xref: /netbsd-src/sys/miscfs/procfs/procfs_vnops.c (revision 7c3f385475147b6e1c4753f2bee961630e2dfc40)
1 /*	$NetBSD: procfs_vnops.c,v 1.166 2008/03/21 21:55:00 ad Exp $	*/
2 
3 /*-
4  * Copyright (c) 2006, 2007, 2008 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Andrew Doran.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the NetBSD
21  *	Foundation, Inc. and its contributors.
22  * 4. Neither the name of The NetBSD Foundation nor the names of its
23  *    contributors may be used to endorse or promote products derived
24  *    from this software without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
27  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
28  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
29  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
30  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36  * POSSIBILITY OF SUCH DAMAGE.
37  */
38 
39 /*
40  * Copyright (c) 1993, 1995
41  *	The Regents of the University of California.  All rights reserved.
42  *
43  * This code is derived from software contributed to Berkeley by
44  * Jan-Simon Pendry.
45  *
46  * Redistribution and use in source and binary forms, with or without
47  * modification, are permitted provided that the following conditions
48  * are met:
49  * 1. Redistributions of source code must retain the above copyright
50  *    notice, this list of conditions and the following disclaimer.
51  * 2. Redistributions in binary form must reproduce the above copyright
52  *    notice, this list of conditions and the following disclaimer in the
53  *    documentation and/or other materials provided with the distribution.
54  * 3. Neither the name of the University nor the names of its contributors
55  *    may be used to endorse or promote products derived from this software
56  *    without specific prior written permission.
57  *
58  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
59  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
60  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
61  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
62  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
63  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
64  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
65  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
66  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
67  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
68  * SUCH DAMAGE.
69  *
70  *	@(#)procfs_vnops.c	8.18 (Berkeley) 5/21/95
71  */
72 
73 /*
74  * Copyright (c) 1993 Jan-Simon Pendry
75  *
76  * This code is derived from software contributed to Berkeley by
77  * Jan-Simon Pendry.
78  *
79  * Redistribution and use in source and binary forms, with or without
80  * modification, are permitted provided that the following conditions
81  * are met:
82  * 1. Redistributions of source code must retain the above copyright
83  *    notice, this list of conditions and the following disclaimer.
84  * 2. Redistributions in binary form must reproduce the above copyright
85  *    notice, this list of conditions and the following disclaimer in the
86  *    documentation and/or other materials provided with the distribution.
87  * 3. All advertising materials mentioning features or use of this software
88  *    must display the following acknowledgement:
89  *	This product includes software developed by the University of
90  *	California, Berkeley and its contributors.
91  * 4. Neither the name of the University nor the names of its contributors
92  *    may be used to endorse or promote products derived from this software
93  *    without specific prior written permission.
94  *
95  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
96  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
97  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
98  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
99  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
100  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
101  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
102  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
103  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
104  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
105  * SUCH DAMAGE.
106  *
107  *	@(#)procfs_vnops.c	8.18 (Berkeley) 5/21/95
108  */
109 
110 /*
111  * procfs vnode interface
112  */
113 
114 #include <sys/cdefs.h>
115 __KERNEL_RCSID(0, "$NetBSD: procfs_vnops.c,v 1.166 2008/03/21 21:55:00 ad Exp $");
116 
117 #include <sys/param.h>
118 #include <sys/systm.h>
119 #include <sys/time.h>
120 #include <sys/kernel.h>
121 #include <sys/file.h>
122 #include <sys/filedesc.h>
123 #include <sys/proc.h>
124 #include <sys/vnode.h>
125 #include <sys/namei.h>
126 #include <sys/malloc.h>
127 #include <sys/mount.h>
128 #include <sys/dirent.h>
129 #include <sys/resourcevar.h>
130 #include <sys/stat.h>
131 #include <sys/ptrace.h>
132 #include <sys/kauth.h>
133 
134 #include <uvm/uvm_extern.h>	/* for PAGE_SIZE */
135 
136 #include <machine/reg.h>
137 
138 #include <miscfs/genfs/genfs.h>
139 #include <miscfs/procfs/procfs.h>
140 
141 /*
142  * Vnode Operations.
143  *
144  */
145 
146 static int procfs_validfile_linux(struct lwp *, struct mount *);
147 static int procfs_root_readdir_callback(struct proc *, void *);
148 static void procfs_dir(pfstype, struct lwp *, struct proc *, char **, char *,
149     size_t);
150 
151 /*
152  * This is a list of the valid names in the
153  * process-specific sub-directories.  It is
154  * used in procfs_lookup and procfs_readdir
155  */
156 static const struct proc_target {
157 	u_char	pt_type;
158 	u_char	pt_namlen;
159 	const char	*pt_name;
160 	pfstype	pt_pfstype;
161 	int	(*pt_valid)(struct lwp *, struct mount *);
162 } proc_targets[] = {
163 #define N(s) sizeof(s)-1, s
164 	/*	  name		type		validp */
165 	{ DT_DIR, N("."),	PFSproc,	NULL },
166 	{ DT_DIR, N(".."),	PFSroot,	NULL },
167 	{ DT_DIR, N("fd"),	PFSfd,		NULL },
168 	{ DT_REG, N("file"),	PFSfile,	procfs_validfile },
169 	{ DT_REG, N("mem"),	PFSmem,		NULL },
170 	{ DT_REG, N("regs"),	PFSregs,	procfs_validregs },
171 	{ DT_REG, N("fpregs"),	PFSfpregs,	procfs_validfpregs },
172 	{ DT_REG, N("ctl"),	PFSctl,		NULL },
173 	{ DT_REG, N("stat"),	PFSstat,	procfs_validfile_linux },
174 	{ DT_REG, N("status"),	PFSstatus,	NULL },
175 	{ DT_REG, N("note"),	PFSnote,	NULL },
176 	{ DT_REG, N("notepg"),	PFSnotepg,	NULL },
177 	{ DT_REG, N("map"),	PFSmap,		procfs_validmap },
178 	{ DT_REG, N("maps"),	PFSmaps,	procfs_validmap },
179 	{ DT_REG, N("cmdline"), PFScmdline,	NULL },
180 	{ DT_REG, N("exe"),	PFSexe,		procfs_validfile },
181 	{ DT_LNK, N("cwd"),	PFScwd,		NULL },
182 	{ DT_LNK, N("root"),	PFSchroot,	NULL },
183 	{ DT_LNK, N("emul"),	PFSemul,	NULL },
184 	{ DT_REG, N("statm"),	PFSstatm,	procfs_validfile_linux },
185 #ifdef __HAVE_PROCFS_MACHDEP
186 	PROCFS_MACHDEP_NODETYPE_DEFNS
187 #endif
188 #undef N
189 };
190 static const int nproc_targets = sizeof(proc_targets) / sizeof(proc_targets[0]);
191 
192 /*
193  * List of files in the root directory. Note: the validate function will
194  * be called with p == NULL for these ones.
195  */
196 static const struct proc_target proc_root_targets[] = {
197 #define N(s) sizeof(s)-1, s
198 	/*	  name		    type	    validp */
199 	{ DT_REG, N("meminfo"),     PFSmeminfo,        procfs_validfile_linux },
200 	{ DT_REG, N("cpuinfo"),     PFScpuinfo,        procfs_validfile_linux },
201 	{ DT_REG, N("uptime"),      PFSuptime,         procfs_validfile_linux },
202 	{ DT_REG, N("mounts"),	    PFSmounts,	       procfs_validfile_linux },
203 	{ DT_REG, N("devices"),     PFSdevices,        procfs_validfile_linux },
204 	{ DT_REG, N("stat"),	    PFScpustat,        procfs_validfile_linux },
205 	{ DT_REG, N("loadavg"),	    PFSloadavg,        procfs_validfile_linux },
206 #undef N
207 };
208 static const int nproc_root_targets =
209     sizeof(proc_root_targets) / sizeof(proc_root_targets[0]);
210 
211 int	procfs_lookup(void *);
212 #define	procfs_create	genfs_eopnotsupp
213 #define	procfs_mknod	genfs_eopnotsupp
214 int	procfs_open(void *);
215 int	procfs_close(void *);
216 int	procfs_access(void *);
217 int	procfs_getattr(void *);
218 int	procfs_setattr(void *);
219 #define	procfs_read	procfs_rw
220 #define	procfs_write	procfs_rw
221 #define	procfs_fcntl	genfs_fcntl
222 #define	procfs_ioctl	genfs_enoioctl
223 #define	procfs_poll	genfs_poll
224 #define procfs_revoke	genfs_revoke
225 #define	procfs_fsync	genfs_nullop
226 #define	procfs_seek	genfs_nullop
227 #define	procfs_remove	genfs_eopnotsupp
228 int	procfs_link(void *);
229 #define	procfs_rename	genfs_eopnotsupp
230 #define	procfs_mkdir	genfs_eopnotsupp
231 #define	procfs_rmdir	genfs_eopnotsupp
232 int	procfs_symlink(void *);
233 int	procfs_readdir(void *);
234 int	procfs_readlink(void *);
235 #define	procfs_abortop	genfs_abortop
236 int	procfs_inactive(void *);
237 int	procfs_reclaim(void *);
238 #define	procfs_lock	genfs_lock
239 #define	procfs_unlock	genfs_unlock
240 #define	procfs_bmap	genfs_badop
241 #define	procfs_strategy	genfs_badop
242 int	procfs_print(void *);
243 int	procfs_pathconf(void *);
244 #define	procfs_islocked	genfs_islocked
245 #define	procfs_advlock	genfs_einval
246 #define	procfs_bwrite	genfs_eopnotsupp
247 #define procfs_putpages	genfs_null_putpages
248 
249 static int atoi(const char *, size_t);
250 
251 /*
252  * procfs vnode operations.
253  */
254 int (**procfs_vnodeop_p)(void *);
255 const struct vnodeopv_entry_desc procfs_vnodeop_entries[] = {
256 	{ &vop_default_desc, vn_default_error },
257 	{ &vop_lookup_desc, procfs_lookup },		/* lookup */
258 	{ &vop_create_desc, procfs_create },		/* create */
259 	{ &vop_mknod_desc, procfs_mknod },		/* mknod */
260 	{ &vop_open_desc, procfs_open },		/* open */
261 	{ &vop_close_desc, procfs_close },		/* close */
262 	{ &vop_access_desc, procfs_access },		/* access */
263 	{ &vop_getattr_desc, procfs_getattr },		/* getattr */
264 	{ &vop_setattr_desc, procfs_setattr },		/* setattr */
265 	{ &vop_read_desc, procfs_read },		/* read */
266 	{ &vop_write_desc, procfs_write },		/* write */
267 	{ &vop_fcntl_desc, procfs_fcntl },		/* fcntl */
268 	{ &vop_ioctl_desc, procfs_ioctl },		/* ioctl */
269 	{ &vop_poll_desc, procfs_poll },		/* poll */
270 	{ &vop_revoke_desc, procfs_revoke },		/* revoke */
271 	{ &vop_fsync_desc, procfs_fsync },		/* fsync */
272 	{ &vop_seek_desc, procfs_seek },		/* seek */
273 	{ &vop_remove_desc, procfs_remove },		/* remove */
274 	{ &vop_link_desc, procfs_link },		/* link */
275 	{ &vop_rename_desc, procfs_rename },		/* rename */
276 	{ &vop_mkdir_desc, procfs_mkdir },		/* mkdir */
277 	{ &vop_rmdir_desc, procfs_rmdir },		/* rmdir */
278 	{ &vop_symlink_desc, procfs_symlink },		/* symlink */
279 	{ &vop_readdir_desc, procfs_readdir },		/* readdir */
280 	{ &vop_readlink_desc, procfs_readlink },	/* readlink */
281 	{ &vop_abortop_desc, procfs_abortop },		/* abortop */
282 	{ &vop_inactive_desc, procfs_inactive },	/* inactive */
283 	{ &vop_reclaim_desc, procfs_reclaim },		/* reclaim */
284 	{ &vop_lock_desc, procfs_lock },		/* lock */
285 	{ &vop_unlock_desc, procfs_unlock },		/* unlock */
286 	{ &vop_bmap_desc, procfs_bmap },		/* bmap */
287 	{ &vop_strategy_desc, procfs_strategy },	/* strategy */
288 	{ &vop_print_desc, procfs_print },		/* print */
289 	{ &vop_islocked_desc, procfs_islocked },	/* islocked */
290 	{ &vop_pathconf_desc, procfs_pathconf },	/* pathconf */
291 	{ &vop_advlock_desc, procfs_advlock },		/* advlock */
292 	{ &vop_putpages_desc, procfs_putpages },	/* putpages */
293 	{ NULL, NULL }
294 };
295 const struct vnodeopv_desc procfs_vnodeop_opv_desc =
296 	{ &procfs_vnodeop_p, procfs_vnodeop_entries };
297 /*
298  * set things up for doing i/o on
299  * the pfsnode (vp).  (vp) is locked
300  * on entry, and should be left locked
301  * on exit.
302  *
303  * for procfs we don't need to do anything
304  * in particular for i/o.  all that is done
305  * is to support exclusive open on process
306  * memory images.
307  */
308 int
309 procfs_open(v)
310 	void *v;
311 {
312 	struct vop_open_args /* {
313 		struct vnode *a_vp;
314 		int  a_mode;
315 		kauth_cred_t a_cred;
316 	} */ *ap = v;
317 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
318 	struct lwp *l1;
319 	struct proc *p2;
320 	int error;
321 
322 	if ((error = procfs_proc_lock(pfs->pfs_pid, &p2, ENOENT)) != 0)
323 		return error;
324 
325 	l1 = curlwp;				/* tracer */
326 
327 #define	M2K(m)	(((m) & FREAD) && ((m) & FWRITE) ? \
328 		 KAUTH_REQ_PROCESS_PROCFS_RW : \
329 		 (m) & FWRITE ? KAUTH_REQ_PROCESS_PROCFS_WRITE : \
330 		 KAUTH_REQ_PROCESS_PROCFS_READ)
331 
332 	mutex_enter(&p2->p_mutex);
333 	error = kauth_authorize_process(l1->l_cred, KAUTH_PROCESS_PROCFS,
334 	    p2, pfs, KAUTH_ARG(M2K(ap->a_mode)), NULL);
335 	mutex_exit(&p2->p_mutex);
336 	if (error) {
337 		procfs_proc_unlock(p2);
338 		return (error);
339 	}
340 
341 #undef M2K
342 
343 	switch (pfs->pfs_type) {
344 	case PFSmem:
345 		if (((pfs->pfs_flags & FWRITE) && (ap->a_mode & O_EXCL)) ||
346 		    ((pfs->pfs_flags & O_EXCL) && (ap->a_mode & FWRITE))) {
347 			error = EBUSY;
348 			break;
349 		}
350 
351 		if (!proc_isunder(p2, l1)) {
352 			error = EPERM;
353 			break;
354 		}
355 
356 		if (ap->a_mode & FWRITE)
357 			pfs->pfs_flags = ap->a_mode & (FWRITE|O_EXCL);
358 
359 		break;
360 
361 	case PFSregs:
362 	case PFSfpregs:
363 		if (!proc_isunder(p2, l1)) {
364 			error = EPERM;
365 			break;
366 		}
367 		break;
368 
369 	default:
370 		break;
371 	}
372 
373 	procfs_proc_unlock(p2);
374 	return (error);
375 }
376 
377 /*
378  * close the pfsnode (vp) after doing i/o.
379  * (vp) is not locked on entry or exit.
380  *
381  * nothing to do for procfs other than undo
382  * any exclusive open flag (see _open above).
383  */
384 int
385 procfs_close(v)
386 	void *v;
387 {
388 	struct vop_close_args /* {
389 		struct vnode *a_vp;
390 		int  a_fflag;
391 		kauth_cred_t a_cred;
392 	} */ *ap = v;
393 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
394 
395 	switch (pfs->pfs_type) {
396 	case PFSmem:
397 		if ((ap->a_fflag & FWRITE) && (pfs->pfs_flags & O_EXCL))
398 			pfs->pfs_flags &= ~(FWRITE|O_EXCL);
399 		break;
400 
401 	default:
402 		break;
403 	}
404 
405 	return (0);
406 }
407 
408 /*
409  * _inactive is called when the pfsnode
410  * is vrele'd and the reference count goes
411  * to zero.  (vp) will be on the vnode free
412  * list, so to get it back vget() must be
413  * used.
414  *
415  * (vp) is locked on entry, but must be unlocked on exit.
416  */
417 int
418 procfs_inactive(v)
419 	void *v;
420 {
421 	struct vop_inactive_args /* {
422 		struct vnode *a_vp;
423 		bool *a_recycle;
424 	} */ *ap = v;
425 	struct vnode *vp = ap->a_vp;
426 	struct pfsnode *pfs = VTOPFS(vp);
427 
428 	mutex_enter(&proclist_lock);
429 	*ap->a_recycle = (p_find(pfs->pfs_pid, PFIND_LOCKED) == NULL);
430 	mutex_exit(&proclist_lock);
431 
432 	VOP_UNLOCK(vp, 0);
433 
434 	return (0);
435 }
436 
437 /*
438  * _reclaim is called when getnewvnode()
439  * wants to make use of an entry on the vnode
440  * free list.  at this time the filesystem needs
441  * to free any private data and remove the node
442  * from any private lists.
443  */
444 int
445 procfs_reclaim(v)
446 	void *v;
447 {
448 	struct vop_reclaim_args /* {
449 		struct vnode *a_vp;
450 	} */ *ap = v;
451 
452 	return (procfs_freevp(ap->a_vp));
453 }
454 
455 /*
456  * Return POSIX pathconf information applicable to special devices.
457  */
458 int
459 procfs_pathconf(v)
460 	void *v;
461 {
462 	struct vop_pathconf_args /* {
463 		struct vnode *a_vp;
464 		int a_name;
465 		register_t *a_retval;
466 	} */ *ap = v;
467 
468 	switch (ap->a_name) {
469 	case _PC_LINK_MAX:
470 		*ap->a_retval = LINK_MAX;
471 		return (0);
472 	case _PC_MAX_CANON:
473 		*ap->a_retval = MAX_CANON;
474 		return (0);
475 	case _PC_MAX_INPUT:
476 		*ap->a_retval = MAX_INPUT;
477 		return (0);
478 	case _PC_PIPE_BUF:
479 		*ap->a_retval = PIPE_BUF;
480 		return (0);
481 	case _PC_CHOWN_RESTRICTED:
482 		*ap->a_retval = 1;
483 		return (0);
484 	case _PC_VDISABLE:
485 		*ap->a_retval = _POSIX_VDISABLE;
486 		return (0);
487 	case _PC_SYNC_IO:
488 		*ap->a_retval = 1;
489 		return (0);
490 	default:
491 		return (EINVAL);
492 	}
493 	/* NOTREACHED */
494 }
495 
496 /*
497  * _print is used for debugging.
498  * just print a readable description
499  * of (vp).
500  */
501 int
502 procfs_print(v)
503 	void *v;
504 {
505 	struct vop_print_args /* {
506 		struct vnode *a_vp;
507 	} */ *ap = v;
508 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
509 
510 	printf("tag VT_PROCFS, type %d, pid %d, mode %x, flags %lx\n",
511 	    pfs->pfs_type, pfs->pfs_pid, pfs->pfs_mode, pfs->pfs_flags);
512 	return 0;
513 }
514 
515 int
516 procfs_link(v)
517 	void *v;
518 {
519 	struct vop_link_args /* {
520 		struct vnode *a_dvp;
521 		struct vnode *a_vp;
522 		struct componentname *a_cnp;
523 	} */ *ap = v;
524 
525 	VOP_ABORTOP(ap->a_dvp, ap->a_cnp);
526 	vput(ap->a_dvp);
527 	return (EROFS);
528 }
529 
530 int
531 procfs_symlink(v)
532 	void *v;
533 {
534 	struct vop_symlink_args /* {
535 		struct vnode *a_dvp;
536 		struct vnode **a_vpp;
537 		struct componentname *a_cnp;
538 		struct vattr *a_vap;
539 		char *a_target;
540 	} */ *ap = v;
541 
542 	VOP_ABORTOP(ap->a_dvp, ap->a_cnp);
543 	vput(ap->a_dvp);
544 	return (EROFS);
545 }
546 
547 /*
548  * Works out the path to (and vnode of) the target process's current
549  * working directory or chroot.  If the caller is in a chroot and
550  * can't "reach" the target's cwd or root (or some other error
551  * occurs), a "/" is returned for the path and a NULL pointer is
552  * returned for the vnode.
553  */
554 static void
555 procfs_dir(pfstype t, struct lwp *caller, struct proc *target, char **bpp,
556     char *path, size_t len)
557 {
558 	struct cwdinfo *cwdi;
559 	struct vnode *vp, *rvp;
560 	char *bp;
561 
562 	cwdi = caller->l_proc->p_cwdi;
563 	rw_enter(&cwdi->cwdi_lock, RW_READER);
564 
565 	rvp = cwdi->cwdi_rdir;
566 	bp = bpp ? *bpp : NULL;
567 
568 	switch (t) {
569 	case PFScwd:
570 		vp = target->p_cwdi->cwdi_cdir;
571 		break;
572 	case PFSchroot:
573 		vp = target->p_cwdi->cwdi_rdir;
574 		break;
575 	case PFSexe:
576 		vp = target->p_textvp;
577 		break;
578 	default:
579 		rw_exit(&cwdi->cwdi_lock);
580 		return;
581 	}
582 
583 	/*
584 	 * XXX: this horrible kludge avoids locking panics when
585 	 * attempting to lookup links that point to within procfs
586 	 */
587 	if (vp != NULL && vp->v_tag == VT_PROCFS) {
588 		if (bpp) {
589 			*--bp = '/';
590 			*bpp = bp;
591 		}
592 		rw_exit(&cwdi->cwdi_lock);
593 		return;
594 	}
595 
596 	if (rvp == NULL)
597 		rvp = rootvnode;
598 	if (vp == NULL || getcwd_common(vp, rvp, bp ? &bp : NULL, path,
599 	    len / 2, 0, caller) != 0) {
600 		vp = NULL;
601 		if (bpp) {
602 /*
603 			if (t == PFSexe) {
604 				snprintf(path, len, "%s/%d/file"
605 				    mp->mnt_stat.f_mntonname, pfs->pfs_pid);
606 			} else */ {
607 				bp = *bpp;
608 				*--bp = '/';
609 			}
610 		}
611 	}
612 
613 	if (bpp)
614 		*bpp = bp;
615 
616 	rw_exit(&cwdi->cwdi_lock);
617 }
618 
619 /*
620  * Invent attributes for pfsnode (vp) and store
621  * them in (vap).
622  * Directories lengths are returned as zero since
623  * any real length would require the genuine size
624  * to be computed, and nothing cares anyway.
625  *
626  * this is relatively minimal for procfs.
627  */
628 int
629 procfs_getattr(v)
630 	void *v;
631 {
632 	struct vop_getattr_args /* {
633 		struct vnode *a_vp;
634 		struct vattr *a_vap;
635 		kauth_cred_t a_cred;
636 	} */ *ap = v;
637 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
638 	struct vattr *vap = ap->a_vap;
639 	struct proc *procp;
640 	char *path;
641 	int error;
642 
643 	/* first check the process still exists */
644 	switch (pfs->pfs_type) {
645 	case PFSroot:
646 	case PFScurproc:
647 	case PFSself:
648 		procp = NULL;
649 		break;
650 
651 	default:
652 		error = procfs_proc_lock(pfs->pfs_pid, &procp, ENOENT);
653 		if (error != 0)
654 			return (error);
655 		break;
656 	}
657 
658 	switch (pfs->pfs_type) {
659 	case PFScwd:
660 	case PFSchroot:
661 	case PFSexe:
662 		MALLOC(path, char *, MAXPATHLEN + 4, M_TEMP,
663 		    M_WAITOK|M_CANFAIL);
664 		if (path == NULL && procp != NULL) {
665 			procfs_proc_unlock(procp);
666 			return (ENOMEM);
667 		}
668 		break;
669 
670 	default:
671 		path = NULL;
672 		break;
673 	}
674 
675 	if (procp != NULL) {
676 		mutex_enter(&procp->p_mutex);
677 		error = kauth_authorize_process(kauth_cred_get(),
678 		    KAUTH_PROCESS_CANSEE, procp,
679 		    KAUTH_ARG(KAUTH_REQ_PROCESS_CANSEE_ENTRY), NULL, NULL);
680 		mutex_exit(&procp->p_mutex);
681 		if (error != 0) {
682 		    	procfs_proc_unlock(procp);
683 		    	if (path != NULL)
684 		    		free(path, M_TEMP);
685 			return (ENOENT);
686 		}
687 	}
688 
689 	error = 0;
690 
691 	/* start by zeroing out the attributes */
692 	VATTR_NULL(vap);
693 
694 	/* next do all the common fields */
695 	vap->va_type = ap->a_vp->v_type;
696 	vap->va_mode = pfs->pfs_mode;
697 	vap->va_fileid = pfs->pfs_fileno;
698 	vap->va_flags = 0;
699 	vap->va_blocksize = PAGE_SIZE;
700 
701 	/*
702 	 * Make all times be current TOD.
703 	 *
704 	 * It would be possible to get the process start
705 	 * time from the p_stats structure, but there's
706 	 * no "file creation" time stamp anyway, and the
707 	 * p_stats structure is not addressable if u. gets
708 	 * swapped out for that process.
709 	 */
710 	getnanotime(&vap->va_ctime);
711 	vap->va_atime = vap->va_mtime = vap->va_ctime;
712 	if (procp)
713 		TIMEVAL_TO_TIMESPEC(&procp->p_stats->p_start,
714 		    &vap->va_birthtime);
715 	else
716 		getnanotime(&vap->va_birthtime);
717 
718 	switch (pfs->pfs_type) {
719 	case PFSmem:
720 	case PFSregs:
721 	case PFSfpregs:
722 #if defined(__HAVE_PROCFS_MACHDEP) && defined(PROCFS_MACHDEP_PROTECT_CASES)
723 	PROCFS_MACHDEP_PROTECT_CASES
724 #endif
725 		/*
726 		 * If the process has exercised some setuid or setgid
727 		 * privilege, then rip away read/write permission so
728 		 * that only root can gain access.
729 		 */
730 		if (procp->p_flag & PK_SUGID)
731 			vap->va_mode &= ~(S_IRUSR|S_IWUSR);
732 		/* FALLTHROUGH */
733 	case PFSctl:
734 	case PFSstatus:
735 	case PFSstat:
736 	case PFSnote:
737 	case PFSnotepg:
738 	case PFSmap:
739 	case PFSmaps:
740 	case PFScmdline:
741 	case PFSemul:
742 	case PFSstatm:
743 		vap->va_nlink = 1;
744 		vap->va_uid = kauth_cred_geteuid(procp->p_cred);
745 		vap->va_gid = kauth_cred_getegid(procp->p_cred);
746 		break;
747 	case PFSmeminfo:
748 	case PFSdevices:
749 	case PFScpuinfo:
750 	case PFSuptime:
751 	case PFSmounts:
752 	case PFScpustat:
753 	case PFSloadavg:
754 		vap->va_nlink = 1;
755 		vap->va_uid = vap->va_gid = 0;
756 		break;
757 
758 	default:
759 		break;
760 	}
761 
762 	/*
763 	 * now do the object specific fields
764 	 *
765 	 * The size could be set from struct reg, but it's hardly
766 	 * worth the trouble, and it puts some (potentially) machine
767 	 * dependent data into this machine-independent code.  If it
768 	 * becomes important then this function should break out into
769 	 * a per-file stat function in the corresponding .c file.
770 	 */
771 
772 	switch (pfs->pfs_type) {
773 	case PFSroot:
774 		/*
775 		 * Set nlink to 1 to tell fts(3) we don't actually know.
776 		 */
777 		vap->va_nlink = 1;
778 		vap->va_uid = 0;
779 		vap->va_gid = 0;
780 		vap->va_bytes = vap->va_size = DEV_BSIZE;
781 		break;
782 
783 	case PFSself:
784 	case PFScurproc: {
785 		char bf[16];		/* should be enough */
786 		vap->va_nlink = 1;
787 		vap->va_uid = 0;
788 		vap->va_gid = 0;
789 		vap->va_bytes = vap->va_size =
790 		    snprintf(bf, sizeof(bf), "%ld", (long)curproc->p_pid);
791 		break;
792 	}
793 
794 	case PFSfd:
795 		if (pfs->pfs_fd != -1) {
796 			file_t *fp;
797 
798 			fp = fd_getfile2(procp, pfs->pfs_fd);
799 			if (fp == NULL) {
800 				error = EBADF;
801 				break;
802 			}
803 			vap->va_nlink = 1;
804 			vap->va_uid = kauth_cred_geteuid(fp->f_cred);
805 			vap->va_gid = kauth_cred_getegid(fp->f_cred);
806 			switch (fp->f_type) {
807 			case DTYPE_VNODE:
808 				vap->va_bytes = vap->va_size =
809 				    ((struct vnode *)fp->f_data)->v_size;
810 				break;
811 			default:
812 				vap->va_bytes = vap->va_size = 0;
813 				break;
814 			}
815 			closef(fp);
816 			break;
817 		}
818 		/*FALLTHROUGH*/
819 	case PFSproc:
820 		vap->va_nlink = 2;
821 		vap->va_uid = kauth_cred_geteuid(procp->p_cred);
822 		vap->va_gid = kauth_cred_getegid(procp->p_cred);
823 		vap->va_bytes = vap->va_size = DEV_BSIZE;
824 		break;
825 
826 	case PFSfile:
827 		error = EOPNOTSUPP;
828 		break;
829 
830 	case PFSmem:
831 		vap->va_bytes = vap->va_size =
832 			ctob(procp->p_vmspace->vm_tsize +
833 				    procp->p_vmspace->vm_dsize +
834 				    procp->p_vmspace->vm_ssize);
835 		break;
836 
837 #if defined(PT_GETREGS) || defined(PT_SETREGS)
838 	case PFSregs:
839 		vap->va_bytes = vap->va_size = sizeof(struct reg);
840 		break;
841 #endif
842 
843 #if defined(PT_GETFPREGS) || defined(PT_SETFPREGS)
844 	case PFSfpregs:
845 		vap->va_bytes = vap->va_size = sizeof(struct fpreg);
846 		break;
847 #endif
848 
849 	case PFSctl:
850 	case PFSstatus:
851 	case PFSstat:
852 	case PFSnote:
853 	case PFSnotepg:
854 	case PFScmdline:
855 	case PFSmeminfo:
856 	case PFSdevices:
857 	case PFScpuinfo:
858 	case PFSuptime:
859 	case PFSmounts:
860 	case PFScpustat:
861 	case PFSloadavg:
862 	case PFSstatm:
863 		vap->va_bytes = vap->va_size = 0;
864 		break;
865 	case PFSmap:
866 	case PFSmaps:
867 		/*
868 		 * Advise a larger blocksize for the map files, so that
869 		 * they may be read in one pass.
870 		 */
871 		vap->va_blocksize = 4 * PAGE_SIZE;
872 		vap->va_bytes = vap->va_size = 0;
873 		break;
874 
875 	case PFScwd:
876 	case PFSchroot:
877 	case PFSexe: {
878 		char *bp;
879 
880 		vap->va_nlink = 1;
881 		vap->va_uid = 0;
882 		vap->va_gid = 0;
883 		bp = path + MAXPATHLEN;
884 		*--bp = '\0';
885 		procfs_dir(pfs->pfs_type, curlwp, procp, &bp, path,
886 		     MAXPATHLEN);
887 		vap->va_bytes = vap->va_size = strlen(bp);
888 		break;
889 	}
890 
891 	case PFSemul:
892 		vap->va_bytes = vap->va_size = strlen(procp->p_emul->e_name);
893 		break;
894 
895 #ifdef __HAVE_PROCFS_MACHDEP
896 	PROCFS_MACHDEP_NODETYPE_CASES
897 		error = procfs_machdep_getattr(ap->a_vp, vap, procp);
898 		break;
899 #endif
900 
901 	default:
902 		panic("procfs_getattr");
903 	}
904 
905 	if (procp != NULL)
906 		procfs_proc_unlock(procp);
907 	if (path != NULL)
908 		free(path, M_TEMP);
909 
910 	return (error);
911 }
912 
913 /*ARGSUSED*/
914 int
915 procfs_setattr(void *v)
916 {
917 	/*
918 	 * just fake out attribute setting
919 	 * it's not good to generate an error
920 	 * return, otherwise things like creat()
921 	 * will fail when they try to set the
922 	 * file length to 0.  worse, this means
923 	 * that echo $note > /proc/$pid/note will fail.
924 	 */
925 
926 	return (0);
927 }
928 
929 /*
930  * implement access checking.
931  *
932  * actually, the check for super-user is slightly
933  * broken since it will allow read access to write-only
934  * objects.  this doesn't cause any particular trouble
935  * but does mean that the i/o entry points need to check
936  * that the operation really does make sense.
937  */
938 int
939 procfs_access(v)
940 	void *v;
941 {
942 	struct vop_access_args /* {
943 		struct vnode *a_vp;
944 		int a_mode;
945 		kauth_cred_t a_cred;
946 	} */ *ap = v;
947 	struct vattr va;
948 	int error;
949 
950 	if ((error = VOP_GETATTR(ap->a_vp, &va, ap->a_cred)) != 0)
951 		return (error);
952 
953 	return (vaccess(va.va_type, va.va_mode,
954 	    va.va_uid, va.va_gid, ap->a_mode, ap->a_cred));
955 }
956 
957 /*
958  * lookup.  this is incredibly complicated in the
959  * general case, however for most pseudo-filesystems
960  * very little needs to be done.
961  *
962  * Locking isn't hard here, just poorly documented.
963  *
964  * If we're looking up ".", just vref the parent & return it.
965  *
966  * If we're looking up "..", unlock the parent, and lock "..". If everything
967  * went ok, and we're on the last component and the caller requested the
968  * parent locked, try to re-lock the parent. We do this to prevent lock
969  * races.
970  *
971  * For anything else, get the needed node. Then unlock the parent if not
972  * the last component or not LOCKPARENT (i.e. if we wouldn't re-lock the
973  * parent in the .. case).
974  *
975  * We try to exit with the parent locked in error cases.
976  */
977 int
978 procfs_lookup(v)
979 	void *v;
980 {
981 	struct vop_lookup_args /* {
982 		struct vnode * a_dvp;
983 		struct vnode ** a_vpp;
984 		struct componentname * a_cnp;
985 	} */ *ap = v;
986 	struct componentname *cnp = ap->a_cnp;
987 	struct vnode **vpp = ap->a_vpp;
988 	struct vnode *dvp = ap->a_dvp;
989 	const char *pname = cnp->cn_nameptr;
990 	const struct proc_target *pt = NULL;
991 	struct vnode *fvp;
992 	pid_t pid, vnpid;
993 	struct pfsnode *pfs;
994 	struct proc *p = NULL;
995 	int i, error;
996 	pfstype type;
997 
998 	*vpp = NULL;
999 
1000 	if (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)
1001 		return (EROFS);
1002 
1003 	if (cnp->cn_namelen == 1 && *pname == '.') {
1004 		*vpp = dvp;
1005 		VREF(dvp);
1006 		return (0);
1007 	}
1008 
1009 	pfs = VTOPFS(dvp);
1010 	switch (pfs->pfs_type) {
1011 	case PFSroot:
1012 		/*
1013 		 * Shouldn't get here with .. in the root node.
1014 		 */
1015 		if (cnp->cn_flags & ISDOTDOT)
1016 			return (EIO);
1017 
1018 		for (i = 0; i < nproc_root_targets; i++) {
1019 			pt = &proc_root_targets[i];
1020 			/*
1021 			 * check for node match.  proc is always NULL here,
1022 			 * so call pt_valid with constant NULL lwp.
1023 			 */
1024 			if (cnp->cn_namelen == pt->pt_namlen &&
1025 			    memcmp(pt->pt_name, pname, cnp->cn_namelen) == 0 &&
1026 			    (pt->pt_valid == NULL ||
1027 			     (*pt->pt_valid)(NULL, dvp->v_mount)))
1028 				break;
1029 		}
1030 
1031 		if (i != nproc_root_targets) {
1032 			error = procfs_allocvp(dvp->v_mount, vpp, 0,
1033 			    pt->pt_pfstype, -1, NULL);
1034 			return (error);
1035 		}
1036 
1037 		if (CNEQ(cnp, "curproc", 7)) {
1038 			pid = curproc->p_pid;
1039 			vnpid = 0;
1040 			type = PFScurproc;
1041 		} else if (CNEQ(cnp, "self", 4)) {
1042 			pid = curproc->p_pid;
1043 			vnpid = 0;
1044 			type = PFSself;
1045 		} else {
1046 			pid = (pid_t)atoi(pname, cnp->cn_namelen);
1047 			vnpid = pid;
1048 			type = PFSproc;
1049 		}
1050 
1051 		if (procfs_proc_lock(pid, &p, ESRCH) != 0)
1052 			break;
1053 		error = procfs_allocvp(dvp->v_mount, vpp, vnpid, type, -1, p);
1054 		procfs_proc_unlock(p);
1055 		return (error);
1056 
1057 	case PFSproc:
1058 		/*
1059 		 * do the .. dance. We unlock the directory, and then
1060 		 * get the root dir. That will automatically return ..
1061 		 * locked. Then if the caller wanted dvp locked, we
1062 		 * re-lock.
1063 		 */
1064 		if (cnp->cn_flags & ISDOTDOT) {
1065 			VOP_UNLOCK(dvp, 0);
1066 			error = procfs_root(dvp->v_mount, vpp);
1067 			vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1068 			return (error);
1069 		}
1070 
1071 		if (procfs_proc_lock(pfs->pfs_pid, &p, ESRCH) != 0)
1072 			break;
1073 
1074 		for (pt = proc_targets, i = 0; i < nproc_targets; pt++, i++) {
1075 			struct lwp *plwp;
1076 			int found;
1077 
1078 			mutex_enter(&p->p_smutex);
1079 			plwp = proc_representative_lwp(p, NULL, 1);
1080 			lwp_addref(plwp);
1081 			mutex_exit(&p->p_smutex);
1082 			found = cnp->cn_namelen == pt->pt_namlen &&
1083 			    memcmp(pt->pt_name, pname, cnp->cn_namelen) == 0 &&
1084 			    (pt->pt_valid == NULL
1085 			      || (*pt->pt_valid)(plwp, dvp->v_mount));
1086 			lwp_delref(plwp);
1087 			if (found)
1088 				break;
1089 		}
1090 		if (i == nproc_targets) {
1091 			procfs_proc_unlock(p);
1092 			break;
1093 		}
1094 		if (pt->pt_pfstype == PFSfile) {
1095 			fvp = p->p_textvp;
1096 			/* We already checked that it exists. */
1097 			VREF(fvp);
1098 			procfs_proc_unlock(p);
1099 			vn_lock(fvp, LK_EXCLUSIVE | LK_RETRY);
1100 			*vpp = fvp;
1101 			return (0);
1102 		}
1103 
1104 		error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
1105 		    pt->pt_pfstype, -1, p);
1106 		procfs_proc_unlock(p);
1107 		return (error);
1108 
1109 	case PFSfd: {
1110 		int fd;
1111 		file_t *fp;
1112 
1113 		if ((error = procfs_proc_lock(pfs->pfs_pid, &p, ENOENT)) != 0)
1114 			return error;
1115 
1116 		/*
1117 		 * do the .. dance. We unlock the directory, and then
1118 		 * get the proc dir. That will automatically return ..
1119 		 * locked. Then re-lock the directory.
1120 		 */
1121 		if (cnp->cn_flags & ISDOTDOT) {
1122 			VOP_UNLOCK(dvp, 0);
1123 			error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
1124 			    PFSproc, -1, p);
1125 			procfs_proc_unlock(p);
1126 			vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1127 			return (error);
1128 		}
1129 		fd = atoi(pname, cnp->cn_namelen);
1130 
1131 		fp = fd_getfile2(p, fd);
1132 		if (fp == NULL) {
1133 			procfs_proc_unlock(p);
1134 			return ENOENT;
1135 		}
1136 		fvp = fp->f_data;
1137 
1138 		/* Don't show directories */
1139 		if (fp->f_type == DTYPE_VNODE && fvp->v_type != VDIR) {
1140 			VREF(fvp);
1141 			closef(fp);
1142 			procfs_proc_unlock(p);
1143 			vn_lock(fvp, LK_EXCLUSIVE | LK_RETRY |
1144 			    (p == curproc ? LK_CANRECURSE : 0));
1145 			*vpp = fvp;
1146 			return 0;
1147 		}
1148 
1149 		closef(fp);
1150 		error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
1151 		    PFSfd, fd, p);
1152 		procfs_proc_unlock(p);
1153 		return error;
1154 	}
1155 	default:
1156 		return (ENOTDIR);
1157 	}
1158 
1159 	return (cnp->cn_nameiop == LOOKUP ? ENOENT : EROFS);
1160 }
1161 
1162 int
1163 procfs_validfile(struct lwp *l, struct mount *mp)
1164 {
1165 	return l != NULL && l->l_proc != NULL && l->l_proc->p_textvp != NULL;
1166 }
1167 
1168 static int
1169 procfs_validfile_linux(l, mp)
1170 	struct lwp *l;
1171 	struct mount *mp;
1172 {
1173 	int flags;
1174 
1175 	flags = VFSTOPROC(mp)->pmnt_flags;
1176 	return (flags & PROCFSMNT_LINUXCOMPAT) &&
1177 	    (l == NULL || l->l_proc == NULL || procfs_validfile(l, mp));
1178 }
1179 
1180 struct procfs_root_readdir_ctx {
1181 	struct uio *uiop;
1182 	off_t *cookies;
1183 	int ncookies;
1184 	off_t off;
1185 	off_t startoff;
1186 	int error;
1187 };
1188 
1189 static int
1190 procfs_root_readdir_callback(struct proc *p, void *arg)
1191 {
1192 	struct procfs_root_readdir_ctx *ctxp = arg;
1193 	struct dirent d;
1194 	struct uio *uiop;
1195 	int error;
1196 
1197 	uiop = ctxp->uiop;
1198 	if (uiop->uio_resid < UIO_MX)
1199 		return -1; /* no space */
1200 
1201 	if (ctxp->off < ctxp->startoff) {
1202 		ctxp->off++;
1203 		return 0;
1204 	}
1205 
1206 	if (kauth_authorize_process(kauth_cred_get(),
1207 	    KAUTH_PROCESS_CANSEE, p,
1208 	    KAUTH_ARG(KAUTH_REQ_PROCESS_CANSEE_ENTRY), NULL, NULL) != 0)
1209 		return 0;
1210 
1211 	memset(&d, 0, UIO_MX);
1212 	d.d_reclen = UIO_MX;
1213 	d.d_fileno = PROCFS_FILENO(p->p_pid, PFSproc, -1);
1214 	d.d_namlen = snprintf(d.d_name,
1215 	    UIO_MX - offsetof(struct dirent, d_name), "%ld", (long)p->p_pid);
1216 	d.d_type = DT_DIR;
1217 
1218 	mutex_exit(&proclist_lock);
1219 	error = uiomove(&d, UIO_MX, uiop);
1220 	mutex_enter(&proclist_lock);
1221 	if (error) {
1222 		ctxp->error = error;
1223 		return -1;
1224 	}
1225 
1226 	ctxp->ncookies++;
1227 	if (ctxp->cookies)
1228 		*(ctxp->cookies)++ = ctxp->off + 1;
1229 	ctxp->off++;
1230 
1231 	return 0;
1232 }
1233 
1234 /*
1235  * readdir returns directory entries from pfsnode (vp).
1236  *
1237  * the strategy here with procfs is to generate a single
1238  * directory entry at a time (struct dirent) and then
1239  * copy that out to userland using uiomove.  a more efficent
1240  * though more complex implementation, would try to minimize
1241  * the number of calls to uiomove().  for procfs, this is
1242  * hardly worth the added code complexity.
1243  *
1244  * this should just be done through read()
1245  */
1246 int
1247 procfs_readdir(v)
1248 	void *v;
1249 {
1250 	struct vop_readdir_args /* {
1251 		struct vnode *a_vp;
1252 		struct uio *a_uio;
1253 		kauth_cred_t a_cred;
1254 		int *a_eofflag;
1255 		off_t **a_cookies;
1256 		int *a_ncookies;
1257 	} */ *ap = v;
1258 	struct uio *uio = ap->a_uio;
1259 	struct dirent d;
1260 	struct pfsnode *pfs;
1261 	off_t i;
1262 	int error;
1263 	off_t *cookies = NULL;
1264 	int ncookies;
1265 	struct vnode *vp;
1266 	const struct proc_target *pt;
1267 	struct procfs_root_readdir_ctx ctx;
1268 	struct lwp *l;
1269 	int nfd;
1270 
1271 	vp = ap->a_vp;
1272 	pfs = VTOPFS(vp);
1273 
1274 	if (uio->uio_resid < UIO_MX)
1275 		return (EINVAL);
1276 	if (uio->uio_offset < 0)
1277 		return (EINVAL);
1278 
1279 	error = 0;
1280 	i = uio->uio_offset;
1281 	memset(&d, 0, UIO_MX);
1282 	d.d_reclen = UIO_MX;
1283 	ncookies = uio->uio_resid / UIO_MX;
1284 
1285 	switch (pfs->pfs_type) {
1286 	/*
1287 	 * this is for the process-specific sub-directories.
1288 	 * all that is needed to is copy out all the entries
1289 	 * from the procent[] table (top of this file).
1290 	 */
1291 	case PFSproc: {
1292 		struct proc *p;
1293 
1294 		if (i >= nproc_targets)
1295 			return 0;
1296 
1297 		if (procfs_proc_lock(pfs->pfs_pid, &p, ESRCH) != 0)
1298 			break;
1299 
1300 		if (ap->a_ncookies) {
1301 			ncookies = min(ncookies, (nproc_targets - i));
1302 			cookies = malloc(ncookies * sizeof (off_t),
1303 			    M_TEMP, M_WAITOK);
1304 			*ap->a_cookies = cookies;
1305 		}
1306 
1307 		for (pt = &proc_targets[i];
1308 		     uio->uio_resid >= UIO_MX && i < nproc_targets; pt++, i++) {
1309 			if (pt->pt_valid) {
1310 				/* XXX LWP can disappear */
1311 				mutex_enter(&p->p_smutex);
1312 				l = proc_representative_lwp(p, NULL, 1);
1313 				mutex_exit(&p->p_smutex);
1314 				if ((*pt->pt_valid)(l, vp->v_mount) == 0)
1315 					continue;
1316 			}
1317 
1318 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid,
1319 			    pt->pt_pfstype, -1);
1320 			d.d_namlen = pt->pt_namlen;
1321 			memcpy(d.d_name, pt->pt_name, pt->pt_namlen + 1);
1322 			d.d_type = pt->pt_type;
1323 
1324 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1325 				break;
1326 			if (cookies)
1327 				*cookies++ = i + 1;
1328 		}
1329 
1330 		procfs_proc_unlock(p);
1331 	    	break;
1332 	}
1333 	case PFSfd: {
1334 		struct proc *p;
1335 		file_t *fp;
1336 		int lim, nc = 0;
1337 
1338 		if ((error = procfs_proc_lock(pfs->pfs_pid, &p, ESRCH)) != 0)
1339 			return error;
1340 
1341 		/* XXX Should this be by file as well? */
1342 		if (kauth_authorize_process(kauth_cred_get(),
1343 		    KAUTH_PROCESS_CANSEE, p,
1344 		    KAUTH_ARG(KAUTH_REQ_PROCESS_CANSEE_OPENFILES), NULL,
1345 		    NULL) != 0) {
1346 		    	procfs_proc_unlock(p);
1347 			return ESRCH;
1348 		}
1349 
1350 		nfd = p->p_fd->fd_nfiles;
1351 
1352 		lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfiles);
1353 		if (i >= lim) {
1354 		    	procfs_proc_unlock(p);
1355 			return 0;
1356 		}
1357 
1358 		if (ap->a_ncookies) {
1359 			ncookies = min(ncookies, (nfd + 2 - i));
1360 			cookies = malloc(ncookies * sizeof (off_t),
1361 			    M_TEMP, M_WAITOK);
1362 			*ap->a_cookies = cookies;
1363 		}
1364 
1365 		for (; i < 2 && uio->uio_resid >= UIO_MX; i++) {
1366 			pt = &proc_targets[i];
1367 			d.d_namlen = pt->pt_namlen;
1368 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid,
1369 			    pt->pt_pfstype, -1);
1370 			(void)memcpy(d.d_name, pt->pt_name, pt->pt_namlen + 1);
1371 			d.d_type = pt->pt_type;
1372 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1373 				break;
1374 			if (cookies)
1375 				*cookies++ = i + 1;
1376 			nc++;
1377 		}
1378 		if (error) {
1379 			ncookies = nc;
1380 			break;
1381 		}
1382 		for (; uio->uio_resid >= UIO_MX && i < nfd; i++) {
1383 			/* check the descriptor exists */
1384 			if ((fp = fd_getfile2(p, i - 2)) == NULL)
1385 				continue;
1386 			closef(fp);
1387 
1388 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid, PFSfd, i - 2);
1389 			d.d_namlen = snprintf(d.d_name, sizeof(d.d_name),
1390 			    "%lld", (long long)(i - 2));
1391 			d.d_type = VREG;
1392 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1393 				break;
1394 			if (cookies)
1395 				*cookies++ = i + 1;
1396 			nc++;
1397 		}
1398 		ncookies = nc;
1399 		procfs_proc_unlock(p);
1400 		break;
1401 	}
1402 
1403 	/*
1404 	 * this is for the root of the procfs filesystem
1405 	 * what is needed are special entries for "curproc"
1406 	 * and "self" followed by an entry for each process
1407 	 * on allproc.
1408 	 */
1409 
1410 	case PFSroot: {
1411 		int nc = 0;
1412 
1413 		if (ap->a_ncookies) {
1414 			/*
1415 			 * XXX Potentially allocating too much space here,
1416 			 * but I'm lazy. This loop needs some work.
1417 			 */
1418 			cookies = malloc(ncookies * sizeof (off_t),
1419 			    M_TEMP, M_WAITOK);
1420 			*ap->a_cookies = cookies;
1421 		}
1422 		error = 0;
1423 		/* 0 ... 3 are static entries. */
1424 		for (; i <= 3 && uio->uio_resid >= UIO_MX; i++) {
1425 			switch (i) {
1426 			case 0:		/* `.' */
1427 			case 1:		/* `..' */
1428 				d.d_fileno = PROCFS_FILENO(0, PFSroot, -1);
1429 				d.d_namlen = i + 1;
1430 				memcpy(d.d_name, "..", d.d_namlen);
1431 				d.d_name[i + 1] = '\0';
1432 				d.d_type = DT_DIR;
1433 				break;
1434 
1435 			case 2:
1436 				d.d_fileno = PROCFS_FILENO(0, PFScurproc, -1);
1437 				d.d_namlen = sizeof("curproc") - 1;
1438 				memcpy(d.d_name, "curproc", sizeof("curproc"));
1439 				d.d_type = DT_LNK;
1440 				break;
1441 
1442 			case 3:
1443 				d.d_fileno = PROCFS_FILENO(0, PFSself, -1);
1444 				d.d_namlen = sizeof("self") - 1;
1445 				memcpy(d.d_name, "self", sizeof("self"));
1446 				d.d_type = DT_LNK;
1447 				break;
1448 			}
1449 
1450 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1451 				break;
1452 			nc++;
1453 			if (cookies)
1454 				*cookies++ = i + 1;
1455 		}
1456 		/* 4 ... are process entries. */
1457 		ctx.uiop = uio;
1458 		ctx.error = 0;
1459 		ctx.off = 4;
1460 		ctx.startoff = i;
1461 		ctx.cookies = cookies;
1462 		ctx.ncookies = nc;
1463 		proclist_foreach_call(&allproc,
1464 		    procfs_root_readdir_callback, &ctx);
1465 		cookies = ctx.cookies;
1466 		nc = ctx.ncookies;
1467 		error = ctx.error;
1468 		if (error)
1469 			break;
1470 
1471 		/* misc entries. */
1472 		if (i < ctx.off)
1473 			i = ctx.off;
1474 		if (i >= ctx.off + nproc_root_targets)
1475 			break;
1476 		for (pt = &proc_root_targets[i - ctx.off];
1477 		    uio->uio_resid >= UIO_MX &&
1478 		    pt < &proc_root_targets[nproc_root_targets];
1479 		    pt++, i++) {
1480 			if (pt->pt_valid &&
1481 			    (*pt->pt_valid)(NULL, vp->v_mount) == 0)
1482 				continue;
1483 			d.d_fileno = PROCFS_FILENO(0, pt->pt_pfstype, -1);
1484 			d.d_namlen = pt->pt_namlen;
1485 			memcpy(d.d_name, pt->pt_name, pt->pt_namlen + 1);
1486 			d.d_type = pt->pt_type;
1487 
1488 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1489 				break;
1490 			nc++;
1491 			if (cookies)
1492 				*cookies++ = i + 1;
1493 		}
1494 
1495 		ncookies = nc;
1496 		break;
1497 	}
1498 
1499 	default:
1500 		error = ENOTDIR;
1501 		break;
1502 	}
1503 
1504 	if (ap->a_ncookies) {
1505 		if (error) {
1506 			if (cookies)
1507 				free(*ap->a_cookies, M_TEMP);
1508 			*ap->a_ncookies = 0;
1509 			*ap->a_cookies = NULL;
1510 		} else
1511 			*ap->a_ncookies = ncookies;
1512 	}
1513 	uio->uio_offset = i;
1514 	return (error);
1515 }
1516 
1517 /*
1518  * readlink reads the link of `curproc' and others
1519  */
1520 int
1521 procfs_readlink(v)
1522 	void *v;
1523 {
1524 	struct vop_readlink_args *ap = v;
1525 	char bf[16];		/* should be enough */
1526 	char *bp = bf;
1527 	char *path = NULL;
1528 	int len = 0;
1529 	int error = 0;
1530 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
1531 	struct proc *pown;
1532 
1533 	if (pfs->pfs_fileno == PROCFS_FILENO(0, PFScurproc, -1))
1534 		len = snprintf(bf, sizeof(bf), "%ld", (long)curproc->p_pid);
1535 	else if (pfs->pfs_fileno == PROCFS_FILENO(0, PFSself, -1))
1536 		len = snprintf(bf, sizeof(bf), "%s", "curproc");
1537 	else if (pfs->pfs_fileno == PROCFS_FILENO(pfs->pfs_pid, PFScwd, -1) ||
1538 	    pfs->pfs_fileno == PROCFS_FILENO(pfs->pfs_pid, PFSchroot, -1) ||
1539 	    pfs->pfs_fileno == PROCFS_FILENO(pfs->pfs_pid, PFSexe, -1)) {
1540 		if ((error = procfs_proc_lock(pfs->pfs_pid, &pown, ESRCH)) != 0)
1541 			return error;
1542 		MALLOC(path, char *, MAXPATHLEN + 4, M_TEMP,
1543 		    M_WAITOK|M_CANFAIL);
1544 		if (path == NULL) {
1545 			procfs_proc_unlock(pown);
1546 			return (ENOMEM);
1547 		}
1548 		bp = path + MAXPATHLEN;
1549 		*--bp = '\0';
1550 		procfs_dir(PROCFS_TYPE(pfs->pfs_fileno), curlwp, pown,
1551 		    &bp, path, MAXPATHLEN);
1552 		procfs_proc_unlock(pown);
1553 		len = strlen(bp);
1554 	} else {
1555 		file_t *fp;
1556 		struct vnode *vxp, *vp;
1557 
1558 		if ((error = procfs_proc_lock(pfs->pfs_pid, &pown, ESRCH)) != 0)
1559 			return error;
1560 
1561 		fp = fd_getfile2(pown, pfs->pfs_fd);
1562 		if (fp == NULL) {
1563 			procfs_proc_unlock(pown);
1564 			return EBADF;
1565 		}
1566 
1567 		switch (fp->f_type) {
1568 		case DTYPE_VNODE:
1569 			vxp = (struct vnode *)fp->f_data;
1570 			if (vxp->v_type != VDIR) {
1571 				error = EINVAL;
1572 				break;
1573 			}
1574 			if ((path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK))
1575 			    == NULL) {
1576 				error = ENOMEM;
1577 				break;
1578 			}
1579 			bp = path + MAXPATHLEN;
1580 			*--bp = '\0';
1581 
1582 			/*
1583 			 * XXX: kludge to avoid locking against ourselves
1584 			 * in getcwd()
1585 			 */
1586 			if (vxp->v_tag == VT_PROCFS) {
1587 				*--bp = '/';
1588 			} else {
1589 				rw_enter(&curproc->p_cwdi->cwdi_lock, RW_READER);
1590 				vp = curproc->p_cwdi->cwdi_rdir;
1591 				if (vp == NULL)
1592 					vp = rootvnode;
1593 				error = getcwd_common(vxp, vp, &bp, path,
1594 				    MAXPATHLEN / 2, 0, curlwp);
1595 				rw_exit(&curproc->p_cwdi->cwdi_lock);
1596 			}
1597 			if (error)
1598 				break;
1599 			len = strlen(bp);
1600 			break;
1601 
1602 		case DTYPE_MISC:
1603 			len = snprintf(bf, sizeof(bf), "%s", "[misc]");
1604 			break;
1605 
1606 		case DTYPE_KQUEUE:
1607 			len = snprintf(bf, sizeof(bf), "%s", "[kqueue]");
1608 			break;
1609 
1610 		default:
1611 			error = EINVAL;
1612 			break;
1613 		}
1614 		closef(fp);
1615 		procfs_proc_unlock(pown);
1616 	}
1617 
1618 	if (error == 0)
1619 		error = uiomove(bp, len, ap->a_uio);
1620 	if (path)
1621 		free(path, M_TEMP);
1622 	return error;
1623 }
1624 
1625 /*
1626  * convert decimal ascii to int
1627  */
1628 static int
1629 atoi(b, len)
1630 	const char *b;
1631 	size_t len;
1632 {
1633 	int p = 0;
1634 
1635 	while (len--) {
1636 		char c = *b++;
1637 		if (c < '0' || c > '9')
1638 			return -1;
1639 		p = 10 * p + (c - '0');
1640 	}
1641 
1642 	return p;
1643 }
1644