1 /* $NetBSD: tmpfs_subr.c,v 1.35 2007/07/09 21:10:50 ad Exp $ */
2
3 /*-
4 * Copyright (c) 2005 The NetBSD Foundation, Inc.
5 * All rights reserved.
6 *
7 * This code is derived from software contributed to The NetBSD Foundation
8 * by Julio M. Merino Vidal, developed as part of Google's Summer of Code
9 * 2005 program.
10 *
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
13 * are met:
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 *
20 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
22 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
23 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
24 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
25 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
26 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
27 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
28 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
29 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
30 * POSSIBILITY OF SUCH DAMAGE.
31 */
32
33 /*
34 * Efficient memory file system supporting functions.
35 */
36 #include <sys/cdefs.h>
37 __FBSDID("$FreeBSD$");
38
39 #include <sys/param.h>
40 #include <sys/namei.h>
41 #include <sys/priv.h>
42 #include <sys/proc.h>
43 #include <sys/stat.h>
44 #include <sys/systm.h>
45 #include <sys/vnode.h>
46 #include <sys/vmmeter.h>
47
48 #include <vm/vm.h>
49 #include <vm/vm_object.h>
50 #include <vm/vm_page.h>
51 #include <vm/vm_pager.h>
52 #include <vm/vm_extern.h>
53
54 #include <fs/tmpfs/tmpfs.h>
55 #include <fs/tmpfs/tmpfs_fifoops.h>
56 #include <fs/tmpfs/tmpfs_vnops.h>
57
58 /* --------------------------------------------------------------------- */
59
60 /*
61 * Allocates a new node of type 'type' inside the 'tmp' mount point, with
62 * its owner set to 'uid', its group to 'gid' and its mode set to 'mode',
63 * using the credentials of the process 'p'.
64 *
65 * If the node type is set to 'VDIR', then the parent parameter must point
66 * to the parent directory of the node being created. It may only be NULL
67 * while allocating the root node.
68 *
69 * If the node type is set to 'VBLK' or 'VCHR', then the rdev parameter
70 * specifies the device the node represents.
71 *
72 * If the node type is set to 'VLNK', then the parameter target specifies
73 * the file name of the target file for the symbolic link that is being
74 * created.
75 *
76 * Note that new nodes are retrieved from the available list if it has
77 * items or, if it is empty, from the node pool as long as there is enough
78 * space to create them.
79 *
80 * Returns zero on success or an appropriate error code on failure.
81 */
82 int
83 tmpfs_alloc_node(struct tmpfs_mount *tmp, enum vtype type,
84 uid_t uid, gid_t gid, mode_t mode, struct tmpfs_node *parent,
85 char *target, dev_t rdev, struct thread *p, struct tmpfs_node **node)
86 {
87 struct tmpfs_node *nnode;
88
89 /* If the root directory of the 'tmp' file system is not yet
90 * allocated, this must be the request to do it. */
91 MPASS(IMPLIES(tmp->tm_root == NULL, parent == NULL && type == VDIR));
92
93 MPASS(IFF(type == VLNK, target != NULL));
94 MPASS(IFF(type == VBLK || type == VCHR, rdev != VNOVAL));
95
96 if (tmp->tm_nodes_inuse > tmp->tm_nodes_max)
97 return (ENOSPC);
98
99 nnode = (struct tmpfs_node *)uma_zalloc_arg(
100 tmp->tm_node_pool, tmp, M_WAITOK);
101
102 /* Generic initialization. */
103 nnode->tn_type = type;
104 vfs_timestamp(&nnode->tn_atime);
105 nnode->tn_birthtime = nnode->tn_ctime = nnode->tn_mtime =
106 nnode->tn_atime;
107 nnode->tn_uid = uid;
108 nnode->tn_gid = gid;
109 nnode->tn_mode = mode;
110 nnode->tn_id = alloc_unr(tmp->tm_ino_unr);
111
112 /* Type-specific initialization. */
113 switch (nnode->tn_type) {
114 case VBLK:
115 case VCHR:
116 nnode->tn_rdev = rdev;
117 break;
118
119 case VDIR:
120 TAILQ_INIT(&nnode->tn_dir.tn_dirhead);
121 MPASS(parent != nnode);
122 MPASS(IMPLIES(parent == NULL, tmp->tm_root == NULL));
123 nnode->tn_dir.tn_parent = (parent == NULL) ? nnode : parent;
124 nnode->tn_dir.tn_readdir_lastn = 0;
125 nnode->tn_dir.tn_readdir_lastp = NULL;
126 nnode->tn_links++;
127 nnode->tn_dir.tn_parent->tn_links++;
128 break;
129
130 case VFIFO:
131 /* FALLTHROUGH */
132 case VSOCK:
133 break;
134
135 case VLNK:
136 MPASS(strlen(target) < MAXPATHLEN);
137 nnode->tn_size = strlen(target);
138 nnode->tn_link = malloc(nnode->tn_size, M_TMPFSNAME,
139 M_WAITOK);
140 memcpy(nnode->tn_link, target, nnode->tn_size);
141 break;
142
143 case VREG:
144 nnode->tn_reg.tn_aobj =
145 vm_pager_allocate(OBJT_SWAP, NULL, 0, VM_PROT_DEFAULT, 0);
146 nnode->tn_reg.tn_aobj_pages = 0;
147 break;
148
149 default:
150 MPASS(0);
151 }
152
153 TMPFS_LOCK(tmp);
154 LIST_INSERT_HEAD(&tmp->tm_nodes_used, nnode, tn_entries);
155 tmp->tm_nodes_inuse++;
156 TMPFS_UNLOCK(tmp);
157
158 *node = nnode;
159 return 0;
160 }
161
162 /* --------------------------------------------------------------------- */
163
164 /*
165 * Destroys the node pointed to by node from the file system 'tmp'.
166 * If the node does not belong to the given mount point, the results are
167 * unpredicted.
168 *
169 * If the node references a directory; no entries are allowed because
170 * their removal could need a recursive algorithm, something forbidden in
171 * kernel space. Furthermore, there is not need to provide such
172 * functionality (recursive removal) because the only primitives offered
173 * to the user are the removal of empty directories and the deletion of
174 * individual files.
175 *
176 * Note that nodes are not really deleted; in fact, when a node has been
177 * allocated, it cannot be deleted during the whole life of the file
178 * system. Instead, they are moved to the available list and remain there
179 * until reused.
180 */
181 void
182 tmpfs_free_node(struct tmpfs_mount *tmp, struct tmpfs_node *node)
183 {
184 size_t pages = 0;
185
186 #ifdef INVARIANTS
187 TMPFS_NODE_LOCK(node);
188 MPASS(node->tn_vnode == NULL);
189 TMPFS_NODE_UNLOCK(node);
190 #endif
191
192 TMPFS_LOCK(tmp);
193 LIST_REMOVE(node, tn_entries);
194 tmp->tm_nodes_inuse--;
195 TMPFS_UNLOCK(tmp);
196
197 switch (node->tn_type) {
198 case VNON:
199 /* Do not do anything. VNON is provided to let the
200 * allocation routine clean itself easily by avoiding
201 * duplicating code in it. */
202 /* FALLTHROUGH */
203 case VBLK:
204 /* FALLTHROUGH */
205 case VCHR:
206 /* FALLTHROUGH */
207 case VDIR:
208 /* FALLTHROUGH */
209 case VFIFO:
210 /* FALLTHROUGH */
211 case VSOCK:
212 break;
213
214 case VLNK:
215 free(node->tn_link, M_TMPFSNAME);
216 break;
217
218 case VREG:
219 if (node->tn_reg.tn_aobj != NULL)
220 vm_object_deallocate(node->tn_reg.tn_aobj);
221 pages = node->tn_reg.tn_aobj_pages;
222 break;
223
224 default:
225 MPASS(0);
226 break;
227 }
228
229 free_unr(tmp->tm_ino_unr, node->tn_id);
230 uma_zfree(tmp->tm_node_pool, node);
231
232 TMPFS_LOCK(tmp);
233 tmp->tm_pages_used -= pages;
234 TMPFS_UNLOCK(tmp);
235 }
236
237 /* --------------------------------------------------------------------- */
238
239 /*
240 * Allocates a new directory entry for the node node with a name of name.
241 * The new directory entry is returned in *de.
242 *
243 * The link count of node is increased by one to reflect the new object
244 * referencing it.
245 *
246 * Returns zero on success or an appropriate error code on failure.
247 */
248 int
249 tmpfs_alloc_dirent(struct tmpfs_mount *tmp, struct tmpfs_node *node,
250 const char *name, uint16_t len, struct tmpfs_dirent **de)
251 {
252 struct tmpfs_dirent *nde;
253
254 nde = (struct tmpfs_dirent *)uma_zalloc(
255 tmp->tm_dirent_pool, M_WAITOK);
256 nde->td_name = malloc(len, M_TMPFSNAME, M_WAITOK);
257 nde->td_namelen = len;
258 memcpy(nde->td_name, name, len);
259
260 nde->td_node = node;
261 node->tn_links++;
262
263 *de = nde;
264
265 return 0;
266 }
267
268 /* --------------------------------------------------------------------- */
269
270 /*
271 * Frees a directory entry. It is the caller's responsibility to destroy
272 * the node referenced by it if needed.
273 *
274 * The link count of node is decreased by one to reflect the removal of an
275 * object that referenced it. This only happens if 'node_exists' is true;
276 * otherwise the function will not access the node referred to by the
277 * directory entry, as it may already have been released from the outside.
278 */
279 void
280 tmpfs_free_dirent(struct tmpfs_mount *tmp, struct tmpfs_dirent *de,
281 boolean_t node_exists)
282 {
283 if (node_exists) {
284 struct tmpfs_node *node;
285
286 node = de->td_node;
287
288 MPASS(node->tn_links > 0);
289 node->tn_links--;
290 }
291
292 free(de->td_name, M_TMPFSNAME);
293 uma_zfree(tmp->tm_dirent_pool, de);
294 }
295
296 /* --------------------------------------------------------------------- */
297
298 /*
299 * Allocates a new vnode for the node node or returns a new reference to
300 * an existing one if the node had already a vnode referencing it. The
301 * resulting locked vnode is returned in *vpp.
302 *
303 * Returns zero on success or an appropriate error code on failure.
304 */
305 int
306 tmpfs_alloc_vp(struct mount *mp, struct tmpfs_node *node, int lkflag,
307 struct vnode **vpp, struct thread *td)
308 {
309 int error = 0;
310 struct vnode *vp;
311
312 loop:
313 TMPFS_NODE_LOCK(node);
314 if ((vp = node->tn_vnode) != NULL) {
315 VI_LOCK(vp);
316 TMPFS_NODE_UNLOCK(node);
317 vholdl(vp);
318 (void) vget(vp, lkflag | LK_INTERLOCK | LK_RETRY, td);
319 vdrop(vp);
320
321 /*
322 * Make sure the vnode is still there after
323 * getting the interlock to avoid racing a free.
324 */
325 if (node->tn_vnode == NULL || node->tn_vnode != vp) {
326 vput(vp);
327 goto loop;
328 }
329
330 goto out;
331 }
332
333 /*
334 * otherwise lock the vp list while we call getnewvnode
335 * since that can block.
336 */
337 if (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) {
338 node->tn_vpstate |= TMPFS_VNODE_WANT;
339 error = msleep((caddr_t) &node->tn_vpstate,
340 TMPFS_NODE_MTX(node), PDROP | PCATCH,
341 "tmpfs_alloc_vp", 0);
342 if (error)
343 return error;
344
345 goto loop;
346 } else
347 node->tn_vpstate |= TMPFS_VNODE_ALLOCATING;
348
349 TMPFS_NODE_UNLOCK(node);
350
351 /* Get a new vnode and associate it with our node. */
352 error = getnewvnode("tmpfs", mp, &tmpfs_vnodeop_entries, &vp);
353 if (error != 0)
354 goto unlock;
355 MPASS(vp != NULL);
356
357 (void) vn_lock(vp, lkflag | LK_RETRY, td);
358
359 vp->v_data = node;
360 vp->v_type = node->tn_type;
361
362 /* Type-specific initialization. */
363 switch (node->tn_type) {
364 case VBLK:
365 /* FALLTHROUGH */
366 case VCHR:
367 /* FALLTHROUGH */
368 case VLNK:
369 /* FALLTHROUGH */
370 case VREG:
371 /* FALLTHROUGH */
372 case VSOCK:
373 break;
374 case VFIFO:
375 vp->v_op = &tmpfs_fifoop_entries;
376 break;
377 case VDIR:
378 if (node->tn_dir.tn_parent == node)
379 vp->v_vflag |= VV_ROOT;
380 break;
381
382 default:
383 MPASS(0);
384 }
385
386 vnode_pager_setsize(vp, node->tn_size);
387 error = insmntque(vp, mp);
388 if (error)
389 vp = NULL;
390
391 unlock:
392 TMPFS_NODE_LOCK(node);
393
394 MPASS(node->tn_vpstate & TMPFS_VNODE_ALLOCATING);
395 node->tn_vpstate &= ~TMPFS_VNODE_ALLOCATING;
396 node->tn_vnode = vp;
397
398 if (node->tn_vpstate & TMPFS_VNODE_WANT) {
399 node->tn_vpstate &= ~TMPFS_VNODE_WANT;
400 TMPFS_NODE_UNLOCK(node);
401 wakeup((caddr_t) &node->tn_vpstate);
402 } else
403 TMPFS_NODE_UNLOCK(node);
404
405 out:
406 *vpp = vp;
407
408 MPASS(IFF(error == 0, *vpp != NULL && VOP_ISLOCKED(*vpp, td)));
409 #ifdef INVARIANTS
410 TMPFS_NODE_LOCK(node);
411 MPASS(*vpp == node->tn_vnode);
412 TMPFS_NODE_UNLOCK(node);
413 #endif
414
415 return error;
416 }
417
418 /* --------------------------------------------------------------------- */
419
420 /*
421 * Destroys the association between the vnode vp and the node it
422 * references.
423 */
424 void
425 tmpfs_free_vp(struct vnode *vp)
426 {
427 struct tmpfs_node *node;
428
429 node = VP_TO_TMPFS_NODE(vp);
430
431 TMPFS_NODE_LOCK(node);
432 node->tn_vnode = NULL;
433 vp->v_data = NULL;
434 TMPFS_NODE_UNLOCK(node);
435 }
436
437 /* --------------------------------------------------------------------- */
438
439 /*
440 * Allocates a new file of type 'type' and adds it to the parent directory
441 * 'dvp'; this addition is done using the component name given in 'cnp'.
442 * The ownership of the new file is automatically assigned based on the
443 * credentials of the caller (through 'cnp'), the group is set based on
444 * the parent directory and the mode is determined from the 'vap' argument.
445 * If successful, *vpp holds a vnode to the newly created file and zero
446 * is returned. Otherwise *vpp is NULL and the function returns an
447 * appropriate error code.
448 */
449 int
450 tmpfs_alloc_file(struct vnode *dvp, struct vnode **vpp, struct vattr *vap,
451 struct componentname *cnp, char *target)
452 {
453 int error;
454 struct tmpfs_dirent *de;
455 struct tmpfs_mount *tmp;
456 struct tmpfs_node *dnode;
457 struct tmpfs_node *node;
458 struct tmpfs_node *parent;
459
460 MPASS(VOP_ISLOCKED(dvp, cnp->cn_thread));
461 MPASS(cnp->cn_flags & HASBUF);
462
463 tmp = VFS_TO_TMPFS(dvp->v_mount);
464 dnode = VP_TO_TMPFS_DIR(dvp);
465 *vpp = NULL;
466
467 /* If the entry we are creating is a directory, we cannot overflow
468 * the number of links of its parent, because it will get a new
469 * link. */
470 if (vap->va_type == VDIR) {
471 /* Ensure that we do not overflow the maximum number of links
472 * imposed by the system. */
473 MPASS(dnode->tn_links <= LINK_MAX);
474 if (dnode->tn_links == LINK_MAX) {
475 error = EMLINK;
476 goto out;
477 }
478
479 parent = dnode;
480 MPASS(parent != NULL);
481 } else
482 parent = NULL;
483
484 /* Allocate a node that represents the new file. */
485 error = tmpfs_alloc_node(tmp, vap->va_type, cnp->cn_cred->cr_uid,
486 dnode->tn_gid, vap->va_mode, parent, target, vap->va_rdev,
487 cnp->cn_thread, &node);
488 if (error != 0)
489 goto out;
490
491 /* Allocate a directory entry that points to the new file. */
492 error = tmpfs_alloc_dirent(tmp, node, cnp->cn_nameptr, cnp->cn_namelen,
493 &de);
494 if (error != 0) {
495 tmpfs_free_node(tmp, node);
496 goto out;
497 }
498
499 /* Allocate a vnode for the new file. */
500 error = tmpfs_alloc_vp(dvp->v_mount, node, LK_EXCLUSIVE, vpp,
501 cnp->cn_thread);
502 if (error != 0) {
503 tmpfs_free_dirent(tmp, de, TRUE);
504 tmpfs_free_node(tmp, node);
505 goto out;
506 }
507
508 /* Now that all required items are allocated, we can proceed to
509 * insert the new node into the directory, an operation that
510 * cannot fail. */
511 tmpfs_dir_attach(dvp, de);
512
513 out:
514
515 return error;
516 }
517
518 /* --------------------------------------------------------------------- */
519
520 /*
521 * Attaches the directory entry de to the directory represented by vp.
522 * Note that this does not change the link count of the node pointed by
523 * the directory entry, as this is done by tmpfs_alloc_dirent.
524 */
525 void
526 tmpfs_dir_attach(struct vnode *vp, struct tmpfs_dirent *de)
527 {
528 struct tmpfs_node *dnode;
529
530 ASSERT_VOP_ELOCKED(vp, __func__);
531 dnode = VP_TO_TMPFS_DIR(vp);
532 TAILQ_INSERT_TAIL(&dnode->tn_dir.tn_dirhead, de, td_entries);
533 dnode->tn_size += sizeof(struct tmpfs_dirent);
534 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
535 TMPFS_NODE_MODIFIED;
536 }
537
538 /* --------------------------------------------------------------------- */
539
540 /*
541 * Detaches the directory entry de from the directory represented by vp.
542 * Note that this does not change the link count of the node pointed by
543 * the directory entry, as this is done by tmpfs_free_dirent.
544 */
545 void
546 tmpfs_dir_detach(struct vnode *vp, struct tmpfs_dirent *de)
547 {
548 struct tmpfs_node *dnode;
549
550 ASSERT_VOP_ELOCKED(vp, __func__);
551 dnode = VP_TO_TMPFS_DIR(vp);
552
553 if (dnode->tn_dir.tn_readdir_lastp == de) {
554 dnode->tn_dir.tn_readdir_lastn = 0;
555 dnode->tn_dir.tn_readdir_lastp = NULL;
556 }
557
558 TAILQ_REMOVE(&dnode->tn_dir.tn_dirhead, de, td_entries);
559 dnode->tn_size -= sizeof(struct tmpfs_dirent);
560 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
561 TMPFS_NODE_MODIFIED;
562 }
563
564 /* --------------------------------------------------------------------- */
565
566 /*
567 * Looks for a directory entry in the directory represented by node.
568 * 'cnp' describes the name of the entry to look for. Note that the .
569 * and .. components are not allowed as they do not physically exist
570 * within directories.
571 *
572 * Returns a pointer to the entry when found, otherwise NULL.
573 */
574 struct tmpfs_dirent *
575 tmpfs_dir_lookup(struct tmpfs_node *node, struct tmpfs_node *f,
576 struct componentname *cnp)
577 {
578 boolean_t found;
579 struct tmpfs_dirent *de;
580
581 MPASS(IMPLIES(cnp->cn_namelen == 1, cnp->cn_nameptr[0] != '.'));
582 MPASS(IMPLIES(cnp->cn_namelen == 2, !(cnp->cn_nameptr[0] == '.' &&
583 cnp->cn_nameptr[1] == '.')));
584 TMPFS_VALIDATE_DIR(node);
585
586 found = 0;
587 TAILQ_FOREACH(de, &node->tn_dir.tn_dirhead, td_entries) {
588 if (f != NULL && de->td_node != f)
589 continue;
590 MPASS(cnp->cn_namelen < 0xffff);
591 if (de->td_namelen == (uint16_t)cnp->cn_namelen &&
592 memcmp(de->td_name, cnp->cn_nameptr, de->td_namelen) == 0) {
593 found = 1;
594 break;
595 }
596 }
597 node->tn_status |= TMPFS_NODE_ACCESSED;
598
599 return found ? de : NULL;
600 }
601
602 /* --------------------------------------------------------------------- */
603
604 /*
605 * Helper function for tmpfs_readdir. Creates a '.' entry for the given
606 * directory and returns it in the uio space. The function returns 0
607 * on success, -1 if there was not enough space in the uio structure to
608 * hold the directory entry or an appropriate error code if another
609 * error happens.
610 */
611 int
612 tmpfs_dir_getdotdent(struct tmpfs_node *node, struct uio *uio)
613 {
614 int error;
615 struct dirent dent;
616
617 TMPFS_VALIDATE_DIR(node);
618 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOT);
619
620 dent.d_fileno = node->tn_id;
621 dent.d_type = DT_DIR;
622 dent.d_namlen = 1;
623 dent.d_name[0] = '.';
624 dent.d_name[1] = '\0';
625 dent.d_reclen = GENERIC_DIRSIZ(&dent);
626
627 if (dent.d_reclen > uio->uio_resid)
628 error = -1;
629 else {
630 error = uiomove(&dent, dent.d_reclen, uio);
631 if (error == 0)
632 uio->uio_offset = TMPFS_DIRCOOKIE_DOTDOT;
633 }
634
635 node->tn_status |= TMPFS_NODE_ACCESSED;
636
637 return error;
638 }
639
640 /* --------------------------------------------------------------------- */
641
642 /*
643 * Helper function for tmpfs_readdir. Creates a '..' entry for the given
644 * directory and returns it in the uio space. The function returns 0
645 * on success, -1 if there was not enough space in the uio structure to
646 * hold the directory entry or an appropriate error code if another
647 * error happens.
648 */
649 int
650 tmpfs_dir_getdotdotdent(struct tmpfs_node *node, struct uio *uio)
651 {
652 int error;
653 struct dirent dent;
654
655 TMPFS_VALIDATE_DIR(node);
656 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT);
657
658 dent.d_fileno = node->tn_dir.tn_parent->tn_id;
659 dent.d_type = DT_DIR;
660 dent.d_namlen = 2;
661 dent.d_name[0] = '.';
662 dent.d_name[1] = '.';
663 dent.d_name[2] = '\0';
664 dent.d_reclen = GENERIC_DIRSIZ(&dent);
665
666 if (dent.d_reclen > uio->uio_resid)
667 error = -1;
668 else {
669 error = uiomove(&dent, dent.d_reclen, uio);
670 if (error == 0) {
671 struct tmpfs_dirent *de;
672
673 de = TAILQ_FIRST(&node->tn_dir.tn_dirhead);
674 if (de == NULL)
675 uio->uio_offset = TMPFS_DIRCOOKIE_EOF;
676 else
677 uio->uio_offset = tmpfs_dircookie(de);
678 }
679 }
680
681 node->tn_status |= TMPFS_NODE_ACCESSED;
682
683 return error;
684 }
685
686 /* --------------------------------------------------------------------- */
687
688 /*
689 * Lookup a directory entry by its associated cookie.
690 */
691 struct tmpfs_dirent *
692 tmpfs_dir_lookupbycookie(struct tmpfs_node *node, off_t cookie)
693 {
694 struct tmpfs_dirent *de;
695
696 if (cookie == node->tn_dir.tn_readdir_lastn &&
697 node->tn_dir.tn_readdir_lastp != NULL) {
698 return node->tn_dir.tn_readdir_lastp;
699 }
700
701 TAILQ_FOREACH(de, &node->tn_dir.tn_dirhead, td_entries) {
702 if (tmpfs_dircookie(de) == cookie) {
703 break;
704 }
705 }
706
707 return de;
708 }
709
710 /* --------------------------------------------------------------------- */
711
712 /*
713 * Helper function for tmpfs_readdir. Returns as much directory entries
714 * as can fit in the uio space. The read starts at uio->uio_offset.
715 * The function returns 0 on success, -1 if there was not enough space
716 * in the uio structure to hold the directory entry or an appropriate
717 * error code if another error happens.
718 */
719 int
720 tmpfs_dir_getdents(struct tmpfs_node *node, struct uio *uio, off_t *cntp)
721 {
722 int error;
723 off_t startcookie;
724 struct tmpfs_dirent *de;
725
726 TMPFS_VALIDATE_DIR(node);
727
728 /* Locate the first directory entry we have to return. We have cached
729 * the last readdir in the node, so use those values if appropriate.
730 * Otherwise do a linear scan to find the requested entry. */
731 startcookie = uio->uio_offset;
732 MPASS(startcookie != TMPFS_DIRCOOKIE_DOT);
733 MPASS(startcookie != TMPFS_DIRCOOKIE_DOTDOT);
734 if (startcookie == TMPFS_DIRCOOKIE_EOF) {
735 return 0;
736 } else {
737 de = tmpfs_dir_lookupbycookie(node, startcookie);
738 }
739 if (de == NULL) {
740 return EINVAL;
741 }
742
743 /* Read as much entries as possible; i.e., until we reach the end of
744 * the directory or we exhaust uio space. */
745 do {
746 struct dirent d;
747
748 /* Create a dirent structure representing the current
749 * tmpfs_node and fill it. */
750 d.d_fileno = de->td_node->tn_id;
751 switch (de->td_node->tn_type) {
752 case VBLK:
753 d.d_type = DT_BLK;
754 break;
755
756 case VCHR:
757 d.d_type = DT_CHR;
758 break;
759
760 case VDIR:
761 d.d_type = DT_DIR;
762 break;
763
764 case VFIFO:
765 d.d_type = DT_FIFO;
766 break;
767
768 case VLNK:
769 d.d_type = DT_LNK;
770 break;
771
772 case VREG:
773 d.d_type = DT_REG;
774 break;
775
776 case VSOCK:
777 d.d_type = DT_SOCK;
778 break;
779
780 default:
781 MPASS(0);
782 }
783 d.d_namlen = de->td_namelen;
784 MPASS(de->td_namelen < sizeof(d.d_name));
785 (void)memcpy(d.d_name, de->td_name, de->td_namelen);
786 d.d_name[de->td_namelen] = '\0';
787 d.d_reclen = GENERIC_DIRSIZ(&d);
788
789 /* Stop reading if the directory entry we are treating is
790 * bigger than the amount of data that can be returned. */
791 if (d.d_reclen > uio->uio_resid) {
792 error = -1;
793 break;
794 }
795
796 /* Copy the new dirent structure into the output buffer and
797 * advance pointers. */
798 error = uiomove(&d, d.d_reclen, uio);
799
800 (*cntp)++;
801 de = TAILQ_NEXT(de, td_entries);
802 } while (error == 0 && uio->uio_resid > 0 && de != NULL);
803
804 /* Update the offset and cache. */
805 if (de == NULL) {
806 uio->uio_offset = TMPFS_DIRCOOKIE_EOF;
807 node->tn_dir.tn_readdir_lastn = 0;
808 node->tn_dir.tn_readdir_lastp = NULL;
809 } else {
810 node->tn_dir.tn_readdir_lastn = uio->uio_offset = tmpfs_dircookie(de);
811 node->tn_dir.tn_readdir_lastp = de;
812 }
813
814 node->tn_status |= TMPFS_NODE_ACCESSED;
815 return error;
816 }
817
818 /* --------------------------------------------------------------------- */
819
820 /*
821 * Resizes the aobj associated to the regular file pointed to by vp to
822 * the size newsize. 'vp' must point to a vnode that represents a regular
823 * file. 'newsize' must be positive.
824 *
825 * Returns zero on success or an appropriate error code on failure.
826 */
827 int
828 tmpfs_reg_resize(struct vnode *vp, off_t newsize)
829 {
830 int error;
831 size_t newpages, oldpages;
832 struct tmpfs_mount *tmp;
833 struct tmpfs_node *node;
834 off_t oldsize;
835
836 MPASS(vp->v_type == VREG);
837 MPASS(newsize >= 0);
838
839 node = VP_TO_TMPFS_NODE(vp);
840 tmp = VFS_TO_TMPFS(vp->v_mount);
841
842 /* Convert the old and new sizes to the number of pages needed to
843 * store them. It may happen that we do not need to do anything
844 * because the last allocated page can accommodate the change on
845 * its own. */
846 oldsize = node->tn_size;
847 oldpages = round_page(oldsize) / PAGE_SIZE;
848 MPASS(oldpages == node->tn_reg.tn_aobj_pages);
849 newpages = round_page(newsize) / PAGE_SIZE;
850
851 if (newpages > oldpages &&
852 newpages - oldpages > TMPFS_PAGES_AVAIL(tmp)) {
853 error = ENOSPC;
854 goto out;
855 }
856
857 node->tn_reg.tn_aobj_pages = newpages;
858
859 TMPFS_LOCK(tmp);
860 tmp->tm_pages_used += (newpages - oldpages);
861 TMPFS_UNLOCK(tmp);
862
863 node->tn_size = newsize;
864 vnode_pager_setsize(vp, newsize);
865 if (newsize < oldsize) {
866 size_t zerolen = round_page(newsize) - newsize;
867 vm_object_t uobj = node->tn_reg.tn_aobj;
868 vm_page_t m;
869
870 /*
871 * free "backing store"
872 */
873 VM_OBJECT_LOCK(uobj);
874 if (newpages < oldpages) {
875 swap_pager_freespace(uobj,
876 newpages, oldpages - newpages);
877 vm_object_page_remove(uobj,
878 OFF_TO_IDX(newsize + PAGE_MASK), 0, FALSE);
879 }
880
881 /*
882 * zero out the truncated part of the last page.
883 */
884
885 if (zerolen > 0) {
886 m = vm_page_grab(uobj, OFF_TO_IDX(newsize),
887 VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
888 pmap_zero_page_area(m, PAGE_SIZE - zerolen,
889 zerolen);
890 vm_page_wakeup(m);
891 }
892 VM_OBJECT_UNLOCK(uobj);
893
894 }
895
896 error = 0;
897
898 out:
899 return error;
900 }
901
902 /* --------------------------------------------------------------------- */
903
904 /*
905 * Change flags of the given vnode.
906 * Caller should execute tmpfs_update on vp after a successful execution.
907 * The vnode must be locked on entry and remain locked on exit.
908 */
909 int
910 tmpfs_chflags(struct vnode *vp, int flags, struct ucred *cred, struct thread *p)
911 {
912 int error;
913 struct tmpfs_node *node;
914
915 MPASS(VOP_ISLOCKED(vp, p));
916
917 node = VP_TO_TMPFS_NODE(vp);
918
919 /* Disallow this operation if the file system is mounted read-only. */
920 if (vp->v_mount->mnt_flag & MNT_RDONLY)
921 return EROFS;
922
923 /*
924 * Callers may only modify the file flags on objects they
925 * have VADMIN rights for.
926 */
927 if ((error = VOP_ACCESS(vp, VADMIN, cred, p)))
928 return (error);
929 /*
930 * Unprivileged processes are not permitted to unset system
931 * flags, or modify flags if any system flags are set.
932 */
933 if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0)) {
934 if (node->tn_flags
935 & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND)) {
936 error = securelevel_gt(cred, 0);
937 if (error)
938 return (error);
939 }
940 /* Snapshot flag cannot be set or cleared */
941 if (((flags & SF_SNAPSHOT) != 0 &&
942 (node->tn_flags & SF_SNAPSHOT) == 0) ||
943 ((flags & SF_SNAPSHOT) == 0 &&
944 (node->tn_flags & SF_SNAPSHOT) != 0))
945 return (EPERM);
946 node->tn_flags = flags;
947 } else {
948 if (node->tn_flags
949 & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND) ||
950 (flags & UF_SETTABLE) != flags)
951 return (EPERM);
952 node->tn_flags &= SF_SETTABLE;
953 node->tn_flags |= (flags & UF_SETTABLE);
954 }
955 node->tn_status |= TMPFS_NODE_CHANGED;
956
957 MPASS(VOP_ISLOCKED(vp, p));
958
959 return 0;
960 }
961
962 /* --------------------------------------------------------------------- */
963
964 /*
965 * Change access mode on the given vnode.
966 * Caller should execute tmpfs_update on vp after a successful execution.
967 * The vnode must be locked on entry and remain locked on exit.
968 */
969 int
970 tmpfs_chmod(struct vnode *vp, mode_t mode, struct ucred *cred, struct thread *p)
971 {
972 int error;
973 struct tmpfs_node *node;
974
975 MPASS(VOP_ISLOCKED(vp, p));
976
977 node = VP_TO_TMPFS_NODE(vp);
978
979 /* Disallow this operation if the file system is mounted read-only. */
980 if (vp->v_mount->mnt_flag & MNT_RDONLY)
981 return EROFS;
982
983 /* Immutable or append-only files cannot be modified, either. */
984 if (node->tn_flags & (IMMUTABLE | APPEND))
985 return EPERM;
986
987 /*
988 * To modify the permissions on a file, must possess VADMIN
989 * for that file.
990 */
991 if ((error = VOP_ACCESS(vp, VADMIN, cred, p)))
992 return (error);
993
994 /*
995 * Privileged processes may set the sticky bit on non-directories,
996 * as well as set the setgid bit on a file with a group that the
997 * process is not a member of.
998 */
999 if (vp->v_type != VDIR && (mode & S_ISTXT)) {
1000 if (priv_check_cred(cred, PRIV_VFS_STICKYFILE, 0))
1001 return (EFTYPE);
1002 }
1003 if (!groupmember(node->tn_gid, cred) && (mode & S_ISGID)) {
1004 error = priv_check_cred(cred, PRIV_VFS_SETGID, 0);
1005 if (error)
1006 return (error);
1007 }
1008
1009
1010 node->tn_mode &= ~ALLPERMS;
1011 node->tn_mode |= mode & ALLPERMS;
1012
1013 node->tn_status |= TMPFS_NODE_CHANGED;
1014
1015 MPASS(VOP_ISLOCKED(vp, p));
1016
1017 return 0;
1018 }
1019
1020 /* --------------------------------------------------------------------- */
1021
1022 /*
1023 * Change ownership of the given vnode. At least one of uid or gid must
1024 * be different than VNOVAL. If one is set to that value, the attribute
1025 * is unchanged.
1026 * Caller should execute tmpfs_update on vp after a successful execution.
1027 * The vnode must be locked on entry and remain locked on exit.
1028 */
1029 int
1030 tmpfs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred,
1031 struct thread *p)
1032 {
1033 int error;
1034 struct tmpfs_node *node;
1035 uid_t ouid;
1036 gid_t ogid;
1037
1038 MPASS(VOP_ISLOCKED(vp, p));
1039
1040 node = VP_TO_TMPFS_NODE(vp);
1041
1042 /* Assign default values if they are unknown. */
1043 MPASS(uid != VNOVAL || gid != VNOVAL);
1044 if (uid == VNOVAL)
1045 uid = node->tn_uid;
1046 if (gid == VNOVAL)
1047 gid = node->tn_gid;
1048 MPASS(uid != VNOVAL && gid != VNOVAL);
1049
1050 /* Disallow this operation if the file system is mounted read-only. */
1051 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1052 return EROFS;
1053
1054 /* Immutable or append-only files cannot be modified, either. */
1055 if (node->tn_flags & (IMMUTABLE | APPEND))
1056 return EPERM;
1057
1058 /*
1059 * To modify the ownership of a file, must possess VADMIN for that
1060 * file.
1061 */
1062 if ((error = VOP_ACCESS(vp, VADMIN, cred, p)))
1063 return (error);
1064
1065 /*
1066 * To change the owner of a file, or change the group of a file to a
1067 * group of which we are not a member, the caller must have
1068 * privilege.
1069 */
1070 if ((uid != node->tn_uid ||
1071 (gid != node->tn_gid && !groupmember(gid, cred))) &&
1072 (error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0)))
1073 return (error);
1074
1075 ogid = node->tn_gid;
1076 ouid = node->tn_uid;
1077
1078 node->tn_uid = uid;
1079 node->tn_gid = gid;
1080
1081 node->tn_status |= TMPFS_NODE_CHANGED;
1082
1083 if ((node->tn_mode & (S_ISUID | S_ISGID)) && (ouid != uid || ogid != gid)) {
1084 if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID, 0))
1085 node->tn_mode &= ~(S_ISUID | S_ISGID);
1086 }
1087
1088 MPASS(VOP_ISLOCKED(vp, p));
1089
1090 return 0;
1091 }
1092
1093 /* --------------------------------------------------------------------- */
1094
1095 /*
1096 * Change size of the given vnode.
1097 * Caller should execute tmpfs_update on vp after a successful execution.
1098 * The vnode must be locked on entry and remain locked on exit.
1099 */
1100 int
1101 tmpfs_chsize(struct vnode *vp, u_quad_t size, struct ucred *cred,
1102 struct thread *p)
1103 {
1104 int error;
1105 struct tmpfs_node *node;
1106
1107 MPASS(VOP_ISLOCKED(vp, p));
1108
1109 node = VP_TO_TMPFS_NODE(vp);
1110
1111 /* Decide whether this is a valid operation based on the file type. */
1112 error = 0;
1113 switch (vp->v_type) {
1114 case VDIR:
1115 return EISDIR;
1116
1117 case VREG:
1118 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1119 return EROFS;
1120 break;
1121
1122 case VBLK:
1123 /* FALLTHROUGH */
1124 case VCHR:
1125 /* FALLTHROUGH */
1126 case VFIFO:
1127 /* Allow modifications of special files even if in the file
1128 * system is mounted read-only (we are not modifying the
1129 * files themselves, but the objects they represent). */
1130 return 0;
1131
1132 default:
1133 /* Anything else is unsupported. */
1134 return EOPNOTSUPP;
1135 }
1136
1137 /* Immutable or append-only files cannot be modified, either. */
1138 if (node->tn_flags & (IMMUTABLE | APPEND))
1139 return EPERM;
1140
1141 error = tmpfs_truncate(vp, size);
1142 /* tmpfs_truncate will raise the NOTE_EXTEND and NOTE_ATTRIB kevents
1143 * for us, as will update tn_status; no need to do that here. */
1144
1145 MPASS(VOP_ISLOCKED(vp, p));
1146
1147 return error;
1148 }
1149
1150 /* --------------------------------------------------------------------- */
1151
1152 /*
1153 * Change access and modification times of the given vnode.
1154 * Caller should execute tmpfs_update on vp after a successful execution.
1155 * The vnode must be locked on entry and remain locked on exit.
1156 */
1157 int
1158 tmpfs_chtimes(struct vnode *vp, struct timespec *atime, struct timespec *mtime,
1159 struct timespec *birthtime, int vaflags, struct ucred *cred, struct thread *l)
1160 {
1161 int error;
1162 struct tmpfs_node *node;
1163
1164 MPASS(VOP_ISLOCKED(vp, l));
1165
1166 node = VP_TO_TMPFS_NODE(vp);
1167
1168 /* Disallow this operation if the file system is mounted read-only. */
1169 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1170 return EROFS;
1171
1172 /* Immutable or append-only files cannot be modified, either. */
1173 if (node->tn_flags & (IMMUTABLE | APPEND))
1174 return EPERM;
1175
1176 /* Determine if the user have proper privilege to update time. */
1177 if (vaflags & VA_UTIMES_NULL) {
1178 error = VOP_ACCESS(vp, VADMIN, cred, l);
1179 if (error)
1180 error = VOP_ACCESS(vp, VWRITE, cred, l);
1181 } else
1182 error = VOP_ACCESS(vp, VADMIN, cred, l);
1183 if (error)
1184 return (error);
1185
1186 if (atime->tv_sec != VNOVAL && atime->tv_nsec != VNOVAL)
1187 node->tn_status |= TMPFS_NODE_ACCESSED;
1188
1189 if (mtime->tv_sec != VNOVAL && mtime->tv_nsec != VNOVAL)
1190 node->tn_status |= TMPFS_NODE_MODIFIED;
1191
1192 if (birthtime->tv_nsec != VNOVAL && birthtime->tv_nsec != VNOVAL)
1193 node->tn_status |= TMPFS_NODE_MODIFIED;
1194
1195 tmpfs_itimes(vp, atime, mtime);
1196
1197 if (birthtime->tv_nsec != VNOVAL && birthtime->tv_nsec != VNOVAL)
1198 node->tn_birthtime = *birthtime;
1199 MPASS(VOP_ISLOCKED(vp, l));
1200
1201 return 0;
1202 }
1203
1204 /* --------------------------------------------------------------------- */
1205 /* Sync timestamps */
1206 void
1207 tmpfs_itimes(struct vnode *vp, const struct timespec *acc,
1208 const struct timespec *mod)
1209 {
1210 struct tmpfs_node *node;
1211 struct timespec now;
1212
1213 node = VP_TO_TMPFS_NODE(vp);
1214
1215 if ((node->tn_status & (TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
1216 TMPFS_NODE_CHANGED)) == 0)
1217 return;
1218
1219 vfs_timestamp(&now);
1220 if (node->tn_status & TMPFS_NODE_ACCESSED) {
1221 if (acc == NULL)
1222 acc = &now;
1223 node->tn_atime = *acc;
1224 }
1225 if (node->tn_status & TMPFS_NODE_MODIFIED) {
1226 if (mod == NULL)
1227 mod = &now;
1228 node->tn_mtime = *mod;
1229 }
1230 if (node->tn_status & TMPFS_NODE_CHANGED) {
1231 node->tn_ctime = now;
1232 }
1233 node->tn_status &=
1234 ~(TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED | TMPFS_NODE_CHANGED);
1235 }
1236
1237 /* --------------------------------------------------------------------- */
1238
1239 void
1240 tmpfs_update(struct vnode *vp)
1241 {
1242
1243 tmpfs_itimes(vp, NULL, NULL);
1244 }
1245
1246 /* --------------------------------------------------------------------- */
1247
1248 int
1249 tmpfs_truncate(struct vnode *vp, off_t length)
1250 {
1251 boolean_t extended;
1252 int error;
1253 struct tmpfs_node *node;
1254
1255 node = VP_TO_TMPFS_NODE(vp);
1256 extended = length > node->tn_size;
1257
1258 if (length < 0) {
1259 error = EINVAL;
1260 goto out;
1261 }
1262
1263 if (node->tn_size == length) {
1264 error = 0;
1265 goto out;
1266 }
1267
1268 if (length > VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize)
1269 return (EFBIG);
1270
1271 error = tmpfs_reg_resize(vp, length);
1272 if (error == 0) {
1273 node->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1274 }
1275
1276 out:
1277 tmpfs_update(vp);
1278
1279 return error;
1280 }
Cache object: b63034cf1a14c1011b47027b9d3db19e
|