xref: /freebsd/sys/fs/pseudofs/pseudofs_vnops.c (revision 0d66206fff44f864ea8a4b220c3a53b4caa959a0)
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 
94 	if (proc == NULL)
95 		return (0);
96 
97 	PROC_LOCK_ASSERT(proc, MA_OWNED);
98 
99 	if ((proc->p_flag & P_WEXIT) != 0)
100 		return (0);
101 	if (p_cansee(td, proc) != 0)
102 		return (0);
103 	return (pn_vis(td, proc, pn));
104 }
105 
106 static int
107 pfs_visible(struct thread *td, struct pfs_node *pn, pid_t pid,
108     struct proc **p)
109 {
110 	struct proc *proc;
111 
112 	PFS_TRACE(("%s (pid: %d, req: %d)",
113 	    pn->pn_name, pid, td->td_proc->p_pid));
114 
115 	if (p)
116 		*p = NULL;
117 	if (pid == NO_PID)
118 		PFS_RETURN (pn_vis(td, NULL, pn));
119 	proc = pfind(pid);
120 	if (proc == NULL)
121 		PFS_RETURN (0);
122 	if (pfs_visible_proc(td, pn, proc)) {
123 		if (p)
124 			*p = proc;
125 		else
126 			PROC_UNLOCK(proc);
127 		PFS_RETURN (1);
128 	}
129 	PROC_UNLOCK(proc);
130 	PFS_RETURN (0);
131 }
132 
133 static int
134 pfs_lookup_proc(pid_t pid, struct proc **p)
135 {
136 	struct proc *proc;
137 
138 	proc = pfind(pid);
139 	if (proc == NULL)
140 		return (0);
141 	if ((proc->p_flag & P_WEXIT) != 0) {
142 		PROC_UNLOCK(proc);
143 		return (0);
144 	}
145 	_PHOLD(proc);
146 	PROC_UNLOCK(proc);
147 	*p = proc;
148 	return (1);
149 }
150 
151 /*
152  * Verify permissions
153  */
154 static int
155 pfs_access(struct vop_access_args *va)
156 {
157 	struct vnode *vn = va->a_vp;
158 	struct pfs_vdata *pvd = vn->v_data;
159 	struct vattr vattr;
160 	int error;
161 
162 	PFS_TRACE(("%s", pvd->pvd_pn->pn_name));
163 	(void)pvd;
164 
165 	error = VOP_GETATTR(vn, &vattr, va->a_cred);
166 	if (error)
167 		PFS_RETURN (error);
168 	error = vaccess(vn->v_type, vattr.va_mode, vattr.va_uid, vattr.va_gid,
169 	    va->a_accmode, va->a_cred);
170 	PFS_RETURN (error);
171 }
172 
173 /*
174  * Close a file or directory
175  */
176 static int
177 pfs_close(struct vop_close_args *va)
178 {
179 	struct vnode *vn = va->a_vp;
180 	struct pfs_vdata *pvd = vn->v_data;
181 	struct pfs_node *pn = pvd->pvd_pn;
182 	struct proc *proc;
183 	int error;
184 
185 	PFS_TRACE(("%s", pn->pn_name));
186 	pfs_assert_not_owned(pn);
187 
188 	/*
189 	 * Do nothing unless this is the last close and the node has a
190 	 * last-close handler.
191 	 */
192 	if (vrefcnt(vn) > 1 || pn->pn_close == NULL)
193 		PFS_RETURN (0);
194 
195 	if (pvd->pvd_pid != NO_PID) {
196 		proc = pfind(pvd->pvd_pid);
197 	} else {
198 		proc = NULL;
199 	}
200 
201 	error = pn_close(va->a_td, proc, pn);
202 
203 	if (proc != NULL)
204 		PROC_UNLOCK(proc);
205 
206 	PFS_RETURN (error);
207 }
208 
209 /*
210  * Get file attributes
211  */
212 static int
213 pfs_getattr(struct vop_getattr_args *va)
214 {
215 	struct vnode *vn = va->a_vp;
216 	struct pfs_vdata *pvd = vn->v_data;
217 	struct pfs_node *pn = pvd->pvd_pn;
218 	struct vattr *vap = va->a_vap;
219 	struct proc *proc;
220 	int error = 0;
221 
222 	PFS_TRACE(("%s", pn->pn_name));
223 	pfs_assert_not_owned(pn);
224 
225 	if (!pfs_visible(curthread, pn, pvd->pvd_pid, &proc))
226 		PFS_RETURN (ENOENT);
227 
228 	vap->va_type = vn->v_type;
229 	vap->va_fileid = pn_fileno(pn, pvd->pvd_pid);
230 	vap->va_flags = 0;
231 	vap->va_blocksize = PAGE_SIZE;
232 	vap->va_bytes = vap->va_size = 0;
233 	vap->va_filerev = 0;
234 	vap->va_fsid = vn->v_mount->mnt_stat.f_fsid.val[0];
235 	vap->va_nlink = 1;
236 	nanotime(&vap->va_ctime);
237 	vap->va_atime = vap->va_mtime = vap->va_ctime;
238 
239 	switch (pn->pn_type) {
240 	case pfstype_procdir:
241 	case pfstype_root:
242 	case pfstype_dir:
243 #if 0
244 		pfs_lock(pn);
245 		/* compute link count */
246 		pfs_unlock(pn);
247 #endif
248 		vap->va_mode = 0555;
249 		break;
250 	case pfstype_file:
251 	case pfstype_symlink:
252 		vap->va_mode = 0444;
253 		break;
254 	default:
255 		printf("shouldn't be here!\n");
256 		vap->va_mode = 0;
257 		break;
258 	}
259 
260 	if (proc != NULL) {
261 		vap->va_uid = proc->p_ucred->cr_ruid;
262 		vap->va_gid = proc->p_ucred->cr_rgid;
263 	} else {
264 		vap->va_uid = 0;
265 		vap->va_gid = 0;
266 	}
267 
268 	if (pn->pn_attr != NULL)
269 		error = pn_attr(curthread, proc, pn, vap);
270 
271 	if(proc != NULL)
272 		PROC_UNLOCK(proc);
273 
274 	PFS_RETURN (error);
275 }
276 
277 /*
278  * Perform an ioctl
279  */
280 static int
281 pfs_ioctl(struct vop_ioctl_args *va)
282 {
283 	struct vnode *vn;
284 	struct pfs_vdata *pvd;
285 	struct pfs_node *pn;
286 	struct proc *proc;
287 	int error;
288 
289 	vn = va->a_vp;
290 	vn_lock(vn, LK_SHARED | LK_RETRY);
291 	if (VN_IS_DOOMED(vn)) {
292 		VOP_UNLOCK(vn);
293 		return (EBADF);
294 	}
295 	pvd = vn->v_data;
296 	pn = pvd->pvd_pn;
297 
298 	PFS_TRACE(("%s: %lx", pn->pn_name, va->a_command));
299 	pfs_assert_not_owned(pn);
300 
301 	if (vn->v_type != VREG) {
302 		VOP_UNLOCK(vn);
303 		PFS_RETURN (EINVAL);
304 	}
305 	KASSERT_PN_IS_FILE(pn);
306 
307 	if (pn->pn_ioctl == NULL) {
308 		VOP_UNLOCK(vn);
309 		PFS_RETURN (ENOTTY);
310 	}
311 
312 	/*
313 	 * This is necessary because process' privileges may
314 	 * have changed since the open() call.
315 	 */
316 	if (!pfs_visible(curthread, pn, pvd->pvd_pid, &proc)) {
317 		VOP_UNLOCK(vn);
318 		PFS_RETURN (EIO);
319 	}
320 
321 	error = pn_ioctl(curthread, proc, pn, va->a_command, va->a_data);
322 
323 	if (proc != NULL)
324 		PROC_UNLOCK(proc);
325 
326 	VOP_UNLOCK(vn);
327 	PFS_RETURN (error);
328 }
329 
330 /*
331  * Perform getextattr
332  */
333 static int
334 pfs_getextattr(struct vop_getextattr_args *va)
335 {
336 	struct vnode *vn = va->a_vp;
337 	struct pfs_vdata *pvd = vn->v_data;
338 	struct pfs_node *pn = pvd->pvd_pn;
339 	struct proc *proc;
340 	int error;
341 
342 	PFS_TRACE(("%s", pn->pn_name));
343 	pfs_assert_not_owned(pn);
344 
345 	/*
346 	 * This is necessary because either process' privileges may
347 	 * have changed since the open() call.
348 	 */
349 	if (!pfs_visible(curthread, pn, pvd->pvd_pid, &proc))
350 		PFS_RETURN (EIO);
351 
352 	if (pn->pn_getextattr == NULL)
353 		error = EOPNOTSUPP;
354 	else
355 		error = pn_getextattr(curthread, proc, pn,
356 		    va->a_attrnamespace, va->a_name, va->a_uio,
357 		    va->a_size, va->a_cred);
358 
359 	if (proc != NULL)
360 		PROC_UNLOCK(proc);
361 
362 	PFS_RETURN (error);
363 }
364 
365 /*
366  * Convert a vnode to its component name
367  */
368 static int
369 pfs_vptocnp(struct vop_vptocnp_args *ap)
370 {
371 	struct vnode *vp = ap->a_vp;
372 	struct vnode **dvp = ap->a_vpp;
373 	struct pfs_vdata *pvd = vp->v_data;
374 	struct pfs_node *pd = pvd->pvd_pn;
375 	struct pfs_node *pn;
376 	struct mount *mp;
377 	char *buf = ap->a_buf;
378 	size_t *buflen = ap->a_buflen;
379 	char pidbuf[PFS_NAMELEN];
380 	pid_t pid = pvd->pvd_pid;
381 	int len, i, error, locked;
382 
383 	i = *buflen;
384 	error = 0;
385 
386 	pfs_lock(pd);
387 
388 	if (vp->v_type == VDIR && pd->pn_type == pfstype_root) {
389 		*dvp = vp;
390 		vhold(*dvp);
391 		pfs_unlock(pd);
392 		PFS_RETURN (0);
393 	} else if (vp->v_type == VDIR && pd->pn_type == pfstype_procdir) {
394 		len = snprintf(pidbuf, sizeof(pidbuf), "%d", pid);
395 		i -= len;
396 		if (i < 0) {
397 			error = ENOMEM;
398 			goto failed;
399 		}
400 		bcopy(pidbuf, buf + i, len);
401 	} else {
402 		len = strlen(pd->pn_name);
403 		i -= len;
404 		if (i < 0) {
405 			error = ENOMEM;
406 			goto failed;
407 		}
408 		bcopy(pd->pn_name, buf + i, len);
409 	}
410 
411 	pn = pd->pn_parent;
412 	pfs_unlock(pd);
413 
414 	mp = vp->v_mount;
415 	error = vfs_busy(mp, 0);
416 	if (error)
417 		return (error);
418 
419 	/*
420 	 * vp is held by caller.
421 	 */
422 	locked = VOP_ISLOCKED(vp);
423 	VOP_UNLOCK(vp);
424 
425 	error = pfs_vncache_alloc(mp, dvp, pn, pid);
426 	if (error) {
427 		vn_lock(vp, locked | LK_RETRY);
428 		vfs_unbusy(mp);
429 		PFS_RETURN(error);
430 	}
431 
432 	*buflen = i;
433 	VOP_UNLOCK(*dvp);
434 	vn_lock(vp, locked | LK_RETRY);
435 	vfs_unbusy(mp);
436 
437 	PFS_RETURN (0);
438 failed:
439 	pfs_unlock(pd);
440 	PFS_RETURN(error);
441 }
442 
443 /*
444  * Look up a file or directory
445  */
446 static int
447 pfs_lookup(struct vop_cachedlookup_args *va)
448 {
449 	struct vnode *vn = va->a_dvp;
450 	struct vnode **vpp = va->a_vpp;
451 	struct componentname *cnp = va->a_cnp;
452 	struct pfs_vdata *pvd = vn->v_data;
453 	struct pfs_node *pd = pvd->pvd_pn;
454 	struct pfs_node *pn, *pdn = NULL;
455 	struct mount *mp;
456 	pid_t pid = pvd->pvd_pid;
457 	char *pname;
458 	int error, i, namelen, visible;
459 
460 	PFS_TRACE(("%.*s", (int)cnp->cn_namelen, cnp->cn_nameptr));
461 	pfs_assert_not_owned(pd);
462 
463 	if (vn->v_type != VDIR)
464 		PFS_RETURN (ENOTDIR);
465 	KASSERT_PN_IS_DIR(pd);
466 
467 	/*
468 	 * Don't support DELETE or RENAME.  CREATE is supported so
469 	 * that O_CREAT will work, but the lookup will still fail if
470 	 * the file does not exist.
471 	 */
472 	if ((cnp->cn_flags & ISLASTCN) &&
473 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
474 		PFS_RETURN (EOPNOTSUPP);
475 
476 	/* shortcut: check if the name is too long */
477 	if (cnp->cn_namelen >= PFS_NAMELEN)
478 		PFS_RETURN (ENOENT);
479 
480 	/* check that parent directory is visible... */
481 	if (!pfs_visible(curthread, pd, pvd->pvd_pid, NULL))
482 		PFS_RETURN (ENOENT);
483 
484 	/* self */
485 	namelen = cnp->cn_namelen;
486 	pname = cnp->cn_nameptr;
487 	if (namelen == 1 && pname[0] == '.') {
488 		pn = pd;
489 		*vpp = vn;
490 		VREF(vn);
491 		PFS_RETURN (0);
492 	}
493 
494 	mp = vn->v_mount;
495 
496 	/* parent */
497 	if (cnp->cn_flags & ISDOTDOT) {
498 		if (pd->pn_type == pfstype_root)
499 			PFS_RETURN (EIO);
500 		error = vfs_busy(mp, MBF_NOWAIT);
501 		if (error != 0) {
502 			vfs_ref(mp);
503 			VOP_UNLOCK(vn);
504 			error = vfs_busy(mp, 0);
505 			vn_lock(vn, LK_EXCLUSIVE | LK_RETRY);
506 			vfs_rel(mp);
507 			if (error != 0)
508 				PFS_RETURN(ENOENT);
509 			if (VN_IS_DOOMED(vn)) {
510 				vfs_unbusy(mp);
511 				PFS_RETURN(ENOENT);
512 			}
513 		}
514 		VOP_UNLOCK(vn);
515 		KASSERT(pd->pn_parent != NULL,
516 		    ("%s(): non-root directory has no parent", __func__));
517 		/*
518 		 * This one is tricky.  Descendents of procdir nodes
519 		 * inherit their parent's process affinity, but
520 		 * there's no easy reverse mapping.  For simplicity,
521 		 * we assume that if this node is a procdir, its
522 		 * parent isn't (which is correct as long as
523 		 * descendents of procdir nodes are never procdir
524 		 * nodes themselves)
525 		 */
526 		if (pd->pn_type == pfstype_procdir)
527 			pid = NO_PID;
528 		pfs_lock(pd);
529 		pn = pd->pn_parent;
530 		pfs_unlock(pd);
531 		goto got_pnode;
532 	}
533 
534 	pfs_lock(pd);
535 
536 	/* named node */
537 	for (pn = pd->pn_nodes; pn != NULL; pn = pn->pn_next)
538 		if (pn->pn_type == pfstype_procdir)
539 			pdn = pn;
540 		else if (pn->pn_name[namelen] == '\0' &&
541 		    bcmp(pname, pn->pn_name, namelen) == 0) {
542 			pfs_unlock(pd);
543 			goto got_pnode;
544 		}
545 
546 	/* process dependent node */
547 	if ((pn = pdn) != NULL) {
548 		pid = 0;
549 		for (pid = 0, i = 0; i < namelen && isdigit(pname[i]); ++i)
550 			if ((pid = pid * 10 + pname[i] - '0') > PID_MAX)
551 				break;
552 		if (i == cnp->cn_namelen) {
553 			pfs_unlock(pd);
554 			goto got_pnode;
555 		}
556 	}
557 
558 	pfs_unlock(pd);
559 
560 	PFS_RETURN (ENOENT);
561 
562  got_pnode:
563 	pfs_assert_not_owned(pd);
564 	pfs_assert_not_owned(pn);
565 	visible = pfs_visible(curthread, pn, pid, NULL);
566 	if (!visible) {
567 		error = ENOENT;
568 		goto failed;
569 	}
570 
571 	error = pfs_vncache_alloc(mp, vpp, pn, pid);
572 	if (error)
573 		goto failed;
574 
575 	if (cnp->cn_flags & ISDOTDOT) {
576 		vfs_unbusy(mp);
577 		vn_lock(vn, LK_EXCLUSIVE | LK_RETRY);
578 		if (VN_IS_DOOMED(vn)) {
579 			vput(*vpp);
580 			*vpp = NULL;
581 			PFS_RETURN(ENOENT);
582 		}
583 	}
584 	if (cnp->cn_flags & MAKEENTRY && !VN_IS_DOOMED(vn))
585 		cache_enter(vn, *vpp, cnp);
586 	PFS_RETURN (0);
587  failed:
588 	if (cnp->cn_flags & ISDOTDOT) {
589 		vfs_unbusy(mp);
590 		vn_lock(vn, LK_EXCLUSIVE | LK_RETRY);
591 		*vpp = NULL;
592 	}
593 	PFS_RETURN(error);
594 }
595 
596 /*
597  * Open a file or directory.
598  */
599 static int
600 pfs_open(struct vop_open_args *va)
601 {
602 	struct vnode *vn = va->a_vp;
603 	struct pfs_vdata *pvd = vn->v_data;
604 	struct pfs_node *pn = pvd->pvd_pn;
605 	int mode = va->a_mode;
606 
607 	PFS_TRACE(("%s (mode 0x%x)", pn->pn_name, mode));
608 	pfs_assert_not_owned(pn);
609 
610 	/* check if the requested mode is permitted */
611 	if (((mode & FREAD) && !(mode & PFS_RD)) ||
612 	    ((mode & FWRITE) && !(mode & PFS_WR)))
613 		PFS_RETURN (EPERM);
614 
615 	/* we don't support locking */
616 	if ((mode & O_SHLOCK) || (mode & O_EXLOCK))
617 		PFS_RETURN (EOPNOTSUPP);
618 
619 	PFS_RETURN (0);
620 }
621 
622 struct sbuf_seek_helper {
623 	off_t		skip_bytes;
624 	struct uio	*uio;
625 };
626 
627 static int
628 pfs_sbuf_uio_drain(void *arg, const char *data, int len)
629 {
630 	struct sbuf_seek_helper *ssh;
631 	struct uio *uio;
632 	int error, skipped;
633 
634 	ssh = arg;
635 	uio = ssh->uio;
636 	skipped = 0;
637 
638 	/* Need to discard first uio_offset bytes. */
639 	if (ssh->skip_bytes > 0) {
640 		if (ssh->skip_bytes >= len) {
641 			ssh->skip_bytes -= len;
642 			return (len);
643 		}
644 
645 		data += ssh->skip_bytes;
646 		len -= ssh->skip_bytes;
647 		skipped = ssh->skip_bytes;
648 		ssh->skip_bytes = 0;
649 	}
650 
651 	error = uiomove(__DECONST(void *, data), len, uio);
652 	if (error != 0)
653 		return (-error);
654 
655 	/*
656 	 * The fill function has more to emit, but the reader is finished.
657 	 * This is similar to the truncated read case for non-draining PFS
658 	 * sbufs, and should be handled appropriately in fill-routines.
659 	 */
660 	if (uio->uio_resid == 0)
661 		return (-ENOBUFS);
662 
663 	return (skipped + len);
664 }
665 
666 /*
667  * Read from a file
668  */
669 static int
670 pfs_read(struct vop_read_args *va)
671 {
672 	struct vnode *vn = va->a_vp;
673 	struct pfs_vdata *pvd = vn->v_data;
674 	struct pfs_node *pn = pvd->pvd_pn;
675 	struct uio *uio = va->a_uio;
676 	struct proc *proc;
677 	struct sbuf *sb = NULL;
678 	int error, locked;
679 	off_t buflen, buflim;
680 	struct sbuf_seek_helper ssh;
681 
682 	PFS_TRACE(("%s", pn->pn_name));
683 	pfs_assert_not_owned(pn);
684 
685 	if (vn->v_type != VREG)
686 		PFS_RETURN (EINVAL);
687 	KASSERT_PN_IS_FILE(pn);
688 
689 	if (!(pn->pn_flags & PFS_RD))
690 		PFS_RETURN (EBADF);
691 
692 	if (pn->pn_fill == NULL)
693 		PFS_RETURN (EIO);
694 
695 	/*
696 	 * This is necessary because either process' privileges may
697 	 * have changed since the open() call.
698 	 */
699 	if (!pfs_visible(curthread, pn, pvd->pvd_pid, &proc))
700 		PFS_RETURN (EIO);
701 	if (proc != NULL) {
702 		_PHOLD(proc);
703 		PROC_UNLOCK(proc);
704 	}
705 
706 	vhold(vn);
707 	locked = VOP_ISLOCKED(vn);
708 	VOP_UNLOCK(vn);
709 
710 	if (pn->pn_flags & PFS_RAWRD) {
711 		PFS_TRACE(("%zd resid", uio->uio_resid));
712 		error = pn_fill(curthread, proc, pn, NULL, uio);
713 		PFS_TRACE(("%zd resid", uio->uio_resid));
714 		goto ret;
715 	}
716 
717 	if (uio->uio_resid < 0 || uio->uio_offset < 0 ||
718 	    uio->uio_resid > OFF_MAX - uio->uio_offset) {
719 		error = EINVAL;
720 		goto ret;
721 	}
722 	buflen = uio->uio_offset + uio->uio_resid + 1;
723 	if (pn->pn_flags & PFS_AUTODRAIN)
724 		/*
725 		 * We can use a smaller buffer if we can stream output to the
726 		 * consumer.
727 		 */
728 		buflim = PAGE_SIZE;
729 	else
730 		buflim = PFS_MAXBUFSIZ;
731 	if (buflen > buflim)
732 		buflen = buflim;
733 
734 	sb = sbuf_new(sb, NULL, buflen, 0);
735 	if (sb == NULL) {
736 		error = EIO;
737 		goto ret;
738 	}
739 
740 	if (pn->pn_flags & PFS_AUTODRAIN) {
741 		ssh.skip_bytes = uio->uio_offset;
742 		ssh.uio = uio;
743 		sbuf_set_drain(sb, pfs_sbuf_uio_drain, &ssh);
744 	}
745 
746 	error = pn_fill(curthread, proc, pn, sb, uio);
747 
748 	if (error) {
749 		sbuf_delete(sb);
750 		goto ret;
751 	}
752 
753 	/*
754 	 * XXX: If the buffer overflowed, sbuf_len() will not return
755 	 * the data length. Then just use the full length because an
756 	 * overflowed sbuf must be full.
757 	 */
758 	error = sbuf_finish(sb);
759 	if ((pn->pn_flags & PFS_AUTODRAIN)) {
760 		/*
761 		 * ENOBUFS just indicates early termination of the fill
762 		 * function as the caller's buffer was already filled.  Squash
763 		 * to zero.
764 		 */
765 		if (uio->uio_resid == 0 && error == ENOBUFS)
766 			error = 0;
767 	} else {
768 		if (error == 0)
769 			buflen = sbuf_len(sb);
770 		else
771 			/* The trailing byte is not valid. */
772 			buflen--;
773 		error = uiomove_frombuf(sbuf_data(sb), buflen, uio);
774 	}
775 	sbuf_delete(sb);
776 ret:
777 	vn_lock(vn, locked | LK_RETRY);
778 	vdrop(vn);
779 	if (proc != NULL)
780 		PRELE(proc);
781 	PFS_RETURN (error);
782 }
783 
784 /*
785  * Iterate through directory entries
786  */
787 static int
788 pfs_iterate(struct thread *td, struct proc *proc, struct pfs_node *pd,
789 	    struct pfs_node **pn, struct proc **p)
790 {
791 	int visible;
792 
793 	sx_assert(&allproc_lock, SX_SLOCKED);
794 	pfs_assert_owned(pd);
795  again:
796 	if (*pn == NULL) {
797 		/* first node */
798 		*pn = pd->pn_nodes;
799 	} else if ((*pn)->pn_type != pfstype_procdir) {
800 		/* next node */
801 		*pn = (*pn)->pn_next;
802 	}
803 	if (*pn != NULL && (*pn)->pn_type == pfstype_procdir) {
804 		/* next process */
805 		if (*p == NULL)
806 			*p = LIST_FIRST(&allproc);
807 		else
808 			*p = LIST_NEXT(*p, p_list);
809 		/* out of processes: next node */
810 		if (*p == NULL)
811 			*pn = (*pn)->pn_next;
812 		else
813 			PROC_LOCK(*p);
814 	}
815 
816 	if ((*pn) == NULL)
817 		return (-1);
818 
819 	if (*p != NULL) {
820 		visible = pfs_visible_proc(td, *pn, *p);
821 		PROC_UNLOCK(*p);
822 	} else if (proc != NULL) {
823 		visible = pfs_visible_proc(td, *pn, proc);
824 	} else {
825 		visible = pn_vis(td, NULL, *pn);
826 	}
827 	if (!visible)
828 		goto again;
829 
830 	return (0);
831 }
832 
833 /* Directory entry list */
834 struct pfsentry {
835 	STAILQ_ENTRY(pfsentry)	link;
836 	struct dirent		entry;
837 };
838 STAILQ_HEAD(pfsdirentlist, pfsentry);
839 
840 /*
841  * Return directory entries.
842  */
843 static int
844 pfs_readdir(struct vop_readdir_args *va)
845 {
846 	struct vnode *vn = va->a_vp;
847 	struct pfs_vdata *pvd = vn->v_data;
848 	struct pfs_node *pd = pvd->pvd_pn;
849 	pid_t pid = pvd->pvd_pid;
850 	struct proc *p, *proc;
851 	struct pfs_node *pn;
852 	struct uio *uio;
853 	struct pfsentry *pfsent, *pfsent2;
854 	struct pfsdirentlist lst;
855 	off_t offset;
856 	int error, i, resid;
857 
858 	STAILQ_INIT(&lst);
859 	error = 0;
860 	KASSERT(pd->pn_info == vn->v_mount->mnt_data,
861 	    ("%s(): pn_info does not match mountpoint", __func__));
862 	PFS_TRACE(("%s pid %lu", pd->pn_name, (unsigned long)pid));
863 	pfs_assert_not_owned(pd);
864 
865 	if (vn->v_type != VDIR)
866 		PFS_RETURN (ENOTDIR);
867 	KASSERT_PN_IS_DIR(pd);
868 	uio = va->a_uio;
869 
870 	/* only allow reading entire entries */
871 	offset = uio->uio_offset;
872 	resid = uio->uio_resid;
873 	if (offset < 0 || offset % PFS_DELEN != 0 ||
874 	    (resid && resid < PFS_DELEN))
875 		PFS_RETURN (EINVAL);
876 	if (resid == 0)
877 		PFS_RETURN (0);
878 
879 	proc = NULL;
880 	if (pid != NO_PID && !pfs_lookup_proc(pid, &proc))
881 		PFS_RETURN (ENOENT);
882 
883 	sx_slock(&allproc_lock);
884 	pfs_lock(pd);
885 
886 	KASSERT(pid == NO_PID || proc != NULL,
887 	    ("%s(): no process for pid %lu", __func__, (unsigned long)pid));
888 
889 	if (pid != NO_PID) {
890 		PROC_LOCK(proc);
891 
892 		/* check if the directory is visible to the caller */
893 		if (!pfs_visible_proc(curthread, pd, proc)) {
894 			_PRELE(proc);
895 			PROC_UNLOCK(proc);
896 			pfs_unlock(pd);
897 			sx_sunlock(&allproc_lock);
898 			PFS_RETURN (ENOENT);
899 		}
900 	}
901 
902 	/* skip unwanted entries */
903 	for (pn = NULL, p = NULL; offset > 0; offset -= PFS_DELEN) {
904 		if (pfs_iterate(curthread, proc, pd, &pn, &p) == -1) {
905 			/* nothing left... */
906 			if (proc != NULL) {
907 				_PRELE(proc);
908 				PROC_UNLOCK(proc);
909 			}
910 			pfs_unlock(pd);
911 			sx_sunlock(&allproc_lock);
912 			PFS_RETURN (0);
913 		}
914 	}
915 
916 	/* fill in entries */
917 	while (pfs_iterate(curthread, proc, pd, &pn, &p) != -1 &&
918 	    resid >= PFS_DELEN) {
919 		if ((pfsent = malloc(sizeof(struct pfsentry), M_IOV,
920 		    M_NOWAIT | M_ZERO)) == NULL) {
921 			error = ENOMEM;
922 			break;
923 		}
924 		pfsent->entry.d_reclen = PFS_DELEN;
925 		pfsent->entry.d_fileno = pn_fileno(pn, pid);
926 		/* PFS_DELEN was picked to fit PFS_NAMLEN */
927 		for (i = 0; i < PFS_NAMELEN - 1 && pn->pn_name[i] != '\0'; ++i)
928 			pfsent->entry.d_name[i] = pn->pn_name[i];
929 		pfsent->entry.d_namlen = i;
930 		/* NOTE: d_off is the offset of the *next* entry. */
931 		pfsent->entry.d_off = offset + PFS_DELEN;
932 		switch (pn->pn_type) {
933 		case pfstype_procdir:
934 			KASSERT(p != NULL,
935 			    ("reached procdir node with p == NULL"));
936 			pfsent->entry.d_namlen = snprintf(pfsent->entry.d_name,
937 			    PFS_NAMELEN, "%d", p->p_pid);
938 			/* fall through */
939 		case pfstype_root:
940 		case pfstype_dir:
941 		case pfstype_this:
942 		case pfstype_parent:
943 			pfsent->entry.d_type = DT_DIR;
944 			break;
945 		case pfstype_file:
946 			pfsent->entry.d_type = DT_REG;
947 			break;
948 		case pfstype_symlink:
949 			pfsent->entry.d_type = DT_LNK;
950 			break;
951 		default:
952 			panic("%s has unexpected node type: %d", pn->pn_name, pn->pn_type);
953 		}
954 		PFS_TRACE(("%s", pfsent->entry.d_name));
955 		dirent_terminate(&pfsent->entry);
956 		STAILQ_INSERT_TAIL(&lst, pfsent, link);
957 		offset += PFS_DELEN;
958 		resid -= PFS_DELEN;
959 	}
960 	if (proc != NULL) {
961 		_PRELE(proc);
962 		PROC_UNLOCK(proc);
963 	}
964 	pfs_unlock(pd);
965 	sx_sunlock(&allproc_lock);
966 	i = 0;
967 	STAILQ_FOREACH_SAFE(pfsent, &lst, link, pfsent2) {
968 		if (error == 0)
969 			error = uiomove(&pfsent->entry, PFS_DELEN, uio);
970 		free(pfsent, M_IOV);
971 		i++;
972 	}
973 	PFS_TRACE(("%ju bytes", (uintmax_t)(i * PFS_DELEN)));
974 	PFS_RETURN (error);
975 }
976 
977 /*
978  * Read a symbolic link
979  */
980 static int
981 pfs_readlink(struct vop_readlink_args *va)
982 {
983 	struct vnode *vn = va->a_vp;
984 	struct pfs_vdata *pvd = vn->v_data;
985 	struct pfs_node *pn = pvd->pvd_pn;
986 	struct uio *uio = va->a_uio;
987 	struct proc *proc = NULL;
988 	char buf[PATH_MAX];
989 	struct sbuf sb;
990 	int error, locked;
991 
992 	PFS_TRACE(("%s", pn->pn_name));
993 	pfs_assert_not_owned(pn);
994 
995 	if (vn->v_type != VLNK)
996 		PFS_RETURN (EINVAL);
997 	KASSERT_PN_IS_LINK(pn);
998 
999 	if (pn->pn_fill == NULL)
1000 		PFS_RETURN (EIO);
1001 
1002 	if (pvd->pvd_pid != NO_PID) {
1003 		if ((proc = pfind(pvd->pvd_pid)) == NULL)
1004 			PFS_RETURN (EIO);
1005 		if (proc->p_flag & P_WEXIT) {
1006 			PROC_UNLOCK(proc);
1007 			PFS_RETURN (EIO);
1008 		}
1009 		_PHOLD(proc);
1010 		PROC_UNLOCK(proc);
1011 	}
1012 	vhold(vn);
1013 	locked = VOP_ISLOCKED(vn);
1014 	VOP_UNLOCK(vn);
1015 
1016 	/* sbuf_new() can't fail with a static buffer */
1017 	sbuf_new(&sb, buf, sizeof buf, 0);
1018 
1019 	error = pn_fill(curthread, proc, pn, &sb, NULL);
1020 
1021 	if (proc != NULL)
1022 		PRELE(proc);
1023 	vn_lock(vn, locked | LK_RETRY);
1024 	vdrop(vn);
1025 
1026 	if (error) {
1027 		sbuf_delete(&sb);
1028 		PFS_RETURN (error);
1029 	}
1030 
1031 	if (sbuf_finish(&sb) != 0) {
1032 		sbuf_delete(&sb);
1033 		PFS_RETURN (ENAMETOOLONG);
1034 	}
1035 
1036 	error = uiomove_frombuf(sbuf_data(&sb), sbuf_len(&sb), uio);
1037 	sbuf_delete(&sb);
1038 	PFS_RETURN (error);
1039 }
1040 
1041 /*
1042  * Reclaim a vnode
1043  */
1044 static int
1045 pfs_reclaim(struct vop_reclaim_args *va)
1046 {
1047 	struct vnode *vn = va->a_vp;
1048 	struct pfs_vdata *pvd = vn->v_data;
1049 	struct pfs_node *pn = pvd->pvd_pn;
1050 
1051 	PFS_TRACE(("%s", pn->pn_name));
1052 	pfs_assert_not_owned(pn);
1053 
1054 	return (pfs_vncache_free(va->a_vp));
1055 }
1056 
1057 /*
1058  * Set attributes
1059  */
1060 static int
1061 pfs_setattr(struct vop_setattr_args *va)
1062 {
1063 	struct vnode *vn = va->a_vp;
1064 	struct pfs_vdata *pvd = vn->v_data;
1065 	struct pfs_node *pn = pvd->pvd_pn;
1066 
1067 	PFS_TRACE(("%s", pn->pn_name));
1068 	pfs_assert_not_owned(pn);
1069 
1070 	/* Silently ignore unchangeable attributes. */
1071 	PFS_RETURN (0);
1072 }
1073 
1074 /*
1075  * Write to a file
1076  */
1077 static int
1078 pfs_write(struct vop_write_args *va)
1079 {
1080 	struct vnode *vn = va->a_vp;
1081 	struct pfs_vdata *pvd = vn->v_data;
1082 	struct pfs_node *pn = pvd->pvd_pn;
1083 	struct uio *uio = va->a_uio;
1084 	struct proc *proc;
1085 	struct sbuf sb;
1086 	int error;
1087 
1088 	PFS_TRACE(("%s", pn->pn_name));
1089 	pfs_assert_not_owned(pn);
1090 
1091 	if (vn->v_type != VREG)
1092 		PFS_RETURN (EINVAL);
1093 	KASSERT_PN_IS_FILE(pn);
1094 
1095 	if (!(pn->pn_flags & PFS_WR))
1096 		PFS_RETURN (EBADF);
1097 
1098 	if (pn->pn_fill == NULL)
1099 		PFS_RETURN (EIO);
1100 
1101 	if (uio->uio_resid > PFS_MAXBUFSIZ)
1102 		PFS_RETURN (EIO);
1103 
1104 	/*
1105 	 * This is necessary because either process' privileges may
1106 	 * have changed since the open() call.
1107 	 */
1108 	if (!pfs_visible(curthread, pn, pvd->pvd_pid, &proc))
1109 		PFS_RETURN (EIO);
1110 	if (proc != NULL) {
1111 		_PHOLD(proc);
1112 		PROC_UNLOCK(proc);
1113 	}
1114 
1115 	if (pn->pn_flags & PFS_RAWWR) {
1116 		error = pn_fill(curthread, proc, pn, NULL, uio);
1117 		if (proc != NULL)
1118 			PRELE(proc);
1119 		PFS_RETURN (error);
1120 	}
1121 
1122 	sbuf_uionew(&sb, uio, &error);
1123 	if (error) {
1124 		if (proc != NULL)
1125 			PRELE(proc);
1126 		PFS_RETURN (error);
1127 	}
1128 
1129 	error = pn_fill(curthread, proc, pn, &sb, uio);
1130 
1131 	sbuf_delete(&sb);
1132 	if (proc != NULL)
1133 		PRELE(proc);
1134 	PFS_RETURN (error);
1135 }
1136 
1137 /*
1138  * Vnode operations
1139  */
1140 struct vop_vector pfs_vnodeops = {
1141 	.vop_default =		&default_vnodeops,
1142 
1143 	.vop_access =		pfs_access,
1144 	.vop_cachedlookup =	pfs_lookup,
1145 	.vop_close =		pfs_close,
1146 	.vop_create =		VOP_EOPNOTSUPP,
1147 	.vop_getattr =		pfs_getattr,
1148 	.vop_getextattr =	pfs_getextattr,
1149 	.vop_ioctl =		pfs_ioctl,
1150 	.vop_link =		VOP_EOPNOTSUPP,
1151 	.vop_lookup =		vfs_cache_lookup,
1152 	.vop_mkdir =		VOP_EOPNOTSUPP,
1153 	.vop_mknod =		VOP_EOPNOTSUPP,
1154 	.vop_open =		pfs_open,
1155 	.vop_read =		pfs_read,
1156 	.vop_readdir =		pfs_readdir,
1157 	.vop_readlink =		pfs_readlink,
1158 	.vop_reclaim =		pfs_reclaim,
1159 	.vop_remove =		VOP_EOPNOTSUPP,
1160 	.vop_rename =		VOP_EOPNOTSUPP,
1161 	.vop_rmdir =		VOP_EOPNOTSUPP,
1162 	.vop_setattr =		pfs_setattr,
1163 	.vop_symlink =		VOP_EOPNOTSUPP,
1164 	.vop_vptocnp =		pfs_vptocnp,
1165 	.vop_write =		pfs_write,
1166 	.vop_add_writecount =	vop_stdadd_writecount_nomsync,
1167 	/* XXX I've probably forgotten a few that need VOP_EOPNOTSUPP */
1168 };
1169 VFS_VOP_VECTOR_REGISTER(pfs_vnodeops);
1170