xref: /netbsd-src/sys/miscfs/procfs/procfs_vnops.c (revision 7fa608457b817eca6e0977b37f758ae064f3c99c)
1 /*	$NetBSD: procfs_vnops.c,v 1.162 2007/11/09 22:45:49 christos Exp $	*/
2 
3 /*-
4  * Copyright (c) 2006, 2007 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  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the NetBSD
21  *	Foundation, Inc. and its contributors.
22  * 4. Neither the name of The NetBSD Foundation nor the names of its
23  *    contributors may be used to endorse or promote products derived
24  *    from this software without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
27  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
28  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
29  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
30  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36  * POSSIBILITY OF SUCH DAMAGE.
37  */
38 
39 /*
40  * Copyright (c) 1993, 1995
41  *	The Regents of the University of California.  All rights reserved.
42  *
43  * This code is derived from software contributed to Berkeley by
44  * Jan-Simon Pendry.
45  *
46  * Redistribution and use in source and binary forms, with or without
47  * modification, are permitted provided that the following conditions
48  * are met:
49  * 1. Redistributions of source code must retain the above copyright
50  *    notice, this list of conditions and the following disclaimer.
51  * 2. Redistributions in binary form must reproduce the above copyright
52  *    notice, this list of conditions and the following disclaimer in the
53  *    documentation and/or other materials provided with the distribution.
54  * 3. Neither the name of the University nor the names of its contributors
55  *    may be used to endorse or promote products derived from this software
56  *    without specific prior written permission.
57  *
58  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
59  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
60  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
61  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
62  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
63  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
64  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
65  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
66  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
67  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
68  * SUCH DAMAGE.
69  *
70  *	@(#)procfs_vnops.c	8.18 (Berkeley) 5/21/95
71  */
72 
73 /*
74  * Copyright (c) 1993 Jan-Simon Pendry
75  *
76  * This code is derived from software contributed to Berkeley by
77  * Jan-Simon Pendry.
78  *
79  * Redistribution and use in source and binary forms, with or without
80  * modification, are permitted provided that the following conditions
81  * are met:
82  * 1. Redistributions of source code must retain the above copyright
83  *    notice, this list of conditions and the following disclaimer.
84  * 2. Redistributions in binary form must reproduce the above copyright
85  *    notice, this list of conditions and the following disclaimer in the
86  *    documentation and/or other materials provided with the distribution.
87  * 3. All advertising materials mentioning features or use of this software
88  *    must display the following acknowledgement:
89  *	This product includes software developed by the University of
90  *	California, Berkeley and its contributors.
91  * 4. Neither the name of the University nor the names of its contributors
92  *    may be used to endorse or promote products derived from this software
93  *    without specific prior written permission.
94  *
95  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
96  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
97  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
98  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
99  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
100  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
101  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
102  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
103  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
104  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
105  * SUCH DAMAGE.
106  *
107  *	@(#)procfs_vnops.c	8.18 (Berkeley) 5/21/95
108  */
109 
110 /*
111  * procfs vnode interface
112  */
113 
114 #include <sys/cdefs.h>
115 __KERNEL_RCSID(0, "$NetBSD: procfs_vnops.c,v 1.162 2007/11/09 22:45:49 christos Exp $");
116 
117 #include <sys/param.h>
118 #include <sys/systm.h>
119 #include <sys/time.h>
120 #include <sys/kernel.h>
121 #include <sys/file.h>
122 #include <sys/filedesc.h>
123 #include <sys/proc.h>
124 #include <sys/vnode.h>
125 #include <sys/namei.h>
126 #include <sys/malloc.h>
127 #include <sys/mount.h>
128 #include <sys/dirent.h>
129 #include <sys/resourcevar.h>
130 #include <sys/stat.h>
131 #include <sys/ptrace.h>
132 #include <sys/kauth.h>
133 
134 #include <uvm/uvm_extern.h>	/* for PAGE_SIZE */
135 
136 #include <machine/reg.h>
137 
138 #include <miscfs/genfs/genfs.h>
139 #include <miscfs/procfs/procfs.h>
140 
141 /*
142  * Vnode Operations.
143  *
144  */
145 
146 static int procfs_validfile_linux(struct lwp *, struct mount *);
147 static int procfs_root_readdir_callback(struct proc *, void *);
148 static void procfs_dir(pfstype, struct lwp *, struct proc *, char **, char *,
149     size_t);
150 
151 /*
152  * This is a list of the valid names in the
153  * process-specific sub-directories.  It is
154  * used in procfs_lookup and procfs_readdir
155  */
156 static const struct proc_target {
157 	u_char	pt_type;
158 	u_char	pt_namlen;
159 	const char	*pt_name;
160 	pfstype	pt_pfstype;
161 	int	(*pt_valid)(struct lwp *, struct mount *);
162 } proc_targets[] = {
163 #define N(s) sizeof(s)-1, s
164 	/*	  name		type		validp */
165 	{ DT_DIR, N("."),	PFSproc,	NULL },
166 	{ DT_DIR, N(".."),	PFSroot,	NULL },
167 	{ DT_DIR, N("fd"),	PFSfd,		NULL },
168 	{ DT_REG, N("file"),	PFSfile,	procfs_validfile },
169 	{ DT_REG, N("mem"),	PFSmem,		NULL },
170 	{ DT_REG, N("regs"),	PFSregs,	procfs_validregs },
171 	{ DT_REG, N("fpregs"),	PFSfpregs,	procfs_validfpregs },
172 	{ DT_REG, N("ctl"),	PFSctl,		NULL },
173 	{ DT_REG, N("stat"),	PFSstat,	procfs_validfile_linux },
174 	{ DT_REG, N("status"),	PFSstatus,	NULL },
175 	{ DT_REG, N("note"),	PFSnote,	NULL },
176 	{ DT_REG, N("notepg"),	PFSnotepg,	NULL },
177 	{ DT_REG, N("map"),	PFSmap,		procfs_validmap },
178 	{ DT_REG, N("maps"),	PFSmaps,	procfs_validmap },
179 	{ DT_REG, N("cmdline"), PFScmdline,	NULL },
180 	{ DT_REG, N("exe"),	PFSexe,		procfs_validfile },
181 	{ DT_LNK, N("cwd"),	PFScwd,		NULL },
182 	{ DT_LNK, N("root"),	PFSchroot,	NULL },
183 	{ DT_LNK, N("emul"),	PFSemul,	NULL },
184 	{ DT_REG, N("statm"),	PFSstatm,	procfs_validfile_linux },
185 #ifdef __HAVE_PROCFS_MACHDEP
186 	PROCFS_MACHDEP_NODETYPE_DEFNS
187 #endif
188 #undef N
189 };
190 static const int nproc_targets = sizeof(proc_targets) / sizeof(proc_targets[0]);
191 
192 /*
193  * List of files in the root directory. Note: the validate function will
194  * be called with p == NULL for these ones.
195  */
196 static const struct proc_target proc_root_targets[] = {
197 #define N(s) sizeof(s)-1, s
198 	/*	  name		    type	    validp */
199 	{ DT_REG, N("meminfo"),     PFSmeminfo,        procfs_validfile_linux },
200 	{ DT_REG, N("cpuinfo"),     PFScpuinfo,        procfs_validfile_linux },
201 	{ DT_REG, N("uptime"),      PFSuptime,         procfs_validfile_linux },
202 	{ DT_REG, N("mounts"),	    PFSmounts,	       procfs_validfile_linux },
203 	{ DT_REG, N("devices"),     PFSdevices,        procfs_validfile_linux },
204 	{ DT_REG, N("stat"),	    PFScpustat,        procfs_validfile_linux },
205 	{ DT_REG, N("loadavg"),	    PFSloadavg,        procfs_validfile_linux },
206 #undef N
207 };
208 static const int nproc_root_targets =
209     sizeof(proc_root_targets) / sizeof(proc_root_targets[0]);
210 
211 int	procfs_lookup(void *);
212 #define	procfs_create	genfs_eopnotsupp
213 #define	procfs_mknod	genfs_eopnotsupp
214 int	procfs_open(void *);
215 int	procfs_close(void *);
216 int	procfs_access(void *);
217 int	procfs_getattr(void *);
218 int	procfs_setattr(void *);
219 #define	procfs_read	procfs_rw
220 #define	procfs_write	procfs_rw
221 #define	procfs_fcntl	genfs_fcntl
222 #define	procfs_ioctl	genfs_enoioctl
223 #define	procfs_poll	genfs_poll
224 #define procfs_revoke	genfs_revoke
225 #define	procfs_fsync	genfs_nullop
226 #define	procfs_seek	genfs_nullop
227 #define	procfs_remove	genfs_eopnotsupp
228 int	procfs_link(void *);
229 #define	procfs_rename	genfs_eopnotsupp
230 #define	procfs_mkdir	genfs_eopnotsupp
231 #define	procfs_rmdir	genfs_eopnotsupp
232 int	procfs_symlink(void *);
233 int	procfs_readdir(void *);
234 int	procfs_readlink(void *);
235 #define	procfs_abortop	genfs_abortop
236 int	procfs_inactive(void *);
237 int	procfs_reclaim(void *);
238 #define	procfs_lock	genfs_lock
239 #define	procfs_unlock	genfs_unlock
240 #define	procfs_bmap	genfs_badop
241 #define	procfs_strategy	genfs_badop
242 int	procfs_print(void *);
243 int	procfs_pathconf(void *);
244 #define	procfs_islocked	genfs_islocked
245 #define	procfs_advlock	genfs_einval
246 #define	procfs_bwrite	genfs_eopnotsupp
247 #define procfs_putpages	genfs_null_putpages
248 
249 static int atoi(const char *, size_t);
250 
251 /*
252  * procfs vnode operations.
253  */
254 int (**procfs_vnodeop_p)(void *);
255 const struct vnodeopv_entry_desc procfs_vnodeop_entries[] = {
256 	{ &vop_default_desc, vn_default_error },
257 	{ &vop_lookup_desc, procfs_lookup },		/* lookup */
258 	{ &vop_create_desc, procfs_create },		/* create */
259 	{ &vop_mknod_desc, procfs_mknod },		/* mknod */
260 	{ &vop_open_desc, procfs_open },		/* open */
261 	{ &vop_close_desc, procfs_close },		/* close */
262 	{ &vop_access_desc, procfs_access },		/* access */
263 	{ &vop_getattr_desc, procfs_getattr },		/* getattr */
264 	{ &vop_setattr_desc, procfs_setattr },		/* setattr */
265 	{ &vop_read_desc, procfs_read },		/* read */
266 	{ &vop_write_desc, procfs_write },		/* write */
267 	{ &vop_fcntl_desc, procfs_fcntl },		/* fcntl */
268 	{ &vop_ioctl_desc, procfs_ioctl },		/* ioctl */
269 	{ &vop_poll_desc, procfs_poll },		/* poll */
270 	{ &vop_revoke_desc, procfs_revoke },		/* revoke */
271 	{ &vop_fsync_desc, procfs_fsync },		/* fsync */
272 	{ &vop_seek_desc, procfs_seek },		/* seek */
273 	{ &vop_remove_desc, procfs_remove },		/* remove */
274 	{ &vop_link_desc, procfs_link },		/* link */
275 	{ &vop_rename_desc, procfs_rename },		/* rename */
276 	{ &vop_mkdir_desc, procfs_mkdir },		/* mkdir */
277 	{ &vop_rmdir_desc, procfs_rmdir },		/* rmdir */
278 	{ &vop_symlink_desc, procfs_symlink },		/* symlink */
279 	{ &vop_readdir_desc, procfs_readdir },		/* readdir */
280 	{ &vop_readlink_desc, procfs_readlink },	/* readlink */
281 	{ &vop_abortop_desc, procfs_abortop },		/* abortop */
282 	{ &vop_inactive_desc, procfs_inactive },	/* inactive */
283 	{ &vop_reclaim_desc, procfs_reclaim },		/* reclaim */
284 	{ &vop_lock_desc, procfs_lock },		/* lock */
285 	{ &vop_unlock_desc, procfs_unlock },		/* unlock */
286 	{ &vop_bmap_desc, procfs_bmap },		/* bmap */
287 	{ &vop_strategy_desc, procfs_strategy },	/* strategy */
288 	{ &vop_print_desc, procfs_print },		/* print */
289 	{ &vop_islocked_desc, procfs_islocked },	/* islocked */
290 	{ &vop_pathconf_desc, procfs_pathconf },	/* pathconf */
291 	{ &vop_advlock_desc, procfs_advlock },		/* advlock */
292 	{ &vop_putpages_desc, procfs_putpages },	/* putpages */
293 	{ NULL, NULL }
294 };
295 const struct vnodeopv_desc procfs_vnodeop_opv_desc =
296 	{ &procfs_vnodeop_p, procfs_vnodeop_entries };
297 /*
298  * set things up for doing i/o on
299  * the pfsnode (vp).  (vp) is locked
300  * on entry, and should be left locked
301  * on exit.
302  *
303  * for procfs we don't need to do anything
304  * in particular for i/o.  all that is done
305  * is to support exclusive open on process
306  * memory images.
307  */
308 int
309 procfs_open(v)
310 	void *v;
311 {
312 	struct vop_open_args /* {
313 		struct vnode *a_vp;
314 		int  a_mode;
315 		kauth_cred_t a_cred;
316 		struct lwp *a_l;
317 	} */ *ap = v;
318 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
319 	struct lwp *l1;
320 	struct proc *p2;
321 	int error;
322 
323 	if ((error = procfs_proc_lock(pfs->pfs_pid, &p2, ENOENT)) != 0)
324 		return error;
325 
326 	l1 = ap->a_l;				/* tracer */
327 
328 #define	M2K(m)	(((m) & FREAD) && ((m) & FWRITE) ? \
329 		 KAUTH_REQ_PROCESS_CANPROCFS_RW : \
330 		 (m) & FWRITE ? KAUTH_REQ_PROCESS_CANPROCFS_WRITE : \
331 		 KAUTH_REQ_PROCESS_CANPROCFS_READ)
332 
333 	mutex_enter(&p2->p_mutex);
334 	error = kauth_authorize_process(l1->l_cred, KAUTH_PROCESS_CANPROCFS,
335 	    p2, pfs, KAUTH_ARG(M2K(ap->a_mode)), NULL);
336 	mutex_exit(&p2->p_mutex);
337 	if (error) {
338 		procfs_proc_unlock(p2);
339 		return (error);
340 	}
341 
342 #undef M2K
343 
344 	switch (pfs->pfs_type) {
345 	case PFSmem:
346 		if (((pfs->pfs_flags & FWRITE) && (ap->a_mode & O_EXCL)) ||
347 		    ((pfs->pfs_flags & O_EXCL) && (ap->a_mode & FWRITE))) {
348 			error = EBUSY;
349 			break;
350 		}
351 
352 		if (!proc_isunder(p2, l1)) {
353 			error = EPERM;
354 			break;
355 		}
356 
357 		if (ap->a_mode & FWRITE)
358 			pfs->pfs_flags = ap->a_mode & (FWRITE|O_EXCL);
359 
360 		break;
361 
362 	case PFSregs:
363 	case PFSfpregs:
364 		if (!proc_isunder(p2, l1)) {
365 			error = EPERM;
366 			break;
367 		}
368 		break;
369 
370 	default:
371 		break;
372 	}
373 
374 	procfs_proc_unlock(p2);
375 	return (error);
376 }
377 
378 /*
379  * close the pfsnode (vp) after doing i/o.
380  * (vp) is not locked on entry or exit.
381  *
382  * nothing to do for procfs other than undo
383  * any exclusive open flag (see _open above).
384  */
385 int
386 procfs_close(v)
387 	void *v;
388 {
389 	struct vop_close_args /* {
390 		struct vnode *a_vp;
391 		int  a_fflag;
392 		kauth_cred_t a_cred;
393 		struct lwp *a_l;
394 	} */ *ap = v;
395 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
396 
397 	switch (pfs->pfs_type) {
398 	case PFSmem:
399 		if ((ap->a_fflag & FWRITE) && (pfs->pfs_flags & O_EXCL))
400 			pfs->pfs_flags &= ~(FWRITE|O_EXCL);
401 		break;
402 
403 	default:
404 		break;
405 	}
406 
407 	return (0);
408 }
409 
410 /*
411  * _inactive is called when the pfsnode
412  * is vrele'd and the reference count goes
413  * to zero.  (vp) will be on the vnode free
414  * list, so to get it back vget() must be
415  * used.
416  *
417  * (vp) is locked on entry, but must be unlocked on exit.
418  */
419 int
420 procfs_inactive(v)
421 	void *v;
422 {
423 	struct vop_inactive_args /* {
424 		struct vnode *a_vp;
425 		bool *a_recycle;
426 	} */ *ap = v;
427 	struct vnode *vp = ap->a_vp;
428 	struct pfsnode *pfs = VTOPFS(vp);
429 	bool recycle;
430 
431 	mutex_enter(&proclist_lock);
432 	recycle = (p_find(pfs->pfs_pid, PFIND_LOCKED) == NULL);
433 	mutex_exit(&proclist_lock);
434 
435 	VOP_UNLOCK(vp, 0);
436 
437 	if (recycle)
438 		vgone(vp);
439 
440 	return (0);
441 }
442 
443 /*
444  * _reclaim is called when getnewvnode()
445  * wants to make use of an entry on the vnode
446  * free list.  at this time the filesystem needs
447  * to free any private data and remove the node
448  * from any private lists.
449  */
450 int
451 procfs_reclaim(v)
452 	void *v;
453 {
454 	struct vop_reclaim_args /* {
455 		struct vnode *a_vp;
456 	} */ *ap = v;
457 
458 	return (procfs_freevp(ap->a_vp));
459 }
460 
461 /*
462  * Return POSIX pathconf information applicable to special devices.
463  */
464 int
465 procfs_pathconf(v)
466 	void *v;
467 {
468 	struct vop_pathconf_args /* {
469 		struct vnode *a_vp;
470 		int a_name;
471 		register_t *a_retval;
472 	} */ *ap = v;
473 
474 	switch (ap->a_name) {
475 	case _PC_LINK_MAX:
476 		*ap->a_retval = LINK_MAX;
477 		return (0);
478 	case _PC_MAX_CANON:
479 		*ap->a_retval = MAX_CANON;
480 		return (0);
481 	case _PC_MAX_INPUT:
482 		*ap->a_retval = MAX_INPUT;
483 		return (0);
484 	case _PC_PIPE_BUF:
485 		*ap->a_retval = PIPE_BUF;
486 		return (0);
487 	case _PC_CHOWN_RESTRICTED:
488 		*ap->a_retval = 1;
489 		return (0);
490 	case _PC_VDISABLE:
491 		*ap->a_retval = _POSIX_VDISABLE;
492 		return (0);
493 	case _PC_SYNC_IO:
494 		*ap->a_retval = 1;
495 		return (0);
496 	default:
497 		return (EINVAL);
498 	}
499 	/* NOTREACHED */
500 }
501 
502 /*
503  * _print is used for debugging.
504  * just print a readable description
505  * of (vp).
506  */
507 int
508 procfs_print(v)
509 	void *v;
510 {
511 	struct vop_print_args /* {
512 		struct vnode *a_vp;
513 	} */ *ap = v;
514 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
515 
516 	printf("tag VT_PROCFS, type %d, pid %d, mode %x, flags %lx\n",
517 	    pfs->pfs_type, pfs->pfs_pid, pfs->pfs_mode, pfs->pfs_flags);
518 	return 0;
519 }
520 
521 int
522 procfs_link(v)
523 	void *v;
524 {
525 	struct vop_link_args /* {
526 		struct vnode *a_dvp;
527 		struct vnode *a_vp;
528 		struct componentname *a_cnp;
529 	} */ *ap = v;
530 
531 	VOP_ABORTOP(ap->a_dvp, ap->a_cnp);
532 	vput(ap->a_dvp);
533 	return (EROFS);
534 }
535 
536 int
537 procfs_symlink(v)
538 	void *v;
539 {
540 	struct vop_symlink_args /* {
541 		struct vnode *a_dvp;
542 		struct vnode **a_vpp;
543 		struct componentname *a_cnp;
544 		struct vattr *a_vap;
545 		char *a_target;
546 	} */ *ap = v;
547 
548 	VOP_ABORTOP(ap->a_dvp, ap->a_cnp);
549 	vput(ap->a_dvp);
550 	return (EROFS);
551 }
552 
553 /*
554  * Works out the path to (and vnode of) the target process's current
555  * working directory or chroot.  If the caller is in a chroot and
556  * can't "reach" the target's cwd or root (or some other error
557  * occurs), a "/" is returned for the path and a NULL pointer is
558  * returned for the vnode.
559  */
560 static void
561 procfs_dir(pfstype t, struct lwp *caller, struct proc *target, char **bpp,
562     char *path, size_t len)
563 {
564 	struct cwdinfo *cwdi;
565 	struct vnode *vp, *rvp;
566 	char *bp;
567 
568 	cwdi = caller->l_proc->p_cwdi;
569 	rw_enter(&cwdi->cwdi_lock, RW_READER);
570 
571 	rvp = cwdi->cwdi_rdir;
572 	bp = bpp ? *bpp : NULL;
573 
574 	switch (t) {
575 	case PFScwd:
576 		vp = target->p_cwdi->cwdi_cdir;
577 		break;
578 	case PFSchroot:
579 		vp = target->p_cwdi->cwdi_rdir;
580 		break;
581 	case PFSexe:
582 		vp = target->p_textvp;
583 		break;
584 	default:
585 		rw_exit(&cwdi->cwdi_lock);
586 		return;
587 	}
588 
589 	/*
590 	 * XXX: this horrible kludge avoids locking panics when
591 	 * attempting to lookup links that point to within procfs
592 	 */
593 	if (vp != NULL && vp->v_tag == VT_PROCFS) {
594 		if (bpp) {
595 			*--bp = '/';
596 			*bpp = bp;
597 		}
598 		rw_exit(&cwdi->cwdi_lock);
599 		return;
600 	}
601 
602 	if (rvp == NULL)
603 		rvp = rootvnode;
604 	if (vp == NULL || getcwd_common(vp, rvp, bp ? &bp : NULL, path,
605 	    len / 2, 0, caller) != 0) {
606 		vp = NULL;
607 		if (bpp) {
608 /*
609 			if (t == PFSexe) {
610 				snprintf(path, len, "%s/%d/file"
611 				    mp->mnt_stat.f_mntonname, pfs->pfs_pid);
612 			} else */ {
613 				bp = *bpp;
614 				*--bp = '/';
615 			}
616 		}
617 	}
618 
619 	if (bpp)
620 		*bpp = bp;
621 
622 	rw_exit(&cwdi->cwdi_lock);
623 }
624 
625 /*
626  * Invent attributes for pfsnode (vp) and store
627  * them in (vap).
628  * Directories lengths are returned as zero since
629  * any real length would require the genuine size
630  * to be computed, and nothing cares anyway.
631  *
632  * this is relatively minimal for procfs.
633  */
634 int
635 procfs_getattr(v)
636 	void *v;
637 {
638 	struct vop_getattr_args /* {
639 		struct vnode *a_vp;
640 		struct vattr *a_vap;
641 		kauth_cred_t a_cred;
642 		struct lwp *a_l;
643 	} */ *ap = v;
644 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
645 	struct vattr *vap = ap->a_vap;
646 	struct proc *procp;
647 	char *path;
648 	int error;
649 
650 	/* first check the process still exists */
651 	switch (pfs->pfs_type) {
652 	case PFSroot:
653 	case PFScurproc:
654 	case PFSself:
655 		procp = NULL;
656 		break;
657 
658 	default:
659 		error = procfs_proc_lock(pfs->pfs_pid, &procp, ENOENT);
660 		if (error != 0)
661 			return (error);
662 		break;
663 	}
664 
665 	switch (pfs->pfs_type) {
666 	case PFScwd:
667 	case PFSchroot:
668 	case PFSexe:
669 		MALLOC(path, char *, MAXPATHLEN + 4, M_TEMP,
670 		    M_WAITOK|M_CANFAIL);
671 		if (path == NULL && procp != NULL) {
672 			procfs_proc_unlock(procp);
673 			return (ENOMEM);
674 		}
675 		break;
676 
677 	default:
678 		path = NULL;
679 		break;
680 	}
681 
682 	if (procp != NULL) {
683 		mutex_enter(&procp->p_mutex);
684 		error = kauth_authorize_process(kauth_cred_get(),
685 		    KAUTH_PROCESS_CANSEE, procp, NULL, NULL, NULL);
686 		mutex_exit(&procp->p_mutex);
687 		if (error != 0) {
688 		    	procfs_proc_unlock(procp);
689 		    	if (path != NULL)
690 		    		free(path, M_TEMP);
691 			return (ENOENT);
692 		}
693 	}
694 
695 	error = 0;
696 
697 	/* start by zeroing out the attributes */
698 	VATTR_NULL(vap);
699 
700 	/* next do all the common fields */
701 	vap->va_type = ap->a_vp->v_type;
702 	vap->va_mode = pfs->pfs_mode;
703 	vap->va_fileid = pfs->pfs_fileno;
704 	vap->va_flags = 0;
705 	vap->va_blocksize = PAGE_SIZE;
706 
707 	/*
708 	 * Make all times be current TOD.
709 	 *
710 	 * It would be possible to get the process start
711 	 * time from the p_stats structure, but there's
712 	 * no "file creation" time stamp anyway, and the
713 	 * p_stats structure is not addressable if u. gets
714 	 * swapped out for that process.
715 	 */
716 	getnanotime(&vap->va_ctime);
717 	vap->va_atime = vap->va_mtime = vap->va_ctime;
718 	if (procp)
719 		TIMEVAL_TO_TIMESPEC(&procp->p_stats->p_start,
720 		    &vap->va_birthtime);
721 	else
722 		getnanotime(&vap->va_birthtime);
723 
724 	switch (pfs->pfs_type) {
725 	case PFSmem:
726 	case PFSregs:
727 	case PFSfpregs:
728 #if defined(__HAVE_PROCFS_MACHDEP) && defined(PROCFS_MACHDEP_PROTECT_CASES)
729 	PROCFS_MACHDEP_PROTECT_CASES
730 #endif
731 		/*
732 		 * If the process has exercised some setuid or setgid
733 		 * privilege, then rip away read/write permission so
734 		 * that only root can gain access.
735 		 */
736 		if (procp->p_flag & PK_SUGID)
737 			vap->va_mode &= ~(S_IRUSR|S_IWUSR);
738 		/* FALLTHROUGH */
739 	case PFSctl:
740 	case PFSstatus:
741 	case PFSstat:
742 	case PFSnote:
743 	case PFSnotepg:
744 	case PFSmap:
745 	case PFSmaps:
746 	case PFScmdline:
747 	case PFSemul:
748 	case PFSstatm:
749 		vap->va_nlink = 1;
750 		vap->va_uid = kauth_cred_geteuid(procp->p_cred);
751 		vap->va_gid = kauth_cred_getegid(procp->p_cred);
752 		break;
753 	case PFSmeminfo:
754 	case PFSdevices:
755 	case PFScpuinfo:
756 	case PFSuptime:
757 	case PFSmounts:
758 	case PFScpustat:
759 	case PFSloadavg:
760 		vap->va_nlink = 1;
761 		vap->va_uid = vap->va_gid = 0;
762 		break;
763 
764 	default:
765 		break;
766 	}
767 
768 	/*
769 	 * now do the object specific fields
770 	 *
771 	 * The size could be set from struct reg, but it's hardly
772 	 * worth the trouble, and it puts some (potentially) machine
773 	 * dependent data into this machine-independent code.  If it
774 	 * becomes important then this function should break out into
775 	 * a per-file stat function in the corresponding .c file.
776 	 */
777 
778 	switch (pfs->pfs_type) {
779 	case PFSroot:
780 		/*
781 		 * Set nlink to 1 to tell fts(3) we don't actually know.
782 		 */
783 		vap->va_nlink = 1;
784 		vap->va_uid = 0;
785 		vap->va_gid = 0;
786 		vap->va_bytes = vap->va_size = DEV_BSIZE;
787 		break;
788 
789 	case PFSself:
790 	case PFScurproc: {
791 		char bf[16];		/* should be enough */
792 		vap->va_nlink = 1;
793 		vap->va_uid = 0;
794 		vap->va_gid = 0;
795 		vap->va_bytes = vap->va_size =
796 		    snprintf(bf, sizeof(bf), "%ld", (long)curproc->p_pid);
797 		break;
798 	}
799 
800 	case PFSfd:
801 		if (pfs->pfs_fd != -1) {
802 			struct file *fp;
803 
804 			fp = fd_getfile(procp->p_fd, pfs->pfs_fd);
805 			if (fp == NULL) {
806 				error = EBADF;
807 				break;
808 			}
809 			FILE_USE(fp);
810 			vap->va_nlink = 1;
811 			vap->va_uid = kauth_cred_geteuid(fp->f_cred);
812 			vap->va_gid = kauth_cred_getegid(fp->f_cred);
813 			switch (fp->f_type) {
814 			case DTYPE_VNODE:
815 				vap->va_bytes = vap->va_size =
816 				    ((struct vnode *)fp->f_data)->v_size;
817 				break;
818 			default:
819 				vap->va_bytes = vap->va_size = 0;
820 				break;
821 			}
822 			FILE_UNUSE(fp, curlwp);
823 			break;
824 		}
825 		/*FALLTHROUGH*/
826 	case PFSproc:
827 		vap->va_nlink = 2;
828 		vap->va_uid = kauth_cred_geteuid(procp->p_cred);
829 		vap->va_gid = kauth_cred_getegid(procp->p_cred);
830 		vap->va_bytes = vap->va_size = DEV_BSIZE;
831 		break;
832 
833 	case PFSfile:
834 		error = EOPNOTSUPP;
835 		break;
836 
837 	case PFSmem:
838 		vap->va_bytes = vap->va_size =
839 			ctob(procp->p_vmspace->vm_tsize +
840 				    procp->p_vmspace->vm_dsize +
841 				    procp->p_vmspace->vm_ssize);
842 		break;
843 
844 #if defined(PT_GETREGS) || defined(PT_SETREGS)
845 	case PFSregs:
846 		vap->va_bytes = vap->va_size = sizeof(struct reg);
847 		break;
848 #endif
849 
850 #if defined(PT_GETFPREGS) || defined(PT_SETFPREGS)
851 	case PFSfpregs:
852 		vap->va_bytes = vap->va_size = sizeof(struct fpreg);
853 		break;
854 #endif
855 
856 	case PFSctl:
857 	case PFSstatus:
858 	case PFSstat:
859 	case PFSnote:
860 	case PFSnotepg:
861 	case PFScmdline:
862 	case PFSmeminfo:
863 	case PFSdevices:
864 	case PFScpuinfo:
865 	case PFSuptime:
866 	case PFSmounts:
867 	case PFScpustat:
868 	case PFSloadavg:
869 	case PFSstatm:
870 		vap->va_bytes = vap->va_size = 0;
871 		break;
872 	case PFSmap:
873 	case PFSmaps:
874 		/*
875 		 * Advise a larger blocksize for the map files, so that
876 		 * they may be read in one pass.
877 		 */
878 		vap->va_blocksize = 4 * PAGE_SIZE;
879 		vap->va_bytes = vap->va_size = 0;
880 		break;
881 
882 	case PFScwd:
883 	case PFSchroot:
884 	case PFSexe: {
885 		char *bp;
886 
887 		vap->va_nlink = 1;
888 		vap->va_uid = 0;
889 		vap->va_gid = 0;
890 		bp = path + MAXPATHLEN;
891 		*--bp = '\0';
892 		procfs_dir(pfs->pfs_type, curlwp, procp, &bp, path,
893 		     MAXPATHLEN);
894 		vap->va_bytes = vap->va_size = strlen(bp);
895 		break;
896 	}
897 
898 	case PFSemul:
899 		vap->va_bytes = vap->va_size = strlen(procp->p_emul->e_name);
900 		break;
901 
902 #ifdef __HAVE_PROCFS_MACHDEP
903 	PROCFS_MACHDEP_NODETYPE_CASES
904 		error = procfs_machdep_getattr(ap->a_vp, vap, procp);
905 		break;
906 #endif
907 
908 	default:
909 		panic("procfs_getattr");
910 	}
911 
912 	if (procp != NULL)
913 		procfs_proc_unlock(procp);
914 	if (path != NULL)
915 		free(path, M_TEMP);
916 
917 	return (error);
918 }
919 
920 /*ARGSUSED*/
921 int
922 procfs_setattr(void *v)
923 {
924 	/*
925 	 * just fake out attribute setting
926 	 * it's not good to generate an error
927 	 * return, otherwise things like creat()
928 	 * will fail when they try to set the
929 	 * file length to 0.  worse, this means
930 	 * that echo $note > /proc/$pid/note will fail.
931 	 */
932 
933 	return (0);
934 }
935 
936 /*
937  * implement access checking.
938  *
939  * actually, the check for super-user is slightly
940  * broken since it will allow read access to write-only
941  * objects.  this doesn't cause any particular trouble
942  * but does mean that the i/o entry points need to check
943  * that the operation really does make sense.
944  */
945 int
946 procfs_access(v)
947 	void *v;
948 {
949 	struct vop_access_args /* {
950 		struct vnode *a_vp;
951 		int a_mode;
952 		kauth_cred_t a_cred;
953 		struct lwp *a_l;
954 	} */ *ap = v;
955 	struct vattr va;
956 	int error;
957 
958 	if ((error = VOP_GETATTR(ap->a_vp, &va, ap->a_cred, ap->a_l)) != 0)
959 		return (error);
960 
961 	return (vaccess(va.va_type, va.va_mode,
962 	    va.va_uid, va.va_gid, ap->a_mode, ap->a_cred));
963 }
964 
965 /*
966  * lookup.  this is incredibly complicated in the
967  * general case, however for most pseudo-filesystems
968  * very little needs to be done.
969  *
970  * Locking isn't hard here, just poorly documented.
971  *
972  * If we're looking up ".", just vref the parent & return it.
973  *
974  * If we're looking up "..", unlock the parent, and lock "..". If everything
975  * went ok, and we're on the last component and the caller requested the
976  * parent locked, try to re-lock the parent. We do this to prevent lock
977  * races.
978  *
979  * For anything else, get the needed node. Then unlock the parent if not
980  * the last component or not LOCKPARENT (i.e. if we wouldn't re-lock the
981  * parent in the .. case).
982  *
983  * We try to exit with the parent locked in error cases.
984  */
985 int
986 procfs_lookup(v)
987 	void *v;
988 {
989 	struct vop_lookup_args /* {
990 		struct vnode * a_dvp;
991 		struct vnode ** a_vpp;
992 		struct componentname * a_cnp;
993 	} */ *ap = v;
994 	struct componentname *cnp = ap->a_cnp;
995 	struct vnode **vpp = ap->a_vpp;
996 	struct vnode *dvp = ap->a_dvp;
997 	const char *pname = cnp->cn_nameptr;
998 	const struct proc_target *pt = NULL;
999 	struct vnode *fvp;
1000 	pid_t pid, vnpid;
1001 	struct pfsnode *pfs;
1002 	struct proc *p = NULL;
1003 	struct lwp *l = NULL;
1004 	int i, error;
1005 	pfstype type;
1006 
1007 	*vpp = NULL;
1008 
1009 	if (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)
1010 		return (EROFS);
1011 
1012 	if (cnp->cn_namelen == 1 && *pname == '.') {
1013 		*vpp = dvp;
1014 		VREF(dvp);
1015 		return (0);
1016 	}
1017 
1018 	pfs = VTOPFS(dvp);
1019 	switch (pfs->pfs_type) {
1020 	case PFSroot:
1021 		/*
1022 		 * Shouldn't get here with .. in the root node.
1023 		 */
1024 		if (cnp->cn_flags & ISDOTDOT)
1025 			return (EIO);
1026 
1027 		for (i = 0; i < nproc_root_targets; i++) {
1028 			pt = &proc_root_targets[i];
1029 			/*
1030 			 * check for node match.  proc is always NULL here,
1031 			 * so call pt_valid with constant NULL lwp.
1032 			 */
1033 			if (cnp->cn_namelen == pt->pt_namlen &&
1034 			    memcmp(pt->pt_name, pname, cnp->cn_namelen) == 0 &&
1035 			    (pt->pt_valid == NULL ||
1036 			     (*pt->pt_valid)(NULL, dvp->v_mount)))
1037 				break;
1038 		}
1039 
1040 		if (i != nproc_root_targets) {
1041 			error = procfs_allocvp(dvp->v_mount, vpp, 0,
1042 			    pt->pt_pfstype, -1, NULL);
1043 			return (error);
1044 		}
1045 
1046 		if (CNEQ(cnp, "curproc", 7)) {
1047 			pid = curproc->p_pid;
1048 			vnpid = 0;
1049 			type = PFScurproc;
1050 		} else if (CNEQ(cnp, "self", 4)) {
1051 			pid = curproc->p_pid;
1052 			vnpid = 0;
1053 			type = PFSself;
1054 		} else {
1055 			pid = (pid_t)atoi(pname, cnp->cn_namelen);
1056 			vnpid = pid;
1057 			type = PFSproc;
1058 		}
1059 
1060 		if (procfs_proc_lock(pid, &p, ESRCH) != 0)
1061 			break;
1062 		error = procfs_allocvp(dvp->v_mount, vpp, vnpid, type, -1, p);
1063 		procfs_proc_unlock(p);
1064 		return (error);
1065 
1066 	case PFSproc:
1067 		/*
1068 		 * do the .. dance. We unlock the directory, and then
1069 		 * get the root dir. That will automatically return ..
1070 		 * locked. Then if the caller wanted dvp locked, we
1071 		 * re-lock.
1072 		 */
1073 		if (cnp->cn_flags & ISDOTDOT) {
1074 			VOP_UNLOCK(dvp, 0);
1075 			error = procfs_root(dvp->v_mount, vpp);
1076 			vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1077 			return (error);
1078 		}
1079 
1080 		if (procfs_proc_lock(pfs->pfs_pid, &p, ESRCH) != 0)
1081 			break;
1082 
1083 		for (pt = proc_targets, i = 0; i < nproc_targets; pt++, i++) {
1084 			struct lwp *plwp;
1085 			int found;
1086 
1087 			mutex_enter(&p->p_smutex);
1088 			plwp = proc_representative_lwp(p, NULL, 1);
1089 			lwp_addref(plwp);
1090 			mutex_exit(&p->p_smutex);
1091 			found = cnp->cn_namelen == pt->pt_namlen &&
1092 			    memcmp(pt->pt_name, pname, cnp->cn_namelen) == 0 &&
1093 			    (pt->pt_valid == NULL
1094 			      || (*pt->pt_valid)(plwp, dvp->v_mount));
1095 			lwp_delref(plwp);
1096 			if (found)
1097 				break;
1098 		}
1099 		if (i == nproc_targets) {
1100 			procfs_proc_unlock(p);
1101 			break;
1102 		}
1103 		if (pt->pt_pfstype == PFSfile) {
1104 			fvp = p->p_textvp;
1105 			/* We already checked that it exists. */
1106 			VREF(fvp);
1107 			procfs_proc_unlock(p);
1108 			vn_lock(fvp, LK_EXCLUSIVE | LK_RETRY);
1109 			*vpp = fvp;
1110 			return (0);
1111 		}
1112 
1113 		error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
1114 		    pt->pt_pfstype, -1, p);
1115 		procfs_proc_unlock(p);
1116 		return (error);
1117 
1118 	case PFSfd: {
1119 		int fd;
1120 		struct file *fp;
1121 
1122 		if ((error = procfs_proc_lock(pfs->pfs_pid, &p, ENOENT)) != 0)
1123 			return error;
1124 
1125 		/*
1126 		 * do the .. dance. We unlock the directory, and then
1127 		 * get the proc dir. That will automatically return ..
1128 		 * locked. Then re-lock the directory.
1129 		 */
1130 		if (cnp->cn_flags & ISDOTDOT) {
1131 			VOP_UNLOCK(dvp, 0);
1132 			error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
1133 			    PFSproc, -1, p);
1134 			procfs_proc_unlock(p);
1135 			vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1136 			return (error);
1137 		}
1138 		fd = atoi(pname, cnp->cn_namelen);
1139 
1140 		fp = fd_getfile(p->p_fd, fd);
1141 		if (fp == NULL) {
1142 			procfs_proc_unlock(p);
1143 			return ENOENT;
1144 		}
1145 
1146 		FILE_USE(fp);
1147 		fvp = (struct vnode *)fp->f_data;
1148 
1149 		/* Don't show directories */
1150 		if (fp->f_type == DTYPE_VNODE && fvp->v_type != VDIR) {
1151 			VREF(fvp);
1152 			FILE_UNUSE(fp, l);
1153 			procfs_proc_unlock(p);
1154 			vn_lock(fvp, LK_EXCLUSIVE | LK_RETRY |
1155 			    (p == curproc ? LK_CANRECURSE : 0));
1156 			*vpp = fvp;
1157 			return 0;
1158 		}
1159 
1160 		FILE_UNUSE(fp, l);
1161 		error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
1162 		    PFSfd, fd, p);
1163 		procfs_proc_unlock(p);
1164 		return error;
1165 	}
1166 	default:
1167 		return (ENOTDIR);
1168 	}
1169 
1170 	return (cnp->cn_nameiop == LOOKUP ? ENOENT : EROFS);
1171 }
1172 
1173 int
1174 procfs_validfile(struct lwp *l, struct mount *mp)
1175 {
1176 	return l != NULL && l->l_proc != NULL && l->l_proc->p_textvp != NULL;
1177 }
1178 
1179 static int
1180 procfs_validfile_linux(l, mp)
1181 	struct lwp *l;
1182 	struct mount *mp;
1183 {
1184 	int flags;
1185 
1186 	flags = VFSTOPROC(mp)->pmnt_flags;
1187 	return (flags & PROCFSMNT_LINUXCOMPAT) &&
1188 	    (l == NULL || l->l_proc == NULL || procfs_validfile(l, mp));
1189 }
1190 
1191 struct procfs_root_readdir_ctx {
1192 	struct uio *uiop;
1193 	off_t *cookies;
1194 	int ncookies;
1195 	off_t off;
1196 	off_t startoff;
1197 	int error;
1198 };
1199 
1200 static int
1201 procfs_root_readdir_callback(struct proc *p, void *arg)
1202 {
1203 	struct procfs_root_readdir_ctx *ctxp = arg;
1204 	struct dirent d;
1205 	struct uio *uiop;
1206 	int error;
1207 
1208 	uiop = ctxp->uiop;
1209 	if (uiop->uio_resid < UIO_MX)
1210 		return -1; /* no space */
1211 
1212 	if (ctxp->off < ctxp->startoff) {
1213 		ctxp->off++;
1214 		return 0;
1215 	}
1216 
1217 	if (kauth_authorize_process(kauth_cred_get(),
1218 	    KAUTH_PROCESS_CANSEE, p, NULL, NULL, NULL) != 0)
1219 		return 0;
1220 
1221 	memset(&d, 0, UIO_MX);
1222 	d.d_reclen = UIO_MX;
1223 	d.d_fileno = PROCFS_FILENO(p->p_pid, PFSproc, -1);
1224 	d.d_namlen = snprintf(d.d_name,
1225 	    UIO_MX - offsetof(struct dirent, d_name), "%ld", (long)p->p_pid);
1226 	d.d_type = DT_DIR;
1227 
1228 	mutex_exit(&proclist_lock);
1229 	error = uiomove(&d, UIO_MX, uiop);
1230 	mutex_enter(&proclist_lock);
1231 	if (error) {
1232 		ctxp->error = error;
1233 		return -1;
1234 	}
1235 
1236 	ctxp->ncookies++;
1237 	if (ctxp->cookies)
1238 		*(ctxp->cookies)++ = ctxp->off + 1;
1239 	ctxp->off++;
1240 
1241 	return 0;
1242 }
1243 
1244 /*
1245  * readdir returns directory entries from pfsnode (vp).
1246  *
1247  * the strategy here with procfs is to generate a single
1248  * directory entry at a time (struct dirent) and then
1249  * copy that out to userland using uiomove.  a more efficent
1250  * though more complex implementation, would try to minimize
1251  * the number of calls to uiomove().  for procfs, this is
1252  * hardly worth the added code complexity.
1253  *
1254  * this should just be done through read()
1255  */
1256 int
1257 procfs_readdir(v)
1258 	void *v;
1259 {
1260 	struct vop_readdir_args /* {
1261 		struct vnode *a_vp;
1262 		struct uio *a_uio;
1263 		kauth_cred_t a_cred;
1264 		int *a_eofflag;
1265 		off_t **a_cookies;
1266 		int *a_ncookies;
1267 	} */ *ap = v;
1268 	struct uio *uio = ap->a_uio;
1269 	struct dirent d;
1270 	struct pfsnode *pfs;
1271 	off_t i;
1272 	int error;
1273 	off_t *cookies = NULL;
1274 	int ncookies;
1275 	struct vnode *vp;
1276 	const struct proc_target *pt;
1277 	struct procfs_root_readdir_ctx ctx;
1278 	struct lwp *l;
1279 	int nfd;
1280 
1281 	vp = ap->a_vp;
1282 	pfs = VTOPFS(vp);
1283 
1284 	if (uio->uio_resid < UIO_MX)
1285 		return (EINVAL);
1286 	if (uio->uio_offset < 0)
1287 		return (EINVAL);
1288 
1289 	error = 0;
1290 	i = uio->uio_offset;
1291 	memset(&d, 0, UIO_MX);
1292 	d.d_reclen = UIO_MX;
1293 	ncookies = uio->uio_resid / UIO_MX;
1294 
1295 	switch (pfs->pfs_type) {
1296 	/*
1297 	 * this is for the process-specific sub-directories.
1298 	 * all that is needed to is copy out all the entries
1299 	 * from the procent[] table (top of this file).
1300 	 */
1301 	case PFSproc: {
1302 		struct proc *p;
1303 
1304 		if (i >= nproc_targets)
1305 			return 0;
1306 
1307 		if (procfs_proc_lock(pfs->pfs_pid, &p, ESRCH) != 0)
1308 			break;
1309 
1310 		if (ap->a_ncookies) {
1311 			ncookies = min(ncookies, (nproc_targets - i));
1312 			cookies = malloc(ncookies * sizeof (off_t),
1313 			    M_TEMP, M_WAITOK);
1314 			*ap->a_cookies = cookies;
1315 		}
1316 
1317 		for (pt = &proc_targets[i];
1318 		     uio->uio_resid >= UIO_MX && i < nproc_targets; pt++, i++) {
1319 			if (pt->pt_valid) {
1320 				/* XXX LWP can disappear */
1321 				mutex_enter(&p->p_smutex);
1322 				l = proc_representative_lwp(p, NULL, 1);
1323 				mutex_exit(&p->p_smutex);
1324 				if ((*pt->pt_valid)(l, vp->v_mount) == 0)
1325 					continue;
1326 			}
1327 
1328 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid,
1329 			    pt->pt_pfstype, -1);
1330 			d.d_namlen = pt->pt_namlen;
1331 			memcpy(d.d_name, pt->pt_name, pt->pt_namlen + 1);
1332 			d.d_type = pt->pt_type;
1333 
1334 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1335 				break;
1336 			if (cookies)
1337 				*cookies++ = i + 1;
1338 		}
1339 
1340 		procfs_proc_unlock(p);
1341 	    	break;
1342 	}
1343 	case PFSfd: {
1344 		struct proc *p;
1345 		struct filedesc	*fdp;
1346 		struct file *fp;
1347 		int lim, nc = 0;
1348 
1349 		if ((error = procfs_proc_lock(pfs->pfs_pid, &p, ESRCH)) != 0)
1350 			return error;
1351 
1352 		if (kauth_authorize_process(kauth_cred_get(),
1353 		    KAUTH_PROCESS_CANSEE, p, NULL, NULL, NULL) != 0) {
1354 		    	procfs_proc_unlock(p);
1355 			return ESRCH;
1356 		}
1357 
1358 		fdp = p->p_fd;
1359 		nfd = fdp->fd_nfiles;
1360 
1361 		lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfiles);
1362 		if (i >= lim) {
1363 		    	procfs_proc_unlock(p);
1364 			return 0;
1365 		}
1366 
1367 		if (ap->a_ncookies) {
1368 			ncookies = min(ncookies, (nfd + 2 - i));
1369 			cookies = malloc(ncookies * sizeof (off_t),
1370 			    M_TEMP, M_WAITOK);
1371 			*ap->a_cookies = cookies;
1372 		}
1373 
1374 		for (; i < 2 && uio->uio_resid >= UIO_MX; i++) {
1375 			pt = &proc_targets[i];
1376 			d.d_namlen = pt->pt_namlen;
1377 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid,
1378 			    pt->pt_pfstype, -1);
1379 			(void)memcpy(d.d_name, pt->pt_name, pt->pt_namlen + 1);
1380 			d.d_type = pt->pt_type;
1381 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1382 				break;
1383 			if (cookies)
1384 				*cookies++ = i + 1;
1385 			nc++;
1386 		}
1387 		if (error) {
1388 			ncookies = nc;
1389 			break;
1390 		}
1391 		for (; uio->uio_resid >= UIO_MX && i < nfd; i++) {
1392 			/* check the descriptor exists */
1393 			if ((fp = fd_getfile(fdp, i - 2)) == NULL)
1394 				continue;
1395 			mutex_exit(&fp->f_lock);
1396 
1397 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid, PFSfd, i - 2);
1398 			d.d_namlen = snprintf(d.d_name, sizeof(d.d_name),
1399 			    "%lld", (long long)(i - 2));
1400 			d.d_type = VREG;
1401 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1402 				break;
1403 			if (cookies)
1404 				*cookies++ = i + 1;
1405 			nc++;
1406 		}
1407 		ncookies = nc;
1408 		procfs_proc_unlock(p);
1409 		break;
1410 	}
1411 
1412 	/*
1413 	 * this is for the root of the procfs filesystem
1414 	 * what is needed are special entries for "curproc"
1415 	 * and "self" followed by an entry for each process
1416 	 * on allproc.
1417 	 */
1418 
1419 	case PFSroot: {
1420 		int nc = 0;
1421 
1422 		if (ap->a_ncookies) {
1423 			/*
1424 			 * XXX Potentially allocating too much space here,
1425 			 * but I'm lazy. This loop needs some work.
1426 			 */
1427 			cookies = malloc(ncookies * sizeof (off_t),
1428 			    M_TEMP, M_WAITOK);
1429 			*ap->a_cookies = cookies;
1430 		}
1431 		error = 0;
1432 		/* 0 ... 3 are static entries. */
1433 		for (; i <= 3 && uio->uio_resid >= UIO_MX; i++) {
1434 			switch (i) {
1435 			case 0:		/* `.' */
1436 			case 1:		/* `..' */
1437 				d.d_fileno = PROCFS_FILENO(0, PFSroot, -1);
1438 				d.d_namlen = i + 1;
1439 				memcpy(d.d_name, "..", d.d_namlen);
1440 				d.d_name[i + 1] = '\0';
1441 				d.d_type = DT_DIR;
1442 				break;
1443 
1444 			case 2:
1445 				d.d_fileno = PROCFS_FILENO(0, PFScurproc, -1);
1446 				d.d_namlen = sizeof("curproc") - 1;
1447 				memcpy(d.d_name, "curproc", sizeof("curproc"));
1448 				d.d_type = DT_LNK;
1449 				break;
1450 
1451 			case 3:
1452 				d.d_fileno = PROCFS_FILENO(0, PFSself, -1);
1453 				d.d_namlen = sizeof("self") - 1;
1454 				memcpy(d.d_name, "self", sizeof("self"));
1455 				d.d_type = DT_LNK;
1456 				break;
1457 			}
1458 
1459 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1460 				break;
1461 			nc++;
1462 			if (cookies)
1463 				*cookies++ = i + 1;
1464 		}
1465 		/* 4 ... are process entries. */
1466 		ctx.uiop = uio;
1467 		ctx.error = 0;
1468 		ctx.off = 4;
1469 		ctx.startoff = i;
1470 		ctx.cookies = cookies;
1471 		ctx.ncookies = nc;
1472 		proclist_foreach_call(&allproc,
1473 		    procfs_root_readdir_callback, &ctx);
1474 		cookies = ctx.cookies;
1475 		nc = ctx.ncookies;
1476 		error = ctx.error;
1477 		if (error)
1478 			break;
1479 
1480 		/* misc entries. */
1481 		if (i < ctx.off)
1482 			i = ctx.off;
1483 		if (i >= ctx.off + nproc_root_targets)
1484 			break;
1485 		for (pt = &proc_root_targets[i - ctx.off];
1486 		    uio->uio_resid >= UIO_MX &&
1487 		    pt < &proc_root_targets[nproc_root_targets];
1488 		    pt++, i++) {
1489 			if (pt->pt_valid &&
1490 			    (*pt->pt_valid)(NULL, vp->v_mount) == 0)
1491 				continue;
1492 			d.d_fileno = PROCFS_FILENO(0, pt->pt_pfstype, -1);
1493 			d.d_namlen = pt->pt_namlen;
1494 			memcpy(d.d_name, pt->pt_name, pt->pt_namlen + 1);
1495 			d.d_type = pt->pt_type;
1496 
1497 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1498 				break;
1499 			nc++;
1500 			if (cookies)
1501 				*cookies++ = i + 1;
1502 		}
1503 
1504 		ncookies = nc;
1505 		break;
1506 	}
1507 
1508 	default:
1509 		error = ENOTDIR;
1510 		break;
1511 	}
1512 
1513 	if (ap->a_ncookies) {
1514 		if (error) {
1515 			if (cookies)
1516 				free(*ap->a_cookies, M_TEMP);
1517 			*ap->a_ncookies = 0;
1518 			*ap->a_cookies = NULL;
1519 		} else
1520 			*ap->a_ncookies = ncookies;
1521 	}
1522 	uio->uio_offset = i;
1523 	return (error);
1524 }
1525 
1526 /*
1527  * readlink reads the link of `curproc' and others
1528  */
1529 int
1530 procfs_readlink(v)
1531 	void *v;
1532 {
1533 	struct vop_readlink_args *ap = v;
1534 	char bf[16];		/* should be enough */
1535 	char *bp = bf;
1536 	char *path = NULL;
1537 	int len = 0;
1538 	int error = 0;
1539 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
1540 	struct proc *pown;
1541 
1542 	if (pfs->pfs_fileno == PROCFS_FILENO(0, PFScurproc, -1))
1543 		len = snprintf(bf, sizeof(bf), "%ld", (long)curproc->p_pid);
1544 	else if (pfs->pfs_fileno == PROCFS_FILENO(0, PFSself, -1))
1545 		len = snprintf(bf, sizeof(bf), "%s", "curproc");
1546 	else if (pfs->pfs_fileno == PROCFS_FILENO(pfs->pfs_pid, PFScwd, -1) ||
1547 	    pfs->pfs_fileno == PROCFS_FILENO(pfs->pfs_pid, PFSchroot, -1) ||
1548 	    pfs->pfs_fileno == PROCFS_FILENO(pfs->pfs_pid, PFSexe, -1)) {
1549 		if ((error = procfs_proc_lock(pfs->pfs_pid, &pown, ESRCH)) != 0)
1550 			return error;
1551 		MALLOC(path, char *, MAXPATHLEN + 4, M_TEMP,
1552 		    M_WAITOK|M_CANFAIL);
1553 		if (path == NULL) {
1554 			procfs_proc_unlock(pown);
1555 			return (ENOMEM);
1556 		}
1557 		bp = path + MAXPATHLEN;
1558 		*--bp = '\0';
1559 		procfs_dir(PROCFS_TYPE(pfs->pfs_fileno), curlwp, pown,
1560 		    &bp, path, MAXPATHLEN);
1561 		procfs_proc_unlock(pown);
1562 		len = strlen(bp);
1563 	} else {
1564 		struct file *fp;
1565 		struct vnode *vxp, *vp;
1566 
1567 		if ((error = procfs_proc_lock(pfs->pfs_pid, &pown, ESRCH)) != 0)
1568 			return error;
1569 
1570 		fp = fd_getfile(pown->p_fd, pfs->pfs_fd);
1571 		if (fp == NULL) {
1572 			procfs_proc_unlock(pown);
1573 			return EBADF;
1574 		}
1575 
1576 		FILE_USE(fp);
1577 		switch (fp->f_type) {
1578 		case DTYPE_VNODE:
1579 			vxp = (struct vnode *)fp->f_data;
1580 			if (vxp->v_type != VDIR) {
1581 				FILE_UNUSE(fp, curlwp);
1582 				error = EINVAL;
1583 				break;
1584 			}
1585 			if ((path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK))
1586 			    == NULL) {
1587 				FILE_UNUSE(fp, curlwp);
1588 				error = ENOMEM;
1589 				break;
1590 			}
1591 			bp = path + MAXPATHLEN;
1592 			*--bp = '\0';
1593 
1594 			/*
1595 			 * XXX: kludge to avoid locking against ourselves
1596 			 * in getcwd()
1597 			 */
1598 			if (vxp->v_tag == VT_PROCFS) {
1599 				*--bp = '/';
1600 			} else {
1601 				rw_enter(&curproc->p_cwdi->cwdi_lock, RW_READER);
1602 				vp = curproc->p_cwdi->cwdi_rdir;
1603 				if (vp == NULL)
1604 					vp = rootvnode;
1605 				error = getcwd_common(vxp, vp, &bp, path,
1606 				    MAXPATHLEN / 2, 0, curlwp);
1607 				rw_exit(&curproc->p_cwdi->cwdi_lock);
1608 			}
1609 			if (error)
1610 				break;
1611 			len = strlen(bp);
1612 			break;
1613 
1614 		case DTYPE_MISC:
1615 			len = snprintf(bf, sizeof(bf), "%s", "[misc]");
1616 			break;
1617 
1618 		case DTYPE_KQUEUE:
1619 			len = snprintf(bf, sizeof(bf), "%s", "[kqueue]");
1620 			break;
1621 
1622 		default:
1623 			error = EINVAL;
1624 			break;
1625 		}
1626 		FILE_UNUSE(fp, curlwp);
1627 		procfs_proc_unlock(pown);
1628 	}
1629 
1630 	if (error == 0)
1631 		error = uiomove(bp, len, ap->a_uio);
1632 	if (path)
1633 		free(path, M_TEMP);
1634 	return error;
1635 }
1636 
1637 /*
1638  * convert decimal ascii to int
1639  */
1640 static int
1641 atoi(b, len)
1642 	const char *b;
1643 	size_t len;
1644 {
1645 	int p = 0;
1646 
1647 	while (len--) {
1648 		char c = *b++;
1649 		if (c < '0' || c > '9')
1650 			return -1;
1651 		p = 10 * p + (c - '0');
1652 	}
1653 
1654 	return p;
1655 }
1656