xref: /freebsd/sys/fs/tmpfs/tmpfs_vnops.c (revision 6e660824a82f590542932de52f128db584029893)
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/lock.h>
43 #include <sys/namei.h>
44 #include <sys/priv.h>
45 #include <sys/proc.h>
46 #include <sys/rwlock.h>
47 #include <sys/sched.h>
48 #include <sys/sf_buf.h>
49 #include <sys/stat.h>
50 #include <sys/systm.h>
51 #include <sys/sysctl.h>
52 #include <sys/unistd.h>
53 #include <sys/vnode.h>
54 
55 #include <vm/vm.h>
56 #include <vm/vm_param.h>
57 #include <vm/vm_object.h>
58 #include <vm/vm_page.h>
59 #include <vm/vm_pager.h>
60 
61 #include <fs/tmpfs/tmpfs_vnops.h>
62 #include <fs/tmpfs/tmpfs.h>
63 
64 SYSCTL_DECL(_vfs_tmpfs);
65 
66 static volatile int tmpfs_rename_restarts;
67 SYSCTL_INT(_vfs_tmpfs, OID_AUTO, rename_restarts, CTLFLAG_RD,
68     __DEVOLATILE(int *, &tmpfs_rename_restarts), 0,
69     "Times rename had to restart due to lock contention");
70 
71 /* --------------------------------------------------------------------- */
72 
73 static int
74 tmpfs_lookup(struct vop_cachedlookup_args *v)
75 {
76 	struct vnode *dvp = v->a_dvp;
77 	struct vnode **vpp = v->a_vpp;
78 	struct componentname *cnp = v->a_cnp;
79 
80 	int error;
81 	struct tmpfs_dirent *de;
82 	struct tmpfs_node *dnode;
83 
84 	dnode = VP_TO_TMPFS_DIR(dvp);
85 	*vpp = NULLVP;
86 
87 	/* Check accessibility of requested node as a first step. */
88 	error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, cnp->cn_thread);
89 	if (error != 0)
90 		goto out;
91 
92 	/* We cannot be requesting the parent directory of the root node. */
93 	MPASS(IMPLIES(dnode->tn_type == VDIR &&
94 	    dnode->tn_dir.tn_parent == dnode,
95 	    !(cnp->cn_flags & ISDOTDOT)));
96 
97 	TMPFS_ASSERT_LOCKED(dnode);
98 	if (dnode->tn_dir.tn_parent == NULL) {
99 		error = ENOENT;
100 		goto out;
101 	}
102 	if (cnp->cn_flags & ISDOTDOT) {
103 		int ltype = 0;
104 
105 		ltype = VOP_ISLOCKED(dvp);
106 		vhold(dvp);
107 		VOP_UNLOCK(dvp, 0);
108 		/* Allocate a new vnode on the matching entry. */
109 		error = tmpfs_alloc_vp(dvp->v_mount, dnode->tn_dir.tn_parent,
110 		    cnp->cn_lkflags, vpp);
111 
112 		vn_lock(dvp, ltype | LK_RETRY);
113 		vdrop(dvp);
114 	} else if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
115 		VREF(dvp);
116 		*vpp = dvp;
117 		error = 0;
118 	} else {
119 		de = tmpfs_dir_lookup(dnode, NULL, cnp);
120 		if (de != NULL && de->td_node == NULL)
121 			cnp->cn_flags |= ISWHITEOUT;
122 		if (de == NULL || de->td_node == NULL) {
123 			/* The entry was not found in the directory.
124 			 * This is OK if we are creating or renaming an
125 			 * entry and are working on the last component of
126 			 * the path name. */
127 			if ((cnp->cn_flags & ISLASTCN) &&
128 			    (cnp->cn_nameiop == CREATE || \
129 			    cnp->cn_nameiop == RENAME ||
130 			    (cnp->cn_nameiop == DELETE &&
131 			    cnp->cn_flags & DOWHITEOUT &&
132 			    cnp->cn_flags & ISWHITEOUT))) {
133 				error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred,
134 				    cnp->cn_thread);
135 				if (error != 0)
136 					goto out;
137 
138 				/* Keep the component name in the buffer for
139 				 * future uses. */
140 				cnp->cn_flags |= SAVENAME;
141 
142 				error = EJUSTRETURN;
143 			} else
144 				error = ENOENT;
145 		} else {
146 			struct tmpfs_node *tnode;
147 
148 			/* The entry was found, so get its associated
149 			 * tmpfs_node. */
150 			tnode = de->td_node;
151 
152 			/* If we are not at the last path component and
153 			 * found a non-directory or non-link entry (which
154 			 * may itself be pointing to a directory), raise
155 			 * an error. */
156 			if ((tnode->tn_type != VDIR &&
157 			    tnode->tn_type != VLNK) &&
158 			    !(cnp->cn_flags & ISLASTCN)) {
159 				error = ENOTDIR;
160 				goto out;
161 			}
162 
163 			/* If we are deleting or renaming the entry, keep
164 			 * track of its tmpfs_dirent so that it can be
165 			 * easily deleted later. */
166 			if ((cnp->cn_flags & ISLASTCN) &&
167 			    (cnp->cn_nameiop == DELETE ||
168 			    cnp->cn_nameiop == RENAME)) {
169 				error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred,
170 				    cnp->cn_thread);
171 				if (error != 0)
172 					goto out;
173 
174 				/* Allocate a new vnode on the matching entry. */
175 				error = tmpfs_alloc_vp(dvp->v_mount, tnode,
176 						cnp->cn_lkflags, vpp);
177 				if (error != 0)
178 					goto out;
179 
180 				if ((dnode->tn_mode & S_ISTXT) &&
181 				  VOP_ACCESS(dvp, VADMIN, cnp->cn_cred, cnp->cn_thread) &&
182 				  VOP_ACCESS(*vpp, VADMIN, cnp->cn_cred, cnp->cn_thread)) {
183 					error = EPERM;
184 					vput(*vpp);
185 					*vpp = NULL;
186 					goto out;
187 				}
188 				cnp->cn_flags |= SAVENAME;
189 			} else {
190 				error = tmpfs_alloc_vp(dvp->v_mount, tnode,
191 						cnp->cn_lkflags, vpp);
192 			}
193 		}
194 	}
195 
196 	/* Store the result of this lookup in the cache.  Avoid this if the
197 	 * request was for creation, as it does not improve timings on
198 	 * emprical tests. */
199 	if ((cnp->cn_flags & MAKEENTRY) && cnp->cn_nameiop != CREATE)
200 		cache_enter(dvp, *vpp, cnp);
201 
202 out:
203 	/* If there were no errors, *vpp cannot be null and it must be
204 	 * locked. */
205 	MPASS(IFF(error == 0, *vpp != NULLVP && VOP_ISLOCKED(*vpp)));
206 
207 	return error;
208 }
209 
210 /* --------------------------------------------------------------------- */
211 
212 static int
213 tmpfs_create(struct vop_create_args *v)
214 {
215 	struct vnode *dvp = v->a_dvp;
216 	struct vnode **vpp = v->a_vpp;
217 	struct componentname *cnp = v->a_cnp;
218 	struct vattr *vap = v->a_vap;
219 
220 	MPASS(vap->va_type == VREG || vap->va_type == VSOCK);
221 
222 	return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
223 }
224 /* --------------------------------------------------------------------- */
225 
226 static int
227 tmpfs_mknod(struct vop_mknod_args *v)
228 {
229 	struct vnode *dvp = v->a_dvp;
230 	struct vnode **vpp = v->a_vpp;
231 	struct componentname *cnp = v->a_cnp;
232 	struct vattr *vap = v->a_vap;
233 
234 	if (vap->va_type != VBLK && vap->va_type != VCHR &&
235 	    vap->va_type != VFIFO)
236 		return EINVAL;
237 
238 	return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
239 }
240 
241 /* --------------------------------------------------------------------- */
242 
243 static int
244 tmpfs_open(struct vop_open_args *v)
245 {
246 	struct vnode *vp = v->a_vp;
247 	int mode = v->a_mode;
248 
249 	int error;
250 	struct tmpfs_node *node;
251 
252 	MPASS(VOP_ISLOCKED(vp));
253 
254 	node = VP_TO_TMPFS_NODE(vp);
255 
256 	/* The file is still active but all its names have been removed
257 	 * (e.g. by a "rmdir $(pwd)").  It cannot be opened any more as
258 	 * it is about to die. */
259 	if (node->tn_links < 1)
260 		return (ENOENT);
261 
262 	/* If the file is marked append-only, deny write requests. */
263 	if (node->tn_flags & APPEND && (mode & (FWRITE | O_APPEND)) == FWRITE)
264 		error = EPERM;
265 	else {
266 		error = 0;
267 		/* For regular files, the call below is nop. */
268 		vnode_create_vobject(vp, node->tn_size, v->a_td);
269 	}
270 
271 	MPASS(VOP_ISLOCKED(vp));
272 	return error;
273 }
274 
275 /* --------------------------------------------------------------------- */
276 
277 static int
278 tmpfs_close(struct vop_close_args *v)
279 {
280 	struct vnode *vp = v->a_vp;
281 
282 	/* Update node times. */
283 	tmpfs_update(vp);
284 
285 	return (0);
286 }
287 
288 /* --------------------------------------------------------------------- */
289 
290 int
291 tmpfs_access(struct vop_access_args *v)
292 {
293 	struct vnode *vp = v->a_vp;
294 	accmode_t accmode = v->a_accmode;
295 	struct ucred *cred = v->a_cred;
296 
297 	int error;
298 	struct tmpfs_node *node;
299 
300 	MPASS(VOP_ISLOCKED(vp));
301 
302 	node = VP_TO_TMPFS_NODE(vp);
303 
304 	switch (vp->v_type) {
305 	case VDIR:
306 		/* FALLTHROUGH */
307 	case VLNK:
308 		/* FALLTHROUGH */
309 	case VREG:
310 		if (accmode & VWRITE && vp->v_mount->mnt_flag & MNT_RDONLY) {
311 			error = EROFS;
312 			goto out;
313 		}
314 		break;
315 
316 	case VBLK:
317 		/* FALLTHROUGH */
318 	case VCHR:
319 		/* FALLTHROUGH */
320 	case VSOCK:
321 		/* FALLTHROUGH */
322 	case VFIFO:
323 		break;
324 
325 	default:
326 		error = EINVAL;
327 		goto out;
328 	}
329 
330 	if (accmode & VWRITE && node->tn_flags & IMMUTABLE) {
331 		error = EPERM;
332 		goto out;
333 	}
334 
335 	error = vaccess(vp->v_type, node->tn_mode, node->tn_uid,
336 	    node->tn_gid, accmode, cred, NULL);
337 
338 out:
339 	MPASS(VOP_ISLOCKED(vp));
340 
341 	return error;
342 }
343 
344 /* --------------------------------------------------------------------- */
345 
346 int
347 tmpfs_getattr(struct vop_getattr_args *v)
348 {
349 	struct vnode *vp = v->a_vp;
350 	struct vattr *vap = v->a_vap;
351 
352 	struct tmpfs_node *node;
353 
354 	node = VP_TO_TMPFS_NODE(vp);
355 
356 	tmpfs_update(vp);
357 
358 	vap->va_type = vp->v_type;
359 	vap->va_mode = node->tn_mode;
360 	vap->va_nlink = node->tn_links;
361 	vap->va_uid = node->tn_uid;
362 	vap->va_gid = node->tn_gid;
363 	vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
364 	vap->va_fileid = node->tn_id;
365 	vap->va_size = node->tn_size;
366 	vap->va_blocksize = PAGE_SIZE;
367 	vap->va_atime = node->tn_atime;
368 	vap->va_mtime = node->tn_mtime;
369 	vap->va_ctime = node->tn_ctime;
370 	vap->va_birthtime = node->tn_birthtime;
371 	vap->va_gen = node->tn_gen;
372 	vap->va_flags = node->tn_flags;
373 	vap->va_rdev = (vp->v_type == VBLK || vp->v_type == VCHR) ?
374 		node->tn_rdev : NODEV;
375 	vap->va_bytes = round_page(node->tn_size);
376 	vap->va_filerev = 0;
377 
378 	return 0;
379 }
380 
381 /* --------------------------------------------------------------------- */
382 
383 /* XXX Should this operation be atomic?  I think it should, but code in
384  * XXX other places (e.g., ufs) doesn't seem to be... */
385 int
386 tmpfs_setattr(struct vop_setattr_args *v)
387 {
388 	struct vnode *vp = v->a_vp;
389 	struct vattr *vap = v->a_vap;
390 	struct ucred *cred = v->a_cred;
391 	struct thread *td = curthread;
392 
393 	int error;
394 
395 	MPASS(VOP_ISLOCKED(vp));
396 
397 	error = 0;
398 
399 	/* Abort if any unsettable attribute is given. */
400 	if (vap->va_type != VNON ||
401 	    vap->va_nlink != VNOVAL ||
402 	    vap->va_fsid != VNOVAL ||
403 	    vap->va_fileid != VNOVAL ||
404 	    vap->va_blocksize != VNOVAL ||
405 	    vap->va_gen != VNOVAL ||
406 	    vap->va_rdev != VNOVAL ||
407 	    vap->va_bytes != VNOVAL)
408 		error = EINVAL;
409 
410 	if (error == 0 && (vap->va_flags != VNOVAL))
411 		error = tmpfs_chflags(vp, vap->va_flags, cred, td);
412 
413 	if (error == 0 && (vap->va_size != VNOVAL))
414 		error = tmpfs_chsize(vp, vap->va_size, cred, td);
415 
416 	if (error == 0 && (vap->va_uid != VNOVAL || vap->va_gid != VNOVAL))
417 		error = tmpfs_chown(vp, vap->va_uid, vap->va_gid, cred, td);
418 
419 	if (error == 0 && (vap->va_mode != (mode_t)VNOVAL))
420 		error = tmpfs_chmod(vp, vap->va_mode, cred, td);
421 
422 	if (error == 0 && ((vap->va_atime.tv_sec != VNOVAL &&
423 	    vap->va_atime.tv_nsec != VNOVAL) ||
424 	    (vap->va_mtime.tv_sec != VNOVAL &&
425 	    vap->va_mtime.tv_nsec != VNOVAL) ||
426 	    (vap->va_birthtime.tv_sec != VNOVAL &&
427 	    vap->va_birthtime.tv_nsec != VNOVAL)))
428 		error = tmpfs_chtimes(vp, &vap->va_atime, &vap->va_mtime,
429 			&vap->va_birthtime, vap->va_vaflags, cred, td);
430 
431 	/* Update the node times.  We give preference to the error codes
432 	 * generated by this function rather than the ones that may arise
433 	 * from tmpfs_update. */
434 	tmpfs_update(vp);
435 
436 	MPASS(VOP_ISLOCKED(vp));
437 
438 	return error;
439 }
440 
441 static int
442 tmpfs_nocacheread(vm_object_t tobj, vm_pindex_t idx,
443     vm_offset_t offset, size_t tlen, struct uio *uio)
444 {
445 	vm_page_t	m;
446 	int		error, rv;
447 
448 	VM_OBJECT_WLOCK(tobj);
449 
450 	/*
451 	 * The kern_sendfile() code calls vn_rdwr() with the page
452 	 * soft-busied.  Ignore the soft-busy state here. Parallel
453 	 * reads of the page content from disk are prevented by
454 	 * VPO_BUSY.
455 	 *
456 	 * Although the tmpfs vnode lock is held here, it is
457 	 * nonetheless safe to sleep waiting for a free page.  The
458 	 * pageout daemon does not need to acquire the tmpfs vnode
459 	 * lock to page out tobj's pages because tobj is a OBJT_SWAP
460 	 * type object.
461 	 */
462 	m = vm_page_grab(tobj, idx, VM_ALLOC_NORMAL | VM_ALLOC_RETRY |
463 	    VM_ALLOC_IGN_SBUSY | VM_ALLOC_NOBUSY);
464 	if (m->valid != VM_PAGE_BITS_ALL) {
465 		vm_page_busy(m);
466 		if (vm_pager_has_page(tobj, idx, NULL, NULL)) {
467 			rv = vm_pager_get_pages(tobj, &m, 1, 0);
468 			m = vm_page_lookup(tobj, idx);
469 			if (m == NULL) {
470 				printf(
471 		    "tmpfs: vm_obj %p idx %jd null lookup rv %d\n",
472 				    tobj, idx, rv);
473 				VM_OBJECT_WUNLOCK(tobj);
474 				return (EIO);
475 			}
476 			if (rv != VM_PAGER_OK) {
477 				printf(
478 		    "tmpfs: vm_obj %p idx %jd valid %x pager error %d\n",
479 				    tobj, idx, m->valid, rv);
480 				vm_page_lock(m);
481 				vm_page_free(m);
482 				vm_page_unlock(m);
483 				VM_OBJECT_WUNLOCK(tobj);
484 				return (EIO);
485 			}
486 		} else
487 			vm_page_zero_invalid(m, TRUE);
488 		vm_page_wakeup(m);
489 	}
490 	vm_page_lock(m);
491 	vm_page_hold(m);
492 	vm_page_unlock(m);
493 	VM_OBJECT_WUNLOCK(tobj);
494 	error = uiomove_fromphys(&m, offset, tlen, uio);
495 	vm_page_lock(m);
496 	vm_page_unhold(m);
497 	if (m->queue == PQ_NONE) {
498 		vm_page_deactivate(m);
499 	} else {
500 		/* Requeue to maintain LRU ordering. */
501 		vm_page_requeue(m);
502 	}
503 	vm_page_unlock(m);
504 
505 	return (error);
506 }
507 
508 static int
509 tmpfs_read(struct vop_read_args *v)
510 {
511 	struct vnode *vp = v->a_vp;
512 	struct uio *uio = v->a_uio;
513 	struct tmpfs_node *node;
514 	vm_object_t uobj;
515 	size_t len;
516 	int resid;
517 	int error = 0;
518 	vm_pindex_t	idx;
519 	vm_offset_t	offset;
520 	off_t		addr;
521 	size_t		tlen;
522 
523 	node = VP_TO_TMPFS_NODE(vp);
524 
525 	if (vp->v_type != VREG) {
526 		error = EISDIR;
527 		goto out;
528 	}
529 
530 	if (uio->uio_offset < 0) {
531 		error = EINVAL;
532 		goto out;
533 	}
534 
535 	node->tn_status |= TMPFS_NODE_ACCESSED;
536 
537 	uobj = node->tn_reg.tn_aobj;
538 	while ((resid = uio->uio_resid) > 0) {
539 		error = 0;
540 		if (node->tn_size <= uio->uio_offset)
541 			break;
542 		len = MIN(node->tn_size - uio->uio_offset, resid);
543 		if (len == 0)
544 			break;
545 		addr = uio->uio_offset;
546 		idx = OFF_TO_IDX(addr);
547 		offset = addr & PAGE_MASK;
548 		tlen = MIN(PAGE_SIZE - offset, len);
549 		error = tmpfs_nocacheread(uobj, idx, offset, tlen, uio);
550 		if ((error != 0) || (resid == uio->uio_resid))
551 			break;
552 	}
553 
554 out:
555 
556 	return error;
557 }
558 
559 /* --------------------------------------------------------------------- */
560 
561 static int
562 tmpfs_mappedwrite(vm_object_t tobj, size_t len, struct uio *uio)
563 {
564 	vm_pindex_t	idx;
565 	vm_page_t	tpg;
566 	vm_offset_t	offset;
567 	off_t		addr;
568 	size_t		tlen;
569 	int		error, rv;
570 
571 	error = 0;
572 
573 	addr = uio->uio_offset;
574 	idx = OFF_TO_IDX(addr);
575 	offset = addr & PAGE_MASK;
576 	tlen = MIN(PAGE_SIZE - offset, len);
577 
578 	VM_OBJECT_WLOCK(tobj);
579 	tpg = vm_page_grab(tobj, idx, VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY |
580 	    VM_ALLOC_RETRY);
581 	if (tpg->valid != VM_PAGE_BITS_ALL) {
582 		vm_page_busy(tpg);
583 		if (vm_pager_has_page(tobj, idx, NULL, NULL)) {
584 			rv = vm_pager_get_pages(tobj, &tpg, 1, 0);
585 			tpg = vm_page_lookup(tobj, idx);
586 			if (tpg == NULL) {
587 				printf(
588 		    "tmpfs: vm_obj %p idx %jd null lookup rv %d\n",
589 				    tobj, idx, rv);
590 				VM_OBJECT_WUNLOCK(tobj);
591 				return (EIO);
592 			}
593 			if (rv != VM_PAGER_OK) {
594 				printf(
595 		    "tmpfs: vm_obj %p idx %jd valid %x pager error %d\n",
596 				    tobj, idx, tpg->valid, rv);
597 				vm_page_lock(tpg);
598 				vm_page_free(tpg);
599 				vm_page_unlock(tpg);
600 				VM_OBJECT_WUNLOCK(tobj);
601 				return (EIO);
602 			}
603 		} else
604 			vm_page_zero_invalid(tpg, TRUE);
605 		vm_page_wakeup(tpg);
606 	}
607 	vm_page_lock(tpg);
608 	vm_page_hold(tpg);
609 	vm_page_unlock(tpg);
610 	VM_OBJECT_WUNLOCK(tobj);
611 	error = uiomove_fromphys(&tpg, offset, tlen, uio);
612 	VM_OBJECT_WLOCK(tobj);
613 	if (error == 0)
614 		vm_page_dirty(tpg);
615 	vm_page_lock(tpg);
616 	vm_page_unhold(tpg);
617 	if (tpg->queue == PQ_NONE) {
618 		vm_page_deactivate(tpg);
619 	} else {
620 		/* Requeue to maintain LRU ordering. */
621 		vm_page_requeue(tpg);
622 	}
623 	vm_page_unlock(tpg);
624 	VM_OBJECT_WUNLOCK(tobj);
625 
626 	return	(error);
627 }
628 
629 static int
630 tmpfs_write(struct vop_write_args *v)
631 {
632 	struct vnode *vp = v->a_vp;
633 	struct uio *uio = v->a_uio;
634 	int ioflag = v->a_ioflag;
635 
636 	boolean_t extended;
637 	int error = 0;
638 	off_t oldsize;
639 	struct tmpfs_node *node;
640 	vm_object_t uobj;
641 	size_t len;
642 	int resid;
643 
644 	node = VP_TO_TMPFS_NODE(vp);
645 	oldsize = node->tn_size;
646 
647 	if (uio->uio_offset < 0 || vp->v_type != VREG) {
648 		error = EINVAL;
649 		goto out;
650 	}
651 
652 	if (uio->uio_resid == 0) {
653 		error = 0;
654 		goto out;
655 	}
656 
657 	if (ioflag & IO_APPEND)
658 		uio->uio_offset = node->tn_size;
659 
660 	if (uio->uio_offset + uio->uio_resid >
661 	  VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize)
662 		return (EFBIG);
663 
664 	if (vn_rlimit_fsize(vp, uio, uio->uio_td))
665 		return (EFBIG);
666 
667 	extended = uio->uio_offset + uio->uio_resid > node->tn_size;
668 	if (extended) {
669 		error = tmpfs_reg_resize(vp, uio->uio_offset + uio->uio_resid,
670 		    FALSE);
671 		if (error != 0)
672 			goto out;
673 	}
674 
675 	uobj = node->tn_reg.tn_aobj;
676 	while ((resid = uio->uio_resid) > 0) {
677 		if (node->tn_size <= uio->uio_offset)
678 			break;
679 		len = MIN(node->tn_size - uio->uio_offset, resid);
680 		if (len == 0)
681 			break;
682 		error = tmpfs_mappedwrite(uobj, len, uio);
683 		if ((error != 0) || (resid == uio->uio_resid))
684 			break;
685 	}
686 
687 	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
688 	    (extended ? TMPFS_NODE_CHANGED : 0);
689 
690 	if (node->tn_mode & (S_ISUID | S_ISGID)) {
691 		if (priv_check_cred(v->a_cred, PRIV_VFS_RETAINSUGID, 0))
692 			node->tn_mode &= ~(S_ISUID | S_ISGID);
693 	}
694 
695 	if (error != 0)
696 		(void)tmpfs_reg_resize(vp, oldsize, TRUE);
697 
698 out:
699 	MPASS(IMPLIES(error == 0, uio->uio_resid == 0));
700 	MPASS(IMPLIES(error != 0, oldsize == node->tn_size));
701 
702 	return error;
703 }
704 
705 /* --------------------------------------------------------------------- */
706 
707 static int
708 tmpfs_fsync(struct vop_fsync_args *v)
709 {
710 	struct vnode *vp = v->a_vp;
711 
712 	MPASS(VOP_ISLOCKED(vp));
713 
714 	tmpfs_update(vp);
715 
716 	return 0;
717 }
718 
719 /* --------------------------------------------------------------------- */
720 
721 static int
722 tmpfs_remove(struct vop_remove_args *v)
723 {
724 	struct vnode *dvp = v->a_dvp;
725 	struct vnode *vp = v->a_vp;
726 
727 	int error;
728 	struct tmpfs_dirent *de;
729 	struct tmpfs_mount *tmp;
730 	struct tmpfs_node *dnode;
731 	struct tmpfs_node *node;
732 
733 	MPASS(VOP_ISLOCKED(dvp));
734 	MPASS(VOP_ISLOCKED(vp));
735 
736 	if (vp->v_type == VDIR) {
737 		error = EISDIR;
738 		goto out;
739 	}
740 
741 	dnode = VP_TO_TMPFS_DIR(dvp);
742 	node = VP_TO_TMPFS_NODE(vp);
743 	tmp = VFS_TO_TMPFS(vp->v_mount);
744 	de = tmpfs_dir_lookup(dnode, node, v->a_cnp);
745 	MPASS(de != NULL);
746 
747 	/* Files marked as immutable or append-only cannot be deleted. */
748 	if ((node->tn_flags & (IMMUTABLE | APPEND | NOUNLINK)) ||
749 	    (dnode->tn_flags & APPEND)) {
750 		error = EPERM;
751 		goto out;
752 	}
753 
754 	/* Remove the entry from the directory; as it is a file, we do not
755 	 * have to change the number of hard links of the directory. */
756 	tmpfs_dir_detach(dvp, de);
757 	if (v->a_cnp->cn_flags & DOWHITEOUT)
758 		tmpfs_dir_whiteout_add(dvp, v->a_cnp);
759 
760 	/* Free the directory entry we just deleted.  Note that the node
761 	 * referred by it will not be removed until the vnode is really
762 	 * reclaimed. */
763 	tmpfs_free_dirent(tmp, de);
764 
765 	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED;
766 	error = 0;
767 
768 out:
769 
770 	return error;
771 }
772 
773 /* --------------------------------------------------------------------- */
774 
775 static int
776 tmpfs_link(struct vop_link_args *v)
777 {
778 	struct vnode *dvp = v->a_tdvp;
779 	struct vnode *vp = v->a_vp;
780 	struct componentname *cnp = v->a_cnp;
781 
782 	int error;
783 	struct tmpfs_dirent *de;
784 	struct tmpfs_node *node;
785 
786 	MPASS(VOP_ISLOCKED(dvp));
787 	MPASS(cnp->cn_flags & HASBUF);
788 	MPASS(dvp != vp); /* XXX When can this be false? */
789 
790 	node = VP_TO_TMPFS_NODE(vp);
791 
792 	/* XXX: Why aren't the following two tests done by the caller? */
793 
794 	/* Hard links of directories are forbidden. */
795 	if (vp->v_type == VDIR) {
796 		error = EPERM;
797 		goto out;
798 	}
799 
800 	/* Cannot create cross-device links. */
801 	if (dvp->v_mount != vp->v_mount) {
802 		error = EXDEV;
803 		goto out;
804 	}
805 
806 	/* Ensure that we do not overflow the maximum number of links imposed
807 	 * by the system. */
808 	MPASS(node->tn_links <= LINK_MAX);
809 	if (node->tn_links == LINK_MAX) {
810 		error = EMLINK;
811 		goto out;
812 	}
813 
814 	/* We cannot create links of files marked immutable or append-only. */
815 	if (node->tn_flags & (IMMUTABLE | APPEND)) {
816 		error = EPERM;
817 		goto out;
818 	}
819 
820 	/* Allocate a new directory entry to represent the node. */
821 	error = tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), node,
822 	    cnp->cn_nameptr, cnp->cn_namelen, &de);
823 	if (error != 0)
824 		goto out;
825 
826 	/* Insert the new directory entry into the appropriate directory. */
827 	if (cnp->cn_flags & ISWHITEOUT)
828 		tmpfs_dir_whiteout_remove(dvp, cnp);
829 	tmpfs_dir_attach(dvp, de);
830 
831 	/* vp link count has changed, so update node times. */
832 	node->tn_status |= TMPFS_NODE_CHANGED;
833 	tmpfs_update(vp);
834 
835 	error = 0;
836 
837 out:
838 	return error;
839 }
840 
841 /* --------------------------------------------------------------------- */
842 
843 /*
844  * We acquire all but fdvp locks using non-blocking acquisitions.  If we
845  * fail to acquire any lock in the path we will drop all held locks,
846  * acquire the new lock in a blocking fashion, and then release it and
847  * restart the rename.  This acquire/release step ensures that we do not
848  * spin on a lock waiting for release.  On error release all vnode locks
849  * and decrement references the way tmpfs_rename() would do.
850  */
851 static int
852 tmpfs_rename_relock(struct vnode *fdvp, struct vnode **fvpp,
853     struct vnode *tdvp, struct vnode **tvpp,
854     struct componentname *fcnp, struct componentname *tcnp)
855 {
856 	struct vnode *nvp;
857 	struct mount *mp;
858 	struct tmpfs_dirent *de;
859 	int error, restarts = 0;
860 
861 	VOP_UNLOCK(tdvp, 0);
862 	if (*tvpp != NULL && *tvpp != tdvp)
863 		VOP_UNLOCK(*tvpp, 0);
864 	mp = fdvp->v_mount;
865 
866 relock:
867 	restarts += 1;
868 	error = vn_lock(fdvp, LK_EXCLUSIVE);
869 	if (error)
870 		goto releout;
871 	if (vn_lock(tdvp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
872 		VOP_UNLOCK(fdvp, 0);
873 		error = vn_lock(tdvp, LK_EXCLUSIVE);
874 		if (error)
875 			goto releout;
876 		VOP_UNLOCK(tdvp, 0);
877 		goto relock;
878 	}
879 	/*
880 	 * Re-resolve fvp to be certain it still exists and fetch the
881 	 * correct vnode.
882 	 */
883 	de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(fdvp), NULL, fcnp);
884 	if (de == NULL) {
885 		VOP_UNLOCK(fdvp, 0);
886 		VOP_UNLOCK(tdvp, 0);
887 		if ((fcnp->cn_flags & ISDOTDOT) != 0 ||
888 		    (fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.'))
889 			error = EINVAL;
890 		else
891 			error = ENOENT;
892 		goto releout;
893 	}
894 	error = tmpfs_alloc_vp(mp, de->td_node, LK_EXCLUSIVE | LK_NOWAIT, &nvp);
895 	if (error != 0) {
896 		VOP_UNLOCK(fdvp, 0);
897 		VOP_UNLOCK(tdvp, 0);
898 		if (error != EBUSY)
899 			goto releout;
900 		error = tmpfs_alloc_vp(mp, de->td_node, LK_EXCLUSIVE, &nvp);
901 		if (error != 0)
902 			goto releout;
903 		VOP_UNLOCK(nvp, 0);
904 		/*
905 		 * Concurrent rename race.
906 		 */
907 		if (nvp == tdvp) {
908 			vrele(nvp);
909 			error = EINVAL;
910 			goto releout;
911 		}
912 		vrele(*fvpp);
913 		*fvpp = nvp;
914 		goto relock;
915 	}
916 	vrele(*fvpp);
917 	*fvpp = nvp;
918 	VOP_UNLOCK(*fvpp, 0);
919 	/*
920 	 * Re-resolve tvp and acquire the vnode lock if present.
921 	 */
922 	de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(tdvp), NULL, tcnp);
923 	/*
924 	 * If tvp disappeared we just carry on.
925 	 */
926 	if (de == NULL && *tvpp != NULL) {
927 		vrele(*tvpp);
928 		*tvpp = NULL;
929 	}
930 	/*
931 	 * Get the tvp ino if the lookup succeeded.  We may have to restart
932 	 * if the non-blocking acquire fails.
933 	 */
934 	if (de != NULL) {
935 		nvp = NULL;
936 		error = tmpfs_alloc_vp(mp, de->td_node,
937 		    LK_EXCLUSIVE | LK_NOWAIT, &nvp);
938 		if (*tvpp != NULL)
939 			vrele(*tvpp);
940 		*tvpp = nvp;
941 		if (error != 0) {
942 			VOP_UNLOCK(fdvp, 0);
943 			VOP_UNLOCK(tdvp, 0);
944 			if (error != EBUSY)
945 				goto releout;
946 			error = tmpfs_alloc_vp(mp, de->td_node, LK_EXCLUSIVE,
947 			    &nvp);
948 			if (error != 0)
949 				goto releout;
950 			VOP_UNLOCK(nvp, 0);
951 			/*
952 			 * fdvp contains fvp, thus tvp (=fdvp) is not empty.
953 			 */
954 			if (nvp == fdvp) {
955 				error = ENOTEMPTY;
956 				goto releout;
957 			}
958 			goto relock;
959 		}
960 	}
961 	tmpfs_rename_restarts += restarts;
962 
963 	return (0);
964 
965 releout:
966 	vrele(fdvp);
967 	vrele(*fvpp);
968 	vrele(tdvp);
969 	if (*tvpp != NULL)
970 		vrele(*tvpp);
971 	tmpfs_rename_restarts += restarts;
972 
973 	return (error);
974 }
975 
976 static int
977 tmpfs_rename(struct vop_rename_args *v)
978 {
979 	struct vnode *fdvp = v->a_fdvp;
980 	struct vnode *fvp = v->a_fvp;
981 	struct componentname *fcnp = v->a_fcnp;
982 	struct vnode *tdvp = v->a_tdvp;
983 	struct vnode *tvp = v->a_tvp;
984 	struct componentname *tcnp = v->a_tcnp;
985 	struct mount *mp = NULL;
986 
987 	char *newname;
988 	int error;
989 	struct tmpfs_dirent *de;
990 	struct tmpfs_mount *tmp;
991 	struct tmpfs_node *fdnode;
992 	struct tmpfs_node *fnode;
993 	struct tmpfs_node *tnode;
994 	struct tmpfs_node *tdnode;
995 
996 	MPASS(VOP_ISLOCKED(tdvp));
997 	MPASS(IMPLIES(tvp != NULL, VOP_ISLOCKED(tvp)));
998 	MPASS(fcnp->cn_flags & HASBUF);
999 	MPASS(tcnp->cn_flags & HASBUF);
1000 
1001 	/* Disallow cross-device renames.
1002 	 * XXX Why isn't this done by the caller? */
1003 	if (fvp->v_mount != tdvp->v_mount ||
1004 	    (tvp != NULL && fvp->v_mount != tvp->v_mount)) {
1005 		error = EXDEV;
1006 		goto out;
1007 	}
1008 
1009 	/* If source and target are the same file, there is nothing to do. */
1010 	if (fvp == tvp) {
1011 		error = 0;
1012 		goto out;
1013 	}
1014 
1015 	/* If we need to move the directory between entries, lock the
1016 	 * source so that we can safely operate on it. */
1017 	if (fdvp != tdvp && fdvp != tvp) {
1018 		if (vn_lock(fdvp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
1019 			mp = tdvp->v_mount;
1020 			error = vfs_busy(mp, 0);
1021 			if (error != 0) {
1022 				mp = NULL;
1023 				goto out;
1024 			}
1025 			error = tmpfs_rename_relock(fdvp, &fvp, tdvp, &tvp,
1026 			    fcnp, tcnp);
1027 			if (error != 0) {
1028 				vfs_unbusy(mp);
1029 				return (error);
1030 			}
1031 			ASSERT_VOP_ELOCKED(fdvp,
1032 			    "tmpfs_rename: fdvp not locked");
1033 			ASSERT_VOP_ELOCKED(tdvp,
1034 			    "tmpfs_rename: tdvp not locked");
1035 			if (tvp != NULL)
1036 				ASSERT_VOP_ELOCKED(tvp,
1037 				    "tmpfs_rename: tvp not locked");
1038 			if (fvp == tvp) {
1039 				error = 0;
1040 				goto out_locked;
1041 			}
1042 		}
1043 	}
1044 
1045 	tmp = VFS_TO_TMPFS(tdvp->v_mount);
1046 	tdnode = VP_TO_TMPFS_DIR(tdvp);
1047 	tnode = (tvp == NULL) ? NULL : VP_TO_TMPFS_NODE(tvp);
1048 	fdnode = VP_TO_TMPFS_DIR(fdvp);
1049 	fnode = VP_TO_TMPFS_NODE(fvp);
1050 	de = tmpfs_dir_lookup(fdnode, fnode, fcnp);
1051 
1052 	/* Entry can disappear before we lock fdvp,
1053 	 * also avoid manipulating '.' and '..' entries. */
1054 	if (de == NULL) {
1055 		if ((fcnp->cn_flags & ISDOTDOT) != 0 ||
1056 		    (fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.'))
1057 			error = EINVAL;
1058 		else
1059 			error = ENOENT;
1060 		goto out_locked;
1061 	}
1062 	MPASS(de->td_node == fnode);
1063 
1064 	/* If re-naming a directory to another preexisting directory
1065 	 * ensure that the target directory is empty so that its
1066 	 * removal causes no side effects.
1067 	 * Kern_rename gurantees the destination to be a directory
1068 	 * if the source is one. */
1069 	if (tvp != NULL) {
1070 		MPASS(tnode != NULL);
1071 
1072 		if ((tnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
1073 		    (tdnode->tn_flags & (APPEND | IMMUTABLE))) {
1074 			error = EPERM;
1075 			goto out_locked;
1076 		}
1077 
1078 		if (fnode->tn_type == VDIR && tnode->tn_type == VDIR) {
1079 			if (tnode->tn_size > 0) {
1080 				error = ENOTEMPTY;
1081 				goto out_locked;
1082 			}
1083 		} else if (fnode->tn_type == VDIR && tnode->tn_type != VDIR) {
1084 			error = ENOTDIR;
1085 			goto out_locked;
1086 		} else if (fnode->tn_type != VDIR && tnode->tn_type == VDIR) {
1087 			error = EISDIR;
1088 			goto out_locked;
1089 		} else {
1090 			MPASS(fnode->tn_type != VDIR &&
1091 				tnode->tn_type != VDIR);
1092 		}
1093 	}
1094 
1095 	if ((fnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND))
1096 	    || (fdnode->tn_flags & (APPEND | IMMUTABLE))) {
1097 		error = EPERM;
1098 		goto out_locked;
1099 	}
1100 
1101 	/* Ensure that we have enough memory to hold the new name, if it
1102 	 * has to be changed. */
1103 	if (fcnp->cn_namelen != tcnp->cn_namelen ||
1104 	    bcmp(fcnp->cn_nameptr, tcnp->cn_nameptr, fcnp->cn_namelen) != 0) {
1105 		newname = malloc(tcnp->cn_namelen, M_TMPFSNAME, M_WAITOK);
1106 	} else
1107 		newname = NULL;
1108 
1109 	/* If the node is being moved to another directory, we have to do
1110 	 * the move. */
1111 	if (fdnode != tdnode) {
1112 		/* In case we are moving a directory, we have to adjust its
1113 		 * parent to point to the new parent. */
1114 		if (de->td_node->tn_type == VDIR) {
1115 			struct tmpfs_node *n;
1116 
1117 			/* Ensure the target directory is not a child of the
1118 			 * directory being moved.  Otherwise, we'd end up
1119 			 * with stale nodes. */
1120 			n = tdnode;
1121 			/* TMPFS_LOCK garanties that no nodes are freed while
1122 			 * traversing the list. Nodes can only be marked as
1123 			 * removed: tn_parent == NULL. */
1124 			TMPFS_LOCK(tmp);
1125 			TMPFS_NODE_LOCK(n);
1126 			while (n != n->tn_dir.tn_parent) {
1127 				struct tmpfs_node *parent;
1128 
1129 				if (n == fnode) {
1130 					TMPFS_NODE_UNLOCK(n);
1131 					TMPFS_UNLOCK(tmp);
1132 					error = EINVAL;
1133 					if (newname != NULL)
1134 						    free(newname, M_TMPFSNAME);
1135 					goto out_locked;
1136 				}
1137 				parent = n->tn_dir.tn_parent;
1138 				TMPFS_NODE_UNLOCK(n);
1139 				if (parent == NULL) {
1140 					n = NULL;
1141 					break;
1142 				}
1143 				TMPFS_NODE_LOCK(parent);
1144 				if (parent->tn_dir.tn_parent == NULL) {
1145 					TMPFS_NODE_UNLOCK(parent);
1146 					n = NULL;
1147 					break;
1148 				}
1149 				n = parent;
1150 			}
1151 			TMPFS_UNLOCK(tmp);
1152 			if (n == NULL) {
1153 				error = EINVAL;
1154 				if (newname != NULL)
1155 					    free(newname, M_TMPFSNAME);
1156 				goto out_locked;
1157 			}
1158 			TMPFS_NODE_UNLOCK(n);
1159 
1160 			/* Adjust the parent pointer. */
1161 			TMPFS_VALIDATE_DIR(fnode);
1162 			TMPFS_NODE_LOCK(de->td_node);
1163 			de->td_node->tn_dir.tn_parent = tdnode;
1164 			TMPFS_NODE_UNLOCK(de->td_node);
1165 
1166 			/* As a result of changing the target of the '..'
1167 			 * entry, the link count of the source and target
1168 			 * directories has to be adjusted. */
1169 			TMPFS_NODE_LOCK(tdnode);
1170 			TMPFS_ASSERT_LOCKED(tdnode);
1171 			tdnode->tn_links++;
1172 			TMPFS_NODE_UNLOCK(tdnode);
1173 
1174 			TMPFS_NODE_LOCK(fdnode);
1175 			TMPFS_ASSERT_LOCKED(fdnode);
1176 			fdnode->tn_links--;
1177 			TMPFS_NODE_UNLOCK(fdnode);
1178 		}
1179 	}
1180 
1181 	/* Do the move: just remove the entry from the source directory
1182 	 * and insert it into the target one. */
1183 	tmpfs_dir_detach(fdvp, de);
1184 
1185 	if (fcnp->cn_flags & DOWHITEOUT)
1186 		tmpfs_dir_whiteout_add(fdvp, fcnp);
1187 	if (tcnp->cn_flags & ISWHITEOUT)
1188 		tmpfs_dir_whiteout_remove(tdvp, tcnp);
1189 
1190 	/* If the name has changed, we need to make it effective by changing
1191 	 * it in the directory entry. */
1192 	if (newname != NULL) {
1193 		MPASS(tcnp->cn_namelen <= MAXNAMLEN);
1194 
1195 		free(de->ud.td_name, M_TMPFSNAME);
1196 		de->ud.td_name = newname;
1197 		tmpfs_dirent_init(de, tcnp->cn_nameptr, tcnp->cn_namelen);
1198 
1199 		fnode->tn_status |= TMPFS_NODE_CHANGED;
1200 		tdnode->tn_status |= TMPFS_NODE_MODIFIED;
1201 	}
1202 
1203 	/* If we are overwriting an entry, we have to remove the old one
1204 	 * from the target directory. */
1205 	if (tvp != NULL) {
1206 		struct tmpfs_dirent *tde;
1207 
1208 		/* Remove the old entry from the target directory. */
1209 		tde = tmpfs_dir_lookup(tdnode, tnode, tcnp);
1210 		tmpfs_dir_detach(tdvp, tde);
1211 
1212 		/* Free the directory entry we just deleted.  Note that the
1213 		 * node referred by it will not be removed until the vnode is
1214 		 * really reclaimed. */
1215 		tmpfs_free_dirent(VFS_TO_TMPFS(tvp->v_mount), tde);
1216 	}
1217 
1218 	tmpfs_dir_attach(tdvp, de);
1219 
1220 	cache_purge(fvp);
1221 	if (tvp != NULL)
1222 		cache_purge(tvp);
1223 	cache_purge_negative(tdvp);
1224 
1225 	error = 0;
1226 
1227 out_locked:
1228 	if (fdvp != tdvp && fdvp != tvp)
1229 		VOP_UNLOCK(fdvp, 0);
1230 
1231 out:
1232 	/* Release target nodes. */
1233 	/* XXX: I don't understand when tdvp can be the same as tvp, but
1234 	 * other code takes care of this... */
1235 	if (tdvp == tvp)
1236 		vrele(tdvp);
1237 	else
1238 		vput(tdvp);
1239 	if (tvp != NULL)
1240 		vput(tvp);
1241 
1242 	/* Release source nodes. */
1243 	vrele(fdvp);
1244 	vrele(fvp);
1245 
1246 	if (mp != NULL)
1247 		vfs_unbusy(mp);
1248 
1249 	return error;
1250 }
1251 
1252 /* --------------------------------------------------------------------- */
1253 
1254 static int
1255 tmpfs_mkdir(struct vop_mkdir_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 
1262 	MPASS(vap->va_type == VDIR);
1263 
1264 	return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
1265 }
1266 
1267 /* --------------------------------------------------------------------- */
1268 
1269 static int
1270 tmpfs_rmdir(struct vop_rmdir_args *v)
1271 {
1272 	struct vnode *dvp = v->a_dvp;
1273 	struct vnode *vp = v->a_vp;
1274 
1275 	int error;
1276 	struct tmpfs_dirent *de;
1277 	struct tmpfs_mount *tmp;
1278 	struct tmpfs_node *dnode;
1279 	struct tmpfs_node *node;
1280 
1281 	MPASS(VOP_ISLOCKED(dvp));
1282 	MPASS(VOP_ISLOCKED(vp));
1283 
1284 	tmp = VFS_TO_TMPFS(dvp->v_mount);
1285 	dnode = VP_TO_TMPFS_DIR(dvp);
1286 	node = VP_TO_TMPFS_DIR(vp);
1287 
1288 	/* Directories with more than two entries ('.' and '..') cannot be
1289 	 * removed. */
1290 	 if (node->tn_size > 0) {
1291 		 error = ENOTEMPTY;
1292 		 goto out;
1293 	 }
1294 
1295 	if ((dnode->tn_flags & APPEND)
1296 	    || (node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1297 		error = EPERM;
1298 		goto out;
1299 	}
1300 
1301 	/* This invariant holds only if we are not trying to remove "..".
1302 	  * We checked for that above so this is safe now. */
1303 	MPASS(node->tn_dir.tn_parent == dnode);
1304 
1305 	/* Get the directory entry associated with node (vp).  This was
1306 	 * filled by tmpfs_lookup while looking up the entry. */
1307 	de = tmpfs_dir_lookup(dnode, node, v->a_cnp);
1308 	MPASS(TMPFS_DIRENT_MATCHES(de,
1309 	    v->a_cnp->cn_nameptr,
1310 	    v->a_cnp->cn_namelen));
1311 
1312 	/* Check flags to see if we are allowed to remove the directory. */
1313 	if (dnode->tn_flags & APPEND
1314 		|| node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) {
1315 		error = EPERM;
1316 		goto out;
1317 	}
1318 
1319 
1320 	/* Detach the directory entry from the directory (dnode). */
1321 	tmpfs_dir_detach(dvp, de);
1322 	if (v->a_cnp->cn_flags & DOWHITEOUT)
1323 		tmpfs_dir_whiteout_add(dvp, v->a_cnp);
1324 
1325 	/* No vnode should be allocated for this entry from this point */
1326 	TMPFS_NODE_LOCK(node);
1327 	TMPFS_ASSERT_ELOCKED(node);
1328 	node->tn_links--;
1329 	node->tn_dir.tn_parent = NULL;
1330 	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
1331 	    TMPFS_NODE_MODIFIED;
1332 
1333 	TMPFS_NODE_UNLOCK(node);
1334 
1335 	TMPFS_NODE_LOCK(dnode);
1336 	TMPFS_ASSERT_ELOCKED(dnode);
1337 	dnode->tn_links--;
1338 	dnode->tn_status |= TMPFS_NODE_ACCESSED | \
1339 	    TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1340 	TMPFS_NODE_UNLOCK(dnode);
1341 
1342 	cache_purge(dvp);
1343 	cache_purge(vp);
1344 
1345 	/* Free the directory entry we just deleted.  Note that the node
1346 	 * referred by it will not be removed until the vnode is really
1347 	 * reclaimed. */
1348 	tmpfs_free_dirent(tmp, de);
1349 
1350 	/* Release the deleted vnode (will destroy the node, notify
1351 	 * interested parties and clean it from the cache). */
1352 
1353 	dnode->tn_status |= TMPFS_NODE_CHANGED;
1354 	tmpfs_update(dvp);
1355 
1356 	error = 0;
1357 
1358 out:
1359 	return error;
1360 }
1361 
1362 /* --------------------------------------------------------------------- */
1363 
1364 static int
1365 tmpfs_symlink(struct vop_symlink_args *v)
1366 {
1367 	struct vnode *dvp = v->a_dvp;
1368 	struct vnode **vpp = v->a_vpp;
1369 	struct componentname *cnp = v->a_cnp;
1370 	struct vattr *vap = v->a_vap;
1371 	char *target = v->a_target;
1372 
1373 #ifdef notyet /* XXX FreeBSD BUG: kern_symlink is not setting VLNK */
1374 	MPASS(vap->va_type == VLNK);
1375 #else
1376 	vap->va_type = VLNK;
1377 #endif
1378 
1379 	return tmpfs_alloc_file(dvp, vpp, vap, cnp, target);
1380 }
1381 
1382 /* --------------------------------------------------------------------- */
1383 
1384 static int
1385 tmpfs_readdir(struct vop_readdir_args *v)
1386 {
1387 	struct vnode *vp = v->a_vp;
1388 	struct uio *uio = v->a_uio;
1389 	int *eofflag = v->a_eofflag;
1390 	u_long **cookies = v->a_cookies;
1391 	int *ncookies = v->a_ncookies;
1392 
1393 	int error;
1394 	ssize_t startresid;
1395 	int cnt = 0;
1396 	struct tmpfs_node *node;
1397 
1398 	/* This operation only makes sense on directory nodes. */
1399 	if (vp->v_type != VDIR)
1400 		return ENOTDIR;
1401 
1402 	node = VP_TO_TMPFS_DIR(vp);
1403 
1404 	startresid = uio->uio_resid;
1405 
1406 	if (cookies != NULL && ncookies != NULL) {
1407 		cnt = howmany(node->tn_size, sizeof(struct tmpfs_dirent)) + 2;
1408 		*cookies = malloc(cnt * sizeof(**cookies), M_TEMP, M_WAITOK);
1409 		*ncookies = 0;
1410 	}
1411 
1412 	if (cnt == 0)
1413 		error = tmpfs_dir_getdents(node, uio, 0, NULL, NULL);
1414 	else
1415 		error = tmpfs_dir_getdents(node, uio, cnt, *cookies, ncookies);
1416 
1417 	if (error == EJUSTRETURN)
1418 		error = (uio->uio_resid != startresid) ? 0 : EINVAL;
1419 
1420 	if (error != 0 && cnt != 0)
1421 		free(*cookies, M_TEMP);
1422 
1423 	if (eofflag != NULL)
1424 		*eofflag =
1425 		    (error == 0 && uio->uio_offset == TMPFS_DIRCOOKIE_EOF);
1426 
1427 	return error;
1428 }
1429 
1430 /* --------------------------------------------------------------------- */
1431 
1432 static int
1433 tmpfs_readlink(struct vop_readlink_args *v)
1434 {
1435 	struct vnode *vp = v->a_vp;
1436 	struct uio *uio = v->a_uio;
1437 
1438 	int error;
1439 	struct tmpfs_node *node;
1440 
1441 	MPASS(uio->uio_offset == 0);
1442 	MPASS(vp->v_type == VLNK);
1443 
1444 	node = VP_TO_TMPFS_NODE(vp);
1445 
1446 	error = uiomove(node->tn_link, MIN(node->tn_size, uio->uio_resid),
1447 	    uio);
1448 	node->tn_status |= TMPFS_NODE_ACCESSED;
1449 
1450 	return error;
1451 }
1452 
1453 /* --------------------------------------------------------------------- */
1454 
1455 static int
1456 tmpfs_inactive(struct vop_inactive_args *v)
1457 {
1458 	struct vnode *vp = v->a_vp;
1459 
1460 	struct tmpfs_node *node;
1461 
1462 	node = VP_TO_TMPFS_NODE(vp);
1463 
1464 	if (node->tn_links == 0)
1465 		vrecycle(vp);
1466 
1467 	return 0;
1468 }
1469 
1470 /* --------------------------------------------------------------------- */
1471 
1472 int
1473 tmpfs_reclaim(struct vop_reclaim_args *v)
1474 {
1475 	struct vnode *vp = v->a_vp;
1476 
1477 	struct tmpfs_mount *tmp;
1478 	struct tmpfs_node *node;
1479 
1480 	node = VP_TO_TMPFS_NODE(vp);
1481 	tmp = VFS_TO_TMPFS(vp->v_mount);
1482 
1483 	if (vp->v_type == VREG)
1484 		tmpfs_destroy_vobject(vp, node->tn_reg.tn_aobj);
1485 	else
1486 		vnode_destroy_vobject(vp);
1487 	vp->v_object = NULL;
1488 	cache_purge(vp);
1489 
1490 	TMPFS_NODE_LOCK(node);
1491 	TMPFS_ASSERT_ELOCKED(node);
1492 	tmpfs_free_vp(vp);
1493 
1494 	/* If the node referenced by this vnode was deleted by the user,
1495 	 * we must free its associated data structures (now that the vnode
1496 	 * is being reclaimed). */
1497 	if (node->tn_links == 0 &&
1498 	    (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0) {
1499 		node->tn_vpstate = TMPFS_VNODE_DOOMED;
1500 		TMPFS_NODE_UNLOCK(node);
1501 		tmpfs_free_node(tmp, node);
1502 	} else
1503 		TMPFS_NODE_UNLOCK(node);
1504 
1505 	MPASS(vp->v_data == NULL);
1506 	return 0;
1507 }
1508 
1509 /* --------------------------------------------------------------------- */
1510 
1511 static int
1512 tmpfs_print(struct vop_print_args *v)
1513 {
1514 	struct vnode *vp = v->a_vp;
1515 
1516 	struct tmpfs_node *node;
1517 
1518 	node = VP_TO_TMPFS_NODE(vp);
1519 
1520 	printf("tag VT_TMPFS, tmpfs_node %p, flags 0x%lx, links %d\n",
1521 	    node, node->tn_flags, node->tn_links);
1522 	printf("\tmode 0%o, owner %d, group %d, size %jd, status 0x%x\n",
1523 	    node->tn_mode, node->tn_uid, node->tn_gid,
1524 	    (intmax_t)node->tn_size, node->tn_status);
1525 
1526 	if (vp->v_type == VFIFO)
1527 		fifo_printinfo(vp);
1528 
1529 	printf("\n");
1530 
1531 	return 0;
1532 }
1533 
1534 /* --------------------------------------------------------------------- */
1535 
1536 static int
1537 tmpfs_pathconf(struct vop_pathconf_args *v)
1538 {
1539 	int name = v->a_name;
1540 	register_t *retval = v->a_retval;
1541 
1542 	int error;
1543 
1544 	error = 0;
1545 
1546 	switch (name) {
1547 	case _PC_LINK_MAX:
1548 		*retval = LINK_MAX;
1549 		break;
1550 
1551 	case _PC_NAME_MAX:
1552 		*retval = NAME_MAX;
1553 		break;
1554 
1555 	case _PC_PATH_MAX:
1556 		*retval = PATH_MAX;
1557 		break;
1558 
1559 	case _PC_PIPE_BUF:
1560 		*retval = PIPE_BUF;
1561 		break;
1562 
1563 	case _PC_CHOWN_RESTRICTED:
1564 		*retval = 1;
1565 		break;
1566 
1567 	case _PC_NO_TRUNC:
1568 		*retval = 1;
1569 		break;
1570 
1571 	case _PC_SYNC_IO:
1572 		*retval = 1;
1573 		break;
1574 
1575 	case _PC_FILESIZEBITS:
1576 		*retval = 0; /* XXX Don't know which value should I return. */
1577 		break;
1578 
1579 	default:
1580 		error = EINVAL;
1581 	}
1582 
1583 	return error;
1584 }
1585 
1586 static int
1587 tmpfs_vptofh(struct vop_vptofh_args *ap)
1588 {
1589 	struct tmpfs_fid *tfhp;
1590 	struct tmpfs_node *node;
1591 
1592 	tfhp = (struct tmpfs_fid *)ap->a_fhp;
1593 	node = VP_TO_TMPFS_NODE(ap->a_vp);
1594 
1595 	tfhp->tf_len = sizeof(struct tmpfs_fid);
1596 	tfhp->tf_id = node->tn_id;
1597 	tfhp->tf_gen = node->tn_gen;
1598 
1599 	return (0);
1600 }
1601 
1602 static int
1603 tmpfs_whiteout(struct vop_whiteout_args *ap)
1604 {
1605 	struct vnode *dvp = ap->a_dvp;
1606 	struct componentname *cnp = ap->a_cnp;
1607 	struct tmpfs_dirent *de;
1608 
1609 	switch (ap->a_flags) {
1610 	case LOOKUP:
1611 		return (0);
1612 	case CREATE:
1613 		de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(dvp), NULL, cnp);
1614 		if (de != NULL)
1615 			return (de->td_node == NULL ? 0 : EEXIST);
1616 		return (tmpfs_dir_whiteout_add(dvp, cnp));
1617 	case DELETE:
1618 		tmpfs_dir_whiteout_remove(dvp, cnp);
1619 		return (0);
1620 	default:
1621 		panic("tmpfs_whiteout: unknown op");
1622 	}
1623 }
1624 
1625 /* --------------------------------------------------------------------- */
1626 
1627 /*
1628  * vnode operations vector used for files stored in a tmpfs file system.
1629  */
1630 struct vop_vector tmpfs_vnodeop_entries = {
1631 	.vop_default =			&default_vnodeops,
1632 	.vop_lookup =			vfs_cache_lookup,
1633 	.vop_cachedlookup =		tmpfs_lookup,
1634 	.vop_create =			tmpfs_create,
1635 	.vop_mknod =			tmpfs_mknod,
1636 	.vop_open =			tmpfs_open,
1637 	.vop_close =			tmpfs_close,
1638 	.vop_access =			tmpfs_access,
1639 	.vop_getattr =			tmpfs_getattr,
1640 	.vop_setattr =			tmpfs_setattr,
1641 	.vop_read =			tmpfs_read,
1642 	.vop_write =			tmpfs_write,
1643 	.vop_fsync =			tmpfs_fsync,
1644 	.vop_remove =			tmpfs_remove,
1645 	.vop_link =			tmpfs_link,
1646 	.vop_rename =			tmpfs_rename,
1647 	.vop_mkdir =			tmpfs_mkdir,
1648 	.vop_rmdir =			tmpfs_rmdir,
1649 	.vop_symlink =			tmpfs_symlink,
1650 	.vop_readdir =			tmpfs_readdir,
1651 	.vop_readlink =			tmpfs_readlink,
1652 	.vop_inactive =			tmpfs_inactive,
1653 	.vop_reclaim =			tmpfs_reclaim,
1654 	.vop_print =			tmpfs_print,
1655 	.vop_pathconf =			tmpfs_pathconf,
1656 	.vop_vptofh =			tmpfs_vptofh,
1657 	.vop_whiteout =			tmpfs_whiteout,
1658 	.vop_bmap =			VOP_EOPNOTSUPP,
1659 };
1660 
1661