1 /*
2 * Copyright (c) 1994 Jan-Simon Pendry
3 * Copyright (c) 1994
4 * The Regents of the University of California. All rights reserved.
5 *
6 * This code is derived from software contributed to Berkeley by
7 * Jan-Simon Pendry.
8 *
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
11 * are met:
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
17 * 3. All advertising materials mentioning features or use of this software
18 * must display the following acknowledgement:
19 * This product includes software developed by the University of
20 * California, Berkeley and its contributors.
21 * 4. Neither the name of the University nor the names of its contributors
22 * may be used to endorse or promote products derived from this software
23 * without specific prior written permission.
24 *
25 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
26 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
29 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35 * SUCH DAMAGE.
36 *
37 * @(#)union_subr.c 8.20 (Berkeley) 5/20/95
38 * $FreeBSD: releng/5.0/sys/fs/unionfs/union_subr.c 105077 2002-10-14 03:20:36Z mckusick $
39 */
40
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/fcntl.h>
44 #include <sys/file.h>
45 #include <sys/filedesc.h>
46 #include <sys/kernel.h>
47 #include <sys/lock.h>
48 #include <sys/malloc.h>
49 #include <sys/module.h>
50 #include <sys/mount.h>
51 #include <sys/mutex.h>
52 #include <sys/namei.h>
53 #include <sys/stat.h>
54 #include <sys/vnode.h>
55
56 #include <vm/vm.h>
57 #include <vm/vm_extern.h> /* for vnode_pager_setsize */
58 #include <vm/vm_object.h> /* for vm cache coherency */
59 #include <vm/uma.h>
60
61 #include <fs/unionfs/union.h>
62
63 #include <sys/proc.h>
64
65 extern int union_init(void);
66
67 /* must be power of two, otherwise change UNION_HASH() */
68 #define NHASH 32
69
70 /* unsigned int ... */
71 #define UNION_HASH(u, l) \
72 (((((uintptr_t) (u)) + ((uintptr_t) l)) >> 8) & (NHASH-1))
73
74 static LIST_HEAD(unhead, union_node) unhead[NHASH];
75 static int unvplock[NHASH];
76
77 static void union_dircache_r(struct vnode *vp, struct vnode ***vppp,
78 int *cntp);
79 static int union_list_lock(int ix);
80 static void union_list_unlock(int ix);
81 static int union_relookup(struct union_mount *um, struct vnode *dvp,
82 struct vnode **vpp,
83 struct componentname *cnp,
84 struct componentname *cn, char *path,
85 int pathlen);
86 static void union_updatevp(struct union_node *un,
87 struct vnode *uppervp,
88 struct vnode *lowervp);
89 static void union_newlower(struct union_node *, struct vnode *);
90 static void union_newupper(struct union_node *, struct vnode *);
91 static int union_copyfile(struct vnode *, struct vnode *,
92 struct ucred *, struct thread *);
93 static int union_vn_create(struct vnode **, struct union_node *,
94 struct thread *);
95 static int union_vn_close(struct vnode *, int, struct ucred *,
96 struct thread *);
97
98 int
99 union_init()
100 {
101 int i;
102
103 for (i = 0; i < NHASH; i++)
104 LIST_INIT(&unhead[i]);
105 bzero((caddr_t)unvplock, sizeof(unvplock));
106 return (0);
107 }
108
109 static int
110 union_list_lock(ix)
111 int ix;
112 {
113 if (unvplock[ix] & UNVP_LOCKED) {
114 unvplock[ix] |= UNVP_WANT;
115 (void) tsleep((caddr_t) &unvplock[ix], PINOD, "unllck", 0);
116 return (1);
117 }
118 unvplock[ix] |= UNVP_LOCKED;
119 return (0);
120 }
121
122 static void
123 union_list_unlock(ix)
124 int ix;
125 {
126 unvplock[ix] &= ~UNVP_LOCKED;
127
128 if (unvplock[ix] & UNVP_WANT) {
129 unvplock[ix] &= ~UNVP_WANT;
130 wakeup((caddr_t) &unvplock[ix]);
131 }
132 }
133
134 /*
135 * union_updatevp:
136 *
137 * The uppervp, if not NULL, must be referenced and not locked by us
138 * The lowervp, if not NULL, must be referenced.
139 *
140 * If uppervp and lowervp match pointers already installed, then
141 * nothing happens. The passed vp's (when matching) are not adjusted.
142 *
143 * This routine may only be called by union_newupper() and
144 * union_newlower().
145 */
146
147 static void
148 union_updatevp(un, uppervp, lowervp)
149 struct union_node *un;
150 struct vnode *uppervp;
151 struct vnode *lowervp;
152 {
153 int ohash = UNION_HASH(un->un_uppervp, un->un_lowervp);
154 int nhash = UNION_HASH(uppervp, lowervp);
155 int docache = (lowervp != NULLVP || uppervp != NULLVP);
156 int lhash, uhash;
157
158 /*
159 * Ensure locking is ordered from lower to higher
160 * to avoid deadlocks.
161 */
162 if (nhash < ohash) {
163 lhash = nhash;
164 uhash = ohash;
165 } else {
166 lhash = ohash;
167 uhash = nhash;
168 }
169
170 if (lhash != uhash) {
171 while (union_list_lock(lhash))
172 continue;
173 }
174
175 while (union_list_lock(uhash))
176 continue;
177
178 if (ohash != nhash || !docache) {
179 if (un->un_flags & UN_CACHED) {
180 un->un_flags &= ~UN_CACHED;
181 LIST_REMOVE(un, un_cache);
182 }
183 }
184
185 if (ohash != nhash)
186 union_list_unlock(ohash);
187
188 if (un->un_lowervp != lowervp) {
189 if (un->un_lowervp) {
190 vrele(un->un_lowervp);
191 if (un->un_path) {
192 free(un->un_path, M_TEMP);
193 un->un_path = 0;
194 }
195 }
196 un->un_lowervp = lowervp;
197 un->un_lowersz = VNOVAL;
198 }
199
200 if (un->un_uppervp != uppervp) {
201 if (un->un_uppervp)
202 vrele(un->un_uppervp);
203 un->un_uppervp = uppervp;
204 un->un_uppersz = VNOVAL;
205 }
206
207 if (docache && (ohash != nhash)) {
208 LIST_INSERT_HEAD(&unhead[nhash], un, un_cache);
209 un->un_flags |= UN_CACHED;
210 }
211
212 union_list_unlock(nhash);
213 }
214
215 /*
216 * Set a new lowervp. The passed lowervp must be referenced and will be
217 * stored in the vp in a referenced state.
218 */
219
220 static void
221 union_newlower(un, lowervp)
222 struct union_node *un;
223 struct vnode *lowervp;
224 {
225 union_updatevp(un, un->un_uppervp, lowervp);
226 }
227
228 /*
229 * Set a new uppervp. The passed uppervp must be locked and will be
230 * stored in the vp in a locked state. The caller should not unlock
231 * uppervp.
232 */
233
234 static void
235 union_newupper(un, uppervp)
236 struct union_node *un;
237 struct vnode *uppervp;
238 {
239 union_updatevp(un, uppervp, un->un_lowervp);
240 }
241
242 /*
243 * Keep track of size changes in the underlying vnodes.
244 * If the size changes, then callback to the vm layer
245 * giving priority to the upper layer size.
246 */
247 void
248 union_newsize(vp, uppersz, lowersz)
249 struct vnode *vp;
250 off_t uppersz, lowersz;
251 {
252 struct union_node *un;
253 off_t sz;
254
255 /* only interested in regular files */
256 if (vp->v_type != VREG)
257 return;
258
259 un = VTOUNION(vp);
260 sz = VNOVAL;
261
262 if ((uppersz != VNOVAL) && (un->un_uppersz != uppersz)) {
263 un->un_uppersz = uppersz;
264 if (sz == VNOVAL)
265 sz = un->un_uppersz;
266 }
267
268 if ((lowersz != VNOVAL) && (un->un_lowersz != lowersz)) {
269 un->un_lowersz = lowersz;
270 if (sz == VNOVAL)
271 sz = un->un_lowersz;
272 }
273
274 if (sz != VNOVAL) {
275 UDEBUG(("union: %s size now %ld\n",
276 (uppersz != VNOVAL ? "upper" : "lower"), (long)sz));
277 /*
278 * There is no need to change size of non-existent object.
279 */
280 /* vnode_pager_setsize(vp, sz); */
281 }
282 }
283
284 /*
285 * union_allocvp: allocate a union_node and associate it with a
286 * parent union_node and one or two vnodes.
287 *
288 * vpp Holds the returned vnode locked and referenced if no
289 * error occurs.
290 *
291 * mp Holds the mount point. mp may or may not be busied.
292 * allocvp() makes no changes to mp.
293 *
294 * dvp Holds the parent union_node to the one we wish to create.
295 * XXX may only be used to traverse an uncopied lowervp-based
296 * tree? XXX
297 *
298 * dvp may or may not be locked. allocvp() makes no changes
299 * to dvp.
300 *
301 * upperdvp Holds the parent vnode to uppervp, generally used along
302 * with path component information to create a shadow of
303 * lowervp when uppervp does not exist.
304 *
305 * upperdvp is referenced but unlocked on entry, and will be
306 * dereferenced on return.
307 *
308 * uppervp Holds the new uppervp vnode to be stored in the
309 * union_node we are allocating. uppervp is referenced but
310 * not locked, and will be dereferenced on return.
311 *
312 * lowervp Holds the new lowervp vnode to be stored in the
313 * union_node we are allocating. lowervp is referenced but
314 * not locked, and will be dereferenced on return.
315 *
316 * cnp Holds path component information to be coupled with
317 * lowervp and upperdvp to allow unionfs to create an uppervp
318 * later on. Only used if lowervp is valid. The contents
319 * of cnp is only valid for the duration of the call.
320 *
321 * docache Determine whether this node should be entered in the
322 * cache or whether it should be destroyed as soon as possible.
323 *
324 * All union_nodes are maintained on a singly-linked
325 * list. New nodes are only allocated when they cannot
326 * be found on this list. Entries on the list are
327 * removed when the vfs reclaim entry is called.
328 *
329 * A single lock is kept for the entire list. This is
330 * needed because the getnewvnode() function can block
331 * waiting for a vnode to become free, in which case there
332 * may be more than one process trying to get the same
333 * vnode. This lock is only taken if we are going to
334 * call getnewvnode(), since the kernel itself is single-threaded.
335 *
336 * If an entry is found on the list, then call vget() to
337 * take a reference. This is done because there may be
338 * zero references to it and so it needs to removed from
339 * the vnode free list.
340 */
341
342 int
343 union_allocvp(vpp, mp, dvp, upperdvp, cnp, uppervp, lowervp, docache)
344 struct vnode **vpp;
345 struct mount *mp;
346 struct vnode *dvp; /* parent union vnode */
347 struct vnode *upperdvp; /* parent vnode of uppervp */
348 struct componentname *cnp; /* may be null */
349 struct vnode *uppervp; /* may be null */
350 struct vnode *lowervp; /* may be null */
351 int docache;
352 {
353 int error;
354 struct union_node *un = 0;
355 struct union_mount *um = MOUNTTOUNIONMOUNT(mp);
356 struct thread *td = (cnp) ? cnp->cn_thread : curthread;
357 int hash = 0;
358 int vflag;
359 int try;
360
361 if (uppervp == NULLVP && lowervp == NULLVP)
362 panic("union: unidentifiable allocation");
363
364 if (uppervp && lowervp && (uppervp->v_type != lowervp->v_type)) {
365 vrele(lowervp);
366 lowervp = NULLVP;
367 }
368
369 /* detect the root vnode (and aliases) */
370 vflag = 0;
371 if ((uppervp == um->um_uppervp) &&
372 ((lowervp == NULLVP) || lowervp == um->um_lowervp)) {
373 if (lowervp == NULLVP) {
374 lowervp = um->um_lowervp;
375 if (lowervp != NULLVP)
376 VREF(lowervp);
377 }
378 vflag = VV_ROOT;
379 }
380
381 loop:
382 if (!docache) {
383 un = 0;
384 } else for (try = 0; try < 3; try++) {
385 switch (try) {
386 case 0:
387 if (lowervp == NULLVP)
388 continue;
389 hash = UNION_HASH(uppervp, lowervp);
390 break;
391
392 case 1:
393 if (uppervp == NULLVP)
394 continue;
395 hash = UNION_HASH(uppervp, NULLVP);
396 break;
397
398 case 2:
399 if (lowervp == NULLVP)
400 continue;
401 hash = UNION_HASH(NULLVP, lowervp);
402 break;
403 }
404
405 while (union_list_lock(hash))
406 continue;
407
408 LIST_FOREACH(un, &unhead[hash], un_cache) {
409 if ((un->un_lowervp == lowervp ||
410 un->un_lowervp == NULLVP) &&
411 (un->un_uppervp == uppervp ||
412 un->un_uppervp == NULLVP) &&
413 (UNIONTOV(un)->v_mount == mp)) {
414 if (vget(UNIONTOV(un), 0,
415 cnp ? cnp->cn_thread : NULL)) {
416 union_list_unlock(hash);
417 goto loop;
418 }
419 break;
420 }
421 }
422
423 union_list_unlock(hash);
424
425 if (un)
426 break;
427 }
428
429 if (un) {
430 /*
431 * Obtain a lock on the union_node. Everything is unlocked
432 * except for dvp, so check that case. If they match, our
433 * new un is already locked. Otherwise we have to lock our
434 * new un.
435 *
436 * A potential deadlock situation occurs when we are holding
437 * one lock while trying to get another. We must follow
438 * strict ordering rules to avoid it. We try to locate dvp
439 * by scanning up from un_vnode, since the most likely
440 * scenario is un being under dvp.
441 */
442
443 if (dvp && un->un_vnode != dvp) {
444 struct vnode *scan = un->un_vnode;
445
446 do {
447 scan = VTOUNION(scan)->un_pvp;
448 } while (scan && scan->v_op == union_vnodeop_p &&
449 scan != dvp);
450 if (scan != dvp) {
451 /*
452 * our new un is above dvp (we never saw dvp
453 * while moving up the tree).
454 */
455 VREF(dvp);
456 VOP_UNLOCK(dvp, 0, td);
457 error = vn_lock(un->un_vnode, LK_EXCLUSIVE, td);
458 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY, td);
459 vrele(dvp);
460 } else {
461 /*
462 * our new un is under dvp
463 */
464 error = vn_lock(un->un_vnode, LK_EXCLUSIVE, td);
465 }
466 } else if (dvp == NULLVP) {
467 /*
468 * dvp is NULL, we need to lock un.
469 */
470 error = vn_lock(un->un_vnode, LK_EXCLUSIVE, td);
471 } else {
472 /*
473 * dvp == un->un_vnode, we are already locked.
474 */
475 error = 0;
476 }
477
478 if (error)
479 goto loop;
480
481 /*
482 * At this point, the union_node is locked and referenced.
483 *
484 * uppervp is locked and referenced or NULL, lowervp is
485 * referenced or NULL.
486 */
487 UDEBUG(("Modify existing un %p vn %p upper %p(refs %d) -> %p(refs %d)\n",
488 un, un->un_vnode, un->un_uppervp,
489 (un->un_uppervp ? vrefcnt(un->un_uppervp) : -99),
490 uppervp,
491 (uppervp ? vrefcnt(uppervp) : -99)
492 ));
493
494 if (uppervp != un->un_uppervp) {
495 KASSERT(uppervp == NULL || vrefcnt(uppervp) > 0, ("union_allocvp: too few refs %d (at least 1 required) on uppervp", vrefcnt(uppervp)));
496 union_newupper(un, uppervp);
497 } else if (uppervp) {
498 KASSERT(vrefcnt(uppervp) > 1, ("union_allocvp: too few refs %d (at least 2 required) on uppervp", vrefcnt(uppervp)));
499 vrele(uppervp);
500 }
501
502 /*
503 * Save information about the lower layer.
504 * This needs to keep track of pathname
505 * and directory information which union_vn_create()
506 * might need.
507 */
508 if (lowervp != un->un_lowervp) {
509 union_newlower(un, lowervp);
510 if (cnp && (lowervp != NULLVP)) {
511 un->un_path = malloc(cnp->cn_namelen+1,
512 M_TEMP, M_WAITOK);
513 bcopy(cnp->cn_nameptr, un->un_path,
514 cnp->cn_namelen);
515 un->un_path[cnp->cn_namelen] = '\0';
516 }
517 } else if (lowervp) {
518 vrele(lowervp);
519 }
520
521 /*
522 * and upperdvp
523 */
524 if (upperdvp != un->un_dirvp) {
525 if (un->un_dirvp)
526 vrele(un->un_dirvp);
527 un->un_dirvp = upperdvp;
528 } else if (upperdvp) {
529 vrele(upperdvp);
530 }
531
532 *vpp = UNIONTOV(un);
533 return (0);
534 }
535
536 if (docache) {
537 /*
538 * Otherwise lock the vp list while we call getnewvnode()
539 * since that can block.
540 */
541 hash = UNION_HASH(uppervp, lowervp);
542
543 if (union_list_lock(hash))
544 goto loop;
545 }
546
547 /*
548 * Create new node rather than replace old node.
549 */
550
551 error = getnewvnode("union", mp, union_vnodeop_p, vpp);
552 if (error) {
553 /*
554 * If an error occurs, clear out vnodes.
555 */
556 if (lowervp)
557 vrele(lowervp);
558 if (uppervp)
559 vrele(uppervp);
560 if (upperdvp)
561 vrele(upperdvp);
562 *vpp = NULL;
563 goto out;
564 }
565
566 MALLOC((*vpp)->v_data, void *, sizeof(struct union_node),
567 M_TEMP, M_WAITOK);
568
569 ASSERT_VOP_LOCKED(*vpp, "union_allocvp");
570 (*vpp)->v_vflag |= vflag;
571 if (uppervp)
572 (*vpp)->v_type = uppervp->v_type;
573 else
574 (*vpp)->v_type = lowervp->v_type;
575
576 un = VTOUNION(*vpp);
577 bzero(un, sizeof(*un));
578
579 vn_lock(*vpp, LK_EXCLUSIVE | LK_RETRY, td);
580
581 un->un_vnode = *vpp;
582 un->un_uppervp = uppervp;
583 un->un_uppersz = VNOVAL;
584 un->un_lowervp = lowervp;
585 un->un_lowersz = VNOVAL;
586 un->un_dirvp = upperdvp;
587 un->un_pvp = dvp; /* only parent dir in new allocation */
588 if (dvp != NULLVP)
589 VREF(dvp);
590 un->un_dircache = 0;
591 un->un_openl = 0;
592
593 if (cnp && (lowervp != NULLVP)) {
594 un->un_path = malloc(cnp->cn_namelen+1, M_TEMP, M_WAITOK);
595 bcopy(cnp->cn_nameptr, un->un_path, cnp->cn_namelen);
596 un->un_path[cnp->cn_namelen] = '\0';
597 } else {
598 un->un_path = 0;
599 un->un_dirvp = NULL;
600 }
601
602 if (docache) {
603 LIST_INSERT_HEAD(&unhead[hash], un, un_cache);
604 un->un_flags |= UN_CACHED;
605 }
606
607 out:
608 if (docache)
609 union_list_unlock(hash);
610
611 return (error);
612 }
613
614 int
615 union_freevp(vp)
616 struct vnode *vp;
617 {
618 struct union_node *un = VTOUNION(vp);
619
620 if (un->un_flags & UN_CACHED) {
621 un->un_flags &= ~UN_CACHED;
622 LIST_REMOVE(un, un_cache);
623 }
624
625 if (un->un_pvp != NULLVP) {
626 vrele(un->un_pvp);
627 un->un_pvp = NULL;
628 }
629 if (un->un_uppervp != NULLVP) {
630 vrele(un->un_uppervp);
631 un->un_uppervp = NULL;
632 }
633 if (un->un_lowervp != NULLVP) {
634 vrele(un->un_lowervp);
635 un->un_lowervp = NULL;
636 }
637 if (un->un_dirvp != NULLVP) {
638 vrele(un->un_dirvp);
639 un->un_dirvp = NULL;
640 }
641 if (un->un_path) {
642 free(un->un_path, M_TEMP);
643 un->un_path = NULL;
644 }
645
646 FREE(vp->v_data, M_TEMP);
647 vp->v_data = 0;
648
649 return (0);
650 }
651
652 /*
653 * copyfile. Copy the vnode (fvp) to the vnode (tvp)
654 * using a sequence of reads and writes. Both (fvp)
655 * and (tvp) are locked on entry and exit.
656 *
657 * fvp and tvp are both exclusive locked on call, but their refcount's
658 * haven't been bumped at all.
659 */
660 static int
661 union_copyfile(fvp, tvp, cred, td)
662 struct vnode *fvp;
663 struct vnode *tvp;
664 struct ucred *cred;
665 struct thread *td;
666 {
667 char *buf;
668 struct uio uio;
669 struct iovec iov;
670 int error = 0;
671
672 /*
673 * strategy:
674 * Allocate a buffer of size MAXBSIZE.
675 * Loop doing reads and writes, keeping track
676 * of the current uio offset.
677 * Give up at the first sign of trouble.
678 */
679
680 bzero(&uio, sizeof(uio));
681
682 uio.uio_td = td;
683 uio.uio_segflg = UIO_SYSSPACE;
684 uio.uio_offset = 0;
685
686 VOP_LEASE(fvp, td, cred, LEASE_READ);
687 VOP_LEASE(tvp, td, cred, LEASE_WRITE);
688
689 buf = malloc(MAXBSIZE, M_TEMP, M_WAITOK);
690
691 /* ugly loop follows... */
692 do {
693 off_t offset = uio.uio_offset;
694 int count;
695 int bufoffset;
696
697 /*
698 * Setup for big read.
699 */
700 uio.uio_iov = &iov;
701 uio.uio_iovcnt = 1;
702 iov.iov_base = buf;
703 iov.iov_len = MAXBSIZE;
704 uio.uio_resid = iov.iov_len;
705 uio.uio_rw = UIO_READ;
706
707 if ((error = VOP_READ(fvp, &uio, 0, cred)) != 0)
708 break;
709
710 /*
711 * Get bytes read, handle read eof case and setup for
712 * write loop.
713 */
714 if ((count = MAXBSIZE - uio.uio_resid) == 0)
715 break;
716 bufoffset = 0;
717
718 /*
719 * Write until an error occurs or our buffer has been
720 * exhausted, then update the offset for the next read.
721 */
722 while (bufoffset < count) {
723 uio.uio_iov = &iov;
724 uio.uio_iovcnt = 1;
725 iov.iov_base = buf + bufoffset;
726 iov.iov_len = count - bufoffset;
727 uio.uio_offset = offset + bufoffset;
728 uio.uio_rw = UIO_WRITE;
729 uio.uio_resid = iov.iov_len;
730
731 if ((error = VOP_WRITE(tvp, &uio, 0, cred)) != 0)
732 break;
733 bufoffset += (count - bufoffset) - uio.uio_resid;
734 }
735 uio.uio_offset = offset + bufoffset;
736 } while (error == 0);
737
738 free(buf, M_TEMP);
739 return (error);
740 }
741
742 /*
743 *
744 * un's vnode is assumed to be locked on entry and remains locked on exit.
745 */
746
747 int
748 union_copyup(un, docopy, cred, td)
749 struct union_node *un;
750 int docopy;
751 struct ucred *cred;
752 struct thread *td;
753 {
754 int error;
755 struct mount *mp;
756 struct vnode *lvp, *uvp;
757
758 /*
759 * If the user does not have read permission, the vnode should not
760 * be copied to upper layer.
761 */
762 vn_lock(un->un_lowervp, LK_EXCLUSIVE | LK_RETRY, td);
763 error = VOP_ACCESS(un->un_lowervp, VREAD, cred, td);
764 VOP_UNLOCK(un->un_lowervp, 0, td);
765 if (error)
766 return (error);
767
768 if ((error = vn_start_write(un->un_dirvp, &mp, V_WAIT | PCATCH)) != 0)
769 return (error);
770 if ((error = union_vn_create(&uvp, un, td)) != 0) {
771 vn_finished_write(mp);
772 return (error);
773 }
774
775 lvp = un->un_lowervp;
776
777 KASSERT(vrefcnt(uvp) > 0, ("copy: uvp refcount 0: %d", vrefcnt(uvp)));
778 if (docopy) {
779 /*
780 * XX - should not ignore errors
781 * from VOP_CLOSE()
782 */
783 vn_lock(lvp, LK_EXCLUSIVE | LK_RETRY, td);
784 error = VOP_OPEN(lvp, FREAD, cred, td);
785 if (error == 0 && vn_canvmio(lvp) == TRUE)
786 error = vfs_object_create(lvp, td, cred);
787 if (error == 0) {
788 error = union_copyfile(lvp, uvp, cred, td);
789 VOP_UNLOCK(lvp, 0, td);
790 (void) VOP_CLOSE(lvp, FREAD, cred, td);
791 }
792 if (error == 0)
793 UDEBUG(("union: copied up %s\n", un->un_path));
794
795 }
796 VOP_UNLOCK(uvp, 0, td);
797 vn_finished_write(mp);
798 union_newupper(un, uvp);
799 KASSERT(vrefcnt(uvp) > 0, ("copy: uvp refcount 0: %d", vrefcnt(uvp)));
800 union_vn_close(uvp, FWRITE, cred, td);
801 KASSERT(vrefcnt(uvp) > 0, ("copy: uvp refcount 0: %d", vrefcnt(uvp)));
802 /*
803 * Subsequent IOs will go to the top layer, so
804 * call close on the lower vnode and open on the
805 * upper vnode to ensure that the filesystem keeps
806 * its references counts right. This doesn't do
807 * the right thing with (cred) and (FREAD) though.
808 * Ignoring error returns is not right, either.
809 */
810 if (error == 0) {
811 int i;
812
813 for (i = 0; i < un->un_openl; i++) {
814 (void) VOP_CLOSE(lvp, FREAD, cred, td);
815 (void) VOP_OPEN(uvp, FREAD, cred, td);
816 }
817 if (un->un_openl) {
818 if (vn_canvmio(uvp) == TRUE)
819 error = vfs_object_create(uvp, td, cred);
820 }
821 un->un_openl = 0;
822 }
823
824 return (error);
825
826 }
827
828 /*
829 * union_relookup:
830 *
831 * dvp should be locked on entry and will be locked on return. No
832 * net change in the ref count will occur.
833 *
834 * If an error is returned, *vpp will be invalid, otherwise it
835 * will hold a locked, referenced vnode. If *vpp == dvp then
836 * remember that only one exclusive lock is held.
837 */
838
839 static int
840 union_relookup(um, dvp, vpp, cnp, cn, path, pathlen)
841 struct union_mount *um;
842 struct vnode *dvp;
843 struct vnode **vpp;
844 struct componentname *cnp;
845 struct componentname *cn;
846 char *path;
847 int pathlen;
848 {
849 int error;
850
851 /*
852 * A new componentname structure must be faked up because
853 * there is no way to know where the upper level cnp came
854 * from or what it is being used for. This must duplicate
855 * some of the work done by NDINIT(), some of the work done
856 * by namei(), some of the work done by lookup() and some of
857 * the work done by VOP_LOOKUP() when given a CREATE flag.
858 * Conclusion: Horrible.
859 */
860 cn->cn_namelen = pathlen;
861 cn->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
862 bcopy(path, cn->cn_pnbuf, cn->cn_namelen);
863 cn->cn_pnbuf[cn->cn_namelen] = '\0';
864
865 cn->cn_nameiop = CREATE;
866 cn->cn_flags = (LOCKPARENT|LOCKLEAF|HASBUF|SAVENAME|ISLASTCN);
867 cn->cn_thread = cnp->cn_thread;
868 if (um->um_op == UNMNT_ABOVE)
869 cn->cn_cred = cnp->cn_cred;
870 else
871 cn->cn_cred = um->um_cred;
872 cn->cn_nameptr = cn->cn_pnbuf;
873 cn->cn_consume = cnp->cn_consume;
874
875 VREF(dvp);
876 VOP_UNLOCK(dvp, 0, cnp->cn_thread);
877
878 /*
879 * Pass dvp unlocked and referenced on call to relookup().
880 *
881 * If an error occurs, dvp will be returned unlocked and dereferenced.
882 */
883
884 if ((error = relookup(dvp, vpp, cn)) != 0) {
885 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY, cnp->cn_thread);
886 return(error);
887 }
888
889 /*
890 * If no error occurs, dvp will be returned locked with the reference
891 * left as before, and vpp will be returned referenced and locked.
892 *
893 * We want to return with dvp as it was passed to us, so we get
894 * rid of our reference.
895 */
896 vrele(dvp);
897 return (0);
898 }
899
900 /*
901 * Create a shadow directory in the upper layer.
902 * The new vnode is returned locked.
903 *
904 * (um) points to the union mount structure for access to the
905 * the mounting process's credentials.
906 * (dvp) is the directory in which to create the shadow directory,
907 * It is locked (but not ref'd) on entry and return.
908 * (cnp) is the component name to be created.
909 * (vpp) is the returned newly created shadow directory, which
910 * is returned locked and ref'd
911 */
912 int
913 union_mkshadow(um, dvp, cnp, vpp)
914 struct union_mount *um;
915 struct vnode *dvp;
916 struct componentname *cnp;
917 struct vnode **vpp;
918 {
919 int error;
920 struct vattr va;
921 struct thread *td = cnp->cn_thread;
922 struct componentname cn;
923 struct mount *mp;
924
925 if ((error = vn_start_write(dvp, &mp, V_WAIT | PCATCH)) != 0)
926 return (error);
927 if ((error = union_relookup(um, dvp, vpp, cnp, &cn,
928 cnp->cn_nameptr, cnp->cn_namelen)) != 0) {
929 vn_finished_write(mp);
930 return (error);
931 }
932
933 if (*vpp) {
934 if (cn.cn_flags & HASBUF) {
935 uma_zfree(namei_zone, cn.cn_pnbuf);
936 cn.cn_flags &= ~HASBUF;
937 }
938 if (dvp == *vpp)
939 vrele(*vpp);
940 else
941 vput(*vpp);
942 vn_finished_write(mp);
943 *vpp = NULLVP;
944 return (EEXIST);
945 }
946
947 /*
948 * Policy: when creating the shadow directory in the
949 * upper layer, create it owned by the user who did
950 * the mount, group from parent directory, and mode
951 * 777 modified by umask (ie mostly identical to the
952 * mkdir syscall). (jsp, kb)
953 */
954
955 VATTR_NULL(&va);
956 va.va_type = VDIR;
957 va.va_mode = um->um_cmode;
958
959 /* VOP_LEASE: dvp is locked */
960 VOP_LEASE(dvp, td, cn.cn_cred, LEASE_WRITE);
961
962 error = VOP_MKDIR(dvp, vpp, &cn, &va);
963 if (cn.cn_flags & HASBUF) {
964 uma_zfree(namei_zone, cn.cn_pnbuf);
965 cn.cn_flags &= ~HASBUF;
966 }
967 /*vput(dvp);*/
968 vn_finished_write(mp);
969 return (error);
970 }
971
972 /*
973 * Create a whiteout entry in the upper layer.
974 *
975 * (um) points to the union mount structure for access to the
976 * the mounting process's credentials.
977 * (dvp) is the directory in which to create the whiteout.
978 * It is locked on entry and return.
979 * (cnp) is the component name to be created.
980 */
981 int
982 union_mkwhiteout(um, dvp, cnp, path)
983 struct union_mount *um;
984 struct vnode *dvp;
985 struct componentname *cnp;
986 char *path;
987 {
988 int error;
989 struct thread *td = cnp->cn_thread;
990 struct vnode *wvp;
991 struct componentname cn;
992 struct mount *mp;
993
994 if ((error = vn_start_write(dvp, &mp, V_WAIT | PCATCH)) != 0)
995 return (error);
996 error = union_relookup(um, dvp, &wvp, cnp, &cn, path, strlen(path));
997 if (error) {
998 vn_finished_write(mp);
999 return (error);
1000 }
1001
1002 if (wvp) {
1003 if (cn.cn_flags & HASBUF) {
1004 uma_zfree(namei_zone, cn.cn_pnbuf);
1005 cn.cn_flags &= ~HASBUF;
1006 }
1007 if (wvp == dvp)
1008 vrele(wvp);
1009 else
1010 vput(wvp);
1011 vn_finished_write(mp);
1012 return (EEXIST);
1013 }
1014
1015 /* VOP_LEASE: dvp is locked */
1016 VOP_LEASE(dvp, td, td->td_ucred, LEASE_WRITE);
1017
1018 error = VOP_WHITEOUT(dvp, &cn, CREATE);
1019 if (cn.cn_flags & HASBUF) {
1020 uma_zfree(namei_zone, cn.cn_pnbuf);
1021 cn.cn_flags &= ~HASBUF;
1022 }
1023 vn_finished_write(mp);
1024 return (error);
1025 }
1026
1027 /*
1028 * union_vn_create: creates and opens a new shadow file
1029 * on the upper union layer. This function is similar
1030 * in spirit to calling vn_open() but it avoids calling namei().
1031 * The problem with calling namei() is that a) it locks too many
1032 * things, and b) it doesn't start at the "right" directory,
1033 * whereas relookup() is told where to start.
1034 *
1035 * On entry, the vnode associated with un is locked. It remains locked
1036 * on return.
1037 *
1038 * If no error occurs, *vpp contains a locked referenced vnode for your
1039 * use. If an error occurs *vpp iis undefined.
1040 */
1041 static int
1042 union_vn_create(vpp, un, td)
1043 struct vnode **vpp;
1044 struct union_node *un;
1045 struct thread *td;
1046 {
1047 struct vnode *vp;
1048 struct ucred *cred = td->td_ucred;
1049 struct vattr vat;
1050 struct vattr *vap = &vat;
1051 int fmode = FFLAGS(O_WRONLY|O_CREAT|O_TRUNC|O_EXCL);
1052 int error;
1053 int cmode;
1054 struct componentname cn;
1055
1056 *vpp = NULLVP;
1057 FILEDESC_LOCK(td->td_proc->p_fd);
1058 cmode = UN_FILEMODE & ~td->td_proc->p_fd->fd_cmask;
1059 FILEDESC_UNLOCK(td->td_proc->p_fd);
1060
1061 /*
1062 * Build a new componentname structure (for the same
1063 * reasons outlines in union_mkshadow()).
1064 * The difference here is that the file is owned by
1065 * the current user, rather than by the person who
1066 * did the mount, since the current user needs to be
1067 * able to write the file (that's why it is being
1068 * copied in the first place).
1069 */
1070 cn.cn_namelen = strlen(un->un_path);
1071 cn.cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
1072 bcopy(un->un_path, cn.cn_pnbuf, cn.cn_namelen+1);
1073 cn.cn_nameiop = CREATE;
1074 cn.cn_flags = (LOCKPARENT|LOCKLEAF|HASBUF|SAVENAME|ISLASTCN);
1075 cn.cn_thread = td;
1076 cn.cn_cred = td->td_ucred;
1077 cn.cn_nameptr = cn.cn_pnbuf;
1078 cn.cn_consume = 0;
1079
1080 /*
1081 * Pass dvp unlocked and referenced on call to relookup().
1082 *
1083 * If an error occurs, dvp will be returned unlocked and dereferenced.
1084 */
1085 VREF(un->un_dirvp);
1086 error = relookup(un->un_dirvp, &vp, &cn);
1087 if (error)
1088 return (error);
1089
1090 /*
1091 * If no error occurs, dvp will be returned locked with the reference
1092 * left as before, and vpp will be returned referenced and locked.
1093 */
1094 if (vp) {
1095 vput(un->un_dirvp);
1096 if (cn.cn_flags & HASBUF) {
1097 uma_zfree(namei_zone, cn.cn_pnbuf);
1098 cn.cn_flags &= ~HASBUF;
1099 }
1100 if (vp == un->un_dirvp)
1101 vrele(vp);
1102 else
1103 vput(vp);
1104 return (EEXIST);
1105 }
1106
1107 /*
1108 * Good - there was no race to create the file
1109 * so go ahead and create it. The permissions
1110 * on the file will be 0666 modified by the
1111 * current user's umask. Access to the file, while
1112 * it is unioned, will require access to the top *and*
1113 * bottom files. Access when not unioned will simply
1114 * require access to the top-level file.
1115 * TODO: confirm choice of access permissions.
1116 */
1117 VATTR_NULL(vap);
1118 vap->va_type = VREG;
1119 vap->va_mode = cmode;
1120 VOP_LEASE(un->un_dirvp, td, cred, LEASE_WRITE);
1121 error = VOP_CREATE(un->un_dirvp, &vp, &cn, vap);
1122 if (cn.cn_flags & HASBUF) {
1123 uma_zfree(namei_zone, cn.cn_pnbuf);
1124 cn.cn_flags &= ~HASBUF;
1125 }
1126 vput(un->un_dirvp);
1127 if (error)
1128 return (error);
1129
1130 error = VOP_OPEN(vp, fmode, cred, td);
1131 if (error == 0 && vn_canvmio(vp) == TRUE)
1132 error = vfs_object_create(vp, td, cred);
1133 if (error) {
1134 vput(vp);
1135 return (error);
1136 }
1137 vp->v_writecount++;
1138 *vpp = vp;
1139 return (0);
1140 }
1141
1142 static int
1143 union_vn_close(vp, fmode, cred, td)
1144 struct vnode *vp;
1145 int fmode;
1146 struct ucred *cred;
1147 struct thread *td;
1148 {
1149
1150 if (fmode & FWRITE)
1151 --vp->v_writecount;
1152 return (VOP_CLOSE(vp, fmode, cred, td));
1153 }
1154
1155 #if 0
1156
1157 /*
1158 * union_removed_upper:
1159 *
1160 * called with union_node unlocked. XXX
1161 */
1162
1163 void
1164 union_removed_upper(un)
1165 struct union_node *un;
1166 {
1167 struct thread *td = curthread; /* XXX */
1168 struct vnode **vpp;
1169
1170 /*
1171 * Do not set the uppervp to NULLVP. If lowervp is NULLVP,
1172 * union node will have neither uppervp nor lowervp. We remove
1173 * the union node from cache, so that it will not be referrenced.
1174 */
1175 union_newupper(un, NULLVP);
1176 if (un->un_dircache != 0) {
1177 for (vpp = un->un_dircache; *vpp != NULLVP; vpp++)
1178 vrele(*vpp);
1179 free(un->un_dircache, M_TEMP);
1180 un->un_dircache = 0;
1181 }
1182
1183 if (un->un_flags & UN_CACHED) {
1184 un->un_flags &= ~UN_CACHED;
1185 LIST_REMOVE(un, un_cache);
1186 }
1187 }
1188
1189 #endif
1190
1191 /*
1192 * Determine whether a whiteout is needed
1193 * during a remove/rmdir operation.
1194 */
1195 int
1196 union_dowhiteout(un, cred, td)
1197 struct union_node *un;
1198 struct ucred *cred;
1199 struct thread *td;
1200 {
1201 struct vattr va;
1202
1203 if (un->un_lowervp != NULLVP)
1204 return (1);
1205
1206 if (VOP_GETATTR(un->un_uppervp, &va, cred, td) == 0 &&
1207 (va.va_flags & OPAQUE))
1208 return (1);
1209
1210 return (0);
1211 }
1212
1213 static void
1214 union_dircache_r(vp, vppp, cntp)
1215 struct vnode *vp;
1216 struct vnode ***vppp;
1217 int *cntp;
1218 {
1219 struct union_node *un;
1220
1221 if (vp->v_op != union_vnodeop_p) {
1222 if (vppp) {
1223 VREF(vp);
1224 *(*vppp)++ = vp;
1225 if (--(*cntp) == 0)
1226 panic("union: dircache table too small");
1227 } else {
1228 (*cntp)++;
1229 }
1230
1231 return;
1232 }
1233
1234 un = VTOUNION(vp);
1235 if (un->un_uppervp != NULLVP)
1236 union_dircache_r(un->un_uppervp, vppp, cntp);
1237 if (un->un_lowervp != NULLVP)
1238 union_dircache_r(un->un_lowervp, vppp, cntp);
1239 }
1240
1241 struct vnode *
1242 union_dircache(vp, td)
1243 struct vnode *vp;
1244 struct thread *td;
1245 {
1246 int cnt;
1247 struct vnode *nvp;
1248 struct vnode **vpp;
1249 struct vnode **dircache;
1250 struct union_node *un;
1251 int error;
1252
1253 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
1254 dircache = VTOUNION(vp)->un_dircache;
1255
1256 nvp = NULLVP;
1257
1258 if (dircache == NULL) {
1259 cnt = 0;
1260 union_dircache_r(vp, 0, &cnt);
1261 cnt++;
1262 dircache = malloc(cnt * sizeof(struct vnode *),
1263 M_TEMP, M_WAITOK);
1264 vpp = dircache;
1265 union_dircache_r(vp, &vpp, &cnt);
1266 *vpp = NULLVP;
1267 vpp = dircache + 1;
1268 } else {
1269 vpp = dircache;
1270 do {
1271 if (*vpp++ == VTOUNION(vp)->un_uppervp)
1272 break;
1273 } while (*vpp != NULLVP);
1274 }
1275
1276 if (*vpp == NULLVP)
1277 goto out;
1278
1279 /*vn_lock(*vpp, LK_EXCLUSIVE | LK_RETRY, td);*/
1280 UDEBUG(("ALLOCVP-3 %p ref %d\n", *vpp, (*vpp ? vrefcnt(*vpp) : -99)));
1281 VREF(*vpp);
1282 error = union_allocvp(&nvp, vp->v_mount, NULLVP, NULLVP, NULL, *vpp, NULLVP, 0);
1283 UDEBUG(("ALLOCVP-3B %p ref %d\n", nvp, (*vpp ? vrefcnt(*vpp) : -99)));
1284 if (error)
1285 goto out;
1286
1287 VTOUNION(vp)->un_dircache = 0;
1288 un = VTOUNION(nvp);
1289 un->un_dircache = dircache;
1290
1291 out:
1292 VOP_UNLOCK(vp, 0, td);
1293 return (nvp);
1294 }
1295
1296 /*
1297 * Module glue to remove #ifdef UNION from vfs_syscalls.c
1298 */
1299 static int
1300 union_dircheck(struct thread *td, struct vnode **vp, struct file *fp)
1301 {
1302 int error = 0;
1303
1304 if ((*vp)->v_op == union_vnodeop_p) {
1305 struct vnode *lvp;
1306
1307 lvp = union_dircache(*vp, td);
1308 if (lvp != NULLVP) {
1309 struct vattr va;
1310
1311 /*
1312 * If the directory is opaque,
1313 * then don't show lower entries
1314 */
1315 error = VOP_GETATTR(*vp, &va, fp->f_cred, td);
1316 if (va.va_flags & OPAQUE) {
1317 vput(lvp);
1318 lvp = NULL;
1319 }
1320 }
1321
1322 if (lvp != NULLVP) {
1323 error = VOP_OPEN(lvp, FREAD, fp->f_cred, td);
1324 if (error == 0 && vn_canvmio(lvp) == TRUE)
1325 error = vfs_object_create(lvp, td, fp->f_cred);
1326 if (error) {
1327 vput(lvp);
1328 return (error);
1329 }
1330 VOP_UNLOCK(lvp, 0, td);
1331 FILE_LOCK(fp);
1332 fp->f_data = (caddr_t) lvp;
1333 fp->f_offset = 0;
1334 FILE_UNLOCK(fp);
1335 error = vn_close(*vp, FREAD, fp->f_cred, td);
1336 if (error)
1337 return (error);
1338 *vp = lvp;
1339 return -1; /* goto unionread */
1340 }
1341 }
1342 return error;
1343 }
1344
1345 static int
1346 union_modevent(module_t mod, int type, void *data)
1347 {
1348 switch (type) {
1349 case MOD_LOAD:
1350 union_dircheckp = union_dircheck;
1351 break;
1352 case MOD_UNLOAD:
1353 union_dircheckp = NULL;
1354 break;
1355 default:
1356 break;
1357 }
1358 return 0;
1359 }
1360
1361 static moduledata_t union_mod = {
1362 "union_dircheck",
1363 union_modevent,
1364 NULL
1365 };
1366
1367 DECLARE_MODULE(union_dircheck, union_mod, SI_SUB_VFS, SI_ORDER_ANY);
Cache object: 21fed0e051e2a10679d47dd21da0b0d4
|