xref: /netbsd-src/sys/miscfs/procfs/procfs_vnops.c (revision 28c37e673e4d9b6cbdc7483062b915cc61d1ccf5)
1 /*	$NetBSD: procfs_vnops.c,v 1.90 2002/08/03 04:52:45 simonb 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 1.90 2002/08/03 04:52:45 simonb 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/proc.h>
55 #include <sys/vnode.h>
56 #include <sys/namei.h>
57 #include <sys/malloc.h>
58 #include <sys/mount.h>
59 #include <sys/dirent.h>
60 #include <sys/resourcevar.h>
61 #include <sys/stat.h>
62 #include <sys/ptrace.h>
63 
64 #include <uvm/uvm_extern.h>	/* for PAGE_SIZE */
65 
66 #include <machine/reg.h>
67 
68 #include <miscfs/genfs/genfs.h>
69 #include <miscfs/procfs/procfs.h>
70 
71 /*
72  * Vnode Operations.
73  *
74  */
75 
76 static int procfs_validfile_linux __P((struct proc *, struct mount *));
77 
78 /*
79  * This is a list of the valid names in the
80  * process-specific sub-directories.  It is
81  * used in procfs_lookup and procfs_readdir
82  */
83 const struct proc_target {
84 	u_char	pt_type;
85 	u_char	pt_namlen;
86 	char	*pt_name;
87 	pfstype	pt_pfstype;
88 	int	(*pt_valid) __P((struct proc *, struct mount *));
89 } proc_targets[] = {
90 #define N(s) sizeof(s)-1, s
91 	/*	  name		type		validp */
92 	{ DT_DIR, N("."),	Pproc,		NULL },
93 	{ DT_DIR, N(".."),	Proot,		NULL },
94 	{ DT_REG, N("file"),	Pfile,		procfs_validfile },
95 	{ DT_REG, N("mem"),	Pmem,		NULL },
96 	{ DT_REG, N("regs"),	Pregs,		procfs_validregs },
97 	{ DT_REG, N("fpregs"),	Pfpregs,	procfs_validfpregs },
98 	{ DT_REG, N("ctl"),	Pctl,		NULL },
99 	{ DT_REG, N("status"),	Pstatus,	NULL },
100 	{ DT_REG, N("note"),	Pnote,		NULL },
101 	{ DT_REG, N("notepg"),	Pnotepg,	NULL },
102 	{ DT_REG, N("map"),	Pmap,		procfs_validmap },
103 	{ DT_REG, N("maps"),	Pmaps,		procfs_validmap },
104 	{ DT_REG, N("cmdline"), Pcmdline,	NULL },
105 	{ DT_REG, N("exe"),	Pfile,		procfs_validfile_linux },
106 #ifdef __HAVE_PROCFS_MACHDEP
107 	PROCFS_MACHDEP_NODETYPE_DEFNS
108 #endif
109 #undef N
110 };
111 static int nproc_targets = sizeof(proc_targets) / sizeof(proc_targets[0]);
112 
113 /*
114  * List of files in the root directory. Note: the validate function will
115  * be called with p == NULL for these ones.
116  */
117 struct proc_target proc_root_targets[] = {
118 #define N(s) sizeof(s)-1, s
119 	/*	  name		    type	    validp */
120 	{ DT_REG, N("meminfo"),     Pmeminfo,        procfs_validfile_linux },
121 	{ DT_REG, N("cpuinfo"),     Pcpuinfo,        procfs_validfile_linux },
122 #undef N
123 };
124 static int nproc_root_targets =
125     sizeof(proc_root_targets) / sizeof(proc_root_targets[0]);
126 
127 int	procfs_lookup	__P((void *));
128 #define	procfs_create	genfs_eopnotsupp_rele
129 #define	procfs_mknod	genfs_eopnotsupp_rele
130 int	procfs_open	__P((void *));
131 int	procfs_close	__P((void *));
132 int	procfs_access	__P((void *));
133 int	procfs_getattr	__P((void *));
134 int	procfs_setattr	__P((void *));
135 #define	procfs_read	procfs_rw
136 #define	procfs_write	procfs_rw
137 #define	procfs_fcntl	genfs_fcntl
138 #define	procfs_ioctl	genfs_enoioctl
139 #define	procfs_poll	genfs_poll
140 #define procfs_revoke	genfs_revoke
141 #define	procfs_fsync	genfs_nullop
142 #define	procfs_seek	genfs_nullop
143 #define	procfs_remove	genfs_eopnotsupp_rele
144 int	procfs_link	__P((void *));
145 #define	procfs_rename	genfs_eopnotsupp_rele
146 #define	procfs_mkdir	genfs_eopnotsupp_rele
147 #define	procfs_rmdir	genfs_eopnotsupp_rele
148 int	procfs_symlink	__P((void *));
149 int	procfs_readdir	__P((void *));
150 int	procfs_readlink	__P((void *));
151 #define	procfs_abortop	genfs_abortop
152 int	procfs_inactive	__P((void *));
153 int	procfs_reclaim	__P((void *));
154 #define	procfs_lock	genfs_lock
155 #define	procfs_unlock	genfs_unlock
156 #define	procfs_bmap	genfs_badop
157 #define	procfs_strategy	genfs_badop
158 int	procfs_print	__P((void *));
159 int	procfs_pathconf	__P((void *));
160 #define	procfs_islocked	genfs_islocked
161 #define	procfs_advlock	genfs_einval
162 #define	procfs_blkatoff	genfs_eopnotsupp
163 #define	procfs_valloc	genfs_eopnotsupp
164 #define	procfs_vfree	genfs_nullop
165 #define	procfs_truncate	genfs_eopnotsupp
166 #define	procfs_update	genfs_nullop
167 #define	procfs_bwrite	genfs_eopnotsupp
168 #define procfs_putpages	genfs_null_putpages
169 
170 static pid_t atopid __P((const char *, u_int));
171 
172 /*
173  * procfs vnode operations.
174  */
175 int (**procfs_vnodeop_p) __P((void *));
176 const struct vnodeopv_entry_desc procfs_vnodeop_entries[] = {
177 	{ &vop_default_desc, vn_default_error },
178 	{ &vop_lookup_desc, procfs_lookup },		/* lookup */
179 	{ &vop_create_desc, procfs_create },		/* create */
180 	{ &vop_mknod_desc, procfs_mknod },		/* mknod */
181 	{ &vop_open_desc, procfs_open },		/* open */
182 	{ &vop_close_desc, procfs_close },		/* close */
183 	{ &vop_access_desc, procfs_access },		/* access */
184 	{ &vop_getattr_desc, procfs_getattr },		/* getattr */
185 	{ &vop_setattr_desc, procfs_setattr },		/* setattr */
186 	{ &vop_read_desc, procfs_read },		/* read */
187 	{ &vop_write_desc, procfs_write },		/* write */
188 	{ &vop_fcntl_desc, procfs_fcntl },		/* fcntl */
189 	{ &vop_ioctl_desc, procfs_ioctl },		/* ioctl */
190 	{ &vop_poll_desc, procfs_poll },		/* poll */
191 	{ &vop_revoke_desc, procfs_revoke },		/* revoke */
192 	{ &vop_fsync_desc, procfs_fsync },		/* fsync */
193 	{ &vop_seek_desc, procfs_seek },		/* seek */
194 	{ &vop_remove_desc, procfs_remove },		/* remove */
195 	{ &vop_link_desc, procfs_link },		/* link */
196 	{ &vop_rename_desc, procfs_rename },		/* rename */
197 	{ &vop_mkdir_desc, procfs_mkdir },		/* mkdir */
198 	{ &vop_rmdir_desc, procfs_rmdir },		/* rmdir */
199 	{ &vop_symlink_desc, procfs_symlink },		/* symlink */
200 	{ &vop_readdir_desc, procfs_readdir },		/* readdir */
201 	{ &vop_readlink_desc, procfs_readlink },	/* readlink */
202 	{ &vop_abortop_desc, procfs_abortop },		/* abortop */
203 	{ &vop_inactive_desc, procfs_inactive },	/* inactive */
204 	{ &vop_reclaim_desc, procfs_reclaim },		/* reclaim */
205 	{ &vop_lock_desc, procfs_lock },		/* lock */
206 	{ &vop_unlock_desc, procfs_unlock },		/* unlock */
207 	{ &vop_bmap_desc, procfs_bmap },		/* bmap */
208 	{ &vop_strategy_desc, procfs_strategy },	/* strategy */
209 	{ &vop_print_desc, procfs_print },		/* print */
210 	{ &vop_islocked_desc, procfs_islocked },	/* islocked */
211 	{ &vop_pathconf_desc, procfs_pathconf },	/* pathconf */
212 	{ &vop_advlock_desc, procfs_advlock },		/* advlock */
213 	{ &vop_blkatoff_desc, procfs_blkatoff },	/* blkatoff */
214 	{ &vop_valloc_desc, procfs_valloc },		/* valloc */
215 	{ &vop_vfree_desc, procfs_vfree },		/* vfree */
216 	{ &vop_truncate_desc, procfs_truncate },	/* truncate */
217 	{ &vop_update_desc, procfs_update },		/* update */
218 	{ &vop_putpages_desc, procfs_putpages },	/* putpages */
219 	{ NULL, NULL }
220 };
221 const struct vnodeopv_desc procfs_vnodeop_opv_desc =
222 	{ &procfs_vnodeop_p, procfs_vnodeop_entries };
223 /*
224  * set things up for doing i/o on
225  * the pfsnode (vp).  (vp) is locked
226  * on entry, and should be left locked
227  * on exit.
228  *
229  * for procfs we don't need to do anything
230  * in particular for i/o.  all that is done
231  * is to support exclusive open on process
232  * memory images.
233  */
234 int
235 procfs_open(v)
236 	void *v;
237 {
238 	struct vop_open_args /* {
239 		struct vnode *a_vp;
240 		int  a_mode;
241 		struct ucred *a_cred;
242 		struct proc *a_p;
243 	} */ *ap = v;
244 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
245 	struct proc *p1, *p2;
246 	int error;
247 
248 	p1 = ap->a_p;				/* tracer */
249 	p2 = PFIND(pfs->pfs_pid);		/* traced */
250 
251 	if (p2 == NULL)
252 		return (ENOENT);		/* was ESRCH, jsp */
253 
254 	switch (pfs->pfs_type) {
255 	case Pmem:
256 		if (((pfs->pfs_flags & FWRITE) && (ap->a_mode & O_EXCL)) ||
257 		    ((pfs->pfs_flags & O_EXCL) && (ap->a_mode & FWRITE)))
258 			return (EBUSY);
259 
260 		if ((error = process_checkioperm(p1, p2)) != 0)
261 			return (error);
262 
263 		if (ap->a_mode & FWRITE)
264 			pfs->pfs_flags = ap->a_mode & (FWRITE|O_EXCL);
265 
266 		return (0);
267 
268 	default:
269 		break;
270 	}
271 
272 	return (0);
273 }
274 
275 /*
276  * close the pfsnode (vp) after doing i/o.
277  * (vp) is not locked on entry or exit.
278  *
279  * nothing to do for procfs other than undo
280  * any exclusive open flag (see _open above).
281  */
282 int
283 procfs_close(v)
284 	void *v;
285 {
286 	struct vop_close_args /* {
287 		struct vnode *a_vp;
288 		int  a_fflag;
289 		struct ucred *a_cred;
290 		struct proc *a_p;
291 	} */ *ap = v;
292 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
293 
294 	switch (pfs->pfs_type) {
295 	case Pmem:
296 		if ((ap->a_fflag & FWRITE) && (pfs->pfs_flags & O_EXCL))
297 			pfs->pfs_flags &= ~(FWRITE|O_EXCL);
298 		break;
299 
300 	default:
301 		break;
302 	}
303 
304 	return (0);
305 }
306 
307 /*
308  * _inactive is called when the pfsnode
309  * is vrele'd and the reference count goes
310  * to zero.  (vp) will be on the vnode free
311  * list, so to get it back vget() must be
312  * used.
313  *
314  * for procfs, check if the process is still
315  * alive and if it isn't then just throw away
316  * the vnode by calling vgone().  this may
317  * be overkill and a waste of time since the
318  * chances are that the process will still be
319  * there and PFIND is not free.
320  *
321  * (vp) is locked on entry, but must be unlocked on exit.
322  */
323 int
324 procfs_inactive(v)
325 	void *v;
326 {
327 	struct vop_inactive_args /* {
328 		struct vnode *a_vp;
329 		struct proc *a_p;
330 	} */ *ap = v;
331 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
332 
333 	VOP_UNLOCK(ap->a_vp, 0);
334 	if (PFIND(pfs->pfs_pid) == 0)
335 		vgone(ap->a_vp);
336 
337 	return (0);
338 }
339 
340 /*
341  * _reclaim is called when getnewvnode()
342  * wants to make use of an entry on the vnode
343  * free list.  at this time the filesystem needs
344  * to free any private data and remove the node
345  * from any private lists.
346  */
347 int
348 procfs_reclaim(v)
349 	void *v;
350 {
351 	struct vop_reclaim_args /* {
352 		struct vnode *a_vp;
353 	} */ *ap = v;
354 
355 	return (procfs_freevp(ap->a_vp));
356 }
357 
358 /*
359  * Return POSIX pathconf information applicable to special devices.
360  */
361 int
362 procfs_pathconf(v)
363 	void *v;
364 {
365 	struct vop_pathconf_args /* {
366 		struct vnode *a_vp;
367 		int a_name;
368 		register_t *a_retval;
369 	} */ *ap = v;
370 
371 	switch (ap->a_name) {
372 	case _PC_LINK_MAX:
373 		*ap->a_retval = LINK_MAX;
374 		return (0);
375 	case _PC_MAX_CANON:
376 		*ap->a_retval = MAX_CANON;
377 		return (0);
378 	case _PC_MAX_INPUT:
379 		*ap->a_retval = MAX_INPUT;
380 		return (0);
381 	case _PC_PIPE_BUF:
382 		*ap->a_retval = PIPE_BUF;
383 		return (0);
384 	case _PC_CHOWN_RESTRICTED:
385 		*ap->a_retval = 1;
386 		return (0);
387 	case _PC_VDISABLE:
388 		*ap->a_retval = _POSIX_VDISABLE;
389 		return (0);
390 	case _PC_SYNC_IO:
391 		*ap->a_retval = 1;
392 		return (0);
393 	default:
394 		return (EINVAL);
395 	}
396 	/* NOTREACHED */
397 }
398 
399 /*
400  * _print is used for debugging.
401  * just print a readable description
402  * of (vp).
403  */
404 int
405 procfs_print(v)
406 	void *v;
407 {
408 	struct vop_print_args /* {
409 		struct vnode *a_vp;
410 	} */ *ap = v;
411 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
412 
413 	printf("tag VT_PROCFS, type %d, pid %d, mode %x, flags %lx\n",
414 	    pfs->pfs_type, pfs->pfs_pid, pfs->pfs_mode, pfs->pfs_flags);
415 	return 0;
416 }
417 
418 int
419 procfs_link(v)
420 	void *v;
421 {
422 	struct vop_link_args /* {
423 		struct vnode *a_dvp;
424 		struct vnode *a_vp;
425 		struct componentname *a_cnp;
426 	} */ *ap = v;
427 
428 	VOP_ABORTOP(ap->a_dvp, ap->a_cnp);
429 	vput(ap->a_dvp);
430 	return (EROFS);
431 }
432 
433 int
434 procfs_symlink(v)
435 	void *v;
436 {
437 	struct vop_symlink_args /* {
438 		struct vnode *a_dvp;
439 		struct vnode **a_vpp;
440 		struct componentname *a_cnp;
441 		struct vattr *a_vap;
442 		char *a_target;
443 	} */ *ap = v;
444 
445 	VOP_ABORTOP(ap->a_dvp, ap->a_cnp);
446 	vput(ap->a_dvp);
447 	return (EROFS);
448 }
449 
450 /*
451  * Invent attributes for pfsnode (vp) and store
452  * them in (vap).
453  * Directories lengths are returned as zero since
454  * any real length would require the genuine size
455  * to be computed, and nothing cares anyway.
456  *
457  * this is relatively minimal for procfs.
458  */
459 int
460 procfs_getattr(v)
461 	void *v;
462 {
463 	struct vop_getattr_args /* {
464 		struct vnode *a_vp;
465 		struct vattr *a_vap;
466 		struct ucred *a_cred;
467 		struct proc *a_p;
468 	} */ *ap = v;
469 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
470 	struct vattr *vap = ap->a_vap;
471 	struct proc *procp;
472 	int error;
473 
474 	/* first check the process still exists */
475 	switch (pfs->pfs_type) {
476 	case Proot:
477 	case Pcurproc:
478 	case Pself:
479 		procp = 0;
480 		break;
481 
482 	default:
483 		procp = PFIND(pfs->pfs_pid);
484 		if (procp == 0)
485 			return (ENOENT);
486 		break;
487 	}
488 
489 	error = 0;
490 
491 	/* start by zeroing out the attributes */
492 	VATTR_NULL(vap);
493 
494 	/* next do all the common fields */
495 	vap->va_type = ap->a_vp->v_type;
496 	vap->va_mode = pfs->pfs_mode;
497 	vap->va_fileid = pfs->pfs_fileno;
498 	vap->va_flags = 0;
499 	vap->va_blocksize = PAGE_SIZE;
500 
501 	/*
502 	 * Make all times be current TOD.  Avoid microtime(9), it's slow.
503 	 * We don't guard the read from time(9) with splclock(9) since we
504 	 * don't actually need to be THAT sure the access is atomic.
505 	 *
506 	 * It would be possible to get the process start
507 	 * time from the p_stat structure, but there's
508 	 * no "file creation" time stamp anyway, and the
509 	 * p_stat structure is not addressible if u. gets
510 	 * swapped out for that process.
511 	 */
512 	TIMEVAL_TO_TIMESPEC(&time, &vap->va_ctime);
513 	vap->va_atime = vap->va_mtime = vap->va_ctime;
514 
515 	switch (pfs->pfs_type) {
516 	case Pmem:
517 	case Pregs:
518 	case Pfpregs:
519 #if defined(__HAVE_PROCFS_MACHDEP) && defined(PROCFS_MACHDEP_PROTECT_CASES)
520 	PROCFS_MACHDEP_PROTECT_CASES
521 #endif
522 		/*
523 		 * If the process has exercised some setuid or setgid
524 		 * privilege, then rip away read/write permission so
525 		 * that only root can gain access.
526 		 */
527 		if (procp->p_flag & P_SUGID)
528 			vap->va_mode &= ~(S_IRUSR|S_IWUSR);
529 		/* FALLTHROUGH */
530 	case Pctl:
531 	case Pstatus:
532 	case Pnote:
533 	case Pnotepg:
534 	case Pmap:
535 	case Pmaps:
536 	case Pcmdline:
537 		vap->va_nlink = 1;
538 		vap->va_uid = procp->p_ucred->cr_uid;
539 		vap->va_gid = procp->p_ucred->cr_gid;
540 		break;
541 	case Pmeminfo:
542 	case Pcpuinfo:
543 		vap->va_nlink = 1;
544 		vap->va_uid = vap->va_gid = 0;
545 		break;
546 
547 	default:
548 		break;
549 	}
550 
551 	/*
552 	 * now do the object specific fields
553 	 *
554 	 * The size could be set from struct reg, but it's hardly
555 	 * worth the trouble, and it puts some (potentially) machine
556 	 * dependent data into this machine-independent code.  If it
557 	 * becomes important then this function should break out into
558 	 * a per-file stat function in the corresponding .c file.
559 	 */
560 
561 	switch (pfs->pfs_type) {
562 	case Proot:
563 		/*
564 		 * Set nlink to 1 to tell fts(3) we don't actually know.
565 		 */
566 		vap->va_nlink = 1;
567 		vap->va_uid = 0;
568 		vap->va_gid = 0;
569 		vap->va_bytes = vap->va_size = DEV_BSIZE;
570 		break;
571 
572 	case Pcurproc: {
573 		char buf[16];		/* should be enough */
574 		vap->va_nlink = 1;
575 		vap->va_uid = 0;
576 		vap->va_gid = 0;
577 		vap->va_bytes = vap->va_size =
578 		    sprintf(buf, "%ld", (long)curproc->p_pid);
579 		break;
580 	}
581 
582 	case Pself:
583 		vap->va_nlink = 1;
584 		vap->va_uid = 0;
585 		vap->va_gid = 0;
586 		vap->va_bytes = vap->va_size = sizeof("curproc");
587 		break;
588 
589 	case Pproc:
590 		vap->va_nlink = 2;
591 		vap->va_uid = procp->p_ucred->cr_uid;
592 		vap->va_gid = procp->p_ucred->cr_gid;
593 		vap->va_bytes = vap->va_size = DEV_BSIZE;
594 		break;
595 
596 	case Pfile:
597 		error = EOPNOTSUPP;
598 		break;
599 
600 	case Pmem:
601 		vap->va_bytes = vap->va_size =
602 			ctob(procp->p_vmspace->vm_tsize +
603 				    procp->p_vmspace->vm_dsize +
604 				    procp->p_vmspace->vm_ssize);
605 		break;
606 
607 #if defined(PT_GETREGS) || defined(PT_SETREGS)
608 	case Pregs:
609 		vap->va_bytes = vap->va_size = sizeof(struct reg);
610 		break;
611 #endif
612 
613 #if defined(PT_GETFPREGS) || defined(PT_SETFPREGS)
614 	case Pfpregs:
615 		vap->va_bytes = vap->va_size = sizeof(struct fpreg);
616 		break;
617 #endif
618 
619 	case Pctl:
620 	case Pstatus:
621 	case Pnote:
622 	case Pnotepg:
623 	case Pcmdline:
624 	case Pmeminfo:
625 	case Pcpuinfo:
626 		vap->va_bytes = vap->va_size = 0;
627 		break;
628 	case Pmap:
629 	case Pmaps:
630 		/*
631 		 * Advise a larger blocksize for the map files, so that
632 		 * they may be read in one pass.
633 		 */
634 		vap->va_blocksize = 4 * PAGE_SIZE;
635 		vap->va_bytes = vap->va_size = 0;
636 		break;
637 
638 #ifdef __HAVE_PROCFS_MACHDEP
639 	PROCFS_MACHDEP_NODETYPE_CASES
640 		error = procfs_machdep_getattr(ap->a_vp, vap, procp);
641 		break;
642 #endif
643 
644 	default:
645 		panic("procfs_getattr");
646 	}
647 
648 	return (error);
649 }
650 
651 /*ARGSUSED*/
652 int
653 procfs_setattr(v)
654 	void *v;
655 {
656 	/*
657 	 * just fake out attribute setting
658 	 * it's not good to generate an error
659 	 * return, otherwise things like creat()
660 	 * will fail when they try to set the
661 	 * file length to 0.  worse, this means
662 	 * that echo $note > /proc/$pid/note will fail.
663 	 */
664 
665 	return (0);
666 }
667 
668 /*
669  * implement access checking.
670  *
671  * actually, the check for super-user is slightly
672  * broken since it will allow read access to write-only
673  * objects.  this doesn't cause any particular trouble
674  * but does mean that the i/o entry points need to check
675  * that the operation really does make sense.
676  */
677 int
678 procfs_access(v)
679 	void *v;
680 {
681 	struct vop_access_args /* {
682 		struct vnode *a_vp;
683 		int a_mode;
684 		struct ucred *a_cred;
685 		struct proc *a_p;
686 	} */ *ap = v;
687 	struct vattr va;
688 	int error;
689 
690 	if ((error = VOP_GETATTR(ap->a_vp, &va, ap->a_cred, ap->a_p)) != 0)
691 		return (error);
692 
693 	return (vaccess(va.va_type, va.va_mode,
694 	    va.va_uid, va.va_gid, ap->a_mode, ap->a_cred));
695 }
696 
697 /*
698  * lookup.  this is incredibly complicated in the
699  * general case, however for most pseudo-filesystems
700  * very little needs to be done.
701  *
702  * Locking isn't hard here, just poorly documented.
703  *
704  * If we're looking up ".", just vref the parent & return it.
705  *
706  * If we're looking up "..", unlock the parent, and lock "..". If everything
707  * went ok, and we're on the last component and the caller requested the
708  * parent locked, try to re-lock the parent. We do this to prevent lock
709  * races.
710  *
711  * For anything else, get the needed node. Then unlock the parent if not
712  * the last component or not LOCKPARENT (i.e. if we wouldn't re-lock the
713  * parent in the .. case).
714  *
715  * We try to exit with the parent locked in error cases.
716  */
717 int
718 procfs_lookup(v)
719 	void *v;
720 {
721 	struct vop_lookup_args /* {
722 		struct vnode * a_dvp;
723 		struct vnode ** a_vpp;
724 		struct componentname * a_cnp;
725 	} */ *ap = v;
726 	struct componentname *cnp = ap->a_cnp;
727 	struct vnode **vpp = ap->a_vpp;
728 	struct vnode *dvp = ap->a_dvp;
729 	const char *pname = cnp->cn_nameptr;
730 	const struct proc_target *pt = NULL;
731 	struct vnode *fvp;
732 	pid_t pid;
733 	struct pfsnode *pfs;
734 	struct proc *p = NULL;
735 	int i, error, wantpunlock, iscurproc = 0, isself = 0;
736 
737 	*vpp = NULL;
738 	cnp->cn_flags &= ~PDIRUNLOCK;
739 
740 	if (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)
741 		return (EROFS);
742 
743 	if (cnp->cn_namelen == 1 && *pname == '.') {
744 		*vpp = dvp;
745 		VREF(dvp);
746 		return (0);
747 	}
748 
749 	wantpunlock = (~cnp->cn_flags & (LOCKPARENT | ISLASTCN));
750 	pfs = VTOPFS(dvp);
751 	switch (pfs->pfs_type) {
752 	case Proot:
753 		/*
754 		 * Shouldn't get here with .. in the root node.
755 		 */
756 		if (cnp->cn_flags & ISDOTDOT)
757 			return (EIO);
758 
759 		iscurproc = CNEQ(cnp, "curproc", 7);
760 		isself = CNEQ(cnp, "self", 4);
761 
762 		if (iscurproc || isself) {
763 			error = procfs_allocvp(dvp->v_mount, vpp, 0,
764 			    iscurproc ? Pcurproc : Pself);
765 			if ((error == 0) && (wantpunlock)) {
766 				VOP_UNLOCK(dvp, 0);
767 				cnp->cn_flags |= PDIRUNLOCK;
768 			}
769 			return (error);
770 		}
771 
772 		for (i = 0; i < nproc_root_targets; i++) {
773 			pt = &proc_root_targets[i];
774 			if (cnp->cn_namelen == pt->pt_namlen &&
775 			    memcmp(pt->pt_name, pname, cnp->cn_namelen) == 0 &&
776 			    (pt->pt_valid == NULL ||
777 			     (*pt->pt_valid)(p, dvp->v_mount)))
778 				break;
779 		}
780 
781 		if (i != nproc_root_targets) {
782 			error = procfs_allocvp(dvp->v_mount, vpp, 0,
783 			    pt->pt_pfstype);
784 			if ((error == 0) && (wantpunlock)) {
785 				VOP_UNLOCK(dvp, 0);
786 				cnp->cn_flags |= PDIRUNLOCK;
787 			}
788 			return (error);
789 		}
790 
791 		pid = atopid(pname, cnp->cn_namelen);
792 		if (pid == NO_PID)
793 			break;
794 
795 		p = PFIND(pid);
796 		if (p == 0)
797 			break;
798 
799 		error = procfs_allocvp(dvp->v_mount, vpp, pid, Pproc);
800 		if ((error == 0) && (wantpunlock)) {
801 			VOP_UNLOCK(dvp, 0);
802 			cnp->cn_flags |= PDIRUNLOCK;
803 		}
804 		return (error);
805 
806 	case Pproc:
807 		/*
808 		 * do the .. dance. We unlock the directory, and then
809 		 * get the root dir. That will automatically return ..
810 		 * locked. Then if the caller wanted dvp locked, we
811 		 * re-lock.
812 		 */
813 		if (cnp->cn_flags & ISDOTDOT) {
814 			VOP_UNLOCK(dvp, 0);
815 			cnp->cn_flags |= PDIRUNLOCK;
816 			error = procfs_root(dvp->v_mount, vpp);
817 			if ((error == 0) && (wantpunlock == 0) &&
818 				    ((error = vn_lock(dvp, LK_EXCLUSIVE)) == 0))
819 				cnp->cn_flags &= ~PDIRUNLOCK;
820 			return (error);
821 		}
822 
823 		p = PFIND(pfs->pfs_pid);
824 		if (p == 0)
825 			break;
826 
827 		for (pt = proc_targets, i = 0; i < nproc_targets; pt++, i++) {
828 			if (cnp->cn_namelen == pt->pt_namlen &&
829 			    memcmp(pt->pt_name, pname, cnp->cn_namelen) == 0 &&
830 			    (pt->pt_valid == NULL ||
831 			     (*pt->pt_valid)(p, dvp->v_mount)))
832 				goto found;
833 		}
834 		break;
835 
836 	found:
837 		if (pt->pt_pfstype == Pfile) {
838 			fvp = p->p_textvp;
839 			/* We already checked that it exists. */
840 			VREF(fvp);
841 			vn_lock(fvp, LK_EXCLUSIVE | LK_RETRY);
842 			if (wantpunlock) {
843 				VOP_UNLOCK(dvp, 0);
844 				cnp->cn_flags |= PDIRUNLOCK;
845 			}
846 			*vpp = fvp;
847 			return (0);
848 		}
849 
850 		error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
851 					pt->pt_pfstype);
852 		if ((error == 0) && (wantpunlock)) {
853 			VOP_UNLOCK(dvp, 0);
854 			cnp->cn_flags |= PDIRUNLOCK;
855 		}
856 		return (error);
857 
858 	default:
859 		return (ENOTDIR);
860 	}
861 
862 	return (cnp->cn_nameiop == LOOKUP ? ENOENT : EROFS);
863 }
864 
865 int
866 procfs_validfile(p, mp)
867 	struct proc *p;
868 	struct mount *mp;
869 {
870 	return (p->p_textvp != NULL);
871 }
872 
873 static int
874 procfs_validfile_linux(p, mp)
875 	struct proc *p;
876 	struct mount *mp;
877 {
878 	int flags;
879 
880 	flags = VFSTOPROC(mp)->pmnt_flags;
881 	return ((flags & PROCFSMNT_LINUXCOMPAT) &&
882 	    (p == NULL || procfs_validfile(p, mp)));
883 }
884 
885 /*
886  * readdir returns directory entries from pfsnode (vp).
887  *
888  * the strategy here with procfs is to generate a single
889  * directory entry at a time (struct dirent) and then
890  * copy that out to userland using uiomove.  a more efficent
891  * though more complex implementation, would try to minimize
892  * the number of calls to uiomove().  for procfs, this is
893  * hardly worth the added code complexity.
894  *
895  * this should just be done through read()
896  */
897 int
898 procfs_readdir(v)
899 	void *v;
900 {
901 	struct vop_readdir_args /* {
902 		struct vnode *a_vp;
903 		struct uio *a_uio;
904 		struct ucred *a_cred;
905 		int *a_eofflag;
906 		off_t **a_cookies;
907 		int *a_ncookies;
908 	} */ *ap = v;
909 	struct uio *uio = ap->a_uio;
910 	struct dirent d;
911 	struct pfsnode *pfs;
912 	off_t i;
913 	int error;
914 	off_t *cookies = NULL;
915 	int ncookies, left, skip, j;
916 	struct vnode *vp;
917 	const struct proc_target *pt;
918 
919 	vp = ap->a_vp;
920 	pfs = VTOPFS(vp);
921 
922 	if (uio->uio_resid < UIO_MX)
923 		return (EINVAL);
924 	if (uio->uio_offset < 0)
925 		return (EINVAL);
926 
927 	error = 0;
928 	i = uio->uio_offset;
929 	memset((caddr_t)&d, 0, UIO_MX);
930 	d.d_reclen = UIO_MX;
931 	ncookies = uio->uio_resid / UIO_MX;
932 
933 	switch (pfs->pfs_type) {
934 	/*
935 	 * this is for the process-specific sub-directories.
936 	 * all that is needed to is copy out all the entries
937 	 * from the procent[] table (top of this file).
938 	 */
939 	case Pproc: {
940 		struct proc *p;
941 
942 		if (i >= nproc_targets)
943 			return 0;
944 
945 		p = PFIND(pfs->pfs_pid);
946 		if (p == NULL)
947 			break;
948 
949 		if (ap->a_ncookies) {
950 			ncookies = min(ncookies, (nproc_targets - i));
951 			cookies = malloc(ncookies * sizeof (off_t),
952 			    M_TEMP, M_WAITOK);
953 			*ap->a_cookies = cookies;
954 		}
955 
956 		for (pt = &proc_targets[i];
957 		     uio->uio_resid >= UIO_MX && i < nproc_targets; pt++, i++) {
958 			if (pt->pt_valid &&
959 			    (*pt->pt_valid)(p, vp->v_mount) == 0)
960 				continue;
961 
962 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid, pt->pt_pfstype);
963 			d.d_namlen = pt->pt_namlen;
964 			memcpy(d.d_name, pt->pt_name, pt->pt_namlen + 1);
965 			d.d_type = pt->pt_type;
966 
967 			if ((error = uiomove((caddr_t)&d, UIO_MX, uio)) != 0)
968 				break;
969 			if (cookies)
970 				*cookies++ = i + 1;
971 		}
972 
973 	    	break;
974 	}
975 
976 	/*
977 	 * this is for the root of the procfs filesystem
978 	 * what is needed are special entries for "curproc"
979 	 * and "self" followed by an entry for each process
980 	 * on allproc
981 #ifdef PROCFS_ZOMBIE
982 	 * and deadproc and zombproc.
983 #endif
984 	 */
985 
986 	case Proot: {
987 		int pcnt = i, nc = 0;
988 		const struct proclist_desc *pd;
989 		volatile struct proc *p;
990 
991 		if (pcnt > 3)
992 			pcnt = 3;
993 		if (ap->a_ncookies) {
994 			/*
995 			 * XXX Potentially allocating too much space here,
996 			 * but I'm lazy. This loop needs some work.
997 			 */
998 			cookies = malloc(ncookies * sizeof (off_t),
999 			    M_TEMP, M_WAITOK);
1000 			*ap->a_cookies = cookies;
1001 		}
1002 		/*
1003 		 * XXX: THIS LOOP ASSUMES THAT allproc IS THE FIRST
1004 		 * PROCLIST IN THE proclists!
1005 		 */
1006 		proclist_lock_read();
1007 		pd = proclists;
1008 #ifdef PROCFS_ZOMBIE
1009 	again:
1010 #endif
1011 		for (p = LIST_FIRST(pd->pd_list);
1012 		     p != NULL && uio->uio_resid >= UIO_MX; i++, pcnt++) {
1013 			switch (i) {
1014 			case 0:		/* `.' */
1015 			case 1:		/* `..' */
1016 				d.d_fileno = PROCFS_FILENO(0, Proot);
1017 				d.d_namlen = i + 1;
1018 				memcpy(d.d_name, "..", d.d_namlen);
1019 				d.d_name[i + 1] = '\0';
1020 				d.d_type = DT_DIR;
1021 				break;
1022 
1023 			case 2:
1024 				d.d_fileno = PROCFS_FILENO(0, Pcurproc);
1025 				d.d_namlen = sizeof("curproc") - 1;
1026 				memcpy(d.d_name, "curproc", sizeof("curproc"));
1027 				d.d_type = DT_LNK;
1028 				break;
1029 
1030 			case 3:
1031 				d.d_fileno = PROCFS_FILENO(0, Pself);
1032 				d.d_namlen = sizeof("self") - 1;
1033 				memcpy(d.d_name, "self", sizeof("self"));
1034 				d.d_type = DT_LNK;
1035 				break;
1036 
1037 			default:
1038 				while (pcnt < i) {
1039 					pcnt++;
1040 					p = LIST_NEXT(p, p_list);
1041 					if (!p)
1042 						goto done;
1043 				}
1044 				d.d_fileno = PROCFS_FILENO(p->p_pid, Pproc);
1045 				d.d_namlen = sprintf(d.d_name, "%ld",
1046 				    (long)p->p_pid);
1047 				d.d_type = DT_DIR;
1048 				p = p->p_list.le_next;
1049 				break;
1050 			}
1051 
1052 			if ((error = uiomove((caddr_t)&d, UIO_MX, uio)) != 0)
1053 				break;
1054 			nc++;
1055 			if (cookies)
1056 				*cookies++ = i + 1;
1057 		}
1058 	done:
1059 
1060 #ifdef PROCFS_ZOMBIE
1061 		pd++;
1062 		if (p == NULL && pd->pd_list != NULL)
1063 			goto again;
1064 #endif
1065 		proclist_unlock_read();
1066 
1067 		skip = i - pcnt;
1068 		if (skip >= nproc_root_targets)
1069 			break;
1070 		left = nproc_root_targets - skip;
1071 		for (j = 0, pt = &proc_root_targets[0];
1072 		     uio->uio_resid >= UIO_MX && j < left;
1073 		     pt++, j++, i++) {
1074 			if (pt->pt_valid &&
1075 			    (*pt->pt_valid)(NULL, vp->v_mount) == 0)
1076 				continue;
1077 			d.d_fileno = PROCFS_FILENO(0, pt->pt_pfstype);
1078 			d.d_namlen = pt->pt_namlen;
1079 			memcpy(d.d_name, pt->pt_name, pt->pt_namlen + 1);
1080 			d.d_type = pt->pt_type;
1081 
1082 			if ((error = uiomove((caddr_t)&d, UIO_MX, uio)) != 0)
1083 				break;
1084 			nc++;
1085 			if (cookies)
1086 				*cookies++ = i + 1;
1087 		}
1088 
1089 		ncookies = nc;
1090 		break;
1091 	}
1092 
1093 	default:
1094 		error = ENOTDIR;
1095 		break;
1096 	}
1097 
1098 	if (ap->a_ncookies) {
1099 		if (error) {
1100 			if (cookies)
1101 				free(*ap->a_cookies, M_TEMP);
1102 			*ap->a_ncookies = 0;
1103 			*ap->a_cookies = NULL;
1104 		} else
1105 			*ap->a_ncookies = ncookies;
1106 	}
1107 	uio->uio_offset = i;
1108 	return (error);
1109 }
1110 
1111 /*
1112  * readlink reads the link of `curproc'
1113  */
1114 int
1115 procfs_readlink(v)
1116 	void *v;
1117 {
1118 	struct vop_readlink_args *ap = v;
1119 	char buf[16];		/* should be enough */
1120 	int len;
1121 
1122 	if (VTOPFS(ap->a_vp)->pfs_fileno == PROCFS_FILENO(0, Pcurproc))
1123 		len = sprintf(buf, "%ld", (long)curproc->p_pid);
1124 	else if (VTOPFS(ap->a_vp)->pfs_fileno == PROCFS_FILENO(0, Pself))
1125 		len = sprintf(buf, "%s", "curproc");
1126 	else
1127 		return (EINVAL);
1128 
1129 	return (uiomove((caddr_t)buf, len, ap->a_uio));
1130 }
1131 
1132 /*
1133  * convert decimal ascii to pid_t
1134  */
1135 static pid_t
1136 atopid(b, len)
1137 	const char *b;
1138 	u_int len;
1139 {
1140 	pid_t p = 0;
1141 
1142 	while (len--) {
1143 		char c = *b++;
1144 		if (c < '0' || c > '9')
1145 			return (NO_PID);
1146 		p = 10 * p + (c - '0');
1147 		if (p > PID_MAX)
1148 			return (NO_PID);
1149 	}
1150 
1151 	return (p);
1152 }
1153