xref: /freebsd/sys/fs/tmpfs/tmpfs_vnops.c (revision 4ed925457ab06e83238a5db33e89ccc94b99a713)
1 /*	$NetBSD: tmpfs_vnops.c,v 1.39 2007/07/23 15:41:01 jmmv Exp $	*/
2 
3 /*-
4  * Copyright (c) 2005, 2006 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 #include <sys/cdefs.h>
37 __FBSDID("$FreeBSD$");
38 
39 #include <sys/param.h>
40 #include <sys/fcntl.h>
41 #include <sys/lockf.h>
42 #include <sys/namei.h>
43 #include <sys/priv.h>
44 #include <sys/proc.h>
45 #include <sys/resourcevar.h>
46 #include <sys/sched.h>
47 #include <sys/sf_buf.h>
48 #include <sys/stat.h>
49 #include <sys/systm.h>
50 #include <sys/unistd.h>
51 #include <sys/vnode.h>
52 
53 #include <vm/vm.h>
54 #include <vm/vm_object.h>
55 #include <vm/vm_page.h>
56 #include <vm/vm_pager.h>
57 
58 #include <machine/_inttypes.h>
59 
60 #include <fs/fifofs/fifo.h>
61 #include <fs/tmpfs/tmpfs_vnops.h>
62 #include <fs/tmpfs/tmpfs.h>
63 
64 /* --------------------------------------------------------------------- */
65 
66 static int
67 tmpfs_lookup(struct vop_cachedlookup_args *v)
68 {
69 	struct vnode *dvp = v->a_dvp;
70 	struct vnode **vpp = v->a_vpp;
71 	struct componentname *cnp = v->a_cnp;
72 
73 	int error;
74 	struct tmpfs_dirent *de;
75 	struct tmpfs_node *dnode;
76 
77 	dnode = VP_TO_TMPFS_DIR(dvp);
78 	*vpp = NULLVP;
79 
80 	/* Check accessibility of requested node as a first step. */
81 	error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, cnp->cn_thread);
82 	if (error != 0)
83 		goto out;
84 
85 	/* We cannot be requesting the parent directory of the root node. */
86 	MPASS(IMPLIES(dnode->tn_type == VDIR &&
87 	    dnode->tn_dir.tn_parent == dnode,
88 	    !(cnp->cn_flags & ISDOTDOT)));
89 
90 	TMPFS_ASSERT_LOCKED(dnode);
91 	if (dnode->tn_dir.tn_parent == NULL) {
92 		error = ENOENT;
93 		goto out;
94 	}
95 	if (cnp->cn_flags & ISDOTDOT) {
96 		int ltype = 0;
97 
98 		ltype = VOP_ISLOCKED(dvp);
99 		vhold(dvp);
100 		VOP_UNLOCK(dvp, 0);
101 		/* Allocate a new vnode on the matching entry. */
102 		error = tmpfs_alloc_vp(dvp->v_mount, dnode->tn_dir.tn_parent,
103 		    cnp->cn_lkflags, vpp);
104 
105 		vn_lock(dvp, ltype | LK_RETRY);
106 		vdrop(dvp);
107 	} else if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
108 		VREF(dvp);
109 		*vpp = dvp;
110 		error = 0;
111 	} else {
112 		de = tmpfs_dir_lookup(dnode, NULL, cnp);
113 		if (de == NULL) {
114 			/* The entry was not found in the directory.
115 			 * This is OK if we are creating or renaming an
116 			 * entry and are working on the last component of
117 			 * the path name. */
118 			if ((cnp->cn_flags & ISLASTCN) &&
119 			    (cnp->cn_nameiop == CREATE || \
120 			    cnp->cn_nameiop == RENAME)) {
121 				error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred,
122 				    cnp->cn_thread);
123 				if (error != 0)
124 					goto out;
125 
126 				/* Keep the component name in the buffer for
127 				 * future uses. */
128 				cnp->cn_flags |= SAVENAME;
129 
130 				error = EJUSTRETURN;
131 			} else
132 				error = ENOENT;
133 		} else {
134 			struct tmpfs_node *tnode;
135 
136 			/* The entry was found, so get its associated
137 			 * tmpfs_node. */
138 			tnode = de->td_node;
139 
140 			/* If we are not at the last path component and
141 			 * found a non-directory or non-link entry (which
142 			 * may itself be pointing to a directory), raise
143 			 * an error. */
144 			if ((tnode->tn_type != VDIR &&
145 			    tnode->tn_type != VLNK) &&
146 			    !(cnp->cn_flags & ISLASTCN)) {
147 				error = ENOTDIR;
148 				goto out;
149 			}
150 
151 			/* If we are deleting or renaming the entry, keep
152 			 * track of its tmpfs_dirent so that it can be
153 			 * easily deleted later. */
154 			if ((cnp->cn_flags & ISLASTCN) &&
155 			    (cnp->cn_nameiop == DELETE ||
156 			    cnp->cn_nameiop == RENAME)) {
157 				error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred,
158 				    cnp->cn_thread);
159 				if (error != 0)
160 					goto out;
161 
162 				/* Allocate a new vnode on the matching entry. */
163 				error = tmpfs_alloc_vp(dvp->v_mount, tnode,
164 						cnp->cn_lkflags, vpp);
165 				if (error != 0)
166 					goto out;
167 
168 				if ((dnode->tn_mode & S_ISTXT) &&
169 				  VOP_ACCESS(dvp, VADMIN, cnp->cn_cred, cnp->cn_thread) &&
170 				  VOP_ACCESS(*vpp, VADMIN, cnp->cn_cred, cnp->cn_thread)) {
171 					error = EPERM;
172 					vput(*vpp);
173 					*vpp = NULL;
174 					goto out;
175 				}
176 				cnp->cn_flags |= SAVENAME;
177 			} else {
178 				error = tmpfs_alloc_vp(dvp->v_mount, tnode,
179 						cnp->cn_lkflags, vpp);
180 			}
181 		}
182 	}
183 
184 	/* Store the result of this lookup in the cache.  Avoid this if the
185 	 * request was for creation, as it does not improve timings on
186 	 * emprical tests. */
187 	if ((cnp->cn_flags & MAKEENTRY) && cnp->cn_nameiop != CREATE)
188 		cache_enter(dvp, *vpp, cnp);
189 
190 out:
191 	/* If there were no errors, *vpp cannot be null and it must be
192 	 * locked. */
193 	MPASS(IFF(error == 0, *vpp != NULLVP && VOP_ISLOCKED(*vpp)));
194 
195 	return error;
196 }
197 
198 /* --------------------------------------------------------------------- */
199 
200 static int
201 tmpfs_create(struct vop_create_args *v)
202 {
203 	struct vnode *dvp = v->a_dvp;
204 	struct vnode **vpp = v->a_vpp;
205 	struct componentname *cnp = v->a_cnp;
206 	struct vattr *vap = v->a_vap;
207 
208 	MPASS(vap->va_type == VREG || vap->va_type == VSOCK);
209 
210 	return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
211 }
212 /* --------------------------------------------------------------------- */
213 
214 static int
215 tmpfs_mknod(struct vop_mknod_args *v)
216 {
217 	struct vnode *dvp = v->a_dvp;
218 	struct vnode **vpp = v->a_vpp;
219 	struct componentname *cnp = v->a_cnp;
220 	struct vattr *vap = v->a_vap;
221 
222 	if (vap->va_type != VBLK && vap->va_type != VCHR &&
223 	    vap->va_type != VFIFO)
224 		return EINVAL;
225 
226 	return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
227 }
228 
229 /* --------------------------------------------------------------------- */
230 
231 static int
232 tmpfs_open(struct vop_open_args *v)
233 {
234 	struct vnode *vp = v->a_vp;
235 	int mode = v->a_mode;
236 
237 	int error;
238 	struct tmpfs_node *node;
239 
240 	MPASS(VOP_ISLOCKED(vp));
241 
242 	node = VP_TO_TMPFS_NODE(vp);
243 
244 	/* The file is still active but all its names have been removed
245 	 * (e.g. by a "rmdir $(pwd)").  It cannot be opened any more as
246 	 * it is about to die. */
247 	if (node->tn_links < 1)
248 		return (ENOENT);
249 
250 	/* If the file is marked append-only, deny write requests. */
251 	if (node->tn_flags & APPEND && (mode & (FWRITE | O_APPEND)) == FWRITE)
252 		error = EPERM;
253 	else {
254 		error = 0;
255 		vnode_create_vobject(vp, node->tn_size, v->a_td);
256 	}
257 
258 	MPASS(VOP_ISLOCKED(vp));
259 	return error;
260 }
261 
262 /* --------------------------------------------------------------------- */
263 
264 static int
265 tmpfs_close(struct vop_close_args *v)
266 {
267 	struct vnode *vp = v->a_vp;
268 
269 	struct tmpfs_node *node;
270 
271 	MPASS(VOP_ISLOCKED(vp));
272 
273 	node = VP_TO_TMPFS_NODE(vp);
274 
275 	if (node->tn_links > 0) {
276 		/* Update node times.  No need to do it if the node has
277 		 * been deleted, because it will vanish after we return. */
278 		tmpfs_update(vp);
279 	}
280 
281 	return 0;
282 }
283 
284 /* --------------------------------------------------------------------- */
285 
286 int
287 tmpfs_access(struct vop_access_args *v)
288 {
289 	struct vnode *vp = v->a_vp;
290 	accmode_t accmode = v->a_accmode;
291 	struct ucred *cred = v->a_cred;
292 
293 	int error;
294 	struct tmpfs_node *node;
295 
296 	MPASS(VOP_ISLOCKED(vp));
297 
298 	node = VP_TO_TMPFS_NODE(vp);
299 
300 	switch (vp->v_type) {
301 	case VDIR:
302 		/* FALLTHROUGH */
303 	case VLNK:
304 		/* FALLTHROUGH */
305 	case VREG:
306 		if (accmode & VWRITE && vp->v_mount->mnt_flag & MNT_RDONLY) {
307 			error = EROFS;
308 			goto out;
309 		}
310 		break;
311 
312 	case VBLK:
313 		/* FALLTHROUGH */
314 	case VCHR:
315 		/* FALLTHROUGH */
316 	case VSOCK:
317 		/* FALLTHROUGH */
318 	case VFIFO:
319 		break;
320 
321 	default:
322 		error = EINVAL;
323 		goto out;
324 	}
325 
326 	if (accmode & VWRITE && node->tn_flags & IMMUTABLE) {
327 		error = EPERM;
328 		goto out;
329 	}
330 
331 	error = vaccess(vp->v_type, node->tn_mode, node->tn_uid,
332 	    node->tn_gid, accmode, cred, NULL);
333 
334 out:
335 	MPASS(VOP_ISLOCKED(vp));
336 
337 	return error;
338 }
339 
340 /* --------------------------------------------------------------------- */
341 
342 int
343 tmpfs_getattr(struct vop_getattr_args *v)
344 {
345 	struct vnode *vp = v->a_vp;
346 	struct vattr *vap = v->a_vap;
347 
348 	struct tmpfs_node *node;
349 
350 	node = VP_TO_TMPFS_NODE(vp);
351 
352 	tmpfs_update(vp);
353 
354 	vap->va_type = vp->v_type;
355 	vap->va_mode = node->tn_mode;
356 	vap->va_nlink = node->tn_links;
357 	vap->va_uid = node->tn_uid;
358 	vap->va_gid = node->tn_gid;
359 	vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
360 	vap->va_fileid = node->tn_id;
361 	vap->va_size = node->tn_size;
362 	vap->va_blocksize = PAGE_SIZE;
363 	vap->va_atime = node->tn_atime;
364 	vap->va_mtime = node->tn_mtime;
365 	vap->va_ctime = node->tn_ctime;
366 	vap->va_birthtime = node->tn_birthtime;
367 	vap->va_gen = node->tn_gen;
368 	vap->va_flags = node->tn_flags;
369 	vap->va_rdev = (vp->v_type == VBLK || vp->v_type == VCHR) ?
370 		node->tn_rdev : NODEV;
371 	vap->va_bytes = round_page(node->tn_size);
372 	vap->va_filerev = 0;
373 
374 	return 0;
375 }
376 
377 /* --------------------------------------------------------------------- */
378 
379 /* XXX Should this operation be atomic?  I think it should, but code in
380  * XXX other places (e.g., ufs) doesn't seem to be... */
381 int
382 tmpfs_setattr(struct vop_setattr_args *v)
383 {
384 	struct vnode *vp = v->a_vp;
385 	struct vattr *vap = v->a_vap;
386 	struct ucred *cred = v->a_cred;
387 	struct thread *td = curthread;
388 
389 	int error;
390 
391 	MPASS(VOP_ISLOCKED(vp));
392 
393 	error = 0;
394 
395 	/* Abort if any unsettable attribute is given. */
396 	if (vap->va_type != VNON ||
397 	    vap->va_nlink != VNOVAL ||
398 	    vap->va_fsid != VNOVAL ||
399 	    vap->va_fileid != VNOVAL ||
400 	    vap->va_blocksize != VNOVAL ||
401 	    vap->va_gen != VNOVAL ||
402 	    vap->va_rdev != VNOVAL ||
403 	    vap->va_bytes != VNOVAL)
404 		error = EINVAL;
405 
406 	if (error == 0 && (vap->va_flags != VNOVAL))
407 		error = tmpfs_chflags(vp, vap->va_flags, cred, td);
408 
409 	if (error == 0 && (vap->va_size != VNOVAL))
410 		error = tmpfs_chsize(vp, vap->va_size, cred, td);
411 
412 	if (error == 0 && (vap->va_uid != VNOVAL || vap->va_gid != VNOVAL))
413 		error = tmpfs_chown(vp, vap->va_uid, vap->va_gid, cred, td);
414 
415 	if (error == 0 && (vap->va_mode != (mode_t)VNOVAL))
416 		error = tmpfs_chmod(vp, vap->va_mode, cred, td);
417 
418 	if (error == 0 && ((vap->va_atime.tv_sec != VNOVAL &&
419 	    vap->va_atime.tv_nsec != VNOVAL) ||
420 	    (vap->va_mtime.tv_sec != VNOVAL &&
421 	    vap->va_mtime.tv_nsec != VNOVAL) ||
422 	    (vap->va_birthtime.tv_sec != VNOVAL &&
423 	    vap->va_birthtime.tv_nsec != VNOVAL)))
424 		error = tmpfs_chtimes(vp, &vap->va_atime, &vap->va_mtime,
425 			&vap->va_birthtime, vap->va_vaflags, cred, td);
426 
427 	/* Update the node times.  We give preference to the error codes
428 	 * generated by this function rather than the ones that may arise
429 	 * from tmpfs_update. */
430 	tmpfs_update(vp);
431 
432 	MPASS(VOP_ISLOCKED(vp));
433 
434 	return error;
435 }
436 
437 /* --------------------------------------------------------------------- */
438 static int
439 tmpfs_nocacheread(vm_object_t tobj, vm_pindex_t idx,
440     vm_offset_t offset, size_t tlen, struct uio *uio)
441 {
442 	vm_page_t	m;
443 	int		error;
444 
445 	VM_OBJECT_LOCK(tobj);
446 	vm_object_pip_add(tobj, 1);
447 	m = vm_page_grab(tobj, idx, VM_ALLOC_WIRED |
448 	    VM_ALLOC_ZERO | VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
449 	if (m->valid != VM_PAGE_BITS_ALL) {
450 		if (vm_pager_has_page(tobj, idx, NULL, NULL)) {
451 			error = vm_pager_get_pages(tobj, &m, 1, 0);
452 			if (error != 0) {
453 				printf("tmpfs get pages from pager error [read]\n");
454 				goto out;
455 			}
456 		} else
457 			vm_page_zero_invalid(m, TRUE);
458 	}
459 	VM_OBJECT_UNLOCK(tobj);
460 	error = uiomove_fromphys(&m, offset, tlen, uio);
461 	VM_OBJECT_LOCK(tobj);
462 out:
463 	vm_page_lock_queues();
464 	vm_page_unwire(m, TRUE);
465 	vm_page_unlock_queues();
466 	vm_page_wakeup(m);
467 	vm_object_pip_subtract(tobj, 1);
468 	VM_OBJECT_UNLOCK(tobj);
469 
470 	return (error);
471 }
472 
473 static __inline int
474 tmpfs_nocacheread_buf(vm_object_t tobj, vm_pindex_t idx,
475     vm_offset_t offset, size_t tlen, void *buf)
476 {
477 	struct uio uio;
478 	struct iovec iov;
479 
480 	uio.uio_iovcnt = 1;
481 	uio.uio_iov = &iov;
482 	iov.iov_base = buf;
483 	iov.iov_len = tlen;
484 
485 	uio.uio_offset = 0;
486 	uio.uio_resid = tlen;
487 	uio.uio_rw = UIO_READ;
488 	uio.uio_segflg = UIO_SYSSPACE;
489 	uio.uio_td = curthread;
490 
491 	return (tmpfs_nocacheread(tobj, idx, offset, tlen, &uio));
492 }
493 
494 static int
495 tmpfs_mappedread(vm_object_t vobj, vm_object_t tobj, size_t len, struct uio *uio)
496 {
497 	struct sf_buf	*sf;
498 	vm_pindex_t	idx;
499 	vm_page_t	m;
500 	vm_offset_t	offset;
501 	off_t		addr;
502 	size_t		tlen;
503 	char		*ma;
504 	int		error;
505 
506 	addr = uio->uio_offset;
507 	idx = OFF_TO_IDX(addr);
508 	offset = addr & PAGE_MASK;
509 	tlen = MIN(PAGE_SIZE - offset, len);
510 
511 	if ((vobj == NULL) ||
512 	    (vobj->resident_page_count == 0 && vobj->cache == NULL))
513 		goto nocache;
514 
515 	VM_OBJECT_LOCK(vobj);
516 lookupvpg:
517 	if (((m = vm_page_lookup(vobj, idx)) != NULL) &&
518 	    vm_page_is_valid(m, offset, tlen)) {
519 		if (vm_page_sleep_if_busy(m, FALSE, "tmfsmr"))
520 			goto lookupvpg;
521 		vm_page_busy(m);
522 		VM_OBJECT_UNLOCK(vobj);
523 		error = uiomove_fromphys(&m, offset, tlen, uio);
524 		VM_OBJECT_LOCK(vobj);
525 		vm_page_wakeup(m);
526 		VM_OBJECT_UNLOCK(vobj);
527 		return	(error);
528 	} else if (m != NULL && uio->uio_segflg == UIO_NOCOPY) {
529 		if (vm_page_sleep_if_busy(m, FALSE, "tmfsmr"))
530 			goto lookupvpg;
531 		vm_page_busy(m);
532 		VM_OBJECT_UNLOCK(vobj);
533 		sched_pin();
534 		sf = sf_buf_alloc(m, SFB_CPUPRIVATE);
535 		ma = (char *)sf_buf_kva(sf);
536 		error = tmpfs_nocacheread_buf(tobj, idx, offset, tlen,
537 		    ma + offset);
538 		if (error == 0) {
539 			uio->uio_offset += tlen;
540 			uio->uio_resid -= tlen;
541 		}
542 		sf_buf_free(sf);
543 		sched_unpin();
544 		VM_OBJECT_LOCK(vobj);
545 		vm_page_wakeup(m);
546 		VM_OBJECT_UNLOCK(vobj);
547 		return	(error);
548 	}
549 	VM_OBJECT_UNLOCK(vobj);
550 nocache:
551 	error = tmpfs_nocacheread(tobj, idx, offset, tlen, uio);
552 
553 	return	(error);
554 }
555 
556 static int
557 tmpfs_read(struct vop_read_args *v)
558 {
559 	struct vnode *vp = v->a_vp;
560 	struct uio *uio = v->a_uio;
561 
562 	struct tmpfs_node *node;
563 	vm_object_t uobj;
564 	size_t len;
565 	int resid;
566 
567 	int error = 0;
568 
569 	node = VP_TO_TMPFS_NODE(vp);
570 
571 	if (vp->v_type != VREG) {
572 		error = EISDIR;
573 		goto out;
574 	}
575 
576 	if (uio->uio_offset < 0) {
577 		error = EINVAL;
578 		goto out;
579 	}
580 
581 	node->tn_status |= TMPFS_NODE_ACCESSED;
582 
583 	uobj = node->tn_reg.tn_aobj;
584 	while ((resid = uio->uio_resid) > 0) {
585 		error = 0;
586 		if (node->tn_size <= uio->uio_offset)
587 			break;
588 		len = MIN(node->tn_size - uio->uio_offset, resid);
589 		if (len == 0)
590 			break;
591 		error = tmpfs_mappedread(vp->v_object, uobj, len, uio);
592 		if ((error != 0) || (resid == uio->uio_resid))
593 			break;
594 	}
595 
596 out:
597 
598 	return error;
599 }
600 
601 /* --------------------------------------------------------------------- */
602 
603 static int
604 tmpfs_mappedwrite(vm_object_t vobj, vm_object_t tobj, size_t len, struct uio *uio)
605 {
606 	vm_pindex_t	idx;
607 	vm_page_t	vpg, tpg;
608 	vm_offset_t	offset;
609 	off_t		addr;
610 	size_t		tlen;
611 	int		error;
612 
613 	error = 0;
614 
615 	addr = uio->uio_offset;
616 	idx = OFF_TO_IDX(addr);
617 	offset = addr & PAGE_MASK;
618 	tlen = MIN(PAGE_SIZE - offset, len);
619 
620 	if ((vobj == NULL) ||
621 	    (vobj->resident_page_count == 0 && vobj->cache == NULL)) {
622 		vpg = NULL;
623 		goto nocache;
624 	}
625 
626 	VM_OBJECT_LOCK(vobj);
627 lookupvpg:
628 	if (((vpg = vm_page_lookup(vobj, idx)) != NULL) &&
629 	    vm_page_is_valid(vpg, offset, tlen)) {
630 		if (vm_page_sleep_if_busy(vpg, FALSE, "tmfsmw"))
631 			goto lookupvpg;
632 		vm_page_busy(vpg);
633 		vm_page_lock_queues();
634 		vm_page_undirty(vpg);
635 		vm_page_unlock_queues();
636 		VM_OBJECT_UNLOCK(vobj);
637 		error = uiomove_fromphys(&vpg, offset, tlen, uio);
638 	} else {
639 		if (__predict_false(vobj->cache != NULL))
640 			vm_page_cache_free(vobj, idx, idx + 1);
641 		VM_OBJECT_UNLOCK(vobj);
642 		vpg = NULL;
643 	}
644 nocache:
645 	VM_OBJECT_LOCK(tobj);
646 	vm_object_pip_add(tobj, 1);
647 	tpg = vm_page_grab(tobj, idx, VM_ALLOC_WIRED |
648 	    VM_ALLOC_ZERO | VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
649 	if (tpg->valid != VM_PAGE_BITS_ALL) {
650 		if (vm_pager_has_page(tobj, idx, NULL, NULL)) {
651 			error = vm_pager_get_pages(tobj, &tpg, 1, 0);
652 			if (error != 0) {
653 				printf("tmpfs get pages from pager error [write]\n");
654 				goto out;
655 			}
656 		} else
657 			vm_page_zero_invalid(tpg, TRUE);
658 	}
659 	VM_OBJECT_UNLOCK(tobj);
660 	if (vpg == NULL)
661 		error = uiomove_fromphys(&tpg, offset, tlen, uio);
662 	else {
663 		KASSERT(vpg->valid == VM_PAGE_BITS_ALL, ("parts of vpg invalid"));
664 		pmap_copy_page(vpg, tpg);
665 	}
666 	VM_OBJECT_LOCK(tobj);
667 out:
668 	if (vobj != NULL)
669 		VM_OBJECT_LOCK(vobj);
670 	vm_page_lock_queues();
671 	if (error == 0) {
672 		KASSERT(tpg->valid == VM_PAGE_BITS_ALL,
673 		    ("parts of tpg invalid"));
674 		vm_page_dirty(tpg);
675 	}
676 	vm_page_unwire(tpg, TRUE);
677 	vm_page_unlock_queues();
678 	vm_page_wakeup(tpg);
679 	if (vpg != NULL)
680 		vm_page_wakeup(vpg);
681 	if (vobj != NULL)
682 		VM_OBJECT_UNLOCK(vobj);
683 	vm_object_pip_subtract(tobj, 1);
684 	VM_OBJECT_UNLOCK(tobj);
685 
686 	return	(error);
687 }
688 
689 static int
690 tmpfs_write(struct vop_write_args *v)
691 {
692 	struct vnode *vp = v->a_vp;
693 	struct uio *uio = v->a_uio;
694 	int ioflag = v->a_ioflag;
695 	struct thread *td = uio->uio_td;
696 
697 	boolean_t extended;
698 	int error = 0;
699 	off_t oldsize;
700 	struct tmpfs_node *node;
701 	vm_object_t uobj;
702 	size_t len;
703 	int resid;
704 
705 	node = VP_TO_TMPFS_NODE(vp);
706 	oldsize = node->tn_size;
707 
708 	if (uio->uio_offset < 0 || vp->v_type != VREG) {
709 		error = EINVAL;
710 		goto out;
711 	}
712 
713 	if (uio->uio_resid == 0) {
714 		error = 0;
715 		goto out;
716 	}
717 
718 	if (ioflag & IO_APPEND)
719 		uio->uio_offset = node->tn_size;
720 
721 	if (uio->uio_offset + uio->uio_resid >
722 	  VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize)
723 		return (EFBIG);
724 
725 	if (vp->v_type == VREG && td != NULL) {
726 		PROC_LOCK(td->td_proc);
727 		if (uio->uio_offset + uio->uio_resid >
728 		  lim_cur(td->td_proc, RLIMIT_FSIZE)) {
729 			psignal(td->td_proc, SIGXFSZ);
730 			PROC_UNLOCK(td->td_proc);
731 			return (EFBIG);
732 		}
733 		PROC_UNLOCK(td->td_proc);
734 	}
735 
736 	extended = uio->uio_offset + uio->uio_resid > node->tn_size;
737 	if (extended) {
738 		error = tmpfs_reg_resize(vp, uio->uio_offset + uio->uio_resid);
739 		if (error != 0)
740 			goto out;
741 	}
742 
743 	uobj = node->tn_reg.tn_aobj;
744 	while ((resid = uio->uio_resid) > 0) {
745 		if (node->tn_size <= uio->uio_offset)
746 			break;
747 		len = MIN(node->tn_size - uio->uio_offset, resid);
748 		if (len == 0)
749 			break;
750 		error = tmpfs_mappedwrite(vp->v_object, uobj, len, uio);
751 		if ((error != 0) || (resid == uio->uio_resid))
752 			break;
753 	}
754 
755 	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
756 	    (extended ? TMPFS_NODE_CHANGED : 0);
757 
758 	if (node->tn_mode & (S_ISUID | S_ISGID)) {
759 		if (priv_check_cred(v->a_cred, PRIV_VFS_RETAINSUGID, 0))
760 			node->tn_mode &= ~(S_ISUID | S_ISGID);
761 	}
762 
763 	if (error != 0)
764 		(void)tmpfs_reg_resize(vp, oldsize);
765 
766 out:
767 	MPASS(IMPLIES(error == 0, uio->uio_resid == 0));
768 	MPASS(IMPLIES(error != 0, oldsize == node->tn_size));
769 
770 	return error;
771 }
772 
773 /* --------------------------------------------------------------------- */
774 
775 static int
776 tmpfs_fsync(struct vop_fsync_args *v)
777 {
778 	struct vnode *vp = v->a_vp;
779 
780 	MPASS(VOP_ISLOCKED(vp));
781 
782 	tmpfs_update(vp);
783 
784 	return 0;
785 }
786 
787 /* --------------------------------------------------------------------- */
788 
789 static int
790 tmpfs_remove(struct vop_remove_args *v)
791 {
792 	struct vnode *dvp = v->a_dvp;
793 	struct vnode *vp = v->a_vp;
794 
795 	int error;
796 	struct tmpfs_dirent *de;
797 	struct tmpfs_mount *tmp;
798 	struct tmpfs_node *dnode;
799 	struct tmpfs_node *node;
800 
801 	MPASS(VOP_ISLOCKED(dvp));
802 	MPASS(VOP_ISLOCKED(vp));
803 
804 	if (vp->v_type == VDIR) {
805 		error = EISDIR;
806 		goto out;
807 	}
808 
809 	dnode = VP_TO_TMPFS_DIR(dvp);
810 	node = VP_TO_TMPFS_NODE(vp);
811 	tmp = VFS_TO_TMPFS(vp->v_mount);
812 	de = tmpfs_dir_lookup(dnode, node, v->a_cnp);
813 	MPASS(de != NULL);
814 
815 	/* Files marked as immutable or append-only cannot be deleted. */
816 	if ((node->tn_flags & (IMMUTABLE | APPEND | NOUNLINK)) ||
817 	    (dnode->tn_flags & APPEND)) {
818 		error = EPERM;
819 		goto out;
820 	}
821 
822 	/* Remove the entry from the directory; as it is a file, we do not
823 	 * have to change the number of hard links of the directory. */
824 	tmpfs_dir_detach(dvp, de);
825 
826 	/* Free the directory entry we just deleted.  Note that the node
827 	 * referred by it will not be removed until the vnode is really
828 	 * reclaimed. */
829 	tmpfs_free_dirent(tmp, de, TRUE);
830 
831 	if (node->tn_links > 0)
832 		node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
833 	    TMPFS_NODE_MODIFIED;
834 	error = 0;
835 
836 out:
837 
838 	return error;
839 }
840 
841 /* --------------------------------------------------------------------- */
842 
843 static int
844 tmpfs_link(struct vop_link_args *v)
845 {
846 	struct vnode *dvp = v->a_tdvp;
847 	struct vnode *vp = v->a_vp;
848 	struct componentname *cnp = v->a_cnp;
849 
850 	int error;
851 	struct tmpfs_dirent *de;
852 	struct tmpfs_node *node;
853 
854 	MPASS(VOP_ISLOCKED(dvp));
855 	MPASS(cnp->cn_flags & HASBUF);
856 	MPASS(dvp != vp); /* XXX When can this be false? */
857 
858 	node = VP_TO_TMPFS_NODE(vp);
859 
860 	/* XXX: Why aren't the following two tests done by the caller? */
861 
862 	/* Hard links of directories are forbidden. */
863 	if (vp->v_type == VDIR) {
864 		error = EPERM;
865 		goto out;
866 	}
867 
868 	/* Cannot create cross-device links. */
869 	if (dvp->v_mount != vp->v_mount) {
870 		error = EXDEV;
871 		goto out;
872 	}
873 
874 	/* Ensure that we do not overflow the maximum number of links imposed
875 	 * by the system. */
876 	MPASS(node->tn_links <= LINK_MAX);
877 	if (node->tn_links == LINK_MAX) {
878 		error = EMLINK;
879 		goto out;
880 	}
881 
882 	/* We cannot create links of files marked immutable or append-only. */
883 	if (node->tn_flags & (IMMUTABLE | APPEND)) {
884 		error = EPERM;
885 		goto out;
886 	}
887 
888 	/* Allocate a new directory entry to represent the node. */
889 	error = tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), node,
890 	    cnp->cn_nameptr, cnp->cn_namelen, &de);
891 	if (error != 0)
892 		goto out;
893 
894 	/* Insert the new directory entry into the appropriate directory. */
895 	tmpfs_dir_attach(dvp, de);
896 
897 	/* vp link count has changed, so update node times. */
898 	node->tn_status |= TMPFS_NODE_CHANGED;
899 	tmpfs_update(vp);
900 
901 	error = 0;
902 
903 out:
904 	return error;
905 }
906 
907 /* --------------------------------------------------------------------- */
908 
909 static int
910 tmpfs_rename(struct vop_rename_args *v)
911 {
912 	struct vnode *fdvp = v->a_fdvp;
913 	struct vnode *fvp = v->a_fvp;
914 	struct componentname *fcnp = v->a_fcnp;
915 	struct vnode *tdvp = v->a_tdvp;
916 	struct vnode *tvp = v->a_tvp;
917 	struct componentname *tcnp = v->a_tcnp;
918 
919 	char *newname;
920 	int error;
921 	struct tmpfs_dirent *de;
922 	struct tmpfs_mount *tmp;
923 	struct tmpfs_node *fdnode;
924 	struct tmpfs_node *fnode;
925 	struct tmpfs_node *tnode;
926 	struct tmpfs_node *tdnode;
927 
928 	MPASS(VOP_ISLOCKED(tdvp));
929 	MPASS(IMPLIES(tvp != NULL, VOP_ISLOCKED(tvp)));
930 	MPASS(fcnp->cn_flags & HASBUF);
931 	MPASS(tcnp->cn_flags & HASBUF);
932 
933   	tnode = (tvp == NULL) ? NULL : VP_TO_TMPFS_NODE(tvp);
934 
935 	/* Disallow cross-device renames.
936 	 * XXX Why isn't this done by the caller? */
937 	if (fvp->v_mount != tdvp->v_mount ||
938 	    (tvp != NULL && fvp->v_mount != tvp->v_mount)) {
939 		error = EXDEV;
940 		goto out;
941 	}
942 
943 	tmp = VFS_TO_TMPFS(tdvp->v_mount);
944 	tdnode = VP_TO_TMPFS_DIR(tdvp);
945 
946 	/* If source and target are the same file, there is nothing to do. */
947 	if (fvp == tvp) {
948 		error = 0;
949 		goto out;
950 	}
951 
952 	/* If we need to move the directory between entries, lock the
953 	 * source so that we can safely operate on it. */
954 	if (tdvp != fdvp) {
955 		error = vn_lock(fdvp, LK_EXCLUSIVE | LK_RETRY);
956 		if (error != 0)
957 			goto out;
958 	}
959 	fdnode = VP_TO_TMPFS_DIR(fdvp);
960 	fnode = VP_TO_TMPFS_NODE(fvp);
961 	de = tmpfs_dir_lookup(fdnode, fnode, fcnp);
962 
963 	/* Avoid manipulating '.' and '..' entries. */
964 	if (de == NULL) {
965 		MPASS(fvp->v_type == VDIR);
966 		error = EINVAL;
967 		goto out_locked;
968 	}
969 	MPASS(de->td_node == fnode);
970 
971 	/* If re-naming a directory to another preexisting directory
972 	 * ensure that the target directory is empty so that its
973 	 * removal causes no side effects.
974 	 * Kern_rename gurantees the destination to be a directory
975 	 * if the source is one. */
976 	if (tvp != NULL) {
977 		MPASS(tnode != NULL);
978 
979 		if ((tnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
980 		    (tdnode->tn_flags & (APPEND | IMMUTABLE))) {
981 			error = EPERM;
982 			goto out_locked;
983 		}
984 
985 		if (fnode->tn_type == VDIR && tnode->tn_type == VDIR) {
986 			if (tnode->tn_size > 0) {
987 				error = ENOTEMPTY;
988 				goto out_locked;
989 			}
990 		} else if (fnode->tn_type == VDIR && tnode->tn_type != VDIR) {
991 			error = ENOTDIR;
992 			goto out_locked;
993 		} else if (fnode->tn_type != VDIR && tnode->tn_type == VDIR) {
994 			error = EISDIR;
995 			goto out_locked;
996 		} else {
997 			MPASS(fnode->tn_type != VDIR &&
998 				tnode->tn_type != VDIR);
999 		}
1000 	}
1001 
1002 	if ((fnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND))
1003 	    || (fdnode->tn_flags & (APPEND | IMMUTABLE))) {
1004 		error = EPERM;
1005 		goto out_locked;
1006 	}
1007 
1008 	/* Ensure that we have enough memory to hold the new name, if it
1009 	 * has to be changed. */
1010 	if (fcnp->cn_namelen != tcnp->cn_namelen ||
1011 	    bcmp(fcnp->cn_nameptr, tcnp->cn_nameptr, fcnp->cn_namelen) != 0) {
1012 		newname = malloc(tcnp->cn_namelen, M_TMPFSNAME, M_WAITOK);
1013 	} else
1014 		newname = NULL;
1015 
1016 	/* If the node is being moved to another directory, we have to do
1017 	 * the move. */
1018 	if (fdnode != tdnode) {
1019 		/* In case we are moving a directory, we have to adjust its
1020 		 * parent to point to the new parent. */
1021 		if (de->td_node->tn_type == VDIR) {
1022 			struct tmpfs_node *n;
1023 
1024 			/* Ensure the target directory is not a child of the
1025 			 * directory being moved.  Otherwise, we'd end up
1026 			 * with stale nodes. */
1027 			n = tdnode;
1028 			/* TMPFS_LOCK garanties that no nodes are freed while
1029 			 * traversing the list. Nodes can only be marked as
1030 			 * removed: tn_parent == NULL. */
1031 			TMPFS_LOCK(tmp);
1032 			TMPFS_NODE_LOCK(n);
1033 			while (n != n->tn_dir.tn_parent) {
1034 				struct tmpfs_node *parent;
1035 
1036 				if (n == fnode) {
1037 					TMPFS_NODE_UNLOCK(n);
1038 					TMPFS_UNLOCK(tmp);
1039 					error = EINVAL;
1040 					if (newname != NULL)
1041 						    free(newname, M_TMPFSNAME);
1042 					goto out_locked;
1043 				}
1044 				parent = n->tn_dir.tn_parent;
1045 				TMPFS_NODE_UNLOCK(n);
1046 				if (parent == NULL) {
1047 					n = NULL;
1048 					break;
1049 				}
1050 				TMPFS_NODE_LOCK(parent);
1051 				if (parent->tn_dir.tn_parent == NULL) {
1052 					TMPFS_NODE_UNLOCK(parent);
1053 					n = NULL;
1054 					break;
1055 				}
1056 				n = parent;
1057 			}
1058 			TMPFS_UNLOCK(tmp);
1059 			if (n == NULL) {
1060 				error = EINVAL;
1061 				if (newname != NULL)
1062 					    free(newname, M_TMPFSNAME);
1063 				goto out_locked;
1064 			}
1065 			TMPFS_NODE_UNLOCK(n);
1066 
1067 			/* Adjust the parent pointer. */
1068 			TMPFS_VALIDATE_DIR(fnode);
1069 			TMPFS_NODE_LOCK(de->td_node);
1070 			de->td_node->tn_dir.tn_parent = tdnode;
1071 			TMPFS_NODE_UNLOCK(de->td_node);
1072 
1073 			/* As a result of changing the target of the '..'
1074 			 * entry, the link count of the source and target
1075 			 * directories has to be adjusted. */
1076 			TMPFS_NODE_LOCK(tdnode);
1077 			TMPFS_ASSERT_LOCKED(tdnode);
1078 			tdnode->tn_links++;
1079 			TMPFS_NODE_UNLOCK(tdnode);
1080 
1081 			TMPFS_NODE_LOCK(fdnode);
1082 			TMPFS_ASSERT_LOCKED(fdnode);
1083 			fdnode->tn_links--;
1084 			TMPFS_NODE_UNLOCK(fdnode);
1085 		}
1086 
1087 		/* Do the move: just remove the entry from the source directory
1088 		 * and insert it into the target one. */
1089 		tmpfs_dir_detach(fdvp, de);
1090 		tmpfs_dir_attach(tdvp, de);
1091 	}
1092 
1093 	/* If the name has changed, we need to make it effective by changing
1094 	 * it in the directory entry. */
1095 	if (newname != NULL) {
1096 		MPASS(tcnp->cn_namelen <= MAXNAMLEN);
1097 
1098 		free(de->td_name, M_TMPFSNAME);
1099 		de->td_namelen = (uint16_t)tcnp->cn_namelen;
1100 		memcpy(newname, tcnp->cn_nameptr, tcnp->cn_namelen);
1101 		de->td_name = newname;
1102 
1103 		fnode->tn_status |= TMPFS_NODE_CHANGED;
1104 		tdnode->tn_status |= TMPFS_NODE_MODIFIED;
1105 	}
1106 
1107 	/* If we are overwriting an entry, we have to remove the old one
1108 	 * from the target directory. */
1109 	if (tvp != NULL) {
1110 		/* Remove the old entry from the target directory. */
1111 		de = tmpfs_dir_lookup(tdnode, tnode, tcnp);
1112 		tmpfs_dir_detach(tdvp, de);
1113 
1114 		/* Free the directory entry we just deleted.  Note that the
1115 		 * node referred by it will not be removed until the vnode is
1116 		 * really reclaimed. */
1117 		tmpfs_free_dirent(VFS_TO_TMPFS(tvp->v_mount), de, TRUE);
1118 	}
1119 
1120 	error = 0;
1121 
1122 out_locked:
1123 	if (fdnode != tdnode)
1124 		VOP_UNLOCK(fdvp, 0);
1125 
1126 out:
1127 	/* Release target nodes. */
1128 	/* XXX: I don't understand when tdvp can be the same as tvp, but
1129 	 * other code takes care of this... */
1130 	if (tdvp == tvp)
1131 		vrele(tdvp);
1132 	else
1133 		vput(tdvp);
1134 	if (tvp != NULL)
1135 		vput(tvp);
1136 
1137 	/* Release source nodes. */
1138 	vrele(fdvp);
1139 	vrele(fvp);
1140 
1141 	return error;
1142 }
1143 
1144 /* --------------------------------------------------------------------- */
1145 
1146 static int
1147 tmpfs_mkdir(struct vop_mkdir_args *v)
1148 {
1149 	struct vnode *dvp = v->a_dvp;
1150 	struct vnode **vpp = v->a_vpp;
1151 	struct componentname *cnp = v->a_cnp;
1152 	struct vattr *vap = v->a_vap;
1153 
1154 	MPASS(vap->va_type == VDIR);
1155 
1156 	return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
1157 }
1158 
1159 /* --------------------------------------------------------------------- */
1160 
1161 static int
1162 tmpfs_rmdir(struct vop_rmdir_args *v)
1163 {
1164 	struct vnode *dvp = v->a_dvp;
1165 	struct vnode *vp = v->a_vp;
1166 
1167 	int error;
1168 	struct tmpfs_dirent *de;
1169 	struct tmpfs_mount *tmp;
1170 	struct tmpfs_node *dnode;
1171 	struct tmpfs_node *node;
1172 
1173 	MPASS(VOP_ISLOCKED(dvp));
1174 	MPASS(VOP_ISLOCKED(vp));
1175 
1176 	tmp = VFS_TO_TMPFS(dvp->v_mount);
1177 	dnode = VP_TO_TMPFS_DIR(dvp);
1178 	node = VP_TO_TMPFS_DIR(vp);
1179 
1180 	/* Directories with more than two entries ('.' and '..') cannot be
1181 	 * removed. */
1182 	 if (node->tn_size > 0) {
1183 		 error = ENOTEMPTY;
1184 		 goto out;
1185 	 }
1186 
1187 	if ((dnode->tn_flags & APPEND)
1188 	    || (node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1189 		error = EPERM;
1190 		goto out;
1191 	}
1192 
1193 	/* This invariant holds only if we are not trying to remove "..".
1194 	  * We checked for that above so this is safe now. */
1195 	MPASS(node->tn_dir.tn_parent == dnode);
1196 
1197 	/* Get the directory entry associated with node (vp).  This was
1198 	 * filled by tmpfs_lookup while looking up the entry. */
1199 	de = tmpfs_dir_lookup(dnode, node, v->a_cnp);
1200 	MPASS(TMPFS_DIRENT_MATCHES(de,
1201 	    v->a_cnp->cn_nameptr,
1202 	    v->a_cnp->cn_namelen));
1203 
1204 	/* Check flags to see if we are allowed to remove the directory. */
1205 	if (dnode->tn_flags & APPEND
1206 		|| node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) {
1207 		error = EPERM;
1208 		goto out;
1209 	}
1210 
1211 
1212 	/* Detach the directory entry from the directory (dnode). */
1213 	tmpfs_dir_detach(dvp, de);
1214 
1215 	/* No vnode should be allocated for this entry from this point */
1216 	TMPFS_NODE_LOCK(node);
1217 	TMPFS_ASSERT_ELOCKED(node);
1218 	node->tn_links--;
1219 	node->tn_dir.tn_parent = NULL;
1220 	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
1221 	    TMPFS_NODE_MODIFIED;
1222 
1223 	TMPFS_NODE_UNLOCK(node);
1224 
1225 	TMPFS_NODE_LOCK(dnode);
1226 	TMPFS_ASSERT_ELOCKED(dnode);
1227 	dnode->tn_links--;
1228 	dnode->tn_status |= TMPFS_NODE_ACCESSED | \
1229 	    TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1230 	TMPFS_NODE_UNLOCK(dnode);
1231 
1232 	cache_purge(dvp);
1233 	cache_purge(vp);
1234 
1235 	/* Free the directory entry we just deleted.  Note that the node
1236 	 * referred by it will not be removed until the vnode is really
1237 	 * reclaimed. */
1238 	tmpfs_free_dirent(tmp, de, TRUE);
1239 
1240 	/* Release the deleted vnode (will destroy the node, notify
1241 	 * interested parties and clean it from the cache). */
1242 
1243 	dnode->tn_status |= TMPFS_NODE_CHANGED;
1244 	tmpfs_update(dvp);
1245 
1246 	error = 0;
1247 
1248 out:
1249 	return error;
1250 }
1251 
1252 /* --------------------------------------------------------------------- */
1253 
1254 static int
1255 tmpfs_symlink(struct vop_symlink_args *v)
1256 {
1257 	struct vnode *dvp = v->a_dvp;
1258 	struct vnode **vpp = v->a_vpp;
1259 	struct componentname *cnp = v->a_cnp;
1260 	struct vattr *vap = v->a_vap;
1261 	char *target = v->a_target;
1262 
1263 #ifdef notyet /* XXX FreeBSD BUG: kern_symlink is not setting VLNK */
1264 	MPASS(vap->va_type == VLNK);
1265 #else
1266 	vap->va_type = VLNK;
1267 #endif
1268 
1269 	return tmpfs_alloc_file(dvp, vpp, vap, cnp, target);
1270 }
1271 
1272 /* --------------------------------------------------------------------- */
1273 
1274 static int
1275 tmpfs_readdir(struct vop_readdir_args *v)
1276 {
1277 	struct vnode *vp = v->a_vp;
1278 	struct uio *uio = v->a_uio;
1279 	int *eofflag = v->a_eofflag;
1280 	u_long **cookies = v->a_cookies;
1281 	int *ncookies = v->a_ncookies;
1282 
1283 	int error;
1284 	off_t startoff;
1285 	off_t cnt = 0;
1286 	struct tmpfs_node *node;
1287 
1288 	/* This operation only makes sense on directory nodes. */
1289 	if (vp->v_type != VDIR)
1290 		return ENOTDIR;
1291 
1292 	node = VP_TO_TMPFS_DIR(vp);
1293 
1294 	startoff = uio->uio_offset;
1295 
1296 	if (uio->uio_offset == TMPFS_DIRCOOKIE_DOT) {
1297 		error = tmpfs_dir_getdotdent(node, uio);
1298 		if (error != 0)
1299 			goto outok;
1300 		cnt++;
1301 	}
1302 
1303 	if (uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT) {
1304 		error = tmpfs_dir_getdotdotdent(node, uio);
1305 		if (error != 0)
1306 			goto outok;
1307 		cnt++;
1308 	}
1309 
1310 	error = tmpfs_dir_getdents(node, uio, &cnt);
1311 
1312 outok:
1313 	MPASS(error >= -1);
1314 
1315 	if (error == -1)
1316 		error = 0;
1317 
1318 	if (eofflag != NULL)
1319 		*eofflag =
1320 		    (error == 0 && uio->uio_offset == TMPFS_DIRCOOKIE_EOF);
1321 
1322 	/* Update NFS-related variables. */
1323 	if (error == 0 && cookies != NULL && ncookies != NULL) {
1324 		off_t i;
1325 		off_t off = startoff;
1326 		struct tmpfs_dirent *de = NULL;
1327 
1328 		*ncookies = cnt;
1329 		*cookies = malloc(cnt * sizeof(off_t), M_TEMP, M_WAITOK);
1330 
1331 		for (i = 0; i < cnt; i++) {
1332 			MPASS(off != TMPFS_DIRCOOKIE_EOF);
1333 			if (off == TMPFS_DIRCOOKIE_DOT) {
1334 				off = TMPFS_DIRCOOKIE_DOTDOT;
1335 			} else {
1336 				if (off == TMPFS_DIRCOOKIE_DOTDOT) {
1337 					de = TAILQ_FIRST(&node->tn_dir.tn_dirhead);
1338 				} else if (de != NULL) {
1339 					de = TAILQ_NEXT(de, td_entries);
1340 				} else {
1341 					de = tmpfs_dir_lookupbycookie(node,
1342 					    off);
1343 					MPASS(de != NULL);
1344 					de = TAILQ_NEXT(de, td_entries);
1345 				}
1346 				if (de == NULL)
1347 					off = TMPFS_DIRCOOKIE_EOF;
1348 				else
1349 					off = tmpfs_dircookie(de);
1350 			}
1351 
1352 			(*cookies)[i] = off;
1353 		}
1354 		MPASS(uio->uio_offset == off);
1355 	}
1356 
1357 	return error;
1358 }
1359 
1360 /* --------------------------------------------------------------------- */
1361 
1362 static int
1363 tmpfs_readlink(struct vop_readlink_args *v)
1364 {
1365 	struct vnode *vp = v->a_vp;
1366 	struct uio *uio = v->a_uio;
1367 
1368 	int error;
1369 	struct tmpfs_node *node;
1370 
1371 	MPASS(uio->uio_offset == 0);
1372 	MPASS(vp->v_type == VLNK);
1373 
1374 	node = VP_TO_TMPFS_NODE(vp);
1375 
1376 	error = uiomove(node->tn_link, MIN(node->tn_size, uio->uio_resid),
1377 	    uio);
1378 	node->tn_status |= TMPFS_NODE_ACCESSED;
1379 
1380 	return error;
1381 }
1382 
1383 /* --------------------------------------------------------------------- */
1384 
1385 static int
1386 tmpfs_inactive(struct vop_inactive_args *v)
1387 {
1388 	struct vnode *vp = v->a_vp;
1389 	struct thread *l = v->a_td;
1390 
1391 	struct tmpfs_node *node;
1392 
1393 	MPASS(VOP_ISLOCKED(vp));
1394 
1395 	node = VP_TO_TMPFS_NODE(vp);
1396 
1397 	if (node->tn_links == 0)
1398 		vrecycle(vp, l);
1399 
1400 	return 0;
1401 }
1402 
1403 /* --------------------------------------------------------------------- */
1404 
1405 int
1406 tmpfs_reclaim(struct vop_reclaim_args *v)
1407 {
1408 	struct vnode *vp = v->a_vp;
1409 
1410 	struct tmpfs_mount *tmp;
1411 	struct tmpfs_node *node;
1412 
1413 	node = VP_TO_TMPFS_NODE(vp);
1414 	tmp = VFS_TO_TMPFS(vp->v_mount);
1415 
1416 	vnode_destroy_vobject(vp);
1417 	cache_purge(vp);
1418 
1419 	TMPFS_NODE_LOCK(node);
1420 	TMPFS_ASSERT_ELOCKED(node);
1421 	tmpfs_free_vp(vp);
1422 
1423 	/* If the node referenced by this vnode was deleted by the user,
1424 	 * we must free its associated data structures (now that the vnode
1425 	 * is being reclaimed). */
1426 	if (node->tn_links == 0 &&
1427 	    (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0) {
1428 		node->tn_vpstate = TMPFS_VNODE_DOOMED;
1429 		TMPFS_NODE_UNLOCK(node);
1430 		tmpfs_free_node(tmp, node);
1431 	} else
1432 		TMPFS_NODE_UNLOCK(node);
1433 
1434 	MPASS(vp->v_data == NULL);
1435 	return 0;
1436 }
1437 
1438 /* --------------------------------------------------------------------- */
1439 
1440 static int
1441 tmpfs_print(struct vop_print_args *v)
1442 {
1443 	struct vnode *vp = v->a_vp;
1444 
1445 	struct tmpfs_node *node;
1446 
1447 	node = VP_TO_TMPFS_NODE(vp);
1448 
1449 	printf("tag VT_TMPFS, tmpfs_node %p, flags 0x%x, links %d\n",
1450 	    node, node->tn_flags, node->tn_links);
1451 	printf("\tmode 0%o, owner %d, group %d, size %" PRIdMAX
1452 	    ", status 0x%x\n",
1453 	    node->tn_mode, node->tn_uid, node->tn_gid,
1454 	    (uintmax_t)node->tn_size, node->tn_status);
1455 
1456 	if (vp->v_type == VFIFO)
1457 		fifo_printinfo(vp);
1458 
1459 	printf("\n");
1460 
1461 	return 0;
1462 }
1463 
1464 /* --------------------------------------------------------------------- */
1465 
1466 static int
1467 tmpfs_pathconf(struct vop_pathconf_args *v)
1468 {
1469 	int name = v->a_name;
1470 	register_t *retval = v->a_retval;
1471 
1472 	int error;
1473 
1474 	error = 0;
1475 
1476 	switch (name) {
1477 	case _PC_LINK_MAX:
1478 		*retval = LINK_MAX;
1479 		break;
1480 
1481 	case _PC_NAME_MAX:
1482 		*retval = NAME_MAX;
1483 		break;
1484 
1485 	case _PC_PATH_MAX:
1486 		*retval = PATH_MAX;
1487 		break;
1488 
1489 	case _PC_PIPE_BUF:
1490 		*retval = PIPE_BUF;
1491 		break;
1492 
1493 	case _PC_CHOWN_RESTRICTED:
1494 		*retval = 1;
1495 		break;
1496 
1497 	case _PC_NO_TRUNC:
1498 		*retval = 1;
1499 		break;
1500 
1501 	case _PC_SYNC_IO:
1502 		*retval = 1;
1503 		break;
1504 
1505 	case _PC_FILESIZEBITS:
1506 		*retval = 0; /* XXX Don't know which value should I return. */
1507 		break;
1508 
1509 	default:
1510 		error = EINVAL;
1511 	}
1512 
1513 	return error;
1514 }
1515 
1516 static int
1517 tmpfs_vptofh(struct vop_vptofh_args *ap)
1518 {
1519 	struct tmpfs_fid *tfhp;
1520 	struct tmpfs_node *node;
1521 
1522 	tfhp = (struct tmpfs_fid *)ap->a_fhp;
1523 	node = VP_TO_TMPFS_NODE(ap->a_vp);
1524 
1525 	tfhp->tf_len = sizeof(struct tmpfs_fid);
1526 	tfhp->tf_id = node->tn_id;
1527 	tfhp->tf_gen = node->tn_gen;
1528 
1529 	return (0);
1530 }
1531 
1532 /* --------------------------------------------------------------------- */
1533 
1534 /*
1535  * vnode operations vector used for files stored in a tmpfs file system.
1536  */
1537 struct vop_vector tmpfs_vnodeop_entries = {
1538 	.vop_default =			&default_vnodeops,
1539 	.vop_lookup =			vfs_cache_lookup,
1540 	.vop_cachedlookup =		tmpfs_lookup,
1541 	.vop_create =			tmpfs_create,
1542 	.vop_mknod =			tmpfs_mknod,
1543 	.vop_open =			tmpfs_open,
1544 	.vop_close =			tmpfs_close,
1545 	.vop_access =			tmpfs_access,
1546 	.vop_getattr =			tmpfs_getattr,
1547 	.vop_setattr =			tmpfs_setattr,
1548 	.vop_read =			tmpfs_read,
1549 	.vop_write =			tmpfs_write,
1550 	.vop_fsync =			tmpfs_fsync,
1551 	.vop_remove =			tmpfs_remove,
1552 	.vop_link =			tmpfs_link,
1553 	.vop_rename =			tmpfs_rename,
1554 	.vop_mkdir =			tmpfs_mkdir,
1555 	.vop_rmdir =			tmpfs_rmdir,
1556 	.vop_symlink =			tmpfs_symlink,
1557 	.vop_readdir =			tmpfs_readdir,
1558 	.vop_readlink =			tmpfs_readlink,
1559 	.vop_inactive =			tmpfs_inactive,
1560 	.vop_reclaim =			tmpfs_reclaim,
1561 	.vop_print =			tmpfs_print,
1562 	.vop_pathconf =			tmpfs_pathconf,
1563 	.vop_vptofh =			tmpfs_vptofh,
1564 	.vop_bmap =			VOP_EOPNOTSUPP,
1565 };
1566 
1567