Home | History | Annotate | Line # | Download | only in procfs
procfs_vnops.c revision 1.1.1.2
      1 /*
      2  * Copyright (c) 1993, 1995 Jan-Simon Pendry
      3  * Copyright (c) 1993, 1995
      4  *	The Regents of the University of California.  All rights reserved.
      5  *
      6  * This code is derived from software contributed to Berkeley by
      7  * Jan-Simon Pendry.
      8  *
      9  * Redistribution and use in source and binary forms, with or without
     10  * modification, are permitted provided that the following conditions
     11  * are met:
     12  * 1. Redistributions of source code must retain the above copyright
     13  *    notice, this list of conditions and the following disclaimer.
     14  * 2. Redistributions in binary form must reproduce the above copyright
     15  *    notice, this list of conditions and the following disclaimer in the
     16  *    documentation and/or other materials provided with the distribution.
     17  * 3. All advertising materials mentioning features or use of this software
     18  *    must display the following acknowledgement:
     19  *	This product includes software developed by the University of
     20  *	California, Berkeley and its contributors.
     21  * 4. Neither the name of the University nor the names of its contributors
     22  *    may be used to endorse or promote products derived from this software
     23  *    without specific prior written permission.
     24  *
     25  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     26  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     27  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     28  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     29  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     30  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     31  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     32  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     33  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     34  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     35  * SUCH DAMAGE.
     36  *
     37  *	@(#)procfs_vnops.c	8.18 (Berkeley) 5/21/95
     38  *
     39  * From:
     40  *	$Id: procfs_vnops.c,v 1.1.1.2 1998/03/01 02:13:19 fvdl Exp $
     41  */
     42 
     43 /*
     44  * procfs vnode interface
     45  */
     46 
     47 #include <sys/param.h>
     48 #include <sys/systm.h>
     49 #include <sys/time.h>
     50 #include <sys/kernel.h>
     51 #include <sys/file.h>
     52 #include <sys/proc.h>
     53 #include <sys/vnode.h>
     54 #include <sys/namei.h>
     55 #include <sys/malloc.h>
     56 #include <sys/dirent.h>
     57 #include <sys/resourcevar.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 		int 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 		int *a_runp;
    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 	if (ap->a_runp != NULL)
    209 		*ap->a_runp = 0;
    210 	return (0);
    211 }
    212 
    213 /*
    214  * procfs_inactive is called when the pfsnode
    215  * is vrele'd and the reference count goes
    216  * to zero.  (vp) will be on the vnode free
    217  * list, so to get it back vget() must be
    218  * used.
    219  *
    220  * for procfs, check if the process is still
    221  * alive and if it isn't then just throw away
    222  * the vnode by calling vgone().  this may
    223  * be overkill and a waste of time since the
    224  * chances are that the process will still be
    225  * there and PFIND is not free.
    226  *
    227  * (vp) is locked on entry, but must be unlocked on exit.
    228  */
    229 procfs_inactive(ap)
    230 	struct vop_inactive_args /* {
    231 		struct vnode *a_vp;
    232 	} */ *ap;
    233 {
    234 	struct vnode *vp = ap->a_vp;
    235 	struct pfsnode *pfs = VTOPFS(vp);
    236 
    237 	VOP_UNLOCK(vp, 0, ap->a_p);
    238 	if (PFIND(pfs->pfs_pid) == 0)
    239 		vgone(vp);
    240 
    241 	return (0);
    242 }
    243 
    244 /*
    245  * _reclaim is called when getnewvnode()
    246  * wants to make use of an entry on the vnode
    247  * free list.  at this time the filesystem needs
    248  * to free any private data and remove the node
    249  * from any private lists.
    250  */
    251 procfs_reclaim(ap)
    252 	struct vop_reclaim_args /* {
    253 		struct vnode *a_vp;
    254 	} */ *ap;
    255 {
    256 
    257 	return (procfs_freevp(ap->a_vp));
    258 }
    259 
    260 /*
    261  * Return POSIX pathconf information applicable to special devices.
    262  */
    263 procfs_pathconf(ap)
    264 	struct vop_pathconf_args /* {
    265 		struct vnode *a_vp;
    266 		int a_name;
    267 		int *a_retval;
    268 	} */ *ap;
    269 {
    270 
    271 	switch (ap->a_name) {
    272 	case _PC_LINK_MAX:
    273 		*ap->a_retval = LINK_MAX;
    274 		return (0);
    275 	case _PC_MAX_CANON:
    276 		*ap->a_retval = MAX_CANON;
    277 		return (0);
    278 	case _PC_MAX_INPUT:
    279 		*ap->a_retval = MAX_INPUT;
    280 		return (0);
    281 	case _PC_PIPE_BUF:
    282 		*ap->a_retval = PIPE_BUF;
    283 		return (0);
    284 	case _PC_CHOWN_RESTRICTED:
    285 		*ap->a_retval = 1;
    286 		return (0);
    287 	case _PC_VDISABLE:
    288 		*ap->a_retval = _POSIX_VDISABLE;
    289 		return (0);
    290 	default:
    291 		return (EINVAL);
    292 	}
    293 	/* NOTREACHED */
    294 }
    295 
    296 /*
    297  * _print is used for debugging.
    298  * just print a readable description
    299  * of (vp).
    300  */
    301 procfs_print(ap)
    302 	struct vop_print_args /* {
    303 		struct vnode *a_vp;
    304 	} */ *ap;
    305 {
    306 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
    307 
    308 	printf("tag VT_PROCFS, type %s, pid %d, mode %x, flags %x\n",
    309 	    pfs->pfs_type, pfs->pfs_pid, pfs->pfs_mode, pfs->pfs_flags);
    310 }
    311 
    312 /*
    313  * _abortop is called when operations such as
    314  * rename and create fail.  this entry is responsible
    315  * for undoing any side-effects caused by the lookup.
    316  * this will always include freeing the pathname buffer.
    317  */
    318 procfs_abortop(ap)
    319 	struct vop_abortop_args /* {
    320 		struct vnode *a_dvp;
    321 		struct componentname *a_cnp;
    322 	} */ *ap;
    323 {
    324 
    325 	if ((ap->a_cnp->cn_flags & (HASBUF | SAVESTART)) == HASBUF)
    326 		FREE(ap->a_cnp->cn_pnbuf, M_NAMEI);
    327 	return (0);
    328 }
    329 
    330 /*
    331  * generic entry point for unsupported operations
    332  */
    333 procfs_badop()
    334 {
    335 
    336 	return (EIO);
    337 }
    338 
    339 /*
    340  * Invent attributes for pfsnode (vp) and store
    341  * them in (vap).
    342  * Directories lengths are returned as zero since
    343  * any real length would require the genuine size
    344  * to be computed, and nothing cares anyway.
    345  *
    346  * this is relatively minimal for procfs.
    347  */
    348 procfs_getattr(ap)
    349 	struct vop_getattr_args /* {
    350 		struct vnode *a_vp;
    351 		struct vattr *a_vap;
    352 		struct ucred *a_cred;
    353 		struct proc *a_p;
    354 	} */ *ap;
    355 {
    356 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
    357 	struct vattr *vap = ap->a_vap;
    358 	struct proc *procp;
    359 	struct timeval tv;
    360 	int error;
    361 
    362 	/* first check the process still exists */
    363 	switch (pfs->pfs_type) {
    364 	case Proot:
    365 	case Pcurproc:
    366 		procp = 0;
    367 		break;
    368 
    369 	default:
    370 		procp = PFIND(pfs->pfs_pid);
    371 		if (procp == 0)
    372 			return (ENOENT);
    373 	}
    374 
    375 	error = 0;
    376 
    377 	/* start by zeroing out the attributes */
    378 	VATTR_NULL(vap);
    379 
    380 	/* next do all the common fields */
    381 	vap->va_type = ap->a_vp->v_type;
    382 	vap->va_mode = pfs->pfs_mode;
    383 	vap->va_fileid = pfs->pfs_fileno;
    384 	vap->va_flags = 0;
    385 	vap->va_blocksize = PAGE_SIZE;
    386 	vap->va_bytes = vap->va_size = 0;
    387 
    388 	/*
    389 	 * Make all times be current TOD.
    390 	 * It would be possible to get the process start
    391 	 * time from the p_stat structure, but there's
    392 	 * no "file creation" time stamp anyway, and the
    393 	 * p_stat structure is not addressible if u. gets
    394 	 * swapped out for that process.
    395 	 */
    396 	microtime(&tv);
    397 	TIMEVAL_TO_TIMESPEC(&tv, &vap->va_ctime);
    398 	vap->va_atime = vap->va_mtime = vap->va_ctime;
    399 
    400 	/*
    401 	 * If the process has exercised some setuid or setgid
    402 	 * privilege, then rip away read/write permission so
    403 	 * that only root can gain access.
    404 	 */
    405 	switch (pfs->pfs_type) {
    406 	case Pmem:
    407 	case Pregs:
    408 	case Pfpregs:
    409 		if (procp->p_flag & P_SUGID)
    410 			vap->va_mode &= ~((VREAD|VWRITE)|
    411 					  ((VREAD|VWRITE)>>3)|
    412 					  ((VREAD|VWRITE)>>6));
    413 	case Pctl:
    414 	case Pstatus:
    415 	case Pnote:
    416 	case Pnotepg:
    417 		vap->va_nlink = 1;
    418 		vap->va_uid = procp->p_ucred->cr_uid;
    419 		vap->va_gid = procp->p_ucred->cr_gid;
    420 		break;
    421 	}
    422 
    423 	/*
    424 	 * now do the object specific fields
    425 	 *
    426 	 * The size could be set from struct reg, but it's hardly
    427 	 * worth the trouble, and it puts some (potentially) machine
    428 	 * dependent data into this machine-independent code.  If it
    429 	 * becomes important then this function should break out into
    430 	 * a per-file stat function in the corresponding .c file.
    431 	 */
    432 
    433 	switch (pfs->pfs_type) {
    434 	case Proot:
    435 		/*
    436 		 * Set nlink to 1 to tell fts(3) we don't actually know.
    437 		 */
    438 		vap->va_nlink = 1;
    439 		vap->va_uid = 0;
    440 		vap->va_gid = 0;
    441 		vap->va_size = vap->va_bytes = DEV_BSIZE;
    442 		break;
    443 
    444 	case Pcurproc: {
    445 		char buf[16];		/* should be enough */
    446 		vap->va_nlink = 1;
    447 		vap->va_uid = 0;
    448 		vap->va_gid = 0;
    449 		vap->va_size = vap->va_bytes =
    450 		    sprintf(buf, "%ld", (long)curproc->p_pid);
    451 		break;
    452 	}
    453 
    454 	case Pproc:
    455 		vap->va_nlink = 2;
    456 		vap->va_uid = procp->p_ucred->cr_uid;
    457 		vap->va_gid = procp->p_ucred->cr_gid;
    458 		vap->va_size = vap->va_bytes = DEV_BSIZE;
    459 		break;
    460 
    461 	case Pfile:
    462 		error = EOPNOTSUPP;
    463 		break;
    464 
    465 	case Pmem:
    466 		vap->va_bytes = vap->va_size =
    467 			ctob(procp->p_vmspace->vm_tsize +
    468 				    procp->p_vmspace->vm_dsize +
    469 				    procp->p_vmspace->vm_ssize);
    470 		break;
    471 
    472 	case Pregs:
    473 		vap->va_bytes = vap->va_size = sizeof(struct reg);
    474 		break;
    475 
    476 	case Pfpregs:
    477 		vap->va_bytes = vap->va_size = sizeof(struct fpreg);
    478 		break;
    479 
    480 	case Pctl:
    481 	case Pstatus:
    482 	case Pnote:
    483 	case Pnotepg:
    484 		break;
    485 
    486 	default:
    487 		panic("procfs_getattr");
    488 	}
    489 
    490 	return (error);
    491 }
    492 
    493 procfs_setattr(ap)
    494 	struct vop_setattr_args /* {
    495 		struct vnode *a_vp;
    496 		struct vattr *a_vap;
    497 		struct ucred *a_cred;
    498 		struct proc *a_p;
    499 	} */ *ap;
    500 {
    501 	/*
    502 	 * just fake out attribute setting
    503 	 * it's not good to generate an error
    504 	 * return, otherwise things like creat()
    505 	 * will fail when they try to set the
    506 	 * file length to 0.  worse, this means
    507 	 * that echo $note > /proc/$pid/note will fail.
    508 	 */
    509 
    510 	return (0);
    511 }
    512 
    513 /*
    514  * implement access checking.
    515  *
    516  * something very similar to this code is duplicated
    517  * throughout the 4bsd kernel and should be moved
    518  * into kern/vfs_subr.c sometime.
    519  *
    520  * actually, the check for super-user is slightly
    521  * broken since it will allow read access to write-only
    522  * objects.  this doesn't cause any particular trouble
    523  * but does mean that the i/o entry points need to check
    524  * that the operation really does make sense.
    525  */
    526 procfs_access(ap)
    527 	struct vop_access_args /* {
    528 		struct vnode *a_vp;
    529 		int a_mode;
    530 		struct ucred *a_cred;
    531 		struct proc *a_p;
    532 	} */ *ap;
    533 {
    534 	struct vattr *vap;
    535 	struct vattr vattr;
    536 	int error;
    537 
    538 	/*
    539 	 * If you're the super-user,
    540 	 * you always get access.
    541 	 */
    542 	if (ap->a_cred->cr_uid == 0)
    543 		return (0);
    544 
    545 	vap = &vattr;
    546 	if (error = VOP_GETATTR(ap->a_vp, vap, ap->a_cred, ap->a_p))
    547 		return (error);
    548 
    549 	/*
    550 	 * Access check is based on only one of owner, group, public.
    551 	 * If not owner, then check group. If not a member of the
    552 	 * group, then check public access.
    553 	 */
    554 	if (ap->a_cred->cr_uid != vap->va_uid) {
    555 		gid_t *gp;
    556 		int i;
    557 
    558 		ap->a_mode >>= 3;
    559 		gp = ap->a_cred->cr_groups;
    560 		for (i = 0; i < ap->a_cred->cr_ngroups; i++, gp++)
    561 			if (vap->va_gid == *gp)
    562 				goto found;
    563 		ap->a_mode >>= 3;
    564 found:
    565 		;
    566 	}
    567 
    568 	if ((vap->va_mode & ap->a_mode) == ap->a_mode)
    569 		return (0);
    570 
    571 	return (EACCES);
    572 }
    573 
    574 /*
    575  * lookup.  this is incredibly complicated in the
    576  * general case, however for most pseudo-filesystems
    577  * very little needs to be done.
    578  *
    579  * unless you want to get a migraine, just make sure your
    580  * filesystem doesn't do any locking of its own.  otherwise
    581  * read and inwardly digest ufs_lookup().
    582  */
    583 procfs_lookup(ap)
    584 	struct vop_lookup_args /* {
    585 		struct vnode * a_dvp;
    586 		struct vnode ** a_vpp;
    587 		struct componentname * a_cnp;
    588 	} */ *ap;
    589 {
    590 	struct componentname *cnp = ap->a_cnp;
    591 	struct vnode **vpp = ap->a_vpp;
    592 	struct vnode *dvp = ap->a_dvp;
    593 	char *pname = cnp->cn_nameptr;
    594 	struct proc *curp = cnp->cn_proc;
    595 	int error = 0;
    596 	struct proc_target *pt;
    597 	struct vnode *fvp;
    598 	pid_t pid;
    599 	struct pfsnode *pfs;
    600 	struct proc *p;
    601 	int i;
    602 
    603 	*vpp = NULL;
    604 
    605 	if (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)
    606 		return (EROFS);
    607 
    608 	if (cnp->cn_namelen == 1 && *pname == '.') {
    609 		*vpp = dvp;
    610 		VREF(dvp);
    611 		/* vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY, curp); */
    612 		return (0);
    613 	}
    614 
    615 	pfs = VTOPFS(dvp);
    616 	switch (pfs->pfs_type) {
    617 	case Proot:
    618 		if (cnp->cn_flags & ISDOTDOT)
    619 			return (EIO);
    620 
    621 		if (CNEQ(cnp, "curproc", 7))
    622 			return (procfs_allocvp(dvp->v_mount, vpp, 0, Pcurproc));
    623 
    624 		pid = atopid(pname, cnp->cn_namelen);
    625 		if (pid == NO_PID)
    626 			break;
    627 
    628 		p = PFIND(pid);
    629 		if (p == 0)
    630 			break;
    631 
    632 		return (procfs_allocvp(dvp->v_mount, vpp, pid, Pproc));
    633 
    634 	case Pproc:
    635 		if (cnp->cn_flags & ISDOTDOT)
    636 			return (procfs_root(dvp->v_mount, vpp));
    637 
    638 		p = PFIND(pfs->pfs_pid);
    639 		if (p == 0)
    640 			break;
    641 
    642 		for (pt = proc_targets, i = 0; i < nproc_targets; pt++, i++) {
    643 			if (cnp->cn_namelen == pt->pt_namlen &&
    644 			    bcmp(pt->pt_name, pname, cnp->cn_namelen) == 0 &&
    645 			    (pt->pt_valid == NULL || (*pt->pt_valid)(p)))
    646 				goto found;
    647 		}
    648 		break;
    649 
    650 	found:
    651 		if (pt->pt_pfstype == Pfile) {
    652 			fvp = procfs_findtextvp(p);
    653 			/* We already checked that it exists. */
    654 			VREF(fvp);
    655 			vn_lock(fvp, LK_EXCLUSIVE | LK_RETRY, curp);
    656 			*vpp = fvp;
    657 			return (0);
    658 		}
    659 
    660 		return (procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
    661 		    pt->pt_pfstype));
    662 
    663 	default:
    664 		return (ENOTDIR);
    665 	}
    666 
    667 	return (cnp->cn_nameiop == LOOKUP ? ENOENT : EROFS);
    668 }
    669 
    670 int
    671 procfs_validfile(p)
    672 	struct proc *p;
    673 {
    674 
    675 	return (procfs_findtextvp(p) != NULLVP);
    676 }
    677 
    678 /*
    679  * readdir returns directory entries from pfsnode (vp).
    680  *
    681  * the strategy here with procfs is to generate a single
    682  * directory entry at a time (struct pfsdent) and then
    683  * copy that out to userland using uiomove.  a more efficent
    684  * though more complex implementation, would try to minimize
    685  * the number of calls to uiomove().  for procfs, this is
    686  * hardly worth the added code complexity.
    687  *
    688  * this should just be done through read()
    689  */
    690 procfs_readdir(ap)
    691 	struct vop_readdir_args /* {
    692 		struct vnode *a_vp;
    693 		struct uio *a_uio;
    694 		struct ucred *a_cred;
    695 		int *a_eofflag;
    696 		u_long *a_cookies;
    697 		int a_ncookies;
    698 	} */ *ap;
    699 {
    700 	struct uio *uio = ap->a_uio;
    701 	struct pfsdent d;
    702 	struct pfsdent *dp = &d;
    703 	struct pfsnode *pfs;
    704 	int error;
    705 	int count;
    706 	int i;
    707 
    708 	/*
    709 	 * We don't allow exporting procfs mounts, and currently local
    710 	 * requests do not need cookies.
    711 	 */
    712 	if (ap->a_ncookies)
    713 		panic("procfs_readdir: not hungry");
    714 
    715 	pfs = VTOPFS(ap->a_vp);
    716 
    717 	if (uio->uio_resid < UIO_MX)
    718 		return (EINVAL);
    719 	if (uio->uio_offset & (UIO_MX-1))
    720 		return (EINVAL);
    721 	if (uio->uio_offset < 0)
    722 		return (EINVAL);
    723 
    724 	error = 0;
    725 	count = 0;
    726 	i = uio->uio_offset / UIO_MX;
    727 
    728 	switch (pfs->pfs_type) {
    729 	/*
    730 	 * this is for the process-specific sub-directories.
    731 	 * all that is needed to is copy out all the entries
    732 	 * from the procent[] table (top of this file).
    733 	 */
    734 	case Pproc: {
    735 		struct proc *p;
    736 		struct proc_target *pt;
    737 
    738 		p = PFIND(pfs->pfs_pid);
    739 		if (p == NULL)
    740 			break;
    741 
    742 		for (pt = &proc_targets[i];
    743 		     uio->uio_resid >= UIO_MX && i < nproc_targets; pt++, i++) {
    744 			if (pt->pt_valid && (*pt->pt_valid)(p) == 0)
    745 				continue;
    746 
    747 			dp->d_reclen = UIO_MX;
    748 			dp->d_fileno = PROCFS_FILENO(pfs->pfs_pid, pt->pt_pfstype);
    749 			dp->d_namlen = pt->pt_namlen;
    750 			bcopy(pt->pt_name, dp->d_name, pt->pt_namlen + 1);
    751 			dp->d_type = pt->pt_type;
    752 
    753 			if (error = uiomove((caddr_t)dp, UIO_MX, uio))
    754 				break;
    755 		}
    756 
    757 	    	break;
    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_revoke_desc, procfs_revoke },		/* revoke */
    900 	{ &vop_fsync_desc, procfs_fsync },		/* fsync */
    901 	{ &vop_seek_desc, procfs_seek },		/* seek */
    902 	{ &vop_remove_desc, procfs_remove },		/* remove */
    903 	{ &vop_link_desc, procfs_link },		/* link */
    904 	{ &vop_rename_desc, procfs_rename },		/* rename */
    905 	{ &vop_mkdir_desc, procfs_mkdir },		/* mkdir */
    906 	{ &vop_rmdir_desc, procfs_rmdir },		/* rmdir */
    907 	{ &vop_symlink_desc, procfs_symlink },		/* symlink */
    908 	{ &vop_readdir_desc, procfs_readdir },		/* readdir */
    909 	{ &vop_readlink_desc, procfs_readlink },	/* readlink */
    910 	{ &vop_abortop_desc, procfs_abortop },		/* abortop */
    911 	{ &vop_inactive_desc, procfs_inactive },	/* inactive */
    912 	{ &vop_reclaim_desc, procfs_reclaim },		/* reclaim */
    913 	{ &vop_lock_desc, procfs_lock },		/* lock */
    914 	{ &vop_unlock_desc, procfs_unlock },		/* unlock */
    915 	{ &vop_bmap_desc, procfs_bmap },		/* bmap */
    916 	{ &vop_strategy_desc, procfs_strategy },	/* strategy */
    917 	{ &vop_print_desc, procfs_print },		/* print */
    918 	{ &vop_islocked_desc, procfs_islocked },	/* islocked */
    919 	{ &vop_pathconf_desc, procfs_pathconf },	/* pathconf */
    920 	{ &vop_advlock_desc, procfs_advlock },		/* advlock */
    921 	{ &vop_blkatoff_desc, procfs_blkatoff },	/* blkatoff */
    922 	{ &vop_valloc_desc, procfs_valloc },		/* valloc */
    923 	{ &vop_vfree_desc, procfs_vfree },		/* vfree */
    924 	{ &vop_truncate_desc, procfs_truncate },	/* truncate */
    925 	{ &vop_update_desc, procfs_update },		/* update */
    926 	{ (struct vnodeop_desc*)NULL, (int(*)())NULL }
    927 };
    928 struct vnodeopv_desc procfs_vnodeop_opv_desc =
    929 	{ &procfs_vnodeop_p, procfs_vnodeop_entries };
    930