xref: /netbsd-src/sys/miscfs/procfs/procfs_vnops.c (revision da9817918ec7e88db2912a2882967c7570a83f47)
1 /*	$NetBSD: procfs_vnops.c,v 1.174 2009/05/24 21:41:26 ad 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 1.174 2009/05/24 21:41:26 ad 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(void *v)
303 {
304 	struct vop_open_args /* {
305 		struct vnode *a_vp;
306 		int  a_mode;
307 		kauth_cred_t a_cred;
308 	} */ *ap = v;
309 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
310 	struct lwp *l1;
311 	struct proc *p2;
312 	int error;
313 
314 	if ((error = procfs_proc_lock(pfs->pfs_pid, &p2, ENOENT)) != 0)
315 		return error;
316 
317 	l1 = curlwp;				/* tracer */
318 
319 #define	M2K(m)	(((m) & FREAD) && ((m) & FWRITE) ? \
320 		 KAUTH_REQ_PROCESS_PROCFS_RW : \
321 		 (m) & FWRITE ? KAUTH_REQ_PROCESS_PROCFS_WRITE : \
322 		 KAUTH_REQ_PROCESS_PROCFS_READ)
323 
324 	mutex_enter(p2->p_lock);
325 	error = kauth_authorize_process(l1->l_cred, KAUTH_PROCESS_PROCFS,
326 	    p2, pfs, KAUTH_ARG(M2K(ap->a_mode)), NULL);
327 	mutex_exit(p2->p_lock);
328 	if (error) {
329 		procfs_proc_unlock(p2);
330 		return (error);
331 	}
332 
333 #undef M2K
334 
335 	switch (pfs->pfs_type) {
336 	case PFSmem:
337 		if (((pfs->pfs_flags & FWRITE) && (ap->a_mode & O_EXCL)) ||
338 		    ((pfs->pfs_flags & O_EXCL) && (ap->a_mode & FWRITE))) {
339 			error = EBUSY;
340 			break;
341 		}
342 
343 		if (!proc_isunder(p2, l1)) {
344 			error = EPERM;
345 			break;
346 		}
347 
348 		if (ap->a_mode & FWRITE)
349 			pfs->pfs_flags = ap->a_mode & (FWRITE|O_EXCL);
350 
351 		break;
352 
353 	case PFSregs:
354 	case PFSfpregs:
355 		if (!proc_isunder(p2, l1)) {
356 			error = EPERM;
357 			break;
358 		}
359 		break;
360 
361 	default:
362 		break;
363 	}
364 
365 	procfs_proc_unlock(p2);
366 	return (error);
367 }
368 
369 /*
370  * close the pfsnode (vp) after doing i/o.
371  * (vp) is not locked on entry or exit.
372  *
373  * nothing to do for procfs other than undo
374  * any exclusive open flag (see _open above).
375  */
376 int
377 procfs_close(void *v)
378 {
379 	struct vop_close_args /* {
380 		struct vnode *a_vp;
381 		int  a_fflag;
382 		kauth_cred_t a_cred;
383 	} */ *ap = v;
384 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
385 
386 	switch (pfs->pfs_type) {
387 	case PFSmem:
388 		if ((ap->a_fflag & FWRITE) && (pfs->pfs_flags & O_EXCL))
389 			pfs->pfs_flags &= ~(FWRITE|O_EXCL);
390 		break;
391 
392 	default:
393 		break;
394 	}
395 
396 	return (0);
397 }
398 
399 /*
400  * _inactive is called when the pfsnode
401  * is vrele'd and the reference count goes
402  * to zero.  (vp) will be on the vnode free
403  * list, so to get it back vget() must be
404  * used.
405  *
406  * (vp) is locked on entry, but must be unlocked on exit.
407  */
408 int
409 procfs_inactive(void *v)
410 {
411 	struct vop_inactive_args /* {
412 		struct vnode *a_vp;
413 		bool *a_recycle;
414 	} */ *ap = v;
415 	struct vnode *vp = ap->a_vp;
416 	struct pfsnode *pfs = VTOPFS(vp);
417 
418 	mutex_enter(proc_lock);
419 	*ap->a_recycle = (p_find(pfs->pfs_pid, PFIND_LOCKED) == NULL);
420 	mutex_exit(proc_lock);
421 
422 	VOP_UNLOCK(vp, 0);
423 
424 	return (0);
425 }
426 
427 /*
428  * _reclaim is called when getnewvnode()
429  * wants to make use of an entry on the vnode
430  * free list.  at this time the filesystem needs
431  * to free any private data and remove the node
432  * from any private lists.
433  */
434 int
435 procfs_reclaim(void *v)
436 {
437 	struct vop_reclaim_args /* {
438 		struct vnode *a_vp;
439 	} */ *ap = v;
440 
441 	return (procfs_freevp(ap->a_vp));
442 }
443 
444 /*
445  * Return POSIX pathconf information applicable to special devices.
446  */
447 int
448 procfs_pathconf(void *v)
449 {
450 	struct vop_pathconf_args /* {
451 		struct vnode *a_vp;
452 		int a_name;
453 		register_t *a_retval;
454 	} */ *ap = v;
455 
456 	switch (ap->a_name) {
457 	case _PC_LINK_MAX:
458 		*ap->a_retval = LINK_MAX;
459 		return (0);
460 	case _PC_MAX_CANON:
461 		*ap->a_retval = MAX_CANON;
462 		return (0);
463 	case _PC_MAX_INPUT:
464 		*ap->a_retval = MAX_INPUT;
465 		return (0);
466 	case _PC_PIPE_BUF:
467 		*ap->a_retval = PIPE_BUF;
468 		return (0);
469 	case _PC_CHOWN_RESTRICTED:
470 		*ap->a_retval = 1;
471 		return (0);
472 	case _PC_VDISABLE:
473 		*ap->a_retval = _POSIX_VDISABLE;
474 		return (0);
475 	case _PC_SYNC_IO:
476 		*ap->a_retval = 1;
477 		return (0);
478 	default:
479 		return (EINVAL);
480 	}
481 	/* NOTREACHED */
482 }
483 
484 /*
485  * _print is used for debugging.
486  * just print a readable description
487  * of (vp).
488  */
489 int
490 procfs_print(void *v)
491 {
492 	struct vop_print_args /* {
493 		struct vnode *a_vp;
494 	} */ *ap = v;
495 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
496 
497 	printf("tag VT_PROCFS, type %d, pid %d, mode %x, flags %lx\n",
498 	    pfs->pfs_type, pfs->pfs_pid, pfs->pfs_mode, pfs->pfs_flags);
499 	return 0;
500 }
501 
502 int
503 procfs_link(void *v)
504 {
505 	struct vop_link_args /* {
506 		struct vnode *a_dvp;
507 		struct vnode *a_vp;
508 		struct componentname *a_cnp;
509 	} */ *ap = v;
510 
511 	VOP_ABORTOP(ap->a_dvp, ap->a_cnp);
512 	vput(ap->a_dvp);
513 	return (EROFS);
514 }
515 
516 int
517 procfs_symlink(void *v)
518 {
519 	struct vop_symlink_args /* {
520 		struct vnode *a_dvp;
521 		struct vnode **a_vpp;
522 		struct componentname *a_cnp;
523 		struct vattr *a_vap;
524 		char *a_target;
525 	} */ *ap = v;
526 
527 	VOP_ABORTOP(ap->a_dvp, ap->a_cnp);
528 	vput(ap->a_dvp);
529 	return (EROFS);
530 }
531 
532 /*
533  * Works out the path to (and vnode of) the target process's current
534  * working directory or chroot.  If the caller is in a chroot and
535  * can't "reach" the target's cwd or root (or some other error
536  * occurs), a "/" is returned for the path and a NULL pointer is
537  * returned for the vnode.
538  */
539 static void
540 procfs_dir(pfstype t, struct lwp *caller, struct proc *target, char **bpp,
541     char *path, size_t len)
542 {
543 	struct cwdinfo *cwdi;
544 	struct vnode *vp, *rvp;
545 	char *bp;
546 
547 	cwdi = caller->l_proc->p_cwdi;
548 	rw_enter(&cwdi->cwdi_lock, RW_READER);
549 
550 	rvp = cwdi->cwdi_rdir;
551 	bp = bpp ? *bpp : NULL;
552 
553 	switch (t) {
554 	case PFScwd:
555 		vp = target->p_cwdi->cwdi_cdir;
556 		break;
557 	case PFSchroot:
558 		vp = target->p_cwdi->cwdi_rdir;
559 		break;
560 	case PFSexe:
561 		vp = target->p_textvp;
562 		break;
563 	default:
564 		rw_exit(&cwdi->cwdi_lock);
565 		return;
566 	}
567 
568 	/*
569 	 * XXX: this horrible kludge avoids locking panics when
570 	 * attempting to lookup links that point to within procfs
571 	 */
572 	if (vp != NULL && vp->v_tag == VT_PROCFS) {
573 		if (bpp) {
574 			*--bp = '/';
575 			*bpp = bp;
576 		}
577 		rw_exit(&cwdi->cwdi_lock);
578 		return;
579 	}
580 
581 	if (rvp == NULL)
582 		rvp = rootvnode;
583 	if (vp == NULL || getcwd_common(vp, rvp, bp ? &bp : NULL, path,
584 	    len / 2, 0, caller) != 0) {
585 		vp = NULL;
586 		if (bpp) {
587 /*
588 			if (t == PFSexe) {
589 				snprintf(path, len, "%s/%d/file"
590 				    mp->mnt_stat.f_mntonname, pfs->pfs_pid);
591 			} else */ {
592 				bp = *bpp;
593 				*--bp = '/';
594 			}
595 		}
596 	}
597 
598 	if (bpp)
599 		*bpp = bp;
600 
601 	rw_exit(&cwdi->cwdi_lock);
602 }
603 
604 /*
605  * Invent attributes for pfsnode (vp) and store
606  * them in (vap).
607  * Directories lengths are returned as zero since
608  * any real length would require the genuine size
609  * to be computed, and nothing cares anyway.
610  *
611  * this is relatively minimal for procfs.
612  */
613 int
614 procfs_getattr(void *v)
615 {
616 	struct vop_getattr_args /* {
617 		struct vnode *a_vp;
618 		struct vattr *a_vap;
619 		kauth_cred_t a_cred;
620 	} */ *ap = v;
621 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
622 	struct vattr *vap = ap->a_vap;
623 	struct proc *procp;
624 	char *path;
625 	int error;
626 
627 	/* first check the process still exists */
628 	switch (pfs->pfs_type) {
629 	case PFSroot:
630 	case PFScurproc:
631 	case PFSself:
632 		procp = NULL;
633 		break;
634 
635 	default:
636 		error = procfs_proc_lock(pfs->pfs_pid, &procp, ENOENT);
637 		if (error != 0)
638 			return (error);
639 		break;
640 	}
641 
642 	switch (pfs->pfs_type) {
643 	case PFScwd:
644 	case PFSchroot:
645 	case PFSexe:
646 		path = malloc(MAXPATHLEN + 4, M_TEMP, M_WAITOK|M_CANFAIL);
647 		if (path == NULL && procp != NULL) {
648 			procfs_proc_unlock(procp);
649 			return (ENOMEM);
650 		}
651 		break;
652 
653 	default:
654 		path = NULL;
655 		break;
656 	}
657 
658 	if (procp != NULL) {
659 		mutex_enter(procp->p_lock);
660 		error = kauth_authorize_process(kauth_cred_get(),
661 		    KAUTH_PROCESS_CANSEE, procp,
662 		    KAUTH_ARG(KAUTH_REQ_PROCESS_CANSEE_ENTRY), NULL, NULL);
663 		mutex_exit(procp->p_lock);
664 		if (error != 0) {
665 		    	procfs_proc_unlock(procp);
666 		    	if (path != NULL)
667 		    		free(path, M_TEMP);
668 			return (ENOENT);
669 		}
670 	}
671 
672 	error = 0;
673 
674 	/* start by zeroing out the attributes */
675 	VATTR_NULL(vap);
676 
677 	/* next do all the common fields */
678 	vap->va_type = ap->a_vp->v_type;
679 	vap->va_mode = pfs->pfs_mode;
680 	vap->va_fileid = pfs->pfs_fileno;
681 	vap->va_flags = 0;
682 	vap->va_blocksize = PAGE_SIZE;
683 
684 	/*
685 	 * Make all times be current TOD.
686 	 *
687 	 * It would be possible to get the process start
688 	 * time from the p_stats structure, but there's
689 	 * no "file creation" time stamp anyway, and the
690 	 * p_stats structure is not addressable if u. gets
691 	 * swapped out for that process.
692 	 */
693 	getnanotime(&vap->va_ctime);
694 	vap->va_atime = vap->va_mtime = vap->va_ctime;
695 	if (procp)
696 		TIMEVAL_TO_TIMESPEC(&procp->p_stats->p_start,
697 		    &vap->va_birthtime);
698 	else
699 		getnanotime(&vap->va_birthtime);
700 
701 	switch (pfs->pfs_type) {
702 	case PFSmem:
703 	case PFSregs:
704 	case PFSfpregs:
705 #if defined(__HAVE_PROCFS_MACHDEP) && defined(PROCFS_MACHDEP_PROTECT_CASES)
706 	PROCFS_MACHDEP_PROTECT_CASES
707 #endif
708 		/*
709 		 * If the process has exercised some setuid or setgid
710 		 * privilege, then rip away read/write permission so
711 		 * that only root can gain access.
712 		 */
713 		if (procp->p_flag & PK_SUGID)
714 			vap->va_mode &= ~(S_IRUSR|S_IWUSR);
715 		/* FALLTHROUGH */
716 	case PFSctl:
717 	case PFSstatus:
718 	case PFSstat:
719 	case PFSnote:
720 	case PFSnotepg:
721 	case PFSmap:
722 	case PFSmaps:
723 	case PFScmdline:
724 	case PFSemul:
725 	case PFSstatm:
726 		vap->va_nlink = 1;
727 		vap->va_uid = kauth_cred_geteuid(procp->p_cred);
728 		vap->va_gid = kauth_cred_getegid(procp->p_cred);
729 		break;
730 	case PFSmeminfo:
731 	case PFSdevices:
732 	case PFScpuinfo:
733 	case PFSuptime:
734 	case PFSmounts:
735 	case PFScpustat:
736 	case PFSloadavg:
737 		vap->va_nlink = 1;
738 		vap->va_uid = vap->va_gid = 0;
739 		break;
740 
741 	default:
742 		break;
743 	}
744 
745 	/*
746 	 * now do the object specific fields
747 	 *
748 	 * The size could be set from struct reg, but it's hardly
749 	 * worth the trouble, and it puts some (potentially) machine
750 	 * dependent data into this machine-independent code.  If it
751 	 * becomes important then this function should break out into
752 	 * a per-file stat function in the corresponding .c file.
753 	 */
754 
755 	switch (pfs->pfs_type) {
756 	case PFSroot:
757 		/*
758 		 * Set nlink to 1 to tell fts(3) we don't actually know.
759 		 */
760 		vap->va_nlink = 1;
761 		vap->va_uid = 0;
762 		vap->va_gid = 0;
763 		vap->va_bytes = vap->va_size = DEV_BSIZE;
764 		break;
765 
766 	case PFSself:
767 	case PFScurproc: {
768 		char bf[16];		/* should be enough */
769 		vap->va_nlink = 1;
770 		vap->va_uid = 0;
771 		vap->va_gid = 0;
772 		vap->va_bytes = vap->va_size =
773 		    snprintf(bf, sizeof(bf), "%ld", (long)curproc->p_pid);
774 		break;
775 	}
776 
777 	case PFSfd:
778 		if (pfs->pfs_fd != -1) {
779 			file_t *fp;
780 
781 			fp = fd_getfile2(procp, pfs->pfs_fd);
782 			if (fp == NULL) {
783 				error = EBADF;
784 				break;
785 			}
786 			vap->va_nlink = 1;
787 			vap->va_uid = kauth_cred_geteuid(fp->f_cred);
788 			vap->va_gid = kauth_cred_getegid(fp->f_cred);
789 			switch (fp->f_type) {
790 			case DTYPE_VNODE:
791 				vap->va_bytes = vap->va_size =
792 				    ((struct vnode *)fp->f_data)->v_size;
793 				break;
794 			default:
795 				vap->va_bytes = vap->va_size = 0;
796 				break;
797 			}
798 			closef(fp);
799 			break;
800 		}
801 		/*FALLTHROUGH*/
802 	case PFSproc:
803 		vap->va_nlink = 2;
804 		vap->va_uid = kauth_cred_geteuid(procp->p_cred);
805 		vap->va_gid = kauth_cred_getegid(procp->p_cred);
806 		vap->va_bytes = vap->va_size = DEV_BSIZE;
807 		break;
808 
809 	case PFSfile:
810 		error = EOPNOTSUPP;
811 		break;
812 
813 	case PFSmem:
814 		vap->va_bytes = vap->va_size =
815 			ctob(procp->p_vmspace->vm_tsize +
816 				    procp->p_vmspace->vm_dsize +
817 				    procp->p_vmspace->vm_ssize);
818 		break;
819 
820 #if defined(PT_GETREGS) || defined(PT_SETREGS)
821 	case PFSregs:
822 		vap->va_bytes = vap->va_size = sizeof(struct reg);
823 		break;
824 #endif
825 
826 #if defined(PT_GETFPREGS) || defined(PT_SETFPREGS)
827 	case PFSfpregs:
828 		vap->va_bytes = vap->va_size = sizeof(struct fpreg);
829 		break;
830 #endif
831 
832 	case PFSctl:
833 	case PFSstatus:
834 	case PFSstat:
835 	case PFSnote:
836 	case PFSnotepg:
837 	case PFScmdline:
838 	case PFSmeminfo:
839 	case PFSdevices:
840 	case PFScpuinfo:
841 	case PFSuptime:
842 	case PFSmounts:
843 	case PFScpustat:
844 	case PFSloadavg:
845 	case PFSstatm:
846 		vap->va_bytes = vap->va_size = 0;
847 		break;
848 	case PFSmap:
849 	case PFSmaps:
850 		/*
851 		 * Advise a larger blocksize for the map files, so that
852 		 * they may be read in one pass.
853 		 */
854 		vap->va_blocksize = 4 * PAGE_SIZE;
855 		vap->va_bytes = vap->va_size = 0;
856 		break;
857 
858 	case PFScwd:
859 	case PFSchroot:
860 	case PFSexe: {
861 		char *bp;
862 
863 		vap->va_nlink = 1;
864 		vap->va_uid = 0;
865 		vap->va_gid = 0;
866 		bp = path + MAXPATHLEN;
867 		*--bp = '\0';
868 		procfs_dir(pfs->pfs_type, curlwp, procp, &bp, path,
869 		     MAXPATHLEN);
870 		vap->va_bytes = vap->va_size = strlen(bp);
871 		break;
872 	}
873 
874 	case PFSemul:
875 		vap->va_bytes = vap->va_size = strlen(procp->p_emul->e_name);
876 		break;
877 
878 #ifdef __HAVE_PROCFS_MACHDEP
879 	PROCFS_MACHDEP_NODETYPE_CASES
880 		error = procfs_machdep_getattr(ap->a_vp, vap, procp);
881 		break;
882 #endif
883 
884 	default:
885 		panic("procfs_getattr");
886 	}
887 
888 	if (procp != NULL)
889 		procfs_proc_unlock(procp);
890 	if (path != NULL)
891 		free(path, M_TEMP);
892 
893 	return (error);
894 }
895 
896 /*ARGSUSED*/
897 int
898 procfs_setattr(void *v)
899 {
900 	/*
901 	 * just fake out attribute setting
902 	 * it's not good to generate an error
903 	 * return, otherwise things like creat()
904 	 * will fail when they try to set the
905 	 * file length to 0.  worse, this means
906 	 * that echo $note > /proc/$pid/note will fail.
907 	 */
908 
909 	return (0);
910 }
911 
912 /*
913  * implement access checking.
914  *
915  * actually, the check for super-user is slightly
916  * broken since it will allow read access to write-only
917  * objects.  this doesn't cause any particular trouble
918  * but does mean that the i/o entry points need to check
919  * that the operation really does make sense.
920  */
921 int
922 procfs_access(void *v)
923 {
924 	struct vop_access_args /* {
925 		struct vnode *a_vp;
926 		int a_mode;
927 		kauth_cred_t a_cred;
928 	} */ *ap = v;
929 	struct vattr va;
930 	int error;
931 
932 	if ((error = VOP_GETATTR(ap->a_vp, &va, ap->a_cred)) != 0)
933 		return (error);
934 
935 	return (vaccess(va.va_type, va.va_mode,
936 	    va.va_uid, va.va_gid, ap->a_mode, ap->a_cred));
937 }
938 
939 /*
940  * lookup.  this is incredibly complicated in the
941  * general case, however for most pseudo-filesystems
942  * very little needs to be done.
943  *
944  * Locking isn't hard here, just poorly documented.
945  *
946  * If we're looking up ".", just vref the parent & return it.
947  *
948  * If we're looking up "..", unlock the parent, and lock "..". If everything
949  * went ok, and we're on the last component and the caller requested the
950  * parent locked, try to re-lock the parent. We do this to prevent lock
951  * races.
952  *
953  * For anything else, get the needed node. Then unlock the parent if not
954  * the last component or not LOCKPARENT (i.e. if we wouldn't re-lock the
955  * parent in the .. case).
956  *
957  * We try to exit with the parent locked in error cases.
958  */
959 int
960 procfs_lookup(void *v)
961 {
962 	struct vop_lookup_args /* {
963 		struct vnode * a_dvp;
964 		struct vnode ** a_vpp;
965 		struct componentname * a_cnp;
966 	} */ *ap = v;
967 	struct componentname *cnp = ap->a_cnp;
968 	struct vnode **vpp = ap->a_vpp;
969 	struct vnode *dvp = ap->a_dvp;
970 	const char *pname = cnp->cn_nameptr;
971 	const struct proc_target *pt = NULL;
972 	struct vnode *fvp;
973 	pid_t pid, vnpid;
974 	struct pfsnode *pfs;
975 	struct proc *p = NULL;
976 	struct lwp *plwp;
977 	int i, error;
978 	pfstype type;
979 
980 	*vpp = NULL;
981 
982 	if (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)
983 		return (EROFS);
984 
985 	if (cnp->cn_namelen == 1 && *pname == '.') {
986 		*vpp = dvp;
987 		VREF(dvp);
988 		return (0);
989 	}
990 
991 	pfs = VTOPFS(dvp);
992 	switch (pfs->pfs_type) {
993 	case PFSroot:
994 		/*
995 		 * Shouldn't get here with .. in the root node.
996 		 */
997 		if (cnp->cn_flags & ISDOTDOT)
998 			return (EIO);
999 
1000 		for (i = 0; i < nproc_root_targets; i++) {
1001 			pt = &proc_root_targets[i];
1002 			/*
1003 			 * check for node match.  proc is always NULL here,
1004 			 * so call pt_valid with constant NULL lwp.
1005 			 */
1006 			if (cnp->cn_namelen == pt->pt_namlen &&
1007 			    memcmp(pt->pt_name, pname, cnp->cn_namelen) == 0 &&
1008 			    (pt->pt_valid == NULL ||
1009 			     (*pt->pt_valid)(NULL, dvp->v_mount)))
1010 				break;
1011 		}
1012 
1013 		if (i != nproc_root_targets) {
1014 			error = procfs_allocvp(dvp->v_mount, vpp, 0,
1015 			    pt->pt_pfstype, -1, NULL);
1016 			return (error);
1017 		}
1018 
1019 		if (CNEQ(cnp, "curproc", 7)) {
1020 			pid = curproc->p_pid;
1021 			vnpid = 0;
1022 			type = PFScurproc;
1023 		} else if (CNEQ(cnp, "self", 4)) {
1024 			pid = curproc->p_pid;
1025 			vnpid = 0;
1026 			type = PFSself;
1027 		} else {
1028 			pid = (pid_t)atoi(pname, cnp->cn_namelen);
1029 			vnpid = pid;
1030 			type = PFSproc;
1031 		}
1032 
1033 		if (procfs_proc_lock(pid, &p, ESRCH) != 0)
1034 			break;
1035 		error = procfs_allocvp(dvp->v_mount, vpp, vnpid, type, -1, p);
1036 		procfs_proc_unlock(p);
1037 		return (error);
1038 
1039 	case PFSproc:
1040 		/*
1041 		 * do the .. dance. We unlock the directory, and then
1042 		 * get the root dir. That will automatically return ..
1043 		 * locked. Then if the caller wanted dvp locked, we
1044 		 * re-lock.
1045 		 */
1046 		if (cnp->cn_flags & ISDOTDOT) {
1047 			VOP_UNLOCK(dvp, 0);
1048 			error = procfs_root(dvp->v_mount, vpp);
1049 			vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1050 			return (error);
1051 		}
1052 
1053 		if (procfs_proc_lock(pfs->pfs_pid, &p, ESRCH) != 0)
1054 			break;
1055 
1056 		mutex_enter(p->p_lock);
1057 		LIST_FOREACH(plwp, &p->p_lwps, l_sibling) {
1058 			if (plwp->l_stat != LSZOMB)
1059 				break;
1060 		}
1061 		/* Process is exiting if no-LWPS or all LWPs are LSZOMB */
1062 		if (plwp == NULL) {
1063 			mutex_exit(p->p_lock);
1064 			procfs_proc_unlock(p);
1065 			return ESRCH;
1066 		}
1067 
1068 		lwp_addref(plwp);
1069 		mutex_exit(p->p_lock);
1070 
1071 		for (pt = proc_targets, i = 0; i < nproc_targets; pt++, i++) {
1072 			int found;
1073 
1074 			found = cnp->cn_namelen == pt->pt_namlen &&
1075 			    memcmp(pt->pt_name, pname, cnp->cn_namelen) == 0 &&
1076 			    (pt->pt_valid == NULL
1077 			      || (*pt->pt_valid)(plwp, dvp->v_mount));
1078 			if (found)
1079 				break;
1080 		}
1081 		lwp_delref(plwp);
1082 
1083 		if (i == nproc_targets) {
1084 			procfs_proc_unlock(p);
1085 			break;
1086 		}
1087 		if (pt->pt_pfstype == PFSfile) {
1088 			fvp = p->p_textvp;
1089 			/* We already checked that it exists. */
1090 			VREF(fvp);
1091 			procfs_proc_unlock(p);
1092 			vn_lock(fvp, LK_EXCLUSIVE | LK_RETRY);
1093 			*vpp = fvp;
1094 			return (0);
1095 		}
1096 
1097 		error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
1098 		    pt->pt_pfstype, -1, p);
1099 		procfs_proc_unlock(p);
1100 		return (error);
1101 
1102 	case PFSfd: {
1103 		int fd;
1104 		file_t *fp;
1105 
1106 		if ((error = procfs_proc_lock(pfs->pfs_pid, &p, ENOENT)) != 0)
1107 			return error;
1108 
1109 		/*
1110 		 * do the .. dance. We unlock the directory, and then
1111 		 * get the proc dir. That will automatically return ..
1112 		 * locked. Then re-lock the directory.
1113 		 */
1114 		if (cnp->cn_flags & ISDOTDOT) {
1115 			VOP_UNLOCK(dvp, 0);
1116 			error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
1117 			    PFSproc, -1, p);
1118 			procfs_proc_unlock(p);
1119 			vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1120 			return (error);
1121 		}
1122 		fd = atoi(pname, cnp->cn_namelen);
1123 
1124 		fp = fd_getfile2(p, fd);
1125 		if (fp == NULL) {
1126 			procfs_proc_unlock(p);
1127 			return ENOENT;
1128 		}
1129 		fvp = fp->f_data;
1130 
1131 		/* Don't show directories */
1132 		if (fp->f_type == DTYPE_VNODE && fvp->v_type != VDIR) {
1133 			VREF(fvp);
1134 			closef(fp);
1135 			procfs_proc_unlock(p);
1136 			vn_lock(fvp, LK_EXCLUSIVE | LK_RETRY |
1137 			    (p == curproc ? LK_CANRECURSE : 0));
1138 			*vpp = fvp;
1139 			return 0;
1140 		}
1141 
1142 		closef(fp);
1143 		error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
1144 		    PFSfd, fd, p);
1145 		procfs_proc_unlock(p);
1146 		return error;
1147 	}
1148 	default:
1149 		return (ENOTDIR);
1150 	}
1151 
1152 	return (cnp->cn_nameiop == LOOKUP ? ENOENT : EROFS);
1153 }
1154 
1155 int
1156 procfs_validfile(struct lwp *l, struct mount *mp)
1157 {
1158 	return l != NULL && l->l_proc != NULL && l->l_proc->p_textvp != NULL;
1159 }
1160 
1161 static int
1162 procfs_validfile_linux(struct lwp *l, struct mount *mp)
1163 {
1164 	int flags;
1165 
1166 	flags = VFSTOPROC(mp)->pmnt_flags;
1167 	return (flags & PROCFSMNT_LINUXCOMPAT) &&
1168 	    (l == NULL || l->l_proc == NULL || procfs_validfile(l, mp));
1169 }
1170 
1171 struct procfs_root_readdir_ctx {
1172 	struct uio *uiop;
1173 	off_t *cookies;
1174 	int ncookies;
1175 	off_t off;
1176 	off_t startoff;
1177 	int error;
1178 };
1179 
1180 static int
1181 procfs_root_readdir_callback(struct proc *p, void *arg)
1182 {
1183 	struct procfs_root_readdir_ctx *ctxp = arg;
1184 	struct dirent d;
1185 	struct uio *uiop;
1186 	int error;
1187 
1188 	uiop = ctxp->uiop;
1189 	if (uiop->uio_resid < UIO_MX)
1190 		return -1; /* no space */
1191 
1192 	if (ctxp->off < ctxp->startoff) {
1193 		ctxp->off++;
1194 		return 0;
1195 	}
1196 
1197 	if (kauth_authorize_process(kauth_cred_get(),
1198 	    KAUTH_PROCESS_CANSEE, p,
1199 	    KAUTH_ARG(KAUTH_REQ_PROCESS_CANSEE_ENTRY), NULL, NULL) != 0)
1200 		return 0;
1201 
1202 	memset(&d, 0, UIO_MX);
1203 	d.d_reclen = UIO_MX;
1204 	d.d_fileno = PROCFS_FILENO(p->p_pid, PFSproc, -1);
1205 	d.d_namlen = snprintf(d.d_name,
1206 	    UIO_MX - offsetof(struct dirent, d_name), "%ld", (long)p->p_pid);
1207 	d.d_type = DT_DIR;
1208 
1209 	mutex_exit(proc_lock);
1210 	error = uiomove(&d, UIO_MX, uiop);
1211 	mutex_enter(proc_lock);
1212 	if (error) {
1213 		ctxp->error = error;
1214 		return -1;
1215 	}
1216 
1217 	ctxp->ncookies++;
1218 	if (ctxp->cookies)
1219 		*(ctxp->cookies)++ = ctxp->off + 1;
1220 	ctxp->off++;
1221 
1222 	return 0;
1223 }
1224 
1225 /*
1226  * readdir returns directory entries from pfsnode (vp).
1227  *
1228  * the strategy here with procfs is to generate a single
1229  * directory entry at a time (struct dirent) and then
1230  * copy that out to userland using uiomove.  a more efficent
1231  * though more complex implementation, would try to minimize
1232  * the number of calls to uiomove().  for procfs, this is
1233  * hardly worth the added code complexity.
1234  *
1235  * this should just be done through read()
1236  */
1237 int
1238 procfs_readdir(void *v)
1239 {
1240 	struct vop_readdir_args /* {
1241 		struct vnode *a_vp;
1242 		struct uio *a_uio;
1243 		kauth_cred_t a_cred;
1244 		int *a_eofflag;
1245 		off_t **a_cookies;
1246 		int *a_ncookies;
1247 	} */ *ap = v;
1248 	struct uio *uio = ap->a_uio;
1249 	struct dirent d;
1250 	struct pfsnode *pfs;
1251 	off_t i;
1252 	int error;
1253 	off_t *cookies = NULL;
1254 	int ncookies;
1255 	struct vnode *vp;
1256 	const struct proc_target *pt;
1257 	struct procfs_root_readdir_ctx ctx;
1258 	struct lwp *l;
1259 	int nfd;
1260 
1261 	vp = ap->a_vp;
1262 	pfs = VTOPFS(vp);
1263 
1264 	if (uio->uio_resid < UIO_MX)
1265 		return (EINVAL);
1266 	if (uio->uio_offset < 0)
1267 		return (EINVAL);
1268 
1269 	error = 0;
1270 	i = uio->uio_offset;
1271 	memset(&d, 0, UIO_MX);
1272 	d.d_reclen = UIO_MX;
1273 	ncookies = uio->uio_resid / UIO_MX;
1274 
1275 	switch (pfs->pfs_type) {
1276 	/*
1277 	 * this is for the process-specific sub-directories.
1278 	 * all that is needed to is copy out all the entries
1279 	 * from the procent[] table (top of this file).
1280 	 */
1281 	case PFSproc: {
1282 		struct proc *p;
1283 
1284 		if (i >= nproc_targets)
1285 			return 0;
1286 
1287 		if (procfs_proc_lock(pfs->pfs_pid, &p, ESRCH) != 0)
1288 			break;
1289 
1290 		if (ap->a_ncookies) {
1291 			ncookies = min(ncookies, (nproc_targets - i));
1292 			cookies = malloc(ncookies * sizeof (off_t),
1293 			    M_TEMP, M_WAITOK);
1294 			*ap->a_cookies = cookies;
1295 		}
1296 
1297 		for (pt = &proc_targets[i];
1298 		     uio->uio_resid >= UIO_MX && i < nproc_targets; pt++, i++) {
1299 			if (pt->pt_valid) {
1300 				/* XXXSMP LWP can disappear */
1301 				mutex_enter(p->p_lock);
1302 				l = LIST_FIRST(&p->p_lwps);
1303 				KASSERT(l != NULL);
1304 				mutex_exit(p->p_lock);
1305 				if ((*pt->pt_valid)(l, vp->v_mount) == 0)
1306 					continue;
1307 			}
1308 
1309 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid,
1310 			    pt->pt_pfstype, -1);
1311 			d.d_namlen = pt->pt_namlen;
1312 			memcpy(d.d_name, pt->pt_name, pt->pt_namlen + 1);
1313 			d.d_type = pt->pt_type;
1314 
1315 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1316 				break;
1317 			if (cookies)
1318 				*cookies++ = i + 1;
1319 		}
1320 
1321 		procfs_proc_unlock(p);
1322 	    	break;
1323 	}
1324 	case PFSfd: {
1325 		struct proc *p;
1326 		file_t *fp;
1327 		int lim, nc = 0;
1328 
1329 		if ((error = procfs_proc_lock(pfs->pfs_pid, &p, ESRCH)) != 0)
1330 			return error;
1331 
1332 		/* XXX Should this be by file as well? */
1333 		if (kauth_authorize_process(kauth_cred_get(),
1334 		    KAUTH_PROCESS_CANSEE, p,
1335 		    KAUTH_ARG(KAUTH_REQ_PROCESS_CANSEE_OPENFILES), NULL,
1336 		    NULL) != 0) {
1337 		    	procfs_proc_unlock(p);
1338 			return ESRCH;
1339 		}
1340 
1341 		nfd = p->p_fd->fd_dt->dt_nfiles;
1342 
1343 		lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfiles);
1344 		if (i >= lim) {
1345 		    	procfs_proc_unlock(p);
1346 			return 0;
1347 		}
1348 
1349 		if (ap->a_ncookies) {
1350 			ncookies = min(ncookies, (nfd + 2 - i));
1351 			cookies = malloc(ncookies * sizeof (off_t),
1352 			    M_TEMP, M_WAITOK);
1353 			*ap->a_cookies = cookies;
1354 		}
1355 
1356 		for (; i < 2 && uio->uio_resid >= UIO_MX; i++) {
1357 			pt = &proc_targets[i];
1358 			d.d_namlen = pt->pt_namlen;
1359 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid,
1360 			    pt->pt_pfstype, -1);
1361 			(void)memcpy(d.d_name, pt->pt_name, pt->pt_namlen + 1);
1362 			d.d_type = pt->pt_type;
1363 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1364 				break;
1365 			if (cookies)
1366 				*cookies++ = i + 1;
1367 			nc++;
1368 		}
1369 		if (error) {
1370 			ncookies = nc;
1371 			break;
1372 		}
1373 		for (; uio->uio_resid >= UIO_MX && i < nfd; i++) {
1374 			/* check the descriptor exists */
1375 			if ((fp = fd_getfile2(p, i - 2)) == NULL)
1376 				continue;
1377 			closef(fp);
1378 
1379 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid, PFSfd, i - 2);
1380 			d.d_namlen = snprintf(d.d_name, sizeof(d.d_name),
1381 			    "%lld", (long long)(i - 2));
1382 			d.d_type = VREG;
1383 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1384 				break;
1385 			if (cookies)
1386 				*cookies++ = i + 1;
1387 			nc++;
1388 		}
1389 		ncookies = nc;
1390 		procfs_proc_unlock(p);
1391 		break;
1392 	}
1393 
1394 	/*
1395 	 * this is for the root of the procfs filesystem
1396 	 * what is needed are special entries for "curproc"
1397 	 * and "self" followed by an entry for each process
1398 	 * on allproc.
1399 	 */
1400 
1401 	case PFSroot: {
1402 		int nc = 0;
1403 
1404 		if (ap->a_ncookies) {
1405 			/*
1406 			 * XXX Potentially allocating too much space here,
1407 			 * but I'm lazy. This loop needs some work.
1408 			 */
1409 			cookies = malloc(ncookies * sizeof (off_t),
1410 			    M_TEMP, M_WAITOK);
1411 			*ap->a_cookies = cookies;
1412 		}
1413 		error = 0;
1414 		/* 0 ... 3 are static entries. */
1415 		for (; i <= 3 && uio->uio_resid >= UIO_MX; i++) {
1416 			switch (i) {
1417 			case 0:		/* `.' */
1418 			case 1:		/* `..' */
1419 				d.d_fileno = PROCFS_FILENO(0, PFSroot, -1);
1420 				d.d_namlen = i + 1;
1421 				memcpy(d.d_name, "..", d.d_namlen);
1422 				d.d_name[i + 1] = '\0';
1423 				d.d_type = DT_DIR;
1424 				break;
1425 
1426 			case 2:
1427 				d.d_fileno = PROCFS_FILENO(0, PFScurproc, -1);
1428 				d.d_namlen = sizeof("curproc") - 1;
1429 				memcpy(d.d_name, "curproc", sizeof("curproc"));
1430 				d.d_type = DT_LNK;
1431 				break;
1432 
1433 			case 3:
1434 				d.d_fileno = PROCFS_FILENO(0, PFSself, -1);
1435 				d.d_namlen = sizeof("self") - 1;
1436 				memcpy(d.d_name, "self", sizeof("self"));
1437 				d.d_type = DT_LNK;
1438 				break;
1439 			}
1440 
1441 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1442 				break;
1443 			nc++;
1444 			if (cookies)
1445 				*cookies++ = i + 1;
1446 		}
1447 		/* 4 ... are process entries. */
1448 		ctx.uiop = uio;
1449 		ctx.error = 0;
1450 		ctx.off = 4;
1451 		ctx.startoff = i;
1452 		ctx.cookies = cookies;
1453 		ctx.ncookies = nc;
1454 		proclist_foreach_call(&allproc,
1455 		    procfs_root_readdir_callback, &ctx);
1456 		cookies = ctx.cookies;
1457 		nc = ctx.ncookies;
1458 		error = ctx.error;
1459 		if (error)
1460 			break;
1461 
1462 		/* misc entries. */
1463 		if (i < ctx.off)
1464 			i = ctx.off;
1465 		if (i >= ctx.off + nproc_root_targets)
1466 			break;
1467 		for (pt = &proc_root_targets[i - ctx.off];
1468 		    uio->uio_resid >= UIO_MX &&
1469 		    pt < &proc_root_targets[nproc_root_targets];
1470 		    pt++, i++) {
1471 			if (pt->pt_valid &&
1472 			    (*pt->pt_valid)(NULL, vp->v_mount) == 0)
1473 				continue;
1474 			d.d_fileno = PROCFS_FILENO(0, pt->pt_pfstype, -1);
1475 			d.d_namlen = pt->pt_namlen;
1476 			memcpy(d.d_name, pt->pt_name, pt->pt_namlen + 1);
1477 			d.d_type = pt->pt_type;
1478 
1479 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1480 				break;
1481 			nc++;
1482 			if (cookies)
1483 				*cookies++ = i + 1;
1484 		}
1485 
1486 		ncookies = nc;
1487 		break;
1488 	}
1489 
1490 	default:
1491 		error = ENOTDIR;
1492 		break;
1493 	}
1494 
1495 	if (ap->a_ncookies) {
1496 		if (error) {
1497 			if (cookies)
1498 				free(*ap->a_cookies, M_TEMP);
1499 			*ap->a_ncookies = 0;
1500 			*ap->a_cookies = NULL;
1501 		} else
1502 			*ap->a_ncookies = ncookies;
1503 	}
1504 	uio->uio_offset = i;
1505 	return (error);
1506 }
1507 
1508 /*
1509  * readlink reads the link of `curproc' and others
1510  */
1511 int
1512 procfs_readlink(void *v)
1513 {
1514 	struct vop_readlink_args *ap = v;
1515 	char bf[16];		/* should be enough */
1516 	char *bp = bf;
1517 	char *path = NULL;
1518 	int len = 0;
1519 	int error = 0;
1520 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
1521 	struct proc *pown;
1522 
1523 	if (pfs->pfs_fileno == PROCFS_FILENO(0, PFScurproc, -1))
1524 		len = snprintf(bf, sizeof(bf), "%ld", (long)curproc->p_pid);
1525 	else if (pfs->pfs_fileno == PROCFS_FILENO(0, PFSself, -1))
1526 		len = snprintf(bf, sizeof(bf), "%s", "curproc");
1527 	else if (pfs->pfs_fileno == PROCFS_FILENO(pfs->pfs_pid, PFScwd, -1) ||
1528 	    pfs->pfs_fileno == PROCFS_FILENO(pfs->pfs_pid, PFSchroot, -1) ||
1529 	    pfs->pfs_fileno == PROCFS_FILENO(pfs->pfs_pid, PFSexe, -1)) {
1530 		if ((error = procfs_proc_lock(pfs->pfs_pid, &pown, ESRCH)) != 0)
1531 			return error;
1532 		path = malloc(MAXPATHLEN + 4, M_TEMP, M_WAITOK|M_CANFAIL);
1533 		if (path == NULL) {
1534 			procfs_proc_unlock(pown);
1535 			return (ENOMEM);
1536 		}
1537 		bp = path + MAXPATHLEN;
1538 		*--bp = '\0';
1539 		procfs_dir(PROCFS_TYPE(pfs->pfs_fileno), curlwp, pown,
1540 		    &bp, path, MAXPATHLEN);
1541 		procfs_proc_unlock(pown);
1542 		len = strlen(bp);
1543 	} else {
1544 		file_t *fp;
1545 		struct vnode *vxp, *vp;
1546 
1547 		if ((error = procfs_proc_lock(pfs->pfs_pid, &pown, ESRCH)) != 0)
1548 			return error;
1549 
1550 		fp = fd_getfile2(pown, pfs->pfs_fd);
1551 		if (fp == NULL) {
1552 			procfs_proc_unlock(pown);
1553 			return EBADF;
1554 		}
1555 
1556 		switch (fp->f_type) {
1557 		case DTYPE_VNODE:
1558 			vxp = (struct vnode *)fp->f_data;
1559 			if (vxp->v_type != VDIR) {
1560 				error = EINVAL;
1561 				break;
1562 			}
1563 			if ((path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK))
1564 			    == NULL) {
1565 				error = ENOMEM;
1566 				break;
1567 			}
1568 			bp = path + MAXPATHLEN;
1569 			*--bp = '\0';
1570 
1571 			/*
1572 			 * XXX: kludge to avoid locking against ourselves
1573 			 * in getcwd()
1574 			 */
1575 			if (vxp->v_tag == VT_PROCFS) {
1576 				*--bp = '/';
1577 			} else {
1578 				rw_enter(&curproc->p_cwdi->cwdi_lock, RW_READER);
1579 				vp = curproc->p_cwdi->cwdi_rdir;
1580 				if (vp == NULL)
1581 					vp = rootvnode;
1582 				error = getcwd_common(vxp, vp, &bp, path,
1583 				    MAXPATHLEN / 2, 0, curlwp);
1584 				rw_exit(&curproc->p_cwdi->cwdi_lock);
1585 			}
1586 			if (error)
1587 				break;
1588 			len = strlen(bp);
1589 			break;
1590 
1591 		case DTYPE_MISC:
1592 			len = snprintf(bf, sizeof(bf), "%s", "[misc]");
1593 			break;
1594 
1595 		case DTYPE_KQUEUE:
1596 			len = snprintf(bf, sizeof(bf), "%s", "[kqueue]");
1597 			break;
1598 
1599 		default:
1600 			error = EINVAL;
1601 			break;
1602 		}
1603 		closef(fp);
1604 		procfs_proc_unlock(pown);
1605 	}
1606 
1607 	if (error == 0)
1608 		error = uiomove(bp, len, ap->a_uio);
1609 	if (path)
1610 		free(path, M_TEMP);
1611 	return error;
1612 }
1613 
1614 /*
1615  * convert decimal ascii to int
1616  */
1617 static int
1618 atoi(const char *b, size_t len)
1619 {
1620 	int p = 0;
1621 
1622 	while (len--) {
1623 		char c = *b++;
1624 		if (c < '0' || c > '9')
1625 			return -1;
1626 		p = 10 * p + (c - '0');
1627 	}
1628 
1629 	return p;
1630 }
1631