xref: /netbsd-src/sys/fs/tmpfs/tmpfs_vnops.c (revision 62a8debe1dc62962e18a1c918def78666141273b)
1 /*	$NetBSD: tmpfs_vnops.c,v 1.66 2010/01/08 11:35:09 pooka Exp $	*/
2 
3 /*
4  * Copyright (c) 2005, 2006, 2007 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  * tmpfs vnode interface.
35  */
36 
37 #include <sys/cdefs.h>
38 __KERNEL_RCSID(0, "$NetBSD: tmpfs_vnops.c,v 1.66 2010/01/08 11:35:09 pooka Exp $");
39 
40 #include <sys/param.h>
41 #include <sys/dirent.h>
42 #include <sys/fcntl.h>
43 #include <sys/event.h>
44 #include <sys/malloc.h>
45 #include <sys/namei.h>
46 #include <sys/proc.h>
47 #include <sys/stat.h>
48 #include <sys/uio.h>
49 #include <sys/unistd.h>
50 #include <sys/vnode.h>
51 #include <sys/lockf.h>
52 #include <sys/kauth.h>
53 
54 #include <uvm/uvm.h>
55 
56 #include <miscfs/fifofs/fifo.h>
57 #include <miscfs/genfs/genfs.h>
58 #include <fs/tmpfs/tmpfs_vnops.h>
59 #include <fs/tmpfs/tmpfs.h>
60 
61 /* --------------------------------------------------------------------- */
62 
63 /*
64  * vnode operations vector used for files stored in a tmpfs file system.
65  */
66 int (**tmpfs_vnodeop_p)(void *);
67 const struct vnodeopv_entry_desc tmpfs_vnodeop_entries[] = {
68 	{ &vop_default_desc,		vn_default_error },
69 	{ &vop_lookup_desc,		tmpfs_lookup },
70 	{ &vop_create_desc,		tmpfs_create },
71 	{ &vop_mknod_desc,		tmpfs_mknod },
72 	{ &vop_open_desc,		tmpfs_open },
73 	{ &vop_close_desc,		tmpfs_close },
74 	{ &vop_access_desc,		tmpfs_access },
75 	{ &vop_getattr_desc,		tmpfs_getattr },
76 	{ &vop_setattr_desc,		tmpfs_setattr },
77 	{ &vop_read_desc,		tmpfs_read },
78 	{ &vop_write_desc,		tmpfs_write },
79 	{ &vop_ioctl_desc,		tmpfs_ioctl },
80 	{ &vop_fcntl_desc,		tmpfs_fcntl },
81 	{ &vop_poll_desc,		tmpfs_poll },
82 	{ &vop_kqfilter_desc,		tmpfs_kqfilter },
83 	{ &vop_revoke_desc,		tmpfs_revoke },
84 	{ &vop_mmap_desc,		tmpfs_mmap },
85 	{ &vop_fsync_desc,		tmpfs_fsync },
86 	{ &vop_seek_desc,		tmpfs_seek },
87 	{ &vop_remove_desc,		tmpfs_remove },
88 	{ &vop_link_desc,		tmpfs_link },
89 	{ &vop_rename_desc,		tmpfs_rename },
90 	{ &vop_mkdir_desc,		tmpfs_mkdir },
91 	{ &vop_rmdir_desc,		tmpfs_rmdir },
92 	{ &vop_symlink_desc,		tmpfs_symlink },
93 	{ &vop_readdir_desc,		tmpfs_readdir },
94 	{ &vop_readlink_desc,		tmpfs_readlink },
95 	{ &vop_abortop_desc,		tmpfs_abortop },
96 	{ &vop_inactive_desc,		tmpfs_inactive },
97 	{ &vop_reclaim_desc,		tmpfs_reclaim },
98 	{ &vop_lock_desc,		tmpfs_lock },
99 	{ &vop_unlock_desc,		tmpfs_unlock },
100 	{ &vop_bmap_desc,		tmpfs_bmap },
101 	{ &vop_strategy_desc,		tmpfs_strategy },
102 	{ &vop_print_desc,		tmpfs_print },
103 	{ &vop_pathconf_desc,		tmpfs_pathconf },
104 	{ &vop_islocked_desc,		tmpfs_islocked },
105 	{ &vop_advlock_desc,		tmpfs_advlock },
106 	{ &vop_bwrite_desc,		tmpfs_bwrite },
107 	{ &vop_getpages_desc,		tmpfs_getpages },
108 	{ &vop_putpages_desc,		tmpfs_putpages },
109 	{ NULL, NULL }
110 };
111 const struct vnodeopv_desc tmpfs_vnodeop_opv_desc =
112 	{ &tmpfs_vnodeop_p, tmpfs_vnodeop_entries };
113 
114 /* --------------------------------------------------------------------- */
115 
116 int
117 tmpfs_lookup(void *v)
118 {
119 	struct vnode *dvp = ((struct vop_lookup_args *)v)->a_dvp;
120 	struct vnode **vpp = ((struct vop_lookup_args *)v)->a_vpp;
121 	struct componentname *cnp = ((struct vop_lookup_args *)v)->a_cnp;
122 
123 	int error;
124 	struct tmpfs_dirent *de;
125 	struct tmpfs_node *dnode;
126 
127 	KASSERT(VOP_ISLOCKED(dvp));
128 
129 	dnode = VP_TO_TMPFS_DIR(dvp);
130 	*vpp = NULL;
131 
132 	/* Check accessibility of requested node as a first step. */
133 	error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred);
134 	if (error != 0)
135 		goto out;
136 
137 	/* If requesting the last path component on a read-only file system
138 	 * with a write operation, deny it. */
139 	if ((cnp->cn_flags & ISLASTCN) &&
140 	    (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
141 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
142 		error = EROFS;
143 		goto out;
144 	}
145 
146 	/* Avoid doing a linear scan of the directory if the requested
147 	 * directory/name couple is already in the cache. */
148 	error = cache_lookup(dvp, vpp, cnp);
149 	if (error >= 0)
150 		goto out;
151 
152 	/* We cannot be requesting the parent directory of the root node. */
153 	KASSERT(IMPLIES(dnode->tn_type == VDIR &&
154 	    dnode->tn_spec.tn_dir.tn_parent == dnode,
155 	    !(cnp->cn_flags & ISDOTDOT)));
156 
157 	if (cnp->cn_flags & ISDOTDOT) {
158 		VOP_UNLOCK(dvp, 0);
159 
160 		/* Allocate a new vnode on the matching entry. */
161 		error = tmpfs_alloc_vp(dvp->v_mount,
162 		    dnode->tn_spec.tn_dir.tn_parent, vpp);
163 
164 		vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
165 	} else if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
166 		vref(dvp);
167 		*vpp = dvp;
168 		error = 0;
169 	} else {
170 		de = tmpfs_dir_lookup(dnode, cnp);
171 		if (de == NULL) {
172 			/* The entry was not found in the directory.
173 			 * This is OK iff we are creating or renaming an
174 			 * entry and are working on the last component of
175 			 * the path name. */
176 			if ((cnp->cn_flags & ISLASTCN) &&
177 			    (cnp->cn_nameiop == CREATE || \
178 			    cnp->cn_nameiop == RENAME)) {
179 				error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred);
180 				if (error != 0)
181 					goto out;
182 
183 				/* Keep the component name in the buffer for
184 				 * future uses. */
185 				cnp->cn_flags |= SAVENAME;
186 
187 				error = EJUSTRETURN;
188 			} else
189 				error = ENOENT;
190 		} else {
191 			struct tmpfs_node *tnode;
192 
193 			/* The entry was found, so get its associated
194 			 * tmpfs_node. */
195 			tnode = de->td_node;
196 
197 			/* If we are not at the last path component and
198 			 * found a non-directory or non-link entry (which
199 			 * may itself be pointing to a directory), raise
200 			 * an error. */
201 			if ((tnode->tn_type != VDIR &&
202 			    tnode->tn_type != VLNK) &&
203 			    !(cnp->cn_flags & ISLASTCN)) {
204 				error = ENOTDIR;
205 				goto out;
206 			}
207 
208 			/* Check permissions */
209 			if ((cnp->cn_flags & ISLASTCN) &&
210 			    (cnp->cn_nameiop == DELETE ||
211 			    cnp->cn_nameiop == RENAME)) {
212 				kauth_action_t action = 0;
213 
214 				/* This is the file-system's decision. */
215 				if ((dnode->tn_mode & S_ISTXT) != 0 &&
216 				    kauth_cred_geteuid(cnp->cn_cred) != dnode->tn_uid &&
217 				    kauth_cred_geteuid(cnp->cn_cred) != tnode->tn_uid)
218 					error = EPERM;
219 				else
220 					error = 0;
221 
222 				/* Only bother if we're not already failing it. */
223 				if (!error) {
224 					error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred);
225 				}
226 
227 				if (cnp->cn_nameiop == DELETE)
228 					action |= KAUTH_VNODE_DELETE;
229 				else /* if (cnp->cn_nameiop == RENAME) */
230 					action |= KAUTH_VNODE_RENAME;
231 
232 				error = kauth_authorize_vnode(cnp->cn_cred,
233 				    action, *vpp, dvp, error);
234 				if (error != 0)
235 					goto out;
236 
237 				cnp->cn_flags |= SAVENAME;
238 			} else
239 				de = NULL;
240 
241 			/* Allocate a new vnode on the matching entry. */
242 			error = tmpfs_alloc_vp(dvp->v_mount, tnode, vpp);
243 		}
244 	}
245 
246 	/* Store the result of this lookup in the cache.  Avoid this if the
247 	 * request was for creation, as it does not improve timings on
248 	 * emprical tests. */
249 	if ((cnp->cn_flags & MAKEENTRY) && cnp->cn_nameiop != CREATE &&
250 	    (cnp->cn_flags & ISDOTDOT) == 0)
251 		cache_enter(dvp, *vpp, cnp);
252 
253 out:
254 	/* If there were no errors, *vpp cannot be null and it must be
255 	 * locked. */
256 	KASSERT(IFF(error == 0, *vpp != NULL && VOP_ISLOCKED(*vpp)));
257 
258 	/* dvp must always be locked. */
259 	KASSERT(VOP_ISLOCKED(dvp));
260 
261 	return error;
262 }
263 
264 /* --------------------------------------------------------------------- */
265 
266 int
267 tmpfs_create(void *v)
268 {
269 	struct vnode *dvp = ((struct vop_create_args *)v)->a_dvp;
270 	struct vnode **vpp = ((struct vop_create_args *)v)->a_vpp;
271 	struct componentname *cnp = ((struct vop_create_args *)v)->a_cnp;
272 	struct vattr *vap = ((struct vop_create_args *)v)->a_vap;
273 
274 	KASSERT(vap->va_type == VREG || vap->va_type == VSOCK);
275 
276 	return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
277 }
278 /* --------------------------------------------------------------------- */
279 
280 int
281 tmpfs_mknod(void *v)
282 {
283 	struct vnode *dvp = ((struct vop_mknod_args *)v)->a_dvp;
284 	struct vnode **vpp = ((struct vop_mknod_args *)v)->a_vpp;
285 	struct componentname *cnp = ((struct vop_mknod_args *)v)->a_cnp;
286 	struct vattr *vap = ((struct vop_mknod_args *)v)->a_vap;
287 
288 	if (vap->va_type != VBLK && vap->va_type != VCHR &&
289 	    vap->va_type != VFIFO) {
290 		vput(dvp);
291 		return EINVAL;
292 	}
293 
294 	return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
295 }
296 
297 /* --------------------------------------------------------------------- */
298 
299 int
300 tmpfs_open(void *v)
301 {
302 	struct vnode *vp = ((struct vop_open_args *)v)->a_vp;
303 	int mode = ((struct vop_open_args *)v)->a_mode;
304 
305 	int error;
306 	struct tmpfs_node *node;
307 
308 	KASSERT(VOP_ISLOCKED(vp));
309 
310 	node = VP_TO_TMPFS_NODE(vp);
311 
312 	/* The file is still active but all its names have been removed
313 	 * (e.g. by a "rmdir $(pwd)").  It cannot be opened any more as
314 	 * it is about to die. */
315 	if (node->tn_links < 1) {
316 		error = ENOENT;
317 		goto out;
318 	}
319 
320 	/* If the file is marked append-only, deny write requests. */
321 	if (node->tn_flags & APPEND && (mode & (FWRITE | O_APPEND)) == FWRITE)
322 		error = EPERM;
323 	else
324 		error = 0;
325 
326 out:
327 	KASSERT(VOP_ISLOCKED(vp));
328 
329 	return error;
330 }
331 
332 /* --------------------------------------------------------------------- */
333 
334 int
335 tmpfs_close(void *v)
336 {
337 	struct vnode *vp = ((struct vop_close_args *)v)->a_vp;
338 
339 	struct tmpfs_node *node;
340 
341 	KASSERT(VOP_ISLOCKED(vp));
342 
343 	node = VP_TO_TMPFS_NODE(vp);
344 
345 	if (node->tn_links > 0) {
346 		/* Update node times.  No need to do it if the node has
347 		 * been deleted, because it will vanish after we return. */
348 		tmpfs_update(vp, NULL, NULL, NULL, UPDATE_CLOSE);
349 	}
350 
351 	return 0;
352 }
353 
354 /* --------------------------------------------------------------------- */
355 
356 static int
357 tmpfs_check_possible(struct vnode *vp, struct tmpfs_node *node, mode_t mode)
358 {
359 	int error = 0;
360 
361 	switch (vp->v_type) {
362 	case VDIR:
363 		/* FALLTHROUGH */
364 	case VLNK:
365 		/* FALLTHROUGH */
366 	case VREG:
367 		if (mode & VWRITE && vp->v_mount->mnt_flag & MNT_RDONLY) {
368 			error = EROFS;
369 			goto out;
370 		}
371 		break;
372 
373 	case VBLK:
374 		/* FALLTHROUGH */
375 	case VCHR:
376 		/* FALLTHROUGH */
377 	case VSOCK:
378 		/* FALLTHROUGH */
379 	case VFIFO:
380 		break;
381 
382 	default:
383 		error = EINVAL;
384 		goto out;
385 	}
386 
387 	if (mode & VWRITE && node->tn_flags & IMMUTABLE) {
388 		error = EPERM;
389 		goto out;
390 	}
391 
392  out:
393 	return error;
394 }
395 
396 static int
397 tmpfs_check_permitted(struct vnode *vp, struct tmpfs_node *node, mode_t mode,
398     kauth_cred_t cred)
399 {
400 
401 	return genfs_can_access(vp->v_type, node->tn_mode, node->tn_uid,
402 	    node->tn_gid, mode, cred);
403 }
404 
405 int
406 tmpfs_access(void *v)
407 {
408 	struct vnode *vp = ((struct vop_access_args *)v)->a_vp;
409 	int mode = ((struct vop_access_args *)v)->a_mode;
410 	kauth_cred_t cred = ((struct vop_access_args *)v)->a_cred;
411 
412 	int error;
413 	struct tmpfs_node *node;
414 
415 	KASSERT(VOP_ISLOCKED(vp));
416 
417 	node = VP_TO_TMPFS_NODE(vp);
418 
419 	error = tmpfs_check_possible(vp, node, mode);
420 	if (error)
421 		goto out;
422 
423 	error = tmpfs_check_permitted(vp, node, mode, cred);
424 
425 	error = kauth_authorize_vnode(cred, kauth_mode_to_action(mode), vp,
426 	    NULL, error);
427 
428 out:
429 	KASSERT(VOP_ISLOCKED(vp));
430 
431 	return error;
432 }
433 
434 /* --------------------------------------------------------------------- */
435 
436 int
437 tmpfs_getattr(void *v)
438 {
439 	struct vnode *vp = ((struct vop_getattr_args *)v)->a_vp;
440 	struct vattr *vap = ((struct vop_getattr_args *)v)->a_vap;
441 
442 	struct tmpfs_node *node;
443 
444 	node = VP_TO_TMPFS_NODE(vp);
445 
446 	vattr_null(vap);
447 
448 	tmpfs_itimes(vp, NULL, NULL, NULL);
449 
450 	vap->va_type = vp->v_type;
451 	vap->va_mode = node->tn_mode;
452 	vap->va_nlink = node->tn_links;
453 	vap->va_uid = node->tn_uid;
454 	vap->va_gid = node->tn_gid;
455 	vap->va_fsid = vp->v_mount->mnt_stat.f_fsidx.__fsid_val[0];
456 	vap->va_fileid = node->tn_id;
457 	vap->va_size = node->tn_size;
458 	vap->va_blocksize = PAGE_SIZE;
459 	vap->va_atime = node->tn_atime;
460 	vap->va_mtime = node->tn_mtime;
461 	vap->va_ctime = node->tn_ctime;
462 	vap->va_birthtime = node->tn_birthtime;
463 	vap->va_gen = node->tn_gen;
464 	vap->va_flags = node->tn_flags;
465 	vap->va_rdev = (vp->v_type == VBLK || vp->v_type == VCHR) ?
466 		node->tn_spec.tn_dev.tn_rdev : VNOVAL;
467 	vap->va_bytes = round_page(node->tn_size);
468 	vap->va_filerev = VNOVAL;
469 	vap->va_vaflags = 0;
470 	vap->va_spare = VNOVAL; /* XXX */
471 
472 	return 0;
473 }
474 
475 /* --------------------------------------------------------------------- */
476 
477 #define GOODTIME(tv)	((tv)->tv_sec != VNOVAL || (tv)->tv_nsec != VNOVAL)
478 /* XXX Should this operation be atomic?  I think it should, but code in
479  * XXX other places (e.g., ufs) doesn't seem to be... */
480 int
481 tmpfs_setattr(void *v)
482 {
483 	struct vnode *vp = ((struct vop_setattr_args *)v)->a_vp;
484 	struct vattr *vap = ((struct vop_setattr_args *)v)->a_vap;
485 	kauth_cred_t cred = ((struct vop_setattr_args *)v)->a_cred;
486 	struct lwp *l = curlwp;
487 
488 	int error;
489 
490 	KASSERT(VOP_ISLOCKED(vp));
491 
492 	error = 0;
493 
494 	/* Abort if any unsettable attribute is given. */
495 	if (vap->va_type != VNON ||
496 	    vap->va_nlink != VNOVAL ||
497 	    vap->va_fsid != VNOVAL ||
498 	    vap->va_fileid != VNOVAL ||
499 	    vap->va_blocksize != VNOVAL ||
500 	    GOODTIME(&vap->va_ctime) ||
501 	    vap->va_gen != VNOVAL ||
502 	    vap->va_rdev != VNOVAL ||
503 	    vap->va_bytes != VNOVAL)
504 		error = EINVAL;
505 
506 	if (error == 0 && (vap->va_flags != VNOVAL))
507 		error = tmpfs_chflags(vp, vap->va_flags, cred, l);
508 
509 	if (error == 0 && (vap->va_size != VNOVAL))
510 		error = tmpfs_chsize(vp, vap->va_size, cred, l);
511 
512 	if (error == 0 && (vap->va_uid != VNOVAL || vap->va_gid != VNOVAL))
513 		error = tmpfs_chown(vp, vap->va_uid, vap->va_gid, cred, l);
514 
515 	if (error == 0 && (vap->va_mode != VNOVAL))
516 		error = tmpfs_chmod(vp, vap->va_mode, cred, l);
517 
518 	if (error == 0 && (GOODTIME(&vap->va_atime) || GOODTIME(&vap->va_mtime)
519 	    || GOODTIME(&vap->va_birthtime)))
520 		if ((error = tmpfs_chtimes(vp, &vap->va_atime, &vap->va_mtime,
521 		    &vap->va_birthtime, vap->va_vaflags, cred, l)) == 0)
522 			return 0;
523 
524 	/* Update the node times.  We give preference to the error codes
525 	 * generated by this function rather than the ones that may arise
526 	 * from tmpfs_update. */
527 	tmpfs_update(vp, NULL, NULL, NULL, 0);
528 
529 	KASSERT(VOP_ISLOCKED(vp));
530 
531 	return error;
532 }
533 
534 /* --------------------------------------------------------------------- */
535 
536 int
537 tmpfs_read(void *v)
538 {
539 	struct vnode *vp = ((struct vop_read_args *)v)->a_vp;
540 	struct uio *uio = ((struct vop_read_args *)v)->a_uio;
541 	int ioflag = ((struct vop_read_args *)v)->a_ioflag;
542 
543 	int error;
544 	struct tmpfs_node *node;
545 	struct uvm_object *uobj;
546 
547 	KASSERT(VOP_ISLOCKED(vp));
548 
549 	node = VP_TO_TMPFS_NODE(vp);
550 
551 	if (vp->v_type != VREG) {
552 		error = EISDIR;
553 		goto out;
554 	}
555 
556 	if (uio->uio_offset < 0) {
557 		error = EINVAL;
558 		goto out;
559 	}
560 
561 	node->tn_status |= TMPFS_NODE_ACCESSED;
562 
563 	uobj = node->tn_spec.tn_reg.tn_aobj;
564 	error = 0;
565 	while (error == 0 && uio->uio_resid > 0) {
566 		vsize_t len;
567 
568 		if (node->tn_size <= uio->uio_offset)
569 			break;
570 
571 		len = MIN(node->tn_size - uio->uio_offset, uio->uio_resid);
572 		if (len == 0)
573 			break;
574 
575 		error = ubc_uiomove(uobj, uio, len, IO_ADV_DECODE(ioflag),
576 		    UBC_READ | UBC_PARTIALOK | UBC_UNMAP_FLAG(vp));
577 	}
578 
579 out:
580 	KASSERT(VOP_ISLOCKED(vp));
581 
582 	return error;
583 }
584 
585 /* --------------------------------------------------------------------- */
586 
587 int
588 tmpfs_write(void *v)
589 {
590 	struct vnode *vp = ((struct vop_write_args *)v)->a_vp;
591 	struct uio *uio = ((struct vop_write_args *)v)->a_uio;
592 	int ioflag = ((struct vop_write_args *)v)->a_ioflag;
593 
594 	bool extended;
595 	int error;
596 	off_t oldsize;
597 	struct proc *p = curproc;
598 	struct tmpfs_node *node;
599 	struct uvm_object *uobj;
600 
601 	KASSERT(VOP_ISLOCKED(vp));
602 
603 	node = VP_TO_TMPFS_NODE(vp);
604 	oldsize = node->tn_size;
605 
606 	if (uio->uio_offset < 0 || vp->v_type != VREG) {
607 		error = EINVAL;
608 		goto out;
609 	}
610 
611 	if (uio->uio_resid == 0) {
612 		error = 0;
613 		goto out;
614 	}
615 
616 	if (((uio->uio_offset + uio->uio_resid) >
617 	    p->p_rlimit[RLIMIT_FSIZE].rlim_cur)) {
618 		mutex_enter(proc_lock);
619 		psignal(p, SIGXFSZ);
620 		mutex_exit(proc_lock);
621 		error = EFBIG;
622 		goto out;
623 	}
624 
625 	if (ioflag & IO_APPEND)
626 		uio->uio_offset = node->tn_size;
627 
628 	extended = uio->uio_offset + uio->uio_resid > node->tn_size;
629 	if (extended) {
630 		error = tmpfs_reg_resize(vp, uio->uio_offset + uio->uio_resid);
631 		if (error != 0)
632 			goto out;
633 	}
634 
635 	uobj = node->tn_spec.tn_reg.tn_aobj;
636 	error = 0;
637 	while (error == 0 && uio->uio_resid > 0) {
638 		vsize_t len;
639 
640 		len = MIN(node->tn_size - uio->uio_offset, uio->uio_resid);
641 		if (len == 0)
642 			break;
643 
644 		error = ubc_uiomove(uobj, uio, len, IO_ADV_DECODE(ioflag),
645 		    UBC_WRITE | UBC_UNMAP_FLAG(vp));
646 	}
647 
648 	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
649 	    (extended ? TMPFS_NODE_CHANGED : 0);
650 
651 	if (error != 0)
652 		(void)tmpfs_reg_resize(vp, oldsize);
653 
654 	VN_KNOTE(vp, NOTE_WRITE);
655 
656 out:
657 	KASSERT(VOP_ISLOCKED(vp));
658 	KASSERT(IMPLIES(error == 0, uio->uio_resid == 0));
659 	KASSERT(IMPLIES(error != 0, oldsize == node->tn_size));
660 
661 	return error;
662 }
663 
664 /* --------------------------------------------------------------------- */
665 
666 int
667 tmpfs_fsync(void *v)
668 {
669 	struct vnode *vp = ((struct vop_fsync_args *)v)->a_vp;
670 
671 	KASSERT(VOP_ISLOCKED(vp));
672 
673 	tmpfs_update(vp, NULL, NULL, NULL, 0);
674 
675 	return 0;
676 }
677 
678 /* --------------------------------------------------------------------- */
679 
680 int
681 tmpfs_remove(void *v)
682 {
683 	struct vnode *dvp = ((struct vop_remove_args *)v)->a_dvp;
684 	struct vnode *vp = ((struct vop_remove_args *)v)->a_vp;
685 	struct componentname *cnp = (((struct vop_remove_args *)v)->a_cnp);
686 
687 	int error;
688 	struct tmpfs_dirent *de;
689 	struct tmpfs_mount *tmp;
690 	struct tmpfs_node *dnode;
691 	struct tmpfs_node *node;
692 
693 	KASSERT(VOP_ISLOCKED(dvp));
694 	KASSERT(VOP_ISLOCKED(vp));
695 
696 	if (vp->v_type == VDIR) {
697 		error = EPERM;
698 		goto out;
699 	}
700 
701 	dnode = VP_TO_TMPFS_DIR(dvp);
702 	node = VP_TO_TMPFS_NODE(vp);
703 	tmp = VFS_TO_TMPFS(vp->v_mount);
704 	de = tmpfs_dir_lookup(dnode, cnp);
705 	KASSERT(de);
706 	KASSERT(de->td_node == node);
707 
708 	/* Files marked as immutable or append-only cannot be deleted. */
709 	if (node->tn_flags & (IMMUTABLE | APPEND)) {
710 		error = EPERM;
711 		goto out;
712 	}
713 
714 	/* Remove the entry from the directory; as it is a file, we do not
715 	 * have to change the number of hard links of the directory. */
716 	tmpfs_dir_detach(dvp, de);
717 
718 	/* Free the directory entry we just deleted.  Note that the node
719 	 * referred by it will not be removed until the vnode is really
720 	 * reclaimed. */
721 	tmpfs_free_dirent(tmp, de, true);
722 
723 	error = 0;
724 
725 out:
726 	vput(vp);
727 	if (dvp == vp)
728 		vrele(dvp);
729 	else
730 		vput(dvp);
731 	if (cnp->cn_flags & HASBUF) {
732 		PNBUF_PUT(cnp->cn_pnbuf);
733 		cnp->cn_flags &= ~HASBUF;
734 	}
735 
736 	return error;
737 }
738 
739 /* --------------------------------------------------------------------- */
740 
741 int
742 tmpfs_link(void *v)
743 {
744 	struct vnode *dvp = ((struct vop_link_args *)v)->a_dvp;
745 	struct vnode *vp = ((struct vop_link_args *)v)->a_vp;
746 	struct componentname *cnp = ((struct vop_link_args *)v)->a_cnp;
747 
748 	int error;
749 	struct tmpfs_dirent *de;
750 	struct tmpfs_node *dnode;
751 	struct tmpfs_node *node;
752 
753 	KASSERT(VOP_ISLOCKED(dvp));
754 	KASSERT(cnp->cn_flags & HASBUF);
755 	KASSERT(dvp != vp); /* XXX When can this be false? */
756 
757 	dnode = VP_TO_TMPFS_DIR(dvp);
758 	node = VP_TO_TMPFS_NODE(vp);
759 
760 	/* Lock vp because we will need to run tmpfs_update over it, which
761 	 * needs the vnode to be locked. */
762 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
763 
764 	/* XXX: Why aren't the following two tests done by the caller? */
765 
766 	/* Hard links of directories are forbidden. */
767 	if (vp->v_type == VDIR) {
768 		error = EPERM;
769 		goto out;
770 	}
771 
772 	/* Cannot create cross-device links. */
773 	if (dvp->v_mount != vp->v_mount) {
774 		error = EXDEV;
775 		goto out;
776 	}
777 
778 	/* Ensure that we do not overflow the maximum number of links imposed
779 	 * by the system. */
780 	KASSERT(node->tn_links <= LINK_MAX);
781 	if (node->tn_links == LINK_MAX) {
782 		error = EMLINK;
783 		goto out;
784 	}
785 
786 	/* We cannot create links of files marked immutable or append-only. */
787 	if (node->tn_flags & (IMMUTABLE | APPEND)) {
788 		error = EPERM;
789 		goto out;
790 	}
791 
792 	/* Allocate a new directory entry to represent the node. */
793 	error = tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), node,
794 	    cnp->cn_nameptr, cnp->cn_namelen, &de);
795 	if (error != 0)
796 		goto out;
797 
798 	/* Insert the new directory entry into the appropriate directory. */
799 	tmpfs_dir_attach(dvp, de);
800 
801 	/* vp link count has changed, so update node times. */
802 	node->tn_status |= TMPFS_NODE_CHANGED;
803 	tmpfs_update(vp, NULL, NULL, NULL, 0);
804 
805 	error = 0;
806 
807 out:
808 	VOP_UNLOCK(vp, 0);
809 	PNBUF_PUT(cnp->cn_pnbuf);
810 	vput(dvp);
811 
812 	return error;
813 }
814 
815 /*
816  * tmpfs_rename: rename routine.
817  *
818  * Arguments: fdvp (from-parent vnode), fvp (from-leaf), tdvp (to-parent)
819  * and tvp (to-leaf), if exists (NULL if not).
820  *
821  * => Caller holds a reference on fdvp and fvp, they are unlocked.
822  *    Note: fdvp and fvp can refer to the same object (i.e. when it is root).
823  *
824  * => Both tdvp and tvp are referenced and locked.  It is our responsibility
825  *    to release the references and unlock them (or destroy).
826  */
827 int
828 tmpfs_rename(void *v)
829 {
830 	struct vnode *fdvp = ((struct vop_rename_args *)v)->a_fdvp;
831 	struct vnode *fvp = ((struct vop_rename_args *)v)->a_fvp;
832 	struct componentname *fcnp = ((struct vop_rename_args *)v)->a_fcnp;
833 	struct vnode *tdvp = ((struct vop_rename_args *)v)->a_tdvp;
834 	struct vnode *tvp = ((struct vop_rename_args *)v)->a_tvp;
835 	struct componentname *tcnp = ((struct vop_rename_args *)v)->a_tcnp;
836 
837 	char *newname;
838 	int error;
839 	struct tmpfs_dirent *de, *de2;
840 	struct tmpfs_mount *tmp;
841 	struct tmpfs_node *fdnode;
842 	struct tmpfs_node *fnode;
843 	struct tmpfs_node *tnode;
844 	struct tmpfs_node *tdnode;
845 	size_t namelen;
846 
847 	KASSERT(VOP_ISLOCKED(tdvp));
848 	KASSERT(IMPLIES(tvp != NULL, VOP_ISLOCKED(tvp) == LK_EXCLUSIVE));
849 	KASSERT(fcnp->cn_flags & HASBUF);
850 	KASSERT(tcnp->cn_flags & HASBUF);
851 
852 	newname = NULL;
853 	namelen = 0;
854 	tmp = NULL;
855 
856 	/* Disallow cross-device renames. */
857 	if (fvp->v_mount != tdvp->v_mount ||
858 	    (tvp != NULL && fvp->v_mount != tvp->v_mount)) {
859 		error = EXDEV;
860 		goto out_unlocked;
861 	}
862 
863 	fnode = VP_TO_TMPFS_NODE(fvp);
864 	fdnode = VP_TO_TMPFS_DIR(fdvp);
865 	tnode = (tvp == NULL) ? NULL : VP_TO_TMPFS_NODE(tvp);
866 	tdnode = VP_TO_TMPFS_DIR(tdvp);
867 	tmp = VFS_TO_TMPFS(tdvp->v_mount);
868 
869 	if (fdvp == tvp) {
870 		error = 0;
871 		goto out_unlocked;
872 	}
873 
874 	/* If we need to move the directory between entries, lock the
875 	 * source so that we can safely operate on it. */
876 
877 	/* XXX: this is a potential locking order violation! */
878 	if (fdnode != tdnode) {
879 		vn_lock(fdvp, LK_EXCLUSIVE | LK_RETRY);
880 	}
881 
882 	/*
883 	 * If the node we were renaming has scarpered, just give up.
884 	 */
885 	de = tmpfs_dir_lookup(fdnode, fcnp);
886 	if (de == NULL || de->td_node != fnode) {
887 		error = ENOENT;
888 		goto out;
889 	}
890 
891 	/* If source and target is the same vnode, remove the source link. */
892 	if (fvp == tvp) {
893 		/*
894 		 * Detach and free the directory entry.  Drops the link
895 		 * count on the node.
896 		 */
897 		tmpfs_dir_detach(fdvp, de);
898 		tmpfs_free_dirent(VFS_TO_TMPFS(fvp->v_mount), de, true);
899 		VN_KNOTE(fdvp, NOTE_WRITE);
900 		goto out_ok;
901 	}
902 
903 	/* If replacing an existing entry, ensure we can do the operation. */
904 	if (tvp != NULL) {
905 		KASSERT(tnode != NULL);
906 		if (fnode->tn_type == VDIR && tnode->tn_type == VDIR) {
907 			if (tnode->tn_size > 0) {
908 				error = ENOTEMPTY;
909 				goto out;
910 			}
911 		} else if (fnode->tn_type == VDIR && tnode->tn_type != VDIR) {
912 			error = ENOTDIR;
913 			goto out;
914 		} else if (fnode->tn_type != VDIR && tnode->tn_type == VDIR) {
915 			error = EISDIR;
916 			goto out;
917 		} else {
918 			KASSERT(fnode->tn_type != VDIR &&
919 			        tnode->tn_type != VDIR);
920 		}
921 	}
922 
923 	/* Ensure that we have enough memory to hold the new name, if it
924 	 * has to be changed. */
925 	namelen = tcnp->cn_namelen;
926 	if (fcnp->cn_namelen != tcnp->cn_namelen ||
927 	    memcmp(fcnp->cn_nameptr, tcnp->cn_nameptr, fcnp->cn_namelen) != 0) {
928 		newname = tmpfs_str_pool_get(&tmp->tm_str_pool, namelen, 0);
929 		if (newname == NULL) {
930 			error = ENOSPC;
931 			goto out;
932 		}
933 	}
934 
935 	/* If the node is being moved to another directory, we have to do
936 	 * the move. */
937 	if (fdnode != tdnode) {
938 		/* In case we are moving a directory, we have to adjust its
939 		 * parent to point to the new parent. */
940 		if (de->td_node->tn_type == VDIR) {
941 			struct tmpfs_node *n;
942 
943 			/* Ensure the target directory is not a child of the
944 			 * directory being moved.  Otherwise, we'd end up
945 			 * with stale nodes. */
946 			n = tdnode;
947 			while (n != n->tn_spec.tn_dir.tn_parent) {
948 				if (n == fnode) {
949 					error = EINVAL;
950 					goto out;
951 				}
952 				n = n->tn_spec.tn_dir.tn_parent;
953 			}
954 
955 			/* Adjust the parent pointer. */
956 			TMPFS_VALIDATE_DIR(fnode);
957 			de->td_node->tn_spec.tn_dir.tn_parent = tdnode;
958 
959 			/* As a result of changing the target of the '..'
960 			 * entry, the link count of the source and target
961 			 * directories has to be adjusted. */
962 			fdnode->tn_links--;
963 			tdnode->tn_links++;
964 		}
965 
966 		/* Do the move: just remove the entry from the source directory
967 		 * and insert it into the target one. */
968 		tmpfs_dir_detach(fdvp, de);
969 		tmpfs_dir_attach(tdvp, de);
970 
971 		/* Notify listeners of fdvp about the change in the directory.
972 		 * We can do it at this point because we aren't touching fdvp
973 		 * any more below. */
974 		VN_KNOTE(fdvp, NOTE_WRITE);
975 	}
976 
977 	/* If we are overwriting an entry, we have to remove the old one
978 	 * from the target directory. */
979 	if (tvp != NULL) {
980 		KASSERT(tnode != NULL);
981 
982 		/* Remove the old entry from the target directory.
983 		 * Note! This relies on tmpfs_dir_attach() putting the new
984 		 * node on the end of the target's node list. */
985 		de2 = tmpfs_dir_lookup(tdnode, tcnp);
986 		KASSERT(de2 != NULL);
987 		KASSERT(de2->td_node == tnode);
988 		tmpfs_dir_detach(tdvp, de2);
989 
990 		/* Free the directory entry we just deleted.  Note that the
991 		 * node referred by it will not be removed until the vnode is
992 		 * really reclaimed. */
993 		tmpfs_free_dirent(VFS_TO_TMPFS(tvp->v_mount), de2, true);
994 	}
995 
996 	/* If the name has changed, we need to make it effective by changing
997 	 * it in the directory entry. */
998 	if (newname != NULL) {
999 		KASSERT(tcnp->cn_namelen < MAXNAMLEN);
1000 		KASSERT(tcnp->cn_namelen < 0xffff);
1001 
1002 		tmpfs_str_pool_put(&tmp->tm_str_pool, de->td_name,
1003 		    de->td_namelen);
1004 		de->td_namelen = (uint16_t)namelen;
1005 		memcpy(newname, tcnp->cn_nameptr, namelen);
1006 		de->td_name = newname;
1007 		newname = NULL;
1008 
1009 		fnode->tn_status |= TMPFS_NODE_CHANGED;
1010 		tdnode->tn_status |= TMPFS_NODE_MODIFIED;
1011 	}
1012  out_ok:
1013 	/* Notify listeners of tdvp about the change in the directory (either
1014 	 * because a new entry was added or because one was removed) and
1015 	 * listeners of fvp about the rename. */
1016 	VN_KNOTE(tdvp, NOTE_WRITE);
1017 	VN_KNOTE(fvp, NOTE_RENAME);
1018 
1019 	error = 0;
1020 
1021  out:
1022 	if (fdnode != tdnode)
1023 		VOP_UNLOCK(fdvp, 0);
1024 
1025  out_unlocked:
1026 	/* Release target nodes. */
1027 	if (tdvp == tvp)
1028 		vrele(tdvp);
1029 	else
1030 		vput(tdvp);
1031 	if (tvp != NULL)
1032 		vput(tvp);
1033 
1034 	/* Release source nodes. */
1035 	vrele(fdvp);
1036 	vrele(fvp);
1037 
1038 	if (newname != NULL)
1039 		tmpfs_str_pool_put(&tmp->tm_str_pool, newname, namelen);
1040 
1041 	return error;
1042 }
1043 
1044 /* --------------------------------------------------------------------- */
1045 
1046 int
1047 tmpfs_mkdir(void *v)
1048 {
1049 	struct vnode *dvp = ((struct vop_mkdir_args *)v)->a_dvp;
1050 	struct vnode **vpp = ((struct vop_mkdir_args *)v)->a_vpp;
1051 	struct componentname *cnp = ((struct vop_mkdir_args *)v)->a_cnp;
1052 	struct vattr *vap = ((struct vop_mkdir_args *)v)->a_vap;
1053 
1054 	KASSERT(vap->va_type == VDIR);
1055 
1056 	return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
1057 }
1058 
1059 /* --------------------------------------------------------------------- */
1060 
1061 int
1062 tmpfs_rmdir(void *v)
1063 {
1064 	struct vnode *dvp = ((struct vop_rmdir_args *)v)->a_dvp;
1065 	struct vnode *vp = ((struct vop_rmdir_args *)v)->a_vp;
1066 	struct componentname *cnp = ((struct vop_rmdir_args *)v)->a_cnp;
1067 
1068 	int error;
1069 	struct tmpfs_dirent *de;
1070 	struct tmpfs_mount *tmp;
1071 	struct tmpfs_node *dnode;
1072 	struct tmpfs_node *node;
1073 
1074 	KASSERT(VOP_ISLOCKED(dvp));
1075 	KASSERT(VOP_ISLOCKED(vp));
1076 
1077 	tmp = VFS_TO_TMPFS(dvp->v_mount);
1078 	dnode = VP_TO_TMPFS_DIR(dvp);
1079 	node = VP_TO_TMPFS_DIR(vp);
1080 	error = 0;
1081 
1082 	/* Directories with more than two entries ('.' and '..') cannot be
1083 	 * removed. */
1084 	if (node->tn_size > 0) {
1085 		error = ENOTEMPTY;
1086 		goto out;
1087 	}
1088 
1089 	/* This invariant holds only if we are not trying to remove "..".
1090 	 * We checked for that above so this is safe now. */
1091 	KASSERT(node->tn_spec.tn_dir.tn_parent == dnode);
1092 
1093 	/* Get the directory entry associated with node (vp). */
1094 	de = tmpfs_dir_lookup(dnode, cnp);
1095 	KASSERT(de);
1096 	KASSERT(de->td_node == node);
1097 
1098 	/* Check flags to see if we are allowed to remove the directory. */
1099 	if (dnode->tn_flags & APPEND || node->tn_flags & (IMMUTABLE | APPEND)) {
1100 		error = EPERM;
1101 		goto out;
1102 	}
1103 
1104 	/* Detach the directory entry from the directory (dnode). */
1105 	tmpfs_dir_detach(dvp, de);
1106 
1107 	node->tn_links--;
1108 	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
1109 	    TMPFS_NODE_MODIFIED;
1110 	node->tn_spec.tn_dir.tn_parent->tn_links--;
1111 	node->tn_spec.tn_dir.tn_parent->tn_status |= TMPFS_NODE_ACCESSED | \
1112 	    TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1113 
1114 	/* Release the parent. */
1115 	cache_purge(dvp); /* XXX Is this needed? */
1116 
1117 	/* Free the directory entry we just deleted.  Note that the node
1118 	 * referred by it will not be removed until the vnode is really
1119 	 * reclaimed. */
1120 	tmpfs_free_dirent(tmp, de, true);
1121 
1122 	KASSERT(node->tn_links == 0);
1123  out:
1124 	/* Release the nodes. */
1125 	vput(dvp);
1126 	vput(vp);
1127 	PNBUF_PUT(cnp->cn_pnbuf);
1128 
1129 	return error;
1130 }
1131 
1132 /* --------------------------------------------------------------------- */
1133 
1134 int
1135 tmpfs_symlink(void *v)
1136 {
1137 	struct vnode *dvp = ((struct vop_symlink_args *)v)->a_dvp;
1138 	struct vnode **vpp = ((struct vop_symlink_args *)v)->a_vpp;
1139 	struct componentname *cnp = ((struct vop_symlink_args *)v)->a_cnp;
1140 	struct vattr *vap = ((struct vop_symlink_args *)v)->a_vap;
1141 	char *target = ((struct vop_symlink_args *)v)->a_target;
1142 
1143 	KASSERT(vap->va_type == VLNK);
1144 
1145 	return tmpfs_alloc_file(dvp, vpp, vap, cnp, target);
1146 }
1147 
1148 /* --------------------------------------------------------------------- */
1149 
1150 int
1151 tmpfs_readdir(void *v)
1152 {
1153 	struct vnode *vp = ((struct vop_readdir_args *)v)->a_vp;
1154 	struct uio *uio = ((struct vop_readdir_args *)v)->a_uio;
1155 	int *eofflag = ((struct vop_readdir_args *)v)->a_eofflag;
1156 	off_t **cookies = ((struct vop_readdir_args *)v)->a_cookies;
1157 	int *ncookies = ((struct vop_readdir_args *)v)->a_ncookies;
1158 
1159 	int error;
1160 	off_t startoff;
1161 	off_t cnt;
1162 	struct tmpfs_node *node;
1163 
1164 	KASSERT(VOP_ISLOCKED(vp));
1165 
1166 	/* This operation only makes sense on directory nodes. */
1167 	if (vp->v_type != VDIR) {
1168 		error = ENOTDIR;
1169 		goto out;
1170 	}
1171 
1172 	node = VP_TO_TMPFS_DIR(vp);
1173 
1174 	startoff = uio->uio_offset;
1175 
1176 	cnt = 0;
1177 	if (uio->uio_offset == TMPFS_DIRCOOKIE_DOT) {
1178 		error = tmpfs_dir_getdotdent(node, uio);
1179 		if (error == -1) {
1180 			error = 0;
1181 			goto outok;
1182 		} else if (error != 0)
1183 			goto outok;
1184 		cnt++;
1185 	}
1186 
1187 	if (uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT) {
1188 		error = tmpfs_dir_getdotdotdent(node, uio);
1189 		if (error == -1) {
1190 			error = 0;
1191 			goto outok;
1192 		} else if (error != 0)
1193 			goto outok;
1194 		cnt++;
1195 	}
1196 
1197 	error = tmpfs_dir_getdents(node, uio, &cnt);
1198 	if (error == -1)
1199 		error = 0;
1200 	KASSERT(error >= 0);
1201 
1202 outok:
1203 	/* This label assumes that startoff has been
1204 	 * initialized.  If the compiler didn't spit out warnings, we'd
1205 	 * simply make this one be 'out' and drop 'outok'. */
1206 
1207 	if (eofflag != NULL)
1208 		*eofflag =
1209 		    (error == 0 && uio->uio_offset == TMPFS_DIRCOOKIE_EOF);
1210 
1211 	/* Update NFS-related variables. */
1212 	if (error == 0 && cookies != NULL && ncookies != NULL) {
1213 		off_t i;
1214 		off_t off = startoff;
1215 		struct tmpfs_dirent *de = NULL;
1216 
1217 		*ncookies = cnt;
1218 		*cookies = malloc(cnt * sizeof(off_t), M_TEMP, M_WAITOK);
1219 
1220 		for (i = 0; i < cnt; i++) {
1221 			KASSERT(off != TMPFS_DIRCOOKIE_EOF);
1222 			if (off == TMPFS_DIRCOOKIE_DOT) {
1223 				off = TMPFS_DIRCOOKIE_DOTDOT;
1224 			} else {
1225 				if (off == TMPFS_DIRCOOKIE_DOTDOT) {
1226 					de = TAILQ_FIRST(&node->tn_spec.
1227 					    tn_dir.tn_dir);
1228 				} else if (de != NULL) {
1229 					de = TAILQ_NEXT(de, td_entries);
1230 				} else {
1231 					de = tmpfs_dir_lookupbycookie(node,
1232 					    off);
1233 					KASSERT(de != NULL);
1234 					de = TAILQ_NEXT(de, td_entries);
1235 				}
1236 				if (de == NULL) {
1237 					off = TMPFS_DIRCOOKIE_EOF;
1238 				} else {
1239 					off = tmpfs_dircookie(de);
1240 				}
1241 			}
1242 
1243 			(*cookies)[i] = off;
1244 		}
1245 		KASSERT(uio->uio_offset == off);
1246 	}
1247 
1248 out:
1249 	KASSERT(VOP_ISLOCKED(vp));
1250 
1251 	return error;
1252 }
1253 
1254 /* --------------------------------------------------------------------- */
1255 
1256 int
1257 tmpfs_readlink(void *v)
1258 {
1259 	struct vnode *vp = ((struct vop_readlink_args *)v)->a_vp;
1260 	struct uio *uio = ((struct vop_readlink_args *)v)->a_uio;
1261 
1262 	int error;
1263 	struct tmpfs_node *node;
1264 
1265 	KASSERT(VOP_ISLOCKED(vp));
1266 	KASSERT(uio->uio_offset == 0);
1267 	KASSERT(vp->v_type == VLNK);
1268 
1269 	node = VP_TO_TMPFS_NODE(vp);
1270 
1271 	error = uiomove(node->tn_spec.tn_lnk.tn_link,
1272 	    MIN(node->tn_size, uio->uio_resid), uio);
1273 	node->tn_status |= TMPFS_NODE_ACCESSED;
1274 
1275 	KASSERT(VOP_ISLOCKED(vp));
1276 
1277 	return error;
1278 }
1279 
1280 /* --------------------------------------------------------------------- */
1281 
1282 int
1283 tmpfs_inactive(void *v)
1284 {
1285 	struct vnode *vp = ((struct vop_inactive_args *)v)->a_vp;
1286 
1287 	struct tmpfs_node *node;
1288 
1289 	KASSERT(VOP_ISLOCKED(vp));
1290 
1291 	node = VP_TO_TMPFS_NODE(vp);
1292 	*((struct vop_inactive_args *)v)->a_recycle = (node->tn_links == 0);
1293 	VOP_UNLOCK(vp, 0);
1294 
1295 	return 0;
1296 }
1297 
1298 /* --------------------------------------------------------------------- */
1299 
1300 int
1301 tmpfs_reclaim(void *v)
1302 {
1303 	struct vnode *vp = ((struct vop_reclaim_args *)v)->a_vp;
1304 
1305 	struct tmpfs_mount *tmp;
1306 	struct tmpfs_node *node;
1307 
1308 	node = VP_TO_TMPFS_NODE(vp);
1309 	tmp = VFS_TO_TMPFS(vp->v_mount);
1310 
1311 	cache_purge(vp);
1312 	tmpfs_free_vp(vp);
1313 
1314 	/* If the node referenced by this vnode was deleted by the user,
1315 	 * we must free its associated data structures (now that the vnode
1316 	 * is being reclaimed). */
1317 	if (node->tn_links == 0)
1318 		tmpfs_free_node(tmp, node);
1319 
1320 	KASSERT(vp->v_data == NULL);
1321 
1322 	return 0;
1323 }
1324 
1325 /* --------------------------------------------------------------------- */
1326 
1327 int
1328 tmpfs_print(void *v)
1329 {
1330 	struct vnode *vp = ((struct vop_print_args *)v)->a_vp;
1331 
1332 	struct tmpfs_node *node;
1333 
1334 	node = VP_TO_TMPFS_NODE(vp);
1335 
1336 	printf("tag VT_TMPFS, tmpfs_node %p, flags 0x%x, links %d\n",
1337 	    node, node->tn_flags, node->tn_links);
1338 	printf("\tmode 0%o, owner %d, group %d, size %" PRIdMAX
1339 	    ", status 0x%x\n",
1340 	    node->tn_mode, node->tn_uid, node->tn_gid,
1341 	    (uintmax_t)node->tn_size, node->tn_status);
1342 	if (vp->v_type == VFIFO)
1343 		fifo_printinfo(vp);
1344 	printf("\n");
1345 
1346 	return 0;
1347 }
1348 
1349 /* --------------------------------------------------------------------- */
1350 
1351 int
1352 tmpfs_pathconf(void *v)
1353 {
1354 	int name = ((struct vop_pathconf_args *)v)->a_name;
1355 	register_t *retval = ((struct vop_pathconf_args *)v)->a_retval;
1356 
1357 	int error;
1358 
1359 	error = 0;
1360 
1361 	switch (name) {
1362 	case _PC_LINK_MAX:
1363 		*retval = LINK_MAX;
1364 		break;
1365 
1366 	case _PC_NAME_MAX:
1367 		*retval = NAME_MAX;
1368 		break;
1369 
1370 	case _PC_PATH_MAX:
1371 		*retval = PATH_MAX;
1372 		break;
1373 
1374 	case _PC_PIPE_BUF:
1375 		*retval = PIPE_BUF;
1376 		break;
1377 
1378 	case _PC_CHOWN_RESTRICTED:
1379 		*retval = 1;
1380 		break;
1381 
1382 	case _PC_NO_TRUNC:
1383 		*retval = 1;
1384 		break;
1385 
1386 	case _PC_SYNC_IO:
1387 		*retval = 1;
1388 		break;
1389 
1390 	case _PC_FILESIZEBITS:
1391 		*retval = 0; /* XXX Don't know which value should I return. */
1392 		break;
1393 
1394 	default:
1395 		error = EINVAL;
1396 	}
1397 
1398 	return error;
1399 }
1400 
1401 /* --------------------------------------------------------------------- */
1402 
1403 int
1404 tmpfs_advlock(void *v)
1405 {
1406 	struct vnode *vp = ((struct vop_advlock_args *)v)->a_vp;
1407 
1408 	struct tmpfs_node *node;
1409 
1410 	node = VP_TO_TMPFS_NODE(vp);
1411 
1412 	return lf_advlock(v, &node->tn_lockf, node->tn_size);
1413 }
1414 
1415 /* --------------------------------------------------------------------- */
1416 
1417 int
1418 tmpfs_getpages(void *v)
1419 {
1420 	struct vnode *vp = ((struct vop_getpages_args *)v)->a_vp;
1421 	voff_t offset = ((struct vop_getpages_args *)v)->a_offset;
1422 	struct vm_page **m = ((struct vop_getpages_args *)v)->a_m;
1423 	int *count = ((struct vop_getpages_args *)v)->a_count;
1424 	int centeridx = ((struct vop_getpages_args *)v)->a_centeridx;
1425 	vm_prot_t access_type = ((struct vop_getpages_args *)v)->a_access_type;
1426 	int advice = ((struct vop_getpages_args *)v)->a_advice;
1427 	int flags = ((struct vop_getpages_args *)v)->a_flags;
1428 
1429 	int error;
1430 	int i;
1431 	struct tmpfs_node *node;
1432 	struct uvm_object *uobj;
1433 	int npages = *count;
1434 
1435 	KASSERT(vp->v_type == VREG);
1436 	KASSERT(mutex_owned(&vp->v_interlock));
1437 
1438 	node = VP_TO_TMPFS_NODE(vp);
1439 	uobj = node->tn_spec.tn_reg.tn_aobj;
1440 
1441 	/* We currently don't rely on PGO_PASTEOF. */
1442 
1443 	if (vp->v_size <= offset + (centeridx << PAGE_SHIFT)) {
1444 		if ((flags & PGO_LOCKED) == 0)
1445 			mutex_exit(&vp->v_interlock);
1446 		return EINVAL;
1447 	}
1448 
1449 	if (vp->v_size < offset + (npages << PAGE_SHIFT)) {
1450 		npages = (round_page(vp->v_size) - offset) >> PAGE_SHIFT;
1451 	}
1452 
1453 	if ((flags & PGO_LOCKED) != 0)
1454 		return EBUSY;
1455 
1456 	if ((flags & PGO_NOTIMESTAMP) == 0) {
1457 		if ((vp->v_mount->mnt_flag & MNT_NOATIME) == 0)
1458 			node->tn_status |= TMPFS_NODE_ACCESSED;
1459 
1460 		if ((access_type & VM_PROT_WRITE) != 0)
1461 			node->tn_status |= TMPFS_NODE_MODIFIED;
1462 	}
1463 
1464 	mutex_exit(&vp->v_interlock);
1465 
1466 	/*
1467 	 * Make sure that the array on which we will store the
1468 	 * gotten pages is clean.  Otherwise uao_get (pointed to by
1469 	 * the pgo_get below) gets confused and does not return the
1470 	 * appropriate pages.
1471 	 *
1472 	 * XXX This shall be revisited when kern/32166 is addressed
1473 	 * because the loop to clean m[i] will most likely be redundant
1474 	 * as well as the PGO_ALLPAGES flag.
1475 	 */
1476 	if (m != NULL)
1477 		for (i = 0; i < npages; i++)
1478 			m[i] = NULL;
1479 	mutex_enter(&uobj->vmobjlock);
1480 	error = (*uobj->pgops->pgo_get)(uobj, offset, m, &npages, centeridx,
1481 	    access_type, advice, flags | PGO_ALLPAGES);
1482 #if defined(DEBUG)
1483 	{
1484 		/* Make sure that all the pages we return are valid. */
1485 		int dbgi;
1486 		if (error == 0 && m != NULL)
1487 			for (dbgi = 0; dbgi < npages; dbgi++)
1488 				KASSERT(m[dbgi] != NULL);
1489 	}
1490 #endif
1491 
1492 	return error;
1493 }
1494 
1495 /* --------------------------------------------------------------------- */
1496 
1497 int
1498 tmpfs_putpages(void *v)
1499 {
1500 	struct vnode *vp = ((struct vop_putpages_args *)v)->a_vp;
1501 	voff_t offlo = ((struct vop_putpages_args *)v)->a_offlo;
1502 	voff_t offhi = ((struct vop_putpages_args *)v)->a_offhi;
1503 	int flags = ((struct vop_putpages_args *)v)->a_flags;
1504 
1505 	int error;
1506 	struct tmpfs_node *node;
1507 	struct uvm_object *uobj;
1508 
1509 	KASSERT(mutex_owned(&vp->v_interlock));
1510 
1511 	node = VP_TO_TMPFS_NODE(vp);
1512 
1513 	if (vp->v_type != VREG) {
1514 		mutex_exit(&vp->v_interlock);
1515 		return 0;
1516 	}
1517 
1518 	uobj = node->tn_spec.tn_reg.tn_aobj;
1519 	mutex_exit(&vp->v_interlock);
1520 
1521 	mutex_enter(&uobj->vmobjlock);
1522 	error = (*uobj->pgops->pgo_put)(uobj, offlo, offhi, flags);
1523 
1524 	/* XXX mtime */
1525 
1526 	return error;
1527 }
1528