Re: Report 2 UBSan bugs found by an automatic tool

Hi James,

On 6/30/21 7:23 PM, He Jingxuan wrote:
> Dear findutils developers,
> 
> We tested findutils with an automatic tool (based on the symbolic execution tool KLEE). 5 test cases triggering UBSan errors were generated. We manually checked those test cases and filtered out benign cases. Finally, we identified and report 2 cases that could trigger bugs. Below is the information for reproducing the bugs.
> 
> - findutils version: 4.7.0
> - operating system: Ubuntu 16.04.7
> - compiler: clang version 6.0.0-1ubuntu2~16.04.1 (tags/RELEASE_600/final)
> - compilation commands:
>    mkdir obj
>    cd obj
>    CC=clang CFLAGS="-g -O1 -Xclang -disable-llvm-passes -D__NO_STRING_INLINES -D_FORTIFY_SOURCE=0 -U__OPTIMIZE__ -fsanitize=signed-integer-overflow -fsanitize=unsigned-integer-overflow -fsanitize=shift -fsanitize=bounds -fsanitize=pointer-overflow -fsanitize=null" ../configure --disable-nls --disable-largefile --disable-threads --without-selinux
>    make
> - inputs: the attached file contains the input file A of the second bug.

Both issues are in CBO code - find's cost-based-optimizer:

> bug 1 command: find -H - -delete
> relevant error message: ../../find/tree.c:538:23: runtime error: member access within null pointer of type 'struct predicate’

This is in consider_arm_swap().

> bug 2 command: find -H -neweraa A
> relevant error message: ../../find/parser.c:698:48: runtime error: signed integer overflow: 1624986826 - -9223372036854775808 cannot be represented in type ‘long'
> Note: I think this bug depends on the timestamp and is triggered only sometimes.

This is in estimate_timestamp_success_rate().

Although I think the first one is a false positive, because the code always ensures
that the left arm is filled, - and considering the other open bugs related to CBO -,
maybe it's time to remove that code?
WDYT?

Have a nice day,
Berny