165524Spendry /*
268232Spendry * Copyright (c) 1993, 1995 Jan-Simon Pendry
368232Spendry * Copyright (c) 1993, 1995
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*69595Smckusick * @(#)procfs_vnops.c 8.18 (Berkeley) 05/21/95
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 */
4669351Smckusick struct proc_target {
4769351Smckusick u_char pt_type;
4869351Smckusick u_char pt_namlen;
4969351Smckusick char *pt_name;
5069351Smckusick pfstype pt_pfstype;
5169351Smckusick int (*pt_valid) __P((struct proc *p));
5269351Smckusick } proc_targets[] = {
5365524Spendry #define N(s) sizeof(s)-1, s
5469351Smckusick /* name type validp */
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 };
6769351Smckusick static int nproc_targets = sizeof(proc_targets) / sizeof(proc_targets[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;
17468232Spendry int *a_runp;
17567367Smckusick } */ *ap;
17665524Spendry {
17765524Spendry
17865524Spendry if (ap->a_vpp != NULL)
17965524Spendry *ap->a_vpp = ap->a_vp;
18065524Spendry if (ap->a_bnp != NULL)
18165524Spendry *ap->a_bnp = ap->a_bn;
18268230Spendry if (ap->a_runp != NULL)
18368230Spendry *ap->a_runp = 0;
18465524Spendry return (0);
18565524Spendry }
18665524Spendry
18765524Spendry /*
188*69595Smckusick * procfs_inactive is called when the pfsnode
18965524Spendry * is vrele'd and the reference count goes
19065524Spendry * to zero. (vp) will be on the vnode free
19165524Spendry * list, so to get it back vget() must be
19265524Spendry * used.
19365524Spendry *
19465524Spendry * for procfs, check if the process is still
19565524Spendry * alive and if it isn't then just throw away
19669362Spendry * the vnode by calling vgone(). this may
19765524Spendry * be overkill and a waste of time since the
19865524Spendry * chances are that the process will still be
19965524Spendry * there and PFIND is not free.
20065524Spendry *
201*69595Smckusick * (vp) is locked on entry, but must be unlocked on exit.
20265524Spendry */
20365524Spendry procfs_inactive(ap)
20467367Smckusick struct vop_inactive_args /* {
20567367Smckusick struct vnode *a_vp;
20667367Smckusick } */ *ap;
20765524Spendry {
208*69595Smckusick struct vnode *vp = ap->a_vp;
209*69595Smckusick struct pfsnode *pfs = VTOPFS(vp);
21065524Spendry
211*69595Smckusick VOP_UNLOCK(vp, 0, ap->a_p);
21265524Spendry if (PFIND(pfs->pfs_pid) == 0)
213*69595Smckusick vgone(vp);
21465524Spendry
21565524Spendry return (0);
21665524Spendry }
21765524Spendry
21865524Spendry /*
21965524Spendry * _reclaim is called when getnewvnode()
22065524Spendry * wants to make use of an entry on the vnode
22165524Spendry * free list. at this time the filesystem needs
22265524Spendry * to free any private data and remove the node
22365524Spendry * from any private lists.
22465524Spendry */
22565524Spendry procfs_reclaim(ap)
22667367Smckusick struct vop_reclaim_args /* {
22767367Smckusick struct vnode *a_vp;
22867367Smckusick } */ *ap;
22965524Spendry {
23065524Spendry
23167389Spendry return (procfs_freevp(ap->a_vp));
23265524Spendry }
23365524Spendry
23465524Spendry /*
23565743Spendry * Return POSIX pathconf information applicable to special devices.
23665743Spendry */
23765743Spendry procfs_pathconf(ap)
23865743Spendry struct vop_pathconf_args /* {
23965743Spendry struct vnode *a_vp;
24065743Spendry int a_name;
24165743Spendry int *a_retval;
24265743Spendry } */ *ap;
24365743Spendry {
24465743Spendry
24565743Spendry switch (ap->a_name) {
24665743Spendry case _PC_LINK_MAX:
24765743Spendry *ap->a_retval = LINK_MAX;
24865743Spendry return (0);
24965743Spendry case _PC_MAX_CANON:
25065743Spendry *ap->a_retval = MAX_CANON;
25165743Spendry return (0);
25265743Spendry case _PC_MAX_INPUT:
25365743Spendry *ap->a_retval = MAX_INPUT;
25465743Spendry return (0);
25565743Spendry case _PC_PIPE_BUF:
25665743Spendry *ap->a_retval = PIPE_BUF;
25765743Spendry return (0);
25865743Spendry case _PC_CHOWN_RESTRICTED:
25965743Spendry *ap->a_retval = 1;
26065743Spendry return (0);
26165743Spendry case _PC_VDISABLE:
26265743Spendry *ap->a_retval = _POSIX_VDISABLE;
26365743Spendry return (0);
26465743Spendry default:
26565743Spendry return (EINVAL);
26665743Spendry }
26765743Spendry /* NOTREACHED */
26865743Spendry }
26965743Spendry
27065743Spendry /*
27165524Spendry * _print is used for debugging.
27265524Spendry * just print a readable description
27365524Spendry * of (vp).
27465524Spendry */
27565524Spendry procfs_print(ap)
27667367Smckusick struct vop_print_args /* {
27767367Smckusick struct vnode *a_vp;
27867367Smckusick } */ *ap;
27965524Spendry {
28065524Spendry struct pfsnode *pfs = VTOPFS(ap->a_vp);
28165524Spendry
28267389Spendry printf("tag VT_PROCFS, type %s, pid %d, mode %x, flags %x\n",
28367389Spendry pfs->pfs_type, pfs->pfs_pid, pfs->pfs_mode, pfs->pfs_flags);
28465524Spendry }
28565524Spendry
28665524Spendry /*
28765524Spendry * _abortop is called when operations such as
28865524Spendry * rename and create fail. this entry is responsible
28965524Spendry * for undoing any side-effects caused by the lookup.
29065524Spendry * this will always include freeing the pathname buffer.
29165524Spendry */
29265524Spendry procfs_abortop(ap)
29367367Smckusick struct vop_abortop_args /* {
29467367Smckusick struct vnode *a_dvp;
29567367Smckusick struct componentname *a_cnp;
29667367Smckusick } */ *ap;
29765524Spendry {
29865524Spendry
29965524Spendry if ((ap->a_cnp->cn_flags & (HASBUF | SAVESTART)) == HASBUF)
30065524Spendry FREE(ap->a_cnp->cn_pnbuf, M_NAMEI);
30165524Spendry return (0);
30265524Spendry }
30365524Spendry
30465524Spendry /*
30565524Spendry * generic entry point for unsupported operations
30665524Spendry */
procfs_badop()30765524Spendry procfs_badop()
30865524Spendry {
30965524Spendry
31065524Spendry return (EIO);
31165524Spendry }
31265524Spendry
31365524Spendry /*
31465524Spendry * Invent attributes for pfsnode (vp) and store
31565524Spendry * them in (vap).
31665524Spendry * Directories lengths are returned as zero since
31765524Spendry * any real length would require the genuine size
31865524Spendry * to be computed, and nothing cares anyway.
31965524Spendry *
32065524Spendry * this is relatively minimal for procfs.
32165524Spendry */
32265524Spendry procfs_getattr(ap)
32367367Smckusick struct vop_getattr_args /* {
32467367Smckusick struct vnode *a_vp;
32567367Smckusick struct vattr *a_vap;
32667367Smckusick struct ucred *a_cred;
32767367Smckusick struct proc *a_p;
32867367Smckusick } */ *ap;
32965524Spendry {
33065524Spendry struct pfsnode *pfs = VTOPFS(ap->a_vp);
33165536Spendry struct vattr *vap = ap->a_vap;
33265524Spendry struct proc *procp;
33368174Scgd struct timeval tv;
33465524Spendry int error;
33565524Spendry
33665524Spendry /* first check the process still exists */
33766025Spendry switch (pfs->pfs_type) {
33866025Spendry case Proot:
33967389Spendry case Pcurproc:
34066025Spendry procp = 0;
34166025Spendry break;
34265524Spendry
34366025Spendry default:
34466025Spendry procp = PFIND(pfs->pfs_pid);
34566025Spendry if (procp == 0)
34666025Spendry return (ENOENT);
34766025Spendry }
34866025Spendry
34965524Spendry error = 0;
35065524Spendry
35165524Spendry /* start by zeroing out the attributes */
35265536Spendry VATTR_NULL(vap);
35365524Spendry
35465524Spendry /* next do all the common fields */
35565536Spendry vap->va_type = ap->a_vp->v_type;
35665536Spendry vap->va_mode = pfs->pfs_mode;
35765536Spendry vap->va_fileid = pfs->pfs_fileno;
35865536Spendry vap->va_flags = 0;
35965536Spendry vap->va_blocksize = PAGE_SIZE;
36065536Spendry vap->va_bytes = vap->va_size = 0;
36165524Spendry
36265524Spendry /*
36367389Spendry * Make all times be current TOD.
36467389Spendry * It would be possible to get the process start
36567389Spendry * time from the p_stat structure, but there's
36667389Spendry * no "file creation" time stamp anyway, and the
36767389Spendry * p_stat structure is not addressible if u. gets
36867389Spendry * swapped out for that process.
36967389Spendry */
37068174Scgd microtime(&tv);
37168174Scgd TIMEVAL_TO_TIMESPEC(&tv, &vap->va_ctime);
37267389Spendry vap->va_atime = vap->va_mtime = vap->va_ctime;
37367389Spendry
37467389Spendry /*
37566025Spendry * If the process has exercised some setuid or setgid
37666025Spendry * privilege, then rip away read/write permission so
37766025Spendry * that only root can gain access.
37866025Spendry */
37966025Spendry switch (pfs->pfs_type) {
38067389Spendry case Pmem:
38166025Spendry case Pregs:
38266025Spendry case Pfpregs:
38366025Spendry if (procp->p_flag & P_SUGID)
38466025Spendry vap->va_mode &= ~((VREAD|VWRITE)|
38566025Spendry ((VREAD|VWRITE)>>3)|
38666025Spendry ((VREAD|VWRITE)>>6));
38767389Spendry case Pctl:
38867389Spendry case Pstatus:
38967389Spendry case Pnote:
39067389Spendry case Pnotepg:
39167389Spendry vap->va_nlink = 1;
39267389Spendry vap->va_uid = procp->p_ucred->cr_uid;
39367389Spendry vap->va_gid = procp->p_ucred->cr_gid;
39466025Spendry break;
39566025Spendry }
39666025Spendry
39766025Spendry /*
39865524Spendry * now do the object specific fields
39965524Spendry *
40065524Spendry * The size could be set from struct reg, but it's hardly
40165524Spendry * worth the trouble, and it puts some (potentially) machine
40265524Spendry * dependent data into this machine-independent code. If it
40365524Spendry * becomes important then this function should break out into
40465524Spendry * a per-file stat function in the corresponding .c file.
40565524Spendry */
40665524Spendry
40765524Spendry switch (pfs->pfs_type) {
40865524Spendry case Proot:
40967389Spendry /*
41067389Spendry * Set nlink to 1 to tell fts(3) we don't actually know.
41167389Spendry */
41267389Spendry vap->va_nlink = 1;
41365536Spendry vap->va_uid = 0;
41465536Spendry vap->va_gid = 0;
41567389Spendry vap->va_size = vap->va_bytes = DEV_BSIZE;
41665524Spendry break;
41765524Spendry
41867389Spendry case Pcurproc: {
41967389Spendry char buf[16]; /* should be enough */
42067389Spendry vap->va_nlink = 1;
42167389Spendry vap->va_uid = 0;
42267389Spendry vap->va_gid = 0;
42367389Spendry vap->va_size = vap->va_bytes =
42467389Spendry sprintf(buf, "%ld", (long)curproc->p_pid);
42567389Spendry break;
42667389Spendry }
42767389Spendry
42865524Spendry case Pproc:
42965536Spendry vap->va_nlink = 2;
43065536Spendry vap->va_uid = procp->p_ucred->cr_uid;
43165536Spendry vap->va_gid = procp->p_ucred->cr_gid;
43267389Spendry vap->va_size = vap->va_bytes = DEV_BSIZE;
43365524Spendry break;
43465524Spendry
43565524Spendry case Pfile:
43665524Spendry error = EOPNOTSUPP;
43765524Spendry break;
43865524Spendry
43965524Spendry case Pmem:
44065536Spendry vap->va_bytes = vap->va_size =
44165524Spendry ctob(procp->p_vmspace->vm_tsize +
44265524Spendry procp->p_vmspace->vm_dsize +
44365524Spendry procp->p_vmspace->vm_ssize);
44465524Spendry break;
44565524Spendry
44665524Spendry case Pregs:
44767389Spendry vap->va_bytes = vap->va_size = sizeof(struct reg);
44867389Spendry break;
44967389Spendry
45065924Spendry case Pfpregs:
45167389Spendry vap->va_bytes = vap->va_size = sizeof(struct fpreg);
45267389Spendry break;
45367389Spendry
45465524Spendry case Pctl:
45565524Spendry case Pstatus:
45665524Spendry case Pnote:
45765524Spendry case Pnotepg:
45865524Spendry break;
45965524Spendry
46065524Spendry default:
46165524Spendry panic("procfs_getattr");
46265524Spendry }
46365524Spendry
46465524Spendry return (error);
46565524Spendry }
46665524Spendry
46765524Spendry procfs_setattr(ap)
46867367Smckusick struct vop_setattr_args /* {
46967367Smckusick struct vnode *a_vp;
47067367Smckusick struct vattr *a_vap;
47167367Smckusick struct ucred *a_cred;
47267367Smckusick struct proc *a_p;
47367367Smckusick } */ *ap;
47465524Spendry {
47565524Spendry /*
47665524Spendry * just fake out attribute setting
47765524Spendry * it's not good to generate an error
47865524Spendry * return, otherwise things like creat()
47965524Spendry * will fail when they try to set the
48065524Spendry * file length to 0. worse, this means
48165524Spendry * that echo $note > /proc/$pid/note will fail.
48265524Spendry */
48365524Spendry
48465524Spendry return (0);
48565524Spendry }
48665524Spendry
48765524Spendry /*
48865524Spendry * implement access checking.
48965524Spendry *
49065524Spendry * something very similar to this code is duplicated
49165524Spendry * throughout the 4bsd kernel and should be moved
49265524Spendry * into kern/vfs_subr.c sometime.
49365524Spendry *
49465524Spendry * actually, the check for super-user is slightly
49565524Spendry * broken since it will allow read access to write-only
49665524Spendry * objects. this doesn't cause any particular trouble
49765524Spendry * but does mean that the i/o entry points need to check
49865524Spendry * that the operation really does make sense.
49965524Spendry */
50065524Spendry procfs_access(ap)
50167367Smckusick struct vop_access_args /* {
50267367Smckusick struct vnode *a_vp;
50367367Smckusick int a_mode;
50467367Smckusick struct ucred *a_cred;
50567367Smckusick struct proc *a_p;
50667367Smckusick } */ *ap;
50765524Spendry {
50865524Spendry struct vattr *vap;
50965524Spendry struct vattr vattr;
51065524Spendry int error;
51165524Spendry
51265524Spendry /*
51365524Spendry * If you're the super-user,
51465524Spendry * you always get access.
51565524Spendry */
51667389Spendry if (ap->a_cred->cr_uid == 0)
51765524Spendry return (0);
51867389Spendry
51965524Spendry vap = &vattr;
52065524Spendry if (error = VOP_GETATTR(ap->a_vp, vap, ap->a_cred, ap->a_p))
52165524Spendry return (error);
52265524Spendry
52365524Spendry /*
52465524Spendry * Access check is based on only one of owner, group, public.
52565524Spendry * If not owner, then check group. If not a member of the
52665524Spendry * group, then check public access.
52765524Spendry */
52865524Spendry if (ap->a_cred->cr_uid != vap->va_uid) {
52965524Spendry gid_t *gp;
53065524Spendry int i;
53165524Spendry
53267389Spendry ap->a_mode >>= 3;
53365524Spendry gp = ap->a_cred->cr_groups;
53465524Spendry for (i = 0; i < ap->a_cred->cr_ngroups; i++, gp++)
53565524Spendry if (vap->va_gid == *gp)
53665524Spendry goto found;
53765524Spendry ap->a_mode >>= 3;
53865524Spendry found:
53965524Spendry ;
54065524Spendry }
54165524Spendry
54265524Spendry if ((vap->va_mode & ap->a_mode) == ap->a_mode)
54365524Spendry return (0);
54465524Spendry
54565524Spendry return (EACCES);
54665524Spendry }
54765524Spendry
54865524Spendry /*
54965524Spendry * lookup. this is incredibly complicated in the
55065524Spendry * general case, however for most pseudo-filesystems
55165524Spendry * very little needs to be done.
55265524Spendry *
55365524Spendry * unless you want to get a migraine, just make sure your
55465524Spendry * filesystem doesn't do any locking of its own. otherwise
55565524Spendry * read and inwardly digest ufs_lookup().
55665524Spendry */
55765524Spendry procfs_lookup(ap)
55867367Smckusick struct vop_lookup_args /* {
55967367Smckusick struct vnode * a_dvp;
56067367Smckusick struct vnode ** a_vpp;
56167367Smckusick struct componentname * a_cnp;
56267367Smckusick } */ *ap;
56365524Spendry {
56465524Spendry struct componentname *cnp = ap->a_cnp;
56565524Spendry struct vnode **vpp = ap->a_vpp;
56665524Spendry struct vnode *dvp = ap->a_dvp;
56765524Spendry char *pname = cnp->cn_nameptr;
56869442Smckusick struct proc *curp = cnp->cn_proc;
56965524Spendry int error = 0;
57069351Smckusick struct proc_target *pt;
57169351Smckusick struct vnode *fvp;
57265524Spendry pid_t pid;
57365524Spendry struct pfsnode *pfs;
57469351Smckusick struct proc *p;
57565524Spendry int i;
57665524Spendry
57769351Smckusick *vpp = NULL;
57869351Smckusick
57969351Smckusick if (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)
58069351Smckusick return (EROFS);
58169351Smckusick
58265524Spendry if (cnp->cn_namelen == 1 && *pname == '.') {
58365524Spendry *vpp = dvp;
58465524Spendry VREF(dvp);
58569442Smckusick /* vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY, curp); */
58665524Spendry return (0);
58765524Spendry }
58865524Spendry
58965524Spendry pfs = VTOPFS(dvp);
59065524Spendry switch (pfs->pfs_type) {
59165524Spendry case Proot:
59265524Spendry if (cnp->cn_flags & ISDOTDOT)
59365524Spendry return (EIO);
59465524Spendry
59565524Spendry if (CNEQ(cnp, "curproc", 7))
59667389Spendry return (procfs_allocvp(dvp->v_mount, vpp, 0, Pcurproc));
59767389Spendry
59867389Spendry pid = atopid(pname, cnp->cn_namelen);
59965524Spendry if (pid == NO_PID)
60069351Smckusick break;
60165524Spendry
60269351Smckusick p = PFIND(pid);
60369351Smckusick if (p == 0)
60469351Smckusick break;
60565524Spendry
60667389Spendry return (procfs_allocvp(dvp->v_mount, vpp, pid, Pproc));
60765524Spendry
60865524Spendry case Pproc:
60969351Smckusick if (cnp->cn_flags & ISDOTDOT)
61069351Smckusick return (procfs_root(dvp->v_mount, vpp));
61165524Spendry
61269351Smckusick p = PFIND(pfs->pfs_pid);
61369351Smckusick if (p == 0)
61469351Smckusick break;
61565524Spendry
61669351Smckusick for (pt = proc_targets, i = 0; i < nproc_targets; pt++, i++) {
61769351Smckusick if (cnp->cn_namelen == pt->pt_namlen &&
61869351Smckusick bcmp(pt->pt_name, pname, cnp->cn_namelen) == 0 &&
61969351Smckusick (pt->pt_valid == NULL || (*pt->pt_valid)(p)))
62065524Spendry goto found;
62165524Spendry }
62269351Smckusick break;
62365524Spendry
62465524Spendry found:
62569351Smckusick if (pt->pt_pfstype == Pfile) {
62669351Smckusick fvp = procfs_findtextvp(p);
62769351Smckusick /* We already checked that it exists. */
62869351Smckusick VREF(fvp);
62969442Smckusick vn_lock(fvp, LK_EXCLUSIVE | LK_RETRY, curp);
63069351Smckusick *vpp = fvp;
63167389Spendry return (0);
63265524Spendry }
63365524Spendry
63467389Spendry return (procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
63569351Smckusick pt->pt_pfstype));
63667389Spendry
63765524Spendry default:
63865524Spendry return (ENOTDIR);
63965524Spendry }
64069351Smckusick
64169351Smckusick return (cnp->cn_nameiop == LOOKUP ? ENOENT : EROFS);
64265524Spendry }
64365524Spendry
64467389Spendry int
procfs_validfile(p)64567389Spendry procfs_validfile(p)
64667389Spendry struct proc *p;
64767389Spendry {
64867389Spendry
64967389Spendry return (procfs_findtextvp(p) != NULLVP);
65067389Spendry }
65167389Spendry
65265524Spendry /*
65365524Spendry * readdir returns directory entries from pfsnode (vp).
65465524Spendry *
65565524Spendry * the strategy here with procfs is to generate a single
65665524Spendry * directory entry at a time (struct pfsdent) and then
65765524Spendry * copy that out to userland using uiomove. a more efficent
65865524Spendry * though more complex implementation, would try to minimize
65965524Spendry * the number of calls to uiomove(). for procfs, this is
66065524Spendry * hardly worth the added code complexity.
66165524Spendry *
66265524Spendry * this should just be done through read()
66365524Spendry */
66465524Spendry procfs_readdir(ap)
66567367Smckusick struct vop_readdir_args /* {
66667367Smckusick struct vnode *a_vp;
66767367Smckusick struct uio *a_uio;
66867367Smckusick struct ucred *a_cred;
66967367Smckusick int *a_eofflag;
67067367Smckusick u_long *a_cookies;
67167367Smckusick int a_ncookies;
67267367Smckusick } */ *ap;
67365524Spendry {
67465524Spendry struct uio *uio = ap->a_uio;
67565524Spendry struct pfsdent d;
67665524Spendry struct pfsdent *dp = &d;
67765524Spendry struct pfsnode *pfs;
67865524Spendry int error;
67965524Spendry int count;
68065524Spendry int i;
68165524Spendry
68267367Smckusick /*
68367367Smckusick * We don't allow exporting procfs mounts, and currently local
68467367Smckusick * requests do not need cookies.
68567367Smckusick */
68667367Smckusick if (ap->a_ncookies)
68767367Smckusick panic("procfs_readdir: not hungry");
68867367Smckusick
68965524Spendry pfs = VTOPFS(ap->a_vp);
69065524Spendry
69165524Spendry if (uio->uio_resid < UIO_MX)
69265524Spendry return (EINVAL);
69365524Spendry if (uio->uio_offset & (UIO_MX-1))
69465524Spendry return (EINVAL);
69565524Spendry if (uio->uio_offset < 0)
69665524Spendry return (EINVAL);
69765524Spendry
69865524Spendry error = 0;
69965524Spendry count = 0;
70065524Spendry i = uio->uio_offset / UIO_MX;
70165524Spendry
70265524Spendry switch (pfs->pfs_type) {
70365524Spendry /*
70465524Spendry * this is for the process-specific sub-directories.
70565524Spendry * all that is needed to is copy out all the entries
70665524Spendry * from the procent[] table (top of this file).
70765524Spendry */
70865524Spendry case Pproc: {
70969351Smckusick struct proc *p;
71069351Smckusick struct proc_target *pt;
71165524Spendry
71269351Smckusick p = PFIND(pfs->pfs_pid);
71369351Smckusick if (p == NULL)
71469351Smckusick break;
71567389Spendry
71669351Smckusick for (pt = &proc_targets[i];
71769351Smckusick uio->uio_resid >= UIO_MX && i < nproc_targets; pt++, i++) {
71869351Smckusick if (pt->pt_valid && (*pt->pt_valid)(p) == 0)
71967389Spendry continue;
72065524Spendry
72165524Spendry dp->d_reclen = UIO_MX;
72269351Smckusick dp->d_fileno = PROCFS_FILENO(pfs->pfs_pid, pt->pt_pfstype);
72369351Smckusick dp->d_namlen = pt->pt_namlen;
72469351Smckusick bcopy(pt->pt_name, dp->d_name, pt->pt_namlen + 1);
72569351Smckusick dp->d_type = pt->pt_type;
72667389Spendry
72767389Spendry if (error = uiomove((caddr_t)dp, UIO_MX, uio))
72865524Spendry break;
72965524Spendry }
73065524Spendry
73165524Spendry break;
73265524Spendry }
73365524Spendry
73465524Spendry /*
73565524Spendry * this is for the root of the procfs filesystem
73665524Spendry * what is needed is a special entry for "curproc"
73765524Spendry * followed by an entry for each process on allproc
73865524Spendry #ifdef PROCFS_ZOMBIE
73965524Spendry * and zombproc.
74065524Spendry #endif
74165524Spendry */
74265524Spendry
74365524Spendry case Proot: {
74465524Spendry #ifdef PROCFS_ZOMBIE
74565524Spendry int doingzomb = 0;
74665524Spendry #endif
74767389Spendry int pcnt = 0;
74867721Smckusick volatile struct proc *p = allproc.lh_first;
74965524Spendry
75067389Spendry again:
75167389Spendry for (; p && uio->uio_resid >= UIO_MX; i++, pcnt++) {
75265524Spendry bzero((char *) dp, UIO_MX);
75365524Spendry dp->d_reclen = UIO_MX;
75465524Spendry
75565524Spendry switch (i) {
75667389Spendry case 0: /* `.' */
75767389Spendry case 1: /* `..' */
75867389Spendry dp->d_fileno = PROCFS_FILENO(0, Proot);
75967389Spendry dp->d_namlen = i + 1;
76067389Spendry bcopy("..", dp->d_name, dp->d_namlen);
76167389Spendry dp->d_name[i + 1] = '\0';
76267389Spendry dp->d_type = DT_DIR;
76365524Spendry break;
76465524Spendry
76567389Spendry case 2:
76667389Spendry dp->d_fileno = PROCFS_FILENO(0, Pcurproc);
76767389Spendry dp->d_namlen = 7;
76867389Spendry bcopy("curproc", dp->d_name, 8);
76967389Spendry dp->d_type = DT_LNK;
77067389Spendry break;
77167389Spendry
77265524Spendry default:
77367389Spendry while (pcnt < i) {
77467389Spendry pcnt++;
77567721Smckusick p = p->p_list.le_next;
77667389Spendry if (!p)
77767389Spendry goto done;
77865524Spendry }
77967389Spendry dp->d_fileno = PROCFS_FILENO(p->p_pid, Pproc);
78067389Spendry dp->d_namlen = sprintf(dp->d_name, "%ld",
78167389Spendry (long)p->p_pid);
78267389Spendry dp->d_type = DT_REG;
78367721Smckusick p = p->p_list.le_next;
78465524Spendry break;
78565524Spendry }
78667389Spendry
78767389Spendry if (error = uiomove((caddr_t)dp, UIO_MX, uio))
78865524Spendry break;
78965524Spendry }
79067389Spendry done:
79165524Spendry
79267389Spendry #ifdef PROCFS_ZOMBIE
79367389Spendry if (p == 0 && doingzomb == 0) {
79467389Spendry doingzomb = 1;
79567721Smckusick p = zombproc.lh_first;
79667389Spendry goto again;
79767389Spendry }
79867389Spendry #endif
79967389Spendry
80065524Spendry break;
80165524Spendry
80265524Spendry }
80365524Spendry
80465524Spendry default:
80565524Spendry error = ENOTDIR;
80665524Spendry break;
80765524Spendry }
80865524Spendry
80965524Spendry uio->uio_offset = i * UIO_MX;
81065524Spendry
81165524Spendry return (error);
81265524Spendry }
81365524Spendry
81465524Spendry /*
81567389Spendry * readlink reads the link of `curproc'
81667389Spendry */
81767389Spendry procfs_readlink(ap)
81867389Spendry struct vop_readlink_args *ap;
81967389Spendry {
82067389Spendry struct uio *uio = ap->a_uio;
82167389Spendry char buf[16]; /* should be enough */
82267389Spendry int len;
82367389Spendry
82467389Spendry if (VTOPFS(ap->a_vp)->pfs_fileno != PROCFS_FILENO(0, Pcurproc))
82567389Spendry return (EINVAL);
82667389Spendry
82767389Spendry len = sprintf(buf, "%ld", (long)curproc->p_pid);
82867389Spendry
82967389Spendry return (uiomove((caddr_t)buf, len, ap->a_uio));
83067389Spendry }
83167389Spendry
83267389Spendry /*
83365524Spendry * convert decimal ascii to pid_t
83465524Spendry */
83565524Spendry static pid_t
atopid(b,len)83665524Spendry atopid(b, len)
83765524Spendry const char *b;
83865524Spendry u_int len;
83965524Spendry {
84065524Spendry pid_t p = 0;
84165524Spendry
84265524Spendry while (len--) {
84365524Spendry char c = *b++;
84465524Spendry if (c < '0' || c > '9')
84565524Spendry return (NO_PID);
84665524Spendry p = 10 * p + (c - '0');
84765524Spendry if (p > PID_MAX)
84865524Spendry return (NO_PID);
84965524Spendry }
85065524Spendry
85165524Spendry return (p);
85265524Spendry }
85365524Spendry
85465524Spendry /*
85565524Spendry * procfs vnode operations.
85665524Spendry */
85765524Spendry int (**procfs_vnodeop_p)();
85865524Spendry struct vnodeopv_entry_desc procfs_vnodeop_entries[] = {
85965524Spendry { &vop_default_desc, vn_default_error },
86065524Spendry { &vop_lookup_desc, procfs_lookup }, /* lookup */
86165524Spendry { &vop_create_desc, procfs_create }, /* create */
86265524Spendry { &vop_mknod_desc, procfs_mknod }, /* mknod */
86365524Spendry { &vop_open_desc, procfs_open }, /* open */
86465524Spendry { &vop_close_desc, procfs_close }, /* close */
86565524Spendry { &vop_access_desc, procfs_access }, /* access */
86665524Spendry { &vop_getattr_desc, procfs_getattr }, /* getattr */
86765524Spendry { &vop_setattr_desc, procfs_setattr }, /* setattr */
86865524Spendry { &vop_read_desc, procfs_read }, /* read */
86965524Spendry { &vop_write_desc, procfs_write }, /* write */
87065524Spendry { &vop_ioctl_desc, procfs_ioctl }, /* ioctl */
87165524Spendry { &vop_select_desc, procfs_select }, /* select */
87265524Spendry { &vop_mmap_desc, procfs_mmap }, /* mmap */
87368732Smckusick { &vop_revoke_desc, procfs_revoke }, /* revoke */
87465524Spendry { &vop_fsync_desc, procfs_fsync }, /* fsync */
87565524Spendry { &vop_seek_desc, procfs_seek }, /* seek */
87665524Spendry { &vop_remove_desc, procfs_remove }, /* remove */
87765524Spendry { &vop_link_desc, procfs_link }, /* link */
87865524Spendry { &vop_rename_desc, procfs_rename }, /* rename */
87965524Spendry { &vop_mkdir_desc, procfs_mkdir }, /* mkdir */
88065524Spendry { &vop_rmdir_desc, procfs_rmdir }, /* rmdir */
88165524Spendry { &vop_symlink_desc, procfs_symlink }, /* symlink */
88265524Spendry { &vop_readdir_desc, procfs_readdir }, /* readdir */
88365524Spendry { &vop_readlink_desc, procfs_readlink }, /* readlink */
88465524Spendry { &vop_abortop_desc, procfs_abortop }, /* abortop */
88565524Spendry { &vop_inactive_desc, procfs_inactive }, /* inactive */
88665524Spendry { &vop_reclaim_desc, procfs_reclaim }, /* reclaim */
88765524Spendry { &vop_lock_desc, procfs_lock }, /* lock */
88865524Spendry { &vop_unlock_desc, procfs_unlock }, /* unlock */
88965524Spendry { &vop_bmap_desc, procfs_bmap }, /* bmap */
89065524Spendry { &vop_strategy_desc, procfs_strategy }, /* strategy */
89165524Spendry { &vop_print_desc, procfs_print }, /* print */
89265524Spendry { &vop_islocked_desc, procfs_islocked }, /* islocked */
89365524Spendry { &vop_pathconf_desc, procfs_pathconf }, /* pathconf */
89465524Spendry { &vop_advlock_desc, procfs_advlock }, /* advlock */
89565524Spendry { &vop_blkatoff_desc, procfs_blkatoff }, /* blkatoff */
89665524Spendry { &vop_valloc_desc, procfs_valloc }, /* valloc */
89765524Spendry { &vop_vfree_desc, procfs_vfree }, /* vfree */
89865524Spendry { &vop_truncate_desc, procfs_truncate }, /* truncate */
89965524Spendry { &vop_update_desc, procfs_update }, /* update */
90065524Spendry { (struct vnodeop_desc*)NULL, (int(*)())NULL }
90165524Spendry };
90265524Spendry struct vnodeopv_desc procfs_vnodeop_opv_desc =
90365524Spendry { &procfs_vnodeop_p, procfs_vnodeop_entries };
904