xref: /freebsd/sys/fs/tmpfs/tmpfs_subr.c (revision 3abc9103eb34adbb48853f2361206eba207d6c4f)
1 /*	$NetBSD: tmpfs_subr.c,v 1.21 2006/06/07 22:33:39 kardel 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  * 3. All advertising materials mentioning features or use of this software
20  *    must display the following acknowledgement:
21  *        This product includes software developed by the NetBSD
22  *        Foundation, Inc. and its contributors.
23  * 4. Neither the name of The NetBSD Foundation nor the names of its
24  *    contributors may be used to endorse or promote products derived
25  *    from this software without specific prior written permission.
26  *
27  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
28  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
29  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
30  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
31  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
32  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
33  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
34  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
35  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
36  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
37  * POSSIBILITY OF SUCH DAMAGE.
38  */
39 
40 /*
41  * Efficient memory file system supporting functions.
42  */
43 #include <sys/cdefs.h>
44 __FBSDID("$FreeBSD$");
45 
46 #include <sys/param.h>
47 #include <sys/namei.h>
48 #include <sys/priv.h>
49 #include <sys/proc.h>
50 #include <sys/stat.h>
51 #include <sys/systm.h>
52 #include <sys/vnode.h>
53 #include <sys/vmmeter.h>
54 
55 #include <vm/vm.h>
56 #include <vm/vm_object.h>
57 #include <vm/vm_page.h>
58 #include <vm/vm_pager.h>
59 #include <vm/vm_extern.h>
60 
61 #include <fs/tmpfs/tmpfs.h>
62 #include <fs/tmpfs/tmpfs_fifoops.h>
63 #include <fs/tmpfs/tmpfs_vnops.h>
64 
65 /* --------------------------------------------------------------------- */
66 
67 /*
68  * Allocates a new node of type 'type' inside the 'tmp' mount point, with
69  * its owner set to 'uid', its group to 'gid' and its mode set to 'mode',
70  * using the credentials of the process 'p'.
71  *
72  * If the node type is set to 'VDIR', then the parent parameter must point
73  * to the parent directory of the node being created.  It may only be NULL
74  * while allocating the root node.
75  *
76  * If the node type is set to 'VBLK' or 'VCHR', then the rdev parameter
77  * specifies the device the node represents.
78  *
79  * If the node type is set to 'VLNK', then the parameter target specifies
80  * the file name of the target file for the symbolic link that is being
81  * created.
82  *
83  * Note that new nodes are retrieved from the available list if it has
84  * items or, if it is empty, from the node pool as long as there is enough
85  * space to create them.
86  *
87  * Returns zero on success or an appropriate error code on failure.
88  */
89 int
90 tmpfs_alloc_node(struct tmpfs_mount *tmp, enum vtype type,
91     uid_t uid, gid_t gid, mode_t mode, struct tmpfs_node *parent,
92     char *target, dev_t rdev, struct thread *p, struct tmpfs_node **node)
93 {
94 	struct tmpfs_node *nnode;
95 
96 	/* If the root directory of the 'tmp' file system is not yet
97 	 * allocated, this must be the request to do it. */
98 	MPASS(IMPLIES(tmp->tm_root == NULL, parent == NULL && type == VDIR));
99 
100 	MPASS(IFF(type == VLNK, target != NULL));
101 	MPASS(IFF(type == VBLK || type == VCHR, rdev != VNOVAL));
102 
103 	if (tmp->tm_nodes_inuse > tmp->tm_nodes_max)
104 		return (ENOSPC);
105 
106 	nnode = (struct tmpfs_node *)uma_zalloc_arg(
107 				tmp->tm_node_pool, tmp, M_WAITOK);
108 
109 	/* Generic initialization. */
110 	nnode->tn_type = type;
111 	vfs_timestamp(&nnode->tn_atime);
112 	nnode->tn_birthtime = nnode->tn_ctime = nnode->tn_mtime =
113 	    nnode->tn_atime;
114 	nnode->tn_uid = uid;
115 	nnode->tn_gid = gid;
116 	nnode->tn_mode = mode;
117 	nnode->tn_id = alloc_unr(tmp->tm_ino_unr);
118 
119 	/* Type-specific initialization. */
120 	switch (nnode->tn_type) {
121 	case VBLK:
122 	case VCHR:
123 		nnode->tn_rdev = rdev;
124 		break;
125 
126 	case VDIR:
127 		TAILQ_INIT(&nnode->tn_dir.tn_dirhead);
128 		nnode->tn_dir.tn_parent = (parent == NULL) ? nnode : parent;
129 		nnode->tn_dir.tn_readdir_lastn = 0;
130 		nnode->tn_dir.tn_readdir_lastp = NULL;
131 		nnode->tn_links++;
132 		nnode->tn_dir.tn_parent->tn_links++;
133 		break;
134 
135 	case VFIFO:
136 		/* FALLTHROUGH */
137 	case VSOCK:
138 		break;
139 
140 	case VLNK:
141 		MPASS(strlen(target) < MAXPATHLEN);
142 		nnode->tn_size = strlen(target);
143 		nnode->tn_link = malloc(nnode->tn_size, M_TMPFSNAME,
144 		    M_WAITOK);
145 		memcpy(nnode->tn_link, target, nnode->tn_size);
146 		break;
147 
148 	case VREG:
149 		nnode->tn_reg.tn_aobj =
150 		    vm_pager_allocate(OBJT_SWAP, NULL, 0, VM_PROT_DEFAULT, 0);
151 		nnode->tn_reg.tn_aobj_pages = 0;
152 		break;
153 
154 	default:
155 		MPASS(0);
156 	}
157 
158 	TMPFS_LOCK(tmp);
159 	LIST_INSERT_HEAD(&tmp->tm_nodes_used, nnode, tn_entries);
160 	tmp->tm_nodes_inuse++;
161 	TMPFS_UNLOCK(tmp);
162 
163 	*node = nnode;
164 	return 0;
165 }
166 
167 /* --------------------------------------------------------------------- */
168 
169 /*
170  * Destroys the node pointed to by node from the file system 'tmp'.
171  * If the node does not belong to the given mount point, the results are
172  * unpredicted.
173  *
174  * If the node references a directory; no entries are allowed because
175  * their removal could need a recursive algorithm, something forbidden in
176  * kernel space.  Furthermore, there is not need to provide such
177  * functionality (recursive removal) because the only primitives offered
178  * to the user are the removal of empty directories and the deletion of
179  * individual files.
180  *
181  * Note that nodes are not really deleted; in fact, when a node has been
182  * allocated, it cannot be deleted during the whole life of the file
183  * system.  Instead, they are moved to the available list and remain there
184  * until reused.
185  */
186 void
187 tmpfs_free_node(struct tmpfs_mount *tmp, struct tmpfs_node *node)
188 {
189 	size_t pages = 0;
190 
191 	TMPFS_LOCK(tmp);
192 	LIST_REMOVE(node, tn_entries);
193 	tmp->tm_nodes_inuse--;
194 	TMPFS_UNLOCK(tmp);
195 
196 	switch (node->tn_type) {
197 	case VNON:
198 		/* Do not do anything.  VNON is provided to let the
199 		 * allocation routine clean itself easily by avoiding
200 		 * duplicating code in it. */
201 		/* FALLTHROUGH */
202 	case VBLK:
203 		/* FALLTHROUGH */
204 	case VCHR:
205 		/* FALLTHROUGH */
206 	case VDIR:
207 		/* FALLTHROUGH */
208 	case VFIFO:
209 		/* FALLTHROUGH */
210 	case VSOCK:
211 		break;
212 
213 	case VLNK:
214 		free(node->tn_link, M_TMPFSNAME);
215 		break;
216 
217 	case VREG:
218 		if (node->tn_reg.tn_aobj != NULL) {
219 			vm_object_deallocate(node->tn_reg.tn_aobj);
220 		}
221 		pages = node->tn_reg.tn_aobj_pages;
222 		break;
223 
224 	default:
225 		MPASS(0);
226 		break;
227 	}
228 
229 	free_unr(tmp->tm_ino_unr, node->tn_id);
230 	uma_zfree(tmp->tm_node_pool, node);
231 
232 	TMPFS_LOCK(tmp);
233 	tmp->tm_pages_used -= pages;
234 	TMPFS_UNLOCK(tmp);
235 }
236 
237 /* --------------------------------------------------------------------- */
238 
239 /*
240  * Allocates a new directory entry for the node node with a name of name.
241  * The new directory entry is returned in *de.
242  *
243  * The link count of node is increased by one to reflect the new object
244  * referencing it.
245  *
246  * Returns zero on success or an appropriate error code on failure.
247  */
248 int
249 tmpfs_alloc_dirent(struct tmpfs_mount *tmp, struct tmpfs_node *node,
250     const char *name, uint16_t len, struct tmpfs_dirent **de)
251 {
252 	struct tmpfs_dirent *nde;
253 
254 	nde = (struct tmpfs_dirent *)uma_zalloc(
255 					tmp->tm_dirent_pool, M_WAITOK);
256 	nde->td_name = malloc(len, M_TMPFSNAME, M_WAITOK);
257 	nde->td_namelen = len;
258 	memcpy(nde->td_name, name, len);
259 
260 	nde->td_node = node;
261 	node->tn_links++;
262 
263 	*de = nde;
264 
265 	return 0;
266 }
267 
268 /* --------------------------------------------------------------------- */
269 
270 /*
271  * Frees a directory entry.  It is the caller's responsibility to destroy
272  * the node referenced by it if needed.
273  *
274  * The link count of node is decreased by one to reflect the removal of an
275  * object that referenced it.  This only happens if 'node_exists' is true;
276  * otherwise the function will not access the node referred to by the
277  * directory entry, as it may already have been released from the outside.
278  */
279 void
280 tmpfs_free_dirent(struct tmpfs_mount *tmp, struct tmpfs_dirent *de,
281     boolean_t node_exists)
282 {
283 	if (node_exists) {
284 		struct tmpfs_node *node;
285 
286 		node = de->td_node;
287 
288 		MPASS(node->tn_links > 0);
289 		node->tn_links--;
290 	}
291 
292 	free(de->td_name, M_TMPFSNAME);
293 	uma_zfree(tmp->tm_dirent_pool, de);
294 }
295 
296 /* --------------------------------------------------------------------- */
297 
298 /*
299  * Allocates a new vnode for the node node or returns a new reference to
300  * an existing one if the node had already a vnode referencing it.  The
301  * resulting locked vnode is returned in *vpp.
302  *
303  * Returns zero on success or an appropriate error code on failure.
304  */
305 int
306 tmpfs_alloc_vp(struct mount *mp, struct tmpfs_node *node, struct vnode **vpp,
307     struct thread *td)
308 {
309 	int error;
310 	struct vnode *vp;
311 
312 	vp = NULL;
313 
314 loop:
315 	if (node->tn_vnode != NULL) {
316 		vp = node->tn_vnode;
317 		error = vget(vp, LK_EXCLUSIVE | LK_RETRY, td);
318 		if (error)
319 			return error;
320 
321 		/*
322 		 * Make sure the vnode is still there after
323 		 * getting the interlock to avoid racing a free.
324 		 */
325 		if (node->tn_vnode == NULL || node->tn_vnode != vp) {
326 			vput(vp);
327 			goto loop;
328 		}
329 
330 		goto out;
331 	}
332 
333 	/*
334 	 * otherwise lock the vp list while we call getnewvnode
335 	 * since that can block.
336 	 */
337 	TMPFS_NODE_LOCK(node);
338 	if (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) {
339 		node->tn_vpstate |= TMPFS_VNODE_WANT;
340 		TMPFS_NODE_UNLOCK(node);
341 		(void) tsleep((caddr_t) &node->tn_vpstate, 0, "tmpfs_vplock", 0);
342 		goto loop;
343 	}
344 
345 	node->tn_vpstate |= TMPFS_VNODE_ALLOCATING;
346 	TMPFS_NODE_UNLOCK(node);
347 
348 	/* Get a new vnode and associate it with our node. */
349 	error = getnewvnode("tmpfs", mp, &tmpfs_vnodeop_entries, &vp);
350 	if (error != 0)
351 		goto unlock;
352 	MPASS(vp != NULL);
353 
354 	error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
355 	if (error != 0) {
356 		vp->v_data = NULL;
357 		vput(vp);
358 		vp = NULL;
359 		goto unlock;
360 	}
361 
362 	vp->v_data = node;
363 	vp->v_type = node->tn_type;
364 
365 	/* Type-specific initialization. */
366 	switch (node->tn_type) {
367 	case VBLK:
368 		/* FALLTHROUGH */
369 	case VCHR:
370 		break;
371 
372 	case VDIR:
373 		break;
374 
375 	case VFIFO:
376 		vp->v_op = &tmpfs_fifoop_entries;
377 		break;
378 
379 	case VLNK:
380 		/* FALLTHROUGH */
381 	case VREG:
382 		/* FALLTHROUGH */
383 	case VSOCK:
384 		break;
385 
386 	default:
387 		MPASS(0);
388 	}
389 
390 	vnode_pager_setsize(vp, node->tn_size);
391 	error = insmntque(vp, mp);
392 	if (error) {
393 		node->tn_vnode = NULL;
394 		if (node->tn_vpstate & TMPFS_VNODE_WANT) {
395 			node->tn_vpstate &= ~TMPFS_VNODE_WANT;
396 			TMPFS_NODE_UNLOCK(node);
397 			wakeup((caddr_t) &node->tn_vpstate);
398 		} else
399 			TMPFS_NODE_UNLOCK(node);
400 		return error;
401 	}
402 	node->tn_vnode = vp;
403 
404 unlock:
405 	MPASS(node->tn_vpstate & TMPFS_VNODE_ALLOCATING);
406 	TMPFS_NODE_LOCK(node);
407 	node->tn_vpstate &= ~TMPFS_VNODE_ALLOCATING;
408 
409 	if (node->tn_vpstate & TMPFS_VNODE_WANT) {
410 		node->tn_vpstate &= ~TMPFS_VNODE_WANT;
411 		TMPFS_NODE_UNLOCK(node);
412 		wakeup((caddr_t) &node->tn_vpstate);
413 	} else
414 		TMPFS_NODE_UNLOCK(node);
415 
416 out:
417 	*vpp = vp;
418 
419 	MPASS(IFF(error == 0, *vpp != NULL && VOP_ISLOCKED(*vpp, td)));
420 	MPASS(*vpp == node->tn_vnode);
421 
422 	return error;
423 }
424 
425 /* --------------------------------------------------------------------- */
426 
427 /*
428  * Destroys the association between the vnode vp and the node it
429  * references.
430  */
431 void
432 tmpfs_free_vp(struct vnode *vp)
433 {
434 	struct tmpfs_node *node;
435 
436 	node = VP_TO_TMPFS_NODE(vp);
437 
438 	node->tn_vnode = NULL;
439 	vp->v_data = NULL;
440 }
441 
442 /* --------------------------------------------------------------------- */
443 
444 /*
445  * Allocates a new file of type 'type' and adds it to the parent directory
446  * 'dvp'; this addition is done using the component name given in 'cnp'.
447  * The ownership of the new file is automatically assigned based on the
448  * credentials of the caller (through 'cnp'), the group is set based on
449  * the parent directory and the mode is determined from the 'vap' argument.
450  * If successful, *vpp holds a vnode to the newly created file and zero
451  * is returned.  Otherwise *vpp is NULL and the function returns an
452  * appropriate error code.
453  */
454 int
455 tmpfs_alloc_file(struct vnode *dvp, struct vnode **vpp, struct vattr *vap,
456     struct componentname *cnp, char *target)
457 {
458 	int error;
459 	struct tmpfs_dirent *de;
460 	struct tmpfs_mount *tmp;
461 	struct tmpfs_node *dnode;
462 	struct tmpfs_node *node;
463 	struct tmpfs_node *parent;
464 
465 	MPASS(VOP_ISLOCKED(dvp, cnp->cn_thread));
466 	MPASS(cnp->cn_flags & HASBUF);
467 
468 	tmp = VFS_TO_TMPFS(dvp->v_mount);
469 	dnode = VP_TO_TMPFS_DIR(dvp);
470 	*vpp = NULL;
471 
472 	/* If the entry we are creating is a directory, we cannot overflow
473 	 * the number of links of its parent, because it will get a new
474 	 * link. */
475 	if (vap->va_type == VDIR) {
476 		/* Ensure that we do not overflow the maximum number of links
477 		 * imposed by the system. */
478 		MPASS(dnode->tn_links <= LINK_MAX);
479 		if (dnode->tn_links == LINK_MAX) {
480 			error = EMLINK;
481 			goto out;
482 		}
483 
484 		parent = dnode;
485 	} else
486 		parent = NULL;
487 
488 	/* Allocate a node that represents the new file. */
489 	error = tmpfs_alloc_node(tmp, vap->va_type, cnp->cn_cred->cr_uid,
490 	    dnode->tn_gid, vap->va_mode, parent, target, vap->va_rdev,
491 	    cnp->cn_thread, &node);
492 	if (error != 0)
493 		goto out;
494 
495 	/* Allocate a directory entry that points to the new file. */
496 	error = tmpfs_alloc_dirent(tmp, node, cnp->cn_nameptr, cnp->cn_namelen,
497 	    &de);
498 	if (error != 0) {
499 		tmpfs_free_node(tmp, node);
500 		goto out;
501 	}
502 
503 	/* Allocate a vnode for the new file. */
504 	error = tmpfs_alloc_vp(dvp->v_mount, node, vpp, cnp->cn_thread);
505 	if (error != 0) {
506 		tmpfs_free_dirent(tmp, de, TRUE);
507 		tmpfs_free_node(tmp, node);
508 		goto out;
509 	}
510 
511 	/* Now that all required items are allocated, we can proceed to
512 	 * insert the new node into the directory, an operation that
513 	 * cannot fail. */
514 	tmpfs_dir_attach(dvp, de);
515 
516 out:
517 
518 	return error;
519 }
520 
521 /* --------------------------------------------------------------------- */
522 
523 /*
524  * Attaches the directory entry de to the directory represented by vp.
525  * Note that this does not change the link count of the node pointed by
526  * the directory entry, as this is done by tmpfs_alloc_dirent.
527  */
528 void
529 tmpfs_dir_attach(struct vnode *vp, struct tmpfs_dirent *de)
530 {
531 	struct tmpfs_node *dnode;
532 
533 	dnode = VP_TO_TMPFS_DIR(vp);
534 	TAILQ_INSERT_TAIL(&dnode->tn_dir.tn_dirhead, de, td_entries);
535 	dnode->tn_size += sizeof(struct tmpfs_dirent);
536 	dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
537 	    TMPFS_NODE_MODIFIED;
538 }
539 
540 /* --------------------------------------------------------------------- */
541 
542 /*
543  * Detaches the directory entry de from the directory represented by vp.
544  * Note that this does not change the link count of the node pointed by
545  * the directory entry, as this is done by tmpfs_free_dirent.
546  */
547 void
548 tmpfs_dir_detach(struct vnode *vp, struct tmpfs_dirent *de)
549 {
550 	struct tmpfs_node *dnode;
551 
552 	dnode = VP_TO_TMPFS_DIR(vp);
553 
554 	if (dnode->tn_dir.tn_readdir_lastp == de) {
555 		dnode->tn_dir.tn_readdir_lastn = 0;
556 		dnode->tn_dir.tn_readdir_lastp = NULL;
557 	}
558 
559 	TAILQ_REMOVE(&dnode->tn_dir.tn_dirhead, de, td_entries);
560 	dnode->tn_size -= sizeof(struct tmpfs_dirent);
561 	dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
562 	    TMPFS_NODE_MODIFIED;
563 }
564 
565 /* --------------------------------------------------------------------- */
566 
567 /*
568  * Looks for a directory entry in the directory represented by node.
569  * 'cnp' describes the name of the entry to look for.  Note that the .
570  * and .. components are not allowed as they do not physically exist
571  * within directories.
572  *
573  * Returns a pointer to the entry when found, otherwise NULL.
574  */
575 struct tmpfs_dirent *
576 tmpfs_dir_lookup(struct tmpfs_node *node, struct componentname *cnp)
577 {
578 	boolean_t found;
579 	struct tmpfs_dirent *de;
580 
581 	MPASS(IMPLIES(cnp->cn_namelen == 1, cnp->cn_nameptr[0] != '.'));
582 	MPASS(IMPLIES(cnp->cn_namelen == 2, !(cnp->cn_nameptr[0] == '.' &&
583 	    cnp->cn_nameptr[1] == '.')));
584 	TMPFS_VALIDATE_DIR(node);
585 
586 	found = 0;
587 	TAILQ_FOREACH(de, &node->tn_dir.tn_dirhead, td_entries) {
588 		MPASS(cnp->cn_namelen < 0xffff);
589 		if (de->td_namelen == (uint16_t)cnp->cn_namelen &&
590 		    memcmp(de->td_name, cnp->cn_nameptr, de->td_namelen) == 0) {
591 			found = 1;
592 			break;
593 		}
594 	}
595 	node->tn_status |= TMPFS_NODE_ACCESSED;
596 
597 	return found ? de : NULL;
598 }
599 
600 /* --------------------------------------------------------------------- */
601 
602 /*
603  * Helper function for tmpfs_readdir.  Creates a '.' entry for the given
604  * directory and returns it in the uio space.  The function returns 0
605  * on success, -1 if there was not enough space in the uio structure to
606  * hold the directory entry or an appropriate error code if another
607  * error happens.
608  */
609 int
610 tmpfs_dir_getdotdent(struct tmpfs_node *node, struct uio *uio)
611 {
612 	int error;
613 	struct dirent dent;
614 
615 	TMPFS_VALIDATE_DIR(node);
616 	MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOT);
617 
618 	dent.d_fileno = node->tn_id;
619 	dent.d_type = DT_DIR;
620 	dent.d_namlen = 1;
621 	dent.d_name[0] = '.';
622 	dent.d_name[1] = '\0';
623 	dent.d_reclen = GENERIC_DIRSIZ(&dent);
624 
625 	if (dent.d_reclen > uio->uio_resid)
626 		error = -1;
627 	else {
628 		error = uiomove(&dent, dent.d_reclen, uio);
629 		if (error == 0)
630 			uio->uio_offset = TMPFS_DIRCOOKIE_DOTDOT;
631 	}
632 
633 	node->tn_status |= TMPFS_NODE_ACCESSED;
634 
635 	return error;
636 }
637 
638 /* --------------------------------------------------------------------- */
639 
640 /*
641  * Helper function for tmpfs_readdir.  Creates a '..' entry for the given
642  * directory and returns it in the uio space.  The function returns 0
643  * on success, -1 if there was not enough space in the uio structure to
644  * hold the directory entry or an appropriate error code if another
645  * error happens.
646  */
647 int
648 tmpfs_dir_getdotdotdent(struct tmpfs_node *node, struct uio *uio)
649 {
650 	int error;
651 	struct dirent dent;
652 
653 	TMPFS_VALIDATE_DIR(node);
654 	MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT);
655 
656 	dent.d_fileno = node->tn_dir.tn_parent->tn_id;
657 	dent.d_type = DT_DIR;
658 	dent.d_namlen = 2;
659 	dent.d_name[0] = '.';
660 	dent.d_name[1] = '.';
661 	dent.d_name[2] = '\0';
662 	dent.d_reclen = GENERIC_DIRSIZ(&dent);
663 
664 	if (dent.d_reclen > uio->uio_resid)
665 		error = -1;
666 	else {
667 		error = uiomove(&dent, dent.d_reclen, uio);
668 		if (error == 0) {
669 			struct tmpfs_dirent *de;
670 
671 			de = TAILQ_FIRST(&node->tn_dir.tn_dirhead);
672 			if (de == NULL)
673 				uio->uio_offset = TMPFS_DIRCOOKIE_EOF;
674 			else
675 				uio->uio_offset = TMPFS_DIRCOOKIE(de);
676 		}
677 	}
678 
679 	node->tn_status |= TMPFS_NODE_ACCESSED;
680 
681 	return error;
682 }
683 
684 /* --------------------------------------------------------------------- */
685 
686 /*
687  * Lookup a directory entry by its associated cookie.
688  */
689 struct tmpfs_dirent *
690 tmpfs_dir_lookupbycookie(struct tmpfs_node *node, off_t cookie)
691 {
692 	struct tmpfs_dirent *de;
693 
694 	if (cookie == node->tn_dir.tn_readdir_lastn &&
695 	    node->tn_dir.tn_readdir_lastp != NULL) {
696 		return node->tn_dir.tn_readdir_lastp;
697 	}
698 
699 	TAILQ_FOREACH(de, &node->tn_dir.tn_dirhead, td_entries) {
700 		if (TMPFS_DIRCOOKIE(de) == cookie) {
701 			break;
702 		}
703 	}
704 
705 	return de;
706 }
707 
708 /* --------------------------------------------------------------------- */
709 
710 /*
711  * Helper function for tmpfs_readdir.  Returns as much directory entries
712  * as can fit in the uio space.  The read starts at uio->uio_offset.
713  * The function returns 0 on success, -1 if there was not enough space
714  * in the uio structure to hold the directory entry or an appropriate
715  * error code if another error happens.
716  */
717 int
718 tmpfs_dir_getdents(struct tmpfs_node *node, struct uio *uio, off_t *cntp)
719 {
720 	int error;
721 	off_t startcookie;
722 	struct tmpfs_dirent *de;
723 
724 	TMPFS_VALIDATE_DIR(node);
725 
726 	/* Locate the first directory entry we have to return.  We have cached
727 	 * the last readdir in the node, so use those values if appropriate.
728 	 * Otherwise do a linear scan to find the requested entry. */
729 	startcookie = uio->uio_offset;
730 	MPASS(startcookie != TMPFS_DIRCOOKIE_DOT);
731 	MPASS(startcookie != TMPFS_DIRCOOKIE_DOTDOT);
732 	if (startcookie == TMPFS_DIRCOOKIE_EOF) {
733 		return 0;
734 	} else {
735 		de = tmpfs_dir_lookupbycookie(node, startcookie);
736 	}
737 	if (de == NULL) {
738 		return EINVAL;
739 	}
740 
741 	/* Read as much entries as possible; i.e., until we reach the end of
742 	 * the directory or we exhaust uio space. */
743 	do {
744 		struct dirent d;
745 
746 		/* Create a dirent structure representing the current
747 		 * tmpfs_node and fill it. */
748 		d.d_fileno = de->td_node->tn_id;
749 		switch (de->td_node->tn_type) {
750 		case VBLK:
751 			d.d_type = DT_BLK;
752 			break;
753 
754 		case VCHR:
755 			d.d_type = DT_CHR;
756 			break;
757 
758 		case VDIR:
759 			d.d_type = DT_DIR;
760 			break;
761 
762 		case VFIFO:
763 			d.d_type = DT_FIFO;
764 			break;
765 
766 		case VLNK:
767 			d.d_type = DT_LNK;
768 			break;
769 
770 		case VREG:
771 			d.d_type = DT_REG;
772 			break;
773 
774 		case VSOCK:
775 			d.d_type = DT_SOCK;
776 			break;
777 
778 		default:
779 			MPASS(0);
780 		}
781 		d.d_namlen = de->td_namelen;
782 		MPASS(de->td_namelen < sizeof(d.d_name));
783 		(void)memcpy(d.d_name, de->td_name, de->td_namelen);
784 		d.d_name[de->td_namelen] = '\0';
785 		d.d_reclen = GENERIC_DIRSIZ(&d);
786 
787 		/* Stop reading if the directory entry we are treating is
788 		 * bigger than the amount of data that can be returned. */
789 		if (d.d_reclen > uio->uio_resid) {
790 			error = -1;
791 			break;
792 		}
793 
794 		/* Copy the new dirent structure into the output buffer and
795 		 * advance pointers. */
796 		error = uiomove(&d, d.d_reclen, uio);
797 
798 		(*cntp)++;
799 		de = TAILQ_NEXT(de, td_entries);
800 	} while (error == 0 && uio->uio_resid > 0 && de != NULL);
801 
802 	/* Update the offset and cache. */
803 	if (de == NULL) {
804 		uio->uio_offset = TMPFS_DIRCOOKIE_EOF;
805 		node->tn_dir.tn_readdir_lastn = 0;
806 		node->tn_dir.tn_readdir_lastp = NULL;
807 	} else {
808 		node->tn_dir.tn_readdir_lastn = uio->uio_offset = TMPFS_DIRCOOKIE(de);
809 		node->tn_dir.tn_readdir_lastp = de;
810 	}
811 
812 	node->tn_status |= TMPFS_NODE_ACCESSED;
813 	return error;
814 }
815 
816 /* --------------------------------------------------------------------- */
817 
818 /*
819  * Resizes the aobj associated to the regular file pointed to by vp to
820  * the size newsize.  'vp' must point to a vnode that represents a regular
821  * file.  'newsize' must be positive.
822  *
823  * Returns zero on success or an appropriate error code on failure.
824  */
825 int
826 tmpfs_reg_resize(struct vnode *vp, off_t newsize)
827 {
828 	int error;
829 	size_t newpages, oldpages;
830 	struct tmpfs_mount *tmp;
831 	struct tmpfs_node *node;
832 	off_t oldsize;
833 
834 	MPASS(vp->v_type == VREG);
835 	MPASS(newsize >= 0);
836 
837 	node = VP_TO_TMPFS_NODE(vp);
838 	tmp = VFS_TO_TMPFS(vp->v_mount);
839 
840 	/* Convert the old and new sizes to the number of pages needed to
841 	 * store them.  It may happen that we do not need to do anything
842 	 * because the last allocated page can accommodate the change on
843 	 * its own. */
844 	oldsize = node->tn_size;
845 	oldpages = round_page(oldsize) / PAGE_SIZE;
846 	MPASS(oldpages == node->tn_reg.tn_aobj_pages);
847 	newpages = round_page(newsize) / PAGE_SIZE;
848 
849 	if (newpages > oldpages &&
850 	    newpages - oldpages > TMPFS_PAGES_AVAIL(tmp)) {
851 		error = ENOSPC;
852 		goto out;
853 	}
854 
855 	node->tn_reg.tn_aobj_pages = newpages;
856 
857 	TMPFS_LOCK(tmp);
858 	tmp->tm_pages_used += (newpages - oldpages);
859 	TMPFS_UNLOCK(tmp);
860 
861 	node->tn_size = newsize;
862 	vnode_pager_setsize(vp, newsize);
863 	if (newsize < oldsize) {
864 		size_t zerolen = round_page(newsize) - newsize;
865 		vm_object_t uobj = node->tn_reg.tn_aobj;
866 		vm_page_t m;
867 
868 		/*
869 		 * free "backing store"
870 		 */
871 		VM_OBJECT_LOCK(uobj);
872 		if (newpages < oldpages) {
873 			swap_pager_freespace(uobj,
874 						newpages, oldpages - newpages);
875 			vm_object_page_remove(uobj,
876 				OFF_TO_IDX(newsize + PAGE_MASK), 0, FALSE);
877 		}
878 
879 		/*
880 		 * zero out the truncated part of the last page.
881 		 */
882 
883 		if (zerolen > 0) {
884 			m = vm_page_grab(uobj, OFF_TO_IDX(newsize),
885 					VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
886 			pmap_zero_page_area(m, PAGE_SIZE - zerolen,
887 				zerolen);
888 			vm_page_wakeup(m);
889 		}
890 		VM_OBJECT_UNLOCK(uobj);
891 
892 	}
893 
894 	error = 0;
895 
896 out:
897 	return error;
898 }
899 
900 /* --------------------------------------------------------------------- */
901 
902 /*
903  * Change flags of the given vnode.
904  * Caller should execute tmpfs_update on vp after a successful execution.
905  * The vnode must be locked on entry and remain locked on exit.
906  */
907 int
908 tmpfs_chflags(struct vnode *vp, int flags, struct ucred *cred, struct thread *p)
909 {
910 	int error;
911 	struct tmpfs_node *node;
912 
913 	MPASS(VOP_ISLOCKED(vp, p));
914 
915 	node = VP_TO_TMPFS_NODE(vp);
916 
917 	/* Disallow this operation if the file system is mounted read-only. */
918 	if (vp->v_mount->mnt_flag & MNT_RDONLY)
919 		return EROFS;
920 
921 	/*
922 	 * Callers may only modify the file flags on objects they
923 	 * have VADMIN rights for.
924 	 */
925 	if ((error = VOP_ACCESS(vp, VADMIN, cred, p)))
926 		return (error);
927 	/*
928 	 * Unprivileged processes are not permitted to unset system
929 	 * flags, or modify flags if any system flags are set.
930 	 */
931 	if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0)) {
932 		if (node->tn_flags
933 		  & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND)) {
934 			error = securelevel_gt(cred, 0);
935 			if (error)
936 				return (error);
937 		}
938 		/* Snapshot flag cannot be set or cleared */
939 		if (((flags & SF_SNAPSHOT) != 0 &&
940 		  (node->tn_flags & SF_SNAPSHOT) == 0) ||
941 		  ((flags & SF_SNAPSHOT) == 0 &&
942 		  (node->tn_flags & SF_SNAPSHOT) != 0))
943 			return (EPERM);
944 		node->tn_flags = flags;
945 	} else {
946 		if (node->tn_flags
947 		  & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND) ||
948 		  (flags & UF_SETTABLE) != flags)
949 			return (EPERM);
950 		node->tn_flags &= SF_SETTABLE;
951 		node->tn_flags |= (flags & UF_SETTABLE);
952 	}
953 	node->tn_status |= TMPFS_NODE_CHANGED;
954 
955 	MPASS(VOP_ISLOCKED(vp, p));
956 
957 	return 0;
958 }
959 
960 /* --------------------------------------------------------------------- */
961 
962 /*
963  * Change access mode on the given vnode.
964  * Caller should execute tmpfs_update on vp after a successful execution.
965  * The vnode must be locked on entry and remain locked on exit.
966  */
967 int
968 tmpfs_chmod(struct vnode *vp, mode_t mode, struct ucred *cred, struct thread *p)
969 {
970 	int error;
971 	struct tmpfs_node *node;
972 
973 	MPASS(VOP_ISLOCKED(vp, p));
974 
975 	node = VP_TO_TMPFS_NODE(vp);
976 
977 	/* Disallow this operation if the file system is mounted read-only. */
978 	if (vp->v_mount->mnt_flag & MNT_RDONLY)
979 		return EROFS;
980 
981 	/* Immutable or append-only files cannot be modified, either. */
982 	if (node->tn_flags & (IMMUTABLE | APPEND))
983 		return EPERM;
984 
985 	/*
986 	 * To modify the permissions on a file, must possess VADMIN
987 	 * for that file.
988 	 */
989 	if ((error = VOP_ACCESS(vp, VADMIN, cred, p)))
990 		return (error);
991 
992 	/*
993 	 * Privileged processes may set the sticky bit on non-directories,
994 	 * as well as set the setgid bit on a file with a group that the
995 	 * process is not a member of.
996 	 */
997 	if (vp->v_type != VDIR && (mode & S_ISTXT)) {
998 		if (priv_check_cred(cred, PRIV_VFS_STICKYFILE, 0))
999 			return (EFTYPE);
1000 	}
1001 	if (!groupmember(node->tn_gid, cred) && (mode & S_ISGID)) {
1002 		error = priv_check_cred(cred, PRIV_VFS_SETGID, 0);
1003 		if (error)
1004 			return (error);
1005 	}
1006 
1007 
1008 	node->tn_mode &= ~ALLPERMS;
1009 	node->tn_mode |= mode & ALLPERMS;
1010 
1011 	node->tn_status |= TMPFS_NODE_CHANGED;
1012 
1013 	MPASS(VOP_ISLOCKED(vp, p));
1014 
1015 	return 0;
1016 }
1017 
1018 /* --------------------------------------------------------------------- */
1019 
1020 /*
1021  * Change ownership of the given vnode.  At least one of uid or gid must
1022  * be different than VNOVAL.  If one is set to that value, the attribute
1023  * is unchanged.
1024  * Caller should execute tmpfs_update on vp after a successful execution.
1025  * The vnode must be locked on entry and remain locked on exit.
1026  */
1027 int
1028 tmpfs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred,
1029     struct thread *p)
1030 {
1031 	int error;
1032 	struct tmpfs_node *node;
1033 	uid_t ouid;
1034 	gid_t ogid;
1035 
1036 	MPASS(VOP_ISLOCKED(vp, p));
1037 
1038 	node = VP_TO_TMPFS_NODE(vp);
1039 
1040 	/* Assign default values if they are unknown. */
1041 	MPASS(uid != VNOVAL || gid != VNOVAL);
1042 	if (uid == VNOVAL)
1043 		uid = node->tn_uid;
1044 	if (gid == VNOVAL)
1045 		gid = node->tn_gid;
1046 	MPASS(uid != VNOVAL && gid != VNOVAL);
1047 
1048 	/* Disallow this operation if the file system is mounted read-only. */
1049 	if (vp->v_mount->mnt_flag & MNT_RDONLY)
1050 		return EROFS;
1051 
1052 	/* Immutable or append-only files cannot be modified, either. */
1053 	if (node->tn_flags & (IMMUTABLE | APPEND))
1054 		return EPERM;
1055 
1056 	/*
1057 	 * To modify the ownership of a file, must possess VADMIN for that
1058 	 * file.
1059 	 */
1060 	if ((error = VOP_ACCESS(vp, VADMIN, cred, p)))
1061 		return (error);
1062 
1063 	/*
1064 	 * To change the owner of a file, or change the group of a file to a
1065 	 * group of which we are not a member, the caller must have
1066 	 * privilege.
1067 	 */
1068 	if ((uid != node->tn_uid ||
1069 	    (gid != node->tn_gid && !groupmember(gid, cred))) &&
1070 	    (error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0)))
1071 		return (error);
1072 
1073 	ogid = node->tn_gid;
1074 	ouid = node->tn_uid;
1075 
1076 	node->tn_uid = uid;
1077 	node->tn_gid = gid;
1078 
1079 	node->tn_status |= TMPFS_NODE_CHANGED;
1080 
1081 	if ((node->tn_mode & (S_ISUID | S_ISGID)) && (ouid != uid || ogid != gid)) {
1082 		if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID, 0))
1083 			node->tn_mode &= ~(S_ISUID | S_ISGID);
1084 	}
1085 
1086 	MPASS(VOP_ISLOCKED(vp, p));
1087 
1088 	return 0;
1089 }
1090 
1091 /* --------------------------------------------------------------------- */
1092 
1093 /*
1094  * Change size of the given vnode.
1095  * Caller should execute tmpfs_update on vp after a successful execution.
1096  * The vnode must be locked on entry and remain locked on exit.
1097  */
1098 int
1099 tmpfs_chsize(struct vnode *vp, u_quad_t size, struct ucred *cred,
1100     struct thread *p)
1101 {
1102 	int error;
1103 	struct tmpfs_node *node;
1104 
1105 	MPASS(VOP_ISLOCKED(vp, p));
1106 
1107 	node = VP_TO_TMPFS_NODE(vp);
1108 
1109 	/* Decide whether this is a valid operation based on the file type. */
1110 	error = 0;
1111 	switch (vp->v_type) {
1112 	case VDIR:
1113 		return EISDIR;
1114 
1115 	case VREG:
1116 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
1117 			return EROFS;
1118 		break;
1119 
1120 	case VBLK:
1121 		/* FALLTHROUGH */
1122 	case VCHR:
1123 		/* FALLTHROUGH */
1124 	case VFIFO:
1125 		/* Allow modifications of special files even if in the file
1126 		 * system is mounted read-only (we are not modifying the
1127 		 * files themselves, but the objects they represent). */
1128 		return 0;
1129 
1130 	default:
1131 		/* Anything else is unsupported. */
1132 		return EOPNOTSUPP;
1133 	}
1134 
1135 	/* Immutable or append-only files cannot be modified, either. */
1136 	if (node->tn_flags & (IMMUTABLE | APPEND))
1137 		return EPERM;
1138 
1139 	error = tmpfs_truncate(vp, size);
1140 	/* tmpfs_truncate will raise the NOTE_EXTEND and NOTE_ATTRIB kevents
1141 	 * for us, as will update tn_status; no need to do that here. */
1142 
1143 	MPASS(VOP_ISLOCKED(vp, p));
1144 
1145 	return error;
1146 }
1147 
1148 /* --------------------------------------------------------------------- */
1149 
1150 /*
1151  * Change access and modification times of the given vnode.
1152  * Caller should execute tmpfs_update on vp after a successful execution.
1153  * The vnode must be locked on entry and remain locked on exit.
1154  */
1155 int
1156 tmpfs_chtimes(struct vnode *vp, struct timespec *atime, struct timespec *mtime,
1157 	struct timespec *birthtime, int vaflags, struct ucred *cred, struct thread *l)
1158 {
1159 	int error;
1160 	struct tmpfs_node *node;
1161 
1162 	MPASS(VOP_ISLOCKED(vp, l));
1163 
1164 	node = VP_TO_TMPFS_NODE(vp);
1165 
1166 	/* Disallow this operation if the file system is mounted read-only. */
1167 	if (vp->v_mount->mnt_flag & MNT_RDONLY)
1168 		return EROFS;
1169 
1170 	/* Immutable or append-only files cannot be modified, either. */
1171 	if (node->tn_flags & (IMMUTABLE | APPEND))
1172 		return EPERM;
1173 
1174 	/* Determine if the user have proper privilege to update time. */
1175 	if (vaflags & VA_UTIMES_NULL) {
1176 		error = VOP_ACCESS(vp, VADMIN, cred, l);
1177 		if (error)
1178 			error = VOP_ACCESS(vp, VWRITE, cred, l);
1179 	} else
1180 		error = VOP_ACCESS(vp, VADMIN, cred, l);
1181 	if (error)
1182 		return (error);
1183 
1184 	if (atime->tv_sec != VNOVAL && atime->tv_nsec != VNOVAL)
1185 		node->tn_status |= TMPFS_NODE_ACCESSED;
1186 
1187 	if (mtime->tv_sec != VNOVAL && mtime->tv_nsec != VNOVAL)
1188 		node->tn_status |= TMPFS_NODE_MODIFIED;
1189 
1190 	if (birthtime->tv_nsec != VNOVAL && birthtime->tv_nsec != VNOVAL)
1191 		node->tn_status |= TMPFS_NODE_MODIFIED;
1192 
1193 	tmpfs_itimes(vp, atime, mtime);
1194 
1195 	if (birthtime->tv_nsec != VNOVAL && birthtime->tv_nsec != VNOVAL)
1196 		node->tn_birthtime = *birthtime;
1197 	MPASS(VOP_ISLOCKED(vp, l));
1198 
1199 	return 0;
1200 }
1201 
1202 /* --------------------------------------------------------------------- */
1203 /* Sync timestamps */
1204 void
1205 tmpfs_itimes(struct vnode *vp, const struct timespec *acc,
1206     const struct timespec *mod)
1207 {
1208 	struct tmpfs_node *node;
1209 	struct timespec now;
1210 
1211 	node = VP_TO_TMPFS_NODE(vp);
1212 
1213 	if ((node->tn_status & (TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
1214 	    TMPFS_NODE_CHANGED)) == 0)
1215 		return;
1216 
1217 	vfs_timestamp(&now);
1218 	if (node->tn_status & TMPFS_NODE_ACCESSED) {
1219 		if (acc == NULL)
1220 			 acc = &now;
1221 		node->tn_atime = *acc;
1222 	}
1223 	if (node->tn_status & TMPFS_NODE_MODIFIED) {
1224 		if (mod == NULL)
1225 			mod = &now;
1226 		node->tn_mtime = *mod;
1227 	}
1228 	if (node->tn_status & TMPFS_NODE_CHANGED) {
1229 		node->tn_ctime = now;
1230 	}
1231 	node->tn_status &=
1232 	    ~(TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED | TMPFS_NODE_CHANGED);
1233 }
1234 
1235 /* --------------------------------------------------------------------- */
1236 
1237 void
1238 tmpfs_update(struct vnode *vp)
1239 {
1240 
1241 	tmpfs_itimes(vp, NULL, NULL);
1242 }
1243 
1244 /* --------------------------------------------------------------------- */
1245 
1246 int
1247 tmpfs_truncate(struct vnode *vp, off_t length)
1248 {
1249 	boolean_t extended;
1250 	int error;
1251 	struct tmpfs_node *node;
1252 
1253 	node = VP_TO_TMPFS_NODE(vp);
1254 	extended = length > node->tn_size;
1255 
1256 	if (length < 0) {
1257 		error = EINVAL;
1258 		goto out;
1259 	}
1260 
1261 	if (node->tn_size == length) {
1262 		error = 0;
1263 		goto out;
1264 	}
1265 
1266 	if (length > VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize)
1267 		return (EFBIG);
1268 
1269 	error = tmpfs_reg_resize(vp, length);
1270 	if (error == 0) {
1271 		node->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1272 	}
1273 
1274 out:
1275 	tmpfs_update(vp);
1276 
1277 	return error;
1278 }
1279