xref: /netbsd-src/sys/miscfs/procfs/procfs_vnops.c (revision 88fcb00c0357f2d7c1774f86a352637bfda96184)
1 /*	$NetBSD: procfs_vnops.c,v 1.181 2011/06/23 17:06:38 christos 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 1.181 2011/06/23 17:06:38 christos 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 = (proc_find(pfs->pfs_pid) == NULL);
420 	mutex_exit(proc_lock);
421 
422 	VOP_UNLOCK(vp);
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 		if (pfs->pfs_type == PFSmap || pfs->pfs_type == PFSmaps)
727 			vap->va_mode = S_IRUSR;
728 		vap->va_nlink = 1;
729 		vap->va_uid = kauth_cred_geteuid(procp->p_cred);
730 		vap->va_gid = kauth_cred_getegid(procp->p_cred);
731 		break;
732 	case PFSmeminfo:
733 	case PFSdevices:
734 	case PFScpuinfo:
735 	case PFSuptime:
736 	case PFSmounts:
737 	case PFScpustat:
738 	case PFSloadavg:
739 		vap->va_nlink = 1;
740 		vap->va_uid = vap->va_gid = 0;
741 		break;
742 
743 	default:
744 		break;
745 	}
746 
747 	/*
748 	 * now do the object specific fields
749 	 *
750 	 * The size could be set from struct reg, but it's hardly
751 	 * worth the trouble, and it puts some (potentially) machine
752 	 * dependent data into this machine-independent code.  If it
753 	 * becomes important then this function should break out into
754 	 * a per-file stat function in the corresponding .c file.
755 	 */
756 
757 	switch (pfs->pfs_type) {
758 	case PFSroot:
759 		/*
760 		 * Set nlink to 1 to tell fts(3) we don't actually know.
761 		 */
762 		vap->va_nlink = 1;
763 		vap->va_uid = 0;
764 		vap->va_gid = 0;
765 		vap->va_bytes = vap->va_size = DEV_BSIZE;
766 		break;
767 
768 	case PFSself:
769 	case PFScurproc: {
770 		char bf[16];		/* should be enough */
771 		vap->va_nlink = 1;
772 		vap->va_uid = 0;
773 		vap->va_gid = 0;
774 		vap->va_bytes = vap->va_size =
775 		    snprintf(bf, sizeof(bf), "%ld", (long)curproc->p_pid);
776 		break;
777 	}
778 
779 	case PFSfd:
780 		if (pfs->pfs_fd != -1) {
781 			file_t *fp;
782 
783 			fp = fd_getfile2(procp, pfs->pfs_fd);
784 			if (fp == NULL) {
785 				error = EBADF;
786 				break;
787 			}
788 			vap->va_nlink = 1;
789 			vap->va_uid = kauth_cred_geteuid(fp->f_cred);
790 			vap->va_gid = kauth_cred_getegid(fp->f_cred);
791 			switch (fp->f_type) {
792 			case DTYPE_VNODE:
793 				vap->va_bytes = vap->va_size =
794 				    ((struct vnode *)fp->f_data)->v_size;
795 				break;
796 			default:
797 				vap->va_bytes = vap->va_size = 0;
798 				break;
799 			}
800 			closef(fp);
801 			break;
802 		}
803 		/*FALLTHROUGH*/
804 	case PFSproc:
805 		vap->va_nlink = 2;
806 		vap->va_uid = kauth_cred_geteuid(procp->p_cred);
807 		vap->va_gid = kauth_cred_getegid(procp->p_cred);
808 		vap->va_bytes = vap->va_size = DEV_BSIZE;
809 		break;
810 
811 	case PFSfile:
812 		error = EOPNOTSUPP;
813 		break;
814 
815 	case PFSmem:
816 		vap->va_bytes = vap->va_size =
817 			ctob(procp->p_vmspace->vm_tsize +
818 				    procp->p_vmspace->vm_dsize +
819 				    procp->p_vmspace->vm_ssize);
820 		break;
821 
822 #if defined(PT_GETREGS) || defined(PT_SETREGS)
823 	case PFSregs:
824 		vap->va_bytes = vap->va_size = sizeof(struct reg);
825 		break;
826 #endif
827 
828 #if defined(PT_GETFPREGS) || defined(PT_SETFPREGS)
829 	case PFSfpregs:
830 		vap->va_bytes = vap->va_size = sizeof(struct fpreg);
831 		break;
832 #endif
833 
834 	case PFSctl:
835 	case PFSstatus:
836 	case PFSstat:
837 	case PFSnote:
838 	case PFSnotepg:
839 	case PFScmdline:
840 	case PFSmeminfo:
841 	case PFSdevices:
842 	case PFScpuinfo:
843 	case PFSuptime:
844 	case PFSmounts:
845 	case PFScpustat:
846 	case PFSloadavg:
847 	case PFSstatm:
848 		vap->va_bytes = vap->va_size = 0;
849 		break;
850 	case PFSmap:
851 	case PFSmaps:
852 		/*
853 		 * Advise a larger blocksize for the map files, so that
854 		 * they may be read in one pass.
855 		 */
856 		vap->va_blocksize = 4 * PAGE_SIZE;
857 		vap->va_bytes = vap->va_size = 0;
858 		break;
859 
860 	case PFScwd:
861 	case PFSchroot:
862 	case PFSexe: {
863 		char *bp;
864 
865 		vap->va_nlink = 1;
866 		vap->va_uid = 0;
867 		vap->va_gid = 0;
868 		bp = path + MAXPATHLEN;
869 		*--bp = '\0';
870 		procfs_dir(pfs->pfs_type, curlwp, procp, &bp, path,
871 		     MAXPATHLEN);
872 		vap->va_bytes = vap->va_size = strlen(bp);
873 		break;
874 	}
875 
876 	case PFSemul:
877 		vap->va_bytes = vap->va_size = strlen(procp->p_emul->e_name);
878 		break;
879 
880 #ifdef __HAVE_PROCFS_MACHDEP
881 	PROCFS_MACHDEP_NODETYPE_CASES
882 		error = procfs_machdep_getattr(ap->a_vp, vap, procp);
883 		break;
884 #endif
885 
886 	default:
887 		panic("procfs_getattr");
888 	}
889 
890 	if (procp != NULL)
891 		procfs_proc_unlock(procp);
892 	if (path != NULL)
893 		free(path, M_TEMP);
894 
895 	return (error);
896 }
897 
898 /*ARGSUSED*/
899 int
900 procfs_setattr(void *v)
901 {
902 	/*
903 	 * just fake out attribute setting
904 	 * it's not good to generate an error
905 	 * return, otherwise things like creat()
906 	 * will fail when they try to set the
907 	 * file length to 0.  worse, this means
908 	 * that echo $note > /proc/$pid/note will fail.
909 	 */
910 
911 	return (0);
912 }
913 
914 static int
915 procfs_check_possible(struct vnode *vp, mode_t mode)
916 {
917 
918 	return 0;
919 }
920 
921 static int
922 procfs_check_permitted(struct vattr *va, mode_t mode, kauth_cred_t cred)
923 {
924 
925 	return genfs_can_access(va->va_type, va->va_mode,
926 	    va->va_uid, va->va_gid, mode, cred);
927 }
928 
929 /*
930  * implement access checking.
931  *
932  * actually, the check for super-user is slightly
933  * broken since it will allow read access to write-only
934  * objects.  this doesn't cause any particular trouble
935  * but does mean that the i/o entry points need to check
936  * that the operation really does make sense.
937  */
938 int
939 procfs_access(void *v)
940 {
941 	struct vop_access_args /* {
942 		struct vnode *a_vp;
943 		int a_mode;
944 		kauth_cred_t a_cred;
945 	} */ *ap = v;
946 	struct vattr va;
947 	int error;
948 
949 	if ((error = VOP_GETATTR(ap->a_vp, &va, ap->a_cred)) != 0)
950 		return (error);
951 
952 	error = procfs_check_possible(ap->a_vp, ap->a_mode);
953 	if (error)
954 		return error;
955 
956 	error = procfs_check_permitted(&va, ap->a_mode, ap->a_cred);
957 
958 	return error;
959 }
960 
961 /*
962  * lookup.  this is incredibly complicated in the
963  * general case, however for most pseudo-filesystems
964  * very little needs to be done.
965  *
966  * Locking isn't hard here, just poorly documented.
967  *
968  * If we're looking up ".", just vref the parent & return it.
969  *
970  * If we're looking up "..", unlock the parent, and lock "..". If everything
971  * went ok, and we're on the last component and the caller requested the
972  * parent locked, try to re-lock the parent. We do this to prevent lock
973  * races.
974  *
975  * For anything else, get the needed node. Then unlock the parent if not
976  * the last component or not LOCKPARENT (i.e. if we wouldn't re-lock the
977  * parent in the .. case).
978  *
979  * We try to exit with the parent locked in error cases.
980  */
981 int
982 procfs_lookup(void *v)
983 {
984 	struct vop_lookup_args /* {
985 		struct vnode * a_dvp;
986 		struct vnode ** a_vpp;
987 		struct componentname * a_cnp;
988 	} */ *ap = v;
989 	struct componentname *cnp = ap->a_cnp;
990 	struct vnode **vpp = ap->a_vpp;
991 	struct vnode *dvp = ap->a_dvp;
992 	const char *pname = cnp->cn_nameptr;
993 	const struct proc_target *pt = NULL;
994 	struct vnode *fvp;
995 	pid_t pid, vnpid;
996 	struct pfsnode *pfs;
997 	struct proc *p = NULL;
998 	struct lwp *plwp;
999 	int i, error;
1000 	pfstype type;
1001 
1002 	*vpp = NULL;
1003 
1004 	if (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)
1005 		return (EROFS);
1006 
1007 	if (cnp->cn_namelen == 1 && *pname == '.') {
1008 		*vpp = dvp;
1009 		vref(dvp);
1010 		return (0);
1011 	}
1012 
1013 	pfs = VTOPFS(dvp);
1014 	switch (pfs->pfs_type) {
1015 	case PFSroot:
1016 		/*
1017 		 * Shouldn't get here with .. in the root node.
1018 		 */
1019 		if (cnp->cn_flags & ISDOTDOT)
1020 			return (EIO);
1021 
1022 		for (i = 0; i < nproc_root_targets; i++) {
1023 			pt = &proc_root_targets[i];
1024 			/*
1025 			 * check for node match.  proc is always NULL here,
1026 			 * so call pt_valid with constant NULL lwp.
1027 			 */
1028 			if (cnp->cn_namelen == pt->pt_namlen &&
1029 			    memcmp(pt->pt_name, pname, cnp->cn_namelen) == 0 &&
1030 			    (pt->pt_valid == NULL ||
1031 			     (*pt->pt_valid)(NULL, dvp->v_mount)))
1032 				break;
1033 		}
1034 
1035 		if (i != nproc_root_targets) {
1036 			error = procfs_allocvp(dvp->v_mount, vpp, 0,
1037 			    pt->pt_pfstype, -1, NULL);
1038 			return (error);
1039 		}
1040 
1041 		if (CNEQ(cnp, "curproc", 7)) {
1042 			pid = curproc->p_pid;
1043 			vnpid = 0;
1044 			type = PFScurproc;
1045 		} else if (CNEQ(cnp, "self", 4)) {
1046 			pid = curproc->p_pid;
1047 			vnpid = 0;
1048 			type = PFSself;
1049 		} else {
1050 			pid = (pid_t)atoi(pname, cnp->cn_namelen);
1051 			vnpid = pid;
1052 			type = PFSproc;
1053 		}
1054 
1055 		if (procfs_proc_lock(pid, &p, ESRCH) != 0)
1056 			break;
1057 		error = procfs_allocvp(dvp->v_mount, vpp, vnpid, type, -1, p);
1058 		procfs_proc_unlock(p);
1059 		return (error);
1060 
1061 	case PFSproc:
1062 		/*
1063 		 * do the .. dance. We unlock the directory, and then
1064 		 * get the root dir. That will automatically return ..
1065 		 * locked. Then if the caller wanted dvp locked, we
1066 		 * re-lock.
1067 		 */
1068 		if (cnp->cn_flags & ISDOTDOT) {
1069 			VOP_UNLOCK(dvp);
1070 			error = procfs_root(dvp->v_mount, vpp);
1071 			vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1072 			return (error);
1073 		}
1074 
1075 		if (procfs_proc_lock(pfs->pfs_pid, &p, ESRCH) != 0)
1076 			break;
1077 
1078 		mutex_enter(p->p_lock);
1079 		LIST_FOREACH(plwp, &p->p_lwps, l_sibling) {
1080 			if (plwp->l_stat != LSZOMB)
1081 				break;
1082 		}
1083 		/* Process is exiting if no-LWPS or all LWPs are LSZOMB */
1084 		if (plwp == NULL) {
1085 			mutex_exit(p->p_lock);
1086 			procfs_proc_unlock(p);
1087 			return ESRCH;
1088 		}
1089 
1090 		lwp_addref(plwp);
1091 		mutex_exit(p->p_lock);
1092 
1093 		for (pt = proc_targets, i = 0; i < nproc_targets; pt++, i++) {
1094 			int found;
1095 
1096 			found = cnp->cn_namelen == pt->pt_namlen &&
1097 			    memcmp(pt->pt_name, pname, cnp->cn_namelen) == 0 &&
1098 			    (pt->pt_valid == NULL
1099 			      || (*pt->pt_valid)(plwp, dvp->v_mount));
1100 			if (found)
1101 				break;
1102 		}
1103 		lwp_delref(plwp);
1104 
1105 		if (i == nproc_targets) {
1106 			procfs_proc_unlock(p);
1107 			break;
1108 		}
1109 		if (pt->pt_pfstype == PFSfile) {
1110 			fvp = p->p_textvp;
1111 			/* We already checked that it exists. */
1112 			vref(fvp);
1113 			procfs_proc_unlock(p);
1114 			vn_lock(fvp, LK_EXCLUSIVE | LK_RETRY);
1115 			*vpp = fvp;
1116 			return (0);
1117 		}
1118 
1119 		error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
1120 		    pt->pt_pfstype, -1, p);
1121 		procfs_proc_unlock(p);
1122 		return (error);
1123 
1124 	case PFSfd: {
1125 		int fd;
1126 		file_t *fp;
1127 
1128 		if ((error = procfs_proc_lock(pfs->pfs_pid, &p, ENOENT)) != 0)
1129 			return error;
1130 
1131 		/*
1132 		 * do the .. dance. We unlock the directory, and then
1133 		 * get the proc dir. That will automatically return ..
1134 		 * locked. Then re-lock the directory.
1135 		 */
1136 		if (cnp->cn_flags & ISDOTDOT) {
1137 			VOP_UNLOCK(dvp);
1138 			error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
1139 			    PFSproc, -1, p);
1140 			procfs_proc_unlock(p);
1141 			vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1142 			return (error);
1143 		}
1144 		fd = atoi(pname, cnp->cn_namelen);
1145 
1146 		fp = fd_getfile2(p, fd);
1147 		if (fp == NULL) {
1148 			procfs_proc_unlock(p);
1149 			return ENOENT;
1150 		}
1151 		fvp = fp->f_data;
1152 
1153 		/* Don't show directories */
1154 		if (fp->f_type == DTYPE_VNODE && fvp->v_type != VDIR) {
1155 			vref(fvp);
1156 			closef(fp);
1157 			procfs_proc_unlock(p);
1158 			vn_lock(fvp, LK_EXCLUSIVE | LK_RETRY);
1159 			*vpp = fvp;
1160 			return 0;
1161 		}
1162 
1163 		closef(fp);
1164 		error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
1165 		    PFSfd, fd, p);
1166 		procfs_proc_unlock(p);
1167 		return error;
1168 	}
1169 	default:
1170 		return (ENOTDIR);
1171 	}
1172 
1173 	return (cnp->cn_nameiop == LOOKUP ? ENOENT : EROFS);
1174 }
1175 
1176 int
1177 procfs_validfile(struct lwp *l, struct mount *mp)
1178 {
1179 	return l != NULL && l->l_proc != NULL && l->l_proc->p_textvp != NULL;
1180 }
1181 
1182 static int
1183 procfs_validfile_linux(struct lwp *l, struct mount *mp)
1184 {
1185 	int flags;
1186 
1187 	flags = VFSTOPROC(mp)->pmnt_flags;
1188 	return (flags & PROCFSMNT_LINUXCOMPAT) &&
1189 	    (l == NULL || l->l_proc == NULL || procfs_validfile(l, mp));
1190 }
1191 
1192 struct procfs_root_readdir_ctx {
1193 	struct uio *uiop;
1194 	off_t *cookies;
1195 	int ncookies;
1196 	off_t off;
1197 	off_t startoff;
1198 	int error;
1199 };
1200 
1201 static int
1202 procfs_root_readdir_callback(struct proc *p, void *arg)
1203 {
1204 	struct procfs_root_readdir_ctx *ctxp = arg;
1205 	struct dirent d;
1206 	struct uio *uiop;
1207 	int error;
1208 
1209 	uiop = ctxp->uiop;
1210 	if (uiop->uio_resid < UIO_MX)
1211 		return -1; /* no space */
1212 
1213 	if (ctxp->off < ctxp->startoff) {
1214 		ctxp->off++;
1215 		return 0;
1216 	}
1217 
1218 	if (kauth_authorize_process(kauth_cred_get(),
1219 	    KAUTH_PROCESS_CANSEE, p,
1220 	    KAUTH_ARG(KAUTH_REQ_PROCESS_CANSEE_ENTRY), NULL, NULL) != 0)
1221 		return 0;
1222 
1223 	memset(&d, 0, UIO_MX);
1224 	d.d_reclen = UIO_MX;
1225 	d.d_fileno = PROCFS_FILENO(p->p_pid, PFSproc, -1);
1226 	d.d_namlen = snprintf(d.d_name,
1227 	    UIO_MX - offsetof(struct dirent, d_name), "%ld", (long)p->p_pid);
1228 	d.d_type = DT_DIR;
1229 
1230 	mutex_exit(proc_lock);
1231 	error = uiomove(&d, UIO_MX, uiop);
1232 	mutex_enter(proc_lock);
1233 	if (error) {
1234 		ctxp->error = error;
1235 		return -1;
1236 	}
1237 
1238 	ctxp->ncookies++;
1239 	if (ctxp->cookies)
1240 		*(ctxp->cookies)++ = ctxp->off + 1;
1241 	ctxp->off++;
1242 
1243 	return 0;
1244 }
1245 
1246 /*
1247  * readdir returns directory entries from pfsnode (vp).
1248  *
1249  * the strategy here with procfs is to generate a single
1250  * directory entry at a time (struct dirent) and then
1251  * copy that out to userland using uiomove.  a more efficent
1252  * though more complex implementation, would try to minimize
1253  * the number of calls to uiomove().  for procfs, this is
1254  * hardly worth the added code complexity.
1255  *
1256  * this should just be done through read()
1257  */
1258 int
1259 procfs_readdir(void *v)
1260 {
1261 	struct vop_readdir_args /* {
1262 		struct vnode *a_vp;
1263 		struct uio *a_uio;
1264 		kauth_cred_t a_cred;
1265 		int *a_eofflag;
1266 		off_t **a_cookies;
1267 		int *a_ncookies;
1268 	} */ *ap = v;
1269 	struct uio *uio = ap->a_uio;
1270 	struct dirent d;
1271 	struct pfsnode *pfs;
1272 	off_t i;
1273 	int error;
1274 	off_t *cookies = NULL;
1275 	int ncookies;
1276 	struct vnode *vp;
1277 	const struct proc_target *pt;
1278 	struct procfs_root_readdir_ctx ctx;
1279 	struct lwp *l;
1280 	int nfd;
1281 
1282 	vp = ap->a_vp;
1283 	pfs = VTOPFS(vp);
1284 
1285 	if (uio->uio_resid < UIO_MX)
1286 		return (EINVAL);
1287 	if (uio->uio_offset < 0)
1288 		return (EINVAL);
1289 
1290 	error = 0;
1291 	i = uio->uio_offset;
1292 	memset(&d, 0, UIO_MX);
1293 	d.d_reclen = UIO_MX;
1294 	ncookies = uio->uio_resid / UIO_MX;
1295 
1296 	switch (pfs->pfs_type) {
1297 	/*
1298 	 * this is for the process-specific sub-directories.
1299 	 * all that is needed to is copy out all the entries
1300 	 * from the procent[] table (top of this file).
1301 	 */
1302 	case PFSproc: {
1303 		struct proc *p;
1304 
1305 		if (i >= nproc_targets)
1306 			return 0;
1307 
1308 		if (procfs_proc_lock(pfs->pfs_pid, &p, ESRCH) != 0)
1309 			break;
1310 
1311 		if (ap->a_ncookies) {
1312 			ncookies = min(ncookies, (nproc_targets - i));
1313 			cookies = malloc(ncookies * sizeof (off_t),
1314 			    M_TEMP, M_WAITOK);
1315 			*ap->a_cookies = cookies;
1316 		}
1317 
1318 		for (pt = &proc_targets[i];
1319 		     uio->uio_resid >= UIO_MX && i < nproc_targets; pt++, i++) {
1320 			if (pt->pt_valid) {
1321 				/* XXXSMP LWP can disappear */
1322 				mutex_enter(p->p_lock);
1323 				l = LIST_FIRST(&p->p_lwps);
1324 				KASSERT(l != NULL);
1325 				mutex_exit(p->p_lock);
1326 				if ((*pt->pt_valid)(l, vp->v_mount) == 0)
1327 					continue;
1328 			}
1329 
1330 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid,
1331 			    pt->pt_pfstype, -1);
1332 			d.d_namlen = pt->pt_namlen;
1333 			memcpy(d.d_name, pt->pt_name, pt->pt_namlen + 1);
1334 			d.d_type = pt->pt_type;
1335 
1336 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1337 				break;
1338 			if (cookies)
1339 				*cookies++ = i + 1;
1340 		}
1341 
1342 		procfs_proc_unlock(p);
1343 	    	break;
1344 	}
1345 	case PFSfd: {
1346 		struct proc *p;
1347 		file_t *fp;
1348 		int lim, nc = 0;
1349 
1350 		if ((error = procfs_proc_lock(pfs->pfs_pid, &p, ESRCH)) != 0)
1351 			return error;
1352 
1353 		/* XXX Should this be by file as well? */
1354 		if (kauth_authorize_process(kauth_cred_get(),
1355 		    KAUTH_PROCESS_CANSEE, p,
1356 		    KAUTH_ARG(KAUTH_REQ_PROCESS_CANSEE_OPENFILES), NULL,
1357 		    NULL) != 0) {
1358 		    	procfs_proc_unlock(p);
1359 			return ESRCH;
1360 		}
1361 
1362 		nfd = p->p_fd->fd_dt->dt_nfiles;
1363 
1364 		lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfiles);
1365 		if (i >= lim) {
1366 		    	procfs_proc_unlock(p);
1367 			return 0;
1368 		}
1369 
1370 		if (ap->a_ncookies) {
1371 			ncookies = min(ncookies, (nfd + 2 - i));
1372 			cookies = malloc(ncookies * sizeof (off_t),
1373 			    M_TEMP, M_WAITOK);
1374 			*ap->a_cookies = cookies;
1375 		}
1376 
1377 		for (; i < 2 && uio->uio_resid >= UIO_MX; i++) {
1378 			pt = &proc_targets[i];
1379 			d.d_namlen = pt->pt_namlen;
1380 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid,
1381 			    pt->pt_pfstype, -1);
1382 			(void)memcpy(d.d_name, pt->pt_name, pt->pt_namlen + 1);
1383 			d.d_type = pt->pt_type;
1384 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1385 				break;
1386 			if (cookies)
1387 				*cookies++ = i + 1;
1388 			nc++;
1389 		}
1390 		if (error) {
1391 			ncookies = nc;
1392 			break;
1393 		}
1394 		for (; uio->uio_resid >= UIO_MX && i < nfd; i++) {
1395 			/* check the descriptor exists */
1396 			if ((fp = fd_getfile2(p, i - 2)) == NULL)
1397 				continue;
1398 			closef(fp);
1399 
1400 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid, PFSfd, i - 2);
1401 			d.d_namlen = snprintf(d.d_name, sizeof(d.d_name),
1402 			    "%lld", (long long)(i - 2));
1403 			d.d_type = VREG;
1404 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1405 				break;
1406 			if (cookies)
1407 				*cookies++ = i + 1;
1408 			nc++;
1409 		}
1410 		ncookies = nc;
1411 		procfs_proc_unlock(p);
1412 		break;
1413 	}
1414 
1415 	/*
1416 	 * this is for the root of the procfs filesystem
1417 	 * what is needed are special entries for "curproc"
1418 	 * and "self" followed by an entry for each process
1419 	 * on allproc.
1420 	 */
1421 
1422 	case PFSroot: {
1423 		int nc = 0;
1424 
1425 		if (ap->a_ncookies) {
1426 			/*
1427 			 * XXX Potentially allocating too much space here,
1428 			 * but I'm lazy. This loop needs some work.
1429 			 */
1430 			cookies = malloc(ncookies * sizeof (off_t),
1431 			    M_TEMP, M_WAITOK);
1432 			*ap->a_cookies = cookies;
1433 		}
1434 		error = 0;
1435 		/* 0 ... 3 are static entries. */
1436 		for (; i <= 3 && uio->uio_resid >= UIO_MX; i++) {
1437 			switch (i) {
1438 			case 0:		/* `.' */
1439 			case 1:		/* `..' */
1440 				d.d_fileno = PROCFS_FILENO(0, PFSroot, -1);
1441 				d.d_namlen = i + 1;
1442 				memcpy(d.d_name, "..", d.d_namlen);
1443 				d.d_name[i + 1] = '\0';
1444 				d.d_type = DT_DIR;
1445 				break;
1446 
1447 			case 2:
1448 				d.d_fileno = PROCFS_FILENO(0, PFScurproc, -1);
1449 				d.d_namlen = sizeof("curproc") - 1;
1450 				memcpy(d.d_name, "curproc", sizeof("curproc"));
1451 				d.d_type = DT_LNK;
1452 				break;
1453 
1454 			case 3:
1455 				d.d_fileno = PROCFS_FILENO(0, PFSself, -1);
1456 				d.d_namlen = sizeof("self") - 1;
1457 				memcpy(d.d_name, "self", sizeof("self"));
1458 				d.d_type = DT_LNK;
1459 				break;
1460 			}
1461 
1462 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1463 				break;
1464 			nc++;
1465 			if (cookies)
1466 				*cookies++ = i + 1;
1467 		}
1468 		/* 4 ... are process entries. */
1469 		ctx.uiop = uio;
1470 		ctx.error = 0;
1471 		ctx.off = 4;
1472 		ctx.startoff = i;
1473 		ctx.cookies = cookies;
1474 		ctx.ncookies = nc;
1475 		proclist_foreach_call(&allproc,
1476 		    procfs_root_readdir_callback, &ctx);
1477 		cookies = ctx.cookies;
1478 		nc = ctx.ncookies;
1479 		error = ctx.error;
1480 		if (error)
1481 			break;
1482 
1483 		/* misc entries. */
1484 		if (i < ctx.off)
1485 			i = ctx.off;
1486 		if (i >= ctx.off + nproc_root_targets)
1487 			break;
1488 		for (pt = &proc_root_targets[i - ctx.off];
1489 		    uio->uio_resid >= UIO_MX &&
1490 		    pt < &proc_root_targets[nproc_root_targets];
1491 		    pt++, i++) {
1492 			if (pt->pt_valid &&
1493 			    (*pt->pt_valid)(NULL, vp->v_mount) == 0)
1494 				continue;
1495 			d.d_fileno = PROCFS_FILENO(0, pt->pt_pfstype, -1);
1496 			d.d_namlen = pt->pt_namlen;
1497 			memcpy(d.d_name, pt->pt_name, pt->pt_namlen + 1);
1498 			d.d_type = pt->pt_type;
1499 
1500 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1501 				break;
1502 			nc++;
1503 			if (cookies)
1504 				*cookies++ = i + 1;
1505 		}
1506 
1507 		ncookies = nc;
1508 		break;
1509 	}
1510 
1511 	default:
1512 		error = ENOTDIR;
1513 		break;
1514 	}
1515 
1516 	if (ap->a_ncookies) {
1517 		if (error) {
1518 			if (cookies)
1519 				free(*ap->a_cookies, M_TEMP);
1520 			*ap->a_ncookies = 0;
1521 			*ap->a_cookies = NULL;
1522 		} else
1523 			*ap->a_ncookies = ncookies;
1524 	}
1525 	uio->uio_offset = i;
1526 	return (error);
1527 }
1528 
1529 /*
1530  * readlink reads the link of `curproc' and others
1531  */
1532 int
1533 procfs_readlink(void *v)
1534 {
1535 	struct vop_readlink_args *ap = v;
1536 	char bf[16];		/* should be enough */
1537 	char *bp = bf;
1538 	char *path = NULL;
1539 	int len = 0;
1540 	int error = 0;
1541 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
1542 	struct proc *pown;
1543 
1544 	if (pfs->pfs_fileno == PROCFS_FILENO(0, PFScurproc, -1))
1545 		len = snprintf(bf, sizeof(bf), "%ld", (long)curproc->p_pid);
1546 	else if (pfs->pfs_fileno == PROCFS_FILENO(0, PFSself, -1))
1547 		len = snprintf(bf, sizeof(bf), "%s", "curproc");
1548 	else if (pfs->pfs_fileno == PROCFS_FILENO(pfs->pfs_pid, PFScwd, -1) ||
1549 	    pfs->pfs_fileno == PROCFS_FILENO(pfs->pfs_pid, PFSchroot, -1) ||
1550 	    pfs->pfs_fileno == PROCFS_FILENO(pfs->pfs_pid, PFSexe, -1)) {
1551 		if ((error = procfs_proc_lock(pfs->pfs_pid, &pown, ESRCH)) != 0)
1552 			return error;
1553 		path = malloc(MAXPATHLEN + 4, M_TEMP, M_WAITOK|M_CANFAIL);
1554 		if (path == NULL) {
1555 			procfs_proc_unlock(pown);
1556 			return (ENOMEM);
1557 		}
1558 		bp = path + MAXPATHLEN;
1559 		*--bp = '\0';
1560 		procfs_dir(PROCFS_TYPE(pfs->pfs_fileno), curlwp, pown,
1561 		    &bp, path, MAXPATHLEN);
1562 		procfs_proc_unlock(pown);
1563 		len = strlen(bp);
1564 	} else {
1565 		file_t *fp;
1566 		struct vnode *vxp, *vp;
1567 
1568 		if ((error = procfs_proc_lock(pfs->pfs_pid, &pown, ESRCH)) != 0)
1569 			return error;
1570 
1571 		fp = fd_getfile2(pown, pfs->pfs_fd);
1572 		if (fp == NULL) {
1573 			procfs_proc_unlock(pown);
1574 			return EBADF;
1575 		}
1576 
1577 		switch (fp->f_type) {
1578 		case DTYPE_VNODE:
1579 			vxp = (struct vnode *)fp->f_data;
1580 			if (vxp->v_type != VDIR) {
1581 				error = EINVAL;
1582 				break;
1583 			}
1584 			if ((path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK))
1585 			    == NULL) {
1586 				error = ENOMEM;
1587 				break;
1588 			}
1589 			bp = path + MAXPATHLEN;
1590 			*--bp = '\0';
1591 
1592 			/*
1593 			 * XXX: kludge to avoid locking against ourselves
1594 			 * in getcwd()
1595 			 */
1596 			if (vxp->v_tag == VT_PROCFS) {
1597 				*--bp = '/';
1598 			} else {
1599 				rw_enter(&curproc->p_cwdi->cwdi_lock, RW_READER);
1600 				vp = curproc->p_cwdi->cwdi_rdir;
1601 				if (vp == NULL)
1602 					vp = rootvnode;
1603 				error = getcwd_common(vxp, vp, &bp, path,
1604 				    MAXPATHLEN / 2, 0, curlwp);
1605 				rw_exit(&curproc->p_cwdi->cwdi_lock);
1606 			}
1607 			if (error)
1608 				break;
1609 			len = strlen(bp);
1610 			break;
1611 
1612 		case DTYPE_MISC:
1613 			len = snprintf(bf, sizeof(bf), "%s", "[misc]");
1614 			break;
1615 
1616 		case DTYPE_KQUEUE:
1617 			len = snprintf(bf, sizeof(bf), "%s", "[kqueue]");
1618 			break;
1619 
1620 		default:
1621 			error = EINVAL;
1622 			break;
1623 		}
1624 		closef(fp);
1625 		procfs_proc_unlock(pown);
1626 	}
1627 
1628 	if (error == 0)
1629 		error = uiomove(bp, len, ap->a_uio);
1630 	if (path)
1631 		free(path, M_TEMP);
1632 	return error;
1633 }
1634 
1635 /*
1636  * convert decimal ascii to int
1637  */
1638 static int
1639 atoi(const char *b, size_t len)
1640 {
1641 	int p = 0;
1642 
1643 	while (len--) {
1644 		char c = *b++;
1645 		if (c < '0' || c > '9')
1646 			return -1;
1647 		p = 10 * p + (c - '0');
1648 	}
1649 
1650 	return p;
1651 }
1652