Home | History | Annotate | Line # | Download | only in procfs
procfs_vnops.c revision 1.28
      1 /*	$NetBSD: procfs_vnops.c,v 1.28 1994/11/14 06:08:21 christos 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 static struct pfsnames {
     73 	u_char	d_type;
     74 	u_char	d_namlen;
     75 	char	d_name[PROCFS_NAMELEN];
     76 	pfstype	d_pfstype;
     77 	int	(*d_valid) __P((struct proc *p));
     78 } procent[] = {
     79 #define N(s) sizeof(s)-1, s
     80 	/* namlen, nam, type */
     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 #define Nprocent (sizeof(procent)/sizeof(procent[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 		struct file *a_fp;
    115 	} */ *ap;
    116 {
    117 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
    118 
    119 	switch (pfs->pfs_type) {
    120 	case Pmem:
    121 		if (PFIND(pfs->pfs_pid) == 0)
    122 			return (ENOENT);	/* was ESRCH, jsp */
    123 
    124 		if ((pfs->pfs_flags & FWRITE) && (ap->a_mode & O_EXCL) ||
    125 		    (pfs->pfs_flags & O_EXCL) && (ap->a_mode & FWRITE))
    126 			return (EBUSY);
    127 
    128 		if (ap->a_mode & FWRITE)
    129 			pfs->pfs_flags = ap->a_mode & (FWRITE|O_EXCL);
    130 
    131 		return (0);
    132 
    133 	default:
    134 		break;
    135 	}
    136 
    137 	return (0);
    138 }
    139 
    140 /*
    141  * close the pfsnode (vp) after doing i/o.
    142  * (vp) is not locked on entry or exit.
    143  *
    144  * nothing to do for procfs other than undo
    145  * any exclusive open flag (see _open above).
    146  */
    147 procfs_close(ap)
    148 	struct vop_close_args /* {
    149 		struct vnode *a_vp;
    150 		int  a_fflag;
    151 		struct ucred *a_cred;
    152 		struct proc *a_p;
    153 	} */ *ap;
    154 {
    155 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
    156 
    157 	switch (pfs->pfs_type) {
    158 	case Pmem:
    159 		if ((ap->a_fflag & FWRITE) && (pfs->pfs_flags & O_EXCL))
    160 			pfs->pfs_flags &= ~(FWRITE|O_EXCL);
    161 		break;
    162 	}
    163 
    164 	return (0);
    165 }
    166 
    167 /*
    168  * do an ioctl operation on pfsnode (vp).
    169  * (vp) is not locked on entry or exit.
    170  */
    171 procfs_ioctl(ap)
    172 	struct vop_ioctl_args /* {
    173 		struct vnode *a_vp;
    174 		u_long a_command;
    175 		caddr_t a_data;
    176 		int a_fflag;
    177 		struct ucred *a_cred;
    178 		struct proc *a_p;
    179 	} */ *ap;
    180 {
    181 
    182 	return (ENOTTY);
    183 }
    184 
    185 /*
    186  * do block mapping for pfsnode (vp).
    187  * since we don't use the buffer cache
    188  * for procfs this function should never
    189  * be called.  in any case, it's not clear
    190  * what part of the kernel ever makes use
    191  * of this function.  for sanity, this is the
    192  * usual no-op bmap, although returning
    193  * (EIO) would be a reasonable alternative.
    194  */
    195 procfs_bmap(ap)
    196 	struct vop_bmap_args /* {
    197 		struct vnode *a_vp;
    198 		daddr_t  a_bn;
    199 		struct vnode **a_vpp;
    200 		daddr_t *a_bnp;
    201 	} */ *ap;
    202 {
    203 
    204 	if (ap->a_vpp != NULL)
    205 		*ap->a_vpp = ap->a_vp;
    206 	if (ap->a_bnp != NULL)
    207 		*ap->a_bnp = ap->a_bn;
    208 	return (0);
    209 }
    210 
    211 /*
    212  * _inactive is called when the pfsnode
    213  * is vrele'd and the reference count goes
    214  * to zero.  (vp) will be on the vnode free
    215  * list, so to get it back vget() must be
    216  * used.
    217  *
    218  * for procfs, check if the process is still
    219  * alive and if it isn't then just throw away
    220  * the vnode by calling vgone().  this may
    221  * be overkill and a waste of time since the
    222  * chances are that the process will still be
    223  * there and PFIND is not free.
    224  *
    225  * (vp) is not locked on entry or exit.
    226  */
    227 procfs_inactive(ap)
    228 	struct vop_inactive_args /* {
    229 		struct vnode *a_vp;
    230 	} */ *ap;
    231 {
    232 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
    233 
    234 	if (PFIND(pfs->pfs_pid) == 0)
    235 		vgone(ap->a_vp);
    236 
    237 	return (0);
    238 }
    239 
    240 /*
    241  * _reclaim is called when getnewvnode()
    242  * wants to make use of an entry on the vnode
    243  * free list.  at this time the filesystem needs
    244  * to free any private data and remove the node
    245  * from any private lists.
    246  */
    247 procfs_reclaim(ap)
    248 	struct vop_reclaim_args /* {
    249 		struct vnode *a_vp;
    250 	} */ *ap;
    251 {
    252 
    253 	return (procfs_freevp(ap->a_vp));
    254 }
    255 
    256 /*
    257  * Return POSIX pathconf information applicable to special devices.
    258  */
    259 procfs_pathconf(ap)
    260 	struct vop_pathconf_args /* {
    261 		struct vnode *a_vp;
    262 		int a_name;
    263 		register_t *a_retval;
    264 	} */ *ap;
    265 {
    266 
    267 	switch (ap->a_name) {
    268 	case _PC_LINK_MAX:
    269 		*ap->a_retval = LINK_MAX;
    270 		return (0);
    271 	case _PC_MAX_CANON:
    272 		*ap->a_retval = MAX_CANON;
    273 		return (0);
    274 	case _PC_MAX_INPUT:
    275 		*ap->a_retval = MAX_INPUT;
    276 		return (0);
    277 	case _PC_PIPE_BUF:
    278 		*ap->a_retval = PIPE_BUF;
    279 		return (0);
    280 	case _PC_CHOWN_RESTRICTED:
    281 		*ap->a_retval = 1;
    282 		return (0);
    283 	case _PC_VDISABLE:
    284 		*ap->a_retval = _POSIX_VDISABLE;
    285 		return (0);
    286 	default:
    287 		return (EINVAL);
    288 	}
    289 	/* NOTREACHED */
    290 }
    291 
    292 /*
    293  * _print is used for debugging.
    294  * just print a readable description
    295  * of (vp).
    296  */
    297 procfs_print(ap)
    298 	struct vop_print_args /* {
    299 		struct vnode *a_vp;
    300 	} */ *ap;
    301 {
    302 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
    303 
    304 	printf("tag VT_PROCFS, type %s, pid %d, mode %x, flags %x\n",
    305 	    pfs->pfs_type, pfs->pfs_pid, pfs->pfs_mode, pfs->pfs_flags);
    306 }
    307 
    308 /*
    309  * _abortop is called when operations such as
    310  * rename and create fail.  this entry is responsible
    311  * for undoing any side-effects caused by the lookup.
    312  * this will always include freeing the pathname buffer.
    313  */
    314 procfs_abortop(ap)
    315 	struct vop_abortop_args /* {
    316 		struct vnode *a_dvp;
    317 		struct componentname *a_cnp;
    318 	} */ *ap;
    319 {
    320 
    321 	if ((ap->a_cnp->cn_flags & (HASBUF | SAVESTART)) == HASBUF)
    322 		FREE(ap->a_cnp->cn_pnbuf, M_NAMEI);
    323 	return (0);
    324 }
    325 
    326 /*
    327  * generic entry point for unsupported operations
    328  */
    329 procfs_badop()
    330 {
    331 
    332 	return (EIO);
    333 }
    334 
    335 /*
    336  * Invent attributes for pfsnode (vp) and store
    337  * them in (vap).
    338  * Directories lengths are returned as zero since
    339  * any real length would require the genuine size
    340  * to be computed, and nothing cares anyway.
    341  *
    342  * this is relatively minimal for procfs.
    343  */
    344 procfs_getattr(ap)
    345 	struct vop_getattr_args /* {
    346 		struct vnode *a_vp;
    347 		struct vattr *a_vap;
    348 		struct ucred *a_cred;
    349 		struct proc *a_p;
    350 	} */ *ap;
    351 {
    352 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
    353 	struct vattr *vap = ap->a_vap;
    354 	struct proc *procp;
    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 	 * XXX
    392 	 * Note that microtime() returns a timeval, not a timespec.
    393 	 */
    394 	microtime(&vap->va_ctime);
    395 	vap->va_atime = vap->va_mtime = vap->va_ctime;
    396 
    397 	/*
    398 	 * If the process has exercised some setuid or setgid
    399 	 * privilege, then rip away read/write permission so
    400 	 * that only root can gain access.
    401 	 */
    402 	switch (pfs->pfs_type) {
    403 	case Pmem:
    404 	case Pregs:
    405 	case Pfpregs:
    406 		if (procp->p_flag & P_SUGID)
    407 			vap->va_mode &= ~((VREAD|VWRITE)|
    408 					  ((VREAD|VWRITE)>>3)|
    409 					  ((VREAD|VWRITE)>>6));
    410 	case Pctl:
    411 	case Pstatus:
    412 	case Pnote:
    413 	case Pnotepg:
    414 		vap->va_nlink = 1;
    415 		vap->va_uid = procp->p_ucred->cr_uid;
    416 		vap->va_gid = procp->p_ucred->cr_gid;
    417 		break;
    418 	}
    419 
    420 	/*
    421 	 * now do the object specific fields
    422 	 *
    423 	 * The size could be set from struct reg, but it's hardly
    424 	 * worth the trouble, and it puts some (potentially) machine
    425 	 * dependent data into this machine-independent code.  If it
    426 	 * becomes important then this function should break out into
    427 	 * a per-file stat function in the corresponding .c file.
    428 	 */
    429 
    430 	switch (pfs->pfs_type) {
    431 	case Proot:
    432 		/*
    433 		 * Set nlink to 1 to tell fts(3) we don't actually know.
    434 		 */
    435 		vap->va_nlink = 1;
    436 		vap->va_uid = 0;
    437 		vap->va_gid = 0;
    438 		vap->va_size = vap->va_bytes = DEV_BSIZE;
    439 		break;
    440 
    441 	case Pcurproc: {
    442 		char buf[16];		/* should be enough */
    443 		vap->va_nlink = 1;
    444 		vap->va_uid = 0;
    445 		vap->va_gid = 0;
    446 		vap->va_size = vap->va_bytes =
    447 		    sprintf(buf, "%ld", (long)curproc->p_pid);
    448 		break;
    449 	}
    450 
    451 	case Pproc:
    452 		vap->va_nlink = 2;
    453 		vap->va_uid = procp->p_ucred->cr_uid;
    454 		vap->va_gid = procp->p_ucred->cr_gid;
    455 		vap->va_size = vap->va_bytes = DEV_BSIZE;
    456 		break;
    457 
    458 	case Pfile:
    459 		error = EOPNOTSUPP;
    460 		break;
    461 
    462 	case Pmem:
    463 		vap->va_bytes = vap->va_size =
    464 			ctob(procp->p_vmspace->vm_tsize +
    465 				    procp->p_vmspace->vm_dsize +
    466 				    procp->p_vmspace->vm_ssize);
    467 		break;
    468 
    469 #if defined(PT_GETREGS) || defined(PT_SETREGS)
    470 	case Pregs:
    471 		vap->va_bytes = vap->va_size = sizeof(struct reg);
    472 		break;
    473 #endif
    474 
    475 #if defined(PT_GETFPREGS) || defined(PT_SETFPREGS)
    476 	case Pfpregs:
    477 		vap->va_bytes = vap->va_size = sizeof(struct fpreg);
    478 		break;
    479 #endif
    480 
    481 	case Pctl:
    482 	case Pstatus:
    483 	case Pnote:
    484 	case Pnotepg:
    485 		break;
    486 
    487 	default:
    488 		panic("procfs_getattr");
    489 	}
    490 
    491 	return (error);
    492 }
    493 
    494 procfs_setattr(ap)
    495 	struct vop_setattr_args /* {
    496 		struct vnode *a_vp;
    497 		struct vattr *a_vap;
    498 		struct ucred *a_cred;
    499 		struct proc *a_p;
    500 	} */ *ap;
    501 {
    502 	/*
    503 	 * just fake out attribute setting
    504 	 * it's not good to generate an error
    505 	 * return, otherwise things like creat()
    506 	 * will fail when they try to set the
    507 	 * file length to 0.  worse, this means
    508 	 * that echo $note > /proc/$pid/note will fail.
    509 	 */
    510 
    511 	return (0);
    512 }
    513 
    514 /*
    515  * implement access checking.
    516  *
    517  * something very similar to this code is duplicated
    518  * throughout the 4bsd kernel and should be moved
    519  * into kern/vfs_subr.c sometime.
    520  *
    521  * actually, the check for super-user is slightly
    522  * broken since it will allow read access to write-only
    523  * objects.  this doesn't cause any particular trouble
    524  * but does mean that the i/o entry points need to check
    525  * that the operation really does make sense.
    526  */
    527 procfs_access(ap)
    528 	struct vop_access_args /* {
    529 		struct vnode *a_vp;
    530 		int a_mode;
    531 		struct ucred *a_cred;
    532 		struct proc *a_p;
    533 	} */ *ap;
    534 {
    535 	struct vattr *vap;
    536 	struct vattr vattr;
    537 	int error;
    538 
    539 	/*
    540 	 * If you're the super-user,
    541 	 * you always get access.
    542 	 */
    543 	if (ap->a_cred->cr_uid == 0)
    544 		return (0);
    545 
    546 	vap = &vattr;
    547 	if (error = VOP_GETATTR(ap->a_vp, vap, ap->a_cred, ap->a_p))
    548 		return (error);
    549 
    550 	/*
    551 	 * Access check is based on only one of owner, group, public.
    552 	 * If not owner, then check group. If not a member of the
    553 	 * group, then check public access.
    554 	 */
    555 	if (ap->a_cred->cr_uid != vap->va_uid) {
    556 		gid_t *gp;
    557 		int i;
    558 
    559 		ap->a_mode >>= 3;
    560 		gp = ap->a_cred->cr_groups;
    561 		for (i = 0; i < ap->a_cred->cr_ngroups; i++, gp++)
    562 			if (vap->va_gid == *gp)
    563 				goto found;
    564 		ap->a_mode >>= 3;
    565 found:
    566 		;
    567 	}
    568 
    569 	if ((vap->va_mode & ap->a_mode) == ap->a_mode)
    570 		return (0);
    571 
    572 	return (EACCES);
    573 }
    574 
    575 /*
    576  * lookup.  this is incredibly complicated in the
    577  * general case, however for most pseudo-filesystems
    578  * very little needs to be done.
    579  *
    580  * unless you want to get a migraine, just make sure your
    581  * filesystem doesn't do any locking of its own.  otherwise
    582  * read and inwardly digest ufs_lookup().
    583  */
    584 procfs_lookup(ap)
    585 	struct vop_lookup_args /* {
    586 		struct vnode * a_dvp;
    587 		struct vnode ** a_vpp;
    588 		struct componentname * a_cnp;
    589 	} */ *ap;
    590 {
    591 	struct componentname *cnp = ap->a_cnp;
    592 	struct vnode **vpp = ap->a_vpp;
    593 	struct vnode *dvp = ap->a_dvp;
    594 	char *pname = cnp->cn_nameptr;
    595 	pid_t pid;
    596 	struct vnode *nvp;
    597 	struct pfsnode *pfs;
    598 	struct proc *procp;
    599 	pfstype pfs_type;
    600 	int i;
    601 
    602 	if (cnp->cn_namelen == 1 && *pname == '.') {
    603 		*vpp = dvp;
    604 		VREF(dvp);
    605 		/*VOP_LOCK(dvp);*/
    606 		return (0);
    607 	}
    608 
    609 	*vpp = NULL;
    610 
    611 	pfs = VTOPFS(dvp);
    612 	switch (pfs->pfs_type) {
    613 	case Proot:
    614 		if (cnp->cn_flags & ISDOTDOT)
    615 			return (EIO);
    616 
    617 		if (CNEQ(cnp, "curproc", 7))
    618 			return (procfs_allocvp(dvp->v_mount, vpp, 0, Pcurproc));
    619 
    620 		pid = atopid(pname, cnp->cn_namelen);
    621 		if (pid == NO_PID)
    622 			return (ENOENT);
    623 
    624 		procp = PFIND(pid);
    625 		if (procp == 0)
    626 			return (ENOENT);
    627 
    628 		return (procfs_allocvp(dvp->v_mount, vpp, pid, Pproc));
    629 
    630 	case Pproc:
    631 		if (cnp->cn_flags & ISDOTDOT)
    632 			return (procfs_root(dvp->v_mount, vpp));
    633 
    634 		procp = PFIND(pfs->pfs_pid);
    635 		if (procp == 0)
    636 			return (ENOENT);
    637 
    638 		for (i = 0; i < Nprocent; i++) {
    639 			struct pfsnames *dp = &procent[i];
    640 
    641 			if (cnp->cn_namelen == dp->d_namlen &&
    642 			    bcmp(pname, dp->d_name, dp->d_namlen) == 0 &&
    643 			    (dp->d_valid == NULL || (*dp->d_valid)(procp))) {
    644 			    	pfs_type = dp->d_pfstype;
    645 				goto found;
    646 			}
    647 		}
    648 		return (ENOENT);
    649 
    650 	found:
    651 		if (pfs_type == Pfile) {
    652 			nvp = procfs_findtextvp(procp);
    653 			/* We already checked that it exists. */
    654 			VREF(nvp);
    655 			VOP_LOCK(nvp);
    656 			*vpp = nvp;
    657 			return (0);
    658 		}
    659 
    660 		return (procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
    661 		    pfs_type));
    662 
    663 	default:
    664 		return (ENOTDIR);
    665 	}
    666 }
    667 
    668 int
    669 procfs_validfile(p)
    670 	struct proc *p;
    671 {
    672 
    673 	return (procfs_findtextvp(p) != NULLVP);
    674 }
    675 
    676 /*
    677  * readdir returns directory entries from pfsnode (vp).
    678  *
    679  * the strategy here with procfs is to generate a single
    680  * directory entry at a time (struct pfsdent) and then
    681  * copy that out to userland using uiomove.  a more efficent
    682  * though more complex implementation, would try to minimize
    683  * the number of calls to uiomove().  for procfs, this is
    684  * hardly worth the added code complexity.
    685  *
    686  * this should just be done through read()
    687  */
    688 procfs_readdir(ap)
    689 	struct vop_readdir_args /* {
    690 		struct vnode *a_vp;
    691 		struct uio *a_uio;
    692 		struct ucred *a_cred;
    693 		int *a_eofflag;
    694 		u_long *a_cookies;
    695 		int a_ncookies;
    696 	} */ *ap;
    697 {
    698 	struct uio *uio = ap->a_uio;
    699 	struct pfsdent d;
    700 	struct pfsdent *dp = &d;
    701 	struct pfsnode *pfs;
    702 	int error;
    703 	int count;
    704 	int i;
    705 
    706 	/*
    707 	 * We don't allow exporting procfs mounts, and currently local
    708 	 * requests do not need cookies.
    709 	 */
    710 	if (ap->a_ncookies)
    711 		panic("procfs_readdir: not hungry");
    712 
    713 	pfs = VTOPFS(ap->a_vp);
    714 
    715 	if (uio->uio_resid < UIO_MX)
    716 		return (EINVAL);
    717 	if (uio->uio_offset & (UIO_MX-1))
    718 		return (EINVAL);
    719 	if (uio->uio_offset < 0)
    720 		return (EINVAL);
    721 
    722 	error = 0;
    723 	count = 0;
    724 	i = uio->uio_offset / UIO_MX;
    725 
    726 	switch (pfs->pfs_type) {
    727 	/*
    728 	 * this is for the process-specific sub-directories.
    729 	 * all that is needed to is copy out all the entries
    730 	 * from the procent[] table (top of this file).
    731 	 */
    732 	case Pproc: {
    733 		pid_t pid = pfs->pfs_pid;
    734 		struct pfsnames *dt;
    735 
    736 		for (dt = &procent[i]; i < Nprocent && uio->uio_resid >= UIO_MX;
    737 		     dt++, i++) {
    738 			struct proc *p = PFIND(pid);
    739 
    740 			if (p == NULL)
    741 				break;
    742 
    743 			if (dt->d_valid && (*dt->d_valid)(p) == 0)
    744 				continue;
    745 
    746 			dp->d_reclen = UIO_MX;
    747 			dp->d_fileno = PROCFS_FILENO(pid, dt->d_pfstype);
    748 			dp->d_namlen = dt->d_namlen;
    749 			bcopy(dt->d_name, dp->d_name, dt->d_namlen + 1);
    750 			dp->d_type = dt->d_type;
    751 
    752 			if (error = uiomove((caddr_t)dp, UIO_MX, uio))
    753 				break;
    754 		}
    755 
    756 	    	break;
    757 
    758 	    }
    759 
    760 	/*
    761 	 * this is for the root of the procfs filesystem
    762 	 * what is needed is a special entry for "curproc"
    763 	 * followed by an entry for each process on allproc
    764 #ifdef PROCFS_ZOMBIE
    765 	 * and zombproc.
    766 #endif
    767 	 */
    768 
    769 	case Proot: {
    770 #ifdef PROCFS_ZOMBIE
    771 		int doingzomb = 0;
    772 #endif
    773 		int pcnt = 0;
    774 		volatile struct proc *p = allproc.lh_first;
    775 
    776 	again:
    777 		for (; p && uio->uio_resid >= UIO_MX; i++, pcnt++) {
    778 			bzero((char *) dp, UIO_MX);
    779 			dp->d_reclen = UIO_MX;
    780 
    781 			switch (i) {
    782 			case 0:		/* `.' */
    783 			case 1:		/* `..' */
    784 				dp->d_fileno = PROCFS_FILENO(0, Proot);
    785 				dp->d_namlen = i + 1;
    786 				bcopy("..", dp->d_name, dp->d_namlen);
    787 				dp->d_name[i + 1] = '\0';
    788 				dp->d_type = DT_DIR;
    789 				break;
    790 
    791 			case 2:
    792 				dp->d_fileno = PROCFS_FILENO(0, Pcurproc);
    793 				dp->d_namlen = 7;
    794 				bcopy("curproc", dp->d_name, 8);
    795 				dp->d_type = DT_LNK;
    796 				break;
    797 
    798 			default:
    799 				while (pcnt < i) {
    800 					pcnt++;
    801 					p = p->p_list.le_next;
    802 					if (!p)
    803 						goto done;
    804 				}
    805 				dp->d_fileno = PROCFS_FILENO(p->p_pid, Pproc);
    806 				dp->d_namlen = sprintf(dp->d_name, "%ld",
    807 				    (long)p->p_pid);
    808 				dp->d_type = DT_REG;
    809 				p = p->p_list.le_next;
    810 				break;
    811 			}
    812 
    813 			if (error = uiomove((caddr_t)dp, UIO_MX, uio))
    814 				break;
    815 		}
    816 	done:
    817 
    818 #ifdef PROCFS_ZOMBIE
    819 		if (p == 0 && doingzomb == 0) {
    820 			doingzomb = 1;
    821 			p = zombproc.lh_first;
    822 			goto again;
    823 		}
    824 #endif
    825 
    826 		break;
    827 
    828 	    }
    829 
    830 	default:
    831 		error = ENOTDIR;
    832 		break;
    833 	}
    834 
    835 	uio->uio_offset = i * UIO_MX;
    836 
    837 	return (error);
    838 }
    839 
    840 /*
    841  * readlink reads the link of `curproc'
    842  */
    843 procfs_readlink(ap)
    844 	struct vop_readlink_args *ap;
    845 {
    846 	struct uio *uio = ap->a_uio;
    847 	char buf[16];		/* should be enough */
    848 	int len;
    849 
    850 	if (VTOPFS(ap->a_vp)->pfs_fileno != PROCFS_FILENO(0, Pcurproc))
    851 		return (EINVAL);
    852 
    853 	len = sprintf(buf, "%ld", (long)curproc->p_pid);
    854 
    855 	return (uiomove((caddr_t)buf, len, ap->a_uio));
    856 }
    857 
    858 /*
    859  * convert decimal ascii to pid_t
    860  */
    861 static pid_t
    862 atopid(b, len)
    863 	const char *b;
    864 	u_int len;
    865 {
    866 	pid_t p = 0;
    867 
    868 	while (len--) {
    869 		char c = *b++;
    870 		if (c < '0' || c > '9')
    871 			return (NO_PID);
    872 		p = 10 * p + (c - '0');
    873 		if (p > PID_MAX)
    874 			return (NO_PID);
    875 	}
    876 
    877 	return (p);
    878 }
    879 
    880 /*
    881  * procfs vnode operations.
    882  */
    883 int (**procfs_vnodeop_p)();
    884 struct vnodeopv_entry_desc procfs_vnodeop_entries[] = {
    885 	{ &vop_default_desc, vn_default_error },
    886 	{ &vop_lookup_desc, procfs_lookup },		/* lookup */
    887 	{ &vop_create_desc, procfs_create },		/* create */
    888 	{ &vop_mknod_desc, procfs_mknod },		/* mknod */
    889 	{ &vop_open_desc, procfs_open },		/* open */
    890 	{ &vop_close_desc, procfs_close },		/* close */
    891 	{ &vop_access_desc, procfs_access },		/* access */
    892 	{ &vop_getattr_desc, procfs_getattr },		/* getattr */
    893 	{ &vop_setattr_desc, procfs_setattr },		/* setattr */
    894 	{ &vop_read_desc, procfs_read },		/* read */
    895 	{ &vop_write_desc, procfs_write },		/* write */
    896 	{ &vop_ioctl_desc, procfs_ioctl },		/* ioctl */
    897 	{ &vop_select_desc, procfs_select },		/* select */
    898 	{ &vop_mmap_desc, procfs_mmap },		/* mmap */
    899 	{ &vop_fsync_desc, procfs_fsync },		/* fsync */
    900 	{ &vop_seek_desc, procfs_seek },		/* seek */
    901 	{ &vop_remove_desc, procfs_remove },		/* remove */
    902 	{ &vop_link_desc, procfs_link },		/* link */
    903 	{ &vop_rename_desc, procfs_rename },		/* rename */
    904 	{ &vop_mkdir_desc, procfs_mkdir },		/* mkdir */
    905 	{ &vop_rmdir_desc, procfs_rmdir },		/* rmdir */
    906 	{ &vop_symlink_desc, procfs_symlink },		/* symlink */
    907 	{ &vop_readdir_desc, procfs_readdir },		/* readdir */
    908 	{ &vop_readlink_desc, procfs_readlink },	/* readlink */
    909 	{ &vop_abortop_desc, procfs_abortop },		/* abortop */
    910 	{ &vop_inactive_desc, procfs_inactive },	/* inactive */
    911 	{ &vop_reclaim_desc, procfs_reclaim },		/* reclaim */
    912 	{ &vop_lock_desc, procfs_lock },		/* lock */
    913 	{ &vop_unlock_desc, procfs_unlock },		/* unlock */
    914 	{ &vop_bmap_desc, procfs_bmap },		/* bmap */
    915 	{ &vop_strategy_desc, procfs_strategy },	/* strategy */
    916 	{ &vop_print_desc, procfs_print },		/* print */
    917 	{ &vop_islocked_desc, procfs_islocked },	/* islocked */
    918 	{ &vop_pathconf_desc, procfs_pathconf },	/* pathconf */
    919 	{ &vop_advlock_desc, procfs_advlock },		/* advlock */
    920 	{ &vop_blkatoff_desc, procfs_blkatoff },	/* blkatoff */
    921 	{ &vop_valloc_desc, procfs_valloc },		/* valloc */
    922 	{ &vop_vfree_desc, procfs_vfree },		/* vfree */
    923 	{ &vop_truncate_desc, procfs_truncate },	/* truncate */
    924 	{ &vop_update_desc, procfs_update },		/* update */
    925 	{ (struct vnodeop_desc*)NULL, (int(*)())NULL }
    926 };
    927 struct vnodeopv_desc procfs_vnodeop_opv_desc =
    928 	{ &procfs_vnodeop_p, procfs_vnodeop_entries };
    929