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/9.0/sys/fs/tmpfs/tmpfs_subr.c 223677 2011-06-29 16:40:41Z alc $");
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 TMPFS_NODE_LOCK(nnode->tn_dir.tn_parent);
128 nnode->tn_dir.tn_parent->tn_links++;
129 TMPFS_NODE_UNLOCK(nnode->tn_dir.tn_parent);
130 break;
131
132 case VFIFO:
133 /* FALLTHROUGH */
134 case VSOCK:
135 break;
136
137 case VLNK:
138 MPASS(strlen(target) < MAXPATHLEN);
139 nnode->tn_size = strlen(target);
140 nnode->tn_link = malloc(nnode->tn_size, M_TMPFSNAME,
141 M_WAITOK);
142 memcpy(nnode->tn_link, target, nnode->tn_size);
143 break;
144
145 case VREG:
146 nnode->tn_reg.tn_aobj =
147 vm_pager_allocate(OBJT_SWAP, NULL, 0, VM_PROT_DEFAULT, 0,
148 NULL /* XXXKIB - tmpfs needs swap reservation */);
149 break;
150
151 default:
152 panic("tmpfs_alloc_node: type %p %d", nnode, (int)nnode->tn_type);
153 }
154
155 TMPFS_LOCK(tmp);
156 LIST_INSERT_HEAD(&tmp->tm_nodes_used, nnode, tn_entries);
157 tmp->tm_nodes_inuse++;
158 TMPFS_UNLOCK(tmp);
159
160 *node = nnode;
161 return 0;
162 }
163
164 /* --------------------------------------------------------------------- */
165
166 /*
167 * Destroys the node pointed to by node from the file system 'tmp'.
168 * If the node does not belong to the given mount point, the results are
169 * unpredicted.
170 *
171 * If the node references a directory; no entries are allowed because
172 * their removal could need a recursive algorithm, something forbidden in
173 * kernel space. Furthermore, there is not need to provide such
174 * functionality (recursive removal) because the only primitives offered
175 * to the user are the removal of empty directories and the deletion of
176 * individual files.
177 *
178 * Note that nodes are not really deleted; in fact, when a node has been
179 * allocated, it cannot be deleted during the whole life of the file
180 * system. Instead, they are moved to the available list and remain there
181 * until reused.
182 */
183 void
184 tmpfs_free_node(struct tmpfs_mount *tmp, struct tmpfs_node *node)
185 {
186 vm_object_t uobj;
187
188 #ifdef INVARIANTS
189 TMPFS_NODE_LOCK(node);
190 MPASS(node->tn_vnode == NULL);
191 MPASS((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0);
192 TMPFS_NODE_UNLOCK(node);
193 #endif
194
195 TMPFS_LOCK(tmp);
196 LIST_REMOVE(node, tn_entries);
197 tmp->tm_nodes_inuse--;
198 TMPFS_UNLOCK(tmp);
199
200 switch (node->tn_type) {
201 case VNON:
202 /* Do not do anything. VNON is provided to let the
203 * allocation routine clean itself easily by avoiding
204 * duplicating code in it. */
205 /* FALLTHROUGH */
206 case VBLK:
207 /* FALLTHROUGH */
208 case VCHR:
209 /* FALLTHROUGH */
210 case VDIR:
211 /* FALLTHROUGH */
212 case VFIFO:
213 /* FALLTHROUGH */
214 case VSOCK:
215 break;
216
217 case VLNK:
218 free(node->tn_link, M_TMPFSNAME);
219 break;
220
221 case VREG:
222 uobj = node->tn_reg.tn_aobj;
223 if (uobj != NULL) {
224 TMPFS_LOCK(tmp);
225 tmp->tm_pages_used -= uobj->size;
226 TMPFS_UNLOCK(tmp);
227 vm_object_deallocate(uobj);
228 }
229 break;
230
231 default:
232 panic("tmpfs_free_node: type %p %d", node, (int)node->tn_type);
233 }
234
235 free_unr(tmp->tm_ino_unr, node->tn_id);
236 uma_zfree(tmp->tm_node_pool, node);
237 }
238
239 /* --------------------------------------------------------------------- */
240
241 /*
242 * Allocates a new directory entry for the node node with a name of name.
243 * The new directory entry is returned in *de.
244 *
245 * The link count of node is increased by one to reflect the new object
246 * referencing it.
247 *
248 * Returns zero on success or an appropriate error code on failure.
249 */
250 int
251 tmpfs_alloc_dirent(struct tmpfs_mount *tmp, struct tmpfs_node *node,
252 const char *name, uint16_t len, struct tmpfs_dirent **de)
253 {
254 struct tmpfs_dirent *nde;
255
256 nde = (struct tmpfs_dirent *)uma_zalloc(
257 tmp->tm_dirent_pool, M_WAITOK);
258 nde->td_name = malloc(len, M_TMPFSNAME, M_WAITOK);
259 nde->td_namelen = len;
260 memcpy(nde->td_name, name, len);
261
262 nde->td_node = node;
263 if (node != NULL)
264 node->tn_links++;
265
266 *de = nde;
267
268 return 0;
269 }
270
271 /* --------------------------------------------------------------------- */
272
273 /*
274 * Frees a directory entry. It is the caller's responsibility to destroy
275 * the node referenced by it if needed.
276 *
277 * The link count of node is decreased by one to reflect the removal of an
278 * object that referenced it. This only happens if 'node_exists' is true;
279 * otherwise the function will not access the node referred to by the
280 * directory entry, as it may already have been released from the outside.
281 */
282 void
283 tmpfs_free_dirent(struct tmpfs_mount *tmp, struct tmpfs_dirent *de,
284 boolean_t node_exists)
285 {
286 if (node_exists) {
287 struct tmpfs_node *node;
288
289 node = de->td_node;
290 if (node != NULL) {
291 MPASS(node->tn_links > 0);
292 node->tn_links--;
293 }
294 }
295
296 free(de->td_name, M_TMPFSNAME);
297 uma_zfree(tmp->tm_dirent_pool, de);
298 }
299
300 /* --------------------------------------------------------------------- */
301
302 /*
303 * Allocates a new vnode for the node node or returns a new reference to
304 * an existing one if the node had already a vnode referencing it. The
305 * resulting locked vnode is returned in *vpp.
306 *
307 * Returns zero on success or an appropriate error code on failure.
308 */
309 int
310 tmpfs_alloc_vp(struct mount *mp, struct tmpfs_node *node, int lkflag,
311 struct vnode **vpp)
312 {
313 int error = 0;
314 struct vnode *vp;
315
316 loop:
317 TMPFS_NODE_LOCK(node);
318 if ((vp = node->tn_vnode) != NULL) {
319 MPASS((node->tn_vpstate & TMPFS_VNODE_DOOMED) == 0);
320 VI_LOCK(vp);
321 TMPFS_NODE_UNLOCK(node);
322 vholdl(vp);
323 (void) vget(vp, lkflag | LK_INTERLOCK | LK_RETRY, curthread);
324 vdrop(vp);
325
326 /*
327 * Make sure the vnode is still there after
328 * getting the interlock to avoid racing a free.
329 */
330 if (node->tn_vnode == NULL || node->tn_vnode != vp) {
331 vput(vp);
332 goto loop;
333 }
334
335 goto out;
336 }
337
338 if ((node->tn_vpstate & TMPFS_VNODE_DOOMED) ||
339 (node->tn_type == VDIR && node->tn_dir.tn_parent == NULL)) {
340 TMPFS_NODE_UNLOCK(node);
341 error = ENOENT;
342 vp = NULL;
343 goto out;
344 }
345
346 /*
347 * otherwise lock the vp list while we call getnewvnode
348 * since that can block.
349 */
350 if (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) {
351 node->tn_vpstate |= TMPFS_VNODE_WANT;
352 error = msleep((caddr_t) &node->tn_vpstate,
353 TMPFS_NODE_MTX(node), PDROP | PCATCH,
354 "tmpfs_alloc_vp", 0);
355 if (error)
356 return error;
357
358 goto loop;
359 } else
360 node->tn_vpstate |= TMPFS_VNODE_ALLOCATING;
361
362 TMPFS_NODE_UNLOCK(node);
363
364 /* Get a new vnode and associate it with our node. */
365 error = getnewvnode("tmpfs", mp, &tmpfs_vnodeop_entries, &vp);
366 if (error != 0)
367 goto unlock;
368 MPASS(vp != NULL);
369
370 (void) vn_lock(vp, lkflag | LK_RETRY);
371
372 vp->v_data = node;
373 vp->v_type = node->tn_type;
374
375 /* Type-specific initialization. */
376 switch (node->tn_type) {
377 case VBLK:
378 /* FALLTHROUGH */
379 case VCHR:
380 /* FALLTHROUGH */
381 case VLNK:
382 /* FALLTHROUGH */
383 case VREG:
384 /* FALLTHROUGH */
385 case VSOCK:
386 break;
387 case VFIFO:
388 vp->v_op = &tmpfs_fifoop_entries;
389 break;
390 case VDIR:
391 MPASS(node->tn_dir.tn_parent != NULL);
392 if (node->tn_dir.tn_parent == node)
393 vp->v_vflag |= VV_ROOT;
394 break;
395
396 default:
397 panic("tmpfs_alloc_vp: type %p %d", node, (int)node->tn_type);
398 }
399
400 vnode_pager_setsize(vp, node->tn_size);
401 error = insmntque(vp, mp);
402 if (error)
403 vp = NULL;
404
405 unlock:
406 TMPFS_NODE_LOCK(node);
407
408 MPASS(node->tn_vpstate & TMPFS_VNODE_ALLOCATING);
409 node->tn_vpstate &= ~TMPFS_VNODE_ALLOCATING;
410 node->tn_vnode = vp;
411
412 if (node->tn_vpstate & TMPFS_VNODE_WANT) {
413 node->tn_vpstate &= ~TMPFS_VNODE_WANT;
414 TMPFS_NODE_UNLOCK(node);
415 wakeup((caddr_t) &node->tn_vpstate);
416 } else
417 TMPFS_NODE_UNLOCK(node);
418
419 out:
420 *vpp = vp;
421
422 MPASS(IFF(error == 0, *vpp != NULL && VOP_ISLOCKED(*vpp)));
423 #ifdef INVARIANTS
424 TMPFS_NODE_LOCK(node);
425 MPASS(*vpp == node->tn_vnode);
426 TMPFS_NODE_UNLOCK(node);
427 #endif
428
429 return error;
430 }
431
432 /* --------------------------------------------------------------------- */
433
434 /*
435 * Destroys the association between the vnode vp and the node it
436 * references.
437 */
438 void
439 tmpfs_free_vp(struct vnode *vp)
440 {
441 struct tmpfs_node *node;
442
443 node = VP_TO_TMPFS_NODE(vp);
444
445 mtx_assert(TMPFS_NODE_MTX(node), MA_OWNED);
446 node->tn_vnode = NULL;
447 vp->v_data = NULL;
448 }
449
450 /* --------------------------------------------------------------------- */
451
452 /*
453 * Allocates a new file of type 'type' and adds it to the parent directory
454 * 'dvp'; this addition is done using the component name given in 'cnp'.
455 * The ownership of the new file is automatically assigned based on the
456 * credentials of the caller (through 'cnp'), the group is set based on
457 * the parent directory and the mode is determined from the 'vap' argument.
458 * If successful, *vpp holds a vnode to the newly created file and zero
459 * is returned. Otherwise *vpp is NULL and the function returns an
460 * appropriate error code.
461 */
462 int
463 tmpfs_alloc_file(struct vnode *dvp, struct vnode **vpp, struct vattr *vap,
464 struct componentname *cnp, char *target)
465 {
466 int error;
467 struct tmpfs_dirent *de;
468 struct tmpfs_mount *tmp;
469 struct tmpfs_node *dnode;
470 struct tmpfs_node *node;
471 struct tmpfs_node *parent;
472
473 MPASS(VOP_ISLOCKED(dvp));
474 MPASS(cnp->cn_flags & HASBUF);
475
476 tmp = VFS_TO_TMPFS(dvp->v_mount);
477 dnode = VP_TO_TMPFS_DIR(dvp);
478 *vpp = NULL;
479
480 /* If the entry we are creating is a directory, we cannot overflow
481 * the number of links of its parent, because it will get a new
482 * link. */
483 if (vap->va_type == VDIR) {
484 /* Ensure that we do not overflow the maximum number of links
485 * imposed by the system. */
486 MPASS(dnode->tn_links <= LINK_MAX);
487 if (dnode->tn_links == LINK_MAX) {
488 error = EMLINK;
489 goto out;
490 }
491
492 parent = dnode;
493 MPASS(parent != NULL);
494 } else
495 parent = NULL;
496
497 /* Allocate a node that represents the new file. */
498 error = tmpfs_alloc_node(tmp, vap->va_type, cnp->cn_cred->cr_uid,
499 dnode->tn_gid, vap->va_mode, parent, target, vap->va_rdev, &node);
500 if (error != 0)
501 goto out;
502
503 /* Allocate a directory entry that points to the new file. */
504 error = tmpfs_alloc_dirent(tmp, node, cnp->cn_nameptr, cnp->cn_namelen,
505 &de);
506 if (error != 0) {
507 tmpfs_free_node(tmp, node);
508 goto out;
509 }
510
511 /* Allocate a vnode for the new file. */
512 error = tmpfs_alloc_vp(dvp->v_mount, node, LK_EXCLUSIVE, vpp);
513 if (error != 0) {
514 tmpfs_free_dirent(tmp, de, TRUE);
515 tmpfs_free_node(tmp, node);
516 goto out;
517 }
518
519 /* Now that all required items are allocated, we can proceed to
520 * insert the new node into the directory, an operation that
521 * cannot fail. */
522 if (cnp->cn_flags & ISWHITEOUT)
523 tmpfs_dir_whiteout_remove(dvp, cnp);
524 tmpfs_dir_attach(dvp, de);
525
526 out:
527
528 return error;
529 }
530
531 /* --------------------------------------------------------------------- */
532
533 /*
534 * Attaches the directory entry de to the directory represented by vp.
535 * Note that this does not change the link count of the node pointed by
536 * the directory entry, as this is done by tmpfs_alloc_dirent.
537 */
538 void
539 tmpfs_dir_attach(struct vnode *vp, struct tmpfs_dirent *de)
540 {
541 struct tmpfs_node *dnode;
542
543 ASSERT_VOP_ELOCKED(vp, __func__);
544 dnode = VP_TO_TMPFS_DIR(vp);
545 TAILQ_INSERT_TAIL(&dnode->tn_dir.tn_dirhead, de, td_entries);
546 dnode->tn_size += sizeof(struct tmpfs_dirent);
547 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
548 TMPFS_NODE_MODIFIED;
549 }
550
551 /* --------------------------------------------------------------------- */
552
553 /*
554 * Detaches the directory entry de from the directory represented by vp.
555 * Note that this does not change the link count of the node pointed by
556 * the directory entry, as this is done by tmpfs_free_dirent.
557 */
558 void
559 tmpfs_dir_detach(struct vnode *vp, struct tmpfs_dirent *de)
560 {
561 struct tmpfs_node *dnode;
562
563 ASSERT_VOP_ELOCKED(vp, __func__);
564 dnode = VP_TO_TMPFS_DIR(vp);
565
566 if (dnode->tn_dir.tn_readdir_lastp == de) {
567 dnode->tn_dir.tn_readdir_lastn = 0;
568 dnode->tn_dir.tn_readdir_lastp = NULL;
569 }
570
571 TAILQ_REMOVE(&dnode->tn_dir.tn_dirhead, de, td_entries);
572 dnode->tn_size -= sizeof(struct tmpfs_dirent);
573 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
574 TMPFS_NODE_MODIFIED;
575 }
576
577 /* --------------------------------------------------------------------- */
578
579 /*
580 * Looks for a directory entry in the directory represented by node.
581 * 'cnp' describes the name of the entry to look for. Note that the .
582 * and .. components are not allowed as they do not physically exist
583 * within directories.
584 *
585 * Returns a pointer to the entry when found, otherwise NULL.
586 */
587 struct tmpfs_dirent *
588 tmpfs_dir_lookup(struct tmpfs_node *node, struct tmpfs_node *f,
589 struct componentname *cnp)
590 {
591 boolean_t found;
592 struct tmpfs_dirent *de;
593
594 MPASS(IMPLIES(cnp->cn_namelen == 1, cnp->cn_nameptr[0] != '.'));
595 MPASS(IMPLIES(cnp->cn_namelen == 2, !(cnp->cn_nameptr[0] == '.' &&
596 cnp->cn_nameptr[1] == '.')));
597 TMPFS_VALIDATE_DIR(node);
598
599 found = 0;
600 TAILQ_FOREACH(de, &node->tn_dir.tn_dirhead, td_entries) {
601 if (f != NULL && de->td_node != f)
602 continue;
603 MPASS(cnp->cn_namelen < 0xffff);
604 if (de->td_namelen == (uint16_t)cnp->cn_namelen &&
605 bcmp(de->td_name, cnp->cn_nameptr, de->td_namelen) == 0) {
606 found = 1;
607 break;
608 }
609 }
610 node->tn_status |= TMPFS_NODE_ACCESSED;
611
612 return found ? de : NULL;
613 }
614
615 /* --------------------------------------------------------------------- */
616
617 /*
618 * Helper function for tmpfs_readdir. Creates a '.' entry for the given
619 * directory and returns it in the uio space. The function returns 0
620 * on success, -1 if there was not enough space in the uio structure to
621 * hold the directory entry or an appropriate error code if another
622 * error happens.
623 */
624 int
625 tmpfs_dir_getdotdent(struct tmpfs_node *node, struct uio *uio)
626 {
627 int error;
628 struct dirent dent;
629
630 TMPFS_VALIDATE_DIR(node);
631 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOT);
632
633 dent.d_fileno = node->tn_id;
634 dent.d_type = DT_DIR;
635 dent.d_namlen = 1;
636 dent.d_name[0] = '.';
637 dent.d_name[1] = '\0';
638 dent.d_reclen = GENERIC_DIRSIZ(&dent);
639
640 if (dent.d_reclen > uio->uio_resid)
641 error = -1;
642 else {
643 error = uiomove(&dent, dent.d_reclen, uio);
644 if (error == 0)
645 uio->uio_offset = TMPFS_DIRCOOKIE_DOTDOT;
646 }
647
648 node->tn_status |= TMPFS_NODE_ACCESSED;
649
650 return error;
651 }
652
653 /* --------------------------------------------------------------------- */
654
655 /*
656 * Helper function for tmpfs_readdir. Creates a '..' entry for the given
657 * directory and returns it in the uio space. The function returns 0
658 * on success, -1 if there was not enough space in the uio structure to
659 * hold the directory entry or an appropriate error code if another
660 * error happens.
661 */
662 int
663 tmpfs_dir_getdotdotdent(struct tmpfs_node *node, struct uio *uio)
664 {
665 int error;
666 struct dirent dent;
667
668 TMPFS_VALIDATE_DIR(node);
669 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT);
670
671 /*
672 * Return ENOENT if the current node is already removed.
673 */
674 TMPFS_ASSERT_LOCKED(node);
675 if (node->tn_dir.tn_parent == NULL) {
676 return (ENOENT);
677 }
678
679 TMPFS_NODE_LOCK(node->tn_dir.tn_parent);
680 dent.d_fileno = node->tn_dir.tn_parent->tn_id;
681 TMPFS_NODE_UNLOCK(node->tn_dir.tn_parent);
682
683 dent.d_type = DT_DIR;
684 dent.d_namlen = 2;
685 dent.d_name[0] = '.';
686 dent.d_name[1] = '.';
687 dent.d_name[2] = '\0';
688 dent.d_reclen = GENERIC_DIRSIZ(&dent);
689
690 if (dent.d_reclen > uio->uio_resid)
691 error = -1;
692 else {
693 error = uiomove(&dent, dent.d_reclen, uio);
694 if (error == 0) {
695 struct tmpfs_dirent *de;
696
697 de = TAILQ_FIRST(&node->tn_dir.tn_dirhead);
698 if (de == NULL)
699 uio->uio_offset = TMPFS_DIRCOOKIE_EOF;
700 else
701 uio->uio_offset = tmpfs_dircookie(de);
702 }
703 }
704
705 node->tn_status |= TMPFS_NODE_ACCESSED;
706
707 return error;
708 }
709
710 /* --------------------------------------------------------------------- */
711
712 /*
713 * Lookup a directory entry by its associated cookie.
714 */
715 struct tmpfs_dirent *
716 tmpfs_dir_lookupbycookie(struct tmpfs_node *node, off_t cookie)
717 {
718 struct tmpfs_dirent *de;
719
720 if (cookie == node->tn_dir.tn_readdir_lastn &&
721 node->tn_dir.tn_readdir_lastp != NULL) {
722 return node->tn_dir.tn_readdir_lastp;
723 }
724
725 TAILQ_FOREACH(de, &node->tn_dir.tn_dirhead, td_entries) {
726 if (tmpfs_dircookie(de) == cookie) {
727 break;
728 }
729 }
730
731 return de;
732 }
733
734 /* --------------------------------------------------------------------- */
735
736 /*
737 * Helper function for tmpfs_readdir. Returns as much directory entries
738 * as can fit in the uio space. The read starts at uio->uio_offset.
739 * The function returns 0 on success, -1 if there was not enough space
740 * in the uio structure to hold the directory entry or an appropriate
741 * error code if another error happens.
742 */
743 int
744 tmpfs_dir_getdents(struct tmpfs_node *node, struct uio *uio, off_t *cntp)
745 {
746 int error;
747 off_t startcookie;
748 struct tmpfs_dirent *de;
749
750 TMPFS_VALIDATE_DIR(node);
751
752 /* Locate the first directory entry we have to return. We have cached
753 * the last readdir in the node, so use those values if appropriate.
754 * Otherwise do a linear scan to find the requested entry. */
755 startcookie = uio->uio_offset;
756 MPASS(startcookie != TMPFS_DIRCOOKIE_DOT);
757 MPASS(startcookie != TMPFS_DIRCOOKIE_DOTDOT);
758 if (startcookie == TMPFS_DIRCOOKIE_EOF) {
759 return 0;
760 } else {
761 de = tmpfs_dir_lookupbycookie(node, startcookie);
762 }
763 if (de == NULL) {
764 return EINVAL;
765 }
766
767 /* Read as much entries as possible; i.e., until we reach the end of
768 * the directory or we exhaust uio space. */
769 do {
770 struct dirent d;
771
772 /* Create a dirent structure representing the current
773 * tmpfs_node and fill it. */
774 if (de->td_node == NULL) {
775 d.d_fileno = 1;
776 d.d_type = DT_WHT;
777 } else {
778 d.d_fileno = de->td_node->tn_id;
779 switch (de->td_node->tn_type) {
780 case VBLK:
781 d.d_type = DT_BLK;
782 break;
783
784 case VCHR:
785 d.d_type = DT_CHR;
786 break;
787
788 case VDIR:
789 d.d_type = DT_DIR;
790 break;
791
792 case VFIFO:
793 d.d_type = DT_FIFO;
794 break;
795
796 case VLNK:
797 d.d_type = DT_LNK;
798 break;
799
800 case VREG:
801 d.d_type = DT_REG;
802 break;
803
804 case VSOCK:
805 d.d_type = DT_SOCK;
806 break;
807
808 default:
809 panic("tmpfs_dir_getdents: type %p %d",
810 de->td_node, (int)de->td_node->tn_type);
811 }
812 }
813 d.d_namlen = de->td_namelen;
814 MPASS(de->td_namelen < sizeof(d.d_name));
815 (void)memcpy(d.d_name, de->td_name, de->td_namelen);
816 d.d_name[de->td_namelen] = '\0';
817 d.d_reclen = GENERIC_DIRSIZ(&d);
818
819 /* Stop reading if the directory entry we are treating is
820 * bigger than the amount of data that can be returned. */
821 if (d.d_reclen > uio->uio_resid) {
822 error = -1;
823 break;
824 }
825
826 /* Copy the new dirent structure into the output buffer and
827 * advance pointers. */
828 error = uiomove(&d, d.d_reclen, uio);
829 if (error == 0) {
830 (*cntp)++;
831 de = TAILQ_NEXT(de, td_entries);
832 }
833 } while (error == 0 && uio->uio_resid > 0 && de != NULL);
834
835 /* Update the offset and cache. */
836 if (de == NULL) {
837 uio->uio_offset = TMPFS_DIRCOOKIE_EOF;
838 node->tn_dir.tn_readdir_lastn = 0;
839 node->tn_dir.tn_readdir_lastp = NULL;
840 } else {
841 node->tn_dir.tn_readdir_lastn = uio->uio_offset = tmpfs_dircookie(de);
842 node->tn_dir.tn_readdir_lastp = de;
843 }
844
845 node->tn_status |= TMPFS_NODE_ACCESSED;
846 return error;
847 }
848
849 int
850 tmpfs_dir_whiteout_add(struct vnode *dvp, struct componentname *cnp)
851 {
852 struct tmpfs_dirent *de;
853 int error;
854
855 error = tmpfs_alloc_dirent(VFS_TO_TMPFS(dvp->v_mount), NULL,
856 cnp->cn_nameptr, cnp->cn_namelen, &de);
857 if (error != 0)
858 return (error);
859 tmpfs_dir_attach(dvp, de);
860 return (0);
861 }
862
863 void
864 tmpfs_dir_whiteout_remove(struct vnode *dvp, struct componentname *cnp)
865 {
866 struct tmpfs_dirent *de;
867
868 de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(dvp), NULL, cnp);
869 MPASS(de != NULL && de->td_node == NULL);
870 tmpfs_dir_detach(dvp, de);
871 tmpfs_free_dirent(VFS_TO_TMPFS(dvp->v_mount), de, TRUE);
872 }
873
874 /* --------------------------------------------------------------------- */
875
876 /*
877 * Resizes the aobj associated with the regular file pointed to by 'vp' to the
878 * size 'newsize'. 'vp' must point to a vnode that represents a regular file.
879 * 'newsize' must be positive.
880 *
881 * Returns zero on success or an appropriate error code on failure.
882 */
883 int
884 tmpfs_reg_resize(struct vnode *vp, off_t newsize)
885 {
886 struct tmpfs_mount *tmp;
887 struct tmpfs_node *node;
888 vm_object_t uobj;
889 vm_page_t m;
890 vm_pindex_t newpages, oldpages;
891 off_t oldsize;
892 size_t zerolen;
893
894 MPASS(vp->v_type == VREG);
895 MPASS(newsize >= 0);
896
897 node = VP_TO_TMPFS_NODE(vp);
898 uobj = node->tn_reg.tn_aobj;
899 tmp = VFS_TO_TMPFS(vp->v_mount);
900
901 /*
902 * Convert the old and new sizes to the number of pages needed to
903 * store them. It may happen that we do not need to do anything
904 * because the last allocated page can accommodate the change on
905 * its own.
906 */
907 oldsize = node->tn_size;
908 oldpages = OFF_TO_IDX(oldsize + PAGE_MASK);
909 MPASS(oldpages == uobj->size);
910 newpages = OFF_TO_IDX(newsize + PAGE_MASK);
911 if (newpages > oldpages &&
912 newpages - oldpages > TMPFS_PAGES_AVAIL(tmp))
913 return (ENOSPC);
914
915 TMPFS_LOCK(tmp);
916 tmp->tm_pages_used += (newpages - oldpages);
917 TMPFS_UNLOCK(tmp);
918
919 node->tn_size = newsize;
920 vnode_pager_setsize(vp, newsize);
921 VM_OBJECT_LOCK(uobj);
922 if (newsize < oldsize) {
923 /*
924 * Release any swap space and free any whole pages.
925 */
926 if (newpages < oldpages) {
927 swap_pager_freespace(uobj, newpages, oldpages -
928 newpages);
929 vm_object_page_remove(uobj, newpages, 0, 0);
930 }
931
932 /*
933 * Zero the truncated part of the last page.
934 */
935 zerolen = round_page(newsize) - newsize;
936 if (zerolen > 0) {
937 m = vm_page_grab(uobj, OFF_TO_IDX(newsize),
938 VM_ALLOC_NOBUSY | VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
939 pmap_zero_page_area(m, PAGE_SIZE - zerolen, zerolen);
940 }
941 }
942 uobj->size = newpages;
943 VM_OBJECT_UNLOCK(uobj);
944 return (0);
945 }
946
947 /* --------------------------------------------------------------------- */
948
949 /*
950 * Change flags of the given vnode.
951 * Caller should execute tmpfs_update on vp after a successful execution.
952 * The vnode must be locked on entry and remain locked on exit.
953 */
954 int
955 tmpfs_chflags(struct vnode *vp, int flags, struct ucred *cred, struct thread *p)
956 {
957 int error;
958 struct tmpfs_node *node;
959
960 MPASS(VOP_ISLOCKED(vp));
961
962 node = VP_TO_TMPFS_NODE(vp);
963
964 /* Disallow this operation if the file system is mounted read-only. */
965 if (vp->v_mount->mnt_flag & MNT_RDONLY)
966 return EROFS;
967
968 /*
969 * Callers may only modify the file flags on objects they
970 * have VADMIN rights for.
971 */
972 if ((error = VOP_ACCESS(vp, VADMIN, cred, p)))
973 return (error);
974 /*
975 * Unprivileged processes are not permitted to unset system
976 * flags, or modify flags if any system flags are set.
977 */
978 if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0)) {
979 if (node->tn_flags
980 & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND)) {
981 error = securelevel_gt(cred, 0);
982 if (error)
983 return (error);
984 }
985 /* Snapshot flag cannot be set or cleared */
986 if (((flags & SF_SNAPSHOT) != 0 &&
987 (node->tn_flags & SF_SNAPSHOT) == 0) ||
988 ((flags & SF_SNAPSHOT) == 0 &&
989 (node->tn_flags & SF_SNAPSHOT) != 0))
990 return (EPERM);
991 node->tn_flags = flags;
992 } else {
993 if (node->tn_flags
994 & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND) ||
995 (flags & UF_SETTABLE) != flags)
996 return (EPERM);
997 node->tn_flags &= SF_SETTABLE;
998 node->tn_flags |= (flags & UF_SETTABLE);
999 }
1000 node->tn_status |= TMPFS_NODE_CHANGED;
1001
1002 MPASS(VOP_ISLOCKED(vp));
1003
1004 return 0;
1005 }
1006
1007 /* --------------------------------------------------------------------- */
1008
1009 /*
1010 * Change access mode on the given vnode.
1011 * Caller should execute tmpfs_update on vp after a successful execution.
1012 * The vnode must be locked on entry and remain locked on exit.
1013 */
1014 int
1015 tmpfs_chmod(struct vnode *vp, mode_t mode, struct ucred *cred, struct thread *p)
1016 {
1017 int error;
1018 struct tmpfs_node *node;
1019
1020 MPASS(VOP_ISLOCKED(vp));
1021
1022 node = VP_TO_TMPFS_NODE(vp);
1023
1024 /* Disallow this operation if the file system is mounted read-only. */
1025 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1026 return EROFS;
1027
1028 /* Immutable or append-only files cannot be modified, either. */
1029 if (node->tn_flags & (IMMUTABLE | APPEND))
1030 return EPERM;
1031
1032 /*
1033 * To modify the permissions on a file, must possess VADMIN
1034 * for that file.
1035 */
1036 if ((error = VOP_ACCESS(vp, VADMIN, cred, p)))
1037 return (error);
1038
1039 /*
1040 * Privileged processes may set the sticky bit on non-directories,
1041 * as well as set the setgid bit on a file with a group that the
1042 * process is not a member of.
1043 */
1044 if (vp->v_type != VDIR && (mode & S_ISTXT)) {
1045 if (priv_check_cred(cred, PRIV_VFS_STICKYFILE, 0))
1046 return (EFTYPE);
1047 }
1048 if (!groupmember(node->tn_gid, cred) && (mode & S_ISGID)) {
1049 error = priv_check_cred(cred, PRIV_VFS_SETGID, 0);
1050 if (error)
1051 return (error);
1052 }
1053
1054
1055 node->tn_mode &= ~ALLPERMS;
1056 node->tn_mode |= mode & ALLPERMS;
1057
1058 node->tn_status |= TMPFS_NODE_CHANGED;
1059
1060 MPASS(VOP_ISLOCKED(vp));
1061
1062 return 0;
1063 }
1064
1065 /* --------------------------------------------------------------------- */
1066
1067 /*
1068 * Change ownership of the given vnode. At least one of uid or gid must
1069 * be different than VNOVAL. If one is set to that value, the attribute
1070 * is unchanged.
1071 * Caller should execute tmpfs_update on vp after a successful execution.
1072 * The vnode must be locked on entry and remain locked on exit.
1073 */
1074 int
1075 tmpfs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred,
1076 struct thread *p)
1077 {
1078 int error;
1079 struct tmpfs_node *node;
1080 uid_t ouid;
1081 gid_t ogid;
1082
1083 MPASS(VOP_ISLOCKED(vp));
1084
1085 node = VP_TO_TMPFS_NODE(vp);
1086
1087 /* Assign default values if they are unknown. */
1088 MPASS(uid != VNOVAL || gid != VNOVAL);
1089 if (uid == VNOVAL)
1090 uid = node->tn_uid;
1091 if (gid == VNOVAL)
1092 gid = node->tn_gid;
1093 MPASS(uid != VNOVAL && gid != VNOVAL);
1094
1095 /* Disallow this operation if the file system is mounted read-only. */
1096 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1097 return EROFS;
1098
1099 /* Immutable or append-only files cannot be modified, either. */
1100 if (node->tn_flags & (IMMUTABLE | APPEND))
1101 return EPERM;
1102
1103 /*
1104 * To modify the ownership of a file, must possess VADMIN for that
1105 * file.
1106 */
1107 if ((error = VOP_ACCESS(vp, VADMIN, cred, p)))
1108 return (error);
1109
1110 /*
1111 * To change the owner of a file, or change the group of a file to a
1112 * group of which we are not a member, the caller must have
1113 * privilege.
1114 */
1115 if ((uid != node->tn_uid ||
1116 (gid != node->tn_gid && !groupmember(gid, cred))) &&
1117 (error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0)))
1118 return (error);
1119
1120 ogid = node->tn_gid;
1121 ouid = node->tn_uid;
1122
1123 node->tn_uid = uid;
1124 node->tn_gid = gid;
1125
1126 node->tn_status |= TMPFS_NODE_CHANGED;
1127
1128 if ((node->tn_mode & (S_ISUID | S_ISGID)) && (ouid != uid || ogid != gid)) {
1129 if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID, 0))
1130 node->tn_mode &= ~(S_ISUID | S_ISGID);
1131 }
1132
1133 MPASS(VOP_ISLOCKED(vp));
1134
1135 return 0;
1136 }
1137
1138 /* --------------------------------------------------------------------- */
1139
1140 /*
1141 * Change size of the given vnode.
1142 * Caller should execute tmpfs_update on vp after a successful execution.
1143 * The vnode must be locked on entry and remain locked on exit.
1144 */
1145 int
1146 tmpfs_chsize(struct vnode *vp, u_quad_t size, struct ucred *cred,
1147 struct thread *p)
1148 {
1149 int error;
1150 struct tmpfs_node *node;
1151
1152 MPASS(VOP_ISLOCKED(vp));
1153
1154 node = VP_TO_TMPFS_NODE(vp);
1155
1156 /* Decide whether this is a valid operation based on the file type. */
1157 error = 0;
1158 switch (vp->v_type) {
1159 case VDIR:
1160 return EISDIR;
1161
1162 case VREG:
1163 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1164 return EROFS;
1165 break;
1166
1167 case VBLK:
1168 /* FALLTHROUGH */
1169 case VCHR:
1170 /* FALLTHROUGH */
1171 case VFIFO:
1172 /* Allow modifications of special files even if in the file
1173 * system is mounted read-only (we are not modifying the
1174 * files themselves, but the objects they represent). */
1175 return 0;
1176
1177 default:
1178 /* Anything else is unsupported. */
1179 return EOPNOTSUPP;
1180 }
1181
1182 /* Immutable or append-only files cannot be modified, either. */
1183 if (node->tn_flags & (IMMUTABLE | APPEND))
1184 return EPERM;
1185
1186 error = tmpfs_truncate(vp, size);
1187 /* tmpfs_truncate will raise the NOTE_EXTEND and NOTE_ATTRIB kevents
1188 * for us, as will update tn_status; no need to do that here. */
1189
1190 MPASS(VOP_ISLOCKED(vp));
1191
1192 return error;
1193 }
1194
1195 /* --------------------------------------------------------------------- */
1196
1197 /*
1198 * Change access and modification times of the given vnode.
1199 * Caller should execute tmpfs_update on vp after a successful execution.
1200 * The vnode must be locked on entry and remain locked on exit.
1201 */
1202 int
1203 tmpfs_chtimes(struct vnode *vp, struct timespec *atime, struct timespec *mtime,
1204 struct timespec *birthtime, int vaflags, struct ucred *cred, struct thread *l)
1205 {
1206 int error;
1207 struct tmpfs_node *node;
1208
1209 MPASS(VOP_ISLOCKED(vp));
1210
1211 node = VP_TO_TMPFS_NODE(vp);
1212
1213 /* Disallow this operation if the file system is mounted read-only. */
1214 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1215 return EROFS;
1216
1217 /* Immutable or append-only files cannot be modified, either. */
1218 if (node->tn_flags & (IMMUTABLE | APPEND))
1219 return EPERM;
1220
1221 /* Determine if the user have proper privilege to update time. */
1222 if (vaflags & VA_UTIMES_NULL) {
1223 error = VOP_ACCESS(vp, VADMIN, cred, l);
1224 if (error)
1225 error = VOP_ACCESS(vp, VWRITE, cred, l);
1226 } else
1227 error = VOP_ACCESS(vp, VADMIN, cred, l);
1228 if (error)
1229 return (error);
1230
1231 if (atime->tv_sec != VNOVAL && atime->tv_nsec != VNOVAL)
1232 node->tn_status |= TMPFS_NODE_ACCESSED;
1233
1234 if (mtime->tv_sec != VNOVAL && mtime->tv_nsec != VNOVAL)
1235 node->tn_status |= TMPFS_NODE_MODIFIED;
1236
1237 if (birthtime->tv_nsec != VNOVAL && birthtime->tv_nsec != VNOVAL)
1238 node->tn_status |= TMPFS_NODE_MODIFIED;
1239
1240 tmpfs_itimes(vp, atime, mtime);
1241
1242 if (birthtime->tv_nsec != VNOVAL && birthtime->tv_nsec != VNOVAL)
1243 node->tn_birthtime = *birthtime;
1244 MPASS(VOP_ISLOCKED(vp));
1245
1246 return 0;
1247 }
1248
1249 /* --------------------------------------------------------------------- */
1250 /* Sync timestamps */
1251 void
1252 tmpfs_itimes(struct vnode *vp, const struct timespec *acc,
1253 const struct timespec *mod)
1254 {
1255 struct tmpfs_node *node;
1256 struct timespec now;
1257
1258 node = VP_TO_TMPFS_NODE(vp);
1259
1260 if ((node->tn_status & (TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
1261 TMPFS_NODE_CHANGED)) == 0)
1262 return;
1263
1264 vfs_timestamp(&now);
1265 if (node->tn_status & TMPFS_NODE_ACCESSED) {
1266 if (acc == NULL)
1267 acc = &now;
1268 node->tn_atime = *acc;
1269 }
1270 if (node->tn_status & TMPFS_NODE_MODIFIED) {
1271 if (mod == NULL)
1272 mod = &now;
1273 node->tn_mtime = *mod;
1274 }
1275 if (node->tn_status & TMPFS_NODE_CHANGED) {
1276 node->tn_ctime = now;
1277 }
1278 node->tn_status &=
1279 ~(TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED | TMPFS_NODE_CHANGED);
1280 }
1281
1282 /* --------------------------------------------------------------------- */
1283
1284 void
1285 tmpfs_update(struct vnode *vp)
1286 {
1287
1288 tmpfs_itimes(vp, NULL, NULL);
1289 }
1290
1291 /* --------------------------------------------------------------------- */
1292
1293 int
1294 tmpfs_truncate(struct vnode *vp, off_t length)
1295 {
1296 int error;
1297 struct tmpfs_node *node;
1298
1299 node = VP_TO_TMPFS_NODE(vp);
1300
1301 if (length < 0) {
1302 error = EINVAL;
1303 goto out;
1304 }
1305
1306 if (node->tn_size == length) {
1307 error = 0;
1308 goto out;
1309 }
1310
1311 if (length > VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize)
1312 return (EFBIG);
1313
1314 error = tmpfs_reg_resize(vp, length);
1315 if (error == 0) {
1316 node->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1317 }
1318
1319 out:
1320 tmpfs_update(vp);
1321
1322 return error;
1323 }
Cache object: 4c697778a7bafac04c2ea4bff29229dd
|