xref: /netbsd-src/sys/miscfs/procfs/procfs_vnops.c (revision 037708cbd4616ccd0d7d0381ebd3964d6696c188)
1 /*	$NetBSD: procfs_vnops.c,v 1.46 1997/04/28 03:49:57 mycroft Exp $	*/
2 
3 /*
4  * Copyright (c) 1993 Jan-Simon Pendry
5  * Copyright (c) 1993
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	8.8 (Berkeley) 6/15/94
40  */
41 
42 /*
43  * procfs vnode interface
44  */
45 
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 #include <sys/time.h>
49 #include <sys/kernel.h>
50 #include <sys/file.h>
51 #include <sys/proc.h>
52 #include <sys/vnode.h>
53 #include <sys/namei.h>
54 #include <sys/malloc.h>
55 #include <sys/dirent.h>
56 #include <sys/resourcevar.h>
57 #include <sys/ptrace.h>
58 #include <vm/vm.h>	/* for PAGE_SIZE */
59 #include <machine/reg.h>
60 
61 #include <miscfs/genfs/genfs.h>
62 #include <miscfs/procfs/procfs.h>
63 
64 /*
65  * Vnode Operations.
66  *
67  */
68 
69 /*
70  * This is a list of the valid names in the
71  * process-specific sub-directories.  It is
72  * used in procfs_lookup and procfs_readdir
73  */
74 struct proc_target {
75 	u_char	pt_type;
76 	u_char	pt_namlen;
77 	char	*pt_name;
78 	pfstype	pt_pfstype;
79 	int	(*pt_valid) __P((struct proc *p));
80 } proc_targets[] = {
81 #define N(s) sizeof(s)-1, s
82 	/*	  name		type		validp */
83 	{ DT_DIR, N("."),	Pproc,		NULL },
84 	{ DT_DIR, N(".."),	Proot,		NULL },
85 	{ DT_REG, N("file"),	Pfile,		procfs_validfile },
86 	{ DT_REG, N("mem"),	Pmem,		NULL },
87 	{ DT_REG, N("regs"),	Pregs,		procfs_validregs },
88 	{ DT_REG, N("fpregs"),	Pfpregs,	procfs_validfpregs },
89 	{ DT_REG, N("ctl"),	Pctl,		NULL },
90 	{ DT_REG, N("status"),	Pstatus,	NULL },
91 	{ DT_REG, N("note"),	Pnote,		NULL },
92 	{ DT_REG, N("notepg"),	Pnotepg,	NULL },
93 #undef N
94 };
95 static int nproc_targets = sizeof(proc_targets) / sizeof(proc_targets[0]);
96 
97 static pid_t atopid __P((const char *, u_int));
98 
99 int	procfs_lookup	__P((void *));
100 #define	procfs_create	genfs_eopnotsupp
101 #define	procfs_mknod	genfs_eopnotsupp
102 int	procfs_open	__P((void *));
103 int	procfs_close	__P((void *));
104 int	procfs_access	__P((void *));
105 int	procfs_getattr	__P((void *));
106 int	procfs_setattr	__P((void *));
107 #define	procfs_read	procfs_rw
108 #define	procfs_write	procfs_rw
109 #define	procfs_ioctl	genfs_eopnotsupp
110 #define	procfs_poll	genfs_poll
111 #define	procfs_mmap	genfs_eopnotsupp
112 #define	procfs_fsync	genfs_nullop
113 #define	procfs_seek	genfs_nullop
114 #define	procfs_remove	genfs_eopnotsupp
115 int	procfs_link	__P((void *));
116 #define	procfs_rename	genfs_eopnotsupp
117 #define	procfs_mkdir	genfs_eopnotsupp
118 #define	procfs_rmdir	genfs_eopnotsupp
119 int	procfs_symlink	__P((void *));
120 int	procfs_readdir	__P((void *));
121 int	procfs_readlink	__P((void *));
122 #define	procfs_abortop	genfs_abortop
123 int	procfs_inactive	__P((void *));
124 int	procfs_reclaim	__P((void *));
125 #define	procfs_lock	genfs_nullop
126 #define	procfs_unlock	genfs_nullop
127 int	procfs_bmap	__P((void *));
128 #define	procfs_strategy	genfs_badop
129 int	procfs_print	__P((void *));
130 int	procfs_pathconf	__P((void *));
131 #define	procfs_islocked	genfs_nullop
132 #define	procfs_advlock	genfs_eopnotsupp
133 #define	procfs_blkatoff	genfs_eopnotsupp
134 #define	procfs_valloc	genfs_eopnotsupp
135 #define	procfs_vfree	genfs_nullop
136 #define	procfs_truncate	genfs_eopnotsupp
137 #define	procfs_update	genfs_nullop
138 #define	procfs_bwrite	genfs_eopnotsupp
139 
140 static pid_t atopid __P((const char *, u_int));
141 
142 /*
143  * procfs vnode operations.
144  */
145 int (**procfs_vnodeop_p) __P((void *));
146 struct vnodeopv_entry_desc procfs_vnodeop_entries[] = {
147 	{ &vop_default_desc, vn_default_error },
148 	{ &vop_lookup_desc, procfs_lookup },		/* lookup */
149 	{ &vop_create_desc, procfs_create },		/* create */
150 	{ &vop_mknod_desc, procfs_mknod },		/* mknod */
151 	{ &vop_open_desc, procfs_open },		/* open */
152 	{ &vop_close_desc, procfs_close },		/* close */
153 	{ &vop_access_desc, procfs_access },		/* access */
154 	{ &vop_getattr_desc, procfs_getattr },		/* getattr */
155 	{ &vop_setattr_desc, procfs_setattr },		/* setattr */
156 	{ &vop_read_desc, procfs_read },		/* read */
157 	{ &vop_write_desc, procfs_write },		/* write */
158 	{ &vop_ioctl_desc, procfs_ioctl },		/* ioctl */
159 	{ &vop_poll_desc, procfs_poll },		/* poll */
160 	{ &vop_mmap_desc, procfs_mmap },		/* mmap */
161 	{ &vop_fsync_desc, procfs_fsync },		/* fsync */
162 	{ &vop_seek_desc, procfs_seek },		/* seek */
163 	{ &vop_remove_desc, procfs_remove },		/* remove */
164 	{ &vop_link_desc, procfs_link },		/* link */
165 	{ &vop_rename_desc, procfs_rename },		/* rename */
166 	{ &vop_mkdir_desc, procfs_mkdir },		/* mkdir */
167 	{ &vop_rmdir_desc, procfs_rmdir },		/* rmdir */
168 	{ &vop_symlink_desc, procfs_symlink },		/* symlink */
169 	{ &vop_readdir_desc, procfs_readdir },		/* readdir */
170 	{ &vop_readlink_desc, procfs_readlink },	/* readlink */
171 	{ &vop_abortop_desc, procfs_abortop },		/* abortop */
172 	{ &vop_inactive_desc, procfs_inactive },	/* inactive */
173 	{ &vop_reclaim_desc, procfs_reclaim },		/* reclaim */
174 	{ &vop_lock_desc, procfs_lock },		/* lock */
175 	{ &vop_unlock_desc, procfs_unlock },		/* unlock */
176 	{ &vop_bmap_desc, procfs_bmap },		/* bmap */
177 	{ &vop_strategy_desc, procfs_strategy },	/* strategy */
178 	{ &vop_print_desc, procfs_print },		/* print */
179 	{ &vop_islocked_desc, procfs_islocked },	/* islocked */
180 	{ &vop_pathconf_desc, procfs_pathconf },	/* pathconf */
181 	{ &vop_advlock_desc, procfs_advlock },		/* advlock */
182 	{ &vop_blkatoff_desc, procfs_blkatoff },	/* blkatoff */
183 	{ &vop_valloc_desc, procfs_valloc },		/* valloc */
184 	{ &vop_vfree_desc, procfs_vfree },		/* vfree */
185 	{ &vop_truncate_desc, procfs_truncate },	/* truncate */
186 	{ &vop_update_desc, procfs_update },		/* update */
187 	{ (struct vnodeop_desc*)NULL, (int(*) __P((void *)))NULL }
188 };
189 struct vnodeopv_desc procfs_vnodeop_opv_desc =
190 	{ &procfs_vnodeop_p, procfs_vnodeop_entries };
191 /*
192  * set things up for doing i/o on
193  * the pfsnode (vp).  (vp) is locked
194  * on entry, and should be left locked
195  * on exit.
196  *
197  * for procfs we don't need to do anything
198  * in particular for i/o.  all that is done
199  * is to support exclusive open on process
200  * memory images.
201  */
202 int
203 procfs_open(v)
204 	void *v;
205 {
206 	struct vop_open_args /* {
207 		struct vnode *a_vp;
208 		int  a_mode;
209 		struct ucred *a_cred;
210 		struct proc *a_p;
211 	} */ *ap = v;
212 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
213 
214 	switch (pfs->pfs_type) {
215 	case Pmem:
216 		if (PFIND(pfs->pfs_pid) == 0)
217 			return (ENOENT);	/* was ESRCH, jsp */
218 
219 		if (((pfs->pfs_flags & FWRITE) && (ap->a_mode & O_EXCL)) ||
220 		    ((pfs->pfs_flags & O_EXCL) && (ap->a_mode & FWRITE)))
221 			return (EBUSY);
222 
223 		if (ap->a_mode & FWRITE)
224 			pfs->pfs_flags = ap->a_mode & (FWRITE|O_EXCL);
225 
226 		return (0);
227 
228 	default:
229 		break;
230 	}
231 
232 	return (0);
233 }
234 
235 /*
236  * close the pfsnode (vp) after doing i/o.
237  * (vp) is not locked on entry or exit.
238  *
239  * nothing to do for procfs other than undo
240  * any exclusive open flag (see _open above).
241  */
242 int
243 procfs_close(v)
244 	void *v;
245 {
246 	struct vop_close_args /* {
247 		struct vnode *a_vp;
248 		int  a_fflag;
249 		struct ucred *a_cred;
250 		struct proc *a_p;
251 	} */ *ap = v;
252 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
253 
254 	switch (pfs->pfs_type) {
255 	case Pmem:
256 		if ((ap->a_fflag & FWRITE) && (pfs->pfs_flags & O_EXCL))
257 			pfs->pfs_flags &= ~(FWRITE|O_EXCL);
258 		break;
259 
260 	default:
261 		break;
262 	}
263 
264 	return (0);
265 }
266 
267 /*
268  * do block mapping for pfsnode (vp).
269  * since we don't use the buffer cache
270  * for procfs this function should never
271  * be called.  in any case, it's not clear
272  * what part of the kernel ever makes use
273  * of this function.  for sanity, this is the
274  * usual no-op bmap, although returning
275  * (EIO) would be a reasonable alternative.
276  */
277 int
278 procfs_bmap(v)
279 	void *v;
280 {
281 	struct vop_bmap_args /* {
282 		struct vnode *a_vp;
283 		daddr_t  a_bn;
284 		struct vnode **a_vpp;
285 		daddr_t *a_bnp;
286 	} */ *ap = v;
287 
288 	if (ap->a_vpp != NULL)
289 		*ap->a_vpp = ap->a_vp;
290 	if (ap->a_bnp != NULL)
291 		*ap->a_bnp = ap->a_bn;
292 	return (0);
293 }
294 
295 /*
296  * _inactive is called when the pfsnode
297  * is vrele'd and the reference count goes
298  * to zero.  (vp) will be on the vnode free
299  * list, so to get it back vget() must be
300  * used.
301  *
302  * for procfs, check if the process is still
303  * alive and if it isn't then just throw away
304  * the vnode by calling vgone().  this may
305  * be overkill and a waste of time since the
306  * chances are that the process will still be
307  * there and PFIND is not free.
308  *
309  * (vp) is not locked on entry or exit.
310  */
311 int
312 procfs_inactive(v)
313 	void *v;
314 {
315 	struct vop_inactive_args /* {
316 		struct vnode *a_vp;
317 	} */ *ap = v;
318 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
319 
320 	if (PFIND(pfs->pfs_pid) == 0)
321 		vgone(ap->a_vp);
322 
323 	return (0);
324 }
325 
326 /*
327  * _reclaim is called when getnewvnode()
328  * wants to make use of an entry on the vnode
329  * free list.  at this time the filesystem needs
330  * to free any private data and remove the node
331  * from any private lists.
332  */
333 int
334 procfs_reclaim(v)
335 	void *v;
336 {
337 	struct vop_reclaim_args /* {
338 		struct vnode *a_vp;
339 	} */ *ap = v;
340 
341 	return (procfs_freevp(ap->a_vp));
342 }
343 
344 /*
345  * Return POSIX pathconf information applicable to special devices.
346  */
347 int
348 procfs_pathconf(v)
349 	void *v;
350 {
351 	struct vop_pathconf_args /* {
352 		struct vnode *a_vp;
353 		int a_name;
354 		register_t *a_retval;
355 	} */ *ap = v;
356 
357 	switch (ap->a_name) {
358 	case _PC_LINK_MAX:
359 		*ap->a_retval = LINK_MAX;
360 		return (0);
361 	case _PC_MAX_CANON:
362 		*ap->a_retval = MAX_CANON;
363 		return (0);
364 	case _PC_MAX_INPUT:
365 		*ap->a_retval = MAX_INPUT;
366 		return (0);
367 	case _PC_PIPE_BUF:
368 		*ap->a_retval = PIPE_BUF;
369 		return (0);
370 	case _PC_CHOWN_RESTRICTED:
371 		*ap->a_retval = 1;
372 		return (0);
373 	case _PC_VDISABLE:
374 		*ap->a_retval = _POSIX_VDISABLE;
375 		return (0);
376 	default:
377 		return (EINVAL);
378 	}
379 	/* NOTREACHED */
380 }
381 
382 /*
383  * _print is used for debugging.
384  * just print a readable description
385  * of (vp).
386  */
387 int
388 procfs_print(v)
389 	void *v;
390 {
391 	struct vop_print_args /* {
392 		struct vnode *a_vp;
393 	} */ *ap = v;
394 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
395 
396 	printf("tag VT_PROCFS, type %d, pid %d, mode %x, flags %lx\n",
397 	    pfs->pfs_type, pfs->pfs_pid, pfs->pfs_mode, pfs->pfs_flags);
398 	return 0;
399 }
400 
401 int
402 procfs_link(v)
403 	void *v;
404 {
405 	struct vop_link_args /* {
406 		struct vnode *a_dvp;
407 		struct vnode *a_vp;
408 		struct componentname *a_cnp;
409 	} */ *ap = v;
410 
411 	VOP_ABORTOP(ap->a_dvp, ap->a_cnp);
412 	vput(ap->a_dvp);
413 	return (EROFS);
414 }
415 
416 int
417 procfs_symlink(v)
418 	void *v;
419 {
420 	struct vop_symlink_args /* {
421 		struct vnode *a_dvp;
422 		struct vnode **a_vpp;
423 		struct componentname *a_cnp;
424 		struct vattr *a_vap;
425 		char *a_target;
426 	} */ *ap = v;
427 
428 	VOP_ABORTOP(ap->a_dvp, ap->a_cnp);
429 	vput(ap->a_dvp);
430 	return (EROFS);
431 }
432 
433 /*
434  * Invent attributes for pfsnode (vp) and store
435  * them in (vap).
436  * Directories lengths are returned as zero since
437  * any real length would require the genuine size
438  * to be computed, and nothing cares anyway.
439  *
440  * this is relatively minimal for procfs.
441  */
442 int
443 procfs_getattr(v)
444 	void *v;
445 {
446 	struct vop_getattr_args /* {
447 		struct vnode *a_vp;
448 		struct vattr *a_vap;
449 		struct ucred *a_cred;
450 		struct proc *a_p;
451 	} */ *ap = v;
452 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
453 	struct vattr *vap = ap->a_vap;
454 	struct proc *procp;
455 	struct timeval tv;
456 	int error;
457 
458 	/* first check the process still exists */
459 	switch (pfs->pfs_type) {
460 	case Proot:
461 	case Pcurproc:
462 		procp = 0;
463 		break;
464 
465 	default:
466 		procp = PFIND(pfs->pfs_pid);
467 		if (procp == 0)
468 			return (ENOENT);
469 		break;
470 	}
471 
472 	error = 0;
473 
474 	/* start by zeroing out the attributes */
475 	VATTR_NULL(vap);
476 
477 	/* next do all the common fields */
478 	vap->va_type = ap->a_vp->v_type;
479 	vap->va_mode = pfs->pfs_mode;
480 	vap->va_fileid = pfs->pfs_fileno;
481 	vap->va_flags = 0;
482 	vap->va_blocksize = PAGE_SIZE;
483 
484 	/*
485 	 * Make all times be current TOD.
486 	 * It would be possible to get the process start
487 	 * time from the p_stat structure, but there's
488 	 * no "file creation" time stamp anyway, and the
489 	 * p_stat structure is not addressible if u. gets
490 	 * swapped out for that process.
491 	 */
492 	microtime(&tv);
493 	TIMEVAL_TO_TIMESPEC(&tv, &vap->va_ctime);
494 	vap->va_atime = vap->va_mtime = vap->va_ctime;
495 
496 	/*
497 	 * If the process has exercised some setuid or setgid
498 	 * privilege, then rip away read/write permission so
499 	 * that only root can gain access.
500 	 */
501 	switch (pfs->pfs_type) {
502 	case Pmem:
503 	case Pregs:
504 	case Pfpregs:
505 		if (procp->p_flag & P_SUGID)
506 			vap->va_mode &= ~((VREAD|VWRITE)|
507 					  ((VREAD|VWRITE)>>3)|
508 					  ((VREAD|VWRITE)>>6));
509 		/* FALLTHROUGH */
510 	case Pctl:
511 	case Pstatus:
512 	case Pnote:
513 	case Pnotepg:
514 		vap->va_nlink = 1;
515 		vap->va_uid = procp->p_ucred->cr_uid;
516 		vap->va_gid = procp->p_ucred->cr_gid;
517 		break;
518 
519 	default:
520 		break;
521 	}
522 
523 	/*
524 	 * now do the object specific fields
525 	 *
526 	 * The size could be set from struct reg, but it's hardly
527 	 * worth the trouble, and it puts some (potentially) machine
528 	 * dependent data into this machine-independent code.  If it
529 	 * becomes important then this function should break out into
530 	 * a per-file stat function in the corresponding .c file.
531 	 */
532 
533 	switch (pfs->pfs_type) {
534 	case Proot:
535 		/*
536 		 * Set nlink to 1 to tell fts(3) we don't actually know.
537 		 */
538 		vap->va_nlink = 1;
539 		vap->va_uid = 0;
540 		vap->va_gid = 0;
541 		vap->va_bytes = vap->va_size = DEV_BSIZE;
542 		break;
543 
544 	case Pcurproc: {
545 		char buf[16];		/* should be enough */
546 		vap->va_nlink = 1;
547 		vap->va_uid = 0;
548 		vap->va_gid = 0;
549 		vap->va_bytes = vap->va_size =
550 		    sprintf(buf, "%ld", (long)curproc->p_pid);
551 		break;
552 	}
553 
554 	case Pproc:
555 		vap->va_nlink = 2;
556 		vap->va_uid = procp->p_ucred->cr_uid;
557 		vap->va_gid = procp->p_ucred->cr_gid;
558 		vap->va_bytes = vap->va_size = DEV_BSIZE;
559 		break;
560 
561 	case Pfile:
562 		error = EOPNOTSUPP;
563 		break;
564 
565 	case Pmem:
566 		vap->va_bytes = vap->va_size =
567 			ctob(procp->p_vmspace->vm_tsize +
568 				    procp->p_vmspace->vm_dsize +
569 				    procp->p_vmspace->vm_ssize);
570 		break;
571 
572 #if defined(PT_GETREGS) || defined(PT_SETREGS)
573 	case Pregs:
574 		vap->va_bytes = vap->va_size = sizeof(struct reg);
575 		break;
576 #endif
577 
578 #if defined(PT_GETFPREGS) || defined(PT_SETFPREGS)
579 	case Pfpregs:
580 		vap->va_bytes = vap->va_size = sizeof(struct fpreg);
581 		break;
582 #endif
583 
584 	case Pctl:
585 	case Pstatus:
586 	case Pnote:
587 	case Pnotepg:
588 		vap->va_bytes = vap->va_size = 0;
589 		break;
590 
591 	default:
592 		panic("procfs_getattr");
593 	}
594 
595 	return (error);
596 }
597 
598 /*ARGSUSED*/
599 int
600 procfs_setattr(v)
601 	void *v;
602 {
603 	/*
604 	 * just fake out attribute setting
605 	 * it's not good to generate an error
606 	 * return, otherwise things like creat()
607 	 * will fail when they try to set the
608 	 * file length to 0.  worse, this means
609 	 * that echo $note > /proc/$pid/note will fail.
610 	 */
611 
612 	return (0);
613 }
614 
615 /*
616  * implement access checking.
617  *
618  * actually, the check for super-user is slightly
619  * broken since it will allow read access to write-only
620  * objects.  this doesn't cause any particular trouble
621  * but does mean that the i/o entry points need to check
622  * that the operation really does make sense.
623  */
624 int
625 procfs_access(v)
626 	void *v;
627 {
628 	struct vop_access_args /* {
629 		struct vnode *a_vp;
630 		int a_mode;
631 		struct ucred *a_cred;
632 		struct proc *a_p;
633 	} */ *ap = v;
634 	struct vattr va;
635 	int error;
636 
637 	if ((error = VOP_GETATTR(ap->a_vp, &va, ap->a_cred, ap->a_p)) != 0)
638 		return (error);
639 
640 	return (vaccess(va.va_mode, va.va_uid, va.va_gid, ap->a_mode,
641 			ap->a_cred));
642 }
643 
644 /*
645  * lookup.  this is incredibly complicated in the
646  * general case, however for most pseudo-filesystems
647  * very little needs to be done.
648  *
649  * unless you want to get a migraine, just make sure your
650  * filesystem doesn't do any locking of its own.  otherwise
651  * read and inwardly digest ufs_lookup().
652  */
653 int
654 procfs_lookup(v)
655 	void *v;
656 {
657 	struct vop_lookup_args /* {
658 		struct vnode * a_dvp;
659 		struct vnode ** a_vpp;
660 		struct componentname * a_cnp;
661 	} */ *ap = v;
662 	struct componentname *cnp = ap->a_cnp;
663 	struct vnode **vpp = ap->a_vpp;
664 	struct vnode *dvp = ap->a_dvp;
665 	const char *pname = cnp->cn_nameptr;
666 	struct proc_target *pt;
667 	struct vnode *fvp;
668 	pid_t pid;
669 	struct pfsnode *pfs;
670 	struct proc *p;
671 	int i;
672 
673 	*vpp = NULL;
674 
675 	if (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)
676 		return (EROFS);
677 
678 	if (cnp->cn_namelen == 1 && *pname == '.') {
679 		*vpp = dvp;
680 		VREF(dvp);
681 		/*VOP_LOCK(dvp);*/
682 		return (0);
683 	}
684 
685 	pfs = VTOPFS(dvp);
686 	switch (pfs->pfs_type) {
687 	case Proot:
688 		if (cnp->cn_flags & ISDOTDOT)
689 			return (EIO);
690 
691 		if (CNEQ(cnp, "curproc", 7))
692 			return (procfs_allocvp(dvp->v_mount, vpp, 0, Pcurproc));
693 
694 		pid = atopid(pname, cnp->cn_namelen);
695 		if (pid == NO_PID)
696 			break;
697 
698 		p = PFIND(pid);
699 		if (p == 0)
700 			break;
701 
702 		return (procfs_allocvp(dvp->v_mount, vpp, pid, Pproc));
703 
704 	case Pproc:
705 		if (cnp->cn_flags & ISDOTDOT)
706 			return (procfs_root(dvp->v_mount, vpp));
707 
708 		p = PFIND(pfs->pfs_pid);
709 		if (p == 0)
710 			break;
711 
712 		for (pt = proc_targets, i = 0; i < nproc_targets; pt++, i++) {
713 			if (cnp->cn_namelen == pt->pt_namlen &&
714 			    bcmp(pt->pt_name, pname, cnp->cn_namelen) == 0 &&
715 			    (pt->pt_valid == NULL || (*pt->pt_valid)(p)))
716 				goto found;
717 		}
718 		break;
719 
720 	found:
721 		if (pt->pt_pfstype == Pfile) {
722 			fvp = procfs_findtextvp(p);
723 			/* We already checked that it exists. */
724 			VREF(fvp);
725 			VOP_LOCK(fvp);
726 			*vpp = fvp;
727 			return (0);
728 		}
729 
730 		return (procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
731 		    pt->pt_pfstype));
732 
733 	default:
734 		return (ENOTDIR);
735 	}
736 
737 	return (cnp->cn_nameiop == LOOKUP ? ENOENT : EROFS);
738 }
739 
740 int
741 procfs_validfile(p)
742 	struct proc *p;
743 {
744 
745 	return (procfs_findtextvp(p) != NULLVP);
746 }
747 
748 /*
749  * readdir returns directory entries from pfsnode (vp).
750  *
751  * the strategy here with procfs is to generate a single
752  * directory entry at a time (struct dirent) and then
753  * copy that out to userland using uiomove.  a more efficent
754  * though more complex implementation, would try to minimize
755  * the number of calls to uiomove().  for procfs, this is
756  * hardly worth the added code complexity.
757  *
758  * this should just be done through read()
759  */
760 int
761 procfs_readdir(v)
762 	void *v;
763 {
764 	struct vop_readdir_args /* {
765 		struct vnode *a_vp;
766 		struct uio *a_uio;
767 		struct ucred *a_cred;
768 		int *a_eofflag;
769 		u_long *a_cookies;
770 		int a_ncookies;
771 	} */ *ap = v;
772 	struct uio *uio = ap->a_uio;
773 	struct dirent d;
774 	struct pfsnode *pfs;
775 	int i;
776 	int error;
777 	u_long *cookies = ap->a_cookies;
778 	int ncookies = ap->a_ncookies;
779 
780 	pfs = VTOPFS(ap->a_vp);
781 
782 	if (uio->uio_resid < UIO_MX)
783 		return (EINVAL);
784 	if (uio->uio_offset < 0)
785 		return (EINVAL);
786 
787 	error = 0;
788 	i = uio->uio_offset;
789 	bzero((caddr_t)&d, UIO_MX);
790 	d.d_reclen = UIO_MX;
791 
792 	switch (pfs->pfs_type) {
793 	/*
794 	 * this is for the process-specific sub-directories.
795 	 * all that is needed to is copy out all the entries
796 	 * from the procent[] table (top of this file).
797 	 */
798 	case Pproc: {
799 		struct proc *p;
800 		struct proc_target *pt;
801 
802 		p = PFIND(pfs->pfs_pid);
803 		if (p == NULL)
804 			break;
805 
806 		for (pt = &proc_targets[i];
807 		     uio->uio_resid >= UIO_MX && i < nproc_targets; pt++, i++) {
808 			if (pt->pt_valid && (*pt->pt_valid)(p) == 0)
809 				continue;
810 
811 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid, pt->pt_pfstype);
812 			d.d_namlen = pt->pt_namlen;
813 			bcopy(pt->pt_name, d.d_name, pt->pt_namlen + 1);
814 			d.d_type = pt->pt_type;
815 
816 			if ((error = uiomove((caddr_t)&d, UIO_MX, uio)) != 0)
817 				break;
818 			if (ncookies-- > 0)
819 				*cookies++ = i + 1;
820 		}
821 
822 	    	break;
823 	}
824 
825 	/*
826 	 * this is for the root of the procfs filesystem
827 	 * what is needed is a special entry for "curproc"
828 	 * followed by an entry for each process on allproc
829 #ifdef PROCFS_ZOMBIE
830 	 * and zombproc.
831 #endif
832 	 */
833 
834 	case Proot: {
835 #ifdef PROCFS_ZOMBIE
836 		int doingzomb = 0;
837 #endif
838 		int pcnt = i;
839 		volatile struct proc *p = allproc.lh_first;
840 
841 		if (pcnt > 3)
842 			pcnt = 3;
843 #ifdef PROCFS_ZOMBIE
844 	again:
845 #endif
846 		for (; p && uio->uio_resid >= UIO_MX; i++, pcnt++) {
847 			switch (i) {
848 			case 0:		/* `.' */
849 			case 1:		/* `..' */
850 				d.d_fileno = PROCFS_FILENO(0, Proot);
851 				d.d_namlen = i + 1;
852 				bcopy("..", d.d_name, d.d_namlen);
853 				d.d_name[i + 1] = '\0';
854 				d.d_type = DT_DIR;
855 				break;
856 
857 			case 2:
858 				d.d_fileno = PROCFS_FILENO(0, Pcurproc);
859 				d.d_namlen = 7;
860 				bcopy("curproc", d.d_name, 8);
861 				d.d_type = DT_LNK;
862 				break;
863 
864 			default:
865 				while (pcnt < i) {
866 					pcnt++;
867 					p = p->p_list.le_next;
868 					if (!p)
869 						goto done;
870 				}
871 				d.d_fileno = PROCFS_FILENO(p->p_pid, Pproc);
872 				d.d_namlen = sprintf(d.d_name, "%ld",
873 				    (long)p->p_pid);
874 				d.d_type = DT_REG;
875 				p = p->p_list.le_next;
876 				break;
877 			}
878 
879 			if ((error = uiomove((caddr_t)&d, UIO_MX, uio)) != 0)
880 				break;
881 			if (ncookies-- > 0)
882 				*cookies++ = i + 1;
883 		}
884 	done:
885 
886 #ifdef PROCFS_ZOMBIE
887 		if (p == 0 && doingzomb == 0) {
888 			doingzomb = 1;
889 			p = zombproc.lh_first;
890 			goto again;
891 		}
892 #endif
893 
894 		break;
895 
896 	}
897 
898 	default:
899 		error = ENOTDIR;
900 		break;
901 	}
902 
903 	uio->uio_offset = i;
904 	return (error);
905 }
906 
907 /*
908  * readlink reads the link of `curproc'
909  */
910 int
911 procfs_readlink(v)
912 	void *v;
913 {
914 	struct vop_readlink_args *ap = v;
915 	char buf[16];		/* should be enough */
916 	int len;
917 
918 	if (VTOPFS(ap->a_vp)->pfs_fileno != PROCFS_FILENO(0, Pcurproc))
919 		return (EINVAL);
920 
921 	len = sprintf(buf, "%ld", (long)curproc->p_pid);
922 
923 	return (uiomove((caddr_t)buf, len, ap->a_uio));
924 }
925 
926 /*
927  * convert decimal ascii to pid_t
928  */
929 static pid_t
930 atopid(b, len)
931 	const char *b;
932 	u_int len;
933 {
934 	pid_t p = 0;
935 
936 	while (len--) {
937 		char c = *b++;
938 		if (c < '0' || c > '9')
939 			return (NO_PID);
940 		p = 10 * p + (c - '0');
941 		if (p > PID_MAX)
942 			return (NO_PID);
943 	}
944 
945 	return (p);
946 }
947