1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21 /*
22 * Copyright 2007 Sun Microsystems, Inc. All rights reserved.
23 * Use is subject to license terms.
24 *
25 * Copyright 2016 RackTop Systems.
26 */
27
28 #include <sys/types.h>
29 #include <sys/param.h>
30 #include <sys/sysmacros.h>
31 #include <sys/systm.h>
32 #include <sys/time.h>
33 #include <sys/vfs.h>
34 #include <sys/vnode.h>
35 #include <sys/errno.h>
36 #include <sys/cmn_err.h>
37 #include <sys/cred.h>
38 #include <sys/stat.h>
39 #include <sys/debug.h>
40 #include <sys/policy.h>
41 #include <sys/fs/tmpnode.h>
42 #include <sys/fs/tmp.h>
43 #include <sys/vtrace.h>
44
45 static int tdircheckpath(struct tmpnode *, struct tmpnode *, struct cred *);
46 static int tdirrename(struct tmpnode *, struct tmpnode *, struct tmpnode *,
47 char *, struct tmpnode *, struct tdirent *, struct cred *);
48 static void tdirfixdotdot(struct tmpnode *, struct tmpnode *, struct tmpnode *);
49 static int tdirmaketnode(struct tmpnode *, struct tmount *, struct vattr *,
50 enum de_op, struct tmpnode **, struct cred *);
51 static int tdiraddentry(struct tmpnode *, struct tmpnode *, char *,
52 enum de_op, struct tmpnode *);
53
54
55 #define T_HASH_SIZE 8192 /* must be power of 2 */
56 #define T_MUTEX_SIZE 64
57
58 static struct tdirent *t_hashtable[T_HASH_SIZE];
59 static kmutex_t t_hashmutex[T_MUTEX_SIZE];
60
61 #define T_HASH_INDEX(a) ((a) & (T_HASH_SIZE-1))
62 #define T_MUTEX_INDEX(a) ((a) & (T_MUTEX_SIZE-1))
63
64 #define TMPFS_HASH(tp, name, hash) \
65 { \
66 char Xc, *Xcp; \
67 hash = (uint_t)(uintptr_t)(tp) >> 8; \
68 for (Xcp = (name); (Xc = *Xcp) != 0; Xcp++) \
69 hash = (hash << 4) + hash + (uint_t)Xc; \
70 }
71
72 void
73 tmpfs_hash_init(void)
74 {
75 int ix;
76
77 for (ix = 0; ix < T_MUTEX_SIZE; ix++)
78 mutex_init(&t_hashmutex[ix], NULL, MUTEX_DEFAULT, NULL);
79 }
80
81 /*
82 * This routine is where the rubber meets the road for identities.
83 */
84 static void
85 tmpfs_hash_in(struct tdirent *t)
86 {
87 uint_t hash;
88 struct tdirent **prevpp;
89 kmutex_t *t_hmtx;
90
91 TMPFS_HASH(t->td_parent, t->td_name, hash);
92 t->td_hash = hash;
93 prevpp = &t_hashtable[T_HASH_INDEX(hash)];
94 t_hmtx = &t_hashmutex[T_MUTEX_INDEX(hash)];
95 mutex_enter(t_hmtx);
96 t->td_link = *prevpp;
97 *prevpp = t;
98 mutex_exit(t_hmtx);
99 }
100
101 /*
102 * Remove tdirent *t from the hash list.
103 */
104 static void
105 tmpfs_hash_out(struct tdirent *t)
106 {
107 uint_t hash;
108 struct tdirent **prevpp;
109 kmutex_t *t_hmtx;
110
111 hash = t->td_hash;
112 prevpp = &t_hashtable[T_HASH_INDEX(hash)];
113 t_hmtx = &t_hashmutex[T_MUTEX_INDEX(hash)];
114 mutex_enter(t_hmtx);
115 while (*prevpp != t)
116 prevpp = &(*prevpp)->td_link;
117 *prevpp = t->td_link;
118 mutex_exit(t_hmtx);
119 }
120
121 /*
122 * Currently called by tdirrename() only.
123 * rename operation needs to be done with lock held, to ensure that
124 * no other operations can access the tmpnode at the same instance.
125 */
126 static void
127 tmpfs_hash_change(struct tdirent *tdp, struct tmpnode *fromtp)
128 {
129 uint_t hash;
130 kmutex_t *t_hmtx;
131
132 hash = tdp->td_hash;
133 t_hmtx = &t_hashmutex[T_MUTEX_INDEX(hash)];
134 mutex_enter(t_hmtx);
135 tdp->td_tmpnode = fromtp;
136 mutex_exit(t_hmtx);
137 }
138
139 static struct tdirent *
140 tmpfs_hash_lookup(char *name, struct tmpnode *parent, uint_t hold,
141 struct tmpnode **found)
142 {
143 struct tdirent *l;
144 uint_t hash;
145 kmutex_t *t_hmtx;
146 struct tmpnode *tnp;
147
148 TMPFS_HASH(parent, name, hash);
149 t_hmtx = &t_hashmutex[T_MUTEX_INDEX(hash)];
150 mutex_enter(t_hmtx);
151 l = t_hashtable[T_HASH_INDEX(hash)];
152 while (l) {
153 if ((l->td_hash == hash) &&
154 (l->td_parent == parent) &&
155 (strcmp(l->td_name, name) == 0)) {
156 /*
157 * We need to make sure that the tmpnode that
158 * we put a hold on is the same one that we pass back.
159 * Hence, temporary variable tnp is necessary.
160 */
161 tnp = l->td_tmpnode;
162 if (hold) {
163 ASSERT(tnp);
164 tmpnode_hold(tnp);
165 }
166 if (found)
167 *found = tnp;
168 mutex_exit(t_hmtx);
169 return (l);
170 } else {
171 l = l->td_link;
172 }
173 }
174 mutex_exit(t_hmtx);
175 return (NULL);
176 }
177
178 /*
179 * Search directory 'parent' for entry 'name'.
180 *
181 * The calling thread can't hold the write version
182 * of the rwlock for the directory being searched
183 *
184 * 0 is returned on success and *foundtp points
185 * to the found tmpnode with its vnode held.
186 */
187 int
188 tdirlookup(
189 struct tmpnode *parent,
190 char *name,
191 struct tmpnode **foundtp,
192 struct cred *cred)
193 {
194 int error;
195
196 *foundtp = NULL;
197 if (parent->tn_type != VDIR)
198 return (ENOTDIR);
199
200 if ((error = tmp_taccess(parent, VEXEC, cred)))
201 return (error);
202
203 if (*name == '\0') {
204 tmpnode_hold(parent);
205 *foundtp = parent;
206 return (0);
207 }
208
209 /*
210 * Search the directory for the matching name
211 * We need the lock protecting the tn_dir list
212 * so that it doesn't change out from underneath us.
213 * tmpfs_hash_lookup() will pass back the tmpnode
214 * with a hold on it.
215 */
216
217 if (tmpfs_hash_lookup(name, parent, 1, foundtp) != NULL) {
218 ASSERT(*foundtp);
219 return (0);
220 }
221
222 return (ENOENT);
223 }
224
225 /*
226 * Enter a directory entry for 'name' and 'tp' into directory 'dir'
227 *
228 * Returns 0 on success.
229 */
230 int
231 tdirenter(
232 struct tmount *tm,
233 struct tmpnode *dir, /* target directory to make entry in */
234 char *name, /* name of entry */
235 enum de_op op, /* entry operation */
236 struct tmpnode *fromparent, /* source directory if rename */
237 struct tmpnode *tp, /* source tmpnode, if link/rename */
238 struct vattr *va,
239 struct tmpnode **tpp, /* return tmpnode, if create/mkdir */
240 struct cred *cred,
241 caller_context_t *ctp)
242 {
243 struct tdirent *tdp;
244 struct tmpnode *found = NULL;
245 int error = 0;
246 char *s;
247
248 /*
249 * tn_rwlock is held to serialize direnter and dirdeletes
250 */
251 ASSERT(RW_WRITE_HELD(&dir->tn_rwlock));
252 ASSERT(dir->tn_type == VDIR);
253
254 /*
255 * Don't allow '/' characters in pathname component
256 * (thus in ufs_direnter()).
257 */
258 for (s = name; *s; s++)
259 if (*s == '/')
260 return (EACCES);
261
262 if (name[0] == '\0')
263 panic("tdirenter: NULL name");
264
265 /*
266 * For link and rename lock the source entry and check the link count
267 * to see if it has been removed while it was unlocked.
268 */
269 if (op == DE_LINK || op == DE_RENAME) {
270 if (tp != dir)
271 rw_enter(&tp->tn_rwlock, RW_WRITER);
272 mutex_enter(&tp->tn_tlock);
273 if (tp->tn_nlink == 0) {
274 mutex_exit(&tp->tn_tlock);
275 if (tp != dir)
276 rw_exit(&tp->tn_rwlock);
277 return (ENOENT);
278 }
279
280 if (tp->tn_nlink == MAXLINK) {
281 mutex_exit(&tp->tn_tlock);
282 if (tp != dir)
283 rw_exit(&tp->tn_rwlock);
284 return (EMLINK);
285 }
286 tp->tn_nlink++;
287 gethrestime(&tp->tn_ctime);
288 mutex_exit(&tp->tn_tlock);
289 if (tp != dir)
290 rw_exit(&tp->tn_rwlock);
291 }
292
293 /*
294 * This might be a "dangling detached directory".
295 * it could have been removed, but a reference
296 * to it kept in u_cwd. don't bother searching
297 * it, and with any luck the user will get tired
298 * of dealing with us and cd to some absolute
299 * pathway. *sigh*, thus in ufs, too.
300 */
301 if (dir->tn_nlink == 0) {
302 error = ENOENT;
303 goto out;
304 }
305
306 /*
307 * If this is a rename of a directory and the parent is
308 * different (".." must be changed), then the source
309 * directory must not be in the directory hierarchy
310 * above the target, as this would orphan everything
311 * below the source directory.
312 */
313 if (op == DE_RENAME) {
314 if (tp == dir) {
315 error = EINVAL;
316 goto out;
317 }
318 if (tp->tn_type == VDIR) {
319 if ((fromparent != dir) &&
320 (error = tdircheckpath(tp, dir, cred))) {
321 goto out;
322 }
323 }
324 }
325
326 /*
327 * Search for the entry. Return "found" if it exists.
328 */
329 tdp = tmpfs_hash_lookup(name, dir, 1, &found);
330
331 if (tdp) {
332 ASSERT(found);
333 switch (op) {
334 case DE_CREATE:
335 case DE_MKDIR:
336 if (tpp) {
337 *tpp = found;
338 error = EEXIST;
339 } else {
340 tmpnode_rele(found);
341 }
342 break;
343
344 case DE_RENAME:
345 error = tdirrename(fromparent, tp,
346 dir, name, found, tdp, cred);
347 if (error == 0) {
348 if (found != NULL) {
349 vnevent_rename_dest(TNTOV(found),
350 TNTOV(dir), name, ctp);
351 }
352 }
353
354 tmpnode_rele(found);
355 break;
356
357 case DE_LINK:
358 /*
359 * Can't link to an existing file.
360 */
361 error = EEXIST;
362 tmpnode_rele(found);
363 break;
364 }
365 } else {
366
367 /*
368 * The entry does not exist. Check write permission in
369 * directory to see if entry can be created.
370 */
371 if (error = tmp_taccess(dir, VWRITE, cred))
372 goto out;
373 if (op == DE_CREATE || op == DE_MKDIR) {
374 /*
375 * Make new tmpnode and directory entry as required.
376 */
377 error = tdirmaketnode(dir, tm, va, op, &tp, cred);
378 if (error)
379 goto out;
380 }
381 if (error = tdiraddentry(dir, tp, name, op, fromparent)) {
382 if (op == DE_CREATE || op == DE_MKDIR) {
383 /*
384 * Unmake the inode we just made.
385 */
386 rw_enter(&tp->tn_rwlock, RW_WRITER);
387 if ((tp->tn_type) == VDIR) {
388 ASSERT(tdp == NULL);
389 /*
390 * cleanup allocs made by tdirinit()
391 */
392 tdirtrunc(tp);
393 }
394 mutex_enter(&tp->tn_tlock);
395 tp->tn_nlink = 0;
396 mutex_exit(&tp->tn_tlock);
397 gethrestime(&tp->tn_ctime);
398 rw_exit(&tp->tn_rwlock);
399 tmpnode_rele(tp);
400 tp = NULL;
401 }
402 } else if (tpp) {
403 *tpp = tp;
404 } else if (op == DE_CREATE || op == DE_MKDIR) {
405 tmpnode_rele(tp);
406 }
407 }
408
409 out:
410 if (error && (op == DE_LINK || op == DE_RENAME)) {
411 /*
412 * Undo bumped link count.
413 */
414 DECR_COUNT(&tp->tn_nlink, &tp->tn_tlock);
415 gethrestime(&tp->tn_ctime);
416 }
417 return (error);
418 }
419
420 /*
421 * Delete entry tp of name "nm" from dir.
422 * Free dir entry space and decrement link count on tmpnode(s).
423 *
424 * Return 0 on success.
425 */
426 int
427 tdirdelete(
428 struct tmpnode *dir,
429 struct tmpnode *tp,
430 char *nm,
431 enum dr_op op,
432 struct cred *cred)
433 {
434 struct tdirent *tpdp;
435 int error;
436 size_t namelen;
437 struct tmpnode *tnp;
438 timestruc_t now;
439
440 ASSERT(RW_WRITE_HELD(&dir->tn_rwlock));
441 ASSERT(RW_WRITE_HELD(&tp->tn_rwlock));
442 ASSERT(dir->tn_type == VDIR);
443
444 if (nm[0] == '\0')
445 panic("tdirdelete: NULL name for %p", (void *)tp);
446
447 /*
448 * return error when removing . and ..
449 */
450 if (nm[0] == '.') {
451 if (nm[1] == '\0')
452 return (EINVAL);
453 if (nm[1] == '.' && nm[2] == '\0')
454 return (EEXIST); /* thus in ufs */
455 }
456
457 if (error = tmp_taccess(dir, VEXEC|VWRITE, cred))
458 return (error);
459
460 /*
461 * If the parent directory is "sticky", then the user must
462 * own the parent directory or the file in it, or else must
463 * have permission to write the file. Otherwise it may not
464 * be deleted (except by privileged users).
465 * Same as ufs_dirremove.
466 */
467 if ((error = tmp_sticky_remove_access(dir, tp, cred)) != 0)
468 return (error);
469
470 if (dir->tn_dir == NULL)
471 return (ENOENT);
472
473 tpdp = tmpfs_hash_lookup(nm, dir, 0, &tnp);
474 if (tpdp == NULL) {
475 /*
476 * If it is gone, some other thread got here first!
477 * Return error ENOENT.
478 */
479 return (ENOENT);
480 }
481
482 /*
483 * If the tmpnode in the tdirent changed, we were probably
484 * the victim of a concurrent rename operation. The original
485 * is gone, so return that status (same as UFS).
486 */
487 if (tp != tnp)
488 return (ENOENT);
489
490 /*
491 * If we are unlinking a directory which isn't empty (it has
492 * more than the minumum '.' and '..' entries) return EEXIST.
493 */
494 if (tp->tn_type == VDIR && tp->tn_dirents > 2)
495 return (EEXIST);
496
497 tmpfs_hash_out(tpdp);
498
499 /*
500 * Take tpdp out of the directory list.
501 */
502 ASSERT(tpdp->td_next != tpdp);
503 ASSERT(tpdp->td_prev != tpdp);
504 if (tpdp->td_prev) {
505 tpdp->td_prev->td_next = tpdp->td_next;
506 }
507 if (tpdp->td_next) {
508 tpdp->td_next->td_prev = tpdp->td_prev;
509 }
510
511 /*
512 * If the roving slot pointer happens to match tpdp,
513 * point it at the previous dirent.
514 */
515 if (dir->tn_dir->td_prev == tpdp) {
516 dir->tn_dir->td_prev = tpdp->td_prev;
517 }
518 ASSERT(tpdp->td_next != tpdp);
519 ASSERT(tpdp->td_prev != tpdp);
520
521 /*
522 * tpdp points to the correct directory entry
523 */
524 namelen = strlen(tpdp->td_name) + 1;
525
526 tmp_memfree(tpdp, sizeof (struct tdirent) + namelen);
527 dir->tn_size -= (sizeof (struct tdirent) + namelen);
528 dir->tn_dirents--;
529
530 gethrestime(&now);
531 dir->tn_mtime = now;
532 dir->tn_ctime = now;
533 tp->tn_ctime = now;
534
535 ASSERT(tp->tn_nlink > 0);
536 DECR_COUNT(&tp->tn_nlink, &tp->tn_tlock);
537 if (tp->tn_type == VDIR) {
538 tdirtrunc(tp);
539 ASSERT(tp->tn_nlink == 0);
540 }
541 return (0);
542 }
543
544 /*
545 * tdirinit is used internally to initialize a directory (dir)
546 * with '.' and '..' entries without checking permissions and locking
547 */
548 void
549 tdirinit(
550 struct tmpnode *parent, /* parent of directory to initialize */
551 struct tmpnode *dir) /* the new directory */
552 {
553 struct tdirent *dot, *dotdot;
554 timestruc_t now;
555
556 ASSERT(RW_WRITE_HELD(&parent->tn_rwlock));
557 ASSERT(dir->tn_type == VDIR);
558
559 dot = tmp_memalloc(sizeof (struct tdirent) + 2, TMP_MUSTHAVE);
560 dotdot = tmp_memalloc(sizeof (struct tdirent) + 3, TMP_MUSTHAVE);
561
562 /*
563 * Initialize the entries
564 */
565 dot->td_tmpnode = dir;
566 dot->td_offset = 0;
567 dot->td_name = (char *)dot + sizeof (struct tdirent);
568 dot->td_name[0] = '.';
569 dot->td_parent = dir;
570 tmpfs_hash_in(dot);
571
572 dotdot->td_tmpnode = parent;
573 dotdot->td_offset = 1;
574 dotdot->td_name = (char *)dotdot + sizeof (struct tdirent);
575 dotdot->td_name[0] = '.';
576 dotdot->td_name[1] = '.';
577 dotdot->td_parent = dir;
578 tmpfs_hash_in(dotdot);
579
580 /*
581 * Initialize directory entry list.
582 */
583 dot->td_next = dotdot;
584 dot->td_prev = dotdot; /* dot's td_prev holds roving slot pointer */
585 dotdot->td_next = NULL;
586 dotdot->td_prev = dot;
587
588 gethrestime(&now);
589 dir->tn_mtime = now;
590 dir->tn_ctime = now;
591
592 /*
593 * Link counts are special for the hidden attribute directory.
594 * The only explicit reference in the name space is "." and
595 * the reference through ".." is not counted on the parent
596 * file. The attrdir is created as a side effect to lookup,
597 * so don't change the ctime of the parent.
598 * Since tdirinit is called with both dir and parent being the
599 * same for the root vnode, we need to increment this before we set
600 * tn_nlink = 2 below.
601 */
602 if (!(dir->tn_vnode->v_flag & V_XATTRDIR)) {
603 INCR_COUNT(&parent->tn_nlink, &parent->tn_tlock);
604 parent->tn_ctime = now;
605 }
606
607 dir->tn_dir = dot;
608 dir->tn_size = 2 * sizeof (struct tdirent) + 5; /* dot and dotdot */
609 dir->tn_dirents = 2;
610 dir->tn_nlink = 2;
611 }
612
613
614 /*
615 * tdirtrunc is called to remove all directory entries under this directory.
616 */
617 void
618 tdirtrunc(struct tmpnode *dir)
619 {
620 struct tdirent *tdp;
621 struct tmpnode *tp;
622 size_t namelen;
623 timestruc_t now;
624 int isvattrdir, isdotdot, skip_decr;
625
626 ASSERT(RW_WRITE_HELD(&dir->tn_rwlock));
627 ASSERT(dir->tn_type == VDIR);
628
629 isvattrdir = (dir->tn_vnode->v_flag & V_XATTRDIR) ? 1 : 0;
630 for (tdp = dir->tn_dir; tdp; tdp = dir->tn_dir) {
631 ASSERT(tdp->td_next != tdp);
632 ASSERT(tdp->td_prev != tdp);
633 ASSERT(tdp->td_tmpnode);
634
635 dir->tn_dir = tdp->td_next;
636 namelen = strlen(tdp->td_name) + 1;
637
638 /*
639 * Adjust the link counts to account for this directory
640 * entry removal. Hidden attribute directories may
641 * not be empty as they may be truncated as a side-
642 * effect of removing the parent. We do hold/rele
643 * operations to free up these tmpnodes.
644 *
645 * Skip the link count adjustment for parents of
646 * attribute directories as those link counts
647 * do not include the ".." reference in the hidden
648 * directories.
649 */
650 tp = tdp->td_tmpnode;
651 isdotdot = (strcmp("..", tdp->td_name) == 0);
652 skip_decr = (isvattrdir && isdotdot);
653 if (!skip_decr) {
654 ASSERT(tp->tn_nlink > 0);
655 DECR_COUNT(&tp->tn_nlink, &tp->tn_tlock);
656 }
657
658 tmpfs_hash_out(tdp);
659
660 tmp_memfree(tdp, sizeof (struct tdirent) + namelen);
661 dir->tn_size -= (sizeof (struct tdirent) + namelen);
662 dir->tn_dirents--;
663 }
664
665 gethrestime(&now);
666 dir->tn_mtime = now;
667 dir->tn_ctime = now;
668
669 ASSERT(dir->tn_dir == NULL);
670 ASSERT(dir->tn_size == 0);
671 ASSERT(dir->tn_dirents == 0);
672 }
673
674 /*
675 * Check if the source directory is in the path of the target directory.
676 * The target directory is locked by the caller.
677 *
678 * XXX - The source and target's should be different upon entry.
679 */
680 static int
681 tdircheckpath(
682 struct tmpnode *fromtp,
683 struct tmpnode *toparent,
684 struct cred *cred)
685 {
686 int error = 0;
687 struct tmpnode *dir, *dotdot;
688 struct tdirent *tdp;
689
690 ASSERT(RW_WRITE_HELD(&toparent->tn_rwlock));
691
692 tdp = tmpfs_hash_lookup("..", toparent, 1, &dotdot);
693 if (tdp == NULL)
694 return (ENOENT);
695
696 ASSERT(dotdot);
697
698 if (dotdot == toparent) {
699 /* root of fs. search trivially satisfied. */
700 tmpnode_rele(dotdot);
701 return (0);
702 }
703 for (;;) {
704 /*
705 * Return error for cases like "mv c c/d",
706 * "mv c c/d/e" and so on.
707 */
708 if (dotdot == fromtp) {
709 tmpnode_rele(dotdot);
710 error = EINVAL;
711 break;
712 }
713 dir = dotdot;
714 error = tdirlookup(dir, "..", &dotdot, cred);
715 if (error) {
716 tmpnode_rele(dir);
717 break;
718 }
719 /*
720 * We're okay if we traverse the directory tree up to
721 * the root directory and don't run into the
722 * parent directory.
723 */
724 if (dir == dotdot) {
725 tmpnode_rele(dir);
726 tmpnode_rele(dotdot);
727 break;
728 }
729 tmpnode_rele(dir);
730 }
731 return (error);
732 }
733
734 static int
735 tdirrename(
736 struct tmpnode *fromparent, /* parent directory of source */
737 struct tmpnode *fromtp, /* source tmpnode */
738 struct tmpnode *toparent, /* parent directory of target */
739 char *nm, /* entry we are trying to change */
740 struct tmpnode *to, /* target tmpnode */
741 struct tdirent *where, /* target tmpnode directory entry */
742 struct cred *cred) /* credentials */
743 {
744 int error = 0;
745 int doingdirectory;
746 timestruc_t now;
747
748 #if defined(lint)
749 nm = nm;
750 #endif
751 ASSERT(RW_WRITE_HELD(&toparent->tn_rwlock));
752
753 /*
754 * Short circuit rename of something to itself.
755 */
756 if (fromtp == to)
757 return (ESAME); /* special KLUDGE error code */
758
759 rw_enter(&fromtp->tn_rwlock, RW_READER);
760 rw_enter(&to->tn_rwlock, RW_READER);
761
762 /*
763 * Check that everything is on the same filesystem.
764 */
765 if (to->tn_vnode->v_vfsp != toparent->tn_vnode->v_vfsp ||
766 to->tn_vnode->v_vfsp != fromtp->tn_vnode->v_vfsp) {
767 error = EXDEV;
768 goto out;
769 }
770
771 /*
772 * Must have write permission to rewrite target entry.
773 * Check for stickyness.
774 */
775 if ((error = tmp_taccess(toparent, VWRITE, cred)) != 0 ||
776 (error = tmp_sticky_remove_access(toparent, to, cred)) != 0)
777 goto out;
778
779 /*
780 * Ensure source and target are compatible (both directories
781 * or both not directories). If target is a directory it must
782 * be empty and have no links to it; in addition it must not
783 * be a mount point, and both the source and target must be
784 * writable.
785 */
786 doingdirectory = (fromtp->tn_type == VDIR);
787 if (to->tn_type == VDIR) {
788 if (!doingdirectory) {
789 error = EISDIR;
790 goto out;
791 }
792 /*
793 * vn_vfswlock will prevent mounts from using the directory
794 * until we are done.
795 */
796 if (vn_vfswlock(TNTOV(to))) {
797 error = EBUSY;
798 goto out;
799 }
800 if (vn_mountedvfs(TNTOV(to)) != NULL) {
801 vn_vfsunlock(TNTOV(to));
802 error = EBUSY;
803 goto out;
804 }
805
806 mutex_enter(&to->tn_tlock);
807 if (to->tn_dirents > 2 || to->tn_nlink > 2) {
808 mutex_exit(&to->tn_tlock);
809 vn_vfsunlock(TNTOV(to));
810 error = EEXIST; /* SIGH should be ENOTEMPTY */
811 /*
812 * Update atime because checking tn_dirents is
813 * logically equivalent to reading the directory
814 */
815 gethrestime(&to->tn_atime);
816 goto out;
817 }
818 mutex_exit(&to->tn_tlock);
819 } else if (doingdirectory) {
820 error = ENOTDIR;
821 goto out;
822 }
823
824 tmpfs_hash_change(where, fromtp);
825 gethrestime(&now);
826 toparent->tn_mtime = now;
827 toparent->tn_ctime = now;
828
829 /*
830 * Upgrade to write lock on "to" (i.e., the target tmpnode).
831 */
832 rw_exit(&to->tn_rwlock);
833 rw_enter(&to->tn_rwlock, RW_WRITER);
834
835 /*
836 * Decrement the link count of the target tmpnode.
837 */
838 DECR_COUNT(&to->tn_nlink, &to->tn_tlock);
839 to->tn_ctime = now;
840
841 if (doingdirectory) {
842 /*
843 * The entry for "to" no longer exists so release the vfslock.
844 */
845 vn_vfsunlock(TNTOV(to));
846
847 /*
848 * Decrement the target link count and delete all entires.
849 */
850 tdirtrunc(to);
851 ASSERT(to->tn_nlink == 0);
852
853 /*
854 * Renaming a directory with the parent different
855 * requires that ".." be rewritten. The window is
856 * still there for ".." to be inconsistent, but this
857 * is unavoidable, and a lot shorter than when it was
858 * done in a user process.
859 */
860 if (fromparent != toparent)
861 tdirfixdotdot(fromtp, fromparent, toparent);
862 }
863 out:
864 rw_exit(&to->tn_rwlock);
865 rw_exit(&fromtp->tn_rwlock);
866 return (error);
867 }
868
869 static void
870 tdirfixdotdot(
871 struct tmpnode *fromtp, /* child directory */
872 struct tmpnode *fromparent, /* old parent directory */
873 struct tmpnode *toparent) /* new parent directory */
874 {
875 struct tdirent *dotdot;
876
877 ASSERT(RW_LOCK_HELD(&toparent->tn_rwlock));
878
879 /*
880 * Increment the link count in the new parent tmpnode
881 */
882 INCR_COUNT(&toparent->tn_nlink, &toparent->tn_tlock);
883 gethrestime(&toparent->tn_ctime);
884
885 dotdot = tmpfs_hash_lookup("..", fromtp, 0, NULL);
886
887 ASSERT(dotdot->td_tmpnode == fromparent);
888 dotdot->td_tmpnode = toparent;
889
890 /*
891 * Decrement the link count of the old parent tmpnode.
892 * If fromparent is NULL, then this is a new directory link;
893 * it has no parent, so we need not do anything.
894 */
895 if (fromparent != NULL) {
896 mutex_enter(&fromparent->tn_tlock);
897 if (fromparent->tn_nlink != 0) {
898 fromparent->tn_nlink--;
899 gethrestime(&fromparent->tn_ctime);
900 }
901 mutex_exit(&fromparent->tn_tlock);
902 }
903 }
904
905 static int
906 tdiraddentry(
907 struct tmpnode *dir, /* target directory to make entry in */
908 struct tmpnode *tp, /* new tmpnode */
909 char *name,
910 enum de_op op,
911 struct tmpnode *fromtp)
912 {
913 struct tdirent *tdp, *tpdp;
914 size_t namelen, alloc_size;
915 timestruc_t now;
916
917 /*
918 * Make sure the parent directory wasn't removed from
919 * underneath the caller.
920 */
921 if (dir->tn_dir == NULL)
922 return (ENOENT);
923
924 /*
925 * Check that everything is on the same filesystem.
926 */
927 if (tp->tn_vnode->v_vfsp != dir->tn_vnode->v_vfsp)
928 return (EXDEV);
929
930 /*
931 * Allocate and initialize directory entry
932 */
933 namelen = strlen(name) + 1;
934 alloc_size = namelen + sizeof (struct tdirent);
935 tdp = tmp_memalloc(alloc_size, 0);
936 if (tdp == NULL)
937 return (ENOSPC);
938
939 if ((op == DE_RENAME) && (tp->tn_type == VDIR))
940 tdirfixdotdot(tp, fromtp, dir);
941
942 dir->tn_size += alloc_size;
943 dir->tn_dirents++;
944 tdp->td_tmpnode = tp;
945 tdp->td_parent = dir;
946
947 /*
948 * The directory entry and its name were allocated sequentially.
949 */
950 tdp->td_name = (char *)tdp + sizeof (struct tdirent);
951 (void) strcpy(tdp->td_name, name);
952
953 tmpfs_hash_in(tdp);
954
955 /*
956 * Some utilities expect the size of a directory to remain
957 * somewhat static. For example, a routine which unlinks
958 * files between calls to readdir(); the size of the
959 * directory changes from underneath it and so the real
960 * directory offset in bytes is invalid. To circumvent
961 * this problem, we initialize a directory entry with an
962 * phony offset, and use this offset to determine end of
963 * file in tmp_readdir.
964 */
965 tpdp = dir->tn_dir->td_prev;
966 /*
967 * Install at first empty "slot" in directory list.
968 */
969 while (tpdp->td_next != NULL && (tpdp->td_next->td_offset -
970 tpdp->td_offset) <= 1) {
971 ASSERT(tpdp->td_next != tpdp);
972 ASSERT(tpdp->td_prev != tpdp);
973 ASSERT(tpdp->td_next->td_offset > tpdp->td_offset);
974 tpdp = tpdp->td_next;
975 }
976 tdp->td_offset = tpdp->td_offset + 1;
977
978 /*
979 * If we're at the end of the dirent list and the offset (which
980 * is necessarily the largest offset in this directory) is more
981 * than twice the number of dirents, that means the directory is
982 * 50% holes. At this point we reset the slot pointer back to
983 * the beginning of the directory so we start using the holes.
984 * The idea is that if there are N dirents, there must also be
985 * N holes, so we can satisfy the next N creates by walking at
986 * most 2N entries; thus the average cost of a create is constant.
987 * Note that we use the first dirent's td_prev as the roving
988 * slot pointer; it's ugly, but it saves a word in every dirent.
989 */
990 if (tpdp->td_next == NULL && tpdp->td_offset > 2 * dir->tn_dirents)
991 dir->tn_dir->td_prev = dir->tn_dir->td_next;
992 else
993 dir->tn_dir->td_prev = tdp;
994
995 ASSERT(tpdp->td_next != tpdp);
996 ASSERT(tpdp->td_prev != tpdp);
997
998 tdp->td_next = tpdp->td_next;
999 if (tdp->td_next) {
1000 tdp->td_next->td_prev = tdp;
1001 }
1002 tdp->td_prev = tpdp;
1003 tpdp->td_next = tdp;
1004
1005 ASSERT(tdp->td_next != tdp);
1006 ASSERT(tdp->td_prev != tdp);
1007 ASSERT(tpdp->td_next != tpdp);
1008 ASSERT(tpdp->td_prev != tpdp);
1009
1010 gethrestime(&now);
1011 dir->tn_mtime = now;
1012 dir->tn_ctime = now;
1013
1014 return (0);
1015 }
1016
1017 static int
1018 tdirmaketnode(
1019 struct tmpnode *dir,
1020 struct tmount *tm,
1021 struct vattr *va,
1022 enum de_op op,
1023 struct tmpnode **newnode,
1024 struct cred *cred)
1025 {
1026 struct tmpnode *tp;
1027 enum vtype type;
1028
1029 ASSERT(va != NULL);
1030 ASSERT(op == DE_CREATE || op == DE_MKDIR);
1031 if (((va->va_mask & AT_ATIME) && TIMESPEC_OVERFLOW(&va->va_atime)) ||
1032 ((va->va_mask & AT_MTIME) && TIMESPEC_OVERFLOW(&va->va_mtime)))
1033 return (EOVERFLOW);
1034 type = va->va_type;
1035 tp = tmp_memalloc(sizeof (struct tmpnode), TMP_MUSTHAVE);
1036 tmpnode_init(tm, tp, va, cred);
1037
1038 /* setup normal file/dir's extended attribute directory */
1039 if (dir->tn_flags & ISXATTR) {
1040 /* parent dir is , mark file as xattr */
1041 tp->tn_flags |= ISXATTR;
1042 }
1043
1044
1045 if (type == VBLK || type == VCHR) {
1046 tp->tn_vnode->v_rdev = tp->tn_rdev = va->va_rdev;
1047 } else {
1048 tp->tn_vnode->v_rdev = tp->tn_rdev = NODEV;
1049 }
1050 tp->tn_vnode->v_type = type;
1051 tp->tn_uid = crgetuid(cred);
1052
1053 /*
1054 * To determine the group-id of the created file:
1055 * 1) If the gid is set in the attribute list (non-Sun & pre-4.0
1056 * clients are not likely to set the gid), then use it if
1057 * the process is privileged, belongs to the target group,
1058 * or the group is the same as the parent directory.
1059 * 2) If the filesystem was not mounted with the Old-BSD-compatible
1060 * GRPID option, and the directory's set-gid bit is clear,
1061 * then use the process's gid.
1062 * 3) Otherwise, set the group-id to the gid of the parent directory.
1063 */
1064 if ((va->va_mask & AT_GID) &&
1065 ((va->va_gid == dir->tn_gid) || groupmember(va->va_gid, cred) ||
1066 secpolicy_vnode_create_gid(cred) == 0)) {
1067 /*
1068 * XXX - is this only the case when a 4.0 NFS client, or a
1069 * client derived from that code, makes a call over the wire?
1070 */
1071 tp->tn_gid = va->va_gid;
1072 } else {
1073 if (dir->tn_mode & VSGID)
1074 tp->tn_gid = dir->tn_gid;
1075 else
1076 tp->tn_gid = crgetgid(cred);
1077 }
1078 /*
1079 * If we're creating a directory, and the parent directory has the
1080 * set-GID bit set, set it on the new directory.
1081 * Otherwise, if the user is neither privileged nor a member of the
1082 * file's new group, clear the file's set-GID bit.
1083 */
1084 if (dir->tn_mode & VSGID && type == VDIR)
1085 tp->tn_mode |= VSGID;
1086 else {
1087 if ((tp->tn_mode & VSGID) &&
1088 secpolicy_vnode_setids_setgids(cred, tp->tn_gid) != 0)
1089 tp->tn_mode &= ~VSGID;
1090 }
1091
1092 if (va->va_mask & AT_ATIME)
1093 tp->tn_atime = va->va_atime;
1094 if (va->va_mask & AT_MTIME)
1095 tp->tn_mtime = va->va_mtime;
1096
1097 if (op == DE_MKDIR)
1098 tdirinit(dir, tp);
1099
1100 *newnode = tp;
1101 return (0);
1102 }