xref: /netbsd-src/sys/miscfs/procfs/procfs_vnops.c (revision 267197ec1eebfcb9810ea27a89625b6ddf68e3e7)
1 /*	$NetBSD: procfs_vnops.c,v 1.165 2008/01/23 15:04:40 elad Exp $	*/
2 
3 /*-
4  * Copyright (c) 2006, 2007 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  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the NetBSD
21  *	Foundation, Inc. and its contributors.
22  * 4. Neither the name of The NetBSD Foundation nor the names of its
23  *    contributors may be used to endorse or promote products derived
24  *    from this software without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
27  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
28  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
29  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
30  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36  * POSSIBILITY OF SUCH DAMAGE.
37  */
38 
39 /*
40  * Copyright (c) 1993, 1995
41  *	The Regents of the University of California.  All rights reserved.
42  *
43  * This code is derived from software contributed to Berkeley by
44  * Jan-Simon Pendry.
45  *
46  * Redistribution and use in source and binary forms, with or without
47  * modification, are permitted provided that the following conditions
48  * are met:
49  * 1. Redistributions of source code must retain the above copyright
50  *    notice, this list of conditions and the following disclaimer.
51  * 2. Redistributions in binary form must reproduce the above copyright
52  *    notice, this list of conditions and the following disclaimer in the
53  *    documentation and/or other materials provided with the distribution.
54  * 3. Neither the name of the University nor the names of its contributors
55  *    may be used to endorse or promote products derived from this software
56  *    without specific prior written permission.
57  *
58  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
59  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
60  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
61  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
62  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
63  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
64  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
65  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
66  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
67  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
68  * SUCH DAMAGE.
69  *
70  *	@(#)procfs_vnops.c	8.18 (Berkeley) 5/21/95
71  */
72 
73 /*
74  * Copyright (c) 1993 Jan-Simon Pendry
75  *
76  * This code is derived from software contributed to Berkeley by
77  * Jan-Simon Pendry.
78  *
79  * Redistribution and use in source and binary forms, with or without
80  * modification, are permitted provided that the following conditions
81  * are met:
82  * 1. Redistributions of source code must retain the above copyright
83  *    notice, this list of conditions and the following disclaimer.
84  * 2. Redistributions in binary form must reproduce the above copyright
85  *    notice, this list of conditions and the following disclaimer in the
86  *    documentation and/or other materials provided with the distribution.
87  * 3. All advertising materials mentioning features or use of this software
88  *    must display the following acknowledgement:
89  *	This product includes software developed by the University of
90  *	California, Berkeley and its contributors.
91  * 4. Neither the name of the University nor the names of its contributors
92  *    may be used to endorse or promote products derived from this software
93  *    without specific prior written permission.
94  *
95  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
96  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
97  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
98  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
99  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
100  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
101  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
102  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
103  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
104  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
105  * SUCH DAMAGE.
106  *
107  *	@(#)procfs_vnops.c	8.18 (Berkeley) 5/21/95
108  */
109 
110 /*
111  * procfs vnode interface
112  */
113 
114 #include <sys/cdefs.h>
115 __KERNEL_RCSID(0, "$NetBSD: procfs_vnops.c,v 1.165 2008/01/23 15:04:40 elad Exp $");
116 
117 #include <sys/param.h>
118 #include <sys/systm.h>
119 #include <sys/time.h>
120 #include <sys/kernel.h>
121 #include <sys/file.h>
122 #include <sys/filedesc.h>
123 #include <sys/proc.h>
124 #include <sys/vnode.h>
125 #include <sys/namei.h>
126 #include <sys/malloc.h>
127 #include <sys/mount.h>
128 #include <sys/dirent.h>
129 #include <sys/resourcevar.h>
130 #include <sys/stat.h>
131 #include <sys/ptrace.h>
132 #include <sys/kauth.h>
133 
134 #include <uvm/uvm_extern.h>	/* for PAGE_SIZE */
135 
136 #include <machine/reg.h>
137 
138 #include <miscfs/genfs/genfs.h>
139 #include <miscfs/procfs/procfs.h>
140 
141 /*
142  * Vnode Operations.
143  *
144  */
145 
146 static int procfs_validfile_linux(struct lwp *, struct mount *);
147 static int procfs_root_readdir_callback(struct proc *, void *);
148 static void procfs_dir(pfstype, struct lwp *, struct proc *, char **, char *,
149     size_t);
150 
151 /*
152  * This is a list of the valid names in the
153  * process-specific sub-directories.  It is
154  * used in procfs_lookup and procfs_readdir
155  */
156 static const struct proc_target {
157 	u_char	pt_type;
158 	u_char	pt_namlen;
159 	const char	*pt_name;
160 	pfstype	pt_pfstype;
161 	int	(*pt_valid)(struct lwp *, struct mount *);
162 } proc_targets[] = {
163 #define N(s) sizeof(s)-1, s
164 	/*	  name		type		validp */
165 	{ DT_DIR, N("."),	PFSproc,	NULL },
166 	{ DT_DIR, N(".."),	PFSroot,	NULL },
167 	{ DT_DIR, N("fd"),	PFSfd,		NULL },
168 	{ DT_REG, N("file"),	PFSfile,	procfs_validfile },
169 	{ DT_REG, N("mem"),	PFSmem,		NULL },
170 	{ DT_REG, N("regs"),	PFSregs,	procfs_validregs },
171 	{ DT_REG, N("fpregs"),	PFSfpregs,	procfs_validfpregs },
172 	{ DT_REG, N("ctl"),	PFSctl,		NULL },
173 	{ DT_REG, N("stat"),	PFSstat,	procfs_validfile_linux },
174 	{ DT_REG, N("status"),	PFSstatus,	NULL },
175 	{ DT_REG, N("note"),	PFSnote,	NULL },
176 	{ DT_REG, N("notepg"),	PFSnotepg,	NULL },
177 	{ DT_REG, N("map"),	PFSmap,		procfs_validmap },
178 	{ DT_REG, N("maps"),	PFSmaps,	procfs_validmap },
179 	{ DT_REG, N("cmdline"), PFScmdline,	NULL },
180 	{ DT_REG, N("exe"),	PFSexe,		procfs_validfile },
181 	{ DT_LNK, N("cwd"),	PFScwd,		NULL },
182 	{ DT_LNK, N("root"),	PFSchroot,	NULL },
183 	{ DT_LNK, N("emul"),	PFSemul,	NULL },
184 	{ DT_REG, N("statm"),	PFSstatm,	procfs_validfile_linux },
185 #ifdef __HAVE_PROCFS_MACHDEP
186 	PROCFS_MACHDEP_NODETYPE_DEFNS
187 #endif
188 #undef N
189 };
190 static const int nproc_targets = sizeof(proc_targets) / sizeof(proc_targets[0]);
191 
192 /*
193  * List of files in the root directory. Note: the validate function will
194  * be called with p == NULL for these ones.
195  */
196 static const struct proc_target proc_root_targets[] = {
197 #define N(s) sizeof(s)-1, s
198 	/*	  name		    type	    validp */
199 	{ DT_REG, N("meminfo"),     PFSmeminfo,        procfs_validfile_linux },
200 	{ DT_REG, N("cpuinfo"),     PFScpuinfo,        procfs_validfile_linux },
201 	{ DT_REG, N("uptime"),      PFSuptime,         procfs_validfile_linux },
202 	{ DT_REG, N("mounts"),	    PFSmounts,	       procfs_validfile_linux },
203 	{ DT_REG, N("devices"),     PFSdevices,        procfs_validfile_linux },
204 	{ DT_REG, N("stat"),	    PFScpustat,        procfs_validfile_linux },
205 	{ DT_REG, N("loadavg"),	    PFSloadavg,        procfs_validfile_linux },
206 #undef N
207 };
208 static const int nproc_root_targets =
209     sizeof(proc_root_targets) / sizeof(proc_root_targets[0]);
210 
211 int	procfs_lookup(void *);
212 #define	procfs_create	genfs_eopnotsupp
213 #define	procfs_mknod	genfs_eopnotsupp
214 int	procfs_open(void *);
215 int	procfs_close(void *);
216 int	procfs_access(void *);
217 int	procfs_getattr(void *);
218 int	procfs_setattr(void *);
219 #define	procfs_read	procfs_rw
220 #define	procfs_write	procfs_rw
221 #define	procfs_fcntl	genfs_fcntl
222 #define	procfs_ioctl	genfs_enoioctl
223 #define	procfs_poll	genfs_poll
224 #define procfs_revoke	genfs_revoke
225 #define	procfs_fsync	genfs_nullop
226 #define	procfs_seek	genfs_nullop
227 #define	procfs_remove	genfs_eopnotsupp
228 int	procfs_link(void *);
229 #define	procfs_rename	genfs_eopnotsupp
230 #define	procfs_mkdir	genfs_eopnotsupp
231 #define	procfs_rmdir	genfs_eopnotsupp
232 int	procfs_symlink(void *);
233 int	procfs_readdir(void *);
234 int	procfs_readlink(void *);
235 #define	procfs_abortop	genfs_abortop
236 int	procfs_inactive(void *);
237 int	procfs_reclaim(void *);
238 #define	procfs_lock	genfs_lock
239 #define	procfs_unlock	genfs_unlock
240 #define	procfs_bmap	genfs_badop
241 #define	procfs_strategy	genfs_badop
242 int	procfs_print(void *);
243 int	procfs_pathconf(void *);
244 #define	procfs_islocked	genfs_islocked
245 #define	procfs_advlock	genfs_einval
246 #define	procfs_bwrite	genfs_eopnotsupp
247 #define procfs_putpages	genfs_null_putpages
248 
249 static int atoi(const char *, size_t);
250 
251 /*
252  * procfs vnode operations.
253  */
254 int (**procfs_vnodeop_p)(void *);
255 const struct vnodeopv_entry_desc procfs_vnodeop_entries[] = {
256 	{ &vop_default_desc, vn_default_error },
257 	{ &vop_lookup_desc, procfs_lookup },		/* lookup */
258 	{ &vop_create_desc, procfs_create },		/* create */
259 	{ &vop_mknod_desc, procfs_mknod },		/* mknod */
260 	{ &vop_open_desc, procfs_open },		/* open */
261 	{ &vop_close_desc, procfs_close },		/* close */
262 	{ &vop_access_desc, procfs_access },		/* access */
263 	{ &vop_getattr_desc, procfs_getattr },		/* getattr */
264 	{ &vop_setattr_desc, procfs_setattr },		/* setattr */
265 	{ &vop_read_desc, procfs_read },		/* read */
266 	{ &vop_write_desc, procfs_write },		/* write */
267 	{ &vop_fcntl_desc, procfs_fcntl },		/* fcntl */
268 	{ &vop_ioctl_desc, procfs_ioctl },		/* ioctl */
269 	{ &vop_poll_desc, procfs_poll },		/* poll */
270 	{ &vop_revoke_desc, procfs_revoke },		/* revoke */
271 	{ &vop_fsync_desc, procfs_fsync },		/* fsync */
272 	{ &vop_seek_desc, procfs_seek },		/* seek */
273 	{ &vop_remove_desc, procfs_remove },		/* remove */
274 	{ &vop_link_desc, procfs_link },		/* link */
275 	{ &vop_rename_desc, procfs_rename },		/* rename */
276 	{ &vop_mkdir_desc, procfs_mkdir },		/* mkdir */
277 	{ &vop_rmdir_desc, procfs_rmdir },		/* rmdir */
278 	{ &vop_symlink_desc, procfs_symlink },		/* symlink */
279 	{ &vop_readdir_desc, procfs_readdir },		/* readdir */
280 	{ &vop_readlink_desc, procfs_readlink },	/* readlink */
281 	{ &vop_abortop_desc, procfs_abortop },		/* abortop */
282 	{ &vop_inactive_desc, procfs_inactive },	/* inactive */
283 	{ &vop_reclaim_desc, procfs_reclaim },		/* reclaim */
284 	{ &vop_lock_desc, procfs_lock },		/* lock */
285 	{ &vop_unlock_desc, procfs_unlock },		/* unlock */
286 	{ &vop_bmap_desc, procfs_bmap },		/* bmap */
287 	{ &vop_strategy_desc, procfs_strategy },	/* strategy */
288 	{ &vop_print_desc, procfs_print },		/* print */
289 	{ &vop_islocked_desc, procfs_islocked },	/* islocked */
290 	{ &vop_pathconf_desc, procfs_pathconf },	/* pathconf */
291 	{ &vop_advlock_desc, procfs_advlock },		/* advlock */
292 	{ &vop_putpages_desc, procfs_putpages },	/* putpages */
293 	{ NULL, NULL }
294 };
295 const struct vnodeopv_desc procfs_vnodeop_opv_desc =
296 	{ &procfs_vnodeop_p, procfs_vnodeop_entries };
297 /*
298  * set things up for doing i/o on
299  * the pfsnode (vp).  (vp) is locked
300  * on entry, and should be left locked
301  * on exit.
302  *
303  * for procfs we don't need to do anything
304  * in particular for i/o.  all that is done
305  * is to support exclusive open on process
306  * memory images.
307  */
308 int
309 procfs_open(v)
310 	void *v;
311 {
312 	struct vop_open_args /* {
313 		struct vnode *a_vp;
314 		int  a_mode;
315 		kauth_cred_t a_cred;
316 	} */ *ap = v;
317 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
318 	struct lwp *l1;
319 	struct proc *p2;
320 	int error;
321 
322 	if ((error = procfs_proc_lock(pfs->pfs_pid, &p2, ENOENT)) != 0)
323 		return error;
324 
325 	l1 = curlwp;				/* tracer */
326 
327 #define	M2K(m)	(((m) & FREAD) && ((m) & FWRITE) ? \
328 		 KAUTH_REQ_PROCESS_PROCFS_RW : \
329 		 (m) & FWRITE ? KAUTH_REQ_PROCESS_PROCFS_WRITE : \
330 		 KAUTH_REQ_PROCESS_PROCFS_READ)
331 
332 	mutex_enter(&p2->p_mutex);
333 	error = kauth_authorize_process(l1->l_cred, KAUTH_PROCESS_PROCFS,
334 	    p2, pfs, KAUTH_ARG(M2K(ap->a_mode)), NULL);
335 	mutex_exit(&p2->p_mutex);
336 	if (error) {
337 		procfs_proc_unlock(p2);
338 		return (error);
339 	}
340 
341 #undef M2K
342 
343 	switch (pfs->pfs_type) {
344 	case PFSmem:
345 		if (((pfs->pfs_flags & FWRITE) && (ap->a_mode & O_EXCL)) ||
346 		    ((pfs->pfs_flags & O_EXCL) && (ap->a_mode & FWRITE))) {
347 			error = EBUSY;
348 			break;
349 		}
350 
351 		if (!proc_isunder(p2, l1)) {
352 			error = EPERM;
353 			break;
354 		}
355 
356 		if (ap->a_mode & FWRITE)
357 			pfs->pfs_flags = ap->a_mode & (FWRITE|O_EXCL);
358 
359 		break;
360 
361 	case PFSregs:
362 	case PFSfpregs:
363 		if (!proc_isunder(p2, l1)) {
364 			error = EPERM;
365 			break;
366 		}
367 		break;
368 
369 	default:
370 		break;
371 	}
372 
373 	procfs_proc_unlock(p2);
374 	return (error);
375 }
376 
377 /*
378  * close the pfsnode (vp) after doing i/o.
379  * (vp) is not locked on entry or exit.
380  *
381  * nothing to do for procfs other than undo
382  * any exclusive open flag (see _open above).
383  */
384 int
385 procfs_close(v)
386 	void *v;
387 {
388 	struct vop_close_args /* {
389 		struct vnode *a_vp;
390 		int  a_fflag;
391 		kauth_cred_t a_cred;
392 	} */ *ap = v;
393 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
394 
395 	switch (pfs->pfs_type) {
396 	case PFSmem:
397 		if ((ap->a_fflag & FWRITE) && (pfs->pfs_flags & O_EXCL))
398 			pfs->pfs_flags &= ~(FWRITE|O_EXCL);
399 		break;
400 
401 	default:
402 		break;
403 	}
404 
405 	return (0);
406 }
407 
408 /*
409  * _inactive is called when the pfsnode
410  * is vrele'd and the reference count goes
411  * to zero.  (vp) will be on the vnode free
412  * list, so to get it back vget() must be
413  * used.
414  *
415  * (vp) is locked on entry, but must be unlocked on exit.
416  */
417 int
418 procfs_inactive(v)
419 	void *v;
420 {
421 	struct vop_inactive_args /* {
422 		struct vnode *a_vp;
423 		bool *a_recycle;
424 	} */ *ap = v;
425 	struct vnode *vp = ap->a_vp;
426 	struct pfsnode *pfs = VTOPFS(vp);
427 
428 	mutex_enter(&proclist_lock);
429 	*ap->a_recycle = (p_find(pfs->pfs_pid, PFIND_LOCKED) == NULL);
430 	mutex_exit(&proclist_lock);
431 
432 	VOP_UNLOCK(vp, 0);
433 
434 	return (0);
435 }
436 
437 /*
438  * _reclaim is called when getnewvnode()
439  * wants to make use of an entry on the vnode
440  * free list.  at this time the filesystem needs
441  * to free any private data and remove the node
442  * from any private lists.
443  */
444 int
445 procfs_reclaim(v)
446 	void *v;
447 {
448 	struct vop_reclaim_args /* {
449 		struct vnode *a_vp;
450 	} */ *ap = v;
451 
452 	return (procfs_freevp(ap->a_vp));
453 }
454 
455 /*
456  * Return POSIX pathconf information applicable to special devices.
457  */
458 int
459 procfs_pathconf(v)
460 	void *v;
461 {
462 	struct vop_pathconf_args /* {
463 		struct vnode *a_vp;
464 		int a_name;
465 		register_t *a_retval;
466 	} */ *ap = v;
467 
468 	switch (ap->a_name) {
469 	case _PC_LINK_MAX:
470 		*ap->a_retval = LINK_MAX;
471 		return (0);
472 	case _PC_MAX_CANON:
473 		*ap->a_retval = MAX_CANON;
474 		return (0);
475 	case _PC_MAX_INPUT:
476 		*ap->a_retval = MAX_INPUT;
477 		return (0);
478 	case _PC_PIPE_BUF:
479 		*ap->a_retval = PIPE_BUF;
480 		return (0);
481 	case _PC_CHOWN_RESTRICTED:
482 		*ap->a_retval = 1;
483 		return (0);
484 	case _PC_VDISABLE:
485 		*ap->a_retval = _POSIX_VDISABLE;
486 		return (0);
487 	case _PC_SYNC_IO:
488 		*ap->a_retval = 1;
489 		return (0);
490 	default:
491 		return (EINVAL);
492 	}
493 	/* NOTREACHED */
494 }
495 
496 /*
497  * _print is used for debugging.
498  * just print a readable description
499  * of (vp).
500  */
501 int
502 procfs_print(v)
503 	void *v;
504 {
505 	struct vop_print_args /* {
506 		struct vnode *a_vp;
507 	} */ *ap = v;
508 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
509 
510 	printf("tag VT_PROCFS, type %d, pid %d, mode %x, flags %lx\n",
511 	    pfs->pfs_type, pfs->pfs_pid, pfs->pfs_mode, pfs->pfs_flags);
512 	return 0;
513 }
514 
515 int
516 procfs_link(v)
517 	void *v;
518 {
519 	struct vop_link_args /* {
520 		struct vnode *a_dvp;
521 		struct vnode *a_vp;
522 		struct componentname *a_cnp;
523 	} */ *ap = v;
524 
525 	VOP_ABORTOP(ap->a_dvp, ap->a_cnp);
526 	vput(ap->a_dvp);
527 	return (EROFS);
528 }
529 
530 int
531 procfs_symlink(v)
532 	void *v;
533 {
534 	struct vop_symlink_args /* {
535 		struct vnode *a_dvp;
536 		struct vnode **a_vpp;
537 		struct componentname *a_cnp;
538 		struct vattr *a_vap;
539 		char *a_target;
540 	} */ *ap = v;
541 
542 	VOP_ABORTOP(ap->a_dvp, ap->a_cnp);
543 	vput(ap->a_dvp);
544 	return (EROFS);
545 }
546 
547 /*
548  * Works out the path to (and vnode of) the target process's current
549  * working directory or chroot.  If the caller is in a chroot and
550  * can't "reach" the target's cwd or root (or some other error
551  * occurs), a "/" is returned for the path and a NULL pointer is
552  * returned for the vnode.
553  */
554 static void
555 procfs_dir(pfstype t, struct lwp *caller, struct proc *target, char **bpp,
556     char *path, size_t len)
557 {
558 	struct cwdinfo *cwdi;
559 	struct vnode *vp, *rvp;
560 	char *bp;
561 
562 	cwdi = caller->l_proc->p_cwdi;
563 	rw_enter(&cwdi->cwdi_lock, RW_READER);
564 
565 	rvp = cwdi->cwdi_rdir;
566 	bp = bpp ? *bpp : NULL;
567 
568 	switch (t) {
569 	case PFScwd:
570 		vp = target->p_cwdi->cwdi_cdir;
571 		break;
572 	case PFSchroot:
573 		vp = target->p_cwdi->cwdi_rdir;
574 		break;
575 	case PFSexe:
576 		vp = target->p_textvp;
577 		break;
578 	default:
579 		rw_exit(&cwdi->cwdi_lock);
580 		return;
581 	}
582 
583 	/*
584 	 * XXX: this horrible kludge avoids locking panics when
585 	 * attempting to lookup links that point to within procfs
586 	 */
587 	if (vp != NULL && vp->v_tag == VT_PROCFS) {
588 		if (bpp) {
589 			*--bp = '/';
590 			*bpp = bp;
591 		}
592 		rw_exit(&cwdi->cwdi_lock);
593 		return;
594 	}
595 
596 	if (rvp == NULL)
597 		rvp = rootvnode;
598 	if (vp == NULL || getcwd_common(vp, rvp, bp ? &bp : NULL, path,
599 	    len / 2, 0, caller) != 0) {
600 		vp = NULL;
601 		if (bpp) {
602 /*
603 			if (t == PFSexe) {
604 				snprintf(path, len, "%s/%d/file"
605 				    mp->mnt_stat.f_mntonname, pfs->pfs_pid);
606 			} else */ {
607 				bp = *bpp;
608 				*--bp = '/';
609 			}
610 		}
611 	}
612 
613 	if (bpp)
614 		*bpp = bp;
615 
616 	rw_exit(&cwdi->cwdi_lock);
617 }
618 
619 /*
620  * Invent attributes for pfsnode (vp) and store
621  * them in (vap).
622  * Directories lengths are returned as zero since
623  * any real length would require the genuine size
624  * to be computed, and nothing cares anyway.
625  *
626  * this is relatively minimal for procfs.
627  */
628 int
629 procfs_getattr(v)
630 	void *v;
631 {
632 	struct vop_getattr_args /* {
633 		struct vnode *a_vp;
634 		struct vattr *a_vap;
635 		kauth_cred_t a_cred;
636 	} */ *ap = v;
637 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
638 	struct vattr *vap = ap->a_vap;
639 	struct proc *procp;
640 	char *path;
641 	int error;
642 
643 	/* first check the process still exists */
644 	switch (pfs->pfs_type) {
645 	case PFSroot:
646 	case PFScurproc:
647 	case PFSself:
648 		procp = NULL;
649 		break;
650 
651 	default:
652 		error = procfs_proc_lock(pfs->pfs_pid, &procp, ENOENT);
653 		if (error != 0)
654 			return (error);
655 		break;
656 	}
657 
658 	switch (pfs->pfs_type) {
659 	case PFScwd:
660 	case PFSchroot:
661 	case PFSexe:
662 		MALLOC(path, char *, MAXPATHLEN + 4, M_TEMP,
663 		    M_WAITOK|M_CANFAIL);
664 		if (path == NULL && procp != NULL) {
665 			procfs_proc_unlock(procp);
666 			return (ENOMEM);
667 		}
668 		break;
669 
670 	default:
671 		path = NULL;
672 		break;
673 	}
674 
675 	if (procp != NULL) {
676 		mutex_enter(&procp->p_mutex);
677 		error = kauth_authorize_process(kauth_cred_get(),
678 		    KAUTH_PROCESS_CANSEE, procp,
679 		    KAUTH_ARG(KAUTH_REQ_PROCESS_CANSEE_ENTRY), NULL, NULL);
680 		mutex_exit(&procp->p_mutex);
681 		if (error != 0) {
682 		    	procfs_proc_unlock(procp);
683 		    	if (path != NULL)
684 		    		free(path, M_TEMP);
685 			return (ENOENT);
686 		}
687 	}
688 
689 	error = 0;
690 
691 	/* start by zeroing out the attributes */
692 	VATTR_NULL(vap);
693 
694 	/* next do all the common fields */
695 	vap->va_type = ap->a_vp->v_type;
696 	vap->va_mode = pfs->pfs_mode;
697 	vap->va_fileid = pfs->pfs_fileno;
698 	vap->va_flags = 0;
699 	vap->va_blocksize = PAGE_SIZE;
700 
701 	/*
702 	 * Make all times be current TOD.
703 	 *
704 	 * It would be possible to get the process start
705 	 * time from the p_stats structure, but there's
706 	 * no "file creation" time stamp anyway, and the
707 	 * p_stats structure is not addressable if u. gets
708 	 * swapped out for that process.
709 	 */
710 	getnanotime(&vap->va_ctime);
711 	vap->va_atime = vap->va_mtime = vap->va_ctime;
712 	if (procp)
713 		TIMEVAL_TO_TIMESPEC(&procp->p_stats->p_start,
714 		    &vap->va_birthtime);
715 	else
716 		getnanotime(&vap->va_birthtime);
717 
718 	switch (pfs->pfs_type) {
719 	case PFSmem:
720 	case PFSregs:
721 	case PFSfpregs:
722 #if defined(__HAVE_PROCFS_MACHDEP) && defined(PROCFS_MACHDEP_PROTECT_CASES)
723 	PROCFS_MACHDEP_PROTECT_CASES
724 #endif
725 		/*
726 		 * If the process has exercised some setuid or setgid
727 		 * privilege, then rip away read/write permission so
728 		 * that only root can gain access.
729 		 */
730 		if (procp->p_flag & PK_SUGID)
731 			vap->va_mode &= ~(S_IRUSR|S_IWUSR);
732 		/* FALLTHROUGH */
733 	case PFSctl:
734 	case PFSstatus:
735 	case PFSstat:
736 	case PFSnote:
737 	case PFSnotepg:
738 	case PFSmap:
739 	case PFSmaps:
740 	case PFScmdline:
741 	case PFSemul:
742 	case PFSstatm:
743 		vap->va_nlink = 1;
744 		vap->va_uid = kauth_cred_geteuid(procp->p_cred);
745 		vap->va_gid = kauth_cred_getegid(procp->p_cred);
746 		break;
747 	case PFSmeminfo:
748 	case PFSdevices:
749 	case PFScpuinfo:
750 	case PFSuptime:
751 	case PFSmounts:
752 	case PFScpustat:
753 	case PFSloadavg:
754 		vap->va_nlink = 1;
755 		vap->va_uid = vap->va_gid = 0;
756 		break;
757 
758 	default:
759 		break;
760 	}
761 
762 	/*
763 	 * now do the object specific fields
764 	 *
765 	 * The size could be set from struct reg, but it's hardly
766 	 * worth the trouble, and it puts some (potentially) machine
767 	 * dependent data into this machine-independent code.  If it
768 	 * becomes important then this function should break out into
769 	 * a per-file stat function in the corresponding .c file.
770 	 */
771 
772 	switch (pfs->pfs_type) {
773 	case PFSroot:
774 		/*
775 		 * Set nlink to 1 to tell fts(3) we don't actually know.
776 		 */
777 		vap->va_nlink = 1;
778 		vap->va_uid = 0;
779 		vap->va_gid = 0;
780 		vap->va_bytes = vap->va_size = DEV_BSIZE;
781 		break;
782 
783 	case PFSself:
784 	case PFScurproc: {
785 		char bf[16];		/* should be enough */
786 		vap->va_nlink = 1;
787 		vap->va_uid = 0;
788 		vap->va_gid = 0;
789 		vap->va_bytes = vap->va_size =
790 		    snprintf(bf, sizeof(bf), "%ld", (long)curproc->p_pid);
791 		break;
792 	}
793 
794 	case PFSfd:
795 		if (pfs->pfs_fd != -1) {
796 			struct file *fp;
797 
798 			fp = fd_getfile(procp->p_fd, pfs->pfs_fd);
799 			if (fp == NULL) {
800 				error = EBADF;
801 				break;
802 			}
803 			FILE_USE(fp);
804 			vap->va_nlink = 1;
805 			vap->va_uid = kauth_cred_geteuid(fp->f_cred);
806 			vap->va_gid = kauth_cred_getegid(fp->f_cred);
807 			switch (fp->f_type) {
808 			case DTYPE_VNODE:
809 				vap->va_bytes = vap->va_size =
810 				    ((struct vnode *)fp->f_data)->v_size;
811 				break;
812 			default:
813 				vap->va_bytes = vap->va_size = 0;
814 				break;
815 			}
816 			FILE_UNUSE(fp, curlwp);
817 			break;
818 		}
819 		/*FALLTHROUGH*/
820 	case PFSproc:
821 		vap->va_nlink = 2;
822 		vap->va_uid = kauth_cred_geteuid(procp->p_cred);
823 		vap->va_gid = kauth_cred_getegid(procp->p_cred);
824 		vap->va_bytes = vap->va_size = DEV_BSIZE;
825 		break;
826 
827 	case PFSfile:
828 		error = EOPNOTSUPP;
829 		break;
830 
831 	case PFSmem:
832 		vap->va_bytes = vap->va_size =
833 			ctob(procp->p_vmspace->vm_tsize +
834 				    procp->p_vmspace->vm_dsize +
835 				    procp->p_vmspace->vm_ssize);
836 		break;
837 
838 #if defined(PT_GETREGS) || defined(PT_SETREGS)
839 	case PFSregs:
840 		vap->va_bytes = vap->va_size = sizeof(struct reg);
841 		break;
842 #endif
843 
844 #if defined(PT_GETFPREGS) || defined(PT_SETFPREGS)
845 	case PFSfpregs:
846 		vap->va_bytes = vap->va_size = sizeof(struct fpreg);
847 		break;
848 #endif
849 
850 	case PFSctl:
851 	case PFSstatus:
852 	case PFSstat:
853 	case PFSnote:
854 	case PFSnotepg:
855 	case PFScmdline:
856 	case PFSmeminfo:
857 	case PFSdevices:
858 	case PFScpuinfo:
859 	case PFSuptime:
860 	case PFSmounts:
861 	case PFScpustat:
862 	case PFSloadavg:
863 	case PFSstatm:
864 		vap->va_bytes = vap->va_size = 0;
865 		break;
866 	case PFSmap:
867 	case PFSmaps:
868 		/*
869 		 * Advise a larger blocksize for the map files, so that
870 		 * they may be read in one pass.
871 		 */
872 		vap->va_blocksize = 4 * PAGE_SIZE;
873 		vap->va_bytes = vap->va_size = 0;
874 		break;
875 
876 	case PFScwd:
877 	case PFSchroot:
878 	case PFSexe: {
879 		char *bp;
880 
881 		vap->va_nlink = 1;
882 		vap->va_uid = 0;
883 		vap->va_gid = 0;
884 		bp = path + MAXPATHLEN;
885 		*--bp = '\0';
886 		procfs_dir(pfs->pfs_type, curlwp, procp, &bp, path,
887 		     MAXPATHLEN);
888 		vap->va_bytes = vap->va_size = strlen(bp);
889 		break;
890 	}
891 
892 	case PFSemul:
893 		vap->va_bytes = vap->va_size = strlen(procp->p_emul->e_name);
894 		break;
895 
896 #ifdef __HAVE_PROCFS_MACHDEP
897 	PROCFS_MACHDEP_NODETYPE_CASES
898 		error = procfs_machdep_getattr(ap->a_vp, vap, procp);
899 		break;
900 #endif
901 
902 	default:
903 		panic("procfs_getattr");
904 	}
905 
906 	if (procp != NULL)
907 		procfs_proc_unlock(procp);
908 	if (path != NULL)
909 		free(path, M_TEMP);
910 
911 	return (error);
912 }
913 
914 /*ARGSUSED*/
915 int
916 procfs_setattr(void *v)
917 {
918 	/*
919 	 * just fake out attribute setting
920 	 * it's not good to generate an error
921 	 * return, otherwise things like creat()
922 	 * will fail when they try to set the
923 	 * file length to 0.  worse, this means
924 	 * that echo $note > /proc/$pid/note will fail.
925 	 */
926 
927 	return (0);
928 }
929 
930 /*
931  * implement access checking.
932  *
933  * actually, the check for super-user is slightly
934  * broken since it will allow read access to write-only
935  * objects.  this doesn't cause any particular trouble
936  * but does mean that the i/o entry points need to check
937  * that the operation really does make sense.
938  */
939 int
940 procfs_access(v)
941 	void *v;
942 {
943 	struct vop_access_args /* {
944 		struct vnode *a_vp;
945 		int a_mode;
946 		kauth_cred_t a_cred;
947 	} */ *ap = v;
948 	struct vattr va;
949 	int error;
950 
951 	if ((error = VOP_GETATTR(ap->a_vp, &va, ap->a_cred)) != 0)
952 		return (error);
953 
954 	return (vaccess(va.va_type, va.va_mode,
955 	    va.va_uid, va.va_gid, ap->a_mode, ap->a_cred));
956 }
957 
958 /*
959  * lookup.  this is incredibly complicated in the
960  * general case, however for most pseudo-filesystems
961  * very little needs to be done.
962  *
963  * Locking isn't hard here, just poorly documented.
964  *
965  * If we're looking up ".", just vref the parent & return it.
966  *
967  * If we're looking up "..", unlock the parent, and lock "..". If everything
968  * went ok, and we're on the last component and the caller requested the
969  * parent locked, try to re-lock the parent. We do this to prevent lock
970  * races.
971  *
972  * For anything else, get the needed node. Then unlock the parent if not
973  * the last component or not LOCKPARENT (i.e. if we wouldn't re-lock the
974  * parent in the .. case).
975  *
976  * We try to exit with the parent locked in error cases.
977  */
978 int
979 procfs_lookup(v)
980 	void *v;
981 {
982 	struct vop_lookup_args /* {
983 		struct vnode * a_dvp;
984 		struct vnode ** a_vpp;
985 		struct componentname * a_cnp;
986 	} */ *ap = v;
987 	struct componentname *cnp = ap->a_cnp;
988 	struct vnode **vpp = ap->a_vpp;
989 	struct vnode *dvp = ap->a_dvp;
990 	const char *pname = cnp->cn_nameptr;
991 	const struct proc_target *pt = NULL;
992 	struct vnode *fvp;
993 	pid_t pid, vnpid;
994 	struct pfsnode *pfs;
995 	struct proc *p = NULL;
996 	struct lwp *l = NULL;
997 	int i, error;
998 	pfstype type;
999 
1000 	*vpp = NULL;
1001 
1002 	if (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)
1003 		return (EROFS);
1004 
1005 	if (cnp->cn_namelen == 1 && *pname == '.') {
1006 		*vpp = dvp;
1007 		VREF(dvp);
1008 		return (0);
1009 	}
1010 
1011 	pfs = VTOPFS(dvp);
1012 	switch (pfs->pfs_type) {
1013 	case PFSroot:
1014 		/*
1015 		 * Shouldn't get here with .. in the root node.
1016 		 */
1017 		if (cnp->cn_flags & ISDOTDOT)
1018 			return (EIO);
1019 
1020 		for (i = 0; i < nproc_root_targets; i++) {
1021 			pt = &proc_root_targets[i];
1022 			/*
1023 			 * check for node match.  proc is always NULL here,
1024 			 * so call pt_valid with constant NULL lwp.
1025 			 */
1026 			if (cnp->cn_namelen == pt->pt_namlen &&
1027 			    memcmp(pt->pt_name, pname, cnp->cn_namelen) == 0 &&
1028 			    (pt->pt_valid == NULL ||
1029 			     (*pt->pt_valid)(NULL, dvp->v_mount)))
1030 				break;
1031 		}
1032 
1033 		if (i != nproc_root_targets) {
1034 			error = procfs_allocvp(dvp->v_mount, vpp, 0,
1035 			    pt->pt_pfstype, -1, NULL);
1036 			return (error);
1037 		}
1038 
1039 		if (CNEQ(cnp, "curproc", 7)) {
1040 			pid = curproc->p_pid;
1041 			vnpid = 0;
1042 			type = PFScurproc;
1043 		} else if (CNEQ(cnp, "self", 4)) {
1044 			pid = curproc->p_pid;
1045 			vnpid = 0;
1046 			type = PFSself;
1047 		} else {
1048 			pid = (pid_t)atoi(pname, cnp->cn_namelen);
1049 			vnpid = pid;
1050 			type = PFSproc;
1051 		}
1052 
1053 		if (procfs_proc_lock(pid, &p, ESRCH) != 0)
1054 			break;
1055 		error = procfs_allocvp(dvp->v_mount, vpp, vnpid, type, -1, p);
1056 		procfs_proc_unlock(p);
1057 		return (error);
1058 
1059 	case PFSproc:
1060 		/*
1061 		 * do the .. dance. We unlock the directory, and then
1062 		 * get the root dir. That will automatically return ..
1063 		 * locked. Then if the caller wanted dvp locked, we
1064 		 * re-lock.
1065 		 */
1066 		if (cnp->cn_flags & ISDOTDOT) {
1067 			VOP_UNLOCK(dvp, 0);
1068 			error = procfs_root(dvp->v_mount, vpp);
1069 			vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1070 			return (error);
1071 		}
1072 
1073 		if (procfs_proc_lock(pfs->pfs_pid, &p, ESRCH) != 0)
1074 			break;
1075 
1076 		for (pt = proc_targets, i = 0; i < nproc_targets; pt++, i++) {
1077 			struct lwp *plwp;
1078 			int found;
1079 
1080 			mutex_enter(&p->p_smutex);
1081 			plwp = proc_representative_lwp(p, NULL, 1);
1082 			lwp_addref(plwp);
1083 			mutex_exit(&p->p_smutex);
1084 			found = cnp->cn_namelen == pt->pt_namlen &&
1085 			    memcmp(pt->pt_name, pname, cnp->cn_namelen) == 0 &&
1086 			    (pt->pt_valid == NULL
1087 			      || (*pt->pt_valid)(plwp, dvp->v_mount));
1088 			lwp_delref(plwp);
1089 			if (found)
1090 				break;
1091 		}
1092 		if (i == nproc_targets) {
1093 			procfs_proc_unlock(p);
1094 			break;
1095 		}
1096 		if (pt->pt_pfstype == PFSfile) {
1097 			fvp = p->p_textvp;
1098 			/* We already checked that it exists. */
1099 			VREF(fvp);
1100 			procfs_proc_unlock(p);
1101 			vn_lock(fvp, LK_EXCLUSIVE | LK_RETRY);
1102 			*vpp = fvp;
1103 			return (0);
1104 		}
1105 
1106 		error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
1107 		    pt->pt_pfstype, -1, p);
1108 		procfs_proc_unlock(p);
1109 		return (error);
1110 
1111 	case PFSfd: {
1112 		int fd;
1113 		struct file *fp;
1114 
1115 		if ((error = procfs_proc_lock(pfs->pfs_pid, &p, ENOENT)) != 0)
1116 			return error;
1117 
1118 		/*
1119 		 * do the .. dance. We unlock the directory, and then
1120 		 * get the proc dir. That will automatically return ..
1121 		 * locked. Then re-lock the directory.
1122 		 */
1123 		if (cnp->cn_flags & ISDOTDOT) {
1124 			VOP_UNLOCK(dvp, 0);
1125 			error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
1126 			    PFSproc, -1, p);
1127 			procfs_proc_unlock(p);
1128 			vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1129 			return (error);
1130 		}
1131 		fd = atoi(pname, cnp->cn_namelen);
1132 
1133 		fp = fd_getfile(p->p_fd, fd);
1134 		if (fp == NULL) {
1135 			procfs_proc_unlock(p);
1136 			return ENOENT;
1137 		}
1138 
1139 		FILE_USE(fp);
1140 		fvp = (struct vnode *)fp->f_data;
1141 
1142 		/* Don't show directories */
1143 		if (fp->f_type == DTYPE_VNODE && fvp->v_type != VDIR) {
1144 			VREF(fvp);
1145 			FILE_UNUSE(fp, l);
1146 			procfs_proc_unlock(p);
1147 			vn_lock(fvp, LK_EXCLUSIVE | LK_RETRY |
1148 			    (p == curproc ? LK_CANRECURSE : 0));
1149 			*vpp = fvp;
1150 			return 0;
1151 		}
1152 
1153 		FILE_UNUSE(fp, l);
1154 		error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
1155 		    PFSfd, fd, p);
1156 		procfs_proc_unlock(p);
1157 		return error;
1158 	}
1159 	default:
1160 		return (ENOTDIR);
1161 	}
1162 
1163 	return (cnp->cn_nameiop == LOOKUP ? ENOENT : EROFS);
1164 }
1165 
1166 int
1167 procfs_validfile(struct lwp *l, struct mount *mp)
1168 {
1169 	return l != NULL && l->l_proc != NULL && l->l_proc->p_textvp != NULL;
1170 }
1171 
1172 static int
1173 procfs_validfile_linux(l, mp)
1174 	struct lwp *l;
1175 	struct mount *mp;
1176 {
1177 	int flags;
1178 
1179 	flags = VFSTOPROC(mp)->pmnt_flags;
1180 	return (flags & PROCFSMNT_LINUXCOMPAT) &&
1181 	    (l == NULL || l->l_proc == NULL || procfs_validfile(l, mp));
1182 }
1183 
1184 struct procfs_root_readdir_ctx {
1185 	struct uio *uiop;
1186 	off_t *cookies;
1187 	int ncookies;
1188 	off_t off;
1189 	off_t startoff;
1190 	int error;
1191 };
1192 
1193 static int
1194 procfs_root_readdir_callback(struct proc *p, void *arg)
1195 {
1196 	struct procfs_root_readdir_ctx *ctxp = arg;
1197 	struct dirent d;
1198 	struct uio *uiop;
1199 	int error;
1200 
1201 	uiop = ctxp->uiop;
1202 	if (uiop->uio_resid < UIO_MX)
1203 		return -1; /* no space */
1204 
1205 	if (ctxp->off < ctxp->startoff) {
1206 		ctxp->off++;
1207 		return 0;
1208 	}
1209 
1210 	if (kauth_authorize_process(kauth_cred_get(),
1211 	    KAUTH_PROCESS_CANSEE, p,
1212 	    KAUTH_ARG(KAUTH_REQ_PROCESS_CANSEE_ENTRY), NULL, NULL) != 0)
1213 		return 0;
1214 
1215 	memset(&d, 0, UIO_MX);
1216 	d.d_reclen = UIO_MX;
1217 	d.d_fileno = PROCFS_FILENO(p->p_pid, PFSproc, -1);
1218 	d.d_namlen = snprintf(d.d_name,
1219 	    UIO_MX - offsetof(struct dirent, d_name), "%ld", (long)p->p_pid);
1220 	d.d_type = DT_DIR;
1221 
1222 	mutex_exit(&proclist_lock);
1223 	error = uiomove(&d, UIO_MX, uiop);
1224 	mutex_enter(&proclist_lock);
1225 	if (error) {
1226 		ctxp->error = error;
1227 		return -1;
1228 	}
1229 
1230 	ctxp->ncookies++;
1231 	if (ctxp->cookies)
1232 		*(ctxp->cookies)++ = ctxp->off + 1;
1233 	ctxp->off++;
1234 
1235 	return 0;
1236 }
1237 
1238 /*
1239  * readdir returns directory entries from pfsnode (vp).
1240  *
1241  * the strategy here with procfs is to generate a single
1242  * directory entry at a time (struct dirent) and then
1243  * copy that out to userland using uiomove.  a more efficent
1244  * though more complex implementation, would try to minimize
1245  * the number of calls to uiomove().  for procfs, this is
1246  * hardly worth the added code complexity.
1247  *
1248  * this should just be done through read()
1249  */
1250 int
1251 procfs_readdir(v)
1252 	void *v;
1253 {
1254 	struct vop_readdir_args /* {
1255 		struct vnode *a_vp;
1256 		struct uio *a_uio;
1257 		kauth_cred_t a_cred;
1258 		int *a_eofflag;
1259 		off_t **a_cookies;
1260 		int *a_ncookies;
1261 	} */ *ap = v;
1262 	struct uio *uio = ap->a_uio;
1263 	struct dirent d;
1264 	struct pfsnode *pfs;
1265 	off_t i;
1266 	int error;
1267 	off_t *cookies = NULL;
1268 	int ncookies;
1269 	struct vnode *vp;
1270 	const struct proc_target *pt;
1271 	struct procfs_root_readdir_ctx ctx;
1272 	struct lwp *l;
1273 	int nfd;
1274 
1275 	vp = ap->a_vp;
1276 	pfs = VTOPFS(vp);
1277 
1278 	if (uio->uio_resid < UIO_MX)
1279 		return (EINVAL);
1280 	if (uio->uio_offset < 0)
1281 		return (EINVAL);
1282 
1283 	error = 0;
1284 	i = uio->uio_offset;
1285 	memset(&d, 0, UIO_MX);
1286 	d.d_reclen = UIO_MX;
1287 	ncookies = uio->uio_resid / UIO_MX;
1288 
1289 	switch (pfs->pfs_type) {
1290 	/*
1291 	 * this is for the process-specific sub-directories.
1292 	 * all that is needed to is copy out all the entries
1293 	 * from the procent[] table (top of this file).
1294 	 */
1295 	case PFSproc: {
1296 		struct proc *p;
1297 
1298 		if (i >= nproc_targets)
1299 			return 0;
1300 
1301 		if (procfs_proc_lock(pfs->pfs_pid, &p, ESRCH) != 0)
1302 			break;
1303 
1304 		if (ap->a_ncookies) {
1305 			ncookies = min(ncookies, (nproc_targets - i));
1306 			cookies = malloc(ncookies * sizeof (off_t),
1307 			    M_TEMP, M_WAITOK);
1308 			*ap->a_cookies = cookies;
1309 		}
1310 
1311 		for (pt = &proc_targets[i];
1312 		     uio->uio_resid >= UIO_MX && i < nproc_targets; pt++, i++) {
1313 			if (pt->pt_valid) {
1314 				/* XXX LWP can disappear */
1315 				mutex_enter(&p->p_smutex);
1316 				l = proc_representative_lwp(p, NULL, 1);
1317 				mutex_exit(&p->p_smutex);
1318 				if ((*pt->pt_valid)(l, vp->v_mount) == 0)
1319 					continue;
1320 			}
1321 
1322 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid,
1323 			    pt->pt_pfstype, -1);
1324 			d.d_namlen = pt->pt_namlen;
1325 			memcpy(d.d_name, pt->pt_name, pt->pt_namlen + 1);
1326 			d.d_type = pt->pt_type;
1327 
1328 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1329 				break;
1330 			if (cookies)
1331 				*cookies++ = i + 1;
1332 		}
1333 
1334 		procfs_proc_unlock(p);
1335 	    	break;
1336 	}
1337 	case PFSfd: {
1338 		struct proc *p;
1339 		struct filedesc	*fdp;
1340 		struct file *fp;
1341 		int lim, nc = 0;
1342 
1343 		if ((error = procfs_proc_lock(pfs->pfs_pid, &p, ESRCH)) != 0)
1344 			return error;
1345 
1346 		/* XXX Should this be by file as well? */
1347 		if (kauth_authorize_process(kauth_cred_get(),
1348 		    KAUTH_PROCESS_CANSEE, p,
1349 		    KAUTH_ARG(KAUTH_REQ_PROCESS_CANSEE_OPENFILES), NULL,
1350 		    NULL) != 0) {
1351 		    	procfs_proc_unlock(p);
1352 			return ESRCH;
1353 		}
1354 
1355 		fdp = p->p_fd;
1356 		nfd = fdp->fd_nfiles;
1357 
1358 		lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfiles);
1359 		if (i >= lim) {
1360 		    	procfs_proc_unlock(p);
1361 			return 0;
1362 		}
1363 
1364 		if (ap->a_ncookies) {
1365 			ncookies = min(ncookies, (nfd + 2 - i));
1366 			cookies = malloc(ncookies * sizeof (off_t),
1367 			    M_TEMP, M_WAITOK);
1368 			*ap->a_cookies = cookies;
1369 		}
1370 
1371 		for (; i < 2 && uio->uio_resid >= UIO_MX; i++) {
1372 			pt = &proc_targets[i];
1373 			d.d_namlen = pt->pt_namlen;
1374 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid,
1375 			    pt->pt_pfstype, -1);
1376 			(void)memcpy(d.d_name, pt->pt_name, pt->pt_namlen + 1);
1377 			d.d_type = pt->pt_type;
1378 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1379 				break;
1380 			if (cookies)
1381 				*cookies++ = i + 1;
1382 			nc++;
1383 		}
1384 		if (error) {
1385 			ncookies = nc;
1386 			break;
1387 		}
1388 		for (; uio->uio_resid >= UIO_MX && i < nfd; i++) {
1389 			/* check the descriptor exists */
1390 			if ((fp = fd_getfile(fdp, i - 2)) == NULL)
1391 				continue;
1392 			mutex_exit(&fp->f_lock);
1393 
1394 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid, PFSfd, i - 2);
1395 			d.d_namlen = snprintf(d.d_name, sizeof(d.d_name),
1396 			    "%lld", (long long)(i - 2));
1397 			d.d_type = VREG;
1398 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1399 				break;
1400 			if (cookies)
1401 				*cookies++ = i + 1;
1402 			nc++;
1403 		}
1404 		ncookies = nc;
1405 		procfs_proc_unlock(p);
1406 		break;
1407 	}
1408 
1409 	/*
1410 	 * this is for the root of the procfs filesystem
1411 	 * what is needed are special entries for "curproc"
1412 	 * and "self" followed by an entry for each process
1413 	 * on allproc.
1414 	 */
1415 
1416 	case PFSroot: {
1417 		int nc = 0;
1418 
1419 		if (ap->a_ncookies) {
1420 			/*
1421 			 * XXX Potentially allocating too much space here,
1422 			 * but I'm lazy. This loop needs some work.
1423 			 */
1424 			cookies = malloc(ncookies * sizeof (off_t),
1425 			    M_TEMP, M_WAITOK);
1426 			*ap->a_cookies = cookies;
1427 		}
1428 		error = 0;
1429 		/* 0 ... 3 are static entries. */
1430 		for (; i <= 3 && uio->uio_resid >= UIO_MX; i++) {
1431 			switch (i) {
1432 			case 0:		/* `.' */
1433 			case 1:		/* `..' */
1434 				d.d_fileno = PROCFS_FILENO(0, PFSroot, -1);
1435 				d.d_namlen = i + 1;
1436 				memcpy(d.d_name, "..", d.d_namlen);
1437 				d.d_name[i + 1] = '\0';
1438 				d.d_type = DT_DIR;
1439 				break;
1440 
1441 			case 2:
1442 				d.d_fileno = PROCFS_FILENO(0, PFScurproc, -1);
1443 				d.d_namlen = sizeof("curproc") - 1;
1444 				memcpy(d.d_name, "curproc", sizeof("curproc"));
1445 				d.d_type = DT_LNK;
1446 				break;
1447 
1448 			case 3:
1449 				d.d_fileno = PROCFS_FILENO(0, PFSself, -1);
1450 				d.d_namlen = sizeof("self") - 1;
1451 				memcpy(d.d_name, "self", sizeof("self"));
1452 				d.d_type = DT_LNK;
1453 				break;
1454 			}
1455 
1456 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1457 				break;
1458 			nc++;
1459 			if (cookies)
1460 				*cookies++ = i + 1;
1461 		}
1462 		/* 4 ... are process entries. */
1463 		ctx.uiop = uio;
1464 		ctx.error = 0;
1465 		ctx.off = 4;
1466 		ctx.startoff = i;
1467 		ctx.cookies = cookies;
1468 		ctx.ncookies = nc;
1469 		proclist_foreach_call(&allproc,
1470 		    procfs_root_readdir_callback, &ctx);
1471 		cookies = ctx.cookies;
1472 		nc = ctx.ncookies;
1473 		error = ctx.error;
1474 		if (error)
1475 			break;
1476 
1477 		/* misc entries. */
1478 		if (i < ctx.off)
1479 			i = ctx.off;
1480 		if (i >= ctx.off + nproc_root_targets)
1481 			break;
1482 		for (pt = &proc_root_targets[i - ctx.off];
1483 		    uio->uio_resid >= UIO_MX &&
1484 		    pt < &proc_root_targets[nproc_root_targets];
1485 		    pt++, i++) {
1486 			if (pt->pt_valid &&
1487 			    (*pt->pt_valid)(NULL, vp->v_mount) == 0)
1488 				continue;
1489 			d.d_fileno = PROCFS_FILENO(0, pt->pt_pfstype, -1);
1490 			d.d_namlen = pt->pt_namlen;
1491 			memcpy(d.d_name, pt->pt_name, pt->pt_namlen + 1);
1492 			d.d_type = pt->pt_type;
1493 
1494 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1495 				break;
1496 			nc++;
1497 			if (cookies)
1498 				*cookies++ = i + 1;
1499 		}
1500 
1501 		ncookies = nc;
1502 		break;
1503 	}
1504 
1505 	default:
1506 		error = ENOTDIR;
1507 		break;
1508 	}
1509 
1510 	if (ap->a_ncookies) {
1511 		if (error) {
1512 			if (cookies)
1513 				free(*ap->a_cookies, M_TEMP);
1514 			*ap->a_ncookies = 0;
1515 			*ap->a_cookies = NULL;
1516 		} else
1517 			*ap->a_ncookies = ncookies;
1518 	}
1519 	uio->uio_offset = i;
1520 	return (error);
1521 }
1522 
1523 /*
1524  * readlink reads the link of `curproc' and others
1525  */
1526 int
1527 procfs_readlink(v)
1528 	void *v;
1529 {
1530 	struct vop_readlink_args *ap = v;
1531 	char bf[16];		/* should be enough */
1532 	char *bp = bf;
1533 	char *path = NULL;
1534 	int len = 0;
1535 	int error = 0;
1536 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
1537 	struct proc *pown;
1538 
1539 	if (pfs->pfs_fileno == PROCFS_FILENO(0, PFScurproc, -1))
1540 		len = snprintf(bf, sizeof(bf), "%ld", (long)curproc->p_pid);
1541 	else if (pfs->pfs_fileno == PROCFS_FILENO(0, PFSself, -1))
1542 		len = snprintf(bf, sizeof(bf), "%s", "curproc");
1543 	else if (pfs->pfs_fileno == PROCFS_FILENO(pfs->pfs_pid, PFScwd, -1) ||
1544 	    pfs->pfs_fileno == PROCFS_FILENO(pfs->pfs_pid, PFSchroot, -1) ||
1545 	    pfs->pfs_fileno == PROCFS_FILENO(pfs->pfs_pid, PFSexe, -1)) {
1546 		if ((error = procfs_proc_lock(pfs->pfs_pid, &pown, ESRCH)) != 0)
1547 			return error;
1548 		MALLOC(path, char *, MAXPATHLEN + 4, M_TEMP,
1549 		    M_WAITOK|M_CANFAIL);
1550 		if (path == NULL) {
1551 			procfs_proc_unlock(pown);
1552 			return (ENOMEM);
1553 		}
1554 		bp = path + MAXPATHLEN;
1555 		*--bp = '\0';
1556 		procfs_dir(PROCFS_TYPE(pfs->pfs_fileno), curlwp, pown,
1557 		    &bp, path, MAXPATHLEN);
1558 		procfs_proc_unlock(pown);
1559 		len = strlen(bp);
1560 	} else {
1561 		struct file *fp;
1562 		struct vnode *vxp, *vp;
1563 
1564 		if ((error = procfs_proc_lock(pfs->pfs_pid, &pown, ESRCH)) != 0)
1565 			return error;
1566 
1567 		fp = fd_getfile(pown->p_fd, pfs->pfs_fd);
1568 		if (fp == NULL) {
1569 			procfs_proc_unlock(pown);
1570 			return EBADF;
1571 		}
1572 
1573 		FILE_USE(fp);
1574 		switch (fp->f_type) {
1575 		case DTYPE_VNODE:
1576 			vxp = (struct vnode *)fp->f_data;
1577 			if (vxp->v_type != VDIR) {
1578 				FILE_UNUSE(fp, curlwp);
1579 				error = EINVAL;
1580 				break;
1581 			}
1582 			if ((path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK))
1583 			    == NULL) {
1584 				FILE_UNUSE(fp, curlwp);
1585 				error = ENOMEM;
1586 				break;
1587 			}
1588 			bp = path + MAXPATHLEN;
1589 			*--bp = '\0';
1590 
1591 			/*
1592 			 * XXX: kludge to avoid locking against ourselves
1593 			 * in getcwd()
1594 			 */
1595 			if (vxp->v_tag == VT_PROCFS) {
1596 				*--bp = '/';
1597 			} else {
1598 				rw_enter(&curproc->p_cwdi->cwdi_lock, RW_READER);
1599 				vp = curproc->p_cwdi->cwdi_rdir;
1600 				if (vp == NULL)
1601 					vp = rootvnode;
1602 				error = getcwd_common(vxp, vp, &bp, path,
1603 				    MAXPATHLEN / 2, 0, curlwp);
1604 				rw_exit(&curproc->p_cwdi->cwdi_lock);
1605 			}
1606 			if (error)
1607 				break;
1608 			len = strlen(bp);
1609 			break;
1610 
1611 		case DTYPE_MISC:
1612 			len = snprintf(bf, sizeof(bf), "%s", "[misc]");
1613 			break;
1614 
1615 		case DTYPE_KQUEUE:
1616 			len = snprintf(bf, sizeof(bf), "%s", "[kqueue]");
1617 			break;
1618 
1619 		default:
1620 			error = EINVAL;
1621 			break;
1622 		}
1623 		FILE_UNUSE(fp, curlwp);
1624 		procfs_proc_unlock(pown);
1625 	}
1626 
1627 	if (error == 0)
1628 		error = uiomove(bp, len, ap->a_uio);
1629 	if (path)
1630 		free(path, M_TEMP);
1631 	return error;
1632 }
1633 
1634 /*
1635  * convert decimal ascii to int
1636  */
1637 static int
1638 atoi(b, len)
1639 	const char *b;
1640 	size_t len;
1641 {
1642 	int p = 0;
1643 
1644 	while (len--) {
1645 		char c = *b++;
1646 		if (c < '0' || c > '9')
1647 			return -1;
1648 		p = 10 * p + (c - '0');
1649 	}
1650 
1651 	return p;
1652 }
1653