Set page permissions for kernel modules

[jtl]

Summary

Set page permissions for kernel modules based on each section's flags.

In order to separate sections with different permissions, we need to put them in different pages. This will result in some extra memory usage to segregate sections on permission boundaries.

The most complicated bit of this is dealing with previous versions of the boot loader which put pages with different permissions on the same page. They can also leave "holes" at any point in the memory.

Test Plan

I created two kernel modules: one which puts executable code in the BSS and an integer in a read-only section; and another module which tries to write to the integer and execute the code. After the changes, I correctly got page faults.

[jhb]

stand/common/load_elf_obj.c

236

This probably breaks debuggers. Debuggers assume loadable sections (SHF_ALLOC) are loaded contiguously into memory honoring alignment, but not with any other gaps. You can tell if kgdb is able to resolve global symbol names to their correct locations (e.g. does 'p foo' where 'foo' is some global variable in a module return the right results). The so-type KLDs that link_elf.c/load_elf.c load probably don't suffer from this as there we load PT_LOAD segments which already account for this sort of thing. You should fix the permissions on those though while you are at it. The amd64 kernel uses link_elf.c/load_elf.c for example rather than the _obj variants. Kernel modules on all other archs besides amd64 and mips also use link_elf.c/load_elf.c (e.g. i386, but also arm)

[jtl]

stand/common/load_elf_obj.c

236

This probably breaks debuggers.

*sigh* You are, of course, correct.

OK, I'll need to rework this quite a bit.

The so-type KLDs that link_elf.c/load_elf.c load probably don't suffer from this as there we load PT_LOAD segments which already account for this sort of thing. You should fix the permissions on those though while you are at it.

Yes, I was planning on tackling them next. But, it sounds like I sort of already wrote the code they need. :-)

FWIW, I probably won't get to revisiting this until Monday.

Thanks for the feedback!

[markj]

I believe this is superseded by commit 1d9eae9fb2e2253ca3d3764a5cc7f124b10e358b .