xref: /freebsd/sys/fs/pseudofs/pseudofs_vnops.c (revision d8a0fe102c0cfdfcd5b818f850eff09d8536c9bc)
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 /*
72  * Returns the fileno, adjusted for target pid
73  */
74 static uint32_t
75 pn_fileno(struct pfs_node *pn, pid_t pid)
76 {
77 
78 	KASSERT(pn->pn_fileno > 0,
79 	    ("%s(): no fileno allocated", __func__));
80 	if (pid != NO_PID)
81 		return (pn->pn_fileno * NO_PID + pid);
82 	return (pn->pn_fileno);
83 }
84 
85 /*
86  * Returns non-zero if given file is visible to given thread.
87  */
88 static int
89 pfs_visible_proc(struct thread *td, struct pfs_node *pn, struct proc *proc)
90 {
91 	int visible;
92 
93 	if (proc == NULL)
94 		return (0);
95 
96 	PROC_LOCK_ASSERT(proc, MA_OWNED);
97 
98 	visible = ((proc->p_flag & P_WEXIT) == 0);
99 	if (visible)
100 		visible = (p_cansee(td, proc) == 0);
101 	if (visible && pn->pn_vis != NULL)
102 		visible = pn_vis(td, proc, pn);
103 	if (!visible)
104 		return (0);
105 	return (1);
106 }
107 
108 static int
109 pfs_visible(struct thread *td, struct pfs_node *pn, pid_t pid,
110     bool allproc_locked, struct proc **p)
111 {
112 	struct proc *proc;
113 
114 	PFS_TRACE(("%s (pid: %d, req: %d)",
115 	    pn->pn_name, pid, td->td_proc->p_pid));
116 
117 	if (p)
118 		*p = NULL;
119 	if (pid == NO_PID)
120 		PFS_RETURN (1);
121 	proc = allproc_locked ? pfind_locked(pid) : pfind(pid);
122 	if (proc == NULL)
123 		PFS_RETURN (0);
124 	if (pfs_visible_proc(td, pn, proc)) {
125 		if (p)
126 			*p = proc;
127 		else
128 			PROC_UNLOCK(proc);
129 		PFS_RETURN (1);
130 	}
131 	PROC_UNLOCK(proc);
132 	PFS_RETURN (0);
133 }
134 
135 /*
136  * Verify permissions
137  */
138 static int
139 pfs_access(struct vop_access_args *va)
140 {
141 	struct vnode *vn = va->a_vp;
142 	struct pfs_vdata *pvd = vn->v_data;
143 	struct vattr vattr;
144 	int error;
145 
146 	PFS_TRACE(("%s", pvd->pvd_pn->pn_name));
147 	(void)pvd;
148 
149 	error = VOP_GETATTR(vn, &vattr, va->a_cred);
150 	if (error)
151 		PFS_RETURN (error);
152 	error = vaccess(vn->v_type, vattr.va_mode, vattr.va_uid,
153 	    vattr.va_gid, va->a_accmode, va->a_cred, NULL);
154 	PFS_RETURN (error);
155 }
156 
157 /*
158  * Close a file or directory
159  */
160 static int
161 pfs_close(struct vop_close_args *va)
162 {
163 	struct vnode *vn = va->a_vp;
164 	struct pfs_vdata *pvd = vn->v_data;
165 	struct pfs_node *pn = pvd->pvd_pn;
166 	struct proc *proc;
167 	int error;
168 
169 	PFS_TRACE(("%s", pn->pn_name));
170 	pfs_assert_not_owned(pn);
171 
172 	/*
173 	 * Do nothing unless this is the last close and the node has a
174 	 * last-close handler.
175 	 */
176 	if (vrefcnt(vn) > 1 || pn->pn_close == NULL)
177 		PFS_RETURN (0);
178 
179 	if (pvd->pvd_pid != NO_PID) {
180 		proc = pfind(pvd->pvd_pid);
181 	} else {
182 		proc = NULL;
183 	}
184 
185 	error = pn_close(va->a_td, proc, pn);
186 
187 	if (proc != NULL)
188 		PROC_UNLOCK(proc);
189 
190 	PFS_RETURN (error);
191 }
192 
193 /*
194  * Get file attributes
195  */
196 static int
197 pfs_getattr(struct vop_getattr_args *va)
198 {
199 	struct vnode *vn = va->a_vp;
200 	struct pfs_vdata *pvd = vn->v_data;
201 	struct pfs_node *pn = pvd->pvd_pn;
202 	struct vattr *vap = va->a_vap;
203 	struct proc *proc;
204 	int error = 0;
205 
206 	PFS_TRACE(("%s", pn->pn_name));
207 	pfs_assert_not_owned(pn);
208 
209 	if (!pfs_visible(curthread, pn, pvd->pvd_pid, false, &proc))
210 		PFS_RETURN (ENOENT);
211 
212 	vap->va_type = vn->v_type;
213 	vap->va_fileid = pn_fileno(pn, pvd->pvd_pid);
214 	vap->va_flags = 0;
215 	vap->va_blocksize = PAGE_SIZE;
216 	vap->va_bytes = vap->va_size = 0;
217 	vap->va_filerev = 0;
218 	vap->va_fsid = vn->v_mount->mnt_stat.f_fsid.val[0];
219 	vap->va_nlink = 1;
220 	nanotime(&vap->va_ctime);
221 	vap->va_atime = vap->va_mtime = vap->va_ctime;
222 
223 	switch (pn->pn_type) {
224 	case pfstype_procdir:
225 	case pfstype_root:
226 	case pfstype_dir:
227 #if 0
228 		pfs_lock(pn);
229 		/* compute link count */
230 		pfs_unlock(pn);
231 #endif
232 		vap->va_mode = 0555;
233 		break;
234 	case pfstype_file:
235 	case pfstype_symlink:
236 		vap->va_mode = 0444;
237 		break;
238 	default:
239 		printf("shouldn't be here!\n");
240 		vap->va_mode = 0;
241 		break;
242 	}
243 
244 	if (proc != NULL) {
245 		vap->va_uid = proc->p_ucred->cr_ruid;
246 		vap->va_gid = proc->p_ucred->cr_rgid;
247 	} else {
248 		vap->va_uid = 0;
249 		vap->va_gid = 0;
250 	}
251 
252 	if (pn->pn_attr != NULL)
253 		error = pn_attr(curthread, proc, pn, vap);
254 
255 	if(proc != NULL)
256 		PROC_UNLOCK(proc);
257 
258 	PFS_RETURN (error);
259 }
260 
261 /*
262  * Perform an ioctl
263  */
264 static int
265 pfs_ioctl(struct vop_ioctl_args *va)
266 {
267 	struct vnode *vn;
268 	struct pfs_vdata *pvd;
269 	struct pfs_node *pn;
270 	struct proc *proc;
271 	int error;
272 
273 	vn = va->a_vp;
274 	vn_lock(vn, LK_SHARED | LK_RETRY);
275 	if (vn->v_iflag & VI_DOOMED) {
276 		VOP_UNLOCK(vn, 0);
277 		return (EBADF);
278 	}
279 	pvd = vn->v_data;
280 	pn = pvd->pvd_pn;
281 
282 	PFS_TRACE(("%s: %lx", pn->pn_name, va->a_command));
283 	pfs_assert_not_owned(pn);
284 
285 	if (vn->v_type != VREG) {
286 		VOP_UNLOCK(vn, 0);
287 		PFS_RETURN (EINVAL);
288 	}
289 	KASSERT_PN_IS_FILE(pn);
290 
291 	if (pn->pn_ioctl == NULL) {
292 		VOP_UNLOCK(vn, 0);
293 		PFS_RETURN (ENOTTY);
294 	}
295 
296 	/*
297 	 * This is necessary because process' privileges may
298 	 * have changed since the open() call.
299 	 */
300 	if (!pfs_visible(curthread, pn, pvd->pvd_pid, false, &proc)) {
301 		VOP_UNLOCK(vn, 0);
302 		PFS_RETURN (EIO);
303 	}
304 
305 	error = pn_ioctl(curthread, proc, pn, va->a_command, va->a_data);
306 
307 	if (proc != NULL)
308 		PROC_UNLOCK(proc);
309 
310 	VOP_UNLOCK(vn, 0);
311 	PFS_RETURN (error);
312 }
313 
314 /*
315  * Perform getextattr
316  */
317 static int
318 pfs_getextattr(struct vop_getextattr_args *va)
319 {
320 	struct vnode *vn = va->a_vp;
321 	struct pfs_vdata *pvd = vn->v_data;
322 	struct pfs_node *pn = pvd->pvd_pn;
323 	struct proc *proc;
324 	int error;
325 
326 	PFS_TRACE(("%s", pn->pn_name));
327 	pfs_assert_not_owned(pn);
328 
329 	/*
330 	 * This is necessary because either process' privileges may
331 	 * have changed since the open() call.
332 	 */
333 	if (!pfs_visible(curthread, pn, pvd->pvd_pid, false, &proc))
334 		PFS_RETURN (EIO);
335 
336 	if (pn->pn_getextattr == NULL)
337 		error = EOPNOTSUPP;
338 	else
339 		error = pn_getextattr(curthread, proc, pn,
340 		    va->a_attrnamespace, va->a_name, va->a_uio,
341 		    va->a_size, va->a_cred);
342 
343 	if (proc != NULL)
344 		PROC_UNLOCK(proc);
345 
346 	PFS_RETURN (error);
347 }
348 
349 /*
350  * Convert a vnode to its component name
351  */
352 static int
353 pfs_vptocnp(struct vop_vptocnp_args *ap)
354 {
355 	struct vnode *vp = ap->a_vp;
356 	struct vnode **dvp = ap->a_vpp;
357 	struct pfs_vdata *pvd = vp->v_data;
358 	struct pfs_node *pd = pvd->pvd_pn;
359 	struct pfs_node *pn;
360 	struct mount *mp;
361 	char *buf = ap->a_buf;
362 	int *buflen = ap->a_buflen;
363 	char pidbuf[PFS_NAMELEN];
364 	pid_t pid = pvd->pvd_pid;
365 	int len, i, error, locked;
366 
367 	i = *buflen;
368 	error = 0;
369 
370 	pfs_lock(pd);
371 
372 	if (vp->v_type == VDIR && pd->pn_type == pfstype_root) {
373 		*dvp = vp;
374 		vhold(*dvp);
375 		pfs_unlock(pd);
376 		PFS_RETURN (0);
377 	} else if (vp->v_type == VDIR && pd->pn_type == pfstype_procdir) {
378 		len = snprintf(pidbuf, sizeof(pidbuf), "%d", pid);
379 		i -= len;
380 		if (i < 0) {
381 			error = ENOMEM;
382 			goto failed;
383 		}
384 		bcopy(pidbuf, buf + i, len);
385 	} else {
386 		len = strlen(pd->pn_name);
387 		i -= len;
388 		if (i < 0) {
389 			error = ENOMEM;
390 			goto failed;
391 		}
392 		bcopy(pd->pn_name, buf + i, len);
393 	}
394 
395 	pn = pd->pn_parent;
396 	pfs_unlock(pd);
397 
398 	mp = vp->v_mount;
399 	error = vfs_busy(mp, 0);
400 	if (error)
401 		return (error);
402 
403 	/*
404 	 * vp is held by caller.
405 	 */
406 	locked = VOP_ISLOCKED(vp);
407 	VOP_UNLOCK(vp, 0);
408 
409 	error = pfs_vncache_alloc(mp, dvp, pn, pid);
410 	if (error) {
411 		vn_lock(vp, locked | LK_RETRY);
412 		vfs_unbusy(mp);
413 		PFS_RETURN(error);
414 	}
415 
416 	*buflen = i;
417 	VOP_UNLOCK(*dvp, 0);
418 	vn_lock(vp, locked | LK_RETRY);
419 	vfs_unbusy(mp);
420 
421 	PFS_RETURN (0);
422 failed:
423 	pfs_unlock(pd);
424 	PFS_RETURN(error);
425 }
426 
427 /*
428  * Look up a file or directory
429  */
430 static int
431 pfs_lookup(struct vop_cachedlookup_args *va)
432 {
433 	struct vnode *vn = va->a_dvp;
434 	struct vnode **vpp = va->a_vpp;
435 	struct componentname *cnp = va->a_cnp;
436 	struct pfs_vdata *pvd = vn->v_data;
437 	struct pfs_node *pd = pvd->pvd_pn;
438 	struct pfs_node *pn, *pdn = NULL;
439 	struct mount *mp;
440 	pid_t pid = pvd->pvd_pid;
441 	char *pname;
442 	int error, i, namelen, visible;
443 
444 	PFS_TRACE(("%.*s", (int)cnp->cn_namelen, cnp->cn_nameptr));
445 	pfs_assert_not_owned(pd);
446 
447 	if (vn->v_type != VDIR)
448 		PFS_RETURN (ENOTDIR);
449 	KASSERT_PN_IS_DIR(pd);
450 
451 	error = VOP_ACCESS(vn, VEXEC, cnp->cn_cred, cnp->cn_thread);
452 	if (error)
453 		PFS_RETURN (error);
454 
455 	/*
456 	 * Don't support DELETE or RENAME.  CREATE is supported so
457 	 * that O_CREAT will work, but the lookup will still fail if
458 	 * the file does not exist.
459 	 */
460 	if ((cnp->cn_flags & ISLASTCN) &&
461 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
462 		PFS_RETURN (EOPNOTSUPP);
463 
464 	/* shortcut: check if the name is too long */
465 	if (cnp->cn_namelen >= PFS_NAMELEN)
466 		PFS_RETURN (ENOENT);
467 
468 	/* check that parent directory is visible... */
469 	if (!pfs_visible(curthread, pd, pvd->pvd_pid, false, NULL))
470 		PFS_RETURN (ENOENT);
471 
472 	/* self */
473 	namelen = cnp->cn_namelen;
474 	pname = cnp->cn_nameptr;
475 	if (namelen == 1 && pname[0] == '.') {
476 		pn = pd;
477 		*vpp = vn;
478 		VREF(vn);
479 		PFS_RETURN (0);
480 	}
481 
482 	mp = vn->v_mount;
483 
484 	/* parent */
485 	if (cnp->cn_flags & ISDOTDOT) {
486 		if (pd->pn_type == pfstype_root)
487 			PFS_RETURN (EIO);
488 		error = vfs_busy(mp, MBF_NOWAIT);
489 		if (error != 0) {
490 			vfs_ref(mp);
491 			VOP_UNLOCK(vn, 0);
492 			error = vfs_busy(mp, 0);
493 			vn_lock(vn, LK_EXCLUSIVE | LK_RETRY);
494 			vfs_rel(mp);
495 			if (error != 0)
496 				PFS_RETURN(ENOENT);
497 			if (vn->v_iflag & VI_DOOMED) {
498 				vfs_unbusy(mp);
499 				PFS_RETURN(ENOENT);
500 			}
501 		}
502 		VOP_UNLOCK(vn, 0);
503 		KASSERT(pd->pn_parent != NULL,
504 		    ("%s(): non-root directory has no parent", __func__));
505 		/*
506 		 * This one is tricky.  Descendents of procdir nodes
507 		 * inherit their parent's process affinity, but
508 		 * there's no easy reverse mapping.  For simplicity,
509 		 * we assume that if this node is a procdir, its
510 		 * parent isn't (which is correct as long as
511 		 * descendents of procdir nodes are never procdir
512 		 * nodes themselves)
513 		 */
514 		if (pd->pn_type == pfstype_procdir)
515 			pid = NO_PID;
516 		pfs_lock(pd);
517 		pn = pd->pn_parent;
518 		pfs_unlock(pd);
519 		goto got_pnode;
520 	}
521 
522 	pfs_lock(pd);
523 
524 	/* named node */
525 	for (pn = pd->pn_nodes; pn != NULL; pn = pn->pn_next)
526 		if (pn->pn_type == pfstype_procdir)
527 			pdn = pn;
528 		else if (pn->pn_name[namelen] == '\0' &&
529 		    bcmp(pname, pn->pn_name, namelen) == 0) {
530 			pfs_unlock(pd);
531 			goto got_pnode;
532 		}
533 
534 	/* process dependent node */
535 	if ((pn = pdn) != NULL) {
536 		pid = 0;
537 		for (pid = 0, i = 0; i < namelen && isdigit(pname[i]); ++i)
538 			if ((pid = pid * 10 + pname[i] - '0') > PID_MAX)
539 				break;
540 		if (i == cnp->cn_namelen) {
541 			pfs_unlock(pd);
542 			goto got_pnode;
543 		}
544 	}
545 
546 	pfs_unlock(pd);
547 
548 	PFS_RETURN (ENOENT);
549 
550  got_pnode:
551 	pfs_assert_not_owned(pd);
552 	pfs_assert_not_owned(pn);
553 	visible = pfs_visible(curthread, pn, pid, false, NULL);
554 	if (!visible) {
555 		error = ENOENT;
556 		goto failed;
557 	}
558 
559 	error = pfs_vncache_alloc(mp, vpp, pn, pid);
560 	if (error)
561 		goto failed;
562 
563 	if (cnp->cn_flags & ISDOTDOT) {
564 		vfs_unbusy(mp);
565 		vn_lock(vn, LK_EXCLUSIVE | LK_RETRY);
566 		if (vn->v_iflag & VI_DOOMED) {
567 			vput(*vpp);
568 			*vpp = NULL;
569 			PFS_RETURN(ENOENT);
570 		}
571 	}
572 	if (cnp->cn_flags & MAKEENTRY && !(vn->v_iflag & VI_DOOMED))
573 		cache_enter(vn, *vpp, cnp);
574 	PFS_RETURN (0);
575  failed:
576 	if (cnp->cn_flags & ISDOTDOT) {
577 		vfs_unbusy(mp);
578 		vn_lock(vn, LK_EXCLUSIVE | LK_RETRY);
579 		*vpp = NULL;
580 	}
581 	PFS_RETURN(error);
582 }
583 
584 /*
585  * Open a file or directory.
586  */
587 static int
588 pfs_open(struct vop_open_args *va)
589 {
590 	struct vnode *vn = va->a_vp;
591 	struct pfs_vdata *pvd = vn->v_data;
592 	struct pfs_node *pn = pvd->pvd_pn;
593 	int mode = va->a_mode;
594 
595 	PFS_TRACE(("%s (mode 0x%x)", pn->pn_name, mode));
596 	pfs_assert_not_owned(pn);
597 
598 	/* check if the requested mode is permitted */
599 	if (((mode & FREAD) && !(mode & PFS_RD)) ||
600 	    ((mode & FWRITE) && !(mode & PFS_WR)))
601 		PFS_RETURN (EPERM);
602 
603 	/* we don't support locking */
604 	if ((mode & O_SHLOCK) || (mode & O_EXLOCK))
605 		PFS_RETURN (EOPNOTSUPP);
606 
607 	PFS_RETURN (0);
608 }
609 
610 /*
611  * Read from a file
612  */
613 static int
614 pfs_read(struct vop_read_args *va)
615 {
616 	struct vnode *vn = va->a_vp;
617 	struct pfs_vdata *pvd = vn->v_data;
618 	struct pfs_node *pn = pvd->pvd_pn;
619 	struct uio *uio = va->a_uio;
620 	struct proc *proc;
621 	struct sbuf *sb = NULL;
622 	int error, locked;
623 	off_t buflen;
624 
625 	PFS_TRACE(("%s", pn->pn_name));
626 	pfs_assert_not_owned(pn);
627 
628 	if (vn->v_type != VREG)
629 		PFS_RETURN (EINVAL);
630 	KASSERT_PN_IS_FILE(pn);
631 
632 	if (!(pn->pn_flags & PFS_RD))
633 		PFS_RETURN (EBADF);
634 
635 	if (pn->pn_fill == NULL)
636 		PFS_RETURN (EIO);
637 
638 	/*
639 	 * This is necessary because either process' privileges may
640 	 * have changed since the open() call.
641 	 */
642 	if (!pfs_visible(curthread, pn, pvd->pvd_pid, false, &proc))
643 		PFS_RETURN (EIO);
644 	if (proc != NULL) {
645 		_PHOLD(proc);
646 		PROC_UNLOCK(proc);
647 	}
648 
649 	vhold(vn);
650 	locked = VOP_ISLOCKED(vn);
651 	VOP_UNLOCK(vn, 0);
652 
653 	if (pn->pn_flags & PFS_RAWRD) {
654 		PFS_TRACE(("%zd resid", uio->uio_resid));
655 		error = pn_fill(curthread, proc, pn, NULL, uio);
656 		PFS_TRACE(("%zd resid", uio->uio_resid));
657 		goto ret;
658 	}
659 
660 	if (uio->uio_resid < 0 || uio->uio_offset < 0 ||
661 	    uio->uio_resid > OFF_MAX - uio->uio_offset) {
662 		error = EINVAL;
663 		goto ret;
664 	}
665 	buflen = uio->uio_offset + uio->uio_resid;
666 	if (buflen > MAXPHYS)
667 		buflen = MAXPHYS;
668 
669 	sb = sbuf_new(sb, NULL, buflen + 1, 0);
670 	if (sb == NULL) {
671 		error = EIO;
672 		goto ret;
673 	}
674 
675 	error = pn_fill(curthread, proc, pn, sb, uio);
676 
677 	if (error) {
678 		sbuf_delete(sb);
679 		goto ret;
680 	}
681 
682 	/*
683 	 * XXX: If the buffer overflowed, sbuf_len() will not return
684 	 * the data length. Then just use the full length because an
685 	 * overflowed sbuf must be full.
686 	 */
687 	if (sbuf_finish(sb) == 0)
688 		buflen = sbuf_len(sb);
689 	error = uiomove_frombuf(sbuf_data(sb), buflen, uio);
690 	sbuf_delete(sb);
691 ret:
692 	vn_lock(vn, locked | LK_RETRY);
693 	vdrop(vn);
694 	if (proc != NULL)
695 		PRELE(proc);
696 	PFS_RETURN (error);
697 }
698 
699 /*
700  * Iterate through directory entries
701  */
702 static int
703 pfs_iterate(struct thread *td, struct proc *proc, struct pfs_node *pd,
704 	    struct pfs_node **pn, struct proc **p)
705 {
706 	int visible;
707 
708 	sx_assert(&allproc_lock, SX_SLOCKED);
709 	pfs_assert_owned(pd);
710  again:
711 	if (*pn == NULL) {
712 		/* first node */
713 		*pn = pd->pn_nodes;
714 	} else if ((*pn)->pn_type != pfstype_procdir) {
715 		/* next node */
716 		*pn = (*pn)->pn_next;
717 	}
718 	if (*pn != NULL && (*pn)->pn_type == pfstype_procdir) {
719 		/* next process */
720 		if (*p == NULL)
721 			*p = LIST_FIRST(&allproc);
722 		else
723 			*p = LIST_NEXT(*p, p_list);
724 		/* out of processes: next node */
725 		if (*p == NULL)
726 			*pn = (*pn)->pn_next;
727 		else
728 			PROC_LOCK(*p);
729 	}
730 
731 	if ((*pn) == NULL)
732 		return (-1);
733 
734 	if (*p != NULL) {
735 		visible = pfs_visible_proc(td, *pn, *p);
736 		PROC_UNLOCK(*p);
737 	} else if (proc != NULL) {
738 		visible = pfs_visible_proc(td, *pn, proc);
739 	} else {
740 		visible = 1;
741 	}
742 	if (!visible)
743 		goto again;
744 
745 	return (0);
746 }
747 
748 /* Directory entry list */
749 struct pfsentry {
750 	STAILQ_ENTRY(pfsentry)	link;
751 	struct dirent		entry;
752 };
753 STAILQ_HEAD(pfsdirentlist, pfsentry);
754 
755 /*
756  * Return directory entries.
757  */
758 static int
759 pfs_readdir(struct vop_readdir_args *va)
760 {
761 	struct vnode *vn = va->a_vp;
762 	struct pfs_vdata *pvd = vn->v_data;
763 	struct pfs_node *pd = pvd->pvd_pn;
764 	pid_t pid = pvd->pvd_pid;
765 	struct proc *p, *proc;
766 	struct pfs_node *pn;
767 	struct uio *uio;
768 	struct pfsentry *pfsent, *pfsent2;
769 	struct pfsdirentlist lst;
770 	off_t offset;
771 	int error, i, resid;
772 
773 	STAILQ_INIT(&lst);
774 	error = 0;
775 	KASSERT(pd->pn_info == vn->v_mount->mnt_data,
776 	    ("%s(): pn_info does not match mountpoint", __func__));
777 	PFS_TRACE(("%s pid %lu", pd->pn_name, (unsigned long)pid));
778 	pfs_assert_not_owned(pd);
779 
780 	if (vn->v_type != VDIR)
781 		PFS_RETURN (ENOTDIR);
782 	KASSERT_PN_IS_DIR(pd);
783 	uio = va->a_uio;
784 
785 	/* only allow reading entire entries */
786 	offset = uio->uio_offset;
787 	resid = uio->uio_resid;
788 	if (offset < 0 || offset % PFS_DELEN != 0 ||
789 	    (resid && resid < PFS_DELEN))
790 		PFS_RETURN (EINVAL);
791 	if (resid == 0)
792 		PFS_RETURN (0);
793 
794 	sx_slock(&allproc_lock);
795 	pfs_lock(pd);
796 
797         /* check if the directory is visible to the caller */
798         if (!pfs_visible(curthread, pd, pid, true, &proc)) {
799 		sx_sunlock(&allproc_lock);
800 		pfs_unlock(pd);
801                 PFS_RETURN (ENOENT);
802 	}
803 	KASSERT(pid == NO_PID || proc != NULL,
804 	    ("%s(): no process for pid %lu", __func__, (unsigned long)pid));
805 
806 	/* skip unwanted entries */
807 	for (pn = NULL, p = NULL; offset > 0; offset -= PFS_DELEN) {
808 		if (pfs_iterate(curthread, proc, pd, &pn, &p) == -1) {
809 			/* nothing left... */
810 			if (proc != NULL)
811 				PROC_UNLOCK(proc);
812 			pfs_unlock(pd);
813 			sx_sunlock(&allproc_lock);
814 			PFS_RETURN (0);
815 		}
816 	}
817 
818 	/* fill in entries */
819 	while (pfs_iterate(curthread, proc, pd, &pn, &p) != -1 &&
820 	    resid >= PFS_DELEN) {
821 		if ((pfsent = malloc(sizeof(struct pfsentry), M_IOV,
822 		    M_NOWAIT | M_ZERO)) == NULL) {
823 			error = ENOMEM;
824 			break;
825 		}
826 		pfsent->entry.d_reclen = PFS_DELEN;
827 		pfsent->entry.d_fileno = pn_fileno(pn, pid);
828 		/* PFS_DELEN was picked to fit PFS_NAMLEN */
829 		for (i = 0; i < PFS_NAMELEN - 1 && pn->pn_name[i] != '\0'; ++i)
830 			pfsent->entry.d_name[i] = pn->pn_name[i];
831 		pfsent->entry.d_name[i] = 0;
832 		pfsent->entry.d_namlen = i;
833 		switch (pn->pn_type) {
834 		case pfstype_procdir:
835 			KASSERT(p != NULL,
836 			    ("reached procdir node with p == NULL"));
837 			pfsent->entry.d_namlen = snprintf(pfsent->entry.d_name,
838 			    PFS_NAMELEN, "%d", p->p_pid);
839 			/* fall through */
840 		case pfstype_root:
841 		case pfstype_dir:
842 		case pfstype_this:
843 		case pfstype_parent:
844 			pfsent->entry.d_type = DT_DIR;
845 			break;
846 		case pfstype_file:
847 			pfsent->entry.d_type = DT_REG;
848 			break;
849 		case pfstype_symlink:
850 			pfsent->entry.d_type = DT_LNK;
851 			break;
852 		default:
853 			panic("%s has unexpected node type: %d", pn->pn_name, pn->pn_type);
854 		}
855 		PFS_TRACE(("%s", pfsent->entry.d_name));
856 		STAILQ_INSERT_TAIL(&lst, pfsent, link);
857 		offset += PFS_DELEN;
858 		resid -= PFS_DELEN;
859 	}
860 	if (proc != NULL)
861 		PROC_UNLOCK(proc);
862 	pfs_unlock(pd);
863 	sx_sunlock(&allproc_lock);
864 	i = 0;
865 	STAILQ_FOREACH_SAFE(pfsent, &lst, link, pfsent2) {
866 		if (error == 0)
867 			error = uiomove(&pfsent->entry, PFS_DELEN, uio);
868 		free(pfsent, M_IOV);
869 		i++;
870 	}
871 	PFS_TRACE(("%ju bytes", (uintmax_t)(i * PFS_DELEN)));
872 	PFS_RETURN (error);
873 }
874 
875 /*
876  * Read a symbolic link
877  */
878 static int
879 pfs_readlink(struct vop_readlink_args *va)
880 {
881 	struct vnode *vn = va->a_vp;
882 	struct pfs_vdata *pvd = vn->v_data;
883 	struct pfs_node *pn = pvd->pvd_pn;
884 	struct uio *uio = va->a_uio;
885 	struct proc *proc = NULL;
886 	char buf[PATH_MAX];
887 	struct sbuf sb;
888 	int error, locked;
889 
890 	PFS_TRACE(("%s", pn->pn_name));
891 	pfs_assert_not_owned(pn);
892 
893 	if (vn->v_type != VLNK)
894 		PFS_RETURN (EINVAL);
895 	KASSERT_PN_IS_LINK(pn);
896 
897 	if (pn->pn_fill == NULL)
898 		PFS_RETURN (EIO);
899 
900 	if (pvd->pvd_pid != NO_PID) {
901 		if ((proc = pfind(pvd->pvd_pid)) == NULL)
902 			PFS_RETURN (EIO);
903 		if (proc->p_flag & P_WEXIT) {
904 			PROC_UNLOCK(proc);
905 			PFS_RETURN (EIO);
906 		}
907 		_PHOLD(proc);
908 		PROC_UNLOCK(proc);
909 	}
910 	vhold(vn);
911 	locked = VOP_ISLOCKED(vn);
912 	VOP_UNLOCK(vn, 0);
913 
914 	/* sbuf_new() can't fail with a static buffer */
915 	sbuf_new(&sb, buf, sizeof buf, 0);
916 
917 	error = pn_fill(curthread, proc, pn, &sb, NULL);
918 
919 	if (proc != NULL)
920 		PRELE(proc);
921 	vn_lock(vn, locked | LK_RETRY);
922 	vdrop(vn);
923 
924 	if (error) {
925 		sbuf_delete(&sb);
926 		PFS_RETURN (error);
927 	}
928 
929 	if (sbuf_finish(&sb) != 0) {
930 		sbuf_delete(&sb);
931 		PFS_RETURN (ENAMETOOLONG);
932 	}
933 
934 	error = uiomove_frombuf(sbuf_data(&sb), sbuf_len(&sb), uio);
935 	sbuf_delete(&sb);
936 	PFS_RETURN (error);
937 }
938 
939 /*
940  * Reclaim a vnode
941  */
942 static int
943 pfs_reclaim(struct vop_reclaim_args *va)
944 {
945 	struct vnode *vn = va->a_vp;
946 	struct pfs_vdata *pvd = vn->v_data;
947 	struct pfs_node *pn = pvd->pvd_pn;
948 
949 	PFS_TRACE(("%s", pn->pn_name));
950 	pfs_assert_not_owned(pn);
951 
952 	return (pfs_vncache_free(va->a_vp));
953 }
954 
955 /*
956  * Set attributes
957  */
958 static int
959 pfs_setattr(struct vop_setattr_args *va)
960 {
961 	struct vnode *vn = va->a_vp;
962 	struct pfs_vdata *pvd = vn->v_data;
963 	struct pfs_node *pn = pvd->pvd_pn;
964 
965 	PFS_TRACE(("%s", pn->pn_name));
966 	pfs_assert_not_owned(pn);
967 
968 	PFS_RETURN (EOPNOTSUPP);
969 }
970 
971 /*
972  * Write to a file
973  */
974 static int
975 pfs_write(struct vop_write_args *va)
976 {
977 	struct vnode *vn = va->a_vp;
978 	struct pfs_vdata *pvd = vn->v_data;
979 	struct pfs_node *pn = pvd->pvd_pn;
980 	struct uio *uio = va->a_uio;
981 	struct proc *proc;
982 	struct sbuf sb;
983 	int error;
984 
985 	PFS_TRACE(("%s", pn->pn_name));
986 	pfs_assert_not_owned(pn);
987 
988 	if (vn->v_type != VREG)
989 		PFS_RETURN (EINVAL);
990 	KASSERT_PN_IS_FILE(pn);
991 
992 	if (!(pn->pn_flags & PFS_WR))
993 		PFS_RETURN (EBADF);
994 
995 	if (pn->pn_fill == NULL)
996 		PFS_RETURN (EIO);
997 
998 	/*
999 	 * This is necessary because either process' privileges may
1000 	 * have changed since the open() call.
1001 	 */
1002 	if (!pfs_visible(curthread, pn, pvd->pvd_pid, false, &proc))
1003 		PFS_RETURN (EIO);
1004 	if (proc != NULL) {
1005 		_PHOLD(proc);
1006 		PROC_UNLOCK(proc);
1007 	}
1008 
1009 	if (pn->pn_flags & PFS_RAWWR) {
1010 		error = pn_fill(curthread, proc, pn, NULL, uio);
1011 		if (proc != NULL)
1012 			PRELE(proc);
1013 		PFS_RETURN (error);
1014 	}
1015 
1016 	sbuf_uionew(&sb, uio, &error);
1017 	if (error) {
1018 		if (proc != NULL)
1019 			PRELE(proc);
1020 		PFS_RETURN (error);
1021 	}
1022 
1023 	error = pn_fill(curthread, proc, pn, &sb, uio);
1024 
1025 	sbuf_delete(&sb);
1026 	if (proc != NULL)
1027 		PRELE(proc);
1028 	PFS_RETURN (error);
1029 }
1030 
1031 /*
1032  * Vnode operations
1033  */
1034 struct vop_vector pfs_vnodeops = {
1035 	.vop_default =		&default_vnodeops,
1036 
1037 	.vop_access =		pfs_access,
1038 	.vop_cachedlookup =	pfs_lookup,
1039 	.vop_close =		pfs_close,
1040 	.vop_create =		VOP_EOPNOTSUPP,
1041 	.vop_getattr =		pfs_getattr,
1042 	.vop_getextattr =	pfs_getextattr,
1043 	.vop_ioctl =		pfs_ioctl,
1044 	.vop_link =		VOP_EOPNOTSUPP,
1045 	.vop_lookup =		vfs_cache_lookup,
1046 	.vop_mkdir =		VOP_EOPNOTSUPP,
1047 	.vop_mknod =		VOP_EOPNOTSUPP,
1048 	.vop_open =		pfs_open,
1049 	.vop_read =		pfs_read,
1050 	.vop_readdir =		pfs_readdir,
1051 	.vop_readlink =		pfs_readlink,
1052 	.vop_reclaim =		pfs_reclaim,
1053 	.vop_remove =		VOP_EOPNOTSUPP,
1054 	.vop_rename =		VOP_EOPNOTSUPP,
1055 	.vop_rmdir =		VOP_EOPNOTSUPP,
1056 	.vop_setattr =		pfs_setattr,
1057 	.vop_symlink =		VOP_EOPNOTSUPP,
1058 	.vop_vptocnp =		pfs_vptocnp,
1059 	.vop_write =		pfs_write,
1060 	/* XXX I've probably forgotten a few that need VOP_EOPNOTSUPP */
1061 };
1062