Elf_Data with d_align of zero passed the alignment check. But an
alignment of zero would cause the padding calculation to be wrong
(resulting in a zero offset). Clamp d_align of zero to one before the
alignment check, the same way sh_addralign of zero is already handled.
Commit c0748eff8473 ("libelf: More checking of valid sh_addralign
values.") fixed a similar issue in elf_getdata. Commit 7369e7d9f171
("elf_newdata: Make sure to set d_type to ELF_T_BYTE and d_align to
one") tried to fix this same issue by making sure elf_newdata set
d_align to one by default.
* libelf/elf32_updatenull.c (update_null): Clamp data->d_align
of zero to one before the powerof2 check.
Signed-off-by: Mark Wielaard <[email protected]>
---
libelf/elf32_updatenull.c | 4 ++++
1 file changed, 4 insertions(+)
diff --git a/libelf/elf32_updatenull.c b/libelf/elf32_updatenull.c
index 57239ee4ac90..a0a9c3bf3afe 100644
--- a/libelf/elf32_updatenull.c
+++ b/libelf/elf32_updatenull.c
@@ -308,6 +308,10 @@ __elfw2(LIBELFBITS,updatenull_wrlock) (Elf *elf, int
*change_bop, size_t shnum)
return -1;
}
+ /* An alignment of zero would cause the padding
+ calculation to be wrong. */
+ if (data->d_align == 0)
+ data->d_align = 1;
if (unlikely (! powerof2 (data->d_align)))
{
__libelf_seterrno (ELF_E_INVALID_ALIGN);
--
2.56.0
