Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

gh-92888: Fix memoryview bad __index__ use after free #92946

Open
wants to merge 12 commits into
base: main
Choose a base branch
from

Conversation

Fidget-Spinner
Copy link
Member

@Fidget-Spinner Fidget-Spinner commented May 19, 2022

Co-Authored-By: chilaxan 35645806+chilaxan@users.noreply.github.com
Fixes #92888

Co-Authored-By: chilaxan <35645806+chilaxan@users.noreply.github.com>
@serhiy-storchaka
Copy link
Member

@serhiy-storchaka serhiy-storchaka commented May 23, 2022

I have other approach, but your approach is more compatible, so it is safer to use it backports. I'll experiment with other approach in the developing branch only.

You can use the following test as a base:

    def test_use_released_memory(self):
        size = 128
        def release():
            m.release()
            nonlocal ba
            ba = bytearray(size)
        class MyIndex:
            def __index__(self):
                release()
                return 4
        class MyFloat:
            def __float__(self):
                release()
                return 4.25
        class MyBool:
            def __bool__(self):
                release()
                return True

        ba = None
        m = memoryview(bytearray(b'\xff'*size))
        with self.assertRaises(ValueError):
            m[MyIndex()]

        ba = None
        m = memoryview(bytearray(b'\xff'*size))
        self.assertEqual(list(m[:MyIndex()]), [255] * 4)

        ba = None
        m = memoryview(bytearray(b'\xff'*size))
        self.assertEqual(list(m[MyIndex():8]), [255] * 4)

        ba = None
        m = memoryview(bytearray(b'\xff'*size))
        m[MyIndex()] = 42
        self.assertEqual(ba[:8], b'\0'*8)

        ba = None
        m = memoryview(bytearray(b'\xff'*size))
        m[:MyIndex()] = b'spam'
        self.assertEqual(ba[:8], b'\0'*8)

        ba = None
        m = memoryview(bytearray(b'\xff'*size))
        m[MyIndex():8] = b'spam'
        self.assertEqual(ba[:8], b'\0'*8)

        ba = None
        m = memoryview(bytearray(b'\xff'*size))
        m[0] = MyIndex()
        self.assertEqual(ba[:8], b'\0'*8)

        for fmt in 'bhilqnBHILQN':
            ba = None
            m = memoryview(bytearray(b'\xff'*size)).cast(fmt)
            m[0] = MyIndex()
            self.assertEqual(ba[:8], b'\0'*8)

        for fmt in 'fd':
            ba = None
            m = memoryview(bytearray(b'\xff'*size)).cast(fmt)
            m[0] = MyFloat()
            self.assertEqual(ba[:8], b'\0'*8)

        ba = None
        m = memoryview(bytearray(b'\xff'*size)).cast('?')
        m[0] = MyBool()
        self.assertEqual(ba[:8], b'\0'*8)

@Fidget-Spinner
Copy link
Member Author

@Fidget-Spinner Fidget-Spinner commented May 23, 2022

@serhiy-storchaka thanks taking the time to review and writing those really comprehensive tests. One part of your test fails on my branch: specifically m[MyIndex()] = 42 raises ValueError. That's supposed to happen right? (Actually, most of the asserts seem to fail on my branch :()

@serhiy-storchaka
Copy link
Member

@serhiy-storchaka serhiy-storchaka commented May 23, 2022

Yes, it is expected that some lines raise exceptions with your changes. Just add assertRaises() if needed.

Co-Authored-By: Serhiy Storchaka <3659035+serhiy-storchaka@users.noreply.github.com>
@Fidget-Spinner
Copy link
Member Author

@Fidget-Spinner Fidget-Spinner commented May 23, 2022

Great, thanks to your tests I found that my fix didn't cover slices. It now does.

Copy link
Member

@serhiy-storchaka serhiy-storchaka left a comment

Yet few tests.

Lib/test/test_memoryview.py Show resolved Hide resolved
Lib/test/test_memoryview.py Show resolved Hide resolved
Fidget-Spinner and others added 2 commits May 23, 2022
Co-Authored-By: Serhiy Storchaka <3659035+serhiy-storchaka@users.noreply.github.com>
Copy link
Member

@serhiy-storchaka serhiy-storchaka left a comment

👍

Lib/test/test_memoryview.py Outdated Show resolved Hide resolved
Lib/test/test_memoryview.py Outdated Show resolved Hide resolved
Lib/test/test_memoryview.py Outdated Show resolved Hide resolved
Lib/test/test_memoryview.py Outdated Show resolved Hide resolved
@serhiy-storchaka
Copy link
Member

@serhiy-storchaka serhiy-storchaka commented May 23, 2022

I am sorry for being so pedantic. 🤦‍♂️

@Fidget-Spinner
Copy link
Member Author

@Fidget-Spinner Fidget-Spinner commented May 23, 2022

I am sorry for being so pedantic. man_facepalming

I didn't feel that you were. Those are errors I ought to have caught anyways. Thanks for the detailed review as always.

Copy link
Member

@vstinner vstinner left a comment

I like your approach to fix memory_ass_sub(). Here is a first review.

Objects/memoryobject.c Outdated Show resolved Hide resolved
{
CHECK_RELEASED_INT(self); /* See gh-92888 for why we need this here */
Copy link
Member

@vstinner vstinner May 23, 2022

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Can you move the check after the equiv_structure() test?

Copy link
Member

@serhiy-storchaka serhiy-storchaka May 25, 2022

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Would not access to Py_buffer of the released object cause reading after free?

Objects/memoryobject.c Outdated Show resolved Hide resolved
@@ -0,0 +1,2 @@
Fix ``memoryview`` use after free when accessing the backing buffer in certain cases.
Copy link
Member

@vstinner vstinner May 23, 2022

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I propose to mention more explicitly that the protection is about released views:

If a memoryview is released while getting or setting an item, raise an exception.
For example, converting an object to an index can execute arbitrary Python
code which can indirectly release the memoryview.

Copy link
Member

@serhiy-storchaka serhiy-storchaka May 25, 2022

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Not always an exception is raised.

The bug was in reading or wring the freed memory. Now it is prevented -- you either get an exception or free the memory after reading. @Fidget-Spinner's description is more correct.

I am going to address such inconsistency in a separate issue.

Lib/test/test_memoryview.py Show resolved Hide resolved
def release():
m.release()
nonlocal ba
ba = bytearray(size)
Copy link
Member

@vstinner vstinner May 23, 2022

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

That's useless, no?

Copy link
Member Author

@Fidget-Spinner Fidget-Spinner May 29, 2022

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

No, we need it for tests below that tests indexing into ba[].

Copy link
Member

@serhiy-storchaka serhiy-storchaka May 29, 2022

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We allocate a bytearray of the same size as the bytearray just released in memoryview in hope that it will be allocated at the same memory. It helps to check that we do nor read/write a freed memory.

return True

ba = None
m = memoryview(bytearray(b'\xff'*size))
Copy link
Member

@vstinner vstinner May 23, 2022

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

In my PR, I tried to make the code more generic to test more cases: https://github.com/python/cpython/pull/93127/files#diff-d41c6bb40a1e03fea5a20d15c4077413e0ddde65651147922b625b03a66a2f16R399:

        tp = self.rw_type
        b = self.rw_type(self._source)
        view = self._view(b)

ba = None
m = memoryview(bytearray(b'\xff'*size))
with self.assertRaises(ValueError):
m[MyIndex()]
Copy link
Member

@vstinner vstinner May 23, 2022

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This test is very long. Can you try to factorize similar code and use loop with subTest(), and put pack operations in one test method and unpack in another test method?

Copy link
Member

@serhiy-storchaka serhiy-storchaka May 25, 2022

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Then we will need to duplicate the definitions of internal classes.

The tested code is so different, that it is difficult to use a loop. And I think that the result will be more complicated.

m = memoryview(bytearray(b'\xff'*size))
with self.assertRaises(ValueError):
m[MyIndex()]

ba = None
Copy link
Member

@serhiy-storchaka serhiy-storchaka May 30, 2022

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It is necessary. The old bytearray should be deallocated before creating a new bytearray, so all bytearrays will be allocated at the same place.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
awaiting merge needs backport to 3.10 needs backport to 3.11 type-crash
Projects
None yet
Development

Successfully merging this pull request may close these issues.

4 participants