Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
7b718769 NS |
2 | * Copyright (c) 2000-2003,2005 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" |
1da177e4 | 19 | #include "xfs_fs.h" |
a844f451 | 20 | #include "xfs_bit.h" |
1da177e4 | 21 | #include "xfs_log.h" |
a844f451 | 22 | #include "xfs_inum.h" |
1da177e4 LT |
23 | #include "xfs_trans.h" |
24 | #include "xfs_sb.h" | |
25 | #include "xfs_ag.h" | |
1da177e4 LT |
26 | #include "xfs_dir2.h" |
27 | #include "xfs_alloc.h" | |
28 | #include "xfs_dmapi.h" | |
29 | #include "xfs_quota.h" | |
30 | #include "xfs_mount.h" | |
1da177e4 | 31 | #include "xfs_bmap_btree.h" |
a844f451 | 32 | #include "xfs_alloc_btree.h" |
1da177e4 | 33 | #include "xfs_ialloc_btree.h" |
1da177e4 | 34 | #include "xfs_dir2_sf.h" |
a844f451 | 35 | #include "xfs_attr_sf.h" |
1da177e4 LT |
36 | #include "xfs_dinode.h" |
37 | #include "xfs_inode.h" | |
38 | #include "xfs_bmap.h" | |
a844f451 NS |
39 | #include "xfs_btree.h" |
40 | #include "xfs_ialloc.h" | |
1da177e4 LT |
41 | #include "xfs_rtalloc.h" |
42 | #include "xfs_error.h" | |
43 | #include "xfs_itable.h" | |
44 | #include "xfs_rw.h" | |
1da177e4 LT |
45 | #include "xfs_attr.h" |
46 | #include "xfs_inode_item.h" | |
47 | #include "xfs_buf_item.h" | |
48 | #include "xfs_utils.h" | |
49 | #include "xfs_iomap.h" | |
993386c1 | 50 | #include "xfs_vnodeops.h" |
1da177e4 LT |
51 | |
52 | #include <linux/capability.h> | |
53 | #include <linux/writeback.h> | |
54 | ||
55 | ||
56 | #if defined(XFS_RW_TRACE) | |
57 | void | |
58 | xfs_rw_enter_trace( | |
59 | int tag, | |
613d7043 | 60 | xfs_inode_t *ip, |
1da177e4 LT |
61 | void *data, |
62 | size_t segs, | |
63 | loff_t offset, | |
64 | int ioflags) | |
65 | { | |
1da177e4 LT |
66 | if (ip->i_rwtrace == NULL) |
67 | return; | |
68 | ktrace_enter(ip->i_rwtrace, | |
69 | (void *)(unsigned long)tag, | |
70 | (void *)ip, | |
71 | (void *)((unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff)), | |
72 | (void *)((unsigned long)(ip->i_d.di_size & 0xffffffff)), | |
73 | (void *)data, | |
74 | (void *)((unsigned long)segs), | |
75 | (void *)((unsigned long)((offset >> 32) & 0xffffffff)), | |
76 | (void *)((unsigned long)(offset & 0xffffffff)), | |
77 | (void *)((unsigned long)ioflags), | |
613d7043 CH |
78 | (void *)((unsigned long)((ip->i_new_size >> 32) & 0xffffffff)), |
79 | (void *)((unsigned long)(ip->i_new_size & 0xffffffff)), | |
f1fdc848 | 80 | (void *)((unsigned long)current_pid()), |
1da177e4 LT |
81 | (void *)NULL, |
82 | (void *)NULL, | |
83 | (void *)NULL, | |
84 | (void *)NULL); | |
85 | } | |
86 | ||
87 | void | |
88 | xfs_inval_cached_trace( | |
613d7043 | 89 | xfs_inode_t *ip, |
1da177e4 LT |
90 | xfs_off_t offset, |
91 | xfs_off_t len, | |
92 | xfs_off_t first, | |
93 | xfs_off_t last) | |
94 | { | |
1da177e4 LT |
95 | |
96 | if (ip->i_rwtrace == NULL) | |
97 | return; | |
98 | ktrace_enter(ip->i_rwtrace, | |
99 | (void *)(__psint_t)XFS_INVAL_CACHED, | |
100 | (void *)ip, | |
101 | (void *)((unsigned long)((offset >> 32) & 0xffffffff)), | |
102 | (void *)((unsigned long)(offset & 0xffffffff)), | |
103 | (void *)((unsigned long)((len >> 32) & 0xffffffff)), | |
104 | (void *)((unsigned long)(len & 0xffffffff)), | |
105 | (void *)((unsigned long)((first >> 32) & 0xffffffff)), | |
106 | (void *)((unsigned long)(first & 0xffffffff)), | |
107 | (void *)((unsigned long)((last >> 32) & 0xffffffff)), | |
108 | (void *)((unsigned long)(last & 0xffffffff)), | |
f1fdc848 | 109 | (void *)((unsigned long)current_pid()), |
1da177e4 LT |
110 | (void *)NULL, |
111 | (void *)NULL, | |
112 | (void *)NULL, | |
113 | (void *)NULL, | |
114 | (void *)NULL); | |
115 | } | |
116 | #endif | |
117 | ||
118 | /* | |
119 | * xfs_iozero | |
120 | * | |
121 | * xfs_iozero clears the specified range of buffer supplied, | |
122 | * and marks all the affected blocks as valid and modified. If | |
123 | * an affected block is not allocated, it will be allocated. If | |
124 | * an affected block is not completely overwritten, and is not | |
125 | * valid before the operation, it will be read from disk before | |
126 | * being partially zeroed. | |
127 | */ | |
128 | STATIC int | |
129 | xfs_iozero( | |
541d7d3c | 130 | struct xfs_inode *ip, /* inode */ |
1da177e4 | 131 | loff_t pos, /* offset in file */ |
68160161 | 132 | size_t count) /* size of data to zero */ |
1da177e4 | 133 | { |
1da177e4 LT |
134 | struct page *page; |
135 | struct address_space *mapping; | |
1da177e4 LT |
136 | int status; |
137 | ||
01651646 | 138 | mapping = VFS_I(ip)->i_mapping; |
1da177e4 | 139 | do { |
d79689c7 NP |
140 | unsigned offset, bytes; |
141 | void *fsdata; | |
1da177e4 LT |
142 | |
143 | offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */ | |
1da177e4 LT |
144 | bytes = PAGE_CACHE_SIZE - offset; |
145 | if (bytes > count) | |
146 | bytes = count; | |
147 | ||
d79689c7 NP |
148 | status = pagecache_write_begin(NULL, mapping, pos, bytes, |
149 | AOP_FLAG_UNINTERRUPTIBLE, | |
150 | &page, &fsdata); | |
e7ff6aed | 151 | if (status) |
d79689c7 | 152 | break; |
1da177e4 | 153 | |
eebd2aa3 | 154 | zero_user(page, offset, bytes); |
e7ff6aed | 155 | |
d79689c7 NP |
156 | status = pagecache_write_end(NULL, mapping, pos, bytes, bytes, |
157 | page, fsdata); | |
158 | WARN_ON(status <= 0); /* can't return less than zero! */ | |
159 | pos += bytes; | |
160 | count -= bytes; | |
161 | status = 0; | |
1da177e4 LT |
162 | } while (count); |
163 | ||
164 | return (-status); | |
165 | } | |
166 | ||
1da177e4 LT |
167 | ssize_t /* bytes read, or (-) error */ |
168 | xfs_read( | |
993386c1 | 169 | xfs_inode_t *ip, |
1da177e4 LT |
170 | struct kiocb *iocb, |
171 | const struct iovec *iovp, | |
172 | unsigned int segs, | |
173 | loff_t *offset, | |
993386c1 | 174 | int ioflags) |
1da177e4 LT |
175 | { |
176 | struct file *file = iocb->ki_filp; | |
177 | struct inode *inode = file->f_mapping->host; | |
993386c1 | 178 | xfs_mount_t *mp = ip->i_mount; |
1da177e4 | 179 | size_t size = 0; |
d3cf2094 | 180 | ssize_t ret = 0; |
1da177e4 | 181 | xfs_fsize_t n; |
1da177e4 LT |
182 | unsigned long seg; |
183 | ||
1da177e4 LT |
184 | |
185 | XFS_STATS_INC(xs_read_calls); | |
186 | ||
187 | /* START copy & waste from filemap.c */ | |
188 | for (seg = 0; seg < segs; seg++) { | |
189 | const struct iovec *iv = &iovp[seg]; | |
190 | ||
191 | /* | |
192 | * If any segment has a negative length, or the cumulative | |
193 | * length ever wraps negative then return -EINVAL. | |
194 | */ | |
195 | size += iv->iov_len; | |
196 | if (unlikely((ssize_t)(size|iv->iov_len) < 0)) | |
197 | return XFS_ERROR(-EINVAL); | |
198 | } | |
199 | /* END copy & waste from filemap.c */ | |
200 | ||
201 | if (unlikely(ioflags & IO_ISDIRECT)) { | |
202 | xfs_buftarg_t *target = | |
71ddabb9 | 203 | XFS_IS_REALTIME_INODE(ip) ? |
1da177e4 | 204 | mp->m_rtdev_targp : mp->m_ddev_targp; |
ce8e922c NS |
205 | if ((*offset & target->bt_smask) || |
206 | (size & target->bt_smask)) { | |
ba87ea69 | 207 | if (*offset == ip->i_size) { |
1da177e4 LT |
208 | return (0); |
209 | } | |
210 | return -XFS_ERROR(EINVAL); | |
211 | } | |
212 | } | |
213 | ||
214 | n = XFS_MAXIOFFSET(mp) - *offset; | |
215 | if ((n <= 0) || (size == 0)) | |
216 | return 0; | |
217 | ||
218 | if (n < size) | |
219 | size = n; | |
220 | ||
a13828b1 | 221 | if (XFS_FORCED_SHUTDOWN(mp)) |
1da177e4 | 222 | return -EIO; |
1da177e4 LT |
223 | |
224 | if (unlikely(ioflags & IO_ISDIRECT)) | |
1b1dcc1b | 225 | mutex_lock(&inode->i_mutex); |
1da177e4 LT |
226 | xfs_ilock(ip, XFS_IOLOCK_SHARED); |
227 | ||
eb9df39d | 228 | if (DM_EVENT_ENABLED(ip, DM_EVENT_READ) && !(ioflags & IO_INVIS)) { |
e1a40fa9 | 229 | int dmflags = FILP_DELAY_FLAG(file) | DM_SEM_FLAG_RD(ioflags); |
126468b1 | 230 | int iolock = XFS_IOLOCK_SHARED; |
1da177e4 | 231 | |
bc4ac74a | 232 | ret = -XFS_SEND_DATA(mp, DM_EVENT_READ, ip, *offset, size, |
126468b1 | 233 | dmflags, &iolock); |
1da177e4 LT |
234 | if (ret) { |
235 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); | |
721259bc LM |
236 | if (unlikely(ioflags & IO_ISDIRECT)) |
237 | mutex_unlock(&inode->i_mutex); | |
238 | return ret; | |
1da177e4 LT |
239 | } |
240 | } | |
241 | ||
edcd4bce | 242 | if (unlikely(ioflags & IO_ISDIRECT)) { |
dcf49cc5 | 243 | if (inode->i_mapping->nrpages) |
2e656092 | 244 | ret = -xfs_flushinval_pages(ip, (*offset & PAGE_CACHE_MASK), |
e6a4b37f | 245 | -1, FI_REMAPF_LOCKED); |
721259bc | 246 | mutex_unlock(&inode->i_mutex); |
d3cf2094 LM |
247 | if (ret) { |
248 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); | |
249 | return ret; | |
250 | } | |
edcd4bce | 251 | } |
721259bc | 252 | |
613d7043 | 253 | xfs_rw_enter_trace(XFS_READ_ENTER, ip, |
1da177e4 | 254 | (void *)iovp, segs, *offset, ioflags); |
543ade1f BP |
255 | |
256 | iocb->ki_pos = *offset; | |
257 | ret = generic_file_aio_read(iocb, iovp, segs, *offset); | |
1da177e4 LT |
258 | if (ret == -EIOCBQUEUED && !(ioflags & IO_ISAIO)) |
259 | ret = wait_on_sync_kiocb(iocb); | |
260 | if (ret > 0) | |
261 | XFS_STATS_ADD(xs_read_bytes, ret); | |
262 | ||
263 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); | |
1da177e4 LT |
264 | return ret; |
265 | } | |
266 | ||
1b895840 NS |
267 | ssize_t |
268 | xfs_splice_read( | |
993386c1 | 269 | xfs_inode_t *ip, |
1b895840 | 270 | struct file *infilp, |
cbb7e577 | 271 | loff_t *ppos, |
3a326a2c | 272 | struct pipe_inode_info *pipe, |
1b895840 NS |
273 | size_t count, |
274 | int flags, | |
993386c1 | 275 | int ioflags) |
1b895840 | 276 | { |
1b895840 NS |
277 | xfs_mount_t *mp = ip->i_mount; |
278 | ssize_t ret; | |
1da177e4 | 279 | |
1b895840 NS |
280 | XFS_STATS_INC(xs_read_calls); |
281 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) | |
282 | return -EIO; | |
283 | ||
284 | xfs_ilock(ip, XFS_IOLOCK_SHARED); | |
285 | ||
eb9df39d | 286 | if (DM_EVENT_ENABLED(ip, DM_EVENT_READ) && !(ioflags & IO_INVIS)) { |
126468b1 | 287 | int iolock = XFS_IOLOCK_SHARED; |
1b895840 NS |
288 | int error; |
289 | ||
bc4ac74a | 290 | error = XFS_SEND_DATA(mp, DM_EVENT_READ, ip, *ppos, count, |
126468b1 | 291 | FILP_DELAY_FLAG(infilp), &iolock); |
1b895840 NS |
292 | if (error) { |
293 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); | |
294 | return -error; | |
295 | } | |
296 | } | |
613d7043 | 297 | xfs_rw_enter_trace(XFS_SPLICE_READ_ENTER, ip, |
cbb7e577 JA |
298 | pipe, count, *ppos, ioflags); |
299 | ret = generic_file_splice_read(infilp, ppos, pipe, count, flags); | |
1da177e4 LT |
300 | if (ret > 0) |
301 | XFS_STATS_ADD(xs_read_bytes, ret); | |
302 | ||
1b895840 NS |
303 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); |
304 | return ret; | |
305 | } | |
306 | ||
307 | ssize_t | |
308 | xfs_splice_write( | |
993386c1 | 309 | xfs_inode_t *ip, |
3a326a2c | 310 | struct pipe_inode_info *pipe, |
1b895840 | 311 | struct file *outfilp, |
cbb7e577 | 312 | loff_t *ppos, |
1b895840 NS |
313 | size_t count, |
314 | int flags, | |
993386c1 | 315 | int ioflags) |
1b895840 | 316 | { |
1b895840 NS |
317 | xfs_mount_t *mp = ip->i_mount; |
318 | ssize_t ret; | |
0a8d17d0 | 319 | struct inode *inode = outfilp->f_mapping->host; |
ba87ea69 | 320 | xfs_fsize_t isize, new_size; |
1b895840 NS |
321 | |
322 | XFS_STATS_INC(xs_write_calls); | |
323 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) | |
324 | return -EIO; | |
325 | ||
326 | xfs_ilock(ip, XFS_IOLOCK_EXCL); | |
327 | ||
eb9df39d | 328 | if (DM_EVENT_ENABLED(ip, DM_EVENT_WRITE) && !(ioflags & IO_INVIS)) { |
126468b1 | 329 | int iolock = XFS_IOLOCK_EXCL; |
1b895840 NS |
330 | int error; |
331 | ||
bc4ac74a | 332 | error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, ip, *ppos, count, |
126468b1 | 333 | FILP_DELAY_FLAG(outfilp), &iolock); |
1b895840 NS |
334 | if (error) { |
335 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
336 | return -error; | |
337 | } | |
338 | } | |
ba87ea69 LM |
339 | |
340 | new_size = *ppos + count; | |
341 | ||
342 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
343 | if (new_size > ip->i_size) | |
613d7043 | 344 | ip->i_new_size = new_size; |
ba87ea69 LM |
345 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
346 | ||
613d7043 | 347 | xfs_rw_enter_trace(XFS_SPLICE_WRITE_ENTER, ip, |
cbb7e577 JA |
348 | pipe, count, *ppos, ioflags); |
349 | ret = generic_file_splice_write(pipe, outfilp, ppos, count, flags); | |
1b895840 NS |
350 | if (ret > 0) |
351 | XFS_STATS_ADD(xs_write_bytes, ret); | |
352 | ||
0a8d17d0 DC |
353 | isize = i_size_read(inode); |
354 | if (unlikely(ret < 0 && ret != -EFAULT && *ppos > isize)) | |
355 | *ppos = isize; | |
356 | ||
ba87ea69 | 357 | if (*ppos > ip->i_size) { |
0a8d17d0 | 358 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
ba87ea69 LM |
359 | if (*ppos > ip->i_size) |
360 | ip->i_size = *ppos; | |
361 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
362 | } | |
363 | ||
613d7043 | 364 | if (ip->i_new_size) { |
ba87ea69 | 365 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
613d7043 | 366 | ip->i_new_size = 0; |
ba87ea69 LM |
367 | if (ip->i_d.di_size > ip->i_size) |
368 | ip->i_d.di_size = ip->i_size; | |
0a8d17d0 DC |
369 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
370 | } | |
1b895840 | 371 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); |
1da177e4 LT |
372 | return ret; |
373 | } | |
374 | ||
375 | /* | |
376 | * This routine is called to handle zeroing any space in the last | |
377 | * block of the file that is beyond the EOF. We do this since the | |
378 | * size is being increased without writing anything to that block | |
379 | * and we don't want anyone to read the garbage on the disk. | |
380 | */ | |
381 | STATIC int /* error (positive) */ | |
382 | xfs_zero_last_block( | |
541d7d3c | 383 | xfs_inode_t *ip, |
68160161 LM |
384 | xfs_fsize_t offset, |
385 | xfs_fsize_t isize) | |
1da177e4 LT |
386 | { |
387 | xfs_fileoff_t last_fsb; | |
541d7d3c | 388 | xfs_mount_t *mp = ip->i_mount; |
1da177e4 LT |
389 | int nimaps; |
390 | int zero_offset; | |
391 | int zero_len; | |
1da177e4 LT |
392 | int error = 0; |
393 | xfs_bmbt_irec_t imap; | |
1da177e4 | 394 | |
579aa9ca | 395 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
1da177e4 | 396 | |
24ee8088 ES |
397 | zero_offset = XFS_B_FSB_OFFSET(mp, isize); |
398 | if (zero_offset == 0) { | |
1da177e4 LT |
399 | /* |
400 | * There are no extra bytes in the last block on disk to | |
401 | * zero, so return. | |
402 | */ | |
403 | return 0; | |
404 | } | |
405 | ||
406 | last_fsb = XFS_B_TO_FSBT(mp, isize); | |
407 | nimaps = 1; | |
541d7d3c | 408 | error = xfs_bmapi(NULL, ip, last_fsb, 1, 0, NULL, 0, &imap, |
3e57ecf6 | 409 | &nimaps, NULL, NULL); |
1da177e4 LT |
410 | if (error) { |
411 | return error; | |
412 | } | |
413 | ASSERT(nimaps > 0); | |
414 | /* | |
415 | * If the block underlying isize is just a hole, then there | |
416 | * is nothing to zero. | |
417 | */ | |
418 | if (imap.br_startblock == HOLESTARTBLOCK) { | |
419 | return 0; | |
420 | } | |
421 | /* | |
422 | * Zero the part of the last block beyond the EOF, and write it | |
423 | * out sync. We need to drop the ilock while we do this so we | |
424 | * don't deadlock when the buffer cache calls back to us. | |
425 | */ | |
579aa9ca | 426 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
1da177e4 | 427 | |
24ee8088 | 428 | zero_len = mp->m_sb.sb_blocksize - zero_offset; |
68160161 LM |
429 | if (isize + zero_len > offset) |
430 | zero_len = offset - isize; | |
431 | error = xfs_iozero(ip, isize, zero_len); | |
1da177e4 | 432 | |
579aa9ca | 433 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
1da177e4 LT |
434 | ASSERT(error >= 0); |
435 | return error; | |
436 | } | |
437 | ||
438 | /* | |
439 | * Zero any on disk space between the current EOF and the new, | |
440 | * larger EOF. This handles the normal case of zeroing the remainder | |
441 | * of the last block in the file and the unusual case of zeroing blocks | |
442 | * out beyond the size of the file. This second case only happens | |
443 | * with fixed size extents and when the system crashes before the inode | |
444 | * size was updated but after blocks were allocated. If fill is set, | |
445 | * then any holes in the range are filled and zeroed. If not, the holes | |
446 | * are left alone as holes. | |
447 | */ | |
448 | ||
449 | int /* error (positive) */ | |
450 | xfs_zero_eof( | |
541d7d3c | 451 | xfs_inode_t *ip, |
1da177e4 | 452 | xfs_off_t offset, /* starting I/O offset */ |
68160161 | 453 | xfs_fsize_t isize) /* current inode size */ |
1da177e4 | 454 | { |
613d7043 | 455 | xfs_mount_t *mp = ip->i_mount; |
1da177e4 LT |
456 | xfs_fileoff_t start_zero_fsb; |
457 | xfs_fileoff_t end_zero_fsb; | |
1da177e4 LT |
458 | xfs_fileoff_t zero_count_fsb; |
459 | xfs_fileoff_t last_fsb; | |
68160161 LM |
460 | xfs_fileoff_t zero_off; |
461 | xfs_fsize_t zero_len; | |
1da177e4 LT |
462 | int nimaps; |
463 | int error = 0; | |
464 | xfs_bmbt_irec_t imap; | |
1da177e4 | 465 | |
579aa9ca | 466 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_IOLOCK_EXCL)); |
24ee8088 | 467 | ASSERT(offset > isize); |
1da177e4 | 468 | |
1da177e4 LT |
469 | /* |
470 | * First handle zeroing the block on which isize resides. | |
471 | * We only zero a part of that block so it is handled specially. | |
472 | */ | |
541d7d3c | 473 | error = xfs_zero_last_block(ip, offset, isize); |
1da177e4 | 474 | if (error) { |
579aa9ca | 475 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_IOLOCK_EXCL)); |
1da177e4 LT |
476 | return error; |
477 | } | |
478 | ||
479 | /* | |
480 | * Calculate the range between the new size and the old | |
481 | * where blocks needing to be zeroed may exist. To get the | |
482 | * block where the last byte in the file currently resides, | |
483 | * we need to subtract one from the size and truncate back | |
484 | * to a block boundary. We subtract 1 in case the size is | |
485 | * exactly on a block boundary. | |
486 | */ | |
487 | last_fsb = isize ? XFS_B_TO_FSBT(mp, isize - 1) : (xfs_fileoff_t)-1; | |
488 | start_zero_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize); | |
489 | end_zero_fsb = XFS_B_TO_FSBT(mp, offset - 1); | |
490 | ASSERT((xfs_sfiloff_t)last_fsb < (xfs_sfiloff_t)start_zero_fsb); | |
491 | if (last_fsb == end_zero_fsb) { | |
492 | /* | |
493 | * The size was only incremented on its last block. | |
494 | * We took care of that above, so just return. | |
495 | */ | |
496 | return 0; | |
497 | } | |
498 | ||
499 | ASSERT(start_zero_fsb <= end_zero_fsb); | |
1da177e4 LT |
500 | while (start_zero_fsb <= end_zero_fsb) { |
501 | nimaps = 1; | |
502 | zero_count_fsb = end_zero_fsb - start_zero_fsb + 1; | |
541d7d3c | 503 | error = xfs_bmapi(NULL, ip, start_zero_fsb, zero_count_fsb, |
3e57ecf6 | 504 | 0, NULL, 0, &imap, &nimaps, NULL, NULL); |
1da177e4 | 505 | if (error) { |
579aa9ca | 506 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_IOLOCK_EXCL)); |
1da177e4 LT |
507 | return error; |
508 | } | |
509 | ASSERT(nimaps > 0); | |
510 | ||
511 | if (imap.br_state == XFS_EXT_UNWRITTEN || | |
512 | imap.br_startblock == HOLESTARTBLOCK) { | |
513 | /* | |
514 | * This loop handles initializing pages that were | |
515 | * partially initialized by the code below this | |
516 | * loop. It basically zeroes the part of the page | |
517 | * that sits on a hole and sets the page as P_HOLE | |
518 | * and calls remapf if it is a mapped file. | |
519 | */ | |
24ee8088 | 520 | start_zero_fsb = imap.br_startoff + imap.br_blockcount; |
1da177e4 LT |
521 | ASSERT(start_zero_fsb <= (end_zero_fsb + 1)); |
522 | continue; | |
523 | } | |
524 | ||
525 | /* | |
3d80ede4 | 526 | * There are blocks we need to zero. |
1da177e4 LT |
527 | * Drop the inode lock while we're doing the I/O. |
528 | * We'll still have the iolock to protect us. | |
529 | */ | |
579aa9ca | 530 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
1da177e4 | 531 | |
68160161 LM |
532 | zero_off = XFS_FSB_TO_B(mp, start_zero_fsb); |
533 | zero_len = XFS_FSB_TO_B(mp, imap.br_blockcount); | |
534 | ||
535 | if ((zero_off + zero_len) > offset) | |
536 | zero_len = offset - zero_off; | |
537 | ||
538 | error = xfs_iozero(ip, zero_off, zero_len); | |
1da177e4 LT |
539 | if (error) { |
540 | goto out_lock; | |
541 | } | |
542 | ||
3d80ede4 | 543 | start_zero_fsb = imap.br_startoff + imap.br_blockcount; |
1da177e4 LT |
544 | ASSERT(start_zero_fsb <= (end_zero_fsb + 1)); |
545 | ||
579aa9ca | 546 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
1da177e4 LT |
547 | } |
548 | ||
549 | return 0; | |
550 | ||
551 | out_lock: | |
579aa9ca | 552 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
1da177e4 LT |
553 | ASSERT(error >= 0); |
554 | return error; | |
555 | } | |
556 | ||
557 | ssize_t /* bytes written, or (-) error */ | |
558 | xfs_write( | |
993386c1 | 559 | struct xfs_inode *xip, |
1da177e4 LT |
560 | struct kiocb *iocb, |
561 | const struct iovec *iovp, | |
562 | unsigned int nsegs, | |
563 | loff_t *offset, | |
993386c1 | 564 | int ioflags) |
1da177e4 LT |
565 | { |
566 | struct file *file = iocb->ki_filp; | |
567 | struct address_space *mapping = file->f_mapping; | |
568 | struct inode *inode = mapping->host; | |
569 | unsigned long segs = nsegs; | |
1da177e4 LT |
570 | xfs_mount_t *mp; |
571 | ssize_t ret = 0, error = 0; | |
572 | xfs_fsize_t isize, new_size; | |
1da177e4 LT |
573 | int iolock; |
574 | int eventsent = 0; | |
1da177e4 LT |
575 | size_t ocount = 0, count; |
576 | loff_t pos; | |
2a329631 | 577 | int need_i_mutex; |
1da177e4 LT |
578 | |
579 | XFS_STATS_INC(xs_write_calls); | |
580 | ||
0ceb3314 DM |
581 | error = generic_segment_checks(iovp, &segs, &ocount, VERIFY_READ); |
582 | if (error) | |
583 | return error; | |
1da177e4 LT |
584 | |
585 | count = ocount; | |
586 | pos = *offset; | |
587 | ||
588 | if (count == 0) | |
589 | return 0; | |
590 | ||
613d7043 | 591 | mp = xip->i_mount; |
1da177e4 | 592 | |
b267ce99 | 593 | xfs_wait_for_freeze(mp, SB_FREEZE_WRITE); |
34327e13 | 594 | |
1da177e4 LT |
595 | if (XFS_FORCED_SHUTDOWN(mp)) |
596 | return -EIO; | |
597 | ||
1da177e4 | 598 | relock: |
2a329631 LM |
599 | if (ioflags & IO_ISDIRECT) { |
600 | iolock = XFS_IOLOCK_SHARED; | |
2a329631 LM |
601 | need_i_mutex = 0; |
602 | } else { | |
1da177e4 | 603 | iolock = XFS_IOLOCK_EXCL; |
2a329631 | 604 | need_i_mutex = 1; |
1b1dcc1b | 605 | mutex_lock(&inode->i_mutex); |
1da177e4 LT |
606 | } |
607 | ||
608 | xfs_ilock(xip, XFS_ILOCK_EXCL|iolock); | |
609 | ||
1da177e4 LT |
610 | start: |
611 | error = -generic_write_checks(file, &pos, &count, | |
612 | S_ISBLK(inode->i_mode)); | |
613 | if (error) { | |
614 | xfs_iunlock(xip, XFS_ILOCK_EXCL|iolock); | |
a13828b1 | 615 | goto out_unlock_mutex; |
1da177e4 LT |
616 | } |
617 | ||
eb9df39d | 618 | if ((DM_EVENT_ENABLED(xip, DM_EVENT_WRITE) && |
1da177e4 | 619 | !(ioflags & IO_INVIS) && !eventsent)) { |
1da177e4 LT |
620 | int dmflags = FILP_DELAY_FLAG(file); |
621 | ||
a13828b1 | 622 | if (need_i_mutex) |
1b1dcc1b | 623 | dmflags |= DM_FLAGS_IMUX; |
1da177e4 LT |
624 | |
625 | xfs_iunlock(xip, XFS_ILOCK_EXCL); | |
bc4ac74a | 626 | error = XFS_SEND_DATA(xip->i_mount, DM_EVENT_WRITE, xip, |
126468b1 | 627 | pos, count, dmflags, &iolock); |
1da177e4 | 628 | if (error) { |
ba87ea69 | 629 | goto out_unlock_internal; |
1da177e4 LT |
630 | } |
631 | xfs_ilock(xip, XFS_ILOCK_EXCL); | |
632 | eventsent = 1; | |
633 | ||
634 | /* | |
c41564b5 | 635 | * The iolock was dropped and reacquired in XFS_SEND_DATA |
1da177e4 LT |
636 | * so we have to recheck the size when appending. |
637 | * We will only "goto start;" once, since having sent the | |
638 | * event prevents another call to XFS_SEND_DATA, which is | |
639 | * what allows the size to change in the first place. | |
640 | */ | |
71dfd5a3 | 641 | if ((file->f_flags & O_APPEND) && pos != xip->i_size) |
1da177e4 | 642 | goto start; |
1da177e4 LT |
643 | } |
644 | ||
71dfd5a3 LM |
645 | if (ioflags & IO_ISDIRECT) { |
646 | xfs_buftarg_t *target = | |
71ddabb9 | 647 | XFS_IS_REALTIME_INODE(xip) ? |
71dfd5a3 LM |
648 | mp->m_rtdev_targp : mp->m_ddev_targp; |
649 | ||
650 | if ((pos & target->bt_smask) || (count & target->bt_smask)) { | |
651 | xfs_iunlock(xip, XFS_ILOCK_EXCL|iolock); | |
652 | return XFS_ERROR(-EINVAL); | |
653 | } | |
654 | ||
dcf49cc5 | 655 | if (!need_i_mutex && (mapping->nrpages || pos > xip->i_size)) { |
71dfd5a3 LM |
656 | xfs_iunlock(xip, XFS_ILOCK_EXCL|iolock); |
657 | iolock = XFS_IOLOCK_EXCL; | |
71dfd5a3 LM |
658 | need_i_mutex = 1; |
659 | mutex_lock(&inode->i_mutex); | |
660 | xfs_ilock(xip, XFS_ILOCK_EXCL|iolock); | |
1da177e4 LT |
661 | goto start; |
662 | } | |
663 | } | |
664 | ||
71dfd5a3 LM |
665 | new_size = pos + count; |
666 | if (new_size > xip->i_size) | |
613d7043 | 667 | xip->i_new_size = new_size; |
71dfd5a3 | 668 | |
2234d54d | 669 | if (likely(!(ioflags & IO_INVIS))) |
3a76c1ea | 670 | xfs_ichgtime(xip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); |
1da177e4 LT |
671 | |
672 | /* | |
673 | * If the offset is beyond the size of the file, we have a couple | |
674 | * of things to do. First, if there is already space allocated | |
675 | * we need to either create holes or zero the disk or ... | |
676 | * | |
677 | * If there is a page where the previous size lands, we need | |
678 | * to zero it out up to the new size. | |
679 | */ | |
680 | ||
ba87ea69 | 681 | if (pos > xip->i_size) { |
541d7d3c | 682 | error = xfs_zero_eof(xip, pos, xip->i_size); |
1da177e4 | 683 | if (error) { |
ba87ea69 LM |
684 | xfs_iunlock(xip, XFS_ILOCK_EXCL); |
685 | goto out_unlock_internal; | |
1da177e4 LT |
686 | } |
687 | } | |
688 | xfs_iunlock(xip, XFS_ILOCK_EXCL); | |
689 | ||
690 | /* | |
691 | * If we're writing the file then make sure to clear the | |
692 | * setuid and setgid bits if the process is not being run | |
693 | * by root. This keeps people from modifying setuid and | |
694 | * setgid binaries. | |
695 | */ | |
696 | ||
697 | if (((xip->i_d.di_mode & S_ISUID) || | |
698 | ((xip->i_d.di_mode & (S_ISGID | S_IXGRP)) == | |
699 | (S_ISGID | S_IXGRP))) && | |
700 | !capable(CAP_FSETID)) { | |
701 | error = xfs_write_clear_setuid(xip); | |
702 | if (likely(!error)) | |
2f1936b8 | 703 | error = -file_remove_suid(file); |
1da177e4 | 704 | if (unlikely(error)) { |
ba87ea69 | 705 | goto out_unlock_internal; |
1da177e4 LT |
706 | } |
707 | } | |
708 | ||
1da177e4 LT |
709 | /* We can write back this queue in page reclaim */ |
710 | current->backing_dev_info = mapping->backing_dev_info; | |
711 | ||
712 | if ((ioflags & IO_ISDIRECT)) { | |
dcf49cc5 | 713 | if (mapping->nrpages) { |
2a329631 | 714 | WARN_ON(need_i_mutex == 0); |
613d7043 | 715 | xfs_inval_cached_trace(xip, pos, -1, |
e6a4b37f | 716 | (pos & PAGE_CACHE_MASK), -1); |
739bfb2a | 717 | error = xfs_flushinval_pages(xip, |
e6a4b37f | 718 | (pos & PAGE_CACHE_MASK), |
1da177e4 | 719 | -1, FI_REMAPF_LOCKED); |
d3cf2094 LM |
720 | if (error) |
721 | goto out_unlock_internal; | |
1da177e4 LT |
722 | } |
723 | ||
a13828b1 | 724 | if (need_i_mutex) { |
1da177e4 | 725 | /* demote the lock now the cached pages are gone */ |
541d7d3c | 726 | xfs_ilock_demote(xip, XFS_IOLOCK_EXCL); |
1b1dcc1b | 727 | mutex_unlock(&inode->i_mutex); |
1da177e4 LT |
728 | |
729 | iolock = XFS_IOLOCK_SHARED; | |
a13828b1 | 730 | need_i_mutex = 0; |
1da177e4 LT |
731 | } |
732 | ||
613d7043 | 733 | xfs_rw_enter_trace(XFS_DIOWR_ENTER, xip, (void *)iovp, segs, |
1da177e4 LT |
734 | *offset, ioflags); |
735 | ret = generic_file_direct_write(iocb, iovp, | |
736 | &segs, pos, offset, count, ocount); | |
737 | ||
738 | /* | |
739 | * direct-io write to a hole: fall through to buffered I/O | |
740 | * for completing the rest of the request. | |
741 | */ | |
742 | if (ret >= 0 && ret != count) { | |
743 | XFS_STATS_ADD(xs_write_bytes, ret); | |
744 | ||
745 | pos += ret; | |
746 | count -= ret; | |
747 | ||
1da177e4 LT |
748 | ioflags &= ~IO_ISDIRECT; |
749 | xfs_iunlock(xip, iolock); | |
750 | goto relock; | |
751 | } | |
752 | } else { | |
5825294e DC |
753 | int enospc = 0; |
754 | ssize_t ret2 = 0; | |
755 | ||
756 | write_retry: | |
613d7043 | 757 | xfs_rw_enter_trace(XFS_WRITE_ENTER, xip, (void *)iovp, segs, |
1da177e4 | 758 | *offset, ioflags); |
5825294e | 759 | ret2 = generic_file_buffered_write(iocb, iovp, segs, |
1da177e4 | 760 | pos, offset, count, ret); |
5825294e DC |
761 | /* |
762 | * if we just got an ENOSPC, flush the inode now we | |
763 | * aren't holding any page locks and retry *once* | |
764 | */ | |
765 | if (ret2 == -ENOSPC && !enospc) { | |
766 | error = xfs_flush_pages(xip, 0, -1, 0, FI_NONE); | |
767 | if (error) | |
768 | goto out_unlock_internal; | |
769 | enospc = 1; | |
770 | goto write_retry; | |
771 | } | |
772 | ret = ret2; | |
1da177e4 LT |
773 | } |
774 | ||
775 | current->backing_dev_info = NULL; | |
776 | ||
777 | if (ret == -EIOCBQUEUED && !(ioflags & IO_ISAIO)) | |
778 | ret = wait_on_sync_kiocb(iocb); | |
779 | ||
25051158 LM |
780 | isize = i_size_read(inode); |
781 | if (unlikely(ret < 0 && ret != -EFAULT && *offset > isize)) | |
782 | *offset = isize; | |
783 | ||
784 | if (*offset > xip->i_size) { | |
785 | xfs_ilock(xip, XFS_ILOCK_EXCL); | |
786 | if (*offset > xip->i_size) | |
787 | xip->i_size = *offset; | |
788 | xfs_iunlock(xip, XFS_ILOCK_EXCL); | |
789 | } | |
790 | ||
eb9df39d CH |
791 | if (ret == -ENOSPC && |
792 | DM_EVENT_ENABLED(xip, DM_EVENT_NOSPACE) && !(ioflags & IO_INVIS)) { | |
126468b1 | 793 | xfs_iunlock(xip, iolock); |
a13828b1 | 794 | if (need_i_mutex) |
1b1dcc1b | 795 | mutex_unlock(&inode->i_mutex); |
bc4ac74a CH |
796 | error = XFS_SEND_NAMESP(xip->i_mount, DM_EVENT_NOSPACE, xip, |
797 | DM_RIGHT_NULL, xip, DM_RIGHT_NULL, NULL, NULL, | |
1da177e4 | 798 | 0, 0, 0); /* Delay flag intentionally unused */ |
a13828b1 | 799 | if (need_i_mutex) |
1b1dcc1b | 800 | mutex_lock(&inode->i_mutex); |
126468b1 | 801 | xfs_ilock(xip, iolock); |
ba87ea69 LM |
802 | if (error) |
803 | goto out_unlock_internal; | |
25051158 | 804 | goto start; |
1da177e4 LT |
805 | } |
806 | ||
807 | error = -ret; | |
808 | if (ret <= 0) | |
809 | goto out_unlock_internal; | |
810 | ||
811 | XFS_STATS_ADD(xs_write_bytes, ret); | |
812 | ||
813 | /* Handle various SYNC-type writes */ | |
814 | if ((file->f_flags & O_SYNC) || IS_SYNC(inode)) { | |
5903c495 | 815 | int error2; |
126468b1 CH |
816 | |
817 | xfs_iunlock(xip, iolock); | |
a13828b1 | 818 | if (need_i_mutex) |
1b1dcc1b | 819 | mutex_unlock(&inode->i_mutex); |
5903c495 | 820 | error2 = sync_page_range(inode, mapping, pos, ret); |
1da177e4 | 821 | if (!error) |
5903c495 | 822 | error = error2; |
ba87ea69 LM |
823 | if (need_i_mutex) |
824 | mutex_lock(&inode->i_mutex); | |
126468b1 | 825 | xfs_ilock(xip, iolock); |
5903c495 LM |
826 | error2 = xfs_write_sync_logforce(mp, xip); |
827 | if (!error) | |
828 | error = error2; | |
1da177e4 LT |
829 | } |
830 | ||
831 | out_unlock_internal: | |
613d7043 | 832 | if (xip->i_new_size) { |
ba87ea69 | 833 | xfs_ilock(xip, XFS_ILOCK_EXCL); |
613d7043 | 834 | xip->i_new_size = 0; |
ba87ea69 LM |
835 | /* |
836 | * If this was a direct or synchronous I/O that failed (such | |
837 | * as ENOSPC) then part of the I/O may have been written to | |
838 | * disk before the error occured. In this case the on-disk | |
839 | * file size may have been adjusted beyond the in-memory file | |
840 | * size and now needs to be truncated back. | |
841 | */ | |
842 | if (xip->i_d.di_size > xip->i_size) | |
843 | xip->i_d.di_size = xip->i_size; | |
844 | xfs_iunlock(xip, XFS_ILOCK_EXCL); | |
845 | } | |
126468b1 | 846 | xfs_iunlock(xip, iolock); |
a13828b1 NS |
847 | out_unlock_mutex: |
848 | if (need_i_mutex) | |
1b1dcc1b | 849 | mutex_unlock(&inode->i_mutex); |
1da177e4 LT |
850 | return -error; |
851 | } | |
852 | ||
853 | /* | |
854 | * All xfs metadata buffers except log state machine buffers | |
855 | * get this attached as their b_bdstrat callback function. | |
856 | * This is so that we can catch a buffer | |
857 | * after prematurely unpinning it to forcibly shutdown the filesystem. | |
858 | */ | |
859 | int | |
860 | xfs_bdstrat_cb(struct xfs_buf *bp) | |
861 | { | |
15ac08a8 | 862 | if (XFS_FORCED_SHUTDOWN(bp->b_mount)) { |
1da177e4 LT |
863 | xfs_buftrace("XFS__BDSTRAT IOERROR", bp); |
864 | /* | |
865 | * Metadata write that didn't get logged but | |
866 | * written delayed anyway. These aren't associated | |
867 | * with a transaction, and can be ignored. | |
868 | */ | |
869 | if (XFS_BUF_IODONE_FUNC(bp) == NULL && | |
870 | (XFS_BUF_ISREAD(bp)) == 0) | |
871 | return (xfs_bioerror_relse(bp)); | |
872 | else | |
873 | return (xfs_bioerror(bp)); | |
874 | } | |
15ac08a8 CH |
875 | |
876 | xfs_buf_iorequest(bp); | |
877 | return 0; | |
1da177e4 LT |
878 | } |
879 | ||
1da177e4 | 880 | /* |
d64e31a2 DC |
881 | * Wrapper around bdstrat so that we can stop data from going to disk in case |
882 | * we are shutting down the filesystem. Typically user data goes thru this | |
883 | * path; one of the exceptions is the superblock. | |
1da177e4 | 884 | */ |
d64e31a2 | 885 | void |
1da177e4 LT |
886 | xfsbdstrat( |
887 | struct xfs_mount *mp, | |
888 | struct xfs_buf *bp) | |
889 | { | |
890 | ASSERT(mp); | |
d4055947 | 891 | if (!XFS_FORCED_SHUTDOWN(mp)) { |
ce8e922c | 892 | xfs_buf_iorequest(bp); |
d4055947 DC |
893 | return; |
894 | } | |
1da177e4 LT |
895 | |
896 | xfs_buftrace("XFSBDSTRAT IOERROR", bp); | |
d64e31a2 | 897 | xfs_bioerror_relse(bp); |
1da177e4 LT |
898 | } |
899 | ||
900 | /* | |
901 | * If the underlying (data/log/rt) device is readonly, there are some | |
902 | * operations that cannot proceed. | |
903 | */ | |
904 | int | |
905 | xfs_dev_is_read_only( | |
906 | xfs_mount_t *mp, | |
907 | char *message) | |
908 | { | |
909 | if (xfs_readonly_buftarg(mp->m_ddev_targp) || | |
910 | xfs_readonly_buftarg(mp->m_logdev_targp) || | |
911 | (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) { | |
912 | cmn_err(CE_NOTE, | |
913 | "XFS: %s required on read-only device.", message); | |
914 | cmn_err(CE_NOTE, | |
915 | "XFS: write access unavailable, cannot proceed."); | |
916 | return EROFS; | |
917 | } | |
918 | return 0; | |
919 | } |