[unix-history] / usr / src / sys / miscfs / procfs / procfs_vnops.c

/*
 * Copyright (c) 1993 Jan-Simon Pendry
 * Copyright (c) 1993
 *	The Regents of the University of California.  All rights reserved.
 *
 * This code is derived from software contributed to Berkeley by
 * Jan-Simon Pendry.
 *
 * %sccs.include.redist.c%
 *
 *	@(#)procfs_vnops.c	8.5 (Berkeley) %G%
 *
 * From:
 *	$Id: procfs_vnops.c,v 3.2 1993/12/15 09:40:17 jsp Exp $
 */

/*
 * procfs vnode interface
 */

#include <sys/param.h>
#include <sys/systm.h>
#include <sys/time.h>
#include <sys/kernel.h>
#include <sys/file.h>
#include <sys/proc.h>
#include <sys/vnode.h>
#include <sys/namei.h>
#include <sys/malloc.h>
#include <sys/dirent.h>
#include <sys/resourcevar.h>
#include <miscfs/procfs/procfs.h>
#include <vm/vm.h>	/* for PAGE_SIZE */

/*
 * Vnode Operations.
 *
 */

/*
 * This is a list of the valid names in the
 * process-specific sub-directories.  It is
 * used in procfs_lookup and procfs_readdir
 */
static struct pfsnames {
	u_short	d_namlen;
	char	d_name[PROCFS_NAMELEN];
	pfstype	d_pfstype;
} procent[] = {
#define N(s) sizeof(s)-1, s
	/* namlen, nam, type */
	{  N("file"),	Pfile },
	{  N("mem"),	Pmem },
	{  N("regs"),	Pregs },
	{  N("fpregs"),	Pfpregs },
	{  N("ctl"),	Pctl },
	{  N("status"),	Pstatus },
	{  N("note"),	Pnote },
	{  N("notepg"),	Pnotepg },
#undef N
};
#define Nprocent (sizeof(procent)/sizeof(procent[0]))

static pid_t atopid __P((const char *, u_int));

/*
 * set things up for doing i/o on
 * the pfsnode (vp).  (vp) is locked
 * on entry, and should be left locked
 * on exit.
 *
 * for procfs we don't need to do anything
 * in particular for i/o.  all that is done
 * is to support exclusive open on process
 * memory images.
 */
procfs_open(ap)
	struct vop_open_args *ap;
{
	struct pfsnode *pfs = VTOPFS(ap->a_vp);

	switch (pfs->pfs_type) {
	case Pmem:
		if (PFIND(pfs->pfs_pid) == 0)
			return (ENOENT);	/* was ESRCH, jsp */

		if ((pfs->pfs_flags & FWRITE) && (ap->a_mode & O_EXCL) ||
				(pfs->pfs_flags & O_EXCL) && (ap->a_mode & FWRITE))
			return (EBUSY);


		if (ap->a_mode & FWRITE)
			pfs->pfs_flags = ap->a_mode & (FWRITE|O_EXCL);

		return (0);

	default:
		break;
	}

	return (0);
}

/*
 * close the pfsnode (vp) after doing i/o.
 * (vp) is not locked on entry or exit.
 *
 * nothing to do for procfs other than undo
 * any exclusive open flag (see _open above).
 */
procfs_close(ap)
	struct vop_close_args *ap;
{
	struct pfsnode *pfs = VTOPFS(ap->a_vp);

	switch (pfs->pfs_type) {
	case Pmem:
		if ((ap->a_fflag & FWRITE) && (pfs->pfs_flags & O_EXCL))
			pfs->pfs_flags &= ~(FWRITE|O_EXCL);
		break;
	}

	return (0);
}

/*
 * do an ioctl operation on pfsnode (vp).
 * (vp) is not locked on entry or exit.
 */
procfs_ioctl(ap)
	struct vop_ioctl_args *ap;
{

	return (ENOTTY);
}

/*
 * do block mapping for pfsnode (vp).
 * since we don't use the buffer cache
 * for procfs this function should never
 * be called.  in any case, it's not clear
 * what part of the kernel ever makes use
 * of this function.  for sanity, this is the
 * usual no-op bmap, although returning
 * (EIO) would be a reasonable alternative.
 */
procfs_bmap(ap)
	struct vop_bmap_args *ap;
{

	if (ap->a_vpp != NULL)
		*ap->a_vpp = ap->a_vp;
	if (ap->a_bnp != NULL)
		*ap->a_bnp = ap->a_bn;
	return (0);
}

/*
 * _inactive is called when the pfsnode
 * is vrele'd and the reference count goes
 * to zero.  (vp) will be on the vnode free
 * list, so to get it back vget() must be
 * used.
 *
 * for procfs, check if the process is still
 * alive and if it isn't then just throw away
 * the vnode by calling vgone().  this may
 * be overkill and a waste of time since the
 * chances are that the process will still be
 * there and PFIND is not free.
 *
 * (vp) is not locked on entry or exit.
 */
procfs_inactive(ap)
	struct vop_inactive_args *ap;
{
	struct pfsnode *pfs = VTOPFS(ap->a_vp);

	if (PFIND(pfs->pfs_pid) == 0)
		vgone(ap->a_vp);

	return (0);
}

/*
 * _reclaim is called when getnewvnode()
 * wants to make use of an entry on the vnode
 * free list.  at this time the filesystem needs
 * to free any private data and remove the node
 * from any private lists.
 */
procfs_reclaim(ap)
	struct vop_reclaim_args *ap;
{
	int error;

	error = procfs_freevp(ap->a_vp);
	return (error);
}

/*
 * Return POSIX pathconf information applicable to special devices.
 */
procfs_pathconf(ap)
	struct vop_pathconf_args /* {
		struct vnode *a_vp;
		int a_name;
		int *a_retval;
	} */ *ap;
{

	switch (ap->a_name) {
	case _PC_LINK_MAX:
		*ap->a_retval = LINK_MAX;
		return (0);
	case _PC_MAX_CANON:
		*ap->a_retval = MAX_CANON;
		return (0);
	case _PC_MAX_INPUT:
		*ap->a_retval = MAX_INPUT;
		return (0);
	case _PC_PIPE_BUF:
		*ap->a_retval = PIPE_BUF;
		return (0);
	case _PC_CHOWN_RESTRICTED:
		*ap->a_retval = 1;
		return (0);
	case _PC_VDISABLE:
		*ap->a_retval = _POSIX_VDISABLE;
		return (0);
	default:
		return (EINVAL);
	}
	/* NOTREACHED */
}

/*
 * _print is used for debugging.
 * just print a readable description
 * of (vp).
 */
procfs_print(ap)
	struct vop_print_args *ap;
{
	struct pfsnode *pfs = VTOPFS(ap->a_vp);

	printf("tag VT_PROCFS, pid %d, mode %x, flags %x\n",
		pfs->pfs_pid,
		pfs->pfs_mode, pfs->pfs_flags);
}

/*
 * _abortop is called when operations such as
 * rename and create fail.  this entry is responsible
 * for undoing any side-effects caused by the lookup.
 * this will always include freeing the pathname buffer.
 */
procfs_abortop(ap)
	struct vop_abortop_args *ap;
{

	if ((ap->a_cnp->cn_flags & (HASBUF | SAVESTART)) == HASBUF)
		FREE(ap->a_cnp->cn_pnbuf, M_NAMEI);
	return (0);
}

/*
 * generic entry point for unsupported operations
 */
procfs_badop()
{

	return (EIO);
}

/*
 * Invent attributes for pfsnode (vp) and store
 * them in (vap).
 * Directories lengths are returned as zero since
 * any real length would require the genuine size
 * to be computed, and nothing cares anyway.
 *
 * this is relatively minimal for procfs.
 */
procfs_getattr(ap)
	struct vop_getattr_args *ap;
{
	struct pfsnode *pfs = VTOPFS(ap->a_vp);
	struct vattr *vap = ap->a_vap;
	struct proc *procp;
	int error;

	/* first check the process still exists */
	procp = PFIND(pfs->pfs_pid);
	if (procp == 0)
		return (ENOENT);

	error = 0;

	/* start by zeroing out the attributes */
	VATTR_NULL(vap);

	/* next do all the common fields */
	vap->va_type = ap->a_vp->v_type;
	vap->va_mode = pfs->pfs_mode;
	vap->va_fileid = pfs->pfs_fileno;
	vap->va_flags = 0;
	vap->va_blocksize = PAGE_SIZE;
	vap->va_bytes = vap->va_size = 0;

	/*
	 * Make all times be current TOD.
	 * It would be possible to get the process start
	 * time from the p_stat structure, but there's
	 * no "file creation" time stamp anyway, and the
	 * p_stat structure is not addressible if u. gets
	 * swapped out for that process.
	 */
	microtime(&vap->va_ctime);
	vap->va_atime = vap->va_mtime = vap->va_ctime;

	/*
	 * now do the object specific fields
	 *
	 * The size could be set from struct reg, but it's hardly
	 * worth the trouble, and it puts some (potentially) machine
	 * dependent data into this machine-independent code.  If it
	 * becomes important then this function should break out into
	 * a per-file stat function in the corresponding .c file.
	 */

	switch (pfs->pfs_type) {
	case Proot:
		vap->va_nlink = 2;
		vap->va_uid = 0;
		vap->va_gid = 0;
		break;

	case Pproc:
		vap->va_nlink = 2;
		vap->va_uid = procp->p_ucred->cr_uid;
		vap->va_gid = procp->p_ucred->cr_gid;
		break;

	case Pfile:
		error = EOPNOTSUPP;
		break;

	case Pmem:
		vap->va_nlink = 1;
		vap->va_bytes = vap->va_size =
			ctob(procp->p_vmspace->vm_tsize +
				    procp->p_vmspace->vm_dsize +
				    procp->p_vmspace->vm_ssize);
		vap->va_uid = procp->p_ucred->cr_uid;
		vap->va_gid = procp->p_ucred->cr_gid;
		break;

	case Pregs:
	case Pfpregs:
	case Pctl:
	case Pstatus:
	case Pnote:
	case Pnotepg:
		vap->va_nlink = 1;
		vap->va_uid = procp->p_ucred->cr_uid;
		vap->va_gid = procp->p_ucred->cr_gid;
		break;

	default:
		panic("procfs_getattr");
	}

	return (error);
}

procfs_setattr(ap)
	struct vop_setattr_args *ap;
{
	/*
	 * just fake out attribute setting
	 * it's not good to generate an error
	 * return, otherwise things like creat()
	 * will fail when they try to set the
	 * file length to 0.  worse, this means
	 * that echo $note > /proc/$pid/note will fail.
	 */

	return (0);
}

/*
 * implement access checking.
 *
 * something very similar to this code is duplicated
 * throughout the 4bsd kernel and should be moved
 * into kern/vfs_subr.c sometime.
 *
 * actually, the check for super-user is slightly
 * broken since it will allow read access to write-only
 * objects.  this doesn't cause any particular trouble
 * but does mean that the i/o entry points need to check
 * that the operation really does make sense.
 */
procfs_access(ap)
	struct vop_access_args *ap;
{
	struct vattr *vap;
	struct vattr vattr;
	int error;

	/*
	 * If you're the super-user,
	 * you always get access.
	 */
	if (ap->a_cred->cr_uid == (uid_t) 0)
		return (0);
	vap = &vattr;
	if (error = VOP_GETATTR(ap->a_vp, vap, ap->a_cred, ap->a_p))
		return (error);

	/*
	 * Access check is based on only one of owner, group, public.
	 * If not owner, then check group. If not a member of the
	 * group, then check public access.
	 */
	if (ap->a_cred->cr_uid != vap->va_uid) {
		gid_t *gp;
		int i;

		(ap->a_mode) >>= 3;
		gp = ap->a_cred->cr_groups;
		for (i = 0; i < ap->a_cred->cr_ngroups; i++, gp++)
			if (vap->va_gid == *gp)
				goto found;
		ap->a_mode >>= 3;
found:
		;
	}

	if ((vap->va_mode & ap->a_mode) == ap->a_mode)
		return (0);

	return (EACCES);
}

/*
 * lookup.  this is incredibly complicated in the
 * general case, however for most pseudo-filesystems
 * very little needs to be done.
 *
 * unless you want to get a migraine, just make sure your
 * filesystem doesn't do any locking of its own.  otherwise
 * read and inwardly digest ufs_lookup().
 */
procfs_lookup(ap)
	struct vop_lookup_args *ap;
{
	struct componentname *cnp = ap->a_cnp;
	struct vnode **vpp = ap->a_vpp;
	struct vnode *dvp = ap->a_dvp;
	char *pname = cnp->cn_nameptr;
	int error = 0;
	pid_t pid;
	struct vnode *nvp;
	struct pfsnode *pfs;
	struct proc *procp;
	pfstype pfs_type;
	int i;

	if (cnp->cn_namelen == 1 && *pname == '.') {
		*vpp = dvp;
		VREF(dvp);
		/*VOP_LOCK(dvp);*/
		return (0);
	}

	*vpp = NULL;

	pfs = VTOPFS(dvp);
	switch (pfs->pfs_type) {
	case Proot:
		if (cnp->cn_flags & ISDOTDOT)
			return (EIO);

		if (CNEQ(cnp, "curproc", 7))
			pid = cnp->cn_proc->p_pid;
		else
			pid = atopid(pname, cnp->cn_namelen);
		if (pid == NO_PID)
			return (ENOENT);

		procp = PFIND(pid);
		if (procp == 0)
			return (ENOENT);

		error = procfs_allocvp(dvp->v_mount, &nvp, pid, Pproc);
		if (error)
			return (error);

		nvp->v_type = VDIR;
		pfs = VTOPFS(nvp);

		*vpp = nvp;
		return (0);

	case Pproc:
		if (cnp->cn_flags & ISDOTDOT) {
			error = procfs_root(dvp->v_mount, vpp);
			return (error);
		}

		procp = PFIND(pfs->pfs_pid);
		if (procp == 0)
			return (ENOENT);

		for (i = 0; i < Nprocent; i++) {
			struct pfsnames *dp = &procent[i];

			if (cnp->cn_namelen == dp->d_namlen &&
			    bcmp(pname, dp->d_name, dp->d_namlen) == 0) {
			    	pfs_type = dp->d_pfstype;
				goto found;
			}
		}
		return (ENOENT);

	found:
		if (pfs_type == Pfile) {
			nvp = procfs_findtextvp(procp);
			if (nvp) {
				VREF(nvp);
				VOP_LOCK(nvp);
			} else {
				error = ENXIO;
			}
		} else {
			error = procfs_allocvp(dvp->v_mount, &nvp,
					pfs->pfs_pid, pfs_type);
			if (error)
				return (error);

			nvp->v_type = VREG;
			pfs = VTOPFS(nvp);
		}
		*vpp = nvp;
		return (error);

	default:
		return (ENOTDIR);
	}
}

/*
 * readdir returns directory entries from pfsnode (vp).
 *
 * the strategy here with procfs is to generate a single
 * directory entry at a time (struct pfsdent) and then
 * copy that out to userland using uiomove.  a more efficent
 * though more complex implementation, would try to minimize
 * the number of calls to uiomove().  for procfs, this is
 * hardly worth the added code complexity.
 *
 * this should just be done through read()
 */
procfs_readdir(ap)
	struct vop_readdir_args *ap;
{
	struct uio *uio = ap->a_uio;
	struct pfsdent d;
	struct pfsdent *dp = &d;
	struct pfsnode *pfs;
	int error;
	int count;
	int i;

	pfs = VTOPFS(ap->a_vp);

	if (uio->uio_resid < UIO_MX)
		return (EINVAL);
	if (uio->uio_offset & (UIO_MX-1))
		return (EINVAL);
	if (uio->uio_offset < 0)
		return (EINVAL);

	error = 0;
	count = 0;
	i = uio->uio_offset / UIO_MX;

	switch (pfs->pfs_type) {
	/*
	 * this is for the process-specific sub-directories.
	 * all that is needed to is copy out all the entries
	 * from the procent[] table (top of this file).
	 */
	case Pproc: {
		while (uio->uio_resid >= UIO_MX) {
			struct pfsnames *dt;

			if (i >= Nprocent)
				break;

			dt = &procent[i];
			
			dp->d_reclen = UIO_MX;
			dp->d_fileno = PROCFS_FILENO(pfs->pfs_pid, dt->d_pfstype);
			dp->d_type = DT_REG;
			dp->d_namlen = dt->d_namlen;
			bcopy(dt->d_name, dp->d_name, sizeof(dt->d_name)-1);
			error = uiomove((caddr_t) dp, UIO_MX, uio);
			if (error)
				break;
			count += UIO_MX;
			i++;
		}

	    	break;

	    }

	/*
	 * this is for the root of the procfs filesystem
	 * what is needed is a special entry for "curproc"
	 * followed by an entry for each process on allproc
#ifdef PROCFS_ZOMBIE
	 * and zombproc.
#endif
	 */

	case Proot: {
		int pcnt;
#ifdef PROCFS_ZOMBIE
		int doingzomb = 0;
#endif
		volatile struct proc *p;

		p = allproc;

#define PROCFS_XFILES	1	/* number of other entries, like "curproc" */
		pcnt = PROCFS_XFILES;

		while (p && uio->uio_resid >= UIO_MX) {
			bzero((char *) dp, UIO_MX);
			dp->d_type = DT_DIR;
			dp->d_reclen = UIO_MX;

			switch (i) {
			case 0:
				/* ship out entry for "curproc" */
				dp->d_fileno = PROCFS_FILENO(PID_MAX+1, Pproc);
				dp->d_namlen = sprintf(dp->d_name, "curproc");
				break;

			default:
				if (pcnt >= i) {
					dp->d_fileno = PROCFS_FILENO(p->p_pid, Pproc);
					dp->d_namlen = sprintf(dp->d_name, "%ld", (long) p->p_pid);
				}

				p = p->p_next;

#ifdef PROCFS_ZOMBIE
				if (p == 0 && doingzomb == 0) {
					doingzomb = 1;
					p = zombproc;
				}
#endif

				if (pcnt++ < i)
					continue;

				break;
			}
			error = uiomove((caddr_t) dp, UIO_MX, uio);
			if (error)
				break;
			count += UIO_MX;
			i++;
		}

		break;

	    }

	default:
		error = ENOTDIR;
		break;
	}

	uio->uio_offset = i * UIO_MX;

	return (error);
}

/*
 * convert decimal ascii to pid_t
 */
static pid_t
atopid(b, len)
	const char *b;
	u_int len;
{
	pid_t p = 0;

	while (len--) {
		char c = *b++;
		if (c < '0' || c > '9')
			return (NO_PID);
		p = 10 * p + (c - '0');
		if (p > PID_MAX)
			return (NO_PID);
	}

	return (p);
}

/*
 * procfs vnode operations.
 */
int (**procfs_vnodeop_p)();
struct vnodeopv_entry_desc procfs_vnodeop_entries[] = {
	{ &vop_default_desc, vn_default_error },
	{ &vop_lookup_desc, procfs_lookup },		/* lookup */
	{ &vop_create_desc, procfs_create },		/* create */
	{ &vop_mknod_desc, procfs_mknod },		/* mknod */
	{ &vop_open_desc, procfs_open },		/* open */
	{ &vop_close_desc, procfs_close },		/* close */
	{ &vop_access_desc, procfs_access },		/* access */
	{ &vop_getattr_desc, procfs_getattr },		/* getattr */
	{ &vop_setattr_desc, procfs_setattr },		/* setattr */
	{ &vop_read_desc, procfs_read },		/* read */
	{ &vop_write_desc, procfs_write },		/* write */
	{ &vop_ioctl_desc, procfs_ioctl },		/* ioctl */
	{ &vop_select_desc, procfs_select },		/* select */
	{ &vop_mmap_desc, procfs_mmap },		/* mmap */
	{ &vop_fsync_desc, procfs_fsync },		/* fsync */
	{ &vop_seek_desc, procfs_seek },		/* seek */
	{ &vop_remove_desc, procfs_remove },		/* remove */
	{ &vop_link_desc, procfs_link },		/* link */
	{ &vop_rename_desc, procfs_rename },		/* rename */
	{ &vop_mkdir_desc, procfs_mkdir },		/* mkdir */
	{ &vop_rmdir_desc, procfs_rmdir },		/* rmdir */
	{ &vop_symlink_desc, procfs_symlink },		/* symlink */
	{ &vop_readdir_desc, procfs_readdir },		/* readdir */
	{ &vop_readlink_desc, procfs_readlink },	/* readlink */
	{ &vop_abortop_desc, procfs_abortop },		/* abortop */
	{ &vop_inactive_desc, procfs_inactive },	/* inactive */
	{ &vop_reclaim_desc, procfs_reclaim },		/* reclaim */
	{ &vop_lock_desc, procfs_lock },		/* lock */
	{ &vop_unlock_desc, procfs_unlock },		/* unlock */
	{ &vop_bmap_desc, procfs_bmap },		/* bmap */
	{ &vop_strategy_desc, procfs_strategy },	/* strategy */
	{ &vop_print_desc, procfs_print },		/* print */
	{ &vop_islocked_desc, procfs_islocked },	/* islocked */
	{ &vop_pathconf_desc, procfs_pathconf },	/* pathconf */
	{ &vop_advlock_desc, procfs_advlock },		/* advlock */
	{ &vop_blkatoff_desc, procfs_blkatoff },	/* blkatoff */
	{ &vop_valloc_desc, procfs_valloc },		/* valloc */
	{ &vop_vfree_desc, procfs_vfree },		/* vfree */
	{ &vop_truncate_desc, procfs_truncate },	/* truncate */
	{ &vop_update_desc, procfs_update },		/* update */
	{ (struct vnodeop_desc*)NULL, (int(*)())NULL }
};
struct vnodeopv_desc procfs_vnodeop_opv_desc =
	{ &procfs_vnodeop_p, procfs_vnodeop_entries };
Commit	Line	Data
76434cf5	1	/*
76434cf5	2	* Copyright (c) 1993 Jan-Simon Pendry
1611db1e KB	3	* Copyright (c) 1993
1611db1e KB	4	* The Regents of the University of California. All rights reserved.
76434cf5 JSP	5	*
	6	* This code is derived from software contributed to Berkeley by
	7	* Jan-Simon Pendry.
	8	*
	9	* %sccs.include.redist.c%
	10	*
6fdd7e21	11	* @(#)procfs_vnops.c 8.5 (Berkeley) %G%
76434cf5 JSP	12	*
	13	* From:
	14	* $Id: procfs_vnops.c,v 3.2 1993/12/15 09:40:17 jsp Exp $
	15	*/
	16
	17	/*
	18	* procfs vnode interface
	19	*/
	20
	21	#include <sys/param.h>
	22	#include <sys/systm.h>
	23	#include <sys/time.h>
	24	#include <sys/kernel.h>
	25	#include <sys/file.h>
	26	#include <sys/proc.h>
	27	#include <sys/vnode.h>
	28	#include <sys/namei.h>
	29	#include <sys/malloc.h>
	30	#include <sys/dirent.h>
	31	#include <sys/resourcevar.h>
	32	#include <miscfs/procfs/procfs.h>
	33	#include <vm/vm.h> /* for PAGE_SIZE */
	34
	35	/*
	36	* Vnode Operations.
	37	*
	38	*/
	39
	40	/*
	41	* This is a list of the valid names in the
	42	* process-specific sub-directories. It is
	43	* used in procfs_lookup and procfs_readdir
	44	*/
	45	static struct pfsnames {
	46	u_short d_namlen;
	47	char d_name[PROCFS_NAMELEN];
	48	pfstype d_pfstype;
	49	} procent[] = {
	50	#define N(s) sizeof(s)-1, s
	51	/* namlen, nam, type */
6fdd7e21 JSP	52	{ N("file"), Pfile },
	53	{ N("mem"), Pmem },
	54	{ N("regs"), Pregs },
	55	{ N("fpregs"), Pfpregs },
	56	{ N("ctl"), Pctl },
	57	{ N("status"), Pstatus },
	58	{ N("note"), Pnote },
	59	{ N("notepg"), Pnotepg },
76434cf5 JSP	60	#undef N
	61	};
	62	#define Nprocent (sizeof(procent)/sizeof(procent[0]))
	63
	64	static pid_t atopid __P((const char *, u_int));
	65
	66	/*
	67	* set things up for doing i/o on
	68	* the pfsnode (vp). (vp) is locked
	69	* on entry, and should be left locked
	70	* on exit.
	71	*
	72	* for procfs we don't need to do anything
	73	* in particular for i/o. all that is done
	74	* is to support exclusive open on process
	75	* memory images.
	76	*/
	77	procfs_open(ap)
	78	struct vop_open_args *ap;
	79	{
	80	struct pfsnode *pfs = VTOPFS(ap->a_vp);
	81
	82	switch (pfs->pfs_type) {
	83	case Pmem:
	84	if (PFIND(pfs->pfs_pid) == 0)
	85	return (ENOENT); /* was ESRCH, jsp */
	86
	87	if ((pfs->pfs_flags & FWRITE) && (ap->a_mode & O_EXCL) \|\|
	88	(pfs->pfs_flags & O_EXCL) && (ap->a_mode & FWRITE))
	89	return (EBUSY);
	90
	91
	92	if (ap->a_mode & FWRITE)
	93	pfs->pfs_flags = ap->a_mode & (FWRITE\|O_EXCL);
	94
	95	return (0);
	96
	97	default:
	98	break;
	99	}
	100
	101	return (0);
	102	}
	103
	104	/*
	105	* close the pfsnode (vp) after doing i/o.
	106	* (vp) is not locked on entry or exit.
	107	*
	108	* nothing to do for procfs other than undo
	109	* any exclusive open flag (see _open above).
	110	*/
	111	procfs_close(ap)
	112	struct vop_close_args *ap;
	113	{
	114	struct pfsnode *pfs = VTOPFS(ap->a_vp);
	115
	116	switch (pfs->pfs_type) {
	117	case Pmem:
	118	if ((ap->a_fflag & FWRITE) && (pfs->pfs_flags & O_EXCL))
	119	pfs->pfs_flags &= ~(FWRITE\|O_EXCL);
	120	break;
	121	}
	122
	123	return (0);
124	}
125
126	/*
127	* do an ioctl operation on pfsnode (vp).
128	* (vp) is not locked on entry or exit.
129	*/
130	procfs_ioctl(ap)
131	struct vop_ioctl_args *ap;
132	{
133
134	return (ENOTTY);
135	}
136
137	/*
138	* do block mapping for pfsnode (vp).
139	* since we don't use the buffer cache
140	* for procfs this function should never
141	* be called. in any case, it's not clear
142	* what part of the kernel ever makes use
143	* of this function. for sanity, this is the
144	* usual no-op bmap, although returning
145	* (EIO) would be a reasonable alternative.
146	*/
147	procfs_bmap(ap)
148	struct vop_bmap_args *ap;
149	{
150
151	if (ap->a_vpp != NULL)
152	*ap->a_vpp = ap->a_vp;
153	if (ap->a_bnp != NULL)
154	*ap->a_bnp = ap->a_bn;
155	return (0);
156	}
157
158	/*
159	* _inactive is called when the pfsnode
160	* is vrele'd and the reference count goes
161	* to zero. (vp) will be on the vnode free
162	* list, so to get it back vget() must be
163	* used.
164	*
165	* for procfs, check if the process is still
166	* alive and if it isn't then just throw away
167	* the vnode by calling vgone(). this may
168	* be overkill and a waste of time since the
169	* chances are that the process will still be
170	* there and PFIND is not free.
171	*
172	* (vp) is not locked on entry or exit.
173	*/
174	procfs_inactive(ap)
175	struct vop_inactive_args *ap;
176	{
177	struct pfsnode *pfs = VTOPFS(ap->a_vp);
178
179	if (PFIND(pfs->pfs_pid) == 0)
180	vgone(ap->a_vp);
181
182	return (0);
183	}
184
185	/*
186	* _reclaim is called when getnewvnode()
187	* wants to make use of an entry on the vnode
188	* free list. at this time the filesystem needs
189	* to free any private data and remove the node
190	* from any private lists.
191	*/
192	procfs_reclaim(ap)
193	struct vop_reclaim_args *ap;
194	{
195	int error;
196
197	error = procfs_freevp(ap->a_vp);
198	return (error);
199	}
200
0020ca6e JSP	201	/*
	202	* Return POSIX pathconf information applicable to special devices.
	203	*/
	204	procfs_pathconf(ap)
	205	struct vop_pathconf_args /* {
	206	struct vnode *a_vp;
	207	int a_name;
	208	int *a_retval;
	209	} / ap;
	210	{
	211
	212	switch (ap->a_name) {
	213	case _PC_LINK_MAX:
	214	*ap->a_retval = LINK_MAX;
	215	return (0);
	216	case _PC_MAX_CANON:
	217	*ap->a_retval = MAX_CANON;
	218	return (0);
	219	case _PC_MAX_INPUT:
	220	*ap->a_retval = MAX_INPUT;
	221	return (0);
	222	case _PC_PIPE_BUF:
	223	*ap->a_retval = PIPE_BUF;
	224	return (0);
	225	case _PC_CHOWN_RESTRICTED:
	226	*ap->a_retval = 1;
	227	return (0);
	228	case _PC_VDISABLE:
	229	*ap->a_retval = _POSIX_VDISABLE;
	230	return (0);
	231	default:
	232	return (EINVAL);
	233	}
	234	/* NOTREACHED */
	235	}
	236
76434cf5 JSP	237	/*
	238	* _print is used for debugging.
	239	* just print a readable description
	240	* of (vp).
	241	*/
	242	procfs_print(ap)
	243	struct vop_print_args *ap;
	244	{
	245	struct pfsnode *pfs = VTOPFS(ap->a_vp);
	246
	247	printf("tag VT_PROCFS, pid %d, mode %x, flags %x\n",
	248	pfs->pfs_pid,
	249	pfs->pfs_mode, pfs->pfs_flags);
	250	}
	251
	252	/*
	253	* _abortop is called when operations such as
	254	* rename and create fail. this entry is responsible
	255	* for undoing any side-effects caused by the lookup.
	256	* this will always include freeing the pathname buffer.
	257	*/
	258	procfs_abortop(ap)
	259	struct vop_abortop_args *ap;
	260	{
	261
	262	if ((ap->a_cnp->cn_flags & (HASBUF \| SAVESTART)) == HASBUF)
	263	FREE(ap->a_cnp->cn_pnbuf, M_NAMEI);
	264	return (0);
	265	}
	266
	267	/*
	268	* generic entry point for unsupported operations
	269	*/
	270	procfs_badop()
	271	{
	272
	273	return (EIO);
	274	}
	275
	276	/*
	277	* Invent attributes for pfsnode (vp) and store
	278	* them in (vap).
	279	* Directories lengths are returned as zero since
	280	* any real length would require the genuine size
	281	* to be computed, and nothing cares anyway.
	282	*
	283	* this is relatively minimal for procfs.
	284	*/
	285	procfs_getattr(ap)
	286	struct vop_getattr_args *ap;
	287	{
	288	struct pfsnode *pfs = VTOPFS(ap->a_vp);
5d86e205	289	struct vattr *vap = ap->a_vap;
76434cf5 JSP	290	struct proc *procp;
	291	int error;
	292
	293	/* first check the process still exists */
	294	procp = PFIND(pfs->pfs_pid);
	295	if (procp == 0)
	296	return (ENOENT);
	297
	298	error = 0;
	299
	300	/* start by zeroing out the attributes */
5d86e205	301	VATTR_NULL(vap);
76434cf5 JSP	302
76434cf5 JSP	303	/* next do all the common fields */
5d86e205 JSP	304	vap->va_type = ap->a_vp->v_type;
	305	vap->va_mode = pfs->pfs_mode;
	306	vap->va_fileid = pfs->pfs_fileno;
	307	vap->va_flags = 0;
	308	vap->va_blocksize = PAGE_SIZE;
	309	vap->va_bytes = vap->va_size = 0;
76434cf5 JSP	310
	311	/*
	312	* Make all times be current TOD.
	313	* It would be possible to get the process start
	314	* time from the p_stat structure, but there's
	315	* no "file creation" time stamp anyway, and the
	316	* p_stat structure is not addressible if u. gets
	317	* swapped out for that process.
	318	*/
5d86e205 JSP	319	microtime(&vap->va_ctime);
5d86e205 JSP	320	vap->va_atime = vap->va_mtime = vap->va_ctime;
76434cf5 JSP	321
	322	/*
	323	* now do the object specific fields
	324	*
	325	* The size could be set from struct reg, but it's hardly
	326	* worth the trouble, and it puts some (potentially) machine
	327	* dependent data into this machine-independent code. If it
	328	* becomes important then this function should break out into
	329	* a per-file stat function in the corresponding .c file.
	330	*/
	331
	332	switch (pfs->pfs_type) {
	333	case Proot:
5d86e205 JSP	334	vap->va_nlink = 2;
	335	vap->va_uid = 0;
	336	vap->va_gid = 0;
76434cf5 JSP	337	break;
	338
	339	case Pproc:
5d86e205 JSP	340	vap->va_nlink = 2;
	341	vap->va_uid = procp->p_ucred->cr_uid;
	342	vap->va_gid = procp->p_ucred->cr_gid;
76434cf5 JSP	343	break;
	344
	345	case Pfile:
	346	error = EOPNOTSUPP;
	347	break;
	348
	349	case Pmem:
5d86e205 JSP	350	vap->va_nlink = 1;
5d86e205 JSP	351	vap->va_bytes = vap->va_size =
76434cf5 JSP	352	ctob(procp->p_vmspace->vm_tsize +
	353	procp->p_vmspace->vm_dsize +
	354	procp->p_vmspace->vm_ssize);
5d86e205 JSP	355	vap->va_uid = procp->p_ucred->cr_uid;
5d86e205 JSP	356	vap->va_gid = procp->p_ucred->cr_gid;
76434cf5 JSP	357	break;
	358
	359	case Pregs:
6fdd7e21	360	case Pfpregs:
76434cf5 JSP	361	case Pctl:
	362	case Pstatus:
	363	case Pnote:
	364	case Pnotepg:
5d86e205 JSP	365	vap->va_nlink = 1;
	366	vap->va_uid = procp->p_ucred->cr_uid;
	367	vap->va_gid = procp->p_ucred->cr_gid;
76434cf5 JSP	368	break;
	369
	370	default:
	371	panic("procfs_getattr");
	372	}
	373
	374	return (error);
	375	}
	376
	377	procfs_setattr(ap)
	378	struct vop_setattr_args *ap;
	379	{
	380	/*
	381	* just fake out attribute setting
	382	* it's not good to generate an error
	383	* return, otherwise things like creat()
	384	* will fail when they try to set the
	385	* file length to 0. worse, this means
	386	* that echo $note > /proc/$pid/note will fail.
	387	*/
	388
	389	return (0);
	390	}
	391
	392	/*
	393	* implement access checking.
	394	*
	395	* something very similar to this code is duplicated
	396	* throughout the 4bsd kernel and should be moved
	397	* into kern/vfs_subr.c sometime.
	398	*
	399	* actually, the check for super-user is slightly
	400	* broken since it will allow read access to write-only
	401	* objects. this doesn't cause any particular trouble
	402	* but does mean that the i/o entry points need to check
	403	* that the operation really does make sense.
	404	*/
	405	procfs_access(ap)
	406	struct vop_access_args *ap;
	407	{
	408	struct vattr *vap;
	409	struct vattr vattr;
	410	int error;
	411
	412	/*
	413	* If you're the super-user,
	414	* you always get access.
	415	*/
	416	if (ap->a_cred->cr_uid == (uid_t) 0)
	417	return (0);
	418	vap = &vattr;
	419	if (error = VOP_GETATTR(ap->a_vp, vap, ap->a_cred, ap->a_p))
	420	return (error);
	421
	422	/*
	423	* Access check is based on only one of owner, group, public.
	424	* If not owner, then check group. If not a member of the
	425	* group, then check public access.
	426	*/
	427	if (ap->a_cred->cr_uid != vap->va_uid) {
	428	gid_t *gp;
	429	int i;
	430
	431	(ap->a_mode) >>= 3;
432	gp = ap->a_cred->cr_groups;
433	for (i = 0; i < ap->a_cred->cr_ngroups; i++, gp++)
434	if (vap->va_gid == *gp)
435	goto found;
436	ap->a_mode >>= 3;
437	found:
438	;
439	}
440
441	if ((vap->va_mode & ap->a_mode) == ap->a_mode)
442	return (0);
443
444	return (EACCES);
445	}
446
447	/*
448	* lookup. this is incredibly complicated in the
449	* general case, however for most pseudo-filesystems
450	* very little needs to be done.
451	*
452	* unless you want to get a migraine, just make sure your
453	* filesystem doesn't do any locking of its own. otherwise
454	* read and inwardly digest ufs_lookup().
455	*/
456	procfs_lookup(ap)
457	struct vop_lookup_args *ap;
458	{
459	struct componentname *cnp = ap->a_cnp;
460	struct vnode **vpp = ap->a_vpp;
461	struct vnode *dvp = ap->a_dvp;
462	char *pname = cnp->cn_nameptr;
463	int error = 0;
76434cf5 JSP	464	pid_t pid;
	465	struct vnode *nvp;
	466	struct pfsnode *pfs;
	467	struct proc *procp;
76434cf5 JSP	468	pfstype pfs_type;
	469	int i;
	470
	471	if (cnp->cn_namelen == 1 && *pname == '.') {
	472	*vpp = dvp;
	473	VREF(dvp);
	474	/VOP_LOCK(dvp);/
	475	return (0);
	476	}
	477
	478	*vpp = NULL;
	479
	480	pfs = VTOPFS(dvp);
	481	switch (pfs->pfs_type) {
	482	case Proot:
	483	if (cnp->cn_flags & ISDOTDOT)
	484	return (EIO);
	485
	486	if (CNEQ(cnp, "curproc", 7))
	487	pid = cnp->cn_proc->p_pid;
	488	else
	489	pid = atopid(pname, cnp->cn_namelen);
	490	if (pid == NO_PID)
	491	return (ENOENT);
	492
	493	procp = PFIND(pid);
	494	if (procp == 0)
	495	return (ENOENT);
	496
	497	error = procfs_allocvp(dvp->v_mount, &nvp, pid, Pproc);
	498	if (error)
	499	return (error);
	500
	501	nvp->v_type = VDIR;
	502	pfs = VTOPFS(nvp);
	503
	504	*vpp = nvp;
	505	return (0);
	506
	507	case Pproc:
	508	if (cnp->cn_flags & ISDOTDOT) {
	509	error = procfs_root(dvp->v_mount, vpp);
	510	return (error);
	511	}
	512
	513	procp = PFIND(pfs->pfs_pid);
	514	if (procp == 0)
	515	return (ENOENT);
	516
	517	for (i = 0; i < Nprocent; i++) {
	518	struct pfsnames *dp = &procent[i];
	519
	520	if (cnp->cn_namelen == dp->d_namlen &&
	521	bcmp(pname, dp->d_name, dp->d_namlen) == 0) {
	522	pfs_type = dp->d_pfstype;
	523	goto found;
	524	}
	525	}
	526	return (ENOENT);
	527
	528	found:
	529	if (pfs_type == Pfile) {
	530	nvp = procfs_findtextvp(procp);
	531	if (nvp) {
532	VREF(nvp);
533	VOP_LOCK(nvp);
534	} else {
535	error = ENXIO;
536	}
537	} else {
538	error = procfs_allocvp(dvp->v_mount, &nvp,
539	pfs->pfs_pid, pfs_type);
540	if (error)
541	return (error);
542
543	nvp->v_type = VREG;
544	pfs = VTOPFS(nvp);
545	}
546	*vpp = nvp;
547	return (error);
548
549	default:
550	return (ENOTDIR);
551	}
552	}
553
554	/*
555	* readdir returns directory entries from pfsnode (vp).
556	*
557	* the strategy here with procfs is to generate a single
558	* directory entry at a time (struct pfsdent) and then
559	* copy that out to userland using uiomove. a more efficent
560	* though more complex implementation, would try to minimize
561	* the number of calls to uiomove(). for procfs, this is
562	* hardly worth the added code complexity.
563	*
564	* this should just be done through read()
565	*/
566	procfs_readdir(ap)
567	struct vop_readdir_args *ap;
568	{
569	struct uio *uio = ap->a_uio;
570	struct pfsdent d;
571	struct pfsdent *dp = &d;
572	struct pfsnode *pfs;
573	int error;
574	int count;
575	int i;
576
577	pfs = VTOPFS(ap->a_vp);
578
579	if (uio->uio_resid < UIO_MX)
580	return (EINVAL);
581	if (uio->uio_offset & (UIO_MX-1))
582	return (EINVAL);
583	if (uio->uio_offset < 0)
584	return (EINVAL);
585
586	error = 0;
587	count = 0;
588	i = uio->uio_offset / UIO_MX;
589
590	switch (pfs->pfs_type) {
591	/*
592	* this is for the process-specific sub-directories.
593	* all that is needed to is copy out all the entries
594	* from the procent[] table (top of this file).
595	*/
596	case Pproc: {
597	while (uio->uio_resid >= UIO_MX) {
598	struct pfsnames *dt;
599
600	if (i >= Nprocent)
601	break;
602
603	dt = &procent[i];
604
605	dp->d_reclen = UIO_MX;
606	dp->d_fileno = PROCFS_FILENO(pfs->pfs_pid, dt->d_pfstype);
607	dp->d_type = DT_REG;
608	dp->d_namlen = dt->d_namlen;
609	bcopy(dt->d_name, dp->d_name, sizeof(dt->d_name)-1);
610	error = uiomove((caddr_t) dp, UIO_MX, uio);
611	if (error)
612	break;
613	count += UIO_MX;
614	i++;
615	}
616
617	break;
618
619	}
620
621	/*
622	* this is for the root of the procfs filesystem
623	* what is needed is a special entry for "curproc"
624	* followed by an entry for each process on allproc
625	#ifdef PROCFS_ZOMBIE
626	* and zombproc.
627	#endif
628	*/
629
630	case Proot: {
631	int pcnt;
632	#ifdef PROCFS_ZOMBIE
633	int doingzomb = 0;
634	#endif
635	volatile struct proc *p;
636
637	p = allproc;
638
639	#define PROCFS_XFILES 1 /* number of other entries, like "curproc" */
640	pcnt = PROCFS_XFILES;
641
642	while (p && uio->uio_resid >= UIO_MX) {
643	bzero((char *) dp, UIO_MX);
644	dp->d_type = DT_DIR;
645	dp->d_reclen = UIO_MX;
646
647	switch (i) {
648	case 0:
649	/* ship out entry for "curproc" */
650	dp->d_fileno = PROCFS_FILENO(PID_MAX+1, Pproc);
5d86e205	651	dp->d_namlen = sprintf(dp->d_name, "curproc");
76434cf5 JSP	652	break;
	653
	654	default:
	655	if (pcnt >= i) {
	656	dp->d_fileno = PROCFS_FILENO(p->p_pid, Pproc);
	657	dp->d_namlen = sprintf(dp->d_name, "%ld", (long) p->p_pid);
	658	}
	659
	660	p = p->p_next;
	661
	662	#ifdef PROCFS_ZOMBIE
	663	if (p == 0 && doingzomb == 0) {
	664	doingzomb = 1;
	665	p = zombproc;
	666	}
	667	#endif
	668
	669	if (pcnt++ < i)
	670	continue;
	671
	672	break;
	673	}
	674	error = uiomove((caddr_t) dp, UIO_MX, uio);
	675	if (error)
	676	break;
	677	count += UIO_MX;
	678	i++;
	679	}
	680
	681	break;
	682
	683	}
	684
	685	default:
	686	error = ENOTDIR;
	687	break;
	688	}
	689
	690	uio->uio_offset = i * UIO_MX;
	691
	692	return (error);
	693	}
	694
	695	/*
	696	* convert decimal ascii to pid_t
	697	*/
	698	static pid_t
	699	atopid(b, len)
	700	const char *b;
	701	u_int len;
	702	{
	703	pid_t p = 0;
	704
	705	while (len--) {
	706	char c = *b++;
	707	if (c < '0' \|\| c > '9')
	708	return (NO_PID);
	709	p = 10 * p + (c - '0');
	710	if (p > PID_MAX)
	711	return (NO_PID);
	712	}
	713
	714	return (p);
	715	}
716
717	/*
718	* procfs vnode operations.
719	*/
720	int (**procfs_vnodeop_p)();
721	struct vnodeopv_entry_desc procfs_vnodeop_entries[] = {
722	{ &vop_default_desc, vn_default_error },
723	{ &vop_lookup_desc, procfs_lookup }, /* lookup */
724	{ &vop_create_desc, procfs_create }, /* create */
725	{ &vop_mknod_desc, procfs_mknod }, /* mknod */
726	{ &vop_open_desc, procfs_open }, /* open */
727	{ &vop_close_desc, procfs_close }, /* close */
728	{ &vop_access_desc, procfs_access }, /* access */
729	{ &vop_getattr_desc, procfs_getattr }, /* getattr */
730	{ &vop_setattr_desc, procfs_setattr }, /* setattr */
731	{ &vop_read_desc, procfs_read }, /* read */
732	{ &vop_write_desc, procfs_write }, /* write */
733	{ &vop_ioctl_desc, procfs_ioctl }, /* ioctl */
734	{ &vop_select_desc, procfs_select }, /* select */
735	{ &vop_mmap_desc, procfs_mmap }, /* mmap */
736	{ &vop_fsync_desc, procfs_fsync }, /* fsync */
737	{ &vop_seek_desc, procfs_seek }, /* seek */
738	{ &vop_remove_desc, procfs_remove }, /* remove */
739	{ &vop_link_desc, procfs_link }, /* link */
740	{ &vop_rename_desc, procfs_rename }, /* rename */
741	{ &vop_mkdir_desc, procfs_mkdir }, /* mkdir */
742	{ &vop_rmdir_desc, procfs_rmdir }, /* rmdir */
743	{ &vop_symlink_desc, procfs_symlink }, /* symlink */
744	{ &vop_readdir_desc, procfs_readdir }, /* readdir */
745	{ &vop_readlink_desc, procfs_readlink }, /* readlink */
746	{ &vop_abortop_desc, procfs_abortop }, /* abortop */
747	{ &vop_inactive_desc, procfs_inactive }, /* inactive */
748	{ &vop_reclaim_desc, procfs_reclaim }, /* reclaim */
749	{ &vop_lock_desc, procfs_lock }, /* lock */
750	{ &vop_unlock_desc, procfs_unlock }, /* unlock */
751	{ &vop_bmap_desc, procfs_bmap }, /* bmap */
752	{ &vop_strategy_desc, procfs_strategy }, /* strategy */
753	{ &vop_print_desc, procfs_print }, /* print */
754	{ &vop_islocked_desc, procfs_islocked }, /* islocked */
755	{ &vop_pathconf_desc, procfs_pathconf }, /* pathconf */
756	{ &vop_advlock_desc, procfs_advlock }, /* advlock */
757	{ &vop_blkatoff_desc, procfs_blkatoff }, /* blkatoff */
758	{ &vop_valloc_desc, procfs_valloc }, /* valloc */
759	{ &vop_vfree_desc, procfs_vfree }, /* vfree */
760	{ &vop_truncate_desc, procfs_truncate }, /* truncate */
761	{ &vop_update_desc, procfs_update }, /* update */
762	{ (struct vnodeop_desc)NULL, (int()())NULL }
763	};
764	struct vnodeopv_desc procfs_vnodeop_opv_desc =
765	{ &procfs_vnodeop_p, procfs_vnodeop_entries };