165524Spendry /*
265524Spendry  * Copyright (c) 1993 Jan-Simon Pendry
365808Sbostic  * Copyright (c) 1993
465808Sbostic  *	The Regents of the University of California.  All rights reserved.
565524Spendry  *
665524Spendry  * This code is derived from software contributed to Berkeley by
765524Spendry  * Jan-Simon Pendry.
865524Spendry  *
965524Spendry  * %sccs.include.redist.c%
1065524Spendry  *
11*67721Smckusick  *	@(#)procfs_vnops.c	8.9 (Berkeley) 08/20/94
1265524Spendry  *
1365524Spendry  * From:
1465524Spendry  *	$Id: procfs_vnops.c,v 3.2 1993/12/15 09:40:17 jsp Exp $
1565524Spendry  */
1665524Spendry 
1765524Spendry /*
1865524Spendry  * procfs vnode interface
1965524Spendry  */
2065524Spendry 
2165524Spendry #include <sys/param.h>
2265524Spendry #include <sys/systm.h>
2365524Spendry #include <sys/time.h>
2465524Spendry #include <sys/kernel.h>
2565524Spendry #include <sys/file.h>
2665524Spendry #include <sys/proc.h>
2765524Spendry #include <sys/vnode.h>
2865524Spendry #include <sys/namei.h>
2965524Spendry #include <sys/malloc.h>
3065524Spendry #include <sys/dirent.h>
3165524Spendry #include <sys/resourcevar.h>
3267389Spendry #include <vm/vm.h>	/* for PAGE_SIZE */
3367389Spendry #include <machine/reg.h>
3465524Spendry #include <miscfs/procfs/procfs.h>
3565524Spendry 
3665524Spendry /*
3765524Spendry  * Vnode Operations.
3865524Spendry  *
3965524Spendry  */
4065524Spendry 
4165524Spendry /*
4265524Spendry  * This is a list of the valid names in the
4365524Spendry  * process-specific sub-directories.  It is
4465524Spendry  * used in procfs_lookup and procfs_readdir
4565524Spendry  */
4665524Spendry static struct pfsnames {
4767389Spendry 	u_char	d_type;
4867389Spendry 	u_char	d_namlen;
4965524Spendry 	char	d_name[PROCFS_NAMELEN];
5065524Spendry 	pfstype	d_pfstype;
5167389Spendry 	int	(*d_valid) __P((struct proc *p));
5265524Spendry } procent[] = {
5365524Spendry #define N(s) sizeof(s)-1, s
5465524Spendry 	/* namlen, nam, type */
5567389Spendry 	{ DT_DIR, N("."),	Pproc,		NULL },
5667389Spendry 	{ DT_DIR, N(".."),	Proot,		NULL },
5767389Spendry 	{ DT_REG, N("file"),	Pfile,		procfs_validfile },
5867389Spendry 	{ DT_REG, N("mem"),	Pmem,		NULL },
5967389Spendry 	{ DT_REG, N("regs"),	Pregs,		procfs_validregs },
6067389Spendry 	{ DT_REG, N("fpregs"),	Pfpregs,	procfs_validfpregs },
6167389Spendry 	{ DT_REG, N("ctl"),	Pctl,		NULL },
6267389Spendry 	{ DT_REG, N("status"),	Pstatus,	NULL },
6367389Spendry 	{ DT_REG, N("note"),	Pnote,		NULL },
6467389Spendry 	{ DT_REG, N("notepg"),	Pnotepg,	NULL },
6565524Spendry #undef N
6665524Spendry };
6765524Spendry #define Nprocent (sizeof(procent)/sizeof(procent[0]))
6865524Spendry 
6965524Spendry static pid_t atopid __P((const char *, u_int));
7065524Spendry 
7165524Spendry /*
7265524Spendry  * set things up for doing i/o on
7365524Spendry  * the pfsnode (vp).  (vp) is locked
7465524Spendry  * on entry, and should be left locked
7565524Spendry  * on exit.
7665524Spendry  *
7765524Spendry  * for procfs we don't need to do anything
7865524Spendry  * in particular for i/o.  all that is done
7965524Spendry  * is to support exclusive open on process
8065524Spendry  * memory images.
8165524Spendry  */
8265524Spendry procfs_open(ap)
8367367Smckusick 	struct vop_open_args /* {
8467367Smckusick 		struct vnode *a_vp;
8567367Smckusick 		int  a_mode;
8667367Smckusick 		struct ucred *a_cred;
8767367Smckusick 		struct proc *a_p;
8867367Smckusick 	} */ *ap;
8965524Spendry {
9065524Spendry 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
9165524Spendry 
9265524Spendry 	switch (pfs->pfs_type) {
9365524Spendry 	case Pmem:
9465524Spendry 		if (PFIND(pfs->pfs_pid) == 0)
9565524Spendry 			return (ENOENT);	/* was ESRCH, jsp */
9665524Spendry 
9765524Spendry 		if ((pfs->pfs_flags & FWRITE) && (ap->a_mode & O_EXCL) ||
9867389Spendry 		    (pfs->pfs_flags & O_EXCL) && (ap->a_mode & FWRITE))
9965524Spendry 			return (EBUSY);
10065524Spendry 
10165524Spendry 		if (ap->a_mode & FWRITE)
10265524Spendry 			pfs->pfs_flags = ap->a_mode & (FWRITE|O_EXCL);
10365524Spendry 
10465524Spendry 		return (0);
10565524Spendry 
10665524Spendry 	default:
10765524Spendry 		break;
10865524Spendry 	}
10965524Spendry 
11065524Spendry 	return (0);
11165524Spendry }
11265524Spendry 
11365524Spendry /*
11465524Spendry  * close the pfsnode (vp) after doing i/o.
11565524Spendry  * (vp) is not locked on entry or exit.
11665524Spendry  *
11765524Spendry  * nothing to do for procfs other than undo
11865524Spendry  * any exclusive open flag (see _open above).
11965524Spendry  */
12065524Spendry procfs_close(ap)
12167367Smckusick 	struct vop_close_args /* {
12267367Smckusick 		struct vnode *a_vp;
12367367Smckusick 		int  a_fflag;
12467367Smckusick 		struct ucred *a_cred;
12567367Smckusick 		struct proc *a_p;
12667367Smckusick 	} */ *ap;
12765524Spendry {
12865524Spendry 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
12965524Spendry 
13065524Spendry 	switch (pfs->pfs_type) {
13165524Spendry 	case Pmem:
13265524Spendry 		if ((ap->a_fflag & FWRITE) && (pfs->pfs_flags & O_EXCL))
13365524Spendry 			pfs->pfs_flags &= ~(FWRITE|O_EXCL);
13465524Spendry 		break;
13565524Spendry 	}
13665524Spendry 
13765524Spendry 	return (0);
13865524Spendry }
13965524Spendry 
14065524Spendry /*
14165524Spendry  * do an ioctl operation on pfsnode (vp).
14265524Spendry  * (vp) is not locked on entry or exit.
14365524Spendry  */
14465524Spendry procfs_ioctl(ap)
14567367Smckusick 	struct vop_ioctl_args /* {
14667367Smckusick 		struct vnode *a_vp;
14767367Smckusick 		int a_command;
14867367Smckusick 		caddr_t a_data;
14967367Smckusick 		int a_fflag;
15067367Smckusick 		struct ucred *a_cred;
15167367Smckusick 		struct proc *a_p;
15267367Smckusick 	} */ *ap;
15365524Spendry {
15465524Spendry 
15565524Spendry 	return (ENOTTY);
15665524Spendry }
15765524Spendry 
15865524Spendry /*
15965524Spendry  * do block mapping for pfsnode (vp).
16065524Spendry  * since we don't use the buffer cache
16165524Spendry  * for procfs this function should never
16265524Spendry  * be called.  in any case, it's not clear
16365524Spendry  * what part of the kernel ever makes use
16465524Spendry  * of this function.  for sanity, this is the
16565524Spendry  * usual no-op bmap, although returning
16665524Spendry  * (EIO) would be a reasonable alternative.
16765524Spendry  */
16865524Spendry procfs_bmap(ap)
16967367Smckusick 	struct vop_bmap_args /* {
17067367Smckusick 		struct vnode *a_vp;
17167367Smckusick 		daddr_t  a_bn;
17267367Smckusick 		struct vnode **a_vpp;
17367367Smckusick 		daddr_t *a_bnp;
17467367Smckusick 	} */ *ap;
17565524Spendry {
17665524Spendry 
17765524Spendry 	if (ap->a_vpp != NULL)
17865524Spendry 		*ap->a_vpp = ap->a_vp;
17965524Spendry 	if (ap->a_bnp != NULL)
18065524Spendry 		*ap->a_bnp = ap->a_bn;
18165524Spendry 	return (0);
18265524Spendry }
18365524Spendry 
18465524Spendry /*
18565524Spendry  * _inactive is called when the pfsnode
18665524Spendry  * is vrele'd and the reference count goes
18765524Spendry  * to zero.  (vp) will be on the vnode free
18865524Spendry  * list, so to get it back vget() must be
18965524Spendry  * used.
19065524Spendry  *
19165524Spendry  * for procfs, check if the process is still
19265524Spendry  * alive and if it isn't then just throw away
19365524Spendry  * the vnode by calling vgone().  this may
19465524Spendry  * be overkill and a waste of time since the
19565524Spendry  * chances are that the process will still be
19665524Spendry  * there and PFIND is not free.
19765524Spendry  *
19865524Spendry  * (vp) is not locked on entry or exit.
19965524Spendry  */
20065524Spendry procfs_inactive(ap)
20167367Smckusick 	struct vop_inactive_args /* {
20267367Smckusick 		struct vnode *a_vp;
20367367Smckusick 	} */ *ap;
20465524Spendry {
20565524Spendry 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
20665524Spendry 
20765524Spendry 	if (PFIND(pfs->pfs_pid) == 0)
20865524Spendry 		vgone(ap->a_vp);
20965524Spendry 
21065524Spendry 	return (0);
21165524Spendry }
21265524Spendry 
21365524Spendry /*
21465524Spendry  * _reclaim is called when getnewvnode()
21565524Spendry  * wants to make use of an entry on the vnode
21665524Spendry  * free list.  at this time the filesystem needs
21765524Spendry  * to free any private data and remove the node
21865524Spendry  * from any private lists.
21965524Spendry  */
22065524Spendry procfs_reclaim(ap)
22167367Smckusick 	struct vop_reclaim_args /* {
22267367Smckusick 		struct vnode *a_vp;
22367367Smckusick 	} */ *ap;
22465524Spendry {
22565524Spendry 
22667389Spendry 	return (procfs_freevp(ap->a_vp));
22765524Spendry }
22865524Spendry 
22965524Spendry /*
23065743Spendry  * Return POSIX pathconf information applicable to special devices.
23165743Spendry  */
23265743Spendry procfs_pathconf(ap)
23365743Spendry 	struct vop_pathconf_args /* {
23465743Spendry 		struct vnode *a_vp;
23565743Spendry 		int a_name;
23665743Spendry 		int *a_retval;
23765743Spendry 	} */ *ap;
23865743Spendry {
23965743Spendry 
24065743Spendry 	switch (ap->a_name) {
24165743Spendry 	case _PC_LINK_MAX:
24265743Spendry 		*ap->a_retval = LINK_MAX;
24365743Spendry 		return (0);
24465743Spendry 	case _PC_MAX_CANON:
24565743Spendry 		*ap->a_retval = MAX_CANON;
24665743Spendry 		return (0);
24765743Spendry 	case _PC_MAX_INPUT:
24865743Spendry 		*ap->a_retval = MAX_INPUT;
24965743Spendry 		return (0);
25065743Spendry 	case _PC_PIPE_BUF:
25165743Spendry 		*ap->a_retval = PIPE_BUF;
25265743Spendry 		return (0);
25365743Spendry 	case _PC_CHOWN_RESTRICTED:
25465743Spendry 		*ap->a_retval = 1;
25565743Spendry 		return (0);
25665743Spendry 	case _PC_VDISABLE:
25765743Spendry 		*ap->a_retval = _POSIX_VDISABLE;
25865743Spendry 		return (0);
25965743Spendry 	default:
26065743Spendry 		return (EINVAL);
26165743Spendry 	}
26265743Spendry 	/* NOTREACHED */
26365743Spendry }
26465743Spendry 
26565743Spendry /*
26665524Spendry  * _print is used for debugging.
26765524Spendry  * just print a readable description
26865524Spendry  * of (vp).
26965524Spendry  */
27065524Spendry procfs_print(ap)
27167367Smckusick 	struct vop_print_args /* {
27267367Smckusick 		struct vnode *a_vp;
27367367Smckusick 	} */ *ap;
27465524Spendry {
27565524Spendry 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
27665524Spendry 
27767389Spendry 	printf("tag VT_PROCFS, type %s, pid %d, mode %x, flags %x\n",
27867389Spendry 	    pfs->pfs_type, pfs->pfs_pid, pfs->pfs_mode, pfs->pfs_flags);
27965524Spendry }
28065524Spendry 
28165524Spendry /*
28265524Spendry  * _abortop is called when operations such as
28365524Spendry  * rename and create fail.  this entry is responsible
28465524Spendry  * for undoing any side-effects caused by the lookup.
28565524Spendry  * this will always include freeing the pathname buffer.
28665524Spendry  */
28765524Spendry procfs_abortop(ap)
28867367Smckusick 	struct vop_abortop_args /* {
28967367Smckusick 		struct vnode *a_dvp;
29067367Smckusick 		struct componentname *a_cnp;
29167367Smckusick 	} */ *ap;
29265524Spendry {
29365524Spendry 
29465524Spendry 	if ((ap->a_cnp->cn_flags & (HASBUF | SAVESTART)) == HASBUF)
29565524Spendry 		FREE(ap->a_cnp->cn_pnbuf, M_NAMEI);
29665524Spendry 	return (0);
29765524Spendry }
29865524Spendry 
29965524Spendry /*
30065524Spendry  * generic entry point for unsupported operations
30165524Spendry  */
30265524Spendry procfs_badop()
30365524Spendry {
30465524Spendry 
30565524Spendry 	return (EIO);
30665524Spendry }
30765524Spendry 
30865524Spendry /*
30965524Spendry  * Invent attributes for pfsnode (vp) and store
31065524Spendry  * them in (vap).
31165524Spendry  * Directories lengths are returned as zero since
31265524Spendry  * any real length would require the genuine size
31365524Spendry  * to be computed, and nothing cares anyway.
31465524Spendry  *
31565524Spendry  * this is relatively minimal for procfs.
31665524Spendry  */
31765524Spendry procfs_getattr(ap)
31867367Smckusick 	struct vop_getattr_args /* {
31967367Smckusick 		struct vnode *a_vp;
32067367Smckusick 		struct vattr *a_vap;
32167367Smckusick 		struct ucred *a_cred;
32267367Smckusick 		struct proc *a_p;
32367367Smckusick 	} */ *ap;
32465524Spendry {
32565524Spendry 	struct pfsnode *pfs = VTOPFS(ap->a_vp);
32665536Spendry 	struct vattr *vap = ap->a_vap;
32765524Spendry 	struct proc *procp;
32865524Spendry 	int error;
32965524Spendry 
33065524Spendry 	/* first check the process still exists */
33166025Spendry 	switch (pfs->pfs_type) {
33266025Spendry 	case Proot:
33367389Spendry 	case Pcurproc:
33466025Spendry 		procp = 0;
33566025Spendry 		break;
33665524Spendry 
33766025Spendry 	default:
33866025Spendry 		procp = PFIND(pfs->pfs_pid);
33966025Spendry 		if (procp == 0)
34066025Spendry 			return (ENOENT);
34166025Spendry 	}
34266025Spendry 
34365524Spendry 	error = 0;
34465524Spendry 
34565524Spendry 	/* start by zeroing out the attributes */
34665536Spendry 	VATTR_NULL(vap);
34765524Spendry 
34865524Spendry 	/* next do all the common fields */
34965536Spendry 	vap->va_type = ap->a_vp->v_type;
35065536Spendry 	vap->va_mode = pfs->pfs_mode;
35165536Spendry 	vap->va_fileid = pfs->pfs_fileno;
35265536Spendry 	vap->va_flags = 0;
35365536Spendry 	vap->va_blocksize = PAGE_SIZE;
35465536Spendry 	vap->va_bytes = vap->va_size = 0;
35565524Spendry 
35665524Spendry 	/*
35767389Spendry 	 * Make all times be current TOD.
35867389Spendry 	 * It would be possible to get the process start
35967389Spendry 	 * time from the p_stat structure, but there's
36067389Spendry 	 * no "file creation" time stamp anyway, and the
36167389Spendry 	 * p_stat structure is not addressible if u. gets
36267389Spendry 	 * swapped out for that process.
36367389Spendry 	 *
36467389Spendry 	 * XXX
36567389Spendry 	 * Note that microtime() returns a timeval, not a timespec.
36667389Spendry 	 */
36767389Spendry 	microtime(&vap->va_ctime);
36867389Spendry 	vap->va_atime = vap->va_mtime = vap->va_ctime;
36967389Spendry 
37067389Spendry 	/*
37166025Spendry 	 * If the process has exercised some setuid or setgid
37266025Spendry 	 * privilege, then rip away read/write permission so
37366025Spendry 	 * that only root can gain access.
37466025Spendry 	 */
37566025Spendry 	switch (pfs->pfs_type) {
37667389Spendry 	case Pmem:
37766025Spendry 	case Pregs:
37866025Spendry 	case Pfpregs:
37966025Spendry 		if (procp->p_flag & P_SUGID)
38066025Spendry 			vap->va_mode &= ~((VREAD|VWRITE)|
38166025Spendry 					  ((VREAD|VWRITE)>>3)|
38266025Spendry 					  ((VREAD|VWRITE)>>6));
38367389Spendry 	case Pctl:
38467389Spendry 	case Pstatus:
38567389Spendry 	case Pnote:
38667389Spendry 	case Pnotepg:
38767389Spendry 		vap->va_nlink = 1;
38867389Spendry 		vap->va_uid = procp->p_ucred->cr_uid;
38967389Spendry 		vap->va_gid = procp->p_ucred->cr_gid;
39066025Spendry 		break;
39166025Spendry 	}
39266025Spendry 
39366025Spendry 	/*
39465524Spendry 	 * now do the object specific fields
39565524Spendry 	 *
39665524Spendry 	 * The size could be set from struct reg, but it's hardly
39765524Spendry 	 * worth the trouble, and it puts some (potentially) machine
39865524Spendry 	 * dependent data into this machine-independent code.  If it
39965524Spendry 	 * becomes important then this function should break out into
40065524Spendry 	 * a per-file stat function in the corresponding .c file.
40165524Spendry 	 */
40265524Spendry 
40365524Spendry 	switch (pfs->pfs_type) {
40465524Spendry 	case Proot:
40567389Spendry 		/*
40667389Spendry 		 * Set nlink to 1 to tell fts(3) we don't actually know.
40767389Spendry 		 */
40867389Spendry 		vap->va_nlink = 1;
40965536Spendry 		vap->va_uid = 0;
41065536Spendry 		vap->va_gid = 0;
41167389Spendry 		vap->va_size = vap->va_bytes = DEV_BSIZE;
41265524Spendry 		break;
41365524Spendry 
41467389Spendry 	case Pcurproc: {
41567389Spendry 		char buf[16];		/* should be enough */
41667389Spendry 		vap->va_nlink = 1;
41767389Spendry 		vap->va_uid = 0;
41867389Spendry 		vap->va_gid = 0;
41967389Spendry 		vap->va_size = vap->va_bytes =
42067389Spendry 		    sprintf(buf, "%ld", (long)curproc->p_pid);
42167389Spendry 		break;
42267389Spendry 	}
42367389Spendry 
42465524Spendry 	case Pproc:
42565536Spendry 		vap->va_nlink = 2;
42665536Spendry 		vap->va_uid = procp->p_ucred->cr_uid;
42765536Spendry 		vap->va_gid = procp->p_ucred->cr_gid;
42867389Spendry 		vap->va_size = vap->va_bytes = DEV_BSIZE;
42965524Spendry 		break;
43065524Spendry 
43165524Spendry 	case Pfile:
43265524Spendry 		error = EOPNOTSUPP;
43365524Spendry 		break;
43465524Spendry 
43565524Spendry 	case Pmem:
43665536Spendry 		vap->va_bytes = vap->va_size =
43765524Spendry 			ctob(procp->p_vmspace->vm_tsize +
43865524Spendry 				    procp->p_vmspace->vm_dsize +
43965524Spendry 				    procp->p_vmspace->vm_ssize);
44065524Spendry 		break;
44165524Spendry 
44265524Spendry 	case Pregs:
44367389Spendry 		vap->va_bytes = vap->va_size = sizeof(struct reg);
44467389Spendry 		break;
44567389Spendry 
44665924Spendry 	case Pfpregs:
44767389Spendry 		vap->va_bytes = vap->va_size = sizeof(struct fpreg);
44867389Spendry 		break;
44967389Spendry 
45065524Spendry 	case Pctl:
45165524Spendry 	case Pstatus:
45265524Spendry 	case Pnote:
45365524Spendry 	case Pnotepg:
45465524Spendry 		break;
45565524Spendry 
45665524Spendry 	default:
45765524Spendry 		panic("procfs_getattr");
45865524Spendry 	}
45965524Spendry 
46065524Spendry 	return (error);
46165524Spendry }
46265524Spendry 
46365524Spendry procfs_setattr(ap)
46467367Smckusick 	struct vop_setattr_args /* {
46567367Smckusick 		struct vnode *a_vp;
46667367Smckusick 		struct vattr *a_vap;
46767367Smckusick 		struct ucred *a_cred;
46867367Smckusick 		struct proc *a_p;
46967367Smckusick 	} */ *ap;
47065524Spendry {
47165524Spendry 	/*
47265524Spendry 	 * just fake out attribute setting
47365524Spendry 	 * it's not good to generate an error
47465524Spendry 	 * return, otherwise things like creat()
47565524Spendry 	 * will fail when they try to set the
47665524Spendry 	 * file length to 0.  worse, this means
47765524Spendry 	 * that echo $note > /proc/$pid/note will fail.
47865524Spendry 	 */
47965524Spendry 
48065524Spendry 	return (0);
48165524Spendry }
48265524Spendry 
48365524Spendry /*
48465524Spendry  * implement access checking.
48565524Spendry  *
48665524Spendry  * something very similar to this code is duplicated
48765524Spendry  * throughout the 4bsd kernel and should be moved
48865524Spendry  * into kern/vfs_subr.c sometime.
48965524Spendry  *
49065524Spendry  * actually, the check for super-user is slightly
49165524Spendry  * broken since it will allow read access to write-only
49265524Spendry  * objects.  this doesn't cause any particular trouble
49365524Spendry  * but does mean that the i/o entry points need to check
49465524Spendry  * that the operation really does make sense.
49565524Spendry  */
49665524Spendry procfs_access(ap)
49767367Smckusick 	struct vop_access_args /* {
49867367Smckusick 		struct vnode *a_vp;
49967367Smckusick 		int a_mode;
50067367Smckusick 		struct ucred *a_cred;
50167367Smckusick 		struct proc *a_p;
50267367Smckusick 	} */ *ap;
50365524Spendry {
50465524Spendry 	struct vattr *vap;
50565524Spendry 	struct vattr vattr;
50665524Spendry 	int error;
50765524Spendry 
50865524Spendry 	/*
50965524Spendry 	 * If you're the super-user,
51065524Spendry 	 * you always get access.
51165524Spendry 	 */
51267389Spendry 	if (ap->a_cred->cr_uid == 0)
51365524Spendry 		return (0);
51467389Spendry 
51565524Spendry 	vap = &vattr;
51665524Spendry 	if (error = VOP_GETATTR(ap->a_vp, vap, ap->a_cred, ap->a_p))
51765524Spendry 		return (error);
51865524Spendry 
51965524Spendry 	/*
52065524Spendry 	 * Access check is based on only one of owner, group, public.
52165524Spendry 	 * If not owner, then check group. If not a member of the
52265524Spendry 	 * group, then check public access.
52365524Spendry 	 */
52465524Spendry 	if (ap->a_cred->cr_uid != vap->va_uid) {
52565524Spendry 		gid_t *gp;
52665524Spendry 		int i;
52765524Spendry 
52867389Spendry 		ap->a_mode >>= 3;
52965524Spendry 		gp = ap->a_cred->cr_groups;
53065524Spendry 		for (i = 0; i < ap->a_cred->cr_ngroups; i++, gp++)
53165524Spendry 			if (vap->va_gid == *gp)
53265524Spendry 				goto found;
53365524Spendry 		ap->a_mode >>= 3;
53465524Spendry found:
53565524Spendry 		;
53665524Spendry 	}
53765524Spendry 
53865524Spendry 	if ((vap->va_mode & ap->a_mode) == ap->a_mode)
53965524Spendry 		return (0);
54065524Spendry 
54165524Spendry 	return (EACCES);
54265524Spendry }
54365524Spendry 
54465524Spendry /*
54565524Spendry  * lookup.  this is incredibly complicated in the
54665524Spendry  * general case, however for most pseudo-filesystems
54765524Spendry  * very little needs to be done.
54865524Spendry  *
54965524Spendry  * unless you want to get a migraine, just make sure your
55065524Spendry  * filesystem doesn't do any locking of its own.  otherwise
55165524Spendry  * read and inwardly digest ufs_lookup().
55265524Spendry  */
55365524Spendry procfs_lookup(ap)
55467367Smckusick 	struct vop_lookup_args /* {
55567367Smckusick 		struct vnode * a_dvp;
55667367Smckusick 		struct vnode ** a_vpp;
55767367Smckusick 		struct componentname * a_cnp;
55867367Smckusick 	} */ *ap;
55965524Spendry {
56065524Spendry 	struct componentname *cnp = ap->a_cnp;
56165524Spendry 	struct vnode **vpp = ap->a_vpp;
56265524Spendry 	struct vnode *dvp = ap->a_dvp;
56365524Spendry 	char *pname = cnp->cn_nameptr;
56465524Spendry 	int error = 0;
56565524Spendry 	pid_t pid;
56665524Spendry 	struct vnode *nvp;
56765524Spendry 	struct pfsnode *pfs;
56865524Spendry 	struct proc *procp;
56965524Spendry 	pfstype pfs_type;
57065524Spendry 	int i;
57165524Spendry 
57265524Spendry 	if (cnp->cn_namelen == 1 && *pname == '.') {
57365524Spendry 		*vpp = dvp;
57465524Spendry 		VREF(dvp);
57565524Spendry 		/*VOP_LOCK(dvp);*/
57665524Spendry 		return (0);
57765524Spendry 	}
57865524Spendry 
57965524Spendry 	*vpp = NULL;
58065524Spendry 
58165524Spendry 	pfs = VTOPFS(dvp);
58265524Spendry 	switch (pfs->pfs_type) {
58365524Spendry 	case Proot:
58465524Spendry 		if (cnp->cn_flags & ISDOTDOT)
58565524Spendry 			return (EIO);
58665524Spendry 
58765524Spendry 		if (CNEQ(cnp, "curproc", 7))
58867389Spendry 			return (procfs_allocvp(dvp->v_mount, vpp, 0, Pcurproc));
58967389Spendry 
59067389Spendry 		pid = atopid(pname, cnp->cn_namelen);
59165524Spendry 		if (pid == NO_PID)
59265524Spendry 			return (ENOENT);
59365524Spendry 
59465524Spendry 		procp = PFIND(pid);
59565524Spendry 		if (procp == 0)
59665524Spendry 			return (ENOENT);
59765524Spendry 
59867389Spendry 		return (procfs_allocvp(dvp->v_mount, vpp, pid, Pproc));
59965524Spendry 
60065524Spendry 	case Pproc:
60165524Spendry 		if (cnp->cn_flags & ISDOTDOT) {
60265524Spendry 			error = procfs_root(dvp->v_mount, vpp);
60365524Spendry 			return (error);
60465524Spendry 		}
60565524Spendry 
60665524Spendry 		procp = PFIND(pfs->pfs_pid);
60765524Spendry 		if (procp == 0)
60865524Spendry 			return (ENOENT);
60965524Spendry 
61065524Spendry 		for (i = 0; i < Nprocent; i++) {
61165524Spendry 			struct pfsnames *dp = &procent[i];
61265524Spendry 
61365524Spendry 			if (cnp->cn_namelen == dp->d_namlen &&
61467389Spendry 			    bcmp(pname, dp->d_name, dp->d_namlen) == 0 &&
61567389Spendry 			    (dp->d_valid == NULL || (*dp->d_valid)(procp))) {
61665524Spendry 			    	pfs_type = dp->d_pfstype;
61765524Spendry 				goto found;
61865524Spendry 			}
61965524Spendry 		}
62065524Spendry 		return (ENOENT);
62165524Spendry 
62265524Spendry 	found:
62365524Spendry 		if (pfs_type == Pfile) {
62465524Spendry 			nvp = procfs_findtextvp(procp);
62567389Spendry 			if (nvp == NULLVP)
62667389Spendry 				return (ENXIO);
62767389Spendry 			VREF(nvp);
62867389Spendry 			VOP_LOCK(nvp);
62967389Spendry 			*vpp = nvp;
63067389Spendry 			return (0);
63165524Spendry 		}
63265524Spendry 
63367389Spendry 		return (procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
63467389Spendry 		    pfs_type));
63567389Spendry 
63665524Spendry 	default:
63765524Spendry 		return (ENOTDIR);
63865524Spendry 	}
63965524Spendry }
64065524Spendry 
64167389Spendry int
64267389Spendry procfs_validfile(p)
64367389Spendry 	struct proc *p;
64467389Spendry {
64567389Spendry 
64667389Spendry 	return (procfs_findtextvp(p) != NULLVP);
64767389Spendry }
64867389Spendry 
64965524Spendry /*
65065524Spendry  * readdir returns directory entries from pfsnode (vp).
65165524Spendry  *
65265524Spendry  * the strategy here with procfs is to generate a single
65365524Spendry  * directory entry at a time (struct pfsdent) and then
65465524Spendry  * copy that out to userland using uiomove.  a more efficent
65565524Spendry  * though more complex implementation, would try to minimize
65665524Spendry  * the number of calls to uiomove().  for procfs, this is
65765524Spendry  * hardly worth the added code complexity.
65865524Spendry  *
65965524Spendry  * this should just be done through read()
66065524Spendry  */
66165524Spendry procfs_readdir(ap)
66267367Smckusick 	struct vop_readdir_args /* {
66367367Smckusick 		struct vnode *a_vp;
66467367Smckusick 		struct uio *a_uio;
66567367Smckusick 		struct ucred *a_cred;
66667367Smckusick 		int *a_eofflag;
66767367Smckusick 		u_long *a_cookies;
66867367Smckusick 		int a_ncookies;
66967367Smckusick 	} */ *ap;
67065524Spendry {
67165524Spendry 	struct uio *uio = ap->a_uio;
67265524Spendry 	struct pfsdent d;
67365524Spendry 	struct pfsdent *dp = &d;
67465524Spendry 	struct pfsnode *pfs;
67565524Spendry 	int error;
67665524Spendry 	int count;
67765524Spendry 	int i;
67865524Spendry 
67967367Smckusick 	/*
68067367Smckusick 	 * We don't allow exporting procfs mounts, and currently local
68167367Smckusick 	 * requests do not need cookies.
68267367Smckusick 	 */
68367367Smckusick 	if (ap->a_ncookies)
68467367Smckusick 		panic("procfs_readdir: not hungry");
68567367Smckusick 
68665524Spendry 	pfs = VTOPFS(ap->a_vp);
68765524Spendry 
68865524Spendry 	if (uio->uio_resid < UIO_MX)
68965524Spendry 		return (EINVAL);
69065524Spendry 	if (uio->uio_offset & (UIO_MX-1))
69165524Spendry 		return (EINVAL);
69265524Spendry 	if (uio->uio_offset < 0)
69365524Spendry 		return (EINVAL);
69465524Spendry 
69565524Spendry 	error = 0;
69665524Spendry 	count = 0;
69765524Spendry 	i = uio->uio_offset / UIO_MX;
69865524Spendry 
69965524Spendry 	switch (pfs->pfs_type) {
70065524Spendry 	/*
70165524Spendry 	 * this is for the process-specific sub-directories.
70265524Spendry 	 * all that is needed to is copy out all the entries
70365524Spendry 	 * from the procent[] table (top of this file).
70465524Spendry 	 */
70565524Spendry 	case Pproc: {
70667389Spendry 		pid_t pid = pfs->pfs_pid;
70767389Spendry 		struct pfsnames *dt;
70865524Spendry 
70967389Spendry 		for (dt = &procent[i]; i < Nprocent && uio->uio_resid >= UIO_MX;
71067389Spendry 		     dt++, i++) {
71167389Spendry 			struct proc *p = PFIND(pid);
71267389Spendry 
71367389Spendry 			if (p == NULL)
71465524Spendry 				break;
71565524Spendry 
71667389Spendry 			if (dt->d_valid && (*dt->d_valid)(p) == 0)
71767389Spendry 				continue;
71865524Spendry 
71965524Spendry 			dp->d_reclen = UIO_MX;
72067389Spendry 			dp->d_fileno = PROCFS_FILENO(pid, dt->d_pfstype);
72165524Spendry 			dp->d_namlen = dt->d_namlen;
72267389Spendry 			bcopy(dt->d_name, dp->d_name, dt->d_namlen + 1);
72367389Spendry 			dp->d_type = dt->d_type;
72467389Spendry 
72567389Spendry 			if (error = uiomove((caddr_t)dp, UIO_MX, uio))
72665524Spendry 				break;
72765524Spendry 		}
72865524Spendry 
72965524Spendry 	    	break;
73065524Spendry 
73165524Spendry 	    }
73265524Spendry 
73365524Spendry 	/*
73465524Spendry 	 * this is for the root of the procfs filesystem
73565524Spendry 	 * what is needed is a special entry for "curproc"
73665524Spendry 	 * followed by an entry for each process on allproc
73765524Spendry #ifdef PROCFS_ZOMBIE
73865524Spendry 	 * and zombproc.
73965524Spendry #endif
74065524Spendry 	 */
74165524Spendry 
74265524Spendry 	case Proot: {
74365524Spendry #ifdef PROCFS_ZOMBIE
74465524Spendry 		int doingzomb = 0;
74565524Spendry #endif
74667389Spendry 		int pcnt = 0;
747*67721Smckusick 		volatile struct proc *p = allproc.lh_first;
74865524Spendry 
74967389Spendry 	again:
75067389Spendry 		for (; p && uio->uio_resid >= UIO_MX; i++, pcnt++) {
75165524Spendry 			bzero((char *) dp, UIO_MX);
75265524Spendry 			dp->d_reclen = UIO_MX;
75365524Spendry 
75465524Spendry 			switch (i) {
75567389Spendry 			case 0:		/* `.' */
75667389Spendry 			case 1:		/* `..' */
75767389Spendry 				dp->d_fileno = PROCFS_FILENO(0, Proot);
75867389Spendry 				dp->d_namlen = i + 1;
75967389Spendry 				bcopy("..", dp->d_name, dp->d_namlen);
76067389Spendry 				dp->d_name[i + 1] = '\0';
76167389Spendry 				dp->d_type = DT_DIR;
76265524Spendry 				break;
76365524Spendry 
76467389Spendry 			case 2:
76567389Spendry 				dp->d_fileno = PROCFS_FILENO(0, Pcurproc);
76667389Spendry 				dp->d_namlen = 7;
76767389Spendry 				bcopy("curproc", dp->d_name, 8);
76867389Spendry 				dp->d_type = DT_LNK;
76967389Spendry 				break;
77067389Spendry 
77165524Spendry 			default:
77267389Spendry 				while (pcnt < i) {
77367389Spendry 					pcnt++;
774*67721Smckusick 					p = p->p_list.le_next;
77567389Spendry 					if (!p)
77667389Spendry 						goto done;
77765524Spendry 				}
77867389Spendry 				dp->d_fileno = PROCFS_FILENO(p->p_pid, Pproc);
77967389Spendry 				dp->d_namlen = sprintf(dp->d_name, "%ld",
78067389Spendry 				    (long)p->p_pid);
78167389Spendry 				dp->d_type = DT_REG;
782*67721Smckusick 				p = p->p_list.le_next;
78365524Spendry 				break;
78465524Spendry 			}
78567389Spendry 
78667389Spendry 			if (error = uiomove((caddr_t)dp, UIO_MX, uio))
78765524Spendry 				break;
78865524Spendry 		}
78967389Spendry 	done:
79065524Spendry 
79167389Spendry #ifdef PROCFS_ZOMBIE
79267389Spendry 		if (p == 0 && doingzomb == 0) {
79367389Spendry 			doingzomb = 1;
794*67721Smckusick 			p = zombproc.lh_first;
79567389Spendry 			goto again;
79667389Spendry 		}
79767389Spendry #endif
79867389Spendry 
79965524Spendry 		break;
80065524Spendry 
80165524Spendry 	    }
80265524Spendry 
80365524Spendry 	default:
80465524Spendry 		error = ENOTDIR;
80565524Spendry 		break;
80665524Spendry 	}
80765524Spendry 
80865524Spendry 	uio->uio_offset = i * UIO_MX;
80965524Spendry 
81065524Spendry 	return (error);
81165524Spendry }
81265524Spendry 
81365524Spendry /*
81467389Spendry  * readlink reads the link of `curproc'
81567389Spendry  */
81667389Spendry procfs_readlink(ap)
81767389Spendry 	struct vop_readlink_args *ap;
81867389Spendry {
81967389Spendry 	struct uio *uio = ap->a_uio;
82067389Spendry 	char buf[16];		/* should be enough */
82167389Spendry 	int len;
82267389Spendry 
82367389Spendry 	if (VTOPFS(ap->a_vp)->pfs_fileno != PROCFS_FILENO(0, Pcurproc))
82467389Spendry 		return (EINVAL);
82567389Spendry 
82667389Spendry 	len = sprintf(buf, "%ld", (long)curproc->p_pid);
82767389Spendry 
82867389Spendry 	return (uiomove((caddr_t)buf, len, ap->a_uio));
82967389Spendry }
83067389Spendry 
83167389Spendry /*
83265524Spendry  * convert decimal ascii to pid_t
83365524Spendry  */
83465524Spendry static pid_t
83565524Spendry atopid(b, len)
83665524Spendry 	const char *b;
83765524Spendry 	u_int len;
83865524Spendry {
83965524Spendry 	pid_t p = 0;
84065524Spendry 
84165524Spendry 	while (len--) {
84265524Spendry 		char c = *b++;
84365524Spendry 		if (c < '0' || c > '9')
84465524Spendry 			return (NO_PID);
84565524Spendry 		p = 10 * p + (c - '0');
84665524Spendry 		if (p > PID_MAX)
84765524Spendry 			return (NO_PID);
84865524Spendry 	}
84965524Spendry 
85065524Spendry 	return (p);
85165524Spendry }
85265524Spendry 
85365524Spendry /*
85465524Spendry  * procfs vnode operations.
85565524Spendry  */
85665524Spendry int (**procfs_vnodeop_p)();
85765524Spendry struct vnodeopv_entry_desc procfs_vnodeop_entries[] = {
85865524Spendry 	{ &vop_default_desc, vn_default_error },
85965524Spendry 	{ &vop_lookup_desc, procfs_lookup },		/* lookup */
86065524Spendry 	{ &vop_create_desc, procfs_create },		/* create */
86165524Spendry 	{ &vop_mknod_desc, procfs_mknod },		/* mknod */
86265524Spendry 	{ &vop_open_desc, procfs_open },		/* open */
86365524Spendry 	{ &vop_close_desc, procfs_close },		/* close */
86465524Spendry 	{ &vop_access_desc, procfs_access },		/* access */
86565524Spendry 	{ &vop_getattr_desc, procfs_getattr },		/* getattr */
86665524Spendry 	{ &vop_setattr_desc, procfs_setattr },		/* setattr */
86765524Spendry 	{ &vop_read_desc, procfs_read },		/* read */
86865524Spendry 	{ &vop_write_desc, procfs_write },		/* write */
86965524Spendry 	{ &vop_ioctl_desc, procfs_ioctl },		/* ioctl */
87065524Spendry 	{ &vop_select_desc, procfs_select },		/* select */
87165524Spendry 	{ &vop_mmap_desc, procfs_mmap },		/* mmap */
87265524Spendry 	{ &vop_fsync_desc, procfs_fsync },		/* fsync */
87365524Spendry 	{ &vop_seek_desc, procfs_seek },		/* seek */
87465524Spendry 	{ &vop_remove_desc, procfs_remove },		/* remove */
87565524Spendry 	{ &vop_link_desc, procfs_link },		/* link */
87665524Spendry 	{ &vop_rename_desc, procfs_rename },		/* rename */
87765524Spendry 	{ &vop_mkdir_desc, procfs_mkdir },		/* mkdir */
87865524Spendry 	{ &vop_rmdir_desc, procfs_rmdir },		/* rmdir */
87965524Spendry 	{ &vop_symlink_desc, procfs_symlink },		/* symlink */
88065524Spendry 	{ &vop_readdir_desc, procfs_readdir },		/* readdir */
88165524Spendry 	{ &vop_readlink_desc, procfs_readlink },	/* readlink */
88265524Spendry 	{ &vop_abortop_desc, procfs_abortop },		/* abortop */
88365524Spendry 	{ &vop_inactive_desc, procfs_inactive },	/* inactive */
88465524Spendry 	{ &vop_reclaim_desc, procfs_reclaim },		/* reclaim */
88565524Spendry 	{ &vop_lock_desc, procfs_lock },		/* lock */
88665524Spendry 	{ &vop_unlock_desc, procfs_unlock },		/* unlock */
88765524Spendry 	{ &vop_bmap_desc, procfs_bmap },		/* bmap */
88865524Spendry 	{ &vop_strategy_desc, procfs_strategy },	/* strategy */
88965524Spendry 	{ &vop_print_desc, procfs_print },		/* print */
89065524Spendry 	{ &vop_islocked_desc, procfs_islocked },	/* islocked */
89165524Spendry 	{ &vop_pathconf_desc, procfs_pathconf },	/* pathconf */
89265524Spendry 	{ &vop_advlock_desc, procfs_advlock },		/* advlock */
89365524Spendry 	{ &vop_blkatoff_desc, procfs_blkatoff },	/* blkatoff */
89465524Spendry 	{ &vop_valloc_desc, procfs_valloc },		/* valloc */
89565524Spendry 	{ &vop_vfree_desc, procfs_vfree },		/* vfree */
89665524Spendry 	{ &vop_truncate_desc, procfs_truncate },	/* truncate */
89765524Spendry 	{ &vop_update_desc, procfs_update },		/* update */
89865524Spendry 	{ (struct vnodeop_desc*)NULL, (int(*)())NULL }
89965524Spendry };
90065524Spendry struct vnodeopv_desc procfs_vnodeop_opv_desc =
90165524Spendry 	{ &procfs_vnodeop_p, procfs_vnodeop_entries };
902