1 /*
2 * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
4 *
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
7 * published by the Free Software Foundation.
8 *
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.
13 *
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
17 */
18 #include "xfs.h"
19 #include "xfs_fs.h"
20 #include "xfs_types.h"
21 #include "xfs_log.h"
22 #include "xfs_inum.h"
23 #include "xfs_trans.h"
24 #include "xfs_sb.h"
25 #include "xfs_dir.h"
26 #include "xfs_dir2.h"
27 #include "xfs_dmapi.h"
28 #include "xfs_mount.h"
29 #include "xfs_da_btree.h"
30 #include "xfs_bmap_btree.h"
31 #include "xfs_dir_sf.h"
32 #include "xfs_dir2_sf.h"
33 #include "xfs_attr_sf.h"
34 #include "xfs_dinode.h"
35 #include "xfs_inode.h"
36 #include "xfs_inode_item.h"
37 #include "xfs_bmap.h"
38 #include "xfs_error.h"
39 #include "xfs_quota.h"
40 #include "xfs_refcache.h"
41 #include "xfs_utils.h"
42 #include "xfs_trans_space.h"
43 #include "xfs_dir_leaf.h"
44
45
46 /*
47 * Given an array of up to 4 inode pointers, unlock the pointed to inodes.
48 * If there are fewer than 4 entries in the array, the empty entries will
49 * be at the end and will have NULL pointers in them.
50 */
51 STATIC void
xfs_rename_unlock4(xfs_inode_t ** i_tab,uint lock_mode)52 xfs_rename_unlock4(
53 xfs_inode_t **i_tab,
54 uint lock_mode)
55 {
56 int i;
57
58 xfs_iunlock(i_tab[0], lock_mode);
59 for (i = 1; i < 4; i++) {
60 if (i_tab[i] == NULL) {
61 break;
62 }
63 /*
64 * Watch out for duplicate entries in the table.
65 */
66 if (i_tab[i] != i_tab[i-1]) {
67 xfs_iunlock(i_tab[i], lock_mode);
68 }
69 }
70 }
71
72 #ifdef DEBUG
73 int xfs_rename_skip, xfs_rename_nskip;
74 #endif
75
76 /*
77 * The following routine will acquire the locks required for a rename
78 * operation. The code understands the semantics of renames and will
79 * validate that name1 exists under dp1 & that name2 may or may not
80 * exist under dp2.
81 *
82 * We are renaming dp1/name1 to dp2/name2.
83 *
84 * Return ENOENT if dp1 does not exist, other lookup errors, or 0 for success.
85 */
86 STATIC int
xfs_lock_for_rename(xfs_inode_t * dp1,xfs_inode_t * dp2,vname_t * vname1,vname_t * vname2,xfs_inode_t ** ipp1,xfs_inode_t ** ipp2,xfs_inode_t ** i_tab,int * num_inodes)87 xfs_lock_for_rename(
88 xfs_inode_t *dp1, /* old (source) directory inode */
89 xfs_inode_t *dp2, /* new (target) directory inode */
90 vname_t *vname1,/* old entry name */
91 vname_t *vname2,/* new entry name */
92 xfs_inode_t **ipp1, /* inode of old entry */
93 xfs_inode_t **ipp2, /* inode of new entry, if it
94 already exists, NULL otherwise. */
95 xfs_inode_t **i_tab,/* array of inode returned, sorted */
96 int *num_inodes) /* number of inodes in array */
97 {
98 xfs_inode_t *ip1, *ip2, *temp;
99 xfs_ino_t inum1, inum2;
100 int error;
101 int i, j;
102 uint lock_mode;
103 int diff_dirs = (dp1 != dp2);
104
105 ip2 = NULL;
106
107 /*
108 * First, find out the current inums of the entries so that we
109 * can determine the initial locking order. We'll have to
110 * sanity check stuff after all the locks have been acquired
111 * to see if we still have the right inodes, directories, etc.
112 */
113 lock_mode = xfs_ilock_map_shared(dp1);
114 error = xfs_get_dir_entry(vname1, &ip1);
115 if (error) {
116 xfs_iunlock_map_shared(dp1, lock_mode);
117 return error;
118 }
119
120 inum1 = ip1->i_ino;
121
122 ASSERT(ip1);
123 ITRACE(ip1);
124
125 /*
126 * Unlock dp1 and lock dp2 if they are different.
127 */
128
129 if (diff_dirs) {
130 xfs_iunlock_map_shared(dp1, lock_mode);
131 lock_mode = xfs_ilock_map_shared(dp2);
132 }
133
134 error = xfs_dir_lookup_int(XFS_ITOBHV(dp2), lock_mode,
135 vname2, &inum2, &ip2);
136 if (error == ENOENT) { /* target does not need to exist. */
137 inum2 = 0;
138 } else if (error) {
139 /*
140 * If dp2 and dp1 are the same, the next line unlocks dp1.
141 * Got it?
142 */
143 xfs_iunlock_map_shared(dp2, lock_mode);
144 IRELE (ip1);
145 return error;
146 } else {
147 ITRACE(ip2);
148 }
149
150 /*
151 * i_tab contains a list of pointers to inodes. We initialize
152 * the table here & we'll sort it. We will then use it to
153 * order the acquisition of the inode locks.
154 *
155 * Note that the table may contain duplicates. e.g., dp1 == dp2.
156 */
157 i_tab[0] = dp1;
158 i_tab[1] = dp2;
159 i_tab[2] = ip1;
160 if (inum2 == 0) {
161 *num_inodes = 3;
162 i_tab[3] = NULL;
163 } else {
164 *num_inodes = 4;
165 i_tab[3] = ip2;
166 }
167
168 /*
169 * Sort the elements via bubble sort. (Remember, there are at
170 * most 4 elements to sort, so this is adequate.)
171 */
172 for (i=0; i < *num_inodes; i++) {
173 for (j=1; j < *num_inodes; j++) {
174 if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) {
175 temp = i_tab[j];
176 i_tab[j] = i_tab[j-1];
177 i_tab[j-1] = temp;
178 }
179 }
180 }
181
182 /*
183 * We have dp2 locked. If it isn't first, unlock it.
184 * If it is first, tell xfs_lock_inodes so it can skip it
185 * when locking. if dp1 == dp2, xfs_lock_inodes will skip both
186 * since they are equal. xfs_lock_inodes needs all these inodes
187 * so that it can unlock and retry if there might be a dead-lock
188 * potential with the log.
189 */
190
191 if (i_tab[0] == dp2 && lock_mode == XFS_ILOCK_SHARED) {
192 #ifdef DEBUG
193 xfs_rename_skip++;
194 #endif
195 xfs_lock_inodes(i_tab, *num_inodes, 1, XFS_ILOCK_SHARED);
196 } else {
197 #ifdef DEBUG
198 xfs_rename_nskip++;
199 #endif
200 xfs_iunlock_map_shared(dp2, lock_mode);
201 xfs_lock_inodes(i_tab, *num_inodes, 0, XFS_ILOCK_SHARED);
202 }
203
204 /*
205 * Set the return value. Null out any unused entries in i_tab.
206 */
207 *ipp1 = *ipp2 = NULL;
208 for (i=0; i < *num_inodes; i++) {
209 if (i_tab[i]->i_ino == inum1) {
210 *ipp1 = i_tab[i];
211 }
212 if (i_tab[i]->i_ino == inum2) {
213 *ipp2 = i_tab[i];
214 }
215 }
216 for (;i < 4; i++) {
217 i_tab[i] = NULL;
218 }
219 return 0;
220 }
221
222 /*
223 * xfs_rename
224 */
225 int
xfs_rename(bhv_desc_t * src_dir_bdp,vname_t * src_vname,xfs_vnode_t * target_dir_vp,vname_t * target_vname,cred_t * credp)226 xfs_rename(
227 bhv_desc_t *src_dir_bdp,
228 vname_t *src_vname,
229 xfs_vnode_t *target_dir_vp,
230 vname_t *target_vname,
231 cred_t *credp)
232 {
233 xfs_trans_t *tp;
234 xfs_inode_t *src_dp, *target_dp, *src_ip, *target_ip;
235 xfs_mount_t *mp;
236 int new_parent; /* moving to a new dir */
237 int src_is_directory; /* src_name is a directory */
238 int error;
239 xfs_bmap_free_t free_list;
240 xfs_fsblock_t first_block;
241 int cancel_flags;
242 int committed;
243 xfs_inode_t *inodes[4];
244 int target_ip_dropped = 0; /* dropped target_ip link? */
245 xfs_vnode_t *src_dir_vp;
246 int spaceres;
247 int target_link_zero = 0;
248 int num_inodes;
249 char *src_name = VNAME(src_vname);
250 char *target_name = VNAME(target_vname);
251 int src_namelen = VNAMELEN(src_vname);
252 int target_namelen = VNAMELEN(target_vname);
253
254 src_dir_vp = BHV_TO_VNODE(src_dir_bdp);
255 vn_trace_entry(src_dir_vp, "xfs_rename", (inst_t *)__return_address);
256 vn_trace_entry(target_dir_vp, "xfs_rename", (inst_t *)__return_address);
257
258 /*
259 * Find the XFS behavior descriptor for the target directory
260 * vnode since it was not handed to us.
261 */
262 target_dp = xfs_vtoi(target_dir_vp);
263 if (target_dp == NULL) {
264 return XFS_ERROR(EXDEV);
265 }
266
267 src_dp = XFS_BHVTOI(src_dir_bdp);
268 mp = src_dp->i_mount;
269
270 if (DM_EVENT_ENABLED(src_dir_vp->v_vfsp, src_dp, DM_EVENT_RENAME) ||
271 DM_EVENT_ENABLED(target_dir_vp->v_vfsp,
272 target_dp, DM_EVENT_RENAME)) {
273 error = XFS_SEND_NAMESP(mp, DM_EVENT_RENAME,
274 src_dir_vp, DM_RIGHT_NULL,
275 target_dir_vp, DM_RIGHT_NULL,
276 src_name, target_name,
277 0, 0, 0);
278 if (error) {
279 return error;
280 }
281 }
282 /* Return through std_return after this point. */
283
284 /*
285 * Lock all the participating inodes. Depending upon whether
286 * the target_name exists in the target directory, and
287 * whether the target directory is the same as the source
288 * directory, we can lock from 2 to 4 inodes.
289 * xfs_lock_for_rename() will return ENOENT if src_name
290 * does not exist in the source directory.
291 */
292 tp = NULL;
293 error = xfs_lock_for_rename(src_dp, target_dp, src_vname,
294 target_vname, &src_ip, &target_ip, inodes,
295 &num_inodes);
296
297 if (error) {
298 /*
299 * We have nothing locked, no inode references, and
300 * no transaction, so just get out.
301 */
302 goto std_return;
303 }
304
305 ASSERT(src_ip != NULL);
306
307 if ((src_ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
308 /*
309 * Check for link count overflow on target_dp
310 */
311 if (target_ip == NULL && (src_dp != target_dp) &&
312 target_dp->i_d.di_nlink >= XFS_MAXLINK) {
313 error = XFS_ERROR(EMLINK);
314 xfs_rename_unlock4(inodes, XFS_ILOCK_SHARED);
315 goto rele_return;
316 }
317 }
318
319 new_parent = (src_dp != target_dp);
320 src_is_directory = ((src_ip->i_d.di_mode & S_IFMT) == S_IFDIR);
321
322 /*
323 * Drop the locks on our inodes so that we can start the transaction.
324 */
325 xfs_rename_unlock4(inodes, XFS_ILOCK_SHARED);
326
327 XFS_BMAP_INIT(&free_list, &first_block);
328 tp = xfs_trans_alloc(mp, XFS_TRANS_RENAME);
329 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
330 spaceres = XFS_RENAME_SPACE_RES(mp, target_namelen);
331 error = xfs_trans_reserve(tp, spaceres, XFS_RENAME_LOG_RES(mp), 0,
332 XFS_TRANS_PERM_LOG_RES, XFS_RENAME_LOG_COUNT);
333 if (error == ENOSPC) {
334 spaceres = 0;
335 error = xfs_trans_reserve(tp, 0, XFS_RENAME_LOG_RES(mp), 0,
336 XFS_TRANS_PERM_LOG_RES, XFS_RENAME_LOG_COUNT);
337 }
338 if (error) {
339 xfs_trans_cancel(tp, 0);
340 goto rele_return;
341 }
342
343 /*
344 * Attach the dquots to the inodes
345 */
346 if ((error = XFS_QM_DQVOPRENAME(mp, inodes))) {
347 xfs_trans_cancel(tp, cancel_flags);
348 goto rele_return;
349 }
350
351 /*
352 * Reacquire the inode locks we dropped above.
353 */
354 xfs_lock_inodes(inodes, num_inodes, 0, XFS_ILOCK_EXCL);
355
356 /*
357 * Join all the inodes to the transaction. From this point on,
358 * we can rely on either trans_commit or trans_cancel to unlock
359 * them. Note that we need to add a vnode reference to the
360 * directories since trans_commit & trans_cancel will decrement
361 * them when they unlock the inodes. Also, we need to be careful
362 * not to add an inode to the transaction more than once.
363 */
364 VN_HOLD(src_dir_vp);
365 xfs_trans_ijoin(tp, src_dp, XFS_ILOCK_EXCL);
366 if (new_parent) {
367 VN_HOLD(target_dir_vp);
368 xfs_trans_ijoin(tp, target_dp, XFS_ILOCK_EXCL);
369 }
370 if ((src_ip != src_dp) && (src_ip != target_dp)) {
371 xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL);
372 }
373 if ((target_ip != NULL) &&
374 (target_ip != src_ip) &&
375 (target_ip != src_dp) &&
376 (target_ip != target_dp)) {
377 xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL);
378 }
379
380 /*
381 * Set up the target.
382 */
383 if (target_ip == NULL) {
384 /*
385 * If there's no space reservation, check the entry will
386 * fit before actually inserting it.
387 */
388 if (spaceres == 0 &&
389 (error = XFS_DIR_CANENTER(mp, tp, target_dp, target_name,
390 target_namelen))) {
391 goto error_return;
392 }
393 /*
394 * If target does not exist and the rename crosses
395 * directories, adjust the target directory link count
396 * to account for the ".." reference from the new entry.
397 */
398 error = XFS_DIR_CREATENAME(mp, tp, target_dp, target_name,
399 target_namelen, src_ip->i_ino,
400 &first_block, &free_list, spaceres);
401 if (error == ENOSPC) {
402 goto error_return;
403 }
404 if (error) {
405 goto abort_return;
406 }
407 xfs_ichgtime(target_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
408
409 if (new_parent && src_is_directory) {
410 error = xfs_bumplink(tp, target_dp);
411 if (error) {
412 goto abort_return;
413 }
414 }
415 } else { /* target_ip != NULL */
416
417 /*
418 * If target exists and it's a directory, check that both
419 * target and source are directories and that target can be
420 * destroyed, or that neither is a directory.
421 */
422 if ((target_ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
423 /*
424 * Make sure target dir is empty.
425 */
426 if (!(XFS_DIR_ISEMPTY(target_ip->i_mount, target_ip)) ||
427 (target_ip->i_d.di_nlink > 2)) {
428 error = XFS_ERROR(EEXIST);
429 goto error_return;
430 }
431 }
432
433 /*
434 * Link the source inode under the target name.
435 * If the source inode is a directory and we are moving
436 * it across directories, its ".." entry will be
437 * inconsistent until we replace that down below.
438 *
439 * In case there is already an entry with the same
440 * name at the destination directory, remove it first.
441 */
442 error = XFS_DIR_REPLACE(mp, tp, target_dp, target_name,
443 target_namelen, src_ip->i_ino, &first_block,
444 &free_list, spaceres);
445 if (error) {
446 goto abort_return;
447 }
448 xfs_ichgtime(target_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
449
450 /*
451 * Decrement the link count on the target since the target
452 * dir no longer points to it.
453 */
454 error = xfs_droplink(tp, target_ip);
455 if (error) {
456 goto abort_return;
457 }
458 target_ip_dropped = 1;
459
460 if (src_is_directory) {
461 /*
462 * Drop the link from the old "." entry.
463 */
464 error = xfs_droplink(tp, target_ip);
465 if (error) {
466 goto abort_return;
467 }
468 }
469
470 /* Do this test while we still hold the locks */
471 target_link_zero = (target_ip)->i_d.di_nlink==0;
472
473 } /* target_ip != NULL */
474
475 /*
476 * Remove the source.
477 */
478 if (new_parent && src_is_directory) {
479
480 /*
481 * Rewrite the ".." entry to point to the new
482 * directory.
483 */
484 error = XFS_DIR_REPLACE(mp, tp, src_ip, "..", 2,
485 target_dp->i_ino, &first_block,
486 &free_list, spaceres);
487 ASSERT(error != EEXIST);
488 if (error) {
489 goto abort_return;
490 }
491 xfs_ichgtime(src_ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
492
493 } else {
494 /*
495 * We always want to hit the ctime on the source inode.
496 * We do it in the if clause above for the 'new_parent &&
497 * src_is_directory' case, and here we get all the other
498 * cases. This isn't strictly required by the standards
499 * since the source inode isn't really being changed,
500 * but old unix file systems did it and some incremental
501 * backup programs won't work without it.
502 */
503 xfs_ichgtime(src_ip, XFS_ICHGTIME_CHG);
504 }
505
506 /*
507 * Adjust the link count on src_dp. This is necessary when
508 * renaming a directory, either within one parent when
509 * the target existed, or across two parent directories.
510 */
511 if (src_is_directory && (new_parent || target_ip != NULL)) {
512
513 /*
514 * Decrement link count on src_directory since the
515 * entry that's moved no longer points to it.
516 */
517 error = xfs_droplink(tp, src_dp);
518 if (error) {
519 goto abort_return;
520 }
521 }
522
523 error = XFS_DIR_REMOVENAME(mp, tp, src_dp, src_name, src_namelen,
524 src_ip->i_ino, &first_block, &free_list, spaceres);
525 if (error) {
526 goto abort_return;
527 }
528 xfs_ichgtime(src_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
529
530 /*
531 * Update the generation counts on all the directory inodes
532 * that we're modifying.
533 */
534 src_dp->i_gen++;
535 xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE);
536
537 if (new_parent) {
538 target_dp->i_gen++;
539 xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE);
540 }
541
542 /*
543 * If there was a target inode, take an extra reference on
544 * it here so that it doesn't go to xfs_inactive() from
545 * within the commit.
546 */
547 if (target_ip != NULL) {
548 IHOLD(target_ip);
549 }
550
551 /*
552 * If this is a synchronous mount, make sure that the
553 * rename transaction goes to disk before returning to
554 * the user.
555 */
556 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
557 xfs_trans_set_sync(tp);
558 }
559
560 /*
561 * Take refs. for vop_link_removed calls below. No need to worry
562 * about directory refs. because the caller holds them.
563 *
564 * Do holds before the xfs_bmap_finish since it might rele them down
565 * to zero.
566 */
567
568 if (target_ip_dropped)
569 IHOLD(target_ip);
570 IHOLD(src_ip);
571
572 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
573 if (error) {
574 xfs_bmap_cancel(&free_list);
575 xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES |
576 XFS_TRANS_ABORT));
577 if (target_ip != NULL) {
578 IRELE(target_ip);
579 }
580 if (target_ip_dropped) {
581 IRELE(target_ip);
582 }
583 IRELE(src_ip);
584 goto std_return;
585 }
586
587 /*
588 * trans_commit will unlock src_ip, target_ip & decrement
589 * the vnode references.
590 */
591 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
592 if (target_ip != NULL) {
593 xfs_refcache_purge_ip(target_ip);
594 IRELE(target_ip);
595 }
596 /*
597 * Let interposed file systems know about removed links.
598 */
599 if (target_ip_dropped) {
600 XVOP_LINK_REMOVED(XFS_ITOV(target_ip), target_dir_vp,
601 target_link_zero);
602 IRELE(target_ip);
603 }
604
605 IRELE(src_ip);
606
607 /* Fall through to std_return with error = 0 or errno from
608 * xfs_trans_commit */
609 std_return:
610 if (DM_EVENT_ENABLED(src_dir_vp->v_vfsp, src_dp, DM_EVENT_POSTRENAME) ||
611 DM_EVENT_ENABLED(target_dir_vp->v_vfsp,
612 target_dp, DM_EVENT_POSTRENAME)) {
613 (void) XFS_SEND_NAMESP (mp, DM_EVENT_POSTRENAME,
614 src_dir_vp, DM_RIGHT_NULL,
615 target_dir_vp, DM_RIGHT_NULL,
616 src_name, target_name,
617 0, error, 0);
618 }
619 return error;
620
621 abort_return:
622 cancel_flags |= XFS_TRANS_ABORT;
623 /* FALLTHROUGH */
624 error_return:
625 xfs_bmap_cancel(&free_list);
626 xfs_trans_cancel(tp, cancel_flags);
627 goto std_return;
628
629 rele_return:
630 IRELE(src_ip);
631 if (target_ip != NULL) {
632 IRELE(target_ip);
633 }
634 goto std_return;
635 }
636
637