Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
3e57ecf6 | 2 | * Copyright (c) 2000-2006 Silicon Graphics, Inc. |
7b718769 | 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 | */ |
16f7e0fe | 18 | |
1da177e4 | 19 | #include "xfs.h" |
a844f451 | 20 | #include "xfs_fs.h" |
1da177e4 | 21 | #include "xfs_types.h" |
a844f451 | 22 | #include "xfs_bit.h" |
1da177e4 LT |
23 | #include "xfs_log.h" |
24 | #include "xfs_trans.h" | |
25 | #include "xfs_sb.h" | |
26 | #include "xfs_ag.h" | |
1da177e4 | 27 | #include "xfs_dir2.h" |
1da177e4 | 28 | #include "xfs_mount.h" |
a844f451 | 29 | #include "xfs_da_btree.h" |
1da177e4 LT |
30 | #include "xfs_bmap_btree.h" |
31 | #include "xfs_ialloc_btree.h" | |
1da177e4 | 32 | #include "xfs_dinode.h" |
1da177e4 | 33 | #include "xfs_inode.h" |
a844f451 | 34 | #include "xfs_inode_item.h" |
a844f451 | 35 | #include "xfs_itable.h" |
a844f451 NS |
36 | #include "xfs_ialloc.h" |
37 | #include "xfs_alloc.h" | |
1da177e4 | 38 | #include "xfs_bmap.h" |
ef14f0c1 | 39 | #include "xfs_acl.h" |
1da177e4 | 40 | #include "xfs_attr.h" |
1da177e4 | 41 | #include "xfs_error.h" |
1da177e4 LT |
42 | #include "xfs_quota.h" |
43 | #include "xfs_utils.h" | |
a844f451 | 44 | #include "xfs_rtalloc.h" |
1da177e4 | 45 | #include "xfs_trans_space.h" |
1da177e4 | 46 | #include "xfs_log_priv.h" |
2a82b8be | 47 | #include "xfs_filestream.h" |
993386c1 | 48 | #include "xfs_vnodeops.h" |
0b1b213f | 49 | #include "xfs_trace.h" |
1da177e4 | 50 | |
7d4fb40a NS |
51 | /* |
52 | * The maximum pathlen is 1024 bytes. Since the minimum file system | |
53 | * blocksize is 512 bytes, we can get a max of 2 extents back from | |
54 | * bmapi. | |
55 | */ | |
56 | #define SYMLINK_MAPS 2 | |
57 | ||
804c83c3 CH |
58 | STATIC int |
59 | xfs_readlink_bmap( | |
60 | xfs_inode_t *ip, | |
61 | char *link) | |
62 | { | |
63 | xfs_mount_t *mp = ip->i_mount; | |
64 | int pathlen = ip->i_d.di_size; | |
65 | int nmaps = SYMLINK_MAPS; | |
66 | xfs_bmbt_irec_t mval[SYMLINK_MAPS]; | |
67 | xfs_daddr_t d; | |
68 | int byte_cnt; | |
69 | int n; | |
70 | xfs_buf_t *bp; | |
71 | int error = 0; | |
72 | ||
5c8ed202 DC |
73 | error = xfs_bmapi_read(ip, 0, XFS_B_TO_FSB(mp, pathlen), mval, &nmaps, |
74 | 0); | |
804c83c3 CH |
75 | if (error) |
76 | goto out; | |
77 | ||
78 | for (n = 0; n < nmaps; n++) { | |
79 | d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock); | |
80 | byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount); | |
81 | ||
611c9946 | 82 | bp = xfs_buf_read(mp->m_ddev_targp, d, BTOBB(byte_cnt), 0); |
ac4d6888 CS |
83 | if (!bp) |
84 | return XFS_ERROR(ENOMEM); | |
e5702805 | 85 | error = bp->b_error; |
804c83c3 | 86 | if (error) { |
901796af | 87 | xfs_buf_ioerror_alert(bp, __func__); |
804c83c3 CH |
88 | xfs_buf_relse(bp); |
89 | goto out; | |
90 | } | |
91 | if (pathlen < byte_cnt) | |
92 | byte_cnt = pathlen; | |
93 | pathlen -= byte_cnt; | |
94 | ||
62926044 | 95 | memcpy(link, bp->b_addr, byte_cnt); |
804c83c3 CH |
96 | xfs_buf_relse(bp); |
97 | } | |
98 | ||
99 | link[ip->i_d.di_size] = '\0'; | |
100 | error = 0; | |
101 | ||
102 | out: | |
103 | return error; | |
104 | } | |
105 | ||
993386c1 | 106 | int |
1da177e4 | 107 | xfs_readlink( |
993386c1 | 108 | xfs_inode_t *ip, |
804c83c3 | 109 | char *link) |
1da177e4 | 110 | { |
804c83c3 | 111 | xfs_mount_t *mp = ip->i_mount; |
b52a360b | 112 | xfs_fsize_t pathlen; |
1da177e4 | 113 | int error = 0; |
1da177e4 | 114 | |
cca28fb8 | 115 | trace_xfs_readlink(ip); |
1da177e4 LT |
116 | |
117 | if (XFS_FORCED_SHUTDOWN(mp)) | |
118 | return XFS_ERROR(EIO); | |
119 | ||
120 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
121 | ||
804c83c3 CH |
122 | pathlen = ip->i_d.di_size; |
123 | if (!pathlen) | |
124 | goto out; | |
1da177e4 | 125 | |
b52a360b CM |
126 | if (pathlen < 0 || pathlen > MAXPATHLEN) { |
127 | xfs_alert(mp, "%s: inode (%llu) bad symlink length (%lld)", | |
128 | __func__, (unsigned long long) ip->i_ino, | |
129 | (long long) pathlen); | |
130 | ASSERT(0); | |
9b025eb3 JK |
131 | error = XFS_ERROR(EFSCORRUPTED); |
132 | goto out; | |
b52a360b CM |
133 | } |
134 | ||
135 | ||
1da177e4 | 136 | if (ip->i_df.if_flags & XFS_IFINLINE) { |
804c83c3 CH |
137 | memcpy(link, ip->i_df.if_u1.if_data, pathlen); |
138 | link[pathlen] = '\0'; | |
139 | } else { | |
140 | error = xfs_readlink_bmap(ip, link); | |
1da177e4 LT |
141 | } |
142 | ||
804c83c3 | 143 | out: |
1da177e4 | 144 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
1da177e4 LT |
145 | return error; |
146 | } | |
147 | ||
c56c9631 CH |
148 | /* |
149 | * Flags for xfs_free_eofblocks | |
150 | */ | |
151 | #define XFS_FREE_EOF_TRYLOCK (1<<0) | |
152 | ||
1da177e4 | 153 | /* |
92dfe8d2 DC |
154 | * This is called by xfs_inactive to free any blocks beyond eof |
155 | * when the link count isn't zero and by xfs_dm_punch_hole() when | |
156 | * punching a hole to EOF. | |
1da177e4 | 157 | */ |
d96f8f89 | 158 | STATIC int |
92dfe8d2 | 159 | xfs_free_eofblocks( |
1da177e4 | 160 | xfs_mount_t *mp, |
92dfe8d2 DC |
161 | xfs_inode_t *ip, |
162 | int flags) | |
1da177e4 LT |
163 | { |
164 | xfs_trans_t *tp; | |
165 | int error; | |
166 | xfs_fileoff_t end_fsb; | |
167 | xfs_fileoff_t last_fsb; | |
168 | xfs_filblks_t map_len; | |
169 | int nimaps; | |
170 | xfs_bmbt_irec_t imap; | |
171 | ||
172 | /* | |
173 | * Figure out if there are any blocks beyond the end | |
174 | * of the file. If not, then there is nothing to do. | |
175 | */ | |
ce7ae151 | 176 | end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_ISIZE(ip)); |
32972383 | 177 | last_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes); |
3f34885c | 178 | if (last_fsb <= end_fsb) |
014c2544 | 179 | return 0; |
3f34885c | 180 | map_len = last_fsb - end_fsb; |
1da177e4 LT |
181 | |
182 | nimaps = 1; | |
183 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
5c8ed202 | 184 | error = xfs_bmapi_read(ip, end_fsb, map_len, &imap, &nimaps, 0); |
1da177e4 LT |
185 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
186 | ||
187 | if (!error && (nimaps != 0) && | |
68bdb6ea YL |
188 | (imap.br_startblock != HOLESTARTBLOCK || |
189 | ip->i_delayed_blks)) { | |
1da177e4 LT |
190 | /* |
191 | * Attach the dquots to the inode up front. | |
192 | */ | |
7d095257 CH |
193 | error = xfs_qm_dqattach(ip, 0); |
194 | if (error) | |
014c2544 | 195 | return error; |
1da177e4 LT |
196 | |
197 | /* | |
198 | * There are blocks after the end of file. | |
199 | * Free them up now by truncating the file to | |
200 | * its current size. | |
201 | */ | |
202 | tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE); | |
203 | ||
c56c9631 CH |
204 | if (flags & XFS_FREE_EOF_TRYLOCK) { |
205 | if (!xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) { | |
206 | xfs_trans_cancel(tp, 0); | |
207 | return 0; | |
208 | } | |
209 | } else { | |
92dfe8d2 | 210 | xfs_ilock(ip, XFS_IOLOCK_EXCL); |
c56c9631 | 211 | } |
1da177e4 LT |
212 | |
213 | error = xfs_trans_reserve(tp, 0, | |
214 | XFS_ITRUNCATE_LOG_RES(mp), | |
215 | 0, XFS_TRANS_PERM_LOG_RES, | |
216 | XFS_ITRUNCATE_LOG_COUNT); | |
217 | if (error) { | |
218 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
219 | xfs_trans_cancel(tp, 0); | |
220 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
014c2544 | 221 | return error; |
1da177e4 LT |
222 | } |
223 | ||
224 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
ddc3415a | 225 | xfs_trans_ijoin(tp, ip, 0); |
1da177e4 | 226 | |
673e8e59 CH |
227 | /* |
228 | * Do not update the on-disk file size. If we update the | |
229 | * on-disk file size and then the system crashes before the | |
230 | * contents of the file are flushed to disk then the files | |
231 | * may be full of holes (ie NULL files bug). | |
232 | */ | |
233 | error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, | |
ce7ae151 | 234 | XFS_ISIZE(ip)); |
1da177e4 | 235 | if (error) { |
8f04c47a CH |
236 | /* |
237 | * If we get an error at this point we simply don't | |
238 | * bother truncating the file. | |
239 | */ | |
1da177e4 LT |
240 | xfs_trans_cancel(tp, |
241 | (XFS_TRANS_RELEASE_LOG_RES | | |
242 | XFS_TRANS_ABORT)); | |
243 | } else { | |
244 | error = xfs_trans_commit(tp, | |
1c72bf90 | 245 | XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 | 246 | } |
c56c9631 | 247 | xfs_iunlock(ip, XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL); |
1da177e4 | 248 | } |
014c2544 | 249 | return error; |
1da177e4 LT |
250 | } |
251 | ||
252 | /* | |
253 | * Free a symlink that has blocks associated with it. | |
254 | */ | |
255 | STATIC int | |
256 | xfs_inactive_symlink_rmt( | |
257 | xfs_inode_t *ip, | |
258 | xfs_trans_t **tpp) | |
259 | { | |
260 | xfs_buf_t *bp; | |
261 | int committed; | |
262 | int done; | |
263 | int error; | |
264 | xfs_fsblock_t first_block; | |
265 | xfs_bmap_free_t free_list; | |
266 | int i; | |
267 | xfs_mount_t *mp; | |
268 | xfs_bmbt_irec_t mval[SYMLINK_MAPS]; | |
269 | int nmaps; | |
270 | xfs_trans_t *ntp; | |
271 | int size; | |
272 | xfs_trans_t *tp; | |
273 | ||
274 | tp = *tpp; | |
275 | mp = ip->i_mount; | |
276 | ASSERT(ip->i_d.di_size > XFS_IFORK_DSIZE(ip)); | |
277 | /* | |
278 | * We're freeing a symlink that has some | |
279 | * blocks allocated to it. Free the | |
280 | * blocks here. We know that we've got | |
281 | * either 1 or 2 extents and that we can | |
282 | * free them all in one bunmapi call. | |
283 | */ | |
284 | ASSERT(ip->i_d.di_nextents > 0 && ip->i_d.di_nextents <= 2); | |
b373e98d | 285 | |
1da177e4 LT |
286 | /* |
287 | * Lock the inode, fix the size, and join it to the transaction. | |
288 | * Hold it so in the normal path, we still have it locked for | |
289 | * the second transaction. In the error paths we need it | |
290 | * held so the cancel won't rele it, see below. | |
291 | */ | |
1da177e4 LT |
292 | size = (int)ip->i_d.di_size; |
293 | ip->i_d.di_size = 0; | |
1da177e4 LT |
294 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); |
295 | /* | |
296 | * Find the block(s) so we can inval and unmap them. | |
297 | */ | |
298 | done = 0; | |
9d87c319 | 299 | xfs_bmap_init(&free_list, &first_block); |
e8c96f8c | 300 | nmaps = ARRAY_SIZE(mval); |
5c8ed202 DC |
301 | error = xfs_bmapi_read(ip, 0, XFS_B_TO_FSB(mp, size), |
302 | mval, &nmaps, 0); | |
303 | if (error) | |
1da177e4 LT |
304 | goto error0; |
305 | /* | |
306 | * Invalidate the block(s). | |
307 | */ | |
308 | for (i = 0; i < nmaps; i++) { | |
309 | bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, | |
310 | XFS_FSB_TO_DADDR(mp, mval[i].br_startblock), | |
311 | XFS_FSB_TO_BB(mp, mval[i].br_blockcount), 0); | |
2a30f36d CS |
312 | if (!bp) { |
313 | error = ENOMEM; | |
314 | goto error1; | |
315 | } | |
1da177e4 LT |
316 | xfs_trans_binval(tp, bp); |
317 | } | |
318 | /* | |
319 | * Unmap the dead block(s) to the free_list. | |
320 | */ | |
321 | if ((error = xfs_bunmapi(tp, ip, 0, size, XFS_BMAPI_METADATA, nmaps, | |
b4e9181e | 322 | &first_block, &free_list, &done))) |
1da177e4 LT |
323 | goto error1; |
324 | ASSERT(done); | |
325 | /* | |
326 | * Commit the first transaction. This logs the EFI and the inode. | |
327 | */ | |
f7c99b6f | 328 | if ((error = xfs_bmap_finish(&tp, &free_list, &committed))) |
1da177e4 LT |
329 | goto error1; |
330 | /* | |
331 | * The transaction must have been committed, since there were | |
332 | * actually extents freed by xfs_bunmapi. See xfs_bmap_finish. | |
333 | * The new tp has the extent freeing and EFDs. | |
334 | */ | |
335 | ASSERT(committed); | |
336 | /* | |
337 | * The first xact was committed, so add the inode to the new one. | |
338 | * Mark it dirty so it will be logged and moved forward in the log as | |
339 | * part of every commit. | |
340 | */ | |
ddc3415a | 341 | xfs_trans_ijoin(tp, ip, 0); |
1da177e4 LT |
342 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); |
343 | /* | |
344 | * Get a new, empty transaction to return to our caller. | |
345 | */ | |
346 | ntp = xfs_trans_dup(tp); | |
347 | /* | |
c41564b5 | 348 | * Commit the transaction containing extent freeing and EFDs. |
1da177e4 LT |
349 | * If we get an error on the commit here or on the reserve below, |
350 | * we need to unlock the inode since the new transaction doesn't | |
351 | * have the inode attached. | |
352 | */ | |
1c72bf90 | 353 | error = xfs_trans_commit(tp, 0); |
1da177e4 LT |
354 | tp = ntp; |
355 | if (error) { | |
356 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
357 | goto error0; | |
358 | } | |
cc09c0dc DC |
359 | /* |
360 | * transaction commit worked ok so we can drop the extra ticket | |
361 | * reference that we gained in xfs_trans_dup() | |
362 | */ | |
363 | xfs_log_ticket_put(tp->t_ticket); | |
364 | ||
1da177e4 LT |
365 | /* |
366 | * Remove the memory for extent descriptions (just bookkeeping). | |
367 | */ | |
368 | if (ip->i_df.if_bytes) | |
369 | xfs_idata_realloc(ip, -ip->i_df.if_bytes, XFS_DATA_FORK); | |
370 | ASSERT(ip->i_df.if_bytes == 0); | |
371 | /* | |
372 | * Put an itruncate log reservation in the new transaction | |
373 | * for our caller. | |
374 | */ | |
375 | if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0, | |
376 | XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) { | |
377 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
378 | goto error0; | |
379 | } | |
b373e98d CH |
380 | |
381 | xfs_trans_ijoin(tp, ip, 0); | |
1da177e4 LT |
382 | *tpp = tp; |
383 | return 0; | |
384 | ||
385 | error1: | |
386 | xfs_bmap_cancel(&free_list); | |
387 | error0: | |
1da177e4 | 388 | return error; |
1da177e4 LT |
389 | } |
390 | ||
1da177e4 LT |
391 | STATIC int |
392 | xfs_inactive_attrs( | |
393 | xfs_inode_t *ip, | |
394 | xfs_trans_t **tpp) | |
395 | { | |
396 | xfs_trans_t *tp; | |
397 | int error; | |
398 | xfs_mount_t *mp; | |
399 | ||
579aa9ca | 400 | ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL)); |
1da177e4 LT |
401 | tp = *tpp; |
402 | mp = ip->i_mount; | |
403 | ASSERT(ip->i_d.di_forkoff != 0); | |
e5720eec | 404 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 | 405 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
e5720eec DC |
406 | if (error) |
407 | goto error_unlock; | |
1da177e4 LT |
408 | |
409 | error = xfs_attr_inactive(ip); | |
e5720eec DC |
410 | if (error) |
411 | goto error_unlock; | |
1da177e4 LT |
412 | |
413 | tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE); | |
414 | error = xfs_trans_reserve(tp, 0, | |
415 | XFS_IFREE_LOG_RES(mp), | |
416 | 0, XFS_TRANS_PERM_LOG_RES, | |
417 | XFS_INACTIVE_LOG_COUNT); | |
e5720eec DC |
418 | if (error) |
419 | goto error_cancel; | |
1da177e4 LT |
420 | |
421 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
ddc3415a | 422 | xfs_trans_ijoin(tp, ip, 0); |
1da177e4 LT |
423 | xfs_idestroy_fork(ip, XFS_ATTR_FORK); |
424 | ||
425 | ASSERT(ip->i_d.di_anextents == 0); | |
426 | ||
427 | *tpp = tp; | |
014c2544 | 428 | return 0; |
e5720eec DC |
429 | |
430 | error_cancel: | |
431 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
432 | xfs_trans_cancel(tp, 0); | |
433 | error_unlock: | |
434 | *tpp = NULL; | |
435 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
436 | return error; | |
1da177e4 LT |
437 | } |
438 | ||
993386c1 | 439 | int |
1da177e4 | 440 | xfs_release( |
993386c1 | 441 | xfs_inode_t *ip) |
1da177e4 | 442 | { |
993386c1 | 443 | xfs_mount_t *mp = ip->i_mount; |
1da177e4 LT |
444 | int error; |
445 | ||
42173f68 | 446 | if (!S_ISREG(ip->i_d.di_mode) || (ip->i_d.di_mode == 0)) |
1da177e4 | 447 | return 0; |
1da177e4 LT |
448 | |
449 | /* If this is a read-only mount, don't do this (would generate I/O) */ | |
bd186aa9 | 450 | if (mp->m_flags & XFS_MOUNT_RDONLY) |
1da177e4 LT |
451 | return 0; |
452 | ||
2a82b8be | 453 | if (!XFS_FORCED_SHUTDOWN(mp)) { |
b3aea4ed CH |
454 | int truncated; |
455 | ||
2a82b8be DC |
456 | /* |
457 | * If we are using filestreams, and we have an unlinked | |
458 | * file that we are processing the last close on, then nothing | |
459 | * will be able to reopen and write to this file. Purge this | |
460 | * inode from the filestreams cache so that it doesn't delay | |
461 | * teardown of the inode. | |
462 | */ | |
463 | if ((ip->i_d.di_nlink == 0) && xfs_inode_is_filestream(ip)) | |
464 | xfs_filestream_deassociate(ip); | |
465 | ||
fbf3ce8d CH |
466 | /* |
467 | * If we previously truncated this file and removed old data | |
468 | * in the process, we want to initiate "early" writeout on | |
469 | * the last close. This is an attempt to combat the notorious | |
25985edc | 470 | * NULL files problem which is particularly noticeable from a |
fbf3ce8d CH |
471 | * truncate down, buffered (re-)write (delalloc), followed by |
472 | * a crash. What we are effectively doing here is | |
473 | * significantly reducing the time window where we'd otherwise | |
474 | * be exposed to that problem. | |
475 | */ | |
09262b43 | 476 | truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED); |
df4368a1 DC |
477 | if (truncated) { |
478 | xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE); | |
479 | if (VN_DIRTY(VFS_I(ip)) && ip->i_delayed_blks > 0) | |
480 | xfs_flush_pages(ip, 0, -1, XBF_ASYNC, FI_NONE); | |
481 | } | |
fbf3ce8d CH |
482 | } |
483 | ||
6e857567 DC |
484 | if (ip->i_d.di_nlink == 0) |
485 | return 0; | |
c56c9631 | 486 | |
abbede1b | 487 | if ((S_ISREG(ip->i_d.di_mode) && |
ce7ae151 CH |
488 | (VFS_I(ip)->i_size > 0 || |
489 | (VN_CACHED(VFS_I(ip)) > 0 || ip->i_delayed_blks > 0)) && | |
6e857567 DC |
490 | (ip->i_df.if_flags & XFS_IFEXTENTS)) && |
491 | (!(ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)))) { | |
492 | ||
493 | /* | |
494 | * If we can't get the iolock just skip truncating the blocks | |
495 | * past EOF because we could deadlock with the mmap_sem | |
496 | * otherwise. We'll get another chance to drop them once the | |
497 | * last reference to the inode is dropped, so we'll never leak | |
498 | * blocks permanently. | |
499 | * | |
500 | * Further, check if the inode is being opened, written and | |
501 | * closed frequently and we have delayed allocation blocks | |
25985edc | 502 | * outstanding (e.g. streaming writes from the NFS server), |
6e857567 DC |
503 | * truncating the blocks past EOF will cause fragmentation to |
504 | * occur. | |
505 | * | |
506 | * In this case don't do the truncation, either, but we have to | |
507 | * be careful how we detect this case. Blocks beyond EOF show | |
508 | * up as i_delayed_blks even when the inode is clean, so we | |
509 | * need to truncate them away first before checking for a dirty | |
510 | * release. Hence on the first dirty close we will still remove | |
511 | * the speculative allocation, but after that we will leave it | |
512 | * in place. | |
513 | */ | |
514 | if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE)) | |
515 | return 0; | |
1da177e4 | 516 | |
6e857567 DC |
517 | error = xfs_free_eofblocks(mp, ip, |
518 | XFS_FREE_EOF_TRYLOCK); | |
519 | if (error) | |
520 | return error; | |
521 | ||
522 | /* delalloc blocks after truncation means it really is dirty */ | |
523 | if (ip->i_delayed_blks) | |
524 | xfs_iflags_set(ip, XFS_IDIRTY_RELEASE); | |
525 | } | |
1da177e4 LT |
526 | return 0; |
527 | } | |
528 | ||
529 | /* | |
530 | * xfs_inactive | |
531 | * | |
532 | * This is called when the vnode reference count for the vnode | |
533 | * goes to zero. If the file has been unlinked, then it must | |
534 | * now be truncated. Also, we clear all of the read-ahead state | |
535 | * kept for the inode here since the file is now closed. | |
536 | */ | |
993386c1 | 537 | int |
1da177e4 | 538 | xfs_inactive( |
993386c1 | 539 | xfs_inode_t *ip) |
1da177e4 | 540 | { |
67fcaa73 | 541 | xfs_bmap_free_t free_list; |
1da177e4 LT |
542 | xfs_fsblock_t first_block; |
543 | int committed; | |
544 | xfs_trans_t *tp; | |
545 | xfs_mount_t *mp; | |
546 | int error; | |
b373e98d | 547 | int truncate = 0; |
1da177e4 | 548 | |
1da177e4 LT |
549 | /* |
550 | * If the inode is already free, then there can be nothing | |
551 | * to clean up here. | |
552 | */ | |
cb4c8cc1 | 553 | if (ip->i_d.di_mode == 0 || is_bad_inode(VFS_I(ip))) { |
1da177e4 LT |
554 | ASSERT(ip->i_df.if_real_bytes == 0); |
555 | ASSERT(ip->i_df.if_broot_bytes == 0); | |
556 | return VN_INACTIVE_CACHE; | |
557 | } | |
558 | ||
1da177e4 LT |
559 | mp = ip->i_mount; |
560 | ||
1da177e4 LT |
561 | error = 0; |
562 | ||
563 | /* If this is a read-only mount, don't do this (would generate I/O) */ | |
bd186aa9 | 564 | if (mp->m_flags & XFS_MOUNT_RDONLY) |
1da177e4 LT |
565 | goto out; |
566 | ||
567 | if (ip->i_d.di_nlink != 0) { | |
abbede1b | 568 | if ((S_ISREG(ip->i_d.di_mode) && |
ce7ae151 CH |
569 | (VFS_I(ip)->i_size > 0 || |
570 | (VN_CACHED(VFS_I(ip)) > 0 || ip->i_delayed_blks > 0)) && | |
571 | (ip->i_df.if_flags & XFS_IFEXTENTS) && | |
572 | (!(ip->i_d.di_flags & | |
dd9f438e | 573 | (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)) || |
ce7ae151 | 574 | ip->i_delayed_blks != 0))) { |
c56c9631 | 575 | error = xfs_free_eofblocks(mp, ip, 0); |
92dfe8d2 | 576 | if (error) |
014c2544 | 577 | return VN_INACTIVE_CACHE; |
1da177e4 LT |
578 | } |
579 | goto out; | |
580 | } | |
581 | ||
b373e98d CH |
582 | if (S_ISREG(ip->i_d.di_mode) && |
583 | (ip->i_d.di_size != 0 || XFS_ISIZE(ip) != 0 || | |
584 | ip->i_d.di_nextents > 0 || ip->i_delayed_blks > 0)) | |
585 | truncate = 1; | |
1da177e4 | 586 | |
7d095257 CH |
587 | error = xfs_qm_dqattach(ip, 0); |
588 | if (error) | |
014c2544 | 589 | return VN_INACTIVE_CACHE; |
1da177e4 LT |
590 | |
591 | tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE); | |
b373e98d CH |
592 | error = xfs_trans_reserve(tp, 0, |
593 | (truncate || S_ISLNK(ip->i_d.di_mode)) ? | |
594 | XFS_ITRUNCATE_LOG_RES(mp) : | |
595 | XFS_IFREE_LOG_RES(mp), | |
596 | 0, | |
597 | XFS_TRANS_PERM_LOG_RES, | |
598 | XFS_ITRUNCATE_LOG_COUNT); | |
599 | if (error) { | |
600 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
601 | xfs_trans_cancel(tp, 0); | |
602 | return VN_INACTIVE_CACHE; | |
603 | } | |
1da177e4 | 604 | |
b373e98d CH |
605 | xfs_ilock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); |
606 | xfs_trans_ijoin(tp, ip, 0); | |
1da177e4 | 607 | |
b373e98d CH |
608 | if (S_ISLNK(ip->i_d.di_mode)) { |
609 | /* | |
610 | * Zero length symlinks _can_ exist. | |
611 | */ | |
612 | if (ip->i_d.di_size > XFS_IFORK_DSIZE(ip)) { | |
613 | error = xfs_inactive_symlink_rmt(ip, &tp); | |
614 | if (error) | |
615 | goto out_cancel; | |
616 | } else if (ip->i_df.if_bytes > 0) { | |
617 | xfs_idata_realloc(ip, -(ip->i_df.if_bytes), | |
618 | XFS_DATA_FORK); | |
619 | ASSERT(ip->i_df.if_bytes == 0); | |
620 | } | |
621 | } else if (truncate) { | |
673e8e59 | 622 | ip->i_d.di_size = 0; |
673e8e59 CH |
623 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); |
624 | ||
625 | error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0); | |
b373e98d CH |
626 | if (error) |
627 | goto out_cancel; | |
673e8e59 CH |
628 | |
629 | ASSERT(ip->i_d.di_nextents == 0); | |
1da177e4 LT |
630 | } |
631 | ||
632 | /* | |
633 | * If there are attributes associated with the file | |
634 | * then blow them away now. The code calls a routine | |
635 | * that recursively deconstructs the attribute fork. | |
636 | * We need to just commit the current transaction | |
637 | * because we can't use it for xfs_attr_inactive(). | |
638 | */ | |
639 | if (ip->i_d.di_anextents > 0) { | |
640 | error = xfs_inactive_attrs(ip, &tp); | |
641 | /* | |
642 | * If we got an error, the transaction is already | |
643 | * cancelled, and the inode is unlocked. Just get out. | |
644 | */ | |
645 | if (error) | |
014c2544 | 646 | return VN_INACTIVE_CACHE; |
1da177e4 LT |
647 | } else if (ip->i_afp) { |
648 | xfs_idestroy_fork(ip, XFS_ATTR_FORK); | |
649 | } | |
650 | ||
651 | /* | |
652 | * Free the inode. | |
653 | */ | |
9d87c319 | 654 | xfs_bmap_init(&free_list, &first_block); |
1da177e4 LT |
655 | error = xfs_ifree(tp, ip, &free_list); |
656 | if (error) { | |
657 | /* | |
658 | * If we fail to free the inode, shut down. The cancel | |
659 | * might do that, we need to make sure. Otherwise the | |
660 | * inode might be lost for a long time or forever. | |
661 | */ | |
662 | if (!XFS_FORCED_SHUTDOWN(mp)) { | |
0b932ccc DC |
663 | xfs_notice(mp, "%s: xfs_ifree returned error %d", |
664 | __func__, error); | |
7d04a335 | 665 | xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR); |
1da177e4 LT |
666 | } |
667 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT); | |
668 | } else { | |
669 | /* | |
670 | * Credit the quota account(s). The inode is gone. | |
671 | */ | |
7d095257 | 672 | xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1); |
1da177e4 LT |
673 | |
674 | /* | |
78e9da77 DC |
675 | * Just ignore errors at this point. There is nothing we can |
676 | * do except to try to keep going. Make sure it's not a silent | |
677 | * error. | |
1da177e4 | 678 | */ |
78e9da77 DC |
679 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
680 | if (error) | |
53487786 DC |
681 | xfs_notice(mp, "%s: xfs_bmap_finish returned error %d", |
682 | __func__, error); | |
78e9da77 DC |
683 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
684 | if (error) | |
53487786 DC |
685 | xfs_notice(mp, "%s: xfs_trans_commit returned error %d", |
686 | __func__, error); | |
1da177e4 | 687 | } |
7d095257 | 688 | |
1da177e4 LT |
689 | /* |
690 | * Release the dquots held by inode, if any. | |
691 | */ | |
7d095257 | 692 | xfs_qm_dqdetach(ip); |
1da177e4 LT |
693 | xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); |
694 | ||
b373e98d CH |
695 | out: |
696 | return VN_INACTIVE_CACHE; | |
697 | out_cancel: | |
698 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT); | |
699 | xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
1da177e4 LT |
700 | return VN_INACTIVE_CACHE; |
701 | } | |
702 | ||
384f3ced BN |
703 | /* |
704 | * Lookups up an inode from "name". If ci_name is not NULL, then a CI match | |
705 | * is allowed, otherwise it has to be an exact match. If a CI match is found, | |
706 | * ci_name->name will point to a the actual name (caller must free) or | |
707 | * will be set to NULL if an exact match is found. | |
708 | */ | |
993386c1 | 709 | int |
1da177e4 | 710 | xfs_lookup( |
993386c1 | 711 | xfs_inode_t *dp, |
556b8b16 | 712 | struct xfs_name *name, |
384f3ced BN |
713 | xfs_inode_t **ipp, |
714 | struct xfs_name *ci_name) | |
1da177e4 | 715 | { |
eca450b7 | 716 | xfs_ino_t inum; |
1da177e4 LT |
717 | int error; |
718 | uint lock_mode; | |
1da177e4 | 719 | |
cca28fb8 | 720 | trace_xfs_lookup(dp, name); |
1da177e4 LT |
721 | |
722 | if (XFS_FORCED_SHUTDOWN(dp->i_mount)) | |
723 | return XFS_ERROR(EIO); | |
724 | ||
725 | lock_mode = xfs_ilock_map_shared(dp); | |
384f3ced | 726 | error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name); |
1da177e4 | 727 | xfs_iunlock_map_shared(dp, lock_mode); |
eca450b7 CH |
728 | |
729 | if (error) | |
730 | goto out; | |
731 | ||
7b6259e7 | 732 | error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp); |
eca450b7 | 733 | if (error) |
384f3ced | 734 | goto out_free_name; |
eca450b7 | 735 | |
eca450b7 CH |
736 | return 0; |
737 | ||
384f3ced BN |
738 | out_free_name: |
739 | if (ci_name) | |
740 | kmem_free(ci_name->name); | |
741 | out: | |
eca450b7 | 742 | *ipp = NULL; |
1da177e4 LT |
743 | return error; |
744 | } | |
745 | ||
993386c1 | 746 | int |
1da177e4 | 747 | xfs_create( |
993386c1 | 748 | xfs_inode_t *dp, |
556b8b16 | 749 | struct xfs_name *name, |
576b1d67 | 750 | umode_t mode, |
3e5daf05 | 751 | xfs_dev_t rdev, |
6c77b0ea | 752 | xfs_inode_t **ipp) |
1da177e4 | 753 | { |
517b5e8c CH |
754 | int is_dir = S_ISDIR(mode); |
755 | struct xfs_mount *mp = dp->i_mount; | |
756 | struct xfs_inode *ip = NULL; | |
757 | struct xfs_trans *tp = NULL; | |
556b8b16 | 758 | int error; |
1da177e4 LT |
759 | xfs_bmap_free_t free_list; |
760 | xfs_fsblock_t first_block; | |
993386c1 | 761 | boolean_t unlock_dp_on_error = B_FALSE; |
1da177e4 LT |
762 | uint cancel_flags; |
763 | int committed; | |
6743099c | 764 | prid_t prid; |
517b5e8c CH |
765 | struct xfs_dquot *udqp = NULL; |
766 | struct xfs_dquot *gdqp = NULL; | |
1da177e4 | 767 | uint resblks; |
517b5e8c CH |
768 | uint log_res; |
769 | uint log_count; | |
1da177e4 | 770 | |
cca28fb8 | 771 | trace_xfs_create(dp, name); |
1da177e4 | 772 | |
517b5e8c CH |
773 | if (XFS_FORCED_SHUTDOWN(mp)) |
774 | return XFS_ERROR(EIO); | |
775 | ||
365ca83d | 776 | if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) |
6743099c | 777 | prid = xfs_get_projid(dp); |
1da177e4 | 778 | else |
6743099c | 779 | prid = XFS_PROJID_DEFAULT; |
1da177e4 LT |
780 | |
781 | /* | |
782 | * Make sure that we have allocated dquot(s) on disk. | |
783 | */ | |
7d095257 | 784 | error = xfs_qm_vop_dqalloc(dp, current_fsuid(), current_fsgid(), prid, |
517b5e8c | 785 | XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp); |
1da177e4 | 786 | if (error) |
ec3ba85f | 787 | return error; |
1da177e4 | 788 | |
517b5e8c CH |
789 | if (is_dir) { |
790 | rdev = 0; | |
791 | resblks = XFS_MKDIR_SPACE_RES(mp, name->len); | |
792 | log_res = XFS_MKDIR_LOG_RES(mp); | |
793 | log_count = XFS_MKDIR_LOG_COUNT; | |
794 | tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR); | |
795 | } else { | |
796 | resblks = XFS_CREATE_SPACE_RES(mp, name->len); | |
797 | log_res = XFS_CREATE_LOG_RES(mp); | |
798 | log_count = XFS_CREATE_LOG_COUNT; | |
799 | tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE); | |
800 | } | |
1da177e4 | 801 | |
1da177e4 | 802 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; |
517b5e8c | 803 | |
1da177e4 LT |
804 | /* |
805 | * Initially assume that the file does not exist and | |
806 | * reserve the resources for that case. If that is not | |
807 | * the case we'll drop the one we have and get a more | |
808 | * appropriate transaction later. | |
809 | */ | |
517b5e8c CH |
810 | error = xfs_trans_reserve(tp, resblks, log_res, 0, |
811 | XFS_TRANS_PERM_LOG_RES, log_count); | |
1da177e4 | 812 | if (error == ENOSPC) { |
153fec43 DC |
813 | /* flush outstanding delalloc blocks and retry */ |
814 | xfs_flush_inodes(dp); | |
4734d401 CH |
815 | error = xfs_trans_reserve(tp, resblks, log_res, 0, |
816 | XFS_TRANS_PERM_LOG_RES, log_count); | |
153fec43 DC |
817 | } |
818 | if (error == ENOSPC) { | |
819 | /* No space at all so try a "no-allocation" reservation */ | |
1da177e4 | 820 | resblks = 0; |
517b5e8c CH |
821 | error = xfs_trans_reserve(tp, 0, log_res, 0, |
822 | XFS_TRANS_PERM_LOG_RES, log_count); | |
1da177e4 LT |
823 | } |
824 | if (error) { | |
825 | cancel_flags = 0; | |
517b5e8c | 826 | goto out_trans_cancel; |
1da177e4 LT |
827 | } |
828 | ||
f7c66ce3 | 829 | xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT); |
993386c1 | 830 | unlock_dp_on_error = B_TRUE; |
1da177e4 | 831 | |
517b5e8c | 832 | xfs_bmap_init(&free_list, &first_block); |
1da177e4 LT |
833 | |
834 | /* | |
835 | * Reserve disk quota and the inode. | |
836 | */ | |
7d095257 | 837 | error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp, resblks, 1, 0); |
1da177e4 | 838 | if (error) |
517b5e8c | 839 | goto out_trans_cancel; |
1da177e4 | 840 | |
556b8b16 BN |
841 | error = xfs_dir_canenter(tp, dp, name, resblks); |
842 | if (error) | |
517b5e8c CH |
843 | goto out_trans_cancel; |
844 | ||
845 | /* | |
846 | * A newly created regular or special file just has one directory | |
847 | * entry pointing to them, but a directory also the "." entry | |
848 | * pointing to itself. | |
849 | */ | |
6c77b0ea | 850 | error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev, |
517b5e8c | 851 | prid, resblks > 0, &ip, &committed); |
1da177e4 LT |
852 | if (error) { |
853 | if (error == ENOSPC) | |
517b5e8c CH |
854 | goto out_trans_cancel; |
855 | goto out_trans_abort; | |
1da177e4 | 856 | } |
1da177e4 | 857 | |
1da177e4 | 858 | /* |
993386c1 CH |
859 | * Now we join the directory inode to the transaction. We do not do it |
860 | * earlier because xfs_dir_ialloc might commit the previous transaction | |
861 | * (and release all the locks). An error from here on will result in | |
862 | * the transaction cancel unlocking dp so don't do it explicitly in the | |
863 | * error path. | |
1da177e4 | 864 | */ |
ddc3415a | 865 | xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); |
993386c1 | 866 | unlock_dp_on_error = B_FALSE; |
1da177e4 | 867 | |
556b8b16 | 868 | error = xfs_dir_createname(tp, dp, name, ip->i_ino, |
f6c2d1fa NS |
869 | &first_block, &free_list, resblks ? |
870 | resblks - XFS_IALLOC_SPACE_RES(mp) : 0); | |
1da177e4 LT |
871 | if (error) { |
872 | ASSERT(error != ENOSPC); | |
517b5e8c | 873 | goto out_trans_abort; |
1da177e4 | 874 | } |
dcd79a14 | 875 | xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); |
1da177e4 LT |
876 | xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); |
877 | ||
517b5e8c CH |
878 | if (is_dir) { |
879 | error = xfs_dir_init(tp, ip, dp); | |
880 | if (error) | |
881 | goto out_bmap_cancel; | |
882 | ||
883 | error = xfs_bumplink(tp, dp); | |
884 | if (error) | |
885 | goto out_bmap_cancel; | |
886 | } | |
887 | ||
1da177e4 LT |
888 | /* |
889 | * If this is a synchronous mount, make sure that the | |
890 | * create transaction goes to disk before returning to | |
891 | * the user. | |
892 | */ | |
517b5e8c | 893 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) |
1da177e4 | 894 | xfs_trans_set_sync(tp); |
1da177e4 | 895 | |
1da177e4 LT |
896 | /* |
897 | * Attach the dquot(s) to the inodes and modify them incore. | |
898 | * These ids of the inode couldn't have changed since the new | |
899 | * inode has been locked ever since it was created. | |
900 | */ | |
7d095257 | 901 | xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp); |
1da177e4 | 902 | |
f7c99b6f | 903 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
517b5e8c | 904 | if (error) |
ec3ba85f | 905 | goto out_bmap_cancel; |
1da177e4 | 906 | |
1c72bf90 | 907 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
ec3ba85f CH |
908 | if (error) |
909 | goto out_release_inode; | |
1da177e4 | 910 | |
7d095257 CH |
911 | xfs_qm_dqrele(udqp); |
912 | xfs_qm_dqrele(gdqp); | |
1da177e4 | 913 | |
979ebab1 | 914 | *ipp = ip; |
288699fe | 915 | return 0; |
1da177e4 | 916 | |
517b5e8c CH |
917 | out_bmap_cancel: |
918 | xfs_bmap_cancel(&free_list); | |
919 | out_trans_abort: | |
1da177e4 | 920 | cancel_flags |= XFS_TRANS_ABORT; |
517b5e8c CH |
921 | out_trans_cancel: |
922 | xfs_trans_cancel(tp, cancel_flags); | |
ec3ba85f CH |
923 | out_release_inode: |
924 | /* | |
925 | * Wait until after the current transaction is aborted to | |
926 | * release the inode. This prevents recursive transactions | |
927 | * and deadlocks from xfs_inactive. | |
928 | */ | |
929 | if (ip) | |
930 | IRELE(ip); | |
931 | ||
7d095257 CH |
932 | xfs_qm_dqrele(udqp); |
933 | xfs_qm_dqrele(gdqp); | |
1da177e4 | 934 | |
993386c1 CH |
935 | if (unlock_dp_on_error) |
936 | xfs_iunlock(dp, XFS_ILOCK_EXCL); | |
288699fe | 937 | return error; |
1da177e4 LT |
938 | } |
939 | ||
1da177e4 LT |
940 | #ifdef DEBUG |
941 | int xfs_locked_n; | |
942 | int xfs_small_retries; | |
943 | int xfs_middle_retries; | |
944 | int xfs_lots_retries; | |
945 | int xfs_lock_delays; | |
946 | #endif | |
947 | ||
f7c66ce3 LM |
948 | /* |
949 | * Bump the subclass so xfs_lock_inodes() acquires each lock with | |
950 | * a different value | |
951 | */ | |
952 | static inline int | |
953 | xfs_lock_inumorder(int lock_mode, int subclass) | |
954 | { | |
955 | if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) | |
0f1145cc | 956 | lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_IOLOCK_SHIFT; |
f7c66ce3 | 957 | if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) |
0f1145cc | 958 | lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_ILOCK_SHIFT; |
f7c66ce3 LM |
959 | |
960 | return lock_mode; | |
961 | } | |
962 | ||
1da177e4 LT |
963 | /* |
964 | * The following routine will lock n inodes in exclusive mode. | |
965 | * We assume the caller calls us with the inodes in i_ino order. | |
966 | * | |
967 | * We need to detect deadlock where an inode that we lock | |
968 | * is in the AIL and we start waiting for another inode that is locked | |
969 | * by a thread in a long running transaction (such as truncate). This can | |
970 | * result in deadlock since the long running trans might need to wait | |
971 | * for the inode we just locked in order to push the tail and free space | |
972 | * in the log. | |
973 | */ | |
974 | void | |
975 | xfs_lock_inodes( | |
976 | xfs_inode_t **ips, | |
977 | int inodes, | |
1da177e4 LT |
978 | uint lock_mode) |
979 | { | |
980 | int attempts = 0, i, j, try_lock; | |
981 | xfs_log_item_t *lp; | |
982 | ||
983 | ASSERT(ips && (inodes >= 2)); /* we need at least two */ | |
984 | ||
cfa853e4 CH |
985 | try_lock = 0; |
986 | i = 0; | |
1da177e4 LT |
987 | |
988 | again: | |
989 | for (; i < inodes; i++) { | |
990 | ASSERT(ips[i]); | |
991 | ||
992 | if (i && (ips[i] == ips[i-1])) /* Already locked */ | |
993 | continue; | |
994 | ||
995 | /* | |
996 | * If try_lock is not set yet, make sure all locked inodes | |
997 | * are not in the AIL. | |
998 | * If any are, set try_lock to be used later. | |
999 | */ | |
1000 | ||
1001 | if (!try_lock) { | |
1002 | for (j = (i - 1); j >= 0 && !try_lock; j--) { | |
1003 | lp = (xfs_log_item_t *)ips[j]->i_itemp; | |
1004 | if (lp && (lp->li_flags & XFS_LI_IN_AIL)) { | |
1005 | try_lock++; | |
1006 | } | |
1007 | } | |
1008 | } | |
1009 | ||
1010 | /* | |
1011 | * If any of the previous locks we have locked is in the AIL, | |
1012 | * we must TRY to get the second and subsequent locks. If | |
1013 | * we can't get any, we must release all we have | |
1014 | * and try again. | |
1015 | */ | |
1016 | ||
1017 | if (try_lock) { | |
1018 | /* try_lock must be 0 if i is 0. */ | |
1019 | /* | |
1020 | * try_lock means we have an inode locked | |
1021 | * that is in the AIL. | |
1022 | */ | |
1023 | ASSERT(i != 0); | |
f7c66ce3 | 1024 | if (!xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i))) { |
1da177e4 LT |
1025 | attempts++; |
1026 | ||
1027 | /* | |
1028 | * Unlock all previous guys and try again. | |
1029 | * xfs_iunlock will try to push the tail | |
1030 | * if the inode is in the AIL. | |
1031 | */ | |
1032 | ||
1033 | for(j = i - 1; j >= 0; j--) { | |
1034 | ||
1035 | /* | |
1036 | * Check to see if we've already | |
1037 | * unlocked this one. | |
1038 | * Not the first one going back, | |
1039 | * and the inode ptr is the same. | |
1040 | */ | |
1041 | if ((j != (i - 1)) && ips[j] == | |
1042 | ips[j+1]) | |
1043 | continue; | |
1044 | ||
1045 | xfs_iunlock(ips[j], lock_mode); | |
1046 | } | |
1047 | ||
1048 | if ((attempts % 5) == 0) { | |
1049 | delay(1); /* Don't just spin the CPU */ | |
1050 | #ifdef DEBUG | |
1051 | xfs_lock_delays++; | |
1052 | #endif | |
1053 | } | |
1054 | i = 0; | |
1055 | try_lock = 0; | |
1056 | goto again; | |
1057 | } | |
1058 | } else { | |
f7c66ce3 | 1059 | xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i)); |
1da177e4 LT |
1060 | } |
1061 | } | |
1062 | ||
1063 | #ifdef DEBUG | |
1064 | if (attempts) { | |
1065 | if (attempts < 5) xfs_small_retries++; | |
1066 | else if (attempts < 100) xfs_middle_retries++; | |
1067 | else xfs_lots_retries++; | |
1068 | } else { | |
1069 | xfs_locked_n++; | |
1070 | } | |
1071 | #endif | |
1072 | } | |
1073 | ||
f9114eba DC |
1074 | /* |
1075 | * xfs_lock_two_inodes() can only be used to lock one type of lock | |
1076 | * at a time - the iolock or the ilock, but not both at once. If | |
1077 | * we lock both at once, lockdep will report false positives saying | |
1078 | * we have violated locking orders. | |
1079 | */ | |
e1cccd91 CH |
1080 | void |
1081 | xfs_lock_two_inodes( | |
1082 | xfs_inode_t *ip0, | |
1083 | xfs_inode_t *ip1, | |
1084 | uint lock_mode) | |
1085 | { | |
1086 | xfs_inode_t *temp; | |
1087 | int attempts = 0; | |
1088 | xfs_log_item_t *lp; | |
1089 | ||
f9114eba DC |
1090 | if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) |
1091 | ASSERT((lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) == 0); | |
e1cccd91 CH |
1092 | ASSERT(ip0->i_ino != ip1->i_ino); |
1093 | ||
1094 | if (ip0->i_ino > ip1->i_ino) { | |
1095 | temp = ip0; | |
1096 | ip0 = ip1; | |
1097 | ip1 = temp; | |
1098 | } | |
1099 | ||
1100 | again: | |
1101 | xfs_ilock(ip0, xfs_lock_inumorder(lock_mode, 0)); | |
1102 | ||
1103 | /* | |
1104 | * If the first lock we have locked is in the AIL, we must TRY to get | |
1105 | * the second lock. If we can't get it, we must release the first one | |
1106 | * and try again. | |
1107 | */ | |
1108 | lp = (xfs_log_item_t *)ip0->i_itemp; | |
1109 | if (lp && (lp->li_flags & XFS_LI_IN_AIL)) { | |
1110 | if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(lock_mode, 1))) { | |
1111 | xfs_iunlock(ip0, lock_mode); | |
1112 | if ((++attempts % 5) == 0) | |
1113 | delay(1); /* Don't just spin the CPU */ | |
1114 | goto again; | |
1115 | } | |
1116 | } else { | |
1117 | xfs_ilock(ip1, xfs_lock_inumorder(lock_mode, 1)); | |
1118 | } | |
1119 | } | |
1120 | ||
993386c1 | 1121 | int |
1da177e4 | 1122 | xfs_remove( |
993386c1 | 1123 | xfs_inode_t *dp, |
556b8b16 BN |
1124 | struct xfs_name *name, |
1125 | xfs_inode_t *ip) | |
1da177e4 | 1126 | { |
993386c1 | 1127 | xfs_mount_t *mp = dp->i_mount; |
1da177e4 | 1128 | xfs_trans_t *tp = NULL; |
8f112e3b | 1129 | int is_dir = S_ISDIR(ip->i_d.di_mode); |
1da177e4 LT |
1130 | int error = 0; |
1131 | xfs_bmap_free_t free_list; | |
1132 | xfs_fsblock_t first_block; | |
1133 | int cancel_flags; | |
1134 | int committed; | |
1da177e4 LT |
1135 | int link_zero; |
1136 | uint resblks; | |
8f112e3b | 1137 | uint log_count; |
1da177e4 | 1138 | |
cca28fb8 | 1139 | trace_xfs_remove(dp, name); |
1da177e4 | 1140 | |
1da177e4 LT |
1141 | if (XFS_FORCED_SHUTDOWN(mp)) |
1142 | return XFS_ERROR(EIO); | |
1143 | ||
7d095257 | 1144 | error = xfs_qm_dqattach(dp, 0); |
8f112e3b CH |
1145 | if (error) |
1146 | goto std_return; | |
1147 | ||
7d095257 | 1148 | error = xfs_qm_dqattach(ip, 0); |
8f112e3b | 1149 | if (error) |
1da177e4 | 1150 | goto std_return; |
1da177e4 | 1151 | |
8f112e3b CH |
1152 | if (is_dir) { |
1153 | tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR); | |
1154 | log_count = XFS_DEFAULT_LOG_COUNT; | |
1155 | } else { | |
1156 | tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE); | |
1157 | log_count = XFS_REMOVE_LOG_COUNT; | |
1158 | } | |
1da177e4 | 1159 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; |
8f112e3b | 1160 | |
1da177e4 LT |
1161 | /* |
1162 | * We try to get the real space reservation first, | |
1163 | * allowing for directory btree deletion(s) implying | |
1164 | * possible bmap insert(s). If we can't get the space | |
1165 | * reservation then we use 0 instead, and avoid the bmap | |
1166 | * btree insert(s) in the directory code by, if the bmap | |
1167 | * insert tries to happen, instead trimming the LAST | |
1168 | * block from the directory. | |
1169 | */ | |
1170 | resblks = XFS_REMOVE_SPACE_RES(mp); | |
1171 | error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0, | |
8f112e3b | 1172 | XFS_TRANS_PERM_LOG_RES, log_count); |
1da177e4 LT |
1173 | if (error == ENOSPC) { |
1174 | resblks = 0; | |
1175 | error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0, | |
8f112e3b | 1176 | XFS_TRANS_PERM_LOG_RES, log_count); |
1da177e4 LT |
1177 | } |
1178 | if (error) { | |
1179 | ASSERT(error != ENOSPC); | |
8f112e3b CH |
1180 | cancel_flags = 0; |
1181 | goto out_trans_cancel; | |
1da177e4 LT |
1182 | } |
1183 | ||
e1cccd91 | 1184 | xfs_lock_two_inodes(dp, ip, XFS_ILOCK_EXCL); |
1da177e4 | 1185 | |
ddc3415a CH |
1186 | xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); |
1187 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
1da177e4 | 1188 | |
8f112e3b CH |
1189 | /* |
1190 | * If we're removing a directory perform some additional validation. | |
1191 | */ | |
1192 | if (is_dir) { | |
1193 | ASSERT(ip->i_d.di_nlink >= 2); | |
1194 | if (ip->i_d.di_nlink != 2) { | |
1195 | error = XFS_ERROR(ENOTEMPTY); | |
1196 | goto out_trans_cancel; | |
1197 | } | |
1198 | if (!xfs_dir_isempty(ip)) { | |
1199 | error = XFS_ERROR(ENOTEMPTY); | |
1200 | goto out_trans_cancel; | |
1201 | } | |
1202 | } | |
1203 | ||
9d87c319 | 1204 | xfs_bmap_init(&free_list, &first_block); |
556b8b16 | 1205 | error = xfs_dir_removename(tp, dp, name, ip->i_ino, |
d4377d84 | 1206 | &first_block, &free_list, resblks); |
1da177e4 LT |
1207 | if (error) { |
1208 | ASSERT(error != ENOENT); | |
8f112e3b | 1209 | goto out_bmap_cancel; |
1da177e4 | 1210 | } |
dcd79a14 | 1211 | xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); |
1da177e4 | 1212 | |
8f112e3b CH |
1213 | if (is_dir) { |
1214 | /* | |
1215 | * Drop the link from ip's "..". | |
1216 | */ | |
1217 | error = xfs_droplink(tp, dp); | |
1218 | if (error) | |
1219 | goto out_bmap_cancel; | |
1220 | ||
1221 | /* | |
2b7035fd | 1222 | * Drop the "." link from ip to self. |
8f112e3b CH |
1223 | */ |
1224 | error = xfs_droplink(tp, ip); | |
1225 | if (error) | |
1226 | goto out_bmap_cancel; | |
1227 | } else { | |
1228 | /* | |
1229 | * When removing a non-directory we need to log the parent | |
26c52951 DC |
1230 | * inode here. For a directory this is done implicitly |
1231 | * by the xfs_droplink call for the ".." entry. | |
8f112e3b CH |
1232 | */ |
1233 | xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); | |
1da177e4 LT |
1234 | } |
1235 | ||
8f112e3b | 1236 | /* |
2b7035fd | 1237 | * Drop the link from dp to ip. |
8f112e3b CH |
1238 | */ |
1239 | error = xfs_droplink(tp, ip); | |
1240 | if (error) | |
1241 | goto out_bmap_cancel; | |
1242 | ||
1243 | /* | |
1244 | * Determine if this is the last link while | |
1da177e4 LT |
1245 | * we are in the transaction. |
1246 | */ | |
8f112e3b | 1247 | link_zero = (ip->i_d.di_nlink == 0); |
1da177e4 | 1248 | |
1da177e4 LT |
1249 | /* |
1250 | * If this is a synchronous mount, make sure that the | |
1251 | * remove transaction goes to disk before returning to | |
1252 | * the user. | |
1253 | */ | |
8f112e3b | 1254 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) |
1da177e4 | 1255 | xfs_trans_set_sync(tp); |
1da177e4 | 1256 | |
f7c99b6f | 1257 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
8f112e3b CH |
1258 | if (error) |
1259 | goto out_bmap_cancel; | |
1da177e4 | 1260 | |
1c72bf90 | 1261 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
5df78e73 | 1262 | if (error) |
1da177e4 | 1263 | goto std_return; |
1da177e4 | 1264 | |
2a82b8be DC |
1265 | /* |
1266 | * If we are using filestreams, kill the stream association. | |
1267 | * If the file is still open it may get a new one but that | |
1268 | * will get killed on last close in xfs_close() so we don't | |
1269 | * have to worry about that. | |
1270 | */ | |
8f112e3b | 1271 | if (!is_dir && link_zero && xfs_inode_is_filestream(ip)) |
2a82b8be DC |
1272 | xfs_filestream_deassociate(ip); |
1273 | ||
288699fe | 1274 | return 0; |
1da177e4 | 1275 | |
8f112e3b | 1276 | out_bmap_cancel: |
1da177e4 LT |
1277 | xfs_bmap_cancel(&free_list); |
1278 | cancel_flags |= XFS_TRANS_ABORT; | |
8f112e3b | 1279 | out_trans_cancel: |
1da177e4 | 1280 | xfs_trans_cancel(tp, cancel_flags); |
288699fe CH |
1281 | std_return: |
1282 | return error; | |
1da177e4 LT |
1283 | } |
1284 | ||
993386c1 | 1285 | int |
1da177e4 | 1286 | xfs_link( |
993386c1 | 1287 | xfs_inode_t *tdp, |
a3da7896 | 1288 | xfs_inode_t *sip, |
556b8b16 | 1289 | struct xfs_name *target_name) |
1da177e4 | 1290 | { |
993386c1 | 1291 | xfs_mount_t *mp = tdp->i_mount; |
1da177e4 | 1292 | xfs_trans_t *tp; |
1da177e4 LT |
1293 | int error; |
1294 | xfs_bmap_free_t free_list; | |
1295 | xfs_fsblock_t first_block; | |
1296 | int cancel_flags; | |
1297 | int committed; | |
1da177e4 | 1298 | int resblks; |
1da177e4 | 1299 | |
cca28fb8 | 1300 | trace_xfs_link(tdp, target_name); |
1da177e4 | 1301 | |
a3da7896 | 1302 | ASSERT(!S_ISDIR(sip->i_d.di_mode)); |
1da177e4 | 1303 | |
1da177e4 LT |
1304 | if (XFS_FORCED_SHUTDOWN(mp)) |
1305 | return XFS_ERROR(EIO); | |
1306 | ||
7d095257 | 1307 | error = xfs_qm_dqattach(sip, 0); |
cb3f35bb CH |
1308 | if (error) |
1309 | goto std_return; | |
1310 | ||
7d095257 | 1311 | error = xfs_qm_dqattach(tdp, 0); |
1da177e4 LT |
1312 | if (error) |
1313 | goto std_return; | |
1314 | ||
1315 | tp = xfs_trans_alloc(mp, XFS_TRANS_LINK); | |
1316 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; | |
556b8b16 | 1317 | resblks = XFS_LINK_SPACE_RES(mp, target_name->len); |
1da177e4 LT |
1318 | error = xfs_trans_reserve(tp, resblks, XFS_LINK_LOG_RES(mp), 0, |
1319 | XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT); | |
1320 | if (error == ENOSPC) { | |
1321 | resblks = 0; | |
1322 | error = xfs_trans_reserve(tp, 0, XFS_LINK_LOG_RES(mp), 0, | |
1323 | XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT); | |
1324 | } | |
1325 | if (error) { | |
1326 | cancel_flags = 0; | |
1327 | goto error_return; | |
1328 | } | |
1329 | ||
e1cccd91 | 1330 | xfs_lock_two_inodes(sip, tdp, XFS_ILOCK_EXCL); |
1da177e4 | 1331 | |
ddc3415a CH |
1332 | xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL); |
1333 | xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL); | |
1da177e4 | 1334 | |
365ca83d NS |
1335 | /* |
1336 | * If we are using project inheritance, we only allow hard link | |
1337 | * creation in our tree when the project IDs are the same; else | |
1338 | * the tree quota mechanism could be circumvented. | |
1339 | */ | |
1340 | if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) && | |
6743099c | 1341 | (xfs_get_projid(tdp) != xfs_get_projid(sip)))) { |
b1ecdda9 | 1342 | error = XFS_ERROR(EXDEV); |
365ca83d NS |
1343 | goto error_return; |
1344 | } | |
1345 | ||
556b8b16 BN |
1346 | error = xfs_dir_canenter(tp, tdp, target_name, resblks); |
1347 | if (error) | |
1da177e4 LT |
1348 | goto error_return; |
1349 | ||
9d87c319 | 1350 | xfs_bmap_init(&free_list, &first_block); |
1da177e4 | 1351 | |
556b8b16 BN |
1352 | error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino, |
1353 | &first_block, &free_list, resblks); | |
1da177e4 LT |
1354 | if (error) |
1355 | goto abort_return; | |
dcd79a14 | 1356 | xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); |
1da177e4 LT |
1357 | xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE); |
1358 | ||
1359 | error = xfs_bumplink(tp, sip); | |
b71d300c | 1360 | if (error) |
1da177e4 | 1361 | goto abort_return; |
1da177e4 LT |
1362 | |
1363 | /* | |
1364 | * If this is a synchronous mount, make sure that the | |
1365 | * link transaction goes to disk before returning to | |
1366 | * the user. | |
1367 | */ | |
1368 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) { | |
1369 | xfs_trans_set_sync(tp); | |
1370 | } | |
1371 | ||
f7c99b6f | 1372 | error = xfs_bmap_finish (&tp, &free_list, &committed); |
1da177e4 LT |
1373 | if (error) { |
1374 | xfs_bmap_cancel(&free_list); | |
1375 | goto abort_return; | |
1376 | } | |
1377 | ||
288699fe | 1378 | return xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 LT |
1379 | |
1380 | abort_return: | |
1381 | cancel_flags |= XFS_TRANS_ABORT; | |
1da177e4 LT |
1382 | error_return: |
1383 | xfs_trans_cancel(tp, cancel_flags); | |
288699fe CH |
1384 | std_return: |
1385 | return error; | |
1da177e4 | 1386 | } |
b71d300c | 1387 | |
993386c1 | 1388 | int |
1da177e4 | 1389 | xfs_symlink( |
993386c1 | 1390 | xfs_inode_t *dp, |
556b8b16 BN |
1391 | struct xfs_name *link_name, |
1392 | const char *target_path, | |
576b1d67 | 1393 | umode_t mode, |
6c77b0ea | 1394 | xfs_inode_t **ipp) |
1da177e4 | 1395 | { |
993386c1 | 1396 | xfs_mount_t *mp = dp->i_mount; |
1da177e4 | 1397 | xfs_trans_t *tp; |
1da177e4 LT |
1398 | xfs_inode_t *ip; |
1399 | int error; | |
1400 | int pathlen; | |
1401 | xfs_bmap_free_t free_list; | |
1402 | xfs_fsblock_t first_block; | |
993386c1 | 1403 | boolean_t unlock_dp_on_error = B_FALSE; |
1da177e4 LT |
1404 | uint cancel_flags; |
1405 | int committed; | |
1406 | xfs_fileoff_t first_fsb; | |
1407 | xfs_filblks_t fs_blocks; | |
1408 | int nmaps; | |
1409 | xfs_bmbt_irec_t mval[SYMLINK_MAPS]; | |
1410 | xfs_daddr_t d; | |
556b8b16 | 1411 | const char *cur_chunk; |
1da177e4 LT |
1412 | int byte_cnt; |
1413 | int n; | |
1414 | xfs_buf_t *bp; | |
6743099c | 1415 | prid_t prid; |
1da177e4 LT |
1416 | struct xfs_dquot *udqp, *gdqp; |
1417 | uint resblks; | |
1da177e4 | 1418 | |
3937be5b | 1419 | *ipp = NULL; |
1da177e4 LT |
1420 | error = 0; |
1421 | ip = NULL; | |
1422 | tp = NULL; | |
1423 | ||
cca28fb8 | 1424 | trace_xfs_symlink(dp, link_name); |
1da177e4 LT |
1425 | |
1426 | if (XFS_FORCED_SHUTDOWN(mp)) | |
1427 | return XFS_ERROR(EIO); | |
1428 | ||
1da177e4 LT |
1429 | /* |
1430 | * Check component lengths of the target path name. | |
1431 | */ | |
1432 | pathlen = strlen(target_path); | |
1433 | if (pathlen >= MAXPATHLEN) /* total string too long */ | |
1434 | return XFS_ERROR(ENAMETOOLONG); | |
1da177e4 | 1435 | |
1da177e4 | 1436 | udqp = gdqp = NULL; |
365ca83d | 1437 | if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) |
6743099c | 1438 | prid = xfs_get_projid(dp); |
1da177e4 | 1439 | else |
6743099c | 1440 | prid = XFS_PROJID_DEFAULT; |
1da177e4 LT |
1441 | |
1442 | /* | |
1443 | * Make sure that we have allocated dquot(s) on disk. | |
1444 | */ | |
7d095257 | 1445 | error = xfs_qm_vop_dqalloc(dp, current_fsuid(), current_fsgid(), prid, |
1da177e4 LT |
1446 | XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp); |
1447 | if (error) | |
1448 | goto std_return; | |
1449 | ||
1450 | tp = xfs_trans_alloc(mp, XFS_TRANS_SYMLINK); | |
1451 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; | |
1452 | /* | |
1453 | * The symlink will fit into the inode data fork? | |
1454 | * There can't be any attributes so we get the whole variable part. | |
1455 | */ | |
1456 | if (pathlen <= XFS_LITINO(mp)) | |
1457 | fs_blocks = 0; | |
1458 | else | |
1459 | fs_blocks = XFS_B_TO_FSB(mp, pathlen); | |
556b8b16 | 1460 | resblks = XFS_SYMLINK_SPACE_RES(mp, link_name->len, fs_blocks); |
1da177e4 LT |
1461 | error = xfs_trans_reserve(tp, resblks, XFS_SYMLINK_LOG_RES(mp), 0, |
1462 | XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT); | |
1463 | if (error == ENOSPC && fs_blocks == 0) { | |
1464 | resblks = 0; | |
1465 | error = xfs_trans_reserve(tp, 0, XFS_SYMLINK_LOG_RES(mp), 0, | |
1466 | XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT); | |
1467 | } | |
1468 | if (error) { | |
1469 | cancel_flags = 0; | |
1da177e4 LT |
1470 | goto error_return; |
1471 | } | |
1472 | ||
f7c66ce3 | 1473 | xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT); |
993386c1 | 1474 | unlock_dp_on_error = B_TRUE; |
1da177e4 LT |
1475 | |
1476 | /* | |
1477 | * Check whether the directory allows new symlinks or not. | |
1478 | */ | |
1479 | if (dp->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) { | |
1480 | error = XFS_ERROR(EPERM); | |
1481 | goto error_return; | |
1482 | } | |
1483 | ||
1484 | /* | |
1485 | * Reserve disk quota : blocks and inode. | |
1486 | */ | |
7d095257 | 1487 | error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp, resblks, 1, 0); |
1da177e4 LT |
1488 | if (error) |
1489 | goto error_return; | |
1490 | ||
1491 | /* | |
1492 | * Check for ability to enter directory entry, if no space reserved. | |
1493 | */ | |
556b8b16 BN |
1494 | error = xfs_dir_canenter(tp, dp, link_name, resblks); |
1495 | if (error) | |
1da177e4 LT |
1496 | goto error_return; |
1497 | /* | |
1498 | * Initialize the bmap freelist prior to calling either | |
1499 | * bmapi or the directory create code. | |
1500 | */ | |
9d87c319 | 1501 | xfs_bmap_init(&free_list, &first_block); |
1da177e4 LT |
1502 | |
1503 | /* | |
1504 | * Allocate an inode for the symlink. | |
1505 | */ | |
6c77b0ea CH |
1506 | error = xfs_dir_ialloc(&tp, dp, S_IFLNK | (mode & ~S_IFMT), 1, 0, |
1507 | prid, resblks > 0, &ip, NULL); | |
1da177e4 LT |
1508 | if (error) { |
1509 | if (error == ENOSPC) | |
1510 | goto error_return; | |
1511 | goto error1; | |
1512 | } | |
1da177e4 | 1513 | |
993386c1 CH |
1514 | /* |
1515 | * An error after we've joined dp to the transaction will result in the | |
1516 | * transaction cancel unlocking dp so don't do it explicitly in the | |
1517 | * error path. | |
1518 | */ | |
ddc3415a | 1519 | xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); |
993386c1 | 1520 | unlock_dp_on_error = B_FALSE; |
1da177e4 LT |
1521 | |
1522 | /* | |
1523 | * Also attach the dquot(s) to it, if applicable. | |
1524 | */ | |
7d095257 | 1525 | xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp); |
1da177e4 LT |
1526 | |
1527 | if (resblks) | |
1528 | resblks -= XFS_IALLOC_SPACE_RES(mp); | |
1529 | /* | |
1530 | * If the symlink will fit into the inode, write it inline. | |
1531 | */ | |
1532 | if (pathlen <= XFS_IFORK_DSIZE(ip)) { | |
1533 | xfs_idata_realloc(ip, pathlen, XFS_DATA_FORK); | |
1534 | memcpy(ip->i_df.if_u1.if_data, target_path, pathlen); | |
1535 | ip->i_d.di_size = pathlen; | |
1536 | ||
1537 | /* | |
1538 | * The inode was initially created in extent format. | |
1539 | */ | |
1540 | ip->i_df.if_flags &= ~(XFS_IFEXTENTS | XFS_IFBROOT); | |
1541 | ip->i_df.if_flags |= XFS_IFINLINE; | |
1542 | ||
1543 | ip->i_d.di_format = XFS_DINODE_FMT_LOCAL; | |
1544 | xfs_trans_log_inode(tp, ip, XFS_ILOG_DDATA | XFS_ILOG_CORE); | |
1545 | ||
1546 | } else { | |
1547 | first_fsb = 0; | |
1548 | nmaps = SYMLINK_MAPS; | |
1549 | ||
c0dc7828 DC |
1550 | error = xfs_bmapi_write(tp, ip, first_fsb, fs_blocks, |
1551 | XFS_BMAPI_METADATA, &first_block, resblks, | |
1552 | mval, &nmaps, &free_list); | |
ec3ba85f CH |
1553 | if (error) |
1554 | goto error2; | |
1da177e4 LT |
1555 | |
1556 | if (resblks) | |
1557 | resblks -= fs_blocks; | |
1558 | ip->i_d.di_size = pathlen; | |
1559 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
1560 | ||
1561 | cur_chunk = target_path; | |
1562 | for (n = 0; n < nmaps; n++) { | |
1563 | d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock); | |
1564 | byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount); | |
1565 | bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d, | |
1566 | BTOBB(byte_cnt), 0); | |
2a30f36d CS |
1567 | if (!bp) { |
1568 | error = ENOMEM; | |
1569 | goto error2; | |
1570 | } | |
1da177e4 LT |
1571 | if (pathlen < byte_cnt) { |
1572 | byte_cnt = pathlen; | |
1573 | } | |
1574 | pathlen -= byte_cnt; | |
1575 | ||
62926044 | 1576 | memcpy(bp->b_addr, cur_chunk, byte_cnt); |
1da177e4 LT |
1577 | cur_chunk += byte_cnt; |
1578 | ||
1579 | xfs_trans_log_buf(tp, bp, 0, byte_cnt - 1); | |
1580 | } | |
1581 | } | |
1582 | ||
1583 | /* | |
1584 | * Create the directory entry for the symlink. | |
1585 | */ | |
556b8b16 BN |
1586 | error = xfs_dir_createname(tp, dp, link_name, ip->i_ino, |
1587 | &first_block, &free_list, resblks); | |
f6c2d1fa | 1588 | if (error) |
ec3ba85f | 1589 | goto error2; |
dcd79a14 | 1590 | xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); |
1da177e4 LT |
1591 | xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); |
1592 | ||
1da177e4 LT |
1593 | /* |
1594 | * If this is a synchronous mount, make sure that the | |
1595 | * symlink transaction goes to disk before returning to | |
1596 | * the user. | |
1597 | */ | |
1598 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) { | |
1599 | xfs_trans_set_sync(tp); | |
1600 | } | |
1601 | ||
f7c99b6f | 1602 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
1da177e4 LT |
1603 | if (error) { |
1604 | goto error2; | |
1605 | } | |
1c72bf90 | 1606 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
7d095257 CH |
1607 | xfs_qm_dqrele(udqp); |
1608 | xfs_qm_dqrele(gdqp); | |
1da177e4 | 1609 | |
288699fe CH |
1610 | *ipp = ip; |
1611 | return 0; | |
1da177e4 LT |
1612 | |
1613 | error2: | |
1614 | IRELE(ip); | |
1615 | error1: | |
1616 | xfs_bmap_cancel(&free_list); | |
1617 | cancel_flags |= XFS_TRANS_ABORT; | |
1618 | error_return: | |
1619 | xfs_trans_cancel(tp, cancel_flags); | |
7d095257 CH |
1620 | xfs_qm_dqrele(udqp); |
1621 | xfs_qm_dqrele(gdqp); | |
1da177e4 | 1622 | |
993386c1 | 1623 | if (unlock_dp_on_error) |
1da177e4 | 1624 | xfs_iunlock(dp, XFS_ILOCK_EXCL); |
288699fe CH |
1625 | std_return: |
1626 | return error; | |
1da177e4 LT |
1627 | } |
1628 | ||
1da177e4 | 1629 | int |
993386c1 CH |
1630 | xfs_set_dmattrs( |
1631 | xfs_inode_t *ip, | |
1da177e4 | 1632 | u_int evmask, |
993386c1 | 1633 | u_int16_t state) |
1da177e4 | 1634 | { |
993386c1 | 1635 | xfs_mount_t *mp = ip->i_mount; |
1da177e4 | 1636 | xfs_trans_t *tp; |
1da177e4 LT |
1637 | int error; |
1638 | ||
1639 | if (!capable(CAP_SYS_ADMIN)) | |
1640 | return XFS_ERROR(EPERM); | |
1641 | ||
1da177e4 LT |
1642 | if (XFS_FORCED_SHUTDOWN(mp)) |
1643 | return XFS_ERROR(EIO); | |
1644 | ||
1645 | tp = xfs_trans_alloc(mp, XFS_TRANS_SET_DMATTRS); | |
1646 | error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES (mp), 0, 0, 0); | |
1647 | if (error) { | |
1648 | xfs_trans_cancel(tp, 0); | |
1649 | return error; | |
1650 | } | |
1651 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
ddc3415a | 1652 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); |
1da177e4 | 1653 | |
613d7043 CH |
1654 | ip->i_d.di_dmevmask = evmask; |
1655 | ip->i_d.di_dmstate = state; | |
1da177e4 LT |
1656 | |
1657 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
1c72bf90 | 1658 | error = xfs_trans_commit(tp, 0); |
1da177e4 LT |
1659 | |
1660 | return error; | |
1661 | } | |
1662 | ||
1da177e4 LT |
1663 | /* |
1664 | * xfs_alloc_file_space() | |
1665 | * This routine allocates disk space for the given file. | |
1666 | * | |
1667 | * If alloc_type == 0, this request is for an ALLOCSP type | |
1668 | * request which will change the file size. In this case, no | |
1669 | * DMAPI event will be generated by the call. A TRUNCATE event | |
1670 | * will be generated later by xfs_setattr. | |
1671 | * | |
1672 | * If alloc_type != 0, this request is for a RESVSP type | |
1673 | * request, and a DMAPI DM_EVENT_WRITE will be generated if the | |
1674 | * lower block boundary byte address is less than the file's | |
1675 | * length. | |
1676 | * | |
1677 | * RETURNS: | |
1678 | * 0 on success | |
1679 | * errno on error | |
1680 | * | |
1681 | */ | |
ba0f32d4 | 1682 | STATIC int |
1da177e4 LT |
1683 | xfs_alloc_file_space( |
1684 | xfs_inode_t *ip, | |
1685 | xfs_off_t offset, | |
1686 | xfs_off_t len, | |
1687 | int alloc_type, | |
1688 | int attr_flags) | |
1689 | { | |
dd9f438e NS |
1690 | xfs_mount_t *mp = ip->i_mount; |
1691 | xfs_off_t count; | |
1da177e4 LT |
1692 | xfs_filblks_t allocated_fsb; |
1693 | xfs_filblks_t allocatesize_fsb; | |
dd9f438e NS |
1694 | xfs_extlen_t extsz, temp; |
1695 | xfs_fileoff_t startoffset_fsb; | |
1da177e4 | 1696 | xfs_fsblock_t firstfsb; |
dd9f438e | 1697 | int nimaps; |
dd9f438e | 1698 | int quota_flag; |
1da177e4 | 1699 | int rt; |
1da177e4 | 1700 | xfs_trans_t *tp; |
dd9f438e NS |
1701 | xfs_bmbt_irec_t imaps[1], *imapp; |
1702 | xfs_bmap_free_t free_list; | |
1703 | uint qblocks, resblks, resrtextents; | |
1704 | int committed; | |
1705 | int error; | |
1da177e4 | 1706 | |
cca28fb8 | 1707 | trace_xfs_alloc_file_space(ip); |
1da177e4 LT |
1708 | |
1709 | if (XFS_FORCED_SHUTDOWN(mp)) | |
1710 | return XFS_ERROR(EIO); | |
1711 | ||
7d095257 CH |
1712 | error = xfs_qm_dqattach(ip, 0); |
1713 | if (error) | |
1da177e4 LT |
1714 | return error; |
1715 | ||
1716 | if (len <= 0) | |
1717 | return XFS_ERROR(EINVAL); | |
1718 | ||
957d0ebe DC |
1719 | rt = XFS_IS_REALTIME_INODE(ip); |
1720 | extsz = xfs_get_extsz_hint(ip); | |
1721 | ||
1da177e4 | 1722 | count = len; |
1da177e4 | 1723 | imapp = &imaps[0]; |
dd9f438e | 1724 | nimaps = 1; |
1da177e4 LT |
1725 | startoffset_fsb = XFS_B_TO_FSBT(mp, offset); |
1726 | allocatesize_fsb = XFS_B_TO_FSB(mp, count); | |
1727 | ||
1da177e4 | 1728 | /* |
dd9f438e | 1729 | * Allocate file space until done or until there is an error |
1da177e4 | 1730 | */ |
1da177e4 | 1731 | while (allocatesize_fsb && !error) { |
dd9f438e NS |
1732 | xfs_fileoff_t s, e; |
1733 | ||
1da177e4 | 1734 | /* |
3ddb8fa9 | 1735 | * Determine space reservations for data/realtime. |
1da177e4 | 1736 | */ |
dd9f438e | 1737 | if (unlikely(extsz)) { |
1da177e4 | 1738 | s = startoffset_fsb; |
dd9f438e NS |
1739 | do_div(s, extsz); |
1740 | s *= extsz; | |
1741 | e = startoffset_fsb + allocatesize_fsb; | |
1742 | if ((temp = do_mod(startoffset_fsb, extsz))) | |
1743 | e += temp; | |
1744 | if ((temp = do_mod(e, extsz))) | |
1745 | e += extsz - temp; | |
1746 | } else { | |
1747 | s = 0; | |
1748 | e = allocatesize_fsb; | |
1749 | } | |
1750 | ||
72656c46 DC |
1751 | /* |
1752 | * The transaction reservation is limited to a 32-bit block | |
1753 | * count, hence we need to limit the number of blocks we are | |
1754 | * trying to reserve to avoid an overflow. We can't allocate | |
1755 | * more than @nimaps extents, and an extent is limited on disk | |
1756 | * to MAXEXTLEN (21 bits), so use that to enforce the limit. | |
1757 | */ | |
1758 | resblks = min_t(xfs_fileoff_t, (e - s), (MAXEXTLEN * nimaps)); | |
dd9f438e | 1759 | if (unlikely(rt)) { |
72656c46 | 1760 | resrtextents = qblocks = resblks; |
dd9f438e NS |
1761 | resrtextents /= mp->m_sb.sb_rextsize; |
1762 | resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0); | |
1763 | quota_flag = XFS_QMOPT_RES_RTBLKS; | |
1da177e4 | 1764 | } else { |
dd9f438e | 1765 | resrtextents = 0; |
72656c46 | 1766 | resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resblks); |
dd9f438e | 1767 | quota_flag = XFS_QMOPT_RES_REGBLKS; |
1da177e4 LT |
1768 | } |
1769 | ||
1770 | /* | |
dd9f438e | 1771 | * Allocate and setup the transaction. |
1da177e4 LT |
1772 | */ |
1773 | tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT); | |
dd9f438e NS |
1774 | error = xfs_trans_reserve(tp, resblks, |
1775 | XFS_WRITE_LOG_RES(mp), resrtextents, | |
1da177e4 LT |
1776 | XFS_TRANS_PERM_LOG_RES, |
1777 | XFS_WRITE_LOG_COUNT); | |
1da177e4 | 1778 | /* |
dd9f438e | 1779 | * Check for running out of space |
1da177e4 LT |
1780 | */ |
1781 | if (error) { | |
1782 | /* | |
1783 | * Free the transaction structure. | |
1784 | */ | |
1785 | ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp)); | |
1786 | xfs_trans_cancel(tp, 0); | |
1787 | break; | |
1788 | } | |
1789 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
7d095257 CH |
1790 | error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks, |
1791 | 0, quota_flag); | |
1da177e4 LT |
1792 | if (error) |
1793 | goto error1; | |
1794 | ||
ddc3415a | 1795 | xfs_trans_ijoin(tp, ip, 0); |
1da177e4 | 1796 | |
9d87c319 | 1797 | xfs_bmap_init(&free_list, &firstfsb); |
c0dc7828 DC |
1798 | error = xfs_bmapi_write(tp, ip, startoffset_fsb, |
1799 | allocatesize_fsb, alloc_type, &firstfsb, | |
1800 | 0, imapp, &nimaps, &free_list); | |
1da177e4 LT |
1801 | if (error) { |
1802 | goto error0; | |
1803 | } | |
1804 | ||
1805 | /* | |
dd9f438e | 1806 | * Complete the transaction |
1da177e4 | 1807 | */ |
f7c99b6f | 1808 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
1da177e4 LT |
1809 | if (error) { |
1810 | goto error0; | |
1811 | } | |
1812 | ||
1c72bf90 | 1813 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 LT |
1814 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
1815 | if (error) { | |
1816 | break; | |
1817 | } | |
1818 | ||
1819 | allocated_fsb = imapp->br_blockcount; | |
1820 | ||
dd9f438e | 1821 | if (nimaps == 0) { |
1da177e4 LT |
1822 | error = XFS_ERROR(ENOSPC); |
1823 | break; | |
1824 | } | |
1825 | ||
1826 | startoffset_fsb += allocated_fsb; | |
1827 | allocatesize_fsb -= allocated_fsb; | |
1828 | } | |
1da177e4 LT |
1829 | |
1830 | return error; | |
1831 | ||
dd9f438e | 1832 | error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */ |
1da177e4 | 1833 | xfs_bmap_cancel(&free_list); |
ea562ed6 | 1834 | xfs_trans_unreserve_quota_nblks(tp, ip, (long)qblocks, 0, quota_flag); |
dd9f438e NS |
1835 | |
1836 | error1: /* Just cancel transaction */ | |
1da177e4 LT |
1837 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT); |
1838 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
288699fe | 1839 | return error; |
1da177e4 LT |
1840 | } |
1841 | ||
1842 | /* | |
1843 | * Zero file bytes between startoff and endoff inclusive. | |
1844 | * The iolock is held exclusive and no blocks are buffered. | |
2fd6f6ec LM |
1845 | * |
1846 | * This function is used by xfs_free_file_space() to zero | |
1847 | * partial blocks when the range to free is not block aligned. | |
1848 | * When unreserving space with boundaries that are not block | |
1849 | * aligned we round up the start and round down the end | |
1850 | * boundaries and then use this function to zero the parts of | |
1851 | * the blocks that got dropped during the rounding. | |
1da177e4 LT |
1852 | */ |
1853 | STATIC int | |
1854 | xfs_zero_remaining_bytes( | |
1855 | xfs_inode_t *ip, | |
1856 | xfs_off_t startoff, | |
1857 | xfs_off_t endoff) | |
1858 | { | |
1859 | xfs_bmbt_irec_t imap; | |
1860 | xfs_fileoff_t offset_fsb; | |
1861 | xfs_off_t lastoffset; | |
1862 | xfs_off_t offset; | |
1863 | xfs_buf_t *bp; | |
1864 | xfs_mount_t *mp = ip->i_mount; | |
1865 | int nimap; | |
1866 | int error = 0; | |
1867 | ||
2fd6f6ec LM |
1868 | /* |
1869 | * Avoid doing I/O beyond eof - it's not necessary | |
1870 | * since nothing can read beyond eof. The space will | |
1871 | * be zeroed when the file is extended anyway. | |
1872 | */ | |
ce7ae151 | 1873 | if (startoff >= XFS_ISIZE(ip)) |
2fd6f6ec LM |
1874 | return 0; |
1875 | ||
ce7ae151 CH |
1876 | if (endoff > XFS_ISIZE(ip)) |
1877 | endoff = XFS_ISIZE(ip); | |
2fd6f6ec | 1878 | |
686865f7 DC |
1879 | bp = xfs_buf_get_uncached(XFS_IS_REALTIME_INODE(ip) ? |
1880 | mp->m_rtdev_targp : mp->m_ddev_targp, | |
aa5c158e | 1881 | BTOBB(mp->m_sb.sb_blocksize), 0); |
c6422617 LM |
1882 | if (!bp) |
1883 | return XFS_ERROR(ENOMEM); | |
1da177e4 | 1884 | |
c8da0faf CH |
1885 | xfs_buf_unlock(bp); |
1886 | ||
1da177e4 LT |
1887 | for (offset = startoff; offset <= endoff; offset = lastoffset + 1) { |
1888 | offset_fsb = XFS_B_TO_FSBT(mp, offset); | |
1889 | nimap = 1; | |
5c8ed202 | 1890 | error = xfs_bmapi_read(ip, offset_fsb, 1, &imap, &nimap, 0); |
1da177e4 LT |
1891 | if (error || nimap < 1) |
1892 | break; | |
1893 | ASSERT(imap.br_blockcount >= 1); | |
1894 | ASSERT(imap.br_startoff == offset_fsb); | |
1895 | lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1; | |
1896 | if (lastoffset > endoff) | |
1897 | lastoffset = endoff; | |
1898 | if (imap.br_startblock == HOLESTARTBLOCK) | |
1899 | continue; | |
1900 | ASSERT(imap.br_startblock != DELAYSTARTBLOCK); | |
1901 | if (imap.br_state == XFS_EXT_UNWRITTEN) | |
1902 | continue; | |
1903 | XFS_BUF_UNDONE(bp); | |
1904 | XFS_BUF_UNWRITE(bp); | |
1905 | XFS_BUF_READ(bp); | |
9d87c319 | 1906 | XFS_BUF_SET_ADDR(bp, xfs_fsb_to_db(ip, imap.br_startblock)); |
1da177e4 | 1907 | xfsbdstrat(mp, bp); |
1a1a3e97 | 1908 | error = xfs_buf_iowait(bp); |
d64e31a2 | 1909 | if (error) { |
901796af CH |
1910 | xfs_buf_ioerror_alert(bp, |
1911 | "xfs_zero_remaining_bytes(read)"); | |
1da177e4 LT |
1912 | break; |
1913 | } | |
62926044 | 1914 | memset(bp->b_addr + |
1da177e4 LT |
1915 | (offset - XFS_FSB_TO_B(mp, imap.br_startoff)), |
1916 | 0, lastoffset - offset + 1); | |
1917 | XFS_BUF_UNDONE(bp); | |
1918 | XFS_BUF_UNREAD(bp); | |
1919 | XFS_BUF_WRITE(bp); | |
1920 | xfsbdstrat(mp, bp); | |
1a1a3e97 | 1921 | error = xfs_buf_iowait(bp); |
d64e31a2 | 1922 | if (error) { |
901796af CH |
1923 | xfs_buf_ioerror_alert(bp, |
1924 | "xfs_zero_remaining_bytes(write)"); | |
1da177e4 LT |
1925 | break; |
1926 | } | |
1927 | } | |
1928 | xfs_buf_free(bp); | |
1929 | return error; | |
1930 | } | |
1931 | ||
1932 | /* | |
1933 | * xfs_free_file_space() | |
1934 | * This routine frees disk space for the given file. | |
1935 | * | |
1936 | * This routine is only called by xfs_change_file_space | |
1937 | * for an UNRESVSP type call. | |
1938 | * | |
1939 | * RETURNS: | |
1940 | * 0 on success | |
1941 | * errno on error | |
1942 | * | |
1943 | */ | |
1944 | STATIC int | |
1945 | xfs_free_file_space( | |
1946 | xfs_inode_t *ip, | |
1947 | xfs_off_t offset, | |
1948 | xfs_off_t len, | |
1949 | int attr_flags) | |
1950 | { | |
1951 | int committed; | |
1952 | int done; | |
1da177e4 LT |
1953 | xfs_fileoff_t endoffset_fsb; |
1954 | int error; | |
1955 | xfs_fsblock_t firstfsb; | |
1956 | xfs_bmap_free_t free_list; | |
1da177e4 LT |
1957 | xfs_bmbt_irec_t imap; |
1958 | xfs_off_t ioffset; | |
1959 | xfs_extlen_t mod=0; | |
1960 | xfs_mount_t *mp; | |
1961 | int nimap; | |
1962 | uint resblks; | |
673cdf5c | 1963 | uint rounding; |
1da177e4 LT |
1964 | int rt; |
1965 | xfs_fileoff_t startoffset_fsb; | |
1966 | xfs_trans_t *tp; | |
5fcbab35 | 1967 | int need_iolock = 1; |
1da177e4 | 1968 | |
1da177e4 LT |
1969 | mp = ip->i_mount; |
1970 | ||
cca28fb8 | 1971 | trace_xfs_free_file_space(ip); |
bd5a876a | 1972 | |
7d095257 CH |
1973 | error = xfs_qm_dqattach(ip, 0); |
1974 | if (error) | |
1da177e4 LT |
1975 | return error; |
1976 | ||
1977 | error = 0; | |
1978 | if (len <= 0) /* if nothing being freed */ | |
1979 | return error; | |
71ddabb9 | 1980 | rt = XFS_IS_REALTIME_INODE(ip); |
1da177e4 | 1981 | startoffset_fsb = XFS_B_TO_FSB(mp, offset); |
288699fe | 1982 | endoffset_fsb = XFS_B_TO_FSBT(mp, offset + len); |
1da177e4 | 1983 | |
0f285c8a | 1984 | if (attr_flags & XFS_ATTR_NOLOCK) |
5fcbab35 | 1985 | need_iolock = 0; |
9fa8046f | 1986 | if (need_iolock) { |
1da177e4 | 1987 | xfs_ilock(ip, XFS_IOLOCK_EXCL); |
25e41b3d | 1988 | /* wait for the completion of any pending DIOs */ |
4a06fd26 | 1989 | inode_dio_wait(VFS_I(ip)); |
9fa8046f | 1990 | } |
5fcbab35 | 1991 | |
e6a4b37f | 1992 | rounding = max_t(uint, 1 << mp->m_sb.sb_blocklog, PAGE_CACHE_SIZE); |
1da177e4 | 1993 | ioffset = offset & ~(rounding - 1); |
bd5a876a | 1994 | |
df80c933 | 1995 | if (VN_CACHED(VFS_I(ip)) != 0) { |
e6a4b37f | 1996 | error = xfs_flushinval_pages(ip, ioffset, -1, FI_REMAPF_LOCKED); |
d3cf2094 LM |
1997 | if (error) |
1998 | goto out_unlock_iolock; | |
bd5a876a CH |
1999 | } |
2000 | ||
1da177e4 LT |
2001 | /* |
2002 | * Need to zero the stuff we're not freeing, on disk. | |
9da096fd | 2003 | * If it's a realtime file & can't use unwritten extents then we |
1da177e4 LT |
2004 | * actually need to zero the extent edges. Otherwise xfs_bunmapi |
2005 | * will take care of it for us. | |
2006 | */ | |
62118709 | 2007 | if (rt && !xfs_sb_version_hasextflgbit(&mp->m_sb)) { |
1da177e4 | 2008 | nimap = 1; |
5c8ed202 DC |
2009 | error = xfs_bmapi_read(ip, startoffset_fsb, 1, |
2010 | &imap, &nimap, 0); | |
1da177e4 LT |
2011 | if (error) |
2012 | goto out_unlock_iolock; | |
2013 | ASSERT(nimap == 0 || nimap == 1); | |
2014 | if (nimap && imap.br_startblock != HOLESTARTBLOCK) { | |
2015 | xfs_daddr_t block; | |
2016 | ||
2017 | ASSERT(imap.br_startblock != DELAYSTARTBLOCK); | |
2018 | block = imap.br_startblock; | |
2019 | mod = do_div(block, mp->m_sb.sb_rextsize); | |
2020 | if (mod) | |
2021 | startoffset_fsb += mp->m_sb.sb_rextsize - mod; | |
2022 | } | |
2023 | nimap = 1; | |
5c8ed202 DC |
2024 | error = xfs_bmapi_read(ip, endoffset_fsb - 1, 1, |
2025 | &imap, &nimap, 0); | |
1da177e4 LT |
2026 | if (error) |
2027 | goto out_unlock_iolock; | |
2028 | ASSERT(nimap == 0 || nimap == 1); | |
2029 | if (nimap && imap.br_startblock != HOLESTARTBLOCK) { | |
2030 | ASSERT(imap.br_startblock != DELAYSTARTBLOCK); | |
2031 | mod++; | |
2032 | if (mod && (mod != mp->m_sb.sb_rextsize)) | |
2033 | endoffset_fsb -= mod; | |
2034 | } | |
2035 | } | |
2036 | if ((done = (endoffset_fsb <= startoffset_fsb))) | |
2037 | /* | |
2038 | * One contiguous piece to clear | |
2039 | */ | |
2040 | error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1); | |
2041 | else { | |
2042 | /* | |
2043 | * Some full blocks, possibly two pieces to clear | |
2044 | */ | |
2045 | if (offset < XFS_FSB_TO_B(mp, startoffset_fsb)) | |
2046 | error = xfs_zero_remaining_bytes(ip, offset, | |
2047 | XFS_FSB_TO_B(mp, startoffset_fsb) - 1); | |
2048 | if (!error && | |
2049 | XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len) | |
2050 | error = xfs_zero_remaining_bytes(ip, | |
2051 | XFS_FSB_TO_B(mp, endoffset_fsb), | |
2052 | offset + len - 1); | |
2053 | } | |
2054 | ||
2055 | /* | |
2056 | * free file space until done or until there is an error | |
2057 | */ | |
2058 | resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0); | |
2059 | while (!error && !done) { | |
2060 | ||
2061 | /* | |
91ebecc7 DC |
2062 | * allocate and setup the transaction. Allow this |
2063 | * transaction to dip into the reserve blocks to ensure | |
2064 | * the freeing of the space succeeds at ENOSPC. | |
1da177e4 LT |
2065 | */ |
2066 | tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT); | |
91ebecc7 | 2067 | tp->t_flags |= XFS_TRANS_RESERVE; |
1da177e4 LT |
2068 | error = xfs_trans_reserve(tp, |
2069 | resblks, | |
2070 | XFS_WRITE_LOG_RES(mp), | |
2071 | 0, | |
2072 | XFS_TRANS_PERM_LOG_RES, | |
2073 | XFS_WRITE_LOG_COUNT); | |
2074 | ||
2075 | /* | |
2076 | * check for running out of space | |
2077 | */ | |
2078 | if (error) { | |
2079 | /* | |
2080 | * Free the transaction structure. | |
2081 | */ | |
2082 | ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp)); | |
2083 | xfs_trans_cancel(tp, 0); | |
2084 | break; | |
2085 | } | |
2086 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
7d095257 CH |
2087 | error = xfs_trans_reserve_quota(tp, mp, |
2088 | ip->i_udquot, ip->i_gdquot, | |
2089 | resblks, 0, XFS_QMOPT_RES_REGBLKS); | |
1da177e4 LT |
2090 | if (error) |
2091 | goto error1; | |
2092 | ||
ddc3415a | 2093 | xfs_trans_ijoin(tp, ip, 0); |
1da177e4 LT |
2094 | |
2095 | /* | |
2096 | * issue the bunmapi() call to free the blocks | |
2097 | */ | |
9d87c319 | 2098 | xfs_bmap_init(&free_list, &firstfsb); |
541d7d3c | 2099 | error = xfs_bunmapi(tp, ip, startoffset_fsb, |
1da177e4 | 2100 | endoffset_fsb - startoffset_fsb, |
b4e9181e | 2101 | 0, 2, &firstfsb, &free_list, &done); |
1da177e4 LT |
2102 | if (error) { |
2103 | goto error0; | |
2104 | } | |
2105 | ||
2106 | /* | |
2107 | * complete the transaction | |
2108 | */ | |
f7c99b6f | 2109 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
1da177e4 LT |
2110 | if (error) { |
2111 | goto error0; | |
2112 | } | |
2113 | ||
1c72bf90 | 2114 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 LT |
2115 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
2116 | } | |
2117 | ||
2118 | out_unlock_iolock: | |
2119 | if (need_iolock) | |
2120 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
2121 | return error; | |
2122 | ||
2123 | error0: | |
2124 | xfs_bmap_cancel(&free_list); | |
2125 | error1: | |
2126 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT); | |
2127 | xfs_iunlock(ip, need_iolock ? (XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL) : | |
2128 | XFS_ILOCK_EXCL); | |
2129 | return error; | |
2130 | } | |
2131 | ||
2132 | /* | |
2133 | * xfs_change_file_space() | |
2134 | * This routine allocates or frees disk space for the given file. | |
2135 | * The user specified parameters are checked for alignment and size | |
2136 | * limitations. | |
2137 | * | |
2138 | * RETURNS: | |
2139 | * 0 on success | |
2140 | * errno on error | |
2141 | * | |
2142 | */ | |
2143 | int | |
2144 | xfs_change_file_space( | |
993386c1 | 2145 | xfs_inode_t *ip, |
1da177e4 LT |
2146 | int cmd, |
2147 | xfs_flock64_t *bf, | |
2148 | xfs_off_t offset, | |
1da177e4 LT |
2149 | int attr_flags) |
2150 | { | |
993386c1 | 2151 | xfs_mount_t *mp = ip->i_mount; |
1da177e4 LT |
2152 | int clrprealloc; |
2153 | int error; | |
2154 | xfs_fsize_t fsize; | |
1da177e4 LT |
2155 | int setprealloc; |
2156 | xfs_off_t startoffset; | |
2157 | xfs_off_t llen; | |
2158 | xfs_trans_t *tp; | |
0f285c8a | 2159 | struct iattr iattr; |
44722352 | 2160 | int prealloc_type; |
1da177e4 | 2161 | |
993386c1 | 2162 | if (!S_ISREG(ip->i_d.di_mode)) |
1da177e4 LT |
2163 | return XFS_ERROR(EINVAL); |
2164 | ||
1da177e4 LT |
2165 | switch (bf->l_whence) { |
2166 | case 0: /*SEEK_SET*/ | |
2167 | break; | |
2168 | case 1: /*SEEK_CUR*/ | |
2169 | bf->l_start += offset; | |
2170 | break; | |
2171 | case 2: /*SEEK_END*/ | |
ce7ae151 | 2172 | bf->l_start += XFS_ISIZE(ip); |
1da177e4 LT |
2173 | break; |
2174 | default: | |
2175 | return XFS_ERROR(EINVAL); | |
2176 | } | |
2177 | ||
2178 | llen = bf->l_len > 0 ? bf->l_len - 1 : bf->l_len; | |
2179 | ||
32972383 DC |
2180 | if (bf->l_start < 0 || |
2181 | bf->l_start > mp->m_super->s_maxbytes || | |
2182 | bf->l_start + llen < 0 || | |
2183 | bf->l_start + llen > mp->m_super->s_maxbytes) | |
1da177e4 LT |
2184 | return XFS_ERROR(EINVAL); |
2185 | ||
2186 | bf->l_whence = 0; | |
2187 | ||
2188 | startoffset = bf->l_start; | |
ce7ae151 | 2189 | fsize = XFS_ISIZE(ip); |
1da177e4 LT |
2190 | |
2191 | /* | |
2192 | * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve | |
2193 | * file space. | |
2194 | * These calls do NOT zero the data space allocated to the file, | |
2195 | * nor do they change the file size. | |
2196 | * | |
2197 | * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file | |
2198 | * space. | |
2199 | * These calls cause the new file data to be zeroed and the file | |
2200 | * size to be changed. | |
2201 | */ | |
2202 | setprealloc = clrprealloc = 0; | |
44722352 | 2203 | prealloc_type = XFS_BMAPI_PREALLOC; |
1da177e4 LT |
2204 | |
2205 | switch (cmd) { | |
44722352 DC |
2206 | case XFS_IOC_ZERO_RANGE: |
2207 | prealloc_type |= XFS_BMAPI_CONVERT; | |
2208 | xfs_tosspages(ip, startoffset, startoffset + bf->l_len, 0); | |
2209 | /* FALLTHRU */ | |
1da177e4 LT |
2210 | case XFS_IOC_RESVSP: |
2211 | case XFS_IOC_RESVSP64: | |
2212 | error = xfs_alloc_file_space(ip, startoffset, bf->l_len, | |
44722352 | 2213 | prealloc_type, attr_flags); |
1da177e4 LT |
2214 | if (error) |
2215 | return error; | |
2216 | setprealloc = 1; | |
2217 | break; | |
2218 | ||
2219 | case XFS_IOC_UNRESVSP: | |
2220 | case XFS_IOC_UNRESVSP64: | |
2221 | if ((error = xfs_free_file_space(ip, startoffset, bf->l_len, | |
2222 | attr_flags))) | |
2223 | return error; | |
2224 | break; | |
2225 | ||
2226 | case XFS_IOC_ALLOCSP: | |
2227 | case XFS_IOC_ALLOCSP64: | |
2228 | case XFS_IOC_FREESP: | |
2229 | case XFS_IOC_FREESP64: | |
bc4010ec DC |
2230 | /* |
2231 | * These operations actually do IO when extending the file, but | |
2232 | * the allocation is done seperately to the zeroing that is | |
2233 | * done. This set of operations need to be serialised against | |
2234 | * other IO operations, such as truncate and buffered IO. We | |
2235 | * need to take the IOLOCK here to serialise the allocation and | |
2236 | * zeroing IO to prevent other IOLOCK holders (e.g. getbmap, | |
2237 | * truncate, direct IO) from racing against the transient | |
2238 | * allocated but not written state we can have here. | |
2239 | */ | |
2240 | xfs_ilock(ip, XFS_IOLOCK_EXCL); | |
1da177e4 LT |
2241 | if (startoffset > fsize) { |
2242 | error = xfs_alloc_file_space(ip, fsize, | |
bc4010ec DC |
2243 | startoffset - fsize, 0, |
2244 | attr_flags | XFS_ATTR_NOLOCK); | |
2245 | if (error) { | |
2246 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
1da177e4 | 2247 | break; |
bc4010ec | 2248 | } |
1da177e4 LT |
2249 | } |
2250 | ||
0f285c8a CH |
2251 | iattr.ia_valid = ATTR_SIZE; |
2252 | iattr.ia_size = startoffset; | |
1da177e4 | 2253 | |
bc4010ec DC |
2254 | error = xfs_setattr_size(ip, &iattr, |
2255 | attr_flags | XFS_ATTR_NOLOCK); | |
2256 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
1da177e4 LT |
2257 | |
2258 | if (error) | |
2259 | return error; | |
2260 | ||
2261 | clrprealloc = 1; | |
2262 | break; | |
2263 | ||
2264 | default: | |
2265 | ASSERT(0); | |
2266 | return XFS_ERROR(EINVAL); | |
2267 | } | |
2268 | ||
2269 | /* | |
2270 | * update the inode timestamp, mode, and prealloc flag bits | |
2271 | */ | |
2272 | tp = xfs_trans_alloc(mp, XFS_TRANS_WRITEID); | |
2273 | ||
2274 | if ((error = xfs_trans_reserve(tp, 0, XFS_WRITEID_LOG_RES(mp), | |
2275 | 0, 0, 0))) { | |
2276 | /* ASSERT(0); */ | |
2277 | xfs_trans_cancel(tp, 0); | |
2278 | return error; | |
2279 | } | |
2280 | ||
2281 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
ddc3415a | 2282 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); |
1da177e4 | 2283 | |
0f285c8a | 2284 | if ((attr_flags & XFS_ATTR_DMI) == 0) { |
1da177e4 LT |
2285 | ip->i_d.di_mode &= ~S_ISUID; |
2286 | ||
2287 | /* | |
2288 | * Note that we don't have to worry about mandatory | |
2289 | * file locking being disabled here because we only | |
2290 | * clear the S_ISGID bit if the Group execute bit is | |
2291 | * on, but if it was on then mandatory locking wouldn't | |
2292 | * have been enabled. | |
2293 | */ | |
2294 | if (ip->i_d.di_mode & S_IXGRP) | |
2295 | ip->i_d.di_mode &= ~S_ISGID; | |
2296 | ||
dcd79a14 | 2297 | xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); |
1da177e4 LT |
2298 | } |
2299 | if (setprealloc) | |
2300 | ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC; | |
2301 | else if (clrprealloc) | |
2302 | ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC; | |
2303 | ||
2304 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
82878897 DC |
2305 | if (attr_flags & XFS_ATTR_SYNC) |
2306 | xfs_trans_set_sync(tp); | |
23bb0be1 | 2307 | return xfs_trans_commit(tp, 0); |
1da177e4 | 2308 | } |