On Mon, Apr 27, 2020 at 6:11 PM Dmitry Torokhov dmitry.torokhov@gmail.com wrote:
On Mon, Apr 27, 2020 at 02:55:48PM -0700, Evan Green wrote:
Fix a use-after-free noticed by running with KASAN enabled. If rmi_irq_fn() is run twice in a row, then rmi_f11_attention() (among others) will end up reading from drvdata->attn_data.data, which was freed and left dangling in rmi_irq_fn().
Commit 55edde9fff1a ("Input: synaptics-rmi4 - prevent UAF reported by KASAN") correctly identified and analyzed this bug. However the attempted fix only NULLed out a local variable, missing the fact that drvdata->attn_data is a struct, not a pointer.
NULL out the correct pointer in the driver data to prevent the attention functions from copying from it.
Fixes: 55edde9fff1a ("Input: synaptics-rmi4 - prevent UAF reported by KASAN") Fixes: b908d3cd812a ("Input: synaptics-rmi4 - allow to add attention data")
Signed-off-by: Evan Green evgreen@chromium.org
Cc: stable@vger.kernel.org Cc: Nick Desaulniers nick.desaulniers@gmail.com
Ugh, this is all kind of ugly, but I guess that's what we have now. Applied, thank you.
Thanks Dmitry. There are other bits that sketch me out in here as well, but I didn't get a chance to really figure out if they're a problem. We call rmi_process_interrupt_requests(), which results in reads from that same attn_data.data pointer, in a few different places. Some of those calls are outside the irq handling path, like the in rmi_enable_irq() and rmi_enable_sensor(). Can they race with the irq handling path? (Meaning they'd be doing those attn_data.data reads as rmi_irq_fn() is kfreeing the data?) There are a smattering of mutexes around, but I'm not sure if they're trying to protect this.
If I can find some time I'll try to submit a patch. Anyone is welcome to beat me to it though. -Evan