xref: /netbsd-src/sys/miscfs/procfs/procfs_vnops.c (revision b8c616269f5ebf18ab2e35cb8099d683130a177c)
1 /*	$NetBSD: procfs_vnops.c,v 1.93 2003/01/04 15:42:35 martin Exp $	*/
2 
3 /*
4  * Copyright (c) 1993 Jan-Simon Pendry
5  * Copyright (c) 1993, 1995
6  *	The Regents of the University of California.  All rights reserved.
7  *
8  * This code is derived from software contributed to Berkeley by
9  * Jan-Simon Pendry.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  * 3. All advertising materials mentioning features or use of this software
20  *    must display the following acknowledgement:
21  *	This product includes software developed by the University of
22  *	California, Berkeley and its contributors.
23  * 4. Neither the name of the University nor the names of its contributors
24  *    may be used to endorse or promote products derived from this software
25  *    without specific prior written permission.
26  *
27  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
28  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
29  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
30  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
31  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
32  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
33  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
34  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
35  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
36  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
37  * SUCH DAMAGE.
38  *
39  *	@(#)procfs_vnops.c	8.18 (Berkeley) 5/21/95
40  */
41 
42 /*
43  * procfs vnode interface
44  */
45 
46 #include <sys/cdefs.h>
47 __KERNEL_RCSID(0, "$NetBSD: procfs_vnops.c,v 1.93 2003/01/04 15:42:35 martin Exp $");
48 
49 #include <sys/param.h>
50 #include <sys/systm.h>
51 #include <sys/time.h>
52 #include <sys/kernel.h>
53 #include <sys/file.h>
54 #include <sys/filedesc.h>
55 #include <sys/proc.h>
56 #include <sys/vnode.h>
57 #include <sys/namei.h>
58 #include <sys/malloc.h>
59 #include <sys/mount.h>
60 #include <sys/dirent.h>
61 #include <sys/resourcevar.h>
62 #include <sys/stat.h>
63 #include <sys/ptrace.h>
64 
65 #include <uvm/uvm_extern.h>	/* for PAGE_SIZE */
66 
67 #include <machine/reg.h>
68 
69 #include <miscfs/genfs/genfs.h>
70 #include <miscfs/procfs/procfs.h>
71 
72 /*
73  * Vnode Operations.
74  *
75  */
76 
77 static int procfs_validfile_linux __P((struct proc *, struct mount *));
78 
79 /*
80  * This is a list of the valid names in the
81  * process-specific sub-directories.  It is
82  * used in procfs_lookup and procfs_readdir
83  */
84 const struct proc_target {
85 	u_char	pt_type;
86 	u_char	pt_namlen;
87 	char	*pt_name;
88 	pfstype	pt_pfstype;
89 	int	(*pt_valid) __P((struct proc *, struct mount *));
90 } proc_targets[] = {
91 #define N(s) sizeof(s)-1, s
92 	/*	  name		type		validp */
93 	{ DT_DIR, N("."),	Pproc,		NULL },
94 	{ DT_DIR, N(".."),	Proot,		NULL },
95 	{ DT_DIR, N("fd"),	Pfd,		NULL },
96 	{ DT_REG, N("file"),	Pfile,		procfs_validfile },
97 	{ DT_REG, N("mem"),	Pmem,		NULL },
98 	{ DT_REG, N("regs"),	Pregs,		procfs_validregs },
99 	{ DT_REG, N("fpregs"),	Pfpregs,	procfs_validfpregs },
100 	{ DT_REG, N("ctl"),	Pctl,		NULL },
101 	{ DT_REG, N("status"),	Pstatus,	NULL },
102 	{ DT_REG, N("note"),	Pnote,		NULL },
103 	{ DT_REG, N("notepg"),	Pnotepg,	NULL },
104 	{ DT_REG, N("map"),	Pmap,		procfs_validmap },
105 	{ DT_REG, N("maps"),	Pmaps,		procfs_validmap },
106 	{ DT_REG, N("cmdline"), Pcmdline,	NULL },
107 	{ DT_REG, N("exe"),	Pfile,		procfs_validfile_linux },
108 #ifdef __HAVE_PROCFS_MACHDEP
109 	PROCFS_MACHDEP_NODETYPE_DEFNS
110 #endif
111 #undef N
112 };
113 static int nproc_targets = sizeof(proc_targets) / sizeof(proc_targets[0]);
114 
115 /*
116  * List of files in the root directory. Note: the validate function will
117  * be called with p == NULL for these ones.
118  */
119 struct proc_target proc_root_targets[] = {
120 #define N(s) sizeof(s)-1, s
121 	/*	  name		    type	    validp */
122 	{ DT_REG, N("meminfo"),     Pmeminfo,        procfs_validfile_linux },
123 	{ DT_REG, N("cpuinfo"),     Pcpuinfo,        procfs_validfile_linux },
124 #undef N
125 };
126 static int nproc_root_targets =
127     sizeof(proc_root_targets) / sizeof(proc_root_targets[0]);
128 
129 int	procfs_lookup	__P((void *));
130 #define	procfs_create	genfs_eopnotsupp_rele
131 #define	procfs_mknod	genfs_eopnotsupp_rele
132 int	procfs_open	__P((void *));
133 int	procfs_close	__P((void *));
134 int	procfs_access	__P((void *));
135 int	procfs_getattr	__P((void *));
136 int	procfs_setattr	__P((void *));
137 #define	procfs_read	procfs_rw
138 #define	procfs_write	procfs_rw
139 #define	procfs_fcntl	genfs_fcntl
140 #define	procfs_ioctl	genfs_enoioctl
141 #define	procfs_poll	genfs_poll
142 #define procfs_revoke	genfs_revoke
143 #define	procfs_fsync	genfs_nullop
144 #define	procfs_seek	genfs_nullop
145 #define	procfs_remove	genfs_eopnotsupp_rele
146 int	procfs_link	__P((void *));
147 #define	procfs_rename	genfs_eopnotsupp_rele
148 #define	procfs_mkdir	genfs_eopnotsupp_rele
149 #define	procfs_rmdir	genfs_eopnotsupp_rele
150 int	procfs_symlink	__P((void *));
151 int	procfs_readdir	__P((void *));
152 int	procfs_readlink	__P((void *));
153 #define	procfs_abortop	genfs_abortop
154 int	procfs_inactive	__P((void *));
155 int	procfs_reclaim	__P((void *));
156 #define	procfs_lock	genfs_lock
157 #define	procfs_unlock	genfs_unlock
158 #define	procfs_bmap	genfs_badop
159 #define	procfs_strategy	genfs_badop
160 int	procfs_print	__P((void *));
161 int	procfs_pathconf	__P((void *));
162 #define	procfs_islocked	genfs_islocked
163 #define	procfs_advlock	genfs_einval
164 #define	procfs_blkatoff	genfs_eopnotsupp
165 #define	procfs_valloc	genfs_eopnotsupp
166 #define	procfs_vfree	genfs_nullop
167 #define	procfs_truncate	genfs_eopnotsupp
168 #define	procfs_update	genfs_nullop
169 #define	procfs_bwrite	genfs_eopnotsupp
170 #define procfs_putpages	genfs_null_putpages
171 
172 static int atoi __P((const char *, size_t));
173 
174 /*
175  * procfs vnode operations.
176  */
177 int (**procfs_vnodeop_p) __P((void *));
178 const struct vnodeopv_entry_desc procfs_vnodeop_entries[] = {
179 	{ &vop_default_desc, vn_default_error },
180 	{ &vop_lookup_desc, procfs_lookup },		/* lookup */
181 	{ &vop_create_desc, procfs_create },		/* create */
182 	{ &vop_mknod_desc, procfs_mknod },		/* mknod */
183 	{ &vop_open_desc, procfs_open },		/* open */
184 	{ &vop_close_desc, procfs_close },		/* close */
185 	{ &vop_access_desc, procfs_access },		/* access */
186 	{ &vop_getattr_desc, procfs_getattr },		/* getattr */
187 	{ &vop_setattr_desc, procfs_setattr },		/* setattr */
188 	{ &vop_read_desc, procfs_read },		/* read */
189 	{ &vop_write_desc, procfs_write },		/* write */
190 	{ &vop_fcntl_desc, procfs_fcntl },		/* fcntl */
191 	{ &vop_ioctl_desc, procfs_ioctl },		/* ioctl */
192 	{ &vop_poll_desc, procfs_poll },		/* poll */
193 	{ &vop_revoke_desc, procfs_revoke },		/* revoke */
194 	{ &vop_fsync_desc, procfs_fsync },		/* fsync */
195 	{ &vop_seek_desc, procfs_seek },		/* seek */
196 	{ &vop_remove_desc, procfs_remove },		/* remove */
197 	{ &vop_link_desc, procfs_link },		/* link */
198 	{ &vop_rename_desc, procfs_rename },		/* rename */
199 	{ &vop_mkdir_desc, procfs_mkdir },		/* mkdir */
200 	{ &vop_rmdir_desc, procfs_rmdir },		/* rmdir */
201 	{ &vop_symlink_desc, procfs_symlink },		/* symlink */
202 	{ &vop_readdir_desc, procfs_readdir },		/* readdir */
203 	{ &vop_readlink_desc, procfs_readlink },	/* readlink */
204 	{ &vop_abortop_desc, procfs_abortop },		/* abortop */
205 	{ &vop_inactive_desc, procfs_inactive },	/* inactive */
206 	{ &vop_reclaim_desc, procfs_reclaim },		/* reclaim */
207 	{ &vop_lock_desc, procfs_lock },		/* lock */
208 	{ &vop_unlock_desc, procfs_unlock },		/* unlock */
209 	{ &vop_bmap_desc, procfs_bmap },		/* bmap */
210 	{ &vop_strategy_desc, procfs_strategy },	/* strategy */
211 	{ &vop_print_desc, procfs_print },		/* print */
212 	{ &vop_islocked_desc, procfs_islocked },	/* islocked */
213 	{ &vop_pathconf_desc, procfs_pathconf },	/* pathconf */
214 	{ &vop_advlock_desc, procfs_advlock },		/* advlock */
215 	{ &vop_blkatoff_desc, procfs_blkatoff },	/* blkatoff */
216 	{ &vop_valloc_desc, procfs_valloc },		/* valloc */
217 	{ &vop_vfree_desc, procfs_vfree },		/* vfree */
218 	{ &vop_truncate_desc, procfs_truncate },	/* truncate */
219 	{ &vop_update_desc, procfs_update },		/* update */
220 	{ &vop_putpages_desc, procfs_putpages },	/* putpages */
221 	{ NULL, NULL }
222 };
223 const struct vnodeopv_desc procfs_vnodeop_opv_desc =
224 	{ &procfs_vnodeop_p, procfs_vnodeop_entries };
225 /*
226  * set things up for doing i/o on
227  * the pfsnode (vp).  (vp) is locked
228  * on entry, and should be left locked
229  * on exit.
230  *
231  * for procfs we don't need to do anything
232  * in particular for i/o.  all that is done
233  * is to support exclusive open on process
234  * memory images.
235  */
236 int
237 procfs_open(v)
238 	void *v;
239 {
240 	struct vop_open_args /* {
241 		struct vnode *a_vp;
242 		int  a_mode;
243 		struct ucred *a_cred;
244 		struct proc *a_p;
245 	} */ *ap = v;
246 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
247 	struct proc *p1, *p2;
248 	int error;
249 
250 	p1 = ap->a_p;				/* tracer */
251 	p2 = PFIND(pfs->pfs_pid);		/* traced */
252 
253 	if (p2 == NULL)
254 		return (ENOENT);		/* was ESRCH, jsp */
255 
256 	switch (pfs->pfs_type) {
257 	case Pmem:
258 		if (((pfs->pfs_flags & FWRITE) && (ap->a_mode & O_EXCL)) ||
259 		    ((pfs->pfs_flags & O_EXCL) && (ap->a_mode & FWRITE)))
260 			return (EBUSY);
261 
262 		if ((error = process_checkioperm(p1, p2)) != 0)
263 			return (error);
264 
265 		if (ap->a_mode & FWRITE)
266 			pfs->pfs_flags = ap->a_mode & (FWRITE|O_EXCL);
267 
268 		return (0);
269 
270 	default:
271 		break;
272 	}
273 
274 	return (0);
275 }
276 
277 /*
278  * close the pfsnode (vp) after doing i/o.
279  * (vp) is not locked on entry or exit.
280  *
281  * nothing to do for procfs other than undo
282  * any exclusive open flag (see _open above).
283  */
284 int
285 procfs_close(v)
286 	void *v;
287 {
288 	struct vop_close_args /* {
289 		struct vnode *a_vp;
290 		int  a_fflag;
291 		struct ucred *a_cred;
292 		struct proc *a_p;
293 	} */ *ap = v;
294 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
295 
296 	switch (pfs->pfs_type) {
297 	case Pmem:
298 		if ((ap->a_fflag & FWRITE) && (pfs->pfs_flags & O_EXCL))
299 			pfs->pfs_flags &= ~(FWRITE|O_EXCL);
300 		break;
301 
302 	default:
303 		break;
304 	}
305 
306 	return (0);
307 }
308 
309 /*
310  * _inactive is called when the pfsnode
311  * is vrele'd and the reference count goes
312  * to zero.  (vp) will be on the vnode free
313  * list, so to get it back vget() must be
314  * used.
315  *
316  * for procfs, check if the process is still
317  * alive and if it isn't then just throw away
318  * the vnode by calling vgone().  this may
319  * be overkill and a waste of time since the
320  * chances are that the process will still be
321  * there and PFIND is not free.
322  *
323  * (vp) is locked on entry, but must be unlocked on exit.
324  */
325 int
326 procfs_inactive(v)
327 	void *v;
328 {
329 	struct vop_inactive_args /* {
330 		struct vnode *a_vp;
331 		struct proc *a_p;
332 	} */ *ap = v;
333 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
334 
335 	VOP_UNLOCK(ap->a_vp, 0);
336 	if (PFIND(pfs->pfs_pid) == NULL)
337 		vgone(ap->a_vp);
338 
339 	return (0);
340 }
341 
342 /*
343  * _reclaim is called when getnewvnode()
344  * wants to make use of an entry on the vnode
345  * free list.  at this time the filesystem needs
346  * to free any private data and remove the node
347  * from any private lists.
348  */
349 int
350 procfs_reclaim(v)
351 	void *v;
352 {
353 	struct vop_reclaim_args /* {
354 		struct vnode *a_vp;
355 	} */ *ap = v;
356 
357 	return (procfs_freevp(ap->a_vp));
358 }
359 
360 /*
361  * Return POSIX pathconf information applicable to special devices.
362  */
363 int
364 procfs_pathconf(v)
365 	void *v;
366 {
367 	struct vop_pathconf_args /* {
368 		struct vnode *a_vp;
369 		int a_name;
370 		register_t *a_retval;
371 	} */ *ap = v;
372 
373 	switch (ap->a_name) {
374 	case _PC_LINK_MAX:
375 		*ap->a_retval = LINK_MAX;
376 		return (0);
377 	case _PC_MAX_CANON:
378 		*ap->a_retval = MAX_CANON;
379 		return (0);
380 	case _PC_MAX_INPUT:
381 		*ap->a_retval = MAX_INPUT;
382 		return (0);
383 	case _PC_PIPE_BUF:
384 		*ap->a_retval = PIPE_BUF;
385 		return (0);
386 	case _PC_CHOWN_RESTRICTED:
387 		*ap->a_retval = 1;
388 		return (0);
389 	case _PC_VDISABLE:
390 		*ap->a_retval = _POSIX_VDISABLE;
391 		return (0);
392 	case _PC_SYNC_IO:
393 		*ap->a_retval = 1;
394 		return (0);
395 	default:
396 		return (EINVAL);
397 	}
398 	/* NOTREACHED */
399 }
400 
401 /*
402  * _print is used for debugging.
403  * just print a readable description
404  * of (vp).
405  */
406 int
407 procfs_print(v)
408 	void *v;
409 {
410 	struct vop_print_args /* {
411 		struct vnode *a_vp;
412 	} */ *ap = v;
413 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
414 
415 	printf("tag VT_PROCFS, type %d, pid %d, mode %x, flags %lx\n",
416 	    pfs->pfs_type, pfs->pfs_pid, pfs->pfs_mode, pfs->pfs_flags);
417 	return 0;
418 }
419 
420 int
421 procfs_link(v)
422 	void *v;
423 {
424 	struct vop_link_args /* {
425 		struct vnode *a_dvp;
426 		struct vnode *a_vp;
427 		struct componentname *a_cnp;
428 	} */ *ap = v;
429 
430 	VOP_ABORTOP(ap->a_dvp, ap->a_cnp);
431 	vput(ap->a_dvp);
432 	return (EROFS);
433 }
434 
435 int
436 procfs_symlink(v)
437 	void *v;
438 {
439 	struct vop_symlink_args /* {
440 		struct vnode *a_dvp;
441 		struct vnode **a_vpp;
442 		struct componentname *a_cnp;
443 		struct vattr *a_vap;
444 		char *a_target;
445 	} */ *ap = v;
446 
447 	VOP_ABORTOP(ap->a_dvp, ap->a_cnp);
448 	vput(ap->a_dvp);
449 	return (EROFS);
450 }
451 
452 /*
453  * Invent attributes for pfsnode (vp) and store
454  * them in (vap).
455  * Directories lengths are returned as zero since
456  * any real length would require the genuine size
457  * to be computed, and nothing cares anyway.
458  *
459  * this is relatively minimal for procfs.
460  */
461 int
462 procfs_getattr(v)
463 	void *v;
464 {
465 	struct vop_getattr_args /* {
466 		struct vnode *a_vp;
467 		struct vattr *a_vap;
468 		struct ucred *a_cred;
469 		struct proc *a_p;
470 	} */ *ap = v;
471 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
472 	struct vattr *vap = ap->a_vap;
473 	struct proc *procp;
474 	int error;
475 
476 	/* first check the process still exists */
477 	switch (pfs->pfs_type) {
478 	case Proot:
479 	case Pcurproc:
480 	case Pself:
481 		procp = 0;
482 		break;
483 
484 	default:
485 		procp = PFIND(pfs->pfs_pid);
486 		if (procp == NULL)
487 			return (ENOENT);
488 		break;
489 	}
490 
491 	error = 0;
492 
493 	/* start by zeroing out the attributes */
494 	VATTR_NULL(vap);
495 
496 	/* next do all the common fields */
497 	vap->va_type = ap->a_vp->v_type;
498 	vap->va_mode = pfs->pfs_mode;
499 	vap->va_fileid = pfs->pfs_fileno;
500 	vap->va_flags = 0;
501 	vap->va_blocksize = PAGE_SIZE;
502 
503 	/*
504 	 * Make all times be current TOD.  Avoid microtime(9), it's slow.
505 	 * We don't guard the read from time(9) with splclock(9) since we
506 	 * don't actually need to be THAT sure the access is atomic.
507 	 *
508 	 * It would be possible to get the process start
509 	 * time from the p_stat structure, but there's
510 	 * no "file creation" time stamp anyway, and the
511 	 * p_stat structure is not addressible if u. gets
512 	 * swapped out for that process.
513 	 */
514 	TIMEVAL_TO_TIMESPEC(&time, &vap->va_ctime);
515 	vap->va_atime = vap->va_mtime = vap->va_ctime;
516 
517 	switch (pfs->pfs_type) {
518 	case Pmem:
519 	case Pregs:
520 	case Pfpregs:
521 #if defined(__HAVE_PROCFS_MACHDEP) && defined(PROCFS_MACHDEP_PROTECT_CASES)
522 	PROCFS_MACHDEP_PROTECT_CASES
523 #endif
524 		/*
525 		 * If the process has exercised some setuid or setgid
526 		 * privilege, then rip away read/write permission so
527 		 * that only root can gain access.
528 		 */
529 		if (procp->p_flag & P_SUGID)
530 			vap->va_mode &= ~(S_IRUSR|S_IWUSR);
531 		/* FALLTHROUGH */
532 	case Pctl:
533 	case Pstatus:
534 	case Pnote:
535 	case Pnotepg:
536 	case Pmap:
537 	case Pmaps:
538 	case Pcmdline:
539 		vap->va_nlink = 1;
540 		vap->va_uid = procp->p_ucred->cr_uid;
541 		vap->va_gid = procp->p_ucred->cr_gid;
542 		break;
543 	case Pmeminfo:
544 	case Pcpuinfo:
545 		vap->va_nlink = 1;
546 		vap->va_uid = vap->va_gid = 0;
547 		break;
548 
549 	default:
550 		break;
551 	}
552 
553 	/*
554 	 * now do the object specific fields
555 	 *
556 	 * The size could be set from struct reg, but it's hardly
557 	 * worth the trouble, and it puts some (potentially) machine
558 	 * dependent data into this machine-independent code.  If it
559 	 * becomes important then this function should break out into
560 	 * a per-file stat function in the corresponding .c file.
561 	 */
562 
563 	switch (pfs->pfs_type) {
564 	case Proot:
565 		/*
566 		 * Set nlink to 1 to tell fts(3) we don't actually know.
567 		 */
568 		vap->va_nlink = 1;
569 		vap->va_uid = 0;
570 		vap->va_gid = 0;
571 		vap->va_bytes = vap->va_size = DEV_BSIZE;
572 		break;
573 
574 	case Pcurproc: {
575 		char buf[16];		/* should be enough */
576 		vap->va_nlink = 1;
577 		vap->va_uid = 0;
578 		vap->va_gid = 0;
579 		vap->va_bytes = vap->va_size =
580 		    sprintf(buf, "%ld", (long)curproc->p_pid);
581 		break;
582 	}
583 
584 	case Pself:
585 		vap->va_nlink = 1;
586 		vap->va_uid = 0;
587 		vap->va_gid = 0;
588 		vap->va_bytes = vap->va_size = sizeof("curproc");
589 		break;
590 
591 	case Pfd:
592 		if (pfs->pfs_fd != -1) {
593 			struct file *fp;
594 			if ((error = procfs_getfp(pfs, &fp)) != 0)
595 				return error;
596 			vap->va_nlink = 1;
597 			vap->va_uid = fp->f_cred->cr_uid;
598 			vap->va_gid = fp->f_cred->cr_gid;
599 			switch (fp->f_type) {
600 			case DTYPE_VNODE:
601 				vap->va_bytes = vap->va_size =
602 				    ((struct vnode *)fp->f_data)->v_size;
603 				break;
604 			default:
605 				vap->va_bytes = vap->va_size = 0;
606 				break;
607 			}
608 			break;
609 		}
610 		/*FALLTHROUGH*/
611 	case Pproc:
612 		vap->va_nlink = 2;
613 		vap->va_uid = procp->p_ucred->cr_uid;
614 		vap->va_gid = procp->p_ucred->cr_gid;
615 		vap->va_bytes = vap->va_size = DEV_BSIZE;
616 		break;
617 
618 	case Pfile:
619 		error = EOPNOTSUPP;
620 		break;
621 
622 	case Pmem:
623 		vap->va_bytes = vap->va_size =
624 			ctob(procp->p_vmspace->vm_tsize +
625 				    procp->p_vmspace->vm_dsize +
626 				    procp->p_vmspace->vm_ssize);
627 		break;
628 
629 #if defined(PT_GETREGS) || defined(PT_SETREGS)
630 	case Pregs:
631 		vap->va_bytes = vap->va_size = sizeof(struct reg);
632 		break;
633 #endif
634 
635 #if defined(PT_GETFPREGS) || defined(PT_SETFPREGS)
636 	case Pfpregs:
637 		vap->va_bytes = vap->va_size = sizeof(struct fpreg);
638 		break;
639 #endif
640 
641 	case Pctl:
642 	case Pstatus:
643 	case Pnote:
644 	case Pnotepg:
645 	case Pcmdline:
646 	case Pmeminfo:
647 	case Pcpuinfo:
648 		vap->va_bytes = vap->va_size = 0;
649 		break;
650 	case Pmap:
651 	case Pmaps:
652 		/*
653 		 * Advise a larger blocksize for the map files, so that
654 		 * they may be read in one pass.
655 		 */
656 		vap->va_blocksize = 4 * PAGE_SIZE;
657 		vap->va_bytes = vap->va_size = 0;
658 		break;
659 
660 #ifdef __HAVE_PROCFS_MACHDEP
661 	PROCFS_MACHDEP_NODETYPE_CASES
662 		error = procfs_machdep_getattr(ap->a_vp, vap, procp);
663 		break;
664 #endif
665 
666 	default:
667 		panic("procfs_getattr");
668 	}
669 
670 	return (error);
671 }
672 
673 /*ARGSUSED*/
674 int
675 procfs_setattr(v)
676 	void *v;
677 {
678 	/*
679 	 * just fake out attribute setting
680 	 * it's not good to generate an error
681 	 * return, otherwise things like creat()
682 	 * will fail when they try to set the
683 	 * file length to 0.  worse, this means
684 	 * that echo $note > /proc/$pid/note will fail.
685 	 */
686 
687 	return (0);
688 }
689 
690 /*
691  * implement access checking.
692  *
693  * actually, the check for super-user is slightly
694  * broken since it will allow read access to write-only
695  * objects.  this doesn't cause any particular trouble
696  * but does mean that the i/o entry points need to check
697  * that the operation really does make sense.
698  */
699 int
700 procfs_access(v)
701 	void *v;
702 {
703 	struct vop_access_args /* {
704 		struct vnode *a_vp;
705 		int a_mode;
706 		struct ucred *a_cred;
707 		struct proc *a_p;
708 	} */ *ap = v;
709 	struct vattr va;
710 	int error;
711 
712 	if ((error = VOP_GETATTR(ap->a_vp, &va, ap->a_cred, ap->a_p)) != 0)
713 		return (error);
714 
715 	return (vaccess(va.va_type, va.va_mode,
716 	    va.va_uid, va.va_gid, ap->a_mode, ap->a_cred));
717 }
718 
719 /*
720  * lookup.  this is incredibly complicated in the
721  * general case, however for most pseudo-filesystems
722  * very little needs to be done.
723  *
724  * Locking isn't hard here, just poorly documented.
725  *
726  * If we're looking up ".", just vref the parent & return it.
727  *
728  * If we're looking up "..", unlock the parent, and lock "..". If everything
729  * went ok, and we're on the last component and the caller requested the
730  * parent locked, try to re-lock the parent. We do this to prevent lock
731  * races.
732  *
733  * For anything else, get the needed node. Then unlock the parent if not
734  * the last component or not LOCKPARENT (i.e. if we wouldn't re-lock the
735  * parent in the .. case).
736  *
737  * We try to exit with the parent locked in error cases.
738  */
739 int
740 procfs_lookup(v)
741 	void *v;
742 {
743 	struct vop_lookup_args /* {
744 		struct vnode * a_dvp;
745 		struct vnode ** a_vpp;
746 		struct componentname * a_cnp;
747 	} */ *ap = v;
748 	struct componentname *cnp = ap->a_cnp;
749 	struct vnode **vpp = ap->a_vpp;
750 	struct vnode *dvp = ap->a_dvp;
751 	const char *pname = cnp->cn_nameptr;
752 	const struct proc_target *pt = NULL;
753 	struct vnode *fvp;
754 	pid_t pid;
755 	struct pfsnode *pfs;
756 	struct proc *p = NULL;
757 	int i, error, wantpunlock, iscurproc = 0, isself = 0;
758 
759 	*vpp = NULL;
760 	cnp->cn_flags &= ~PDIRUNLOCK;
761 
762 	if (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)
763 		return (EROFS);
764 
765 	if (cnp->cn_namelen == 1 && *pname == '.') {
766 		*vpp = dvp;
767 		VREF(dvp);
768 		return (0);
769 	}
770 
771 	wantpunlock = (~cnp->cn_flags & (LOCKPARENT | ISLASTCN));
772 	pfs = VTOPFS(dvp);
773 	switch (pfs->pfs_type) {
774 	case Proot:
775 		/*
776 		 * Shouldn't get here with .. in the root node.
777 		 */
778 		if (cnp->cn_flags & ISDOTDOT)
779 			return (EIO);
780 
781 		iscurproc = CNEQ(cnp, "curproc", 7);
782 		isself = CNEQ(cnp, "self", 4);
783 
784 		if (iscurproc || isself) {
785 			error = procfs_allocvp(dvp->v_mount, vpp, 0,
786 			    iscurproc ? Pcurproc : Pself, -1);
787 			if ((error == 0) && (wantpunlock)) {
788 				VOP_UNLOCK(dvp, 0);
789 				cnp->cn_flags |= PDIRUNLOCK;
790 			}
791 			return (error);
792 		}
793 
794 		for (i = 0; i < nproc_root_targets; i++) {
795 			pt = &proc_root_targets[i];
796 			if (cnp->cn_namelen == pt->pt_namlen &&
797 			    memcmp(pt->pt_name, pname, cnp->cn_namelen) == 0 &&
798 			    (pt->pt_valid == NULL ||
799 			     (*pt->pt_valid)(p, dvp->v_mount)))
800 				break;
801 		}
802 
803 		if (i != nproc_root_targets) {
804 			error = procfs_allocvp(dvp->v_mount, vpp, 0,
805 			    pt->pt_pfstype, -1);
806 			if ((error == 0) && (wantpunlock)) {
807 				VOP_UNLOCK(dvp, 0);
808 				cnp->cn_flags |= PDIRUNLOCK;
809 			}
810 			return (error);
811 		}
812 
813 		pid = (pid_t)atoi(pname, cnp->cn_namelen);
814 		if (pid > PID_MAX)
815 			break;
816 
817 		p = PFIND(pid);
818 		if (p == NULL)
819 			break;
820 
821 		error = procfs_allocvp(dvp->v_mount, vpp, pid, Pproc, -1);
822 		if ((error == 0) && (wantpunlock)) {
823 			VOP_UNLOCK(dvp, 0);
824 			cnp->cn_flags |= PDIRUNLOCK;
825 		}
826 		return (error);
827 
828 	case Pproc:
829 		/*
830 		 * do the .. dance. We unlock the directory, and then
831 		 * get the root dir. That will automatically return ..
832 		 * locked. Then if the caller wanted dvp locked, we
833 		 * re-lock.
834 		 */
835 		if (cnp->cn_flags & ISDOTDOT) {
836 			VOP_UNLOCK(dvp, 0);
837 			cnp->cn_flags |= PDIRUNLOCK;
838 			error = procfs_root(dvp->v_mount, vpp);
839 			if ((error == 0) && (wantpunlock == 0) &&
840 				    ((error = vn_lock(dvp, LK_EXCLUSIVE)) == 0))
841 				cnp->cn_flags &= ~PDIRUNLOCK;
842 			return (error);
843 		}
844 
845 		p = PFIND(pfs->pfs_pid);
846 		if (p == NULL)
847 			break;
848 
849 		for (pt = proc_targets, i = 0; i < nproc_targets; pt++, i++) {
850 			if (cnp->cn_namelen == pt->pt_namlen &&
851 			    memcmp(pt->pt_name, pname, cnp->cn_namelen) == 0 &&
852 			    (pt->pt_valid == NULL ||
853 			     (*pt->pt_valid)(p, dvp->v_mount)))
854 				goto found;
855 		}
856 		break;
857 
858 	found:
859 		if (pt->pt_pfstype == Pfile) {
860 			fvp = p->p_textvp;
861 			/* We already checked that it exists. */
862 			VREF(fvp);
863 			vn_lock(fvp, LK_EXCLUSIVE | LK_RETRY);
864 			if (wantpunlock) {
865 				VOP_UNLOCK(dvp, 0);
866 				cnp->cn_flags |= PDIRUNLOCK;
867 			}
868 			*vpp = fvp;
869 			return (0);
870 		}
871 
872 		error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
873 		    pt->pt_pfstype, -1);
874 		if ((error == 0) && (wantpunlock)) {
875 			VOP_UNLOCK(dvp, 0);
876 			cnp->cn_flags |= PDIRUNLOCK;
877 		}
878 		return (error);
879 
880 	case Pfd: {
881 		int fd;
882 		struct file *fp;
883 		/*
884 		 * do the .. dance. We unlock the directory, and then
885 		 * get the proc dir. That will automatically return ..
886 		 * locked. Then if the caller wanted dvp locked, we
887 		 * re-lock.
888 		 */
889 		if (cnp->cn_flags & ISDOTDOT) {
890 			VOP_UNLOCK(dvp, 0);
891 			cnp->cn_flags |= PDIRUNLOCK;
892 			error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
893 			    Pproc, -1);
894 			if ((error == 0) && (wantpunlock == 0) &&
895 				    ((error = vn_lock(dvp, LK_EXCLUSIVE)) == 0))
896 				cnp->cn_flags &= ~PDIRUNLOCK;
897 			return (error);
898 		}
899 		fd = atoi(pname, cnp->cn_namelen);
900 		if (fd == -1)
901 			return EINVAL;
902 		p = PFIND(pfs->pfs_pid);
903 		if (p == NULL)
904 			return ESRCH;
905 		if ((fp = p->p_fd->fd_ofiles[fd]) == NULL)
906 			return ENOENT;
907 		switch (fp->f_type) {
908 		case DTYPE_VNODE:
909 			fvp = (struct vnode *)fp->f_data;
910 			VREF(fvp);
911 			vn_lock(fvp, LK_EXCLUSIVE | LK_RETRY |
912 			    (p == curproc ? LK_CANRECURSE : 0));
913 			*vpp = fvp;
914 			error = 0;
915 			break;
916 		default:
917 			error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
918 			    Pfd, fd);
919 			break;
920 		}
921 		if ((error == 0) && (wantpunlock)) {
922 			VOP_UNLOCK(dvp, 0);
923 			cnp->cn_flags |= PDIRUNLOCK;
924 		}
925 		return error;
926 	}
927 	default:
928 		return (ENOTDIR);
929 	}
930 
931 	return (cnp->cn_nameiop == LOOKUP ? ENOENT : EROFS);
932 }
933 
934 int
935 procfs_validfile(p, mp)
936 	struct proc *p;
937 	struct mount *mp;
938 {
939 	return (p->p_textvp != NULL);
940 }
941 
942 static int
943 procfs_validfile_linux(p, mp)
944 	struct proc *p;
945 	struct mount *mp;
946 {
947 	int flags;
948 
949 	flags = VFSTOPROC(mp)->pmnt_flags;
950 	return ((flags & PROCFSMNT_LINUXCOMPAT) &&
951 	    (p == NULL || procfs_validfile(p, mp)));
952 }
953 
954 /*
955  * readdir returns directory entries from pfsnode (vp).
956  *
957  * the strategy here with procfs is to generate a single
958  * directory entry at a time (struct dirent) and then
959  * copy that out to userland using uiomove.  a more efficent
960  * though more complex implementation, would try to minimize
961  * the number of calls to uiomove().  for procfs, this is
962  * hardly worth the added code complexity.
963  *
964  * this should just be done through read()
965  */
966 int
967 procfs_readdir(v)
968 	void *v;
969 {
970 	struct vop_readdir_args /* {
971 		struct vnode *a_vp;
972 		struct uio *a_uio;
973 		struct ucred *a_cred;
974 		int *a_eofflag;
975 		off_t **a_cookies;
976 		int *a_ncookies;
977 	} */ *ap = v;
978 	struct uio *uio = ap->a_uio;
979 	struct dirent d;
980 	struct pfsnode *pfs;
981 	off_t i;
982 	int error;
983 	off_t *cookies = NULL;
984 	int ncookies, left, skip, j;
985 	struct vnode *vp;
986 	const struct proc_target *pt;
987 
988 	vp = ap->a_vp;
989 	pfs = VTOPFS(vp);
990 
991 	if (uio->uio_resid < UIO_MX)
992 		return (EINVAL);
993 	if (uio->uio_offset < 0)
994 		return (EINVAL);
995 
996 	error = 0;
997 	i = uio->uio_offset;
998 	memset((caddr_t)&d, 0, UIO_MX);
999 	d.d_reclen = UIO_MX;
1000 	ncookies = uio->uio_resid / UIO_MX;
1001 
1002 	switch (pfs->pfs_type) {
1003 	/*
1004 	 * this is for the process-specific sub-directories.
1005 	 * all that is needed to is copy out all the entries
1006 	 * from the procent[] table (top of this file).
1007 	 */
1008 	case Pproc: {
1009 		struct proc *p;
1010 
1011 		if (i >= nproc_targets)
1012 			return 0;
1013 
1014 		p = PFIND(pfs->pfs_pid);
1015 		if (p == NULL)
1016 			break;
1017 
1018 		if (ap->a_ncookies) {
1019 			ncookies = min(ncookies, (nproc_targets - i));
1020 			cookies = malloc(ncookies * sizeof (off_t),
1021 			    M_TEMP, M_WAITOK);
1022 			*ap->a_cookies = cookies;
1023 		}
1024 
1025 		for (pt = &proc_targets[i];
1026 		     uio->uio_resid >= UIO_MX && i < nproc_targets; pt++, i++) {
1027 			if (pt->pt_valid &&
1028 			    (*pt->pt_valid)(p, vp->v_mount) == 0)
1029 				continue;
1030 
1031 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid,
1032 			    pt->pt_pfstype, -1);
1033 			d.d_namlen = pt->pt_namlen;
1034 			memcpy(d.d_name, pt->pt_name, pt->pt_namlen + 1);
1035 			d.d_type = pt->pt_type;
1036 
1037 			if ((error = uiomove((caddr_t)&d, UIO_MX, uio)) != 0)
1038 				break;
1039 			if (cookies)
1040 				*cookies++ = i + 1;
1041 		}
1042 
1043 	    	break;
1044 	}
1045 	case Pfd: {
1046 		struct proc *p;
1047 		struct filedesc	*fdp;
1048 		int lim, last, nc = 0;
1049 
1050 		p = PFIND(pfs->pfs_pid);
1051 		if (p == NULL)
1052 			return ESRCH;
1053 
1054 		fdp = p->p_fd;
1055 
1056 		lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfiles);
1057 		last = min(fdp->fd_nfiles, lim);
1058 		if (i >= lim)
1059 			return 0;
1060 
1061 		if (ap->a_ncookies) {
1062 			ncookies = min(ncookies, (fdp->fd_nfiles + 2 - i));
1063 			cookies = malloc(ncookies * sizeof (off_t),
1064 			    M_TEMP, M_WAITOK);
1065 			*ap->a_cookies = cookies;
1066 		}
1067 
1068 		for (; i < 2 && uio->uio_resid >= UIO_MX; i++) {
1069 			pt = &proc_targets[i];
1070 			d.d_fileno = (pfs->pfs_pid << 8) + i;
1071 			d.d_namlen = pt->pt_namlen;
1072 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid,
1073 			    pt->pt_pfstype, -1);
1074 			(void)memcpy(d.d_name, pt->pt_name, pt->pt_namlen + 1);
1075 			d.d_type = pt->pt_type;
1076 			if ((error = uiomove((caddr_t)&d, UIO_MX, uio)) != 0)
1077 				break;
1078 			if (cookies)
1079 				*cookies++ = i + 1;
1080 			nc++;
1081 		}
1082 		if (error) {
1083 			ncookies = nc;
1084 			break;
1085 		}
1086 		for (; uio->uio_resid >= UIO_MX && i < fdp->fd_nfiles; i++) {
1087 			if (fdp->fd_ofiles[i - 2] == NULL)
1088 				continue;
1089 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid, Pfd, i - 2);
1090 			d.d_namlen = snprintf(d.d_name, sizeof(d.d_name),
1091 			    "%lld", (long long)(i - 2));
1092 			d.d_type = VREG;
1093 			if ((error = uiomove((caddr_t)&d, UIO_MX, uio)) != 0)
1094 				break;
1095 			if (cookies)
1096 				*cookies++ = i + 1;
1097 			nc++;
1098 		}
1099 		ncookies = nc;
1100 		break;
1101 	}
1102 
1103 	/*
1104 	 * this is for the root of the procfs filesystem
1105 	 * what is needed are special entries for "curproc"
1106 	 * and "self" followed by an entry for each process
1107 	 * on allproc
1108 #ifdef PROCFS_ZOMBIE
1109 	 * and deadproc and zombproc.
1110 #endif
1111 	 */
1112 
1113 	case Proot: {
1114 		int pcnt = i, nc = 0;
1115 		const struct proclist_desc *pd;
1116 		volatile struct proc *p;
1117 
1118 		if (pcnt > 3)
1119 			pcnt = 3;
1120 		if (ap->a_ncookies) {
1121 			/*
1122 			 * XXX Potentially allocating too much space here,
1123 			 * but I'm lazy. This loop needs some work.
1124 			 */
1125 			cookies = malloc(ncookies * sizeof (off_t),
1126 			    M_TEMP, M_WAITOK);
1127 			*ap->a_cookies = cookies;
1128 		}
1129 		/*
1130 		 * XXX: THIS LOOP ASSUMES THAT allproc IS THE FIRST
1131 		 * PROCLIST IN THE proclists!
1132 		 */
1133 		proclist_lock_read();
1134 		pd = proclists;
1135 #ifdef PROCFS_ZOMBIE
1136 	again:
1137 #endif
1138 		for (p = LIST_FIRST(pd->pd_list);
1139 		     p != NULL && uio->uio_resid >= UIO_MX; i++, pcnt++) {
1140 			switch (i) {
1141 			case 0:		/* `.' */
1142 			case 1:		/* `..' */
1143 				d.d_fileno = PROCFS_FILENO(0, Proot, -1);
1144 				d.d_namlen = i + 1;
1145 				memcpy(d.d_name, "..", d.d_namlen);
1146 				d.d_name[i + 1] = '\0';
1147 				d.d_type = DT_DIR;
1148 				break;
1149 
1150 			case 2:
1151 				d.d_fileno = PROCFS_FILENO(0, Pcurproc, -1);
1152 				d.d_namlen = sizeof("curproc") - 1;
1153 				memcpy(d.d_name, "curproc", sizeof("curproc"));
1154 				d.d_type = DT_LNK;
1155 				break;
1156 
1157 			case 3:
1158 				d.d_fileno = PROCFS_FILENO(0, Pself, -1);
1159 				d.d_namlen = sizeof("self") - 1;
1160 				memcpy(d.d_name, "self", sizeof("self"));
1161 				d.d_type = DT_LNK;
1162 				break;
1163 
1164 			default:
1165 				while (pcnt < i) {
1166 					pcnt++;
1167 					p = LIST_NEXT(p, p_list);
1168 					if (!p)
1169 						goto done;
1170 				}
1171 				d.d_fileno = PROCFS_FILENO(p->p_pid, Pproc, -1);
1172 				d.d_namlen = sprintf(d.d_name, "%ld",
1173 				    (long)p->p_pid);
1174 				d.d_type = DT_DIR;
1175 				p = p->p_list.le_next;
1176 				break;
1177 			}
1178 
1179 			if ((error = uiomove((caddr_t)&d, UIO_MX, uio)) != 0)
1180 				break;
1181 			nc++;
1182 			if (cookies)
1183 				*cookies++ = i + 1;
1184 		}
1185 	done:
1186 
1187 #ifdef PROCFS_ZOMBIE
1188 		pd++;
1189 		if (p == NULL && pd->pd_list != NULL)
1190 			goto again;
1191 #endif
1192 		proclist_unlock_read();
1193 
1194 		skip = i - pcnt;
1195 		if (skip >= nproc_root_targets)
1196 			break;
1197 		left = nproc_root_targets - skip;
1198 		for (j = 0, pt = &proc_root_targets[0];
1199 		     uio->uio_resid >= UIO_MX && j < left;
1200 		     pt++, j++, i++) {
1201 			if (pt->pt_valid &&
1202 			    (*pt->pt_valid)(NULL, vp->v_mount) == 0)
1203 				continue;
1204 			d.d_fileno = PROCFS_FILENO(0, pt->pt_pfstype, -1);
1205 			d.d_namlen = pt->pt_namlen;
1206 			memcpy(d.d_name, pt->pt_name, pt->pt_namlen + 1);
1207 			d.d_type = pt->pt_type;
1208 
1209 			if ((error = uiomove((caddr_t)&d, UIO_MX, uio)) != 0)
1210 				break;
1211 			nc++;
1212 			if (cookies)
1213 				*cookies++ = i + 1;
1214 		}
1215 
1216 		ncookies = nc;
1217 		break;
1218 	}
1219 
1220 	default:
1221 		error = ENOTDIR;
1222 		break;
1223 	}
1224 
1225 	if (ap->a_ncookies) {
1226 		if (error) {
1227 			if (cookies)
1228 				free(*ap->a_cookies, M_TEMP);
1229 			*ap->a_ncookies = 0;
1230 			*ap->a_cookies = NULL;
1231 		} else
1232 			*ap->a_ncookies = ncookies;
1233 	}
1234 	uio->uio_offset = i;
1235 	return (error);
1236 }
1237 
1238 /*
1239  * readlink reads the link of `curproc'
1240  */
1241 int
1242 procfs_readlink(v)
1243 	void *v;
1244 {
1245 	struct vop_readlink_args *ap = v;
1246 	char buf[16];		/* should be enough */
1247 	int len;
1248 
1249 	if (VTOPFS(ap->a_vp)->pfs_fileno == PROCFS_FILENO(0, Pcurproc, -1))
1250 		len = sprintf(buf, "%ld", (long)curproc->p_pid);
1251 	else if (VTOPFS(ap->a_vp)->pfs_fileno == PROCFS_FILENO(0, Pself, -1))
1252 		len = sprintf(buf, "%s", "curproc");
1253 	else
1254 		return (EINVAL);
1255 
1256 	return (uiomove((caddr_t)buf, len, ap->a_uio));
1257 }
1258 
1259 /*
1260  * convert decimal ascii to int
1261  */
1262 static int
1263 atoi(b, len)
1264 	const char *b;
1265 	size_t len;
1266 {
1267 	int p = 0;
1268 
1269 	while (len--) {
1270 		char c = *b++;
1271 		if (c < '0' || c > '9')
1272 			return -1;
1273 		p = 10 * p + (c - '0');
1274 	}
1275 
1276 	return p;
1277 }
1278