xref: /netbsd-src/sys/miscfs/procfs/procfs_vnops.c (revision 7c192b2a5e1093666e67801684f930ef49b3b363)
1 /*	$NetBSD: procfs_vnops.c,v 1.197 2017/05/26 14:21:01 riastradh 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  *
19  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29  * POSSIBILITY OF SUCH DAMAGE.
30  */
31 
32 /*
33  * Copyright (c) 1993, 1995
34  *	The Regents of the University of California.  All rights reserved.
35  *
36  * This code is derived from software contributed to Berkeley by
37  * Jan-Simon Pendry.
38  *
39  * Redistribution and use in source and binary forms, with or without
40  * modification, are permitted provided that the following conditions
41  * are met:
42  * 1. Redistributions of source code must retain the above copyright
43  *    notice, this list of conditions and the following disclaimer.
44  * 2. Redistributions in binary form must reproduce the above copyright
45  *    notice, this list of conditions and the following disclaimer in the
46  *    documentation and/or other materials provided with the distribution.
47  * 3. Neither the name of the University nor the names of its contributors
48  *    may be used to endorse or promote products derived from this software
49  *    without specific prior written permission.
50  *
51  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
52  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
53  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
54  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
55  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
56  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
57  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
58  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
59  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
60  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
61  * SUCH DAMAGE.
62  *
63  *	@(#)procfs_vnops.c	8.18 (Berkeley) 5/21/95
64  */
65 
66 /*
67  * Copyright (c) 1993 Jan-Simon Pendry
68  *
69  * This code is derived from software contributed to Berkeley by
70  * Jan-Simon Pendry.
71  *
72  * Redistribution and use in source and binary forms, with or without
73  * modification, are permitted provided that the following conditions
74  * are met:
75  * 1. Redistributions of source code must retain the above copyright
76  *    notice, this list of conditions and the following disclaimer.
77  * 2. Redistributions in binary form must reproduce the above copyright
78  *    notice, this list of conditions and the following disclaimer in the
79  *    documentation and/or other materials provided with the distribution.
80  * 3. All advertising materials mentioning features or use of this software
81  *    must display the following acknowledgement:
82  *	This product includes software developed by the University of
83  *	California, Berkeley and its contributors.
84  * 4. Neither the name of the University nor the names of its contributors
85  *    may be used to endorse or promote products derived from this software
86  *    without specific prior written permission.
87  *
88  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
89  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
90  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
91  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
92  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
93  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
94  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
95  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
96  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
97  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
98  * SUCH DAMAGE.
99  *
100  *	@(#)procfs_vnops.c	8.18 (Berkeley) 5/21/95
101  */
102 
103 /*
104  * procfs vnode interface
105  */
106 
107 #include <sys/cdefs.h>
108 __KERNEL_RCSID(0, "$NetBSD: procfs_vnops.c,v 1.197 2017/05/26 14:21:01 riastradh Exp $");
109 
110 #include <sys/param.h>
111 #include <sys/systm.h>
112 #include <sys/time.h>
113 #include <sys/kernel.h>
114 #include <sys/file.h>
115 #include <sys/filedesc.h>
116 #include <sys/proc.h>
117 #include <sys/vnode.h>
118 #include <sys/namei.h>
119 #include <sys/malloc.h>
120 #include <sys/mount.h>
121 #include <sys/dirent.h>
122 #include <sys/resourcevar.h>
123 #include <sys/stat.h>
124 #include <sys/ptrace.h>
125 #include <sys/kauth.h>
126 #include <sys/exec.h>
127 
128 #include <uvm/uvm_extern.h>	/* for PAGE_SIZE */
129 
130 #include <machine/reg.h>
131 
132 #include <miscfs/genfs/genfs.h>
133 #include <miscfs/procfs/procfs.h>
134 
135 /*
136  * Vnode Operations.
137  *
138  */
139 
140 static int procfs_validfile_linux(struct lwp *, struct mount *);
141 static int procfs_root_readdir_callback(struct proc *, void *);
142 static void procfs_dir(pfstype, struct lwp *, struct proc *, char **, char *,
143     size_t);
144 
145 /*
146  * This is a list of the valid names in the
147  * process-specific sub-directories.  It is
148  * used in procfs_lookup and procfs_readdir
149  */
150 static const struct proc_target {
151 	u_char	pt_type;
152 	u_char	pt_namlen;
153 	const char	*pt_name;
154 	pfstype	pt_pfstype;
155 	int	(*pt_valid)(struct lwp *, struct mount *);
156 } proc_targets[] = {
157 #define N(s) sizeof(s)-1, s
158 	/*	  name		type		validp */
159 	{ DT_DIR, N("."),	PFSproc,	NULL },
160 	{ DT_DIR, N(".."),	PFSroot,	NULL },
161 	{ DT_DIR, N("fd"),	PFSfd,		NULL },
162 	{ DT_REG, N("file"),	PFSfile,	procfs_validfile },
163 	{ DT_REG, N("auxv"),	PFSauxv,	procfs_validauxv },
164 	{ DT_REG, N("mem"),	PFSmem,		NULL },
165 	{ DT_REG, N("regs"),	PFSregs,	procfs_validregs },
166 	{ DT_REG, N("fpregs"),	PFSfpregs,	procfs_validfpregs },
167 	{ DT_REG, N("ctl"),	PFSctl,		NULL },
168 	{ DT_REG, N("stat"),	PFSstat,	procfs_validfile_linux },
169 	{ DT_REG, N("status"),	PFSstatus,	NULL },
170 	{ DT_REG, N("note"),	PFSnote,	NULL },
171 	{ DT_REG, N("notepg"),	PFSnotepg,	NULL },
172 	{ DT_REG, N("map"),	PFSmap,		procfs_validmap },
173 	{ DT_REG, N("maps"),	PFSmaps,	procfs_validmap },
174 	{ DT_REG, N("cmdline"), PFScmdline,	NULL },
175 	{ DT_REG, N("exe"),	PFSexe,		procfs_validfile },
176 	{ DT_LNK, N("cwd"),	PFScwd,		NULL },
177 	{ DT_LNK, N("root"),	PFSchroot,	NULL },
178 	{ DT_LNK, N("emul"),	PFSemul,	NULL },
179 	{ DT_REG, N("statm"),	PFSstatm,	procfs_validfile_linux },
180 	{ DT_DIR, N("task"),	PFStask,	procfs_validfile_linux },
181 #ifdef __HAVE_PROCFS_MACHDEP
182 	PROCFS_MACHDEP_NODETYPE_DEFNS
183 #endif
184 #undef N
185 };
186 static const int nproc_targets = sizeof(proc_targets) / sizeof(proc_targets[0]);
187 
188 /*
189  * List of files in the root directory. Note: the validate function will
190  * be called with p == NULL for these ones.
191  */
192 static const struct proc_target proc_root_targets[] = {
193 #define N(s) sizeof(s)-1, s
194 	/*	  name		    type	    validp */
195 	{ DT_REG, N("meminfo"),     PFSmeminfo,        procfs_validfile_linux },
196 	{ DT_REG, N("cpuinfo"),     PFScpuinfo,        procfs_validfile_linux },
197 	{ DT_REG, N("uptime"),      PFSuptime,         procfs_validfile_linux },
198 	{ DT_REG, N("mounts"),	    PFSmounts,	       procfs_validfile_linux },
199 	{ DT_REG, N("devices"),     PFSdevices,        procfs_validfile_linux },
200 	{ DT_REG, N("stat"),	    PFScpustat,        procfs_validfile_linux },
201 	{ DT_REG, N("loadavg"),	    PFSloadavg,        procfs_validfile_linux },
202 	{ DT_REG, N("version"),     PFSversion,        procfs_validfile_linux },
203 #undef N
204 };
205 static const int nproc_root_targets =
206     sizeof(proc_root_targets) / sizeof(proc_root_targets[0]);
207 
208 int	procfs_lookup(void *);
209 #define	procfs_create	genfs_eopnotsupp
210 #define	procfs_mknod	genfs_eopnotsupp
211 int	procfs_open(void *);
212 int	procfs_close(void *);
213 int	procfs_access(void *);
214 int	procfs_getattr(void *);
215 int	procfs_setattr(void *);
216 #define	procfs_read	procfs_rw
217 #define	procfs_write	procfs_rw
218 #define	procfs_fcntl	genfs_fcntl
219 #define	procfs_ioctl	genfs_enoioctl
220 #define	procfs_poll	genfs_poll
221 #define procfs_revoke	genfs_revoke
222 #define	procfs_fsync	genfs_nullop
223 #define	procfs_seek	genfs_nullop
224 #define	procfs_remove	genfs_eopnotsupp
225 int	procfs_link(void *);
226 #define	procfs_rename	genfs_eopnotsupp
227 #define	procfs_mkdir	genfs_eopnotsupp
228 #define	procfs_rmdir	genfs_eopnotsupp
229 int	procfs_symlink(void *);
230 int	procfs_readdir(void *);
231 int	procfs_readlink(void *);
232 #define	procfs_abortop	genfs_abortop
233 int	procfs_inactive(void *);
234 int	procfs_reclaim(void *);
235 #define	procfs_lock	genfs_lock
236 #define	procfs_unlock	genfs_unlock
237 #define	procfs_bmap	genfs_badop
238 #define	procfs_strategy	genfs_badop
239 int	procfs_print(void *);
240 int	procfs_pathconf(void *);
241 #define	procfs_islocked	genfs_islocked
242 #define	procfs_advlock	genfs_einval
243 #define	procfs_bwrite	genfs_eopnotsupp
244 #define procfs_putpages	genfs_null_putpages
245 
246 static int atoi(const char *, size_t);
247 
248 /*
249  * procfs vnode operations.
250  */
251 int (**procfs_vnodeop_p)(void *);
252 const struct vnodeopv_entry_desc procfs_vnodeop_entries[] = {
253 	{ &vop_default_desc, vn_default_error },
254 	{ &vop_lookup_desc, procfs_lookup },		/* lookup */
255 	{ &vop_create_desc, procfs_create },		/* create */
256 	{ &vop_mknod_desc, procfs_mknod },		/* mknod */
257 	{ &vop_open_desc, procfs_open },		/* open */
258 	{ &vop_close_desc, procfs_close },		/* close */
259 	{ &vop_access_desc, procfs_access },		/* access */
260 	{ &vop_getattr_desc, procfs_getattr },		/* getattr */
261 	{ &vop_setattr_desc, procfs_setattr },		/* setattr */
262 	{ &vop_read_desc, procfs_read },		/* read */
263 	{ &vop_write_desc, procfs_write },		/* write */
264 	{ &vop_fallocate_desc, genfs_eopnotsupp },	/* fallocate */
265 	{ &vop_fdiscard_desc, genfs_eopnotsupp },	/* fdiscard */
266 	{ &vop_fcntl_desc, procfs_fcntl },		/* fcntl */
267 	{ &vop_ioctl_desc, procfs_ioctl },		/* ioctl */
268 	{ &vop_poll_desc, procfs_poll },		/* poll */
269 	{ &vop_revoke_desc, procfs_revoke },		/* revoke */
270 	{ &vop_fsync_desc, procfs_fsync },		/* fsync */
271 	{ &vop_seek_desc, procfs_seek },		/* seek */
272 	{ &vop_remove_desc, procfs_remove },		/* remove */
273 	{ &vop_link_desc, procfs_link },		/* link */
274 	{ &vop_rename_desc, procfs_rename },		/* rename */
275 	{ &vop_mkdir_desc, procfs_mkdir },		/* mkdir */
276 	{ &vop_rmdir_desc, procfs_rmdir },		/* rmdir */
277 	{ &vop_symlink_desc, procfs_symlink },		/* symlink */
278 	{ &vop_readdir_desc, procfs_readdir },		/* readdir */
279 	{ &vop_readlink_desc, procfs_readlink },	/* readlink */
280 	{ &vop_abortop_desc, procfs_abortop },		/* abortop */
281 	{ &vop_inactive_desc, procfs_inactive },	/* inactive */
282 	{ &vop_reclaim_desc, procfs_reclaim },		/* reclaim */
283 	{ &vop_lock_desc, procfs_lock },		/* lock */
284 	{ &vop_unlock_desc, procfs_unlock },		/* unlock */
285 	{ &vop_bmap_desc, procfs_bmap },		/* bmap */
286 	{ &vop_strategy_desc, procfs_strategy },	/* strategy */
287 	{ &vop_print_desc, procfs_print },		/* print */
288 	{ &vop_islocked_desc, procfs_islocked },	/* islocked */
289 	{ &vop_pathconf_desc, procfs_pathconf },	/* pathconf */
290 	{ &vop_advlock_desc, procfs_advlock },		/* advlock */
291 	{ &vop_putpages_desc, procfs_putpages },	/* putpages */
292 	{ NULL, NULL }
293 };
294 const struct vnodeopv_desc procfs_vnodeop_opv_desc =
295 	{ &procfs_vnodeop_p, procfs_vnodeop_entries };
296 /*
297  * set things up for doing i/o on
298  * the pfsnode (vp).  (vp) is locked
299  * on entry, and should be left locked
300  * on exit.
301  *
302  * for procfs we don't need to do anything
303  * in particular for i/o.  all that is done
304  * is to support exclusive open on process
305  * memory images.
306  */
307 int
308 procfs_open(void *v)
309 {
310 	struct vop_open_args /* {
311 		struct vnode *a_vp;
312 		int  a_mode;
313 		kauth_cred_t a_cred;
314 	} */ *ap = v;
315 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
316 	struct lwp *l1;
317 	struct proc *p2;
318 	int error;
319 
320 	if ((error = procfs_proc_lock(pfs->pfs_pid, &p2, ENOENT)) != 0)
321 		return error;
322 
323 	l1 = curlwp;				/* tracer */
324 
325 #define	M2K(m)	(((m) & FREAD) && ((m) & FWRITE) ? \
326 		 KAUTH_REQ_PROCESS_PROCFS_RW : \
327 		 (m) & FWRITE ? KAUTH_REQ_PROCESS_PROCFS_WRITE : \
328 		 KAUTH_REQ_PROCESS_PROCFS_READ)
329 
330 	mutex_enter(p2->p_lock);
331 	error = kauth_authorize_process(l1->l_cred, KAUTH_PROCESS_PROCFS,
332 	    p2, pfs, KAUTH_ARG(M2K(ap->a_mode)), NULL);
333 	mutex_exit(p2->p_lock);
334 	if (error) {
335 		procfs_proc_unlock(p2);
336 		return (error);
337 	}
338 
339 #undef M2K
340 
341 	switch (pfs->pfs_type) {
342 	case PFSmem:
343 		if (((pfs->pfs_flags & FWRITE) && (ap->a_mode & O_EXCL)) ||
344 		    ((pfs->pfs_flags & O_EXCL) && (ap->a_mode & FWRITE))) {
345 			error = EBUSY;
346 			break;
347 		}
348 
349 		if (!proc_isunder(p2, l1)) {
350 			error = EPERM;
351 			break;
352 		}
353 
354 		if (ap->a_mode & FWRITE)
355 			pfs->pfs_flags = ap->a_mode & (FWRITE|O_EXCL);
356 
357 		break;
358 
359 	case PFSregs:
360 	case PFSfpregs:
361 		if (!proc_isunder(p2, l1)) {
362 			error = EPERM;
363 			break;
364 		}
365 		break;
366 
367 	default:
368 		break;
369 	}
370 
371 	procfs_proc_unlock(p2);
372 	return (error);
373 }
374 
375 /*
376  * close the pfsnode (vp) after doing i/o.
377  * (vp) is not locked on entry or exit.
378  *
379  * nothing to do for procfs other than undo
380  * any exclusive open flag (see _open above).
381  */
382 int
383 procfs_close(void *v)
384 {
385 	struct vop_close_args /* {
386 		struct vnode *a_vp;
387 		int  a_fflag;
388 		kauth_cred_t a_cred;
389 	} */ *ap = v;
390 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
391 
392 	switch (pfs->pfs_type) {
393 	case PFSmem:
394 		if ((ap->a_fflag & FWRITE) && (pfs->pfs_flags & O_EXCL))
395 			pfs->pfs_flags &= ~(FWRITE|O_EXCL);
396 		break;
397 
398 	default:
399 		break;
400 	}
401 
402 	return (0);
403 }
404 
405 /*
406  * _inactive is called when the pfsnode
407  * is vrele'd and the reference count goes
408  * to zero.  (vp) will be on the vnode free
409  * list, so to get it back vget() must be
410  * used.
411  *
412  * (vp) is locked on entry, but must be unlocked on exit.
413  */
414 int
415 procfs_inactive(void *v)
416 {
417 	struct vop_inactive_v2_args /* {
418 		struct vnode *a_vp;
419 		bool *a_recycle;
420 	} */ *ap = v;
421 	struct vnode *vp = ap->a_vp;
422 	struct pfsnode *pfs = VTOPFS(vp);
423 
424 	mutex_enter(proc_lock);
425 	*ap->a_recycle = (proc_find(pfs->pfs_pid) == NULL);
426 	mutex_exit(proc_lock);
427 
428 	return (0);
429 }
430 
431 /*
432  * _reclaim is called when getnewvnode()
433  * wants to make use of an entry on the vnode
434  * free list.  at this time the filesystem needs
435  * to free any private data and remove the node
436  * from any private lists.
437  */
438 int
439 procfs_reclaim(void *v)
440 {
441 	struct vop_reclaim_v2_args /* {
442 		struct vnode *a_vp;
443 	} */ *ap = v;
444 	struct vnode *vp = ap->a_vp;
445 	struct pfsnode *pfs = VTOPFS(vp);
446 
447 	VOP_UNLOCK(vp);
448 
449 	/*
450 	 * To interlock with procfs_revoke_vnodes().
451 	 */
452 	mutex_enter(vp->v_interlock);
453 	vp->v_data = NULL;
454 	mutex_exit(vp->v_interlock);
455 	kmem_free(pfs, sizeof(*pfs));
456 	return 0;
457 }
458 
459 /*
460  * Return POSIX pathconf information applicable to special devices.
461  */
462 int
463 procfs_pathconf(void *v)
464 {
465 	struct vop_pathconf_args /* {
466 		struct vnode *a_vp;
467 		int a_name;
468 		register_t *a_retval;
469 	} */ *ap = v;
470 
471 	switch (ap->a_name) {
472 	case _PC_LINK_MAX:
473 		*ap->a_retval = LINK_MAX;
474 		return (0);
475 	case _PC_MAX_CANON:
476 		*ap->a_retval = MAX_CANON;
477 		return (0);
478 	case _PC_MAX_INPUT:
479 		*ap->a_retval = MAX_INPUT;
480 		return (0);
481 	case _PC_PIPE_BUF:
482 		*ap->a_retval = PIPE_BUF;
483 		return (0);
484 	case _PC_CHOWN_RESTRICTED:
485 		*ap->a_retval = 1;
486 		return (0);
487 	case _PC_VDISABLE:
488 		*ap->a_retval = _POSIX_VDISABLE;
489 		return (0);
490 	case _PC_SYNC_IO:
491 		*ap->a_retval = 1;
492 		return (0);
493 	default:
494 		return (EINVAL);
495 	}
496 	/* NOTREACHED */
497 }
498 
499 /*
500  * _print is used for debugging.
501  * just print a readable description
502  * of (vp).
503  */
504 int
505 procfs_print(void *v)
506 {
507 	struct vop_print_args /* {
508 		struct vnode *a_vp;
509 	} */ *ap = v;
510 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
511 
512 	printf("tag VT_PROCFS, type %d, pid %d, mode %x, flags %lx\n",
513 	    pfs->pfs_type, pfs->pfs_pid, pfs->pfs_mode, pfs->pfs_flags);
514 	return 0;
515 }
516 
517 int
518 procfs_link(void *v)
519 {
520 	struct vop_link_v2_args /* {
521 		struct vnode *a_dvp;
522 		struct vnode *a_vp;
523 		struct componentname *a_cnp;
524 	} */ *ap = v;
525 
526 	VOP_ABORTOP(ap->a_dvp, ap->a_cnp);
527 	return (EROFS);
528 }
529 
530 int
531 procfs_symlink(void *v)
532 {
533 	struct vop_symlink_v3_args /* {
534 		struct vnode *a_dvp;
535 		struct vnode **a_vpp;
536 		struct componentname *a_cnp;
537 		struct vattr *a_vap;
538 		char *a_target;
539 	} */ *ap = v;
540 
541 	VOP_ABORTOP(ap->a_dvp, ap->a_cnp);
542 	return (EROFS);
543 }
544 
545 /*
546  * Works out the path to (and vnode of) the target process's current
547  * working directory or chroot.  If the caller is in a chroot and
548  * can't "reach" the target's cwd or root (or some other error
549  * occurs), a "/" is returned for the path and a NULL pointer is
550  * returned for the vnode.
551  */
552 static void
553 procfs_dir(pfstype t, struct lwp *caller, struct proc *target, char **bpp,
554     char *path, size_t len)
555 {
556 	struct cwdinfo *cwdi;
557 	struct vnode *vp, *rvp;
558 	char *bp;
559 
560 	cwdi = caller->l_proc->p_cwdi;
561 	rw_enter(&cwdi->cwdi_lock, RW_READER);
562 
563 	rvp = cwdi->cwdi_rdir;
564 	bp = bpp ? *bpp : NULL;
565 
566 	switch (t) {
567 	case PFScwd:
568 		vp = target->p_cwdi->cwdi_cdir;
569 		break;
570 	case PFSchroot:
571 		vp = target->p_cwdi->cwdi_rdir;
572 		break;
573 	case PFSexe:
574 		vp = target->p_textvp;
575 		break;
576 	default:
577 		rw_exit(&cwdi->cwdi_lock);
578 		return;
579 	}
580 
581 	/*
582 	 * XXX: this horrible kludge avoids locking panics when
583 	 * attempting to lookup links that point to within procfs
584 	 */
585 	if (vp != NULL && vp->v_tag == VT_PROCFS) {
586 		if (bpp) {
587 			*--bp = '/';
588 			*bpp = bp;
589 		}
590 		rw_exit(&cwdi->cwdi_lock);
591 		return;
592 	}
593 
594 	if (rvp == NULL)
595 		rvp = rootvnode;
596 	if (vp == NULL || getcwd_common(vp, rvp, bp ? &bp : NULL, path,
597 	    len / 2, 0, caller) != 0) {
598 		vp = NULL;
599 		if (bpp) {
600 /*
601 			if (t == PFSexe) {
602 				snprintf(path, len, "%s/%d/file"
603 				    mp->mnt_stat.f_mntonname, pfs->pfs_pid);
604 			} else */ {
605 				bp = *bpp;
606 				*--bp = '/';
607 			}
608 		}
609 	}
610 
611 	if (bpp)
612 		*bpp = bp;
613 
614 	rw_exit(&cwdi->cwdi_lock);
615 }
616 
617 /*
618  * Invent attributes for pfsnode (vp) and store
619  * them in (vap).
620  * Directories lengths are returned as zero since
621  * any real length would require the genuine size
622  * to be computed, and nothing cares anyway.
623  *
624  * this is relatively minimal for procfs.
625  */
626 int
627 procfs_getattr(void *v)
628 {
629 	struct vop_getattr_args /* {
630 		struct vnode *a_vp;
631 		struct vattr *a_vap;
632 		kauth_cred_t a_cred;
633 	} */ *ap = v;
634 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
635 	struct vattr *vap = ap->a_vap;
636 	struct proc *procp;
637 	char *path;
638 	int error;
639 
640 	/* first check the process still exists */
641 	switch (pfs->pfs_type) {
642 	case PFSroot:
643 	case PFScurproc:
644 	case PFSself:
645 		procp = NULL;
646 		break;
647 
648 	default:
649 		error = procfs_proc_lock(pfs->pfs_pid, &procp, ENOENT);
650 		if (error != 0)
651 			return (error);
652 		break;
653 	}
654 
655 	switch (pfs->pfs_type) {
656 	case PFStask:
657 		if (pfs->pfs_fd == -1) {
658 			path = NULL;
659 			break;
660 		}
661 		/*FALLTHROUGH*/
662 	case PFScwd:
663 	case PFSchroot:
664 	case PFSexe:
665 		path = malloc(MAXPATHLEN + 4, M_TEMP, M_WAITOK|M_CANFAIL);
666 		if (path == NULL && procp != NULL) {
667 			procfs_proc_unlock(procp);
668 			return (ENOMEM);
669 		}
670 		break;
671 
672 	default:
673 		path = NULL;
674 		break;
675 	}
676 
677 	if (procp != NULL) {
678 		mutex_enter(procp->p_lock);
679 		error = kauth_authorize_process(kauth_cred_get(),
680 		    KAUTH_PROCESS_CANSEE, procp,
681 		    KAUTH_ARG(KAUTH_REQ_PROCESS_CANSEE_ENTRY), NULL, NULL);
682 		mutex_exit(procp->p_lock);
683 		if (error != 0) {
684 		    	procfs_proc_unlock(procp);
685 		    	if (path != NULL)
686 		    		free(path, M_TEMP);
687 			return (ENOENT);
688 		}
689 	}
690 
691 	error = 0;
692 
693 	/* start by zeroing out the attributes */
694 	vattr_null(vap);
695 
696 	/* next do all the common fields */
697 	vap->va_type = ap->a_vp->v_type;
698 	vap->va_mode = pfs->pfs_mode;
699 	vap->va_fileid = pfs->pfs_fileno;
700 	vap->va_flags = 0;
701 	vap->va_blocksize = PAGE_SIZE;
702 
703 	/*
704 	 * Make all times be current TOD.
705 	 *
706 	 * It would be possible to get the process start
707 	 * time from the p_stats structure, but there's
708 	 * no "file creation" time stamp anyway, and the
709 	 * p_stats structure is not addressable if u. gets
710 	 * swapped out for that process.
711 	 */
712 	getnanotime(&vap->va_ctime);
713 	vap->va_atime = vap->va_mtime = vap->va_ctime;
714 	if (procp)
715 		TIMEVAL_TO_TIMESPEC(&procp->p_stats->p_start,
716 		    &vap->va_birthtime);
717 	else
718 		getnanotime(&vap->va_birthtime);
719 
720 	switch (pfs->pfs_type) {
721 	case PFSmem:
722 	case PFSregs:
723 	case PFSfpregs:
724 #if defined(__HAVE_PROCFS_MACHDEP) && defined(PROCFS_MACHDEP_PROTECT_CASES)
725 	PROCFS_MACHDEP_PROTECT_CASES
726 #endif
727 		/*
728 		 * If the process has exercised some setuid or setgid
729 		 * privilege, then rip away read/write permission so
730 		 * that only root can gain access.
731 		 */
732 		if (procp->p_flag & PK_SUGID)
733 			vap->va_mode &= ~(S_IRUSR|S_IWUSR);
734 		/* FALLTHROUGH */
735 	case PFSctl:
736 	case PFSstatus:
737 	case PFSstat:
738 	case PFSnote:
739 	case PFSnotepg:
740 	case PFScmdline:
741 	case PFSemul:
742 	case PFSstatm:
743 
744 	case PFSmap:
745 	case PFSmaps:
746 	case PFSauxv:
747 		vap->va_nlink = 1;
748 		vap->va_uid = kauth_cred_geteuid(procp->p_cred);
749 		vap->va_gid = kauth_cred_getegid(procp->p_cred);
750 		break;
751 	case PFSmeminfo:
752 	case PFSdevices:
753 	case PFScpuinfo:
754 	case PFSuptime:
755 	case PFSmounts:
756 	case PFScpustat:
757 	case PFSloadavg:
758 	case PFSversion:
759 		vap->va_nlink = 1;
760 		vap->va_uid = vap->va_gid = 0;
761 		break;
762 
763 	case PFSproc:
764 	case PFStask:
765 	case PFSexe:
766 	case PFSfile:
767 	case PFSself:
768 	case PFScurproc:
769 	case PFScwd:
770 	case PFSroot:
771 	case PFSchroot:
772 	case PFSfd:
773 		break;
774 
775 	default:
776 		panic("%s: %d/1", __func__, pfs->pfs_type);
777 	}
778 
779 	/*
780 	 * now do the object specific fields
781 	 *
782 	 * The size could be set from struct reg, but it's hardly
783 	 * worth the trouble, and it puts some (potentially) machine
784 	 * dependent data into this machine-independent code.  If it
785 	 * becomes important then this function should break out into
786 	 * a per-file stat function in the corresponding .c file.
787 	 */
788 
789 	switch (pfs->pfs_type) {
790 	case PFSroot:
791 		/*
792 		 * Set nlink to 1 to tell fts(3) we don't actually know.
793 		 */
794 		vap->va_nlink = 1;
795 		vap->va_uid = 0;
796 		vap->va_gid = 0;
797 		vap->va_bytes = vap->va_size = DEV_BSIZE;
798 		break;
799 
800 	case PFSself:
801 	case PFScurproc: {
802 		char bf[16];		/* should be enough */
803 		vap->va_nlink = 1;
804 		vap->va_uid = 0;
805 		vap->va_gid = 0;
806 		vap->va_bytes = vap->va_size =
807 		    snprintf(bf, sizeof(bf), "%ld", (long)curproc->p_pid);
808 		break;
809 	}
810 	case PFStask:
811 		if (pfs->pfs_fd != -1) {
812 			char bf[4];		/* should be enough */
813 			vap->va_nlink = 1;
814 			vap->va_uid = 0;
815 			vap->va_gid = 0;
816 			vap->va_bytes = vap->va_size =
817 			    snprintf(bf, sizeof(bf), "..");
818 			break;
819 		}
820 		/*FALLTHROUGH*/
821 	case PFSfd:
822 		if (pfs->pfs_fd != -1) {
823 			file_t *fp;
824 
825 			fp = fd_getfile2(procp, pfs->pfs_fd);
826 			if (fp == NULL) {
827 				error = EBADF;
828 				break;
829 			}
830 			vap->va_nlink = 1;
831 			vap->va_uid = kauth_cred_geteuid(fp->f_cred);
832 			vap->va_gid = kauth_cred_getegid(fp->f_cred);
833 			switch (fp->f_type) {
834 			case DTYPE_VNODE:
835 				vap->va_bytes = vap->va_size =
836 				    fp->f_vnode->v_size;
837 				break;
838 			default:
839 				vap->va_bytes = vap->va_size = 0;
840 				break;
841 			}
842 			closef(fp);
843 			break;
844 		}
845 		/*FALLTHROUGH*/
846 	case PFSproc:
847 		vap->va_nlink = 2;
848 		vap->va_uid = kauth_cred_geteuid(procp->p_cred);
849 		vap->va_gid = kauth_cred_getegid(procp->p_cred);
850 		vap->va_bytes = vap->va_size = DEV_BSIZE;
851 		break;
852 
853 	case PFSfile:
854 		error = EOPNOTSUPP;
855 		break;
856 
857 	case PFSmem:
858 		vap->va_bytes = vap->va_size =
859 			ctob(procp->p_vmspace->vm_tsize +
860 				    procp->p_vmspace->vm_dsize +
861 				    procp->p_vmspace->vm_ssize);
862 		break;
863 
864 	case PFSauxv:
865 		vap->va_bytes = vap->va_size = procp->p_execsw->es_arglen;
866 		break;
867 
868 
869 #if defined(PT_GETREGS) || defined(PT_SETREGS)
870 	case PFSregs:
871 		vap->va_bytes = vap->va_size = sizeof(struct reg);
872 		break;
873 #endif
874 
875 #if defined(PT_GETFPREGS) || defined(PT_SETFPREGS)
876 	case PFSfpregs:
877 		vap->va_bytes = vap->va_size = sizeof(struct fpreg);
878 		break;
879 #endif
880 
881 	case PFSctl:
882 	case PFSstatus:
883 	case PFSstat:
884 	case PFSnote:
885 	case PFSnotepg:
886 	case PFScmdline:
887 	case PFSmeminfo:
888 	case PFSdevices:
889 	case PFScpuinfo:
890 	case PFSuptime:
891 	case PFSmounts:
892 	case PFScpustat:
893 	case PFSloadavg:
894 	case PFSstatm:
895 	case PFSversion:
896 		vap->va_bytes = vap->va_size = 0;
897 		break;
898 	case PFSmap:
899 	case PFSmaps:
900 		/*
901 		 * Advise a larger blocksize for the map files, so that
902 		 * they may be read in one pass.
903 		 */
904 		vap->va_blocksize = 4 * PAGE_SIZE;
905 		vap->va_bytes = vap->va_size = 0;
906 		break;
907 
908 	case PFScwd:
909 	case PFSchroot:
910 	case PFSexe: {
911 		char *bp;
912 
913 		vap->va_nlink = 1;
914 		vap->va_uid = 0;
915 		vap->va_gid = 0;
916 		bp = path + MAXPATHLEN;
917 		*--bp = '\0';
918 		procfs_dir(pfs->pfs_type, curlwp, procp, &bp, path,
919 		     MAXPATHLEN);
920 		vap->va_bytes = vap->va_size = strlen(bp);
921 		break;
922 	}
923 
924 	case PFSemul:
925 		vap->va_bytes = vap->va_size = strlen(procp->p_emul->e_name);
926 		break;
927 
928 #ifdef __HAVE_PROCFS_MACHDEP
929 	PROCFS_MACHDEP_NODETYPE_CASES
930 		error = procfs_machdep_getattr(ap->a_vp, vap, procp);
931 		break;
932 #endif
933 
934 	default:
935 		panic("%s: %d/2", __func__, pfs->pfs_type);
936 	}
937 
938 	if (procp != NULL)
939 		procfs_proc_unlock(procp);
940 	if (path != NULL)
941 		free(path, M_TEMP);
942 
943 	return (error);
944 }
945 
946 /*ARGSUSED*/
947 int
948 procfs_setattr(void *v)
949 {
950 	/*
951 	 * just fake out attribute setting
952 	 * it's not good to generate an error
953 	 * return, otherwise things like creat()
954 	 * will fail when they try to set the
955 	 * file length to 0.  worse, this means
956 	 * that echo $note > /proc/$pid/note will fail.
957 	 */
958 
959 	return (0);
960 }
961 
962 /*
963  * implement access checking.
964  *
965  * actually, the check for super-user is slightly
966  * broken since it will allow read access to write-only
967  * objects.  this doesn't cause any particular trouble
968  * but does mean that the i/o entry points need to check
969  * that the operation really does make sense.
970  */
971 int
972 procfs_access(void *v)
973 {
974 	struct vop_access_args /* {
975 		struct vnode *a_vp;
976 		int a_mode;
977 		kauth_cred_t a_cred;
978 	} */ *ap = v;
979 	struct vattr va;
980 	int error;
981 
982 	if ((error = VOP_GETATTR(ap->a_vp, &va, ap->a_cred)) != 0)
983 		return (error);
984 
985 	return kauth_authorize_vnode(ap->a_cred,
986 	    KAUTH_ACCESS_ACTION(ap->a_mode, ap->a_vp->v_type, va.va_mode),
987 	    ap->a_vp, NULL, genfs_can_access(va.va_type, va.va_mode,
988 	    va.va_uid, va.va_gid, ap->a_mode, ap->a_cred));
989 }
990 
991 /*
992  * lookup.  this is incredibly complicated in the
993  * general case, however for most pseudo-filesystems
994  * very little needs to be done.
995  *
996  * Locking isn't hard here, just poorly documented.
997  *
998  * If we're looking up ".", just vref the parent & return it.
999  *
1000  * If we're looking up "..", unlock the parent, and lock "..". If everything
1001  * went ok, and we're on the last component and the caller requested the
1002  * parent locked, try to re-lock the parent. We do this to prevent lock
1003  * races.
1004  *
1005  * For anything else, get the needed node. Then unlock the parent if not
1006  * the last component or not LOCKPARENT (i.e. if we wouldn't re-lock the
1007  * parent in the .. case).
1008  *
1009  * We try to exit with the parent locked in error cases.
1010  */
1011 int
1012 procfs_lookup(void *v)
1013 {
1014 	struct vop_lookup_v2_args /* {
1015 		struct vnode * a_dvp;
1016 		struct vnode ** a_vpp;
1017 		struct componentname * a_cnp;
1018 	} */ *ap = v;
1019 	struct componentname *cnp = ap->a_cnp;
1020 	struct vnode **vpp = ap->a_vpp;
1021 	struct vnode *dvp = ap->a_dvp;
1022 	const char *pname = cnp->cn_nameptr;
1023 	const struct proc_target *pt = NULL;
1024 	struct vnode *fvp;
1025 	pid_t pid, vnpid;
1026 	struct pfsnode *pfs;
1027 	struct proc *p = NULL;
1028 	struct lwp *plwp;
1029 	int i, error;
1030 	pfstype type;
1031 
1032 	*vpp = NULL;
1033 
1034 	if (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)
1035 		return (EROFS);
1036 
1037 	if (cnp->cn_namelen == 1 && *pname == '.') {
1038 		*vpp = dvp;
1039 		vref(dvp);
1040 		return (0);
1041 	}
1042 
1043 	pfs = VTOPFS(dvp);
1044 	switch (pfs->pfs_type) {
1045 	case PFSroot:
1046 		/*
1047 		 * Shouldn't get here with .. in the root node.
1048 		 */
1049 		if (cnp->cn_flags & ISDOTDOT)
1050 			return (EIO);
1051 
1052 		for (i = 0; i < nproc_root_targets; i++) {
1053 			pt = &proc_root_targets[i];
1054 			/*
1055 			 * check for node match.  proc is always NULL here,
1056 			 * so call pt_valid with constant NULL lwp.
1057 			 */
1058 			if (cnp->cn_namelen == pt->pt_namlen &&
1059 			    memcmp(pt->pt_name, pname, cnp->cn_namelen) == 0 &&
1060 			    (pt->pt_valid == NULL ||
1061 			     (*pt->pt_valid)(NULL, dvp->v_mount)))
1062 				break;
1063 		}
1064 
1065 		if (i != nproc_root_targets) {
1066 			error = procfs_allocvp(dvp->v_mount, vpp, 0,
1067 			    pt->pt_pfstype, -1);
1068 			return (error);
1069 		}
1070 
1071 		if (CNEQ(cnp, "curproc", 7)) {
1072 			pid = curproc->p_pid;
1073 			vnpid = 0;
1074 			type = PFScurproc;
1075 		} else if (CNEQ(cnp, "self", 4)) {
1076 			pid = curproc->p_pid;
1077 			vnpid = 0;
1078 			type = PFSself;
1079 		} else {
1080 			pid = (pid_t)atoi(pname, cnp->cn_namelen);
1081 			vnpid = pid;
1082 			type = PFSproc;
1083 		}
1084 
1085 		if (procfs_proc_lock(pid, &p, ESRCH) != 0)
1086 			break;
1087 		error = procfs_allocvp(dvp->v_mount, vpp, vnpid, type, -1);
1088 		procfs_proc_unlock(p);
1089 		return (error);
1090 
1091 	case PFSproc:
1092 		if (cnp->cn_flags & ISDOTDOT) {
1093 			error = procfs_allocvp(dvp->v_mount, vpp, 0, PFSroot,
1094 			    -1);
1095 			return (error);
1096 		}
1097 
1098 		if (procfs_proc_lock(pfs->pfs_pid, &p, ESRCH) != 0)
1099 			break;
1100 
1101 		mutex_enter(p->p_lock);
1102 		LIST_FOREACH(plwp, &p->p_lwps, l_sibling) {
1103 			if (plwp->l_stat != LSZOMB)
1104 				break;
1105 		}
1106 		/* Process is exiting if no-LWPS or all LWPs are LSZOMB */
1107 		if (plwp == NULL) {
1108 			mutex_exit(p->p_lock);
1109 			procfs_proc_unlock(p);
1110 			return ESRCH;
1111 		}
1112 
1113 		lwp_addref(plwp);
1114 		mutex_exit(p->p_lock);
1115 
1116 		for (pt = proc_targets, i = 0; i < nproc_targets; pt++, i++) {
1117 			int found;
1118 
1119 			found = cnp->cn_namelen == pt->pt_namlen &&
1120 			    memcmp(pt->pt_name, pname, cnp->cn_namelen) == 0 &&
1121 			    (pt->pt_valid == NULL
1122 			      || (*pt->pt_valid)(plwp, dvp->v_mount));
1123 			if (found)
1124 				break;
1125 		}
1126 		lwp_delref(plwp);
1127 
1128 		if (i == nproc_targets) {
1129 			procfs_proc_unlock(p);
1130 			break;
1131 		}
1132 		if (pt->pt_pfstype == PFSfile) {
1133 			fvp = p->p_textvp;
1134 			/* We already checked that it exists. */
1135 			vref(fvp);
1136 			procfs_proc_unlock(p);
1137 			*vpp = fvp;
1138 			return (0);
1139 		}
1140 
1141 		error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
1142 		    pt->pt_pfstype, -1);
1143 		procfs_proc_unlock(p);
1144 		return (error);
1145 
1146 	case PFSfd: {
1147 		int fd;
1148 		file_t *fp;
1149 
1150 		if ((error = procfs_proc_lock(pfs->pfs_pid, &p, ENOENT)) != 0)
1151 			return error;
1152 
1153 		if (cnp->cn_flags & ISDOTDOT) {
1154 			error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
1155 			    PFSproc, -1);
1156 			procfs_proc_unlock(p);
1157 			return (error);
1158 		}
1159 		fd = atoi(pname, cnp->cn_namelen);
1160 
1161 		fp = fd_getfile2(p, fd);
1162 		if (fp == NULL) {
1163 			procfs_proc_unlock(p);
1164 			return ENOENT;
1165 		}
1166 		fvp = fp->f_vnode;
1167 
1168 		/* Don't show directories */
1169 		if (fp->f_type == DTYPE_VNODE && fvp->v_type != VDIR) {
1170 			vref(fvp);
1171 			closef(fp);
1172 			procfs_proc_unlock(p);
1173 			*vpp = fvp;
1174 			return 0;
1175 		}
1176 
1177 		closef(fp);
1178 		error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
1179 		    PFSfd, fd);
1180 		procfs_proc_unlock(p);
1181 		return error;
1182 	}
1183 	case PFStask: {
1184 		int xpid;
1185 
1186 		if ((error = procfs_proc_lock(pfs->pfs_pid, &p, ENOENT)) != 0)
1187 			return error;
1188 
1189 		if (cnp->cn_flags & ISDOTDOT) {
1190 			error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
1191 			    PFSproc, -1);
1192 			procfs_proc_unlock(p);
1193 			return (error);
1194 		}
1195 		xpid = atoi(pname, cnp->cn_namelen);
1196 
1197 		if (xpid != pfs->pfs_pid) {
1198 			procfs_proc_unlock(p);
1199 			return ENOENT;
1200 		}
1201 		error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
1202 		    PFStask, 0);
1203 		procfs_proc_unlock(p);
1204 		return error;
1205 	}
1206 	default:
1207 		return (ENOTDIR);
1208 	}
1209 
1210 	return (cnp->cn_nameiop == LOOKUP ? ENOENT : EROFS);
1211 }
1212 
1213 int
1214 procfs_validfile(struct lwp *l, struct mount *mp)
1215 {
1216 	return l != NULL && l->l_proc != NULL && l->l_proc->p_textvp != NULL;
1217 }
1218 
1219 static int
1220 procfs_validfile_linux(struct lwp *l, struct mount *mp)
1221 {
1222 	int flags;
1223 
1224 	flags = VFSTOPROC(mp)->pmnt_flags;
1225 	return (flags & PROCFSMNT_LINUXCOMPAT) &&
1226 	    (l == NULL || l->l_proc == NULL || procfs_validfile(l, mp));
1227 }
1228 
1229 struct procfs_root_readdir_ctx {
1230 	struct uio *uiop;
1231 	off_t *cookies;
1232 	int ncookies;
1233 	off_t off;
1234 	off_t startoff;
1235 	int error;
1236 };
1237 
1238 static int
1239 procfs_root_readdir_callback(struct proc *p, void *arg)
1240 {
1241 	struct procfs_root_readdir_ctx *ctxp = arg;
1242 	struct dirent d;
1243 	struct uio *uiop;
1244 	int error;
1245 
1246 	uiop = ctxp->uiop;
1247 	if (uiop->uio_resid < UIO_MX)
1248 		return -1; /* no space */
1249 
1250 	if (ctxp->off < ctxp->startoff) {
1251 		ctxp->off++;
1252 		return 0;
1253 	}
1254 
1255 	if (kauth_authorize_process(kauth_cred_get(),
1256 	    KAUTH_PROCESS_CANSEE, p,
1257 	    KAUTH_ARG(KAUTH_REQ_PROCESS_CANSEE_ENTRY), NULL, NULL) != 0)
1258 		return 0;
1259 
1260 	memset(&d, 0, UIO_MX);
1261 	d.d_reclen = UIO_MX;
1262 	d.d_fileno = PROCFS_FILENO(p->p_pid, PFSproc, -1);
1263 	d.d_namlen = snprintf(d.d_name,
1264 	    UIO_MX - offsetof(struct dirent, d_name), "%ld", (long)p->p_pid);
1265 	d.d_type = DT_DIR;
1266 
1267 	mutex_exit(proc_lock);
1268 	error = uiomove(&d, UIO_MX, uiop);
1269 	mutex_enter(proc_lock);
1270 	if (error) {
1271 		ctxp->error = error;
1272 		return -1;
1273 	}
1274 
1275 	ctxp->ncookies++;
1276 	if (ctxp->cookies)
1277 		*(ctxp->cookies)++ = ctxp->off + 1;
1278 	ctxp->off++;
1279 
1280 	return 0;
1281 }
1282 
1283 /*
1284  * readdir returns directory entries from pfsnode (vp).
1285  *
1286  * the strategy here with procfs is to generate a single
1287  * directory entry at a time (struct dirent) and then
1288  * copy that out to userland using uiomove.  a more efficent
1289  * though more complex implementation, would try to minimize
1290  * the number of calls to uiomove().  for procfs, this is
1291  * hardly worth the added code complexity.
1292  *
1293  * this should just be done through read()
1294  */
1295 int
1296 procfs_readdir(void *v)
1297 {
1298 	struct vop_readdir_args /* {
1299 		struct vnode *a_vp;
1300 		struct uio *a_uio;
1301 		kauth_cred_t a_cred;
1302 		int *a_eofflag;
1303 		off_t **a_cookies;
1304 		int *a_ncookies;
1305 	} */ *ap = v;
1306 	struct uio *uio = ap->a_uio;
1307 	struct dirent d;
1308 	struct pfsnode *pfs;
1309 	off_t i;
1310 	int error;
1311 	off_t *cookies = NULL;
1312 	int ncookies;
1313 	struct vnode *vp;
1314 	const struct proc_target *pt;
1315 	struct procfs_root_readdir_ctx ctx;
1316 	struct lwp *l;
1317 	int nfd;
1318 
1319 	vp = ap->a_vp;
1320 	pfs = VTOPFS(vp);
1321 
1322 	if (uio->uio_resid < UIO_MX)
1323 		return (EINVAL);
1324 	if (uio->uio_offset < 0)
1325 		return (EINVAL);
1326 
1327 	error = 0;
1328 	i = uio->uio_offset;
1329 	memset(&d, 0, UIO_MX);
1330 	d.d_reclen = UIO_MX;
1331 	ncookies = uio->uio_resid / UIO_MX;
1332 
1333 	switch (pfs->pfs_type) {
1334 	/*
1335 	 * this is for the process-specific sub-directories.
1336 	 * all that is needed to is copy out all the entries
1337 	 * from the procent[] table (top of this file).
1338 	 */
1339 	case PFSproc: {
1340 		struct proc *p;
1341 
1342 		if (i >= nproc_targets)
1343 			return 0;
1344 
1345 		if (procfs_proc_lock(pfs->pfs_pid, &p, ESRCH) != 0)
1346 			break;
1347 
1348 		if (ap->a_ncookies) {
1349 			ncookies = min(ncookies, (nproc_targets - i));
1350 			cookies = malloc(ncookies * sizeof (off_t),
1351 			    M_TEMP, M_WAITOK);
1352 			*ap->a_cookies = cookies;
1353 		}
1354 
1355 		for (pt = &proc_targets[i];
1356 		     uio->uio_resid >= UIO_MX && i < nproc_targets; pt++, i++) {
1357 			if (pt->pt_valid) {
1358 				/* XXXSMP LWP can disappear */
1359 				mutex_enter(p->p_lock);
1360 				l = LIST_FIRST(&p->p_lwps);
1361 				KASSERT(l != NULL);
1362 				mutex_exit(p->p_lock);
1363 				if ((*pt->pt_valid)(l, vp->v_mount) == 0)
1364 					continue;
1365 			}
1366 
1367 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid,
1368 			    pt->pt_pfstype, -1);
1369 			d.d_namlen = pt->pt_namlen;
1370 			memcpy(d.d_name, pt->pt_name, pt->pt_namlen + 1);
1371 			d.d_type = pt->pt_type;
1372 
1373 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1374 				break;
1375 			if (cookies)
1376 				*cookies++ = i + 1;
1377 		}
1378 
1379 		procfs_proc_unlock(p);
1380 	    	break;
1381 	}
1382 	case PFSfd: {
1383 		struct proc *p;
1384 		file_t *fp;
1385 		int lim, nc = 0;
1386 
1387 		if ((error = procfs_proc_lock(pfs->pfs_pid, &p, ESRCH)) != 0)
1388 			return error;
1389 
1390 		/* XXX Should this be by file as well? */
1391 		if (kauth_authorize_process(kauth_cred_get(),
1392 		    KAUTH_PROCESS_CANSEE, p,
1393 		    KAUTH_ARG(KAUTH_REQ_PROCESS_CANSEE_OPENFILES), NULL,
1394 		    NULL) != 0) {
1395 		    	procfs_proc_unlock(p);
1396 			return ESRCH;
1397 		}
1398 
1399 		nfd = p->p_fd->fd_dt->dt_nfiles;
1400 
1401 		lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfiles);
1402 		if (i >= lim) {
1403 		    	procfs_proc_unlock(p);
1404 			return 0;
1405 		}
1406 
1407 		if (ap->a_ncookies) {
1408 			ncookies = min(ncookies, (nfd + 2 - i));
1409 			cookies = malloc(ncookies * sizeof (off_t),
1410 			    M_TEMP, M_WAITOK);
1411 			*ap->a_cookies = cookies;
1412 		}
1413 
1414 		for (; i < 2 && uio->uio_resid >= UIO_MX; i++) {
1415 			pt = &proc_targets[i];
1416 			d.d_namlen = pt->pt_namlen;
1417 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid,
1418 			    pt->pt_pfstype, -1);
1419 			(void)memcpy(d.d_name, pt->pt_name, pt->pt_namlen + 1);
1420 			d.d_type = pt->pt_type;
1421 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1422 				break;
1423 			if (cookies)
1424 				*cookies++ = i + 1;
1425 			nc++;
1426 		}
1427 		if (error) {
1428 			ncookies = nc;
1429 			break;
1430 		}
1431 		for (; uio->uio_resid >= UIO_MX && i < nfd; i++) {
1432 			/* check the descriptor exists */
1433 			if ((fp = fd_getfile2(p, i - 2)) == NULL)
1434 				continue;
1435 			closef(fp);
1436 
1437 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid, PFSfd, i - 2);
1438 			d.d_namlen = snprintf(d.d_name, sizeof(d.d_name),
1439 			    "%lld", (long long)(i - 2));
1440 			d.d_type = VREG;
1441 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1442 				break;
1443 			if (cookies)
1444 				*cookies++ = i + 1;
1445 			nc++;
1446 		}
1447 		ncookies = nc;
1448 		procfs_proc_unlock(p);
1449 		break;
1450 	}
1451 	case PFStask: {
1452 		struct proc *p;
1453 		int nc = 0;
1454 
1455 		if ((error = procfs_proc_lock(pfs->pfs_pid, &p, ESRCH)) != 0)
1456 			return error;
1457 
1458 		nfd = 3;	/* ., .., pid */
1459 
1460 		if (ap->a_ncookies) {
1461 			ncookies = min(ncookies, (nfd + 2 - i));
1462 			cookies = malloc(ncookies * sizeof (off_t),
1463 			    M_TEMP, M_WAITOK);
1464 			*ap->a_cookies = cookies;
1465 		}
1466 
1467 		for (; i < 2 && uio->uio_resid >= UIO_MX; i++) {
1468 			pt = &proc_targets[i];
1469 			d.d_namlen = pt->pt_namlen;
1470 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid,
1471 			    pt->pt_pfstype, -1);
1472 			(void)memcpy(d.d_name, pt->pt_name, pt->pt_namlen + 1);
1473 			d.d_type = pt->pt_type;
1474 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1475 				break;
1476 			if (cookies)
1477 				*cookies++ = i + 1;
1478 			nc++;
1479 		}
1480 		if (error) {
1481 			ncookies = nc;
1482 			break;
1483 		}
1484 		for (; uio->uio_resid >= UIO_MX && i < nfd; i++) {
1485 			/* check the descriptor exists */
1486 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid, PFStask,
1487 			    i - 2);
1488 			d.d_namlen = snprintf(d.d_name, sizeof(d.d_name),
1489 			    "%ld", (long)pfs->pfs_pid);
1490 			d.d_type = DT_LNK;
1491 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1492 				break;
1493 			if (cookies)
1494 				*cookies++ = i + 1;
1495 			nc++;
1496 		}
1497 		ncookies = nc;
1498 		procfs_proc_unlock(p);
1499 		break;
1500 	}
1501 
1502 	/*
1503 	 * this is for the root of the procfs filesystem
1504 	 * what is needed are special entries for "curproc"
1505 	 * and "self" followed by an entry for each process
1506 	 * on allproc.
1507 	 */
1508 
1509 	case PFSroot: {
1510 		int nc = 0;
1511 
1512 		if (ap->a_ncookies) {
1513 			/*
1514 			 * XXX Potentially allocating too much space here,
1515 			 * but I'm lazy. This loop needs some work.
1516 			 */
1517 			cookies = malloc(ncookies * sizeof (off_t),
1518 			    M_TEMP, M_WAITOK);
1519 			*ap->a_cookies = cookies;
1520 		}
1521 		error = 0;
1522 		/* 0 ... 3 are static entries. */
1523 		for (; i <= 3 && uio->uio_resid >= UIO_MX; i++) {
1524 			switch (i) {
1525 			case 0:		/* `.' */
1526 			case 1:		/* `..' */
1527 				d.d_fileno = PROCFS_FILENO(0, PFSroot, -1);
1528 				d.d_namlen = i + 1;
1529 				memcpy(d.d_name, "..", d.d_namlen);
1530 				d.d_name[i + 1] = '\0';
1531 				d.d_type = DT_DIR;
1532 				break;
1533 
1534 			case 2:
1535 				d.d_fileno = PROCFS_FILENO(0, PFScurproc, -1);
1536 				d.d_namlen = sizeof("curproc") - 1;
1537 				memcpy(d.d_name, "curproc", sizeof("curproc"));
1538 				d.d_type = DT_LNK;
1539 				break;
1540 
1541 			case 3:
1542 				d.d_fileno = PROCFS_FILENO(0, PFSself, -1);
1543 				d.d_namlen = sizeof("self") - 1;
1544 				memcpy(d.d_name, "self", sizeof("self"));
1545 				d.d_type = DT_LNK;
1546 				break;
1547 			}
1548 
1549 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1550 				break;
1551 			nc++;
1552 			if (cookies)
1553 				*cookies++ = i + 1;
1554 		}
1555 		/* 4 ... are process entries. */
1556 		ctx.uiop = uio;
1557 		ctx.error = 0;
1558 		ctx.off = 4;
1559 		ctx.startoff = i;
1560 		ctx.cookies = cookies;
1561 		ctx.ncookies = nc;
1562 		proclist_foreach_call(&allproc,
1563 		    procfs_root_readdir_callback, &ctx);
1564 		cookies = ctx.cookies;
1565 		nc = ctx.ncookies;
1566 		error = ctx.error;
1567 		if (error)
1568 			break;
1569 
1570 		/* misc entries. */
1571 		if (i < ctx.off)
1572 			i = ctx.off;
1573 		if (i >= ctx.off + nproc_root_targets)
1574 			break;
1575 		for (pt = &proc_root_targets[i - ctx.off];
1576 		    uio->uio_resid >= UIO_MX &&
1577 		    pt < &proc_root_targets[nproc_root_targets];
1578 		    pt++, i++) {
1579 			if (pt->pt_valid &&
1580 			    (*pt->pt_valid)(NULL, vp->v_mount) == 0)
1581 				continue;
1582 			d.d_fileno = PROCFS_FILENO(0, pt->pt_pfstype, -1);
1583 			d.d_namlen = pt->pt_namlen;
1584 			memcpy(d.d_name, pt->pt_name, pt->pt_namlen + 1);
1585 			d.d_type = pt->pt_type;
1586 
1587 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1588 				break;
1589 			nc++;
1590 			if (cookies)
1591 				*cookies++ = i + 1;
1592 		}
1593 
1594 		ncookies = nc;
1595 		break;
1596 	}
1597 
1598 	default:
1599 		error = ENOTDIR;
1600 		break;
1601 	}
1602 
1603 	if (ap->a_ncookies) {
1604 		if (error) {
1605 			if (cookies)
1606 				free(*ap->a_cookies, M_TEMP);
1607 			*ap->a_ncookies = 0;
1608 			*ap->a_cookies = NULL;
1609 		} else
1610 			*ap->a_ncookies = ncookies;
1611 	}
1612 	uio->uio_offset = i;
1613 	return (error);
1614 }
1615 
1616 /*
1617  * readlink reads the link of `curproc' and others
1618  */
1619 int
1620 procfs_readlink(void *v)
1621 {
1622 	struct vop_readlink_args *ap = v;
1623 	char bf[16];		/* should be enough */
1624 	char *bp = bf;
1625 	char *path = NULL;
1626 	int len = 0;
1627 	int error = 0;
1628 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
1629 	struct proc *pown;
1630 
1631 	if (pfs->pfs_fileno == PROCFS_FILENO(0, PFScurproc, -1))
1632 		len = snprintf(bf, sizeof(bf), "%ld", (long)curproc->p_pid);
1633 	else if (pfs->pfs_fileno == PROCFS_FILENO(0, PFSself, -1))
1634 		len = snprintf(bf, sizeof(bf), "%s", "curproc");
1635 	else if (pfs->pfs_fileno == PROCFS_FILENO(pfs->pfs_pid, PFStask, 0))
1636 		len = snprintf(bf, sizeof(bf), "..");
1637 	else if (pfs->pfs_fileno == PROCFS_FILENO(pfs->pfs_pid, PFScwd, -1) ||
1638 	    pfs->pfs_fileno == PROCFS_FILENO(pfs->pfs_pid, PFSchroot, -1) ||
1639 	    pfs->pfs_fileno == PROCFS_FILENO(pfs->pfs_pid, PFSexe, -1)) {
1640 		if ((error = procfs_proc_lock(pfs->pfs_pid, &pown, ESRCH)) != 0)
1641 			return error;
1642 		path = malloc(MAXPATHLEN + 4, M_TEMP, M_WAITOK|M_CANFAIL);
1643 		if (path == NULL) {
1644 			procfs_proc_unlock(pown);
1645 			return (ENOMEM);
1646 		}
1647 		bp = path + MAXPATHLEN;
1648 		*--bp = '\0';
1649 		procfs_dir(PROCFS_TYPE(pfs->pfs_fileno), curlwp, pown,
1650 		    &bp, path, MAXPATHLEN);
1651 		procfs_proc_unlock(pown);
1652 		len = strlen(bp);
1653 	} else {
1654 		file_t *fp;
1655 		struct vnode *vxp, *vp;
1656 
1657 		if ((error = procfs_proc_lock(pfs->pfs_pid, &pown, ESRCH)) != 0)
1658 			return error;
1659 
1660 		fp = fd_getfile2(pown, pfs->pfs_fd);
1661 		if (fp == NULL) {
1662 			procfs_proc_unlock(pown);
1663 			return EBADF;
1664 		}
1665 
1666 		switch (fp->f_type) {
1667 		case DTYPE_VNODE:
1668 			vxp = fp->f_vnode;
1669 			if (vxp->v_type != VDIR) {
1670 				error = EINVAL;
1671 				break;
1672 			}
1673 			if ((path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK))
1674 			    == NULL) {
1675 				error = ENOMEM;
1676 				break;
1677 			}
1678 			bp = path + MAXPATHLEN;
1679 			*--bp = '\0';
1680 
1681 			/*
1682 			 * XXX: kludge to avoid locking against ourselves
1683 			 * in getcwd()
1684 			 */
1685 			if (vxp->v_tag == VT_PROCFS) {
1686 				*--bp = '/';
1687 			} else {
1688 				rw_enter(&curproc->p_cwdi->cwdi_lock,
1689 				    RW_READER);
1690 				vp = curproc->p_cwdi->cwdi_rdir;
1691 				if (vp == NULL)
1692 					vp = rootvnode;
1693 				error = getcwd_common(vxp, vp, &bp, path,
1694 				    MAXPATHLEN / 2, 0, curlwp);
1695 				rw_exit(&curproc->p_cwdi->cwdi_lock);
1696 			}
1697 			if (error)
1698 				break;
1699 			len = strlen(bp);
1700 			break;
1701 
1702 		case DTYPE_MISC:
1703 			len = snprintf(bf, sizeof(bf), "%s", "[misc]");
1704 			break;
1705 
1706 		case DTYPE_KQUEUE:
1707 			len = snprintf(bf, sizeof(bf), "%s", "[kqueue]");
1708 			break;
1709 
1710 		case DTYPE_SEM:
1711 			len = snprintf(bf, sizeof(bf), "%s", "[ksem]");
1712 			break;
1713 
1714 		default:
1715 			error = EINVAL;
1716 			break;
1717 		}
1718 		closef(fp);
1719 		procfs_proc_unlock(pown);
1720 	}
1721 
1722 	if (error == 0)
1723 		error = uiomove(bp, len, ap->a_uio);
1724 	if (path)
1725 		free(path, M_TEMP);
1726 	return error;
1727 }
1728 
1729 /*
1730  * convert decimal ascii to int
1731  */
1732 static int
1733 atoi(const char *b, size_t len)
1734 {
1735 	int p = 0;
1736 
1737 	while (len--) {
1738 		char c = *b++;
1739 		if (c < '0' || c > '9')
1740 			return -1;
1741 		p = 10 * p + (c - '0');
1742 	}
1743 
1744 	return p;
1745 }
1746