xref: /netbsd-src/sys/miscfs/procfs/procfs_vnops.c (revision aad9773e38ed2370a628a6416e098f9008fc10a7)
1 /*	$NetBSD: procfs_vnops.c,v 1.192 2014/09/05 09:26:16 matt 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 1.192 2014/09/05 09:26:16 matt 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 
127 #include <uvm/uvm_extern.h>	/* for PAGE_SIZE */
128 
129 #include <machine/reg.h>
130 
131 #include <miscfs/genfs/genfs.h>
132 #include <miscfs/procfs/procfs.h>
133 
134 /*
135  * Vnode Operations.
136  *
137  */
138 
139 static int procfs_validfile_linux(struct lwp *, struct mount *);
140 static int procfs_root_readdir_callback(struct proc *, void *);
141 static void procfs_dir(pfstype, struct lwp *, struct proc *, char **, char *,
142     size_t);
143 
144 /*
145  * This is a list of the valid names in the
146  * process-specific sub-directories.  It is
147  * used in procfs_lookup and procfs_readdir
148  */
149 static const struct proc_target {
150 	u_char	pt_type;
151 	u_char	pt_namlen;
152 	const char	*pt_name;
153 	pfstype	pt_pfstype;
154 	int	(*pt_valid)(struct lwp *, struct mount *);
155 } proc_targets[] = {
156 #define N(s) sizeof(s)-1, s
157 	/*	  name		type		validp */
158 	{ DT_DIR, N("."),	PFSproc,	NULL },
159 	{ DT_DIR, N(".."),	PFSroot,	NULL },
160 	{ DT_DIR, N("fd"),	PFSfd,		NULL },
161 	{ DT_REG, N("file"),	PFSfile,	procfs_validfile },
162 	{ DT_REG, N("mem"),	PFSmem,		NULL },
163 	{ DT_REG, N("regs"),	PFSregs,	procfs_validregs },
164 	{ DT_REG, N("fpregs"),	PFSfpregs,	procfs_validfpregs },
165 	{ DT_REG, N("ctl"),	PFSctl,		NULL },
166 	{ DT_REG, N("stat"),	PFSstat,	procfs_validfile_linux },
167 	{ DT_REG, N("status"),	PFSstatus,	NULL },
168 	{ DT_REG, N("note"),	PFSnote,	NULL },
169 	{ DT_REG, N("notepg"),	PFSnotepg,	NULL },
170 	{ DT_REG, N("map"),	PFSmap,		procfs_validmap },
171 	{ DT_REG, N("maps"),	PFSmaps,	procfs_validmap },
172 	{ DT_REG, N("cmdline"), PFScmdline,	NULL },
173 	{ DT_REG, N("exe"),	PFSexe,		procfs_validfile },
174 	{ DT_LNK, N("cwd"),	PFScwd,		NULL },
175 	{ DT_LNK, N("root"),	PFSchroot,	NULL },
176 	{ DT_LNK, N("emul"),	PFSemul,	NULL },
177 	{ DT_REG, N("statm"),	PFSstatm,	procfs_validfile_linux },
178 	{ DT_DIR, N("task"),	PFStask,	procfs_validfile_linux },
179 #ifdef __HAVE_PROCFS_MACHDEP
180 	PROCFS_MACHDEP_NODETYPE_DEFNS
181 #endif
182 #undef N
183 };
184 static const int nproc_targets = sizeof(proc_targets) / sizeof(proc_targets[0]);
185 
186 /*
187  * List of files in the root directory. Note: the validate function will
188  * be called with p == NULL for these ones.
189  */
190 static const struct proc_target proc_root_targets[] = {
191 #define N(s) sizeof(s)-1, s
192 	/*	  name		    type	    validp */
193 	{ DT_REG, N("meminfo"),     PFSmeminfo,        procfs_validfile_linux },
194 	{ DT_REG, N("cpuinfo"),     PFScpuinfo,        procfs_validfile_linux },
195 	{ DT_REG, N("uptime"),      PFSuptime,         procfs_validfile_linux },
196 	{ DT_REG, N("mounts"),	    PFSmounts,	       procfs_validfile_linux },
197 	{ DT_REG, N("devices"),     PFSdevices,        procfs_validfile_linux },
198 	{ DT_REG, N("stat"),	    PFScpustat,        procfs_validfile_linux },
199 	{ DT_REG, N("loadavg"),	    PFSloadavg,        procfs_validfile_linux },
200 	{ DT_REG, N("version"),     PFSversion,        procfs_validfile_linux },
201 #undef N
202 };
203 static const int nproc_root_targets =
204     sizeof(proc_root_targets) / sizeof(proc_root_targets[0]);
205 
206 int	procfs_lookup(void *);
207 #define	procfs_create	genfs_eopnotsupp
208 #define	procfs_mknod	genfs_eopnotsupp
209 int	procfs_open(void *);
210 int	procfs_close(void *);
211 int	procfs_access(void *);
212 int	procfs_getattr(void *);
213 int	procfs_setattr(void *);
214 #define	procfs_read	procfs_rw
215 #define	procfs_write	procfs_rw
216 #define	procfs_fcntl	genfs_fcntl
217 #define	procfs_ioctl	genfs_enoioctl
218 #define	procfs_poll	genfs_poll
219 #define procfs_revoke	genfs_revoke
220 #define	procfs_fsync	genfs_nullop
221 #define	procfs_seek	genfs_nullop
222 #define	procfs_remove	genfs_eopnotsupp
223 int	procfs_link(void *);
224 #define	procfs_rename	genfs_eopnotsupp
225 #define	procfs_mkdir	genfs_eopnotsupp
226 #define	procfs_rmdir	genfs_eopnotsupp
227 int	procfs_symlink(void *);
228 int	procfs_readdir(void *);
229 int	procfs_readlink(void *);
230 #define	procfs_abortop	genfs_abortop
231 int	procfs_inactive(void *);
232 int	procfs_reclaim(void *);
233 #define	procfs_lock	genfs_lock
234 #define	procfs_unlock	genfs_unlock
235 #define	procfs_bmap	genfs_badop
236 #define	procfs_strategy	genfs_badop
237 int	procfs_print(void *);
238 int	procfs_pathconf(void *);
239 #define	procfs_islocked	genfs_islocked
240 #define	procfs_advlock	genfs_einval
241 #define	procfs_bwrite	genfs_eopnotsupp
242 #define procfs_putpages	genfs_null_putpages
243 
244 static int atoi(const char *, size_t);
245 
246 /*
247  * procfs vnode operations.
248  */
249 int (**procfs_vnodeop_p)(void *);
250 const struct vnodeopv_entry_desc procfs_vnodeop_entries[] = {
251 	{ &vop_default_desc, vn_default_error },
252 	{ &vop_lookup_desc, procfs_lookup },		/* lookup */
253 	{ &vop_create_desc, procfs_create },		/* create */
254 	{ &vop_mknod_desc, procfs_mknod },		/* mknod */
255 	{ &vop_open_desc, procfs_open },		/* open */
256 	{ &vop_close_desc, procfs_close },		/* close */
257 	{ &vop_access_desc, procfs_access },		/* access */
258 	{ &vop_getattr_desc, procfs_getattr },		/* getattr */
259 	{ &vop_setattr_desc, procfs_setattr },		/* setattr */
260 	{ &vop_read_desc, procfs_read },		/* read */
261 	{ &vop_write_desc, procfs_write },		/* write */
262 	{ &vop_fallocate_desc, genfs_eopnotsupp },	/* fallocate */
263 	{ &vop_fdiscard_desc, genfs_eopnotsupp },	/* fdiscard */
264 	{ &vop_fcntl_desc, procfs_fcntl },		/* fcntl */
265 	{ &vop_ioctl_desc, procfs_ioctl },		/* ioctl */
266 	{ &vop_poll_desc, procfs_poll },		/* poll */
267 	{ &vop_revoke_desc, procfs_revoke },		/* revoke */
268 	{ &vop_fsync_desc, procfs_fsync },		/* fsync */
269 	{ &vop_seek_desc, procfs_seek },		/* seek */
270 	{ &vop_remove_desc, procfs_remove },		/* remove */
271 	{ &vop_link_desc, procfs_link },		/* link */
272 	{ &vop_rename_desc, procfs_rename },		/* rename */
273 	{ &vop_mkdir_desc, procfs_mkdir },		/* mkdir */
274 	{ &vop_rmdir_desc, procfs_rmdir },		/* rmdir */
275 	{ &vop_symlink_desc, procfs_symlink },		/* symlink */
276 	{ &vop_readdir_desc, procfs_readdir },		/* readdir */
277 	{ &vop_readlink_desc, procfs_readlink },	/* readlink */
278 	{ &vop_abortop_desc, procfs_abortop },		/* abortop */
279 	{ &vop_inactive_desc, procfs_inactive },	/* inactive */
280 	{ &vop_reclaim_desc, procfs_reclaim },		/* reclaim */
281 	{ &vop_lock_desc, procfs_lock },		/* lock */
282 	{ &vop_unlock_desc, procfs_unlock },		/* unlock */
283 	{ &vop_bmap_desc, procfs_bmap },		/* bmap */
284 	{ &vop_strategy_desc, procfs_strategy },	/* strategy */
285 	{ &vop_print_desc, procfs_print },		/* print */
286 	{ &vop_islocked_desc, procfs_islocked },	/* islocked */
287 	{ &vop_pathconf_desc, procfs_pathconf },	/* pathconf */
288 	{ &vop_advlock_desc, procfs_advlock },		/* advlock */
289 	{ &vop_putpages_desc, procfs_putpages },	/* putpages */
290 	{ NULL, NULL }
291 };
292 const struct vnodeopv_desc procfs_vnodeop_opv_desc =
293 	{ &procfs_vnodeop_p, procfs_vnodeop_entries };
294 /*
295  * set things up for doing i/o on
296  * the pfsnode (vp).  (vp) is locked
297  * on entry, and should be left locked
298  * on exit.
299  *
300  * for procfs we don't need to do anything
301  * in particular for i/o.  all that is done
302  * is to support exclusive open on process
303  * memory images.
304  */
305 int
306 procfs_open(void *v)
307 {
308 	struct vop_open_args /* {
309 		struct vnode *a_vp;
310 		int  a_mode;
311 		kauth_cred_t a_cred;
312 	} */ *ap = v;
313 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
314 	struct lwp *l1;
315 	struct proc *p2;
316 	int error;
317 
318 	if ((error = procfs_proc_lock(pfs->pfs_pid, &p2, ENOENT)) != 0)
319 		return error;
320 
321 	l1 = curlwp;				/* tracer */
322 
323 #define	M2K(m)	(((m) & FREAD) && ((m) & FWRITE) ? \
324 		 KAUTH_REQ_PROCESS_PROCFS_RW : \
325 		 (m) & FWRITE ? KAUTH_REQ_PROCESS_PROCFS_WRITE : \
326 		 KAUTH_REQ_PROCESS_PROCFS_READ)
327 
328 	mutex_enter(p2->p_lock);
329 	error = kauth_authorize_process(l1->l_cred, KAUTH_PROCESS_PROCFS,
330 	    p2, pfs, KAUTH_ARG(M2K(ap->a_mode)), NULL);
331 	mutex_exit(p2->p_lock);
332 	if (error) {
333 		procfs_proc_unlock(p2);
334 		return (error);
335 	}
336 
337 #undef M2K
338 
339 	switch (pfs->pfs_type) {
340 	case PFSmem:
341 		if (((pfs->pfs_flags & FWRITE) && (ap->a_mode & O_EXCL)) ||
342 		    ((pfs->pfs_flags & O_EXCL) && (ap->a_mode & FWRITE))) {
343 			error = EBUSY;
344 			break;
345 		}
346 
347 		if (!proc_isunder(p2, l1)) {
348 			error = EPERM;
349 			break;
350 		}
351 
352 		if (ap->a_mode & FWRITE)
353 			pfs->pfs_flags = ap->a_mode & (FWRITE|O_EXCL);
354 
355 		break;
356 
357 	case PFSregs:
358 	case PFSfpregs:
359 		if (!proc_isunder(p2, l1)) {
360 			error = EPERM;
361 			break;
362 		}
363 		break;
364 
365 	default:
366 		break;
367 	}
368 
369 	procfs_proc_unlock(p2);
370 	return (error);
371 }
372 
373 /*
374  * close the pfsnode (vp) after doing i/o.
375  * (vp) is not locked on entry or exit.
376  *
377  * nothing to do for procfs other than undo
378  * any exclusive open flag (see _open above).
379  */
380 int
381 procfs_close(void *v)
382 {
383 	struct vop_close_args /* {
384 		struct vnode *a_vp;
385 		int  a_fflag;
386 		kauth_cred_t a_cred;
387 	} */ *ap = v;
388 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
389 
390 	switch (pfs->pfs_type) {
391 	case PFSmem:
392 		if ((ap->a_fflag & FWRITE) && (pfs->pfs_flags & O_EXCL))
393 			pfs->pfs_flags &= ~(FWRITE|O_EXCL);
394 		break;
395 
396 	default:
397 		break;
398 	}
399 
400 	return (0);
401 }
402 
403 /*
404  * _inactive is called when the pfsnode
405  * is vrele'd and the reference count goes
406  * to zero.  (vp) will be on the vnode free
407  * list, so to get it back vget() must be
408  * used.
409  *
410  * (vp) is locked on entry, but must be unlocked on exit.
411  */
412 int
413 procfs_inactive(void *v)
414 {
415 	struct vop_inactive_args /* {
416 		struct vnode *a_vp;
417 		bool *a_recycle;
418 	} */ *ap = v;
419 	struct vnode *vp = ap->a_vp;
420 	struct pfsnode *pfs = VTOPFS(vp);
421 
422 	mutex_enter(proc_lock);
423 	*ap->a_recycle = (proc_find(pfs->pfs_pid) == NULL);
424 	mutex_exit(proc_lock);
425 
426 	VOP_UNLOCK(vp);
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_args /* {
442 		struct vnode *a_vp;
443 	} */ *ap = v;
444 	struct vnode *vp = ap->a_vp;
445 	struct pfsnode *pfs = VTOPFS(vp);
446 
447 	/*
448 	 * To interlock with procfs_revoke_vnodes().
449 	 */
450 	mutex_enter(vp->v_interlock);
451 	vp->v_data = NULL;
452 	mutex_exit(vp->v_interlock);
453 	vcache_remove(vp->v_mount, &pfs->pfs_key, sizeof(pfs->pfs_key));
454 	kmem_free(pfs, sizeof(*pfs));
455 	return 0;
456 }
457 
458 /*
459  * Return POSIX pathconf information applicable to special devices.
460  */
461 int
462 procfs_pathconf(void *v)
463 {
464 	struct vop_pathconf_args /* {
465 		struct vnode *a_vp;
466 		int a_name;
467 		register_t *a_retval;
468 	} */ *ap = v;
469 
470 	switch (ap->a_name) {
471 	case _PC_LINK_MAX:
472 		*ap->a_retval = LINK_MAX;
473 		return (0);
474 	case _PC_MAX_CANON:
475 		*ap->a_retval = MAX_CANON;
476 		return (0);
477 	case _PC_MAX_INPUT:
478 		*ap->a_retval = MAX_INPUT;
479 		return (0);
480 	case _PC_PIPE_BUF:
481 		*ap->a_retval = PIPE_BUF;
482 		return (0);
483 	case _PC_CHOWN_RESTRICTED:
484 		*ap->a_retval = 1;
485 		return (0);
486 	case _PC_VDISABLE:
487 		*ap->a_retval = _POSIX_VDISABLE;
488 		return (0);
489 	case _PC_SYNC_IO:
490 		*ap->a_retval = 1;
491 		return (0);
492 	default:
493 		return (EINVAL);
494 	}
495 	/* NOTREACHED */
496 }
497 
498 /*
499  * _print is used for debugging.
500  * just print a readable description
501  * of (vp).
502  */
503 int
504 procfs_print(void *v)
505 {
506 	struct vop_print_args /* {
507 		struct vnode *a_vp;
508 	} */ *ap = v;
509 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
510 
511 	printf("tag VT_PROCFS, type %d, pid %d, mode %x, flags %lx\n",
512 	    pfs->pfs_type, pfs->pfs_pid, pfs->pfs_mode, pfs->pfs_flags);
513 	return 0;
514 }
515 
516 int
517 procfs_link(void *v)
518 {
519 	struct vop_link_args /* {
520 		struct vnode *a_dvp;
521 		struct vnode *a_vp;
522 		struct componentname *a_cnp;
523 	} */ *ap = v;
524 
525 	VOP_ABORTOP(ap->a_dvp, ap->a_cnp);
526 	vput(ap->a_dvp);
527 	return (EROFS);
528 }
529 
530 int
531 procfs_symlink(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 PFSmap:
741 	case PFSmaps:
742 	case PFScmdline:
743 	case PFSemul:
744 	case PFSstatm:
745 		if (pfs->pfs_type == PFSmap || pfs->pfs_type == PFSmaps)
746 			vap->va_mode = S_IRUSR;
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 	default:
764 		break;
765 	}
766 
767 	/*
768 	 * now do the object specific fields
769 	 *
770 	 * The size could be set from struct reg, but it's hardly
771 	 * worth the trouble, and it puts some (potentially) machine
772 	 * dependent data into this machine-independent code.  If it
773 	 * becomes important then this function should break out into
774 	 * a per-file stat function in the corresponding .c file.
775 	 */
776 
777 	switch (pfs->pfs_type) {
778 	case PFSroot:
779 		/*
780 		 * Set nlink to 1 to tell fts(3) we don't actually know.
781 		 */
782 		vap->va_nlink = 1;
783 		vap->va_uid = 0;
784 		vap->va_gid = 0;
785 		vap->va_bytes = vap->va_size = DEV_BSIZE;
786 		break;
787 
788 	case PFSself:
789 	case PFScurproc: {
790 		char bf[16];		/* should be enough */
791 		vap->va_nlink = 1;
792 		vap->va_uid = 0;
793 		vap->va_gid = 0;
794 		vap->va_bytes = vap->va_size =
795 		    snprintf(bf, sizeof(bf), "%ld", (long)curproc->p_pid);
796 		break;
797 	}
798 	case PFStask:
799 		if (pfs->pfs_fd != -1) {
800 			char bf[4];		/* should be enough */
801 			vap->va_nlink = 1;
802 			vap->va_uid = 0;
803 			vap->va_gid = 0;
804 			vap->va_bytes = vap->va_size =
805 			    snprintf(bf, sizeof(bf), "..");
806 			break;
807 		}
808 		/*FALLTHROUGH*/
809 	case PFSfd:
810 		if (pfs->pfs_fd != -1) {
811 			file_t *fp;
812 
813 			fp = fd_getfile2(procp, pfs->pfs_fd);
814 			if (fp == NULL) {
815 				error = EBADF;
816 				break;
817 			}
818 			vap->va_nlink = 1;
819 			vap->va_uid = kauth_cred_geteuid(fp->f_cred);
820 			vap->va_gid = kauth_cred_getegid(fp->f_cred);
821 			switch (fp->f_type) {
822 			case DTYPE_VNODE:
823 				vap->va_bytes = vap->va_size =
824 				    fp->f_vnode->v_size;
825 				break;
826 			default:
827 				vap->va_bytes = vap->va_size = 0;
828 				break;
829 			}
830 			closef(fp);
831 			break;
832 		}
833 		/*FALLTHROUGH*/
834 	case PFSproc:
835 		vap->va_nlink = 2;
836 		vap->va_uid = kauth_cred_geteuid(procp->p_cred);
837 		vap->va_gid = kauth_cred_getegid(procp->p_cred);
838 		vap->va_bytes = vap->va_size = DEV_BSIZE;
839 		break;
840 
841 	case PFSfile:
842 		error = EOPNOTSUPP;
843 		break;
844 
845 	case PFSmem:
846 		vap->va_bytes = vap->va_size =
847 			ctob(procp->p_vmspace->vm_tsize +
848 				    procp->p_vmspace->vm_dsize +
849 				    procp->p_vmspace->vm_ssize);
850 		break;
851 
852 #if defined(PT_GETREGS) || defined(PT_SETREGS)
853 	case PFSregs:
854 		vap->va_bytes = vap->va_size = sizeof(struct reg);
855 		break;
856 #endif
857 
858 #if defined(PT_GETFPREGS) || defined(PT_SETFPREGS)
859 	case PFSfpregs:
860 		vap->va_bytes = vap->va_size = sizeof(struct fpreg);
861 		break;
862 #endif
863 
864 	case PFSctl:
865 	case PFSstatus:
866 	case PFSstat:
867 	case PFSnote:
868 	case PFSnotepg:
869 	case PFScmdline:
870 	case PFSmeminfo:
871 	case PFSdevices:
872 	case PFScpuinfo:
873 	case PFSuptime:
874 	case PFSmounts:
875 	case PFScpustat:
876 	case PFSloadavg:
877 	case PFSstatm:
878 	case PFSversion:
879 		vap->va_bytes = vap->va_size = 0;
880 		break;
881 	case PFSmap:
882 	case PFSmaps:
883 		/*
884 		 * Advise a larger blocksize for the map files, so that
885 		 * they may be read in one pass.
886 		 */
887 		vap->va_blocksize = 4 * PAGE_SIZE;
888 		vap->va_bytes = vap->va_size = 0;
889 		break;
890 
891 	case PFScwd:
892 	case PFSchroot:
893 	case PFSexe: {
894 		char *bp;
895 
896 		vap->va_nlink = 1;
897 		vap->va_uid = 0;
898 		vap->va_gid = 0;
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 
907 	case PFSemul:
908 		vap->va_bytes = vap->va_size = strlen(procp->p_emul->e_name);
909 		break;
910 
911 #ifdef __HAVE_PROCFS_MACHDEP
912 	PROCFS_MACHDEP_NODETYPE_CASES
913 		error = procfs_machdep_getattr(ap->a_vp, vap, procp);
914 		break;
915 #endif
916 
917 	default:
918 		panic("procfs_getattr");
919 	}
920 
921 	if (procp != NULL)
922 		procfs_proc_unlock(procp);
923 	if (path != NULL)
924 		free(path, M_TEMP);
925 
926 	return (error);
927 }
928 
929 /*ARGSUSED*/
930 int
931 procfs_setattr(void *v)
932 {
933 	/*
934 	 * just fake out attribute setting
935 	 * it's not good to generate an error
936 	 * return, otherwise things like creat()
937 	 * will fail when they try to set the
938 	 * file length to 0.  worse, this means
939 	 * that echo $note > /proc/$pid/note will fail.
940 	 */
941 
942 	return (0);
943 }
944 
945 /*
946  * implement access checking.
947  *
948  * actually, the check for super-user is slightly
949  * broken since it will allow read access to write-only
950  * objects.  this doesn't cause any particular trouble
951  * but does mean that the i/o entry points need to check
952  * that the operation really does make sense.
953  */
954 int
955 procfs_access(void *v)
956 {
957 	struct vop_access_args /* {
958 		struct vnode *a_vp;
959 		int a_mode;
960 		kauth_cred_t a_cred;
961 	} */ *ap = v;
962 	struct vattr va;
963 	int error;
964 
965 	if ((error = VOP_GETATTR(ap->a_vp, &va, ap->a_cred)) != 0)
966 		return (error);
967 
968 	return kauth_authorize_vnode(ap->a_cred,
969 	    KAUTH_ACCESS_ACTION(ap->a_mode, ap->a_vp->v_type, va.va_mode),
970 	    ap->a_vp, NULL, genfs_can_access(va.va_type, va.va_mode,
971 	    va.va_uid, va.va_gid, ap->a_mode, ap->a_cred));
972 }
973 
974 /*
975  * lookup.  this is incredibly complicated in the
976  * general case, however for most pseudo-filesystems
977  * very little needs to be done.
978  *
979  * Locking isn't hard here, just poorly documented.
980  *
981  * If we're looking up ".", just vref the parent & return it.
982  *
983  * If we're looking up "..", unlock the parent, and lock "..". If everything
984  * went ok, and we're on the last component and the caller requested the
985  * parent locked, try to re-lock the parent. We do this to prevent lock
986  * races.
987  *
988  * For anything else, get the needed node. Then unlock the parent if not
989  * the last component or not LOCKPARENT (i.e. if we wouldn't re-lock the
990  * parent in the .. case).
991  *
992  * We try to exit with the parent locked in error cases.
993  */
994 int
995 procfs_lookup(void *v)
996 {
997 	struct vop_lookup_v2_args /* {
998 		struct vnode * a_dvp;
999 		struct vnode ** a_vpp;
1000 		struct componentname * a_cnp;
1001 	} */ *ap = v;
1002 	struct componentname *cnp = ap->a_cnp;
1003 	struct vnode **vpp = ap->a_vpp;
1004 	struct vnode *dvp = ap->a_dvp;
1005 	const char *pname = cnp->cn_nameptr;
1006 	const struct proc_target *pt = NULL;
1007 	struct vnode *fvp;
1008 	pid_t pid, vnpid;
1009 	struct pfsnode *pfs;
1010 	struct proc *p = NULL;
1011 	struct lwp *plwp;
1012 	int i, error;
1013 	pfstype type;
1014 
1015 	*vpp = NULL;
1016 
1017 	if (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)
1018 		return (EROFS);
1019 
1020 	if (cnp->cn_namelen == 1 && *pname == '.') {
1021 		*vpp = dvp;
1022 		vref(dvp);
1023 		return (0);
1024 	}
1025 
1026 	pfs = VTOPFS(dvp);
1027 	switch (pfs->pfs_type) {
1028 	case PFSroot:
1029 		/*
1030 		 * Shouldn't get here with .. in the root node.
1031 		 */
1032 		if (cnp->cn_flags & ISDOTDOT)
1033 			return (EIO);
1034 
1035 		for (i = 0; i < nproc_root_targets; i++) {
1036 			pt = &proc_root_targets[i];
1037 			/*
1038 			 * check for node match.  proc is always NULL here,
1039 			 * so call pt_valid with constant NULL lwp.
1040 			 */
1041 			if (cnp->cn_namelen == pt->pt_namlen &&
1042 			    memcmp(pt->pt_name, pname, cnp->cn_namelen) == 0 &&
1043 			    (pt->pt_valid == NULL ||
1044 			     (*pt->pt_valid)(NULL, dvp->v_mount)))
1045 				break;
1046 		}
1047 
1048 		if (i != nproc_root_targets) {
1049 			error = procfs_allocvp(dvp->v_mount, vpp, 0,
1050 			    pt->pt_pfstype, -1);
1051 			return (error);
1052 		}
1053 
1054 		if (CNEQ(cnp, "curproc", 7)) {
1055 			pid = curproc->p_pid;
1056 			vnpid = 0;
1057 			type = PFScurproc;
1058 		} else if (CNEQ(cnp, "self", 4)) {
1059 			pid = curproc->p_pid;
1060 			vnpid = 0;
1061 			type = PFSself;
1062 		} else {
1063 			pid = (pid_t)atoi(pname, cnp->cn_namelen);
1064 			vnpid = pid;
1065 			type = PFSproc;
1066 		}
1067 
1068 		if (procfs_proc_lock(pid, &p, ESRCH) != 0)
1069 			break;
1070 		error = procfs_allocvp(dvp->v_mount, vpp, vnpid, type, -1);
1071 		procfs_proc_unlock(p);
1072 		return (error);
1073 
1074 	case PFSproc:
1075 		if (cnp->cn_flags & ISDOTDOT) {
1076 			error = procfs_allocvp(dvp->v_mount, vpp, 0, PFSroot,
1077 			    -1);
1078 			return (error);
1079 		}
1080 
1081 		if (procfs_proc_lock(pfs->pfs_pid, &p, ESRCH) != 0)
1082 			break;
1083 
1084 		mutex_enter(p->p_lock);
1085 		LIST_FOREACH(plwp, &p->p_lwps, l_sibling) {
1086 			if (plwp->l_stat != LSZOMB)
1087 				break;
1088 		}
1089 		/* Process is exiting if no-LWPS or all LWPs are LSZOMB */
1090 		if (plwp == NULL) {
1091 			mutex_exit(p->p_lock);
1092 			procfs_proc_unlock(p);
1093 			return ESRCH;
1094 		}
1095 
1096 		lwp_addref(plwp);
1097 		mutex_exit(p->p_lock);
1098 
1099 		for (pt = proc_targets, i = 0; i < nproc_targets; pt++, i++) {
1100 			int found;
1101 
1102 			found = cnp->cn_namelen == pt->pt_namlen &&
1103 			    memcmp(pt->pt_name, pname, cnp->cn_namelen) == 0 &&
1104 			    (pt->pt_valid == NULL
1105 			      || (*pt->pt_valid)(plwp, dvp->v_mount));
1106 			if (found)
1107 				break;
1108 		}
1109 		lwp_delref(plwp);
1110 
1111 		if (i == nproc_targets) {
1112 			procfs_proc_unlock(p);
1113 			break;
1114 		}
1115 		if (pt->pt_pfstype == PFSfile) {
1116 			fvp = p->p_textvp;
1117 			/* We already checked that it exists. */
1118 			vref(fvp);
1119 			procfs_proc_unlock(p);
1120 			*vpp = fvp;
1121 			return (0);
1122 		}
1123 
1124 		error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
1125 		    pt->pt_pfstype, -1);
1126 		procfs_proc_unlock(p);
1127 		return (error);
1128 
1129 	case PFSfd: {
1130 		int fd;
1131 		file_t *fp;
1132 
1133 		if ((error = procfs_proc_lock(pfs->pfs_pid, &p, ENOENT)) != 0)
1134 			return error;
1135 
1136 		if (cnp->cn_flags & ISDOTDOT) {
1137 			error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
1138 			    PFSproc, -1);
1139 			procfs_proc_unlock(p);
1140 			return (error);
1141 		}
1142 		fd = atoi(pname, cnp->cn_namelen);
1143 
1144 		fp = fd_getfile2(p, fd);
1145 		if (fp == NULL) {
1146 			procfs_proc_unlock(p);
1147 			return ENOENT;
1148 		}
1149 		fvp = fp->f_vnode;
1150 
1151 		/* Don't show directories */
1152 		if (fp->f_type == DTYPE_VNODE && fvp->v_type != VDIR) {
1153 			vref(fvp);
1154 			closef(fp);
1155 			procfs_proc_unlock(p);
1156 			*vpp = fvp;
1157 			return 0;
1158 		}
1159 
1160 		closef(fp);
1161 		error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
1162 		    PFSfd, fd);
1163 		procfs_proc_unlock(p);
1164 		return error;
1165 	}
1166 	case PFStask: {
1167 		int xpid;
1168 
1169 		if ((error = procfs_proc_lock(pfs->pfs_pid, &p, ENOENT)) != 0)
1170 			return error;
1171 
1172 		if (cnp->cn_flags & ISDOTDOT) {
1173 			error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
1174 			    PFSproc, -1);
1175 			procfs_proc_unlock(p);
1176 			return (error);
1177 		}
1178 		xpid = atoi(pname, cnp->cn_namelen);
1179 
1180 		if (xpid != pfs->pfs_pid) {
1181 			procfs_proc_unlock(p);
1182 			return ENOENT;
1183 		}
1184 		error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
1185 		    PFStask, 0);
1186 		procfs_proc_unlock(p);
1187 		return error;
1188 	}
1189 	default:
1190 		return (ENOTDIR);
1191 	}
1192 
1193 	return (cnp->cn_nameiop == LOOKUP ? ENOENT : EROFS);
1194 }
1195 
1196 int
1197 procfs_validfile(struct lwp *l, struct mount *mp)
1198 {
1199 	return l != NULL && l->l_proc != NULL && l->l_proc->p_textvp != NULL;
1200 }
1201 
1202 static int
1203 procfs_validfile_linux(struct lwp *l, struct mount *mp)
1204 {
1205 	int flags;
1206 
1207 	flags = VFSTOPROC(mp)->pmnt_flags;
1208 	return (flags & PROCFSMNT_LINUXCOMPAT) &&
1209 	    (l == NULL || l->l_proc == NULL || procfs_validfile(l, mp));
1210 }
1211 
1212 struct procfs_root_readdir_ctx {
1213 	struct uio *uiop;
1214 	off_t *cookies;
1215 	int ncookies;
1216 	off_t off;
1217 	off_t startoff;
1218 	int error;
1219 };
1220 
1221 static int
1222 procfs_root_readdir_callback(struct proc *p, void *arg)
1223 {
1224 	struct procfs_root_readdir_ctx *ctxp = arg;
1225 	struct dirent d;
1226 	struct uio *uiop;
1227 	int error;
1228 
1229 	uiop = ctxp->uiop;
1230 	if (uiop->uio_resid < UIO_MX)
1231 		return -1; /* no space */
1232 
1233 	if (ctxp->off < ctxp->startoff) {
1234 		ctxp->off++;
1235 		return 0;
1236 	}
1237 
1238 	if (kauth_authorize_process(kauth_cred_get(),
1239 	    KAUTH_PROCESS_CANSEE, p,
1240 	    KAUTH_ARG(KAUTH_REQ_PROCESS_CANSEE_ENTRY), NULL, NULL) != 0)
1241 		return 0;
1242 
1243 	memset(&d, 0, UIO_MX);
1244 	d.d_reclen = UIO_MX;
1245 	d.d_fileno = PROCFS_FILENO(p->p_pid, PFSproc, -1);
1246 	d.d_namlen = snprintf(d.d_name,
1247 	    UIO_MX - offsetof(struct dirent, d_name), "%ld", (long)p->p_pid);
1248 	d.d_type = DT_DIR;
1249 
1250 	mutex_exit(proc_lock);
1251 	error = uiomove(&d, UIO_MX, uiop);
1252 	mutex_enter(proc_lock);
1253 	if (error) {
1254 		ctxp->error = error;
1255 		return -1;
1256 	}
1257 
1258 	ctxp->ncookies++;
1259 	if (ctxp->cookies)
1260 		*(ctxp->cookies)++ = ctxp->off + 1;
1261 	ctxp->off++;
1262 
1263 	return 0;
1264 }
1265 
1266 /*
1267  * readdir returns directory entries from pfsnode (vp).
1268  *
1269  * the strategy here with procfs is to generate a single
1270  * directory entry at a time (struct dirent) and then
1271  * copy that out to userland using uiomove.  a more efficent
1272  * though more complex implementation, would try to minimize
1273  * the number of calls to uiomove().  for procfs, this is
1274  * hardly worth the added code complexity.
1275  *
1276  * this should just be done through read()
1277  */
1278 int
1279 procfs_readdir(void *v)
1280 {
1281 	struct vop_readdir_args /* {
1282 		struct vnode *a_vp;
1283 		struct uio *a_uio;
1284 		kauth_cred_t a_cred;
1285 		int *a_eofflag;
1286 		off_t **a_cookies;
1287 		int *a_ncookies;
1288 	} */ *ap = v;
1289 	struct uio *uio = ap->a_uio;
1290 	struct dirent d;
1291 	struct pfsnode *pfs;
1292 	off_t i;
1293 	int error;
1294 	off_t *cookies = NULL;
1295 	int ncookies;
1296 	struct vnode *vp;
1297 	const struct proc_target *pt;
1298 	struct procfs_root_readdir_ctx ctx;
1299 	struct lwp *l;
1300 	int nfd;
1301 
1302 	vp = ap->a_vp;
1303 	pfs = VTOPFS(vp);
1304 
1305 	if (uio->uio_resid < UIO_MX)
1306 		return (EINVAL);
1307 	if (uio->uio_offset < 0)
1308 		return (EINVAL);
1309 
1310 	error = 0;
1311 	i = uio->uio_offset;
1312 	memset(&d, 0, UIO_MX);
1313 	d.d_reclen = UIO_MX;
1314 	ncookies = uio->uio_resid / UIO_MX;
1315 
1316 	switch (pfs->pfs_type) {
1317 	/*
1318 	 * this is for the process-specific sub-directories.
1319 	 * all that is needed to is copy out all the entries
1320 	 * from the procent[] table (top of this file).
1321 	 */
1322 	case PFSproc: {
1323 		struct proc *p;
1324 
1325 		if (i >= nproc_targets)
1326 			return 0;
1327 
1328 		if (procfs_proc_lock(pfs->pfs_pid, &p, ESRCH) != 0)
1329 			break;
1330 
1331 		if (ap->a_ncookies) {
1332 			ncookies = min(ncookies, (nproc_targets - i));
1333 			cookies = malloc(ncookies * sizeof (off_t),
1334 			    M_TEMP, M_WAITOK);
1335 			*ap->a_cookies = cookies;
1336 		}
1337 
1338 		for (pt = &proc_targets[i];
1339 		     uio->uio_resid >= UIO_MX && i < nproc_targets; pt++, i++) {
1340 			if (pt->pt_valid) {
1341 				/* XXXSMP LWP can disappear */
1342 				mutex_enter(p->p_lock);
1343 				l = LIST_FIRST(&p->p_lwps);
1344 				KASSERT(l != NULL);
1345 				mutex_exit(p->p_lock);
1346 				if ((*pt->pt_valid)(l, vp->v_mount) == 0)
1347 					continue;
1348 			}
1349 
1350 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid,
1351 			    pt->pt_pfstype, -1);
1352 			d.d_namlen = pt->pt_namlen;
1353 			memcpy(d.d_name, pt->pt_name, pt->pt_namlen + 1);
1354 			d.d_type = pt->pt_type;
1355 
1356 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1357 				break;
1358 			if (cookies)
1359 				*cookies++ = i + 1;
1360 		}
1361 
1362 		procfs_proc_unlock(p);
1363 	    	break;
1364 	}
1365 	case PFSfd: {
1366 		struct proc *p;
1367 		file_t *fp;
1368 		int lim, nc = 0;
1369 
1370 		if ((error = procfs_proc_lock(pfs->pfs_pid, &p, ESRCH)) != 0)
1371 			return error;
1372 
1373 		/* XXX Should this be by file as well? */
1374 		if (kauth_authorize_process(kauth_cred_get(),
1375 		    KAUTH_PROCESS_CANSEE, p,
1376 		    KAUTH_ARG(KAUTH_REQ_PROCESS_CANSEE_OPENFILES), NULL,
1377 		    NULL) != 0) {
1378 		    	procfs_proc_unlock(p);
1379 			return ESRCH;
1380 		}
1381 
1382 		nfd = p->p_fd->fd_dt->dt_nfiles;
1383 
1384 		lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfiles);
1385 		if (i >= lim) {
1386 		    	procfs_proc_unlock(p);
1387 			return 0;
1388 		}
1389 
1390 		if (ap->a_ncookies) {
1391 			ncookies = min(ncookies, (nfd + 2 - i));
1392 			cookies = malloc(ncookies * sizeof (off_t),
1393 			    M_TEMP, M_WAITOK);
1394 			*ap->a_cookies = cookies;
1395 		}
1396 
1397 		for (; i < 2 && uio->uio_resid >= UIO_MX; i++) {
1398 			pt = &proc_targets[i];
1399 			d.d_namlen = pt->pt_namlen;
1400 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid,
1401 			    pt->pt_pfstype, -1);
1402 			(void)memcpy(d.d_name, pt->pt_name, pt->pt_namlen + 1);
1403 			d.d_type = pt->pt_type;
1404 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1405 				break;
1406 			if (cookies)
1407 				*cookies++ = i + 1;
1408 			nc++;
1409 		}
1410 		if (error) {
1411 			ncookies = nc;
1412 			break;
1413 		}
1414 		for (; uio->uio_resid >= UIO_MX && i < nfd; i++) {
1415 			/* check the descriptor exists */
1416 			if ((fp = fd_getfile2(p, i - 2)) == NULL)
1417 				continue;
1418 			closef(fp);
1419 
1420 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid, PFSfd, i - 2);
1421 			d.d_namlen = snprintf(d.d_name, sizeof(d.d_name),
1422 			    "%lld", (long long)(i - 2));
1423 			d.d_type = VREG;
1424 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1425 				break;
1426 			if (cookies)
1427 				*cookies++ = i + 1;
1428 			nc++;
1429 		}
1430 		ncookies = nc;
1431 		procfs_proc_unlock(p);
1432 		break;
1433 	}
1434 	case PFStask: {
1435 		struct proc *p;
1436 		int nc = 0;
1437 
1438 		if ((error = procfs_proc_lock(pfs->pfs_pid, &p, ESRCH)) != 0)
1439 			return error;
1440 
1441 		nfd = 3;	/* ., .., pid */
1442 
1443 		if (ap->a_ncookies) {
1444 			ncookies = min(ncookies, (nfd + 2 - i));
1445 			cookies = malloc(ncookies * sizeof (off_t),
1446 			    M_TEMP, M_WAITOK);
1447 			*ap->a_cookies = cookies;
1448 		}
1449 
1450 		for (; i < 2 && uio->uio_resid >= UIO_MX; i++) {
1451 			pt = &proc_targets[i];
1452 			d.d_namlen = pt->pt_namlen;
1453 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid,
1454 			    pt->pt_pfstype, -1);
1455 			(void)memcpy(d.d_name, pt->pt_name, pt->pt_namlen + 1);
1456 			d.d_type = pt->pt_type;
1457 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1458 				break;
1459 			if (cookies)
1460 				*cookies++ = i + 1;
1461 			nc++;
1462 		}
1463 		if (error) {
1464 			ncookies = nc;
1465 			break;
1466 		}
1467 		for (; uio->uio_resid >= UIO_MX && i < nfd; i++) {
1468 			/* check the descriptor exists */
1469 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid, PFStask,
1470 			    i - 2);
1471 			d.d_namlen = snprintf(d.d_name, sizeof(d.d_name),
1472 			    "%ld", (long)pfs->pfs_pid);
1473 			d.d_type = DT_LNK;
1474 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1475 				break;
1476 			if (cookies)
1477 				*cookies++ = i + 1;
1478 			nc++;
1479 		}
1480 		ncookies = nc;
1481 		procfs_proc_unlock(p);
1482 		break;
1483 	}
1484 
1485 	/*
1486 	 * this is for the root of the procfs filesystem
1487 	 * what is needed are special entries for "curproc"
1488 	 * and "self" followed by an entry for each process
1489 	 * on allproc.
1490 	 */
1491 
1492 	case PFSroot: {
1493 		int nc = 0;
1494 
1495 		if (ap->a_ncookies) {
1496 			/*
1497 			 * XXX Potentially allocating too much space here,
1498 			 * but I'm lazy. This loop needs some work.
1499 			 */
1500 			cookies = malloc(ncookies * sizeof (off_t),
1501 			    M_TEMP, M_WAITOK);
1502 			*ap->a_cookies = cookies;
1503 		}
1504 		error = 0;
1505 		/* 0 ... 3 are static entries. */
1506 		for (; i <= 3 && uio->uio_resid >= UIO_MX; i++) {
1507 			switch (i) {
1508 			case 0:		/* `.' */
1509 			case 1:		/* `..' */
1510 				d.d_fileno = PROCFS_FILENO(0, PFSroot, -1);
1511 				d.d_namlen = i + 1;
1512 				memcpy(d.d_name, "..", d.d_namlen);
1513 				d.d_name[i + 1] = '\0';
1514 				d.d_type = DT_DIR;
1515 				break;
1516 
1517 			case 2:
1518 				d.d_fileno = PROCFS_FILENO(0, PFScurproc, -1);
1519 				d.d_namlen = sizeof("curproc") - 1;
1520 				memcpy(d.d_name, "curproc", sizeof("curproc"));
1521 				d.d_type = DT_LNK;
1522 				break;
1523 
1524 			case 3:
1525 				d.d_fileno = PROCFS_FILENO(0, PFSself, -1);
1526 				d.d_namlen = sizeof("self") - 1;
1527 				memcpy(d.d_name, "self", sizeof("self"));
1528 				d.d_type = DT_LNK;
1529 				break;
1530 			}
1531 
1532 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1533 				break;
1534 			nc++;
1535 			if (cookies)
1536 				*cookies++ = i + 1;
1537 		}
1538 		/* 4 ... are process entries. */
1539 		ctx.uiop = uio;
1540 		ctx.error = 0;
1541 		ctx.off = 4;
1542 		ctx.startoff = i;
1543 		ctx.cookies = cookies;
1544 		ctx.ncookies = nc;
1545 		proclist_foreach_call(&allproc,
1546 		    procfs_root_readdir_callback, &ctx);
1547 		cookies = ctx.cookies;
1548 		nc = ctx.ncookies;
1549 		error = ctx.error;
1550 		if (error)
1551 			break;
1552 
1553 		/* misc entries. */
1554 		if (i < ctx.off)
1555 			i = ctx.off;
1556 		if (i >= ctx.off + nproc_root_targets)
1557 			break;
1558 		for (pt = &proc_root_targets[i - ctx.off];
1559 		    uio->uio_resid >= UIO_MX &&
1560 		    pt < &proc_root_targets[nproc_root_targets];
1561 		    pt++, i++) {
1562 			if (pt->pt_valid &&
1563 			    (*pt->pt_valid)(NULL, vp->v_mount) == 0)
1564 				continue;
1565 			d.d_fileno = PROCFS_FILENO(0, pt->pt_pfstype, -1);
1566 			d.d_namlen = pt->pt_namlen;
1567 			memcpy(d.d_name, pt->pt_name, pt->pt_namlen + 1);
1568 			d.d_type = pt->pt_type;
1569 
1570 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1571 				break;
1572 			nc++;
1573 			if (cookies)
1574 				*cookies++ = i + 1;
1575 		}
1576 
1577 		ncookies = nc;
1578 		break;
1579 	}
1580 
1581 	default:
1582 		error = ENOTDIR;
1583 		break;
1584 	}
1585 
1586 	if (ap->a_ncookies) {
1587 		if (error) {
1588 			if (cookies)
1589 				free(*ap->a_cookies, M_TEMP);
1590 			*ap->a_ncookies = 0;
1591 			*ap->a_cookies = NULL;
1592 		} else
1593 			*ap->a_ncookies = ncookies;
1594 	}
1595 	uio->uio_offset = i;
1596 	return (error);
1597 }
1598 
1599 /*
1600  * readlink reads the link of `curproc' and others
1601  */
1602 int
1603 procfs_readlink(void *v)
1604 {
1605 	struct vop_readlink_args *ap = v;
1606 	char bf[16];		/* should be enough */
1607 	char *bp = bf;
1608 	char *path = NULL;
1609 	int len = 0;
1610 	int error = 0;
1611 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
1612 	struct proc *pown;
1613 
1614 	if (pfs->pfs_fileno == PROCFS_FILENO(0, PFScurproc, -1))
1615 		len = snprintf(bf, sizeof(bf), "%ld", (long)curproc->p_pid);
1616 	else if (pfs->pfs_fileno == PROCFS_FILENO(0, PFSself, -1))
1617 		len = snprintf(bf, sizeof(bf), "%s", "curproc");
1618 	else if (pfs->pfs_fileno == PROCFS_FILENO(pfs->pfs_pid, PFStask, 0))
1619 		len = snprintf(bf, sizeof(bf), "..");
1620 	else if (pfs->pfs_fileno == PROCFS_FILENO(pfs->pfs_pid, PFScwd, -1) ||
1621 	    pfs->pfs_fileno == PROCFS_FILENO(pfs->pfs_pid, PFSchroot, -1) ||
1622 	    pfs->pfs_fileno == PROCFS_FILENO(pfs->pfs_pid, PFSexe, -1)) {
1623 		if ((error = procfs_proc_lock(pfs->pfs_pid, &pown, ESRCH)) != 0)
1624 			return error;
1625 		path = malloc(MAXPATHLEN + 4, M_TEMP, M_WAITOK|M_CANFAIL);
1626 		if (path == NULL) {
1627 			procfs_proc_unlock(pown);
1628 			return (ENOMEM);
1629 		}
1630 		bp = path + MAXPATHLEN;
1631 		*--bp = '\0';
1632 		procfs_dir(PROCFS_TYPE(pfs->pfs_fileno), curlwp, pown,
1633 		    &bp, path, MAXPATHLEN);
1634 		procfs_proc_unlock(pown);
1635 		len = strlen(bp);
1636 	} else {
1637 		file_t *fp;
1638 		struct vnode *vxp, *vp;
1639 
1640 		if ((error = procfs_proc_lock(pfs->pfs_pid, &pown, ESRCH)) != 0)
1641 			return error;
1642 
1643 		fp = fd_getfile2(pown, pfs->pfs_fd);
1644 		if (fp == NULL) {
1645 			procfs_proc_unlock(pown);
1646 			return EBADF;
1647 		}
1648 
1649 		switch (fp->f_type) {
1650 		case DTYPE_VNODE:
1651 			vxp = fp->f_vnode;
1652 			if (vxp->v_type != VDIR) {
1653 				error = EINVAL;
1654 				break;
1655 			}
1656 			if ((path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK))
1657 			    == NULL) {
1658 				error = ENOMEM;
1659 				break;
1660 			}
1661 			bp = path + MAXPATHLEN;
1662 			*--bp = '\0';
1663 
1664 			/*
1665 			 * XXX: kludge to avoid locking against ourselves
1666 			 * in getcwd()
1667 			 */
1668 			if (vxp->v_tag == VT_PROCFS) {
1669 				*--bp = '/';
1670 			} else {
1671 				rw_enter(&curproc->p_cwdi->cwdi_lock,
1672 				    RW_READER);
1673 				vp = curproc->p_cwdi->cwdi_rdir;
1674 				if (vp == NULL)
1675 					vp = rootvnode;
1676 				error = getcwd_common(vxp, vp, &bp, path,
1677 				    MAXPATHLEN / 2, 0, curlwp);
1678 				rw_exit(&curproc->p_cwdi->cwdi_lock);
1679 			}
1680 			if (error)
1681 				break;
1682 			len = strlen(bp);
1683 			break;
1684 
1685 		case DTYPE_MISC:
1686 			len = snprintf(bf, sizeof(bf), "%s", "[misc]");
1687 			break;
1688 
1689 		case DTYPE_KQUEUE:
1690 			len = snprintf(bf, sizeof(bf), "%s", "[kqueue]");
1691 			break;
1692 
1693 		case DTYPE_SEM:
1694 			len = snprintf(bf, sizeof(bf), "%s", "[ksem]");
1695 			break;
1696 
1697 		default:
1698 			error = EINVAL;
1699 			break;
1700 		}
1701 		closef(fp);
1702 		procfs_proc_unlock(pown);
1703 	}
1704 
1705 	if (error == 0)
1706 		error = uiomove(bp, len, ap->a_uio);
1707 	if (path)
1708 		free(path, M_TEMP);
1709 	return error;
1710 }
1711 
1712 /*
1713  * convert decimal ascii to int
1714  */
1715 static int
1716 atoi(const char *b, size_t len)
1717 {
1718 	int p = 0;
1719 
1720 	while (len--) {
1721 		char c = *b++;
1722 		if (c < '0' || c > '9')
1723 			return -1;
1724 		p = 10 * p + (c - '0');
1725 	}
1726 
1727 	return p;
1728 }
1729