Fix issue #66: ignore 0th section header when sorting, don't overwrite NOBITS #149
Conversation
… and not overwriting NOBITS entries.
|
Can this be merged? I can't patch hardly any recently-built static go binaries without this. On the other hand, I'm having complete success with this PR's branch of Thanks! |
|
same here, i could really use this! |
|
Worked for me checking out the branch and building locally in at least one case. > file go
go: ELF 64-bit LSB executable, x86-64, version 1 (SYSV), dynamically linked, interpreter
/lib64/ld-linux-x86-64.so.2, with debug_info, not stripped
> /usr/local/bin/patchelf go --set-interpreter /nix/store/j2yzfcd4wli8lj5i6yvyg6h1sbvjr55i-glibc-2.27/lib/ld-2.27.so
> file go
go: ELF 64-bit LSB executable, x86-64, version 1 (SYSV), dynamically linked, interpreter
/nix/store/j2yzfcd4wli8lj5i6yvyg6h1sbvjr55i-glibc-2.27/lib/ld-2.27.so, with debug_info, not stripped |
|
Thank you @ezquat, this works for my test cases as well! Can we please get this merged? |
|
+1 I'm unable to build the binaries for the pulumi CLI which is written in Go. |
|
@edolstra, any thoughts on this? |
|
This patch worked for me as well. Any chance on getting this merged eventually? |
|
@puffnfresh: could you explain why this hasn't been merged yet? |
He has approved the changes, it's @edolstra we are waiting for. |
|
@edolstra @puffnfresh |
|
Released in 0.10 for anyone wondering |
|
Great! Thanks for the heads-up. |
To recount the problem symptoms: Given an executable with certain properties, including a section at offset 0, patchelf fails to set the interpreter. (With patchelf 0.8 you get an assertion failure, (off_t) rdi(hdr->e_shoff) >= startOffset'. With patchelf 0.9 you get cannot find section, which is referring to the section with the empty name, which is some kind of dummy placeholder in ELF files, at section index 0.)
Here's my understanding of the bug: Most of the code in patchelf ignores the 0th entry in the section header table, but there is a call to C++
sort()which can move the entry at index 0, in the case where some other entry also has offset 0 (which is the case for Go binaries, a .tbss section has offset 0). It is easy to skip entry 0 in the sort. Then another problem kicks in, which is that patchelf tries to overwrite some bytes (with 'X's) at that offset 0, even though the.tbsssection is marked as NOBITS and thus has no content.