xref: /freebsd/sys/fs/tmpfs/tmpfs_subr.c (revision 072b47f31c4f9b331ccc09113ea77bc2ba849c05)
1 /*	$NetBSD: tmpfs_subr.c,v 1.35 2007/07/09 21:10:50 ad Exp $	*/
2 
3 /*-
4  * Copyright (c) 2005 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Julio M. Merino Vidal, developed as part of Google's Summer of Code
9  * 2005 program.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
21  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
22  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
23  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
24  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
25  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
26  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
27  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
28  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
29  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
30  * POSSIBILITY OF SUCH DAMAGE.
31  */
32 
33 /*
34  * Efficient memory file system supporting functions.
35  */
36 #include <sys/cdefs.h>
37 __FBSDID("$FreeBSD$");
38 
39 #include <sys/param.h>
40 #include <sys/fnv_hash.h>
41 #include <sys/lock.h>
42 #include <sys/namei.h>
43 #include <sys/priv.h>
44 #include <sys/proc.h>
45 #include <sys/rwlock.h>
46 #include <sys/stat.h>
47 #include <sys/systm.h>
48 #include <sys/sysctl.h>
49 #include <sys/vnode.h>
50 #include <sys/vmmeter.h>
51 
52 #include <vm/vm.h>
53 #include <vm/vm_param.h>
54 #include <vm/vm_object.h>
55 #include <vm/vm_page.h>
56 #include <vm/vm_pageout.h>
57 #include <vm/vm_pager.h>
58 #include <vm/vm_extern.h>
59 
60 #include <fs/tmpfs/tmpfs.h>
61 #include <fs/tmpfs/tmpfs_fifoops.h>
62 #include <fs/tmpfs/tmpfs_vnops.h>
63 
64 struct tmpfs_dir_cursor {
65 	struct tmpfs_dirent	*tdc_current;
66 	struct tmpfs_dirent	*tdc_tree;
67 };
68 
69 SYSCTL_NODE(_vfs, OID_AUTO, tmpfs, CTLFLAG_RW, 0, "tmpfs file system");
70 
71 static long tmpfs_pages_reserved = TMPFS_PAGES_MINRESERVED;
72 
73 static int
74 sysctl_mem_reserved(SYSCTL_HANDLER_ARGS)
75 {
76 	int error;
77 	long pages, bytes;
78 
79 	pages = *(long *)arg1;
80 	bytes = pages * PAGE_SIZE;
81 
82 	error = sysctl_handle_long(oidp, &bytes, 0, req);
83 	if (error || !req->newptr)
84 		return (error);
85 
86 	pages = bytes / PAGE_SIZE;
87 	if (pages < TMPFS_PAGES_MINRESERVED)
88 		return (EINVAL);
89 
90 	*(long *)arg1 = pages;
91 	return (0);
92 }
93 
94 SYSCTL_PROC(_vfs_tmpfs, OID_AUTO, memory_reserved, CTLTYPE_LONG|CTLFLAG_RW,
95     &tmpfs_pages_reserved, 0, sysctl_mem_reserved, "L",
96     "Amount of available memory and swap below which tmpfs growth stops");
97 
98 static __inline int tmpfs_dirtree_cmp(struct tmpfs_dirent *a,
99     struct tmpfs_dirent *b);
100 RB_PROTOTYPE_STATIC(tmpfs_dir, tmpfs_dirent, uh.td_entries, tmpfs_dirtree_cmp);
101 
102 size_t
103 tmpfs_mem_avail(void)
104 {
105 	vm_ooffset_t avail;
106 
107 	avail = swap_pager_avail + cnt.v_free_count + cnt.v_cache_count -
108 	    tmpfs_pages_reserved;
109 	if (__predict_false(avail < 0))
110 		avail = 0;
111 	return (avail);
112 }
113 
114 size_t
115 tmpfs_pages_used(struct tmpfs_mount *tmp)
116 {
117 	const size_t node_size = sizeof(struct tmpfs_node) +
118 	    sizeof(struct tmpfs_dirent);
119 	size_t meta_pages;
120 
121 	meta_pages = howmany((uintmax_t)tmp->tm_nodes_inuse * node_size,
122 	    PAGE_SIZE);
123 	return (meta_pages + tmp->tm_pages_used);
124 }
125 
126 static size_t
127 tmpfs_pages_check_avail(struct tmpfs_mount *tmp, size_t req_pages)
128 {
129 	if (tmpfs_mem_avail() < req_pages)
130 		return (0);
131 
132 	if (tmp->tm_pages_max != SIZE_MAX &&
133 	    tmp->tm_pages_max < req_pages + tmpfs_pages_used(tmp))
134 			return (0);
135 
136 	return (1);
137 }
138 
139 /* --------------------------------------------------------------------- */
140 
141 /*
142  * Allocates a new node of type 'type' inside the 'tmp' mount point, with
143  * its owner set to 'uid', its group to 'gid' and its mode set to 'mode',
144  * using the credentials of the process 'p'.
145  *
146  * If the node type is set to 'VDIR', then the parent parameter must point
147  * to the parent directory of the node being created.  It may only be NULL
148  * while allocating the root node.
149  *
150  * If the node type is set to 'VBLK' or 'VCHR', then the rdev parameter
151  * specifies the device the node represents.
152  *
153  * If the node type is set to 'VLNK', then the parameter target specifies
154  * the file name of the target file for the symbolic link that is being
155  * created.
156  *
157  * Note that new nodes are retrieved from the available list if it has
158  * items or, if it is empty, from the node pool as long as there is enough
159  * space to create them.
160  *
161  * Returns zero on success or an appropriate error code on failure.
162  */
163 int
164 tmpfs_alloc_node(struct tmpfs_mount *tmp, enum vtype type,
165     uid_t uid, gid_t gid, mode_t mode, struct tmpfs_node *parent,
166     char *target, dev_t rdev, struct tmpfs_node **node)
167 {
168 	struct tmpfs_node *nnode;
169 	vm_object_t obj;
170 
171 	/* If the root directory of the 'tmp' file system is not yet
172 	 * allocated, this must be the request to do it. */
173 	MPASS(IMPLIES(tmp->tm_root == NULL, parent == NULL && type == VDIR));
174 
175 	MPASS(IFF(type == VLNK, target != NULL));
176 	MPASS(IFF(type == VBLK || type == VCHR, rdev != VNOVAL));
177 
178 	if (tmp->tm_nodes_inuse >= tmp->tm_nodes_max)
179 		return (ENOSPC);
180 	if (tmpfs_pages_check_avail(tmp, 1) == 0)
181 		return (ENOSPC);
182 
183 	nnode = (struct tmpfs_node *)uma_zalloc_arg(
184 				tmp->tm_node_pool, tmp, M_WAITOK);
185 
186 	/* Generic initialization. */
187 	nnode->tn_type = type;
188 	vfs_timestamp(&nnode->tn_atime);
189 	nnode->tn_birthtime = nnode->tn_ctime = nnode->tn_mtime =
190 	    nnode->tn_atime;
191 	nnode->tn_uid = uid;
192 	nnode->tn_gid = gid;
193 	nnode->tn_mode = mode;
194 	nnode->tn_id = alloc_unr(tmp->tm_ino_unr);
195 
196 	/* Type-specific initialization. */
197 	switch (nnode->tn_type) {
198 	case VBLK:
199 	case VCHR:
200 		nnode->tn_rdev = rdev;
201 		break;
202 
203 	case VDIR:
204 		RB_INIT(&nnode->tn_dir.tn_dirhead);
205 		LIST_INIT(&nnode->tn_dir.tn_dupindex);
206 		MPASS(parent != nnode);
207 		MPASS(IMPLIES(parent == NULL, tmp->tm_root == NULL));
208 		nnode->tn_dir.tn_parent = (parent == NULL) ? nnode : parent;
209 		nnode->tn_dir.tn_readdir_lastn = 0;
210 		nnode->tn_dir.tn_readdir_lastp = NULL;
211 		nnode->tn_links++;
212 		TMPFS_NODE_LOCK(nnode->tn_dir.tn_parent);
213 		nnode->tn_dir.tn_parent->tn_links++;
214 		TMPFS_NODE_UNLOCK(nnode->tn_dir.tn_parent);
215 		break;
216 
217 	case VFIFO:
218 		/* FALLTHROUGH */
219 	case VSOCK:
220 		break;
221 
222 	case VLNK:
223 		MPASS(strlen(target) < MAXPATHLEN);
224 		nnode->tn_size = strlen(target);
225 		nnode->tn_link = malloc(nnode->tn_size, M_TMPFSNAME,
226 		    M_WAITOK);
227 		memcpy(nnode->tn_link, target, nnode->tn_size);
228 		break;
229 
230 	case VREG:
231 		obj = nnode->tn_reg.tn_aobj =
232 		    vm_pager_allocate(OBJT_SWAP, NULL, 0, VM_PROT_DEFAULT, 0,
233 			NULL /* XXXKIB - tmpfs needs swap reservation */);
234 		VM_OBJECT_WLOCK(obj);
235 		/* OBJ_TMPFS is set together with the setting of vp->v_object */
236 		vm_object_set_flag(obj, OBJ_NOSPLIT);
237 		vm_object_clear_flag(obj, OBJ_ONEMAPPING);
238 		VM_OBJECT_WUNLOCK(obj);
239 		break;
240 
241 	default:
242 		panic("tmpfs_alloc_node: type %p %d", nnode, (int)nnode->tn_type);
243 	}
244 
245 	TMPFS_LOCK(tmp);
246 	LIST_INSERT_HEAD(&tmp->tm_nodes_used, nnode, tn_entries);
247 	tmp->tm_nodes_inuse++;
248 	TMPFS_UNLOCK(tmp);
249 
250 	*node = nnode;
251 	return 0;
252 }
253 
254 /* --------------------------------------------------------------------- */
255 
256 /*
257  * Destroys the node pointed to by node from the file system 'tmp'.
258  * If the node does not belong to the given mount point, the results are
259  * unpredicted.
260  *
261  * If the node references a directory; no entries are allowed because
262  * their removal could need a recursive algorithm, something forbidden in
263  * kernel space.  Furthermore, there is not need to provide such
264  * functionality (recursive removal) because the only primitives offered
265  * to the user are the removal of empty directories and the deletion of
266  * individual files.
267  *
268  * Note that nodes are not really deleted; in fact, when a node has been
269  * allocated, it cannot be deleted during the whole life of the file
270  * system.  Instead, they are moved to the available list and remain there
271  * until reused.
272  */
273 void
274 tmpfs_free_node(struct tmpfs_mount *tmp, struct tmpfs_node *node)
275 {
276 	vm_object_t uobj;
277 
278 #ifdef INVARIANTS
279 	TMPFS_NODE_LOCK(node);
280 	MPASS(node->tn_vnode == NULL);
281 	MPASS((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0);
282 	TMPFS_NODE_UNLOCK(node);
283 #endif
284 
285 	TMPFS_LOCK(tmp);
286 	LIST_REMOVE(node, tn_entries);
287 	tmp->tm_nodes_inuse--;
288 	TMPFS_UNLOCK(tmp);
289 
290 	switch (node->tn_type) {
291 	case VNON:
292 		/* Do not do anything.  VNON is provided to let the
293 		 * allocation routine clean itself easily by avoiding
294 		 * duplicating code in it. */
295 		/* FALLTHROUGH */
296 	case VBLK:
297 		/* FALLTHROUGH */
298 	case VCHR:
299 		/* FALLTHROUGH */
300 	case VDIR:
301 		/* FALLTHROUGH */
302 	case VFIFO:
303 		/* FALLTHROUGH */
304 	case VSOCK:
305 		break;
306 
307 	case VLNK:
308 		free(node->tn_link, M_TMPFSNAME);
309 		break;
310 
311 	case VREG:
312 		uobj = node->tn_reg.tn_aobj;
313 		if (uobj != NULL) {
314 			TMPFS_LOCK(tmp);
315 			tmp->tm_pages_used -= uobj->size;
316 			TMPFS_UNLOCK(tmp);
317 			KASSERT((uobj->flags & OBJ_TMPFS) == 0,
318 			    ("leaked OBJ_TMPFS node %p vm_obj %p", node, uobj));
319 			vm_object_deallocate(uobj);
320 		}
321 		break;
322 
323 	default:
324 		panic("tmpfs_free_node: type %p %d", node, (int)node->tn_type);
325 	}
326 
327 	free_unr(tmp->tm_ino_unr, node->tn_id);
328 	uma_zfree(tmp->tm_node_pool, node);
329 }
330 
331 /* --------------------------------------------------------------------- */
332 
333 static __inline uint32_t
334 tmpfs_dirent_hash(const char *name, u_int len)
335 {
336 	uint32_t hash;
337 
338 	hash = fnv_32_buf(name, len, FNV1_32_INIT + len) & TMPFS_DIRCOOKIE_MASK;
339 #ifdef TMPFS_DEBUG_DIRCOOKIE_DUP
340 	hash &= 0xf;
341 #endif
342 	if (hash < TMPFS_DIRCOOKIE_MIN)
343 		hash += TMPFS_DIRCOOKIE_MIN;
344 
345 	return (hash);
346 }
347 
348 static __inline off_t
349 tmpfs_dirent_cookie(struct tmpfs_dirent *de)
350 {
351 	if (de == NULL)
352 		return (TMPFS_DIRCOOKIE_EOF);
353 
354 	MPASS(de->td_cookie >= TMPFS_DIRCOOKIE_MIN);
355 
356 	return (de->td_cookie);
357 }
358 
359 static __inline boolean_t
360 tmpfs_dirent_dup(struct tmpfs_dirent *de)
361 {
362 	return ((de->td_cookie & TMPFS_DIRCOOKIE_DUP) != 0);
363 }
364 
365 static __inline boolean_t
366 tmpfs_dirent_duphead(struct tmpfs_dirent *de)
367 {
368 	return ((de->td_cookie & TMPFS_DIRCOOKIE_DUPHEAD) != 0);
369 }
370 
371 void
372 tmpfs_dirent_init(struct tmpfs_dirent *de, const char *name, u_int namelen)
373 {
374 	de->td_hash = de->td_cookie = tmpfs_dirent_hash(name, namelen);
375 	memcpy(de->ud.td_name, name, namelen);
376 	de->td_namelen = namelen;
377 }
378 
379 /*
380  * Allocates a new directory entry for the node node with a name of name.
381  * The new directory entry is returned in *de.
382  *
383  * The link count of node is increased by one to reflect the new object
384  * referencing it.
385  *
386  * Returns zero on success or an appropriate error code on failure.
387  */
388 int
389 tmpfs_alloc_dirent(struct tmpfs_mount *tmp, struct tmpfs_node *node,
390     const char *name, u_int len, struct tmpfs_dirent **de)
391 {
392 	struct tmpfs_dirent *nde;
393 
394 	nde = uma_zalloc(tmp->tm_dirent_pool, M_WAITOK);
395 	nde->td_node = node;
396 	if (name != NULL) {
397 		nde->ud.td_name = malloc(len, M_TMPFSNAME, M_WAITOK);
398 		tmpfs_dirent_init(nde, name, len);
399 	} else
400 		nde->td_namelen = 0;
401 	if (node != NULL)
402 		node->tn_links++;
403 
404 	*de = nde;
405 
406 	return 0;
407 }
408 
409 /* --------------------------------------------------------------------- */
410 
411 /*
412  * Frees a directory entry.  It is the caller's responsibility to destroy
413  * the node referenced by it if needed.
414  *
415  * The link count of node is decreased by one to reflect the removal of an
416  * object that referenced it.  This only happens if 'node_exists' is true;
417  * otherwise the function will not access the node referred to by the
418  * directory entry, as it may already have been released from the outside.
419  */
420 void
421 tmpfs_free_dirent(struct tmpfs_mount *tmp, struct tmpfs_dirent *de)
422 {
423 	struct tmpfs_node *node;
424 
425 	node = de->td_node;
426 	if (node != NULL) {
427 		MPASS(node->tn_links > 0);
428 		node->tn_links--;
429 	}
430 	if (!tmpfs_dirent_duphead(de) && de->ud.td_name != NULL)
431 		free(de->ud.td_name, M_TMPFSNAME);
432 	uma_zfree(tmp->tm_dirent_pool, de);
433 }
434 
435 /* --------------------------------------------------------------------- */
436 
437 void
438 tmpfs_destroy_vobject(struct vnode *vp, vm_object_t obj)
439 {
440 
441 	if (vp->v_type != VREG || obj == NULL)
442 		return;
443 
444 	VM_OBJECT_WLOCK(obj);
445 	VI_LOCK(vp);
446 	vm_object_clear_flag(obj, OBJ_TMPFS);
447 	obj->un_pager.swp.swp_tmpfs = NULL;
448 	VI_UNLOCK(vp);
449 	VM_OBJECT_WUNLOCK(obj);
450 }
451 
452 /*
453  * Need to clear v_object for insmntque failure.
454  */
455 static void
456 tmpfs_insmntque_dtr(struct vnode *vp, void *dtr_arg)
457 {
458 
459 	tmpfs_destroy_vobject(vp, vp->v_object);
460 	vp->v_object = NULL;
461 	vp->v_data = NULL;
462 	vp->v_op = &dead_vnodeops;
463 	vgone(vp);
464 	vput(vp);
465 }
466 
467 /*
468  * Allocates a new vnode for the node node or returns a new reference to
469  * an existing one if the node had already a vnode referencing it.  The
470  * resulting locked vnode is returned in *vpp.
471  *
472  * Returns zero on success or an appropriate error code on failure.
473  */
474 int
475 tmpfs_alloc_vp(struct mount *mp, struct tmpfs_node *node, int lkflag,
476     struct vnode **vpp)
477 {
478 	struct vnode *vp;
479 	vm_object_t object;
480 	int error;
481 
482 	error = 0;
483 loop:
484 	TMPFS_NODE_LOCK(node);
485 loop1:
486 	if ((vp = node->tn_vnode) != NULL) {
487 		MPASS((node->tn_vpstate & TMPFS_VNODE_DOOMED) == 0);
488 		VI_LOCK(vp);
489 		if ((node->tn_type == VDIR && node->tn_dir.tn_parent == NULL) ||
490 		    ((vp->v_iflag & VI_DOOMED) != 0 &&
491 		    (lkflag & LK_NOWAIT) != 0)) {
492 			VI_UNLOCK(vp);
493 			TMPFS_NODE_UNLOCK(node);
494 			error = ENOENT;
495 			vp = NULL;
496 			goto out;
497 		}
498 		if ((vp->v_iflag & VI_DOOMED) != 0) {
499 			VI_UNLOCK(vp);
500 			node->tn_vpstate |= TMPFS_VNODE_WRECLAIM;
501 			while ((node->tn_vpstate & TMPFS_VNODE_WRECLAIM) != 0) {
502 				msleep(&node->tn_vnode, TMPFS_NODE_MTX(node),
503 				    0, "tmpfsE", 0);
504 			}
505 			goto loop1;
506 		}
507 		TMPFS_NODE_UNLOCK(node);
508 		error = vget(vp, lkflag | LK_INTERLOCK, curthread);
509 		if (error == ENOENT)
510 			goto loop;
511 		if (error != 0) {
512 			vp = NULL;
513 			goto out;
514 		}
515 
516 		/*
517 		 * Make sure the vnode is still there after
518 		 * getting the interlock to avoid racing a free.
519 		 */
520 		if (node->tn_vnode == NULL || node->tn_vnode != vp) {
521 			vput(vp);
522 			goto loop;
523 		}
524 
525 		goto out;
526 	}
527 
528 	if ((node->tn_vpstate & TMPFS_VNODE_DOOMED) ||
529 	    (node->tn_type == VDIR && node->tn_dir.tn_parent == NULL)) {
530 		TMPFS_NODE_UNLOCK(node);
531 		error = ENOENT;
532 		vp = NULL;
533 		goto out;
534 	}
535 
536 	/*
537 	 * otherwise lock the vp list while we call getnewvnode
538 	 * since that can block.
539 	 */
540 	if (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) {
541 		node->tn_vpstate |= TMPFS_VNODE_WANT;
542 		error = msleep((caddr_t) &node->tn_vpstate,
543 		    TMPFS_NODE_MTX(node), PDROP | PCATCH,
544 		    "tmpfs_alloc_vp", 0);
545 		if (error)
546 			return error;
547 
548 		goto loop;
549 	} else
550 		node->tn_vpstate |= TMPFS_VNODE_ALLOCATING;
551 
552 	TMPFS_NODE_UNLOCK(node);
553 
554 	/* Get a new vnode and associate it with our node. */
555 	error = getnewvnode("tmpfs", mp, &tmpfs_vnodeop_entries, &vp);
556 	if (error != 0)
557 		goto unlock;
558 	MPASS(vp != NULL);
559 
560 	(void) vn_lock(vp, lkflag | LK_RETRY);
561 
562 	vp->v_data = node;
563 	vp->v_type = node->tn_type;
564 
565 	/* Type-specific initialization. */
566 	switch (node->tn_type) {
567 	case VBLK:
568 		/* FALLTHROUGH */
569 	case VCHR:
570 		/* FALLTHROUGH */
571 	case VLNK:
572 		/* FALLTHROUGH */
573 	case VSOCK:
574 		break;
575 	case VFIFO:
576 		vp->v_op = &tmpfs_fifoop_entries;
577 		break;
578 	case VREG:
579 		object = node->tn_reg.tn_aobj;
580 		VM_OBJECT_WLOCK(object);
581 		VI_LOCK(vp);
582 		KASSERT(vp->v_object == NULL, ("Not NULL v_object in tmpfs"));
583 		vp->v_object = object;
584 		object->un_pager.swp.swp_tmpfs = vp;
585 		vm_object_set_flag(object, OBJ_TMPFS);
586 		VI_UNLOCK(vp);
587 		VM_OBJECT_WUNLOCK(object);
588 		break;
589 	case VDIR:
590 		MPASS(node->tn_dir.tn_parent != NULL);
591 		if (node->tn_dir.tn_parent == node)
592 			vp->v_vflag |= VV_ROOT;
593 		break;
594 
595 	default:
596 		panic("tmpfs_alloc_vp: type %p %d", node, (int)node->tn_type);
597 	}
598 
599 	error = insmntque1(vp, mp, tmpfs_insmntque_dtr, NULL);
600 	if (error)
601 		vp = NULL;
602 
603 unlock:
604 	TMPFS_NODE_LOCK(node);
605 
606 	MPASS(node->tn_vpstate & TMPFS_VNODE_ALLOCATING);
607 	node->tn_vpstate &= ~TMPFS_VNODE_ALLOCATING;
608 	node->tn_vnode = vp;
609 
610 	if (node->tn_vpstate & TMPFS_VNODE_WANT) {
611 		node->tn_vpstate &= ~TMPFS_VNODE_WANT;
612 		TMPFS_NODE_UNLOCK(node);
613 		wakeup((caddr_t) &node->tn_vpstate);
614 	} else
615 		TMPFS_NODE_UNLOCK(node);
616 
617 out:
618 	*vpp = vp;
619 
620 #ifdef INVARIANTS
621 	if (error == 0) {
622 		MPASS(*vpp != NULL && VOP_ISLOCKED(*vpp));
623 		TMPFS_NODE_LOCK(node);
624 		MPASS(*vpp == node->tn_vnode);
625 		TMPFS_NODE_UNLOCK(node);
626 	}
627 #endif
628 
629 	return error;
630 }
631 
632 /* --------------------------------------------------------------------- */
633 
634 /*
635  * Destroys the association between the vnode vp and the node it
636  * references.
637  */
638 void
639 tmpfs_free_vp(struct vnode *vp)
640 {
641 	struct tmpfs_node *node;
642 
643 	node = VP_TO_TMPFS_NODE(vp);
644 
645 	mtx_assert(TMPFS_NODE_MTX(node), MA_OWNED);
646 	node->tn_vnode = NULL;
647 	if ((node->tn_vpstate & TMPFS_VNODE_WRECLAIM) != 0)
648 		wakeup(&node->tn_vnode);
649 	node->tn_vpstate &= ~TMPFS_VNODE_WRECLAIM;
650 	vp->v_data = NULL;
651 }
652 
653 /* --------------------------------------------------------------------- */
654 
655 /*
656  * Allocates a new file of type 'type' and adds it to the parent directory
657  * 'dvp'; this addition is done using the component name given in 'cnp'.
658  * The ownership of the new file is automatically assigned based on the
659  * credentials of the caller (through 'cnp'), the group is set based on
660  * the parent directory and the mode is determined from the 'vap' argument.
661  * If successful, *vpp holds a vnode to the newly created file and zero
662  * is returned.  Otherwise *vpp is NULL and the function returns an
663  * appropriate error code.
664  */
665 int
666 tmpfs_alloc_file(struct vnode *dvp, struct vnode **vpp, struct vattr *vap,
667     struct componentname *cnp, char *target)
668 {
669 	int error;
670 	struct tmpfs_dirent *de;
671 	struct tmpfs_mount *tmp;
672 	struct tmpfs_node *dnode;
673 	struct tmpfs_node *node;
674 	struct tmpfs_node *parent;
675 
676 	MPASS(VOP_ISLOCKED(dvp));
677 	MPASS(cnp->cn_flags & HASBUF);
678 
679 	tmp = VFS_TO_TMPFS(dvp->v_mount);
680 	dnode = VP_TO_TMPFS_DIR(dvp);
681 	*vpp = NULL;
682 
683 	/* If the entry we are creating is a directory, we cannot overflow
684 	 * the number of links of its parent, because it will get a new
685 	 * link. */
686 	if (vap->va_type == VDIR) {
687 		/* Ensure that we do not overflow the maximum number of links
688 		 * imposed by the system. */
689 		MPASS(dnode->tn_links <= LINK_MAX);
690 		if (dnode->tn_links == LINK_MAX) {
691 			error = EMLINK;
692 			goto out;
693 		}
694 
695 		parent = dnode;
696 		MPASS(parent != NULL);
697 	} else
698 		parent = NULL;
699 
700 	/* Allocate a node that represents the new file. */
701 	error = tmpfs_alloc_node(tmp, vap->va_type, cnp->cn_cred->cr_uid,
702 	    dnode->tn_gid, vap->va_mode, parent, target, vap->va_rdev, &node);
703 	if (error != 0)
704 		goto out;
705 
706 	/* Allocate a directory entry that points to the new file. */
707 	error = tmpfs_alloc_dirent(tmp, node, cnp->cn_nameptr, cnp->cn_namelen,
708 	    &de);
709 	if (error != 0) {
710 		tmpfs_free_node(tmp, node);
711 		goto out;
712 	}
713 
714 	/* Allocate a vnode for the new file. */
715 	error = tmpfs_alloc_vp(dvp->v_mount, node, LK_EXCLUSIVE, vpp);
716 	if (error != 0) {
717 		tmpfs_free_dirent(tmp, de);
718 		tmpfs_free_node(tmp, node);
719 		goto out;
720 	}
721 
722 	/* Now that all required items are allocated, we can proceed to
723 	 * insert the new node into the directory, an operation that
724 	 * cannot fail. */
725 	if (cnp->cn_flags & ISWHITEOUT)
726 		tmpfs_dir_whiteout_remove(dvp, cnp);
727 	tmpfs_dir_attach(dvp, de);
728 
729 out:
730 
731 	return error;
732 }
733 
734 /* --------------------------------------------------------------------- */
735 
736 static struct tmpfs_dirent *
737 tmpfs_dir_first(struct tmpfs_node *dnode, struct tmpfs_dir_cursor *dc)
738 {
739 	struct tmpfs_dirent *de;
740 
741 	de = RB_MIN(tmpfs_dir, &dnode->tn_dir.tn_dirhead);
742 	dc->tdc_tree = de;
743 	if (de != NULL && tmpfs_dirent_duphead(de))
744 		de = LIST_FIRST(&de->ud.td_duphead);
745 	dc->tdc_current = de;
746 
747 	return (dc->tdc_current);
748 }
749 
750 static struct tmpfs_dirent *
751 tmpfs_dir_next(struct tmpfs_node *dnode, struct tmpfs_dir_cursor *dc)
752 {
753 	struct tmpfs_dirent *de;
754 
755 	MPASS(dc->tdc_tree != NULL);
756 	if (tmpfs_dirent_dup(dc->tdc_current)) {
757 		dc->tdc_current = LIST_NEXT(dc->tdc_current, uh.td_dup.entries);
758 		if (dc->tdc_current != NULL)
759 			return (dc->tdc_current);
760 	}
761 	dc->tdc_tree = dc->tdc_current = RB_NEXT(tmpfs_dir,
762 	    &dnode->tn_dir.tn_dirhead, dc->tdc_tree);
763 	if ((de = dc->tdc_current) != NULL && tmpfs_dirent_duphead(de)) {
764 		dc->tdc_current = LIST_FIRST(&de->ud.td_duphead);
765 		MPASS(dc->tdc_current != NULL);
766 	}
767 
768 	return (dc->tdc_current);
769 }
770 
771 /* Lookup directory entry in RB-Tree. Function may return duphead entry. */
772 static struct tmpfs_dirent *
773 tmpfs_dir_xlookup_hash(struct tmpfs_node *dnode, uint32_t hash)
774 {
775 	struct tmpfs_dirent *de, dekey;
776 
777 	dekey.td_hash = hash;
778 	de = RB_FIND(tmpfs_dir, &dnode->tn_dir.tn_dirhead, &dekey);
779 	return (de);
780 }
781 
782 /* Lookup directory entry by cookie, initialize directory cursor accordingly. */
783 static struct tmpfs_dirent *
784 tmpfs_dir_lookup_cookie(struct tmpfs_node *node, off_t cookie,
785     struct tmpfs_dir_cursor *dc)
786 {
787 	struct tmpfs_dir *dirhead = &node->tn_dir.tn_dirhead;
788 	struct tmpfs_dirent *de, dekey;
789 
790 	MPASS(cookie >= TMPFS_DIRCOOKIE_MIN);
791 
792 	if (cookie == node->tn_dir.tn_readdir_lastn &&
793 	    (de = node->tn_dir.tn_readdir_lastp) != NULL) {
794 		/* Protect against possible race, tn_readdir_last[pn]
795 		 * may be updated with only shared vnode lock held. */
796 		if (cookie == tmpfs_dirent_cookie(de))
797 			goto out;
798 	}
799 
800 	if ((cookie & TMPFS_DIRCOOKIE_DUP) != 0) {
801 		LIST_FOREACH(de, &node->tn_dir.tn_dupindex,
802 		    uh.td_dup.index_entries) {
803 			MPASS(tmpfs_dirent_dup(de));
804 			if (de->td_cookie == cookie)
805 				goto out;
806 			/* dupindex list is sorted. */
807 			if (de->td_cookie < cookie) {
808 				de = NULL;
809 				goto out;
810 			}
811 		}
812 		MPASS(de == NULL);
813 		goto out;
814 	}
815 
816 	MPASS((cookie & TMPFS_DIRCOOKIE_MASK) == cookie);
817 	dekey.td_hash = cookie;
818 	/* Recover if direntry for cookie was removed */
819 	de = RB_NFIND(tmpfs_dir, dirhead, &dekey);
820 	dc->tdc_tree = de;
821 	dc->tdc_current = de;
822 	if (de != NULL && tmpfs_dirent_duphead(de)) {
823 		dc->tdc_current = LIST_FIRST(&de->ud.td_duphead);
824 		MPASS(dc->tdc_current != NULL);
825 	}
826 	return (dc->tdc_current);
827 
828 out:
829 	dc->tdc_tree = de;
830 	dc->tdc_current = de;
831 	if (de != NULL && tmpfs_dirent_dup(de))
832 		dc->tdc_tree = tmpfs_dir_xlookup_hash(node,
833 		    de->td_hash);
834 	return (dc->tdc_current);
835 }
836 
837 /*
838  * Looks for a directory entry in the directory represented by node.
839  * 'cnp' describes the name of the entry to look for.  Note that the .
840  * and .. components are not allowed as they do not physically exist
841  * within directories.
842  *
843  * Returns a pointer to the entry when found, otherwise NULL.
844  */
845 struct tmpfs_dirent *
846 tmpfs_dir_lookup(struct tmpfs_node *node, struct tmpfs_node *f,
847     struct componentname *cnp)
848 {
849 	struct tmpfs_dir_duphead *duphead;
850 	struct tmpfs_dirent *de;
851 	uint32_t hash;
852 
853 	MPASS(IMPLIES(cnp->cn_namelen == 1, cnp->cn_nameptr[0] != '.'));
854 	MPASS(IMPLIES(cnp->cn_namelen == 2, !(cnp->cn_nameptr[0] == '.' &&
855 	    cnp->cn_nameptr[1] == '.')));
856 	TMPFS_VALIDATE_DIR(node);
857 
858 	hash = tmpfs_dirent_hash(cnp->cn_nameptr, cnp->cn_namelen);
859 	de = tmpfs_dir_xlookup_hash(node, hash);
860 	if (de != NULL && tmpfs_dirent_duphead(de)) {
861 		duphead = &de->ud.td_duphead;
862 		LIST_FOREACH(de, duphead, uh.td_dup.entries) {
863 			if (TMPFS_DIRENT_MATCHES(de, cnp->cn_nameptr,
864 			    cnp->cn_namelen))
865 				break;
866 		}
867 	} else if (de != NULL) {
868 		if (!TMPFS_DIRENT_MATCHES(de, cnp->cn_nameptr,
869 		    cnp->cn_namelen))
870 			de = NULL;
871 	}
872 	if (de != NULL && f != NULL && de->td_node != f)
873 		de = NULL;
874 
875 	return (de);
876 }
877 
878 /*
879  * Attach duplicate-cookie directory entry nde to dnode and insert to dupindex
880  * list, allocate new cookie value.
881  */
882 static void
883 tmpfs_dir_attach_dup(struct tmpfs_node *dnode,
884     struct tmpfs_dir_duphead *duphead, struct tmpfs_dirent *nde)
885 {
886 	struct tmpfs_dir_duphead *dupindex;
887 	struct tmpfs_dirent *de, *pde;
888 
889 	dupindex = &dnode->tn_dir.tn_dupindex;
890 	de = LIST_FIRST(dupindex);
891 	if (de == NULL || de->td_cookie < TMPFS_DIRCOOKIE_DUP_MAX) {
892 		if (de == NULL)
893 			nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MIN;
894 		else
895 			nde->td_cookie = de->td_cookie + 1;
896 		MPASS(tmpfs_dirent_dup(nde));
897 		LIST_INSERT_HEAD(dupindex, nde, uh.td_dup.index_entries);
898 		LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries);
899 		return;
900 	}
901 
902 	/*
903 	 * Cookie numbers are near exhaustion. Scan dupindex list for unused
904 	 * numbers. dupindex list is sorted in descending order. Keep it so
905 	 * after inserting nde.
906 	 */
907 	while (1) {
908 		pde = de;
909 		de = LIST_NEXT(de, uh.td_dup.index_entries);
910 		if (de == NULL && pde->td_cookie != TMPFS_DIRCOOKIE_DUP_MIN) {
911 			/*
912 			 * Last element of the index doesn't have minimal cookie
913 			 * value, use it.
914 			 */
915 			nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MIN;
916 			LIST_INSERT_AFTER(pde, nde, uh.td_dup.index_entries);
917 			LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries);
918 			return;
919 		} else if (de == NULL) {
920 			/*
921 			 * We are so lucky have 2^30 hash duplicates in single
922 			 * directory :) Return largest possible cookie value.
923 			 * It should be fine except possible issues with
924 			 * VOP_READDIR restart.
925 			 */
926 			nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MAX;
927 			LIST_INSERT_HEAD(dupindex, nde,
928 			    uh.td_dup.index_entries);
929 			LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries);
930 			return;
931 		}
932 		if (de->td_cookie + 1 == pde->td_cookie ||
933 		    de->td_cookie >= TMPFS_DIRCOOKIE_DUP_MAX)
934 			continue;	/* No hole or invalid cookie. */
935 		nde->td_cookie = de->td_cookie + 1;
936 		MPASS(tmpfs_dirent_dup(nde));
937 		MPASS(pde->td_cookie > nde->td_cookie);
938 		MPASS(nde->td_cookie > de->td_cookie);
939 		LIST_INSERT_BEFORE(de, nde, uh.td_dup.index_entries);
940 		LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries);
941 		return;
942 	};
943 }
944 
945 /*
946  * Attaches the directory entry de to the directory represented by vp.
947  * Note that this does not change the link count of the node pointed by
948  * the directory entry, as this is done by tmpfs_alloc_dirent.
949  */
950 void
951 tmpfs_dir_attach(struct vnode *vp, struct tmpfs_dirent *de)
952 {
953 	struct tmpfs_node *dnode;
954 	struct tmpfs_dirent *xde, *nde;
955 
956 	ASSERT_VOP_ELOCKED(vp, __func__);
957 	MPASS(de->td_namelen > 0);
958 	MPASS(de->td_hash >= TMPFS_DIRCOOKIE_MIN);
959 	MPASS(de->td_cookie == de->td_hash);
960 
961 	dnode = VP_TO_TMPFS_DIR(vp);
962 	dnode->tn_dir.tn_readdir_lastn = 0;
963 	dnode->tn_dir.tn_readdir_lastp = NULL;
964 
965 	MPASS(!tmpfs_dirent_dup(de));
966 	xde = RB_INSERT(tmpfs_dir, &dnode->tn_dir.tn_dirhead, de);
967 	if (xde != NULL && tmpfs_dirent_duphead(xde))
968 		tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, de);
969 	else if (xde != NULL) {
970 		/*
971 		 * Allocate new duphead. Swap xde with duphead to avoid
972 		 * adding/removing elements with the same hash.
973 		 */
974 		MPASS(!tmpfs_dirent_dup(xde));
975 		tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), NULL, NULL, 0,
976 		    &nde);
977 		/* *nde = *xde; XXX gcc 4.2.1 may generate invalid code. */
978 		memcpy(nde, xde, sizeof(*xde));
979 		xde->td_cookie |= TMPFS_DIRCOOKIE_DUPHEAD;
980 		LIST_INIT(&xde->ud.td_duphead);
981 		xde->td_namelen = 0;
982 		xde->td_node = NULL;
983 		tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, nde);
984 		tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, de);
985 	}
986 	dnode->tn_size += sizeof(struct tmpfs_dirent);
987 	dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
988 	    TMPFS_NODE_MODIFIED;
989 }
990 
991 /* --------------------------------------------------------------------- */
992 
993 /*
994  * Detaches the directory entry de from the directory represented by vp.
995  * Note that this does not change the link count of the node pointed by
996  * the directory entry, as this is done by tmpfs_free_dirent.
997  */
998 void
999 tmpfs_dir_detach(struct vnode *vp, struct tmpfs_dirent *de)
1000 {
1001 	struct tmpfs_mount *tmp;
1002 	struct tmpfs_dir *head;
1003 	struct tmpfs_node *dnode;
1004 	struct tmpfs_dirent *xde;
1005 
1006 	ASSERT_VOP_ELOCKED(vp, __func__);
1007 
1008 	dnode = VP_TO_TMPFS_DIR(vp);
1009 	head = &dnode->tn_dir.tn_dirhead;
1010 	dnode->tn_dir.tn_readdir_lastn = 0;
1011 	dnode->tn_dir.tn_readdir_lastp = NULL;
1012 
1013 	if (tmpfs_dirent_dup(de)) {
1014 		/* Remove duphead if de was last entry. */
1015 		if (LIST_NEXT(de, uh.td_dup.entries) == NULL) {
1016 			xde = tmpfs_dir_xlookup_hash(dnode, de->td_hash);
1017 			MPASS(tmpfs_dirent_duphead(xde));
1018 		} else
1019 			xde = NULL;
1020 		LIST_REMOVE(de, uh.td_dup.entries);
1021 		LIST_REMOVE(de, uh.td_dup.index_entries);
1022 		if (xde != NULL) {
1023 			if (LIST_EMPTY(&xde->ud.td_duphead)) {
1024 				RB_REMOVE(tmpfs_dir, head, xde);
1025 				tmp = VFS_TO_TMPFS(vp->v_mount);
1026 				MPASS(xde->td_node == NULL);
1027 				tmpfs_free_dirent(tmp, xde);
1028 			}
1029 		}
1030 	} else
1031 		RB_REMOVE(tmpfs_dir, head, de);
1032 
1033 	dnode->tn_size -= sizeof(struct tmpfs_dirent);
1034 	dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
1035 	    TMPFS_NODE_MODIFIED;
1036 }
1037 
1038 void
1039 tmpfs_dir_destroy(struct tmpfs_mount *tmp, struct tmpfs_node *dnode)
1040 {
1041 	struct tmpfs_dirent *de, *dde, *nde;
1042 
1043 	RB_FOREACH_SAFE(de, tmpfs_dir, &dnode->tn_dir.tn_dirhead, nde) {
1044 		RB_REMOVE(tmpfs_dir, &dnode->tn_dir.tn_dirhead, de);
1045 		/* Node may already be destroyed. */
1046 		de->td_node = NULL;
1047 		if (tmpfs_dirent_duphead(de)) {
1048 			while ((dde = LIST_FIRST(&de->ud.td_duphead)) != NULL) {
1049 				LIST_REMOVE(dde, uh.td_dup.entries);
1050 				dde->td_node = NULL;
1051 				tmpfs_free_dirent(tmp, dde);
1052 			}
1053 		}
1054 		tmpfs_free_dirent(tmp, de);
1055 	}
1056 }
1057 
1058 /* --------------------------------------------------------------------- */
1059 
1060 /*
1061  * Helper function for tmpfs_readdir.  Creates a '.' entry for the given
1062  * directory and returns it in the uio space.  The function returns 0
1063  * on success, -1 if there was not enough space in the uio structure to
1064  * hold the directory entry or an appropriate error code if another
1065  * error happens.
1066  */
1067 static int
1068 tmpfs_dir_getdotdent(struct tmpfs_node *node, struct uio *uio)
1069 {
1070 	int error;
1071 	struct dirent dent;
1072 
1073 	TMPFS_VALIDATE_DIR(node);
1074 	MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOT);
1075 
1076 	dent.d_fileno = node->tn_id;
1077 	dent.d_type = DT_DIR;
1078 	dent.d_namlen = 1;
1079 	dent.d_name[0] = '.';
1080 	dent.d_name[1] = '\0';
1081 	dent.d_reclen = GENERIC_DIRSIZ(&dent);
1082 
1083 	if (dent.d_reclen > uio->uio_resid)
1084 		error = EJUSTRETURN;
1085 	else
1086 		error = uiomove(&dent, dent.d_reclen, uio);
1087 
1088 	node->tn_status |= TMPFS_NODE_ACCESSED;
1089 
1090 	return error;
1091 }
1092 
1093 /* --------------------------------------------------------------------- */
1094 
1095 /*
1096  * Helper function for tmpfs_readdir.  Creates a '..' entry for the given
1097  * directory and returns it in the uio space.  The function returns 0
1098  * on success, -1 if there was not enough space in the uio structure to
1099  * hold the directory entry or an appropriate error code if another
1100  * error happens.
1101  */
1102 static int
1103 tmpfs_dir_getdotdotdent(struct tmpfs_node *node, struct uio *uio)
1104 {
1105 	int error;
1106 	struct dirent dent;
1107 
1108 	TMPFS_VALIDATE_DIR(node);
1109 	MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT);
1110 
1111 	/*
1112 	 * Return ENOENT if the current node is already removed.
1113 	 */
1114 	TMPFS_ASSERT_LOCKED(node);
1115 	if (node->tn_dir.tn_parent == NULL) {
1116 		return (ENOENT);
1117 	}
1118 
1119 	TMPFS_NODE_LOCK(node->tn_dir.tn_parent);
1120 	dent.d_fileno = node->tn_dir.tn_parent->tn_id;
1121 	TMPFS_NODE_UNLOCK(node->tn_dir.tn_parent);
1122 
1123 	dent.d_type = DT_DIR;
1124 	dent.d_namlen = 2;
1125 	dent.d_name[0] = '.';
1126 	dent.d_name[1] = '.';
1127 	dent.d_name[2] = '\0';
1128 	dent.d_reclen = GENERIC_DIRSIZ(&dent);
1129 
1130 	if (dent.d_reclen > uio->uio_resid)
1131 		error = EJUSTRETURN;
1132 	else
1133 		error = uiomove(&dent, dent.d_reclen, uio);
1134 
1135 	node->tn_status |= TMPFS_NODE_ACCESSED;
1136 
1137 	return error;
1138 }
1139 
1140 /* --------------------------------------------------------------------- */
1141 
1142 /*
1143  * Helper function for tmpfs_readdir.  Returns as much directory entries
1144  * as can fit in the uio space.  The read starts at uio->uio_offset.
1145  * The function returns 0 on success, -1 if there was not enough space
1146  * in the uio structure to hold the directory entry or an appropriate
1147  * error code if another error happens.
1148  */
1149 int
1150 tmpfs_dir_getdents(struct tmpfs_node *node, struct uio *uio, int cnt,
1151     u_long *cookies, int *ncookies)
1152 {
1153 	struct tmpfs_dir_cursor dc;
1154 	struct tmpfs_dirent *de;
1155 	off_t off;
1156 	int error;
1157 
1158 	TMPFS_VALIDATE_DIR(node);
1159 
1160 	off = 0;
1161 
1162 	/*
1163 	 * Lookup the node from the current offset.  The starting offset of
1164 	 * 0 will lookup both '.' and '..', and then the first real entry,
1165 	 * or EOF if there are none.  Then find all entries for the dir that
1166 	 * fit into the buffer.  Once no more entries are found (de == NULL),
1167 	 * the offset is set to TMPFS_DIRCOOKIE_EOF, which will cause the next
1168 	 * call to return 0.
1169 	 */
1170 	switch (uio->uio_offset) {
1171 	case TMPFS_DIRCOOKIE_DOT:
1172 		error = tmpfs_dir_getdotdent(node, uio);
1173 		if (error != 0)
1174 			return (error);
1175 		uio->uio_offset = TMPFS_DIRCOOKIE_DOTDOT;
1176 		if (cnt != 0)
1177 			cookies[(*ncookies)++] = off = uio->uio_offset;
1178 	case TMPFS_DIRCOOKIE_DOTDOT:
1179 		error = tmpfs_dir_getdotdotdent(node, uio);
1180 		if (error != 0)
1181 			return (error);
1182 		de = tmpfs_dir_first(node, &dc);
1183 		uio->uio_offset = tmpfs_dirent_cookie(de);
1184 		if (cnt != 0)
1185 			cookies[(*ncookies)++] = off = uio->uio_offset;
1186 		/* EOF. */
1187 		if (de == NULL)
1188 			return (0);
1189 		break;
1190 	case TMPFS_DIRCOOKIE_EOF:
1191 		return (0);
1192 	default:
1193 		de = tmpfs_dir_lookup_cookie(node, uio->uio_offset, &dc);
1194 		if (de == NULL)
1195 			return (EINVAL);
1196 		if (cnt != 0)
1197 			off = tmpfs_dirent_cookie(de);
1198 	}
1199 
1200 	/* Read as much entries as possible; i.e., until we reach the end of
1201 	 * the directory or we exhaust uio space. */
1202 	do {
1203 		struct dirent d;
1204 
1205 		/* Create a dirent structure representing the current
1206 		 * tmpfs_node and fill it. */
1207 		if (de->td_node == NULL) {
1208 			d.d_fileno = 1;
1209 			d.d_type = DT_WHT;
1210 		} else {
1211 			d.d_fileno = de->td_node->tn_id;
1212 			switch (de->td_node->tn_type) {
1213 			case VBLK:
1214 				d.d_type = DT_BLK;
1215 				break;
1216 
1217 			case VCHR:
1218 				d.d_type = DT_CHR;
1219 				break;
1220 
1221 			case VDIR:
1222 				d.d_type = DT_DIR;
1223 				break;
1224 
1225 			case VFIFO:
1226 				d.d_type = DT_FIFO;
1227 				break;
1228 
1229 			case VLNK:
1230 				d.d_type = DT_LNK;
1231 				break;
1232 
1233 			case VREG:
1234 				d.d_type = DT_REG;
1235 				break;
1236 
1237 			case VSOCK:
1238 				d.d_type = DT_SOCK;
1239 				break;
1240 
1241 			default:
1242 				panic("tmpfs_dir_getdents: type %p %d",
1243 				    de->td_node, (int)de->td_node->tn_type);
1244 			}
1245 		}
1246 		d.d_namlen = de->td_namelen;
1247 		MPASS(de->td_namelen < sizeof(d.d_name));
1248 		(void)memcpy(d.d_name, de->ud.td_name, de->td_namelen);
1249 		d.d_name[de->td_namelen] = '\0';
1250 		d.d_reclen = GENERIC_DIRSIZ(&d);
1251 
1252 		/* Stop reading if the directory entry we are treating is
1253 		 * bigger than the amount of data that can be returned. */
1254 		if (d.d_reclen > uio->uio_resid) {
1255 			error = EJUSTRETURN;
1256 			break;
1257 		}
1258 
1259 		/* Copy the new dirent structure into the output buffer and
1260 		 * advance pointers. */
1261 		error = uiomove(&d, d.d_reclen, uio);
1262 		if (error == 0) {
1263 			de = tmpfs_dir_next(node, &dc);
1264 			if (cnt != 0) {
1265 				off = tmpfs_dirent_cookie(de);
1266 				MPASS(*ncookies < cnt);
1267 				cookies[(*ncookies)++] = off;
1268 			}
1269 		}
1270 	} while (error == 0 && uio->uio_resid > 0 && de != NULL);
1271 
1272 	/* Update the offset and cache. */
1273 	if (cnt == 0)
1274 		off = tmpfs_dirent_cookie(de);
1275 
1276 	uio->uio_offset = off;
1277 	node->tn_dir.tn_readdir_lastn = off;
1278 	node->tn_dir.tn_readdir_lastp = de;
1279 
1280 	node->tn_status |= TMPFS_NODE_ACCESSED;
1281 	return error;
1282 }
1283 
1284 int
1285 tmpfs_dir_whiteout_add(struct vnode *dvp, struct componentname *cnp)
1286 {
1287 	struct tmpfs_dirent *de;
1288 	int error;
1289 
1290 	error = tmpfs_alloc_dirent(VFS_TO_TMPFS(dvp->v_mount), NULL,
1291 	    cnp->cn_nameptr, cnp->cn_namelen, &de);
1292 	if (error != 0)
1293 		return (error);
1294 	tmpfs_dir_attach(dvp, de);
1295 	return (0);
1296 }
1297 
1298 void
1299 tmpfs_dir_whiteout_remove(struct vnode *dvp, struct componentname *cnp)
1300 {
1301 	struct tmpfs_dirent *de;
1302 
1303 	de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(dvp), NULL, cnp);
1304 	MPASS(de != NULL && de->td_node == NULL);
1305 	tmpfs_dir_detach(dvp, de);
1306 	tmpfs_free_dirent(VFS_TO_TMPFS(dvp->v_mount), de);
1307 }
1308 
1309 /* --------------------------------------------------------------------- */
1310 
1311 /*
1312  * Resizes the aobj associated with the regular file pointed to by 'vp' to the
1313  * size 'newsize'.  'vp' must point to a vnode that represents a regular file.
1314  * 'newsize' must be positive.
1315  *
1316  * Returns zero on success or an appropriate error code on failure.
1317  */
1318 int
1319 tmpfs_reg_resize(struct vnode *vp, off_t newsize, boolean_t ignerr)
1320 {
1321 	struct tmpfs_mount *tmp;
1322 	struct tmpfs_node *node;
1323 	vm_object_t uobj;
1324 	vm_page_t m, ma[1];
1325 	vm_pindex_t idx, newpages, oldpages;
1326 	off_t oldsize;
1327 	int base, rv;
1328 
1329 	MPASS(vp->v_type == VREG);
1330 	MPASS(newsize >= 0);
1331 
1332 	node = VP_TO_TMPFS_NODE(vp);
1333 	uobj = node->tn_reg.tn_aobj;
1334 	tmp = VFS_TO_TMPFS(vp->v_mount);
1335 
1336 	/*
1337 	 * Convert the old and new sizes to the number of pages needed to
1338 	 * store them.  It may happen that we do not need to do anything
1339 	 * because the last allocated page can accommodate the change on
1340 	 * its own.
1341 	 */
1342 	oldsize = node->tn_size;
1343 	oldpages = OFF_TO_IDX(oldsize + PAGE_MASK);
1344 	MPASS(oldpages == uobj->size);
1345 	newpages = OFF_TO_IDX(newsize + PAGE_MASK);
1346 	if (newpages > oldpages &&
1347 	    tmpfs_pages_check_avail(tmp, newpages - oldpages) == 0)
1348 		return (ENOSPC);
1349 
1350 	VM_OBJECT_WLOCK(uobj);
1351 	if (newsize < oldsize) {
1352 		/*
1353 		 * Zero the truncated part of the last page.
1354 		 */
1355 		base = newsize & PAGE_MASK;
1356 		if (base != 0) {
1357 			idx = OFF_TO_IDX(newsize);
1358 retry:
1359 			m = vm_page_lookup(uobj, idx);
1360 			if (m != NULL) {
1361 				if (vm_page_sleep_if_busy(m, "tmfssz"))
1362 					goto retry;
1363 				MPASS(m->valid == VM_PAGE_BITS_ALL);
1364 			} else if (vm_pager_has_page(uobj, idx, NULL, NULL)) {
1365 				m = vm_page_alloc(uobj, idx, VM_ALLOC_NORMAL);
1366 				if (m == NULL) {
1367 					VM_OBJECT_WUNLOCK(uobj);
1368 					VM_WAIT;
1369 					VM_OBJECT_WLOCK(uobj);
1370 					goto retry;
1371 				} else if (m->valid != VM_PAGE_BITS_ALL) {
1372 					ma[0] = m;
1373 					rv = vm_pager_get_pages(uobj, ma, 1, 0);
1374 					m = vm_page_lookup(uobj, idx);
1375 				} else
1376 					/* A cached page was reactivated. */
1377 					rv = VM_PAGER_OK;
1378 				vm_page_lock(m);
1379 				if (rv == VM_PAGER_OK) {
1380 					vm_page_deactivate(m);
1381 					vm_page_unlock(m);
1382 					vm_page_xunbusy(m);
1383 				} else {
1384 					vm_page_free(m);
1385 					vm_page_unlock(m);
1386 					if (ignerr)
1387 						m = NULL;
1388 					else {
1389 						VM_OBJECT_WUNLOCK(uobj);
1390 						return (EIO);
1391 					}
1392 				}
1393 			}
1394 			if (m != NULL) {
1395 				pmap_zero_page_area(m, base, PAGE_SIZE - base);
1396 				vm_page_dirty(m);
1397 				vm_pager_page_unswapped(m);
1398 			}
1399 		}
1400 
1401 		/*
1402 		 * Release any swap space and free any whole pages.
1403 		 */
1404 		if (newpages < oldpages) {
1405 			swap_pager_freespace(uobj, newpages, oldpages -
1406 			    newpages);
1407 			vm_object_page_remove(uobj, newpages, 0, 0);
1408 		}
1409 	}
1410 	uobj->size = newpages;
1411 	VM_OBJECT_WUNLOCK(uobj);
1412 
1413 	TMPFS_LOCK(tmp);
1414 	tmp->tm_pages_used += (newpages - oldpages);
1415 	TMPFS_UNLOCK(tmp);
1416 
1417 	node->tn_size = newsize;
1418 	return (0);
1419 }
1420 
1421 /* --------------------------------------------------------------------- */
1422 
1423 /*
1424  * Change flags of the given vnode.
1425  * Caller should execute tmpfs_update on vp after a successful execution.
1426  * The vnode must be locked on entry and remain locked on exit.
1427  */
1428 int
1429 tmpfs_chflags(struct vnode *vp, u_long flags, struct ucred *cred,
1430     struct thread *p)
1431 {
1432 	int error;
1433 	struct tmpfs_node *node;
1434 
1435 	MPASS(VOP_ISLOCKED(vp));
1436 
1437 	node = VP_TO_TMPFS_NODE(vp);
1438 
1439 	if ((flags & ~(SF_APPEND | SF_ARCHIVED | SF_IMMUTABLE | SF_NOUNLINK |
1440 	    UF_APPEND | UF_ARCHIVE | UF_HIDDEN | UF_IMMUTABLE | UF_NODUMP |
1441 	    UF_NOUNLINK | UF_OFFLINE | UF_OPAQUE | UF_READONLY | UF_REPARSE |
1442 	    UF_SPARSE | UF_SYSTEM)) != 0)
1443 		return (EOPNOTSUPP);
1444 
1445 	/* Disallow this operation if the file system is mounted read-only. */
1446 	if (vp->v_mount->mnt_flag & MNT_RDONLY)
1447 		return EROFS;
1448 
1449 	/*
1450 	 * Callers may only modify the file flags on objects they
1451 	 * have VADMIN rights for.
1452 	 */
1453 	if ((error = VOP_ACCESS(vp, VADMIN, cred, p)))
1454 		return (error);
1455 	/*
1456 	 * Unprivileged processes are not permitted to unset system
1457 	 * flags, or modify flags if any system flags are set.
1458 	 */
1459 	if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0)) {
1460 		if (node->tn_flags &
1461 		    (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND)) {
1462 			error = securelevel_gt(cred, 0);
1463 			if (error)
1464 				return (error);
1465 		}
1466 	} else {
1467 		if (node->tn_flags &
1468 		    (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND) ||
1469 		    ((flags ^ node->tn_flags) & SF_SETTABLE))
1470 			return (EPERM);
1471 	}
1472 	node->tn_flags = flags;
1473 	node->tn_status |= TMPFS_NODE_CHANGED;
1474 
1475 	MPASS(VOP_ISLOCKED(vp));
1476 
1477 	return 0;
1478 }
1479 
1480 /* --------------------------------------------------------------------- */
1481 
1482 /*
1483  * Change access mode on the given vnode.
1484  * Caller should execute tmpfs_update on vp after a successful execution.
1485  * The vnode must be locked on entry and remain locked on exit.
1486  */
1487 int
1488 tmpfs_chmod(struct vnode *vp, mode_t mode, struct ucred *cred, struct thread *p)
1489 {
1490 	int error;
1491 	struct tmpfs_node *node;
1492 
1493 	MPASS(VOP_ISLOCKED(vp));
1494 
1495 	node = VP_TO_TMPFS_NODE(vp);
1496 
1497 	/* Disallow this operation if the file system is mounted read-only. */
1498 	if (vp->v_mount->mnt_flag & MNT_RDONLY)
1499 		return EROFS;
1500 
1501 	/* Immutable or append-only files cannot be modified, either. */
1502 	if (node->tn_flags & (IMMUTABLE | APPEND))
1503 		return EPERM;
1504 
1505 	/*
1506 	 * To modify the permissions on a file, must possess VADMIN
1507 	 * for that file.
1508 	 */
1509 	if ((error = VOP_ACCESS(vp, VADMIN, cred, p)))
1510 		return (error);
1511 
1512 	/*
1513 	 * Privileged processes may set the sticky bit on non-directories,
1514 	 * as well as set the setgid bit on a file with a group that the
1515 	 * process is not a member of.
1516 	 */
1517 	if (vp->v_type != VDIR && (mode & S_ISTXT)) {
1518 		if (priv_check_cred(cred, PRIV_VFS_STICKYFILE, 0))
1519 			return (EFTYPE);
1520 	}
1521 	if (!groupmember(node->tn_gid, cred) && (mode & S_ISGID)) {
1522 		error = priv_check_cred(cred, PRIV_VFS_SETGID, 0);
1523 		if (error)
1524 			return (error);
1525 	}
1526 
1527 
1528 	node->tn_mode &= ~ALLPERMS;
1529 	node->tn_mode |= mode & ALLPERMS;
1530 
1531 	node->tn_status |= TMPFS_NODE_CHANGED;
1532 
1533 	MPASS(VOP_ISLOCKED(vp));
1534 
1535 	return 0;
1536 }
1537 
1538 /* --------------------------------------------------------------------- */
1539 
1540 /*
1541  * Change ownership of the given vnode.  At least one of uid or gid must
1542  * be different than VNOVAL.  If one is set to that value, the attribute
1543  * is unchanged.
1544  * Caller should execute tmpfs_update on vp after a successful execution.
1545  * The vnode must be locked on entry and remain locked on exit.
1546  */
1547 int
1548 tmpfs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred,
1549     struct thread *p)
1550 {
1551 	int error;
1552 	struct tmpfs_node *node;
1553 	uid_t ouid;
1554 	gid_t ogid;
1555 
1556 	MPASS(VOP_ISLOCKED(vp));
1557 
1558 	node = VP_TO_TMPFS_NODE(vp);
1559 
1560 	/* Assign default values if they are unknown. */
1561 	MPASS(uid != VNOVAL || gid != VNOVAL);
1562 	if (uid == VNOVAL)
1563 		uid = node->tn_uid;
1564 	if (gid == VNOVAL)
1565 		gid = node->tn_gid;
1566 	MPASS(uid != VNOVAL && gid != VNOVAL);
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 ownership of a file, must possess VADMIN for that
1578 	 * file.
1579 	 */
1580 	if ((error = VOP_ACCESS(vp, VADMIN, cred, p)))
1581 		return (error);
1582 
1583 	/*
1584 	 * To change the owner of a file, or change the group of a file to a
1585 	 * group of which we are not a member, the caller must have
1586 	 * privilege.
1587 	 */
1588 	if ((uid != node->tn_uid ||
1589 	    (gid != node->tn_gid && !groupmember(gid, cred))) &&
1590 	    (error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0)))
1591 		return (error);
1592 
1593 	ogid = node->tn_gid;
1594 	ouid = node->tn_uid;
1595 
1596 	node->tn_uid = uid;
1597 	node->tn_gid = gid;
1598 
1599 	node->tn_status |= TMPFS_NODE_CHANGED;
1600 
1601 	if ((node->tn_mode & (S_ISUID | S_ISGID)) && (ouid != uid || ogid != gid)) {
1602 		if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID, 0))
1603 			node->tn_mode &= ~(S_ISUID | S_ISGID);
1604 	}
1605 
1606 	MPASS(VOP_ISLOCKED(vp));
1607 
1608 	return 0;
1609 }
1610 
1611 /* --------------------------------------------------------------------- */
1612 
1613 /*
1614  * Change size of the given vnode.
1615  * Caller should execute tmpfs_update on vp after a successful execution.
1616  * The vnode must be locked on entry and remain locked on exit.
1617  */
1618 int
1619 tmpfs_chsize(struct vnode *vp, u_quad_t size, struct ucred *cred,
1620     struct thread *p)
1621 {
1622 	int error;
1623 	struct tmpfs_node *node;
1624 
1625 	MPASS(VOP_ISLOCKED(vp));
1626 
1627 	node = VP_TO_TMPFS_NODE(vp);
1628 
1629 	/* Decide whether this is a valid operation based on the file type. */
1630 	error = 0;
1631 	switch (vp->v_type) {
1632 	case VDIR:
1633 		return EISDIR;
1634 
1635 	case VREG:
1636 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
1637 			return EROFS;
1638 		break;
1639 
1640 	case VBLK:
1641 		/* FALLTHROUGH */
1642 	case VCHR:
1643 		/* FALLTHROUGH */
1644 	case VFIFO:
1645 		/* Allow modifications of special files even if in the file
1646 		 * system is mounted read-only (we are not modifying the
1647 		 * files themselves, but the objects they represent). */
1648 		return 0;
1649 
1650 	default:
1651 		/* Anything else is unsupported. */
1652 		return EOPNOTSUPP;
1653 	}
1654 
1655 	/* Immutable or append-only files cannot be modified, either. */
1656 	if (node->tn_flags & (IMMUTABLE | APPEND))
1657 		return EPERM;
1658 
1659 	error = tmpfs_truncate(vp, size);
1660 	/* tmpfs_truncate will raise the NOTE_EXTEND and NOTE_ATTRIB kevents
1661 	 * for us, as will update tn_status; no need to do that here. */
1662 
1663 	MPASS(VOP_ISLOCKED(vp));
1664 
1665 	return error;
1666 }
1667 
1668 /* --------------------------------------------------------------------- */
1669 
1670 /*
1671  * Change access and modification times of the given vnode.
1672  * Caller should execute tmpfs_update on vp after a successful execution.
1673  * The vnode must be locked on entry and remain locked on exit.
1674  */
1675 int
1676 tmpfs_chtimes(struct vnode *vp, struct timespec *atime, struct timespec *mtime,
1677 	struct timespec *birthtime, int vaflags, struct ucred *cred, struct thread *l)
1678 {
1679 	int error;
1680 	struct tmpfs_node *node;
1681 
1682 	MPASS(VOP_ISLOCKED(vp));
1683 
1684 	node = VP_TO_TMPFS_NODE(vp);
1685 
1686 	/* Disallow this operation if the file system is mounted read-only. */
1687 	if (vp->v_mount->mnt_flag & MNT_RDONLY)
1688 		return EROFS;
1689 
1690 	/* Immutable or append-only files cannot be modified, either. */
1691 	if (node->tn_flags & (IMMUTABLE | APPEND))
1692 		return EPERM;
1693 
1694 	/* Determine if the user have proper privilege to update time. */
1695 	if (vaflags & VA_UTIMES_NULL) {
1696 		error = VOP_ACCESS(vp, VADMIN, cred, l);
1697 		if (error)
1698 			error = VOP_ACCESS(vp, VWRITE, cred, l);
1699 	} else
1700 		error = VOP_ACCESS(vp, VADMIN, cred, l);
1701 	if (error)
1702 		return (error);
1703 
1704 	if (atime->tv_sec != VNOVAL && atime->tv_nsec != VNOVAL)
1705 		node->tn_status |= TMPFS_NODE_ACCESSED;
1706 
1707 	if (mtime->tv_sec != VNOVAL && mtime->tv_nsec != VNOVAL)
1708 		node->tn_status |= TMPFS_NODE_MODIFIED;
1709 
1710 	if (birthtime->tv_nsec != VNOVAL && birthtime->tv_nsec != VNOVAL)
1711 		node->tn_status |= TMPFS_NODE_MODIFIED;
1712 
1713 	tmpfs_itimes(vp, atime, mtime);
1714 
1715 	if (birthtime->tv_nsec != VNOVAL && birthtime->tv_nsec != VNOVAL)
1716 		node->tn_birthtime = *birthtime;
1717 	MPASS(VOP_ISLOCKED(vp));
1718 
1719 	return 0;
1720 }
1721 
1722 /* --------------------------------------------------------------------- */
1723 /* Sync timestamps */
1724 void
1725 tmpfs_itimes(struct vnode *vp, const struct timespec *acc,
1726     const struct timespec *mod)
1727 {
1728 	struct tmpfs_node *node;
1729 	struct timespec now;
1730 
1731 	node = VP_TO_TMPFS_NODE(vp);
1732 
1733 	if ((node->tn_status & (TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
1734 	    TMPFS_NODE_CHANGED)) == 0)
1735 		return;
1736 
1737 	vfs_timestamp(&now);
1738 	if (node->tn_status & TMPFS_NODE_ACCESSED) {
1739 		if (acc == NULL)
1740 			 acc = &now;
1741 		node->tn_atime = *acc;
1742 	}
1743 	if (node->tn_status & TMPFS_NODE_MODIFIED) {
1744 		if (mod == NULL)
1745 			mod = &now;
1746 		node->tn_mtime = *mod;
1747 	}
1748 	if (node->tn_status & TMPFS_NODE_CHANGED) {
1749 		node->tn_ctime = now;
1750 	}
1751 	node->tn_status &=
1752 	    ~(TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED | TMPFS_NODE_CHANGED);
1753 }
1754 
1755 /* --------------------------------------------------------------------- */
1756 
1757 void
1758 tmpfs_update(struct vnode *vp)
1759 {
1760 
1761 	tmpfs_itimes(vp, NULL, NULL);
1762 }
1763 
1764 /* --------------------------------------------------------------------- */
1765 
1766 int
1767 tmpfs_truncate(struct vnode *vp, off_t length)
1768 {
1769 	int error;
1770 	struct tmpfs_node *node;
1771 
1772 	node = VP_TO_TMPFS_NODE(vp);
1773 
1774 	if (length < 0) {
1775 		error = EINVAL;
1776 		goto out;
1777 	}
1778 
1779 	if (node->tn_size == length) {
1780 		error = 0;
1781 		goto out;
1782 	}
1783 
1784 	if (length > VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize)
1785 		return (EFBIG);
1786 
1787 	error = tmpfs_reg_resize(vp, length, FALSE);
1788 	if (error == 0) {
1789 		node->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1790 	}
1791 
1792 out:
1793 	tmpfs_update(vp);
1794 
1795 	return error;
1796 }
1797 
1798 static __inline int
1799 tmpfs_dirtree_cmp(struct tmpfs_dirent *a, struct tmpfs_dirent *b)
1800 {
1801 	if (a->td_hash > b->td_hash)
1802 		return (1);
1803 	else if (a->td_hash < b->td_hash)
1804 		return (-1);
1805 	return (0);
1806 }
1807 
1808 RB_GENERATE_STATIC(tmpfs_dir, tmpfs_dirent, uh.td_entries, tmpfs_dirtree_cmp);
1809