Produits Entr'ouvert » Lasso

Hello,

On some systems, such as SUSE Linux Enterprise 15 SP5, we're seeing that lasso-2.6.1 and lasso-2.8.2 fail to build, as its test suite hits a segmentation fault .
Through careful analysis, we've discovered the exact instruction that triggers the segfault.

0x7ffff7c84b00 <xmlDictOwns>    test   %rsi,%rsi                                                                                                                                                 
    0x7ffff7c84b03 <xmlDictOwns+3>  sete   %cl                                                                                                                                                       
    0x7ffff7c84b06 <xmlDictOwns+6>  test   %rdi,%rdi                                                                                                                                                 
    0x7ffff7c84b09 <xmlDictOwns+9>  je     0x7ffff7c84b4e <xmlDictOwns+78>                                                                                                                           
    0x7ffff7c84b0b <xmlDictOwns+11> test   %cl,%cl                                                                                                                                                   
    0x7ffff7c84b0d <xmlDictOwns+13> jne    0x7ffff7c84b4e <xmlDictOwns+78>                                                                                                                           
--> 0x7ffff7c84b0f <xmlDictOwns+15> mov    0x20(%rdi),%rax                                                                                                                                           
    0x7ffff7c84b13 <xmlDictOwns+19> test   %rax,%rax                                                                                                                                                 
    0x7ffff7c84b16 <xmlDictOwns+22> je     0x7ffff7c84b37 <xmlDictOwns+55>                                                                                                                           
    0x7ffff7c84b18 <xmlDictOwns+24> nopl   0x0(%rax,%rax,1)                                                                                                                                          
    0x7ffff7c84b20 <xmlDictOwns+32> lea    0x28(%rax),%rdx                                                                                                                                           
    0x7ffff7c84b24 <xmlDictOwns+36> cmp    %rdx,%rsi                                                                                                                                                 
    0x7ffff7c84b27 <xmlDictOwns+39> jb     0x7ffff7c84b2f <xmlDictOwns+47>                                                                                                                           
    0x7ffff7c84b29 <xmlDictOwns+41> cmp    %rsi,0x8(%rax)                                                                                                                                            
    0x7ffff7c84b2d <xmlDictOwns+45> jae    0x7ffff7c84b48 <xmlDictOwns+72>

This happens when running the following test: test16_test_get_issuer_fn (basic_tests.c:1019) .
This test contains a loop, and it doesn't happen the first time the instruction is executed. It takes over 30 thousand attempts, but then it always causes a segfault.
Various other tests, typically login-related ones, also trigger this issue.

On SLE 15 SP5 specifically, we're using libxml2 version 2.10.3.

I'm including a patch we've written that seems to mitigate the issue, hopefully you can have a look and see how it might mitigate this problem, and maybe other people are also being affected.

We have a build of lasso-2.8.2 running here, with the patch:
https://build.opensuse.org/package/show/home:pgajdos/lasso

Thanks.