xref: /freebsd/sys/fs/pseudofs/pseudofs_vnops.c (revision 716fd348e01c5f2ba125f878a634a753436c2994)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 2001 Dag-Erling Coïdan Smørgrav
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer
12  *    in this position and unchanged.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. The name of the author may not be used to endorse or promote products
17  *    derived from this software without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
20  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
21  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
22  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
23  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
24  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
25  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
26  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
27  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
28  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
29  */
30 
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
33 
34 #include "opt_pseudofs.h"
35 
36 #include <sys/param.h>
37 #include <sys/kernel.h>
38 #include <sys/systm.h>
39 #include <sys/ctype.h>
40 #include <sys/dirent.h>
41 #include <sys/fcntl.h>
42 #include <sys/limits.h>
43 #include <sys/lock.h>
44 #include <sys/malloc.h>
45 #include <sys/mount.h>
46 #include <sys/mutex.h>
47 #include <sys/namei.h>
48 #include <sys/proc.h>
49 #include <sys/sbuf.h>
50 #include <sys/sx.h>
51 #include <sys/sysctl.h>
52 #include <sys/vnode.h>
53 
54 #include <fs/pseudofs/pseudofs.h>
55 #include <fs/pseudofs/pseudofs_internal.h>
56 
57 #define KASSERT_PN_IS_DIR(pn)						\
58 	KASSERT((pn)->pn_type == pfstype_root ||			\
59 	    (pn)->pn_type == pfstype_dir ||				\
60 	    (pn)->pn_type == pfstype_procdir,				\
61 	    ("%s(): VDIR vnode refers to non-directory pfs_node", __func__))
62 
63 #define KASSERT_PN_IS_FILE(pn)						\
64 	KASSERT((pn)->pn_type == pfstype_file,				\
65 	    ("%s(): VREG vnode refers to non-file pfs_node", __func__))
66 
67 #define KASSERT_PN_IS_LINK(pn)						\
68 	KASSERT((pn)->pn_type == pfstype_symlink,			\
69 	    ("%s(): VLNK vnode refers to non-link pfs_node", __func__))
70 
71 #define	PFS_MAXBUFSIZ		1024 * 1024
72 
73 /*
74  * Returns the fileno, adjusted for target pid
75  */
76 static uint32_t
77 pn_fileno(struct pfs_node *pn, pid_t pid)
78 {
79 
80 	KASSERT(pn->pn_fileno > 0,
81 	    ("%s(): no fileno allocated", __func__));
82 	if (pid != NO_PID)
83 		return (pn->pn_fileno * NO_PID + pid);
84 	return (pn->pn_fileno);
85 }
86 
87 /*
88  * Returns non-zero if given file is visible to given thread.
89  */
90 static int
91 pfs_visible_proc(struct thread *td, struct pfs_node *pn, struct proc *proc)
92 {
93 	int visible;
94 
95 	if (proc == NULL)
96 		return (0);
97 
98 	PROC_LOCK_ASSERT(proc, MA_OWNED);
99 
100 	visible = ((proc->p_flag & P_WEXIT) == 0);
101 	if (visible)
102 		visible = (p_cansee(td, proc) == 0);
103 	if (visible && pn->pn_vis != NULL)
104 		visible = pn_vis(td, proc, pn);
105 	if (!visible)
106 		return (0);
107 	return (1);
108 }
109 
110 static int
111 pfs_visible(struct thread *td, struct pfs_node *pn, pid_t pid,
112     struct proc **p)
113 {
114 	struct proc *proc;
115 
116 	PFS_TRACE(("%s (pid: %d, req: %d)",
117 	    pn->pn_name, pid, td->td_proc->p_pid));
118 
119 	if (p)
120 		*p = NULL;
121 	if (pid == NO_PID)
122 		PFS_RETURN (1);
123 	proc = pfind(pid);
124 	if (proc == NULL)
125 		PFS_RETURN (0);
126 	if (pfs_visible_proc(td, pn, proc)) {
127 		if (p)
128 			*p = proc;
129 		else
130 			PROC_UNLOCK(proc);
131 		PFS_RETURN (1);
132 	}
133 	PROC_UNLOCK(proc);
134 	PFS_RETURN (0);
135 }
136 
137 static int
138 pfs_lookup_proc(pid_t pid, struct proc **p)
139 {
140 	struct proc *proc;
141 
142 	proc = pfind(pid);
143 	if (proc == NULL)
144 		return (0);
145 	if ((proc->p_flag & P_WEXIT) != 0) {
146 		PROC_UNLOCK(proc);
147 		return (0);
148 	}
149 	_PHOLD(proc);
150 	PROC_UNLOCK(proc);
151 	*p = proc;
152 	return (1);
153 }
154 
155 /*
156  * Verify permissions
157  */
158 static int
159 pfs_access(struct vop_access_args *va)
160 {
161 	struct vnode *vn = va->a_vp;
162 	struct pfs_vdata *pvd = vn->v_data;
163 	struct vattr vattr;
164 	int error;
165 
166 	PFS_TRACE(("%s", pvd->pvd_pn->pn_name));
167 	(void)pvd;
168 
169 	error = VOP_GETATTR(vn, &vattr, va->a_cred);
170 	if (error)
171 		PFS_RETURN (error);
172 	error = vaccess(vn->v_type, vattr.va_mode, vattr.va_uid, vattr.va_gid,
173 	    va->a_accmode, va->a_cred);
174 	PFS_RETURN (error);
175 }
176 
177 /*
178  * Close a file or directory
179  */
180 static int
181 pfs_close(struct vop_close_args *va)
182 {
183 	struct vnode *vn = va->a_vp;
184 	struct pfs_vdata *pvd = vn->v_data;
185 	struct pfs_node *pn = pvd->pvd_pn;
186 	struct proc *proc;
187 	int error;
188 
189 	PFS_TRACE(("%s", pn->pn_name));
190 	pfs_assert_not_owned(pn);
191 
192 	/*
193 	 * Do nothing unless this is the last close and the node has a
194 	 * last-close handler.
195 	 */
196 	if (vrefcnt(vn) > 1 || pn->pn_close == NULL)
197 		PFS_RETURN (0);
198 
199 	if (pvd->pvd_pid != NO_PID) {
200 		proc = pfind(pvd->pvd_pid);
201 	} else {
202 		proc = NULL;
203 	}
204 
205 	error = pn_close(va->a_td, proc, pn);
206 
207 	if (proc != NULL)
208 		PROC_UNLOCK(proc);
209 
210 	PFS_RETURN (error);
211 }
212 
213 /*
214  * Get file attributes
215  */
216 static int
217 pfs_getattr(struct vop_getattr_args *va)
218 {
219 	struct vnode *vn = va->a_vp;
220 	struct pfs_vdata *pvd = vn->v_data;
221 	struct pfs_node *pn = pvd->pvd_pn;
222 	struct vattr *vap = va->a_vap;
223 	struct proc *proc;
224 	int error = 0;
225 
226 	PFS_TRACE(("%s", pn->pn_name));
227 	pfs_assert_not_owned(pn);
228 
229 	if (!pfs_visible(curthread, pn, pvd->pvd_pid, &proc))
230 		PFS_RETURN (ENOENT);
231 
232 	vap->va_type = vn->v_type;
233 	vap->va_fileid = pn_fileno(pn, pvd->pvd_pid);
234 	vap->va_flags = 0;
235 	vap->va_blocksize = PAGE_SIZE;
236 	vap->va_bytes = vap->va_size = 0;
237 	vap->va_filerev = 0;
238 	vap->va_fsid = vn->v_mount->mnt_stat.f_fsid.val[0];
239 	vap->va_nlink = 1;
240 	nanotime(&vap->va_ctime);
241 	vap->va_atime = vap->va_mtime = vap->va_ctime;
242 
243 	switch (pn->pn_type) {
244 	case pfstype_procdir:
245 	case pfstype_root:
246 	case pfstype_dir:
247 #if 0
248 		pfs_lock(pn);
249 		/* compute link count */
250 		pfs_unlock(pn);
251 #endif
252 		vap->va_mode = 0555;
253 		break;
254 	case pfstype_file:
255 	case pfstype_symlink:
256 		vap->va_mode = 0444;
257 		break;
258 	default:
259 		printf("shouldn't be here!\n");
260 		vap->va_mode = 0;
261 		break;
262 	}
263 
264 	if (proc != NULL) {
265 		vap->va_uid = proc->p_ucred->cr_ruid;
266 		vap->va_gid = proc->p_ucred->cr_rgid;
267 	} else {
268 		vap->va_uid = 0;
269 		vap->va_gid = 0;
270 	}
271 
272 	if (pn->pn_attr != NULL)
273 		error = pn_attr(curthread, proc, pn, vap);
274 
275 	if(proc != NULL)
276 		PROC_UNLOCK(proc);
277 
278 	PFS_RETURN (error);
279 }
280 
281 /*
282  * Perform an ioctl
283  */
284 static int
285 pfs_ioctl(struct vop_ioctl_args *va)
286 {
287 	struct vnode *vn;
288 	struct pfs_vdata *pvd;
289 	struct pfs_node *pn;
290 	struct proc *proc;
291 	int error;
292 
293 	vn = va->a_vp;
294 	vn_lock(vn, LK_SHARED | LK_RETRY);
295 	if (VN_IS_DOOMED(vn)) {
296 		VOP_UNLOCK(vn);
297 		return (EBADF);
298 	}
299 	pvd = vn->v_data;
300 	pn = pvd->pvd_pn;
301 
302 	PFS_TRACE(("%s: %lx", pn->pn_name, va->a_command));
303 	pfs_assert_not_owned(pn);
304 
305 	if (vn->v_type != VREG) {
306 		VOP_UNLOCK(vn);
307 		PFS_RETURN (EINVAL);
308 	}
309 	KASSERT_PN_IS_FILE(pn);
310 
311 	if (pn->pn_ioctl == NULL) {
312 		VOP_UNLOCK(vn);
313 		PFS_RETURN (ENOTTY);
314 	}
315 
316 	/*
317 	 * This is necessary because process' privileges may
318 	 * have changed since the open() call.
319 	 */
320 	if (!pfs_visible(curthread, pn, pvd->pvd_pid, &proc)) {
321 		VOP_UNLOCK(vn);
322 		PFS_RETURN (EIO);
323 	}
324 
325 	error = pn_ioctl(curthread, proc, pn, va->a_command, va->a_data);
326 
327 	if (proc != NULL)
328 		PROC_UNLOCK(proc);
329 
330 	VOP_UNLOCK(vn);
331 	PFS_RETURN (error);
332 }
333 
334 /*
335  * Perform getextattr
336  */
337 static int
338 pfs_getextattr(struct vop_getextattr_args *va)
339 {
340 	struct vnode *vn = va->a_vp;
341 	struct pfs_vdata *pvd = vn->v_data;
342 	struct pfs_node *pn = pvd->pvd_pn;
343 	struct proc *proc;
344 	int error;
345 
346 	PFS_TRACE(("%s", pn->pn_name));
347 	pfs_assert_not_owned(pn);
348 
349 	/*
350 	 * This is necessary because either process' privileges may
351 	 * have changed since the open() call.
352 	 */
353 	if (!pfs_visible(curthread, pn, pvd->pvd_pid, &proc))
354 		PFS_RETURN (EIO);
355 
356 	if (pn->pn_getextattr == NULL)
357 		error = EOPNOTSUPP;
358 	else
359 		error = pn_getextattr(curthread, proc, pn,
360 		    va->a_attrnamespace, va->a_name, va->a_uio,
361 		    va->a_size, va->a_cred);
362 
363 	if (proc != NULL)
364 		PROC_UNLOCK(proc);
365 
366 	PFS_RETURN (error);
367 }
368 
369 /*
370  * Convert a vnode to its component name
371  */
372 static int
373 pfs_vptocnp(struct vop_vptocnp_args *ap)
374 {
375 	struct vnode *vp = ap->a_vp;
376 	struct vnode **dvp = ap->a_vpp;
377 	struct pfs_vdata *pvd = vp->v_data;
378 	struct pfs_node *pd = pvd->pvd_pn;
379 	struct pfs_node *pn;
380 	struct mount *mp;
381 	char *buf = ap->a_buf;
382 	size_t *buflen = ap->a_buflen;
383 	char pidbuf[PFS_NAMELEN];
384 	pid_t pid = pvd->pvd_pid;
385 	int len, i, error, locked;
386 
387 	i = *buflen;
388 	error = 0;
389 
390 	pfs_lock(pd);
391 
392 	if (vp->v_type == VDIR && pd->pn_type == pfstype_root) {
393 		*dvp = vp;
394 		vhold(*dvp);
395 		pfs_unlock(pd);
396 		PFS_RETURN (0);
397 	} else if (vp->v_type == VDIR && pd->pn_type == pfstype_procdir) {
398 		len = snprintf(pidbuf, sizeof(pidbuf), "%d", pid);
399 		i -= len;
400 		if (i < 0) {
401 			error = ENOMEM;
402 			goto failed;
403 		}
404 		bcopy(pidbuf, buf + i, len);
405 	} else {
406 		len = strlen(pd->pn_name);
407 		i -= len;
408 		if (i < 0) {
409 			error = ENOMEM;
410 			goto failed;
411 		}
412 		bcopy(pd->pn_name, buf + i, len);
413 	}
414 
415 	pn = pd->pn_parent;
416 	pfs_unlock(pd);
417 
418 	mp = vp->v_mount;
419 	error = vfs_busy(mp, 0);
420 	if (error)
421 		return (error);
422 
423 	/*
424 	 * vp is held by caller.
425 	 */
426 	locked = VOP_ISLOCKED(vp);
427 	VOP_UNLOCK(vp);
428 
429 	error = pfs_vncache_alloc(mp, dvp, pn, pid);
430 	if (error) {
431 		vn_lock(vp, locked | LK_RETRY);
432 		vfs_unbusy(mp);
433 		PFS_RETURN(error);
434 	}
435 
436 	*buflen = i;
437 	VOP_UNLOCK(*dvp);
438 	vn_lock(vp, locked | LK_RETRY);
439 	vfs_unbusy(mp);
440 
441 	PFS_RETURN (0);
442 failed:
443 	pfs_unlock(pd);
444 	PFS_RETURN(error);
445 }
446 
447 /*
448  * Look up a file or directory
449  */
450 static int
451 pfs_lookup(struct vop_cachedlookup_args *va)
452 {
453 	struct vnode *vn = va->a_dvp;
454 	struct vnode **vpp = va->a_vpp;
455 	struct componentname *cnp = va->a_cnp;
456 	struct pfs_vdata *pvd = vn->v_data;
457 	struct pfs_node *pd = pvd->pvd_pn;
458 	struct pfs_node *pn, *pdn = NULL;
459 	struct mount *mp;
460 	pid_t pid = pvd->pvd_pid;
461 	char *pname;
462 	int error, i, namelen, visible;
463 
464 	PFS_TRACE(("%.*s", (int)cnp->cn_namelen, cnp->cn_nameptr));
465 	pfs_assert_not_owned(pd);
466 
467 	if (vn->v_type != VDIR)
468 		PFS_RETURN (ENOTDIR);
469 	KASSERT_PN_IS_DIR(pd);
470 
471 	/*
472 	 * Don't support DELETE or RENAME.  CREATE is supported so
473 	 * that O_CREAT will work, but the lookup will still fail if
474 	 * the file does not exist.
475 	 */
476 	if ((cnp->cn_flags & ISLASTCN) &&
477 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
478 		PFS_RETURN (EOPNOTSUPP);
479 
480 	/* shortcut: check if the name is too long */
481 	if (cnp->cn_namelen >= PFS_NAMELEN)
482 		PFS_RETURN (ENOENT);
483 
484 	/* check that parent directory is visible... */
485 	if (!pfs_visible(curthread, pd, pvd->pvd_pid, NULL))
486 		PFS_RETURN (ENOENT);
487 
488 	/* self */
489 	namelen = cnp->cn_namelen;
490 	pname = cnp->cn_nameptr;
491 	if (namelen == 1 && pname[0] == '.') {
492 		pn = pd;
493 		*vpp = vn;
494 		VREF(vn);
495 		PFS_RETURN (0);
496 	}
497 
498 	mp = vn->v_mount;
499 
500 	/* parent */
501 	if (cnp->cn_flags & ISDOTDOT) {
502 		if (pd->pn_type == pfstype_root)
503 			PFS_RETURN (EIO);
504 		error = vfs_busy(mp, MBF_NOWAIT);
505 		if (error != 0) {
506 			vfs_ref(mp);
507 			VOP_UNLOCK(vn);
508 			error = vfs_busy(mp, 0);
509 			vn_lock(vn, LK_EXCLUSIVE | LK_RETRY);
510 			vfs_rel(mp);
511 			if (error != 0)
512 				PFS_RETURN(ENOENT);
513 			if (VN_IS_DOOMED(vn)) {
514 				vfs_unbusy(mp);
515 				PFS_RETURN(ENOENT);
516 			}
517 		}
518 		VOP_UNLOCK(vn);
519 		KASSERT(pd->pn_parent != NULL,
520 		    ("%s(): non-root directory has no parent", __func__));
521 		/*
522 		 * This one is tricky.  Descendents of procdir nodes
523 		 * inherit their parent's process affinity, but
524 		 * there's no easy reverse mapping.  For simplicity,
525 		 * we assume that if this node is a procdir, its
526 		 * parent isn't (which is correct as long as
527 		 * descendents of procdir nodes are never procdir
528 		 * nodes themselves)
529 		 */
530 		if (pd->pn_type == pfstype_procdir)
531 			pid = NO_PID;
532 		pfs_lock(pd);
533 		pn = pd->pn_parent;
534 		pfs_unlock(pd);
535 		goto got_pnode;
536 	}
537 
538 	pfs_lock(pd);
539 
540 	/* named node */
541 	for (pn = pd->pn_nodes; pn != NULL; pn = pn->pn_next)
542 		if (pn->pn_type == pfstype_procdir)
543 			pdn = pn;
544 		else if (pn->pn_name[namelen] == '\0' &&
545 		    bcmp(pname, pn->pn_name, namelen) == 0) {
546 			pfs_unlock(pd);
547 			goto got_pnode;
548 		}
549 
550 	/* process dependent node */
551 	if ((pn = pdn) != NULL) {
552 		pid = 0;
553 		for (pid = 0, i = 0; i < namelen && isdigit(pname[i]); ++i)
554 			if ((pid = pid * 10 + pname[i] - '0') > PID_MAX)
555 				break;
556 		if (i == cnp->cn_namelen) {
557 			pfs_unlock(pd);
558 			goto got_pnode;
559 		}
560 	}
561 
562 	pfs_unlock(pd);
563 
564 	PFS_RETURN (ENOENT);
565 
566  got_pnode:
567 	pfs_assert_not_owned(pd);
568 	pfs_assert_not_owned(pn);
569 	visible = pfs_visible(curthread, pn, pid, NULL);
570 	if (!visible) {
571 		error = ENOENT;
572 		goto failed;
573 	}
574 
575 	error = pfs_vncache_alloc(mp, vpp, pn, pid);
576 	if (error)
577 		goto failed;
578 
579 	if (cnp->cn_flags & ISDOTDOT) {
580 		vfs_unbusy(mp);
581 		vn_lock(vn, LK_EXCLUSIVE | LK_RETRY);
582 		if (VN_IS_DOOMED(vn)) {
583 			vput(*vpp);
584 			*vpp = NULL;
585 			PFS_RETURN(ENOENT);
586 		}
587 	}
588 	if (cnp->cn_flags & MAKEENTRY && !VN_IS_DOOMED(vn))
589 		cache_enter(vn, *vpp, cnp);
590 	PFS_RETURN (0);
591  failed:
592 	if (cnp->cn_flags & ISDOTDOT) {
593 		vfs_unbusy(mp);
594 		vn_lock(vn, LK_EXCLUSIVE | LK_RETRY);
595 		*vpp = NULL;
596 	}
597 	PFS_RETURN(error);
598 }
599 
600 /*
601  * Open a file or directory.
602  */
603 static int
604 pfs_open(struct vop_open_args *va)
605 {
606 	struct vnode *vn = va->a_vp;
607 	struct pfs_vdata *pvd = vn->v_data;
608 	struct pfs_node *pn = pvd->pvd_pn;
609 	int mode = va->a_mode;
610 
611 	PFS_TRACE(("%s (mode 0x%x)", pn->pn_name, mode));
612 	pfs_assert_not_owned(pn);
613 
614 	/* check if the requested mode is permitted */
615 	if (((mode & FREAD) && !(mode & PFS_RD)) ||
616 	    ((mode & FWRITE) && !(mode & PFS_WR)))
617 		PFS_RETURN (EPERM);
618 
619 	/* we don't support locking */
620 	if ((mode & O_SHLOCK) || (mode & O_EXLOCK))
621 		PFS_RETURN (EOPNOTSUPP);
622 
623 	PFS_RETURN (0);
624 }
625 
626 struct sbuf_seek_helper {
627 	off_t		skip_bytes;
628 	struct uio	*uio;
629 };
630 
631 static int
632 pfs_sbuf_uio_drain(void *arg, const char *data, int len)
633 {
634 	struct sbuf_seek_helper *ssh;
635 	struct uio *uio;
636 	int error, skipped;
637 
638 	ssh = arg;
639 	uio = ssh->uio;
640 	skipped = 0;
641 
642 	/* Need to discard first uio_offset bytes. */
643 	if (ssh->skip_bytes > 0) {
644 		if (ssh->skip_bytes >= len) {
645 			ssh->skip_bytes -= len;
646 			return (len);
647 		}
648 
649 		data += ssh->skip_bytes;
650 		len -= ssh->skip_bytes;
651 		skipped = ssh->skip_bytes;
652 		ssh->skip_bytes = 0;
653 	}
654 
655 	error = uiomove(__DECONST(void *, data), len, uio);
656 	if (error != 0)
657 		return (-error);
658 
659 	/*
660 	 * The fill function has more to emit, but the reader is finished.
661 	 * This is similar to the truncated read case for non-draining PFS
662 	 * sbufs, and should be handled appropriately in fill-routines.
663 	 */
664 	if (uio->uio_resid == 0)
665 		return (-ENOBUFS);
666 
667 	return (skipped + len);
668 }
669 
670 /*
671  * Read from a file
672  */
673 static int
674 pfs_read(struct vop_read_args *va)
675 {
676 	struct vnode *vn = va->a_vp;
677 	struct pfs_vdata *pvd = vn->v_data;
678 	struct pfs_node *pn = pvd->pvd_pn;
679 	struct uio *uio = va->a_uio;
680 	struct proc *proc;
681 	struct sbuf *sb = NULL;
682 	int error, locked;
683 	off_t buflen, buflim;
684 	struct sbuf_seek_helper ssh;
685 
686 	PFS_TRACE(("%s", pn->pn_name));
687 	pfs_assert_not_owned(pn);
688 
689 	if (vn->v_type != VREG)
690 		PFS_RETURN (EINVAL);
691 	KASSERT_PN_IS_FILE(pn);
692 
693 	if (!(pn->pn_flags & PFS_RD))
694 		PFS_RETURN (EBADF);
695 
696 	if (pn->pn_fill == NULL)
697 		PFS_RETURN (EIO);
698 
699 	/*
700 	 * This is necessary because either process' privileges may
701 	 * have changed since the open() call.
702 	 */
703 	if (!pfs_visible(curthread, pn, pvd->pvd_pid, &proc))
704 		PFS_RETURN (EIO);
705 	if (proc != NULL) {
706 		_PHOLD(proc);
707 		PROC_UNLOCK(proc);
708 	}
709 
710 	vhold(vn);
711 	locked = VOP_ISLOCKED(vn);
712 	VOP_UNLOCK(vn);
713 
714 	if (pn->pn_flags & PFS_RAWRD) {
715 		PFS_TRACE(("%zd resid", uio->uio_resid));
716 		error = pn_fill(curthread, proc, pn, NULL, uio);
717 		PFS_TRACE(("%zd resid", uio->uio_resid));
718 		goto ret;
719 	}
720 
721 	if (uio->uio_resid < 0 || uio->uio_offset < 0 ||
722 	    uio->uio_resid > OFF_MAX - uio->uio_offset) {
723 		error = EINVAL;
724 		goto ret;
725 	}
726 	buflen = uio->uio_offset + uio->uio_resid + 1;
727 	if (pn->pn_flags & PFS_AUTODRAIN)
728 		/*
729 		 * We can use a smaller buffer if we can stream output to the
730 		 * consumer.
731 		 */
732 		buflim = PAGE_SIZE;
733 	else
734 		buflim = PFS_MAXBUFSIZ;
735 	if (buflen > buflim)
736 		buflen = buflim;
737 
738 	sb = sbuf_new(sb, NULL, buflen, 0);
739 	if (sb == NULL) {
740 		error = EIO;
741 		goto ret;
742 	}
743 
744 	if (pn->pn_flags & PFS_AUTODRAIN) {
745 		ssh.skip_bytes = uio->uio_offset;
746 		ssh.uio = uio;
747 		sbuf_set_drain(sb, pfs_sbuf_uio_drain, &ssh);
748 	}
749 
750 	error = pn_fill(curthread, proc, pn, sb, uio);
751 
752 	if (error) {
753 		sbuf_delete(sb);
754 		goto ret;
755 	}
756 
757 	/*
758 	 * XXX: If the buffer overflowed, sbuf_len() will not return
759 	 * the data length. Then just use the full length because an
760 	 * overflowed sbuf must be full.
761 	 */
762 	error = sbuf_finish(sb);
763 	if ((pn->pn_flags & PFS_AUTODRAIN)) {
764 		/*
765 		 * ENOBUFS just indicates early termination of the fill
766 		 * function as the caller's buffer was already filled.  Squash
767 		 * to zero.
768 		 */
769 		if (uio->uio_resid == 0 && error == ENOBUFS)
770 			error = 0;
771 	} else {
772 		if (error == 0)
773 			buflen = sbuf_len(sb);
774 		else
775 			/* The trailing byte is not valid. */
776 			buflen--;
777 		error = uiomove_frombuf(sbuf_data(sb), buflen, uio);
778 	}
779 	sbuf_delete(sb);
780 ret:
781 	vn_lock(vn, locked | LK_RETRY);
782 	vdrop(vn);
783 	if (proc != NULL)
784 		PRELE(proc);
785 	PFS_RETURN (error);
786 }
787 
788 /*
789  * Iterate through directory entries
790  */
791 static int
792 pfs_iterate(struct thread *td, struct proc *proc, struct pfs_node *pd,
793 	    struct pfs_node **pn, struct proc **p)
794 {
795 	int visible;
796 
797 	sx_assert(&allproc_lock, SX_SLOCKED);
798 	pfs_assert_owned(pd);
799  again:
800 	if (*pn == NULL) {
801 		/* first node */
802 		*pn = pd->pn_nodes;
803 	} else if ((*pn)->pn_type != pfstype_procdir) {
804 		/* next node */
805 		*pn = (*pn)->pn_next;
806 	}
807 	if (*pn != NULL && (*pn)->pn_type == pfstype_procdir) {
808 		/* next process */
809 		if (*p == NULL)
810 			*p = LIST_FIRST(&allproc);
811 		else
812 			*p = LIST_NEXT(*p, p_list);
813 		/* out of processes: next node */
814 		if (*p == NULL)
815 			*pn = (*pn)->pn_next;
816 		else
817 			PROC_LOCK(*p);
818 	}
819 
820 	if ((*pn) == NULL)
821 		return (-1);
822 
823 	if (*p != NULL) {
824 		visible = pfs_visible_proc(td, *pn, *p);
825 		PROC_UNLOCK(*p);
826 	} else if (proc != NULL) {
827 		visible = pfs_visible_proc(td, *pn, proc);
828 	} else {
829 		visible = 1;
830 	}
831 	if (!visible)
832 		goto again;
833 
834 	return (0);
835 }
836 
837 /* Directory entry list */
838 struct pfsentry {
839 	STAILQ_ENTRY(pfsentry)	link;
840 	struct dirent		entry;
841 };
842 STAILQ_HEAD(pfsdirentlist, pfsentry);
843 
844 /*
845  * Return directory entries.
846  */
847 static int
848 pfs_readdir(struct vop_readdir_args *va)
849 {
850 	struct vnode *vn = va->a_vp;
851 	struct pfs_vdata *pvd = vn->v_data;
852 	struct pfs_node *pd = pvd->pvd_pn;
853 	pid_t pid = pvd->pvd_pid;
854 	struct proc *p, *proc;
855 	struct pfs_node *pn;
856 	struct uio *uio;
857 	struct pfsentry *pfsent, *pfsent2;
858 	struct pfsdirentlist lst;
859 	off_t offset;
860 	int error, i, resid;
861 
862 	STAILQ_INIT(&lst);
863 	error = 0;
864 	KASSERT(pd->pn_info == vn->v_mount->mnt_data,
865 	    ("%s(): pn_info does not match mountpoint", __func__));
866 	PFS_TRACE(("%s pid %lu", pd->pn_name, (unsigned long)pid));
867 	pfs_assert_not_owned(pd);
868 
869 	if (vn->v_type != VDIR)
870 		PFS_RETURN (ENOTDIR);
871 	KASSERT_PN_IS_DIR(pd);
872 	uio = va->a_uio;
873 
874 	/* only allow reading entire entries */
875 	offset = uio->uio_offset;
876 	resid = uio->uio_resid;
877 	if (offset < 0 || offset % PFS_DELEN != 0 ||
878 	    (resid && resid < PFS_DELEN))
879 		PFS_RETURN (EINVAL);
880 	if (resid == 0)
881 		PFS_RETURN (0);
882 
883 	proc = NULL;
884 	if (pid != NO_PID && !pfs_lookup_proc(pid, &proc))
885 		PFS_RETURN (ENOENT);
886 
887 	sx_slock(&allproc_lock);
888 	pfs_lock(pd);
889 
890 	KASSERT(pid == NO_PID || proc != NULL,
891 	    ("%s(): no process for pid %lu", __func__, (unsigned long)pid));
892 
893 	if (pid != NO_PID) {
894 		PROC_LOCK(proc);
895 
896 		/* check if the directory is visible to the caller */
897 		if (!pfs_visible_proc(curthread, pd, proc)) {
898 			_PRELE(proc);
899 			PROC_UNLOCK(proc);
900 			sx_sunlock(&allproc_lock);
901 			pfs_unlock(pd);
902 			PFS_RETURN (ENOENT);
903 		}
904 	}
905 
906 	/* skip unwanted entries */
907 	for (pn = NULL, p = NULL; offset > 0; offset -= PFS_DELEN) {
908 		if (pfs_iterate(curthread, proc, pd, &pn, &p) == -1) {
909 			/* nothing left... */
910 			if (proc != NULL) {
911 				_PRELE(proc);
912 				PROC_UNLOCK(proc);
913 			}
914 			pfs_unlock(pd);
915 			sx_sunlock(&allproc_lock);
916 			PFS_RETURN (0);
917 		}
918 	}
919 
920 	/* fill in entries */
921 	while (pfs_iterate(curthread, proc, pd, &pn, &p) != -1 &&
922 	    resid >= PFS_DELEN) {
923 		if ((pfsent = malloc(sizeof(struct pfsentry), M_IOV,
924 		    M_NOWAIT | M_ZERO)) == NULL) {
925 			error = ENOMEM;
926 			break;
927 		}
928 		pfsent->entry.d_reclen = PFS_DELEN;
929 		pfsent->entry.d_fileno = pn_fileno(pn, pid);
930 		/* PFS_DELEN was picked to fit PFS_NAMLEN */
931 		for (i = 0; i < PFS_NAMELEN - 1 && pn->pn_name[i] != '\0'; ++i)
932 			pfsent->entry.d_name[i] = pn->pn_name[i];
933 		pfsent->entry.d_namlen = i;
934 		/* NOTE: d_off is the offset of the *next* entry. */
935 		pfsent->entry.d_off = offset + PFS_DELEN;
936 		switch (pn->pn_type) {
937 		case pfstype_procdir:
938 			KASSERT(p != NULL,
939 			    ("reached procdir node with p == NULL"));
940 			pfsent->entry.d_namlen = snprintf(pfsent->entry.d_name,
941 			    PFS_NAMELEN, "%d", p->p_pid);
942 			/* fall through */
943 		case pfstype_root:
944 		case pfstype_dir:
945 		case pfstype_this:
946 		case pfstype_parent:
947 			pfsent->entry.d_type = DT_DIR;
948 			break;
949 		case pfstype_file:
950 			pfsent->entry.d_type = DT_REG;
951 			break;
952 		case pfstype_symlink:
953 			pfsent->entry.d_type = DT_LNK;
954 			break;
955 		default:
956 			panic("%s has unexpected node type: %d", pn->pn_name, pn->pn_type);
957 		}
958 		PFS_TRACE(("%s", pfsent->entry.d_name));
959 		dirent_terminate(&pfsent->entry);
960 		STAILQ_INSERT_TAIL(&lst, pfsent, link);
961 		offset += PFS_DELEN;
962 		resid -= PFS_DELEN;
963 	}
964 	if (proc != NULL) {
965 		_PRELE(proc);
966 		PROC_UNLOCK(proc);
967 	}
968 	pfs_unlock(pd);
969 	sx_sunlock(&allproc_lock);
970 	i = 0;
971 	STAILQ_FOREACH_SAFE(pfsent, &lst, link, pfsent2) {
972 		if (error == 0)
973 			error = uiomove(&pfsent->entry, PFS_DELEN, uio);
974 		free(pfsent, M_IOV);
975 		i++;
976 	}
977 	PFS_TRACE(("%ju bytes", (uintmax_t)(i * PFS_DELEN)));
978 	PFS_RETURN (error);
979 }
980 
981 /*
982  * Read a symbolic link
983  */
984 static int
985 pfs_readlink(struct vop_readlink_args *va)
986 {
987 	struct vnode *vn = va->a_vp;
988 	struct pfs_vdata *pvd = vn->v_data;
989 	struct pfs_node *pn = pvd->pvd_pn;
990 	struct uio *uio = va->a_uio;
991 	struct proc *proc = NULL;
992 	char buf[PATH_MAX];
993 	struct sbuf sb;
994 	int error, locked;
995 
996 	PFS_TRACE(("%s", pn->pn_name));
997 	pfs_assert_not_owned(pn);
998 
999 	if (vn->v_type != VLNK)
1000 		PFS_RETURN (EINVAL);
1001 	KASSERT_PN_IS_LINK(pn);
1002 
1003 	if (pn->pn_fill == NULL)
1004 		PFS_RETURN (EIO);
1005 
1006 	if (pvd->pvd_pid != NO_PID) {
1007 		if ((proc = pfind(pvd->pvd_pid)) == NULL)
1008 			PFS_RETURN (EIO);
1009 		if (proc->p_flag & P_WEXIT) {
1010 			PROC_UNLOCK(proc);
1011 			PFS_RETURN (EIO);
1012 		}
1013 		_PHOLD(proc);
1014 		PROC_UNLOCK(proc);
1015 	}
1016 	vhold(vn);
1017 	locked = VOP_ISLOCKED(vn);
1018 	VOP_UNLOCK(vn);
1019 
1020 	/* sbuf_new() can't fail with a static buffer */
1021 	sbuf_new(&sb, buf, sizeof buf, 0);
1022 
1023 	error = pn_fill(curthread, proc, pn, &sb, NULL);
1024 
1025 	if (proc != NULL)
1026 		PRELE(proc);
1027 	vn_lock(vn, locked | LK_RETRY);
1028 	vdrop(vn);
1029 
1030 	if (error) {
1031 		sbuf_delete(&sb);
1032 		PFS_RETURN (error);
1033 	}
1034 
1035 	if (sbuf_finish(&sb) != 0) {
1036 		sbuf_delete(&sb);
1037 		PFS_RETURN (ENAMETOOLONG);
1038 	}
1039 
1040 	error = uiomove_frombuf(sbuf_data(&sb), sbuf_len(&sb), uio);
1041 	sbuf_delete(&sb);
1042 	PFS_RETURN (error);
1043 }
1044 
1045 /*
1046  * Reclaim a vnode
1047  */
1048 static int
1049 pfs_reclaim(struct vop_reclaim_args *va)
1050 {
1051 	struct vnode *vn = va->a_vp;
1052 	struct pfs_vdata *pvd = vn->v_data;
1053 	struct pfs_node *pn = pvd->pvd_pn;
1054 
1055 	PFS_TRACE(("%s", pn->pn_name));
1056 	pfs_assert_not_owned(pn);
1057 
1058 	return (pfs_vncache_free(va->a_vp));
1059 }
1060 
1061 /*
1062  * Set attributes
1063  */
1064 static int
1065 pfs_setattr(struct vop_setattr_args *va)
1066 {
1067 	struct vnode *vn = va->a_vp;
1068 	struct pfs_vdata *pvd = vn->v_data;
1069 	struct pfs_node *pn = pvd->pvd_pn;
1070 
1071 	PFS_TRACE(("%s", pn->pn_name));
1072 	pfs_assert_not_owned(pn);
1073 
1074 	/* Silently ignore unchangeable attributes. */
1075 	PFS_RETURN (0);
1076 }
1077 
1078 /*
1079  * Write to a file
1080  */
1081 static int
1082 pfs_write(struct vop_write_args *va)
1083 {
1084 	struct vnode *vn = va->a_vp;
1085 	struct pfs_vdata *pvd = vn->v_data;
1086 	struct pfs_node *pn = pvd->pvd_pn;
1087 	struct uio *uio = va->a_uio;
1088 	struct proc *proc;
1089 	struct sbuf sb;
1090 	int error;
1091 
1092 	PFS_TRACE(("%s", pn->pn_name));
1093 	pfs_assert_not_owned(pn);
1094 
1095 	if (vn->v_type != VREG)
1096 		PFS_RETURN (EINVAL);
1097 	KASSERT_PN_IS_FILE(pn);
1098 
1099 	if (!(pn->pn_flags & PFS_WR))
1100 		PFS_RETURN (EBADF);
1101 
1102 	if (pn->pn_fill == NULL)
1103 		PFS_RETURN (EIO);
1104 
1105 	if (uio->uio_resid > PFS_MAXBUFSIZ)
1106 		PFS_RETURN (EIO);
1107 
1108 	/*
1109 	 * This is necessary because either process' privileges may
1110 	 * have changed since the open() call.
1111 	 */
1112 	if (!pfs_visible(curthread, pn, pvd->pvd_pid, &proc))
1113 		PFS_RETURN (EIO);
1114 	if (proc != NULL) {
1115 		_PHOLD(proc);
1116 		PROC_UNLOCK(proc);
1117 	}
1118 
1119 	if (pn->pn_flags & PFS_RAWWR) {
1120 		error = pn_fill(curthread, proc, pn, NULL, uio);
1121 		if (proc != NULL)
1122 			PRELE(proc);
1123 		PFS_RETURN (error);
1124 	}
1125 
1126 	sbuf_uionew(&sb, uio, &error);
1127 	if (error) {
1128 		if (proc != NULL)
1129 			PRELE(proc);
1130 		PFS_RETURN (error);
1131 	}
1132 
1133 	error = pn_fill(curthread, proc, pn, &sb, uio);
1134 
1135 	sbuf_delete(&sb);
1136 	if (proc != NULL)
1137 		PRELE(proc);
1138 	PFS_RETURN (error);
1139 }
1140 
1141 /*
1142  * Vnode operations
1143  */
1144 struct vop_vector pfs_vnodeops = {
1145 	.vop_default =		&default_vnodeops,
1146 
1147 	.vop_access =		pfs_access,
1148 	.vop_cachedlookup =	pfs_lookup,
1149 	.vop_close =		pfs_close,
1150 	.vop_create =		VOP_EOPNOTSUPP,
1151 	.vop_getattr =		pfs_getattr,
1152 	.vop_getextattr =	pfs_getextattr,
1153 	.vop_ioctl =		pfs_ioctl,
1154 	.vop_link =		VOP_EOPNOTSUPP,
1155 	.vop_lookup =		vfs_cache_lookup,
1156 	.vop_mkdir =		VOP_EOPNOTSUPP,
1157 	.vop_mknod =		VOP_EOPNOTSUPP,
1158 	.vop_open =		pfs_open,
1159 	.vop_read =		pfs_read,
1160 	.vop_readdir =		pfs_readdir,
1161 	.vop_readlink =		pfs_readlink,
1162 	.vop_reclaim =		pfs_reclaim,
1163 	.vop_remove =		VOP_EOPNOTSUPP,
1164 	.vop_rename =		VOP_EOPNOTSUPP,
1165 	.vop_rmdir =		VOP_EOPNOTSUPP,
1166 	.vop_setattr =		pfs_setattr,
1167 	.vop_symlink =		VOP_EOPNOTSUPP,
1168 	.vop_vptocnp =		pfs_vptocnp,
1169 	.vop_write =		pfs_write,
1170 	.vop_add_writecount =	vop_stdadd_writecount_nomsync,
1171 	/* XXX I've probably forgotten a few that need VOP_EOPNOTSUPP */
1172 };
1173 VFS_VOP_VECTOR_REGISTER(pfs_vnodeops);
1174