Bug 210540 – Fix an integer overflow in WebCrypto AES-CTR Mac implementation, which may detect a false loop

• Status: RESOLVED
• Resolution: FIXED
• Component: WebCore Misc.

[Tomoki Imai]

(1 << counterLength) causes an integer overflow, and the undefined behavior.
The longest valid counterLength on 64 bit machine is 63, and the literal 1 is considered as 32-bit signed integer.
Left shifting 1 beyond or to sign-bit is undefined behavior in C++ spec.

This issue is originally found in https://bugs.webkit.org/show_bug.cgi?id=208186#c2

Unfortunately, I don't own Mac so I can't test this locally.
For OpenSSL implementation the added test case could catch this issue.

[Tomoki Imai]

Created attachment 396514 [details]
patch

[Jiewen Tan]

Comment on attachment 396514 [details]
patch

View in context: https://bugs.webkit.org/attachment.cgi?id=396514&action=review

Good catch! Please address my comments below.

> Source/WebCore/ChangeLog:6
> +        (1 << counterLength) causes an integer overflow, and the undefined behavior.

Maybe you could reference here: https://en.cppreference.com/w/cpp/language/integer_literal.

> Source/WebCore/crypto/mac/CryptoAlgorithmAES_CTRMac.cpp:48
> +    if (counterLength < sizeof(size_t) * 8 && numberOfBlocks > ((size_t)1 << counterLength))

((size_t)1 => 1ull

[Radar WebKit Bug Importer]

<rdar://problem/61857969>

[Tomoki Imai]

(In reply to Jiewen Tan from comment #2)
> Comment on attachment 396514 [details]
> patch
> 
> View in context:
> https://bugs.webkit.org/attachment.cgi?id=396514&action=review
> 
> Good catch! Please address my comments below.

Thanks for your review!

> 
> > Source/WebCore/ChangeLog:6
> > +        (1 << counterLength) causes an integer overflow, and the undefined behavior.
> 
> Maybe you could reference here:
> https://en.cppreference.com/w/cpp/language/integer_literal.

Will do.

> 
> > Source/WebCore/crypto/mac/CryptoAlgorithmAES_CTRMac.cpp:48
> > +    if (counterLength < sizeof(size_t) * 8 && numberOfBlocks > ((size_t)1 << counterLength))
> 
> ((size_t)1 => 1ull

I used ((size_t) 1) rather than 1ull, because I thought using size_t type makes more sense than "unsigned long long".
- "counterLength < sizeof(size_t) * 8" checks whether size_t has enough size to store (1 << counterLength)
- In 32-bit environment, 1ull, which is 64-bit or larger, is a bit overkill, because counterLength < 31 in 32-bit environment.

How do you think about it?
Of course, 1ull should work fine both on 32-bit and 64-bit environment, so I'm fine to change it to 1ull.

[Jiewen Tan]

Comment on attachment 396514 [details]
patch

View in context: https://bugs.webkit.org/attachment.cgi?id=396514&action=review

>>> Source/WebCore/crypto/mac/CryptoAlgorithmAES_CTRMac.cpp:48
>>> +    if (counterLength < sizeof(size_t) * 8 && numberOfBlocks > ((size_t)1 << counterLength))
>> 
>> ((size_t)1 => 1ull
> 
> I used ((size_t) 1) rather than 1ull, because I thought using size_t type makes more sense than "unsigned long long".
> - "counterLength < sizeof(size_t) * 8" checks whether size_t has enough size to store (1 << counterLength)
> - In 32-bit environment, 1ull, which is 64-bit or larger, is a bit overkill, because counterLength < 31 in 32-bit environment.
> 
> How do you think about it?
> Of course, 1ull should work fine both on 32-bit and 64-bit environment, so I'm fine to change it to 1ull.

I was not aware we are still supporting 32 bit machines. Doing a C style cast is not appropriate in WebKit, please change it to static_cast<size_t>(1).

[Tomoki Imai]

Created attachment 396737 [details]
patch

Reflected the review comments

[Tomoki Imai]

(In reply to Jiewen Tan from comment #5)
> Comment on attachment 396514 [details]
> patch
> 
> View in context:
> https://bugs.webkit.org/attachment.cgi?id=396514&action=review
> 
> >>> Source/WebCore/crypto/mac/CryptoAlgorithmAES_CTRMac.cpp:48
> >>> +    if (counterLength < sizeof(size_t) * 8 && numberOfBlocks > ((size_t)1 << counterLength))
> >> 
> >> ((size_t)1 => 1ull
> > 
> > I used ((size_t) 1) rather than 1ull, because I thought using size_t type makes more sense than "unsigned long long".
> > - "counterLength < sizeof(size_t) * 8" checks whether size_t has enough size to store (1 << counterLength)
> > - In 32-bit environment, 1ull, which is 64-bit or larger, is a bit overkill, because counterLength < 31 in 32-bit environment.
> > 
> > How do you think about it?
> > Of course, 1ull should work fine both on 32-bit and 64-bit environment, so I'm fine to change it to 1ull.
> 
> I was not aware we are still supporting 32 bit machines. Doing a C style
> cast is not appropriate in WebKit, please change it to
> static_cast<size_t>(1).

I changed it to static_cast<size_t>(1).
I don't know about Mac implementation, but the other code seems to try to support 32-bit by using size_t.

> 
> > Source/WebCore/ChangeLog:6
> > +        (1 << counterLength) causes an integer overflow, and the undefined behavior.
> 
> Maybe you could reference here:
> https://en.cppreference.com/w/cpp/language/integer_literal.

I added 
- https://en.cppreference.com/w/cpp/language/integer_literal and 
- https://en.cppreference.com/w/cpp/language/operator_arithmetic#Bitwise_shift_operators 
as references.

[Jiewen Tan]

(In reply to Tomoki Imai from comment #7)
> (In reply to Jiewen Tan from comment #5)
> > Comment on attachment 396514 [details]
> > patch
> > 
> > View in context:
> > https://bugs.webkit.org/attachment.cgi?id=396514&action=review
> > 
> > >>> Source/WebCore/crypto/mac/CryptoAlgorithmAES_CTRMac.cpp:48
> > >>> +    if (counterLength < sizeof(size_t) * 8 && numberOfBlocks > ((size_t)1 << counterLength))
> > >> 
> > >> ((size_t)1 => 1ull
> > > 
> > > I used ((size_t) 1) rather than 1ull, because I thought using size_t type makes more sense than "unsigned long long".
> > > - "counterLength < sizeof(size_t) * 8" checks whether size_t has enough size to store (1 << counterLength)
> > > - In 32-bit environment, 1ull, which is 64-bit or larger, is a bit overkill, because counterLength < 31 in 32-bit environment.
> > > 
> > > How do you think about it?
> > > Of course, 1ull should work fine both on 32-bit and 64-bit environment, so I'm fine to change it to 1ull.
> > 
> > I was not aware we are still supporting 32 bit machines. Doing a C style
> > cast is not appropriate in WebKit, please change it to
> > static_cast<size_t>(1).
> 
> I changed it to static_cast<size_t>(1).
> I don't know about Mac implementation, but the other code seems to try to
> support 32-bit by using size_t.
> 

That was true 4 years ago for Apple ports but not now. Anyway, I think it is better to keep the consistency.

[Jiewen Tan]

Comment on attachment 396737 [details]
patch

r=me.

[EWS]

Committed r260238: <https://trac.webkit.org/changeset/260238>

All reviewed patches have been landed. Closing bug and clearing flags on attachment 396737 [details].