inoptr newfile(inoptr pino, char *name)
{
inoptr nindex;
uint8_t j;
/* First see if parent is writeable */
if(!(getperm(pino) & OTH_WR))
goto nogood;
if(!(nindex = i_open(pino->c_dev, 0)))
goto nogood;
/* BUG FIX: user/group setting was missing SN */
nindex->c_node.i_uid = udata.u_euid;
nindex->c_node.i_gid = udata.u_egid;
nindex->c_node.i_mode = F_REG; /* For the time being */
nindex->c_node.i_nlink = 1;
nindex->c_node.i_size = 0;
for(j=0; j <20; j++)
nindex->c_node.i_addr[j] = 0;
wr_inode(nindex);
if(!ch_link(pino, "", name, nindex)) {
i_deref(nindex);
goto nogood;
}
i_deref(pino);
return nindex;
nogood:
i_deref(pino);
return NULLINODE;
}
static arg_t chdiroot_op(inoptr ino, inoptr * p)
{
if (getmode(ino) != F_DIR) {
udata.u_error = ENOTDIR;
i_deref(ino);
return (-1);
}
i_deref(*p);
*p = ino;
return 0;
}
int16_t _stat(void)
{
inoptr ino;
if (!(ino = n_open(path, NULLINOPTR)))
return (-1);
stcpy(ino, buf);
i_deref(ino);
return (0);
}
void oft_deref(int8_t of)
{
struct oft *ofptr;
ofptr = of_tab + of;
if(!(--ofptr->o_refs) && ofptr->o_inode) {
i_deref(ofptr->o_inode);
ofptr->o_inode = NULLINODE;
}
}
arg_t _stat(void)
{
inoptr ino;
int err;
if (!(ino = n_open(path, NULLINOPTR)))
return (-1);
err = stcpy(ino, buf);
i_deref(ino);
return err;
}
arg_t _mknod(void)
{
inoptr ino;
inoptr parent;
char fname[FILENAME_LEN + 1];
udata.u_error = 0;
if (!super() && ((mode & F_MASK) != F_PIPE)) {
udata.u_error = EPERM;
goto nogood3;
}
if ((ino = n_open(name, &parent)) != NULL) {
udata.u_error = EEXIST;
goto nogood;
}
if (!parent) {
udata.u_error = ENOENT;
return (-1);
}
filename(name, fname);
ino = newfile(parent, fname);
if(!ino)
goto nogood3; /* parent inode is derefed in newfile. SN */
/* Initialize mode and dev */
ino->c_node.i_mode = mode & ~udata.u_mask;
ino->c_node.i_addr[0] = isdevice(ino) ? dev : 0;
setftime(ino, A_TIME | M_TIME | C_TIME);
wr_inode(ino);
i_deref(ino);
return (0);
nogood:
i_deref(ino);
i_deref(parent);
nogood3:
return (-1);
}
arg_t _chown(void)
{
inoptr ino;
int ret;
if (!(ino = n_open(path, NULLINOPTR)))
return (-1);
ret = chown_op(ino);
i_deref(ino);
return ret;
}
static int chown_op(inoptr ino)
{
if (ino->c_node.i_uid != udata.u_euid && esuper()) {
i_deref(ino);
return (-1);
}
ino->c_node.i_uid = owner;
ino->c_node.i_gid = group;
setftime(ino, C_TIME);
return 0;
}
arg_t _unlink(void)
{
inoptr ino;
inoptr pino;
int r;
char fname[FILENAME_LEN + 1];
ino = n_open(path, &pino);
if (!(pino && ino)) {
if (pino) /* parent exist */
i_deref(pino);
udata.u_error = ENOENT;
return (-1);
}
filename(path, fname);
r = unlinki(ino, pino, fname);
i_deref(pino);
i_deref(ino);
return r;
}
static int16_t chmod_op(inoptr ino)
{
if (ino->c_node.i_uid != udata.u_euid && esuper()) {
i_deref(ino);
return (-1);
}
ino->c_node.i_mode =
(mode & MODE_MASK) | (ino->c_node.i_mode & F_MASK);
setftime(ino, C_TIME);
return 0;
}
uint8_t write_core_image(void)
{
inoptr parent = NULLINODE;
inoptr ino;
udata.u_error = 0;
/* FIXME: need to think more about the case sp is lala */
if (uput("core", (uint8_t *)udata.u_syscall_sp - 5, 5))
return 0;
ino = n_open((char *)udata.u_syscall_sp - 5, &parent);
if (ino) {
i_deref(parent);
return 0;
}
if (parent) {
i_lock(parent);
if ((ino = newfile(parent, "core")) != NULL) {
ino->c_node.i_mode = F_REG | 0400;
setftime(ino, A_TIME | M_TIME | C_TIME);
wr_inode(ino);
f_trunc(ino);
#if 0
/* FIXME address ranges for different models - move core writer
for address spaces into helpers ? */
/* FIXME: need to add some arch specific header bits, and
also pull stuff like the true sp and registers out of
the return stack properly */
corehdr.ch_base = pagemap_base;
corehdr.ch_break = udata.u_break;
corehdr.ch_sp = udata.u_syscall_sp;
corehdr.ch_top = PROGTOP;
udata.u_offset = 0;
udata.u_base = (uint8_t *)&corehdr;
udata.u_sysio = true;
udata.u_count = sizeof(corehdr);
writei(ino, 0);
udata.u_sysio = false;
udata.u_base = (uint8_t *)pagemap_base;
udata.u_count = udata.u_break - pagemap_base;
writei(ino, 0);
udata.u_base = udata.u_sp;
udata.u_count = PROGTOP - (uint32_t)udata.u_sp;
writei(ino, 0);
#endif
i_unlock_deref(ino);
return W_COREDUMP;
}
}
return 0;
}
arg_t _link(void)
{
inoptr ino;
inoptr ino2;
inoptr parent2;
char fname[FILENAME_LEN + 1];
if (!(ino = n_open(name1, NULLINOPTR)))
return (-1);
if (getmode(ino) == F_DIR && esuper())
goto nogood;
if (ino->c_node.i_nlink == 0xFFFF) {
udata.u_error = EMLINK;
goto nogood;
}
/* Make sure file2 doesn't exist, and get its parent */
if ((ino2 = n_open(name2, &parent2)) != NULL) {
i_deref(ino2);
i_deref(parent2);
udata.u_error = EEXIST;
goto nogood;
}
if (!parent2)
goto nogood;
if (ino->c_dev != parent2->c_dev) {
i_deref(parent2);
udata.u_error = EXDEV;
goto nogood;
}
filename(name2, fname);
if (!ch_link(parent2, "", fname, ino)) {
i_deref(parent2);
goto nogood;
}
/* Update the link count. */
++ino->c_node.i_nlink;
wr_inode(ino);
setftime(ino, C_TIME);
i_deref(parent2);
i_deref(ino);
return 0;
nogood:
i_deref(ino);
return -1;
}
int16_t _utime(void)
{
inoptr ino;
time_t t[2];
if (!valaddr(buf, 2 * sizeof(time_t)))
return (-1);
if (!(ino = n_open(file, NULLINOPTR)))
return (-1);
if (ino->c_node.i_uid != udata.u_euid && esuper()) {
i_deref(ino);
return (-1);
}
uget(buf, t, 2 * sizeof(time_t));
/* FIXME: needs updating once we pack top bits
elsewhere in the inode */
ino->c_node.i_atime = t[0];
ino->c_node.i_mtime = t[1];
setftime(ino, C_TIME);
i_deref(ino);
return (0);
}
inoptr srch_mt(inoptr ino)
{
uint8_t j;
struct mount *m = &fs_tab[0];
for(j=0; j < NMOUNTS; ++j){
if(m->m_dev != NO_DEVICE && m->m_fs->s_mntpt == ino) {
i_deref(ino);
return i_open(m->m_dev, ROOTINODE);
}
m++;
};
return ino;
}
uint8_t write_core_image(void)
{
inoptr parent = NULLINODE;
inoptr ino;
udata.u_error = 0;
/* FIXME: need to think more about the case sp is lala */
if (uput("core", udata.u_syscall_sp - 5, 5))
return 0;
ino = n_open(udata.u_syscall_sp - 5, &parent);
if (ino) {
i_deref(parent);
return 0;
}
if (parent) {
i_lock(parent);
if (ino = newfile(parent, "core")) {
ino->c_node.i_mode = F_REG | 0400;
setftime(ino, A_TIME | M_TIME | C_TIME);
wr_inode(ino);
f_trunc(ino);
/* FIXME: need to add some arch specific header bits, and
also pull stuff like the true sp and registers out of
the return stack properly */
corehdr.ch_base = MAPBASE;
corehdr.ch_break = udata.u_break;
corehdr.ch_sp = udata.u_syscall_sp;
corehdr.ch_top = PROGTOP;
udata.u_offset = 0;
udata.u_base = (uint8_t *)&corehdr;
udata.u_sysio = true;
udata.u_count = sizeof(corehdr);
writei(ino, 0);
udata.u_sysio = false;
udata.u_base = MAPBASE;
udata.u_count = udata.u_break - MAPBASE;
writei(ino, 0);
udata.u_base = udata.u_sp;
udata.u_count = PROGTOP - (uint16_t)udata.u_sp;
writei(ino, 0);
i_unlock_deref(ino);
return W_COREDUMP;
}
}
return 0;
}
arg_t _utime(void)
{
inoptr ino;
time_t t[2];
if (!(ino = n_open(file, NULLINOPTR)))
return (-1);
if (ino->c_flags & CRDONLY) {
udata.u_error = EROFS;
goto out2;
}
/* Special case in the Unix API - NULL means now */
if (buf) {
if (ino->c_node.i_uid != udata.u_euid && esuper())
goto out;
if (!valaddr(buf, 2 * sizeof(time_t)))
goto out2;
uget(buf, t, 2 * sizeof(time_t));
} else {
if (!(getperm(ino) & OTH_WR))
goto out;
rdtime(&t[0]);
memcpy(&t[1], &t[0], sizeof(t[1]));
}
/* FIXME: needs updating once we pack top bits
elsewhere in the inode */
ino->c_node.i_atime = t[0].low;
ino->c_node.i_mtime = t[1].low;
setftime(ino, C_TIME);
i_deref(ino);
return (0);
out:
udata.u_error = EPERM;
out2:
i_deref(ino);
return -1;
}
arg_t _access(void)
{
inoptr ino;
uint16_t euid;
uint16_t egid;
int16_t retval;
if ((mode & 07) && !ugetc(path)) {
udata.u_error = ENOENT;
return (-1);
}
/* Temporarily make eff. id real id. */
euid = udata.u_euid;
egid = udata.u_egid;
udata.u_euid = udata.u_ptab->p_uid;
udata.u_egid = udata.u_gid;
if (!(ino = n_open(path, NULLINOPTR))) {
retval = -1;
goto nogood;
}
retval = 0;
if (~getperm(ino) & (mode & 07)) {
udata.u_error = EACCES;
retval = -1;
}
i_deref(ino);
nogood:
udata.u_euid = euid;
udata.u_egid = egid;
return (retval);
}
void doexit(int16_t val, int16_t val2)
{
int16_t j;
ptptr p;
irqflags_t irq;
#ifdef DEBUG
kprintf("process %d exiting\n", udata.u_ptab->p_pid);
kprintf
("udata.u_page %u, udata.u_ptab %x, udata.u_ptab->p_page %u\n",
udata.u_page, udata.u_ptab, udata.u_ptab->p_page);
#endif
if (udata.u_ptab->p_pid == 1)
panic("killed init");
_sync(); /* Not necessary, but a good idea. */
irq = di();
/* Discard our memory before we blow away and reuse the memory */
pagemap_free(udata.u_ptab);
for (j = 0; j < UFTSIZE; ++j) {
if (udata.u_files[j] != NO_FILE)
doclose(j);
}
udata.u_ptab->p_exitval = (val << 8) | (val2 & 0xff);
i_deref(udata.u_cwd);
i_deref(udata.u_root);
/* Stash away child's execution tick counts in process table,
* overlaying some no longer necessary stuff.
*
* Pedantically POSIX says we should do this at the point of wait()
*/
udata.u_utime += udata.u_cutime;
udata.u_stime += udata.u_cstime;
memcpy(&(udata.u_ptab->p_priority), &udata.u_utime,
2 * sizeof(clock_t));
for (p = ptab; p < ptab_end; ++p) {
if (p->p_status == P_EMPTY || p == udata.u_ptab)
continue;
/* Set any child's parents to our parent */
if (p->p_pptr == udata.u_ptab)
p->p_pptr = udata.u_ptab->p_pptr;
/* Send SIGHUP to any pgrp members and remove
them from our pgrp */
if (p->p_pgrp == udata.u_ptab->p_pid) {
p->p_pgrp = 0;
ssig(p, SIGHUP);
}
}
tty_exit();
irqrestore(irq);
#ifdef DEBUG
kprintf
("udata.u_page %u, udata.u_ptab %x, udata.u_ptab->p_page %u\n",
udata.u_page, udata.u_ptab, udata.u_ptab->p_page);
#endif
#ifdef CONFIG_ACCT
acctexit(p);
#endif
udata.u_page = 0xFFFFU;
udata.u_page2 = 0xFFFFU;
/* FIXME: send SIGCLD here */
/* FIXME: POSIX.1 says that SIG_IGN for SIGCLD means don't go
zombie, just clean up as we go */
/* Wake up a waiting parent, if any. */
wakeup((char *) udata.u_ptab->p_pptr);
udata.u_ptab->p_status = P_ZOMBIE;
nready--;
nproc--;
switchin(getproc());
panic("doexit: won't exit");
}
void doexit(uint16_t val)
{
int16_t j;
ptptr p;
irqflags_t irq;
#ifdef DEBUG
kprintf("process %d exiting\n", udata.u_ptab->p_pid);
kprintf
("udata.u_page %u, udata.u_ptab %x, udata.u_ptab->p_page %u\n",
udata.u_page, udata.u_ptab, udata.u_ptab->p_page);
#endif
if (udata.u_ptab->p_pid == 1)
panic(PANIC_KILLED_INIT);
sync(); /* Not necessary, but a good idea. */
irq = di();
/* Discard our memory before we blow away and reuse the memory */
pagemap_free(udata.u_ptab);
for (j = 0; j < UFTSIZE; ++j) {
if (udata.u_files[j] != NO_FILE)
doclose(j);
}
udata.u_ptab->p_exitval = val;
i_deref(udata.u_cwd);
i_deref(udata.u_root);
/* Stash away child's execution tick counts in process table,
* overlaying some no longer necessary stuff.
*
* Pedantically POSIX says we should do this at the point of wait()
*/
udata.u_utime += udata.u_cutime;
udata.u_stime += udata.u_cstime;
memcpy(&(udata.u_ptab->p_priority), &udata.u_utime,
2 * sizeof(clock_t));
for (p = ptab; p < ptab_end; ++p) {
if (p->p_status == P_EMPTY || p == udata.u_ptab)
continue;
/* Set any child's parents to our parent */
/* FIXME: do we need to wakeup the parent if we do this ? */
if (p->p_pptr == udata.u_ptab)
p->p_pptr = udata.u_ptab->p_pptr;
/* Send SIGHUP to any pgrp members and remove
them from our pgrp */
if (p->p_pgrp == udata.u_ptab->p_pid) {
p->p_pgrp = 0;
ssig(p, SIGHUP);
ssig(p, SIGCONT);
}
}
tty_exit();
irqrestore(irq);
#ifdef DEBUG
kprintf
("udata.u_page %u, udata.u_ptab %x, udata.u_ptab->p_page %u\n",
udata.u_page, udata.u_ptab, udata.u_ptab->p_page);
#endif
#ifdef CONFIG_ACCT
acctexit(p);
#endif
udata.u_page = 0xFFFFU;
udata.u_page2 = 0xFFFFU;
signal_parent(udata.u_ptab->p_pptr);
nready--;
nproc--;
switchin(getproc());
panic(PANIC_DOEXIT);
}
inoptr kn_open(char *namep, inoptr *parent)
{
staticfast inoptr wd; /* the directory we are currently searching. */
staticfast inoptr ninode;
inoptr temp;
staticfast char *name;
name = namep;
#ifdef DEBUG
kprintf("kn_open(\"%s\")\n", name);
#endif
if(*name == '/')
wd = udata.u_root;
else
wd = udata.u_cwd;
i_ref(ninode = wd);
i_ref(ninode);
for(;;)
{
if(ninode)
magic(ninode);
/* cheap way to spot rename inside yourself */
if (udata.u_rename == ninode)
udata.u_rename = NULLINODE;
/* See if we are at a mount point */
if(ninode)
ninode = srch_mt(ninode);
while(*name == '/') /* Skip(possibly repeated) slashes */
++name;
if(!*name) /* No more components of path? */
break;
if(!ninode){
udata.u_error = ENOENT;
goto nodir;
}
i_deref(wd);
wd = ninode;
if(getmode(wd) != F_DIR){
udata.u_error = ENOTDIR;
goto nodir;
}
if(!(getperm(wd) & OTH_EX)){
udata.u_error = EPERM;
goto nodir;
}
/* See if we are going up through a mount point */
if((wd == udata.u_root || (wd->c_num == ROOTINODE && wd->c_dev != root_dev)) &&
name[0] == '.' && name[1] == '.' &&
(name[2] == '/' || name[2] == '\0')){
if (wd == udata.u_root) {
ninode = wd;
name += 2;
continue;
}
temp = fs_tab_get(wd->c_dev)->m_fs->s_mntpt;
++temp->c_refs;
i_deref(wd);
wd = temp;
}
ninode = srch_dir(wd, name);
while(*name != '/' && *name)
++name;
}
if(parent)
*parent = wd;
else
i_deref(wd);
if(!(parent || ninode))
udata.u_error = ENOENT;
return ninode;
nodir:
if(parent)
*parent = NULLINODE;
i_deref(wd);
return NULLINODE;
}
void doexit(uint16_t val)
{
int16_t j;
ptptr p;
irqflags_t irq;
#ifdef DEBUG
kprintf("process %d exiting %d\n", udata.u_ptab->p_pid, val);
kprintf
("udata.u_page %u, udata.u_ptab %p, udata.u_ptab->p_page %u\n",
udata.u_page, udata.u_ptab, udata.u_ptab->p_page);
#endif
if (udata.u_ptab->p_pid == 1)
panic(PANIC_KILLED_INIT);
sync(); /* Not necessary, but a good idea. */
irq = di();
/* We are exiting, hold all signals (they will never be
delivered). If we don't do this we might take a signal
while exiting which would be ... unfortunate */
udata.u_ptab->p_held = 0xFFFFFFFFUL;
udata.u_cursig = 0;
/* Discard our memory before we blow away and reuse the memory */
pagemap_free(udata.u_ptab);
for (j = 0; j < UFTSIZE; ++j) {
if (udata.u_files[j] != NO_FILE)
doclose(j);
}
udata.u_ptab->p_exitval = val;
i_deref(udata.u_cwd);
i_deref(udata.u_root);
/* Stash away child's execution tick counts in process table,
* overlaying some no longer necessary stuff.
*
* Pedantically POSIX says we should do this at the point of wait()
*/
udata.u_utime += udata.u_cutime;
udata.u_stime += udata.u_cstime;
memcpy(&(udata.u_ptab->p_priority), &udata.u_utime,
2 * sizeof(clock_t));
for (p = ptab; p < ptab_end; ++p) {
if (p->p_status == P_EMPTY || p == udata.u_ptab)
continue;
/* Set any child's parents to init */
if (p->p_pptr == udata.u_ptab) {
p->p_pptr = ptab; /* ptab is always init */
/* Suppose our child is a zombie and init has
SIGCLD blocked */
if (ptab[0].p_ignored & (1UL << SIGCHLD)) {
p->p_status = P_EMPTY;
} else {
ssig(&ptab[0], SIGCHLD);
wakeup(&ptab[0]);
}
}
/* Send SIGHUP to any pgrp members and remove
them from our pgrp */
if (p->p_pgrp == udata.u_ptab->p_pid) {
p->p_pgrp = 0;
ssig(p, SIGHUP);
ssig(p, SIGCONT);
}
}
tty_exit();
irqrestore(irq);
#ifdef DEBUG
kprintf
("udata.u_page %u, udata.u_ptab %p, udata.u_ptab->p_page %u\n",
udata.u_page, udata.u_ptab, udata.u_ptab->p_page);
#endif
#ifdef CONFIG_ACCT
acctexit(p);
#endif
udata.u_page = 0xFFFFU;
udata.u_page2 = 0xFFFFU;
signal_parent(udata.u_ptab->p_pptr);
nready--;
nproc--;
switchin(getproc());
panic(PANIC_DOEXIT);
}
arg_t _open(void)
{
int8_t uindex;
int8_t oftindex;
staticfast inoptr ino;
int16_t perm;
staticfast inoptr parent;
char fname[FILENAME_LEN + 1];
int trunc;
int r;
int w;
int j;
parent = NULLINODE;
trunc = flag & O_TRUNC;
r = (flag + 1) & 1;
w = (flag + 1) & 2;
if (O_ACCMODE(flag) == 3 || (flag & O_BADBITS)) {
udata.u_error = EINVAL;
return (-1);
}
if ((uindex = uf_alloc()) == -1)
return (-1);
if ((oftindex = oft_alloc()) == -1)
goto nooft;
ino = n_open(name, &parent);
if (ino) {
i_deref(parent);
/* O_EXCL test */
if ((flag & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL)) {
udata.u_error = EEXIST;
goto idrop;
}
} else {
/* The n_open failed */
if (udata.u_error == EFAULT)
goto cantopen;
/* New file */
if (!(flag & O_CREAT)) {
udata.u_error = ENOENT;
goto cantopen;
}
filename(name, fname);
/* newfile drops parent for us */
if (parent && (ino = newfile(parent, fname))) {
ino->c_node.i_mode =
(F_REG | (mode & MODE_MASK & ~udata.u_mask));
setftime(ino, A_TIME | M_TIME | C_TIME);
wr_inode(ino);
} else {
udata.u_error = ENFILE; /* FIXME, should be set in newfile
not bodged to a guessed code */
goto cantopen;
}
}
/* Book our slot in case we block opening a device */
of_tab[oftindex].o_inode = ino;
perm = getperm(ino);
if ((r && !(perm & OTH_RD)) || (w && !(perm & OTH_WR))) {
udata.u_error = EACCES;
goto cantopen;
}
if (w) {
if (getmode(ino) == F_DIR ) {
udata.u_error = EISDIR;
goto cantopen;
}
if (ino->c_flags & CRDONLY) {
udata.u_error = EROFS;
goto cantopen;
}
}
if (isdevice(ino)) {
inoptr *iptr = &of_tab[oftindex].o_inode;
/* d_open may block and thus ino may become invalid as may
parent (but we don't need it again). It may also be changed
by the call to dev_openi */
if (dev_openi(iptr, flag) != 0)
goto cantopen;
/* May have changed */
/* get the static pointer back in case it changed via dev
usage or just because we blocked */
ino = *iptr;
}
if (trunc && getmode(ino) == F_REG) {
if (f_trunc(ino))
goto idrop;
for (j = 0; j < OFTSIZE; ++j)
/* Arguably should fix at read/write */
if (of_tab[j].o_inode == ino)
of_tab[j].o_ptr = 0;
}
udata.u_files[uindex] = oftindex;
of_tab[oftindex].o_ptr = 0;
of_tab[oftindex].o_access = flag; /* Save the low bits only */
if (flag & O_CLOEXEC)
udata.u_cloexec |= (1 << oftindex);
/* FIXME: ATIME ? */
/*
* Sleep process if no writer or reader
*/
if (getmode(ino) == F_PIPE && of_tab[oftindex].o_refs == 1
&& !(flag & O_NDELAY))
psleep(ino);
/* From the moment of the psleep ino is invalid */
return (uindex);
idrop:
i_deref(ino);
cantopen:
oft_deref(oftindex); /* This will call i_deref() */
nooft:
udata.u_files[uindex] = NO_FILE;
return (-1);
}
arg_t _execve(void)
{
/* We aren't re-entrant where this matters */
uint8_t hdr[16];
staticfast inoptr ino;
char **nargv; /* In user space */
char **nenvp; /* In user space */
struct s_argblk *abuf, *ebuf;
int argc;
uint16_t progptr;
uint16_t progload;
staticfast uint16_t top;
uint16_t bin_size; /* Will need to be bigger on some cpus */
uint16_t bss;
top = ramtop;
if (!(ino = n_open_lock(name, NULLINOPTR)))
return (-1);
if (!((getperm(ino) & OTH_EX) &&
(ino->c_node.i_mode & F_REG) &&
(ino->c_node.i_mode & (OWN_EX | OTH_EX | GRP_EX)))) {
udata.u_error = EACCES;
goto nogood;
}
setftime(ino, A_TIME);
udata.u_offset = 0;
udata.u_count = 16;
udata.u_base = hdr;
udata.u_sysio = true;
readi(ino, 0);
if (udata.u_done != 16) {
udata.u_error = ENOEXEC;
goto nogood;
}
if (!header_ok(hdr)) {
udata.u_error = ENOEXEC;
goto nogood2;
}
progload = hdr[7] << 8;
if (progload == 0)
progload = PROGLOAD;
top = *(uint16_t *)(hdr + 8);
if (top == 0) /* Legacy 'all space' binary */
top = ramtop;
else /* Requested an amount, so adjust for the base */
top += progload;
bss = *(uint16_t *)(hdr + 14);
/* Binary doesn't fit */
/* FIXME: review overflows */
bin_size = ino->c_node.i_size;
progptr = bin_size + 1024 + bss;
if (progload < PROGLOAD || top - progload < progptr || progptr < bin_size) {
udata.u_error = ENOMEM;
goto nogood2;
}
udata.u_ptab->p_status = P_NOSLEEP;
/* If we made pagemap_realloc keep hold of some defined area we
could in theory just move the arguments up or down as part of
the process - that would save us all this hassle but replace it
with new hassle */
/* Gather the arguments, and put them in temporary buffers. */
abuf = (struct s_argblk *) tmpbuf();
/* Put environment in another buffer. */
ebuf = (struct s_argblk *) tmpbuf();
/* Read args and environment from process memory */
if (rargs(argv, abuf) || rargs(envp, ebuf))
goto nogood3; /* SN */
/* This must be the last test as it makes changes if it works */
/* FIXME: once we sort out chmem we can make stack and data
two elements. We never allocate 'code' as there is no split I/D */
/* This is only safe from deadlocks providing pagemap_realloc doesn't
sleep */
if (pagemap_realloc(0, top - MAPBASE, 0))
goto nogood3;
/* From this point on we are commmited to the exec() completing */
/* Core dump and ptrace permission logic */
#ifdef CONFIG_LEVEL_2
/* Q: should uid == 0 mean we always allow core */
if ((!(getperm(ino) & OTH_RD)) ||
(ino->c_node.i_mode & (SET_UID | SET_GID)))
udata.u_flags |= U_FLAG_NOCORE;
else
udata.u_flags &= ~U_FLAG_NOCORE;
#endif
udata.u_top = top;
udata.u_ptab->p_top = top;
/* setuid, setgid if executable requires it */
if (ino->c_node.i_mode & SET_UID)
udata.u_euid = ino->c_node.i_uid;
if (ino->c_node.i_mode & SET_GID)
udata.u_egid = ino->c_node.i_gid;
/* FIXME: In the execve case we may on some platforms have space
below PROGLOAD to clear... */
/* We are definitely going to succeed with the exec,
* so we can start writing over the old program
*/
uput(hdr, (uint8_t *)progload, 16);
/* At this point, we are committed to reading in and
* executing the program. This call must not block. */
close_on_exec();
/*
* Read in the rest of the program, block by block. We rely upon
* the optimization path in readi to spot this is a big move to user
* space and move it directly.
*/
progptr = progload + 16;
if (bin_size > 16) {
bin_size -= 16;
udata.u_base = (uint8_t *)progptr; /* We copied the first block already */
udata.u_count = bin_size;
udata.u_sysio = false;
readi(ino, 0);
if (udata.u_done != bin_size)
goto nogood4;
progptr += bin_size;
}
/* Wipe the memory in the BSS. We don't wipe the memory above
that on 8bit boxes, but defer it to brk/sbrk() */
uzero((uint8_t *)progptr, bss);
/* Set initial break for program */
udata.u_break = (int)ALIGNUP(progptr + bss);
/* Turn off caught signals */
memset(udata.u_sigvec, 0, sizeof(udata.u_sigvec));
// place the arguments, environment and stack at the top of userspace memory,
// Write back the arguments and the environment
nargv = wargs(((char *) top - 2), abuf, &argc);
nenvp = wargs((char *) (nargv), ebuf, NULL);
// Fill in udata.u_name with program invocation name
uget((void *) ugetw(nargv), udata.u_name, 8);
memcpy(udata.u_ptab->p_name, udata.u_name, 8);
tmpfree(abuf);
tmpfree(ebuf);
i_deref(ino);
/* Shove argc and the address of argv just below envp
FIXME: should flip them in crt0.S of app for R2L setups
so we can get rid of the ifdefs */
#ifdef CONFIG_CALL_R2L /* Arguments are stacked the 'wrong' way around */
uputw((uint16_t) nargv, nenvp - 2);
uputw((uint16_t) argc, nenvp - 1);
#else
uputw((uint16_t) nargv, nenvp - 1);
uputw((uint16_t) argc, nenvp - 2);
#endif
/* Set stack pointer for the program */
udata.u_isp = nenvp - 2;
/* Start execution (never returns) */
udata.u_ptab->p_status = P_RUNNING;
doexec(progload);
/* tidy up in various failure modes */
nogood4:
/* Must not run userspace */
ssig(udata.u_ptab, SIGKILL);
nogood3:
udata.u_ptab->p_status = P_RUNNING;
tmpfree(abuf);
tmpfree(ebuf);
nogood2:
nogood:
i_unlock_deref(ino);
return (-1);
}
arg_t _execve(void)
{
/* Not ideal on stack */
struct binfmt_flat binflat;
inoptr ino;
char **nargv; /* In user space */
char **nenvp; /* In user space */
struct s_argblk *abuf, *ebuf;
int argc;
uint32_t bin_size; /* Will need to be bigger on some cpus */
uaddr_t progbase, top;
uaddr_t go;
uint32_t true_brk;
if (!(ino = n_open(name, NULLINOPTR)))
return (-1);
if (!((getperm(ino) & OTH_EX) &&
(ino->c_node.i_mode & F_REG) &&
(ino->c_node.i_mode & (OWN_EX | OTH_EX | GRP_EX)))) {
udata.u_error = EACCES;
goto nogood;
}
setftime(ino, A_TIME);
udata.u_offset = 0;
udata.u_count = sizeof(struct binfmt_flat);
udata.u_base = (void *)&binflat;
udata.u_sysio = true;
readi(ino, 0);
if (udata.u_count != sizeof(struct binfmt_flat)) {
udata.u_error = ENOEXEC;
goto nogood;
}
/* FIXME: ugly - save this as valid_hdr modifies it */
true_brk = binflat.bss_end;
/* Hard coded for our 68K format. We don't quite use the ucLinux
names, we don't want to load a ucLinux binary in error! */
if (memcmp(binflat.magic, "bFLT", 4) || !valid_hdr(ino, &binflat)) {
udata.u_error = ENOEXEC;
goto nogood2;
}
/* Memory needed */
bin_size = binflat.bss_end + binflat.stack_size;
/* Overflow ? */
if (bin_size < binflat.bss_end) {
udata.u_error = ENOEXEC;
goto nogood2;
}
/* Gather the arguments, and put them in temporary buffers. */
abuf = (struct s_argblk *) tmpbuf();
/* Put environment in another buffer. */
ebuf = (struct s_argblk *) tmpbuf();
/* Read args and environment from process memory */
if (rargs(argv, abuf) || rargs(envp, ebuf))
goto nogood3;
/* This must be the last test as it makes changes if it works */
/* FIXME: need to update this to support split code/data and to fix
stack handling nicely */
if (pagemap_realloc(0, bin_size, 0))
goto nogood3;
/* Core dump and ptrace permission logic */
#ifdef CONFIG_LEVEL_2
/* Q: should uid == 0 mean we always allow core */
if ((!(getperm(ino) & OTH_RD)) ||
(ino->c_node.i_mode & (SET_UID | SET_GID)))
udata.u_flags |= U_FLAG_NOCORE;
else
udata.u_flags &= ~U_FLAG_NOCORE;
#endif
udata.u_codebase = progbase = pagemap_base();
/* From this point on we are commmited to the exec() completing
so we can start writing over the old program */
uput(&binflat, (uint8_t *)progbase, sizeof(struct binfmt_flat));
/* setuid, setgid if executable requires it */
if (ino->c_node.i_mode & SET_UID)
udata.u_euid = ino->c_node.i_uid;
if (ino->c_node.i_mode & SET_GID)
udata.u_egid = ino->c_node.i_gid;
top = progbase + bin_size;
// kprintf("user space at %p\n", progbase);
// kprintf("top at %p\n", progbase + bin_size);
bin_size = binflat.reloc_start + 4 * binflat.reloc_count;
go = (uint32_t)progbase + binflat.entry;
close_on_exec();
/*
* Read in the rest of the program, block by block. We rely upon
* the optimization path in readi to spot this is a big move to user
* space and move it directly.
*/
if (bin_size > sizeof(struct binfmt_flat)) {
/* We copied the header already */
bin_size -= sizeof(struct binfmt_flat);
udata.u_base = (uint8_t *)progbase +
sizeof(struct binfmt_flat);
udata.u_count = bin_size;
udata.u_sysio = false;
readi(ino, 0);
if (udata.u_count != bin_size)
goto nogood4;
}
/* Header isn't counted in relocations */
relocate(&binflat, progbase, bin_size);
/* This may wipe the relocations */
uzero((uint8_t *)progbase + binflat.data_end,
binflat.bss_end - binflat.data_end + binflat.stack_size);
/* Use of brk eats into the stack allocation */
/* Use the temporary we saved (hack) as we mangled bss_end */
udata.u_break = udata.u_codebase + true_brk;
/* Turn off caught signals */
memset(udata.u_sigvec, 0, sizeof(udata.u_sigvec));
/* place the arguments, environment and stack at the top of userspace memory. */
/* Write back the arguments and the environment */
nargv = wargs(((char *) top - 4), abuf, &argc);
nenvp = wargs((char *) (nargv), ebuf, NULL);
/* Fill in udata.u_name with Program invocation name */
uget((void *) ugetl(nargv, NULL), udata.u_name, 8);
memcpy(udata.u_ptab->p_name, udata.u_name, 8);
tmpfree(abuf);
tmpfree(ebuf);
i_deref(ino);
/* Shove argc and the address of argv just below envp */
uputl((uint32_t) nargv, nenvp - 1);
uputl((uint32_t) argc, nenvp - 2);
// Set stack pointer for the program
udata.u_isp = nenvp - 2;
/*
* Sort of - it's a good way to deal with all the stupidity of
* random 68K platforms we will have to handle, and a nice place
* to stuff the signal trampoline 8)
*/
install_vdso();
// kprintf("Go = %p ISP = %p\n", go, udata.u_isp);
doexec(go);
nogood4:
/* Must not run userspace */
ssig(udata.u_ptab, SIGKILL);
nogood3:
tmpfree(abuf);
tmpfree(ebuf);
nogood2:
nogood:
i_deref(ino);
return (-1);
}
