xref: /freebsd/sys/fs/pseudofs/pseudofs_vnops.c (revision f4b37ed0f8b307b1f3f0f630ca725d68f1dff30d)
1 /*-
2  * Copyright (c) 2001 Dag-Erling Coïdan Smørgrav
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 	VOP_UNLOCK(*dvp, 0);
414 	vn_lock(vp, locked | LK_RETRY);
415 	vfs_unbusy(mp);
416 
417 	PFS_RETURN (0);
418 failed:
419 	pfs_unlock(pd);
420 	PFS_RETURN(error);
421 }
422 
423 /*
424  * Look up a file or directory
425  */
426 static int
427 pfs_lookup(struct vop_cachedlookup_args *va)
428 {
429 	struct vnode *vn = va->a_dvp;
430 	struct vnode **vpp = va->a_vpp;
431 	struct componentname *cnp = va->a_cnp;
432 	struct pfs_vdata *pvd = vn->v_data;
433 	struct pfs_node *pd = pvd->pvd_pn;
434 	struct pfs_node *pn, *pdn = NULL;
435 	struct mount *mp;
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 	mp = vn->v_mount;
479 
480 	/* parent */
481 	if (cnp->cn_flags & ISDOTDOT) {
482 		if (pd->pn_type == pfstype_root)
483 			PFS_RETURN (EIO);
484 		error = vfs_busy(mp, MBF_NOWAIT);
485 		if (error != 0) {
486 			vfs_ref(mp);
487 			VOP_UNLOCK(vn, 0);
488 			error = vfs_busy(mp, 0);
489 			vn_lock(vn, LK_EXCLUSIVE | LK_RETRY);
490 			vfs_rel(mp);
491 			if (error != 0)
492 				PFS_RETURN(ENOENT);
493 			if (vn->v_iflag & VI_DOOMED) {
494 				vfs_unbusy(mp);
495 				PFS_RETURN(ENOENT);
496 			}
497 		}
498 		VOP_UNLOCK(vn, 0);
499 		KASSERT(pd->pn_parent != NULL,
500 		    ("%s(): non-root directory has no parent", __func__));
501 		/*
502 		 * This one is tricky.  Descendents of procdir nodes
503 		 * inherit their parent's process affinity, but
504 		 * there's no easy reverse mapping.  For simplicity,
505 		 * we assume that if this node is a procdir, its
506 		 * parent isn't (which is correct as long as
507 		 * descendents of procdir nodes are never procdir
508 		 * nodes themselves)
509 		 */
510 		if (pd->pn_type == pfstype_procdir)
511 			pid = NO_PID;
512 		pfs_lock(pd);
513 		pn = pd->pn_parent;
514 		pfs_unlock(pd);
515 		goto got_pnode;
516 	}
517 
518 	pfs_lock(pd);
519 
520 	/* named node */
521 	for (pn = pd->pn_nodes; pn != NULL; pn = pn->pn_next)
522 		if (pn->pn_type == pfstype_procdir)
523 			pdn = pn;
524 		else if (pn->pn_name[namelen] == '\0' &&
525 		    bcmp(pname, pn->pn_name, namelen) == 0) {
526 			pfs_unlock(pd);
527 			goto got_pnode;
528 		}
529 
530 	/* process dependent node */
531 	if ((pn = pdn) != NULL) {
532 		pid = 0;
533 		for (pid = 0, i = 0; i < namelen && isdigit(pname[i]); ++i)
534 			if ((pid = pid * 10 + pname[i] - '0') > PID_MAX)
535 				break;
536 		if (i == cnp->cn_namelen) {
537 			pfs_unlock(pd);
538 			goto got_pnode;
539 		}
540 	}
541 
542 	pfs_unlock(pd);
543 
544 	PFS_RETURN (ENOENT);
545 
546  got_pnode:
547 	pfs_assert_not_owned(pd);
548 	pfs_assert_not_owned(pn);
549 	visible = pfs_visible(curthread, pn, pid, NULL);
550 	if (!visible) {
551 		error = ENOENT;
552 		goto failed;
553 	}
554 
555 	error = pfs_vncache_alloc(mp, vpp, pn, pid);
556 	if (error)
557 		goto failed;
558 
559 	if (cnp->cn_flags & ISDOTDOT) {
560 		vfs_unbusy(mp);
561 		vn_lock(vn, LK_EXCLUSIVE | LK_RETRY);
562 		if (vn->v_iflag & VI_DOOMED) {
563 			vput(*vpp);
564 			*vpp = NULL;
565 			PFS_RETURN(ENOENT);
566 		}
567 	}
568 	if (cnp->cn_flags & MAKEENTRY && !(vn->v_iflag & VI_DOOMED))
569 		cache_enter(vn, *vpp, cnp);
570 	PFS_RETURN (0);
571  failed:
572 	if (cnp->cn_flags & ISDOTDOT) {
573 		vfs_unbusy(mp);
574 		vn_lock(vn, LK_EXCLUSIVE | LK_RETRY);
575 		*vpp = NULL;
576 	}
577 	PFS_RETURN(error);
578 }
579 
580 /*
581  * Open a file or directory.
582  */
583 static int
584 pfs_open(struct vop_open_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 	int mode = va->a_mode;
590 
591 	PFS_TRACE(("%s (mode 0x%x)", pn->pn_name, mode));
592 	pfs_assert_not_owned(pn);
593 
594 	/* check if the requested mode is permitted */
595 	if (((mode & FREAD) && !(mode & PFS_RD)) ||
596 	    ((mode & FWRITE) && !(mode & PFS_WR)))
597 		PFS_RETURN (EPERM);
598 
599 	/* we don't support locking */
600 	if ((mode & O_SHLOCK) || (mode & O_EXLOCK))
601 		PFS_RETURN (EOPNOTSUPP);
602 
603 	PFS_RETURN (0);
604 }
605 
606 /*
607  * Read from a file
608  */
609 static int
610 pfs_read(struct vop_read_args *va)
611 {
612 	struct vnode *vn = va->a_vp;
613 	struct pfs_vdata *pvd = vn->v_data;
614 	struct pfs_node *pn = pvd->pvd_pn;
615 	struct uio *uio = va->a_uio;
616 	struct proc *proc;
617 	struct sbuf *sb = NULL;
618 	int error, locked;
619 	off_t buflen;
620 
621 	PFS_TRACE(("%s", pn->pn_name));
622 	pfs_assert_not_owned(pn);
623 
624 	if (vn->v_type != VREG)
625 		PFS_RETURN (EINVAL);
626 	KASSERT_PN_IS_FILE(pn);
627 
628 	if (!(pn->pn_flags & PFS_RD))
629 		PFS_RETURN (EBADF);
630 
631 	if (pn->pn_fill == NULL)
632 		PFS_RETURN (EIO);
633 
634 	/*
635 	 * This is necessary because either process' privileges may
636 	 * have changed since the open() call.
637 	 */
638 	if (!pfs_visible(curthread, pn, pvd->pvd_pid, &proc))
639 		PFS_RETURN (EIO);
640 	if (proc != NULL) {
641 		_PHOLD(proc);
642 		PROC_UNLOCK(proc);
643 	}
644 
645 	vhold(vn);
646 	locked = VOP_ISLOCKED(vn);
647 	VOP_UNLOCK(vn, 0);
648 
649 	if (pn->pn_flags & PFS_RAWRD) {
650 		PFS_TRACE(("%zd resid", uio->uio_resid));
651 		error = pn_fill(curthread, proc, pn, NULL, uio);
652 		PFS_TRACE(("%zd resid", uio->uio_resid));
653 		goto ret;
654 	}
655 
656 	if (uio->uio_resid < 0 || uio->uio_offset < 0 ||
657 	    uio->uio_resid > OFF_MAX - uio->uio_offset) {
658 		error = EINVAL;
659 		goto ret;
660 	}
661 	buflen = uio->uio_offset + uio->uio_resid;
662 	if (buflen > MAXPHYS)
663 		buflen = MAXPHYS;
664 
665 	sb = sbuf_new(sb, NULL, buflen + 1, 0);
666 	if (sb == NULL) {
667 		error = EIO;
668 		goto ret;
669 	}
670 
671 	error = pn_fill(curthread, proc, pn, sb, uio);
672 
673 	if (error) {
674 		sbuf_delete(sb);
675 		goto ret;
676 	}
677 
678 	/*
679 	 * XXX: If the buffer overflowed, sbuf_len() will not return
680 	 * the data length. Then just use the full length because an
681 	 * overflowed sbuf must be full.
682 	 */
683 	if (sbuf_finish(sb) == 0)
684 		buflen = sbuf_len(sb);
685 	error = uiomove_frombuf(sbuf_data(sb), buflen, uio);
686 	sbuf_delete(sb);
687 ret:
688 	vn_lock(vn, locked | LK_RETRY);
689 	vdrop(vn);
690 	if (proc != NULL)
691 		PRELE(proc);
692 	PFS_RETURN (error);
693 }
694 
695 /*
696  * Iterate through directory entries
697  */
698 static int
699 pfs_iterate(struct thread *td, struct proc *proc, struct pfs_node *pd,
700 	    struct pfs_node **pn, struct proc **p)
701 {
702 	int visible;
703 
704 	sx_assert(&allproc_lock, SX_SLOCKED);
705 	pfs_assert_owned(pd);
706  again:
707 	if (*pn == NULL) {
708 		/* first node */
709 		*pn = pd->pn_nodes;
710 	} else if ((*pn)->pn_type != pfstype_procdir) {
711 		/* next node */
712 		*pn = (*pn)->pn_next;
713 	}
714 	if (*pn != NULL && (*pn)->pn_type == pfstype_procdir) {
715 		/* next process */
716 		if (*p == NULL)
717 			*p = LIST_FIRST(&allproc);
718 		else
719 			*p = LIST_NEXT(*p, p_list);
720 		/* out of processes: next node */
721 		if (*p == NULL)
722 			*pn = (*pn)->pn_next;
723 		else
724 			PROC_LOCK(*p);
725 	}
726 
727 	if ((*pn) == NULL)
728 		return (-1);
729 
730 	if (*p != NULL) {
731 		visible = pfs_visible_proc(td, *pn, *p);
732 		PROC_UNLOCK(*p);
733 	} else if (proc != NULL) {
734 		visible = pfs_visible_proc(td, *pn, proc);
735 	} else {
736 		visible = 1;
737 	}
738 	if (!visible)
739 		goto again;
740 
741 	return (0);
742 }
743 
744 /* Directory entry list */
745 struct pfsentry {
746 	STAILQ_ENTRY(pfsentry)	link;
747 	struct dirent		entry;
748 };
749 STAILQ_HEAD(pfsdirentlist, pfsentry);
750 
751 /*
752  * Return directory entries.
753  */
754 static int
755 pfs_readdir(struct vop_readdir_args *va)
756 {
757 	struct vnode *vn = va->a_vp;
758 	struct pfs_vdata *pvd = vn->v_data;
759 	struct pfs_node *pd = pvd->pvd_pn;
760 	pid_t pid = pvd->pvd_pid;
761 	struct proc *p, *proc;
762 	struct pfs_node *pn;
763 	struct uio *uio;
764 	struct pfsentry *pfsent, *pfsent2;
765 	struct pfsdirentlist lst;
766 	off_t offset;
767 	int error, i, resid;
768 
769 	STAILQ_INIT(&lst);
770 	error = 0;
771 	KASSERT(pd->pn_info == vn->v_mount->mnt_data,
772 	    ("%s(): pn_info does not match mountpoint", __func__));
773 	PFS_TRACE(("%s pid %lu", pd->pn_name, (unsigned long)pid));
774 	pfs_assert_not_owned(pd);
775 
776 	if (vn->v_type != VDIR)
777 		PFS_RETURN (ENOTDIR);
778 	KASSERT_PN_IS_DIR(pd);
779 	uio = va->a_uio;
780 
781 	/* only allow reading entire entries */
782 	offset = uio->uio_offset;
783 	resid = uio->uio_resid;
784 	if (offset < 0 || offset % PFS_DELEN != 0 ||
785 	    (resid && resid < PFS_DELEN))
786 		PFS_RETURN (EINVAL);
787 	if (resid == 0)
788 		PFS_RETURN (0);
789 
790 	sx_slock(&allproc_lock);
791 	pfs_lock(pd);
792 
793         /* check if the directory is visible to the caller */
794         if (!pfs_visible(curthread, pd, pid, &proc)) {
795 		sx_sunlock(&allproc_lock);
796 		pfs_unlock(pd);
797                 PFS_RETURN (ENOENT);
798 	}
799 	KASSERT(pid == NO_PID || proc != NULL,
800 	    ("%s(): no process for pid %lu", __func__, (unsigned long)pid));
801 
802 	/* skip unwanted entries */
803 	for (pn = NULL, p = NULL; offset > 0; offset -= PFS_DELEN) {
804 		if (pfs_iterate(curthread, proc, pd, &pn, &p) == -1) {
805 			/* nothing left... */
806 			if (proc != NULL)
807 				PROC_UNLOCK(proc);
808 			pfs_unlock(pd);
809 			sx_sunlock(&allproc_lock);
810 			PFS_RETURN (0);
811 		}
812 	}
813 
814 	/* fill in entries */
815 	while (pfs_iterate(curthread, proc, pd, &pn, &p) != -1 &&
816 	    resid >= PFS_DELEN) {
817 		if ((pfsent = malloc(sizeof(struct pfsentry), M_IOV,
818 		    M_NOWAIT | M_ZERO)) == NULL) {
819 			error = ENOMEM;
820 			break;
821 		}
822 		pfsent->entry.d_reclen = PFS_DELEN;
823 		pfsent->entry.d_fileno = pn_fileno(pn, pid);
824 		/* PFS_DELEN was picked to fit PFS_NAMLEN */
825 		for (i = 0; i < PFS_NAMELEN - 1 && pn->pn_name[i] != '\0'; ++i)
826 			pfsent->entry.d_name[i] = pn->pn_name[i];
827 		pfsent->entry.d_name[i] = 0;
828 		pfsent->entry.d_namlen = i;
829 		switch (pn->pn_type) {
830 		case pfstype_procdir:
831 			KASSERT(p != NULL,
832 			    ("reached procdir node with p == NULL"));
833 			pfsent->entry.d_namlen = snprintf(pfsent->entry.d_name,
834 			    PFS_NAMELEN, "%d", p->p_pid);
835 			/* fall through */
836 		case pfstype_root:
837 		case pfstype_dir:
838 		case pfstype_this:
839 		case pfstype_parent:
840 			pfsent->entry.d_type = DT_DIR;
841 			break;
842 		case pfstype_file:
843 			pfsent->entry.d_type = DT_REG;
844 			break;
845 		case pfstype_symlink:
846 			pfsent->entry.d_type = DT_LNK;
847 			break;
848 		default:
849 			panic("%s has unexpected node type: %d", pn->pn_name, pn->pn_type);
850 		}
851 		PFS_TRACE(("%s", pfsent->entry.d_name));
852 		STAILQ_INSERT_TAIL(&lst, pfsent, link);
853 		offset += PFS_DELEN;
854 		resid -= PFS_DELEN;
855 	}
856 	if (proc != NULL)
857 		PROC_UNLOCK(proc);
858 	pfs_unlock(pd);
859 	sx_sunlock(&allproc_lock);
860 	i = 0;
861 	STAILQ_FOREACH_SAFE(pfsent, &lst, link, pfsent2) {
862 		if (error == 0)
863 			error = uiomove(&pfsent->entry, PFS_DELEN, uio);
864 		free(pfsent, M_IOV);
865 		i++;
866 	}
867 	PFS_TRACE(("%d bytes", i * PFS_DELEN));
868 	PFS_RETURN (error);
869 }
870 
871 /*
872  * Read a symbolic link
873  */
874 static int
875 pfs_readlink(struct vop_readlink_args *va)
876 {
877 	struct vnode *vn = va->a_vp;
878 	struct pfs_vdata *pvd = vn->v_data;
879 	struct pfs_node *pn = pvd->pvd_pn;
880 	struct uio *uio = va->a_uio;
881 	struct proc *proc = NULL;
882 	char buf[PATH_MAX];
883 	struct sbuf sb;
884 	int error, locked;
885 
886 	PFS_TRACE(("%s", pn->pn_name));
887 	pfs_assert_not_owned(pn);
888 
889 	if (vn->v_type != VLNK)
890 		PFS_RETURN (EINVAL);
891 	KASSERT_PN_IS_LINK(pn);
892 
893 	if (pn->pn_fill == NULL)
894 		PFS_RETURN (EIO);
895 
896 	if (pvd->pvd_pid != NO_PID) {
897 		if ((proc = pfind(pvd->pvd_pid)) == NULL)
898 			PFS_RETURN (EIO);
899 		if (proc->p_flag & P_WEXIT) {
900 			PROC_UNLOCK(proc);
901 			PFS_RETURN (EIO);
902 		}
903 		_PHOLD(proc);
904 		PROC_UNLOCK(proc);
905 	}
906 	vhold(vn);
907 	locked = VOP_ISLOCKED(vn);
908 	VOP_UNLOCK(vn, 0);
909 
910 	/* sbuf_new() can't fail with a static buffer */
911 	sbuf_new(&sb, buf, sizeof buf, 0);
912 
913 	error = pn_fill(curthread, proc, pn, &sb, NULL);
914 
915 	if (proc != NULL)
916 		PRELE(proc);
917 	vn_lock(vn, locked | LK_RETRY);
918 	vdrop(vn);
919 
920 	if (error) {
921 		sbuf_delete(&sb);
922 		PFS_RETURN (error);
923 	}
924 
925 	if (sbuf_finish(&sb) != 0) {
926 		sbuf_delete(&sb);
927 		PFS_RETURN (ENAMETOOLONG);
928 	}
929 
930 	error = uiomove_frombuf(sbuf_data(&sb), sbuf_len(&sb), uio);
931 	sbuf_delete(&sb);
932 	PFS_RETURN (error);
933 }
934 
935 /*
936  * Reclaim a vnode
937  */
938 static int
939 pfs_reclaim(struct vop_reclaim_args *va)
940 {
941 	struct vnode *vn = va->a_vp;
942 	struct pfs_vdata *pvd = vn->v_data;
943 	struct pfs_node *pn = pvd->pvd_pn;
944 
945 	PFS_TRACE(("%s", pn->pn_name));
946 	pfs_assert_not_owned(pn);
947 
948 	return (pfs_vncache_free(va->a_vp));
949 }
950 
951 /*
952  * Set attributes
953  */
954 static int
955 pfs_setattr(struct vop_setattr_args *va)
956 {
957 	struct vnode *vn = va->a_vp;
958 	struct pfs_vdata *pvd = vn->v_data;
959 	struct pfs_node *pn = pvd->pvd_pn;
960 
961 	PFS_TRACE(("%s", pn->pn_name));
962 	pfs_assert_not_owned(pn);
963 
964 	PFS_RETURN (EOPNOTSUPP);
965 }
966 
967 /*
968  * Write to a file
969  */
970 static int
971 pfs_write(struct vop_write_args *va)
972 {
973 	struct vnode *vn = va->a_vp;
974 	struct pfs_vdata *pvd = vn->v_data;
975 	struct pfs_node *pn = pvd->pvd_pn;
976 	struct uio *uio = va->a_uio;
977 	struct proc *proc;
978 	struct sbuf sb;
979 	int error;
980 
981 	PFS_TRACE(("%s", pn->pn_name));
982 	pfs_assert_not_owned(pn);
983 
984 	if (vn->v_type != VREG)
985 		PFS_RETURN (EINVAL);
986 	KASSERT_PN_IS_FILE(pn);
987 
988 	if (!(pn->pn_flags & PFS_WR))
989 		PFS_RETURN (EBADF);
990 
991 	if (pn->pn_fill == NULL)
992 		PFS_RETURN (EIO);
993 
994 	/*
995 	 * This is necessary because either process' privileges may
996 	 * have changed since the open() call.
997 	 */
998 	if (!pfs_visible(curthread, pn, pvd->pvd_pid, &proc))
999 		PFS_RETURN (EIO);
1000 	if (proc != NULL) {
1001 		_PHOLD(proc);
1002 		PROC_UNLOCK(proc);
1003 	}
1004 
1005 	if (pn->pn_flags & PFS_RAWWR) {
1006 		error = pn_fill(curthread, proc, pn, NULL, uio);
1007 		if (proc != NULL)
1008 			PRELE(proc);
1009 		PFS_RETURN (error);
1010 	}
1011 
1012 	sbuf_uionew(&sb, uio, &error);
1013 	if (error) {
1014 		if (proc != NULL)
1015 			PRELE(proc);
1016 		PFS_RETURN (error);
1017 	}
1018 
1019 	error = pn_fill(curthread, proc, pn, &sb, uio);
1020 
1021 	sbuf_delete(&sb);
1022 	if (proc != NULL)
1023 		PRELE(proc);
1024 	PFS_RETURN (error);
1025 }
1026 
1027 /*
1028  * Vnode operations
1029  */
1030 struct vop_vector pfs_vnodeops = {
1031 	.vop_default =		&default_vnodeops,
1032 
1033 	.vop_access =		pfs_access,
1034 	.vop_cachedlookup =	pfs_lookup,
1035 	.vop_close =		pfs_close,
1036 	.vop_create =		VOP_EOPNOTSUPP,
1037 	.vop_getattr =		pfs_getattr,
1038 	.vop_getextattr =	pfs_getextattr,
1039 	.vop_ioctl =		pfs_ioctl,
1040 	.vop_link =		VOP_EOPNOTSUPP,
1041 	.vop_lookup =		vfs_cache_lookup,
1042 	.vop_mkdir =		VOP_EOPNOTSUPP,
1043 	.vop_mknod =		VOP_EOPNOTSUPP,
1044 	.vop_open =		pfs_open,
1045 	.vop_read =		pfs_read,
1046 	.vop_readdir =		pfs_readdir,
1047 	.vop_readlink =		pfs_readlink,
1048 	.vop_reclaim =		pfs_reclaim,
1049 	.vop_remove =		VOP_EOPNOTSUPP,
1050 	.vop_rename =		VOP_EOPNOTSUPP,
1051 	.vop_rmdir =		VOP_EOPNOTSUPP,
1052 	.vop_setattr =		pfs_setattr,
1053 	.vop_symlink =		VOP_EOPNOTSUPP,
1054 	.vop_vptocnp =		pfs_vptocnp,
1055 	.vop_write =		pfs_write,
1056 	/* XXX I've probably forgotten a few that need VOP_EOPNOTSUPP */
1057 };
1058