xref: /netbsd-src/sys/miscfs/procfs/procfs_vnops.c (revision deb6f0161a9109e7de9b519dc8dfb9478668dcdd)
1 /*	$NetBSD: procfs_vnops.c,v 1.205 2018/10/14 17:37:40 jdolecek 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 1.205 2018/10/14 17:37:40 jdolecek 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("stat"),	PFSstat,	procfs_validfile_linux },
168 	{ DT_REG, N("status"),	PFSstatus,	NULL },
169 	{ DT_REG, N("note"),	PFSnote,	NULL },
170 	{ DT_REG, N("notepg"),	PFSnotepg,	NULL },
171 	{ DT_REG, N("map"),	PFSmap,		procfs_validmap },
172 	{ DT_REG, N("maps"),	PFSmaps,	procfs_validmap },
173 	{ DT_REG, N("cmdline"), PFScmdline,	NULL },
174 	{ DT_REG, N("environ"), PFSenviron,	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_kqfilter	genfs_kqfilter
222 #define	procfs_revoke	genfs_revoke
223 #define	procfs_fsync	genfs_nullop
224 #define	procfs_seek	genfs_nullop
225 #define	procfs_remove	genfs_eopnotsupp
226 int	procfs_link(void *);
227 #define	procfs_rename	genfs_eopnotsupp
228 #define	procfs_mkdir	genfs_eopnotsupp
229 #define	procfs_rmdir	genfs_eopnotsupp
230 int	procfs_symlink(void *);
231 int	procfs_readdir(void *);
232 int	procfs_readlink(void *);
233 #define	procfs_abortop	genfs_abortop
234 int	procfs_inactive(void *);
235 int	procfs_reclaim(void *);
236 #define	procfs_lock	genfs_lock
237 #define	procfs_unlock	genfs_unlock
238 #define	procfs_bmap	genfs_badop
239 #define	procfs_strategy	genfs_badop
240 int	procfs_print(void *);
241 int	procfs_pathconf(void *);
242 #define	procfs_islocked	genfs_islocked
243 #define	procfs_advlock	genfs_einval
244 #define	procfs_bwrite	genfs_eopnotsupp
245 #define procfs_putpages	genfs_null_putpages
246 
247 static int atoi(const char *, size_t);
248 
249 /*
250  * procfs vnode operations.
251  */
252 int (**procfs_vnodeop_p)(void *);
253 const struct vnodeopv_entry_desc procfs_vnodeop_entries[] = {
254 	{ &vop_default_desc, vn_default_error },
255 	{ &vop_lookup_desc, procfs_lookup },		/* lookup */
256 	{ &vop_create_desc, procfs_create },		/* create */
257 	{ &vop_mknod_desc, procfs_mknod },		/* mknod */
258 	{ &vop_open_desc, procfs_open },		/* open */
259 	{ &vop_close_desc, procfs_close },		/* close */
260 	{ &vop_access_desc, procfs_access },		/* access */
261 	{ &vop_getattr_desc, procfs_getattr },		/* getattr */
262 	{ &vop_setattr_desc, procfs_setattr },		/* setattr */
263 	{ &vop_read_desc, procfs_read },		/* read */
264 	{ &vop_write_desc, procfs_write },		/* write */
265 	{ &vop_fallocate_desc, genfs_eopnotsupp },	/* fallocate */
266 	{ &vop_fdiscard_desc, genfs_eopnotsupp },	/* fdiscard */
267 	{ &vop_fcntl_desc, procfs_fcntl },		/* fcntl */
268 	{ &vop_ioctl_desc, procfs_ioctl },		/* ioctl */
269 	{ &vop_poll_desc, procfs_poll },		/* poll */
270 	{ &vop_kqfilter_desc, procfs_kqfilter },	/* kqfilter */
271 	{ &vop_revoke_desc, procfs_revoke },		/* revoke */
272 	{ &vop_fsync_desc, procfs_fsync },		/* fsync */
273 	{ &vop_seek_desc, procfs_seek },		/* seek */
274 	{ &vop_remove_desc, procfs_remove },		/* remove */
275 	{ &vop_link_desc, procfs_link },		/* link */
276 	{ &vop_rename_desc, procfs_rename },		/* rename */
277 	{ &vop_mkdir_desc, procfs_mkdir },		/* mkdir */
278 	{ &vop_rmdir_desc, procfs_rmdir },		/* rmdir */
279 	{ &vop_symlink_desc, procfs_symlink },		/* symlink */
280 	{ &vop_readdir_desc, procfs_readdir },		/* readdir */
281 	{ &vop_readlink_desc, procfs_readlink },	/* readlink */
282 	{ &vop_abortop_desc, procfs_abortop },		/* abortop */
283 	{ &vop_inactive_desc, procfs_inactive },	/* inactive */
284 	{ &vop_reclaim_desc, procfs_reclaim },		/* reclaim */
285 	{ &vop_lock_desc, procfs_lock },		/* lock */
286 	{ &vop_unlock_desc, procfs_unlock },		/* unlock */
287 	{ &vop_bmap_desc, procfs_bmap },		/* bmap */
288 	{ &vop_strategy_desc, procfs_strategy },	/* strategy */
289 	{ &vop_print_desc, procfs_print },		/* print */
290 	{ &vop_islocked_desc, procfs_islocked },	/* islocked */
291 	{ &vop_pathconf_desc, procfs_pathconf },	/* pathconf */
292 	{ &vop_advlock_desc, procfs_advlock },		/* advlock */
293 	{ &vop_putpages_desc, procfs_putpages },	/* putpages */
294 	{ NULL, NULL }
295 };
296 const struct vnodeopv_desc procfs_vnodeop_opv_desc =
297 	{ &procfs_vnodeop_p, procfs_vnodeop_entries };
298 /*
299  * set things up for doing i/o on
300  * the pfsnode (vp).  (vp) is locked
301  * on entry, and should be left locked
302  * on exit.
303  *
304  * for procfs we don't need to do anything
305  * in particular for i/o.  all that is done
306  * is to support exclusive open on process
307  * memory images.
308  */
309 int
310 procfs_open(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_lock);
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_lock);
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(void *v)
386 {
387 	struct vop_close_args /* {
388 		struct vnode *a_vp;
389 		int  a_fflag;
390 		kauth_cred_t a_cred;
391 	} */ *ap = v;
392 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
393 
394 	switch (pfs->pfs_type) {
395 	case PFSmem:
396 		if ((ap->a_fflag & FWRITE) && (pfs->pfs_flags & O_EXCL))
397 			pfs->pfs_flags &= ~(FWRITE|O_EXCL);
398 		break;
399 
400 	default:
401 		break;
402 	}
403 
404 	return (0);
405 }
406 
407 /*
408  * _inactive is called when the pfsnode
409  * is vrele'd and the reference count goes
410  * to zero.  (vp) will be on the vnode free
411  * list, so to get it back vget() must be
412  * used.
413  *
414  * (vp) is locked on entry, but must be unlocked on exit.
415  */
416 int
417 procfs_inactive(void *v)
418 {
419 	struct vop_inactive_v2_args /* {
420 		struct vnode *a_vp;
421 		bool *a_recycle;
422 	} */ *ap = v;
423 	struct vnode *vp = ap->a_vp;
424 	struct pfsnode *pfs = VTOPFS(vp);
425 
426 	mutex_enter(proc_lock);
427 	*ap->a_recycle = (proc_find(pfs->pfs_pid) == NULL);
428 	mutex_exit(proc_lock);
429 
430 	return (0);
431 }
432 
433 /*
434  * _reclaim is called when getnewvnode()
435  * wants to make use of an entry on the vnode
436  * free list.  at this time the filesystem needs
437  * to free any private data and remove the node
438  * from any private lists.
439  */
440 int
441 procfs_reclaim(void *v)
442 {
443 	struct vop_reclaim_v2_args /* {
444 		struct vnode *a_vp;
445 	} */ *ap = v;
446 	struct vnode *vp = ap->a_vp;
447 	struct pfsnode *pfs = VTOPFS(vp);
448 
449 	VOP_UNLOCK(vp);
450 
451 	/*
452 	 * To interlock with procfs_revoke_vnodes().
453 	 */
454 	mutex_enter(vp->v_interlock);
455 	vp->v_data = NULL;
456 	mutex_exit(vp->v_interlock);
457 	kmem_free(pfs, sizeof(*pfs));
458 	return 0;
459 }
460 
461 /*
462  * Return POSIX pathconf information applicable to special devices.
463  */
464 int
465 procfs_pathconf(void *v)
466 {
467 	struct vop_pathconf_args /* {
468 		struct vnode *a_vp;
469 		int a_name;
470 		register_t *a_retval;
471 	} */ *ap = v;
472 
473 	switch (ap->a_name) {
474 	case _PC_LINK_MAX:
475 		*ap->a_retval = LINK_MAX;
476 		return (0);
477 	case _PC_MAX_CANON:
478 		*ap->a_retval = MAX_CANON;
479 		return (0);
480 	case _PC_MAX_INPUT:
481 		*ap->a_retval = MAX_INPUT;
482 		return (0);
483 	case _PC_PIPE_BUF:
484 		*ap->a_retval = PIPE_BUF;
485 		return (0);
486 	case _PC_CHOWN_RESTRICTED:
487 		*ap->a_retval = 1;
488 		return (0);
489 	case _PC_VDISABLE:
490 		*ap->a_retval = _POSIX_VDISABLE;
491 		return (0);
492 	case _PC_SYNC_IO:
493 		*ap->a_retval = 1;
494 		return (0);
495 	default:
496 		return (EINVAL);
497 	}
498 	/* NOTREACHED */
499 }
500 
501 /*
502  * _print is used for debugging.
503  * just print a readable description
504  * of (vp).
505  */
506 int
507 procfs_print(void *v)
508 {
509 	struct vop_print_args /* {
510 		struct vnode *a_vp;
511 	} */ *ap = v;
512 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
513 
514 	printf("tag VT_PROCFS, type %d, pid %d, mode %x, flags %lx\n",
515 	    pfs->pfs_type, pfs->pfs_pid, pfs->pfs_mode, pfs->pfs_flags);
516 	return 0;
517 }
518 
519 int
520 procfs_link(void *v)
521 {
522 	struct vop_link_v2_args /* {
523 		struct vnode *a_dvp;
524 		struct vnode *a_vp;
525 		struct componentname *a_cnp;
526 	} */ *ap = v;
527 
528 	VOP_ABORTOP(ap->a_dvp, ap->a_cnp);
529 	return (EROFS);
530 }
531 
532 int
533 procfs_symlink(void *v)
534 {
535 	struct vop_symlink_v3_args /* {
536 		struct vnode *a_dvp;
537 		struct vnode **a_vpp;
538 		struct componentname *a_cnp;
539 		struct vattr *a_vap;
540 		char *a_target;
541 	} */ *ap = v;
542 
543 	VOP_ABORTOP(ap->a_dvp, ap->a_cnp);
544 	return (EROFS);
545 }
546 
547 /*
548  * Works out the path to 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.
552  */
553 static void
554 procfs_dir(pfstype t, struct lwp *caller, struct proc *target, char **bpp,
555     char *path, size_t len)
556 {
557 	struct cwdinfo *cwdi;
558 	struct vnode *vp, *rvp;
559 	char *bp;
560 
561 	/*
562 	 * Lock target cwdi and take a reference to the vnode
563 	 * we are interested in to prevent it from disappearing
564 	 * before getcwd_common() below.
565 	 */
566 	rw_enter(&target->p_cwdi->cwdi_lock, RW_READER);
567 	switch (t) {
568 	case PFScwd:
569 		vp = target->p_cwdi->cwdi_cdir;
570 		break;
571 	case PFSchroot:
572 		vp = target->p_cwdi->cwdi_rdir;
573 		break;
574 	default:
575 		rw_exit(&target->p_cwdi->cwdi_lock);
576 		return;
577 	}
578 	if (vp != NULL)
579 		vref(vp);
580 	rw_exit(&target->p_cwdi->cwdi_lock);
581 
582 	cwdi = caller->l_proc->p_cwdi;
583 	rw_enter(&cwdi->cwdi_lock, RW_READER);
584 
585 	rvp = cwdi->cwdi_rdir;
586 	bp = bpp ? *bpp : NULL;
587 
588 	/*
589 	 * XXX: this horrible kludge avoids locking panics when
590 	 * attempting to lookup links that point to within procfs
591 	 */
592 	if (vp != NULL && vp->v_tag == VT_PROCFS) {
593 		if (bpp) {
594 			*--bp = '/';
595 			*bpp = bp;
596 		}
597 		vrele(vp);
598 		rw_exit(&cwdi->cwdi_lock);
599 		return;
600 	}
601 
602 	if (rvp == NULL)
603 		rvp = rootvnode;
604 	if (vp == NULL || getcwd_common(vp, rvp, bp ? &bp : NULL, path,
605 	    len / 2, 0, caller) != 0) {
606 		if (bpp) {
607 			bp = *bpp;
608 			*--bp = '/';
609 		}
610 	}
611 
612 	if (bpp)
613 		*bpp = bp;
614 
615 	if (vp != NULL)
616 		vrele(vp);
617 	rw_exit(&cwdi->cwdi_lock);
618 }
619 
620 /*
621  * Invent attributes for pfsnode (vp) and store
622  * them in (vap).
623  * Directories lengths are returned as zero since
624  * any real length would require the genuine size
625  * to be computed, and nothing cares anyway.
626  *
627  * this is relatively minimal for procfs.
628  */
629 int
630 procfs_getattr(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, *bp, bf[16];
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 PFStask:
660 		if (pfs->pfs_fd == -1) {
661 			path = NULL;
662 			break;
663 		}
664 		/*FALLTHROUGH*/
665 	case PFScwd:
666 	case PFSchroot:
667 		path = malloc(MAXPATHLEN + 4, M_TEMP, M_WAITOK);
668 		if (path == NULL && procp != NULL) {
669 			procfs_proc_unlock(procp);
670 			return (ENOMEM);
671 		}
672 		break;
673 
674 	default:
675 		path = NULL;
676 		break;
677 	}
678 
679 	if (procp != NULL) {
680 		mutex_enter(procp->p_lock);
681 		error = kauth_authorize_process(kauth_cred_get(),
682 		    KAUTH_PROCESS_CANSEE, procp,
683 		    KAUTH_ARG(KAUTH_REQ_PROCESS_CANSEE_ENTRY), NULL, NULL);
684 		mutex_exit(procp->p_lock);
685 		if (error != 0) {
686 		    	procfs_proc_unlock(procp);
687 		    	if (path != NULL)
688 		    		free(path, M_TEMP);
689 			return (ENOENT);
690 		}
691 	}
692 
693 	error = 0;
694 
695 	/* start by zeroing out the attributes */
696 	vattr_null(vap);
697 
698 	/* next do all the common fields */
699 	vap->va_type = ap->a_vp->v_type;
700 	vap->va_mode = pfs->pfs_mode;
701 	vap->va_fileid = pfs->pfs_fileno;
702 	vap->va_flags = 0;
703 	vap->va_blocksize = PAGE_SIZE;
704 
705 	/*
706 	 * Make all times be current TOD.
707 	 *
708 	 * It would be possible to get the process start
709 	 * time from the p_stats structure, but there's
710 	 * no "file creation" time stamp anyway, and the
711 	 * p_stats structure is not addressable if u. gets
712 	 * swapped out for that process.
713 	 */
714 	getnanotime(&vap->va_ctime);
715 	vap->va_atime = vap->va_mtime = vap->va_ctime;
716 	if (procp)
717 		TIMEVAL_TO_TIMESPEC(&procp->p_stats->p_start,
718 		    &vap->va_birthtime);
719 	else
720 		getnanotime(&vap->va_birthtime);
721 
722 	switch (pfs->pfs_type) {
723 	case PFSmem:
724 	case PFSregs:
725 	case PFSfpregs:
726 #if defined(__HAVE_PROCFS_MACHDEP) && defined(PROCFS_MACHDEP_PROTECT_CASES)
727 	PROCFS_MACHDEP_PROTECT_CASES
728 #endif
729 		/*
730 		 * If the process has exercised some setuid or setgid
731 		 * privilege, then rip away read/write permission so
732 		 * that only root can gain access.
733 		 */
734 		if (procp->p_flag & PK_SUGID)
735 			vap->va_mode &= ~(S_IRUSR|S_IWUSR);
736 		/* FALLTHROUGH */
737 	case PFSstatus:
738 	case PFSstat:
739 	case PFSnote:
740 	case PFSnotepg:
741 	case PFScmdline:
742 	case PFSenviron:
743 	case PFSemul:
744 	case PFSstatm:
745 
746 	case PFSmap:
747 	case PFSmaps:
748 	case PFSauxv:
749 		vap->va_nlink = 1;
750 		vap->va_uid = kauth_cred_geteuid(procp->p_cred);
751 		vap->va_gid = kauth_cred_getegid(procp->p_cred);
752 		break;
753 	case PFScwd:
754 	case PFSchroot:
755 	case PFSmeminfo:
756 	case PFSdevices:
757 	case PFScpuinfo:
758 	case PFSuptime:
759 	case PFSmounts:
760 	case PFScpustat:
761 	case PFSloadavg:
762 	case PFSversion:
763 	case PFSexe:
764 	case PFSself:
765 	case PFScurproc:
766 	case PFSroot:
767 		vap->va_nlink = 1;
768 		vap->va_uid = vap->va_gid = 0;
769 		break;
770 
771 	case PFSproc:
772 	case PFStask:
773 	case PFSfile:
774 	case PFSfd:
775 		break;
776 
777 	default:
778 		panic("%s: %d/1", __func__, pfs->pfs_type);
779 	}
780 
781 	/*
782 	 * now do the object specific fields
783 	 *
784 	 * The size could be set from struct reg, but it's hardly
785 	 * worth the trouble, and it puts some (potentially) machine
786 	 * dependent data into this machine-independent code.  If it
787 	 * becomes important then this function should break out into
788 	 * a per-file stat function in the corresponding .c file.
789 	 */
790 
791 	switch (pfs->pfs_type) {
792 	case PFSroot:
793 		vap->va_bytes = vap->va_size = DEV_BSIZE;
794 		break;
795 
796 	case PFSself:
797 	case PFScurproc:
798 		vap->va_bytes = vap->va_size =
799 		    snprintf(bf, sizeof(bf), "%ld", (long)curproc->p_pid);
800 		break;
801 	case PFStask:
802 		if (pfs->pfs_fd != -1) {
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), "..");
808 			break;
809 		}
810 		/*FALLTHROUGH*/
811 	case PFSfd:
812 		if (pfs->pfs_fd != -1) {
813 			file_t *fp;
814 
815 			fp = fd_getfile2(procp, pfs->pfs_fd);
816 			if (fp == NULL) {
817 				error = EBADF;
818 				break;
819 			}
820 			vap->va_nlink = 1;
821 			vap->va_uid = kauth_cred_geteuid(fp->f_cred);
822 			vap->va_gid = kauth_cred_getegid(fp->f_cred);
823 			switch (fp->f_type) {
824 			case DTYPE_VNODE:
825 				vap->va_bytes = vap->va_size =
826 				    fp->f_vnode->v_size;
827 				break;
828 			default:
829 				vap->va_bytes = vap->va_size = 0;
830 				break;
831 			}
832 			closef(fp);
833 			break;
834 		}
835 		/*FALLTHROUGH*/
836 	case PFSproc:
837 		vap->va_nlink = 2;
838 		vap->va_uid = kauth_cred_geteuid(procp->p_cred);
839 		vap->va_gid = kauth_cred_getegid(procp->p_cred);
840 		vap->va_bytes = vap->va_size = DEV_BSIZE;
841 		break;
842 
843 	case PFSfile:
844 		error = EOPNOTSUPP;
845 		break;
846 
847 	case PFSmem:
848 		vap->va_bytes = vap->va_size =
849 			ctob(procp->p_vmspace->vm_tsize +
850 				    procp->p_vmspace->vm_dsize +
851 				    procp->p_vmspace->vm_ssize);
852 		break;
853 
854 	case PFSauxv:
855 		vap->va_bytes = vap->va_size = procp->p_execsw->es_arglen;
856 		break;
857 
858 #if defined(PT_GETREGS) || defined(PT_SETREGS)
859 	case PFSregs:
860 		vap->va_bytes = vap->va_size = sizeof(struct reg);
861 		break;
862 #endif
863 
864 #if defined(PT_GETFPREGS) || defined(PT_SETFPREGS)
865 	case PFSfpregs:
866 		vap->va_bytes = vap->va_size = sizeof(struct fpreg);
867 		break;
868 #endif
869 
870 	case PFSstatus:
871 	case PFSstat:
872 	case PFSnote:
873 	case PFSnotepg:
874 	case PFScmdline:
875 	case PFSenviron:
876 	case PFSmeminfo:
877 	case PFSdevices:
878 	case PFScpuinfo:
879 	case PFSuptime:
880 	case PFSmounts:
881 	case PFScpustat:
882 	case PFSloadavg:
883 	case PFSstatm:
884 	case PFSversion:
885 		vap->va_bytes = vap->va_size = 0;
886 		break;
887 	case PFSmap:
888 	case PFSmaps:
889 		/*
890 		 * Advise a larger blocksize for the map files, so that
891 		 * they may be read in one pass.
892 		 */
893 		vap->va_blocksize = 4 * PAGE_SIZE;
894 		vap->va_bytes = vap->va_size = 0;
895 		break;
896 
897 	case PFScwd:
898 	case PFSchroot:
899 		bp = path + MAXPATHLEN;
900 		*--bp = '\0';
901 		procfs_dir(pfs->pfs_type, curlwp, procp, &bp, path,
902 		     MAXPATHLEN);
903 		vap->va_bytes = vap->va_size = strlen(bp);
904 		break;
905 
906 	case PFSexe:
907 		vap->va_bytes = vap->va_size = strlen(procp->p_path);
908 		break;
909 
910 	case PFSemul:
911 		vap->va_bytes = vap->va_size = strlen(procp->p_emul->e_name);
912 		break;
913 
914 #ifdef __HAVE_PROCFS_MACHDEP
915 	PROCFS_MACHDEP_NODETYPE_CASES
916 		error = procfs_machdep_getattr(ap->a_vp, vap, procp);
917 		break;
918 #endif
919 
920 	default:
921 		panic("%s: %d/2", __func__, pfs->pfs_type);
922 	}
923 
924 	if (procp != NULL)
925 		procfs_proc_unlock(procp);
926 	if (path != NULL)
927 		free(path, M_TEMP);
928 
929 	return (error);
930 }
931 
932 /*ARGSUSED*/
933 int
934 procfs_setattr(void *v)
935 {
936 	/*
937 	 * just fake out attribute setting
938 	 * it's not good to generate an error
939 	 * return, otherwise things like creat()
940 	 * will fail when they try to set the
941 	 * file length to 0.  worse, this means
942 	 * that echo $note > /proc/$pid/note will fail.
943 	 */
944 
945 	return (0);
946 }
947 
948 /*
949  * implement access checking.
950  *
951  * actually, the check for super-user is slightly
952  * broken since it will allow read access to write-only
953  * objects.  this doesn't cause any particular trouble
954  * but does mean that the i/o entry points need to check
955  * that the operation really does make sense.
956  */
957 int
958 procfs_access(void *v)
959 {
960 	struct vop_access_args /* {
961 		struct vnode *a_vp;
962 		int a_mode;
963 		kauth_cred_t a_cred;
964 	} */ *ap = v;
965 	struct vattr va;
966 	int error;
967 
968 	if ((error = VOP_GETATTR(ap->a_vp, &va, ap->a_cred)) != 0)
969 		return (error);
970 
971 	return kauth_authorize_vnode(ap->a_cred,
972 	    KAUTH_ACCESS_ACTION(ap->a_mode, ap->a_vp->v_type, va.va_mode),
973 	    ap->a_vp, NULL, genfs_can_access(va.va_type, va.va_mode,
974 	    va.va_uid, va.va_gid, ap->a_mode, ap->a_cred));
975 }
976 
977 /*
978  * lookup.  this is incredibly complicated in the
979  * general case, however for most pseudo-filesystems
980  * very little needs to be done.
981  *
982  * Locking isn't hard here, just poorly documented.
983  *
984  * If we're looking up ".", just vref the parent & return it.
985  *
986  * If we're looking up "..", unlock the parent, and lock "..". If everything
987  * went ok, and we're on the last component and the caller requested the
988  * parent locked, try to re-lock the parent. We do this to prevent lock
989  * races.
990  *
991  * For anything else, get the needed node. Then unlock the parent if not
992  * the last component or not LOCKPARENT (i.e. if we wouldn't re-lock the
993  * parent in the .. case).
994  *
995  * We try to exit with the parent locked in error cases.
996  */
997 int
998 procfs_lookup(void *v)
999 {
1000 	struct vop_lookup_v2_args /* {
1001 		struct vnode * a_dvp;
1002 		struct vnode ** a_vpp;
1003 		struct componentname * a_cnp;
1004 	} */ *ap = v;
1005 	struct componentname *cnp = ap->a_cnp;
1006 	struct vnode **vpp = ap->a_vpp;
1007 	struct vnode *dvp = ap->a_dvp;
1008 	const char *pname = cnp->cn_nameptr;
1009 	const struct proc_target *pt = NULL;
1010 	struct vnode *fvp;
1011 	pid_t pid, vnpid;
1012 	struct pfsnode *pfs;
1013 	struct proc *p = NULL;
1014 	struct lwp *plwp;
1015 	int i, error;
1016 	pfstype type;
1017 
1018 	*vpp = NULL;
1019 
1020 	if (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)
1021 		return (EROFS);
1022 
1023 	if (cnp->cn_namelen == 1 && *pname == '.') {
1024 		*vpp = dvp;
1025 		vref(dvp);
1026 		return (0);
1027 	}
1028 
1029 	pfs = VTOPFS(dvp);
1030 	switch (pfs->pfs_type) {
1031 	case PFSroot:
1032 		/*
1033 		 * Shouldn't get here with .. in the root node.
1034 		 */
1035 		if (cnp->cn_flags & ISDOTDOT)
1036 			return (EIO);
1037 
1038 		for (i = 0; i < nproc_root_targets; i++) {
1039 			pt = &proc_root_targets[i];
1040 			/*
1041 			 * check for node match.  proc is always NULL here,
1042 			 * so call pt_valid with constant NULL lwp.
1043 			 */
1044 			if (cnp->cn_namelen == pt->pt_namlen &&
1045 			    memcmp(pt->pt_name, pname, cnp->cn_namelen) == 0 &&
1046 			    (pt->pt_valid == NULL ||
1047 			     (*pt->pt_valid)(NULL, dvp->v_mount)))
1048 				break;
1049 		}
1050 
1051 		if (i != nproc_root_targets) {
1052 			error = procfs_allocvp(dvp->v_mount, vpp, 0,
1053 			    pt->pt_pfstype, -1);
1054 			return (error);
1055 		}
1056 
1057 		if (CNEQ(cnp, "curproc", 7)) {
1058 			pid = curproc->p_pid;
1059 			vnpid = 0;
1060 			type = PFScurproc;
1061 		} else if (CNEQ(cnp, "self", 4)) {
1062 			pid = curproc->p_pid;
1063 			vnpid = 0;
1064 			type = PFSself;
1065 		} else {
1066 			pid = (pid_t)atoi(pname, cnp->cn_namelen);
1067 			vnpid = pid;
1068 			type = PFSproc;
1069 		}
1070 
1071 		if (procfs_proc_lock(pid, &p, ESRCH) != 0)
1072 			break;
1073 		error = procfs_allocvp(dvp->v_mount, vpp, vnpid, type, -1);
1074 		procfs_proc_unlock(p);
1075 		return (error);
1076 
1077 	case PFSproc:
1078 		if (cnp->cn_flags & ISDOTDOT) {
1079 			error = procfs_allocvp(dvp->v_mount, vpp, 0, PFSroot,
1080 			    -1);
1081 			return (error);
1082 		}
1083 
1084 		if (procfs_proc_lock(pfs->pfs_pid, &p, ESRCH) != 0)
1085 			break;
1086 
1087 		mutex_enter(p->p_lock);
1088 		LIST_FOREACH(plwp, &p->p_lwps, l_sibling) {
1089 			if (plwp->l_stat != LSZOMB)
1090 				break;
1091 		}
1092 		/* Process is exiting if no-LWPS or all LWPs are LSZOMB */
1093 		if (plwp == NULL) {
1094 			mutex_exit(p->p_lock);
1095 			procfs_proc_unlock(p);
1096 			return ESRCH;
1097 		}
1098 
1099 		lwp_addref(plwp);
1100 		mutex_exit(p->p_lock);
1101 
1102 		for (pt = proc_targets, i = 0; i < nproc_targets; pt++, i++) {
1103 			int found;
1104 
1105 			found = cnp->cn_namelen == pt->pt_namlen &&
1106 			    memcmp(pt->pt_name, pname, cnp->cn_namelen) == 0 &&
1107 			    (pt->pt_valid == NULL
1108 			      || (*pt->pt_valid)(plwp, dvp->v_mount));
1109 			if (found)
1110 				break;
1111 		}
1112 		lwp_delref(plwp);
1113 
1114 		if (i == nproc_targets) {
1115 			procfs_proc_unlock(p);
1116 			break;
1117 		}
1118 		if (pt->pt_pfstype == PFSfile) {
1119 			fvp = p->p_textvp;
1120 			/* We already checked that it exists. */
1121 			vref(fvp);
1122 			procfs_proc_unlock(p);
1123 			*vpp = fvp;
1124 			return (0);
1125 		}
1126 
1127 		error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
1128 		    pt->pt_pfstype, -1);
1129 		procfs_proc_unlock(p);
1130 		return (error);
1131 
1132 	case PFSfd: {
1133 		int fd;
1134 		file_t *fp;
1135 
1136 		if ((error = procfs_proc_lock(pfs->pfs_pid, &p, ENOENT)) != 0)
1137 			return error;
1138 
1139 		if (cnp->cn_flags & ISDOTDOT) {
1140 			error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
1141 			    PFSproc, -1);
1142 			procfs_proc_unlock(p);
1143 			return (error);
1144 		}
1145 		fd = atoi(pname, cnp->cn_namelen);
1146 
1147 		fp = fd_getfile2(p, fd);
1148 		if (fp == NULL) {
1149 			procfs_proc_unlock(p);
1150 			return ENOENT;
1151 		}
1152 		fvp = fp->f_vnode;
1153 
1154 		/* Don't show directories */
1155 		if (fp->f_type == DTYPE_VNODE && fvp->v_type != VDIR) {
1156 			vref(fvp);
1157 			closef(fp);
1158 			procfs_proc_unlock(p);
1159 			*vpp = fvp;
1160 			return 0;
1161 		}
1162 
1163 		closef(fp);
1164 		error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
1165 		    PFSfd, fd);
1166 		procfs_proc_unlock(p);
1167 		return error;
1168 	}
1169 	case PFStask: {
1170 		int xpid;
1171 
1172 		if ((error = procfs_proc_lock(pfs->pfs_pid, &p, ENOENT)) != 0)
1173 			return error;
1174 
1175 		if (cnp->cn_flags & ISDOTDOT) {
1176 			error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
1177 			    PFSproc, -1);
1178 			procfs_proc_unlock(p);
1179 			return (error);
1180 		}
1181 		xpid = atoi(pname, cnp->cn_namelen);
1182 
1183 		if (xpid != pfs->pfs_pid) {
1184 			procfs_proc_unlock(p);
1185 			return ENOENT;
1186 		}
1187 		error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
1188 		    PFStask, 0);
1189 		procfs_proc_unlock(p);
1190 		return error;
1191 	}
1192 	default:
1193 		return (ENOTDIR);
1194 	}
1195 
1196 	return (cnp->cn_nameiop == LOOKUP ? ENOENT : EROFS);
1197 }
1198 
1199 int
1200 procfs_validfile(struct lwp *l, struct mount *mp)
1201 {
1202 	return l != NULL && l->l_proc != NULL && l->l_proc->p_textvp != NULL;
1203 }
1204 
1205 static int
1206 procfs_validfile_linux(struct lwp *l, struct mount *mp)
1207 {
1208 	int flags;
1209 
1210 	flags = VFSTOPROC(mp)->pmnt_flags;
1211 	return (flags & PROCFSMNT_LINUXCOMPAT) &&
1212 	    (l == NULL || l->l_proc == NULL || procfs_validfile(l, mp));
1213 }
1214 
1215 struct procfs_root_readdir_ctx {
1216 	struct uio *uiop;
1217 	off_t *cookies;
1218 	int ncookies;
1219 	off_t off;
1220 	off_t startoff;
1221 	int error;
1222 };
1223 
1224 static int
1225 procfs_root_readdir_callback(struct proc *p, void *arg)
1226 {
1227 	struct procfs_root_readdir_ctx *ctxp = arg;
1228 	struct dirent d;
1229 	struct uio *uiop;
1230 	int error;
1231 
1232 	uiop = ctxp->uiop;
1233 	if (uiop->uio_resid < UIO_MX)
1234 		return -1; /* no space */
1235 
1236 	if (ctxp->off < ctxp->startoff) {
1237 		ctxp->off++;
1238 		return 0;
1239 	}
1240 
1241 	if (kauth_authorize_process(kauth_cred_get(),
1242 	    KAUTH_PROCESS_CANSEE, p,
1243 	    KAUTH_ARG(KAUTH_REQ_PROCESS_CANSEE_ENTRY), NULL, NULL) != 0)
1244 		return 0;
1245 
1246 	memset(&d, 0, UIO_MX);
1247 	d.d_reclen = UIO_MX;
1248 	d.d_fileno = PROCFS_FILENO(p->p_pid, PFSproc, -1);
1249 	d.d_namlen = snprintf(d.d_name,
1250 	    UIO_MX - offsetof(struct dirent, d_name), "%ld", (long)p->p_pid);
1251 	d.d_type = DT_DIR;
1252 
1253 	mutex_exit(proc_lock);
1254 	error = uiomove(&d, UIO_MX, uiop);
1255 	mutex_enter(proc_lock);
1256 	if (error) {
1257 		ctxp->error = error;
1258 		return -1;
1259 	}
1260 
1261 	ctxp->ncookies++;
1262 	if (ctxp->cookies)
1263 		*(ctxp->cookies)++ = ctxp->off + 1;
1264 	ctxp->off++;
1265 
1266 	return 0;
1267 }
1268 
1269 /*
1270  * readdir returns directory entries from pfsnode (vp).
1271  *
1272  * the strategy here with procfs is to generate a single
1273  * directory entry at a time (struct dirent) and then
1274  * copy that out to userland using uiomove.  a more efficent
1275  * though more complex implementation, would try to minimize
1276  * the number of calls to uiomove().  for procfs, this is
1277  * hardly worth the added code complexity.
1278  *
1279  * this should just be done through read()
1280  */
1281 int
1282 procfs_readdir(void *v)
1283 {
1284 	struct vop_readdir_args /* {
1285 		struct vnode *a_vp;
1286 		struct uio *a_uio;
1287 		kauth_cred_t a_cred;
1288 		int *a_eofflag;
1289 		off_t **a_cookies;
1290 		int *a_ncookies;
1291 	} */ *ap = v;
1292 	struct uio *uio = ap->a_uio;
1293 	struct dirent d;
1294 	struct pfsnode *pfs;
1295 	off_t i;
1296 	int error;
1297 	off_t *cookies = NULL;
1298 	int ncookies;
1299 	struct vnode *vp;
1300 	const struct proc_target *pt;
1301 	struct procfs_root_readdir_ctx ctx;
1302 	struct lwp *l;
1303 	int nfd;
1304 
1305 	vp = ap->a_vp;
1306 	pfs = VTOPFS(vp);
1307 
1308 	if (uio->uio_resid < UIO_MX)
1309 		return (EINVAL);
1310 	if (uio->uio_offset < 0)
1311 		return (EINVAL);
1312 
1313 	error = 0;
1314 	i = uio->uio_offset;
1315 	memset(&d, 0, UIO_MX);
1316 	d.d_reclen = UIO_MX;
1317 	ncookies = uio->uio_resid / UIO_MX;
1318 
1319 	switch (pfs->pfs_type) {
1320 	/*
1321 	 * this is for the process-specific sub-directories.
1322 	 * all that is needed to is copy out all the entries
1323 	 * from the procent[] table (top of this file).
1324 	 */
1325 	case PFSproc: {
1326 		struct proc *p;
1327 
1328 		if (i >= nproc_targets)
1329 			return 0;
1330 
1331 		if (procfs_proc_lock(pfs->pfs_pid, &p, ESRCH) != 0)
1332 			break;
1333 
1334 		if (ap->a_ncookies) {
1335 			ncookies = uimin(ncookies, (nproc_targets - i));
1336 			cookies = malloc(ncookies * sizeof (off_t),
1337 			    M_TEMP, M_WAITOK);
1338 			*ap->a_cookies = cookies;
1339 		}
1340 
1341 		for (pt = &proc_targets[i];
1342 		     uio->uio_resid >= UIO_MX && i < nproc_targets; pt++, i++) {
1343 			if (pt->pt_valid) {
1344 				/* XXXSMP LWP can disappear */
1345 				mutex_enter(p->p_lock);
1346 				l = LIST_FIRST(&p->p_lwps);
1347 				KASSERT(l != NULL);
1348 				mutex_exit(p->p_lock);
1349 				if ((*pt->pt_valid)(l, vp->v_mount) == 0)
1350 					continue;
1351 			}
1352 
1353 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid,
1354 			    pt->pt_pfstype, -1);
1355 			d.d_namlen = pt->pt_namlen;
1356 			memcpy(d.d_name, pt->pt_name, pt->pt_namlen + 1);
1357 			d.d_type = pt->pt_type;
1358 
1359 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1360 				break;
1361 			if (cookies)
1362 				*cookies++ = i + 1;
1363 		}
1364 
1365 		procfs_proc_unlock(p);
1366 	    	break;
1367 	}
1368 	case PFSfd: {
1369 		struct proc *p;
1370 		file_t *fp;
1371 		int lim, nc = 0;
1372 
1373 		if ((error = procfs_proc_lock(pfs->pfs_pid, &p, ESRCH)) != 0)
1374 			return error;
1375 
1376 		/* XXX Should this be by file as well? */
1377 		if (kauth_authorize_process(kauth_cred_get(),
1378 		    KAUTH_PROCESS_CANSEE, p,
1379 		    KAUTH_ARG(KAUTH_REQ_PROCESS_CANSEE_OPENFILES), NULL,
1380 		    NULL) != 0) {
1381 		    	procfs_proc_unlock(p);
1382 			return ESRCH;
1383 		}
1384 
1385 		nfd = p->p_fd->fd_dt->dt_nfiles;
1386 
1387 		lim = uimin((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfiles);
1388 		if (i >= lim) {
1389 		    	procfs_proc_unlock(p);
1390 			return 0;
1391 		}
1392 
1393 		if (ap->a_ncookies) {
1394 			ncookies = uimin(ncookies, (nfd + 2 - i));
1395 			cookies = malloc(ncookies * sizeof (off_t),
1396 			    M_TEMP, M_WAITOK);
1397 			*ap->a_cookies = cookies;
1398 		}
1399 
1400 		for (; i < 2 && uio->uio_resid >= UIO_MX; i++) {
1401 			pt = &proc_targets[i];
1402 			d.d_namlen = pt->pt_namlen;
1403 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid,
1404 			    pt->pt_pfstype, -1);
1405 			(void)memcpy(d.d_name, pt->pt_name, pt->pt_namlen + 1);
1406 			d.d_type = pt->pt_type;
1407 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1408 				break;
1409 			if (cookies)
1410 				*cookies++ = i + 1;
1411 			nc++;
1412 		}
1413 		if (error) {
1414 			ncookies = nc;
1415 			break;
1416 		}
1417 		for (; uio->uio_resid >= UIO_MX && i < nfd; i++) {
1418 			/* check the descriptor exists */
1419 			if ((fp = fd_getfile2(p, i - 2)) == NULL)
1420 				continue;
1421 			closef(fp);
1422 
1423 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid, PFSfd, i - 2);
1424 			d.d_namlen = snprintf(d.d_name, sizeof(d.d_name),
1425 			    "%lld", (long long)(i - 2));
1426 			d.d_type = VREG;
1427 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1428 				break;
1429 			if (cookies)
1430 				*cookies++ = i + 1;
1431 			nc++;
1432 		}
1433 		ncookies = nc;
1434 		procfs_proc_unlock(p);
1435 		break;
1436 	}
1437 	case PFStask: {
1438 		struct proc *p;
1439 		int nc = 0;
1440 
1441 		if ((error = procfs_proc_lock(pfs->pfs_pid, &p, ESRCH)) != 0)
1442 			return error;
1443 
1444 		nfd = 3;	/* ., .., pid */
1445 
1446 		if (ap->a_ncookies) {
1447 			ncookies = uimin(ncookies, (nfd + 2 - i));
1448 			cookies = malloc(ncookies * sizeof (off_t),
1449 			    M_TEMP, M_WAITOK);
1450 			*ap->a_cookies = cookies;
1451 		}
1452 
1453 		for (; i < 2 && uio->uio_resid >= UIO_MX; i++) {
1454 			pt = &proc_targets[i];
1455 			d.d_namlen = pt->pt_namlen;
1456 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid,
1457 			    pt->pt_pfstype, -1);
1458 			(void)memcpy(d.d_name, pt->pt_name, pt->pt_namlen + 1);
1459 			d.d_type = pt->pt_type;
1460 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1461 				break;
1462 			if (cookies)
1463 				*cookies++ = i + 1;
1464 			nc++;
1465 		}
1466 		if (error) {
1467 			ncookies = nc;
1468 			break;
1469 		}
1470 		for (; uio->uio_resid >= UIO_MX && i < nfd; i++) {
1471 			/* check the descriptor exists */
1472 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid, PFStask,
1473 			    i - 2);
1474 			d.d_namlen = snprintf(d.d_name, sizeof(d.d_name),
1475 			    "%ld", (long)pfs->pfs_pid);
1476 			d.d_type = DT_LNK;
1477 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1478 				break;
1479 			if (cookies)
1480 				*cookies++ = i + 1;
1481 			nc++;
1482 		}
1483 		ncookies = nc;
1484 		procfs_proc_unlock(p);
1485 		break;
1486 	}
1487 
1488 	/*
1489 	 * this is for the root of the procfs filesystem
1490 	 * what is needed are special entries for "curproc"
1491 	 * and "self" followed by an entry for each process
1492 	 * on allproc.
1493 	 */
1494 
1495 	case PFSroot: {
1496 		int nc = 0;
1497 
1498 		if (ap->a_ncookies) {
1499 			/*
1500 			 * XXX Potentially allocating too much space here,
1501 			 * but I'm lazy. This loop needs some work.
1502 			 */
1503 			cookies = malloc(ncookies * sizeof (off_t),
1504 			    M_TEMP, M_WAITOK);
1505 			*ap->a_cookies = cookies;
1506 		}
1507 		error = 0;
1508 		/* 0 ... 3 are static entries. */
1509 		for (; i <= 3 && uio->uio_resid >= UIO_MX; i++) {
1510 			switch (i) {
1511 			case 0:		/* `.' */
1512 			case 1:		/* `..' */
1513 				d.d_fileno = PROCFS_FILENO(0, PFSroot, -1);
1514 				d.d_namlen = i + 1;
1515 				memcpy(d.d_name, "..", d.d_namlen);
1516 				d.d_name[i + 1] = '\0';
1517 				d.d_type = DT_DIR;
1518 				break;
1519 
1520 			case 2:
1521 				d.d_fileno = PROCFS_FILENO(0, PFScurproc, -1);
1522 				d.d_namlen = sizeof("curproc") - 1;
1523 				memcpy(d.d_name, "curproc", sizeof("curproc"));
1524 				d.d_type = DT_LNK;
1525 				break;
1526 
1527 			case 3:
1528 				d.d_fileno = PROCFS_FILENO(0, PFSself, -1);
1529 				d.d_namlen = sizeof("self") - 1;
1530 				memcpy(d.d_name, "self", sizeof("self"));
1531 				d.d_type = DT_LNK;
1532 				break;
1533 			}
1534 
1535 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1536 				break;
1537 			nc++;
1538 			if (cookies)
1539 				*cookies++ = i + 1;
1540 		}
1541 		/* 4 ... are process entries. */
1542 		ctx.uiop = uio;
1543 		ctx.error = 0;
1544 		ctx.off = 4;
1545 		ctx.startoff = i;
1546 		ctx.cookies = cookies;
1547 		ctx.ncookies = nc;
1548 		proclist_foreach_call(&allproc,
1549 		    procfs_root_readdir_callback, &ctx);
1550 		cookies = ctx.cookies;
1551 		nc = ctx.ncookies;
1552 		error = ctx.error;
1553 		if (error)
1554 			break;
1555 
1556 		/* misc entries. */
1557 		if (i < ctx.off)
1558 			i = ctx.off;
1559 		if (i >= ctx.off + nproc_root_targets)
1560 			break;
1561 		for (pt = &proc_root_targets[i - ctx.off];
1562 		    uio->uio_resid >= UIO_MX &&
1563 		    pt < &proc_root_targets[nproc_root_targets];
1564 		    pt++, i++) {
1565 			if (pt->pt_valid &&
1566 			    (*pt->pt_valid)(NULL, vp->v_mount) == 0)
1567 				continue;
1568 			d.d_fileno = PROCFS_FILENO(0, pt->pt_pfstype, -1);
1569 			d.d_namlen = pt->pt_namlen;
1570 			memcpy(d.d_name, pt->pt_name, pt->pt_namlen + 1);
1571 			d.d_type = pt->pt_type;
1572 
1573 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1574 				break;
1575 			nc++;
1576 			if (cookies)
1577 				*cookies++ = i + 1;
1578 		}
1579 
1580 		ncookies = nc;
1581 		break;
1582 	}
1583 
1584 	default:
1585 		error = ENOTDIR;
1586 		break;
1587 	}
1588 
1589 	if (ap->a_ncookies) {
1590 		if (error) {
1591 			if (cookies)
1592 				free(*ap->a_cookies, M_TEMP);
1593 			*ap->a_ncookies = 0;
1594 			*ap->a_cookies = NULL;
1595 		} else
1596 			*ap->a_ncookies = ncookies;
1597 	}
1598 	uio->uio_offset = i;
1599 	return (error);
1600 }
1601 
1602 /*
1603  * readlink reads the link of `curproc' and others
1604  */
1605 int
1606 procfs_readlink(void *v)
1607 {
1608 	struct vop_readlink_args *ap = v;
1609 	char bf[16];		/* should be enough */
1610 	char *bp = bf;
1611 	char *path = NULL;
1612 	int len = 0;
1613 	int error = 0;
1614 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
1615 	struct proc *pown = NULL;
1616 
1617 	if (pfs->pfs_fileno == PROCFS_FILENO(0, PFScurproc, -1))
1618 		len = snprintf(bf, sizeof(bf), "%ld", (long)curproc->p_pid);
1619 	else if (pfs->pfs_fileno == PROCFS_FILENO(0, PFSself, -1))
1620 		len = snprintf(bf, sizeof(bf), "%s", "curproc");
1621 	else if (pfs->pfs_fileno == PROCFS_FILENO(pfs->pfs_pid, PFStask, 0))
1622 		len = snprintf(bf, sizeof(bf), "..");
1623 	else if (pfs->pfs_fileno == PROCFS_FILENO(pfs->pfs_pid, PFSexe, -1)) {
1624 		if ((error = procfs_proc_lock(pfs->pfs_pid, &pown, ESRCH)) != 0)
1625 			return error;
1626 		bp = pown->p_path;
1627 		len = strlen(bp);
1628 	} else if (pfs->pfs_fileno == PROCFS_FILENO(pfs->pfs_pid, PFScwd, -1) ||
1629 	    pfs->pfs_fileno == PROCFS_FILENO(pfs->pfs_pid, PFSchroot, -1)) {
1630 		if ((error = procfs_proc_lock(pfs->pfs_pid, &pown, ESRCH)) != 0)
1631 			return error;
1632 		path = malloc(MAXPATHLEN + 4, M_TEMP, M_WAITOK);
1633 		if (path == NULL) {
1634 			procfs_proc_unlock(pown);
1635 			return (ENOMEM);
1636 		}
1637 		bp = path + MAXPATHLEN;
1638 		*--bp = '\0';
1639 		procfs_dir(PROCFS_TYPE(pfs->pfs_fileno), curlwp, pown,
1640 		    &bp, path, MAXPATHLEN);
1641 		len = strlen(bp);
1642 	} else {
1643 		file_t *fp;
1644 		struct vnode *vxp, *vp;
1645 
1646 		if ((error = procfs_proc_lock(pfs->pfs_pid, &pown, ESRCH)) != 0)
1647 			return error;
1648 
1649 		fp = fd_getfile2(pown, pfs->pfs_fd);
1650 		if (fp == NULL) {
1651 			procfs_proc_unlock(pown);
1652 			return EBADF;
1653 		}
1654 
1655 		switch (fp->f_type) {
1656 		case DTYPE_VNODE:
1657 			vxp = fp->f_vnode;
1658 			if (vxp->v_type != VDIR) {
1659 				error = EINVAL;
1660 				break;
1661 			}
1662 			if ((path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK))
1663 			    == NULL) {
1664 				error = ENOMEM;
1665 				break;
1666 			}
1667 			bp = path + MAXPATHLEN;
1668 			*--bp = '\0';
1669 
1670 			/*
1671 			 * XXX: kludge to avoid locking against ourselves
1672 			 * in getcwd()
1673 			 */
1674 			if (vxp->v_tag == VT_PROCFS) {
1675 				*--bp = '/';
1676 			} else {
1677 				rw_enter(&curproc->p_cwdi->cwdi_lock,
1678 				    RW_READER);
1679 				vp = curproc->p_cwdi->cwdi_rdir;
1680 				if (vp == NULL)
1681 					vp = rootvnode;
1682 				error = getcwd_common(vxp, vp, &bp, path,
1683 				    MAXPATHLEN / 2, 0, curlwp);
1684 				rw_exit(&curproc->p_cwdi->cwdi_lock);
1685 			}
1686 			if (error)
1687 				break;
1688 			len = strlen(bp);
1689 			break;
1690 
1691 		case DTYPE_MISC:
1692 			len = snprintf(bf, sizeof(bf), "%s", "[misc]");
1693 			break;
1694 
1695 		case DTYPE_KQUEUE:
1696 			len = snprintf(bf, sizeof(bf), "%s", "[kqueue]");
1697 			break;
1698 
1699 		case DTYPE_SEM:
1700 			len = snprintf(bf, sizeof(bf), "%s", "[ksem]");
1701 			break;
1702 
1703 		default:
1704 			error = EINVAL;
1705 			break;
1706 		}
1707 		closef(fp);
1708 	}
1709 
1710 	if (error == 0)
1711 		error = uiomove(bp, len, ap->a_uio);
1712 	if (pown)
1713 		procfs_proc_unlock(pown);
1714 	if (path)
1715 		free(path, M_TEMP);
1716 	return error;
1717 }
1718 
1719 /*
1720  * convert decimal ascii to int
1721  */
1722 static int
1723 atoi(const char *b, size_t len)
1724 {
1725 	int p = 0;
1726 
1727 	while (len--) {
1728 		char c = *b++;
1729 		if (c < '0' || c > '9')
1730 			return -1;
1731 		p = 10 * p + (c - '0');
1732 	}
1733 
1734 	return p;
1735 }
1736