xref: /freebsd/sys/fs/tmpfs/tmpfs_vnops.c (revision b4663a8d111767206bb3ebcfec5b95a6b88bc720)
1 /*	$NetBSD: tmpfs_vnops.c,v 1.39 2007/07/23 15:41:01 jmmv Exp $	*/
2 
3 /*-
4  * SPDX-License-Identifier: BSD-2-Clause
5  *
6  * Copyright (c) 2005, 2006 The NetBSD Foundation, Inc.
7  * All rights reserved.
8  *
9  * This code is derived from software contributed to The NetBSD Foundation
10  * by Julio M. Merino Vidal, developed as part of Google's Summer of Code
11  * 2005 program.
12  *
13  * Redistribution and use in source and binary forms, with or without
14  * modification, are permitted provided that the following conditions
15  * are met:
16  * 1. Redistributions of source code must retain the above copyright
17  *    notice, this list of conditions and the following disclaimer.
18  * 2. Redistributions in binary form must reproduce the above copyright
19  *    notice, this list of conditions and the following disclaimer in the
20  *    documentation and/or other materials provided with the distribution.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
23  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
24  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
25  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
26  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
27  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
28  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
29  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
30  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
31  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
32  * POSSIBILITY OF SUCH DAMAGE.
33  */
34 
35 /*
36  * tmpfs vnode interface.
37  */
38 
39 #include <sys/param.h>
40 #include <sys/systm.h>
41 #include <sys/dirent.h>
42 #include <sys/extattr.h>
43 #include <sys/fcntl.h>
44 #include <sys/file.h>
45 #include <sys/filio.h>
46 #include <sys/limits.h>
47 #include <sys/lockf.h>
48 #include <sys/lock.h>
49 #include <sys/mount.h>
50 #include <sys/namei.h>
51 #include <sys/priv.h>
52 #include <sys/proc.h>
53 #include <sys/rwlock.h>
54 #include <sys/sched.h>
55 #include <sys/smr.h>
56 #include <sys/stat.h>
57 #include <sys/sysctl.h>
58 #include <sys/unistd.h>
59 #include <sys/vnode.h>
60 #include <security/audit/audit.h>
61 #include <security/mac/mac_framework.h>
62 
63 #include <vm/vm.h>
64 #include <vm/vm_param.h>
65 #include <vm/vm_object.h>
66 #include <vm/vm_page.h>
67 #include <vm/vm_pager.h>
68 #include <vm/swap_pager.h>
69 
70 #include <fs/tmpfs/tmpfs_vnops.h>
71 #include <fs/tmpfs/tmpfs.h>
72 
73 SYSCTL_DECL(_vfs_tmpfs);
74 VFS_SMR_DECLARE;
75 
76 static volatile int tmpfs_rename_restarts;
77 SYSCTL_INT(_vfs_tmpfs, OID_AUTO, rename_restarts, CTLFLAG_RD,
78     __DEVOLATILE(int *, &tmpfs_rename_restarts), 0,
79     "Times rename had to restart due to lock contention");
80 
81 MALLOC_DEFINE(M_TMPFSEA, "tmpfs extattr", "tmpfs extattr structure");
82 
83 static int
84 tmpfs_vn_get_ino_alloc(struct mount *mp, void *arg, int lkflags,
85     struct vnode **rvp)
86 {
87 
88 	return (tmpfs_alloc_vp(mp, arg, lkflags, rvp));
89 }
90 
91 static int
92 tmpfs_lookup1(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp)
93 {
94 	struct tmpfs_dirent *de;
95 	struct tmpfs_node *dnode, *pnode;
96 	struct tmpfs_mount *tm;
97 	int error;
98 
99 	/* Caller assumes responsibility for ensuring access (VEXEC). */
100 	dnode = VP_TO_TMPFS_DIR(dvp);
101 	*vpp = NULLVP;
102 
103 	/* We cannot be requesting the parent directory of the root node. */
104 	MPASS(IMPLIES(dnode->tn_type == VDIR &&
105 	    dnode->tn_dir.tn_parent == dnode,
106 	    !(cnp->cn_flags & ISDOTDOT)));
107 
108 	TMPFS_ASSERT_LOCKED(dnode);
109 	if (dnode->tn_dir.tn_parent == NULL) {
110 		error = ENOENT;
111 		goto out;
112 	}
113 	if (cnp->cn_flags & ISDOTDOT) {
114 		tm = VFS_TO_TMPFS(dvp->v_mount);
115 		pnode = dnode->tn_dir.tn_parent;
116 		tmpfs_ref_node(pnode);
117 		error = vn_vget_ino_gen(dvp, tmpfs_vn_get_ino_alloc,
118 		    pnode, cnp->cn_lkflags, vpp);
119 		tmpfs_free_node(tm, pnode);
120 		if (error != 0)
121 			goto out;
122 	} else if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
123 		VREF(dvp);
124 		*vpp = dvp;
125 		error = 0;
126 	} else {
127 		de = tmpfs_dir_lookup(dnode, NULL, cnp);
128 		if (de != NULL && de->td_node == NULL)
129 			cnp->cn_flags |= ISWHITEOUT;
130 		if (de == NULL || de->td_node == NULL) {
131 			/*
132 			 * The entry was not found in the directory.
133 			 * This is OK if we are creating or renaming an
134 			 * entry and are working on the last component of
135 			 * the path name.
136 			 */
137 			if ((cnp->cn_flags & ISLASTCN) &&
138 			    (cnp->cn_nameiop == CREATE || \
139 			    cnp->cn_nameiop == RENAME ||
140 			    (cnp->cn_nameiop == DELETE &&
141 			    cnp->cn_flags & DOWHITEOUT &&
142 			    cnp->cn_flags & ISWHITEOUT))) {
143 				error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred,
144 				    curthread);
145 				if (error != 0)
146 					goto out;
147 
148 				error = EJUSTRETURN;
149 			} else
150 				error = ENOENT;
151 		} else {
152 			struct tmpfs_node *tnode;
153 
154 			/*
155 			 * The entry was found, so get its associated
156 			 * tmpfs_node.
157 			 */
158 			tnode = de->td_node;
159 
160 			/*
161 			 * If we are not at the last path component and
162 			 * found a non-directory or non-link entry (which
163 			 * may itself be pointing to a directory), raise
164 			 * an error.
165 			 */
166 			if ((tnode->tn_type != VDIR &&
167 			    tnode->tn_type != VLNK) &&
168 			    !(cnp->cn_flags & ISLASTCN)) {
169 				error = ENOTDIR;
170 				goto out;
171 			}
172 
173 			/*
174 			 * If we are deleting or renaming the entry, keep
175 			 * track of its tmpfs_dirent so that it can be
176 			 * easily deleted later.
177 			 */
178 			if ((cnp->cn_flags & ISLASTCN) &&
179 			    (cnp->cn_nameiop == DELETE ||
180 			    cnp->cn_nameiop == RENAME)) {
181 				error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred,
182 				    curthread);
183 				if (error != 0)
184 					goto out;
185 
186 				/* Allocate a new vnode on the matching entry. */
187 				error = tmpfs_alloc_vp(dvp->v_mount, tnode,
188 				    cnp->cn_lkflags, vpp);
189 				if (error != 0)
190 					goto out;
191 
192 				if ((dnode->tn_mode & S_ISTXT) &&
193 				  VOP_ACCESS(dvp, VADMIN, cnp->cn_cred,
194 				  curthread) && VOP_ACCESS(*vpp, VADMIN,
195 				  cnp->cn_cred, curthread)) {
196 					error = EPERM;
197 					vput(*vpp);
198 					*vpp = NULL;
199 					goto out;
200 				}
201 			} else {
202 				error = tmpfs_alloc_vp(dvp->v_mount, tnode,
203 				    cnp->cn_lkflags, vpp);
204 				if (error != 0)
205 					goto out;
206 			}
207 		}
208 	}
209 
210 	/*
211 	 * Store the result of this lookup in the cache.  Avoid this if the
212 	 * request was for creation, as it does not improve timings on
213 	 * emprical tests.
214 	 */
215 	if ((cnp->cn_flags & MAKEENTRY) != 0 && tmpfs_use_nc(dvp))
216 		cache_enter(dvp, *vpp, cnp);
217 
218 out:
219 #ifdef INVARIANTS
220 	/*
221 	 * If there were no errors, *vpp cannot be null and it must be
222 	 * locked.
223 	 */
224 	if (error == 0) {
225 		MPASS(*vpp != NULLVP);
226 		ASSERT_VOP_LOCKED(*vpp, __func__);
227 	} else {
228 		MPASS(*vpp == NULL);
229 	}
230 #endif
231 
232 	return (error);
233 }
234 
235 static int
236 tmpfs_cached_lookup(struct vop_cachedlookup_args *v)
237 {
238 
239 	return (tmpfs_lookup1(v->a_dvp, v->a_vpp, v->a_cnp));
240 }
241 
242 static int
243 tmpfs_lookup(struct vop_lookup_args *v)
244 {
245 	struct vnode *dvp = v->a_dvp;
246 	struct vnode **vpp = v->a_vpp;
247 	struct componentname *cnp = v->a_cnp;
248 	int error;
249 
250 	/* Check accessibility of requested node as a first step. */
251 	error = vn_dir_check_exec(dvp, cnp);
252 	if (error != 0)
253 		return (error);
254 
255 	return (tmpfs_lookup1(dvp, vpp, cnp));
256 }
257 
258 static int
259 tmpfs_create(struct vop_create_args *v)
260 {
261 	struct vnode *dvp = v->a_dvp;
262 	struct vnode **vpp = v->a_vpp;
263 	struct componentname *cnp = v->a_cnp;
264 	struct vattr *vap = v->a_vap;
265 	int error;
266 
267 	MPASS(vap->va_type == VREG || vap->va_type == VSOCK);
268 
269 	error = tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
270 	if (error == 0 && (cnp->cn_flags & MAKEENTRY) != 0 && tmpfs_use_nc(dvp))
271 		cache_enter(dvp, *vpp, cnp);
272 	return (error);
273 }
274 
275 static int
276 tmpfs_mknod(struct vop_mknod_args *v)
277 {
278 	struct vnode *dvp = v->a_dvp;
279 	struct vnode **vpp = v->a_vpp;
280 	struct componentname *cnp = v->a_cnp;
281 	struct vattr *vap = v->a_vap;
282 
283 	if (vap->va_type != VBLK && vap->va_type != VCHR &&
284 	    vap->va_type != VFIFO)
285 		return (EINVAL);
286 
287 	return (tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL));
288 }
289 
290 struct fileops tmpfs_fnops;
291 
292 static int
293 tmpfs_open(struct vop_open_args *v)
294 {
295 	struct vnode *vp;
296 	struct tmpfs_node *node;
297 	struct file *fp;
298 	int error, mode;
299 
300 	vp = v->a_vp;
301 	mode = v->a_mode;
302 	node = VP_TO_TMPFS_NODE(vp);
303 
304 	/*
305 	 * The file is still active but all its names have been removed
306 	 * (e.g. by a "rmdir $(pwd)").  It cannot be opened any more as
307 	 * it is about to die.
308 	 */
309 	if (node->tn_links < 1)
310 		return (ENOENT);
311 
312 	/* If the file is marked append-only, deny write requests. */
313 	if (node->tn_flags & APPEND && (mode & (FWRITE | O_APPEND)) == FWRITE)
314 		error = EPERM;
315 	else {
316 		error = 0;
317 		/* For regular files, the call below is nop. */
318 		KASSERT(vp->v_type != VREG || (node->tn_reg.tn_aobj->flags &
319 		    OBJ_DEAD) == 0, ("dead object"));
320 		vnode_create_vobject(vp, node->tn_size, v->a_td);
321 	}
322 
323 	fp = v->a_fp;
324 	MPASS(fp == NULL || fp->f_data == NULL);
325 	if (error == 0 && fp != NULL && vp->v_type == VREG) {
326 		tmpfs_ref_node(node);
327 		finit_vnode(fp, mode, node, &tmpfs_fnops);
328 	}
329 
330 	return (error);
331 }
332 
333 static int
334 tmpfs_close(struct vop_close_args *v)
335 {
336 	struct vnode *vp = v->a_vp;
337 
338 	/* Update node times. */
339 	tmpfs_update(vp);
340 
341 	return (0);
342 }
343 
344 int
345 tmpfs_fo_close(struct file *fp, struct thread *td)
346 {
347 	struct tmpfs_node *node;
348 
349 	node = fp->f_data;
350 	if (node != NULL) {
351 		MPASS(node->tn_type == VREG);
352 		tmpfs_free_node(node->tn_reg.tn_tmp, node);
353 	}
354 	return (vnops.fo_close(fp, td));
355 }
356 
357 /*
358  * VOP_FPLOOKUP_VEXEC routines are subject to special circumstances, see
359  * the comment above cache_fplookup for details.
360  */
361 int
362 tmpfs_fplookup_vexec(struct vop_fplookup_vexec_args *v)
363 {
364 	struct vnode *vp;
365 	struct tmpfs_node *node;
366 	struct ucred *cred;
367 	mode_t all_x, mode;
368 
369 	vp = v->a_vp;
370 	node = VP_TO_TMPFS_NODE_SMR(vp);
371 	if (__predict_false(node == NULL))
372 		return (EAGAIN);
373 
374 	all_x = S_IXUSR | S_IXGRP | S_IXOTH;
375 	mode = atomic_load_short(&node->tn_mode);
376 	if (__predict_true((mode & all_x) == all_x))
377 		return (0);
378 
379 	cred = v->a_cred;
380 	return (vaccess_vexec_smr(mode, node->tn_uid, node->tn_gid, cred));
381 }
382 
383 static int
384 tmpfs_access_locked(struct vnode *vp, struct tmpfs_node *node,
385     accmode_t accmode, struct ucred *cred)
386 {
387 #ifdef DEBUG_VFS_LOCKS
388 	if (!mtx_owned(TMPFS_NODE_MTX(node))) {
389 		ASSERT_VOP_LOCKED(vp,
390 		    "tmpfs_access_locked needs locked vnode or node");
391 	}
392 #endif
393 
394 	if ((accmode & VWRITE) != 0 && (node->tn_flags & IMMUTABLE) != 0)
395 		return (EPERM);
396 	return (vaccess(vp->v_type, node->tn_mode, node->tn_uid, node->tn_gid,
397 	    accmode, cred));
398 }
399 
400 int
401 tmpfs_access(struct vop_access_args *v)
402 {
403 	struct vnode *vp = v->a_vp;
404 	struct ucred *cred = v->a_cred;
405 	struct tmpfs_node *node = VP_TO_TMPFS_NODE(vp);
406 	mode_t all_x = S_IXUSR | S_IXGRP | S_IXOTH;
407 	accmode_t accmode = v->a_accmode;
408 
409 	/*
410 	 * Common case path lookup.
411 	 */
412 	if (__predict_true(accmode == VEXEC &&
413 	    (node->tn_mode & all_x) == all_x))
414 		return (0);
415 
416 	switch (vp->v_type) {
417 	case VDIR:
418 		/* FALLTHROUGH */
419 	case VLNK:
420 		/* FALLTHROUGH */
421 	case VREG:
422 		if ((accmode & VWRITE) != 0 &&
423 		    (vp->v_mount->mnt_flag & MNT_RDONLY) != 0)
424 			return (EROFS);
425 		break;
426 
427 	case VBLK:
428 		/* FALLTHROUGH */
429 	case VCHR:
430 		/* FALLTHROUGH */
431 	case VSOCK:
432 		/* FALLTHROUGH */
433 	case VFIFO:
434 		break;
435 
436 	default:
437 		return (EINVAL);
438 	}
439 
440 	return (tmpfs_access_locked(vp, node, accmode, cred));
441 }
442 
443 int
444 tmpfs_stat(struct vop_stat_args *v)
445 {
446 	struct vnode *vp = v->a_vp;
447 	struct stat *sb = v->a_sb;
448 	struct tmpfs_node *node;
449 	int error;
450 
451 	node = VP_TO_TMPFS_NODE(vp);
452 
453 	tmpfs_update_getattr(vp);
454 
455 	error = vop_stat_helper_pre(v);
456 	if (__predict_false(error))
457 		return (error);
458 
459 	sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0];
460 	sb->st_ino = node->tn_id;
461 	sb->st_mode = node->tn_mode | VTTOIF(vp->v_type);
462 	sb->st_nlink = node->tn_links;
463 	sb->st_uid = node->tn_uid;
464 	sb->st_gid = node->tn_gid;
465 	sb->st_rdev = (vp->v_type == VBLK || vp->v_type == VCHR) ?
466 		node->tn_rdev : NODEV;
467 	sb->st_size = node->tn_size;
468 	sb->st_atim.tv_sec = node->tn_atime.tv_sec;
469 	sb->st_atim.tv_nsec = node->tn_atime.tv_nsec;
470 	sb->st_mtim.tv_sec = node->tn_mtime.tv_sec;
471 	sb->st_mtim.tv_nsec = node->tn_mtime.tv_nsec;
472 	sb->st_ctim.tv_sec = node->tn_ctime.tv_sec;
473 	sb->st_ctim.tv_nsec = node->tn_ctime.tv_nsec;
474 	sb->st_birthtim.tv_sec = node->tn_birthtime.tv_sec;
475 	sb->st_birthtim.tv_nsec = node->tn_birthtime.tv_nsec;
476 	sb->st_blksize = PAGE_SIZE;
477 	sb->st_flags = node->tn_flags;
478 	sb->st_gen = node->tn_gen;
479 	sb->st_filerev = 0;
480 	if (vp->v_type == VREG) {
481 #ifdef __ILP32__
482 		vm_object_t obj = node->tn_reg.tn_aobj;
483 
484 		/* Handle torn read */
485 		VM_OBJECT_RLOCK(obj);
486 #endif
487 		sb->st_blocks = ptoa(node->tn_reg.tn_pages);
488 #ifdef __ILP32__
489 		VM_OBJECT_RUNLOCK(obj);
490 #endif
491 	} else {
492 		sb->st_blocks = node->tn_size;
493 	}
494 	sb->st_blocks /= S_BLKSIZE;
495 	return (vop_stat_helper_post(v, error));
496 }
497 
498 int
499 tmpfs_getattr(struct vop_getattr_args *v)
500 {
501 	struct vnode *vp = v->a_vp;
502 	struct vattr *vap = v->a_vap;
503 	struct tmpfs_node *node;
504 
505 	node = VP_TO_TMPFS_NODE(vp);
506 
507 	tmpfs_update_getattr(vp);
508 
509 	vap->va_type = vp->v_type;
510 	vap->va_mode = node->tn_mode;
511 	vap->va_nlink = node->tn_links;
512 	vap->va_uid = node->tn_uid;
513 	vap->va_gid = node->tn_gid;
514 	vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
515 	vap->va_fileid = node->tn_id;
516 	vap->va_size = node->tn_size;
517 	vap->va_blocksize = PAGE_SIZE;
518 	vap->va_atime = node->tn_atime;
519 	vap->va_mtime = node->tn_mtime;
520 	vap->va_ctime = node->tn_ctime;
521 	vap->va_birthtime = node->tn_birthtime;
522 	vap->va_gen = node->tn_gen;
523 	vap->va_flags = node->tn_flags;
524 	vap->va_rdev = (vp->v_type == VBLK || vp->v_type == VCHR) ?
525 	    node->tn_rdev : NODEV;
526 	if (vp->v_type == VREG) {
527 #ifdef __ILP32__
528 		vm_object_t obj = node->tn_reg.tn_aobj;
529 
530 		VM_OBJECT_RLOCK(obj);
531 #endif
532 		vap->va_bytes = ptoa(node->tn_reg.tn_pages);
533 #ifdef __ILP32__
534 		VM_OBJECT_RUNLOCK(obj);
535 #endif
536 	} else {
537 		vap->va_bytes = node->tn_size;
538 	}
539 	vap->va_filerev = 0;
540 
541 	return (0);
542 }
543 
544 int
545 tmpfs_setattr(struct vop_setattr_args *v)
546 {
547 	struct vnode *vp = v->a_vp;
548 	struct vattr *vap = v->a_vap;
549 	struct ucred *cred = v->a_cred;
550 	struct thread *td = curthread;
551 
552 	int error;
553 
554 	ASSERT_VOP_IN_SEQC(vp);
555 
556 	error = 0;
557 
558 	/* Abort if any unsettable attribute is given. */
559 	if (vap->va_type != VNON ||
560 	    vap->va_nlink != VNOVAL ||
561 	    vap->va_fsid != VNOVAL ||
562 	    vap->va_fileid != VNOVAL ||
563 	    vap->va_blocksize != VNOVAL ||
564 	    vap->va_gen != VNOVAL ||
565 	    vap->va_rdev != VNOVAL ||
566 	    vap->va_bytes != VNOVAL)
567 		error = EINVAL;
568 
569 	if (error == 0 && (vap->va_flags != VNOVAL))
570 		error = tmpfs_chflags(vp, vap->va_flags, cred, td);
571 
572 	if (error == 0 && (vap->va_size != VNOVAL))
573 		error = tmpfs_chsize(vp, vap->va_size, cred, td);
574 
575 	if (error == 0 && (vap->va_uid != VNOVAL || vap->va_gid != VNOVAL))
576 		error = tmpfs_chown(vp, vap->va_uid, vap->va_gid, cred, td);
577 
578 	if (error == 0 && (vap->va_mode != (mode_t)VNOVAL))
579 		error = tmpfs_chmod(vp, vap->va_mode, cred, td);
580 
581 	if (error == 0 && ((vap->va_atime.tv_sec != VNOVAL &&
582 	    vap->va_atime.tv_nsec != VNOVAL) ||
583 	    (vap->va_mtime.tv_sec != VNOVAL &&
584 	    vap->va_mtime.tv_nsec != VNOVAL) ||
585 	    (vap->va_birthtime.tv_sec != VNOVAL &&
586 	    vap->va_birthtime.tv_nsec != VNOVAL)))
587 		error = tmpfs_chtimes(vp, vap, cred, td);
588 
589 	/*
590 	 * Update the node times.  We give preference to the error codes
591 	 * generated by this function rather than the ones that may arise
592 	 * from tmpfs_update.
593 	 */
594 	tmpfs_update(vp);
595 
596 	return (error);
597 }
598 
599 static int
600 tmpfs_read(struct vop_read_args *v)
601 {
602 	struct vnode *vp;
603 	struct uio *uio;
604 	struct tmpfs_node *node;
605 
606 	vp = v->a_vp;
607 	if (vp->v_type != VREG)
608 		return (EISDIR);
609 	uio = v->a_uio;
610 	if (uio->uio_offset < 0)
611 		return (EINVAL);
612 	node = VP_TO_TMPFS_NODE(vp);
613 	tmpfs_set_accessed(VFS_TO_TMPFS(vp->v_mount), node);
614 	return (uiomove_object(node->tn_reg.tn_aobj, node->tn_size, uio));
615 }
616 
617 static int
618 tmpfs_read_pgcache(struct vop_read_pgcache_args *v)
619 {
620 	struct vnode *vp;
621 	struct tmpfs_node *node;
622 	vm_object_t object;
623 	off_t size;
624 	int error;
625 
626 	vp = v->a_vp;
627 	VNPASS((vn_irflag_read(vp) & VIRF_PGREAD) != 0, vp);
628 
629 	if (v->a_uio->uio_offset < 0)
630 		return (EINVAL);
631 
632 	error = EJUSTRETURN;
633 	vfs_smr_enter();
634 
635 	node = VP_TO_TMPFS_NODE_SMR(vp);
636 	if (node == NULL)
637 		goto out_smr;
638 	MPASS(node->tn_type == VREG);
639 	MPASS(node->tn_refcount >= 1);
640 	object = node->tn_reg.tn_aobj;
641 	if (object == NULL)
642 		goto out_smr;
643 
644 	MPASS(object->type == tmpfs_pager_type);
645 	MPASS((object->flags & (OBJ_ANON | OBJ_DEAD | OBJ_SWAP)) ==
646 	    OBJ_SWAP);
647 	if (!VN_IS_DOOMED(vp)) {
648 		/* size cannot become shorter due to rangelock. */
649 		size = node->tn_size;
650 		tmpfs_set_accessed(node->tn_reg.tn_tmp, node);
651 		vfs_smr_exit();
652 		error = uiomove_object(object, size, v->a_uio);
653 		return (error);
654 	}
655 out_smr:
656 	vfs_smr_exit();
657 	return (error);
658 }
659 
660 static int
661 tmpfs_write(struct vop_write_args *v)
662 {
663 	struct vnode *vp;
664 	struct uio *uio;
665 	struct tmpfs_node *node;
666 	off_t oldsize;
667 	ssize_t r;
668 	int error, ioflag;
669 	mode_t newmode;
670 
671 	vp = v->a_vp;
672 	uio = v->a_uio;
673 	ioflag = v->a_ioflag;
674 	error = 0;
675 	node = VP_TO_TMPFS_NODE(vp);
676 	oldsize = node->tn_size;
677 
678 	if (uio->uio_offset < 0 || vp->v_type != VREG)
679 		return (EINVAL);
680 	if (uio->uio_resid == 0)
681 		return (0);
682 	if (ioflag & IO_APPEND)
683 		uio->uio_offset = node->tn_size;
684 	error = vn_rlimit_fsizex(vp, uio, VFS_TO_TMPFS(vp->v_mount)->
685 	    tm_maxfilesize, &r, uio->uio_td);
686 	if (error != 0) {
687 		vn_rlimit_fsizex_res(uio, r);
688 		return (error);
689 	}
690 
691 	if (uio->uio_offset + uio->uio_resid > node->tn_size) {
692 		error = tmpfs_reg_resize(vp, uio->uio_offset + uio->uio_resid,
693 		    FALSE);
694 		if (error != 0)
695 			goto out;
696 	}
697 
698 	error = uiomove_object(node->tn_reg.tn_aobj, node->tn_size, uio);
699 	node->tn_status |= TMPFS_NODE_MODIFIED | TMPFS_NODE_CHANGED;
700 	node->tn_accessed = true;
701 	if (node->tn_mode & (S_ISUID | S_ISGID)) {
702 		if (priv_check_cred(v->a_cred, PRIV_VFS_RETAINSUGID)) {
703 			newmode = node->tn_mode & ~(S_ISUID | S_ISGID);
704 			vn_seqc_write_begin(vp);
705 			atomic_store_short(&node->tn_mode, newmode);
706 			vn_seqc_write_end(vp);
707 		}
708 	}
709 	if (error != 0)
710 		(void)tmpfs_reg_resize(vp, oldsize, TRUE);
711 
712 out:
713 	MPASS(IMPLIES(error == 0, uio->uio_resid == 0));
714 	MPASS(IMPLIES(error != 0, oldsize == node->tn_size));
715 
716 	vn_rlimit_fsizex_res(uio, r);
717 	return (error);
718 }
719 
720 static int
721 tmpfs_deallocate(struct vop_deallocate_args *v)
722 {
723 	return (tmpfs_reg_punch_hole(v->a_vp, v->a_offset, v->a_len));
724 }
725 
726 static int
727 tmpfs_fsync(struct vop_fsync_args *v)
728 {
729 	struct vnode *vp = v->a_vp;
730 
731 	tmpfs_check_mtime(vp);
732 	tmpfs_update(vp);
733 
734 	return (0);
735 }
736 
737 static int
738 tmpfs_remove(struct vop_remove_args *v)
739 {
740 	struct vnode *dvp = v->a_dvp;
741 	struct vnode *vp = v->a_vp;
742 
743 	int error;
744 	struct tmpfs_dirent *de;
745 	struct tmpfs_mount *tmp;
746 	struct tmpfs_node *dnode;
747 	struct tmpfs_node *node;
748 
749 	if (vp->v_type == VDIR) {
750 		error = EISDIR;
751 		goto out;
752 	}
753 
754 	dnode = VP_TO_TMPFS_DIR(dvp);
755 	node = VP_TO_TMPFS_NODE(vp);
756 	tmp = VFS_TO_TMPFS(vp->v_mount);
757 	de = tmpfs_dir_lookup(dnode, node, v->a_cnp);
758 	MPASS(de != NULL);
759 
760 	/* Files marked as immutable or append-only cannot be deleted. */
761 	if ((node->tn_flags & (IMMUTABLE | APPEND | NOUNLINK)) ||
762 	    (dnode->tn_flags & APPEND)) {
763 		error = EPERM;
764 		goto out;
765 	}
766 
767 	/* Remove the entry from the directory; as it is a file, we do not
768 	 * have to change the number of hard links of the directory. */
769 	tmpfs_dir_detach(dvp, de);
770 	if (v->a_cnp->cn_flags & DOWHITEOUT)
771 		tmpfs_dir_whiteout_add(dvp, v->a_cnp);
772 
773 	/* Free the directory entry we just deleted.  Note that the node
774 	 * referred by it will not be removed until the vnode is really
775 	 * reclaimed. */
776 	tmpfs_free_dirent(tmp, de);
777 
778 	node->tn_status |= TMPFS_NODE_CHANGED;
779 	node->tn_accessed = true;
780 	error = 0;
781 
782 out:
783 	return (error);
784 }
785 
786 static int
787 tmpfs_link(struct vop_link_args *v)
788 {
789 	struct vnode *dvp = v->a_tdvp;
790 	struct vnode *vp = v->a_vp;
791 	struct componentname *cnp = v->a_cnp;
792 
793 	int error;
794 	struct tmpfs_dirent *de;
795 	struct tmpfs_node *node;
796 
797 	MPASS(dvp != vp); /* XXX When can this be false? */
798 	node = VP_TO_TMPFS_NODE(vp);
799 
800 	/* Ensure that we do not overflow the maximum number of links imposed
801 	 * by the system. */
802 	MPASS(node->tn_links <= TMPFS_LINK_MAX);
803 	if (node->tn_links == TMPFS_LINK_MAX) {
804 		error = EMLINK;
805 		goto out;
806 	}
807 
808 	/* We cannot create links of files marked immutable or append-only. */
809 	if (node->tn_flags & (IMMUTABLE | APPEND)) {
810 		error = EPERM;
811 		goto out;
812 	}
813 
814 	/* Allocate a new directory entry to represent the node. */
815 	error = tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), node,
816 	    cnp->cn_nameptr, cnp->cn_namelen, &de);
817 	if (error != 0)
818 		goto out;
819 
820 	/* Insert the new directory entry into the appropriate directory. */
821 	if (cnp->cn_flags & ISWHITEOUT)
822 		tmpfs_dir_whiteout_remove(dvp, cnp);
823 	tmpfs_dir_attach(dvp, de);
824 
825 	/* vp link count has changed, so update node times. */
826 	node->tn_status |= TMPFS_NODE_CHANGED;
827 	tmpfs_update(vp);
828 
829 	error = 0;
830 
831 out:
832 	return (error);
833 }
834 
835 /*
836  * We acquire all but fdvp locks using non-blocking acquisitions.  If we
837  * fail to acquire any lock in the path we will drop all held locks,
838  * acquire the new lock in a blocking fashion, and then release it and
839  * restart the rename.  This acquire/release step ensures that we do not
840  * spin on a lock waiting for release.  On error release all vnode locks
841  * and decrement references the way tmpfs_rename() would do.
842  */
843 static int
844 tmpfs_rename_relock(struct vnode *fdvp, struct vnode **fvpp,
845     struct vnode *tdvp, struct vnode **tvpp,
846     struct componentname *fcnp, struct componentname *tcnp)
847 {
848 	struct vnode *nvp;
849 	struct mount *mp;
850 	struct tmpfs_dirent *de;
851 	int error, restarts = 0;
852 
853 	VOP_UNLOCK(tdvp);
854 	if (*tvpp != NULL && *tvpp != tdvp)
855 		VOP_UNLOCK(*tvpp);
856 	mp = fdvp->v_mount;
857 
858 relock:
859 	restarts += 1;
860 	error = vn_lock(fdvp, LK_EXCLUSIVE);
861 	if (error)
862 		goto releout;
863 	if (vn_lock(tdvp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
864 		VOP_UNLOCK(fdvp);
865 		error = vn_lock(tdvp, LK_EXCLUSIVE);
866 		if (error)
867 			goto releout;
868 		VOP_UNLOCK(tdvp);
869 		goto relock;
870 	}
871 	/*
872 	 * Re-resolve fvp to be certain it still exists and fetch the
873 	 * correct vnode.
874 	 */
875 	de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(fdvp), NULL, fcnp);
876 	if (de == NULL) {
877 		VOP_UNLOCK(fdvp);
878 		VOP_UNLOCK(tdvp);
879 		if ((fcnp->cn_flags & ISDOTDOT) != 0 ||
880 		    (fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.'))
881 			error = EINVAL;
882 		else
883 			error = ENOENT;
884 		goto releout;
885 	}
886 	error = tmpfs_alloc_vp(mp, de->td_node, LK_EXCLUSIVE | LK_NOWAIT, &nvp);
887 	if (error != 0) {
888 		VOP_UNLOCK(fdvp);
889 		VOP_UNLOCK(tdvp);
890 		if (error != EBUSY)
891 			goto releout;
892 		error = tmpfs_alloc_vp(mp, de->td_node, LK_EXCLUSIVE, &nvp);
893 		if (error != 0)
894 			goto releout;
895 		VOP_UNLOCK(nvp);
896 		/*
897 		 * Concurrent rename race.
898 		 */
899 		if (nvp == tdvp) {
900 			vrele(nvp);
901 			error = EINVAL;
902 			goto releout;
903 		}
904 		vrele(*fvpp);
905 		*fvpp = nvp;
906 		goto relock;
907 	}
908 	vrele(*fvpp);
909 	*fvpp = nvp;
910 	VOP_UNLOCK(*fvpp);
911 	/*
912 	 * Re-resolve tvp and acquire the vnode lock if present.
913 	 */
914 	de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(tdvp), NULL, tcnp);
915 	/*
916 	 * If tvp disappeared we just carry on.
917 	 */
918 	if (de == NULL && *tvpp != NULL) {
919 		vrele(*tvpp);
920 		*tvpp = NULL;
921 	}
922 	/*
923 	 * Get the tvp ino if the lookup succeeded.  We may have to restart
924 	 * if the non-blocking acquire fails.
925 	 */
926 	if (de != NULL) {
927 		nvp = NULL;
928 		error = tmpfs_alloc_vp(mp, de->td_node,
929 		    LK_EXCLUSIVE | LK_NOWAIT, &nvp);
930 		if (*tvpp != NULL)
931 			vrele(*tvpp);
932 		*tvpp = nvp;
933 		if (error != 0) {
934 			VOP_UNLOCK(fdvp);
935 			VOP_UNLOCK(tdvp);
936 			if (error != EBUSY)
937 				goto releout;
938 			error = tmpfs_alloc_vp(mp, de->td_node, LK_EXCLUSIVE,
939 			    &nvp);
940 			if (error != 0)
941 				goto releout;
942 			VOP_UNLOCK(nvp);
943 			/*
944 			 * fdvp contains fvp, thus tvp (=fdvp) is not empty.
945 			 */
946 			if (nvp == fdvp) {
947 				error = ENOTEMPTY;
948 				goto releout;
949 			}
950 			goto relock;
951 		}
952 	}
953 	tmpfs_rename_restarts += restarts;
954 
955 	return (0);
956 
957 releout:
958 	vrele(fdvp);
959 	vrele(*fvpp);
960 	vrele(tdvp);
961 	if (*tvpp != NULL)
962 		vrele(*tvpp);
963 	tmpfs_rename_restarts += restarts;
964 
965 	return (error);
966 }
967 
968 static int
969 tmpfs_rename(struct vop_rename_args *v)
970 {
971 	struct vnode *fdvp = v->a_fdvp;
972 	struct vnode *fvp = v->a_fvp;
973 	struct componentname *fcnp = v->a_fcnp;
974 	struct vnode *tdvp = v->a_tdvp;
975 	struct vnode *tvp = v->a_tvp;
976 	struct componentname *tcnp = v->a_tcnp;
977 	char *newname;
978 	struct tmpfs_dirent *de;
979 	struct tmpfs_mount *tmp;
980 	struct tmpfs_node *fdnode;
981 	struct tmpfs_node *fnode;
982 	struct tmpfs_node *tnode;
983 	struct tmpfs_node *tdnode;
984 	int error;
985 	bool want_seqc_end;
986 
987 	want_seqc_end = false;
988 
989 	/*
990 	 * Disallow cross-device renames.
991 	 * XXX Why isn't this done by the caller?
992 	 */
993 	if (fvp->v_mount != tdvp->v_mount ||
994 	    (tvp != NULL && fvp->v_mount != tvp->v_mount)) {
995 		error = EXDEV;
996 		goto out;
997 	}
998 
999 	/* If source and target are the same file, there is nothing to do. */
1000 	if (fvp == tvp) {
1001 		error = 0;
1002 		goto out;
1003 	}
1004 
1005 	/*
1006 	 * If we need to move the directory between entries, lock the
1007 	 * source so that we can safely operate on it.
1008 	 */
1009 	if (fdvp != tdvp && fdvp != tvp) {
1010 		if (vn_lock(fdvp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
1011 			error = tmpfs_rename_relock(fdvp, &fvp, tdvp, &tvp,
1012 			    fcnp, tcnp);
1013 			if (error != 0)
1014 				return (error);
1015 			ASSERT_VOP_ELOCKED(fdvp,
1016 			    "tmpfs_rename: fdvp not locked");
1017 			ASSERT_VOP_ELOCKED(tdvp,
1018 			    "tmpfs_rename: tdvp not locked");
1019 			if (tvp != NULL)
1020 				ASSERT_VOP_ELOCKED(tvp,
1021 				    "tmpfs_rename: tvp not locked");
1022 			if (fvp == tvp) {
1023 				error = 0;
1024 				goto out_locked;
1025 			}
1026 		}
1027 	}
1028 
1029 	/*
1030 	 * Avoid manipulating '.' and '..' entries.
1031 	 */
1032 	if ((fcnp->cn_flags & ISDOTDOT) != 0 ||
1033 	    (fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.')) {
1034 		error = EINVAL;
1035 		goto out_locked;
1036 	}
1037 
1038 	if (tvp != NULL)
1039 		vn_seqc_write_begin(tvp);
1040 	vn_seqc_write_begin(tdvp);
1041 	vn_seqc_write_begin(fvp);
1042 	vn_seqc_write_begin(fdvp);
1043 	want_seqc_end = true;
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 	/*
1053 	 * Entry can disappear before we lock fdvp.
1054 	 */
1055 	if (de == NULL) {
1056 		if ((fcnp->cn_flags & ISDOTDOT) != 0 ||
1057 		    (fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.'))
1058 			error = EINVAL;
1059 		else
1060 			error = ENOENT;
1061 		goto out_locked;
1062 	}
1063 	MPASS(de->td_node == fnode);
1064 
1065 	/*
1066 	 * If re-naming a directory to another preexisting directory
1067 	 * ensure that the target directory is empty so that its
1068 	 * removal causes no side effects.
1069 	 * Kern_rename guarantees the destination to be a directory
1070 	 * if the source is one.
1071 	 */
1072 	if (tvp != NULL) {
1073 		MPASS(tnode != NULL);
1074 
1075 		if ((tnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
1076 		    (tdnode->tn_flags & (APPEND | IMMUTABLE))) {
1077 			error = EPERM;
1078 			goto out_locked;
1079 		}
1080 
1081 		if (fnode->tn_type == VDIR && tnode->tn_type == VDIR) {
1082 			if (tnode->tn_size != 0 &&
1083 			    ((tcnp->cn_flags & IGNOREWHITEOUT) == 0 ||
1084 			    tnode->tn_size > tnode->tn_dir.tn_wht_size)) {
1085 				error = ENOTEMPTY;
1086 				goto out_locked;
1087 			}
1088 		} else if (fnode->tn_type == VDIR && tnode->tn_type != VDIR) {
1089 			error = ENOTDIR;
1090 			goto out_locked;
1091 		} else if (fnode->tn_type != VDIR && tnode->tn_type == VDIR) {
1092 			error = EISDIR;
1093 			goto out_locked;
1094 		} else {
1095 			MPASS(fnode->tn_type != VDIR &&
1096 				tnode->tn_type != VDIR);
1097 		}
1098 	}
1099 
1100 	if ((fnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND))
1101 	    || (fdnode->tn_flags & (APPEND | IMMUTABLE))) {
1102 		error = EPERM;
1103 		goto out_locked;
1104 	}
1105 
1106 	/*
1107 	 * Ensure that we have enough memory to hold the new name, if it
1108 	 * has to be changed.
1109 	 */
1110 	if (fcnp->cn_namelen != tcnp->cn_namelen ||
1111 	    bcmp(fcnp->cn_nameptr, tcnp->cn_nameptr, fcnp->cn_namelen) != 0) {
1112 		newname = malloc(tcnp->cn_namelen, M_TMPFSNAME, M_WAITOK);
1113 	} else
1114 		newname = NULL;
1115 
1116 	/*
1117 	 * If the node is being moved to another directory, we have to do
1118 	 * the move.
1119 	 */
1120 	if (fdnode != tdnode) {
1121 		/*
1122 		 * In case we are moving a directory, we have to adjust its
1123 		 * parent to point to the new parent.
1124 		 */
1125 		if (de->td_node->tn_type == VDIR) {
1126 			struct tmpfs_node *n;
1127 
1128 			TMPFS_NODE_LOCK(fnode);
1129 			error = tmpfs_access_locked(fvp, fnode, VWRITE,
1130 			    tcnp->cn_cred);
1131 			TMPFS_NODE_UNLOCK(fnode);
1132 			if (error) {
1133 				if (newname != NULL)
1134 					free(newname, M_TMPFSNAME);
1135 				goto out_locked;
1136 			}
1137 
1138 			/*
1139 			 * Ensure the target directory is not a child of the
1140 			 * directory being moved.  Otherwise, we'd end up
1141 			 * with stale nodes.
1142 			 */
1143 			n = tdnode;
1144 			/*
1145 			 * TMPFS_LOCK guaranties that no nodes are freed while
1146 			 * traversing the list. Nodes can only be marked as
1147 			 * removed: tn_parent == NULL.
1148 			 */
1149 			TMPFS_LOCK(tmp);
1150 			TMPFS_NODE_LOCK(n);
1151 			while (n != n->tn_dir.tn_parent) {
1152 				struct tmpfs_node *parent;
1153 
1154 				if (n == fnode) {
1155 					TMPFS_NODE_UNLOCK(n);
1156 					TMPFS_UNLOCK(tmp);
1157 					error = EINVAL;
1158 					if (newname != NULL)
1159 						free(newname, M_TMPFSNAME);
1160 					goto out_locked;
1161 				}
1162 				parent = n->tn_dir.tn_parent;
1163 				TMPFS_NODE_UNLOCK(n);
1164 				if (parent == NULL) {
1165 					n = NULL;
1166 					break;
1167 				}
1168 				TMPFS_NODE_LOCK(parent);
1169 				if (parent->tn_dir.tn_parent == NULL) {
1170 					TMPFS_NODE_UNLOCK(parent);
1171 					n = NULL;
1172 					break;
1173 				}
1174 				n = parent;
1175 			}
1176 			TMPFS_UNLOCK(tmp);
1177 			if (n == NULL) {
1178 				error = EINVAL;
1179 				if (newname != NULL)
1180 					    free(newname, M_TMPFSNAME);
1181 				goto out_locked;
1182 			}
1183 			TMPFS_NODE_UNLOCK(n);
1184 
1185 			/* Adjust the parent pointer. */
1186 			TMPFS_VALIDATE_DIR(fnode);
1187 			TMPFS_NODE_LOCK(de->td_node);
1188 			de->td_node->tn_dir.tn_parent = tdnode;
1189 			TMPFS_NODE_UNLOCK(de->td_node);
1190 
1191 			/*
1192 			 * As a result of changing the target of the '..'
1193 			 * entry, the link count of the source and target
1194 			 * directories has to be adjusted.
1195 			 */
1196 			TMPFS_NODE_LOCK(tdnode);
1197 			TMPFS_ASSERT_LOCKED(tdnode);
1198 			tdnode->tn_links++;
1199 			TMPFS_NODE_UNLOCK(tdnode);
1200 
1201 			TMPFS_NODE_LOCK(fdnode);
1202 			TMPFS_ASSERT_LOCKED(fdnode);
1203 			fdnode->tn_links--;
1204 			TMPFS_NODE_UNLOCK(fdnode);
1205 		}
1206 	}
1207 
1208 	/*
1209 	 * Do the move: just remove the entry from the source directory
1210 	 * and insert it into the target one.
1211 	 */
1212 	tmpfs_dir_detach(fdvp, de);
1213 
1214 	if (fcnp->cn_flags & DOWHITEOUT)
1215 		tmpfs_dir_whiteout_add(fdvp, fcnp);
1216 	if (tcnp->cn_flags & ISWHITEOUT)
1217 		tmpfs_dir_whiteout_remove(tdvp, tcnp);
1218 
1219 	/*
1220 	 * If the name has changed, we need to make it effective by changing
1221 	 * it in the directory entry.
1222 	 */
1223 	if (newname != NULL) {
1224 		MPASS(tcnp->cn_namelen <= MAXNAMLEN);
1225 
1226 		free(de->ud.td_name, M_TMPFSNAME);
1227 		de->ud.td_name = newname;
1228 		tmpfs_dirent_init(de, tcnp->cn_nameptr, tcnp->cn_namelen);
1229 
1230 		fnode->tn_status |= TMPFS_NODE_CHANGED;
1231 		tdnode->tn_status |= TMPFS_NODE_MODIFIED;
1232 	}
1233 
1234 	/*
1235 	 * If we are overwriting an entry, we have to remove the old one
1236 	 * from the target directory.
1237 	 */
1238 	if (tvp != NULL) {
1239 		struct tmpfs_dirent *tde;
1240 
1241 		/* Remove the old entry from the target directory. */
1242 		tde = tmpfs_dir_lookup(tdnode, tnode, tcnp);
1243 		tmpfs_dir_detach(tdvp, tde);
1244 
1245 		/*
1246 		 * If we are overwriting a directory, per the ENOTEMPTY check
1247 		 * above it must either be empty or contain only whiteout
1248 		 * entries.  In the latter case (which can only happen if
1249 		 * IGNOREWHITEOUT was passed in tcnp->cn_flags), clear the
1250 		 * whiteout entries to avoid leaking memory.
1251 		 */
1252 		if (tnode->tn_type == VDIR && tnode->tn_size > 0)
1253 			tmpfs_dir_clear_whiteouts(tvp);
1254 
1255 		/* Update node's ctime because of possible hardlinks. */
1256 		tnode->tn_status |= TMPFS_NODE_CHANGED;
1257 		tmpfs_update(tvp);
1258 
1259 		/*
1260 		 * Free the directory entry we just deleted.  Note that the
1261 		 * node referred by it will not be removed until the vnode is
1262 		 * really reclaimed.
1263 		 */
1264 		tmpfs_free_dirent(VFS_TO_TMPFS(tvp->v_mount), tde);
1265 	}
1266 
1267 	tmpfs_dir_attach(tdvp, de);
1268 
1269 	if (tmpfs_use_nc(fvp)) {
1270 		cache_vop_rename(fdvp, fvp, tdvp, tvp, fcnp, tcnp);
1271 	}
1272 
1273 	error = 0;
1274 
1275 out_locked:
1276 	if (fdvp != tdvp && fdvp != tvp)
1277 		VOP_UNLOCK(fdvp);
1278 
1279 out:
1280 	if (want_seqc_end) {
1281 		if (tvp != NULL)
1282 			vn_seqc_write_end(tvp);
1283 		vn_seqc_write_end(tdvp);
1284 		vn_seqc_write_end(fvp);
1285 		vn_seqc_write_end(fdvp);
1286 	}
1287 
1288 	/*
1289 	 * Release target nodes.
1290 	 * XXX: I don't understand when tdvp can be the same as tvp, but
1291 	 * other code takes care of this...
1292 	 */
1293 	if (tdvp == tvp)
1294 		vrele(tdvp);
1295 	else
1296 		vput(tdvp);
1297 	if (tvp != NULL)
1298 		vput(tvp);
1299 
1300 	/* Release source nodes. */
1301 	vrele(fdvp);
1302 	vrele(fvp);
1303 
1304 	return (error);
1305 }
1306 
1307 static int
1308 tmpfs_mkdir(struct vop_mkdir_args *v)
1309 {
1310 	struct vnode *dvp = v->a_dvp;
1311 	struct vnode **vpp = v->a_vpp;
1312 	struct componentname *cnp = v->a_cnp;
1313 	struct vattr *vap = v->a_vap;
1314 
1315 	MPASS(vap->va_type == VDIR);
1316 
1317 	return (tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL));
1318 }
1319 
1320 static int
1321 tmpfs_rmdir(struct vop_rmdir_args *v)
1322 {
1323 	struct vnode *dvp = v->a_dvp;
1324 	struct vnode *vp = v->a_vp;
1325 	struct componentname *cnp = v->a_cnp;
1326 
1327 	int error;
1328 	struct tmpfs_dirent *de;
1329 	struct tmpfs_mount *tmp;
1330 	struct tmpfs_node *dnode;
1331 	struct tmpfs_node *node;
1332 
1333 	tmp = VFS_TO_TMPFS(dvp->v_mount);
1334 	dnode = VP_TO_TMPFS_DIR(dvp);
1335 	node = VP_TO_TMPFS_DIR(vp);
1336 
1337 	/*
1338 	 * Directories with more than two non-whiteout entries ('.' and '..')
1339 	 * cannot be removed.
1340 	 */
1341 	if (node->tn_size != 0 &&
1342 	    ((cnp->cn_flags & IGNOREWHITEOUT) == 0 ||
1343 	    node->tn_size > node->tn_dir.tn_wht_size)) {
1344 		error = ENOTEMPTY;
1345 		goto out;
1346 	}
1347 
1348 	/* Check flags to see if we are allowed to remove the directory. */
1349 	if ((dnode->tn_flags & APPEND)
1350 	    || (node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1351 		error = EPERM;
1352 		goto out;
1353 	}
1354 
1355 	/* This invariant holds only if we are not trying to remove "..".
1356 	 * We checked for that above so this is safe now. */
1357 	MPASS(node->tn_dir.tn_parent == dnode);
1358 
1359 	/* Get the directory entry associated with node (vp).  This was
1360 	 * filled by tmpfs_lookup while looking up the entry. */
1361 	de = tmpfs_dir_lookup(dnode, node, cnp);
1362 	MPASS(TMPFS_DIRENT_MATCHES(de,
1363 	    cnp->cn_nameptr,
1364 	    cnp->cn_namelen));
1365 
1366 	/* Detach the directory entry from the directory (dnode). */
1367 	tmpfs_dir_detach(dvp, de);
1368 
1369 	/*
1370 	 * If we are removing a directory, per the ENOTEMPTY check above it
1371 	 * must either be empty or contain only whiteout entries.  In the
1372 	 * latter case (which can only happen if IGNOREWHITEOUT was passed
1373 	 * in cnp->cn_flags), clear the whiteout entries to avoid leaking
1374 	 * memory.
1375 	 */
1376 	if (node->tn_size > 0)
1377 		tmpfs_dir_clear_whiteouts(vp);
1378 
1379 	if (cnp->cn_flags & DOWHITEOUT)
1380 		tmpfs_dir_whiteout_add(dvp, cnp);
1381 
1382 	/* No vnode should be allocated for this entry from this point */
1383 	TMPFS_NODE_LOCK(node);
1384 	node->tn_links--;
1385 	node->tn_dir.tn_parent = NULL;
1386 	node->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1387 	node->tn_accessed = true;
1388 
1389 	TMPFS_NODE_UNLOCK(node);
1390 
1391 	TMPFS_NODE_LOCK(dnode);
1392 	dnode->tn_links--;
1393 	dnode->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1394 	dnode->tn_accessed = true;
1395 	TMPFS_NODE_UNLOCK(dnode);
1396 
1397 	if (tmpfs_use_nc(dvp)) {
1398 		cache_vop_rmdir(dvp, vp);
1399 	}
1400 
1401 	/* Free the directory entry we just deleted.  Note that the node
1402 	 * referred by it will not be removed until the vnode is really
1403 	 * reclaimed. */
1404 	tmpfs_free_dirent(tmp, de);
1405 
1406 	/* Release the deleted vnode (will destroy the node, notify
1407 	 * interested parties and clean it from the cache). */
1408 
1409 	dnode->tn_status |= TMPFS_NODE_CHANGED;
1410 	tmpfs_update(dvp);
1411 
1412 	error = 0;
1413 
1414 out:
1415 	return (error);
1416 }
1417 
1418 static int
1419 tmpfs_symlink(struct vop_symlink_args *v)
1420 {
1421 	struct vnode *dvp = v->a_dvp;
1422 	struct vnode **vpp = v->a_vpp;
1423 	struct componentname *cnp = v->a_cnp;
1424 	struct vattr *vap = v->a_vap;
1425 	const char *target = v->a_target;
1426 
1427 #ifdef notyet /* XXX FreeBSD BUG: kern_symlink is not setting VLNK */
1428 	MPASS(vap->va_type == VLNK);
1429 #else
1430 	vap->va_type = VLNK;
1431 #endif
1432 
1433 	return (tmpfs_alloc_file(dvp, vpp, vap, cnp, target));
1434 }
1435 
1436 static int
1437 tmpfs_readdir(struct vop_readdir_args *va)
1438 {
1439 	struct vnode *vp;
1440 	struct uio *uio;
1441 	struct tmpfs_mount *tm;
1442 	struct tmpfs_node *node;
1443 	uint64_t **cookies;
1444 	int *eofflag, *ncookies;
1445 	ssize_t startresid;
1446 	int error, maxcookies;
1447 
1448 	vp = va->a_vp;
1449 	uio = va->a_uio;
1450 	eofflag = va->a_eofflag;
1451 	cookies = va->a_cookies;
1452 	ncookies = va->a_ncookies;
1453 
1454 	/* This operation only makes sense on directory nodes. */
1455 	if (vp->v_type != VDIR)
1456 		return (ENOTDIR);
1457 
1458 	maxcookies = 0;
1459 	node = VP_TO_TMPFS_DIR(vp);
1460 	tm = VFS_TO_TMPFS(vp->v_mount);
1461 
1462 	startresid = uio->uio_resid;
1463 
1464 	/* Allocate cookies for NFS and compat modules. */
1465 	if (cookies != NULL && ncookies != NULL) {
1466 		maxcookies = howmany(node->tn_size,
1467 		    sizeof(struct tmpfs_dirent)) + 2;
1468 		*cookies = malloc(maxcookies * sizeof(**cookies), M_TEMP,
1469 		    M_WAITOK);
1470 		*ncookies = 0;
1471 	}
1472 
1473 	if (cookies == NULL)
1474 		error = tmpfs_dir_getdents(tm, node, uio, 0, NULL, NULL);
1475 	else
1476 		error = tmpfs_dir_getdents(tm, node, uio, maxcookies, *cookies,
1477 		    ncookies);
1478 
1479 	/* Buffer was filled without hitting EOF. */
1480 	if (error == EJUSTRETURN)
1481 		error = (uio->uio_resid != startresid) ? 0 : EINVAL;
1482 
1483 	if (error != 0 && cookies != NULL && ncookies != NULL) {
1484 		free(*cookies, M_TEMP);
1485 		*cookies = NULL;
1486 		*ncookies = 0;
1487 	}
1488 
1489 	if (eofflag != NULL)
1490 		*eofflag =
1491 		    (error == 0 && uio->uio_offset == TMPFS_DIRCOOKIE_EOF);
1492 
1493 	return (error);
1494 }
1495 
1496 static int
1497 tmpfs_readlink(struct vop_readlink_args *v)
1498 {
1499 	struct vnode *vp = v->a_vp;
1500 	struct uio *uio = v->a_uio;
1501 
1502 	int error;
1503 	struct tmpfs_node *node;
1504 
1505 	MPASS(uio->uio_offset == 0);
1506 	MPASS(vp->v_type == VLNK);
1507 
1508 	node = VP_TO_TMPFS_NODE(vp);
1509 
1510 	error = uiomove(node->tn_link_target, MIN(node->tn_size, uio->uio_resid),
1511 	    uio);
1512 	tmpfs_set_accessed(VFS_TO_TMPFS(vp->v_mount), node);
1513 
1514 	return (error);
1515 }
1516 
1517 /*
1518  * VOP_FPLOOKUP_SYMLINK routines are subject to special circumstances, see
1519  * the comment above cache_fplookup for details.
1520  *
1521  * Check tmpfs_alloc_node for tmpfs-specific synchronisation notes.
1522  */
1523 static int
1524 tmpfs_fplookup_symlink(struct vop_fplookup_symlink_args *v)
1525 {
1526 	struct vnode *vp;
1527 	struct tmpfs_node *node;
1528 	char *symlink;
1529 
1530 	vp = v->a_vp;
1531 	node = VP_TO_TMPFS_NODE_SMR(vp);
1532 	if (__predict_false(node == NULL))
1533 		return (EAGAIN);
1534 	if (!atomic_load_char(&node->tn_link_smr))
1535 		return (EAGAIN);
1536 	symlink = atomic_load_ptr(&node->tn_link_target);
1537 	if (symlink == NULL)
1538 		return (EAGAIN);
1539 
1540 	return (cache_symlink_resolve(v->a_fpl, symlink, node->tn_size));
1541 }
1542 
1543 static int
1544 tmpfs_inactive(struct vop_inactive_args *v)
1545 {
1546 	struct vnode *vp;
1547 	struct tmpfs_node *node;
1548 
1549 	vp = v->a_vp;
1550 	node = VP_TO_TMPFS_NODE(vp);
1551 	if (node->tn_links == 0)
1552 		vrecycle(vp);
1553 	else
1554 		tmpfs_check_mtime(vp);
1555 	return (0);
1556 }
1557 
1558 static int
1559 tmpfs_need_inactive(struct vop_need_inactive_args *ap)
1560 {
1561 	struct vnode *vp;
1562 	struct tmpfs_node *node;
1563 	struct vm_object *obj;
1564 
1565 	vp = ap->a_vp;
1566 	node = VP_TO_TMPFS_NODE(vp);
1567 	if (node->tn_links == 0)
1568 		goto need;
1569 	if (vp->v_type == VREG) {
1570 		obj = vp->v_object;
1571 		if (obj->generation != obj->cleangeneration)
1572 			goto need;
1573 	}
1574 	return (0);
1575 need:
1576 	return (1);
1577 }
1578 
1579 int
1580 tmpfs_reclaim(struct vop_reclaim_args *v)
1581 {
1582 	struct vnode *vp;
1583 	struct tmpfs_mount *tmp;
1584 	struct tmpfs_node *node;
1585 	bool unlock;
1586 
1587 	vp = v->a_vp;
1588 	node = VP_TO_TMPFS_NODE(vp);
1589 	tmp = VFS_TO_TMPFS(vp->v_mount);
1590 
1591 	if (vp->v_type == VREG)
1592 		tmpfs_destroy_vobject(vp, node->tn_reg.tn_aobj);
1593 	vp->v_object = NULL;
1594 
1595 	TMPFS_LOCK(tmp);
1596 	TMPFS_NODE_LOCK(node);
1597 	tmpfs_free_vp(vp);
1598 
1599 	/*
1600 	 * If the node referenced by this vnode was deleted by the user,
1601 	 * we must free its associated data structures (now that the vnode
1602 	 * is being reclaimed).
1603 	 */
1604 	unlock = true;
1605 	if (node->tn_links == 0 &&
1606 	    (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0) {
1607 		node->tn_vpstate = TMPFS_VNODE_DOOMED;
1608 		unlock = !tmpfs_free_node_locked(tmp, node, true);
1609 	}
1610 
1611 	if (unlock) {
1612 		TMPFS_NODE_UNLOCK(node);
1613 		TMPFS_UNLOCK(tmp);
1614 	}
1615 
1616 	MPASS(vp->v_data == NULL);
1617 	return (0);
1618 }
1619 
1620 int
1621 tmpfs_print(struct vop_print_args *v)
1622 {
1623 	struct vnode *vp = v->a_vp;
1624 
1625 	struct tmpfs_node *node;
1626 
1627 	node = VP_TO_TMPFS_NODE(vp);
1628 
1629 	printf("tag VT_TMPFS, tmpfs_node %p, flags 0x%lx, links %jd\n",
1630 	    node, node->tn_flags, (uintmax_t)node->tn_links);
1631 	printf("\tmode 0%o, owner %d, group %d, size %jd, status 0x%x\n",
1632 	    node->tn_mode, node->tn_uid, node->tn_gid,
1633 	    (intmax_t)node->tn_size, node->tn_status);
1634 
1635 	if (vp->v_type == VFIFO)
1636 		fifo_printinfo(vp);
1637 
1638 	printf("\n");
1639 
1640 	return (0);
1641 }
1642 
1643 int
1644 tmpfs_pathconf(struct vop_pathconf_args *v)
1645 {
1646 	struct vnode *vp = v->a_vp;
1647 	int name = v->a_name;
1648 	long *retval = v->a_retval;
1649 
1650 	int error;
1651 
1652 	error = 0;
1653 
1654 	switch (name) {
1655 	case _PC_LINK_MAX:
1656 		*retval = TMPFS_LINK_MAX;
1657 		break;
1658 
1659 	case _PC_SYMLINK_MAX:
1660 		*retval = MAXPATHLEN;
1661 		break;
1662 
1663 	case _PC_NAME_MAX:
1664 		*retval = NAME_MAX;
1665 		break;
1666 
1667 	case _PC_PIPE_BUF:
1668 		if (vp->v_type == VDIR || vp->v_type == VFIFO)
1669 			*retval = PIPE_BUF;
1670 		else
1671 			error = EINVAL;
1672 		break;
1673 
1674 	case _PC_CHOWN_RESTRICTED:
1675 		*retval = 1;
1676 		break;
1677 
1678 	case _PC_NO_TRUNC:
1679 		*retval = 1;
1680 		break;
1681 
1682 	case _PC_SYNC_IO:
1683 		*retval = 1;
1684 		break;
1685 
1686 	case _PC_FILESIZEBITS:
1687 		*retval = 64;
1688 		break;
1689 
1690 	case _PC_MIN_HOLE_SIZE:
1691 		*retval = PAGE_SIZE;
1692 		break;
1693 
1694 	default:
1695 		error = vop_stdpathconf(v);
1696 	}
1697 
1698 	return (error);
1699 }
1700 
1701 static int
1702 tmpfs_vptofh(struct vop_vptofh_args *ap)
1703 /*
1704 vop_vptofh {
1705 	IN struct vnode *a_vp;
1706 	IN struct fid *a_fhp;
1707 };
1708 */
1709 {
1710 	struct tmpfs_fid_data *const tfd = (struct tmpfs_fid_data *)ap->a_fhp;
1711 	struct tmpfs_node *node;
1712 	_Static_assert(sizeof(struct tmpfs_fid_data) <= sizeof(struct fid),
1713 	    "struct tmpfs_fid_data cannot be larger than struct fid");
1714 
1715 	node = VP_TO_TMPFS_NODE(ap->a_vp);
1716 	tfd->tfd_len = sizeof(*tfd);
1717 	tfd->tfd_gen = node->tn_gen;
1718 	tfd->tfd_id = node->tn_id;
1719 
1720 	return (0);
1721 }
1722 
1723 static int
1724 tmpfs_whiteout(struct vop_whiteout_args *ap)
1725 {
1726 	struct vnode *dvp = ap->a_dvp;
1727 	struct componentname *cnp = ap->a_cnp;
1728 	struct tmpfs_dirent *de;
1729 
1730 	switch (ap->a_flags) {
1731 	case LOOKUP:
1732 		return (0);
1733 	case CREATE:
1734 		de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(dvp), NULL, cnp);
1735 		if (de != NULL)
1736 			return (de->td_node == NULL ? 0 : EEXIST);
1737 		return (tmpfs_dir_whiteout_add(dvp, cnp));
1738 	case DELETE:
1739 		tmpfs_dir_whiteout_remove(dvp, cnp);
1740 		return (0);
1741 	default:
1742 		panic("tmpfs_whiteout: unknown op");
1743 	}
1744 }
1745 
1746 static int
1747 tmpfs_vptocnp_dir(struct tmpfs_node *tn, struct tmpfs_node *tnp,
1748     struct tmpfs_dirent **pde)
1749 {
1750 	struct tmpfs_dir_cursor dc;
1751 	struct tmpfs_dirent *de;
1752 
1753 	for (de = tmpfs_dir_first(tnp, &dc); de != NULL;
1754 	     de = tmpfs_dir_next(tnp, &dc)) {
1755 		if (de->td_node == tn) {
1756 			*pde = de;
1757 			return (0);
1758 		}
1759 	}
1760 	return (ENOENT);
1761 }
1762 
1763 static int
1764 tmpfs_vptocnp_fill(struct vnode *vp, struct tmpfs_node *tn,
1765     struct tmpfs_node *tnp, char *buf, size_t *buflen, struct vnode **dvp)
1766 {
1767 	struct tmpfs_dirent *de;
1768 	int error, i;
1769 
1770 	error = vn_vget_ino_gen(vp, tmpfs_vn_get_ino_alloc, tnp, LK_SHARED,
1771 	    dvp);
1772 	if (error != 0)
1773 		return (error);
1774 	error = tmpfs_vptocnp_dir(tn, tnp, &de);
1775 	if (error == 0) {
1776 		i = *buflen;
1777 		i -= de->td_namelen;
1778 		if (i < 0) {
1779 			error = ENOMEM;
1780 		} else {
1781 			bcopy(de->ud.td_name, buf + i, de->td_namelen);
1782 			*buflen = i;
1783 		}
1784 	}
1785 	if (error == 0) {
1786 		if (vp != *dvp)
1787 			VOP_UNLOCK(*dvp);
1788 	} else {
1789 		if (vp != *dvp)
1790 			vput(*dvp);
1791 		else
1792 			vrele(vp);
1793 	}
1794 	return (error);
1795 }
1796 
1797 static int
1798 tmpfs_vptocnp(struct vop_vptocnp_args *ap)
1799 {
1800 	struct vnode *vp, **dvp;
1801 	struct tmpfs_node *tn, *tnp, *tnp1;
1802 	struct tmpfs_dirent *de;
1803 	struct tmpfs_mount *tm;
1804 	char *buf;
1805 	size_t *buflen;
1806 	int error;
1807 
1808 	vp = ap->a_vp;
1809 	dvp = ap->a_vpp;
1810 	buf = ap->a_buf;
1811 	buflen = ap->a_buflen;
1812 
1813 	tm = VFS_TO_TMPFS(vp->v_mount);
1814 	tn = VP_TO_TMPFS_NODE(vp);
1815 	if (tn->tn_type == VDIR) {
1816 		tnp = tn->tn_dir.tn_parent;
1817 		if (tnp == NULL)
1818 			return (ENOENT);
1819 		tmpfs_ref_node(tnp);
1820 		error = tmpfs_vptocnp_fill(vp, tn, tn->tn_dir.tn_parent, buf,
1821 		    buflen, dvp);
1822 		tmpfs_free_node(tm, tnp);
1823 		return (error);
1824 	}
1825 restart:
1826 	TMPFS_LOCK(tm);
1827 restart_locked:
1828 	LIST_FOREACH_SAFE(tnp, &tm->tm_nodes_used, tn_entries, tnp1) {
1829 		if (tnp->tn_type != VDIR)
1830 			continue;
1831 		TMPFS_NODE_LOCK(tnp);
1832 		tmpfs_ref_node(tnp);
1833 
1834 		/*
1835 		 * tn_vnode cannot be instantiated while we hold the
1836 		 * node lock, so the directory cannot be changed while
1837 		 * we iterate over it.  Do this to avoid instantiating
1838 		 * vnode for directories which cannot point to our
1839 		 * node.
1840 		 */
1841 		error = tnp->tn_vnode == NULL ? tmpfs_vptocnp_dir(tn, tnp,
1842 		    &de) : 0;
1843 
1844 		if (error == 0) {
1845 			TMPFS_NODE_UNLOCK(tnp);
1846 			TMPFS_UNLOCK(tm);
1847 			error = tmpfs_vptocnp_fill(vp, tn, tnp, buf, buflen,
1848 			    dvp);
1849 			if (error == 0) {
1850 				tmpfs_free_node(tm, tnp);
1851 				return (0);
1852 			}
1853 			if (VN_IS_DOOMED(vp)) {
1854 				tmpfs_free_node(tm, tnp);
1855 				return (ENOENT);
1856 			}
1857 			TMPFS_LOCK(tm);
1858 			TMPFS_NODE_LOCK(tnp);
1859 		}
1860 		if (tmpfs_free_node_locked(tm, tnp, false)) {
1861 			goto restart;
1862 		} else {
1863 			KASSERT(tnp->tn_refcount > 0,
1864 			    ("node %p refcount zero", tnp));
1865 			if (tnp->tn_attached) {
1866 				tnp1 = LIST_NEXT(tnp, tn_entries);
1867 				TMPFS_NODE_UNLOCK(tnp);
1868 			} else {
1869 				TMPFS_NODE_UNLOCK(tnp);
1870 				goto restart_locked;
1871 			}
1872 		}
1873 	}
1874 	TMPFS_UNLOCK(tm);
1875 	return (ENOENT);
1876 }
1877 
1878 void
1879 tmpfs_extattr_free(struct tmpfs_extattr *ea)
1880 {
1881 	free(ea->ea_name, M_TMPFSEA);
1882 	free(ea->ea_value, M_TMPFSEA);
1883 	free(ea, M_TMPFSEA);
1884 }
1885 
1886 static bool
1887 tmpfs_extattr_update_mem(struct tmpfs_mount *tmp, ssize_t size)
1888 {
1889 	TMPFS_LOCK(tmp);
1890 	if (size > 0 &&
1891 	    !tmpfs_pages_check_avail(tmp, howmany(size, PAGE_SIZE))) {
1892 		TMPFS_UNLOCK(tmp);
1893 		return (false);
1894 	}
1895 	if (tmp->tm_ea_memory_inuse + size > tmp->tm_ea_memory_max) {
1896 		TMPFS_UNLOCK(tmp);
1897 		return (false);
1898 	}
1899 	tmp->tm_ea_memory_inuse += size;
1900 	TMPFS_UNLOCK(tmp);
1901 	return (true);
1902 }
1903 
1904 static int
1905 tmpfs_deleteextattr(struct vop_deleteextattr_args *ap)
1906 {
1907 	struct vnode *vp = ap->a_vp;
1908 	struct tmpfs_mount *tmp;
1909 	struct tmpfs_node *node;
1910 	struct tmpfs_extattr *ea;
1911 	size_t namelen;
1912 	ssize_t diff;
1913 	int error;
1914 
1915 	node = VP_TO_TMPFS_NODE(vp);
1916 	tmp = VFS_TO_TMPFS(vp->v_mount);
1917 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1918 		return (EOPNOTSUPP);
1919 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1920 	    ap->a_cred, ap->a_td, VWRITE);
1921 	if (error != 0)
1922 		return (error);
1923 	if (ap->a_name == NULL || ap->a_name[0] == '\0')
1924 		return (EINVAL);
1925 	namelen = strlen(ap->a_name);
1926 	if (namelen > EXTATTR_MAXNAMELEN)
1927 		return (EINVAL);
1928 
1929 	LIST_FOREACH(ea, &node->tn_extattrs, ea_extattrs) {
1930 		if (ea->ea_namespace == ap->a_attrnamespace &&
1931 		    namelen == ea->ea_namelen &&
1932 		    memcmp(ap->a_name, ea->ea_name, namelen) == 0)
1933 			break;
1934 	}
1935 
1936 	if (ea == NULL)
1937 		return (ENOATTR);
1938 	LIST_REMOVE(ea, ea_extattrs);
1939 	diff = -(sizeof(struct tmpfs_extattr) + namelen + ea->ea_size);
1940 	tmpfs_extattr_update_mem(tmp, diff);
1941 	tmpfs_extattr_free(ea);
1942 	return (0);
1943 }
1944 
1945 static int
1946 tmpfs_getextattr(struct vop_getextattr_args *ap)
1947 {
1948 	struct vnode *vp = ap->a_vp;
1949 	struct tmpfs_node *node;
1950 	struct tmpfs_extattr *ea;
1951 	size_t namelen;
1952 	int error;
1953 
1954 	node = VP_TO_TMPFS_NODE(vp);
1955 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1956 		return (EOPNOTSUPP);
1957 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1958 	    ap->a_cred, ap->a_td, VREAD);
1959 	if (error != 0)
1960 		return (error);
1961 	if (ap->a_name == NULL || ap->a_name[0] == '\0')
1962 		return (EINVAL);
1963 	namelen = strlen(ap->a_name);
1964 	if (namelen > EXTATTR_MAXNAMELEN)
1965 		return (EINVAL);
1966 
1967 	LIST_FOREACH(ea, &node->tn_extattrs, ea_extattrs) {
1968 		if (ea->ea_namespace == ap->a_attrnamespace &&
1969 		    namelen == ea->ea_namelen &&
1970 		    memcmp(ap->a_name, ea->ea_name, namelen) == 0)
1971 			break;
1972 	}
1973 
1974 	if (ea == NULL)
1975 		return (ENOATTR);
1976 	if (ap->a_size != NULL)
1977 		*ap->a_size = ea->ea_size;
1978 	if (ap->a_uio != NULL && ea->ea_size != 0)
1979 		error = uiomove(ea->ea_value, ea->ea_size, ap->a_uio);
1980 	return (error);
1981 }
1982 
1983 static int
1984 tmpfs_listextattr(struct vop_listextattr_args *ap)
1985 {
1986 	struct vnode *vp = ap->a_vp;
1987 	struct tmpfs_node *node;
1988 	struct tmpfs_extattr *ea;
1989 	int error;
1990 
1991 	node = VP_TO_TMPFS_NODE(vp);
1992 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1993 		return (EOPNOTSUPP);
1994 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1995 	    ap->a_cred, ap->a_td, VREAD);
1996 	if (error != 0)
1997 		return (error);
1998 	if (ap->a_size != NULL)
1999 		*ap->a_size = 0;
2000 
2001 	LIST_FOREACH(ea, &node->tn_extattrs, ea_extattrs) {
2002 		if (ea->ea_namespace != ap->a_attrnamespace)
2003 			continue;
2004 		if (ap->a_size != NULL)
2005 			*ap->a_size += ea->ea_namelen + 1;
2006 		if (ap->a_uio != NULL) {
2007 			error = uiomove(&ea->ea_namelen, 1, ap->a_uio);
2008 			if (error != 0)
2009 				break;
2010 			error = uiomove(ea->ea_name, ea->ea_namelen, ap->a_uio);
2011 			if (error != 0)
2012 				break;
2013 		}
2014 	}
2015 
2016 	return (error);
2017 }
2018 
2019 static int
2020 tmpfs_setextattr(struct vop_setextattr_args *ap)
2021 {
2022 	struct vnode *vp = ap->a_vp;
2023 	struct tmpfs_mount *tmp;
2024 	struct tmpfs_node *node;
2025 	struct tmpfs_extattr *ea;
2026 	struct tmpfs_extattr *new_ea;
2027 	size_t attr_size;
2028 	size_t namelen;
2029 	ssize_t diff;
2030 	int error;
2031 
2032 	node = VP_TO_TMPFS_NODE(vp);
2033 	tmp = VFS_TO_TMPFS(vp->v_mount);
2034 	attr_size = ap->a_uio->uio_resid;
2035 	diff = 0;
2036 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
2037 		return (EOPNOTSUPP);
2038 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
2039 	    ap->a_cred, ap->a_td, VWRITE);
2040 	if (error != 0)
2041 		return (error);
2042 	if (ap->a_name == NULL || ap->a_name[0] == '\0')
2043 		return (EINVAL);
2044 	namelen = strlen(ap->a_name);
2045 	if (namelen > EXTATTR_MAXNAMELEN)
2046 		return (EINVAL);
2047 
2048 	LIST_FOREACH(ea, &node->tn_extattrs, ea_extattrs) {
2049 		if (ea->ea_namespace == ap->a_attrnamespace &&
2050 		    namelen == ea->ea_namelen &&
2051 		    memcmp(ap->a_name, ea->ea_name, namelen) == 0) {
2052 			diff -= sizeof(struct tmpfs_extattr) + ea->ea_namelen +
2053 			    ea->ea_size;
2054 			break;
2055 		}
2056 	}
2057 
2058 	diff += sizeof(struct tmpfs_extattr) + namelen + attr_size;
2059 	if (!tmpfs_extattr_update_mem(tmp, diff))
2060 		return (ENOSPC);
2061 	new_ea = malloc(sizeof(struct tmpfs_extattr), M_TMPFSEA, M_WAITOK);
2062 	new_ea->ea_namespace = ap->a_attrnamespace;
2063 	new_ea->ea_name = malloc(namelen, M_TMPFSEA, M_WAITOK);
2064 	new_ea->ea_namelen = namelen;
2065 	memcpy(new_ea->ea_name, ap->a_name, namelen);
2066 	if (attr_size != 0) {
2067 		new_ea->ea_value = malloc(attr_size, M_TMPFSEA, M_WAITOK);
2068 		new_ea->ea_size = attr_size;
2069 		error = uiomove(new_ea->ea_value, attr_size, ap->a_uio);
2070 	} else {
2071 		new_ea->ea_value = NULL;
2072 		new_ea->ea_size = 0;
2073 	}
2074 	if (error != 0) {
2075 		tmpfs_extattr_update_mem(tmp, -diff);
2076 		tmpfs_extattr_free(new_ea);
2077 		return (error);
2078 	}
2079 	if (ea != NULL) {
2080 		LIST_REMOVE(ea, ea_extattrs);
2081 		tmpfs_extattr_free(ea);
2082 	}
2083 	LIST_INSERT_HEAD(&node->tn_extattrs, new_ea, ea_extattrs);
2084 	return (0);
2085 }
2086 
2087 static off_t
2088 tmpfs_seek_data_locked(vm_object_t obj, off_t noff)
2089 {
2090 	vm_pindex_t p;
2091 
2092 	p = swap_pager_seek_data(obj, OFF_TO_IDX(noff));
2093 	return (p == OFF_TO_IDX(noff) ? noff : IDX_TO_OFF(p));
2094 }
2095 
2096 static int
2097 tmpfs_seek_clamp(struct tmpfs_node *tn, off_t *noff, bool seekdata)
2098 {
2099 	if (*noff < tn->tn_size)
2100 		return (0);
2101 	if (seekdata)
2102 		return (ENXIO);
2103 	*noff = tn->tn_size;
2104 	return (0);
2105 }
2106 
2107 static off_t
2108 tmpfs_seek_hole_locked(vm_object_t obj, off_t noff)
2109 {
2110 
2111 	return (IDX_TO_OFF(swap_pager_seek_hole(obj, OFF_TO_IDX(noff))));
2112 }
2113 
2114 static int
2115 tmpfs_seek_datahole(struct vnode *vp, off_t *off, bool seekdata)
2116 {
2117 	struct tmpfs_node *tn;
2118 	vm_object_t obj;
2119 	off_t noff;
2120 	int error;
2121 
2122 	if (vp->v_type != VREG)
2123 		return (ENOTTY);
2124 	tn = VP_TO_TMPFS_NODE(vp);
2125 	noff = *off;
2126 	if (noff < 0)
2127 		return (ENXIO);
2128 	error = tmpfs_seek_clamp(tn, &noff, seekdata);
2129 	if (error != 0)
2130 		return (error);
2131 	obj = tn->tn_reg.tn_aobj;
2132 
2133 	VM_OBJECT_RLOCK(obj);
2134 	noff = seekdata ? tmpfs_seek_data_locked(obj, noff) :
2135 	    tmpfs_seek_hole_locked(obj, noff);
2136 	VM_OBJECT_RUNLOCK(obj);
2137 
2138 	error = tmpfs_seek_clamp(tn, &noff, seekdata);
2139 	if (error == 0)
2140 		*off = noff;
2141 	return (error);
2142 }
2143 
2144 static int
2145 tmpfs_ioctl(struct vop_ioctl_args *ap)
2146 {
2147 	struct vnode *vp = ap->a_vp;
2148 	int error = 0;
2149 
2150 	switch (ap->a_command) {
2151 	case FIOSEEKDATA:
2152 	case FIOSEEKHOLE:
2153 		error = vn_lock(vp, LK_SHARED);
2154 		if (error != 0) {
2155 			error = EBADF;
2156 			break;
2157 		}
2158 		error = tmpfs_seek_datahole(vp, (off_t *)ap->a_data,
2159 		    ap->a_command == FIOSEEKDATA);
2160 		VOP_UNLOCK(vp);
2161 		break;
2162 	default:
2163 		error = ENOTTY;
2164 		break;
2165 	}
2166 	return (error);
2167 }
2168 
2169 /*
2170  * Vnode operations vector used for files stored in a tmpfs file system.
2171  */
2172 struct vop_vector tmpfs_vnodeop_entries = {
2173 	.vop_default =			&default_vnodeops,
2174 	.vop_lookup =			vfs_cache_lookup,
2175 	.vop_cachedlookup =		tmpfs_cached_lookup,
2176 	.vop_create =			tmpfs_create,
2177 	.vop_mknod =			tmpfs_mknod,
2178 	.vop_open =			tmpfs_open,
2179 	.vop_close =			tmpfs_close,
2180 	.vop_fplookup_vexec =		tmpfs_fplookup_vexec,
2181 	.vop_fplookup_symlink =		tmpfs_fplookup_symlink,
2182 	.vop_access =			tmpfs_access,
2183 	.vop_stat =			tmpfs_stat,
2184 	.vop_getattr =			tmpfs_getattr,
2185 	.vop_setattr =			tmpfs_setattr,
2186 	.vop_read =			tmpfs_read,
2187 	.vop_read_pgcache =		tmpfs_read_pgcache,
2188 	.vop_write =			tmpfs_write,
2189 	.vop_deallocate =		tmpfs_deallocate,
2190 	.vop_fsync =			tmpfs_fsync,
2191 	.vop_remove =			tmpfs_remove,
2192 	.vop_link =			tmpfs_link,
2193 	.vop_rename =			tmpfs_rename,
2194 	.vop_mkdir =			tmpfs_mkdir,
2195 	.vop_rmdir =			tmpfs_rmdir,
2196 	.vop_symlink =			tmpfs_symlink,
2197 	.vop_readdir =			tmpfs_readdir,
2198 	.vop_readlink =			tmpfs_readlink,
2199 	.vop_inactive =			tmpfs_inactive,
2200 	.vop_need_inactive =		tmpfs_need_inactive,
2201 	.vop_reclaim =			tmpfs_reclaim,
2202 	.vop_print =			tmpfs_print,
2203 	.vop_pathconf =			tmpfs_pathconf,
2204 	.vop_vptofh =			tmpfs_vptofh,
2205 	.vop_whiteout =			tmpfs_whiteout,
2206 	.vop_bmap =			VOP_EOPNOTSUPP,
2207 	.vop_vptocnp =			tmpfs_vptocnp,
2208 	.vop_lock1 =			vop_lock,
2209 	.vop_unlock = 			vop_unlock,
2210 	.vop_islocked = 		vop_islocked,
2211 	.vop_deleteextattr =		tmpfs_deleteextattr,
2212 	.vop_getextattr =		tmpfs_getextattr,
2213 	.vop_listextattr =		tmpfs_listextattr,
2214 	.vop_setextattr =		tmpfs_setextattr,
2215 	.vop_add_writecount =		vop_stdadd_writecount_nomsync,
2216 	.vop_ioctl =			tmpfs_ioctl,
2217 };
2218 VFS_VOP_VECTOR_REGISTER(tmpfs_vnodeop_entries);
2219 
2220 /*
2221  * Same vector for mounts which do not use namecache.
2222  */
2223 struct vop_vector tmpfs_vnodeop_nonc_entries = {
2224 	.vop_default =			&tmpfs_vnodeop_entries,
2225 	.vop_lookup =			tmpfs_lookup,
2226 };
2227 VFS_VOP_VECTOR_REGISTER(tmpfs_vnodeop_nonc_entries);
2228