1 /* $NetBSD: tmpfs_subr.c,v 1.35 2007/07/09 21:10:50 ad Exp $ */
2
3 /*-
4 * SPDX-License-Identifier: BSD-2-Clause-NetBSD
5 *
6 * Copyright (c) 2005 The NetBSD Foundation, Inc.
7 * All rights reserved.
8 *
9 * This code is derived from software contributed to The NetBSD Foundation
10 * by Julio M. Merino Vidal, developed as part of Google's Summer of Code
11 * 2005 program.
12 *
13 * Redistribution and use in source and binary forms, with or without
14 * modification, are permitted provided that the following conditions
15 * are met:
16 * 1. Redistributions of source code must retain the above copyright
17 * notice, this list of conditions and the following disclaimer.
18 * 2. Redistributions in binary form must reproduce the above copyright
19 * notice, this list of conditions and the following disclaimer in the
20 * documentation and/or other materials provided with the distribution.
21 *
22 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
23 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
24 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
25 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
26 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
27 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
28 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
29 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
30 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
31 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
32 * POSSIBILITY OF SUCH DAMAGE.
33 */
34
35 /*
36 * Efficient memory file system supporting functions.
37 */
38 #include <sys/cdefs.h>
39 __FBSDID("$FreeBSD$");
40
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/dirent.h>
44 #include <sys/fnv_hash.h>
45 #include <sys/lock.h>
46 #include <sys/limits.h>
47 #include <sys/mount.h>
48 #include <sys/namei.h>
49 #include <sys/priv.h>
50 #include <sys/proc.h>
51 #include <sys/random.h>
52 #include <sys/refcount.h>
53 #include <sys/rwlock.h>
54 #include <sys/smr.h>
55 #include <sys/stat.h>
56 #include <sys/sysctl.h>
57 #include <sys/user.h>
58 #include <sys/vnode.h>
59 #include <sys/vmmeter.h>
60
61 #include <vm/vm.h>
62 #include <vm/vm_param.h>
63 #include <vm/vm_object.h>
64 #include <vm/vm_page.h>
65 #include <vm/vm_pageout.h>
66 #include <vm/vm_pager.h>
67 #include <vm/vm_extern.h>
68 #include <vm/swap_pager.h>
69
70 #include <fs/tmpfs/tmpfs.h>
71 #include <fs/tmpfs/tmpfs_fifoops.h>
72 #include <fs/tmpfs/tmpfs_vnops.h>
73
74 SYSCTL_NODE(_vfs, OID_AUTO, tmpfs, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
75 "tmpfs file system");
76
77 static long tmpfs_pages_reserved = TMPFS_PAGES_MINRESERVED;
78
79 MALLOC_DEFINE(M_TMPFSDIR, "tmpfs dir", "tmpfs dirent structure");
80 static uma_zone_t tmpfs_node_pool;
81 VFS_SMR_DECLARE;
82
83 int tmpfs_pager_type = -1;
84
85 static vm_object_t
86 tmpfs_pager_alloc(void *handle, vm_ooffset_t size, vm_prot_t prot,
87 vm_ooffset_t offset, struct ucred *cred)
88 {
89 vm_object_t object;
90
91 MPASS(handle == NULL);
92 MPASS(offset == 0);
93 object = vm_object_allocate_dyn(tmpfs_pager_type, size,
94 OBJ_COLORED | OBJ_SWAP);
95 if (!swap_pager_init_object(object, NULL, NULL, size, 0)) {
96 vm_object_deallocate(object);
97 object = NULL;
98 }
99 return (object);
100 }
101
102 /*
103 * Make sure tmpfs vnodes with writable mappings can be found on the lazy list.
104 *
105 * This allows for periodic mtime updates while only scanning vnodes which are
106 * plausibly dirty, see tmpfs_update_mtime_lazy.
107 */
108 static void
109 tmpfs_pager_writecount_recalc(vm_object_t object, vm_offset_t old,
110 vm_offset_t new)
111 {
112 struct vnode *vp;
113
114 VM_OBJECT_ASSERT_WLOCKED(object);
115
116 vp = VM_TO_TMPFS_VP(object);
117
118 /*
119 * Forced unmount?
120 */
121 if (vp == NULL) {
122 KASSERT((object->flags & OBJ_TMPFS_VREF) == 0,
123 ("object %p with OBJ_TMPFS_VREF but without vnode",
124 object));
125 VM_OBJECT_WUNLOCK(object);
126 return;
127 }
128
129 if (old == 0) {
130 VNASSERT((object->flags & OBJ_TMPFS_VREF) == 0, vp,
131 ("object without writable mappings has a reference"));
132 VNPASS(vp->v_usecount > 0, vp);
133 } else {
134 VNASSERT((object->flags & OBJ_TMPFS_VREF) != 0, vp,
135 ("object with writable mappings does not "
136 "have a reference"));
137 }
138
139 if (old == new) {
140 VM_OBJECT_WUNLOCK(object);
141 return;
142 }
143
144 if (new == 0) {
145 vm_object_clear_flag(object, OBJ_TMPFS_VREF);
146 VM_OBJECT_WUNLOCK(object);
147 vrele(vp);
148 } else {
149 if ((object->flags & OBJ_TMPFS_VREF) == 0) {
150 vref(vp);
151 vlazy(vp);
152 vm_object_set_flag(object, OBJ_TMPFS_VREF);
153 }
154 VM_OBJECT_WUNLOCK(object);
155 }
156 }
157
158 static void
159 tmpfs_pager_update_writecount(vm_object_t object, vm_offset_t start,
160 vm_offset_t end)
161 {
162 vm_offset_t new, old;
163
164 VM_OBJECT_WLOCK(object);
165 KASSERT((object->flags & OBJ_ANON) == 0,
166 ("%s: object %p with OBJ_ANON", __func__, object));
167 old = object->un_pager.swp.writemappings;
168 object->un_pager.swp.writemappings += (vm_ooffset_t)end - start;
169 new = object->un_pager.swp.writemappings;
170 tmpfs_pager_writecount_recalc(object, old, new);
171 VM_OBJECT_ASSERT_UNLOCKED(object);
172 }
173
174 static void
175 tmpfs_pager_release_writecount(vm_object_t object, vm_offset_t start,
176 vm_offset_t end)
177 {
178 vm_offset_t new, old;
179
180 VM_OBJECT_WLOCK(object);
181 KASSERT((object->flags & OBJ_ANON) == 0,
182 ("%s: object %p with OBJ_ANON", __func__, object));
183 old = object->un_pager.swp.writemappings;
184 object->un_pager.swp.writemappings -= (vm_ooffset_t)end - start;
185 new = object->un_pager.swp.writemappings;
186 tmpfs_pager_writecount_recalc(object, old, new);
187 VM_OBJECT_ASSERT_UNLOCKED(object);
188 }
189
190 static void
191 tmpfs_pager_getvp(vm_object_t object, struct vnode **vpp, bool *vp_heldp)
192 {
193 struct vnode *vp;
194
195 /*
196 * Tmpfs VREG node, which was reclaimed, has tmpfs_pager_type
197 * type. In this case there is no v_writecount to adjust.
198 */
199 if (vp_heldp != NULL)
200 VM_OBJECT_RLOCK(object);
201 else
202 VM_OBJECT_ASSERT_LOCKED(object);
203 if ((object->flags & OBJ_TMPFS) != 0) {
204 vp = VM_TO_TMPFS_VP(object);
205 if (vp != NULL) {
206 *vpp = vp;
207 if (vp_heldp != NULL) {
208 vhold(vp);
209 *vp_heldp = true;
210 }
211 }
212 }
213 if (vp_heldp != NULL)
214 VM_OBJECT_RUNLOCK(object);
215 }
216
217 static void
218 tmpfs_pager_freespace(vm_object_t obj, vm_pindex_t start, vm_size_t size)
219 {
220 struct tmpfs_node *node;
221 struct tmpfs_mount *tm;
222 vm_size_t c;
223
224 swap_pager_freespace(obj, start, size, &c);
225 if ((obj->flags & OBJ_TMPFS) == 0 || c == 0)
226 return;
227
228 node = obj->un_pager.swp.swp_priv;
229 MPASS(node->tn_type == VREG);
230 tm = node->tn_reg.tn_tmp;
231
232 KASSERT(tm->tm_pages_used >= c,
233 ("tmpfs tm %p pages %jd free %jd", tm,
234 (uintmax_t)tm->tm_pages_used, (uintmax_t)c));
235 atomic_add_long(&tm->tm_pages_used, -c);
236 KASSERT(node->tn_reg.tn_pages >= c,
237 ("tmpfs node %p pages %jd free %jd", node,
238 (uintmax_t)node->tn_reg.tn_pages, (uintmax_t)c));
239 node->tn_reg.tn_pages -= c;
240 }
241
242 static void
243 tmpfs_page_inserted(vm_object_t obj, vm_page_t m)
244 {
245 struct tmpfs_node *node;
246 struct tmpfs_mount *tm;
247
248 if ((obj->flags & OBJ_TMPFS) == 0)
249 return;
250
251 node = obj->un_pager.swp.swp_priv;
252 MPASS(node->tn_type == VREG);
253 tm = node->tn_reg.tn_tmp;
254
255 if (!vm_pager_has_page(obj, m->pindex, NULL, NULL)) {
256 atomic_add_long(&tm->tm_pages_used, 1);
257 node->tn_reg.tn_pages += 1;
258 }
259 }
260
261 static void
262 tmpfs_page_removed(vm_object_t obj, vm_page_t m)
263 {
264 struct tmpfs_node *node;
265 struct tmpfs_mount *tm;
266
267 if ((obj->flags & OBJ_TMPFS) == 0)
268 return;
269
270 node = obj->un_pager.swp.swp_priv;
271 MPASS(node->tn_type == VREG);
272 tm = node->tn_reg.tn_tmp;
273
274 if (!vm_pager_has_page(obj, m->pindex, NULL, NULL)) {
275 KASSERT(tm->tm_pages_used >= 1,
276 ("tmpfs tm %p pages %jd free 1", tm,
277 (uintmax_t)tm->tm_pages_used));
278 atomic_add_long(&tm->tm_pages_used, -1);
279 KASSERT(node->tn_reg.tn_pages >= 1,
280 ("tmpfs node %p pages %jd free 1", node,
281 (uintmax_t)node->tn_reg.tn_pages));
282 node->tn_reg.tn_pages -= 1;
283 }
284 }
285
286 static boolean_t
287 tmpfs_can_alloc_page(vm_object_t obj, vm_pindex_t pindex)
288 {
289 struct tmpfs_mount *tm;
290
291 tm = VM_TO_TMPFS_MP(obj);
292 if (tm == NULL || vm_pager_has_page(obj, pindex, NULL, NULL) ||
293 tm->tm_pages_max == 0)
294 return (true);
295 return (tm->tm_pages_max > atomic_load_long(&tm->tm_pages_used));
296 }
297
298 struct pagerops tmpfs_pager_ops = {
299 .pgo_kvme_type = KVME_TYPE_VNODE,
300 .pgo_alloc = tmpfs_pager_alloc,
301 .pgo_set_writeable_dirty = vm_object_set_writeable_dirty_,
302 .pgo_update_writecount = tmpfs_pager_update_writecount,
303 .pgo_release_writecount = tmpfs_pager_release_writecount,
304 .pgo_mightbedirty = vm_object_mightbedirty_,
305 .pgo_getvp = tmpfs_pager_getvp,
306 .pgo_freespace = tmpfs_pager_freespace,
307 .pgo_page_inserted = tmpfs_page_inserted,
308 .pgo_page_removed = tmpfs_page_removed,
309 .pgo_can_alloc_page = tmpfs_can_alloc_page,
310 };
311
312 static int
313 tmpfs_node_ctor(void *mem, int size, void *arg, int flags)
314 {
315 struct tmpfs_node *node;
316
317 node = mem;
318 node->tn_gen++;
319 node->tn_size = 0;
320 node->tn_status = 0;
321 node->tn_accessed = false;
322 node->tn_flags = 0;
323 node->tn_links = 0;
324 node->tn_vnode = NULL;
325 node->tn_vpstate = 0;
326 return (0);
327 }
328
329 static void
330 tmpfs_node_dtor(void *mem, int size, void *arg)
331 {
332 struct tmpfs_node *node;
333
334 node = mem;
335 node->tn_type = VNON;
336 }
337
338 static int
339 tmpfs_node_init(void *mem, int size, int flags)
340 {
341 struct tmpfs_node *node;
342
343 node = mem;
344 node->tn_id = 0;
345 mtx_init(&node->tn_interlock, "tmpfsni", NULL, MTX_DEF);
346 node->tn_gen = arc4random();
347 return (0);
348 }
349
350 static void
351 tmpfs_node_fini(void *mem, int size)
352 {
353 struct tmpfs_node *node;
354
355 node = mem;
356 mtx_destroy(&node->tn_interlock);
357 }
358
359 int
360 tmpfs_subr_init(void)
361 {
362 tmpfs_pager_type = vm_pager_alloc_dyn_type(&tmpfs_pager_ops,
363 OBJT_SWAP);
364 if (tmpfs_pager_type == -1)
365 return (EINVAL);
366 tmpfs_node_pool = uma_zcreate("TMPFS node",
367 sizeof(struct tmpfs_node), tmpfs_node_ctor, tmpfs_node_dtor,
368 tmpfs_node_init, tmpfs_node_fini, UMA_ALIGN_PTR, 0);
369 VFS_SMR_ZONE_SET(tmpfs_node_pool);
370 return (0);
371 }
372
373 void
374 tmpfs_subr_uninit(void)
375 {
376 if (tmpfs_pager_type != -1)
377 vm_pager_free_dyn_type(tmpfs_pager_type);
378 tmpfs_pager_type = -1;
379 uma_zdestroy(tmpfs_node_pool);
380 }
381
382 static int
383 sysctl_mem_reserved(SYSCTL_HANDLER_ARGS)
384 {
385 int error;
386 long pages, bytes;
387
388 pages = *(long *)arg1;
389 bytes = pages * PAGE_SIZE;
390
391 error = sysctl_handle_long(oidp, &bytes, 0, req);
392 if (error || !req->newptr)
393 return (error);
394
395 pages = bytes / PAGE_SIZE;
396 if (pages < TMPFS_PAGES_MINRESERVED)
397 return (EINVAL);
398
399 *(long *)arg1 = pages;
400 return (0);
401 }
402
403 SYSCTL_PROC(_vfs_tmpfs, OID_AUTO, memory_reserved,
404 CTLTYPE_LONG|CTLFLAG_MPSAFE|CTLFLAG_RW, &tmpfs_pages_reserved, 0,
405 sysctl_mem_reserved, "L",
406 "Amount of available memory and swap below which tmpfs growth stops");
407
408 static __inline int tmpfs_dirtree_cmp(struct tmpfs_dirent *a,
409 struct tmpfs_dirent *b);
410 RB_PROTOTYPE_STATIC(tmpfs_dir, tmpfs_dirent, uh.td_entries, tmpfs_dirtree_cmp);
411
412 size_t
413 tmpfs_mem_avail(void)
414 {
415 size_t avail;
416 long reserved;
417
418 avail = swap_pager_avail + vm_free_count();
419 reserved = atomic_load_long(&tmpfs_pages_reserved);
420 if (__predict_false(avail < reserved))
421 return (0);
422 return (avail - reserved);
423 }
424
425 size_t
426 tmpfs_pages_used(struct tmpfs_mount *tmp)
427 {
428 const size_t node_size = sizeof(struct tmpfs_node) +
429 sizeof(struct tmpfs_dirent);
430 size_t meta_pages;
431
432 meta_pages = howmany((uintmax_t)tmp->tm_nodes_inuse * node_size,
433 PAGE_SIZE);
434 return (meta_pages + tmp->tm_pages_used);
435 }
436
437 static bool
438 tmpfs_pages_check_avail(struct tmpfs_mount *tmp, size_t req_pages)
439 {
440 if (tmpfs_mem_avail() < req_pages)
441 return (false);
442
443 if (tmp->tm_pages_max != ULONG_MAX &&
444 tmp->tm_pages_max < req_pages + tmpfs_pages_used(tmp))
445 return (false);
446
447 return (true);
448 }
449
450 static int
451 tmpfs_partial_page_invalidate(vm_object_t object, vm_pindex_t idx, int base,
452 int end, boolean_t ignerr)
453 {
454 vm_page_t m;
455 int rv, error;
456
457 VM_OBJECT_ASSERT_WLOCKED(object);
458 KASSERT(base >= 0, ("%s: base %d", __func__, base));
459 KASSERT(end - base <= PAGE_SIZE, ("%s: base %d end %d", __func__, base,
460 end));
461 error = 0;
462
463 retry:
464 m = vm_page_grab(object, idx, VM_ALLOC_NOCREAT);
465 if (m != NULL) {
466 MPASS(vm_page_all_valid(m));
467 } else if (vm_pager_has_page(object, idx, NULL, NULL)) {
468 m = vm_page_alloc(object, idx, VM_ALLOC_NORMAL |
469 VM_ALLOC_WAITFAIL);
470 if (m == NULL)
471 goto retry;
472 vm_object_pip_add(object, 1);
473 VM_OBJECT_WUNLOCK(object);
474 rv = vm_pager_get_pages(object, &m, 1, NULL, NULL);
475 VM_OBJECT_WLOCK(object);
476 vm_object_pip_wakeup(object);
477 if (rv == VM_PAGER_OK) {
478 /*
479 * Since the page was not resident, and therefore not
480 * recently accessed, immediately enqueue it for
481 * asynchronous laundering. The current operation is
482 * not regarded as an access.
483 */
484 vm_page_launder(m);
485 } else {
486 vm_page_free(m);
487 m = NULL;
488 if (!ignerr)
489 error = EIO;
490 }
491 }
492 if (m != NULL) {
493 pmap_zero_page_area(m, base, end - base);
494 vm_page_set_dirty(m);
495 vm_page_xunbusy(m);
496 }
497
498 return (error);
499 }
500
501 void
502 tmpfs_ref_node(struct tmpfs_node *node)
503 {
504 #ifdef INVARIANTS
505 u_int old;
506
507 old =
508 #endif
509 refcount_acquire(&node->tn_refcount);
510 #ifdef INVARIANTS
511 KASSERT(old > 0, ("node %p zero refcount", node));
512 #endif
513 }
514
515 /*
516 * Allocates a new node of type 'type' inside the 'tmp' mount point, with
517 * its owner set to 'uid', its group to 'gid' and its mode set to 'mode',
518 * using the credentials of the process 'p'.
519 *
520 * If the node type is set to 'VDIR', then the parent parameter must point
521 * to the parent directory of the node being created. It may only be NULL
522 * while allocating the root node.
523 *
524 * If the node type is set to 'VBLK' or 'VCHR', then the rdev parameter
525 * specifies the device the node represents.
526 *
527 * If the node type is set to 'VLNK', then the parameter target specifies
528 * the file name of the target file for the symbolic link that is being
529 * created.
530 *
531 * Note that new nodes are retrieved from the available list if it has
532 * items or, if it is empty, from the node pool as long as there is enough
533 * space to create them.
534 *
535 * Returns zero on success or an appropriate error code on failure.
536 */
537 int
538 tmpfs_alloc_node(struct mount *mp, struct tmpfs_mount *tmp, enum vtype type,
539 uid_t uid, gid_t gid, mode_t mode, struct tmpfs_node *parent,
540 const char *target, dev_t rdev, struct tmpfs_node **node)
541 {
542 struct tmpfs_node *nnode;
543 char *symlink;
544 char symlink_smr;
545
546 /* If the root directory of the 'tmp' file system is not yet
547 * allocated, this must be the request to do it. */
548 MPASS(IMPLIES(tmp->tm_root == NULL, parent == NULL && type == VDIR));
549
550 MPASS(IFF(type == VLNK, target != NULL));
551 MPASS(IFF(type == VBLK || type == VCHR, rdev != VNOVAL));
552
553 if (tmp->tm_nodes_inuse >= tmp->tm_nodes_max)
554 return (ENOSPC);
555 if (!tmpfs_pages_check_avail(tmp, 1))
556 return (ENOSPC);
557
558 if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) {
559 /*
560 * When a new tmpfs node is created for fully
561 * constructed mount point, there must be a parent
562 * node, which vnode is locked exclusively. As
563 * consequence, if the unmount is executing in
564 * parallel, vflush() cannot reclaim the parent vnode.
565 * Due to this, the check for MNTK_UNMOUNT flag is not
566 * racy: if we did not see MNTK_UNMOUNT flag, then tmp
567 * cannot be destroyed until node construction is
568 * finished and the parent vnode unlocked.
569 *
570 * Tmpfs does not need to instantiate new nodes during
571 * unmount.
572 */
573 return (EBUSY);
574 }
575 if ((mp->mnt_kern_flag & MNT_RDONLY) != 0)
576 return (EROFS);
577
578 nnode = uma_zalloc_smr(tmpfs_node_pool, M_WAITOK);
579
580 /* Generic initialization. */
581 nnode->tn_type = type;
582 vfs_timestamp(&nnode->tn_atime);
583 nnode->tn_birthtime = nnode->tn_ctime = nnode->tn_mtime =
584 nnode->tn_atime;
585 nnode->tn_uid = uid;
586 nnode->tn_gid = gid;
587 nnode->tn_mode = mode;
588 nnode->tn_id = alloc_unr64(&tmp->tm_ino_unr);
589 nnode->tn_refcount = 1;
590
591 /* Type-specific initialization. */
592 switch (nnode->tn_type) {
593 case VBLK:
594 case VCHR:
595 nnode->tn_rdev = rdev;
596 break;
597
598 case VDIR:
599 RB_INIT(&nnode->tn_dir.tn_dirhead);
600 LIST_INIT(&nnode->tn_dir.tn_dupindex);
601 MPASS(parent != nnode);
602 MPASS(IMPLIES(parent == NULL, tmp->tm_root == NULL));
603 nnode->tn_dir.tn_parent = (parent == NULL) ? nnode : parent;
604 nnode->tn_dir.tn_readdir_lastn = 0;
605 nnode->tn_dir.tn_readdir_lastp = NULL;
606 nnode->tn_links++;
607 TMPFS_NODE_LOCK(nnode->tn_dir.tn_parent);
608 nnode->tn_dir.tn_parent->tn_links++;
609 TMPFS_NODE_UNLOCK(nnode->tn_dir.tn_parent);
610 break;
611
612 case VFIFO:
613 /* FALLTHROUGH */
614 case VSOCK:
615 break;
616
617 case VLNK:
618 MPASS(strlen(target) < MAXPATHLEN);
619 nnode->tn_size = strlen(target);
620
621 symlink = NULL;
622 if (!tmp->tm_nonc) {
623 symlink = cache_symlink_alloc(nnode->tn_size + 1,
624 M_WAITOK);
625 symlink_smr = true;
626 }
627 if (symlink == NULL) {
628 symlink = malloc(nnode->tn_size + 1, M_TMPFSNAME,
629 M_WAITOK);
630 symlink_smr = false;
631 }
632 memcpy(symlink, target, nnode->tn_size + 1);
633
634 /*
635 * Allow safe symlink resolving for lockless lookup.
636 * tmpfs_fplookup_symlink references this comment.
637 *
638 * 1. nnode is not yet visible to the world
639 * 2. both tn_link_target and tn_link_smr get populated
640 * 3. release fence publishes their content
641 * 4. tn_link_target content is immutable until node
642 * destruction, where the pointer gets set to NULL
643 * 5. tn_link_smr is never changed once set
644 *
645 * As a result it is sufficient to issue load consume
646 * on the node pointer to also get the above content
647 * in a stable manner. Worst case tn_link_smr flag
648 * may be set to true despite being stale, while the
649 * target buffer is already cleared out.
650 */
651 atomic_store_ptr(&nnode->tn_link_target, symlink);
652 atomic_store_char((char *)&nnode->tn_link_smr, symlink_smr);
653 atomic_thread_fence_rel();
654 break;
655
656 case VREG:
657 nnode->tn_reg.tn_aobj =
658 vm_pager_allocate(tmpfs_pager_type, NULL, 0,
659 VM_PROT_DEFAULT, 0,
660 NULL /* XXXKIB - tmpfs needs swap reservation */);
661 nnode->tn_reg.tn_aobj->un_pager.swp.swp_priv = nnode;
662 vm_object_set_flag(nnode->tn_reg.tn_aobj, OBJ_TMPFS);
663 nnode->tn_reg.tn_tmp = tmp;
664 nnode->tn_reg.tn_pages = 0;
665 break;
666
667 default:
668 panic("tmpfs_alloc_node: type %p %d", nnode,
669 (int)nnode->tn_type);
670 }
671
672 TMPFS_LOCK(tmp);
673 LIST_INSERT_HEAD(&tmp->tm_nodes_used, nnode, tn_entries);
674 nnode->tn_attached = true;
675 tmp->tm_nodes_inuse++;
676 tmp->tm_refcount++;
677 TMPFS_UNLOCK(tmp);
678
679 *node = nnode;
680 return (0);
681 }
682
683 /*
684 * Destroys the node pointed to by node from the file system 'tmp'.
685 * If the node references a directory, no entries are allowed.
686 */
687 void
688 tmpfs_free_node(struct tmpfs_mount *tmp, struct tmpfs_node *node)
689 {
690 if (refcount_release_if_not_last(&node->tn_refcount))
691 return;
692
693 TMPFS_LOCK(tmp);
694 TMPFS_NODE_LOCK(node);
695 if (!tmpfs_free_node_locked(tmp, node, false)) {
696 TMPFS_NODE_UNLOCK(node);
697 TMPFS_UNLOCK(tmp);
698 }
699 }
700
701 bool
702 tmpfs_free_node_locked(struct tmpfs_mount *tmp, struct tmpfs_node *node,
703 bool detach)
704 {
705 vm_object_t uobj;
706 char *symlink;
707 bool last;
708
709 TMPFS_MP_ASSERT_LOCKED(tmp);
710 TMPFS_NODE_ASSERT_LOCKED(node);
711
712 last = refcount_release(&node->tn_refcount);
713 if (node->tn_attached && (detach || last)) {
714 MPASS(tmp->tm_nodes_inuse > 0);
715 tmp->tm_nodes_inuse--;
716 LIST_REMOVE(node, tn_entries);
717 node->tn_attached = false;
718 }
719 if (!last)
720 return (false);
721
722 TMPFS_NODE_UNLOCK(node);
723
724 #ifdef INVARIANTS
725 MPASS(node->tn_vnode == NULL);
726 MPASS((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0);
727
728 /*
729 * Make sure this is a node type we can deal with. Everything
730 * is explicitly enumerated without the 'default' clause so
731 * the compiler can throw an error in case a new type is
732 * added.
733 */
734 switch (node->tn_type) {
735 case VBLK:
736 case VCHR:
737 case VDIR:
738 case VFIFO:
739 case VSOCK:
740 case VLNK:
741 case VREG:
742 break;
743 case VNON:
744 case VBAD:
745 case VMARKER:
746 panic("%s: bad type %d for node %p", __func__,
747 (int)node->tn_type, node);
748 }
749 #endif
750
751 switch (node->tn_type) {
752 case VREG:
753 uobj = node->tn_reg.tn_aobj;
754 node->tn_reg.tn_aobj = NULL;
755 if (uobj != NULL) {
756 VM_OBJECT_WLOCK(uobj);
757 KASSERT((uobj->flags & OBJ_TMPFS) != 0,
758 ("tmpfs node %p uobj %p not tmpfs", node, uobj));
759 vm_object_clear_flag(uobj, OBJ_TMPFS);
760 KASSERT(tmp->tm_pages_used >= node->tn_reg.tn_pages,
761 ("tmpfs tmp %p node %p pages %jd free %jd", tmp,
762 node, (uintmax_t)tmp->tm_pages_used,
763 (uintmax_t)node->tn_reg.tn_pages));
764 atomic_add_long(&tmp->tm_pages_used,
765 -node->tn_reg.tn_pages);
766 VM_OBJECT_WUNLOCK(uobj);
767 }
768 tmpfs_free_tmp(tmp);
769
770 /*
771 * vm_object_deallocate() must not be called while
772 * owning tm_allnode_lock, because deallocate might
773 * sleep. Call it after tmpfs_free_tmp() does the
774 * unlock.
775 */
776 if (uobj != NULL)
777 vm_object_deallocate(uobj);
778
779 break;
780 case VLNK:
781 tmpfs_free_tmp(tmp);
782
783 symlink = node->tn_link_target;
784 atomic_store_ptr(&node->tn_link_target, NULL);
785 if (atomic_load_char(&node->tn_link_smr)) {
786 cache_symlink_free(symlink, node->tn_size + 1);
787 } else {
788 free(symlink, M_TMPFSNAME);
789 }
790 break;
791 default:
792 tmpfs_free_tmp(tmp);
793 break;
794 }
795
796 uma_zfree_smr(tmpfs_node_pool, node);
797 return (true);
798 }
799
800 static __inline uint32_t
801 tmpfs_dirent_hash(const char *name, u_int len)
802 {
803 uint32_t hash;
804
805 hash = fnv_32_buf(name, len, FNV1_32_INIT + len) & TMPFS_DIRCOOKIE_MASK;
806 #ifdef TMPFS_DEBUG_DIRCOOKIE_DUP
807 hash &= 0xf;
808 #endif
809 if (hash < TMPFS_DIRCOOKIE_MIN)
810 hash += TMPFS_DIRCOOKIE_MIN;
811
812 return (hash);
813 }
814
815 static __inline off_t
816 tmpfs_dirent_cookie(struct tmpfs_dirent *de)
817 {
818 if (de == NULL)
819 return (TMPFS_DIRCOOKIE_EOF);
820
821 MPASS(de->td_cookie >= TMPFS_DIRCOOKIE_MIN);
822
823 return (de->td_cookie);
824 }
825
826 static __inline boolean_t
827 tmpfs_dirent_dup(struct tmpfs_dirent *de)
828 {
829 return ((de->td_cookie & TMPFS_DIRCOOKIE_DUP) != 0);
830 }
831
832 static __inline boolean_t
833 tmpfs_dirent_duphead(struct tmpfs_dirent *de)
834 {
835 return ((de->td_cookie & TMPFS_DIRCOOKIE_DUPHEAD) != 0);
836 }
837
838 void
839 tmpfs_dirent_init(struct tmpfs_dirent *de, const char *name, u_int namelen)
840 {
841 de->td_hash = de->td_cookie = tmpfs_dirent_hash(name, namelen);
842 memcpy(de->ud.td_name, name, namelen);
843 de->td_namelen = namelen;
844 }
845
846 /*
847 * Allocates a new directory entry for the node node with a name of name.
848 * The new directory entry is returned in *de.
849 *
850 * The link count of node is increased by one to reflect the new object
851 * referencing it.
852 *
853 * Returns zero on success or an appropriate error code on failure.
854 */
855 int
856 tmpfs_alloc_dirent(struct tmpfs_mount *tmp, struct tmpfs_node *node,
857 const char *name, u_int len, struct tmpfs_dirent **de)
858 {
859 struct tmpfs_dirent *nde;
860
861 nde = malloc(sizeof(*nde), M_TMPFSDIR, M_WAITOK);
862 nde->td_node = node;
863 if (name != NULL) {
864 nde->ud.td_name = malloc(len, M_TMPFSNAME, M_WAITOK);
865 tmpfs_dirent_init(nde, name, len);
866 } else
867 nde->td_namelen = 0;
868 if (node != NULL)
869 node->tn_links++;
870
871 *de = nde;
872
873 return (0);
874 }
875
876 /*
877 * Frees a directory entry. It is the caller's responsibility to destroy
878 * the node referenced by it if needed.
879 *
880 * The link count of node is decreased by one to reflect the removal of an
881 * object that referenced it. This only happens if 'node_exists' is true;
882 * otherwise the function will not access the node referred to by the
883 * directory entry, as it may already have been released from the outside.
884 */
885 void
886 tmpfs_free_dirent(struct tmpfs_mount *tmp, struct tmpfs_dirent *de)
887 {
888 struct tmpfs_node *node;
889
890 node = de->td_node;
891 if (node != NULL) {
892 MPASS(node->tn_links > 0);
893 node->tn_links--;
894 }
895 if (!tmpfs_dirent_duphead(de) && de->ud.td_name != NULL)
896 free(de->ud.td_name, M_TMPFSNAME);
897 free(de, M_TMPFSDIR);
898 }
899
900 void
901 tmpfs_destroy_vobject(struct vnode *vp, vm_object_t obj)
902 {
903 bool want_vrele;
904
905 ASSERT_VOP_ELOCKED(vp, "tmpfs_destroy_vobject");
906 if (vp->v_type != VREG || obj == NULL)
907 return;
908
909 VM_OBJECT_WLOCK(obj);
910 VI_LOCK(vp);
911 /*
912 * May be going through forced unmount.
913 */
914 want_vrele = false;
915 if ((obj->flags & OBJ_TMPFS_VREF) != 0) {
916 vm_object_clear_flag(obj, OBJ_TMPFS_VREF);
917 want_vrele = true;
918 }
919
920 if (vp->v_writecount < 0)
921 vp->v_writecount = 0;
922 VI_UNLOCK(vp);
923 VM_OBJECT_WUNLOCK(obj);
924 if (want_vrele) {
925 vrele(vp);
926 }
927 }
928
929 /*
930 * Allocates a new vnode for the node node or returns a new reference to
931 * an existing one if the node had already a vnode referencing it. The
932 * resulting locked vnode is returned in *vpp.
933 *
934 * Returns zero on success or an appropriate error code on failure.
935 */
936 int
937 tmpfs_alloc_vp(struct mount *mp, struct tmpfs_node *node, int lkflag,
938 struct vnode **vpp)
939 {
940 struct vnode *vp;
941 enum vgetstate vs;
942 struct tmpfs_mount *tm;
943 vm_object_t object;
944 int error;
945
946 error = 0;
947 tm = VFS_TO_TMPFS(mp);
948 TMPFS_NODE_LOCK(node);
949 tmpfs_ref_node(node);
950 loop:
951 TMPFS_NODE_ASSERT_LOCKED(node);
952 if ((vp = node->tn_vnode) != NULL) {
953 MPASS((node->tn_vpstate & TMPFS_VNODE_DOOMED) == 0);
954 if ((node->tn_type == VDIR && node->tn_dir.tn_parent == NULL) ||
955 (VN_IS_DOOMED(vp) &&
956 (lkflag & LK_NOWAIT) != 0)) {
957 TMPFS_NODE_UNLOCK(node);
958 error = ENOENT;
959 vp = NULL;
960 goto out;
961 }
962 if (VN_IS_DOOMED(vp)) {
963 node->tn_vpstate |= TMPFS_VNODE_WRECLAIM;
964 while ((node->tn_vpstate & TMPFS_VNODE_WRECLAIM) != 0) {
965 msleep(&node->tn_vnode, TMPFS_NODE_MTX(node),
966 0, "tmpfsE", 0);
967 }
968 goto loop;
969 }
970 vs = vget_prep(vp);
971 TMPFS_NODE_UNLOCK(node);
972 error = vget_finish(vp, lkflag, vs);
973 if (error == ENOENT) {
974 TMPFS_NODE_LOCK(node);
975 goto loop;
976 }
977 if (error != 0) {
978 vp = NULL;
979 goto out;
980 }
981
982 /*
983 * Make sure the vnode is still there after
984 * getting the interlock to avoid racing a free.
985 */
986 if (node->tn_vnode != vp) {
987 vput(vp);
988 TMPFS_NODE_LOCK(node);
989 goto loop;
990 }
991
992 goto out;
993 }
994
995 if ((node->tn_vpstate & TMPFS_VNODE_DOOMED) ||
996 (node->tn_type == VDIR && node->tn_dir.tn_parent == NULL)) {
997 TMPFS_NODE_UNLOCK(node);
998 error = ENOENT;
999 vp = NULL;
1000 goto out;
1001 }
1002
1003 /*
1004 * otherwise lock the vp list while we call getnewvnode
1005 * since that can block.
1006 */
1007 if (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) {
1008 node->tn_vpstate |= TMPFS_VNODE_WANT;
1009 error = msleep((caddr_t) &node->tn_vpstate,
1010 TMPFS_NODE_MTX(node), 0, "tmpfs_alloc_vp", 0);
1011 if (error != 0)
1012 goto out;
1013 goto loop;
1014 } else
1015 node->tn_vpstate |= TMPFS_VNODE_ALLOCATING;
1016
1017 TMPFS_NODE_UNLOCK(node);
1018
1019 /* Get a new vnode and associate it with our node. */
1020 error = getnewvnode("tmpfs", mp, VFS_TO_TMPFS(mp)->tm_nonc ?
1021 &tmpfs_vnodeop_nonc_entries : &tmpfs_vnodeop_entries, &vp);
1022 if (error != 0)
1023 goto unlock;
1024 MPASS(vp != NULL);
1025
1026 /* lkflag is ignored, the lock is exclusive */
1027 (void) vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1028
1029 vp->v_data = node;
1030 vp->v_type = node->tn_type;
1031
1032 /* Type-specific initialization. */
1033 switch (node->tn_type) {
1034 case VBLK:
1035 /* FALLTHROUGH */
1036 case VCHR:
1037 /* FALLTHROUGH */
1038 case VLNK:
1039 /* FALLTHROUGH */
1040 case VSOCK:
1041 break;
1042 case VFIFO:
1043 vp->v_op = &tmpfs_fifoop_entries;
1044 break;
1045 case VREG:
1046 object = node->tn_reg.tn_aobj;
1047 VM_OBJECT_WLOCK(object);
1048 KASSERT((object->flags & OBJ_TMPFS_VREF) == 0,
1049 ("%s: object %p with OBJ_TMPFS_VREF but without vnode",
1050 __func__, object));
1051 KASSERT(object->un_pager.swp.writemappings == 0,
1052 ("%s: object %p has writemappings",
1053 __func__, object));
1054 VI_LOCK(vp);
1055 KASSERT(vp->v_object == NULL, ("Not NULL v_object in tmpfs"));
1056 vp->v_object = object;
1057 vn_irflag_set_locked(vp, VIRF_PGREAD | VIRF_TEXT_REF);
1058 VI_UNLOCK(vp);
1059 VM_OBJECT_WUNLOCK(object);
1060 break;
1061 case VDIR:
1062 MPASS(node->tn_dir.tn_parent != NULL);
1063 if (node->tn_dir.tn_parent == node)
1064 vp->v_vflag |= VV_ROOT;
1065 break;
1066
1067 default:
1068 panic("tmpfs_alloc_vp: type %p %d", node, (int)node->tn_type);
1069 }
1070 if (vp->v_type != VFIFO)
1071 VN_LOCK_ASHARE(vp);
1072
1073 error = insmntque1(vp, mp);
1074 if (error != 0) {
1075 /* Need to clear v_object for insmntque failure. */
1076 tmpfs_destroy_vobject(vp, vp->v_object);
1077 vp->v_object = NULL;
1078 vp->v_data = NULL;
1079 vp->v_op = &dead_vnodeops;
1080 vgone(vp);
1081 vput(vp);
1082 vp = NULL;
1083 } else {
1084 vn_set_state(vp, VSTATE_CONSTRUCTED);
1085 }
1086
1087 unlock:
1088 TMPFS_NODE_LOCK(node);
1089
1090 MPASS(node->tn_vpstate & TMPFS_VNODE_ALLOCATING);
1091 node->tn_vpstate &= ~TMPFS_VNODE_ALLOCATING;
1092 node->tn_vnode = vp;
1093
1094 if (node->tn_vpstate & TMPFS_VNODE_WANT) {
1095 node->tn_vpstate &= ~TMPFS_VNODE_WANT;
1096 TMPFS_NODE_UNLOCK(node);
1097 wakeup((caddr_t) &node->tn_vpstate);
1098 } else
1099 TMPFS_NODE_UNLOCK(node);
1100
1101 out:
1102 if (error == 0) {
1103 *vpp = vp;
1104
1105 #ifdef INVARIANTS
1106 MPASS(*vpp != NULL && VOP_ISLOCKED(*vpp));
1107 TMPFS_NODE_LOCK(node);
1108 MPASS(*vpp == node->tn_vnode);
1109 TMPFS_NODE_UNLOCK(node);
1110 #endif
1111 }
1112 tmpfs_free_node(tm, node);
1113
1114 return (error);
1115 }
1116
1117 /*
1118 * Destroys the association between the vnode vp and the node it
1119 * references.
1120 */
1121 void
1122 tmpfs_free_vp(struct vnode *vp)
1123 {
1124 struct tmpfs_node *node;
1125
1126 node = VP_TO_TMPFS_NODE(vp);
1127
1128 TMPFS_NODE_ASSERT_LOCKED(node);
1129 node->tn_vnode = NULL;
1130 if ((node->tn_vpstate & TMPFS_VNODE_WRECLAIM) != 0)
1131 wakeup(&node->tn_vnode);
1132 node->tn_vpstate &= ~TMPFS_VNODE_WRECLAIM;
1133 vp->v_data = NULL;
1134 }
1135
1136 /*
1137 * Allocates a new file of type 'type' and adds it to the parent directory
1138 * 'dvp'; this addition is done using the component name given in 'cnp'.
1139 * The ownership of the new file is automatically assigned based on the
1140 * credentials of the caller (through 'cnp'), the group is set based on
1141 * the parent directory and the mode is determined from the 'vap' argument.
1142 * If successful, *vpp holds a vnode to the newly created file and zero
1143 * is returned. Otherwise *vpp is NULL and the function returns an
1144 * appropriate error code.
1145 */
1146 int
1147 tmpfs_alloc_file(struct vnode *dvp, struct vnode **vpp, struct vattr *vap,
1148 struct componentname *cnp, const char *target)
1149 {
1150 int error;
1151 struct tmpfs_dirent *de;
1152 struct tmpfs_mount *tmp;
1153 struct tmpfs_node *dnode;
1154 struct tmpfs_node *node;
1155 struct tmpfs_node *parent;
1156
1157 ASSERT_VOP_ELOCKED(dvp, "tmpfs_alloc_file");
1158
1159 tmp = VFS_TO_TMPFS(dvp->v_mount);
1160 dnode = VP_TO_TMPFS_DIR(dvp);
1161 *vpp = NULL;
1162
1163 /* If the entry we are creating is a directory, we cannot overflow
1164 * the number of links of its parent, because it will get a new
1165 * link. */
1166 if (vap->va_type == VDIR) {
1167 /* Ensure that we do not overflow the maximum number of links
1168 * imposed by the system. */
1169 MPASS(dnode->tn_links <= TMPFS_LINK_MAX);
1170 if (dnode->tn_links == TMPFS_LINK_MAX) {
1171 return (EMLINK);
1172 }
1173
1174 parent = dnode;
1175 MPASS(parent != NULL);
1176 } else
1177 parent = NULL;
1178
1179 /* Allocate a node that represents the new file. */
1180 error = tmpfs_alloc_node(dvp->v_mount, tmp, vap->va_type,
1181 cnp->cn_cred->cr_uid, dnode->tn_gid, vap->va_mode, parent,
1182 target, vap->va_rdev, &node);
1183 if (error != 0)
1184 return (error);
1185
1186 /* Allocate a directory entry that points to the new file. */
1187 error = tmpfs_alloc_dirent(tmp, node, cnp->cn_nameptr, cnp->cn_namelen,
1188 &de);
1189 if (error != 0) {
1190 tmpfs_free_node(tmp, node);
1191 return (error);
1192 }
1193
1194 /* Allocate a vnode for the new file. */
1195 error = tmpfs_alloc_vp(dvp->v_mount, node, LK_EXCLUSIVE, vpp);
1196 if (error != 0) {
1197 tmpfs_free_dirent(tmp, de);
1198 tmpfs_free_node(tmp, node);
1199 return (error);
1200 }
1201
1202 /* Now that all required items are allocated, we can proceed to
1203 * insert the new node into the directory, an operation that
1204 * cannot fail. */
1205 if (cnp->cn_flags & ISWHITEOUT)
1206 tmpfs_dir_whiteout_remove(dvp, cnp);
1207 tmpfs_dir_attach(dvp, de);
1208 return (0);
1209 }
1210
1211 struct tmpfs_dirent *
1212 tmpfs_dir_first(struct tmpfs_node *dnode, struct tmpfs_dir_cursor *dc)
1213 {
1214 struct tmpfs_dirent *de;
1215
1216 de = RB_MIN(tmpfs_dir, &dnode->tn_dir.tn_dirhead);
1217 dc->tdc_tree = de;
1218 if (de != NULL && tmpfs_dirent_duphead(de))
1219 de = LIST_FIRST(&de->ud.td_duphead);
1220 dc->tdc_current = de;
1221
1222 return (dc->tdc_current);
1223 }
1224
1225 struct tmpfs_dirent *
1226 tmpfs_dir_next(struct tmpfs_node *dnode, struct tmpfs_dir_cursor *dc)
1227 {
1228 struct tmpfs_dirent *de;
1229
1230 MPASS(dc->tdc_tree != NULL);
1231 if (tmpfs_dirent_dup(dc->tdc_current)) {
1232 dc->tdc_current = LIST_NEXT(dc->tdc_current, uh.td_dup.entries);
1233 if (dc->tdc_current != NULL)
1234 return (dc->tdc_current);
1235 }
1236 dc->tdc_tree = dc->tdc_current = RB_NEXT(tmpfs_dir,
1237 &dnode->tn_dir.tn_dirhead, dc->tdc_tree);
1238 if ((de = dc->tdc_current) != NULL && tmpfs_dirent_duphead(de)) {
1239 dc->tdc_current = LIST_FIRST(&de->ud.td_duphead);
1240 MPASS(dc->tdc_current != NULL);
1241 }
1242
1243 return (dc->tdc_current);
1244 }
1245
1246 /* Lookup directory entry in RB-Tree. Function may return duphead entry. */
1247 static struct tmpfs_dirent *
1248 tmpfs_dir_xlookup_hash(struct tmpfs_node *dnode, uint32_t hash)
1249 {
1250 struct tmpfs_dirent *de, dekey;
1251
1252 dekey.td_hash = hash;
1253 de = RB_FIND(tmpfs_dir, &dnode->tn_dir.tn_dirhead, &dekey);
1254 return (de);
1255 }
1256
1257 /* Lookup directory entry by cookie, initialize directory cursor accordingly. */
1258 static struct tmpfs_dirent *
1259 tmpfs_dir_lookup_cookie(struct tmpfs_node *node, off_t cookie,
1260 struct tmpfs_dir_cursor *dc)
1261 {
1262 struct tmpfs_dir *dirhead = &node->tn_dir.tn_dirhead;
1263 struct tmpfs_dirent *de, dekey;
1264
1265 MPASS(cookie >= TMPFS_DIRCOOKIE_MIN);
1266
1267 if (cookie == node->tn_dir.tn_readdir_lastn &&
1268 (de = node->tn_dir.tn_readdir_lastp) != NULL) {
1269 /* Protect against possible race, tn_readdir_last[pn]
1270 * may be updated with only shared vnode lock held. */
1271 if (cookie == tmpfs_dirent_cookie(de))
1272 goto out;
1273 }
1274
1275 if ((cookie & TMPFS_DIRCOOKIE_DUP) != 0) {
1276 LIST_FOREACH(de, &node->tn_dir.tn_dupindex,
1277 uh.td_dup.index_entries) {
1278 MPASS(tmpfs_dirent_dup(de));
1279 if (de->td_cookie == cookie)
1280 goto out;
1281 /* dupindex list is sorted. */
1282 if (de->td_cookie < cookie) {
1283 de = NULL;
1284 goto out;
1285 }
1286 }
1287 MPASS(de == NULL);
1288 goto out;
1289 }
1290
1291 if ((cookie & TMPFS_DIRCOOKIE_MASK) != cookie) {
1292 de = NULL;
1293 } else {
1294 dekey.td_hash = cookie;
1295 /* Recover if direntry for cookie was removed */
1296 de = RB_NFIND(tmpfs_dir, dirhead, &dekey);
1297 }
1298 dc->tdc_tree = de;
1299 dc->tdc_current = de;
1300 if (de != NULL && tmpfs_dirent_duphead(de)) {
1301 dc->tdc_current = LIST_FIRST(&de->ud.td_duphead);
1302 MPASS(dc->tdc_current != NULL);
1303 }
1304 return (dc->tdc_current);
1305
1306 out:
1307 dc->tdc_tree = de;
1308 dc->tdc_current = de;
1309 if (de != NULL && tmpfs_dirent_dup(de))
1310 dc->tdc_tree = tmpfs_dir_xlookup_hash(node,
1311 de->td_hash);
1312 return (dc->tdc_current);
1313 }
1314
1315 /*
1316 * Looks for a directory entry in the directory represented by node.
1317 * 'cnp' describes the name of the entry to look for. Note that the .
1318 * and .. components are not allowed as they do not physically exist
1319 * within directories.
1320 *
1321 * Returns a pointer to the entry when found, otherwise NULL.
1322 */
1323 struct tmpfs_dirent *
1324 tmpfs_dir_lookup(struct tmpfs_node *node, struct tmpfs_node *f,
1325 struct componentname *cnp)
1326 {
1327 struct tmpfs_dir_duphead *duphead;
1328 struct tmpfs_dirent *de;
1329 uint32_t hash;
1330
1331 MPASS(IMPLIES(cnp->cn_namelen == 1, cnp->cn_nameptr[0] != '.'));
1332 MPASS(IMPLIES(cnp->cn_namelen == 2, !(cnp->cn_nameptr[0] == '.' &&
1333 cnp->cn_nameptr[1] == '.')));
1334 TMPFS_VALIDATE_DIR(node);
1335
1336 hash = tmpfs_dirent_hash(cnp->cn_nameptr, cnp->cn_namelen);
1337 de = tmpfs_dir_xlookup_hash(node, hash);
1338 if (de != NULL && tmpfs_dirent_duphead(de)) {
1339 duphead = &de->ud.td_duphead;
1340 LIST_FOREACH(de, duphead, uh.td_dup.entries) {
1341 if (TMPFS_DIRENT_MATCHES(de, cnp->cn_nameptr,
1342 cnp->cn_namelen))
1343 break;
1344 }
1345 } else if (de != NULL) {
1346 if (!TMPFS_DIRENT_MATCHES(de, cnp->cn_nameptr,
1347 cnp->cn_namelen))
1348 de = NULL;
1349 }
1350 if (de != NULL && f != NULL && de->td_node != f)
1351 de = NULL;
1352
1353 return (de);
1354 }
1355
1356 /*
1357 * Attach duplicate-cookie directory entry nde to dnode and insert to dupindex
1358 * list, allocate new cookie value.
1359 */
1360 static void
1361 tmpfs_dir_attach_dup(struct tmpfs_node *dnode,
1362 struct tmpfs_dir_duphead *duphead, struct tmpfs_dirent *nde)
1363 {
1364 struct tmpfs_dir_duphead *dupindex;
1365 struct tmpfs_dirent *de, *pde;
1366
1367 dupindex = &dnode->tn_dir.tn_dupindex;
1368 de = LIST_FIRST(dupindex);
1369 if (de == NULL || de->td_cookie < TMPFS_DIRCOOKIE_DUP_MAX) {
1370 if (de == NULL)
1371 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MIN;
1372 else
1373 nde->td_cookie = de->td_cookie + 1;
1374 MPASS(tmpfs_dirent_dup(nde));
1375 LIST_INSERT_HEAD(dupindex, nde, uh.td_dup.index_entries);
1376 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries);
1377 return;
1378 }
1379
1380 /*
1381 * Cookie numbers are near exhaustion. Scan dupindex list for unused
1382 * numbers. dupindex list is sorted in descending order. Keep it so
1383 * after inserting nde.
1384 */
1385 while (1) {
1386 pde = de;
1387 de = LIST_NEXT(de, uh.td_dup.index_entries);
1388 if (de == NULL && pde->td_cookie != TMPFS_DIRCOOKIE_DUP_MIN) {
1389 /*
1390 * Last element of the index doesn't have minimal cookie
1391 * value, use it.
1392 */
1393 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MIN;
1394 LIST_INSERT_AFTER(pde, nde, uh.td_dup.index_entries);
1395 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries);
1396 return;
1397 } else if (de == NULL) {
1398 /*
1399 * We are so lucky have 2^30 hash duplicates in single
1400 * directory :) Return largest possible cookie value.
1401 * It should be fine except possible issues with
1402 * VOP_READDIR restart.
1403 */
1404 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MAX;
1405 LIST_INSERT_HEAD(dupindex, nde,
1406 uh.td_dup.index_entries);
1407 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries);
1408 return;
1409 }
1410 if (de->td_cookie + 1 == pde->td_cookie ||
1411 de->td_cookie >= TMPFS_DIRCOOKIE_DUP_MAX)
1412 continue; /* No hole or invalid cookie. */
1413 nde->td_cookie = de->td_cookie + 1;
1414 MPASS(tmpfs_dirent_dup(nde));
1415 MPASS(pde->td_cookie > nde->td_cookie);
1416 MPASS(nde->td_cookie > de->td_cookie);
1417 LIST_INSERT_BEFORE(de, nde, uh.td_dup.index_entries);
1418 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries);
1419 return;
1420 }
1421 }
1422
1423 /*
1424 * Attaches the directory entry de to the directory represented by vp.
1425 * Note that this does not change the link count of the node pointed by
1426 * the directory entry, as this is done by tmpfs_alloc_dirent.
1427 */
1428 void
1429 tmpfs_dir_attach(struct vnode *vp, struct tmpfs_dirent *de)
1430 {
1431 struct tmpfs_node *dnode;
1432 struct tmpfs_dirent *xde, *nde;
1433
1434 ASSERT_VOP_ELOCKED(vp, __func__);
1435 MPASS(de->td_namelen > 0);
1436 MPASS(de->td_hash >= TMPFS_DIRCOOKIE_MIN);
1437 MPASS(de->td_cookie == de->td_hash);
1438
1439 dnode = VP_TO_TMPFS_DIR(vp);
1440 dnode->tn_dir.tn_readdir_lastn = 0;
1441 dnode->tn_dir.tn_readdir_lastp = NULL;
1442
1443 MPASS(!tmpfs_dirent_dup(de));
1444 xde = RB_INSERT(tmpfs_dir, &dnode->tn_dir.tn_dirhead, de);
1445 if (xde != NULL && tmpfs_dirent_duphead(xde))
1446 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, de);
1447 else if (xde != NULL) {
1448 /*
1449 * Allocate new duphead. Swap xde with duphead to avoid
1450 * adding/removing elements with the same hash.
1451 */
1452 MPASS(!tmpfs_dirent_dup(xde));
1453 tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), NULL, NULL, 0,
1454 &nde);
1455 /* *nde = *xde; XXX gcc 4.2.1 may generate invalid code. */
1456 memcpy(nde, xde, sizeof(*xde));
1457 xde->td_cookie |= TMPFS_DIRCOOKIE_DUPHEAD;
1458 LIST_INIT(&xde->ud.td_duphead);
1459 xde->td_namelen = 0;
1460 xde->td_node = NULL;
1461 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, nde);
1462 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, de);
1463 }
1464 dnode->tn_size += sizeof(struct tmpfs_dirent);
1465 dnode->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1466 dnode->tn_accessed = true;
1467 tmpfs_update(vp);
1468 }
1469
1470 /*
1471 * Detaches the directory entry de from the directory represented by vp.
1472 * Note that this does not change the link count of the node pointed by
1473 * the directory entry, as this is done by tmpfs_free_dirent.
1474 */
1475 void
1476 tmpfs_dir_detach(struct vnode *vp, struct tmpfs_dirent *de)
1477 {
1478 struct tmpfs_mount *tmp;
1479 struct tmpfs_dir *head;
1480 struct tmpfs_node *dnode;
1481 struct tmpfs_dirent *xde;
1482
1483 ASSERT_VOP_ELOCKED(vp, __func__);
1484
1485 dnode = VP_TO_TMPFS_DIR(vp);
1486 head = &dnode->tn_dir.tn_dirhead;
1487 dnode->tn_dir.tn_readdir_lastn = 0;
1488 dnode->tn_dir.tn_readdir_lastp = NULL;
1489
1490 if (tmpfs_dirent_dup(de)) {
1491 /* Remove duphead if de was last entry. */
1492 if (LIST_NEXT(de, uh.td_dup.entries) == NULL) {
1493 xde = tmpfs_dir_xlookup_hash(dnode, de->td_hash);
1494 MPASS(tmpfs_dirent_duphead(xde));
1495 } else
1496 xde = NULL;
1497 LIST_REMOVE(de, uh.td_dup.entries);
1498 LIST_REMOVE(de, uh.td_dup.index_entries);
1499 if (xde != NULL) {
1500 if (LIST_EMPTY(&xde->ud.td_duphead)) {
1501 RB_REMOVE(tmpfs_dir, head, xde);
1502 tmp = VFS_TO_TMPFS(vp->v_mount);
1503 MPASS(xde->td_node == NULL);
1504 tmpfs_free_dirent(tmp, xde);
1505 }
1506 }
1507 de->td_cookie = de->td_hash;
1508 } else
1509 RB_REMOVE(tmpfs_dir, head, de);
1510
1511 dnode->tn_size -= sizeof(struct tmpfs_dirent);
1512 dnode->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1513 dnode->tn_accessed = true;
1514 tmpfs_update(vp);
1515 }
1516
1517 void
1518 tmpfs_dir_destroy(struct tmpfs_mount *tmp, struct tmpfs_node *dnode)
1519 {
1520 struct tmpfs_dirent *de, *dde, *nde;
1521
1522 RB_FOREACH_SAFE(de, tmpfs_dir, &dnode->tn_dir.tn_dirhead, nde) {
1523 RB_REMOVE(tmpfs_dir, &dnode->tn_dir.tn_dirhead, de);
1524 /* Node may already be destroyed. */
1525 de->td_node = NULL;
1526 if (tmpfs_dirent_duphead(de)) {
1527 while ((dde = LIST_FIRST(&de->ud.td_duphead)) != NULL) {
1528 LIST_REMOVE(dde, uh.td_dup.entries);
1529 dde->td_node = NULL;
1530 tmpfs_free_dirent(tmp, dde);
1531 }
1532 }
1533 tmpfs_free_dirent(tmp, de);
1534 }
1535 }
1536
1537 /*
1538 * Helper function for tmpfs_readdir. Creates a '.' entry for the given
1539 * directory and returns it in the uio space. The function returns 0
1540 * on success, -1 if there was not enough space in the uio structure to
1541 * hold the directory entry or an appropriate error code if another
1542 * error happens.
1543 */
1544 static int
1545 tmpfs_dir_getdotdent(struct tmpfs_mount *tm, struct tmpfs_node *node,
1546 struct uio *uio)
1547 {
1548 int error;
1549 struct dirent dent;
1550
1551 TMPFS_VALIDATE_DIR(node);
1552 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOT);
1553
1554 dent.d_fileno = node->tn_id;
1555 dent.d_off = TMPFS_DIRCOOKIE_DOTDOT;
1556 dent.d_type = DT_DIR;
1557 dent.d_namlen = 1;
1558 dent.d_name[0] = '.';
1559 dent.d_reclen = GENERIC_DIRSIZ(&dent);
1560 dirent_terminate(&dent);
1561
1562 if (dent.d_reclen > uio->uio_resid)
1563 error = EJUSTRETURN;
1564 else
1565 error = uiomove(&dent, dent.d_reclen, uio);
1566
1567 tmpfs_set_accessed(tm, node);
1568
1569 return (error);
1570 }
1571
1572 /*
1573 * Helper function for tmpfs_readdir. Creates a '..' entry for the given
1574 * directory and returns it in the uio space. The function returns 0
1575 * on success, -1 if there was not enough space in the uio structure to
1576 * hold the directory entry or an appropriate error code if another
1577 * error happens.
1578 */
1579 static int
1580 tmpfs_dir_getdotdotdent(struct tmpfs_mount *tm, struct tmpfs_node *node,
1581 struct uio *uio, off_t next)
1582 {
1583 struct tmpfs_node *parent;
1584 struct dirent dent;
1585 int error;
1586
1587 TMPFS_VALIDATE_DIR(node);
1588 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT);
1589
1590 /*
1591 * Return ENOENT if the current node is already removed.
1592 */
1593 TMPFS_ASSERT_LOCKED(node);
1594 parent = node->tn_dir.tn_parent;
1595 if (parent == NULL)
1596 return (ENOENT);
1597
1598 dent.d_fileno = parent->tn_id;
1599 dent.d_off = next;
1600 dent.d_type = DT_DIR;
1601 dent.d_namlen = 2;
1602 dent.d_name[0] = '.';
1603 dent.d_name[1] = '.';
1604 dent.d_reclen = GENERIC_DIRSIZ(&dent);
1605 dirent_terminate(&dent);
1606
1607 if (dent.d_reclen > uio->uio_resid)
1608 error = EJUSTRETURN;
1609 else
1610 error = uiomove(&dent, dent.d_reclen, uio);
1611
1612 tmpfs_set_accessed(tm, node);
1613
1614 return (error);
1615 }
1616
1617 /*
1618 * Helper function for tmpfs_readdir. Returns as much directory entries
1619 * as can fit in the uio space. The read starts at uio->uio_offset.
1620 * The function returns 0 on success, -1 if there was not enough space
1621 * in the uio structure to hold the directory entry or an appropriate
1622 * error code if another error happens.
1623 */
1624 int
1625 tmpfs_dir_getdents(struct tmpfs_mount *tm, struct tmpfs_node *node,
1626 struct uio *uio, int maxcookies, uint64_t *cookies, int *ncookies)
1627 {
1628 struct tmpfs_dir_cursor dc;
1629 struct tmpfs_dirent *de, *nde;
1630 off_t off;
1631 int error;
1632
1633 TMPFS_VALIDATE_DIR(node);
1634
1635 off = 0;
1636
1637 /*
1638 * Lookup the node from the current offset. The starting offset of
1639 * 0 will lookup both '.' and '..', and then the first real entry,
1640 * or EOF if there are none. Then find all entries for the dir that
1641 * fit into the buffer. Once no more entries are found (de == NULL),
1642 * the offset is set to TMPFS_DIRCOOKIE_EOF, which will cause the next
1643 * call to return 0.
1644 */
1645 switch (uio->uio_offset) {
1646 case TMPFS_DIRCOOKIE_DOT:
1647 error = tmpfs_dir_getdotdent(tm, node, uio);
1648 if (error != 0)
1649 return (error);
1650 uio->uio_offset = off = TMPFS_DIRCOOKIE_DOTDOT;
1651 if (cookies != NULL)
1652 cookies[(*ncookies)++] = off;
1653 /* FALLTHROUGH */
1654 case TMPFS_DIRCOOKIE_DOTDOT:
1655 de = tmpfs_dir_first(node, &dc);
1656 off = tmpfs_dirent_cookie(de);
1657 error = tmpfs_dir_getdotdotdent(tm, node, uio, off);
1658 if (error != 0)
1659 return (error);
1660 uio->uio_offset = off;
1661 if (cookies != NULL)
1662 cookies[(*ncookies)++] = off;
1663 /* EOF. */
1664 if (de == NULL)
1665 return (0);
1666 break;
1667 case TMPFS_DIRCOOKIE_EOF:
1668 return (0);
1669 default:
1670 de = tmpfs_dir_lookup_cookie(node, uio->uio_offset, &dc);
1671 if (de == NULL)
1672 return (EINVAL);
1673 if (cookies != NULL)
1674 off = tmpfs_dirent_cookie(de);
1675 }
1676
1677 /*
1678 * Read as much entries as possible; i.e., until we reach the end of the
1679 * directory or we exhaust uio space.
1680 */
1681 do {
1682 struct dirent d;
1683
1684 /*
1685 * Create a dirent structure representing the current tmpfs_node
1686 * and fill it.
1687 */
1688 if (de->td_node == NULL) {
1689 d.d_fileno = 1;
1690 d.d_type = DT_WHT;
1691 } else {
1692 d.d_fileno = de->td_node->tn_id;
1693 switch (de->td_node->tn_type) {
1694 case VBLK:
1695 d.d_type = DT_BLK;
1696 break;
1697
1698 case VCHR:
1699 d.d_type = DT_CHR;
1700 break;
1701
1702 case VDIR:
1703 d.d_type = DT_DIR;
1704 break;
1705
1706 case VFIFO:
1707 d.d_type = DT_FIFO;
1708 break;
1709
1710 case VLNK:
1711 d.d_type = DT_LNK;
1712 break;
1713
1714 case VREG:
1715 d.d_type = DT_REG;
1716 break;
1717
1718 case VSOCK:
1719 d.d_type = DT_SOCK;
1720 break;
1721
1722 default:
1723 panic("tmpfs_dir_getdents: type %p %d",
1724 de->td_node, (int)de->td_node->tn_type);
1725 }
1726 }
1727 d.d_namlen = de->td_namelen;
1728 MPASS(de->td_namelen < sizeof(d.d_name));
1729 (void)memcpy(d.d_name, de->ud.td_name, de->td_namelen);
1730 d.d_reclen = GENERIC_DIRSIZ(&d);
1731
1732 /*
1733 * Stop reading if the directory entry we are treating is bigger
1734 * than the amount of data that can be returned.
1735 */
1736 if (d.d_reclen > uio->uio_resid) {
1737 error = EJUSTRETURN;
1738 break;
1739 }
1740
1741 nde = tmpfs_dir_next(node, &dc);
1742 d.d_off = tmpfs_dirent_cookie(nde);
1743 dirent_terminate(&d);
1744
1745 /*
1746 * Copy the new dirent structure into the output buffer and
1747 * advance pointers.
1748 */
1749 error = uiomove(&d, d.d_reclen, uio);
1750 if (error == 0) {
1751 de = nde;
1752 if (cookies != NULL) {
1753 off = tmpfs_dirent_cookie(de);
1754 MPASS(*ncookies < maxcookies);
1755 cookies[(*ncookies)++] = off;
1756 }
1757 }
1758 } while (error == 0 && uio->uio_resid > 0 && de != NULL);
1759
1760 /* Skip setting off when using cookies as it is already done above. */
1761 if (cookies == NULL)
1762 off = tmpfs_dirent_cookie(de);
1763
1764 /* Update the offset and cache. */
1765 uio->uio_offset = off;
1766 node->tn_dir.tn_readdir_lastn = off;
1767 node->tn_dir.tn_readdir_lastp = de;
1768
1769 tmpfs_set_accessed(tm, node);
1770 return (error);
1771 }
1772
1773 int
1774 tmpfs_dir_whiteout_add(struct vnode *dvp, struct componentname *cnp)
1775 {
1776 struct tmpfs_dirent *de;
1777 int error;
1778
1779 error = tmpfs_alloc_dirent(VFS_TO_TMPFS(dvp->v_mount), NULL,
1780 cnp->cn_nameptr, cnp->cn_namelen, &de);
1781 if (error != 0)
1782 return (error);
1783 tmpfs_dir_attach(dvp, de);
1784 return (0);
1785 }
1786
1787 void
1788 tmpfs_dir_whiteout_remove(struct vnode *dvp, struct componentname *cnp)
1789 {
1790 struct tmpfs_dirent *de;
1791
1792 de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(dvp), NULL, cnp);
1793 MPASS(de != NULL && de->td_node == NULL);
1794 tmpfs_dir_detach(dvp, de);
1795 tmpfs_free_dirent(VFS_TO_TMPFS(dvp->v_mount), de);
1796 }
1797
1798 /*
1799 * Resizes the aobj associated with the regular file pointed to by 'vp' to the
1800 * size 'newsize'. 'vp' must point to a vnode that represents a regular file.
1801 * 'newsize' must be positive.
1802 *
1803 * Returns zero on success or an appropriate error code on failure.
1804 */
1805 int
1806 tmpfs_reg_resize(struct vnode *vp, off_t newsize, boolean_t ignerr)
1807 {
1808 struct tmpfs_node *node;
1809 vm_object_t uobj;
1810 vm_pindex_t idx, newpages, oldpages;
1811 off_t oldsize;
1812 int base, error;
1813
1814 MPASS(vp->v_type == VREG);
1815 MPASS(newsize >= 0);
1816
1817 node = VP_TO_TMPFS_NODE(vp);
1818 uobj = node->tn_reg.tn_aobj;
1819
1820 /*
1821 * Convert the old and new sizes to the number of pages needed to
1822 * store them. It may happen that we do not need to do anything
1823 * because the last allocated page can accommodate the change on
1824 * its own.
1825 */
1826 oldsize = node->tn_size;
1827 oldpages = OFF_TO_IDX(oldsize + PAGE_MASK);
1828 MPASS(oldpages == uobj->size);
1829 newpages = OFF_TO_IDX(newsize + PAGE_MASK);
1830
1831 if (__predict_true(newpages == oldpages && newsize >= oldsize)) {
1832 node->tn_size = newsize;
1833 return (0);
1834 }
1835
1836 VM_OBJECT_WLOCK(uobj);
1837 if (newsize < oldsize) {
1838 /*
1839 * Zero the truncated part of the last page.
1840 */
1841 base = newsize & PAGE_MASK;
1842 if (base != 0) {
1843 idx = OFF_TO_IDX(newsize);
1844 error = tmpfs_partial_page_invalidate(uobj, idx, base,
1845 PAGE_SIZE, ignerr);
1846 if (error != 0) {
1847 VM_OBJECT_WUNLOCK(uobj);
1848 return (error);
1849 }
1850 }
1851
1852 /*
1853 * Release any swap space and free any whole pages.
1854 */
1855 if (newpages < oldpages)
1856 vm_object_page_remove(uobj, newpages, 0, 0);
1857 }
1858 uobj->size = newpages;
1859 VM_OBJECT_WUNLOCK(uobj);
1860
1861 node->tn_size = newsize;
1862 return (0);
1863 }
1864
1865 /*
1866 * Punch hole in the aobj associated with the regular file pointed to by 'vp'.
1867 * Requests completely beyond the end-of-file are converted to no-op.
1868 *
1869 * Returns 0 on success or error code from tmpfs_partial_page_invalidate() on
1870 * failure.
1871 */
1872 int
1873 tmpfs_reg_punch_hole(struct vnode *vp, off_t *offset, off_t *length)
1874 {
1875 struct tmpfs_node *node;
1876 vm_object_t object;
1877 vm_pindex_t pistart, pi, piend;
1878 int startofs, endofs, end;
1879 off_t off, len;
1880 int error;
1881
1882 KASSERT(*length <= OFF_MAX - *offset, ("%s: offset + length overflows",
1883 __func__));
1884 node = VP_TO_TMPFS_NODE(vp);
1885 KASSERT(node->tn_type == VREG, ("%s: node is not regular file",
1886 __func__));
1887 object = node->tn_reg.tn_aobj;
1888 off = *offset;
1889 len = omin(node->tn_size - off, *length);
1890 startofs = off & PAGE_MASK;
1891 endofs = (off + len) & PAGE_MASK;
1892 pistart = OFF_TO_IDX(off);
1893 piend = OFF_TO_IDX(off + len);
1894 pi = OFF_TO_IDX((vm_ooffset_t)off + PAGE_MASK);
1895 error = 0;
1896
1897 /* Handle the case when offset is on or beyond file size. */
1898 if (len <= 0) {
1899 *length = 0;
1900 return (0);
1901 }
1902
1903 VM_OBJECT_WLOCK(object);
1904
1905 /*
1906 * If there is a partial page at the beginning of the hole-punching
1907 * request, fill the partial page with zeroes.
1908 */
1909 if (startofs != 0) {
1910 end = pistart != piend ? PAGE_SIZE : endofs;
1911 error = tmpfs_partial_page_invalidate(object, pistart, startofs,
1912 end, FALSE);
1913 if (error != 0)
1914 goto out;
1915 off += end - startofs;
1916 len -= end - startofs;
1917 }
1918
1919 /*
1920 * Toss away the full pages in the affected area.
1921 */
1922 if (pi < piend) {
1923 vm_object_page_remove(object, pi, piend, 0);
1924 off += IDX_TO_OFF(piend - pi);
1925 len -= IDX_TO_OFF(piend - pi);
1926 }
1927
1928 /*
1929 * If there is a partial page at the end of the hole-punching request,
1930 * fill the partial page with zeroes.
1931 */
1932 if (endofs != 0 && pistart != piend) {
1933 error = tmpfs_partial_page_invalidate(object, piend, 0, endofs,
1934 FALSE);
1935 if (error != 0)
1936 goto out;
1937 off += endofs;
1938 len -= endofs;
1939 }
1940
1941 out:
1942 VM_OBJECT_WUNLOCK(object);
1943 *offset = off;
1944 *length = len;
1945 return (error);
1946 }
1947
1948 void
1949 tmpfs_check_mtime(struct vnode *vp)
1950 {
1951 struct tmpfs_node *node;
1952 struct vm_object *obj;
1953
1954 ASSERT_VOP_ELOCKED(vp, "check_mtime");
1955 if (vp->v_type != VREG)
1956 return;
1957 obj = vp->v_object;
1958 KASSERT(obj->type == tmpfs_pager_type &&
1959 (obj->flags & (OBJ_SWAP | OBJ_TMPFS)) ==
1960 (OBJ_SWAP | OBJ_TMPFS), ("non-tmpfs obj"));
1961 /* unlocked read */
1962 if (obj->generation != obj->cleangeneration) {
1963 VM_OBJECT_WLOCK(obj);
1964 if (obj->generation != obj->cleangeneration) {
1965 obj->cleangeneration = obj->generation;
1966 node = VP_TO_TMPFS_NODE(vp);
1967 node->tn_status |= TMPFS_NODE_MODIFIED |
1968 TMPFS_NODE_CHANGED;
1969 }
1970 VM_OBJECT_WUNLOCK(obj);
1971 }
1972 }
1973
1974 /*
1975 * Change flags of the given vnode.
1976 * Caller should execute tmpfs_update on vp after a successful execution.
1977 * The vnode must be locked on entry and remain locked on exit.
1978 */
1979 int
1980 tmpfs_chflags(struct vnode *vp, u_long flags, struct ucred *cred,
1981 struct thread *td)
1982 {
1983 int error;
1984 struct tmpfs_node *node;
1985
1986 ASSERT_VOP_ELOCKED(vp, "chflags");
1987
1988 node = VP_TO_TMPFS_NODE(vp);
1989
1990 if ((flags & ~(SF_APPEND | SF_ARCHIVED | SF_IMMUTABLE | SF_NOUNLINK |
1991 UF_APPEND | UF_ARCHIVE | UF_HIDDEN | UF_IMMUTABLE | UF_NODUMP |
1992 UF_NOUNLINK | UF_OFFLINE | UF_OPAQUE | UF_READONLY | UF_REPARSE |
1993 UF_SPARSE | UF_SYSTEM)) != 0)
1994 return (EOPNOTSUPP);
1995
1996 /* Disallow this operation if the file system is mounted read-only. */
1997 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1998 return (EROFS);
1999
2000 /*
2001 * Callers may only modify the file flags on objects they
2002 * have VADMIN rights for.
2003 */
2004 if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
2005 return (error);
2006 /*
2007 * Unprivileged processes are not permitted to unset system
2008 * flags, or modify flags if any system flags are set.
2009 */
2010 if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS)) {
2011 if (node->tn_flags &
2012 (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND)) {
2013 error = securelevel_gt(cred, 0);
2014 if (error)
2015 return (error);
2016 }
2017 } else {
2018 if (node->tn_flags &
2019 (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND) ||
2020 ((flags ^ node->tn_flags) & SF_SETTABLE))
2021 return (EPERM);
2022 }
2023 node->tn_flags = flags;
2024 node->tn_status |= TMPFS_NODE_CHANGED;
2025
2026 ASSERT_VOP_ELOCKED(vp, "chflags2");
2027
2028 return (0);
2029 }
2030
2031 /*
2032 * Change access mode on the given vnode.
2033 * Caller should execute tmpfs_update on vp after a successful execution.
2034 * The vnode must be locked on entry and remain locked on exit.
2035 */
2036 int
2037 tmpfs_chmod(struct vnode *vp, mode_t mode, struct ucred *cred,
2038 struct thread *td)
2039 {
2040 int error;
2041 struct tmpfs_node *node;
2042 mode_t newmode;
2043
2044 ASSERT_VOP_ELOCKED(vp, "chmod");
2045 ASSERT_VOP_IN_SEQC(vp);
2046
2047 node = VP_TO_TMPFS_NODE(vp);
2048
2049 /* Disallow this operation if the file system is mounted read-only. */
2050 if (vp->v_mount->mnt_flag & MNT_RDONLY)
2051 return (EROFS);
2052
2053 /* Immutable or append-only files cannot be modified, either. */
2054 if (node->tn_flags & (IMMUTABLE | APPEND))
2055 return (EPERM);
2056
2057 /*
2058 * To modify the permissions on a file, must possess VADMIN
2059 * for that file.
2060 */
2061 if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
2062 return (error);
2063
2064 /*
2065 * Privileged processes may set the sticky bit on non-directories,
2066 * as well as set the setgid bit on a file with a group that the
2067 * process is not a member of.
2068 */
2069 if (vp->v_type != VDIR && (mode & S_ISTXT)) {
2070 if (priv_check_cred(cred, PRIV_VFS_STICKYFILE))
2071 return (EFTYPE);
2072 }
2073 if (!groupmember(node->tn_gid, cred) && (mode & S_ISGID)) {
2074 error = priv_check_cred(cred, PRIV_VFS_SETGID);
2075 if (error)
2076 return (error);
2077 }
2078
2079 newmode = node->tn_mode & ~ALLPERMS;
2080 newmode |= mode & ALLPERMS;
2081 atomic_store_short(&node->tn_mode, newmode);
2082
2083 node->tn_status |= TMPFS_NODE_CHANGED;
2084
2085 ASSERT_VOP_ELOCKED(vp, "chmod2");
2086
2087 return (0);
2088 }
2089
2090 /*
2091 * Change ownership of the given vnode. At least one of uid or gid must
2092 * be different than VNOVAL. If one is set to that value, the attribute
2093 * is unchanged.
2094 * Caller should execute tmpfs_update on vp after a successful execution.
2095 * The vnode must be locked on entry and remain locked on exit.
2096 */
2097 int
2098 tmpfs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred,
2099 struct thread *td)
2100 {
2101 int error;
2102 struct tmpfs_node *node;
2103 uid_t ouid;
2104 gid_t ogid;
2105 mode_t newmode;
2106
2107 ASSERT_VOP_ELOCKED(vp, "chown");
2108 ASSERT_VOP_IN_SEQC(vp);
2109
2110 node = VP_TO_TMPFS_NODE(vp);
2111
2112 /* Assign default values if they are unknown. */
2113 MPASS(uid != VNOVAL || gid != VNOVAL);
2114 if (uid == VNOVAL)
2115 uid = node->tn_uid;
2116 if (gid == VNOVAL)
2117 gid = node->tn_gid;
2118 MPASS(uid != VNOVAL && gid != VNOVAL);
2119
2120 /* Disallow this operation if the file system is mounted read-only. */
2121 if (vp->v_mount->mnt_flag & MNT_RDONLY)
2122 return (EROFS);
2123
2124 /* Immutable or append-only files cannot be modified, either. */
2125 if (node->tn_flags & (IMMUTABLE | APPEND))
2126 return (EPERM);
2127
2128 /*
2129 * To modify the ownership of a file, must possess VADMIN for that
2130 * file.
2131 */
2132 if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
2133 return (error);
2134
2135 /*
2136 * To change the owner of a file, or change the group of a file to a
2137 * group of which we are not a member, the caller must have
2138 * privilege.
2139 */
2140 if ((uid != node->tn_uid ||
2141 (gid != node->tn_gid && !groupmember(gid, cred))) &&
2142 (error = priv_check_cred(cred, PRIV_VFS_CHOWN)))
2143 return (error);
2144
2145 ogid = node->tn_gid;
2146 ouid = node->tn_uid;
2147
2148 node->tn_uid = uid;
2149 node->tn_gid = gid;
2150
2151 node->tn_status |= TMPFS_NODE_CHANGED;
2152
2153 if ((node->tn_mode & (S_ISUID | S_ISGID)) != 0 &&
2154 (ouid != uid || ogid != gid)) {
2155 if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID)) {
2156 newmode = node->tn_mode & ~(S_ISUID | S_ISGID);
2157 atomic_store_short(&node->tn_mode, newmode);
2158 }
2159 }
2160
2161 ASSERT_VOP_ELOCKED(vp, "chown2");
2162
2163 return (0);
2164 }
2165
2166 /*
2167 * Change size of the given vnode.
2168 * Caller should execute tmpfs_update on vp after a successful execution.
2169 * The vnode must be locked on entry and remain locked on exit.
2170 */
2171 int
2172 tmpfs_chsize(struct vnode *vp, u_quad_t size, struct ucred *cred,
2173 struct thread *td)
2174 {
2175 int error;
2176 struct tmpfs_node *node;
2177
2178 ASSERT_VOP_ELOCKED(vp, "chsize");
2179
2180 node = VP_TO_TMPFS_NODE(vp);
2181
2182 /* Decide whether this is a valid operation based on the file type. */
2183 error = 0;
2184 switch (vp->v_type) {
2185 case VDIR:
2186 return (EISDIR);
2187
2188 case VREG:
2189 if (vp->v_mount->mnt_flag & MNT_RDONLY)
2190 return (EROFS);
2191 break;
2192
2193 case VBLK:
2194 /* FALLTHROUGH */
2195 case VCHR:
2196 /* FALLTHROUGH */
2197 case VFIFO:
2198 /*
2199 * Allow modifications of special files even if in the file
2200 * system is mounted read-only (we are not modifying the
2201 * files themselves, but the objects they represent).
2202 */
2203 return (0);
2204
2205 default:
2206 /* Anything else is unsupported. */
2207 return (EOPNOTSUPP);
2208 }
2209
2210 /* Immutable or append-only files cannot be modified, either. */
2211 if (node->tn_flags & (IMMUTABLE | APPEND))
2212 return (EPERM);
2213
2214 error = vn_rlimit_trunc(size, td);
2215 if (error != 0)
2216 return (error);
2217
2218 error = tmpfs_truncate(vp, size);
2219 /*
2220 * tmpfs_truncate will raise the NOTE_EXTEND and NOTE_ATTRIB kevents
2221 * for us, as will update tn_status; no need to do that here.
2222 */
2223
2224 ASSERT_VOP_ELOCKED(vp, "chsize2");
2225
2226 return (error);
2227 }
2228
2229 /*
2230 * Change access and modification times of the given vnode.
2231 * Caller should execute tmpfs_update on vp after a successful execution.
2232 * The vnode must be locked on entry and remain locked on exit.
2233 */
2234 int
2235 tmpfs_chtimes(struct vnode *vp, struct vattr *vap,
2236 struct ucred *cred, struct thread *td)
2237 {
2238 int error;
2239 struct tmpfs_node *node;
2240
2241 ASSERT_VOP_ELOCKED(vp, "chtimes");
2242
2243 node = VP_TO_TMPFS_NODE(vp);
2244
2245 /* Disallow this operation if the file system is mounted read-only. */
2246 if (vp->v_mount->mnt_flag & MNT_RDONLY)
2247 return (EROFS);
2248
2249 /* Immutable or append-only files cannot be modified, either. */
2250 if (node->tn_flags & (IMMUTABLE | APPEND))
2251 return (EPERM);
2252
2253 error = vn_utimes_perm(vp, vap, cred, td);
2254 if (error != 0)
2255 return (error);
2256
2257 if (vap->va_atime.tv_sec != VNOVAL)
2258 node->tn_accessed = true;
2259 if (vap->va_mtime.tv_sec != VNOVAL)
2260 node->tn_status |= TMPFS_NODE_MODIFIED;
2261 if (vap->va_birthtime.tv_sec != VNOVAL)
2262 node->tn_status |= TMPFS_NODE_MODIFIED;
2263 tmpfs_itimes(vp, &vap->va_atime, &vap->va_mtime);
2264 if (vap->va_birthtime.tv_sec != VNOVAL)
2265 node->tn_birthtime = vap->va_birthtime;
2266 ASSERT_VOP_ELOCKED(vp, "chtimes2");
2267
2268 return (0);
2269 }
2270
2271 void
2272 tmpfs_set_status(struct tmpfs_mount *tm, struct tmpfs_node *node, int status)
2273 {
2274
2275 if ((node->tn_status & status) == status || tm->tm_ronly)
2276 return;
2277 TMPFS_NODE_LOCK(node);
2278 node->tn_status |= status;
2279 TMPFS_NODE_UNLOCK(node);
2280 }
2281
2282 void
2283 tmpfs_set_accessed(struct tmpfs_mount *tm, struct tmpfs_node *node)
2284 {
2285 if (node->tn_accessed || tm->tm_ronly)
2286 return;
2287 atomic_store_8(&node->tn_accessed, true);
2288 }
2289
2290 /* Sync timestamps */
2291 void
2292 tmpfs_itimes(struct vnode *vp, const struct timespec *acc,
2293 const struct timespec *mod)
2294 {
2295 struct tmpfs_node *node;
2296 struct timespec now;
2297
2298 ASSERT_VOP_LOCKED(vp, "tmpfs_itimes");
2299 node = VP_TO_TMPFS_NODE(vp);
2300
2301 if (!node->tn_accessed &&
2302 (node->tn_status & (TMPFS_NODE_MODIFIED | TMPFS_NODE_CHANGED)) == 0)
2303 return;
2304
2305 vfs_timestamp(&now);
2306 TMPFS_NODE_LOCK(node);
2307 if (node->tn_accessed) {
2308 if (acc == NULL)
2309 acc = &now;
2310 node->tn_atime = *acc;
2311 }
2312 if (node->tn_status & TMPFS_NODE_MODIFIED) {
2313 if (mod == NULL)
2314 mod = &now;
2315 node->tn_mtime = *mod;
2316 }
2317 if (node->tn_status & TMPFS_NODE_CHANGED)
2318 node->tn_ctime = now;
2319 node->tn_status &= ~(TMPFS_NODE_MODIFIED | TMPFS_NODE_CHANGED);
2320 node->tn_accessed = false;
2321 TMPFS_NODE_UNLOCK(node);
2322
2323 /* XXX: FIX? The entropy here is desirable, but the harvesting may be expensive */
2324 random_harvest_queue(node, sizeof(*node), RANDOM_FS_ATIME);
2325 }
2326
2327 int
2328 tmpfs_truncate(struct vnode *vp, off_t length)
2329 {
2330 struct tmpfs_node *node;
2331 int error;
2332
2333 if (length < 0)
2334 return (EINVAL);
2335 if (length > VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize)
2336 return (EFBIG);
2337
2338 node = VP_TO_TMPFS_NODE(vp);
2339 error = node->tn_size == length ? 0 : tmpfs_reg_resize(vp, length,
2340 FALSE);
2341 if (error == 0)
2342 node->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
2343 tmpfs_update(vp);
2344
2345 return (error);
2346 }
2347
2348 static __inline int
2349 tmpfs_dirtree_cmp(struct tmpfs_dirent *a, struct tmpfs_dirent *b)
2350 {
2351 if (a->td_hash > b->td_hash)
2352 return (1);
2353 else if (a->td_hash < b->td_hash)
2354 return (-1);
2355 return (0);
2356 }
2357
2358 RB_GENERATE_STATIC(tmpfs_dir, tmpfs_dirent, uh.td_entries, tmpfs_dirtree_cmp);
Cache object: 555a36e1e8ba19b901ccc923a0abc688
|