liveupdate: sanitize incoming session count
PATCH 1/1 liveupdate: sanitize incoming session count
luo_session_deserialize() iterates incoming sessions using luo_session_header_ser::count. The header physical address is provided by the previous kernel via the KHO FDT node.
If the header is corrupted, count may become arbitrarily large and the new kernel can read past the preserved session array (sh->ser[i]). This is an OOB read that can crash or hang early boot.
This can happen if the FDT node is corrupted or mis-parsed and points to a wrong header address, if stale/incompatible handover data is interpreted with the wrong layout, or if the preserved region is scribbled by memory corruption or DMA after kexec.
Clamp the incoming count to LUO_SESSION_MAX before iterating.
--- kernel/liveupdate/luo_session.c | 10 +++++++++- 1 file changed, 9 insertions(+), 1 deletion(-)
kernel/liveupdate/luo_session.c +9 -1
@@ -515,6 +515,7 @@ int luo_session_deserialize(void)struct luo_session_header *sh = &luo_session_global.incoming;static bool is_deserialized;static int err; +u64 count;/* If has been deserialized, always return the same error code */ if (is_deserialized)
@@ -524,6 +525,13 @@ int luo_session_deserialize(void)if (!sh->active)return 0 ; + count = sh->header_ser->count; +if (count >LUO_SESSION_MAX ) { +pr_warn ("incoming session count %llu exceeds max %lu\n" , + count,LUO_SESSION_MAX ); + count =LUO_SESSION_MAX ; + } +/* * Note on error handling: *
@@ -539,7 +547,7 @@ int luo_session_deserialize(void)* userspace to detect the failure and trigger a reboot, which will * reliably reset devices and reclaim memory. */ -for (int i =0 ; i < sh->header_ser->count; i++) { +for (u64 i =0 ; i < count; i++) {struct luo_session *session; session =luo_session_alloc (sh->ser[i].name);
Signature · 1 line
2.52.0
Pratyush Yadav
maintainerHi Li,
Quoted 4 lines · Li Chen
luo_session_deserialize()iterates incoming sessions usingluo_session_header_ser::count. The header physical address is provided by the previous kernel via the KHO FDT node.
Quoted 8 lines
If the header is corrupted, count may become arbitrarily large and the new kernel can read past the preserved session array (sh->ser[i]). This is an OOB read that can crash or hang early boot.
This can happen if the FDT node is corrupted or mis-parsed and points to a wrong header address, if stale/incompatible handover data is interpreted with the wrong layout, or if the preserved region is scribbled by memory corruption or DMA after kexec.
If the header is corrupted, won't the FDT magic checks fail when doing any of the FDT operations like getting the compatible? Or perhaps we should call fdt_check_header() in luo_early_startup()?
I think the sanity check might still be a useful thing, but I'd like to clarify _why_ we are doing this.
Quoted 4 lines
Clamp the incoming count to
LUO_SESSION_MAXbefore iterating.Signed-off-by Li Chen <[email protected]>
[...]
Signature · 2 lines
Regards, Pratyush Yadav