[RFC][PATCH] SI_ASYNCIO: should be a kernel signal ?

[Sukadev Bhattiprolu <sukadev@linux.vnet.ibm.com>]

From: Sukadev Bhattiprolu <sukadev@linux.vnet.ibm.com>
Date: Fri, 19 Dec 2008 20:45:49 -0800
Subject: [RFC][PATCH] SI_ASYNCIO: should be a kernel signal ?

SI_ASYNCIO is currently defined as -4 in the kernel. This makes it appear
like a "user" signal (i.e SI_FROMUSER() is true). SI_ASYNCIO is generated
by the kernel for async events like SI_MESGQ and SI_POLL, but unlike
SI_ASYNCIO, SI_MESGQ and SI_POLL are "kernel" signals (i.e SI_FROMKERNEL()
is true).

Shouldn't SI_ASYNCIO be a "kernel" signal too ? It is currently generated
from USB core code.

This quick/untested RFC patch changes the in-kernel value of SI_ASYNCIO
as follows so that it becomes a "kernel" signal.

	(7 << 16)|(-4 & 0xffff) = 0x7fffc which is SI_FROMKERNEL().

The user-space value of SI_ASYNCIO continues to be -4.

Known side-effects:

	Is this required to be SI_FROMUSER() to enable the uid checks in
	kill_pid_info_as_uid() ? Also, changing to "kernel" signal would skip
	the permission checks in check_kill_permission().  Would that be a
	problem ?

Why bother now ? (Sigh. Condensed long story)

	Besides the consistency with SI_POLL and SI_MESGQ this could simplify
	implementation of special signal semantics for container-init.  When a
	signal is sent to container-init from user-space, we need to check the
	pid namespace of the sender in send_signal(). But since send_signal()
	can also be called from interrupt context,  we have no way of knowing
	if it is safe to check the pid namespace of the caller.

	If SI_ASYNCIO signal appears as a kernel signal, we could possibly use
	SI_FROMUSER() to check if it safe to reference the pid namespace of
	the sender.

	If this change has no other side-effects/breakage we will explore this
	path further for the signal semantics for container-init. (There could
	be other hurdles along the way...) 

	See also http://lkml.org/lkml/2008/12/20/183

Appreciate any comments on this.

TODO:
	If this makes sense, make corresponding change to the SI_ASYNCIO
	in arch/mips/siginfo.h.

	SI_DETHREAD and SI_SIGIO are currently unused in the kernel. Should
	we similarly make them "kernel" signals too ?

---
 include/asm-generic/siginfo.h |    4 +++-
 1 files changed, 3 insertions(+), 1 deletions(-)

diff --git a/include/asm-generic/siginfo.h b/include/asm-generic/siginfo.h
index 9695701..7b69598 100644
--- a/include/asm-generic/siginfo.h
+++ b/include/asm-generic/siginfo.h
@@ -124,6 +124,7 @@ typedef struct siginfo {
 #define __SI_CHLD	(4 << 16)
 #define __SI_RT		(5 << 16)
 #define __SI_MESGQ	(6 << 16)
+#define __SI_ASYNCIO	(7 << 16)
 #define __SI_CODE(T,N)	((T) | ((N) & 0xffff))
 #else
 #define __SI_KILL	0
@@ -133,6 +134,7 @@ typedef struct siginfo {
 #define __SI_CHLD	0
 #define __SI_RT		0
 #define __SI_MESGQ	0
+#define __SI_ASYNCIO	0
 #define __SI_CODE(T,N)	(N)
 #endif
 
@@ -145,7 +147,7 @@ typedef struct siginfo {
 #define SI_QUEUE	-1		/* sent by sigqueue */
 #define SI_TIMER __SI_CODE(__SI_TIMER,-2) /* sent by timer expiration */
 #define SI_MESGQ __SI_CODE(__SI_MESGQ,-3) /* sent by real time mesq state change */
-#define SI_ASYNCIO	-4		/* sent by AIO completion */
+#define SI_ASYNCIO __SI_CODE(__SI_ASYNCIO, -4)	/* sent by AIO completion */
 #define SI_SIGIO	-5		/* sent by queued SIGIO */
 #define SI_TKILL	-6		/* sent by tkill system call */
 #define SI_DETHREAD	-7		/* sent by execve() killing subsidiary threads */
-- 
1.5.2.5