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/systm.h>
41 #include <sys/dirent.h>
42 #include <sys/fnv_hash.h>
43 #include <sys/lock.h>
44 #include <sys/namei.h>
45 #include <sys/priv.h>
46 #include <sys/proc.h>
47 #include <sys/random.h>
48 #include <sys/rwlock.h>
49 #include <sys/stat.h>
50 #include <sys/sysctl.h>
51 #include <sys/vnode.h>
52 #include <sys/vmmeter.h>
53
54 #include <vm/vm.h>
55 #include <vm/vm_param.h>
56 #include <vm/vm_object.h>
57 #include <vm/vm_page.h>
58 #include <vm/vm_pageout.h>
59 #include <vm/vm_pager.h>
60 #include <vm/vm_extern.h>
61
62 #include <fs/tmpfs/tmpfs.h>
63 #include <fs/tmpfs/tmpfs_fifoops.h>
64 #include <fs/tmpfs/tmpfs_vnops.h>
65
66 SYSCTL_NODE(_vfs, OID_AUTO, tmpfs, CTLFLAG_RW, 0, "tmpfs file system");
67
68 static long tmpfs_pages_reserved = TMPFS_PAGES_MINRESERVED;
69
70 static int
71 sysctl_mem_reserved(SYSCTL_HANDLER_ARGS)
72 {
73 int error;
74 long pages, bytes;
75
76 pages = *(long *)arg1;
77 bytes = pages * PAGE_SIZE;
78
79 error = sysctl_handle_long(oidp, &bytes, 0, req);
80 if (error || !req->newptr)
81 return (error);
82
83 pages = bytes / PAGE_SIZE;
84 if (pages < TMPFS_PAGES_MINRESERVED)
85 return (EINVAL);
86
87 *(long *)arg1 = pages;
88 return (0);
89 }
90
91 SYSCTL_PROC(_vfs_tmpfs, OID_AUTO, memory_reserved, CTLTYPE_LONG|CTLFLAG_RW,
92 &tmpfs_pages_reserved, 0, sysctl_mem_reserved, "L",
93 "Amount of available memory and swap below which tmpfs growth stops");
94
95 static __inline int tmpfs_dirtree_cmp(struct tmpfs_dirent *a,
96 struct tmpfs_dirent *b);
97 RB_PROTOTYPE_STATIC(tmpfs_dir, tmpfs_dirent, uh.td_entries, tmpfs_dirtree_cmp);
98
99 size_t
100 tmpfs_mem_avail(void)
101 {
102 vm_ooffset_t avail;
103
104 avail = swap_pager_avail + vm_cnt.v_free_count - tmpfs_pages_reserved;
105 if (__predict_false(avail < 0))
106 avail = 0;
107 return (avail);
108 }
109
110 size_t
111 tmpfs_pages_used(struct tmpfs_mount *tmp)
112 {
113 const size_t node_size = sizeof(struct tmpfs_node) +
114 sizeof(struct tmpfs_dirent);
115 size_t meta_pages;
116
117 meta_pages = howmany((uintmax_t)tmp->tm_nodes_inuse * node_size,
118 PAGE_SIZE);
119 return (meta_pages + tmp->tm_pages_used);
120 }
121
122 static size_t
123 tmpfs_pages_check_avail(struct tmpfs_mount *tmp, size_t req_pages)
124 {
125 if (tmpfs_mem_avail() < req_pages)
126 return (0);
127
128 if (tmp->tm_pages_max != ULONG_MAX &&
129 tmp->tm_pages_max < req_pages + tmpfs_pages_used(tmp))
130 return (0);
131
132 return (1);
133 }
134
135 void
136 tmpfs_ref_node(struct tmpfs_node *node)
137 {
138
139 TMPFS_NODE_LOCK(node);
140 tmpfs_ref_node_locked(node);
141 TMPFS_NODE_UNLOCK(node);
142 }
143
144 void
145 tmpfs_ref_node_locked(struct tmpfs_node *node)
146 {
147
148 TMPFS_NODE_ASSERT_LOCKED(node);
149 KASSERT(node->tn_refcount > 0, ("node %p zero refcount", node));
150 KASSERT(node->tn_refcount < UINT_MAX, ("node %p refcount %u", node,
151 node->tn_refcount));
152 node->tn_refcount++;
153 }
154
155 /*
156 * Allocates a new node of type 'type' inside the 'tmp' mount point, with
157 * its owner set to 'uid', its group to 'gid' and its mode set to 'mode',
158 * using the credentials of the process 'p'.
159 *
160 * If the node type is set to 'VDIR', then the parent parameter must point
161 * to the parent directory of the node being created. It may only be NULL
162 * while allocating the root node.
163 *
164 * If the node type is set to 'VBLK' or 'VCHR', then the rdev parameter
165 * specifies the device the node represents.
166 *
167 * If the node type is set to 'VLNK', then the parameter target specifies
168 * the file name of the target file for the symbolic link that is being
169 * created.
170 *
171 * Note that new nodes are retrieved from the available list if it has
172 * items or, if it is empty, from the node pool as long as there is enough
173 * space to create them.
174 *
175 * Returns zero on success or an appropriate error code on failure.
176 */
177 int
178 tmpfs_alloc_node(struct mount *mp, struct tmpfs_mount *tmp, enum vtype type,
179 uid_t uid, gid_t gid, mode_t mode, struct tmpfs_node *parent,
180 char *target, dev_t rdev, struct tmpfs_node **node)
181 {
182 struct tmpfs_node *nnode;
183 vm_object_t obj;
184
185 /* If the root directory of the 'tmp' file system is not yet
186 * allocated, this must be the request to do it. */
187 MPASS(IMPLIES(tmp->tm_root == NULL, parent == NULL && type == VDIR));
188 KASSERT(tmp->tm_root == NULL || mp->mnt_writeopcount > 0,
189 ("creating node not under vn_start_write"));
190
191 MPASS(IFF(type == VLNK, target != NULL));
192 MPASS(IFF(type == VBLK || type == VCHR, rdev != VNOVAL));
193
194 if (tmp->tm_nodes_inuse >= tmp->tm_nodes_max)
195 return (ENOSPC);
196 if (tmpfs_pages_check_avail(tmp, 1) == 0)
197 return (ENOSPC);
198
199 if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) {
200 /*
201 * When a new tmpfs node is created for fully
202 * constructed mount point, there must be a parent
203 * node, which vnode is locked exclusively. As
204 * consequence, if the unmount is executing in
205 * parallel, vflush() cannot reclaim the parent vnode.
206 * Due to this, the check for MNTK_UNMOUNT flag is not
207 * racy: if we did not see MNTK_UNMOUNT flag, then tmp
208 * cannot be destroyed until node construction is
209 * finished and the parent vnode unlocked.
210 *
211 * Tmpfs does not need to instantiate new nodes during
212 * unmount.
213 */
214 return (EBUSY);
215 }
216 if ((mp->mnt_kern_flag & MNT_RDONLY) != 0)
217 return (EROFS);
218
219 nnode = (struct tmpfs_node *)uma_zalloc_arg(tmp->tm_node_pool, tmp,
220 M_WAITOK);
221
222 /* Generic initialization. */
223 nnode->tn_type = type;
224 vfs_timestamp(&nnode->tn_atime);
225 nnode->tn_birthtime = nnode->tn_ctime = nnode->tn_mtime =
226 nnode->tn_atime;
227 nnode->tn_uid = uid;
228 nnode->tn_gid = gid;
229 nnode->tn_mode = mode;
230 nnode->tn_id = alloc_unr(tmp->tm_ino_unr);
231 nnode->tn_refcount = 1;
232
233 /* Type-specific initialization. */
234 switch (nnode->tn_type) {
235 case VBLK:
236 case VCHR:
237 nnode->tn_rdev = rdev;
238 break;
239
240 case VDIR:
241 RB_INIT(&nnode->tn_dir.tn_dirhead);
242 LIST_INIT(&nnode->tn_dir.tn_dupindex);
243 MPASS(parent != nnode);
244 MPASS(IMPLIES(parent == NULL, tmp->tm_root == NULL));
245 nnode->tn_dir.tn_parent = (parent == NULL) ? nnode : parent;
246 nnode->tn_dir.tn_readdir_lastn = 0;
247 nnode->tn_dir.tn_readdir_lastp = NULL;
248 nnode->tn_links++;
249 TMPFS_NODE_LOCK(nnode->tn_dir.tn_parent);
250 nnode->tn_dir.tn_parent->tn_links++;
251 TMPFS_NODE_UNLOCK(nnode->tn_dir.tn_parent);
252 break;
253
254 case VFIFO:
255 /* FALLTHROUGH */
256 case VSOCK:
257 break;
258
259 case VLNK:
260 MPASS(strlen(target) < MAXPATHLEN);
261 nnode->tn_size = strlen(target);
262 nnode->tn_link = malloc(nnode->tn_size, M_TMPFSNAME,
263 M_WAITOK);
264 memcpy(nnode->tn_link, target, nnode->tn_size);
265 break;
266
267 case VREG:
268 obj = nnode->tn_reg.tn_aobj =
269 vm_pager_allocate(OBJT_SWAP, NULL, 0, VM_PROT_DEFAULT, 0,
270 NULL /* XXXKIB - tmpfs needs swap reservation */);
271 VM_OBJECT_WLOCK(obj);
272 /* OBJ_TMPFS is set together with the setting of vp->v_object */
273 vm_object_set_flag(obj, OBJ_NOSPLIT | OBJ_TMPFS_NODE);
274 vm_object_clear_flag(obj, OBJ_ONEMAPPING);
275 VM_OBJECT_WUNLOCK(obj);
276 break;
277
278 default:
279 panic("tmpfs_alloc_node: type %p %d", nnode,
280 (int)nnode->tn_type);
281 }
282
283 TMPFS_LOCK(tmp);
284 LIST_INSERT_HEAD(&tmp->tm_nodes_used, nnode, tn_entries);
285 nnode->tn_attached = true;
286 tmp->tm_nodes_inuse++;
287 tmp->tm_refcount++;
288 TMPFS_UNLOCK(tmp);
289
290 *node = nnode;
291 return (0);
292 }
293
294 /*
295 * Destroys the node pointed to by node from the file system 'tmp'.
296 * If the node references a directory, no entries are allowed.
297 */
298 void
299 tmpfs_free_node(struct tmpfs_mount *tmp, struct tmpfs_node *node)
300 {
301
302 TMPFS_LOCK(tmp);
303 TMPFS_NODE_LOCK(node);
304 if (!tmpfs_free_node_locked(tmp, node, false)) {
305 TMPFS_NODE_UNLOCK(node);
306 TMPFS_UNLOCK(tmp);
307 }
308 }
309
310 bool
311 tmpfs_free_node_locked(struct tmpfs_mount *tmp, struct tmpfs_node *node,
312 bool detach)
313 {
314 vm_object_t uobj;
315
316 TMPFS_MP_ASSERT_LOCKED(tmp);
317 TMPFS_NODE_ASSERT_LOCKED(node);
318 KASSERT(node->tn_refcount > 0, ("node %p refcount zero", node));
319
320 node->tn_refcount--;
321 if (node->tn_attached && (detach || node->tn_refcount == 0)) {
322 MPASS(tmp->tm_nodes_inuse > 0);
323 tmp->tm_nodes_inuse--;
324 LIST_REMOVE(node, tn_entries);
325 node->tn_attached = false;
326 }
327 if (node->tn_refcount > 0)
328 return (false);
329
330 #ifdef INVARIANTS
331 MPASS(node->tn_vnode == NULL);
332 MPASS((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0);
333 #endif
334 TMPFS_NODE_UNLOCK(node);
335 TMPFS_UNLOCK(tmp);
336
337 switch (node->tn_type) {
338 case VBLK:
339 /* FALLTHROUGH */
340 case VCHR:
341 /* FALLTHROUGH */
342 case VDIR:
343 /* FALLTHROUGH */
344 case VFIFO:
345 /* FALLTHROUGH */
346 case VSOCK:
347 break;
348
349 case VLNK:
350 free(node->tn_link, M_TMPFSNAME);
351 break;
352
353 case VREG:
354 uobj = node->tn_reg.tn_aobj;
355 if (uobj != NULL) {
356 if (uobj->size != 0)
357 atomic_subtract_long(&tmp->tm_pages_used, uobj->size);
358 KASSERT((uobj->flags & OBJ_TMPFS) == 0,
359 ("leaked OBJ_TMPFS node %p vm_obj %p", node, uobj));
360 vm_object_deallocate(uobj);
361 }
362 break;
363
364 default:
365 panic("tmpfs_free_node: type %p %d", node, (int)node->tn_type);
366 }
367
368 free_unr(tmp->tm_ino_unr, node->tn_id);
369 uma_zfree(tmp->tm_node_pool, node);
370 TMPFS_LOCK(tmp);
371 tmpfs_free_tmp(tmp);
372 return (true);
373 }
374
375 static __inline uint32_t
376 tmpfs_dirent_hash(const char *name, u_int len)
377 {
378 uint32_t hash;
379
380 hash = fnv_32_buf(name, len, FNV1_32_INIT + len) & TMPFS_DIRCOOKIE_MASK;
381 #ifdef TMPFS_DEBUG_DIRCOOKIE_DUP
382 hash &= 0xf;
383 #endif
384 if (hash < TMPFS_DIRCOOKIE_MIN)
385 hash += TMPFS_DIRCOOKIE_MIN;
386
387 return (hash);
388 }
389
390 static __inline off_t
391 tmpfs_dirent_cookie(struct tmpfs_dirent *de)
392 {
393 if (de == NULL)
394 return (TMPFS_DIRCOOKIE_EOF);
395
396 MPASS(de->td_cookie >= TMPFS_DIRCOOKIE_MIN);
397
398 return (de->td_cookie);
399 }
400
401 static __inline boolean_t
402 tmpfs_dirent_dup(struct tmpfs_dirent *de)
403 {
404 return ((de->td_cookie & TMPFS_DIRCOOKIE_DUP) != 0);
405 }
406
407 static __inline boolean_t
408 tmpfs_dirent_duphead(struct tmpfs_dirent *de)
409 {
410 return ((de->td_cookie & TMPFS_DIRCOOKIE_DUPHEAD) != 0);
411 }
412
413 void
414 tmpfs_dirent_init(struct tmpfs_dirent *de, const char *name, u_int namelen)
415 {
416 de->td_hash = de->td_cookie = tmpfs_dirent_hash(name, namelen);
417 memcpy(de->ud.td_name, name, namelen);
418 de->td_namelen = namelen;
419 }
420
421 /*
422 * Allocates a new directory entry for the node node with a name of name.
423 * The new directory entry is returned in *de.
424 *
425 * The link count of node is increased by one to reflect the new object
426 * referencing it.
427 *
428 * Returns zero on success or an appropriate error code on failure.
429 */
430 int
431 tmpfs_alloc_dirent(struct tmpfs_mount *tmp, struct tmpfs_node *node,
432 const char *name, u_int len, struct tmpfs_dirent **de)
433 {
434 struct tmpfs_dirent *nde;
435
436 nde = uma_zalloc(tmp->tm_dirent_pool, M_WAITOK);
437 nde->td_node = node;
438 if (name != NULL) {
439 nde->ud.td_name = malloc(len, M_TMPFSNAME, M_WAITOK);
440 tmpfs_dirent_init(nde, name, len);
441 } else
442 nde->td_namelen = 0;
443 if (node != NULL)
444 node->tn_links++;
445
446 *de = nde;
447
448 return 0;
449 }
450
451 /*
452 * Frees a directory entry. It is the caller's responsibility to destroy
453 * the node referenced by it if needed.
454 *
455 * The link count of node is decreased by one to reflect the removal of an
456 * object that referenced it. This only happens if 'node_exists' is true;
457 * otherwise the function will not access the node referred to by the
458 * directory entry, as it may already have been released from the outside.
459 */
460 void
461 tmpfs_free_dirent(struct tmpfs_mount *tmp, struct tmpfs_dirent *de)
462 {
463 struct tmpfs_node *node;
464
465 node = de->td_node;
466 if (node != NULL) {
467 MPASS(node->tn_links > 0);
468 node->tn_links--;
469 }
470 if (!tmpfs_dirent_duphead(de) && de->ud.td_name != NULL)
471 free(de->ud.td_name, M_TMPFSNAME);
472 uma_zfree(tmp->tm_dirent_pool, de);
473 }
474
475 void
476 tmpfs_destroy_vobject(struct vnode *vp, vm_object_t obj)
477 {
478
479 ASSERT_VOP_ELOCKED(vp, "tmpfs_destroy_vobject");
480 if (vp->v_type != VREG || obj == NULL)
481 return;
482
483 VM_OBJECT_WLOCK(obj);
484 VI_LOCK(vp);
485 vm_object_clear_flag(obj, OBJ_TMPFS);
486 obj->un_pager.swp.swp_tmpfs = NULL;
487 VI_UNLOCK(vp);
488 VM_OBJECT_WUNLOCK(obj);
489 }
490
491 /*
492 * Need to clear v_object for insmntque failure.
493 */
494 static void
495 tmpfs_insmntque_dtr(struct vnode *vp, void *dtr_arg)
496 {
497
498 tmpfs_destroy_vobject(vp, vp->v_object);
499 vp->v_object = NULL;
500 vp->v_data = NULL;
501 vp->v_op = &dead_vnodeops;
502 vgone(vp);
503 vput(vp);
504 }
505
506 /*
507 * Allocates a new vnode for the node node or returns a new reference to
508 * an existing one if the node had already a vnode referencing it. The
509 * resulting locked vnode is returned in *vpp.
510 *
511 * Returns zero on success or an appropriate error code on failure.
512 */
513 int
514 tmpfs_alloc_vp(struct mount *mp, struct tmpfs_node *node, int lkflag,
515 struct vnode **vpp)
516 {
517 struct vnode *vp;
518 struct tmpfs_mount *tm;
519 vm_object_t object;
520 int error;
521
522 error = 0;
523 tm = VFS_TO_TMPFS(mp);
524 TMPFS_NODE_LOCK(node);
525 tmpfs_ref_node_locked(node);
526 loop:
527 TMPFS_NODE_ASSERT_LOCKED(node);
528 if ((vp = node->tn_vnode) != NULL) {
529 MPASS((node->tn_vpstate & TMPFS_VNODE_DOOMED) == 0);
530 VI_LOCK(vp);
531 if ((node->tn_type == VDIR && node->tn_dir.tn_parent == NULL) ||
532 ((vp->v_iflag & VI_DOOMED) != 0 &&
533 (lkflag & LK_NOWAIT) != 0)) {
534 VI_UNLOCK(vp);
535 TMPFS_NODE_UNLOCK(node);
536 error = ENOENT;
537 vp = NULL;
538 goto out;
539 }
540 if ((vp->v_iflag & VI_DOOMED) != 0) {
541 VI_UNLOCK(vp);
542 node->tn_vpstate |= TMPFS_VNODE_WRECLAIM;
543 while ((node->tn_vpstate & TMPFS_VNODE_WRECLAIM) != 0) {
544 msleep(&node->tn_vnode, TMPFS_NODE_MTX(node),
545 0, "tmpfsE", 0);
546 }
547 goto loop;
548 }
549 TMPFS_NODE_UNLOCK(node);
550 error = vget(vp, lkflag | LK_INTERLOCK, curthread);
551 if (error == ENOENT) {
552 TMPFS_NODE_LOCK(node);
553 goto loop;
554 }
555 if (error != 0) {
556 vp = NULL;
557 goto out;
558 }
559
560 /*
561 * Make sure the vnode is still there after
562 * getting the interlock to avoid racing a free.
563 */
564 if (node->tn_vnode == NULL || node->tn_vnode != vp) {
565 vput(vp);
566 TMPFS_NODE_LOCK(node);
567 goto loop;
568 }
569
570 goto out;
571 }
572
573 if ((node->tn_vpstate & TMPFS_VNODE_DOOMED) ||
574 (node->tn_type == VDIR && node->tn_dir.tn_parent == NULL)) {
575 TMPFS_NODE_UNLOCK(node);
576 error = ENOENT;
577 vp = NULL;
578 goto out;
579 }
580
581 /*
582 * otherwise lock the vp list while we call getnewvnode
583 * since that can block.
584 */
585 if (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) {
586 node->tn_vpstate |= TMPFS_VNODE_WANT;
587 error = msleep((caddr_t) &node->tn_vpstate,
588 TMPFS_NODE_MTX(node), 0, "tmpfs_alloc_vp", 0);
589 if (error != 0)
590 goto out;
591 goto loop;
592 } else
593 node->tn_vpstate |= TMPFS_VNODE_ALLOCATING;
594
595 TMPFS_NODE_UNLOCK(node);
596
597 /* Get a new vnode and associate it with our node. */
598 error = getnewvnode("tmpfs", mp, VFS_TO_TMPFS(mp)->tm_nonc ?
599 &tmpfs_vnodeop_nonc_entries : &tmpfs_vnodeop_entries, &vp);
600 if (error != 0)
601 goto unlock;
602 MPASS(vp != NULL);
603
604 /* lkflag is ignored, the lock is exclusive */
605 (void) vn_lock(vp, lkflag | LK_RETRY);
606
607 vp->v_data = node;
608 vp->v_type = node->tn_type;
609
610 /* Type-specific initialization. */
611 switch (node->tn_type) {
612 case VBLK:
613 /* FALLTHROUGH */
614 case VCHR:
615 /* FALLTHROUGH */
616 case VLNK:
617 /* FALLTHROUGH */
618 case VSOCK:
619 break;
620 case VFIFO:
621 vp->v_op = &tmpfs_fifoop_entries;
622 break;
623 case VREG:
624 object = node->tn_reg.tn_aobj;
625 VM_OBJECT_WLOCK(object);
626 VI_LOCK(vp);
627 KASSERT(vp->v_object == NULL, ("Not NULL v_object in tmpfs"));
628 vp->v_object = object;
629 object->un_pager.swp.swp_tmpfs = vp;
630 vm_object_set_flag(object, OBJ_TMPFS);
631 VI_UNLOCK(vp);
632 VM_OBJECT_WUNLOCK(object);
633 break;
634 case VDIR:
635 MPASS(node->tn_dir.tn_parent != NULL);
636 if (node->tn_dir.tn_parent == node)
637 vp->v_vflag |= VV_ROOT;
638 break;
639
640 default:
641 panic("tmpfs_alloc_vp: type %p %d", node, (int)node->tn_type);
642 }
643 if (vp->v_type != VFIFO)
644 VN_LOCK_ASHARE(vp);
645
646 error = insmntque1(vp, mp, tmpfs_insmntque_dtr, NULL);
647 if (error != 0)
648 vp = NULL;
649
650 unlock:
651 TMPFS_NODE_LOCK(node);
652
653 MPASS(node->tn_vpstate & TMPFS_VNODE_ALLOCATING);
654 node->tn_vpstate &= ~TMPFS_VNODE_ALLOCATING;
655 node->tn_vnode = vp;
656
657 if (node->tn_vpstate & TMPFS_VNODE_WANT) {
658 node->tn_vpstate &= ~TMPFS_VNODE_WANT;
659 TMPFS_NODE_UNLOCK(node);
660 wakeup((caddr_t) &node->tn_vpstate);
661 } else
662 TMPFS_NODE_UNLOCK(node);
663
664 out:
665 if (error == 0) {
666 *vpp = vp;
667
668 #ifdef INVARIANTS
669 MPASS(*vpp != NULL && VOP_ISLOCKED(*vpp));
670 TMPFS_NODE_LOCK(node);
671 MPASS(*vpp == node->tn_vnode);
672 TMPFS_NODE_UNLOCK(node);
673 #endif
674 }
675 tmpfs_free_node(tm, node);
676
677 return (error);
678 }
679
680 /*
681 * Destroys the association between the vnode vp and the node it
682 * references.
683 */
684 void
685 tmpfs_free_vp(struct vnode *vp)
686 {
687 struct tmpfs_node *node;
688
689 node = VP_TO_TMPFS_NODE(vp);
690
691 TMPFS_NODE_ASSERT_LOCKED(node);
692 node->tn_vnode = NULL;
693 if ((node->tn_vpstate & TMPFS_VNODE_WRECLAIM) != 0)
694 wakeup(&node->tn_vnode);
695 node->tn_vpstate &= ~TMPFS_VNODE_WRECLAIM;
696 vp->v_data = NULL;
697 }
698
699 /*
700 * Allocates a new file of type 'type' and adds it to the parent directory
701 * 'dvp'; this addition is done using the component name given in 'cnp'.
702 * The ownership of the new file is automatically assigned based on the
703 * credentials of the caller (through 'cnp'), the group is set based on
704 * the parent directory and the mode is determined from the 'vap' argument.
705 * If successful, *vpp holds a vnode to the newly created file and zero
706 * is returned. Otherwise *vpp is NULL and the function returns an
707 * appropriate error code.
708 */
709 int
710 tmpfs_alloc_file(struct vnode *dvp, struct vnode **vpp, struct vattr *vap,
711 struct componentname *cnp, char *target)
712 {
713 int error;
714 struct tmpfs_dirent *de;
715 struct tmpfs_mount *tmp;
716 struct tmpfs_node *dnode;
717 struct tmpfs_node *node;
718 struct tmpfs_node *parent;
719
720 ASSERT_VOP_ELOCKED(dvp, "tmpfs_alloc_file");
721 MPASS(cnp->cn_flags & HASBUF);
722
723 tmp = VFS_TO_TMPFS(dvp->v_mount);
724 dnode = VP_TO_TMPFS_DIR(dvp);
725 *vpp = NULL;
726
727 /* If the entry we are creating is a directory, we cannot overflow
728 * the number of links of its parent, because it will get a new
729 * link. */
730 if (vap->va_type == VDIR) {
731 /* Ensure that we do not overflow the maximum number of links
732 * imposed by the system. */
733 MPASS(dnode->tn_links <= LINK_MAX);
734 if (dnode->tn_links == LINK_MAX) {
735 return (EMLINK);
736 }
737
738 parent = dnode;
739 MPASS(parent != NULL);
740 } else
741 parent = NULL;
742
743 /* Allocate a node that represents the new file. */
744 error = tmpfs_alloc_node(dvp->v_mount, tmp, vap->va_type,
745 cnp->cn_cred->cr_uid, dnode->tn_gid, vap->va_mode, parent,
746 target, vap->va_rdev, &node);
747 if (error != 0)
748 return (error);
749
750 /* Allocate a directory entry that points to the new file. */
751 error = tmpfs_alloc_dirent(tmp, node, cnp->cn_nameptr, cnp->cn_namelen,
752 &de);
753 if (error != 0) {
754 tmpfs_free_node(tmp, node);
755 return (error);
756 }
757
758 /* Allocate a vnode for the new file. */
759 error = tmpfs_alloc_vp(dvp->v_mount, node, LK_EXCLUSIVE, vpp);
760 if (error != 0) {
761 tmpfs_free_dirent(tmp, de);
762 tmpfs_free_node(tmp, node);
763 return (error);
764 }
765
766 /* Now that all required items are allocated, we can proceed to
767 * insert the new node into the directory, an operation that
768 * cannot fail. */
769 if (cnp->cn_flags & ISWHITEOUT)
770 tmpfs_dir_whiteout_remove(dvp, cnp);
771 tmpfs_dir_attach(dvp, de);
772 return (0);
773 }
774
775 struct tmpfs_dirent *
776 tmpfs_dir_first(struct tmpfs_node *dnode, struct tmpfs_dir_cursor *dc)
777 {
778 struct tmpfs_dirent *de;
779
780 de = RB_MIN(tmpfs_dir, &dnode->tn_dir.tn_dirhead);
781 dc->tdc_tree = de;
782 if (de != NULL && tmpfs_dirent_duphead(de))
783 de = LIST_FIRST(&de->ud.td_duphead);
784 dc->tdc_current = de;
785
786 return (dc->tdc_current);
787 }
788
789 struct tmpfs_dirent *
790 tmpfs_dir_next(struct tmpfs_node *dnode, struct tmpfs_dir_cursor *dc)
791 {
792 struct tmpfs_dirent *de;
793
794 MPASS(dc->tdc_tree != NULL);
795 if (tmpfs_dirent_dup(dc->tdc_current)) {
796 dc->tdc_current = LIST_NEXT(dc->tdc_current, uh.td_dup.entries);
797 if (dc->tdc_current != NULL)
798 return (dc->tdc_current);
799 }
800 dc->tdc_tree = dc->tdc_current = RB_NEXT(tmpfs_dir,
801 &dnode->tn_dir.tn_dirhead, dc->tdc_tree);
802 if ((de = dc->tdc_current) != NULL && tmpfs_dirent_duphead(de)) {
803 dc->tdc_current = LIST_FIRST(&de->ud.td_duphead);
804 MPASS(dc->tdc_current != NULL);
805 }
806
807 return (dc->tdc_current);
808 }
809
810 /* Lookup directory entry in RB-Tree. Function may return duphead entry. */
811 static struct tmpfs_dirent *
812 tmpfs_dir_xlookup_hash(struct tmpfs_node *dnode, uint32_t hash)
813 {
814 struct tmpfs_dirent *de, dekey;
815
816 dekey.td_hash = hash;
817 de = RB_FIND(tmpfs_dir, &dnode->tn_dir.tn_dirhead, &dekey);
818 return (de);
819 }
820
821 /* Lookup directory entry by cookie, initialize directory cursor accordingly. */
822 static struct tmpfs_dirent *
823 tmpfs_dir_lookup_cookie(struct tmpfs_node *node, off_t cookie,
824 struct tmpfs_dir_cursor *dc)
825 {
826 struct tmpfs_dir *dirhead = &node->tn_dir.tn_dirhead;
827 struct tmpfs_dirent *de, dekey;
828
829 MPASS(cookie >= TMPFS_DIRCOOKIE_MIN);
830
831 if (cookie == node->tn_dir.tn_readdir_lastn &&
832 (de = node->tn_dir.tn_readdir_lastp) != NULL) {
833 /* Protect against possible race, tn_readdir_last[pn]
834 * may be updated with only shared vnode lock held. */
835 if (cookie == tmpfs_dirent_cookie(de))
836 goto out;
837 }
838
839 if ((cookie & TMPFS_DIRCOOKIE_DUP) != 0) {
840 LIST_FOREACH(de, &node->tn_dir.tn_dupindex,
841 uh.td_dup.index_entries) {
842 MPASS(tmpfs_dirent_dup(de));
843 if (de->td_cookie == cookie)
844 goto out;
845 /* dupindex list is sorted. */
846 if (de->td_cookie < cookie) {
847 de = NULL;
848 goto out;
849 }
850 }
851 MPASS(de == NULL);
852 goto out;
853 }
854
855 if ((cookie & TMPFS_DIRCOOKIE_MASK) != cookie) {
856 de = NULL;
857 } else {
858 dekey.td_hash = cookie;
859 /* Recover if direntry for cookie was removed */
860 de = RB_NFIND(tmpfs_dir, dirhead, &dekey);
861 }
862 dc->tdc_tree = de;
863 dc->tdc_current = de;
864 if (de != NULL && tmpfs_dirent_duphead(de)) {
865 dc->tdc_current = LIST_FIRST(&de->ud.td_duphead);
866 MPASS(dc->tdc_current != NULL);
867 }
868 return (dc->tdc_current);
869
870 out:
871 dc->tdc_tree = de;
872 dc->tdc_current = de;
873 if (de != NULL && tmpfs_dirent_dup(de))
874 dc->tdc_tree = tmpfs_dir_xlookup_hash(node,
875 de->td_hash);
876 return (dc->tdc_current);
877 }
878
879 /*
880 * Looks for a directory entry in the directory represented by node.
881 * 'cnp' describes the name of the entry to look for. Note that the .
882 * and .. components are not allowed as they do not physically exist
883 * within directories.
884 *
885 * Returns a pointer to the entry when found, otherwise NULL.
886 */
887 struct tmpfs_dirent *
888 tmpfs_dir_lookup(struct tmpfs_node *node, struct tmpfs_node *f,
889 struct componentname *cnp)
890 {
891 struct tmpfs_dir_duphead *duphead;
892 struct tmpfs_dirent *de;
893 uint32_t hash;
894
895 MPASS(IMPLIES(cnp->cn_namelen == 1, cnp->cn_nameptr[0] != '.'));
896 MPASS(IMPLIES(cnp->cn_namelen == 2, !(cnp->cn_nameptr[0] == '.' &&
897 cnp->cn_nameptr[1] == '.')));
898 TMPFS_VALIDATE_DIR(node);
899
900 hash = tmpfs_dirent_hash(cnp->cn_nameptr, cnp->cn_namelen);
901 de = tmpfs_dir_xlookup_hash(node, hash);
902 if (de != NULL && tmpfs_dirent_duphead(de)) {
903 duphead = &de->ud.td_duphead;
904 LIST_FOREACH(de, duphead, uh.td_dup.entries) {
905 if (TMPFS_DIRENT_MATCHES(de, cnp->cn_nameptr,
906 cnp->cn_namelen))
907 break;
908 }
909 } else if (de != NULL) {
910 if (!TMPFS_DIRENT_MATCHES(de, cnp->cn_nameptr,
911 cnp->cn_namelen))
912 de = NULL;
913 }
914 if (de != NULL && f != NULL && de->td_node != f)
915 de = NULL;
916
917 return (de);
918 }
919
920 /*
921 * Attach duplicate-cookie directory entry nde to dnode and insert to dupindex
922 * list, allocate new cookie value.
923 */
924 static void
925 tmpfs_dir_attach_dup(struct tmpfs_node *dnode,
926 struct tmpfs_dir_duphead *duphead, struct tmpfs_dirent *nde)
927 {
928 struct tmpfs_dir_duphead *dupindex;
929 struct tmpfs_dirent *de, *pde;
930
931 dupindex = &dnode->tn_dir.tn_dupindex;
932 de = LIST_FIRST(dupindex);
933 if (de == NULL || de->td_cookie < TMPFS_DIRCOOKIE_DUP_MAX) {
934 if (de == NULL)
935 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MIN;
936 else
937 nde->td_cookie = de->td_cookie + 1;
938 MPASS(tmpfs_dirent_dup(nde));
939 LIST_INSERT_HEAD(dupindex, nde, uh.td_dup.index_entries);
940 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries);
941 return;
942 }
943
944 /*
945 * Cookie numbers are near exhaustion. Scan dupindex list for unused
946 * numbers. dupindex list is sorted in descending order. Keep it so
947 * after inserting nde.
948 */
949 while (1) {
950 pde = de;
951 de = LIST_NEXT(de, uh.td_dup.index_entries);
952 if (de == NULL && pde->td_cookie != TMPFS_DIRCOOKIE_DUP_MIN) {
953 /*
954 * Last element of the index doesn't have minimal cookie
955 * value, use it.
956 */
957 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MIN;
958 LIST_INSERT_AFTER(pde, nde, uh.td_dup.index_entries);
959 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries);
960 return;
961 } else if (de == NULL) {
962 /*
963 * We are so lucky have 2^30 hash duplicates in single
964 * directory :) Return largest possible cookie value.
965 * It should be fine except possible issues with
966 * VOP_READDIR restart.
967 */
968 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MAX;
969 LIST_INSERT_HEAD(dupindex, nde,
970 uh.td_dup.index_entries);
971 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries);
972 return;
973 }
974 if (de->td_cookie + 1 == pde->td_cookie ||
975 de->td_cookie >= TMPFS_DIRCOOKIE_DUP_MAX)
976 continue; /* No hole or invalid cookie. */
977 nde->td_cookie = de->td_cookie + 1;
978 MPASS(tmpfs_dirent_dup(nde));
979 MPASS(pde->td_cookie > nde->td_cookie);
980 MPASS(nde->td_cookie > de->td_cookie);
981 LIST_INSERT_BEFORE(de, nde, uh.td_dup.index_entries);
982 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries);
983 return;
984 }
985 }
986
987 /*
988 * Attaches the directory entry de to the directory represented by vp.
989 * Note that this does not change the link count of the node pointed by
990 * the directory entry, as this is done by tmpfs_alloc_dirent.
991 */
992 void
993 tmpfs_dir_attach(struct vnode *vp, struct tmpfs_dirent *de)
994 {
995 struct tmpfs_node *dnode;
996 struct tmpfs_dirent *xde, *nde;
997
998 ASSERT_VOP_ELOCKED(vp, __func__);
999 MPASS(de->td_namelen > 0);
1000 MPASS(de->td_hash >= TMPFS_DIRCOOKIE_MIN);
1001 MPASS(de->td_cookie == de->td_hash);
1002
1003 dnode = VP_TO_TMPFS_DIR(vp);
1004 dnode->tn_dir.tn_readdir_lastn = 0;
1005 dnode->tn_dir.tn_readdir_lastp = NULL;
1006
1007 MPASS(!tmpfs_dirent_dup(de));
1008 xde = RB_INSERT(tmpfs_dir, &dnode->tn_dir.tn_dirhead, de);
1009 if (xde != NULL && tmpfs_dirent_duphead(xde))
1010 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, de);
1011 else if (xde != NULL) {
1012 /*
1013 * Allocate new duphead. Swap xde with duphead to avoid
1014 * adding/removing elements with the same hash.
1015 */
1016 MPASS(!tmpfs_dirent_dup(xde));
1017 tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), NULL, NULL, 0,
1018 &nde);
1019 /* *nde = *xde; XXX gcc 4.2.1 may generate invalid code. */
1020 memcpy(nde, xde, sizeof(*xde));
1021 xde->td_cookie |= TMPFS_DIRCOOKIE_DUPHEAD;
1022 LIST_INIT(&xde->ud.td_duphead);
1023 xde->td_namelen = 0;
1024 xde->td_node = NULL;
1025 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, nde);
1026 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, de);
1027 }
1028 dnode->tn_size += sizeof(struct tmpfs_dirent);
1029 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
1030 TMPFS_NODE_MODIFIED;
1031 tmpfs_update(vp);
1032 }
1033
1034 /*
1035 * Detaches the directory entry de from the directory represented by vp.
1036 * Note that this does not change the link count of the node pointed by
1037 * the directory entry, as this is done by tmpfs_free_dirent.
1038 */
1039 void
1040 tmpfs_dir_detach(struct vnode *vp, struct tmpfs_dirent *de)
1041 {
1042 struct tmpfs_mount *tmp;
1043 struct tmpfs_dir *head;
1044 struct tmpfs_node *dnode;
1045 struct tmpfs_dirent *xde;
1046
1047 ASSERT_VOP_ELOCKED(vp, __func__);
1048
1049 dnode = VP_TO_TMPFS_DIR(vp);
1050 head = &dnode->tn_dir.tn_dirhead;
1051 dnode->tn_dir.tn_readdir_lastn = 0;
1052 dnode->tn_dir.tn_readdir_lastp = NULL;
1053
1054 if (tmpfs_dirent_dup(de)) {
1055 /* Remove duphead if de was last entry. */
1056 if (LIST_NEXT(de, uh.td_dup.entries) == NULL) {
1057 xde = tmpfs_dir_xlookup_hash(dnode, de->td_hash);
1058 MPASS(tmpfs_dirent_duphead(xde));
1059 } else
1060 xde = NULL;
1061 LIST_REMOVE(de, uh.td_dup.entries);
1062 LIST_REMOVE(de, uh.td_dup.index_entries);
1063 if (xde != NULL) {
1064 if (LIST_EMPTY(&xde->ud.td_duphead)) {
1065 RB_REMOVE(tmpfs_dir, head, xde);
1066 tmp = VFS_TO_TMPFS(vp->v_mount);
1067 MPASS(xde->td_node == NULL);
1068 tmpfs_free_dirent(tmp, xde);
1069 }
1070 }
1071 de->td_cookie = de->td_hash;
1072 } else
1073 RB_REMOVE(tmpfs_dir, head, de);
1074
1075 dnode->tn_size -= sizeof(struct tmpfs_dirent);
1076 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
1077 TMPFS_NODE_MODIFIED;
1078 tmpfs_update(vp);
1079 }
1080
1081 void
1082 tmpfs_dir_destroy(struct tmpfs_mount *tmp, struct tmpfs_node *dnode)
1083 {
1084 struct tmpfs_dirent *de, *dde, *nde;
1085
1086 RB_FOREACH_SAFE(de, tmpfs_dir, &dnode->tn_dir.tn_dirhead, nde) {
1087 RB_REMOVE(tmpfs_dir, &dnode->tn_dir.tn_dirhead, de);
1088 /* Node may already be destroyed. */
1089 de->td_node = NULL;
1090 if (tmpfs_dirent_duphead(de)) {
1091 while ((dde = LIST_FIRST(&de->ud.td_duphead)) != NULL) {
1092 LIST_REMOVE(dde, uh.td_dup.entries);
1093 dde->td_node = NULL;
1094 tmpfs_free_dirent(tmp, dde);
1095 }
1096 }
1097 tmpfs_free_dirent(tmp, de);
1098 }
1099 }
1100
1101 /*
1102 * Helper function for tmpfs_readdir. Creates a '.' entry for the given
1103 * directory and returns it in the uio space. The function returns 0
1104 * on success, -1 if there was not enough space in the uio structure to
1105 * hold the directory entry or an appropriate error code if another
1106 * error happens.
1107 */
1108 static int
1109 tmpfs_dir_getdotdent(struct tmpfs_mount *tm, struct tmpfs_node *node,
1110 struct uio *uio)
1111 {
1112 int error;
1113 struct dirent dent;
1114
1115 TMPFS_VALIDATE_DIR(node);
1116 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOT);
1117
1118 dent.d_fileno = node->tn_id;
1119 dent.d_type = DT_DIR;
1120 dent.d_namlen = 1;
1121 dent.d_name[0] = '.';
1122 dent.d_reclen = GENERIC_DIRSIZ(&dent);
1123 dirent_terminate(&dent);
1124
1125 if (dent.d_reclen > uio->uio_resid)
1126 error = EJUSTRETURN;
1127 else
1128 error = uiomove(&dent, dent.d_reclen, uio);
1129
1130 tmpfs_set_status(tm, node, TMPFS_NODE_ACCESSED);
1131
1132 return (error);
1133 }
1134
1135 /*
1136 * Helper function for tmpfs_readdir. Creates a '..' entry for the given
1137 * directory and returns it in the uio space. The function returns 0
1138 * on success, -1 if there was not enough space in the uio structure to
1139 * hold the directory entry or an appropriate error code if another
1140 * error happens.
1141 */
1142 static int
1143 tmpfs_dir_getdotdotdent(struct tmpfs_mount *tm, struct tmpfs_node *node,
1144 struct uio *uio)
1145 {
1146 struct tmpfs_node *parent;
1147 struct dirent dent;
1148 int error;
1149
1150 TMPFS_VALIDATE_DIR(node);
1151 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT);
1152
1153 /*
1154 * Return ENOENT if the current node is already removed.
1155 */
1156 TMPFS_ASSERT_LOCKED(node);
1157 parent = node->tn_dir.tn_parent;
1158 if (parent == NULL)
1159 return (ENOENT);
1160
1161 TMPFS_NODE_LOCK(parent);
1162 dent.d_fileno = parent->tn_id;
1163 TMPFS_NODE_UNLOCK(parent);
1164
1165 dent.d_type = DT_DIR;
1166 dent.d_namlen = 2;
1167 dent.d_name[0] = '.';
1168 dent.d_name[1] = '.';
1169 dent.d_reclen = GENERIC_DIRSIZ(&dent);
1170 dirent_terminate(&dent);
1171
1172 if (dent.d_reclen > uio->uio_resid)
1173 error = EJUSTRETURN;
1174 else
1175 error = uiomove(&dent, dent.d_reclen, uio);
1176
1177 tmpfs_set_status(tm, node, TMPFS_NODE_ACCESSED);
1178
1179 return (error);
1180 }
1181
1182 /*
1183 * Helper function for tmpfs_readdir. Returns as much directory entries
1184 * as can fit in the uio space. The read starts at uio->uio_offset.
1185 * The function returns 0 on success, -1 if there was not enough space
1186 * in the uio structure to hold the directory entry or an appropriate
1187 * error code if another error happens.
1188 */
1189 int
1190 tmpfs_dir_getdents(struct tmpfs_mount *tm, struct tmpfs_node *node,
1191 struct uio *uio, int maxcookies, u_long *cookies, int *ncookies)
1192 {
1193 struct tmpfs_dir_cursor dc;
1194 struct tmpfs_dirent *de;
1195 off_t off;
1196 int error;
1197
1198 TMPFS_VALIDATE_DIR(node);
1199
1200 off = 0;
1201
1202 /*
1203 * Lookup the node from the current offset. The starting offset of
1204 * 0 will lookup both '.' and '..', and then the first real entry,
1205 * or EOF if there are none. Then find all entries for the dir that
1206 * fit into the buffer. Once no more entries are found (de == NULL),
1207 * the offset is set to TMPFS_DIRCOOKIE_EOF, which will cause the next
1208 * call to return 0.
1209 */
1210 switch (uio->uio_offset) {
1211 case TMPFS_DIRCOOKIE_DOT:
1212 error = tmpfs_dir_getdotdent(tm, node, uio);
1213 if (error != 0)
1214 return (error);
1215 uio->uio_offset = TMPFS_DIRCOOKIE_DOTDOT;
1216 if (cookies != NULL)
1217 cookies[(*ncookies)++] = off = uio->uio_offset;
1218 /* FALLTHROUGH */
1219 case TMPFS_DIRCOOKIE_DOTDOT:
1220 error = tmpfs_dir_getdotdotdent(tm, node, uio);
1221 if (error != 0)
1222 return (error);
1223 de = tmpfs_dir_first(node, &dc);
1224 uio->uio_offset = tmpfs_dirent_cookie(de);
1225 if (cookies != NULL)
1226 cookies[(*ncookies)++] = off = uio->uio_offset;
1227 /* EOF. */
1228 if (de == NULL)
1229 return (0);
1230 break;
1231 case TMPFS_DIRCOOKIE_EOF:
1232 return (0);
1233 default:
1234 de = tmpfs_dir_lookup_cookie(node, uio->uio_offset, &dc);
1235 if (de == NULL)
1236 return (EINVAL);
1237 if (cookies != NULL)
1238 off = tmpfs_dirent_cookie(de);
1239 }
1240
1241 /* Read as much entries as possible; i.e., until we reach the end of
1242 * the directory or we exhaust uio space. */
1243 do {
1244 struct dirent d;
1245
1246 /* Create a dirent structure representing the current
1247 * tmpfs_node and fill it. */
1248 if (de->td_node == NULL) {
1249 d.d_fileno = 1;
1250 d.d_type = DT_WHT;
1251 } else {
1252 d.d_fileno = de->td_node->tn_id;
1253 switch (de->td_node->tn_type) {
1254 case VBLK:
1255 d.d_type = DT_BLK;
1256 break;
1257
1258 case VCHR:
1259 d.d_type = DT_CHR;
1260 break;
1261
1262 case VDIR:
1263 d.d_type = DT_DIR;
1264 break;
1265
1266 case VFIFO:
1267 d.d_type = DT_FIFO;
1268 break;
1269
1270 case VLNK:
1271 d.d_type = DT_LNK;
1272 break;
1273
1274 case VREG:
1275 d.d_type = DT_REG;
1276 break;
1277
1278 case VSOCK:
1279 d.d_type = DT_SOCK;
1280 break;
1281
1282 default:
1283 panic("tmpfs_dir_getdents: type %p %d",
1284 de->td_node, (int)de->td_node->tn_type);
1285 }
1286 }
1287 d.d_namlen = de->td_namelen;
1288 MPASS(de->td_namelen < sizeof(d.d_name));
1289 (void)memcpy(d.d_name, de->ud.td_name, de->td_namelen);
1290 d.d_reclen = GENERIC_DIRSIZ(&d);
1291 dirent_terminate(&d);
1292
1293 /* Stop reading if the directory entry we are treating is
1294 * bigger than the amount of data that can be returned. */
1295 if (d.d_reclen > uio->uio_resid) {
1296 error = EJUSTRETURN;
1297 break;
1298 }
1299
1300 /* Copy the new dirent structure into the output buffer and
1301 * advance pointers. */
1302 error = uiomove(&d, d.d_reclen, uio);
1303 if (error == 0) {
1304 de = tmpfs_dir_next(node, &dc);
1305 if (cookies != NULL) {
1306 off = tmpfs_dirent_cookie(de);
1307 MPASS(*ncookies < maxcookies);
1308 cookies[(*ncookies)++] = off;
1309 }
1310 }
1311 } while (error == 0 && uio->uio_resid > 0 && de != NULL);
1312
1313 /* Skip setting off when using cookies as it is already done above. */
1314 if (cookies == NULL)
1315 off = tmpfs_dirent_cookie(de);
1316
1317 /* Update the offset and cache. */
1318 uio->uio_offset = off;
1319 node->tn_dir.tn_readdir_lastn = off;
1320 node->tn_dir.tn_readdir_lastp = de;
1321
1322 tmpfs_set_status(tm, node, TMPFS_NODE_ACCESSED);
1323 return error;
1324 }
1325
1326 int
1327 tmpfs_dir_whiteout_add(struct vnode *dvp, struct componentname *cnp)
1328 {
1329 struct tmpfs_dirent *de;
1330 int error;
1331
1332 error = tmpfs_alloc_dirent(VFS_TO_TMPFS(dvp->v_mount), NULL,
1333 cnp->cn_nameptr, cnp->cn_namelen, &de);
1334 if (error != 0)
1335 return (error);
1336 tmpfs_dir_attach(dvp, de);
1337 return (0);
1338 }
1339
1340 void
1341 tmpfs_dir_whiteout_remove(struct vnode *dvp, struct componentname *cnp)
1342 {
1343 struct tmpfs_dirent *de;
1344
1345 de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(dvp), NULL, cnp);
1346 MPASS(de != NULL && de->td_node == NULL);
1347 tmpfs_dir_detach(dvp, de);
1348 tmpfs_free_dirent(VFS_TO_TMPFS(dvp->v_mount), de);
1349 }
1350
1351 /*
1352 * Resizes the aobj associated with the regular file pointed to by 'vp' to the
1353 * size 'newsize'. 'vp' must point to a vnode that represents a regular file.
1354 * 'newsize' must be positive.
1355 *
1356 * Returns zero on success or an appropriate error code on failure.
1357 */
1358 int
1359 tmpfs_reg_resize(struct vnode *vp, off_t newsize, boolean_t ignerr)
1360 {
1361 struct tmpfs_mount *tmp;
1362 struct tmpfs_node *node;
1363 vm_object_t uobj;
1364 vm_page_t m;
1365 vm_pindex_t idx, newpages, oldpages;
1366 off_t oldsize;
1367 int base, rv;
1368
1369 MPASS(vp->v_type == VREG);
1370 MPASS(newsize >= 0);
1371
1372 node = VP_TO_TMPFS_NODE(vp);
1373 uobj = node->tn_reg.tn_aobj;
1374 tmp = VFS_TO_TMPFS(vp->v_mount);
1375
1376 /*
1377 * Convert the old and new sizes to the number of pages needed to
1378 * store them. It may happen that we do not need to do anything
1379 * because the last allocated page can accommodate the change on
1380 * its own.
1381 */
1382 oldsize = node->tn_size;
1383 oldpages = OFF_TO_IDX(oldsize + PAGE_MASK);
1384 MPASS(oldpages == uobj->size);
1385 newpages = OFF_TO_IDX(newsize + PAGE_MASK);
1386
1387 if (__predict_true(newpages == oldpages && newsize >= oldsize)) {
1388 node->tn_size = newsize;
1389 return (0);
1390 }
1391
1392 if (newpages > oldpages &&
1393 tmpfs_pages_check_avail(tmp, newpages - oldpages) == 0)
1394 return (ENOSPC);
1395
1396 VM_OBJECT_WLOCK(uobj);
1397 if (newsize < oldsize) {
1398 /*
1399 * Zero the truncated part of the last page.
1400 */
1401 base = newsize & PAGE_MASK;
1402 if (base != 0) {
1403 idx = OFF_TO_IDX(newsize);
1404 retry:
1405 m = vm_page_lookup(uobj, idx);
1406 if (m != NULL) {
1407 if (vm_page_sleep_if_busy(m, "tmfssz"))
1408 goto retry;
1409 MPASS(m->valid == VM_PAGE_BITS_ALL);
1410 } else if (vm_pager_has_page(uobj, idx, NULL, NULL)) {
1411 m = vm_page_alloc(uobj, idx, VM_ALLOC_NORMAL |
1412 VM_ALLOC_WAITFAIL);
1413 if (m == NULL)
1414 goto retry;
1415 rv = vm_pager_get_pages(uobj, &m, 1, NULL,
1416 NULL);
1417 vm_page_lock(m);
1418 if (rv == VM_PAGER_OK) {
1419 vm_page_deactivate(m);
1420 vm_page_unlock(m);
1421 vm_page_xunbusy(m);
1422 } else {
1423 vm_page_free(m);
1424 vm_page_unlock(m);
1425 if (ignerr)
1426 m = NULL;
1427 else {
1428 VM_OBJECT_WUNLOCK(uobj);
1429 return (EIO);
1430 }
1431 }
1432 }
1433 if (m != NULL) {
1434 pmap_zero_page_area(m, base, PAGE_SIZE - base);
1435 vm_page_dirty(m);
1436 vm_pager_page_unswapped(m);
1437 }
1438 }
1439
1440 /*
1441 * Release any swap space and free any whole pages.
1442 */
1443 if (newpages < oldpages) {
1444 swap_pager_freespace(uobj, newpages, oldpages -
1445 newpages);
1446 vm_object_page_remove(uobj, newpages, 0, 0);
1447 }
1448 }
1449 uobj->size = newpages;
1450 VM_OBJECT_WUNLOCK(uobj);
1451
1452 atomic_add_long(&tmp->tm_pages_used, newpages - oldpages);
1453
1454 node->tn_size = newsize;
1455 return (0);
1456 }
1457
1458 void
1459 tmpfs_check_mtime(struct vnode *vp)
1460 {
1461 struct tmpfs_node *node;
1462 struct vm_object *obj;
1463
1464 ASSERT_VOP_ELOCKED(vp, "check_mtime");
1465 if (vp->v_type != VREG)
1466 return;
1467 obj = vp->v_object;
1468 KASSERT((obj->flags & (OBJ_TMPFS_NODE | OBJ_TMPFS)) ==
1469 (OBJ_TMPFS_NODE | OBJ_TMPFS), ("non-tmpfs obj"));
1470 /* unlocked read */
1471 if ((obj->flags & OBJ_TMPFS_DIRTY) != 0) {
1472 VM_OBJECT_WLOCK(obj);
1473 if ((obj->flags & OBJ_TMPFS_DIRTY) != 0) {
1474 obj->flags &= ~OBJ_TMPFS_DIRTY;
1475 node = VP_TO_TMPFS_NODE(vp);
1476 node->tn_status |= TMPFS_NODE_MODIFIED |
1477 TMPFS_NODE_CHANGED;
1478 }
1479 VM_OBJECT_WUNLOCK(obj);
1480 }
1481 }
1482
1483 /*
1484 * Change flags of the given vnode.
1485 * Caller should execute tmpfs_update on vp after a successful execution.
1486 * The vnode must be locked on entry and remain locked on exit.
1487 */
1488 int
1489 tmpfs_chflags(struct vnode *vp, u_long flags, struct ucred *cred,
1490 struct thread *p)
1491 {
1492 int error;
1493 struct tmpfs_node *node;
1494
1495 ASSERT_VOP_ELOCKED(vp, "chflags");
1496
1497 node = VP_TO_TMPFS_NODE(vp);
1498
1499 if ((flags & ~(SF_APPEND | SF_ARCHIVED | SF_IMMUTABLE | SF_NOUNLINK |
1500 UF_APPEND | UF_ARCHIVE | UF_HIDDEN | UF_IMMUTABLE | UF_NODUMP |
1501 UF_NOUNLINK | UF_OFFLINE | UF_OPAQUE | UF_READONLY | UF_REPARSE |
1502 UF_SPARSE | UF_SYSTEM)) != 0)
1503 return (EOPNOTSUPP);
1504
1505 /* Disallow this operation if the file system is mounted read-only. */
1506 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1507 return EROFS;
1508
1509 /*
1510 * Callers may only modify the file flags on objects they
1511 * have VADMIN rights for.
1512 */
1513 if ((error = VOP_ACCESS(vp, VADMIN, cred, p)))
1514 return (error);
1515 /*
1516 * Unprivileged processes are not permitted to unset system
1517 * flags, or modify flags if any system flags are set.
1518 */
1519 if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0)) {
1520 if (node->tn_flags &
1521 (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND)) {
1522 error = securelevel_gt(cred, 0);
1523 if (error)
1524 return (error);
1525 }
1526 } else {
1527 if (node->tn_flags &
1528 (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND) ||
1529 ((flags ^ node->tn_flags) & SF_SETTABLE))
1530 return (EPERM);
1531 }
1532 node->tn_flags = flags;
1533 node->tn_status |= TMPFS_NODE_CHANGED;
1534
1535 ASSERT_VOP_ELOCKED(vp, "chflags2");
1536
1537 return (0);
1538 }
1539
1540 /*
1541 * Change access mode on the given vnode.
1542 * Caller should execute tmpfs_update on vp after a successful execution.
1543 * The vnode must be locked on entry and remain locked on exit.
1544 */
1545 int
1546 tmpfs_chmod(struct vnode *vp, mode_t mode, struct ucred *cred, struct thread *p)
1547 {
1548 int error;
1549 struct tmpfs_node *node;
1550
1551 ASSERT_VOP_ELOCKED(vp, "chmod");
1552
1553 node = VP_TO_TMPFS_NODE(vp);
1554
1555 /* Disallow this operation if the file system is mounted read-only. */
1556 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1557 return EROFS;
1558
1559 /* Immutable or append-only files cannot be modified, either. */
1560 if (node->tn_flags & (IMMUTABLE | APPEND))
1561 return EPERM;
1562
1563 /*
1564 * To modify the permissions on a file, must possess VADMIN
1565 * for that file.
1566 */
1567 if ((error = VOP_ACCESS(vp, VADMIN, cred, p)))
1568 return (error);
1569
1570 /*
1571 * Privileged processes may set the sticky bit on non-directories,
1572 * as well as set the setgid bit on a file with a group that the
1573 * process is not a member of.
1574 */
1575 if (vp->v_type != VDIR && (mode & S_ISTXT)) {
1576 if (priv_check_cred(cred, PRIV_VFS_STICKYFILE, 0))
1577 return (EFTYPE);
1578 }
1579 if (!groupmember(node->tn_gid, cred) && (mode & S_ISGID)) {
1580 error = priv_check_cred(cred, PRIV_VFS_SETGID, 0);
1581 if (error)
1582 return (error);
1583 }
1584
1585
1586 node->tn_mode &= ~ALLPERMS;
1587 node->tn_mode |= mode & ALLPERMS;
1588
1589 node->tn_status |= TMPFS_NODE_CHANGED;
1590
1591 ASSERT_VOP_ELOCKED(vp, "chmod2");
1592
1593 return (0);
1594 }
1595
1596 /*
1597 * Change ownership of the given vnode. At least one of uid or gid must
1598 * be different than VNOVAL. If one is set to that value, the attribute
1599 * is unchanged.
1600 * Caller should execute tmpfs_update on vp after a successful execution.
1601 * The vnode must be locked on entry and remain locked on exit.
1602 */
1603 int
1604 tmpfs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred,
1605 struct thread *p)
1606 {
1607 int error;
1608 struct tmpfs_node *node;
1609 uid_t ouid;
1610 gid_t ogid;
1611
1612 ASSERT_VOP_ELOCKED(vp, "chown");
1613
1614 node = VP_TO_TMPFS_NODE(vp);
1615
1616 /* Assign default values if they are unknown. */
1617 MPASS(uid != VNOVAL || gid != VNOVAL);
1618 if (uid == VNOVAL)
1619 uid = node->tn_uid;
1620 if (gid == VNOVAL)
1621 gid = node->tn_gid;
1622 MPASS(uid != VNOVAL && gid != VNOVAL);
1623
1624 /* Disallow this operation if the file system is mounted read-only. */
1625 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1626 return EROFS;
1627
1628 /* Immutable or append-only files cannot be modified, either. */
1629 if (node->tn_flags & (IMMUTABLE | APPEND))
1630 return EPERM;
1631
1632 /*
1633 * To modify the ownership of a file, must possess VADMIN for that
1634 * file.
1635 */
1636 if ((error = VOP_ACCESS(vp, VADMIN, cred, p)))
1637 return (error);
1638
1639 /*
1640 * To change the owner of a file, or change the group of a file to a
1641 * group of which we are not a member, the caller must have
1642 * privilege.
1643 */
1644 if ((uid != node->tn_uid ||
1645 (gid != node->tn_gid && !groupmember(gid, cred))) &&
1646 (error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0)))
1647 return (error);
1648
1649 ogid = node->tn_gid;
1650 ouid = node->tn_uid;
1651
1652 node->tn_uid = uid;
1653 node->tn_gid = gid;
1654
1655 node->tn_status |= TMPFS_NODE_CHANGED;
1656
1657 if ((node->tn_mode & (S_ISUID | S_ISGID)) && (ouid != uid || ogid != gid)) {
1658 if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID, 0))
1659 node->tn_mode &= ~(S_ISUID | S_ISGID);
1660 }
1661
1662 ASSERT_VOP_ELOCKED(vp, "chown2");
1663
1664 return (0);
1665 }
1666
1667 /*
1668 * Change size of the given vnode.
1669 * Caller should execute tmpfs_update on vp after a successful execution.
1670 * The vnode must be locked on entry and remain locked on exit.
1671 */
1672 int
1673 tmpfs_chsize(struct vnode *vp, u_quad_t size, struct ucred *cred,
1674 struct thread *p)
1675 {
1676 int error;
1677 struct tmpfs_node *node;
1678
1679 ASSERT_VOP_ELOCKED(vp, "chsize");
1680
1681 node = VP_TO_TMPFS_NODE(vp);
1682
1683 /* Decide whether this is a valid operation based on the file type. */
1684 error = 0;
1685 switch (vp->v_type) {
1686 case VDIR:
1687 return EISDIR;
1688
1689 case VREG:
1690 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1691 return EROFS;
1692 break;
1693
1694 case VBLK:
1695 /* FALLTHROUGH */
1696 case VCHR:
1697 /* FALLTHROUGH */
1698 case VFIFO:
1699 /* Allow modifications of special files even if in the file
1700 * system is mounted read-only (we are not modifying the
1701 * files themselves, but the objects they represent). */
1702 return 0;
1703
1704 default:
1705 /* Anything else is unsupported. */
1706 return EOPNOTSUPP;
1707 }
1708
1709 /* Immutable or append-only files cannot be modified, either. */
1710 if (node->tn_flags & (IMMUTABLE | APPEND))
1711 return EPERM;
1712
1713 error = tmpfs_truncate(vp, size);
1714 /* tmpfs_truncate will raise the NOTE_EXTEND and NOTE_ATTRIB kevents
1715 * for us, as will update tn_status; no need to do that here. */
1716
1717 ASSERT_VOP_ELOCKED(vp, "chsize2");
1718
1719 return (error);
1720 }
1721
1722 /*
1723 * Change access and modification times of the given vnode.
1724 * Caller should execute tmpfs_update on vp after a successful execution.
1725 * The vnode must be locked on entry and remain locked on exit.
1726 */
1727 int
1728 tmpfs_chtimes(struct vnode *vp, struct vattr *vap,
1729 struct ucred *cred, struct thread *l)
1730 {
1731 int error;
1732 struct tmpfs_node *node;
1733
1734 ASSERT_VOP_ELOCKED(vp, "chtimes");
1735
1736 node = VP_TO_TMPFS_NODE(vp);
1737
1738 /* Disallow this operation if the file system is mounted read-only. */
1739 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1740 return EROFS;
1741
1742 /* Immutable or append-only files cannot be modified, either. */
1743 if (node->tn_flags & (IMMUTABLE | APPEND))
1744 return EPERM;
1745
1746 error = vn_utimes_perm(vp, vap, cred, l);
1747 if (error != 0)
1748 return (error);
1749
1750 if (vap->va_atime.tv_sec != VNOVAL)
1751 node->tn_status |= TMPFS_NODE_ACCESSED;
1752
1753 if (vap->va_mtime.tv_sec != VNOVAL)
1754 node->tn_status |= TMPFS_NODE_MODIFIED;
1755
1756 if (vap->va_birthtime.tv_sec != VNOVAL)
1757 node->tn_status |= TMPFS_NODE_MODIFIED;
1758
1759 tmpfs_itimes(vp, &vap->va_atime, &vap->va_mtime);
1760
1761 if (vap->va_birthtime.tv_sec != VNOVAL)
1762 node->tn_birthtime = vap->va_birthtime;
1763 ASSERT_VOP_ELOCKED(vp, "chtimes2");
1764
1765 return (0);
1766 }
1767
1768 void
1769 tmpfs_set_status(struct tmpfs_mount *tm, struct tmpfs_node *node, int status)
1770 {
1771
1772 if ((node->tn_status & status) == status || tm->tm_ronly)
1773 return;
1774 TMPFS_NODE_LOCK(node);
1775 node->tn_status |= status;
1776 TMPFS_NODE_UNLOCK(node);
1777 }
1778
1779 /* Sync timestamps */
1780 void
1781 tmpfs_itimes(struct vnode *vp, const struct timespec *acc,
1782 const struct timespec *mod)
1783 {
1784 struct tmpfs_node *node;
1785 struct timespec now;
1786
1787 ASSERT_VOP_LOCKED(vp, "tmpfs_itimes");
1788 node = VP_TO_TMPFS_NODE(vp);
1789
1790 if ((node->tn_status & (TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
1791 TMPFS_NODE_CHANGED)) == 0)
1792 return;
1793
1794 vfs_timestamp(&now);
1795 TMPFS_NODE_LOCK(node);
1796 if (node->tn_status & TMPFS_NODE_ACCESSED) {
1797 if (acc == NULL)
1798 acc = &now;
1799 node->tn_atime = *acc;
1800 }
1801 if (node->tn_status & TMPFS_NODE_MODIFIED) {
1802 if (mod == NULL)
1803 mod = &now;
1804 node->tn_mtime = *mod;
1805 }
1806 if (node->tn_status & TMPFS_NODE_CHANGED)
1807 node->tn_ctime = now;
1808 node->tn_status &= ~(TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
1809 TMPFS_NODE_CHANGED);
1810 TMPFS_NODE_UNLOCK(node);
1811
1812 /* XXX: FIX? The entropy here is desirable, but the harvesting may be expensive */
1813 random_harvest_queue(node, sizeof(*node), 1, RANDOM_FS_ATIME);
1814 }
1815
1816 void
1817 tmpfs_update(struct vnode *vp)
1818 {
1819
1820 tmpfs_itimes(vp, NULL, NULL);
1821 }
1822
1823 int
1824 tmpfs_truncate(struct vnode *vp, off_t length)
1825 {
1826 int error;
1827 struct tmpfs_node *node;
1828
1829 node = VP_TO_TMPFS_NODE(vp);
1830
1831 if (length < 0) {
1832 error = EINVAL;
1833 goto out;
1834 }
1835
1836 if (node->tn_size == length) {
1837 error = 0;
1838 goto out;
1839 }
1840
1841 if (length > VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize)
1842 return (EFBIG);
1843
1844 error = tmpfs_reg_resize(vp, length, FALSE);
1845 if (error == 0)
1846 node->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1847
1848 out:
1849 tmpfs_update(vp);
1850
1851 return (error);
1852 }
1853
1854 static __inline int
1855 tmpfs_dirtree_cmp(struct tmpfs_dirent *a, struct tmpfs_dirent *b)
1856 {
1857 if (a->td_hash > b->td_hash)
1858 return (1);
1859 else if (a->td_hash < b->td_hash)
1860 return (-1);
1861 return (0);
1862 }
1863
1864 RB_GENERATE_STATIC(tmpfs_dir, tmpfs_dirent, uh.td_entries, tmpfs_dirtree_cmp);
Cache object: 29e9d895ecd335e80cef867edfab85b6
|