xref: /netbsd-src/sys/miscfs/procfs/procfs_vnops.c (revision b7ae68fde0d8ef1c03714e8bbb1ee7c6118ea93b)
1 /*	$NetBSD: procfs_vnops.c,v 1.134 2006/09/20 08:09:05 manu 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 1.134 2006/09/20 08:09:05 manu 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 	{ DT_REG, N("devices"),     PFSdevices,        procfs_validfile_linux },
167 #undef N
168 };
169 static const int nproc_root_targets =
170     sizeof(proc_root_targets) / sizeof(proc_root_targets[0]);
171 
172 int	procfs_lookup(void *);
173 #define	procfs_create	genfs_eopnotsupp
174 #define	procfs_mknod	genfs_eopnotsupp
175 int	procfs_open(void *);
176 int	procfs_close(void *);
177 int	procfs_access(void *);
178 int	procfs_getattr(void *);
179 int	procfs_setattr(void *);
180 #define	procfs_read	procfs_rw
181 #define	procfs_write	procfs_rw
182 #define	procfs_fcntl	genfs_fcntl
183 #define	procfs_ioctl	genfs_enoioctl
184 #define	procfs_poll	genfs_poll
185 #define procfs_revoke	genfs_revoke
186 #define	procfs_fsync	genfs_nullop
187 #define	procfs_seek	genfs_nullop
188 #define	procfs_remove	genfs_eopnotsupp
189 int	procfs_link(void *);
190 #define	procfs_rename	genfs_eopnotsupp
191 #define	procfs_mkdir	genfs_eopnotsupp
192 #define	procfs_rmdir	genfs_eopnotsupp
193 int	procfs_symlink(void *);
194 int	procfs_readdir(void *);
195 int	procfs_readlink(void *);
196 #define	procfs_abortop	genfs_abortop
197 int	procfs_inactive(void *);
198 int	procfs_reclaim(void *);
199 #define	procfs_lock	genfs_lock
200 #define	procfs_unlock	genfs_unlock
201 #define	procfs_bmap	genfs_badop
202 #define	procfs_strategy	genfs_badop
203 int	procfs_print(void *);
204 int	procfs_pathconf(void *);
205 #define	procfs_islocked	genfs_islocked
206 #define	procfs_advlock	genfs_einval
207 #define	procfs_bwrite	genfs_eopnotsupp
208 #define procfs_putpages	genfs_null_putpages
209 
210 static int atoi(const char *, size_t);
211 
212 /*
213  * procfs vnode operations.
214  */
215 int (**procfs_vnodeop_p)(void *);
216 const struct vnodeopv_entry_desc procfs_vnodeop_entries[] = {
217 	{ &vop_default_desc, vn_default_error },
218 	{ &vop_lookup_desc, procfs_lookup },		/* lookup */
219 	{ &vop_create_desc, procfs_create },		/* create */
220 	{ &vop_mknod_desc, procfs_mknod },		/* mknod */
221 	{ &vop_open_desc, procfs_open },		/* open */
222 	{ &vop_close_desc, procfs_close },		/* close */
223 	{ &vop_access_desc, procfs_access },		/* access */
224 	{ &vop_getattr_desc, procfs_getattr },		/* getattr */
225 	{ &vop_setattr_desc, procfs_setattr },		/* setattr */
226 	{ &vop_read_desc, procfs_read },		/* read */
227 	{ &vop_write_desc, procfs_write },		/* write */
228 	{ &vop_fcntl_desc, procfs_fcntl },		/* fcntl */
229 	{ &vop_ioctl_desc, procfs_ioctl },		/* ioctl */
230 	{ &vop_poll_desc, procfs_poll },		/* poll */
231 	{ &vop_revoke_desc, procfs_revoke },		/* revoke */
232 	{ &vop_fsync_desc, procfs_fsync },		/* fsync */
233 	{ &vop_seek_desc, procfs_seek },		/* seek */
234 	{ &vop_remove_desc, procfs_remove },		/* remove */
235 	{ &vop_link_desc, procfs_link },		/* link */
236 	{ &vop_rename_desc, procfs_rename },		/* rename */
237 	{ &vop_mkdir_desc, procfs_mkdir },		/* mkdir */
238 	{ &vop_rmdir_desc, procfs_rmdir },		/* rmdir */
239 	{ &vop_symlink_desc, procfs_symlink },		/* symlink */
240 	{ &vop_readdir_desc, procfs_readdir },		/* readdir */
241 	{ &vop_readlink_desc, procfs_readlink },	/* readlink */
242 	{ &vop_abortop_desc, procfs_abortop },		/* abortop */
243 	{ &vop_inactive_desc, procfs_inactive },	/* inactive */
244 	{ &vop_reclaim_desc, procfs_reclaim },		/* reclaim */
245 	{ &vop_lock_desc, procfs_lock },		/* lock */
246 	{ &vop_unlock_desc, procfs_unlock },		/* unlock */
247 	{ &vop_bmap_desc, procfs_bmap },		/* bmap */
248 	{ &vop_strategy_desc, procfs_strategy },	/* strategy */
249 	{ &vop_print_desc, procfs_print },		/* print */
250 	{ &vop_islocked_desc, procfs_islocked },	/* islocked */
251 	{ &vop_pathconf_desc, procfs_pathconf },	/* pathconf */
252 	{ &vop_advlock_desc, procfs_advlock },		/* advlock */
253 	{ &vop_putpages_desc, procfs_putpages },	/* putpages */
254 	{ NULL, NULL }
255 };
256 const struct vnodeopv_desc procfs_vnodeop_opv_desc =
257 	{ &procfs_vnodeop_p, procfs_vnodeop_entries };
258 /*
259  * set things up for doing i/o on
260  * the pfsnode (vp).  (vp) is locked
261  * on entry, and should be left locked
262  * on exit.
263  *
264  * for procfs we don't need to do anything
265  * in particular for i/o.  all that is done
266  * is to support exclusive open on process
267  * memory images.
268  */
269 int
270 procfs_open(v)
271 	void *v;
272 {
273 	struct vop_open_args /* {
274 		struct vnode *a_vp;
275 		int  a_mode;
276 		kauth_cred_t a_cred;
277 		struct lwp *a_l;
278 	} */ *ap = v;
279 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
280 	struct lwp *l1;
281 	struct proc *p2;
282 	int error;
283 
284 	l1 = ap->a_l;				/* tracer */
285 	p2 = PFIND(pfs->pfs_pid);		/* traced */
286 
287 	if (p2 == NULL)
288 		return (ENOENT);		/* was ESRCH, jsp */
289 
290 	switch (pfs->pfs_type) {
291 	case PFSmem:
292 		if (((pfs->pfs_flags & FWRITE) && (ap->a_mode & O_EXCL)) ||
293 		    ((pfs->pfs_flags & O_EXCL) && (ap->a_mode & FWRITE)))
294 			return (EBUSY);
295 
296 		if ((error = process_checkioperm(l1, p2)) != 0)
297 			return (error);
298 
299 		if (ap->a_mode & FWRITE)
300 			pfs->pfs_flags = ap->a_mode & (FWRITE|O_EXCL);
301 
302 		return (0);
303 
304 	default:
305 		break;
306 	}
307 
308 	return (0);
309 }
310 
311 /*
312  * close the pfsnode (vp) after doing i/o.
313  * (vp) is not locked on entry or exit.
314  *
315  * nothing to do for procfs other than undo
316  * any exclusive open flag (see _open above).
317  */
318 int
319 procfs_close(v)
320 	void *v;
321 {
322 	struct vop_close_args /* {
323 		struct vnode *a_vp;
324 		int  a_fflag;
325 		kauth_cred_t a_cred;
326 		struct lwp *a_l;
327 	} */ *ap = v;
328 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
329 
330 	switch (pfs->pfs_type) {
331 	case PFSmem:
332 		if ((ap->a_fflag & FWRITE) && (pfs->pfs_flags & O_EXCL))
333 			pfs->pfs_flags &= ~(FWRITE|O_EXCL);
334 		break;
335 
336 	default:
337 		break;
338 	}
339 
340 	return (0);
341 }
342 
343 /*
344  * _inactive is called when the pfsnode
345  * is vrele'd and the reference count goes
346  * to zero.  (vp) will be on the vnode free
347  * list, so to get it back vget() must be
348  * used.
349  *
350  * for procfs, check if the process is still
351  * alive and if it isn't then just throw away
352  * the vnode by calling vgone().  this may
353  * be overkill and a waste of time since the
354  * chances are that the process will still be
355  * there and PFIND is not free.
356  *
357  * (vp) is locked on entry, but must be unlocked on exit.
358  */
359 int
360 procfs_inactive(v)
361 	void *v;
362 {
363 	struct vop_inactive_args /* {
364 		struct vnode *a_vp;
365 		struct proc *a_p;
366 	} */ *ap = v;
367 	struct vnode *vp = ap->a_vp;
368 	struct pfsnode *pfs = VTOPFS(vp);
369 
370 	VOP_UNLOCK(vp, 0);
371 	if (PFIND(pfs->pfs_pid) == NULL && (vp->v_flag & VXLOCK) == 0)
372 		vgone(vp);
373 
374 	return (0);
375 }
376 
377 /*
378  * _reclaim is called when getnewvnode()
379  * wants to make use of an entry on the vnode
380  * free list.  at this time the filesystem needs
381  * to free any private data and remove the node
382  * from any private lists.
383  */
384 int
385 procfs_reclaim(v)
386 	void *v;
387 {
388 	struct vop_reclaim_args /* {
389 		struct vnode *a_vp;
390 	} */ *ap = v;
391 
392 	return (procfs_freevp(ap->a_vp));
393 }
394 
395 /*
396  * Return POSIX pathconf information applicable to special devices.
397  */
398 int
399 procfs_pathconf(v)
400 	void *v;
401 {
402 	struct vop_pathconf_args /* {
403 		struct vnode *a_vp;
404 		int a_name;
405 		register_t *a_retval;
406 	} */ *ap = v;
407 
408 	switch (ap->a_name) {
409 	case _PC_LINK_MAX:
410 		*ap->a_retval = LINK_MAX;
411 		return (0);
412 	case _PC_MAX_CANON:
413 		*ap->a_retval = MAX_CANON;
414 		return (0);
415 	case _PC_MAX_INPUT:
416 		*ap->a_retval = MAX_INPUT;
417 		return (0);
418 	case _PC_PIPE_BUF:
419 		*ap->a_retval = PIPE_BUF;
420 		return (0);
421 	case _PC_CHOWN_RESTRICTED:
422 		*ap->a_retval = 1;
423 		return (0);
424 	case _PC_VDISABLE:
425 		*ap->a_retval = _POSIX_VDISABLE;
426 		return (0);
427 	case _PC_SYNC_IO:
428 		*ap->a_retval = 1;
429 		return (0);
430 	default:
431 		return (EINVAL);
432 	}
433 	/* NOTREACHED */
434 }
435 
436 /*
437  * _print is used for debugging.
438  * just print a readable description
439  * of (vp).
440  */
441 int
442 procfs_print(v)
443 	void *v;
444 {
445 	struct vop_print_args /* {
446 		struct vnode *a_vp;
447 	} */ *ap = v;
448 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
449 
450 	printf("tag VT_PROCFS, type %d, pid %d, mode %x, flags %lx\n",
451 	    pfs->pfs_type, pfs->pfs_pid, pfs->pfs_mode, pfs->pfs_flags);
452 	return 0;
453 }
454 
455 int
456 procfs_link(v)
457 	void *v;
458 {
459 	struct vop_link_args /* {
460 		struct vnode *a_dvp;
461 		struct vnode *a_vp;
462 		struct componentname *a_cnp;
463 	} */ *ap = v;
464 
465 	VOP_ABORTOP(ap->a_dvp, ap->a_cnp);
466 	vput(ap->a_dvp);
467 	return (EROFS);
468 }
469 
470 int
471 procfs_symlink(v)
472 	void *v;
473 {
474 	struct vop_symlink_args /* {
475 		struct vnode *a_dvp;
476 		struct vnode **a_vpp;
477 		struct componentname *a_cnp;
478 		struct vattr *a_vap;
479 		char *a_target;
480 	} */ *ap = v;
481 
482 	VOP_ABORTOP(ap->a_dvp, ap->a_cnp);
483 	vput(ap->a_dvp);
484 	return (EROFS);
485 }
486 
487 /*
488  * Works out the path to (and vnode of) the target process's current
489  * working directory or chroot.  If the caller is in a chroot and
490  * can't "reach" the target's cwd or root (or some other error
491  * occurs), a "/" is returned for the path and a NULL pointer is
492  * returned for the vnode.
493  */
494 static struct vnode *
495 procfs_dir(pfstype t, struct lwp *caller, struct proc *target,
496 	   char **bpp, char *path, int len)
497 {
498 	struct vnode *vp, *rvp = caller->l_proc->p_cwdi->cwdi_rdir;
499 	char *bp;
500 
501 	bp = bpp ? *bpp : NULL;
502 
503 	switch (t) {
504 	case PFScwd:
505 		vp = target->p_cwdi->cwdi_cdir;
506 		break;
507 	case PFSchroot:
508 		vp = target->p_cwdi->cwdi_rdir;
509 		break;
510 	default:
511 		return (NULL);
512 	}
513 
514 	if (rvp == NULL)
515 		rvp = rootvnode;
516 	if (vp == NULL || getcwd_common(vp, rvp, bp ? &bp : NULL, path,
517 	    len / 2, 0, caller) != 0) {
518 		vp = NULL;
519 		if (bpp) {
520 			bp = *bpp;
521 			*--bp = '/';
522 		}
523 	}
524 
525 	if (bpp)
526 		*bpp = bp;
527 
528 	return (vp);
529 }
530 
531 /*
532  * Invent attributes for pfsnode (vp) and store
533  * them in (vap).
534  * Directories lengths are returned as zero since
535  * any real length would require the genuine size
536  * to be computed, and nothing cares anyway.
537  *
538  * this is relatively minimal for procfs.
539  */
540 int
541 procfs_getattr(v)
542 	void *v;
543 {
544 	struct vop_getattr_args /* {
545 		struct vnode *a_vp;
546 		struct vattr *a_vap;
547 		kauth_cred_t a_cred;
548 		struct lwp *a_l;
549 	} */ *ap = v;
550 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
551 	struct vattr *vap = ap->a_vap;
552 	struct proc *procp;
553 	int error;
554 
555 	/* first check the process still exists */
556 	switch (pfs->pfs_type) {
557 	case PFSroot:
558 	case PFScurproc:
559 	case PFSself:
560 		procp = 0;
561 		break;
562 
563 	default:
564 		procp = PFIND(pfs->pfs_pid);
565 		if (procp == NULL)
566 			return (ENOENT);
567 		break;
568 	}
569 
570 	if (procp != NULL) {
571 		if (kauth_authorize_process(kauth_cred_get(),
572 		    KAUTH_PROCESS_CANSEE, procp, NULL, NULL, NULL) != 0)
573 			return (ENOENT);
574 	}
575 
576 	error = 0;
577 
578 	/* start by zeroing out the attributes */
579 	VATTR_NULL(vap);
580 
581 	/* next do all the common fields */
582 	vap->va_type = ap->a_vp->v_type;
583 	vap->va_mode = pfs->pfs_mode;
584 	vap->va_fileid = pfs->pfs_fileno;
585 	vap->va_flags = 0;
586 	vap->va_blocksize = PAGE_SIZE;
587 
588 	/*
589 	 * Make all times be current TOD.
590 	 *
591 	 * It would be possible to get the process start
592 	 * time from the p_stats structure, but there's
593 	 * no "file creation" time stamp anyway, and the
594 	 * p_stats structure is not addressable if u. gets
595 	 * swapped out for that process.
596 	 */
597 	getnanotime(&vap->va_ctime);
598 	vap->va_atime = vap->va_mtime = vap->va_ctime;
599 	if (procp)
600 		TIMEVAL_TO_TIMESPEC(&procp->p_stats->p_start,
601 		    &vap->va_birthtime);
602 	else
603 		getnanotime(&vap->va_birthtime);
604 
605 	switch (pfs->pfs_type) {
606 	case PFSmem:
607 	case PFSregs:
608 	case PFSfpregs:
609 #if defined(__HAVE_PROCFS_MACHDEP) && defined(PROCFS_MACHDEP_PROTECT_CASES)
610 	PROCFS_MACHDEP_PROTECT_CASES
611 #endif
612 		/*
613 		 * If the process has exercised some setuid or setgid
614 		 * privilege, then rip away read/write permission so
615 		 * that only root can gain access.
616 		 */
617 		if (procp->p_flag & P_SUGID)
618 			vap->va_mode &= ~(S_IRUSR|S_IWUSR);
619 		/* FALLTHROUGH */
620 	case PFSctl:
621 	case PFSstatus:
622 	case PFSstat:
623 	case PFSnote:
624 	case PFSnotepg:
625 	case PFSmap:
626 	case PFSmaps:
627 	case PFScmdline:
628 		vap->va_nlink = 1;
629 		vap->va_uid = kauth_cred_geteuid(procp->p_cred);
630 		vap->va_gid = kauth_cred_getegid(procp->p_cred);
631 		break;
632 	case PFSmeminfo:
633 	case PFSdevices:
634 	case PFScpuinfo:
635 	case PFSuptime:
636 	case PFSmounts:
637 		vap->va_nlink = 1;
638 		vap->va_uid = vap->va_gid = 0;
639 		break;
640 
641 	default:
642 		break;
643 	}
644 
645 	/*
646 	 * now do the object specific fields
647 	 *
648 	 * The size could be set from struct reg, but it's hardly
649 	 * worth the trouble, and it puts some (potentially) machine
650 	 * dependent data into this machine-independent code.  If it
651 	 * becomes important then this function should break out into
652 	 * a per-file stat function in the corresponding .c file.
653 	 */
654 
655 	switch (pfs->pfs_type) {
656 	case PFSroot:
657 		/*
658 		 * Set nlink to 1 to tell fts(3) we don't actually know.
659 		 */
660 		vap->va_nlink = 1;
661 		vap->va_uid = 0;
662 		vap->va_gid = 0;
663 		vap->va_bytes = vap->va_size = DEV_BSIZE;
664 		break;
665 
666 	case PFSself:
667 	case PFScurproc: {
668 		char bf[16];		/* should be enough */
669 		vap->va_nlink = 1;
670 		vap->va_uid = 0;
671 		vap->va_gid = 0;
672 		vap->va_bytes = vap->va_size =
673 		    snprintf(bf, sizeof(bf), "%ld", (long)curproc->p_pid);
674 		break;
675 	}
676 
677 	case PFSfd:
678 		if (pfs->pfs_fd != -1) {
679 			struct file *fp;
680 			struct proc *pown;
681 
682 			if ((error = procfs_getfp(pfs, &pown, &fp)) != 0)
683 				return error;
684 			FILE_USE(fp);
685 			vap->va_nlink = 1;
686 			vap->va_uid = kauth_cred_geteuid(fp->f_cred);
687 			vap->va_gid = kauth_cred_getegid(fp->f_cred);
688 			switch (fp->f_type) {
689 			case DTYPE_VNODE:
690 				vap->va_bytes = vap->va_size =
691 				    ((struct vnode *)fp->f_data)->v_size;
692 				break;
693 			default:
694 				vap->va_bytes = vap->va_size = 0;
695 				break;
696 			}
697 			FILE_UNUSE(fp, proc_representative_lwp(pown));
698 			break;
699 		}
700 		/*FALLTHROUGH*/
701 	case PFSproc:
702 		vap->va_nlink = 2;
703 		vap->va_uid = kauth_cred_geteuid(procp->p_cred);
704 		vap->va_gid = kauth_cred_getegid(procp->p_cred);
705 		vap->va_bytes = vap->va_size = DEV_BSIZE;
706 		break;
707 
708 	case PFSfile:
709 		error = EOPNOTSUPP;
710 		break;
711 
712 	case PFSmem:
713 		vap->va_bytes = vap->va_size =
714 			ctob(procp->p_vmspace->vm_tsize +
715 				    procp->p_vmspace->vm_dsize +
716 				    procp->p_vmspace->vm_ssize);
717 		break;
718 
719 #if defined(PT_GETREGS) || defined(PT_SETREGS)
720 	case PFSregs:
721 		vap->va_bytes = vap->va_size = sizeof(struct reg);
722 		break;
723 #endif
724 
725 #if defined(PT_GETFPREGS) || defined(PT_SETFPREGS)
726 	case PFSfpregs:
727 		vap->va_bytes = vap->va_size = sizeof(struct fpreg);
728 		break;
729 #endif
730 
731 	case PFSctl:
732 	case PFSstatus:
733 	case PFSstat:
734 	case PFSnote:
735 	case PFSnotepg:
736 	case PFScmdline:
737 	case PFSmeminfo:
738 	case PFSdevices:
739 	case PFScpuinfo:
740 	case PFSuptime:
741 	case PFSmounts:
742 		vap->va_bytes = vap->va_size = 0;
743 		break;
744 	case PFSmap:
745 	case PFSmaps:
746 		/*
747 		 * Advise a larger blocksize for the map files, so that
748 		 * they may be read in one pass.
749 		 */
750 		vap->va_blocksize = 4 * PAGE_SIZE;
751 		vap->va_bytes = vap->va_size = 0;
752 		break;
753 
754 	case PFScwd:
755 	case PFSchroot: {
756 		char *path, *bp;
757 
758 		MALLOC(path, char *, MAXPATHLEN + 4, M_TEMP,
759 		    M_WAITOK|M_CANFAIL);
760 		if (path == NULL)
761 			return (ENOMEM);
762 		vap->va_nlink = 1;
763 		vap->va_uid = 0;
764 		vap->va_gid = 0;
765 		bp = path + MAXPATHLEN;
766 		*--bp = '\0';
767 		(void)procfs_dir(pfs->pfs_type, curlwp, procp, &bp, path,
768 		     MAXPATHLEN);
769 		vap->va_bytes = vap->va_size = strlen(bp);
770 		free(path, M_TEMP);
771 		break;
772 	}
773 
774 #ifdef __HAVE_PROCFS_MACHDEP
775 	PROCFS_MACHDEP_NODETYPE_CASES
776 		error = procfs_machdep_getattr(ap->a_vp, vap, procp);
777 		break;
778 #endif
779 
780 	default:
781 		panic("procfs_getattr");
782 	}
783 
784 	return (error);
785 }
786 
787 /*ARGSUSED*/
788 int
789 procfs_setattr(v)
790 	void *v;
791 {
792 	/*
793 	 * just fake out attribute setting
794 	 * it's not good to generate an error
795 	 * return, otherwise things like creat()
796 	 * will fail when they try to set the
797 	 * file length to 0.  worse, this means
798 	 * that echo $note > /proc/$pid/note will fail.
799 	 */
800 
801 	return (0);
802 }
803 
804 /*
805  * implement access checking.
806  *
807  * actually, the check for super-user is slightly
808  * broken since it will allow read access to write-only
809  * objects.  this doesn't cause any particular trouble
810  * but does mean that the i/o entry points need to check
811  * that the operation really does make sense.
812  */
813 int
814 procfs_access(v)
815 	void *v;
816 {
817 	struct vop_access_args /* {
818 		struct vnode *a_vp;
819 		int a_mode;
820 		kauth_cred_t a_cred;
821 		struct lwp *a_l;
822 	} */ *ap = v;
823 	struct vattr va;
824 	int error;
825 
826 	if ((error = VOP_GETATTR(ap->a_vp, &va, ap->a_cred, ap->a_l)) != 0)
827 		return (error);
828 
829 	return (vaccess(va.va_type, va.va_mode,
830 	    va.va_uid, va.va_gid, ap->a_mode, ap->a_cred));
831 }
832 
833 /*
834  * lookup.  this is incredibly complicated in the
835  * general case, however for most pseudo-filesystems
836  * very little needs to be done.
837  *
838  * Locking isn't hard here, just poorly documented.
839  *
840  * If we're looking up ".", just vref the parent & return it.
841  *
842  * If we're looking up "..", unlock the parent, and lock "..". If everything
843  * went ok, and we're on the last component and the caller requested the
844  * parent locked, try to re-lock the parent. We do this to prevent lock
845  * races.
846  *
847  * For anything else, get the needed node. Then unlock the parent if not
848  * the last component or not LOCKPARENT (i.e. if we wouldn't re-lock the
849  * parent in the .. case).
850  *
851  * We try to exit with the parent locked in error cases.
852  */
853 int
854 procfs_lookup(v)
855 	void *v;
856 {
857 	struct vop_lookup_args /* {
858 		struct vnode * a_dvp;
859 		struct vnode ** a_vpp;
860 		struct componentname * a_cnp;
861 	} */ *ap = v;
862 	struct componentname *cnp = ap->a_cnp;
863 	struct vnode **vpp = ap->a_vpp;
864 	struct vnode *dvp = ap->a_dvp;
865 	const char *pname = cnp->cn_nameptr;
866 	const struct proc_target *pt = NULL;
867 	struct vnode *fvp;
868 	pid_t pid;
869 	struct pfsnode *pfs;
870 	struct proc *p = NULL;
871 	struct lwp *l = NULL;
872 	int i, error, wantpunlock, iscurproc = 0, isself = 0;
873 
874 	*vpp = NULL;
875 	cnp->cn_flags &= ~PDIRUNLOCK;
876 
877 	if (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)
878 		return (EROFS);
879 
880 	if (cnp->cn_namelen == 1 && *pname == '.') {
881 		*vpp = dvp;
882 		VREF(dvp);
883 		return (0);
884 	}
885 
886 	wantpunlock = (~cnp->cn_flags & (LOCKPARENT | ISLASTCN));
887 	pfs = VTOPFS(dvp);
888 	switch (pfs->pfs_type) {
889 	case PFSroot:
890 		/*
891 		 * Shouldn't get here with .. in the root node.
892 		 */
893 		if (cnp->cn_flags & ISDOTDOT)
894 			return (EIO);
895 
896 		iscurproc = CNEQ(cnp, "curproc", 7);
897 		isself = CNEQ(cnp, "self", 4);
898 
899 		if (iscurproc || isself) {
900 			error = procfs_allocvp(dvp->v_mount, vpp, 0,
901 			    iscurproc ? PFScurproc : PFSself, -1);
902 			if ((error == 0) && (wantpunlock)) {
903 				VOP_UNLOCK(dvp, 0);
904 				cnp->cn_flags |= PDIRUNLOCK;
905 			}
906 			return (error);
907 		}
908 
909 		for (i = 0; i < nproc_root_targets; i++) {
910 			pt = &proc_root_targets[i];
911 			if (cnp->cn_namelen == pt->pt_namlen &&
912 			    memcmp(pt->pt_name, pname, cnp->cn_namelen) == 0 &&
913 			    (pt->pt_valid == NULL ||
914 			     (*pt->pt_valid)(cnp->cn_lwp, dvp->v_mount)))
915 				break;
916 		}
917 
918 		if (i != nproc_root_targets) {
919 			error = procfs_allocvp(dvp->v_mount, vpp, 0,
920 			    pt->pt_pfstype, -1);
921 			if ((error == 0) && (wantpunlock)) {
922 				VOP_UNLOCK(dvp, 0);
923 				cnp->cn_flags |= PDIRUNLOCK;
924 			}
925 			return (error);
926 		}
927 
928 		pid = (pid_t)atoi(pname, cnp->cn_namelen);
929 
930 		p = PFIND(pid);
931 		if (p == NULL)
932 			break;
933 
934 		error = procfs_allocvp(dvp->v_mount, vpp, pid, PFSproc, -1);
935 		if ((error == 0) && (wantpunlock)) {
936 			VOP_UNLOCK(dvp, 0);
937 			cnp->cn_flags |= PDIRUNLOCK;
938 		}
939 		return (error);
940 
941 	case PFSproc:
942 		/*
943 		 * do the .. dance. We unlock the directory, and then
944 		 * get the root dir. That will automatically return ..
945 		 * locked. Then if the caller wanted dvp locked, we
946 		 * re-lock.
947 		 */
948 		if (cnp->cn_flags & ISDOTDOT) {
949 			VOP_UNLOCK(dvp, 0);
950 			cnp->cn_flags |= PDIRUNLOCK;
951 			error = procfs_root(dvp->v_mount, vpp);
952 			if ((error == 0) && (wantpunlock == 0) &&
953 				    ((error = vn_lock(dvp, LK_EXCLUSIVE)) == 0))
954 				cnp->cn_flags &= ~PDIRUNLOCK;
955 			return (error);
956 		}
957 
958 		p = PFIND(pfs->pfs_pid);
959 		if (p == NULL)
960 			break;
961 		l = proc_representative_lwp(p);
962 
963 		for (pt = proc_targets, i = 0; i < nproc_targets; pt++, i++) {
964 			if (cnp->cn_namelen == pt->pt_namlen &&
965 			    memcmp(pt->pt_name, pname, cnp->cn_namelen) == 0 &&
966 			    (pt->pt_valid == NULL ||
967 			     (*pt->pt_valid)(cnp->cn_lwp, dvp->v_mount)))
968 				goto found;
969 		}
970 		break;
971 
972 	found:
973 		if (pt->pt_pfstype == PFSfile) {
974 			fvp = p->p_textvp;
975 			/* We already checked that it exists. */
976 			VREF(fvp);
977 			vn_lock(fvp, LK_EXCLUSIVE | LK_RETRY);
978 			if (wantpunlock) {
979 				VOP_UNLOCK(dvp, 0);
980 				cnp->cn_flags |= PDIRUNLOCK;
981 			}
982 			*vpp = fvp;
983 			return (0);
984 		}
985 
986 		error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
987 		    pt->pt_pfstype, -1);
988 		if ((error == 0) && (wantpunlock)) {
989 			VOP_UNLOCK(dvp, 0);
990 			cnp->cn_flags |= PDIRUNLOCK;
991 		}
992 		return (error);
993 
994 	case PFSfd: {
995 		int fd;
996 		struct file *fp;
997 		/*
998 		 * do the .. dance. We unlock the directory, and then
999 		 * get the proc dir. That will automatically return ..
1000 		 * locked. Then if the caller wanted dvp locked, we
1001 		 * re-lock.
1002 		 */
1003 		if (cnp->cn_flags & ISDOTDOT) {
1004 			VOP_UNLOCK(dvp, 0);
1005 			cnp->cn_flags |= PDIRUNLOCK;
1006 			error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
1007 			    PFSproc, -1);
1008 			if ((error == 0) && (wantpunlock == 0) &&
1009 				    ((error = vn_lock(dvp, LK_EXCLUSIVE)) == 0))
1010 				cnp->cn_flags &= ~PDIRUNLOCK;
1011 			return (error);
1012 		}
1013 		fd = atoi(pname, cnp->cn_namelen);
1014 		p = PFIND(pfs->pfs_pid);
1015 		if (p == NULL || (fp = fd_getfile(p->p_fd, fd)) == NULL)
1016 			return ENOENT;
1017 		FILE_USE(fp);
1018 
1019 		switch (fp->f_type) {
1020 		case DTYPE_VNODE:
1021 			fvp = (struct vnode *)fp->f_data;
1022 
1023 			/* Don't show directories */
1024 			if (fvp->v_type == VDIR)
1025 				goto symlink;
1026 
1027 			VREF(fvp);
1028 			FILE_UNUSE(fp, l);
1029 			vn_lock(fvp, LK_EXCLUSIVE | LK_RETRY |
1030 			    (p == curproc ? LK_CANRECURSE : 0));
1031 			*vpp = fvp;
1032 			error = 0;
1033 			break;
1034 		default:
1035 		symlink:
1036 			FILE_UNUSE(fp, l);
1037 			error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
1038 			    PFSfd, fd);
1039 			break;
1040 		}
1041 		if ((error == 0) && (wantpunlock)) {
1042 			VOP_UNLOCK(dvp, 0);
1043 			cnp->cn_flags |= PDIRUNLOCK;
1044 		}
1045 		return error;
1046 	}
1047 	default:
1048 		return (ENOTDIR);
1049 	}
1050 
1051 	return (cnp->cn_nameiop == LOOKUP ? ENOENT : EROFS);
1052 }
1053 
1054 int
1055 procfs_validfile(l, mp)
1056 	struct lwp *l;
1057 	struct mount *mp;
1058 {
1059 	return (l->l_proc->p_textvp != NULL);
1060 }
1061 
1062 static int
1063 procfs_validfile_linux(l, mp)
1064 	struct lwp *l;
1065 	struct mount *mp;
1066 {
1067 	int flags;
1068 
1069 	flags = VFSTOPROC(mp)->pmnt_flags;
1070 	return ((flags & PROCFSMNT_LINUXCOMPAT) &&
1071 	    (l == NULL || l->l_proc == NULL || procfs_validfile(l, mp)));
1072 }
1073 
1074 struct procfs_root_readdir_ctx {
1075 	struct uio *uiop;
1076 	off_t *cookies;
1077 	int ncookies;
1078 	off_t off;
1079 	off_t startoff;
1080 	int error;
1081 };
1082 
1083 static int
1084 procfs_root_readdir_callback(struct proc *p, void *arg)
1085 {
1086 	struct procfs_root_readdir_ctx *ctxp = arg;
1087 	struct dirent d;
1088 	struct uio *uiop;
1089 	int error;
1090 
1091 	uiop = ctxp->uiop;
1092 	if (uiop->uio_resid < UIO_MX)
1093 		return -1; /* no space */
1094 
1095 	if (ctxp->off < ctxp->startoff) {
1096 		ctxp->off++;
1097 		return 0;
1098 	}
1099 
1100 	if (kauth_authorize_process(kauth_cred_get(),
1101 	    KAUTH_PROCESS_CANSEE, p, NULL, NULL, NULL) != 0)
1102 		return 0;
1103 
1104 	memset(&d, 0, UIO_MX);
1105 	d.d_reclen = UIO_MX;
1106 	d.d_fileno = PROCFS_FILENO(p->p_pid, PFSproc, -1);
1107 	d.d_namlen = snprintf(d.d_name,
1108 	    UIO_MX - offsetof(struct dirent, d_name), "%ld", (long)p->p_pid);
1109 	d.d_type = DT_DIR;
1110 
1111 	proclist_unlock_read();
1112 	error = uiomove(&d, UIO_MX, uiop);
1113 	proclist_lock_read();
1114 	if (error) {
1115 		ctxp->error = error;
1116 		return -1;
1117 	}
1118 
1119 	ctxp->ncookies++;
1120 	if (ctxp->cookies)
1121 		*(ctxp->cookies)++ = ctxp->off + 1;
1122 	ctxp->off++;
1123 
1124 	return 0;
1125 }
1126 
1127 /*
1128  * readdir returns directory entries from pfsnode (vp).
1129  *
1130  * the strategy here with procfs is to generate a single
1131  * directory entry at a time (struct dirent) and then
1132  * copy that out to userland using uiomove.  a more efficent
1133  * though more complex implementation, would try to minimize
1134  * the number of calls to uiomove().  for procfs, this is
1135  * hardly worth the added code complexity.
1136  *
1137  * this should just be done through read()
1138  */
1139 int
1140 procfs_readdir(v)
1141 	void *v;
1142 {
1143 	struct vop_readdir_args /* {
1144 		struct vnode *a_vp;
1145 		struct uio *a_uio;
1146 		kauth_cred_t a_cred;
1147 		int *a_eofflag;
1148 		off_t **a_cookies;
1149 		int *a_ncookies;
1150 	} */ *ap = v;
1151 	struct uio *uio = ap->a_uio;
1152 	struct dirent d;
1153 	struct pfsnode *pfs;
1154 	off_t i;
1155 	int error;
1156 	off_t *cookies = NULL;
1157 	int ncookies;
1158 	struct vnode *vp;
1159 	const struct proc_target *pt;
1160 	struct procfs_root_readdir_ctx ctx;
1161 
1162 	vp = ap->a_vp;
1163 	pfs = VTOPFS(vp);
1164 
1165 	if (uio->uio_resid < UIO_MX)
1166 		return (EINVAL);
1167 	if (uio->uio_offset < 0)
1168 		return (EINVAL);
1169 
1170 	error = 0;
1171 	i = uio->uio_offset;
1172 	memset(&d, 0, UIO_MX);
1173 	d.d_reclen = UIO_MX;
1174 	ncookies = uio->uio_resid / UIO_MX;
1175 
1176 	switch (pfs->pfs_type) {
1177 	/*
1178 	 * this is for the process-specific sub-directories.
1179 	 * all that is needed to is copy out all the entries
1180 	 * from the procent[] table (top of this file).
1181 	 */
1182 	case PFSproc: {
1183 		struct proc *p;
1184 
1185 		if (i >= nproc_targets)
1186 			return 0;
1187 
1188 		p = PFIND(pfs->pfs_pid);
1189 		if (p == NULL)
1190 			break;
1191 
1192 		if (ap->a_ncookies) {
1193 			ncookies = min(ncookies, (nproc_targets - i));
1194 			cookies = malloc(ncookies * sizeof (off_t),
1195 			    M_TEMP, M_WAITOK);
1196 			*ap->a_cookies = cookies;
1197 		}
1198 
1199 		for (pt = &proc_targets[i];
1200 		     uio->uio_resid >= UIO_MX && i < nproc_targets; pt++, i++) {
1201 			if (pt->pt_valid &&
1202 			    (*pt->pt_valid)(proc_representative_lwp(p), vp->v_mount) == 0)
1203 				continue;
1204 
1205 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid,
1206 			    pt->pt_pfstype, -1);
1207 			d.d_namlen = pt->pt_namlen;
1208 			memcpy(d.d_name, pt->pt_name, pt->pt_namlen + 1);
1209 			d.d_type = pt->pt_type;
1210 
1211 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1212 				break;
1213 			if (cookies)
1214 				*cookies++ = i + 1;
1215 		}
1216 
1217 	    	break;
1218 	}
1219 	case PFSfd: {
1220 		struct proc *p;
1221 		struct filedesc	*fdp;
1222 		struct file *fp;
1223 		int lim, nc = 0;
1224 
1225 		p = PFIND(pfs->pfs_pid);
1226 		if (p == NULL)
1227 			return ESRCH;
1228 
1229 		if (kauth_authorize_process(kauth_cred_get(),
1230 		    KAUTH_PROCESS_CANSEE, p, NULL, NULL, NULL) != 0)
1231 			return ESRCH;
1232 
1233 		fdp = p->p_fd;
1234 
1235 		lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfiles);
1236 		if (i >= lim)
1237 			return 0;
1238 
1239 		if (ap->a_ncookies) {
1240 			ncookies = min(ncookies, (fdp->fd_nfiles + 2 - i));
1241 			cookies = malloc(ncookies * sizeof (off_t),
1242 			    M_TEMP, M_WAITOK);
1243 			*ap->a_cookies = cookies;
1244 		}
1245 
1246 		for (; i < 2 && uio->uio_resid >= UIO_MX; i++) {
1247 			pt = &proc_targets[i];
1248 			d.d_namlen = pt->pt_namlen;
1249 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid,
1250 			    pt->pt_pfstype, -1);
1251 			(void)memcpy(d.d_name, pt->pt_name, pt->pt_namlen + 1);
1252 			d.d_type = pt->pt_type;
1253 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1254 				break;
1255 			if (cookies)
1256 				*cookies++ = i + 1;
1257 			nc++;
1258 		}
1259 		if (error) {
1260 			ncookies = nc;
1261 			break;
1262 		}
1263 		for (; uio->uio_resid >= UIO_MX && i < fdp->fd_nfiles; i++) {
1264 			/* check the descriptor exists */
1265 			if ((fp = fd_getfile(fdp, i - 2)) == NULL)
1266 				continue;
1267 			simple_unlock(&fp->f_slock);
1268 
1269 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid, PFSfd, i - 2);
1270 			d.d_namlen = snprintf(d.d_name, sizeof(d.d_name),
1271 			    "%lld", (long long)(i - 2));
1272 			d.d_type = VREG;
1273 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1274 				break;
1275 			if (cookies)
1276 				*cookies++ = i + 1;
1277 			nc++;
1278 		}
1279 		ncookies = nc;
1280 		break;
1281 	}
1282 
1283 	/*
1284 	 * this is for the root of the procfs filesystem
1285 	 * what is needed are special entries for "curproc"
1286 	 * and "self" followed by an entry for each process
1287 	 * on allproc.
1288 	 */
1289 
1290 	case PFSroot: {
1291 		int nc = 0;
1292 
1293 		if (ap->a_ncookies) {
1294 			/*
1295 			 * XXX Potentially allocating too much space here,
1296 			 * but I'm lazy. This loop needs some work.
1297 			 */
1298 			cookies = malloc(ncookies * sizeof (off_t),
1299 			    M_TEMP, M_WAITOK);
1300 			*ap->a_cookies = cookies;
1301 		}
1302 		error = 0;
1303 		/* 0 ... 3 are static entries. */
1304 		for (; i <= 3 && uio->uio_resid >= UIO_MX; i++) {
1305 			switch (i) {
1306 			case 0:		/* `.' */
1307 			case 1:		/* `..' */
1308 				d.d_fileno = PROCFS_FILENO(0, PFSroot, -1);
1309 				d.d_namlen = i + 1;
1310 				memcpy(d.d_name, "..", d.d_namlen);
1311 				d.d_name[i + 1] = '\0';
1312 				d.d_type = DT_DIR;
1313 				break;
1314 
1315 			case 2:
1316 				d.d_fileno = PROCFS_FILENO(0, PFScurproc, -1);
1317 				d.d_namlen = sizeof("curproc") - 1;
1318 				memcpy(d.d_name, "curproc", sizeof("curproc"));
1319 				d.d_type = DT_LNK;
1320 				break;
1321 
1322 			case 3:
1323 				d.d_fileno = PROCFS_FILENO(0, PFSself, -1);
1324 				d.d_namlen = sizeof("self") - 1;
1325 				memcpy(d.d_name, "self", sizeof("self"));
1326 				d.d_type = DT_LNK;
1327 				break;
1328 			}
1329 
1330 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1331 				break;
1332 			nc++;
1333 			if (cookies)
1334 				*cookies++ = i + 1;
1335 		}
1336 		/* 4 ... are process entries. */
1337 		ctx.uiop = uio;
1338 		ctx.error = 0;
1339 		ctx.off = 4;
1340 		ctx.startoff = i;
1341 		ctx.cookies = cookies;
1342 		ctx.ncookies = nc;
1343 		proclist_foreach_call(&allproc,
1344 		    procfs_root_readdir_callback, &ctx);
1345 		cookies = ctx.cookies;
1346 		nc = ctx.ncookies;
1347 		error = ctx.error;
1348 		if (error)
1349 			break;
1350 
1351 		/* misc entries. */
1352 		if (i < ctx.off)
1353 			i = ctx.off;
1354 		if (i >= ctx.off + nproc_root_targets)
1355 			break;
1356 		for (pt = &proc_root_targets[i - ctx.off];
1357 		    uio->uio_resid >= UIO_MX &&
1358 		    pt < &proc_root_targets[nproc_root_targets];
1359 		    pt++, i++) {
1360 			if (pt->pt_valid &&
1361 			    (*pt->pt_valid)(NULL, vp->v_mount) == 0)
1362 				continue;
1363 			d.d_fileno = PROCFS_FILENO(0, pt->pt_pfstype, -1);
1364 			d.d_namlen = pt->pt_namlen;
1365 			memcpy(d.d_name, pt->pt_name, pt->pt_namlen + 1);
1366 			d.d_type = pt->pt_type;
1367 
1368 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1369 				break;
1370 			nc++;
1371 			if (cookies)
1372 				*cookies++ = i + 1;
1373 		}
1374 
1375 		ncookies = nc;
1376 		break;
1377 	}
1378 
1379 	default:
1380 		error = ENOTDIR;
1381 		break;
1382 	}
1383 
1384 	if (ap->a_ncookies) {
1385 		if (error) {
1386 			if (cookies)
1387 				free(*ap->a_cookies, M_TEMP);
1388 			*ap->a_ncookies = 0;
1389 			*ap->a_cookies = NULL;
1390 		} else
1391 			*ap->a_ncookies = ncookies;
1392 	}
1393 	uio->uio_offset = i;
1394 	return (error);
1395 }
1396 
1397 /*
1398  * readlink reads the link of `curproc' and others
1399  */
1400 int
1401 procfs_readlink(v)
1402 	void *v;
1403 {
1404 	struct vop_readlink_args *ap = v;
1405 	char bf[16];		/* should be enough */
1406 	char *bp = bf;
1407 	char *path = NULL;
1408 	int len;
1409 	int error = 0;
1410 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
1411 	struct proc *pown;
1412 
1413 	if (pfs->pfs_fileno == PROCFS_FILENO(0, PFScurproc, -1))
1414 		len = snprintf(bf, sizeof(bf), "%ld", (long)curproc->p_pid);
1415 	else if (pfs->pfs_fileno == PROCFS_FILENO(0, PFSself, -1))
1416 		len = snprintf(bf, sizeof(bf), "%s", "curproc");
1417 	else if (pfs->pfs_fileno == PROCFS_FILENO(pfs->pfs_pid, PFScwd, -1)) {
1418 		pown = PFIND(pfs->pfs_pid);
1419 		if (pown == NULL)
1420 			return (ESRCH);
1421 		MALLOC(path, char *, MAXPATHLEN + 4, M_TEMP,
1422 		    M_WAITOK|M_CANFAIL);
1423 		if (path == NULL)
1424 			return (ENOMEM);
1425 		bp = path + MAXPATHLEN;
1426 		*--bp = '\0';
1427 		(void)procfs_dir(PFScwd, curlwp, pown, &bp, path,
1428 		     MAXPATHLEN);
1429 		len = strlen(bp);
1430 	}
1431 	else if (pfs->pfs_fileno == PROCFS_FILENO(pfs->pfs_pid, PFSchroot, -1)) {
1432 		pown = PFIND(pfs->pfs_pid);
1433 		if (pown == NULL)
1434 			return (ESRCH);
1435 		MALLOC(path, char *, MAXPATHLEN + 4, M_TEMP,
1436 		    M_WAITOK|M_CANFAIL);
1437 		if (path == NULL)
1438 			return (ENOMEM);
1439 		bp = path + MAXPATHLEN;
1440 		*--bp = '\0';
1441 		(void)procfs_dir(PFSchroot, curlwp, pown, &bp, path,
1442 		     MAXPATHLEN);
1443 		len = strlen(bp);
1444 	}
1445 	else {
1446 		struct file *fp;
1447 		struct vnode *vxp, *vp;
1448 
1449 		if ((error = procfs_getfp(pfs, &pown, &fp)) != 0)
1450 			return error;
1451 		FILE_USE(fp);
1452 		switch (fp->f_type) {
1453 		case DTYPE_VNODE:
1454 			vxp = (struct vnode *)fp->f_data;
1455 			if (vxp->v_type != VDIR) {
1456 				FILE_UNUSE(fp, proc_representative_lwp(pown));
1457 				return EINVAL;
1458 			}
1459 			if ((path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK))
1460 			    == NULL) {
1461 				FILE_UNUSE(fp, proc_representative_lwp(pown));
1462 				return ENOMEM;
1463 			}
1464 			bp = path + MAXPATHLEN;
1465 			*--bp = '\0';
1466 			vp = curproc->p_cwdi->cwdi_rdir;
1467 			if (vp == NULL)
1468 				vp = rootvnode;
1469 			error = getcwd_common(vxp, vp, &bp, path,
1470 			    MAXPATHLEN / 2, 0, curlwp);
1471 			FILE_UNUSE(fp, proc_representative_lwp(pown));
1472 			if (error) {
1473 				free(path, M_TEMP);
1474 				return error;
1475 			}
1476 			len = strlen(bp);
1477 			break;
1478 
1479 		case DTYPE_MISC:
1480 			len = snprintf(bf, sizeof(bf), "%s", "[misc]");
1481 			break;
1482 
1483 		case DTYPE_KQUEUE:
1484 			len = snprintf(bf, sizeof(bf), "%s", "[kqueue]");
1485 			break;
1486 
1487 		default:
1488 			return EINVAL;
1489 		}
1490 	}
1491 
1492 	error = uiomove(bp, len, ap->a_uio);
1493 	if (path)
1494 		free(path, M_TEMP);
1495 	return error;
1496 }
1497 
1498 /*
1499  * convert decimal ascii to int
1500  */
1501 static int
1502 atoi(b, len)
1503 	const char *b;
1504 	size_t len;
1505 {
1506 	int p = 0;
1507 
1508 	while (len--) {
1509 		char c = *b++;
1510 		if (c < '0' || c > '9')
1511 			return -1;
1512 		p = 10 * p + (c - '0');
1513 	}
1514 
1515 	return p;
1516 }
1517