Home | History | Annotate | Line # | Download | only in procfs
procfs_vnops.c revision 1.36
      1 /*	$NetBSD: procfs_vnops.c,v 1.36 1996/02/09 14:45:53 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 int
    308 procfs_link(ap)
    309 	struct vop_link_args /* {
    310 		struct vnode *a_dvp;
    311 		struct vnode *a_vp;
    312 		struct componentname *a_cnp;
    313 	} */ *ap;
    314 {
    315 
    316 	VOP_ABORTOP(ap->a_dvp, ap->a_cnp);
    317 	vput(ap->a_dvp);
    318 	return (EROFS);
    319 }
    320 
    321 int
    322 procfs_symlink(ap)
    323 	struct vop_symlink_args /* {
    324 		struct vnode *a_dvp;
    325 		struct vnode **a_vpp;
    326 		struct componentname *a_cnp;
    327 		struct vattr *a_vap;
    328 		char *a_target;
    329 	} */ *ap;
    330 {
    331 
    332 	VOP_ABORTOP(ap->a_dvp, ap->a_cnp);
    333 	vput(ap->a_dvp);
    334 	return (EROFS);
    335 }
    336 
    337 /*
    338  * _abortop is called when operations such as
    339  * rename and create fail.  this entry is responsible
    340  * for undoing any side-effects caused by the lookup.
    341  * this will always include freeing the pathname buffer.
    342  */
    343 procfs_abortop(ap)
    344 	struct vop_abortop_args /* {
    345 		struct vnode *a_dvp;
    346 		struct componentname *a_cnp;
    347 	} */ *ap;
    348 {
    349 
    350 	if ((ap->a_cnp->cn_flags & (HASBUF | SAVESTART)) == HASBUF)
    351 		FREE(ap->a_cnp->cn_pnbuf, M_NAMEI);
    352 	return (0);
    353 }
    354 
    355 /*
    356  * generic entry point for unsupported operations
    357  */
    358 procfs_badop()
    359 {
    360 
    361 	return (EIO);
    362 }
    363 
    364 /*
    365  * Invent attributes for pfsnode (vp) and store
    366  * them in (vap).
    367  * Directories lengths are returned as zero since
    368  * any real length would require the genuine size
    369  * to be computed, and nothing cares anyway.
    370  *
    371  * this is relatively minimal for procfs.
    372  */
    373 procfs_getattr(ap)
    374 	struct vop_getattr_args /* {
    375 		struct vnode *a_vp;
    376 		struct vattr *a_vap;
    377 		struct ucred *a_cred;
    378 		struct proc *a_p;
    379 	} */ *ap;
    380 {
    381 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
    382 	struct vattr *vap = ap->a_vap;
    383 	struct proc *procp;
    384 	struct timeval tv;
    385 	int error;
    386 
    387 	/* first check the process still exists */
    388 	switch (pfs->pfs_type) {
    389 	case Proot:
    390 	case Pcurproc:
    391 		procp = 0;
    392 		break;
    393 
    394 	default:
    395 		procp = PFIND(pfs->pfs_pid);
    396 		if (procp == 0)
    397 			return (ENOENT);
    398 	}
    399 
    400 	error = 0;
    401 
    402 	/* start by zeroing out the attributes */
    403 	VATTR_NULL(vap);
    404 
    405 	/* next do all the common fields */
    406 	vap->va_type = ap->a_vp->v_type;
    407 	vap->va_mode = pfs->pfs_mode;
    408 	vap->va_fileid = pfs->pfs_fileno;
    409 	vap->va_flags = 0;
    410 	vap->va_blocksize = PAGE_SIZE;
    411 	vap->va_bytes = vap->va_size = 0;
    412 
    413 	/*
    414 	 * Make all times be current TOD.
    415 	 * It would be possible to get the process start
    416 	 * time from the p_stat structure, but there's
    417 	 * no "file creation" time stamp anyway, and the
    418 	 * p_stat structure is not addressible if u. gets
    419 	 * swapped out for that process.
    420 	 */
    421 	microtime(&tv);
    422 	TIMEVAL_TO_TIMESPEC(&tv, &vap->va_ctime);
    423 	vap->va_atime = vap->va_mtime = vap->va_ctime;
    424 
    425 	/*
    426 	 * If the process has exercised some setuid or setgid
    427 	 * privilege, then rip away read/write permission so
    428 	 * that only root can gain access.
    429 	 */
    430 	switch (pfs->pfs_type) {
    431 	case Pmem:
    432 	case Pregs:
    433 	case Pfpregs:
    434 		if (procp->p_flag & P_SUGID)
    435 			vap->va_mode &= ~((VREAD|VWRITE)|
    436 					  ((VREAD|VWRITE)>>3)|
    437 					  ((VREAD|VWRITE)>>6));
    438 	case Pctl:
    439 	case Pstatus:
    440 	case Pnote:
    441 	case Pnotepg:
    442 		vap->va_nlink = 1;
    443 		vap->va_uid = procp->p_ucred->cr_uid;
    444 		vap->va_gid = procp->p_ucred->cr_gid;
    445 		break;
    446 	}
    447 
    448 	/*
    449 	 * now do the object specific fields
    450 	 *
    451 	 * The size could be set from struct reg, but it's hardly
    452 	 * worth the trouble, and it puts some (potentially) machine
    453 	 * dependent data into this machine-independent code.  If it
    454 	 * becomes important then this function should break out into
    455 	 * a per-file stat function in the corresponding .c file.
    456 	 */
    457 
    458 	switch (pfs->pfs_type) {
    459 	case Proot:
    460 		/*
    461 		 * Set nlink to 1 to tell fts(3) we don't actually know.
    462 		 */
    463 		vap->va_nlink = 1;
    464 		vap->va_uid = 0;
    465 		vap->va_gid = 0;
    466 		vap->va_size = vap->va_bytes = DEV_BSIZE;
    467 		break;
    468 
    469 	case Pcurproc: {
    470 		char buf[16];		/* should be enough */
    471 		vap->va_nlink = 1;
    472 		vap->va_uid = 0;
    473 		vap->va_gid = 0;
    474 		vap->va_size = vap->va_bytes =
    475 		    sprintf(buf, "%ld", (long)curproc->p_pid);
    476 		break;
    477 	}
    478 
    479 	case Pproc:
    480 		vap->va_nlink = 2;
    481 		vap->va_uid = procp->p_ucred->cr_uid;
    482 		vap->va_gid = procp->p_ucred->cr_gid;
    483 		vap->va_size = vap->va_bytes = DEV_BSIZE;
    484 		break;
    485 
    486 	case Pfile:
    487 		error = EOPNOTSUPP;
    488 		break;
    489 
    490 	case Pmem:
    491 		vap->va_bytes = vap->va_size =
    492 			ctob(procp->p_vmspace->vm_tsize +
    493 				    procp->p_vmspace->vm_dsize +
    494 				    procp->p_vmspace->vm_ssize);
    495 		break;
    496 
    497 #if defined(PT_GETREGS) || defined(PT_SETREGS)
    498 	case Pregs:
    499 		vap->va_bytes = vap->va_size = sizeof(struct reg);
    500 		break;
    501 #endif
    502 
    503 #if defined(PT_GETFPREGS) || defined(PT_SETFPREGS)
    504 	case Pfpregs:
    505 		vap->va_bytes = vap->va_size = sizeof(struct fpreg);
    506 		break;
    507 #endif
    508 
    509 	case Pctl:
    510 	case Pstatus:
    511 	case Pnote:
    512 	case Pnotepg:
    513 		break;
    514 
    515 	default:
    516 		panic("procfs_getattr");
    517 	}
    518 
    519 	return (error);
    520 }
    521 
    522 procfs_setattr(ap)
    523 	struct vop_setattr_args /* {
    524 		struct vnode *a_vp;
    525 		struct vattr *a_vap;
    526 		struct ucred *a_cred;
    527 		struct proc *a_p;
    528 	} */ *ap;
    529 {
    530 	/*
    531 	 * just fake out attribute setting
    532 	 * it's not good to generate an error
    533 	 * return, otherwise things like creat()
    534 	 * will fail when they try to set the
    535 	 * file length to 0.  worse, this means
    536 	 * that echo $note > /proc/$pid/note will fail.
    537 	 */
    538 
    539 	return (0);
    540 }
    541 
    542 /*
    543  * implement access checking.
    544  *
    545  * actually, the check for super-user is slightly
    546  * broken since it will allow read access to write-only
    547  * objects.  this doesn't cause any particular trouble
    548  * but does mean that the i/o entry points need to check
    549  * that the operation really does make sense.
    550  */
    551 procfs_access(ap)
    552 	struct vop_access_args /* {
    553 		struct vnode *a_vp;
    554 		int a_mode;
    555 		struct ucred *a_cred;
    556 		struct proc *a_p;
    557 	} */ *ap;
    558 {
    559 	struct vattr va;
    560 	int error;
    561 
    562 	if (error = VOP_GETATTR(ap->a_vp, &va, ap->a_cred, ap->a_p))
    563 		return (error);
    564 
    565 	return (vaccess(va.va_mode, va.va_uid, va.va_gid, ap->a_mode,
    566 	    ap->a_cred));
    567 }
    568 
    569 /*
    570  * lookup.  this is incredibly complicated in the
    571  * general case, however for most pseudo-filesystems
    572  * very little needs to be done.
    573  *
    574  * unless you want to get a migraine, just make sure your
    575  * filesystem doesn't do any locking of its own.  otherwise
    576  * read and inwardly digest ufs_lookup().
    577  */
    578 procfs_lookup(ap)
    579 	struct vop_lookup_args /* {
    580 		struct vnode * a_dvp;
    581 		struct vnode ** a_vpp;
    582 		struct componentname * a_cnp;
    583 	} */ *ap;
    584 {
    585 	struct componentname *cnp = ap->a_cnp;
    586 	struct vnode **vpp = ap->a_vpp;
    587 	struct vnode *dvp = ap->a_dvp;
    588 	char *pname = cnp->cn_nameptr;
    589 	struct proc_target *pt;
    590 	struct vnode *fvp;
    591 	pid_t pid;
    592 	struct pfsnode *pfs;
    593 	struct proc *p;
    594 	int i;
    595 
    596 	*vpp = NULL;
    597 
    598 	if (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)
    599 		return (EROFS);
    600 
    601 	if (cnp->cn_namelen == 1 && *pname == '.') {
    602 		*vpp = dvp;
    603 		VREF(dvp);
    604 		/*VOP_LOCK(dvp);*/
    605 		return (0);
    606 	}
    607 
    608 	pfs = VTOPFS(dvp);
    609 	switch (pfs->pfs_type) {
    610 	case Proot:
    611 		if (cnp->cn_flags & ISDOTDOT)
    612 			return (EIO);
    613 
    614 		if (CNEQ(cnp, "curproc", 7))
    615 			return (procfs_allocvp(dvp->v_mount, vpp, 0, Pcurproc));
    616 
    617 		pid = atopid(pname, cnp->cn_namelen);
    618 		if (pid == NO_PID)
    619 			break;
    620 
    621 		p = PFIND(pid);
    622 		if (p == 0)
    623 			break;
    624 
    625 		return (procfs_allocvp(dvp->v_mount, vpp, pid, Pproc));
    626 
    627 	case Pproc:
    628 		if (cnp->cn_flags & ISDOTDOT)
    629 			return (procfs_root(dvp->v_mount, vpp));
    630 
    631 		p = PFIND(pfs->pfs_pid);
    632 		if (p == 0)
    633 			break;
    634 
    635 		for (pt = proc_targets, i = 0; i < nproc_targets; pt++, i++) {
    636 			if (cnp->cn_namelen == pt->pt_namlen &&
    637 			    bcmp(pt->pt_name, pname, cnp->cn_namelen) == 0 &&
    638 			    (pt->pt_valid == NULL || (*pt->pt_valid)(p)))
    639 				goto found;
    640 		}
    641 		break;
    642 
    643 	found:
    644 		if (pt->pt_pfstype == Pfile) {
    645 			fvp = procfs_findtextvp(p);
    646 			/* We already checked that it exists. */
    647 			VREF(fvp);
    648 			VOP_LOCK(fvp);
    649 			*vpp = fvp;
    650 			return (0);
    651 		}
    652 
    653 		return (procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
    654 		    pt->pt_pfstype));
    655 
    656 	default:
    657 		return (ENOTDIR);
    658 	}
    659 
    660 	return (cnp->cn_nameiop == LOOKUP ? ENOENT : EROFS);
    661 }
    662 
    663 int
    664 procfs_validfile(p)
    665 	struct proc *p;
    666 {
    667 
    668 	return (procfs_findtextvp(p) != NULLVP);
    669 }
    670 
    671 /*
    672  * readdir returns directory entries from pfsnode (vp).
    673  *
    674  * the strategy here with procfs is to generate a single
    675  * directory entry at a time (struct dirent) and then
    676  * copy that out to userland using uiomove.  a more efficent
    677  * though more complex implementation, would try to minimize
    678  * the number of calls to uiomove().  for procfs, this is
    679  * hardly worth the added code complexity.
    680  *
    681  * this should just be done through read()
    682  */
    683 procfs_readdir(ap)
    684 	struct vop_readdir_args /* {
    685 		struct vnode *a_vp;
    686 		struct uio *a_uio;
    687 		struct ucred *a_cred;
    688 		int *a_eofflag;
    689 		u_long *a_cookies;
    690 		int a_ncookies;
    691 	} */ *ap;
    692 {
    693 	struct uio *uio = ap->a_uio;
    694 	struct dirent d;
    695 	struct pfsnode *pfs;
    696 	int i;
    697 	int error;
    698 	u_long *cookies = ap->a_cookies;
    699 	int ncookies = ap->a_ncookies;
    700 
    701 	pfs = VTOPFS(ap->a_vp);
    702 
    703 	if (uio->uio_resid < UIO_MX)
    704 		return (EINVAL);
    705 	if (uio->uio_offset < 0)
    706 		return (EINVAL);
    707 
    708 	error = 0;
    709 	i = uio->uio_offset;
    710 	bzero((caddr_t)&d, UIO_MX);
    711 	d.d_reclen = UIO_MX;
    712 
    713 	switch (pfs->pfs_type) {
    714 	/*
    715 	 * this is for the process-specific sub-directories.
    716 	 * all that is needed to is copy out all the entries
    717 	 * from the procent[] table (top of this file).
    718 	 */
    719 	case Pproc: {
    720 		struct proc *p;
    721 		struct proc_target *pt;
    722 
    723 		p = PFIND(pfs->pfs_pid);
    724 		if (p == NULL)
    725 			break;
    726 
    727 		for (pt = &proc_targets[i];
    728 		     uio->uio_resid >= UIO_MX && i < nproc_targets; pt++, i++) {
    729 			if (pt->pt_valid && (*pt->pt_valid)(p) == 0)
    730 				continue;
    731 
    732 			d.d_fileno = PROCFS_FILENO(pfs->pfs_pid, pt->pt_pfstype);
    733 			d.d_namlen = pt->pt_namlen;
    734 			bcopy(pt->pt_name, d.d_name, pt->pt_namlen + 1);
    735 			d.d_type = pt->pt_type;
    736 
    737 			if (error = uiomove((caddr_t)&d, UIO_MX, uio))
    738 				break;
    739 			if (ncookies-- > 0)
    740 				*cookies++ = i + 1;
    741 		}
    742 
    743 	    	break;
    744 	}
    745 
    746 	/*
    747 	 * this is for the root of the procfs filesystem
    748 	 * what is needed is a special entry for "curproc"
    749 	 * followed by an entry for each process on allproc
    750 #ifdef PROCFS_ZOMBIE
    751 	 * and zombproc.
    752 #endif
    753 	 */
    754 
    755 	case Proot: {
    756 #ifdef PROCFS_ZOMBIE
    757 		int doingzomb = 0;
    758 #endif
    759 		int pcnt = i;
    760 		volatile struct proc *p = allproc.lh_first;
    761 
    762 		if (pcnt > 3)
    763 			pcnt = 3;
    764 	again:
    765 		for (; p && uio->uio_resid >= UIO_MX; i++, pcnt++) {
    766 			switch (i) {
    767 			case 0:		/* `.' */
    768 			case 1:		/* `..' */
    769 				d.d_fileno = PROCFS_FILENO(0, Proot);
    770 				d.d_namlen = i + 1;
    771 				bcopy("..", d.d_name, d.d_namlen);
    772 				d.d_name[i + 1] = '\0';
    773 				d.d_type = DT_DIR;
    774 				break;
    775 
    776 			case 2:
    777 				d.d_fileno = PROCFS_FILENO(0, Pcurproc);
    778 				d.d_namlen = 7;
    779 				bcopy("curproc", d.d_name, 8);
    780 				d.d_type = DT_LNK;
    781 				break;
    782 
    783 			default:
    784 				while (pcnt < i) {
    785 					pcnt++;
    786 					p = p->p_list.le_next;
    787 					if (!p)
    788 						goto done;
    789 				}
    790 				d.d_fileno = PROCFS_FILENO(p->p_pid, Pproc);
    791 				d.d_namlen = sprintf(d.d_name, "%ld",
    792 				    (long)p->p_pid);
    793 				d.d_type = DT_REG;
    794 				p = p->p_list.le_next;
    795 				break;
    796 			}
    797 
    798 			if (error = uiomove((caddr_t)&d, UIO_MX, uio))
    799 				break;
    800 			if (ncookies-- > 0)
    801 				*cookies++ = i + 1;
    802 		}
    803 	done:
    804 
    805 #ifdef PROCFS_ZOMBIE
    806 		if (p == 0 && doingzomb == 0) {
    807 			doingzomb = 1;
    808 			p = zombproc.lh_first;
    809 			goto again;
    810 		}
    811 #endif
    812 
    813 		break;
    814 
    815 	}
    816 
    817 	default:
    818 		error = ENOTDIR;
    819 		break;
    820 	}
    821 
    822 	uio->uio_offset = i;
    823 	return (error);
    824 }
    825 
    826 /*
    827  * readlink reads the link of `curproc'
    828  */
    829 procfs_readlink(ap)
    830 	struct vop_readlink_args *ap;
    831 {
    832 	struct uio *uio = ap->a_uio;
    833 	char buf[16];		/* should be enough */
    834 	int len;
    835 
    836 	if (VTOPFS(ap->a_vp)->pfs_fileno != PROCFS_FILENO(0, Pcurproc))
    837 		return (EINVAL);
    838 
    839 	len = sprintf(buf, "%ld", (long)curproc->p_pid);
    840 
    841 	return (uiomove((caddr_t)buf, len, ap->a_uio));
    842 }
    843 
    844 /*
    845  * convert decimal ascii to pid_t
    846  */
    847 static pid_t
    848 atopid(b, len)
    849 	const char *b;
    850 	u_int len;
    851 {
    852 	pid_t p = 0;
    853 
    854 	while (len--) {
    855 		char c = *b++;
    856 		if (c < '0' || c > '9')
    857 			return (NO_PID);
    858 		p = 10 * p + (c - '0');
    859 		if (p > PID_MAX)
    860 			return (NO_PID);
    861 	}
    862 
    863 	return (p);
    864 }
    865 
    866 /*
    867  * procfs vnode operations.
    868  */
    869 int (**procfs_vnodeop_p)();
    870 struct vnodeopv_entry_desc procfs_vnodeop_entries[] = {
    871 	{ &vop_default_desc, vn_default_error },
    872 	{ &vop_lookup_desc, procfs_lookup },		/* lookup */
    873 	{ &vop_create_desc, procfs_create },		/* create */
    874 	{ &vop_mknod_desc, procfs_mknod },		/* mknod */
    875 	{ &vop_open_desc, procfs_open },		/* open */
    876 	{ &vop_close_desc, procfs_close },		/* close */
    877 	{ &vop_access_desc, procfs_access },		/* access */
    878 	{ &vop_getattr_desc, procfs_getattr },		/* getattr */
    879 	{ &vop_setattr_desc, procfs_setattr },		/* setattr */
    880 	{ &vop_read_desc, procfs_read },		/* read */
    881 	{ &vop_write_desc, procfs_write },		/* write */
    882 	{ &vop_ioctl_desc, procfs_ioctl },		/* ioctl */
    883 	{ &vop_select_desc, procfs_select },		/* select */
    884 	{ &vop_mmap_desc, procfs_mmap },		/* mmap */
    885 	{ &vop_fsync_desc, procfs_fsync },		/* fsync */
    886 	{ &vop_seek_desc, procfs_seek },		/* seek */
    887 	{ &vop_remove_desc, procfs_remove },		/* remove */
    888 	{ &vop_link_desc, procfs_link },		/* link */
    889 	{ &vop_rename_desc, procfs_rename },		/* rename */
    890 	{ &vop_mkdir_desc, procfs_mkdir },		/* mkdir */
    891 	{ &vop_rmdir_desc, procfs_rmdir },		/* rmdir */
    892 	{ &vop_symlink_desc, procfs_symlink },		/* symlink */
    893 	{ &vop_readdir_desc, procfs_readdir },		/* readdir */
    894 	{ &vop_readlink_desc, procfs_readlink },	/* readlink */
    895 	{ &vop_abortop_desc, procfs_abortop },		/* abortop */
    896 	{ &vop_inactive_desc, procfs_inactive },	/* inactive */
    897 	{ &vop_reclaim_desc, procfs_reclaim },		/* reclaim */
    898 	{ &vop_lock_desc, procfs_lock },		/* lock */
    899 	{ &vop_unlock_desc, procfs_unlock },		/* unlock */
    900 	{ &vop_bmap_desc, procfs_bmap },		/* bmap */
    901 	{ &vop_strategy_desc, procfs_strategy },	/* strategy */
    902 	{ &vop_print_desc, procfs_print },		/* print */
    903 	{ &vop_islocked_desc, procfs_islocked },	/* islocked */
    904 	{ &vop_pathconf_desc, procfs_pathconf },	/* pathconf */
    905 	{ &vop_advlock_desc, procfs_advlock },		/* advlock */
    906 	{ &vop_blkatoff_desc, procfs_blkatoff },	/* blkatoff */
    907 	{ &vop_valloc_desc, procfs_valloc },		/* valloc */
    908 	{ &vop_vfree_desc, procfs_vfree },		/* vfree */
    909 	{ &vop_truncate_desc, procfs_truncate },	/* truncate */
    910 	{ &vop_update_desc, procfs_update },		/* update */
    911 	{ (struct vnodeop_desc*)NULL, (int(*)())NULL }
    912 };
    913 struct vnodeopv_desc procfs_vnodeop_opv_desc =
    914 	{ &procfs_vnodeop_p, procfs_vnodeop_entries };
    915