Check that the frame pointer is within the current stack.

[jhb]

Summary

This same check is used on other architectures. Previously this would
permit a stack frame to unwind into any arbitrary kernel address
(including unmapped addresses).

[jhb]

RISC-V has similar changes in D27357

[andrew]

sys/cddl/dev/dtrace/aarch64/dtrace_isa.c

91 ↗

(On Diff #79958)

This was changed in rS333570 as unwind_frame could be probed by the FBT provider. It might be enough to exclude it from being probed in fbt_provide_module_function

[markj]

sys/cddl/dev/dtrace/aarch64/dtrace_isa.c

91 ↗

(On Diff #79958)

I think that would be fine. Alternately, perhaps unwind_frame() could move to stack.h?

[jhb]

sys/cddl/dev/dtrace/aarch64/dtrace_isa.c

91 ↗

(On Diff #79958)

Hmm, risc-v would need the same approach as I have made it also use unwind_frame. Do you have a preference Mark on how to fix it?

[markj]

sys/cddl/dev/dtrace/aarch64/dtrace_isa.c

91 ↗

(On Diff #79958)

I think I prefer to explicitly exclude unwind_frame() from FBT by modifying fbt_provide_module_function() for the relevant arches. All of the solutions (keep the code as-is, make unwind_frame() inlinable, exclude unwind_frame() from FBT) are kind of fragile, but modifying FBT at least makes it easy to identify these problematic functions.

[jhb]

• Add unwind_frame to fbt_excluded.