Home | History | Annotate | Line # | Download | only in procfs
procfs_vnops.c revision 1.35
      1 /*	$NetBSD: procfs_vnops.c,v 1.35 1995/10/09 14:03:38 mycroft Exp $	*/
      2 
      3 /*
      4  * Copyright (c) 1993 Jan-Simon Pendry
      5  * Copyright (c) 1993
      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	8.8 (Berkeley) 6/15/94
     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/file.h>
     51 #include <sys/proc.h>
     52 #include <sys/vnode.h>
     53 #include <sys/namei.h>
     54 #include <sys/malloc.h>
     55 #include <sys/dirent.h>
     56 #include <sys/resourcevar.h>
     57 #include <sys/ptrace.h>
     58 #include <vm/vm.h>	/* for PAGE_SIZE */
     59 #include <machine/reg.h>
     60 #include <miscfs/procfs/procfs.h>
     61 
     62 /*
     63  * Vnode Operations.
     64  *
     65  */
     66 
     67 /*
     68  * This is a list of the valid names in the
     69  * process-specific sub-directories.  It is
     70  * used in procfs_lookup and procfs_readdir
     71  */
     72 struct proc_target {
     73 	u_char	pt_type;
     74 	u_char	pt_namlen;
     75 	char	*pt_name;
     76 	pfstype	pt_pfstype;
     77 	int	(*pt_valid) __P((struct proc *p));
     78 } proc_targets[] = {
     79 #define N(s) sizeof(s)-1, s
     80 	/*	  name		type		validp */
     81 	{ DT_DIR, N("."),	Pproc,		NULL },
     82 	{ DT_DIR, N(".."),	Proot,		NULL },
     83 	{ DT_REG, N("file"),	Pfile,		procfs_validfile },
     84 	{ DT_REG, N("mem"),	Pmem,		NULL },
     85 	{ DT_REG, N("regs"),	Pregs,		procfs_validregs },
     86 	{ DT_REG, N("fpregs"),	Pfpregs,	procfs_validfpregs },
     87 	{ DT_REG, N("ctl"),	Pctl,		NULL },
     88 	{ DT_REG, N("status"),	Pstatus,	NULL },
     89 	{ DT_REG, N("note"),	Pnote,		NULL },
     90 	{ DT_REG, N("notepg"),	Pnotepg,	NULL },
     91 #undef N
     92 };
     93 static int nproc_targets = sizeof(proc_targets) / sizeof(proc_targets[0]);
     94 
     95 static pid_t atopid __P((const char *, u_int));
     96 
     97 /*
     98  * set things up for doing i/o on
     99  * the pfsnode (vp).  (vp) is locked
    100  * on entry, and should be left locked
    101  * on exit.
    102  *
    103  * for procfs we don't need to do anything
    104  * in particular for i/o.  all that is done
    105  * is to support exclusive open on process
    106  * memory images.
    107  */
    108 procfs_open(ap)
    109 	struct vop_open_args /* {
    110 		struct vnode *a_vp;
    111 		int  a_mode;
    112 		struct ucred *a_cred;
    113 		struct proc *a_p;
    114 	} */ *ap;
    115 {
    116 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
    117 
    118 	switch (pfs->pfs_type) {
    119 	case Pmem:
    120 		if (PFIND(pfs->pfs_pid) == 0)
    121 			return (ENOENT);	/* was ESRCH, jsp */
    122 
    123 		if ((pfs->pfs_flags & FWRITE) && (ap->a_mode & O_EXCL) ||
    124 		    (pfs->pfs_flags & O_EXCL) && (ap->a_mode & FWRITE))
    125 			return (EBUSY);
    126 
    127 		if (ap->a_mode & FWRITE)
    128 			pfs->pfs_flags = ap->a_mode & (FWRITE|O_EXCL);
    129 
    130 		return (0);
    131 
    132 	default:
    133 		break;
    134 	}
    135 
    136 	return (0);
    137 }
    138 
    139 /*
    140  * close the pfsnode (vp) after doing i/o.
    141  * (vp) is not locked on entry or exit.
    142  *
    143  * nothing to do for procfs other than undo
    144  * any exclusive open flag (see _open above).
    145  */
    146 procfs_close(ap)
    147 	struct vop_close_args /* {
    148 		struct vnode *a_vp;
    149 		int  a_fflag;
    150 		struct ucred *a_cred;
    151 		struct proc *a_p;
    152 	} */ *ap;
    153 {
    154 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
    155 
    156 	switch (pfs->pfs_type) {
    157 	case Pmem:
    158 		if ((ap->a_fflag & FWRITE) && (pfs->pfs_flags & O_EXCL))
    159 			pfs->pfs_flags &= ~(FWRITE|O_EXCL);
    160 		break;
    161 	}
    162 
    163 	return (0);
    164 }
    165 
    166 /*
    167  * do an ioctl operation on pfsnode (vp).
    168  * (vp) is not locked on entry or exit.
    169  */
    170 procfs_ioctl(ap)
    171 	struct vop_ioctl_args /* {
    172 		struct vnode *a_vp;
    173 		u_long a_command;
    174 		caddr_t a_data;
    175 		int a_fflag;
    176 		struct ucred *a_cred;
    177 		struct proc *a_p;
    178 	} */ *ap;
    179 {
    180 
    181 	return (ENOTTY);
    182 }
    183 
    184 /*
    185  * do block mapping for pfsnode (vp).
    186  * since we don't use the buffer cache
    187  * for procfs this function should never
    188  * be called.  in any case, it's not clear
    189  * what part of the kernel ever makes use
    190  * of this function.  for sanity, this is the
    191  * usual no-op bmap, although returning
    192  * (EIO) would be a reasonable alternative.
    193  */
    194 procfs_bmap(ap)
    195 	struct vop_bmap_args /* {
    196 		struct vnode *a_vp;
    197 		daddr_t  a_bn;
    198 		struct vnode **a_vpp;
    199 		daddr_t *a_bnp;
    200 	} */ *ap;
    201 {
    202 
    203 	if (ap->a_vpp != NULL)
    204 		*ap->a_vpp = ap->a_vp;
    205 	if (ap->a_bnp != NULL)
    206 		*ap->a_bnp = ap->a_bn;
    207 	return (0);
    208 }
    209 
    210 /*
    211  * _inactive is called when the pfsnode
    212  * is vrele'd and the reference count goes
    213  * to zero.  (vp) will be on the vnode free
    214  * list, so to get it back vget() must be
    215  * used.
    216  *
    217  * for procfs, check if the process is still
    218  * alive and if it isn't then just throw away
    219  * the vnode by calling vgone().  this may
    220  * be overkill and a waste of time since the
    221  * chances are that the process will still be
    222  * there and PFIND is not free.
    223  *
    224  * (vp) is not locked on entry or exit.
    225  */
    226 procfs_inactive(ap)
    227 	struct vop_inactive_args /* {
    228 		struct vnode *a_vp;
    229 	} */ *ap;
    230 {
    231 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
    232 
    233 	if (PFIND(pfs->pfs_pid) == 0)
    234 		vgone(ap->a_vp);
    235 
    236 	return (0);
    237 }
    238 
    239 /*
    240  * _reclaim is called when getnewvnode()
    241  * wants to make use of an entry on the vnode
    242  * free list.  at this time the filesystem needs
    243  * to free any private data and remove the node
    244  * from any private lists.
    245  */
    246 procfs_reclaim(ap)
    247 	struct vop_reclaim_args /* {
    248 		struct vnode *a_vp;
    249 	} */ *ap;
    250 {
    251 
    252 	return (procfs_freevp(ap->a_vp));
    253 }
    254 
    255 /*
    256  * Return POSIX pathconf information applicable to special devices.
    257  */
    258 procfs_pathconf(ap)
    259 	struct vop_pathconf_args /* {
    260 		struct vnode *a_vp;
    261 		int a_name;
    262 		register_t *a_retval;
    263 	} */ *ap;
    264 {
    265 
    266 	switch (ap->a_name) {
    267 	case _PC_LINK_MAX:
    268 		*ap->a_retval = LINK_MAX;
    269 		return (0);
    270 	case _PC_MAX_CANON:
    271 		*ap->a_retval = MAX_CANON;
    272 		return (0);
    273 	case _PC_MAX_INPUT:
    274 		*ap->a_retval = MAX_INPUT;
    275 		return (0);
    276 	case _PC_PIPE_BUF:
    277 		*ap->a_retval = PIPE_BUF;
    278 		return (0);
    279 	case _PC_CHOWN_RESTRICTED:
    280 		*ap->a_retval = 1;
    281 		return (0);
    282 	case _PC_VDISABLE:
    283 		*ap->a_retval = _POSIX_VDISABLE;
    284 		return (0);
    285 	default:
    286 		return (EINVAL);
    287 	}
    288 	/* NOTREACHED */
    289 }
    290 
    291 /*
    292  * _print is used for debugging.
    293  * just print a readable description
    294  * of (vp).
    295  */
    296 procfs_print(ap)
    297 	struct vop_print_args /* {
    298 		struct vnode *a_vp;
    299 	} */ *ap;
    300 {
    301 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
    302 
    303 	printf("tag VT_PROCFS, type %s, pid %d, mode %x, flags %x\n",
    304 	    pfs->pfs_type, pfs->pfs_pid, pfs->pfs_mode, pfs->pfs_flags);
    305 }
    306 
    307 /*
    308  * _abortop is called when operations such as
    309  * rename and create fail.  this entry is responsible
    310  * for undoing any side-effects caused by the lookup.
    311  * this will always include freeing the pathname buffer.
    312  */
    313 procfs_abortop(ap)
    314 	struct vop_abortop_args /* {
    315 		struct vnode *a_dvp;
    316 		struct componentname *a_cnp;
    317 	} */ *ap;
    318 {
    319 
    320 	if ((ap->a_cnp->cn_flags & (HASBUF | SAVESTART)) == HASBUF)
    321 		FREE(ap->a_cnp->cn_pnbuf, M_NAMEI);
    322 	return (0);
    323 }
    324 
    325 /*
    326  * generic entry point for unsupported operations
    327  */
    328 procfs_badop()
    329 {
    330 
    331 	return (EIO);
    332 }
    333 
    334 /*
    335  * Invent attributes for pfsnode (vp) and store
    336  * them in (vap).
    337  * Directories lengths are returned as zero since
    338  * any real length would require the genuine size
    339  * to be computed, and nothing cares anyway.
    340  *
    341  * this is relatively minimal for procfs.
    342  */
    343 procfs_getattr(ap)
    344 	struct vop_getattr_args /* {
    345 		struct vnode *a_vp;
    346 		struct vattr *a_vap;
    347 		struct ucred *a_cred;
    348 		struct proc *a_p;
    349 	} */ *ap;
    350 {
    351 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
    352 	struct vattr *vap = ap->a_vap;
    353 	struct proc *procp;
    354 	struct timeval tv;
    355 	int error;
    356 
    357 	/* first check the process still exists */
    358 	switch (pfs->pfs_type) {
    359 	case Proot:
    360 	case Pcurproc:
    361 		procp = 0;
    362 		break;
    363 
    364 	default:
    365 		procp = PFIND(pfs->pfs_pid);
    366 		if (procp == 0)
    367 			return (ENOENT);
    368 	}
    369 
    370 	error = 0;
    371 
    372 	/* start by zeroing out the attributes */
    373 	VATTR_NULL(vap);
    374 
    375 	/* next do all the common fields */
    376 	vap->va_type = ap->a_vp->v_type;
    377 	vap->va_mode = pfs->pfs_mode;
    378 	vap->va_fileid = pfs->pfs_fileno;
    379 	vap->va_flags = 0;
    380 	vap->va_blocksize = PAGE_SIZE;
    381 	vap->va_bytes = vap->va_size = 0;
    382 
    383 	/*
    384 	 * Make all times be current TOD.
    385 	 * It would be possible to get the process start
    386 	 * time from the p_stat structure, but there's
    387 	 * no "file creation" time stamp anyway, and the
    388 	 * p_stat structure is not addressible if u. gets
    389 	 * swapped out for that process.
    390 	 */
    391 	microtime(&tv);
    392 	TIMEVAL_TO_TIMESPEC(&tv, &vap->va_ctime);
    393 	vap->va_atime = vap->va_mtime = vap->va_ctime;
    394 
    395 	/*
    396 	 * If the process has exercised some setuid or setgid
    397 	 * privilege, then rip away read/write permission so
    398 	 * that only root can gain access.
    399 	 */
    400 	switch (pfs->pfs_type) {
    401 	case Pmem:
    402 	case Pregs:
    403 	case Pfpregs:
    404 		if (procp->p_flag & P_SUGID)
    405 			vap->va_mode &= ~((VREAD|VWRITE)|
    406 					  ((VREAD|VWRITE)>>3)|
    407 					  ((VREAD|VWRITE)>>6));
    408 	case Pctl:
    409 	case Pstatus:
    410 	case Pnote:
    411 	case Pnotepg:
    412 		vap->va_nlink = 1;
    413 		vap->va_uid = procp->p_ucred->cr_uid;
    414 		vap->va_gid = procp->p_ucred->cr_gid;
    415 		break;
    416 	}
    417 
    418 	/*
    419 	 * now do the object specific fields
    420 	 *
    421 	 * The size could be set from struct reg, but it's hardly
    422 	 * worth the trouble, and it puts some (potentially) machine
    423 	 * dependent data into this machine-independent code.  If it
    424 	 * becomes important then this function should break out into
    425 	 * a per-file stat function in the corresponding .c file.
    426 	 */
    427 
    428 	switch (pfs->pfs_type) {
    429 	case Proot:
    430 		/*
    431 		 * Set nlink to 1 to tell fts(3) we don't actually know.
    432 		 */
    433 		vap->va_nlink = 1;
    434 		vap->va_uid = 0;
    435 		vap->va_gid = 0;
    436 		vap->va_size = vap->va_bytes = DEV_BSIZE;
    437 		break;
    438 
    439 	case Pcurproc: {
    440 		char buf[16];		/* should be enough */
    441 		vap->va_nlink = 1;
    442 		vap->va_uid = 0;
    443 		vap->va_gid = 0;
    444 		vap->va_size = vap->va_bytes =
    445 		    sprintf(buf, "%ld", (long)curproc->p_pid);
    446 		break;
    447 	}
    448 
    449 	case Pproc:
    450 		vap->va_nlink = 2;
    451 		vap->va_uid = procp->p_ucred->cr_uid;
    452 		vap->va_gid = procp->p_ucred->cr_gid;
    453 		vap->va_size = vap->va_bytes = DEV_BSIZE;
    454 		break;
    455 
    456 	case Pfile:
    457 		error = EOPNOTSUPP;
    458 		break;
    459 
    460 	case Pmem:
    461 		vap->va_bytes = vap->va_size =
    462 			ctob(procp->p_vmspace->vm_tsize +
    463 				    procp->p_vmspace->vm_dsize +
    464 				    procp->p_vmspace->vm_ssize);
    465 		break;
    466 
    467 #if defined(PT_GETREGS) || defined(PT_SETREGS)
    468 	case Pregs:
    469 		vap->va_bytes = vap->va_size = sizeof(struct reg);
    470 		break;
    471 #endif
    472 
    473 #if defined(PT_GETFPREGS) || defined(PT_SETFPREGS)
    474 	case Pfpregs:
    475 		vap->va_bytes = vap->va_size = sizeof(struct fpreg);
    476 		break;
    477 #endif
    478 
    479 	case Pctl:
    480 	case Pstatus:
    481 	case Pnote:
    482 	case Pnotepg:
    483 		break;
    484 
    485 	default:
    486 		panic("procfs_getattr");
    487 	}
    488 
    489 	return (error);
    490 }
    491 
    492 procfs_setattr(ap)
    493 	struct vop_setattr_args /* {
    494 		struct vnode *a_vp;
    495 		struct vattr *a_vap;
    496 		struct ucred *a_cred;
    497 		struct proc *a_p;
    498 	} */ *ap;
    499 {
    500 	/*
    501 	 * just fake out attribute setting
    502 	 * it's not good to generate an error
    503 	 * return, otherwise things like creat()
    504 	 * will fail when they try to set the
    505 	 * file length to 0.  worse, this means
    506 	 * that echo $note > /proc/$pid/note will fail.
    507 	 */
    508 
    509 	return (0);
    510 }
    511 
    512 /*
    513  * implement access checking.
    514  *
    515  * actually, the check for super-user is slightly
    516  * broken since it will allow read access to write-only
    517  * objects.  this doesn't cause any particular trouble
    518  * but does mean that the i/o entry points need to check
    519  * that the operation really does make sense.
    520  */
    521 procfs_access(ap)
    522 	struct vop_access_args /* {
    523 		struct vnode *a_vp;
    524 		int a_mode;
    525 		struct ucred *a_cred;
    526 		struct proc *a_p;
    527 	} */ *ap;
    528 {
    529 	struct vattr va;
    530 	int error;
    531 
    532 	if (error = VOP_GETATTR(ap->a_vp, &va, ap->a_cred, ap->a_p))
    533 		return (error);
    534 
    535 	return (vaccess(va.va_mode, va.va_uid, va.va_gid, ap->a_mode,
    536 	    ap->a_cred));
    537 }
    538 
    539 /*
    540  * lookup.  this is incredibly complicated in the
    541  * general case, however for most pseudo-filesystems
    542  * very little needs to be done.
    543  *
    544  * unless you want to get a migraine, just make sure your
    545  * filesystem doesn't do any locking of its own.  otherwise
    546  * read and inwardly digest ufs_lookup().
    547  */
    548 procfs_lookup(ap)
    549 	struct vop_lookup_args /* {
    550 		struct vnode * a_dvp;
    551 		struct vnode ** a_vpp;
    552 		struct componentname * a_cnp;
    553 	} */ *ap;
    554 {
    555 	struct componentname *cnp = ap->a_cnp;
    556 	struct vnode **vpp = ap->a_vpp;
    557 	struct vnode *dvp = ap->a_dvp;
    558 	char *pname = cnp->cn_nameptr;
    559 	struct proc_target *pt;
    560 	struct vnode *fvp;
    561 	pid_t pid;
    562 	struct pfsnode *pfs;
    563 	struct proc *p;
    564 	int i;
    565 
    566 	*vpp = NULL;
    567 
    568 	if (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)
    569 		return (EROFS);
    570 
    571 	if (cnp->cn_namelen == 1 && *pname == '.') {
    572 		*vpp = dvp;
    573 		VREF(dvp);
    574 		/*VOP_LOCK(dvp);*/
    575 		return (0);
    576 	}
    577 
    578 	pfs = VTOPFS(dvp);
    579 	switch (pfs->pfs_type) {
    580 	case Proot:
    581 		if (cnp->cn_flags & ISDOTDOT)
    582 			return (EIO);
    583 
    584 		if (CNEQ(cnp, "curproc", 7))
    585 			return (procfs_allocvp(dvp->v_mount, vpp, 0, Pcurproc));
    586 
    587 		pid = atopid(pname, cnp->cn_namelen);
    588 		if (pid == NO_PID)
    589 			break;
    590 
    591 		p = PFIND(pid);
    592 		if (p == 0)
    593 			break;
    594 
    595 		return (procfs_allocvp(dvp->v_mount, vpp, pid, Pproc));
    596 
    597 	case Pproc:
    598 		if (cnp->cn_flags & ISDOTDOT)
    599 			return (procfs_root(dvp->v_mount, vpp));
    600 
    601 		p = PFIND(pfs->pfs_pid);
    602 		if (p == 0)
    603 			break;
    604 
    605 		for (pt = proc_targets, i = 0; i < nproc_targets; pt++, i++) {
    606 			if (cnp->cn_namelen == pt->pt_namlen &&
    607 			    bcmp(pt->pt_name, pname, cnp->cn_namelen) == 0 &&
    608 			    (pt->pt_valid == NULL || (*pt->pt_valid)(p)))
    609 				goto found;
    610 		}
    611 		break;
    612 
    613 	found:
    614 		if (pt->pt_pfstype == Pfile) {
    615 			fvp = procfs_findtextvp(p);
    616 			/* We already checked that it exists. */
    617 			VREF(fvp);
    618 			VOP_LOCK(fvp);
    619 			*vpp = fvp;
    620 			return (0);
    621 		}
    622 
    623 		return (procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
    624 		    pt->pt_pfstype));
    625 
    626 	default:
    627 		return (ENOTDIR);
    628 	}
    629 
    630 	return (cnp->cn_nameiop == LOOKUP ? ENOENT : EROFS);
    631 }
    632 
    633 int
    634 procfs_validfile(p)
    635 	struct proc *p;
    636 {
    637 
    638 	return (procfs_findtextvp(p) != NULLVP);
    639 }
    640 
    641 /*
    642  * readdir returns directory entries from pfsnode (vp).
    643  *
    644  * the strategy here with procfs is to generate a single
    645  * directory entry at a time (struct dirent) and then
    646  * copy that out to userland using uiomove.  a more efficent
    647  * though more complex implementation, would try to minimize
    648  * the number of calls to uiomove().  for procfs, this is
    649  * hardly worth the added code complexity.
    650  *
    651  * this should just be done through read()
    652  */
    653 procfs_readdir(ap)
    654 	struct vop_readdir_args /* {
    655 		struct vnode *a_vp;
    656 		struct uio *a_uio;
    657 		struct ucred *a_cred;
    658 		int *a_eofflag;
    659 		u_long *a_cookies;
    660 		int a_ncookies;
    661 	} */ *ap;
    662 {
    663 	struct uio *uio = ap->a_uio;
    664 	struct dirent d;
    665 	struct pfsnode *pfs;
    666 	int i;
    667 	int error;
    668 	u_long *cookies = ap->a_cookies;
    669 	int ncookies = ap->a_ncookies;
    670 
    671 	pfs = VTOPFS(ap->a_vp);
    672 
    673 	if (uio->uio_resid < UIO_MX)
    674 		return (EINVAL);
    675 	if (uio->uio_offset < 0)
    676 		return (EINVAL);
    677 
    678 	error = 0;
    679 	i = uio->uio_offset;
    680 	bzero((caddr_t)&d, UIO_MX);
    681 	d.d_reclen = UIO_MX;
    682 
    683 	switch (pfs->pfs_type) {
    684 	/*
    685 	 * this is for the process-specific sub-directories.
    686 	 * all that is needed to is copy out all the entries
    687 	 * from the procent[] table (top of this file).
    688 	 */
    689 	case Pproc: {
    690 		struct proc *p;
    691 		struct proc_target *pt;
    692 
    693 		p = PFIND(pfs->pfs_pid);
    694 		if (p == NULL)
    695 			break;
    696 
    697 		for (pt = &proc_targets[i];
    698 		     uio->uio_resid >= UIO_MX && i < nproc_targets; pt++, i++) {
    699 			if (pt->pt_valid && (*pt->pt_valid)(p) == 0)
    700 				continue;
    701 
    702 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid, pt->pt_pfstype);
    703 			d.d_namlen = pt->pt_namlen;
    704 			bcopy(pt->pt_name, d.d_name, pt->pt_namlen + 1);
    705 			d.d_type = pt->pt_type;
    706 
    707 			if (error = uiomove((caddr_t)&d, UIO_MX, uio))
    708 				break;
    709 			if (ncookies-- > 0)
    710 				*cookies++ = i + 1;
    711 		}
    712 
    713 	    	break;
    714 	}
    715 
    716 	/*
    717 	 * this is for the root of the procfs filesystem
    718 	 * what is needed is a special entry for "curproc"
    719 	 * followed by an entry for each process on allproc
    720 #ifdef PROCFS_ZOMBIE
    721 	 * and zombproc.
    722 #endif
    723 	 */
    724 
    725 	case Proot: {
    726 #ifdef PROCFS_ZOMBIE
    727 		int doingzomb = 0;
    728 #endif
    729 		int pcnt = i;
    730 		volatile struct proc *p = allproc.lh_first;
    731 
    732 		if (pcnt > 3)
    733 			pcnt = 3;
    734 	again:
    735 		for (; p && uio->uio_resid >= UIO_MX; i++, pcnt++) {
    736 			switch (i) {
    737 			case 0:		/* `.' */
    738 			case 1:		/* `..' */
    739 				d.d_fileno = PROCFS_FILENO(0, Proot);
    740 				d.d_namlen = i + 1;
    741 				bcopy("..", d.d_name, d.d_namlen);
    742 				d.d_name[i + 1] = '\0';
    743 				d.d_type = DT_DIR;
    744 				break;
    745 
    746 			case 2:
    747 				d.d_fileno = PROCFS_FILENO(0, Pcurproc);
    748 				d.d_namlen = 7;
    749 				bcopy("curproc", d.d_name, 8);
    750 				d.d_type = DT_LNK;
    751 				break;
    752 
    753 			default:
    754 				while (pcnt < i) {
    755 					pcnt++;
    756 					p = p->p_list.le_next;
    757 					if (!p)
    758 						goto done;
    759 				}
    760 				d.d_fileno = PROCFS_FILENO(p->p_pid, Pproc);
    761 				d.d_namlen = sprintf(d.d_name, "%ld",
    762 				    (long)p->p_pid);
    763 				d.d_type = DT_REG;
    764 				p = p->p_list.le_next;
    765 				break;
    766 			}
    767 
    768 			if (error = uiomove((caddr_t)&d, UIO_MX, uio))
    769 				break;
    770 			if (ncookies-- > 0)
    771 				*cookies++ = i + 1;
    772 		}
    773 	done:
    774 
    775 #ifdef PROCFS_ZOMBIE
    776 		if (p == 0 && doingzomb == 0) {
    777 			doingzomb = 1;
    778 			p = zombproc.lh_first;
    779 			goto again;
    780 		}
    781 #endif
    782 
    783 		break;
    784 
    785 	}
    786 
    787 	default:
    788 		error = ENOTDIR;
    789 		break;
    790 	}
    791 
    792 	uio->uio_offset = i;
    793 	return (error);
    794 }
    795 
    796 /*
    797  * readlink reads the link of `curproc'
    798  */
    799 procfs_readlink(ap)
    800 	struct vop_readlink_args *ap;
    801 {
    802 	struct uio *uio = ap->a_uio;
    803 	char buf[16];		/* should be enough */
    804 	int len;
    805 
    806 	if (VTOPFS(ap->a_vp)->pfs_fileno != PROCFS_FILENO(0, Pcurproc))
    807 		return (EINVAL);
    808 
    809 	len = sprintf(buf, "%ld", (long)curproc->p_pid);
    810 
    811 	return (uiomove((caddr_t)buf, len, ap->a_uio));
    812 }
    813 
    814 /*
    815  * convert decimal ascii to pid_t
    816  */
    817 static pid_t
    818 atopid(b, len)
    819 	const char *b;
    820 	u_int len;
    821 {
    822 	pid_t p = 0;
    823 
    824 	while (len--) {
    825 		char c = *b++;
    826 		if (c < '0' || c > '9')
    827 			return (NO_PID);
    828 		p = 10 * p + (c - '0');
    829 		if (p > PID_MAX)
    830 			return (NO_PID);
    831 	}
    832 
    833 	return (p);
    834 }
    835 
    836 /*
    837  * procfs vnode operations.
    838  */
    839 int (**procfs_vnodeop_p)();
    840 struct vnodeopv_entry_desc procfs_vnodeop_entries[] = {
    841 	{ &vop_default_desc, vn_default_error },
    842 	{ &vop_lookup_desc, procfs_lookup },		/* lookup */
    843 	{ &vop_create_desc, procfs_create },		/* create */
    844 	{ &vop_mknod_desc, procfs_mknod },		/* mknod */
    845 	{ &vop_open_desc, procfs_open },		/* open */
    846 	{ &vop_close_desc, procfs_close },		/* close */
    847 	{ &vop_access_desc, procfs_access },		/* access */
    848 	{ &vop_getattr_desc, procfs_getattr },		/* getattr */
    849 	{ &vop_setattr_desc, procfs_setattr },		/* setattr */
    850 	{ &vop_read_desc, procfs_read },		/* read */
    851 	{ &vop_write_desc, procfs_write },		/* write */
    852 	{ &vop_ioctl_desc, procfs_ioctl },		/* ioctl */
    853 	{ &vop_select_desc, procfs_select },		/* select */
    854 	{ &vop_mmap_desc, procfs_mmap },		/* mmap */
    855 	{ &vop_fsync_desc, procfs_fsync },		/* fsync */
    856 	{ &vop_seek_desc, procfs_seek },		/* seek */
    857 	{ &vop_remove_desc, procfs_remove },		/* remove */
    858 	{ &vop_link_desc, procfs_link },		/* link */
    859 	{ &vop_rename_desc, procfs_rename },		/* rename */
    860 	{ &vop_mkdir_desc, procfs_mkdir },		/* mkdir */
    861 	{ &vop_rmdir_desc, procfs_rmdir },		/* rmdir */
    862 	{ &vop_symlink_desc, procfs_symlink },		/* symlink */
    863 	{ &vop_readdir_desc, procfs_readdir },		/* readdir */
    864 	{ &vop_readlink_desc, procfs_readlink },	/* readlink */
    865 	{ &vop_abortop_desc, procfs_abortop },		/* abortop */
    866 	{ &vop_inactive_desc, procfs_inactive },	/* inactive */
    867 	{ &vop_reclaim_desc, procfs_reclaim },		/* reclaim */
    868 	{ &vop_lock_desc, procfs_lock },		/* lock */
    869 	{ &vop_unlock_desc, procfs_unlock },		/* unlock */
    870 	{ &vop_bmap_desc, procfs_bmap },		/* bmap */
    871 	{ &vop_strategy_desc, procfs_strategy },	/* strategy */
    872 	{ &vop_print_desc, procfs_print },		/* print */
    873 	{ &vop_islocked_desc, procfs_islocked },	/* islocked */
    874 	{ &vop_pathconf_desc, procfs_pathconf },	/* pathconf */
    875 	{ &vop_advlock_desc, procfs_advlock },		/* advlock */
    876 	{ &vop_blkatoff_desc, procfs_blkatoff },	/* blkatoff */
    877 	{ &vop_valloc_desc, procfs_valloc },		/* valloc */
    878 	{ &vop_vfree_desc, procfs_vfree },		/* vfree */
    879 	{ &vop_truncate_desc, procfs_truncate },	/* truncate */
    880 	{ &vop_update_desc, procfs_update },		/* update */
    881 	{ (struct vnodeop_desc*)NULL, (int(*)())NULL }
    882 };
    883 struct vnodeopv_desc procfs_vnodeop_opv_desc =
    884 	{ &procfs_vnodeop_p, procfs_vnodeop_entries };
    885