xref: /netbsd-src/sys/miscfs/procfs/procfs_vnops.c (revision c9496f6b604074a9451a67df576a5b423068e71e)
1 /*	$NetBSD: procfs_vnops.c,v 1.202 2017/12/31 03:02:23 christos 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 1.202 2017/12/31 03:02:23 christos Exp $");
109 
110 #include <sys/param.h>
111 #include <sys/systm.h>
112 #include <sys/time.h>
113 #include <sys/kernel.h>
114 #include <sys/file.h>
115 #include <sys/filedesc.h>
116 #include <sys/proc.h>
117 #include <sys/vnode.h>
118 #include <sys/namei.h>
119 #include <sys/malloc.h>
120 #include <sys/mount.h>
121 #include <sys/dirent.h>
122 #include <sys/resourcevar.h>
123 #include <sys/stat.h>
124 #include <sys/ptrace.h>
125 #include <sys/kauth.h>
126 #include <sys/exec.h>
127 
128 #include <uvm/uvm_extern.h>	/* for PAGE_SIZE */
129 
130 #include <machine/reg.h>
131 
132 #include <miscfs/genfs/genfs.h>
133 #include <miscfs/procfs/procfs.h>
134 
135 /*
136  * Vnode Operations.
137  *
138  */
139 
140 static int procfs_validfile_linux(struct lwp *, struct mount *);
141 static int procfs_root_readdir_callback(struct proc *, void *);
142 static void procfs_dir(pfstype, struct lwp *, struct proc *, char **, char *,
143     size_t);
144 
145 /*
146  * This is a list of the valid names in the
147  * process-specific sub-directories.  It is
148  * used in procfs_lookup and procfs_readdir
149  */
150 static const struct proc_target {
151 	u_char	pt_type;
152 	u_char	pt_namlen;
153 	const char	*pt_name;
154 	pfstype	pt_pfstype;
155 	int	(*pt_valid)(struct lwp *, struct mount *);
156 } proc_targets[] = {
157 #define N(s) sizeof(s)-1, s
158 	/*	  name		type		validp */
159 	{ DT_DIR, N("."),	PFSproc,	NULL },
160 	{ DT_DIR, N(".."),	PFSroot,	NULL },
161 	{ DT_DIR, N("fd"),	PFSfd,		NULL },
162 	{ DT_REG, N("file"),	PFSfile,	procfs_validfile },
163 	{ DT_REG, N("auxv"),	PFSauxv,	procfs_validauxv },
164 	{ DT_REG, N("mem"),	PFSmem,		NULL },
165 	{ DT_REG, N("regs"),	PFSregs,	procfs_validregs },
166 	{ DT_REG, N("fpregs"),	PFSfpregs,	procfs_validfpregs },
167 	{ DT_REG, N("stat"),	PFSstat,	procfs_validfile_linux },
168 	{ DT_REG, N("status"),	PFSstatus,	NULL },
169 	{ DT_REG, N("note"),	PFSnote,	NULL },
170 	{ DT_REG, N("notepg"),	PFSnotepg,	NULL },
171 	{ DT_REG, N("map"),	PFSmap,		procfs_validmap },
172 	{ DT_REG, N("maps"),	PFSmaps,	procfs_validmap },
173 	{ DT_REG, N("cmdline"), PFScmdline,	NULL },
174 	{ DT_REG, N("environ"), PFSenviron,	NULL },
175 	{ DT_REG, N("exe"),	PFSexe,		procfs_validfile },
176 	{ DT_LNK, N("cwd"),	PFScwd,		NULL },
177 	{ DT_LNK, N("root"),	PFSchroot,	NULL },
178 	{ DT_LNK, N("emul"),	PFSemul,	NULL },
179 	{ DT_REG, N("statm"),	PFSstatm,	procfs_validfile_linux },
180 	{ DT_DIR, N("task"),	PFStask,	procfs_validfile_linux },
181 #ifdef __HAVE_PROCFS_MACHDEP
182 	PROCFS_MACHDEP_NODETYPE_DEFNS
183 #endif
184 #undef N
185 };
186 static const int nproc_targets = sizeof(proc_targets) / sizeof(proc_targets[0]);
187 
188 /*
189  * List of files in the root directory. Note: the validate function will
190  * be called with p == NULL for these ones.
191  */
192 static const struct proc_target proc_root_targets[] = {
193 #define N(s) sizeof(s)-1, s
194 	/*	  name		    type	    validp */
195 	{ DT_REG, N("meminfo"),     PFSmeminfo,        procfs_validfile_linux },
196 	{ DT_REG, N("cpuinfo"),     PFScpuinfo,        procfs_validfile_linux },
197 	{ DT_REG, N("uptime"),      PFSuptime,         procfs_validfile_linux },
198 	{ DT_REG, N("mounts"),	    PFSmounts,	       procfs_validfile_linux },
199 	{ DT_REG, N("devices"),     PFSdevices,        procfs_validfile_linux },
200 	{ DT_REG, N("stat"),	    PFScpustat,        procfs_validfile_linux },
201 	{ DT_REG, N("loadavg"),	    PFSloadavg,        procfs_validfile_linux },
202 	{ DT_REG, N("version"),     PFSversion,        procfs_validfile_linux },
203 #undef N
204 };
205 static const int nproc_root_targets =
206     sizeof(proc_root_targets) / sizeof(proc_root_targets[0]);
207 
208 int	procfs_lookup(void *);
209 #define	procfs_create	genfs_eopnotsupp
210 #define	procfs_mknod	genfs_eopnotsupp
211 int	procfs_open(void *);
212 int	procfs_close(void *);
213 int	procfs_access(void *);
214 int	procfs_getattr(void *);
215 int	procfs_setattr(void *);
216 #define	procfs_read	procfs_rw
217 #define	procfs_write	procfs_rw
218 #define	procfs_fcntl	genfs_fcntl
219 #define	procfs_ioctl	genfs_enoioctl
220 #define	procfs_poll	genfs_poll
221 #define	procfs_kqfilter	genfs_kqfilter
222 #define	procfs_revoke	genfs_revoke
223 #define	procfs_fsync	genfs_nullop
224 #define	procfs_seek	genfs_nullop
225 #define	procfs_remove	genfs_eopnotsupp
226 int	procfs_link(void *);
227 #define	procfs_rename	genfs_eopnotsupp
228 #define	procfs_mkdir	genfs_eopnotsupp
229 #define	procfs_rmdir	genfs_eopnotsupp
230 int	procfs_symlink(void *);
231 int	procfs_readdir(void *);
232 int	procfs_readlink(void *);
233 #define	procfs_abortop	genfs_abortop
234 int	procfs_inactive(void *);
235 int	procfs_reclaim(void *);
236 #define	procfs_lock	genfs_lock
237 #define	procfs_unlock	genfs_unlock
238 #define	procfs_bmap	genfs_badop
239 #define	procfs_strategy	genfs_badop
240 int	procfs_print(void *);
241 int	procfs_pathconf(void *);
242 #define	procfs_islocked	genfs_islocked
243 #define	procfs_advlock	genfs_einval
244 #define	procfs_bwrite	genfs_eopnotsupp
245 #define procfs_putpages	genfs_null_putpages
246 
247 static int atoi(const char *, size_t);
248 
249 /*
250  * procfs vnode operations.
251  */
252 int (**procfs_vnodeop_p)(void *);
253 const struct vnodeopv_entry_desc procfs_vnodeop_entries[] = {
254 	{ &vop_default_desc, vn_default_error },
255 	{ &vop_lookup_desc, procfs_lookup },		/* lookup */
256 	{ &vop_create_desc, procfs_create },		/* create */
257 	{ &vop_mknod_desc, procfs_mknod },		/* mknod */
258 	{ &vop_open_desc, procfs_open },		/* open */
259 	{ &vop_close_desc, procfs_close },		/* close */
260 	{ &vop_access_desc, procfs_access },		/* access */
261 	{ &vop_getattr_desc, procfs_getattr },		/* getattr */
262 	{ &vop_setattr_desc, procfs_setattr },		/* setattr */
263 	{ &vop_read_desc, procfs_read },		/* read */
264 	{ &vop_write_desc, procfs_write },		/* write */
265 	{ &vop_fallocate_desc, genfs_eopnotsupp },	/* fallocate */
266 	{ &vop_fdiscard_desc, genfs_eopnotsupp },	/* fdiscard */
267 	{ &vop_fcntl_desc, procfs_fcntl },		/* fcntl */
268 	{ &vop_ioctl_desc, procfs_ioctl },		/* ioctl */
269 	{ &vop_poll_desc, procfs_poll },		/* poll */
270 	{ &vop_kqfilter_desc, procfs_kqfilter },	/* kqfilter */
271 	{ &vop_revoke_desc, procfs_revoke },		/* revoke */
272 	{ &vop_fsync_desc, procfs_fsync },		/* fsync */
273 	{ &vop_seek_desc, procfs_seek },		/* seek */
274 	{ &vop_remove_desc, procfs_remove },		/* remove */
275 	{ &vop_link_desc, procfs_link },		/* link */
276 	{ &vop_rename_desc, procfs_rename },		/* rename */
277 	{ &vop_mkdir_desc, procfs_mkdir },		/* mkdir */
278 	{ &vop_rmdir_desc, procfs_rmdir },		/* rmdir */
279 	{ &vop_symlink_desc, procfs_symlink },		/* symlink */
280 	{ &vop_readdir_desc, procfs_readdir },		/* readdir */
281 	{ &vop_readlink_desc, procfs_readlink },	/* readlink */
282 	{ &vop_abortop_desc, procfs_abortop },		/* abortop */
283 	{ &vop_inactive_desc, procfs_inactive },	/* inactive */
284 	{ &vop_reclaim_desc, procfs_reclaim },		/* reclaim */
285 	{ &vop_lock_desc, procfs_lock },		/* lock */
286 	{ &vop_unlock_desc, procfs_unlock },		/* unlock */
287 	{ &vop_bmap_desc, procfs_bmap },		/* bmap */
288 	{ &vop_strategy_desc, procfs_strategy },	/* strategy */
289 	{ &vop_print_desc, procfs_print },		/* print */
290 	{ &vop_islocked_desc, procfs_islocked },	/* islocked */
291 	{ &vop_pathconf_desc, procfs_pathconf },	/* pathconf */
292 	{ &vop_advlock_desc, procfs_advlock },		/* advlock */
293 	{ &vop_putpages_desc, procfs_putpages },	/* putpages */
294 	{ NULL, NULL }
295 };
296 const struct vnodeopv_desc procfs_vnodeop_opv_desc =
297 	{ &procfs_vnodeop_p, procfs_vnodeop_entries };
298 /*
299  * set things up for doing i/o on
300  * the pfsnode (vp).  (vp) is locked
301  * on entry, and should be left locked
302  * on exit.
303  *
304  * for procfs we don't need to do anything
305  * in particular for i/o.  all that is done
306  * is to support exclusive open on process
307  * memory images.
308  */
309 int
310 procfs_open(void *v)
311 {
312 	struct vop_open_args /* {
313 		struct vnode *a_vp;
314 		int  a_mode;
315 		kauth_cred_t a_cred;
316 	} */ *ap = v;
317 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
318 	struct lwp *l1;
319 	struct proc *p2;
320 	int error;
321 
322 	if ((error = procfs_proc_lock(pfs->pfs_pid, &p2, ENOENT)) != 0)
323 		return error;
324 
325 	l1 = curlwp;				/* tracer */
326 
327 #define	M2K(m)	(((m) & FREAD) && ((m) & FWRITE) ? \
328 		 KAUTH_REQ_PROCESS_PROCFS_RW : \
329 		 (m) & FWRITE ? KAUTH_REQ_PROCESS_PROCFS_WRITE : \
330 		 KAUTH_REQ_PROCESS_PROCFS_READ)
331 
332 	mutex_enter(p2->p_lock);
333 	error = kauth_authorize_process(l1->l_cred, KAUTH_PROCESS_PROCFS,
334 	    p2, pfs, KAUTH_ARG(M2K(ap->a_mode)), NULL);
335 	mutex_exit(p2->p_lock);
336 	if (error) {
337 		procfs_proc_unlock(p2);
338 		return (error);
339 	}
340 
341 #undef M2K
342 
343 	switch (pfs->pfs_type) {
344 	case PFSmem:
345 		if (((pfs->pfs_flags & FWRITE) && (ap->a_mode & O_EXCL)) ||
346 		    ((pfs->pfs_flags & O_EXCL) && (ap->a_mode & FWRITE))) {
347 			error = EBUSY;
348 			break;
349 		}
350 
351 		if (!proc_isunder(p2, l1)) {
352 			error = EPERM;
353 			break;
354 		}
355 
356 		if (ap->a_mode & FWRITE)
357 			pfs->pfs_flags = ap->a_mode & (FWRITE|O_EXCL);
358 
359 		break;
360 
361 	case PFSregs:
362 	case PFSfpregs:
363 		if (!proc_isunder(p2, l1)) {
364 			error = EPERM;
365 			break;
366 		}
367 		break;
368 
369 	default:
370 		break;
371 	}
372 
373 	procfs_proc_unlock(p2);
374 	return (error);
375 }
376 
377 /*
378  * close the pfsnode (vp) after doing i/o.
379  * (vp) is not locked on entry or exit.
380  *
381  * nothing to do for procfs other than undo
382  * any exclusive open flag (see _open above).
383  */
384 int
385 procfs_close(void *v)
386 {
387 	struct vop_close_args /* {
388 		struct vnode *a_vp;
389 		int  a_fflag;
390 		kauth_cred_t a_cred;
391 	} */ *ap = v;
392 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
393 
394 	switch (pfs->pfs_type) {
395 	case PFSmem:
396 		if ((ap->a_fflag & FWRITE) && (pfs->pfs_flags & O_EXCL))
397 			pfs->pfs_flags &= ~(FWRITE|O_EXCL);
398 		break;
399 
400 	default:
401 		break;
402 	}
403 
404 	return (0);
405 }
406 
407 /*
408  * _inactive is called when the pfsnode
409  * is vrele'd and the reference count goes
410  * to zero.  (vp) will be on the vnode free
411  * list, so to get it back vget() must be
412  * used.
413  *
414  * (vp) is locked on entry, but must be unlocked on exit.
415  */
416 int
417 procfs_inactive(void *v)
418 {
419 	struct vop_inactive_v2_args /* {
420 		struct vnode *a_vp;
421 		bool *a_recycle;
422 	} */ *ap = v;
423 	struct vnode *vp = ap->a_vp;
424 	struct pfsnode *pfs = VTOPFS(vp);
425 
426 	mutex_enter(proc_lock);
427 	*ap->a_recycle = (proc_find(pfs->pfs_pid) == NULL);
428 	mutex_exit(proc_lock);
429 
430 	return (0);
431 }
432 
433 /*
434  * _reclaim is called when getnewvnode()
435  * wants to make use of an entry on the vnode
436  * free list.  at this time the filesystem needs
437  * to free any private data and remove the node
438  * from any private lists.
439  */
440 int
441 procfs_reclaim(void *v)
442 {
443 	struct vop_reclaim_v2_args /* {
444 		struct vnode *a_vp;
445 	} */ *ap = v;
446 	struct vnode *vp = ap->a_vp;
447 	struct pfsnode *pfs = VTOPFS(vp);
448 
449 	VOP_UNLOCK(vp);
450 
451 	/*
452 	 * To interlock with procfs_revoke_vnodes().
453 	 */
454 	mutex_enter(vp->v_interlock);
455 	vp->v_data = NULL;
456 	mutex_exit(vp->v_interlock);
457 	kmem_free(pfs, sizeof(*pfs));
458 	return 0;
459 }
460 
461 /*
462  * Return POSIX pathconf information applicable to special devices.
463  */
464 int
465 procfs_pathconf(void *v)
466 {
467 	struct vop_pathconf_args /* {
468 		struct vnode *a_vp;
469 		int a_name;
470 		register_t *a_retval;
471 	} */ *ap = v;
472 
473 	switch (ap->a_name) {
474 	case _PC_LINK_MAX:
475 		*ap->a_retval = LINK_MAX;
476 		return (0);
477 	case _PC_MAX_CANON:
478 		*ap->a_retval = MAX_CANON;
479 		return (0);
480 	case _PC_MAX_INPUT:
481 		*ap->a_retval = MAX_INPUT;
482 		return (0);
483 	case _PC_PIPE_BUF:
484 		*ap->a_retval = PIPE_BUF;
485 		return (0);
486 	case _PC_CHOWN_RESTRICTED:
487 		*ap->a_retval = 1;
488 		return (0);
489 	case _PC_VDISABLE:
490 		*ap->a_retval = _POSIX_VDISABLE;
491 		return (0);
492 	case _PC_SYNC_IO:
493 		*ap->a_retval = 1;
494 		return (0);
495 	default:
496 		return (EINVAL);
497 	}
498 	/* NOTREACHED */
499 }
500 
501 /*
502  * _print is used for debugging.
503  * just print a readable description
504  * of (vp).
505  */
506 int
507 procfs_print(void *v)
508 {
509 	struct vop_print_args /* {
510 		struct vnode *a_vp;
511 	} */ *ap = v;
512 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
513 
514 	printf("tag VT_PROCFS, type %d, pid %d, mode %x, flags %lx\n",
515 	    pfs->pfs_type, pfs->pfs_pid, pfs->pfs_mode, pfs->pfs_flags);
516 	return 0;
517 }
518 
519 int
520 procfs_link(void *v)
521 {
522 	struct vop_link_v2_args /* {
523 		struct vnode *a_dvp;
524 		struct vnode *a_vp;
525 		struct componentname *a_cnp;
526 	} */ *ap = v;
527 
528 	VOP_ABORTOP(ap->a_dvp, ap->a_cnp);
529 	return (EROFS);
530 }
531 
532 int
533 procfs_symlink(void *v)
534 {
535 	struct vop_symlink_v3_args /* {
536 		struct vnode *a_dvp;
537 		struct vnode **a_vpp;
538 		struct componentname *a_cnp;
539 		struct vattr *a_vap;
540 		char *a_target;
541 	} */ *ap = v;
542 
543 	VOP_ABORTOP(ap->a_dvp, ap->a_cnp);
544 	return (EROFS);
545 }
546 
547 /*
548  * Works out the path to (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 	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 			bp = *bpp;
600 			*--bp = '/';
601 		}
602 	}
603 
604 	if (bpp)
605 		*bpp = bp;
606 
607 	rw_exit(&cwdi->cwdi_lock);
608 }
609 
610 /*
611  * Invent attributes for pfsnode (vp) and store
612  * them in (vap).
613  * Directories lengths are returned as zero since
614  * any real length would require the genuine size
615  * to be computed, and nothing cares anyway.
616  *
617  * this is relatively minimal for procfs.
618  */
619 int
620 procfs_getattr(void *v)
621 {
622 	struct vop_getattr_args /* {
623 		struct vnode *a_vp;
624 		struct vattr *a_vap;
625 		kauth_cred_t a_cred;
626 	} */ *ap = v;
627 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
628 	struct vattr *vap = ap->a_vap;
629 	struct proc *procp;
630 	char *path, *bp, bf[16];
631 	int error;
632 
633 	/* first check the process still exists */
634 	switch (pfs->pfs_type) {
635 	case PFSroot:
636 	case PFScurproc:
637 	case PFSself:
638 		procp = NULL;
639 		break;
640 
641 	default:
642 		error = procfs_proc_lock(pfs->pfs_pid, &procp, ENOENT);
643 		if (error != 0)
644 			return (error);
645 		break;
646 	}
647 
648 	switch (pfs->pfs_type) {
649 	case PFStask:
650 		if (pfs->pfs_fd == -1) {
651 			path = NULL;
652 			break;
653 		}
654 		/*FALLTHROUGH*/
655 	case PFScwd:
656 	case PFSchroot:
657 		path = malloc(MAXPATHLEN + 4, M_TEMP, M_WAITOK|M_CANFAIL);
658 		if (path == NULL && procp != NULL) {
659 			procfs_proc_unlock(procp);
660 			return (ENOMEM);
661 		}
662 		break;
663 
664 	default:
665 		path = NULL;
666 		break;
667 	}
668 
669 	if (procp != NULL) {
670 		mutex_enter(procp->p_lock);
671 		error = kauth_authorize_process(kauth_cred_get(),
672 		    KAUTH_PROCESS_CANSEE, procp,
673 		    KAUTH_ARG(KAUTH_REQ_PROCESS_CANSEE_ENTRY), NULL, NULL);
674 		mutex_exit(procp->p_lock);
675 		if (error != 0) {
676 		    	procfs_proc_unlock(procp);
677 		    	if (path != NULL)
678 		    		free(path, M_TEMP);
679 			return (ENOENT);
680 		}
681 	}
682 
683 	error = 0;
684 
685 	/* start by zeroing out the attributes */
686 	vattr_null(vap);
687 
688 	/* next do all the common fields */
689 	vap->va_type = ap->a_vp->v_type;
690 	vap->va_mode = pfs->pfs_mode;
691 	vap->va_fileid = pfs->pfs_fileno;
692 	vap->va_flags = 0;
693 	vap->va_blocksize = PAGE_SIZE;
694 
695 	/*
696 	 * Make all times be current TOD.
697 	 *
698 	 * It would be possible to get the process start
699 	 * time from the p_stats structure, but there's
700 	 * no "file creation" time stamp anyway, and the
701 	 * p_stats structure is not addressable if u. gets
702 	 * swapped out for that process.
703 	 */
704 	getnanotime(&vap->va_ctime);
705 	vap->va_atime = vap->va_mtime = vap->va_ctime;
706 	if (procp)
707 		TIMEVAL_TO_TIMESPEC(&procp->p_stats->p_start,
708 		    &vap->va_birthtime);
709 	else
710 		getnanotime(&vap->va_birthtime);
711 
712 	switch (pfs->pfs_type) {
713 	case PFSmem:
714 	case PFSregs:
715 	case PFSfpregs:
716 #if defined(__HAVE_PROCFS_MACHDEP) && defined(PROCFS_MACHDEP_PROTECT_CASES)
717 	PROCFS_MACHDEP_PROTECT_CASES
718 #endif
719 		/*
720 		 * If the process has exercised some setuid or setgid
721 		 * privilege, then rip away read/write permission so
722 		 * that only root can gain access.
723 		 */
724 		if (procp->p_flag & PK_SUGID)
725 			vap->va_mode &= ~(S_IRUSR|S_IWUSR);
726 		/* FALLTHROUGH */
727 	case PFSstatus:
728 	case PFSstat:
729 	case PFSnote:
730 	case PFSnotepg:
731 	case PFScmdline:
732 	case PFSenviron:
733 	case PFSemul:
734 	case PFSstatm:
735 
736 	case PFSmap:
737 	case PFSmaps:
738 	case PFSauxv:
739 		vap->va_nlink = 1;
740 		vap->va_uid = kauth_cred_geteuid(procp->p_cred);
741 		vap->va_gid = kauth_cred_getegid(procp->p_cred);
742 		break;
743 	case PFScwd:
744 	case PFSchroot:
745 	case PFSmeminfo:
746 	case PFSdevices:
747 	case PFScpuinfo:
748 	case PFSuptime:
749 	case PFSmounts:
750 	case PFScpustat:
751 	case PFSloadavg:
752 	case PFSversion:
753 	case PFSexe:
754 	case PFSself:
755 	case PFScurproc:
756 	case PFSroot:
757 		vap->va_nlink = 1;
758 		vap->va_uid = vap->va_gid = 0;
759 		break;
760 
761 	case PFSproc:
762 	case PFStask:
763 	case PFSfile:
764 	case PFSfd:
765 		break;
766 
767 	default:
768 		panic("%s: %d/1", __func__, pfs->pfs_type);
769 	}
770 
771 	/*
772 	 * now do the object specific fields
773 	 *
774 	 * The size could be set from struct reg, but it's hardly
775 	 * worth the trouble, and it puts some (potentially) machine
776 	 * dependent data into this machine-independent code.  If it
777 	 * becomes important then this function should break out into
778 	 * a per-file stat function in the corresponding .c file.
779 	 */
780 
781 	switch (pfs->pfs_type) {
782 	case PFSroot:
783 		vap->va_bytes = vap->va_size = DEV_BSIZE;
784 		break;
785 
786 	case PFSself:
787 	case PFScurproc:
788 		vap->va_bytes = vap->va_size =
789 		    snprintf(bf, sizeof(bf), "%ld", (long)curproc->p_pid);
790 		break;
791 	case PFStask:
792 		if (pfs->pfs_fd != -1) {
793 			vap->va_nlink = 1;
794 			vap->va_uid = 0;
795 			vap->va_gid = 0;
796 			vap->va_bytes = vap->va_size =
797 			    snprintf(bf, sizeof(bf), "..");
798 			break;
799 		}
800 		/*FALLTHROUGH*/
801 	case PFSfd:
802 		if (pfs->pfs_fd != -1) {
803 			file_t *fp;
804 
805 			fp = fd_getfile2(procp, pfs->pfs_fd);
806 			if (fp == NULL) {
807 				error = EBADF;
808 				break;
809 			}
810 			vap->va_nlink = 1;
811 			vap->va_uid = kauth_cred_geteuid(fp->f_cred);
812 			vap->va_gid = kauth_cred_getegid(fp->f_cred);
813 			switch (fp->f_type) {
814 			case DTYPE_VNODE:
815 				vap->va_bytes = vap->va_size =
816 				    fp->f_vnode->v_size;
817 				break;
818 			default:
819 				vap->va_bytes = vap->va_size = 0;
820 				break;
821 			}
822 			closef(fp);
823 			break;
824 		}
825 		/*FALLTHROUGH*/
826 	case PFSproc:
827 		vap->va_nlink = 2;
828 		vap->va_uid = kauth_cred_geteuid(procp->p_cred);
829 		vap->va_gid = kauth_cred_getegid(procp->p_cred);
830 		vap->va_bytes = vap->va_size = DEV_BSIZE;
831 		break;
832 
833 	case PFSfile:
834 		error = EOPNOTSUPP;
835 		break;
836 
837 	case PFSmem:
838 		vap->va_bytes = vap->va_size =
839 			ctob(procp->p_vmspace->vm_tsize +
840 				    procp->p_vmspace->vm_dsize +
841 				    procp->p_vmspace->vm_ssize);
842 		break;
843 
844 	case PFSauxv:
845 		vap->va_bytes = vap->va_size = procp->p_execsw->es_arglen;
846 		break;
847 
848 #if defined(PT_GETREGS) || defined(PT_SETREGS)
849 	case PFSregs:
850 		vap->va_bytes = vap->va_size = sizeof(struct reg);
851 		break;
852 #endif
853 
854 #if defined(PT_GETFPREGS) || defined(PT_SETFPREGS)
855 	case PFSfpregs:
856 		vap->va_bytes = vap->va_size = sizeof(struct fpreg);
857 		break;
858 #endif
859 
860 	case PFSstatus:
861 	case PFSstat:
862 	case PFSnote:
863 	case PFSnotepg:
864 	case PFScmdline:
865 	case PFSenviron:
866 	case PFSmeminfo:
867 	case PFSdevices:
868 	case PFScpuinfo:
869 	case PFSuptime:
870 	case PFSmounts:
871 	case PFScpustat:
872 	case PFSloadavg:
873 	case PFSstatm:
874 	case PFSversion:
875 		vap->va_bytes = vap->va_size = 0;
876 		break;
877 	case PFSmap:
878 	case PFSmaps:
879 		/*
880 		 * Advise a larger blocksize for the map files, so that
881 		 * they may be read in one pass.
882 		 */
883 		vap->va_blocksize = 4 * PAGE_SIZE;
884 		vap->va_bytes = vap->va_size = 0;
885 		break;
886 
887 	case PFScwd:
888 	case PFSchroot:
889 		bp = path + MAXPATHLEN;
890 		*--bp = '\0';
891 		procfs_dir(pfs->pfs_type, curlwp, procp, &bp, path,
892 		     MAXPATHLEN);
893 		vap->va_bytes = vap->va_size = strlen(bp);
894 		break;
895 
896 	case PFSexe:
897 		vap->va_bytes = vap->va_size = strlen(procp->p_path);
898 		break;
899 
900 	case PFSemul:
901 		vap->va_bytes = vap->va_size = strlen(procp->p_emul->e_name);
902 		break;
903 
904 #ifdef __HAVE_PROCFS_MACHDEP
905 	PROCFS_MACHDEP_NODETYPE_CASES
906 		error = procfs_machdep_getattr(ap->a_vp, vap, procp);
907 		break;
908 #endif
909 
910 	default:
911 		panic("%s: %d/2", __func__, pfs->pfs_type);
912 	}
913 
914 	if (procp != NULL)
915 		procfs_proc_unlock(procp);
916 	if (path != NULL)
917 		free(path, M_TEMP);
918 
919 	return (error);
920 }
921 
922 /*ARGSUSED*/
923 int
924 procfs_setattr(void *v)
925 {
926 	/*
927 	 * just fake out attribute setting
928 	 * it's not good to generate an error
929 	 * return, otherwise things like creat()
930 	 * will fail when they try to set the
931 	 * file length to 0.  worse, this means
932 	 * that echo $note > /proc/$pid/note will fail.
933 	 */
934 
935 	return (0);
936 }
937 
938 /*
939  * implement access checking.
940  *
941  * actually, the check for super-user is slightly
942  * broken since it will allow read access to write-only
943  * objects.  this doesn't cause any particular trouble
944  * but does mean that the i/o entry points need to check
945  * that the operation really does make sense.
946  */
947 int
948 procfs_access(void *v)
949 {
950 	struct vop_access_args /* {
951 		struct vnode *a_vp;
952 		int a_mode;
953 		kauth_cred_t a_cred;
954 	} */ *ap = v;
955 	struct vattr va;
956 	int error;
957 
958 	if ((error = VOP_GETATTR(ap->a_vp, &va, ap->a_cred)) != 0)
959 		return (error);
960 
961 	return kauth_authorize_vnode(ap->a_cred,
962 	    KAUTH_ACCESS_ACTION(ap->a_mode, ap->a_vp->v_type, va.va_mode),
963 	    ap->a_vp, NULL, genfs_can_access(va.va_type, va.va_mode,
964 	    va.va_uid, va.va_gid, ap->a_mode, ap->a_cred));
965 }
966 
967 /*
968  * lookup.  this is incredibly complicated in the
969  * general case, however for most pseudo-filesystems
970  * very little needs to be done.
971  *
972  * Locking isn't hard here, just poorly documented.
973  *
974  * If we're looking up ".", just vref the parent & return it.
975  *
976  * If we're looking up "..", unlock the parent, and lock "..". If everything
977  * went ok, and we're on the last component and the caller requested the
978  * parent locked, try to re-lock the parent. We do this to prevent lock
979  * races.
980  *
981  * For anything else, get the needed node. Then unlock the parent if not
982  * the last component or not LOCKPARENT (i.e. if we wouldn't re-lock the
983  * parent in the .. case).
984  *
985  * We try to exit with the parent locked in error cases.
986  */
987 int
988 procfs_lookup(void *v)
989 {
990 	struct vop_lookup_v2_args /* {
991 		struct vnode * a_dvp;
992 		struct vnode ** a_vpp;
993 		struct componentname * a_cnp;
994 	} */ *ap = v;
995 	struct componentname *cnp = ap->a_cnp;
996 	struct vnode **vpp = ap->a_vpp;
997 	struct vnode *dvp = ap->a_dvp;
998 	const char *pname = cnp->cn_nameptr;
999 	const struct proc_target *pt = NULL;
1000 	struct vnode *fvp;
1001 	pid_t pid, vnpid;
1002 	struct pfsnode *pfs;
1003 	struct proc *p = NULL;
1004 	struct lwp *plwp;
1005 	int i, error;
1006 	pfstype type;
1007 
1008 	*vpp = NULL;
1009 
1010 	if (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)
1011 		return (EROFS);
1012 
1013 	if (cnp->cn_namelen == 1 && *pname == '.') {
1014 		*vpp = dvp;
1015 		vref(dvp);
1016 		return (0);
1017 	}
1018 
1019 	pfs = VTOPFS(dvp);
1020 	switch (pfs->pfs_type) {
1021 	case PFSroot:
1022 		/*
1023 		 * Shouldn't get here with .. in the root node.
1024 		 */
1025 		if (cnp->cn_flags & ISDOTDOT)
1026 			return (EIO);
1027 
1028 		for (i = 0; i < nproc_root_targets; i++) {
1029 			pt = &proc_root_targets[i];
1030 			/*
1031 			 * check for node match.  proc is always NULL here,
1032 			 * so call pt_valid with constant NULL lwp.
1033 			 */
1034 			if (cnp->cn_namelen == pt->pt_namlen &&
1035 			    memcmp(pt->pt_name, pname, cnp->cn_namelen) == 0 &&
1036 			    (pt->pt_valid == NULL ||
1037 			     (*pt->pt_valid)(NULL, dvp->v_mount)))
1038 				break;
1039 		}
1040 
1041 		if (i != nproc_root_targets) {
1042 			error = procfs_allocvp(dvp->v_mount, vpp, 0,
1043 			    pt->pt_pfstype, -1);
1044 			return (error);
1045 		}
1046 
1047 		if (CNEQ(cnp, "curproc", 7)) {
1048 			pid = curproc->p_pid;
1049 			vnpid = 0;
1050 			type = PFScurproc;
1051 		} else if (CNEQ(cnp, "self", 4)) {
1052 			pid = curproc->p_pid;
1053 			vnpid = 0;
1054 			type = PFSself;
1055 		} else {
1056 			pid = (pid_t)atoi(pname, cnp->cn_namelen);
1057 			vnpid = pid;
1058 			type = PFSproc;
1059 		}
1060 
1061 		if (procfs_proc_lock(pid, &p, ESRCH) != 0)
1062 			break;
1063 		error = procfs_allocvp(dvp->v_mount, vpp, vnpid, type, -1);
1064 		procfs_proc_unlock(p);
1065 		return (error);
1066 
1067 	case PFSproc:
1068 		if (cnp->cn_flags & ISDOTDOT) {
1069 			error = procfs_allocvp(dvp->v_mount, vpp, 0, PFSroot,
1070 			    -1);
1071 			return (error);
1072 		}
1073 
1074 		if (procfs_proc_lock(pfs->pfs_pid, &p, ESRCH) != 0)
1075 			break;
1076 
1077 		mutex_enter(p->p_lock);
1078 		LIST_FOREACH(plwp, &p->p_lwps, l_sibling) {
1079 			if (plwp->l_stat != LSZOMB)
1080 				break;
1081 		}
1082 		/* Process is exiting if no-LWPS or all LWPs are LSZOMB */
1083 		if (plwp == NULL) {
1084 			mutex_exit(p->p_lock);
1085 			procfs_proc_unlock(p);
1086 			return ESRCH;
1087 		}
1088 
1089 		lwp_addref(plwp);
1090 		mutex_exit(p->p_lock);
1091 
1092 		for (pt = proc_targets, i = 0; i < nproc_targets; pt++, i++) {
1093 			int found;
1094 
1095 			found = cnp->cn_namelen == pt->pt_namlen &&
1096 			    memcmp(pt->pt_name, pname, cnp->cn_namelen) == 0 &&
1097 			    (pt->pt_valid == NULL
1098 			      || (*pt->pt_valid)(plwp, dvp->v_mount));
1099 			if (found)
1100 				break;
1101 		}
1102 		lwp_delref(plwp);
1103 
1104 		if (i == nproc_targets) {
1105 			procfs_proc_unlock(p);
1106 			break;
1107 		}
1108 		if (pt->pt_pfstype == PFSfile) {
1109 			fvp = p->p_textvp;
1110 			/* We already checked that it exists. */
1111 			vref(fvp);
1112 			procfs_proc_unlock(p);
1113 			*vpp = fvp;
1114 			return (0);
1115 		}
1116 
1117 		error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
1118 		    pt->pt_pfstype, -1);
1119 		procfs_proc_unlock(p);
1120 		return (error);
1121 
1122 	case PFSfd: {
1123 		int fd;
1124 		file_t *fp;
1125 
1126 		if ((error = procfs_proc_lock(pfs->pfs_pid, &p, ENOENT)) != 0)
1127 			return error;
1128 
1129 		if (cnp->cn_flags & ISDOTDOT) {
1130 			error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
1131 			    PFSproc, -1);
1132 			procfs_proc_unlock(p);
1133 			return (error);
1134 		}
1135 		fd = atoi(pname, cnp->cn_namelen);
1136 
1137 		fp = fd_getfile2(p, fd);
1138 		if (fp == NULL) {
1139 			procfs_proc_unlock(p);
1140 			return ENOENT;
1141 		}
1142 		fvp = fp->f_vnode;
1143 
1144 		/* Don't show directories */
1145 		if (fp->f_type == DTYPE_VNODE && fvp->v_type != VDIR) {
1146 			vref(fvp);
1147 			closef(fp);
1148 			procfs_proc_unlock(p);
1149 			*vpp = fvp;
1150 			return 0;
1151 		}
1152 
1153 		closef(fp);
1154 		error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
1155 		    PFSfd, fd);
1156 		procfs_proc_unlock(p);
1157 		return error;
1158 	}
1159 	case PFStask: {
1160 		int xpid;
1161 
1162 		if ((error = procfs_proc_lock(pfs->pfs_pid, &p, ENOENT)) != 0)
1163 			return error;
1164 
1165 		if (cnp->cn_flags & ISDOTDOT) {
1166 			error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
1167 			    PFSproc, -1);
1168 			procfs_proc_unlock(p);
1169 			return (error);
1170 		}
1171 		xpid = atoi(pname, cnp->cn_namelen);
1172 
1173 		if (xpid != pfs->pfs_pid) {
1174 			procfs_proc_unlock(p);
1175 			return ENOENT;
1176 		}
1177 		error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
1178 		    PFStask, 0);
1179 		procfs_proc_unlock(p);
1180 		return error;
1181 	}
1182 	default:
1183 		return (ENOTDIR);
1184 	}
1185 
1186 	return (cnp->cn_nameiop == LOOKUP ? ENOENT : EROFS);
1187 }
1188 
1189 int
1190 procfs_validfile(struct lwp *l, struct mount *mp)
1191 {
1192 	return l != NULL && l->l_proc != NULL && l->l_proc->p_textvp != NULL;
1193 }
1194 
1195 static int
1196 procfs_validfile_linux(struct lwp *l, struct mount *mp)
1197 {
1198 	int flags;
1199 
1200 	flags = VFSTOPROC(mp)->pmnt_flags;
1201 	return (flags & PROCFSMNT_LINUXCOMPAT) &&
1202 	    (l == NULL || l->l_proc == NULL || procfs_validfile(l, mp));
1203 }
1204 
1205 struct procfs_root_readdir_ctx {
1206 	struct uio *uiop;
1207 	off_t *cookies;
1208 	int ncookies;
1209 	off_t off;
1210 	off_t startoff;
1211 	int error;
1212 };
1213 
1214 static int
1215 procfs_root_readdir_callback(struct proc *p, void *arg)
1216 {
1217 	struct procfs_root_readdir_ctx *ctxp = arg;
1218 	struct dirent d;
1219 	struct uio *uiop;
1220 	int error;
1221 
1222 	uiop = ctxp->uiop;
1223 	if (uiop->uio_resid < UIO_MX)
1224 		return -1; /* no space */
1225 
1226 	if (ctxp->off < ctxp->startoff) {
1227 		ctxp->off++;
1228 		return 0;
1229 	}
1230 
1231 	if (kauth_authorize_process(kauth_cred_get(),
1232 	    KAUTH_PROCESS_CANSEE, p,
1233 	    KAUTH_ARG(KAUTH_REQ_PROCESS_CANSEE_ENTRY), NULL, NULL) != 0)
1234 		return 0;
1235 
1236 	memset(&d, 0, UIO_MX);
1237 	d.d_reclen = UIO_MX;
1238 	d.d_fileno = PROCFS_FILENO(p->p_pid, PFSproc, -1);
1239 	d.d_namlen = snprintf(d.d_name,
1240 	    UIO_MX - offsetof(struct dirent, d_name), "%ld", (long)p->p_pid);
1241 	d.d_type = DT_DIR;
1242 
1243 	mutex_exit(proc_lock);
1244 	error = uiomove(&d, UIO_MX, uiop);
1245 	mutex_enter(proc_lock);
1246 	if (error) {
1247 		ctxp->error = error;
1248 		return -1;
1249 	}
1250 
1251 	ctxp->ncookies++;
1252 	if (ctxp->cookies)
1253 		*(ctxp->cookies)++ = ctxp->off + 1;
1254 	ctxp->off++;
1255 
1256 	return 0;
1257 }
1258 
1259 /*
1260  * readdir returns directory entries from pfsnode (vp).
1261  *
1262  * the strategy here with procfs is to generate a single
1263  * directory entry at a time (struct dirent) and then
1264  * copy that out to userland using uiomove.  a more efficent
1265  * though more complex implementation, would try to minimize
1266  * the number of calls to uiomove().  for procfs, this is
1267  * hardly worth the added code complexity.
1268  *
1269  * this should just be done through read()
1270  */
1271 int
1272 procfs_readdir(void *v)
1273 {
1274 	struct vop_readdir_args /* {
1275 		struct vnode *a_vp;
1276 		struct uio *a_uio;
1277 		kauth_cred_t a_cred;
1278 		int *a_eofflag;
1279 		off_t **a_cookies;
1280 		int *a_ncookies;
1281 	} */ *ap = v;
1282 	struct uio *uio = ap->a_uio;
1283 	struct dirent d;
1284 	struct pfsnode *pfs;
1285 	off_t i;
1286 	int error;
1287 	off_t *cookies = NULL;
1288 	int ncookies;
1289 	struct vnode *vp;
1290 	const struct proc_target *pt;
1291 	struct procfs_root_readdir_ctx ctx;
1292 	struct lwp *l;
1293 	int nfd;
1294 
1295 	vp = ap->a_vp;
1296 	pfs = VTOPFS(vp);
1297 
1298 	if (uio->uio_resid < UIO_MX)
1299 		return (EINVAL);
1300 	if (uio->uio_offset < 0)
1301 		return (EINVAL);
1302 
1303 	error = 0;
1304 	i = uio->uio_offset;
1305 	memset(&d, 0, UIO_MX);
1306 	d.d_reclen = UIO_MX;
1307 	ncookies = uio->uio_resid / UIO_MX;
1308 
1309 	switch (pfs->pfs_type) {
1310 	/*
1311 	 * this is for the process-specific sub-directories.
1312 	 * all that is needed to is copy out all the entries
1313 	 * from the procent[] table (top of this file).
1314 	 */
1315 	case PFSproc: {
1316 		struct proc *p;
1317 
1318 		if (i >= nproc_targets)
1319 			return 0;
1320 
1321 		if (procfs_proc_lock(pfs->pfs_pid, &p, ESRCH) != 0)
1322 			break;
1323 
1324 		if (ap->a_ncookies) {
1325 			ncookies = min(ncookies, (nproc_targets - i));
1326 			cookies = malloc(ncookies * sizeof (off_t),
1327 			    M_TEMP, M_WAITOK);
1328 			*ap->a_cookies = cookies;
1329 		}
1330 
1331 		for (pt = &proc_targets[i];
1332 		     uio->uio_resid >= UIO_MX && i < nproc_targets; pt++, i++) {
1333 			if (pt->pt_valid) {
1334 				/* XXXSMP LWP can disappear */
1335 				mutex_enter(p->p_lock);
1336 				l = LIST_FIRST(&p->p_lwps);
1337 				KASSERT(l != NULL);
1338 				mutex_exit(p->p_lock);
1339 				if ((*pt->pt_valid)(l, vp->v_mount) == 0)
1340 					continue;
1341 			}
1342 
1343 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid,
1344 			    pt->pt_pfstype, -1);
1345 			d.d_namlen = pt->pt_namlen;
1346 			memcpy(d.d_name, pt->pt_name, pt->pt_namlen + 1);
1347 			d.d_type = pt->pt_type;
1348 
1349 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1350 				break;
1351 			if (cookies)
1352 				*cookies++ = i + 1;
1353 		}
1354 
1355 		procfs_proc_unlock(p);
1356 	    	break;
1357 	}
1358 	case PFSfd: {
1359 		struct proc *p;
1360 		file_t *fp;
1361 		int lim, nc = 0;
1362 
1363 		if ((error = procfs_proc_lock(pfs->pfs_pid, &p, ESRCH)) != 0)
1364 			return error;
1365 
1366 		/* XXX Should this be by file as well? */
1367 		if (kauth_authorize_process(kauth_cred_get(),
1368 		    KAUTH_PROCESS_CANSEE, p,
1369 		    KAUTH_ARG(KAUTH_REQ_PROCESS_CANSEE_OPENFILES), NULL,
1370 		    NULL) != 0) {
1371 		    	procfs_proc_unlock(p);
1372 			return ESRCH;
1373 		}
1374 
1375 		nfd = p->p_fd->fd_dt->dt_nfiles;
1376 
1377 		lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfiles);
1378 		if (i >= lim) {
1379 		    	procfs_proc_unlock(p);
1380 			return 0;
1381 		}
1382 
1383 		if (ap->a_ncookies) {
1384 			ncookies = min(ncookies, (nfd + 2 - i));
1385 			cookies = malloc(ncookies * sizeof (off_t),
1386 			    M_TEMP, M_WAITOK);
1387 			*ap->a_cookies = cookies;
1388 		}
1389 
1390 		for (; i < 2 && uio->uio_resid >= UIO_MX; i++) {
1391 			pt = &proc_targets[i];
1392 			d.d_namlen = pt->pt_namlen;
1393 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid,
1394 			    pt->pt_pfstype, -1);
1395 			(void)memcpy(d.d_name, pt->pt_name, pt->pt_namlen + 1);
1396 			d.d_type = pt->pt_type;
1397 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1398 				break;
1399 			if (cookies)
1400 				*cookies++ = i + 1;
1401 			nc++;
1402 		}
1403 		if (error) {
1404 			ncookies = nc;
1405 			break;
1406 		}
1407 		for (; uio->uio_resid >= UIO_MX && i < nfd; i++) {
1408 			/* check the descriptor exists */
1409 			if ((fp = fd_getfile2(p, i - 2)) == NULL)
1410 				continue;
1411 			closef(fp);
1412 
1413 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid, PFSfd, i - 2);
1414 			d.d_namlen = snprintf(d.d_name, sizeof(d.d_name),
1415 			    "%lld", (long long)(i - 2));
1416 			d.d_type = VREG;
1417 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1418 				break;
1419 			if (cookies)
1420 				*cookies++ = i + 1;
1421 			nc++;
1422 		}
1423 		ncookies = nc;
1424 		procfs_proc_unlock(p);
1425 		break;
1426 	}
1427 	case PFStask: {
1428 		struct proc *p;
1429 		int nc = 0;
1430 
1431 		if ((error = procfs_proc_lock(pfs->pfs_pid, &p, ESRCH)) != 0)
1432 			return error;
1433 
1434 		nfd = 3;	/* ., .., pid */
1435 
1436 		if (ap->a_ncookies) {
1437 			ncookies = min(ncookies, (nfd + 2 - i));
1438 			cookies = malloc(ncookies * sizeof (off_t),
1439 			    M_TEMP, M_WAITOK);
1440 			*ap->a_cookies = cookies;
1441 		}
1442 
1443 		for (; i < 2 && uio->uio_resid >= UIO_MX; i++) {
1444 			pt = &proc_targets[i];
1445 			d.d_namlen = pt->pt_namlen;
1446 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid,
1447 			    pt->pt_pfstype, -1);
1448 			(void)memcpy(d.d_name, pt->pt_name, pt->pt_namlen + 1);
1449 			d.d_type = pt->pt_type;
1450 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1451 				break;
1452 			if (cookies)
1453 				*cookies++ = i + 1;
1454 			nc++;
1455 		}
1456 		if (error) {
1457 			ncookies = nc;
1458 			break;
1459 		}
1460 		for (; uio->uio_resid >= UIO_MX && i < nfd; i++) {
1461 			/* check the descriptor exists */
1462 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid, PFStask,
1463 			    i - 2);
1464 			d.d_namlen = snprintf(d.d_name, sizeof(d.d_name),
1465 			    "%ld", (long)pfs->pfs_pid);
1466 			d.d_type = DT_LNK;
1467 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1468 				break;
1469 			if (cookies)
1470 				*cookies++ = i + 1;
1471 			nc++;
1472 		}
1473 		ncookies = nc;
1474 		procfs_proc_unlock(p);
1475 		break;
1476 	}
1477 
1478 	/*
1479 	 * this is for the root of the procfs filesystem
1480 	 * what is needed are special entries for "curproc"
1481 	 * and "self" followed by an entry for each process
1482 	 * on allproc.
1483 	 */
1484 
1485 	case PFSroot: {
1486 		int nc = 0;
1487 
1488 		if (ap->a_ncookies) {
1489 			/*
1490 			 * XXX Potentially allocating too much space here,
1491 			 * but I'm lazy. This loop needs some work.
1492 			 */
1493 			cookies = malloc(ncookies * sizeof (off_t),
1494 			    M_TEMP, M_WAITOK);
1495 			*ap->a_cookies = cookies;
1496 		}
1497 		error = 0;
1498 		/* 0 ... 3 are static entries. */
1499 		for (; i <= 3 && uio->uio_resid >= UIO_MX; i++) {
1500 			switch (i) {
1501 			case 0:		/* `.' */
1502 			case 1:		/* `..' */
1503 				d.d_fileno = PROCFS_FILENO(0, PFSroot, -1);
1504 				d.d_namlen = i + 1;
1505 				memcpy(d.d_name, "..", d.d_namlen);
1506 				d.d_name[i + 1] = '\0';
1507 				d.d_type = DT_DIR;
1508 				break;
1509 
1510 			case 2:
1511 				d.d_fileno = PROCFS_FILENO(0, PFScurproc, -1);
1512 				d.d_namlen = sizeof("curproc") - 1;
1513 				memcpy(d.d_name, "curproc", sizeof("curproc"));
1514 				d.d_type = DT_LNK;
1515 				break;
1516 
1517 			case 3:
1518 				d.d_fileno = PROCFS_FILENO(0, PFSself, -1);
1519 				d.d_namlen = sizeof("self") - 1;
1520 				memcpy(d.d_name, "self", sizeof("self"));
1521 				d.d_type = DT_LNK;
1522 				break;
1523 			}
1524 
1525 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1526 				break;
1527 			nc++;
1528 			if (cookies)
1529 				*cookies++ = i + 1;
1530 		}
1531 		/* 4 ... are process entries. */
1532 		ctx.uiop = uio;
1533 		ctx.error = 0;
1534 		ctx.off = 4;
1535 		ctx.startoff = i;
1536 		ctx.cookies = cookies;
1537 		ctx.ncookies = nc;
1538 		proclist_foreach_call(&allproc,
1539 		    procfs_root_readdir_callback, &ctx);
1540 		cookies = ctx.cookies;
1541 		nc = ctx.ncookies;
1542 		error = ctx.error;
1543 		if (error)
1544 			break;
1545 
1546 		/* misc entries. */
1547 		if (i < ctx.off)
1548 			i = ctx.off;
1549 		if (i >= ctx.off + nproc_root_targets)
1550 			break;
1551 		for (pt = &proc_root_targets[i - ctx.off];
1552 		    uio->uio_resid >= UIO_MX &&
1553 		    pt < &proc_root_targets[nproc_root_targets];
1554 		    pt++, i++) {
1555 			if (pt->pt_valid &&
1556 			    (*pt->pt_valid)(NULL, vp->v_mount) == 0)
1557 				continue;
1558 			d.d_fileno = PROCFS_FILENO(0, pt->pt_pfstype, -1);
1559 			d.d_namlen = pt->pt_namlen;
1560 			memcpy(d.d_name, pt->pt_name, pt->pt_namlen + 1);
1561 			d.d_type = pt->pt_type;
1562 
1563 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1564 				break;
1565 			nc++;
1566 			if (cookies)
1567 				*cookies++ = i + 1;
1568 		}
1569 
1570 		ncookies = nc;
1571 		break;
1572 	}
1573 
1574 	default:
1575 		error = ENOTDIR;
1576 		break;
1577 	}
1578 
1579 	if (ap->a_ncookies) {
1580 		if (error) {
1581 			if (cookies)
1582 				free(*ap->a_cookies, M_TEMP);
1583 			*ap->a_ncookies = 0;
1584 			*ap->a_cookies = NULL;
1585 		} else
1586 			*ap->a_ncookies = ncookies;
1587 	}
1588 	uio->uio_offset = i;
1589 	return (error);
1590 }
1591 
1592 /*
1593  * readlink reads the link of `curproc' and others
1594  */
1595 int
1596 procfs_readlink(void *v)
1597 {
1598 	struct vop_readlink_args *ap = v;
1599 	char bf[16];		/* should be enough */
1600 	char *bp = bf;
1601 	char *path = NULL;
1602 	int len = 0;
1603 	int error = 0;
1604 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
1605 	struct proc *pown = NULL;
1606 
1607 	if (pfs->pfs_fileno == PROCFS_FILENO(0, PFScurproc, -1))
1608 		len = snprintf(bf, sizeof(bf), "%ld", (long)curproc->p_pid);
1609 	else if (pfs->pfs_fileno == PROCFS_FILENO(0, PFSself, -1))
1610 		len = snprintf(bf, sizeof(bf), "%s", "curproc");
1611 	else if (pfs->pfs_fileno == PROCFS_FILENO(pfs->pfs_pid, PFStask, 0))
1612 		len = snprintf(bf, sizeof(bf), "..");
1613 	else if (pfs->pfs_fileno == PROCFS_FILENO(pfs->pfs_pid, PFSexe, -1)) {
1614 		if ((error = procfs_proc_lock(pfs->pfs_pid, &pown, ESRCH)) != 0)
1615 			return error;
1616 		bp = pown->p_path;
1617 		len = strlen(bp);
1618 	} else if (pfs->pfs_fileno == PROCFS_FILENO(pfs->pfs_pid, PFScwd, -1) ||
1619 	    pfs->pfs_fileno == PROCFS_FILENO(pfs->pfs_pid, PFSchroot, -1)) {
1620 		if ((error = procfs_proc_lock(pfs->pfs_pid, &pown, ESRCH)) != 0)
1621 			return error;
1622 		path = malloc(MAXPATHLEN + 4, M_TEMP, M_WAITOK|M_CANFAIL);
1623 		if (path == NULL) {
1624 			procfs_proc_unlock(pown);
1625 			return (ENOMEM);
1626 		}
1627 		bp = path + MAXPATHLEN;
1628 		*--bp = '\0';
1629 		procfs_dir(PROCFS_TYPE(pfs->pfs_fileno), curlwp, pown,
1630 		    &bp, path, MAXPATHLEN);
1631 		len = strlen(bp);
1632 	} else {
1633 		file_t *fp;
1634 		struct vnode *vxp, *vp;
1635 
1636 		if ((error = procfs_proc_lock(pfs->pfs_pid, &pown, ESRCH)) != 0)
1637 			return error;
1638 
1639 		fp = fd_getfile2(pown, pfs->pfs_fd);
1640 		if (fp == NULL) {
1641 			procfs_proc_unlock(pown);
1642 			return EBADF;
1643 		}
1644 
1645 		switch (fp->f_type) {
1646 		case DTYPE_VNODE:
1647 			vxp = fp->f_vnode;
1648 			if (vxp->v_type != VDIR) {
1649 				error = EINVAL;
1650 				break;
1651 			}
1652 			if ((path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK))
1653 			    == NULL) {
1654 				error = ENOMEM;
1655 				break;
1656 			}
1657 			bp = path + MAXPATHLEN;
1658 			*--bp = '\0';
1659 
1660 			/*
1661 			 * XXX: kludge to avoid locking against ourselves
1662 			 * in getcwd()
1663 			 */
1664 			if (vxp->v_tag == VT_PROCFS) {
1665 				*--bp = '/';
1666 			} else {
1667 				rw_enter(&curproc->p_cwdi->cwdi_lock,
1668 				    RW_READER);
1669 				vp = curproc->p_cwdi->cwdi_rdir;
1670 				if (vp == NULL)
1671 					vp = rootvnode;
1672 				error = getcwd_common(vxp, vp, &bp, path,
1673 				    MAXPATHLEN / 2, 0, curlwp);
1674 				rw_exit(&curproc->p_cwdi->cwdi_lock);
1675 			}
1676 			if (error)
1677 				break;
1678 			len = strlen(bp);
1679 			break;
1680 
1681 		case DTYPE_MISC:
1682 			len = snprintf(bf, sizeof(bf), "%s", "[misc]");
1683 			break;
1684 
1685 		case DTYPE_KQUEUE:
1686 			len = snprintf(bf, sizeof(bf), "%s", "[kqueue]");
1687 			break;
1688 
1689 		case DTYPE_SEM:
1690 			len = snprintf(bf, sizeof(bf), "%s", "[ksem]");
1691 			break;
1692 
1693 		default:
1694 			error = EINVAL;
1695 			break;
1696 		}
1697 		closef(fp);
1698 	}
1699 
1700 	if (error == 0)
1701 		error = uiomove(bp, len, ap->a_uio);
1702 	if (pown)
1703 		procfs_proc_unlock(pown);
1704 	if (path)
1705 		free(path, M_TEMP);
1706 	return error;
1707 }
1708 
1709 /*
1710  * convert decimal ascii to int
1711  */
1712 static int
1713 atoi(const char *b, size_t len)
1714 {
1715 	int p = 0;
1716 
1717 	while (len--) {
1718 		char c = *b++;
1719 		if (c < '0' || c > '9')
1720 			return -1;
1721 		p = 10 * p + (c - '0');
1722 	}
1723 
1724 	return p;
1725 }
1726