xref: /netbsd-src/sys/miscfs/procfs/procfs_vnops.c (revision 23c8222edbfb0f0932d88a8351d3a0cf817dfb9e)
1 /*	$NetBSD: procfs_vnops.c,v 1.120 2004/10/04 08:40:18 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 1.120 2004/10/04 08:40:18 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 
97 #include <uvm/uvm_extern.h>	/* for PAGE_SIZE */
98 
99 #include <machine/reg.h>
100 
101 #include <miscfs/genfs/genfs.h>
102 #include <miscfs/procfs/procfs.h>
103 
104 /*
105  * Vnode Operations.
106  *
107  */
108 
109 static int procfs_validfile_linux __P((struct proc *, struct mount *));
110 static int procfs_root_readdir_callback(struct proc *, void *);
111 
112 /*
113  * This is a list of the valid names in the
114  * process-specific sub-directories.  It is
115  * used in procfs_lookup and procfs_readdir
116  */
117 static const struct proc_target {
118 	u_char	pt_type;
119 	u_char	pt_namlen;
120 	const char	*pt_name;
121 	pfstype	pt_pfstype;
122 	int	(*pt_valid) __P((struct proc *, struct mount *));
123 } proc_targets[] = {
124 #define N(s) sizeof(s)-1, s
125 	/*	  name		type		validp */
126 	{ DT_DIR, N("."),	PFSproc,		NULL },
127 	{ DT_DIR, N(".."),	PFSroot,		NULL },
128 	{ DT_DIR, N("fd"),	PFSfd,		NULL },
129 	{ DT_REG, N("file"),	PFSfile,		procfs_validfile },
130 	{ DT_REG, N("mem"),	PFSmem,		NULL },
131 	{ DT_REG, N("regs"),	PFSregs,		procfs_validregs },
132 	{ DT_REG, N("fpregs"),	PFSfpregs,	procfs_validfpregs },
133 	{ DT_REG, N("ctl"),	PFSctl,		NULL },
134 	{ DT_REG, N("stat"),	PFSstat,		procfs_validfile_linux },
135 	{ DT_REG, N("status"),	PFSstatus,	NULL },
136 	{ DT_REG, N("note"),	PFSnote,		NULL },
137 	{ DT_REG, N("notepg"),	PFSnotepg,	NULL },
138 	{ DT_REG, N("map"),	PFSmap,		procfs_validmap },
139 	{ DT_REG, N("maps"),	PFSmaps,		procfs_validmap },
140 	{ DT_REG, N("cmdline"), PFScmdline,	NULL },
141 	{ DT_REG, N("exe"),	PFSfile,		procfs_validfile_linux },
142 #ifdef __HAVE_PROCFS_MACHDEP
143 	PROCFS_MACHDEP_NODETYPE_DEFNS
144 #endif
145 #undef N
146 };
147 static const int nproc_targets = sizeof(proc_targets) / sizeof(proc_targets[0]);
148 
149 /*
150  * List of files in the root directory. Note: the validate function will
151  * be called with p == NULL for these ones.
152  */
153 static const struct proc_target proc_root_targets[] = {
154 #define N(s) sizeof(s)-1, s
155 	/*	  name		    type	    validp */
156 	{ DT_REG, N("meminfo"),     PFSmeminfo,        procfs_validfile_linux },
157 	{ DT_REG, N("cpuinfo"),     PFScpuinfo,        procfs_validfile_linux },
158 	{ DT_REG, N("uptime"),      PFSuptime,         procfs_validfile_linux },
159 	{ DT_REG, N("mounts"),	    PFSmounts,	       procfs_validfile_linux },
160 #undef N
161 };
162 static const int nproc_root_targets =
163     sizeof(proc_root_targets) / sizeof(proc_root_targets[0]);
164 
165 int	procfs_lookup	__P((void *));
166 #define	procfs_create	genfs_eopnotsupp
167 #define	procfs_mknod	genfs_eopnotsupp
168 int	procfs_open	__P((void *));
169 int	procfs_close	__P((void *));
170 int	procfs_access	__P((void *));
171 int	procfs_getattr	__P((void *));
172 int	procfs_setattr	__P((void *));
173 #define	procfs_read	procfs_rw
174 #define	procfs_write	procfs_rw
175 #define	procfs_fcntl	genfs_fcntl
176 #define	procfs_ioctl	genfs_enoioctl
177 #define	procfs_poll	genfs_poll
178 #define procfs_revoke	genfs_revoke
179 #define	procfs_fsync	genfs_nullop
180 #define	procfs_seek	genfs_nullop
181 #define	procfs_remove	genfs_eopnotsupp
182 int	procfs_link	__P((void *));
183 #define	procfs_rename	genfs_eopnotsupp
184 #define	procfs_mkdir	genfs_eopnotsupp
185 #define	procfs_rmdir	genfs_eopnotsupp
186 int	procfs_symlink	__P((void *));
187 int	procfs_readdir	__P((void *));
188 int	procfs_readlink	__P((void *));
189 #define	procfs_abortop	genfs_abortop
190 int	procfs_inactive	__P((void *));
191 int	procfs_reclaim	__P((void *));
192 #define	procfs_lock	genfs_lock
193 #define	procfs_unlock	genfs_unlock
194 #define	procfs_bmap	genfs_badop
195 #define	procfs_strategy	genfs_badop
196 int	procfs_print	__P((void *));
197 int	procfs_pathconf	__P((void *));
198 #define	procfs_islocked	genfs_islocked
199 #define	procfs_advlock	genfs_einval
200 #define	procfs_blkatoff	genfs_eopnotsupp
201 #define	procfs_valloc	genfs_eopnotsupp
202 #define	procfs_vfree	genfs_nullop
203 #define	procfs_truncate	genfs_eopnotsupp
204 #define	procfs_update	genfs_nullop
205 #define	procfs_bwrite	genfs_eopnotsupp
206 #define procfs_putpages	genfs_null_putpages
207 
208 static int atoi __P((const char *, size_t));
209 
210 /*
211  * procfs vnode operations.
212  */
213 int (**procfs_vnodeop_p) __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_blkatoff_desc, procfs_blkatoff },	/* blkatoff */
252 	{ &vop_valloc_desc, procfs_valloc },		/* valloc */
253 	{ &vop_vfree_desc, procfs_vfree },		/* vfree */
254 	{ &vop_truncate_desc, procfs_truncate },	/* truncate */
255 	{ &vop_update_desc, procfs_update },		/* update */
256 	{ &vop_putpages_desc, procfs_putpages },	/* putpages */
257 	{ NULL, NULL }
258 };
259 const struct vnodeopv_desc procfs_vnodeop_opv_desc =
260 	{ &procfs_vnodeop_p, procfs_vnodeop_entries };
261 /*
262  * set things up for doing i/o on
263  * the pfsnode (vp).  (vp) is locked
264  * on entry, and should be left locked
265  * on exit.
266  *
267  * for procfs we don't need to do anything
268  * in particular for i/o.  all that is done
269  * is to support exclusive open on process
270  * memory images.
271  */
272 int
273 procfs_open(v)
274 	void *v;
275 {
276 	struct vop_open_args /* {
277 		struct vnode *a_vp;
278 		int  a_mode;
279 		struct ucred *a_cred;
280 		struct proc *a_p;
281 	} */ *ap = v;
282 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
283 	struct proc *p1, *p2;
284 	int error;
285 
286 	p1 = ap->a_p;				/* tracer */
287 	p2 = PFIND(pfs->pfs_pid);		/* traced */
288 
289 	if (p2 == NULL)
290 		return (ENOENT);		/* was ESRCH, jsp */
291 
292 	switch (pfs->pfs_type) {
293 	case PFSmem:
294 		if (((pfs->pfs_flags & FWRITE) && (ap->a_mode & O_EXCL)) ||
295 		    ((pfs->pfs_flags & O_EXCL) && (ap->a_mode & FWRITE)))
296 			return (EBUSY);
297 
298 		if ((error = process_checkioperm(p1, p2)) != 0)
299 			return (error);
300 
301 		if (ap->a_mode & FWRITE)
302 			pfs->pfs_flags = ap->a_mode & (FWRITE|O_EXCL);
303 
304 		return (0);
305 
306 	default:
307 		break;
308 	}
309 
310 	return (0);
311 }
312 
313 /*
314  * close the pfsnode (vp) after doing i/o.
315  * (vp) is not locked on entry or exit.
316  *
317  * nothing to do for procfs other than undo
318  * any exclusive open flag (see _open above).
319  */
320 int
321 procfs_close(v)
322 	void *v;
323 {
324 	struct vop_close_args /* {
325 		struct vnode *a_vp;
326 		int  a_fflag;
327 		struct ucred *a_cred;
328 		struct proc *a_p;
329 	} */ *ap = v;
330 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
331 
332 	switch (pfs->pfs_type) {
333 	case PFSmem:
334 		if ((ap->a_fflag & FWRITE) && (pfs->pfs_flags & O_EXCL))
335 			pfs->pfs_flags &= ~(FWRITE|O_EXCL);
336 		break;
337 
338 	default:
339 		break;
340 	}
341 
342 	return (0);
343 }
344 
345 /*
346  * _inactive is called when the pfsnode
347  * is vrele'd and the reference count goes
348  * to zero.  (vp) will be on the vnode free
349  * list, so to get it back vget() must be
350  * used.
351  *
352  * for procfs, check if the process is still
353  * alive and if it isn't then just throw away
354  * the vnode by calling vgone().  this may
355  * be overkill and a waste of time since the
356  * chances are that the process will still be
357  * there and PFIND is not free.
358  *
359  * (vp) is locked on entry, but must be unlocked on exit.
360  */
361 int
362 procfs_inactive(v)
363 	void *v;
364 {
365 	struct vop_inactive_args /* {
366 		struct vnode *a_vp;
367 		struct proc *a_p;
368 	} */ *ap = v;
369 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
370 
371 	VOP_UNLOCK(ap->a_vp, 0);
372 	if (PFIND(pfs->pfs_pid) == NULL)
373 		vgone(ap->a_vp);
374 
375 	return (0);
376 }
377 
378 /*
379  * _reclaim is called when getnewvnode()
380  * wants to make use of an entry on the vnode
381  * free list.  at this time the filesystem needs
382  * to free any private data and remove the node
383  * from any private lists.
384  */
385 int
386 procfs_reclaim(v)
387 	void *v;
388 {
389 	struct vop_reclaim_args /* {
390 		struct vnode *a_vp;
391 	} */ *ap = v;
392 
393 	return (procfs_freevp(ap->a_vp));
394 }
395 
396 /*
397  * Return POSIX pathconf information applicable to special devices.
398  */
399 int
400 procfs_pathconf(v)
401 	void *v;
402 {
403 	struct vop_pathconf_args /* {
404 		struct vnode *a_vp;
405 		int a_name;
406 		register_t *a_retval;
407 	} */ *ap = v;
408 
409 	switch (ap->a_name) {
410 	case _PC_LINK_MAX:
411 		*ap->a_retval = LINK_MAX;
412 		return (0);
413 	case _PC_MAX_CANON:
414 		*ap->a_retval = MAX_CANON;
415 		return (0);
416 	case _PC_MAX_INPUT:
417 		*ap->a_retval = MAX_INPUT;
418 		return (0);
419 	case _PC_PIPE_BUF:
420 		*ap->a_retval = PIPE_BUF;
421 		return (0);
422 	case _PC_CHOWN_RESTRICTED:
423 		*ap->a_retval = 1;
424 		return (0);
425 	case _PC_VDISABLE:
426 		*ap->a_retval = _POSIX_VDISABLE;
427 		return (0);
428 	case _PC_SYNC_IO:
429 		*ap->a_retval = 1;
430 		return (0);
431 	default:
432 		return (EINVAL);
433 	}
434 	/* NOTREACHED */
435 }
436 
437 /*
438  * _print is used for debugging.
439  * just print a readable description
440  * of (vp).
441  */
442 int
443 procfs_print(v)
444 	void *v;
445 {
446 	struct vop_print_args /* {
447 		struct vnode *a_vp;
448 	} */ *ap = v;
449 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
450 
451 	printf("tag VT_PROCFS, type %d, pid %d, mode %x, flags %lx\n",
452 	    pfs->pfs_type, pfs->pfs_pid, pfs->pfs_mode, pfs->pfs_flags);
453 	return 0;
454 }
455 
456 int
457 procfs_link(v)
458 	void *v;
459 {
460 	struct vop_link_args /* {
461 		struct vnode *a_dvp;
462 		struct vnode *a_vp;
463 		struct componentname *a_cnp;
464 	} */ *ap = v;
465 
466 	VOP_ABORTOP(ap->a_dvp, ap->a_cnp);
467 	vput(ap->a_dvp);
468 	return (EROFS);
469 }
470 
471 int
472 procfs_symlink(v)
473 	void *v;
474 {
475 	struct vop_symlink_args /* {
476 		struct vnode *a_dvp;
477 		struct vnode **a_vpp;
478 		struct componentname *a_cnp;
479 		struct vattr *a_vap;
480 		char *a_target;
481 	} */ *ap = v;
482 
483 	VOP_ABORTOP(ap->a_dvp, ap->a_cnp);
484 	vput(ap->a_dvp);
485 	return (EROFS);
486 }
487 
488 /*
489  * Invent attributes for pfsnode (vp) and store
490  * them in (vap).
491  * Directories lengths are returned as zero since
492  * any real length would require the genuine size
493  * to be computed, and nothing cares anyway.
494  *
495  * this is relatively minimal for procfs.
496  */
497 int
498 procfs_getattr(v)
499 	void *v;
500 {
501 	struct vop_getattr_args /* {
502 		struct vnode *a_vp;
503 		struct vattr *a_vap;
504 		struct ucred *a_cred;
505 		struct proc *a_p;
506 	} */ *ap = v;
507 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
508 	struct vattr *vap = ap->a_vap;
509 	struct proc *procp;
510 	int error;
511 
512 	/* first check the process still exists */
513 	switch (pfs->pfs_type) {
514 	case PFSroot:
515 	case PFScurproc:
516 	case PFSself:
517 		procp = 0;
518 		break;
519 
520 	default:
521 		procp = PFIND(pfs->pfs_pid);
522 		if (procp == NULL)
523 			return (ENOENT);
524 		break;
525 	}
526 
527 	error = 0;
528 
529 	/* start by zeroing out the attributes */
530 	VATTR_NULL(vap);
531 
532 	/* next do all the common fields */
533 	vap->va_type = ap->a_vp->v_type;
534 	vap->va_mode = pfs->pfs_mode;
535 	vap->va_fileid = pfs->pfs_fileno;
536 	vap->va_flags = 0;
537 	vap->va_blocksize = PAGE_SIZE;
538 
539 	/*
540 	 * Make all times be current TOD.  Avoid microtime(9), it's slow.
541 	 * We don't guard the read from time(9) with splclock(9) since we
542 	 * don't actually need to be THAT sure the access is atomic.
543 	 *
544 	 * It would be possible to get the process start
545 	 * time from the p_stat structure, but there's
546 	 * no "file creation" time stamp anyway, and the
547 	 * p_stat structure is not addressible if u. gets
548 	 * swapped out for that process.
549 	 */
550 	TIMEVAL_TO_TIMESPEC(&time, &vap->va_ctime);
551 	vap->va_atime = vap->va_mtime = vap->va_ctime;
552 
553 	switch (pfs->pfs_type) {
554 	case PFSmem:
555 	case PFSregs:
556 	case PFSfpregs:
557 #if defined(__HAVE_PROCFS_MACHDEP) && defined(PROCFS_MACHDEP_PROTECT_CASES)
558 	PROCFS_MACHDEP_PROTECT_CASES
559 #endif
560 		/*
561 		 * If the process has exercised some setuid or setgid
562 		 * privilege, then rip away read/write permission so
563 		 * that only root can gain access.
564 		 */
565 		if (procp->p_flag & P_SUGID)
566 			vap->va_mode &= ~(S_IRUSR|S_IWUSR);
567 		/* FALLTHROUGH */
568 	case PFSctl:
569 	case PFSstatus:
570 	case PFSstat:
571 	case PFSnote:
572 	case PFSnotepg:
573 	case PFSmap:
574 	case PFSmaps:
575 	case PFScmdline:
576 		vap->va_nlink = 1;
577 		vap->va_uid = procp->p_ucred->cr_uid;
578 		vap->va_gid = procp->p_ucred->cr_gid;
579 		break;
580 	case PFSmeminfo:
581 	case PFScpuinfo:
582 	case PFSuptime:
583 	case PFSmounts:
584 		vap->va_nlink = 1;
585 		vap->va_uid = vap->va_gid = 0;
586 		break;
587 
588 	default:
589 		break;
590 	}
591 
592 	/*
593 	 * now do the object specific fields
594 	 *
595 	 * The size could be set from struct reg, but it's hardly
596 	 * worth the trouble, and it puts some (potentially) machine
597 	 * dependent data into this machine-independent code.  If it
598 	 * becomes important then this function should break out into
599 	 * a per-file stat function in the corresponding .c file.
600 	 */
601 
602 	switch (pfs->pfs_type) {
603 	case PFSroot:
604 		/*
605 		 * Set nlink to 1 to tell fts(3) we don't actually know.
606 		 */
607 		vap->va_nlink = 1;
608 		vap->va_uid = 0;
609 		vap->va_gid = 0;
610 		vap->va_bytes = vap->va_size = DEV_BSIZE;
611 		break;
612 
613 	case PFScurproc: {
614 		char buf[16];		/* should be enough */
615 		vap->va_nlink = 1;
616 		vap->va_uid = 0;
617 		vap->va_gid = 0;
618 		vap->va_bytes = vap->va_size =
619 		    snprintf(buf, sizeof(buf), "%ld", (long)curproc->p_pid);
620 		break;
621 	}
622 
623 	case PFSself:
624 		vap->va_nlink = 1;
625 		vap->va_uid = 0;
626 		vap->va_gid = 0;
627 		vap->va_bytes = vap->va_size = sizeof("curproc") - 1;
628 		break;
629 
630 	case PFSfd:
631 		if (pfs->pfs_fd != -1) {
632 			struct file *fp;
633 			struct proc *pown;
634 
635 			if ((error = procfs_getfp(pfs, &pown, &fp)) != 0)
636 				return error;
637 			FILE_USE(fp);
638 			vap->va_nlink = 1;
639 			vap->va_uid = fp->f_cred->cr_uid;
640 			vap->va_gid = fp->f_cred->cr_gid;
641 			switch (fp->f_type) {
642 			case DTYPE_VNODE:
643 				vap->va_bytes = vap->va_size =
644 				    ((struct vnode *)fp->f_data)->v_size;
645 				break;
646 			default:
647 				vap->va_bytes = vap->va_size = 0;
648 				break;
649 			}
650 			FILE_UNUSE(fp, pown);
651 			break;
652 		}
653 		/*FALLTHROUGH*/
654 	case PFSproc:
655 		vap->va_nlink = 2;
656 		vap->va_uid = procp->p_ucred->cr_uid;
657 		vap->va_gid = procp->p_ucred->cr_gid;
658 		vap->va_bytes = vap->va_size = DEV_BSIZE;
659 		break;
660 
661 	case PFSfile:
662 		error = EOPNOTSUPP;
663 		break;
664 
665 	case PFSmem:
666 		vap->va_bytes = vap->va_size =
667 			ctob(procp->p_vmspace->vm_tsize +
668 				    procp->p_vmspace->vm_dsize +
669 				    procp->p_vmspace->vm_ssize);
670 		break;
671 
672 #if defined(PT_GETREGS) || defined(PT_SETREGS)
673 	case PFSregs:
674 		vap->va_bytes = vap->va_size = sizeof(struct reg);
675 		break;
676 #endif
677 
678 #if defined(PT_GETFPREGS) || defined(PT_SETFPREGS)
679 	case PFSfpregs:
680 		vap->va_bytes = vap->va_size = sizeof(struct fpreg);
681 		break;
682 #endif
683 
684 	case PFSctl:
685 	case PFSstatus:
686 	case PFSstat:
687 	case PFSnote:
688 	case PFSnotepg:
689 	case PFScmdline:
690 	case PFSmeminfo:
691 	case PFScpuinfo:
692 	case PFSuptime:
693 	case PFSmounts:
694 		vap->va_bytes = vap->va_size = 0;
695 		break;
696 	case PFSmap:
697 	case PFSmaps:
698 		/*
699 		 * Advise a larger blocksize for the map files, so that
700 		 * they may be read in one pass.
701 		 */
702 		vap->va_blocksize = 4 * PAGE_SIZE;
703 		vap->va_bytes = vap->va_size = 0;
704 		break;
705 
706 #ifdef __HAVE_PROCFS_MACHDEP
707 	PROCFS_MACHDEP_NODETYPE_CASES
708 		error = procfs_machdep_getattr(ap->a_vp, vap, procp);
709 		break;
710 #endif
711 
712 	default:
713 		panic("procfs_getattr");
714 	}
715 
716 	return (error);
717 }
718 
719 /*ARGSUSED*/
720 int
721 procfs_setattr(v)
722 	void *v;
723 {
724 	/*
725 	 * just fake out attribute setting
726 	 * it's not good to generate an error
727 	 * return, otherwise things like creat()
728 	 * will fail when they try to set the
729 	 * file length to 0.  worse, this means
730 	 * that echo $note > /proc/$pid/note will fail.
731 	 */
732 
733 	return (0);
734 }
735 
736 /*
737  * implement access checking.
738  *
739  * actually, the check for super-user is slightly
740  * broken since it will allow read access to write-only
741  * objects.  this doesn't cause any particular trouble
742  * but does mean that the i/o entry points need to check
743  * that the operation really does make sense.
744  */
745 int
746 procfs_access(v)
747 	void *v;
748 {
749 	struct vop_access_args /* {
750 		struct vnode *a_vp;
751 		int a_mode;
752 		struct ucred *a_cred;
753 		struct proc *a_p;
754 	} */ *ap = v;
755 	struct vattr va;
756 	int error;
757 
758 	if ((error = VOP_GETATTR(ap->a_vp, &va, ap->a_cred, ap->a_p)) != 0)
759 		return (error);
760 
761 	return (vaccess(va.va_type, va.va_mode,
762 	    va.va_uid, va.va_gid, ap->a_mode, ap->a_cred));
763 }
764 
765 /*
766  * lookup.  this is incredibly complicated in the
767  * general case, however for most pseudo-filesystems
768  * very little needs to be done.
769  *
770  * Locking isn't hard here, just poorly documented.
771  *
772  * If we're looking up ".", just vref the parent & return it.
773  *
774  * If we're looking up "..", unlock the parent, and lock "..". If everything
775  * went ok, and we're on the last component and the caller requested the
776  * parent locked, try to re-lock the parent. We do this to prevent lock
777  * races.
778  *
779  * For anything else, get the needed node. Then unlock the parent if not
780  * the last component or not LOCKPARENT (i.e. if we wouldn't re-lock the
781  * parent in the .. case).
782  *
783  * We try to exit with the parent locked in error cases.
784  */
785 int
786 procfs_lookup(v)
787 	void *v;
788 {
789 	struct vop_lookup_args /* {
790 		struct vnode * a_dvp;
791 		struct vnode ** a_vpp;
792 		struct componentname * a_cnp;
793 	} */ *ap = v;
794 	struct componentname *cnp = ap->a_cnp;
795 	struct vnode **vpp = ap->a_vpp;
796 	struct vnode *dvp = ap->a_dvp;
797 	const char *pname = cnp->cn_nameptr;
798 	const struct proc_target *pt = NULL;
799 	struct vnode *fvp;
800 	pid_t pid;
801 	struct pfsnode *pfs;
802 	struct proc *p = NULL;
803 	int i, error, wantpunlock, iscurproc = 0, isself = 0;
804 
805 	*vpp = NULL;
806 	cnp->cn_flags &= ~PDIRUNLOCK;
807 
808 	if (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)
809 		return (EROFS);
810 
811 	if (cnp->cn_namelen == 1 && *pname == '.') {
812 		*vpp = dvp;
813 		VREF(dvp);
814 		return (0);
815 	}
816 
817 	wantpunlock = (~cnp->cn_flags & (LOCKPARENT | ISLASTCN));
818 	pfs = VTOPFS(dvp);
819 	switch (pfs->pfs_type) {
820 	case PFSroot:
821 		/*
822 		 * Shouldn't get here with .. in the root node.
823 		 */
824 		if (cnp->cn_flags & ISDOTDOT)
825 			return (EIO);
826 
827 		iscurproc = CNEQ(cnp, "curproc", 7);
828 		isself = CNEQ(cnp, "self", 4);
829 
830 		if (iscurproc || isself) {
831 			error = procfs_allocvp(dvp->v_mount, vpp, 0,
832 			    iscurproc ? PFScurproc : PFSself, -1);
833 			if ((error == 0) && (wantpunlock)) {
834 				VOP_UNLOCK(dvp, 0);
835 				cnp->cn_flags |= PDIRUNLOCK;
836 			}
837 			return (error);
838 		}
839 
840 		for (i = 0; i < nproc_root_targets; i++) {
841 			pt = &proc_root_targets[i];
842 			if (cnp->cn_namelen == pt->pt_namlen &&
843 			    memcmp(pt->pt_name, pname, cnp->cn_namelen) == 0 &&
844 			    (pt->pt_valid == NULL ||
845 			     (*pt->pt_valid)(p, dvp->v_mount)))
846 				break;
847 		}
848 
849 		if (i != nproc_root_targets) {
850 			error = procfs_allocvp(dvp->v_mount, vpp, 0,
851 			    pt->pt_pfstype, -1);
852 			if ((error == 0) && (wantpunlock)) {
853 				VOP_UNLOCK(dvp, 0);
854 				cnp->cn_flags |= PDIRUNLOCK;
855 			}
856 			return (error);
857 		}
858 
859 		pid = (pid_t)atoi(pname, cnp->cn_namelen);
860 
861 		p = PFIND(pid);
862 		if (p == NULL)
863 			break;
864 
865 		error = procfs_allocvp(dvp->v_mount, vpp, pid, PFSproc, -1);
866 		if ((error == 0) && (wantpunlock)) {
867 			VOP_UNLOCK(dvp, 0);
868 			cnp->cn_flags |= PDIRUNLOCK;
869 		}
870 		return (error);
871 
872 	case PFSproc:
873 		/*
874 		 * do the .. dance. We unlock the directory, and then
875 		 * get the root dir. That will automatically return ..
876 		 * locked. Then if the caller wanted dvp locked, we
877 		 * re-lock.
878 		 */
879 		if (cnp->cn_flags & ISDOTDOT) {
880 			VOP_UNLOCK(dvp, 0);
881 			cnp->cn_flags |= PDIRUNLOCK;
882 			error = procfs_root(dvp->v_mount, vpp);
883 			if ((error == 0) && (wantpunlock == 0) &&
884 				    ((error = vn_lock(dvp, LK_EXCLUSIVE)) == 0))
885 				cnp->cn_flags &= ~PDIRUNLOCK;
886 			return (error);
887 		}
888 
889 		p = PFIND(pfs->pfs_pid);
890 		if (p == NULL)
891 			break;
892 
893 		for (pt = proc_targets, i = 0; i < nproc_targets; pt++, i++) {
894 			if (cnp->cn_namelen == pt->pt_namlen &&
895 			    memcmp(pt->pt_name, pname, cnp->cn_namelen) == 0 &&
896 			    (pt->pt_valid == NULL ||
897 			     (*pt->pt_valid)(p, dvp->v_mount)))
898 				goto found;
899 		}
900 		break;
901 
902 	found:
903 		if (pt->pt_pfstype == PFSfile) {
904 			fvp = p->p_textvp;
905 			/* We already checked that it exists. */
906 			VREF(fvp);
907 			vn_lock(fvp, LK_EXCLUSIVE | LK_RETRY);
908 			if (wantpunlock) {
909 				VOP_UNLOCK(dvp, 0);
910 				cnp->cn_flags |= PDIRUNLOCK;
911 			}
912 			*vpp = fvp;
913 			return (0);
914 		}
915 
916 		error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
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 	case PFSfd: {
925 		int fd;
926 		struct file *fp;
927 		/*
928 		 * do the .. dance. We unlock the directory, and then
929 		 * get the proc dir. That will automatically return ..
930 		 * locked. Then if the caller wanted dvp locked, we
931 		 * re-lock.
932 		 */
933 		if (cnp->cn_flags & ISDOTDOT) {
934 			VOP_UNLOCK(dvp, 0);
935 			cnp->cn_flags |= PDIRUNLOCK;
936 			error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
937 			    PFSproc, -1);
938 			if ((error == 0) && (wantpunlock == 0) &&
939 				    ((error = vn_lock(dvp, LK_EXCLUSIVE)) == 0))
940 				cnp->cn_flags &= ~PDIRUNLOCK;
941 			return (error);
942 		}
943 		fd = atoi(pname, cnp->cn_namelen);
944 		p = PFIND(pfs->pfs_pid);
945 		if (p == NULL || (fp = fd_getfile(p->p_fd, fd)) == NULL)
946 			return ENOENT;
947 		FILE_USE(fp);
948 
949 		switch (fp->f_type) {
950 		case DTYPE_VNODE:
951 			fvp = (struct vnode *)fp->f_data;
952 
953 			/* Don't show directories */
954 			if (fvp->v_type == VDIR)
955 				goto symlink;
956 
957 			VREF(fvp);
958 			FILE_UNUSE(fp, p);
959 			vn_lock(fvp, LK_EXCLUSIVE | LK_RETRY |
960 			    (p == curproc ? LK_CANRECURSE : 0));
961 			*vpp = fvp;
962 			error = 0;
963 			break;
964 		default:
965 		symlink:
966 			FILE_UNUSE(fp, p);
967 			error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
968 			    PFSfd, fd);
969 			break;
970 		}
971 		if ((error == 0) && (wantpunlock)) {
972 			VOP_UNLOCK(dvp, 0);
973 			cnp->cn_flags |= PDIRUNLOCK;
974 		}
975 		return error;
976 	}
977 	default:
978 		return (ENOTDIR);
979 	}
980 
981 	return (cnp->cn_nameiop == LOOKUP ? ENOENT : EROFS);
982 }
983 
984 int
985 procfs_validfile(p, mp)
986 	struct proc *p;
987 	struct mount *mp;
988 {
989 	return (p->p_textvp != NULL);
990 }
991 
992 static int
993 procfs_validfile_linux(p, mp)
994 	struct proc *p;
995 	struct mount *mp;
996 {
997 	int flags;
998 
999 	flags = VFSTOPROC(mp)->pmnt_flags;
1000 	return ((flags & PROCFSMNT_LINUXCOMPAT) &&
1001 	    (p == NULL || procfs_validfile(p, mp)));
1002 }
1003 
1004 struct procfs_root_readdir_ctx {
1005 	struct uio *uiop;
1006 	off_t *cookies;
1007 	int ncookies;
1008 	off_t off;
1009 	off_t startoff;
1010 	int error;
1011 };
1012 
1013 static int
1014 procfs_root_readdir_callback(struct proc *p, void *arg)
1015 {
1016 	struct procfs_root_readdir_ctx *ctxp = arg;
1017 	struct dirent d;
1018 	struct uio *uiop;
1019 	int error;
1020 
1021 	uiop = ctxp->uiop;
1022 	if (uiop->uio_resid < UIO_MX)
1023 		return -1; /* no space */
1024 
1025 	if (ctxp->off < ctxp->startoff) {
1026 		ctxp->off++;
1027 		return 0;
1028 	}
1029 
1030 	memset(&d, 0, UIO_MX);
1031 	d.d_reclen = UIO_MX;
1032 	d.d_fileno = PROCFS_FILENO(p->p_pid, PFSproc, -1);
1033 	d.d_namlen = snprintf(d.d_name,
1034 	    UIO_MX - offsetof(struct dirent, d_name), "%ld", (long)p->p_pid);
1035 	d.d_type = DT_DIR;
1036 
1037 	proclist_unlock_read();
1038 	error = uiomove(&d, UIO_MX, uiop);
1039 	proclist_lock_read();
1040 	if (error) {
1041 		ctxp->error = error;
1042 		return -1;
1043 	}
1044 
1045 	ctxp->ncookies++;
1046 	if (ctxp->cookies)
1047 		*(ctxp->cookies)++ = ctxp->off + 1;
1048 	ctxp->off++;
1049 
1050 	return 0;
1051 }
1052 
1053 /*
1054  * readdir returns directory entries from pfsnode (vp).
1055  *
1056  * the strategy here with procfs is to generate a single
1057  * directory entry at a time (struct dirent) and then
1058  * copy that out to userland using uiomove.  a more efficent
1059  * though more complex implementation, would try to minimize
1060  * the number of calls to uiomove().  for procfs, this is
1061  * hardly worth the added code complexity.
1062  *
1063  * this should just be done through read()
1064  */
1065 int
1066 procfs_readdir(v)
1067 	void *v;
1068 {
1069 	struct vop_readdir_args /* {
1070 		struct vnode *a_vp;
1071 		struct uio *a_uio;
1072 		struct ucred *a_cred;
1073 		int *a_eofflag;
1074 		off_t **a_cookies;
1075 		int *a_ncookies;
1076 	} */ *ap = v;
1077 	struct uio *uio = ap->a_uio;
1078 	struct dirent d;
1079 	struct pfsnode *pfs;
1080 	off_t i;
1081 	int error;
1082 	off_t *cookies = NULL;
1083 	int ncookies;
1084 	struct vnode *vp;
1085 	const struct proc_target *pt;
1086 	struct procfs_root_readdir_ctx ctx;
1087 
1088 	vp = ap->a_vp;
1089 	pfs = VTOPFS(vp);
1090 
1091 	if (uio->uio_resid < UIO_MX)
1092 		return (EINVAL);
1093 	if (uio->uio_offset < 0)
1094 		return (EINVAL);
1095 
1096 	error = 0;
1097 	i = uio->uio_offset;
1098 	memset(&d, 0, UIO_MX);
1099 	d.d_reclen = UIO_MX;
1100 	ncookies = uio->uio_resid / UIO_MX;
1101 
1102 	switch (pfs->pfs_type) {
1103 	/*
1104 	 * this is for the process-specific sub-directories.
1105 	 * all that is needed to is copy out all the entries
1106 	 * from the procent[] table (top of this file).
1107 	 */
1108 	case PFSproc: {
1109 		struct proc *p;
1110 
1111 		if (i >= nproc_targets)
1112 			return 0;
1113 
1114 		p = PFIND(pfs->pfs_pid);
1115 		if (p == NULL)
1116 			break;
1117 
1118 		if (ap->a_ncookies) {
1119 			ncookies = min(ncookies, (nproc_targets - i));
1120 			cookies = malloc(ncookies * sizeof (off_t),
1121 			    M_TEMP, M_WAITOK);
1122 			*ap->a_cookies = cookies;
1123 		}
1124 
1125 		for (pt = &proc_targets[i];
1126 		     uio->uio_resid >= UIO_MX && i < nproc_targets; pt++, i++) {
1127 			if (pt->pt_valid &&
1128 			    (*pt->pt_valid)(p, vp->v_mount) == 0)
1129 				continue;
1130 
1131 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid,
1132 			    pt->pt_pfstype, -1);
1133 			d.d_namlen = pt->pt_namlen;
1134 			memcpy(d.d_name, pt->pt_name, pt->pt_namlen + 1);
1135 			d.d_type = pt->pt_type;
1136 
1137 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1138 				break;
1139 			if (cookies)
1140 				*cookies++ = i + 1;
1141 		}
1142 
1143 	    	break;
1144 	}
1145 	case PFSfd: {
1146 		struct proc *p;
1147 		struct filedesc	*fdp;
1148 		struct file *fp;
1149 		int lim, nc = 0;
1150 
1151 		p = PFIND(pfs->pfs_pid);
1152 		if (p == NULL)
1153 			return ESRCH;
1154 
1155 		fdp = p->p_fd;
1156 
1157 		lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfiles);
1158 		if (i >= lim)
1159 			return 0;
1160 
1161 		if (ap->a_ncookies) {
1162 			ncookies = min(ncookies, (fdp->fd_nfiles + 2 - i));
1163 			cookies = malloc(ncookies * sizeof (off_t),
1164 			    M_TEMP, M_WAITOK);
1165 			*ap->a_cookies = cookies;
1166 		}
1167 
1168 		for (; i < 2 && uio->uio_resid >= UIO_MX; i++) {
1169 			pt = &proc_targets[i];
1170 			d.d_namlen = pt->pt_namlen;
1171 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid,
1172 			    pt->pt_pfstype, -1);
1173 			(void)memcpy(d.d_name, pt->pt_name, pt->pt_namlen + 1);
1174 			d.d_type = pt->pt_type;
1175 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1176 				break;
1177 			if (cookies)
1178 				*cookies++ = i + 1;
1179 			nc++;
1180 		}
1181 		if (error) {
1182 			ncookies = nc;
1183 			break;
1184 		}
1185 		for (; uio->uio_resid >= UIO_MX && i < fdp->fd_nfiles; i++) {
1186 			/* check the descriptor exists */
1187 			if ((fp = fd_getfile(fdp, i - 2)) == NULL)
1188 				continue;
1189 			simple_unlock(&fp->f_slock);
1190 
1191 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid, PFSfd, i - 2);
1192 			d.d_namlen = snprintf(d.d_name, sizeof(d.d_name),
1193 			    "%lld", (long long)(i - 2));
1194 			d.d_type = VREG;
1195 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1196 				break;
1197 			if (cookies)
1198 				*cookies++ = i + 1;
1199 			nc++;
1200 		}
1201 		ncookies = nc;
1202 		break;
1203 	}
1204 
1205 	/*
1206 	 * this is for the root of the procfs filesystem
1207 	 * what is needed are special entries for "curproc"
1208 	 * and "self" followed by an entry for each process
1209 	 * on allproc.
1210 	 */
1211 
1212 	case PFSroot: {
1213 		int nc = 0;
1214 
1215 		if (ap->a_ncookies) {
1216 			/*
1217 			 * XXX Potentially allocating too much space here,
1218 			 * but I'm lazy. This loop needs some work.
1219 			 */
1220 			cookies = malloc(ncookies * sizeof (off_t),
1221 			    M_TEMP, M_WAITOK);
1222 			*ap->a_cookies = cookies;
1223 		}
1224 		error = 0;
1225 		/* 0 ... 3 are static entries. */
1226 		for (; i <= 3 && uio->uio_resid >= UIO_MX; i++) {
1227 			switch (i) {
1228 			case 0:		/* `.' */
1229 			case 1:		/* `..' */
1230 				d.d_fileno = PROCFS_FILENO(0, PFSroot, -1);
1231 				d.d_namlen = i + 1;
1232 				memcpy(d.d_name, "..", d.d_namlen);
1233 				d.d_name[i + 1] = '\0';
1234 				d.d_type = DT_DIR;
1235 				break;
1236 
1237 			case 2:
1238 				d.d_fileno = PROCFS_FILENO(0, PFScurproc, -1);
1239 				d.d_namlen = sizeof("curproc") - 1;
1240 				memcpy(d.d_name, "curproc", sizeof("curproc"));
1241 				d.d_type = DT_LNK;
1242 				break;
1243 
1244 			case 3:
1245 				d.d_fileno = PROCFS_FILENO(0, PFSself, -1);
1246 				d.d_namlen = sizeof("self") - 1;
1247 				memcpy(d.d_name, "self", sizeof("self"));
1248 				d.d_type = DT_LNK;
1249 				break;
1250 			}
1251 
1252 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1253 				break;
1254 			nc++;
1255 			if (cookies)
1256 				*cookies++ = i + 1;
1257 		}
1258 		/* 4 ... are process entries. */
1259 		ctx.uiop = uio;
1260 		ctx.error = 0;
1261 		ctx.off = 4;
1262 		ctx.startoff = i;
1263 		ctx.cookies = cookies;
1264 		ctx.ncookies = nc;
1265 		proclist_foreach_call(&allproc,
1266 		    procfs_root_readdir_callback, &ctx);
1267 		cookies = ctx.cookies;
1268 		nc = ctx.ncookies;
1269 		error = ctx.error;
1270 		if (error)
1271 			break;
1272 
1273 		/* misc entries. */
1274 		if (i < ctx.off)
1275 			i = ctx.off;
1276 		if (i >= ctx.off + nproc_root_targets)
1277 			break;
1278 		for (pt = &proc_root_targets[i - ctx.off];
1279 		    uio->uio_resid >= UIO_MX &&
1280 		    pt < &proc_root_targets[nproc_root_targets];
1281 		    pt++, i++) {
1282 			if (pt->pt_valid &&
1283 			    (*pt->pt_valid)(NULL, vp->v_mount) == 0)
1284 				continue;
1285 			d.d_fileno = PROCFS_FILENO(0, pt->pt_pfstype, -1);
1286 			d.d_namlen = pt->pt_namlen;
1287 			memcpy(d.d_name, pt->pt_name, pt->pt_namlen + 1);
1288 			d.d_type = pt->pt_type;
1289 
1290 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1291 				break;
1292 			nc++;
1293 			if (cookies)
1294 				*cookies++ = i + 1;
1295 		}
1296 
1297 		ncookies = nc;
1298 		break;
1299 	}
1300 
1301 	default:
1302 		error = ENOTDIR;
1303 		break;
1304 	}
1305 
1306 	if (ap->a_ncookies) {
1307 		if (error) {
1308 			if (cookies)
1309 				free(*ap->a_cookies, M_TEMP);
1310 			*ap->a_ncookies = 0;
1311 			*ap->a_cookies = NULL;
1312 		} else
1313 			*ap->a_ncookies = ncookies;
1314 	}
1315 	uio->uio_offset = i;
1316 	return (error);
1317 }
1318 
1319 /*
1320  * readlink reads the link of `curproc'
1321  */
1322 int
1323 procfs_readlink(v)
1324 	void *v;
1325 {
1326 	struct vop_readlink_args *ap = v;
1327 	char buf[16];		/* should be enough */
1328 	char *bp = buf;
1329 	char *path = NULL;
1330 	int len;
1331 	int error = 0;
1332 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
1333 
1334 	if (pfs->pfs_fileno == PROCFS_FILENO(0, PFScurproc, -1))
1335 		len = snprintf(buf, sizeof(buf), "%ld", (long)curproc->p_pid);
1336 	else if (pfs->pfs_fileno == PROCFS_FILENO(0, PFSself, -1))
1337 		len = snprintf(buf, sizeof(buf), "%s", "curproc");
1338 	else {
1339 		struct file *fp;
1340 		struct proc *pown;
1341 		struct vnode *vxp, *vp;
1342 
1343 		if ((error = procfs_getfp(pfs, &pown, &fp)) != 0)
1344 			return error;
1345 		FILE_USE(fp);
1346 		switch (fp->f_type) {
1347 		case DTYPE_VNODE:
1348 			vxp = (struct vnode *)fp->f_data;
1349 			if (vxp->v_type != VDIR) {
1350 				FILE_UNUSE(fp, pown);
1351 				return EINVAL;
1352 			}
1353 			if ((path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK))
1354 			    == NULL) {
1355 				FILE_UNUSE(fp, pown);
1356 				return ENOMEM;
1357 			}
1358 			bp = path + MAXPATHLEN;
1359 			*--bp = '\0';
1360 			vp = curproc->p_cwdi->cwdi_rdir;
1361 			if (vp == NULL)
1362 				vp = rootvnode;
1363 			error = getcwd_common(vxp, vp, &bp, path,
1364 			    MAXPATHLEN / 2, 0, curproc);
1365 			FILE_UNUSE(fp, pown);
1366 			if (error) {
1367 				free(path, M_TEMP);
1368 				return error;
1369 			}
1370 			len = strlen(bp);
1371 			break;
1372 
1373 		case DTYPE_MISC:
1374 			len = snprintf(buf, sizeof(buf), "%s", "[misc]");
1375 			break;
1376 
1377 		case DTYPE_KQUEUE:
1378 			len = snprintf(buf, sizeof(buf), "%s", "[kqueue]");
1379 			break;
1380 
1381 		default:
1382 			return EINVAL;
1383 		}
1384 	}
1385 
1386 	error = uiomove(bp, len, ap->a_uio);
1387 	if (path)
1388 		free(path, M_TEMP);
1389 	return error;
1390 }
1391 
1392 /*
1393  * convert decimal ascii to int
1394  */
1395 static int
1396 atoi(b, len)
1397 	const char *b;
1398 	size_t len;
1399 {
1400 	int p = 0;
1401 
1402 	while (len--) {
1403 		char c = *b++;
1404 		if (c < '0' || c > '9')
1405 			return -1;
1406 		p = 10 * p + (c - '0');
1407 	}
1408 
1409 	return p;
1410 }
1411