xref: /dflybsd-src/sys/vfs/procfs/procfs_vnops.c (revision 48d201a5a8c1dab4aa7166b0812594c101fc43c3)
1 /*
2  * Copyright (c) 1993, 1995 Jan-Simon Pendry
3  * Copyright (c) 1993, 1995
4  *	The Regents of the University of California.  All rights reserved.
5  *
6  * This code is derived from software contributed to Berkeley by
7  * Jan-Simon Pendry.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  * 3. All advertising materials mentioning features or use of this software
18  *    must display the following acknowledgement:
19  *	This product includes software developed by the University of
20  *	California, Berkeley and its contributors.
21  * 4. Neither the name of the University nor the names of its contributors
22  *    may be used to endorse or promote products derived from this software
23  *    without specific prior written permission.
24  *
25  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
26  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
29  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35  * SUCH DAMAGE.
36  *
37  *	@(#)procfs_vnops.c	8.18 (Berkeley) 5/21/95
38  *
39  * $FreeBSD: src/sys/miscfs/procfs/procfs_vnops.c,v 1.76.2.7 2002/01/22 17:22:59 nectar Exp $
40  * $DragonFly: src/sys/vfs/procfs/procfs_vnops.c,v 1.19 2004/08/28 19:02:27 dillon Exp $
41  */
42 
43 /*
44  * procfs vnode interface
45  */
46 
47 #include <sys/param.h>
48 #include <sys/systm.h>
49 #include <sys/time.h>
50 #include <sys/kernel.h>
51 #include <sys/lock.h>
52 #include <sys/fcntl.h>
53 #include <sys/proc.h>
54 #include <sys/signalvar.h>
55 #include <sys/vnode.h>
56 #include <sys/uio.h>
57 #include <sys/mount.h>
58 #include <sys/namei.h>
59 #include <sys/dirent.h>
60 #include <sys/malloc.h>
61 #include <machine/reg.h>
62 #include <vm/vm_zone.h>
63 #include <vfs/procfs/procfs.h>
64 #include <sys/pioctl.h>
65 
66 static int	procfs_access (struct vop_access_args *);
67 static int	procfs_badop (void);
68 static int	procfs_bmap (struct vop_bmap_args *);
69 static int	procfs_close (struct vop_close_args *);
70 static int	procfs_getattr (struct vop_getattr_args *);
71 static int	procfs_inactive (struct vop_inactive_args *);
72 static int	procfs_ioctl (struct vop_ioctl_args *);
73 static int	procfs_lookup (struct vop_lookup_args *);
74 static int	procfs_open (struct vop_open_args *);
75 static int	procfs_print (struct vop_print_args *);
76 static int	procfs_readdir (struct vop_readdir_args *);
77 static int	procfs_readlink (struct vop_readlink_args *);
78 static int	procfs_reclaim (struct vop_reclaim_args *);
79 static int	procfs_setattr (struct vop_setattr_args *);
80 
81 /*
82  * This is a list of the valid names in the
83  * process-specific sub-directories.  It is
84  * used in procfs_lookup and procfs_readdir
85  */
86 static struct proc_target {
87 	u_char	pt_type;
88 	u_char	pt_namlen;
89 	char	*pt_name;
90 	pfstype	pt_pfstype;
91 	int	(*pt_valid) (struct proc *p);
92 } proc_targets[] = {
93 #define N(s) sizeof(s)-1, s
94 	/*	  name		type		validp */
95 	{ DT_DIR, N("."),	Pproc,		NULL },
96 	{ DT_DIR, N(".."),	Proot,		NULL },
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("dbregs"),	Pdbregs,	procfs_validdbregs },
101 	{ DT_REG, N("ctl"),	Pctl,		NULL },
102 	{ DT_REG, N("status"),	Pstatus,	NULL },
103 	{ DT_REG, N("note"),	Pnote,		NULL },
104 	{ DT_REG, N("notepg"),	Pnotepg,	NULL },
105 	{ DT_REG, N("map"), 	Pmap,		procfs_validmap },
106 	{ DT_REG, N("etype"),	Ptype,		procfs_validtype },
107 	{ DT_REG, N("cmdline"),	Pcmdline,	NULL },
108 	{ DT_REG, N("rlimit"),	Prlimit,	NULL },
109 	{ DT_LNK, N("file"),	Pfile,		NULL },
110 #undef N
111 };
112 static const int nproc_targets = sizeof(proc_targets) / sizeof(proc_targets[0]);
113 
114 static pid_t atopid (const char *, u_int);
115 
116 /*
117  * set things up for doing i/o on
118  * the pfsnode (vp).  (vp) is locked
119  * on entry, and should be left locked
120  * on exit.
121  *
122  * for procfs we don't need to do anything
123  * in particular for i/o.  all that is done
124  * is to support exclusive open on process
125  * memory images.
126  *
127  * procfs_open(struct vnode *a_vp, int a_mode, struct ucred *a_cred,
128  *		struct thread *a_td)
129  */
130 static int
131 procfs_open(struct vop_open_args *ap)
132 {
133 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
134 	struct proc *p1, *p2;
135 
136 	p2 = PFIND(pfs->pfs_pid);
137 	if (p2 == NULL)
138 		return (ENOENT);
139 	if (pfs->pfs_pid && !PRISON_CHECK(ap->a_cred, p2->p_ucred))
140 		return (ENOENT);
141 
142 	switch (pfs->pfs_type) {
143 	case Pmem:
144 		if (((pfs->pfs_flags & FWRITE) && (ap->a_mode & O_EXCL)) ||
145 		    ((pfs->pfs_flags & O_EXCL) && (ap->a_mode & FWRITE)))
146 			return (EBUSY);
147 
148 		p1 = ap->a_td->td_proc;
149 		KKASSERT(p1);
150 		/* Can't trace a process that's currently exec'ing. */
151 		if ((p2->p_flag & P_INEXEC) != 0)
152 			return EAGAIN;
153 		if (!CHECKIO(p1, p2) || p_trespass(ap->a_cred, p2->p_ucred))
154 			return (EPERM);
155 
156 		if (ap->a_mode & FWRITE)
157 			pfs->pfs_flags = ap->a_mode & (FWRITE|O_EXCL);
158 
159 		return (0);
160 
161 	default:
162 		break;
163 	}
164 
165 	return (0);
166 }
167 
168 /*
169  * close the pfsnode (vp) after doing i/o.
170  * (vp) is not locked on entry or exit.
171  *
172  * nothing to do for procfs other than undo
173  * any exclusive open flag (see _open above).
174  *
175  * procfs_close(struct vnode *a_vp, int a_fflag, struct ucred *a_cred,
176  *		struct thread *a_td)
177  */
178 static int
179 procfs_close(struct vop_close_args *ap)
180 {
181 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
182 	struct proc *p;
183 
184 	switch (pfs->pfs_type) {
185 	case Pmem:
186 		if ((ap->a_fflag & FWRITE) && (pfs->pfs_flags & O_EXCL))
187 			pfs->pfs_flags &= ~(FWRITE|O_EXCL);
188 		/*
189 		 * This rather complicated-looking code is trying to
190 		 * determine if this was the last close on this particular
191 		 * vnode.  While one would expect v_usecount to be 1 at
192 		 * that point, it seems that (according to John Dyson)
193 		 * the VM system will bump up the usecount.  So:  if the
194 		 * usecount is 2, and VOBJBUF is set, then this is really
195 		 * the last close.  Otherwise, if the usecount is < 2
196 		 * then it is definitely the last close.
197 		 * If this is the last close, then it checks to see if
198 		 * the target process has PF_LINGER set in p_pfsflags,
199 		 * if this is *not* the case, then the process' stop flags
200 		 * are cleared, and the process is woken up.  This is
201 		 * to help prevent the case where a process has been
202 		 * told to stop on an event, but then the requesting process
203 		 * has gone away or forgotten about it.
204 		 */
205 		if ((ap->a_vp->v_usecount < 2)
206 		    && (p = pfind(pfs->pfs_pid))
207 		    && !(p->p_pfsflags & PF_LINGER)) {
208 			p->p_stops = 0;
209 			p->p_step = 0;
210 			wakeup(&p->p_step);
211 		}
212 		break;
213 	default:
214 		break;
215 	}
216 
217 	return (0);
218 }
219 
220 /*
221  * do an ioctl operation on a pfsnode (vp).
222  * (vp) is not locked on entry or exit.
223  */
224 static int
225 procfs_ioctl(struct vop_ioctl_args *ap)
226 {
227 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
228 	struct proc *procp;
229 	struct proc *p;
230 	int error;
231 	int signo;
232 	struct procfs_status *psp;
233 	unsigned char flags;
234 
235 	procp = pfind(pfs->pfs_pid);
236 	if (procp == NULL)
237 		return ENOTTY;
238 	p = ap->a_td->td_proc;
239 	if (p == NULL)
240 		return EINVAL;
241 
242 	/* Can't trace a process that's currently exec'ing. */
243 	if ((procp->p_flag & P_INEXEC) != 0)
244 		return EAGAIN;
245 	if (!CHECKIO(p, procp) || p_trespass(ap->a_cred, procp->p_ucred))
246 		return EPERM;
247 
248 	switch (ap->a_command) {
249 	case PIOCBIS:
250 	  procp->p_stops |= *(unsigned int*)ap->a_data;
251 	  break;
252 	case PIOCBIC:
253 	  procp->p_stops &= ~*(unsigned int*)ap->a_data;
254 	  break;
255 	case PIOCSFL:
256 	  /*
257 	   * NFLAGS is "non-suser_xxx flags" -- currently, only
258 	   * PFS_ISUGID ("ignore set u/g id");
259 	   */
260 #define NFLAGS	(PF_ISUGID)
261 	  flags = (unsigned char)*(unsigned int*)ap->a_data;
262 	  if (flags & NFLAGS && (error = suser_cred(ap->a_cred, 0)))
263 	    return error;
264 	  procp->p_pfsflags = flags;
265 	  break;
266 	case PIOCGFL:
267 	  *(unsigned int*)ap->a_data = (unsigned int)procp->p_pfsflags;
268 	  break;
269 	case PIOCSTATUS:
270 	  psp = (struct procfs_status *)ap->a_data;
271 	  psp->state = (procp->p_step == 0);
272 	  psp->flags = procp->p_pfsflags;
273 	  psp->events = procp->p_stops;
274 	  if (procp->p_step) {
275 	    psp->why = procp->p_stype;
276 	    psp->val = procp->p_xstat;
277 	  } else {
278 	    psp->why = psp->val = 0;	/* Not defined values */
279 	  }
280 	  break;
281 	case PIOCWAIT:
282 	  psp = (struct procfs_status *)ap->a_data;
283 	  if (procp->p_step == 0) {
284 	    error = tsleep(&procp->p_stype, PCATCH, "piocwait", 0);
285 	    if (error)
286 	      return error;
287 	  }
288 	  psp->state = 1;	/* It stopped */
289 	  psp->flags = procp->p_pfsflags;
290 	  psp->events = procp->p_stops;
291 	  psp->why = procp->p_stype;	/* why it stopped */
292 	  psp->val = procp->p_xstat;	/* any extra info */
293 	  break;
294 	case PIOCCONT:	/* Restart a proc */
295 	  if (procp->p_step == 0)
296 	    return EINVAL;	/* Can only start a stopped process */
297 	  if ((signo = *(int*)ap->a_data) != 0) {
298 	    if (signo >= NSIG || signo <= 0)
299 	      return EINVAL;
300 	    psignal(procp, signo);
301 	  }
302 	  procp->p_step = 0;
303 	  wakeup(&procp->p_step);
304 	  break;
305 	default:
306 	  return (ENOTTY);
307 	}
308 	return 0;
309 }
310 
311 /*
312  * do block mapping for pfsnode (vp).
313  * since we don't use the buffer cache
314  * for procfs this function should never
315  * be called.  in any case, it's not clear
316  * what part of the kernel ever makes use
317  * of this function.  for sanity, this is the
318  * usual no-op bmap, although returning
319  * (EIO) would be a reasonable alternative.
320  *
321  * procfs_bmap(struct vnode *a_vp, daddr_t a_bn, struct vnode **a_vpp,
322  *		daddr_t *a_bnp, int *a_runp)
323  */
324 static int
325 procfs_bmap(struct vop_bmap_args *ap)
326 {
327 	if (ap->a_vpp != NULL)
328 		*ap->a_vpp = ap->a_vp;
329 	if (ap->a_bnp != NULL)
330 		*ap->a_bnp = ap->a_bn;
331 	if (ap->a_runp != NULL)
332 		*ap->a_runp = 0;
333 	return (0);
334 }
335 
336 /*
337  * procfs_inactive is called when the pfsnode
338  * is vrele'd and the reference count goes
339  * to zero.  (vp) will be on the vnode free
340  * list, so to get it back vget() must be
341  * used.
342  *
343  * (vp) is locked on entry, but must be unlocked on exit.
344  *
345  * procfs_inactive(struct vnode *a_vp, struct thread *a_td)
346  */
347 static int
348 procfs_inactive(struct vop_inactive_args *ap)
349 {
350 	struct vnode *vp = ap->a_vp;
351 
352 	VOP_UNLOCK(vp, NULL, 0, ap->a_td);
353 
354 	return (0);
355 }
356 
357 /*
358  * _reclaim is called when getnewvnode()
359  * wants to make use of an entry on the vnode
360  * free list.  at this time the filesystem needs
361  * to free any private data and remove the node
362  * from any private lists.
363  *
364  * procfs_reclaim(struct vnode *a_vp)
365  */
366 static int
367 procfs_reclaim(struct vop_reclaim_args *ap)
368 {
369 	return (procfs_freevp(ap->a_vp));
370 }
371 
372 /*
373  * _print is used for debugging.
374  * just print a readable description
375  * of (vp).
376  *
377  * procfs_print(struct vnode *a_vp)
378  */
379 static int
380 procfs_print(struct vop_print_args *ap)
381 {
382 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
383 
384 	printf("tag VT_PROCFS, type %d, pid %ld, mode %x, flags %lx\n",
385 	    pfs->pfs_type, (long)pfs->pfs_pid, pfs->pfs_mode, pfs->pfs_flags);
386 	return (0);
387 }
388 
389 /*
390  * generic entry point for unsupported operations
391  */
392 static int
393 procfs_badop(void)
394 {
395 	return (EIO);
396 }
397 
398 /*
399  * Invent attributes for pfsnode (vp) and store
400  * them in (vap).
401  * Directories lengths are returned as zero since
402  * any real length would require the genuine size
403  * to be computed, and nothing cares anyway.
404  *
405  * this is relatively minimal for procfs.
406  *
407  * procfs_getattr(struct vnode *a_vp, struct vattr *a_vap,
408  *		  struct ucred *a_cred,	struct thread *a_td)
409  */
410 static int
411 procfs_getattr(struct vop_getattr_args *ap)
412 {
413 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
414 	struct vattr *vap = ap->a_vap;
415 	struct proc *procp;
416 	int error;
417 
418 	/*
419 	 * First make sure that the process and its credentials
420 	 * still exist.
421 	 */
422 	switch (pfs->pfs_type) {
423 	case Proot:
424 	case Pcurproc:
425 		procp = 0;
426 		break;
427 
428 	default:
429 		procp = PFIND(pfs->pfs_pid);
430 		if (procp == NULL || procp->p_ucred == NULL)
431 			return (ENOENT);
432 	}
433 
434 	error = 0;
435 
436 	/* start by zeroing out the attributes */
437 	VATTR_NULL(vap);
438 
439 	/* next do all the common fields */
440 	vap->va_type = ap->a_vp->v_type;
441 	vap->va_mode = pfs->pfs_mode;
442 	vap->va_fileid = pfs->pfs_fileno;
443 	vap->va_flags = 0;
444 	vap->va_blocksize = PAGE_SIZE;
445 	vap->va_bytes = vap->va_size = 0;
446 	vap->va_fsid = ap->a_vp->v_mount->mnt_stat.f_fsid.val[0];
447 
448 	/*
449 	 * Make all times be current TOD.
450 	 * It would be possible to get the process start
451 	 * time from the p_stat structure, but there's
452 	 * no "file creation" time stamp anyway, and the
453 	 * p_stat structure is not addressible if u. gets
454 	 * swapped out for that process.
455 	 */
456 	nanotime(&vap->va_ctime);
457 	vap->va_atime = vap->va_mtime = vap->va_ctime;
458 
459 	/*
460 	 * If the process has exercised some setuid or setgid
461 	 * privilege, then rip away read/write permission so
462 	 * that only root can gain access.
463 	 */
464 	switch (pfs->pfs_type) {
465 	case Pctl:
466 	case Pregs:
467 	case Pfpregs:
468 	case Pdbregs:
469 	case Pmem:
470 		if (procp->p_flag & P_SUGID)
471 			vap->va_mode &= ~((VREAD|VWRITE)|
472 					  ((VREAD|VWRITE)>>3)|
473 					  ((VREAD|VWRITE)>>6));
474 		break;
475 	default:
476 		break;
477 	}
478 
479 	/*
480 	 * now do the object specific fields
481 	 *
482 	 * The size could be set from struct reg, but it's hardly
483 	 * worth the trouble, and it puts some (potentially) machine
484 	 * dependent data into this machine-independent code.  If it
485 	 * becomes important then this function should break out into
486 	 * a per-file stat function in the corresponding .c file.
487 	 */
488 
489 	vap->va_nlink = 1;
490 	if (procp) {
491 		vap->va_uid = procp->p_ucred->cr_uid;
492 		vap->va_gid = procp->p_ucred->cr_gid;
493 	}
494 
495 	switch (pfs->pfs_type) {
496 	case Proot:
497 		/*
498 		 * Set nlink to 1 to tell fts(3) we don't actually know.
499 		 */
500 		vap->va_nlink = 1;
501 		vap->va_uid = 0;
502 		vap->va_gid = 0;
503 		vap->va_size = vap->va_bytes = DEV_BSIZE;
504 		break;
505 
506 	case Pcurproc: {
507 		char buf[16];		/* should be enough */
508 		vap->va_uid = 0;
509 		vap->va_gid = 0;
510 		vap->va_size = vap->va_bytes =
511 		    snprintf(buf, sizeof(buf), "%ld", (long)curproc->p_pid);
512 		break;
513 	}
514 
515 	case Pproc:
516 		vap->va_nlink = nproc_targets;
517 		vap->va_size = vap->va_bytes = DEV_BSIZE;
518 		break;
519 
520 	case Pfile: {
521 		char *fullpath, *freepath;
522 		error = vn_fullpath(procp, NULL, &fullpath, &freepath);
523 		if (error == 0) {
524 			vap->va_size = strlen(fullpath);
525 			free(freepath, M_TEMP);
526 		} else {
527 			vap->va_size = sizeof("unknown") - 1;
528 			error = 0;
529 		}
530 		vap->va_bytes = vap->va_size;
531 		break;
532 	}
533 
534 	case Pmem:
535 		/*
536 		 * If we denied owner access earlier, then we have to
537 		 * change the owner to root - otherwise 'ps' and friends
538 		 * will break even though they are setgid kmem. *SIGH*
539 		 */
540 		if (procp->p_flag & P_SUGID)
541 			vap->va_uid = 0;
542 		else
543 			vap->va_uid = procp->p_ucred->cr_uid;
544 		break;
545 
546 	case Pregs:
547 		vap->va_bytes = vap->va_size = sizeof(struct reg);
548 		break;
549 
550 	case Pfpregs:
551 		vap->va_bytes = vap->va_size = sizeof(struct fpreg);
552 		break;
553 
554         case Pdbregs:
555                 vap->va_bytes = vap->va_size = sizeof(struct dbreg);
556                 break;
557 
558 	case Ptype:
559 	case Pmap:
560 	case Pctl:
561 	case Pstatus:
562 	case Pnote:
563 	case Pnotepg:
564 	case Pcmdline:
565 	case Prlimit:
566 		break;
567 
568 	default:
569 		panic("procfs_getattr");
570 	}
571 
572 	return (error);
573 }
574 
575 /*
576  * procfs_setattr(struct vnode *a_vp, struct vattr *a_vap,
577  *		  struct ucred *a_cred,	struct thread *a_td)
578  */
579 static int
580 procfs_setattr(struct vop_setattr_args *ap)
581 {
582 	if (ap->a_vap->va_flags != VNOVAL)
583 		return (EOPNOTSUPP);
584 
585 	/*
586 	 * just fake out attribute setting
587 	 * it's not good to generate an error
588 	 * return, otherwise things like creat()
589 	 * will fail when they try to set the
590 	 * file length to 0.  worse, this means
591 	 * that echo $note > /proc/$pid/note will fail.
592 	 */
593 
594 	return (0);
595 }
596 
597 /*
598  * implement access checking.
599  *
600  * something very similar to this code is duplicated
601  * throughout the 4bsd kernel and should be moved
602  * into kern/vfs_subr.c sometime.
603  *
604  * actually, the check for super-user is slightly
605  * broken since it will allow read access to write-only
606  * objects.  this doesn't cause any particular trouble
607  * but does mean that the i/o entry points need to check
608  * that the operation really does make sense.
609  *
610  * procfs_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred,
611  *		 struct thread *a_td)
612  */
613 static int
614 procfs_access(struct vop_access_args *ap)
615 {
616 	struct vattr *vap;
617 	struct vattr vattr;
618 	int error;
619 
620 	/*
621 	 * If you're the super-user,
622 	 * you always get access.
623 	 */
624 	if (ap->a_cred->cr_uid == 0)
625 		return (0);
626 
627 	vap = &vattr;
628 	error = VOP_GETATTR(ap->a_vp, vap, ap->a_td);
629 	if (error)
630 		return (error);
631 
632 	/*
633 	 * Access check is based on only one of owner, group, public.
634 	 * If not owner, then check group. If not a member of the
635 	 * group, then check public access.
636 	 */
637 	if (ap->a_cred->cr_uid != vap->va_uid) {
638 		gid_t *gp;
639 		int i;
640 
641 		ap->a_mode >>= 3;
642 		gp = ap->a_cred->cr_groups;
643 		for (i = 0; i < ap->a_cred->cr_ngroups; i++, gp++)
644 			if (vap->va_gid == *gp)
645 				goto found;
646 		ap->a_mode >>= 3;
647 found:
648 		;
649 	}
650 
651 	if ((vap->va_mode & ap->a_mode) == ap->a_mode)
652 		return (0);
653 
654 	return (EACCES);
655 }
656 
657 /*
658  * lookup.  this is incredibly complicated in the general case, however
659  * for most pseudo-filesystems very little needs to be done.
660  *
661  * procfs_lookup(struct vnode *a_dvp, struct vnode **a_vpp,
662  *		 struct componentname *a_cnp)
663  */
664 static int
665 procfs_lookup(struct vop_lookup_args *ap)
666 {
667 	struct componentname *cnp = ap->a_cnp;
668 	struct vnode **vpp = ap->a_vpp;
669 	struct vnode *dvp = ap->a_dvp;
670 	char *pname = cnp->cn_nameptr;
671 	/* struct proc *curp = cnp->cn_proc; */
672 	struct proc_target *pt;
673 	pid_t pid;
674 	struct pfsnode *pfs;
675 	struct proc *p;
676 	int i;
677 	int error;
678 
679 	*vpp = NULL;
680 
681 	if (cnp->cn_nameiop == NAMEI_DELETE || cnp->cn_nameiop == NAMEI_RENAME)
682 		return (EROFS);
683 
684 	error = 0;
685 	if (cnp->cn_namelen == 1 && *pname == '.') {
686 		*vpp = dvp;
687 		vref(*vpp);
688 		goto out;
689 	}
690 
691 	pfs = VTOPFS(dvp);
692 	switch (pfs->pfs_type) {
693 	case Proot:
694 		if (cnp->cn_flags & CNP_ISDOTDOT)
695 			return (EIO);
696 
697 		if (CNEQ(cnp, "curproc", 7)) {
698 			error = procfs_allocvp(dvp->v_mount, vpp, 0, Pcurproc);
699 			goto out;
700 		}
701 
702 		pid = atopid(pname, cnp->cn_namelen);
703 		if (pid == NO_PID)
704 			break;
705 
706 		p = PFIND(pid);
707 		if (p == NULL)
708 			break;
709 
710 		if (ps_showallprocs == 0 && ap->a_cnp->cn_cred->cr_uid != 0 &&
711 		    ap->a_cnp->cn_cred->cr_uid != p->p_ucred->cr_uid)
712 			break;
713 
714 		error = procfs_allocvp(dvp->v_mount, vpp, pid, Pproc);
715 		goto out;
716 
717 	case Pproc:
718 		if (cnp->cn_flags & CNP_ISDOTDOT) {
719 			error = procfs_root(dvp->v_mount, vpp);
720 			goto out;
721 		}
722 
723 		p = PFIND(pfs->pfs_pid);
724 		if (p == NULL)
725 			break;
726 
727 		if (ps_showallprocs == 0 && ap->a_cnp->cn_cred->cr_uid != 0 &&
728 		    ap->a_cnp->cn_cred->cr_uid != p->p_ucred->cr_uid)
729 			break;
730 
731 		for (pt = proc_targets, i = 0; i < nproc_targets; pt++, i++) {
732 			if (cnp->cn_namelen == pt->pt_namlen &&
733 			    bcmp(pt->pt_name, pname, cnp->cn_namelen) == 0 &&
734 			    (pt->pt_valid == NULL || (*pt->pt_valid)(p)))
735 				goto found;
736 		}
737 		break;
738 	found:
739 		error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
740 					pt->pt_pfstype);
741 		goto out;
742 
743 	default:
744 		error = ENOTDIR;
745 		goto out;
746 	}
747 	if (cnp->cn_nameiop == NAMEI_LOOKUP)
748 		error = ENOENT;
749 	else
750 		error = EROFS;
751 	/*
752 	 * If no error occured *vpp will hold a referenced locked vnode.
753 	 * dvp was passed to us locked and *vpp must be returned locked
754 	 * so if dvp != *vpp and CNP_LOCKPARENT is not set, unlock dvp.
755 	 */
756 out:
757 	if (error == 0) {
758 		if (*vpp != dvp && (cnp->cn_flags & CNP_LOCKPARENT) == 0) {
759 			cnp->cn_flags |= CNP_PDIRUNLOCK;
760 			VOP_UNLOCK(dvp, NULL, 0, cnp->cn_td);
761 		}
762 	}
763 	return (error);
764 }
765 
766 /*
767  * Does this process have a text file?
768  */
769 int
770 procfs_validfile(struct proc *p)
771 {
772 	return (procfs_findtextvp(p) != NULLVP);
773 }
774 
775 /*
776  * readdir() returns directory entries from pfsnode (vp).
777  *
778  * We generate just one directory entry at a time, as it would probably
779  * not pay off to buffer several entries locally to save uiomove calls.
780  *
781  * procfs_readdir(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred,
782  *		  int *a_eofflag, int *a_ncookies, u_long **a_cookies)
783  */
784 static int
785 procfs_readdir(struct vop_readdir_args *ap)
786 {
787 	struct uio *uio = ap->a_uio;
788 	struct dirent d;
789 	struct dirent *dp = &d;
790 	struct pfsnode *pfs;
791 	int count, error, i, off;
792 	static u_int delen;
793 
794 	if (!delen) {
795 
796 		d.d_namlen = PROCFS_NAMELEN;
797 		delen = GENERIC_DIRSIZ(&d);
798 	}
799 
800 	pfs = VTOPFS(ap->a_vp);
801 
802 	off = (int)uio->uio_offset;
803 	if (off != uio->uio_offset || off < 0 ||
804 	    off % delen != 0 || uio->uio_resid < delen)
805 		return (EINVAL);
806 
807 	error = 0;
808 	count = 0;
809 	i = off / delen;
810 
811 	switch (pfs->pfs_type) {
812 	/*
813 	 * this is for the process-specific sub-directories.
814 	 * all that is needed to is copy out all the entries
815 	 * from the procent[] table (top of this file).
816 	 */
817 	case Pproc: {
818 		struct proc *p;
819 		struct proc_target *pt;
820 
821 		p = PFIND(pfs->pfs_pid);
822 		if (p == NULL)
823 			break;
824 		if (!PRISON_CHECK(ap->a_cred, p->p_ucred))
825 			break;
826 
827 		for (pt = &proc_targets[i];
828 		     uio->uio_resid >= delen && i < nproc_targets; pt++, i++) {
829 			if (pt->pt_valid && (*pt->pt_valid)(p) == 0)
830 				continue;
831 
832 			dp->d_reclen = delen;
833 			dp->d_fileno = PROCFS_FILENO(pfs->pfs_pid, pt->pt_pfstype);
834 			dp->d_namlen = pt->pt_namlen;
835 			bcopy(pt->pt_name, dp->d_name, pt->pt_namlen + 1);
836 			dp->d_type = pt->pt_type;
837 
838 			if ((error = uiomove((caddr_t)dp, delen, uio)) != 0)
839 				break;
840 		}
841 
842 	    	break;
843 	    }
844 
845 	/*
846 	 * this is for the root of the procfs filesystem
847 	 * what is needed is a special entry for "curproc"
848 	 * followed by an entry for each process on allproc
849 #ifdef PROCFS_ZOMBIE
850 	 * and zombproc.
851 #endif
852 	 */
853 
854 	case Proot: {
855 #ifdef PROCFS_ZOMBIE
856 		int doingzomb = 0;
857 #endif
858 		int pcnt = 0;
859 		volatile struct proc *p = allproc.lh_first;
860 
861 		for (; p && uio->uio_resid >= delen; i++, pcnt++) {
862 			bzero((char *) dp, delen);
863 			dp->d_reclen = delen;
864 
865 			switch (i) {
866 			case 0:		/* `.' */
867 			case 1:		/* `..' */
868 				dp->d_fileno = PROCFS_FILENO(0, Proot);
869 				dp->d_namlen = i + 1;
870 				bcopy("..", dp->d_name, dp->d_namlen);
871 				dp->d_name[i + 1] = '\0';
872 				dp->d_type = DT_DIR;
873 				break;
874 
875 			case 2:
876 				dp->d_fileno = PROCFS_FILENO(0, Pcurproc);
877 				dp->d_namlen = 7;
878 				bcopy("curproc", dp->d_name, 8);
879 				dp->d_type = DT_LNK;
880 				break;
881 
882 			default:
883 				while (pcnt < i) {
884 					p = p->p_list.le_next;
885 					if (!p)
886 						goto done;
887 					if (!PRISON_CHECK(ap->a_cred, p->p_ucred))
888 						continue;
889 					pcnt++;
890 				}
891 				while (!PRISON_CHECK(ap->a_cred, p->p_ucred)) {
892 					p = p->p_list.le_next;
893 					if (!p)
894 						goto done;
895 				}
896 				if (ps_showallprocs == 0 &&
897 				    ap->a_cred->cr_uid != 0 &&
898 				    ap->a_cred->cr_uid !=
899 				    p->p_ucred->cr_uid) {
900 					p = p->p_list.le_next;
901 					if (!p)
902 						goto done;
903 					break;
904 				}
905 
906 				dp->d_fileno = PROCFS_FILENO(p->p_pid, Pproc);
907 				dp->d_namlen = sprintf(dp->d_name, "%ld",
908 				    (long)p->p_pid);
909 				dp->d_type = DT_DIR;
910 				p = p->p_list.le_next;
911 				break;
912 			}
913 
914 			if ((error = uiomove((caddr_t)dp, delen, uio)) != 0)
915 				break;
916 		}
917 	done:
918 
919 #ifdef PROCFS_ZOMBIE
920 		if (p == NULL && doingzomb == 0) {
921 			doingzomb = 1;
922 			p = zombproc.lh_first;
923 			goto again;
924 		}
925 #endif
926 
927 		break;
928 
929 	    }
930 
931 	default:
932 		error = ENOTDIR;
933 		break;
934 	}
935 
936 	uio->uio_offset = i * delen;
937 
938 	return (error);
939 }
940 
941 /*
942  * readlink reads the link of `curproc' or `file'
943  */
944 static int
945 procfs_readlink(struct vop_readlink_args *ap)
946 {
947 	char buf[16];		/* should be enough */
948 	struct proc *procp;
949 	struct vnode *vp = ap->a_vp;
950 	struct pfsnode *pfs = VTOPFS(vp);
951 	char *fullpath, *freepath;
952 	int error, len;
953 
954 	switch (pfs->pfs_type) {
955 	case Pcurproc:
956 		if (pfs->pfs_fileno != PROCFS_FILENO(0, Pcurproc))
957 			return (EINVAL);
958 
959 		len = snprintf(buf, sizeof(buf), "%ld", (long)curproc->p_pid);
960 
961 		return (uiomove(buf, len, ap->a_uio));
962 	/*
963 	 * There _should_ be no way for an entire process to disappear
964 	 * from under us...
965 	 */
966 	case Pfile:
967 		procp = PFIND(pfs->pfs_pid);
968 		if (procp == NULL || procp->p_ucred == NULL) {
969 			printf("procfs_readlink: pid %d disappeared\n",
970 			    pfs->pfs_pid);
971 			return (uiomove("unknown", sizeof("unknown") - 1,
972 			    ap->a_uio));
973 		}
974 		error = vn_fullpath(procp, NULL, &fullpath, &freepath);
975 		if (error != 0)
976 			return (uiomove("unknown", sizeof("unknown") - 1,
977 			    ap->a_uio));
978 		error = uiomove(fullpath, strlen(fullpath), ap->a_uio);
979 		free(freepath, M_TEMP);
980 		return (error);
981 	default:
982 		return (EINVAL);
983 	}
984 }
985 
986 /*
987  * convert decimal ascii to pid_t
988  */
989 static pid_t
990 atopid(const char *b, u_int len)
991 {
992 	pid_t p = 0;
993 
994 	while (len--) {
995 		char c = *b++;
996 		if (c < '0' || c > '9')
997 			return (NO_PID);
998 		p = 10 * p + (c - '0');
999 		if (p > PID_MAX)
1000 			return (NO_PID);
1001 	}
1002 
1003 	return (p);
1004 }
1005 
1006 /*
1007  * procfs vnode operations.
1008  */
1009 struct vnodeopv_entry_desc procfs_vnodeop_entries[] = {
1010 	{ &vop_default_desc,		vop_defaultop },
1011 	{ &vop_access_desc,		(void *) procfs_access },
1012 	{ &vop_advlock_desc,		(void *) procfs_badop },
1013 	{ &vop_bmap_desc,		(void *) procfs_bmap },
1014 	{ &vop_close_desc,		(void *) procfs_close },
1015 	{ &vop_create_desc,		(void *) procfs_badop },
1016 	{ &vop_getattr_desc,		(void *) procfs_getattr },
1017 	{ &vop_inactive_desc,		(void *) procfs_inactive },
1018 	{ &vop_link_desc,		(void *) procfs_badop },
1019 	{ &vop_lookup_desc,		(void *) procfs_lookup },
1020 	{ &vop_mkdir_desc,		(void *) procfs_badop },
1021 	{ &vop_mknod_desc,		(void *) procfs_badop },
1022 	{ &vop_open_desc,		(void *) procfs_open },
1023 	{ &vop_pathconf_desc,		(void *) vop_stdpathconf },
1024 	{ &vop_print_desc,		(void *) procfs_print },
1025 	{ &vop_read_desc,		(void *) procfs_rw },
1026 	{ &vop_readdir_desc,		(void *) procfs_readdir },
1027 	{ &vop_readlink_desc,		(void *) procfs_readlink },
1028 	{ &vop_reclaim_desc,		(void *) procfs_reclaim },
1029 	{ &vop_remove_desc,		(void *) procfs_badop },
1030 	{ &vop_rename_desc,		(void *) procfs_badop },
1031 	{ &vop_rmdir_desc,		(void *) procfs_badop },
1032 	{ &vop_setattr_desc,		(void *) procfs_setattr },
1033 	{ &vop_symlink_desc,		(void *) procfs_badop },
1034 	{ &vop_write_desc,		(void *) procfs_rw },
1035 	{ &vop_ioctl_desc,		(void *) procfs_ioctl },
1036 	{ NULL, NULL }
1037 };
1038 
1039