xref: /netbsd-src/sys/miscfs/procfs/procfs_vnops.c (revision b7b7574d3bf8eeb51a1fa3977b59142ec6434a55)
1 /*	$NetBSD: procfs_vnops.c,v 1.189 2014/02/07 15:29:22 hannken 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 1.189 2014/02/07 15:29:22 hannken 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_fcntl_desc, procfs_fcntl },		/* fcntl */
263 	{ &vop_ioctl_desc, procfs_ioctl },		/* ioctl */
264 	{ &vop_poll_desc, procfs_poll },		/* poll */
265 	{ &vop_revoke_desc, procfs_revoke },		/* revoke */
266 	{ &vop_fsync_desc, procfs_fsync },		/* fsync */
267 	{ &vop_seek_desc, procfs_seek },		/* seek */
268 	{ &vop_remove_desc, procfs_remove },		/* remove */
269 	{ &vop_link_desc, procfs_link },		/* link */
270 	{ &vop_rename_desc, procfs_rename },		/* rename */
271 	{ &vop_mkdir_desc, procfs_mkdir },		/* mkdir */
272 	{ &vop_rmdir_desc, procfs_rmdir },		/* rmdir */
273 	{ &vop_symlink_desc, procfs_symlink },		/* symlink */
274 	{ &vop_readdir_desc, procfs_readdir },		/* readdir */
275 	{ &vop_readlink_desc, procfs_readlink },	/* readlink */
276 	{ &vop_abortop_desc, procfs_abortop },		/* abortop */
277 	{ &vop_inactive_desc, procfs_inactive },	/* inactive */
278 	{ &vop_reclaim_desc, procfs_reclaim },		/* reclaim */
279 	{ &vop_lock_desc, procfs_lock },		/* lock */
280 	{ &vop_unlock_desc, procfs_unlock },		/* unlock */
281 	{ &vop_bmap_desc, procfs_bmap },		/* bmap */
282 	{ &vop_strategy_desc, procfs_strategy },	/* strategy */
283 	{ &vop_print_desc, procfs_print },		/* print */
284 	{ &vop_islocked_desc, procfs_islocked },	/* islocked */
285 	{ &vop_pathconf_desc, procfs_pathconf },	/* pathconf */
286 	{ &vop_advlock_desc, procfs_advlock },		/* advlock */
287 	{ &vop_putpages_desc, procfs_putpages },	/* putpages */
288 	{ NULL, NULL }
289 };
290 const struct vnodeopv_desc procfs_vnodeop_opv_desc =
291 	{ &procfs_vnodeop_p, procfs_vnodeop_entries };
292 /*
293  * set things up for doing i/o on
294  * the pfsnode (vp).  (vp) is locked
295  * on entry, and should be left locked
296  * on exit.
297  *
298  * for procfs we don't need to do anything
299  * in particular for i/o.  all that is done
300  * is to support exclusive open on process
301  * memory images.
302  */
303 int
304 procfs_open(void *v)
305 {
306 	struct vop_open_args /* {
307 		struct vnode *a_vp;
308 		int  a_mode;
309 		kauth_cred_t a_cred;
310 	} */ *ap = v;
311 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
312 	struct lwp *l1;
313 	struct proc *p2;
314 	int error;
315 
316 	if ((error = procfs_proc_lock(pfs->pfs_pid, &p2, ENOENT)) != 0)
317 		return error;
318 
319 	l1 = curlwp;				/* tracer */
320 
321 #define	M2K(m)	(((m) & FREAD) && ((m) & FWRITE) ? \
322 		 KAUTH_REQ_PROCESS_PROCFS_RW : \
323 		 (m) & FWRITE ? KAUTH_REQ_PROCESS_PROCFS_WRITE : \
324 		 KAUTH_REQ_PROCESS_PROCFS_READ)
325 
326 	mutex_enter(p2->p_lock);
327 	error = kauth_authorize_process(l1->l_cred, KAUTH_PROCESS_PROCFS,
328 	    p2, pfs, KAUTH_ARG(M2K(ap->a_mode)), NULL);
329 	mutex_exit(p2->p_lock);
330 	if (error) {
331 		procfs_proc_unlock(p2);
332 		return (error);
333 	}
334 
335 #undef M2K
336 
337 	switch (pfs->pfs_type) {
338 	case PFSmem:
339 		if (((pfs->pfs_flags & FWRITE) && (ap->a_mode & O_EXCL)) ||
340 		    ((pfs->pfs_flags & O_EXCL) && (ap->a_mode & FWRITE))) {
341 			error = EBUSY;
342 			break;
343 		}
344 
345 		if (!proc_isunder(p2, l1)) {
346 			error = EPERM;
347 			break;
348 		}
349 
350 		if (ap->a_mode & FWRITE)
351 			pfs->pfs_flags = ap->a_mode & (FWRITE|O_EXCL);
352 
353 		break;
354 
355 	case PFSregs:
356 	case PFSfpregs:
357 		if (!proc_isunder(p2, l1)) {
358 			error = EPERM;
359 			break;
360 		}
361 		break;
362 
363 	default:
364 		break;
365 	}
366 
367 	procfs_proc_unlock(p2);
368 	return (error);
369 }
370 
371 /*
372  * close the pfsnode (vp) after doing i/o.
373  * (vp) is not locked on entry or exit.
374  *
375  * nothing to do for procfs other than undo
376  * any exclusive open flag (see _open above).
377  */
378 int
379 procfs_close(void *v)
380 {
381 	struct vop_close_args /* {
382 		struct vnode *a_vp;
383 		int  a_fflag;
384 		kauth_cred_t a_cred;
385 	} */ *ap = v;
386 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
387 
388 	switch (pfs->pfs_type) {
389 	case PFSmem:
390 		if ((ap->a_fflag & FWRITE) && (pfs->pfs_flags & O_EXCL))
391 			pfs->pfs_flags &= ~(FWRITE|O_EXCL);
392 		break;
393 
394 	default:
395 		break;
396 	}
397 
398 	return (0);
399 }
400 
401 /*
402  * _inactive is called when the pfsnode
403  * is vrele'd and the reference count goes
404  * to zero.  (vp) will be on the vnode free
405  * list, so to get it back vget() must be
406  * used.
407  *
408  * (vp) is locked on entry, but must be unlocked on exit.
409  */
410 int
411 procfs_inactive(void *v)
412 {
413 	struct vop_inactive_args /* {
414 		struct vnode *a_vp;
415 		bool *a_recycle;
416 	} */ *ap = v;
417 	struct vnode *vp = ap->a_vp;
418 	struct pfsnode *pfs = VTOPFS(vp);
419 
420 	mutex_enter(proc_lock);
421 	*ap->a_recycle = (proc_find(pfs->pfs_pid) == NULL);
422 	mutex_exit(proc_lock);
423 
424 	VOP_UNLOCK(vp);
425 
426 	return (0);
427 }
428 
429 /*
430  * _reclaim is called when getnewvnode()
431  * wants to make use of an entry on the vnode
432  * free list.  at this time the filesystem needs
433  * to free any private data and remove the node
434  * from any private lists.
435  */
436 int
437 procfs_reclaim(void *v)
438 {
439 	struct vop_reclaim_args /* {
440 		struct vnode *a_vp;
441 	} */ *ap = v;
442 
443 	return (procfs_freevp(ap->a_vp));
444 }
445 
446 /*
447  * Return POSIX pathconf information applicable to special devices.
448  */
449 int
450 procfs_pathconf(void *v)
451 {
452 	struct vop_pathconf_args /* {
453 		struct vnode *a_vp;
454 		int a_name;
455 		register_t *a_retval;
456 	} */ *ap = v;
457 
458 	switch (ap->a_name) {
459 	case _PC_LINK_MAX:
460 		*ap->a_retval = LINK_MAX;
461 		return (0);
462 	case _PC_MAX_CANON:
463 		*ap->a_retval = MAX_CANON;
464 		return (0);
465 	case _PC_MAX_INPUT:
466 		*ap->a_retval = MAX_INPUT;
467 		return (0);
468 	case _PC_PIPE_BUF:
469 		*ap->a_retval = PIPE_BUF;
470 		return (0);
471 	case _PC_CHOWN_RESTRICTED:
472 		*ap->a_retval = 1;
473 		return (0);
474 	case _PC_VDISABLE:
475 		*ap->a_retval = _POSIX_VDISABLE;
476 		return (0);
477 	case _PC_SYNC_IO:
478 		*ap->a_retval = 1;
479 		return (0);
480 	default:
481 		return (EINVAL);
482 	}
483 	/* NOTREACHED */
484 }
485 
486 /*
487  * _print is used for debugging.
488  * just print a readable description
489  * of (vp).
490  */
491 int
492 procfs_print(void *v)
493 {
494 	struct vop_print_args /* {
495 		struct vnode *a_vp;
496 	} */ *ap = v;
497 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
498 
499 	printf("tag VT_PROCFS, type %d, pid %d, mode %x, flags %lx\n",
500 	    pfs->pfs_type, pfs->pfs_pid, pfs->pfs_mode, pfs->pfs_flags);
501 	return 0;
502 }
503 
504 int
505 procfs_link(void *v)
506 {
507 	struct vop_link_args /* {
508 		struct vnode *a_dvp;
509 		struct vnode *a_vp;
510 		struct componentname *a_cnp;
511 	} */ *ap = v;
512 
513 	VOP_ABORTOP(ap->a_dvp, ap->a_cnp);
514 	vput(ap->a_dvp);
515 	return (EROFS);
516 }
517 
518 int
519 procfs_symlink(void *v)
520 {
521 	struct vop_symlink_v3_args /* {
522 		struct vnode *a_dvp;
523 		struct vnode **a_vpp;
524 		struct componentname *a_cnp;
525 		struct vattr *a_vap;
526 		char *a_target;
527 	} */ *ap = v;
528 
529 	VOP_ABORTOP(ap->a_dvp, ap->a_cnp);
530 	return (EROFS);
531 }
532 
533 /*
534  * Works out the path to (and vnode of) the target process's current
535  * working directory or chroot.  If the caller is in a chroot and
536  * can't "reach" the target's cwd or root (or some other error
537  * occurs), a "/" is returned for the path and a NULL pointer is
538  * returned for the vnode.
539  */
540 static void
541 procfs_dir(pfstype t, struct lwp *caller, struct proc *target, char **bpp,
542     char *path, size_t len)
543 {
544 	struct cwdinfo *cwdi;
545 	struct vnode *vp, *rvp;
546 	char *bp;
547 
548 	cwdi = caller->l_proc->p_cwdi;
549 	rw_enter(&cwdi->cwdi_lock, RW_READER);
550 
551 	rvp = cwdi->cwdi_rdir;
552 	bp = bpp ? *bpp : NULL;
553 
554 	switch (t) {
555 	case PFScwd:
556 		vp = target->p_cwdi->cwdi_cdir;
557 		break;
558 	case PFSchroot:
559 		vp = target->p_cwdi->cwdi_rdir;
560 		break;
561 	case PFSexe:
562 		vp = target->p_textvp;
563 		break;
564 	default:
565 		rw_exit(&cwdi->cwdi_lock);
566 		return;
567 	}
568 
569 	/*
570 	 * XXX: this horrible kludge avoids locking panics when
571 	 * attempting to lookup links that point to within procfs
572 	 */
573 	if (vp != NULL && vp->v_tag == VT_PROCFS) {
574 		if (bpp) {
575 			*--bp = '/';
576 			*bpp = bp;
577 		}
578 		rw_exit(&cwdi->cwdi_lock);
579 		return;
580 	}
581 
582 	if (rvp == NULL)
583 		rvp = rootvnode;
584 	if (vp == NULL || getcwd_common(vp, rvp, bp ? &bp : NULL, path,
585 	    len / 2, 0, caller) != 0) {
586 		vp = NULL;
587 		if (bpp) {
588 /*
589 			if (t == PFSexe) {
590 				snprintf(path, len, "%s/%d/file"
591 				    mp->mnt_stat.f_mntonname, pfs->pfs_pid);
592 			} else */ {
593 				bp = *bpp;
594 				*--bp = '/';
595 			}
596 		}
597 	}
598 
599 	if (bpp)
600 		*bpp = bp;
601 
602 	rw_exit(&cwdi->cwdi_lock);
603 }
604 
605 /*
606  * Invent attributes for pfsnode (vp) and store
607  * them in (vap).
608  * Directories lengths are returned as zero since
609  * any real length would require the genuine size
610  * to be computed, and nothing cares anyway.
611  *
612  * this is relatively minimal for procfs.
613  */
614 int
615 procfs_getattr(void *v)
616 {
617 	struct vop_getattr_args /* {
618 		struct vnode *a_vp;
619 		struct vattr *a_vap;
620 		kauth_cred_t a_cred;
621 	} */ *ap = v;
622 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
623 	struct vattr *vap = ap->a_vap;
624 	struct proc *procp;
625 	char *path;
626 	int error;
627 
628 	/* first check the process still exists */
629 	switch (pfs->pfs_type) {
630 	case PFSroot:
631 	case PFScurproc:
632 	case PFSself:
633 		procp = NULL;
634 		break;
635 
636 	default:
637 		error = procfs_proc_lock(pfs->pfs_pid, &procp, ENOENT);
638 		if (error != 0)
639 			return (error);
640 		break;
641 	}
642 
643 	switch (pfs->pfs_type) {
644 	case PFStask:
645 		if (pfs->pfs_fd == -1) {
646 			path = NULL;
647 			break;
648 		}
649 		/*FALLTHROUGH*/
650 	case PFScwd:
651 	case PFSchroot:
652 	case PFSexe:
653 		path = malloc(MAXPATHLEN + 4, M_TEMP, M_WAITOK|M_CANFAIL);
654 		if (path == NULL && procp != NULL) {
655 			procfs_proc_unlock(procp);
656 			return (ENOMEM);
657 		}
658 		break;
659 
660 	default:
661 		path = NULL;
662 		break;
663 	}
664 
665 	if (procp != NULL) {
666 		mutex_enter(procp->p_lock);
667 		error = kauth_authorize_process(kauth_cred_get(),
668 		    KAUTH_PROCESS_CANSEE, procp,
669 		    KAUTH_ARG(KAUTH_REQ_PROCESS_CANSEE_ENTRY), NULL, NULL);
670 		mutex_exit(procp->p_lock);
671 		if (error != 0) {
672 		    	procfs_proc_unlock(procp);
673 		    	if (path != NULL)
674 		    		free(path, M_TEMP);
675 			return (ENOENT);
676 		}
677 	}
678 
679 	error = 0;
680 
681 	/* start by zeroing out the attributes */
682 	vattr_null(vap);
683 
684 	/* next do all the common fields */
685 	vap->va_type = ap->a_vp->v_type;
686 	vap->va_mode = pfs->pfs_mode;
687 	vap->va_fileid = pfs->pfs_fileno;
688 	vap->va_flags = 0;
689 	vap->va_blocksize = PAGE_SIZE;
690 
691 	/*
692 	 * Make all times be current TOD.
693 	 *
694 	 * It would be possible to get the process start
695 	 * time from the p_stats structure, but there's
696 	 * no "file creation" time stamp anyway, and the
697 	 * p_stats structure is not addressable if u. gets
698 	 * swapped out for that process.
699 	 */
700 	getnanotime(&vap->va_ctime);
701 	vap->va_atime = vap->va_mtime = vap->va_ctime;
702 	if (procp)
703 		TIMEVAL_TO_TIMESPEC(&procp->p_stats->p_start,
704 		    &vap->va_birthtime);
705 	else
706 		getnanotime(&vap->va_birthtime);
707 
708 	switch (pfs->pfs_type) {
709 	case PFSmem:
710 	case PFSregs:
711 	case PFSfpregs:
712 #if defined(__HAVE_PROCFS_MACHDEP) && defined(PROCFS_MACHDEP_PROTECT_CASES)
713 	PROCFS_MACHDEP_PROTECT_CASES
714 #endif
715 		/*
716 		 * If the process has exercised some setuid or setgid
717 		 * privilege, then rip away read/write permission so
718 		 * that only root can gain access.
719 		 */
720 		if (procp->p_flag & PK_SUGID)
721 			vap->va_mode &= ~(S_IRUSR|S_IWUSR);
722 		/* FALLTHROUGH */
723 	case PFSctl:
724 	case PFSstatus:
725 	case PFSstat:
726 	case PFSnote:
727 	case PFSnotepg:
728 	case PFSmap:
729 	case PFSmaps:
730 	case PFScmdline:
731 	case PFSemul:
732 	case PFSstatm:
733 		if (pfs->pfs_type == PFSmap || pfs->pfs_type == PFSmaps)
734 			vap->va_mode = S_IRUSR;
735 		vap->va_nlink = 1;
736 		vap->va_uid = kauth_cred_geteuid(procp->p_cred);
737 		vap->va_gid = kauth_cred_getegid(procp->p_cred);
738 		break;
739 	case PFSmeminfo:
740 	case PFSdevices:
741 	case PFScpuinfo:
742 	case PFSuptime:
743 	case PFSmounts:
744 	case PFScpustat:
745 	case PFSloadavg:
746 	case PFSversion:
747 		vap->va_nlink = 1;
748 		vap->va_uid = vap->va_gid = 0;
749 		break;
750 
751 	default:
752 		break;
753 	}
754 
755 	/*
756 	 * now do the object specific fields
757 	 *
758 	 * The size could be set from struct reg, but it's hardly
759 	 * worth the trouble, and it puts some (potentially) machine
760 	 * dependent data into this machine-independent code.  If it
761 	 * becomes important then this function should break out into
762 	 * a per-file stat function in the corresponding .c file.
763 	 */
764 
765 	switch (pfs->pfs_type) {
766 	case PFSroot:
767 		/*
768 		 * Set nlink to 1 to tell fts(3) we don't actually know.
769 		 */
770 		vap->va_nlink = 1;
771 		vap->va_uid = 0;
772 		vap->va_gid = 0;
773 		vap->va_bytes = vap->va_size = DEV_BSIZE;
774 		break;
775 
776 	case PFSself:
777 	case PFScurproc: {
778 		char bf[16];		/* should be enough */
779 		vap->va_nlink = 1;
780 		vap->va_uid = 0;
781 		vap->va_gid = 0;
782 		vap->va_bytes = vap->va_size =
783 		    snprintf(bf, sizeof(bf), "%ld", (long)curproc->p_pid);
784 		break;
785 	}
786 	case PFStask:
787 		if (pfs->pfs_fd != -1) {
788 			char bf[4];		/* should be enough */
789 			vap->va_nlink = 1;
790 			vap->va_uid = 0;
791 			vap->va_gid = 0;
792 			vap->va_bytes = vap->va_size =
793 			    snprintf(bf, sizeof(bf), "..");
794 			break;
795 		}
796 		/*FALLTHROUGH*/
797 	case PFSfd:
798 		if (pfs->pfs_fd != -1) {
799 			file_t *fp;
800 
801 			fp = fd_getfile2(procp, pfs->pfs_fd);
802 			if (fp == NULL) {
803 				error = EBADF;
804 				break;
805 			}
806 			vap->va_nlink = 1;
807 			vap->va_uid = kauth_cred_geteuid(fp->f_cred);
808 			vap->va_gid = kauth_cred_getegid(fp->f_cred);
809 			switch (fp->f_type) {
810 			case DTYPE_VNODE:
811 				vap->va_bytes = vap->va_size =
812 				    ((struct vnode *)fp->f_data)->v_size;
813 				break;
814 			default:
815 				vap->va_bytes = vap->va_size = 0;
816 				break;
817 			}
818 			closef(fp);
819 			break;
820 		}
821 		/*FALLTHROUGH*/
822 	case PFSproc:
823 		vap->va_nlink = 2;
824 		vap->va_uid = kauth_cred_geteuid(procp->p_cred);
825 		vap->va_gid = kauth_cred_getegid(procp->p_cred);
826 		vap->va_bytes = vap->va_size = DEV_BSIZE;
827 		break;
828 
829 	case PFSfile:
830 		error = EOPNOTSUPP;
831 		break;
832 
833 	case PFSmem:
834 		vap->va_bytes = vap->va_size =
835 			ctob(procp->p_vmspace->vm_tsize +
836 				    procp->p_vmspace->vm_dsize +
837 				    procp->p_vmspace->vm_ssize);
838 		break;
839 
840 #if defined(PT_GETREGS) || defined(PT_SETREGS)
841 	case PFSregs:
842 		vap->va_bytes = vap->va_size = sizeof(struct reg);
843 		break;
844 #endif
845 
846 #if defined(PT_GETFPREGS) || defined(PT_SETFPREGS)
847 	case PFSfpregs:
848 		vap->va_bytes = vap->va_size = sizeof(struct fpreg);
849 		break;
850 #endif
851 
852 	case PFSctl:
853 	case PFSstatus:
854 	case PFSstat:
855 	case PFSnote:
856 	case PFSnotepg:
857 	case PFScmdline:
858 	case PFSmeminfo:
859 	case PFSdevices:
860 	case PFScpuinfo:
861 	case PFSuptime:
862 	case PFSmounts:
863 	case PFScpustat:
864 	case PFSloadavg:
865 	case PFSstatm:
866 	case PFSversion:
867 		vap->va_bytes = vap->va_size = 0;
868 		break;
869 	case PFSmap:
870 	case PFSmaps:
871 		/*
872 		 * Advise a larger blocksize for the map files, so that
873 		 * they may be read in one pass.
874 		 */
875 		vap->va_blocksize = 4 * PAGE_SIZE;
876 		vap->va_bytes = vap->va_size = 0;
877 		break;
878 
879 	case PFScwd:
880 	case PFSchroot:
881 	case PFSexe: {
882 		char *bp;
883 
884 		vap->va_nlink = 1;
885 		vap->va_uid = 0;
886 		vap->va_gid = 0;
887 		bp = path + MAXPATHLEN;
888 		*--bp = '\0';
889 		procfs_dir(pfs->pfs_type, curlwp, procp, &bp, path,
890 		     MAXPATHLEN);
891 		vap->va_bytes = vap->va_size = strlen(bp);
892 		break;
893 	}
894 
895 	case PFSemul:
896 		vap->va_bytes = vap->va_size = strlen(procp->p_emul->e_name);
897 		break;
898 
899 #ifdef __HAVE_PROCFS_MACHDEP
900 	PROCFS_MACHDEP_NODETYPE_CASES
901 		error = procfs_machdep_getattr(ap->a_vp, vap, procp);
902 		break;
903 #endif
904 
905 	default:
906 		panic("procfs_getattr");
907 	}
908 
909 	if (procp != NULL)
910 		procfs_proc_unlock(procp);
911 	if (path != NULL)
912 		free(path, M_TEMP);
913 
914 	return (error);
915 }
916 
917 /*ARGSUSED*/
918 int
919 procfs_setattr(void *v)
920 {
921 	/*
922 	 * just fake out attribute setting
923 	 * it's not good to generate an error
924 	 * return, otherwise things like creat()
925 	 * will fail when they try to set the
926 	 * file length to 0.  worse, this means
927 	 * that echo $note > /proc/$pid/note will fail.
928 	 */
929 
930 	return (0);
931 }
932 
933 /*
934  * implement access checking.
935  *
936  * actually, the check for super-user is slightly
937  * broken since it will allow read access to write-only
938  * objects.  this doesn't cause any particular trouble
939  * but does mean that the i/o entry points need to check
940  * that the operation really does make sense.
941  */
942 int
943 procfs_access(void *v)
944 {
945 	struct vop_access_args /* {
946 		struct vnode *a_vp;
947 		int a_mode;
948 		kauth_cred_t a_cred;
949 	} */ *ap = v;
950 	struct vattr va;
951 	int error;
952 
953 	if ((error = VOP_GETATTR(ap->a_vp, &va, ap->a_cred)) != 0)
954 		return (error);
955 
956 	return kauth_authorize_vnode(ap->a_cred,
957 	    KAUTH_ACCESS_ACTION(ap->a_mode, ap->a_vp->v_type, va.va_mode),
958 	    ap->a_vp, NULL, genfs_can_access(va.va_type, va.va_mode,
959 	    va.va_uid, va.va_gid, ap->a_mode, ap->a_cred));
960 }
961 
962 /*
963  * lookup.  this is incredibly complicated in the
964  * general case, however for most pseudo-filesystems
965  * very little needs to be done.
966  *
967  * Locking isn't hard here, just poorly documented.
968  *
969  * If we're looking up ".", just vref the parent & return it.
970  *
971  * If we're looking up "..", unlock the parent, and lock "..". If everything
972  * went ok, and we're on the last component and the caller requested the
973  * parent locked, try to re-lock the parent. We do this to prevent lock
974  * races.
975  *
976  * For anything else, get the needed node. Then unlock the parent if not
977  * the last component or not LOCKPARENT (i.e. if we wouldn't re-lock the
978  * parent in the .. case).
979  *
980  * We try to exit with the parent locked in error cases.
981  */
982 int
983 procfs_lookup(void *v)
984 {
985 	struct vop_lookup_v2_args /* {
986 		struct vnode * a_dvp;
987 		struct vnode ** a_vpp;
988 		struct componentname * a_cnp;
989 	} */ *ap = v;
990 	struct componentname *cnp = ap->a_cnp;
991 	struct vnode **vpp = ap->a_vpp;
992 	struct vnode *dvp = ap->a_dvp;
993 	const char *pname = cnp->cn_nameptr;
994 	const struct proc_target *pt = NULL;
995 	struct vnode *fvp;
996 	pid_t pid, vnpid;
997 	struct pfsnode *pfs;
998 	struct proc *p = NULL;
999 	struct lwp *plwp;
1000 	int i, error;
1001 	pfstype type;
1002 
1003 	*vpp = NULL;
1004 
1005 	if (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)
1006 		return (EROFS);
1007 
1008 	if (cnp->cn_namelen == 1 && *pname == '.') {
1009 		*vpp = dvp;
1010 		vref(dvp);
1011 		return (0);
1012 	}
1013 
1014 	pfs = VTOPFS(dvp);
1015 	switch (pfs->pfs_type) {
1016 	case PFSroot:
1017 		/*
1018 		 * Shouldn't get here with .. in the root node.
1019 		 */
1020 		if (cnp->cn_flags & ISDOTDOT)
1021 			return (EIO);
1022 
1023 		for (i = 0; i < nproc_root_targets; i++) {
1024 			pt = &proc_root_targets[i];
1025 			/*
1026 			 * check for node match.  proc is always NULL here,
1027 			 * so call pt_valid with constant NULL lwp.
1028 			 */
1029 			if (cnp->cn_namelen == pt->pt_namlen &&
1030 			    memcmp(pt->pt_name, pname, cnp->cn_namelen) == 0 &&
1031 			    (pt->pt_valid == NULL ||
1032 			     (*pt->pt_valid)(NULL, dvp->v_mount)))
1033 				break;
1034 		}
1035 
1036 		if (i != nproc_root_targets) {
1037 			error = procfs_allocvp(dvp->v_mount, vpp, 0,
1038 			    pt->pt_pfstype, -1, NULL);
1039 			return (error);
1040 		}
1041 
1042 		if (CNEQ(cnp, "curproc", 7)) {
1043 			pid = curproc->p_pid;
1044 			vnpid = 0;
1045 			type = PFScurproc;
1046 		} else if (CNEQ(cnp, "self", 4)) {
1047 			pid = curproc->p_pid;
1048 			vnpid = 0;
1049 			type = PFSself;
1050 		} else {
1051 			pid = (pid_t)atoi(pname, cnp->cn_namelen);
1052 			vnpid = pid;
1053 			type = PFSproc;
1054 		}
1055 
1056 		if (procfs_proc_lock(pid, &p, ESRCH) != 0)
1057 			break;
1058 		error = procfs_allocvp(dvp->v_mount, vpp, vnpid, type, -1, p);
1059 		procfs_proc_unlock(p);
1060 		return (error);
1061 
1062 	case PFSproc:
1063 		if (cnp->cn_flags & ISDOTDOT) {
1064 			error = procfs_allocvp(dvp->v_mount, vpp, 0, PFSroot,
1065 			    -1, NULL);
1066 			return (error);
1067 		}
1068 
1069 		if (procfs_proc_lock(pfs->pfs_pid, &p, ESRCH) != 0)
1070 			break;
1071 
1072 		mutex_enter(p->p_lock);
1073 		LIST_FOREACH(plwp, &p->p_lwps, l_sibling) {
1074 			if (plwp->l_stat != LSZOMB)
1075 				break;
1076 		}
1077 		/* Process is exiting if no-LWPS or all LWPs are LSZOMB */
1078 		if (plwp == NULL) {
1079 			mutex_exit(p->p_lock);
1080 			procfs_proc_unlock(p);
1081 			return ESRCH;
1082 		}
1083 
1084 		lwp_addref(plwp);
1085 		mutex_exit(p->p_lock);
1086 
1087 		for (pt = proc_targets, i = 0; i < nproc_targets; pt++, i++) {
1088 			int found;
1089 
1090 			found = cnp->cn_namelen == pt->pt_namlen &&
1091 			    memcmp(pt->pt_name, pname, cnp->cn_namelen) == 0 &&
1092 			    (pt->pt_valid == NULL
1093 			      || (*pt->pt_valid)(plwp, dvp->v_mount));
1094 			if (found)
1095 				break;
1096 		}
1097 		lwp_delref(plwp);
1098 
1099 		if (i == nproc_targets) {
1100 			procfs_proc_unlock(p);
1101 			break;
1102 		}
1103 		if (pt->pt_pfstype == PFSfile) {
1104 			fvp = p->p_textvp;
1105 			/* We already checked that it exists. */
1106 			vref(fvp);
1107 			procfs_proc_unlock(p);
1108 			*vpp = fvp;
1109 			return (0);
1110 		}
1111 
1112 		error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
1113 		    pt->pt_pfstype, -1, p);
1114 		procfs_proc_unlock(p);
1115 		return (error);
1116 
1117 	case PFSfd: {
1118 		int fd;
1119 		file_t *fp;
1120 
1121 		if ((error = procfs_proc_lock(pfs->pfs_pid, &p, ENOENT)) != 0)
1122 			return error;
1123 
1124 		if (cnp->cn_flags & ISDOTDOT) {
1125 			error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
1126 			    PFSproc, -1, p);
1127 			procfs_proc_unlock(p);
1128 			return (error);
1129 		}
1130 		fd = atoi(pname, cnp->cn_namelen);
1131 
1132 		fp = fd_getfile2(p, fd);
1133 		if (fp == NULL) {
1134 			procfs_proc_unlock(p);
1135 			return ENOENT;
1136 		}
1137 		fvp = fp->f_data;
1138 
1139 		/* Don't show directories */
1140 		if (fp->f_type == DTYPE_VNODE && fvp->v_type != VDIR) {
1141 			vref(fvp);
1142 			closef(fp);
1143 			procfs_proc_unlock(p);
1144 			*vpp = fvp;
1145 			return 0;
1146 		}
1147 
1148 		closef(fp);
1149 		error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
1150 		    PFSfd, fd, p);
1151 		procfs_proc_unlock(p);
1152 		return error;
1153 	}
1154 	case PFStask: {
1155 		int xpid;
1156 
1157 		if ((error = procfs_proc_lock(pfs->pfs_pid, &p, ENOENT)) != 0)
1158 			return error;
1159 
1160 		if (cnp->cn_flags & ISDOTDOT) {
1161 			error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
1162 			    PFSproc, -1, p);
1163 			procfs_proc_unlock(p);
1164 			return (error);
1165 		}
1166 		xpid = atoi(pname, cnp->cn_namelen);
1167 
1168 		if (xpid != pfs->pfs_pid) {
1169 			procfs_proc_unlock(p);
1170 			return ENOENT;
1171 		}
1172 		error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
1173 		    PFStask, 0, p);
1174 		procfs_proc_unlock(p);
1175 		return error;
1176 	}
1177 	default:
1178 		return (ENOTDIR);
1179 	}
1180 
1181 	return (cnp->cn_nameiop == LOOKUP ? ENOENT : EROFS);
1182 }
1183 
1184 int
1185 procfs_validfile(struct lwp *l, struct mount *mp)
1186 {
1187 	return l != NULL && l->l_proc != NULL && l->l_proc->p_textvp != NULL;
1188 }
1189 
1190 static int
1191 procfs_validfile_linux(struct lwp *l, struct mount *mp)
1192 {
1193 	int flags;
1194 
1195 	flags = VFSTOPROC(mp)->pmnt_flags;
1196 	return (flags & PROCFSMNT_LINUXCOMPAT) &&
1197 	    (l == NULL || l->l_proc == NULL || procfs_validfile(l, mp));
1198 }
1199 
1200 struct procfs_root_readdir_ctx {
1201 	struct uio *uiop;
1202 	off_t *cookies;
1203 	int ncookies;
1204 	off_t off;
1205 	off_t startoff;
1206 	int error;
1207 };
1208 
1209 static int
1210 procfs_root_readdir_callback(struct proc *p, void *arg)
1211 {
1212 	struct procfs_root_readdir_ctx *ctxp = arg;
1213 	struct dirent d;
1214 	struct uio *uiop;
1215 	int error;
1216 
1217 	uiop = ctxp->uiop;
1218 	if (uiop->uio_resid < UIO_MX)
1219 		return -1; /* no space */
1220 
1221 	if (ctxp->off < ctxp->startoff) {
1222 		ctxp->off++;
1223 		return 0;
1224 	}
1225 
1226 	if (kauth_authorize_process(kauth_cred_get(),
1227 	    KAUTH_PROCESS_CANSEE, p,
1228 	    KAUTH_ARG(KAUTH_REQ_PROCESS_CANSEE_ENTRY), NULL, NULL) != 0)
1229 		return 0;
1230 
1231 	memset(&d, 0, UIO_MX);
1232 	d.d_reclen = UIO_MX;
1233 	d.d_fileno = PROCFS_FILENO(p->p_pid, PFSproc, -1);
1234 	d.d_namlen = snprintf(d.d_name,
1235 	    UIO_MX - offsetof(struct dirent, d_name), "%ld", (long)p->p_pid);
1236 	d.d_type = DT_DIR;
1237 
1238 	mutex_exit(proc_lock);
1239 	error = uiomove(&d, UIO_MX, uiop);
1240 	mutex_enter(proc_lock);
1241 	if (error) {
1242 		ctxp->error = error;
1243 		return -1;
1244 	}
1245 
1246 	ctxp->ncookies++;
1247 	if (ctxp->cookies)
1248 		*(ctxp->cookies)++ = ctxp->off + 1;
1249 	ctxp->off++;
1250 
1251 	return 0;
1252 }
1253 
1254 /*
1255  * readdir returns directory entries from pfsnode (vp).
1256  *
1257  * the strategy here with procfs is to generate a single
1258  * directory entry at a time (struct dirent) and then
1259  * copy that out to userland using uiomove.  a more efficent
1260  * though more complex implementation, would try to minimize
1261  * the number of calls to uiomove().  for procfs, this is
1262  * hardly worth the added code complexity.
1263  *
1264  * this should just be done through read()
1265  */
1266 int
1267 procfs_readdir(void *v)
1268 {
1269 	struct vop_readdir_args /* {
1270 		struct vnode *a_vp;
1271 		struct uio *a_uio;
1272 		kauth_cred_t a_cred;
1273 		int *a_eofflag;
1274 		off_t **a_cookies;
1275 		int *a_ncookies;
1276 	} */ *ap = v;
1277 	struct uio *uio = ap->a_uio;
1278 	struct dirent d;
1279 	struct pfsnode *pfs;
1280 	off_t i;
1281 	int error;
1282 	off_t *cookies = NULL;
1283 	int ncookies;
1284 	struct vnode *vp;
1285 	const struct proc_target *pt;
1286 	struct procfs_root_readdir_ctx ctx;
1287 	struct lwp *l;
1288 	int nfd;
1289 
1290 	vp = ap->a_vp;
1291 	pfs = VTOPFS(vp);
1292 
1293 	if (uio->uio_resid < UIO_MX)
1294 		return (EINVAL);
1295 	if (uio->uio_offset < 0)
1296 		return (EINVAL);
1297 
1298 	error = 0;
1299 	i = uio->uio_offset;
1300 	memset(&d, 0, UIO_MX);
1301 	d.d_reclen = UIO_MX;
1302 	ncookies = uio->uio_resid / UIO_MX;
1303 
1304 	switch (pfs->pfs_type) {
1305 	/*
1306 	 * this is for the process-specific sub-directories.
1307 	 * all that is needed to is copy out all the entries
1308 	 * from the procent[] table (top of this file).
1309 	 */
1310 	case PFSproc: {
1311 		struct proc *p;
1312 
1313 		if (i >= nproc_targets)
1314 			return 0;
1315 
1316 		if (procfs_proc_lock(pfs->pfs_pid, &p, ESRCH) != 0)
1317 			break;
1318 
1319 		if (ap->a_ncookies) {
1320 			ncookies = min(ncookies, (nproc_targets - i));
1321 			cookies = malloc(ncookies * sizeof (off_t),
1322 			    M_TEMP, M_WAITOK);
1323 			*ap->a_cookies = cookies;
1324 		}
1325 
1326 		for (pt = &proc_targets[i];
1327 		     uio->uio_resid >= UIO_MX && i < nproc_targets; pt++, i++) {
1328 			if (pt->pt_valid) {
1329 				/* XXXSMP LWP can disappear */
1330 				mutex_enter(p->p_lock);
1331 				l = LIST_FIRST(&p->p_lwps);
1332 				KASSERT(l != NULL);
1333 				mutex_exit(p->p_lock);
1334 				if ((*pt->pt_valid)(l, vp->v_mount) == 0)
1335 					continue;
1336 			}
1337 
1338 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid,
1339 			    pt->pt_pfstype, -1);
1340 			d.d_namlen = pt->pt_namlen;
1341 			memcpy(d.d_name, pt->pt_name, pt->pt_namlen + 1);
1342 			d.d_type = pt->pt_type;
1343 
1344 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1345 				break;
1346 			if (cookies)
1347 				*cookies++ = i + 1;
1348 		}
1349 
1350 		procfs_proc_unlock(p);
1351 	    	break;
1352 	}
1353 	case PFSfd: {
1354 		struct proc *p;
1355 		file_t *fp;
1356 		int lim, nc = 0;
1357 
1358 		if ((error = procfs_proc_lock(pfs->pfs_pid, &p, ESRCH)) != 0)
1359 			return error;
1360 
1361 		/* XXX Should this be by file as well? */
1362 		if (kauth_authorize_process(kauth_cred_get(),
1363 		    KAUTH_PROCESS_CANSEE, p,
1364 		    KAUTH_ARG(KAUTH_REQ_PROCESS_CANSEE_OPENFILES), NULL,
1365 		    NULL) != 0) {
1366 		    	procfs_proc_unlock(p);
1367 			return ESRCH;
1368 		}
1369 
1370 		nfd = p->p_fd->fd_dt->dt_nfiles;
1371 
1372 		lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfiles);
1373 		if (i >= lim) {
1374 		    	procfs_proc_unlock(p);
1375 			return 0;
1376 		}
1377 
1378 		if (ap->a_ncookies) {
1379 			ncookies = min(ncookies, (nfd + 2 - i));
1380 			cookies = malloc(ncookies * sizeof (off_t),
1381 			    M_TEMP, M_WAITOK);
1382 			*ap->a_cookies = cookies;
1383 		}
1384 
1385 		for (; i < 2 && uio->uio_resid >= UIO_MX; i++) {
1386 			pt = &proc_targets[i];
1387 			d.d_namlen = pt->pt_namlen;
1388 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid,
1389 			    pt->pt_pfstype, -1);
1390 			(void)memcpy(d.d_name, pt->pt_name, pt->pt_namlen + 1);
1391 			d.d_type = pt->pt_type;
1392 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1393 				break;
1394 			if (cookies)
1395 				*cookies++ = i + 1;
1396 			nc++;
1397 		}
1398 		if (error) {
1399 			ncookies = nc;
1400 			break;
1401 		}
1402 		for (; uio->uio_resid >= UIO_MX && i < nfd; i++) {
1403 			/* check the descriptor exists */
1404 			if ((fp = fd_getfile2(p, i - 2)) == NULL)
1405 				continue;
1406 			closef(fp);
1407 
1408 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid, PFSfd, i - 2);
1409 			d.d_namlen = snprintf(d.d_name, sizeof(d.d_name),
1410 			    "%lld", (long long)(i - 2));
1411 			d.d_type = VREG;
1412 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1413 				break;
1414 			if (cookies)
1415 				*cookies++ = i + 1;
1416 			nc++;
1417 		}
1418 		ncookies = nc;
1419 		procfs_proc_unlock(p);
1420 		break;
1421 	}
1422 	case PFStask: {
1423 		struct proc *p;
1424 		int nc = 0;
1425 
1426 		if ((error = procfs_proc_lock(pfs->pfs_pid, &p, ESRCH)) != 0)
1427 			return error;
1428 
1429 		nfd = 3;	/* ., .., pid */
1430 
1431 		if (ap->a_ncookies) {
1432 			ncookies = min(ncookies, (nfd + 2 - i));
1433 			cookies = malloc(ncookies * sizeof (off_t),
1434 			    M_TEMP, M_WAITOK);
1435 			*ap->a_cookies = cookies;
1436 		}
1437 
1438 		for (; i < 2 && uio->uio_resid >= UIO_MX; i++) {
1439 			pt = &proc_targets[i];
1440 			d.d_namlen = pt->pt_namlen;
1441 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid,
1442 			    pt->pt_pfstype, -1);
1443 			(void)memcpy(d.d_name, pt->pt_name, pt->pt_namlen + 1);
1444 			d.d_type = pt->pt_type;
1445 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1446 				break;
1447 			if (cookies)
1448 				*cookies++ = i + 1;
1449 			nc++;
1450 		}
1451 		if (error) {
1452 			ncookies = nc;
1453 			break;
1454 		}
1455 		for (; uio->uio_resid >= UIO_MX && i < nfd; i++) {
1456 			/* check the descriptor exists */
1457 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid, PFStask,
1458 			    i - 2);
1459 			d.d_namlen = snprintf(d.d_name, sizeof(d.d_name),
1460 			    "%ld", (long)pfs->pfs_pid);
1461 			d.d_type = DT_LNK;
1462 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1463 				break;
1464 			if (cookies)
1465 				*cookies++ = i + 1;
1466 			nc++;
1467 		}
1468 		ncookies = nc;
1469 		procfs_proc_unlock(p);
1470 		break;
1471 	}
1472 
1473 	/*
1474 	 * this is for the root of the procfs filesystem
1475 	 * what is needed are special entries for "curproc"
1476 	 * and "self" followed by an entry for each process
1477 	 * on allproc.
1478 	 */
1479 
1480 	case PFSroot: {
1481 		int nc = 0;
1482 
1483 		if (ap->a_ncookies) {
1484 			/*
1485 			 * XXX Potentially allocating too much space here,
1486 			 * but I'm lazy. This loop needs some work.
1487 			 */
1488 			cookies = malloc(ncookies * sizeof (off_t),
1489 			    M_TEMP, M_WAITOK);
1490 			*ap->a_cookies = cookies;
1491 		}
1492 		error = 0;
1493 		/* 0 ... 3 are static entries. */
1494 		for (; i <= 3 && uio->uio_resid >= UIO_MX; i++) {
1495 			switch (i) {
1496 			case 0:		/* `.' */
1497 			case 1:		/* `..' */
1498 				d.d_fileno = PROCFS_FILENO(0, PFSroot, -1);
1499 				d.d_namlen = i + 1;
1500 				memcpy(d.d_name, "..", d.d_namlen);
1501 				d.d_name[i + 1] = '\0';
1502 				d.d_type = DT_DIR;
1503 				break;
1504 
1505 			case 2:
1506 				d.d_fileno = PROCFS_FILENO(0, PFScurproc, -1);
1507 				d.d_namlen = sizeof("curproc") - 1;
1508 				memcpy(d.d_name, "curproc", sizeof("curproc"));
1509 				d.d_type = DT_LNK;
1510 				break;
1511 
1512 			case 3:
1513 				d.d_fileno = PROCFS_FILENO(0, PFSself, -1);
1514 				d.d_namlen = sizeof("self") - 1;
1515 				memcpy(d.d_name, "self", sizeof("self"));
1516 				d.d_type = DT_LNK;
1517 				break;
1518 			}
1519 
1520 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1521 				break;
1522 			nc++;
1523 			if (cookies)
1524 				*cookies++ = i + 1;
1525 		}
1526 		/* 4 ... are process entries. */
1527 		ctx.uiop = uio;
1528 		ctx.error = 0;
1529 		ctx.off = 4;
1530 		ctx.startoff = i;
1531 		ctx.cookies = cookies;
1532 		ctx.ncookies = nc;
1533 		proclist_foreach_call(&allproc,
1534 		    procfs_root_readdir_callback, &ctx);
1535 		cookies = ctx.cookies;
1536 		nc = ctx.ncookies;
1537 		error = ctx.error;
1538 		if (error)
1539 			break;
1540 
1541 		/* misc entries. */
1542 		if (i < ctx.off)
1543 			i = ctx.off;
1544 		if (i >= ctx.off + nproc_root_targets)
1545 			break;
1546 		for (pt = &proc_root_targets[i - ctx.off];
1547 		    uio->uio_resid >= UIO_MX &&
1548 		    pt < &proc_root_targets[nproc_root_targets];
1549 		    pt++, i++) {
1550 			if (pt->pt_valid &&
1551 			    (*pt->pt_valid)(NULL, vp->v_mount) == 0)
1552 				continue;
1553 			d.d_fileno = PROCFS_FILENO(0, pt->pt_pfstype, -1);
1554 			d.d_namlen = pt->pt_namlen;
1555 			memcpy(d.d_name, pt->pt_name, pt->pt_namlen + 1);
1556 			d.d_type = pt->pt_type;
1557 
1558 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1559 				break;
1560 			nc++;
1561 			if (cookies)
1562 				*cookies++ = i + 1;
1563 		}
1564 
1565 		ncookies = nc;
1566 		break;
1567 	}
1568 
1569 	default:
1570 		error = ENOTDIR;
1571 		break;
1572 	}
1573 
1574 	if (ap->a_ncookies) {
1575 		if (error) {
1576 			if (cookies)
1577 				free(*ap->a_cookies, M_TEMP);
1578 			*ap->a_ncookies = 0;
1579 			*ap->a_cookies = NULL;
1580 		} else
1581 			*ap->a_ncookies = ncookies;
1582 	}
1583 	uio->uio_offset = i;
1584 	return (error);
1585 }
1586 
1587 /*
1588  * readlink reads the link of `curproc' and others
1589  */
1590 int
1591 procfs_readlink(void *v)
1592 {
1593 	struct vop_readlink_args *ap = v;
1594 	char bf[16];		/* should be enough */
1595 	char *bp = bf;
1596 	char *path = NULL;
1597 	int len = 0;
1598 	int error = 0;
1599 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
1600 	struct proc *pown;
1601 
1602 	if (pfs->pfs_fileno == PROCFS_FILENO(0, PFScurproc, -1))
1603 		len = snprintf(bf, sizeof(bf), "%ld", (long)curproc->p_pid);
1604 	else if (pfs->pfs_fileno == PROCFS_FILENO(0, PFSself, -1))
1605 		len = snprintf(bf, sizeof(bf), "%s", "curproc");
1606 	else if (pfs->pfs_fileno == PROCFS_FILENO(pfs->pfs_pid, PFStask, 0))
1607 		len = snprintf(bf, sizeof(bf), "..");
1608 	else if (pfs->pfs_fileno == PROCFS_FILENO(pfs->pfs_pid, PFScwd, -1) ||
1609 	    pfs->pfs_fileno == PROCFS_FILENO(pfs->pfs_pid, PFSchroot, -1) ||
1610 	    pfs->pfs_fileno == PROCFS_FILENO(pfs->pfs_pid, PFSexe, -1)) {
1611 		if ((error = procfs_proc_lock(pfs->pfs_pid, &pown, ESRCH)) != 0)
1612 			return error;
1613 		path = malloc(MAXPATHLEN + 4, M_TEMP, M_WAITOK|M_CANFAIL);
1614 		if (path == NULL) {
1615 			procfs_proc_unlock(pown);
1616 			return (ENOMEM);
1617 		}
1618 		bp = path + MAXPATHLEN;
1619 		*--bp = '\0';
1620 		procfs_dir(PROCFS_TYPE(pfs->pfs_fileno), curlwp, pown,
1621 		    &bp, path, MAXPATHLEN);
1622 		procfs_proc_unlock(pown);
1623 		len = strlen(bp);
1624 	} else {
1625 		file_t *fp;
1626 		struct vnode *vxp, *vp;
1627 
1628 		if ((error = procfs_proc_lock(pfs->pfs_pid, &pown, ESRCH)) != 0)
1629 			return error;
1630 
1631 		fp = fd_getfile2(pown, pfs->pfs_fd);
1632 		if (fp == NULL) {
1633 			procfs_proc_unlock(pown);
1634 			return EBADF;
1635 		}
1636 
1637 		switch (fp->f_type) {
1638 		case DTYPE_VNODE:
1639 			vxp = (struct vnode *)fp->f_data;
1640 			if (vxp->v_type != VDIR) {
1641 				error = EINVAL;
1642 				break;
1643 			}
1644 			if ((path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK))
1645 			    == NULL) {
1646 				error = ENOMEM;
1647 				break;
1648 			}
1649 			bp = path + MAXPATHLEN;
1650 			*--bp = '\0';
1651 
1652 			/*
1653 			 * XXX: kludge to avoid locking against ourselves
1654 			 * in getcwd()
1655 			 */
1656 			if (vxp->v_tag == VT_PROCFS) {
1657 				*--bp = '/';
1658 			} else {
1659 				rw_enter(&curproc->p_cwdi->cwdi_lock,
1660 				    RW_READER);
1661 				vp = curproc->p_cwdi->cwdi_rdir;
1662 				if (vp == NULL)
1663 					vp = rootvnode;
1664 				error = getcwd_common(vxp, vp, &bp, path,
1665 				    MAXPATHLEN / 2, 0, curlwp);
1666 				rw_exit(&curproc->p_cwdi->cwdi_lock);
1667 			}
1668 			if (error)
1669 				break;
1670 			len = strlen(bp);
1671 			break;
1672 
1673 		case DTYPE_MISC:
1674 			len = snprintf(bf, sizeof(bf), "%s", "[misc]");
1675 			break;
1676 
1677 		case DTYPE_KQUEUE:
1678 			len = snprintf(bf, sizeof(bf), "%s", "[kqueue]");
1679 			break;
1680 
1681 		case DTYPE_SEM:
1682 			len = snprintf(bf, sizeof(bf), "%s", "[ksem]");
1683 			break;
1684 
1685 		default:
1686 			error = EINVAL;
1687 			break;
1688 		}
1689 		closef(fp);
1690 		procfs_proc_unlock(pown);
1691 	}
1692 
1693 	if (error == 0)
1694 		error = uiomove(bp, len, ap->a_uio);
1695 	if (path)
1696 		free(path, M_TEMP);
1697 	return error;
1698 }
1699 
1700 /*
1701  * convert decimal ascii to int
1702  */
1703 static int
1704 atoi(const char *b, size_t len)
1705 {
1706 	int p = 0;
1707 
1708 	while (len--) {
1709 		char c = *b++;
1710 		if (c < '0' || c > '9')
1711 			return -1;
1712 		p = 10 * p + (c - '0');
1713 	}
1714 
1715 	return p;
1716 }
1717