xref: /netbsd-src/sys/miscfs/procfs/procfs_vnops.c (revision d9ac053afa92f3d7a493f791e10558f738234217)
1 /*	$NetBSD: procfs_vnops.c,v 1.133 2006/06/13 13:57:33 yamt Exp $	*/
2 
3 /*
4  * Copyright (c) 1993, 1995
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Jan-Simon Pendry.
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. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)procfs_vnops.c	8.18 (Berkeley) 5/21/95
35  */
36 
37 /*
38  * Copyright (c) 1993 Jan-Simon Pendry
39  *
40  * This code is derived from software contributed to Berkeley by
41  * Jan-Simon Pendry.
42  *
43  * Redistribution and use in source and binary forms, with or without
44  * modification, are permitted provided that the following conditions
45  * are met:
46  * 1. Redistributions of source code must retain the above copyright
47  *    notice, this list of conditions and the following disclaimer.
48  * 2. Redistributions in binary form must reproduce the above copyright
49  *    notice, this list of conditions and the following disclaimer in the
50  *    documentation and/or other materials provided with the distribution.
51  * 3. All advertising materials mentioning features or use of this software
52  *    must display the following acknowledgement:
53  *	This product includes software developed by the University of
54  *	California, Berkeley and its contributors.
55  * 4. Neither the name of the University nor the names of its contributors
56  *    may be used to endorse or promote products derived from this software
57  *    without specific prior written permission.
58  *
59  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
60  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
61  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
62  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
63  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
64  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
65  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
66  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
67  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
68  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
69  * SUCH DAMAGE.
70  *
71  *	@(#)procfs_vnops.c	8.18 (Berkeley) 5/21/95
72  */
73 
74 /*
75  * procfs vnode interface
76  */
77 
78 #include <sys/cdefs.h>
79 __KERNEL_RCSID(0, "$NetBSD: procfs_vnops.c,v 1.133 2006/06/13 13:57:33 yamt Exp $");
80 
81 #include <sys/param.h>
82 #include <sys/systm.h>
83 #include <sys/time.h>
84 #include <sys/kernel.h>
85 #include <sys/file.h>
86 #include <sys/filedesc.h>
87 #include <sys/proc.h>
88 #include <sys/vnode.h>
89 #include <sys/namei.h>
90 #include <sys/malloc.h>
91 #include <sys/mount.h>
92 #include <sys/dirent.h>
93 #include <sys/resourcevar.h>
94 #include <sys/stat.h>
95 #include <sys/ptrace.h>
96 #include <sys/sysctl.h> /* XXX for curtain */
97 #include <sys/kauth.h>
98 
99 #include <uvm/uvm_extern.h>	/* for PAGE_SIZE */
100 
101 #include <machine/reg.h>
102 
103 #include <miscfs/genfs/genfs.h>
104 #include <miscfs/procfs/procfs.h>
105 
106 /*
107  * Vnode Operations.
108  *
109  */
110 
111 static int procfs_validfile_linux(struct lwp *, struct mount *);
112 static int procfs_root_readdir_callback(struct proc *, void *);
113 static struct vnode *procfs_dir(pfstype, struct lwp *, struct proc *,
114 				char **, char *, int);
115 
116 /*
117  * This is a list of the valid names in the
118  * process-specific sub-directories.  It is
119  * used in procfs_lookup and procfs_readdir
120  */
121 static const struct proc_target {
122 	u_char	pt_type;
123 	u_char	pt_namlen;
124 	const char	*pt_name;
125 	pfstype	pt_pfstype;
126 	int	(*pt_valid)(struct lwp *, struct mount *);
127 } proc_targets[] = {
128 #define N(s) sizeof(s)-1, s
129 	/*	  name		type		validp */
130 	{ DT_DIR, N("."),	PFSproc,		NULL },
131 	{ DT_DIR, N(".."),	PFSroot,		NULL },
132 	{ DT_DIR, N("fd"),	PFSfd,		NULL },
133 	{ DT_REG, N("file"),	PFSfile,		procfs_validfile },
134 	{ DT_REG, N("mem"),	PFSmem,		NULL },
135 	{ DT_REG, N("regs"),	PFSregs,		procfs_validregs },
136 	{ DT_REG, N("fpregs"),	PFSfpregs,	procfs_validfpregs },
137 	{ DT_REG, N("ctl"),	PFSctl,		NULL },
138 	{ DT_REG, N("stat"),	PFSstat,		procfs_validfile_linux },
139 	{ DT_REG, N("status"),	PFSstatus,	NULL },
140 	{ DT_REG, N("note"),	PFSnote,		NULL },
141 	{ DT_REG, N("notepg"),	PFSnotepg,	NULL },
142 	{ DT_REG, N("map"),	PFSmap,		procfs_validmap },
143 	{ DT_REG, N("maps"),	PFSmaps,		procfs_validmap },
144 	{ DT_REG, N("cmdline"), PFScmdline,	NULL },
145 	{ DT_REG, N("exe"),	PFSfile,		procfs_validfile_linux },
146 	{ DT_LNK, N("cwd"),	PFScwd,		NULL },
147 	{ DT_LNK, N("root"),	PFSchroot,	NULL },
148 #ifdef __HAVE_PROCFS_MACHDEP
149 	PROCFS_MACHDEP_NODETYPE_DEFNS
150 #endif
151 #undef N
152 };
153 static const int nproc_targets = sizeof(proc_targets) / sizeof(proc_targets[0]);
154 
155 /*
156  * List of files in the root directory. Note: the validate function will
157  * be called with p == NULL for these ones.
158  */
159 static const struct proc_target proc_root_targets[] = {
160 #define N(s) sizeof(s)-1, s
161 	/*	  name		    type	    validp */
162 	{ DT_REG, N("meminfo"),     PFSmeminfo,        procfs_validfile_linux },
163 	{ DT_REG, N("cpuinfo"),     PFScpuinfo,        procfs_validfile_linux },
164 	{ DT_REG, N("uptime"),      PFSuptime,         procfs_validfile_linux },
165 	{ DT_REG, N("mounts"),	    PFSmounts,	       procfs_validfile_linux },
166 #undef N
167 };
168 static const int nproc_root_targets =
169     sizeof(proc_root_targets) / sizeof(proc_root_targets[0]);
170 
171 int	procfs_lookup(void *);
172 #define	procfs_create	genfs_eopnotsupp
173 #define	procfs_mknod	genfs_eopnotsupp
174 int	procfs_open(void *);
175 int	procfs_close(void *);
176 int	procfs_access(void *);
177 int	procfs_getattr(void *);
178 int	procfs_setattr(void *);
179 #define	procfs_read	procfs_rw
180 #define	procfs_write	procfs_rw
181 #define	procfs_fcntl	genfs_fcntl
182 #define	procfs_ioctl	genfs_enoioctl
183 #define	procfs_poll	genfs_poll
184 #define procfs_revoke	genfs_revoke
185 #define	procfs_fsync	genfs_nullop
186 #define	procfs_seek	genfs_nullop
187 #define	procfs_remove	genfs_eopnotsupp
188 int	procfs_link(void *);
189 #define	procfs_rename	genfs_eopnotsupp
190 #define	procfs_mkdir	genfs_eopnotsupp
191 #define	procfs_rmdir	genfs_eopnotsupp
192 int	procfs_symlink(void *);
193 int	procfs_readdir(void *);
194 int	procfs_readlink(void *);
195 #define	procfs_abortop	genfs_abortop
196 int	procfs_inactive(void *);
197 int	procfs_reclaim(void *);
198 #define	procfs_lock	genfs_lock
199 #define	procfs_unlock	genfs_unlock
200 #define	procfs_bmap	genfs_badop
201 #define	procfs_strategy	genfs_badop
202 int	procfs_print(void *);
203 int	procfs_pathconf(void *);
204 #define	procfs_islocked	genfs_islocked
205 #define	procfs_advlock	genfs_einval
206 #define	procfs_bwrite	genfs_eopnotsupp
207 #define procfs_putpages	genfs_null_putpages
208 
209 static int atoi(const char *, size_t);
210 
211 /*
212  * procfs vnode operations.
213  */
214 int (**procfs_vnodeop_p)(void *);
215 const struct vnodeopv_entry_desc procfs_vnodeop_entries[] = {
216 	{ &vop_default_desc, vn_default_error },
217 	{ &vop_lookup_desc, procfs_lookup },		/* lookup */
218 	{ &vop_create_desc, procfs_create },		/* create */
219 	{ &vop_mknod_desc, procfs_mknod },		/* mknod */
220 	{ &vop_open_desc, procfs_open },		/* open */
221 	{ &vop_close_desc, procfs_close },		/* close */
222 	{ &vop_access_desc, procfs_access },		/* access */
223 	{ &vop_getattr_desc, procfs_getattr },		/* getattr */
224 	{ &vop_setattr_desc, procfs_setattr },		/* setattr */
225 	{ &vop_read_desc, procfs_read },		/* read */
226 	{ &vop_write_desc, procfs_write },		/* write */
227 	{ &vop_fcntl_desc, procfs_fcntl },		/* fcntl */
228 	{ &vop_ioctl_desc, procfs_ioctl },		/* ioctl */
229 	{ &vop_poll_desc, procfs_poll },		/* poll */
230 	{ &vop_revoke_desc, procfs_revoke },		/* revoke */
231 	{ &vop_fsync_desc, procfs_fsync },		/* fsync */
232 	{ &vop_seek_desc, procfs_seek },		/* seek */
233 	{ &vop_remove_desc, procfs_remove },		/* remove */
234 	{ &vop_link_desc, procfs_link },		/* link */
235 	{ &vop_rename_desc, procfs_rename },		/* rename */
236 	{ &vop_mkdir_desc, procfs_mkdir },		/* mkdir */
237 	{ &vop_rmdir_desc, procfs_rmdir },		/* rmdir */
238 	{ &vop_symlink_desc, procfs_symlink },		/* symlink */
239 	{ &vop_readdir_desc, procfs_readdir },		/* readdir */
240 	{ &vop_readlink_desc, procfs_readlink },	/* readlink */
241 	{ &vop_abortop_desc, procfs_abortop },		/* abortop */
242 	{ &vop_inactive_desc, procfs_inactive },	/* inactive */
243 	{ &vop_reclaim_desc, procfs_reclaim },		/* reclaim */
244 	{ &vop_lock_desc, procfs_lock },		/* lock */
245 	{ &vop_unlock_desc, procfs_unlock },		/* unlock */
246 	{ &vop_bmap_desc, procfs_bmap },		/* bmap */
247 	{ &vop_strategy_desc, procfs_strategy },	/* strategy */
248 	{ &vop_print_desc, procfs_print },		/* print */
249 	{ &vop_islocked_desc, procfs_islocked },	/* islocked */
250 	{ &vop_pathconf_desc, procfs_pathconf },	/* pathconf */
251 	{ &vop_advlock_desc, procfs_advlock },		/* advlock */
252 	{ &vop_putpages_desc, procfs_putpages },	/* putpages */
253 	{ NULL, NULL }
254 };
255 const struct vnodeopv_desc procfs_vnodeop_opv_desc =
256 	{ &procfs_vnodeop_p, procfs_vnodeop_entries };
257 /*
258  * set things up for doing i/o on
259  * the pfsnode (vp).  (vp) is locked
260  * on entry, and should be left locked
261  * on exit.
262  *
263  * for procfs we don't need to do anything
264  * in particular for i/o.  all that is done
265  * is to support exclusive open on process
266  * memory images.
267  */
268 int
269 procfs_open(v)
270 	void *v;
271 {
272 	struct vop_open_args /* {
273 		struct vnode *a_vp;
274 		int  a_mode;
275 		kauth_cred_t a_cred;
276 		struct lwp *a_l;
277 	} */ *ap = v;
278 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
279 	struct lwp *l1;
280 	struct proc *p2;
281 	int error;
282 
283 	l1 = ap->a_l;				/* tracer */
284 	p2 = PFIND(pfs->pfs_pid);		/* traced */
285 
286 	if (p2 == NULL)
287 		return (ENOENT);		/* was ESRCH, jsp */
288 
289 	switch (pfs->pfs_type) {
290 	case PFSmem:
291 		if (((pfs->pfs_flags & FWRITE) && (ap->a_mode & O_EXCL)) ||
292 		    ((pfs->pfs_flags & O_EXCL) && (ap->a_mode & FWRITE)))
293 			return (EBUSY);
294 
295 		if ((error = process_checkioperm(l1, p2)) != 0)
296 			return (error);
297 
298 		if (ap->a_mode & FWRITE)
299 			pfs->pfs_flags = ap->a_mode & (FWRITE|O_EXCL);
300 
301 		return (0);
302 
303 	default:
304 		break;
305 	}
306 
307 	return (0);
308 }
309 
310 /*
311  * close the pfsnode (vp) after doing i/o.
312  * (vp) is not locked on entry or exit.
313  *
314  * nothing to do for procfs other than undo
315  * any exclusive open flag (see _open above).
316  */
317 int
318 procfs_close(v)
319 	void *v;
320 {
321 	struct vop_close_args /* {
322 		struct vnode *a_vp;
323 		int  a_fflag;
324 		kauth_cred_t a_cred;
325 		struct lwp *a_l;
326 	} */ *ap = v;
327 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
328 
329 	switch (pfs->pfs_type) {
330 	case PFSmem:
331 		if ((ap->a_fflag & FWRITE) && (pfs->pfs_flags & O_EXCL))
332 			pfs->pfs_flags &= ~(FWRITE|O_EXCL);
333 		break;
334 
335 	default:
336 		break;
337 	}
338 
339 	return (0);
340 }
341 
342 /*
343  * _inactive is called when the pfsnode
344  * is vrele'd and the reference count goes
345  * to zero.  (vp) will be on the vnode free
346  * list, so to get it back vget() must be
347  * used.
348  *
349  * for procfs, check if the process is still
350  * alive and if it isn't then just throw away
351  * the vnode by calling vgone().  this may
352  * be overkill and a waste of time since the
353  * chances are that the process will still be
354  * there and PFIND is not free.
355  *
356  * (vp) is locked on entry, but must be unlocked on exit.
357  */
358 int
359 procfs_inactive(v)
360 	void *v;
361 {
362 	struct vop_inactive_args /* {
363 		struct vnode *a_vp;
364 		struct proc *a_p;
365 	} */ *ap = v;
366 	struct vnode *vp = ap->a_vp;
367 	struct pfsnode *pfs = VTOPFS(vp);
368 
369 	VOP_UNLOCK(vp, 0);
370 	if (PFIND(pfs->pfs_pid) == NULL && (vp->v_flag & VXLOCK) == 0)
371 		vgone(vp);
372 
373 	return (0);
374 }
375 
376 /*
377  * _reclaim is called when getnewvnode()
378  * wants to make use of an entry on the vnode
379  * free list.  at this time the filesystem needs
380  * to free any private data and remove the node
381  * from any private lists.
382  */
383 int
384 procfs_reclaim(v)
385 	void *v;
386 {
387 	struct vop_reclaim_args /* {
388 		struct vnode *a_vp;
389 	} */ *ap = v;
390 
391 	return (procfs_freevp(ap->a_vp));
392 }
393 
394 /*
395  * Return POSIX pathconf information applicable to special devices.
396  */
397 int
398 procfs_pathconf(v)
399 	void *v;
400 {
401 	struct vop_pathconf_args /* {
402 		struct vnode *a_vp;
403 		int a_name;
404 		register_t *a_retval;
405 	} */ *ap = v;
406 
407 	switch (ap->a_name) {
408 	case _PC_LINK_MAX:
409 		*ap->a_retval = LINK_MAX;
410 		return (0);
411 	case _PC_MAX_CANON:
412 		*ap->a_retval = MAX_CANON;
413 		return (0);
414 	case _PC_MAX_INPUT:
415 		*ap->a_retval = MAX_INPUT;
416 		return (0);
417 	case _PC_PIPE_BUF:
418 		*ap->a_retval = PIPE_BUF;
419 		return (0);
420 	case _PC_CHOWN_RESTRICTED:
421 		*ap->a_retval = 1;
422 		return (0);
423 	case _PC_VDISABLE:
424 		*ap->a_retval = _POSIX_VDISABLE;
425 		return (0);
426 	case _PC_SYNC_IO:
427 		*ap->a_retval = 1;
428 		return (0);
429 	default:
430 		return (EINVAL);
431 	}
432 	/* NOTREACHED */
433 }
434 
435 /*
436  * _print is used for debugging.
437  * just print a readable description
438  * of (vp).
439  */
440 int
441 procfs_print(v)
442 	void *v;
443 {
444 	struct vop_print_args /* {
445 		struct vnode *a_vp;
446 	} */ *ap = v;
447 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
448 
449 	printf("tag VT_PROCFS, type %d, pid %d, mode %x, flags %lx\n",
450 	    pfs->pfs_type, pfs->pfs_pid, pfs->pfs_mode, pfs->pfs_flags);
451 	return 0;
452 }
453 
454 int
455 procfs_link(v)
456 	void *v;
457 {
458 	struct vop_link_args /* {
459 		struct vnode *a_dvp;
460 		struct vnode *a_vp;
461 		struct componentname *a_cnp;
462 	} */ *ap = v;
463 
464 	VOP_ABORTOP(ap->a_dvp, ap->a_cnp);
465 	vput(ap->a_dvp);
466 	return (EROFS);
467 }
468 
469 int
470 procfs_symlink(v)
471 	void *v;
472 {
473 	struct vop_symlink_args /* {
474 		struct vnode *a_dvp;
475 		struct vnode **a_vpp;
476 		struct componentname *a_cnp;
477 		struct vattr *a_vap;
478 		char *a_target;
479 	} */ *ap = v;
480 
481 	VOP_ABORTOP(ap->a_dvp, ap->a_cnp);
482 	vput(ap->a_dvp);
483 	return (EROFS);
484 }
485 
486 /*
487  * Works out the path to (and vnode of) the target process's current
488  * working directory or chroot.  If the caller is in a chroot and
489  * can't "reach" the target's cwd or root (or some other error
490  * occurs), a "/" is returned for the path and a NULL pointer is
491  * returned for the vnode.
492  */
493 static struct vnode *
494 procfs_dir(pfstype t, struct lwp *caller, struct proc *target,
495 	   char **bpp, char *path, int len)
496 {
497 	struct vnode *vp, *rvp = caller->l_proc->p_cwdi->cwdi_rdir;
498 	char *bp;
499 
500 	bp = bpp ? *bpp : NULL;
501 
502 	switch (t) {
503 	case PFScwd:
504 		vp = target->p_cwdi->cwdi_cdir;
505 		break;
506 	case PFSchroot:
507 		vp = target->p_cwdi->cwdi_rdir;
508 		break;
509 	default:
510 		return (NULL);
511 	}
512 
513 	if (rvp == NULL)
514 		rvp = rootvnode;
515 	if (vp == NULL || getcwd_common(vp, rvp, bp ? &bp : NULL, path,
516 	    len / 2, 0, caller) != 0) {
517 		vp = NULL;
518 		if (bpp) {
519 			bp = *bpp;
520 			*--bp = '/';
521 		}
522 	}
523 
524 	if (bpp)
525 		*bpp = bp;
526 
527 	return (vp);
528 }
529 
530 /*
531  * Invent attributes for pfsnode (vp) and store
532  * them in (vap).
533  * Directories lengths are returned as zero since
534  * any real length would require the genuine size
535  * to be computed, and nothing cares anyway.
536  *
537  * this is relatively minimal for procfs.
538  */
539 int
540 procfs_getattr(v)
541 	void *v;
542 {
543 	struct vop_getattr_args /* {
544 		struct vnode *a_vp;
545 		struct vattr *a_vap;
546 		kauth_cred_t a_cred;
547 		struct lwp *a_l;
548 	} */ *ap = v;
549 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
550 	struct vattr *vap = ap->a_vap;
551 	struct proc *procp;
552 	int error;
553 
554 	/* first check the process still exists */
555 	switch (pfs->pfs_type) {
556 	case PFSroot:
557 	case PFScurproc:
558 	case PFSself:
559 		procp = 0;
560 		break;
561 
562 	default:
563 		procp = PFIND(pfs->pfs_pid);
564 		if (procp == NULL)
565 			return (ENOENT);
566 		break;
567 	}
568 
569 	if (procp != NULL) {
570 		if (kauth_authorize_process(kauth_cred_get(),
571 		    KAUTH_PROCESS_CANSEE, procp, NULL, NULL, NULL) != 0)
572 			return (ENOENT);
573 	}
574 
575 	error = 0;
576 
577 	/* start by zeroing out the attributes */
578 	VATTR_NULL(vap);
579 
580 	/* next do all the common fields */
581 	vap->va_type = ap->a_vp->v_type;
582 	vap->va_mode = pfs->pfs_mode;
583 	vap->va_fileid = pfs->pfs_fileno;
584 	vap->va_flags = 0;
585 	vap->va_blocksize = PAGE_SIZE;
586 
587 	/*
588 	 * Make all times be current TOD.
589 	 *
590 	 * It would be possible to get the process start
591 	 * time from the p_stats structure, but there's
592 	 * no "file creation" time stamp anyway, and the
593 	 * p_stats structure is not addressable if u. gets
594 	 * swapped out for that process.
595 	 */
596 	getnanotime(&vap->va_ctime);
597 	vap->va_atime = vap->va_mtime = vap->va_ctime;
598 	if (procp)
599 		TIMEVAL_TO_TIMESPEC(&procp->p_stats->p_start,
600 		    &vap->va_birthtime);
601 	else
602 		getnanotime(&vap->va_birthtime);
603 
604 	switch (pfs->pfs_type) {
605 	case PFSmem:
606 	case PFSregs:
607 	case PFSfpregs:
608 #if defined(__HAVE_PROCFS_MACHDEP) && defined(PROCFS_MACHDEP_PROTECT_CASES)
609 	PROCFS_MACHDEP_PROTECT_CASES
610 #endif
611 		/*
612 		 * If the process has exercised some setuid or setgid
613 		 * privilege, then rip away read/write permission so
614 		 * that only root can gain access.
615 		 */
616 		if (procp->p_flag & P_SUGID)
617 			vap->va_mode &= ~(S_IRUSR|S_IWUSR);
618 		/* FALLTHROUGH */
619 	case PFSctl:
620 	case PFSstatus:
621 	case PFSstat:
622 	case PFSnote:
623 	case PFSnotepg:
624 	case PFSmap:
625 	case PFSmaps:
626 	case PFScmdline:
627 		vap->va_nlink = 1;
628 		vap->va_uid = kauth_cred_geteuid(procp->p_cred);
629 		vap->va_gid = kauth_cred_getegid(procp->p_cred);
630 		break;
631 	case PFSmeminfo:
632 	case PFScpuinfo:
633 	case PFSuptime:
634 	case PFSmounts:
635 		vap->va_nlink = 1;
636 		vap->va_uid = vap->va_gid = 0;
637 		break;
638 
639 	default:
640 		break;
641 	}
642 
643 	/*
644 	 * now do the object specific fields
645 	 *
646 	 * The size could be set from struct reg, but it's hardly
647 	 * worth the trouble, and it puts some (potentially) machine
648 	 * dependent data into this machine-independent code.  If it
649 	 * becomes important then this function should break out into
650 	 * a per-file stat function in the corresponding .c file.
651 	 */
652 
653 	switch (pfs->pfs_type) {
654 	case PFSroot:
655 		/*
656 		 * Set nlink to 1 to tell fts(3) we don't actually know.
657 		 */
658 		vap->va_nlink = 1;
659 		vap->va_uid = 0;
660 		vap->va_gid = 0;
661 		vap->va_bytes = vap->va_size = DEV_BSIZE;
662 		break;
663 
664 	case PFSself:
665 	case PFScurproc: {
666 		char bf[16];		/* should be enough */
667 		vap->va_nlink = 1;
668 		vap->va_uid = 0;
669 		vap->va_gid = 0;
670 		vap->va_bytes = vap->va_size =
671 		    snprintf(bf, sizeof(bf), "%ld", (long)curproc->p_pid);
672 		break;
673 	}
674 
675 	case PFSfd:
676 		if (pfs->pfs_fd != -1) {
677 			struct file *fp;
678 			struct proc *pown;
679 
680 			if ((error = procfs_getfp(pfs, &pown, &fp)) != 0)
681 				return error;
682 			FILE_USE(fp);
683 			vap->va_nlink = 1;
684 			vap->va_uid = kauth_cred_geteuid(fp->f_cred);
685 			vap->va_gid = kauth_cred_getegid(fp->f_cred);
686 			switch (fp->f_type) {
687 			case DTYPE_VNODE:
688 				vap->va_bytes = vap->va_size =
689 				    ((struct vnode *)fp->f_data)->v_size;
690 				break;
691 			default:
692 				vap->va_bytes = vap->va_size = 0;
693 				break;
694 			}
695 			FILE_UNUSE(fp, proc_representative_lwp(pown));
696 			break;
697 		}
698 		/*FALLTHROUGH*/
699 	case PFSproc:
700 		vap->va_nlink = 2;
701 		vap->va_uid = kauth_cred_geteuid(procp->p_cred);
702 		vap->va_gid = kauth_cred_getegid(procp->p_cred);
703 		vap->va_bytes = vap->va_size = DEV_BSIZE;
704 		break;
705 
706 	case PFSfile:
707 		error = EOPNOTSUPP;
708 		break;
709 
710 	case PFSmem:
711 		vap->va_bytes = vap->va_size =
712 			ctob(procp->p_vmspace->vm_tsize +
713 				    procp->p_vmspace->vm_dsize +
714 				    procp->p_vmspace->vm_ssize);
715 		break;
716 
717 #if defined(PT_GETREGS) || defined(PT_SETREGS)
718 	case PFSregs:
719 		vap->va_bytes = vap->va_size = sizeof(struct reg);
720 		break;
721 #endif
722 
723 #if defined(PT_GETFPREGS) || defined(PT_SETFPREGS)
724 	case PFSfpregs:
725 		vap->va_bytes = vap->va_size = sizeof(struct fpreg);
726 		break;
727 #endif
728 
729 	case PFSctl:
730 	case PFSstatus:
731 	case PFSstat:
732 	case PFSnote:
733 	case PFSnotepg:
734 	case PFScmdline:
735 	case PFSmeminfo:
736 	case PFScpuinfo:
737 	case PFSuptime:
738 	case PFSmounts:
739 		vap->va_bytes = vap->va_size = 0;
740 		break;
741 	case PFSmap:
742 	case PFSmaps:
743 		/*
744 		 * Advise a larger blocksize for the map files, so that
745 		 * they may be read in one pass.
746 		 */
747 		vap->va_blocksize = 4 * PAGE_SIZE;
748 		vap->va_bytes = vap->va_size = 0;
749 		break;
750 
751 	case PFScwd:
752 	case PFSchroot: {
753 		char *path, *bp;
754 
755 		MALLOC(path, char *, MAXPATHLEN + 4, M_TEMP,
756 		    M_WAITOK|M_CANFAIL);
757 		if (path == NULL)
758 			return (ENOMEM);
759 		vap->va_nlink = 1;
760 		vap->va_uid = 0;
761 		vap->va_gid = 0;
762 		bp = path + MAXPATHLEN;
763 		*--bp = '\0';
764 		(void)procfs_dir(pfs->pfs_type, curlwp, procp, &bp, path,
765 		     MAXPATHLEN);
766 		vap->va_bytes = vap->va_size = strlen(bp);
767 		free(path, M_TEMP);
768 		break;
769 	}
770 
771 #ifdef __HAVE_PROCFS_MACHDEP
772 	PROCFS_MACHDEP_NODETYPE_CASES
773 		error = procfs_machdep_getattr(ap->a_vp, vap, procp);
774 		break;
775 #endif
776 
777 	default:
778 		panic("procfs_getattr");
779 	}
780 
781 	return (error);
782 }
783 
784 /*ARGSUSED*/
785 int
786 procfs_setattr(v)
787 	void *v;
788 {
789 	/*
790 	 * just fake out attribute setting
791 	 * it's not good to generate an error
792 	 * return, otherwise things like creat()
793 	 * will fail when they try to set the
794 	 * file length to 0.  worse, this means
795 	 * that echo $note > /proc/$pid/note will fail.
796 	 */
797 
798 	return (0);
799 }
800 
801 /*
802  * implement access checking.
803  *
804  * actually, the check for super-user is slightly
805  * broken since it will allow read access to write-only
806  * objects.  this doesn't cause any particular trouble
807  * but does mean that the i/o entry points need to check
808  * that the operation really does make sense.
809  */
810 int
811 procfs_access(v)
812 	void *v;
813 {
814 	struct vop_access_args /* {
815 		struct vnode *a_vp;
816 		int a_mode;
817 		kauth_cred_t a_cred;
818 		struct lwp *a_l;
819 	} */ *ap = v;
820 	struct vattr va;
821 	int error;
822 
823 	if ((error = VOP_GETATTR(ap->a_vp, &va, ap->a_cred, ap->a_l)) != 0)
824 		return (error);
825 
826 	return (vaccess(va.va_type, va.va_mode,
827 	    va.va_uid, va.va_gid, ap->a_mode, ap->a_cred));
828 }
829 
830 /*
831  * lookup.  this is incredibly complicated in the
832  * general case, however for most pseudo-filesystems
833  * very little needs to be done.
834  *
835  * Locking isn't hard here, just poorly documented.
836  *
837  * If we're looking up ".", just vref the parent & return it.
838  *
839  * If we're looking up "..", unlock the parent, and lock "..". If everything
840  * went ok, and we're on the last component and the caller requested the
841  * parent locked, try to re-lock the parent. We do this to prevent lock
842  * races.
843  *
844  * For anything else, get the needed node. Then unlock the parent if not
845  * the last component or not LOCKPARENT (i.e. if we wouldn't re-lock the
846  * parent in the .. case).
847  *
848  * We try to exit with the parent locked in error cases.
849  */
850 int
851 procfs_lookup(v)
852 	void *v;
853 {
854 	struct vop_lookup_args /* {
855 		struct vnode * a_dvp;
856 		struct vnode ** a_vpp;
857 		struct componentname * a_cnp;
858 	} */ *ap = v;
859 	struct componentname *cnp = ap->a_cnp;
860 	struct vnode **vpp = ap->a_vpp;
861 	struct vnode *dvp = ap->a_dvp;
862 	const char *pname = cnp->cn_nameptr;
863 	const struct proc_target *pt = NULL;
864 	struct vnode *fvp;
865 	pid_t pid;
866 	struct pfsnode *pfs;
867 	struct proc *p = NULL;
868 	struct lwp *l = NULL;
869 	int i, error, wantpunlock, iscurproc = 0, isself = 0;
870 
871 	*vpp = NULL;
872 	cnp->cn_flags &= ~PDIRUNLOCK;
873 
874 	if (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)
875 		return (EROFS);
876 
877 	if (cnp->cn_namelen == 1 && *pname == '.') {
878 		*vpp = dvp;
879 		VREF(dvp);
880 		return (0);
881 	}
882 
883 	wantpunlock = (~cnp->cn_flags & (LOCKPARENT | ISLASTCN));
884 	pfs = VTOPFS(dvp);
885 	switch (pfs->pfs_type) {
886 	case PFSroot:
887 		/*
888 		 * Shouldn't get here with .. in the root node.
889 		 */
890 		if (cnp->cn_flags & ISDOTDOT)
891 			return (EIO);
892 
893 		iscurproc = CNEQ(cnp, "curproc", 7);
894 		isself = CNEQ(cnp, "self", 4);
895 
896 		if (iscurproc || isself) {
897 			error = procfs_allocvp(dvp->v_mount, vpp, 0,
898 			    iscurproc ? PFScurproc : PFSself, -1);
899 			if ((error == 0) && (wantpunlock)) {
900 				VOP_UNLOCK(dvp, 0);
901 				cnp->cn_flags |= PDIRUNLOCK;
902 			}
903 			return (error);
904 		}
905 
906 		for (i = 0; i < nproc_root_targets; i++) {
907 			pt = &proc_root_targets[i];
908 			if (cnp->cn_namelen == pt->pt_namlen &&
909 			    memcmp(pt->pt_name, pname, cnp->cn_namelen) == 0 &&
910 			    (pt->pt_valid == NULL ||
911 			     (*pt->pt_valid)(cnp->cn_lwp, dvp->v_mount)))
912 				break;
913 		}
914 
915 		if (i != nproc_root_targets) {
916 			error = procfs_allocvp(dvp->v_mount, vpp, 0,
917 			    pt->pt_pfstype, -1);
918 			if ((error == 0) && (wantpunlock)) {
919 				VOP_UNLOCK(dvp, 0);
920 				cnp->cn_flags |= PDIRUNLOCK;
921 			}
922 			return (error);
923 		}
924 
925 		pid = (pid_t)atoi(pname, cnp->cn_namelen);
926 
927 		p = PFIND(pid);
928 		if (p == NULL)
929 			break;
930 
931 		error = procfs_allocvp(dvp->v_mount, vpp, pid, PFSproc, -1);
932 		if ((error == 0) && (wantpunlock)) {
933 			VOP_UNLOCK(dvp, 0);
934 			cnp->cn_flags |= PDIRUNLOCK;
935 		}
936 		return (error);
937 
938 	case PFSproc:
939 		/*
940 		 * do the .. dance. We unlock the directory, and then
941 		 * get the root dir. That will automatically return ..
942 		 * locked. Then if the caller wanted dvp locked, we
943 		 * re-lock.
944 		 */
945 		if (cnp->cn_flags & ISDOTDOT) {
946 			VOP_UNLOCK(dvp, 0);
947 			cnp->cn_flags |= PDIRUNLOCK;
948 			error = procfs_root(dvp->v_mount, vpp);
949 			if ((error == 0) && (wantpunlock == 0) &&
950 				    ((error = vn_lock(dvp, LK_EXCLUSIVE)) == 0))
951 				cnp->cn_flags &= ~PDIRUNLOCK;
952 			return (error);
953 		}
954 
955 		p = PFIND(pfs->pfs_pid);
956 		if (p == NULL)
957 			break;
958 		l = proc_representative_lwp(p);
959 
960 		for (pt = proc_targets, i = 0; i < nproc_targets; pt++, i++) {
961 			if (cnp->cn_namelen == pt->pt_namlen &&
962 			    memcmp(pt->pt_name, pname, cnp->cn_namelen) == 0 &&
963 			    (pt->pt_valid == NULL ||
964 			     (*pt->pt_valid)(cnp->cn_lwp, dvp->v_mount)))
965 				goto found;
966 		}
967 		break;
968 
969 	found:
970 		if (pt->pt_pfstype == PFSfile) {
971 			fvp = p->p_textvp;
972 			/* We already checked that it exists. */
973 			VREF(fvp);
974 			vn_lock(fvp, LK_EXCLUSIVE | LK_RETRY);
975 			if (wantpunlock) {
976 				VOP_UNLOCK(dvp, 0);
977 				cnp->cn_flags |= PDIRUNLOCK;
978 			}
979 			*vpp = fvp;
980 			return (0);
981 		}
982 
983 		error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
984 		    pt->pt_pfstype, -1);
985 		if ((error == 0) && (wantpunlock)) {
986 			VOP_UNLOCK(dvp, 0);
987 			cnp->cn_flags |= PDIRUNLOCK;
988 		}
989 		return (error);
990 
991 	case PFSfd: {
992 		int fd;
993 		struct file *fp;
994 		/*
995 		 * do the .. dance. We unlock the directory, and then
996 		 * get the proc dir. That will automatically return ..
997 		 * locked. Then if the caller wanted dvp locked, we
998 		 * re-lock.
999 		 */
1000 		if (cnp->cn_flags & ISDOTDOT) {
1001 			VOP_UNLOCK(dvp, 0);
1002 			cnp->cn_flags |= PDIRUNLOCK;
1003 			error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
1004 			    PFSproc, -1);
1005 			if ((error == 0) && (wantpunlock == 0) &&
1006 				    ((error = vn_lock(dvp, LK_EXCLUSIVE)) == 0))
1007 				cnp->cn_flags &= ~PDIRUNLOCK;
1008 			return (error);
1009 		}
1010 		fd = atoi(pname, cnp->cn_namelen);
1011 		p = PFIND(pfs->pfs_pid);
1012 		if (p == NULL || (fp = fd_getfile(p->p_fd, fd)) == NULL)
1013 			return ENOENT;
1014 		FILE_USE(fp);
1015 
1016 		switch (fp->f_type) {
1017 		case DTYPE_VNODE:
1018 			fvp = (struct vnode *)fp->f_data;
1019 
1020 			/* Don't show directories */
1021 			if (fvp->v_type == VDIR)
1022 				goto symlink;
1023 
1024 			VREF(fvp);
1025 			FILE_UNUSE(fp, l);
1026 			vn_lock(fvp, LK_EXCLUSIVE | LK_RETRY |
1027 			    (p == curproc ? LK_CANRECURSE : 0));
1028 			*vpp = fvp;
1029 			error = 0;
1030 			break;
1031 		default:
1032 		symlink:
1033 			FILE_UNUSE(fp, l);
1034 			error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
1035 			    PFSfd, fd);
1036 			break;
1037 		}
1038 		if ((error == 0) && (wantpunlock)) {
1039 			VOP_UNLOCK(dvp, 0);
1040 			cnp->cn_flags |= PDIRUNLOCK;
1041 		}
1042 		return error;
1043 	}
1044 	default:
1045 		return (ENOTDIR);
1046 	}
1047 
1048 	return (cnp->cn_nameiop == LOOKUP ? ENOENT : EROFS);
1049 }
1050 
1051 int
1052 procfs_validfile(l, mp)
1053 	struct lwp *l;
1054 	struct mount *mp;
1055 {
1056 	return (l->l_proc->p_textvp != NULL);
1057 }
1058 
1059 static int
1060 procfs_validfile_linux(l, mp)
1061 	struct lwp *l;
1062 	struct mount *mp;
1063 {
1064 	int flags;
1065 
1066 	flags = VFSTOPROC(mp)->pmnt_flags;
1067 	return ((flags & PROCFSMNT_LINUXCOMPAT) &&
1068 	    (l == NULL || l->l_proc == NULL || procfs_validfile(l, mp)));
1069 }
1070 
1071 struct procfs_root_readdir_ctx {
1072 	struct uio *uiop;
1073 	off_t *cookies;
1074 	int ncookies;
1075 	off_t off;
1076 	off_t startoff;
1077 	int error;
1078 };
1079 
1080 static int
1081 procfs_root_readdir_callback(struct proc *p, void *arg)
1082 {
1083 	struct procfs_root_readdir_ctx *ctxp = arg;
1084 	struct dirent d;
1085 	struct uio *uiop;
1086 	int error;
1087 
1088 	uiop = ctxp->uiop;
1089 	if (uiop->uio_resid < UIO_MX)
1090 		return -1; /* no space */
1091 
1092 	if (ctxp->off < ctxp->startoff) {
1093 		ctxp->off++;
1094 		return 0;
1095 	}
1096 
1097 	if (kauth_authorize_process(kauth_cred_get(),
1098 	    KAUTH_PROCESS_CANSEE, p, NULL, NULL, NULL) != 0)
1099 		return 0;
1100 
1101 	memset(&d, 0, UIO_MX);
1102 	d.d_reclen = UIO_MX;
1103 	d.d_fileno = PROCFS_FILENO(p->p_pid, PFSproc, -1);
1104 	d.d_namlen = snprintf(d.d_name,
1105 	    UIO_MX - offsetof(struct dirent, d_name), "%ld", (long)p->p_pid);
1106 	d.d_type = DT_DIR;
1107 
1108 	proclist_unlock_read();
1109 	error = uiomove(&d, UIO_MX, uiop);
1110 	proclist_lock_read();
1111 	if (error) {
1112 		ctxp->error = error;
1113 		return -1;
1114 	}
1115 
1116 	ctxp->ncookies++;
1117 	if (ctxp->cookies)
1118 		*(ctxp->cookies)++ = ctxp->off + 1;
1119 	ctxp->off++;
1120 
1121 	return 0;
1122 }
1123 
1124 /*
1125  * readdir returns directory entries from pfsnode (vp).
1126  *
1127  * the strategy here with procfs is to generate a single
1128  * directory entry at a time (struct dirent) and then
1129  * copy that out to userland using uiomove.  a more efficent
1130  * though more complex implementation, would try to minimize
1131  * the number of calls to uiomove().  for procfs, this is
1132  * hardly worth the added code complexity.
1133  *
1134  * this should just be done through read()
1135  */
1136 int
1137 procfs_readdir(v)
1138 	void *v;
1139 {
1140 	struct vop_readdir_args /* {
1141 		struct vnode *a_vp;
1142 		struct uio *a_uio;
1143 		kauth_cred_t a_cred;
1144 		int *a_eofflag;
1145 		off_t **a_cookies;
1146 		int *a_ncookies;
1147 	} */ *ap = v;
1148 	struct uio *uio = ap->a_uio;
1149 	struct dirent d;
1150 	struct pfsnode *pfs;
1151 	off_t i;
1152 	int error;
1153 	off_t *cookies = NULL;
1154 	int ncookies;
1155 	struct vnode *vp;
1156 	const struct proc_target *pt;
1157 	struct procfs_root_readdir_ctx ctx;
1158 
1159 	vp = ap->a_vp;
1160 	pfs = VTOPFS(vp);
1161 
1162 	if (uio->uio_resid < UIO_MX)
1163 		return (EINVAL);
1164 	if (uio->uio_offset < 0)
1165 		return (EINVAL);
1166 
1167 	error = 0;
1168 	i = uio->uio_offset;
1169 	memset(&d, 0, UIO_MX);
1170 	d.d_reclen = UIO_MX;
1171 	ncookies = uio->uio_resid / UIO_MX;
1172 
1173 	switch (pfs->pfs_type) {
1174 	/*
1175 	 * this is for the process-specific sub-directories.
1176 	 * all that is needed to is copy out all the entries
1177 	 * from the procent[] table (top of this file).
1178 	 */
1179 	case PFSproc: {
1180 		struct proc *p;
1181 
1182 		if (i >= nproc_targets)
1183 			return 0;
1184 
1185 		p = PFIND(pfs->pfs_pid);
1186 		if (p == NULL)
1187 			break;
1188 
1189 		if (ap->a_ncookies) {
1190 			ncookies = min(ncookies, (nproc_targets - i));
1191 			cookies = malloc(ncookies * sizeof (off_t),
1192 			    M_TEMP, M_WAITOK);
1193 			*ap->a_cookies = cookies;
1194 		}
1195 
1196 		for (pt = &proc_targets[i];
1197 		     uio->uio_resid >= UIO_MX && i < nproc_targets; pt++, i++) {
1198 			if (pt->pt_valid &&
1199 			    (*pt->pt_valid)(proc_representative_lwp(p), vp->v_mount) == 0)
1200 				continue;
1201 
1202 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid,
1203 			    pt->pt_pfstype, -1);
1204 			d.d_namlen = pt->pt_namlen;
1205 			memcpy(d.d_name, pt->pt_name, pt->pt_namlen + 1);
1206 			d.d_type = pt->pt_type;
1207 
1208 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1209 				break;
1210 			if (cookies)
1211 				*cookies++ = i + 1;
1212 		}
1213 
1214 	    	break;
1215 	}
1216 	case PFSfd: {
1217 		struct proc *p;
1218 		struct filedesc	*fdp;
1219 		struct file *fp;
1220 		int lim, nc = 0;
1221 
1222 		p = PFIND(pfs->pfs_pid);
1223 		if (p == NULL)
1224 			return ESRCH;
1225 
1226 		if (kauth_authorize_process(kauth_cred_get(),
1227 		    KAUTH_PROCESS_CANSEE, p, NULL, NULL, NULL) != 0)
1228 			return ESRCH;
1229 
1230 		fdp = p->p_fd;
1231 
1232 		lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfiles);
1233 		if (i >= lim)
1234 			return 0;
1235 
1236 		if (ap->a_ncookies) {
1237 			ncookies = min(ncookies, (fdp->fd_nfiles + 2 - i));
1238 			cookies = malloc(ncookies * sizeof (off_t),
1239 			    M_TEMP, M_WAITOK);
1240 			*ap->a_cookies = cookies;
1241 		}
1242 
1243 		for (; i < 2 && uio->uio_resid >= UIO_MX; i++) {
1244 			pt = &proc_targets[i];
1245 			d.d_namlen = pt->pt_namlen;
1246 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid,
1247 			    pt->pt_pfstype, -1);
1248 			(void)memcpy(d.d_name, pt->pt_name, pt->pt_namlen + 1);
1249 			d.d_type = pt->pt_type;
1250 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1251 				break;
1252 			if (cookies)
1253 				*cookies++ = i + 1;
1254 			nc++;
1255 		}
1256 		if (error) {
1257 			ncookies = nc;
1258 			break;
1259 		}
1260 		for (; uio->uio_resid >= UIO_MX && i < fdp->fd_nfiles; i++) {
1261 			/* check the descriptor exists */
1262 			if ((fp = fd_getfile(fdp, i - 2)) == NULL)
1263 				continue;
1264 			simple_unlock(&fp->f_slock);
1265 
1266 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid, PFSfd, i - 2);
1267 			d.d_namlen = snprintf(d.d_name, sizeof(d.d_name),
1268 			    "%lld", (long long)(i - 2));
1269 			d.d_type = VREG;
1270 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1271 				break;
1272 			if (cookies)
1273 				*cookies++ = i + 1;
1274 			nc++;
1275 		}
1276 		ncookies = nc;
1277 		break;
1278 	}
1279 
1280 	/*
1281 	 * this is for the root of the procfs filesystem
1282 	 * what is needed are special entries for "curproc"
1283 	 * and "self" followed by an entry for each process
1284 	 * on allproc.
1285 	 */
1286 
1287 	case PFSroot: {
1288 		int nc = 0;
1289 
1290 		if (ap->a_ncookies) {
1291 			/*
1292 			 * XXX Potentially allocating too much space here,
1293 			 * but I'm lazy. This loop needs some work.
1294 			 */
1295 			cookies = malloc(ncookies * sizeof (off_t),
1296 			    M_TEMP, M_WAITOK);
1297 			*ap->a_cookies = cookies;
1298 		}
1299 		error = 0;
1300 		/* 0 ... 3 are static entries. */
1301 		for (; i <= 3 && uio->uio_resid >= UIO_MX; i++) {
1302 			switch (i) {
1303 			case 0:		/* `.' */
1304 			case 1:		/* `..' */
1305 				d.d_fileno = PROCFS_FILENO(0, PFSroot, -1);
1306 				d.d_namlen = i + 1;
1307 				memcpy(d.d_name, "..", d.d_namlen);
1308 				d.d_name[i + 1] = '\0';
1309 				d.d_type = DT_DIR;
1310 				break;
1311 
1312 			case 2:
1313 				d.d_fileno = PROCFS_FILENO(0, PFScurproc, -1);
1314 				d.d_namlen = sizeof("curproc") - 1;
1315 				memcpy(d.d_name, "curproc", sizeof("curproc"));
1316 				d.d_type = DT_LNK;
1317 				break;
1318 
1319 			case 3:
1320 				d.d_fileno = PROCFS_FILENO(0, PFSself, -1);
1321 				d.d_namlen = sizeof("self") - 1;
1322 				memcpy(d.d_name, "self", sizeof("self"));
1323 				d.d_type = DT_LNK;
1324 				break;
1325 			}
1326 
1327 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1328 				break;
1329 			nc++;
1330 			if (cookies)
1331 				*cookies++ = i + 1;
1332 		}
1333 		/* 4 ... are process entries. */
1334 		ctx.uiop = uio;
1335 		ctx.error = 0;
1336 		ctx.off = 4;
1337 		ctx.startoff = i;
1338 		ctx.cookies = cookies;
1339 		ctx.ncookies = nc;
1340 		proclist_foreach_call(&allproc,
1341 		    procfs_root_readdir_callback, &ctx);
1342 		cookies = ctx.cookies;
1343 		nc = ctx.ncookies;
1344 		error = ctx.error;
1345 		if (error)
1346 			break;
1347 
1348 		/* misc entries. */
1349 		if (i < ctx.off)
1350 			i = ctx.off;
1351 		if (i >= ctx.off + nproc_root_targets)
1352 			break;
1353 		for (pt = &proc_root_targets[i - ctx.off];
1354 		    uio->uio_resid >= UIO_MX &&
1355 		    pt < &proc_root_targets[nproc_root_targets];
1356 		    pt++, i++) {
1357 			if (pt->pt_valid &&
1358 			    (*pt->pt_valid)(NULL, vp->v_mount) == 0)
1359 				continue;
1360 			d.d_fileno = PROCFS_FILENO(0, pt->pt_pfstype, -1);
1361 			d.d_namlen = pt->pt_namlen;
1362 			memcpy(d.d_name, pt->pt_name, pt->pt_namlen + 1);
1363 			d.d_type = pt->pt_type;
1364 
1365 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1366 				break;
1367 			nc++;
1368 			if (cookies)
1369 				*cookies++ = i + 1;
1370 		}
1371 
1372 		ncookies = nc;
1373 		break;
1374 	}
1375 
1376 	default:
1377 		error = ENOTDIR;
1378 		break;
1379 	}
1380 
1381 	if (ap->a_ncookies) {
1382 		if (error) {
1383 			if (cookies)
1384 				free(*ap->a_cookies, M_TEMP);
1385 			*ap->a_ncookies = 0;
1386 			*ap->a_cookies = NULL;
1387 		} else
1388 			*ap->a_ncookies = ncookies;
1389 	}
1390 	uio->uio_offset = i;
1391 	return (error);
1392 }
1393 
1394 /*
1395  * readlink reads the link of `curproc' and others
1396  */
1397 int
1398 procfs_readlink(v)
1399 	void *v;
1400 {
1401 	struct vop_readlink_args *ap = v;
1402 	char bf[16];		/* should be enough */
1403 	char *bp = bf;
1404 	char *path = NULL;
1405 	int len;
1406 	int error = 0;
1407 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
1408 	struct proc *pown;
1409 
1410 	if (pfs->pfs_fileno == PROCFS_FILENO(0, PFScurproc, -1))
1411 		len = snprintf(bf, sizeof(bf), "%ld", (long)curproc->p_pid);
1412 	else if (pfs->pfs_fileno == PROCFS_FILENO(0, PFSself, -1))
1413 		len = snprintf(bf, sizeof(bf), "%s", "curproc");
1414 	else if (pfs->pfs_fileno == PROCFS_FILENO(pfs->pfs_pid, PFScwd, -1)) {
1415 		pown = PFIND(pfs->pfs_pid);
1416 		if (pown == NULL)
1417 			return (ESRCH);
1418 		MALLOC(path, char *, MAXPATHLEN + 4, M_TEMP,
1419 		    M_WAITOK|M_CANFAIL);
1420 		if (path == NULL)
1421 			return (ENOMEM);
1422 		bp = path + MAXPATHLEN;
1423 		*--bp = '\0';
1424 		(void)procfs_dir(PFScwd, curlwp, pown, &bp, path,
1425 		     MAXPATHLEN);
1426 		len = strlen(bp);
1427 	}
1428 	else if (pfs->pfs_fileno == PROCFS_FILENO(pfs->pfs_pid, PFSchroot, -1)) {
1429 		pown = PFIND(pfs->pfs_pid);
1430 		if (pown == NULL)
1431 			return (ESRCH);
1432 		MALLOC(path, char *, MAXPATHLEN + 4, M_TEMP,
1433 		    M_WAITOK|M_CANFAIL);
1434 		if (path == NULL)
1435 			return (ENOMEM);
1436 		bp = path + MAXPATHLEN;
1437 		*--bp = '\0';
1438 		(void)procfs_dir(PFSchroot, curlwp, pown, &bp, path,
1439 		     MAXPATHLEN);
1440 		len = strlen(bp);
1441 	}
1442 	else {
1443 		struct file *fp;
1444 		struct vnode *vxp, *vp;
1445 
1446 		if ((error = procfs_getfp(pfs, &pown, &fp)) != 0)
1447 			return error;
1448 		FILE_USE(fp);
1449 		switch (fp->f_type) {
1450 		case DTYPE_VNODE:
1451 			vxp = (struct vnode *)fp->f_data;
1452 			if (vxp->v_type != VDIR) {
1453 				FILE_UNUSE(fp, proc_representative_lwp(pown));
1454 				return EINVAL;
1455 			}
1456 			if ((path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK))
1457 			    == NULL) {
1458 				FILE_UNUSE(fp, proc_representative_lwp(pown));
1459 				return ENOMEM;
1460 			}
1461 			bp = path + MAXPATHLEN;
1462 			*--bp = '\0';
1463 			vp = curproc->p_cwdi->cwdi_rdir;
1464 			if (vp == NULL)
1465 				vp = rootvnode;
1466 			error = getcwd_common(vxp, vp, &bp, path,
1467 			    MAXPATHLEN / 2, 0, curlwp);
1468 			FILE_UNUSE(fp, proc_representative_lwp(pown));
1469 			if (error) {
1470 				free(path, M_TEMP);
1471 				return error;
1472 			}
1473 			len = strlen(bp);
1474 			break;
1475 
1476 		case DTYPE_MISC:
1477 			len = snprintf(bf, sizeof(bf), "%s", "[misc]");
1478 			break;
1479 
1480 		case DTYPE_KQUEUE:
1481 			len = snprintf(bf, sizeof(bf), "%s", "[kqueue]");
1482 			break;
1483 
1484 		default:
1485 			return EINVAL;
1486 		}
1487 	}
1488 
1489 	error = uiomove(bp, len, ap->a_uio);
1490 	if (path)
1491 		free(path, M_TEMP);
1492 	return error;
1493 }
1494 
1495 /*
1496  * convert decimal ascii to int
1497  */
1498 static int
1499 atoi(b, len)
1500 	const char *b;
1501 	size_t len;
1502 {
1503 	int p = 0;
1504 
1505 	while (len--) {
1506 		char c = *b++;
1507 		if (c < '0' || c > '9')
1508 			return -1;
1509 		p = 10 * p + (c - '0');
1510 	}
1511 
1512 	return p;
1513 }
1514