xref: /dflybsd-src/sys/vfs/procfs/procfs_vnops.c (revision 41871674d0079dec70d55eb824f39d07dc7b3310)
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 1.29 2006/04/01 20:46:53 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 #include <machine/limits.h>
67 
68 static int	procfs_access (struct vop_access_args *);
69 static int	procfs_badop (void);
70 static int	procfs_bmap (struct vop_bmap_args *);
71 static int	procfs_close (struct vop_close_args *);
72 static int	procfs_getattr (struct vop_getattr_args *);
73 static int	procfs_inactive (struct vop_inactive_args *);
74 static int	procfs_ioctl (struct vop_ioctl_args *);
75 static int	procfs_lookup (struct vop_old_lookup_args *);
76 static int	procfs_open (struct vop_open_args *);
77 static int	procfs_print (struct vop_print_args *);
78 static int	procfs_readdir (struct vop_readdir_args *);
79 static int	procfs_readlink (struct vop_readlink_args *);
80 static int	procfs_reclaim (struct vop_reclaim_args *);
81 static int	procfs_setattr (struct vop_setattr_args *);
82 
83 static int	procfs_readdir_proc(struct vop_readdir_args *);
84 static int	procfs_readdir_root(struct vop_readdir_args *);
85 
86 /*
87  * This is a list of the valid names in the
88  * process-specific sub-directories.  It is
89  * used in procfs_lookup and procfs_readdir
90  */
91 static struct proc_target {
92 	u_char	pt_type;
93 	u_char	pt_namlen;
94 	char	*pt_name;
95 	pfstype	pt_pfstype;
96 	int	(*pt_valid) (struct proc *p);
97 } proc_targets[] = {
98 #define N(s) sizeof(s)-1, s
99 	/*	  name		type		validp */
100 	{ DT_DIR, N("."),	Pproc,		NULL },
101 	{ DT_DIR, N(".."),	Proot,		NULL },
102 	{ DT_REG, N("mem"),	Pmem,		NULL },
103 	{ DT_REG, N("regs"),	Pregs,		procfs_validregs },
104 	{ DT_REG, N("fpregs"),	Pfpregs,	procfs_validfpregs },
105 	{ DT_REG, N("dbregs"),	Pdbregs,	procfs_validdbregs },
106 	{ DT_REG, N("ctl"),	Pctl,		NULL },
107 	{ DT_REG, N("status"),	Pstatus,	NULL },
108 	{ DT_REG, N("note"),	Pnote,		NULL },
109 	{ DT_REG, N("notepg"),	Pnotepg,	NULL },
110 	{ DT_REG, N("map"), 	Pmap,		procfs_validmap },
111 	{ DT_REG, N("etype"),	Ptype,		procfs_validtype },
112 	{ DT_REG, N("cmdline"),	Pcmdline,	NULL },
113 	{ DT_REG, N("rlimit"),	Prlimit,	NULL },
114 	{ DT_LNK, N("file"),	Pfile,		NULL },
115 #undef N
116 };
117 static const int nproc_targets = sizeof(proc_targets) / sizeof(proc_targets[0]);
118 
119 static pid_t atopid (const char *, u_int);
120 
121 /*
122  * set things up for doing i/o on
123  * the pfsnode (vp).  (vp) is locked
124  * on entry, and should be left locked
125  * on exit.
126  *
127  * for procfs we don't need to do anything
128  * in particular for i/o.  all that is done
129  * is to support exclusive open on process
130  * memory images.
131  *
132  * procfs_open(struct vnode *a_vp, int a_mode, struct ucred *a_cred,
133  *		struct thread *a_td)
134  */
135 static int
136 procfs_open(struct vop_open_args *ap)
137 {
138 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
139 	struct proc *p1, *p2;
140 
141 	p2 = PFIND(pfs->pfs_pid);
142 	if (p2 == NULL)
143 		return (ENOENT);
144 	if (pfs->pfs_pid && !PRISON_CHECK(ap->a_cred, p2->p_ucred))
145 		return (ENOENT);
146 
147 	switch (pfs->pfs_type) {
148 	case Pmem:
149 		if (((pfs->pfs_flags & FWRITE) && (ap->a_mode & O_EXCL)) ||
150 		    ((pfs->pfs_flags & O_EXCL) && (ap->a_mode & FWRITE)))
151 			return (EBUSY);
152 
153 		p1 = ap->a_td->td_proc;
154 		KKASSERT(p1);
155 		/* Can't trace a process that's currently exec'ing. */
156 		if ((p2->p_flag & P_INEXEC) != 0)
157 			return EAGAIN;
158 		if (!CHECKIO(p1, p2) || p_trespass(ap->a_cred, p2->p_ucred))
159 			return (EPERM);
160 
161 		if (ap->a_mode & FWRITE)
162 			pfs->pfs_flags = ap->a_mode & (FWRITE|O_EXCL);
163 
164 		break;
165 
166 	default:
167 		break;
168 	}
169 
170 	return (vop_stdopen(ap));
171 }
172 
173 /*
174  * close the pfsnode (vp) after doing i/o.
175  * (vp) is not locked on entry or exit.
176  *
177  * nothing to do for procfs other than undo
178  * any exclusive open flag (see _open above).
179  *
180  * procfs_close(struct vnode *a_vp, int a_fflag, struct ucred *a_cred,
181  *		struct thread *a_td)
182  */
183 static int
184 procfs_close(struct vop_close_args *ap)
185 {
186 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
187 	struct proc *p;
188 
189 	switch (pfs->pfs_type) {
190 	case Pmem:
191 		if ((ap->a_fflag & FWRITE) && (pfs->pfs_flags & O_EXCL))
192 			pfs->pfs_flags &= ~(FWRITE|O_EXCL);
193 		/*
194 		 * This rather complicated-looking code is trying to
195 		 * determine if this was the last close on this particular
196 		 * vnode.  While one would expect v_usecount to be 1 at
197 		 * that point, it seems that (according to John Dyson)
198 		 * the VM system will bump up the usecount.  So:  if the
199 		 * usecount is 2, and VOBJBUF is set, then this is really
200 		 * the last close.  Otherwise, if the usecount is < 2
201 		 * then it is definitely the last close.
202 		 * If this is the last close, then it checks to see if
203 		 * the target process has PF_LINGER set in p_pfsflags,
204 		 * if this is *not* the case, then the process' stop flags
205 		 * are cleared, and the process is woken up.  This is
206 		 * to help prevent the case where a process has been
207 		 * told to stop on an event, but then the requesting process
208 		 * has gone away or forgotten about it.
209 		 */
210 		if ((ap->a_vp->v_usecount < 2)
211 		    && (p = pfind(pfs->pfs_pid))
212 		    && !(p->p_pfsflags & PF_LINGER)) {
213 			p->p_stops = 0;
214 			p->p_step = 0;
215 			wakeup(&p->p_step);
216 		}
217 		break;
218 	default:
219 		break;
220 	}
221 
222 	return (vop_stdclose(ap));
223 }
224 
225 /*
226  * do an ioctl operation on a pfsnode (vp).
227  * (vp) is not locked on entry or exit.
228  */
229 static int
230 procfs_ioctl(struct vop_ioctl_args *ap)
231 {
232 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
233 	struct proc *procp;
234 	struct proc *p;
235 	int error;
236 	int signo;
237 	struct procfs_status *psp;
238 	unsigned char flags;
239 
240 	procp = pfind(pfs->pfs_pid);
241 	if (procp == NULL)
242 		return ENOTTY;
243 	p = ap->a_td->td_proc;
244 	if (p == NULL)
245 		return EINVAL;
246 
247 	/* Can't trace a process that's currently exec'ing. */
248 	if ((procp->p_flag & P_INEXEC) != 0)
249 		return EAGAIN;
250 	if (!CHECKIO(p, procp) || p_trespass(ap->a_cred, procp->p_ucred))
251 		return EPERM;
252 
253 	switch (ap->a_command) {
254 	case PIOCBIS:
255 	  procp->p_stops |= *(unsigned int*)ap->a_data;
256 	  break;
257 	case PIOCBIC:
258 	  procp->p_stops &= ~*(unsigned int*)ap->a_data;
259 	  break;
260 	case PIOCSFL:
261 	  /*
262 	   * NFLAGS is "non-suser_xxx flags" -- currently, only
263 	   * PFS_ISUGID ("ignore set u/g id");
264 	   */
265 #define NFLAGS	(PF_ISUGID)
266 	  flags = (unsigned char)*(unsigned int*)ap->a_data;
267 	  if (flags & NFLAGS && (error = suser_cred(ap->a_cred, 0)))
268 	    return error;
269 	  procp->p_pfsflags = flags;
270 	  break;
271 	case PIOCGFL:
272 	  *(unsigned int*)ap->a_data = (unsigned int)procp->p_pfsflags;
273 	  break;
274 	case PIOCSTATUS:
275 	  psp = (struct procfs_status *)ap->a_data;
276 	  psp->state = (procp->p_step == 0);
277 	  psp->flags = procp->p_pfsflags;
278 	  psp->events = procp->p_stops;
279 	  if (procp->p_step) {
280 	    psp->why = procp->p_stype;
281 	    psp->val = procp->p_xstat;
282 	  } else {
283 	    psp->why = psp->val = 0;	/* Not defined values */
284 	  }
285 	  break;
286 	case PIOCWAIT:
287 	  psp = (struct procfs_status *)ap->a_data;
288 	  if (procp->p_step == 0) {
289 	    error = tsleep(&procp->p_stype, PCATCH, "piocwait", 0);
290 	    if (error)
291 	      return error;
292 	  }
293 	  psp->state = 1;	/* It stopped */
294 	  psp->flags = procp->p_pfsflags;
295 	  psp->events = procp->p_stops;
296 	  psp->why = procp->p_stype;	/* why it stopped */
297 	  psp->val = procp->p_xstat;	/* any extra info */
298 	  break;
299 	case PIOCCONT:	/* Restart a proc */
300 	  if (procp->p_step == 0)
301 	    return EINVAL;	/* Can only start a stopped process */
302 	  if ((signo = *(int*)ap->a_data) != 0) {
303 	    if (signo >= NSIG || signo <= 0)
304 	      return EINVAL;
305 	    psignal(procp, signo);
306 	  }
307 	  procp->p_step = 0;
308 	  wakeup(&procp->p_step);
309 	  break;
310 	default:
311 	  return (ENOTTY);
312 	}
313 	return 0;
314 }
315 
316 /*
317  * do block mapping for pfsnode (vp).
318  * since we don't use the buffer cache
319  * for procfs this function should never
320  * be called.  in any case, it's not clear
321  * what part of the kernel ever makes use
322  * of this function.  for sanity, this is the
323  * usual no-op bmap, although returning
324  * (EIO) would be a reasonable alternative.
325  *
326  * procfs_bmap(struct vnode *a_vp, off_t a_loffset, struct vnode **a_vpp,
327  *		off_t *a_doffsetp, int *a_runp)
328  */
329 static int
330 procfs_bmap(struct vop_bmap_args *ap)
331 {
332 	if (ap->a_vpp != NULL)
333 		*ap->a_vpp = ap->a_vp;
334 	if (ap->a_doffsetp != NULL)
335 		*ap->a_doffsetp = ap->a_loffset;
336 	if (ap->a_runp != NULL)
337 		*ap->a_runp = 0;
338 	if (ap->a_runb != NULL)
339 		*ap->a_runb = 0;
340 	return (0);
341 }
342 
343 /*
344  * procfs_inactive is called when the pfsnode
345  * is vrele'd and the reference count goes
346  * to zero.  (vp) will be on the vnode free
347  * list, so to get it back vget() must be
348  * used.
349  *
350  * (vp) is locked on entry, but must be unlocked on exit.
351  *
352  * procfs_inactive(struct vnode *a_vp, struct thread *a_td)
353  */
354 static int
355 procfs_inactive(struct vop_inactive_args *ap)
356 {
357 	/*struct vnode *vp = ap->a_vp;*/
358 
359 	return (0);
360 }
361 
362 /*
363  * _reclaim is called when getnewvnode()
364  * wants to make use of an entry on the vnode
365  * free list.  at this time the filesystem needs
366  * to free any private data and remove the node
367  * from any private lists.
368  *
369  * procfs_reclaim(struct vnode *a_vp)
370  */
371 static int
372 procfs_reclaim(struct vop_reclaim_args *ap)
373 {
374 	return (procfs_freevp(ap->a_vp));
375 }
376 
377 /*
378  * _print is used for debugging.
379  * just print a readable description
380  * of (vp).
381  *
382  * procfs_print(struct vnode *a_vp)
383  */
384 static int
385 procfs_print(struct vop_print_args *ap)
386 {
387 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
388 
389 	printf("tag VT_PROCFS, type %d, pid %ld, mode %x, flags %lx\n",
390 	    pfs->pfs_type, (long)pfs->pfs_pid, pfs->pfs_mode, pfs->pfs_flags);
391 	return (0);
392 }
393 
394 /*
395  * generic entry point for unsupported operations
396  */
397 static int
398 procfs_badop(void)
399 {
400 	return (EIO);
401 }
402 
403 /*
404  * Invent attributes for pfsnode (vp) and store
405  * them in (vap).
406  * Directories lengths are returned as zero since
407  * any real length would require the genuine size
408  * to be computed, and nothing cares anyway.
409  *
410  * this is relatively minimal for procfs.
411  *
412  * procfs_getattr(struct vnode *a_vp, struct vattr *a_vap,
413  *		  struct ucred *a_cred,	struct thread *a_td)
414  */
415 static int
416 procfs_getattr(struct vop_getattr_args *ap)
417 {
418 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
419 	struct vattr *vap = ap->a_vap;
420 	struct proc *procp;
421 	int error;
422 
423 	/*
424 	 * First make sure that the process and its credentials
425 	 * still exist.
426 	 */
427 	switch (pfs->pfs_type) {
428 	case Proot:
429 	case Pcurproc:
430 		procp = 0;
431 		break;
432 
433 	default:
434 		procp = PFIND(pfs->pfs_pid);
435 		if (procp == NULL || procp->p_ucred == NULL)
436 			return (ENOENT);
437 	}
438 
439 	error = 0;
440 
441 	/* start by zeroing out the attributes */
442 	VATTR_NULL(vap);
443 
444 	/* next do all the common fields */
445 	vap->va_type = ap->a_vp->v_type;
446 	vap->va_mode = pfs->pfs_mode;
447 	vap->va_fileid = pfs->pfs_fileno;
448 	vap->va_flags = 0;
449 	vap->va_blocksize = PAGE_SIZE;
450 	vap->va_bytes = vap->va_size = 0;
451 	vap->va_fsid = ap->a_vp->v_mount->mnt_stat.f_fsid.val[0];
452 
453 	/*
454 	 * Make all times be current TOD.
455 	 * It would be possible to get the process start
456 	 * time from the p_stat structure, but there's
457 	 * no "file creation" time stamp anyway, and the
458 	 * p_stat structure is not addressible if u. gets
459 	 * swapped out for that process.
460 	 */
461 	nanotime(&vap->va_ctime);
462 	vap->va_atime = vap->va_mtime = vap->va_ctime;
463 
464 	/*
465 	 * If the process has exercised some setuid or setgid
466 	 * privilege, then rip away read/write permission so
467 	 * that only root can gain access.
468 	 */
469 	switch (pfs->pfs_type) {
470 	case Pctl:
471 	case Pregs:
472 	case Pfpregs:
473 	case Pdbregs:
474 	case Pmem:
475 		if (procp->p_flag & P_SUGID)
476 			vap->va_mode &= ~((VREAD|VWRITE)|
477 					  ((VREAD|VWRITE)>>3)|
478 					  ((VREAD|VWRITE)>>6));
479 		break;
480 	default:
481 		break;
482 	}
483 
484 	/*
485 	 * now do the object specific fields
486 	 *
487 	 * The size could be set from struct reg, but it's hardly
488 	 * worth the trouble, and it puts some (potentially) machine
489 	 * dependent data into this machine-independent code.  If it
490 	 * becomes important then this function should break out into
491 	 * a per-file stat function in the corresponding .c file.
492 	 */
493 
494 	vap->va_nlink = 1;
495 	if (procp) {
496 		vap->va_uid = procp->p_ucred->cr_uid;
497 		vap->va_gid = procp->p_ucred->cr_gid;
498 	}
499 
500 	switch (pfs->pfs_type) {
501 	case Proot:
502 		/*
503 		 * Set nlink to 1 to tell fts(3) we don't actually know.
504 		 */
505 		vap->va_nlink = 1;
506 		vap->va_uid = 0;
507 		vap->va_gid = 0;
508 		vap->va_size = vap->va_bytes = DEV_BSIZE;
509 		break;
510 
511 	case Pcurproc: {
512 		char buf[16];		/* should be enough */
513 		vap->va_uid = 0;
514 		vap->va_gid = 0;
515 		vap->va_size = vap->va_bytes =
516 		    snprintf(buf, sizeof(buf), "%ld", (long)curproc->p_pid);
517 		break;
518 	}
519 
520 	case Pproc:
521 		vap->va_nlink = nproc_targets;
522 		vap->va_size = vap->va_bytes = DEV_BSIZE;
523 		break;
524 
525 	case Pfile: {
526 		char *fullpath, *freepath;
527 		error = vn_fullpath(procp, NULL, &fullpath, &freepath);
528 		if (error == 0) {
529 			vap->va_size = strlen(fullpath);
530 			free(freepath, M_TEMP);
531 		} else {
532 			vap->va_size = sizeof("unknown") - 1;
533 			error = 0;
534 		}
535 		vap->va_bytes = vap->va_size;
536 		break;
537 	}
538 
539 	case Pmem:
540 		/*
541 		 * If we denied owner access earlier, then we have to
542 		 * change the owner to root - otherwise 'ps' and friends
543 		 * will break even though they are setgid kmem. *SIGH*
544 		 */
545 		if (procp->p_flag & P_SUGID)
546 			vap->va_uid = 0;
547 		else
548 			vap->va_uid = procp->p_ucred->cr_uid;
549 		break;
550 
551 	case Pregs:
552 		vap->va_bytes = vap->va_size = sizeof(struct reg);
553 		break;
554 
555 	case Pfpregs:
556 		vap->va_bytes = vap->va_size = sizeof(struct fpreg);
557 		break;
558 
559         case Pdbregs:
560                 vap->va_bytes = vap->va_size = sizeof(struct dbreg);
561                 break;
562 
563 	case Ptype:
564 	case Pmap:
565 	case Pctl:
566 	case Pstatus:
567 	case Pnote:
568 	case Pnotepg:
569 	case Pcmdline:
570 	case Prlimit:
571 		break;
572 
573 	default:
574 		panic("procfs_getattr");
575 	}
576 
577 	return (error);
578 }
579 
580 /*
581  * procfs_setattr(struct vnode *a_vp, struct vattr *a_vap,
582  *		  struct ucred *a_cred,	struct thread *a_td)
583  */
584 static int
585 procfs_setattr(struct vop_setattr_args *ap)
586 {
587 	if (ap->a_vap->va_flags != VNOVAL)
588 		return (EOPNOTSUPP);
589 
590 	/*
591 	 * just fake out attribute setting
592 	 * it's not good to generate an error
593 	 * return, otherwise things like creat()
594 	 * will fail when they try to set the
595 	 * file length to 0.  worse, this means
596 	 * that echo $note > /proc/$pid/note will fail.
597 	 */
598 
599 	return (0);
600 }
601 
602 /*
603  * implement access checking.
604  *
605  * something very similar to this code is duplicated
606  * throughout the 4bsd kernel and should be moved
607  * into kern/vfs_subr.c sometime.
608  *
609  * actually, the check for super-user is slightly
610  * broken since it will allow read access to write-only
611  * objects.  this doesn't cause any particular trouble
612  * but does mean that the i/o entry points need to check
613  * that the operation really does make sense.
614  *
615  * procfs_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred,
616  *		 struct thread *a_td)
617  */
618 static int
619 procfs_access(struct vop_access_args *ap)
620 {
621 	struct vattr *vap;
622 	struct vattr vattr;
623 	int error;
624 
625 	/*
626 	 * If you're the super-user,
627 	 * you always get access.
628 	 */
629 	if (ap->a_cred->cr_uid == 0)
630 		return (0);
631 
632 	vap = &vattr;
633 	error = VOP_GETATTR(ap->a_vp, vap, ap->a_td);
634 	if (error)
635 		return (error);
636 
637 	/*
638 	 * Access check is based on only one of owner, group, public.
639 	 * If not owner, then check group. If not a member of the
640 	 * group, then check public access.
641 	 */
642 	if (ap->a_cred->cr_uid != vap->va_uid) {
643 		gid_t *gp;
644 		int i;
645 
646 		ap->a_mode >>= 3;
647 		gp = ap->a_cred->cr_groups;
648 		for (i = 0; i < ap->a_cred->cr_ngroups; i++, gp++)
649 			if (vap->va_gid == *gp)
650 				goto found;
651 		ap->a_mode >>= 3;
652 found:
653 		;
654 	}
655 
656 	if ((vap->va_mode & ap->a_mode) == ap->a_mode)
657 		return (0);
658 
659 	return (EACCES);
660 }
661 
662 /*
663  * lookup.  this is incredibly complicated in the general case, however
664  * for most pseudo-filesystems very little needs to be done.
665  *
666  * procfs_lookup(struct vnode *a_dvp, struct vnode **a_vpp,
667  *		 struct componentname *a_cnp)
668  */
669 static int
670 procfs_lookup(struct vop_old_lookup_args *ap)
671 {
672 	struct componentname *cnp = ap->a_cnp;
673 	struct vnode **vpp = ap->a_vpp;
674 	struct vnode *dvp = ap->a_dvp;
675 	char *pname = cnp->cn_nameptr;
676 	/* struct proc *curp = cnp->cn_proc; */
677 	struct proc_target *pt;
678 	pid_t pid;
679 	struct pfsnode *pfs;
680 	struct proc *p;
681 	int i;
682 	int error;
683 
684 	*vpp = NULL;
685 
686 	if (cnp->cn_nameiop == NAMEI_DELETE || cnp->cn_nameiop == NAMEI_RENAME)
687 		return (EROFS);
688 
689 	error = 0;
690 	if (cnp->cn_namelen == 1 && *pname == '.') {
691 		*vpp = dvp;
692 		vref(*vpp);
693 		goto out;
694 	}
695 
696 	pfs = VTOPFS(dvp);
697 	switch (pfs->pfs_type) {
698 	case Proot:
699 		if (cnp->cn_flags & CNP_ISDOTDOT)
700 			return (EIO);
701 
702 		if (CNEQ(cnp, "curproc", 7)) {
703 			error = procfs_allocvp(dvp->v_mount, vpp, 0, Pcurproc);
704 			goto out;
705 		}
706 
707 		pid = atopid(pname, cnp->cn_namelen);
708 		if (pid == NO_PID)
709 			break;
710 
711 		p = PFIND(pid);
712 		if (p == NULL)
713 			break;
714 
715 		if (!PRISON_CHECK(ap->a_cnp->cn_cred, p->p_ucred))
716 			break;
717 
718 		if (ps_showallprocs == 0 && ap->a_cnp->cn_cred->cr_uid != 0 &&
719 		    ap->a_cnp->cn_cred->cr_uid != p->p_ucred->cr_uid)
720 			break;
721 
722 		error = procfs_allocvp(dvp->v_mount, vpp, pid, Pproc);
723 		goto out;
724 
725 	case Pproc:
726 		if (cnp->cn_flags & CNP_ISDOTDOT) {
727 			error = procfs_root(dvp->v_mount, vpp);
728 			goto out;
729 		}
730 
731 		p = PFIND(pfs->pfs_pid);
732 		if (p == NULL)
733 			break;
734 
735 		if (!PRISON_CHECK(ap->a_cnp->cn_cred, p->p_ucred))
736 			break;
737 
738 		if (ps_showallprocs == 0 && ap->a_cnp->cn_cred->cr_uid != 0 &&
739 		    ap->a_cnp->cn_cred->cr_uid != p->p_ucred->cr_uid)
740 			break;
741 
742 		for (pt = proc_targets, i = 0; i < nproc_targets; pt++, i++) {
743 			if (cnp->cn_namelen == pt->pt_namlen &&
744 			    bcmp(pt->pt_name, pname, cnp->cn_namelen) == 0 &&
745 			    (pt->pt_valid == NULL || (*pt->pt_valid)(p)))
746 				goto found;
747 		}
748 		break;
749 	found:
750 		error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
751 					pt->pt_pfstype);
752 		goto out;
753 
754 	default:
755 		error = ENOTDIR;
756 		goto out;
757 	}
758 	if (cnp->cn_nameiop == NAMEI_LOOKUP)
759 		error = ENOENT;
760 	else
761 		error = EROFS;
762 	/*
763 	 * If no error occured *vpp will hold a referenced locked vnode.
764 	 * dvp was passed to us locked and *vpp must be returned locked.
765 	 * If *vpp != dvp then we should unlock dvp if (1) this is not the
766 	 * last component or (2) CNP_LOCKPARENT is not set.
767 	 */
768 out:
769 	if (error == 0 && *vpp != dvp) {
770 		if ((cnp->cn_flags & CNP_LOCKPARENT) == 0) {
771 			cnp->cn_flags |= CNP_PDIRUNLOCK;
772 			VOP_UNLOCK(dvp, 0, cnp->cn_td);
773 		}
774 	}
775 	return (error);
776 }
777 
778 /*
779  * Does this process have a text file?
780  */
781 int
782 procfs_validfile(struct proc *p)
783 {
784 	return (procfs_findtextvp(p) != NULLVP);
785 }
786 
787 /*
788  * readdir() returns directory entries from pfsnode (vp).
789  *
790  * We generate just one directory entry at a time, as it would probably
791  * not pay off to buffer several entries locally to save uiomove calls.
792  *
793  * procfs_readdir(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred,
794  *		  int *a_eofflag, int *a_ncookies, u_long **a_cookies)
795  */
796 static int
797 procfs_readdir(struct vop_readdir_args *ap)
798 {
799 	struct pfsnode *pfs;
800 	int error;
801 
802 	if (ap->a_uio->uio_offset < 0 || ap->a_uio->uio_offset > INT_MAX)
803 		return (EINVAL);
804 
805 	pfs = VTOPFS(ap->a_vp);
806 
807 	switch (pfs->pfs_type) {
808 	/*
809 	 * this is for the process-specific sub-directories.
810 	 * all that is needed to is copy out all the entries
811 	 * from the procent[] table (top of this file).
812 	 */
813 	case Pproc:
814 		error = procfs_readdir_proc(ap);
815 		break;
816 
817 	/*
818 	 * this is for the root of the procfs filesystem
819 	 * what is needed is a special entry for "curproc"
820 	 * followed by an entry for each process on allproc
821 	 */
822 
823 	case Proot:
824 		error = procfs_readdir_root(ap);
825 		break;
826 
827 	default:
828 		error = ENOTDIR;
829 		break;
830 	}
831 
832 	return (error);
833 }
834 
835 static int
836 procfs_readdir_proc(struct vop_readdir_args *ap)
837 {
838 	struct pfsnode *pfs;
839 	int error, i, retval;
840 	struct proc *p;
841 	struct proc_target *pt;
842 	struct uio *uio = ap->a_uio;
843 
844 	pfs = VTOPFS(ap->a_vp);
845 	p = PFIND(pfs->pfs_pid);
846 	if (p == NULL)
847 		return(0);
848 	if (!PRISON_CHECK(ap->a_cred, p->p_ucred))
849 		return(0);
850 
851 	error = 0;
852 	i = uio->uio_offset;
853 
854 	for (pt = &proc_targets[i];
855 	     !error && uio->uio_resid > 0 && i < nproc_targets; pt++, i++) {
856 		if (pt->pt_valid && (*pt->pt_valid)(p) == 0)
857 			continue;
858 
859 		retval = vop_write_dirent(&error, uio,
860 		    PROCFS_FILENO(pfs->pfs_pid, pt->pt_pfstype), pt->pt_type,
861 		    pt->pt_namlen, pt->pt_name);
862 		if (retval)
863 			break;
864 	}
865 
866 	uio->uio_offset = i;
867 
868 	return(0);
869 }
870 
871 static int
872 procfs_readdir_root(struct vop_readdir_args *ap)
873 {
874 	int error, i, pcnt, retval;
875 	struct uio *uio = ap->a_uio;
876 	volatile struct proc *p = LIST_FIRST(&allproc);
877 	ino_t d_ino;
878 	const char *d_name;
879 	char d_name_pid[20];
880 	size_t d_namlen;
881 	uint8_t d_type;
882 
883 	error = 0;
884 	pcnt = 0;
885 	i = uio->uio_offset;
886 
887 	for (; p && uio->uio_resid > 0 && !error; i++, pcnt++) {
888 		switch (i) {
889 		case 0:		/* `.' */
890 			d_ino = PROCFS_FILENO(0, Proot);
891 			d_name = ".";
892 			d_namlen = 1;
893 			d_type = DT_DIR;
894 			break;
895 		case 1:		/* `..' */
896 			d_ino = PROCFS_FILENO(0, Proot);
897 			d_name = "..";
898 			d_namlen = 2;
899 			d_type = DT_DIR;
900 			break;
901 
902 		case 2:
903 			d_ino = PROCFS_FILENO(0, Pcurproc);
904 			d_namlen = 7;
905 			d_name = "curproc";
906 			d_type = DT_LNK;
907 			break;
908 
909 
910 		default:
911 			while (pcnt < i) {
912 				p = LIST_NEXT(p, p_list);
913 				if (!p)
914 					goto done;
915 				if (!PRISON_CHECK(ap->a_cred, p->p_ucred))
916 					continue;
917 				pcnt++;
918 			}
919 			while (!PRISON_CHECK(ap->a_cred, p->p_ucred)) {
920 				p = LIST_NEXT(p, p_list);
921 				if (!p)
922 					goto done;
923 			}
924 			if (ps_showallprocs == 0 &&
925 			    ap->a_cred->cr_uid != 0 &&
926 			    ap->a_cred->cr_uid != p->p_ucred->cr_uid) {
927 				p = LIST_NEXT(p, p_list);
928 				if (!p)
929 					goto done;
930 				continue;
931 			}
932 
933 			d_ino = PROCFS_FILENO(p->p_pid, Pproc);
934 			d_namlen = snprintf(d_name_pid, sizeof(d_name_pid),
935 			    "%ld", (long)p->p_pid);
936 			d_name = d_name_pid;
937 			d_type = DT_DIR;
938 			p = LIST_NEXT(p, p_list);
939 			break;
940 		}
941 
942 		if (p)
943 		    PHOLD(p);
944 		retval = vop_write_dirent(&error, uio,
945 		    d_ino, d_type, d_namlen, d_name);
946 		if (p)
947 		    PRELE(p);
948 		if (retval)
949 			break;
950 	}
951 done:
952 	uio->uio_offset = i;
953 	return(0);
954 }
955 
956 /*
957  * readlink reads the link of `curproc' or `file'
958  */
959 static int
960 procfs_readlink(struct vop_readlink_args *ap)
961 {
962 	char buf[16];		/* should be enough */
963 	struct proc *procp;
964 	struct vnode *vp = ap->a_vp;
965 	struct pfsnode *pfs = VTOPFS(vp);
966 	char *fullpath, *freepath;
967 	int error, len;
968 
969 	switch (pfs->pfs_type) {
970 	case Pcurproc:
971 		if (pfs->pfs_fileno != PROCFS_FILENO(0, Pcurproc))
972 			return (EINVAL);
973 
974 		len = snprintf(buf, sizeof(buf), "%ld", (long)curproc->p_pid);
975 
976 		return (uiomove(buf, len, ap->a_uio));
977 	/*
978 	 * There _should_ be no way for an entire process to disappear
979 	 * from under us...
980 	 */
981 	case Pfile:
982 		procp = PFIND(pfs->pfs_pid);
983 		if (procp == NULL || procp->p_ucred == NULL) {
984 			printf("procfs_readlink: pid %d disappeared\n",
985 			    pfs->pfs_pid);
986 			return (uiomove("unknown", sizeof("unknown") - 1,
987 			    ap->a_uio));
988 		}
989 		error = vn_fullpath(procp, NULL, &fullpath, &freepath);
990 		if (error != 0)
991 			return (uiomove("unknown", sizeof("unknown") - 1,
992 			    ap->a_uio));
993 		error = uiomove(fullpath, strlen(fullpath), ap->a_uio);
994 		free(freepath, M_TEMP);
995 		return (error);
996 	default:
997 		return (EINVAL);
998 	}
999 }
1000 
1001 /*
1002  * convert decimal ascii to pid_t
1003  */
1004 static pid_t
1005 atopid(const char *b, u_int len)
1006 {
1007 	pid_t p = 0;
1008 
1009 	while (len--) {
1010 		char c = *b++;
1011 		if (c < '0' || c > '9')
1012 			return (NO_PID);
1013 		p = 10 * p + (c - '0');
1014 		if (p > PID_MAX)
1015 			return (NO_PID);
1016 	}
1017 
1018 	return (p);
1019 }
1020 
1021 /*
1022  * procfs vnode operations.
1023  */
1024 struct vnodeopv_entry_desc procfs_vnodeop_entries[] = {
1025 	{ &vop_default_desc,		vop_defaultop },
1026 	{ &vop_access_desc,		(vnodeopv_entry_t) procfs_access },
1027 	{ &vop_advlock_desc,		(vnodeopv_entry_t) procfs_badop },
1028 	{ &vop_bmap_desc,		(vnodeopv_entry_t) procfs_bmap },
1029 	{ &vop_close_desc,		(vnodeopv_entry_t) procfs_close },
1030 	{ &vop_old_create_desc,		(vnodeopv_entry_t) procfs_badop },
1031 	{ &vop_getattr_desc,		(vnodeopv_entry_t) procfs_getattr },
1032 	{ &vop_inactive_desc,		(vnodeopv_entry_t) procfs_inactive },
1033 	{ &vop_old_link_desc,		(vnodeopv_entry_t) procfs_badop },
1034 	{ &vop_old_lookup_desc,		(vnodeopv_entry_t) procfs_lookup },
1035 	{ &vop_old_mkdir_desc,		(vnodeopv_entry_t) procfs_badop },
1036 	{ &vop_old_mknod_desc,		(vnodeopv_entry_t) procfs_badop },
1037 	{ &vop_open_desc,		(vnodeopv_entry_t) procfs_open },
1038 	{ &vop_pathconf_desc,		(vnodeopv_entry_t) vop_stdpathconf },
1039 	{ &vop_print_desc,		(vnodeopv_entry_t) procfs_print },
1040 	{ &vop_read_desc,		(vnodeopv_entry_t) procfs_rw },
1041 	{ &vop_readdir_desc,		(vnodeopv_entry_t) procfs_readdir },
1042 	{ &vop_readlink_desc,		(vnodeopv_entry_t) procfs_readlink },
1043 	{ &vop_reclaim_desc,		(vnodeopv_entry_t) procfs_reclaim },
1044 	{ &vop_old_remove_desc,		(vnodeopv_entry_t) procfs_badop },
1045 	{ &vop_old_rename_desc,		(vnodeopv_entry_t) procfs_badop },
1046 	{ &vop_old_rmdir_desc,		(vnodeopv_entry_t) procfs_badop },
1047 	{ &vop_setattr_desc,		(vnodeopv_entry_t) procfs_setattr },
1048 	{ &vop_old_symlink_desc,	(vnodeopv_entry_t) procfs_badop },
1049 	{ &vop_write_desc,		(vnodeopv_entry_t) procfs_rw },
1050 	{ &vop_ioctl_desc,		(vnodeopv_entry_t) procfs_ioctl },
1051 	{ NULL, NULL }
1052 };
1053 
1054