On Fri, Jun 16, 2017 at 11:47:41AM +0100, Catalin Marinas wrote:

On Thu, Jun 15, 2017 at 03:03:42PM +0100, Dave P Martin wrote:

quoted

--- a/arch/arm64/include/uapi/asm/sigcontext.h
+++ b/arch/arm64/include/uapi/asm/sigcontext.h

@@ -41,9 +41,10 @@ struct sigcontext {
  *
  *	0x210		fpsimd_context
  *	 0x10		esr_context
+ *	 0x20		extra_context (optional)
  *	 0x10		terminator (null _aarch64_ctx)
  *
- *	0xdd0		(reserved for future allocation)
+ *	0xdb0		(reserved for future allocation)
  *
  * New records that can exceed this space need to be opt-in for userspace, so
  * that an expanded signal frame is not generated unexpectedly.  The mechanism

@@ -80,4 +81,39 @@ struct esr_context {
 	__u64 esr;
 };
 
+/*
+ * extra_context: describes extra space in the signal frame for
+ * additional structures that don't fit in sigcontext.__reserved[].
+ *
+ * Note:
+ *
+ * 1) fpsimd_context, esr_context and extra_context must be placed in
+ * sigcontext.__reserved[] if present.  They cannot be placed in the
+ * extra space.  Any other record can be placed either in the extra
+ * space or in sigcontext.__reserved[], unless otherwise specified in
+ * this file.
+ *
+ * 2) There must not be more than one extra_context.
+ *
+ * 3) If extra_context is present, it must be followed immediately in
+ * sigcontext.__reserved[] by the terminating null _aarch64_ctx.
+ *
+ * 4) The extra space to which data points must start at the first
+ * 16-byte aligned address immediately after the terminating null
+ * _aarch64_ctx that follows the extra_context structure in
+ * __reserved[].  The extra space may overrun the end of __reserved[],
+ * as indicated by a sufficiently large value for the size field.
+ *
+ * 5) The extra space must itself be terminated with a null
+ * _aarch64_ctx.
+ */
+#define EXTRA_MAGIC	0x45585401
+
+struct extra_context {
+	struct _aarch64_ctx head;
+	void __user *data;	/* 16-byte aligned pointer to extra space */
+	__u32 size;		/* size in bytes of the extra space */
+	__u32 __reserved[3];
+};

Some more thoughts on the type of "data" above. It's the first time we
introduce pointers in sigcontext and wonder whether __u64 would make
more sense (invariable size with the ABI).

I'm happy to change this if you want -- it can be a __u64 without
impacting the semantics.

Also for discussion - whether using an offset vs absolute address would
be better.

My rationale here related to the ucontext semantics associated with
the signal frame, rather than signal handling itself.

Consider:

void signal_handler(int n, siginfo_t *si, void *uc)
{
	*signal_context = *(ucontext_t *)uc;
}

Will this "work"?  On most machines, yes it will. 

With an expanded signal frame on arm64, it won't work: the copied
context will be truncated at the end of __reserved[].  If extra_data
uses an offset to locate the additional data then someone parsing
*signal_context won't be able to detect the error and will overrun:
this includes rt_sigreturn (if that is run on a stack that has the
copied context).

If a pointer is used in extra_context, this provides some chance of
handling this: naive code like the above will result in a ucontext_t
whose extra_context data pointer is "wrong": extra_context aware
parsers can detect that and error out or ignore the remaining data.
extra_context aware code that wants to copy the signal context can walk
the signal frame to determine its true size, allocate memory, copy, and
then update the extra_context data pointer in the new copy.

This isn't foolproof, but it at least gives us some protection.

Cheers
---Dave