | /* | |
1067 | * Unwritten extents that have data in the page | |
1068 | * cache covering them can be identified by the | |
1069 | * BH_Unwritten state flag. Pages with multiple | |
1070 | * buffers might have a mix of holes, data and | |
1071 | * unwritten extents - any buffer with valid | |
1072 | * data in it should have BH_Uptodate flag set | |
1073 | * on it. | |
1074 | */ | |
1075 | if (buffer_unwritten(bh) || | |
1076 | buffer_uptodate(bh)) { | |
1077 | if (type == DATA_OFF) | |
1078 | found = true; | |
1079 | } else { | |
1080 | if (type == HOLE_OFF) | |
1081 | found = true; | |
1082 | } | |
1083 | ||
1084 | if (found) { | |
1085 | *offset = lastoff; | |
1086 | break; | |
1087 | } | |
1088 | lastoff += bh->b_size; | |
1089 | } while ((bh = bh->b_this_page) != head); | |
1090 | ||
1091 | return found; | |
1092 | } | |
1093 | ||
1094 | /* | |
1095 | * This routine is called to find out and return a data or hole offset | |
1096 | * from the page cache for unwritten extents according to the desired | |
49c69591 | 1097 | * type for xfs_seek_hole_data(). |
d126d43f JL |
1098 | * |
1099 | * The argument offset is used to tell where we start to search from the | |
1100 | * page cache. Map is used to figure out the end points of the range to | |
1101 | * lookup pages. | |
1102 | * | |
1103 | * Return true if the desired type of offset was found, and the argument | |
1104 | * offset is filled with that address. Otherwise, return false and keep | |
1105 | * offset unchanged. | |
1106 | */ | |
1107 | STATIC bool | |
1108 | xfs_find_get_desired_pgoff( | |
1109 | struct inode *inode, | |
1110 | struct xfs_bmbt_irec *map, | |
1111 | unsigned int type, | |
1112 | loff_t *offset) | |
1113 | { | |
1114 | struct xfs_inode *ip = XFS_I(inode); | |
1115 | struct xfs_mount *mp = ip->i_mount; | |
1116 | struct pagevec pvec; | |
1117 | pgoff_t index; | |
1118 | pgoff_t end; | |
1119 | loff_t endoff; | |
1120 | loff_t startoff = *offset; | |
1121 | loff_t lastoff = startoff; | |
1122 | bool found = false; | |
1123 | ||
1124 | pagevec_init(&pvec, 0); | |
1125 | ||
09cbfeaf | 1126 | index = startoff >> PAGE_SHIFT; |
d126d43f | 1127 | endoff = XFS_FSB_TO_B(mp, map->br_startoff + map->br_blockcount); |
09cbfeaf | 1128 | end = endoff >> PAGE_SHIFT; |
d126d43f JL |
1129 | do { |
1130 | int want; | |
1131 | unsigned nr_pages; | |
1132 | unsigned int i; | |
1133 | ||
1134 | want = min_t(pgoff_t, end - index, PAGEVEC_SIZE); | |
1135 | nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index, | |
1136 | want); | |
1137 | /* | |
1138 | * No page mapped into given range. If we are searching holes | |
1139 | * and if this is the first time we got into the loop, it means | |
1140 | * that the given offset is landed in a hole, return it. | |
1141 | * | |
1142 | * If we have already stepped through some block buffers to find | |
1143 | * holes but they all contains data. In this case, the last | |
1144 | * offset is already updated and pointed to the end of the last | |
1145 | * mapped page, if it does not reach the endpoint to search, | |
1146 | * that means there should be a hole between them. | |
1147 | */ | |
1148 | if (nr_pages == 0) { | |
1149 | /* Data search found nothing */ | |
1150 | if (type == DATA_OFF) | |
1151 | break; | |
1152 | ||
1153 | ASSERT(type == HOLE_OFF); | |
1154 | if (lastoff == startoff || lastoff < endoff) { | |
1155 | found = true; | |
1156 | *offset = lastoff; | |
1157 | } | |
1158 | break; | |
1159 | } | |
1160 | ||
1161 | /* | |
1162 | * At lease we found one page. If this is the first time we | |
1163 | * step into the loop, and if the first page index offset is | |
1164 | * greater than the given search offset, a hole was found. | |
1165 | */ | |
1166 | if (type == HOLE_OFF && lastoff == startoff && | |
1167 | lastoff < page_offset(pvec.pages[0])) { | |
1168 | found = true; | |
1169 | break; | |
1170 | } | |
1171 | ||
1172 | for (i = 0; i < nr_pages; i++) { | |
1173 | struct page *page = pvec.pages[i]; | |
1174 | loff_t b_offset; | |
1175 | ||
1176 | /* | |
1177 | * At this point, the page may be truncated or | |
1178 | * invalidated (changing page->mapping to NULL), | |
1179 | * or even swizzled back from swapper_space to tmpfs | |
1180 | * file mapping. However, page->index will not change | |
1181 | * because we have a reference on the page. | |
1182 | * | |
1183 | * Searching done if the page index is out of range. | |
1184 | * If the current offset is not reaches the end of | |
1185 | * the specified search range, there should be a hole | |
1186 | * between them. | |
1187 | */ | |
1188 | if (page->index > end) { | |
1189 | if (type == HOLE_OFF && lastoff < endoff) { | |
1190 | *offset = lastoff; | |
1191 | found = true; | |
1192 | } | |
1193 | goto out; | |
1194 | } | |
1195 | ||
1196 | lock_page(page); | |
1197 | /* | |
1198 | * Page truncated or invalidated(page->mapping == NULL). | |
1199 | * We can freely skip it and proceed to check the next | |
1200 | * page. | |
1201 | */ | |
1202 | if (unlikely(page->mapping != inode->i_mapping)) { | |
1203 | unlock_page(page); | |
1204 | continue; | |
1205 | } | |
1206 | ||
1207 | if (!page_has_buffers(page)) { | |
1208 | unlock_page(page); | |
1209 | continue; | |
1210 | } | |
1211 | ||
1212 | found = xfs_lookup_buffer_offset(page, &b_offset, type); | |
1213 | if (found) { | |
1214 | /* | |
1215 | * The found offset may be less than the start | |
1216 | * point to search if this is the first time to | |
1217 | * come here. | |
1218 | */ | |
1219 | *offset = max_t(loff_t, startoff, b_offset); | |
1220 | unlock_page(page); | |
1221 | goto out; | |
1222 | } | |
1223 | ||
1224 | /* | |
1225 | * We either searching data but nothing was found, or | |
1226 | * searching hole but found a data buffer. In either | |
1227 | * case, probably the next page contains the desired | |
1228 | * things, update the last offset to it so. | |
1229 | */ | |
1230 | lastoff = page_offset(page) + PAGE_SIZE; | |
1231 | unlock_page(page); | |
1232 | } | |
1233 | ||
1234 | /* | |
1235 | * The number of returned pages less than our desired, search | |
1236 | * done. In this case, nothing was found for searching data, | |
1237 | * but we found a hole behind the last offset. | |
1238 | */ | |
1239 | if (nr_pages < want) { | |
1240 | if (type == HOLE_OFF) { | |
1241 | *offset = lastoff; | |
1242 | found = true; | |
1243 | } | |
1244 | break; | |
1245 | } | |
1246 | ||
1247 | index = pvec.pages[i - 1]->index + 1; | |
1248 | pagevec_release(&pvec); | |
1249 | } while (index <= end); | |
1250 | ||
1251 | out: | |
1252 | pagevec_release(&pvec); | |
1253 | return found; | |
1254 | } | |
1255 | ||
8aa7d37e ES |
1256 | /* |
1257 | * caller must lock inode with xfs_ilock_data_map_shared, | |
1258 | * can we craft an appropriate ASSERT? | |
1259 | * | |
1260 | * end is because the VFS-level lseek interface is defined such that any | |
1261 | * offset past i_size shall return -ENXIO, but we use this for quota code | |
1262 | * which does not maintain i_size, and we want to SEEK_DATA past i_size. | |
1263 | */ | |
1264 | loff_t | |
1265 | __xfs_seek_hole_data( | |
1266 | struct inode *inode, | |
49c69591 | 1267 | loff_t start, |
8aa7d37e | 1268 | loff_t end, |
49c69591 | 1269 | int whence) |
3fe3e6b1 | 1270 | { |
3fe3e6b1 JL |
1271 | struct xfs_inode *ip = XFS_I(inode); |
1272 | struct xfs_mount *mp = ip->i_mount; | |
3fe3e6b1 | 1273 | loff_t uninitialized_var(offset); |
3fe3e6b1 | 1274 | xfs_fileoff_t fsbno; |
8aa7d37e | 1275 | xfs_filblks_t lastbno; |
3fe3e6b1 JL |
1276 | int error; |
1277 | ||
8aa7d37e | 1278 | if (start >= end) { |
2451337d | 1279 | error = -ENXIO; |
8aa7d37e | 1280 | goto out_error; |
3fe3e6b1 JL |
1281 | } |
1282 | ||
3fe3e6b1 JL |
1283 | /* |
1284 | * Try to read extents from the first block indicated | |
1285 | * by fsbno to the end block of the file. | |
1286 | */ | |
52f1acc8 | 1287 | fsbno = XFS_B_TO_FSBT(mp, start); |
8aa7d37e | 1288 | lastbno = XFS_B_TO_FSB(mp, end); |
49c69591 | 1289 | |
52f1acc8 JL |
1290 | for (;;) { |
1291 | struct xfs_bmbt_irec map[2]; | |
1292 | int nmap = 2; | |
1293 | unsigned int i; | |
3fe3e6b1 | 1294 | |
8aa7d37e | 1295 | error = xfs_bmapi_read(ip, fsbno, lastbno - fsbno, map, &nmap, |
52f1acc8 JL |
1296 | XFS_BMAPI_ENTIRE); |
1297 | if (error) | |
8aa7d37e | 1298 | goto out_error; |
3fe3e6b1 | 1299 | |
52f1acc8 JL |
1300 | /* No extents at given offset, must be beyond EOF */ |
1301 | if (nmap == 0) { | |
2451337d | 1302 | error = -ENXIO; |
8aa7d37e | 1303 | goto out_error; |
52f1acc8 JL |
1304 | } |
1305 | ||
1306 | for (i = 0; i < nmap; i++) { | |
1307 | offset = max_t(loff_t, start, | |
1308 | XFS_FSB_TO_B(mp, map[i].br_startoff)); | |
1309 | ||
49c69591 ES |
1310 | /* Landed in the hole we wanted? */ |
1311 | if (whence == SEEK_HOLE && | |
1312 | map[i].br_startblock == HOLESTARTBLOCK) | |
1313 | goto out; | |
1314 | ||
1315 | /* Landed in the data extent we wanted? */ | |
1316 | if (whence == SEEK_DATA && | |
1317 | (map[i].br_startblock == DELAYSTARTBLOCK || | |
1318 | (map[i].br_state == XFS_EXT_NORM && | |
1319 | !isnullstartblock(map[i].br_startblock)))) | |
52f1acc8 JL |
1320 | goto out; |
1321 | ||
1322 | /* | |
49c69591 ES |
1323 | * Landed in an unwritten extent, try to search |
1324 | * for hole or data from page cache. | |
52f1acc8 JL |
1325 | */ |
1326 | if (map[i].br_state == XFS_EXT_UNWRITTEN) { | |
1327 | if (xfs_find_get_desired_pgoff(inode, &map[i], | |
49c69591 ES |
1328 | whence == SEEK_HOLE ? HOLE_OFF : DATA_OFF, |
1329 | &offset)) | |
52f1acc8 JL |
1330 | goto out; |
1331 | } | |
1332 | } | |
1333 | ||
1334 | /* | |
49c69591 ES |
1335 | * We only received one extent out of the two requested. This |
1336 | * means we've hit EOF and didn't find what we are looking for. | |
52f1acc8 | 1337 | */ |
3fe3e6b1 | 1338 | if (nmap == 1) { |
49c69591 ES |
1339 | /* |
1340 | * If we were looking for a hole, set offset to | |
1341 | * the end of the file (i.e., there is an implicit | |
1342 | * hole at the end of any file). | |
1343 | */ | |
1344 | if (whence == SEEK_HOLE) { | |
8aa7d37e | 1345 | offset = end; |
49c69591 ES |
1346 | break; |
1347 | } | |
1348 | /* | |
1349 | * If we were looking for data, it's nowhere to be found | |
1350 | */ | |
1351 | ASSERT(whence == SEEK_DATA); | |
2451337d | 1352 | error = -ENXIO; |
8aa7d37e | 1353 | goto out_error; |
3fe3e6b1 JL |
1354 | } |
1355 | ||
52f1acc8 JL |
1356 | ASSERT(i > 1); |
1357 | ||
1358 | /* | |
1359 | * Nothing was found, proceed to the next round of search | |
49c69591 | 1360 | * if the next reading offset is not at or beyond EOF. |
52f1acc8 JL |
1361 | */ |
1362 | fsbno = map[i - 1].br_startoff + map[i - 1].br_blockcount; | |
1363 | start = XFS_FSB_TO_B(mp, fsbno); | |
8aa7d37e | 1364 | if (start >= end) { |
49c69591 | 1365 | if (whence == SEEK_HOLE) { |
8aa7d37e | 1366 | offset = end; |
49c69591 ES |
1367 | break; |
1368 | } | |
1369 | ASSERT(whence == SEEK_DATA); | |
2451337d | 1370 | error = -ENXIO; |
8aa7d37e | 1371 | goto out_error; |
52f1acc8 | 1372 | } |
3fe3e6b1 JL |
1373 | } |
1374 | ||
b686d1f7 JL |
1375 | out: |
1376 | /* | |
49c69591 | 1377 | * If at this point we have found the hole we wanted, the returned |
b686d1f7 | 1378 | * offset may be bigger than the file size as it may be aligned to |
49c69591 | 1379 | * page boundary for unwritten extents. We need to deal with this |
b686d1f7 JL |
1380 | * situation in particular. |
1381 | */ | |
49c69591 | 1382 | if (whence == SEEK_HOLE) |
8aa7d37e ES |
1383 | offset = min_t(loff_t, offset, end); |
1384 | ||
1385 | return offset; | |
1386 | ||
1387 | out_error: | |
1388 | return error; | |
1389 | } | |
1390 | ||
1391 | STATIC loff_t | |
1392 | xfs_seek_hole_data( | |
1393 | struct file *file, | |
1394 | loff_t start, | |
1395 | int whence) | |
1396 | { | |
1397 | struct inode *inode = file->f_mapping->host; | |
1398 | struct xfs_inode *ip = XFS_I(inode); | |
1399 | struct xfs_mount *mp = ip->i_mount; | |
1400 | uint lock; | |
1401 | loff_t offset, end; | |
1402 | int error = 0; | |
1403 | ||
1404 | if (XFS_FORCED_SHUTDOWN(mp)) | |
1405 | return -EIO; | |
1406 | ||
1407 | lock = xfs_ilock_data_map_shared(ip); | |
1408 | ||
1409 | end = i_size_read(inode); | |
1410 | offset = __xfs_seek_hole_data(inode, start, end, whence); | |
1411 | if (offset < 0) { | |
1412 | error = offset; | |
1413 | goto out_unlock; | |
1414 | } | |
1415 | ||
46a1c2c7 | 1416 | offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes); |
3fe3e6b1 JL |
1417 | |
1418 | out_unlock: | |
01f4f327 | 1419 | xfs_iunlock(ip, lock); |
3fe3e6b1 JL |
1420 | |
1421 | if (error) | |
2451337d | 1422 | return error; |
3fe3e6b1 JL |
1423 | return offset; |
1424 | } | |
1425 | ||
1426 | STATIC loff_t | |
1427 | xfs_file_llseek( | |
1428 | struct file *file, | |
1429 | loff_t offset, | |
59f9c004 | 1430 | int whence) |
3fe3e6b1 | 1431 | { |
59f9c004 | 1432 | switch (whence) { |
3fe3e6b1 JL |
1433 | case SEEK_END: |
1434 | case SEEK_CUR: | |
1435 | case SEEK_SET: | |
59f9c004 | 1436 | return generic_file_llseek(file, offset, whence); |
3fe3e6b1 | 1437 | case SEEK_HOLE: |
49c69591 | 1438 | case SEEK_DATA: |
59f9c004 | 1439 | return xfs_seek_hole_data(file, offset, whence); |
3fe3e6b1 JL |
1440 | default: |
1441 | return -EINVAL; | |
1442 | } | |
1443 | } | |
1444 | ||
de0e8c20 DC |
1445 | /* |
1446 | * Locking for serialisation of IO during page faults. This results in a lock | |
1447 | * ordering of: | |
1448 | * | |
1449 | * mmap_sem (MM) | |
6b698ede | 1450 | * sb_start_pagefault(vfs, freeze) |
13ad4fe3 | 1451 | * i_mmaplock (XFS - truncate serialisation) |
6b698ede DC |
1452 | * page_lock (MM) |
1453 | * i_lock (XFS - extent map serialisation) | |
de0e8c20 | 1454 | */ |
de0e8c20 | 1455 | |
075a924d DC |
1456 | /* |
1457 | * mmap()d file has taken write protection fault and is being made writable. We | |
1458 | * can set the page state up correctly for a writable page, which means we can | |
1459 | * do correct delalloc accounting (ENOSPC checking!) and unwritten extent | |
1460 | * mapping. | |
de0e8c20 DC |
1461 | */ |
1462 | STATIC int | |
075a924d | 1463 | xfs_filemap_page_mkwrite( |
de0e8c20 DC |
1464 | struct vm_area_struct *vma, |
1465 | struct vm_fault *vmf) | |
1466 | { | |
6b698ede | 1467 | struct inode *inode = file_inode(vma->vm_file); |
ec56b1f1 | 1468 | int ret; |
de0e8c20 | 1469 | |
6b698ede | 1470 | trace_xfs_filemap_page_mkwrite(XFS_I(inode)); |
de0e8c20 | 1471 | |
6b698ede | 1472 | sb_start_pagefault(inode->i_sb); |
ec56b1f1 | 1473 | file_update_time(vma->vm_file); |
6b698ede | 1474 | xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED); |
de0e8c20 | 1475 | |
6b698ede | 1476 | if (IS_DAX(inode)) { |
6b524995 | 1477 | ret = dax_mkwrite(vma, vmf, xfs_get_blocks_dax_fault); |
6b698ede | 1478 | } else { |
68a9f5e7 | 1479 | ret = iomap_page_mkwrite(vma, vmf, &xfs_iomap_ops); |
6b698ede DC |
1480 | ret = block_page_mkwrite_return(ret); |
1481 | } | |
1482 | ||
1483 | xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED); | |
1484 | sb_end_pagefault(inode->i_sb); | |
1485 | ||
1486 | return ret; | |
de0e8c20 DC |
1487 | } |
1488 | ||
075a924d | 1489 | STATIC int |
6b698ede | 1490 | xfs_filemap_fault( |
075a924d DC |
1491 | struct vm_area_struct *vma, |
1492 | struct vm_fault *vmf) | |
1493 | { | |
b2442c5a | 1494 | struct inode *inode = file_inode(vma->vm_file); |
6b698ede | 1495 | int ret; |
ec56b1f1 | 1496 | |
b2442c5a | 1497 | trace_xfs_filemap_fault(XFS_I(inode)); |
075a924d | 1498 | |
6b698ede | 1499 | /* DAX can shortcut the normal fault path on write faults! */ |
b2442c5a | 1500 | if ((vmf->flags & FAULT_FLAG_WRITE) && IS_DAX(inode)) |
6b698ede | 1501 | return xfs_filemap_page_mkwrite(vma, vmf); |
075a924d | 1502 | |
b2442c5a DC |
1503 | xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED); |
1504 | if (IS_DAX(inode)) { | |
1505 | /* | |
1506 | * we do not want to trigger unwritten extent conversion on read | |
1507 | * faults - that is unnecessary overhead and would also require | |
1508 | * changes to xfs_get_blocks_direct() to map unwritten extent | |
1509 | * ioend for conversion on read-only mappings. | |
1510 | */ | |
6b524995 | 1511 | ret = dax_fault(vma, vmf, xfs_get_blocks_dax_fault); |
b2442c5a DC |
1512 | } else |
1513 | ret = filemap_fault(vma, vmf); | |
1514 | xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED); | |
075a924d | 1515 | |
6b698ede DC |
1516 | return ret; |
1517 | } | |
1518 | ||
13ad4fe3 DC |
1519 | /* |
1520 | * Similar to xfs_filemap_fault(), the DAX fault path can call into here on | |
1521 | * both read and write faults. Hence we need to handle both cases. There is no | |
1522 | * ->pmd_mkwrite callout for huge pages, so we have a single function here to | |
1523 | * handle both cases here. @flags carries the information on the type of fault | |
1524 | * occuring. | |
1525 | */ | |
acd76e74 MW |
1526 | STATIC int |
1527 | xfs_filemap_pmd_fault( | |
1528 | struct vm_area_struct *vma, | |
1529 | unsigned long addr, | |
1530 | pmd_t *pmd, | |
1531 | unsigned int flags) | |
1532 | { | |
1533 | struct inode *inode = file_inode(vma->vm_file); | |
1534 | struct xfs_inode *ip = XFS_I(inode); | |
1535 | int ret; | |
1536 | ||
1537 | if (!IS_DAX(inode)) | |
1538 | return VM_FAULT_FALLBACK; | |
1539 | ||
1540 | trace_xfs_filemap_pmd_fault(ip); | |
1541 | ||
13ad4fe3 DC |
1542 | if (flags & FAULT_FLAG_WRITE) { |
1543 | sb_start_pagefault(inode->i_sb); | |
1544 | file_update_time(vma->vm_file); | |
1545 | } | |
1546 | ||
acd76e74 | 1547 | xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED); |
6b524995 | 1548 | ret = dax_pmd_fault(vma, addr, pmd, flags, xfs_get_blocks_dax_fault); |
acd76e74 | 1549 | xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED); |
acd76e74 | 1550 | |
13ad4fe3 DC |
1551 | if (flags & FAULT_FLAG_WRITE) |
1552 | sb_end_pagefault(inode->i_sb); | |
acd76e74 MW |
1553 | |
1554 | return ret; | |
1555 | } | |
1556 | ||
3af49285 DC |
1557 | /* |
1558 | * pfn_mkwrite was originally inteneded to ensure we capture time stamp | |
1559 | * updates on write faults. In reality, it's need to serialise against | |
5eb88dca RZ |
1560 | * truncate similar to page_mkwrite. Hence we cycle the XFS_MMAPLOCK_SHARED |
1561 | * to ensure we serialise the fault barrier in place. | |
3af49285 DC |
1562 | */ |
1563 | static int | |
1564 | xfs_filemap_pfn_mkwrite( | |
1565 | struct vm_area_struct *vma, | |
1566 | struct vm_fault *vmf) | |
1567 | { | |
1568 | ||
1569 | struct inode *inode = file_inode(vma->vm_file); | |
1570 | struct xfs_inode *ip = XFS_I(inode); | |
1571 | int ret = VM_FAULT_NOPAGE; | |
1572 | loff_t size; | |
1573 | ||
1574 | trace_xfs_filemap_pfn_mkwrite(ip); | |
1575 | ||
1576 | sb_start_pagefault(inode->i_sb); | |
1577 | file_update_time(vma->vm_file); | |
1578 | ||
1579 | /* check if the faulting page hasn't raced with truncate */ | |
1580 | xfs_ilock(ip, XFS_MMAPLOCK_SHARED); | |
1581 | size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
1582 | if (vmf->pgoff >= size) | |
1583 | ret = VM_FAULT_SIGBUS; | |
5eb88dca RZ |
1584 | else if (IS_DAX(inode)) |
1585 | ret = dax_pfn_mkwrite(vma, vmf); | |
3af49285 DC |
1586 | xfs_iunlock(ip, XFS_MMAPLOCK_SHARED); |
1587 | sb_end_pagefault(inode->i_sb); | |
acd76e74 | 1588 | return ret; |
3af49285 | 1589 | |
acd76e74 MW |
1590 | } |
1591 | ||
6b698ede DC |
1592 | static const struct vm_operations_struct xfs_file_vm_ops = { |
1593 | .fault = xfs_filemap_fault, | |
acd76e74 | 1594 | .pmd_fault = xfs_filemap_pmd_fault, |
6b698ede DC |
1595 | .map_pages = filemap_map_pages, |
1596 | .page_mkwrite = xfs_filemap_page_mkwrite, | |
3af49285 | 1597 | .pfn_mkwrite = xfs_filemap_pfn_mkwrite, |
6b698ede DC |
1598 | }; |
1599 | ||
1600 | STATIC int | |
1601 | xfs_file_mmap( | |
1602 | struct file *filp, | |
1603 | struct vm_area_struct *vma) | |
1604 | { | |
1605 | file_accessed(filp); | |
1606 | vma->vm_ops = &xfs_file_vm_ops; | |
1607 | if (IS_DAX(file_inode(filp))) | |
acd76e74 | 1608 | vma->vm_flags |= VM_MIXEDMAP | VM_HUGEPAGE; |
6b698ede | 1609 | return 0; |
075a924d DC |
1610 | } |
1611 | ||
4b6f5d20 | 1612 | const struct file_operations xfs_file_operations = { |
3fe3e6b1 | 1613 | .llseek = xfs_file_llseek, |
b4f5d2c6 | 1614 | .read_iter = xfs_file_read_iter, |
bf97f3bc | 1615 | .write_iter = xfs_file_write_iter, |
82c156f8 | 1616 | .splice_read = generic_file_splice_read, |
8d020765 | 1617 | .splice_write = iter_file_splice_write, |
3562fd45 | 1618 | .unlocked_ioctl = xfs_file_ioctl, |
1da177e4 | 1619 | #ifdef CONFIG_COMPAT |
3562fd45 | 1620 | .compat_ioctl = xfs_file_compat_ioctl, |
1da177e4 | 1621 | #endif |
3562fd45 NS |
1622 | .mmap = xfs_file_mmap, |
1623 | .open = xfs_file_open, | |
1624 | .release = xfs_file_release, | |
1625 | .fsync = xfs_file_fsync, | |
2fe17c10 | 1626 | .fallocate = xfs_file_fallocate, |
1da177e4 LT |
1627 | }; |
1628 | ||
4b6f5d20 | 1629 | const struct file_operations xfs_dir_file_operations = { |
f999a5bf | 1630 | .open = xfs_dir_open, |
1da177e4 | 1631 | .read = generic_read_dir, |
3b0a3c1a | 1632 | .iterate_shared = xfs_file_readdir, |
59af1584 | 1633 | .llseek = generic_file_llseek, |
3562fd45 | 1634 | .unlocked_ioctl = xfs_file_ioctl, |
d3870398 | 1635 | #ifdef CONFIG_COMPAT |
3562fd45 | 1636 | .compat_ioctl = xfs_file_compat_ioctl, |
d3870398 | 1637 | #endif |
1da2f2db | 1638 | .fsync = xfs_dir_fsync, |
1da177e4 | 1639 | }; |