xref: /dragonfly/sys/vfs/procfs/procfs_vnops.c (revision f9a239ec)
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 1.22 2004/11/12 00:09:47 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 	return (0);
353 }
354 
355 /*
356  * _reclaim is called when getnewvnode()
357  * wants to make use of an entry on the vnode
358  * free list.  at this time the filesystem needs
359  * to free any private data and remove the node
360  * from any private lists.
361  *
362  * procfs_reclaim(struct vnode *a_vp)
363  */
364 static int
365 procfs_reclaim(struct vop_reclaim_args *ap)
366 {
367 	return (procfs_freevp(ap->a_vp));
368 }
369 
370 /*
371  * _print is used for debugging.
372  * just print a readable description
373  * of (vp).
374  *
375  * procfs_print(struct vnode *a_vp)
376  */
377 static int
378 procfs_print(struct vop_print_args *ap)
379 {
380 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
381 
382 	printf("tag VT_PROCFS, type %d, pid %ld, mode %x, flags %lx\n",
383 	    pfs->pfs_type, (long)pfs->pfs_pid, pfs->pfs_mode, pfs->pfs_flags);
384 	return (0);
385 }
386 
387 /*
388  * generic entry point for unsupported operations
389  */
390 static int
391 procfs_badop(void)
392 {
393 	return (EIO);
394 }
395 
396 /*
397  * Invent attributes for pfsnode (vp) and store
398  * them in (vap).
399  * Directories lengths are returned as zero since
400  * any real length would require the genuine size
401  * to be computed, and nothing cares anyway.
402  *
403  * this is relatively minimal for procfs.
404  *
405  * procfs_getattr(struct vnode *a_vp, struct vattr *a_vap,
406  *		  struct ucred *a_cred,	struct thread *a_td)
407  */
408 static int
409 procfs_getattr(struct vop_getattr_args *ap)
410 {
411 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
412 	struct vattr *vap = ap->a_vap;
413 	struct proc *procp;
414 	int error;
415 
416 	/*
417 	 * First make sure that the process and its credentials
418 	 * still exist.
419 	 */
420 	switch (pfs->pfs_type) {
421 	case Proot:
422 	case Pcurproc:
423 		procp = 0;
424 		break;
425 
426 	default:
427 		procp = PFIND(pfs->pfs_pid);
428 		if (procp == NULL || procp->p_ucred == NULL)
429 			return (ENOENT);
430 	}
431 
432 	error = 0;
433 
434 	/* start by zeroing out the attributes */
435 	VATTR_NULL(vap);
436 
437 	/* next do all the common fields */
438 	vap->va_type = ap->a_vp->v_type;
439 	vap->va_mode = pfs->pfs_mode;
440 	vap->va_fileid = pfs->pfs_fileno;
441 	vap->va_flags = 0;
442 	vap->va_blocksize = PAGE_SIZE;
443 	vap->va_bytes = vap->va_size = 0;
444 	vap->va_fsid = ap->a_vp->v_mount->mnt_stat.f_fsid.val[0];
445 
446 	/*
447 	 * Make all times be current TOD.
448 	 * It would be possible to get the process start
449 	 * time from the p_stat structure, but there's
450 	 * no "file creation" time stamp anyway, and the
451 	 * p_stat structure is not addressible if u. gets
452 	 * swapped out for that process.
453 	 */
454 	nanotime(&vap->va_ctime);
455 	vap->va_atime = vap->va_mtime = vap->va_ctime;
456 
457 	/*
458 	 * If the process has exercised some setuid or setgid
459 	 * privilege, then rip away read/write permission so
460 	 * that only root can gain access.
461 	 */
462 	switch (pfs->pfs_type) {
463 	case Pctl:
464 	case Pregs:
465 	case Pfpregs:
466 	case Pdbregs:
467 	case Pmem:
468 		if (procp->p_flag & P_SUGID)
469 			vap->va_mode &= ~((VREAD|VWRITE)|
470 					  ((VREAD|VWRITE)>>3)|
471 					  ((VREAD|VWRITE)>>6));
472 		break;
473 	default:
474 		break;
475 	}
476 
477 	/*
478 	 * now do the object specific fields
479 	 *
480 	 * The size could be set from struct reg, but it's hardly
481 	 * worth the trouble, and it puts some (potentially) machine
482 	 * dependent data into this machine-independent code.  If it
483 	 * becomes important then this function should break out into
484 	 * a per-file stat function in the corresponding .c file.
485 	 */
486 
487 	vap->va_nlink = 1;
488 	if (procp) {
489 		vap->va_uid = procp->p_ucred->cr_uid;
490 		vap->va_gid = procp->p_ucred->cr_gid;
491 	}
492 
493 	switch (pfs->pfs_type) {
494 	case Proot:
495 		/*
496 		 * Set nlink to 1 to tell fts(3) we don't actually know.
497 		 */
498 		vap->va_nlink = 1;
499 		vap->va_uid = 0;
500 		vap->va_gid = 0;
501 		vap->va_size = vap->va_bytes = DEV_BSIZE;
502 		break;
503 
504 	case Pcurproc: {
505 		char buf[16];		/* should be enough */
506 		vap->va_uid = 0;
507 		vap->va_gid = 0;
508 		vap->va_size = vap->va_bytes =
509 		    snprintf(buf, sizeof(buf), "%ld", (long)curproc->p_pid);
510 		break;
511 	}
512 
513 	case Pproc:
514 		vap->va_nlink = nproc_targets;
515 		vap->va_size = vap->va_bytes = DEV_BSIZE;
516 		break;
517 
518 	case Pfile: {
519 		char *fullpath, *freepath;
520 		error = vn_fullpath(procp, NULL, &fullpath, &freepath);
521 		if (error == 0) {
522 			vap->va_size = strlen(fullpath);
523 			free(freepath, M_TEMP);
524 		} else {
525 			vap->va_size = sizeof("unknown") - 1;
526 			error = 0;
527 		}
528 		vap->va_bytes = vap->va_size;
529 		break;
530 	}
531 
532 	case Pmem:
533 		/*
534 		 * If we denied owner access earlier, then we have to
535 		 * change the owner to root - otherwise 'ps' and friends
536 		 * will break even though they are setgid kmem. *SIGH*
537 		 */
538 		if (procp->p_flag & P_SUGID)
539 			vap->va_uid = 0;
540 		else
541 			vap->va_uid = procp->p_ucred->cr_uid;
542 		break;
543 
544 	case Pregs:
545 		vap->va_bytes = vap->va_size = sizeof(struct reg);
546 		break;
547 
548 	case Pfpregs:
549 		vap->va_bytes = vap->va_size = sizeof(struct fpreg);
550 		break;
551 
552         case Pdbregs:
553                 vap->va_bytes = vap->va_size = sizeof(struct dbreg);
554                 break;
555 
556 	case Ptype:
557 	case Pmap:
558 	case Pctl:
559 	case Pstatus:
560 	case Pnote:
561 	case Pnotepg:
562 	case Pcmdline:
563 	case Prlimit:
564 		break;
565 
566 	default:
567 		panic("procfs_getattr");
568 	}
569 
570 	return (error);
571 }
572 
573 /*
574  * procfs_setattr(struct vnode *a_vp, struct vattr *a_vap,
575  *		  struct ucred *a_cred,	struct thread *a_td)
576  */
577 static int
578 procfs_setattr(struct vop_setattr_args *ap)
579 {
580 	if (ap->a_vap->va_flags != VNOVAL)
581 		return (EOPNOTSUPP);
582 
583 	/*
584 	 * just fake out attribute setting
585 	 * it's not good to generate an error
586 	 * return, otherwise things like creat()
587 	 * will fail when they try to set the
588 	 * file length to 0.  worse, this means
589 	 * that echo $note > /proc/$pid/note will fail.
590 	 */
591 
592 	return (0);
593 }
594 
595 /*
596  * implement access checking.
597  *
598  * something very similar to this code is duplicated
599  * throughout the 4bsd kernel and should be moved
600  * into kern/vfs_subr.c sometime.
601  *
602  * actually, the check for super-user is slightly
603  * broken since it will allow read access to write-only
604  * objects.  this doesn't cause any particular trouble
605  * but does mean that the i/o entry points need to check
606  * that the operation really does make sense.
607  *
608  * procfs_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred,
609  *		 struct thread *a_td)
610  */
611 static int
612 procfs_access(struct vop_access_args *ap)
613 {
614 	struct vattr *vap;
615 	struct vattr vattr;
616 	int error;
617 
618 	/*
619 	 * If you're the super-user,
620 	 * you always get access.
621 	 */
622 	if (ap->a_cred->cr_uid == 0)
623 		return (0);
624 
625 	vap = &vattr;
626 	error = VOP_GETATTR(ap->a_vp, vap, ap->a_td);
627 	if (error)
628 		return (error);
629 
630 	/*
631 	 * Access check is based on only one of owner, group, public.
632 	 * If not owner, then check group. If not a member of the
633 	 * group, then check public access.
634 	 */
635 	if (ap->a_cred->cr_uid != vap->va_uid) {
636 		gid_t *gp;
637 		int i;
638 
639 		ap->a_mode >>= 3;
640 		gp = ap->a_cred->cr_groups;
641 		for (i = 0; i < ap->a_cred->cr_ngroups; i++, gp++)
642 			if (vap->va_gid == *gp)
643 				goto found;
644 		ap->a_mode >>= 3;
645 found:
646 		;
647 	}
648 
649 	if ((vap->va_mode & ap->a_mode) == ap->a_mode)
650 		return (0);
651 
652 	return (EACCES);
653 }
654 
655 /*
656  * lookup.  this is incredibly complicated in the general case, however
657  * for most pseudo-filesystems very little needs to be done.
658  *
659  * procfs_lookup(struct vnode *a_dvp, struct vnode **a_vpp,
660  *		 struct componentname *a_cnp)
661  */
662 static int
663 procfs_lookup(struct vop_lookup_args *ap)
664 {
665 	struct componentname *cnp = ap->a_cnp;
666 	struct vnode **vpp = ap->a_vpp;
667 	struct vnode *dvp = ap->a_dvp;
668 	char *pname = cnp->cn_nameptr;
669 	/* struct proc *curp = cnp->cn_proc; */
670 	struct proc_target *pt;
671 	pid_t pid;
672 	struct pfsnode *pfs;
673 	struct proc *p;
674 	int i;
675 	int error;
676 
677 	*vpp = NULL;
678 
679 	if (cnp->cn_nameiop == NAMEI_DELETE || cnp->cn_nameiop == NAMEI_RENAME)
680 		return (EROFS);
681 
682 	error = 0;
683 	if (cnp->cn_namelen == 1 && *pname == '.') {
684 		*vpp = dvp;
685 		vref(*vpp);
686 		goto out;
687 	}
688 
689 	pfs = VTOPFS(dvp);
690 	switch (pfs->pfs_type) {
691 	case Proot:
692 		if (cnp->cn_flags & CNP_ISDOTDOT)
693 			return (EIO);
694 
695 		if (CNEQ(cnp, "curproc", 7)) {
696 			error = procfs_allocvp(dvp->v_mount, vpp, 0, Pcurproc);
697 			goto out;
698 		}
699 
700 		pid = atopid(pname, cnp->cn_namelen);
701 		if (pid == NO_PID)
702 			break;
703 
704 		p = PFIND(pid);
705 		if (p == NULL)
706 			break;
707 
708 		if (ps_showallprocs == 0 && ap->a_cnp->cn_cred->cr_uid != 0 &&
709 		    ap->a_cnp->cn_cred->cr_uid != p->p_ucred->cr_uid)
710 			break;
711 
712 		error = procfs_allocvp(dvp->v_mount, vpp, pid, Pproc);
713 		goto out;
714 
715 	case Pproc:
716 		if (cnp->cn_flags & CNP_ISDOTDOT) {
717 			error = procfs_root(dvp->v_mount, vpp);
718 			goto out;
719 		}
720 
721 		p = PFIND(pfs->pfs_pid);
722 		if (p == NULL)
723 			break;
724 
725 		if (ps_showallprocs == 0 && ap->a_cnp->cn_cred->cr_uid != 0 &&
726 		    ap->a_cnp->cn_cred->cr_uid != p->p_ucred->cr_uid)
727 			break;
728 
729 		for (pt = proc_targets, i = 0; i < nproc_targets; pt++, i++) {
730 			if (cnp->cn_namelen == pt->pt_namlen &&
731 			    bcmp(pt->pt_name, pname, cnp->cn_namelen) == 0 &&
732 			    (pt->pt_valid == NULL || (*pt->pt_valid)(p)))
733 				goto found;
734 		}
735 		break;
736 	found:
737 		error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
738 					pt->pt_pfstype);
739 		goto out;
740 
741 	default:
742 		error = ENOTDIR;
743 		goto out;
744 	}
745 	if (cnp->cn_nameiop == NAMEI_LOOKUP)
746 		error = ENOENT;
747 	else
748 		error = EROFS;
749 	/*
750 	 * If no error occured *vpp will hold a referenced locked vnode.
751 	 * dvp was passed to us locked and *vpp must be returned locked.
752 	 * If *vpp != dvp then we should unlock dvp if (1) this is not the
753 	 * last component or (2) CNP_LOCKPARENT is not set.
754 	 */
755 out:
756 	if (error == 0 && *vpp != dvp) {
757 		if ((cnp->cn_flags & CNP_LOCKPARENT) == 0) {
758 			cnp->cn_flags |= CNP_PDIRUNLOCK;
759 			VOP_UNLOCK(dvp, 0, cnp->cn_td);
760 		}
761 	}
762 	return (error);
763 }
764 
765 /*
766  * Does this process have a text file?
767  */
768 int
769 procfs_validfile(struct proc *p)
770 {
771 	return (procfs_findtextvp(p) != NULLVP);
772 }
773 
774 /*
775  * readdir() returns directory entries from pfsnode (vp).
776  *
777  * We generate just one directory entry at a time, as it would probably
778  * not pay off to buffer several entries locally to save uiomove calls.
779  *
780  * procfs_readdir(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred,
781  *		  int *a_eofflag, int *a_ncookies, u_long **a_cookies)
782  */
783 static int
784 procfs_readdir(struct vop_readdir_args *ap)
785 {
786 	struct uio *uio = ap->a_uio;
787 	struct dirent d;
788 	struct dirent *dp = &d;
789 	struct pfsnode *pfs;
790 	int count, error, i, off;
791 	static u_int delen;
792 
793 	if (!delen) {
794 
795 		d.d_namlen = PROCFS_NAMELEN;
796 		delen = GENERIC_DIRSIZ(&d);
797 	}
798 
799 	pfs = VTOPFS(ap->a_vp);
800 
801 	off = (int)uio->uio_offset;
802 	if (off != uio->uio_offset || off < 0 ||
803 	    off % delen != 0 || uio->uio_resid < delen)
804 		return (EINVAL);
805 
806 	error = 0;
807 	count = 0;
808 	i = off / delen;
809 
810 	switch (pfs->pfs_type) {
811 	/*
812 	 * this is for the process-specific sub-directories.
813 	 * all that is needed to is copy out all the entries
814 	 * from the procent[] table (top of this file).
815 	 */
816 	case Pproc: {
817 		struct proc *p;
818 		struct proc_target *pt;
819 
820 		p = PFIND(pfs->pfs_pid);
821 		if (p == NULL)
822 			break;
823 		if (!PRISON_CHECK(ap->a_cred, p->p_ucred))
824 			break;
825 
826 		for (pt = &proc_targets[i];
827 		     uio->uio_resid >= delen && i < nproc_targets; pt++, i++) {
828 			if (pt->pt_valid && (*pt->pt_valid)(p) == 0)
829 				continue;
830 
831 			dp->d_reclen = delen;
832 			dp->d_fileno = PROCFS_FILENO(pfs->pfs_pid, pt->pt_pfstype);
833 			dp->d_namlen = pt->pt_namlen;
834 			bcopy(pt->pt_name, dp->d_name, pt->pt_namlen + 1);
835 			dp->d_type = pt->pt_type;
836 
837 			if ((error = uiomove((caddr_t)dp, delen, uio)) != 0)
838 				break;
839 		}
840 
841 	    	break;
842 	    }
843 
844 	/*
845 	 * this is for the root of the procfs filesystem
846 	 * what is needed is a special entry for "curproc"
847 	 * followed by an entry for each process on allproc
848 #ifdef PROCFS_ZOMBIE
849 	 * and zombproc.
850 #endif
851 	 */
852 
853 	case Proot: {
854 #ifdef PROCFS_ZOMBIE
855 		int doingzomb = 0;
856 #endif
857 		int pcnt = 0;
858 		volatile struct proc *p = allproc.lh_first;
859 
860 		for (; p && uio->uio_resid >= delen; i++, pcnt++) {
861 			bzero((char *) dp, delen);
862 			dp->d_reclen = delen;
863 
864 			switch (i) {
865 			case 0:		/* `.' */
866 			case 1:		/* `..' */
867 				dp->d_fileno = PROCFS_FILENO(0, Proot);
868 				dp->d_namlen = i + 1;
869 				bcopy("..", dp->d_name, dp->d_namlen);
870 				dp->d_name[i + 1] = '\0';
871 				dp->d_type = DT_DIR;
872 				break;
873 
874 			case 2:
875 				dp->d_fileno = PROCFS_FILENO(0, Pcurproc);
876 				dp->d_namlen = 7;
877 				bcopy("curproc", dp->d_name, 8);
878 				dp->d_type = DT_LNK;
879 				break;
880 
881 			default:
882 				while (pcnt < i) {
883 					p = p->p_list.le_next;
884 					if (!p)
885 						goto done;
886 					if (!PRISON_CHECK(ap->a_cred, p->p_ucred))
887 						continue;
888 					pcnt++;
889 				}
890 				while (!PRISON_CHECK(ap->a_cred, p->p_ucred)) {
891 					p = p->p_list.le_next;
892 					if (!p)
893 						goto done;
894 				}
895 				if (ps_showallprocs == 0 &&
896 				    ap->a_cred->cr_uid != 0 &&
897 				    ap->a_cred->cr_uid !=
898 				    p->p_ucred->cr_uid) {
899 					p = p->p_list.le_next;
900 					if (!p)
901 						goto done;
902 					break;
903 				}
904 
905 				dp->d_fileno = PROCFS_FILENO(p->p_pid, Pproc);
906 				dp->d_namlen = sprintf(dp->d_name, "%ld",
907 				    (long)p->p_pid);
908 				dp->d_type = DT_DIR;
909 				p = p->p_list.le_next;
910 				break;
911 			}
912 
913 			if ((error = uiomove((caddr_t)dp, delen, uio)) != 0)
914 				break;
915 		}
916 	done:
917 
918 #ifdef PROCFS_ZOMBIE
919 		if (p == NULL && doingzomb == 0) {
920 			doingzomb = 1;
921 			p = zombproc.lh_first;
922 			goto again;
923 		}
924 #endif
925 
926 		break;
927 
928 	    }
929 
930 	default:
931 		error = ENOTDIR;
932 		break;
933 	}
934 
935 	uio->uio_offset = i * delen;
936 
937 	return (error);
938 }
939 
940 /*
941  * readlink reads the link of `curproc' or `file'
942  */
943 static int
944 procfs_readlink(struct vop_readlink_args *ap)
945 {
946 	char buf[16];		/* should be enough */
947 	struct proc *procp;
948 	struct vnode *vp = ap->a_vp;
949 	struct pfsnode *pfs = VTOPFS(vp);
950 	char *fullpath, *freepath;
951 	int error, len;
952 
953 	switch (pfs->pfs_type) {
954 	case Pcurproc:
955 		if (pfs->pfs_fileno != PROCFS_FILENO(0, Pcurproc))
956 			return (EINVAL);
957 
958 		len = snprintf(buf, sizeof(buf), "%ld", (long)curproc->p_pid);
959 
960 		return (uiomove(buf, len, ap->a_uio));
961 	/*
962 	 * There _should_ be no way for an entire process to disappear
963 	 * from under us...
964 	 */
965 	case Pfile:
966 		procp = PFIND(pfs->pfs_pid);
967 		if (procp == NULL || procp->p_ucred == NULL) {
968 			printf("procfs_readlink: pid %d disappeared\n",
969 			    pfs->pfs_pid);
970 			return (uiomove("unknown", sizeof("unknown") - 1,
971 			    ap->a_uio));
972 		}
973 		error = vn_fullpath(procp, NULL, &fullpath, &freepath);
974 		if (error != 0)
975 			return (uiomove("unknown", sizeof("unknown") - 1,
976 			    ap->a_uio));
977 		error = uiomove(fullpath, strlen(fullpath), ap->a_uio);
978 		free(freepath, M_TEMP);
979 		return (error);
980 	default:
981 		return (EINVAL);
982 	}
983 }
984 
985 /*
986  * convert decimal ascii to pid_t
987  */
988 static pid_t
989 atopid(const char *b, u_int len)
990 {
991 	pid_t p = 0;
992 
993 	while (len--) {
994 		char c = *b++;
995 		if (c < '0' || c > '9')
996 			return (NO_PID);
997 		p = 10 * p + (c - '0');
998 		if (p > PID_MAX)
999 			return (NO_PID);
1000 	}
1001 
1002 	return (p);
1003 }
1004 
1005 /*
1006  * procfs vnode operations.
1007  */
1008 struct vnodeopv_entry_desc procfs_vnodeop_entries[] = {
1009 	{ &vop_default_desc,		vop_defaultop },
1010 	{ &vop_access_desc,		(void *) procfs_access },
1011 	{ &vop_advlock_desc,		(void *) procfs_badop },
1012 	{ &vop_bmap_desc,		(void *) procfs_bmap },
1013 	{ &vop_close_desc,		(void *) procfs_close },
1014 	{ &vop_create_desc,		(void *) procfs_badop },
1015 	{ &vop_getattr_desc,		(void *) procfs_getattr },
1016 	{ &vop_inactive_desc,		(void *) procfs_inactive },
1017 	{ &vop_link_desc,		(void *) procfs_badop },
1018 	{ &vop_lookup_desc,		(void *) procfs_lookup },
1019 	{ &vop_mkdir_desc,		(void *) procfs_badop },
1020 	{ &vop_mknod_desc,		(void *) procfs_badop },
1021 	{ &vop_open_desc,		(void *) procfs_open },
1022 	{ &vop_pathconf_desc,		(void *) vop_stdpathconf },
1023 	{ &vop_print_desc,		(void *) procfs_print },
1024 	{ &vop_read_desc,		(void *) procfs_rw },
1025 	{ &vop_readdir_desc,		(void *) procfs_readdir },
1026 	{ &vop_readlink_desc,		(void *) procfs_readlink },
1027 	{ &vop_reclaim_desc,		(void *) procfs_reclaim },
1028 	{ &vop_remove_desc,		(void *) procfs_badop },
1029 	{ &vop_rename_desc,		(void *) procfs_badop },
1030 	{ &vop_rmdir_desc,		(void *) procfs_badop },
1031 	{ &vop_setattr_desc,		(void *) procfs_setattr },
1032 	{ &vop_symlink_desc,		(void *) procfs_badop },
1033 	{ &vop_write_desc,		(void *) procfs_rw },
1034 	{ &vop_ioctl_desc,		(void *) procfs_ioctl },
1035 	{ NULL, NULL }
1036 };
1037 
1038