xref: /dflybsd-src/sys/vfs/procfs/procfs_vnops.c (revision 211d4362597aee676ecea315377d5cb13da26bb5)
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  */
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/lock.h>
51 #include <sys/fcntl.h>
52 #include <sys/proc.h>
53 #include <sys/priv.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 <sys/reg.h>
62 #include <vm/vm_zone.h>
63 #include <vfs/procfs/procfs.h>
64 #include <sys/pioctl.h>
65 
66 #include <sys/spinlock2.h>
67 
68 #include <machine/limits.h>
69 
70 static int	procfs_access (struct vop_access_args *);
71 static int	procfs_badop (struct vop_generic_args *);
72 static int	procfs_bmap (struct vop_bmap_args *);
73 static int	procfs_close (struct vop_close_args *);
74 static int	procfs_getattr (struct vop_getattr_args *);
75 static int	procfs_inactive (struct vop_inactive_args *);
76 static int	procfs_ioctl (struct vop_ioctl_args *);
77 static int	procfs_lookup (struct vop_old_lookup_args *);
78 static int	procfs_open (struct vop_open_args *);
79 static int	procfs_print (struct vop_print_args *);
80 static int	procfs_readdir (struct vop_readdir_args *);
81 static int	procfs_readlink (struct vop_readlink_args *);
82 static int	procfs_reclaim (struct vop_reclaim_args *);
83 static int	procfs_setattr (struct vop_setattr_args *);
84 
85 static int	procfs_readdir_proc(struct vop_readdir_args *);
86 static int	procfs_readdir_root(struct vop_readdir_args *);
87 
88 /*
89  * procfs vnode operations.
90  */
91 struct vop_ops procfs_vnode_vops = {
92 	.vop_default =		vop_defaultop,
93 	.vop_access =		procfs_access,
94 	.vop_advlock =		(void *)procfs_badop,
95 	.vop_bmap =		procfs_bmap,
96 	.vop_close =		procfs_close,
97 	.vop_old_create =	(void *)procfs_badop,
98 	.vop_getattr =		procfs_getattr,
99 	.vop_inactive =		procfs_inactive,
100 	.vop_old_link =		(void *)procfs_badop,
101 	.vop_old_lookup =	procfs_lookup,
102 	.vop_old_mkdir =	(void *)procfs_badop,
103 	.vop_old_mknod =	(void *)procfs_badop,
104 	.vop_open =		procfs_open,
105 	.vop_pathconf =		vop_stdpathconf,
106 	.vop_print =		procfs_print,
107 	.vop_read =		procfs_rw,
108 	.vop_readdir =		procfs_readdir,
109 	.vop_readlink =		procfs_readlink,
110 	.vop_reclaim =		procfs_reclaim,
111 	.vop_old_remove =	(void *)procfs_badop,
112 	.vop_old_rename =	(void *)procfs_badop,
113 	.vop_old_rmdir =	(void *)procfs_badop,
114 	.vop_setattr =		procfs_setattr,
115 	.vop_old_symlink =	(void *)procfs_badop,
116 	.vop_write =		(void *)procfs_rw,
117 	.vop_ioctl =		procfs_ioctl
118 };
119 
120 
121 /*
122  * This is a list of the valid names in the
123  * process-specific sub-directories.  It is
124  * used in procfs_lookup and procfs_readdir
125  */
126 static struct proc_target {
127 	u_char	pt_type;
128 	u_char	pt_namlen;
129 	char	*pt_name;
130 	pfstype	pt_pfstype;
131 	int	(*pt_valid) (struct lwp *p);
132 } proc_targets[] = {
133 #define N(s) sizeof(s)-1, s
134 	/*	  name		type		validp */
135 	{ DT_DIR, N("."),	Pproc,		NULL },
136 	{ DT_DIR, N(".."),	Proot,		NULL },
137 	{ DT_REG, N("mem"),	Pmem,		NULL },
138 	{ DT_REG, N("regs"),	Pregs,		procfs_validregs },
139 	{ DT_REG, N("fpregs"),	Pfpregs,	procfs_validfpregs },
140 	{ DT_REG, N("dbregs"),	Pdbregs,	procfs_validdbregs },
141 	{ DT_REG, N("ctl"),	Pctl,		NULL },
142 	{ DT_REG, N("status"),	Pstatus,	NULL },
143 	{ DT_REG, N("note"),	Pnote,		NULL },
144 	{ DT_REG, N("notepg"),	Pnotepg,	NULL },
145 	{ DT_REG, N("map"), 	Pmap,		procfs_validmap },
146 	{ DT_REG, N("etype"),	Ptype,		procfs_validtype },
147 	{ DT_REG, N("cmdline"),	Pcmdline,	NULL },
148 	{ DT_REG, N("rlimit"),	Prlimit,	NULL },
149 	{ DT_LNK, N("file"),	Pfile,		NULL },
150 #undef N
151 };
152 static const int nproc_targets = NELEM(proc_targets);
153 
154 static pid_t atopid (const char *, u_int);
155 
156 /*
157  * set things up for doing i/o on
158  * the pfsnode (vp).  (vp) is locked
159  * on entry, and should be left locked
160  * on exit.
161  *
162  * for procfs we don't need to do anything
163  * in particular for i/o.  all that is done
164  * is to support exclusive open on process
165  * memory images.
166  *
167  * procfs_open(struct vnode *a_vp, int a_mode, struct ucred *a_cred,
168  *	       struct file *a_fp)
169  */
170 static int
171 procfs_open(struct vop_open_args *ap)
172 {
173 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
174 	struct proc *p1, *p2;
175 	int error;
176 
177 	p2 = pfs_pfind(pfs->pfs_pid);
178 	if (p2 == NULL)
179 		return (ENOENT);
180 	if (pfs->pfs_pid && !PRISON_CHECK(ap->a_cred, p2->p_ucred)) {
181 		error = ENOENT;
182 		goto done;
183 	}
184 
185 	switch (pfs->pfs_type) {
186 	case Pmem:
187 		if (((pfs->pfs_flags & FWRITE) && (ap->a_mode & O_EXCL)) ||
188 		    ((pfs->pfs_flags & O_EXCL) && (ap->a_mode & FWRITE))) {
189 			error = EBUSY;
190 			goto done;
191 		}
192 
193 		p1 = curproc;
194 		KKASSERT(p1);
195 		/* Can't trace a process that's currently exec'ing. */
196 		if ((p2->p_flags & P_INEXEC) != 0) {
197 			error = EAGAIN;
198 			goto done;
199 		}
200 		if (!CHECKIO(p1, p2) || p_trespass(ap->a_cred, p2->p_ucred)) {
201 			error = EPERM;
202 			goto done;
203 		}
204 
205 		if (ap->a_mode & FWRITE)
206 			pfs->pfs_flags = ap->a_mode & (FWRITE|O_EXCL);
207 
208 		break;
209 
210 	default:
211 		break;
212 	}
213 	error = vop_stdopen(ap);
214 done:
215 	pfs_pdone(p2);
216 	return error;
217 }
218 
219 /*
220  * close the pfsnode (vp) after doing i/o.
221  * (vp) is not locked on entry or exit.
222  *
223  * nothing to do for procfs other than undo
224  * any exclusive open flag (see _open above).
225  *
226  * procfs_close(struct vnode *a_vp, int a_fflag, struct ucred *a_cred)
227  */
228 static int
229 procfs_close(struct vop_close_args *ap)
230 {
231 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
232 	struct proc *p;
233 
234 	switch (pfs->pfs_type) {
235 	case Pmem:
236 		if ((ap->a_fflag & FWRITE) && (pfs->pfs_flags & O_EXCL))
237 			pfs->pfs_flags &= ~(FWRITE|O_EXCL);
238 		/*
239 		 * v_opencount determines the last real close on the vnode.
240 		 *
241 		 * If this is the last close, then it checks to see if
242 		 * the target process has PF_LINGER set in p_pfsflags,
243 		 * if this is *not* the case, then the process' stop flags
244 		 * are cleared, and the process is woken up.  This is
245 		 * to help prevent the case where a process has been
246 		 * told to stop on an event, but then the requesting process
247 		 * has gone away or forgotten about it.
248 		 */
249 		p = NULL;
250 		if ((ap->a_vp->v_opencount < 2)
251 		    && ((p = pfs_pfind(pfs->pfs_pid)) != NULL ||
252 		        (p = pfs_zpfind(pfs->pfs_pid)) != NULL)
253 		    && !(p->p_pfsflags & PF_LINGER)) {
254 			spin_lock(&p->p_spin);
255 			p->p_stops = 0;
256 			p->p_step = 0;
257 			spin_unlock(&p->p_spin);
258 			wakeup(&p->p_stype);
259 		}
260 		pfs_pdone(p);
261 		break;
262 	default:
263 		break;
264 	}
265 
266 	return (vop_stdclose(ap));
267 }
268 
269 /*
270  * do an ioctl operation on a pfsnode (vp).
271  * (vp) is not locked on entry or exit.
272  */
273 static int
274 procfs_ioctl(struct vop_ioctl_args *ap)
275 {
276 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
277 	struct proc *procp;
278 	struct proc *p;
279 	int error;
280 	int signo;
281 	struct procfs_status *psp;
282 	unsigned char flags;
283 
284 	procp = pfs_pfind(pfs->pfs_pid);
285 	if (procp == NULL)
286 		return ENOTTY;
287 	p = curproc;
288 	if (p == NULL) {
289 		error = EINVAL;
290 		goto done;
291 	}
292 
293 	/* Can't trace a process that's currently exec'ing. */
294 	if ((procp->p_flags & P_INEXEC) != 0) {
295 		error = EAGAIN;
296 		goto done;
297 	}
298 	if (!CHECKIO(p, procp) || p_trespass(ap->a_cred, procp->p_ucred)) {
299 		error = EPERM;
300 		goto done;
301 	}
302 
303 	switch (ap->a_command) {
304 	case PIOCBIS:
305 	  spin_lock(&procp->p_spin);
306 	  procp->p_stops |= *(unsigned int*)ap->a_data;
307 	  spin_unlock(&procp->p_spin);
308 	  break;
309 	case PIOCBIC:
310 	  spin_lock(&procp->p_spin);
311 	  procp->p_stops &= ~*(unsigned int*)ap->a_data;
312 	  spin_unlock(&procp->p_spin);
313 	  break;
314 	case PIOCSFL:
315 	  /*
316 	   * NFLAGS is "non-suser_xxx flags" -- currently, only
317 	   * PFS_ISUGID ("ignore set u/g id");
318 	   */
319 #define NFLAGS	(PF_ISUGID)
320 	  flags = (unsigned char)*(unsigned int*)ap->a_data;
321 	  if (flags & NFLAGS && (error = priv_check_cred(ap->a_cred, PRIV_ROOT, 0)))
322 	    goto done;
323 	  procp->p_pfsflags = flags;
324 	  break;
325 	case PIOCGFL:
326 	  *(unsigned int*)ap->a_data = (unsigned int)procp->p_pfsflags;
327 	  break;
328 	case PIOCSTATUS:
329 	  /*
330 	   * NOTE: syscall entry deals with stopevents and may run without
331 	   *	   the MP lock.
332 	   */
333 	  psp = (struct procfs_status *)ap->a_data;
334 	  psp->flags = procp->p_pfsflags;
335 	  psp->events = procp->p_stops;
336 	  spin_lock(&procp->p_spin);
337 	  if (procp->p_step) {
338 	    psp->state = 0;
339 	    psp->why = procp->p_stype;
340 	    psp->val = procp->p_xstat;
341 	    spin_unlock(&procp->p_spin);
342 	  } else {
343 	    psp->state = 1;
344 	    spin_unlock(&procp->p_spin);
345 	    psp->why = 0;	/* Not defined values */
346 	    psp->val = 0;	/* Not defined values */
347 	  }
348 	  break;
349 	case PIOCWAIT:
350 	  /*
351 	   * NOTE: syscall entry deals with stopevents and may run without
352 	   *	   the MP lock.
353 	   */
354 	  psp = (struct procfs_status *)ap->a_data;
355 	  spin_lock(&procp->p_spin);
356 	  while (procp->p_step == 0) {
357 	    tsleep_interlock(&procp->p_stype, PCATCH);
358 	    spin_unlock(&procp->p_spin);
359 	    if (procp->p_stops == 0) {
360 		error = EINVAL;
361 		goto done;
362 	    }
363 	    if (procp->p_flags & P_POSTEXIT) {
364 		error = EINVAL;
365 		goto done;
366 	    }
367 	    if (procp->p_flags & P_INEXEC) {
368 		error = EAGAIN;
369 		goto done;
370 	    }
371 	    error = tsleep(&procp->p_stype, PCATCH | PINTERLOCKED,
372 			   "piocwait", 0);
373 	    if (error)
374 	      goto done;
375 	    spin_lock(&procp->p_spin);
376 	  }
377 	  spin_unlock(&procp->p_spin);
378 	  psp->state = 1;	/* It stopped */
379 	  psp->flags = procp->p_pfsflags;
380 	  psp->events = procp->p_stops;
381 	  psp->why = procp->p_stype;	/* why it stopped */
382 	  psp->val = procp->p_xstat;	/* any extra info */
383 	  break;
384 	case PIOCCONT:	/* Restart a proc */
385 	  /*
386 	   * NOTE: syscall entry deals with stopevents and may run without
387 	   *	   the MP lock.  However, the caller is presumably interlocked
388 	   *	   by having waited.
389 	   */
390 	  if (procp->p_step == 0) {
391 	    error = EINVAL;	/* Can only start a stopped process */
392 	    goto done;
393 	  }
394 	  if ((signo = *(int*)ap->a_data) != 0) {
395 	    if (signo >= NSIG || signo <= 0) {
396 	      error = EINVAL;
397 	      goto done;
398 	    }
399 	    ksignal(procp, signo);
400 	  }
401 	  procp->p_step = 0;
402 	  wakeup(&procp->p_step);
403 	  break;
404 	default:
405 	  error = ENOTTY;
406 	  goto done;
407 	}
408 	error = 0;
409 done:
410 	pfs_pdone(procp);
411 	return 0;
412 }
413 
414 /*
415  * do block mapping for pfsnode (vp).
416  * since we don't use the buffer cache
417  * for procfs this function should never
418  * be called.  in any case, it's not clear
419  * what part of the kernel ever makes use
420  * of this function.  for sanity, this is the
421  * usual no-op bmap, although returning
422  * (EIO) would be a reasonable alternative.
423  *
424  * XXX mmap assumes buffer cache operation
425  *
426  * procfs_bmap(struct vnode *a_vp, off_t a_loffset,
427  *		off_t *a_doffsetp, int *a_runp, int *a_runb)
428  */
429 static int
430 procfs_bmap(struct vop_bmap_args *ap)
431 {
432 	if (ap->a_doffsetp != NULL)
433 		*ap->a_doffsetp = ap->a_loffset;
434 	if (ap->a_runp != NULL)
435 		*ap->a_runp = 0;
436 	if (ap->a_runb != NULL)
437 		*ap->a_runb = 0;
438 	return (0);
439 }
440 
441 /*
442  * procfs_inactive is called when the pfsnode
443  * is vrele'd and the reference count goes
444  * to zero.  (vp) will be on the vnode free
445  * list, so to get it back vget() must be
446  * used.
447  *
448  * (vp) is locked on entry, but must be unlocked on exit.
449  *
450  * procfs_inactive(struct vnode *a_vp)
451  */
452 static int
453 procfs_inactive(struct vop_inactive_args *ap)
454 {
455 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
456 
457 	if (pfs->pfs_pid & PFS_DEAD)
458 		vrecycle(ap->a_vp);
459 	return (0);
460 }
461 
462 /*
463  * _reclaim is called when getnewvnode()
464  * wants to make use of an entry on the vnode
465  * free list.  at this time the filesystem needs
466  * to free any private data and remove the node
467  * from any private lists.
468  *
469  * procfs_reclaim(struct vnode *a_vp)
470  */
471 static int
472 procfs_reclaim(struct vop_reclaim_args *ap)
473 {
474 	return (procfs_freevp(ap->a_vp));
475 }
476 
477 /*
478  * _print is used for debugging.
479  * just print a readable description
480  * of (vp).
481  *
482  * procfs_print(struct vnode *a_vp)
483  */
484 static int
485 procfs_print(struct vop_print_args *ap)
486 {
487 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
488 
489 	kprintf("tag VT_PROCFS, type %d, pid %ld, mode %x, flags %lx\n",
490 	    pfs->pfs_type, (long)pfs->pfs_pid, pfs->pfs_mode, pfs->pfs_flags);
491 	return (0);
492 }
493 
494 /*
495  * generic entry point for unsupported operations
496  */
497 static int
498 procfs_badop(struct vop_generic_args *ap)
499 {
500 	return (EIO);
501 }
502 
503 /*
504  * Invent attributes for pfsnode (vp) and store
505  * them in (vap).
506  * Directories lengths are returned as zero since
507  * any real length would require the genuine size
508  * to be computed, and nothing cares anyway.
509  *
510  * this is relatively minimal for procfs.
511  *
512  * procfs_getattr(struct vnode *a_vp, struct vattr *a_vap)
513  */
514 static int
515 procfs_getattr(struct vop_getattr_args *ap)
516 {
517 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
518 	struct vattr *vap = ap->a_vap;
519 	struct proc *procp;
520 	int error;
521 
522 	/*
523 	 * First make sure that the process and its credentials
524 	 * still exist.
525 	 */
526 	switch (pfs->pfs_type) {
527 	case Proot:
528 	case Pcurproc:
529 		procp = NULL;
530 		break;
531 	default:
532 		procp = pfs_pfind(pfs->pfs_pid);
533 		if (procp == NULL || procp->p_ucred == NULL) {
534 			error = ENOENT;
535 			goto done;
536 		}
537 		break;
538 	}
539 
540 	error = 0;
541 
542 	/* start by zeroing out the attributes */
543 	VATTR_NULL(vap);
544 
545 	/* next do all the common fields */
546 	vap->va_type = ap->a_vp->v_type;
547 	vap->va_mode = pfs->pfs_mode;
548 	vap->va_fileid = pfs->pfs_fileno;
549 	vap->va_flags = 0;
550 	vap->va_blocksize = PAGE_SIZE;
551 	vap->va_bytes = vap->va_size = 0;
552 	vap->va_fsid = ap->a_vp->v_mount->mnt_stat.f_fsid.val[0];
553 
554 	/*
555 	 * Make all times be current TOD.
556 	 * It would be possible to get the process start
557 	 * time from the p_stat structure, but there's
558 	 * no "file creation" time stamp anyway, and the
559 	 * p_stat structure is not addressible if u. gets
560 	 * swapped out for that process.
561 	 */
562 	nanotime(&vap->va_ctime);
563 	vap->va_atime = vap->va_mtime = vap->va_ctime;
564 
565 	/*
566 	 * If the process has exercised some setuid or setgid
567 	 * privilege, then rip away read/write permission so
568 	 * that only root can gain access.
569 	 */
570 	switch (pfs->pfs_type) {
571 	case Pctl:
572 	case Pregs:
573 	case Pfpregs:
574 	case Pdbregs:
575 	case Pmem:
576 		if (procp->p_flags & P_SUGID) {
577 			vap->va_mode &= ~((VREAD|VWRITE)|
578 					  ((VREAD|VWRITE)>>3)|
579 					  ((VREAD|VWRITE)>>6));
580 		}
581 		break;
582 	default:
583 		break;
584 	}
585 
586 	/*
587 	 * now do the object specific fields
588 	 *
589 	 * The size could be set from struct reg, but it's hardly
590 	 * worth the trouble, and it puts some (potentially) machine
591 	 * dependent data into this machine-independent code.  If it
592 	 * becomes important then this function should break out into
593 	 * a per-file stat function in the corresponding .c file.
594 	 */
595 
596 	vap->va_nlink = 1;
597 	if (procp) {
598 		if (procp->p_ucred) {
599 			vap->va_uid = procp->p_ucred->cr_uid;
600 			vap->va_gid = procp->p_ucred->cr_gid;
601 		} else {
602 			vap->va_uid = -1;
603 			vap->va_gid = -1;
604 		}
605 	}
606 
607 	switch (pfs->pfs_type) {
608 	case Proot:
609 		/*
610 		 * Set nlink to 1 to tell fts(3) we don't actually know.
611 		 */
612 		vap->va_nlink = 1;
613 		vap->va_uid = 0;
614 		vap->va_gid = 0;
615 		vap->va_size = vap->va_bytes = DEV_BSIZE;
616 		break;
617 
618 	case Pcurproc: {
619 		char buf[16];		/* should be enough */
620 
621 		vap->va_uid = 0;
622 		vap->va_gid = 0;
623 		vap->va_size = ksnprintf(buf, sizeof(buf),
624 					 "%ld", (long)curproc->p_pid);
625 		vap->va_bytes = vap->va_size;
626 		break;
627 	}
628 
629 	case Pproc:
630 		vap->va_nlink = nproc_targets;
631 		vap->va_size = vap->va_bytes = DEV_BSIZE;
632 		break;
633 
634 	case Pfile: {
635 		char *fullpath, *freepath;
636 
637 		if (procp->p_textnch.ncp) {
638 			struct nchandle nch;
639 
640 			cache_copy(&procp->p_textnch, &nch);
641 			error = cache_fullpath(procp, &nch, NULL,
642 					       &fullpath, &freepath, 0);
643 			cache_drop(&nch);
644 		} else {
645 			error = EINVAL;
646 		}
647 
648 		if (error == 0) {
649 			vap->va_size = strlen(fullpath);
650 			kfree(freepath, M_TEMP);
651 		} else {
652 			vap->va_size = sizeof("unknown") - 1;
653 			error = 0;
654 		}
655 		vap->va_bytes = vap->va_size;
656 		break;
657 	}
658 
659 	case Pmem:
660 		/*
661 		 * If we denied owner access earlier, then we have to
662 		 * change the owner to root - otherwise 'ps' and friends
663 		 * will break even though they are setgid kmem. *SIGH*
664 		 */
665 		if (procp->p_flags & P_SUGID)
666 			vap->va_uid = 0;
667 		else if (procp->p_ucred)
668 			vap->va_uid = procp->p_ucred->cr_uid;
669 		else
670 			vap->va_uid = -1;
671 		break;
672 
673 	case Pregs:
674 		vap->va_bytes = vap->va_size = sizeof(struct reg);
675 		break;
676 
677 	case Pfpregs:
678 		vap->va_bytes = vap->va_size = sizeof(struct fpreg);
679 		break;
680 
681         case Pdbregs:
682                 vap->va_bytes = vap->va_size = sizeof(struct dbreg);
683                 break;
684 
685 	case Ptype:
686 	case Pmap:
687 	case Pctl:
688 	case Pstatus:
689 	case Pnote:
690 	case Pnotepg:
691 	case Pcmdline:
692 	case Prlimit:
693 		break;
694 
695 	default:
696 		panic("procfs_getattr");
697 	}
698 done:
699 	pfs_pdone(procp);
700 	return (error);
701 }
702 
703 /*
704  * procfs_setattr(struct vnode *a_vp, struct vattr *a_vap,
705  *		  struct ucred *a_cred)
706  */
707 static int
708 procfs_setattr(struct vop_setattr_args *ap)
709 {
710 	if (ap->a_vap->va_flags != VNOVAL)
711 		return (EOPNOTSUPP);
712 
713 	/*
714 	 * just fake out attribute setting
715 	 * it's not good to generate an error
716 	 * return, otherwise things like creat()
717 	 * will fail when they try to set the
718 	 * file length to 0.  worse, this means
719 	 * that echo $note > /proc/$pid/note will fail.
720 	 */
721 
722 	return (0);
723 }
724 
725 /*
726  * implement access checking.
727  *
728  * procfs_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred)
729  */
730 static int
731 procfs_access(struct vop_access_args *ap)
732 {
733 	struct vattr vattr;
734 	int error;
735 
736 	error = VOP_GETATTR(ap->a_vp, &vattr);
737 	if (!error)
738 		error = vop_helper_access(ap, vattr.va_uid, vattr.va_gid,
739 				vattr.va_mode, 0);
740 	return (error);
741 }
742 
743 /*
744  * lookup.  this is incredibly complicated in the general case, however
745  * for most pseudo-filesystems very little needs to be done.
746  *
747  * procfs_lookup(struct vnode *a_dvp, struct vnode **a_vpp,
748  *		 struct componentname *a_cnp)
749  */
750 static int
751 procfs_lookup(struct vop_old_lookup_args *ap)
752 {
753 	struct componentname *cnp = ap->a_cnp;
754 	struct vnode **vpp = ap->a_vpp;
755 	struct vnode *dvp = ap->a_dvp;
756 	char *pname = cnp->cn_nameptr;
757 	/* struct proc *curp = cnp->cn_proc; */
758 	struct proc_target *pt;
759 	pid_t pid;
760 	struct pfsnode *pfs;
761 	struct proc *p;
762 	struct lwp *lp;
763 	int i;
764 	int error;
765 
766 	*vpp = NULL;
767 
768 	if (cnp->cn_nameiop == NAMEI_DELETE || cnp->cn_nameiop == NAMEI_RENAME)
769 		return (EROFS);
770 
771 	p = NULL;
772 	error = 0;
773 	if (cnp->cn_namelen == 1 && *pname == '.') {
774 		*vpp = dvp;
775 		vref(*vpp);
776 		goto out;
777 	}
778 
779 	pfs = VTOPFS(dvp);
780 	switch (pfs->pfs_type) {
781 	case Proot:
782 		if (cnp->cn_flags & CNP_ISDOTDOT)
783 			return (EIO);
784 
785 		if (CNEQ(cnp, "curproc", 7)) {
786 			error = procfs_allocvp(dvp->v_mount, vpp, 0, Pcurproc);
787 			goto out;
788 		}
789 
790 		pid = atopid(pname, cnp->cn_namelen);
791 		if (pid == NO_PID)
792 			break;
793 
794 		p = pfs_pfind(pid);
795 		if (p == NULL)
796 			break;
797 
798 		if (!PRISON_CHECK(ap->a_cnp->cn_cred, p->p_ucred))
799 			break;
800 
801 		if (ps_showallprocs == 0 && ap->a_cnp->cn_cred->cr_uid != 0 &&
802 		    ap->a_cnp->cn_cred->cr_uid != p->p_ucred->cr_uid)
803 			break;
804 
805 		error = procfs_allocvp(dvp->v_mount, vpp, pid, Pproc);
806 		goto out;
807 
808 	case Pproc:
809 		if (cnp->cn_flags & CNP_ISDOTDOT) {
810 			error = procfs_root(dvp->v_mount, vpp);
811 			goto out;
812 		}
813 
814 		p = pfs_pfind(pfs->pfs_pid);
815 		if (p == NULL)
816 			break;
817 		/* XXX lwp */
818 		lp = FIRST_LWP_IN_PROC(p);
819 
820 		if (!PRISON_CHECK(ap->a_cnp->cn_cred, p->p_ucred))
821 			break;
822 
823 		if (ps_showallprocs == 0 && ap->a_cnp->cn_cred->cr_uid != 0 &&
824 		    ap->a_cnp->cn_cred->cr_uid != p->p_ucred->cr_uid)
825 			break;
826 
827 		for (pt = proc_targets, i = 0; i < nproc_targets; pt++, i++) {
828 			if (cnp->cn_namelen == pt->pt_namlen &&
829 			    bcmp(pt->pt_name, pname, cnp->cn_namelen) == 0 &&
830 			    (pt->pt_valid == NULL || (*pt->pt_valid)(lp)))
831 				goto found;
832 		}
833 		break;
834 	found:
835 		error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
836 					pt->pt_pfstype);
837 		goto out;
838 
839 	default:
840 		error = ENOTDIR;
841 		goto out;
842 	}
843 	if (cnp->cn_nameiop == NAMEI_LOOKUP)
844 		error = ENOENT;
845 	else
846 		error = EROFS;
847 	/*
848 	 * If no error occured *vpp will hold a referenced locked vnode.
849 	 * dvp was passed to us locked and *vpp must be returned locked.
850 	 * If *vpp != dvp then we should unlock dvp if (1) this is not the
851 	 * last component or (2) CNP_LOCKPARENT is not set.
852 	 */
853 out:
854 	if (error == 0 && *vpp != dvp) {
855 		if ((cnp->cn_flags & CNP_LOCKPARENT) == 0) {
856 			cnp->cn_flags |= CNP_PDIRUNLOCK;
857 			vn_unlock(dvp);
858 		}
859 	}
860 	pfs_pdone(p);
861 	return (error);
862 }
863 
864 /*
865  * Does this process have a text file?
866  */
867 int
868 procfs_validfile(struct lwp *lp)
869 {
870 	return (procfs_findtextvp(lp->lwp_proc) != NULLVP);
871 }
872 
873 /*
874  * readdir() returns directory entries from pfsnode (vp).
875  *
876  * We generate just one directory entry at a time, as it would probably
877  * not pay off to buffer several entries locally to save uiomove calls.
878  *
879  * procfs_readdir(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred,
880  *		  int *a_eofflag, int *a_ncookies, off_t **a_cookies)
881  */
882 static int
883 procfs_readdir(struct vop_readdir_args *ap)
884 {
885 	struct pfsnode *pfs;
886 	int error;
887 
888 	if (ap->a_uio->uio_offset < 0 || ap->a_uio->uio_offset > INT_MAX)
889 		return (EINVAL);
890 	if ((error = vn_lock(ap->a_vp, LK_EXCLUSIVE | LK_RETRY)) != 0)
891 		return (error);
892 	pfs = VTOPFS(ap->a_vp);
893 
894 	switch (pfs->pfs_type) {
895 	case Pproc:
896 		/*
897 		 * this is for the process-specific sub-directories.
898 		 * all that is needed to is copy out all the entries
899 		 * from the procent[] table (top of this file).
900 		 */
901 		error = procfs_readdir_proc(ap);
902 		break;
903 	case Proot:
904 		/*
905 		 * this is for the root of the procfs filesystem
906 		 * what is needed is a special entry for "curproc"
907 		 * followed by an entry for each process on allproc
908 		 */
909 		error = procfs_readdir_root(ap);
910 		break;
911 	default:
912 		error = ENOTDIR;
913 		break;
914 	}
915 
916 	vn_unlock(ap->a_vp);
917 	return (error);
918 }
919 
920 static int
921 procfs_readdir_proc(struct vop_readdir_args *ap)
922 {
923 	struct pfsnode *pfs;
924 	int error, i, retval;
925 	struct proc *p;
926 	struct lwp *lp;
927 	struct proc_target *pt;
928 	struct uio *uio = ap->a_uio;
929 
930 	pfs = VTOPFS(ap->a_vp);
931 	p = pfs_pfind(pfs->pfs_pid);
932 	if (p == NULL)
933 		return(0);
934 	if (!PRISON_CHECK(ap->a_cred, p->p_ucred)) {
935 		error = 0;
936 		goto done;
937 	}
938 	/* XXX lwp, not MPSAFE */
939 	lp = FIRST_LWP_IN_PROC(p);
940 
941 	error = 0;
942 	i = (int)uio->uio_offset;
943 	if (i < 0) {
944 		error = EINVAL;
945 		goto done;
946 	}
947 
948 	for (pt = &proc_targets[i];
949 	     !error && uio->uio_resid > 0 && i < nproc_targets; pt++, i++) {
950 		if (pt->pt_valid && (*pt->pt_valid)(lp) == 0)
951 			continue;
952 
953 		retval = vop_write_dirent(&error, uio,
954 		    PROCFS_FILENO(pfs->pfs_pid, pt->pt_pfstype), pt->pt_type,
955 		    pt->pt_namlen, pt->pt_name);
956 		if (retval)
957 			break;
958 	}
959 
960 	uio->uio_offset = (off_t)i;
961 	error = 0;
962 done:
963 	pfs_pdone(p);
964 	return error;
965 }
966 
967 struct procfs_readdir_root_info {
968 	int error;
969 	int i;
970 	int pcnt;
971 	struct uio *uio;
972 	struct ucred *cred;
973 };
974 
975 static int procfs_readdir_root_callback(struct proc *p, void *data);
976 
977 static int
978 procfs_readdir_root(struct vop_readdir_args *ap)
979 {
980 	struct procfs_readdir_root_info info;
981 	struct uio *uio = ap->a_uio;
982 	int res;
983 
984 	res = 0;
985 	info.error = 0;
986 	info.i = (int)uio->uio_offset;
987 
988 	if (info.i < 0)
989 		return (EINVAL);
990 
991 	info.pcnt = 0;
992 	info.uio = uio;
993 	info.cred = ap->a_cred;
994 	while (info.pcnt < 3) {
995 		res = procfs_readdir_root_callback(NULL, &info);
996 		if (res < 0)
997 			break;
998 	}
999 	if (res >= 0)
1000 		allproc_scan(procfs_readdir_root_callback, &info);
1001 	uio->uio_offset = (off_t)info.i;
1002 
1003 	return (info.error);
1004 }
1005 
1006 static int
1007 procfs_readdir_root_callback(struct proc *p, void *data)
1008 {
1009 	struct procfs_readdir_root_info *info = data;
1010 	struct uio *uio;
1011 	int retval;
1012 	ino_t d_ino;
1013 	const char *d_name;
1014 	char d_name_pid[20];
1015 	size_t d_namlen;
1016 	uint8_t d_type;
1017 
1018 	uio = info->uio;
1019 
1020 	if (uio->uio_resid <= 0 || info->error)
1021 		return(-1);
1022 
1023 	switch (info->pcnt) {
1024 	case 0:		/* `.' */
1025 		d_ino = PROCFS_FILENO(0, Proot);
1026 		d_name = ".";
1027 		d_namlen = 1;
1028 		d_type = DT_DIR;
1029 		break;
1030 	case 1:		/* `..' */
1031 		d_ino = PROCFS_FILENO(0, Proot);
1032 		d_name = "..";
1033 		d_namlen = 2;
1034 		d_type = DT_DIR;
1035 		break;
1036 
1037 	case 2:
1038 		d_ino = PROCFS_FILENO(0, Pcurproc);
1039 		d_namlen = 7;
1040 		d_name = "curproc";
1041 		d_type = DT_LNK;
1042 		break;
1043 
1044 
1045 	default:
1046 		if (!PRISON_CHECK(info->cred, p->p_ucred))
1047 			return(0);
1048 		if (ps_showallprocs == 0 &&
1049 		    info->cred->cr_uid != 0 &&
1050 		    info->cred->cr_uid != p->p_ucred->cr_uid) {
1051 			return(0);
1052 		}
1053 
1054 		/*
1055 		 * Skip entries we have already returned (optimization)
1056 		 */
1057 		if (info->pcnt < info->i) {
1058 			++info->pcnt;
1059 			return(0);
1060 		}
1061 
1062 		d_ino = PROCFS_FILENO(p->p_pid, Pproc);
1063 		d_namlen = ksnprintf(d_name_pid, sizeof(d_name_pid),
1064 		    "%ld", (long)p->p_pid);
1065 		d_name = d_name_pid;
1066 		d_type = DT_DIR;
1067 		break;
1068 	}
1069 
1070 	/*
1071 	 * Skip entries we have already returned (optimization)
1072 	 */
1073 	if (info->pcnt < info->i) {
1074 		++info->pcnt;
1075 		return(0);
1076 	}
1077 
1078 	retval = vop_write_dirent(&info->error, uio,
1079 				  d_ino, d_type, d_namlen, d_name);
1080 	if (retval)
1081 		return(-1);
1082 	++info->pcnt;
1083 	++info->i;
1084 	return(0);
1085 }
1086 
1087 /*
1088  * readlink reads the link of `curproc' or `file'
1089  */
1090 static int
1091 procfs_readlink(struct vop_readlink_args *ap)
1092 {
1093 	char buf[16];		/* should be enough */
1094 	struct proc *procp;
1095 	struct vnode *vp = ap->a_vp;
1096 	struct pfsnode *pfs = VTOPFS(vp);
1097 	char *fullpath, *freepath;
1098 	int error, len;
1099 
1100 	switch (pfs->pfs_type) {
1101 	case Pcurproc:
1102 		if (pfs->pfs_fileno != PROCFS_FILENO(0, Pcurproc))
1103 			return (EINVAL);
1104 
1105 		len = ksnprintf(buf, sizeof(buf), "%ld", (long)curproc->p_pid);
1106 
1107 		return (uiomove(buf, len, ap->a_uio));
1108 	/*
1109 	 * There _should_ be no way for an entire process to disappear
1110 	 * from under us...
1111 	 */
1112 	case Pfile:
1113 		procp = pfs_pfind(pfs->pfs_pid);
1114 		if (procp == NULL || procp->p_ucred == NULL) {
1115 			kprintf("procfs_readlink: pid %d disappeared\n",
1116 			    pfs->pfs_pid);
1117 			pfs_pdone(procp);
1118 			return (uiomove("unknown", sizeof("unknown") - 1,
1119 					ap->a_uio));
1120 		}
1121 		if (procp->p_textnch.ncp) {
1122 			struct nchandle nch;
1123 
1124 			cache_copy(&procp->p_textnch, &nch);
1125 			error = cache_fullpath(procp, &nch, NULL,
1126 					       &fullpath, &freepath, 0);
1127 			cache_drop(&nch);
1128 		} else {
1129 			error = EINVAL;
1130 		}
1131 
1132 		if (error != 0) {
1133 			pfs_pdone(procp);
1134 			return (uiomove("unknown", sizeof("unknown") - 1,
1135 					ap->a_uio));
1136 		}
1137 		error = uiomove(fullpath, strlen(fullpath), ap->a_uio);
1138 		kfree(freepath, M_TEMP);
1139 		pfs_pdone(procp);
1140 		return (error);
1141 	default:
1142 		return (EINVAL);
1143 	}
1144 }
1145 
1146 /*
1147  * convert decimal ascii to pid_t
1148  */
1149 static pid_t
1150 atopid(const char *b, u_int len)
1151 {
1152 	pid_t p = 0;
1153 
1154 	while (len--) {
1155 		char c = *b++;
1156 		if (c < '0' || c > '9')
1157 			return (NO_PID);
1158 		p = 10 * p + (c - '0');
1159 		if (p > PID_MAX)
1160 			return (NO_PID);
1161 	}
1162 
1163 	return (p);
1164 }
1165 
1166