xref: /dflybsd-src/sys/vfs/procfs/procfs_vnops.c (revision c0d274d062fd959993bf623f25f7cb6a8a676c4e)
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_flag & 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 	PRELE(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 = pfind(pfs->pfs_pid))
252 		    && !(p->p_pfsflags & PF_LINGER)) {
253 			spin_lock(&p->p_spin);
254 			p->p_stops = 0;
255 			p->p_step = 0;
256 			spin_unlock(&p->p_spin);
257 			wakeup(&p->p_step);
258 		}
259 		if (p)
260 			PRELE(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 = 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_flag & 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 	  procp->p_stops |= *(unsigned int*)ap->a_data;
306 	  break;
307 	case PIOCBIC:
308 	  procp->p_stops &= ~*(unsigned int*)ap->a_data;
309 	  break;
310 	case PIOCSFL:
311 	  /*
312 	   * NFLAGS is "non-suser_xxx flags" -- currently, only
313 	   * PFS_ISUGID ("ignore set u/g id");
314 	   */
315 #define NFLAGS	(PF_ISUGID)
316 	  flags = (unsigned char)*(unsigned int*)ap->a_data;
317 	  if (flags & NFLAGS && (error = priv_check_cred(ap->a_cred, PRIV_ROOT, 0)))
318 	    goto done;
319 	  procp->p_pfsflags = flags;
320 	  break;
321 	case PIOCGFL:
322 	  *(unsigned int*)ap->a_data = (unsigned int)procp->p_pfsflags;
323 	  break;
324 	case PIOCSTATUS:
325 	  /*
326 	   * NOTE: syscall entry deals with stopevents and may run without
327 	   *	   the MP lock.
328 	   */
329 	  psp = (struct procfs_status *)ap->a_data;
330 	  psp->flags = procp->p_pfsflags;
331 	  psp->events = procp->p_stops;
332 	  spin_lock(&procp->p_spin);
333 	  if (procp->p_step) {
334 	    psp->state = 0;
335 	    psp->why = procp->p_stype;
336 	    psp->val = procp->p_xstat;
337 	    spin_unlock(&procp->p_spin);
338 	  } else {
339 	    psp->state = 1;
340 	    spin_unlock(&procp->p_spin);
341 	    psp->why = 0;	/* Not defined values */
342 	    psp->val = 0;	/* Not defined values */
343 	  }
344 	  break;
345 	case PIOCWAIT:
346 	  /*
347 	   * NOTE: syscall entry deals with stopevents and may run without
348 	   *	   the MP lock.
349 	   */
350 	  psp = (struct procfs_status *)ap->a_data;
351 	  spin_lock(&procp->p_spin);
352 	  while (procp->p_step == 0) {
353 	    tsleep_interlock(&procp->p_stype, PCATCH);
354 	    spin_unlock(&procp->p_spin);
355 	    error = tsleep(&procp->p_stype, PCATCH | PINTERLOCKED, "piocwait", 0);
356 	    if (error)
357 	      goto done;
358 	    spin_lock(&procp->p_spin);
359 	  }
360 	  spin_unlock(&procp->p_spin);
361 	  psp->state = 1;	/* It stopped */
362 	  psp->flags = procp->p_pfsflags;
363 	  psp->events = procp->p_stops;
364 	  psp->why = procp->p_stype;	/* why it stopped */
365 	  psp->val = procp->p_xstat;	/* any extra info */
366 	  break;
367 	case PIOCCONT:	/* Restart a proc */
368 	  /*
369 	   * NOTE: syscall entry deals with stopevents and may run without
370 	   *	   the MP lock.  However, the caller is presumably interlocked
371 	   *	   by having waited.
372 	   */
373 	  if (procp->p_step == 0) {
374 	    error = EINVAL;	/* Can only start a stopped process */
375 	    goto done;
376 	  }
377 	  if ((signo = *(int*)ap->a_data) != 0) {
378 	    if (signo >= NSIG || signo <= 0) {
379 	      error = EINVAL;
380 	      goto done;
381 	    }
382 	    ksignal(procp, signo);
383 	  }
384 	  procp->p_step = 0;
385 	  wakeup(&procp->p_step);
386 	  break;
387 	default:
388 	  error = ENOTTY;
389 	  goto done;
390 	}
391 	error = 0;
392 done:
393 	PRELE(procp);
394 	return 0;
395 }
396 
397 /*
398  * do block mapping for pfsnode (vp).
399  * since we don't use the buffer cache
400  * for procfs this function should never
401  * be called.  in any case, it's not clear
402  * what part of the kernel ever makes use
403  * of this function.  for sanity, this is the
404  * usual no-op bmap, although returning
405  * (EIO) would be a reasonable alternative.
406  *
407  * XXX mmap assumes buffer cache operation
408  *
409  * procfs_bmap(struct vnode *a_vp, off_t a_loffset,
410  *		off_t *a_doffsetp, int *a_runp, int *a_runb)
411  */
412 static int
413 procfs_bmap(struct vop_bmap_args *ap)
414 {
415 	if (ap->a_doffsetp != NULL)
416 		*ap->a_doffsetp = ap->a_loffset;
417 	if (ap->a_runp != NULL)
418 		*ap->a_runp = 0;
419 	if (ap->a_runb != NULL)
420 		*ap->a_runb = 0;
421 	return (0);
422 }
423 
424 /*
425  * procfs_inactive is called when the pfsnode
426  * is vrele'd and the reference count goes
427  * to zero.  (vp) will be on the vnode free
428  * list, so to get it back vget() must be
429  * used.
430  *
431  * (vp) is locked on entry, but must be unlocked on exit.
432  *
433  * procfs_inactive(struct vnode *a_vp)
434  */
435 static int
436 procfs_inactive(struct vop_inactive_args *ap)
437 {
438 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
439 
440 	if (pfs->pfs_pid & PFS_DEAD)
441 		vrecycle(ap->a_vp);
442 	return (0);
443 }
444 
445 /*
446  * _reclaim is called when getnewvnode()
447  * wants to make use of an entry on the vnode
448  * free list.  at this time the filesystem needs
449  * to free any private data and remove the node
450  * from any private lists.
451  *
452  * procfs_reclaim(struct vnode *a_vp)
453  */
454 static int
455 procfs_reclaim(struct vop_reclaim_args *ap)
456 {
457 	return (procfs_freevp(ap->a_vp));
458 }
459 
460 /*
461  * _print is used for debugging.
462  * just print a readable description
463  * of (vp).
464  *
465  * procfs_print(struct vnode *a_vp)
466  */
467 static int
468 procfs_print(struct vop_print_args *ap)
469 {
470 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
471 
472 	kprintf("tag VT_PROCFS, type %d, pid %ld, mode %x, flags %lx\n",
473 	    pfs->pfs_type, (long)pfs->pfs_pid, pfs->pfs_mode, pfs->pfs_flags);
474 	return (0);
475 }
476 
477 /*
478  * generic entry point for unsupported operations
479  */
480 static int
481 procfs_badop(struct vop_generic_args *ap)
482 {
483 	return (EIO);
484 }
485 
486 /*
487  * Invent attributes for pfsnode (vp) and store
488  * them in (vap).
489  * Directories lengths are returned as zero since
490  * any real length would require the genuine size
491  * to be computed, and nothing cares anyway.
492  *
493  * this is relatively minimal for procfs.
494  *
495  * procfs_getattr(struct vnode *a_vp, struct vattr *a_vap)
496  */
497 static int
498 procfs_getattr(struct vop_getattr_args *ap)
499 {
500 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
501 	struct vattr *vap = ap->a_vap;
502 	struct proc *procp;
503 	int error;
504 
505 	/*
506 	 * First make sure that the process and its credentials
507 	 * still exist.
508 	 */
509 	switch (pfs->pfs_type) {
510 	case Proot:
511 	case Pcurproc:
512 		procp = NULL;
513 		break;
514 
515 	default:
516 		procp = pfs_pfind(pfs->pfs_pid);
517 		if (procp == NULL || procp->p_ucred == NULL) {
518 			error = ENOENT;
519 			goto done;
520 		}
521 	}
522 
523 	error = 0;
524 
525 	/* start by zeroing out the attributes */
526 	VATTR_NULL(vap);
527 
528 	/* next do all the common fields */
529 	vap->va_type = ap->a_vp->v_type;
530 	vap->va_mode = pfs->pfs_mode;
531 	vap->va_fileid = pfs->pfs_fileno;
532 	vap->va_flags = 0;
533 	vap->va_blocksize = PAGE_SIZE;
534 	vap->va_bytes = vap->va_size = 0;
535 	vap->va_fsid = ap->a_vp->v_mount->mnt_stat.f_fsid.val[0];
536 
537 	/*
538 	 * Make all times be current TOD.
539 	 * It would be possible to get the process start
540 	 * time from the p_stat structure, but there's
541 	 * no "file creation" time stamp anyway, and the
542 	 * p_stat structure is not addressible if u. gets
543 	 * swapped out for that process.
544 	 */
545 	nanotime(&vap->va_ctime);
546 	vap->va_atime = vap->va_mtime = vap->va_ctime;
547 
548 	/*
549 	 * If the process has exercised some setuid or setgid
550 	 * privilege, then rip away read/write permission so
551 	 * that only root can gain access.
552 	 */
553 	switch (pfs->pfs_type) {
554 	case Pctl:
555 	case Pregs:
556 	case Pfpregs:
557 	case Pdbregs:
558 	case Pmem:
559 		if (procp->p_flag & P_SUGID)
560 			vap->va_mode &= ~((VREAD|VWRITE)|
561 					  ((VREAD|VWRITE)>>3)|
562 					  ((VREAD|VWRITE)>>6));
563 		break;
564 	default:
565 		break;
566 	}
567 
568 	/*
569 	 * now do the object specific fields
570 	 *
571 	 * The size could be set from struct reg, but it's hardly
572 	 * worth the trouble, and it puts some (potentially) machine
573 	 * dependent data into this machine-independent code.  If it
574 	 * becomes important then this function should break out into
575 	 * a per-file stat function in the corresponding .c file.
576 	 */
577 
578 	vap->va_nlink = 1;
579 	if (procp) {
580 		vap->va_uid = procp->p_ucred->cr_uid;
581 		vap->va_gid = procp->p_ucred->cr_gid;
582 	}
583 
584 	switch (pfs->pfs_type) {
585 	case Proot:
586 		/*
587 		 * Set nlink to 1 to tell fts(3) we don't actually know.
588 		 */
589 		vap->va_nlink = 1;
590 		vap->va_uid = 0;
591 		vap->va_gid = 0;
592 		vap->va_size = vap->va_bytes = DEV_BSIZE;
593 		break;
594 
595 	case Pcurproc: {
596 		char buf[16];		/* should be enough */
597 		vap->va_uid = 0;
598 		vap->va_gid = 0;
599 		vap->va_size = vap->va_bytes =
600 		    ksnprintf(buf, sizeof(buf), "%ld", (long)curproc->p_pid);
601 		break;
602 	}
603 
604 	case Pproc:
605 		vap->va_nlink = nproc_targets;
606 		vap->va_size = vap->va_bytes = DEV_BSIZE;
607 		break;
608 
609 	case Pfile: {
610 		char *fullpath, *freepath;
611 		error = cache_fullpath(procp, &procp->p_textnch, &fullpath, &freepath, 0);
612 		if (error == 0) {
613 			vap->va_size = strlen(fullpath);
614 			kfree(freepath, M_TEMP);
615 		} else {
616 			vap->va_size = sizeof("unknown") - 1;
617 			error = 0;
618 		}
619 		vap->va_bytes = vap->va_size;
620 		break;
621 	}
622 
623 	case Pmem:
624 		/*
625 		 * If we denied owner access earlier, then we have to
626 		 * change the owner to root - otherwise 'ps' and friends
627 		 * will break even though they are setgid kmem. *SIGH*
628 		 */
629 		if (procp->p_flag & P_SUGID)
630 			vap->va_uid = 0;
631 		else
632 			vap->va_uid = procp->p_ucred->cr_uid;
633 		break;
634 
635 	case Pregs:
636 		vap->va_bytes = vap->va_size = sizeof(struct reg);
637 		break;
638 
639 	case Pfpregs:
640 		vap->va_bytes = vap->va_size = sizeof(struct fpreg);
641 		break;
642 
643         case Pdbregs:
644                 vap->va_bytes = vap->va_size = sizeof(struct dbreg);
645                 break;
646 
647 	case Ptype:
648 	case Pmap:
649 	case Pctl:
650 	case Pstatus:
651 	case Pnote:
652 	case Pnotepg:
653 	case Pcmdline:
654 	case Prlimit:
655 		break;
656 
657 	default:
658 		panic("procfs_getattr");
659 	}
660 done:
661 	if (procp)
662 		PRELE(procp);
663 	return (error);
664 }
665 
666 /*
667  * procfs_setattr(struct vnode *a_vp, struct vattr *a_vap,
668  *		  struct ucred *a_cred)
669  */
670 static int
671 procfs_setattr(struct vop_setattr_args *ap)
672 {
673 	if (ap->a_vap->va_flags != VNOVAL)
674 		return (EOPNOTSUPP);
675 
676 	/*
677 	 * just fake out attribute setting
678 	 * it's not good to generate an error
679 	 * return, otherwise things like creat()
680 	 * will fail when they try to set the
681 	 * file length to 0.  worse, this means
682 	 * that echo $note > /proc/$pid/note will fail.
683 	 */
684 
685 	return (0);
686 }
687 
688 /*
689  * implement access checking.
690  *
691  * procfs_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred)
692  */
693 static int
694 procfs_access(struct vop_access_args *ap)
695 {
696 	struct vattr vattr;
697 	int error;
698 
699 	error = VOP_GETATTR(ap->a_vp, &vattr);
700 	if (!error)
701 		error = vop_helper_access(ap, vattr.va_uid, vattr.va_gid,
702 				vattr.va_mode, 0);
703 	return (error);
704 }
705 
706 /*
707  * lookup.  this is incredibly complicated in the general case, however
708  * for most pseudo-filesystems very little needs to be done.
709  *
710  * procfs_lookup(struct vnode *a_dvp, struct vnode **a_vpp,
711  *		 struct componentname *a_cnp)
712  */
713 static int
714 procfs_lookup(struct vop_old_lookup_args *ap)
715 {
716 	struct componentname *cnp = ap->a_cnp;
717 	struct vnode **vpp = ap->a_vpp;
718 	struct vnode *dvp = ap->a_dvp;
719 	char *pname = cnp->cn_nameptr;
720 	/* struct proc *curp = cnp->cn_proc; */
721 	struct proc_target *pt;
722 	pid_t pid;
723 	struct pfsnode *pfs;
724 	struct proc *p;
725 	struct lwp *lp;
726 	int i;
727 	int error;
728 
729 	*vpp = NULL;
730 
731 	if (cnp->cn_nameiop == NAMEI_DELETE || cnp->cn_nameiop == NAMEI_RENAME)
732 		return (EROFS);
733 
734 	p = NULL;
735 	error = 0;
736 	if (cnp->cn_namelen == 1 && *pname == '.') {
737 		*vpp = dvp;
738 		vref(*vpp);
739 		goto out;
740 	}
741 
742 	pfs = VTOPFS(dvp);
743 	switch (pfs->pfs_type) {
744 	case Proot:
745 		if (cnp->cn_flags & CNP_ISDOTDOT)
746 			return (EIO);
747 
748 		if (CNEQ(cnp, "curproc", 7)) {
749 			error = procfs_allocvp(dvp->v_mount, vpp, 0, Pcurproc);
750 			goto out;
751 		}
752 
753 		pid = atopid(pname, cnp->cn_namelen);
754 		if (pid == NO_PID)
755 			break;
756 
757 		p = pfs_pfind(pid);
758 		if (p == NULL)
759 			break;
760 
761 		if (!PRISON_CHECK(ap->a_cnp->cn_cred, p->p_ucred))
762 			break;
763 
764 		if (ps_showallprocs == 0 && ap->a_cnp->cn_cred->cr_uid != 0 &&
765 		    ap->a_cnp->cn_cred->cr_uid != p->p_ucred->cr_uid)
766 			break;
767 
768 		error = procfs_allocvp(dvp->v_mount, vpp, pid, Pproc);
769 		goto out;
770 
771 	case Pproc:
772 		if (cnp->cn_flags & CNP_ISDOTDOT) {
773 			error = procfs_root(dvp->v_mount, vpp);
774 			goto out;
775 		}
776 
777 		p = pfs_pfind(pfs->pfs_pid);
778 		if (p == NULL)
779 			break;
780 		/* XXX lwp */
781 		lp = FIRST_LWP_IN_PROC(p);
782 
783 		if (!PRISON_CHECK(ap->a_cnp->cn_cred, p->p_ucred))
784 			break;
785 
786 		if (ps_showallprocs == 0 && ap->a_cnp->cn_cred->cr_uid != 0 &&
787 		    ap->a_cnp->cn_cred->cr_uid != p->p_ucred->cr_uid)
788 			break;
789 
790 		for (pt = proc_targets, i = 0; i < nproc_targets; pt++, i++) {
791 			if (cnp->cn_namelen == pt->pt_namlen &&
792 			    bcmp(pt->pt_name, pname, cnp->cn_namelen) == 0 &&
793 			    (pt->pt_valid == NULL || (*pt->pt_valid)(lp)))
794 				goto found;
795 		}
796 		break;
797 	found:
798 		error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
799 					pt->pt_pfstype);
800 		goto out;
801 
802 	default:
803 		error = ENOTDIR;
804 		goto out;
805 	}
806 	if (cnp->cn_nameiop == NAMEI_LOOKUP)
807 		error = ENOENT;
808 	else
809 		error = EROFS;
810 	/*
811 	 * If no error occured *vpp will hold a referenced locked vnode.
812 	 * dvp was passed to us locked and *vpp must be returned locked.
813 	 * If *vpp != dvp then we should unlock dvp if (1) this is not the
814 	 * last component or (2) CNP_LOCKPARENT is not set.
815 	 */
816 out:
817 	if (error == 0 && *vpp != dvp) {
818 		if ((cnp->cn_flags & CNP_LOCKPARENT) == 0) {
819 			cnp->cn_flags |= CNP_PDIRUNLOCK;
820 			vn_unlock(dvp);
821 		}
822 	}
823 	if (p)
824 		PRELE(p);
825 	return (error);
826 }
827 
828 /*
829  * Does this process have a text file?
830  */
831 int
832 procfs_validfile(struct lwp *lp)
833 {
834 	return (procfs_findtextvp(lp->lwp_proc) != NULLVP);
835 }
836 
837 /*
838  * readdir() returns directory entries from pfsnode (vp).
839  *
840  * We generate just one directory entry at a time, as it would probably
841  * not pay off to buffer several entries locally to save uiomove calls.
842  *
843  * procfs_readdir(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred,
844  *		  int *a_eofflag, int *a_ncookies, off_t **a_cookies)
845  */
846 static int
847 procfs_readdir(struct vop_readdir_args *ap)
848 {
849 	struct pfsnode *pfs;
850 	int error;
851 
852 	if (ap->a_uio->uio_offset < 0 || ap->a_uio->uio_offset > INT_MAX)
853 		return (EINVAL);
854 	if ((error = vn_lock(ap->a_vp, LK_EXCLUSIVE | LK_RETRY)) != 0)
855 		return (error);
856 	pfs = VTOPFS(ap->a_vp);
857 
858 	switch (pfs->pfs_type) {
859 	case Pproc:
860 		/*
861 		 * this is for the process-specific sub-directories.
862 		 * all that is needed to is copy out all the entries
863 		 * from the procent[] table (top of this file).
864 		 */
865 		error = procfs_readdir_proc(ap);
866 		break;
867 	case Proot:
868 		/*
869 		 * this is for the root of the procfs filesystem
870 		 * what is needed is a special entry for "curproc"
871 		 * followed by an entry for each process on allproc
872 		 */
873 		error = procfs_readdir_root(ap);
874 		break;
875 	default:
876 		error = ENOTDIR;
877 		break;
878 	}
879 
880 	vn_unlock(ap->a_vp);
881 	return (error);
882 }
883 
884 static int
885 procfs_readdir_proc(struct vop_readdir_args *ap)
886 {
887 	struct pfsnode *pfs;
888 	int error, i, retval;
889 	struct proc *p;
890 	struct lwp *lp;
891 	struct proc_target *pt;
892 	struct uio *uio = ap->a_uio;
893 
894 	pfs = VTOPFS(ap->a_vp);
895 	p = pfs_pfind(pfs->pfs_pid);
896 	if (p == NULL)
897 		return(0);
898 	if (!PRISON_CHECK(ap->a_cred, p->p_ucred)) {
899 		error = 0;
900 		goto done;
901 	}
902 	/* XXX lwp, not MPSAFE */
903 	lp = FIRST_LWP_IN_PROC(p);
904 
905 	error = 0;
906 	i = (int)uio->uio_offset;
907 	if (i < 0) {
908 		error = EINVAL;
909 		goto done;
910 	}
911 
912 	for (pt = &proc_targets[i];
913 	     !error && uio->uio_resid > 0 && i < nproc_targets; pt++, i++) {
914 		if (pt->pt_valid && (*pt->pt_valid)(lp) == 0)
915 			continue;
916 
917 		retval = vop_write_dirent(&error, uio,
918 		    PROCFS_FILENO(pfs->pfs_pid, pt->pt_pfstype), pt->pt_type,
919 		    pt->pt_namlen, pt->pt_name);
920 		if (retval)
921 			break;
922 	}
923 
924 	uio->uio_offset = (off_t)i;
925 	error = 0;
926 done:
927 	PRELE(p);
928 	return error;
929 }
930 
931 struct procfs_readdir_root_info {
932 	int error;
933 	int i;
934 	int pcnt;
935 	struct uio *uio;
936 	struct ucred *cred;
937 };
938 
939 static int procfs_readdir_root_callback(struct proc *p, void *data);
940 
941 static int
942 procfs_readdir_root(struct vop_readdir_args *ap)
943 {
944 	struct procfs_readdir_root_info info;
945 	struct uio *uio = ap->a_uio;
946 	int res;
947 
948 	info.error = 0;
949 	info.i = (int)uio->uio_offset;
950 
951 	if (info.i < 0)
952 		return (EINVAL);
953 
954 	info.pcnt = 0;
955 	info.uio = uio;
956 	info.cred = ap->a_cred;
957 	while (info.pcnt < 3) {
958 		res = procfs_readdir_root_callback(NULL, &info);
959 		if (res < 0)
960 			break;
961 	}
962 	if (res >= 0)
963 		allproc_scan(procfs_readdir_root_callback, &info);
964 	uio->uio_offset = (off_t)info.i;
965 
966 	return (info.error);
967 }
968 
969 static int
970 procfs_readdir_root_callback(struct proc *p, void *data)
971 {
972 	struct procfs_readdir_root_info *info = data;
973 	struct uio *uio;
974 	int retval;
975 	ino_t d_ino;
976 	const char *d_name;
977 	char d_name_pid[20];
978 	size_t d_namlen;
979 	uint8_t d_type;
980 
981 	uio = info->uio;
982 
983 	if (uio->uio_resid <= 0 || info->error)
984 		return(-1);
985 
986 	switch (info->pcnt) {
987 	case 0:		/* `.' */
988 		d_ino = PROCFS_FILENO(0, Proot);
989 		d_name = ".";
990 		d_namlen = 1;
991 		d_type = DT_DIR;
992 		break;
993 	case 1:		/* `..' */
994 		d_ino = PROCFS_FILENO(0, Proot);
995 		d_name = "..";
996 		d_namlen = 2;
997 		d_type = DT_DIR;
998 		break;
999 
1000 	case 2:
1001 		d_ino = PROCFS_FILENO(0, Pcurproc);
1002 		d_namlen = 7;
1003 		d_name = "curproc";
1004 		d_type = DT_LNK;
1005 		break;
1006 
1007 
1008 	default:
1009 		if (!PRISON_CHECK(info->cred, p->p_ucred))
1010 			return(0);
1011 		if (ps_showallprocs == 0 &&
1012 		    info->cred->cr_uid != 0 &&
1013 		    info->cred->cr_uid != p->p_ucred->cr_uid) {
1014 			return(0);
1015 		}
1016 
1017 		/*
1018 		 * Skip entries we have already returned (optimization)
1019 		 */
1020 		if (info->pcnt < info->i) {
1021 			++info->pcnt;
1022 			return(0);
1023 		}
1024 
1025 		d_ino = PROCFS_FILENO(p->p_pid, Pproc);
1026 		d_namlen = ksnprintf(d_name_pid, sizeof(d_name_pid),
1027 		    "%ld", (long)p->p_pid);
1028 		d_name = d_name_pid;
1029 		d_type = DT_DIR;
1030 		break;
1031 	}
1032 
1033 	/*
1034 	 * Skip entries we have already returned (optimization)
1035 	 */
1036 	if (info->pcnt < info->i) {
1037 		++info->pcnt;
1038 		return(0);
1039 	}
1040 
1041 	retval = vop_write_dirent(&info->error, uio,
1042 				  d_ino, d_type, d_namlen, d_name);
1043 	if (retval)
1044 		return(-1);
1045 	++info->pcnt;
1046 	++info->i;
1047 	return(0);
1048 }
1049 
1050 /*
1051  * readlink reads the link of `curproc' or `file'
1052  */
1053 static int
1054 procfs_readlink(struct vop_readlink_args *ap)
1055 {
1056 	char buf[16];		/* should be enough */
1057 	struct proc *procp;
1058 	struct vnode *vp = ap->a_vp;
1059 	struct pfsnode *pfs = VTOPFS(vp);
1060 	char *fullpath, *freepath;
1061 	int error, len;
1062 
1063 	switch (pfs->pfs_type) {
1064 	case Pcurproc:
1065 		if (pfs->pfs_fileno != PROCFS_FILENO(0, Pcurproc))
1066 			return (EINVAL);
1067 
1068 		len = ksnprintf(buf, sizeof(buf), "%ld", (long)curproc->p_pid);
1069 
1070 		return (uiomove(buf, len, ap->a_uio));
1071 	/*
1072 	 * There _should_ be no way for an entire process to disappear
1073 	 * from under us...
1074 	 */
1075 	case Pfile:
1076 		procp = pfs_pfind(pfs->pfs_pid);
1077 		if (procp == NULL || procp->p_ucred == NULL) {
1078 			kprintf("procfs_readlink: pid %d disappeared\n",
1079 			    pfs->pfs_pid);
1080 			if (procp)
1081 				PRELE(procp);
1082 			return (uiomove("unknown", sizeof("unknown") - 1,
1083 			    ap->a_uio));
1084 		}
1085 		error = cache_fullpath(procp, &procp->p_textnch, &fullpath, &freepath, 0);
1086 		if (error != 0) {
1087 			if (procp)
1088 				PRELE(procp);
1089 			return (uiomove("unknown", sizeof("unknown") - 1,
1090 			    ap->a_uio));
1091 		}
1092 		error = uiomove(fullpath, strlen(fullpath), ap->a_uio);
1093 		kfree(freepath, M_TEMP);
1094 		if (procp)
1095 			PRELE(procp);
1096 		return (error);
1097 	default:
1098 		return (EINVAL);
1099 	}
1100 }
1101 
1102 /*
1103  * convert decimal ascii to pid_t
1104  */
1105 static pid_t
1106 atopid(const char *b, u_int len)
1107 {
1108 	pid_t p = 0;
1109 
1110 	while (len--) {
1111 		char c = *b++;
1112 		if (c < '0' || c > '9')
1113 			return (NO_PID);
1114 		p = 10 * p + (c - '0');
1115 		if (p > PID_MAX)
1116 			return (NO_PID);
1117 	}
1118 
1119 	return (p);
1120 }
1121 
1122