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: releng/12.0/sys/fs/tmpfs/tmpfs_subr.c 341085 2018-11-27 17:58:25Z markj $");
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/rwlock.h>
53 #include <sys/stat.h>
54 #include <sys/sysctl.h>
55 #include <sys/vnode.h>
56 #include <sys/vmmeter.h>
57
58 #include <vm/vm.h>
59 #include <vm/vm_param.h>
60 #include <vm/vm_object.h>
61 #include <vm/vm_page.h>
62 #include <vm/vm_pageout.h>
63 #include <vm/vm_pager.h>
64 #include <vm/vm_extern.h>
65 #include <vm/swap_pager.h>
66
67 #include <fs/tmpfs/tmpfs.h>
68 #include <fs/tmpfs/tmpfs_fifoops.h>
69 #include <fs/tmpfs/tmpfs_vnops.h>
70
71 SYSCTL_NODE(_vfs, OID_AUTO, tmpfs, CTLFLAG_RW, 0, "tmpfs file system");
72
73 static long tmpfs_pages_reserved = TMPFS_PAGES_MINRESERVED;
74
75 static int
76 sysctl_mem_reserved(SYSCTL_HANDLER_ARGS)
77 {
78 int error;
79 long pages, bytes;
80
81 pages = *(long *)arg1;
82 bytes = pages * PAGE_SIZE;
83
84 error = sysctl_handle_long(oidp, &bytes, 0, req);
85 if (error || !req->newptr)
86 return (error);
87
88 pages = bytes / PAGE_SIZE;
89 if (pages < TMPFS_PAGES_MINRESERVED)
90 return (EINVAL);
91
92 *(long *)arg1 = pages;
93 return (0);
94 }
95
96 SYSCTL_PROC(_vfs_tmpfs, OID_AUTO, memory_reserved, CTLTYPE_LONG|CTLFLAG_RW,
97 &tmpfs_pages_reserved, 0, sysctl_mem_reserved, "L",
98 "Amount of available memory and swap below which tmpfs growth stops");
99
100 static __inline int tmpfs_dirtree_cmp(struct tmpfs_dirent *a,
101 struct tmpfs_dirent *b);
102 RB_PROTOTYPE_STATIC(tmpfs_dir, tmpfs_dirent, uh.td_entries, tmpfs_dirtree_cmp);
103
104 size_t
105 tmpfs_mem_avail(void)
106 {
107 vm_ooffset_t avail;
108
109 avail = swap_pager_avail + vm_free_count() - tmpfs_pages_reserved;
110 if (__predict_false(avail < 0))
111 avail = 0;
112 return (avail);
113 }
114
115 size_t
116 tmpfs_pages_used(struct tmpfs_mount *tmp)
117 {
118 const size_t node_size = sizeof(struct tmpfs_node) +
119 sizeof(struct tmpfs_dirent);
120 size_t meta_pages;
121
122 meta_pages = howmany((uintmax_t)tmp->tm_nodes_inuse * node_size,
123 PAGE_SIZE);
124 return (meta_pages + tmp->tm_pages_used);
125 }
126
127 static size_t
128 tmpfs_pages_check_avail(struct tmpfs_mount *tmp, size_t req_pages)
129 {
130 if (tmpfs_mem_avail() < req_pages)
131 return (0);
132
133 if (tmp->tm_pages_max != ULONG_MAX &&
134 tmp->tm_pages_max < req_pages + tmpfs_pages_used(tmp))
135 return (0);
136
137 return (1);
138 }
139
140 void
141 tmpfs_ref_node(struct tmpfs_node *node)
142 {
143
144 TMPFS_NODE_LOCK(node);
145 tmpfs_ref_node_locked(node);
146 TMPFS_NODE_UNLOCK(node);
147 }
148
149 void
150 tmpfs_ref_node_locked(struct tmpfs_node *node)
151 {
152
153 TMPFS_NODE_ASSERT_LOCKED(node);
154 KASSERT(node->tn_refcount > 0, ("node %p zero refcount", node));
155 KASSERT(node->tn_refcount < UINT_MAX, ("node %p refcount %u", node,
156 node->tn_refcount));
157 node->tn_refcount++;
158 }
159
160 /*
161 * Allocates a new node of type 'type' inside the 'tmp' mount point, with
162 * its owner set to 'uid', its group to 'gid' and its mode set to 'mode',
163 * using the credentials of the process 'p'.
164 *
165 * If the node type is set to 'VDIR', then the parent parameter must point
166 * to the parent directory of the node being created. It may only be NULL
167 * while allocating the root node.
168 *
169 * If the node type is set to 'VBLK' or 'VCHR', then the rdev parameter
170 * specifies the device the node represents.
171 *
172 * If the node type is set to 'VLNK', then the parameter target specifies
173 * the file name of the target file for the symbolic link that is being
174 * created.
175 *
176 * Note that new nodes are retrieved from the available list if it has
177 * items or, if it is empty, from the node pool as long as there is enough
178 * space to create them.
179 *
180 * Returns zero on success or an appropriate error code on failure.
181 */
182 int
183 tmpfs_alloc_node(struct mount *mp, struct tmpfs_mount *tmp, enum vtype type,
184 uid_t uid, gid_t gid, mode_t mode, struct tmpfs_node *parent,
185 char *target, dev_t rdev, struct tmpfs_node **node)
186 {
187 struct tmpfs_node *nnode;
188 vm_object_t obj;
189
190 /* If the root directory of the 'tmp' file system is not yet
191 * allocated, this must be the request to do it. */
192 MPASS(IMPLIES(tmp->tm_root == NULL, parent == NULL && type == VDIR));
193 KASSERT(tmp->tm_root == NULL || mp->mnt_writeopcount > 0,
194 ("creating node not under vn_start_write"));
195
196 MPASS(IFF(type == VLNK, target != NULL));
197 MPASS(IFF(type == VBLK || type == VCHR, rdev != VNOVAL));
198
199 if (tmp->tm_nodes_inuse >= tmp->tm_nodes_max)
200 return (ENOSPC);
201 if (tmpfs_pages_check_avail(tmp, 1) == 0)
202 return (ENOSPC);
203
204 if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) {
205 /*
206 * When a new tmpfs node is created for fully
207 * constructed mount point, there must be a parent
208 * node, which vnode is locked exclusively. As
209 * consequence, if the unmount is executing in
210 * parallel, vflush() cannot reclaim the parent vnode.
211 * Due to this, the check for MNTK_UNMOUNT flag is not
212 * racy: if we did not see MNTK_UNMOUNT flag, then tmp
213 * cannot be destroyed until node construction is
214 * finished and the parent vnode unlocked.
215 *
216 * Tmpfs does not need to instantiate new nodes during
217 * unmount.
218 */
219 return (EBUSY);
220 }
221
222 nnode = (struct tmpfs_node *)uma_zalloc_arg(tmp->tm_node_pool, tmp,
223 M_WAITOK);
224
225 /* Generic initialization. */
226 nnode->tn_type = type;
227 vfs_timestamp(&nnode->tn_atime);
228 nnode->tn_birthtime = nnode->tn_ctime = nnode->tn_mtime =
229 nnode->tn_atime;
230 nnode->tn_uid = uid;
231 nnode->tn_gid = gid;
232 nnode->tn_mode = mode;
233 nnode->tn_id = alloc_unr(tmp->tm_ino_unr);
234 nnode->tn_refcount = 1;
235
236 /* Type-specific initialization. */
237 switch (nnode->tn_type) {
238 case VBLK:
239 case VCHR:
240 nnode->tn_rdev = rdev;
241 break;
242
243 case VDIR:
244 RB_INIT(&nnode->tn_dir.tn_dirhead);
245 LIST_INIT(&nnode->tn_dir.tn_dupindex);
246 MPASS(parent != nnode);
247 MPASS(IMPLIES(parent == NULL, tmp->tm_root == NULL));
248 nnode->tn_dir.tn_parent = (parent == NULL) ? nnode : parent;
249 nnode->tn_dir.tn_readdir_lastn = 0;
250 nnode->tn_dir.tn_readdir_lastp = NULL;
251 nnode->tn_links++;
252 TMPFS_NODE_LOCK(nnode->tn_dir.tn_parent);
253 nnode->tn_dir.tn_parent->tn_links++;
254 TMPFS_NODE_UNLOCK(nnode->tn_dir.tn_parent);
255 break;
256
257 case VFIFO:
258 /* FALLTHROUGH */
259 case VSOCK:
260 break;
261
262 case VLNK:
263 MPASS(strlen(target) < MAXPATHLEN);
264 nnode->tn_size = strlen(target);
265 nnode->tn_link = malloc(nnode->tn_size, M_TMPFSNAME,
266 M_WAITOK);
267 memcpy(nnode->tn_link, target, nnode->tn_size);
268 break;
269
270 case VREG:
271 obj = nnode->tn_reg.tn_aobj =
272 vm_pager_allocate(OBJT_SWAP, NULL, 0, VM_PROT_DEFAULT, 0,
273 NULL /* XXXKIB - tmpfs needs swap reservation */);
274 VM_OBJECT_WLOCK(obj);
275 /* OBJ_TMPFS is set together with the setting of vp->v_object */
276 vm_object_set_flag(obj, OBJ_NOSPLIT | OBJ_TMPFS_NODE);
277 vm_object_clear_flag(obj, OBJ_ONEMAPPING);
278 VM_OBJECT_WUNLOCK(obj);
279 break;
280
281 default:
282 panic("tmpfs_alloc_node: type %p %d", nnode,
283 (int)nnode->tn_type);
284 }
285
286 TMPFS_LOCK(tmp);
287 LIST_INSERT_HEAD(&tmp->tm_nodes_used, nnode, tn_entries);
288 nnode->tn_attached = true;
289 tmp->tm_nodes_inuse++;
290 tmp->tm_refcount++;
291 TMPFS_UNLOCK(tmp);
292
293 *node = nnode;
294 return (0);
295 }
296
297 /*
298 * Destroys the node pointed to by node from the file system 'tmp'.
299 * If the node references a directory, no entries are allowed.
300 */
301 void
302 tmpfs_free_node(struct tmpfs_mount *tmp, struct tmpfs_node *node)
303 {
304
305 TMPFS_LOCK(tmp);
306 TMPFS_NODE_LOCK(node);
307 if (!tmpfs_free_node_locked(tmp, node, false)) {
308 TMPFS_NODE_UNLOCK(node);
309 TMPFS_UNLOCK(tmp);
310 }
311 }
312
313 bool
314 tmpfs_free_node_locked(struct tmpfs_mount *tmp, struct tmpfs_node *node,
315 bool detach)
316 {
317 vm_object_t uobj;
318
319 TMPFS_MP_ASSERT_LOCKED(tmp);
320 TMPFS_NODE_ASSERT_LOCKED(node);
321 KASSERT(node->tn_refcount > 0, ("node %p refcount zero", node));
322
323 node->tn_refcount--;
324 if (node->tn_attached && (detach || node->tn_refcount == 0)) {
325 MPASS(tmp->tm_nodes_inuse > 0);
326 tmp->tm_nodes_inuse--;
327 LIST_REMOVE(node, tn_entries);
328 node->tn_attached = false;
329 }
330 if (node->tn_refcount > 0)
331 return (false);
332
333 #ifdef INVARIANTS
334 MPASS(node->tn_vnode == NULL);
335 MPASS((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0);
336 #endif
337 TMPFS_NODE_UNLOCK(node);
338 TMPFS_UNLOCK(tmp);
339
340 switch (node->tn_type) {
341 case VBLK:
342 /* FALLTHROUGH */
343 case VCHR:
344 /* FALLTHROUGH */
345 case VDIR:
346 /* FALLTHROUGH */
347 case VFIFO:
348 /* FALLTHROUGH */
349 case VSOCK:
350 break;
351
352 case VLNK:
353 free(node->tn_link, M_TMPFSNAME);
354 break;
355
356 case VREG:
357 uobj = node->tn_reg.tn_aobj;
358 if (uobj != NULL) {
359 if (uobj->size != 0)
360 atomic_subtract_long(&tmp->tm_pages_used, uobj->size);
361 KASSERT((uobj->flags & OBJ_TMPFS) == 0,
362 ("leaked OBJ_TMPFS node %p vm_obj %p", node, uobj));
363 vm_object_deallocate(uobj);
364 }
365 break;
366
367 default:
368 panic("tmpfs_free_node: type %p %d", node, (int)node->tn_type);
369 }
370
371 /*
372 * If we are unmounting there is no need for going through the overhead
373 * of freeing the inodes from the unr individually, so free them all in
374 * one go later.
375 */
376 if (!detach)
377 free_unr(tmp->tm_ino_unr, node->tn_id);
378 uma_zfree(tmp->tm_node_pool, node);
379 TMPFS_LOCK(tmp);
380 tmpfs_free_tmp(tmp);
381 return (true);
382 }
383
384 static __inline uint32_t
385 tmpfs_dirent_hash(const char *name, u_int len)
386 {
387 uint32_t hash;
388
389 hash = fnv_32_buf(name, len, FNV1_32_INIT + len) & TMPFS_DIRCOOKIE_MASK;
390 #ifdef TMPFS_DEBUG_DIRCOOKIE_DUP
391 hash &= 0xf;
392 #endif
393 if (hash < TMPFS_DIRCOOKIE_MIN)
394 hash += TMPFS_DIRCOOKIE_MIN;
395
396 return (hash);
397 }
398
399 static __inline off_t
400 tmpfs_dirent_cookie(struct tmpfs_dirent *de)
401 {
402 if (de == NULL)
403 return (TMPFS_DIRCOOKIE_EOF);
404
405 MPASS(de->td_cookie >= TMPFS_DIRCOOKIE_MIN);
406
407 return (de->td_cookie);
408 }
409
410 static __inline boolean_t
411 tmpfs_dirent_dup(struct tmpfs_dirent *de)
412 {
413 return ((de->td_cookie & TMPFS_DIRCOOKIE_DUP) != 0);
414 }
415
416 static __inline boolean_t
417 tmpfs_dirent_duphead(struct tmpfs_dirent *de)
418 {
419 return ((de->td_cookie & TMPFS_DIRCOOKIE_DUPHEAD) != 0);
420 }
421
422 void
423 tmpfs_dirent_init(struct tmpfs_dirent *de, const char *name, u_int namelen)
424 {
425 de->td_hash = de->td_cookie = tmpfs_dirent_hash(name, namelen);
426 memcpy(de->ud.td_name, name, namelen);
427 de->td_namelen = namelen;
428 }
429
430 /*
431 * Allocates a new directory entry for the node node with a name of name.
432 * The new directory entry is returned in *de.
433 *
434 * The link count of node is increased by one to reflect the new object
435 * referencing it.
436 *
437 * Returns zero on success or an appropriate error code on failure.
438 */
439 int
440 tmpfs_alloc_dirent(struct tmpfs_mount *tmp, struct tmpfs_node *node,
441 const char *name, u_int len, struct tmpfs_dirent **de)
442 {
443 struct tmpfs_dirent *nde;
444
445 nde = uma_zalloc(tmp->tm_dirent_pool, M_WAITOK);
446 nde->td_node = node;
447 if (name != NULL) {
448 nde->ud.td_name = malloc(len, M_TMPFSNAME, M_WAITOK);
449 tmpfs_dirent_init(nde, name, len);
450 } else
451 nde->td_namelen = 0;
452 if (node != NULL)
453 node->tn_links++;
454
455 *de = nde;
456
457 return 0;
458 }
459
460 /*
461 * Frees a directory entry. It is the caller's responsibility to destroy
462 * the node referenced by it if needed.
463 *
464 * The link count of node is decreased by one to reflect the removal of an
465 * object that referenced it. This only happens if 'node_exists' is true;
466 * otherwise the function will not access the node referred to by the
467 * directory entry, as it may already have been released from the outside.
468 */
469 void
470 tmpfs_free_dirent(struct tmpfs_mount *tmp, struct tmpfs_dirent *de)
471 {
472 struct tmpfs_node *node;
473
474 node = de->td_node;
475 if (node != NULL) {
476 MPASS(node->tn_links > 0);
477 node->tn_links--;
478 }
479 if (!tmpfs_dirent_duphead(de) && de->ud.td_name != NULL)
480 free(de->ud.td_name, M_TMPFSNAME);
481 uma_zfree(tmp->tm_dirent_pool, de);
482 }
483
484 void
485 tmpfs_destroy_vobject(struct vnode *vp, vm_object_t obj)
486 {
487
488 ASSERT_VOP_ELOCKED(vp, "tmpfs_destroy_vobject");
489 if (vp->v_type != VREG || obj == NULL)
490 return;
491
492 VM_OBJECT_WLOCK(obj);
493 VI_LOCK(vp);
494 vm_object_clear_flag(obj, OBJ_TMPFS);
495 obj->un_pager.swp.swp_tmpfs = NULL;
496 VI_UNLOCK(vp);
497 VM_OBJECT_WUNLOCK(obj);
498 }
499
500 /*
501 * Need to clear v_object for insmntque failure.
502 */
503 static void
504 tmpfs_insmntque_dtr(struct vnode *vp, void *dtr_arg)
505 {
506
507 tmpfs_destroy_vobject(vp, vp->v_object);
508 vp->v_object = NULL;
509 vp->v_data = NULL;
510 vp->v_op = &dead_vnodeops;
511 vgone(vp);
512 vput(vp);
513 }
514
515 /*
516 * Allocates a new vnode for the node node or returns a new reference to
517 * an existing one if the node had already a vnode referencing it. The
518 * resulting locked vnode is returned in *vpp.
519 *
520 * Returns zero on success or an appropriate error code on failure.
521 */
522 int
523 tmpfs_alloc_vp(struct mount *mp, struct tmpfs_node *node, int lkflag,
524 struct vnode **vpp)
525 {
526 struct vnode *vp;
527 struct tmpfs_mount *tm;
528 vm_object_t object;
529 int error;
530
531 error = 0;
532 tm = VFS_TO_TMPFS(mp);
533 TMPFS_NODE_LOCK(node);
534 tmpfs_ref_node_locked(node);
535 loop:
536 TMPFS_NODE_ASSERT_LOCKED(node);
537 if ((vp = node->tn_vnode) != NULL) {
538 MPASS((node->tn_vpstate & TMPFS_VNODE_DOOMED) == 0);
539 VI_LOCK(vp);
540 if ((node->tn_type == VDIR && node->tn_dir.tn_parent == NULL) ||
541 ((vp->v_iflag & VI_DOOMED) != 0 &&
542 (lkflag & LK_NOWAIT) != 0)) {
543 VI_UNLOCK(vp);
544 TMPFS_NODE_UNLOCK(node);
545 error = ENOENT;
546 vp = NULL;
547 goto out;
548 }
549 if ((vp->v_iflag & VI_DOOMED) != 0) {
550 VI_UNLOCK(vp);
551 node->tn_vpstate |= TMPFS_VNODE_WRECLAIM;
552 while ((node->tn_vpstate & TMPFS_VNODE_WRECLAIM) != 0) {
553 msleep(&node->tn_vnode, TMPFS_NODE_MTX(node),
554 0, "tmpfsE", 0);
555 }
556 goto loop;
557 }
558 TMPFS_NODE_UNLOCK(node);
559 error = vget(vp, lkflag | LK_INTERLOCK, curthread);
560 if (error == ENOENT) {
561 TMPFS_NODE_LOCK(node);
562 goto loop;
563 }
564 if (error != 0) {
565 vp = NULL;
566 goto out;
567 }
568
569 /*
570 * Make sure the vnode is still there after
571 * getting the interlock to avoid racing a free.
572 */
573 if (node->tn_vnode == NULL || node->tn_vnode != vp) {
574 vput(vp);
575 TMPFS_NODE_LOCK(node);
576 goto loop;
577 }
578
579 goto out;
580 }
581
582 if ((node->tn_vpstate & TMPFS_VNODE_DOOMED) ||
583 (node->tn_type == VDIR && node->tn_dir.tn_parent == NULL)) {
584 TMPFS_NODE_UNLOCK(node);
585 error = ENOENT;
586 vp = NULL;
587 goto out;
588 }
589
590 /*
591 * otherwise lock the vp list while we call getnewvnode
592 * since that can block.
593 */
594 if (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) {
595 node->tn_vpstate |= TMPFS_VNODE_WANT;
596 error = msleep((caddr_t) &node->tn_vpstate,
597 TMPFS_NODE_MTX(node), 0, "tmpfs_alloc_vp", 0);
598 if (error != 0)
599 goto out;
600 goto loop;
601 } else
602 node->tn_vpstate |= TMPFS_VNODE_ALLOCATING;
603
604 TMPFS_NODE_UNLOCK(node);
605
606 /* Get a new vnode and associate it with our node. */
607 error = getnewvnode("tmpfs", mp, VFS_TO_TMPFS(mp)->tm_nonc ?
608 &tmpfs_vnodeop_nonc_entries : &tmpfs_vnodeop_entries, &vp);
609 if (error != 0)
610 goto unlock;
611 MPASS(vp != NULL);
612
613 /* lkflag is ignored, the lock is exclusive */
614 (void) vn_lock(vp, lkflag | LK_RETRY);
615
616 vp->v_data = node;
617 vp->v_type = node->tn_type;
618
619 /* Type-specific initialization. */
620 switch (node->tn_type) {
621 case VBLK:
622 /* FALLTHROUGH */
623 case VCHR:
624 /* FALLTHROUGH */
625 case VLNK:
626 /* FALLTHROUGH */
627 case VSOCK:
628 break;
629 case VFIFO:
630 vp->v_op = &tmpfs_fifoop_entries;
631 break;
632 case VREG:
633 object = node->tn_reg.tn_aobj;
634 VM_OBJECT_WLOCK(object);
635 VI_LOCK(vp);
636 KASSERT(vp->v_object == NULL, ("Not NULL v_object in tmpfs"));
637 vp->v_object = object;
638 object->un_pager.swp.swp_tmpfs = vp;
639 vm_object_set_flag(object, OBJ_TMPFS);
640 VI_UNLOCK(vp);
641 VM_OBJECT_WUNLOCK(object);
642 break;
643 case VDIR:
644 MPASS(node->tn_dir.tn_parent != NULL);
645 if (node->tn_dir.tn_parent == node)
646 vp->v_vflag |= VV_ROOT;
647 break;
648
649 default:
650 panic("tmpfs_alloc_vp: type %p %d", node, (int)node->tn_type);
651 }
652 if (vp->v_type != VFIFO)
653 VN_LOCK_ASHARE(vp);
654
655 error = insmntque1(vp, mp, tmpfs_insmntque_dtr, NULL);
656 if (error != 0)
657 vp = NULL;
658
659 unlock:
660 TMPFS_NODE_LOCK(node);
661
662 MPASS(node->tn_vpstate & TMPFS_VNODE_ALLOCATING);
663 node->tn_vpstate &= ~TMPFS_VNODE_ALLOCATING;
664 node->tn_vnode = vp;
665
666 if (node->tn_vpstate & TMPFS_VNODE_WANT) {
667 node->tn_vpstate &= ~TMPFS_VNODE_WANT;
668 TMPFS_NODE_UNLOCK(node);
669 wakeup((caddr_t) &node->tn_vpstate);
670 } else
671 TMPFS_NODE_UNLOCK(node);
672
673 out:
674 if (error == 0) {
675 *vpp = vp;
676
677 #ifdef INVARIANTS
678 MPASS(*vpp != NULL && VOP_ISLOCKED(*vpp));
679 TMPFS_NODE_LOCK(node);
680 MPASS(*vpp == node->tn_vnode);
681 TMPFS_NODE_UNLOCK(node);
682 #endif
683 }
684 tmpfs_free_node(tm, node);
685
686 return (error);
687 }
688
689 /*
690 * Destroys the association between the vnode vp and the node it
691 * references.
692 */
693 void
694 tmpfs_free_vp(struct vnode *vp)
695 {
696 struct tmpfs_node *node;
697
698 node = VP_TO_TMPFS_NODE(vp);
699
700 TMPFS_NODE_ASSERT_LOCKED(node);
701 node->tn_vnode = NULL;
702 if ((node->tn_vpstate & TMPFS_VNODE_WRECLAIM) != 0)
703 wakeup(&node->tn_vnode);
704 node->tn_vpstate &= ~TMPFS_VNODE_WRECLAIM;
705 vp->v_data = NULL;
706 }
707
708 /*
709 * Allocates a new file of type 'type' and adds it to the parent directory
710 * 'dvp'; this addition is done using the component name given in 'cnp'.
711 * The ownership of the new file is automatically assigned based on the
712 * credentials of the caller (through 'cnp'), the group is set based on
713 * the parent directory and the mode is determined from the 'vap' argument.
714 * If successful, *vpp holds a vnode to the newly created file and zero
715 * is returned. Otherwise *vpp is NULL and the function returns an
716 * appropriate error code.
717 */
718 int
719 tmpfs_alloc_file(struct vnode *dvp, struct vnode **vpp, struct vattr *vap,
720 struct componentname *cnp, char *target)
721 {
722 int error;
723 struct tmpfs_dirent *de;
724 struct tmpfs_mount *tmp;
725 struct tmpfs_node *dnode;
726 struct tmpfs_node *node;
727 struct tmpfs_node *parent;
728
729 ASSERT_VOP_ELOCKED(dvp, "tmpfs_alloc_file");
730 MPASS(cnp->cn_flags & HASBUF);
731
732 tmp = VFS_TO_TMPFS(dvp->v_mount);
733 dnode = VP_TO_TMPFS_DIR(dvp);
734 *vpp = NULL;
735
736 /* If the entry we are creating is a directory, we cannot overflow
737 * the number of links of its parent, because it will get a new
738 * link. */
739 if (vap->va_type == VDIR) {
740 /* Ensure that we do not overflow the maximum number of links
741 * imposed by the system. */
742 MPASS(dnode->tn_links <= TMPFS_LINK_MAX);
743 if (dnode->tn_links == TMPFS_LINK_MAX) {
744 return (EMLINK);
745 }
746
747 parent = dnode;
748 MPASS(parent != NULL);
749 } else
750 parent = NULL;
751
752 /* Allocate a node that represents the new file. */
753 error = tmpfs_alloc_node(dvp->v_mount, tmp, vap->va_type,
754 cnp->cn_cred->cr_uid, dnode->tn_gid, vap->va_mode, parent,
755 target, vap->va_rdev, &node);
756 if (error != 0)
757 return (error);
758
759 /* Allocate a directory entry that points to the new file. */
760 error = tmpfs_alloc_dirent(tmp, node, cnp->cn_nameptr, cnp->cn_namelen,
761 &de);
762 if (error != 0) {
763 tmpfs_free_node(tmp, node);
764 return (error);
765 }
766
767 /* Allocate a vnode for the new file. */
768 error = tmpfs_alloc_vp(dvp->v_mount, node, LK_EXCLUSIVE, vpp);
769 if (error != 0) {
770 tmpfs_free_dirent(tmp, de);
771 tmpfs_free_node(tmp, node);
772 return (error);
773 }
774
775 /* Now that all required items are allocated, we can proceed to
776 * insert the new node into the directory, an operation that
777 * cannot fail. */
778 if (cnp->cn_flags & ISWHITEOUT)
779 tmpfs_dir_whiteout_remove(dvp, cnp);
780 tmpfs_dir_attach(dvp, de);
781 return (0);
782 }
783
784 struct tmpfs_dirent *
785 tmpfs_dir_first(struct tmpfs_node *dnode, struct tmpfs_dir_cursor *dc)
786 {
787 struct tmpfs_dirent *de;
788
789 de = RB_MIN(tmpfs_dir, &dnode->tn_dir.tn_dirhead);
790 dc->tdc_tree = de;
791 if (de != NULL && tmpfs_dirent_duphead(de))
792 de = LIST_FIRST(&de->ud.td_duphead);
793 dc->tdc_current = de;
794
795 return (dc->tdc_current);
796 }
797
798 struct tmpfs_dirent *
799 tmpfs_dir_next(struct tmpfs_node *dnode, struct tmpfs_dir_cursor *dc)
800 {
801 struct tmpfs_dirent *de;
802
803 MPASS(dc->tdc_tree != NULL);
804 if (tmpfs_dirent_dup(dc->tdc_current)) {
805 dc->tdc_current = LIST_NEXT(dc->tdc_current, uh.td_dup.entries);
806 if (dc->tdc_current != NULL)
807 return (dc->tdc_current);
808 }
809 dc->tdc_tree = dc->tdc_current = RB_NEXT(tmpfs_dir,
810 &dnode->tn_dir.tn_dirhead, dc->tdc_tree);
811 if ((de = dc->tdc_current) != NULL && tmpfs_dirent_duphead(de)) {
812 dc->tdc_current = LIST_FIRST(&de->ud.td_duphead);
813 MPASS(dc->tdc_current != NULL);
814 }
815
816 return (dc->tdc_current);
817 }
818
819 /* Lookup directory entry in RB-Tree. Function may return duphead entry. */
820 static struct tmpfs_dirent *
821 tmpfs_dir_xlookup_hash(struct tmpfs_node *dnode, uint32_t hash)
822 {
823 struct tmpfs_dirent *de, dekey;
824
825 dekey.td_hash = hash;
826 de = RB_FIND(tmpfs_dir, &dnode->tn_dir.tn_dirhead, &dekey);
827 return (de);
828 }
829
830 /* Lookup directory entry by cookie, initialize directory cursor accordingly. */
831 static struct tmpfs_dirent *
832 tmpfs_dir_lookup_cookie(struct tmpfs_node *node, off_t cookie,
833 struct tmpfs_dir_cursor *dc)
834 {
835 struct tmpfs_dir *dirhead = &node->tn_dir.tn_dirhead;
836 struct tmpfs_dirent *de, dekey;
837
838 MPASS(cookie >= TMPFS_DIRCOOKIE_MIN);
839
840 if (cookie == node->tn_dir.tn_readdir_lastn &&
841 (de = node->tn_dir.tn_readdir_lastp) != NULL) {
842 /* Protect against possible race, tn_readdir_last[pn]
843 * may be updated with only shared vnode lock held. */
844 if (cookie == tmpfs_dirent_cookie(de))
845 goto out;
846 }
847
848 if ((cookie & TMPFS_DIRCOOKIE_DUP) != 0) {
849 LIST_FOREACH(de, &node->tn_dir.tn_dupindex,
850 uh.td_dup.index_entries) {
851 MPASS(tmpfs_dirent_dup(de));
852 if (de->td_cookie == cookie)
853 goto out;
854 /* dupindex list is sorted. */
855 if (de->td_cookie < cookie) {
856 de = NULL;
857 goto out;
858 }
859 }
860 MPASS(de == NULL);
861 goto out;
862 }
863
864 if ((cookie & TMPFS_DIRCOOKIE_MASK) != cookie) {
865 de = NULL;
866 } else {
867 dekey.td_hash = cookie;
868 /* Recover if direntry for cookie was removed */
869 de = RB_NFIND(tmpfs_dir, dirhead, &dekey);
870 }
871 dc->tdc_tree = de;
872 dc->tdc_current = de;
873 if (de != NULL && tmpfs_dirent_duphead(de)) {
874 dc->tdc_current = LIST_FIRST(&de->ud.td_duphead);
875 MPASS(dc->tdc_current != NULL);
876 }
877 return (dc->tdc_current);
878
879 out:
880 dc->tdc_tree = de;
881 dc->tdc_current = de;
882 if (de != NULL && tmpfs_dirent_dup(de))
883 dc->tdc_tree = tmpfs_dir_xlookup_hash(node,
884 de->td_hash);
885 return (dc->tdc_current);
886 }
887
888 /*
889 * Looks for a directory entry in the directory represented by node.
890 * 'cnp' describes the name of the entry to look for. Note that the .
891 * and .. components are not allowed as they do not physically exist
892 * within directories.
893 *
894 * Returns a pointer to the entry when found, otherwise NULL.
895 */
896 struct tmpfs_dirent *
897 tmpfs_dir_lookup(struct tmpfs_node *node, struct tmpfs_node *f,
898 struct componentname *cnp)
899 {
900 struct tmpfs_dir_duphead *duphead;
901 struct tmpfs_dirent *de;
902 uint32_t hash;
903
904 MPASS(IMPLIES(cnp->cn_namelen == 1, cnp->cn_nameptr[0] != '.'));
905 MPASS(IMPLIES(cnp->cn_namelen == 2, !(cnp->cn_nameptr[0] == '.' &&
906 cnp->cn_nameptr[1] == '.')));
907 TMPFS_VALIDATE_DIR(node);
908
909 hash = tmpfs_dirent_hash(cnp->cn_nameptr, cnp->cn_namelen);
910 de = tmpfs_dir_xlookup_hash(node, hash);
911 if (de != NULL && tmpfs_dirent_duphead(de)) {
912 duphead = &de->ud.td_duphead;
913 LIST_FOREACH(de, duphead, uh.td_dup.entries) {
914 if (TMPFS_DIRENT_MATCHES(de, cnp->cn_nameptr,
915 cnp->cn_namelen))
916 break;
917 }
918 } else if (de != NULL) {
919 if (!TMPFS_DIRENT_MATCHES(de, cnp->cn_nameptr,
920 cnp->cn_namelen))
921 de = NULL;
922 }
923 if (de != NULL && f != NULL && de->td_node != f)
924 de = NULL;
925
926 return (de);
927 }
928
929 /*
930 * Attach duplicate-cookie directory entry nde to dnode and insert to dupindex
931 * list, allocate new cookie value.
932 */
933 static void
934 tmpfs_dir_attach_dup(struct tmpfs_node *dnode,
935 struct tmpfs_dir_duphead *duphead, struct tmpfs_dirent *nde)
936 {
937 struct tmpfs_dir_duphead *dupindex;
938 struct tmpfs_dirent *de, *pde;
939
940 dupindex = &dnode->tn_dir.tn_dupindex;
941 de = LIST_FIRST(dupindex);
942 if (de == NULL || de->td_cookie < TMPFS_DIRCOOKIE_DUP_MAX) {
943 if (de == NULL)
944 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MIN;
945 else
946 nde->td_cookie = de->td_cookie + 1;
947 MPASS(tmpfs_dirent_dup(nde));
948 LIST_INSERT_HEAD(dupindex, nde, uh.td_dup.index_entries);
949 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries);
950 return;
951 }
952
953 /*
954 * Cookie numbers are near exhaustion. Scan dupindex list for unused
955 * numbers. dupindex list is sorted in descending order. Keep it so
956 * after inserting nde.
957 */
958 while (1) {
959 pde = de;
960 de = LIST_NEXT(de, uh.td_dup.index_entries);
961 if (de == NULL && pde->td_cookie != TMPFS_DIRCOOKIE_DUP_MIN) {
962 /*
963 * Last element of the index doesn't have minimal cookie
964 * value, use it.
965 */
966 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MIN;
967 LIST_INSERT_AFTER(pde, nde, uh.td_dup.index_entries);
968 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries);
969 return;
970 } else if (de == NULL) {
971 /*
972 * We are so lucky have 2^30 hash duplicates in single
973 * directory :) Return largest possible cookie value.
974 * It should be fine except possible issues with
975 * VOP_READDIR restart.
976 */
977 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MAX;
978 LIST_INSERT_HEAD(dupindex, nde,
979 uh.td_dup.index_entries);
980 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries);
981 return;
982 }
983 if (de->td_cookie + 1 == pde->td_cookie ||
984 de->td_cookie >= TMPFS_DIRCOOKIE_DUP_MAX)
985 continue; /* No hole or invalid cookie. */
986 nde->td_cookie = de->td_cookie + 1;
987 MPASS(tmpfs_dirent_dup(nde));
988 MPASS(pde->td_cookie > nde->td_cookie);
989 MPASS(nde->td_cookie > de->td_cookie);
990 LIST_INSERT_BEFORE(de, nde, uh.td_dup.index_entries);
991 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries);
992 return;
993 }
994 }
995
996 /*
997 * Attaches the directory entry de to the directory represented by vp.
998 * Note that this does not change the link count of the node pointed by
999 * the directory entry, as this is done by tmpfs_alloc_dirent.
1000 */
1001 void
1002 tmpfs_dir_attach(struct vnode *vp, struct tmpfs_dirent *de)
1003 {
1004 struct tmpfs_node *dnode;
1005 struct tmpfs_dirent *xde, *nde;
1006
1007 ASSERT_VOP_ELOCKED(vp, __func__);
1008 MPASS(de->td_namelen > 0);
1009 MPASS(de->td_hash >= TMPFS_DIRCOOKIE_MIN);
1010 MPASS(de->td_cookie == de->td_hash);
1011
1012 dnode = VP_TO_TMPFS_DIR(vp);
1013 dnode->tn_dir.tn_readdir_lastn = 0;
1014 dnode->tn_dir.tn_readdir_lastp = NULL;
1015
1016 MPASS(!tmpfs_dirent_dup(de));
1017 xde = RB_INSERT(tmpfs_dir, &dnode->tn_dir.tn_dirhead, de);
1018 if (xde != NULL && tmpfs_dirent_duphead(xde))
1019 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, de);
1020 else if (xde != NULL) {
1021 /*
1022 * Allocate new duphead. Swap xde with duphead to avoid
1023 * adding/removing elements with the same hash.
1024 */
1025 MPASS(!tmpfs_dirent_dup(xde));
1026 tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), NULL, NULL, 0,
1027 &nde);
1028 /* *nde = *xde; XXX gcc 4.2.1 may generate invalid code. */
1029 memcpy(nde, xde, sizeof(*xde));
1030 xde->td_cookie |= TMPFS_DIRCOOKIE_DUPHEAD;
1031 LIST_INIT(&xde->ud.td_duphead);
1032 xde->td_namelen = 0;
1033 xde->td_node = NULL;
1034 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, nde);
1035 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, de);
1036 }
1037 dnode->tn_size += sizeof(struct tmpfs_dirent);
1038 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
1039 TMPFS_NODE_MODIFIED;
1040 tmpfs_update(vp);
1041 }
1042
1043 /*
1044 * Detaches the directory entry de from the directory represented by vp.
1045 * Note that this does not change the link count of the node pointed by
1046 * the directory entry, as this is done by tmpfs_free_dirent.
1047 */
1048 void
1049 tmpfs_dir_detach(struct vnode *vp, struct tmpfs_dirent *de)
1050 {
1051 struct tmpfs_mount *tmp;
1052 struct tmpfs_dir *head;
1053 struct tmpfs_node *dnode;
1054 struct tmpfs_dirent *xde;
1055
1056 ASSERT_VOP_ELOCKED(vp, __func__);
1057
1058 dnode = VP_TO_TMPFS_DIR(vp);
1059 head = &dnode->tn_dir.tn_dirhead;
1060 dnode->tn_dir.tn_readdir_lastn = 0;
1061 dnode->tn_dir.tn_readdir_lastp = NULL;
1062
1063 if (tmpfs_dirent_dup(de)) {
1064 /* Remove duphead if de was last entry. */
1065 if (LIST_NEXT(de, uh.td_dup.entries) == NULL) {
1066 xde = tmpfs_dir_xlookup_hash(dnode, de->td_hash);
1067 MPASS(tmpfs_dirent_duphead(xde));
1068 } else
1069 xde = NULL;
1070 LIST_REMOVE(de, uh.td_dup.entries);
1071 LIST_REMOVE(de, uh.td_dup.index_entries);
1072 if (xde != NULL) {
1073 if (LIST_EMPTY(&xde->ud.td_duphead)) {
1074 RB_REMOVE(tmpfs_dir, head, xde);
1075 tmp = VFS_TO_TMPFS(vp->v_mount);
1076 MPASS(xde->td_node == NULL);
1077 tmpfs_free_dirent(tmp, xde);
1078 }
1079 }
1080 de->td_cookie = de->td_hash;
1081 } else
1082 RB_REMOVE(tmpfs_dir, head, de);
1083
1084 dnode->tn_size -= sizeof(struct tmpfs_dirent);
1085 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
1086 TMPFS_NODE_MODIFIED;
1087 tmpfs_update(vp);
1088 }
1089
1090 void
1091 tmpfs_dir_destroy(struct tmpfs_mount *tmp, struct tmpfs_node *dnode)
1092 {
1093 struct tmpfs_dirent *de, *dde, *nde;
1094
1095 RB_FOREACH_SAFE(de, tmpfs_dir, &dnode->tn_dir.tn_dirhead, nde) {
1096 RB_REMOVE(tmpfs_dir, &dnode->tn_dir.tn_dirhead, de);
1097 /* Node may already be destroyed. */
1098 de->td_node = NULL;
1099 if (tmpfs_dirent_duphead(de)) {
1100 while ((dde = LIST_FIRST(&de->ud.td_duphead)) != NULL) {
1101 LIST_REMOVE(dde, uh.td_dup.entries);
1102 dde->td_node = NULL;
1103 tmpfs_free_dirent(tmp, dde);
1104 }
1105 }
1106 tmpfs_free_dirent(tmp, de);
1107 }
1108 }
1109
1110 /*
1111 * Helper function for tmpfs_readdir. Creates a '.' entry for the given
1112 * directory and returns it in the uio space. The function returns 0
1113 * on success, -1 if there was not enough space in the uio structure to
1114 * hold the directory entry or an appropriate error code if another
1115 * error happens.
1116 */
1117 static int
1118 tmpfs_dir_getdotdent(struct tmpfs_node *node, struct uio *uio)
1119 {
1120 int error;
1121 struct dirent dent;
1122
1123 TMPFS_VALIDATE_DIR(node);
1124 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOT);
1125
1126 dent.d_fileno = node->tn_id;
1127 dent.d_type = DT_DIR;
1128 dent.d_namlen = 1;
1129 dent.d_name[0] = '.';
1130 dent.d_reclen = GENERIC_DIRSIZ(&dent);
1131 dirent_terminate(&dent);
1132
1133 if (dent.d_reclen > uio->uio_resid)
1134 error = EJUSTRETURN;
1135 else
1136 error = uiomove(&dent, dent.d_reclen, uio);
1137
1138 tmpfs_set_status(node, TMPFS_NODE_ACCESSED);
1139
1140 return (error);
1141 }
1142
1143 /*
1144 * Helper function for tmpfs_readdir. Creates a '..' entry for the given
1145 * directory and returns it in the uio space. The function returns 0
1146 * on success, -1 if there was not enough space in the uio structure to
1147 * hold the directory entry or an appropriate error code if another
1148 * error happens.
1149 */
1150 static int
1151 tmpfs_dir_getdotdotdent(struct tmpfs_node *node, struct uio *uio)
1152 {
1153 int error;
1154 struct dirent dent;
1155
1156 TMPFS_VALIDATE_DIR(node);
1157 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT);
1158
1159 /*
1160 * Return ENOENT if the current node is already removed.
1161 */
1162 TMPFS_ASSERT_LOCKED(node);
1163 if (node->tn_dir.tn_parent == NULL)
1164 return (ENOENT);
1165
1166 TMPFS_NODE_LOCK(node->tn_dir.tn_parent);
1167 dent.d_fileno = node->tn_dir.tn_parent->tn_id;
1168 TMPFS_NODE_UNLOCK(node->tn_dir.tn_parent);
1169
1170 dent.d_type = DT_DIR;
1171 dent.d_namlen = 2;
1172 dent.d_name[0] = '.';
1173 dent.d_name[1] = '.';
1174 dent.d_reclen = GENERIC_DIRSIZ(&dent);
1175 dirent_terminate(&dent);
1176
1177 if (dent.d_reclen > uio->uio_resid)
1178 error = EJUSTRETURN;
1179 else
1180 error = uiomove(&dent, dent.d_reclen, uio);
1181
1182 tmpfs_set_status(node, TMPFS_NODE_ACCESSED);
1183
1184 return (error);
1185 }
1186
1187 /*
1188 * Helper function for tmpfs_readdir. Returns as much directory entries
1189 * as can fit in the uio space. The read starts at uio->uio_offset.
1190 * The function returns 0 on success, -1 if there was not enough space
1191 * in the uio structure to hold the directory entry or an appropriate
1192 * error code if another error happens.
1193 */
1194 int
1195 tmpfs_dir_getdents(struct tmpfs_node *node, struct uio *uio, int maxcookies,
1196 u_long *cookies, int *ncookies)
1197 {
1198 struct tmpfs_dir_cursor dc;
1199 struct tmpfs_dirent *de;
1200 off_t off;
1201 int error;
1202
1203 TMPFS_VALIDATE_DIR(node);
1204
1205 off = 0;
1206
1207 /*
1208 * Lookup the node from the current offset. The starting offset of
1209 * 0 will lookup both '.' and '..', and then the first real entry,
1210 * or EOF if there are none. Then find all entries for the dir that
1211 * fit into the buffer. Once no more entries are found (de == NULL),
1212 * the offset is set to TMPFS_DIRCOOKIE_EOF, which will cause the next
1213 * call to return 0.
1214 */
1215 switch (uio->uio_offset) {
1216 case TMPFS_DIRCOOKIE_DOT:
1217 error = tmpfs_dir_getdotdent(node, uio);
1218 if (error != 0)
1219 return (error);
1220 uio->uio_offset = TMPFS_DIRCOOKIE_DOTDOT;
1221 if (cookies != NULL)
1222 cookies[(*ncookies)++] = off = uio->uio_offset;
1223 /* FALLTHROUGH */
1224 case TMPFS_DIRCOOKIE_DOTDOT:
1225 error = tmpfs_dir_getdotdotdent(node, uio);
1226 if (error != 0)
1227 return (error);
1228 de = tmpfs_dir_first(node, &dc);
1229 uio->uio_offset = tmpfs_dirent_cookie(de);
1230 if (cookies != NULL)
1231 cookies[(*ncookies)++] = off = uio->uio_offset;
1232 /* EOF. */
1233 if (de == NULL)
1234 return (0);
1235 break;
1236 case TMPFS_DIRCOOKIE_EOF:
1237 return (0);
1238 default:
1239 de = tmpfs_dir_lookup_cookie(node, uio->uio_offset, &dc);
1240 if (de == NULL)
1241 return (EINVAL);
1242 if (cookies != NULL)
1243 off = tmpfs_dirent_cookie(de);
1244 }
1245
1246 /* Read as much entries as possible; i.e., until we reach the end of
1247 * the directory or we exhaust uio space. */
1248 do {
1249 struct dirent d;
1250
1251 /* Create a dirent structure representing the current
1252 * tmpfs_node and fill it. */
1253 if (de->td_node == NULL) {
1254 d.d_fileno = 1;
1255 d.d_type = DT_WHT;
1256 } else {
1257 d.d_fileno = de->td_node->tn_id;
1258 switch (de->td_node->tn_type) {
1259 case VBLK:
1260 d.d_type = DT_BLK;
1261 break;
1262
1263 case VCHR:
1264 d.d_type = DT_CHR;
1265 break;
1266
1267 case VDIR:
1268 d.d_type = DT_DIR;
1269 break;
1270
1271 case VFIFO:
1272 d.d_type = DT_FIFO;
1273 break;
1274
1275 case VLNK:
1276 d.d_type = DT_LNK;
1277 break;
1278
1279 case VREG:
1280 d.d_type = DT_REG;
1281 break;
1282
1283 case VSOCK:
1284 d.d_type = DT_SOCK;
1285 break;
1286
1287 default:
1288 panic("tmpfs_dir_getdents: type %p %d",
1289 de->td_node, (int)de->td_node->tn_type);
1290 }
1291 }
1292 d.d_namlen = de->td_namelen;
1293 MPASS(de->td_namelen < sizeof(d.d_name));
1294 (void)memcpy(d.d_name, de->ud.td_name, de->td_namelen);
1295 d.d_reclen = GENERIC_DIRSIZ(&d);
1296 dirent_terminate(&d);
1297
1298 /* Stop reading if the directory entry we are treating is
1299 * bigger than the amount of data that can be returned. */
1300 if (d.d_reclen > uio->uio_resid) {
1301 error = EJUSTRETURN;
1302 break;
1303 }
1304
1305 /* Copy the new dirent structure into the output buffer and
1306 * advance pointers. */
1307 error = uiomove(&d, d.d_reclen, uio);
1308 if (error == 0) {
1309 de = tmpfs_dir_next(node, &dc);
1310 if (cookies != NULL) {
1311 off = tmpfs_dirent_cookie(de);
1312 MPASS(*ncookies < maxcookies);
1313 cookies[(*ncookies)++] = off;
1314 }
1315 }
1316 } while (error == 0 && uio->uio_resid > 0 && de != NULL);
1317
1318 /* Skip setting off when using cookies as it is already done above. */
1319 if (cookies == NULL)
1320 off = tmpfs_dirent_cookie(de);
1321
1322 /* Update the offset and cache. */
1323 uio->uio_offset = off;
1324 node->tn_dir.tn_readdir_lastn = off;
1325 node->tn_dir.tn_readdir_lastp = de;
1326
1327 tmpfs_set_status(node, TMPFS_NODE_ACCESSED);
1328 return error;
1329 }
1330
1331 int
1332 tmpfs_dir_whiteout_add(struct vnode *dvp, struct componentname *cnp)
1333 {
1334 struct tmpfs_dirent *de;
1335 int error;
1336
1337 error = tmpfs_alloc_dirent(VFS_TO_TMPFS(dvp->v_mount), NULL,
1338 cnp->cn_nameptr, cnp->cn_namelen, &de);
1339 if (error != 0)
1340 return (error);
1341 tmpfs_dir_attach(dvp, de);
1342 return (0);
1343 }
1344
1345 void
1346 tmpfs_dir_whiteout_remove(struct vnode *dvp, struct componentname *cnp)
1347 {
1348 struct tmpfs_dirent *de;
1349
1350 de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(dvp), NULL, cnp);
1351 MPASS(de != NULL && de->td_node == NULL);
1352 tmpfs_dir_detach(dvp, de);
1353 tmpfs_free_dirent(VFS_TO_TMPFS(dvp->v_mount), de);
1354 }
1355
1356 /*
1357 * Resizes the aobj associated with the regular file pointed to by 'vp' to the
1358 * size 'newsize'. 'vp' must point to a vnode that represents a regular file.
1359 * 'newsize' must be positive.
1360 *
1361 * Returns zero on success or an appropriate error code on failure.
1362 */
1363 int
1364 tmpfs_reg_resize(struct vnode *vp, off_t newsize, boolean_t ignerr)
1365 {
1366 struct tmpfs_mount *tmp;
1367 struct tmpfs_node *node;
1368 vm_object_t uobj;
1369 vm_page_t m;
1370 vm_pindex_t idx, newpages, oldpages;
1371 off_t oldsize;
1372 int base, rv;
1373
1374 MPASS(vp->v_type == VREG);
1375 MPASS(newsize >= 0);
1376
1377 node = VP_TO_TMPFS_NODE(vp);
1378 uobj = node->tn_reg.tn_aobj;
1379 tmp = VFS_TO_TMPFS(vp->v_mount);
1380
1381 /*
1382 * Convert the old and new sizes to the number of pages needed to
1383 * store them. It may happen that we do not need to do anything
1384 * because the last allocated page can accommodate the change on
1385 * its own.
1386 */
1387 oldsize = node->tn_size;
1388 oldpages = OFF_TO_IDX(oldsize + PAGE_MASK);
1389 MPASS(oldpages == uobj->size);
1390 newpages = OFF_TO_IDX(newsize + PAGE_MASK);
1391
1392 if (__predict_true(newpages == oldpages && newsize >= oldsize)) {
1393 node->tn_size = newsize;
1394 return (0);
1395 }
1396
1397 if (newpages > oldpages &&
1398 tmpfs_pages_check_avail(tmp, newpages - oldpages) == 0)
1399 return (ENOSPC);
1400
1401 VM_OBJECT_WLOCK(uobj);
1402 if (newsize < oldsize) {
1403 /*
1404 * Zero the truncated part of the last page.
1405 */
1406 base = newsize & PAGE_MASK;
1407 if (base != 0) {
1408 idx = OFF_TO_IDX(newsize);
1409 retry:
1410 m = vm_page_lookup(uobj, idx);
1411 if (m != NULL) {
1412 if (vm_page_sleep_if_busy(m, "tmfssz"))
1413 goto retry;
1414 MPASS(m->valid == VM_PAGE_BITS_ALL);
1415 } else if (vm_pager_has_page(uobj, idx, NULL, NULL)) {
1416 m = vm_page_alloc(uobj, idx, VM_ALLOC_NORMAL |
1417 VM_ALLOC_WAITFAIL);
1418 if (m == NULL)
1419 goto retry;
1420 rv = vm_pager_get_pages(uobj, &m, 1, NULL,
1421 NULL);
1422 vm_page_lock(m);
1423 if (rv == VM_PAGER_OK) {
1424 /*
1425 * Since the page was not resident,
1426 * and therefore not recently
1427 * accessed, immediately enqueue it
1428 * for asynchronous laundering. The
1429 * current operation is not regarded
1430 * as an access.
1431 */
1432 vm_page_launder(m);
1433 vm_page_unlock(m);
1434 vm_page_xunbusy(m);
1435 } else {
1436 vm_page_free(m);
1437 vm_page_unlock(m);
1438 if (ignerr)
1439 m = NULL;
1440 else {
1441 VM_OBJECT_WUNLOCK(uobj);
1442 return (EIO);
1443 }
1444 }
1445 }
1446 if (m != NULL) {
1447 pmap_zero_page_area(m, base, PAGE_SIZE - base);
1448 vm_page_dirty(m);
1449 vm_pager_page_unswapped(m);
1450 }
1451 }
1452
1453 /*
1454 * Release any swap space and free any whole pages.
1455 */
1456 if (newpages < oldpages) {
1457 swap_pager_freespace(uobj, newpages, oldpages -
1458 newpages);
1459 vm_object_page_remove(uobj, newpages, 0, 0);
1460 }
1461 }
1462 uobj->size = newpages;
1463 VM_OBJECT_WUNLOCK(uobj);
1464
1465 atomic_add_long(&tmp->tm_pages_used, newpages - oldpages);
1466
1467 node->tn_size = newsize;
1468 return (0);
1469 }
1470
1471 void
1472 tmpfs_check_mtime(struct vnode *vp)
1473 {
1474 struct tmpfs_node *node;
1475 struct vm_object *obj;
1476
1477 ASSERT_VOP_ELOCKED(vp, "check_mtime");
1478 if (vp->v_type != VREG)
1479 return;
1480 obj = vp->v_object;
1481 KASSERT((obj->flags & (OBJ_TMPFS_NODE | OBJ_TMPFS)) ==
1482 (OBJ_TMPFS_NODE | OBJ_TMPFS), ("non-tmpfs obj"));
1483 /* unlocked read */
1484 if ((obj->flags & OBJ_TMPFS_DIRTY) != 0) {
1485 VM_OBJECT_WLOCK(obj);
1486 if ((obj->flags & OBJ_TMPFS_DIRTY) != 0) {
1487 obj->flags &= ~OBJ_TMPFS_DIRTY;
1488 node = VP_TO_TMPFS_NODE(vp);
1489 node->tn_status |= TMPFS_NODE_MODIFIED |
1490 TMPFS_NODE_CHANGED;
1491 }
1492 VM_OBJECT_WUNLOCK(obj);
1493 }
1494 }
1495
1496 /*
1497 * Change flags of the given vnode.
1498 * Caller should execute tmpfs_update on vp after a successful execution.
1499 * The vnode must be locked on entry and remain locked on exit.
1500 */
1501 int
1502 tmpfs_chflags(struct vnode *vp, u_long flags, struct ucred *cred,
1503 struct thread *p)
1504 {
1505 int error;
1506 struct tmpfs_node *node;
1507
1508 ASSERT_VOP_ELOCKED(vp, "chflags");
1509
1510 node = VP_TO_TMPFS_NODE(vp);
1511
1512 if ((flags & ~(SF_APPEND | SF_ARCHIVED | SF_IMMUTABLE | SF_NOUNLINK |
1513 UF_APPEND | UF_ARCHIVE | UF_HIDDEN | UF_IMMUTABLE | UF_NODUMP |
1514 UF_NOUNLINK | UF_OFFLINE | UF_OPAQUE | UF_READONLY | UF_REPARSE |
1515 UF_SPARSE | UF_SYSTEM)) != 0)
1516 return (EOPNOTSUPP);
1517
1518 /* Disallow this operation if the file system is mounted read-only. */
1519 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1520 return EROFS;
1521
1522 /*
1523 * Callers may only modify the file flags on objects they
1524 * have VADMIN rights for.
1525 */
1526 if ((error = VOP_ACCESS(vp, VADMIN, cred, p)))
1527 return (error);
1528 /*
1529 * Unprivileged processes are not permitted to unset system
1530 * flags, or modify flags if any system flags are set.
1531 */
1532 if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0)) {
1533 if (node->tn_flags &
1534 (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND)) {
1535 error = securelevel_gt(cred, 0);
1536 if (error)
1537 return (error);
1538 }
1539 } else {
1540 if (node->tn_flags &
1541 (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND) ||
1542 ((flags ^ node->tn_flags) & SF_SETTABLE))
1543 return (EPERM);
1544 }
1545 node->tn_flags = flags;
1546 node->tn_status |= TMPFS_NODE_CHANGED;
1547
1548 ASSERT_VOP_ELOCKED(vp, "chflags2");
1549
1550 return (0);
1551 }
1552
1553 /*
1554 * Change access mode on the given vnode.
1555 * Caller should execute tmpfs_update on vp after a successful execution.
1556 * The vnode must be locked on entry and remain locked on exit.
1557 */
1558 int
1559 tmpfs_chmod(struct vnode *vp, mode_t mode, struct ucred *cred, struct thread *p)
1560 {
1561 int error;
1562 struct tmpfs_node *node;
1563
1564 ASSERT_VOP_ELOCKED(vp, "chmod");
1565
1566 node = VP_TO_TMPFS_NODE(vp);
1567
1568 /* Disallow this operation if the file system is mounted read-only. */
1569 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1570 return EROFS;
1571
1572 /* Immutable or append-only files cannot be modified, either. */
1573 if (node->tn_flags & (IMMUTABLE | APPEND))
1574 return EPERM;
1575
1576 /*
1577 * To modify the permissions on a file, must possess VADMIN
1578 * for that file.
1579 */
1580 if ((error = VOP_ACCESS(vp, VADMIN, cred, p)))
1581 return (error);
1582
1583 /*
1584 * Privileged processes may set the sticky bit on non-directories,
1585 * as well as set the setgid bit on a file with a group that the
1586 * process is not a member of.
1587 */
1588 if (vp->v_type != VDIR && (mode & S_ISTXT)) {
1589 if (priv_check_cred(cred, PRIV_VFS_STICKYFILE, 0))
1590 return (EFTYPE);
1591 }
1592 if (!groupmember(node->tn_gid, cred) && (mode & S_ISGID)) {
1593 error = priv_check_cred(cred, PRIV_VFS_SETGID, 0);
1594 if (error)
1595 return (error);
1596 }
1597
1598
1599 node->tn_mode &= ~ALLPERMS;
1600 node->tn_mode |= mode & ALLPERMS;
1601
1602 node->tn_status |= TMPFS_NODE_CHANGED;
1603
1604 ASSERT_VOP_ELOCKED(vp, "chmod2");
1605
1606 return (0);
1607 }
1608
1609 /*
1610 * Change ownership of the given vnode. At least one of uid or gid must
1611 * be different than VNOVAL. If one is set to that value, the attribute
1612 * is unchanged.
1613 * Caller should execute tmpfs_update on vp after a successful execution.
1614 * The vnode must be locked on entry and remain locked on exit.
1615 */
1616 int
1617 tmpfs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred,
1618 struct thread *p)
1619 {
1620 int error;
1621 struct tmpfs_node *node;
1622 uid_t ouid;
1623 gid_t ogid;
1624
1625 ASSERT_VOP_ELOCKED(vp, "chown");
1626
1627 node = VP_TO_TMPFS_NODE(vp);
1628
1629 /* Assign default values if they are unknown. */
1630 MPASS(uid != VNOVAL || gid != VNOVAL);
1631 if (uid == VNOVAL)
1632 uid = node->tn_uid;
1633 if (gid == VNOVAL)
1634 gid = node->tn_gid;
1635 MPASS(uid != VNOVAL && gid != VNOVAL);
1636
1637 /* Disallow this operation if the file system is mounted read-only. */
1638 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1639 return EROFS;
1640
1641 /* Immutable or append-only files cannot be modified, either. */
1642 if (node->tn_flags & (IMMUTABLE | APPEND))
1643 return EPERM;
1644
1645 /*
1646 * To modify the ownership of a file, must possess VADMIN for that
1647 * file.
1648 */
1649 if ((error = VOP_ACCESS(vp, VADMIN, cred, p)))
1650 return (error);
1651
1652 /*
1653 * To change the owner of a file, or change the group of a file to a
1654 * group of which we are not a member, the caller must have
1655 * privilege.
1656 */
1657 if ((uid != node->tn_uid ||
1658 (gid != node->tn_gid && !groupmember(gid, cred))) &&
1659 (error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0)))
1660 return (error);
1661
1662 ogid = node->tn_gid;
1663 ouid = node->tn_uid;
1664
1665 node->tn_uid = uid;
1666 node->tn_gid = gid;
1667
1668 node->tn_status |= TMPFS_NODE_CHANGED;
1669
1670 if ((node->tn_mode & (S_ISUID | S_ISGID)) && (ouid != uid || ogid != gid)) {
1671 if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID, 0))
1672 node->tn_mode &= ~(S_ISUID | S_ISGID);
1673 }
1674
1675 ASSERT_VOP_ELOCKED(vp, "chown2");
1676
1677 return (0);
1678 }
1679
1680 /*
1681 * Change size of the given vnode.
1682 * Caller should execute tmpfs_update on vp after a successful execution.
1683 * The vnode must be locked on entry and remain locked on exit.
1684 */
1685 int
1686 tmpfs_chsize(struct vnode *vp, u_quad_t size, struct ucred *cred,
1687 struct thread *p)
1688 {
1689 int error;
1690 struct tmpfs_node *node;
1691
1692 ASSERT_VOP_ELOCKED(vp, "chsize");
1693
1694 node = VP_TO_TMPFS_NODE(vp);
1695
1696 /* Decide whether this is a valid operation based on the file type. */
1697 error = 0;
1698 switch (vp->v_type) {
1699 case VDIR:
1700 return EISDIR;
1701
1702 case VREG:
1703 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1704 return EROFS;
1705 break;
1706
1707 case VBLK:
1708 /* FALLTHROUGH */
1709 case VCHR:
1710 /* FALLTHROUGH */
1711 case VFIFO:
1712 /* Allow modifications of special files even if in the file
1713 * system is mounted read-only (we are not modifying the
1714 * files themselves, but the objects they represent). */
1715 return 0;
1716
1717 default:
1718 /* Anything else is unsupported. */
1719 return EOPNOTSUPP;
1720 }
1721
1722 /* Immutable or append-only files cannot be modified, either. */
1723 if (node->tn_flags & (IMMUTABLE | APPEND))
1724 return EPERM;
1725
1726 error = tmpfs_truncate(vp, size);
1727 /* tmpfs_truncate will raise the NOTE_EXTEND and NOTE_ATTRIB kevents
1728 * for us, as will update tn_status; no need to do that here. */
1729
1730 ASSERT_VOP_ELOCKED(vp, "chsize2");
1731
1732 return (error);
1733 }
1734
1735 /*
1736 * Change access and modification times of the given vnode.
1737 * Caller should execute tmpfs_update on vp after a successful execution.
1738 * The vnode must be locked on entry and remain locked on exit.
1739 */
1740 int
1741 tmpfs_chtimes(struct vnode *vp, struct vattr *vap,
1742 struct ucred *cred, struct thread *l)
1743 {
1744 int error;
1745 struct tmpfs_node *node;
1746
1747 ASSERT_VOP_ELOCKED(vp, "chtimes");
1748
1749 node = VP_TO_TMPFS_NODE(vp);
1750
1751 /* Disallow this operation if the file system is mounted read-only. */
1752 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1753 return EROFS;
1754
1755 /* Immutable or append-only files cannot be modified, either. */
1756 if (node->tn_flags & (IMMUTABLE | APPEND))
1757 return EPERM;
1758
1759 error = vn_utimes_perm(vp, vap, cred, l);
1760 if (error != 0)
1761 return (error);
1762
1763 if (vap->va_atime.tv_sec != VNOVAL)
1764 node->tn_status |= TMPFS_NODE_ACCESSED;
1765
1766 if (vap->va_mtime.tv_sec != VNOVAL)
1767 node->tn_status |= TMPFS_NODE_MODIFIED;
1768
1769 if (vap->va_birthtime.tv_sec != VNOVAL)
1770 node->tn_status |= TMPFS_NODE_MODIFIED;
1771
1772 tmpfs_itimes(vp, &vap->va_atime, &vap->va_mtime);
1773
1774 if (vap->va_birthtime.tv_sec != VNOVAL)
1775 node->tn_birthtime = vap->va_birthtime;
1776 ASSERT_VOP_ELOCKED(vp, "chtimes2");
1777
1778 return (0);
1779 }
1780
1781 void
1782 tmpfs_set_status(struct tmpfs_node *node, int status)
1783 {
1784
1785 if ((node->tn_status & status) == status)
1786 return;
1787 TMPFS_NODE_LOCK(node);
1788 node->tn_status |= status;
1789 TMPFS_NODE_UNLOCK(node);
1790 }
1791
1792 /* Sync timestamps */
1793 void
1794 tmpfs_itimes(struct vnode *vp, const struct timespec *acc,
1795 const struct timespec *mod)
1796 {
1797 struct tmpfs_node *node;
1798 struct timespec now;
1799
1800 ASSERT_VOP_LOCKED(vp, "tmpfs_itimes");
1801 node = VP_TO_TMPFS_NODE(vp);
1802
1803 if ((node->tn_status & (TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
1804 TMPFS_NODE_CHANGED)) == 0)
1805 return;
1806
1807 vfs_timestamp(&now);
1808 TMPFS_NODE_LOCK(node);
1809 if (node->tn_status & TMPFS_NODE_ACCESSED) {
1810 if (acc == NULL)
1811 acc = &now;
1812 node->tn_atime = *acc;
1813 }
1814 if (node->tn_status & TMPFS_NODE_MODIFIED) {
1815 if (mod == NULL)
1816 mod = &now;
1817 node->tn_mtime = *mod;
1818 }
1819 if (node->tn_status & TMPFS_NODE_CHANGED)
1820 node->tn_ctime = now;
1821 node->tn_status &= ~(TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
1822 TMPFS_NODE_CHANGED);
1823 TMPFS_NODE_UNLOCK(node);
1824
1825 /* XXX: FIX? The entropy here is desirable, but the harvesting may be expensive */
1826 random_harvest_queue(node, sizeof(*node), RANDOM_FS_ATIME);
1827 }
1828
1829 void
1830 tmpfs_update(struct vnode *vp)
1831 {
1832
1833 tmpfs_itimes(vp, NULL, NULL);
1834 }
1835
1836 int
1837 tmpfs_truncate(struct vnode *vp, off_t length)
1838 {
1839 int error;
1840 struct tmpfs_node *node;
1841
1842 node = VP_TO_TMPFS_NODE(vp);
1843
1844 if (length < 0) {
1845 error = EINVAL;
1846 goto out;
1847 }
1848
1849 if (node->tn_size == length) {
1850 error = 0;
1851 goto out;
1852 }
1853
1854 if (length > VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize)
1855 return (EFBIG);
1856
1857 error = tmpfs_reg_resize(vp, length, FALSE);
1858 if (error == 0)
1859 node->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1860
1861 out:
1862 tmpfs_update(vp);
1863
1864 return (error);
1865 }
1866
1867 static __inline int
1868 tmpfs_dirtree_cmp(struct tmpfs_dirent *a, struct tmpfs_dirent *b)
1869 {
1870 if (a->td_hash > b->td_hash)
1871 return (1);
1872 else if (a->td_hash < b->td_hash)
1873 return (-1);
1874 return (0);
1875 }
1876
1877 RB_GENERATE_STATIC(tmpfs_dir, tmpfs_dirent, uh.td_entries, tmpfs_dirtree_cmp);
Cache object: 710bdcc04998b8324ecc0ebee1cc8c29
|