Hello,
Manolo de Medici, le dim. 05 avril 2026 13:39:31 +0100, a ecrit:
> ipc_kmsg_copyin_body() updated OOL port arrays from 32-bit
> mach_port_name_t elements to 64-bit mach_port_t elements only in the
> non-empty path. For zero-length arrays, the code skipped that
> normalization and left msgtl_size at 32.
Ok, but what is the consequence?
In commit logs we not only need to know the "what" of the patch, but
also the "why".
> Normalize the descriptor size for OOL port arrays before the length == 0
> branch so empty and non-empty arrays are handled consistently.
> ---
> ipc/ipc_kmsg.c | 13 +++++++++++--
> 1 file changed, 11 insertions(+), 2 deletions(-)
>
> diff --git a/ipc/ipc_kmsg.c b/ipc/ipc_kmsg.c
> index fe6582d1..daf2ef4b 100644
> --- a/ipc/ipc_kmsg.c
> +++ b/ipc/ipc_kmsg.c
> @@ -1355,6 +1355,15 @@ ipc_kmsg_copyin_body(
>
> is_port = MACH_MSG_TYPE_PORT_ANY(name);
>
> + if (is_port && !is_inline &&
> + sizeof(mach_port_name_t) != sizeof(mach_port_t)) {
> + if (longform)
> + type->msgtl_size = sizeof(mach_port_t) * 8;
> + else
> + ((mach_msg_type_t *)type)->msgt_size =
> + sizeof(mach_port_t) * 8;
> + }
This looks quite early, can't wedo it after the check for types, and
then it can be inside the !is_inline if branch? That'd make the code
more simple by keeping all the out-of-line code together.
Samuel
> if ((is_port && !is_inline && (size !=
> PORT_NAME_T_SIZE_IN_BITS)) ||
> (is_port && is_inline && (size != PORT_T_SIZE_IN_BITS)) ||
> #ifndef __LP64__
> @@ -1398,14 +1407,14 @@ ipc_kmsg_copyin_body(
> data = 0;
> else if (is_port) {
> const vm_size_t user_length = length;
> +
> /*
> * In 64 bit architectures, out of line port
> names are
> * represented as an array of mach_port_name_t
> which are
> * smaller than mach_port_t.
> - */
> + */
> if (sizeof(mach_port_name_t) !=
> sizeof(mach_port_t)) {
> length = sizeof(mach_port_t) * number;
> - type->msgtl_size = sizeof(mach_port_t)
> * 8;
> }
>
> data = kalloc(length);
> --
> 2.53.0
>
>
--
Samuel
In mutt, type cthis
Dans mutt, taper cceci
