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" |
33479e05 | 50 | #include "xfs_icache.h" |
1da177e4 | 51 | |
7d4fb40a NS |
52 | /* |
53 | * The maximum pathlen is 1024 bytes. Since the minimum file system | |
54 | * blocksize is 512 bytes, we can get a max of 2 extents back from | |
55 | * bmapi. | |
56 | */ | |
57 | #define SYMLINK_MAPS 2 | |
58 | ||
804c83c3 CH |
59 | STATIC int |
60 | xfs_readlink_bmap( | |
61 | xfs_inode_t *ip, | |
62 | char *link) | |
63 | { | |
64 | xfs_mount_t *mp = ip->i_mount; | |
65 | int pathlen = ip->i_d.di_size; | |
66 | int nmaps = SYMLINK_MAPS; | |
67 | xfs_bmbt_irec_t mval[SYMLINK_MAPS]; | |
68 | xfs_daddr_t d; | |
69 | int byte_cnt; | |
70 | int n; | |
71 | xfs_buf_t *bp; | |
72 | int error = 0; | |
73 | ||
5c8ed202 DC |
74 | error = xfs_bmapi_read(ip, 0, XFS_B_TO_FSB(mp, pathlen), mval, &nmaps, |
75 | 0); | |
804c83c3 CH |
76 | if (error) |
77 | goto out; | |
78 | ||
79 | for (n = 0; n < nmaps; n++) { | |
80 | d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock); | |
81 | byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount); | |
82 | ||
611c9946 | 83 | bp = xfs_buf_read(mp->m_ddev_targp, d, BTOBB(byte_cnt), 0); |
ac4d6888 CS |
84 | if (!bp) |
85 | return XFS_ERROR(ENOMEM); | |
e5702805 | 86 | error = bp->b_error; |
804c83c3 | 87 | if (error) { |
901796af | 88 | xfs_buf_ioerror_alert(bp, __func__); |
804c83c3 CH |
89 | xfs_buf_relse(bp); |
90 | goto out; | |
91 | } | |
92 | if (pathlen < byte_cnt) | |
93 | byte_cnt = pathlen; | |
94 | pathlen -= byte_cnt; | |
95 | ||
62926044 | 96 | memcpy(link, bp->b_addr, byte_cnt); |
804c83c3 CH |
97 | xfs_buf_relse(bp); |
98 | } | |
99 | ||
100 | link[ip->i_d.di_size] = '\0'; | |
101 | error = 0; | |
102 | ||
103 | out: | |
104 | return error; | |
105 | } | |
106 | ||
993386c1 | 107 | int |
1da177e4 | 108 | xfs_readlink( |
993386c1 | 109 | xfs_inode_t *ip, |
804c83c3 | 110 | char *link) |
1da177e4 | 111 | { |
804c83c3 | 112 | xfs_mount_t *mp = ip->i_mount; |
b52a360b | 113 | xfs_fsize_t pathlen; |
1da177e4 | 114 | int error = 0; |
1da177e4 | 115 | |
cca28fb8 | 116 | trace_xfs_readlink(ip); |
1da177e4 LT |
117 | |
118 | if (XFS_FORCED_SHUTDOWN(mp)) | |
119 | return XFS_ERROR(EIO); | |
120 | ||
121 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
122 | ||
804c83c3 CH |
123 | pathlen = ip->i_d.di_size; |
124 | if (!pathlen) | |
125 | goto out; | |
1da177e4 | 126 | |
b52a360b CM |
127 | if (pathlen < 0 || pathlen > MAXPATHLEN) { |
128 | xfs_alert(mp, "%s: inode (%llu) bad symlink length (%lld)", | |
129 | __func__, (unsigned long long) ip->i_ino, | |
130 | (long long) pathlen); | |
131 | ASSERT(0); | |
9b025eb3 JK |
132 | error = XFS_ERROR(EFSCORRUPTED); |
133 | goto out; | |
b52a360b CM |
134 | } |
135 | ||
136 | ||
1da177e4 | 137 | if (ip->i_df.if_flags & XFS_IFINLINE) { |
804c83c3 CH |
138 | memcpy(link, ip->i_df.if_u1.if_data, pathlen); |
139 | link[pathlen] = '\0'; | |
140 | } else { | |
141 | error = xfs_readlink_bmap(ip, link); | |
1da177e4 LT |
142 | } |
143 | ||
804c83c3 | 144 | out: |
1da177e4 | 145 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
1da177e4 LT |
146 | return error; |
147 | } | |
148 | ||
1da177e4 | 149 | /* |
92dfe8d2 DC |
150 | * This is called by xfs_inactive to free any blocks beyond eof |
151 | * when the link count isn't zero and by xfs_dm_punch_hole() when | |
152 | * punching a hole to EOF. | |
1da177e4 | 153 | */ |
d96f8f89 | 154 | STATIC int |
92dfe8d2 | 155 | xfs_free_eofblocks( |
1da177e4 | 156 | xfs_mount_t *mp, |
92dfe8d2 | 157 | xfs_inode_t *ip, |
5a15322d | 158 | bool need_iolock) |
1da177e4 LT |
159 | { |
160 | xfs_trans_t *tp; | |
161 | int error; | |
162 | xfs_fileoff_t end_fsb; | |
163 | xfs_fileoff_t last_fsb; | |
164 | xfs_filblks_t map_len; | |
165 | int nimaps; | |
166 | xfs_bmbt_irec_t imap; | |
167 | ||
168 | /* | |
169 | * Figure out if there are any blocks beyond the end | |
170 | * of the file. If not, then there is nothing to do. | |
171 | */ | |
ce7ae151 | 172 | end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_ISIZE(ip)); |
32972383 | 173 | last_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes); |
3f34885c | 174 | if (last_fsb <= end_fsb) |
014c2544 | 175 | return 0; |
3f34885c | 176 | map_len = last_fsb - end_fsb; |
1da177e4 LT |
177 | |
178 | nimaps = 1; | |
179 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
5c8ed202 | 180 | error = xfs_bmapi_read(ip, end_fsb, map_len, &imap, &nimaps, 0); |
1da177e4 LT |
181 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
182 | ||
183 | if (!error && (nimaps != 0) && | |
68bdb6ea YL |
184 | (imap.br_startblock != HOLESTARTBLOCK || |
185 | ip->i_delayed_blks)) { | |
1da177e4 LT |
186 | /* |
187 | * Attach the dquots to the inode up front. | |
188 | */ | |
7d095257 CH |
189 | error = xfs_qm_dqattach(ip, 0); |
190 | if (error) | |
014c2544 | 191 | return error; |
1da177e4 LT |
192 | |
193 | /* | |
194 | * There are blocks after the end of file. | |
195 | * Free them up now by truncating the file to | |
196 | * its current size. | |
197 | */ | |
198 | tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE); | |
199 | ||
5a15322d | 200 | if (need_iolock) { |
c56c9631 CH |
201 | if (!xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) { |
202 | xfs_trans_cancel(tp, 0); | |
203 | return 0; | |
204 | } | |
c56c9631 | 205 | } |
1da177e4 LT |
206 | |
207 | error = xfs_trans_reserve(tp, 0, | |
208 | XFS_ITRUNCATE_LOG_RES(mp), | |
209 | 0, XFS_TRANS_PERM_LOG_RES, | |
210 | XFS_ITRUNCATE_LOG_COUNT); | |
211 | if (error) { | |
212 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
213 | xfs_trans_cancel(tp, 0); | |
5a15322d CH |
214 | if (need_iolock) |
215 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
014c2544 | 216 | return error; |
1da177e4 LT |
217 | } |
218 | ||
219 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
ddc3415a | 220 | xfs_trans_ijoin(tp, ip, 0); |
1da177e4 | 221 | |
673e8e59 CH |
222 | /* |
223 | * Do not update the on-disk file size. If we update the | |
224 | * on-disk file size and then the system crashes before the | |
225 | * contents of the file are flushed to disk then the files | |
226 | * may be full of holes (ie NULL files bug). | |
227 | */ | |
228 | error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, | |
ce7ae151 | 229 | XFS_ISIZE(ip)); |
1da177e4 | 230 | if (error) { |
8f04c47a CH |
231 | /* |
232 | * If we get an error at this point we simply don't | |
233 | * bother truncating the file. | |
234 | */ | |
1da177e4 LT |
235 | xfs_trans_cancel(tp, |
236 | (XFS_TRANS_RELEASE_LOG_RES | | |
237 | XFS_TRANS_ABORT)); | |
238 | } else { | |
239 | error = xfs_trans_commit(tp, | |
1c72bf90 | 240 | XFS_TRANS_RELEASE_LOG_RES); |
27b52867 BF |
241 | if (!error) |
242 | xfs_inode_clear_eofblocks_tag(ip); | |
1da177e4 | 243 | } |
5a15322d CH |
244 | |
245 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
246 | if (need_iolock) | |
247 | xfs_iunlock(ip, XFS_IOLOCK_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 | ||
993386c1 | 391 | int |
1da177e4 | 392 | xfs_release( |
993386c1 | 393 | xfs_inode_t *ip) |
1da177e4 | 394 | { |
993386c1 | 395 | xfs_mount_t *mp = ip->i_mount; |
1da177e4 LT |
396 | int error; |
397 | ||
42173f68 | 398 | if (!S_ISREG(ip->i_d.di_mode) || (ip->i_d.di_mode == 0)) |
1da177e4 | 399 | return 0; |
1da177e4 LT |
400 | |
401 | /* If this is a read-only mount, don't do this (would generate I/O) */ | |
bd186aa9 | 402 | if (mp->m_flags & XFS_MOUNT_RDONLY) |
1da177e4 LT |
403 | return 0; |
404 | ||
2a82b8be | 405 | if (!XFS_FORCED_SHUTDOWN(mp)) { |
b3aea4ed CH |
406 | int truncated; |
407 | ||
2a82b8be DC |
408 | /* |
409 | * If we are using filestreams, and we have an unlinked | |
410 | * file that we are processing the last close on, then nothing | |
411 | * will be able to reopen and write to this file. Purge this | |
412 | * inode from the filestreams cache so that it doesn't delay | |
413 | * teardown of the inode. | |
414 | */ | |
415 | if ((ip->i_d.di_nlink == 0) && xfs_inode_is_filestream(ip)) | |
416 | xfs_filestream_deassociate(ip); | |
417 | ||
fbf3ce8d CH |
418 | /* |
419 | * If we previously truncated this file and removed old data | |
420 | * in the process, we want to initiate "early" writeout on | |
421 | * the last close. This is an attempt to combat the notorious | |
25985edc | 422 | * NULL files problem which is particularly noticeable from a |
fbf3ce8d CH |
423 | * truncate down, buffered (re-)write (delalloc), followed by |
424 | * a crash. What we are effectively doing here is | |
425 | * significantly reducing the time window where we'd otherwise | |
426 | * be exposed to that problem. | |
427 | */ | |
09262b43 | 428 | truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED); |
df4368a1 DC |
429 | if (truncated) { |
430 | xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE); | |
431 | if (VN_DIRTY(VFS_I(ip)) && ip->i_delayed_blks > 0) | |
432 | xfs_flush_pages(ip, 0, -1, XBF_ASYNC, FI_NONE); | |
433 | } | |
fbf3ce8d CH |
434 | } |
435 | ||
6e857567 DC |
436 | if (ip->i_d.di_nlink == 0) |
437 | return 0; | |
c56c9631 | 438 | |
abbede1b | 439 | if ((S_ISREG(ip->i_d.di_mode) && |
ce7ae151 CH |
440 | (VFS_I(ip)->i_size > 0 || |
441 | (VN_CACHED(VFS_I(ip)) > 0 || ip->i_delayed_blks > 0)) && | |
6e857567 DC |
442 | (ip->i_df.if_flags & XFS_IFEXTENTS)) && |
443 | (!(ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)))) { | |
444 | ||
445 | /* | |
446 | * If we can't get the iolock just skip truncating the blocks | |
447 | * past EOF because we could deadlock with the mmap_sem | |
448 | * otherwise. We'll get another chance to drop them once the | |
449 | * last reference to the inode is dropped, so we'll never leak | |
450 | * blocks permanently. | |
451 | * | |
452 | * Further, check if the inode is being opened, written and | |
453 | * closed frequently and we have delayed allocation blocks | |
25985edc | 454 | * outstanding (e.g. streaming writes from the NFS server), |
6e857567 DC |
455 | * truncating the blocks past EOF will cause fragmentation to |
456 | * occur. | |
457 | * | |
458 | * In this case don't do the truncation, either, but we have to | |
459 | * be careful how we detect this case. Blocks beyond EOF show | |
460 | * up as i_delayed_blks even when the inode is clean, so we | |
461 | * need to truncate them away first before checking for a dirty | |
462 | * release. Hence on the first dirty close we will still remove | |
463 | * the speculative allocation, but after that we will leave it | |
464 | * in place. | |
465 | */ | |
466 | if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE)) | |
467 | return 0; | |
1da177e4 | 468 | |
5a15322d | 469 | error = xfs_free_eofblocks(mp, ip, true); |
6e857567 DC |
470 | if (error) |
471 | return error; | |
472 | ||
473 | /* delalloc blocks after truncation means it really is dirty */ | |
474 | if (ip->i_delayed_blks) | |
475 | xfs_iflags_set(ip, XFS_IDIRTY_RELEASE); | |
476 | } | |
1da177e4 LT |
477 | return 0; |
478 | } | |
479 | ||
480 | /* | |
481 | * xfs_inactive | |
482 | * | |
483 | * This is called when the vnode reference count for the vnode | |
484 | * goes to zero. If the file has been unlinked, then it must | |
485 | * now be truncated. Also, we clear all of the read-ahead state | |
486 | * kept for the inode here since the file is now closed. | |
487 | */ | |
993386c1 | 488 | int |
1da177e4 | 489 | xfs_inactive( |
993386c1 | 490 | xfs_inode_t *ip) |
1da177e4 | 491 | { |
67fcaa73 | 492 | xfs_bmap_free_t free_list; |
1da177e4 LT |
493 | xfs_fsblock_t first_block; |
494 | int committed; | |
495 | xfs_trans_t *tp; | |
496 | xfs_mount_t *mp; | |
497 | int error; | |
b373e98d | 498 | int truncate = 0; |
1da177e4 | 499 | |
1da177e4 LT |
500 | /* |
501 | * If the inode is already free, then there can be nothing | |
502 | * to clean up here. | |
503 | */ | |
cb4c8cc1 | 504 | if (ip->i_d.di_mode == 0 || is_bad_inode(VFS_I(ip))) { |
1da177e4 LT |
505 | ASSERT(ip->i_df.if_real_bytes == 0); |
506 | ASSERT(ip->i_df.if_broot_bytes == 0); | |
507 | return VN_INACTIVE_CACHE; | |
508 | } | |
509 | ||
1da177e4 LT |
510 | mp = ip->i_mount; |
511 | ||
1da177e4 LT |
512 | error = 0; |
513 | ||
514 | /* If this is a read-only mount, don't do this (would generate I/O) */ | |
bd186aa9 | 515 | if (mp->m_flags & XFS_MOUNT_RDONLY) |
1da177e4 LT |
516 | goto out; |
517 | ||
518 | if (ip->i_d.di_nlink != 0) { | |
abbede1b | 519 | if ((S_ISREG(ip->i_d.di_mode) && |
ce7ae151 CH |
520 | (VFS_I(ip)->i_size > 0 || |
521 | (VN_CACHED(VFS_I(ip)) > 0 || ip->i_delayed_blks > 0)) && | |
522 | (ip->i_df.if_flags & XFS_IFEXTENTS) && | |
523 | (!(ip->i_d.di_flags & | |
dd9f438e | 524 | (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)) || |
ce7ae151 | 525 | ip->i_delayed_blks != 0))) { |
5a15322d | 526 | error = xfs_free_eofblocks(mp, ip, false); |
92dfe8d2 | 527 | if (error) |
014c2544 | 528 | return VN_INACTIVE_CACHE; |
1da177e4 LT |
529 | } |
530 | goto out; | |
531 | } | |
532 | ||
b373e98d CH |
533 | if (S_ISREG(ip->i_d.di_mode) && |
534 | (ip->i_d.di_size != 0 || XFS_ISIZE(ip) != 0 || | |
535 | ip->i_d.di_nextents > 0 || ip->i_delayed_blks > 0)) | |
536 | truncate = 1; | |
1da177e4 | 537 | |
7d095257 CH |
538 | error = xfs_qm_dqattach(ip, 0); |
539 | if (error) | |
014c2544 | 540 | return VN_INACTIVE_CACHE; |
1da177e4 LT |
541 | |
542 | tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE); | |
b373e98d CH |
543 | error = xfs_trans_reserve(tp, 0, |
544 | (truncate || S_ISLNK(ip->i_d.di_mode)) ? | |
545 | XFS_ITRUNCATE_LOG_RES(mp) : | |
546 | XFS_IFREE_LOG_RES(mp), | |
547 | 0, | |
548 | XFS_TRANS_PERM_LOG_RES, | |
549 | XFS_ITRUNCATE_LOG_COUNT); | |
550 | if (error) { | |
551 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
552 | xfs_trans_cancel(tp, 0); | |
553 | return VN_INACTIVE_CACHE; | |
554 | } | |
1da177e4 | 555 | |
0b56185b | 556 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
b373e98d | 557 | xfs_trans_ijoin(tp, ip, 0); |
1da177e4 | 558 | |
b373e98d CH |
559 | if (S_ISLNK(ip->i_d.di_mode)) { |
560 | /* | |
561 | * Zero length symlinks _can_ exist. | |
562 | */ | |
563 | if (ip->i_d.di_size > XFS_IFORK_DSIZE(ip)) { | |
564 | error = xfs_inactive_symlink_rmt(ip, &tp); | |
565 | if (error) | |
566 | goto out_cancel; | |
567 | } else if (ip->i_df.if_bytes > 0) { | |
568 | xfs_idata_realloc(ip, -(ip->i_df.if_bytes), | |
569 | XFS_DATA_FORK); | |
570 | ASSERT(ip->i_df.if_bytes == 0); | |
571 | } | |
572 | } else if (truncate) { | |
673e8e59 | 573 | ip->i_d.di_size = 0; |
673e8e59 CH |
574 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); |
575 | ||
576 | error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0); | |
b373e98d CH |
577 | if (error) |
578 | goto out_cancel; | |
673e8e59 CH |
579 | |
580 | ASSERT(ip->i_d.di_nextents == 0); | |
1da177e4 LT |
581 | } |
582 | ||
583 | /* | |
fe67be03 CH |
584 | * If there are attributes associated with the file then blow them away |
585 | * now. The code calls a routine that recursively deconstructs the | |
586 | * attribute fork. We need to just commit the current transaction | |
1da177e4 LT |
587 | * because we can't use it for xfs_attr_inactive(). |
588 | */ | |
589 | if (ip->i_d.di_anextents > 0) { | |
fe67be03 CH |
590 | ASSERT(ip->i_d.di_forkoff != 0); |
591 | ||
592 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); | |
fe67be03 | 593 | if (error) |
0b56185b CH |
594 | goto out_unlock; |
595 | ||
596 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
fe67be03 CH |
597 | |
598 | error = xfs_attr_inactive(ip); | |
599 | if (error) | |
0b56185b | 600 | goto out; |
fe67be03 CH |
601 | |
602 | tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE); | |
603 | error = xfs_trans_reserve(tp, 0, | |
604 | XFS_IFREE_LOG_RES(mp), | |
605 | 0, XFS_TRANS_PERM_LOG_RES, | |
606 | XFS_INACTIVE_LOG_COUNT); | |
0b56185b CH |
607 | if (error) { |
608 | xfs_trans_cancel(tp, 0); | |
609 | goto out; | |
610 | } | |
fe67be03 CH |
611 | |
612 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
613 | xfs_trans_ijoin(tp, ip, 0); | |
1da177e4 LT |
614 | } |
615 | ||
fe67be03 CH |
616 | if (ip->i_afp) |
617 | xfs_idestroy_fork(ip, XFS_ATTR_FORK); | |
618 | ||
619 | ASSERT(ip->i_d.di_anextents == 0); | |
620 | ||
1da177e4 LT |
621 | /* |
622 | * Free the inode. | |
623 | */ | |
9d87c319 | 624 | xfs_bmap_init(&free_list, &first_block); |
1da177e4 LT |
625 | error = xfs_ifree(tp, ip, &free_list); |
626 | if (error) { | |
627 | /* | |
628 | * If we fail to free the inode, shut down. The cancel | |
629 | * might do that, we need to make sure. Otherwise the | |
630 | * inode might be lost for a long time or forever. | |
631 | */ | |
632 | if (!XFS_FORCED_SHUTDOWN(mp)) { | |
0b932ccc DC |
633 | xfs_notice(mp, "%s: xfs_ifree returned error %d", |
634 | __func__, error); | |
7d04a335 | 635 | xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR); |
1da177e4 LT |
636 | } |
637 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT); | |
638 | } else { | |
639 | /* | |
640 | * Credit the quota account(s). The inode is gone. | |
641 | */ | |
7d095257 | 642 | xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1); |
1da177e4 LT |
643 | |
644 | /* | |
78e9da77 DC |
645 | * Just ignore errors at this point. There is nothing we can |
646 | * do except to try to keep going. Make sure it's not a silent | |
647 | * error. | |
1da177e4 | 648 | */ |
78e9da77 DC |
649 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
650 | if (error) | |
53487786 DC |
651 | xfs_notice(mp, "%s: xfs_bmap_finish returned error %d", |
652 | __func__, error); | |
78e9da77 DC |
653 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
654 | if (error) | |
53487786 DC |
655 | xfs_notice(mp, "%s: xfs_trans_commit returned error %d", |
656 | __func__, error); | |
1da177e4 | 657 | } |
7d095257 | 658 | |
1da177e4 LT |
659 | /* |
660 | * Release the dquots held by inode, if any. | |
661 | */ | |
7d095257 | 662 | xfs_qm_dqdetach(ip); |
0b56185b CH |
663 | out_unlock: |
664 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
b373e98d CH |
665 | out: |
666 | return VN_INACTIVE_CACHE; | |
667 | out_cancel: | |
668 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT); | |
0b56185b | 669 | goto out_unlock; |
1da177e4 LT |
670 | } |
671 | ||
384f3ced BN |
672 | /* |
673 | * Lookups up an inode from "name". If ci_name is not NULL, then a CI match | |
674 | * is allowed, otherwise it has to be an exact match. If a CI match is found, | |
675 | * ci_name->name will point to a the actual name (caller must free) or | |
676 | * will be set to NULL if an exact match is found. | |
677 | */ | |
993386c1 | 678 | int |
1da177e4 | 679 | xfs_lookup( |
993386c1 | 680 | xfs_inode_t *dp, |
556b8b16 | 681 | struct xfs_name *name, |
384f3ced BN |
682 | xfs_inode_t **ipp, |
683 | struct xfs_name *ci_name) | |
1da177e4 | 684 | { |
eca450b7 | 685 | xfs_ino_t inum; |
1da177e4 LT |
686 | int error; |
687 | uint lock_mode; | |
1da177e4 | 688 | |
cca28fb8 | 689 | trace_xfs_lookup(dp, name); |
1da177e4 LT |
690 | |
691 | if (XFS_FORCED_SHUTDOWN(dp->i_mount)) | |
692 | return XFS_ERROR(EIO); | |
693 | ||
694 | lock_mode = xfs_ilock_map_shared(dp); | |
384f3ced | 695 | error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name); |
1da177e4 | 696 | xfs_iunlock_map_shared(dp, lock_mode); |
eca450b7 CH |
697 | |
698 | if (error) | |
699 | goto out; | |
700 | ||
7b6259e7 | 701 | error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp); |
eca450b7 | 702 | if (error) |
384f3ced | 703 | goto out_free_name; |
eca450b7 | 704 | |
eca450b7 CH |
705 | return 0; |
706 | ||
384f3ced BN |
707 | out_free_name: |
708 | if (ci_name) | |
709 | kmem_free(ci_name->name); | |
710 | out: | |
eca450b7 | 711 | *ipp = NULL; |
1da177e4 LT |
712 | return error; |
713 | } | |
714 | ||
993386c1 | 715 | int |
1da177e4 | 716 | xfs_create( |
993386c1 | 717 | xfs_inode_t *dp, |
556b8b16 | 718 | struct xfs_name *name, |
576b1d67 | 719 | umode_t mode, |
3e5daf05 | 720 | xfs_dev_t rdev, |
6c77b0ea | 721 | xfs_inode_t **ipp) |
1da177e4 | 722 | { |
517b5e8c CH |
723 | int is_dir = S_ISDIR(mode); |
724 | struct xfs_mount *mp = dp->i_mount; | |
725 | struct xfs_inode *ip = NULL; | |
726 | struct xfs_trans *tp = NULL; | |
556b8b16 | 727 | int error; |
1da177e4 LT |
728 | xfs_bmap_free_t free_list; |
729 | xfs_fsblock_t first_block; | |
993386c1 | 730 | boolean_t unlock_dp_on_error = B_FALSE; |
1da177e4 LT |
731 | uint cancel_flags; |
732 | int committed; | |
6743099c | 733 | prid_t prid; |
517b5e8c CH |
734 | struct xfs_dquot *udqp = NULL; |
735 | struct xfs_dquot *gdqp = NULL; | |
1da177e4 | 736 | uint resblks; |
517b5e8c CH |
737 | uint log_res; |
738 | uint log_count; | |
1da177e4 | 739 | |
cca28fb8 | 740 | trace_xfs_create(dp, name); |
1da177e4 | 741 | |
517b5e8c CH |
742 | if (XFS_FORCED_SHUTDOWN(mp)) |
743 | return XFS_ERROR(EIO); | |
744 | ||
365ca83d | 745 | if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) |
6743099c | 746 | prid = xfs_get_projid(dp); |
1da177e4 | 747 | else |
6743099c | 748 | prid = XFS_PROJID_DEFAULT; |
1da177e4 LT |
749 | |
750 | /* | |
751 | * Make sure that we have allocated dquot(s) on disk. | |
752 | */ | |
7d095257 | 753 | error = xfs_qm_vop_dqalloc(dp, current_fsuid(), current_fsgid(), prid, |
517b5e8c | 754 | XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp); |
1da177e4 | 755 | if (error) |
ec3ba85f | 756 | return error; |
1da177e4 | 757 | |
517b5e8c CH |
758 | if (is_dir) { |
759 | rdev = 0; | |
760 | resblks = XFS_MKDIR_SPACE_RES(mp, name->len); | |
761 | log_res = XFS_MKDIR_LOG_RES(mp); | |
762 | log_count = XFS_MKDIR_LOG_COUNT; | |
763 | tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR); | |
764 | } else { | |
765 | resblks = XFS_CREATE_SPACE_RES(mp, name->len); | |
766 | log_res = XFS_CREATE_LOG_RES(mp); | |
767 | log_count = XFS_CREATE_LOG_COUNT; | |
768 | tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE); | |
769 | } | |
1da177e4 | 770 | |
1da177e4 | 771 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; |
517b5e8c | 772 | |
1da177e4 LT |
773 | /* |
774 | * Initially assume that the file does not exist and | |
775 | * reserve the resources for that case. If that is not | |
776 | * the case we'll drop the one we have and get a more | |
777 | * appropriate transaction later. | |
778 | */ | |
517b5e8c CH |
779 | error = xfs_trans_reserve(tp, resblks, log_res, 0, |
780 | XFS_TRANS_PERM_LOG_RES, log_count); | |
1da177e4 | 781 | if (error == ENOSPC) { |
153fec43 | 782 | /* flush outstanding delalloc blocks and retry */ |
9aa05000 | 783 | xfs_flush_inodes(mp); |
4734d401 CH |
784 | error = xfs_trans_reserve(tp, resblks, log_res, 0, |
785 | XFS_TRANS_PERM_LOG_RES, log_count); | |
153fec43 DC |
786 | } |
787 | if (error == ENOSPC) { | |
788 | /* No space at all so try a "no-allocation" reservation */ | |
1da177e4 | 789 | resblks = 0; |
517b5e8c CH |
790 | error = xfs_trans_reserve(tp, 0, log_res, 0, |
791 | XFS_TRANS_PERM_LOG_RES, log_count); | |
1da177e4 LT |
792 | } |
793 | if (error) { | |
794 | cancel_flags = 0; | |
517b5e8c | 795 | goto out_trans_cancel; |
1da177e4 LT |
796 | } |
797 | ||
f7c66ce3 | 798 | xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT); |
993386c1 | 799 | unlock_dp_on_error = B_TRUE; |
1da177e4 | 800 | |
517b5e8c | 801 | xfs_bmap_init(&free_list, &first_block); |
1da177e4 LT |
802 | |
803 | /* | |
804 | * Reserve disk quota and the inode. | |
805 | */ | |
7d095257 | 806 | error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp, resblks, 1, 0); |
1da177e4 | 807 | if (error) |
517b5e8c | 808 | goto out_trans_cancel; |
1da177e4 | 809 | |
556b8b16 BN |
810 | error = xfs_dir_canenter(tp, dp, name, resblks); |
811 | if (error) | |
517b5e8c CH |
812 | goto out_trans_cancel; |
813 | ||
814 | /* | |
815 | * A newly created regular or special file just has one directory | |
816 | * entry pointing to them, but a directory also the "." entry | |
817 | * pointing to itself. | |
818 | */ | |
6c77b0ea | 819 | error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev, |
517b5e8c | 820 | prid, resblks > 0, &ip, &committed); |
1da177e4 LT |
821 | if (error) { |
822 | if (error == ENOSPC) | |
517b5e8c CH |
823 | goto out_trans_cancel; |
824 | goto out_trans_abort; | |
1da177e4 | 825 | } |
1da177e4 | 826 | |
1da177e4 | 827 | /* |
993386c1 CH |
828 | * Now we join the directory inode to the transaction. We do not do it |
829 | * earlier because xfs_dir_ialloc might commit the previous transaction | |
830 | * (and release all the locks). An error from here on will result in | |
831 | * the transaction cancel unlocking dp so don't do it explicitly in the | |
832 | * error path. | |
1da177e4 | 833 | */ |
ddc3415a | 834 | xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); |
993386c1 | 835 | unlock_dp_on_error = B_FALSE; |
1da177e4 | 836 | |
556b8b16 | 837 | error = xfs_dir_createname(tp, dp, name, ip->i_ino, |
f6c2d1fa NS |
838 | &first_block, &free_list, resblks ? |
839 | resblks - XFS_IALLOC_SPACE_RES(mp) : 0); | |
1da177e4 LT |
840 | if (error) { |
841 | ASSERT(error != ENOSPC); | |
517b5e8c | 842 | goto out_trans_abort; |
1da177e4 | 843 | } |
dcd79a14 | 844 | xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); |
1da177e4 LT |
845 | xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); |
846 | ||
517b5e8c CH |
847 | if (is_dir) { |
848 | error = xfs_dir_init(tp, ip, dp); | |
849 | if (error) | |
850 | goto out_bmap_cancel; | |
851 | ||
852 | error = xfs_bumplink(tp, dp); | |
853 | if (error) | |
854 | goto out_bmap_cancel; | |
855 | } | |
856 | ||
1da177e4 LT |
857 | /* |
858 | * If this is a synchronous mount, make sure that the | |
859 | * create transaction goes to disk before returning to | |
860 | * the user. | |
861 | */ | |
517b5e8c | 862 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) |
1da177e4 | 863 | xfs_trans_set_sync(tp); |
1da177e4 | 864 | |
1da177e4 LT |
865 | /* |
866 | * Attach the dquot(s) to the inodes and modify them incore. | |
867 | * These ids of the inode couldn't have changed since the new | |
868 | * inode has been locked ever since it was created. | |
869 | */ | |
7d095257 | 870 | xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp); |
1da177e4 | 871 | |
f7c99b6f | 872 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
517b5e8c | 873 | if (error) |
ec3ba85f | 874 | goto out_bmap_cancel; |
1da177e4 | 875 | |
1c72bf90 | 876 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
ec3ba85f CH |
877 | if (error) |
878 | goto out_release_inode; | |
1da177e4 | 879 | |
7d095257 CH |
880 | xfs_qm_dqrele(udqp); |
881 | xfs_qm_dqrele(gdqp); | |
1da177e4 | 882 | |
979ebab1 | 883 | *ipp = ip; |
288699fe | 884 | return 0; |
1da177e4 | 885 | |
517b5e8c CH |
886 | out_bmap_cancel: |
887 | xfs_bmap_cancel(&free_list); | |
888 | out_trans_abort: | |
1da177e4 | 889 | cancel_flags |= XFS_TRANS_ABORT; |
517b5e8c CH |
890 | out_trans_cancel: |
891 | xfs_trans_cancel(tp, cancel_flags); | |
ec3ba85f CH |
892 | out_release_inode: |
893 | /* | |
894 | * Wait until after the current transaction is aborted to | |
895 | * release the inode. This prevents recursive transactions | |
896 | * and deadlocks from xfs_inactive. | |
897 | */ | |
898 | if (ip) | |
899 | IRELE(ip); | |
900 | ||
7d095257 CH |
901 | xfs_qm_dqrele(udqp); |
902 | xfs_qm_dqrele(gdqp); | |
1da177e4 | 903 | |
993386c1 CH |
904 | if (unlock_dp_on_error) |
905 | xfs_iunlock(dp, XFS_ILOCK_EXCL); | |
288699fe | 906 | return error; |
1da177e4 LT |
907 | } |
908 | ||
1da177e4 LT |
909 | #ifdef DEBUG |
910 | int xfs_locked_n; | |
911 | int xfs_small_retries; | |
912 | int xfs_middle_retries; | |
913 | int xfs_lots_retries; | |
914 | int xfs_lock_delays; | |
915 | #endif | |
916 | ||
f7c66ce3 LM |
917 | /* |
918 | * Bump the subclass so xfs_lock_inodes() acquires each lock with | |
919 | * a different value | |
920 | */ | |
921 | static inline int | |
922 | xfs_lock_inumorder(int lock_mode, int subclass) | |
923 | { | |
924 | if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) | |
0f1145cc | 925 | lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_IOLOCK_SHIFT; |
f7c66ce3 | 926 | if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) |
0f1145cc | 927 | lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_ILOCK_SHIFT; |
f7c66ce3 LM |
928 | |
929 | return lock_mode; | |
930 | } | |
931 | ||
1da177e4 LT |
932 | /* |
933 | * The following routine will lock n inodes in exclusive mode. | |
934 | * We assume the caller calls us with the inodes in i_ino order. | |
935 | * | |
936 | * We need to detect deadlock where an inode that we lock | |
937 | * is in the AIL and we start waiting for another inode that is locked | |
938 | * by a thread in a long running transaction (such as truncate). This can | |
939 | * result in deadlock since the long running trans might need to wait | |
940 | * for the inode we just locked in order to push the tail and free space | |
941 | * in the log. | |
942 | */ | |
943 | void | |
944 | xfs_lock_inodes( | |
945 | xfs_inode_t **ips, | |
946 | int inodes, | |
1da177e4 LT |
947 | uint lock_mode) |
948 | { | |
949 | int attempts = 0, i, j, try_lock; | |
950 | xfs_log_item_t *lp; | |
951 | ||
952 | ASSERT(ips && (inodes >= 2)); /* we need at least two */ | |
953 | ||
cfa853e4 CH |
954 | try_lock = 0; |
955 | i = 0; | |
1da177e4 LT |
956 | |
957 | again: | |
958 | for (; i < inodes; i++) { | |
959 | ASSERT(ips[i]); | |
960 | ||
961 | if (i && (ips[i] == ips[i-1])) /* Already locked */ | |
962 | continue; | |
963 | ||
964 | /* | |
965 | * If try_lock is not set yet, make sure all locked inodes | |
966 | * are not in the AIL. | |
967 | * If any are, set try_lock to be used later. | |
968 | */ | |
969 | ||
970 | if (!try_lock) { | |
971 | for (j = (i - 1); j >= 0 && !try_lock; j--) { | |
972 | lp = (xfs_log_item_t *)ips[j]->i_itemp; | |
973 | if (lp && (lp->li_flags & XFS_LI_IN_AIL)) { | |
974 | try_lock++; | |
975 | } | |
976 | } | |
977 | } | |
978 | ||
979 | /* | |
980 | * If any of the previous locks we have locked is in the AIL, | |
981 | * we must TRY to get the second and subsequent locks. If | |
982 | * we can't get any, we must release all we have | |
983 | * and try again. | |
984 | */ | |
985 | ||
986 | if (try_lock) { | |
987 | /* try_lock must be 0 if i is 0. */ | |
988 | /* | |
989 | * try_lock means we have an inode locked | |
990 | * that is in the AIL. | |
991 | */ | |
992 | ASSERT(i != 0); | |
f7c66ce3 | 993 | if (!xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i))) { |
1da177e4 LT |
994 | attempts++; |
995 | ||
996 | /* | |
997 | * Unlock all previous guys and try again. | |
998 | * xfs_iunlock will try to push the tail | |
999 | * if the inode is in the AIL. | |
1000 | */ | |
1001 | ||
1002 | for(j = i - 1; j >= 0; j--) { | |
1003 | ||
1004 | /* | |
1005 | * Check to see if we've already | |
1006 | * unlocked this one. | |
1007 | * Not the first one going back, | |
1008 | * and the inode ptr is the same. | |
1009 | */ | |
1010 | if ((j != (i - 1)) && ips[j] == | |
1011 | ips[j+1]) | |
1012 | continue; | |
1013 | ||
1014 | xfs_iunlock(ips[j], lock_mode); | |
1015 | } | |
1016 | ||
1017 | if ((attempts % 5) == 0) { | |
1018 | delay(1); /* Don't just spin the CPU */ | |
1019 | #ifdef DEBUG | |
1020 | xfs_lock_delays++; | |
1021 | #endif | |
1022 | } | |
1023 | i = 0; | |
1024 | try_lock = 0; | |
1025 | goto again; | |
1026 | } | |
1027 | } else { | |
f7c66ce3 | 1028 | xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i)); |
1da177e4 LT |
1029 | } |
1030 | } | |
1031 | ||
1032 | #ifdef DEBUG | |
1033 | if (attempts) { | |
1034 | if (attempts < 5) xfs_small_retries++; | |
1035 | else if (attempts < 100) xfs_middle_retries++; | |
1036 | else xfs_lots_retries++; | |
1037 | } else { | |
1038 | xfs_locked_n++; | |
1039 | } | |
1040 | #endif | |
1041 | } | |
1042 | ||
f9114eba DC |
1043 | /* |
1044 | * xfs_lock_two_inodes() can only be used to lock one type of lock | |
1045 | * at a time - the iolock or the ilock, but not both at once. If | |
1046 | * we lock both at once, lockdep will report false positives saying | |
1047 | * we have violated locking orders. | |
1048 | */ | |
e1cccd91 CH |
1049 | void |
1050 | xfs_lock_two_inodes( | |
1051 | xfs_inode_t *ip0, | |
1052 | xfs_inode_t *ip1, | |
1053 | uint lock_mode) | |
1054 | { | |
1055 | xfs_inode_t *temp; | |
1056 | int attempts = 0; | |
1057 | xfs_log_item_t *lp; | |
1058 | ||
f9114eba DC |
1059 | if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) |
1060 | ASSERT((lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) == 0); | |
e1cccd91 CH |
1061 | ASSERT(ip0->i_ino != ip1->i_ino); |
1062 | ||
1063 | if (ip0->i_ino > ip1->i_ino) { | |
1064 | temp = ip0; | |
1065 | ip0 = ip1; | |
1066 | ip1 = temp; | |
1067 | } | |
1068 | ||
1069 | again: | |
1070 | xfs_ilock(ip0, xfs_lock_inumorder(lock_mode, 0)); | |
1071 | ||
1072 | /* | |
1073 | * If the first lock we have locked is in the AIL, we must TRY to get | |
1074 | * the second lock. If we can't get it, we must release the first one | |
1075 | * and try again. | |
1076 | */ | |
1077 | lp = (xfs_log_item_t *)ip0->i_itemp; | |
1078 | if (lp && (lp->li_flags & XFS_LI_IN_AIL)) { | |
1079 | if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(lock_mode, 1))) { | |
1080 | xfs_iunlock(ip0, lock_mode); | |
1081 | if ((++attempts % 5) == 0) | |
1082 | delay(1); /* Don't just spin the CPU */ | |
1083 | goto again; | |
1084 | } | |
1085 | } else { | |
1086 | xfs_ilock(ip1, xfs_lock_inumorder(lock_mode, 1)); | |
1087 | } | |
1088 | } | |
1089 | ||
993386c1 | 1090 | int |
1da177e4 | 1091 | xfs_remove( |
993386c1 | 1092 | xfs_inode_t *dp, |
556b8b16 BN |
1093 | struct xfs_name *name, |
1094 | xfs_inode_t *ip) | |
1da177e4 | 1095 | { |
993386c1 | 1096 | xfs_mount_t *mp = dp->i_mount; |
1da177e4 | 1097 | xfs_trans_t *tp = NULL; |
8f112e3b | 1098 | int is_dir = S_ISDIR(ip->i_d.di_mode); |
1da177e4 LT |
1099 | int error = 0; |
1100 | xfs_bmap_free_t free_list; | |
1101 | xfs_fsblock_t first_block; | |
1102 | int cancel_flags; | |
1103 | int committed; | |
1da177e4 LT |
1104 | int link_zero; |
1105 | uint resblks; | |
8f112e3b | 1106 | uint log_count; |
1da177e4 | 1107 | |
cca28fb8 | 1108 | trace_xfs_remove(dp, name); |
1da177e4 | 1109 | |
1da177e4 LT |
1110 | if (XFS_FORCED_SHUTDOWN(mp)) |
1111 | return XFS_ERROR(EIO); | |
1112 | ||
7d095257 | 1113 | error = xfs_qm_dqattach(dp, 0); |
8f112e3b CH |
1114 | if (error) |
1115 | goto std_return; | |
1116 | ||
7d095257 | 1117 | error = xfs_qm_dqattach(ip, 0); |
8f112e3b | 1118 | if (error) |
1da177e4 | 1119 | goto std_return; |
1da177e4 | 1120 | |
8f112e3b CH |
1121 | if (is_dir) { |
1122 | tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR); | |
1123 | log_count = XFS_DEFAULT_LOG_COUNT; | |
1124 | } else { | |
1125 | tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE); | |
1126 | log_count = XFS_REMOVE_LOG_COUNT; | |
1127 | } | |
1da177e4 | 1128 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; |
8f112e3b | 1129 | |
1da177e4 LT |
1130 | /* |
1131 | * We try to get the real space reservation first, | |
1132 | * allowing for directory btree deletion(s) implying | |
1133 | * possible bmap insert(s). If we can't get the space | |
1134 | * reservation then we use 0 instead, and avoid the bmap | |
1135 | * btree insert(s) in the directory code by, if the bmap | |
1136 | * insert tries to happen, instead trimming the LAST | |
1137 | * block from the directory. | |
1138 | */ | |
1139 | resblks = XFS_REMOVE_SPACE_RES(mp); | |
1140 | error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0, | |
8f112e3b | 1141 | XFS_TRANS_PERM_LOG_RES, log_count); |
1da177e4 LT |
1142 | if (error == ENOSPC) { |
1143 | resblks = 0; | |
1144 | error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0, | |
8f112e3b | 1145 | XFS_TRANS_PERM_LOG_RES, log_count); |
1da177e4 LT |
1146 | } |
1147 | if (error) { | |
1148 | ASSERT(error != ENOSPC); | |
8f112e3b CH |
1149 | cancel_flags = 0; |
1150 | goto out_trans_cancel; | |
1da177e4 LT |
1151 | } |
1152 | ||
e1cccd91 | 1153 | xfs_lock_two_inodes(dp, ip, XFS_ILOCK_EXCL); |
1da177e4 | 1154 | |
ddc3415a CH |
1155 | xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); |
1156 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
1da177e4 | 1157 | |
8f112e3b CH |
1158 | /* |
1159 | * If we're removing a directory perform some additional validation. | |
1160 | */ | |
1161 | if (is_dir) { | |
1162 | ASSERT(ip->i_d.di_nlink >= 2); | |
1163 | if (ip->i_d.di_nlink != 2) { | |
1164 | error = XFS_ERROR(ENOTEMPTY); | |
1165 | goto out_trans_cancel; | |
1166 | } | |
1167 | if (!xfs_dir_isempty(ip)) { | |
1168 | error = XFS_ERROR(ENOTEMPTY); | |
1169 | goto out_trans_cancel; | |
1170 | } | |
1171 | } | |
1172 | ||
9d87c319 | 1173 | xfs_bmap_init(&free_list, &first_block); |
556b8b16 | 1174 | error = xfs_dir_removename(tp, dp, name, ip->i_ino, |
d4377d84 | 1175 | &first_block, &free_list, resblks); |
1da177e4 LT |
1176 | if (error) { |
1177 | ASSERT(error != ENOENT); | |
8f112e3b | 1178 | goto out_bmap_cancel; |
1da177e4 | 1179 | } |
dcd79a14 | 1180 | xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); |
1da177e4 | 1181 | |
8f112e3b CH |
1182 | if (is_dir) { |
1183 | /* | |
1184 | * Drop the link from ip's "..". | |
1185 | */ | |
1186 | error = xfs_droplink(tp, dp); | |
1187 | if (error) | |
1188 | goto out_bmap_cancel; | |
1189 | ||
1190 | /* | |
2b7035fd | 1191 | * Drop the "." link from ip to self. |
8f112e3b CH |
1192 | */ |
1193 | error = xfs_droplink(tp, ip); | |
1194 | if (error) | |
1195 | goto out_bmap_cancel; | |
1196 | } else { | |
1197 | /* | |
1198 | * When removing a non-directory we need to log the parent | |
26c52951 DC |
1199 | * inode here. For a directory this is done implicitly |
1200 | * by the xfs_droplink call for the ".." entry. | |
8f112e3b CH |
1201 | */ |
1202 | xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); | |
1da177e4 LT |
1203 | } |
1204 | ||
8f112e3b | 1205 | /* |
2b7035fd | 1206 | * Drop the link from dp to ip. |
8f112e3b CH |
1207 | */ |
1208 | error = xfs_droplink(tp, ip); | |
1209 | if (error) | |
1210 | goto out_bmap_cancel; | |
1211 | ||
1212 | /* | |
1213 | * Determine if this is the last link while | |
1da177e4 LT |
1214 | * we are in the transaction. |
1215 | */ | |
8f112e3b | 1216 | link_zero = (ip->i_d.di_nlink == 0); |
1da177e4 | 1217 | |
1da177e4 LT |
1218 | /* |
1219 | * If this is a synchronous mount, make sure that the | |
1220 | * remove transaction goes to disk before returning to | |
1221 | * the user. | |
1222 | */ | |
8f112e3b | 1223 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) |
1da177e4 | 1224 | xfs_trans_set_sync(tp); |
1da177e4 | 1225 | |
f7c99b6f | 1226 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
8f112e3b CH |
1227 | if (error) |
1228 | goto out_bmap_cancel; | |
1da177e4 | 1229 | |
1c72bf90 | 1230 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
5df78e73 | 1231 | if (error) |
1da177e4 | 1232 | goto std_return; |
1da177e4 | 1233 | |
2a82b8be DC |
1234 | /* |
1235 | * If we are using filestreams, kill the stream association. | |
1236 | * If the file is still open it may get a new one but that | |
1237 | * will get killed on last close in xfs_close() so we don't | |
1238 | * have to worry about that. | |
1239 | */ | |
8f112e3b | 1240 | if (!is_dir && link_zero && xfs_inode_is_filestream(ip)) |
2a82b8be DC |
1241 | xfs_filestream_deassociate(ip); |
1242 | ||
288699fe | 1243 | return 0; |
1da177e4 | 1244 | |
8f112e3b | 1245 | out_bmap_cancel: |
1da177e4 LT |
1246 | xfs_bmap_cancel(&free_list); |
1247 | cancel_flags |= XFS_TRANS_ABORT; | |
8f112e3b | 1248 | out_trans_cancel: |
1da177e4 | 1249 | xfs_trans_cancel(tp, cancel_flags); |
288699fe CH |
1250 | std_return: |
1251 | return error; | |
1da177e4 LT |
1252 | } |
1253 | ||
993386c1 | 1254 | int |
1da177e4 | 1255 | xfs_link( |
993386c1 | 1256 | xfs_inode_t *tdp, |
a3da7896 | 1257 | xfs_inode_t *sip, |
556b8b16 | 1258 | struct xfs_name *target_name) |
1da177e4 | 1259 | { |
993386c1 | 1260 | xfs_mount_t *mp = tdp->i_mount; |
1da177e4 | 1261 | xfs_trans_t *tp; |
1da177e4 LT |
1262 | int error; |
1263 | xfs_bmap_free_t free_list; | |
1264 | xfs_fsblock_t first_block; | |
1265 | int cancel_flags; | |
1266 | int committed; | |
1da177e4 | 1267 | int resblks; |
1da177e4 | 1268 | |
cca28fb8 | 1269 | trace_xfs_link(tdp, target_name); |
1da177e4 | 1270 | |
a3da7896 | 1271 | ASSERT(!S_ISDIR(sip->i_d.di_mode)); |
1da177e4 | 1272 | |
1da177e4 LT |
1273 | if (XFS_FORCED_SHUTDOWN(mp)) |
1274 | return XFS_ERROR(EIO); | |
1275 | ||
7d095257 | 1276 | error = xfs_qm_dqattach(sip, 0); |
cb3f35bb CH |
1277 | if (error) |
1278 | goto std_return; | |
1279 | ||
7d095257 | 1280 | error = xfs_qm_dqattach(tdp, 0); |
1da177e4 LT |
1281 | if (error) |
1282 | goto std_return; | |
1283 | ||
1284 | tp = xfs_trans_alloc(mp, XFS_TRANS_LINK); | |
1285 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; | |
556b8b16 | 1286 | resblks = XFS_LINK_SPACE_RES(mp, target_name->len); |
1da177e4 LT |
1287 | error = xfs_trans_reserve(tp, resblks, XFS_LINK_LOG_RES(mp), 0, |
1288 | XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT); | |
1289 | if (error == ENOSPC) { | |
1290 | resblks = 0; | |
1291 | error = xfs_trans_reserve(tp, 0, XFS_LINK_LOG_RES(mp), 0, | |
1292 | XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT); | |
1293 | } | |
1294 | if (error) { | |
1295 | cancel_flags = 0; | |
1296 | goto error_return; | |
1297 | } | |
1298 | ||
e1cccd91 | 1299 | xfs_lock_two_inodes(sip, tdp, XFS_ILOCK_EXCL); |
1da177e4 | 1300 | |
ddc3415a CH |
1301 | xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL); |
1302 | xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL); | |
1da177e4 | 1303 | |
365ca83d NS |
1304 | /* |
1305 | * If we are using project inheritance, we only allow hard link | |
1306 | * creation in our tree when the project IDs are the same; else | |
1307 | * the tree quota mechanism could be circumvented. | |
1308 | */ | |
1309 | if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) && | |
6743099c | 1310 | (xfs_get_projid(tdp) != xfs_get_projid(sip)))) { |
b1ecdda9 | 1311 | error = XFS_ERROR(EXDEV); |
365ca83d NS |
1312 | goto error_return; |
1313 | } | |
1314 | ||
556b8b16 BN |
1315 | error = xfs_dir_canenter(tp, tdp, target_name, resblks); |
1316 | if (error) | |
1da177e4 LT |
1317 | goto error_return; |
1318 | ||
9d87c319 | 1319 | xfs_bmap_init(&free_list, &first_block); |
1da177e4 | 1320 | |
556b8b16 BN |
1321 | error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino, |
1322 | &first_block, &free_list, resblks); | |
1da177e4 LT |
1323 | if (error) |
1324 | goto abort_return; | |
dcd79a14 | 1325 | xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); |
1da177e4 LT |
1326 | xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE); |
1327 | ||
1328 | error = xfs_bumplink(tp, sip); | |
b71d300c | 1329 | if (error) |
1da177e4 | 1330 | goto abort_return; |
1da177e4 LT |
1331 | |
1332 | /* | |
1333 | * If this is a synchronous mount, make sure that the | |
1334 | * link transaction goes to disk before returning to | |
1335 | * the user. | |
1336 | */ | |
1337 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) { | |
1338 | xfs_trans_set_sync(tp); | |
1339 | } | |
1340 | ||
f7c99b6f | 1341 | error = xfs_bmap_finish (&tp, &free_list, &committed); |
1da177e4 LT |
1342 | if (error) { |
1343 | xfs_bmap_cancel(&free_list); | |
1344 | goto abort_return; | |
1345 | } | |
1346 | ||
288699fe | 1347 | return xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 LT |
1348 | |
1349 | abort_return: | |
1350 | cancel_flags |= XFS_TRANS_ABORT; | |
1da177e4 LT |
1351 | error_return: |
1352 | xfs_trans_cancel(tp, cancel_flags); | |
288699fe CH |
1353 | std_return: |
1354 | return error; | |
1da177e4 | 1355 | } |
b71d300c | 1356 | |
993386c1 | 1357 | int |
1da177e4 | 1358 | xfs_symlink( |
993386c1 | 1359 | xfs_inode_t *dp, |
556b8b16 BN |
1360 | struct xfs_name *link_name, |
1361 | const char *target_path, | |
576b1d67 | 1362 | umode_t mode, |
6c77b0ea | 1363 | xfs_inode_t **ipp) |
1da177e4 | 1364 | { |
993386c1 | 1365 | xfs_mount_t *mp = dp->i_mount; |
1da177e4 | 1366 | xfs_trans_t *tp; |
1da177e4 LT |
1367 | xfs_inode_t *ip; |
1368 | int error; | |
1369 | int pathlen; | |
1370 | xfs_bmap_free_t free_list; | |
1371 | xfs_fsblock_t first_block; | |
993386c1 | 1372 | boolean_t unlock_dp_on_error = B_FALSE; |
1da177e4 LT |
1373 | uint cancel_flags; |
1374 | int committed; | |
1375 | xfs_fileoff_t first_fsb; | |
1376 | xfs_filblks_t fs_blocks; | |
1377 | int nmaps; | |
1378 | xfs_bmbt_irec_t mval[SYMLINK_MAPS]; | |
1379 | xfs_daddr_t d; | |
556b8b16 | 1380 | const char *cur_chunk; |
1da177e4 LT |
1381 | int byte_cnt; |
1382 | int n; | |
1383 | xfs_buf_t *bp; | |
6743099c | 1384 | prid_t prid; |
1da177e4 LT |
1385 | struct xfs_dquot *udqp, *gdqp; |
1386 | uint resblks; | |
1da177e4 | 1387 | |
3937be5b | 1388 | *ipp = NULL; |
1da177e4 LT |
1389 | error = 0; |
1390 | ip = NULL; | |
1391 | tp = NULL; | |
1392 | ||
cca28fb8 | 1393 | trace_xfs_symlink(dp, link_name); |
1da177e4 LT |
1394 | |
1395 | if (XFS_FORCED_SHUTDOWN(mp)) | |
1396 | return XFS_ERROR(EIO); | |
1397 | ||
1da177e4 LT |
1398 | /* |
1399 | * Check component lengths of the target path name. | |
1400 | */ | |
1401 | pathlen = strlen(target_path); | |
1402 | if (pathlen >= MAXPATHLEN) /* total string too long */ | |
1403 | return XFS_ERROR(ENAMETOOLONG); | |
1da177e4 | 1404 | |
1da177e4 | 1405 | udqp = gdqp = NULL; |
365ca83d | 1406 | if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) |
6743099c | 1407 | prid = xfs_get_projid(dp); |
1da177e4 | 1408 | else |
6743099c | 1409 | prid = XFS_PROJID_DEFAULT; |
1da177e4 LT |
1410 | |
1411 | /* | |
1412 | * Make sure that we have allocated dquot(s) on disk. | |
1413 | */ | |
7d095257 | 1414 | error = xfs_qm_vop_dqalloc(dp, current_fsuid(), current_fsgid(), prid, |
1da177e4 LT |
1415 | XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp); |
1416 | if (error) | |
1417 | goto std_return; | |
1418 | ||
1419 | tp = xfs_trans_alloc(mp, XFS_TRANS_SYMLINK); | |
1420 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; | |
1421 | /* | |
1422 | * The symlink will fit into the inode data fork? | |
1423 | * There can't be any attributes so we get the whole variable part. | |
1424 | */ | |
1425 | if (pathlen <= XFS_LITINO(mp)) | |
1426 | fs_blocks = 0; | |
1427 | else | |
1428 | fs_blocks = XFS_B_TO_FSB(mp, pathlen); | |
556b8b16 | 1429 | resblks = XFS_SYMLINK_SPACE_RES(mp, link_name->len, fs_blocks); |
1da177e4 LT |
1430 | error = xfs_trans_reserve(tp, resblks, XFS_SYMLINK_LOG_RES(mp), 0, |
1431 | XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT); | |
1432 | if (error == ENOSPC && fs_blocks == 0) { | |
1433 | resblks = 0; | |
1434 | error = xfs_trans_reserve(tp, 0, XFS_SYMLINK_LOG_RES(mp), 0, | |
1435 | XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT); | |
1436 | } | |
1437 | if (error) { | |
1438 | cancel_flags = 0; | |
1da177e4 LT |
1439 | goto error_return; |
1440 | } | |
1441 | ||
f7c66ce3 | 1442 | xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT); |
993386c1 | 1443 | unlock_dp_on_error = B_TRUE; |
1da177e4 LT |
1444 | |
1445 | /* | |
1446 | * Check whether the directory allows new symlinks or not. | |
1447 | */ | |
1448 | if (dp->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) { | |
1449 | error = XFS_ERROR(EPERM); | |
1450 | goto error_return; | |
1451 | } | |
1452 | ||
1453 | /* | |
1454 | * Reserve disk quota : blocks and inode. | |
1455 | */ | |
7d095257 | 1456 | error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp, resblks, 1, 0); |
1da177e4 LT |
1457 | if (error) |
1458 | goto error_return; | |
1459 | ||
1460 | /* | |
1461 | * Check for ability to enter directory entry, if no space reserved. | |
1462 | */ | |
556b8b16 BN |
1463 | error = xfs_dir_canenter(tp, dp, link_name, resblks); |
1464 | if (error) | |
1da177e4 LT |
1465 | goto error_return; |
1466 | /* | |
1467 | * Initialize the bmap freelist prior to calling either | |
1468 | * bmapi or the directory create code. | |
1469 | */ | |
9d87c319 | 1470 | xfs_bmap_init(&free_list, &first_block); |
1da177e4 LT |
1471 | |
1472 | /* | |
1473 | * Allocate an inode for the symlink. | |
1474 | */ | |
6c77b0ea CH |
1475 | error = xfs_dir_ialloc(&tp, dp, S_IFLNK | (mode & ~S_IFMT), 1, 0, |
1476 | prid, resblks > 0, &ip, NULL); | |
1da177e4 LT |
1477 | if (error) { |
1478 | if (error == ENOSPC) | |
1479 | goto error_return; | |
1480 | goto error1; | |
1481 | } | |
1da177e4 | 1482 | |
993386c1 CH |
1483 | /* |
1484 | * An error after we've joined dp to the transaction will result in the | |
1485 | * transaction cancel unlocking dp so don't do it explicitly in the | |
1486 | * error path. | |
1487 | */ | |
ddc3415a | 1488 | xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); |
993386c1 | 1489 | unlock_dp_on_error = B_FALSE; |
1da177e4 LT |
1490 | |
1491 | /* | |
1492 | * Also attach the dquot(s) to it, if applicable. | |
1493 | */ | |
7d095257 | 1494 | xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp); |
1da177e4 LT |
1495 | |
1496 | if (resblks) | |
1497 | resblks -= XFS_IALLOC_SPACE_RES(mp); | |
1498 | /* | |
1499 | * If the symlink will fit into the inode, write it inline. | |
1500 | */ | |
1501 | if (pathlen <= XFS_IFORK_DSIZE(ip)) { | |
1502 | xfs_idata_realloc(ip, pathlen, XFS_DATA_FORK); | |
1503 | memcpy(ip->i_df.if_u1.if_data, target_path, pathlen); | |
1504 | ip->i_d.di_size = pathlen; | |
1505 | ||
1506 | /* | |
1507 | * The inode was initially created in extent format. | |
1508 | */ | |
1509 | ip->i_df.if_flags &= ~(XFS_IFEXTENTS | XFS_IFBROOT); | |
1510 | ip->i_df.if_flags |= XFS_IFINLINE; | |
1511 | ||
1512 | ip->i_d.di_format = XFS_DINODE_FMT_LOCAL; | |
1513 | xfs_trans_log_inode(tp, ip, XFS_ILOG_DDATA | XFS_ILOG_CORE); | |
1514 | ||
1515 | } else { | |
1516 | first_fsb = 0; | |
1517 | nmaps = SYMLINK_MAPS; | |
1518 | ||
c0dc7828 DC |
1519 | error = xfs_bmapi_write(tp, ip, first_fsb, fs_blocks, |
1520 | XFS_BMAPI_METADATA, &first_block, resblks, | |
1521 | mval, &nmaps, &free_list); | |
ec3ba85f CH |
1522 | if (error) |
1523 | goto error2; | |
1da177e4 LT |
1524 | |
1525 | if (resblks) | |
1526 | resblks -= fs_blocks; | |
1527 | ip->i_d.di_size = pathlen; | |
1528 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
1529 | ||
1530 | cur_chunk = target_path; | |
1531 | for (n = 0; n < nmaps; n++) { | |
1532 | d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock); | |
1533 | byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount); | |
1534 | bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d, | |
1535 | BTOBB(byte_cnt), 0); | |
2a30f36d CS |
1536 | if (!bp) { |
1537 | error = ENOMEM; | |
1538 | goto error2; | |
1539 | } | |
1da177e4 LT |
1540 | if (pathlen < byte_cnt) { |
1541 | byte_cnt = pathlen; | |
1542 | } | |
1543 | pathlen -= byte_cnt; | |
1544 | ||
62926044 | 1545 | memcpy(bp->b_addr, cur_chunk, byte_cnt); |
1da177e4 LT |
1546 | cur_chunk += byte_cnt; |
1547 | ||
1548 | xfs_trans_log_buf(tp, bp, 0, byte_cnt - 1); | |
1549 | } | |
1550 | } | |
1551 | ||
1552 | /* | |
1553 | * Create the directory entry for the symlink. | |
1554 | */ | |
556b8b16 BN |
1555 | error = xfs_dir_createname(tp, dp, link_name, ip->i_ino, |
1556 | &first_block, &free_list, resblks); | |
f6c2d1fa | 1557 | if (error) |
ec3ba85f | 1558 | goto error2; |
dcd79a14 | 1559 | xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); |
1da177e4 LT |
1560 | xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); |
1561 | ||
1da177e4 LT |
1562 | /* |
1563 | * If this is a synchronous mount, make sure that the | |
1564 | * symlink transaction goes to disk before returning to | |
1565 | * the user. | |
1566 | */ | |
1567 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) { | |
1568 | xfs_trans_set_sync(tp); | |
1569 | } | |
1570 | ||
f7c99b6f | 1571 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
1da177e4 LT |
1572 | if (error) { |
1573 | goto error2; | |
1574 | } | |
1c72bf90 | 1575 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
7d095257 CH |
1576 | xfs_qm_dqrele(udqp); |
1577 | xfs_qm_dqrele(gdqp); | |
1da177e4 | 1578 | |
288699fe CH |
1579 | *ipp = ip; |
1580 | return 0; | |
1da177e4 LT |
1581 | |
1582 | error2: | |
1583 | IRELE(ip); | |
1584 | error1: | |
1585 | xfs_bmap_cancel(&free_list); | |
1586 | cancel_flags |= XFS_TRANS_ABORT; | |
1587 | error_return: | |
1588 | xfs_trans_cancel(tp, cancel_flags); | |
7d095257 CH |
1589 | xfs_qm_dqrele(udqp); |
1590 | xfs_qm_dqrele(gdqp); | |
1da177e4 | 1591 | |
993386c1 | 1592 | if (unlock_dp_on_error) |
1da177e4 | 1593 | xfs_iunlock(dp, XFS_ILOCK_EXCL); |
288699fe CH |
1594 | std_return: |
1595 | return error; | |
1da177e4 LT |
1596 | } |
1597 | ||
1da177e4 | 1598 | int |
993386c1 CH |
1599 | xfs_set_dmattrs( |
1600 | xfs_inode_t *ip, | |
1da177e4 | 1601 | u_int evmask, |
993386c1 | 1602 | u_int16_t state) |
1da177e4 | 1603 | { |
993386c1 | 1604 | xfs_mount_t *mp = ip->i_mount; |
1da177e4 | 1605 | xfs_trans_t *tp; |
1da177e4 LT |
1606 | int error; |
1607 | ||
1608 | if (!capable(CAP_SYS_ADMIN)) | |
1609 | return XFS_ERROR(EPERM); | |
1610 | ||
1da177e4 LT |
1611 | if (XFS_FORCED_SHUTDOWN(mp)) |
1612 | return XFS_ERROR(EIO); | |
1613 | ||
1614 | tp = xfs_trans_alloc(mp, XFS_TRANS_SET_DMATTRS); | |
1615 | error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES (mp), 0, 0, 0); | |
1616 | if (error) { | |
1617 | xfs_trans_cancel(tp, 0); | |
1618 | return error; | |
1619 | } | |
1620 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
ddc3415a | 1621 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); |
1da177e4 | 1622 | |
613d7043 CH |
1623 | ip->i_d.di_dmevmask = evmask; |
1624 | ip->i_d.di_dmstate = state; | |
1da177e4 LT |
1625 | |
1626 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
1c72bf90 | 1627 | error = xfs_trans_commit(tp, 0); |
1da177e4 LT |
1628 | |
1629 | return error; | |
1630 | } | |
1631 | ||
1da177e4 LT |
1632 | /* |
1633 | * xfs_alloc_file_space() | |
1634 | * This routine allocates disk space for the given file. | |
1635 | * | |
1636 | * If alloc_type == 0, this request is for an ALLOCSP type | |
1637 | * request which will change the file size. In this case, no | |
1638 | * DMAPI event will be generated by the call. A TRUNCATE event | |
1639 | * will be generated later by xfs_setattr. | |
1640 | * | |
1641 | * If alloc_type != 0, this request is for a RESVSP type | |
1642 | * request, and a DMAPI DM_EVENT_WRITE will be generated if the | |
1643 | * lower block boundary byte address is less than the file's | |
1644 | * length. | |
1645 | * | |
1646 | * RETURNS: | |
1647 | * 0 on success | |
1648 | * errno on error | |
1649 | * | |
1650 | */ | |
ba0f32d4 | 1651 | STATIC int |
1da177e4 LT |
1652 | xfs_alloc_file_space( |
1653 | xfs_inode_t *ip, | |
1654 | xfs_off_t offset, | |
1655 | xfs_off_t len, | |
1656 | int alloc_type, | |
1657 | int attr_flags) | |
1658 | { | |
dd9f438e NS |
1659 | xfs_mount_t *mp = ip->i_mount; |
1660 | xfs_off_t count; | |
1da177e4 LT |
1661 | xfs_filblks_t allocated_fsb; |
1662 | xfs_filblks_t allocatesize_fsb; | |
dd9f438e NS |
1663 | xfs_extlen_t extsz, temp; |
1664 | xfs_fileoff_t startoffset_fsb; | |
1da177e4 | 1665 | xfs_fsblock_t firstfsb; |
dd9f438e | 1666 | int nimaps; |
dd9f438e | 1667 | int quota_flag; |
1da177e4 | 1668 | int rt; |
1da177e4 | 1669 | xfs_trans_t *tp; |
dd9f438e NS |
1670 | xfs_bmbt_irec_t imaps[1], *imapp; |
1671 | xfs_bmap_free_t free_list; | |
1672 | uint qblocks, resblks, resrtextents; | |
1673 | int committed; | |
1674 | int error; | |
1da177e4 | 1675 | |
cca28fb8 | 1676 | trace_xfs_alloc_file_space(ip); |
1da177e4 LT |
1677 | |
1678 | if (XFS_FORCED_SHUTDOWN(mp)) | |
1679 | return XFS_ERROR(EIO); | |
1680 | ||
7d095257 CH |
1681 | error = xfs_qm_dqattach(ip, 0); |
1682 | if (error) | |
1da177e4 LT |
1683 | return error; |
1684 | ||
1685 | if (len <= 0) | |
1686 | return XFS_ERROR(EINVAL); | |
1687 | ||
957d0ebe DC |
1688 | rt = XFS_IS_REALTIME_INODE(ip); |
1689 | extsz = xfs_get_extsz_hint(ip); | |
1690 | ||
1da177e4 | 1691 | count = len; |
1da177e4 | 1692 | imapp = &imaps[0]; |
dd9f438e | 1693 | nimaps = 1; |
1da177e4 LT |
1694 | startoffset_fsb = XFS_B_TO_FSBT(mp, offset); |
1695 | allocatesize_fsb = XFS_B_TO_FSB(mp, count); | |
1696 | ||
1da177e4 | 1697 | /* |
dd9f438e | 1698 | * Allocate file space until done or until there is an error |
1da177e4 | 1699 | */ |
1da177e4 | 1700 | while (allocatesize_fsb && !error) { |
dd9f438e NS |
1701 | xfs_fileoff_t s, e; |
1702 | ||
1da177e4 | 1703 | /* |
3ddb8fa9 | 1704 | * Determine space reservations for data/realtime. |
1da177e4 | 1705 | */ |
dd9f438e | 1706 | if (unlikely(extsz)) { |
1da177e4 | 1707 | s = startoffset_fsb; |
dd9f438e NS |
1708 | do_div(s, extsz); |
1709 | s *= extsz; | |
1710 | e = startoffset_fsb + allocatesize_fsb; | |
1711 | if ((temp = do_mod(startoffset_fsb, extsz))) | |
1712 | e += temp; | |
1713 | if ((temp = do_mod(e, extsz))) | |
1714 | e += extsz - temp; | |
1715 | } else { | |
1716 | s = 0; | |
1717 | e = allocatesize_fsb; | |
1718 | } | |
1719 | ||
72656c46 DC |
1720 | /* |
1721 | * The transaction reservation is limited to a 32-bit block | |
1722 | * count, hence we need to limit the number of blocks we are | |
1723 | * trying to reserve to avoid an overflow. We can't allocate | |
1724 | * more than @nimaps extents, and an extent is limited on disk | |
1725 | * to MAXEXTLEN (21 bits), so use that to enforce the limit. | |
1726 | */ | |
1727 | resblks = min_t(xfs_fileoff_t, (e - s), (MAXEXTLEN * nimaps)); | |
dd9f438e | 1728 | if (unlikely(rt)) { |
72656c46 | 1729 | resrtextents = qblocks = resblks; |
dd9f438e NS |
1730 | resrtextents /= mp->m_sb.sb_rextsize; |
1731 | resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0); | |
1732 | quota_flag = XFS_QMOPT_RES_RTBLKS; | |
1da177e4 | 1733 | } else { |
dd9f438e | 1734 | resrtextents = 0; |
72656c46 | 1735 | resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resblks); |
dd9f438e | 1736 | quota_flag = XFS_QMOPT_RES_REGBLKS; |
1da177e4 LT |
1737 | } |
1738 | ||
1739 | /* | |
dd9f438e | 1740 | * Allocate and setup the transaction. |
1da177e4 LT |
1741 | */ |
1742 | tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT); | |
dd9f438e NS |
1743 | error = xfs_trans_reserve(tp, resblks, |
1744 | XFS_WRITE_LOG_RES(mp), resrtextents, | |
1da177e4 LT |
1745 | XFS_TRANS_PERM_LOG_RES, |
1746 | XFS_WRITE_LOG_COUNT); | |
1da177e4 | 1747 | /* |
dd9f438e | 1748 | * Check for running out of space |
1da177e4 LT |
1749 | */ |
1750 | if (error) { | |
1751 | /* | |
1752 | * Free the transaction structure. | |
1753 | */ | |
1754 | ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp)); | |
1755 | xfs_trans_cancel(tp, 0); | |
1756 | break; | |
1757 | } | |
1758 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
7d095257 CH |
1759 | error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks, |
1760 | 0, quota_flag); | |
1da177e4 LT |
1761 | if (error) |
1762 | goto error1; | |
1763 | ||
ddc3415a | 1764 | xfs_trans_ijoin(tp, ip, 0); |
1da177e4 | 1765 | |
9d87c319 | 1766 | xfs_bmap_init(&free_list, &firstfsb); |
c0dc7828 DC |
1767 | error = xfs_bmapi_write(tp, ip, startoffset_fsb, |
1768 | allocatesize_fsb, alloc_type, &firstfsb, | |
1769 | 0, imapp, &nimaps, &free_list); | |
1da177e4 LT |
1770 | if (error) { |
1771 | goto error0; | |
1772 | } | |
1773 | ||
1774 | /* | |
dd9f438e | 1775 | * Complete the transaction |
1da177e4 | 1776 | */ |
f7c99b6f | 1777 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
1da177e4 LT |
1778 | if (error) { |
1779 | goto error0; | |
1780 | } | |
1781 | ||
1c72bf90 | 1782 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 LT |
1783 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
1784 | if (error) { | |
1785 | break; | |
1786 | } | |
1787 | ||
1788 | allocated_fsb = imapp->br_blockcount; | |
1789 | ||
dd9f438e | 1790 | if (nimaps == 0) { |
1da177e4 LT |
1791 | error = XFS_ERROR(ENOSPC); |
1792 | break; | |
1793 | } | |
1794 | ||
1795 | startoffset_fsb += allocated_fsb; | |
1796 | allocatesize_fsb -= allocated_fsb; | |
1797 | } | |
1da177e4 LT |
1798 | |
1799 | return error; | |
1800 | ||
dd9f438e | 1801 | error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */ |
1da177e4 | 1802 | xfs_bmap_cancel(&free_list); |
ea562ed6 | 1803 | xfs_trans_unreserve_quota_nblks(tp, ip, (long)qblocks, 0, quota_flag); |
dd9f438e NS |
1804 | |
1805 | error1: /* Just cancel transaction */ | |
1da177e4 LT |
1806 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT); |
1807 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
288699fe | 1808 | return error; |
1da177e4 LT |
1809 | } |
1810 | ||
1811 | /* | |
1812 | * Zero file bytes between startoff and endoff inclusive. | |
1813 | * The iolock is held exclusive and no blocks are buffered. | |
2fd6f6ec LM |
1814 | * |
1815 | * This function is used by xfs_free_file_space() to zero | |
1816 | * partial blocks when the range to free is not block aligned. | |
1817 | * When unreserving space with boundaries that are not block | |
1818 | * aligned we round up the start and round down the end | |
1819 | * boundaries and then use this function to zero the parts of | |
1820 | * the blocks that got dropped during the rounding. | |
1da177e4 LT |
1821 | */ |
1822 | STATIC int | |
1823 | xfs_zero_remaining_bytes( | |
1824 | xfs_inode_t *ip, | |
1825 | xfs_off_t startoff, | |
1826 | xfs_off_t endoff) | |
1827 | { | |
1828 | xfs_bmbt_irec_t imap; | |
1829 | xfs_fileoff_t offset_fsb; | |
1830 | xfs_off_t lastoffset; | |
1831 | xfs_off_t offset; | |
1832 | xfs_buf_t *bp; | |
1833 | xfs_mount_t *mp = ip->i_mount; | |
1834 | int nimap; | |
1835 | int error = 0; | |
1836 | ||
2fd6f6ec LM |
1837 | /* |
1838 | * Avoid doing I/O beyond eof - it's not necessary | |
1839 | * since nothing can read beyond eof. The space will | |
1840 | * be zeroed when the file is extended anyway. | |
1841 | */ | |
ce7ae151 | 1842 | if (startoff >= XFS_ISIZE(ip)) |
2fd6f6ec LM |
1843 | return 0; |
1844 | ||
ce7ae151 CH |
1845 | if (endoff > XFS_ISIZE(ip)) |
1846 | endoff = XFS_ISIZE(ip); | |
2fd6f6ec | 1847 | |
686865f7 DC |
1848 | bp = xfs_buf_get_uncached(XFS_IS_REALTIME_INODE(ip) ? |
1849 | mp->m_rtdev_targp : mp->m_ddev_targp, | |
aa5c158e | 1850 | BTOBB(mp->m_sb.sb_blocksize), 0); |
c6422617 LM |
1851 | if (!bp) |
1852 | return XFS_ERROR(ENOMEM); | |
1da177e4 | 1853 | |
c8da0faf CH |
1854 | xfs_buf_unlock(bp); |
1855 | ||
1da177e4 LT |
1856 | for (offset = startoff; offset <= endoff; offset = lastoffset + 1) { |
1857 | offset_fsb = XFS_B_TO_FSBT(mp, offset); | |
1858 | nimap = 1; | |
5c8ed202 | 1859 | error = xfs_bmapi_read(ip, offset_fsb, 1, &imap, &nimap, 0); |
1da177e4 LT |
1860 | if (error || nimap < 1) |
1861 | break; | |
1862 | ASSERT(imap.br_blockcount >= 1); | |
1863 | ASSERT(imap.br_startoff == offset_fsb); | |
1864 | lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1; | |
1865 | if (lastoffset > endoff) | |
1866 | lastoffset = endoff; | |
1867 | if (imap.br_startblock == HOLESTARTBLOCK) | |
1868 | continue; | |
1869 | ASSERT(imap.br_startblock != DELAYSTARTBLOCK); | |
1870 | if (imap.br_state == XFS_EXT_UNWRITTEN) | |
1871 | continue; | |
1872 | XFS_BUF_UNDONE(bp); | |
1873 | XFS_BUF_UNWRITE(bp); | |
1874 | XFS_BUF_READ(bp); | |
9d87c319 | 1875 | XFS_BUF_SET_ADDR(bp, xfs_fsb_to_db(ip, imap.br_startblock)); |
1da177e4 | 1876 | xfsbdstrat(mp, bp); |
1a1a3e97 | 1877 | error = xfs_buf_iowait(bp); |
d64e31a2 | 1878 | if (error) { |
901796af CH |
1879 | xfs_buf_ioerror_alert(bp, |
1880 | "xfs_zero_remaining_bytes(read)"); | |
1da177e4 LT |
1881 | break; |
1882 | } | |
62926044 | 1883 | memset(bp->b_addr + |
1da177e4 LT |
1884 | (offset - XFS_FSB_TO_B(mp, imap.br_startoff)), |
1885 | 0, lastoffset - offset + 1); | |
1886 | XFS_BUF_UNDONE(bp); | |
1887 | XFS_BUF_UNREAD(bp); | |
1888 | XFS_BUF_WRITE(bp); | |
1889 | xfsbdstrat(mp, bp); | |
1a1a3e97 | 1890 | error = xfs_buf_iowait(bp); |
d64e31a2 | 1891 | if (error) { |
901796af CH |
1892 | xfs_buf_ioerror_alert(bp, |
1893 | "xfs_zero_remaining_bytes(write)"); | |
1da177e4 LT |
1894 | break; |
1895 | } | |
1896 | } | |
1897 | xfs_buf_free(bp); | |
1898 | return error; | |
1899 | } | |
1900 | ||
1901 | /* | |
1902 | * xfs_free_file_space() | |
1903 | * This routine frees disk space for the given file. | |
1904 | * | |
1905 | * This routine is only called by xfs_change_file_space | |
1906 | * for an UNRESVSP type call. | |
1907 | * | |
1908 | * RETURNS: | |
1909 | * 0 on success | |
1910 | * errno on error | |
1911 | * | |
1912 | */ | |
1913 | STATIC int | |
1914 | xfs_free_file_space( | |
1915 | xfs_inode_t *ip, | |
1916 | xfs_off_t offset, | |
1917 | xfs_off_t len, | |
1918 | int attr_flags) | |
1919 | { | |
1920 | int committed; | |
1921 | int done; | |
1da177e4 LT |
1922 | xfs_fileoff_t endoffset_fsb; |
1923 | int error; | |
1924 | xfs_fsblock_t firstfsb; | |
1925 | xfs_bmap_free_t free_list; | |
1da177e4 LT |
1926 | xfs_bmbt_irec_t imap; |
1927 | xfs_off_t ioffset; | |
1928 | xfs_extlen_t mod=0; | |
1929 | xfs_mount_t *mp; | |
1930 | int nimap; | |
1931 | uint resblks; | |
673cdf5c | 1932 | uint rounding; |
1da177e4 LT |
1933 | int rt; |
1934 | xfs_fileoff_t startoffset_fsb; | |
1935 | xfs_trans_t *tp; | |
5fcbab35 | 1936 | int need_iolock = 1; |
1da177e4 | 1937 | |
1da177e4 LT |
1938 | mp = ip->i_mount; |
1939 | ||
cca28fb8 | 1940 | trace_xfs_free_file_space(ip); |
bd5a876a | 1941 | |
7d095257 CH |
1942 | error = xfs_qm_dqattach(ip, 0); |
1943 | if (error) | |
1da177e4 LT |
1944 | return error; |
1945 | ||
1946 | error = 0; | |
1947 | if (len <= 0) /* if nothing being freed */ | |
1948 | return error; | |
71ddabb9 | 1949 | rt = XFS_IS_REALTIME_INODE(ip); |
1da177e4 | 1950 | startoffset_fsb = XFS_B_TO_FSB(mp, offset); |
288699fe | 1951 | endoffset_fsb = XFS_B_TO_FSBT(mp, offset + len); |
1da177e4 | 1952 | |
0f285c8a | 1953 | if (attr_flags & XFS_ATTR_NOLOCK) |
5fcbab35 | 1954 | need_iolock = 0; |
9fa8046f | 1955 | if (need_iolock) { |
1da177e4 | 1956 | xfs_ilock(ip, XFS_IOLOCK_EXCL); |
25e41b3d | 1957 | /* wait for the completion of any pending DIOs */ |
4a06fd26 | 1958 | inode_dio_wait(VFS_I(ip)); |
9fa8046f | 1959 | } |
5fcbab35 | 1960 | |
e6a4b37f | 1961 | rounding = max_t(uint, 1 << mp->m_sb.sb_blocklog, PAGE_CACHE_SIZE); |
1da177e4 | 1962 | ioffset = offset & ~(rounding - 1); |
bd5a876a | 1963 | |
df80c933 | 1964 | if (VN_CACHED(VFS_I(ip)) != 0) { |
e6a4b37f | 1965 | error = xfs_flushinval_pages(ip, ioffset, -1, FI_REMAPF_LOCKED); |
d3cf2094 LM |
1966 | if (error) |
1967 | goto out_unlock_iolock; | |
bd5a876a CH |
1968 | } |
1969 | ||
1da177e4 LT |
1970 | /* |
1971 | * Need to zero the stuff we're not freeing, on disk. | |
9da096fd | 1972 | * If it's a realtime file & can't use unwritten extents then we |
1da177e4 LT |
1973 | * actually need to zero the extent edges. Otherwise xfs_bunmapi |
1974 | * will take care of it for us. | |
1975 | */ | |
62118709 | 1976 | if (rt && !xfs_sb_version_hasextflgbit(&mp->m_sb)) { |
1da177e4 | 1977 | nimap = 1; |
5c8ed202 DC |
1978 | error = xfs_bmapi_read(ip, startoffset_fsb, 1, |
1979 | &imap, &nimap, 0); | |
1da177e4 LT |
1980 | if (error) |
1981 | goto out_unlock_iolock; | |
1982 | ASSERT(nimap == 0 || nimap == 1); | |
1983 | if (nimap && imap.br_startblock != HOLESTARTBLOCK) { | |
1984 | xfs_daddr_t block; | |
1985 | ||
1986 | ASSERT(imap.br_startblock != DELAYSTARTBLOCK); | |
1987 | block = imap.br_startblock; | |
1988 | mod = do_div(block, mp->m_sb.sb_rextsize); | |
1989 | if (mod) | |
1990 | startoffset_fsb += mp->m_sb.sb_rextsize - mod; | |
1991 | } | |
1992 | nimap = 1; | |
5c8ed202 DC |
1993 | error = xfs_bmapi_read(ip, endoffset_fsb - 1, 1, |
1994 | &imap, &nimap, 0); | |
1da177e4 LT |
1995 | if (error) |
1996 | goto out_unlock_iolock; | |
1997 | ASSERT(nimap == 0 || nimap == 1); | |
1998 | if (nimap && imap.br_startblock != HOLESTARTBLOCK) { | |
1999 | ASSERT(imap.br_startblock != DELAYSTARTBLOCK); | |
2000 | mod++; | |
2001 | if (mod && (mod != mp->m_sb.sb_rextsize)) | |
2002 | endoffset_fsb -= mod; | |
2003 | } | |
2004 | } | |
2005 | if ((done = (endoffset_fsb <= startoffset_fsb))) | |
2006 | /* | |
2007 | * One contiguous piece to clear | |
2008 | */ | |
2009 | error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1); | |
2010 | else { | |
2011 | /* | |
2012 | * Some full blocks, possibly two pieces to clear | |
2013 | */ | |
2014 | if (offset < XFS_FSB_TO_B(mp, startoffset_fsb)) | |
2015 | error = xfs_zero_remaining_bytes(ip, offset, | |
2016 | XFS_FSB_TO_B(mp, startoffset_fsb) - 1); | |
2017 | if (!error && | |
2018 | XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len) | |
2019 | error = xfs_zero_remaining_bytes(ip, | |
2020 | XFS_FSB_TO_B(mp, endoffset_fsb), | |
2021 | offset + len - 1); | |
2022 | } | |
2023 | ||
2024 | /* | |
2025 | * free file space until done or until there is an error | |
2026 | */ | |
2027 | resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0); | |
2028 | while (!error && !done) { | |
2029 | ||
2030 | /* | |
91ebecc7 DC |
2031 | * allocate and setup the transaction. Allow this |
2032 | * transaction to dip into the reserve blocks to ensure | |
2033 | * the freeing of the space succeeds at ENOSPC. | |
1da177e4 LT |
2034 | */ |
2035 | tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT); | |
91ebecc7 | 2036 | tp->t_flags |= XFS_TRANS_RESERVE; |
1da177e4 LT |
2037 | error = xfs_trans_reserve(tp, |
2038 | resblks, | |
2039 | XFS_WRITE_LOG_RES(mp), | |
2040 | 0, | |
2041 | XFS_TRANS_PERM_LOG_RES, | |
2042 | XFS_WRITE_LOG_COUNT); | |
2043 | ||
2044 | /* | |
2045 | * check for running out of space | |
2046 | */ | |
2047 | if (error) { | |
2048 | /* | |
2049 | * Free the transaction structure. | |
2050 | */ | |
2051 | ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp)); | |
2052 | xfs_trans_cancel(tp, 0); | |
2053 | break; | |
2054 | } | |
2055 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
7d095257 CH |
2056 | error = xfs_trans_reserve_quota(tp, mp, |
2057 | ip->i_udquot, ip->i_gdquot, | |
2058 | resblks, 0, XFS_QMOPT_RES_REGBLKS); | |
1da177e4 LT |
2059 | if (error) |
2060 | goto error1; | |
2061 | ||
ddc3415a | 2062 | xfs_trans_ijoin(tp, ip, 0); |
1da177e4 LT |
2063 | |
2064 | /* | |
2065 | * issue the bunmapi() call to free the blocks | |
2066 | */ | |
9d87c319 | 2067 | xfs_bmap_init(&free_list, &firstfsb); |
541d7d3c | 2068 | error = xfs_bunmapi(tp, ip, startoffset_fsb, |
1da177e4 | 2069 | endoffset_fsb - startoffset_fsb, |
b4e9181e | 2070 | 0, 2, &firstfsb, &free_list, &done); |
1da177e4 LT |
2071 | if (error) { |
2072 | goto error0; | |
2073 | } | |
2074 | ||
2075 | /* | |
2076 | * complete the transaction | |
2077 | */ | |
f7c99b6f | 2078 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
1da177e4 LT |
2079 | if (error) { |
2080 | goto error0; | |
2081 | } | |
2082 | ||
1c72bf90 | 2083 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 LT |
2084 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
2085 | } | |
2086 | ||
2087 | out_unlock_iolock: | |
2088 | if (need_iolock) | |
2089 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
2090 | return error; | |
2091 | ||
2092 | error0: | |
2093 | xfs_bmap_cancel(&free_list); | |
2094 | error1: | |
2095 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT); | |
2096 | xfs_iunlock(ip, need_iolock ? (XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL) : | |
2097 | XFS_ILOCK_EXCL); | |
2098 | return error; | |
2099 | } | |
2100 | ||
2101 | /* | |
2102 | * xfs_change_file_space() | |
2103 | * This routine allocates or frees disk space for the given file. | |
2104 | * The user specified parameters are checked for alignment and size | |
2105 | * limitations. | |
2106 | * | |
2107 | * RETURNS: | |
2108 | * 0 on success | |
2109 | * errno on error | |
2110 | * | |
2111 | */ | |
2112 | int | |
2113 | xfs_change_file_space( | |
993386c1 | 2114 | xfs_inode_t *ip, |
1da177e4 LT |
2115 | int cmd, |
2116 | xfs_flock64_t *bf, | |
2117 | xfs_off_t offset, | |
1da177e4 LT |
2118 | int attr_flags) |
2119 | { | |
993386c1 | 2120 | xfs_mount_t *mp = ip->i_mount; |
1da177e4 LT |
2121 | int clrprealloc; |
2122 | int error; | |
2123 | xfs_fsize_t fsize; | |
1da177e4 LT |
2124 | int setprealloc; |
2125 | xfs_off_t startoffset; | |
2126 | xfs_off_t llen; | |
2127 | xfs_trans_t *tp; | |
0f285c8a | 2128 | struct iattr iattr; |
44722352 | 2129 | int prealloc_type; |
1da177e4 | 2130 | |
993386c1 | 2131 | if (!S_ISREG(ip->i_d.di_mode)) |
1da177e4 LT |
2132 | return XFS_ERROR(EINVAL); |
2133 | ||
1da177e4 LT |
2134 | switch (bf->l_whence) { |
2135 | case 0: /*SEEK_SET*/ | |
2136 | break; | |
2137 | case 1: /*SEEK_CUR*/ | |
2138 | bf->l_start += offset; | |
2139 | break; | |
2140 | case 2: /*SEEK_END*/ | |
ce7ae151 | 2141 | bf->l_start += XFS_ISIZE(ip); |
1da177e4 LT |
2142 | break; |
2143 | default: | |
2144 | return XFS_ERROR(EINVAL); | |
2145 | } | |
2146 | ||
2147 | llen = bf->l_len > 0 ? bf->l_len - 1 : bf->l_len; | |
2148 | ||
32972383 DC |
2149 | if (bf->l_start < 0 || |
2150 | bf->l_start > mp->m_super->s_maxbytes || | |
2151 | bf->l_start + llen < 0 || | |
2152 | bf->l_start + llen > mp->m_super->s_maxbytes) | |
1da177e4 LT |
2153 | return XFS_ERROR(EINVAL); |
2154 | ||
2155 | bf->l_whence = 0; | |
2156 | ||
2157 | startoffset = bf->l_start; | |
ce7ae151 | 2158 | fsize = XFS_ISIZE(ip); |
1da177e4 LT |
2159 | |
2160 | /* | |
2161 | * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve | |
2162 | * file space. | |
2163 | * These calls do NOT zero the data space allocated to the file, | |
2164 | * nor do they change the file size. | |
2165 | * | |
2166 | * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file | |
2167 | * space. | |
2168 | * These calls cause the new file data to be zeroed and the file | |
2169 | * size to be changed. | |
2170 | */ | |
2171 | setprealloc = clrprealloc = 0; | |
44722352 | 2172 | prealloc_type = XFS_BMAPI_PREALLOC; |
1da177e4 LT |
2173 | |
2174 | switch (cmd) { | |
44722352 DC |
2175 | case XFS_IOC_ZERO_RANGE: |
2176 | prealloc_type |= XFS_BMAPI_CONVERT; | |
2177 | xfs_tosspages(ip, startoffset, startoffset + bf->l_len, 0); | |
2178 | /* FALLTHRU */ | |
1da177e4 LT |
2179 | case XFS_IOC_RESVSP: |
2180 | case XFS_IOC_RESVSP64: | |
2181 | error = xfs_alloc_file_space(ip, startoffset, bf->l_len, | |
44722352 | 2182 | prealloc_type, attr_flags); |
1da177e4 LT |
2183 | if (error) |
2184 | return error; | |
2185 | setprealloc = 1; | |
2186 | break; | |
2187 | ||
2188 | case XFS_IOC_UNRESVSP: | |
2189 | case XFS_IOC_UNRESVSP64: | |
2190 | if ((error = xfs_free_file_space(ip, startoffset, bf->l_len, | |
2191 | attr_flags))) | |
2192 | return error; | |
2193 | break; | |
2194 | ||
2195 | case XFS_IOC_ALLOCSP: | |
2196 | case XFS_IOC_ALLOCSP64: | |
2197 | case XFS_IOC_FREESP: | |
2198 | case XFS_IOC_FREESP64: | |
bc4010ec DC |
2199 | /* |
2200 | * These operations actually do IO when extending the file, but | |
2201 | * the allocation is done seperately to the zeroing that is | |
2202 | * done. This set of operations need to be serialised against | |
2203 | * other IO operations, such as truncate and buffered IO. We | |
2204 | * need to take the IOLOCK here to serialise the allocation and | |
2205 | * zeroing IO to prevent other IOLOCK holders (e.g. getbmap, | |
2206 | * truncate, direct IO) from racing against the transient | |
2207 | * allocated but not written state we can have here. | |
2208 | */ | |
2209 | xfs_ilock(ip, XFS_IOLOCK_EXCL); | |
1da177e4 LT |
2210 | if (startoffset > fsize) { |
2211 | error = xfs_alloc_file_space(ip, fsize, | |
bc4010ec DC |
2212 | startoffset - fsize, 0, |
2213 | attr_flags | XFS_ATTR_NOLOCK); | |
2214 | if (error) { | |
2215 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
1da177e4 | 2216 | break; |
bc4010ec | 2217 | } |
1da177e4 LT |
2218 | } |
2219 | ||
0f285c8a CH |
2220 | iattr.ia_valid = ATTR_SIZE; |
2221 | iattr.ia_size = startoffset; | |
1da177e4 | 2222 | |
bc4010ec DC |
2223 | error = xfs_setattr_size(ip, &iattr, |
2224 | attr_flags | XFS_ATTR_NOLOCK); | |
2225 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
1da177e4 LT |
2226 | |
2227 | if (error) | |
2228 | return error; | |
2229 | ||
2230 | clrprealloc = 1; | |
2231 | break; | |
2232 | ||
2233 | default: | |
2234 | ASSERT(0); | |
2235 | return XFS_ERROR(EINVAL); | |
2236 | } | |
2237 | ||
2238 | /* | |
2239 | * update the inode timestamp, mode, and prealloc flag bits | |
2240 | */ | |
2241 | tp = xfs_trans_alloc(mp, XFS_TRANS_WRITEID); | |
2242 | ||
2243 | if ((error = xfs_trans_reserve(tp, 0, XFS_WRITEID_LOG_RES(mp), | |
2244 | 0, 0, 0))) { | |
2245 | /* ASSERT(0); */ | |
2246 | xfs_trans_cancel(tp, 0); | |
2247 | return error; | |
2248 | } | |
2249 | ||
2250 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
ddc3415a | 2251 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); |
1da177e4 | 2252 | |
0f285c8a | 2253 | if ((attr_flags & XFS_ATTR_DMI) == 0) { |
1da177e4 LT |
2254 | ip->i_d.di_mode &= ~S_ISUID; |
2255 | ||
2256 | /* | |
2257 | * Note that we don't have to worry about mandatory | |
2258 | * file locking being disabled here because we only | |
2259 | * clear the S_ISGID bit if the Group execute bit is | |
2260 | * on, but if it was on then mandatory locking wouldn't | |
2261 | * have been enabled. | |
2262 | */ | |
2263 | if (ip->i_d.di_mode & S_IXGRP) | |
2264 | ip->i_d.di_mode &= ~S_ISGID; | |
2265 | ||
dcd79a14 | 2266 | xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); |
1da177e4 LT |
2267 | } |
2268 | if (setprealloc) | |
2269 | ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC; | |
2270 | else if (clrprealloc) | |
2271 | ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC; | |
2272 | ||
2273 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
82878897 DC |
2274 | if (attr_flags & XFS_ATTR_SYNC) |
2275 | xfs_trans_set_sync(tp); | |
23bb0be1 | 2276 | return xfs_trans_commit(tp, 0); |
1da177e4 | 2277 | } |