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: releng/8.0/sys/fs/tmpfs/tmpfs_subr.c 194766 2009-06-23 20:45:22Z kib $");
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 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 NULL /* XXXKIB - tmpfs needs swap reservation */);
147 nnode->tn_reg.tn_aobj_pages = 0;
148 break;
149
150 default:
151 panic("tmpfs_alloc_node: type %p %d", nnode, (int)nnode->tn_type);
152 }
153
154 TMPFS_LOCK(tmp);
155 LIST_INSERT_HEAD(&tmp->tm_nodes_used, nnode, tn_entries);
156 tmp->tm_nodes_inuse++;
157 TMPFS_UNLOCK(tmp);
158
159 *node = nnode;
160 return 0;
161 }
162
163 /* --------------------------------------------------------------------- */
164
165 /*
166 * Destroys the node pointed to by node from the file system 'tmp'.
167 * If the node does not belong to the given mount point, the results are
168 * unpredicted.
169 *
170 * If the node references a directory; no entries are allowed because
171 * their removal could need a recursive algorithm, something forbidden in
172 * kernel space. Furthermore, there is not need to provide such
173 * functionality (recursive removal) because the only primitives offered
174 * to the user are the removal of empty directories and the deletion of
175 * individual files.
176 *
177 * Note that nodes are not really deleted; in fact, when a node has been
178 * allocated, it cannot be deleted during the whole life of the file
179 * system. Instead, they are moved to the available list and remain there
180 * until reused.
181 */
182 void
183 tmpfs_free_node(struct tmpfs_mount *tmp, struct tmpfs_node *node)
184 {
185 size_t pages = 0;
186
187 #ifdef INVARIANTS
188 TMPFS_NODE_LOCK(node);
189 MPASS(node->tn_vnode == NULL);
190 TMPFS_NODE_UNLOCK(node);
191 #endif
192
193 TMPFS_LOCK(tmp);
194 LIST_REMOVE(node, tn_entries);
195 tmp->tm_nodes_inuse--;
196 TMPFS_UNLOCK(tmp);
197
198 switch (node->tn_type) {
199 case VNON:
200 /* Do not do anything. VNON is provided to let the
201 * allocation routine clean itself easily by avoiding
202 * duplicating code in it. */
203 /* FALLTHROUGH */
204 case VBLK:
205 /* FALLTHROUGH */
206 case VCHR:
207 /* FALLTHROUGH */
208 case VDIR:
209 /* FALLTHROUGH */
210 case VFIFO:
211 /* FALLTHROUGH */
212 case VSOCK:
213 break;
214
215 case VLNK:
216 free(node->tn_link, M_TMPFSNAME);
217 break;
218
219 case VREG:
220 if (node->tn_reg.tn_aobj != NULL)
221 vm_object_deallocate(node->tn_reg.tn_aobj);
222 pages = node->tn_reg.tn_aobj_pages;
223 break;
224
225 default:
226 panic("tmpfs_free_node: type %p %d", node, (int)node->tn_type);
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)
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, curthread);
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);
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 panic("tmpfs_alloc_vp: type %p %d", node, (int)node->tn_type);
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)));
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));
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, &node);
487 if (error != 0)
488 goto out;
489
490 /* Allocate a directory entry that points to the new file. */
491 error = tmpfs_alloc_dirent(tmp, node, cnp->cn_nameptr, cnp->cn_namelen,
492 &de);
493 if (error != 0) {
494 tmpfs_free_node(tmp, node);
495 goto out;
496 }
497
498 /* Allocate a vnode for the new file. */
499 error = tmpfs_alloc_vp(dvp->v_mount, node, LK_EXCLUSIVE, vpp);
500 if (error != 0) {
501 tmpfs_free_dirent(tmp, de, TRUE);
502 tmpfs_free_node(tmp, node);
503 goto out;
504 }
505
506 /* Now that all required items are allocated, we can proceed to
507 * insert the new node into the directory, an operation that
508 * cannot fail. */
509 tmpfs_dir_attach(dvp, de);
510
511 out:
512
513 return error;
514 }
515
516 /* --------------------------------------------------------------------- */
517
518 /*
519 * Attaches the directory entry de to the directory represented by vp.
520 * Note that this does not change the link count of the node pointed by
521 * the directory entry, as this is done by tmpfs_alloc_dirent.
522 */
523 void
524 tmpfs_dir_attach(struct vnode *vp, struct tmpfs_dirent *de)
525 {
526 struct tmpfs_node *dnode;
527
528 ASSERT_VOP_ELOCKED(vp, __func__);
529 dnode = VP_TO_TMPFS_DIR(vp);
530 TAILQ_INSERT_TAIL(&dnode->tn_dir.tn_dirhead, de, td_entries);
531 dnode->tn_size += sizeof(struct tmpfs_dirent);
532 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
533 TMPFS_NODE_MODIFIED;
534 }
535
536 /* --------------------------------------------------------------------- */
537
538 /*
539 * Detaches the directory entry de from the directory represented by vp.
540 * Note that this does not change the link count of the node pointed by
541 * the directory entry, as this is done by tmpfs_free_dirent.
542 */
543 void
544 tmpfs_dir_detach(struct vnode *vp, struct tmpfs_dirent *de)
545 {
546 struct tmpfs_node *dnode;
547
548 ASSERT_VOP_ELOCKED(vp, __func__);
549 dnode = VP_TO_TMPFS_DIR(vp);
550
551 if (dnode->tn_dir.tn_readdir_lastp == de) {
552 dnode->tn_dir.tn_readdir_lastn = 0;
553 dnode->tn_dir.tn_readdir_lastp = NULL;
554 }
555
556 TAILQ_REMOVE(&dnode->tn_dir.tn_dirhead, de, td_entries);
557 dnode->tn_size -= sizeof(struct tmpfs_dirent);
558 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
559 TMPFS_NODE_MODIFIED;
560 }
561
562 /* --------------------------------------------------------------------- */
563
564 /*
565 * Looks for a directory entry in the directory represented by node.
566 * 'cnp' describes the name of the entry to look for. Note that the .
567 * and .. components are not allowed as they do not physically exist
568 * within directories.
569 *
570 * Returns a pointer to the entry when found, otherwise NULL.
571 */
572 struct tmpfs_dirent *
573 tmpfs_dir_lookup(struct tmpfs_node *node, struct tmpfs_node *f,
574 struct componentname *cnp)
575 {
576 boolean_t found;
577 struct tmpfs_dirent *de;
578
579 MPASS(IMPLIES(cnp->cn_namelen == 1, cnp->cn_nameptr[0] != '.'));
580 MPASS(IMPLIES(cnp->cn_namelen == 2, !(cnp->cn_nameptr[0] == '.' &&
581 cnp->cn_nameptr[1] == '.')));
582 TMPFS_VALIDATE_DIR(node);
583
584 found = 0;
585 TAILQ_FOREACH(de, &node->tn_dir.tn_dirhead, td_entries) {
586 if (f != NULL && de->td_node != f)
587 continue;
588 MPASS(cnp->cn_namelen < 0xffff);
589 if (de->td_namelen == (uint16_t)cnp->cn_namelen &&
590 bcmp(de->td_name, cnp->cn_nameptr, de->td_namelen) == 0) {
591 found = 1;
592 break;
593 }
594 }
595 node->tn_status |= TMPFS_NODE_ACCESSED;
596
597 return found ? de : NULL;
598 }
599
600 /* --------------------------------------------------------------------- */
601
602 /*
603 * Helper function for tmpfs_readdir. Creates a '.' entry for the given
604 * directory and returns it in the uio space. The function returns 0
605 * on success, -1 if there was not enough space in the uio structure to
606 * hold the directory entry or an appropriate error code if another
607 * error happens.
608 */
609 int
610 tmpfs_dir_getdotdent(struct tmpfs_node *node, struct uio *uio)
611 {
612 int error;
613 struct dirent dent;
614
615 TMPFS_VALIDATE_DIR(node);
616 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOT);
617
618 dent.d_fileno = node->tn_id;
619 dent.d_type = DT_DIR;
620 dent.d_namlen = 1;
621 dent.d_name[0] = '.';
622 dent.d_name[1] = '\0';
623 dent.d_reclen = GENERIC_DIRSIZ(&dent);
624
625 if (dent.d_reclen > uio->uio_resid)
626 error = -1;
627 else {
628 error = uiomove(&dent, dent.d_reclen, uio);
629 if (error == 0)
630 uio->uio_offset = TMPFS_DIRCOOKIE_DOTDOT;
631 }
632
633 node->tn_status |= TMPFS_NODE_ACCESSED;
634
635 return error;
636 }
637
638 /* --------------------------------------------------------------------- */
639
640 /*
641 * Helper function for tmpfs_readdir. Creates a '..' entry for the given
642 * directory and returns it in the uio space. The function returns 0
643 * on success, -1 if there was not enough space in the uio structure to
644 * hold the directory entry or an appropriate error code if another
645 * error happens.
646 */
647 int
648 tmpfs_dir_getdotdotdent(struct tmpfs_node *node, struct uio *uio)
649 {
650 int error;
651 struct dirent dent;
652
653 TMPFS_VALIDATE_DIR(node);
654 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT);
655
656 dent.d_fileno = node->tn_dir.tn_parent->tn_id;
657 dent.d_type = DT_DIR;
658 dent.d_namlen = 2;
659 dent.d_name[0] = '.';
660 dent.d_name[1] = '.';
661 dent.d_name[2] = '\0';
662 dent.d_reclen = GENERIC_DIRSIZ(&dent);
663
664 if (dent.d_reclen > uio->uio_resid)
665 error = -1;
666 else {
667 error = uiomove(&dent, dent.d_reclen, uio);
668 if (error == 0) {
669 struct tmpfs_dirent *de;
670
671 de = TAILQ_FIRST(&node->tn_dir.tn_dirhead);
672 if (de == NULL)
673 uio->uio_offset = TMPFS_DIRCOOKIE_EOF;
674 else
675 uio->uio_offset = tmpfs_dircookie(de);
676 }
677 }
678
679 node->tn_status |= TMPFS_NODE_ACCESSED;
680
681 return error;
682 }
683
684 /* --------------------------------------------------------------------- */
685
686 /*
687 * Lookup a directory entry by its associated cookie.
688 */
689 struct tmpfs_dirent *
690 tmpfs_dir_lookupbycookie(struct tmpfs_node *node, off_t cookie)
691 {
692 struct tmpfs_dirent *de;
693
694 if (cookie == node->tn_dir.tn_readdir_lastn &&
695 node->tn_dir.tn_readdir_lastp != NULL) {
696 return node->tn_dir.tn_readdir_lastp;
697 }
698
699 TAILQ_FOREACH(de, &node->tn_dir.tn_dirhead, td_entries) {
700 if (tmpfs_dircookie(de) == cookie) {
701 break;
702 }
703 }
704
705 return de;
706 }
707
708 /* --------------------------------------------------------------------- */
709
710 /*
711 * Helper function for tmpfs_readdir. Returns as much directory entries
712 * as can fit in the uio space. The read starts at uio->uio_offset.
713 * The function returns 0 on success, -1 if there was not enough space
714 * in the uio structure to hold the directory entry or an appropriate
715 * error code if another error happens.
716 */
717 int
718 tmpfs_dir_getdents(struct tmpfs_node *node, struct uio *uio, off_t *cntp)
719 {
720 int error;
721 off_t startcookie;
722 struct tmpfs_dirent *de;
723
724 TMPFS_VALIDATE_DIR(node);
725
726 /* Locate the first directory entry we have to return. We have cached
727 * the last readdir in the node, so use those values if appropriate.
728 * Otherwise do a linear scan to find the requested entry. */
729 startcookie = uio->uio_offset;
730 MPASS(startcookie != TMPFS_DIRCOOKIE_DOT);
731 MPASS(startcookie != TMPFS_DIRCOOKIE_DOTDOT);
732 if (startcookie == TMPFS_DIRCOOKIE_EOF) {
733 return 0;
734 } else {
735 de = tmpfs_dir_lookupbycookie(node, startcookie);
736 }
737 if (de == NULL) {
738 return EINVAL;
739 }
740
741 /* Read as much entries as possible; i.e., until we reach the end of
742 * the directory or we exhaust uio space. */
743 do {
744 struct dirent d;
745
746 /* Create a dirent structure representing the current
747 * tmpfs_node and fill it. */
748 d.d_fileno = de->td_node->tn_id;
749 switch (de->td_node->tn_type) {
750 case VBLK:
751 d.d_type = DT_BLK;
752 break;
753
754 case VCHR:
755 d.d_type = DT_CHR;
756 break;
757
758 case VDIR:
759 d.d_type = DT_DIR;
760 break;
761
762 case VFIFO:
763 d.d_type = DT_FIFO;
764 break;
765
766 case VLNK:
767 d.d_type = DT_LNK;
768 break;
769
770 case VREG:
771 d.d_type = DT_REG;
772 break;
773
774 case VSOCK:
775 d.d_type = DT_SOCK;
776 break;
777
778 default:
779 panic("tmpfs_dir_getdents: type %p %d",
780 de->td_node, (int)de->td_node->tn_type);
781 }
782 d.d_namlen = de->td_namelen;
783 MPASS(de->td_namelen < sizeof(d.d_name));
784 (void)memcpy(d.d_name, de->td_name, de->td_namelen);
785 d.d_name[de->td_namelen] = '\0';
786 d.d_reclen = GENERIC_DIRSIZ(&d);
787
788 /* Stop reading if the directory entry we are treating is
789 * bigger than the amount of data that can be returned. */
790 if (d.d_reclen > uio->uio_resid) {
791 error = -1;
792 break;
793 }
794
795 /* Copy the new dirent structure into the output buffer and
796 * advance pointers. */
797 error = uiomove(&d, d.d_reclen, uio);
798
799 (*cntp)++;
800 de = TAILQ_NEXT(de, td_entries);
801 } while (error == 0 && uio->uio_resid > 0 && de != NULL);
802
803 /* Update the offset and cache. */
804 if (de == NULL) {
805 uio->uio_offset = TMPFS_DIRCOOKIE_EOF;
806 node->tn_dir.tn_readdir_lastn = 0;
807 node->tn_dir.tn_readdir_lastp = NULL;
808 } else {
809 node->tn_dir.tn_readdir_lastn = uio->uio_offset = tmpfs_dircookie(de);
810 node->tn_dir.tn_readdir_lastp = de;
811 }
812
813 node->tn_status |= TMPFS_NODE_ACCESSED;
814 return error;
815 }
816
817 /* --------------------------------------------------------------------- */
818
819 /*
820 * Resizes the aobj associated to the regular file pointed to by vp to
821 * the size newsize. 'vp' must point to a vnode that represents a regular
822 * file. 'newsize' must be positive.
823 *
824 * Returns zero on success or an appropriate error code on failure.
825 */
826 int
827 tmpfs_reg_resize(struct vnode *vp, off_t newsize)
828 {
829 int error;
830 size_t newpages, oldpages;
831 struct tmpfs_mount *tmp;
832 struct tmpfs_node *node;
833 off_t oldsize;
834
835 MPASS(vp->v_type == VREG);
836 MPASS(newsize >= 0);
837
838 node = VP_TO_TMPFS_NODE(vp);
839 tmp = VFS_TO_TMPFS(vp->v_mount);
840
841 /* Convert the old and new sizes to the number of pages needed to
842 * store them. It may happen that we do not need to do anything
843 * because the last allocated page can accommodate the change on
844 * its own. */
845 oldsize = node->tn_size;
846 oldpages = round_page(oldsize) / PAGE_SIZE;
847 MPASS(oldpages == node->tn_reg.tn_aobj_pages);
848 newpages = round_page(newsize) / PAGE_SIZE;
849
850 if (newpages > oldpages &&
851 newpages - oldpages > TMPFS_PAGES_AVAIL(tmp)) {
852 error = ENOSPC;
853 goto out;
854 }
855
856 node->tn_reg.tn_aobj_pages = newpages;
857
858 TMPFS_LOCK(tmp);
859 tmp->tm_pages_used += (newpages - oldpages);
860 TMPFS_UNLOCK(tmp);
861
862 node->tn_size = newsize;
863 vnode_pager_setsize(vp, newsize);
864 if (newsize < oldsize) {
865 size_t zerolen = round_page(newsize) - newsize;
866 vm_object_t uobj = node->tn_reg.tn_aobj;
867 vm_page_t m;
868
869 /*
870 * free "backing store"
871 */
872 VM_OBJECT_LOCK(uobj);
873 if (newpages < oldpages) {
874 swap_pager_freespace(uobj,
875 newpages, oldpages - newpages);
876 vm_object_page_remove(uobj,
877 OFF_TO_IDX(newsize + PAGE_MASK), 0, FALSE);
878 }
879
880 /*
881 * zero out the truncated part of the last page.
882 */
883
884 if (zerolen > 0) {
885 m = vm_page_grab(uobj, OFF_TO_IDX(newsize),
886 VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
887 pmap_zero_page_area(m, PAGE_SIZE - zerolen,
888 zerolen);
889 vm_page_wakeup(m);
890 }
891 VM_OBJECT_UNLOCK(uobj);
892
893 }
894
895 error = 0;
896
897 out:
898 return error;
899 }
900
901 /* --------------------------------------------------------------------- */
902
903 /*
904 * Change flags of the given vnode.
905 * Caller should execute tmpfs_update on vp after a successful execution.
906 * The vnode must be locked on entry and remain locked on exit.
907 */
908 int
909 tmpfs_chflags(struct vnode *vp, int flags, struct ucred *cred, struct thread *p)
910 {
911 int error;
912 struct tmpfs_node *node;
913
914 MPASS(VOP_ISLOCKED(vp));
915
916 node = VP_TO_TMPFS_NODE(vp);
917
918 /* Disallow this operation if the file system is mounted read-only. */
919 if (vp->v_mount->mnt_flag & MNT_RDONLY)
920 return EROFS;
921
922 /*
923 * Callers may only modify the file flags on objects they
924 * have VADMIN rights for.
925 */
926 if ((error = VOP_ACCESS(vp, VADMIN, cred, p)))
927 return (error);
928 /*
929 * Unprivileged processes are not permitted to unset system
930 * flags, or modify flags if any system flags are set.
931 */
932 if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0)) {
933 if (node->tn_flags
934 & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND)) {
935 error = securelevel_gt(cred, 0);
936 if (error)
937 return (error);
938 }
939 /* Snapshot flag cannot be set or cleared */
940 if (((flags & SF_SNAPSHOT) != 0 &&
941 (node->tn_flags & SF_SNAPSHOT) == 0) ||
942 ((flags & SF_SNAPSHOT) == 0 &&
943 (node->tn_flags & SF_SNAPSHOT) != 0))
944 return (EPERM);
945 node->tn_flags = flags;
946 } else {
947 if (node->tn_flags
948 & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND) ||
949 (flags & UF_SETTABLE) != flags)
950 return (EPERM);
951 node->tn_flags &= SF_SETTABLE;
952 node->tn_flags |= (flags & UF_SETTABLE);
953 }
954 node->tn_status |= TMPFS_NODE_CHANGED;
955
956 MPASS(VOP_ISLOCKED(vp));
957
958 return 0;
959 }
960
961 /* --------------------------------------------------------------------- */
962
963 /*
964 * Change access mode on the given vnode.
965 * Caller should execute tmpfs_update on vp after a successful execution.
966 * The vnode must be locked on entry and remain locked on exit.
967 */
968 int
969 tmpfs_chmod(struct vnode *vp, mode_t mode, struct ucred *cred, struct thread *p)
970 {
971 int error;
972 struct tmpfs_node *node;
973
974 MPASS(VOP_ISLOCKED(vp));
975
976 node = VP_TO_TMPFS_NODE(vp);
977
978 /* Disallow this operation if the file system is mounted read-only. */
979 if (vp->v_mount->mnt_flag & MNT_RDONLY)
980 return EROFS;
981
982 /* Immutable or append-only files cannot be modified, either. */
983 if (node->tn_flags & (IMMUTABLE | APPEND))
984 return EPERM;
985
986 /*
987 * To modify the permissions on a file, must possess VADMIN
988 * for that file.
989 */
990 if ((error = VOP_ACCESS(vp, VADMIN, cred, p)))
991 return (error);
992
993 /*
994 * Privileged processes may set the sticky bit on non-directories,
995 * as well as set the setgid bit on a file with a group that the
996 * process is not a member of.
997 */
998 if (vp->v_type != VDIR && (mode & S_ISTXT)) {
999 if (priv_check_cred(cred, PRIV_VFS_STICKYFILE, 0))
1000 return (EFTYPE);
1001 }
1002 if (!groupmember(node->tn_gid, cred) && (mode & S_ISGID)) {
1003 error = priv_check_cred(cred, PRIV_VFS_SETGID, 0);
1004 if (error)
1005 return (error);
1006 }
1007
1008
1009 node->tn_mode &= ~ALLPERMS;
1010 node->tn_mode |= mode & ALLPERMS;
1011
1012 node->tn_status |= TMPFS_NODE_CHANGED;
1013
1014 MPASS(VOP_ISLOCKED(vp));
1015
1016 return 0;
1017 }
1018
1019 /* --------------------------------------------------------------------- */
1020
1021 /*
1022 * Change ownership of the given vnode. At least one of uid or gid must
1023 * be different than VNOVAL. If one is set to that value, the attribute
1024 * is unchanged.
1025 * Caller should execute tmpfs_update on vp after a successful execution.
1026 * The vnode must be locked on entry and remain locked on exit.
1027 */
1028 int
1029 tmpfs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred,
1030 struct thread *p)
1031 {
1032 int error;
1033 struct tmpfs_node *node;
1034 uid_t ouid;
1035 gid_t ogid;
1036
1037 MPASS(VOP_ISLOCKED(vp));
1038
1039 node = VP_TO_TMPFS_NODE(vp);
1040
1041 /* Assign default values if they are unknown. */
1042 MPASS(uid != VNOVAL || gid != VNOVAL);
1043 if (uid == VNOVAL)
1044 uid = node->tn_uid;
1045 if (gid == VNOVAL)
1046 gid = node->tn_gid;
1047 MPASS(uid != VNOVAL && gid != VNOVAL);
1048
1049 /* Disallow this operation if the file system is mounted read-only. */
1050 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1051 return EROFS;
1052
1053 /* Immutable or append-only files cannot be modified, either. */
1054 if (node->tn_flags & (IMMUTABLE | APPEND))
1055 return EPERM;
1056
1057 /*
1058 * To modify the ownership of a file, must possess VADMIN for that
1059 * file.
1060 */
1061 if ((error = VOP_ACCESS(vp, VADMIN, cred, p)))
1062 return (error);
1063
1064 /*
1065 * To change the owner of a file, or change the group of a file to a
1066 * group of which we are not a member, the caller must have
1067 * privilege.
1068 */
1069 if ((uid != node->tn_uid ||
1070 (gid != node->tn_gid && !groupmember(gid, cred))) &&
1071 (error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0)))
1072 return (error);
1073
1074 ogid = node->tn_gid;
1075 ouid = node->tn_uid;
1076
1077 node->tn_uid = uid;
1078 node->tn_gid = gid;
1079
1080 node->tn_status |= TMPFS_NODE_CHANGED;
1081
1082 if ((node->tn_mode & (S_ISUID | S_ISGID)) && (ouid != uid || ogid != gid)) {
1083 if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID, 0))
1084 node->tn_mode &= ~(S_ISUID | S_ISGID);
1085 }
1086
1087 MPASS(VOP_ISLOCKED(vp));
1088
1089 return 0;
1090 }
1091
1092 /* --------------------------------------------------------------------- */
1093
1094 /*
1095 * Change size of the given vnode.
1096 * Caller should execute tmpfs_update on vp after a successful execution.
1097 * The vnode must be locked on entry and remain locked on exit.
1098 */
1099 int
1100 tmpfs_chsize(struct vnode *vp, u_quad_t size, struct ucred *cred,
1101 struct thread *p)
1102 {
1103 int error;
1104 struct tmpfs_node *node;
1105
1106 MPASS(VOP_ISLOCKED(vp));
1107
1108 node = VP_TO_TMPFS_NODE(vp);
1109
1110 /* Decide whether this is a valid operation based on the file type. */
1111 error = 0;
1112 switch (vp->v_type) {
1113 case VDIR:
1114 return EISDIR;
1115
1116 case VREG:
1117 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1118 return EROFS;
1119 break;
1120
1121 case VBLK:
1122 /* FALLTHROUGH */
1123 case VCHR:
1124 /* FALLTHROUGH */
1125 case VFIFO:
1126 /* Allow modifications of special files even if in the file
1127 * system is mounted read-only (we are not modifying the
1128 * files themselves, but the objects they represent). */
1129 return 0;
1130
1131 default:
1132 /* Anything else is unsupported. */
1133 return EOPNOTSUPP;
1134 }
1135
1136 /* Immutable or append-only files cannot be modified, either. */
1137 if (node->tn_flags & (IMMUTABLE | APPEND))
1138 return EPERM;
1139
1140 error = tmpfs_truncate(vp, size);
1141 /* tmpfs_truncate will raise the NOTE_EXTEND and NOTE_ATTRIB kevents
1142 * for us, as will update tn_status; no need to do that here. */
1143
1144 MPASS(VOP_ISLOCKED(vp));
1145
1146 return error;
1147 }
1148
1149 /* --------------------------------------------------------------------- */
1150
1151 /*
1152 * Change access and modification times of the given vnode.
1153 * Caller should execute tmpfs_update on vp after a successful execution.
1154 * The vnode must be locked on entry and remain locked on exit.
1155 */
1156 int
1157 tmpfs_chtimes(struct vnode *vp, struct timespec *atime, struct timespec *mtime,
1158 struct timespec *birthtime, int vaflags, struct ucred *cred, struct thread *l)
1159 {
1160 int error;
1161 struct tmpfs_node *node;
1162
1163 MPASS(VOP_ISLOCKED(vp));
1164
1165 node = VP_TO_TMPFS_NODE(vp);
1166
1167 /* Disallow this operation if the file system is mounted read-only. */
1168 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1169 return EROFS;
1170
1171 /* Immutable or append-only files cannot be modified, either. */
1172 if (node->tn_flags & (IMMUTABLE | APPEND))
1173 return EPERM;
1174
1175 /* Determine if the user have proper privilege to update time. */
1176 if (vaflags & VA_UTIMES_NULL) {
1177 error = VOP_ACCESS(vp, VADMIN, cred, l);
1178 if (error)
1179 error = VOP_ACCESS(vp, VWRITE, cred, l);
1180 } else
1181 error = VOP_ACCESS(vp, VADMIN, cred, l);
1182 if (error)
1183 return (error);
1184
1185 if (atime->tv_sec != VNOVAL && atime->tv_nsec != VNOVAL)
1186 node->tn_status |= TMPFS_NODE_ACCESSED;
1187
1188 if (mtime->tv_sec != VNOVAL && mtime->tv_nsec != VNOVAL)
1189 node->tn_status |= TMPFS_NODE_MODIFIED;
1190
1191 if (birthtime->tv_nsec != VNOVAL && birthtime->tv_nsec != VNOVAL)
1192 node->tn_status |= TMPFS_NODE_MODIFIED;
1193
1194 tmpfs_itimes(vp, atime, mtime);
1195
1196 if (birthtime->tv_nsec != VNOVAL && birthtime->tv_nsec != VNOVAL)
1197 node->tn_birthtime = *birthtime;
1198 MPASS(VOP_ISLOCKED(vp));
1199
1200 return 0;
1201 }
1202
1203 /* --------------------------------------------------------------------- */
1204 /* Sync timestamps */
1205 void
1206 tmpfs_itimes(struct vnode *vp, const struct timespec *acc,
1207 const struct timespec *mod)
1208 {
1209 struct tmpfs_node *node;
1210 struct timespec now;
1211
1212 node = VP_TO_TMPFS_NODE(vp);
1213
1214 if ((node->tn_status & (TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
1215 TMPFS_NODE_CHANGED)) == 0)
1216 return;
1217
1218 vfs_timestamp(&now);
1219 if (node->tn_status & TMPFS_NODE_ACCESSED) {
1220 if (acc == NULL)
1221 acc = &now;
1222 node->tn_atime = *acc;
1223 }
1224 if (node->tn_status & TMPFS_NODE_MODIFIED) {
1225 if (mod == NULL)
1226 mod = &now;
1227 node->tn_mtime = *mod;
1228 }
1229 if (node->tn_status & TMPFS_NODE_CHANGED) {
1230 node->tn_ctime = now;
1231 }
1232 node->tn_status &=
1233 ~(TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED | TMPFS_NODE_CHANGED);
1234 }
1235
1236 /* --------------------------------------------------------------------- */
1237
1238 void
1239 tmpfs_update(struct vnode *vp)
1240 {
1241
1242 tmpfs_itimes(vp, NULL, NULL);
1243 }
1244
1245 /* --------------------------------------------------------------------- */
1246
1247 int
1248 tmpfs_truncate(struct vnode *vp, off_t length)
1249 {
1250 int error;
1251 struct tmpfs_node *node;
1252
1253 node = VP_TO_TMPFS_NODE(vp);
1254
1255 if (length < 0) {
1256 error = EINVAL;
1257 goto out;
1258 }
1259
1260 if (node->tn_size == length) {
1261 error = 0;
1262 goto out;
1263 }
1264
1265 if (length > VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize)
1266 return (EFBIG);
1267
1268 error = tmpfs_reg_resize(vp, length);
1269 if (error == 0) {
1270 node->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1271 }
1272
1273 out:
1274 tmpfs_update(vp);
1275
1276 return error;
1277 }
Cache object: 96551f8b0c5fc77694327cf8be33cdb9
|