Looks like if a bpf program gets inserted and then removed, hdev->bpf.device_data is then allocated, but the loop iterating over the bpf program is never assigning ret. This is a problem and also revealed another bug in which only the last value of ret was checked. This effectively meant than only the last program in the chain could change the size of the incoming buffer. Reported-by: Dan Carpenter <dan.carpenter@xxxxxxxxxx> Link: https://lore.kernel.org/all/00f7b624-219f-4a05-a7ad-5335f15a41c7@moroto.mountain Fixes: 4a86220e046d ("HID: bpf: remove tracing HID-BPF capability") Signed-off-by: Benjamin Tissoires <bentiss@xxxxxxxxxx> --- no changes in v2 --- drivers/hid/bpf/hid_bpf_dispatch.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/drivers/hid/bpf/hid_bpf_dispatch.c b/drivers/hid/bpf/hid_bpf_dispatch.c index 06cc628e7bb4..b7b11a7c69db 100644 --- a/drivers/hid/bpf/hid_bpf_dispatch.c +++ b/drivers/hid/bpf/hid_bpf_dispatch.c @@ -57,11 +57,12 @@ dispatch_hid_bpf_device_event(struct hid_device *hdev, enum hid_report_type type } if (ret) - ctx_kern.ctx.retval = ret; + ctx_kern.ctx.size = ret; } } rcu_read_unlock(); + ret = ctx_kern.ctx.size; if (ret) { if (ret > ctx_kern.ctx.allocated_size) return ERR_PTR(-EINVAL); -- 2.44.0
