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> |
1da177e4 | 44 | |
f0f37e2f | 45 | static const struct vm_operations_struct xfs_file_vm_ops; |
1da177e4 | 46 | |
487f84f3 DC |
47 | /* |
48 | * Locking primitives for read and write IO paths to ensure we consistently use | |
49 | * and order the inode->i_mutex, ip->i_lock and ip->i_iolock. | |
50 | */ | |
51 | static inline void | |
52 | xfs_rw_ilock( | |
53 | struct xfs_inode *ip, | |
54 | int type) | |
55 | { | |
56 | if (type & XFS_IOLOCK_EXCL) | |
57 | mutex_lock(&VFS_I(ip)->i_mutex); | |
58 | xfs_ilock(ip, type); | |
59 | } | |
60 | ||
61 | static inline void | |
62 | xfs_rw_iunlock( | |
63 | struct xfs_inode *ip, | |
64 | int type) | |
65 | { | |
66 | xfs_iunlock(ip, type); | |
67 | if (type & XFS_IOLOCK_EXCL) | |
68 | mutex_unlock(&VFS_I(ip)->i_mutex); | |
69 | } | |
70 | ||
71 | static inline void | |
72 | xfs_rw_ilock_demote( | |
73 | struct xfs_inode *ip, | |
74 | int type) | |
75 | { | |
76 | xfs_ilock_demote(ip, type); | |
77 | if (type & XFS_IOLOCK_EXCL) | |
78 | mutex_unlock(&VFS_I(ip)->i_mutex); | |
79 | } | |
80 | ||
dda35b8f CH |
81 | /* |
82 | * xfs_iozero | |
83 | * | |
84 | * xfs_iozero clears the specified range of buffer supplied, | |
85 | * and marks all the affected blocks as valid and modified. If | |
86 | * an affected block is not allocated, it will be allocated. If | |
87 | * an affected block is not completely overwritten, and is not | |
88 | * valid before the operation, it will be read from disk before | |
89 | * being partially zeroed. | |
90 | */ | |
ef9d8733 | 91 | int |
dda35b8f CH |
92 | xfs_iozero( |
93 | struct xfs_inode *ip, /* inode */ | |
94 | loff_t pos, /* offset in file */ | |
95 | size_t count) /* size of data to zero */ | |
96 | { | |
97 | struct page *page; | |
98 | struct address_space *mapping; | |
99 | int status; | |
100 | ||
101 | mapping = VFS_I(ip)->i_mapping; | |
102 | do { | |
103 | unsigned offset, bytes; | |
104 | void *fsdata; | |
105 | ||
106 | offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */ | |
107 | bytes = PAGE_CACHE_SIZE - offset; | |
108 | if (bytes > count) | |
109 | bytes = count; | |
110 | ||
111 | status = pagecache_write_begin(NULL, mapping, pos, bytes, | |
112 | AOP_FLAG_UNINTERRUPTIBLE, | |
113 | &page, &fsdata); | |
114 | if (status) | |
115 | break; | |
116 | ||
117 | zero_user(page, offset, bytes); | |
118 | ||
119 | status = pagecache_write_end(NULL, mapping, pos, bytes, bytes, | |
120 | page, fsdata); | |
121 | WARN_ON(status <= 0); /* can't return less than zero! */ | |
122 | pos += bytes; | |
123 | count -= bytes; | |
124 | status = 0; | |
125 | } while (count); | |
126 | ||
127 | return (-status); | |
128 | } | |
129 | ||
8add71ca CH |
130 | int |
131 | xfs_update_prealloc_flags( | |
132 | struct xfs_inode *ip, | |
133 | enum xfs_prealloc_flags flags) | |
134 | { | |
135 | struct xfs_trans *tp; | |
136 | int error; | |
137 | ||
138 | tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_WRITEID); | |
139 | error = xfs_trans_reserve(tp, &M_RES(ip->i_mount)->tr_writeid, 0, 0); | |
140 | if (error) { | |
141 | xfs_trans_cancel(tp, 0); | |
142 | return error; | |
143 | } | |
144 | ||
145 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
146 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
147 | ||
148 | if (!(flags & XFS_PREALLOC_INVISIBLE)) { | |
149 | ip->i_d.di_mode &= ~S_ISUID; | |
150 | if (ip->i_d.di_mode & S_IXGRP) | |
151 | ip->i_d.di_mode &= ~S_ISGID; | |
152 | xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
153 | } | |
154 | ||
155 | if (flags & XFS_PREALLOC_SET) | |
156 | ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC; | |
157 | if (flags & XFS_PREALLOC_CLEAR) | |
158 | ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC; | |
159 | ||
160 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
161 | if (flags & XFS_PREALLOC_SYNC) | |
162 | xfs_trans_set_sync(tp); | |
163 | return xfs_trans_commit(tp, 0); | |
164 | } | |
165 | ||
1da2f2db CH |
166 | /* |
167 | * Fsync operations on directories are much simpler than on regular files, | |
168 | * as there is no file data to flush, and thus also no need for explicit | |
169 | * cache flush operations, and there are no non-transaction metadata updates | |
170 | * on directories either. | |
171 | */ | |
172 | STATIC int | |
173 | xfs_dir_fsync( | |
174 | struct file *file, | |
175 | loff_t start, | |
176 | loff_t end, | |
177 | int datasync) | |
178 | { | |
179 | struct xfs_inode *ip = XFS_I(file->f_mapping->host); | |
180 | struct xfs_mount *mp = ip->i_mount; | |
181 | xfs_lsn_t lsn = 0; | |
182 | ||
183 | trace_xfs_dir_fsync(ip); | |
184 | ||
185 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
186 | if (xfs_ipincount(ip)) | |
187 | lsn = ip->i_itemp->ili_last_lsn; | |
188 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
189 | ||
190 | if (!lsn) | |
191 | return 0; | |
2451337d | 192 | return _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, NULL); |
1da2f2db CH |
193 | } |
194 | ||
fd3200be CH |
195 | STATIC int |
196 | xfs_file_fsync( | |
197 | struct file *file, | |
02c24a82 JB |
198 | loff_t start, |
199 | loff_t end, | |
fd3200be CH |
200 | int datasync) |
201 | { | |
7ea80859 CH |
202 | struct inode *inode = file->f_mapping->host; |
203 | struct xfs_inode *ip = XFS_I(inode); | |
a27a263b | 204 | struct xfs_mount *mp = ip->i_mount; |
fd3200be CH |
205 | int error = 0; |
206 | int log_flushed = 0; | |
b1037058 | 207 | xfs_lsn_t lsn = 0; |
fd3200be | 208 | |
cca28fb8 | 209 | trace_xfs_file_fsync(ip); |
fd3200be | 210 | |
02c24a82 JB |
211 | error = filemap_write_and_wait_range(inode->i_mapping, start, end); |
212 | if (error) | |
213 | return error; | |
214 | ||
a27a263b | 215 | if (XFS_FORCED_SHUTDOWN(mp)) |
b474c7ae | 216 | return -EIO; |
fd3200be CH |
217 | |
218 | xfs_iflags_clear(ip, XFS_ITRUNCATED); | |
219 | ||
a27a263b CH |
220 | if (mp->m_flags & XFS_MOUNT_BARRIER) { |
221 | /* | |
222 | * If we have an RT and/or log subvolume we need to make sure | |
223 | * to flush the write cache the device used for file data | |
224 | * first. This is to ensure newly written file data make | |
225 | * it to disk before logging the new inode size in case of | |
226 | * an extending write. | |
227 | */ | |
228 | if (XFS_IS_REALTIME_INODE(ip)) | |
229 | xfs_blkdev_issue_flush(mp->m_rtdev_targp); | |
230 | else if (mp->m_logdev_targp != mp->m_ddev_targp) | |
231 | xfs_blkdev_issue_flush(mp->m_ddev_targp); | |
232 | } | |
233 | ||
fd3200be | 234 | /* |
8a9c9980 CH |
235 | * All metadata updates are logged, which means that we just have |
236 | * to flush the log up to the latest LSN that touched the inode. | |
fd3200be CH |
237 | */ |
238 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
8f639dde CH |
239 | if (xfs_ipincount(ip)) { |
240 | if (!datasync || | |
241 | (ip->i_itemp->ili_fields & ~XFS_ILOG_TIMESTAMP)) | |
242 | lsn = ip->i_itemp->ili_last_lsn; | |
243 | } | |
8a9c9980 | 244 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
fd3200be | 245 | |
8a9c9980 | 246 | if (lsn) |
b1037058 CH |
247 | error = _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, &log_flushed); |
248 | ||
a27a263b CH |
249 | /* |
250 | * If we only have a single device, and the log force about was | |
251 | * a no-op we might have to flush the data device cache here. | |
252 | * This can only happen for fdatasync/O_DSYNC if we were overwriting | |
253 | * an already allocated file and thus do not have any metadata to | |
254 | * commit. | |
255 | */ | |
256 | if ((mp->m_flags & XFS_MOUNT_BARRIER) && | |
257 | mp->m_logdev_targp == mp->m_ddev_targp && | |
258 | !XFS_IS_REALTIME_INODE(ip) && | |
259 | !log_flushed) | |
260 | xfs_blkdev_issue_flush(mp->m_ddev_targp); | |
fd3200be | 261 | |
2451337d | 262 | return error; |
fd3200be CH |
263 | } |
264 | ||
00258e36 | 265 | STATIC ssize_t |
b4f5d2c6 | 266 | xfs_file_read_iter( |
dda35b8f | 267 | struct kiocb *iocb, |
b4f5d2c6 | 268 | struct iov_iter *to) |
dda35b8f CH |
269 | { |
270 | struct file *file = iocb->ki_filp; | |
271 | struct inode *inode = file->f_mapping->host; | |
00258e36 CH |
272 | struct xfs_inode *ip = XFS_I(inode); |
273 | struct xfs_mount *mp = ip->i_mount; | |
b4f5d2c6 | 274 | size_t size = iov_iter_count(to); |
dda35b8f | 275 | ssize_t ret = 0; |
00258e36 | 276 | int ioflags = 0; |
dda35b8f | 277 | xfs_fsize_t n; |
b4f5d2c6 | 278 | loff_t pos = iocb->ki_pos; |
dda35b8f | 279 | |
dda35b8f CH |
280 | XFS_STATS_INC(xs_read_calls); |
281 | ||
2ba48ce5 | 282 | if (unlikely(iocb->ki_flags & IOCB_DIRECT)) |
b92cc59f | 283 | ioflags |= XFS_IO_ISDIRECT; |
00258e36 | 284 | if (file->f_mode & FMODE_NOCMTIME) |
b92cc59f | 285 | ioflags |= XFS_IO_INVIS; |
00258e36 | 286 | |
b92cc59f | 287 | if (unlikely(ioflags & XFS_IO_ISDIRECT)) { |
dda35b8f CH |
288 | xfs_buftarg_t *target = |
289 | XFS_IS_REALTIME_INODE(ip) ? | |
290 | mp->m_rtdev_targp : mp->m_ddev_targp; | |
7c71ee78 ES |
291 | /* DIO must be aligned to device logical sector size */ |
292 | if ((pos | size) & target->bt_logical_sectormask) { | |
fb595814 | 293 | if (pos == i_size_read(inode)) |
00258e36 | 294 | return 0; |
b474c7ae | 295 | return -EINVAL; |
dda35b8f CH |
296 | } |
297 | } | |
298 | ||
fb595814 | 299 | n = mp->m_super->s_maxbytes - pos; |
00258e36 | 300 | if (n <= 0 || size == 0) |
dda35b8f CH |
301 | return 0; |
302 | ||
303 | if (n < size) | |
304 | size = n; | |
305 | ||
306 | if (XFS_FORCED_SHUTDOWN(mp)) | |
307 | return -EIO; | |
308 | ||
0c38a251 DC |
309 | /* |
310 | * Locking is a bit tricky here. If we take an exclusive lock | |
311 | * for direct IO, we effectively serialise all new concurrent | |
312 | * read IO to this file and block it behind IO that is currently in | |
313 | * progress because IO in progress holds the IO lock shared. We only | |
314 | * need to hold the lock exclusive to blow away the page cache, so | |
315 | * only take lock exclusively if the page cache needs invalidation. | |
316 | * This allows the normal direct IO case of no page cache pages to | |
317 | * proceeed concurrently without serialisation. | |
318 | */ | |
319 | xfs_rw_ilock(ip, XFS_IOLOCK_SHARED); | |
b92cc59f | 320 | if ((ioflags & XFS_IO_ISDIRECT) && inode->i_mapping->nrpages) { |
0c38a251 | 321 | xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED); |
487f84f3 DC |
322 | xfs_rw_ilock(ip, XFS_IOLOCK_EXCL); |
323 | ||
00258e36 | 324 | if (inode->i_mapping->nrpages) { |
8ff1e670 | 325 | ret = filemap_write_and_wait_range( |
fb595814 | 326 | VFS_I(ip)->i_mapping, |
7d4ea3ce | 327 | pos, pos + size - 1); |
487f84f3 DC |
328 | if (ret) { |
329 | xfs_rw_iunlock(ip, XFS_IOLOCK_EXCL); | |
330 | return ret; | |
331 | } | |
85e584da CM |
332 | |
333 | /* | |
334 | * Invalidate whole pages. This can return an error if | |
335 | * we fail to invalidate a page, but this should never | |
336 | * happen on XFS. Warn if it does fail. | |
337 | */ | |
338 | ret = invalidate_inode_pages2_range(VFS_I(ip)->i_mapping, | |
7d4ea3ce DC |
339 | pos >> PAGE_CACHE_SHIFT, |
340 | (pos + size - 1) >> PAGE_CACHE_SHIFT); | |
85e584da CM |
341 | WARN_ON_ONCE(ret); |
342 | ret = 0; | |
00258e36 | 343 | } |
487f84f3 | 344 | xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL); |
0c38a251 | 345 | } |
dda35b8f | 346 | |
fb595814 | 347 | trace_xfs_file_read(ip, size, pos, ioflags); |
dda35b8f | 348 | |
b4f5d2c6 | 349 | ret = generic_file_read_iter(iocb, to); |
dda35b8f CH |
350 | if (ret > 0) |
351 | XFS_STATS_ADD(xs_read_bytes, ret); | |
352 | ||
487f84f3 | 353 | xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED); |
dda35b8f CH |
354 | return ret; |
355 | } | |
356 | ||
00258e36 CH |
357 | STATIC ssize_t |
358 | xfs_file_splice_read( | |
dda35b8f CH |
359 | struct file *infilp, |
360 | loff_t *ppos, | |
361 | struct pipe_inode_info *pipe, | |
362 | size_t count, | |
00258e36 | 363 | unsigned int flags) |
dda35b8f | 364 | { |
00258e36 | 365 | struct xfs_inode *ip = XFS_I(infilp->f_mapping->host); |
00258e36 | 366 | int ioflags = 0; |
dda35b8f CH |
367 | ssize_t ret; |
368 | ||
369 | XFS_STATS_INC(xs_read_calls); | |
00258e36 CH |
370 | |
371 | if (infilp->f_mode & FMODE_NOCMTIME) | |
b92cc59f | 372 | ioflags |= XFS_IO_INVIS; |
00258e36 | 373 | |
dda35b8f CH |
374 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
375 | return -EIO; | |
376 | ||
487f84f3 | 377 | xfs_rw_ilock(ip, XFS_IOLOCK_SHARED); |
dda35b8f | 378 | |
dda35b8f CH |
379 | trace_xfs_file_splice_read(ip, count, *ppos, ioflags); |
380 | ||
381 | ret = generic_file_splice_read(infilp, ppos, pipe, count, flags); | |
382 | if (ret > 0) | |
383 | XFS_STATS_ADD(xs_read_bytes, ret); | |
384 | ||
487f84f3 | 385 | xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED); |
dda35b8f CH |
386 | return ret; |
387 | } | |
388 | ||
dda35b8f | 389 | /* |
193aec10 CH |
390 | * This routine is called to handle zeroing any space in the last block of the |
391 | * file that is beyond the EOF. We do this since the size is being increased | |
392 | * without writing anything to that block and we don't want to read the | |
393 | * garbage on the disk. | |
dda35b8f CH |
394 | */ |
395 | STATIC int /* error (positive) */ | |
396 | xfs_zero_last_block( | |
193aec10 CH |
397 | struct xfs_inode *ip, |
398 | xfs_fsize_t offset, | |
5885ebda DC |
399 | xfs_fsize_t isize, |
400 | bool *did_zeroing) | |
dda35b8f | 401 | { |
193aec10 CH |
402 | struct xfs_mount *mp = ip->i_mount; |
403 | xfs_fileoff_t last_fsb = XFS_B_TO_FSBT(mp, isize); | |
404 | int zero_offset = XFS_B_FSB_OFFSET(mp, isize); | |
405 | int zero_len; | |
406 | int nimaps = 1; | |
407 | int error = 0; | |
408 | struct xfs_bmbt_irec imap; | |
dda35b8f | 409 | |
193aec10 | 410 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
5c8ed202 | 411 | error = xfs_bmapi_read(ip, last_fsb, 1, &imap, &nimaps, 0); |
193aec10 | 412 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
5c8ed202 | 413 | if (error) |
dda35b8f | 414 | return error; |
193aec10 | 415 | |
dda35b8f | 416 | ASSERT(nimaps > 0); |
193aec10 | 417 | |
dda35b8f CH |
418 | /* |
419 | * If the block underlying isize is just a hole, then there | |
420 | * is nothing to zero. | |
421 | */ | |
193aec10 | 422 | if (imap.br_startblock == HOLESTARTBLOCK) |
dda35b8f | 423 | return 0; |
dda35b8f CH |
424 | |
425 | zero_len = mp->m_sb.sb_blocksize - zero_offset; | |
426 | if (isize + zero_len > offset) | |
427 | zero_len = offset - isize; | |
5885ebda | 428 | *did_zeroing = true; |
193aec10 | 429 | return xfs_iozero(ip, isize, zero_len); |
dda35b8f CH |
430 | } |
431 | ||
432 | /* | |
193aec10 CH |
433 | * Zero any on disk space between the current EOF and the new, larger EOF. |
434 | * | |
435 | * This handles the normal case of zeroing the remainder of the last block in | |
436 | * the file and the unusual case of zeroing blocks out beyond the size of the | |
437 | * file. This second case only happens with fixed size extents and when the | |
438 | * system crashes before the inode size was updated but after blocks were | |
439 | * allocated. | |
440 | * | |
441 | * Expects the iolock to be held exclusive, and will take the ilock internally. | |
dda35b8f | 442 | */ |
dda35b8f CH |
443 | int /* error (positive) */ |
444 | xfs_zero_eof( | |
193aec10 CH |
445 | struct xfs_inode *ip, |
446 | xfs_off_t offset, /* starting I/O offset */ | |
5885ebda DC |
447 | xfs_fsize_t isize, /* current inode size */ |
448 | bool *did_zeroing) | |
dda35b8f | 449 | { |
193aec10 CH |
450 | struct xfs_mount *mp = ip->i_mount; |
451 | xfs_fileoff_t start_zero_fsb; | |
452 | xfs_fileoff_t end_zero_fsb; | |
453 | xfs_fileoff_t zero_count_fsb; | |
454 | xfs_fileoff_t last_fsb; | |
455 | xfs_fileoff_t zero_off; | |
456 | xfs_fsize_t zero_len; | |
457 | int nimaps; | |
458 | int error = 0; | |
459 | struct xfs_bmbt_irec imap; | |
460 | ||
461 | ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL)); | |
dda35b8f CH |
462 | ASSERT(offset > isize); |
463 | ||
464 | /* | |
465 | * First handle zeroing the block on which isize resides. | |
193aec10 | 466 | * |
dda35b8f CH |
467 | * We only zero a part of that block so it is handled specially. |
468 | */ | |
193aec10 | 469 | if (XFS_B_FSB_OFFSET(mp, isize) != 0) { |
5885ebda | 470 | error = xfs_zero_last_block(ip, offset, isize, did_zeroing); |
193aec10 CH |
471 | if (error) |
472 | return error; | |
dda35b8f CH |
473 | } |
474 | ||
475 | /* | |
193aec10 CH |
476 | * Calculate the range between the new size and the old where blocks |
477 | * needing to be zeroed may exist. | |
478 | * | |
479 | * To get the block where the last byte in the file currently resides, | |
480 | * we need to subtract one from the size and truncate back to a block | |
481 | * boundary. We subtract 1 in case the size is exactly on a block | |
482 | * boundary. | |
dda35b8f CH |
483 | */ |
484 | last_fsb = isize ? XFS_B_TO_FSBT(mp, isize - 1) : (xfs_fileoff_t)-1; | |
485 | start_zero_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize); | |
486 | end_zero_fsb = XFS_B_TO_FSBT(mp, offset - 1); | |
487 | ASSERT((xfs_sfiloff_t)last_fsb < (xfs_sfiloff_t)start_zero_fsb); | |
488 | if (last_fsb == end_zero_fsb) { | |
489 | /* | |
490 | * The size was only incremented on its last block. | |
491 | * We took care of that above, so just return. | |
492 | */ | |
493 | return 0; | |
494 | } | |
495 | ||
496 | ASSERT(start_zero_fsb <= end_zero_fsb); | |
497 | while (start_zero_fsb <= end_zero_fsb) { | |
498 | nimaps = 1; | |
499 | zero_count_fsb = end_zero_fsb - start_zero_fsb + 1; | |
193aec10 CH |
500 | |
501 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
5c8ed202 DC |
502 | error = xfs_bmapi_read(ip, start_zero_fsb, zero_count_fsb, |
503 | &imap, &nimaps, 0); | |
193aec10 CH |
504 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
505 | if (error) | |
dda35b8f | 506 | return error; |
193aec10 | 507 | |
dda35b8f CH |
508 | ASSERT(nimaps > 0); |
509 | ||
510 | if (imap.br_state == XFS_EXT_UNWRITTEN || | |
511 | imap.br_startblock == HOLESTARTBLOCK) { | |
dda35b8f CH |
512 | start_zero_fsb = imap.br_startoff + imap.br_blockcount; |
513 | ASSERT(start_zero_fsb <= (end_zero_fsb + 1)); | |
514 | continue; | |
515 | } | |
516 | ||
517 | /* | |
518 | * There are blocks we need to zero. | |
dda35b8f | 519 | */ |
dda35b8f CH |
520 | zero_off = XFS_FSB_TO_B(mp, start_zero_fsb); |
521 | zero_len = XFS_FSB_TO_B(mp, imap.br_blockcount); | |
522 | ||
523 | if ((zero_off + zero_len) > offset) | |
524 | zero_len = offset - zero_off; | |
525 | ||
526 | error = xfs_iozero(ip, zero_off, zero_len); | |
193aec10 CH |
527 | if (error) |
528 | return error; | |
dda35b8f | 529 | |
5885ebda | 530 | *did_zeroing = true; |
dda35b8f CH |
531 | start_zero_fsb = imap.br_startoff + imap.br_blockcount; |
532 | ASSERT(start_zero_fsb <= (end_zero_fsb + 1)); | |
dda35b8f CH |
533 | } |
534 | ||
535 | return 0; | |
dda35b8f CH |
536 | } |
537 | ||
4d8d1581 DC |
538 | /* |
539 | * Common pre-write limit and setup checks. | |
540 | * | |
5bf1f262 CH |
541 | * Called with the iolocked held either shared and exclusive according to |
542 | * @iolock, and returns with it held. Might upgrade the iolock to exclusive | |
543 | * if called for a direct write beyond i_size. | |
4d8d1581 DC |
544 | */ |
545 | STATIC ssize_t | |
546 | xfs_file_aio_write_checks( | |
99733fa3 AV |
547 | struct kiocb *iocb, |
548 | struct iov_iter *from, | |
4d8d1581 DC |
549 | int *iolock) |
550 | { | |
99733fa3 | 551 | struct file *file = iocb->ki_filp; |
4d8d1581 DC |
552 | struct inode *inode = file->f_mapping->host; |
553 | struct xfs_inode *ip = XFS_I(inode); | |
3309dd04 | 554 | ssize_t error = 0; |
99733fa3 | 555 | size_t count = iov_iter_count(from); |
4d8d1581 | 556 | |
7271d243 | 557 | restart: |
3309dd04 AV |
558 | error = generic_write_checks(iocb, from); |
559 | if (error <= 0) | |
4d8d1581 | 560 | return error; |
4d8d1581 | 561 | |
781355c6 CH |
562 | error = xfs_break_layouts(inode, iolock); |
563 | if (error) | |
564 | return error; | |
565 | ||
4d8d1581 DC |
566 | /* |
567 | * If the offset is beyond the size of the file, we need to zero any | |
568 | * blocks that fall between the existing EOF and the start of this | |
2813d682 | 569 | * write. If zeroing is needed and we are currently holding the |
467f7899 CH |
570 | * iolock shared, we need to update it to exclusive which implies |
571 | * having to redo all checks before. | |
4d8d1581 | 572 | */ |
99733fa3 | 573 | if (iocb->ki_pos > i_size_read(inode)) { |
5885ebda DC |
574 | bool zero = false; |
575 | ||
7271d243 | 576 | if (*iolock == XFS_IOLOCK_SHARED) { |
467f7899 | 577 | xfs_rw_iunlock(ip, *iolock); |
7271d243 | 578 | *iolock = XFS_IOLOCK_EXCL; |
467f7899 | 579 | xfs_rw_ilock(ip, *iolock); |
3309dd04 | 580 | iov_iter_reexpand(from, count); |
7271d243 DC |
581 | goto restart; |
582 | } | |
99733fa3 | 583 | error = xfs_zero_eof(ip, iocb->ki_pos, i_size_read(inode), &zero); |
467f7899 CH |
584 | if (error) |
585 | return error; | |
7271d243 | 586 | } |
4d8d1581 | 587 | |
8a9c9980 CH |
588 | /* |
589 | * Updating the timestamps will grab the ilock again from | |
590 | * xfs_fs_dirty_inode, so we have to call it after dropping the | |
591 | * lock above. Eventually we should look into a way to avoid | |
592 | * the pointless lock roundtrip. | |
593 | */ | |
c3b2da31 JB |
594 | if (likely(!(file->f_mode & FMODE_NOCMTIME))) { |
595 | error = file_update_time(file); | |
596 | if (error) | |
597 | return error; | |
598 | } | |
8a9c9980 | 599 | |
4d8d1581 DC |
600 | /* |
601 | * If we're writing the file then make sure to clear the setuid and | |
602 | * setgid bits if the process is not being run by root. This keeps | |
603 | * people from modifying setuid and setgid binaries. | |
604 | */ | |
605 | return file_remove_suid(file); | |
4d8d1581 DC |
606 | } |
607 | ||
f0d26e86 DC |
608 | /* |
609 | * xfs_file_dio_aio_write - handle direct IO writes | |
610 | * | |
611 | * Lock the inode appropriately to prepare for and issue a direct IO write. | |
eda77982 | 612 | * By separating it from the buffered write path we remove all the tricky to |
f0d26e86 DC |
613 | * follow locking changes and looping. |
614 | * | |
eda77982 DC |
615 | * If there are cached pages or we're extending the file, we need IOLOCK_EXCL |
616 | * until we're sure the bytes at the new EOF have been zeroed and/or the cached | |
617 | * pages are flushed out. | |
618 | * | |
619 | * In most cases the direct IO writes will be done holding IOLOCK_SHARED | |
620 | * allowing them to be done in parallel with reads and other direct IO writes. | |
621 | * However, if the IO is not aligned to filesystem blocks, the direct IO layer | |
622 | * needs to do sub-block zeroing and that requires serialisation against other | |
623 | * direct IOs to the same block. In this case we need to serialise the | |
624 | * submission of the unaligned IOs so that we don't get racing block zeroing in | |
625 | * the dio layer. To avoid the problem with aio, we also need to wait for | |
626 | * outstanding IOs to complete so that unwritten extent conversion is completed | |
627 | * before we try to map the overlapping block. This is currently implemented by | |
4a06fd26 | 628 | * hitting it with a big hammer (i.e. inode_dio_wait()). |
eda77982 | 629 | * |
f0d26e86 DC |
630 | * Returns with locks held indicated by @iolock and errors indicated by |
631 | * negative return values. | |
632 | */ | |
633 | STATIC ssize_t | |
634 | xfs_file_dio_aio_write( | |
635 | struct kiocb *iocb, | |
b3188919 | 636 | struct iov_iter *from) |
f0d26e86 DC |
637 | { |
638 | struct file *file = iocb->ki_filp; | |
639 | struct address_space *mapping = file->f_mapping; | |
640 | struct inode *inode = mapping->host; | |
641 | struct xfs_inode *ip = XFS_I(inode); | |
642 | struct xfs_mount *mp = ip->i_mount; | |
643 | ssize_t ret = 0; | |
eda77982 | 644 | int unaligned_io = 0; |
d0606464 | 645 | int iolock; |
b3188919 AV |
646 | size_t count = iov_iter_count(from); |
647 | loff_t pos = iocb->ki_pos; | |
f0d26e86 DC |
648 | struct xfs_buftarg *target = XFS_IS_REALTIME_INODE(ip) ? |
649 | mp->m_rtdev_targp : mp->m_ddev_targp; | |
650 | ||
7c71ee78 ES |
651 | /* DIO must be aligned to device logical sector size */ |
652 | if ((pos | count) & target->bt_logical_sectormask) | |
b474c7ae | 653 | return -EINVAL; |
f0d26e86 | 654 | |
7c71ee78 | 655 | /* "unaligned" here means not aligned to a filesystem block */ |
eda77982 DC |
656 | if ((pos & mp->m_blockmask) || ((pos + count) & mp->m_blockmask)) |
657 | unaligned_io = 1; | |
658 | ||
7271d243 DC |
659 | /* |
660 | * We don't need to take an exclusive lock unless there page cache needs | |
661 | * to be invalidated or unaligned IO is being executed. We don't need to | |
662 | * consider the EOF extension case here because | |
663 | * xfs_file_aio_write_checks() will relock the inode as necessary for | |
664 | * EOF zeroing cases and fill out the new inode size as appropriate. | |
665 | */ | |
666 | if (unaligned_io || mapping->nrpages) | |
d0606464 | 667 | iolock = XFS_IOLOCK_EXCL; |
f0d26e86 | 668 | else |
d0606464 CH |
669 | iolock = XFS_IOLOCK_SHARED; |
670 | xfs_rw_ilock(ip, iolock); | |
c58cb165 CH |
671 | |
672 | /* | |
673 | * Recheck if there are cached pages that need invalidate after we got | |
674 | * the iolock to protect against other threads adding new pages while | |
675 | * we were waiting for the iolock. | |
676 | */ | |
d0606464 CH |
677 | if (mapping->nrpages && iolock == XFS_IOLOCK_SHARED) { |
678 | xfs_rw_iunlock(ip, iolock); | |
679 | iolock = XFS_IOLOCK_EXCL; | |
680 | xfs_rw_ilock(ip, iolock); | |
c58cb165 | 681 | } |
f0d26e86 | 682 | |
99733fa3 | 683 | ret = xfs_file_aio_write_checks(iocb, from, &iolock); |
4d8d1581 | 684 | if (ret) |
d0606464 | 685 | goto out; |
99733fa3 AV |
686 | count = iov_iter_count(from); |
687 | pos = iocb->ki_pos; | |
f0d26e86 DC |
688 | |
689 | if (mapping->nrpages) { | |
07d5035a | 690 | ret = filemap_write_and_wait_range(VFS_I(ip)->i_mapping, |
7d4ea3ce | 691 | pos, pos + count - 1); |
f0d26e86 | 692 | if (ret) |
d0606464 | 693 | goto out; |
834ffca6 DC |
694 | /* |
695 | * Invalidate whole pages. This can return an error if | |
696 | * we fail to invalidate a page, but this should never | |
697 | * happen on XFS. Warn if it does fail. | |
698 | */ | |
699 | ret = invalidate_inode_pages2_range(VFS_I(ip)->i_mapping, | |
7d4ea3ce DC |
700 | pos >> PAGE_CACHE_SHIFT, |
701 | (pos + count - 1) >> PAGE_CACHE_SHIFT); | |
834ffca6 DC |
702 | WARN_ON_ONCE(ret); |
703 | ret = 0; | |
f0d26e86 DC |
704 | } |
705 | ||
eda77982 DC |
706 | /* |
707 | * If we are doing unaligned IO, wait for all other IO to drain, | |
708 | * otherwise demote the lock if we had to flush cached pages | |
709 | */ | |
710 | if (unaligned_io) | |
4a06fd26 | 711 | inode_dio_wait(inode); |
d0606464 | 712 | else if (iolock == XFS_IOLOCK_EXCL) { |
f0d26e86 | 713 | xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL); |
d0606464 | 714 | iolock = XFS_IOLOCK_SHARED; |
f0d26e86 DC |
715 | } |
716 | ||
717 | trace_xfs_file_direct_write(ip, count, iocb->ki_pos, 0); | |
b3188919 | 718 | ret = generic_file_direct_write(iocb, from, pos); |
f0d26e86 | 719 | |
d0606464 CH |
720 | out: |
721 | xfs_rw_iunlock(ip, iolock); | |
722 | ||
f0d26e86 DC |
723 | /* No fallback to buffered IO on errors for XFS. */ |
724 | ASSERT(ret < 0 || ret == count); | |
725 | return ret; | |
726 | } | |
727 | ||
00258e36 | 728 | STATIC ssize_t |
637bbc75 | 729 | xfs_file_buffered_aio_write( |
dda35b8f | 730 | struct kiocb *iocb, |
b3188919 | 731 | struct iov_iter *from) |
dda35b8f CH |
732 | { |
733 | struct file *file = iocb->ki_filp; | |
734 | struct address_space *mapping = file->f_mapping; | |
735 | struct inode *inode = mapping->host; | |
00258e36 | 736 | struct xfs_inode *ip = XFS_I(inode); |
637bbc75 DC |
737 | ssize_t ret; |
738 | int enospc = 0; | |
d0606464 | 739 | int iolock = XFS_IOLOCK_EXCL; |
dda35b8f | 740 | |
d0606464 | 741 | xfs_rw_ilock(ip, iolock); |
dda35b8f | 742 | |
99733fa3 | 743 | ret = xfs_file_aio_write_checks(iocb, from, &iolock); |
4d8d1581 | 744 | if (ret) |
d0606464 | 745 | goto out; |
dda35b8f CH |
746 | |
747 | /* We can write back this queue in page reclaim */ | |
de1414a6 | 748 | current->backing_dev_info = inode_to_bdi(inode); |
dda35b8f | 749 | |
dda35b8f | 750 | write_retry: |
99733fa3 AV |
751 | trace_xfs_file_buffered_write(ip, iov_iter_count(from), |
752 | iocb->ki_pos, 0); | |
753 | ret = generic_perform_write(file, from, iocb->ki_pos); | |
0a64bc2c | 754 | if (likely(ret >= 0)) |
99733fa3 | 755 | iocb->ki_pos += ret; |
dc06f398 | 756 | |
637bbc75 | 757 | /* |
dc06f398 BF |
758 | * If we hit a space limit, try to free up some lingering preallocated |
759 | * space before returning an error. In the case of ENOSPC, first try to | |
760 | * write back all dirty inodes to free up some of the excess reserved | |
761 | * metadata space. This reduces the chances that the eofblocks scan | |
762 | * waits on dirty mappings. Since xfs_flush_inodes() is serialized, this | |
763 | * also behaves as a filter to prevent too many eofblocks scans from | |
764 | * running at the same time. | |
637bbc75 | 765 | */ |
dc06f398 BF |
766 | if (ret == -EDQUOT && !enospc) { |
767 | enospc = xfs_inode_free_quota_eofblocks(ip); | |
768 | if (enospc) | |
769 | goto write_retry; | |
770 | } else if (ret == -ENOSPC && !enospc) { | |
771 | struct xfs_eofblocks eofb = {0}; | |
772 | ||
637bbc75 | 773 | enospc = 1; |
9aa05000 | 774 | xfs_flush_inodes(ip->i_mount); |
dc06f398 BF |
775 | eofb.eof_scan_owner = ip->i_ino; /* for locking */ |
776 | eofb.eof_flags = XFS_EOF_FLAGS_SYNC; | |
777 | xfs_icache_free_eofblocks(ip->i_mount, &eofb); | |
9aa05000 | 778 | goto write_retry; |
dda35b8f | 779 | } |
d0606464 | 780 | |
dda35b8f | 781 | current->backing_dev_info = NULL; |
d0606464 CH |
782 | out: |
783 | xfs_rw_iunlock(ip, iolock); | |
637bbc75 DC |
784 | return ret; |
785 | } | |
786 | ||
787 | STATIC ssize_t | |
bf97f3bc | 788 | xfs_file_write_iter( |
637bbc75 | 789 | struct kiocb *iocb, |
bf97f3bc | 790 | struct iov_iter *from) |
637bbc75 DC |
791 | { |
792 | struct file *file = iocb->ki_filp; | |
793 | struct address_space *mapping = file->f_mapping; | |
794 | struct inode *inode = mapping->host; | |
795 | struct xfs_inode *ip = XFS_I(inode); | |
796 | ssize_t ret; | |
bf97f3bc | 797 | size_t ocount = iov_iter_count(from); |
637bbc75 DC |
798 | |
799 | XFS_STATS_INC(xs_write_calls); | |
800 | ||
637bbc75 DC |
801 | if (ocount == 0) |
802 | return 0; | |
803 | ||
bf97f3bc AV |
804 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
805 | return -EIO; | |
637bbc75 | 806 | |
2ba48ce5 | 807 | if (unlikely(iocb->ki_flags & IOCB_DIRECT)) |
bf97f3bc | 808 | ret = xfs_file_dio_aio_write(iocb, from); |
637bbc75 | 809 | else |
bf97f3bc | 810 | ret = xfs_file_buffered_aio_write(iocb, from); |
dda35b8f | 811 | |
d0606464 CH |
812 | if (ret > 0) { |
813 | ssize_t err; | |
dda35b8f | 814 | |
d0606464 | 815 | XFS_STATS_ADD(xs_write_bytes, ret); |
dda35b8f | 816 | |
d0606464 | 817 | /* Handle various SYNC-type writes */ |
d311d79d | 818 | err = generic_write_sync(file, iocb->ki_pos - ret, ret); |
d0606464 CH |
819 | if (err < 0) |
820 | ret = err; | |
dda35b8f | 821 | } |
a363f0c2 | 822 | return ret; |
dda35b8f CH |
823 | } |
824 | ||
2fe17c10 CH |
825 | STATIC long |
826 | xfs_file_fallocate( | |
83aee9e4 CH |
827 | struct file *file, |
828 | int mode, | |
829 | loff_t offset, | |
830 | loff_t len) | |
2fe17c10 | 831 | { |
83aee9e4 CH |
832 | struct inode *inode = file_inode(file); |
833 | struct xfs_inode *ip = XFS_I(inode); | |
83aee9e4 | 834 | long error; |
8add71ca | 835 | enum xfs_prealloc_flags flags = 0; |
781355c6 | 836 | uint iolock = XFS_IOLOCK_EXCL; |
83aee9e4 | 837 | loff_t new_size = 0; |
2fe17c10 | 838 | |
83aee9e4 CH |
839 | if (!S_ISREG(inode->i_mode)) |
840 | return -EINVAL; | |
e1d8fb88 | 841 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | |
376ba313 | 842 | FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE)) |
2fe17c10 CH |
843 | return -EOPNOTSUPP; |
844 | ||
781355c6 CH |
845 | xfs_ilock(ip, iolock); |
846 | error = xfs_break_layouts(inode, &iolock); | |
847 | if (error) | |
848 | goto out_unlock; | |
849 | ||
83aee9e4 CH |
850 | if (mode & FALLOC_FL_PUNCH_HOLE) { |
851 | error = xfs_free_file_space(ip, offset, len); | |
852 | if (error) | |
853 | goto out_unlock; | |
e1d8fb88 NJ |
854 | } else if (mode & FALLOC_FL_COLLAPSE_RANGE) { |
855 | unsigned blksize_mask = (1 << inode->i_blkbits) - 1; | |
856 | ||
857 | if (offset & blksize_mask || len & blksize_mask) { | |
2451337d | 858 | error = -EINVAL; |
e1d8fb88 NJ |
859 | goto out_unlock; |
860 | } | |
861 | ||
23fffa92 LC |
862 | /* |
863 | * There is no need to overlap collapse range with EOF, | |
864 | * in which case it is effectively a truncate operation | |
865 | */ | |
866 | if (offset + len >= i_size_read(inode)) { | |
2451337d | 867 | error = -EINVAL; |
23fffa92 LC |
868 | goto out_unlock; |
869 | } | |
870 | ||
e1d8fb88 NJ |
871 | new_size = i_size_read(inode) - len; |
872 | ||
873 | error = xfs_collapse_file_space(ip, offset, len); | |
874 | if (error) | |
875 | goto out_unlock; | |
83aee9e4 | 876 | } else { |
8add71ca CH |
877 | flags |= XFS_PREALLOC_SET; |
878 | ||
83aee9e4 CH |
879 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
880 | offset + len > i_size_read(inode)) { | |
881 | new_size = offset + len; | |
2451337d | 882 | error = inode_newsize_ok(inode, new_size); |
83aee9e4 CH |
883 | if (error) |
884 | goto out_unlock; | |
885 | } | |
2fe17c10 | 886 | |
376ba313 LC |
887 | if (mode & FALLOC_FL_ZERO_RANGE) |
888 | error = xfs_zero_file_space(ip, offset, len); | |
889 | else | |
890 | error = xfs_alloc_file_space(ip, offset, len, | |
891 | XFS_BMAPI_PREALLOC); | |
2fe17c10 CH |
892 | if (error) |
893 | goto out_unlock; | |
894 | } | |
895 | ||
83aee9e4 | 896 | if (file->f_flags & O_DSYNC) |
8add71ca CH |
897 | flags |= XFS_PREALLOC_SYNC; |
898 | ||
899 | error = xfs_update_prealloc_flags(ip, flags); | |
2fe17c10 CH |
900 | if (error) |
901 | goto out_unlock; | |
902 | ||
903 | /* Change file size if needed */ | |
904 | if (new_size) { | |
905 | struct iattr iattr; | |
906 | ||
907 | iattr.ia_valid = ATTR_SIZE; | |
908 | iattr.ia_size = new_size; | |
83aee9e4 | 909 | error = xfs_setattr_size(ip, &iattr); |
2fe17c10 CH |
910 | } |
911 | ||
912 | out_unlock: | |
781355c6 | 913 | xfs_iunlock(ip, iolock); |
2451337d | 914 | return error; |
2fe17c10 CH |
915 | } |
916 | ||
917 | ||
1da177e4 | 918 | STATIC int |
3562fd45 | 919 | xfs_file_open( |
1da177e4 | 920 | struct inode *inode, |
f999a5bf | 921 | struct file *file) |
1da177e4 | 922 | { |
f999a5bf | 923 | if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS) |
1da177e4 | 924 | return -EFBIG; |
f999a5bf CH |
925 | if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb))) |
926 | return -EIO; | |
927 | return 0; | |
928 | } | |
929 | ||
930 | STATIC int | |
931 | xfs_dir_open( | |
932 | struct inode *inode, | |
933 | struct file *file) | |
934 | { | |
935 | struct xfs_inode *ip = XFS_I(inode); | |
936 | int mode; | |
937 | int error; | |
938 | ||
939 | error = xfs_file_open(inode, file); | |
940 | if (error) | |
941 | return error; | |
942 | ||
943 | /* | |
944 | * If there are any blocks, read-ahead block 0 as we're almost | |
945 | * certain to have the next operation be a read there. | |
946 | */ | |
309ecac8 | 947 | mode = xfs_ilock_data_map_shared(ip); |
f999a5bf | 948 | if (ip->i_d.di_nextents > 0) |
9df2dd0b | 949 | xfs_dir3_data_readahead(ip, 0, -1); |
f999a5bf CH |
950 | xfs_iunlock(ip, mode); |
951 | return 0; | |
1da177e4 LT |
952 | } |
953 | ||
1da177e4 | 954 | STATIC int |
3562fd45 | 955 | xfs_file_release( |
1da177e4 LT |
956 | struct inode *inode, |
957 | struct file *filp) | |
958 | { | |
2451337d | 959 | return xfs_release(XFS_I(inode)); |
1da177e4 LT |
960 | } |
961 | ||
1da177e4 | 962 | STATIC int |
3562fd45 | 963 | xfs_file_readdir( |
b8227554 AV |
964 | struct file *file, |
965 | struct dir_context *ctx) | |
1da177e4 | 966 | { |
b8227554 | 967 | struct inode *inode = file_inode(file); |
739bfb2a | 968 | xfs_inode_t *ip = XFS_I(inode); |
051e7cd4 CH |
969 | size_t bufsize; |
970 | ||
971 | /* | |
972 | * The Linux API doesn't pass down the total size of the buffer | |
973 | * we read into down to the filesystem. With the filldir concept | |
974 | * it's not needed for correct information, but the XFS dir2 leaf | |
975 | * code wants an estimate of the buffer size to calculate it's | |
976 | * readahead window and size the buffers used for mapping to | |
977 | * physical blocks. | |
978 | * | |
979 | * Try to give it an estimate that's good enough, maybe at some | |
980 | * point we can change the ->readdir prototype to include the | |
a9cc799e | 981 | * buffer size. For now we use the current glibc buffer size. |
051e7cd4 | 982 | */ |
a9cc799e | 983 | bufsize = (size_t)min_t(loff_t, 32768, ip->i_d.di_size); |
051e7cd4 | 984 | |
8300475e | 985 | return xfs_readdir(ip, ctx, bufsize); |
1da177e4 LT |
986 | } |
987 | ||
1da177e4 | 988 | STATIC int |
3562fd45 | 989 | xfs_file_mmap( |
1da177e4 LT |
990 | struct file *filp, |
991 | struct vm_area_struct *vma) | |
992 | { | |
3562fd45 | 993 | vma->vm_ops = &xfs_file_vm_ops; |
6fac0cb4 | 994 | |
fbc1462b | 995 | file_accessed(filp); |
1da177e4 LT |
996 | return 0; |
997 | } | |
998 | ||
4f57dbc6 DC |
999 | /* |
1000 | * mmap()d file has taken write protection fault and is being made | |
1001 | * writable. We can set the page state up correctly for a writable | |
1002 | * page, which means we can do correct delalloc accounting (ENOSPC | |
1003 | * checking!) and unwritten extent mapping. | |
1004 | */ | |
1005 | STATIC int | |
1006 | xfs_vm_page_mkwrite( | |
1007 | struct vm_area_struct *vma, | |
c2ec175c | 1008 | struct vm_fault *vmf) |
4f57dbc6 | 1009 | { |
c2ec175c | 1010 | return block_page_mkwrite(vma, vmf, xfs_get_blocks); |
4f57dbc6 DC |
1011 | } |
1012 | ||
d126d43f JL |
1013 | /* |
1014 | * This type is designed to indicate the type of offset we would like | |
49c69591 | 1015 | * to search from page cache for xfs_seek_hole_data(). |
d126d43f JL |
1016 | */ |
1017 | enum { | |
1018 | HOLE_OFF = 0, | |
1019 | DATA_OFF, | |
1020 | }; | |
1021 | ||
1022 | /* | |
1023 | * Lookup the desired type of offset from the given page. | |
1024 | * | |
1025 | * On success, return true and the offset argument will point to the | |
1026 | * start of the region that was found. Otherwise this function will | |
1027 | * return false and keep the offset argument unchanged. | |
1028 | */ | |
1029 | STATIC bool | |
1030 | xfs_lookup_buffer_offset( | |
1031 | struct page *page, | |
1032 | loff_t *offset, | |
1033 | unsigned int type) | |
1034 | { | |
1035 | loff_t lastoff = page_offset(page); | |
1036 | bool found = false; | |
1037 | struct buffer_head *bh, *head; | |
1038 | ||
1039 | bh = head = page_buffers(page); | |
1040 | do { | |
1041 | /* | |
1042 | * Unwritten extents that have data in the page | |
1043 | * cache covering them can be identified by the | |
1044 | * BH_Unwritten state flag. Pages with multiple | |
1045 | * buffers might have a mix of holes, data and | |
1046 | * unwritten extents - any buffer with valid | |
1047 | * data in it should have BH_Uptodate flag set | |
1048 | * on it. | |
1049 | */ | |
1050 | if (buffer_unwritten(bh) || | |
1051 | buffer_uptodate(bh)) { | |
1052 | if (type == DATA_OFF) | |
1053 | found = true; | |
1054 | } else { | |
1055 | if (type == HOLE_OFF) | |
1056 | found = true; | |
1057 | } | |
1058 | ||
1059 | if (found) { | |
1060 | *offset = lastoff; | |
1061 | break; | |
1062 | } | |
1063 | lastoff += bh->b_size; | |
1064 | } while ((bh = bh->b_this_page) != head); | |
1065 | ||
1066 | return found; | |
1067 | } | |
1068 | ||
1069 | /* | |
1070 | * This routine is called to find out and return a data or hole offset | |
1071 | * from the page cache for unwritten extents according to the desired | |
49c69591 | 1072 | * type for xfs_seek_hole_data(). |
d126d43f JL |
1073 | * |
1074 | * The argument offset is used to tell where we start to search from the | |
1075 | * page cache. Map is used to figure out the end points of the range to | |
1076 | * lookup pages. | |
1077 | * | |
1078 | * Return true if the desired type of offset was found, and the argument | |
1079 | * offset is filled with that address. Otherwise, return false and keep | |
1080 | * offset unchanged. | |
1081 | */ | |
1082 | STATIC bool | |
1083 | xfs_find_get_desired_pgoff( | |
1084 | struct inode *inode, | |
1085 | struct xfs_bmbt_irec *map, | |
1086 | unsigned int type, | |
1087 | loff_t *offset) | |
1088 | { | |
1089 | struct xfs_inode *ip = XFS_I(inode); | |
1090 | struct xfs_mount *mp = ip->i_mount; | |
1091 | struct pagevec pvec; | |
1092 | pgoff_t index; | |
1093 | pgoff_t end; | |
1094 | loff_t endoff; | |
1095 | loff_t startoff = *offset; | |
1096 | loff_t lastoff = startoff; | |
1097 | bool found = false; | |
1098 | ||
1099 | pagevec_init(&pvec, 0); | |
1100 | ||
1101 | index = startoff >> PAGE_CACHE_SHIFT; | |
1102 | endoff = XFS_FSB_TO_B(mp, map->br_startoff + map->br_blockcount); | |
1103 | end = endoff >> PAGE_CACHE_SHIFT; | |
1104 | do { | |
1105 | int want; | |
1106 | unsigned nr_pages; | |
1107 | unsigned int i; | |
1108 | ||
1109 | want = min_t(pgoff_t, end - index, PAGEVEC_SIZE); | |
1110 | nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index, | |
1111 | want); | |
1112 | /* | |
1113 | * No page mapped into given range. If we are searching holes | |
1114 | * and if this is the first time we got into the loop, it means | |
1115 | * that the given offset is landed in a hole, return it. | |
1116 | * | |
1117 | * If we have already stepped through some block buffers to find | |
1118 | * holes but they all contains data. In this case, the last | |
1119 | * offset is already updated and pointed to the end of the last | |
1120 | * mapped page, if it does not reach the endpoint to search, | |
1121 | * that means there should be a hole between them. | |
1122 | */ | |
1123 | if (nr_pages == 0) { | |
1124 | /* Data search found nothing */ | |
1125 | if (type == DATA_OFF) | |
1126 | break; | |
1127 | ||
1128 | ASSERT(type == HOLE_OFF); | |
1129 | if (lastoff == startoff || lastoff < endoff) { | |
1130 | found = true; | |
1131 | *offset = lastoff; | |
1132 | } | |
1133 | break; | |
1134 | } | |
1135 | ||
1136 | /* | |
1137 | * At lease we found one page. If this is the first time we | |
1138 | * step into the loop, and if the first page index offset is | |
1139 | * greater than the given search offset, a hole was found. | |
1140 | */ | |
1141 | if (type == HOLE_OFF && lastoff == startoff && | |
1142 | lastoff < page_offset(pvec.pages[0])) { | |
1143 | found = true; | |
1144 | break; | |
1145 | } | |
1146 | ||
1147 | for (i = 0; i < nr_pages; i++) { | |
1148 | struct page *page = pvec.pages[i]; | |
1149 | loff_t b_offset; | |
1150 | ||
1151 | /* | |
1152 | * At this point, the page may be truncated or | |
1153 | * invalidated (changing page->mapping to NULL), | |
1154 | * or even swizzled back from swapper_space to tmpfs | |
1155 | * file mapping. However, page->index will not change | |
1156 | * because we have a reference on the page. | |
1157 | * | |
1158 | * Searching done if the page index is out of range. | |
1159 | * If the current offset is not reaches the end of | |
1160 | * the specified search range, there should be a hole | |
1161 | * between them. | |
1162 | */ | |
1163 | if (page->index > end) { | |
1164 | if (type == HOLE_OFF && lastoff < endoff) { | |
1165 | *offset = lastoff; | |
1166 | found = true; | |
1167 | } | |
1168 | goto out; | |
1169 | } | |
1170 | ||
1171 | lock_page(page); | |
1172 | /* | |
1173 | * Page truncated or invalidated(page->mapping == NULL). | |
1174 | * We can freely skip it and proceed to check the next | |
1175 | * page. | |
1176 | */ | |
1177 | if (unlikely(page->mapping != inode->i_mapping)) { | |
1178 | unlock_page(page); | |
1179 | continue; | |
1180 | } | |
1181 | ||
1182 | if (!page_has_buffers(page)) { | |
1183 | unlock_page(page); | |
1184 | continue; | |
1185 | } | |
1186 | ||
1187 | found = xfs_lookup_buffer_offset(page, &b_offset, type); | |
1188 | if (found) { | |
1189 | /* | |
1190 | * The found offset may be less than the start | |
1191 | * point to search if this is the first time to | |
1192 | * come here. | |
1193 | */ | |
1194 | *offset = max_t(loff_t, startoff, b_offset); | |
1195 | unlock_page(page); | |
1196 | goto out; | |
1197 | } | |
1198 | ||
1199 | /* | |
1200 | * We either searching data but nothing was found, or | |
1201 | * searching hole but found a data buffer. In either | |
1202 | * case, probably the next page contains the desired | |
1203 | * things, update the last offset to it so. | |
1204 | */ | |
1205 | lastoff = page_offset(page) + PAGE_SIZE; | |
1206 | unlock_page(page); | |
1207 | } | |
1208 | ||
1209 | /* | |
1210 | * The number of returned pages less than our desired, search | |
1211 | * done. In this case, nothing was found for searching data, | |
1212 | * but we found a hole behind the last offset. | |
1213 | */ | |
1214 | if (nr_pages < want) { | |
1215 | if (type == HOLE_OFF) { | |
1216 | *offset = lastoff; | |
1217 | found = true; | |
1218 | } | |
1219 | break; | |
1220 | } | |
1221 | ||
1222 | index = pvec.pages[i - 1]->index + 1; | |
1223 | pagevec_release(&pvec); | |
1224 | } while (index <= end); | |
1225 | ||
1226 | out: | |
1227 | pagevec_release(&pvec); | |
1228 | return found; | |
1229 | } | |
1230 | ||
3fe3e6b1 | 1231 | STATIC loff_t |
49c69591 | 1232 | xfs_seek_hole_data( |
3fe3e6b1 | 1233 | struct file *file, |
49c69591 ES |
1234 | loff_t start, |
1235 | int whence) | |
3fe3e6b1 JL |
1236 | { |
1237 | struct inode *inode = file->f_mapping->host; | |
1238 | struct xfs_inode *ip = XFS_I(inode); | |
1239 | struct xfs_mount *mp = ip->i_mount; | |
3fe3e6b1 JL |
1240 | loff_t uninitialized_var(offset); |
1241 | xfs_fsize_t isize; | |
1242 | xfs_fileoff_t fsbno; | |
1243 | xfs_filblks_t end; | |
1244 | uint lock; | |
1245 | int error; | |
1246 | ||
49c69591 ES |
1247 | if (XFS_FORCED_SHUTDOWN(mp)) |
1248 | return -EIO; | |
1249 | ||
309ecac8 | 1250 | lock = xfs_ilock_data_map_shared(ip); |
3fe3e6b1 JL |
1251 | |
1252 | isize = i_size_read(inode); | |
1253 | if (start >= isize) { | |
2451337d | 1254 | error = -ENXIO; |
3fe3e6b1 JL |
1255 | goto out_unlock; |
1256 | } | |
1257 | ||
3fe3e6b1 JL |
1258 | /* |
1259 | * Try to read extents from the first block indicated | |
1260 | * by fsbno to the end block of the file. | |
1261 | */ | |
52f1acc8 | 1262 | fsbno = XFS_B_TO_FSBT(mp, start); |
3fe3e6b1 | 1263 | end = XFS_B_TO_FSB(mp, isize); |
49c69591 | 1264 | |
52f1acc8 JL |
1265 | for (;;) { |
1266 | struct xfs_bmbt_irec map[2]; | |
1267 | int nmap = 2; | |
1268 | unsigned int i; | |
3fe3e6b1 | 1269 | |
52f1acc8 JL |
1270 | error = xfs_bmapi_read(ip, fsbno, end - fsbno, map, &nmap, |
1271 | XFS_BMAPI_ENTIRE); | |
1272 | if (error) | |
1273 | goto out_unlock; | |
3fe3e6b1 | 1274 | |
52f1acc8 JL |
1275 | /* No extents at given offset, must be beyond EOF */ |
1276 | if (nmap == 0) { | |
2451337d | 1277 | error = -ENXIO; |
52f1acc8 JL |
1278 | goto out_unlock; |
1279 | } | |
1280 | ||
1281 | for (i = 0; i < nmap; i++) { | |
1282 | offset = max_t(loff_t, start, | |
1283 | XFS_FSB_TO_B(mp, map[i].br_startoff)); | |
1284 | ||
49c69591 ES |
1285 | /* Landed in the hole we wanted? */ |
1286 | if (whence == SEEK_HOLE && | |
1287 | map[i].br_startblock == HOLESTARTBLOCK) | |
1288 | goto out; | |
1289 | ||
1290 | /* Landed in the data extent we wanted? */ | |
1291 | if (whence == SEEK_DATA && | |
1292 | (map[i].br_startblock == DELAYSTARTBLOCK || | |
1293 | (map[i].br_state == XFS_EXT_NORM && | |
1294 | !isnullstartblock(map[i].br_startblock)))) | |
52f1acc8 JL |
1295 | goto out; |
1296 | ||
1297 | /* | |
49c69591 ES |
1298 | * Landed in an unwritten extent, try to search |
1299 | * for hole or data from page cache. | |
52f1acc8 JL |
1300 | */ |
1301 | if (map[i].br_state == XFS_EXT_UNWRITTEN) { | |
1302 | if (xfs_find_get_desired_pgoff(inode, &map[i], | |
49c69591 ES |
1303 | whence == SEEK_HOLE ? HOLE_OFF : DATA_OFF, |
1304 | &offset)) | |
52f1acc8 JL |
1305 | goto out; |
1306 | } | |
1307 | } | |
1308 | ||
1309 | /* | |
49c69591 ES |
1310 | * We only received one extent out of the two requested. This |
1311 | * means we've hit EOF and didn't find what we are looking for. | |
52f1acc8 | 1312 | */ |
3fe3e6b1 | 1313 | if (nmap == 1) { |
49c69591 ES |
1314 | /* |
1315 | * If we were looking for a hole, set offset to | |
1316 | * the end of the file (i.e., there is an implicit | |
1317 | * hole at the end of any file). | |
1318 | */ | |
1319 | if (whence == SEEK_HOLE) { | |
1320 | offset = isize; | |
1321 | break; | |
1322 | } | |
1323 | /* | |
1324 | * If we were looking for data, it's nowhere to be found | |
1325 | */ | |
1326 | ASSERT(whence == SEEK_DATA); | |
2451337d | 1327 | error = -ENXIO; |
3fe3e6b1 JL |
1328 | goto out_unlock; |
1329 | } | |
1330 | ||
52f1acc8 JL |
1331 | ASSERT(i > 1); |
1332 | ||
1333 | /* | |
1334 | * Nothing was found, proceed to the next round of search | |
49c69591 | 1335 | * if the next reading offset is not at or beyond EOF. |
52f1acc8 JL |
1336 | */ |
1337 | fsbno = map[i - 1].br_startoff + map[i - 1].br_blockcount; | |
1338 | start = XFS_FSB_TO_B(mp, fsbno); | |
1339 | if (start >= isize) { | |
49c69591 ES |
1340 | if (whence == SEEK_HOLE) { |
1341 | offset = isize; | |
1342 | break; | |
1343 | } | |
1344 | ASSERT(whence == SEEK_DATA); | |
2451337d | 1345 | error = -ENXIO; |
52f1acc8 JL |
1346 | goto out_unlock; |
1347 | } | |
3fe3e6b1 JL |
1348 | } |
1349 | ||
b686d1f7 JL |
1350 | out: |
1351 | /* | |
49c69591 | 1352 | * If at this point we have found the hole we wanted, the returned |
b686d1f7 | 1353 | * offset may be bigger than the file size as it may be aligned to |
49c69591 | 1354 | * page boundary for unwritten extents. We need to deal with this |
b686d1f7 JL |
1355 | * situation in particular. |
1356 | */ | |
49c69591 ES |
1357 | if (whence == SEEK_HOLE) |
1358 | offset = min_t(loff_t, offset, isize); | |
46a1c2c7 | 1359 | offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes); |
3fe3e6b1 JL |
1360 | |
1361 | out_unlock: | |
01f4f327 | 1362 | xfs_iunlock(ip, lock); |
3fe3e6b1 JL |
1363 | |
1364 | if (error) | |
2451337d | 1365 | return error; |
3fe3e6b1 JL |
1366 | return offset; |
1367 | } | |
1368 | ||
1369 | STATIC loff_t | |
1370 | xfs_file_llseek( | |
1371 | struct file *file, | |
1372 | loff_t offset, | |
59f9c004 | 1373 | int whence) |
3fe3e6b1 | 1374 | { |
59f9c004 | 1375 | switch (whence) { |
3fe3e6b1 JL |
1376 | case SEEK_END: |
1377 | case SEEK_CUR: | |
1378 | case SEEK_SET: | |
59f9c004 | 1379 | return generic_file_llseek(file, offset, whence); |
3fe3e6b1 | 1380 | case SEEK_HOLE: |
49c69591 | 1381 | case SEEK_DATA: |
59f9c004 | 1382 | return xfs_seek_hole_data(file, offset, whence); |
3fe3e6b1 JL |
1383 | default: |
1384 | return -EINVAL; | |
1385 | } | |
1386 | } | |
1387 | ||
4b6f5d20 | 1388 | const struct file_operations xfs_file_operations = { |
3fe3e6b1 | 1389 | .llseek = xfs_file_llseek, |
b4f5d2c6 | 1390 | .read_iter = xfs_file_read_iter, |
bf97f3bc | 1391 | .write_iter = xfs_file_write_iter, |
1b895840 | 1392 | .splice_read = xfs_file_splice_read, |
8d020765 | 1393 | .splice_write = iter_file_splice_write, |
3562fd45 | 1394 | .unlocked_ioctl = xfs_file_ioctl, |
1da177e4 | 1395 | #ifdef CONFIG_COMPAT |
3562fd45 | 1396 | .compat_ioctl = xfs_file_compat_ioctl, |
1da177e4 | 1397 | #endif |
3562fd45 NS |
1398 | .mmap = xfs_file_mmap, |
1399 | .open = xfs_file_open, | |
1400 | .release = xfs_file_release, | |
1401 | .fsync = xfs_file_fsync, | |
2fe17c10 | 1402 | .fallocate = xfs_file_fallocate, |
1da177e4 LT |
1403 | }; |
1404 | ||
4b6f5d20 | 1405 | const struct file_operations xfs_dir_file_operations = { |
f999a5bf | 1406 | .open = xfs_dir_open, |
1da177e4 | 1407 | .read = generic_read_dir, |
b8227554 | 1408 | .iterate = xfs_file_readdir, |
59af1584 | 1409 | .llseek = generic_file_llseek, |
3562fd45 | 1410 | .unlocked_ioctl = xfs_file_ioctl, |
d3870398 | 1411 | #ifdef CONFIG_COMPAT |
3562fd45 | 1412 | .compat_ioctl = xfs_file_compat_ioctl, |
d3870398 | 1413 | #endif |
1da2f2db | 1414 | .fsync = xfs_dir_fsync, |
1da177e4 LT |
1415 | }; |
1416 | ||
f0f37e2f | 1417 | static const struct vm_operations_struct xfs_file_vm_ops = { |
54cb8821 | 1418 | .fault = filemap_fault, |
f1820361 | 1419 | .map_pages = filemap_map_pages, |
4f57dbc6 | 1420 | .page_mkwrite = xfs_vm_page_mkwrite, |
6fac0cb4 | 1421 | }; |