xref: /netbsd-src/sys/miscfs/procfs/procfs_vnops.c (revision 0dd5877adce57db949b16ae963e5a6831cccdfb6)
1 /*	$NetBSD: procfs_vnops.c,v 1.88 2002/01/12 18:52:30 christos Exp $	*/
2 
3 /*
4  * Copyright (c) 1993 Jan-Simon Pendry
5  * Copyright (c) 1993, 1995
6  *	The Regents of the University of California.  All rights reserved.
7  *
8  * This code is derived from software contributed to Berkeley by
9  * Jan-Simon Pendry.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  * 3. All advertising materials mentioning features or use of this software
20  *    must display the following acknowledgement:
21  *	This product includes software developed by the University of
22  *	California, Berkeley and its contributors.
23  * 4. Neither the name of the University nor the names of its contributors
24  *    may be used to endorse or promote products derived from this software
25  *    without specific prior written permission.
26  *
27  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
28  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
29  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
30  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
31  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
32  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
33  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
34  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
35  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
36  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
37  * SUCH DAMAGE.
38  *
39  *	@(#)procfs_vnops.c	8.18 (Berkeley) 5/21/95
40  */
41 
42 /*
43  * procfs vnode interface
44  */
45 
46 #include <sys/cdefs.h>
47 __KERNEL_RCSID(0, "$NetBSD: procfs_vnops.c,v 1.88 2002/01/12 18:52:30 christos Exp $");
48 
49 #include <sys/param.h>
50 #include <sys/systm.h>
51 #include <sys/time.h>
52 #include <sys/kernel.h>
53 #include <sys/file.h>
54 #include <sys/proc.h>
55 #include <sys/vnode.h>
56 #include <sys/namei.h>
57 #include <sys/malloc.h>
58 #include <sys/mount.h>
59 #include <sys/dirent.h>
60 #include <sys/resourcevar.h>
61 #include <sys/stat.h>
62 
63 #include <uvm/uvm_extern.h>	/* for PAGE_SIZE */
64 
65 #include <machine/reg.h>
66 
67 #include <miscfs/genfs/genfs.h>
68 #include <miscfs/procfs/procfs.h>
69 
70 /*
71  * Vnode Operations.
72  *
73  */
74 
75 static int procfs_validfile_linux __P((struct proc *, struct mount *));
76 
77 /*
78  * This is a list of the valid names in the
79  * process-specific sub-directories.  It is
80  * used in procfs_lookup and procfs_readdir
81  */
82 const struct proc_target {
83 	u_char	pt_type;
84 	u_char	pt_namlen;
85 	char	*pt_name;
86 	pfstype	pt_pfstype;
87 	int	(*pt_valid) __P((struct proc *, struct mount *));
88 } proc_targets[] = {
89 #define N(s) sizeof(s)-1, s
90 	/*	  name		type		validp */
91 	{ DT_DIR, N("."),	Pproc,		NULL },
92 	{ DT_DIR, N(".."),	Proot,		NULL },
93 	{ DT_REG, N("file"),	Pfile,		procfs_validfile },
94 	{ DT_REG, N("mem"),	Pmem,		NULL },
95 	{ DT_REG, N("regs"),	Pregs,		procfs_validregs },
96 	{ DT_REG, N("fpregs"),	Pfpregs,	procfs_validfpregs },
97 	{ DT_REG, N("ctl"),	Pctl,		NULL },
98 	{ DT_REG, N("status"),	Pstatus,	NULL },
99 	{ DT_REG, N("note"),	Pnote,		NULL },
100 	{ DT_REG, N("notepg"),	Pnotepg,	NULL },
101 	{ DT_REG, N("map"),	Pmap,		procfs_validmap },
102 	{ DT_REG, N("maps"),	Pmaps,		procfs_validmap },
103 	{ DT_REG, N("cmdline"), Pcmdline,	NULL },
104 	{ DT_REG, N("exe"),	Pfile,		procfs_validfile_linux },
105 #ifdef __HAVE_PROCFS_MACHDEP
106 	PROCFS_MACHDEP_NODETYPE_DEFNS
107 #endif
108 #undef N
109 };
110 static int nproc_targets = sizeof(proc_targets) / sizeof(proc_targets[0]);
111 
112 /*
113  * List of files in the root directory. Note: the validate function will
114  * be called with p == NULL for these ones.
115  */
116 struct proc_target proc_root_targets[] = {
117 #define N(s) sizeof(s)-1, s
118 	/*	  name		    type	    validp */
119 	{ DT_REG, N("meminfo"),     Pmeminfo,        procfs_validfile_linux },
120 	{ DT_REG, N("cpuinfo"),     Pcpuinfo,        procfs_validfile_linux },
121 #undef N
122 };
123 static int nproc_root_targets =
124     sizeof(proc_root_targets) / sizeof(proc_root_targets[0]);
125 
126 int	procfs_lookup	__P((void *));
127 #define	procfs_create	genfs_eopnotsupp_rele
128 #define	procfs_mknod	genfs_eopnotsupp_rele
129 int	procfs_open	__P((void *));
130 int	procfs_close	__P((void *));
131 int	procfs_access	__P((void *));
132 int	procfs_getattr	__P((void *));
133 int	procfs_setattr	__P((void *));
134 #define	procfs_read	procfs_rw
135 #define	procfs_write	procfs_rw
136 #define	procfs_fcntl	genfs_fcntl
137 #define	procfs_ioctl	genfs_enoioctl
138 #define	procfs_poll	genfs_poll
139 #define procfs_revoke	genfs_revoke
140 #define	procfs_fsync	genfs_nullop
141 #define	procfs_seek	genfs_nullop
142 #define	procfs_remove	genfs_eopnotsupp_rele
143 int	procfs_link	__P((void *));
144 #define	procfs_rename	genfs_eopnotsupp_rele
145 #define	procfs_mkdir	genfs_eopnotsupp_rele
146 #define	procfs_rmdir	genfs_eopnotsupp_rele
147 int	procfs_symlink	__P((void *));
148 int	procfs_readdir	__P((void *));
149 int	procfs_readlink	__P((void *));
150 #define	procfs_abortop	genfs_abortop
151 int	procfs_inactive	__P((void *));
152 int	procfs_reclaim	__P((void *));
153 #define	procfs_lock	genfs_lock
154 #define	procfs_unlock	genfs_unlock
155 #define	procfs_bmap	genfs_badop
156 #define	procfs_strategy	genfs_badop
157 int	procfs_print	__P((void *));
158 int	procfs_pathconf	__P((void *));
159 #define	procfs_islocked	genfs_islocked
160 #define	procfs_advlock	genfs_einval
161 #define	procfs_blkatoff	genfs_eopnotsupp
162 #define	procfs_valloc	genfs_eopnotsupp
163 #define	procfs_vfree	genfs_nullop
164 #define	procfs_truncate	genfs_eopnotsupp
165 #define	procfs_update	genfs_nullop
166 #define	procfs_bwrite	genfs_eopnotsupp
167 #define procfs_putpages	genfs_null_putpages
168 
169 static pid_t atopid __P((const char *, u_int));
170 
171 /*
172  * procfs vnode operations.
173  */
174 int (**procfs_vnodeop_p) __P((void *));
175 const struct vnodeopv_entry_desc procfs_vnodeop_entries[] = {
176 	{ &vop_default_desc, vn_default_error },
177 	{ &vop_lookup_desc, procfs_lookup },		/* lookup */
178 	{ &vop_create_desc, procfs_create },		/* create */
179 	{ &vop_mknod_desc, procfs_mknod },		/* mknod */
180 	{ &vop_open_desc, procfs_open },		/* open */
181 	{ &vop_close_desc, procfs_close },		/* close */
182 	{ &vop_access_desc, procfs_access },		/* access */
183 	{ &vop_getattr_desc, procfs_getattr },		/* getattr */
184 	{ &vop_setattr_desc, procfs_setattr },		/* setattr */
185 	{ &vop_read_desc, procfs_read },		/* read */
186 	{ &vop_write_desc, procfs_write },		/* write */
187 	{ &vop_fcntl_desc, procfs_fcntl },		/* fcntl */
188 	{ &vop_ioctl_desc, procfs_ioctl },		/* ioctl */
189 	{ &vop_poll_desc, procfs_poll },		/* poll */
190 	{ &vop_revoke_desc, procfs_revoke },		/* revoke */
191 	{ &vop_fsync_desc, procfs_fsync },		/* fsync */
192 	{ &vop_seek_desc, procfs_seek },		/* seek */
193 	{ &vop_remove_desc, procfs_remove },		/* remove */
194 	{ &vop_link_desc, procfs_link },		/* link */
195 	{ &vop_rename_desc, procfs_rename },		/* rename */
196 	{ &vop_mkdir_desc, procfs_mkdir },		/* mkdir */
197 	{ &vop_rmdir_desc, procfs_rmdir },		/* rmdir */
198 	{ &vop_symlink_desc, procfs_symlink },		/* symlink */
199 	{ &vop_readdir_desc, procfs_readdir },		/* readdir */
200 	{ &vop_readlink_desc, procfs_readlink },	/* readlink */
201 	{ &vop_abortop_desc, procfs_abortop },		/* abortop */
202 	{ &vop_inactive_desc, procfs_inactive },	/* inactive */
203 	{ &vop_reclaim_desc, procfs_reclaim },		/* reclaim */
204 	{ &vop_lock_desc, procfs_lock },		/* lock */
205 	{ &vop_unlock_desc, procfs_unlock },		/* unlock */
206 	{ &vop_bmap_desc, procfs_bmap },		/* bmap */
207 	{ &vop_strategy_desc, procfs_strategy },	/* strategy */
208 	{ &vop_print_desc, procfs_print },		/* print */
209 	{ &vop_islocked_desc, procfs_islocked },	/* islocked */
210 	{ &vop_pathconf_desc, procfs_pathconf },	/* pathconf */
211 	{ &vop_advlock_desc, procfs_advlock },		/* advlock */
212 	{ &vop_blkatoff_desc, procfs_blkatoff },	/* blkatoff */
213 	{ &vop_valloc_desc, procfs_valloc },		/* valloc */
214 	{ &vop_vfree_desc, procfs_vfree },		/* vfree */
215 	{ &vop_truncate_desc, procfs_truncate },	/* truncate */
216 	{ &vop_update_desc, procfs_update },		/* update */
217 	{ &vop_putpages_desc, procfs_putpages },	/* putpages */
218 	{ NULL, NULL }
219 };
220 const struct vnodeopv_desc procfs_vnodeop_opv_desc =
221 	{ &procfs_vnodeop_p, procfs_vnodeop_entries };
222 /*
223  * set things up for doing i/o on
224  * the pfsnode (vp).  (vp) is locked
225  * on entry, and should be left locked
226  * on exit.
227  *
228  * for procfs we don't need to do anything
229  * in particular for i/o.  all that is done
230  * is to support exclusive open on process
231  * memory images.
232  */
233 int
234 procfs_open(v)
235 	void *v;
236 {
237 	struct vop_open_args /* {
238 		struct vnode *a_vp;
239 		int  a_mode;
240 		struct ucred *a_cred;
241 		struct proc *a_p;
242 	} */ *ap = v;
243 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
244 	struct proc *p1, *p2;
245 	int error;
246 
247 	p1 = ap->a_p;				/* tracer */
248 	p2 = PFIND(pfs->pfs_pid);		/* traced */
249 
250 	if (p2 == NULL)
251 		return (ENOENT);		/* was ESRCH, jsp */
252 
253 	switch (pfs->pfs_type) {
254 	case Pmem:
255 		if (((pfs->pfs_flags & FWRITE) && (ap->a_mode & O_EXCL)) ||
256 		    ((pfs->pfs_flags & O_EXCL) && (ap->a_mode & FWRITE)))
257 			return (EBUSY);
258 
259 		if ((error = procfs_checkioperm(p1, p2)) != 0)
260 			return (error);
261 
262 		if (ap->a_mode & FWRITE)
263 			pfs->pfs_flags = ap->a_mode & (FWRITE|O_EXCL);
264 
265 		return (0);
266 
267 	default:
268 		break;
269 	}
270 
271 	return (0);
272 }
273 
274 /*
275  * close the pfsnode (vp) after doing i/o.
276  * (vp) is not locked on entry or exit.
277  *
278  * nothing to do for procfs other than undo
279  * any exclusive open flag (see _open above).
280  */
281 int
282 procfs_close(v)
283 	void *v;
284 {
285 	struct vop_close_args /* {
286 		struct vnode *a_vp;
287 		int  a_fflag;
288 		struct ucred *a_cred;
289 		struct proc *a_p;
290 	} */ *ap = v;
291 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
292 
293 	switch (pfs->pfs_type) {
294 	case Pmem:
295 		if ((ap->a_fflag & FWRITE) && (pfs->pfs_flags & O_EXCL))
296 			pfs->pfs_flags &= ~(FWRITE|O_EXCL);
297 		break;
298 
299 	default:
300 		break;
301 	}
302 
303 	return (0);
304 }
305 
306 /*
307  * _inactive is called when the pfsnode
308  * is vrele'd and the reference count goes
309  * to zero.  (vp) will be on the vnode free
310  * list, so to get it back vget() must be
311  * used.
312  *
313  * for procfs, check if the process is still
314  * alive and if it isn't then just throw away
315  * the vnode by calling vgone().  this may
316  * be overkill and a waste of time since the
317  * chances are that the process will still be
318  * there and PFIND is not free.
319  *
320  * (vp) is locked on entry, but must be unlocked on exit.
321  */
322 int
323 procfs_inactive(v)
324 	void *v;
325 {
326 	struct vop_inactive_args /* {
327 		struct vnode *a_vp;
328 		struct proc *a_p;
329 	} */ *ap = v;
330 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
331 
332 	VOP_UNLOCK(ap->a_vp, 0);
333 	if (PFIND(pfs->pfs_pid) == 0)
334 		vgone(ap->a_vp);
335 
336 	return (0);
337 }
338 
339 /*
340  * _reclaim is called when getnewvnode()
341  * wants to make use of an entry on the vnode
342  * free list.  at this time the filesystem needs
343  * to free any private data and remove the node
344  * from any private lists.
345  */
346 int
347 procfs_reclaim(v)
348 	void *v;
349 {
350 	struct vop_reclaim_args /* {
351 		struct vnode *a_vp;
352 	} */ *ap = v;
353 
354 	return (procfs_freevp(ap->a_vp));
355 }
356 
357 /*
358  * Return POSIX pathconf information applicable to special devices.
359  */
360 int
361 procfs_pathconf(v)
362 	void *v;
363 {
364 	struct vop_pathconf_args /* {
365 		struct vnode *a_vp;
366 		int a_name;
367 		register_t *a_retval;
368 	} */ *ap = v;
369 
370 	switch (ap->a_name) {
371 	case _PC_LINK_MAX:
372 		*ap->a_retval = LINK_MAX;
373 		return (0);
374 	case _PC_MAX_CANON:
375 		*ap->a_retval = MAX_CANON;
376 		return (0);
377 	case _PC_MAX_INPUT:
378 		*ap->a_retval = MAX_INPUT;
379 		return (0);
380 	case _PC_PIPE_BUF:
381 		*ap->a_retval = PIPE_BUF;
382 		return (0);
383 	case _PC_CHOWN_RESTRICTED:
384 		*ap->a_retval = 1;
385 		return (0);
386 	case _PC_VDISABLE:
387 		*ap->a_retval = _POSIX_VDISABLE;
388 		return (0);
389 	case _PC_SYNC_IO:
390 		*ap->a_retval = 1;
391 		return (0);
392 	default:
393 		return (EINVAL);
394 	}
395 	/* NOTREACHED */
396 }
397 
398 /*
399  * _print is used for debugging.
400  * just print a readable description
401  * of (vp).
402  */
403 int
404 procfs_print(v)
405 	void *v;
406 {
407 	struct vop_print_args /* {
408 		struct vnode *a_vp;
409 	} */ *ap = v;
410 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
411 
412 	printf("tag VT_PROCFS, type %d, pid %d, mode %x, flags %lx\n",
413 	    pfs->pfs_type, pfs->pfs_pid, pfs->pfs_mode, pfs->pfs_flags);
414 	return 0;
415 }
416 
417 int
418 procfs_link(v)
419 	void *v;
420 {
421 	struct vop_link_args /* {
422 		struct vnode *a_dvp;
423 		struct vnode *a_vp;
424 		struct componentname *a_cnp;
425 	} */ *ap = v;
426 
427 	VOP_ABORTOP(ap->a_dvp, ap->a_cnp);
428 	vput(ap->a_dvp);
429 	return (EROFS);
430 }
431 
432 int
433 procfs_symlink(v)
434 	void *v;
435 {
436 	struct vop_symlink_args /* {
437 		struct vnode *a_dvp;
438 		struct vnode **a_vpp;
439 		struct componentname *a_cnp;
440 		struct vattr *a_vap;
441 		char *a_target;
442 	} */ *ap = v;
443 
444 	VOP_ABORTOP(ap->a_dvp, ap->a_cnp);
445 	vput(ap->a_dvp);
446 	return (EROFS);
447 }
448 
449 /*
450  * Invent attributes for pfsnode (vp) and store
451  * them in (vap).
452  * Directories lengths are returned as zero since
453  * any real length would require the genuine size
454  * to be computed, and nothing cares anyway.
455  *
456  * this is relatively minimal for procfs.
457  */
458 int
459 procfs_getattr(v)
460 	void *v;
461 {
462 	struct vop_getattr_args /* {
463 		struct vnode *a_vp;
464 		struct vattr *a_vap;
465 		struct ucred *a_cred;
466 		struct proc *a_p;
467 	} */ *ap = v;
468 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
469 	struct vattr *vap = ap->a_vap;
470 	struct proc *procp;
471 	struct timeval tv;
472 	int error;
473 
474 	/* first check the process still exists */
475 	switch (pfs->pfs_type) {
476 	case Proot:
477 	case Pcurproc:
478 	case Pself:
479 		procp = 0;
480 		break;
481 
482 	default:
483 		procp = PFIND(pfs->pfs_pid);
484 		if (procp == 0)
485 			return (ENOENT);
486 		break;
487 	}
488 
489 	error = 0;
490 
491 	/* start by zeroing out the attributes */
492 	VATTR_NULL(vap);
493 
494 	/* next do all the common fields */
495 	vap->va_type = ap->a_vp->v_type;
496 	vap->va_mode = pfs->pfs_mode;
497 	vap->va_fileid = pfs->pfs_fileno;
498 	vap->va_flags = 0;
499 	vap->va_blocksize = PAGE_SIZE;
500 
501 	/*
502 	 * Make all times be current TOD.
503 	 * It would be possible to get the process start
504 	 * time from the p_stat structure, but there's
505 	 * no "file creation" time stamp anyway, and the
506 	 * p_stat structure is not addressible if u. gets
507 	 * swapped out for that process.
508 	 */
509 	microtime(&tv);
510 	TIMEVAL_TO_TIMESPEC(&tv, &vap->va_ctime);
511 	vap->va_atime = vap->va_mtime = vap->va_ctime;
512 
513 	switch (pfs->pfs_type) {
514 	case Pmem:
515 	case Pregs:
516 	case Pfpregs:
517 #if defined(__HAVE_PROCFS_MACHDEP) && defined(PROCFS_MACHDEP_PROTECT_CASES)
518 	PROCFS_MACHDEP_PROTECT_CASES
519 #endif
520 		/*
521 		 * If the process has exercised some setuid or setgid
522 		 * privilege, then rip away read/write permission so
523 		 * that only root can gain access.
524 		 */
525 		if (procp->p_flag & P_SUGID)
526 			vap->va_mode &= ~(S_IRUSR|S_IWUSR);
527 		/* FALLTHROUGH */
528 	case Pctl:
529 	case Pstatus:
530 	case Pnote:
531 	case Pnotepg:
532 	case Pmap:
533 	case Pmaps:
534 	case Pcmdline:
535 		vap->va_nlink = 1;
536 		vap->va_uid = procp->p_ucred->cr_uid;
537 		vap->va_gid = procp->p_ucred->cr_gid;
538 		break;
539 	case Pmeminfo:
540 	case Pcpuinfo:
541 		vap->va_nlink = 1;
542 		vap->va_uid = vap->va_gid = 0;
543 		break;
544 
545 	default:
546 		break;
547 	}
548 
549 	/*
550 	 * now do the object specific fields
551 	 *
552 	 * The size could be set from struct reg, but it's hardly
553 	 * worth the trouble, and it puts some (potentially) machine
554 	 * dependent data into this machine-independent code.  If it
555 	 * becomes important then this function should break out into
556 	 * a per-file stat function in the corresponding .c file.
557 	 */
558 
559 	switch (pfs->pfs_type) {
560 	case Proot:
561 		/*
562 		 * Set nlink to 1 to tell fts(3) we don't actually know.
563 		 */
564 		vap->va_nlink = 1;
565 		vap->va_uid = 0;
566 		vap->va_gid = 0;
567 		vap->va_bytes = vap->va_size = DEV_BSIZE;
568 		break;
569 
570 	case Pcurproc: {
571 		char buf[16];		/* should be enough */
572 		vap->va_nlink = 1;
573 		vap->va_uid = 0;
574 		vap->va_gid = 0;
575 		vap->va_bytes = vap->va_size =
576 		    sprintf(buf, "%ld", (long)curproc->p_pid);
577 		break;
578 	}
579 
580 	case Pself:
581 		vap->va_nlink = 1;
582 		vap->va_uid = 0;
583 		vap->va_gid = 0;
584 		vap->va_bytes = vap->va_size = sizeof("curproc");
585 		break;
586 
587 	case Pproc:
588 		vap->va_nlink = 2;
589 		vap->va_uid = procp->p_ucred->cr_uid;
590 		vap->va_gid = procp->p_ucred->cr_gid;
591 		vap->va_bytes = vap->va_size = DEV_BSIZE;
592 		break;
593 
594 	case Pfile:
595 		error = EOPNOTSUPP;
596 		break;
597 
598 	case Pmem:
599 		vap->va_bytes = vap->va_size =
600 			ctob(procp->p_vmspace->vm_tsize +
601 				    procp->p_vmspace->vm_dsize +
602 				    procp->p_vmspace->vm_ssize);
603 		break;
604 
605 #if defined(PT_GETREGS) || defined(PT_SETREGS)
606 	case Pregs:
607 		vap->va_bytes = vap->va_size = sizeof(struct reg);
608 		break;
609 #endif
610 
611 #if defined(PT_GETFPREGS) || defined(PT_SETFPREGS)
612 	case Pfpregs:
613 		vap->va_bytes = vap->va_size = sizeof(struct fpreg);
614 		break;
615 #endif
616 
617 	case Pctl:
618 	case Pstatus:
619 	case Pnote:
620 	case Pnotepg:
621 	case Pcmdline:
622 	case Pmeminfo:
623 	case Pcpuinfo:
624 		vap->va_bytes = vap->va_size = 0;
625 		break;
626 	case Pmap:
627 	case Pmaps:
628 		/*
629 		 * Advise a larger blocksize for the map files, so that
630 		 * they may be read in one pass.
631 		 */
632 		vap->va_blocksize = 4 * PAGE_SIZE;
633 		vap->va_bytes = vap->va_size = 0;
634 		break;
635 
636 #ifdef __HAVE_PROCFS_MACHDEP
637 	PROCFS_MACHDEP_NODETYPE_CASES
638 		error = procfs_machdep_getattr(ap->a_vp, vap, procp);
639 		break;
640 #endif
641 
642 	default:
643 		panic("procfs_getattr");
644 	}
645 
646 	return (error);
647 }
648 
649 /*ARGSUSED*/
650 int
651 procfs_setattr(v)
652 	void *v;
653 {
654 	/*
655 	 * just fake out attribute setting
656 	 * it's not good to generate an error
657 	 * return, otherwise things like creat()
658 	 * will fail when they try to set the
659 	 * file length to 0.  worse, this means
660 	 * that echo $note > /proc/$pid/note will fail.
661 	 */
662 
663 	return (0);
664 }
665 
666 /*
667  * implement access checking.
668  *
669  * actually, the check for super-user is slightly
670  * broken since it will allow read access to write-only
671  * objects.  this doesn't cause any particular trouble
672  * but does mean that the i/o entry points need to check
673  * that the operation really does make sense.
674  */
675 int
676 procfs_access(v)
677 	void *v;
678 {
679 	struct vop_access_args /* {
680 		struct vnode *a_vp;
681 		int a_mode;
682 		struct ucred *a_cred;
683 		struct proc *a_p;
684 	} */ *ap = v;
685 	struct vattr va;
686 	int error;
687 
688 	if ((error = VOP_GETATTR(ap->a_vp, &va, ap->a_cred, ap->a_p)) != 0)
689 		return (error);
690 
691 	return (vaccess(va.va_type, va.va_mode,
692 	    va.va_uid, va.va_gid, ap->a_mode, ap->a_cred));
693 }
694 
695 /*
696  * lookup.  this is incredibly complicated in the
697  * general case, however for most pseudo-filesystems
698  * very little needs to be done.
699  *
700  * Locking isn't hard here, just poorly documented.
701  *
702  * If we're looking up ".", just vref the parent & return it.
703  *
704  * If we're looking up "..", unlock the parent, and lock "..". If everything
705  * went ok, and we're on the last component and the caller requested the
706  * parent locked, try to re-lock the parent. We do this to prevent lock
707  * races.
708  *
709  * For anything else, get the needed node. Then unlock the parent if not
710  * the last component or not LOCKPARENT (i.e. if we wouldn't re-lock the
711  * parent in the .. case).
712  *
713  * We try to exit with the parent locked in error cases.
714  */
715 int
716 procfs_lookup(v)
717 	void *v;
718 {
719 	struct vop_lookup_args /* {
720 		struct vnode * a_dvp;
721 		struct vnode ** a_vpp;
722 		struct componentname * a_cnp;
723 	} */ *ap = v;
724 	struct componentname *cnp = ap->a_cnp;
725 	struct vnode **vpp = ap->a_vpp;
726 	struct vnode *dvp = ap->a_dvp;
727 	const char *pname = cnp->cn_nameptr;
728 	const struct proc_target *pt = NULL;
729 	struct vnode *fvp;
730 	pid_t pid;
731 	struct pfsnode *pfs;
732 	struct proc *p = NULL;
733 	int i, error, wantpunlock, iscurproc = 0, isself = 0;
734 
735 	*vpp = NULL;
736 	cnp->cn_flags &= ~PDIRUNLOCK;
737 
738 	if (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)
739 		return (EROFS);
740 
741 	if (cnp->cn_namelen == 1 && *pname == '.') {
742 		*vpp = dvp;
743 		VREF(dvp);
744 		return (0);
745 	}
746 
747 	wantpunlock = (~cnp->cn_flags & (LOCKPARENT | ISLASTCN));
748 	pfs = VTOPFS(dvp);
749 	switch (pfs->pfs_type) {
750 	case Proot:
751 		/*
752 		 * Shouldn't get here with .. in the root node.
753 		 */
754 		if (cnp->cn_flags & ISDOTDOT)
755 			return (EIO);
756 
757 		iscurproc = CNEQ(cnp, "curproc", 7);
758 		isself = CNEQ(cnp, "self", 4);
759 
760 		if (iscurproc || isself) {
761 			error = procfs_allocvp(dvp->v_mount, vpp, 0,
762 			    iscurproc ? Pcurproc : Pself);
763 			if ((error == 0) && (wantpunlock)) {
764 				VOP_UNLOCK(dvp, 0);
765 				cnp->cn_flags |= PDIRUNLOCK;
766 			}
767 			return (error);
768 		}
769 
770 		for (i = 0; i < nproc_root_targets; i++) {
771 			pt = &proc_root_targets[i];
772 			if (cnp->cn_namelen == pt->pt_namlen &&
773 			    memcmp(pt->pt_name, pname, cnp->cn_namelen) == 0 &&
774 			    (pt->pt_valid == NULL ||
775 			     (*pt->pt_valid)(p, dvp->v_mount)))
776 				break;
777 		}
778 
779 		if (i != nproc_root_targets) {
780 			error = procfs_allocvp(dvp->v_mount, vpp, 0,
781 			    pt->pt_pfstype);
782 			if ((error == 0) && (wantpunlock)) {
783 				VOP_UNLOCK(dvp, 0);
784 				cnp->cn_flags |= PDIRUNLOCK;
785 			}
786 			return (error);
787 		}
788 
789 		pid = atopid(pname, cnp->cn_namelen);
790 		if (pid == NO_PID)
791 			break;
792 
793 		p = PFIND(pid);
794 		if (p == 0)
795 			break;
796 
797 		error = procfs_allocvp(dvp->v_mount, vpp, pid, Pproc);
798 		if ((error == 0) && (wantpunlock)) {
799 			VOP_UNLOCK(dvp, 0);
800 			cnp->cn_flags |= PDIRUNLOCK;
801 		}
802 		return (error);
803 
804 	case Pproc:
805 		/*
806 		 * do the .. dance. We unlock the directory, and then
807 		 * get the root dir. That will automatically return ..
808 		 * locked. Then if the caller wanted dvp locked, we
809 		 * re-lock.
810 		 */
811 		if (cnp->cn_flags & ISDOTDOT) {
812 			VOP_UNLOCK(dvp, 0);
813 			cnp->cn_flags |= PDIRUNLOCK;
814 			error = procfs_root(dvp->v_mount, vpp);
815 			if ((error == 0) && (wantpunlock == 0) &&
816 				    ((error = vn_lock(dvp, LK_EXCLUSIVE)) == 0))
817 				cnp->cn_flags &= ~PDIRUNLOCK;
818 			return (error);
819 		}
820 
821 		p = PFIND(pfs->pfs_pid);
822 		if (p == 0)
823 			break;
824 
825 		for (pt = proc_targets, i = 0; i < nproc_targets; pt++, i++) {
826 			if (cnp->cn_namelen == pt->pt_namlen &&
827 			    memcmp(pt->pt_name, pname, cnp->cn_namelen) == 0 &&
828 			    (pt->pt_valid == NULL ||
829 			     (*pt->pt_valid)(p, dvp->v_mount)))
830 				goto found;
831 		}
832 		break;
833 
834 	found:
835 		if (pt->pt_pfstype == Pfile) {
836 			fvp = p->p_textvp;
837 			/* We already checked that it exists. */
838 			VREF(fvp);
839 			vn_lock(fvp, LK_EXCLUSIVE | LK_RETRY);
840 			if (wantpunlock) {
841 				VOP_UNLOCK(dvp, 0);
842 				cnp->cn_flags |= PDIRUNLOCK;
843 			}
844 			*vpp = fvp;
845 			return (0);
846 		}
847 
848 		error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
849 					pt->pt_pfstype);
850 		if ((error == 0) && (wantpunlock)) {
851 			VOP_UNLOCK(dvp, 0);
852 			cnp->cn_flags |= PDIRUNLOCK;
853 		}
854 		return (error);
855 
856 	default:
857 		return (ENOTDIR);
858 	}
859 
860 	return (cnp->cn_nameiop == LOOKUP ? ENOENT : EROFS);
861 }
862 
863 int
864 procfs_validfile(p, mp)
865 	struct proc *p;
866 	struct mount *mp;
867 {
868 	return (p->p_textvp != NULL);
869 }
870 
871 static int
872 procfs_validfile_linux(p, mp)
873 	struct proc *p;
874 	struct mount *mp;
875 {
876 	int flags;
877 
878 	flags = VFSTOPROC(mp)->pmnt_flags;
879 	return ((flags & PROCFSMNT_LINUXCOMPAT) &&
880 	    (p == NULL || procfs_validfile(p, mp)));
881 }
882 
883 /*
884  * readdir returns directory entries from pfsnode (vp).
885  *
886  * the strategy here with procfs is to generate a single
887  * directory entry at a time (struct dirent) and then
888  * copy that out to userland using uiomove.  a more efficent
889  * though more complex implementation, would try to minimize
890  * the number of calls to uiomove().  for procfs, this is
891  * hardly worth the added code complexity.
892  *
893  * this should just be done through read()
894  */
895 int
896 procfs_readdir(v)
897 	void *v;
898 {
899 	struct vop_readdir_args /* {
900 		struct vnode *a_vp;
901 		struct uio *a_uio;
902 		struct ucred *a_cred;
903 		int *a_eofflag;
904 		off_t **a_cookies;
905 		int *a_ncookies;
906 	} */ *ap = v;
907 	struct uio *uio = ap->a_uio;
908 	struct dirent d;
909 	struct pfsnode *pfs;
910 	off_t i;
911 	int error;
912 	off_t *cookies = NULL;
913 	int ncookies, left, skip, j;
914 	struct vnode *vp;
915 	const struct proc_target *pt;
916 
917 	vp = ap->a_vp;
918 	pfs = VTOPFS(vp);
919 
920 	if (uio->uio_resid < UIO_MX)
921 		return (EINVAL);
922 	if (uio->uio_offset < 0)
923 		return (EINVAL);
924 
925 	error = 0;
926 	i = uio->uio_offset;
927 	memset((caddr_t)&d, 0, UIO_MX);
928 	d.d_reclen = UIO_MX;
929 	ncookies = uio->uio_resid / UIO_MX;
930 
931 	switch (pfs->pfs_type) {
932 	/*
933 	 * this is for the process-specific sub-directories.
934 	 * all that is needed to is copy out all the entries
935 	 * from the procent[] table (top of this file).
936 	 */
937 	case Pproc: {
938 		struct proc *p;
939 
940 		if (i >= nproc_targets)
941 			return 0;
942 
943 		p = PFIND(pfs->pfs_pid);
944 		if (p == NULL)
945 			break;
946 
947 		if (ap->a_ncookies) {
948 			ncookies = min(ncookies, (nproc_targets - i));
949 			cookies = malloc(ncookies * sizeof (off_t),
950 			    M_TEMP, M_WAITOK);
951 			*ap->a_cookies = cookies;
952 		}
953 
954 		for (pt = &proc_targets[i];
955 		     uio->uio_resid >= UIO_MX && i < nproc_targets; pt++, i++) {
956 			if (pt->pt_valid &&
957 			    (*pt->pt_valid)(p, vp->v_mount) == 0)
958 				continue;
959 
960 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid, pt->pt_pfstype);
961 			d.d_namlen = pt->pt_namlen;
962 			memcpy(d.d_name, pt->pt_name, pt->pt_namlen + 1);
963 			d.d_type = pt->pt_type;
964 
965 			if ((error = uiomove((caddr_t)&d, UIO_MX, uio)) != 0)
966 				break;
967 			if (cookies)
968 				*cookies++ = i + 1;
969 		}
970 
971 	    	break;
972 	}
973 
974 	/*
975 	 * this is for the root of the procfs filesystem
976 	 * what is needed are special entries for "curproc"
977 	 * and "self" followed by an entry for each process
978 	 * on allproc
979 #ifdef PROCFS_ZOMBIE
980 	 * and deadproc and zombproc.
981 #endif
982 	 */
983 
984 	case Proot: {
985 		int pcnt = i, nc = 0;
986 		const struct proclist_desc *pd;
987 		volatile struct proc *p;
988 
989 		if (pcnt > 3)
990 			pcnt = 3;
991 		if (ap->a_ncookies) {
992 			/*
993 			 * XXX Potentially allocating too much space here,
994 			 * but I'm lazy. This loop needs some work.
995 			 */
996 			cookies = malloc(ncookies * sizeof (off_t),
997 			    M_TEMP, M_WAITOK);
998 			*ap->a_cookies = cookies;
999 		}
1000 		/*
1001 		 * XXX: THIS LOOP ASSUMES THAT allproc IS THE FIRST
1002 		 * PROCLIST IN THE proclists!
1003 		 */
1004 		proclist_lock_read();
1005 		pd = proclists;
1006 #ifdef PROCFS_ZOMBIE
1007 	again:
1008 #endif
1009 		for (p = LIST_FIRST(pd->pd_list);
1010 		     p != NULL && uio->uio_resid >= UIO_MX; i++, pcnt++) {
1011 			switch (i) {
1012 			case 0:		/* `.' */
1013 			case 1:		/* `..' */
1014 				d.d_fileno = PROCFS_FILENO(0, Proot);
1015 				d.d_namlen = i + 1;
1016 				memcpy(d.d_name, "..", d.d_namlen);
1017 				d.d_name[i + 1] = '\0';
1018 				d.d_type = DT_DIR;
1019 				break;
1020 
1021 			case 2:
1022 				d.d_fileno = PROCFS_FILENO(0, Pcurproc);
1023 				d.d_namlen = sizeof("curproc") - 1;
1024 				memcpy(d.d_name, "curproc", sizeof("curproc"));
1025 				d.d_type = DT_LNK;
1026 				break;
1027 
1028 			case 3:
1029 				d.d_fileno = PROCFS_FILENO(0, Pself);
1030 				d.d_namlen = sizeof("self") - 1;
1031 				memcpy(d.d_name, "self", sizeof("self"));
1032 				d.d_type = DT_LNK;
1033 				break;
1034 
1035 			default:
1036 				while (pcnt < i) {
1037 					pcnt++;
1038 					p = LIST_NEXT(p, p_list);
1039 					if (!p)
1040 						goto done;
1041 				}
1042 				d.d_fileno = PROCFS_FILENO(p->p_pid, Pproc);
1043 				d.d_namlen = sprintf(d.d_name, "%ld",
1044 				    (long)p->p_pid);
1045 				d.d_type = DT_DIR;
1046 				p = p->p_list.le_next;
1047 				break;
1048 			}
1049 
1050 			if ((error = uiomove((caddr_t)&d, UIO_MX, uio)) != 0)
1051 				break;
1052 			nc++;
1053 			if (cookies)
1054 				*cookies++ = i + 1;
1055 		}
1056 	done:
1057 
1058 #ifdef PROCFS_ZOMBIE
1059 		pd++;
1060 		if (p == NULL && pd->pd_list != NULL)
1061 			goto again;
1062 #endif
1063 		proclist_unlock_read();
1064 
1065 		skip = i - pcnt;
1066 		if (skip >= nproc_root_targets)
1067 			break;
1068 		left = nproc_root_targets - skip;
1069 		for (j = 0, pt = &proc_root_targets[0];
1070 		     uio->uio_resid >= UIO_MX && j < left;
1071 		     pt++, j++, i++) {
1072 			if (pt->pt_valid &&
1073 			    (*pt->pt_valid)(NULL, vp->v_mount) == 0)
1074 				continue;
1075 			d.d_fileno = PROCFS_FILENO(0, pt->pt_pfstype);
1076 			d.d_namlen = pt->pt_namlen;
1077 			memcpy(d.d_name, pt->pt_name, pt->pt_namlen + 1);
1078 			d.d_type = pt->pt_type;
1079 
1080 			if ((error = uiomove((caddr_t)&d, UIO_MX, uio)) != 0)
1081 				break;
1082 			nc++;
1083 			if (cookies)
1084 				*cookies++ = i + 1;
1085 		}
1086 
1087 		ncookies = nc;
1088 		break;
1089 	}
1090 
1091 	default:
1092 		error = ENOTDIR;
1093 		break;
1094 	}
1095 
1096 	if (ap->a_ncookies) {
1097 		if (error) {
1098 			if (cookies)
1099 				free(*ap->a_cookies, M_TEMP);
1100 			*ap->a_ncookies = 0;
1101 			*ap->a_cookies = NULL;
1102 		} else
1103 			*ap->a_ncookies = ncookies;
1104 	}
1105 	uio->uio_offset = i;
1106 	return (error);
1107 }
1108 
1109 /*
1110  * readlink reads the link of `curproc'
1111  */
1112 int
1113 procfs_readlink(v)
1114 	void *v;
1115 {
1116 	struct vop_readlink_args *ap = v;
1117 	char buf[16];		/* should be enough */
1118 	int len;
1119 
1120 	if (VTOPFS(ap->a_vp)->pfs_fileno == PROCFS_FILENO(0, Pcurproc))
1121 		len = sprintf(buf, "%ld", (long)curproc->p_pid);
1122 	else if (VTOPFS(ap->a_vp)->pfs_fileno == PROCFS_FILENO(0, Pself))
1123 		len = sprintf(buf, "%s", "curproc");
1124 	else
1125 		return (EINVAL);
1126 
1127 	return (uiomove((caddr_t)buf, len, ap->a_uio));
1128 }
1129 
1130 /*
1131  * convert decimal ascii to pid_t
1132  */
1133 static pid_t
1134 atopid(b, len)
1135 	const char *b;
1136 	u_int len;
1137 {
1138 	pid_t p = 0;
1139 
1140 	while (len--) {
1141 		char c = *b++;
1142 		if (c < '0' || c > '9')
1143 			return (NO_PID);
1144 		p = 10 * p + (c - '0');
1145 		if (p > PID_MAX)
1146 			return (NO_PID);
1147 	}
1148 
1149 	return (p);
1150 }
1151