Fix ELF segment end computation #422
Conversation
ELF segment end computation was using the page-aligned start, not the virtual address start (which may not be page-aligned). This was resulting in not protecting the BSS section.
|
wait do we need an analogous fix for |
|
I don't remember off-hand how TLS is laid out in memory so I'm not sure. That might be a good thing to document and maybe point out the pkey_mprotected parts in permissive mode logs (#421). |
|
TLS looks ok to me, but RELRO needed the same fix. Not that it matters there because RELRO start is page-aligned already in the ELF, but this just ensures we do it right even if it's not. |
I don't really understand, what do you want documented? |
|
Oh I meant mark the segments or parts of a segment that are protected in the /proc/pid/maps that we write to the end of each permissive mode log. But I can do it. |
|
Oops, wrong branch. I'll just push these two commits to main as well. |
ELF segment end computation was using the page-aligned start, not the virtual address start (which may not be page-aligned). This was resulting in not protecting the BSS section.