Re: [PATCH 00/10] RFC: assorted bcachefs patches

From: Kent Overstreet
Date: Fri May 18 2018 - 13:32:50 EST


On Fri, May 18, 2018 at 02:03:25PM -0400, Josef Bacik wrote:
> There's nothing stopping us from doing that, it just uses a kprobe to override
> the function with our helper, so we could conceivably put it anywhere in the
> function. The reason I limited it to individual functions was because it was
> easier than trying to figure out the side-effects of stopping mid-function. If
> I needed to fail mid-function I just added a helper where I needed it and failed
> that instead. I imagine safety is going to be of larger concern if we allow bpf
> scripts to randomly return anywhere inside a function, even if the function is
> marked as allowing error injection. Thanks,

Ahh no, that's not what I want... here's an example:

https://evilpiepirate.org/git/bcachefs.git/tree/fs/bcachefs/btree_cache.c#n674

Here we've got to do this thing which can race - which is fine, we just need to
check for and handle the race, on line 709 - but actually exercising that with a
test is difficult since it requires a heavily multithreaded workload with btree
nodes getting evicted to see it happen, so - it pretends the race happened if
race_fault() returns true. The race_fault() invocation shows up in debugfs,
where userspace can tell it to fire.

the way it works is dynamic_fault() is a macro that expands to a static struct
dfault_descriptor, stuck in a particular linker section so the dynamic fault
code can find them and stick them in debugfs (which is also the way dynamic
debug works).

#define dynamic_fault(_class) \
({ \
static struct _dfault descriptor \
__used __aligned(8) __attribute__((section("__faults"))) = { \
.modname = KBUILD_MODNAME, \
.function = __func__, \
.filename = __FILE__, \
.line = __LINE__, \
.class = _class, \
}; \
\
static_key_false(&descriptor.enabled) && \
__dynamic_fault_enabled(&descriptor); \
})

Honestly it still seems like the cleanest and safest way of doing it to me...