uintptr
sysexec(va_list list)
{
Segment *s, *ts;
int i;
Chan *tc;
char **argv, **argp, **argp0;
char *a, *e, *charp, *args, *file, *file0;
char *progarg[sizeof(Exec)/2+1], *elem, progelem[64];
ulong magic, ssize, nargs, nbytes, n;
uintptr t, d, b, entry, bssend, text, data, bss, tstk, align;
int indir;
Exec exec;
char line[sizeof(Exec)];
Fgrp *f;
Image *img;
Tos *tos;
args = elem = nil;
file0 = va_arg(list, char*);
validaddr((uintptr)file0, 1, 0);
argp0 = va_arg(list, char**);
evenaddr((uintptr)argp0);
validaddr((uintptr)argp0, 2*BY2WD, 0);
if(*argp0 == nil)
error(Ebadarg);
file0 = validnamedup(file0, 1);
if(waserror()){
free(file0);
free(elem);
free(args);
/* Disaster after commit */
if(up->seg[SSEG] == nil)
pexit(up->errstr, 1);
s = up->seg[ESEG];
if(s != nil){
putseg(s);
up->seg[ESEG] = nil;
}
nexterror();
}
align = BY2PG;
indir = 0;
file = file0;
for(;;){
tc = namec(file, Aopen, OEXEC, 0);
if(waserror()){
cclose(tc);
nexterror();
}
if(!indir)
kstrdup(&elem, up->genbuf);
n = devtab[tc->type]->read(tc, &exec, sizeof(Exec), 0);
if(n <= 2)
error(Ebadexec);
magic = l2be(exec.magic);
if(n == sizeof(Exec) && (magic == AOUT_MAGIC)){
entry = l2be(exec.entry);
text = l2be(exec.text);
if(magic & HDR_MAGIC)
text += 8;
switch(magic){
case S_MAGIC: /* 2MB segment alignment for amd64 */
align = 0x200000;
break;
case V_MAGIC: /* 16K segment alignment for mips */
align = 0x4000;
break;
}
if(text >= (USTKTOP-USTKSIZE)-(UTZERO+sizeof(Exec))
|| entry < UTZERO+sizeof(Exec)
|| entry >= UTZERO+sizeof(Exec)+text)
error(Ebadexec);
break; /* for binary */
}
/*
* Process #! /bin/sh args ...
*/
memmove(line, &exec, n);
if(indir || line[0]!='#' || line[1]!='!')
error(Ebadexec);
n = shargs(line, n, progarg);
if(n < 1)
error(Ebadexec);
indir = 1;
/*
* First arg becomes complete file name
*/
progarg[n++] = file;
progarg[n] = nil;
argp0++;
file = progarg[0];
if(strlen(elem) >= sizeof progelem)
error(Ebadexec);
strcpy(progelem, elem);
progarg[0] = progelem;
poperror();
cclose(tc);
}
data = l2be(exec.data);
bss = l2be(exec.bss);
align--;
t = (UTZERO+sizeof(Exec)+text+align) & ~align;
align = BY2PG-1;
d = (t + data + align) & ~align;
bssend = t + data + bss;
b = (bssend + align) & ~align;
if(t >= (USTKTOP-USTKSIZE) || d >= (USTKTOP-USTKSIZE) || b >= (USTKTOP-USTKSIZE))
error(Ebadexec);
/*
* Args: pass 1: count
*/
nbytes = sizeof(Tos); /* hole for profiling clock at top of stack (and more) */
nargs = 0;
if(indir){
argp = progarg;
while(*argp != nil){
a = *argp++;
nbytes += strlen(a) + 1;
nargs++;
}
}
argp = argp0;
while(*argp != nil){
a = *argp++;
if(((uintptr)argp&(BY2PG-1)) < BY2WD)
validaddr((uintptr)argp, BY2WD, 0);
validaddr((uintptr)a, 1, 0);
e = vmemchr(a, 0, USTKSIZE);
if(e == nil)
error(Ebadarg);
nbytes += (e - a) + 1;
if(nbytes >= USTKSIZE)
error(Enovmem);
nargs++;
}
ssize = BY2WD*(nargs+1) + ((nbytes+(BY2WD-1)) & ~(BY2WD-1));
/*
* 8-byte align SP for those (e.g. sparc) that need it.
* execregs() will subtract another 4 bytes for argc.
*/
if(BY2WD == 4 && (ssize+4) & 7)
ssize += 4;
if(PGROUND(ssize) >= USTKSIZE)
error(Enovmem);
/*
* Build the stack segment, putting it in kernel virtual for the moment
*/
qlock(&up->seglock);
if(waserror()){
qunlock(&up->seglock);
nexterror();
}
s = up->seg[SSEG];
do {
tstk = s->base;
if(tstk <= USTKSIZE)
error(Enovmem);
} while((s = isoverlap(up, tstk-USTKSIZE, USTKSIZE)) != nil);
up->seg[ESEG] = newseg(SG_STACK, tstk-USTKSIZE, USTKSIZE/BY2PG);
/*
* Args: pass 2: assemble; the pages will be faulted in
*/
tos = (Tos*)(tstk - sizeof(Tos));
tos->cyclefreq = m->cyclefreq;
tos->kcycles = 0;
tos->pcycles = 0;
tos->clock = 0;
argv = (char**)(tstk - ssize);
charp = (char*)(tstk - nbytes);
if(indir)
argp = progarg;
else
argp = argp0;
for(i=0; i<nargs; i++){
if(indir && *argp==nil) {
indir = 0;
argp = argp0;
}
*argv++ = charp + (USTKTOP-tstk);
a = *argp++;
if(indir)
e = strchr(a, 0);
else {
validaddr((uintptr)a, 1, 0);
e = vmemchr(a, 0, (char*)tstk - charp);
if(e == nil)
error(Ebadarg);
}
n = (e - a) + 1;
memmove(charp, a, n);
charp += n;
}
/* copy args; easiest from new process's stack */
a = (char*)(tstk - nbytes);
n = charp - a;
if(n > 128) /* don't waste too much space on huge arg lists */
n = 128;
args = smalloc(n);
memmove(args, a, n);
if(n>0 && args[n-1]!='\0'){
/* make sure last arg is NUL-terminated */
/* put NUL at UTF-8 character boundary */
for(i=n-1; i>0; --i)
if(fullrune(args+i, n-i))
break;
args[i] = 0;
n = i+1;
}
/*
* Committed.
* Free old memory.
* Special segments are maintained across exec
*/
for(i = SSEG; i <= BSEG; i++) {
putseg(up->seg[i]);
/* prevent a second free if we have an error */
up->seg[i] = nil;
}
for(i = ESEG+1; i < NSEG; i++) {
s = up->seg[i];
if(s != nil && (s->type&SG_CEXEC) != 0) {
putseg(s);
up->seg[i] = nil;
}
}
/*
* Close on exec
*/
if((f = up->fgrp) != nil) {
for(i=0; i<=f->maxfd; i++)
fdclose(i, CCEXEC);
}
/* Text. Shared. Attaches to cache image if possible */
/* attachimage returns a locked cache image */
img = attachimage(SG_TEXT|SG_RONLY, tc, UTZERO, (t-UTZERO)>>PGSHIFT);
ts = img->s;
up->seg[TSEG] = ts;
ts->flushme = 1;
ts->fstart = 0;
ts->flen = sizeof(Exec)+text;
unlock(img);
/* Data. Shared. */
s = newseg(SG_DATA, t, (d-t)>>PGSHIFT);
up->seg[DSEG] = s;
/* Attached by hand */
incref(img);
s->image = img;
s->fstart = ts->fstart+ts->flen;
s->flen = data;
/* BSS. Zero fill on demand */
up->seg[BSEG] = newseg(SG_BSS, d, (b-d)>>PGSHIFT);
/*
* Move the stack
*/
s = up->seg[ESEG];
up->seg[ESEG] = nil;
s->base = USTKTOP-USTKSIZE;
s->top = USTKTOP;
relocateseg(s, USTKTOP-tstk);
up->seg[SSEG] = s;
qunlock(&up->seglock);
poperror(); /* seglock */
/*
* '/' processes are higher priority (hack to make /ip more responsive).
*/
if(devtab[tc->type]->dc == L'/')
up->basepri = PriRoot;
up->priority = up->basepri;
poperror(); /* tc */
cclose(tc);
poperror(); /* file0 */
free(file0);
qlock(&up->debug);
free(up->text);
up->text = elem;
free(up->args);
up->args = args;
up->nargs = n;
up->setargs = 0;
up->nnote = 0;
up->notify = 0;
up->notified = 0;
up->privatemem = 0;
up->noswap = 0;
procsetup(up);
qunlock(&up->debug);
/*
* At this point, the mmu contains info about the old address
* space and needs to be flushed
*/
flushmmu();
if(up->hang)
up->procctl = Proc_stopme;
return execregs(entry, ssize, nargs);
}
long
sysexec(ulong *arg)
{
Segment *s, *ts;
ulong t, d, b;
int i;
Chan *tc;
char **argv, **argp;
char *a, *charp, *args, *file, *file0;
char *progarg[sizeof(Exec)/2+1], *elem, progelem[64];
ulong ssize, spage, nargs, nbytes, n, bssend;
int indir;
Exec exec;
char line[sizeof(Exec)];
Fgrp *f;
Image *img;
ulong magic, text, entry, data, bss;
Tos *tos;
indir = 0;
elem = nil;
validaddr(arg[0], 1, 0);
file0 = validnamedup((char*)arg[0], 1);
if(waserror()){
free(file0);
free(elem);
nexterror();
}
file = file0;
for(;;){
tc = namec(file, Aopen, OEXEC, 0);
if(waserror()){
cclose(tc);
nexterror();
}
if(!indir)
kstrdup(&elem, up->genbuf);
n = devtab[tc->type]->read(tc, &exec, sizeof(Exec), 0);
if(n < 2)
error(Ebadexec);
magic = l2be(exec.magic);
text = l2be(exec.text);
entry = l2be(exec.entry);
if(n==sizeof(Exec) && (magic == AOUT_MAGIC)){
if(text >= USTKTOP-UTZERO
|| entry < UTZERO+sizeof(Exec)
|| entry >= UTZERO+sizeof(Exec)+text)
error(Ebadexec);
break; /* for binary */
}
/*
* Process #! /bin/sh args ...
*/
memmove(line, &exec, sizeof(Exec));
if(indir || line[0]!='#' || line[1]!='!')
error(Ebadexec);
n = shargs(line, n, progarg);
if(n == 0)
error(Ebadexec);
indir = 1;
/*
* First arg becomes complete file name
*/
progarg[n++] = file;
progarg[n] = 0;
validaddr(arg[1], BY2WD, 1);
arg[1] += BY2WD;
file = progarg[0];
if(strlen(elem) >= sizeof progelem)
error(Ebadexec);
strcpy(progelem, elem);
progarg[0] = progelem;
poperror();
cclose(tc);
}
data = l2be(exec.data);
bss = l2be(exec.bss);
t = (UTZERO+sizeof(Exec)+text+(BY2PG-1)) & ~(BY2PG-1);
d = (t + data + (BY2PG-1)) & ~(BY2PG-1);
bssend = t + data + bss;
b = (bssend + (BY2PG-1)) & ~(BY2PG-1);
if(t >= KZERO || d >= KZERO || b >= KZERO)
error(Ebadexec);
/*
* Args: pass 1: count
*/
nbytes = sizeof(Tos); /* hole for profiling clock at top of stack (and more) */
nargs = 0;
if(indir){
argp = progarg;
while(*argp){
a = *argp++;
nbytes += strlen(a) + 1;
nargs++;
}
}
evenaddr(arg[1]);
argp = (char**)arg[1];
validaddr((ulong)argp, BY2WD, 0);
while(*argp){
a = *argp++;
if(((ulong)argp&(BY2PG-1)) < BY2WD)
validaddr((ulong)argp, BY2WD, 0);
validaddr((ulong)a, 1, 0);
nbytes += ((char*)vmemchr(a, 0, 0x7FFFFFFF) - a) + 1;
nargs++;
}
ssize = BY2WD*(nargs+1) + ((nbytes+(BY2WD-1)) & ~(BY2WD-1));
/*
* 8-byte align SP for those (e.g. sparc) that need it.
* execregs() will subtract another 4 bytes for argc.
*/
if((ssize+4) & 7)
ssize += 4;
spage = (ssize+(BY2PG-1)) >> PGSHIFT;
/*
* Build the stack segment, putting it in kernel virtual for the moment
*/
if(spage > TSTKSIZ)
error(Enovmem);
qlock(&up->seglock);
if(waserror()){
qunlock(&up->seglock);
nexterror();
}
up->seg[ESEG] = newseg(SG_STACK, TSTKTOP-USTKSIZE, USTKSIZE/BY2PG);
/*
* Args: pass 2: assemble; the pages will be faulted in
*/
tos = (Tos*)(TSTKTOP - sizeof(Tos));
tos->cyclefreq = m->cyclefreq;
cycles((uvlong*)&tos->pcycles);
tos->pcycles = -tos->pcycles;
tos->kcycles = tos->pcycles;
tos->clock = 0;
argv = (char**)(TSTKTOP - ssize);
charp = (char*)(TSTKTOP - nbytes);
args = charp;
if(indir)
argp = progarg;
else
argp = (char**)arg[1];
for(i=0; i<nargs; i++){
if(indir && *argp==0) {
indir = 0;
argp = (char**)arg[1];
}
*argv++ = charp + (USTKTOP-TSTKTOP);
n = strlen(*argp) + 1;
memmove(charp, *argp++, n);
charp += n;
}
free(file0);
free(up->text);
up->text = elem;
elem = nil; /* so waserror() won't free elem */
USED(elem);
/* copy args; easiest from new process's stack */
n = charp - args;
if(n > 128) /* don't waste too much space on huge arg lists */
n = 128;
a = up->args;
up->args = nil;
free(a);
up->args = smalloc(n);
memmove(up->args, args, n);
if(n>0 && up->args[n-1]!='\0'){
/* make sure last arg is NUL-terminated */
/* put NUL at UTF-8 character boundary */
for(i=n-1; i>0; --i)
if(fullrune(up->args+i, n-i))
break;
up->args[i] = 0;
n = i+1;
}
up->nargs = n;
/*
* Committed.
* Free old memory.
* Special segments are maintained across exec
*/
for(i = SSEG; i <= BSEG; i++) {
putseg(up->seg[i]);
/* prevent a second free if we have an error */
up->seg[i] = 0;
}
for(i = BSEG+1; i < NSEG; i++) {
s = up->seg[i];
if(s != 0 && (s->type&SG_CEXEC)) {
putseg(s);
up->seg[i] = 0;
}
}
/*
* Close on exec
*/
f = up->fgrp;
for(i=0; i<=f->maxfd; i++)
fdclose(i, CCEXEC);
/* Text. Shared. Attaches to cache image if possible */
/* attachimage returns a locked cache image */
img = attachimage(SG_TEXT|SG_RONLY, tc, UTZERO, (t-UTZERO)>>PGSHIFT);
ts = img->s;
up->seg[TSEG] = ts;
ts->flushme = 1;
ts->fstart = 0;
ts->flen = sizeof(Exec)+text;
unlock(img);
/* Data. Shared. */
s = newseg(SG_DATA, t, (d-t)>>PGSHIFT);
up->seg[DSEG] = s;
/* Attached by hand */
incref(img);
s->image = img;
s->fstart = ts->fstart+ts->flen;
s->flen = data;
/* BSS. Zero fill on demand */
up->seg[BSEG] = newseg(SG_BSS, d, (b-d)>>PGSHIFT);
/*
* Move the stack
*/
s = up->seg[ESEG];
up->seg[ESEG] = 0;
up->seg[SSEG] = s;
qunlock(&up->seglock);
poperror(); /* seglock */
poperror(); /* elem */
s->base = USTKTOP-USTKSIZE;
s->top = USTKTOP;
relocateseg(s, USTKTOP-TSTKTOP);
/*
* '/' processes are higher priority (hack to make /ip more responsive).
*/
if(devtab[tc->type]->dc == L'/')
up->basepri = PriRoot;
up->priority = up->basepri;
poperror();
cclose(tc);
/*
* At this point, the mmu contains info about the old address
* space and needs to be flushed
*/
flushmmu();
qlock(&up->debug);
up->nnote = 0;
up->notify = 0;
up->notified = 0;
up->privatemem = 0;
procsetup(up);
qunlock(&up->debug);
if(up->hang)
up->procctl = Proc_stopme;
return execregs(entry, ssize, nargs);
}