xref: /netbsd-src/sys/miscfs/procfs/procfs_vnops.c (revision 0c4ddb1599a0bea866fde8522a74cfbd2f68cd1b)
1 /*	$NetBSD: procfs_vnops.c,v 1.170 2008/07/02 19:49:58 rmind Exp $	*/
2 
3 /*-
4  * Copyright (c) 2006, 2007, 2008 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Andrew Doran.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29  * POSSIBILITY OF SUCH DAMAGE.
30  */
31 
32 /*
33  * Copyright (c) 1993, 1995
34  *	The Regents of the University of California.  All rights reserved.
35  *
36  * This code is derived from software contributed to Berkeley by
37  * Jan-Simon Pendry.
38  *
39  * Redistribution and use in source and binary forms, with or without
40  * modification, are permitted provided that the following conditions
41  * are met:
42  * 1. Redistributions of source code must retain the above copyright
43  *    notice, this list of conditions and the following disclaimer.
44  * 2. Redistributions in binary form must reproduce the above copyright
45  *    notice, this list of conditions and the following disclaimer in the
46  *    documentation and/or other materials provided with the distribution.
47  * 3. Neither the name of the University nor the names of its contributors
48  *    may be used to endorse or promote products derived from this software
49  *    without specific prior written permission.
50  *
51  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
52  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
53  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
54  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
55  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
56  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
57  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
58  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
59  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
60  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
61  * SUCH DAMAGE.
62  *
63  *	@(#)procfs_vnops.c	8.18 (Berkeley) 5/21/95
64  */
65 
66 /*
67  * Copyright (c) 1993 Jan-Simon Pendry
68  *
69  * This code is derived from software contributed to Berkeley by
70  * Jan-Simon Pendry.
71  *
72  * Redistribution and use in source and binary forms, with or without
73  * modification, are permitted provided that the following conditions
74  * are met:
75  * 1. Redistributions of source code must retain the above copyright
76  *    notice, this list of conditions and the following disclaimer.
77  * 2. Redistributions in binary form must reproduce the above copyright
78  *    notice, this list of conditions and the following disclaimer in the
79  *    documentation and/or other materials provided with the distribution.
80  * 3. All advertising materials mentioning features or use of this software
81  *    must display the following acknowledgement:
82  *	This product includes software developed by the University of
83  *	California, Berkeley and its contributors.
84  * 4. Neither the name of the University nor the names of its contributors
85  *    may be used to endorse or promote products derived from this software
86  *    without specific prior written permission.
87  *
88  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
89  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
90  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
91  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
92  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
93  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
94  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
95  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
96  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
97  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
98  * SUCH DAMAGE.
99  *
100  *	@(#)procfs_vnops.c	8.18 (Berkeley) 5/21/95
101  */
102 
103 /*
104  * procfs vnode interface
105  */
106 
107 #include <sys/cdefs.h>
108 __KERNEL_RCSID(0, "$NetBSD: procfs_vnops.c,v 1.170 2008/07/02 19:49:58 rmind Exp $");
109 
110 #include <sys/param.h>
111 #include <sys/systm.h>
112 #include <sys/time.h>
113 #include <sys/kernel.h>
114 #include <sys/file.h>
115 #include <sys/filedesc.h>
116 #include <sys/proc.h>
117 #include <sys/vnode.h>
118 #include <sys/namei.h>
119 #include <sys/malloc.h>
120 #include <sys/mount.h>
121 #include <sys/dirent.h>
122 #include <sys/resourcevar.h>
123 #include <sys/stat.h>
124 #include <sys/ptrace.h>
125 #include <sys/kauth.h>
126 
127 #include <uvm/uvm_extern.h>	/* for PAGE_SIZE */
128 
129 #include <machine/reg.h>
130 
131 #include <miscfs/genfs/genfs.h>
132 #include <miscfs/procfs/procfs.h>
133 
134 /*
135  * Vnode Operations.
136  *
137  */
138 
139 static int procfs_validfile_linux(struct lwp *, struct mount *);
140 static int procfs_root_readdir_callback(struct proc *, void *);
141 static void procfs_dir(pfstype, struct lwp *, struct proc *, char **, char *,
142     size_t);
143 
144 /*
145  * This is a list of the valid names in the
146  * process-specific sub-directories.  It is
147  * used in procfs_lookup and procfs_readdir
148  */
149 static const struct proc_target {
150 	u_char	pt_type;
151 	u_char	pt_namlen;
152 	const char	*pt_name;
153 	pfstype	pt_pfstype;
154 	int	(*pt_valid)(struct lwp *, struct mount *);
155 } proc_targets[] = {
156 #define N(s) sizeof(s)-1, s
157 	/*	  name		type		validp */
158 	{ DT_DIR, N("."),	PFSproc,	NULL },
159 	{ DT_DIR, N(".."),	PFSroot,	NULL },
160 	{ DT_DIR, N("fd"),	PFSfd,		NULL },
161 	{ DT_REG, N("file"),	PFSfile,	procfs_validfile },
162 	{ DT_REG, N("mem"),	PFSmem,		NULL },
163 	{ DT_REG, N("regs"),	PFSregs,	procfs_validregs },
164 	{ DT_REG, N("fpregs"),	PFSfpregs,	procfs_validfpregs },
165 	{ DT_REG, N("ctl"),	PFSctl,		NULL },
166 	{ DT_REG, N("stat"),	PFSstat,	procfs_validfile_linux },
167 	{ DT_REG, N("status"),	PFSstatus,	NULL },
168 	{ DT_REG, N("note"),	PFSnote,	NULL },
169 	{ DT_REG, N("notepg"),	PFSnotepg,	NULL },
170 	{ DT_REG, N("map"),	PFSmap,		procfs_validmap },
171 	{ DT_REG, N("maps"),	PFSmaps,	procfs_validmap },
172 	{ DT_REG, N("cmdline"), PFScmdline,	NULL },
173 	{ DT_REG, N("exe"),	PFSexe,		procfs_validfile },
174 	{ DT_LNK, N("cwd"),	PFScwd,		NULL },
175 	{ DT_LNK, N("root"),	PFSchroot,	NULL },
176 	{ DT_LNK, N("emul"),	PFSemul,	NULL },
177 	{ DT_REG, N("statm"),	PFSstatm,	procfs_validfile_linux },
178 #ifdef __HAVE_PROCFS_MACHDEP
179 	PROCFS_MACHDEP_NODETYPE_DEFNS
180 #endif
181 #undef N
182 };
183 static const int nproc_targets = sizeof(proc_targets) / sizeof(proc_targets[0]);
184 
185 /*
186  * List of files in the root directory. Note: the validate function will
187  * be called with p == NULL for these ones.
188  */
189 static const struct proc_target proc_root_targets[] = {
190 #define N(s) sizeof(s)-1, s
191 	/*	  name		    type	    validp */
192 	{ DT_REG, N("meminfo"),     PFSmeminfo,        procfs_validfile_linux },
193 	{ DT_REG, N("cpuinfo"),     PFScpuinfo,        procfs_validfile_linux },
194 	{ DT_REG, N("uptime"),      PFSuptime,         procfs_validfile_linux },
195 	{ DT_REG, N("mounts"),	    PFSmounts,	       procfs_validfile_linux },
196 	{ DT_REG, N("devices"),     PFSdevices,        procfs_validfile_linux },
197 	{ DT_REG, N("stat"),	    PFScpustat,        procfs_validfile_linux },
198 	{ DT_REG, N("loadavg"),	    PFSloadavg,        procfs_validfile_linux },
199 #undef N
200 };
201 static const int nproc_root_targets =
202     sizeof(proc_root_targets) / sizeof(proc_root_targets[0]);
203 
204 int	procfs_lookup(void *);
205 #define	procfs_create	genfs_eopnotsupp
206 #define	procfs_mknod	genfs_eopnotsupp
207 int	procfs_open(void *);
208 int	procfs_close(void *);
209 int	procfs_access(void *);
210 int	procfs_getattr(void *);
211 int	procfs_setattr(void *);
212 #define	procfs_read	procfs_rw
213 #define	procfs_write	procfs_rw
214 #define	procfs_fcntl	genfs_fcntl
215 #define	procfs_ioctl	genfs_enoioctl
216 #define	procfs_poll	genfs_poll
217 #define procfs_revoke	genfs_revoke
218 #define	procfs_fsync	genfs_nullop
219 #define	procfs_seek	genfs_nullop
220 #define	procfs_remove	genfs_eopnotsupp
221 int	procfs_link(void *);
222 #define	procfs_rename	genfs_eopnotsupp
223 #define	procfs_mkdir	genfs_eopnotsupp
224 #define	procfs_rmdir	genfs_eopnotsupp
225 int	procfs_symlink(void *);
226 int	procfs_readdir(void *);
227 int	procfs_readlink(void *);
228 #define	procfs_abortop	genfs_abortop
229 int	procfs_inactive(void *);
230 int	procfs_reclaim(void *);
231 #define	procfs_lock	genfs_lock
232 #define	procfs_unlock	genfs_unlock
233 #define	procfs_bmap	genfs_badop
234 #define	procfs_strategy	genfs_badop
235 int	procfs_print(void *);
236 int	procfs_pathconf(void *);
237 #define	procfs_islocked	genfs_islocked
238 #define	procfs_advlock	genfs_einval
239 #define	procfs_bwrite	genfs_eopnotsupp
240 #define procfs_putpages	genfs_null_putpages
241 
242 static int atoi(const char *, size_t);
243 
244 /*
245  * procfs vnode operations.
246  */
247 int (**procfs_vnodeop_p)(void *);
248 const struct vnodeopv_entry_desc procfs_vnodeop_entries[] = {
249 	{ &vop_default_desc, vn_default_error },
250 	{ &vop_lookup_desc, procfs_lookup },		/* lookup */
251 	{ &vop_create_desc, procfs_create },		/* create */
252 	{ &vop_mknod_desc, procfs_mknod },		/* mknod */
253 	{ &vop_open_desc, procfs_open },		/* open */
254 	{ &vop_close_desc, procfs_close },		/* close */
255 	{ &vop_access_desc, procfs_access },		/* access */
256 	{ &vop_getattr_desc, procfs_getattr },		/* getattr */
257 	{ &vop_setattr_desc, procfs_setattr },		/* setattr */
258 	{ &vop_read_desc, procfs_read },		/* read */
259 	{ &vop_write_desc, procfs_write },		/* write */
260 	{ &vop_fcntl_desc, procfs_fcntl },		/* fcntl */
261 	{ &vop_ioctl_desc, procfs_ioctl },		/* ioctl */
262 	{ &vop_poll_desc, procfs_poll },		/* poll */
263 	{ &vop_revoke_desc, procfs_revoke },		/* revoke */
264 	{ &vop_fsync_desc, procfs_fsync },		/* fsync */
265 	{ &vop_seek_desc, procfs_seek },		/* seek */
266 	{ &vop_remove_desc, procfs_remove },		/* remove */
267 	{ &vop_link_desc, procfs_link },		/* link */
268 	{ &vop_rename_desc, procfs_rename },		/* rename */
269 	{ &vop_mkdir_desc, procfs_mkdir },		/* mkdir */
270 	{ &vop_rmdir_desc, procfs_rmdir },		/* rmdir */
271 	{ &vop_symlink_desc, procfs_symlink },		/* symlink */
272 	{ &vop_readdir_desc, procfs_readdir },		/* readdir */
273 	{ &vop_readlink_desc, procfs_readlink },	/* readlink */
274 	{ &vop_abortop_desc, procfs_abortop },		/* abortop */
275 	{ &vop_inactive_desc, procfs_inactive },	/* inactive */
276 	{ &vop_reclaim_desc, procfs_reclaim },		/* reclaim */
277 	{ &vop_lock_desc, procfs_lock },		/* lock */
278 	{ &vop_unlock_desc, procfs_unlock },		/* unlock */
279 	{ &vop_bmap_desc, procfs_bmap },		/* bmap */
280 	{ &vop_strategy_desc, procfs_strategy },	/* strategy */
281 	{ &vop_print_desc, procfs_print },		/* print */
282 	{ &vop_islocked_desc, procfs_islocked },	/* islocked */
283 	{ &vop_pathconf_desc, procfs_pathconf },	/* pathconf */
284 	{ &vop_advlock_desc, procfs_advlock },		/* advlock */
285 	{ &vop_putpages_desc, procfs_putpages },	/* putpages */
286 	{ NULL, NULL }
287 };
288 const struct vnodeopv_desc procfs_vnodeop_opv_desc =
289 	{ &procfs_vnodeop_p, procfs_vnodeop_entries };
290 /*
291  * set things up for doing i/o on
292  * the pfsnode (vp).  (vp) is locked
293  * on entry, and should be left locked
294  * on exit.
295  *
296  * for procfs we don't need to do anything
297  * in particular for i/o.  all that is done
298  * is to support exclusive open on process
299  * memory images.
300  */
301 int
302 procfs_open(v)
303 	void *v;
304 {
305 	struct vop_open_args /* {
306 		struct vnode *a_vp;
307 		int  a_mode;
308 		kauth_cred_t a_cred;
309 	} */ *ap = v;
310 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
311 	struct lwp *l1;
312 	struct proc *p2;
313 	int error;
314 
315 	if ((error = procfs_proc_lock(pfs->pfs_pid, &p2, ENOENT)) != 0)
316 		return error;
317 
318 	l1 = curlwp;				/* tracer */
319 
320 #define	M2K(m)	(((m) & FREAD) && ((m) & FWRITE) ? \
321 		 KAUTH_REQ_PROCESS_PROCFS_RW : \
322 		 (m) & FWRITE ? KAUTH_REQ_PROCESS_PROCFS_WRITE : \
323 		 KAUTH_REQ_PROCESS_PROCFS_READ)
324 
325 	mutex_enter(p2->p_lock);
326 	error = kauth_authorize_process(l1->l_cred, KAUTH_PROCESS_PROCFS,
327 	    p2, pfs, KAUTH_ARG(M2K(ap->a_mode)), NULL);
328 	mutex_exit(p2->p_lock);
329 	if (error) {
330 		procfs_proc_unlock(p2);
331 		return (error);
332 	}
333 
334 #undef M2K
335 
336 	switch (pfs->pfs_type) {
337 	case PFSmem:
338 		if (((pfs->pfs_flags & FWRITE) && (ap->a_mode & O_EXCL)) ||
339 		    ((pfs->pfs_flags & O_EXCL) && (ap->a_mode & FWRITE))) {
340 			error = EBUSY;
341 			break;
342 		}
343 
344 		if (!proc_isunder(p2, l1)) {
345 			error = EPERM;
346 			break;
347 		}
348 
349 		if (ap->a_mode & FWRITE)
350 			pfs->pfs_flags = ap->a_mode & (FWRITE|O_EXCL);
351 
352 		break;
353 
354 	case PFSregs:
355 	case PFSfpregs:
356 		if (!proc_isunder(p2, l1)) {
357 			error = EPERM;
358 			break;
359 		}
360 		break;
361 
362 	default:
363 		break;
364 	}
365 
366 	procfs_proc_unlock(p2);
367 	return (error);
368 }
369 
370 /*
371  * close the pfsnode (vp) after doing i/o.
372  * (vp) is not locked on entry or exit.
373  *
374  * nothing to do for procfs other than undo
375  * any exclusive open flag (see _open above).
376  */
377 int
378 procfs_close(v)
379 	void *v;
380 {
381 	struct vop_close_args /* {
382 		struct vnode *a_vp;
383 		int  a_fflag;
384 		kauth_cred_t a_cred;
385 	} */ *ap = v;
386 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
387 
388 	switch (pfs->pfs_type) {
389 	case PFSmem:
390 		if ((ap->a_fflag & FWRITE) && (pfs->pfs_flags & O_EXCL))
391 			pfs->pfs_flags &= ~(FWRITE|O_EXCL);
392 		break;
393 
394 	default:
395 		break;
396 	}
397 
398 	return (0);
399 }
400 
401 /*
402  * _inactive is called when the pfsnode
403  * is vrele'd and the reference count goes
404  * to zero.  (vp) will be on the vnode free
405  * list, so to get it back vget() must be
406  * used.
407  *
408  * (vp) is locked on entry, but must be unlocked on exit.
409  */
410 int
411 procfs_inactive(v)
412 	void *v;
413 {
414 	struct vop_inactive_args /* {
415 		struct vnode *a_vp;
416 		bool *a_recycle;
417 	} */ *ap = v;
418 	struct vnode *vp = ap->a_vp;
419 	struct pfsnode *pfs = VTOPFS(vp);
420 
421 	mutex_enter(proc_lock);
422 	*ap->a_recycle = (p_find(pfs->pfs_pid, PFIND_LOCKED) == NULL);
423 	mutex_exit(proc_lock);
424 
425 	VOP_UNLOCK(vp, 0);
426 
427 	return (0);
428 }
429 
430 /*
431  * _reclaim is called when getnewvnode()
432  * wants to make use of an entry on the vnode
433  * free list.  at this time the filesystem needs
434  * to free any private data and remove the node
435  * from any private lists.
436  */
437 int
438 procfs_reclaim(v)
439 	void *v;
440 {
441 	struct vop_reclaim_args /* {
442 		struct vnode *a_vp;
443 	} */ *ap = v;
444 
445 	return (procfs_freevp(ap->a_vp));
446 }
447 
448 /*
449  * Return POSIX pathconf information applicable to special devices.
450  */
451 int
452 procfs_pathconf(v)
453 	void *v;
454 {
455 	struct vop_pathconf_args /* {
456 		struct vnode *a_vp;
457 		int a_name;
458 		register_t *a_retval;
459 	} */ *ap = v;
460 
461 	switch (ap->a_name) {
462 	case _PC_LINK_MAX:
463 		*ap->a_retval = LINK_MAX;
464 		return (0);
465 	case _PC_MAX_CANON:
466 		*ap->a_retval = MAX_CANON;
467 		return (0);
468 	case _PC_MAX_INPUT:
469 		*ap->a_retval = MAX_INPUT;
470 		return (0);
471 	case _PC_PIPE_BUF:
472 		*ap->a_retval = PIPE_BUF;
473 		return (0);
474 	case _PC_CHOWN_RESTRICTED:
475 		*ap->a_retval = 1;
476 		return (0);
477 	case _PC_VDISABLE:
478 		*ap->a_retval = _POSIX_VDISABLE;
479 		return (0);
480 	case _PC_SYNC_IO:
481 		*ap->a_retval = 1;
482 		return (0);
483 	default:
484 		return (EINVAL);
485 	}
486 	/* NOTREACHED */
487 }
488 
489 /*
490  * _print is used for debugging.
491  * just print a readable description
492  * of (vp).
493  */
494 int
495 procfs_print(v)
496 	void *v;
497 {
498 	struct vop_print_args /* {
499 		struct vnode *a_vp;
500 	} */ *ap = v;
501 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
502 
503 	printf("tag VT_PROCFS, type %d, pid %d, mode %x, flags %lx\n",
504 	    pfs->pfs_type, pfs->pfs_pid, pfs->pfs_mode, pfs->pfs_flags);
505 	return 0;
506 }
507 
508 int
509 procfs_link(v)
510 	void *v;
511 {
512 	struct vop_link_args /* {
513 		struct vnode *a_dvp;
514 		struct vnode *a_vp;
515 		struct componentname *a_cnp;
516 	} */ *ap = v;
517 
518 	VOP_ABORTOP(ap->a_dvp, ap->a_cnp);
519 	vput(ap->a_dvp);
520 	return (EROFS);
521 }
522 
523 int
524 procfs_symlink(v)
525 	void *v;
526 {
527 	struct vop_symlink_args /* {
528 		struct vnode *a_dvp;
529 		struct vnode **a_vpp;
530 		struct componentname *a_cnp;
531 		struct vattr *a_vap;
532 		char *a_target;
533 	} */ *ap = v;
534 
535 	VOP_ABORTOP(ap->a_dvp, ap->a_cnp);
536 	vput(ap->a_dvp);
537 	return (EROFS);
538 }
539 
540 /*
541  * Works out the path to (and vnode of) the target process's current
542  * working directory or chroot.  If the caller is in a chroot and
543  * can't "reach" the target's cwd or root (or some other error
544  * occurs), a "/" is returned for the path and a NULL pointer is
545  * returned for the vnode.
546  */
547 static void
548 procfs_dir(pfstype t, struct lwp *caller, struct proc *target, char **bpp,
549     char *path, size_t len)
550 {
551 	struct cwdinfo *cwdi;
552 	struct vnode *vp, *rvp;
553 	char *bp;
554 
555 	cwdi = caller->l_proc->p_cwdi;
556 	rw_enter(&cwdi->cwdi_lock, RW_READER);
557 
558 	rvp = cwdi->cwdi_rdir;
559 	bp = bpp ? *bpp : NULL;
560 
561 	switch (t) {
562 	case PFScwd:
563 		vp = target->p_cwdi->cwdi_cdir;
564 		break;
565 	case PFSchroot:
566 		vp = target->p_cwdi->cwdi_rdir;
567 		break;
568 	case PFSexe:
569 		vp = target->p_textvp;
570 		break;
571 	default:
572 		rw_exit(&cwdi->cwdi_lock);
573 		return;
574 	}
575 
576 	/*
577 	 * XXX: this horrible kludge avoids locking panics when
578 	 * attempting to lookup links that point to within procfs
579 	 */
580 	if (vp != NULL && vp->v_tag == VT_PROCFS) {
581 		if (bpp) {
582 			*--bp = '/';
583 			*bpp = bp;
584 		}
585 		rw_exit(&cwdi->cwdi_lock);
586 		return;
587 	}
588 
589 	if (rvp == NULL)
590 		rvp = rootvnode;
591 	if (vp == NULL || getcwd_common(vp, rvp, bp ? &bp : NULL, path,
592 	    len / 2, 0, caller) != 0) {
593 		vp = NULL;
594 		if (bpp) {
595 /*
596 			if (t == PFSexe) {
597 				snprintf(path, len, "%s/%d/file"
598 				    mp->mnt_stat.f_mntonname, pfs->pfs_pid);
599 			} else */ {
600 				bp = *bpp;
601 				*--bp = '/';
602 			}
603 		}
604 	}
605 
606 	if (bpp)
607 		*bpp = bp;
608 
609 	rw_exit(&cwdi->cwdi_lock);
610 }
611 
612 /*
613  * Invent attributes for pfsnode (vp) and store
614  * them in (vap).
615  * Directories lengths are returned as zero since
616  * any real length would require the genuine size
617  * to be computed, and nothing cares anyway.
618  *
619  * this is relatively minimal for procfs.
620  */
621 int
622 procfs_getattr(v)
623 	void *v;
624 {
625 	struct vop_getattr_args /* {
626 		struct vnode *a_vp;
627 		struct vattr *a_vap;
628 		kauth_cred_t a_cred;
629 	} */ *ap = v;
630 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
631 	struct vattr *vap = ap->a_vap;
632 	struct proc *procp;
633 	char *path;
634 	int error;
635 
636 	/* first check the process still exists */
637 	switch (pfs->pfs_type) {
638 	case PFSroot:
639 	case PFScurproc:
640 	case PFSself:
641 		procp = NULL;
642 		break;
643 
644 	default:
645 		error = procfs_proc_lock(pfs->pfs_pid, &procp, ENOENT);
646 		if (error != 0)
647 			return (error);
648 		break;
649 	}
650 
651 	switch (pfs->pfs_type) {
652 	case PFScwd:
653 	case PFSchroot:
654 	case PFSexe:
655 		MALLOC(path, char *, MAXPATHLEN + 4, M_TEMP,
656 		    M_WAITOK|M_CANFAIL);
657 		if (path == NULL && procp != NULL) {
658 			procfs_proc_unlock(procp);
659 			return (ENOMEM);
660 		}
661 		break;
662 
663 	default:
664 		path = NULL;
665 		break;
666 	}
667 
668 	if (procp != NULL) {
669 		mutex_enter(procp->p_lock);
670 		error = kauth_authorize_process(kauth_cred_get(),
671 		    KAUTH_PROCESS_CANSEE, procp,
672 		    KAUTH_ARG(KAUTH_REQ_PROCESS_CANSEE_ENTRY), NULL, NULL);
673 		mutex_exit(procp->p_lock);
674 		if (error != 0) {
675 		    	procfs_proc_unlock(procp);
676 		    	if (path != NULL)
677 		    		free(path, M_TEMP);
678 			return (ENOENT);
679 		}
680 	}
681 
682 	error = 0;
683 
684 	/* start by zeroing out the attributes */
685 	VATTR_NULL(vap);
686 
687 	/* next do all the common fields */
688 	vap->va_type = ap->a_vp->v_type;
689 	vap->va_mode = pfs->pfs_mode;
690 	vap->va_fileid = pfs->pfs_fileno;
691 	vap->va_flags = 0;
692 	vap->va_blocksize = PAGE_SIZE;
693 
694 	/*
695 	 * Make all times be current TOD.
696 	 *
697 	 * It would be possible to get the process start
698 	 * time from the p_stats structure, but there's
699 	 * no "file creation" time stamp anyway, and the
700 	 * p_stats structure is not addressable if u. gets
701 	 * swapped out for that process.
702 	 */
703 	getnanotime(&vap->va_ctime);
704 	vap->va_atime = vap->va_mtime = vap->va_ctime;
705 	if (procp)
706 		TIMEVAL_TO_TIMESPEC(&procp->p_stats->p_start,
707 		    &vap->va_birthtime);
708 	else
709 		getnanotime(&vap->va_birthtime);
710 
711 	switch (pfs->pfs_type) {
712 	case PFSmem:
713 	case PFSregs:
714 	case PFSfpregs:
715 #if defined(__HAVE_PROCFS_MACHDEP) && defined(PROCFS_MACHDEP_PROTECT_CASES)
716 	PROCFS_MACHDEP_PROTECT_CASES
717 #endif
718 		/*
719 		 * If the process has exercised some setuid or setgid
720 		 * privilege, then rip away read/write permission so
721 		 * that only root can gain access.
722 		 */
723 		if (procp->p_flag & PK_SUGID)
724 			vap->va_mode &= ~(S_IRUSR|S_IWUSR);
725 		/* FALLTHROUGH */
726 	case PFSctl:
727 	case PFSstatus:
728 	case PFSstat:
729 	case PFSnote:
730 	case PFSnotepg:
731 	case PFSmap:
732 	case PFSmaps:
733 	case PFScmdline:
734 	case PFSemul:
735 	case PFSstatm:
736 		vap->va_nlink = 1;
737 		vap->va_uid = kauth_cred_geteuid(procp->p_cred);
738 		vap->va_gid = kauth_cred_getegid(procp->p_cred);
739 		break;
740 	case PFSmeminfo:
741 	case PFSdevices:
742 	case PFScpuinfo:
743 	case PFSuptime:
744 	case PFSmounts:
745 	case PFScpustat:
746 	case PFSloadavg:
747 		vap->va_nlink = 1;
748 		vap->va_uid = vap->va_gid = 0;
749 		break;
750 
751 	default:
752 		break;
753 	}
754 
755 	/*
756 	 * now do the object specific fields
757 	 *
758 	 * The size could be set from struct reg, but it's hardly
759 	 * worth the trouble, and it puts some (potentially) machine
760 	 * dependent data into this machine-independent code.  If it
761 	 * becomes important then this function should break out into
762 	 * a per-file stat function in the corresponding .c file.
763 	 */
764 
765 	switch (pfs->pfs_type) {
766 	case PFSroot:
767 		/*
768 		 * Set nlink to 1 to tell fts(3) we don't actually know.
769 		 */
770 		vap->va_nlink = 1;
771 		vap->va_uid = 0;
772 		vap->va_gid = 0;
773 		vap->va_bytes = vap->va_size = DEV_BSIZE;
774 		break;
775 
776 	case PFSself:
777 	case PFScurproc: {
778 		char bf[16];		/* should be enough */
779 		vap->va_nlink = 1;
780 		vap->va_uid = 0;
781 		vap->va_gid = 0;
782 		vap->va_bytes = vap->va_size =
783 		    snprintf(bf, sizeof(bf), "%ld", (long)curproc->p_pid);
784 		break;
785 	}
786 
787 	case PFSfd:
788 		if (pfs->pfs_fd != -1) {
789 			file_t *fp;
790 
791 			fp = fd_getfile2(procp, pfs->pfs_fd);
792 			if (fp == NULL) {
793 				error = EBADF;
794 				break;
795 			}
796 			vap->va_nlink = 1;
797 			vap->va_uid = kauth_cred_geteuid(fp->f_cred);
798 			vap->va_gid = kauth_cred_getegid(fp->f_cred);
799 			switch (fp->f_type) {
800 			case DTYPE_VNODE:
801 				vap->va_bytes = vap->va_size =
802 				    ((struct vnode *)fp->f_data)->v_size;
803 				break;
804 			default:
805 				vap->va_bytes = vap->va_size = 0;
806 				break;
807 			}
808 			closef(fp);
809 			break;
810 		}
811 		/*FALLTHROUGH*/
812 	case PFSproc:
813 		vap->va_nlink = 2;
814 		vap->va_uid = kauth_cred_geteuid(procp->p_cred);
815 		vap->va_gid = kauth_cred_getegid(procp->p_cred);
816 		vap->va_bytes = vap->va_size = DEV_BSIZE;
817 		break;
818 
819 	case PFSfile:
820 		error = EOPNOTSUPP;
821 		break;
822 
823 	case PFSmem:
824 		vap->va_bytes = vap->va_size =
825 			ctob(procp->p_vmspace->vm_tsize +
826 				    procp->p_vmspace->vm_dsize +
827 				    procp->p_vmspace->vm_ssize);
828 		break;
829 
830 #if defined(PT_GETREGS) || defined(PT_SETREGS)
831 	case PFSregs:
832 		vap->va_bytes = vap->va_size = sizeof(struct reg);
833 		break;
834 #endif
835 
836 #if defined(PT_GETFPREGS) || defined(PT_SETFPREGS)
837 	case PFSfpregs:
838 		vap->va_bytes = vap->va_size = sizeof(struct fpreg);
839 		break;
840 #endif
841 
842 	case PFSctl:
843 	case PFSstatus:
844 	case PFSstat:
845 	case PFSnote:
846 	case PFSnotepg:
847 	case PFScmdline:
848 	case PFSmeminfo:
849 	case PFSdevices:
850 	case PFScpuinfo:
851 	case PFSuptime:
852 	case PFSmounts:
853 	case PFScpustat:
854 	case PFSloadavg:
855 	case PFSstatm:
856 		vap->va_bytes = vap->va_size = 0;
857 		break;
858 	case PFSmap:
859 	case PFSmaps:
860 		/*
861 		 * Advise a larger blocksize for the map files, so that
862 		 * they may be read in one pass.
863 		 */
864 		vap->va_blocksize = 4 * PAGE_SIZE;
865 		vap->va_bytes = vap->va_size = 0;
866 		break;
867 
868 	case PFScwd:
869 	case PFSchroot:
870 	case PFSexe: {
871 		char *bp;
872 
873 		vap->va_nlink = 1;
874 		vap->va_uid = 0;
875 		vap->va_gid = 0;
876 		bp = path + MAXPATHLEN;
877 		*--bp = '\0';
878 		procfs_dir(pfs->pfs_type, curlwp, procp, &bp, path,
879 		     MAXPATHLEN);
880 		vap->va_bytes = vap->va_size = strlen(bp);
881 		break;
882 	}
883 
884 	case PFSemul:
885 		vap->va_bytes = vap->va_size = strlen(procp->p_emul->e_name);
886 		break;
887 
888 #ifdef __HAVE_PROCFS_MACHDEP
889 	PROCFS_MACHDEP_NODETYPE_CASES
890 		error = procfs_machdep_getattr(ap->a_vp, vap, procp);
891 		break;
892 #endif
893 
894 	default:
895 		panic("procfs_getattr");
896 	}
897 
898 	if (procp != NULL)
899 		procfs_proc_unlock(procp);
900 	if (path != NULL)
901 		free(path, M_TEMP);
902 
903 	return (error);
904 }
905 
906 /*ARGSUSED*/
907 int
908 procfs_setattr(void *v)
909 {
910 	/*
911 	 * just fake out attribute setting
912 	 * it's not good to generate an error
913 	 * return, otherwise things like creat()
914 	 * will fail when they try to set the
915 	 * file length to 0.  worse, this means
916 	 * that echo $note > /proc/$pid/note will fail.
917 	 */
918 
919 	return (0);
920 }
921 
922 /*
923  * implement access checking.
924  *
925  * actually, the check for super-user is slightly
926  * broken since it will allow read access to write-only
927  * objects.  this doesn't cause any particular trouble
928  * but does mean that the i/o entry points need to check
929  * that the operation really does make sense.
930  */
931 int
932 procfs_access(v)
933 	void *v;
934 {
935 	struct vop_access_args /* {
936 		struct vnode *a_vp;
937 		int a_mode;
938 		kauth_cred_t a_cred;
939 	} */ *ap = v;
940 	struct vattr va;
941 	int error;
942 
943 	if ((error = VOP_GETATTR(ap->a_vp, &va, ap->a_cred)) != 0)
944 		return (error);
945 
946 	return (vaccess(va.va_type, va.va_mode,
947 	    va.va_uid, va.va_gid, ap->a_mode, ap->a_cred));
948 }
949 
950 /*
951  * lookup.  this is incredibly complicated in the
952  * general case, however for most pseudo-filesystems
953  * very little needs to be done.
954  *
955  * Locking isn't hard here, just poorly documented.
956  *
957  * If we're looking up ".", just vref the parent & return it.
958  *
959  * If we're looking up "..", unlock the parent, and lock "..". If everything
960  * went ok, and we're on the last component and the caller requested the
961  * parent locked, try to re-lock the parent. We do this to prevent lock
962  * races.
963  *
964  * For anything else, get the needed node. Then unlock the parent if not
965  * the last component or not LOCKPARENT (i.e. if we wouldn't re-lock the
966  * parent in the .. case).
967  *
968  * We try to exit with the parent locked in error cases.
969  */
970 int
971 procfs_lookup(v)
972 	void *v;
973 {
974 	struct vop_lookup_args /* {
975 		struct vnode * a_dvp;
976 		struct vnode ** a_vpp;
977 		struct componentname * a_cnp;
978 	} */ *ap = v;
979 	struct componentname *cnp = ap->a_cnp;
980 	struct vnode **vpp = ap->a_vpp;
981 	struct vnode *dvp = ap->a_dvp;
982 	const char *pname = cnp->cn_nameptr;
983 	const struct proc_target *pt = NULL;
984 	struct vnode *fvp;
985 	pid_t pid, vnpid;
986 	struct pfsnode *pfs;
987 	struct proc *p = NULL;
988 	int i, error;
989 	pfstype type;
990 
991 	*vpp = NULL;
992 
993 	if (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)
994 		return (EROFS);
995 
996 	if (cnp->cn_namelen == 1 && *pname == '.') {
997 		*vpp = dvp;
998 		VREF(dvp);
999 		return (0);
1000 	}
1001 
1002 	pfs = VTOPFS(dvp);
1003 	switch (pfs->pfs_type) {
1004 	case PFSroot:
1005 		/*
1006 		 * Shouldn't get here with .. in the root node.
1007 		 */
1008 		if (cnp->cn_flags & ISDOTDOT)
1009 			return (EIO);
1010 
1011 		for (i = 0; i < nproc_root_targets; i++) {
1012 			pt = &proc_root_targets[i];
1013 			/*
1014 			 * check for node match.  proc is always NULL here,
1015 			 * so call pt_valid with constant NULL lwp.
1016 			 */
1017 			if (cnp->cn_namelen == pt->pt_namlen &&
1018 			    memcmp(pt->pt_name, pname, cnp->cn_namelen) == 0 &&
1019 			    (pt->pt_valid == NULL ||
1020 			     (*pt->pt_valid)(NULL, dvp->v_mount)))
1021 				break;
1022 		}
1023 
1024 		if (i != nproc_root_targets) {
1025 			error = procfs_allocvp(dvp->v_mount, vpp, 0,
1026 			    pt->pt_pfstype, -1, NULL);
1027 			return (error);
1028 		}
1029 
1030 		if (CNEQ(cnp, "curproc", 7)) {
1031 			pid = curproc->p_pid;
1032 			vnpid = 0;
1033 			type = PFScurproc;
1034 		} else if (CNEQ(cnp, "self", 4)) {
1035 			pid = curproc->p_pid;
1036 			vnpid = 0;
1037 			type = PFSself;
1038 		} else {
1039 			pid = (pid_t)atoi(pname, cnp->cn_namelen);
1040 			vnpid = pid;
1041 			type = PFSproc;
1042 		}
1043 
1044 		if (procfs_proc_lock(pid, &p, ESRCH) != 0)
1045 			break;
1046 		error = procfs_allocvp(dvp->v_mount, vpp, vnpid, type, -1, p);
1047 		procfs_proc_unlock(p);
1048 		return (error);
1049 
1050 	case PFSproc:
1051 		/*
1052 		 * do the .. dance. We unlock the directory, and then
1053 		 * get the root dir. That will automatically return ..
1054 		 * locked. Then if the caller wanted dvp locked, we
1055 		 * re-lock.
1056 		 */
1057 		if (cnp->cn_flags & ISDOTDOT) {
1058 			VOP_UNLOCK(dvp, 0);
1059 			error = procfs_root(dvp->v_mount, vpp);
1060 			vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1061 			return (error);
1062 		}
1063 
1064 		if (procfs_proc_lock(pfs->pfs_pid, &p, ESRCH) != 0)
1065 			break;
1066 
1067 		for (pt = proc_targets, i = 0; i < nproc_targets; pt++, i++) {
1068 			struct lwp *plwp;
1069 			int found;
1070 
1071 			mutex_enter(p->p_lock);
1072 			plwp = LIST_FIRST(&p->p_lwps);
1073 			KASSERT(plwp != NULL);
1074 			lwp_addref(plwp);
1075 			mutex_exit(p->p_lock);
1076 			found = cnp->cn_namelen == pt->pt_namlen &&
1077 			    memcmp(pt->pt_name, pname, cnp->cn_namelen) == 0 &&
1078 			    (pt->pt_valid == NULL
1079 			      || (*pt->pt_valid)(plwp, dvp->v_mount));
1080 			lwp_delref(plwp);
1081 			if (found)
1082 				break;
1083 		}
1084 		if (i == nproc_targets) {
1085 			procfs_proc_unlock(p);
1086 			break;
1087 		}
1088 		if (pt->pt_pfstype == PFSfile) {
1089 			fvp = p->p_textvp;
1090 			/* We already checked that it exists. */
1091 			VREF(fvp);
1092 			procfs_proc_unlock(p);
1093 			vn_lock(fvp, LK_EXCLUSIVE | LK_RETRY);
1094 			*vpp = fvp;
1095 			return (0);
1096 		}
1097 
1098 		error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
1099 		    pt->pt_pfstype, -1, p);
1100 		procfs_proc_unlock(p);
1101 		return (error);
1102 
1103 	case PFSfd: {
1104 		int fd;
1105 		file_t *fp;
1106 
1107 		if ((error = procfs_proc_lock(pfs->pfs_pid, &p, ENOENT)) != 0)
1108 			return error;
1109 
1110 		/*
1111 		 * do the .. dance. We unlock the directory, and then
1112 		 * get the proc dir. That will automatically return ..
1113 		 * locked. Then re-lock the directory.
1114 		 */
1115 		if (cnp->cn_flags & ISDOTDOT) {
1116 			VOP_UNLOCK(dvp, 0);
1117 			error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
1118 			    PFSproc, -1, p);
1119 			procfs_proc_unlock(p);
1120 			vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1121 			return (error);
1122 		}
1123 		fd = atoi(pname, cnp->cn_namelen);
1124 
1125 		fp = fd_getfile2(p, fd);
1126 		if (fp == NULL) {
1127 			procfs_proc_unlock(p);
1128 			return ENOENT;
1129 		}
1130 		fvp = fp->f_data;
1131 
1132 		/* Don't show directories */
1133 		if (fp->f_type == DTYPE_VNODE && fvp->v_type != VDIR) {
1134 			VREF(fvp);
1135 			closef(fp);
1136 			procfs_proc_unlock(p);
1137 			vn_lock(fvp, LK_EXCLUSIVE | LK_RETRY |
1138 			    (p == curproc ? LK_CANRECURSE : 0));
1139 			*vpp = fvp;
1140 			return 0;
1141 		}
1142 
1143 		closef(fp);
1144 		error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
1145 		    PFSfd, fd, p);
1146 		procfs_proc_unlock(p);
1147 		return error;
1148 	}
1149 	default:
1150 		return (ENOTDIR);
1151 	}
1152 
1153 	return (cnp->cn_nameiop == LOOKUP ? ENOENT : EROFS);
1154 }
1155 
1156 int
1157 procfs_validfile(struct lwp *l, struct mount *mp)
1158 {
1159 	return l != NULL && l->l_proc != NULL && l->l_proc->p_textvp != NULL;
1160 }
1161 
1162 static int
1163 procfs_validfile_linux(l, mp)
1164 	struct lwp *l;
1165 	struct mount *mp;
1166 {
1167 	int flags;
1168 
1169 	flags = VFSTOPROC(mp)->pmnt_flags;
1170 	return (flags & PROCFSMNT_LINUXCOMPAT) &&
1171 	    (l == NULL || l->l_proc == NULL || procfs_validfile(l, mp));
1172 }
1173 
1174 struct procfs_root_readdir_ctx {
1175 	struct uio *uiop;
1176 	off_t *cookies;
1177 	int ncookies;
1178 	off_t off;
1179 	off_t startoff;
1180 	int error;
1181 };
1182 
1183 static int
1184 procfs_root_readdir_callback(struct proc *p, void *arg)
1185 {
1186 	struct procfs_root_readdir_ctx *ctxp = arg;
1187 	struct dirent d;
1188 	struct uio *uiop;
1189 	int error;
1190 
1191 	uiop = ctxp->uiop;
1192 	if (uiop->uio_resid < UIO_MX)
1193 		return -1; /* no space */
1194 
1195 	if (ctxp->off < ctxp->startoff) {
1196 		ctxp->off++;
1197 		return 0;
1198 	}
1199 
1200 	if (kauth_authorize_process(kauth_cred_get(),
1201 	    KAUTH_PROCESS_CANSEE, p,
1202 	    KAUTH_ARG(KAUTH_REQ_PROCESS_CANSEE_ENTRY), NULL, NULL) != 0)
1203 		return 0;
1204 
1205 	memset(&d, 0, UIO_MX);
1206 	d.d_reclen = UIO_MX;
1207 	d.d_fileno = PROCFS_FILENO(p->p_pid, PFSproc, -1);
1208 	d.d_namlen = snprintf(d.d_name,
1209 	    UIO_MX - offsetof(struct dirent, d_name), "%ld", (long)p->p_pid);
1210 	d.d_type = DT_DIR;
1211 
1212 	mutex_exit(proc_lock);
1213 	error = uiomove(&d, UIO_MX, uiop);
1214 	mutex_enter(proc_lock);
1215 	if (error) {
1216 		ctxp->error = error;
1217 		return -1;
1218 	}
1219 
1220 	ctxp->ncookies++;
1221 	if (ctxp->cookies)
1222 		*(ctxp->cookies)++ = ctxp->off + 1;
1223 	ctxp->off++;
1224 
1225 	return 0;
1226 }
1227 
1228 /*
1229  * readdir returns directory entries from pfsnode (vp).
1230  *
1231  * the strategy here with procfs is to generate a single
1232  * directory entry at a time (struct dirent) and then
1233  * copy that out to userland using uiomove.  a more efficent
1234  * though more complex implementation, would try to minimize
1235  * the number of calls to uiomove().  for procfs, this is
1236  * hardly worth the added code complexity.
1237  *
1238  * this should just be done through read()
1239  */
1240 int
1241 procfs_readdir(v)
1242 	void *v;
1243 {
1244 	struct vop_readdir_args /* {
1245 		struct vnode *a_vp;
1246 		struct uio *a_uio;
1247 		kauth_cred_t a_cred;
1248 		int *a_eofflag;
1249 		off_t **a_cookies;
1250 		int *a_ncookies;
1251 	} */ *ap = v;
1252 	struct uio *uio = ap->a_uio;
1253 	struct dirent d;
1254 	struct pfsnode *pfs;
1255 	off_t i;
1256 	int error;
1257 	off_t *cookies = NULL;
1258 	int ncookies;
1259 	struct vnode *vp;
1260 	const struct proc_target *pt;
1261 	struct procfs_root_readdir_ctx ctx;
1262 	struct lwp *l;
1263 	int nfd;
1264 
1265 	vp = ap->a_vp;
1266 	pfs = VTOPFS(vp);
1267 
1268 	if (uio->uio_resid < UIO_MX)
1269 		return (EINVAL);
1270 	if (uio->uio_offset < 0)
1271 		return (EINVAL);
1272 
1273 	error = 0;
1274 	i = uio->uio_offset;
1275 	memset(&d, 0, UIO_MX);
1276 	d.d_reclen = UIO_MX;
1277 	ncookies = uio->uio_resid / UIO_MX;
1278 
1279 	switch (pfs->pfs_type) {
1280 	/*
1281 	 * this is for the process-specific sub-directories.
1282 	 * all that is needed to is copy out all the entries
1283 	 * from the procent[] table (top of this file).
1284 	 */
1285 	case PFSproc: {
1286 		struct proc *p;
1287 
1288 		if (i >= nproc_targets)
1289 			return 0;
1290 
1291 		if (procfs_proc_lock(pfs->pfs_pid, &p, ESRCH) != 0)
1292 			break;
1293 
1294 		if (ap->a_ncookies) {
1295 			ncookies = min(ncookies, (nproc_targets - i));
1296 			cookies = malloc(ncookies * sizeof (off_t),
1297 			    M_TEMP, M_WAITOK);
1298 			*ap->a_cookies = cookies;
1299 		}
1300 
1301 		for (pt = &proc_targets[i];
1302 		     uio->uio_resid >= UIO_MX && i < nproc_targets; pt++, i++) {
1303 			if (pt->pt_valid) {
1304 				/* XXXSMP LWP can disappear */
1305 				mutex_enter(p->p_lock);
1306 				l = LIST_FIRST(&p->p_lwps);
1307 				KASSERT(l != NULL);
1308 				mutex_exit(p->p_lock);
1309 				if ((*pt->pt_valid)(l, vp->v_mount) == 0)
1310 					continue;
1311 			}
1312 
1313 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid,
1314 			    pt->pt_pfstype, -1);
1315 			d.d_namlen = pt->pt_namlen;
1316 			memcpy(d.d_name, pt->pt_name, pt->pt_namlen + 1);
1317 			d.d_type = pt->pt_type;
1318 
1319 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1320 				break;
1321 			if (cookies)
1322 				*cookies++ = i + 1;
1323 		}
1324 
1325 		procfs_proc_unlock(p);
1326 	    	break;
1327 	}
1328 	case PFSfd: {
1329 		struct proc *p;
1330 		file_t *fp;
1331 		int lim, nc = 0;
1332 
1333 		if ((error = procfs_proc_lock(pfs->pfs_pid, &p, ESRCH)) != 0)
1334 			return error;
1335 
1336 		/* XXX Should this be by file as well? */
1337 		if (kauth_authorize_process(kauth_cred_get(),
1338 		    KAUTH_PROCESS_CANSEE, p,
1339 		    KAUTH_ARG(KAUTH_REQ_PROCESS_CANSEE_OPENFILES), NULL,
1340 		    NULL) != 0) {
1341 		    	procfs_proc_unlock(p);
1342 			return ESRCH;
1343 		}
1344 
1345 		nfd = p->p_fd->fd_nfiles;
1346 
1347 		lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfiles);
1348 		if (i >= lim) {
1349 		    	procfs_proc_unlock(p);
1350 			return 0;
1351 		}
1352 
1353 		if (ap->a_ncookies) {
1354 			ncookies = min(ncookies, (nfd + 2 - i));
1355 			cookies = malloc(ncookies * sizeof (off_t),
1356 			    M_TEMP, M_WAITOK);
1357 			*ap->a_cookies = cookies;
1358 		}
1359 
1360 		for (; i < 2 && uio->uio_resid >= UIO_MX; i++) {
1361 			pt = &proc_targets[i];
1362 			d.d_namlen = pt->pt_namlen;
1363 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid,
1364 			    pt->pt_pfstype, -1);
1365 			(void)memcpy(d.d_name, pt->pt_name, pt->pt_namlen + 1);
1366 			d.d_type = pt->pt_type;
1367 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1368 				break;
1369 			if (cookies)
1370 				*cookies++ = i + 1;
1371 			nc++;
1372 		}
1373 		if (error) {
1374 			ncookies = nc;
1375 			break;
1376 		}
1377 		for (; uio->uio_resid >= UIO_MX && i < nfd; i++) {
1378 			/* check the descriptor exists */
1379 			if ((fp = fd_getfile2(p, i - 2)) == NULL)
1380 				continue;
1381 			closef(fp);
1382 
1383 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid, PFSfd, i - 2);
1384 			d.d_namlen = snprintf(d.d_name, sizeof(d.d_name),
1385 			    "%lld", (long long)(i - 2));
1386 			d.d_type = VREG;
1387 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1388 				break;
1389 			if (cookies)
1390 				*cookies++ = i + 1;
1391 			nc++;
1392 		}
1393 		ncookies = nc;
1394 		procfs_proc_unlock(p);
1395 		break;
1396 	}
1397 
1398 	/*
1399 	 * this is for the root of the procfs filesystem
1400 	 * what is needed are special entries for "curproc"
1401 	 * and "self" followed by an entry for each process
1402 	 * on allproc.
1403 	 */
1404 
1405 	case PFSroot: {
1406 		int nc = 0;
1407 
1408 		if (ap->a_ncookies) {
1409 			/*
1410 			 * XXX Potentially allocating too much space here,
1411 			 * but I'm lazy. This loop needs some work.
1412 			 */
1413 			cookies = malloc(ncookies * sizeof (off_t),
1414 			    M_TEMP, M_WAITOK);
1415 			*ap->a_cookies = cookies;
1416 		}
1417 		error = 0;
1418 		/* 0 ... 3 are static entries. */
1419 		for (; i <= 3 && uio->uio_resid >= UIO_MX; i++) {
1420 			switch (i) {
1421 			case 0:		/* `.' */
1422 			case 1:		/* `..' */
1423 				d.d_fileno = PROCFS_FILENO(0, PFSroot, -1);
1424 				d.d_namlen = i + 1;
1425 				memcpy(d.d_name, "..", d.d_namlen);
1426 				d.d_name[i + 1] = '\0';
1427 				d.d_type = DT_DIR;
1428 				break;
1429 
1430 			case 2:
1431 				d.d_fileno = PROCFS_FILENO(0, PFScurproc, -1);
1432 				d.d_namlen = sizeof("curproc") - 1;
1433 				memcpy(d.d_name, "curproc", sizeof("curproc"));
1434 				d.d_type = DT_LNK;
1435 				break;
1436 
1437 			case 3:
1438 				d.d_fileno = PROCFS_FILENO(0, PFSself, -1);
1439 				d.d_namlen = sizeof("self") - 1;
1440 				memcpy(d.d_name, "self", sizeof("self"));
1441 				d.d_type = DT_LNK;
1442 				break;
1443 			}
1444 
1445 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1446 				break;
1447 			nc++;
1448 			if (cookies)
1449 				*cookies++ = i + 1;
1450 		}
1451 		/* 4 ... are process entries. */
1452 		ctx.uiop = uio;
1453 		ctx.error = 0;
1454 		ctx.off = 4;
1455 		ctx.startoff = i;
1456 		ctx.cookies = cookies;
1457 		ctx.ncookies = nc;
1458 		proclist_foreach_call(&allproc,
1459 		    procfs_root_readdir_callback, &ctx);
1460 		cookies = ctx.cookies;
1461 		nc = ctx.ncookies;
1462 		error = ctx.error;
1463 		if (error)
1464 			break;
1465 
1466 		/* misc entries. */
1467 		if (i < ctx.off)
1468 			i = ctx.off;
1469 		if (i >= ctx.off + nproc_root_targets)
1470 			break;
1471 		for (pt = &proc_root_targets[i - ctx.off];
1472 		    uio->uio_resid >= UIO_MX &&
1473 		    pt < &proc_root_targets[nproc_root_targets];
1474 		    pt++, i++) {
1475 			if (pt->pt_valid &&
1476 			    (*pt->pt_valid)(NULL, vp->v_mount) == 0)
1477 				continue;
1478 			d.d_fileno = PROCFS_FILENO(0, pt->pt_pfstype, -1);
1479 			d.d_namlen = pt->pt_namlen;
1480 			memcpy(d.d_name, pt->pt_name, pt->pt_namlen + 1);
1481 			d.d_type = pt->pt_type;
1482 
1483 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1484 				break;
1485 			nc++;
1486 			if (cookies)
1487 				*cookies++ = i + 1;
1488 		}
1489 
1490 		ncookies = nc;
1491 		break;
1492 	}
1493 
1494 	default:
1495 		error = ENOTDIR;
1496 		break;
1497 	}
1498 
1499 	if (ap->a_ncookies) {
1500 		if (error) {
1501 			if (cookies)
1502 				free(*ap->a_cookies, M_TEMP);
1503 			*ap->a_ncookies = 0;
1504 			*ap->a_cookies = NULL;
1505 		} else
1506 			*ap->a_ncookies = ncookies;
1507 	}
1508 	uio->uio_offset = i;
1509 	return (error);
1510 }
1511 
1512 /*
1513  * readlink reads the link of `curproc' and others
1514  */
1515 int
1516 procfs_readlink(v)
1517 	void *v;
1518 {
1519 	struct vop_readlink_args *ap = v;
1520 	char bf[16];		/* should be enough */
1521 	char *bp = bf;
1522 	char *path = NULL;
1523 	int len = 0;
1524 	int error = 0;
1525 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
1526 	struct proc *pown;
1527 
1528 	if (pfs->pfs_fileno == PROCFS_FILENO(0, PFScurproc, -1))
1529 		len = snprintf(bf, sizeof(bf), "%ld", (long)curproc->p_pid);
1530 	else if (pfs->pfs_fileno == PROCFS_FILENO(0, PFSself, -1))
1531 		len = snprintf(bf, sizeof(bf), "%s", "curproc");
1532 	else if (pfs->pfs_fileno == PROCFS_FILENO(pfs->pfs_pid, PFScwd, -1) ||
1533 	    pfs->pfs_fileno == PROCFS_FILENO(pfs->pfs_pid, PFSchroot, -1) ||
1534 	    pfs->pfs_fileno == PROCFS_FILENO(pfs->pfs_pid, PFSexe, -1)) {
1535 		if ((error = procfs_proc_lock(pfs->pfs_pid, &pown, ESRCH)) != 0)
1536 			return error;
1537 		MALLOC(path, char *, MAXPATHLEN + 4, M_TEMP,
1538 		    M_WAITOK|M_CANFAIL);
1539 		if (path == NULL) {
1540 			procfs_proc_unlock(pown);
1541 			return (ENOMEM);
1542 		}
1543 		bp = path + MAXPATHLEN;
1544 		*--bp = '\0';
1545 		procfs_dir(PROCFS_TYPE(pfs->pfs_fileno), curlwp, pown,
1546 		    &bp, path, MAXPATHLEN);
1547 		procfs_proc_unlock(pown);
1548 		len = strlen(bp);
1549 	} else {
1550 		file_t *fp;
1551 		struct vnode *vxp, *vp;
1552 
1553 		if ((error = procfs_proc_lock(pfs->pfs_pid, &pown, ESRCH)) != 0)
1554 			return error;
1555 
1556 		fp = fd_getfile2(pown, pfs->pfs_fd);
1557 		if (fp == NULL) {
1558 			procfs_proc_unlock(pown);
1559 			return EBADF;
1560 		}
1561 
1562 		switch (fp->f_type) {
1563 		case DTYPE_VNODE:
1564 			vxp = (struct vnode *)fp->f_data;
1565 			if (vxp->v_type != VDIR) {
1566 				error = EINVAL;
1567 				break;
1568 			}
1569 			if ((path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK))
1570 			    == NULL) {
1571 				error = ENOMEM;
1572 				break;
1573 			}
1574 			bp = path + MAXPATHLEN;
1575 			*--bp = '\0';
1576 
1577 			/*
1578 			 * XXX: kludge to avoid locking against ourselves
1579 			 * in getcwd()
1580 			 */
1581 			if (vxp->v_tag == VT_PROCFS) {
1582 				*--bp = '/';
1583 			} else {
1584 				rw_enter(&curproc->p_cwdi->cwdi_lock, RW_READER);
1585 				vp = curproc->p_cwdi->cwdi_rdir;
1586 				if (vp == NULL)
1587 					vp = rootvnode;
1588 				error = getcwd_common(vxp, vp, &bp, path,
1589 				    MAXPATHLEN / 2, 0, curlwp);
1590 				rw_exit(&curproc->p_cwdi->cwdi_lock);
1591 			}
1592 			if (error)
1593 				break;
1594 			len = strlen(bp);
1595 			break;
1596 
1597 		case DTYPE_MISC:
1598 			len = snprintf(bf, sizeof(bf), "%s", "[misc]");
1599 			break;
1600 
1601 		case DTYPE_KQUEUE:
1602 			len = snprintf(bf, sizeof(bf), "%s", "[kqueue]");
1603 			break;
1604 
1605 		default:
1606 			error = EINVAL;
1607 			break;
1608 		}
1609 		closef(fp);
1610 		procfs_proc_unlock(pown);
1611 	}
1612 
1613 	if (error == 0)
1614 		error = uiomove(bp, len, ap->a_uio);
1615 	if (path)
1616 		free(path, M_TEMP);
1617 	return error;
1618 }
1619 
1620 /*
1621  * convert decimal ascii to int
1622  */
1623 static int
1624 atoi(b, len)
1625 	const char *b;
1626 	size_t len;
1627 {
1628 	int p = 0;
1629 
1630 	while (len--) {
1631 		char c = *b++;
1632 		if (c < '0' || c > '9')
1633 			return -1;
1634 		p = 10 * p + (c - '0');
1635 	}
1636 
1637 	return p;
1638 }
1639