xref: /netbsd-src/sys/miscfs/procfs/procfs_vnops.c (revision 1f2744e6e4915c9da2a3f980279398c4cf7d5e6d)
1 /*	$NetBSD: procfs_vnops.c,v 1.32 1995/02/03 16:18:55 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 #include <miscfs/procfs/procfs.h>
61 
62 /*
63  * Vnode Operations.
64  *
65  */
66 
67 /*
68  * This is a list of the valid names in the
69  * process-specific sub-directories.  It is
70  * used in procfs_lookup and procfs_readdir
71  */
72 struct proc_target {
73 	u_char	pt_type;
74 	u_char	pt_namlen;
75 	char	*pt_name;
76 	pfstype	pt_pfstype;
77 	int	(*pt_valid) __P((struct proc *p));
78 } proc_targets[] = {
79 #define N(s) sizeof(s)-1, s
80 	/*	  name		type		validp */
81 	{ DT_DIR, N("."),	Pproc,		NULL },
82 	{ DT_DIR, N(".."),	Proot,		NULL },
83 	{ DT_REG, N("file"),	Pfile,		procfs_validfile },
84 	{ DT_REG, N("mem"),	Pmem,		NULL },
85 	{ DT_REG, N("regs"),	Pregs,		procfs_validregs },
86 	{ DT_REG, N("fpregs"),	Pfpregs,	procfs_validfpregs },
87 	{ DT_REG, N("ctl"),	Pctl,		NULL },
88 	{ DT_REG, N("status"),	Pstatus,	NULL },
89 	{ DT_REG, N("note"),	Pnote,		NULL },
90 	{ DT_REG, N("notepg"),	Pnotepg,	NULL },
91 #undef N
92 };
93 static int nproc_targets = sizeof(proc_targets) / sizeof(proc_targets[0]);
94 
95 static pid_t atopid __P((const char *, u_int));
96 
97 /*
98  * set things up for doing i/o on
99  * the pfsnode (vp).  (vp) is locked
100  * on entry, and should be left locked
101  * on exit.
102  *
103  * for procfs we don't need to do anything
104  * in particular for i/o.  all that is done
105  * is to support exclusive open on process
106  * memory images.
107  */
108 procfs_open(ap)
109 	struct vop_open_args /* {
110 		struct vnode *a_vp;
111 		int  a_mode;
112 		struct ucred *a_cred;
113 		struct proc *a_p;
114 	} */ *ap;
115 {
116 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
117 
118 	switch (pfs->pfs_type) {
119 	case Pmem:
120 		if (PFIND(pfs->pfs_pid) == 0)
121 			return (ENOENT);	/* was ESRCH, jsp */
122 
123 		if ((pfs->pfs_flags & FWRITE) && (ap->a_mode & O_EXCL) ||
124 		    (pfs->pfs_flags & O_EXCL) && (ap->a_mode & FWRITE))
125 			return (EBUSY);
126 
127 		if (ap->a_mode & FWRITE)
128 			pfs->pfs_flags = ap->a_mode & (FWRITE|O_EXCL);
129 
130 		return (0);
131 
132 	default:
133 		break;
134 	}
135 
136 	return (0);
137 }
138 
139 /*
140  * close the pfsnode (vp) after doing i/o.
141  * (vp) is not locked on entry or exit.
142  *
143  * nothing to do for procfs other than undo
144  * any exclusive open flag (see _open above).
145  */
146 procfs_close(ap)
147 	struct vop_close_args /* {
148 		struct vnode *a_vp;
149 		int  a_fflag;
150 		struct ucred *a_cred;
151 		struct proc *a_p;
152 	} */ *ap;
153 {
154 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
155 
156 	switch (pfs->pfs_type) {
157 	case Pmem:
158 		if ((ap->a_fflag & FWRITE) && (pfs->pfs_flags & O_EXCL))
159 			pfs->pfs_flags &= ~(FWRITE|O_EXCL);
160 		break;
161 	}
162 
163 	return (0);
164 }
165 
166 /*
167  * do an ioctl operation on pfsnode (vp).
168  * (vp) is not locked on entry or exit.
169  */
170 procfs_ioctl(ap)
171 	struct vop_ioctl_args /* {
172 		struct vnode *a_vp;
173 		u_long a_command;
174 		caddr_t a_data;
175 		int a_fflag;
176 		struct ucred *a_cred;
177 		struct proc *a_p;
178 	} */ *ap;
179 {
180 
181 	return (ENOTTY);
182 }
183 
184 /*
185  * do block mapping for pfsnode (vp).
186  * since we don't use the buffer cache
187  * for procfs this function should never
188  * be called.  in any case, it's not clear
189  * what part of the kernel ever makes use
190  * of this function.  for sanity, this is the
191  * usual no-op bmap, although returning
192  * (EIO) would be a reasonable alternative.
193  */
194 procfs_bmap(ap)
195 	struct vop_bmap_args /* {
196 		struct vnode *a_vp;
197 		daddr_t  a_bn;
198 		struct vnode **a_vpp;
199 		daddr_t *a_bnp;
200 	} */ *ap;
201 {
202 
203 	if (ap->a_vpp != NULL)
204 		*ap->a_vpp = ap->a_vp;
205 	if (ap->a_bnp != NULL)
206 		*ap->a_bnp = ap->a_bn;
207 	return (0);
208 }
209 
210 /*
211  * _inactive is called when the pfsnode
212  * is vrele'd and the reference count goes
213  * to zero.  (vp) will be on the vnode free
214  * list, so to get it back vget() must be
215  * used.
216  *
217  * for procfs, check if the process is still
218  * alive and if it isn't then just throw away
219  * the vnode by calling vgone().  this may
220  * be overkill and a waste of time since the
221  * chances are that the process will still be
222  * there and PFIND is not free.
223  *
224  * (vp) is not locked on entry or exit.
225  */
226 procfs_inactive(ap)
227 	struct vop_inactive_args /* {
228 		struct vnode *a_vp;
229 	} */ *ap;
230 {
231 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
232 
233 	if (PFIND(pfs->pfs_pid) == 0)
234 		vgone(ap->a_vp);
235 
236 	return (0);
237 }
238 
239 /*
240  * _reclaim is called when getnewvnode()
241  * wants to make use of an entry on the vnode
242  * free list.  at this time the filesystem needs
243  * to free any private data and remove the node
244  * from any private lists.
245  */
246 procfs_reclaim(ap)
247 	struct vop_reclaim_args /* {
248 		struct vnode *a_vp;
249 	} */ *ap;
250 {
251 
252 	return (procfs_freevp(ap->a_vp));
253 }
254 
255 /*
256  * Return POSIX pathconf information applicable to special devices.
257  */
258 procfs_pathconf(ap)
259 	struct vop_pathconf_args /* {
260 		struct vnode *a_vp;
261 		int a_name;
262 		register_t *a_retval;
263 	} */ *ap;
264 {
265 
266 	switch (ap->a_name) {
267 	case _PC_LINK_MAX:
268 		*ap->a_retval = LINK_MAX;
269 		return (0);
270 	case _PC_MAX_CANON:
271 		*ap->a_retval = MAX_CANON;
272 		return (0);
273 	case _PC_MAX_INPUT:
274 		*ap->a_retval = MAX_INPUT;
275 		return (0);
276 	case _PC_PIPE_BUF:
277 		*ap->a_retval = PIPE_BUF;
278 		return (0);
279 	case _PC_CHOWN_RESTRICTED:
280 		*ap->a_retval = 1;
281 		return (0);
282 	case _PC_VDISABLE:
283 		*ap->a_retval = _POSIX_VDISABLE;
284 		return (0);
285 	default:
286 		return (EINVAL);
287 	}
288 	/* NOTREACHED */
289 }
290 
291 /*
292  * _print is used for debugging.
293  * just print a readable description
294  * of (vp).
295  */
296 procfs_print(ap)
297 	struct vop_print_args /* {
298 		struct vnode *a_vp;
299 	} */ *ap;
300 {
301 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
302 
303 	printf("tag VT_PROCFS, type %s, pid %d, mode %x, flags %x\n",
304 	    pfs->pfs_type, pfs->pfs_pid, pfs->pfs_mode, pfs->pfs_flags);
305 }
306 
307 /*
308  * _abortop is called when operations such as
309  * rename and create fail.  this entry is responsible
310  * for undoing any side-effects caused by the lookup.
311  * this will always include freeing the pathname buffer.
312  */
313 procfs_abortop(ap)
314 	struct vop_abortop_args /* {
315 		struct vnode *a_dvp;
316 		struct componentname *a_cnp;
317 	} */ *ap;
318 {
319 
320 	if ((ap->a_cnp->cn_flags & (HASBUF | SAVESTART)) == HASBUF)
321 		FREE(ap->a_cnp->cn_pnbuf, M_NAMEI);
322 	return (0);
323 }
324 
325 /*
326  * generic entry point for unsupported operations
327  */
328 procfs_badop()
329 {
330 
331 	return (EIO);
332 }
333 
334 /*
335  * Invent attributes for pfsnode (vp) and store
336  * them in (vap).
337  * Directories lengths are returned as zero since
338  * any real length would require the genuine size
339  * to be computed, and nothing cares anyway.
340  *
341  * this is relatively minimal for procfs.
342  */
343 procfs_getattr(ap)
344 	struct vop_getattr_args /* {
345 		struct vnode *a_vp;
346 		struct vattr *a_vap;
347 		struct ucred *a_cred;
348 		struct proc *a_p;
349 	} */ *ap;
350 {
351 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
352 	struct vattr *vap = ap->a_vap;
353 	struct proc *procp;
354 	int error;
355 
356 	/* first check the process still exists */
357 	switch (pfs->pfs_type) {
358 	case Proot:
359 	case Pcurproc:
360 		procp = 0;
361 		break;
362 
363 	default:
364 		procp = PFIND(pfs->pfs_pid);
365 		if (procp == 0)
366 			return (ENOENT);
367 	}
368 
369 	error = 0;
370 
371 	/* start by zeroing out the attributes */
372 	VATTR_NULL(vap);
373 
374 	/* next do all the common fields */
375 	vap->va_type = ap->a_vp->v_type;
376 	vap->va_mode = pfs->pfs_mode;
377 	vap->va_fileid = pfs->pfs_fileno;
378 	vap->va_flags = 0;
379 	vap->va_blocksize = PAGE_SIZE;
380 	vap->va_bytes = vap->va_size = 0;
381 
382 	/*
383 	 * Make all times be current TOD.
384 	 * It would be possible to get the process start
385 	 * time from the p_stat structure, but there's
386 	 * no "file creation" time stamp anyway, and the
387 	 * p_stat structure is not addressible if u. gets
388 	 * swapped out for that process.
389 	 *
390 	 * XXX
391 	 * Note that microtime() returns a timeval, not a timespec.
392 	 */
393 	microtime(&vap->va_ctime);
394 	vap->va_atime = vap->va_mtime = vap->va_ctime;
395 
396 	/*
397 	 * If the process has exercised some setuid or setgid
398 	 * privilege, then rip away read/write permission so
399 	 * that only root can gain access.
400 	 */
401 	switch (pfs->pfs_type) {
402 	case Pmem:
403 	case Pregs:
404 	case Pfpregs:
405 		if (procp->p_flag & P_SUGID)
406 			vap->va_mode &= ~((VREAD|VWRITE)|
407 					  ((VREAD|VWRITE)>>3)|
408 					  ((VREAD|VWRITE)>>6));
409 	case Pctl:
410 	case Pstatus:
411 	case Pnote:
412 	case Pnotepg:
413 		vap->va_nlink = 1;
414 		vap->va_uid = procp->p_ucred->cr_uid;
415 		vap->va_gid = procp->p_ucred->cr_gid;
416 		break;
417 	}
418 
419 	/*
420 	 * now do the object specific fields
421 	 *
422 	 * The size could be set from struct reg, but it's hardly
423 	 * worth the trouble, and it puts some (potentially) machine
424 	 * dependent data into this machine-independent code.  If it
425 	 * becomes important then this function should break out into
426 	 * a per-file stat function in the corresponding .c file.
427 	 */
428 
429 	switch (pfs->pfs_type) {
430 	case Proot:
431 		/*
432 		 * Set nlink to 1 to tell fts(3) we don't actually know.
433 		 */
434 		vap->va_nlink = 1;
435 		vap->va_uid = 0;
436 		vap->va_gid = 0;
437 		vap->va_size = vap->va_bytes = DEV_BSIZE;
438 		break;
439 
440 	case Pcurproc: {
441 		char buf[16];		/* should be enough */
442 		vap->va_nlink = 1;
443 		vap->va_uid = 0;
444 		vap->va_gid = 0;
445 		vap->va_size = vap->va_bytes =
446 		    sprintf(buf, "%ld", (long)curproc->p_pid);
447 		break;
448 	}
449 
450 	case Pproc:
451 		vap->va_nlink = 2;
452 		vap->va_uid = procp->p_ucred->cr_uid;
453 		vap->va_gid = procp->p_ucred->cr_gid;
454 		vap->va_size = vap->va_bytes = DEV_BSIZE;
455 		break;
456 
457 	case Pfile:
458 		error = EOPNOTSUPP;
459 		break;
460 
461 	case Pmem:
462 		vap->va_bytes = vap->va_size =
463 			ctob(procp->p_vmspace->vm_tsize +
464 				    procp->p_vmspace->vm_dsize +
465 				    procp->p_vmspace->vm_ssize);
466 		break;
467 
468 #if defined(PT_GETREGS) || defined(PT_SETREGS)
469 	case Pregs:
470 		vap->va_bytes = vap->va_size = sizeof(struct reg);
471 		break;
472 #endif
473 
474 #if defined(PT_GETFPREGS) || defined(PT_SETFPREGS)
475 	case Pfpregs:
476 		vap->va_bytes = vap->va_size = sizeof(struct fpreg);
477 		break;
478 #endif
479 
480 	case Pctl:
481 	case Pstatus:
482 	case Pnote:
483 	case Pnotepg:
484 		break;
485 
486 	default:
487 		panic("procfs_getattr");
488 	}
489 
490 	return (error);
491 }
492 
493 procfs_setattr(ap)
494 	struct vop_setattr_args /* {
495 		struct vnode *a_vp;
496 		struct vattr *a_vap;
497 		struct ucred *a_cred;
498 		struct proc *a_p;
499 	} */ *ap;
500 {
501 	/*
502 	 * just fake out attribute setting
503 	 * it's not good to generate an error
504 	 * return, otherwise things like creat()
505 	 * will fail when they try to set the
506 	 * file length to 0.  worse, this means
507 	 * that echo $note > /proc/$pid/note will fail.
508 	 */
509 
510 	return (0);
511 }
512 
513 /*
514  * implement access checking.
515  *
516  * actually, the check for super-user is slightly
517  * broken since it will allow read access to write-only
518  * objects.  this doesn't cause any particular trouble
519  * but does mean that the i/o entry points need to check
520  * that the operation really does make sense.
521  */
522 procfs_access(ap)
523 	struct vop_access_args /* {
524 		struct vnode *a_vp;
525 		int a_mode;
526 		struct ucred *a_cred;
527 		struct proc *a_p;
528 	} */ *ap;
529 {
530 	struct vattr va;
531 	int error;
532 
533 	if (error = VOP_GETATTR(ap->a_vp, &va, ap->a_cred, ap->a_p))
534 		return (error);
535 
536 	return (vaccess(va.va_mode, va.va_uid, va.va_gid, ap->a_mode,
537 	    ap->a_cred));
538 }
539 
540 /*
541  * lookup.  this is incredibly complicated in the
542  * general case, however for most pseudo-filesystems
543  * very little needs to be done.
544  *
545  * unless you want to get a migraine, just make sure your
546  * filesystem doesn't do any locking of its own.  otherwise
547  * read and inwardly digest ufs_lookup().
548  */
549 procfs_lookup(ap)
550 	struct vop_lookup_args /* {
551 		struct vnode * a_dvp;
552 		struct vnode ** a_vpp;
553 		struct componentname * a_cnp;
554 	} */ *ap;
555 {
556 	struct componentname *cnp = ap->a_cnp;
557 	struct vnode **vpp = ap->a_vpp;
558 	struct vnode *dvp = ap->a_dvp;
559 	char *pname = cnp->cn_nameptr;
560 	struct proc_target *pt;
561 	struct vnode *fvp;
562 	pid_t pid;
563 	struct pfsnode *pfs;
564 	struct proc *p;
565 	int i;
566 
567 	*vpp = NULL;
568 
569 	if (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)
570 		return (EROFS);
571 
572 	if (cnp->cn_namelen == 1 && *pname == '.') {
573 		*vpp = dvp;
574 		VREF(dvp);
575 		/*VOP_LOCK(dvp);*/
576 		return (0);
577 	}
578 
579 	pfs = VTOPFS(dvp);
580 	switch (pfs->pfs_type) {
581 	case Proot:
582 		if (cnp->cn_flags & ISDOTDOT)
583 			return (EIO);
584 
585 		if (CNEQ(cnp, "curproc", 7))
586 			return (procfs_allocvp(dvp->v_mount, vpp, 0, Pcurproc));
587 
588 		pid = atopid(pname, cnp->cn_namelen);
589 		if (pid == NO_PID)
590 			break;
591 
592 		p = PFIND(pid);
593 		if (p == 0)
594 			break;
595 
596 		return (procfs_allocvp(dvp->v_mount, vpp, pid, Pproc));
597 
598 	case Pproc:
599 		if (cnp->cn_flags & ISDOTDOT)
600 			return (procfs_root(dvp->v_mount, vpp));
601 
602 		p = PFIND(pfs->pfs_pid);
603 		if (p == 0)
604 			break;
605 
606 		for (pt = proc_targets, i = 0; i < nproc_targets; pt++, i++) {
607 			if (cnp->cn_namelen == pt->pt_namlen &&
608 			    bcmp(pt->pt_name, pname, cnp->cn_namelen) == 0 &&
609 			    (pt->pt_valid == NULL || (*pt->pt_valid)(p)))
610 				goto found;
611 		}
612 		break;
613 
614 	found:
615 		if (pt->pt_pfstype == Pfile) {
616 			fvp = procfs_findtextvp(p);
617 			/* We already checked that it exists. */
618 			VREF(fvp);
619 			VOP_LOCK(fvp);
620 			*vpp = fvp;
621 			return (0);
622 		}
623 
624 		return (procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
625 		    pt->pt_pfstype));
626 
627 	default:
628 		return (ENOTDIR);
629 	}
630 
631 	return (cnp->cn_nameiop == LOOKUP ? ENOENT : EROFS);
632 }
633 
634 int
635 procfs_validfile(p)
636 	struct proc *p;
637 {
638 
639 	return (procfs_findtextvp(p) != NULLVP);
640 }
641 
642 /*
643  * readdir returns directory entries from pfsnode (vp).
644  *
645  * the strategy here with procfs is to generate a single
646  * directory entry at a time (struct pfsdent) and then
647  * copy that out to userland using uiomove.  a more efficent
648  * though more complex implementation, would try to minimize
649  * the number of calls to uiomove().  for procfs, this is
650  * hardly worth the added code complexity.
651  *
652  * this should just be done through read()
653  */
654 procfs_readdir(ap)
655 	struct vop_readdir_args /* {
656 		struct vnode *a_vp;
657 		struct uio *a_uio;
658 		struct ucred *a_cred;
659 		int *a_eofflag;
660 		u_long *a_cookies;
661 		int a_ncookies;
662 	} */ *ap;
663 {
664 	struct uio *uio = ap->a_uio;
665 	struct pfsdent d;
666 	struct pfsdent *dp = &d;
667 	struct pfsnode *pfs;
668 	int error;
669 	int count;
670 	int i;
671 
672 	/*
673 	 * We don't allow exporting procfs mounts, and currently local
674 	 * requests do not need cookies.
675 	 */
676 	if (ap->a_ncookies)
677 		panic("procfs_readdir: not hungry");
678 
679 	pfs = VTOPFS(ap->a_vp);
680 
681 	if (uio->uio_resid < UIO_MX)
682 		return (EINVAL);
683 	if (uio->uio_offset & (UIO_MX-1))
684 		return (EINVAL);
685 	if (uio->uio_offset < 0)
686 		return (EINVAL);
687 
688 	error = 0;
689 	count = 0;
690 	i = uio->uio_offset / UIO_MX;
691 
692 	switch (pfs->pfs_type) {
693 	/*
694 	 * this is for the process-specific sub-directories.
695 	 * all that is needed to is copy out all the entries
696 	 * from the procent[] table (top of this file).
697 	 */
698 	case Pproc: {
699 		struct proc *p;
700 		struct proc_target *pt;
701 
702 		p = PFIND(pfs->pfs_pid);
703 		if (p == NULL)
704 			break;
705 
706 		for (pt = &proc_targets[i];
707 		     uio->uio_resid >= UIO_MX && i < nproc_targets; pt++, i++) {
708 			if (pt->pt_valid && (*pt->pt_valid)(p) == 0)
709 				continue;
710 
711 			dp->d_reclen = UIO_MX;
712 			dp->d_fileno = PROCFS_FILENO(pfs->pfs_pid, pt->pt_pfstype);
713 			dp->d_namlen = pt->pt_namlen;
714 			bcopy(pt->pt_name, dp->d_name, pt->pt_namlen + 1);
715 			dp->d_type = pt->pt_type;
716 
717 			if (error = uiomove((caddr_t)dp, UIO_MX, uio))
718 				break;
719 		}
720 
721 	    	break;
722 	    }
723 
724 	/*
725 	 * this is for the root of the procfs filesystem
726 	 * what is needed is a special entry for "curproc"
727 	 * followed by an entry for each process on allproc
728 #ifdef PROCFS_ZOMBIE
729 	 * and zombproc.
730 #endif
731 	 */
732 
733 	case Proot: {
734 #ifdef PROCFS_ZOMBIE
735 		int doingzomb = 0;
736 #endif
737 		int pcnt = 0;
738 		volatile struct proc *p = allproc.lh_first;
739 
740 	again:
741 		for (; p && uio->uio_resid >= UIO_MX; i++, pcnt++) {
742 			bzero((char *) dp, UIO_MX);
743 			dp->d_reclen = UIO_MX;
744 
745 			switch (i) {
746 			case 0:		/* `.' */
747 			case 1:		/* `..' */
748 				dp->d_fileno = PROCFS_FILENO(0, Proot);
749 				dp->d_namlen = i + 1;
750 				bcopy("..", dp->d_name, dp->d_namlen);
751 				dp->d_name[i + 1] = '\0';
752 				dp->d_type = DT_DIR;
753 				break;
754 
755 			case 2:
756 				dp->d_fileno = PROCFS_FILENO(0, Pcurproc);
757 				dp->d_namlen = 7;
758 				bcopy("curproc", dp->d_name, 8);
759 				dp->d_type = DT_LNK;
760 				break;
761 
762 			default:
763 				while (pcnt < i) {
764 					pcnt++;
765 					p = p->p_list.le_next;
766 					if (!p)
767 						goto done;
768 				}
769 				dp->d_fileno = PROCFS_FILENO(p->p_pid, Pproc);
770 				dp->d_namlen = sprintf(dp->d_name, "%ld",
771 				    (long)p->p_pid);
772 				dp->d_type = DT_REG;
773 				p = p->p_list.le_next;
774 				break;
775 			}
776 
777 			if (error = uiomove((caddr_t)dp, UIO_MX, uio))
778 				break;
779 		}
780 	done:
781 
782 #ifdef PROCFS_ZOMBIE
783 		if (p == 0 && doingzomb == 0) {
784 			doingzomb = 1;
785 			p = zombproc.lh_first;
786 			goto again;
787 		}
788 #endif
789 
790 		break;
791 
792 	    }
793 
794 	default:
795 		error = ENOTDIR;
796 		break;
797 	}
798 
799 	uio->uio_offset = i * UIO_MX;
800 
801 	return (error);
802 }
803 
804 /*
805  * readlink reads the link of `curproc'
806  */
807 procfs_readlink(ap)
808 	struct vop_readlink_args *ap;
809 {
810 	struct uio *uio = ap->a_uio;
811 	char buf[16];		/* should be enough */
812 	int len;
813 
814 	if (VTOPFS(ap->a_vp)->pfs_fileno != PROCFS_FILENO(0, Pcurproc))
815 		return (EINVAL);
816 
817 	len = sprintf(buf, "%ld", (long)curproc->p_pid);
818 
819 	return (uiomove((caddr_t)buf, len, ap->a_uio));
820 }
821 
822 /*
823  * convert decimal ascii to pid_t
824  */
825 static pid_t
826 atopid(b, len)
827 	const char *b;
828 	u_int len;
829 {
830 	pid_t p = 0;
831 
832 	while (len--) {
833 		char c = *b++;
834 		if (c < '0' || c > '9')
835 			return (NO_PID);
836 		p = 10 * p + (c - '0');
837 		if (p > PID_MAX)
838 			return (NO_PID);
839 	}
840 
841 	return (p);
842 }
843 
844 /*
845  * procfs vnode operations.
846  */
847 int (**procfs_vnodeop_p)();
848 struct vnodeopv_entry_desc procfs_vnodeop_entries[] = {
849 	{ &vop_default_desc, vn_default_error },
850 	{ &vop_lookup_desc, procfs_lookup },		/* lookup */
851 	{ &vop_create_desc, procfs_create },		/* create */
852 	{ &vop_mknod_desc, procfs_mknod },		/* mknod */
853 	{ &vop_open_desc, procfs_open },		/* open */
854 	{ &vop_close_desc, procfs_close },		/* close */
855 	{ &vop_access_desc, procfs_access },		/* access */
856 	{ &vop_getattr_desc, procfs_getattr },		/* getattr */
857 	{ &vop_setattr_desc, procfs_setattr },		/* setattr */
858 	{ &vop_read_desc, procfs_read },		/* read */
859 	{ &vop_write_desc, procfs_write },		/* write */
860 	{ &vop_ioctl_desc, procfs_ioctl },		/* ioctl */
861 	{ &vop_select_desc, procfs_select },		/* select */
862 	{ &vop_mmap_desc, procfs_mmap },		/* mmap */
863 	{ &vop_fsync_desc, procfs_fsync },		/* fsync */
864 	{ &vop_seek_desc, procfs_seek },		/* seek */
865 	{ &vop_remove_desc, procfs_remove },		/* remove */
866 	{ &vop_link_desc, procfs_link },		/* link */
867 	{ &vop_rename_desc, procfs_rename },		/* rename */
868 	{ &vop_mkdir_desc, procfs_mkdir },		/* mkdir */
869 	{ &vop_rmdir_desc, procfs_rmdir },		/* rmdir */
870 	{ &vop_symlink_desc, procfs_symlink },		/* symlink */
871 	{ &vop_readdir_desc, procfs_readdir },		/* readdir */
872 	{ &vop_readlink_desc, procfs_readlink },	/* readlink */
873 	{ &vop_abortop_desc, procfs_abortop },		/* abortop */
874 	{ &vop_inactive_desc, procfs_inactive },	/* inactive */
875 	{ &vop_reclaim_desc, procfs_reclaim },		/* reclaim */
876 	{ &vop_lock_desc, procfs_lock },		/* lock */
877 	{ &vop_unlock_desc, procfs_unlock },		/* unlock */
878 	{ &vop_bmap_desc, procfs_bmap },		/* bmap */
879 	{ &vop_strategy_desc, procfs_strategy },	/* strategy */
880 	{ &vop_print_desc, procfs_print },		/* print */
881 	{ &vop_islocked_desc, procfs_islocked },	/* islocked */
882 	{ &vop_pathconf_desc, procfs_pathconf },	/* pathconf */
883 	{ &vop_advlock_desc, procfs_advlock },		/* advlock */
884 	{ &vop_blkatoff_desc, procfs_blkatoff },	/* blkatoff */
885 	{ &vop_valloc_desc, procfs_valloc },		/* valloc */
886 	{ &vop_vfree_desc, procfs_vfree },		/* vfree */
887 	{ &vop_truncate_desc, procfs_truncate },	/* truncate */
888 	{ &vop_update_desc, procfs_update },		/* update */
889 	{ (struct vnodeop_desc*)NULL, (int(*)())NULL }
890 };
891 struct vnodeopv_desc procfs_vnodeop_opv_desc =
892 	{ &procfs_vnodeop_p, procfs_vnodeop_entries };
893