tmpfs_vnops.c revision 1.52 
1/*	$NetBSD: tmpfs_vnops.c,v 1.52 2008/11/26 20:17:33 pooka Exp $	*/
2
3/*
4 * Copyright (c) 2005, 2006, 2007 The NetBSD Foundation, Inc.
5 * All rights reserved.
6 *
7 * This code is derived from software contributed to The NetBSD Foundation
8 * by Julio M. Merino Vidal, developed as part of Google's Summer of Code
9 * 2005 program.
10 *
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
13 * are met:
14 * 1. Redistributions of source code must retain the above copyright
15 *    notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 *    notice, this list of conditions and the following disclaimer in the
18 *    documentation and/or other materials provided with the distribution.
19 *
20 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
22 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
23 * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
24 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
25 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
26 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
27 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
28 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
29 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
30 * POSSIBILITY OF SUCH DAMAGE.
31 */
32
33/*
34 * tmpfs vnode interface.
35 */
36
37#include <sys/cdefs.h>
38__KERNEL_RCSID(0, "$NetBSD: tmpfs_vnops.c,v 1.52 2008/11/26 20:17:33 pooka Exp $");
39
40#include <sys/param.h>
41#include <sys/dirent.h>
42#include <sys/fcntl.h>
43#include <sys/event.h>
44#include <sys/malloc.h>
45#include <sys/namei.h>
46#include <sys/proc.h>
47#include <sys/stat.h>
48#include <sys/uio.h>
49#include <sys/unistd.h>
50#include <sys/vnode.h>
51#include <sys/lockf.h>
52#include <sys/kauth.h>
53
54#include <uvm/uvm.h>
55
56#include <miscfs/fifofs/fifo.h>
57#include <fs/tmpfs/tmpfs_vnops.h>
58#include <fs/tmpfs/tmpfs.h>
59
60/* --------------------------------------------------------------------- */
61
62/*
63 * vnode operations vector used for files stored in a tmpfs file system.
64 */
65int (**tmpfs_vnodeop_p)(void *);
66const struct vnodeopv_entry_desc tmpfs_vnodeop_entries[] = {
67	{ &vop_default_desc,		vn_default_error },
68	{ &vop_lookup_desc,		tmpfs_lookup },
69	{ &vop_create_desc,		tmpfs_create },
70	{ &vop_mknod_desc,		tmpfs_mknod },
71	{ &vop_open_desc,		tmpfs_open },
72	{ &vop_close_desc,		tmpfs_close },
73	{ &vop_access_desc,		tmpfs_access },
74	{ &vop_getattr_desc,		tmpfs_getattr },
75	{ &vop_setattr_desc,		tmpfs_setattr },
76	{ &vop_read_desc,		tmpfs_read },
77	{ &vop_write_desc,		tmpfs_write },
78	{ &vop_ioctl_desc,		tmpfs_ioctl },
79	{ &vop_fcntl_desc,		tmpfs_fcntl },
80	{ &vop_poll_desc,		tmpfs_poll },
81	{ &vop_kqfilter_desc,		tmpfs_kqfilter },
82	{ &vop_revoke_desc,		tmpfs_revoke },
83	{ &vop_mmap_desc,		tmpfs_mmap },
84	{ &vop_fsync_desc,		tmpfs_fsync },
85	{ &vop_seek_desc,		tmpfs_seek },
86	{ &vop_remove_desc,		tmpfs_remove },
87	{ &vop_link_desc,		tmpfs_link },
88	{ &vop_rename_desc,		tmpfs_rename },
89	{ &vop_mkdir_desc,		tmpfs_mkdir },
90	{ &vop_rmdir_desc,		tmpfs_rmdir },
91	{ &vop_symlink_desc,		tmpfs_symlink },
92	{ &vop_readdir_desc,		tmpfs_readdir },
93	{ &vop_readlink_desc,		tmpfs_readlink },
94	{ &vop_abortop_desc,		tmpfs_abortop },
95	{ &vop_inactive_desc,		tmpfs_inactive },
96	{ &vop_reclaim_desc,		tmpfs_reclaim },
97	{ &vop_lock_desc,		tmpfs_lock },
98	{ &vop_unlock_desc,		tmpfs_unlock },
99	{ &vop_bmap_desc,		tmpfs_bmap },
100	{ &vop_strategy_desc,		tmpfs_strategy },
101	{ &vop_print_desc,		tmpfs_print },
102	{ &vop_pathconf_desc,		tmpfs_pathconf },
103	{ &vop_islocked_desc,		tmpfs_islocked },
104	{ &vop_advlock_desc,		tmpfs_advlock },
105	{ &vop_bwrite_desc,		tmpfs_bwrite },
106	{ &vop_getpages_desc,		tmpfs_getpages },
107	{ &vop_putpages_desc,		tmpfs_putpages },
108	{ NULL, NULL }
109};
110const struct vnodeopv_desc tmpfs_vnodeop_opv_desc =
111	{ &tmpfs_vnodeop_p, tmpfs_vnodeop_entries };
112
113/* --------------------------------------------------------------------- */
114
115int
116tmpfs_lookup(void *v)
117{
118	struct vnode *dvp = ((struct vop_lookup_args *)v)->a_dvp;
119	struct vnode **vpp = ((struct vop_lookup_args *)v)->a_vpp;
120	struct componentname *cnp = ((struct vop_lookup_args *)v)->a_cnp;
121
122	int error;
123	struct tmpfs_dirent *de;
124	struct tmpfs_node *dnode;
125
126	KASSERT(VOP_ISLOCKED(dvp));
127
128	dnode = VP_TO_TMPFS_DIR(dvp);
129	*vpp = NULL;
130
131	/* Check accessibility of requested node as a first step. */
132	error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred);
133	if (error != 0)
134		goto out;
135
136	/* If requesting the last path component on a read-only file system
137	 * with a write operation, deny it. */
138	if ((cnp->cn_flags & ISLASTCN) &&
139	    (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
140	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
141		error = EROFS;
142		goto out;
143	}
144
145	/* Avoid doing a linear scan of the directory if the requested
146	 * directory/name couple is already in the cache. */
147	error = cache_lookup(dvp, vpp, cnp);
148	if (error >= 0)
149		goto out;
150
151	/* We cannot be requesting the parent directory of the root node. */
152	KASSERT(IMPLIES(dnode->tn_type == VDIR &&
153	    dnode->tn_spec.tn_dir.tn_parent == dnode,
154	    !(cnp->cn_flags & ISDOTDOT)));
155
156	if (cnp->cn_flags & ISDOTDOT) {
157		VOP_UNLOCK(dvp, 0);
158
159		/* Allocate a new vnode on the matching entry. */
160		error = tmpfs_alloc_vp(dvp->v_mount,
161		    dnode->tn_spec.tn_dir.tn_parent, vpp);
162
163		vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
164	} else if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
165		VREF(dvp);
166		*vpp = dvp;
167		error = 0;
168	} else {
169		de = tmpfs_dir_lookup(dnode, cnp);
170		if (de == NULL) {
171			/* The entry was not found in the directory.
172			 * This is OK iff we are creating or renaming an
173			 * entry and are working on the last component of
174			 * the path name. */
175			if ((cnp->cn_flags & ISLASTCN) &&
176			    (cnp->cn_nameiop == CREATE || \
177			    cnp->cn_nameiop == RENAME)) {
178				error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred);
179				if (error != 0)
180					goto out;
181
182				/* Keep the component name in the buffer for
183				 * future uses. */
184				cnp->cn_flags |= SAVENAME;
185
186				error = EJUSTRETURN;
187			} else
188				error = ENOENT;
189		} else {
190			struct tmpfs_node *tnode;
191
192			/* The entry was found, so get its associated
193			 * tmpfs_node. */
194			tnode = de->td_node;
195
196			/* If we are not at the last path component and
197			 * found a non-directory or non-link entry (which
198			 * may itself be pointing to a directory), raise
199			 * an error. */
200			if ((tnode->tn_type != VDIR &&
201			    tnode->tn_type != VLNK) &&
202			    !(cnp->cn_flags & ISLASTCN)) {
203				error = ENOTDIR;
204				goto out;
205			}
206
207			/* If we are deleting or renaming the entry, keep
208			 * track of its tmpfs_dirent so that it can be
209			 * easily deleted later. */
210			if ((cnp->cn_flags & ISLASTCN) &&
211			    (cnp->cn_nameiop == DELETE ||
212			    cnp->cn_nameiop == RENAME)) {
213				if ((dnode->tn_mode & S_ISTXT) != 0 &&
214				    kauth_authorize_generic(cnp->cn_cred,
215				     KAUTH_GENERIC_ISSUSER, NULL) != 0 &&
216				    kauth_cred_geteuid(cnp->cn_cred) != dnode->tn_uid &&
217				    kauth_cred_geteuid(cnp->cn_cred) != tnode->tn_uid)
218					return EPERM;
219				error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred);
220				if (error != 0)
221					goto out;
222			} else
223				de = NULL;
224
225			/* Allocate a new vnode on the matching entry. */
226			error = tmpfs_alloc_vp(dvp->v_mount, tnode, vpp);
227		}
228	}
229
230	/* Store the result of this lookup in the cache.  Avoid this if the
231	 * request was for creation, as it does not improve timings on
232	 * emprical tests. */
233	if ((cnp->cn_flags & MAKEENTRY) && cnp->cn_nameiop != CREATE &&
234	    (cnp->cn_flags & ISDOTDOT) == 0)
235		cache_enter(dvp, *vpp, cnp);
236
237out:
238	/* If there were no errors, *vpp cannot be null and it must be
239	 * locked. */
240	KASSERT(IFF(error == 0, *vpp != NULL && VOP_ISLOCKED(*vpp)));
241
242	/* dvp must always be locked. */
243	KASSERT(VOP_ISLOCKED(dvp));
244
245	return error;
246}
247
248/* --------------------------------------------------------------------- */
249
250int
251tmpfs_create(void *v)
252{
253	struct vnode *dvp = ((struct vop_create_args *)v)->a_dvp;
254	struct vnode **vpp = ((struct vop_create_args *)v)->a_vpp;
255	struct componentname *cnp = ((struct vop_create_args *)v)->a_cnp;
256	struct vattr *vap = ((struct vop_create_args *)v)->a_vap;
257
258	KASSERT(vap->va_type == VREG || vap->va_type == VSOCK);
259
260	return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
261}
262/* --------------------------------------------------------------------- */
263
264int
265tmpfs_mknod(void *v)
266{
267	struct vnode *dvp = ((struct vop_mknod_args *)v)->a_dvp;
268	struct vnode **vpp = ((struct vop_mknod_args *)v)->a_vpp;
269	struct componentname *cnp = ((struct vop_mknod_args *)v)->a_cnp;
270	struct vattr *vap = ((struct vop_mknod_args *)v)->a_vap;
271
272	if (vap->va_type != VBLK && vap->va_type != VCHR &&
273	    vap->va_type != VFIFO)
274		return EINVAL;
275
276	return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
277}
278
279/* --------------------------------------------------------------------- */
280
281int
282tmpfs_open(void *v)
283{
284	struct vnode *vp = ((struct vop_open_args *)v)->a_vp;
285	int mode = ((struct vop_open_args *)v)->a_mode;
286
287	int error;
288	struct tmpfs_node *node;
289
290	KASSERT(VOP_ISLOCKED(vp));
291
292	node = VP_TO_TMPFS_NODE(vp);
293
294	/* The file is still active but all its names have been removed
295	 * (e.g. by a "rmdir $(pwd)").  It cannot be opened any more as
296	 * it is about to die. */
297	if (node->tn_links < 1) {
298		error = ENOENT;
299		goto out;
300	}
301
302	/* If the file is marked append-only, deny write requests. */
303	if (node->tn_flags & APPEND && (mode & (FWRITE | O_APPEND)) == FWRITE)
304		error = EPERM;
305	else
306		error = 0;
307
308out:
309	KASSERT(VOP_ISLOCKED(vp));
310
311	return error;
312}
313
314/* --------------------------------------------------------------------- */
315
316int
317tmpfs_close(void *v)
318{
319	struct vnode *vp = ((struct vop_close_args *)v)->a_vp;
320
321	struct tmpfs_node *node;
322
323	KASSERT(VOP_ISLOCKED(vp));
324
325	node = VP_TO_TMPFS_NODE(vp);
326
327	if (node->tn_links > 0) {
328		/* Update node times.  No need to do it if the node has
329		 * been deleted, because it will vanish after we return. */
330		tmpfs_update(vp, NULL, NULL, NULL, UPDATE_CLOSE);
331	}
332
333	return 0;
334}
335
336/* --------------------------------------------------------------------- */
337
338int
339tmpfs_access(void *v)
340{
341	struct vnode *vp = ((struct vop_access_args *)v)->a_vp;
342	int mode = ((struct vop_access_args *)v)->a_mode;
343	kauth_cred_t cred = ((struct vop_access_args *)v)->a_cred;
344
345	int error;
346	struct tmpfs_node *node;
347
348	KASSERT(VOP_ISLOCKED(vp));
349
350	node = VP_TO_TMPFS_NODE(vp);
351
352	switch (vp->v_type) {
353	case VDIR:
354		/* FALLTHROUGH */
355	case VLNK:
356		/* FALLTHROUGH */
357	case VREG:
358		if (mode & VWRITE && vp->v_mount->mnt_flag & MNT_RDONLY) {
359			error = EROFS;
360			goto out;
361		}
362		break;
363
364	case VBLK:
365		/* FALLTHROUGH */
366	case VCHR:
367		/* FALLTHROUGH */
368	case VSOCK:
369		/* FALLTHROUGH */
370	case VFIFO:
371		break;
372
373	default:
374		error = EINVAL;
375		goto out;
376	}
377
378	if (mode & VWRITE && node->tn_flags & IMMUTABLE) {
379		error = EPERM;
380		goto out;
381	}
382
383	error = vaccess(vp->v_type, node->tn_mode, node->tn_uid,
384	    node->tn_gid, mode, cred);
385
386out:
387	KASSERT(VOP_ISLOCKED(vp));
388
389	return error;
390}
391
392/* --------------------------------------------------------------------- */
393
394int
395tmpfs_getattr(void *v)
396{
397	struct vnode *vp = ((struct vop_getattr_args *)v)->a_vp;
398	struct vattr *vap = ((struct vop_getattr_args *)v)->a_vap;
399
400	struct tmpfs_node *node;
401
402	node = VP_TO_TMPFS_NODE(vp);
403
404	VATTR_NULL(vap);
405
406	tmpfs_itimes(vp, NULL, NULL, NULL);
407
408	vap->va_type = vp->v_type;
409	vap->va_mode = node->tn_mode;
410	vap->va_nlink = node->tn_links;
411	vap->va_uid = node->tn_uid;
412	vap->va_gid = node->tn_gid;
413	vap->va_fsid = vp->v_mount->mnt_stat.f_fsidx.__fsid_val[0];
414	vap->va_fileid = node->tn_id;
415	vap->va_size = node->tn_size;
416	vap->va_blocksize = PAGE_SIZE;
417	vap->va_atime = node->tn_atime;
418	vap->va_mtime = node->tn_mtime;
419	vap->va_ctime = node->tn_ctime;
420	vap->va_birthtime = node->tn_birthtime;
421	vap->va_gen = node->tn_gen;
422	vap->va_flags = node->tn_flags;
423	vap->va_rdev = (vp->v_type == VBLK || vp->v_type == VCHR) ?
424		node->tn_spec.tn_dev.tn_rdev : VNOVAL;
425	vap->va_bytes = round_page(node->tn_size);
426	vap->va_filerev = VNOVAL;
427	vap->va_vaflags = 0;
428	vap->va_spare = VNOVAL; /* XXX */
429
430	return 0;
431}
432
433/* --------------------------------------------------------------------- */
434
435#define GOODTIME(tv)	((tv)->tv_sec != VNOVAL || (tv)->tv_nsec != VNOVAL)
436/* XXX Should this operation be atomic?  I think it should, but code in
437 * XXX other places (e.g., ufs) doesn't seem to be... */
438int
439tmpfs_setattr(void *v)
440{
441	struct vnode *vp = ((struct vop_setattr_args *)v)->a_vp;
442	struct vattr *vap = ((struct vop_setattr_args *)v)->a_vap;
443	kauth_cred_t cred = ((struct vop_setattr_args *)v)->a_cred;
444	struct lwp *l = curlwp;
445
446	int error;
447
448	KASSERT(VOP_ISLOCKED(vp));
449
450	error = 0;
451
452	/* Abort if any unsettable attribute is given. */
453	if (vap->va_type != VNON ||
454	    vap->va_nlink != VNOVAL ||
455	    vap->va_fsid != VNOVAL ||
456	    vap->va_fileid != VNOVAL ||
457	    vap->va_blocksize != VNOVAL ||
458	    GOODTIME(&vap->va_ctime) ||
459	    vap->va_gen != VNOVAL ||
460	    vap->va_rdev != VNOVAL ||
461	    vap->va_bytes != VNOVAL)
462		error = EINVAL;
463
464	if (error == 0 && (vap->va_flags != VNOVAL))
465		error = tmpfs_chflags(vp, vap->va_flags, cred, l);
466
467	if (error == 0 && (vap->va_size != VNOVAL))
468		error = tmpfs_chsize(vp, vap->va_size, cred, l);
469
470	if (error == 0 && (vap->va_uid != VNOVAL || vap->va_gid != VNOVAL))
471		error = tmpfs_chown(vp, vap->va_uid, vap->va_gid, cred, l);
472
473	if (error == 0 && (vap->va_mode != VNOVAL))
474		error = tmpfs_chmod(vp, vap->va_mode, cred, l);
475
476	if (error == 0 && (GOODTIME(&vap->va_atime) || GOODTIME(&vap->va_mtime)
477	    || GOODTIME(&vap->va_birthtime)))
478		if ((error = tmpfs_chtimes(vp, &vap->va_atime, &vap->va_mtime,
479		    &vap->va_birthtime, vap->va_vaflags, cred, l)) == 0)
480			return 0;
481
482	/* Update the node times.  We give preference to the error codes
483	 * generated by this function rather than the ones that may arise
484	 * from tmpfs_update. */
485	tmpfs_update(vp, NULL, NULL, NULL, 0);
486
487	KASSERT(VOP_ISLOCKED(vp));
488
489	return error;
490}
491
492/* --------------------------------------------------------------------- */
493
494int
495tmpfs_read(void *v)
496{
497	struct vnode *vp = ((struct vop_read_args *)v)->a_vp;
498	struct uio *uio = ((struct vop_read_args *)v)->a_uio;
499	int ioflag = ((struct vop_read_args *)v)->a_ioflag;
500
501	int error;
502	struct tmpfs_node *node;
503	struct uvm_object *uobj;
504
505	KASSERT(VOP_ISLOCKED(vp));
506
507	node = VP_TO_TMPFS_NODE(vp);
508
509	if (vp->v_type != VREG) {
510		error = EISDIR;
511		goto out;
512	}
513
514	if (uio->uio_offset < 0) {
515		error = EINVAL;
516		goto out;
517	}
518
519	node->tn_status |= TMPFS_NODE_ACCESSED;
520
521	uobj = node->tn_spec.tn_reg.tn_aobj;
522	error = 0;
523	while (error == 0 && uio->uio_resid > 0) {
524		vsize_t len;
525
526		if (node->tn_size <= uio->uio_offset)
527			break;
528
529		len = MIN(node->tn_size - uio->uio_offset, uio->uio_resid);
530		if (len == 0)
531			break;
532
533		error = ubc_uiomove(uobj, uio, len, IO_ADV_DECODE(ioflag),
534		    UBC_READ | UBC_PARTIALOK | UBC_UNMAP_FLAG(vp));
535	}
536
537out:
538	KASSERT(VOP_ISLOCKED(vp));
539
540	return error;
541}
542
543/* --------------------------------------------------------------------- */
544
545int
546tmpfs_write(void *v)
547{
548	struct vnode *vp = ((struct vop_write_args *)v)->a_vp;
549	struct uio *uio = ((struct vop_write_args *)v)->a_uio;
550	int ioflag = ((struct vop_write_args *)v)->a_ioflag;
551
552	bool extended;
553	int error;
554	off_t oldsize;
555	struct tmpfs_node *node;
556	struct uvm_object *uobj;
557
558	KASSERT(VOP_ISLOCKED(vp));
559
560	node = VP_TO_TMPFS_NODE(vp);
561	oldsize = node->tn_size;
562
563	if (uio->uio_offset < 0 || vp->v_type != VREG) {
564		error = EINVAL;
565		goto out;
566	}
567
568	if (uio->uio_resid == 0) {
569		error = 0;
570		goto out;
571	}
572
573	if (ioflag & IO_APPEND)
574		uio->uio_offset = node->tn_size;
575
576	extended = uio->uio_offset + uio->uio_resid > node->tn_size;
577	if (extended) {
578		error = tmpfs_reg_resize(vp, uio->uio_offset + uio->uio_resid);
579		if (error != 0)
580			goto out;
581	}
582
583	uobj = node->tn_spec.tn_reg.tn_aobj;
584	error = 0;
585	while (error == 0 && uio->uio_resid > 0) {
586		vsize_t len;
587
588		len = MIN(node->tn_size - uio->uio_offset, uio->uio_resid);
589		if (len == 0)
590			break;
591
592		error = ubc_uiomove(uobj, uio, len, IO_ADV_DECODE(ioflag),
593		    UBC_WRITE | UBC_UNMAP_FLAG(vp));
594	}
595
596	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
597	    (extended ? TMPFS_NODE_CHANGED : 0);
598
599	if (error != 0)
600		(void)tmpfs_reg_resize(vp, oldsize);
601
602	VN_KNOTE(vp, NOTE_WRITE);
603
604out:
605	KASSERT(VOP_ISLOCKED(vp));
606	KASSERT(IMPLIES(error == 0, uio->uio_resid == 0));
607	KASSERT(IMPLIES(error != 0, oldsize == node->tn_size));
608
609	return error;
610}
611
612/* --------------------------------------------------------------------- */
613
614int
615tmpfs_fsync(void *v)
616{
617	struct vnode *vp = ((struct vop_fsync_args *)v)->a_vp;
618
619	KASSERT(VOP_ISLOCKED(vp));
620
621	tmpfs_update(vp, NULL, NULL, NULL, 0);
622
623	return 0;
624}
625
626/* --------------------------------------------------------------------- */
627
628int
629tmpfs_remove(void *v)
630{
631	struct vnode *dvp = ((struct vop_remove_args *)v)->a_dvp;
632	struct vnode *vp = ((struct vop_remove_args *)v)->a_vp;
633	struct componentname *cnp = (((struct vop_remove_args *)v)->a_cnp);
634
635	int error;
636	struct tmpfs_dirent *de;
637	struct tmpfs_mount *tmp;
638	struct tmpfs_node *dnode;
639	struct tmpfs_node *node;
640
641	KASSERT(VOP_ISLOCKED(dvp));
642	KASSERT(VOP_ISLOCKED(vp));
643
644	if (vp->v_type == VDIR) {
645		error = EPERM;
646		goto out;
647	}
648
649	dnode = VP_TO_TMPFS_DIR(dvp);
650	node = VP_TO_TMPFS_NODE(vp);
651	tmp = VFS_TO_TMPFS(vp->v_mount);
652	de = tmpfs_dir_lookup(dnode, cnp);
653	if (de == NULL) {
654		error = ENOENT;
655		goto out;
656	}
657	KASSERT(de->td_node == node);
658
659	/* Files marked as immutable or append-only cannot be deleted. */
660	if (node->tn_flags & (IMMUTABLE | APPEND)) {
661		error = EPERM;
662		goto out;
663	}
664
665	/* Remove the entry from the directory; as it is a file, we do not
666	 * have to change the number of hard links of the directory. */
667	tmpfs_dir_detach(dvp, de);
668
669	/* Free the directory entry we just deleted.  Note that the node
670	 * referred by it will not be removed until the vnode is really
671	 * reclaimed. */
672	tmpfs_free_dirent(tmp, de, true);
673
674	error = 0;
675
676out:
677	vput(vp);
678	if (dvp == vp)
679		vrele(dvp);
680	else
681		vput(dvp);
682
683	return error;
684}
685
686/* --------------------------------------------------------------------- */
687
688int
689tmpfs_link(void *v)
690{
691	struct vnode *dvp = ((struct vop_link_args *)v)->a_dvp;
692	struct vnode *vp = ((struct vop_link_args *)v)->a_vp;
693	struct componentname *cnp = ((struct vop_link_args *)v)->a_cnp;
694
695	int error;
696	struct tmpfs_dirent *de;
697	struct tmpfs_node *dnode;
698	struct tmpfs_node *node;
699
700	KASSERT(VOP_ISLOCKED(dvp));
701	KASSERT(cnp->cn_flags & HASBUF);
702	KASSERT(dvp != vp); /* XXX When can this be false? */
703
704	dnode = VP_TO_TMPFS_DIR(dvp);
705	node = VP_TO_TMPFS_NODE(vp);
706
707	/* Lock vp because we will need to run tmpfs_update over it, which
708	 * needs the vnode to be locked. */
709	error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
710	if (error != 0)
711		goto out1;
712
713	/* XXX: Why aren't the following two tests done by the caller? */
714
715	/* Hard links of directories are forbidden. */
716	if (vp->v_type == VDIR) {
717		error = EPERM;
718		goto out;
719	}
720
721	/* Cannot create cross-device links. */
722	if (dvp->v_mount != vp->v_mount) {
723		error = EXDEV;
724		goto out;
725	}
726
727	/* Ensure that we do not overflow the maximum number of links imposed
728	 * by the system. */
729	KASSERT(node->tn_links <= LINK_MAX);
730	if (node->tn_links == LINK_MAX) {
731		error = EMLINK;
732		goto out;
733	}
734
735	/* We cannot create links of files marked immutable or append-only. */
736	if (node->tn_flags & (IMMUTABLE | APPEND)) {
737		error = EPERM;
738		goto out;
739	}
740
741	/* Allocate a new directory entry to represent the node. */
742	error = tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), node,
743	    cnp->cn_nameptr, cnp->cn_namelen, &de);
744	if (error != 0)
745		goto out;
746
747	/* Insert the new directory entry into the appropriate directory. */
748	tmpfs_dir_attach(dvp, de);
749
750	/* vp link count has changed, so update node times. */
751	node->tn_status |= TMPFS_NODE_CHANGED;
752	tmpfs_update(vp, NULL, NULL, NULL, 0);
753
754	error = 0;
755
756out:
757	VOP_UNLOCK(vp, 0);
758out1:
759	PNBUF_PUT(cnp->cn_pnbuf);
760
761	vput(dvp);
762
763	return error;
764}
765
766/* --------------------------------------------------------------------- */
767
768int
769tmpfs_rename(void *v)
770{
771	struct vnode *fdvp = ((struct vop_rename_args *)v)->a_fdvp;
772	struct vnode *fvp = ((struct vop_rename_args *)v)->a_fvp;
773	struct componentname *fcnp = ((struct vop_rename_args *)v)->a_fcnp;
774	struct vnode *tdvp = ((struct vop_rename_args *)v)->a_tdvp;
775	struct vnode *tvp = ((struct vop_rename_args *)v)->a_tvp;
776	struct componentname *tcnp = ((struct vop_rename_args *)v)->a_tcnp;
777
778	char *newname;
779	int error;
780	struct tmpfs_dirent *de, *de2;
781	struct tmpfs_mount *tmp;
782	struct tmpfs_node *fdnode;
783	struct tmpfs_node *fnode;
784	struct tmpfs_node *tnode;
785	struct tmpfs_node *tdnode;
786	size_t namelen;
787
788	KASSERT(VOP_ISLOCKED(tdvp));
789	KASSERT(IMPLIES(tvp != NULL, VOP_ISLOCKED(tvp) == LK_EXCLUSIVE));
790	KASSERT(fcnp->cn_flags & HASBUF);
791	KASSERT(tcnp->cn_flags & HASBUF);
792
793	newname = NULL;
794	namelen = 0;
795	tmp = NULL;
796
797	/* Disallow cross-device renames. */
798	if (fvp->v_mount != tdvp->v_mount ||
799	    (tvp != NULL && fvp->v_mount != tvp->v_mount)) {
800		error = EXDEV;
801		goto out_unlocked;
802	}
803
804	fnode = VP_TO_TMPFS_NODE(fvp);
805	fdnode = VP_TO_TMPFS_DIR(fdvp);
806	tnode = (tvp == NULL) ? NULL : VP_TO_TMPFS_NODE(tvp);
807	tdnode = VP_TO_TMPFS_DIR(tdvp);
808	tmp = VFS_TO_TMPFS(tdvp->v_mount);
809
810	/* If we need to move the directory between entries, lock the
811	 * source so that we can safely operate on it. */
812
813	/* XXX: this is a potential locking order violation! */
814	if (fdnode != tdnode) {
815		error = vn_lock(fdvp, LK_EXCLUSIVE | LK_RETRY);
816		if (error != 0)
817			goto out_unlocked;
818	}
819
820	de = tmpfs_dir_lookup(fdnode, fcnp);
821	if (de == NULL) {
822		error = ENOENT;
823		goto out;
824	}
825	KASSERT(de->td_node == fnode);
826
827	/* If source and target are the same file, there is nothing to do. */
828	if (fvp == tvp) {
829		error = 0;
830		goto out;
831	}
832
833	/* Avoid manipulating '.' and '..' entries. */
834	if (de == NULL) {
835		KASSERT(fvp->v_type == VDIR);
836		error = EINVAL;
837		goto out;
838	}
839	KASSERT(de->td_node == fnode);
840
841	/* If replacing an existing entry, ensure we can do the operation. */
842	if (tvp != NULL) {
843		KASSERT(tnode != NULL);
844		if (fnode->tn_type == VDIR && tnode->tn_type == VDIR) {
845			if (tnode->tn_size > 0) {
846				error = ENOTEMPTY;
847				goto out;
848			}
849		} else if (fnode->tn_type == VDIR && tnode->tn_type != VDIR) {
850			error = ENOTDIR;
851			goto out;
852		} else if (fnode->tn_type != VDIR && tnode->tn_type == VDIR) {
853			error = EISDIR;
854			goto out;
855		} else {
856			KASSERT(fnode->tn_type != VDIR &&
857			        tnode->tn_type != VDIR);
858		}
859	}
860
861	/* Ensure that we have enough memory to hold the new name, if it
862	 * has to be changed. */
863	namelen = tcnp->cn_namelen;
864	if (fcnp->cn_namelen != tcnp->cn_namelen ||
865	    memcmp(fcnp->cn_nameptr, tcnp->cn_nameptr, fcnp->cn_namelen) != 0) {
866		newname = tmpfs_str_pool_get(&tmp->tm_str_pool, namelen, 0);
867		if (newname == NULL) {
868			error = ENOSPC;
869			goto out;
870		}
871	}
872
873	/* If the node is being moved to another directory, we have to do
874	 * the move. */
875	if (fdnode != tdnode) {
876		/* In case we are moving a directory, we have to adjust its
877		 * parent to point to the new parent. */
878		if (de->td_node->tn_type == VDIR) {
879			struct tmpfs_node *n;
880
881			/* Ensure the target directory is not a child of the
882			 * directory being moved.  Otherwise, we'd end up
883			 * with stale nodes. */
884			n = tdnode;
885			while (n != n->tn_spec.tn_dir.tn_parent) {
886				if (n == fnode) {
887					error = EINVAL;
888					goto out;
889				}
890				n = n->tn_spec.tn_dir.tn_parent;
891			}
892
893			/* Adjust the parent pointer. */
894			TMPFS_VALIDATE_DIR(fnode);
895			de->td_node->tn_spec.tn_dir.tn_parent = tdnode;
896
897			/* As a result of changing the target of the '..'
898			 * entry, the link count of the source and target
899			 * directories has to be adjusted. */
900			fdnode->tn_links--;
901			tdnode->tn_links++;
902		}
903
904		/* Do the move: just remove the entry from the source directory
905		 * and insert it into the target one. */
906		tmpfs_dir_detach(fdvp, de);
907		tmpfs_dir_attach(tdvp, de);
908
909		/* Notify listeners of fdvp about the change in the directory.
910		 * We can do it at this point because we aren't touching fdvp
911		 * any more below. */
912		VN_KNOTE(fdvp, NOTE_WRITE);
913	}
914
915	/* If we are overwriting an entry, we have to remove the old one
916	 * from the target directory. */
917	if (tvp != NULL) {
918		KASSERT(tnode != NULL);
919
920		/* Remove the old entry from the target directory.
921		 * Note! This relies on tmpfs_dir_attach() putting the new
922		 * node on the end of the target's node list. */
923		de2 = tmpfs_dir_lookup(tdnode, tcnp);
924		KASSERT(de2 != NULL);
925		KASSERT(de2->td_node == tnode);
926		tmpfs_dir_detach(tdvp, de2);
927
928		/* Free the directory entry we just deleted.  Note that the
929		 * node referred by it will not be removed until the vnode is
930		 * really reclaimed. */
931		tmpfs_free_dirent(VFS_TO_TMPFS(tvp->v_mount), de2, true);
932	}
933
934	/* If the name has changed, we need to make it effective by changing
935	 * it in the directory entry. */
936	if (newname != NULL) {
937		KASSERT(tcnp->cn_namelen < MAXNAMLEN);
938		KASSERT(tcnp->cn_namelen < 0xffff);
939
940		tmpfs_str_pool_put(&tmp->tm_str_pool, de->td_name,
941		    de->td_namelen);
942		de->td_namelen = (uint16_t)namelen;
943		memcpy(newname, tcnp->cn_nameptr, namelen);
944		de->td_name = newname;
945		newname = NULL;
946
947		fnode->tn_status |= TMPFS_NODE_CHANGED;
948		tdnode->tn_status |= TMPFS_NODE_MODIFIED;
949	}
950
951	/* Notify listeners of tdvp about the change in the directory (either
952	 * because a new entry was added or because one was removed) and
953	 * listeners of fvp about the rename. */
954	VN_KNOTE(tdvp, NOTE_WRITE);
955	VN_KNOTE(fvp, NOTE_RENAME);
956
957	error = 0;
958
959 out:
960	if (fdnode != tdnode)
961		VOP_UNLOCK(fdvp, 0);
962
963 out_unlocked:
964	/* Release target nodes. */
965	if (tdvp == tvp)
966		vrele(tdvp);
967	else
968		vput(tdvp);
969	if (tvp != NULL)
970		vput(tvp);
971
972	/* Release source nodes. */
973	vrele(fdvp);
974	vrele(fvp);
975
976	if (newname != NULL)
977		tmpfs_str_pool_put(&tmp->tm_str_pool, newname, namelen);
978
979	return error;
980}
981
982/* --------------------------------------------------------------------- */
983
984int
985tmpfs_mkdir(void *v)
986{
987	struct vnode *dvp = ((struct vop_mkdir_args *)v)->a_dvp;
988	struct vnode **vpp = ((struct vop_mkdir_args *)v)->a_vpp;
989	struct componentname *cnp = ((struct vop_mkdir_args *)v)->a_cnp;
990	struct vattr *vap = ((struct vop_mkdir_args *)v)->a_vap;
991
992	KASSERT(vap->va_type == VDIR);
993
994	return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
995}
996
997/* --------------------------------------------------------------------- */
998
999int
1000tmpfs_rmdir(void *v)
1001{
1002	struct vnode *dvp = ((struct vop_rmdir_args *)v)->a_dvp;
1003	struct vnode *vp = ((struct vop_rmdir_args *)v)->a_vp;
1004	struct componentname *cnp = ((struct vop_rmdir_args *)v)->a_cnp;
1005
1006	int error;
1007	struct tmpfs_dirent *de;
1008	struct tmpfs_mount *tmp;
1009	struct tmpfs_node *dnode;
1010	struct tmpfs_node *node;
1011
1012	KASSERT(VOP_ISLOCKED(dvp));
1013	KASSERT(VOP_ISLOCKED(vp));
1014
1015	tmp = VFS_TO_TMPFS(dvp->v_mount);
1016	dnode = VP_TO_TMPFS_DIR(dvp);
1017	node = VP_TO_TMPFS_DIR(vp);
1018	error = 0;
1019
1020	/* Directories with more than two entries ('.' and '..') cannot be
1021	 * removed. */
1022	if (node->tn_size > 0) {
1023		error = ENOTEMPTY;
1024		goto out;
1025	}
1026
1027	/* This invariant holds only if we are not trying to remove "..".
1028	 * We checked for that above so this is safe now. */
1029	KASSERT(node->tn_spec.tn_dir.tn_parent == dnode);
1030
1031	/* Get the directory entry associated with node (vp). */
1032	de = tmpfs_dir_lookup(dnode, cnp);
1033	if (de == NULL) {
1034		error = ENOENT;
1035		goto out;
1036	}
1037	KASSERT(de->td_node == node);
1038
1039	/* Check flags to see if we are allowed to remove the directory. */
1040	if (dnode->tn_flags & APPEND || node->tn_flags & (IMMUTABLE | APPEND)) {
1041		error = EPERM;
1042		goto out;
1043	}
1044
1045	/* Detach the directory entry from the directory (dnode). */
1046	tmpfs_dir_detach(dvp, de);
1047
1048	node->tn_links--;
1049	node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
1050	    TMPFS_NODE_MODIFIED;
1051	node->tn_spec.tn_dir.tn_parent->tn_links--;
1052	node->tn_spec.tn_dir.tn_parent->tn_status |= TMPFS_NODE_ACCESSED | \
1053	    TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1054
1055	/* Release the parent. */
1056	cache_purge(dvp); /* XXX Is this needed? */
1057
1058	/* Free the directory entry we just deleted.  Note that the node
1059	 * referred by it will not be removed until the vnode is really
1060	 * reclaimed. */
1061	tmpfs_free_dirent(tmp, de, true);
1062
1063	KASSERT(node->tn_links == 0);
1064 out:
1065	/* Release the nodes. */
1066	vput(dvp);
1067	vput(vp);
1068
1069	return error;
1070}
1071
1072/* --------------------------------------------------------------------- */
1073
1074int
1075tmpfs_symlink(void *v)
1076{
1077	struct vnode *dvp = ((struct vop_symlink_args *)v)->a_dvp;
1078	struct vnode **vpp = ((struct vop_symlink_args *)v)->a_vpp;
1079	struct componentname *cnp = ((struct vop_symlink_args *)v)->a_cnp;
1080	struct vattr *vap = ((struct vop_symlink_args *)v)->a_vap;
1081	char *target = ((struct vop_symlink_args *)v)->a_target;
1082
1083	KASSERT(vap->va_type == VLNK);
1084
1085	return tmpfs_alloc_file(dvp, vpp, vap, cnp, target);
1086}
1087
1088/* --------------------------------------------------------------------- */
1089
1090int
1091tmpfs_readdir(void *v)
1092{
1093	struct vnode *vp = ((struct vop_readdir_args *)v)->a_vp;
1094	struct uio *uio = ((struct vop_readdir_args *)v)->a_uio;
1095	int *eofflag = ((struct vop_readdir_args *)v)->a_eofflag;
1096	off_t **cookies = ((struct vop_readdir_args *)v)->a_cookies;
1097	int *ncookies = ((struct vop_readdir_args *)v)->a_ncookies;
1098
1099	int error;
1100	off_t startoff;
1101	off_t cnt;
1102	struct tmpfs_node *node;
1103
1104	KASSERT(VOP_ISLOCKED(vp));
1105
1106	/* This operation only makes sense on directory nodes. */
1107	if (vp->v_type != VDIR) {
1108		error = ENOTDIR;
1109		goto out;
1110	}
1111
1112	node = VP_TO_TMPFS_DIR(vp);
1113
1114	startoff = uio->uio_offset;
1115
1116	cnt = 0;
1117	if (uio->uio_offset == TMPFS_DIRCOOKIE_DOT) {
1118		error = tmpfs_dir_getdotdent(node, uio);
1119		if (error == -1) {
1120			error = 0;
1121			goto outok;
1122		} else if (error != 0)
1123			goto outok;
1124		cnt++;
1125	}
1126
1127	if (uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT) {
1128		error = tmpfs_dir_getdotdotdent(node, uio);
1129		if (error == -1) {
1130			error = 0;
1131			goto outok;
1132		} else if (error != 0)
1133			goto outok;
1134		cnt++;
1135	}
1136
1137	error = tmpfs_dir_getdents(node, uio, &cnt);
1138	if (error == -1)
1139		error = 0;
1140	KASSERT(error >= 0);
1141
1142outok:
1143	/* This label assumes that startoff has been
1144	 * initialized.  If the compiler didn't spit out warnings, we'd
1145	 * simply make this one be 'out' and drop 'outok'. */
1146
1147	if (eofflag != NULL)
1148		*eofflag =
1149		    (error == 0 && uio->uio_offset == TMPFS_DIRCOOKIE_EOF);
1150
1151	/* Update NFS-related variables. */
1152	if (error == 0 && cookies != NULL && ncookies != NULL) {
1153		off_t i;
1154		off_t off = startoff;
1155		struct tmpfs_dirent *de = NULL;
1156
1157		*ncookies = cnt;
1158		*cookies = malloc(cnt * sizeof(off_t), M_TEMP, M_WAITOK);
1159
1160		for (i = 0; i < cnt; i++) {
1161			KASSERT(off != TMPFS_DIRCOOKIE_EOF);
1162			if (off == TMPFS_DIRCOOKIE_DOT) {
1163				off = TMPFS_DIRCOOKIE_DOTDOT;
1164			} else {
1165				if (off == TMPFS_DIRCOOKIE_DOTDOT) {
1166					de = TAILQ_FIRST(&node->tn_spec.
1167					    tn_dir.tn_dir);
1168				} else if (de != NULL) {
1169					de = TAILQ_NEXT(de, td_entries);
1170				} else {
1171					de = tmpfs_dir_lookupbycookie(node,
1172					    off);
1173					KASSERT(de != NULL);
1174					de = TAILQ_NEXT(de, td_entries);
1175				}
1176				if (de == NULL) {
1177					off = TMPFS_DIRCOOKIE_EOF;
1178				} else {
1179					off = tmpfs_dircookie(de);
1180				}
1181			}
1182
1183			(*cookies)[i] = off;
1184		}
1185		KASSERT(uio->uio_offset == off);
1186	}
1187
1188out:
1189	KASSERT(VOP_ISLOCKED(vp));
1190
1191	return error;
1192}
1193
1194/* --------------------------------------------------------------------- */
1195
1196int
1197tmpfs_readlink(void *v)
1198{
1199	struct vnode *vp = ((struct vop_readlink_args *)v)->a_vp;
1200	struct uio *uio = ((struct vop_readlink_args *)v)->a_uio;
1201
1202	int error;
1203	struct tmpfs_node *node;
1204
1205	KASSERT(VOP_ISLOCKED(vp));
1206	KASSERT(uio->uio_offset == 0);
1207	KASSERT(vp->v_type == VLNK);
1208
1209	node = VP_TO_TMPFS_NODE(vp);
1210
1211	error = uiomove(node->tn_spec.tn_lnk.tn_link,
1212	    MIN(node->tn_size, uio->uio_resid), uio);
1213	node->tn_status |= TMPFS_NODE_ACCESSED;
1214
1215	KASSERT(VOP_ISLOCKED(vp));
1216
1217	return error;
1218}
1219
1220/* --------------------------------------------------------------------- */
1221
1222int
1223tmpfs_inactive(void *v)
1224{
1225	struct vnode *vp = ((struct vop_inactive_args *)v)->a_vp;
1226
1227	struct tmpfs_node *node;
1228
1229	KASSERT(VOP_ISLOCKED(vp));
1230
1231	node = VP_TO_TMPFS_NODE(vp);
1232	*((struct vop_inactive_args *)v)->a_recycle = (node->tn_links == 0);
1233	VOP_UNLOCK(vp, 0);
1234
1235	return 0;
1236}
1237
1238/* --------------------------------------------------------------------- */
1239
1240int
1241tmpfs_reclaim(void *v)
1242{
1243	struct vnode *vp = ((struct vop_reclaim_args *)v)->a_vp;
1244
1245	struct tmpfs_mount *tmp;
1246	struct tmpfs_node *node;
1247
1248	node = VP_TO_TMPFS_NODE(vp);
1249	tmp = VFS_TO_TMPFS(vp->v_mount);
1250
1251	cache_purge(vp);
1252	tmpfs_free_vp(vp);
1253
1254	/* If the node referenced by this vnode was deleted by the user,
1255	 * we must free its associated data structures (now that the vnode
1256	 * is being reclaimed). */
1257	if (node->tn_links == 0)
1258		tmpfs_free_node(tmp, node);
1259
1260	KASSERT(vp->v_data == NULL);
1261
1262	return 0;
1263}
1264
1265/* --------------------------------------------------------------------- */
1266
1267int
1268tmpfs_print(void *v)
1269{
1270	struct vnode *vp = ((struct vop_print_args *)v)->a_vp;
1271
1272	struct tmpfs_node *node;
1273
1274	node = VP_TO_TMPFS_NODE(vp);
1275
1276	printf("tag VT_TMPFS, tmpfs_node %p, flags 0x%x, links %d\n",
1277	    node, node->tn_flags, node->tn_links);
1278	printf("\tmode 0%o, owner %d, group %d, size %" PRIdMAX
1279	    ", status 0x%x\n",
1280	    node->tn_mode, node->tn_uid, node->tn_gid,
1281	    (uintmax_t)node->tn_size, node->tn_status);
1282	if (vp->v_type == VFIFO)
1283		fifo_printinfo(vp);
1284	printf("\n");
1285
1286	return 0;
1287}
1288
1289/* --------------------------------------------------------------------- */
1290
1291int
1292tmpfs_pathconf(void *v)
1293{
1294	int name = ((struct vop_pathconf_args *)v)->a_name;
1295	register_t *retval = ((struct vop_pathconf_args *)v)->a_retval;
1296
1297	int error;
1298
1299	error = 0;
1300
1301	switch (name) {
1302	case _PC_LINK_MAX:
1303		*retval = LINK_MAX;
1304		break;
1305
1306	case _PC_NAME_MAX:
1307		*retval = NAME_MAX;
1308		break;
1309
1310	case _PC_PATH_MAX:
1311		*retval = PATH_MAX;
1312		break;
1313
1314	case _PC_PIPE_BUF:
1315		*retval = PIPE_BUF;
1316		break;
1317
1318	case _PC_CHOWN_RESTRICTED:
1319		*retval = 1;
1320		break;
1321
1322	case _PC_NO_TRUNC:
1323		*retval = 1;
1324		break;
1325
1326	case _PC_SYNC_IO:
1327		*retval = 1;
1328		break;
1329
1330	case _PC_FILESIZEBITS:
1331		*retval = 0; /* XXX Don't know which value should I return. */
1332		break;
1333
1334	default:
1335		error = EINVAL;
1336	}
1337
1338	return error;
1339}
1340
1341/* --------------------------------------------------------------------- */
1342
1343int
1344tmpfs_advlock(void *v)
1345{
1346	struct vnode *vp = ((struct vop_advlock_args *)v)->a_vp;
1347
1348	struct tmpfs_node *node;
1349
1350	node = VP_TO_TMPFS_NODE(vp);
1351
1352	return lf_advlock(v, &node->tn_lockf, node->tn_size);
1353}
1354
1355/* --------------------------------------------------------------------- */
1356
1357int
1358tmpfs_getpages(void *v)
1359{
1360	struct vnode *vp = ((struct vop_getpages_args *)v)->a_vp;
1361	voff_t offset = ((struct vop_getpages_args *)v)->a_offset;
1362	struct vm_page **m = ((struct vop_getpages_args *)v)->a_m;
1363	int *count = ((struct vop_getpages_args *)v)->a_count;
1364	int centeridx = ((struct vop_getpages_args *)v)->a_centeridx;
1365	vm_prot_t access_type = ((struct vop_getpages_args *)v)->a_access_type;
1366	int advice = ((struct vop_getpages_args *)v)->a_advice;
1367	int flags = ((struct vop_getpages_args *)v)->a_flags;
1368
1369	int error;
1370	int i;
1371	struct tmpfs_node *node;
1372	struct uvm_object *uobj;
1373	int npages = *count;
1374
1375	KASSERT(vp->v_type == VREG);
1376	KASSERT(mutex_owned(&vp->v_interlock));
1377
1378	node = VP_TO_TMPFS_NODE(vp);
1379	uobj = node->tn_spec.tn_reg.tn_aobj;
1380
1381	/* We currently don't rely on PGO_PASTEOF. */
1382
1383	if (vp->v_size <= offset + (centeridx << PAGE_SHIFT)) {
1384		if ((flags & PGO_LOCKED) == 0)
1385			mutex_exit(&vp->v_interlock);
1386		return EINVAL;
1387	}
1388
1389	if (vp->v_size < offset + (npages << PAGE_SHIFT)) {
1390		npages = (round_page(vp->v_size) - offset) >> PAGE_SHIFT;
1391	}
1392
1393	if ((flags & PGO_LOCKED) != 0)
1394		return EBUSY;
1395
1396	if ((flags & PGO_NOTIMESTAMP) == 0) {
1397		if ((vp->v_mount->mnt_flag & MNT_NOATIME) == 0)
1398			node->tn_status |= TMPFS_NODE_ACCESSED;
1399
1400		if ((access_type & VM_PROT_WRITE) != 0)
1401			node->tn_status |= TMPFS_NODE_MODIFIED;
1402	}
1403
1404	mutex_exit(&vp->v_interlock);
1405
1406	/*
1407	 * Make sure that the array on which we will store the
1408	 * gotten pages is clean.  Otherwise uao_get (pointed to by
1409	 * the pgo_get below) gets confused and does not return the
1410	 * appropriate pages.
1411	 *
1412	 * XXX This shall be revisited when kern/32166 is addressed
1413	 * because the loop to clean m[i] will most likely be redundant
1414	 * as well as the PGO_ALLPAGES flag.
1415	 */
1416	if (m != NULL)
1417		for (i = 0; i < npages; i++)
1418			m[i] = NULL;
1419	mutex_enter(&uobj->vmobjlock);
1420	error = (*uobj->pgops->pgo_get)(uobj, offset, m, &npages, centeridx,
1421	    access_type, advice, flags | PGO_ALLPAGES);
1422#if defined(DEBUG)
1423	{
1424		/* Make sure that all the pages we return are valid. */
1425		int dbgi;
1426		if (error == 0 && m != NULL)
1427			for (dbgi = 0; dbgi < npages; dbgi++)
1428				KASSERT(m[dbgi] != NULL);
1429	}
1430#endif
1431
1432	return error;
1433}
1434
1435/* --------------------------------------------------------------------- */
1436
1437int
1438tmpfs_putpages(void *v)
1439{
1440	struct vnode *vp = ((struct vop_putpages_args *)v)->a_vp;
1441	voff_t offlo = ((struct vop_putpages_args *)v)->a_offlo;
1442	voff_t offhi = ((struct vop_putpages_args *)v)->a_offhi;
1443	int flags = ((struct vop_putpages_args *)v)->a_flags;
1444
1445	int error;
1446	struct tmpfs_node *node;
1447	struct uvm_object *uobj;
1448
1449	KASSERT(mutex_owned(&vp->v_interlock));
1450
1451	node = VP_TO_TMPFS_NODE(vp);
1452
1453	if (vp->v_type != VREG) {
1454		mutex_exit(&vp->v_interlock);
1455		return 0;
1456	}
1457
1458	uobj = node->tn_spec.tn_reg.tn_aobj;
1459	mutex_exit(&vp->v_interlock);
1460
1461	mutex_enter(&uobj->vmobjlock);
1462	error = (*uobj->pgops->pgo_put)(uobj, offlo, offhi, flags);
1463
1464	/* XXX mtime */
1465
1466	return error;
1467}
1468