Hello. In my latest set of 32-bit UID support patches, I've changed the
msgctl(), semctl(), and shmctl() functions to no longer use the same
structures for both kernel and user space (for IPC_STAT, IPC_SET,
MSG_STAT, SEM_STAT, and SHM_STAT). Instead, the msg_queue, semid_ds, and
shmid_ds structures are considered private to the kernel, and there are
now 2 sets of structures for communicating with user space:
user_msqid_ds, user_semid_ds, user_shmid_ds
old_user_msqid_ds, old_user_semid_ds, old_user_shmid_ds
The former are used for the "new" IPC_STAT, IPC_SET, et. al, while the
latter are used for backwards compatibility (i.e. 16-bit UIDs).
I favored this approach because it made adding 32-bit UID support as
simple as possible. I've created functions such as
kernel2user_semid_ds(struct semid_ds *in, struct user_semid_ds *out)
kernel2user_ipc_perm(struct ipc_perm *in, struct user_ipc_perm *out)
kernel2old_user_ipc_perm(struct ipc_perm *in, struct old_user_ipc_perm *out)
rather than toss all the new code into the already hairy msgctl(),
semctl(), and shmctl() functions. This makes the 32-bit UID patch much
smaller and easier to manage.
If anyone is interested in looking over this patch, you can find it at:
The patch doesn't define the user_ipc_perm, user_msqid_ds, user_semid_ds,
and user_shmid_ds structures; these are defined on an
architecture-by-architecture basis for greatest flexibility. You can
examine the architectural patches at:
to see the actual definitions.
- does anyone think that this is a bad idea?
- should there be more pad space left in the user_ipc_perm,
user_msqid_ds, user_semid_ds, and user_shmid_ds structures for
future use? At present, I've left 2 machine words worth of pad
space in the msqid, semid, and shmid structures, and no extra
padding in user_ipc_perm.