xref: /freebsd/sys/fs/pseudofs/pseudofs_vnops.c (revision 9bd497b8354567454e075076d40c996e21bd6095)
1 /*-
2  * Copyright (c) 2001 Dag-Erling Co�dan Sm�rgrav
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer
10  *    in this position and unchanged.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  * 3. The name of the author may not be used to endorse or promote products
15  *    derived from this software without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
18  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
19  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
20  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
21  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
22  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
26  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27  */
28 
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
31 
32 #include "opt_pseudofs.h"
33 
34 #include <sys/param.h>
35 #include <sys/kernel.h>
36 #include <sys/systm.h>
37 #include <sys/ctype.h>
38 #include <sys/dirent.h>
39 #include <sys/fcntl.h>
40 #include <sys/limits.h>
41 #include <sys/lock.h>
42 #include <sys/malloc.h>
43 #include <sys/mount.h>
44 #include <sys/mutex.h>
45 #include <sys/namei.h>
46 #include <sys/proc.h>
47 #include <sys/sbuf.h>
48 #include <sys/sx.h>
49 #include <sys/sysctl.h>
50 #include <sys/vnode.h>
51 
52 #include <fs/pseudofs/pseudofs.h>
53 #include <fs/pseudofs/pseudofs_internal.h>
54 
55 #define KASSERT_PN_IS_DIR(pn)						\
56 	KASSERT((pn)->pn_type == pfstype_root ||			\
57 	    (pn)->pn_type == pfstype_dir ||				\
58 	    (pn)->pn_type == pfstype_procdir,				\
59 	    ("%s(): VDIR vnode refers to non-directory pfs_node", __func__))
60 
61 #define KASSERT_PN_IS_FILE(pn)						\
62 	KASSERT((pn)->pn_type == pfstype_file,				\
63 	    ("%s(): VREG vnode refers to non-file pfs_node", __func__))
64 
65 #define KASSERT_PN_IS_LINK(pn)						\
66 	KASSERT((pn)->pn_type == pfstype_symlink,			\
67 	    ("%s(): VLNK vnode refers to non-link pfs_node", __func__))
68 
69 /*
70  * Returns the fileno, adjusted for target pid
71  */
72 static uint32_t
73 pn_fileno(struct pfs_node *pn, pid_t pid)
74 {
75 
76 	KASSERT(pn->pn_fileno > 0,
77 	    ("%s(): no fileno allocated", __func__));
78 	if (pid != NO_PID)
79 		return (pn->pn_fileno * NO_PID + pid);
80 	return (pn->pn_fileno);
81 }
82 
83 /*
84  * Returns non-zero if given file is visible to given thread.
85  */
86 static int
87 pfs_visible_proc(struct thread *td, struct pfs_node *pn, struct proc *proc)
88 {
89 	int visible;
90 
91 	if (proc == NULL)
92 		return (0);
93 
94 	PROC_LOCK_ASSERT(proc, MA_OWNED);
95 
96 	visible = ((proc->p_flag & P_WEXIT) == 0);
97 	if (visible)
98 		visible = (p_cansee(td, proc) == 0);
99 	if (visible && pn->pn_vis != NULL)
100 		visible = pn_vis(td, proc, pn);
101 	if (!visible)
102 		return (0);
103 	return (1);
104 }
105 
106 static int
107 pfs_visible(struct thread *td, struct pfs_node *pn, pid_t pid, struct proc **p)
108 {
109 	struct proc *proc;
110 
111 	PFS_TRACE(("%s (pid: %d, req: %d)",
112 	    pn->pn_name, pid, td->td_proc->p_pid));
113 
114 	if (p)
115 		*p = NULL;
116 	if (pid == NO_PID)
117 		PFS_RETURN (1);
118 	if ((proc = pfind(pid)) == NULL)
119 		PFS_RETURN (0);
120 	if (pfs_visible_proc(td, pn, proc)) {
121 		if (p)
122 			*p = proc;
123 		else
124 			PROC_UNLOCK(proc);
125 		PFS_RETURN (1);
126 	}
127 	PROC_UNLOCK(proc);
128 	PFS_RETURN (0);
129 }
130 
131 /*
132  * Verify permissions
133  */
134 static int
135 pfs_access(struct vop_access_args *va)
136 {
137 	struct vnode *vn = va->a_vp;
138 	struct pfs_vdata *pvd = vn->v_data;
139 	struct vattr vattr;
140 	int error;
141 
142 	PFS_TRACE(("%s", pvd->pvd_pn->pn_name));
143 	(void)pvd;
144 
145 	error = VOP_GETATTR(vn, &vattr, va->a_cred);
146 	if (error)
147 		PFS_RETURN (error);
148 	error = vaccess(vn->v_type, vattr.va_mode, vattr.va_uid,
149 	    vattr.va_gid, va->a_accmode, va->a_cred, NULL);
150 	PFS_RETURN (error);
151 }
152 
153 /*
154  * Close a file or directory
155  */
156 static int
157 pfs_close(struct vop_close_args *va)
158 {
159 	struct vnode *vn = va->a_vp;
160 	struct pfs_vdata *pvd = vn->v_data;
161 	struct pfs_node *pn = pvd->pvd_pn;
162 	struct proc *proc;
163 	int error;
164 
165 	PFS_TRACE(("%s", pn->pn_name));
166 	pfs_assert_not_owned(pn);
167 
168 	/*
169 	 * Do nothing unless this is the last close and the node has a
170 	 * last-close handler.
171 	 */
172 	if (vrefcnt(vn) > 1 || pn->pn_close == NULL)
173 		PFS_RETURN (0);
174 
175 	if (pvd->pvd_pid != NO_PID) {
176 		proc = pfind(pvd->pvd_pid);
177 	} else {
178 		proc = NULL;
179 	}
180 
181 	error = pn_close(va->a_td, proc, pn);
182 
183 	if (proc != NULL)
184 		PROC_UNLOCK(proc);
185 
186 	PFS_RETURN (error);
187 }
188 
189 /*
190  * Get file attributes
191  */
192 static int
193 pfs_getattr(struct vop_getattr_args *va)
194 {
195 	struct vnode *vn = va->a_vp;
196 	struct pfs_vdata *pvd = vn->v_data;
197 	struct pfs_node *pn = pvd->pvd_pn;
198 	struct vattr *vap = va->a_vap;
199 	struct proc *proc;
200 	int error = 0;
201 
202 	PFS_TRACE(("%s", pn->pn_name));
203 	pfs_assert_not_owned(pn);
204 
205 	if (!pfs_visible(curthread, pn, pvd->pvd_pid, &proc))
206 		PFS_RETURN (ENOENT);
207 
208 	vap->va_type = vn->v_type;
209 	vap->va_fileid = pn_fileno(pn, pvd->pvd_pid);
210 	vap->va_flags = 0;
211 	vap->va_blocksize = PAGE_SIZE;
212 	vap->va_bytes = vap->va_size = 0;
213 	vap->va_filerev = 0;
214 	vap->va_fsid = vn->v_mount->mnt_stat.f_fsid.val[0];
215 	vap->va_nlink = 1;
216 	nanotime(&vap->va_ctime);
217 	vap->va_atime = vap->va_mtime = vap->va_ctime;
218 
219 	switch (pn->pn_type) {
220 	case pfstype_procdir:
221 	case pfstype_root:
222 	case pfstype_dir:
223 #if 0
224 		pfs_lock(pn);
225 		/* compute link count */
226 		pfs_unlock(pn);
227 #endif
228 		vap->va_mode = 0555;
229 		break;
230 	case pfstype_file:
231 	case pfstype_symlink:
232 		vap->va_mode = 0444;
233 		break;
234 	default:
235 		printf("shouldn't be here!\n");
236 		vap->va_mode = 0;
237 		break;
238 	}
239 
240 	if (proc != NULL) {
241 		vap->va_uid = proc->p_ucred->cr_ruid;
242 		vap->va_gid = proc->p_ucred->cr_rgid;
243 	} else {
244 		vap->va_uid = 0;
245 		vap->va_gid = 0;
246 	}
247 
248 	if (pn->pn_attr != NULL)
249 		error = pn_attr(curthread, proc, pn, vap);
250 
251 	if(proc != NULL)
252 		PROC_UNLOCK(proc);
253 
254 	PFS_RETURN (error);
255 }
256 
257 /*
258  * Perform an ioctl
259  */
260 static int
261 pfs_ioctl(struct vop_ioctl_args *va)
262 {
263 	struct vnode *vn;
264 	struct pfs_vdata *pvd;
265 	struct pfs_node *pn;
266 	struct proc *proc;
267 	int error;
268 
269 	vn = va->a_vp;
270 	vn_lock(vn, LK_SHARED | LK_RETRY);
271 	if (vn->v_iflag & VI_DOOMED) {
272 		VOP_UNLOCK(vn, 0);
273 		return (EBADF);
274 	}
275 	pvd = vn->v_data;
276 	pn = pvd->pvd_pn;
277 
278 	PFS_TRACE(("%s: %lx", pn->pn_name, va->a_command));
279 	pfs_assert_not_owned(pn);
280 
281 	if (vn->v_type != VREG) {
282 		VOP_UNLOCK(vn, 0);
283 		PFS_RETURN (EINVAL);
284 	}
285 	KASSERT_PN_IS_FILE(pn);
286 
287 	if (pn->pn_ioctl == NULL) {
288 		VOP_UNLOCK(vn, 0);
289 		PFS_RETURN (ENOTTY);
290 	}
291 
292 	/*
293 	 * This is necessary because process' privileges may
294 	 * have changed since the open() call.
295 	 */
296 	if (!pfs_visible(curthread, pn, pvd->pvd_pid, &proc)) {
297 		VOP_UNLOCK(vn, 0);
298 		PFS_RETURN (EIO);
299 	}
300 
301 	error = pn_ioctl(curthread, proc, pn, va->a_command, va->a_data);
302 
303 	if (proc != NULL)
304 		PROC_UNLOCK(proc);
305 
306 	VOP_UNLOCK(vn, 0);
307 	PFS_RETURN (error);
308 }
309 
310 /*
311  * Perform getextattr
312  */
313 static int
314 pfs_getextattr(struct vop_getextattr_args *va)
315 {
316 	struct vnode *vn = va->a_vp;
317 	struct pfs_vdata *pvd = vn->v_data;
318 	struct pfs_node *pn = pvd->pvd_pn;
319 	struct proc *proc;
320 	int error;
321 
322 	PFS_TRACE(("%s", pn->pn_name));
323 	pfs_assert_not_owned(pn);
324 
325 	/*
326 	 * This is necessary because either process' privileges may
327 	 * have changed since the open() call.
328 	 */
329 	if (!pfs_visible(curthread, pn, pvd->pvd_pid, &proc))
330 		PFS_RETURN (EIO);
331 
332 	if (pn->pn_getextattr == NULL)
333 		error = EOPNOTSUPP;
334 	else
335 		error = pn_getextattr(curthread, proc, pn,
336 		    va->a_attrnamespace, va->a_name, va->a_uio,
337 		    va->a_size, va->a_cred);
338 
339 	if (proc != NULL)
340 		PROC_UNLOCK(proc);
341 
342 	PFS_RETURN (error);
343 }
344 
345 /*
346  * Convert a vnode to its component name
347  */
348 static int
349 pfs_vptocnp(struct vop_vptocnp_args *ap)
350 {
351 	struct vnode *vp = ap->a_vp;
352 	struct vnode **dvp = ap->a_vpp;
353 	struct pfs_vdata *pvd = vp->v_data;
354 	struct pfs_node *pd = pvd->pvd_pn;
355 	struct pfs_node *pn;
356 	struct mount *mp;
357 	char *buf = ap->a_buf;
358 	int *buflen = ap->a_buflen;
359 	char pidbuf[PFS_NAMELEN];
360 	pid_t pid = pvd->pvd_pid;
361 	int len, i, error, locked;
362 
363 	i = *buflen;
364 	error = 0;
365 
366 	pfs_lock(pd);
367 
368 	if (vp->v_type == VDIR && pd->pn_type == pfstype_root) {
369 		*dvp = vp;
370 		vhold(*dvp);
371 		pfs_unlock(pd);
372 		PFS_RETURN (0);
373 	} else if (vp->v_type == VDIR && pd->pn_type == pfstype_procdir) {
374 		len = snprintf(pidbuf, sizeof(pidbuf), "%d", pid);
375 		i -= len;
376 		if (i < 0) {
377 			error = ENOMEM;
378 			goto failed;
379 		}
380 		bcopy(pidbuf, buf + i, len);
381 	} else {
382 		len = strlen(pd->pn_name);
383 		i -= len;
384 		if (i < 0) {
385 			error = ENOMEM;
386 			goto failed;
387 		}
388 		bcopy(pd->pn_name, buf + i, len);
389 	}
390 
391 	pn = pd->pn_parent;
392 	pfs_unlock(pd);
393 
394 	mp = vp->v_mount;
395 	error = vfs_busy(mp, 0);
396 	if (error)
397 		return (error);
398 
399 	/*
400 	 * vp is held by caller.
401 	 */
402 	locked = VOP_ISLOCKED(vp);
403 	VOP_UNLOCK(vp, 0);
404 
405 	error = pfs_vncache_alloc(mp, dvp, pn, pid);
406 	if (error) {
407 		vn_lock(vp, locked | LK_RETRY);
408 		vfs_unbusy(mp);
409 		PFS_RETURN(error);
410 	}
411 
412 	*buflen = i;
413 	vhold(*dvp);
414 	vput(*dvp);
415 	vn_lock(vp, locked | LK_RETRY);
416 	vfs_unbusy(mp);
417 
418 	PFS_RETURN (0);
419 failed:
420 	pfs_unlock(pd);
421 	PFS_RETURN(error);
422 }
423 
424 /*
425  * Look up a file or directory
426  */
427 static int
428 pfs_lookup(struct vop_cachedlookup_args *va)
429 {
430 	struct vnode *vn = va->a_dvp;
431 	struct vnode **vpp = va->a_vpp;
432 	struct componentname *cnp = va->a_cnp;
433 	struct pfs_vdata *pvd = vn->v_data;
434 	struct pfs_node *pd = pvd->pvd_pn;
435 	struct pfs_node *pn, *pdn = NULL;
436 	pid_t pid = pvd->pvd_pid;
437 	char *pname;
438 	int error, i, namelen, visible;
439 
440 	PFS_TRACE(("%.*s", (int)cnp->cn_namelen, cnp->cn_nameptr));
441 	pfs_assert_not_owned(pd);
442 
443 	if (vn->v_type != VDIR)
444 		PFS_RETURN (ENOTDIR);
445 	KASSERT_PN_IS_DIR(pd);
446 
447 	error = VOP_ACCESS(vn, VEXEC, cnp->cn_cred, cnp->cn_thread);
448 	if (error)
449 		PFS_RETURN (error);
450 
451 	/*
452 	 * Don't support DELETE or RENAME.  CREATE is supported so
453 	 * that O_CREAT will work, but the lookup will still fail if
454 	 * the file does not exist.
455 	 */
456 	if ((cnp->cn_flags & ISLASTCN) &&
457 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
458 		PFS_RETURN (EOPNOTSUPP);
459 
460 	/* shortcut: check if the name is too long */
461 	if (cnp->cn_namelen >= PFS_NAMELEN)
462 		PFS_RETURN (ENOENT);
463 
464 	/* check that parent directory is visible... */
465 	if (!pfs_visible(curthread, pd, pvd->pvd_pid, NULL))
466 		PFS_RETURN (ENOENT);
467 
468 	/* self */
469 	namelen = cnp->cn_namelen;
470 	pname = cnp->cn_nameptr;
471 	if (namelen == 1 && pname[0] == '.') {
472 		pn = pd;
473 		*vpp = vn;
474 		VREF(vn);
475 		PFS_RETURN (0);
476 	}
477 
478 	/* parent */
479 	if (cnp->cn_flags & ISDOTDOT) {
480 		if (pd->pn_type == pfstype_root)
481 			PFS_RETURN (EIO);
482 		VOP_UNLOCK(vn, 0);
483 		KASSERT(pd->pn_parent != NULL,
484 		    ("%s(): non-root directory has no parent", __func__));
485 		/*
486 		 * This one is tricky.  Descendents of procdir nodes
487 		 * inherit their parent's process affinity, but
488 		 * there's no easy reverse mapping.  For simplicity,
489 		 * we assume that if this node is a procdir, its
490 		 * parent isn't (which is correct as long as
491 		 * descendents of procdir nodes are never procdir
492 		 * nodes themselves)
493 		 */
494 		if (pd->pn_type == pfstype_procdir)
495 			pid = NO_PID;
496 		pfs_lock(pd);
497 		pn = pd->pn_parent;
498 		pfs_unlock(pd);
499 		goto got_pnode;
500 	}
501 
502 	pfs_lock(pd);
503 
504 	/* named node */
505 	for (pn = pd->pn_nodes; pn != NULL; pn = pn->pn_next)
506 		if (pn->pn_type == pfstype_procdir)
507 			pdn = pn;
508 		else if (pn->pn_name[namelen] == '\0' &&
509 		    bcmp(pname, pn->pn_name, namelen) == 0) {
510 			pfs_unlock(pd);
511 			goto got_pnode;
512 		}
513 
514 	/* process dependent node */
515 	if ((pn = pdn) != NULL) {
516 		pid = 0;
517 		for (pid = 0, i = 0; i < namelen && isdigit(pname[i]); ++i)
518 			if ((pid = pid * 10 + pname[i] - '0') > PID_MAX)
519 				break;
520 		if (i == cnp->cn_namelen) {
521 			pfs_unlock(pd);
522 			goto got_pnode;
523 		}
524 	}
525 
526 	pfs_unlock(pd);
527 
528 	PFS_RETURN (ENOENT);
529 
530  got_pnode:
531 	pfs_assert_not_owned(pd);
532 	pfs_assert_not_owned(pn);
533 	visible = pfs_visible(curthread, pn, pid, NULL);
534 	if (!visible) {
535 		error = ENOENT;
536 		goto failed;
537 	}
538 
539 	error = pfs_vncache_alloc(vn->v_mount, vpp, pn, pid);
540 	if (error)
541 		goto failed;
542 
543 	if (cnp->cn_flags & ISDOTDOT)
544 		vn_lock(vn, LK_EXCLUSIVE|LK_RETRY);
545 	if (cnp->cn_flags & MAKEENTRY)
546 		cache_enter(vn, *vpp, cnp);
547 	PFS_RETURN (0);
548  failed:
549 	if (cnp->cn_flags & ISDOTDOT)
550 		vn_lock(vn, LK_EXCLUSIVE|LK_RETRY);
551 	PFS_RETURN(error);
552 }
553 
554 /*
555  * Open a file or directory.
556  */
557 static int
558 pfs_open(struct vop_open_args *va)
559 {
560 	struct vnode *vn = va->a_vp;
561 	struct pfs_vdata *pvd = vn->v_data;
562 	struct pfs_node *pn = pvd->pvd_pn;
563 	int mode = va->a_mode;
564 
565 	PFS_TRACE(("%s (mode 0x%x)", pn->pn_name, mode));
566 	pfs_assert_not_owned(pn);
567 
568 	/* check if the requested mode is permitted */
569 	if (((mode & FREAD) && !(mode & PFS_RD)) ||
570 	    ((mode & FWRITE) && !(mode & PFS_WR)))
571 		PFS_RETURN (EPERM);
572 
573 	/* we don't support locking */
574 	if ((mode & O_SHLOCK) || (mode & O_EXLOCK))
575 		PFS_RETURN (EOPNOTSUPP);
576 
577 	PFS_RETURN (0);
578 }
579 
580 /*
581  * Read from a file
582  */
583 static int
584 pfs_read(struct vop_read_args *va)
585 {
586 	struct vnode *vn = va->a_vp;
587 	struct pfs_vdata *pvd = vn->v_data;
588 	struct pfs_node *pn = pvd->pvd_pn;
589 	struct uio *uio = va->a_uio;
590 	struct proc *proc;
591 	struct sbuf *sb = NULL;
592 	int error, locked;
593 	unsigned int buflen, offset, resid;
594 
595 	PFS_TRACE(("%s", pn->pn_name));
596 	pfs_assert_not_owned(pn);
597 
598 	if (vn->v_type != VREG)
599 		PFS_RETURN (EINVAL);
600 	KASSERT_PN_IS_FILE(pn);
601 
602 	if (!(pn->pn_flags & PFS_RD))
603 		PFS_RETURN (EBADF);
604 
605 	if (pn->pn_fill == NULL)
606 		PFS_RETURN (EIO);
607 
608 	/*
609 	 * This is necessary because either process' privileges may
610 	 * have changed since the open() call.
611 	 */
612 	if (!pfs_visible(curthread, pn, pvd->pvd_pid, &proc))
613 		PFS_RETURN (EIO);
614 	if (proc != NULL) {
615 		_PHOLD(proc);
616 		PROC_UNLOCK(proc);
617 	}
618 
619 	vhold(vn);
620 	locked = VOP_ISLOCKED(vn);
621 	VOP_UNLOCK(vn, 0);
622 
623 	if (pn->pn_flags & PFS_RAWRD) {
624 		PFS_TRACE(("%zd resid", uio->uio_resid));
625 		error = pn_fill(curthread, proc, pn, NULL, uio);
626 		PFS_TRACE(("%zd resid", uio->uio_resid));
627 		goto ret;
628 	}
629 
630 	/* beaucoup sanity checks so we don't ask for bogus allocation */
631 	if (uio->uio_offset < 0 || uio->uio_resid < 0 ||
632 	    (offset = uio->uio_offset) != uio->uio_offset ||
633 	    (resid = uio->uio_resid) != uio->uio_resid ||
634 	    (buflen = offset + resid + 1) < offset || buflen > INT_MAX) {
635 		if (proc != NULL)
636 			PRELE(proc);
637 		error = EINVAL;
638 		goto ret;
639 	}
640 	if (buflen > MAXPHYS + 1)
641 		buflen = MAXPHYS + 1;
642 
643 	sb = sbuf_new(sb, NULL, buflen, 0);
644 	if (sb == NULL) {
645 		error = EIO;
646 		goto ret;
647 	}
648 
649 	error = pn_fill(curthread, proc, pn, sb, uio);
650 
651 	if (error) {
652 		sbuf_delete(sb);
653 		goto ret;
654 	}
655 
656 	sbuf_finish(sb);
657 	error = uiomove_frombuf(sbuf_data(sb), sbuf_len(sb), uio);
658 	sbuf_delete(sb);
659 ret:
660 	vn_lock(vn, locked | LK_RETRY);
661 	vdrop(vn);
662 	if (proc != NULL)
663 		PRELE(proc);
664 	PFS_RETURN (error);
665 }
666 
667 /*
668  * Iterate through directory entries
669  */
670 static int
671 pfs_iterate(struct thread *td, struct proc *proc, struct pfs_node *pd,
672 	    struct pfs_node **pn, struct proc **p)
673 {
674 	int visible;
675 
676 	sx_assert(&allproc_lock, SX_SLOCKED);
677 	pfs_assert_owned(pd);
678  again:
679 	if (*pn == NULL) {
680 		/* first node */
681 		*pn = pd->pn_nodes;
682 	} else if ((*pn)->pn_type != pfstype_procdir) {
683 		/* next node */
684 		*pn = (*pn)->pn_next;
685 	}
686 	if (*pn != NULL && (*pn)->pn_type == pfstype_procdir) {
687 		/* next process */
688 		if (*p == NULL)
689 			*p = LIST_FIRST(&allproc);
690 		else
691 			*p = LIST_NEXT(*p, p_list);
692 		/* out of processes: next node */
693 		if (*p == NULL)
694 			*pn = (*pn)->pn_next;
695 		else
696 			PROC_LOCK(*p);
697 	}
698 
699 	if ((*pn) == NULL)
700 		return (-1);
701 
702 	if (*p != NULL) {
703 		visible = pfs_visible_proc(td, *pn, *p);
704 		PROC_UNLOCK(*p);
705 	} else if (proc != NULL) {
706 		visible = pfs_visible_proc(td, *pn, proc);
707 	} else {
708 		visible = 1;
709 	}
710 	if (!visible)
711 		goto again;
712 
713 	return (0);
714 }
715 
716 /*
717  * Return directory entries.
718  */
719 static int
720 pfs_readdir(struct vop_readdir_args *va)
721 {
722 	struct vnode *vn = va->a_vp;
723 	struct pfs_vdata *pvd = vn->v_data;
724 	struct pfs_node *pd = pvd->pvd_pn;
725 	pid_t pid = pvd->pvd_pid;
726 	struct proc *p, *proc;
727 	struct pfs_node *pn;
728 	struct dirent *entry;
729 	struct uio *uio;
730 	off_t offset;
731 	int error, i, resid;
732 	char *buf, *ent;
733 
734 	KASSERT(pd->pn_info == vn->v_mount->mnt_data,
735 	    ("%s(): pn_info does not match mountpoint", __func__));
736 	PFS_TRACE(("%s pid %lu", pd->pn_name, (unsigned long)pid));
737 	pfs_assert_not_owned(pd);
738 
739 	if (vn->v_type != VDIR)
740 		PFS_RETURN (ENOTDIR);
741 	KASSERT_PN_IS_DIR(pd);
742 	uio = va->a_uio;
743 
744 	/* only allow reading entire entries */
745 	offset = uio->uio_offset;
746 	resid = uio->uio_resid;
747 	if (offset < 0 || offset % PFS_DELEN != 0 ||
748 	    (resid && resid < PFS_DELEN))
749 		PFS_RETURN (EINVAL);
750 	if (resid == 0)
751 		PFS_RETURN (0);
752 
753 	/* can't do this while holding the proc lock... */
754 	buf = malloc(resid, M_IOV, M_WAITOK | M_ZERO);
755 	sx_slock(&allproc_lock);
756 	pfs_lock(pd);
757 
758         /* check if the directory is visible to the caller */
759         if (!pfs_visible(curthread, pd, pid, &proc)) {
760 		sx_sunlock(&allproc_lock);
761 		pfs_unlock(pd);
762 		free(buf, M_IOV);
763                 PFS_RETURN (ENOENT);
764 	}
765 	KASSERT(pid == NO_PID || proc != NULL,
766 	    ("%s(): no process for pid %lu", __func__, (unsigned long)pid));
767 
768 	/* skip unwanted entries */
769 	for (pn = NULL, p = NULL; offset > 0; offset -= PFS_DELEN) {
770 		if (pfs_iterate(curthread, proc, pd, &pn, &p) == -1) {
771 			/* nothing left... */
772 			if (proc != NULL)
773 				PROC_UNLOCK(proc);
774 			pfs_unlock(pd);
775 			sx_sunlock(&allproc_lock);
776 			free(buf, M_IOV);
777 			PFS_RETURN (0);
778 		}
779 	}
780 
781 	/* fill in entries */
782 	ent = buf;
783 	while (pfs_iterate(curthread, proc, pd, &pn, &p) != -1 &&
784 	    resid >= PFS_DELEN) {
785 		entry = (struct dirent *)ent;
786 		entry->d_reclen = PFS_DELEN;
787 		entry->d_fileno = pn_fileno(pn, pid);
788 		/* PFS_DELEN was picked to fit PFS_NAMLEN */
789 		for (i = 0; i < PFS_NAMELEN - 1 && pn->pn_name[i] != '\0'; ++i)
790 			entry->d_name[i] = pn->pn_name[i];
791 		entry->d_name[i] = 0;
792 		entry->d_namlen = i;
793 		switch (pn->pn_type) {
794 		case pfstype_procdir:
795 			KASSERT(p != NULL,
796 			    ("reached procdir node with p == NULL"));
797 			entry->d_namlen = snprintf(entry->d_name,
798 			    PFS_NAMELEN, "%d", p->p_pid);
799 			/* fall through */
800 		case pfstype_root:
801 		case pfstype_dir:
802 		case pfstype_this:
803 		case pfstype_parent:
804 			entry->d_type = DT_DIR;
805 			break;
806 		case pfstype_file:
807 			entry->d_type = DT_REG;
808 			break;
809 		case pfstype_symlink:
810 			entry->d_type = DT_LNK;
811 			break;
812 		default:
813 			panic("%s has unexpected node type: %d", pn->pn_name, pn->pn_type);
814 		}
815 		PFS_TRACE(("%s", entry->d_name));
816 		offset += PFS_DELEN;
817 		resid -= PFS_DELEN;
818 		ent += PFS_DELEN;
819 	}
820 	if (proc != NULL)
821 		PROC_UNLOCK(proc);
822 	pfs_unlock(pd);
823 	sx_sunlock(&allproc_lock);
824 	PFS_TRACE(("%zd bytes", ent - buf));
825 	error = uiomove(buf, ent - buf, uio);
826 	free(buf, M_IOV);
827 	PFS_RETURN (error);
828 }
829 
830 /*
831  * Read a symbolic link
832  */
833 static int
834 pfs_readlink(struct vop_readlink_args *va)
835 {
836 	struct vnode *vn = va->a_vp;
837 	struct pfs_vdata *pvd = vn->v_data;
838 	struct pfs_node *pn = pvd->pvd_pn;
839 	struct uio *uio = va->a_uio;
840 	struct proc *proc = NULL;
841 	char buf[PATH_MAX];
842 	struct sbuf sb;
843 	int error, locked;
844 
845 	PFS_TRACE(("%s", pn->pn_name));
846 	pfs_assert_not_owned(pn);
847 
848 	if (vn->v_type != VLNK)
849 		PFS_RETURN (EINVAL);
850 	KASSERT_PN_IS_LINK(pn);
851 
852 	if (pn->pn_fill == NULL)
853 		PFS_RETURN (EIO);
854 
855 	if (pvd->pvd_pid != NO_PID) {
856 		if ((proc = pfind(pvd->pvd_pid)) == NULL)
857 			PFS_RETURN (EIO);
858 		if (proc->p_flag & P_WEXIT) {
859 			PROC_UNLOCK(proc);
860 			PFS_RETURN (EIO);
861 		}
862 		_PHOLD(proc);
863 		PROC_UNLOCK(proc);
864 	}
865 	vhold(vn);
866 	locked = VOP_ISLOCKED(vn);
867 	VOP_UNLOCK(vn, 0);
868 
869 	/* sbuf_new() can't fail with a static buffer */
870 	sbuf_new(&sb, buf, sizeof buf, 0);
871 
872 	error = pn_fill(curthread, proc, pn, &sb, NULL);
873 
874 	if (proc != NULL)
875 		PRELE(proc);
876 	vn_lock(vn, locked | LK_RETRY);
877 	vdrop(vn);
878 
879 	if (error) {
880 		sbuf_delete(&sb);
881 		PFS_RETURN (error);
882 	}
883 
884 	sbuf_finish(&sb);
885 	error = uiomove_frombuf(sbuf_data(&sb), sbuf_len(&sb), uio);
886 	sbuf_delete(&sb);
887 	PFS_RETURN (error);
888 }
889 
890 /*
891  * Reclaim a vnode
892  */
893 static int
894 pfs_reclaim(struct vop_reclaim_args *va)
895 {
896 	struct vnode *vn = va->a_vp;
897 	struct pfs_vdata *pvd = vn->v_data;
898 	struct pfs_node *pn = pvd->pvd_pn;
899 
900 	PFS_TRACE(("%s", pn->pn_name));
901 	pfs_assert_not_owned(pn);
902 
903 	return (pfs_vncache_free(va->a_vp));
904 }
905 
906 /*
907  * Set attributes
908  */
909 static int
910 pfs_setattr(struct vop_setattr_args *va)
911 {
912 	struct vnode *vn = va->a_vp;
913 	struct pfs_vdata *pvd = vn->v_data;
914 	struct pfs_node *pn = pvd->pvd_pn;
915 
916 	PFS_TRACE(("%s", pn->pn_name));
917 	pfs_assert_not_owned(pn);
918 
919 	PFS_RETURN (EOPNOTSUPP);
920 }
921 
922 /*
923  * Write to a file
924  */
925 static int
926 pfs_write(struct vop_write_args *va)
927 {
928 	struct vnode *vn = va->a_vp;
929 	struct pfs_vdata *pvd = vn->v_data;
930 	struct pfs_node *pn = pvd->pvd_pn;
931 	struct uio *uio = va->a_uio;
932 	struct proc *proc;
933 	struct sbuf sb;
934 	int error;
935 
936 	PFS_TRACE(("%s", pn->pn_name));
937 	pfs_assert_not_owned(pn);
938 
939 	if (vn->v_type != VREG)
940 		PFS_RETURN (EINVAL);
941 	KASSERT_PN_IS_FILE(pn);
942 
943 	if (!(pn->pn_flags & PFS_WR))
944 		PFS_RETURN (EBADF);
945 
946 	if (pn->pn_fill == NULL)
947 		PFS_RETURN (EIO);
948 
949 	/*
950 	 * This is necessary because either process' privileges may
951 	 * have changed since the open() call.
952 	 */
953 	if (!pfs_visible(curthread, pn, pvd->pvd_pid, &proc))
954 		PFS_RETURN (EIO);
955 	if (proc != NULL) {
956 		_PHOLD(proc);
957 		PROC_UNLOCK(proc);
958 	}
959 
960 	if (pn->pn_flags & PFS_RAWWR) {
961 		error = pn_fill(curthread, proc, pn, NULL, uio);
962 		if (proc != NULL)
963 			PRELE(proc);
964 		PFS_RETURN (error);
965 	}
966 
967 	sbuf_uionew(&sb, uio, &error);
968 	if (error) {
969 		if (proc != NULL)
970 			PRELE(proc);
971 		PFS_RETURN (error);
972 	}
973 
974 	error = pn_fill(curthread, proc, pn, &sb, uio);
975 
976 	sbuf_delete(&sb);
977 	if (proc != NULL)
978 		PRELE(proc);
979 	PFS_RETURN (error);
980 }
981 
982 /*
983  * Vnode operations
984  */
985 struct vop_vector pfs_vnodeops = {
986 	.vop_default =		&default_vnodeops,
987 
988 	.vop_access =		pfs_access,
989 	.vop_cachedlookup =	pfs_lookup,
990 	.vop_close =		pfs_close,
991 	.vop_create =		VOP_EOPNOTSUPP,
992 	.vop_getattr =		pfs_getattr,
993 	.vop_getextattr =	pfs_getextattr,
994 	.vop_ioctl =		pfs_ioctl,
995 	.vop_link =		VOP_EOPNOTSUPP,
996 	.vop_lookup =		vfs_cache_lookup,
997 	.vop_mkdir =		VOP_EOPNOTSUPP,
998 	.vop_mknod =		VOP_EOPNOTSUPP,
999 	.vop_open =		pfs_open,
1000 	.vop_read =		pfs_read,
1001 	.vop_readdir =		pfs_readdir,
1002 	.vop_readlink =		pfs_readlink,
1003 	.vop_reclaim =		pfs_reclaim,
1004 	.vop_remove =		VOP_EOPNOTSUPP,
1005 	.vop_rename =		VOP_EOPNOTSUPP,
1006 	.vop_rmdir =		VOP_EOPNOTSUPP,
1007 	.vop_setattr =		pfs_setattr,
1008 	.vop_symlink =		VOP_EOPNOTSUPP,
1009 	.vop_vptocnp =		pfs_vptocnp,
1010 	.vop_write =		pfs_write,
1011 	/* XXX I've probably forgotten a few that need VOP_EOPNOTSUPP */
1012 };
1013