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