xref: /netbsd-src/sys/miscfs/procfs/procfs_vnops.c (revision e5548b402ae4c44fb816de42c7bba9581ce23ef5)
1 /*	$NetBSD: procfs_vnops.c,v 1.128 2005/12/11 12:24:51 christos 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 1.128 2005/12/11 12:24:51 christos 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 
98 #include <uvm/uvm_extern.h>	/* for PAGE_SIZE */
99 
100 #include <machine/reg.h>
101 
102 #include <miscfs/genfs/genfs.h>
103 #include <miscfs/procfs/procfs.h>
104 
105 /*
106  * Vnode Operations.
107  *
108  */
109 
110 static int procfs_validfile_linux(struct lwp *, struct mount *);
111 static int procfs_root_readdir_callback(struct proc *, void *);
112 static struct vnode *procfs_dir(pfstype, struct lwp *, struct proc *,
113 				char **, char *, int);
114 
115 /*
116  * This is a list of the valid names in the
117  * process-specific sub-directories.  It is
118  * used in procfs_lookup and procfs_readdir
119  */
120 static const struct proc_target {
121 	u_char	pt_type;
122 	u_char	pt_namlen;
123 	const char	*pt_name;
124 	pfstype	pt_pfstype;
125 	int	(*pt_valid)(struct lwp *, struct mount *);
126 } proc_targets[] = {
127 #define N(s) sizeof(s)-1, s
128 	/*	  name		type		validp */
129 	{ DT_DIR, N("."),	PFSproc,		NULL },
130 	{ DT_DIR, N(".."),	PFSroot,		NULL },
131 	{ DT_DIR, N("fd"),	PFSfd,		NULL },
132 	{ DT_REG, N("file"),	PFSfile,		procfs_validfile },
133 	{ DT_REG, N("mem"),	PFSmem,		NULL },
134 	{ DT_REG, N("regs"),	PFSregs,		procfs_validregs },
135 	{ DT_REG, N("fpregs"),	PFSfpregs,	procfs_validfpregs },
136 	{ DT_REG, N("ctl"),	PFSctl,		NULL },
137 	{ DT_REG, N("stat"),	PFSstat,		procfs_validfile_linux },
138 	{ DT_REG, N("status"),	PFSstatus,	NULL },
139 	{ DT_REG, N("note"),	PFSnote,		NULL },
140 	{ DT_REG, N("notepg"),	PFSnotepg,	NULL },
141 	{ DT_REG, N("map"),	PFSmap,		procfs_validmap },
142 	{ DT_REG, N("maps"),	PFSmaps,		procfs_validmap },
143 	{ DT_REG, N("cmdline"), PFScmdline,	NULL },
144 	{ DT_REG, N("exe"),	PFSfile,		procfs_validfile_linux },
145 	{ DT_LNK, N("cwd"),	PFScwd,		NULL },
146 	{ DT_LNK, N("root"),	PFSchroot,	NULL },
147 #ifdef __HAVE_PROCFS_MACHDEP
148 	PROCFS_MACHDEP_NODETYPE_DEFNS
149 #endif
150 #undef N
151 };
152 static const int nproc_targets = sizeof(proc_targets) / sizeof(proc_targets[0]);
153 
154 /*
155  * List of files in the root directory. Note: the validate function will
156  * be called with p == NULL for these ones.
157  */
158 static const struct proc_target proc_root_targets[] = {
159 #define N(s) sizeof(s)-1, s
160 	/*	  name		    type	    validp */
161 	{ DT_REG, N("meminfo"),     PFSmeminfo,        procfs_validfile_linux },
162 	{ DT_REG, N("cpuinfo"),     PFScpuinfo,        procfs_validfile_linux },
163 	{ DT_REG, N("uptime"),      PFSuptime,         procfs_validfile_linux },
164 	{ DT_REG, N("mounts"),	    PFSmounts,	       procfs_validfile_linux },
165 #undef N
166 };
167 static const int nproc_root_targets =
168     sizeof(proc_root_targets) / sizeof(proc_root_targets[0]);
169 
170 int	procfs_lookup(void *);
171 #define	procfs_create	genfs_eopnotsupp
172 #define	procfs_mknod	genfs_eopnotsupp
173 int	procfs_open(void *);
174 int	procfs_close(void *);
175 int	procfs_access(void *);
176 int	procfs_getattr(void *);
177 int	procfs_setattr(void *);
178 #define	procfs_read	procfs_rw
179 #define	procfs_write	procfs_rw
180 #define	procfs_fcntl	genfs_fcntl
181 #define	procfs_ioctl	genfs_enoioctl
182 #define	procfs_poll	genfs_poll
183 #define procfs_revoke	genfs_revoke
184 #define	procfs_fsync	genfs_nullop
185 #define	procfs_seek	genfs_nullop
186 #define	procfs_remove	genfs_eopnotsupp
187 int	procfs_link(void *);
188 #define	procfs_rename	genfs_eopnotsupp
189 #define	procfs_mkdir	genfs_eopnotsupp
190 #define	procfs_rmdir	genfs_eopnotsupp
191 int	procfs_symlink(void *);
192 int	procfs_readdir(void *);
193 int	procfs_readlink(void *);
194 #define	procfs_abortop	genfs_abortop
195 int	procfs_inactive(void *);
196 int	procfs_reclaim(void *);
197 #define	procfs_lock	genfs_lock
198 #define	procfs_unlock	genfs_unlock
199 #define	procfs_bmap	genfs_badop
200 #define	procfs_strategy	genfs_badop
201 int	procfs_print(void *);
202 int	procfs_pathconf(void *);
203 #define	procfs_islocked	genfs_islocked
204 #define	procfs_advlock	genfs_einval
205 #define	procfs_bwrite	genfs_eopnotsupp
206 #define procfs_putpages	genfs_null_putpages
207 
208 static int atoi(const char *, size_t);
209 
210 /*
211  * procfs vnode operations.
212  */
213 int (**procfs_vnodeop_p)(void *);
214 const struct vnodeopv_entry_desc procfs_vnodeop_entries[] = {
215 	{ &vop_default_desc, vn_default_error },
216 	{ &vop_lookup_desc, procfs_lookup },		/* lookup */
217 	{ &vop_create_desc, procfs_create },		/* create */
218 	{ &vop_mknod_desc, procfs_mknod },		/* mknod */
219 	{ &vop_open_desc, procfs_open },		/* open */
220 	{ &vop_close_desc, procfs_close },		/* close */
221 	{ &vop_access_desc, procfs_access },		/* access */
222 	{ &vop_getattr_desc, procfs_getattr },		/* getattr */
223 	{ &vop_setattr_desc, procfs_setattr },		/* setattr */
224 	{ &vop_read_desc, procfs_read },		/* read */
225 	{ &vop_write_desc, procfs_write },		/* write */
226 	{ &vop_fcntl_desc, procfs_fcntl },		/* fcntl */
227 	{ &vop_ioctl_desc, procfs_ioctl },		/* ioctl */
228 	{ &vop_poll_desc, procfs_poll },		/* poll */
229 	{ &vop_revoke_desc, procfs_revoke },		/* revoke */
230 	{ &vop_fsync_desc, procfs_fsync },		/* fsync */
231 	{ &vop_seek_desc, procfs_seek },		/* seek */
232 	{ &vop_remove_desc, procfs_remove },		/* remove */
233 	{ &vop_link_desc, procfs_link },		/* link */
234 	{ &vop_rename_desc, procfs_rename },		/* rename */
235 	{ &vop_mkdir_desc, procfs_mkdir },		/* mkdir */
236 	{ &vop_rmdir_desc, procfs_rmdir },		/* rmdir */
237 	{ &vop_symlink_desc, procfs_symlink },		/* symlink */
238 	{ &vop_readdir_desc, procfs_readdir },		/* readdir */
239 	{ &vop_readlink_desc, procfs_readlink },	/* readlink */
240 	{ &vop_abortop_desc, procfs_abortop },		/* abortop */
241 	{ &vop_inactive_desc, procfs_inactive },	/* inactive */
242 	{ &vop_reclaim_desc, procfs_reclaim },		/* reclaim */
243 	{ &vop_lock_desc, procfs_lock },		/* lock */
244 	{ &vop_unlock_desc, procfs_unlock },		/* unlock */
245 	{ &vop_bmap_desc, procfs_bmap },		/* bmap */
246 	{ &vop_strategy_desc, procfs_strategy },	/* strategy */
247 	{ &vop_print_desc, procfs_print },		/* print */
248 	{ &vop_islocked_desc, procfs_islocked },	/* islocked */
249 	{ &vop_pathconf_desc, procfs_pathconf },	/* pathconf */
250 	{ &vop_advlock_desc, procfs_advlock },		/* advlock */
251 	{ &vop_putpages_desc, procfs_putpages },	/* putpages */
252 	{ NULL, NULL }
253 };
254 const struct vnodeopv_desc procfs_vnodeop_opv_desc =
255 	{ &procfs_vnodeop_p, procfs_vnodeop_entries };
256 /*
257  * set things up for doing i/o on
258  * the pfsnode (vp).  (vp) is locked
259  * on entry, and should be left locked
260  * on exit.
261  *
262  * for procfs we don't need to do anything
263  * in particular for i/o.  all that is done
264  * is to support exclusive open on process
265  * memory images.
266  */
267 int
268 procfs_open(v)
269 	void *v;
270 {
271 	struct vop_open_args /* {
272 		struct vnode *a_vp;
273 		int  a_mode;
274 		struct ucred *a_cred;
275 		struct lwp *a_l;
276 	} */ *ap = v;
277 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
278 	struct lwp *l1;
279 	struct proc *p2;
280 	int error;
281 
282 	l1 = ap->a_l;				/* tracer */
283 	p2 = PFIND(pfs->pfs_pid);		/* traced */
284 
285 	if (p2 == NULL)
286 		return (ENOENT);		/* was ESRCH, jsp */
287 
288 	switch (pfs->pfs_type) {
289 	case PFSmem:
290 		if (((pfs->pfs_flags & FWRITE) && (ap->a_mode & O_EXCL)) ||
291 		    ((pfs->pfs_flags & O_EXCL) && (ap->a_mode & FWRITE)))
292 			return (EBUSY);
293 
294 		if ((error = process_checkioperm(l1, p2)) != 0)
295 			return (error);
296 
297 		if (ap->a_mode & FWRITE)
298 			pfs->pfs_flags = ap->a_mode & (FWRITE|O_EXCL);
299 
300 		return (0);
301 
302 	default:
303 		break;
304 	}
305 
306 	return (0);
307 }
308 
309 /*
310  * close the pfsnode (vp) after doing i/o.
311  * (vp) is not locked on entry or exit.
312  *
313  * nothing to do for procfs other than undo
314  * any exclusive open flag (see _open above).
315  */
316 int
317 procfs_close(v)
318 	void *v;
319 {
320 	struct vop_close_args /* {
321 		struct vnode *a_vp;
322 		int  a_fflag;
323 		struct ucred *a_cred;
324 		struct lwp *a_l;
325 	} */ *ap = v;
326 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
327 
328 	switch (pfs->pfs_type) {
329 	case PFSmem:
330 		if ((ap->a_fflag & FWRITE) && (pfs->pfs_flags & O_EXCL))
331 			pfs->pfs_flags &= ~(FWRITE|O_EXCL);
332 		break;
333 
334 	default:
335 		break;
336 	}
337 
338 	return (0);
339 }
340 
341 /*
342  * _inactive is called when the pfsnode
343  * is vrele'd and the reference count goes
344  * to zero.  (vp) will be on the vnode free
345  * list, so to get it back vget() must be
346  * used.
347  *
348  * for procfs, check if the process is still
349  * alive and if it isn't then just throw away
350  * the vnode by calling vgone().  this may
351  * be overkill and a waste of time since the
352  * chances are that the process will still be
353  * there and PFIND is not free.
354  *
355  * (vp) is locked on entry, but must be unlocked on exit.
356  */
357 int
358 procfs_inactive(v)
359 	void *v;
360 {
361 	struct vop_inactive_args /* {
362 		struct vnode *a_vp;
363 		struct proc *a_p;
364 	} */ *ap = v;
365 	struct vnode *vp = ap->a_vp;
366 	struct pfsnode *pfs = VTOPFS(vp);
367 
368 	VOP_UNLOCK(vp, 0);
369 	if (PFIND(pfs->pfs_pid) == NULL && (vp->v_flag & VXLOCK) == 0)
370 		vgone(vp);
371 
372 	return (0);
373 }
374 
375 /*
376  * _reclaim is called when getnewvnode()
377  * wants to make use of an entry on the vnode
378  * free list.  at this time the filesystem needs
379  * to free any private data and remove the node
380  * from any private lists.
381  */
382 int
383 procfs_reclaim(v)
384 	void *v;
385 {
386 	struct vop_reclaim_args /* {
387 		struct vnode *a_vp;
388 	} */ *ap = v;
389 
390 	return (procfs_freevp(ap->a_vp));
391 }
392 
393 /*
394  * Return POSIX pathconf information applicable to special devices.
395  */
396 int
397 procfs_pathconf(v)
398 	void *v;
399 {
400 	struct vop_pathconf_args /* {
401 		struct vnode *a_vp;
402 		int a_name;
403 		register_t *a_retval;
404 	} */ *ap = v;
405 
406 	switch (ap->a_name) {
407 	case _PC_LINK_MAX:
408 		*ap->a_retval = LINK_MAX;
409 		return (0);
410 	case _PC_MAX_CANON:
411 		*ap->a_retval = MAX_CANON;
412 		return (0);
413 	case _PC_MAX_INPUT:
414 		*ap->a_retval = MAX_INPUT;
415 		return (0);
416 	case _PC_PIPE_BUF:
417 		*ap->a_retval = PIPE_BUF;
418 		return (0);
419 	case _PC_CHOWN_RESTRICTED:
420 		*ap->a_retval = 1;
421 		return (0);
422 	case _PC_VDISABLE:
423 		*ap->a_retval = _POSIX_VDISABLE;
424 		return (0);
425 	case _PC_SYNC_IO:
426 		*ap->a_retval = 1;
427 		return (0);
428 	default:
429 		return (EINVAL);
430 	}
431 	/* NOTREACHED */
432 }
433 
434 /*
435  * _print is used for debugging.
436  * just print a readable description
437  * of (vp).
438  */
439 int
440 procfs_print(v)
441 	void *v;
442 {
443 	struct vop_print_args /* {
444 		struct vnode *a_vp;
445 	} */ *ap = v;
446 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
447 
448 	printf("tag VT_PROCFS, type %d, pid %d, mode %x, flags %lx\n",
449 	    pfs->pfs_type, pfs->pfs_pid, pfs->pfs_mode, pfs->pfs_flags);
450 	return 0;
451 }
452 
453 int
454 procfs_link(v)
455 	void *v;
456 {
457 	struct vop_link_args /* {
458 		struct vnode *a_dvp;
459 		struct vnode *a_vp;
460 		struct componentname *a_cnp;
461 	} */ *ap = v;
462 
463 	VOP_ABORTOP(ap->a_dvp, ap->a_cnp);
464 	vput(ap->a_dvp);
465 	return (EROFS);
466 }
467 
468 int
469 procfs_symlink(v)
470 	void *v;
471 {
472 	struct vop_symlink_args /* {
473 		struct vnode *a_dvp;
474 		struct vnode **a_vpp;
475 		struct componentname *a_cnp;
476 		struct vattr *a_vap;
477 		char *a_target;
478 	} */ *ap = v;
479 
480 	VOP_ABORTOP(ap->a_dvp, ap->a_cnp);
481 	vput(ap->a_dvp);
482 	return (EROFS);
483 }
484 
485 /*
486  * Works out the path to (and vnode of) the target process's current
487  * working directory or chroot.  If the caller is in a chroot and
488  * can't "reach" the target's cwd or root (or some other error
489  * occurs), a "/" is returned for the path and a NULL pointer is
490  * returned for the vnode.
491  */
492 static struct vnode *
493 procfs_dir(pfstype t, struct lwp *caller, struct proc *target,
494 	   char **bpp, char *path, int len)
495 {
496 	struct vnode *vp, *rvp = caller->l_proc->p_cwdi->cwdi_rdir;
497 	char *bp;
498 
499 	bp = bpp ? *bpp : NULL;
500 
501 	switch (t) {
502 	case PFScwd:
503 		vp = target->p_cwdi->cwdi_cdir;
504 		break;
505 	case PFSchroot:
506 		vp = target->p_cwdi->cwdi_rdir;
507 		break;
508 	default:
509 		return (NULL);
510 	}
511 
512 	if (rvp == NULL)
513 		rvp = rootvnode;
514 	if (vp == NULL || getcwd_common(vp, rvp, bp ? &bp : NULL, path,
515 	    len / 2, 0, caller) != 0) {
516 		vp = NULL;
517 		if (bpp) {
518 			bp = *bpp;
519 			*--bp = '/';
520 		}
521 	}
522 
523 	if (bpp)
524 		*bpp = bp;
525 
526 	return (vp);
527 }
528 
529 /*
530  * Invent attributes for pfsnode (vp) and store
531  * them in (vap).
532  * Directories lengths are returned as zero since
533  * any real length would require the genuine size
534  * to be computed, and nothing cares anyway.
535  *
536  * this is relatively minimal for procfs.
537  */
538 int
539 procfs_getattr(v)
540 	void *v;
541 {
542 	struct vop_getattr_args /* {
543 		struct vnode *a_vp;
544 		struct vattr *a_vap;
545 		struct ucred *a_cred;
546 		struct lwp *a_l;
547 	} */ *ap = v;
548 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
549 	struct vattr *vap = ap->a_vap;
550 	struct proc *procp;
551 	int error;
552 
553 	/* first check the process still exists */
554 	switch (pfs->pfs_type) {
555 	case PFSroot:
556 	case PFScurproc:
557 	case PFSself:
558 		procp = 0;
559 		break;
560 
561 	default:
562 		procp = PFIND(pfs->pfs_pid);
563 		if (procp == NULL)
564 			return (ENOENT);
565 		break;
566 	}
567 
568 	if (procp != NULL) {
569 		if (CURTAIN(curlwp->l_proc->p_ucred->cr_uid,
570 			    procp->p_ucred->cr_uid))
571 			return (ENOENT);
572 	}
573 
574 	error = 0;
575 
576 	/* start by zeroing out the attributes */
577 	VATTR_NULL(vap);
578 
579 	/* next do all the common fields */
580 	vap->va_type = ap->a_vp->v_type;
581 	vap->va_mode = pfs->pfs_mode;
582 	vap->va_fileid = pfs->pfs_fileno;
583 	vap->va_flags = 0;
584 	vap->va_blocksize = PAGE_SIZE;
585 
586 	/*
587 	 * Make all times be current TOD.  Avoid microtime(9), it's slow.
588 	 * We don't guard the read from time(9) with splclock(9) since we
589 	 * don't actually need to be THAT sure the access is atomic.
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 	TIMEVAL_TO_TIMESPEC(&time, &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 		TIMEVAL_TO_TIMESPEC(&boottime, &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 = procp->p_ucred->cr_uid;
630 		vap->va_gid = procp->p_ucred->cr_gid;
631 		break;
632 	case PFSmeminfo:
633 	case PFScpuinfo:
634 	case PFSuptime:
635 	case PFSmounts:
636 		vap->va_nlink = 1;
637 		vap->va_uid = vap->va_gid = 0;
638 		break;
639 
640 	default:
641 		break;
642 	}
643 
644 	/*
645 	 * now do the object specific fields
646 	 *
647 	 * The size could be set from struct reg, but it's hardly
648 	 * worth the trouble, and it puts some (potentially) machine
649 	 * dependent data into this machine-independent code.  If it
650 	 * becomes important then this function should break out into
651 	 * a per-file stat function in the corresponding .c file.
652 	 */
653 
654 	switch (pfs->pfs_type) {
655 	case PFSroot:
656 		/*
657 		 * Set nlink to 1 to tell fts(3) we don't actually know.
658 		 */
659 		vap->va_nlink = 1;
660 		vap->va_uid = 0;
661 		vap->va_gid = 0;
662 		vap->va_bytes = vap->va_size = DEV_BSIZE;
663 		break;
664 
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 PFSself:
676 		vap->va_nlink = 1;
677 		vap->va_uid = 0;
678 		vap->va_gid = 0;
679 		vap->va_bytes = vap->va_size = sizeof("curproc") - 1;
680 		break;
681 
682 	case PFSfd:
683 		if (pfs->pfs_fd != -1) {
684 			struct file *fp;
685 			struct proc *pown;
686 
687 			if ((error = procfs_getfp(pfs, &pown, &fp)) != 0)
688 				return error;
689 			FILE_USE(fp);
690 			vap->va_nlink = 1;
691 			vap->va_uid = fp->f_cred->cr_uid;
692 			vap->va_gid = fp->f_cred->cr_gid;
693 			switch (fp->f_type) {
694 			case DTYPE_VNODE:
695 				vap->va_bytes = vap->va_size =
696 				    ((struct vnode *)fp->f_data)->v_size;
697 				break;
698 			default:
699 				vap->va_bytes = vap->va_size = 0;
700 				break;
701 			}
702 			FILE_UNUSE(fp, proc_representative_lwp(pown));
703 			break;
704 		}
705 		/*FALLTHROUGH*/
706 	case PFSproc:
707 		vap->va_nlink = 2;
708 		vap->va_uid = procp->p_ucred->cr_uid;
709 		vap->va_gid = procp->p_ucred->cr_gid;
710 		vap->va_bytes = vap->va_size = DEV_BSIZE;
711 		break;
712 
713 	case PFSfile:
714 		error = EOPNOTSUPP;
715 		break;
716 
717 	case PFSmem:
718 		vap->va_bytes = vap->va_size =
719 			ctob(procp->p_vmspace->vm_tsize +
720 				    procp->p_vmspace->vm_dsize +
721 				    procp->p_vmspace->vm_ssize);
722 		break;
723 
724 #if defined(PT_GETREGS) || defined(PT_SETREGS)
725 	case PFSregs:
726 		vap->va_bytes = vap->va_size = sizeof(struct reg);
727 		break;
728 #endif
729 
730 #if defined(PT_GETFPREGS) || defined(PT_SETFPREGS)
731 	case PFSfpregs:
732 		vap->va_bytes = vap->va_size = sizeof(struct fpreg);
733 		break;
734 #endif
735 
736 	case PFSctl:
737 	case PFSstatus:
738 	case PFSstat:
739 	case PFSnote:
740 	case PFSnotepg:
741 	case PFScmdline:
742 	case PFSmeminfo:
743 	case PFScpuinfo:
744 	case PFSuptime:
745 	case PFSmounts:
746 		vap->va_bytes = vap->va_size = 0;
747 		break;
748 	case PFSmap:
749 	case PFSmaps:
750 		/*
751 		 * Advise a larger blocksize for the map files, so that
752 		 * they may be read in one pass.
753 		 */
754 		vap->va_blocksize = 4 * PAGE_SIZE;
755 		vap->va_bytes = vap->va_size = 0;
756 		break;
757 
758 	case PFScwd:
759 	case PFSchroot: {
760 		char *path, *bp;
761 
762 		MALLOC(path, char *, MAXPATHLEN + 4, M_TEMP,
763 		    M_WAITOK|M_CANFAIL);
764 		if (path == NULL)
765 			return (ENOMEM);
766 		vap->va_nlink = 1;
767 		vap->va_uid = 0;
768 		vap->va_gid = 0;
769 		bp = path + MAXPATHLEN;
770 		*--bp = '\0';
771 		(void)procfs_dir(pfs->pfs_type, curlwp, procp, &bp, path,
772 		     MAXPATHLEN);
773 		vap->va_bytes = vap->va_size = strlen(bp);
774 		free(path, M_TEMP);
775 		break;
776 	}
777 
778 #ifdef __HAVE_PROCFS_MACHDEP
779 	PROCFS_MACHDEP_NODETYPE_CASES
780 		error = procfs_machdep_getattr(ap->a_vp, vap, procp);
781 		break;
782 #endif
783 
784 	default:
785 		panic("procfs_getattr");
786 	}
787 
788 	return (error);
789 }
790 
791 /*ARGSUSED*/
792 int
793 procfs_setattr(v)
794 	void *v;
795 {
796 	/*
797 	 * just fake out attribute setting
798 	 * it's not good to generate an error
799 	 * return, otherwise things like creat()
800 	 * will fail when they try to set the
801 	 * file length to 0.  worse, this means
802 	 * that echo $note > /proc/$pid/note will fail.
803 	 */
804 
805 	return (0);
806 }
807 
808 /*
809  * implement access checking.
810  *
811  * actually, the check for super-user is slightly
812  * broken since it will allow read access to write-only
813  * objects.  this doesn't cause any particular trouble
814  * but does mean that the i/o entry points need to check
815  * that the operation really does make sense.
816  */
817 int
818 procfs_access(v)
819 	void *v;
820 {
821 	struct vop_access_args /* {
822 		struct vnode *a_vp;
823 		int a_mode;
824 		struct ucred *a_cred;
825 		struct lwp *a_l;
826 	} */ *ap = v;
827 	struct vattr va;
828 	int error;
829 
830 	if ((error = VOP_GETATTR(ap->a_vp, &va, ap->a_cred, ap->a_l)) != 0)
831 		return (error);
832 
833 	return (vaccess(va.va_type, va.va_mode,
834 	    va.va_uid, va.va_gid, ap->a_mode, ap->a_cred));
835 }
836 
837 /*
838  * lookup.  this is incredibly complicated in the
839  * general case, however for most pseudo-filesystems
840  * very little needs to be done.
841  *
842  * Locking isn't hard here, just poorly documented.
843  *
844  * If we're looking up ".", just vref the parent & return it.
845  *
846  * If we're looking up "..", unlock the parent, and lock "..". If everything
847  * went ok, and we're on the last component and the caller requested the
848  * parent locked, try to re-lock the parent. We do this to prevent lock
849  * races.
850  *
851  * For anything else, get the needed node. Then unlock the parent if not
852  * the last component or not LOCKPARENT (i.e. if we wouldn't re-lock the
853  * parent in the .. case).
854  *
855  * We try to exit with the parent locked in error cases.
856  */
857 int
858 procfs_lookup(v)
859 	void *v;
860 {
861 	struct vop_lookup_args /* {
862 		struct vnode * a_dvp;
863 		struct vnode ** a_vpp;
864 		struct componentname * a_cnp;
865 	} */ *ap = v;
866 	struct componentname *cnp = ap->a_cnp;
867 	struct vnode **vpp = ap->a_vpp;
868 	struct vnode *dvp = ap->a_dvp;
869 	const char *pname = cnp->cn_nameptr;
870 	const struct proc_target *pt = NULL;
871 	struct vnode *fvp;
872 	pid_t pid;
873 	struct pfsnode *pfs;
874 	struct proc *p = NULL;
875 	struct lwp *l = NULL;
876 	int i, error, wantpunlock, iscurproc = 0, isself = 0;
877 
878 	*vpp = NULL;
879 	cnp->cn_flags &= ~PDIRUNLOCK;
880 
881 	if (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)
882 		return (EROFS);
883 
884 	if (cnp->cn_namelen == 1 && *pname == '.') {
885 		*vpp = dvp;
886 		VREF(dvp);
887 		return (0);
888 	}
889 
890 	wantpunlock = (~cnp->cn_flags & (LOCKPARENT | ISLASTCN));
891 	pfs = VTOPFS(dvp);
892 	switch (pfs->pfs_type) {
893 	case PFSroot:
894 		/*
895 		 * Shouldn't get here with .. in the root node.
896 		 */
897 		if (cnp->cn_flags & ISDOTDOT)
898 			return (EIO);
899 
900 		iscurproc = CNEQ(cnp, "curproc", 7);
901 		isself = CNEQ(cnp, "self", 4);
902 
903 		if (iscurproc || isself) {
904 			error = procfs_allocvp(dvp->v_mount, vpp, 0,
905 			    iscurproc ? PFScurproc : PFSself, -1);
906 			if ((error == 0) && (wantpunlock)) {
907 				VOP_UNLOCK(dvp, 0);
908 				cnp->cn_flags |= PDIRUNLOCK;
909 			}
910 			return (error);
911 		}
912 
913 		for (i = 0; i < nproc_root_targets; i++) {
914 			pt = &proc_root_targets[i];
915 			if (cnp->cn_namelen == pt->pt_namlen &&
916 			    memcmp(pt->pt_name, pname, cnp->cn_namelen) == 0 &&
917 			    (pt->pt_valid == NULL ||
918 			     (*pt->pt_valid)(cnp->cn_lwp, dvp->v_mount)))
919 				break;
920 		}
921 
922 		if (i != nproc_root_targets) {
923 			error = procfs_allocvp(dvp->v_mount, vpp, 0,
924 			    pt->pt_pfstype, -1);
925 			if ((error == 0) && (wantpunlock)) {
926 				VOP_UNLOCK(dvp, 0);
927 				cnp->cn_flags |= PDIRUNLOCK;
928 			}
929 			return (error);
930 		}
931 
932 		pid = (pid_t)atoi(pname, cnp->cn_namelen);
933 
934 		p = PFIND(pid);
935 		if (p == NULL)
936 			break;
937 
938 		error = procfs_allocvp(dvp->v_mount, vpp, pid, PFSproc, -1);
939 		if ((error == 0) && (wantpunlock)) {
940 			VOP_UNLOCK(dvp, 0);
941 			cnp->cn_flags |= PDIRUNLOCK;
942 		}
943 		return (error);
944 
945 	case PFSproc:
946 		/*
947 		 * do the .. dance. We unlock the directory, and then
948 		 * get the root dir. That will automatically return ..
949 		 * locked. Then if the caller wanted dvp locked, we
950 		 * re-lock.
951 		 */
952 		if (cnp->cn_flags & ISDOTDOT) {
953 			VOP_UNLOCK(dvp, 0);
954 			cnp->cn_flags |= PDIRUNLOCK;
955 			error = procfs_root(dvp->v_mount, vpp);
956 			if ((error == 0) && (wantpunlock == 0) &&
957 				    ((error = vn_lock(dvp, LK_EXCLUSIVE)) == 0))
958 				cnp->cn_flags &= ~PDIRUNLOCK;
959 			return (error);
960 		}
961 
962 		p = PFIND(pfs->pfs_pid);
963 		if (p == NULL)
964 			break;
965 		l = proc_representative_lwp(p);
966 
967 		for (pt = proc_targets, i = 0; i < nproc_targets; pt++, i++) {
968 			if (cnp->cn_namelen == pt->pt_namlen &&
969 			    memcmp(pt->pt_name, pname, cnp->cn_namelen) == 0 &&
970 			    (pt->pt_valid == NULL ||
971 			     (*pt->pt_valid)(cnp->cn_lwp, dvp->v_mount)))
972 				goto found;
973 		}
974 		break;
975 
976 	found:
977 		if (pt->pt_pfstype == PFSfile) {
978 			fvp = p->p_textvp;
979 			/* We already checked that it exists. */
980 			VREF(fvp);
981 			vn_lock(fvp, LK_EXCLUSIVE | LK_RETRY);
982 			if (wantpunlock) {
983 				VOP_UNLOCK(dvp, 0);
984 				cnp->cn_flags |= PDIRUNLOCK;
985 			}
986 			*vpp = fvp;
987 			return (0);
988 		}
989 
990 		error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
991 		    pt->pt_pfstype, -1);
992 		if ((error == 0) && (wantpunlock)) {
993 			VOP_UNLOCK(dvp, 0);
994 			cnp->cn_flags |= PDIRUNLOCK;
995 		}
996 		return (error);
997 
998 	case PFSfd: {
999 		int fd;
1000 		struct file *fp;
1001 		/*
1002 		 * do the .. dance. We unlock the directory, and then
1003 		 * get the proc dir. That will automatically return ..
1004 		 * locked. Then if the caller wanted dvp locked, we
1005 		 * re-lock.
1006 		 */
1007 		if (cnp->cn_flags & ISDOTDOT) {
1008 			VOP_UNLOCK(dvp, 0);
1009 			cnp->cn_flags |= PDIRUNLOCK;
1010 			error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
1011 			    PFSproc, -1);
1012 			if ((error == 0) && (wantpunlock == 0) &&
1013 				    ((error = vn_lock(dvp, LK_EXCLUSIVE)) == 0))
1014 				cnp->cn_flags &= ~PDIRUNLOCK;
1015 			return (error);
1016 		}
1017 		fd = atoi(pname, cnp->cn_namelen);
1018 		p = PFIND(pfs->pfs_pid);
1019 		if (p == NULL || (fp = fd_getfile(p->p_fd, fd)) == NULL)
1020 			return ENOENT;
1021 		FILE_USE(fp);
1022 
1023 		switch (fp->f_type) {
1024 		case DTYPE_VNODE:
1025 			fvp = (struct vnode *)fp->f_data;
1026 
1027 			/* Don't show directories */
1028 			if (fvp->v_type == VDIR)
1029 				goto symlink;
1030 
1031 			VREF(fvp);
1032 			FILE_UNUSE(fp, l);
1033 			vn_lock(fvp, LK_EXCLUSIVE | LK_RETRY |
1034 			    (p == curproc ? LK_CANRECURSE : 0));
1035 			*vpp = fvp;
1036 			error = 0;
1037 			break;
1038 		default:
1039 		symlink:
1040 			FILE_UNUSE(fp, l);
1041 			error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
1042 			    PFSfd, fd);
1043 			break;
1044 		}
1045 		if ((error == 0) && (wantpunlock)) {
1046 			VOP_UNLOCK(dvp, 0);
1047 			cnp->cn_flags |= PDIRUNLOCK;
1048 		}
1049 		return error;
1050 	}
1051 	default:
1052 		return (ENOTDIR);
1053 	}
1054 
1055 	return (cnp->cn_nameiop == LOOKUP ? ENOENT : EROFS);
1056 }
1057 
1058 int
1059 procfs_validfile(l, mp)
1060 	struct lwp *l;
1061 	struct mount *mp;
1062 {
1063 	return (l->l_proc->p_textvp != NULL);
1064 }
1065 
1066 static int
1067 procfs_validfile_linux(l, mp)
1068 	struct lwp *l;
1069 	struct mount *mp;
1070 {
1071 	int flags;
1072 
1073 	flags = VFSTOPROC(mp)->pmnt_flags;
1074 	return ((flags & PROCFSMNT_LINUXCOMPAT) &&
1075 	    (l == NULL || l->l_proc == NULL || procfs_validfile(l, mp)));
1076 }
1077 
1078 struct procfs_root_readdir_ctx {
1079 	struct uio *uiop;
1080 	off_t *cookies;
1081 	int ncookies;
1082 	off_t off;
1083 	off_t startoff;
1084 	int error;
1085 };
1086 
1087 static int
1088 procfs_root_readdir_callback(struct proc *p, void *arg)
1089 {
1090 	struct procfs_root_readdir_ctx *ctxp = arg;
1091 	struct dirent d;
1092 	struct uio *uiop;
1093 	int error;
1094 
1095 	uiop = ctxp->uiop;
1096 	if (uiop->uio_resid < UIO_MX)
1097 		return -1; /* no space */
1098 
1099 	if (ctxp->off < ctxp->startoff) {
1100 		ctxp->off++;
1101 		return 0;
1102 	}
1103 
1104 	if (CURTAIN(curlwp->l_proc->p_ucred->cr_uid, p->p_ucred->cr_uid))
1105 		return (0);
1106 
1107 	memset(&d, 0, UIO_MX);
1108 	d.d_reclen = UIO_MX;
1109 	d.d_fileno = PROCFS_FILENO(p->p_pid, PFSproc, -1);
1110 	d.d_namlen = snprintf(d.d_name,
1111 	    UIO_MX - offsetof(struct dirent, d_name), "%ld", (long)p->p_pid);
1112 	d.d_type = DT_DIR;
1113 
1114 	proclist_unlock_read();
1115 	error = uiomove(&d, UIO_MX, uiop);
1116 	proclist_lock_read();
1117 	if (error) {
1118 		ctxp->error = error;
1119 		return -1;
1120 	}
1121 
1122 	ctxp->ncookies++;
1123 	if (ctxp->cookies)
1124 		*(ctxp->cookies)++ = ctxp->off + 1;
1125 	ctxp->off++;
1126 
1127 	return 0;
1128 }
1129 
1130 /*
1131  * readdir returns directory entries from pfsnode (vp).
1132  *
1133  * the strategy here with procfs is to generate a single
1134  * directory entry at a time (struct dirent) and then
1135  * copy that out to userland using uiomove.  a more efficent
1136  * though more complex implementation, would try to minimize
1137  * the number of calls to uiomove().  for procfs, this is
1138  * hardly worth the added code complexity.
1139  *
1140  * this should just be done through read()
1141  */
1142 int
1143 procfs_readdir(v)
1144 	void *v;
1145 {
1146 	struct vop_readdir_args /* {
1147 		struct vnode *a_vp;
1148 		struct uio *a_uio;
1149 		struct ucred *a_cred;
1150 		int *a_eofflag;
1151 		off_t **a_cookies;
1152 		int *a_ncookies;
1153 	} */ *ap = v;
1154 	struct uio *uio = ap->a_uio;
1155 	struct dirent d;
1156 	struct pfsnode *pfs;
1157 	off_t i;
1158 	int error;
1159 	off_t *cookies = NULL;
1160 	int ncookies;
1161 	struct vnode *vp;
1162 	const struct proc_target *pt;
1163 	struct procfs_root_readdir_ctx ctx;
1164 
1165 	vp = ap->a_vp;
1166 	pfs = VTOPFS(vp);
1167 
1168 	if (uio->uio_resid < UIO_MX)
1169 		return (EINVAL);
1170 	if (uio->uio_offset < 0)
1171 		return (EINVAL);
1172 
1173 	error = 0;
1174 	i = uio->uio_offset;
1175 	memset(&d, 0, UIO_MX);
1176 	d.d_reclen = UIO_MX;
1177 	ncookies = uio->uio_resid / UIO_MX;
1178 
1179 	switch (pfs->pfs_type) {
1180 	/*
1181 	 * this is for the process-specific sub-directories.
1182 	 * all that is needed to is copy out all the entries
1183 	 * from the procent[] table (top of this file).
1184 	 */
1185 	case PFSproc: {
1186 		struct proc *p;
1187 
1188 		if (i >= nproc_targets)
1189 			return 0;
1190 
1191 		p = PFIND(pfs->pfs_pid);
1192 		if (p == NULL)
1193 			break;
1194 
1195 		if (ap->a_ncookies) {
1196 			ncookies = min(ncookies, (nproc_targets - i));
1197 			cookies = malloc(ncookies * sizeof (off_t),
1198 			    M_TEMP, M_WAITOK);
1199 			*ap->a_cookies = cookies;
1200 		}
1201 
1202 		for (pt = &proc_targets[i];
1203 		     uio->uio_resid >= UIO_MX && i < nproc_targets; pt++, i++) {
1204 			if (pt->pt_valid &&
1205 			    (*pt->pt_valid)(proc_representative_lwp(p), vp->v_mount) == 0)
1206 				continue;
1207 
1208 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid,
1209 			    pt->pt_pfstype, -1);
1210 			d.d_namlen = pt->pt_namlen;
1211 			memcpy(d.d_name, pt->pt_name, pt->pt_namlen + 1);
1212 			d.d_type = pt->pt_type;
1213 
1214 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1215 				break;
1216 			if (cookies)
1217 				*cookies++ = i + 1;
1218 		}
1219 
1220 	    	break;
1221 	}
1222 	case PFSfd: {
1223 		struct proc *p;
1224 		struct filedesc	*fdp;
1225 		struct file *fp;
1226 		int lim, nc = 0;
1227 
1228 		p = PFIND(pfs->pfs_pid);
1229 		if (p == NULL)
1230 			return ESRCH;
1231 
1232 		if (CURTAIN(curlwp->l_proc->p_ucred->cr_uid,
1233 			    p->p_ucred->cr_uid))
1234 			return (ESRCH);
1235 
1236 		fdp = p->p_fd;
1237 
1238 		lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfiles);
1239 		if (i >= lim)
1240 			return 0;
1241 
1242 		if (ap->a_ncookies) {
1243 			ncookies = min(ncookies, (fdp->fd_nfiles + 2 - i));
1244 			cookies = malloc(ncookies * sizeof (off_t),
1245 			    M_TEMP, M_WAITOK);
1246 			*ap->a_cookies = cookies;
1247 		}
1248 
1249 		for (; i < 2 && uio->uio_resid >= UIO_MX; i++) {
1250 			pt = &proc_targets[i];
1251 			d.d_namlen = pt->pt_namlen;
1252 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid,
1253 			    pt->pt_pfstype, -1);
1254 			(void)memcpy(d.d_name, pt->pt_name, pt->pt_namlen + 1);
1255 			d.d_type = pt->pt_type;
1256 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1257 				break;
1258 			if (cookies)
1259 				*cookies++ = i + 1;
1260 			nc++;
1261 		}
1262 		if (error) {
1263 			ncookies = nc;
1264 			break;
1265 		}
1266 		for (; uio->uio_resid >= UIO_MX && i < fdp->fd_nfiles; i++) {
1267 			/* check the descriptor exists */
1268 			if ((fp = fd_getfile(fdp, i - 2)) == NULL)
1269 				continue;
1270 			simple_unlock(&fp->f_slock);
1271 
1272 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid, PFSfd, i - 2);
1273 			d.d_namlen = snprintf(d.d_name, sizeof(d.d_name),
1274 			    "%lld", (long long)(i - 2));
1275 			d.d_type = VREG;
1276 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1277 				break;
1278 			if (cookies)
1279 				*cookies++ = i + 1;
1280 			nc++;
1281 		}
1282 		ncookies = nc;
1283 		break;
1284 	}
1285 
1286 	/*
1287 	 * this is for the root of the procfs filesystem
1288 	 * what is needed are special entries for "curproc"
1289 	 * and "self" followed by an entry for each process
1290 	 * on allproc.
1291 	 */
1292 
1293 	case PFSroot: {
1294 		int nc = 0;
1295 
1296 		if (ap->a_ncookies) {
1297 			/*
1298 			 * XXX Potentially allocating too much space here,
1299 			 * but I'm lazy. This loop needs some work.
1300 			 */
1301 			cookies = malloc(ncookies * sizeof (off_t),
1302 			    M_TEMP, M_WAITOK);
1303 			*ap->a_cookies = cookies;
1304 		}
1305 		error = 0;
1306 		/* 0 ... 3 are static entries. */
1307 		for (; i <= 3 && uio->uio_resid >= UIO_MX; i++) {
1308 			switch (i) {
1309 			case 0:		/* `.' */
1310 			case 1:		/* `..' */
1311 				d.d_fileno = PROCFS_FILENO(0, PFSroot, -1);
1312 				d.d_namlen = i + 1;
1313 				memcpy(d.d_name, "..", d.d_namlen);
1314 				d.d_name[i + 1] = '\0';
1315 				d.d_type = DT_DIR;
1316 				break;
1317 
1318 			case 2:
1319 				d.d_fileno = PROCFS_FILENO(0, PFScurproc, -1);
1320 				d.d_namlen = sizeof("curproc") - 1;
1321 				memcpy(d.d_name, "curproc", sizeof("curproc"));
1322 				d.d_type = DT_LNK;
1323 				break;
1324 
1325 			case 3:
1326 				d.d_fileno = PROCFS_FILENO(0, PFSself, -1);
1327 				d.d_namlen = sizeof("self") - 1;
1328 				memcpy(d.d_name, "self", sizeof("self"));
1329 				d.d_type = DT_LNK;
1330 				break;
1331 			}
1332 
1333 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1334 				break;
1335 			nc++;
1336 			if (cookies)
1337 				*cookies++ = i + 1;
1338 		}
1339 		/* 4 ... are process entries. */
1340 		ctx.uiop = uio;
1341 		ctx.error = 0;
1342 		ctx.off = 4;
1343 		ctx.startoff = i;
1344 		ctx.cookies = cookies;
1345 		ctx.ncookies = nc;
1346 		proclist_foreach_call(&allproc,
1347 		    procfs_root_readdir_callback, &ctx);
1348 		cookies = ctx.cookies;
1349 		nc = ctx.ncookies;
1350 		error = ctx.error;
1351 		if (error)
1352 			break;
1353 
1354 		/* misc entries. */
1355 		if (i < ctx.off)
1356 			i = ctx.off;
1357 		if (i >= ctx.off + nproc_root_targets)
1358 			break;
1359 		for (pt = &proc_root_targets[i - ctx.off];
1360 		    uio->uio_resid >= UIO_MX &&
1361 		    pt < &proc_root_targets[nproc_root_targets];
1362 		    pt++, i++) {
1363 			if (pt->pt_valid &&
1364 			    (*pt->pt_valid)(NULL, vp->v_mount) == 0)
1365 				continue;
1366 			d.d_fileno = PROCFS_FILENO(0, pt->pt_pfstype, -1);
1367 			d.d_namlen = pt->pt_namlen;
1368 			memcpy(d.d_name, pt->pt_name, pt->pt_namlen + 1);
1369 			d.d_type = pt->pt_type;
1370 
1371 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1372 				break;
1373 			nc++;
1374 			if (cookies)
1375 				*cookies++ = i + 1;
1376 		}
1377 
1378 		ncookies = nc;
1379 		break;
1380 	}
1381 
1382 	default:
1383 		error = ENOTDIR;
1384 		break;
1385 	}
1386 
1387 	if (ap->a_ncookies) {
1388 		if (error) {
1389 			if (cookies)
1390 				free(*ap->a_cookies, M_TEMP);
1391 			*ap->a_ncookies = 0;
1392 			*ap->a_cookies = NULL;
1393 		} else
1394 			*ap->a_ncookies = ncookies;
1395 	}
1396 	uio->uio_offset = i;
1397 	return (error);
1398 }
1399 
1400 /*
1401  * readlink reads the link of `curproc' and others
1402  */
1403 int
1404 procfs_readlink(v)
1405 	void *v;
1406 {
1407 	struct vop_readlink_args *ap = v;
1408 	char bf[16];		/* should be enough */
1409 	char *bp = bf;
1410 	char *path = NULL;
1411 	int len;
1412 	int error = 0;
1413 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
1414 	struct proc *pown;
1415 
1416 	if (pfs->pfs_fileno == PROCFS_FILENO(0, PFScurproc, -1))
1417 		len = snprintf(bf, sizeof(bf), "%ld", (long)curproc->p_pid);
1418 	else if (pfs->pfs_fileno == PROCFS_FILENO(0, PFSself, -1))
1419 		len = snprintf(bf, sizeof(bf), "%s", "curproc");
1420 	else if (pfs->pfs_fileno == PROCFS_FILENO(pfs->pfs_pid, PFScwd, -1)) {
1421 		pown = PFIND(pfs->pfs_pid);
1422 		if (pown == NULL)
1423 			return (ESRCH);
1424 		MALLOC(path, char *, MAXPATHLEN + 4, M_TEMP,
1425 		    M_WAITOK|M_CANFAIL);
1426 		if (path == NULL)
1427 			return (ENOMEM);
1428 		bp = path + MAXPATHLEN;
1429 		*--bp = '\0';
1430 		(void)procfs_dir(PFScwd, curlwp, pown, &bp, path,
1431 		     MAXPATHLEN);
1432 		len = strlen(bp);
1433 	}
1434 	else if (pfs->pfs_fileno == PROCFS_FILENO(pfs->pfs_pid, PFSchroot, -1)) {
1435 		pown = PFIND(pfs->pfs_pid);
1436 		if (pown == NULL)
1437 			return (ESRCH);
1438 		MALLOC(path, char *, MAXPATHLEN + 4, M_TEMP,
1439 		    M_WAITOK|M_CANFAIL);
1440 		if (path == NULL)
1441 			return (ENOMEM);
1442 		bp = path + MAXPATHLEN;
1443 		*--bp = '\0';
1444 		(void)procfs_dir(PFSchroot, curlwp, pown, &bp, path,
1445 		     MAXPATHLEN);
1446 		len = strlen(bp);
1447 	}
1448 	else {
1449 		struct file *fp;
1450 		struct vnode *vxp, *vp;
1451 
1452 		if ((error = procfs_getfp(pfs, &pown, &fp)) != 0)
1453 			return error;
1454 		FILE_USE(fp);
1455 		switch (fp->f_type) {
1456 		case DTYPE_VNODE:
1457 			vxp = (struct vnode *)fp->f_data;
1458 			if (vxp->v_type != VDIR) {
1459 				FILE_UNUSE(fp, proc_representative_lwp(pown));
1460 				return EINVAL;
1461 			}
1462 			if ((path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK))
1463 			    == NULL) {
1464 				FILE_UNUSE(fp, proc_representative_lwp(pown));
1465 				return ENOMEM;
1466 			}
1467 			bp = path + MAXPATHLEN;
1468 			*--bp = '\0';
1469 			vp = curproc->p_cwdi->cwdi_rdir;
1470 			if (vp == NULL)
1471 				vp = rootvnode;
1472 			error = getcwd_common(vxp, vp, &bp, path,
1473 			    MAXPATHLEN / 2, 0, curlwp);
1474 			FILE_UNUSE(fp, proc_representative_lwp(pown));
1475 			if (error) {
1476 				free(path, M_TEMP);
1477 				return error;
1478 			}
1479 			len = strlen(bp);
1480 			break;
1481 
1482 		case DTYPE_MISC:
1483 			len = snprintf(bf, sizeof(bf), "%s", "[misc]");
1484 			break;
1485 
1486 		case DTYPE_KQUEUE:
1487 			len = snprintf(bf, sizeof(bf), "%s", "[kqueue]");
1488 			break;
1489 
1490 		default:
1491 			return EINVAL;
1492 		}
1493 	}
1494 
1495 	error = uiomove(bp, len, ap->a_uio);
1496 	if (path)
1497 		free(path, M_TEMP);
1498 	return error;
1499 }
1500 
1501 /*
1502  * convert decimal ascii to int
1503  */
1504 static int
1505 atoi(b, len)
1506 	const char *b;
1507 	size_t len;
1508 {
1509 	int p = 0;
1510 
1511 	while (len--) {
1512 		char c = *b++;
1513 		if (c < '0' || c > '9')
1514 			return -1;
1515 		p = 10 * p + (c - '0');
1516 	}
1517 
1518 	return p;
1519 }
1520