Fixes for MIPS binaries #180
Conversation
|
UPD: replaced |
|
@edolstra: are you ok with this pull request? Unavailability of patchelf on MIPS is leading to a bunch of issues downstream in Debian. To expedite a fix, we could take this patch and implement it downstream until a newer version of patchelf is released. However, I'm hesitant to do that if you have a fundamental disagreement with how this solves #82. Thanks! |
|
@darealshinji would you review/test this? |
|
I can confirm this makes tests pass on debian mipsel, on a Rhino Labs UTM8 |
|
It would be great if someone stepped up to be maintainer of MIPS as per #199 |
|
UPD: rebased onto the current master with a couple of minor cosmetic changes. |
src/patchelf.cc
Outdated
| else if (d_tag == DT_MIPS_RLD_MAP_REL) { | ||
| /* the MIPS_RLD_MAP_REL tag stores the offset to the debug | ||
| pointer, relative to the address of the tag */ | ||
| debug("Updating DT_MIPS_RLD_MAP_REL"); |
There was a problem hiding this comment.
nitpick. we don't do any debug printing for other sections:
| debug("Updating DT_MIPS_RLD_MAP_REL"); |
There was a problem hiding this comment.
Ok, I'm going to drop this.
| @@ -1129,6 +1141,14 @@ void ElfFile<ElfFileParamNames>::rewriteHeaders(Elf_Addr phdrAddress) | |||
| dyn->d_un.d_ptr = findSection(".gnu.version_r").sh_addr; | |||
| else if (d_tag == DT_VERSYM) | |||
| dyn->d_un.d_ptr = findSection(".gnu.version").sh_addr; | |||
| else if (d_tag == DT_MIPS_RLD_MAP_REL) { | |||
There was a problem hiding this comment.
Could the .rld_map section be stripped just like DT_JMPREL or DT_REL?
There was a problem hiding this comment.
What I am asking is, is it possible to that someone has binaries where this section was stripped.
I fixed a similar issue here: #296
There was a problem hiding this comment.
@Mic92, thank you for the review.
I understood what you're asking, and I need to give it some thought and run a couple of experiments.
I used to believe that elf with DT_MIPS_RLD_MAP_REL in dynamic section but without .rld_map is simply broken, but maybe patchelf needs to handle such files. If that's the case, I'll update the code here and add a test for that.
There was a problem hiding this comment.
Well, elf file with DT_MIPS_RLD_MAP_REL in dynamic section but without .rld_map is indeed broken: dynamic loader (I've checked glibc 2.32) just writes the debug pointer to the address it calculates from DT_MIPS_RLD_MAP_REL entry. I guess no tool produces such binaries.
But when I just removed .rtld_map section with objcopy, the resulting binaries mostly worked -- out of sheer luck probably, but still most of them did not crash.
So I rewrote the code to handle such case. Removing DT_MIPS_RLD_MAP_REL might have been a cleaner solution, but patchelf does not remove any dynamic section entries yet and I don't think supporting binaries that were broken in the fist place worth the burden of starting to do so.
And @Mic92, thank you for pointing this case out.
|
I also gained now merge rights for this repo. |
`patchelf --set-rpath` corrupted executables on mips32el: the dynamic liker crushed with Segmentation fault when loading any executable with RPATH added that way. The problem was around the MIPS-specific mechanism of setting up the debug map pointer. When DT_MIPS_RLD_MAP_REL entry in the dynamic section is present, it holds the relative address of __RLD_MAP -- an offset relative to this dynamic section entry. Dynamic linker puts the pointer to the `r_debug` structure there. When patchelf updates the executable RPATH, it moves the .dynamic section both in the binary and in memory, while __RLD_MAP is not moved in memory, since it belongs to special .rld_map section that has type PROGBITS. So, the offset stored in DT_MIPS_RLD_MAP_REL entry is not valid anymore and should be updated. This commit adds the necessary update. In the corner case when DT_MIPS_RLD_MAP_REL is present, but .rld_map section is not, the dynamic loader writes the debug pointer to some arbitrary bytes in memory. To avoid crushes on otherwise "working" binaries, we set offset to zero so that the dynamic loader would just overwrite the dynamic section. Here we also import DT_MIPS_RLD_MAP_REL definition in elf.h form glibc commit a2057c984e4314c3740f04cf54e36c824e4c8f32. Refs: NixOS#82 Signed-off-by: Ivan A. Melnikov <iv@altlinux.org>
When loading the executable on MIPS, the dynamic loader looks for MIPS ABI flags using PT_MIPS_ABIFLAGS header. The flags themselves are stored in the .MIPS.abiflags section, so the header must be updated when the section is moved. Here we also import PT_MIPS_ABIFLAGS definition from glibc commit 0bd956720c457ff054325b48f26ac7c91cb060e8. Closes: NixOS#82 Signed-off-by: Ivan A. Melnikov <iv@altlinux.org>
|
Thanks! |
ghost
mentioned this pull request
A couple of changes that address problems described in #82.
Tested on two machines, Baikal BE-T1000 and Loongson 3A, running mipsel port of ALT Linux Sisyphus. With these two changes, all tests pass on both of the machines, and on my x86_64 machine.