x86/vdso: Error out if the vDSO isn't a valid DSO

Some distros has been playing with toolchain changes that can affect
the type of ELF objects built.  Occasionally, this goes wrong and
the vDSO ends up not being a DSO at all.  This causes the kernel to
end up broken in a surprisingly subtle way -- glibc apparently
silently ignores a vDSO that isn't a DSO, so everything works,
albeit slowly, until users try a different libc implementation.

Make the kernel build process a bit more robust: fail outright if
the vDSO isn't ET_DYN or is missing its PT_DYNAMIC segment.  I've
never seen this in an unmodified kernel.

See: https://github.com/docker/docker/issues/23378

Signed-off-by: Andy Lutomirski <luto@kernel.org>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Brian Gerst <brgerst@gmail.com>
Cc: Denys Vlasenko <dvlasenk@redhat.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Josh Poimboeuf <jpoimboe@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/8a30e0a07c3b47ff917a8daa2df5e407cc0c6698.1468878336.git.luto@kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
This commit is contained in:
Andy Lutomirski 2016-07-18 14:46:35 -07:00 committed by Ingo Molnar
parent be8a18e2e9
commit 57f90c3dfc
1 changed files with 6 additions and 0 deletions

View File

@ -22,6 +22,9 @@ static void BITSFUNC(go)(void *raw_addr, size_t raw_len,
ELF(Phdr) *pt = (ELF(Phdr) *)(raw_addr + GET_LE(&hdr->e_phoff));
if (hdr->e_type != ET_DYN)
fail("input is not a shared object\n");
/* Walk the segment table. */
for (i = 0; i < GET_LE(&hdr->e_phnum); i++) {
if (GET_LE(&pt[i].p_type) == PT_LOAD) {
@ -49,6 +52,9 @@ static void BITSFUNC(go)(void *raw_addr, size_t raw_len,
if (stripped_len < load_size)
fail("stripped input is too short\n");
if (!dyn)
fail("input has no PT_DYNAMIC section -- your toolchain is buggy\n");
/* Walk the dynamic table */
for (i = 0; dyn + i < dyn_end &&
GET_LE(&dyn[i].d_tag) != DT_NULL; i++) {