/*
* 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.
*/
static int
linprocfs_open(struct vop_open_args *ap)
{
struct pfsnode *pfs = VTOPFS(ap->a_vp);
struct proc *p2;
p2 = PFIND(pfs->pfs_pid);
if (p2 == NULL)
return (ENOENT);
if (pfs->pfs_pid && !PRISON_CHECK(ap->a_cred, p2->p_ucred))
return (ENOENT);
switch (pfs->pfs_type) {
case Pmem:
if (((pfs->pfs_flags & FWRITE) && (ap->a_mode & O_EXCL)) ||
((pfs->pfs_flags & O_EXCL) && (ap->a_mode & FWRITE)))
return (EBUSY);
if (p_trespass(ap->a_cred, p2->p_ucred))
return (EPERM);
if (ap->a_mode & FWRITE)
pfs->pfs_flags = ap->a_mode & (FWRITE|O_EXCL);
break;
default:
break;
}
return (vop_stdopen(ap));
}
/*
* 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).
*/
static int
linprocfs_close(struct vop_close_args *ap)
{
struct pfsnode *pfs = VTOPFS(ap->a_vp);
struct proc *p;
switch (pfs->pfs_type) {
case Pmem:
if ((ap->a_fflag & FWRITE) && (pfs->pfs_flags & O_EXCL))
pfs->pfs_flags &= ~(FWRITE|O_EXCL);
/*
* If this is the last close, then it checks to see if
* the target process has PF_LINGER set in p_pfsflags,
* if this is *not* the case, then the process' stop flags
* are cleared, and the process is woken up. This is
* to help prevent the case where a process has been
* told to stop on an event, but then the requesting process
* has gone away or forgotten about it.
*/
p = NULL;
if ((ap->a_vp->v_opencount < 2)
&& (p = pfind(pfs->pfs_pid))
&& !(p->p_pfsflags & PF_LINGER)) {
p->p_stops = 0;
p->p_step = 0;
wakeup(&p->p_step);
}
if (p)
PRELE(p);
break;
default:
break;
}
return (vop_stdclose(ap));
}
/*
* 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.
*/
int
procfs_open(void *v)
{
struct vop_open_args *ap = v;
struct pfsnode *pfs = VTOPFS(ap->a_vp);
struct proc *p1 = ap->a_p; /* tracer */
struct proc *p2; /* traced */
int error;
if ((p2 = pfind(pfs->pfs_pid)) == 0)
return (ENOENT); /* was ESRCH, jsp */
switch (pfs->pfs_type) {
case Pmem:
if (((pfs->pfs_flags & FWRITE) && (ap->a_mode & O_EXCL)) ||
((pfs->pfs_flags & O_EXCL) && (ap->a_mode & FWRITE)))
return (EBUSY);
if ((error = process_checkioperm(p1, p2)) != 0)
return (error);
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).
*/
int
procfs_close(void *v)
{
struct vop_close_args *ap = v;
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;
case Pctl:
case Pstatus:
case Pnotepg:
case Pnote:
case Proot:
case Pcurproc:
case Pself:
case Pproc:
case Pfile:
case Pregs:
case Pfpregs:
case Pcmdline:
case Pmeminfo:
case Pcpuinfo:
break;
}
return (0);
}
/*
* procfs_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.
*
* (vp) is locked on entry, but must be unlocked on exit.
*
* procfs_inactive(struct vnode *a_vp)
*/
static int
procfs_inactive(struct vop_inactive_args *ap)
{
struct pfsnode *pfs = VTOPFS(ap->a_vp);
if (pfs->pfs_pid & PFS_DEAD)
vrecycle(ap->a_vp);
return (0);
}
/*
* _print is used for debugging.
* just print a readable description
* of (vp).
*/
static int
linprocfs_print(struct vop_print_args *ap)
{
struct pfsnode *pfs = VTOPFS(ap->a_vp);
kprintf("tag VT_PROCFS, type %d, pid %ld, mode %x, flags %lx\n",
pfs->pfs_type, (long)pfs->pfs_pid, pfs->pfs_mode, pfs->pfs_flags);
return (0);
}
/*
* _print is used for debugging.
* just print a readable description
* of (vp).
*/
int
procfs_print(void *v)
{
struct vop_print_args *ap = v;
struct pfsnode *pfs = VTOPFS(ap->a_vp);
printf("tag VT_PROCFS, type %d, pid %d, mode %x, flags %lx\n",
pfs->pfs_type, pfs->pfs_pid, pfs->pfs_mode, pfs->pfs_flags);
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.
*/
int
procfs_inactive(void *v)
{
struct vop_inactive_args *ap = v;
struct vnode *vp = ap->a_vp;
struct pfsnode *pfs = VTOPFS(vp);
if (pfind(pfs->pfs_pid) == NULL && !(vp->v_flag & VXLOCK))
vgone(vp);
return (0);
}
int
procfs_machdep_getattr(struct vnode *vp, struct vattr *vap, struct proc *procp)
{
struct pfsnode *pfs = VTOPFS(vp);
switch (pfs->pfs_type) {
case Pmachdep_xmmregs:
vap->va_bytes = vap->va_size = sizeof(struct xmmregs);
break;
default:
KASSERT(false);
}
return 0;
}
/*
* readlink reads the link of `curproc' or `file'
*/
static int
procfs_readlink(struct vop_readlink_args *ap)
{
char buf[16]; /* should be enough */
struct proc *procp;
struct vnode *vp = ap->a_vp;
struct pfsnode *pfs = VTOPFS(vp);
char *fullpath, *freepath;
int error, len;
switch (pfs->pfs_type) {
case Pcurproc:
if (pfs->pfs_fileno != PROCFS_FILENO(0, Pcurproc))
return (EINVAL);
len = ksnprintf(buf, sizeof(buf), "%ld", (long)curproc->p_pid);
return (uiomove(buf, len, ap->a_uio));
/*
* There _should_ be no way for an entire process to disappear
* from under us...
*/
case Pfile:
procp = pfs_pfind(pfs->pfs_pid);
if (procp == NULL || procp->p_ucred == NULL) {
kprintf("procfs_readlink: pid %d disappeared\n",
pfs->pfs_pid);
if (procp)
PRELE(procp);
return (uiomove("unknown", sizeof("unknown") - 1,
ap->a_uio));
}
error = cache_fullpath(procp, &procp->p_textnch, &fullpath, &freepath, 0);
if (error != 0) {
if (procp)
PRELE(procp);
return (uiomove("unknown", sizeof("unknown") - 1,
ap->a_uio));
}
error = uiomove(fullpath, strlen(fullpath), ap->a_uio);
kfree(freepath, M_TEMP);
if (procp)
PRELE(procp);
return (error);
default:
return (EINVAL);
}
}
/*
* 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(struct vnode *a_vp, int a_mode, struct ucred *a_cred,
* struct file *a_fp)
*/
static int
procfs_open(struct vop_open_args *ap)
{
struct pfsnode *pfs = VTOPFS(ap->a_vp);
struct proc *p1, *p2;
int error;
p2 = pfs_pfind(pfs->pfs_pid);
if (p2 == NULL)
return (ENOENT);
if (pfs->pfs_pid && !PRISON_CHECK(ap->a_cred, p2->p_ucred)) {
error = ENOENT;
goto done;
}
switch (pfs->pfs_type) {
case Pmem:
if (((pfs->pfs_flags & FWRITE) && (ap->a_mode & O_EXCL)) ||
((pfs->pfs_flags & O_EXCL) && (ap->a_mode & FWRITE))) {
error = EBUSY;
goto done;
}
p1 = curproc;
KKASSERT(p1);
/* Can't trace a process that's currently exec'ing. */
if ((p2->p_flags & P_INEXEC) != 0) {
error = EAGAIN;
goto done;
}
if (!CHECKIO(p1, p2) || p_trespass(ap->a_cred, p2->p_ucred)) {
error = EPERM;
goto done;
}
if (ap->a_mode & FWRITE)
pfs->pfs_flags = ap->a_mode & (FWRITE|O_EXCL);
break;
default:
break;
}
error = vop_stdopen(ap);
done:
pfs_pdone(p2);
return error;
}
static int
procfs_readdir_proc(struct vop_readdir_args *ap)
{
struct pfsnode *pfs;
int error, i, retval;
struct proc *p;
struct lwp *lp;
struct proc_target *pt;
struct uio *uio = ap->a_uio;
pfs = VTOPFS(ap->a_vp);
p = pfs_pfind(pfs->pfs_pid);
if (p == NULL)
return(0);
if (!PRISON_CHECK(ap->a_cred, p->p_ucred)) {
error = 0;
goto done;
}
/* XXX lwp, not MPSAFE */
lp = FIRST_LWP_IN_PROC(p);
error = 0;
i = (int)uio->uio_offset;
if (i < 0) {
error = EINVAL;
goto done;
}
for (pt = &proc_targets[i];
!error && uio->uio_resid > 0 && i < nproc_targets; pt++, i++) {
if (pt->pt_valid && (*pt->pt_valid)(lp) == 0)
continue;
retval = vop_write_dirent(&error, uio,
PROCFS_FILENO(pfs->pfs_pid, pt->pt_pfstype), pt->pt_type,
pt->pt_namlen, pt->pt_name);
if (retval)
break;
}
uio->uio_offset = (off_t)i;
error = 0;
done:
PRELE(p);
return error;
}
/*
* readdir() returns directory entries from pfsnode (vp).
*
* We generate just one directory entry at a time, as it would probably
* not pay off to buffer several entries locally to save uiomove calls.
*
* procfs_readdir(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred,
* int *a_eofflag, int *a_ncookies, off_t **a_cookies)
*/
static int
procfs_readdir(struct vop_readdir_args *ap)
{
struct pfsnode *pfs;
int error;
if (ap->a_uio->uio_offset < 0 || ap->a_uio->uio_offset > INT_MAX)
return (EINVAL);
if ((error = vn_lock(ap->a_vp, LK_EXCLUSIVE | LK_RETRY)) != 0)
return (error);
pfs = VTOPFS(ap->a_vp);
switch (pfs->pfs_type) {
case Pproc:
/*
* 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).
*/
error = procfs_readdir_proc(ap);
break;
case Proot:
/*
* 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
*/
error = procfs_readdir_root(ap);
break;
default:
error = ENOTDIR;
break;
}
vn_unlock(ap->a_vp);
return (error);
}
/*
* lookup. this is incredibly complicated in the general case, however
* for most pseudo-filesystems very little needs to be done.
*
* procfs_lookup(struct vnode *a_dvp, struct vnode **a_vpp,
* struct componentname *a_cnp)
*/
static int
procfs_lookup(struct vop_old_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;
/* struct proc *curp = cnp->cn_proc; */
struct proc_target *pt;
pid_t pid;
struct pfsnode *pfs;
struct proc *p;
struct lwp *lp;
int i;
int error;
*vpp = NULL;
if (cnp->cn_nameiop == NAMEI_DELETE || cnp->cn_nameiop == NAMEI_RENAME)
return (EROFS);
p = NULL;
error = 0;
if (cnp->cn_namelen == 1 && *pname == '.') {
*vpp = dvp;
vref(*vpp);
goto out;
}
pfs = VTOPFS(dvp);
switch (pfs->pfs_type) {
case Proot:
if (cnp->cn_flags & CNP_ISDOTDOT)
return (EIO);
if (CNEQ(cnp, "curproc", 7)) {
error = procfs_allocvp(dvp->v_mount, vpp, 0, Pcurproc);
goto out;
}
pid = atopid(pname, cnp->cn_namelen);
if (pid == NO_PID)
break;
p = pfs_pfind(pid);
if (p == NULL)
break;
if (!PRISON_CHECK(ap->a_cnp->cn_cred, p->p_ucred))
break;
if (ps_showallprocs == 0 && ap->a_cnp->cn_cred->cr_uid != 0 &&
ap->a_cnp->cn_cred->cr_uid != p->p_ucred->cr_uid)
break;
error = procfs_allocvp(dvp->v_mount, vpp, pid, Pproc);
goto out;
case Pproc:
if (cnp->cn_flags & CNP_ISDOTDOT) {
error = procfs_root(dvp->v_mount, vpp);
goto out;
}
p = pfs_pfind(pfs->pfs_pid);
if (p == NULL)
break;
/* XXX lwp */
lp = FIRST_LWP_IN_PROC(p);
if (!PRISON_CHECK(ap->a_cnp->cn_cred, p->p_ucred))
break;
if (ps_showallprocs == 0 && ap->a_cnp->cn_cred->cr_uid != 0 &&
ap->a_cnp->cn_cred->cr_uid != p->p_ucred->cr_uid)
break;
for (pt = proc_targets, i = 0; i < nproc_targets; pt++, i++) {
if (cnp->cn_namelen == pt->pt_namlen &&
bcmp(pt->pt_name, pname, cnp->cn_namelen) == 0 &&
(pt->pt_valid == NULL || (*pt->pt_valid)(lp)))
goto found;
}
break;
found:
error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
pt->pt_pfstype);
goto out;
default:
error = ENOTDIR;
goto out;
}
if (cnp->cn_nameiop == NAMEI_LOOKUP)
error = ENOENT;
else
error = EROFS;
/*
* If no error occured *vpp will hold a referenced locked vnode.
* dvp was passed to us locked and *vpp must be returned locked.
* If *vpp != dvp then we should unlock dvp if (1) this is not the
* last component or (2) CNP_LOCKPARENT is not set.
*/
out:
if (error == 0 && *vpp != dvp) {
if ((cnp->cn_flags & CNP_LOCKPARENT) == 0) {
cnp->cn_flags |= CNP_PDIRUNLOCK;
vn_unlock(dvp);
}
}
if (p)
PRELE(p);
return (error);
}
/*
* do an ioctl operation on a pfsnode (vp).
* (vp) is not locked on entry or exit.
*/
static int
linprocfs_ioctl(struct vop_ioctl_args *ap)
{
struct pfsnode *pfs = VTOPFS(ap->a_vp);
struct proc *procp;
int error;
int signo;
struct procfs_status *psp;
unsigned char flags;
procp = pfind(pfs->pfs_pid);
if (procp == NULL) {
return ENOTTY;
}
if (p_trespass(ap->a_cred, procp->p_ucred)) {
error = EPERM;
goto done;
}
switch (ap->a_command) {
case PIOCBIS:
procp->p_stops |= *(unsigned int*)ap->a_data;
break;
case PIOCBIC:
procp->p_stops &= ~*(unsigned int*)ap->a_data;
break;
case PIOCSFL:
/*
* NFLAGS is "non-suser_xxx flags" -- currently, only
* PFS_ISUGID ("ignore set u/g id");
*/
#define NFLAGS (PF_ISUGID)
flags = (unsigned char)*(unsigned int*)ap->a_data;
if (flags & NFLAGS && (error = priv_check_cred(ap->a_cred, PRIV_ROOT, 0)))
goto done;
procp->p_pfsflags = flags;
break;
case PIOCGFL:
*(unsigned int*)ap->a_data = (unsigned int)procp->p_pfsflags;
case PIOCSTATUS:
psp = (struct procfs_status *)ap->a_data;
psp->state = (procp->p_step == 0);
psp->flags = procp->p_pfsflags;
psp->events = procp->p_stops;
if (procp->p_step) {
psp->why = procp->p_stype;
psp->val = procp->p_xstat;
} else {
psp->why = psp->val = 0; /* Not defined values */
}
break;
case PIOCWAIT:
psp = (struct procfs_status *)ap->a_data;
if (procp->p_step == 0) {
error = tsleep(&procp->p_stype, PCATCH, "piocwait", 0);
if (error)
goto done;
}
psp->state = 1; /* It stopped */
psp->flags = procp->p_pfsflags;
psp->events = procp->p_stops;
psp->why = procp->p_stype; /* why it stopped */
psp->val = procp->p_xstat; /* any extra info */
break;
case PIOCCONT: /* Restart a proc */
if (procp->p_step == 0) {
error = EINVAL; /* Can only start a stopped process */
goto done;
}
if ((signo = *(int*)ap->a_data) != 0) {
if (signo >= NSIG || signo <= 0) {
error = EINVAL;
goto done;
}
ksignal(procp, signo);
}
procp->p_step = 0;
wakeup(&procp->p_step);
break;
default:
error = ENOTTY;
goto done;
}
error = 0;
done:
if (procp)
PRELE(procp);
return error;
}
/*
* readdir returns directory entries from pfsnode (vp).
*
* the strategy here with procfs is to generate a single
* directory entry at a time (struct dirent) 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()
*/
int
procfs_readdir(void *v)
{
struct vop_readdir_args *ap = v;
struct uio *uio = ap->a_uio;
struct dirent d;
struct pfsnode *pfs;
struct vnode *vp;
int i;
int error;
vp = ap->a_vp;
pfs = VTOPFS(vp);
if (uio->uio_resid < UIO_MX)
return (EINVAL);
error = 0;
i = uio->uio_offset;
if (i < 0)
return (EINVAL);
bzero(&d, UIO_MX);
d.d_reclen = 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: {
struct proc *p;
const struct proc_target *pt;
p = pfind(pfs->pfs_pid);
if (p == NULL)
break;
for (pt = &proc_targets[i];
uio->uio_resid >= UIO_MX && i < nproc_targets; pt++, i++) {
if (pt->pt_valid &&
(*pt->pt_valid)(p, vp->v_mount) == 0)
continue;
d.d_fileno = PROCFS_FILENO(pfs->pfs_pid, pt->pt_pfstype);
d.d_namlen = pt->pt_namlen;
bcopy(pt->pt_name, d.d_name, pt->pt_namlen + 1);
d.d_type = pt->pt_type;
if ((error = uiomove(&d, UIO_MX, uio)) != 0)
break;
}
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: {
#ifdef PROCFS_ZOMBIE
int doingzomb = 0;
#endif
int pcnt = i;
volatile struct proc *p = LIST_FIRST(&allproc);
if (pcnt > 3)
pcnt = 3;
#ifdef PROCFS_ZOMBIE
again:
#endif
for (; p && uio->uio_resid >= UIO_MX; i++, pcnt++) {
switch (i) {
case 0: /* `.' */
case 1: /* `..' */
d.d_fileno = PROCFS_FILENO(0, Proot);
d.d_namlen = i + 1;
bcopy("..", d.d_name, d.d_namlen);
d.d_name[i + 1] = '\0';
d.d_type = DT_DIR;
break;
case 2:
d.d_fileno = PROCFS_FILENO(0, Pcurproc);
d.d_namlen = 7;
bcopy("curproc", d.d_name, 8);
d.d_type = DT_LNK;
break;
case 3:
d.d_fileno = PROCFS_FILENO(0, Pself);
d.d_namlen = 4;
bcopy("self", d.d_name, 5);
d.d_type = DT_LNK;
break;
case 4:
if (VFSTOPROC(vp->v_mount)->pmnt_flags &
PROCFSMNT_LINUXCOMPAT) {
d.d_fileno = PROCFS_FILENO(0, Pcpuinfo);
d.d_namlen = 7;
bcopy("cpuinfo", d.d_name, 8);
d.d_type = DT_REG;
break;
}
/* fall through */
case 5:
if (VFSTOPROC(vp->v_mount)->pmnt_flags &
PROCFSMNT_LINUXCOMPAT) {
d.d_fileno = PROCFS_FILENO(0, Pmeminfo);
d.d_namlen = 7;
bcopy("meminfo", d.d_name, 8);
d.d_type = DT_REG;
break;
}
/* fall through */
default:
while (pcnt < i) {
pcnt++;
p = LIST_NEXT(p, p_list);
if (!p)
goto done;
}
d.d_fileno = PROCFS_FILENO(p->p_pid, Pproc);
d.d_namlen = snprintf(d.d_name, sizeof(d.d_name),
"%ld", (long)p->p_pid);
d.d_type = DT_REG;
p = LIST_NEXT(p, p_list);
break;
}
if ((error = uiomove(&d, UIO_MX, uio)) != 0)
break;
}
done:
#ifdef PROCFS_ZOMBIE
if (p == 0 && doingzomb == 0) {
doingzomb = 1;
p = LIST_FIRST(&zombproc);
goto again;
}
#endif
break;
}
default:
error = ENOTDIR;
break;
}
uio->uio_offset = i;
return (error);
}
/*
* 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().
*/
int
procfs_lookup(void *v)
{
struct vop_lookup_args *ap = v;
struct componentname *cnp = ap->a_cnp;
struct vnode **vpp = ap->a_vpp;
struct vnode *dvp = ap->a_dvp;
char *pname = cnp->cn_nameptr;
struct proc *curp = curproc;
const struct proc_target *pt;
struct vnode *fvp;
pid_t pid;
struct pfsnode *pfs;
struct proc *p = NULL;
int i, error, wantpunlock, iscurproc = 0, isself = 0;
*vpp = NULL;
cnp->cn_flags &= ~PDIRUNLOCK;
if (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)
return (EROFS);
if (cnp->cn_namelen == 1 && *pname == '.') {
*vpp = dvp;
VREF(dvp);
return (0);
}
wantpunlock = (~cnp->cn_flags & (LOCKPARENT | ISLASTCN));
pfs = VTOPFS(dvp);
switch (pfs->pfs_type) {
case Proot:
if (cnp->cn_flags & ISDOTDOT)
return (EIO);
iscurproc = CNEQ(cnp, "curproc", 7);
isself = CNEQ(cnp, "self", 4);
if (iscurproc || isself) {
error = procfs_allocvp(dvp->v_mount, vpp, 0,
iscurproc ? Pcurproc : Pself);
if ((error == 0) && (wantpunlock)) {
VOP_UNLOCK(dvp, 0, curp);
cnp->cn_flags |= PDIRUNLOCK;
}
return (error);
}
for (i = 0; i < nproc_root_targets; i++) {
pt = &proc_root_targets[i];
if (cnp->cn_namelen == pt->pt_namlen &&
memcmp(pt->pt_name, pname, cnp->cn_namelen) == 0 &&
(pt->pt_valid == NULL ||
(*pt->pt_valid)(p, dvp->v_mount)))
break;
}
if (i != nproc_root_targets) {
error = procfs_allocvp(dvp->v_mount, vpp, 0,
pt->pt_pfstype);
if ((error == 0) && (wantpunlock)) {
VOP_UNLOCK(dvp, 0, curp);
cnp->cn_flags |= PDIRUNLOCK;
}
return (error);
}
pid = atopid(pname, cnp->cn_namelen);
if (pid == NO_PID)
break;
p = pfind(pid);
if (p == 0)
break;
error = procfs_allocvp(dvp->v_mount, vpp, pid, Pproc);
if ((error == 0) && wantpunlock) {
VOP_UNLOCK(dvp, 0, curp);
cnp->cn_flags |= PDIRUNLOCK;
}
return (error);
case Pproc:
/*
* do the .. dance. We unlock the directory, and then
* get the root dir. That will automatically return ..
* locked. Then if the caller wanted dvp locked, we
* re-lock.
*/
if (cnp->cn_flags & ISDOTDOT) {
VOP_UNLOCK(dvp, 0, p);
cnp->cn_flags |= PDIRUNLOCK;
error = procfs_root(dvp->v_mount, vpp);
if ((error == 0) && (wantpunlock == 0) &&
((error = vn_lock(dvp, LK_EXCLUSIVE, curp)) == 0))
cnp->cn_flags &= ~PDIRUNLOCK;
return (error);
}
p = pfind(pfs->pfs_pid);
if (p == 0)
break;
for (pt = proc_targets, i = 0; i < nproc_targets; pt++, i++) {
if (cnp->cn_namelen == pt->pt_namlen &&
bcmp(pt->pt_name, pname, cnp->cn_namelen) == 0 &&
(pt->pt_valid == NULL ||
(*pt->pt_valid)(p, dvp->v_mount)))
goto found;
}
break;
found:
if (pt->pt_pfstype == Pfile) {
fvp = p->p_textvp;
/* We already checked that it exists. */
VREF(fvp);
vn_lock(fvp, LK_EXCLUSIVE | LK_RETRY, curp);
if (wantpunlock) {
VOP_UNLOCK(dvp, 0, curp);
cnp->cn_flags |= PDIRUNLOCK;
}
*vpp = fvp;
return (0);
}
error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid,
pt->pt_pfstype);
if ((error == 0) && (wantpunlock)) {
VOP_UNLOCK(dvp, 0, curp);
cnp->cn_flags |= PDIRUNLOCK;
}
return (error);
default:
return (ENOTDIR);
}
return (cnp->cn_nameiop == LOOKUP ? ENOENT : EROFS);
}
/*
* 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.
*/
int
procfs_getattr(void *v)
{
struct vop_getattr_args *ap = v;
struct pfsnode *pfs = VTOPFS(ap->a_vp);
struct vattr *vap = ap->a_vap;
struct proc *procp;
int error;
/* first check the process still exists */
switch (pfs->pfs_type) {
case Proot:
case Pcurproc:
case Pcpuinfo:
case Pmeminfo:
procp = 0;
break;
default:
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.
*/
getnanotime(&vap->va_ctime);
vap->va_atime = vap->va_mtime = vap->va_ctime;
switch (pfs->pfs_type) {
case Pregs:
case Pfpregs:
#ifndef PTRACE
break;
#endif
case Pmem:
/*
* If the process has exercised some setuid or setgid
* privilege, then rip away read/write permission so
* that only root can gain access.
*/
if (procp->p_flag & P_SUGID)
vap->va_mode &= ~(S_IRUSR|S_IWUSR);
/* FALLTHROUGH */
case Pctl:
case Pstatus:
case Pnote:
case Pnotepg:
case Pcmdline:
vap->va_nlink = 1;
vap->va_uid = procp->p_ucred->cr_uid;
vap->va_gid = procp->p_ucred->cr_gid;
break;
case Pmeminfo:
case Pcpuinfo:
vap->va_nlink = 1;
vap->va_uid = vap->va_gid = 0;
break;
case Pproc:
case Pfile:
case Proot:
case Pcurproc:
case Pself:
break;
}
/*
* 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:
/*
* Set nlink to 1 to tell fts(3) we don't actually know.
*/
vap->va_nlink = 1;
vap->va_uid = 0;
vap->va_gid = 0;
vap->va_size = vap->va_bytes = DEV_BSIZE;
break;
case Pcurproc: {
char buf[16]; /* should be enough */
int len;
len = snprintf(buf, sizeof buf, "%ld", (long)curproc->p_pid);
if (len == -1 || len >= sizeof buf) {
error = EINVAL;
break;
}
vap->va_nlink = 1;
vap->va_uid = 0;
vap->va_gid = 0;
vap->va_size = vap->va_bytes = len;
break;
}
case Pself:
vap->va_nlink = 1;
vap->va_uid = 0;
vap->va_gid = 0;
vap->va_size = vap->va_bytes = sizeof("curproc");
break;
case Pproc:
vap->va_nlink = 2;
vap->va_uid = procp->p_ucred->cr_uid;
vap->va_gid = procp->p_ucred->cr_gid;
vap->va_size = vap->va_bytes = DEV_BSIZE;
break;
case Pfile:
error = EOPNOTSUPP;
break;
case Pmem:
vap->va_bytes = vap->va_size =
ptoa(procp->p_vmspace->vm_tsize +
procp->p_vmspace->vm_dsize +
procp->p_vmspace->vm_ssize);
break;
case Pregs:
#ifdef PTRACE
vap->va_bytes = vap->va_size = sizeof(struct reg);
#endif
break;
case Pfpregs:
#if defined(PT_GETFPREGS) || defined(PT_SETFPREGS)
#ifdef PTRACE
vap->va_bytes = vap->va_size = sizeof(struct fpreg);
#endif
#endif
break;
case Pctl:
case Pstatus:
case Pnote:
case Pnotepg:
case Pcmdline:
case Pmeminfo:
case Pcpuinfo:
vap->va_bytes = vap->va_size = 0;
break;
#ifdef DIAGNOSTIC
default:
panic("procfs_getattr");
#endif
}
return (error);
}
/*
* do an ioctl operation on a pfsnode (vp).
* (vp) is not locked on entry or exit.
*/
static int
procfs_ioctl(struct vop_ioctl_args *ap)
{
struct pfsnode *pfs = VTOPFS(ap->a_vp);
struct proc *procp;
struct proc *p;
int error;
int signo;
struct procfs_status *psp;
unsigned char flags;
procp = pfs_pfind(pfs->pfs_pid);
if (procp == NULL)
return ENOTTY;
p = curproc;
if (p == NULL) {
error = EINVAL;
goto done;
}
/* Can't trace a process that's currently exec'ing. */
if ((procp->p_flags & P_INEXEC) != 0) {
error = EAGAIN;
goto done;
}
if (!CHECKIO(p, procp) || p_trespass(ap->a_cred, procp->p_ucred)) {
error = EPERM;
goto done;
}
switch (ap->a_command) {
case PIOCBIS:
spin_lock(&procp->p_spin);
procp->p_stops |= *(unsigned int*)ap->a_data;
spin_unlock(&procp->p_spin);
break;
case PIOCBIC:
spin_lock(&procp->p_spin);
procp->p_stops &= ~*(unsigned int*)ap->a_data;
spin_unlock(&procp->p_spin);
break;
case PIOCSFL:
/*
* NFLAGS is "non-suser_xxx flags" -- currently, only
* PFS_ISUGID ("ignore set u/g id");
*/
#define NFLAGS (PF_ISUGID)
flags = (unsigned char)*(unsigned int*)ap->a_data;
if (flags & NFLAGS && (error = priv_check_cred(ap->a_cred, PRIV_ROOT, 0)))
goto done;
procp->p_pfsflags = flags;
break;
case PIOCGFL:
*(unsigned int*)ap->a_data = (unsigned int)procp->p_pfsflags;
break;
case PIOCSTATUS:
/*
* NOTE: syscall entry deals with stopevents and may run without
* the MP lock.
*/
psp = (struct procfs_status *)ap->a_data;
psp->flags = procp->p_pfsflags;
psp->events = procp->p_stops;
spin_lock(&procp->p_spin);
if (procp->p_step) {
psp->state = 0;
psp->why = procp->p_stype;
psp->val = procp->p_xstat;
spin_unlock(&procp->p_spin);
} else {
psp->state = 1;
spin_unlock(&procp->p_spin);
psp->why = 0; /* Not defined values */
psp->val = 0; /* Not defined values */
}
break;
case PIOCWAIT:
/*
* NOTE: syscall entry deals with stopevents and may run without
* the MP lock.
*/
psp = (struct procfs_status *)ap->a_data;
spin_lock(&procp->p_spin);
while (procp->p_step == 0) {
tsleep_interlock(&procp->p_stype, PCATCH);
spin_unlock(&procp->p_spin);
if (procp->p_stops == 0) {
error = EINVAL;
goto done;
}
if (procp->p_flags & P_POSTEXIT) {
error = EINVAL;
goto done;
}
if (procp->p_flags & P_INEXEC) {
error = EAGAIN;
goto done;
}
error = tsleep(&procp->p_stype, PCATCH | PINTERLOCKED,
"piocwait", 0);
if (error)
goto done;
spin_lock(&procp->p_spin);
}
spin_unlock(&procp->p_spin);
psp->state = 1; /* It stopped */
psp->flags = procp->p_pfsflags;
psp->events = procp->p_stops;
psp->why = procp->p_stype; /* why it stopped */
psp->val = procp->p_xstat; /* any extra info */
break;
case PIOCCONT: /* Restart a proc */
/*
* NOTE: syscall entry deals with stopevents and may run without
* the MP lock. However, the caller is presumably interlocked
* by having waited.
*/
if (procp->p_step == 0) {
error = EINVAL; /* Can only start a stopped process */
goto done;
}
if ((signo = *(int*)ap->a_data) != 0) {
if (signo >= NSIG || signo <= 0) {
error = EINVAL;
goto done;
}
ksignal(procp, signo);
}
procp->p_step = 0;
wakeup(&procp->p_step);
break;
default:
error = ENOTTY;
goto done;
}
error = 0;
done:
pfs_pdone(procp);
return 0;
}
/*
* 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(struct vnode *a_vp, struct vattr *a_vap)
*/
static int
procfs_getattr(struct vop_getattr_args *ap)
{
struct pfsnode *pfs = VTOPFS(ap->a_vp);
struct vattr *vap = ap->a_vap;
struct proc *procp;
int error;
/*
* First make sure that the process and its credentials
* still exist.
*/
switch (pfs->pfs_type) {
case Proot:
case Pcurproc:
procp = NULL;
break;
default:
procp = pfs_pfind(pfs->pfs_pid);
if (procp == NULL || procp->p_ucred == NULL) {
error = ENOENT;
goto done;
}
}
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;
vap->va_fsid = ap->a_vp->v_mount->mnt_stat.f_fsid.val[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.
*/
nanotime(&vap->va_ctime);
vap->va_atime = vap->va_mtime = vap->va_ctime;
/*
* If the process has exercised some setuid or setgid
* privilege, then rip away read/write permission so
* that only root can gain access.
*/
switch (pfs->pfs_type) {
case Pctl:
case Pregs:
case Pfpregs:
case Pdbregs:
case Pmem:
if (procp->p_flag & P_SUGID)
vap->va_mode &= ~((VREAD|VWRITE)|
((VREAD|VWRITE)>>3)|
((VREAD|VWRITE)>>6));
break;
default:
break;
}
/*
* 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.
*/
vap->va_nlink = 1;
if (procp) {
vap->va_uid = procp->p_ucred->cr_uid;
vap->va_gid = procp->p_ucred->cr_gid;
}
switch (pfs->pfs_type) {
case Proot:
/*
* Set nlink to 1 to tell fts(3) we don't actually know.
*/
vap->va_nlink = 1;
vap->va_uid = 0;
vap->va_gid = 0;
vap->va_size = vap->va_bytes = DEV_BSIZE;
break;
case Pcurproc: {
char buf[16]; /* should be enough */
vap->va_uid = 0;
vap->va_gid = 0;
vap->va_size = vap->va_bytes =
ksnprintf(buf, sizeof(buf), "%ld", (long)curproc->p_pid);
break;
}
case Pproc:
vap->va_nlink = nproc_targets;
vap->va_size = vap->va_bytes = DEV_BSIZE;
break;
case Pfile: {
char *fullpath, *freepath;
error = cache_fullpath(procp, &procp->p_textnch, &fullpath, &freepath, 0);
if (error == 0) {
vap->va_size = strlen(fullpath);
kfree(freepath, M_TEMP);
} else {
vap->va_size = sizeof("unknown") - 1;
error = 0;
}
vap->va_bytes = vap->va_size;
break;
}
case Pmem:
/*
* If we denied owner access earlier, then we have to
* change the owner to root - otherwise 'ps' and friends
* will break even though they are setgid kmem. *SIGH*
*/
if (procp->p_flag & P_SUGID)
vap->va_uid = 0;
else
vap->va_uid = procp->p_ucred->cr_uid;
break;
case Pregs:
vap->va_bytes = vap->va_size = sizeof(struct reg);
break;
case Pfpregs:
vap->va_bytes = vap->va_size = sizeof(struct fpreg);
break;
case Pdbregs:
vap->va_bytes = vap->va_size = sizeof(struct dbreg);
break;
case Ptype:
case Pmap:
case Pctl:
case Pstatus:
case Pnote:
case Pnotepg:
case Pcmdline:
case Prlimit:
break;
default:
panic("procfs_getattr");
}
done:
if (procp)
PRELE(procp);
return (error);
}
/*
* 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(struct vnode *a_vp, struct vattr *a_vap)
*/
static int
procfs_getattr(struct vop_getattr_args *ap)
{
struct pfsnode *pfs = VTOPFS(ap->a_vp);
struct vattr *vap = ap->a_vap;
struct proc *procp;
int error;
/*
* First make sure that the process and its credentials
* still exist.
*/
switch (pfs->pfs_type) {
case Proot:
case Pcurproc:
procp = NULL;
break;
default:
procp = pfs_pfind(pfs->pfs_pid);
if (procp == NULL || procp->p_ucred == NULL) {
error = ENOENT;
goto done;
}
break;
}
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;
vap->va_fsid = ap->a_vp->v_mount->mnt_stat.f_fsid.val[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.
*/
nanotime(&vap->va_ctime);
vap->va_atime = vap->va_mtime = vap->va_ctime;
/*
* If the process has exercised some setuid or setgid
* privilege, then rip away read/write permission so
* that only root can gain access.
*/
switch (pfs->pfs_type) {
case Pctl:
case Pregs:
case Pfpregs:
case Pdbregs:
case Pmem:
if (procp->p_flags & P_SUGID) {
vap->va_mode &= ~((VREAD|VWRITE)|
((VREAD|VWRITE)>>3)|
((VREAD|VWRITE)>>6));
}
break;
default:
break;
}
