xref: /netbsd-src/sys/miscfs/procfs/procfs_vnops.c (revision 53b02e147d4ed531c0d2a5ca9b3e8026ba3e99b5)
1 /*	$NetBSD: procfs_vnops.c,v 1.220 2021/12/08 20:11:54 andvar Exp $	*/
2 
3 /*-
4  * Copyright (c) 2006, 2007, 2008, 2020 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 1.220 2021/12/08 20:11:54 andvar Exp $");
109 
110 #include <sys/param.h>
111 #include <sys/atomic.h>
112 #include <sys/systm.h>
113 #include <sys/time.h>
114 #include <sys/kernel.h>
115 #include <sys/file.h>
116 #include <sys/filedesc.h>
117 #include <sys/proc.h>
118 #include <sys/vnode.h>
119 #include <sys/namei.h>
120 #include <sys/malloc.h>
121 #include <sys/mount.h>
122 #include <sys/dirent.h>
123 #include <sys/resourcevar.h>
124 #include <sys/stat.h>
125 #include <sys/ptrace.h>
126 #include <sys/kauth.h>
127 #include <sys/exec.h>
128 
129 #include <uvm/uvm_extern.h>	/* for PAGE_SIZE */
130 
131 #include <machine/reg.h>
132 
133 #include <miscfs/genfs/genfs.h>
134 #include <miscfs/procfs/procfs.h>
135 
136 /*
137  * Vnode Operations.
138  *
139  */
140 
141 static int procfs_validfile_linux(struct lwp *, struct mount *);
142 static int procfs_root_readdir_callback(struct proc *, void *);
143 static void procfs_dir(pfstype, struct lwp *, struct proc *, char **, char *,
144     size_t);
145 
146 /*
147  * This is a list of the valid names in the
148  * process-specific sub-directories.  It is
149  * used in procfs_lookup and procfs_readdir
150  */
151 static const struct proc_target {
152 	u_char	pt_type;
153 	u_char	pt_namlen;
154 	const char	*pt_name;
155 	pfstype	pt_pfstype;
156 	int	(*pt_valid)(struct lwp *, struct mount *);
157 } proc_targets[] = {
158 #define N(s) sizeof(s)-1, s
159 	/*	  name		type		validp */
160 	{ DT_DIR, N("."),	PFSproc,	NULL },
161 	{ DT_DIR, N(".."),	PFSroot,	NULL },
162 	{ DT_DIR, N("fd"),	PFSfd,		NULL },
163 	{ DT_DIR, N("task"),	PFStask,	procfs_validfile_linux },
164 	{ DT_LNK, N("cwd"),	PFScwd,		NULL },
165 	{ DT_LNK, N("emul"),	PFSemul,	NULL },
166 	{ DT_LNK, N("root"),	PFSchroot,	NULL },
167 	{ DT_REG, N("auxv"),	PFSauxv,	procfs_validauxv },
168 	{ DT_REG, N("cmdline"), PFScmdline,	NULL },
169 	{ DT_REG, N("environ"), PFSenviron,	NULL },
170 	{ DT_REG, N("exe"),	PFSexe,		procfs_validfile },
171 	{ DT_REG, N("file"),	PFSfile,	procfs_validfile },
172 	{ DT_REG, N("fpregs"),	PFSfpregs,	procfs_validfpregs },
173 	{ DT_REG, N("limit"),	PFSlimit,	NULL },
174 	{ DT_REG, N("map"),	PFSmap,		procfs_validmap },
175 	{ DT_REG, N("maps"),	PFSmaps,	procfs_validmap },
176 	{ DT_REG, N("mem"),	PFSmem,		NULL },
177 	{ DT_REG, N("note"),	PFSnote,	NULL },
178 	{ DT_REG, N("notepg"),	PFSnotepg,	NULL },
179 	{ DT_REG, N("regs"),	PFSregs,	procfs_validregs },
180 	{ DT_REG, N("stat"),	PFSstat,	procfs_validfile_linux },
181 	{ DT_REG, N("statm"),	PFSstatm,	procfs_validfile_linux },
182 	{ DT_REG, N("status"),	PFSstatus,	NULL },
183 #ifdef __HAVE_PROCFS_MACHDEP
184 	PROCFS_MACHDEP_NODETYPE_DEFNS
185 #endif
186 #undef N
187 };
188 static const int nproc_targets = sizeof(proc_targets) / sizeof(proc_targets[0]);
189 
190 /*
191  * List of files in the root directory. Note: the validate function will
192  * be called with p == NULL for these ones.
193  */
194 static const struct proc_target proc_root_targets[] = {
195 #define N(s) sizeof(s)-1, s
196 	/*	  name		    type	    validp */
197 	{ DT_REG, N("meminfo"),     PFSmeminfo,        procfs_validfile_linux },
198 	{ DT_REG, N("cpuinfo"),     PFScpuinfo,        procfs_validfile_linux },
199 	{ DT_REG, N("uptime"),      PFSuptime,         procfs_validfile_linux },
200 	{ DT_REG, N("mounts"),	    PFSmounts,	       procfs_validfile_linux },
201 	{ DT_REG, N("devices"),     PFSdevices,        procfs_validfile_linux },
202 	{ DT_REG, N("stat"),	    PFScpustat,        procfs_validfile_linux },
203 	{ DT_REG, N("loadavg"),	    PFSloadavg,        procfs_validfile_linux },
204 	{ DT_REG, N("version"),     PFSversion,        procfs_validfile_linux },
205 #undef N
206 };
207 static const int nproc_root_targets =
208     sizeof(proc_root_targets) / sizeof(proc_root_targets[0]);
209 
210 int	procfs_lookup(void *);
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 int	procfs_link(void *);
217 int	procfs_symlink(void *);
218 int	procfs_readdir(void *);
219 int	procfs_readlink(void *);
220 int	procfs_inactive(void *);
221 int	procfs_reclaim(void *);
222 int	procfs_print(void *);
223 int	procfs_pathconf(void *);
224 int	procfs_getpages(void *);
225 
226 static int atoi(const char *, size_t);
227 
228 /*
229  * procfs vnode operations.
230  */
231 int (**procfs_vnodeop_p)(void *);
232 const struct vnodeopv_entry_desc procfs_vnodeop_entries[] = {
233 	{ &vop_default_desc, vn_default_error },
234 	{ &vop_parsepath_desc, genfs_parsepath },	/* parsepath */
235 	{ &vop_lookup_desc, procfs_lookup },		/* lookup */
236 	{ &vop_create_desc, genfs_eopnotsupp },		/* create */
237 	{ &vop_mknod_desc, genfs_eopnotsupp },		/* mknod */
238 	{ &vop_open_desc, procfs_open },		/* open */
239 	{ &vop_close_desc, procfs_close },		/* close */
240 	{ &vop_access_desc, procfs_access },		/* access */
241 	{ &vop_accessx_desc, genfs_accessx },		/* accessx */
242 	{ &vop_getattr_desc, procfs_getattr },		/* getattr */
243 	{ &vop_setattr_desc, procfs_setattr },		/* setattr */
244 	{ &vop_read_desc, procfs_rw },			/* read */
245 	{ &vop_write_desc, procfs_rw },			/* write */
246 	{ &vop_fallocate_desc, genfs_eopnotsupp },	/* fallocate */
247 	{ &vop_fdiscard_desc, genfs_eopnotsupp },	/* fdiscard */
248 	{ &vop_fcntl_desc, genfs_fcntl },		/* fcntl */
249 	{ &vop_ioctl_desc, genfs_enoioctl },		/* ioctl */
250 	{ &vop_poll_desc, genfs_poll },			/* poll */
251 	{ &vop_kqfilter_desc, genfs_kqfilter },		/* kqfilter */
252 	{ &vop_revoke_desc, genfs_revoke },		/* revoke */
253 	{ &vop_fsync_desc, genfs_nullop },		/* fsync */
254 	{ &vop_seek_desc, genfs_nullop },		/* seek */
255 	{ &vop_remove_desc, genfs_eopnotsupp },		/* remove */
256 	{ &vop_link_desc, procfs_link },		/* link */
257 	{ &vop_rename_desc, genfs_eopnotsupp },		/* rename */
258 	{ &vop_mkdir_desc, genfs_eopnotsupp },		/* mkdir */
259 	{ &vop_rmdir_desc, genfs_eopnotsupp },		/* rmdir */
260 	{ &vop_symlink_desc, procfs_symlink },		/* symlink */
261 	{ &vop_readdir_desc, procfs_readdir },		/* readdir */
262 	{ &vop_readlink_desc, procfs_readlink },	/* readlink */
263 	{ &vop_abortop_desc, genfs_abortop },		/* abortop */
264 	{ &vop_inactive_desc, procfs_inactive },	/* inactive */
265 	{ &vop_reclaim_desc, procfs_reclaim },		/* reclaim */
266 	{ &vop_lock_desc, genfs_lock },			/* lock */
267 	{ &vop_unlock_desc, genfs_unlock },		/* unlock */
268 	{ &vop_bmap_desc, genfs_eopnotsupp },		/* bmap */
269 	{ &vop_strategy_desc, genfs_badop },		/* strategy */
270 	{ &vop_print_desc, procfs_print },		/* print */
271 	{ &vop_islocked_desc, genfs_islocked },		/* islocked */
272 	{ &vop_pathconf_desc, procfs_pathconf },	/* pathconf */
273 	{ &vop_advlock_desc, genfs_einval },		/* advlock */
274 	{ &vop_getpages_desc, procfs_getpages },	/* getpages */
275 	{ &vop_putpages_desc, genfs_null_putpages },	/* putpages */
276 	{ NULL, NULL }
277 };
278 const struct vnodeopv_desc procfs_vnodeop_opv_desc =
279 	{ &procfs_vnodeop_p, procfs_vnodeop_entries };
280 /*
281  * set things up for doing i/o on
282  * the pfsnode (vp).  (vp) is locked
283  * on entry, and should be left locked
284  * on exit.
285  *
286  * for procfs we don't need to do anything
287  * in particular for i/o.  all that is done
288  * is to support exclusive open on process
289  * memory images.
290  */
291 int
292 procfs_open(void *v)
293 {
294 	struct vop_open_args /* {
295 		struct vnode *a_vp;
296 		int  a_mode;
297 		kauth_cred_t a_cred;
298 	} */ *ap = v;
299 	struct vnode *vp = ap->a_vp;
300 	struct pfsnode *pfs = VTOPFS(vp);
301 	struct lwp *l1;
302 	struct proc *p2;
303 	int error;
304 
305 	if ((error =
306 	     procfs_proc_lock(vp->v_mount, pfs->pfs_pid, &p2, ENOENT)) != 0)
307 		return error;
308 
309 	l1 = curlwp;				/* tracer */
310 
311 #define	M2K(m)	(((m) & FREAD) && ((m) & FWRITE) ? \
312 		 KAUTH_REQ_PROCESS_PROCFS_RW : \
313 		 (m) & FWRITE ? KAUTH_REQ_PROCESS_PROCFS_WRITE : \
314 		 KAUTH_REQ_PROCESS_PROCFS_READ)
315 
316 	mutex_enter(p2->p_lock);
317 	error = kauth_authorize_process(l1->l_cred, KAUTH_PROCESS_PROCFS,
318 	    p2, pfs, KAUTH_ARG(M2K(ap->a_mode)), NULL);
319 	mutex_exit(p2->p_lock);
320 	if (error) {
321 		procfs_proc_unlock(p2);
322 		return (error);
323 	}
324 
325 #undef M2K
326 
327 	switch (pfs->pfs_type) {
328 	case PFSmem:
329 		if (((pfs->pfs_flags & FWRITE) && (ap->a_mode & O_EXCL)) ||
330 		    ((pfs->pfs_flags & O_EXCL) && (ap->a_mode & FWRITE))) {
331 			error = EBUSY;
332 			break;
333 		}
334 
335 		if (!proc_isunder(p2, l1)) {
336 			error = EPERM;
337 			break;
338 		}
339 
340 		if (ap->a_mode & FWRITE)
341 			pfs->pfs_flags = ap->a_mode & (FWRITE|O_EXCL);
342 
343 		break;
344 
345 	case PFSregs:
346 	case PFSfpregs:
347 		if (!proc_isunder(p2, l1)) {
348 			error = EPERM;
349 			break;
350 		}
351 		break;
352 
353 	default:
354 		break;
355 	}
356 
357 	procfs_proc_unlock(p2);
358 	return (error);
359 }
360 
361 /*
362  * close the pfsnode (vp) after doing i/o.
363  * (vp) is not locked on entry or exit.
364  *
365  * nothing to do for procfs other than undo
366  * any exclusive open flag (see _open above).
367  */
368 int
369 procfs_close(void *v)
370 {
371 	struct vop_close_args /* {
372 		struct vnode *a_vp;
373 		int  a_fflag;
374 		kauth_cred_t a_cred;
375 	} */ *ap = v;
376 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
377 
378 	switch (pfs->pfs_type) {
379 	case PFSmem:
380 		if ((ap->a_fflag & FWRITE) && (pfs->pfs_flags & O_EXCL))
381 			pfs->pfs_flags &= ~(FWRITE|O_EXCL);
382 		break;
383 
384 	default:
385 		break;
386 	}
387 
388 	return (0);
389 }
390 
391 /*
392  * _inactive is called when the pfsnode
393  * is vrele'd and the reference count goes
394  * to zero.  (vp) will be on the vnode free
395  * list, so to get it back vget() must be
396  * used.
397  *
398  * (vp) is locked on entry, but must be unlocked on exit.
399  */
400 int
401 procfs_inactive(void *v)
402 {
403 	struct vop_inactive_v2_args /* {
404 		struct vnode *a_vp;
405 		bool *a_recycle;
406 	} */ *ap = v;
407 	struct vnode *vp = ap->a_vp;
408 	struct pfsnode *pfs = VTOPFS(vp);
409 
410 	mutex_enter(&proc_lock);
411 	*ap->a_recycle = (procfs_proc_find(vp->v_mount, pfs->pfs_pid) == NULL);
412 	mutex_exit(&proc_lock);
413 
414 	return (0);
415 }
416 
417 /*
418  * _reclaim is called when getnewvnode()
419  * wants to make use of an entry on the vnode
420  * free list.  at this time the filesystem needs
421  * to free any private data and remove the node
422  * from any private lists.
423  */
424 int
425 procfs_reclaim(void *v)
426 {
427 	struct vop_reclaim_v2_args /* {
428 		struct vnode *a_vp;
429 	} */ *ap = v;
430 	struct vnode *vp = ap->a_vp;
431 	struct pfsnode *pfs = VTOPFS(vp);
432 
433 	VOP_UNLOCK(vp);
434 
435 	/*
436 	 * To interlock with procfs_revoke_vnodes().
437 	 */
438 	mutex_enter(vp->v_interlock);
439 	vp->v_data = NULL;
440 	mutex_exit(vp->v_interlock);
441 	kmem_free(pfs, sizeof(*pfs));
442 	return 0;
443 }
444 
445 /*
446  * Return POSIX pathconf information applicable to special devices.
447  */
448 int
449 procfs_pathconf(void *v)
450 {
451 	struct vop_pathconf_args /* {
452 		struct vnode *a_vp;
453 		int a_name;
454 		register_t *a_retval;
455 	} */ *ap = v;
456 
457 	switch (ap->a_name) {
458 	case _PC_LINK_MAX:
459 		*ap->a_retval = LINK_MAX;
460 		return (0);
461 	case _PC_MAX_CANON:
462 		*ap->a_retval = MAX_CANON;
463 		return (0);
464 	case _PC_MAX_INPUT:
465 		*ap->a_retval = MAX_INPUT;
466 		return (0);
467 	case _PC_PIPE_BUF:
468 		*ap->a_retval = PIPE_BUF;
469 		return (0);
470 	case _PC_CHOWN_RESTRICTED:
471 		*ap->a_retval = 1;
472 		return (0);
473 	case _PC_VDISABLE:
474 		*ap->a_retval = _POSIX_VDISABLE;
475 		return (0);
476 	case _PC_SYNC_IO:
477 		*ap->a_retval = 1;
478 		return (0);
479 	default:
480 		return genfs_pathconf(ap);
481 	}
482 	/* NOTREACHED */
483 }
484 
485 /*
486  * _print is used for debugging.
487  * just print a readable description
488  * of (vp).
489  */
490 int
491 procfs_print(void *v)
492 {
493 	struct vop_print_args /* {
494 		struct vnode *a_vp;
495 	} */ *ap = v;
496 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
497 
498 	printf("tag VT_PROCFS, type %d, pid %d, mode %x, flags %lx\n",
499 	    pfs->pfs_type, pfs->pfs_pid, pfs->pfs_mode, pfs->pfs_flags);
500 	return 0;
501 }
502 
503 int
504 procfs_link(void *v)
505 {
506 	struct vop_link_v2_args /* {
507 		struct vnode *a_dvp;
508 		struct vnode *a_vp;
509 		struct componentname *a_cnp;
510 	} */ *ap = v;
511 
512 	VOP_ABORTOP(ap->a_dvp, ap->a_cnp);
513 	return (EROFS);
514 }
515 
516 int
517 procfs_symlink(void *v)
518 {
519 	struct vop_symlink_v3_args /* {
520 		struct vnode *a_dvp;
521 		struct vnode **a_vpp;
522 		struct componentname *a_cnp;
523 		struct vattr *a_vap;
524 		char *a_target;
525 	} */ *ap = v;
526 
527 	VOP_ABORTOP(ap->a_dvp, ap->a_cnp);
528 	return (EROFS);
529 }
530 
531 /*
532  * Works out the path to the target process's current
533  * working directory or chroot.  If the caller is in a chroot and
534  * can't "reach" the target's cwd or root (or some other error
535  * occurs), a "/" is returned for the path.
536  */
537 static void
538 procfs_dir(pfstype t, struct lwp *caller, struct proc *target, char **bpp,
539     char *path, size_t len)
540 {
541 	struct cwdinfo *cwdi;
542 	struct vnode *vp, *rvp;
543 	char *bp;
544 
545 	/*
546 	 * Lock target cwdi and take a reference to the vnode
547 	 * we are interested in to prevent it from disappearing
548 	 * before getcwd_common() below.
549 	 */
550 	rw_enter(&target->p_cwdi->cwdi_lock, RW_READER);
551 	switch (t) {
552 	case PFScwd:
553 		vp = target->p_cwdi->cwdi_cdir;
554 		break;
555 	case PFSchroot:
556 		vp = target->p_cwdi->cwdi_rdir;
557 		break;
558 	default:
559 		rw_exit(&target->p_cwdi->cwdi_lock);
560 		return;
561 	}
562 	if (vp != NULL)
563 		vref(vp);
564 	rw_exit(&target->p_cwdi->cwdi_lock);
565 
566 	cwdi = caller->l_proc->p_cwdi;
567 	rw_enter(&cwdi->cwdi_lock, RW_READER);
568 
569 	rvp = cwdi->cwdi_rdir;
570 	bp = bpp ? *bpp : NULL;
571 
572 	/*
573 	 * XXX: this horrible kludge avoids locking panics when
574 	 * attempting to lookup links that point to within procfs
575 	 */
576 	if (vp != NULL && vp->v_tag == VT_PROCFS) {
577 		if (bpp) {
578 			*--bp = '/';
579 			*bpp = bp;
580 		}
581 		vrele(vp);
582 		rw_exit(&cwdi->cwdi_lock);
583 		return;
584 	}
585 
586 	if (rvp == NULL)
587 		rvp = rootvnode;
588 	if (vp == NULL || getcwd_common(vp, rvp, bp ? &bp : NULL, path,
589 	    len / 2, 0, caller) != 0) {
590 		if (bpp) {
591 			bp = *bpp;
592 			*--bp = '/';
593 		}
594 	}
595 
596 	if (bpp)
597 		*bpp = bp;
598 
599 	if (vp != NULL)
600 		vrele(vp);
601 	rw_exit(&cwdi->cwdi_lock);
602 }
603 
604 /*
605  * Invent attributes for pfsnode (vp) and store
606  * them in (vap).
607  * Directories lengths are returned as zero since
608  * any real length would require the genuine size
609  * to be computed, and nothing cares anyway.
610  *
611  * this is relatively minimal for procfs.
612  */
613 int
614 procfs_getattr(void *v)
615 {
616 	struct vop_getattr_args /* {
617 		struct vnode *a_vp;
618 		struct vattr *a_vap;
619 		kauth_cred_t a_cred;
620 	} */ *ap = v;
621 	struct vnode *vp = ap->a_vp;
622 	struct pfsnode *pfs = VTOPFS(vp);
623 	struct vattr *vap = ap->a_vap;
624 	struct proc *procp;
625 	char *path, *bp, bf[16];
626 	int error;
627 
628 	/* first check the process still exists */
629 	switch (pfs->pfs_type) {
630 	case PFSroot:
631 	case PFScurproc:
632 	case PFSself:
633 		procp = NULL;
634 		break;
635 
636 	default:
637 		error =
638 		    procfs_proc_lock(vp->v_mount, pfs->pfs_pid, &procp, ENOENT);
639 		if (error != 0)
640 			return (error);
641 		break;
642 	}
643 
644 	switch (pfs->pfs_type) {
645 	case PFStask:
646 		if (pfs->pfs_fd == -1) {
647 			path = NULL;
648 			break;
649 		}
650 		/*FALLTHROUGH*/
651 	case PFScwd:
652 	case PFSchroot:
653 		path = malloc(MAXPATHLEN + 4, M_TEMP, M_WAITOK);
654 		if (path == NULL && procp != NULL) {
655 			procfs_proc_unlock(procp);
656 			return (ENOMEM);
657 		}
658 		break;
659 
660 	default:
661 		path = NULL;
662 		break;
663 	}
664 
665 	if (procp != NULL) {
666 		mutex_enter(procp->p_lock);
667 		error = kauth_authorize_process(kauth_cred_get(),
668 		    KAUTH_PROCESS_CANSEE, procp,
669 		    KAUTH_ARG(KAUTH_REQ_PROCESS_CANSEE_ENTRY), NULL, NULL);
670 		mutex_exit(procp->p_lock);
671 		if (error != 0) {
672 		    	procfs_proc_unlock(procp);
673 		    	if (path != NULL)
674 		    		free(path, M_TEMP);
675 			return (ENOENT);
676 		}
677 	}
678 
679 	error = 0;
680 
681 	/* start by zeroing out the attributes */
682 	vattr_null(vap);
683 
684 	/* next do all the common fields */
685 	vap->va_type = ap->a_vp->v_type;
686 	vap->va_mode = pfs->pfs_mode;
687 	vap->va_fileid = pfs->pfs_fileno;
688 	vap->va_flags = 0;
689 	vap->va_blocksize = PAGE_SIZE;
690 
691 	/*
692 	 * Make all times be current TOD.
693 	 *
694 	 * It would be possible to get the process start
695 	 * time from the p_stats structure, but there's
696 	 * no "file creation" time stamp anyway, and the
697 	 * p_stats structure is not addressable if u. gets
698 	 * swapped out for that process.
699 	 */
700 	getnanotime(&vap->va_ctime);
701 	vap->va_atime = vap->va_mtime = vap->va_ctime;
702 	if (procp)
703 		TIMEVAL_TO_TIMESPEC(&procp->p_stats->p_start,
704 		    &vap->va_birthtime);
705 	else
706 		getnanotime(&vap->va_birthtime);
707 
708 	switch (pfs->pfs_type) {
709 	case PFSmem:
710 	case PFSregs:
711 	case PFSfpregs:
712 #if defined(__HAVE_PROCFS_MACHDEP) && defined(PROCFS_MACHDEP_PROTECT_CASES)
713 	PROCFS_MACHDEP_PROTECT_CASES
714 #endif
715 		/*
716 		 * If the process has exercised some setuid or setgid
717 		 * privilege, then rip away read/write permission so
718 		 * that only root can gain access.
719 		 */
720 		if (procp->p_flag & PK_SUGID)
721 			vap->va_mode &= ~(S_IRUSR|S_IWUSR);
722 		/* FALLTHROUGH */
723 	case PFSstatus:
724 	case PFSstat:
725 	case PFSnote:
726 	case PFSnotepg:
727 	case PFScmdline:
728 	case PFSenviron:
729 	case PFSemul:
730 	case PFSstatm:
731 
732 	case PFSmap:
733 	case PFSmaps:
734 	case PFSlimit:
735 	case PFSauxv:
736 		vap->va_nlink = 1;
737 		vap->va_uid = kauth_cred_geteuid(procp->p_cred);
738 		vap->va_gid = kauth_cred_getegid(procp->p_cred);
739 		break;
740 	case PFScwd:
741 	case PFSchroot:
742 	case PFSmeminfo:
743 	case PFSdevices:
744 	case PFScpuinfo:
745 	case PFSuptime:
746 	case PFSmounts:
747 	case PFScpustat:
748 	case PFSloadavg:
749 	case PFSversion:
750 	case PFSexe:
751 	case PFSself:
752 	case PFScurproc:
753 	case PFSroot:
754 		vap->va_nlink = 1;
755 		vap->va_uid = vap->va_gid = 0;
756 		break;
757 
758 	case PFSproc:
759 	case PFStask:
760 	case PFSfile:
761 	case PFSfd:
762 		break;
763 
764 	default:
765 		panic("%s: %d/1", __func__, pfs->pfs_type);
766 	}
767 
768 	/*
769 	 * now do the object specific fields
770 	 *
771 	 * The size could be set from struct reg, but it's hardly
772 	 * worth the trouble, and it puts some (potentially) machine
773 	 * dependent data into this machine-independent code.  If it
774 	 * becomes important then this function should break out into
775 	 * a per-file stat function in the corresponding .c file.
776 	 */
777 
778 	switch (pfs->pfs_type) {
779 	case PFSroot:
780 		vap->va_bytes = vap->va_size = DEV_BSIZE;
781 		break;
782 
783 	case PFSself:
784 	case PFScurproc:
785 		vap->va_bytes = vap->va_size =
786 		    snprintf(bf, sizeof(bf), "%ld", (long)curproc->p_pid);
787 		break;
788 	case PFStask:
789 		if (pfs->pfs_fd != -1) {
790 			vap->va_nlink = 1;
791 			vap->va_uid = 0;
792 			vap->va_gid = 0;
793 			vap->va_bytes = vap->va_size =
794 			    snprintf(bf, sizeof(bf), "..");
795 			break;
796 		}
797 		/*FALLTHROUGH*/
798 	case PFSfd:
799 		if (pfs->pfs_fd != -1) {
800 			file_t *fp;
801 
802 			fp = fd_getfile2(procp, pfs->pfs_fd);
803 			if (fp == NULL) {
804 				error = EBADF;
805 				break;
806 			}
807 			vap->va_nlink = 1;
808 			vap->va_uid = kauth_cred_geteuid(fp->f_cred);
809 			vap->va_gid = kauth_cred_getegid(fp->f_cred);
810 			switch (fp->f_type) {
811 			case DTYPE_VNODE:
812 				vap->va_bytes = vap->va_size =
813 				    fp->f_vnode->v_size;
814 				break;
815 			default:
816 				vap->va_bytes = vap->va_size = 0;
817 				break;
818 			}
819 			closef(fp);
820 			break;
821 		}
822 		/*FALLTHROUGH*/
823 	case PFSproc:
824 		vap->va_nlink = 2;
825 		vap->va_uid = kauth_cred_geteuid(procp->p_cred);
826 		vap->va_gid = kauth_cred_getegid(procp->p_cred);
827 		vap->va_bytes = vap->va_size = DEV_BSIZE;
828 		break;
829 
830 	case PFSfile:
831 		error = EOPNOTSUPP;
832 		break;
833 
834 	case PFSmem:
835 		vap->va_bytes = vap->va_size =
836 			ctob(procp->p_vmspace->vm_tsize +
837 				    procp->p_vmspace->vm_dsize +
838 				    procp->p_vmspace->vm_ssize);
839 		break;
840 
841 	case PFSauxv:
842 		vap->va_bytes = vap->va_size = procp->p_execsw->es_arglen;
843 		break;
844 
845 #if defined(PT_GETREGS) || defined(PT_SETREGS)
846 	case PFSregs:
847 		vap->va_bytes = vap->va_size = sizeof(struct reg);
848 		break;
849 #endif
850 
851 #if defined(PT_GETFPREGS) || defined(PT_SETFPREGS)
852 	case PFSfpregs:
853 		vap->va_bytes = vap->va_size = sizeof(struct fpreg);
854 		break;
855 #endif
856 
857 	case PFSstatus:
858 	case PFSstat:
859 	case PFSnote:
860 	case PFSnotepg:
861 	case PFScmdline:
862 	case PFSenviron:
863 	case PFSmeminfo:
864 	case PFSdevices:
865 	case PFScpuinfo:
866 	case PFSuptime:
867 	case PFSmounts:
868 	case PFScpustat:
869 	case PFSloadavg:
870 	case PFSstatm:
871 	case PFSversion:
872 		vap->va_bytes = vap->va_size = 0;
873 		break;
874 	case PFSlimit:
875 	case PFSmap:
876 	case PFSmaps:
877 		/*
878 		 * Advise a larger blocksize for the map files, so that
879 		 * they may be read in one pass.
880 		 */
881 		vap->va_blocksize = 4 * PAGE_SIZE;
882 		vap->va_bytes = vap->va_size = 0;
883 		break;
884 
885 	case PFScwd:
886 	case PFSchroot:
887 		bp = path + MAXPATHLEN;
888 		*--bp = '\0';
889 		procfs_dir(pfs->pfs_type, curlwp, procp, &bp, path,
890 		     MAXPATHLEN);
891 		vap->va_bytes = vap->va_size = strlen(bp);
892 		break;
893 
894 	case PFSexe:
895 		vap->va_bytes = vap->va_size = strlen(procp->p_path);
896 		break;
897 
898 	case PFSemul:
899 		vap->va_bytes = vap->va_size = strlen(procp->p_emul->e_name);
900 		break;
901 
902 #ifdef __HAVE_PROCFS_MACHDEP
903 	PROCFS_MACHDEP_NODETYPE_CASES
904 		error = procfs_machdep_getattr(ap->a_vp, vap, procp);
905 		break;
906 #endif
907 
908 	default:
909 		panic("%s: %d/2", __func__, pfs->pfs_type);
910 	}
911 
912 	if (procp != NULL)
913 		procfs_proc_unlock(procp);
914 	if (path != NULL)
915 		free(path, M_TEMP);
916 
917 	return (error);
918 }
919 
920 /*ARGSUSED*/
921 int
922 procfs_setattr(void *v)
923 {
924 	/*
925 	 * just fake out attribute setting
926 	 * it's not good to generate an error
927 	 * return, otherwise things like creat()
928 	 * will fail when they try to set the
929 	 * file length to 0.  worse, this means
930 	 * that echo $note > /proc/$pid/note will fail.
931 	 */
932 
933 	return (0);
934 }
935 
936 /*
937  * implement access checking.
938  *
939  * actually, the check for super-user is slightly
940  * broken since it will allow read access to write-only
941  * objects.  this doesn't cause any particular trouble
942  * but does mean that the i/o entry points need to check
943  * that the operation really does make sense.
944  */
945 int
946 procfs_access(void *v)
947 {
948 	struct vop_access_args /* {
949 		struct vnode *a_vp;
950 		accmode_t a_accmode;
951 		kauth_cred_t a_cred;
952 	} */ *ap = v;
953 	struct vattr va;
954 	int error;
955 
956 	if ((error = VOP_GETATTR(ap->a_vp, &va, ap->a_cred)) != 0)
957 		return (error);
958 
959 	return kauth_authorize_vnode(ap->a_cred,
960 	    KAUTH_ACCESS_ACTION(ap->a_accmode, ap->a_vp->v_type, va.va_mode),
961 	    ap->a_vp, NULL, genfs_can_access(ap->a_vp, ap->a_cred,
962 	    va.va_uid, va.va_gid, va.va_mode, NULL, ap->a_accmode));
963 }
964 
965 /*
966  * lookup.  this is incredibly complicated in the
967  * general case, however for most pseudo-filesystems
968  * very little needs to be done.
969  *
970  * Locking isn't hard here, just poorly documented.
971  *
972  * If we're looking up ".", just vref the parent & return it.
973  *
974  * If we're looking up "..", unlock the parent, and lock "..". If everything
975  * went ok, and we're on the last component and the caller requested the
976  * parent locked, try to re-lock the parent. We do this to prevent lock
977  * races.
978  *
979  * For anything else, get the needed node. Then unlock the parent if not
980  * the last component or not LOCKPARENT (i.e. if we wouldn't re-lock the
981  * parent in the .. case).
982  *
983  * We try to exit with the parent locked in error cases.
984  */
985 int
986 procfs_lookup(void *v)
987 {
988 	struct vop_lookup_v2_args /* {
989 		struct vnode * a_dvp;
990 		struct vnode ** a_vpp;
991 		struct componentname * a_cnp;
992 	} */ *ap = v;
993 	struct componentname *cnp = ap->a_cnp;
994 	struct vnode **vpp = ap->a_vpp;
995 	struct vnode *dvp = ap->a_dvp;
996 	const char *pname = cnp->cn_nameptr;
997 	const struct proc_target *pt = NULL;
998 	struct vnode *fvp;
999 	pid_t pid, vnpid;
1000 	struct pfsnode *pfs;
1001 	struct proc *p = NULL;
1002 	struct lwp *plwp;
1003 	int i, error;
1004 	pfstype type;
1005 
1006 	*vpp = NULL;
1007 
1008 	if (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)
1009 		return (EROFS);
1010 
1011 	if (cnp->cn_namelen == 1 && *pname == '.') {
1012 		*vpp = dvp;
1013 		vref(dvp);
1014 		return (0);
1015 	}
1016 
1017 	pfs = VTOPFS(dvp);
1018 	switch (pfs->pfs_type) {
1019 	case PFSroot:
1020 		/*
1021 		 * Shouldn't get here with .. in the root node.
1022 		 */
1023 		if (cnp->cn_flags & ISDOTDOT)
1024 			return (EIO);
1025 
1026 		for (i = 0; i < nproc_root_targets; i++) {
1027 			pt = &proc_root_targets[i];
1028 			/*
1029 			 * check for node match.  proc is always NULL here,
1030 			 * so call pt_valid with constant NULL lwp.
1031 			 */
1032 			if (cnp->cn_namelen == pt->pt_namlen &&
1033 			    memcmp(pt->pt_name, pname, cnp->cn_namelen) == 0 &&
1034 			    (pt->pt_valid == NULL ||
1035 			     (*pt->pt_valid)(NULL, dvp->v_mount)))
1036 				break;
1037 		}
1038 
1039 		if (i != nproc_root_targets) {
1040 			error = procfs_allocvp(dvp->v_mount, vpp, 0,
1041 			    pt->pt_pfstype, -1);
1042 			return (error);
1043 		}
1044 
1045 		if (CNEQ(cnp, "curproc", 7)) {
1046 			pid = curproc->p_pid;
1047 			vnpid = 0;
1048 			type = PFScurproc;
1049 		} else if (CNEQ(cnp, "self", 4)) {
1050 			pid = curproc->p_pid;
1051 			vnpid = 0;
1052 			type = PFSself;
1053 		} else {
1054 			pid = (pid_t)atoi(pname, cnp->cn_namelen);
1055 			vnpid = pid;
1056 			type = PFSproc;
1057 		}
1058 
1059 		if (procfs_proc_lock(dvp->v_mount, pid, &p, ESRCH) != 0)
1060 			break;
1061 		error = procfs_allocvp(dvp->v_mount, vpp, vnpid, type, -1);
1062 		procfs_proc_unlock(p);
1063 		return (error);
1064 
1065 	case PFSproc:
1066 		if (cnp->cn_flags & ISDOTDOT) {
1067 			error = procfs_allocvp(dvp->v_mount, vpp, 0, PFSroot,
1068 			    -1);
1069 			return (error);
1070 		}
1071 
1072 		if (procfs_proc_lock(dvp->v_mount, pfs->pfs_pid, &p,
1073 				     ESRCH) != 0)
1074 			break;
1075 
1076 		mutex_enter(p->p_lock);
1077 		LIST_FOREACH(plwp, &p->p_lwps, l_sibling) {
1078 			if (plwp->l_stat != LSZOMB)
1079 				break;
1080 		}
1081 		/* Process is exiting if no-LWPS or all LWPs are LSZOMB */
1082 		if (plwp == NULL) {
1083 			mutex_exit(p->p_lock);
1084 			procfs_proc_unlock(p);
1085 			return ESRCH;
1086 		}
1087 
1088 		lwp_addref(plwp);
1089 		mutex_exit(p->p_lock);
1090 
1091 		for (pt = proc_targets, i = 0; i < nproc_targets; pt++, i++) {
1092 			int found;
1093 
1094 			found = cnp->cn_namelen == pt->pt_namlen &&
1095 			    memcmp(pt->pt_name, pname, cnp->cn_namelen) == 0 &&
1096 			    (pt->pt_valid == NULL
1097 			      || (*pt->pt_valid)(plwp, dvp->v_mount));
1098 			if (found)
1099 				break;
1100 		}
1101 		lwp_delref(plwp);
1102 
1103 		if (i == nproc_targets) {
1104 			procfs_proc_unlock(p);
1105 			break;
1106 		}
1107 		if (pt->pt_pfstype == PFSfile) {
1108 			fvp = p->p_textvp;
1109 			/* We already checked that it exists. */
1110 			vref(fvp);
1111 			procfs_proc_unlock(p);
1112 			*vpp = fvp;
1113 			return (0);
1114 		}
1115 
1116 		error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
1117 		    pt->pt_pfstype, -1);
1118 		procfs_proc_unlock(p);
1119 		return (error);
1120 
1121 	case PFSfd: {
1122 		int fd;
1123 		file_t *fp;
1124 
1125 		if ((error = procfs_proc_lock(dvp->v_mount, pfs->pfs_pid, &p,
1126 					      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(dvp->v_mount, pfs->pfs_pid, &p,
1163 					      ENOENT)) != 0)
1164 			return error;
1165 
1166 		if (cnp->cn_flags & ISDOTDOT) {
1167 			error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
1168 			    PFSproc, -1);
1169 			procfs_proc_unlock(p);
1170 			return (error);
1171 		}
1172 		xpid = atoi(pname, cnp->cn_namelen);
1173 
1174 		if (xpid != pfs->pfs_pid) {
1175 			procfs_proc_unlock(p);
1176 			return ENOENT;
1177 		}
1178 		error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
1179 		    PFStask, 0);
1180 		procfs_proc_unlock(p);
1181 		return error;
1182 	}
1183 	default:
1184 		return (ENOTDIR);
1185 	}
1186 
1187 	return (cnp->cn_nameiop == LOOKUP ? ENOENT : EROFS);
1188 }
1189 
1190 int
1191 procfs_validfile(struct lwp *l, struct mount *mp)
1192 {
1193 	return l != NULL && l->l_proc != NULL && l->l_proc->p_textvp != NULL;
1194 }
1195 
1196 static int
1197 procfs_validfile_linux(struct lwp *l, struct mount *mp)
1198 {
1199 	return procfs_use_linux_compat(mp) &&
1200 	    (l == NULL || l->l_proc == NULL || procfs_validfile(l, mp));
1201 }
1202 
1203 struct procfs_root_readdir_ctx {
1204 	struct uio *uiop;
1205 	off_t *cookies;
1206 	int ncookies;
1207 	off_t off;
1208 	off_t startoff;
1209 	int error;
1210 };
1211 
1212 static int
1213 procfs_root_readdir_callback(struct proc *p, void *arg)
1214 {
1215 	struct procfs_root_readdir_ctx *ctxp = arg;
1216 	struct dirent d;
1217 	struct uio *uiop;
1218 	int error;
1219 
1220 	uiop = ctxp->uiop;
1221 	if (uiop->uio_resid < UIO_MX)
1222 		return -1; /* no space */
1223 
1224 	if (kauth_authorize_process(kauth_cred_get(),
1225 	    KAUTH_PROCESS_CANSEE, p,
1226 	    KAUTH_ARG(KAUTH_REQ_PROCESS_CANSEE_ENTRY), NULL, NULL) != 0)
1227 		return 0;
1228 
1229 	if (ctxp->off < ctxp->startoff) {
1230 		ctxp->off++;
1231 		return 0;
1232 	}
1233 
1234 	memset(&d, 0, UIO_MX);
1235 	d.d_reclen = UIO_MX;
1236 	d.d_fileno = PROCFS_FILENO(p->p_pid, PFSproc, -1);
1237 	d.d_namlen = snprintf(d.d_name,
1238 	    UIO_MX - offsetof(struct dirent, d_name), "%ld", (long)p->p_pid);
1239 	d.d_type = DT_DIR;
1240 
1241 	mutex_exit(&proc_lock);
1242 	error = uiomove(&d, UIO_MX, uiop);
1243 	mutex_enter(&proc_lock);
1244 	if (error) {
1245 		ctxp->error = error;
1246 		return -1;
1247 	}
1248 
1249 	ctxp->ncookies++;
1250 	if (ctxp->cookies)
1251 		*(ctxp->cookies)++ = ctxp->off + 1;
1252 	ctxp->off++;
1253 
1254 	return 0;
1255 }
1256 
1257 /*
1258  * readdir returns directory entries from pfsnode (vp).
1259  *
1260  * the strategy here with procfs is to generate a single
1261  * directory entry at a time (struct dirent) and then
1262  * copy that out to userland using uiomove.  a more efficient
1263  * though more complex implementation, would try to minimize
1264  * the number of calls to uiomove().  for procfs, this is
1265  * hardly worth the added code complexity.
1266  *
1267  * this should just be done through read()
1268  */
1269 int
1270 procfs_readdir(void *v)
1271 {
1272 	struct vop_readdir_args /* {
1273 		struct vnode *a_vp;
1274 		struct uio *a_uio;
1275 		kauth_cred_t a_cred;
1276 		int *a_eofflag;
1277 		off_t **a_cookies;
1278 		int *a_ncookies;
1279 	} */ *ap = v;
1280 	struct uio *uio = ap->a_uio;
1281 	struct dirent d;
1282 	struct pfsnode *pfs;
1283 	off_t i;
1284 	int error;
1285 	off_t *cookies = NULL;
1286 	int ncookies;
1287 	struct vnode *vp;
1288 	const struct proc_target *pt;
1289 	struct procfs_root_readdir_ctx ctx;
1290 	struct lwp *l;
1291 	int nfd;
1292 
1293 	vp = ap->a_vp;
1294 	pfs = VTOPFS(vp);
1295 
1296 	if (uio->uio_resid < UIO_MX)
1297 		return (EINVAL);
1298 	if (uio->uio_offset < 0)
1299 		return (EINVAL);
1300 
1301 	error = 0;
1302 	i = uio->uio_offset;
1303 	memset(&d, 0, UIO_MX);
1304 	d.d_reclen = UIO_MX;
1305 	ncookies = uio->uio_resid / UIO_MX;
1306 
1307 	switch (pfs->pfs_type) {
1308 	/*
1309 	 * this is for the process-specific sub-directories.
1310 	 * all that is needed to is copy out all the entries
1311 	 * from the procent[] table (top of this file).
1312 	 */
1313 	case PFSproc: {
1314 		struct proc *p;
1315 
1316 		if (i >= nproc_targets)
1317 			return 0;
1318 
1319 		if (procfs_proc_lock(vp->v_mount, pfs->pfs_pid, &p, ESRCH) != 0)
1320 			break;
1321 
1322 		if (ap->a_ncookies) {
1323 			ncookies = uimin(ncookies, (nproc_targets - i));
1324 			cookies = malloc(ncookies * sizeof (off_t),
1325 			    M_TEMP, M_WAITOK);
1326 			*ap->a_cookies = cookies;
1327 		}
1328 
1329 		for (pt = &proc_targets[i];
1330 		     uio->uio_resid >= UIO_MX && i < nproc_targets; pt++, i++) {
1331 			if (pt->pt_valid) {
1332 				/* XXXSMP LWP can disappear */
1333 				mutex_enter(p->p_lock);
1334 				l = LIST_FIRST(&p->p_lwps);
1335 				KASSERT(l != NULL);
1336 				mutex_exit(p->p_lock);
1337 				if ((*pt->pt_valid)(l, vp->v_mount) == 0)
1338 					continue;
1339 			}
1340 
1341 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid,
1342 			    pt->pt_pfstype, -1);
1343 			d.d_namlen = pt->pt_namlen;
1344 			memcpy(d.d_name, pt->pt_name, pt->pt_namlen + 1);
1345 			d.d_type = pt->pt_type;
1346 
1347 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1348 				break;
1349 			if (cookies)
1350 				*cookies++ = i + 1;
1351 		}
1352 
1353 		procfs_proc_unlock(p);
1354 	    	break;
1355 	}
1356 	case PFSfd: {
1357 		struct proc *p;
1358 		file_t *fp;
1359 		int lim, nc = 0;
1360 
1361 		if ((error = procfs_proc_lock(vp->v_mount, pfs->pfs_pid, &p,
1362 					      ESRCH)) != 0)
1363 			return error;
1364 
1365 		/* XXX Should this be by file as well? */
1366 		if (kauth_authorize_process(kauth_cred_get(),
1367 		    KAUTH_PROCESS_CANSEE, p,
1368 		    KAUTH_ARG(KAUTH_REQ_PROCESS_CANSEE_OPENFILES), NULL,
1369 		    NULL) != 0) {
1370 		    	procfs_proc_unlock(p);
1371 			return ESRCH;
1372 		}
1373 
1374 		nfd = atomic_load_consume(&p->p_fd->fd_dt)->dt_nfiles;
1375 
1376 		lim = uimin((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfiles);
1377 		if (i >= lim) {
1378 		    	procfs_proc_unlock(p);
1379 			return 0;
1380 		}
1381 
1382 		if (ap->a_ncookies) {
1383 			ncookies = uimin(ncookies, (nfd + 2 - i));
1384 			cookies = malloc(ncookies * sizeof (off_t),
1385 			    M_TEMP, M_WAITOK);
1386 			*ap->a_cookies = cookies;
1387 		}
1388 
1389 		for (; i < 2 && uio->uio_resid >= UIO_MX; i++) {
1390 			pt = &proc_targets[i];
1391 			d.d_namlen = pt->pt_namlen;
1392 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid,
1393 			    pt->pt_pfstype, -1);
1394 			(void)memcpy(d.d_name, pt->pt_name, pt->pt_namlen + 1);
1395 			d.d_type = pt->pt_type;
1396 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1397 				break;
1398 			if (cookies)
1399 				*cookies++ = i + 1;
1400 			nc++;
1401 		}
1402 		if (error) {
1403 			ncookies = nc;
1404 			break;
1405 		}
1406 		for (; uio->uio_resid >= UIO_MX && i < nfd; i++) {
1407 			/* check the descriptor exists */
1408 			if ((fp = fd_getfile2(p, i - 2)) == NULL)
1409 				continue;
1410 			closef(fp);
1411 
1412 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid, PFSfd, i - 2);
1413 			d.d_namlen = snprintf(d.d_name, sizeof(d.d_name),
1414 			    "%lld", (long long)(i - 2));
1415 			d.d_type = VREG;
1416 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1417 				break;
1418 			if (cookies)
1419 				*cookies++ = i + 1;
1420 			nc++;
1421 		}
1422 		ncookies = nc;
1423 		procfs_proc_unlock(p);
1424 		break;
1425 	}
1426 	case PFStask: {
1427 		struct proc *p;
1428 		int nc = 0;
1429 
1430 		if ((error = procfs_proc_lock(vp->v_mount, pfs->pfs_pid, &p,
1431 					      ESRCH)) != 0)
1432 			return error;
1433 
1434 		nfd = 3;	/* ., .., pid */
1435 
1436 		if (ap->a_ncookies) {
1437 			ncookies = uimin(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 		struct proc *p;
1487 		int nc = 0;
1488 
1489 		if (ap->a_ncookies) {
1490 			/*
1491 			 * XXX Potentially allocating too much space here,
1492 			 * but I'm lazy. This loop needs some work.
1493 			 */
1494 			cookies = malloc(ncookies * sizeof (off_t),
1495 			    M_TEMP, M_WAITOK);
1496 			*ap->a_cookies = cookies;
1497 		}
1498 		error = 0;
1499 		/* 0 ... 3 are static entries. */
1500 		for (; i <= 3 && uio->uio_resid >= UIO_MX; i++) {
1501 			switch (i) {
1502 			case 0:		/* `.' */
1503 			case 1:		/* `..' */
1504 				d.d_fileno = PROCFS_FILENO(0, PFSroot, -1);
1505 				d.d_namlen = i + 1;
1506 				memcpy(d.d_name, "..", d.d_namlen);
1507 				d.d_name[i + 1] = '\0';
1508 				d.d_type = DT_DIR;
1509 				break;
1510 
1511 			case 2:
1512 				d.d_fileno = PROCFS_FILENO(0, PFScurproc, -1);
1513 				d.d_namlen = sizeof("curproc") - 1;
1514 				memcpy(d.d_name, "curproc", sizeof("curproc"));
1515 				d.d_type = DT_LNK;
1516 				break;
1517 
1518 			case 3:
1519 				d.d_fileno = PROCFS_FILENO(0, PFSself, -1);
1520 				d.d_namlen = sizeof("self") - 1;
1521 				memcpy(d.d_name, "self", sizeof("self"));
1522 				d.d_type = DT_LNK;
1523 				break;
1524 			}
1525 
1526 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1527 				break;
1528 			nc++;
1529 			if (cookies)
1530 				*cookies++ = i + 1;
1531 		}
1532 		/* 4 ... are process entries. */
1533 		ctx.uiop = uio;
1534 		ctx.error = 0;
1535 		ctx.off = 4;
1536 		ctx.startoff = i;
1537 		ctx.cookies = cookies;
1538 		ctx.ncookies = nc;
1539 		proclist_foreach_call(&allproc,
1540 		    procfs_root_readdir_callback, &ctx);
1541 		cookies = ctx.cookies;
1542 		nc = ctx.ncookies;
1543 		error = ctx.error;
1544 		if (error)
1545 			break;
1546 
1547 		/* misc entries. */
1548 		if (i < ctx.off)
1549 			i = ctx.off;
1550 		if (i >= ctx.off + nproc_root_targets)
1551 			break;
1552 		error = procfs_proc_lock(vp->v_mount, pfs->pfs_pid, &p, ESRCH);
1553 		if (error)
1554 			break;
1555 		for (pt = &proc_root_targets[i - ctx.off];
1556 		    uio->uio_resid >= UIO_MX &&
1557 		    pt < &proc_root_targets[nproc_root_targets];
1558 		    pt++, i++) {
1559 			if (pt->pt_valid &&
1560 			    (*pt->pt_valid)(NULL, vp->v_mount) == 0)
1561 				continue;
1562 			if (kauth_authorize_process(kauth_cred_get(),
1563 			    KAUTH_PROCESS_CANSEE, p,
1564 			    KAUTH_ARG(KAUTH_REQ_PROCESS_CANSEE_ENTRY),
1565 			    NULL, NULL) != 0)
1566 				continue;
1567 			d.d_fileno = PROCFS_FILENO(0, pt->pt_pfstype, -1);
1568 			d.d_namlen = pt->pt_namlen;
1569 			memcpy(d.d_name, pt->pt_name, pt->pt_namlen + 1);
1570 			d.d_type = pt->pt_type;
1571 
1572 			if ((error = uiomove(&d, UIO_MX, uio)) != 0)
1573 				break;
1574 			nc++;
1575 			if (cookies)
1576 				*cookies++ = i + 1;
1577 		}
1578 
1579 		ncookies = nc;
1580 		procfs_proc_unlock(p);
1581 		break;
1582 	}
1583 
1584 	default:
1585 		error = ENOTDIR;
1586 		break;
1587 	}
1588 
1589 	if (ap->a_ncookies) {
1590 		if (error) {
1591 			if (cookies)
1592 				free(*ap->a_cookies, M_TEMP);
1593 			*ap->a_ncookies = 0;
1594 			*ap->a_cookies = NULL;
1595 		} else
1596 			*ap->a_ncookies = ncookies;
1597 	}
1598 	uio->uio_offset = i;
1599 	return (error);
1600 }
1601 
1602 /*
1603  * readlink reads the link of `curproc' and others
1604  */
1605 int
1606 procfs_readlink(void *v)
1607 {
1608 	struct vop_readlink_args *ap = v;
1609 	char bf[16];		/* should be enough */
1610 	char *bp = bf;
1611 	char *path = NULL;
1612 	int len = 0;
1613 	int error = 0;
1614 	struct vnode *vp = ap->a_vp;
1615 	struct pfsnode *pfs = VTOPFS(vp);
1616 	struct proc *pown = NULL;
1617 
1618 	if (pfs->pfs_fileno == PROCFS_FILENO(0, PFScurproc, -1))
1619 		len = snprintf(bf, sizeof(bf), "%ld", (long)curproc->p_pid);
1620 	else if (pfs->pfs_fileno == PROCFS_FILENO(0, PFSself, -1))
1621 		len = snprintf(bf, sizeof(bf), "%s", "curproc");
1622 	else if (pfs->pfs_fileno == PROCFS_FILENO(pfs->pfs_pid, PFStask, 0))
1623 		len = snprintf(bf, sizeof(bf), "..");
1624 	else if (pfs->pfs_fileno == PROCFS_FILENO(pfs->pfs_pid, PFSexe, -1)) {
1625 		if ((error = procfs_proc_lock(vp->v_mount, pfs->pfs_pid, &pown,
1626 					      ESRCH)) != 0)
1627 			return error;
1628 		bp = pown->p_path;
1629 		len = strlen(bp);
1630 	} else if (pfs->pfs_fileno == PROCFS_FILENO(pfs->pfs_pid, PFScwd, -1) ||
1631 	    pfs->pfs_fileno == PROCFS_FILENO(pfs->pfs_pid, PFSchroot, -1)) {
1632 		if ((error = procfs_proc_lock(vp->v_mount, pfs->pfs_pid, &pown,
1633 					      ESRCH)) != 0)
1634 			return error;
1635 		path = malloc(MAXPATHLEN + 4, M_TEMP, M_WAITOK);
1636 		if (path == NULL) {
1637 			procfs_proc_unlock(pown);
1638 			return (ENOMEM);
1639 		}
1640 		bp = path + MAXPATHLEN;
1641 		*--bp = '\0';
1642 		procfs_dir(PROCFS_TYPE(pfs->pfs_fileno), curlwp, pown,
1643 		    &bp, path, MAXPATHLEN);
1644 		len = strlen(bp);
1645 	} else {
1646 		file_t *fp;
1647 		struct vnode *vxp;
1648 
1649 		if ((error = procfs_proc_lock(vp->v_mount, pfs->pfs_pid, &pown,
1650 					      ESRCH)) != 0)
1651 			return error;
1652 
1653 		fp = fd_getfile2(pown, pfs->pfs_fd);
1654 		if (fp == NULL) {
1655 			procfs_proc_unlock(pown);
1656 			return EBADF;
1657 		}
1658 
1659 		switch (fp->f_type) {
1660 		case DTYPE_VNODE:
1661 			vxp = fp->f_vnode;
1662 			if (vxp->v_type != VDIR) {
1663 				error = EINVAL;
1664 				break;
1665 			}
1666 			if ((path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK))
1667 			    == NULL) {
1668 				error = ENOMEM;
1669 				break;
1670 			}
1671 			bp = path + MAXPATHLEN;
1672 			*--bp = '\0';
1673 
1674 			/*
1675 			 * XXX: kludge to avoid locking against ourselves
1676 			 * in getcwd()
1677 			 */
1678 			if (vxp->v_tag == VT_PROCFS) {
1679 				*--bp = '/';
1680 			} else {
1681 				rw_enter(&curproc->p_cwdi->cwdi_lock,
1682 				    RW_READER);
1683 				vp = curproc->p_cwdi->cwdi_rdir;
1684 				if (vp == NULL)
1685 					vp = rootvnode;
1686 				error = getcwd_common(vxp, vp, &bp, path,
1687 				    MAXPATHLEN / 2, 0, curlwp);
1688 				rw_exit(&curproc->p_cwdi->cwdi_lock);
1689 			}
1690 			if (error)
1691 				break;
1692 			len = strlen(bp);
1693 			break;
1694 
1695 		case DTYPE_MISC:
1696 			len = snprintf(bf, sizeof(bf), "%s", "[misc]");
1697 			break;
1698 
1699 		case DTYPE_KQUEUE:
1700 			len = snprintf(bf, sizeof(bf), "%s", "[kqueue]");
1701 			break;
1702 
1703 		case DTYPE_SEM:
1704 			len = snprintf(bf, sizeof(bf), "%s", "[ksem]");
1705 			break;
1706 
1707 		default:
1708 			error = EINVAL;
1709 			break;
1710 		}
1711 		closef(fp);
1712 	}
1713 
1714 	if (error == 0)
1715 		error = uiomove(bp, len, ap->a_uio);
1716 	if (pown)
1717 		procfs_proc_unlock(pown);
1718 	if (path)
1719 		free(path, M_TEMP);
1720 	return error;
1721 }
1722 
1723 int
1724 procfs_getpages(void *v)
1725 {
1726 	struct vop_getpages_args /* {
1727 		struct vnode *a_vp;
1728 		voff_t a_offset;
1729 		struct vm_page **a_m;
1730 		int *a_count;
1731 		int a_centeridx;
1732 		vm_prot_t a_access_type;
1733 		int a_advice;
1734 		int a_flags;
1735 	} */ *ap = v;
1736 
1737 	if ((ap->a_flags & PGO_LOCKED) == 0)
1738 		rw_exit(ap->a_vp->v_uobj.vmobjlock);
1739 
1740 	return (EFAULT);
1741 }
1742 
1743 /*
1744  * convert decimal ascii to int
1745  */
1746 static int
1747 atoi(const char *b, size_t len)
1748 {
1749 	int p = 0;
1750 
1751 	while (len--) {
1752 		char c = *b++;
1753 		if (c < '0' || c > '9')
1754 			return -1;
1755 		p = 10 * p + (c - '0');
1756 	}
1757 
1758 	return p;
1759 }
1760