xref: /netbsd-src/sys/miscfs/procfs/procfs_vnops.c (revision 946379e7b37692fc43f68eb0d1c10daa0a7f3b6c)
1 /*	$NetBSD: procfs_vnops.c,v 1.193 2015/04/20 23:03:08 riastradh Exp $	*/
2 
3 /*-
4  * Copyright (c) 2006, 2007, 2008 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Andrew Doran.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29  * POSSIBILITY OF SUCH DAMAGE.
30  */
31 
32 /*
33  * Copyright (c) 1993, 1995
34  *	The Regents of the University of California.  All rights reserved.
35  *
36  * This code is derived from software contributed to Berkeley by
37  * Jan-Simon Pendry.
38  *
39  * Redistribution and use in source and binary forms, with or without
40  * modification, are permitted provided that the following conditions
41  * are met:
42  * 1. Redistributions of source code must retain the above copyright
43  *    notice, this list of conditions and the following disclaimer.
44  * 2. Redistributions in binary form must reproduce the above copyright
45  *    notice, this list of conditions and the following disclaimer in the
46  *    documentation and/or other materials provided with the distribution.
47  * 3. Neither the name of the University nor the names of its contributors
48  *    may be used to endorse or promote products derived from this software
49  *    without specific prior written permission.
50  *
51  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
52  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
53  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
54  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
55  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
56  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
57  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
58  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
59  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
60  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
61  * SUCH DAMAGE.
62  *
63  *	@(#)procfs_vnops.c	8.18 (Berkeley) 5/21/95
64  */
65 
66 /*
67  * Copyright (c) 1993 Jan-Simon Pendry
68  *
69  * This code is derived from software contributed to Berkeley by
70  * Jan-Simon Pendry.
71  *
72  * Redistribution and use in source and binary forms, with or without
73  * modification, are permitted provided that the following conditions
74  * are met:
75  * 1. Redistributions of source code must retain the above copyright
76  *    notice, this list of conditions and the following disclaimer.
77  * 2. Redistributions in binary form must reproduce the above copyright
78  *    notice, this list of conditions and the following disclaimer in the
79  *    documentation and/or other materials provided with the distribution.
80  * 3. All advertising materials mentioning features or use of this software
81  *    must display the following acknowledgement:
82  *	This product includes software developed by the University of
83  *	California, Berkeley and its contributors.
84  * 4. Neither the name of the University nor the names of its contributors
85  *    may be used to endorse or promote products derived from this software
86  *    without specific prior written permission.
87  *
88  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
89  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
90  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
91  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
92  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
93  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
94  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
95  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
96  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
97  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
98  * SUCH DAMAGE.
99  *
100  *	@(#)procfs_vnops.c	8.18 (Berkeley) 5/21/95
101  */
102 
103 /*
104  * procfs vnode interface
105  */
106 
107 #include <sys/cdefs.h>
108 __KERNEL_RCSID(0, "$NetBSD: procfs_vnops.c,v 1.193 2015/04/20 23:03:08 riastradh Exp $");
109 
110 #include <sys/param.h>
111 #include <sys/systm.h>
112 #include <sys/time.h>
113 #include <sys/kernel.h>
114 #include <sys/file.h>
115 #include <sys/filedesc.h>
116 #include <sys/proc.h>
117 #include <sys/vnode.h>
118 #include <sys/namei.h>
119 #include <sys/malloc.h>
120 #include <sys/mount.h>
121 #include <sys/dirent.h>
122 #include <sys/resourcevar.h>
123 #include <sys/stat.h>
124 #include <sys/ptrace.h>
125 #include <sys/kauth.h>
126 
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_v2_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 	return (EROFS);
527 }
528 
529 int
530 procfs_symlink(void *v)
531 {
532 	struct vop_symlink_v3_args /* {
533 		struct vnode *a_dvp;
534 		struct vnode **a_vpp;
535 		struct componentname *a_cnp;
536 		struct vattr *a_vap;
537 		char *a_target;
538 	} */ *ap = v;
539 
540 	VOP_ABORTOP(ap->a_dvp, ap->a_cnp);
541 	return (EROFS);
542 }
543 
544 /*
545  * Works out the path to (and vnode of) the target process's current
546  * working directory or chroot.  If the caller is in a chroot and
547  * can't "reach" the target's cwd or root (or some other error
548  * occurs), a "/" is returned for the path and a NULL pointer is
549  * returned for the vnode.
550  */
551 static void
552 procfs_dir(pfstype t, struct lwp *caller, struct proc *target, char **bpp,
553     char *path, size_t len)
554 {
555 	struct cwdinfo *cwdi;
556 	struct vnode *vp, *rvp;
557 	char *bp;
558 
559 	cwdi = caller->l_proc->p_cwdi;
560 	rw_enter(&cwdi->cwdi_lock, RW_READER);
561 
562 	rvp = cwdi->cwdi_rdir;
563 	bp = bpp ? *bpp : NULL;
564 
565 	switch (t) {
566 	case PFScwd:
567 		vp = target->p_cwdi->cwdi_cdir;
568 		break;
569 	case PFSchroot:
570 		vp = target->p_cwdi->cwdi_rdir;
571 		break;
572 	case PFSexe:
573 		vp = target->p_textvp;
574 		break;
575 	default:
576 		rw_exit(&cwdi->cwdi_lock);
577 		return;
578 	}
579 
580 	/*
581 	 * XXX: this horrible kludge avoids locking panics when
582 	 * attempting to lookup links that point to within procfs
583 	 */
584 	if (vp != NULL && vp->v_tag == VT_PROCFS) {
585 		if (bpp) {
586 			*--bp = '/';
587 			*bpp = bp;
588 		}
589 		rw_exit(&cwdi->cwdi_lock);
590 		return;
591 	}
592 
593 	if (rvp == NULL)
594 		rvp = rootvnode;
595 	if (vp == NULL || getcwd_common(vp, rvp, bp ? &bp : NULL, path,
596 	    len / 2, 0, caller) != 0) {
597 		vp = NULL;
598 		if (bpp) {
599 /*
600 			if (t == PFSexe) {
601 				snprintf(path, len, "%s/%d/file"
602 				    mp->mnt_stat.f_mntonname, pfs->pfs_pid);
603 			} else */ {
604 				bp = *bpp;
605 				*--bp = '/';
606 			}
607 		}
608 	}
609 
610 	if (bpp)
611 		*bpp = bp;
612 
613 	rw_exit(&cwdi->cwdi_lock);
614 }
615 
616 /*
617  * Invent attributes for pfsnode (vp) and store
618  * them in (vap).
619  * Directories lengths are returned as zero since
620  * any real length would require the genuine size
621  * to be computed, and nothing cares anyway.
622  *
623  * this is relatively minimal for procfs.
624  */
625 int
626 procfs_getattr(void *v)
627 {
628 	struct vop_getattr_args /* {
629 		struct vnode *a_vp;
630 		struct vattr *a_vap;
631 		kauth_cred_t a_cred;
632 	} */ *ap = v;
633 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
634 	struct vattr *vap = ap->a_vap;
635 	struct proc *procp;
636 	char *path;
637 	int error;
638 
639 	/* first check the process still exists */
640 	switch (pfs->pfs_type) {
641 	case PFSroot:
642 	case PFScurproc:
643 	case PFSself:
644 		procp = NULL;
645 		break;
646 
647 	default:
648 		error = procfs_proc_lock(pfs->pfs_pid, &procp, ENOENT);
649 		if (error != 0)
650 			return (error);
651 		break;
652 	}
653 
654 	switch (pfs->pfs_type) {
655 	case PFStask:
656 		if (pfs->pfs_fd == -1) {
657 			path = NULL;
658 			break;
659 		}
660 		/*FALLTHROUGH*/
661 	case PFScwd:
662 	case PFSchroot:
663 	case PFSexe:
664 		path = malloc(MAXPATHLEN + 4, M_TEMP, M_WAITOK|M_CANFAIL);
665 		if (path == NULL && procp != NULL) {
666 			procfs_proc_unlock(procp);
667 			return (ENOMEM);
668 		}
669 		break;
670 
671 	default:
672 		path = NULL;
673 		break;
674 	}
675 
676 	if (procp != NULL) {
677 		mutex_enter(procp->p_lock);
678 		error = kauth_authorize_process(kauth_cred_get(),
679 		    KAUTH_PROCESS_CANSEE, procp,
680 		    KAUTH_ARG(KAUTH_REQ_PROCESS_CANSEE_ENTRY), NULL, NULL);
681 		mutex_exit(procp->p_lock);
682 		if (error != 0) {
683 		    	procfs_proc_unlock(procp);
684 		    	if (path != NULL)
685 		    		free(path, M_TEMP);
686 			return (ENOENT);
687 		}
688 	}
689 
690 	error = 0;
691 
692 	/* start by zeroing out the attributes */
693 	vattr_null(vap);
694 
695 	/* next do all the common fields */
696 	vap->va_type = ap->a_vp->v_type;
697 	vap->va_mode = pfs->pfs_mode;
698 	vap->va_fileid = pfs->pfs_fileno;
699 	vap->va_flags = 0;
700 	vap->va_blocksize = PAGE_SIZE;
701 
702 	/*
703 	 * Make all times be current TOD.
704 	 *
705 	 * It would be possible to get the process start
706 	 * time from the p_stats structure, but there's
707 	 * no "file creation" time stamp anyway, and the
708 	 * p_stats structure is not addressable if u. gets
709 	 * swapped out for that process.
710 	 */
711 	getnanotime(&vap->va_ctime);
712 	vap->va_atime = vap->va_mtime = vap->va_ctime;
713 	if (procp)
714 		TIMEVAL_TO_TIMESPEC(&procp->p_stats->p_start,
715 		    &vap->va_birthtime);
716 	else
717 		getnanotime(&vap->va_birthtime);
718 
719 	switch (pfs->pfs_type) {
720 	case PFSmem:
721 	case PFSregs:
722 	case PFSfpregs:
723 #if defined(__HAVE_PROCFS_MACHDEP) && defined(PROCFS_MACHDEP_PROTECT_CASES)
724 	PROCFS_MACHDEP_PROTECT_CASES
725 #endif
726 		/*
727 		 * If the process has exercised some setuid or setgid
728 		 * privilege, then rip away read/write permission so
729 		 * that only root can gain access.
730 		 */
731 		if (procp->p_flag & PK_SUGID)
732 			vap->va_mode &= ~(S_IRUSR|S_IWUSR);
733 		/* FALLTHROUGH */
734 	case PFSctl:
735 	case PFSstatus:
736 	case PFSstat:
737 	case PFSnote:
738 	case PFSnotepg:
739 	case PFSmap:
740 	case PFSmaps:
741 	case PFScmdline:
742 	case PFSemul:
743 	case PFSstatm:
744 		if (pfs->pfs_type == PFSmap || pfs->pfs_type == PFSmaps)
745 			vap->va_mode = S_IRUSR;
746 		vap->va_nlink = 1;
747 		vap->va_uid = kauth_cred_geteuid(procp->p_cred);
748 		vap->va_gid = kauth_cred_getegid(procp->p_cred);
749 		break;
750 	case PFSmeminfo:
751 	case PFSdevices:
752 	case PFScpuinfo:
753 	case PFSuptime:
754 	case PFSmounts:
755 	case PFScpustat:
756 	case PFSloadavg:
757 	case PFSversion:
758 		vap->va_nlink = 1;
759 		vap->va_uid = vap->va_gid = 0;
760 		break;
761 
762 	default:
763 		break;
764 	}
765 
766 	/*
767 	 * now do the object specific fields
768 	 *
769 	 * The size could be set from struct reg, but it's hardly
770 	 * worth the trouble, and it puts some (potentially) machine
771 	 * dependent data into this machine-independent code.  If it
772 	 * becomes important then this function should break out into
773 	 * a per-file stat function in the corresponding .c file.
774 	 */
775 
776 	switch (pfs->pfs_type) {
777 	case PFSroot:
778 		/*
779 		 * Set nlink to 1 to tell fts(3) we don't actually know.
780 		 */
781 		vap->va_nlink = 1;
782 		vap->va_uid = 0;
783 		vap->va_gid = 0;
784 		vap->va_bytes = vap->va_size = DEV_BSIZE;
785 		break;
786 
787 	case PFSself:
788 	case PFScurproc: {
789 		char bf[16];		/* should be enough */
790 		vap->va_nlink = 1;
791 		vap->va_uid = 0;
792 		vap->va_gid = 0;
793 		vap->va_bytes = vap->va_size =
794 		    snprintf(bf, sizeof(bf), "%ld", (long)curproc->p_pid);
795 		break;
796 	}
797 	case PFStask:
798 		if (pfs->pfs_fd != -1) {
799 			char bf[4];		/* should be enough */
800 			vap->va_nlink = 1;
801 			vap->va_uid = 0;
802 			vap->va_gid = 0;
803 			vap->va_bytes = vap->va_size =
804 			    snprintf(bf, sizeof(bf), "..");
805 			break;
806 		}
807 		/*FALLTHROUGH*/
808 	case PFSfd:
809 		if (pfs->pfs_fd != -1) {
810 			file_t *fp;
811 
812 			fp = fd_getfile2(procp, pfs->pfs_fd);
813 			if (fp == NULL) {
814 				error = EBADF;
815 				break;
816 			}
817 			vap->va_nlink = 1;
818 			vap->va_uid = kauth_cred_geteuid(fp->f_cred);
819 			vap->va_gid = kauth_cred_getegid(fp->f_cred);
820 			switch (fp->f_type) {
821 			case DTYPE_VNODE:
822 				vap->va_bytes = vap->va_size =
823 				    fp->f_vnode->v_size;
824 				break;
825 			default:
826 				vap->va_bytes = vap->va_size = 0;
827 				break;
828 			}
829 			closef(fp);
830 			break;
831 		}
832 		/*FALLTHROUGH*/
833 	case PFSproc:
834 		vap->va_nlink = 2;
835 		vap->va_uid = kauth_cred_geteuid(procp->p_cred);
836 		vap->va_gid = kauth_cred_getegid(procp->p_cred);
837 		vap->va_bytes = vap->va_size = DEV_BSIZE;
838 		break;
839 
840 	case PFSfile:
841 		error = EOPNOTSUPP;
842 		break;
843 
844 	case PFSmem:
845 		vap->va_bytes = vap->va_size =
846 			ctob(procp->p_vmspace->vm_tsize +
847 				    procp->p_vmspace->vm_dsize +
848 				    procp->p_vmspace->vm_ssize);
849 		break;
850 
851 #if defined(PT_GETREGS) || defined(PT_SETREGS)
852 	case PFSregs:
853 		vap->va_bytes = vap->va_size = sizeof(struct reg);
854 		break;
855 #endif
856 
857 #if defined(PT_GETFPREGS) || defined(PT_SETFPREGS)
858 	case PFSfpregs:
859 		vap->va_bytes = vap->va_size = sizeof(struct fpreg);
860 		break;
861 #endif
862 
863 	case PFSctl:
864 	case PFSstatus:
865 	case PFSstat:
866 	case PFSnote:
867 	case PFSnotepg:
868 	case PFScmdline:
869 	case PFSmeminfo:
870 	case PFSdevices:
871 	case PFScpuinfo:
872 	case PFSuptime:
873 	case PFSmounts:
874 	case PFScpustat:
875 	case PFSloadavg:
876 	case PFSstatm:
877 	case PFSversion:
878 		vap->va_bytes = vap->va_size = 0;
879 		break;
880 	case PFSmap:
881 	case PFSmaps:
882 		/*
883 		 * Advise a larger blocksize for the map files, so that
884 		 * they may be read in one pass.
885 		 */
886 		vap->va_blocksize = 4 * PAGE_SIZE;
887 		vap->va_bytes = vap->va_size = 0;
888 		break;
889 
890 	case PFScwd:
891 	case PFSchroot:
892 	case PFSexe: {
893 		char *bp;
894 
895 		vap->va_nlink = 1;
896 		vap->va_uid = 0;
897 		vap->va_gid = 0;
898 		bp = path + MAXPATHLEN;
899 		*--bp = '\0';
900 		procfs_dir(pfs->pfs_type, curlwp, procp, &bp, path,
901 		     MAXPATHLEN);
902 		vap->va_bytes = vap->va_size = strlen(bp);
903 		break;
904 	}
905 
906 	case PFSemul:
907 		vap->va_bytes = vap->va_size = strlen(procp->p_emul->e_name);
908 		break;
909 
910 #ifdef __HAVE_PROCFS_MACHDEP
911 	PROCFS_MACHDEP_NODETYPE_CASES
912 		error = procfs_machdep_getattr(ap->a_vp, vap, procp);
913 		break;
914 #endif
915 
916 	default:
917 		panic("procfs_getattr");
918 	}
919 
920 	if (procp != NULL)
921 		procfs_proc_unlock(procp);
922 	if (path != NULL)
923 		free(path, M_TEMP);
924 
925 	return (error);
926 }
927 
928 /*ARGSUSED*/
929 int
930 procfs_setattr(void *v)
931 {
932 	/*
933 	 * just fake out attribute setting
934 	 * it's not good to generate an error
935 	 * return, otherwise things like creat()
936 	 * will fail when they try to set the
937 	 * file length to 0.  worse, this means
938 	 * that echo $note > /proc/$pid/note will fail.
939 	 */
940 
941 	return (0);
942 }
943 
944 /*
945  * implement access checking.
946  *
947  * actually, the check for super-user is slightly
948  * broken since it will allow read access to write-only
949  * objects.  this doesn't cause any particular trouble
950  * but does mean that the i/o entry points need to check
951  * that the operation really does make sense.
952  */
953 int
954 procfs_access(void *v)
955 {
956 	struct vop_access_args /* {
957 		struct vnode *a_vp;
958 		int a_mode;
959 		kauth_cred_t a_cred;
960 	} */ *ap = v;
961 	struct vattr va;
962 	int error;
963 
964 	if ((error = VOP_GETATTR(ap->a_vp, &va, ap->a_cred)) != 0)
965 		return (error);
966 
967 	return kauth_authorize_vnode(ap->a_cred,
968 	    KAUTH_ACCESS_ACTION(ap->a_mode, ap->a_vp->v_type, va.va_mode),
969 	    ap->a_vp, NULL, genfs_can_access(va.va_type, va.va_mode,
970 	    va.va_uid, va.va_gid, ap->a_mode, ap->a_cred));
971 }
972 
973 /*
974  * lookup.  this is incredibly complicated in the
975  * general case, however for most pseudo-filesystems
976  * very little needs to be done.
977  *
978  * Locking isn't hard here, just poorly documented.
979  *
980  * If we're looking up ".", just vref the parent & return it.
981  *
982  * If we're looking up "..", unlock the parent, and lock "..". If everything
983  * went ok, and we're on the last component and the caller requested the
984  * parent locked, try to re-lock the parent. We do this to prevent lock
985  * races.
986  *
987  * For anything else, get the needed node. Then unlock the parent if not
988  * the last component or not LOCKPARENT (i.e. if we wouldn't re-lock the
989  * parent in the .. case).
990  *
991  * We try to exit with the parent locked in error cases.
992  */
993 int
994 procfs_lookup(void *v)
995 {
996 	struct vop_lookup_v2_args /* {
997 		struct vnode * a_dvp;
998 		struct vnode ** a_vpp;
999 		struct componentname * a_cnp;
1000 	} */ *ap = v;
1001 	struct componentname *cnp = ap->a_cnp;
1002 	struct vnode **vpp = ap->a_vpp;
1003 	struct vnode *dvp = ap->a_dvp;
1004 	const char *pname = cnp->cn_nameptr;
1005 	const struct proc_target *pt = NULL;
1006 	struct vnode *fvp;
1007 	pid_t pid, vnpid;
1008 	struct pfsnode *pfs;
1009 	struct proc *p = NULL;
1010 	struct lwp *plwp;
1011 	int i, error;
1012 	pfstype type;
1013 
1014 	*vpp = NULL;
1015 
1016 	if (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)
1017 		return (EROFS);
1018 
1019 	if (cnp->cn_namelen == 1 && *pname == '.') {
1020 		*vpp = dvp;
1021 		vref(dvp);
1022 		return (0);
1023 	}
1024 
1025 	pfs = VTOPFS(dvp);
1026 	switch (pfs->pfs_type) {
1027 	case PFSroot:
1028 		/*
1029 		 * Shouldn't get here with .. in the root node.
1030 		 */
1031 		if (cnp->cn_flags & ISDOTDOT)
1032 			return (EIO);
1033 
1034 		for (i = 0; i < nproc_root_targets; i++) {
1035 			pt = &proc_root_targets[i];
1036 			/*
1037 			 * check for node match.  proc is always NULL here,
1038 			 * so call pt_valid with constant NULL lwp.
1039 			 */
1040 			if (cnp->cn_namelen == pt->pt_namlen &&
1041 			    memcmp(pt->pt_name, pname, cnp->cn_namelen) == 0 &&
1042 			    (pt->pt_valid == NULL ||
1043 			     (*pt->pt_valid)(NULL, dvp->v_mount)))
1044 				break;
1045 		}
1046 
1047 		if (i != nproc_root_targets) {
1048 			error = procfs_allocvp(dvp->v_mount, vpp, 0,
1049 			    pt->pt_pfstype, -1);
1050 			return (error);
1051 		}
1052 
1053 		if (CNEQ(cnp, "curproc", 7)) {
1054 			pid = curproc->p_pid;
1055 			vnpid = 0;
1056 			type = PFScurproc;
1057 		} else if (CNEQ(cnp, "self", 4)) {
1058 			pid = curproc->p_pid;
1059 			vnpid = 0;
1060 			type = PFSself;
1061 		} else {
1062 			pid = (pid_t)atoi(pname, cnp->cn_namelen);
1063 			vnpid = pid;
1064 			type = PFSproc;
1065 		}
1066 
1067 		if (procfs_proc_lock(pid, &p, ESRCH) != 0)
1068 			break;
1069 		error = procfs_allocvp(dvp->v_mount, vpp, vnpid, type, -1);
1070 		procfs_proc_unlock(p);
1071 		return (error);
1072 
1073 	case PFSproc:
1074 		if (cnp->cn_flags & ISDOTDOT) {
1075 			error = procfs_allocvp(dvp->v_mount, vpp, 0, PFSroot,
1076 			    -1);
1077 			return (error);
1078 		}
1079 
1080 		if (procfs_proc_lock(pfs->pfs_pid, &p, ESRCH) != 0)
1081 			break;
1082 
1083 		mutex_enter(p->p_lock);
1084 		LIST_FOREACH(plwp, &p->p_lwps, l_sibling) {
1085 			if (plwp->l_stat != LSZOMB)
1086 				break;
1087 		}
1088 		/* Process is exiting if no-LWPS or all LWPs are LSZOMB */
1089 		if (plwp == NULL) {
1090 			mutex_exit(p->p_lock);
1091 			procfs_proc_unlock(p);
1092 			return ESRCH;
1093 		}
1094 
1095 		lwp_addref(plwp);
1096 		mutex_exit(p->p_lock);
1097 
1098 		for (pt = proc_targets, i = 0; i < nproc_targets; pt++, i++) {
1099 			int found;
1100 
1101 			found = cnp->cn_namelen == pt->pt_namlen &&
1102 			    memcmp(pt->pt_name, pname, cnp->cn_namelen) == 0 &&
1103 			    (pt->pt_valid == NULL
1104 			      || (*pt->pt_valid)(plwp, dvp->v_mount));
1105 			if (found)
1106 				break;
1107 		}
1108 		lwp_delref(plwp);
1109 
1110 		if (i == nproc_targets) {
1111 			procfs_proc_unlock(p);
1112 			break;
1113 		}
1114 		if (pt->pt_pfstype == PFSfile) {
1115 			fvp = p->p_textvp;
1116 			/* We already checked that it exists. */
1117 			vref(fvp);
1118 			procfs_proc_unlock(p);
1119 			*vpp = fvp;
1120 			return (0);
1121 		}
1122 
1123 		error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
1124 		    pt->pt_pfstype, -1);
1125 		procfs_proc_unlock(p);
1126 		return (error);
1127 
1128 	case PFSfd: {
1129 		int fd;
1130 		file_t *fp;
1131 
1132 		if ((error = procfs_proc_lock(pfs->pfs_pid, &p, ENOENT)) != 0)
1133 			return error;
1134 
1135 		if (cnp->cn_flags & ISDOTDOT) {
1136 			error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
1137 			    PFSproc, -1);
1138 			procfs_proc_unlock(p);
1139 			return (error);
1140 		}
1141 		fd = atoi(pname, cnp->cn_namelen);
1142 
1143 		fp = fd_getfile2(p, fd);
1144 		if (fp == NULL) {
1145 			procfs_proc_unlock(p);
1146 			return ENOENT;
1147 		}
1148 		fvp = fp->f_vnode;
1149 
1150 		/* Don't show directories */
1151 		if (fp->f_type == DTYPE_VNODE && fvp->v_type != VDIR) {
1152 			vref(fvp);
1153 			closef(fp);
1154 			procfs_proc_unlock(p);
1155 			*vpp = fvp;
1156 			return 0;
1157 		}
1158 
1159 		closef(fp);
1160 		error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
1161 		    PFSfd, fd);
1162 		procfs_proc_unlock(p);
1163 		return error;
1164 	}
1165 	case PFStask: {
1166 		int xpid;
1167 
1168 		if ((error = procfs_proc_lock(pfs->pfs_pid, &p, ENOENT)) != 0)
1169 			return error;
1170 
1171 		if (cnp->cn_flags & ISDOTDOT) {
1172 			error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
1173 			    PFSproc, -1);
1174 			procfs_proc_unlock(p);
1175 			return (error);
1176 		}
1177 		xpid = atoi(pname, cnp->cn_namelen);
1178 
1179 		if (xpid != pfs->pfs_pid) {
1180 			procfs_proc_unlock(p);
1181 			return ENOENT;
1182 		}
1183 		error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
1184 		    PFStask, 0);
1185 		procfs_proc_unlock(p);
1186 		return error;
1187 	}
1188 	default:
1189 		return (ENOTDIR);
1190 	}
1191 
1192 	return (cnp->cn_nameiop == LOOKUP ? ENOENT : EROFS);
1193 }
1194 
1195 int
1196 procfs_validfile(struct lwp *l, struct mount *mp)
1197 {
1198 	return l != NULL && l->l_proc != NULL && l->l_proc->p_textvp != NULL;
1199 }
1200 
1201 static int
1202 procfs_validfile_linux(struct lwp *l, struct mount *mp)
1203 {
1204 	int flags;
1205 
1206 	flags = VFSTOPROC(mp)->pmnt_flags;
1207 	return (flags & PROCFSMNT_LINUXCOMPAT) &&
1208 	    (l == NULL || l->l_proc == NULL || procfs_validfile(l, mp));
1209 }
1210 
1211 struct procfs_root_readdir_ctx {
1212 	struct uio *uiop;
1213 	off_t *cookies;
1214 	int ncookies;
1215 	off_t off;
1216 	off_t startoff;
1217 	int error;
1218 };
1219 
1220 static int
1221 procfs_root_readdir_callback(struct proc *p, void *arg)
1222 {
1223 	struct procfs_root_readdir_ctx *ctxp = arg;
1224 	struct dirent d;
1225 	struct uio *uiop;
1226 	int error;
1227 
1228 	uiop = ctxp->uiop;
1229 	if (uiop->uio_resid < UIO_MX)
1230 		return -1; /* no space */
1231 
1232 	if (ctxp->off < ctxp->startoff) {
1233 		ctxp->off++;
1234 		return 0;
1235 	}
1236 
1237 	if (kauth_authorize_process(kauth_cred_get(),
1238 	    KAUTH_PROCESS_CANSEE, p,
1239 	    KAUTH_ARG(KAUTH_REQ_PROCESS_CANSEE_ENTRY), NULL, NULL) != 0)
1240 		return 0;
1241 
1242 	memset(&d, 0, UIO_MX);
1243 	d.d_reclen = UIO_MX;
1244 	d.d_fileno = PROCFS_FILENO(p->p_pid, PFSproc, -1);
1245 	d.d_namlen = snprintf(d.d_name,
1246 	    UIO_MX - offsetof(struct dirent, d_name), "%ld", (long)p->p_pid);
1247 	d.d_type = DT_DIR;
1248 
1249 	mutex_exit(proc_lock);
1250 	error = uiomove(&d, UIO_MX, uiop);
1251 	mutex_enter(proc_lock);
1252 	if (error) {
1253 		ctxp->error = error;
1254 		return -1;
1255 	}
1256 
1257 	ctxp->ncookies++;
1258 	if (ctxp->cookies)
1259 		*(ctxp->cookies)++ = ctxp->off + 1;
1260 	ctxp->off++;
1261 
1262 	return 0;
1263 }
1264 
1265 /*
1266  * readdir returns directory entries from pfsnode (vp).
1267  *
1268  * the strategy here with procfs is to generate a single
1269  * directory entry at a time (struct dirent) and then
1270  * copy that out to userland using uiomove.  a more efficent
1271  * though more complex implementation, would try to minimize
1272  * the number of calls to uiomove().  for procfs, this is
1273  * hardly worth the added code complexity.
1274  *
1275  * this should just be done through read()
1276  */
1277 int
1278 procfs_readdir(void *v)
1279 {
1280 	struct vop_readdir_args /* {
1281 		struct vnode *a_vp;
1282 		struct uio *a_uio;
1283 		kauth_cred_t a_cred;
1284 		int *a_eofflag;
1285 		off_t **a_cookies;
1286 		int *a_ncookies;
1287 	} */ *ap = v;
1288 	struct uio *uio = ap->a_uio;
1289 	struct dirent d;
1290 	struct pfsnode *pfs;
1291 	off_t i;
1292 	int error;
1293 	off_t *cookies = NULL;
1294 	int ncookies;
1295 	struct vnode *vp;
1296 	const struct proc_target *pt;
1297 	struct procfs_root_readdir_ctx ctx;
1298 	struct lwp *l;
1299 	int nfd;
1300 
1301 	vp = ap->a_vp;
1302 	pfs = VTOPFS(vp);
1303 
1304 	if (uio->uio_resid < UIO_MX)
1305 		return (EINVAL);
1306 	if (uio->uio_offset < 0)
1307 		return (EINVAL);
1308 
1309 	error = 0;
1310 	i = uio->uio_offset;
1311 	memset(&d, 0, UIO_MX);
1312 	d.d_reclen = UIO_MX;
1313 	ncookies = uio->uio_resid / UIO_MX;
1314 
1315 	switch (pfs->pfs_type) {
1316 	/*
1317 	 * this is for the process-specific sub-directories.
1318 	 * all that is needed to is copy out all the entries
1319 	 * from the procent[] table (top of this file).
1320 	 */
1321 	case PFSproc: {
1322 		struct proc *p;
1323 
1324 		if (i >= nproc_targets)
1325 			return 0;
1326 
1327 		if (procfs_proc_lock(pfs->pfs_pid, &p, ESRCH) != 0)
1328 			break;
1329 
1330 		if (ap->a_ncookies) {
1331 			ncookies = min(ncookies, (nproc_targets - i));
1332 			cookies = malloc(ncookies * sizeof (off_t),
1333 			    M_TEMP, M_WAITOK);
1334 			*ap->a_cookies = cookies;
1335 		}
1336 
1337 		for (pt = &proc_targets[i];
1338 		     uio->uio_resid >= UIO_MX && i < nproc_targets; pt++, i++) {
1339 			if (pt->pt_valid) {
1340 				/* XXXSMP LWP can disappear */
1341 				mutex_enter(p->p_lock);
1342 				l = LIST_FIRST(&p->p_lwps);
1343 				KASSERT(l != NULL);
1344 				mutex_exit(p->p_lock);
1345 				if ((*pt->pt_valid)(l, vp->v_mount) == 0)
1346 					continue;
1347 			}
1348 
1349 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid,
1350 			    pt->pt_pfstype, -1);
1351 			d.d_namlen = pt->pt_namlen;
1352 			memcpy(d.d_name, pt->pt_name, pt->pt_namlen + 1);
1353 			d.d_type = pt->pt_type;
1354 
1355 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1356 				break;
1357 			if (cookies)
1358 				*cookies++ = i + 1;
1359 		}
1360 
1361 		procfs_proc_unlock(p);
1362 	    	break;
1363 	}
1364 	case PFSfd: {
1365 		struct proc *p;
1366 		file_t *fp;
1367 		int lim, nc = 0;
1368 
1369 		if ((error = procfs_proc_lock(pfs->pfs_pid, &p, ESRCH)) != 0)
1370 			return error;
1371 
1372 		/* XXX Should this be by file as well? */
1373 		if (kauth_authorize_process(kauth_cred_get(),
1374 		    KAUTH_PROCESS_CANSEE, p,
1375 		    KAUTH_ARG(KAUTH_REQ_PROCESS_CANSEE_OPENFILES), NULL,
1376 		    NULL) != 0) {
1377 		    	procfs_proc_unlock(p);
1378 			return ESRCH;
1379 		}
1380 
1381 		nfd = p->p_fd->fd_dt->dt_nfiles;
1382 
1383 		lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfiles);
1384 		if (i >= lim) {
1385 		    	procfs_proc_unlock(p);
1386 			return 0;
1387 		}
1388 
1389 		if (ap->a_ncookies) {
1390 			ncookies = min(ncookies, (nfd + 2 - i));
1391 			cookies = malloc(ncookies * sizeof (off_t),
1392 			    M_TEMP, M_WAITOK);
1393 			*ap->a_cookies = cookies;
1394 		}
1395 
1396 		for (; i < 2 && uio->uio_resid >= UIO_MX; i++) {
1397 			pt = &proc_targets[i];
1398 			d.d_namlen = pt->pt_namlen;
1399 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid,
1400 			    pt->pt_pfstype, -1);
1401 			(void)memcpy(d.d_name, pt->pt_name, pt->pt_namlen + 1);
1402 			d.d_type = pt->pt_type;
1403 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1404 				break;
1405 			if (cookies)
1406 				*cookies++ = i + 1;
1407 			nc++;
1408 		}
1409 		if (error) {
1410 			ncookies = nc;
1411 			break;
1412 		}
1413 		for (; uio->uio_resid >= UIO_MX && i < nfd; i++) {
1414 			/* check the descriptor exists */
1415 			if ((fp = fd_getfile2(p, i - 2)) == NULL)
1416 				continue;
1417 			closef(fp);
1418 
1419 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid, PFSfd, i - 2);
1420 			d.d_namlen = snprintf(d.d_name, sizeof(d.d_name),
1421 			    "%lld", (long long)(i - 2));
1422 			d.d_type = VREG;
1423 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1424 				break;
1425 			if (cookies)
1426 				*cookies++ = i + 1;
1427 			nc++;
1428 		}
1429 		ncookies = nc;
1430 		procfs_proc_unlock(p);
1431 		break;
1432 	}
1433 	case PFStask: {
1434 		struct proc *p;
1435 		int nc = 0;
1436 
1437 		if ((error = procfs_proc_lock(pfs->pfs_pid, &p, ESRCH)) != 0)
1438 			return error;
1439 
1440 		nfd = 3;	/* ., .., pid */
1441 
1442 		if (ap->a_ncookies) {
1443 			ncookies = min(ncookies, (nfd + 2 - i));
1444 			cookies = malloc(ncookies * sizeof (off_t),
1445 			    M_TEMP, M_WAITOK);
1446 			*ap->a_cookies = cookies;
1447 		}
1448 
1449 		for (; i < 2 && uio->uio_resid >= UIO_MX; i++) {
1450 			pt = &proc_targets[i];
1451 			d.d_namlen = pt->pt_namlen;
1452 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid,
1453 			    pt->pt_pfstype, -1);
1454 			(void)memcpy(d.d_name, pt->pt_name, pt->pt_namlen + 1);
1455 			d.d_type = pt->pt_type;
1456 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1457 				break;
1458 			if (cookies)
1459 				*cookies++ = i + 1;
1460 			nc++;
1461 		}
1462 		if (error) {
1463 			ncookies = nc;
1464 			break;
1465 		}
1466 		for (; uio->uio_resid >= UIO_MX && i < nfd; i++) {
1467 			/* check the descriptor exists */
1468 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid, PFStask,
1469 			    i - 2);
1470 			d.d_namlen = snprintf(d.d_name, sizeof(d.d_name),
1471 			    "%ld", (long)pfs->pfs_pid);
1472 			d.d_type = DT_LNK;
1473 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1474 				break;
1475 			if (cookies)
1476 				*cookies++ = i + 1;
1477 			nc++;
1478 		}
1479 		ncookies = nc;
1480 		procfs_proc_unlock(p);
1481 		break;
1482 	}
1483 
1484 	/*
1485 	 * this is for the root of the procfs filesystem
1486 	 * what is needed are special entries for "curproc"
1487 	 * and "self" followed by an entry for each process
1488 	 * on allproc.
1489 	 */
1490 
1491 	case PFSroot: {
1492 		int nc = 0;
1493 
1494 		if (ap->a_ncookies) {
1495 			/*
1496 			 * XXX Potentially allocating too much space here,
1497 			 * but I'm lazy. This loop needs some work.
1498 			 */
1499 			cookies = malloc(ncookies * sizeof (off_t),
1500 			    M_TEMP, M_WAITOK);
1501 			*ap->a_cookies = cookies;
1502 		}
1503 		error = 0;
1504 		/* 0 ... 3 are static entries. */
1505 		for (; i <= 3 && uio->uio_resid >= UIO_MX; i++) {
1506 			switch (i) {
1507 			case 0:		/* `.' */
1508 			case 1:		/* `..' */
1509 				d.d_fileno = PROCFS_FILENO(0, PFSroot, -1);
1510 				d.d_namlen = i + 1;
1511 				memcpy(d.d_name, "..", d.d_namlen);
1512 				d.d_name[i + 1] = '\0';
1513 				d.d_type = DT_DIR;
1514 				break;
1515 
1516 			case 2:
1517 				d.d_fileno = PROCFS_FILENO(0, PFScurproc, -1);
1518 				d.d_namlen = sizeof("curproc") - 1;
1519 				memcpy(d.d_name, "curproc", sizeof("curproc"));
1520 				d.d_type = DT_LNK;
1521 				break;
1522 
1523 			case 3:
1524 				d.d_fileno = PROCFS_FILENO(0, PFSself, -1);
1525 				d.d_namlen = sizeof("self") - 1;
1526 				memcpy(d.d_name, "self", sizeof("self"));
1527 				d.d_type = DT_LNK;
1528 				break;
1529 			}
1530 
1531 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1532 				break;
1533 			nc++;
1534 			if (cookies)
1535 				*cookies++ = i + 1;
1536 		}
1537 		/* 4 ... are process entries. */
1538 		ctx.uiop = uio;
1539 		ctx.error = 0;
1540 		ctx.off = 4;
1541 		ctx.startoff = i;
1542 		ctx.cookies = cookies;
1543 		ctx.ncookies = nc;
1544 		proclist_foreach_call(&allproc,
1545 		    procfs_root_readdir_callback, &ctx);
1546 		cookies = ctx.cookies;
1547 		nc = ctx.ncookies;
1548 		error = ctx.error;
1549 		if (error)
1550 			break;
1551 
1552 		/* misc entries. */
1553 		if (i < ctx.off)
1554 			i = ctx.off;
1555 		if (i >= ctx.off + nproc_root_targets)
1556 			break;
1557 		for (pt = &proc_root_targets[i - ctx.off];
1558 		    uio->uio_resid >= UIO_MX &&
1559 		    pt < &proc_root_targets[nproc_root_targets];
1560 		    pt++, i++) {
1561 			if (pt->pt_valid &&
1562 			    (*pt->pt_valid)(NULL, vp->v_mount) == 0)
1563 				continue;
1564 			d.d_fileno = PROCFS_FILENO(0, pt->pt_pfstype, -1);
1565 			d.d_namlen = pt->pt_namlen;
1566 			memcpy(d.d_name, pt->pt_name, pt->pt_namlen + 1);
1567 			d.d_type = pt->pt_type;
1568 
1569 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1570 				break;
1571 			nc++;
1572 			if (cookies)
1573 				*cookies++ = i + 1;
1574 		}
1575 
1576 		ncookies = nc;
1577 		break;
1578 	}
1579 
1580 	default:
1581 		error = ENOTDIR;
1582 		break;
1583 	}
1584 
1585 	if (ap->a_ncookies) {
1586 		if (error) {
1587 			if (cookies)
1588 				free(*ap->a_cookies, M_TEMP);
1589 			*ap->a_ncookies = 0;
1590 			*ap->a_cookies = NULL;
1591 		} else
1592 			*ap->a_ncookies = ncookies;
1593 	}
1594 	uio->uio_offset = i;
1595 	return (error);
1596 }
1597 
1598 /*
1599  * readlink reads the link of `curproc' and others
1600  */
1601 int
1602 procfs_readlink(void *v)
1603 {
1604 	struct vop_readlink_args *ap = v;
1605 	char bf[16];		/* should be enough */
1606 	char *bp = bf;
1607 	char *path = NULL;
1608 	int len = 0;
1609 	int error = 0;
1610 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
1611 	struct proc *pown;
1612 
1613 	if (pfs->pfs_fileno == PROCFS_FILENO(0, PFScurproc, -1))
1614 		len = snprintf(bf, sizeof(bf), "%ld", (long)curproc->p_pid);
1615 	else if (pfs->pfs_fileno == PROCFS_FILENO(0, PFSself, -1))
1616 		len = snprintf(bf, sizeof(bf), "%s", "curproc");
1617 	else if (pfs->pfs_fileno == PROCFS_FILENO(pfs->pfs_pid, PFStask, 0))
1618 		len = snprintf(bf, sizeof(bf), "..");
1619 	else if (pfs->pfs_fileno == PROCFS_FILENO(pfs->pfs_pid, PFScwd, -1) ||
1620 	    pfs->pfs_fileno == PROCFS_FILENO(pfs->pfs_pid, PFSchroot, -1) ||
1621 	    pfs->pfs_fileno == PROCFS_FILENO(pfs->pfs_pid, PFSexe, -1)) {
1622 		if ((error = procfs_proc_lock(pfs->pfs_pid, &pown, ESRCH)) != 0)
1623 			return error;
1624 		path = malloc(MAXPATHLEN + 4, M_TEMP, M_WAITOK|M_CANFAIL);
1625 		if (path == NULL) {
1626 			procfs_proc_unlock(pown);
1627 			return (ENOMEM);
1628 		}
1629 		bp = path + MAXPATHLEN;
1630 		*--bp = '\0';
1631 		procfs_dir(PROCFS_TYPE(pfs->pfs_fileno), curlwp, pown,
1632 		    &bp, path, MAXPATHLEN);
1633 		procfs_proc_unlock(pown);
1634 		len = strlen(bp);
1635 	} else {
1636 		file_t *fp;
1637 		struct vnode *vxp, *vp;
1638 
1639 		if ((error = procfs_proc_lock(pfs->pfs_pid, &pown, ESRCH)) != 0)
1640 			return error;
1641 
1642 		fp = fd_getfile2(pown, pfs->pfs_fd);
1643 		if (fp == NULL) {
1644 			procfs_proc_unlock(pown);
1645 			return EBADF;
1646 		}
1647 
1648 		switch (fp->f_type) {
1649 		case DTYPE_VNODE:
1650 			vxp = fp->f_vnode;
1651 			if (vxp->v_type != VDIR) {
1652 				error = EINVAL;
1653 				break;
1654 			}
1655 			if ((path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK))
1656 			    == NULL) {
1657 				error = ENOMEM;
1658 				break;
1659 			}
1660 			bp = path + MAXPATHLEN;
1661 			*--bp = '\0';
1662 
1663 			/*
1664 			 * XXX: kludge to avoid locking against ourselves
1665 			 * in getcwd()
1666 			 */
1667 			if (vxp->v_tag == VT_PROCFS) {
1668 				*--bp = '/';
1669 			} else {
1670 				rw_enter(&curproc->p_cwdi->cwdi_lock,
1671 				    RW_READER);
1672 				vp = curproc->p_cwdi->cwdi_rdir;
1673 				if (vp == NULL)
1674 					vp = rootvnode;
1675 				error = getcwd_common(vxp, vp, &bp, path,
1676 				    MAXPATHLEN / 2, 0, curlwp);
1677 				rw_exit(&curproc->p_cwdi->cwdi_lock);
1678 			}
1679 			if (error)
1680 				break;
1681 			len = strlen(bp);
1682 			break;
1683 
1684 		case DTYPE_MISC:
1685 			len = snprintf(bf, sizeof(bf), "%s", "[misc]");
1686 			break;
1687 
1688 		case DTYPE_KQUEUE:
1689 			len = snprintf(bf, sizeof(bf), "%s", "[kqueue]");
1690 			break;
1691 
1692 		case DTYPE_SEM:
1693 			len = snprintf(bf, sizeof(bf), "%s", "[ksem]");
1694 			break;
1695 
1696 		default:
1697 			error = EINVAL;
1698 			break;
1699 		}
1700 		closef(fp);
1701 		procfs_proc_unlock(pown);
1702 	}
1703 
1704 	if (error == 0)
1705 		error = uiomove(bp, len, ap->a_uio);
1706 	if (path)
1707 		free(path, M_TEMP);
1708 	return error;
1709 }
1710 
1711 /*
1712  * convert decimal ascii to int
1713  */
1714 static int
1715 atoi(const char *b, size_t len)
1716 {
1717 	int p = 0;
1718 
1719 	while (len--) {
1720 		char c = *b++;
1721 		if (c < '0' || c > '9')
1722 			return -1;
1723 		p = 10 * p + (c - '0');
1724 	}
1725 
1726 	return p;
1727 }
1728