]>
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" |
1da177e4 LT |
40 | |
41 | #include <linux/dcache.h> | |
2fe17c10 | 42 | #include <linux/falloc.h> |
d126d43f | 43 | #include <linux/pagevec.h> |
66114cad | 44 | #include <linux/backing-dev.h> |
1da177e4 | 45 | |
f0f37e2f | 46 | static const struct vm_operations_struct xfs_file_vm_ops; |
1da177e4 | 47 | |
487f84f3 DC |
48 | /* |
49 | * Locking primitives for read and write IO paths to ensure we consistently use | |
50 | * and order the inode->i_mutex, ip->i_lock and ip->i_iolock. | |
51 | */ | |
52 | static inline void | |
53 | xfs_rw_ilock( | |
54 | struct xfs_inode *ip, | |
55 | int type) | |
56 | { | |
57 | if (type & XFS_IOLOCK_EXCL) | |
58 | mutex_lock(&VFS_I(ip)->i_mutex); | |
59 | xfs_ilock(ip, type); | |
60 | } | |
61 | ||
62 | static inline void | |
63 | xfs_rw_iunlock( | |
64 | struct xfs_inode *ip, | |
65 | int type) | |
66 | { | |
67 | xfs_iunlock(ip, type); | |
68 | if (type & XFS_IOLOCK_EXCL) | |
69 | mutex_unlock(&VFS_I(ip)->i_mutex); | |
70 | } | |
71 | ||
72 | static inline void | |
73 | xfs_rw_ilock_demote( | |
74 | struct xfs_inode *ip, | |
75 | int type) | |
76 | { | |
77 | xfs_ilock_demote(ip, type); | |
78 | if (type & XFS_IOLOCK_EXCL) | |
79 | mutex_unlock(&VFS_I(ip)->i_mutex); | |
80 | } | |
81 | ||
dda35b8f | 82 | /* |
4f69f578 DC |
83 | * xfs_iozero clears the specified range supplied via the page cache (except in |
84 | * the DAX case). Writes through the page cache will allocate blocks over holes, | |
85 | * though the callers usually map the holes first and avoid them. If a block is | |
86 | * not completely zeroed, then it will be read from disk before being partially | |
87 | * zeroed. | |
dda35b8f | 88 | * |
4f69f578 DC |
89 | * In the DAX case, we can just directly write to the underlying pages. This |
90 | * will not allocate blocks, but will avoid holes and unwritten extents and so | |
91 | * not do unnecessary work. | |
dda35b8f | 92 | */ |
ef9d8733 | 93 | int |
dda35b8f CH |
94 | xfs_iozero( |
95 | struct xfs_inode *ip, /* inode */ | |
96 | loff_t pos, /* offset in file */ | |
97 | size_t count) /* size of data to zero */ | |
98 | { | |
99 | struct page *page; | |
100 | struct address_space *mapping; | |
4f69f578 DC |
101 | int status = 0; |
102 | ||
dda35b8f CH |
103 | |
104 | mapping = VFS_I(ip)->i_mapping; | |
105 | do { | |
106 | unsigned offset, bytes; | |
107 | void *fsdata; | |
108 | ||
109 | offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */ | |
110 | bytes = PAGE_CACHE_SIZE - offset; | |
111 | if (bytes > count) | |
112 | bytes = count; | |
113 | ||
4f69f578 DC |
114 | if (IS_DAX(VFS_I(ip))) { |
115 | status = dax_zero_page_range(VFS_I(ip), pos, bytes, | |
116 | xfs_get_blocks_direct); | |
117 | if (status) | |
118 | break; | |
119 | } else { | |
120 | status = pagecache_write_begin(NULL, mapping, pos, bytes, | |
121 | AOP_FLAG_UNINTERRUPTIBLE, | |
122 | &page, &fsdata); | |
123 | if (status) | |
124 | break; | |
dda35b8f | 125 | |
4f69f578 | 126 | zero_user(page, offset, bytes); |
dda35b8f | 127 | |
4f69f578 DC |
128 | status = pagecache_write_end(NULL, mapping, pos, bytes, |
129 | bytes, page, fsdata); | |
130 | WARN_ON(status <= 0); /* can't return less than zero! */ | |
131 | status = 0; | |
132 | } | |
dda35b8f CH |
133 | pos += bytes; |
134 | count -= bytes; | |
dda35b8f CH |
135 | } while (count); |
136 | ||
cddc1162 | 137 | return status; |
dda35b8f CH |
138 | } |
139 | ||
8add71ca CH |
140 | int |
141 | xfs_update_prealloc_flags( | |
142 | struct xfs_inode *ip, | |
143 | enum xfs_prealloc_flags flags) | |
144 | { | |
145 | struct xfs_trans *tp; | |
146 | int error; | |
147 | ||
148 | tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_WRITEID); | |
149 | error = xfs_trans_reserve(tp, &M_RES(ip->i_mount)->tr_writeid, 0, 0); | |
150 | if (error) { | |
4906e215 | 151 | xfs_trans_cancel(tp); |
8add71ca CH |
152 | return error; |
153 | } | |
154 | ||
155 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
156 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
157 | ||
158 | if (!(flags & XFS_PREALLOC_INVISIBLE)) { | |
159 | ip->i_d.di_mode &= ~S_ISUID; | |
160 | if (ip->i_d.di_mode & S_IXGRP) | |
161 | ip->i_d.di_mode &= ~S_ISGID; | |
162 | xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
163 | } | |
164 | ||
165 | if (flags & XFS_PREALLOC_SET) | |
166 | ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC; | |
167 | if (flags & XFS_PREALLOC_CLEAR) | |
168 | ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC; | |
169 | ||
170 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
171 | if (flags & XFS_PREALLOC_SYNC) | |
172 | xfs_trans_set_sync(tp); | |
70393313 | 173 | return xfs_trans_commit(tp); |
8add71ca CH |
174 | } |
175 | ||
1da2f2db CH |
176 | /* |
177 | * Fsync operations on directories are much simpler than on regular files, | |
178 | * as there is no file data to flush, and thus also no need for explicit | |
179 | * cache flush operations, and there are no non-transaction metadata updates | |
180 | * on directories either. | |
181 | */ | |
182 | STATIC int | |
183 | xfs_dir_fsync( | |
184 | struct file *file, | |
185 | loff_t start, | |
186 | loff_t end, | |
187 | int datasync) | |
188 | { | |
189 | struct xfs_inode *ip = XFS_I(file->f_mapping->host); | |
190 | struct xfs_mount *mp = ip->i_mount; | |
191 | xfs_lsn_t lsn = 0; | |
192 | ||
193 | trace_xfs_dir_fsync(ip); | |
194 | ||
195 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
196 | if (xfs_ipincount(ip)) | |
197 | lsn = ip->i_itemp->ili_last_lsn; | |
198 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
199 | ||
200 | if (!lsn) | |
201 | return 0; | |
2451337d | 202 | return _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, NULL); |
1da2f2db CH |
203 | } |
204 | ||
fd3200be CH |
205 | STATIC int |
206 | xfs_file_fsync( | |
207 | struct file *file, | |
02c24a82 JB |
208 | loff_t start, |
209 | loff_t end, | |
fd3200be CH |
210 | int datasync) |
211 | { | |
7ea80859 CH |
212 | struct inode *inode = file->f_mapping->host; |
213 | struct xfs_inode *ip = XFS_I(inode); | |
a27a263b | 214 | struct xfs_mount *mp = ip->i_mount; |
fd3200be CH |
215 | int error = 0; |
216 | int log_flushed = 0; | |
b1037058 | 217 | xfs_lsn_t lsn = 0; |
fd3200be | 218 | |
cca28fb8 | 219 | trace_xfs_file_fsync(ip); |
fd3200be | 220 | |
02c24a82 JB |
221 | error = filemap_write_and_wait_range(inode->i_mapping, start, end); |
222 | if (error) | |
223 | return error; | |
224 | ||
a27a263b | 225 | if (XFS_FORCED_SHUTDOWN(mp)) |
b474c7ae | 226 | return -EIO; |
fd3200be CH |
227 | |
228 | xfs_iflags_clear(ip, XFS_ITRUNCATED); | |
229 | ||
a27a263b CH |
230 | if (mp->m_flags & XFS_MOUNT_BARRIER) { |
231 | /* | |
232 | * If we have an RT and/or log subvolume we need to make sure | |
233 | * to flush the write cache the device used for file data | |
234 | * first. This is to ensure newly written file data make | |
235 | * it to disk before logging the new inode size in case of | |
236 | * an extending write. | |
237 | */ | |
238 | if (XFS_IS_REALTIME_INODE(ip)) | |
239 | xfs_blkdev_issue_flush(mp->m_rtdev_targp); | |
240 | else if (mp->m_logdev_targp != mp->m_ddev_targp) | |
241 | xfs_blkdev_issue_flush(mp->m_ddev_targp); | |
242 | } | |
243 | ||
fd3200be | 244 | /* |
8a9c9980 CH |
245 | * All metadata updates are logged, which means that we just have |
246 | * to flush the log up to the latest LSN that touched the inode. | |
fd3200be CH |
247 | */ |
248 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
8f639dde CH |
249 | if (xfs_ipincount(ip)) { |
250 | if (!datasync || | |
251 | (ip->i_itemp->ili_fields & ~XFS_ILOG_TIMESTAMP)) | |
252 | lsn = ip->i_itemp->ili_last_lsn; | |
253 | } | |
8a9c9980 | 254 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
fd3200be | 255 | |
8a9c9980 | 256 | if (lsn) |
b1037058 CH |
257 | error = _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, &log_flushed); |
258 | ||
a27a263b CH |
259 | /* |
260 | * If we only have a single device, and the log force about was | |
261 | * a no-op we might have to flush the data device cache here. | |
262 | * This can only happen for fdatasync/O_DSYNC if we were overwriting | |
263 | * an already allocated file and thus do not have any metadata to | |
264 | * commit. | |
265 | */ | |
266 | if ((mp->m_flags & XFS_MOUNT_BARRIER) && | |
267 | mp->m_logdev_targp == mp->m_ddev_targp && | |
268 | !XFS_IS_REALTIME_INODE(ip) && | |
269 | !log_flushed) | |
270 | xfs_blkdev_issue_flush(mp->m_ddev_targp); | |
fd3200be | 271 | |
2451337d | 272 | return error; |
fd3200be CH |
273 | } |
274 | ||
00258e36 | 275 | STATIC ssize_t |
b4f5d2c6 | 276 | xfs_file_read_iter( |
dda35b8f | 277 | struct kiocb *iocb, |
b4f5d2c6 | 278 | struct iov_iter *to) |
dda35b8f CH |
279 | { |
280 | struct file *file = iocb->ki_filp; | |
281 | struct inode *inode = file->f_mapping->host; | |
00258e36 CH |
282 | struct xfs_inode *ip = XFS_I(inode); |
283 | struct xfs_mount *mp = ip->i_mount; | |
b4f5d2c6 | 284 | size_t size = iov_iter_count(to); |
dda35b8f | 285 | ssize_t ret = 0; |
00258e36 | 286 | int ioflags = 0; |
dda35b8f | 287 | xfs_fsize_t n; |
b4f5d2c6 | 288 | loff_t pos = iocb->ki_pos; |
dda35b8f | 289 | |
dda35b8f CH |
290 | XFS_STATS_INC(xs_read_calls); |
291 | ||
2ba48ce5 | 292 | if (unlikely(iocb->ki_flags & IOCB_DIRECT)) |
b92cc59f | 293 | ioflags |= XFS_IO_ISDIRECT; |
00258e36 | 294 | if (file->f_mode & FMODE_NOCMTIME) |
b92cc59f | 295 | ioflags |= XFS_IO_INVIS; |
00258e36 | 296 | |
6b698ede | 297 | if ((ioflags & XFS_IO_ISDIRECT) && !IS_DAX(inode)) { |
dda35b8f CH |
298 | xfs_buftarg_t *target = |
299 | XFS_IS_REALTIME_INODE(ip) ? | |
300 | mp->m_rtdev_targp : mp->m_ddev_targp; | |
7c71ee78 ES |
301 | /* DIO must be aligned to device logical sector size */ |
302 | if ((pos | size) & target->bt_logical_sectormask) { | |
fb595814 | 303 | if (pos == i_size_read(inode)) |
00258e36 | 304 | return 0; |
b474c7ae | 305 | return -EINVAL; |
dda35b8f CH |
306 | } |
307 | } | |
308 | ||
fb595814 | 309 | n = mp->m_super->s_maxbytes - pos; |
00258e36 | 310 | if (n <= 0 || size == 0) |
dda35b8f CH |
311 | return 0; |
312 | ||
313 | if (n < size) | |
314 | size = n; | |
315 | ||
316 | if (XFS_FORCED_SHUTDOWN(mp)) | |
317 | return -EIO; | |
318 | ||
0c38a251 DC |
319 | /* |
320 | * Locking is a bit tricky here. If we take an exclusive lock | |
321 | * for direct IO, we effectively serialise all new concurrent | |
322 | * read IO to this file and block it behind IO that is currently in | |
323 | * progress because IO in progress holds the IO lock shared. We only | |
324 | * need to hold the lock exclusive to blow away the page cache, so | |
325 | * only take lock exclusively if the page cache needs invalidation. | |
326 | * This allows the normal direct IO case of no page cache pages to | |
327 | * proceeed concurrently without serialisation. | |
328 | */ | |
329 | xfs_rw_ilock(ip, XFS_IOLOCK_SHARED); | |
b92cc59f | 330 | if ((ioflags & XFS_IO_ISDIRECT) && inode->i_mapping->nrpages) { |
0c38a251 | 331 | xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED); |
487f84f3 DC |
332 | xfs_rw_ilock(ip, XFS_IOLOCK_EXCL); |
333 | ||
00258e36 | 334 | if (inode->i_mapping->nrpages) { |
8ff1e670 | 335 | ret = filemap_write_and_wait_range( |
fb595814 | 336 | VFS_I(ip)->i_mapping, |
7d4ea3ce | 337 | pos, pos + size - 1); |
487f84f3 DC |
338 | if (ret) { |
339 | xfs_rw_iunlock(ip, XFS_IOLOCK_EXCL); | |
340 | return ret; | |
341 | } | |
85e584da CM |
342 | |
343 | /* | |
344 | * Invalidate whole pages. This can return an error if | |
345 | * we fail to invalidate a page, but this should never | |
346 | * happen on XFS. Warn if it does fail. | |
347 | */ | |
348 | ret = invalidate_inode_pages2_range(VFS_I(ip)->i_mapping, | |
7d4ea3ce DC |
349 | pos >> PAGE_CACHE_SHIFT, |
350 | (pos + size - 1) >> PAGE_CACHE_SHIFT); | |
85e584da CM |
351 | WARN_ON_ONCE(ret); |
352 | ret = 0; | |
00258e36 | 353 | } |
487f84f3 | 354 | xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL); |
0c38a251 | 355 | } |
dda35b8f | 356 | |
fb595814 | 357 | trace_xfs_file_read(ip, size, pos, ioflags); |
dda35b8f | 358 | |
b4f5d2c6 | 359 | ret = generic_file_read_iter(iocb, to); |
dda35b8f CH |
360 | if (ret > 0) |
361 | XFS_STATS_ADD(xs_read_bytes, ret); | |
362 | ||
487f84f3 | 363 | xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED); |
dda35b8f CH |
364 | return ret; |
365 | } | |
366 | ||
00258e36 CH |
367 | STATIC ssize_t |
368 | xfs_file_splice_read( | |
dda35b8f CH |
369 | struct file *infilp, |
370 | loff_t *ppos, | |
371 | struct pipe_inode_info *pipe, | |
372 | size_t count, | |
00258e36 | 373 | unsigned int flags) |
dda35b8f | 374 | { |
00258e36 | 375 | struct xfs_inode *ip = XFS_I(infilp->f_mapping->host); |
00258e36 | 376 | int ioflags = 0; |
dda35b8f CH |
377 | ssize_t ret; |
378 | ||
379 | XFS_STATS_INC(xs_read_calls); | |
00258e36 CH |
380 | |
381 | if (infilp->f_mode & FMODE_NOCMTIME) | |
b92cc59f | 382 | ioflags |= XFS_IO_INVIS; |
00258e36 | 383 | |
dda35b8f CH |
384 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
385 | return -EIO; | |
386 | ||
487f84f3 | 387 | xfs_rw_ilock(ip, XFS_IOLOCK_SHARED); |
dda35b8f | 388 | |
dda35b8f CH |
389 | trace_xfs_file_splice_read(ip, count, *ppos, ioflags); |
390 | ||
6b698ede DC |
391 | /* for dax, we need to avoid the page cache */ |
392 | if (IS_DAX(VFS_I(ip))) | |
393 | ret = default_file_splice_read(infilp, ppos, pipe, count, flags); | |
394 | else | |
395 | ret = generic_file_splice_read(infilp, ppos, pipe, count, flags); | |
dda35b8f CH |
396 | if (ret > 0) |
397 | XFS_STATS_ADD(xs_read_bytes, ret); | |
398 | ||
487f84f3 | 399 | xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED); |
dda35b8f CH |
400 | return ret; |
401 | } | |
402 | ||
dda35b8f | 403 | /* |
193aec10 CH |
404 | * This routine is called to handle zeroing any space in the last block of the |
405 | * file that is beyond the EOF. We do this since the size is being increased | |
406 | * without writing anything to that block and we don't want to read the | |
407 | * garbage on the disk. | |
dda35b8f CH |
408 | */ |
409 | STATIC int /* error (positive) */ | |
410 | xfs_zero_last_block( | |
193aec10 CH |
411 | struct xfs_inode *ip, |
412 | xfs_fsize_t offset, | |
5885ebda DC |
413 | xfs_fsize_t isize, |
414 | bool *did_zeroing) | |
dda35b8f | 415 | { |
193aec10 CH |
416 | struct xfs_mount *mp = ip->i_mount; |
417 | xfs_fileoff_t last_fsb = XFS_B_TO_FSBT(mp, isize); | |
418 | int zero_offset = XFS_B_FSB_OFFSET(mp, isize); | |
419 | int zero_len; | |
420 | int nimaps = 1; | |
421 | int error = 0; | |
422 | struct xfs_bmbt_irec imap; | |
dda35b8f | 423 | |
193aec10 | 424 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
5c8ed202 | 425 | error = xfs_bmapi_read(ip, last_fsb, 1, &imap, &nimaps, 0); |
193aec10 | 426 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
5c8ed202 | 427 | if (error) |
dda35b8f | 428 | return error; |
193aec10 | 429 | |
dda35b8f | 430 | ASSERT(nimaps > 0); |
193aec10 | 431 | |
dda35b8f CH |
432 | /* |
433 | * If the block underlying isize is just a hole, then there | |
434 | * is nothing to zero. | |
435 | */ | |
193aec10 | 436 | if (imap.br_startblock == HOLESTARTBLOCK) |
dda35b8f | 437 | return 0; |
dda35b8f CH |
438 | |
439 | zero_len = mp->m_sb.sb_blocksize - zero_offset; | |
440 | if (isize + zero_len > offset) | |
441 | zero_len = offset - isize; | |
5885ebda | 442 | *did_zeroing = true; |
193aec10 | 443 | return xfs_iozero(ip, isize, zero_len); |
dda35b8f CH |
444 | } |
445 | ||
446 | /* | |
193aec10 CH |
447 | * Zero any on disk space between the current EOF and the new, larger EOF. |
448 | * | |
449 | * This handles the normal case of zeroing the remainder of the last block in | |
450 | * the file and the unusual case of zeroing blocks out beyond the size of the | |
451 | * file. This second case only happens with fixed size extents and when the | |
452 | * system crashes before the inode size was updated but after blocks were | |
453 | * allocated. | |
454 | * | |
455 | * Expects the iolock to be held exclusive, and will take the ilock internally. | |
dda35b8f | 456 | */ |
dda35b8f CH |
457 | int /* error (positive) */ |
458 | xfs_zero_eof( | |
193aec10 CH |
459 | struct xfs_inode *ip, |
460 | xfs_off_t offset, /* starting I/O offset */ | |
5885ebda DC |
461 | xfs_fsize_t isize, /* current inode size */ |
462 | bool *did_zeroing) | |
dda35b8f | 463 | { |
193aec10 CH |
464 | struct xfs_mount *mp = ip->i_mount; |
465 | xfs_fileoff_t start_zero_fsb; | |
466 | xfs_fileoff_t end_zero_fsb; | |
467 | xfs_fileoff_t zero_count_fsb; | |
468 | xfs_fileoff_t last_fsb; | |
469 | xfs_fileoff_t zero_off; | |
470 | xfs_fsize_t zero_len; | |
471 | int nimaps; | |
472 | int error = 0; | |
473 | struct xfs_bmbt_irec imap; | |
474 | ||
475 | ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL)); | |
dda35b8f CH |
476 | ASSERT(offset > isize); |
477 | ||
478 | /* | |
479 | * First handle zeroing the block on which isize resides. | |
193aec10 | 480 | * |
dda35b8f CH |
481 | * We only zero a part of that block so it is handled specially. |
482 | */ | |
193aec10 | 483 | if (XFS_B_FSB_OFFSET(mp, isize) != 0) { |
5885ebda | 484 | error = xfs_zero_last_block(ip, offset, isize, did_zeroing); |
193aec10 CH |
485 | if (error) |
486 | return error; | |
dda35b8f CH |
487 | } |
488 | ||
489 | /* | |
193aec10 CH |
490 | * Calculate the range between the new size and the old where blocks |
491 | * needing to be zeroed may exist. | |
492 | * | |
493 | * To get the block where the last byte in the file currently resides, | |
494 | * we need to subtract one from the size and truncate back to a block | |
495 | * boundary. We subtract 1 in case the size is exactly on a block | |
496 | * boundary. | |
dda35b8f CH |
497 | */ |
498 | last_fsb = isize ? XFS_B_TO_FSBT(mp, isize - 1) : (xfs_fileoff_t)-1; | |
499 | start_zero_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize); | |
500 | end_zero_fsb = XFS_B_TO_FSBT(mp, offset - 1); | |
501 | ASSERT((xfs_sfiloff_t)last_fsb < (xfs_sfiloff_t)start_zero_fsb); | |
502 | if (last_fsb == end_zero_fsb) { | |
503 | /* | |
504 | * The size was only incremented on its last block. | |
505 | * We took care of that above, so just return. | |
506 | */ | |
507 | return 0; | |
508 | } | |
509 | ||
510 | ASSERT(start_zero_fsb <= end_zero_fsb); | |
511 | while (start_zero_fsb <= end_zero_fsb) { | |
512 | nimaps = 1; | |
513 | zero_count_fsb = end_zero_fsb - start_zero_fsb + 1; | |
193aec10 CH |
514 | |
515 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
5c8ed202 DC |
516 | error = xfs_bmapi_read(ip, start_zero_fsb, zero_count_fsb, |
517 | &imap, &nimaps, 0); | |
193aec10 CH |
518 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
519 | if (error) | |
dda35b8f | 520 | return error; |
193aec10 | 521 | |
dda35b8f CH |
522 | ASSERT(nimaps > 0); |
523 | ||
524 | if (imap.br_state == XFS_EXT_UNWRITTEN || | |
525 | imap.br_startblock == HOLESTARTBLOCK) { | |
dda35b8f CH |
526 | start_zero_fsb = imap.br_startoff + imap.br_blockcount; |
527 | ASSERT(start_zero_fsb <= (end_zero_fsb + 1)); | |
528 | continue; | |
529 | } | |
530 | ||
531 | /* | |
532 | * There are blocks we need to zero. | |
dda35b8f | 533 | */ |
dda35b8f CH |
534 | zero_off = XFS_FSB_TO_B(mp, start_zero_fsb); |
535 | zero_len = XFS_FSB_TO_B(mp, imap.br_blockcount); | |
536 | ||
537 | if ((zero_off + zero_len) > offset) | |
538 | zero_len = offset - zero_off; | |
539 | ||
540 | error = xfs_iozero(ip, zero_off, zero_len); | |
193aec10 CH |
541 | if (error) |
542 | return error; | |
dda35b8f | 543 | |
5885ebda | 544 | *did_zeroing = true; |
dda35b8f CH |
545 | start_zero_fsb = imap.br_startoff + imap.br_blockcount; |
546 | ASSERT(start_zero_fsb <= (end_zero_fsb + 1)); | |
dda35b8f CH |
547 | } |
548 | ||
549 | return 0; | |
dda35b8f CH |
550 | } |
551 | ||
4d8d1581 DC |
552 | /* |
553 | * Common pre-write limit and setup checks. | |
554 | * | |
5bf1f262 CH |
555 | * Called with the iolocked held either shared and exclusive according to |
556 | * @iolock, and returns with it held. Might upgrade the iolock to exclusive | |
557 | * if called for a direct write beyond i_size. | |
4d8d1581 DC |
558 | */ |
559 | STATIC ssize_t | |
560 | xfs_file_aio_write_checks( | |
99733fa3 AV |
561 | struct kiocb *iocb, |
562 | struct iov_iter *from, | |
4d8d1581 DC |
563 | int *iolock) |
564 | { | |
99733fa3 | 565 | struct file *file = iocb->ki_filp; |
4d8d1581 DC |
566 | struct inode *inode = file->f_mapping->host; |
567 | struct xfs_inode *ip = XFS_I(inode); | |
3309dd04 | 568 | ssize_t error = 0; |
99733fa3 | 569 | size_t count = iov_iter_count(from); |
4d8d1581 | 570 | |
7271d243 | 571 | restart: |
3309dd04 AV |
572 | error = generic_write_checks(iocb, from); |
573 | if (error <= 0) | |
4d8d1581 | 574 | return error; |
4d8d1581 | 575 | |
21c3ea18 | 576 | error = xfs_break_layouts(inode, iolock, true); |
781355c6 CH |
577 | if (error) |
578 | return error; | |
579 | ||
a6de82ca JK |
580 | /* For changing security info in file_remove_privs() we need i_mutex */ |
581 | if (*iolock == XFS_IOLOCK_SHARED && !IS_NOSEC(inode)) { | |
582 | xfs_rw_iunlock(ip, *iolock); | |
583 | *iolock = XFS_IOLOCK_EXCL; | |
584 | xfs_rw_ilock(ip, *iolock); | |
585 | goto restart; | |
586 | } | |
4d8d1581 DC |
587 | /* |
588 | * If the offset is beyond the size of the file, we need to zero any | |
589 | * blocks that fall between the existing EOF and the start of this | |
2813d682 | 590 | * write. If zeroing is needed and we are currently holding the |
467f7899 CH |
591 | * iolock shared, we need to update it to exclusive which implies |
592 | * having to redo all checks before. | |
b9d59846 DC |
593 | * |
594 | * We need to serialise against EOF updates that occur in IO | |
595 | * completions here. We want to make sure that nobody is changing the | |
596 | * size while we do this check until we have placed an IO barrier (i.e. | |
597 | * hold the XFS_IOLOCK_EXCL) that prevents new IO from being dispatched. | |
598 | * The spinlock effectively forms a memory barrier once we have the | |
599 | * XFS_IOLOCK_EXCL so we are guaranteed to see the latest EOF value | |
600 | * and hence be able to correctly determine if we need to run zeroing. | |
4d8d1581 | 601 | */ |
b9d59846 | 602 | spin_lock(&ip->i_flags_lock); |
99733fa3 | 603 | if (iocb->ki_pos > i_size_read(inode)) { |
5885ebda DC |
604 | bool zero = false; |
605 | ||
b9d59846 | 606 | spin_unlock(&ip->i_flags_lock); |
7271d243 | 607 | if (*iolock == XFS_IOLOCK_SHARED) { |
467f7899 | 608 | xfs_rw_iunlock(ip, *iolock); |
7271d243 | 609 | *iolock = XFS_IOLOCK_EXCL; |
467f7899 | 610 | xfs_rw_ilock(ip, *iolock); |
3309dd04 | 611 | iov_iter_reexpand(from, count); |
40c63fbc DC |
612 | |
613 | /* | |
614 | * We now have an IO submission barrier in place, but | |
615 | * AIO can do EOF updates during IO completion and hence | |
616 | * we now need to wait for all of them to drain. Non-AIO | |
617 | * DIO will have drained before we are given the | |
618 | * XFS_IOLOCK_EXCL, and so for most cases this wait is a | |
619 | * no-op. | |
620 | */ | |
621 | inode_dio_wait(inode); | |
7271d243 DC |
622 | goto restart; |
623 | } | |
99733fa3 | 624 | error = xfs_zero_eof(ip, iocb->ki_pos, i_size_read(inode), &zero); |
467f7899 CH |
625 | if (error) |
626 | return error; | |
b9d59846 DC |
627 | } else |
628 | spin_unlock(&ip->i_flags_lock); | |
4d8d1581 | 629 | |
8a9c9980 CH |
630 | /* |
631 | * Updating the timestamps will grab the ilock again from | |
632 | * xfs_fs_dirty_inode, so we have to call it after dropping the | |
633 | * lock above. Eventually we should look into a way to avoid | |
634 | * the pointless lock roundtrip. | |
635 | */ | |
c3b2da31 JB |
636 | if (likely(!(file->f_mode & FMODE_NOCMTIME))) { |
637 | error = file_update_time(file); | |
638 | if (error) | |
639 | return error; | |
640 | } | |
8a9c9980 | 641 | |
4d8d1581 DC |
642 | /* |
643 | * If we're writing the file then make sure to clear the setuid and | |
644 | * setgid bits if the process is not being run by root. This keeps | |
645 | * people from modifying setuid and setgid binaries. | |
646 | */ | |
a6de82ca JK |
647 | if (!IS_NOSEC(inode)) |
648 | return file_remove_privs(file); | |
649 | return 0; | |
4d8d1581 DC |
650 | } |
651 | ||
f0d26e86 DC |
652 | /* |
653 | * xfs_file_dio_aio_write - handle direct IO writes | |
654 | * | |
655 | * Lock the inode appropriately to prepare for and issue a direct IO write. | |
eda77982 | 656 | * By separating it from the buffered write path we remove all the tricky to |
f0d26e86 DC |
657 | * follow locking changes and looping. |
658 | * | |
eda77982 DC |
659 | * If there are cached pages or we're extending the file, we need IOLOCK_EXCL |
660 | * until we're sure the bytes at the new EOF have been zeroed and/or the cached | |
661 | * pages are flushed out. | |
662 | * | |
663 | * In most cases the direct IO writes will be done holding IOLOCK_SHARED | |
664 | * allowing them to be done in parallel with reads and other direct IO writes. | |
665 | * However, if the IO is not aligned to filesystem blocks, the direct IO layer | |
666 | * needs to do sub-block zeroing and that requires serialisation against other | |
667 | * direct IOs to the same block. In this case we need to serialise the | |
668 | * submission of the unaligned IOs so that we don't get racing block zeroing in | |
669 | * the dio layer. To avoid the problem with aio, we also need to wait for | |
670 | * outstanding IOs to complete so that unwritten extent conversion is completed | |
671 | * before we try to map the overlapping block. This is currently implemented by | |
4a06fd26 | 672 | * hitting it with a big hammer (i.e. inode_dio_wait()). |
eda77982 | 673 | * |
f0d26e86 DC |
674 | * Returns with locks held indicated by @iolock and errors indicated by |
675 | * negative return values. | |
676 | */ | |
677 | STATIC ssize_t | |
678 | xfs_file_dio_aio_write( | |
679 | struct kiocb *iocb, | |
b3188919 | 680 | struct iov_iter *from) |
f0d26e86 DC |
681 | { |
682 | struct file *file = iocb->ki_filp; | |
683 | struct address_space *mapping = file->f_mapping; | |
684 | struct inode *inode = mapping->host; | |
685 | struct xfs_inode *ip = XFS_I(inode); | |
686 | struct xfs_mount *mp = ip->i_mount; | |
687 | ssize_t ret = 0; | |
eda77982 | 688 | int unaligned_io = 0; |
d0606464 | 689 | int iolock; |
b3188919 AV |
690 | size_t count = iov_iter_count(from); |
691 | loff_t pos = iocb->ki_pos; | |
0cefb29e DC |
692 | loff_t end; |
693 | struct iov_iter data; | |
f0d26e86 DC |
694 | struct xfs_buftarg *target = XFS_IS_REALTIME_INODE(ip) ? |
695 | mp->m_rtdev_targp : mp->m_ddev_targp; | |
696 | ||
7c71ee78 | 697 | /* DIO must be aligned to device logical sector size */ |
6b698ede | 698 | if (!IS_DAX(inode) && ((pos | count) & target->bt_logical_sectormask)) |
b474c7ae | 699 | return -EINVAL; |
f0d26e86 | 700 | |
7c71ee78 | 701 | /* "unaligned" here means not aligned to a filesystem block */ |
eda77982 DC |
702 | if ((pos & mp->m_blockmask) || ((pos + count) & mp->m_blockmask)) |
703 | unaligned_io = 1; | |
704 | ||
7271d243 DC |
705 | /* |
706 | * We don't need to take an exclusive lock unless there page cache needs | |
707 | * to be invalidated or unaligned IO is being executed. We don't need to | |
708 | * consider the EOF extension case here because | |
709 | * xfs_file_aio_write_checks() will relock the inode as necessary for | |
710 | * EOF zeroing cases and fill out the new inode size as appropriate. | |
711 | */ | |
712 | if (unaligned_io || mapping->nrpages) | |
d0606464 | 713 | iolock = XFS_IOLOCK_EXCL; |
f0d26e86 | 714 | else |
d0606464 CH |
715 | iolock = XFS_IOLOCK_SHARED; |
716 | xfs_rw_ilock(ip, iolock); | |
c58cb165 CH |
717 | |
718 | /* | |
719 | * Recheck if there are cached pages that need invalidate after we got | |
720 | * the iolock to protect against other threads adding new pages while | |
721 | * we were waiting for the iolock. | |
722 | */ | |
d0606464 CH |
723 | if (mapping->nrpages && iolock == XFS_IOLOCK_SHARED) { |
724 | xfs_rw_iunlock(ip, iolock); | |
725 | iolock = XFS_IOLOCK_EXCL; | |
726 | xfs_rw_ilock(ip, iolock); | |
c58cb165 | 727 | } |
f0d26e86 | 728 | |
99733fa3 | 729 | ret = xfs_file_aio_write_checks(iocb, from, &iolock); |
4d8d1581 | 730 | if (ret) |
d0606464 | 731 | goto out; |
99733fa3 AV |
732 | count = iov_iter_count(from); |
733 | pos = iocb->ki_pos; | |
0cefb29e | 734 | end = pos + count - 1; |
f0d26e86 DC |
735 | |
736 | if (mapping->nrpages) { | |
07d5035a | 737 | ret = filemap_write_and_wait_range(VFS_I(ip)->i_mapping, |
0cefb29e | 738 | pos, end); |
f0d26e86 | 739 | if (ret) |
d0606464 | 740 | goto out; |
834ffca6 DC |
741 | /* |
742 | * Invalidate whole pages. This can return an error if | |
743 | * we fail to invalidate a page, but this should never | |
744 | * happen on XFS. Warn if it does fail. | |
745 | */ | |
746 | ret = invalidate_inode_pages2_range(VFS_I(ip)->i_mapping, | |
7d4ea3ce | 747 | pos >> PAGE_CACHE_SHIFT, |
0cefb29e | 748 | end >> PAGE_CACHE_SHIFT); |
834ffca6 DC |
749 | WARN_ON_ONCE(ret); |
750 | ret = 0; | |
f0d26e86 DC |
751 | } |
752 | ||
eda77982 DC |
753 | /* |
754 | * If we are doing unaligned IO, wait for all other IO to drain, | |
755 | * otherwise demote the lock if we had to flush cached pages | |
756 | */ | |
757 | if (unaligned_io) | |
4a06fd26 | 758 | inode_dio_wait(inode); |
d0606464 | 759 | else if (iolock == XFS_IOLOCK_EXCL) { |
f0d26e86 | 760 | xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL); |
d0606464 | 761 | iolock = XFS_IOLOCK_SHARED; |
f0d26e86 DC |
762 | } |
763 | ||
764 | trace_xfs_file_direct_write(ip, count, iocb->ki_pos, 0); | |
f0d26e86 | 765 | |
0cefb29e | 766 | data = *from; |
1aef882f | 767 | ret = mapping->a_ops->direct_IO(iocb, &data, pos); |
0cefb29e DC |
768 | |
769 | /* see generic_file_direct_write() for why this is necessary */ | |
770 | if (mapping->nrpages) { | |
771 | invalidate_inode_pages2_range(mapping, | |
772 | pos >> PAGE_CACHE_SHIFT, | |
773 | end >> PAGE_CACHE_SHIFT); | |
774 | } | |
775 | ||
776 | if (ret > 0) { | |
777 | pos += ret; | |
778 | iov_iter_advance(from, ret); | |
779 | iocb->ki_pos = pos; | |
780 | } | |
d0606464 CH |
781 | out: |
782 | xfs_rw_iunlock(ip, iolock); | |
783 | ||
6b698ede DC |
784 | /* |
785 | * No fallback to buffered IO on errors for XFS. DAX can result in | |
786 | * partial writes, but direct IO will either complete fully or fail. | |
787 | */ | |
788 | ASSERT(ret < 0 || ret == count || IS_DAX(VFS_I(ip))); | |
f0d26e86 DC |
789 | return ret; |
790 | } | |
791 | ||
00258e36 | 792 | STATIC ssize_t |
637bbc75 | 793 | xfs_file_buffered_aio_write( |
dda35b8f | 794 | struct kiocb *iocb, |
b3188919 | 795 | struct iov_iter *from) |
dda35b8f CH |
796 | { |
797 | struct file *file = iocb->ki_filp; | |
798 | struct address_space *mapping = file->f_mapping; | |
799 | struct inode *inode = mapping->host; | |
00258e36 | 800 | struct xfs_inode *ip = XFS_I(inode); |
637bbc75 DC |
801 | ssize_t ret; |
802 | int enospc = 0; | |
d0606464 | 803 | int iolock = XFS_IOLOCK_EXCL; |
dda35b8f | 804 | |
d0606464 | 805 | xfs_rw_ilock(ip, iolock); |
dda35b8f | 806 | |
99733fa3 | 807 | ret = xfs_file_aio_write_checks(iocb, from, &iolock); |
4d8d1581 | 808 | if (ret) |
d0606464 | 809 | goto out; |
dda35b8f CH |
810 | |
811 | /* We can write back this queue in page reclaim */ | |
de1414a6 | 812 | current->backing_dev_info = inode_to_bdi(inode); |
dda35b8f | 813 | |
dda35b8f | 814 | write_retry: |
99733fa3 AV |
815 | trace_xfs_file_buffered_write(ip, iov_iter_count(from), |
816 | iocb->ki_pos, 0); | |
817 | ret = generic_perform_write(file, from, iocb->ki_pos); | |
0a64bc2c | 818 | if (likely(ret >= 0)) |
99733fa3 | 819 | iocb->ki_pos += ret; |
dc06f398 | 820 | |
637bbc75 | 821 | /* |
dc06f398 BF |
822 | * If we hit a space limit, try to free up some lingering preallocated |
823 | * space before returning an error. In the case of ENOSPC, first try to | |
824 | * write back all dirty inodes to free up some of the excess reserved | |
825 | * metadata space. This reduces the chances that the eofblocks scan | |
826 | * waits on dirty mappings. Since xfs_flush_inodes() is serialized, this | |
827 | * also behaves as a filter to prevent too many eofblocks scans from | |
828 | * running at the same time. | |
637bbc75 | 829 | */ |
dc06f398 BF |
830 | if (ret == -EDQUOT && !enospc) { |
831 | enospc = xfs_inode_free_quota_eofblocks(ip); | |
832 | if (enospc) | |
833 | goto write_retry; | |
834 | } else if (ret == -ENOSPC && !enospc) { | |
835 | struct xfs_eofblocks eofb = {0}; | |
836 | ||
637bbc75 | 837 | enospc = 1; |
9aa05000 | 838 | xfs_flush_inodes(ip->i_mount); |
dc06f398 BF |
839 | eofb.eof_scan_owner = ip->i_ino; /* for locking */ |
840 | eofb.eof_flags = XFS_EOF_FLAGS_SYNC; | |
841 | xfs_icache_free_eofblocks(ip->i_mount, &eofb); | |
9aa05000 | 842 | goto write_retry; |
dda35b8f | 843 | } |
d0606464 | 844 | |
dda35b8f | 845 | current->backing_dev_info = NULL; |
d0606464 CH |
846 | out: |
847 | xfs_rw_iunlock(ip, iolock); | |
637bbc75 DC |
848 | return ret; |
849 | } | |
850 | ||
851 | STATIC ssize_t | |
bf97f3bc | 852 | xfs_file_write_iter( |
637bbc75 | 853 | struct kiocb *iocb, |
bf97f3bc | 854 | struct iov_iter *from) |
637bbc75 DC |
855 | { |
856 | struct file *file = iocb->ki_filp; | |
857 | struct address_space *mapping = file->f_mapping; | |
858 | struct inode *inode = mapping->host; | |
859 | struct xfs_inode *ip = XFS_I(inode); | |
860 | ssize_t ret; | |
bf97f3bc | 861 | size_t ocount = iov_iter_count(from); |
637bbc75 DC |
862 | |
863 | XFS_STATS_INC(xs_write_calls); | |
864 | ||
637bbc75 DC |
865 | if (ocount == 0) |
866 | return 0; | |
867 | ||
bf97f3bc AV |
868 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
869 | return -EIO; | |
637bbc75 | 870 | |
6b698ede | 871 | if ((iocb->ki_flags & IOCB_DIRECT) || IS_DAX(inode)) |
bf97f3bc | 872 | ret = xfs_file_dio_aio_write(iocb, from); |
637bbc75 | 873 | else |
bf97f3bc | 874 | ret = xfs_file_buffered_aio_write(iocb, from); |
dda35b8f | 875 | |
d0606464 CH |
876 | if (ret > 0) { |
877 | ssize_t err; | |
dda35b8f | 878 | |
d0606464 | 879 | XFS_STATS_ADD(xs_write_bytes, ret); |
dda35b8f | 880 | |
d0606464 | 881 | /* Handle various SYNC-type writes */ |
d311d79d | 882 | err = generic_write_sync(file, iocb->ki_pos - ret, ret); |
d0606464 CH |
883 | if (err < 0) |
884 | ret = err; | |
dda35b8f | 885 | } |
a363f0c2 | 886 | return ret; |
dda35b8f CH |
887 | } |
888 | ||
a904b1ca NJ |
889 | #define XFS_FALLOC_FL_SUPPORTED \ |
890 | (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \ | |
891 | FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE | \ | |
892 | FALLOC_FL_INSERT_RANGE) | |
893 | ||
2fe17c10 CH |
894 | STATIC long |
895 | xfs_file_fallocate( | |
83aee9e4 CH |
896 | struct file *file, |
897 | int mode, | |
898 | loff_t offset, | |
899 | loff_t len) | |
2fe17c10 | 900 | { |
83aee9e4 CH |
901 | struct inode *inode = file_inode(file); |
902 | struct xfs_inode *ip = XFS_I(inode); | |
83aee9e4 | 903 | long error; |
8add71ca | 904 | enum xfs_prealloc_flags flags = 0; |
781355c6 | 905 | uint iolock = XFS_IOLOCK_EXCL; |
83aee9e4 | 906 | loff_t new_size = 0; |
a904b1ca | 907 | bool do_file_insert = 0; |
2fe17c10 | 908 | |
83aee9e4 CH |
909 | if (!S_ISREG(inode->i_mode)) |
910 | return -EINVAL; | |
a904b1ca | 911 | if (mode & ~XFS_FALLOC_FL_SUPPORTED) |
2fe17c10 CH |
912 | return -EOPNOTSUPP; |
913 | ||
781355c6 | 914 | xfs_ilock(ip, iolock); |
21c3ea18 | 915 | error = xfs_break_layouts(inode, &iolock, false); |
781355c6 CH |
916 | if (error) |
917 | goto out_unlock; | |
918 | ||
e8e9ad42 DC |
919 | xfs_ilock(ip, XFS_MMAPLOCK_EXCL); |
920 | iolock |= XFS_MMAPLOCK_EXCL; | |
921 | ||
83aee9e4 CH |
922 | if (mode & FALLOC_FL_PUNCH_HOLE) { |
923 | error = xfs_free_file_space(ip, offset, len); | |
924 | if (error) | |
925 | goto out_unlock; | |
e1d8fb88 NJ |
926 | } else if (mode & FALLOC_FL_COLLAPSE_RANGE) { |
927 | unsigned blksize_mask = (1 << inode->i_blkbits) - 1; | |
928 | ||
929 | if (offset & blksize_mask || len & blksize_mask) { | |
2451337d | 930 | error = -EINVAL; |
e1d8fb88 NJ |
931 | goto out_unlock; |
932 | } | |
933 | ||
23fffa92 LC |
934 | /* |
935 | * There is no need to overlap collapse range with EOF, | |
936 | * in which case it is effectively a truncate operation | |
937 | */ | |
938 | if (offset + len >= i_size_read(inode)) { | |
2451337d | 939 | error = -EINVAL; |
23fffa92 LC |
940 | goto out_unlock; |
941 | } | |
942 | ||
e1d8fb88 NJ |
943 | new_size = i_size_read(inode) - len; |
944 | ||
945 | error = xfs_collapse_file_space(ip, offset, len); | |
946 | if (error) | |
947 | goto out_unlock; | |
a904b1ca NJ |
948 | } else if (mode & FALLOC_FL_INSERT_RANGE) { |
949 | unsigned blksize_mask = (1 << inode->i_blkbits) - 1; | |
950 | ||
951 | new_size = i_size_read(inode) + len; | |
952 | if (offset & blksize_mask || len & blksize_mask) { | |
953 | error = -EINVAL; | |
954 | goto out_unlock; | |
955 | } | |
956 | ||
957 | /* check the new inode size does not wrap through zero */ | |
958 | if (new_size > inode->i_sb->s_maxbytes) { | |
959 | error = -EFBIG; | |
960 | goto out_unlock; | |
961 | } | |
962 | ||
963 | /* Offset should be less than i_size */ | |
964 | if (offset >= i_size_read(inode)) { | |
965 | error = -EINVAL; | |
966 | goto out_unlock; | |
967 | } | |
968 | do_file_insert = 1; | |
83aee9e4 | 969 | } else { |
8add71ca CH |
970 | flags |= XFS_PREALLOC_SET; |
971 | ||
83aee9e4 CH |
972 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
973 | offset + len > i_size_read(inode)) { | |
974 | new_size = offset + len; | |
2451337d | 975 | error = inode_newsize_ok(inode, new_size); |
83aee9e4 CH |
976 | if (error) |
977 | goto out_unlock; | |
978 | } | |
2fe17c10 | 979 | |
376ba313 LC |
980 | if (mode & FALLOC_FL_ZERO_RANGE) |
981 | error = xfs_zero_file_space(ip, offset, len); | |
982 | else | |
983 | error = xfs_alloc_file_space(ip, offset, len, | |
984 | XFS_BMAPI_PREALLOC); | |
2fe17c10 CH |
985 | if (error) |
986 | goto out_unlock; | |
987 | } | |
988 | ||
83aee9e4 | 989 | if (file->f_flags & O_DSYNC) |
8add71ca CH |
990 | flags |= XFS_PREALLOC_SYNC; |
991 | ||
992 | error = xfs_update_prealloc_flags(ip, flags); | |
2fe17c10 CH |
993 | if (error) |
994 | goto out_unlock; | |
995 | ||
996 | /* Change file size if needed */ | |
997 | if (new_size) { | |
998 | struct iattr iattr; | |
999 | ||
1000 | iattr.ia_valid = ATTR_SIZE; | |
1001 | iattr.ia_size = new_size; | |
83aee9e4 | 1002 | error = xfs_setattr_size(ip, &iattr); |
a904b1ca NJ |
1003 | if (error) |
1004 | goto out_unlock; | |
2fe17c10 CH |
1005 | } |
1006 | ||
a904b1ca NJ |
1007 | /* |
1008 | * Perform hole insertion now that the file size has been | |
1009 | * updated so that if we crash during the operation we don't | |
1010 | * leave shifted extents past EOF and hence losing access to | |
1011 | * the data that is contained within them. | |
1012 | */ | |
1013 | if (do_file_insert) | |
1014 | error = xfs_insert_file_space(ip, offset, len); | |
1015 | ||
2fe17c10 | 1016 | out_unlock: |
781355c6 | 1017 | xfs_iunlock(ip, iolock); |
2451337d | 1018 | return error; |
2fe17c10 CH |
1019 | } |
1020 | ||
1021 | ||
1da177e4 | 1022 | STATIC int |
3562fd45 | 1023 | xfs_file_open( |
1da177e4 | 1024 | struct inode *inode, |
f999a5bf | 1025 | struct file *file) |
1da177e4 | 1026 | { |
f999a5bf | 1027 | if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS) |
1da177e4 | 1028 | return -EFBIG; |
f999a5bf CH |
1029 | if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb))) |
1030 | return -EIO; | |
1031 | return 0; | |
1032 | } | |
1033 | ||
1034 | STATIC int | |
1035 | xfs_dir_open( | |
1036 | struct inode *inode, | |
1037 | struct file *file) | |
1038 | { | |
1039 | struct xfs_inode *ip = XFS_I(inode); | |
1040 | int mode; | |
1041 | int error; | |
1042 | ||
1043 | error = xfs_file_open(inode, file); | |
1044 | if (error) | |
1045 | return error; | |
1046 | ||
1047 | /* | |
1048 | * If there are any blocks, read-ahead block 0 as we're almost | |
1049 | * certain to have the next operation be a read there. | |
1050 | */ | |
309ecac8 | 1051 | mode = xfs_ilock_data_map_shared(ip); |
f999a5bf | 1052 | if (ip->i_d.di_nextents > 0) |
9df2dd0b | 1053 | xfs_dir3_data_readahead(ip, 0, -1); |
f999a5bf CH |
1054 | xfs_iunlock(ip, mode); |
1055 | return 0; | |
1da177e4 LT |
1056 | } |
1057 | ||
1da177e4 | 1058 | STATIC int |
3562fd45 | 1059 | xfs_file_release( |
1da177e4 LT |
1060 | struct inode *inode, |
1061 | struct file *filp) | |
1062 | { | |
2451337d | 1063 | return xfs_release(XFS_I(inode)); |
1da177e4 LT |
1064 | } |
1065 | ||
1da177e4 | 1066 | STATIC int |
3562fd45 | 1067 | xfs_file_readdir( |
b8227554 AV |
1068 | struct file *file, |
1069 | struct dir_context *ctx) | |
1da177e4 | 1070 | { |
b8227554 | 1071 | struct inode *inode = file_inode(file); |
739bfb2a | 1072 | xfs_inode_t *ip = XFS_I(inode); |
051e7cd4 CH |
1073 | size_t bufsize; |
1074 | ||
1075 | /* | |
1076 | * The Linux API doesn't pass down the total size of the buffer | |
1077 | * we read into down to the filesystem. With the filldir concept | |
1078 | * it's not needed for correct information, but the XFS dir2 leaf | |
1079 | * code wants an estimate of the buffer size to calculate it's | |
1080 | * readahead window and size the buffers used for mapping to | |
1081 | * physical blocks. | |
1082 | * | |
1083 | * Try to give it an estimate that's good enough, maybe at some | |
1084 | * point we can change the ->readdir prototype to include the | |
a9cc799e | 1085 | * buffer size. For now we use the current glibc buffer size. |
051e7cd4 | 1086 | */ |
a9cc799e | 1087 | bufsize = (size_t)min_t(loff_t, 32768, ip->i_d.di_size); |
051e7cd4 | 1088 | |
8300475e | 1089 | return xfs_readdir(ip, ctx, bufsize); |
1da177e4 LT |
1090 | } |
1091 | ||
d126d43f JL |
1092 | /* |
1093 | * This type is designed to indicate the type of offset we would like | |
49c69591 | 1094 | * to search from page cache for xfs_seek_hole_data(). |
d126d43f JL |
1095 | */ |
1096 | enum { | |
1097 | HOLE_OFF = 0, | |
1098 | DATA_OFF, | |
1099 | }; | |
1100 | ||
1101 | /* | |
1102 | * Lookup the desired type of offset from the given page. | |
1103 | * | |
1104 | * On success, return true and the offset argument will point to the | |
1105 | * start of the region that was found. Otherwise this function will | |
1106 | * return false and keep the offset argument unchanged. | |
1107 | */ | |
1108 | STATIC bool | |
1109 | xfs_lookup_buffer_offset( | |
1110 | struct page *page, | |
1111 | loff_t *offset, | |
1112 | unsigned int type) | |
1113 | { | |
1114 | loff_t lastoff = page_offset(page); | |
1115 | bool found = false; | |
1116 | struct buffer_head *bh, *head; | |
1117 | ||
1118 | bh = head = page_buffers(page); | |
1119 | do { | |
1120 | /* | |
1121 | * Unwritten extents that have data in the page | |
1122 | * cache covering them can be identified by the | |
1123 | * BH_Unwritten state flag. Pages with multiple | |
1124 | * buffers might have a mix of holes, data and | |
1125 | * unwritten extents - any buffer with valid | |
1126 | * data in it should have BH_Uptodate flag set | |
1127 | * on it. | |
1128 | */ | |
1129 | if (buffer_unwritten(bh) || | |
1130 | buffer_uptodate(bh)) { | |
1131 | if (type == DATA_OFF) | |
1132 | found = true; | |
1133 | } else { | |
1134 | if (type == HOLE_OFF) | |
1135 | found = true; | |
1136 | } | |
1137 | ||
1138 | if (found) { | |
1139 | *offset = lastoff; | |
1140 | break; | |
1141 | } | |
1142 | lastoff += bh->b_size; | |
1143 | } while ((bh = bh->b_this_page) != head); | |
1144 | ||
1145 | return found; | |
1146 | } | |
1147 | ||
1148 | /* | |
1149 | * This routine is called to find out and return a data or hole offset | |
1150 | * from the page cache for unwritten extents according to the desired | |
49c69591 | 1151 | * type for xfs_seek_hole_data(). |
d126d43f JL |
1152 | * |
1153 | * The argument offset is used to tell where we start to search from the | |
1154 | * page cache. Map is used to figure out the end points of the range to | |
1155 | * lookup pages. | |
1156 | * | |
1157 | * Return true if the desired type of offset was found, and the argument | |
1158 | * offset is filled with that address. Otherwise, return false and keep | |
1159 | * offset unchanged. | |
1160 | */ | |
1161 | STATIC bool | |
1162 | xfs_find_get_desired_pgoff( | |
1163 | struct inode *inode, | |
1164 | struct xfs_bmbt_irec *map, | |
1165 | unsigned int type, | |
1166 | loff_t *offset) | |
1167 | { | |
1168 | struct xfs_inode *ip = XFS_I(inode); | |
1169 | struct xfs_mount *mp = ip->i_mount; | |
1170 | struct pagevec pvec; | |
1171 | pgoff_t index; | |
1172 | pgoff_t end; | |
1173 | loff_t endoff; | |
1174 | loff_t startoff = *offset; | |
1175 | loff_t lastoff = startoff; | |
1176 | bool found = false; | |
1177 | ||
1178 | pagevec_init(&pvec, 0); | |
1179 | ||
1180 | index = startoff >> PAGE_CACHE_SHIFT; | |
1181 | endoff = XFS_FSB_TO_B(mp, map->br_startoff + map->br_blockcount); | |
1182 | end = endoff >> PAGE_CACHE_SHIFT; | |
1183 | do { | |
1184 | int want; | |
1185 | unsigned nr_pages; | |
1186 | unsigned int i; | |
1187 | ||
1188 | want = min_t(pgoff_t, end - index, PAGEVEC_SIZE); | |
1189 | nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index, | |
1190 | want); | |
1191 | /* | |
1192 | * No page mapped into given range. If we are searching holes | |
1193 | * and if this is the first time we got into the loop, it means | |
1194 | * that the given offset is landed in a hole, return it. | |
1195 | * | |
1196 | * If we have already stepped through some block buffers to find | |
1197 | * holes but they all contains data. In this case, the last | |
1198 | * offset is already updated and pointed to the end of the last | |
1199 | * mapped page, if it does not reach the endpoint to search, | |
1200 | * that means there should be a hole between them. | |
1201 | */ | |
1202 | if (nr_pages == 0) { | |
1203 | /* Data search found nothing */ | |
1204 | if (type == DATA_OFF) | |
1205 | break; | |
1206 | ||
1207 | ASSERT(type == HOLE_OFF); | |
1208 | if (lastoff == startoff || lastoff < endoff) { | |
1209 | found = true; | |
1210 | *offset = lastoff; | |
1211 | } | |
1212 | break; | |
1213 | } | |
1214 | ||
1215 | /* | |
1216 | * At lease we found one page. If this is the first time we | |
1217 | * step into the loop, and if the first page index offset is | |
1218 | * greater than the given search offset, a hole was found. | |
1219 | */ | |
1220 | if (type == HOLE_OFF && lastoff == startoff && | |
1221 | lastoff < page_offset(pvec.pages[0])) { | |
1222 | found = true; | |
1223 | break; | |
1224 | } | |
1225 | ||
1226 | for (i = 0; i < nr_pages; i++) { | |
1227 | struct page *page = pvec.pages[i]; | |
1228 | loff_t b_offset; | |
1229 | ||
1230 | /* | |
1231 | * At this point, the page may be truncated or | |
1232 | * invalidated (changing page->mapping to NULL), | |
1233 | * or even swizzled back from swapper_space to tmpfs | |
1234 | * file mapping. However, page->index will not change | |
1235 | * because we have a reference on the page. | |
1236 | * | |
1237 | * Searching done if the page index is out of range. | |
1238 | * If the current offset is not reaches the end of | |
1239 | * the specified search range, there should be a hole | |
1240 | * between them. | |
1241 | */ | |
1242 | if (page->index > end) { | |
1243 | if (type == HOLE_OFF && lastoff < endoff) { | |
1244 | *offset = lastoff; | |
1245 | found = true; | |
1246 | } | |
1247 | goto out; | |
1248 | } | |
1249 | ||
1250 | lock_page(page); | |
1251 | /* | |
1252 | * Page truncated or invalidated(page->mapping == NULL). | |
1253 | * We can freely skip it and proceed to check the next | |
1254 | * page. | |
1255 | */ | |
1256 | if (unlikely(page->mapping != inode->i_mapping)) { | |
1257 | unlock_page(page); | |
1258 | continue; | |
1259 | } | |
1260 | ||
1261 | if (!page_has_buffers(page)) { | |
1262 | unlock_page(page); | |
1263 | continue; | |
1264 | } | |
1265 | ||
1266 | found = xfs_lookup_buffer_offset(page, &b_offset, type); | |
1267 | if (found) { | |
1268 | /* | |
1269 | * The found offset may be less than the start | |
1270 | * point to search if this is the first time to | |
1271 | * come here. | |
1272 | */ | |
1273 | *offset = max_t(loff_t, startoff, b_offset); | |
1274 | unlock_page(page); | |
1275 | goto out; | |
1276 | } | |
1277 | ||
1278 | /* | |
1279 | * We either searching data but nothing was found, or | |
1280 | * searching hole but found a data buffer. In either | |
1281 | * case, probably the next page contains the desired | |
1282 | * things, update the last offset to it so. | |
1283 | */ | |
1284 | lastoff = page_offset(page) + PAGE_SIZE; | |
1285 | unlock_page(page); | |
1286 | } | |
1287 | ||
1288 | /* | |
1289 | * The number of returned pages less than our desired, search | |
1290 | * done. In this case, nothing was found for searching data, | |
1291 | * but we found a hole behind the last offset. | |
1292 | */ | |
1293 | if (nr_pages < want) { | |
1294 | if (type == HOLE_OFF) { | |
1295 | *offset = lastoff; | |
1296 | found = true; | |
1297 | } | |
1298 | break; | |
1299 | } | |
1300 | ||
1301 | index = pvec.pages[i - 1]->index + 1; | |
1302 | pagevec_release(&pvec); | |
1303 | } while (index <= end); | |
1304 | ||
1305 | out: | |
1306 | pagevec_release(&pvec); | |
1307 | return found; | |
1308 | } | |
1309 | ||
3fe3e6b1 | 1310 | STATIC loff_t |
49c69591 | 1311 | xfs_seek_hole_data( |
3fe3e6b1 | 1312 | struct file *file, |
49c69591 ES |
1313 | loff_t start, |
1314 | int whence) | |
3fe3e6b1 JL |
1315 | { |
1316 | struct inode *inode = file->f_mapping->host; | |
1317 | struct xfs_inode *ip = XFS_I(inode); | |
1318 | struct xfs_mount *mp = ip->i_mount; | |
3fe3e6b1 JL |
1319 | loff_t uninitialized_var(offset); |
1320 | xfs_fsize_t isize; | |
1321 | xfs_fileoff_t fsbno; | |
1322 | xfs_filblks_t end; | |
1323 | uint lock; | |
1324 | int error; | |
1325 | ||
49c69591 ES |
1326 | if (XFS_FORCED_SHUTDOWN(mp)) |
1327 | return -EIO; | |
1328 | ||
309ecac8 | 1329 | lock = xfs_ilock_data_map_shared(ip); |
3fe3e6b1 JL |
1330 | |
1331 | isize = i_size_read(inode); | |
1332 | if (start >= isize) { | |
2451337d | 1333 | error = -ENXIO; |
3fe3e6b1 JL |
1334 | goto out_unlock; |
1335 | } | |
1336 | ||
3fe3e6b1 JL |
1337 | /* |
1338 | * Try to read extents from the first block indicated | |
1339 | * by fsbno to the end block of the file. | |
1340 | */ | |
52f1acc8 | 1341 | fsbno = XFS_B_TO_FSBT(mp, start); |
3fe3e6b1 | 1342 | end = XFS_B_TO_FSB(mp, isize); |
49c69591 | 1343 | |
52f1acc8 JL |
1344 | for (;;) { |
1345 | struct xfs_bmbt_irec map[2]; | |
1346 | int nmap = 2; | |
1347 | unsigned int i; | |
3fe3e6b1 | 1348 | |
52f1acc8 JL |
1349 | error = xfs_bmapi_read(ip, fsbno, end - fsbno, map, &nmap, |
1350 | XFS_BMAPI_ENTIRE); | |
1351 | if (error) | |
1352 | goto out_unlock; | |
3fe3e6b1 | 1353 | |
52f1acc8 JL |
1354 | /* No extents at given offset, must be beyond EOF */ |
1355 | if (nmap == 0) { | |
2451337d | 1356 | error = -ENXIO; |
52f1acc8 JL |
1357 | goto out_unlock; |
1358 | } | |
1359 | ||
1360 | for (i = 0; i < nmap; i++) { | |
1361 | offset = max_t(loff_t, start, | |
1362 | XFS_FSB_TO_B(mp, map[i].br_startoff)); | |
1363 | ||
49c69591 ES |
1364 | /* Landed in the hole we wanted? */ |
1365 | if (whence == SEEK_HOLE && | |
1366 | map[i].br_startblock == HOLESTARTBLOCK) | |
1367 | goto out; | |
1368 | ||
1369 | /* Landed in the data extent we wanted? */ | |
1370 | if (whence == SEEK_DATA && | |
1371 | (map[i].br_startblock == DELAYSTARTBLOCK || | |
1372 | (map[i].br_state == XFS_EXT_NORM && | |
1373 | !isnullstartblock(map[i].br_startblock)))) | |
52f1acc8 JL |
1374 | goto out; |
1375 | ||
1376 | /* | |
49c69591 ES |
1377 | * Landed in an unwritten extent, try to search |
1378 | * for hole or data from page cache. | |
52f1acc8 JL |
1379 | */ |
1380 | if (map[i].br_state == XFS_EXT_UNWRITTEN) { | |
1381 | if (xfs_find_get_desired_pgoff(inode, &map[i], | |
49c69591 ES |
1382 | whence == SEEK_HOLE ? HOLE_OFF : DATA_OFF, |
1383 | &offset)) | |
52f1acc8 JL |
1384 | goto out; |
1385 | } | |
1386 | } | |
1387 | ||
1388 | /* | |
49c69591 ES |
1389 | * We only received one extent out of the two requested. This |
1390 | * means we've hit EOF and didn't find what we are looking for. | |
52f1acc8 | 1391 | */ |
3fe3e6b1 | 1392 | if (nmap == 1) { |
49c69591 ES |
1393 | /* |
1394 | * If we were looking for a hole, set offset to | |
1395 | * the end of the file (i.e., there is an implicit | |
1396 | * hole at the end of any file). | |
1397 | */ | |
1398 | if (whence == SEEK_HOLE) { | |
1399 | offset = isize; | |
1400 | break; | |
1401 | } | |
1402 | /* | |
1403 | * If we were looking for data, it's nowhere to be found | |
1404 | */ | |
1405 | ASSERT(whence == SEEK_DATA); | |
2451337d | 1406 | error = -ENXIO; |
3fe3e6b1 JL |
1407 | goto out_unlock; |
1408 | } | |
1409 | ||
52f1acc8 JL |
1410 | ASSERT(i > 1); |
1411 | ||
1412 | /* | |
1413 | * Nothing was found, proceed to the next round of search | |
49c69591 | 1414 | * if the next reading offset is not at or beyond EOF. |
52f1acc8 JL |
1415 | */ |
1416 | fsbno = map[i - 1].br_startoff + map[i - 1].br_blockcount; | |
1417 | start = XFS_FSB_TO_B(mp, fsbno); | |
1418 | if (start >= isize) { | |
49c69591 ES |
1419 | if (whence == SEEK_HOLE) { |
1420 | offset = isize; | |
1421 | break; | |
1422 | } | |
1423 | ASSERT(whence == SEEK_DATA); | |
2451337d | 1424 | error = -ENXIO; |
52f1acc8 JL |
1425 | goto out_unlock; |
1426 | } | |
3fe3e6b1 JL |
1427 | } |
1428 | ||
b686d1f7 JL |
1429 | out: |
1430 | /* | |
49c69591 | 1431 | * If at this point we have found the hole we wanted, the returned |
b686d1f7 | 1432 | * offset may be bigger than the file size as it may be aligned to |
49c69591 | 1433 | * page boundary for unwritten extents. We need to deal with this |
b686d1f7 JL |
1434 | * situation in particular. |
1435 | */ | |
49c69591 ES |
1436 | if (whence == SEEK_HOLE) |
1437 | offset = min_t(loff_t, offset, isize); | |
46a1c2c7 | 1438 | offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes); |
3fe3e6b1 JL |
1439 | |
1440 | out_unlock: | |
01f4f327 | 1441 | xfs_iunlock(ip, lock); |
3fe3e6b1 JL |
1442 | |
1443 | if (error) | |
2451337d | 1444 | return error; |
3fe3e6b1 JL |
1445 | return offset; |
1446 | } | |
1447 | ||
1448 | STATIC loff_t | |
1449 | xfs_file_llseek( | |
1450 | struct file *file, | |
1451 | loff_t offset, | |
59f9c004 | 1452 | int whence) |
3fe3e6b1 | 1453 | { |
59f9c004 | 1454 | switch (whence) { |
3fe3e6b1 JL |
1455 | case SEEK_END: |
1456 | case SEEK_CUR: | |
1457 | case SEEK_SET: | |
59f9c004 | 1458 | return generic_file_llseek(file, offset, whence); |
3fe3e6b1 | 1459 | case SEEK_HOLE: |
49c69591 | 1460 | case SEEK_DATA: |
59f9c004 | 1461 | return xfs_seek_hole_data(file, offset, whence); |
3fe3e6b1 JL |
1462 | default: |
1463 | return -EINVAL; | |
1464 | } | |
1465 | } | |
1466 | ||
de0e8c20 DC |
1467 | /* |
1468 | * Locking for serialisation of IO during page faults. This results in a lock | |
1469 | * ordering of: | |
1470 | * | |
1471 | * mmap_sem (MM) | |
6b698ede DC |
1472 | * sb_start_pagefault(vfs, freeze) |
1473 | * i_mmap_lock (XFS - truncate serialisation) | |
1474 | * page_lock (MM) | |
1475 | * i_lock (XFS - extent map serialisation) | |
de0e8c20 | 1476 | */ |
de0e8c20 | 1477 | |
075a924d DC |
1478 | /* |
1479 | * mmap()d file has taken write protection fault and is being made writable. We | |
1480 | * can set the page state up correctly for a writable page, which means we can | |
1481 | * do correct delalloc accounting (ENOSPC checking!) and unwritten extent | |
1482 | * mapping. | |
de0e8c20 DC |
1483 | */ |
1484 | STATIC int | |
075a924d | 1485 | xfs_filemap_page_mkwrite( |
de0e8c20 DC |
1486 | struct vm_area_struct *vma, |
1487 | struct vm_fault *vmf) | |
1488 | { | |
6b698ede | 1489 | struct inode *inode = file_inode(vma->vm_file); |
ec56b1f1 | 1490 | int ret; |
de0e8c20 | 1491 | |
6b698ede | 1492 | trace_xfs_filemap_page_mkwrite(XFS_I(inode)); |
de0e8c20 | 1493 | |
6b698ede | 1494 | sb_start_pagefault(inode->i_sb); |
ec56b1f1 | 1495 | file_update_time(vma->vm_file); |
6b698ede | 1496 | xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED); |
de0e8c20 | 1497 | |
6b698ede DC |
1498 | if (IS_DAX(inode)) { |
1499 | ret = __dax_mkwrite(vma, vmf, xfs_get_blocks_direct, | |
1500 | xfs_end_io_dax_write); | |
1501 | } else { | |
1502 | ret = __block_page_mkwrite(vma, vmf, xfs_get_blocks); | |
1503 | ret = block_page_mkwrite_return(ret); | |
1504 | } | |
1505 | ||
1506 | xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED); | |
1507 | sb_end_pagefault(inode->i_sb); | |
1508 | ||
1509 | return ret; | |
de0e8c20 DC |
1510 | } |
1511 | ||
075a924d | 1512 | STATIC int |
6b698ede | 1513 | xfs_filemap_fault( |
075a924d DC |
1514 | struct vm_area_struct *vma, |
1515 | struct vm_fault *vmf) | |
1516 | { | |
b2442c5a | 1517 | struct inode *inode = file_inode(vma->vm_file); |
6b698ede | 1518 | int ret; |
ec56b1f1 | 1519 | |
b2442c5a | 1520 | trace_xfs_filemap_fault(XFS_I(inode)); |
075a924d | 1521 | |
6b698ede | 1522 | /* DAX can shortcut the normal fault path on write faults! */ |
b2442c5a | 1523 | if ((vmf->flags & FAULT_FLAG_WRITE) && IS_DAX(inode)) |
6b698ede | 1524 | return xfs_filemap_page_mkwrite(vma, vmf); |
075a924d | 1525 | |
b2442c5a DC |
1526 | xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED); |
1527 | if (IS_DAX(inode)) { | |
1528 | /* | |
1529 | * we do not want to trigger unwritten extent conversion on read | |
1530 | * faults - that is unnecessary overhead and would also require | |
1531 | * changes to xfs_get_blocks_direct() to map unwritten extent | |
1532 | * ioend for conversion on read-only mappings. | |
1533 | */ | |
1534 | ret = __dax_fault(vma, vmf, xfs_get_blocks_direct, NULL); | |
1535 | } else | |
1536 | ret = filemap_fault(vma, vmf); | |
1537 | xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED); | |
075a924d | 1538 | |
6b698ede DC |
1539 | return ret; |
1540 | } | |
1541 | ||
1542 | static const struct vm_operations_struct xfs_file_vm_ops = { | |
1543 | .fault = xfs_filemap_fault, | |
1544 | .map_pages = filemap_map_pages, | |
1545 | .page_mkwrite = xfs_filemap_page_mkwrite, | |
1546 | }; | |
1547 | ||
1548 | STATIC int | |
1549 | xfs_file_mmap( | |
1550 | struct file *filp, | |
1551 | struct vm_area_struct *vma) | |
1552 | { | |
1553 | file_accessed(filp); | |
1554 | vma->vm_ops = &xfs_file_vm_ops; | |
1555 | if (IS_DAX(file_inode(filp))) | |
1556 | vma->vm_flags |= VM_MIXEDMAP; | |
1557 | return 0; | |
075a924d DC |
1558 | } |
1559 | ||
4b6f5d20 | 1560 | const struct file_operations xfs_file_operations = { |
3fe3e6b1 | 1561 | .llseek = xfs_file_llseek, |
b4f5d2c6 | 1562 | .read_iter = xfs_file_read_iter, |
bf97f3bc | 1563 | .write_iter = xfs_file_write_iter, |
1b895840 | 1564 | .splice_read = xfs_file_splice_read, |
8d020765 | 1565 | .splice_write = iter_file_splice_write, |
3562fd45 | 1566 | .unlocked_ioctl = xfs_file_ioctl, |
1da177e4 | 1567 | #ifdef CONFIG_COMPAT |
3562fd45 | 1568 | .compat_ioctl = xfs_file_compat_ioctl, |
1da177e4 | 1569 | #endif |
3562fd45 NS |
1570 | .mmap = xfs_file_mmap, |
1571 | .open = xfs_file_open, | |
1572 | .release = xfs_file_release, | |
1573 | .fsync = xfs_file_fsync, | |
2fe17c10 | 1574 | .fallocate = xfs_file_fallocate, |
1da177e4 LT |
1575 | }; |
1576 | ||
4b6f5d20 | 1577 | const struct file_operations xfs_dir_file_operations = { |
f999a5bf | 1578 | .open = xfs_dir_open, |
1da177e4 | 1579 | .read = generic_read_dir, |
b8227554 | 1580 | .iterate = xfs_file_readdir, |
59af1584 | 1581 | .llseek = generic_file_llseek, |
3562fd45 | 1582 | .unlocked_ioctl = xfs_file_ioctl, |
d3870398 | 1583 | #ifdef CONFIG_COMPAT |
3562fd45 | 1584 | .compat_ioctl = xfs_file_compat_ioctl, |
d3870398 | 1585 | #endif |
1da2f2db | 1586 | .fsync = xfs_dir_fsync, |
1da177e4 | 1587 | }; |