Hi,
This is a followup of my v1 at [0].
The short summary of the previous cover letter and discussions is that HID could benefit from BPF for the following use cases:
- simple fixup of report descriptor: benefits are faster development time and testing, with the produced bpf program being shipped in the kernel directly (the shipping part is *not* addressed here).
- Universal Stylus Interface: allows a user-space program to define its own kernel interface
- Surface Dial: somehow similar to the previous one except that userspace can decide to change the shape of the exported device
- firewall: still partly missing there, there is not yet interception of hidraw calls, but it's coming in a followup series, I promise
- tracing: well, tracing.
I tried to address as many comments as I could and here is the short log of changes:
v2: ===
- split the series by subsystem (bpf, HID, libbpf, selftests and samples)
- Added an extra patch at the beginning to not require CAP_NET_ADMIN for BPF_PROG_TYPE_LIRC_MODE2 (please shout if this is wrong)
- made the bpf context attached to HID program of dynamic size: * the first 1 kB will be able to be addressed directly * the rest can be retrieved through bpf_hid_{set|get}_data (note that I am definitivey not happy with that API, because there is part of it in bits and other in bytes. ouch)
- added an extra patch to prevent non GPL HID bpf programs to be loaded of type BPF_PROG_TYPE_HID * same here, not really happy but I don't know where to put that check in verifier.c
- added a new flag BPF_F_INSERT_HEAD for BPF_LINK_CREATE syscall when in used with HID program types. * this flag is used for tracing, to be able to load a program before any others that might already have been inserted and that might change the data stream.
Cheers, Benjamin
[0] https://lore.kernel.org/linux-input/20220224110828.2168231-1-benjamin.tissoi...
Benjamin Tissoires (28): bpf: add new is_sys_admin_prog_type() helper bpf: introduce hid program type HID: hook up with bpf libbpf: add HID program type and API selftests/bpf: add tests for the HID-bpf initial implementation samples/bpf: add new hid_mouse example bpf/hid: add a new attach type to change the report descriptor HID: allow to change the report descriptor from an eBPF program libbpf: add new attach type BPF_HID_RDESC_FIXUP selftests/bpf: add report descriptor fixup tests samples/bpf: add a report descriptor fixup bpf/hid: add hid_{get|set}_data helpers HID: bpf: implement hid_bpf_get|set_data selftests/bpf: add tests for hid_{get|set}_data helpers bpf/hid: add new BPF type to trigger commands from userspace libbpf: add new attach type BPF_HID_USER_EVENT selftests/bpf: add test for user call of HID bpf programs selftests/bpf: hid: rely on uhid event to know if a test device is ready bpf/hid: add bpf_hid_raw_request helper function HID: add implementation of bpf_hid_raw_request selftests/bpf: add tests for bpf_hid_hw_request bpf/verifier: prevent non GPL programs to be loaded against HID HID: bpf: compute only the required buffer size for the device HID: bpf: only call hid_bpf_raw_event() if a ctx is available bpf/hid: Add a flag to add the program at the beginning of the list libbpf: add handling for BPF_F_INSERT_HEAD in HID programs selftests/bpf: Add a test for BPF_F_INSERT_HEAD samples/bpf: fix bpf_program__attach_hid() api change
drivers/hid/Makefile | 1 + drivers/hid/hid-bpf.c | 361 +++++++++ drivers/hid/hid-core.c | 34 +- include/linux/bpf-hid.h | 129 +++ include/linux/bpf_types.h | 4 + include/linux/hid.h | 25 + include/uapi/linux/bpf.h | 59 ++ include/uapi/linux/bpf_hid.h | 50 ++ kernel/bpf/Makefile | 3 + kernel/bpf/hid.c | 652 +++++++++++++++ kernel/bpf/syscall.c | 26 +- kernel/bpf/verifier.c | 7 + samples/bpf/.gitignore | 1 + samples/bpf/Makefile | 4 + samples/bpf/hid_mouse_kern.c | 91 +++ samples/bpf/hid_mouse_user.c | 129 +++ tools/include/uapi/linux/bpf.h | 59 ++ tools/lib/bpf/libbpf.c | 22 +- tools/lib/bpf/libbpf.h | 2 + tools/lib/bpf/libbpf.map | 1 + tools/testing/selftests/bpf/prog_tests/hid.c | 788 +++++++++++++++++++ tools/testing/selftests/bpf/progs/hid.c | 216 +++++ 22 files changed, 2649 insertions(+), 15 deletions(-) create mode 100644 drivers/hid/hid-bpf.c create mode 100644 include/linux/bpf-hid.h create mode 100644 include/uapi/linux/bpf_hid.h create mode 100644 kernel/bpf/hid.c create mode 100644 samples/bpf/hid_mouse_kern.c create mode 100644 samples/bpf/hid_mouse_user.c create mode 100644 tools/testing/selftests/bpf/prog_tests/hid.c create mode 100644 tools/testing/selftests/bpf/progs/hid.c
LIRC_MODE2 does not really need net_admin capability, but only sys_admin.
Extract a new helper for it, it will be also used for the HID bpf implementation.
Cc: Sean Young sean@mess.org Signed-off-by: Benjamin Tissoires benjamin.tissoires@redhat.com
---
new in v2 --- kernel/bpf/syscall.c | 14 +++++++++++++- 1 file changed, 13 insertions(+), 1 deletion(-)
diff --git a/kernel/bpf/syscall.c b/kernel/bpf/syscall.c index db402ebc5570..cc570891322b 100644 --- a/kernel/bpf/syscall.c +++ b/kernel/bpf/syscall.c @@ -2165,7 +2165,6 @@ static bool is_net_admin_prog_type(enum bpf_prog_type prog_type) case BPF_PROG_TYPE_LWT_SEG6LOCAL: case BPF_PROG_TYPE_SK_SKB: case BPF_PROG_TYPE_SK_MSG: - case BPF_PROG_TYPE_LIRC_MODE2: case BPF_PROG_TYPE_FLOW_DISSECTOR: case BPF_PROG_TYPE_CGROUP_DEVICE: case BPF_PROG_TYPE_CGROUP_SOCK: @@ -2202,6 +2201,17 @@ static bool is_perfmon_prog_type(enum bpf_prog_type prog_type) } }
+static bool is_sys_admin_prog_type(enum bpf_prog_type prog_type) +{ + switch (prog_type) { + case BPF_PROG_TYPE_LIRC_MODE2: + case BPF_PROG_TYPE_EXT: /* extends any prog */ + return true; + default: + return false; + } +} + /* last field in 'union bpf_attr' used by this command */ #define BPF_PROG_LOAD_LAST_FIELD core_relo_rec_size
@@ -2252,6 +2262,8 @@ static int bpf_prog_load(union bpf_attr *attr, bpfptr_t uattr) return -EPERM; if (is_perfmon_prog_type(type) && !perfmon_capable()) return -EPERM; + if (is_sys_admin_prog_type(type) && !capable(CAP_SYS_ADMIN)) + return -EPERM;
/* attach_prog_fd/attach_btf_obj_fd can specify fd of either bpf_prog * or btf, we need to check which one it is
On Fri, Mar 4, 2022 at 9:30 AM Benjamin Tissoires benjamin.tissoires@redhat.com wrote:
LIRC_MODE2 does not really need net_admin capability, but only sys_admin.
Extract a new helper for it, it will be also used for the HID bpf implementation.
Cc: Sean Young sean@mess.org Signed-off-by: Benjamin Tissoires benjamin.tissoires@redhat.com
new in v2
kernel/bpf/syscall.c | 14 +++++++++++++- 1 file changed, 13 insertions(+), 1 deletion(-)
diff --git a/kernel/bpf/syscall.c b/kernel/bpf/syscall.c index db402ebc5570..cc570891322b 100644 --- a/kernel/bpf/syscall.c +++ b/kernel/bpf/syscall.c @@ -2165,7 +2165,6 @@ static bool is_net_admin_prog_type(enum bpf_prog_type prog_type) case BPF_PROG_TYPE_LWT_SEG6LOCAL: case BPF_PROG_TYPE_SK_SKB: case BPF_PROG_TYPE_SK_MSG:
case BPF_PROG_TYPE_LIRC_MODE2: case BPF_PROG_TYPE_FLOW_DISSECTOR: case BPF_PROG_TYPE_CGROUP_DEVICE: case BPF_PROG_TYPE_CGROUP_SOCK:@@ -2202,6 +2201,17 @@ static bool is_perfmon_prog_type(enum bpf_prog_type prog_type) } }
+static bool is_sys_admin_prog_type(enum bpf_prog_type prog_type) +{
switch (prog_type) {case BPF_PROG_TYPE_LIRC_MODE2:case BPF_PROG_TYPE_EXT: /* extends any prog */return true;default:return false;}+}
I am not sure whether we should do this. This is a behavior change, that may break some user space. Also, BPF_PROG_TYPE_EXT is checked in is_perfmon_prog_type(), and this change will make that case useless.
Thanks, Song
[...]
On Sat, Mar 5, 2022 at 12:12 AM Song Liu song@kernel.org wrote:
On Fri, Mar 4, 2022 at 9:30 AM Benjamin Tissoires benjamin.tissoires@redhat.com wrote:
LIRC_MODE2 does not really need net_admin capability, but only sys_admin.
Extract a new helper for it, it will be also used for the HID bpf implementation.
Cc: Sean Young sean@mess.org Signed-off-by: Benjamin Tissoires benjamin.tissoires@redhat.com
new in v2
kernel/bpf/syscall.c | 14 +++++++++++++- 1 file changed, 13 insertions(+), 1 deletion(-)
diff --git a/kernel/bpf/syscall.c b/kernel/bpf/syscall.c index db402ebc5570..cc570891322b 100644 --- a/kernel/bpf/syscall.c +++ b/kernel/bpf/syscall.c @@ -2165,7 +2165,6 @@ static bool is_net_admin_prog_type(enum bpf_prog_type prog_type) case BPF_PROG_TYPE_LWT_SEG6LOCAL: case BPF_PROG_TYPE_SK_SKB: case BPF_PROG_TYPE_SK_MSG:
case BPF_PROG_TYPE_LIRC_MODE2: case BPF_PROG_TYPE_FLOW_DISSECTOR: case BPF_PROG_TYPE_CGROUP_DEVICE: case BPF_PROG_TYPE_CGROUP_SOCK:@@ -2202,6 +2201,17 @@ static bool is_perfmon_prog_type(enum bpf_prog_type prog_type) } }
+static bool is_sys_admin_prog_type(enum bpf_prog_type prog_type) +{
switch (prog_type) {case BPF_PROG_TYPE_LIRC_MODE2:case BPF_PROG_TYPE_EXT: /* extends any prog */return true;default:return false;}+}
I am not sure whether we should do this. This is a behavior change, that may break some user space. Also, BPF_PROG_TYPE_EXT is checked in is_perfmon_prog_type(), and this change will make that case useless.
Sure, I can drop it from v3 and make this function appear for HID only.
Regarding BPF_PROG_TYPE_EXT, it was already in both is_net_admin_prog_type() and is_perfmon_prog_type(), so I duplicated it here, but I agree, given that it's already in the first function there, CPA_SYS_ADMIN is already checked.
Cheers, Benjamin
Thanks, Song
[...]
On Sat, Mar 05, 2022 at 11:07:04AM +0100, Benjamin Tissoires wrote:
On Sat, Mar 5, 2022 at 12:12 AM Song Liu song@kernel.org wrote:
On Fri, Mar 4, 2022 at 9:30 AM Benjamin Tissoires benjamin.tissoires@redhat.com wrote:
LIRC_MODE2 does not really need net_admin capability, but only sys_admin.
Extract a new helper for it, it will be also used for the HID bpf implementation.
Cc: Sean Young sean@mess.org Signed-off-by: Benjamin Tissoires benjamin.tissoires@redhat.com
new in v2
kernel/bpf/syscall.c | 14 +++++++++++++- 1 file changed, 13 insertions(+), 1 deletion(-)
diff --git a/kernel/bpf/syscall.c b/kernel/bpf/syscall.c index db402ebc5570..cc570891322b 100644 --- a/kernel/bpf/syscall.c +++ b/kernel/bpf/syscall.c @@ -2165,7 +2165,6 @@ static bool is_net_admin_prog_type(enum bpf_prog_type prog_type) case BPF_PROG_TYPE_LWT_SEG6LOCAL: case BPF_PROG_TYPE_SK_SKB: case BPF_PROG_TYPE_SK_MSG:
case BPF_PROG_TYPE_LIRC_MODE2: case BPF_PROG_TYPE_FLOW_DISSECTOR: case BPF_PROG_TYPE_CGROUP_DEVICE: case BPF_PROG_TYPE_CGROUP_SOCK:@@ -2202,6 +2201,17 @@ static bool is_perfmon_prog_type(enum bpf_prog_type prog_type) } }
+static bool is_sys_admin_prog_type(enum bpf_prog_type prog_type) +{
switch (prog_type) {case BPF_PROG_TYPE_LIRC_MODE2:case BPF_PROG_TYPE_EXT: /* extends any prog */return true;default:return false;}+}
I am not sure whether we should do this. This is a behavior change, that may break some user space. Also, BPF_PROG_TYPE_EXT is checked in is_perfmon_prog_type(), and this change will make that case useless.
Sure, I can drop it from v3 and make this function appear for HID only.
For BPF_PROG_TYPE_LIRC_MODE2, I don't think this change will break userspace. This is called from ir-keytable(1) which is called from udev. It should have all the necessary permissions.
In addition, the vast majority IR decoders are non-bpf. bpf ir decoders have very few users at the moment.
I am working on completely new userspace tooling which will make extensive use of bpf ir decoding with full lircd and IRP compatibility, but this is not finished yet (see https://github.com/seanyoung/cir).
Thanks
Sean
On Sat, Mar 5, 2022 at 8:58 AM Sean Young sean@mess.org wrote:
On Sat, Mar 05, 2022 at 11:07:04AM +0100, Benjamin Tissoires wrote:
On Sat, Mar 5, 2022 at 12:12 AM Song Liu song@kernel.org wrote:
On Fri, Mar 4, 2022 at 9:30 AM Benjamin Tissoires benjamin.tissoires@redhat.com wrote:
LIRC_MODE2 does not really need net_admin capability, but only sys_admin.
Extract a new helper for it, it will be also used for the HID bpf implementation.
Cc: Sean Young sean@mess.org Signed-off-by: Benjamin Tissoires benjamin.tissoires@redhat.com
new in v2
kernel/bpf/syscall.c | 14 +++++++++++++- 1 file changed, 13 insertions(+), 1 deletion(-)
diff --git a/kernel/bpf/syscall.c b/kernel/bpf/syscall.c index db402ebc5570..cc570891322b 100644 --- a/kernel/bpf/syscall.c +++ b/kernel/bpf/syscall.c @@ -2165,7 +2165,6 @@ static bool is_net_admin_prog_type(enum bpf_prog_type prog_type) case BPF_PROG_TYPE_LWT_SEG6LOCAL: case BPF_PROG_TYPE_SK_SKB: case BPF_PROG_TYPE_SK_MSG:
case BPF_PROG_TYPE_LIRC_MODE2: case BPF_PROG_TYPE_FLOW_DISSECTOR: case BPF_PROG_TYPE_CGROUP_DEVICE: case BPF_PROG_TYPE_CGROUP_SOCK:@@ -2202,6 +2201,17 @@ static bool is_perfmon_prog_type(enum bpf_prog_type prog_type) } }
+static bool is_sys_admin_prog_type(enum bpf_prog_type prog_type) +{
switch (prog_type) {case BPF_PROG_TYPE_LIRC_MODE2:case BPF_PROG_TYPE_EXT: /* extends any prog */return true;default:return false;}+}
I am not sure whether we should do this. This is a behavior change, that may break some user space. Also, BPF_PROG_TYPE_EXT is checked in is_perfmon_prog_type(), and this change will make that case useless.
Sure, I can drop it from v3 and make this function appear for HID only.
For BPF_PROG_TYPE_LIRC_MODE2, I don't think this change will break userspace. This is called from ir-keytable(1) which is called from udev. It should have all the necessary permissions.
In addition, the vast majority IR decoders are non-bpf. bpf ir decoders have very few users at the moment.
I am working on completely new userspace tooling which will make extensive use of bpf ir decoding with full lircd and IRP compatibility, but this is not finished yet (see https://github.com/seanyoung/cir).
Thanks for these information. I guess change for BPF_PROG_TYPE_LIRC_MODE2 is ok then. Would you mind ack or review this change (either current version or a later version)?
Thanks, Song
On Sat, Mar 5, 2022 at 2:07 AM Benjamin Tissoires benjamin.tissoires@redhat.com wrote:
On Sat, Mar 5, 2022 at 12:12 AM Song Liu song@kernel.org wrote:
On Fri, Mar 4, 2022 at 9:30 AM Benjamin Tissoires benjamin.tissoires@redhat.com wrote:
LIRC_MODE2 does not really need net_admin capability, but only sys_admin.
Extract a new helper for it, it will be also used for the HID bpf implementation.
Cc: Sean Young sean@mess.org Signed-off-by: Benjamin Tissoires benjamin.tissoires@redhat.com
new in v2
kernel/bpf/syscall.c | 14 +++++++++++++- 1 file changed, 13 insertions(+), 1 deletion(-)
diff --git a/kernel/bpf/syscall.c b/kernel/bpf/syscall.c index db402ebc5570..cc570891322b 100644 --- a/kernel/bpf/syscall.c +++ b/kernel/bpf/syscall.c @@ -2165,7 +2165,6 @@ static bool is_net_admin_prog_type(enum bpf_prog_type prog_type) case BPF_PROG_TYPE_LWT_SEG6LOCAL: case BPF_PROG_TYPE_SK_SKB: case BPF_PROG_TYPE_SK_MSG:
case BPF_PROG_TYPE_LIRC_MODE2: case BPF_PROG_TYPE_FLOW_DISSECTOR: case BPF_PROG_TYPE_CGROUP_DEVICE: case BPF_PROG_TYPE_CGROUP_SOCK:@@ -2202,6 +2201,17 @@ static bool is_perfmon_prog_type(enum bpf_prog_type prog_type) } }
+static bool is_sys_admin_prog_type(enum bpf_prog_type prog_type) +{
switch (prog_type) {case BPF_PROG_TYPE_LIRC_MODE2:case BPF_PROG_TYPE_EXT: /* extends any prog */return true;default:return false;}+}
I am not sure whether we should do this. This is a behavior change, that may break some user space. Also, BPF_PROG_TYPE_EXT is checked in is_perfmon_prog_type(), and this change will make that case useless.
Sure, I can drop it from v3 and make this function appear for HID only.
Regarding BPF_PROG_TYPE_EXT, it was already in both is_net_admin_prog_type() and is_perfmon_prog_type(), so I duplicated it here, but I agree, given that it's already in the first function there, CPA_SYS_ADMIN is already checked.
I think with current code, a user with CAP_BPF, CAP_NET_ADMIN, and CAP_PERFMON (but not CAP_SYS_ADMIN) can load programs of type BPF_PROG_TYPE_EXT. But after the patch, the same user will not be able to do it. Did I misread it? It is not a common case though.
On Fri, Mar 04, 2022 at 06:28:25PM +0100, Benjamin Tissoires wrote:
LIRC_MODE2 does not really need net_admin capability, but only sys_admin.
Extract a new helper for it, it will be also used for the HID bpf implementation.
Cc: Sean Young sean@mess.org Signed-off-by: Benjamin Tissoires benjamin.tissoires@redhat.com
For BPF_PROG_TYPE_LIRC_MODE2, I don't think this change will break userspace. This is called from ir-keytable(1) which is called from udev. It should have all the necessary permissions.
In addition, the vast majority IR decoders are non-bpf. bpf ir decoders have very few users at the moment.
Acked-by: Sean Young sean@mess.org
Sean
new in v2
kernel/bpf/syscall.c | 14 +++++++++++++- 1 file changed, 13 insertions(+), 1 deletion(-)
diff --git a/kernel/bpf/syscall.c b/kernel/bpf/syscall.c index db402ebc5570..cc570891322b 100644 --- a/kernel/bpf/syscall.c +++ b/kernel/bpf/syscall.c @@ -2165,7 +2165,6 @@ static bool is_net_admin_prog_type(enum bpf_prog_type prog_type) case BPF_PROG_TYPE_LWT_SEG6LOCAL: case BPF_PROG_TYPE_SK_SKB: case BPF_PROG_TYPE_SK_MSG:
- case BPF_PROG_TYPE_LIRC_MODE2: case BPF_PROG_TYPE_FLOW_DISSECTOR: case BPF_PROG_TYPE_CGROUP_DEVICE: case BPF_PROG_TYPE_CGROUP_SOCK:
@@ -2202,6 +2201,17 @@ static bool is_perfmon_prog_type(enum bpf_prog_type prog_type) } } +static bool is_sys_admin_prog_type(enum bpf_prog_type prog_type) +{
- switch (prog_type) {
- case BPF_PROG_TYPE_LIRC_MODE2:
- case BPF_PROG_TYPE_EXT: /* extends any prog */
return true;- default:
return false;- }
+}
/* last field in 'union bpf_attr' used by this command */ #define BPF_PROG_LOAD_LAST_FIELD core_relo_rec_size @@ -2252,6 +2262,8 @@ static int bpf_prog_load(union bpf_attr *attr, bpfptr_t uattr) return -EPERM; if (is_perfmon_prog_type(type) && !perfmon_capable()) return -EPERM;
- if (is_sys_admin_prog_type(type) && !capable(CAP_SYS_ADMIN))
return -EPERM;/* attach_prog_fd/attach_btf_obj_fd can specify fd of either bpf_prog * or btf, we need to check which one it is -- 2.35.1
HID is a protocol that could benefit from using BPF too.
This patch implements a net-like use of BPF capability for HID. Any incoming report coming from the device can be injected into a series of BPF programs that can modify it or even discard it by setting the size in the context to 0.
The kernel/bpf implementation is based on net-namespace.c, with only the bpf_link part kept, there is no real points in keeping the bpf_prog_{attach|detach} API.
The implementation here is only focusing on the bpf changes. The HID changes that hooks onto this are coming in a separate patch.
Given that HID can be compiled in as a module, and the functions that kernel/bpf/hid.c needs to call in hid.ko are exported in struct hid_hooks.
Signed-off-by: Benjamin Tissoires benjamin.tissoires@redhat.com
---
changes in v2: - split the series by bpf/libbpf/hid/selftests and samples - unsigned long -> __u16 in uapi/linux/bpf_hid.h - change the bpf_ctx to be of variable size, with a min of 1024 bytes - make this 1 kB available directly from bpf program, the rest will need a helper - add some more doc comments in uapi --- include/linux/bpf-hid.h | 108 ++++++++ include/linux/bpf_types.h | 4 + include/linux/hid.h | 5 + include/uapi/linux/bpf.h | 7 + include/uapi/linux/bpf_hid.h | 39 +++ kernel/bpf/Makefile | 3 + kernel/bpf/hid.c | 437 +++++++++++++++++++++++++++++++++ kernel/bpf/syscall.c | 8 + tools/include/uapi/linux/bpf.h | 7 + 9 files changed, 618 insertions(+) create mode 100644 include/linux/bpf-hid.h create mode 100644 include/uapi/linux/bpf_hid.h create mode 100644 kernel/bpf/hid.c
diff --git a/include/linux/bpf-hid.h b/include/linux/bpf-hid.h new file mode 100644 index 000000000000..3cda78051b5f --- /dev/null +++ b/include/linux/bpf-hid.h @@ -0,0 +1,108 @@ +/* SPDX-License-Identifier: GPL-2.0 */ +#ifndef _BPF_HID_H +#define _BPF_HID_H + +#include <linux/mutex.h> +#include <uapi/linux/bpf.h> +#include <uapi/linux/bpf_hid.h> +#include <linux/list.h> +#include <linux/slab.h> + +struct bpf_prog; +struct bpf_prog_array; +struct hid_device; + +enum bpf_hid_attach_type { + BPF_HID_ATTACH_INVALID = -1, + BPF_HID_ATTACH_DEVICE_EVENT = 0, + MAX_BPF_HID_ATTACH_TYPE +}; + +struct bpf_hid { + struct hid_bpf_ctx *ctx; + + /* Array of programs to run compiled from links */ + struct bpf_prog_array __rcu *run_array[MAX_BPF_HID_ATTACH_TYPE]; + struct list_head links[MAX_BPF_HID_ATTACH_TYPE]; +}; + +static inline enum bpf_hid_attach_type +to_bpf_hid_attach_type(enum bpf_attach_type attach_type) +{ + switch (attach_type) { + case BPF_HID_DEVICE_EVENT: + return BPF_HID_ATTACH_DEVICE_EVENT; + default: + return BPF_HID_ATTACH_INVALID; + } +} + +static inline struct hid_bpf_ctx *bpf_hid_allocate_ctx(struct hid_device *hdev, + size_t data_size) +{ + struct hid_bpf_ctx *ctx; + + /* ensure data_size is between min and max */ + data_size = clamp_val(data_size, + HID_BPF_MIN_BUFFER_SIZE, + HID_BPF_MAX_BUFFER_SIZE); + + ctx = kzalloc(sizeof(*ctx) + data_size, GFP_KERNEL); + if (!ctx) + return ERR_PTR(-ENOMEM); + + ctx->hdev = hdev; + ctx->allocated_size = data_size; + + return ctx; +} + +union bpf_attr; +struct bpf_prog; + +#if IS_ENABLED(CONFIG_HID) +int bpf_hid_prog_query(const union bpf_attr *attr, + union bpf_attr __user *uattr); +int bpf_hid_link_create(const union bpf_attr *attr, + struct bpf_prog *prog); +#else +static inline int bpf_hid_prog_query(const union bpf_attr *attr, + union bpf_attr __user *uattr) +{ + return -EOPNOTSUPP; +} + +static inline int bpf_hid_link_create(const union bpf_attr *attr, + struct bpf_prog *prog) +{ + return -EOPNOTSUPP; +} +#endif + +static inline bool bpf_hid_link_empty(struct bpf_hid *bpf, + enum bpf_hid_attach_type type) +{ + return list_empty(&bpf->links[type]); +} + +struct bpf_hid_hooks { + struct hid_device *(*hdev_from_fd)(int fd); + int (*link_attach)(struct hid_device *hdev, enum bpf_hid_attach_type type); + void (*array_detached)(struct hid_device *hdev, enum bpf_hid_attach_type type); +}; + +#ifdef CONFIG_BPF +int bpf_hid_init(struct hid_device *hdev); +void bpf_hid_exit(struct hid_device *hdev); +void bpf_hid_set_hooks(struct bpf_hid_hooks *hooks); +#else +static inline int bpf_hid_init(struct hid_device *hdev) +{ + return 0; +} + +static inline void bpf_hid_exit(struct hid_device *hdev) {} +static inline void bpf_hid_set_hooks(struct bpf_hid_hooks *hooks) {} +#endif + +#endif /* _BPF_HID_H */ diff --git a/include/linux/bpf_types.h b/include/linux/bpf_types.h index 48a91c51c015..1509862aacc4 100644 --- a/include/linux/bpf_types.h +++ b/include/linux/bpf_types.h @@ -76,6 +76,10 @@ BPF_PROG_TYPE(BPF_PROG_TYPE_EXT, bpf_extension, BPF_PROG_TYPE(BPF_PROG_TYPE_LSM, lsm, void *, void *) #endif /* CONFIG_BPF_LSM */ +#if IS_ENABLED(CONFIG_HID) +BPF_PROG_TYPE(BPF_PROG_TYPE_HID, hid, + __u32, u32) +#endif #endif BPF_PROG_TYPE(BPF_PROG_TYPE_SYSCALL, bpf_syscall, void *, void *) diff --git a/include/linux/hid.h b/include/linux/hid.h index 7487b0586fe6..56f6f4ad45a7 100644 --- a/include/linux/hid.h +++ b/include/linux/hid.h @@ -15,6 +15,7 @@
#include <linux/bitops.h> +#include <linux/bpf-hid.h> #include <linux/types.h> #include <linux/slab.h> #include <linux/list.h> @@ -639,6 +640,10 @@ struct hid_device { /* device report descriptor */ struct list_head debug_list; spinlock_t debug_list_lock; wait_queue_head_t debug_wait; + +#ifdef CONFIG_BPF + struct bpf_hid bpf; +#endif };
#define to_hid_device(pdev) \ diff --git a/include/uapi/linux/bpf.h b/include/uapi/linux/bpf.h index afe3d0d7f5f2..5978b92cacd3 100644 --- a/include/uapi/linux/bpf.h +++ b/include/uapi/linux/bpf.h @@ -952,6 +952,7 @@ enum bpf_prog_type { BPF_PROG_TYPE_LSM, BPF_PROG_TYPE_SK_LOOKUP, BPF_PROG_TYPE_SYSCALL, /* a program that can execute syscalls */ + BPF_PROG_TYPE_HID, };
enum bpf_attach_type { @@ -997,6 +998,7 @@ enum bpf_attach_type { BPF_SK_REUSEPORT_SELECT, BPF_SK_REUSEPORT_SELECT_OR_MIGRATE, BPF_PERF_EVENT, + BPF_HID_DEVICE_EVENT, __MAX_BPF_ATTACH_TYPE };
@@ -1011,6 +1013,7 @@ enum bpf_link_type { BPF_LINK_TYPE_NETNS = 5, BPF_LINK_TYPE_XDP = 6, BPF_LINK_TYPE_PERF_EVENT = 7, + BPF_LINK_TYPE_HID = 8,
MAX_BPF_LINK_TYPE, }; @@ -5870,6 +5873,10 @@ struct bpf_link_info { struct { __u32 ifindex; } xdp; + struct { + __s32 hidraw_ino; + __u32 attach_type; + } hid; }; } __attribute__((aligned(8)));
diff --git a/include/uapi/linux/bpf_hid.h b/include/uapi/linux/bpf_hid.h new file mode 100644 index 000000000000..975ca5bd526f --- /dev/null +++ b/include/uapi/linux/bpf_hid.h @@ -0,0 +1,39 @@ +/* SPDX-License-Identifier: GPL-2.0-or-later WITH Linux-syscall-note */ + +/* + * HID BPF public headers + * + * Copyright (c) 2021 Benjamin Tissoires + */ + +#ifndef _UAPI__LINUX_BPF_HID_H__ +#define _UAPI__LINUX_BPF_HID_H__ + +#include <linux/types.h> + +/* + * The first 1024 bytes are available directly in the bpf programs. + * To access the rest of the data (if allocated_size is bigger + * than 1024, you need to use bpf_hid_ helpers. + */ +#define HID_BPF_MIN_BUFFER_SIZE 1024 +#define HID_BPF_MAX_BUFFER_SIZE 16384 /* in sync with HID_MAX_BUFFER_SIZE */ + +struct hid_device; + +enum hid_bpf_event { + HID_BPF_UNDEF = 0, + HID_BPF_DEVICE_EVENT, /* when attach type is BPF_HID_DEVICE_EVENT */ +}; + +struct hid_bpf_ctx { + enum hid_bpf_event type; /* read-only */ + __u16 allocated_size; /* the allocated size of data below (RO) */ + struct hid_device *hdev; /* read-only */ + + __u16 size; /* used size in data (RW) */ + __u8 data[]; /* data buffer (RW) */ +}; + +#endif /* _UAPI__LINUX_BPF_HID_H__ */ + diff --git a/kernel/bpf/Makefile b/kernel/bpf/Makefile index c1a9be6a4b9f..8d5619d3d7e5 100644 --- a/kernel/bpf/Makefile +++ b/kernel/bpf/Makefile @@ -35,6 +35,9 @@ ifeq ($(CONFIG_BPF_JIT),y) obj-$(CONFIG_BPF_SYSCALL) += bpf_struct_ops.o obj-${CONFIG_BPF_LSM} += bpf_lsm.o endif +ifneq ($(CONFIG_HID),) +obj-$(CONFIG_BPF_SYSCALL) += hid.o +endif obj-$(CONFIG_BPF_PRELOAD) += preload/
obj-$(CONFIG_BPF_SYSCALL) += relo_core.o diff --git a/kernel/bpf/hid.c b/kernel/bpf/hid.c new file mode 100644 index 000000000000..db7f75a0a812 --- /dev/null +++ b/kernel/bpf/hid.c @@ -0,0 +1,437 @@ +// SPDX-License-Identifier: GPL-2.0 +/* + * based on kernel/bpf/net-namespace.c + */ + +#include <linux/bpf.h> +#include <linux/bpf-hid.h> +#include <linux/filter.h> +#include <linux/hid.h> +#include <linux/hidraw.h> + +/* + * Functions to manage BPF programs attached to hid + */ + +struct bpf_hid_link { + struct bpf_link link; + enum bpf_attach_type type; + enum bpf_hid_attach_type hid_type; + + /* Must be accessed with bpf_hid_mutex held. */ + struct hid_device *hdev; + struct list_head node; /* node in list of links attached to hid */ +}; + +/* Protects updates to bpf_hid */ +DEFINE_MUTEX(bpf_hid_mutex); + +static struct bpf_hid_hooks hid_hooks = {0}; + +void bpf_hid_set_hooks(struct bpf_hid_hooks *hooks) +{ + if (hooks) + hid_hooks = *hooks; + else + memset(&hid_hooks, 0, sizeof(hid_hooks)); +} +EXPORT_SYMBOL_GPL(bpf_hid_set_hooks); + +static const struct bpf_func_proto * +hid_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) +{ + switch (func_id) { + default: + return bpf_base_func_proto(func_id); + } +} + +static bool hid_is_valid_access(int off, int size, + enum bpf_access_type access_type, + const struct bpf_prog *prog, + struct bpf_insn_access_aux *info) +{ + /* everything not in ctx is prohibited */ + if (off < 0 || off + size > sizeof(struct hid_bpf_ctx) + HID_BPF_MIN_BUFFER_SIZE) + return false; + + switch (off) { + /* type, allocated_size, hdev are read-only */ + case bpf_ctx_range_till(struct hid_bpf_ctx, type, hdev): + return access_type == BPF_READ; + } + + /* everything else is read/write */ + return true; +} + +const struct bpf_verifier_ops hid_verifier_ops = { + .get_func_proto = hid_func_proto, + .is_valid_access = hid_is_valid_access +}; + +/* Must be called with bpf_hid_mutex held. */ +static void bpf_hid_run_array_detach(struct hid_device *hdev, + enum bpf_hid_attach_type type) +{ + struct bpf_prog_array *run_array; + + run_array = rcu_replace_pointer(hdev->bpf.run_array[type], NULL, + lockdep_is_held(&bpf_hid_mutex)); + bpf_prog_array_free(run_array); + + if (hid_hooks.array_detached) + hid_hooks.array_detached(hdev, type); +} + +static int link_index(struct hid_device *hdev, enum bpf_hid_attach_type type, + struct bpf_hid_link *link) +{ + struct bpf_hid_link *pos; + int i = 0; + + list_for_each_entry(pos, &hdev->bpf.links[type], node) { + if (pos == link) + return i; + i++; + } + return -ENOENT; +} + +static int link_count(struct hid_device *hdev, enum bpf_hid_attach_type type) +{ + struct list_head *pos; + int i = 0; + + list_for_each(pos, &hdev->bpf.links[type]) + i++; + return i; +} + +static void fill_prog_array(struct hid_device *hdev, enum bpf_hid_attach_type type, + struct bpf_prog_array *prog_array) +{ + struct bpf_hid_link *pos; + unsigned int i = 0; + + list_for_each_entry(pos, &hdev->bpf.links[type], node) { + prog_array->items[i].prog = pos->link.prog; + i++; + } +} + +static void bpf_hid_link_release(struct bpf_link *link) +{ + struct bpf_hid_link *hid_link = + container_of(link, struct bpf_hid_link, link); + enum bpf_hid_attach_type type = hid_link->hid_type; + struct bpf_prog_array *old_array, *new_array; + struct hid_device *hdev; + int cnt, idx; + + mutex_lock(&bpf_hid_mutex); + + hdev = hid_link->hdev; + if (!hdev) + goto out_unlock; + + /* Remember link position in case of safe delete */ + idx = link_index(hdev, type, hid_link); + list_del(&hid_link->node); + + cnt = link_count(hdev, type); + if (!cnt) { + bpf_hid_run_array_detach(hdev, type); + goto out_unlock; + } + + old_array = rcu_dereference_protected(hdev->bpf.run_array[type], + lockdep_is_held(&bpf_hid_mutex)); + new_array = bpf_prog_array_alloc(cnt, GFP_KERNEL); + if (!new_array) { + WARN_ON(bpf_prog_array_delete_safe_at(old_array, idx)); + goto out_unlock; + } + fill_prog_array(hdev, type, new_array); + rcu_assign_pointer(hdev->bpf.run_array[type], new_array); + bpf_prog_array_free(old_array); + +out_unlock: + hid_link->hdev = NULL; + mutex_unlock(&bpf_hid_mutex); +} + +static int bpf_hid_link_detach(struct bpf_link *link) +{ + bpf_hid_link_release(link); + return 0; +} + +static void bpf_hid_link_dealloc(struct bpf_link *link) +{ + struct bpf_hid_link *hid_link = + container_of(link, struct bpf_hid_link, link); + + kfree(hid_link); +} + +static int bpf_hid_link_update_prog(struct bpf_link *link, + struct bpf_prog *new_prog, + struct bpf_prog *old_prog) +{ + struct bpf_hid_link *hid_link = + container_of(link, struct bpf_hid_link, link); + enum bpf_hid_attach_type type = hid_link->hid_type; + struct bpf_prog_array *run_array; + struct hid_device *hdev; + int idx, ret; + + if (old_prog && old_prog != link->prog) + return -EPERM; + if (new_prog->type != link->prog->type) + return -EINVAL; + + mutex_lock(&bpf_hid_mutex); + + hdev = hid_link->hdev; + if (!hdev) { + /* hid dying */ + ret = -ENOLINK; + goto out_unlock; + } + + run_array = rcu_dereference_protected(hdev->bpf.run_array[type], + lockdep_is_held(&bpf_hid_mutex)); + idx = link_index(hdev, type, hid_link); + ret = bpf_prog_array_update_at(run_array, idx, new_prog); + if (ret) + goto out_unlock; + + old_prog = xchg(&link->prog, new_prog); + bpf_prog_put(old_prog); + +out_unlock: + mutex_unlock(&bpf_hid_mutex); + return ret; +} + +static int bpf_hid_link_fill_info(const struct bpf_link *link, + struct bpf_link_info *info) +{ + const struct bpf_hid_link *hid_link = + container_of(link, struct bpf_hid_link, link); + int hidraw_ino = -1; + struct hid_device *hdev; + struct hidraw *hidraw; + + mutex_lock(&bpf_hid_mutex); + hdev = hid_link->hdev; + if (hdev && hdev->hidraw) { + hidraw = hdev->hidraw; + hidraw_ino = hidraw->minor; + } + mutex_unlock(&bpf_hid_mutex); + + info->hid.hidraw_ino = hidraw_ino; + info->hid.attach_type = hid_link->type; + return 0; +} + +static void bpf_hid_link_show_fdinfo(const struct bpf_link *link, + struct seq_file *seq) +{ + struct bpf_link_info info = {}; + + bpf_hid_link_fill_info(link, &info); + seq_printf(seq, + "hidraw_ino:\t%u\n" + "attach_type:\t%u\n", + info.hid.hidraw_ino, + info.hid.attach_type); +} + +static const struct bpf_link_ops bpf_hid_link_ops = { + .release = bpf_hid_link_release, + .dealloc = bpf_hid_link_dealloc, + .detach = bpf_hid_link_detach, + .update_prog = bpf_hid_link_update_prog, + .fill_link_info = bpf_hid_link_fill_info, + .show_fdinfo = bpf_hid_link_show_fdinfo, +}; + +/* Must be called with bpf_hid_mutex held. */ +static int __bpf_hid_prog_query(const union bpf_attr *attr, + union bpf_attr __user *uattr, + struct hid_device *hdev, + enum bpf_hid_attach_type type) +{ + __u32 __user *prog_ids = u64_to_user_ptr(attr->query.prog_ids); + struct bpf_prog_array *run_array; + u32 prog_cnt = 0, flags = 0; + + run_array = rcu_dereference_protected(hdev->bpf.run_array[type], + lockdep_is_held(&bpf_hid_mutex)); + if (run_array) + prog_cnt = bpf_prog_array_length(run_array); + + if (copy_to_user(&uattr->query.attach_flags, &flags, sizeof(flags))) + return -EFAULT; + if (copy_to_user(&uattr->query.prog_cnt, &prog_cnt, sizeof(prog_cnt))) + return -EFAULT; + if (!attr->query.prog_cnt || !prog_ids || !prog_cnt) + return 0; + + return bpf_prog_array_copy_to_user(run_array, prog_ids, + attr->query.prog_cnt); +} + +int bpf_hid_prog_query(const union bpf_attr *attr, + union bpf_attr __user *uattr) +{ + enum bpf_hid_attach_type type; + struct hid_device *hdev; + int ret; + + if (attr->query.query_flags || !hid_hooks.hdev_from_fd) + return -EINVAL; + + type = to_bpf_hid_attach_type(attr->query.attach_type); + if (type < 0) + return -EINVAL; + + hdev = hid_hooks.hdev_from_fd(attr->query.target_fd); + if (IS_ERR(hdev)) + return PTR_ERR(hdev); + + mutex_lock(&bpf_hid_mutex); + ret = __bpf_hid_prog_query(attr, uattr, hdev, type); + mutex_unlock(&bpf_hid_mutex); + + return ret; +} + +static int bpf_hid_max_progs(enum bpf_hid_attach_type type) +{ + switch (type) { + case BPF_HID_ATTACH_DEVICE_EVENT: + return 64; + default: + return 0; + } +} + +static int bpf_hid_link_attach(struct hid_device *hdev, struct bpf_link *link, + enum bpf_hid_attach_type type) +{ + struct bpf_hid_link *hid_link = + container_of(link, struct bpf_hid_link, link); + struct bpf_prog_array *run_array; + int cnt, err = 0; + + mutex_lock(&bpf_hid_mutex); + + cnt = link_count(hdev, type); + if (cnt >= bpf_hid_max_progs(type)) { + err = -E2BIG; + goto out_unlock; + } + + if (hid_hooks.link_attach) { + err = hid_hooks.link_attach(hdev, type); + if (err) + goto out_unlock; + } + + run_array = bpf_prog_array_alloc(cnt + 1, GFP_KERNEL); + if (!run_array) { + err = -ENOMEM; + goto out_unlock; + } + + list_add_tail(&hid_link->node, &hdev->bpf.links[type]); + + fill_prog_array(hdev, type, run_array); + run_array = rcu_replace_pointer(hdev->bpf.run_array[type], run_array, + lockdep_is_held(&bpf_hid_mutex)); + bpf_prog_array_free(run_array); + +out_unlock: + mutex_unlock(&bpf_hid_mutex); + return err; +} + +int bpf_hid_link_create(const union bpf_attr *attr, struct bpf_prog *prog) +{ + enum bpf_hid_attach_type hid_type; + struct bpf_link_primer link_primer; + struct bpf_hid_link *hid_link; + enum bpf_attach_type type; + struct hid_device *hdev; + int err; + + if (attr->link_create.flags || !hid_hooks.hdev_from_fd) + return -EINVAL; + + type = attr->link_create.attach_type; + hid_type = to_bpf_hid_attach_type(type); + if (hid_type < 0) + return -EINVAL; + + hdev = hid_hooks.hdev_from_fd(attr->link_create.target_fd); + if (IS_ERR(hdev)) + return PTR_ERR(hdev); + + hid_link = kzalloc(sizeof(*hid_link), GFP_USER); + if (!hid_link) + return -ENOMEM; + + bpf_link_init(&hid_link->link, BPF_LINK_TYPE_HID, + &bpf_hid_link_ops, prog); + hid_link->hdev = hdev; + hid_link->type = type; + hid_link->hid_type = hid_type; + + err = bpf_link_prime(&hid_link->link, &link_primer); + if (err) { + kfree(hid_link); + return err; + } + + err = bpf_hid_link_attach(hdev, &hid_link->link, hid_type); + if (err) { + bpf_link_cleanup(&link_primer); + return err; + } + + return bpf_link_settle(&link_primer); +} + +const struct bpf_prog_ops hid_prog_ops = { +}; + +int bpf_hid_init(struct hid_device *hdev) +{ + int type; + + for (type = 0; type < MAX_BPF_HID_ATTACH_TYPE; type++) + INIT_LIST_HEAD(&hdev->bpf.links[type]); + + return 0; +} +EXPORT_SYMBOL_GPL(bpf_hid_init); + +void bpf_hid_exit(struct hid_device *hdev) +{ + enum bpf_hid_attach_type type; + struct bpf_hid_link *hid_link; + + mutex_lock(&bpf_hid_mutex); + for (type = 0; type < MAX_BPF_HID_ATTACH_TYPE; type++) { + bpf_hid_run_array_detach(hdev, type); + list_for_each_entry(hid_link, &hdev->bpf.links[type], node) { + hid_link->hdev = NULL; /* auto-detach link */ + } + } + mutex_unlock(&bpf_hid_mutex); +} +EXPORT_SYMBOL_GPL(bpf_hid_exit); diff --git a/kernel/bpf/syscall.c b/kernel/bpf/syscall.c index cc570891322b..a94e78ec3211 100644 --- a/kernel/bpf/syscall.c +++ b/kernel/bpf/syscall.c @@ -3,6 +3,7 @@ */ #include <linux/bpf.h> #include <linux/bpf-cgroup.h> +#include <linux/bpf-hid.h> #include <linux/bpf_trace.h> #include <linux/bpf_lirc.h> #include <linux/bpf_verifier.h> @@ -2205,6 +2206,7 @@ static bool is_sys_admin_prog_type(enum bpf_prog_type prog_type) { switch (prog_type) { case BPF_PROG_TYPE_LIRC_MODE2: + case BPF_PROG_TYPE_HID: case BPF_PROG_TYPE_EXT: /* extends any prog */ return true; default: @@ -3200,6 +3202,8 @@ attach_type_to_prog_type(enum bpf_attach_type attach_type) return BPF_PROG_TYPE_SK_LOOKUP; case BPF_XDP: return BPF_PROG_TYPE_XDP; + case BPF_HID_DEVICE_EVENT: + return BPF_PROG_TYPE_HID; default: return BPF_PROG_TYPE_UNSPEC; } @@ -3343,6 +3347,8 @@ static int bpf_prog_query(const union bpf_attr *attr, case BPF_SK_MSG_VERDICT: case BPF_SK_SKB_VERDICT: return sock_map_bpf_prog_query(attr, uattr); + case BPF_HID_DEVICE_EVENT: + return bpf_hid_prog_query(attr, uattr); default: return -EINVAL; } @@ -4337,6 +4343,8 @@ static int link_create(union bpf_attr *attr, bpfptr_t uattr) ret = bpf_perf_link_attach(attr, prog); break; #endif + case BPF_PROG_TYPE_HID: + return bpf_hid_link_create(attr, prog); default: ret = -EINVAL; } diff --git a/tools/include/uapi/linux/bpf.h b/tools/include/uapi/linux/bpf.h index afe3d0d7f5f2..5978b92cacd3 100644 --- a/tools/include/uapi/linux/bpf.h +++ b/tools/include/uapi/linux/bpf.h @@ -952,6 +952,7 @@ enum bpf_prog_type { BPF_PROG_TYPE_LSM, BPF_PROG_TYPE_SK_LOOKUP, BPF_PROG_TYPE_SYSCALL, /* a program that can execute syscalls */ + BPF_PROG_TYPE_HID, };
enum bpf_attach_type { @@ -997,6 +998,7 @@ enum bpf_attach_type { BPF_SK_REUSEPORT_SELECT, BPF_SK_REUSEPORT_SELECT_OR_MIGRATE, BPF_PERF_EVENT, + BPF_HID_DEVICE_EVENT, __MAX_BPF_ATTACH_TYPE };
@@ -1011,6 +1013,7 @@ enum bpf_link_type { BPF_LINK_TYPE_NETNS = 5, BPF_LINK_TYPE_XDP = 6, BPF_LINK_TYPE_PERF_EVENT = 7, + BPF_LINK_TYPE_HID = 8,
MAX_BPF_LINK_TYPE, }; @@ -5870,6 +5873,10 @@ struct bpf_link_info { struct { __u32 ifindex; } xdp; + struct { + __s32 hidraw_ino; + __u32 attach_type; + } hid; }; } __attribute__((aligned(8)));
On Fri, Mar 04, 2022 at 06:28:26PM +0100, Benjamin Tissoires wrote:
HID is a protocol that could benefit from using BPF too.
This patch implements a net-like use of BPF capability for HID. Any incoming report coming from the device can be injected into a series of BPF programs that can modify it or even discard it by setting the size in the context to 0.
The kernel/bpf implementation is based on net-namespace.c, with only the bpf_link part kept, there is no real points in keeping the bpf_prog_{attach|detach} API.
The implementation here is only focusing on the bpf changes. The HID changes that hooks onto this are coming in a separate patch.
Given that HID can be compiled in as a module, and the functions that kernel/bpf/hid.c needs to call in hid.ko are exported in struct hid_hooks.
Signed-off-by: Benjamin Tissoires benjamin.tissoires@redhat.com
changes in v2:
- split the series by bpf/libbpf/hid/selftests and samples
- unsigned long -> __u16 in uapi/linux/bpf_hid.h
- change the bpf_ctx to be of variable size, with a min of 1024 bytes
- make this 1 kB available directly from bpf program, the rest will need a helper
- add some more doc comments in uapi
include/linux/bpf-hid.h | 108 ++++++++ include/linux/bpf_types.h | 4 + include/linux/hid.h | 5 + include/uapi/linux/bpf.h | 7 + include/uapi/linux/bpf_hid.h | 39 +++ kernel/bpf/Makefile | 3 + kernel/bpf/hid.c | 437 +++++++++++++++++++++++++++++++++ kernel/bpf/syscall.c | 8 + tools/include/uapi/linux/bpf.h | 7 + 9 files changed, 618 insertions(+) create mode 100644 include/linux/bpf-hid.h create mode 100644 include/uapi/linux/bpf_hid.h create mode 100644 kernel/bpf/hid.c
diff --git a/include/linux/bpf-hid.h b/include/linux/bpf-hid.h new file mode 100644 index 000000000000..3cda78051b5f --- /dev/null +++ b/include/linux/bpf-hid.h @@ -0,0 +1,108 @@ +/* SPDX-License-Identifier: GPL-2.0 */ +#ifndef _BPF_HID_H +#define _BPF_HID_H
+#include <linux/mutex.h> +#include <uapi/linux/bpf.h> +#include <uapi/linux/bpf_hid.h> +#include <linux/list.h> +#include <linux/slab.h>
+struct bpf_prog; +struct bpf_prog_array; +struct hid_device;
+enum bpf_hid_attach_type {
- BPF_HID_ATTACH_INVALID = -1,
- BPF_HID_ATTACH_DEVICE_EVENT = 0,
- MAX_BPF_HID_ATTACH_TYPE
+};
+struct bpf_hid {
- struct hid_bpf_ctx *ctx;
- /* Array of programs to run compiled from links */
- struct bpf_prog_array __rcu *run_array[MAX_BPF_HID_ATTACH_TYPE];
- struct list_head links[MAX_BPF_HID_ATTACH_TYPE];
+};
+static inline enum bpf_hid_attach_type +to_bpf_hid_attach_type(enum bpf_attach_type attach_type) +{
- switch (attach_type) {
- case BPF_HID_DEVICE_EVENT:
return BPF_HID_ATTACH_DEVICE_EVENT;- default:
return BPF_HID_ATTACH_INVALID;- }
+}
+static inline struct hid_bpf_ctx *bpf_hid_allocate_ctx(struct hid_device *hdev,
size_t data_size)+{
- struct hid_bpf_ctx *ctx;
- /* ensure data_size is between min and max */
- data_size = clamp_val(data_size,
HID_BPF_MIN_BUFFER_SIZE,HID_BPF_MAX_BUFFER_SIZE);
Do you want to return an error if the data size is not within the range? Otherwise people will just start to use crazy values and you will always be limiting them?
- ctx = kzalloc(sizeof(*ctx) + data_size, GFP_KERNEL);
- if (!ctx)
return ERR_PTR(-ENOMEM);- ctx->hdev = hdev;
- ctx->allocated_size = data_size;
- return ctx;
+}
And why is this an inline function? Why not put it in a .c file?
+union bpf_attr; +struct bpf_prog;
+#if IS_ENABLED(CONFIG_HID) +int bpf_hid_prog_query(const union bpf_attr *attr,
union bpf_attr __user *uattr);+int bpf_hid_link_create(const union bpf_attr *attr,
struct bpf_prog *prog);+#else +static inline int bpf_hid_prog_query(const union bpf_attr *attr,
union bpf_attr __user *uattr)+{
- return -EOPNOTSUPP;
+}
+static inline int bpf_hid_link_create(const union bpf_attr *attr,
struct bpf_prog *prog)+{
- return -EOPNOTSUPP;
+} +#endif
+static inline bool bpf_hid_link_empty(struct bpf_hid *bpf,
enum bpf_hid_attach_type type)+{
- return list_empty(&bpf->links[type]);
+}
+struct bpf_hid_hooks {
- struct hid_device *(*hdev_from_fd)(int fd);
- int (*link_attach)(struct hid_device *hdev, enum bpf_hid_attach_type type);
- void (*array_detached)(struct hid_device *hdev, enum bpf_hid_attach_type type);
+};
+#ifdef CONFIG_BPF +int bpf_hid_init(struct hid_device *hdev); +void bpf_hid_exit(struct hid_device *hdev); +void bpf_hid_set_hooks(struct bpf_hid_hooks *hooks); +#else +static inline int bpf_hid_init(struct hid_device *hdev) +{
- return 0;
+}
+static inline void bpf_hid_exit(struct hid_device *hdev) {} +static inline void bpf_hid_set_hooks(struct bpf_hid_hooks *hooks) {} +#endif
+#endif /* _BPF_HID_H */ diff --git a/include/linux/bpf_types.h b/include/linux/bpf_types.h index 48a91c51c015..1509862aacc4 100644 --- a/include/linux/bpf_types.h +++ b/include/linux/bpf_types.h @@ -76,6 +76,10 @@ BPF_PROG_TYPE(BPF_PROG_TYPE_EXT, bpf_extension, BPF_PROG_TYPE(BPF_PROG_TYPE_LSM, lsm, void *, void *) #endif /* CONFIG_BPF_LSM */ +#if IS_ENABLED(CONFIG_HID) +BPF_PROG_TYPE(BPF_PROG_TYPE_HID, hid,
__u32, u32)
Why the mix of __u32 and u32 here?
+#endif #endif BPF_PROG_TYPE(BPF_PROG_TYPE_SYSCALL, bpf_syscall, void *, void *) diff --git a/include/linux/hid.h b/include/linux/hid.h index 7487b0586fe6..56f6f4ad45a7 100644 --- a/include/linux/hid.h +++ b/include/linux/hid.h @@ -15,6 +15,7 @@ #include <linux/bitops.h> +#include <linux/bpf-hid.h> #include <linux/types.h> #include <linux/slab.h> #include <linux/list.h> @@ -639,6 +640,10 @@ struct hid_device { /* device report descriptor */ struct list_head debug_list; spinlock_t debug_list_lock; wait_queue_head_t debug_wait;
+#ifdef CONFIG_BPF
- struct bpf_hid bpf;
+#endif }; #define to_hid_device(pdev) \ diff --git a/include/uapi/linux/bpf.h b/include/uapi/linux/bpf.h index afe3d0d7f5f2..5978b92cacd3 100644 --- a/include/uapi/linux/bpf.h +++ b/include/uapi/linux/bpf.h @@ -952,6 +952,7 @@ enum bpf_prog_type { BPF_PROG_TYPE_LSM, BPF_PROG_TYPE_SK_LOOKUP, BPF_PROG_TYPE_SYSCALL, /* a program that can execute syscalls */
- BPF_PROG_TYPE_HID,
}; enum bpf_attach_type { @@ -997,6 +998,7 @@ enum bpf_attach_type { BPF_SK_REUSEPORT_SELECT, BPF_SK_REUSEPORT_SELECT_OR_MIGRATE, BPF_PERF_EVENT,
- BPF_HID_DEVICE_EVENT, __MAX_BPF_ATTACH_TYPE
}; @@ -1011,6 +1013,7 @@ enum bpf_link_type { BPF_LINK_TYPE_NETNS = 5, BPF_LINK_TYPE_XDP = 6, BPF_LINK_TYPE_PERF_EVENT = 7,
- BPF_LINK_TYPE_HID = 8,
MAX_BPF_LINK_TYPE, }; @@ -5870,6 +5873,10 @@ struct bpf_link_info { struct { __u32 ifindex; } xdp;
struct {__s32 hidraw_ino;
"ino"? We have lots of letters to spell words out :)
__u32 attach_type;} hid;} __attribute__((aligned(8))); diff --git a/include/uapi/linux/bpf_hid.h b/include/uapi/linux/bpf_hid.h new file mode 100644 index 000000000000..975ca5bd526f --- /dev/null +++ b/include/uapi/linux/bpf_hid.h @@ -0,0 +1,39 @@ +/* SPDX-License-Identifier: GPL-2.0-or-later WITH Linux-syscall-note */
+/*
- HID BPF public headers
- Copyright (c) 2021 Benjamin Tissoires
Did you forget the copyright line on the other .h file above?
- */
+#ifndef _UAPI__LINUX_BPF_HID_H__ +#define _UAPI__LINUX_BPF_HID_H__
+#include <linux/types.h>
+/*
- The first 1024 bytes are available directly in the bpf programs.
- To access the rest of the data (if allocated_size is bigger
- than 1024, you need to use bpf_hid_ helpers.
- */
+#define HID_BPF_MIN_BUFFER_SIZE 1024 +#define HID_BPF_MAX_BUFFER_SIZE 16384 /* in sync with HID_MAX_BUFFER_SIZE */
Can't you just use HID_MAX_BUFFER_SIZE?
Anyway, all minor stuff, looks good!
Reviewed-by: Greg Kroah-Hartman gregkh@linuxfoundation.org
On Fri, Mar 4, 2022 at 7:21 PM Greg KH gregkh@linuxfoundation.org wrote:
On Fri, Mar 04, 2022 at 06:28:26PM +0100, Benjamin Tissoires wrote:
HID is a protocol that could benefit from using BPF too.
This patch implements a net-like use of BPF capability for HID. Any incoming report coming from the device can be injected into a series of BPF programs that can modify it or even discard it by setting the size in the context to 0.
The kernel/bpf implementation is based on net-namespace.c, with only the bpf_link part kept, there is no real points in keeping the bpf_prog_{attach|detach} API.
The implementation here is only focusing on the bpf changes. The HID changes that hooks onto this are coming in a separate patch.
Given that HID can be compiled in as a module, and the functions that kernel/bpf/hid.c needs to call in hid.ko are exported in struct hid_hooks.
Signed-off-by: Benjamin Tissoires benjamin.tissoires@redhat.com
changes in v2:
- split the series by bpf/libbpf/hid/selftests and samples
- unsigned long -> __u16 in uapi/linux/bpf_hid.h
- change the bpf_ctx to be of variable size, with a min of 1024 bytes
- make this 1 kB available directly from bpf program, the rest will need a helper
- add some more doc comments in uapi
include/linux/bpf-hid.h | 108 ++++++++ include/linux/bpf_types.h | 4 + include/linux/hid.h | 5 + include/uapi/linux/bpf.h | 7 + include/uapi/linux/bpf_hid.h | 39 +++ kernel/bpf/Makefile | 3 + kernel/bpf/hid.c | 437 +++++++++++++++++++++++++++++++++ kernel/bpf/syscall.c | 8 + tools/include/uapi/linux/bpf.h | 7 + 9 files changed, 618 insertions(+) create mode 100644 include/linux/bpf-hid.h create mode 100644 include/uapi/linux/bpf_hid.h create mode 100644 kernel/bpf/hid.c
diff --git a/include/linux/bpf-hid.h b/include/linux/bpf-hid.h new file mode 100644 index 000000000000..3cda78051b5f --- /dev/null +++ b/include/linux/bpf-hid.h @@ -0,0 +1,108 @@ +/* SPDX-License-Identifier: GPL-2.0 */ +#ifndef _BPF_HID_H +#define _BPF_HID_H
+#include <linux/mutex.h> +#include <uapi/linux/bpf.h> +#include <uapi/linux/bpf_hid.h> +#include <linux/list.h> +#include <linux/slab.h>
+struct bpf_prog; +struct bpf_prog_array; +struct hid_device;
+enum bpf_hid_attach_type {
BPF_HID_ATTACH_INVALID = -1,BPF_HID_ATTACH_DEVICE_EVENT = 0,MAX_BPF_HID_ATTACH_TYPE+};
+struct bpf_hid {
struct hid_bpf_ctx *ctx;/* Array of programs to run compiled from links */struct bpf_prog_array __rcu *run_array[MAX_BPF_HID_ATTACH_TYPE];struct list_head links[MAX_BPF_HID_ATTACH_TYPE];+};
+static inline enum bpf_hid_attach_type +to_bpf_hid_attach_type(enum bpf_attach_type attach_type) +{
switch (attach_type) {case BPF_HID_DEVICE_EVENT:return BPF_HID_ATTACH_DEVICE_EVENT;default:return BPF_HID_ATTACH_INVALID;}+}
+static inline struct hid_bpf_ctx *bpf_hid_allocate_ctx(struct hid_device *hdev,
size_t data_size)+{
struct hid_bpf_ctx *ctx;/* ensure data_size is between min and max */data_size = clamp_val(data_size,HID_BPF_MIN_BUFFER_SIZE,HID_BPF_MAX_BUFFER_SIZE);Do you want to return an error if the data size is not within the range?
That was not something I was counting on. Though I am thinking of not necessarily clamping this value in the end because I might have found a way to not do the initial memcpy when running a prog, and so not having to limit the size of the data.
Otherwise people will just start to use crazy values and you will always be limiting them?
The users of this helper are really limited to drivers/hid/hid_pbf.c and kernel/bpf/hid.c. And they are known in advance and there must be only one user per attach type. The only thing where the data might explode is when in used with SEC(hid/device_event), because we statically allocate one bpf_ctx based on the device report descriptor. But if the required size is bigger than HID_BPF_MAX_BUFFER_SIZE, the device will not probe or at least already logs something in the dmesg that we are using a too big buffer.
ctx = kzalloc(sizeof(*ctx) + data_size, GFP_KERNEL);if (!ctx)return ERR_PTR(-ENOMEM);ctx->hdev = hdev;ctx->allocated_size = data_size;return ctx;+}
And why is this an inline function? Why not put it in a .c file?
The problem I have here is that the hid module can be loaded as an external module. So I can not directly use that helper from hid.ko from kernel/bpf/hid.c (I need it there once for the SEC(hid/user_event) bprogram attach type).
So the solution would be to have the code in the c part of kernel/bpf/hid.c and export the function as GPL, but I wanted to have the minimum of knowledge of HID-BPF internals in that file. So I ended up using an inline so I can reuse it independently in kernel/bpf/hid.c and drivers/hid/hid-bpf.c.
+union bpf_attr; +struct bpf_prog;
+#if IS_ENABLED(CONFIG_HID) +int bpf_hid_prog_query(const union bpf_attr *attr,
union bpf_attr __user *uattr);+int bpf_hid_link_create(const union bpf_attr *attr,
struct bpf_prog *prog);+#else +static inline int bpf_hid_prog_query(const union bpf_attr *attr,
union bpf_attr __user *uattr)+{
return -EOPNOTSUPP;+}
+static inline int bpf_hid_link_create(const union bpf_attr *attr,
struct bpf_prog *prog)+{
return -EOPNOTSUPP;+} +#endif
+static inline bool bpf_hid_link_empty(struct bpf_hid *bpf,
enum bpf_hid_attach_type type)+{
return list_empty(&bpf->links[type]);+}
+struct bpf_hid_hooks {
struct hid_device *(*hdev_from_fd)(int fd);int (*link_attach)(struct hid_device *hdev, enum bpf_hid_attach_type type);void (*array_detached)(struct hid_device *hdev, enum bpf_hid_attach_type type);+};
+#ifdef CONFIG_BPF +int bpf_hid_init(struct hid_device *hdev); +void bpf_hid_exit(struct hid_device *hdev); +void bpf_hid_set_hooks(struct bpf_hid_hooks *hooks); +#else +static inline int bpf_hid_init(struct hid_device *hdev) +{
return 0;+}
+static inline void bpf_hid_exit(struct hid_device *hdev) {} +static inline void bpf_hid_set_hooks(struct bpf_hid_hooks *hooks) {} +#endif
+#endif /* _BPF_HID_H */ diff --git a/include/linux/bpf_types.h b/include/linux/bpf_types.h index 48a91c51c015..1509862aacc4 100644 --- a/include/linux/bpf_types.h +++ b/include/linux/bpf_types.h @@ -76,6 +76,10 @@ BPF_PROG_TYPE(BPF_PROG_TYPE_EXT, bpf_extension, BPF_PROG_TYPE(BPF_PROG_TYPE_LSM, lsm, void *, void *) #endif /* CONFIG_BPF_LSM */ +#if IS_ENABLED(CONFIG_HID) +BPF_PROG_TYPE(BPF_PROG_TYPE_HID, hid,
__u32, u32)Why the mix of __u32 and u32 here?
This is actually valid. I tracked it down to kernel/bpf/btf.c with:
static union { struct bpf_ctx_convert { #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type) \ prog_ctx_type _id##_prog; \ kern_ctx_type _id##_kern; #include <linux/bpf_types.h> #undef BPF_PROG_TYPE } *__t; /* 't' is written once under lock. Read many times. */ const struct btf_type *t; } bpf_ctx_convert;
So prog_ctx_type represents a user API, while kern_ctx_type represents the kernel counterpart.
That being said, this is plain wrong, because I am not using u32 as bpf context, but a properly defined struct :o)
So I probably need to amend this to be either "void *, void *)" or something better (I'll ask Song in my reply to him).
+#endif #endif BPF_PROG_TYPE(BPF_PROG_TYPE_SYSCALL, bpf_syscall, void *, void *) diff --git a/include/linux/hid.h b/include/linux/hid.h index 7487b0586fe6..56f6f4ad45a7 100644 --- a/include/linux/hid.h +++ b/include/linux/hid.h @@ -15,6 +15,7 @@
#include <linux/bitops.h> +#include <linux/bpf-hid.h> #include <linux/types.h> #include <linux/slab.h> #include <linux/list.h> @@ -639,6 +640,10 @@ struct hid_device { /* device report descriptor */ struct list_head debug_list; spinlock_t debug_list_lock; wait_queue_head_t debug_wait;
+#ifdef CONFIG_BPF
struct bpf_hid bpf;+#endif };
#define to_hid_device(pdev) \ diff --git a/include/uapi/linux/bpf.h b/include/uapi/linux/bpf.h index afe3d0d7f5f2..5978b92cacd3 100644 --- a/include/uapi/linux/bpf.h +++ b/include/uapi/linux/bpf.h @@ -952,6 +952,7 @@ enum bpf_prog_type { BPF_PROG_TYPE_LSM, BPF_PROG_TYPE_SK_LOOKUP, BPF_PROG_TYPE_SYSCALL, /* a program that can execute syscalls */
BPF_PROG_TYPE_HID,};
enum bpf_attach_type { @@ -997,6 +998,7 @@ enum bpf_attach_type { BPF_SK_REUSEPORT_SELECT, BPF_SK_REUSEPORT_SELECT_OR_MIGRATE, BPF_PERF_EVENT,
BPF_HID_DEVICE_EVENT, __MAX_BPF_ATTACH_TYPE};
@@ -1011,6 +1013,7 @@ enum bpf_link_type { BPF_LINK_TYPE_NETNS = 5, BPF_LINK_TYPE_XDP = 6, BPF_LINK_TYPE_PERF_EVENT = 7,
BPF_LINK_TYPE_HID = 8, MAX_BPF_LINK_TYPE,}; @@ -5870,6 +5873,10 @@ struct bpf_link_info { struct { __u32 ifindex; } xdp;
struct {__s32 hidraw_ino;"ino"? We have lots of letters to spell words out :)
no comments... :)
__u32 attach_type;} hid; };} __attribute__((aligned(8)));
diff --git a/include/uapi/linux/bpf_hid.h b/include/uapi/linux/bpf_hid.h new file mode 100644 index 000000000000..975ca5bd526f --- /dev/null +++ b/include/uapi/linux/bpf_hid.h @@ -0,0 +1,39 @@ +/* SPDX-License-Identifier: GPL-2.0-or-later WITH Linux-syscall-note */
+/*
- HID BPF public headers
- Copyright (c) 2021 Benjamin Tissoires
Did you forget the copyright line on the other .h file above?
oops
- */
+#ifndef _UAPI__LINUX_BPF_HID_H__ +#define _UAPI__LINUX_BPF_HID_H__
+#include <linux/types.h>
+/*
- The first 1024 bytes are available directly in the bpf programs.
- To access the rest of the data (if allocated_size is bigger
- than 1024, you need to use bpf_hid_ helpers.
- */
+#define HID_BPF_MIN_BUFFER_SIZE 1024 +#define HID_BPF_MAX_BUFFER_SIZE 16384 /* in sync with HID_MAX_BUFFER_SIZE */
Can't you just use HID_MAX_BUFFER_SIZE?
Call me dumb, but curiously I could not get my code to compile there. If I include linux/hid.h, things are getting messy and either the tools or the kernel itself was not compiling properly (couldn't really remember what was failing exactly, sorry).
Anyway, all minor stuff, looks good!
Thanks. Not sure I'll keep the bpf_ctx the same after further thoughts, but I appreciate the review :)
Cheers, Benjamin
Reviewed-by: Greg Kroah-Hartman gregkh@linuxfoundation.org
On Fri, Mar 4, 2022 at 9:31 AM Benjamin Tissoires benjamin.tissoires@redhat.com wrote:
HID is a protocol that could benefit from using BPF too.
[...]
+#include <linux/list.h> +#include <linux/slab.h>
+struct bpf_prog; +struct bpf_prog_array; +struct hid_device;
+enum bpf_hid_attach_type {
BPF_HID_ATTACH_INVALID = -1,BPF_HID_ATTACH_DEVICE_EVENT = 0,MAX_BPF_HID_ATTACH_TYPE
Is it typical to have different BPF programs for different attach types? Otherwise, (different types may have similar BPF programs), maybe we can pass type as an argument to the program (shared among different types)?
[...]
+struct hid_device;
+enum hid_bpf_event {
HID_BPF_UNDEF = 0,HID_BPF_DEVICE_EVENT, /* when attach type is BPF_HID_DEVICE_EVENT */+};
+struct hid_bpf_ctx {
enum hid_bpf_event type; /* read-only */__u16 allocated_size; /* the allocated size of data below (RO) */
There is a (6-byte?) hole here.
struct hid_device *hdev; /* read-only */__u16 size; /* used size in data (RW) */__u8 data[]; /* data buffer (RW) */+};
Do we really need hit_bpf_ctx in uapi? Maybe we can just use it from vmlinuxh?
[...]
+static bool hid_is_valid_access(int off, int size,
enum bpf_access_type access_type,const struct bpf_prog *prog,struct bpf_insn_access_aux *info)+{
/* everything not in ctx is prohibited */if (off < 0 || off + size > sizeof(struct hid_bpf_ctx) + HID_BPF_MIN_BUFFER_SIZE)return false;
Mabe add the following here to fail unaligned accesses
if (off % size != 0) return false; [...]
On Sat, Mar 5, 2022 at 1:03 AM Song Liu song@kernel.org wrote:
On Fri, Mar 4, 2022 at 9:31 AM Benjamin Tissoires benjamin.tissoires@redhat.com wrote:
HID is a protocol that could benefit from using BPF too.
[...]
+#include <linux/list.h> +#include <linux/slab.h>
+struct bpf_prog; +struct bpf_prog_array; +struct hid_device;
+enum bpf_hid_attach_type {
BPF_HID_ATTACH_INVALID = -1,BPF_HID_ATTACH_DEVICE_EVENT = 0,MAX_BPF_HID_ATTACH_TYPEIs it typical to have different BPF programs for different attach types? Otherwise, (different types may have similar BPF programs), maybe we can pass type as an argument to the program (shared among different types)?
Not quite sure I am entirely following you, but I consider the various attach types to be quite different and thus you can not really reuse the same BPF program with 2 different attach types.
In my view, we have 4 attach types: - BPF_HID_ATTACH_DEVICE_EVENT: called whenever we receive an IRQ from the given device (so this is net-like event stream) - BPF_HID_ATTACH_RDESC_FIXUP: there can be only one of this type, and this is called to change the device capabilities. So you can not reuse the other programs for this one - BPF_HID_ATTACH_USER_EVENT: called explicitly by the userspace process owning the program. There we can use functions that are sleeping (we are not in IRQ context), so this is also fundamentally different from the 3 others. - BPF_HID_ATTACH_DRIVER_EVENT: whenever the driver gets called into, we get a bpf program run. This can be suspend/resume, or even specific request to the device (change a feature on the device or get its current state). Again, IMO fundamentally different from the others.
So I'm open to any suggestions, but if we can keep the userspace API being defined with different SEC in libbpf, that would be the best.
[...]
+struct hid_device;
+enum hid_bpf_event {
HID_BPF_UNDEF = 0,HID_BPF_DEVICE_EVENT, /* when attach type is BPF_HID_DEVICE_EVENT */+};
+struct hid_bpf_ctx {
enum hid_bpf_event type; /* read-only */__u16 allocated_size; /* the allocated size of data below (RO) */There is a (6-byte?) hole here.
struct hid_device *hdev; /* read-only */__u16 size; /* used size in data (RW) */__u8 data[]; /* data buffer (RW) */+};
Do we really need hit_bpf_ctx in uapi? Maybe we can just use it from vmlinuxh?
I had a thought at this context today, and I think I am getting to the limit of what I understand.
My first worry is that the way I wrote it there, with a variable data field length is that this is not forward compatible. Unless BTF and CORE are making magic, this will bite me in the long run IMO.
But then, you are talking about not using uapi, and I am starting to wonder: am I doing the things correctly?
To solve my first issue (and the weird API I had to introduce in the bpf_hid_get/set_data), I came up to the following: instead of exporting the data directly in the context, I could create a helper bpf_hid_get_data_buf(ctx, const uint size) that returns a RET_PTR_TO_ALLOC_MEM_OR_NULL in the same way bpf_ringbuf_reserve() does.
This way, I can directly access the fields within the bpf program without having to worry about the size.
But now, I am wondering whether the uapi I defined here is correct in the way CORE works.
My goal is to have HID-BPF programs to be CORE compatible, and not have to recompile them depending on the underlying kernel.
I can not understand right now if I need to add some other BTF helpers in the same way the access to struct xdp_md and struct xdp_buff are converted between one and other, or if defining a forward compatible struct hid_bpf_ctx is enough. As far as I understand, .convert_ctx_access allows to export a stable uapi to the bpf prog users with the verifier doing the conversion between the structs for me. But is this really required for all the BPF programs if we want them to be CORE?
Also, I am starting to wonder if I should not hide fields in the context to the users. The .data field could be a pointer and only accessed through the helper I mentioned above. This would be forward compatible, and also allows to use whatever available memory in the kernel to be forwarded to the BPF program. This way I can skip the memcpy part and work directly with the incoming dma data buffer from the IRQ.
But is it best practice to do such a thing?
Cheers, Benjamin
[...]
+static bool hid_is_valid_access(int off, int size,
enum bpf_access_type access_type,const struct bpf_prog *prog,struct bpf_insn_access_aux *info)+{
/* everything not in ctx is prohibited */if (off < 0 || off + size > sizeof(struct hid_bpf_ctx) + HID_BPF_MIN_BUFFER_SIZE)return false;Mabe add the following here to fail unaligned accesses
if (off % size != 0) return false;[...]
On Mon, Mar 7, 2022 at 10:39 AM Benjamin Tissoires benjamin.tissoires@redhat.com wrote:
On Sat, Mar 5, 2022 at 1:03 AM Song Liu song@kernel.org wrote:
On Fri, Mar 4, 2022 at 9:31 AM Benjamin Tissoires benjamin.tissoires@redhat.com wrote:
HID is a protocol that could benefit from using BPF too.
[...]
+#include <linux/list.h> +#include <linux/slab.h>
+struct bpf_prog; +struct bpf_prog_array; +struct hid_device;
+enum bpf_hid_attach_type {
BPF_HID_ATTACH_INVALID = -1,BPF_HID_ATTACH_DEVICE_EVENT = 0,MAX_BPF_HID_ATTACH_TYPEIs it typical to have different BPF programs for different attach types? Otherwise, (different types may have similar BPF programs), maybe we can pass type as an argument to the program (shared among different types)?
Not quite sure I am entirely following you, but I consider the various attach types to be quite different and thus you can not really reuse the same BPF program with 2 different attach types.
In my view, we have 4 attach types:
- BPF_HID_ATTACH_DEVICE_EVENT: called whenever we receive an IRQ from
the given device (so this is net-like event stream)
- BPF_HID_ATTACH_RDESC_FIXUP: there can be only one of this type, and
this is called to change the device capabilities. So you can not reuse the other programs for this one
- BPF_HID_ATTACH_USER_EVENT: called explicitly by the userspace
process owning the program. There we can use functions that are sleeping (we are not in IRQ context), so this is also fundamentally different from the 3 others.
- BPF_HID_ATTACH_DRIVER_EVENT: whenever the driver gets called into,
we get a bpf program run. This can be suspend/resume, or even specific request to the device (change a feature on the device or get its current state). Again, IMO fundamentally different from the others.
So I'm open to any suggestions, but if we can keep the userspace API being defined with different SEC in libbpf, that would be the best.
Thanks for this information. Different attach_types sound right for the use case.
[...]
+struct hid_device;
+enum hid_bpf_event {
HID_BPF_UNDEF = 0,HID_BPF_DEVICE_EVENT, /* when attach type is BPF_HID_DEVICE_EVENT */+};
+struct hid_bpf_ctx {
enum hid_bpf_event type; /* read-only */__u16 allocated_size; /* the allocated size of data below (RO) */There is a (6-byte?) hole here.
struct hid_device *hdev; /* read-only */__u16 size; /* used size in data (RW) */__u8 data[]; /* data buffer (RW) */+};
Do we really need hit_bpf_ctx in uapi? Maybe we can just use it from vmlinuxh?
I had a thought at this context today, and I think I am getting to the limit of what I understand.
My first worry is that the way I wrote it there, with a variable data field length is that this is not forward compatible. Unless BTF and CORE are making magic, this will bite me in the long run IMO.
But then, you are talking about not using uapi, and I am starting to wonder: am I doing the things correctly?
To solve my first issue (and the weird API I had to introduce in the bpf_hid_get/set_data), I came up to the following: instead of exporting the data directly in the context, I could create a helper bpf_hid_get_data_buf(ctx, const uint size) that returns a RET_PTR_TO_ALLOC_MEM_OR_NULL in the same way bpf_ringbuf_reserve() does.
This way, I can directly access the fields within the bpf program without having to worry about the size.
But now, I am wondering whether the uapi I defined here is correct in the way CORE works.
My goal is to have HID-BPF programs to be CORE compatible, and not have to recompile them depending on the underlying kernel.
I can not understand right now if I need to add some other BTF helpers in the same way the access to struct xdp_md and struct xdp_buff are converted between one and other, or if defining a forward compatible struct hid_bpf_ctx is enough. As far as I understand, .convert_ctx_access allows to export a stable uapi to the bpf prog users with the verifier doing the conversion between the structs for me. But is this really required for all the BPF programs if we want them to be CORE?
Also, I am starting to wonder if I should not hide fields in the context to the users. The .data field could be a pointer and only accessed through the helper I mentioned above. This would be forward compatible, and also allows to use whatever available memory in the kernel to be forwarded to the BPF program. This way I can skip the memcpy part and work directly with the incoming dma data buffer from the IRQ.
But is it best practice to do such a thing?
I think .convert_ctx_access is the way to go if we want to access the data buffer without memcpy. I am not sure how much work is needed to make it compatible with CORE though.
To make sure I understand the case, do we want something like
bpf_prog(struct hid_bpf_ctx *ctx) { /* makes sure n < ctx->size */ x = ctx->data[n]; /* read data */ ctx->data[n] = <something>; /* write data */ ctx->size = <something <= n>; /* change data size */ }
We also need it to be CORE, so that we may modify hid_bpf_ctx by inserting more members to it before data.
Is this accurate?
Song
On Tue, Mar 8, 2022 at 1:57 AM Song Liu song@kernel.org wrote:
On Mon, Mar 7, 2022 at 10:39 AM Benjamin Tissoires benjamin.tissoires@redhat.com wrote:
On Sat, Mar 5, 2022 at 1:03 AM Song Liu song@kernel.org wrote:
On Fri, Mar 4, 2022 at 9:31 AM Benjamin Tissoires benjamin.tissoires@redhat.com wrote:
HID is a protocol that could benefit from using BPF too.
[...]
+#include <linux/list.h> +#include <linux/slab.h>
+struct bpf_prog; +struct bpf_prog_array; +struct hid_device;
+enum bpf_hid_attach_type {
BPF_HID_ATTACH_INVALID = -1,BPF_HID_ATTACH_DEVICE_EVENT = 0,MAX_BPF_HID_ATTACH_TYPEIs it typical to have different BPF programs for different attach types? Otherwise, (different types may have similar BPF programs), maybe we can pass type as an argument to the program (shared among different types)?
Not quite sure I am entirely following you, but I consider the various attach types to be quite different and thus you can not really reuse the same BPF program with 2 different attach types.
In my view, we have 4 attach types:
- BPF_HID_ATTACH_DEVICE_EVENT: called whenever we receive an IRQ from
the given device (so this is net-like event stream)
- BPF_HID_ATTACH_RDESC_FIXUP: there can be only one of this type, and
this is called to change the device capabilities. So you can not reuse the other programs for this one
- BPF_HID_ATTACH_USER_EVENT: called explicitly by the userspace
process owning the program. There we can use functions that are sleeping (we are not in IRQ context), so this is also fundamentally different from the 3 others.
- BPF_HID_ATTACH_DRIVER_EVENT: whenever the driver gets called into,
we get a bpf program run. This can be suspend/resume, or even specific request to the device (change a feature on the device or get its current state). Again, IMO fundamentally different from the others.
So I'm open to any suggestions, but if we can keep the userspace API being defined with different SEC in libbpf, that would be the best.
Thanks for this information. Different attach_types sound right for the use case.
[...]
+struct hid_device;
+enum hid_bpf_event {
HID_BPF_UNDEF = 0,HID_BPF_DEVICE_EVENT, /* when attach type is BPF_HID_DEVICE_EVENT */+};
+struct hid_bpf_ctx {
enum hid_bpf_event type; /* read-only */__u16 allocated_size; /* the allocated size of data below (RO) */There is a (6-byte?) hole here.
struct hid_device *hdev; /* read-only */__u16 size; /* used size in data (RW) */__u8 data[]; /* data buffer (RW) */+};
Do we really need hit_bpf_ctx in uapi? Maybe we can just use it from vmlinuxh?
I had a thought at this context today, and I think I am getting to the limit of what I understand.
My first worry is that the way I wrote it there, with a variable data field length is that this is not forward compatible. Unless BTF and CORE are making magic, this will bite me in the long run IMO.
But then, you are talking about not using uapi, and I am starting to wonder: am I doing the things correctly?
To solve my first issue (and the weird API I had to introduce in the bpf_hid_get/set_data), I came up to the following: instead of exporting the data directly in the context, I could create a helper bpf_hid_get_data_buf(ctx, const uint size) that returns a RET_PTR_TO_ALLOC_MEM_OR_NULL in the same way bpf_ringbuf_reserve() does.
This way, I can directly access the fields within the bpf program without having to worry about the size.
But now, I am wondering whether the uapi I defined here is correct in the way CORE works.
My goal is to have HID-BPF programs to be CORE compatible, and not have to recompile them depending on the underlying kernel.
I can not understand right now if I need to add some other BTF helpers in the same way the access to struct xdp_md and struct xdp_buff are converted between one and other, or if defining a forward compatible struct hid_bpf_ctx is enough. As far as I understand, .convert_ctx_access allows to export a stable uapi to the bpf prog users with the verifier doing the conversion between the structs for me. But is this really required for all the BPF programs if we want them to be CORE?
Also, I am starting to wonder if I should not hide fields in the context to the users. The .data field could be a pointer and only accessed through the helper I mentioned above. This would be forward compatible, and also allows to use whatever available memory in the kernel to be forwarded to the BPF program. This way I can skip the memcpy part and work directly with the incoming dma data buffer from the IRQ.
But is it best practice to do such a thing?
I think .convert_ctx_access is the way to go if we want to access the data buffer without memcpy. I am not sure how much work is needed to make it compatible with CORE though.
To make sure I understand the case, do we want something like
bpf_prog(struct hid_bpf_ctx *ctx) { /* makes sure n < ctx->size */ x = ctx->data[n]; /* read data */ ctx->data[n] = <something>; /* write data */ ctx->size = <something <= n>; /* change data size */ }
We also need it to be CORE, so that we may modify hid_bpf_ctx by inserting more members to it before data.
Is this accurate?
Yes, you pretty much summed it all (except maybe that we might want to have allocated_size in addition to size so we can also grow the value of .size within the allocated limit).
All in all, what I want for HID bpf programs is to be able to read and write an array of bytes, and change its size within an allocated kernel limit.
This will apply to every HID bpf attach type, to the exception of some BPF_HID_ATTACH_DRIVER_EVENT when we are receiving a suspend/resume notification. Though in the suspend/resume case we won't have the data array available, so it won't matter much.
I want the HID bpf programs to be CORE, but if you tell me that it would matter only if we need to reshuffle hid_bpf_ctx, I would be fine simply put a comment "new fields must be added at the end" like some other definitions of contexts are doing.
Besides that, I currently do not want to allow access to the content of struct hid_device (or any other kernel struct) in HID BPF programs. That might be of interest at some point for debugging, but with just the array capability I should be able to achieve all of my use cases.
Cheers, Benjamin
On Tue, Mar 8, 2022 at 10:20 AM Benjamin Tissoires benjamin.tissoires@redhat.com wrote:
On Tue, Mar 8, 2022 at 1:57 AM Song Liu song@kernel.org wrote:
On Mon, Mar 7, 2022 at 10:39 AM Benjamin Tissoires benjamin.tissoires@redhat.com wrote:
On Sat, Mar 5, 2022 at 1:03 AM Song Liu song@kernel.org wrote:
On Fri, Mar 4, 2022 at 9:31 AM Benjamin Tissoires benjamin.tissoires@redhat.com wrote:
[...]
+struct hid_bpf_ctx {
enum hid_bpf_event type; /* read-only */__u16 allocated_size; /* the allocated size of data below (RO) */There is a (6-byte?) hole here.
struct hid_device *hdev; /* read-only */__u16 size; /* used size in data (RW) */__u8 data[]; /* data buffer (RW) */+};
Do we really need hit_bpf_ctx in uapi? Maybe we can just use it from vmlinuxh?
I had a thought at this context today, and I think I am getting to the limit of what I understand.
My first worry is that the way I wrote it there, with a variable data field length is that this is not forward compatible. Unless BTF and CORE are making magic, this will bite me in the long run IMO.
But then, you are talking about not using uapi, and I am starting to wonder: am I doing the things correctly?
To solve my first issue (and the weird API I had to introduce in the bpf_hid_get/set_data), I came up to the following: instead of exporting the data directly in the context, I could create a helper bpf_hid_get_data_buf(ctx, const uint size) that returns a RET_PTR_TO_ALLOC_MEM_OR_NULL in the same way bpf_ringbuf_reserve() does.
This way, I can directly access the fields within the bpf program without having to worry about the size.
But now, I am wondering whether the uapi I defined here is correct in the way CORE works.
My goal is to have HID-BPF programs to be CORE compatible, and not have to recompile them depending on the underlying kernel.
I can not understand right now if I need to add some other BTF helpers in the same way the access to struct xdp_md and struct xdp_buff are converted between one and other, or if defining a forward compatible struct hid_bpf_ctx is enough. As far as I understand, .convert_ctx_access allows to export a stable uapi to the bpf prog users with the verifier doing the conversion between the structs for me. But is this really required for all the BPF programs if we want them to be CORE?
Also, I am starting to wonder if I should not hide fields in the context to the users. The .data field could be a pointer and only accessed through the helper I mentioned above. This would be forward compatible, and also allows to use whatever available memory in the kernel to be forwarded to the BPF program. This way I can skip the memcpy part and work directly with the incoming dma data buffer from the IRQ.
But is it best practice to do such a thing?
I think .convert_ctx_access is the way to go if we want to access the data buffer without memcpy. I am not sure how much work is needed to make it compatible with CORE though.
I spent the week working on that, and I am really amazed at how smart and simple the .convert_ctx_access is working.
With that in place, I can hide the internal fields and export "virtual" public fields that are remapped on the fly by the kernel at load time :)
To make sure I understand the case, do we want something like
bpf_prog(struct hid_bpf_ctx *ctx) { /* makes sure n < ctx->size */ x = ctx->data[n]; /* read data */ ctx->data[n] = <something>; /* write data */ ctx->size = <something <= n>; /* change data size */ }
We also need it to be CORE, so that we may modify hid_bpf_ctx by inserting more members to it before data.
Is this accurate?
I have been trying to implement that, based on the PTR_TO_PACKET implementation. It works, but is kind of verbose while using it because we need to teach the verifier about the size of the arrays.
For instance, I have the following:
int bpf_prog(struct hid_bpf_ctx *ctx) { /* we need to store a pointer on the stack to store its accessible size */ __u8 *data = ctx->data;
if (data + 3 > ctx->data_end) return 0; /* EPERM, bounds check */
data[1] = data[0] + 3;
return 0; }
This is OK, but it gets worse if I want to access a random offset:
__s64 offset = 0; int bpf_prog(struct hid_bpf_ctx *ctx) { __u8 *data = ctx->data; __u16 *x;
/* first assign the size of data */ if (data + 4 > ctx->data_end) return 0; /* EPERM, bounds check */
/* teach the verifier the range of offset (needs to be a s64) */ if (offset >= 0 && offset < 2) { x = (__u16 *)&data[offset];
/* now store the size of x in its register */ if (x + 2 > ctx->data_end) return 0;
/* finally we can read/write the data */ *x += 1; }
return 0; }
OTOH, I managed to define a simpler helper that allows me to return a pointer to the internal data:
BPF_CALL_3(bpf_hid_get_data, struct hid_bpf_ctx_kern*, ctx, u64, offset, u64, size) { if (!size) return (unsigned long)0;
if (offset + size > ctx->data_end - ctx->data) return (unsigned long)0;
return (unsigned long)(ctx->data + offset); }
static const struct bpf_func_proto bpf_hid_get_data_proto = { .func = bpf_hid_get_data, .gpl_only = true, .ret_type = RET_PTR_TO_ALLOC_MEM_OR_NULL, .arg1_type = ARG_PTR_TO_CTX, .arg2_type = ARG_ANYTHING, .arg3_type = ARG_CONST_ALLOC_SIZE_OR_ZERO, };
Which makes the previous bpf code into:
__u64 offset = 0; int bpf_prog(struct hid_bpf_ctx *ctx) { __u16 *x = bpf_hid_get_data(ctx, offset, 2);
if (!x) return 0; /* EPERM, bounds check */
*x += 1;
return 0; }
The advantage of both of those solutions is that they are removing the need for bpf_hid_{get|set}_bytes().
The second solution is much simpler in terms of usage, but it doesn't feel "right" to have to reimplement the wheel when we should have direct array accesses. OTOH, the first solution with the packet implementation is bulky and requiring users to do that for every single access of the data is IMO way too much...
Any thoughts?
Cheers, Benjamin
On Fri, Mar 4, 2022 at 9:31 AM Benjamin Tissoires benjamin.tissoires@redhat.com wrote:
[...]
+#endif
+static inline bool bpf_hid_link_empty(struct bpf_hid *bpf,
enum bpf_hid_attach_type type)+{
return list_empty(&bpf->links[type]);+}
+struct bpf_hid_hooks {
struct hid_device *(*hdev_from_fd)(int fd);int (*link_attach)(struct hid_device *hdev, enum bpf_hid_attach_type type);void (*array_detached)(struct hid_device *hdev, enum bpf_hid_attach_type type);
shall we call this array_detach()? detached sounds like a function that checks the status of link/hook.
[...]
--- /dev/null +++ b/kernel/bpf/hid.c @@ -0,0 +1,437 @@ +// SPDX-License-Identifier: GPL-2.0 +/*
- based on kernel/bpf/net-namespace.c
- */
I guess we don't need this comment.
+#include <linux/bpf.h> +#include <linux/bpf-hid.h> +#include <linux/filter.h> +#include <linux/hid.h> +#include <linux/hidraw.h>
[...]
Now that BPF can be compatible with HID, add the capability into HID. drivers/hid/hid-bpf.c takes care of the glue between bpf and HID, and hid-core can then inject any incoming event from the device into a BPF program to filter/analyze it.
Signed-off-by: Benjamin Tissoires benjamin.tissoires@redhat.com
---
changes in v2: - split the series by bpf/libbpf/hid/selftests and samples - addressed review comments from v1 --- drivers/hid/Makefile | 1 + drivers/hid/hid-bpf.c | 157 +++++++++++++++++++++++++++++++++++++++++ drivers/hid/hid-core.c | 21 +++++- include/linux/hid.h | 11 +++ 4 files changed, 187 insertions(+), 3 deletions(-) create mode 100644 drivers/hid/hid-bpf.c
diff --git a/drivers/hid/Makefile b/drivers/hid/Makefile index 6d3e630e81af..08d2d7619937 100644 --- a/drivers/hid/Makefile +++ b/drivers/hid/Makefile @@ -4,6 +4,7 @@ # hid-y := hid-core.o hid-input.o hid-quirks.o hid-$(CONFIG_DEBUG_FS) += hid-debug.o +hid-$(CONFIG_BPF) += hid-bpf.o
obj-$(CONFIG_HID) += hid.o obj-$(CONFIG_UHID) += uhid.o diff --git a/drivers/hid/hid-bpf.c b/drivers/hid/hid-bpf.c new file mode 100644 index 000000000000..8120e598de9f --- /dev/null +++ b/drivers/hid/hid-bpf.c @@ -0,0 +1,157 @@ +// SPDX-License-Identifier: GPL-2.0-or-later +/* + * BPF in HID support for Linux + * + * Copyright (c) 2022 Benjamin Tissoires + */ + +#include <linux/filter.h> +#include <linux/mutex.h> +#include <linux/slab.h> + +#include <uapi/linux/bpf_hid.h> +#include <linux/hid.h> + +static int __hid_bpf_match_sysfs(struct device *dev, const void *data) +{ + struct kernfs_node *kn = dev->kobj.sd; + struct kernfs_node *uevent_kn; + + uevent_kn = kernfs_find_and_get_ns(kn, "uevent", NULL); + + return uevent_kn == data; +} + +static struct hid_device *hid_bpf_fd_to_hdev(int fd) +{ + struct device *dev; + struct hid_device *hdev; + struct fd f = fdget(fd); + struct inode *inode; + struct kernfs_node *node; + + if (!f.file) { + hdev = ERR_PTR(-EBADF); + goto out; + } + + inode = file_inode(f.file); + node = inode->i_private; + + dev = bus_find_device(&hid_bus_type, NULL, node, __hid_bpf_match_sysfs); + + if (dev) + hdev = to_hid_device(dev); + else + hdev = ERR_PTR(-EINVAL); + + out: + fdput(f); + return hdev; +} + +static int hid_bpf_link_attach(struct hid_device *hdev, enum bpf_hid_attach_type type) +{ + int err = 0; + + switch (type) { + case BPF_HID_ATTACH_DEVICE_EVENT: + if (!hdev->bpf.ctx) { + hdev->bpf.ctx = bpf_hid_allocate_ctx(hdev, HID_BPF_MAX_BUFFER_SIZE); + if (IS_ERR(hdev->bpf.ctx)) { + err = PTR_ERR(hdev->bpf.ctx); + hdev->bpf.ctx = NULL; + } + } + break; + default: + /* do nothing */ + } + + return err; +} + +static void hid_bpf_array_detached(struct hid_device *hdev, enum bpf_hid_attach_type type) +{ + switch (type) { + case BPF_HID_ATTACH_DEVICE_EVENT: + kfree(hdev->bpf.ctx); + hdev->bpf.ctx = NULL; + break; + default: + /* do nothing */ + } +} + +static int hid_bpf_run_progs(struct hid_device *hdev, enum bpf_hid_attach_type type, + struct hid_bpf_ctx *ctx, u8 *data, int size) +{ + enum hid_bpf_event event = HID_BPF_UNDEF; + + if (type < 0 || !ctx) + return -EINVAL; + + if (size > ctx->allocated_size) + return -E2BIG; + + switch (type) { + case BPF_HID_ATTACH_DEVICE_EVENT: + event = HID_BPF_DEVICE_EVENT; + break; + default: + return -EINVAL; + } + + if (!hdev->bpf.run_array[type]) + return 0; + + memset(ctx->data, 0, ctx->allocated_size); + ctx->type = event; + + if (size && data) { + memcpy(ctx->data, data, size); + ctx->size = size; + } else { + ctx->size = 0; + } + + return BPF_PROG_RUN_ARRAY(hdev->bpf.run_array[type], ctx, bpf_prog_run); +} + +u8 *hid_bpf_raw_event(struct hid_device *hdev, u8 *data, int *size) +{ + int ret; + + if (bpf_hid_link_empty(&hdev->bpf, BPF_HID_ATTACH_DEVICE_EVENT)) + return data; + + ret = hid_bpf_run_progs(hdev, BPF_HID_ATTACH_DEVICE_EVENT, + hdev->bpf.ctx, data, *size); + if (ret) + return ERR_PTR(ret); + + if (!hdev->bpf.ctx->size) + return ERR_PTR(-EINVAL); + + *size = hdev->bpf.ctx->size; + + return hdev->bpf.ctx->data; +} + +int __init hid_bpf_module_init(void) +{ + struct bpf_hid_hooks hooks = { + .hdev_from_fd = hid_bpf_fd_to_hdev, + .link_attach = hid_bpf_link_attach, + .array_detached = hid_bpf_array_detached, + }; + + bpf_hid_set_hooks(&hooks); + + return 0; +} + +void __exit hid_bpf_module_exit(void) +{ + bpf_hid_set_hooks(NULL); +} diff --git a/drivers/hid/hid-core.c b/drivers/hid/hid-core.c index f1aed5bbd000..a80bffe6ce4a 100644 --- a/drivers/hid/hid-core.c +++ b/drivers/hid/hid-core.c @@ -1748,13 +1748,21 @@ int hid_report_raw_event(struct hid_device *hid, int type, u8 *data, u32 size, struct hid_driver *hdrv; unsigned int a; u32 rsize, csize = size; - u8 *cdata = data; + u8 *cdata; int ret = 0;
+ data = hid_bpf_raw_event(hid, data, &size); + if (IS_ERR(data)) { + ret = PTR_ERR(data); + goto out; + } + report = hid_get_report(report_enum, data); if (!report) goto out;
+ cdata = data; + if (report_enum->numbered) { cdata++; csize--; @@ -2528,10 +2536,12 @@ int hid_add_device(struct hid_device *hdev)
hid_debug_register(hdev, dev_name(&hdev->dev)); ret = device_add(&hdev->dev); - if (!ret) + if (!ret) { hdev->status |= HID_STAT_ADDED; - else + } else { hid_debug_unregister(hdev); + bpf_hid_exit(hdev); + }
return ret; } @@ -2567,6 +2577,7 @@ struct hid_device *hid_allocate_device(void) spin_lock_init(&hdev->debug_list_lock); sema_init(&hdev->driver_input_lock, 1); mutex_init(&hdev->ll_open_lock); + bpf_hid_init(hdev);
return hdev; } @@ -2574,6 +2585,7 @@ EXPORT_SYMBOL_GPL(hid_allocate_device);
static void hid_remove_device(struct hid_device *hdev) { + bpf_hid_exit(hdev); if (hdev->status & HID_STAT_ADDED) { device_del(&hdev->dev); hid_debug_unregister(hdev); @@ -2700,6 +2712,8 @@ static int __init hid_init(void)
hid_debug_init();
+ hid_bpf_module_init(); + return 0; err_bus: bus_unregister(&hid_bus_type); @@ -2709,6 +2723,7 @@ static int __init hid_init(void)
static void __exit hid_exit(void) { + hid_bpf_module_exit(); hid_debug_exit(); hidraw_exit(); bus_unregister(&hid_bus_type); diff --git a/include/linux/hid.h b/include/linux/hid.h index 56f6f4ad45a7..8fd79011f461 100644 --- a/include/linux/hid.h +++ b/include/linux/hid.h @@ -27,6 +27,7 @@ #include <linux/mutex.h> #include <linux/power_supply.h> #include <uapi/linux/hid.h> +#include <uapi/linux/bpf_hid.h>
/* * We parse each description item into this structure. Short items data @@ -1210,4 +1211,14 @@ do { \ #define hid_dbg_once(hid, fmt, ...) \ dev_dbg_once(&(hid)->dev, fmt, ##__VA_ARGS__)
+#ifdef CONFIG_BPF +u8 *hid_bpf_raw_event(struct hid_device *hdev, u8 *rd, int *size); +int hid_bpf_module_init(void); +void hid_bpf_module_exit(void); +#else +static inline u8 *hid_bpf_raw_event(struct hid_device *hdev, u8 *rd, int *size) { return rd; } +static inline int hid_bpf_module_init(void) { return 0; } +static inline void hid_bpf_module_exit(void) {} +#endif + #endif
On Fri, Mar 04, 2022 at 06:28:27PM +0100, Benjamin Tissoires wrote:
Now that BPF can be compatible with HID, add the capability into HID. drivers/hid/hid-bpf.c takes care of the glue between bpf and HID, and hid-core can then inject any incoming event from the device into a BPF program to filter/analyze it.
Signed-off-by: Benjamin Tissoires benjamin.tissoires@redhat.com
Reviewed-by: Greg Kroah-Hartman gregkh@linuxfoundation.org
On Fri, Mar 4, 2022 at 9:31 AM Benjamin Tissoires benjamin.tissoires@redhat.com wrote:
Now that BPF can be compatible with HID, add the capability into HID. drivers/hid/hid-bpf.c takes care of the glue between bpf and HID, and hid-core can then inject any incoming event from the device into a BPF program to filter/analyze it.
Signed-off-by: Benjamin Tissoires benjamin.tissoires@redhat.com
[...]
+static int hid_bpf_link_attach(struct hid_device *hdev, enum bpf_hid_attach_type type) +{
int err = 0;switch (type) {case BPF_HID_ATTACH_DEVICE_EVENT:if (!hdev->bpf.ctx) {hdev->bpf.ctx = bpf_hid_allocate_ctx(hdev, HID_BPF_MAX_BUFFER_SIZE);if (IS_ERR(hdev->bpf.ctx)) {err = PTR_ERR(hdev->bpf.ctx);hdev->bpf.ctx = NULL;}}break;default:/* do nothing */
Do we need to show warning and/or return EINVAL here?
}return err;+}
+static void hid_bpf_array_detached(struct hid_device *hdev, enum bpf_hid_attach_type type) +{
switch (type) {case BPF_HID_ATTACH_DEVICE_EVENT:kfree(hdev->bpf.ctx);hdev->bpf.ctx = NULL;break;default:/* do nothing */
ditto
[...]
Hi Benjamin,
On 04/03/2022 19:28, Benjamin Tissoires wrote:
Now that BPF can be compatible with HID, add the capability into HID. drivers/hid/hid-bpf.c takes care of the glue between bpf and HID, and hid-core can then inject any incoming event from the device into a BPF program to filter/analyze it.
Signed-off-by: Benjamin Tissoires benjamin.tissoires@redhat.com
changes in v2:
- split the series by bpf/libbpf/hid/selftests and samples
- addressed review comments from v1
drivers/hid/Makefile | 1 + drivers/hid/hid-bpf.c | 157 +++++++++++++++++++++++++++++++++++++++++ drivers/hid/hid-core.c | 21 +++++- include/linux/hid.h | 11 +++ 4 files changed, 187 insertions(+), 3 deletions(-) create mode 100644 drivers/hid/hid-bpf.c
diff --git a/drivers/hid/Makefile b/drivers/hid/Makefile index 6d3e630e81af..08d2d7619937 100644 --- a/drivers/hid/Makefile +++ b/drivers/hid/Makefile @@ -4,6 +4,7 @@ # hid-y := hid-core.o hid-input.o hid-quirks.o hid-$(CONFIG_DEBUG_FS) += hid-debug.o +hid-$(CONFIG_BPF) += hid-bpf.o obj-$(CONFIG_HID) += hid.o obj-$(CONFIG_UHID) += uhid.o diff --git a/drivers/hid/hid-bpf.c b/drivers/hid/hid-bpf.c new file mode 100644 index 000000000000..8120e598de9f --- /dev/null +++ b/drivers/hid/hid-bpf.c @@ -0,0 +1,157 @@ +// SPDX-License-Identifier: GPL-2.0-or-later +/*
- BPF in HID support for Linux
- Copyright (c) 2022 Benjamin Tissoires
- */
+#include <linux/filter.h> +#include <linux/mutex.h> +#include <linux/slab.h>
+#include <uapi/linux/bpf_hid.h> +#include <linux/hid.h>
+static int __hid_bpf_match_sysfs(struct device *dev, const void *data) +{
- struct kernfs_node *kn = dev->kobj.sd;
- struct kernfs_node *uevent_kn;
- uevent_kn = kernfs_find_and_get_ns(kn, "uevent", NULL);
- return uevent_kn == data;
+}
+static struct hid_device *hid_bpf_fd_to_hdev(int fd) +{
- struct device *dev;
- struct hid_device *hdev;
- struct fd f = fdget(fd);
- struct inode *inode;
- struct kernfs_node *node;
- if (!f.file) {
hdev = ERR_PTR(-EBADF);goto out;- }
- inode = file_inode(f.file);
- node = inode->i_private;
- dev = bus_find_device(&hid_bus_type, NULL, node, __hid_bpf_match_sysfs);
- if (dev)
hdev = to_hid_device(dev);- else
hdev = ERR_PTR(-EINVAL);- out:
- fdput(f);
- return hdev;
+}
+static int hid_bpf_link_attach(struct hid_device *hdev, enum bpf_hid_attach_type type) +{
- int err = 0;
- switch (type) {
- case BPF_HID_ATTACH_DEVICE_EVENT:
if (!hdev->bpf.ctx) {hdev->bpf.ctx = bpf_hid_allocate_ctx(hdev, HID_BPF_MAX_BUFFER_SIZE);if (IS_ERR(hdev->bpf.ctx)) {err = PTR_ERR(hdev->bpf.ctx);hdev->bpf.ctx = NULL;}}break;- default:
/* do nothing */
These cause following error:
CC     drivers/hid/hid-bpf.o drivers/hid/hid-bpf.c: In function ‘hid_bpf_link_attach’: drivers/hid/hid-bpf.c:88:2: error: label at end of compound statement   88 | default:      | ^~~~~~~ drivers/hid/hid-bpf.c: In function ‘hid_bpf_link_attached’: drivers/hid/hid-bpf.c:101:2: error: label at end of compound statement  101 | default:      | ^~~~~~~ drivers/hid/hid-bpf.c: In function ‘hid_bpf_array_detached’: drivers/hid/hid-bpf.c:116:2: error: label at end of compound statement  116 | default:      | ^~~~~~~ make[2]: *** [scripts/Makefile.build:288: drivers/hid/hid-bpf.o] Error 1 make[1]: *** [scripts/Makefile.build:550: drivers/hid] Error 2 make[1]: *** Waiting for unfinished jobs.... make: *** [Makefile:1831: drivers] Error 2
To fix that, you need to add a break statement at end:
default:
/* do nothing */
break;
Same for couple of other occurrences in the file.
-Tero
HID-bpf program type are needing a new SEC. To bind a hid-bpf program, we can rely on bpf_program__attach_fd() so export a new function to the API.
Signed-off-by: Benjamin Tissoires benjamin.tissoires@redhat.com
---
changes in v2: - split the series by bpf/libbpf/hid/selftests and samples --- tools/lib/bpf/libbpf.c | 7 +++++++ tools/lib/bpf/libbpf.h | 2 ++ tools/lib/bpf/libbpf.map | 1 + 3 files changed, 10 insertions(+)
diff --git a/tools/lib/bpf/libbpf.c b/tools/lib/bpf/libbpf.c index 81bf01d67671..356bbd3ad2c7 100644 --- a/tools/lib/bpf/libbpf.c +++ b/tools/lib/bpf/libbpf.c @@ -8680,6 +8680,7 @@ static const struct bpf_sec_def section_defs[] = { SEC_DEF("cgroup/setsockopt", CGROUP_SOCKOPT, BPF_CGROUP_SETSOCKOPT, SEC_ATTACHABLE | SEC_SLOPPY_PFX), SEC_DEF("struct_ops+", STRUCT_OPS, 0, SEC_NONE), SEC_DEF("sk_lookup", SK_LOOKUP, BPF_SK_LOOKUP, SEC_ATTACHABLE | SEC_SLOPPY_PFX), + SEC_DEF("hid/device_event", HID, BPF_HID_DEVICE_EVENT, SEC_ATTACHABLE_OPT | SEC_SLOPPY_PFX), };
#define MAX_TYPE_NAME_SIZE 32 @@ -10659,6 +10660,12 @@ static struct bpf_link *attach_iter(const struct bpf_program *prog, long cookie) return bpf_program__attach_iter(prog, NULL); }
+struct bpf_link * +bpf_program__attach_hid(const struct bpf_program *prog, int hid_fd) +{ + return bpf_program__attach_fd(prog, hid_fd, 0, "hid"); +} + struct bpf_link *bpf_program__attach(const struct bpf_program *prog) { if (!prog->sec_def || !prog->sec_def->attach_fn) diff --git a/tools/lib/bpf/libbpf.h b/tools/lib/bpf/libbpf.h index c8d8daad212e..f677ac0a9ede 100644 --- a/tools/lib/bpf/libbpf.h +++ b/tools/lib/bpf/libbpf.h @@ -529,6 +529,8 @@ struct bpf_iter_attach_opts { LIBBPF_API struct bpf_link * bpf_program__attach_iter(const struct bpf_program *prog, const struct bpf_iter_attach_opts *opts); +LIBBPF_API struct bpf_link * +bpf_program__attach_hid(const struct bpf_program *prog, int hid_fd);
/* * Libbpf allows callers to adjust BPF programs before being loaded diff --git a/tools/lib/bpf/libbpf.map b/tools/lib/bpf/libbpf.map index 47e70c9058d9..fdc6fa743953 100644 --- a/tools/lib/bpf/libbpf.map +++ b/tools/lib/bpf/libbpf.map @@ -424,6 +424,7 @@ LIBBPF_0.6.0 { LIBBPF_0.7.0 { global: bpf_btf_load; + bpf_program__attach_hid; bpf_program__expected_attach_type; bpf_program__log_buf; bpf_program__log_level;
On Fri, Mar 4, 2022 at 9:31 AM Benjamin Tissoires benjamin.tissoires@redhat.com wrote:
HID-bpf program type are needing a new SEC. To bind a hid-bpf program, we can rely on bpf_program__attach_fd() so export a new function to the API.
Signed-off-by: Benjamin Tissoires benjamin.tissoires@redhat.com
Acked-by: Song Liu songliubraving@fb.com
changes in v2:
- split the series by bpf/libbpf/hid/selftests and samples
tools/lib/bpf/libbpf.c | 7 +++++++ tools/lib/bpf/libbpf.h | 2 ++ tools/lib/bpf/libbpf.map | 1 + 3 files changed, 10 insertions(+)
diff --git a/tools/lib/bpf/libbpf.c b/tools/lib/bpf/libbpf.c index 81bf01d67671..356bbd3ad2c7 100644 --- a/tools/lib/bpf/libbpf.c +++ b/tools/lib/bpf/libbpf.c @@ -8680,6 +8680,7 @@ static const struct bpf_sec_def section_defs[] = { SEC_DEF("cgroup/setsockopt", CGROUP_SOCKOPT, BPF_CGROUP_SETSOCKOPT, SEC_ATTACHABLE | SEC_SLOPPY_PFX), SEC_DEF("struct_ops+", STRUCT_OPS, 0, SEC_NONE), SEC_DEF("sk_lookup", SK_LOOKUP, BPF_SK_LOOKUP, SEC_ATTACHABLE | SEC_SLOPPY_PFX),
SEC_DEF("hid/device_event", HID, BPF_HID_DEVICE_EVENT, SEC_ATTACHABLE_OPT | SEC_SLOPPY_PFX),};
#define MAX_TYPE_NAME_SIZE 32 @@ -10659,6 +10660,12 @@ static struct bpf_link *attach_iter(const struct bpf_program *prog, long cookie) return bpf_program__attach_iter(prog, NULL); }
+struct bpf_link * +bpf_program__attach_hid(const struct bpf_program *prog, int hid_fd) +{
return bpf_program__attach_fd(prog, hid_fd, 0, "hid");+}
struct bpf_link *bpf_program__attach(const struct bpf_program *prog) { if (!prog->sec_def || !prog->sec_def->attach_fn) diff --git a/tools/lib/bpf/libbpf.h b/tools/lib/bpf/libbpf.h index c8d8daad212e..f677ac0a9ede 100644 --- a/tools/lib/bpf/libbpf.h +++ b/tools/lib/bpf/libbpf.h @@ -529,6 +529,8 @@ struct bpf_iter_attach_opts { LIBBPF_API struct bpf_link * bpf_program__attach_iter(const struct bpf_program *prog, const struct bpf_iter_attach_opts *opts); +LIBBPF_API struct bpf_link * +bpf_program__attach_hid(const struct bpf_program *prog, int hid_fd);
/*
- Libbpf allows callers to adjust BPF programs before being loaded
diff --git a/tools/lib/bpf/libbpf.map b/tools/lib/bpf/libbpf.map index 47e70c9058d9..fdc6fa743953 100644 --- a/tools/lib/bpf/libbpf.map +++ b/tools/lib/bpf/libbpf.map @@ -424,6 +424,7 @@ LIBBPF_0.6.0 { LIBBPF_0.7.0 { global: bpf_btf_load;
bpf_program__attach_hid; bpf_program__expected_attach_type; bpf_program__log_buf; bpf_program__log_level;-- 2.35.1
On Fri, Mar 4, 2022 at 9:31 AM Benjamin Tissoires benjamin.tissoires@redhat.com wrote:
HID-bpf program type are needing a new SEC. To bind a hid-bpf program, we can rely on bpf_program__attach_fd() so export a new function to the API.
Signed-off-by: Benjamin Tissoires benjamin.tissoires@redhat.com
changes in v2:
- split the series by bpf/libbpf/hid/selftests and samples
tools/lib/bpf/libbpf.c | 7 +++++++ tools/lib/bpf/libbpf.h | 2 ++ tools/lib/bpf/libbpf.map | 1 + 3 files changed, 10 insertions(+)
diff --git a/tools/lib/bpf/libbpf.c b/tools/lib/bpf/libbpf.c index 81bf01d67671..356bbd3ad2c7 100644 --- a/tools/lib/bpf/libbpf.c +++ b/tools/lib/bpf/libbpf.c @@ -8680,6 +8680,7 @@ static const struct bpf_sec_def section_defs[] = { SEC_DEF("cgroup/setsockopt", CGROUP_SOCKOPT, BPF_CGROUP_SETSOCKOPT, SEC_ATTACHABLE | SEC_SLOPPY_PFX), SEC_DEF("struct_ops+", STRUCT_OPS, 0, SEC_NONE), SEC_DEF("sk_lookup", SK_LOOKUP, BPF_SK_LOOKUP, SEC_ATTACHABLE | SEC_SLOPPY_PFX),
SEC_DEF("hid/device_event", HID, BPF_HID_DEVICE_EVENT, SEC_ATTACHABLE_OPT | SEC_SLOPPY_PFX),
no SEC_SLOPPY_PFX for any new program type, please
};
#define MAX_TYPE_NAME_SIZE 32 @@ -10659,6 +10660,12 @@ static struct bpf_link *attach_iter(const struct bpf_program *prog, long cookie) return bpf_program__attach_iter(prog, NULL); }
+struct bpf_link * +bpf_program__attach_hid(const struct bpf_program *prog, int hid_fd) +{
return bpf_program__attach_fd(prog, hid_fd, 0, "hid");+}
struct bpf_link *bpf_program__attach(const struct bpf_program *prog) { if (!prog->sec_def || !prog->sec_def->attach_fn) diff --git a/tools/lib/bpf/libbpf.h b/tools/lib/bpf/libbpf.h index c8d8daad212e..f677ac0a9ede 100644 --- a/tools/lib/bpf/libbpf.h +++ b/tools/lib/bpf/libbpf.h @@ -529,6 +529,8 @@ struct bpf_iter_attach_opts { LIBBPF_API struct bpf_link * bpf_program__attach_iter(const struct bpf_program *prog, const struct bpf_iter_attach_opts *opts); +LIBBPF_API struct bpf_link * +bpf_program__attach_hid(const struct bpf_program *prog, int hid_fd);
/*
- Libbpf allows callers to adjust BPF programs before being loaded
diff --git a/tools/lib/bpf/libbpf.map b/tools/lib/bpf/libbpf.map index 47e70c9058d9..fdc6fa743953 100644 --- a/tools/lib/bpf/libbpf.map +++ b/tools/lib/bpf/libbpf.map @@ -424,6 +424,7 @@ LIBBPF_0.6.0 { LIBBPF_0.7.0 { global: bpf_btf_load;
bpf_program__attach_hid;
should go into 0.8.0
bpf_program__expected_attach_type; bpf_program__log_buf; bpf_program__log_level;-- 2.35.1
On Mar 7, 2022, at 5:30 PM, Andrii Nakryiko andrii.nakryiko@gmail.com wrote:
On Fri, Mar 4, 2022 at 9:31 AM Benjamin Tissoires benjamin.tissoires@redhat.com wrote:
HID-bpf program type are needing a new SEC. To bind a hid-bpf program, we can rely on bpf_program__attach_fd() so export a new function to the API.
Signed-off-by: Benjamin Tissoires benjamin.tissoires@redhat.com
changes in v2:
- split the series by bpf/libbpf/hid/selftests and samples
tools/lib/bpf/libbpf.c | 7 +++++++ tools/lib/bpf/libbpf.h | 2 ++ tools/lib/bpf/libbpf.map | 1 + 3 files changed, 10 insertions(+)
diff --git a/tools/lib/bpf/libbpf.c b/tools/lib/bpf/libbpf.c index 81bf01d67671..356bbd3ad2c7 100644 --- a/tools/lib/bpf/libbpf.c +++ b/tools/lib/bpf/libbpf.c @@ -8680,6 +8680,7 @@ static const struct bpf_sec_def section_defs[] = { SEC_DEF("cgroup/setsockopt", CGROUP_SOCKOPT, BPF_CGROUP_SETSOCKOPT, SEC_ATTACHABLE | SEC_SLOPPY_PFX), SEC_DEF("struct_ops+", STRUCT_OPS, 0, SEC_NONE), SEC_DEF("sk_lookup", SK_LOOKUP, BPF_SK_LOOKUP, SEC_ATTACHABLE | SEC_SLOPPY_PFX),
SEC_DEF("hid/device_event", HID, BPF_HID_DEVICE_EVENT, SEC_ATTACHABLE_OPT | SEC_SLOPPY_PFX),no SEC_SLOPPY_PFX for any new program type, please
};
#define MAX_TYPE_NAME_SIZE 32 @@ -10659,6 +10660,12 @@ static struct bpf_link *attach_iter(const struct bpf_program *prog, long cookie) return bpf_program__attach_iter(prog, NULL); }
+struct bpf_link * +bpf_program__attach_hid(const struct bpf_program *prog, int hid_fd) +{
return bpf_program__attach_fd(prog, hid_fd, 0, "hid");+}
struct bpf_link *bpf_program__attach(const struct bpf_program *prog) { if (!prog->sec_def || !prog->sec_def->attach_fn) diff --git a/tools/lib/bpf/libbpf.h b/tools/lib/bpf/libbpf.h index c8d8daad212e..f677ac0a9ede 100644 --- a/tools/lib/bpf/libbpf.h +++ b/tools/lib/bpf/libbpf.h @@ -529,6 +529,8 @@ struct bpf_iter_attach_opts { LIBBPF_API struct bpf_link * bpf_program__attach_iter(const struct bpf_program *prog, const struct bpf_iter_attach_opts *opts); +LIBBPF_API struct bpf_link * +bpf_program__attach_hid(const struct bpf_program *prog, int hid_fd);
/*
- Libbpf allows callers to adjust BPF programs before being loaded
diff --git a/tools/lib/bpf/libbpf.map b/tools/lib/bpf/libbpf.map index 47e70c9058d9..fdc6fa743953 100644 --- a/tools/lib/bpf/libbpf.map +++ b/tools/lib/bpf/libbpf.map @@ -424,6 +424,7 @@ LIBBPF_0.6.0 { LIBBPF_0.7.0 { global: bpf_btf_load;
bpf_program__attach_hid;should go into 0.8.0
Ah, I missed this one.
btw, bpf_xdp_attach and buddies should also go into 0.8.0, no?
bpf_program__expected_attach_type; bpf_program__log_buf; bpf_program__log_level;-- 2.35.1
On Mon, Mar 7, 2022 at 5:38 PM Song Liu songliubraving@fb.com wrote:
On Mar 7, 2022, at 5:30 PM, Andrii Nakryiko andrii.nakryiko@gmail.com wrote:
On Fri, Mar 4, 2022 at 9:31 AM Benjamin Tissoires benjamin.tissoires@redhat.com wrote:
HID-bpf program type are needing a new SEC. To bind a hid-bpf program, we can rely on bpf_program__attach_fd() so export a new function to the API.
Signed-off-by: Benjamin Tissoires benjamin.tissoires@redhat.com
changes in v2:
- split the series by bpf/libbpf/hid/selftests and samples
tools/lib/bpf/libbpf.c | 7 +++++++ tools/lib/bpf/libbpf.h | 2 ++ tools/lib/bpf/libbpf.map | 1 + 3 files changed, 10 insertions(+)
diff --git a/tools/lib/bpf/libbpf.c b/tools/lib/bpf/libbpf.c index 81bf01d67671..356bbd3ad2c7 100644 --- a/tools/lib/bpf/libbpf.c +++ b/tools/lib/bpf/libbpf.c @@ -8680,6 +8680,7 @@ static const struct bpf_sec_def section_defs[] = { SEC_DEF("cgroup/setsockopt", CGROUP_SOCKOPT, BPF_CGROUP_SETSOCKOPT, SEC_ATTACHABLE | SEC_SLOPPY_PFX), SEC_DEF("struct_ops+", STRUCT_OPS, 0, SEC_NONE), SEC_DEF("sk_lookup", SK_LOOKUP, BPF_SK_LOOKUP, SEC_ATTACHABLE | SEC_SLOPPY_PFX),
SEC_DEF("hid/device_event", HID, BPF_HID_DEVICE_EVENT, SEC_ATTACHABLE_OPT | SEC_SLOPPY_PFX),no SEC_SLOPPY_PFX for any new program type, please
};
#define MAX_TYPE_NAME_SIZE 32 @@ -10659,6 +10660,12 @@ static struct bpf_link *attach_iter(const struct bpf_program *prog, long cookie) return bpf_program__attach_iter(prog, NULL); }
+struct bpf_link * +bpf_program__attach_hid(const struct bpf_program *prog, int hid_fd) +{
return bpf_program__attach_fd(prog, hid_fd, 0, "hid");+}
struct bpf_link *bpf_program__attach(const struct bpf_program *prog) { if (!prog->sec_def || !prog->sec_def->attach_fn) diff --git a/tools/lib/bpf/libbpf.h b/tools/lib/bpf/libbpf.h index c8d8daad212e..f677ac0a9ede 100644 --- a/tools/lib/bpf/libbpf.h +++ b/tools/lib/bpf/libbpf.h @@ -529,6 +529,8 @@ struct bpf_iter_attach_opts { LIBBPF_API struct bpf_link * bpf_program__attach_iter(const struct bpf_program *prog, const struct bpf_iter_attach_opts *opts); +LIBBPF_API struct bpf_link * +bpf_program__attach_hid(const struct bpf_program *prog, int hid_fd);
/*
- Libbpf allows callers to adjust BPF programs before being loaded
diff --git a/tools/lib/bpf/libbpf.map b/tools/lib/bpf/libbpf.map index 47e70c9058d9..fdc6fa743953 100644 --- a/tools/lib/bpf/libbpf.map +++ b/tools/lib/bpf/libbpf.map @@ -424,6 +424,7 @@ LIBBPF_0.6.0 { LIBBPF_0.7.0 { global: bpf_btf_load;
bpf_program__attach_hid;should go into 0.8.0
Ah, I missed this one.
btw, bpf_xdp_attach and buddies should also go into 0.8.0, no?
not really, they were released in libbpf v0.7, it's just any new incoming API that should go into 0.8.0
bpf_program__expected_attach_type; bpf_program__log_buf; bpf_program__log_level;-- 2.35.1
The test is pretty basic: - create a virtual uhid device that no userspace will like (to not mess up the running system) - attach a BPF prog to it - open the matching hidraw node - inject one event and check: * that the BPF program can do something on the event stream * can modify the event stream
Signed-off-by: Benjamin Tissoires benjamin.tissoires@redhat.com
---
changes in v2: - split the series by bpf/libbpf/hid/selftests and samples --- tools/testing/selftests/bpf/prog_tests/hid.c | 335 +++++++++++++++++++ tools/testing/selftests/bpf/progs/hid.c | 20 ++ 2 files changed, 355 insertions(+) create mode 100644 tools/testing/selftests/bpf/prog_tests/hid.c create mode 100644 tools/testing/selftests/bpf/progs/hid.c
diff --git a/tools/testing/selftests/bpf/prog_tests/hid.c b/tools/testing/selftests/bpf/prog_tests/hid.c new file mode 100644 index 000000000000..ee495d5a8bd5 --- /dev/null +++ b/tools/testing/selftests/bpf/prog_tests/hid.c @@ -0,0 +1,335 @@ +// SPDX-License-Identifier: GPL-2.0 +/* Copyright (c) 2022 Red Hat */ +#include <test_progs.h> +#include <testing_helpers.h> +#include "hid.skel.h" + +#include <fcntl.h> +#include <fnmatch.h> +#include <dirent.h> +#include <poll.h> +#include <stdbool.h> +#include <linux/uhid.h> + +static unsigned char rdesc[] = { + 0x06, 0x00, 0xff, /* Usage Page (Vendor Defined Page 1) */ + 0x09, 0x21, /* Usage (Vendor Usage 0x21) */ + 0xa1, 0x01, /* COLLECTION (Application) */ + 0x09, 0x01, /* Usage (Vendor Usage 0x01) */ + 0xa1, 0x00, /* COLLECTION (Physical) */ + 0x85, 0x01, /* REPORT_ID (1) */ + 0x06, 0x00, 0xff, /* Usage Page (Vendor Defined Page 1) */ + 0x19, 0x01, /* USAGE_MINIMUM (1) */ + 0x29, 0x03, /* USAGE_MAXIMUM (3) */ + 0x15, 0x00, /* LOGICAL_MINIMUM (0) */ + 0x25, 0x01, /* LOGICAL_MAXIMUM (1) */ + 0x95, 0x03, /* REPORT_COUNT (3) */ + 0x75, 0x01, /* REPORT_SIZE (1) */ + 0x81, 0x02, /* INPUT (Data,Var,Abs) */ + 0x95, 0x01, /* REPORT_COUNT (1) */ + 0x75, 0x05, /* REPORT_SIZE (5) */ + 0x81, 0x01, /* INPUT (Cnst,Var,Abs) */ + 0x05, 0x01, /* USAGE_PAGE (Generic Desktop) */ + 0x09, 0x30, /* USAGE (X) */ + 0x09, 0x31, /* USAGE (Y) */ + 0x15, 0x81, /* LOGICAL_MINIMUM (-127) */ + 0x25, 0x7f, /* LOGICAL_MAXIMUM (127) */ + 0x75, 0x10, /* REPORT_SIZE (16) */ + 0x95, 0x02, /* REPORT_COUNT (2) */ + 0x81, 0x06, /* INPUT (Data,Var,Rel) */ + + 0x06, 0x00, 0xff, /* Usage Page (Vendor Defined Page 1) */ + 0x19, 0x01, /* USAGE_MINIMUM (1) */ + 0x29, 0x03, /* USAGE_MAXIMUM (3) */ + 0x15, 0x00, /* LOGICAL_MINIMUM (0) */ + 0x25, 0x01, /* LOGICAL_MAXIMUM (1) */ + 0x95, 0x03, /* REPORT_COUNT (3) */ + 0x75, 0x01, /* REPORT_SIZE (1) */ + 0x91, 0x02, /* Output (Data,Var,Abs) */ + 0x95, 0x01, /* REPORT_COUNT (1) */ + 0x75, 0x05, /* REPORT_SIZE (5) */ + 0x91, 0x01, /* Output (Cnst,Var,Abs) */ + + 0x06, 0x00, 0xff, /* Usage Page (Vendor Defined Page 1) */ + 0x19, 0x06, /* USAGE_MINIMUM (6) */ + 0x29, 0x08, /* USAGE_MAXIMUM (8) */ + 0x15, 0x00, /* LOGICAL_MINIMUM (0) */ + 0x25, 0x01, /* LOGICAL_MAXIMUM (1) */ + 0x95, 0x03, /* REPORT_COUNT (3) */ + 0x75, 0x01, /* REPORT_SIZE (1) */ + 0xb1, 0x02, /* Feature (Data,Var,Abs) */ + 0x95, 0x01, /* REPORT_COUNT (1) */ + 0x75, 0x05, /* REPORT_SIZE (5) */ + 0x91, 0x01, /* Output (Cnst,Var,Abs) */ + + 0xc0, /* END_COLLECTION */ + 0xc0, /* END_COLLECTION */ +}; + +static int uhid_write(int fd, const struct uhid_event *ev) +{ + ssize_t ret; + + ret = write(fd, ev, sizeof(*ev)); + if (ret < 0) { + fprintf(stderr, "Cannot write to uhid: %m\n"); + return -errno; + } else if (ret != sizeof(*ev)) { + fprintf(stderr, "Wrong size written to uhid: %zd != %zu\n", + ret, sizeof(ev)); + return -EFAULT; + } else { + return 0; + } +} + +static int create(int fd, int rand_nb) +{ + struct uhid_event ev; + char buf[25]; + + sprintf(buf, "test-uhid-device-%d", rand_nb); + + memset(&ev, 0, sizeof(ev)); + ev.type = UHID_CREATE; + strcpy((char *)ev.u.create.name, buf); + ev.u.create.rd_data = rdesc; + ev.u.create.rd_size = sizeof(rdesc); + ev.u.create.bus = BUS_USB; + ev.u.create.vendor = 0x0001; + ev.u.create.product = 0x0a37; + ev.u.create.version = 0; + ev.u.create.country = 0; + + sprintf(buf, "%d", rand_nb); + strcpy((char *)ev.u.create.phys, buf); + + return uhid_write(fd, &ev); +} + +static void destroy(int fd) +{ + struct uhid_event ev; + + memset(&ev, 0, sizeof(ev)); + ev.type = UHID_DESTROY; + + uhid_write(fd, &ev); +} + +static int send_event(int fd, u8 *buf, size_t size) +{ + struct uhid_event ev; + + if (size > sizeof(ev.u.input.data)) + return -E2BIG; + + memset(&ev, 0, sizeof(ev)); + ev.type = UHID_INPUT2; + ev.u.input2.size = size; + + memcpy(ev.u.input2.data, buf, size); + + return uhid_write(fd, &ev); +} + +static int setup_uhid(int rand_nb) +{ + int fd; + const char *path = "/dev/uhid"; + int ret; + + fd = open(path, O_RDWR | O_CLOEXEC); + if (!ASSERT_GE(fd, 0, "open uhid-cdev")) + return -EPERM; + + ret = create(fd, rand_nb); + if (!ASSERT_OK(ret, "create uhid device")) { + close(fd); + return -EPERM; + } + + return fd; +} + +static int get_sysfs_fd(int rand_nb) +{ + const char *workdir = "/sys/devices/virtual/misc/uhid"; + const char *target = "0003:0001:0A37.*"; + char uevent[1024]; + char temp[512]; + char phys[512]; + DIR *d; + struct dirent *dir; + int fd, nread; + int found = -1; + + /* it would be nice to be able to use nftw, but the no_alu32 target doesn't support it */ + + sprintf(phys, "PHYS=%d", rand_nb); + + d = opendir(workdir); + if (d) { + while ((dir = readdir(d)) != NULL) { + if (fnmatch(target, dir->d_name, 0)) + continue; + + /* we found the correct VID/PID, now check for phys */ + sprintf(uevent, "%s/%s/uevent", workdir, dir->d_name); + fd = open(uevent, O_RDONLY | O_NONBLOCK); + if (fd < 0) + continue; + + nread = read(fd, temp, ARRAY_SIZE(temp)); + if (nread > 0 && (strstr(temp, phys)) != NULL) { + found = fd; + break; + } + + close(fd); + fd = 0; + } + closedir(d); + } + + return found; +} + +static int get_hidraw(struct bpf_link *link) +{ + struct bpf_link_info info = {0}; + int prog_id, i; + + /* retry 5 times in case the system is loaded */ + for (i = 5; i > 0; i--) { + usleep(10); + prog_id = link_info_prog_id(link, &info); + if (!prog_id) + continue; + if (info.hid.hidraw_ino >= 0) + break; + } + + if (!prog_id) + return -1; + + return info.hid.hidraw_ino; +} + +/* + * Attach hid_first_event to the given uhid device, + * retrieve and open the matching hidraw node, + * inject one event in the uhid device, + * check that the program sees it and can change the data + */ +static int test_hid_raw_event(struct hid *hid_skel, int uhid_fd, int sysfs_fd) +{ + int err, hidraw_ino, hidraw_fd = -1; + char hidraw_path[64] = {0}; + u8 buf[10] = {0}; + int ret = -1; + + /* check that the program is correctly loaded */ + ASSERT_EQ(hid_skel->data->callback_check, 52, "callback_check1"); + ASSERT_EQ(hid_skel->data->callback2_check, 52, "callback2_check1"); + + /* attach the first program */ + hid_skel->links.hid_first_event = + bpf_program__attach_hid(hid_skel->progs.hid_first_event, sysfs_fd); + if (!ASSERT_OK_PTR(hid_skel->links.hid_first_event, + "attach_hid(hid_first_event)")) + return PTR_ERR(hid_skel->links.hid_first_event); + + hidraw_ino = get_hidraw(hid_skel->links.hid_first_event); + if (!ASSERT_GE(hidraw_ino, 0, "get_hidraw")) + goto cleanup; + + /* open hidraw node to check the other side of the pipe */ + sprintf(hidraw_path, "/dev/hidraw%d", hidraw_ino); + hidraw_fd = open(hidraw_path, O_RDWR | O_NONBLOCK); + + if (!ASSERT_GE(hidraw_fd, 0, "open_hidraw")) + goto cleanup; + + /* inject one event */ + buf[0] = 1; + buf[1] = 42; + send_event(uhid_fd, buf, 6); + + /* check that hid_first_event() was executed */ + ASSERT_EQ(hid_skel->data->callback_check, 42, "callback_check1"); + + /* read the data from hidraw */ + memset(buf, 0, sizeof(buf)); + err = read(hidraw_fd, buf, sizeof(buf)); + if (!ASSERT_EQ(err, 6, "read_hidraw")) + goto cleanup; + + if (!ASSERT_EQ(buf[2], 47, "hid_first_event")) + goto cleanup; + + /* inject another event */ + buf[0] = 1; + buf[1] = 47; + send_event(uhid_fd, buf, 6); + + /* check that hid_first_event() was executed */ + ASSERT_EQ(hid_skel->data->callback_check, 47, "callback_check1"); + + /* read the data from hidraw */ + memset(buf, 0, sizeof(buf)); + err = read(hidraw_fd, buf, sizeof(buf)); + if (!ASSERT_EQ(err, 6, "read_hidraw")) + goto cleanup; + + if (!ASSERT_EQ(buf[2], 52, "hid_first_event")) + goto cleanup; + + ret = 0; + +cleanup: + if (hidraw_fd >= 0) + close(hidraw_fd); + + hid__detach(hid_skel); + + return ret; +} + +void serial_test_hid_bpf(void) +{ + struct hid *hid_skel = NULL; + int err, uhid_fd, sysfs_fd; + time_t t; + int rand_nb; + + /* initialize random number generator */ + srand((unsigned int)time(&t)); + + rand_nb = rand() % 1024; + + uhid_fd = setup_uhid(rand_nb); + if (!ASSERT_GE(uhid_fd, 0, "setup uhid")) + return; + + /* give a little bit of time for the device to appear */ + /* TODO: check on uhid events */ + usleep(1000); + + /* locate the uevent file of the created device */ + sysfs_fd = get_sysfs_fd(rand_nb); + if (!ASSERT_GE(sysfs_fd, 0, "locate sysfs uhid device")) + goto cleanup; + + hid_skel = hid__open_and_load(); + if (!ASSERT_OK_PTR(hid_skel, "hid_skel_load")) + goto cleanup; + + /* start the tests! */ + err = test_hid_raw_event(hid_skel, uhid_fd, sysfs_fd); + ASSERT_OK(err, "hid"); + +cleanup: + hid__destroy(hid_skel); + destroy(uhid_fd); +} diff --git a/tools/testing/selftests/bpf/progs/hid.c b/tools/testing/selftests/bpf/progs/hid.c new file mode 100644 index 000000000000..2201dd3b105d --- /dev/null +++ b/tools/testing/selftests/bpf/progs/hid.c @@ -0,0 +1,20 @@ +// SPDX-License-Identifier: GPL-2.0 +/* Copyright (c) 2022 Red hat */ +#include <linux/bpf.h> +#include <bpf/bpf_helpers.h> +#include <linux/bpf_hid.h> + +char _license[] SEC("license") = "GPL"; + +__u64 callback_check = 52; +__u64 callback2_check = 52; + +SEC("hid/device_event") +int hid_first_event(struct hid_bpf_ctx *ctx) +{ + callback_check = ctx->data[1]; + + ctx->data[2] = ctx->data[1] + 5; + + return 0; +}
On Fri, Mar 4, 2022 at 9:31 AM Benjamin Tissoires benjamin.tissoires@redhat.com wrote:
The test is pretty basic:
- create a virtual uhid device that no userspace will like (to not mess up the running system)
- attach a BPF prog to it
- open the matching hidraw node
- inject one event and check:
- that the BPF program can do something on the event stream
- can modify the event stream
Signed-off-by: Benjamin Tissoires benjamin.tissoires@redhat.com
Does this test run with vm (qemu, etc.)? Maybe we need to update tools/testing/selftests/bpf/config ?
Thanks, Song
On Sat, Mar 5, 2022 at 1:41 AM Song Liu song@kernel.org wrote:
On Fri, Mar 4, 2022 at 9:31 AM Benjamin Tissoires benjamin.tissoires@redhat.com wrote:
The test is pretty basic:
- create a virtual uhid device that no userspace will like (to not mess up the running system)
- attach a BPF prog to it
- open the matching hidraw node
- inject one event and check:
- that the BPF program can do something on the event stream
- can modify the event stream
Signed-off-by: Benjamin Tissoires benjamin.tissoires@redhat.com
Does this test run with vm (qemu, etc.)? Maybe we need to update tools/testing/selftests/bpf/config ?
Good point. I'll give it a shot. I was testing on my devel machine so it was quicker to iterate, but I completely forgot to make a run at qemu with the scripts in place.
Cheers, Benjamin
Thanks, Song
Everything should be available in the selftest part of the tree, but providing an example without uhid and hidraw will be more easy to follow for users.
Signed-off-by: Benjamin Tissoires benjamin.tissoires@redhat.com
---
changes in v2: - split the series by bpf/libbpf/hid/selftests and samples --- samples/bpf/.gitignore | 1 + samples/bpf/Makefile | 4 ++ samples/bpf/hid_mouse_kern.c | 66 ++++++++++++++++++ samples/bpf/hid_mouse_user.c | 129 +++++++++++++++++++++++++++++++++++ 4 files changed, 200 insertions(+) create mode 100644 samples/bpf/hid_mouse_kern.c create mode 100644 samples/bpf/hid_mouse_user.c
diff --git a/samples/bpf/.gitignore b/samples/bpf/.gitignore index 0e7bfdbff80a..65440bd618b2 100644 --- a/samples/bpf/.gitignore +++ b/samples/bpf/.gitignore @@ -2,6 +2,7 @@ cpustat fds_example hbm +hid_mouse ibumad lathist lwt_len_hist diff --git a/samples/bpf/Makefile b/samples/bpf/Makefile index 38638845db9d..84ef458487df 100644 --- a/samples/bpf/Makefile +++ b/samples/bpf/Makefile @@ -60,6 +60,8 @@ tprogs-y += xdp_redirect_map tprogs-y += xdp_redirect tprogs-y += xdp_monitor
+tprogs-y += hid_mouse + # Libbpf dependencies LIBBPF_SRC = $(TOOLS_PATH)/lib/bpf LIBBPF_OUTPUT = $(abspath $(BPF_SAMPLES_PATH))/libbpf @@ -124,6 +126,7 @@ xdp_redirect_cpu-objs := xdp_redirect_cpu_user.o $(XDP_SAMPLE) xdp_redirect_map-objs := xdp_redirect_map_user.o $(XDP_SAMPLE) xdp_redirect-objs := xdp_redirect_user.o $(XDP_SAMPLE) xdp_monitor-objs := xdp_monitor_user.o $(XDP_SAMPLE) +hid_mouse-objs := hid_mouse_user.o
# Tell kbuild to always build the programs always-y := $(tprogs-y) @@ -181,6 +184,7 @@ always-y += ibumad_kern.o always-y += hbm_out_kern.o always-y += hbm_edt_kern.o always-y += xdpsock_kern.o +always-y += hid_mouse_kern.o
ifeq ($(ARCH), arm) # Strip all except -D__LINUX_ARM_ARCH__ option needed to handle linux diff --git a/samples/bpf/hid_mouse_kern.c b/samples/bpf/hid_mouse_kern.c new file mode 100644 index 000000000000..c24a12e06b40 --- /dev/null +++ b/samples/bpf/hid_mouse_kern.c @@ -0,0 +1,66 @@ +// SPDX-License-Identifier: GPL-2.0-only +/* Copyright (c) 2021 Benjamin Tissoires + */ +#include <linux/version.h> +#include <uapi/linux/bpf.h> +#include <uapi/linux/bpf_hid.h> +#include <bpf/bpf_helpers.h> + +SEC("hid/device_event") +int hid_y_event(struct hid_bpf_ctx *ctx) +{ + s16 y; + + bpf_printk("event: %02x size: %d", ctx->type, ctx->size); + bpf_printk("incoming event: %02x %02x %02x", + ctx->data[0], + ctx->data[1], + ctx->data[2]); + bpf_printk(" %02x %02x %02x", + ctx->data[3], + ctx->data[4], + ctx->data[5]); + bpf_printk(" %02x %02x %02x", + ctx->data[6], + ctx->data[7], + ctx->data[8]); + + y = ctx->data[3] | (ctx->data[4] << 8); + + y = -y; + + ctx->data[3] = y & 0xFF; + ctx->data[4] = (y >> 8) & 0xFF; + + bpf_printk("modified event: %02x %02x %02x", + ctx->data[0], + ctx->data[1], + ctx->data[2]); + bpf_printk(" %02x %02x %02x", + ctx->data[3], + ctx->data[4], + ctx->data[5]); + bpf_printk(" %02x %02x %02x", + ctx->data[6], + ctx->data[7], + ctx->data[8]); + + return 0; +} + +SEC("hid/device_event") +int hid_x_event(struct hid_bpf_ctx *ctx) +{ + s16 x; + + x = ctx->data[1] | (ctx->data[2] << 8); + + x = -x; + + ctx->data[1] = x & 0xFF; + ctx->data[2] = (x >> 8) & 0xFF; + return 0; +} + +char _license[] SEC("license") = "GPL"; +u32 _version SEC("version") = LINUX_VERSION_CODE; diff --git a/samples/bpf/hid_mouse_user.c b/samples/bpf/hid_mouse_user.c new file mode 100644 index 000000000000..d4f37caca2fa --- /dev/null +++ b/samples/bpf/hid_mouse_user.c @@ -0,0 +1,129 @@ +// SPDX-License-Identifier: GPL-2.0-only +/* Copyright (c) 2021 Benjamin Tissoires + */ +#include <linux/bpf.h> +#include <linux/if_link.h> +#include <assert.h> +#include <errno.h> +#include <fcntl.h> +#include <signal.h> +#include <stdio.h> +#include <stdlib.h> +#include <string.h> +#include <unistd.h> +#include <libgen.h> +#include <sys/resource.h> + +#include "bpf_util.h" +#include <bpf/bpf.h> +#include <bpf/libbpf.h> + +static char *sysfs_path; +static int sysfs_fd; +static int prog_count; + +struct prog { + int fd; + struct bpf_link *link; + enum bpf_attach_type type; +}; + +static struct prog progs[10]; + +static void int_exit(int sig) +{ + for (prog_count--; prog_count >= 0; prog_count--) + bpf_link__destroy(progs[prog_count].link); + + close(sysfs_fd); + exit(0); +} + +static void usage(const char *prog) +{ + fprintf(stderr, + "%s: %s /sys/bus/hid/devices/0BUS:0VID:0PID:00ID/uevent\n\n", + __func__, prog); +} + +int main(int argc, char **argv) +{ + struct bpf_prog_info info = {}; + __u32 info_len = sizeof(info); + const char *optstr = ""; + struct bpf_object *obj; + struct bpf_program *prog; + int opt; + char filename[256]; + int err; + + while ((opt = getopt(argc, argv, optstr)) != -1) { + switch (opt) { + default: + usage(basename(argv[0])); + return 1; + } + } + + if (optind == argc) { + usage(basename(argv[0])); + return 1; + } + + sysfs_path = argv[optind]; + if (!sysfs_path) { + perror("sysfs"); + return 1; + } + + snprintf(filename, sizeof(filename), "%s_kern.o", argv[0]); + obj = bpf_object__open_file(filename, NULL); + err = libbpf_get_error(obj); + if (err) { + fprintf(stderr, "ERROR: opening BPF object file failed\n"); + obj = NULL; + err = 1; + goto cleanup; + } + + /* load BPF program */ + err = bpf_object__load(obj); + if (err) { + fprintf(stderr, "ERROR: loading BPF object file failed\n"); + goto cleanup; + } + + sysfs_fd = open(sysfs_path, O_RDONLY); + + bpf_object__for_each_program(prog, obj) { + progs[prog_count].fd = bpf_program__fd(prog); + progs[prog_count].type = bpf_program__get_expected_attach_type(prog); + progs[prog_count].link = bpf_program__attach_hid(prog, sysfs_fd); + if (libbpf_get_error(progs[prog_count].link)) { + fprintf(stderr, "bpf_prog_attach: err=%m\n"); + progs[prog_count].fd = 0; + progs[prog_count].link = NULL; + goto cleanup; + } + prog_count++; + } + + signal(SIGINT, int_exit); + signal(SIGTERM, int_exit); + + err = bpf_obj_get_info_by_fd(progs[0].fd, &info, &info_len); + if (err) { + printf("can't get prog info - %s\n", strerror(errno)); + goto cleanup; + } + + while (1) + ; + + cleanup: + for (prog_count--; prog_count >= 0; prog_count--) + bpf_link__destroy(progs[prog_count].link); + + bpf_object__close(obj); + return err; +}
On Fri, Mar 04, 2022 at 06:28:30PM +0100, Benjamin Tissoires wrote:
Everything should be available in the selftest part of the tree, but providing an example without uhid and hidraw will be more easy to follow for users.
Signed-off-by: Benjamin Tissoires benjamin.tissoires@redhat.com
changes in v2:
- split the series by bpf/libbpf/hid/selftests and samples
samples/bpf/.gitignore | 1 + samples/bpf/Makefile | 4 ++ samples/bpf/hid_mouse_kern.c | 66 ++++++++++++++++++ samples/bpf/hid_mouse_user.c | 129 +++++++++++++++++++++++++++++++++++ 4 files changed, 200 insertions(+) create mode 100644 samples/bpf/hid_mouse_kern.c create mode 100644 samples/bpf/hid_mouse_user.c
diff --git a/samples/bpf/.gitignore b/samples/bpf/.gitignore index 0e7bfdbff80a..65440bd618b2 100644 --- a/samples/bpf/.gitignore +++ b/samples/bpf/.gitignore @@ -2,6 +2,7 @@ cpustat fds_example hbm +hid_mouse ibumad lathist lwt_len_hist diff --git a/samples/bpf/Makefile b/samples/bpf/Makefile index 38638845db9d..84ef458487df 100644 --- a/samples/bpf/Makefile +++ b/samples/bpf/Makefile @@ -60,6 +60,8 @@ tprogs-y += xdp_redirect_map tprogs-y += xdp_redirect tprogs-y += xdp_monitor +tprogs-y += hid_mouse
# Libbpf dependencies LIBBPF_SRC = $(TOOLS_PATH)/lib/bpf LIBBPF_OUTPUT = $(abspath $(BPF_SAMPLES_PATH))/libbpf @@ -124,6 +126,7 @@ xdp_redirect_cpu-objs := xdp_redirect_cpu_user.o $(XDP_SAMPLE) xdp_redirect_map-objs := xdp_redirect_map_user.o $(XDP_SAMPLE) xdp_redirect-objs := xdp_redirect_user.o $(XDP_SAMPLE) xdp_monitor-objs := xdp_monitor_user.o $(XDP_SAMPLE) +hid_mouse-objs := hid_mouse_user.o # Tell kbuild to always build the programs always-y := $(tprogs-y) @@ -181,6 +184,7 @@ always-y += ibumad_kern.o always-y += hbm_out_kern.o always-y += hbm_edt_kern.o always-y += xdpsock_kern.o +always-y += hid_mouse_kern.o ifeq ($(ARCH), arm) # Strip all except -D__LINUX_ARM_ARCH__ option needed to handle linux diff --git a/samples/bpf/hid_mouse_kern.c b/samples/bpf/hid_mouse_kern.c new file mode 100644 index 000000000000..c24a12e06b40 --- /dev/null +++ b/samples/bpf/hid_mouse_kern.c @@ -0,0 +1,66 @@ +// SPDX-License-Identifier: GPL-2.0-only +/* Copyright (c) 2021 Benjamin Tissoires
It's 2022 now :(
Other than that, looks nice and simple, good work!
Reviewed-by: Greg Kroah-Hartman gregkh@linuxfoundation.org
The report descriptor is the dictionary of the HID protocol specific to the given device. Changing it is a common habit in the HID world, and making that feature accessible from eBPF allows to fix devices without having to install a new kernel.
However, the report descriptor is supposed to be static on a device. To be able to change it, we need to reconnect the device at the HID level. So whenever the report descriptor program type is attached or detached, we call on a hook on HID to notify it that there is something to be done.
Signed-off-by: Benjamin Tissoires benjamin.tissoires@redhat.com
---
changes in v2: - split the series by bpf/libbpf/hid/selftests and samples - unsigned long -> __u16 in uapi/linux/bpf_hid.h --- include/linux/bpf-hid.h | 4 ++++ include/uapi/linux/bpf.h | 1 + include/uapi/linux/bpf_hid.h | 1 + kernel/bpf/hid.c | 5 +++++ kernel/bpf/syscall.c | 2 ++ tools/include/uapi/linux/bpf.h | 1 + 6 files changed, 14 insertions(+)
diff --git a/include/linux/bpf-hid.h b/include/linux/bpf-hid.h index 3cda78051b5f..0c5000b28b20 100644 --- a/include/linux/bpf-hid.h +++ b/include/linux/bpf-hid.h @@ -15,6 +15,7 @@ struct hid_device; enum bpf_hid_attach_type { BPF_HID_ATTACH_INVALID = -1, BPF_HID_ATTACH_DEVICE_EVENT = 0, + BPF_HID_ATTACH_RDESC_FIXUP, MAX_BPF_HID_ATTACH_TYPE };
@@ -32,6 +33,8 @@ to_bpf_hid_attach_type(enum bpf_attach_type attach_type) switch (attach_type) { case BPF_HID_DEVICE_EVENT: return BPF_HID_ATTACH_DEVICE_EVENT; + case BPF_HID_RDESC_FIXUP: + return BPF_HID_ATTACH_RDESC_FIXUP; default: return BPF_HID_ATTACH_INVALID; } @@ -88,6 +91,7 @@ static inline bool bpf_hid_link_empty(struct bpf_hid *bpf, struct bpf_hid_hooks { struct hid_device *(*hdev_from_fd)(int fd); int (*link_attach)(struct hid_device *hdev, enum bpf_hid_attach_type type); + void (*link_attached)(struct hid_device *hdev, enum bpf_hid_attach_type type); void (*array_detached)(struct hid_device *hdev, enum bpf_hid_attach_type type); };
diff --git a/include/uapi/linux/bpf.h b/include/uapi/linux/bpf.h index 5978b92cacd3..a7a8d9cfcf24 100644 --- a/include/uapi/linux/bpf.h +++ b/include/uapi/linux/bpf.h @@ -999,6 +999,7 @@ enum bpf_attach_type { BPF_SK_REUSEPORT_SELECT_OR_MIGRATE, BPF_PERF_EVENT, BPF_HID_DEVICE_EVENT, + BPF_HID_RDESC_FIXUP, __MAX_BPF_ATTACH_TYPE };
diff --git a/include/uapi/linux/bpf_hid.h b/include/uapi/linux/bpf_hid.h index 975ca5bd526f..634f17c0b1cb 100644 --- a/include/uapi/linux/bpf_hid.h +++ b/include/uapi/linux/bpf_hid.h @@ -24,6 +24,7 @@ struct hid_device; enum hid_bpf_event { HID_BPF_UNDEF = 0, HID_BPF_DEVICE_EVENT, /* when attach type is BPF_HID_DEVICE_EVENT */ + HID_BPF_RDESC_FIXUP, /* ................... BPF_HID_RDESC_FIXUP */ };
struct hid_bpf_ctx { diff --git a/kernel/bpf/hid.c b/kernel/bpf/hid.c index db7f75a0a812..37500313e270 100644 --- a/kernel/bpf/hid.c +++ b/kernel/bpf/hid.c @@ -315,6 +315,8 @@ static int bpf_hid_max_progs(enum bpf_hid_attach_type type) switch (type) { case BPF_HID_ATTACH_DEVICE_EVENT: return 64; + case BPF_HID_ATTACH_RDESC_FIXUP: + return 1; default: return 0; } @@ -355,6 +357,9 @@ static int bpf_hid_link_attach(struct hid_device *hdev, struct bpf_link *link, lockdep_is_held(&bpf_hid_mutex)); bpf_prog_array_free(run_array);
+ if (hid_hooks.link_attached) + hid_hooks.link_attached(hdev, type); + out_unlock: mutex_unlock(&bpf_hid_mutex); return err; diff --git a/kernel/bpf/syscall.c b/kernel/bpf/syscall.c index a94e78ec3211..7428a1a512c6 100644 --- a/kernel/bpf/syscall.c +++ b/kernel/bpf/syscall.c @@ -3203,6 +3203,7 @@ attach_type_to_prog_type(enum bpf_attach_type attach_type) case BPF_XDP: return BPF_PROG_TYPE_XDP; case BPF_HID_DEVICE_EVENT: + case BPF_HID_RDESC_FIXUP: return BPF_PROG_TYPE_HID; default: return BPF_PROG_TYPE_UNSPEC; @@ -3348,6 +3349,7 @@ static int bpf_prog_query(const union bpf_attr *attr, case BPF_SK_SKB_VERDICT: return sock_map_bpf_prog_query(attr, uattr); case BPF_HID_DEVICE_EVENT: + case BPF_HID_RDESC_FIXUP: return bpf_hid_prog_query(attr, uattr); default: return -EINVAL; diff --git a/tools/include/uapi/linux/bpf.h b/tools/include/uapi/linux/bpf.h index 5978b92cacd3..a7a8d9cfcf24 100644 --- a/tools/include/uapi/linux/bpf.h +++ b/tools/include/uapi/linux/bpf.h @@ -999,6 +999,7 @@ enum bpf_attach_type { BPF_SK_REUSEPORT_SELECT_OR_MIGRATE, BPF_PERF_EVENT, BPF_HID_DEVICE_EVENT, + BPF_HID_RDESC_FIXUP, __MAX_BPF_ATTACH_TYPE };
On Fri, Mar 04, 2022 at 06:28:31PM +0100, Benjamin Tissoires wrote:
The report descriptor is the dictionary of the HID protocol specific to the given device. Changing it is a common habit in the HID world, and making that feature accessible from eBPF allows to fix devices without having to install a new kernel.
However, the report descriptor is supposed to be static on a device. To be able to change it, we need to reconnect the device at the HID level. So whenever the report descriptor program type is attached or detached, we call on a hook on HID to notify it that there is something to be done.
Signed-off-by: Benjamin Tissoires benjamin.tissoires@redhat.com
Reviewed-by: Greg Kroah-Hartman gregkh@linuxfoundation.org
Make use of BPF_HID_ATTACH_RDESC_FIXUP so we can trigger an rdesc fixup in the bpf world.
Whenever the program gets attached/detached, the device is reconnected meaning that userspace will see it disappearing and reappearing with the new report descriptor.
Signed-off-by: Benjamin Tissoires benjamin.tissoires@redhat.com
---
changes in v2: - split the series by bpf/libbpf/hid/selftests and samples --- drivers/hid/hid-bpf.c | 60 ++++++++++++++++++++++++++++++++++++++++++ drivers/hid/hid-core.c | 3 ++- include/linux/hid.h | 6 +++++ 3 files changed, 68 insertions(+), 1 deletion(-)
diff --git a/drivers/hid/hid-bpf.c b/drivers/hid/hid-bpf.c index 8120e598de9f..510e24f4307c 100644 --- a/drivers/hid/hid-bpf.c +++ b/drivers/hid/hid-bpf.c @@ -50,6 +50,14 @@ static struct hid_device *hid_bpf_fd_to_hdev(int fd) return hdev; }
+static int hid_reconnect(struct hid_device *hdev) +{ + if (!test_and_set_bit(ffs(HID_STAT_REPROBED), &hdev->status)) + return device_reprobe(&hdev->dev); + + return 0; +} + static int hid_bpf_link_attach(struct hid_device *hdev, enum bpf_hid_attach_type type) { int err = 0; @@ -71,6 +79,17 @@ static int hid_bpf_link_attach(struct hid_device *hdev, enum bpf_hid_attach_type return err; }
+static void hid_bpf_link_attached(struct hid_device *hdev, enum bpf_hid_attach_type type) +{ + switch (type) { + case BPF_HID_ATTACH_RDESC_FIXUP: + hid_reconnect(hdev); + break; + default: + /* do nothing */ + } +} + static void hid_bpf_array_detached(struct hid_device *hdev, enum bpf_hid_attach_type type) { switch (type) { @@ -78,6 +97,9 @@ static void hid_bpf_array_detached(struct hid_device *hdev, enum bpf_hid_attach_ kfree(hdev->bpf.ctx); hdev->bpf.ctx = NULL; break; + case BPF_HID_ATTACH_RDESC_FIXUP: + hid_reconnect(hdev); + break; default: /* do nothing */ } @@ -98,6 +120,9 @@ static int hid_bpf_run_progs(struct hid_device *hdev, enum bpf_hid_attach_type t case BPF_HID_ATTACH_DEVICE_EVENT: event = HID_BPF_DEVICE_EVENT; break; + case BPF_HID_ATTACH_RDESC_FIXUP: + event = HID_BPF_RDESC_FIXUP; + break; default: return -EINVAL; } @@ -138,11 +163,46 @@ u8 *hid_bpf_raw_event(struct hid_device *hdev, u8 *data, int *size) return hdev->bpf.ctx->data; }
+u8 *hid_bpf_report_fixup(struct hid_device *hdev, u8 *rdesc, unsigned int *size) +{ + struct hid_bpf_ctx *ctx = NULL; + int ret; + + if (bpf_hid_link_empty(&hdev->bpf, BPF_HID_ATTACH_RDESC_FIXUP)) + goto ignore_bpf; + + ctx = bpf_hid_allocate_ctx(hdev, HID_MAX_DESCRIPTOR_SIZE); + if (IS_ERR(ctx)) + goto ignore_bpf; + + ret = hid_bpf_run_progs(hdev, BPF_HID_ATTACH_RDESC_FIXUP, ctx, rdesc, *size); + if (ret) + goto ignore_bpf; + + *size = ctx->size; + + if (!*size) { + rdesc = NULL; + goto unlock; + } + + rdesc = kmemdup(ctx->data, *size, GFP_KERNEL); + + unlock: + kfree(ctx); + return rdesc; + + ignore_bpf: + kfree(ctx); + return kmemdup(rdesc, *size, GFP_KERNEL); +} + int __init hid_bpf_module_init(void) { struct bpf_hid_hooks hooks = { .hdev_from_fd = hid_bpf_fd_to_hdev, .link_attach = hid_bpf_link_attach, + .link_attached = hid_bpf_link_attached, .array_detached = hid_bpf_array_detached, };
diff --git a/drivers/hid/hid-core.c b/drivers/hid/hid-core.c index a80bffe6ce4a..0eb8189faaee 100644 --- a/drivers/hid/hid-core.c +++ b/drivers/hid/hid-core.c @@ -1213,7 +1213,8 @@ int hid_open_report(struct hid_device *device) return -ENODEV; size = device->dev_rsize;
- buf = kmemdup(start, size, GFP_KERNEL); + /* hid_bpf_report_fixup() ensures we work on a copy of rdesc */ + buf = hid_bpf_report_fixup(device, start, &size); if (buf == NULL) return -ENOMEM;
diff --git a/include/linux/hid.h b/include/linux/hid.h index 8fd79011f461..66d949d10b78 100644 --- a/include/linux/hid.h +++ b/include/linux/hid.h @@ -1213,10 +1213,16 @@ do { \
#ifdef CONFIG_BPF u8 *hid_bpf_raw_event(struct hid_device *hdev, u8 *rd, int *size); +u8 *hid_bpf_report_fixup(struct hid_device *hdev, u8 *rdesc, unsigned int *size); int hid_bpf_module_init(void); void hid_bpf_module_exit(void); #else static inline u8 *hid_bpf_raw_event(struct hid_device *hdev, u8 *rd, int *size) { return rd; } +static inline u8 *hid_bpf_report_fixup(struct hid_device *hdev, u8 *rdesc, + unsigned int *size) +{ + return kmemdup(rdesc, *size, GFP_KERNEL); +} static inline int hid_bpf_module_init(void) { return 0; } static inline void hid_bpf_module_exit(void) {} #endif
On Fri, Mar 04, 2022 at 06:28:32PM +0100, Benjamin Tissoires wrote:
Make use of BPF_HID_ATTACH_RDESC_FIXUP so we can trigger an rdesc fixup in the bpf world.
Whenever the program gets attached/detached, the device is reconnected meaning that userspace will see it disappearing and reappearing with the new report descriptor.
Signed-off-by: Benjamin Tissoires benjamin.tissoires@redhat.com
Reviewed-by: Greg Kroah-Hartman gregkh@linuxfoundation.org
Adds the SEC definiton for report descriptor fixups.
Signed-off-by: Benjamin Tissoires benjamin.tissoires@redhat.com
---
changes in v2: - split the series by bpf/libbpf/hid/selftests and samples --- tools/lib/bpf/libbpf.c | 1 + 1 file changed, 1 insertion(+)
diff --git a/tools/lib/bpf/libbpf.c b/tools/lib/bpf/libbpf.c index 356bbd3ad2c7..192ef3901251 100644 --- a/tools/lib/bpf/libbpf.c +++ b/tools/lib/bpf/libbpf.c @@ -8681,6 +8681,7 @@ static const struct bpf_sec_def section_defs[] = { SEC_DEF("struct_ops+", STRUCT_OPS, 0, SEC_NONE), SEC_DEF("sk_lookup", SK_LOOKUP, BPF_SK_LOOKUP, SEC_ATTACHABLE | SEC_SLOPPY_PFX), SEC_DEF("hid/device_event", HID, BPF_HID_DEVICE_EVENT, SEC_ATTACHABLE_OPT | SEC_SLOPPY_PFX), + SEC_DEF("hid/rdesc_fixup", HID, BPF_HID_RDESC_FIXUP, SEC_ATTACHABLE_OPT | SEC_SLOPPY_PFX), };
#define MAX_TYPE_NAME_SIZE 32
Simple report descriptor override in HID: replace part of the report descriptor from a static definition in the bpf kernel program.
Note that this test should be run last because we disconnect/reconnect the device, meaning that it changes the overall uhid device.
Signed-off-by: Benjamin Tissoires benjamin.tissoires@redhat.com
---
changes in v2: - split the series by bpf/libbpf/hid/selftests and samples --- tools/testing/selftests/bpf/prog_tests/hid.c | 69 ++++++++++++++++++++ tools/testing/selftests/bpf/progs/hid.c | 48 ++++++++++++++ 2 files changed, 117 insertions(+)
diff --git a/tools/testing/selftests/bpf/prog_tests/hid.c b/tools/testing/selftests/bpf/prog_tests/hid.c index ee495d5a8bd5..91543b8078ca 100644 --- a/tools/testing/selftests/bpf/prog_tests/hid.c +++ b/tools/testing/selftests/bpf/prog_tests/hid.c @@ -9,6 +9,7 @@ #include <dirent.h> #include <poll.h> #include <stdbool.h> +#include <linux/hidraw.h> #include <linux/uhid.h>
static unsigned char rdesc[] = { @@ -296,6 +297,71 @@ static int test_hid_raw_event(struct hid *hid_skel, int uhid_fd, int sysfs_fd) return ret; }
+/* + * Attach hid_rdesc_fixup to the given uhid device, + * retrieve and open the matching hidraw node, + * check that the hidraw report descriptor has been updated. + */ +static int test_rdesc_fixup(struct hid *hid_skel, int uhid_fd, int sysfs_fd) +{ + struct hidraw_report_descriptor rpt_desc = {0}; + int err, desc_size, hidraw_ino, hidraw_fd = -1; + char hidraw_path[64] = {0}; + int ret = -1; + + /* attach the program */ + hid_skel->links.hid_rdesc_fixup = + bpf_program__attach_hid(hid_skel->progs.hid_rdesc_fixup, sysfs_fd); + if (!ASSERT_OK_PTR(hid_skel->links.hid_rdesc_fixup, + "attach_hid(hid_rdesc_fixup)")) + return PTR_ERR(hid_skel->links.hid_rdesc_fixup); + + /* give a little bit of time for the device to appear */ + /* TODO: check on uhid events */ + usleep(1000); + + hidraw_ino = get_hidraw(hid_skel->links.hid_rdesc_fixup); + if (!ASSERT_GE(hidraw_ino, 0, "get_hidraw")) + goto cleanup; + + /* open hidraw node to check the other side of the pipe */ + sprintf(hidraw_path, "/dev/hidraw%d", hidraw_ino); + hidraw_fd = open(hidraw_path, O_RDWR | O_NONBLOCK); + + if (!ASSERT_GE(hidraw_fd, 0, "open_hidraw")) + goto cleanup; + + /* check that hid_rdesc_fixup() was executed */ + ASSERT_EQ(hid_skel->data->callback2_check, 0x21, "callback_check2"); + + /* read the exposed report descriptor from hidraw */ + err = ioctl(hidraw_fd, HIDIOCGRDESCSIZE, &desc_size); + if (!ASSERT_GE(err, 0, "HIDIOCGRDESCSIZE")) + goto cleanup; + + /* ensure the new size of the rdesc is bigger than the old one */ + if (!ASSERT_GT(desc_size, sizeof(rdesc), "new_rdesc_size")) + goto cleanup; + + rpt_desc.size = desc_size; + err = ioctl(hidraw_fd, HIDIOCGRDESC, &rpt_desc); + if (!ASSERT_GE(err, 0, "HIDIOCGRDESC")) + goto cleanup; + + if (!ASSERT_EQ(rpt_desc.value[4], 0x42, "hid_rdesc_fixup")) + goto cleanup; + + ret = 0; + +cleanup: + if (hidraw_fd >= 0) + close(hidraw_fd); + + hid__detach(hid_skel); + + return ret; +} + void serial_test_hid_bpf(void) { struct hid *hid_skel = NULL; @@ -329,6 +395,9 @@ void serial_test_hid_bpf(void) err = test_hid_raw_event(hid_skel, uhid_fd, sysfs_fd); ASSERT_OK(err, "hid");
+ err = test_rdesc_fixup(hid_skel, uhid_fd, sysfs_fd); + ASSERT_OK(err, "hid_rdesc_fixup"); + cleanup: hid__destroy(hid_skel); destroy(uhid_fd); diff --git a/tools/testing/selftests/bpf/progs/hid.c b/tools/testing/selftests/bpf/progs/hid.c index 2201dd3b105d..2270448d0d3f 100644 --- a/tools/testing/selftests/bpf/progs/hid.c +++ b/tools/testing/selftests/bpf/progs/hid.c @@ -18,3 +18,51 @@ int hid_first_event(struct hid_bpf_ctx *ctx)
return 0; } + +static __u8 rdesc[] = { + 0x05, 0x01, /* USAGE_PAGE (Generic Desktop) */ + 0x09, 0x32, /* USAGE (Z) */ + 0x95, 0x01, /* REPORT_COUNT (1) */ + 0x81, 0x06, /* INPUT (Data,Var,Rel) */ + + 0x06, 0x00, 0xff, /* Usage Page (Vendor Defined Page 1) */ + 0x19, 0x01, /* USAGE_MINIMUM (1) */ + 0x29, 0x03, /* USAGE_MAXIMUM (3) */ + 0x15, 0x00, /* LOGICAL_MINIMUM (0) */ + 0x25, 0x01, /* LOGICAL_MAXIMUM (1) */ + 0x95, 0x03, /* REPORT_COUNT (3) */ + 0x75, 0x01, /* REPORT_SIZE (1) */ + 0x91, 0x02, /* Output (Data,Var,Abs) */ + 0x95, 0x01, /* REPORT_COUNT (1) */ + 0x75, 0x05, /* REPORT_SIZE (5) */ + 0x91, 0x01, /* Output (Cnst,Var,Abs) */ + + 0x06, 0x00, 0xff, /* Usage Page (Vendor Defined Page 1) */ + 0x19, 0x06, /* USAGE_MINIMUM (6) */ + 0x29, 0x08, /* USAGE_MAXIMUM (8) */ + 0x15, 0x00, /* LOGICAL_MINIMUM (0) */ + 0x25, 0x01, /* LOGICAL_MAXIMUM (1) */ + 0x95, 0x03, /* REPORT_COUNT (3) */ + 0x75, 0x01, /* REPORT_SIZE (1) */ + 0xb1, 0x02, /* Feature (Data,Var,Abs) */ + 0x95, 0x01, /* REPORT_COUNT (1) */ + 0x75, 0x05, /* REPORT_SIZE (5) */ + 0x91, 0x01, /* Output (Cnst,Var,Abs) */ + + 0xc0, /* END_COLLECTION */ + 0xc0, /* END_COLLECTION */ +}; + +SEC("hid/rdesc_fixup") +int hid_rdesc_fixup(struct hid_bpf_ctx *ctx) +{ + callback2_check = ctx->data[4]; + + /* insert rdesc at offset 52 */ + __builtin_memcpy(&ctx->data[52], rdesc, sizeof(rdesc)); + ctx->size = sizeof(rdesc) + 52; + + ctx->data[4] = 0x42; + + return 0; +}
On Fri, Mar 04, 2022 at 06:28:34PM +0100, Benjamin Tissoires wrote:
Simple report descriptor override in HID: replace part of the report descriptor from a static definition in the bpf kernel program.
Note that this test should be run last because we disconnect/reconnect the device, meaning that it changes the overall uhid device.
You might want to add that in a comment right before:
@@ -329,6 +395,9 @@ void serial_test_hid_bpf(void) err = test_hid_raw_event(hid_skel, uhid_fd, sysfs_fd); ASSERT_OK(err, "hid");
- err = test_rdesc_fixup(hid_skel, uhid_fd, sysfs_fd);
- ASSERT_OK(err, "hid_rdesc_fixup");
There, so that you don't add another test after this later on in 5 years.
Anyway, just a nit:
Reviewed-by: Greg Kroah-Hartman gregkh@linuxfoundation.org
the program inverts the definition of X and Y at a given place in the report descriptor of my mouse.
Signed-off-by: Benjamin Tissoires benjamin.tissoires@redhat.com
---
changes in v2: - split the series by bpf/libbpf/hid/selftests and samples --- samples/bpf/hid_mouse_kern.c | 25 +++++++++++++++++++++++++ 1 file changed, 25 insertions(+)
diff --git a/samples/bpf/hid_mouse_kern.c b/samples/bpf/hid_mouse_kern.c index c24a12e06b40..958820caaf5d 100644 --- a/samples/bpf/hid_mouse_kern.c +++ b/samples/bpf/hid_mouse_kern.c @@ -62,5 +62,30 @@ int hid_x_event(struct hid_bpf_ctx *ctx) return 0; }
+SEC("hid/rdesc_fixup") +int hid_rdesc_fixup(struct hid_bpf_ctx *ctx) +{ + if (ctx->type != HID_BPF_RDESC_FIXUP) + return 0; + + bpf_printk("rdesc: %02x %02x %02x", + ctx->data[0], + ctx->data[1], + ctx->data[2]); + bpf_printk(" %02x %02x %02x", + ctx->data[3], + ctx->data[4], + ctx->data[5]); + bpf_printk(" %02x %02x %02x ...", + ctx->data[6], + ctx->data[7], + ctx->data[8]); + + ctx->data[39] = 0x31; + ctx->data[41] = 0x30; + + return 0; +} + char _license[] SEC("license") = "GPL"; u32 _version SEC("version") = LINUX_VERSION_CODE;
On Fri, Mar 04, 2022 at 06:28:35PM +0100, Benjamin Tissoires wrote:
the program inverts the definition of X and Y at a given place in the report descriptor of my mouse.
Signed-off-by: Benjamin Tissoires benjamin.tissoires@redhat.com
changes in v2:
- split the series by bpf/libbpf/hid/selftests and samples
samples/bpf/hid_mouse_kern.c | 25 +++++++++++++++++++++++++ 1 file changed, 25 insertions(+)
diff --git a/samples/bpf/hid_mouse_kern.c b/samples/bpf/hid_mouse_kern.c index c24a12e06b40..958820caaf5d 100644 --- a/samples/bpf/hid_mouse_kern.c +++ b/samples/bpf/hid_mouse_kern.c @@ -62,5 +62,30 @@ int hid_x_event(struct hid_bpf_ctx *ctx) return 0; } +SEC("hid/rdesc_fixup") +int hid_rdesc_fixup(struct hid_bpf_ctx *ctx)
No comment here to show the same as you put in the changelog saying what this function is doing?
Otherwise it's hard for a non-HID developer to know that:
- ctx->data[39] = 0x31;
- ctx->data[41] = 0x30;
Is flipping things.
thanks,
greg k-h
When we process an incoming HID report, it is common to have to account for fields that are not aligned in the report. HID is using 2 helpers hid_field_extract() and implement() to pick up any data at any offset within the report.
Export those 2 helpers in BPF programs so users can also rely on them. The second net worth advantage of those helpers is that now we can fetch data anywhere in the report without knowing at compile time the location of it. The boundary checks are done in hid-bpf.c, to prevent a memory leak.
Signed-off-by: Benjamin Tissoires benjamin.tissoires@redhat.com
---
changes in v2: - split the patch with libbpf and HID left outside. --- include/linux/bpf-hid.h | 4 +++ include/uapi/linux/bpf.h | 32 ++++++++++++++++++++ kernel/bpf/hid.c | 53 ++++++++++++++++++++++++++++++++++ tools/include/uapi/linux/bpf.h | 32 ++++++++++++++++++++ 4 files changed, 121 insertions(+)
diff --git a/include/linux/bpf-hid.h b/include/linux/bpf-hid.h index 0c5000b28b20..69bb28523ceb 100644 --- a/include/linux/bpf-hid.h +++ b/include/linux/bpf-hid.h @@ -93,6 +93,10 @@ struct bpf_hid_hooks { int (*link_attach)(struct hid_device *hdev, enum bpf_hid_attach_type type); void (*link_attached)(struct hid_device *hdev, enum bpf_hid_attach_type type); void (*array_detached)(struct hid_device *hdev, enum bpf_hid_attach_type type); + int (*hid_get_data)(struct hid_device *hdev, u8 *buf, size_t buf_size, + u64 offset, u32 n, u8 *data, u64 data_size); + int (*hid_set_data)(struct hid_device *hdev, u8 *buf, size_t buf_size, + u64 offset, u32 n, u8 *data, u64 data_size); };
#ifdef CONFIG_BPF diff --git a/include/uapi/linux/bpf.h b/include/uapi/linux/bpf.h index a7a8d9cfcf24..4845a20e6f96 100644 --- a/include/uapi/linux/bpf.h +++ b/include/uapi/linux/bpf.h @@ -5090,6 +5090,36 @@ union bpf_attr { * Return * 0 on success, or a negative error in case of failure. On error * *dst* buffer is zeroed out. + * + * int bpf_hid_get_data(void *ctx, u64 offset, u32 n, u8 *data, u64 size) + * Description + * Get the data of size n (in bits) at the given offset (bits) in the + * ctx->event.data field and store it into data. + * + * if n is less or equal than 32, we can address with bit precision, + * the value in the buffer. However, data must be a pointer to a u32 + * and size must be 4. + * + * if n is greater than 32, offset and n must be a multiple of 8 + * and the result is working with a memcpy internally. + * Return + * The length of data copied into data. On error, a negative value + * is returned. + * + * int bpf_hid_set_data(void *ctx, u64 offset, u32 n, u8 *data, u64 size) + * Description + * Set the data of size n (in bits) at the given offset (bits) in the + * ctx->event.data field. + * + * if n is less or equal than 32, we can address with bit precision, + * the value in the buffer. However, data must be a pointer to a u32 + * and size must be 4. + * + * if n is greater than 32, offset and n must be a multiple of 8 + * and the result is working with a memcpy internally. + * Return + * The length of data copied into ctx->event.data. On error, a negative + * value is returned. */ #define __BPF_FUNC_MAPPER(FN) \ FN(unspec), \ @@ -5284,6 +5314,8 @@ union bpf_attr { FN(xdp_load_bytes), \ FN(xdp_store_bytes), \ FN(copy_from_user_task), \ + FN(hid_get_data), \ + FN(hid_set_data), \ /* */
/* integer value in 'imm' field of BPF_CALL instruction selects which helper diff --git a/kernel/bpf/hid.c b/kernel/bpf/hid.c index 37500313e270..640e55ba66ec 100644 --- a/kernel/bpf/hid.c +++ b/kernel/bpf/hid.c @@ -37,10 +37,63 @@ void bpf_hid_set_hooks(struct bpf_hid_hooks *hooks) } EXPORT_SYMBOL_GPL(bpf_hid_set_hooks);
+BPF_CALL_5(bpf_hid_get_data, void*, ctx, u64, offset, u32, n, void*, data, u64, size) +{ + struct hid_bpf_ctx *bpf_ctx = ctx; + + if (!hid_hooks.hid_get_data) + return -EOPNOTSUPP; + + return hid_hooks.hid_get_data(bpf_ctx->hdev, + bpf_ctx->data, bpf_ctx->allocated_size, + offset, n, + data, size); +} + +static const struct bpf_func_proto bpf_hid_get_data_proto = { + .func = bpf_hid_get_data, + .gpl_only = true, + .ret_type = RET_INTEGER, + .arg1_type = ARG_PTR_TO_CTX, + .arg2_type = ARG_ANYTHING, + .arg3_type = ARG_ANYTHING, + .arg4_type = ARG_PTR_TO_MEM, + .arg5_type = ARG_CONST_SIZE_OR_ZERO, +}; + +BPF_CALL_5(bpf_hid_set_data, void*, ctx, u64, offset, u32, n, void*, data, u64, size) +{ + struct hid_bpf_ctx *bpf_ctx = ctx; + + if (!hid_hooks.hid_set_data) + return -EOPNOTSUPP; + + hid_hooks.hid_set_data(bpf_ctx->hdev, + bpf_ctx->data, bpf_ctx->allocated_size, + offset, n, + data, size); + return 0; +} + +static const struct bpf_func_proto bpf_hid_set_data_proto = { + .func = bpf_hid_set_data, + .gpl_only = true, + .ret_type = RET_INTEGER, + .arg1_type = ARG_PTR_TO_CTX, + .arg2_type = ARG_ANYTHING, + .arg3_type = ARG_ANYTHING, + .arg4_type = ARG_PTR_TO_MEM, + .arg5_type = ARG_CONST_SIZE_OR_ZERO, +}; + static const struct bpf_func_proto * hid_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) { switch (func_id) { + case BPF_FUNC_hid_get_data: + return &bpf_hid_get_data_proto; + case BPF_FUNC_hid_set_data: + return &bpf_hid_set_data_proto; default: return bpf_base_func_proto(func_id); } diff --git a/tools/include/uapi/linux/bpf.h b/tools/include/uapi/linux/bpf.h index a7a8d9cfcf24..4845a20e6f96 100644 --- a/tools/include/uapi/linux/bpf.h +++ b/tools/include/uapi/linux/bpf.h @@ -5090,6 +5090,36 @@ union bpf_attr { * Return * 0 on success, or a negative error in case of failure. On error * *dst* buffer is zeroed out. + * + * int bpf_hid_get_data(void *ctx, u64 offset, u32 n, u8 *data, u64 size) + * Description + * Get the data of size n (in bits) at the given offset (bits) in the + * ctx->event.data field and store it into data. + * + * if n is less or equal than 32, we can address with bit precision, + * the value in the buffer. However, data must be a pointer to a u32 + * and size must be 4. + * + * if n is greater than 32, offset and n must be a multiple of 8 + * and the result is working with a memcpy internally. + * Return + * The length of data copied into data. On error, a negative value + * is returned. + * + * int bpf_hid_set_data(void *ctx, u64 offset, u32 n, u8 *data, u64 size) + * Description + * Set the data of size n (in bits) at the given offset (bits) in the + * ctx->event.data field. + * + * if n is less or equal than 32, we can address with bit precision, + * the value in the buffer. However, data must be a pointer to a u32 + * and size must be 4. + * + * if n is greater than 32, offset and n must be a multiple of 8 + * and the result is working with a memcpy internally. + * Return + * The length of data copied into ctx->event.data. On error, a negative + * value is returned. */ #define __BPF_FUNC_MAPPER(FN) \ FN(unspec), \ @@ -5284,6 +5314,8 @@ union bpf_attr { FN(xdp_load_bytes), \ FN(xdp_store_bytes), \ FN(copy_from_user_task), \ + FN(hid_get_data), \ + FN(hid_set_data), \ /* */
/* integer value in 'imm' field of BPF_CALL instruction selects which helper
On Fri, Mar 04, 2022 at 06:28:36PM +0100, Benjamin Tissoires wrote:
When we process an incoming HID report, it is common to have to account for fields that are not aligned in the report. HID is using 2 helpers hid_field_extract() and implement() to pick up any data at any offset within the report.
Export those 2 helpers in BPF programs so users can also rely on them. The second net worth advantage of those helpers is that now we can fetch data anywhere in the report without knowing at compile time the location of it. The boundary checks are done in hid-bpf.c, to prevent a memory leak.
Signed-off-by: Benjamin Tissoires benjamin.tissoires@redhat.com
Reviewed-by: Greg Kroah-Hartman gregkh@linuxfoundation.org
On Fri, Mar 04, 2022 at 06:28:36PM +0100, Benjamin Tissoires wrote:
When we process an incoming HID report, it is common to have to account for fields that are not aligned in the report. HID is using 2 helpers hid_field_extract() and implement() to pick up any data at any offset within the report.
Export those 2 helpers in BPF programs so users can also rely on them. The second net worth advantage of those helpers is that now we can fetch data anywhere in the report without knowing at compile time the location of it. The boundary checks are done in hid-bpf.c, to prevent a memory leak.
Signed-off-by: Benjamin Tissoires benjamin.tissoires@redhat.com
changes in v2:
- split the patch with libbpf and HID left outside.
include/linux/bpf-hid.h | 4 +++ include/uapi/linux/bpf.h | 32 ++++++++++++++++++++ kernel/bpf/hid.c | 53 ++++++++++++++++++++++++++++++++++ tools/include/uapi/linux/bpf.h | 32 ++++++++++++++++++++ 4 files changed, 121 insertions(+)
diff --git a/include/linux/bpf-hid.h b/include/linux/bpf-hid.h index 0c5000b28b20..69bb28523ceb 100644 --- a/include/linux/bpf-hid.h +++ b/include/linux/bpf-hid.h @@ -93,6 +93,10 @@ struct bpf_hid_hooks { int (*link_attach)(struct hid_device *hdev, enum bpf_hid_attach_type type); void (*link_attached)(struct hid_device *hdev, enum bpf_hid_attach_type type); void (*array_detached)(struct hid_device *hdev, enum bpf_hid_attach_type type);
- int (*hid_get_data)(struct hid_device *hdev, u8 *buf, size_t buf_size,
u64 offset, u32 n, u8 *data, u64 data_size);- int (*hid_set_data)(struct hid_device *hdev, u8 *buf, size_t buf_size,
u64 offset, u32 n, u8 *data, u64 data_size);}; #ifdef CONFIG_BPF diff --git a/include/uapi/linux/bpf.h b/include/uapi/linux/bpf.h index a7a8d9cfcf24..4845a20e6f96 100644 --- a/include/uapi/linux/bpf.h +++ b/include/uapi/linux/bpf.h @@ -5090,6 +5090,36 @@ union bpf_attr {
- Return
0 on success, or a negative error in case of failure. On error*dst* buffer is zeroed out.
- int bpf_hid_get_data(void *ctx, u64 offset, u32 n, u8 *data, u64 size)
- Description
Get the data of size n (in bits) at the given offset (bits) in the
ctx->event.data field and store it into data.
if n is less or equal than 32, we can address with bit precision,
the value in the buffer. However, data must be a pointer to a u32
and size must be 4.
if n is greater than 32, offset and n must be a multiple of 8
and the result is working with a memcpy internally.
- Return
The length of data copied into data. On error, a negative value
is returned.
- int bpf_hid_set_data(void *ctx, u64 offset, u32 n, u8 *data, u64 size)
- Description
Set the data of size n (in bits) at the given offset (bits) in the
ctx->event.data field.
if n is less or equal than 32, we can address with bit precision,
the value in the buffer. However, data must be a pointer to a u32
and size must be 4.
if n is greater than 32, offset and n must be a multiple of 8
and the result is working with a memcpy internally.
- Return
The length of data copied into ctx->event.data. On error, a negative
value is returned.
Wait, nevermind my reviewed-by previously, see my comment about how this might be split into 4: bpf_hid_set_bytes() bpf_hid_get_bytes() bpf_hid_set_bits() bpf_hid_get_bits()
Should be easier to understand and maintain over time, right?
thanks,
greg k-h
On Fri, Mar 4, 2022 at 7:41 PM Greg KH gregkh@linuxfoundation.org wrote:
On Fri, Mar 04, 2022 at 06:28:36PM +0100, Benjamin Tissoires wrote:
When we process an incoming HID report, it is common to have to account for fields that are not aligned in the report. HID is using 2 helpers hid_field_extract() and implement() to pick up any data at any offset within the report.
Export those 2 helpers in BPF programs so users can also rely on them. The second net worth advantage of those helpers is that now we can fetch data anywhere in the report without knowing at compile time the location of it. The boundary checks are done in hid-bpf.c, to prevent a memory leak.
Signed-off-by: Benjamin Tissoires benjamin.tissoires@redhat.com
changes in v2:
- split the patch with libbpf and HID left outside.
include/linux/bpf-hid.h | 4 +++ include/uapi/linux/bpf.h | 32 ++++++++++++++++++++ kernel/bpf/hid.c | 53 ++++++++++++++++++++++++++++++++++ tools/include/uapi/linux/bpf.h | 32 ++++++++++++++++++++ 4 files changed, 121 insertions(+)
diff --git a/include/linux/bpf-hid.h b/include/linux/bpf-hid.h index 0c5000b28b20..69bb28523ceb 100644 --- a/include/linux/bpf-hid.h +++ b/include/linux/bpf-hid.h @@ -93,6 +93,10 @@ struct bpf_hid_hooks { int (*link_attach)(struct hid_device *hdev, enum bpf_hid_attach_type type); void (*link_attached)(struct hid_device *hdev, enum bpf_hid_attach_type type); void (*array_detached)(struct hid_device *hdev, enum bpf_hid_attach_type type);
int (*hid_get_data)(struct hid_device *hdev, u8 *buf, size_t buf_size,u64 offset, u32 n, u8 *data, u64 data_size);int (*hid_set_data)(struct hid_device *hdev, u8 *buf, size_t buf_size,u64 offset, u32 n, u8 *data, u64 data_size);};
#ifdef CONFIG_BPF diff --git a/include/uapi/linux/bpf.h b/include/uapi/linux/bpf.h index a7a8d9cfcf24..4845a20e6f96 100644 --- a/include/uapi/linux/bpf.h +++ b/include/uapi/linux/bpf.h @@ -5090,6 +5090,36 @@ union bpf_attr {
- Return
0 on success, or a negative error in case of failure. On error*dst* buffer is zeroed out.
- int bpf_hid_get_data(void *ctx, u64 offset, u32 n, u8 *data, u64 size)
- Description
Get the data of size n (in bits) at the given offset (bits) in the
ctx->event.data field and store it into data.
if n is less or equal than 32, we can address with bit precision,
the value in the buffer. However, data must be a pointer to a u32
and size must be 4.
if n is greater than 32, offset and n must be a multiple of 8
and the result is working with a memcpy internally.
- Return
The length of data copied into data. On error, a negative value
is returned.
- int bpf_hid_set_data(void *ctx, u64 offset, u32 n, u8 *data, u64 size)
- Description
Set the data of size n (in bits) at the given offset (bits) in the
ctx->event.data field.
if n is less or equal than 32, we can address with bit precision,
the value in the buffer. However, data must be a pointer to a u32
and size must be 4.
if n is greater than 32, offset and n must be a multiple of 8
and the result is working with a memcpy internally.
- Return
The length of data copied into ctx->event.data. On error, a negative
value is returned.
Quick answer on this one (before going deeper with the other remarks next week):
Wait, nevermind my reviewed-by previously, see my comment about how this might be split into 4: bpf_hid_set_bytes() bpf_hid_get_bytes() bpf_hid_set_bits() bpf_hid_get_bits()
Should be easier to understand and maintain over time, right?
Yes, definitively. I thought about adding a `bytes` suffix to the function name for n > 32, but not the `bits` one, meaning the API was still bunkers in my head.
And as I mentioned in the commit notes, I knew the API proposed in this patch was not correct, simply because while working on the selftests it was completely wrong :)
In terms of API usage, does it feel wrong for you to have only a subset of the array available "for free" and enforce the rest to be used through these helpers? That's the point I am not sure but I feel like 1024 (or slightly more) would be enough for most use cases, and when we are dealing with bigger data sets the helpers can be justified.
Thanks for the suggestion.
Cheers, Benjamin
thanks,
greg k-h
On Sat, Mar 05, 2022 at 11:33:07AM +0100, Benjamin Tissoires wrote:
On Fri, Mar 4, 2022 at 7:41 PM Greg KH gregkh@linuxfoundation.org wrote:
On Fri, Mar 04, 2022 at 06:28:36PM +0100, Benjamin Tissoires wrote:
When we process an incoming HID report, it is common to have to account for fields that are not aligned in the report. HID is using 2 helpers hid_field_extract() and implement() to pick up any data at any offset within the report.
Export those 2 helpers in BPF programs so users can also rely on them. The second net worth advantage of those helpers is that now we can fetch data anywhere in the report without knowing at compile time the location of it. The boundary checks are done in hid-bpf.c, to prevent a memory leak.
Signed-off-by: Benjamin Tissoires benjamin.tissoires@redhat.com
changes in v2:
- split the patch with libbpf and HID left outside.
include/linux/bpf-hid.h | 4 +++ include/uapi/linux/bpf.h | 32 ++++++++++++++++++++ kernel/bpf/hid.c | 53 ++++++++++++++++++++++++++++++++++ tools/include/uapi/linux/bpf.h | 32 ++++++++++++++++++++ 4 files changed, 121 insertions(+)
diff --git a/include/linux/bpf-hid.h b/include/linux/bpf-hid.h index 0c5000b28b20..69bb28523ceb 100644 --- a/include/linux/bpf-hid.h +++ b/include/linux/bpf-hid.h @@ -93,6 +93,10 @@ struct bpf_hid_hooks { int (*link_attach)(struct hid_device *hdev, enum bpf_hid_attach_type type); void (*link_attached)(struct hid_device *hdev, enum bpf_hid_attach_type type); void (*array_detached)(struct hid_device *hdev, enum bpf_hid_attach_type type);
int (*hid_get_data)(struct hid_device *hdev, u8 *buf, size_t buf_size,u64 offset, u32 n, u8 *data, u64 data_size);int (*hid_set_data)(struct hid_device *hdev, u8 *buf, size_t buf_size,u64 offset, u32 n, u8 *data, u64 data_size);};
#ifdef CONFIG_BPF diff --git a/include/uapi/linux/bpf.h b/include/uapi/linux/bpf.h index a7a8d9cfcf24..4845a20e6f96 100644 --- a/include/uapi/linux/bpf.h +++ b/include/uapi/linux/bpf.h @@ -5090,6 +5090,36 @@ union bpf_attr {
- Return
0 on success, or a negative error in case of failure. On error*dst* buffer is zeroed out.
- int bpf_hid_get_data(void *ctx, u64 offset, u32 n, u8 *data, u64 size)
- Description
Get the data of size n (in bits) at the given offset (bits) in the
ctx->event.data field and store it into data.
if n is less or equal than 32, we can address with bit precision,
the value in the buffer. However, data must be a pointer to a u32
and size must be 4.
if n is greater than 32, offset and n must be a multiple of 8
and the result is working with a memcpy internally.
- Return
The length of data copied into data. On error, a negative value
is returned.
- int bpf_hid_set_data(void *ctx, u64 offset, u32 n, u8 *data, u64 size)
- Description
Set the data of size n (in bits) at the given offset (bits) in the
ctx->event.data field.
if n is less or equal than 32, we can address with bit precision,
the value in the buffer. However, data must be a pointer to a u32
and size must be 4.
if n is greater than 32, offset and n must be a multiple of 8
and the result is working with a memcpy internally.
- Return
The length of data copied into ctx->event.data. On error, a negative
value is returned.Quick answer on this one (before going deeper with the other remarks next week):
Wait, nevermind my reviewed-by previously, see my comment about how this might be split into 4: bpf_hid_set_bytes() bpf_hid_get_bytes() bpf_hid_set_bits() bpf_hid_get_bits()
Should be easier to understand and maintain over time, right?
Yes, definitively. I thought about adding a `bytes` suffix to the function name for n > 32, but not the `bits` one, meaning the API was still bunkers in my head.
And as I mentioned in the commit notes, I knew the API proposed in this patch was not correct, simply because while working on the selftests it was completely wrong :)
In terms of API usage, does it feel wrong for you to have only a subset of the array available "for free" and enforce the rest to be used through these helpers?
It does feel "odd" but:
That's the point I am not sure but I feel like 1024 (or slightly more) would be enough for most use cases, and when we are dealing with bigger data sets the helpers can be justified.
How often are hid reports that big? If providing access to the first 1024 or so bytes for free would handle the huge majority of fixup cases, and only if you have crazy devices with large reports would you have to use the functions, it might be fine.
The problem is, only time will tell, and then, it's too late to change the api :)
So I agree with this for now, stick with what you have (if you split this into 4 functions), and let's see what happens.
thanks,
greg k-h
On Sat, Mar 5, 2022 at 2:47 AM Greg KH gregkh@linuxfoundation.org wrote:
On Sat, Mar 05, 2022 at 11:33:07AM +0100, Benjamin Tissoires wrote:
On Fri, Mar 4, 2022 at 7:41 PM Greg KH gregkh@linuxfoundation.org wrote:
On Fri, Mar 04, 2022 at 06:28:36PM +0100, Benjamin Tissoires wrote:
When we process an incoming HID report, it is common to have to account for fields that are not aligned in the report. HID is using 2 helpers hid_field_extract() and implement() to pick up any data at any offset within the report.
Export those 2 helpers in BPF programs so users can also rely on them. The second net worth advantage of those helpers is that now we can fetch data anywhere in the report without knowing at compile time the location of it. The boundary checks are done in hid-bpf.c, to prevent a memory leak.
Signed-off-by: Benjamin Tissoires benjamin.tissoires@redhat.com
changes in v2:
- split the patch with libbpf and HID left outside.
include/linux/bpf-hid.h | 4 +++ include/uapi/linux/bpf.h | 32 ++++++++++++++++++++ kernel/bpf/hid.c | 53 ++++++++++++++++++++++++++++++++++ tools/include/uapi/linux/bpf.h | 32 ++++++++++++++++++++ 4 files changed, 121 insertions(+)
diff --git a/include/linux/bpf-hid.h b/include/linux/bpf-hid.h index 0c5000b28b20..69bb28523ceb 100644 --- a/include/linux/bpf-hid.h +++ b/include/linux/bpf-hid.h @@ -93,6 +93,10 @@ struct bpf_hid_hooks { int (*link_attach)(struct hid_device *hdev, enum bpf_hid_attach_type type); void (*link_attached)(struct hid_device *hdev, enum bpf_hid_attach_type type); void (*array_detached)(struct hid_device *hdev, enum bpf_hid_attach_type type);
int (*hid_get_data)(struct hid_device *hdev, u8 *buf, size_t buf_size,u64 offset, u32 n, u8 *data, u64 data_size);int (*hid_set_data)(struct hid_device *hdev, u8 *buf, size_t buf_size,u64 offset, u32 n, u8 *data, u64 data_size);};
#ifdef CONFIG_BPF diff --git a/include/uapi/linux/bpf.h b/include/uapi/linux/bpf.h index a7a8d9cfcf24..4845a20e6f96 100644 --- a/include/uapi/linux/bpf.h +++ b/include/uapi/linux/bpf.h @@ -5090,6 +5090,36 @@ union bpf_attr {
- Return
0 on success, or a negative error in case of failure. On error*dst* buffer is zeroed out.
- int bpf_hid_get_data(void *ctx, u64 offset, u32 n, u8 *data, u64 size)
- Description
Get the data of size n (in bits) at the given offset (bits) in the
ctx->event.data field and store it into data.
if n is less or equal than 32, we can address with bit precision,
the value in the buffer. However, data must be a pointer to a u32
and size must be 4.
if n is greater than 32, offset and n must be a multiple of 8
and the result is working with a memcpy internally.
- Return
The length of data copied into data. On error, a negative value
is returned.
- int bpf_hid_set_data(void *ctx, u64 offset, u32 n, u8 *data, u64 size)
- Description
Set the data of size n (in bits) at the given offset (bits) in the
ctx->event.data field.
if n is less or equal than 32, we can address with bit precision,
the value in the buffer. However, data must be a pointer to a u32
and size must be 4.
if n is greater than 32, offset and n must be a multiple of 8
and the result is working with a memcpy internally.
- Return
The length of data copied into ctx->event.data. On error, a negative
value is returned.Quick answer on this one (before going deeper with the other remarks next week):
Wait, nevermind my reviewed-by previously, see my comment about how this might be split into 4: bpf_hid_set_bytes() bpf_hid_get_bytes() bpf_hid_set_bits() bpf_hid_get_bits()
Should be easier to understand and maintain over time, right?
Yes, definitively. I thought about adding a `bytes` suffix to the function name for n > 32, but not the `bits` one, meaning the API was still bunkers in my head.
Do we really need per-bit access? I was under the impression that only one BPF program is working on a ctx/buffer at a time, so we can just do read-modify-write at byte level, no?
Thanks, Song
On Fri, Mar 11, 2022 at 1:41 AM Song Liu song@kernel.org wrote:
On Sat, Mar 5, 2022 at 2:47 AM Greg KH gregkh@linuxfoundation.org wrote:
On Sat, Mar 05, 2022 at 11:33:07AM +0100, Benjamin Tissoires wrote:
On Fri, Mar 4, 2022 at 7:41 PM Greg KH gregkh@linuxfoundation.org wrote:
On Fri, Mar 04, 2022 at 06:28:36PM +0100, Benjamin Tissoires wrote:
When we process an incoming HID report, it is common to have to account for fields that are not aligned in the report. HID is using 2 helpers hid_field_extract() and implement() to pick up any data at any offset within the report.
Export those 2 helpers in BPF programs so users can also rely on them. The second net worth advantage of those helpers is that now we can fetch data anywhere in the report without knowing at compile time the location of it. The boundary checks are done in hid-bpf.c, to prevent a memory leak.
Signed-off-by: Benjamin Tissoires benjamin.tissoires@redhat.com
changes in v2:
- split the patch with libbpf and HID left outside.
include/linux/bpf-hid.h | 4 +++ include/uapi/linux/bpf.h | 32 ++++++++++++++++++++ kernel/bpf/hid.c | 53 ++++++++++++++++++++++++++++++++++ tools/include/uapi/linux/bpf.h | 32 ++++++++++++++++++++ 4 files changed, 121 insertions(+)
diff --git a/include/linux/bpf-hid.h b/include/linux/bpf-hid.h index 0c5000b28b20..69bb28523ceb 100644 --- a/include/linux/bpf-hid.h +++ b/include/linux/bpf-hid.h @@ -93,6 +93,10 @@ struct bpf_hid_hooks { int (*link_attach)(struct hid_device *hdev, enum bpf_hid_attach_type type); void (*link_attached)(struct hid_device *hdev, enum bpf_hid_attach_type type); void (*array_detached)(struct hid_device *hdev, enum bpf_hid_attach_type type);
int (*hid_get_data)(struct hid_device *hdev, u8 *buf, size_t buf_size,u64 offset, u32 n, u8 *data, u64 data_size);int (*hid_set_data)(struct hid_device *hdev, u8 *buf, size_t buf_size,u64 offset, u32 n, u8 *data, u64 data_size);};
#ifdef CONFIG_BPF diff --git a/include/uapi/linux/bpf.h b/include/uapi/linux/bpf.h index a7a8d9cfcf24..4845a20e6f96 100644 --- a/include/uapi/linux/bpf.h +++ b/include/uapi/linux/bpf.h @@ -5090,6 +5090,36 @@ union bpf_attr {
- Return
0 on success, or a negative error in case of failure. On error*dst* buffer is zeroed out.
- int bpf_hid_get_data(void *ctx, u64 offset, u32 n, u8 *data, u64 size)
- Description
Get the data of size n (in bits) at the given offset (bits) in the
ctx->event.data field and store it into data.
if n is less or equal than 32, we can address with bit precision,
the value in the buffer. However, data must be a pointer to a u32
and size must be 4.
if n is greater than 32, offset and n must be a multiple of 8
and the result is working with a memcpy internally.
- Return
The length of data copied into data. On error, a negative value
is returned.
- int bpf_hid_set_data(void *ctx, u64 offset, u32 n, u8 *data, u64 size)
- Description
Set the data of size n (in bits) at the given offset (bits) in the
ctx->event.data field.
if n is less or equal than 32, we can address with bit precision,
the value in the buffer. However, data must be a pointer to a u32
and size must be 4.
if n is greater than 32, offset and n must be a multiple of 8
and the result is working with a memcpy internally.
- Return
The length of data copied into ctx->event.data. On error, a negative
value is returned.Quick answer on this one (before going deeper with the other remarks next week):
Wait, nevermind my reviewed-by previously, see my comment about how this might be split into 4: bpf_hid_set_bytes() bpf_hid_get_bytes() bpf_hid_set_bits() bpf_hid_get_bits()
Should be easier to understand and maintain over time, right?
Yes, definitively. I thought about adding a `bytes` suffix to the function name for n > 32, but not the `bits` one, meaning the API was still bunkers in my head.
Do we really need per-bit access? I was under the impression that only one BPF program is working on a ctx/buffer at a time, so we can just do read-modify-write at byte level, no?
Yes, we really need per-bit access, and yes only one BPF program is working on a ctx/buffer at a time.
The per-bit access is a HID requirement and a much more convenient way of accessing data in the buffer. Well, there is another advantage too that I'll add later.
Basically, in the HID world, HW makers are trying to 'compact' the reports their device is sending to a minimum value.
For instance, when you have a 3 buttons + wheel mouse you may need: 3 bits of information for the 3 buttons 4 bits for the wheel 16 bits for X 16 bits for Y.
This usually translates almost verbatim in the report (we can add one bit of padding between buttons and wheel), which means that accessing the wheel data requires the user to access the offset 4 (bits) of size 4 bits in the report.
Some HW vendors are not even bothering aligning the data, so this can be messy from time to time with just plain byte access. All in all, the HID report descriptor gives you that information, and internally, the HID stack stores the offset in bits and the sizes in bits to access them without too much trouble.
The second advantage I have with these 2 accessors is that it allows me to not know statically the offset and size values. Because the helper in the kernel checks them for me, I can use registers values that are unknown to the verifier and can basically have:
``` __u64 offsetX = 0; __u64 offsetY = 0; __u32 sizeX = 0; __u32 sizeY = 0;
SEC("hid/device_event") int invert_xy(struct hid_bpf_ctx *ctx) { __u16 x, y;
if (sizeX == 16) { x = bpf_hid_get_bits(ctx, offsetX, sizeX); bpf_hid_set_bits(ctx, offsetX, -x); } if (sizeY == 16) { y = bpf_hid_get_bits(ctx, offsetY, sizeY); bpf_hid_set_bits(ctx, offsetY, -y); } return 0; } ```
Then, I have my userspace program parse the report descriptor, set the correct values for size{X|Y} and offset{X|Y} and I can have this program compiled once and redistributed many times.
Cheers, Benjamin
We have 2 cases of usage here: - either n <= 32: we are addressing individual bits at the given offset
- either n > 32: we are using a memcpy to transmit the data to the caller, meaning that we need to be byte-aligned.
Signed-off-by: Benjamin Tissoires benjamin.tissoires@redhat.com
---
changes in v2: - split the series by bpf/libbpf/hid/selftests and samples - allow for n > 32, by relying on memcpy --- drivers/hid/hid-bpf.c | 68 ++++++++++++++++++++++++++++++++++++++++++ drivers/hid/hid-core.c | 4 +-- include/linux/hid.h | 2 ++ 3 files changed, 72 insertions(+), 2 deletions(-)
diff --git a/drivers/hid/hid-bpf.c b/drivers/hid/hid-bpf.c index 510e24f4307c..8ae247fba5bc 100644 --- a/drivers/hid/hid-bpf.c +++ b/drivers/hid/hid-bpf.c @@ -105,6 +105,72 @@ static void hid_bpf_array_detached(struct hid_device *hdev, enum bpf_hid_attach_ } }
+int hid_bpf_get_data(struct hid_device *hdev, u8 *buf, size_t buf_size, u64 offset, u32 n, + u8 *data, u64 data_size) +{ + u32 *value = (u32 *)data; + + if (((offset + n) >> 3) >= buf_size) + return -E2BIG; + + if (n <= 32) { + /* data must be a pointer to a u32 */ + if (data_size != 4) + return -EINVAL; + + *value = hid_field_extract(hdev, buf, offset, n); + return 4; + } + + /* if n > 32, use memcpy, but ensure we are dealing with full bytes */ + if ((n | offset) & 0x7) + return -EINVAL; + + /* work on bytes now */ + offset = offset >> 3; + n = n >> 3; + + if (n > data_size) + return -EINVAL; + + memcpy(data, buf + offset, n); + + return n; +} + +int hid_bpf_set_data(struct hid_device *hdev, u8 *buf, size_t buf_size, u64 offset, u32 n, + u8 *data, u64 data_size) +{ + u32 *value = (u32 *)data; + + if (((offset + n) >> 3) >= buf_size) + return -E2BIG; + + if (n <= 32) { + /* data must be a pointer to a u32 */ + if (data_size != 4) + return -EINVAL; + + implement(hdev, buf, offset, n, *value); + return 4; + } + + /* if n > 32, use memcpy, but ensure we are dealing with full bytes */ + if ((n | offset) & 0x7) + return -EINVAL; + + /* work on bytes now */ + offset = offset >> 3; + n = n >> 3; + + if (n > data_size) + return -EINVAL; + + memcpy(buf + offset, data, n); + + return n; +} + static int hid_bpf_run_progs(struct hid_device *hdev, enum bpf_hid_attach_type type, struct hid_bpf_ctx *ctx, u8 *data, int size) { @@ -204,6 +270,8 @@ int __init hid_bpf_module_init(void) .link_attach = hid_bpf_link_attach, .link_attached = hid_bpf_link_attached, .array_detached = hid_bpf_array_detached, + .hid_get_data = hid_bpf_get_data, + .hid_set_data = hid_bpf_set_data, };
bpf_hid_set_hooks(&hooks); diff --git a/drivers/hid/hid-core.c b/drivers/hid/hid-core.c index 0eb8189faaee..d3f4499ee4cd 100644 --- a/drivers/hid/hid-core.c +++ b/drivers/hid/hid-core.c @@ -1416,8 +1416,8 @@ static void __implement(u8 *report, unsigned offset, int n, u32 value) } }
-static void implement(const struct hid_device *hid, u8 *report, - unsigned offset, unsigned n, u32 value) +void implement(const struct hid_device *hid, u8 *report, unsigned int offset, unsigned int n, + u32 value) { if (unlikely(n > 32)) { hid_warn(hid, "%s() called with n (%d) > 32! (%s)\n", diff --git a/include/linux/hid.h b/include/linux/hid.h index 66d949d10b78..7454e844324c 100644 --- a/include/linux/hid.h +++ b/include/linux/hid.h @@ -944,6 +944,8 @@ bool hid_compare_device_paths(struct hid_device *hdev_a, s32 hid_snto32(__u32 value, unsigned n); __u32 hid_field_extract(const struct hid_device *hid, __u8 *report, unsigned offset, unsigned n); +void implement(const struct hid_device *hid, u8 *report, unsigned int offset, unsigned int n, + u32 value);
#ifdef CONFIG_PM int hid_driver_suspend(struct hid_device *hdev, pm_message_t state);
On Fri, Mar 04, 2022 at 06:28:37PM +0100, Benjamin Tissoires wrote:
We have 2 cases of usage here:
either n <= 32: we are addressing individual bits at the given offset
either n > 32: we are using a memcpy to transmit the data to the caller, meaning that we need to be byte-aligned.
That is crazy. Can you split this into a bits/bytes function set to have 4 of them? You are using each function in two totally different ways.
thanks,
greg k-h
Simple test added here, with one use of each helper.
Signed-off-by: Benjamin Tissoires benjamin.tissoires@redhat.com
---
changes in v2: - split the patch with libbpf left outside. --- tools/testing/selftests/bpf/prog_tests/hid.c | 65 ++++++++++++++++++++ tools/testing/selftests/bpf/progs/hid.c | 45 ++++++++++++++ 2 files changed, 110 insertions(+)
diff --git a/tools/testing/selftests/bpf/prog_tests/hid.c b/tools/testing/selftests/bpf/prog_tests/hid.c index 91543b8078ca..74426523dd6f 100644 --- a/tools/testing/selftests/bpf/prog_tests/hid.c +++ b/tools/testing/selftests/bpf/prog_tests/hid.c @@ -297,6 +297,68 @@ static int test_hid_raw_event(struct hid *hid_skel, int uhid_fd, int sysfs_fd) return ret; }
+/* + * Attach hid_set_get_data to the given uhid device, + * retrieve and open the matching hidraw node, + * inject one event in the uhid device, + * check that the program makes correct use of bpf_hid_{set|get}_data. + */ +static int test_hid_set_get_data(struct hid *hid_skel, int uhid_fd, int sysfs_fd) +{ + int err, hidraw_ino, hidraw_fd = -1; + char hidraw_path[64] = {0}; + u8 buf[10] = {0}; + int ret = -1; + + /* attach hid_set_get_data program */ + hid_skel->links.hid_set_get_data = + bpf_program__attach_hid(hid_skel->progs.hid_set_get_data, sysfs_fd); + if (!ASSERT_OK_PTR(hid_skel->links.hid_set_get_data, + "attach_hid(hid_set_get_data)")) + return PTR_ERR(hid_skel->links.hid_set_get_data); + + hidraw_ino = get_hidraw(hid_skel->links.hid_set_get_data); + if (!ASSERT_GE(hidraw_ino, 0, "get_hidraw")) + goto cleanup; + + /* open hidraw node to check the other side of the pipe */ + sprintf(hidraw_path, "/dev/hidraw%d", hidraw_ino); + hidraw_fd = open(hidraw_path, O_RDWR | O_NONBLOCK); + + if (!ASSERT_GE(hidraw_fd, 0, "open_hidraw")) + goto cleanup; + + /* inject one event */ + buf[0] = 1; + buf[1] = 42; + send_event(uhid_fd, buf, 6); + + /* read the data from hidraw */ + memset(buf, 0, sizeof(buf)); + err = read(hidraw_fd, buf, sizeof(buf)); + if (!ASSERT_EQ(err, 6, "read_hidraw")) + goto cleanup; + + if (!ASSERT_EQ(buf[2], (42 >> 2), "hid_set_get_data")) + goto cleanup; + + if (!ASSERT_EQ(buf[3], 1, "hid_set_get_data")) + goto cleanup; + + if (!ASSERT_EQ(buf[4], 42, "hid_set_get_data")) + goto cleanup; + + ret = 0; + +cleanup: + if (hidraw_fd >= 0) + close(hidraw_fd); + + hid__detach(hid_skel); + + return ret; +} + /* * Attach hid_rdesc_fixup to the given uhid device, * retrieve and open the matching hidraw node, @@ -395,6 +457,9 @@ void serial_test_hid_bpf(void) err = test_hid_raw_event(hid_skel, uhid_fd, sysfs_fd); ASSERT_OK(err, "hid");
+ err = test_hid_set_get_data(hid_skel, uhid_fd, sysfs_fd); + ASSERT_OK(err, "hid_set_get_data"); + err = test_rdesc_fixup(hid_skel, uhid_fd, sysfs_fd); ASSERT_OK(err, "hid_rdesc_fixup");
diff --git a/tools/testing/selftests/bpf/progs/hid.c b/tools/testing/selftests/bpf/progs/hid.c index 2270448d0d3f..de6668471940 100644 --- a/tools/testing/selftests/bpf/progs/hid.c +++ b/tools/testing/selftests/bpf/progs/hid.c @@ -66,3 +66,48 @@ int hid_rdesc_fixup(struct hid_bpf_ctx *ctx)
return 0; } + +SEC("hid/device_event") +int hid_set_get_data(struct hid_bpf_ctx *ctx) +{ + int ret; + __u8 *buf; + + buf = bpf_ringbuf_reserve(&ringbuf, 8, 0); + if (!buf) + return -12; /* -ENOMEM */ + + /* first try read/write with n > 32 */ + ret = bpf_hid_get_data(ctx, 0, 64, buf, 8); + if (ret < 0) + goto discard; + + /* reinject it */ + ret = bpf_hid_set_data(ctx, 24, 64, buf, 8); + if (ret < 0) + goto discard; + + /* extract data at bit offset 10 of size 4 (half a byte) */ + ret = bpf_hid_get_data(ctx, 10, 4, buf, 8); /* expected to fail */ + if (ret > 0) { + ret = -1; + goto discard; + } + + ret = bpf_hid_get_data(ctx, 10, 4, buf, 4); + if (ret < 0) + goto discard; + + /* reinject it */ + ret = bpf_hid_set_data(ctx, 16, 4, buf, 4); + if (ret < 0) + goto discard; + + ret = 0; + + discard: + + bpf_ringbuf_discard(buf, 0); + + return ret; +}
Hi Benjamin,
On 04/03/2022 19:28, Benjamin Tissoires wrote:
Simple test added here, with one use of each helper.
Signed-off-by: Benjamin Tissoires benjamin.tissoires@redhat.com
changes in v2:
- split the patch with libbpf left outside.
tools/testing/selftests/bpf/prog_tests/hid.c | 65 ++++++++++++++++++++ tools/testing/selftests/bpf/progs/hid.c | 45 ++++++++++++++ 2 files changed, 110 insertions(+)
diff --git a/tools/testing/selftests/bpf/prog_tests/hid.c b/tools/testing/selftests/bpf/prog_tests/hid.c index 91543b8078ca..74426523dd6f 100644 --- a/tools/testing/selftests/bpf/prog_tests/hid.c +++ b/tools/testing/selftests/bpf/prog_tests/hid.c @@ -297,6 +297,68 @@ static int test_hid_raw_event(struct hid *hid_skel, int uhid_fd, int sysfs_fd) return ret; } +/*
- Attach hid_set_get_data to the given uhid device,
- retrieve and open the matching hidraw node,
- inject one event in the uhid device,
- check that the program makes correct use of bpf_hid_{set|get}_data.
- */
+static int test_hid_set_get_data(struct hid *hid_skel, int uhid_fd, int sysfs_fd) +{
- int err, hidraw_ino, hidraw_fd = -1;
- char hidraw_path[64] = {0};
- u8 buf[10] = {0};
- int ret = -1;
- /* attach hid_set_get_data program */
- hid_skel->links.hid_set_get_data =
bpf_program__attach_hid(hid_skel->progs.hid_set_get_data, sysfs_fd);- if (!ASSERT_OK_PTR(hid_skel->links.hid_set_get_data,
"attach_hid(hid_set_get_data)"))return PTR_ERR(hid_skel->links.hid_set_get_data);- hidraw_ino = get_hidraw(hid_skel->links.hid_set_get_data);
- if (!ASSERT_GE(hidraw_ino, 0, "get_hidraw"))
goto cleanup;- /* open hidraw node to check the other side of the pipe */
- sprintf(hidraw_path, "/dev/hidraw%d", hidraw_ino);
- hidraw_fd = open(hidraw_path, O_RDWR | O_NONBLOCK);
- if (!ASSERT_GE(hidraw_fd, 0, "open_hidraw"))
goto cleanup;- /* inject one event */
- buf[0] = 1;
- buf[1] = 42;
- send_event(uhid_fd, buf, 6);
- /* read the data from hidraw */
- memset(buf, 0, sizeof(buf));
- err = read(hidraw_fd, buf, sizeof(buf));
- if (!ASSERT_EQ(err, 6, "read_hidraw"))
goto cleanup;- if (!ASSERT_EQ(buf[2], (42 >> 2), "hid_set_get_data"))
goto cleanup;- if (!ASSERT_EQ(buf[3], 1, "hid_set_get_data"))
goto cleanup;- if (!ASSERT_EQ(buf[4], 42, "hid_set_get_data"))
goto cleanup;- ret = 0;
+cleanup:
- if (hidraw_fd >= 0)
close(hidraw_fd);- hid__detach(hid_skel);
- return ret;
+}
- /*
- Attach hid_rdesc_fixup to the given uhid device,
- retrieve and open the matching hidraw node,
@@ -395,6 +457,9 @@ void serial_test_hid_bpf(void) err = test_hid_raw_event(hid_skel, uhid_fd, sysfs_fd); ASSERT_OK(err, "hid");
- err = test_hid_set_get_data(hid_skel, uhid_fd, sysfs_fd);
- ASSERT_OK(err, "hid_set_get_data");
- err = test_rdesc_fixup(hid_skel, uhid_fd, sysfs_fd); ASSERT_OK(err, "hid_rdesc_fixup");
diff --git a/tools/testing/selftests/bpf/progs/hid.c b/tools/testing/selftests/bpf/progs/hid.c index 2270448d0d3f..de6668471940 100644 --- a/tools/testing/selftests/bpf/progs/hid.c +++ b/tools/testing/selftests/bpf/progs/hid.c @@ -66,3 +66,48 @@ int hid_rdesc_fixup(struct hid_bpf_ctx *ctx) return 0; }
+SEC("hid/device_event") +int hid_set_get_data(struct hid_bpf_ctx *ctx) +{
- int ret;
- __u8 *buf;
- buf = bpf_ringbuf_reserve(&ringbuf, 8, 0);
Ordering of patches is probably wrong, it seems the ringbuf is defined in patch #21 but used here.
Also, this usage of ringbuf leads into running out of available memory in the buffer if used for long time, it is not evident from the test case written here but I spent a couple of hours debugging my own BPF program that used ringbuf in similar way as what is done here. Basically the producer idx is increased with the bpf_ringbuf_reserve / discard, but the consumer index is not if you don't have a consumer in place.
I ended up using a global statically allocated buffer for the purpose for now.
-Tero
- if (!buf)
return -12; /* -ENOMEM */- /* first try read/write with n > 32 */
- ret = bpf_hid_get_data(ctx, 0, 64, buf, 8);
- if (ret < 0)
goto discard;- /* reinject it */
- ret = bpf_hid_set_data(ctx, 24, 64, buf, 8);
- if (ret < 0)
goto discard;- /* extract data at bit offset 10 of size 4 (half a byte) */
- ret = bpf_hid_get_data(ctx, 10, 4, buf, 8); /* expected to fail */
- if (ret > 0) {
ret = -1;goto discard;- }
- ret = bpf_hid_get_data(ctx, 10, 4, buf, 4);
- if (ret < 0)
goto discard;- /* reinject it */
- ret = bpf_hid_set_data(ctx, 16, 4, buf, 4);
- if (ret < 0)
goto discard;- ret = 0;
- discard:
- bpf_ringbuf_discard(buf, 0);
- return ret;
+}
On Tue, Mar 15, 2022 at 5:51 PM Tero Kristo tero.kristo@linux.intel.com wrote:
Hi Benjamin,
On 04/03/2022 19:28, Benjamin Tissoires wrote:
Simple test added here, with one use of each helper.
Signed-off-by: Benjamin Tissoires benjamin.tissoires@redhat.com
changes in v2:
- split the patch with libbpf left outside.
tools/testing/selftests/bpf/prog_tests/hid.c | 65 ++++++++++++++++++++ tools/testing/selftests/bpf/progs/hid.c | 45 ++++++++++++++ 2 files changed, 110 insertions(+)
diff --git a/tools/testing/selftests/bpf/prog_tests/hid.c b/tools/testing/selftests/bpf/prog_tests/hid.c index 91543b8078ca..74426523dd6f 100644 --- a/tools/testing/selftests/bpf/prog_tests/hid.c +++ b/tools/testing/selftests/bpf/prog_tests/hid.c @@ -297,6 +297,68 @@ static int test_hid_raw_event(struct hid *hid_skel, int uhid_fd, int sysfs_fd) return ret; }
+/*
- Attach hid_set_get_data to the given uhid device,
- retrieve and open the matching hidraw node,
- inject one event in the uhid device,
- check that the program makes correct use of bpf_hid_{set|get}_data.
- */
+static int test_hid_set_get_data(struct hid *hid_skel, int uhid_fd, int sysfs_fd) +{
int err, hidraw_ino, hidraw_fd = -1;char hidraw_path[64] = {0};u8 buf[10] = {0};int ret = -1;/* attach hid_set_get_data program */hid_skel->links.hid_set_get_data =bpf_program__attach_hid(hid_skel->progs.hid_set_get_data, sysfs_fd);if (!ASSERT_OK_PTR(hid_skel->links.hid_set_get_data,"attach_hid(hid_set_get_data)"))return PTR_ERR(hid_skel->links.hid_set_get_data);hidraw_ino = get_hidraw(hid_skel->links.hid_set_get_data);if (!ASSERT_GE(hidraw_ino, 0, "get_hidraw"))goto cleanup;/* open hidraw node to check the other side of the pipe */sprintf(hidraw_path, "/dev/hidraw%d", hidraw_ino);hidraw_fd = open(hidraw_path, O_RDWR | O_NONBLOCK);if (!ASSERT_GE(hidraw_fd, 0, "open_hidraw"))goto cleanup;/* inject one event */buf[0] = 1;buf[1] = 42;send_event(uhid_fd, buf, 6);/* read the data from hidraw */memset(buf, 0, sizeof(buf));err = read(hidraw_fd, buf, sizeof(buf));if (!ASSERT_EQ(err, 6, "read_hidraw"))goto cleanup;if (!ASSERT_EQ(buf[2], (42 >> 2), "hid_set_get_data"))goto cleanup;if (!ASSERT_EQ(buf[3], 1, "hid_set_get_data"))goto cleanup;if (!ASSERT_EQ(buf[4], 42, "hid_set_get_data"))goto cleanup;ret = 0;+cleanup:
if (hidraw_fd >= 0)close(hidraw_fd);hid__detach(hid_skel);return ret;+}
- /*
- Attach hid_rdesc_fixup to the given uhid device,
- retrieve and open the matching hidraw node,
@@ -395,6 +457,9 @@ void serial_test_hid_bpf(void) err = test_hid_raw_event(hid_skel, uhid_fd, sysfs_fd); ASSERT_OK(err, "hid");
err = test_hid_set_get_data(hid_skel, uhid_fd, sysfs_fd);ASSERT_OK(err, "hid_set_get_data");err = test_rdesc_fixup(hid_skel, uhid_fd, sysfs_fd); ASSERT_OK(err, "hid_rdesc_fixup");diff --git a/tools/testing/selftests/bpf/progs/hid.c b/tools/testing/selftests/bpf/progs/hid.c index 2270448d0d3f..de6668471940 100644 --- a/tools/testing/selftests/bpf/progs/hid.c +++ b/tools/testing/selftests/bpf/progs/hid.c @@ -66,3 +66,48 @@ int hid_rdesc_fixup(struct hid_bpf_ctx *ctx)
return 0;}
+SEC("hid/device_event") +int hid_set_get_data(struct hid_bpf_ctx *ctx) +{
int ret;__u8 *buf;buf = bpf_ringbuf_reserve(&ringbuf, 8, 0);Ordering of patches is probably wrong, it seems the ringbuf is defined in patch #21 but used here.
Also, this usage of ringbuf leads into running out of available memory in the buffer if used for long time, it is not evident from the test case written here but I spent a couple of hours debugging my own BPF program that used ringbuf in similar way as what is done here. Basically the producer idx is increased with the bpf_ringbuf_reserve / discard, but the consumer index is not if you don't have a consumer in place.
Oh, that's good to know.
FWIW, on v3, we won't have to use a ringbuf at all. This is a new API break (sorry for that), but I think this will be better in the long run. I have chosen to use the hid_bpf_get_data() from [0]. The net advantage is that we get a pointer to the internal buffer that can be of any size, and we can even re-inject that same pointer into bpf_hid_raw_request() without having to use another buffer.
I'm currently working on the documentation part and small fixes now that the big rewrite with that implementation is now done.
Cheers, Benjamin
I ended up using a global statically allocated buffer for the purpose for now.
-Tero
[0] https://lore.kernel.org/linux-input/CAO-hwJJqP5iivQZOu0LTYa1D5OuM_aVi=LH27Ud...
Given that we can not call bpf_hid_raw_request() from within an IRQ, userspace needs to have a way to communicate with the device when it needs.
Implement a new type that the caller can run at will without being in an IRQ context.
Signed-off-by: Benjamin Tissoires benjamin.tissoires@redhat.com
---
changes in v2: - split the series by bpf/libbpf/hid/selftests and samples - unsigned long -> __u16 in uapi/linux/bpf_hid.h - int -> __32 in uapi/linux/bpf_hid.h --- include/linux/bpf-hid.h | 3 + include/uapi/linux/bpf.h | 1 + include/uapi/linux/bpf_hid.h | 10 +++ kernel/bpf/hid.c | 116 +++++++++++++++++++++++++++++++++ kernel/bpf/syscall.c | 2 + tools/include/uapi/linux/bpf.h | 1 + 6 files changed, 133 insertions(+)
diff --git a/include/linux/bpf-hid.h b/include/linux/bpf-hid.h index 69bb28523ceb..4cf2e99109fe 100644 --- a/include/linux/bpf-hid.h +++ b/include/linux/bpf-hid.h @@ -16,6 +16,7 @@ enum bpf_hid_attach_type { BPF_HID_ATTACH_INVALID = -1, BPF_HID_ATTACH_DEVICE_EVENT = 0, BPF_HID_ATTACH_RDESC_FIXUP, + BPF_HID_ATTACH_USER_EVENT, MAX_BPF_HID_ATTACH_TYPE };
@@ -35,6 +36,8 @@ to_bpf_hid_attach_type(enum bpf_attach_type attach_type) return BPF_HID_ATTACH_DEVICE_EVENT; case BPF_HID_RDESC_FIXUP: return BPF_HID_ATTACH_RDESC_FIXUP; + case BPF_HID_USER_EVENT: + return BPF_HID_ATTACH_USER_EVENT; default: return BPF_HID_ATTACH_INVALID; } diff --git a/include/uapi/linux/bpf.h b/include/uapi/linux/bpf.h index 4845a20e6f96..b3063384d380 100644 --- a/include/uapi/linux/bpf.h +++ b/include/uapi/linux/bpf.h @@ -1000,6 +1000,7 @@ enum bpf_attach_type { BPF_PERF_EVENT, BPF_HID_DEVICE_EVENT, BPF_HID_RDESC_FIXUP, + BPF_HID_USER_EVENT, __MAX_BPF_ATTACH_TYPE };
diff --git a/include/uapi/linux/bpf_hid.h b/include/uapi/linux/bpf_hid.h index 634f17c0b1cb..14a3c0405345 100644 --- a/include/uapi/linux/bpf_hid.h +++ b/include/uapi/linux/bpf_hid.h @@ -25,6 +25,12 @@ enum hid_bpf_event { HID_BPF_UNDEF = 0, HID_BPF_DEVICE_EVENT, /* when attach type is BPF_HID_DEVICE_EVENT */ HID_BPF_RDESC_FIXUP, /* ................... BPF_HID_RDESC_FIXUP */ + HID_BPF_USER_EVENT, /* ................... BPF_HID_USER_EVENT */ +}; + +/* type is HID_BPF_USER_EVENT */ +struct hid_bpf_ctx_user_event { + __s32 retval; };
struct hid_bpf_ctx { @@ -32,6 +38,10 @@ struct hid_bpf_ctx { __u16 allocated_size; /* the allocated size of data below (RO) */ struct hid_device *hdev; /* read-only */
+ union { + struct hid_bpf_ctx_user_event user; /* read-write */ + } u; + __u16 size; /* used size in data (RW) */ __u8 data[]; /* data buffer (RW) */ }; diff --git a/kernel/bpf/hid.c b/kernel/bpf/hid.c index 640e55ba66ec..de003dbd7d01 100644 --- a/kernel/bpf/hid.c +++ b/kernel/bpf/hid.c @@ -370,6 +370,8 @@ static int bpf_hid_max_progs(enum bpf_hid_attach_type type) return 64; case BPF_HID_ATTACH_RDESC_FIXUP: return 1; + case BPF_HID_ATTACH_USER_EVENT: + return 64; default: return 0; } @@ -464,7 +466,121 @@ int bpf_hid_link_create(const union bpf_attr *attr, struct bpf_prog *prog) return bpf_link_settle(&link_primer); }
+static int hid_bpf_prog_test_run(struct bpf_prog *prog, + const union bpf_attr *attr, + union bpf_attr __user *uattr) +{ + struct hid_device *hdev = NULL; + struct bpf_prog_array *progs; + struct hid_bpf_ctx *ctx = NULL; + bool valid_prog = false; + int i; + int target_fd, ret; + void __user *data_out = u64_to_user_ptr(attr->test.data_out); + void __user *data_in = u64_to_user_ptr(attr->test.data_in); + u32 user_size_in = attr->test.data_size_in; + u32 user_size_out = attr->test.data_size_out; + + if (!hid_hooks.hdev_from_fd) + return -EOPNOTSUPP; + + if (attr->test.ctx_size_in != sizeof(int)) + return -EINVAL; + + if (user_size_in > HID_BPF_MAX_BUFFER_SIZE) + return -E2BIG; + + if (copy_from_user(&target_fd, (void *)attr->test.ctx_in, attr->test.ctx_size_in)) + return -EFAULT; + + hdev = hid_hooks.hdev_from_fd(target_fd); + if (IS_ERR(hdev)) + return PTR_ERR(hdev); + + ret = mutex_lock_interruptible(&bpf_hid_mutex); + if (ret) + return ret; + + /* check if the given program is of correct type and registered */ + progs = rcu_dereference_protected(hdev->bpf.run_array[BPF_HID_ATTACH_USER_EVENT], + lockdep_is_held(&bpf_hid_mutex)); + if (!progs) { + ret = -EFAULT; + goto unlock; + } + + for (i = 0; i < bpf_prog_array_length(progs); i++) { + if (progs->items[i].prog == prog) { + valid_prog = true; + break; + } + } + + if (!valid_prog) { + ret = -EINVAL; + goto unlock; + } + + ctx = bpf_hid_allocate_ctx(hdev, max(user_size_in, user_size_out)); + if (IS_ERR(ctx)) { + ret = PTR_ERR(ctx); + goto unlock; + } + + ctx->type = HID_BPF_USER_EVENT; + + /* copy data_in from userspace */ + if (user_size_in) { + if (user_size_in > ctx->allocated_size) { + /* should never happen, given that size is < HID_BPF_MAX_BUFFER_SIZE */ + ret = -E2BIG; + goto unlock; + } + + if (copy_from_user(ctx->data, data_in, user_size_in)) { + ret = -EFAULT; + goto unlock; + } + + ctx->size = user_size_in; + } + + migrate_disable(); + + ret = bpf_prog_run(prog, ctx); + + migrate_enable(); + + if (user_size_out && data_out) { + user_size_out = min3(user_size_out, (u32)ctx->size, (u32)ctx->allocated_size); + + if (copy_to_user(data_out, ctx->data, user_size_out)) { + ret = -EFAULT; + goto unlock; + } + + if (copy_to_user(&uattr->test.data_size_out, + &user_size_out, + sizeof(user_size_out))) { + ret = -EFAULT; + goto unlock; + } + } + + if (copy_to_user(&uattr->test.retval, &ctx->u.user.retval, sizeof(ctx->u.user.retval))) { + ret = -EFAULT; + goto unlock; + } + +unlock: + kfree(ctx); + + mutex_unlock(&bpf_hid_mutex); + return ret; +} + const struct bpf_prog_ops hid_prog_ops = { + .test_run = hid_bpf_prog_test_run, };
int bpf_hid_init(struct hid_device *hdev) diff --git a/kernel/bpf/syscall.c b/kernel/bpf/syscall.c index 7428a1a512c6..74d13ec826df 100644 --- a/kernel/bpf/syscall.c +++ b/kernel/bpf/syscall.c @@ -3204,6 +3204,7 @@ attach_type_to_prog_type(enum bpf_attach_type attach_type) return BPF_PROG_TYPE_XDP; case BPF_HID_DEVICE_EVENT: case BPF_HID_RDESC_FIXUP: + case BPF_HID_USER_EVENT: return BPF_PROG_TYPE_HID; default: return BPF_PROG_TYPE_UNSPEC; @@ -3350,6 +3351,7 @@ static int bpf_prog_query(const union bpf_attr *attr, return sock_map_bpf_prog_query(attr, uattr); case BPF_HID_DEVICE_EVENT: case BPF_HID_RDESC_FIXUP: + case BPF_HID_USER_EVENT: return bpf_hid_prog_query(attr, uattr); default: return -EINVAL; diff --git a/tools/include/uapi/linux/bpf.h b/tools/include/uapi/linux/bpf.h index 4845a20e6f96..b3063384d380 100644 --- a/tools/include/uapi/linux/bpf.h +++ b/tools/include/uapi/linux/bpf.h @@ -1000,6 +1000,7 @@ enum bpf_attach_type { BPF_PERF_EVENT, BPF_HID_DEVICE_EVENT, BPF_HID_RDESC_FIXUP, + BPF_HID_USER_EVENT, __MAX_BPF_ATTACH_TYPE };
On Fri, Mar 4, 2022 at 9:33 AM Benjamin Tissoires benjamin.tissoires@redhat.com wrote:
Given that we can not call bpf_hid_raw_request() from within an IRQ, userspace needs to have a way to communicate with the device when it needs.
Implement a new type that the caller can run at will without being in an IRQ context.
Signed-off-by: Benjamin Tissoires benjamin.tissoires@redhat.com
[...]
if (user_size_out && data_out) {user_size_out = min3(user_size_out, (u32)ctx->size, (u32)ctx->allocated_size);if (copy_to_user(data_out, ctx->data, user_size_out)) {ret = -EFAULT;goto unlock;}if (copy_to_user(&uattr->test.data_size_out,&user_size_out,sizeof(user_size_out))) {ret = -EFAULT;goto unlock;}}if (copy_to_user(&uattr->test.retval, &ctx->u.user.retval, sizeof(ctx->u.user.retval))) {ret = -EFAULT;goto unlock;
nit: this goto is not really needed.
}+unlock:
kfree(ctx);mutex_unlock(&bpf_hid_mutex);return ret;+}
[...]
Adds the SEC definiton for the user callbacks of HID bpf.
Signed-off-by: Benjamin Tissoires benjamin.tissoires@redhat.com
---
changes in v2: - split the series by bpf/libbpf/hid/selftests and samples --- tools/lib/bpf/libbpf.c | 1 + 1 file changed, 1 insertion(+)
diff --git a/tools/lib/bpf/libbpf.c b/tools/lib/bpf/libbpf.c index 192ef3901251..d1e305f760bb 100644 --- a/tools/lib/bpf/libbpf.c +++ b/tools/lib/bpf/libbpf.c @@ -8682,6 +8682,7 @@ static const struct bpf_sec_def section_defs[] = { SEC_DEF("sk_lookup", SK_LOOKUP, BPF_SK_LOOKUP, SEC_ATTACHABLE | SEC_SLOPPY_PFX), SEC_DEF("hid/device_event", HID, BPF_HID_DEVICE_EVENT, SEC_ATTACHABLE_OPT | SEC_SLOPPY_PFX), SEC_DEF("hid/rdesc_fixup", HID, BPF_HID_RDESC_FIXUP, SEC_ATTACHABLE_OPT | SEC_SLOPPY_PFX), + SEC_DEF("hid/user_event", HID, BPF_HID_USER_EVENT, SEC_ATTACHABLE_OPT | SEC_SLOPPY_PFX), };
#define MAX_TYPE_NAME_SIZE 32
Add a simple test to see if we can trigger a bpf program of type "hid/user_event".
Signed-off-by: Benjamin Tissoires benjamin.tissoires@redhat.com
---
changes in v2: - split the series by bpf/libbpf/hid/selftests and samples --- tools/testing/selftests/bpf/prog_tests/hid.c | 56 ++++++++++++++++++++ tools/testing/selftests/bpf/progs/hid.c | 10 ++++ 2 files changed, 66 insertions(+)
diff --git a/tools/testing/selftests/bpf/prog_tests/hid.c b/tools/testing/selftests/bpf/prog_tests/hid.c index 74426523dd6f..eb4c0d0a4666 100644 --- a/tools/testing/selftests/bpf/prog_tests/hid.c +++ b/tools/testing/selftests/bpf/prog_tests/hid.c @@ -359,6 +359,59 @@ static int test_hid_set_get_data(struct hid *hid_skel, int uhid_fd, int sysfs_fd return ret; }
+/* + * Attach hid_user to the given uhid device, + * call the bpf program from userspace + * check that the program is called and does the expected. + */ +static int test_hid_user_call(struct hid *hid_skel, int uhid_fd, int sysfs_fd) +{ + int err, prog_fd; + u8 buf[10] = {0}; + int ret = -1; + + LIBBPF_OPTS(bpf_test_run_opts, run_attrs, + .repeat = 1, + .ctx_in = &sysfs_fd, + .ctx_size_in = sizeof(sysfs_fd), + .data_in = buf, + .data_size_in = sizeof(buf), + .data_out = buf, + .data_size_out = sizeof(buf), + ); + + /* attach hid_user program */ + hid_skel->links.hid_user = bpf_program__attach_hid(hid_skel->progs.hid_user, sysfs_fd); + if (!ASSERT_OK_PTR(hid_skel->links.hid_user, + "attach_hid(hid_user)")) + return PTR_ERR(hid_skel->links.hid_user); + + buf[0] = 39; + + prog_fd = bpf_program__fd(hid_skel->progs.hid_user); + + err = bpf_prog_test_run_opts(prog_fd, &run_attrs); + if (!ASSERT_EQ(err, 0, "bpf_prog_test_run_xattr")) + goto cleanup; + + if (!ASSERT_EQ(run_attrs.retval, 72, "bpf_prog_test_run_opts")) + goto cleanup; + + if (!ASSERT_EQ(buf[1], 42, "hid_user_check_in")) + goto cleanup; + + if (!ASSERT_EQ(buf[2], 4, "hid_user_check_static_out")) + goto cleanup; + + ret = 0; + +cleanup: + + hid__detach(hid_skel); + + return ret; +} + /* * Attach hid_rdesc_fixup to the given uhid device, * retrieve and open the matching hidraw node, @@ -460,6 +513,9 @@ void serial_test_hid_bpf(void) err = test_hid_set_get_data(hid_skel, uhid_fd, sysfs_fd); ASSERT_OK(err, "hid_set_get_data");
+ err = test_hid_user_call(hid_skel, uhid_fd, sysfs_fd); + ASSERT_OK(err, "hid_user"); + err = test_rdesc_fixup(hid_skel, uhid_fd, sysfs_fd); ASSERT_OK(err, "hid_rdesc_fixup");
diff --git a/tools/testing/selftests/bpf/progs/hid.c b/tools/testing/selftests/bpf/progs/hid.c index de6668471940..fabfaf0f2526 100644 --- a/tools/testing/selftests/bpf/progs/hid.c +++ b/tools/testing/selftests/bpf/progs/hid.c @@ -111,3 +111,13 @@ int hid_set_get_data(struct hid_bpf_ctx *ctx)
return ret; } + +SEC("hid/user_event") +int hid_user(struct hid_bpf_ctx *ctx) +{ + ctx->data[1] = ctx->data[0] + 3; + ctx->data[2] = 4; + ctx->u.user.retval = 72; + + return 0; +}
We need this for 2 reasons: - first we remove the ugly sleeps - then when we try to communicate with the device, we need to have another thread that handles that communication and simulate a real device
Signed-off-by: Benjamin Tissoires benjamin.tissoires@redhat.com
---
changes in v2: - split the series by bpf/libbpf/hid/selftests and samples --- tools/testing/selftests/bpf/prog_tests/hid.c | 126 ++++++++++++++++++- 1 file changed, 120 insertions(+), 6 deletions(-)
diff --git a/tools/testing/selftests/bpf/prog_tests/hid.c b/tools/testing/selftests/bpf/prog_tests/hid.c index eb4c0d0a4666..cc01601c1168 100644 --- a/tools/testing/selftests/bpf/prog_tests/hid.c +++ b/tools/testing/selftests/bpf/prog_tests/hid.c @@ -67,6 +67,12 @@ static unsigned char rdesc[] = { 0xc0, /* END_COLLECTION */ };
+static pthread_mutex_t uhid_started_mtx = PTHREAD_MUTEX_INITIALIZER; +static pthread_cond_t uhid_started = PTHREAD_COND_INITIALIZER; + +/* no need to protect uhid_stopped, only one thread accesses it */ +static bool uhid_stopped; + static int uhid_write(int fd, const struct uhid_event *ev) { ssize_t ret; @@ -118,6 +124,104 @@ static void destroy(int fd) uhid_write(fd, &ev); }
+static int event(int fd) +{ + struct uhid_event ev; + ssize_t ret; + + memset(&ev, 0, sizeof(ev)); + ret = read(fd, &ev, sizeof(ev)); + if (ret == 0) { + fprintf(stderr, "Read HUP on uhid-cdev\n"); + return -EFAULT; + } else if (ret < 0) { + fprintf(stderr, "Cannot read uhid-cdev: %m\n"); + return -errno; + } else if (ret != sizeof(ev)) { + fprintf(stderr, "Invalid size read from uhid-dev: %zd != %zu\n", + ret, sizeof(ev)); + return -EFAULT; + } + + switch (ev.type) { + case UHID_START: + pthread_mutex_lock(&uhid_started_mtx); + pthread_cond_signal(&uhid_started); + pthread_mutex_unlock(&uhid_started_mtx); + + fprintf(stderr, "UHID_START from uhid-dev\n"); + break; + case UHID_STOP: + uhid_stopped = true; + + fprintf(stderr, "UHID_STOP from uhid-dev\n"); + break; + case UHID_OPEN: + fprintf(stderr, "UHID_OPEN from uhid-dev\n"); + break; + case UHID_CLOSE: + fprintf(stderr, "UHID_CLOSE from uhid-dev\n"); + break; + case UHID_OUTPUT: + fprintf(stderr, "UHID_OUTPUT from uhid-dev\n"); + break; + case UHID_GET_REPORT: + fprintf(stderr, "UHID_GET_REPORT from uhid-dev\n"); + break; + case UHID_SET_REPORT: + fprintf(stderr, "UHID_SET_REPORT from uhid-dev\n"); + break; + default: + fprintf(stderr, "Invalid event from uhid-dev: %u\n", ev.type); + } + + return 0; +} + +static void *read_uhid_events_thread(void *arg) +{ + int fd = *(int *)arg; + struct pollfd pfds[1]; + int ret = 0; + + pfds[0].fd = fd; + pfds[0].events = POLLIN; + + uhid_stopped = false; + + while (!uhid_stopped) { + ret = poll(pfds, 1, 100); + if (ret < 0) { + fprintf(stderr, "Cannot poll for fds: %m\n"); + break; + } + if (pfds[0].revents & POLLIN) { + ret = event(fd); + if (ret) + break; + } + } + + return (void *)(long)ret; +} + +static int uhid_start_listener(pthread_t *tid, int uhid_fd) +{ + int fd = uhid_fd; + + pthread_mutex_lock(&uhid_started_mtx); + if (CHECK_FAIL(pthread_create(tid, NULL, read_uhid_events_thread, + (void *)&fd))) { + pthread_mutex_unlock(&uhid_started_mtx); + close(fd); + return -EIO; + } + pthread_cond_wait(&uhid_started, &uhid_started_mtx); + pthread_mutex_unlock(&uhid_started_mtx); + + return 0; +} + static int send_event(int fd, u8 *buf, size_t size) { struct uhid_event ev; @@ -422,7 +526,9 @@ static int test_rdesc_fixup(struct hid *hid_skel, int uhid_fd, int sysfs_fd) struct hidraw_report_descriptor rpt_desc = {0}; int err, desc_size, hidraw_ino, hidraw_fd = -1; char hidraw_path[64] = {0}; + void *uhid_err; int ret = -1; + pthread_t tid;
/* attach the program */ hid_skel->links.hid_rdesc_fixup = @@ -431,9 +537,8 @@ static int test_rdesc_fixup(struct hid *hid_skel, int uhid_fd, int sysfs_fd) "attach_hid(hid_rdesc_fixup)")) return PTR_ERR(hid_skel->links.hid_rdesc_fixup);
- /* give a little bit of time for the device to appear */ - /* TODO: check on uhid events */ - usleep(1000); + err = uhid_start_listener(&tid, uhid_fd); + ASSERT_OK(err, "uhid_start_listener");
hidraw_ino = get_hidraw(hid_skel->links.hid_rdesc_fixup); if (!ASSERT_GE(hidraw_ino, 0, "get_hidraw")) @@ -474,6 +579,10 @@ static int test_rdesc_fixup(struct hid *hid_skel, int uhid_fd, int sysfs_fd)
hid__detach(hid_skel);
+ pthread_join(tid, &uhid_err); + err = (int)(long)uhid_err; + CHECK_FAIL(err); + return ret; }
@@ -481,7 +590,9 @@ void serial_test_hid_bpf(void) { struct hid *hid_skel = NULL; int err, uhid_fd, sysfs_fd; + void *uhid_err; time_t t; + pthread_t tid; int rand_nb;
/* initialize random number generator */ @@ -493,9 +604,8 @@ void serial_test_hid_bpf(void) if (!ASSERT_GE(uhid_fd, 0, "setup uhid")) return;
- /* give a little bit of time for the device to appear */ - /* TODO: check on uhid events */ - usleep(1000); + err = uhid_start_listener(&tid, uhid_fd); + ASSERT_OK(err, "uhid_start_listener");
/* locate the uevent file of the created device */ sysfs_fd = get_sysfs_fd(rand_nb); @@ -522,4 +632,8 @@ void serial_test_hid_bpf(void) cleanup: hid__destroy(hid_skel); destroy(uhid_fd); + + pthread_join(tid, &uhid_err); + err = (int)(long)uhid_err; + CHECK_FAIL(err); }
When we are in a user_event context, we can talk to the device to fetch or set features/outputs/inputs reports. Add a bpf helper to do so. This helper is thus only available to user_events, because calling this function while in IRQ context (any other BPF type) is forbidden.
Signed-off-by: Benjamin Tissoires benjamin.tissoires@redhat.com
---
changes in v2: - split the series by bpf/libbpf/hid/selftests and samples --- include/linux/bpf-hid.h | 2 ++ include/uapi/linux/bpf.h | 8 ++++++++ kernel/bpf/hid.c | 26 ++++++++++++++++++++++++++ tools/include/uapi/linux/bpf.h | 8 ++++++++ 4 files changed, 44 insertions(+)
diff --git a/include/linux/bpf-hid.h b/include/linux/bpf-hid.h index 4cf2e99109fe..bd548f6a4a26 100644 --- a/include/linux/bpf-hid.h +++ b/include/linux/bpf-hid.h @@ -100,6 +100,8 @@ struct bpf_hid_hooks { u64 offset, u32 n, u8 *data, u64 data_size); int (*hid_set_data)(struct hid_device *hdev, u8 *buf, size_t buf_size, u64 offset, u32 n, u8 *data, u64 data_size); + int (*hid_raw_request)(struct hid_device *hdev, u8 *buf, size_t size, + u8 rtype, u8 reqtype); };
#ifdef CONFIG_BPF diff --git a/include/uapi/linux/bpf.h b/include/uapi/linux/bpf.h index b3063384d380..417cf1c31579 100644 --- a/include/uapi/linux/bpf.h +++ b/include/uapi/linux/bpf.h @@ -5121,6 +5121,13 @@ union bpf_attr { * Return * The length of data copied into ctx->event.data. On error, a negative * value is returned. + * + * int bpf_hid_raw_request(void *ctx, void *buf, u64 size, u8 rtype, u8 reqtype) + * Description + * communicate with the HID device + * Return + * 0 on success. + * negative value on error. */ #define __BPF_FUNC_MAPPER(FN) \ FN(unspec), \ @@ -5317,6 +5324,7 @@ union bpf_attr { FN(copy_from_user_task), \ FN(hid_get_data), \ FN(hid_set_data), \ + FN(hid_raw_request), \ /* */
/* integer value in 'imm' field of BPF_CALL instruction selects which helper diff --git a/kernel/bpf/hid.c b/kernel/bpf/hid.c index de003dbd7d01..653d10c0f4e6 100644 --- a/kernel/bpf/hid.c +++ b/kernel/bpf/hid.c @@ -86,6 +86,28 @@ static const struct bpf_func_proto bpf_hid_set_data_proto = { .arg5_type = ARG_CONST_SIZE_OR_ZERO, };
+BPF_CALL_5(bpf_hid_raw_request, void*, ctx, void*, buf, u64, size, + u8, rtype, u8, reqtype) +{ + struct hid_bpf_ctx *bpf_ctx = ctx; + + if (!hid_hooks.hid_raw_request) + return -EOPNOTSUPP; + + return hid_hooks.hid_raw_request(bpf_ctx->hdev, buf, size, rtype, reqtype); +} + +static const struct bpf_func_proto bpf_hid_raw_request_proto = { + .func = bpf_hid_raw_request, + .gpl_only = true, /* hid_raw_request is EXPORT_SYMBOL_GPL */ + .ret_type = RET_INTEGER, + .arg1_type = ARG_PTR_TO_CTX, + .arg2_type = ARG_PTR_TO_MEM, + .arg3_type = ARG_CONST_SIZE_OR_ZERO, + .arg4_type = ARG_ANYTHING, + .arg5_type = ARG_ANYTHING, +}; + static const struct bpf_func_proto * hid_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) { @@ -94,6 +116,10 @@ hid_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) return &bpf_hid_get_data_proto; case BPF_FUNC_hid_set_data: return &bpf_hid_set_data_proto; + case BPF_FUNC_hid_raw_request: + if (prog->expected_attach_type != BPF_HID_DEVICE_EVENT) + return &bpf_hid_raw_request_proto; + return NULL; default: return bpf_base_func_proto(func_id); } diff --git a/tools/include/uapi/linux/bpf.h b/tools/include/uapi/linux/bpf.h index b3063384d380..417cf1c31579 100644 --- a/tools/include/uapi/linux/bpf.h +++ b/tools/include/uapi/linux/bpf.h @@ -5121,6 +5121,13 @@ union bpf_attr { * Return * The length of data copied into ctx->event.data. On error, a negative * value is returned. + * + * int bpf_hid_raw_request(void *ctx, void *buf, u64 size, u8 rtype, u8 reqtype) + * Description + * communicate with the HID device + * Return + * 0 on success. + * negative value on error. */ #define __BPF_FUNC_MAPPER(FN) \ FN(unspec), \ @@ -5317,6 +5324,7 @@ union bpf_attr { FN(copy_from_user_task), \ FN(hid_get_data), \ FN(hid_set_data), \ + FN(hid_raw_request), \ /* */
/* integer value in 'imm' field of BPF_CALL instruction selects which helper
On Fri, Mar 4, 2022 at 9:35 AM Benjamin Tissoires benjamin.tissoires@redhat.com wrote:
When we are in a user_event context, we can talk to the device to fetch or set features/outputs/inputs reports. Add a bpf helper to do so. This helper is thus only available to user_events, because calling this function while in IRQ context (any other BPF type) is forbidden.
Signed-off-by: Benjamin Tissoires benjamin.tissoires@redhat.com
changes in v2:
- split the series by bpf/libbpf/hid/selftests and samples
include/linux/bpf-hid.h | 2 ++ include/uapi/linux/bpf.h | 8 ++++++++ kernel/bpf/hid.c | 26 ++++++++++++++++++++++++++ tools/include/uapi/linux/bpf.h | 8 ++++++++ 4 files changed, 44 insertions(+)
diff --git a/include/linux/bpf-hid.h b/include/linux/bpf-hid.h index 4cf2e99109fe..bd548f6a4a26 100644 --- a/include/linux/bpf-hid.h +++ b/include/linux/bpf-hid.h @@ -100,6 +100,8 @@ struct bpf_hid_hooks { u64 offset, u32 n, u8 *data, u64 data_size); int (*hid_set_data)(struct hid_device *hdev, u8 *buf, size_t buf_size, u64 offset, u32 n, u8 *data, u64 data_size);
int (*hid_raw_request)(struct hid_device *hdev, u8 *buf, size_t size,u8 rtype, u8 reqtype);};
#ifdef CONFIG_BPF diff --git a/include/uapi/linux/bpf.h b/include/uapi/linux/bpf.h index b3063384d380..417cf1c31579 100644 --- a/include/uapi/linux/bpf.h +++ b/include/uapi/linux/bpf.h @@ -5121,6 +5121,13 @@ union bpf_attr {
ReturnThe length of data copied into ctx->event.data. On error, a negativevalue is returned.
- int bpf_hid_raw_request(void *ctx, void *buf, u64 size, u8 rtype, u8 reqtype)
Description
communicate with the HID device
I think we need more description here, e.g. what are rtype and reqtype here?
Return
0 on success.*/
negative value on error.#define __BPF_FUNC_MAPPER(FN) \ FN(unspec), \ @@ -5317,6 +5324,7 @@ union bpf_attr { FN(copy_from_user_task), \ FN(hid_get_data), \ FN(hid_set_data), \
FN(hid_raw_request), \ /* */
[...]
Hook up BPF to hid_hw_raw_request. Not much to report here except that we need to export hid_get_report from hid-core so it gets available in hid-bpf.c
Signed-off-by: Benjamin Tissoires benjamin.tissoires@redhat.com
---
changes in v2: - split the series by bpf/libbpf/hid/selftests and samples --- drivers/hid/hid-bpf.c | 63 ++++++++++++++++++++++++++++++++++++++++++ drivers/hid/hid-core.c | 3 +- include/linux/hid.h | 1 + 3 files changed, 65 insertions(+), 2 deletions(-)
diff --git a/drivers/hid/hid-bpf.c b/drivers/hid/hid-bpf.c index 8ae247fba5bc..b8c0060f3180 100644 --- a/drivers/hid/hid-bpf.c +++ b/drivers/hid/hid-bpf.c @@ -171,6 +171,68 @@ int hid_bpf_set_data(struct hid_device *hdev, u8 *buf, size_t buf_size, u64 offs return n; }
+int hid_bpf_raw_request(struct hid_device *hdev, u8 *buf, size_t size, + u8 rtype, u8 reqtype) +{ + struct hid_report *report; + struct hid_report_enum *report_enum; + u8 *dma_data; + u32 report_len; + int ret; + + /* check arguments */ + switch (rtype) { + case HID_INPUT_REPORT: + case HID_OUTPUT_REPORT: + case HID_FEATURE_REPORT: + break; + default: + return -EINVAL; + } + + switch (reqtype) { + case HID_REQ_GET_REPORT: + case HID_REQ_GET_IDLE: + case HID_REQ_GET_PROTOCOL: + case HID_REQ_SET_REPORT: + case HID_REQ_SET_IDLE: + case HID_REQ_SET_PROTOCOL: + break; + default: + return -EINVAL; + } + + if (size < 1) + return -EINVAL; + + report_enum = hdev->report_enum + rtype; + report = hid_get_report(report_enum, buf); + if (!report) + return -EINVAL; + + report_len = hid_report_len(report); + + if (size > report_len) + size = report_len; + + dma_data = kmemdup(buf, size, GFP_KERNEL); + if (!dma_data) + return -ENOMEM; + + ret = hid_hw_raw_request(hdev, + dma_data[0], + dma_data, + size, + rtype, + reqtype); + + if (ret > 0) + memcpy(buf, dma_data, ret); + + kfree(dma_data); + return ret; +} + static int hid_bpf_run_progs(struct hid_device *hdev, enum bpf_hid_attach_type type, struct hid_bpf_ctx *ctx, u8 *data, int size) { @@ -272,6 +334,7 @@ int __init hid_bpf_module_init(void) .array_detached = hid_bpf_array_detached, .hid_get_data = hid_bpf_get_data, .hid_set_data = hid_bpf_set_data, + .hid_raw_request = hid_bpf_raw_request, };
bpf_hid_set_hooks(&hooks); diff --git a/drivers/hid/hid-core.c b/drivers/hid/hid-core.c index d3f4499ee4cd..d0e015986e17 100644 --- a/drivers/hid/hid-core.c +++ b/drivers/hid/hid-core.c @@ -1686,8 +1686,7 @@ int hid_set_field(struct hid_field *field, unsigned offset, __s32 value) } EXPORT_SYMBOL_GPL(hid_set_field);
-static struct hid_report *hid_get_report(struct hid_report_enum *report_enum, - const u8 *data) +struct hid_report *hid_get_report(struct hid_report_enum *report_enum, const u8 *data) { struct hid_report *report; unsigned int n = 0; /* Normally report number is 0 */ diff --git a/include/linux/hid.h b/include/linux/hid.h index 7454e844324c..b2698df31e5b 100644 --- a/include/linux/hid.h +++ b/include/linux/hid.h @@ -946,6 +946,7 @@ __u32 hid_field_extract(const struct hid_device *hid, __u8 *report, unsigned offset, unsigned n); void implement(const struct hid_device *hid, u8 *report, unsigned int offset, unsigned int n, u32 value); +struct hid_report *hid_get_report(struct hid_report_enum *report_enum, const u8 *data);
#ifdef CONFIG_PM int hid_driver_suspend(struct hid_device *hdev, pm_message_t state);
Add tests for the newly implemented function. We test here only the GET_REPORT part because the other calls are pure HID protocol and won't infer the result of the test of the bpf hook.
Signed-off-by: Benjamin Tissoires benjamin.tissoires@redhat.com
---
changes in v2: - split the series by bpf/libbpf/hid/selftests and samples --- tools/testing/selftests/bpf/prog_tests/hid.c | 71 +++++++++++++++++++- tools/testing/selftests/bpf/progs/hid.c | 57 ++++++++++++++++ 2 files changed, 127 insertions(+), 1 deletion(-)
diff --git a/tools/testing/selftests/bpf/prog_tests/hid.c b/tools/testing/selftests/bpf/prog_tests/hid.c index cc01601c1168..a2bab6a799a1 100644 --- a/tools/testing/selftests/bpf/prog_tests/hid.c +++ b/tools/testing/selftests/bpf/prog_tests/hid.c @@ -67,6 +67,8 @@ static unsigned char rdesc[] = { 0xc0, /* END_COLLECTION */ };
+static u8 feature_data[] = { 1, 2 }; + static pthread_mutex_t uhid_started_mtx = PTHREAD_MUTEX_INITIALIZER; static pthread_cond_t uhid_started = PTHREAD_COND_INITIALIZER;
@@ -126,7 +128,7 @@ static void destroy(int fd)
static int event(int fd) { - struct uhid_event ev; + struct uhid_event ev, answer; ssize_t ret;
memset(&ev, 0, sizeof(ev)); @@ -143,6 +145,8 @@ static int event(int fd) return -EFAULT; }
+ memset(&answer, 0, sizeof(answer)); + switch (ev.type) { case UHID_START: pthread_mutex_lock(&uhid_started_mtx); @@ -167,6 +171,15 @@ static int event(int fd) break; case UHID_GET_REPORT: fprintf(stderr, "UHID_GET_REPORT from uhid-dev\n"); + + answer.type = UHID_GET_REPORT_REPLY; + answer.u.get_report_reply.id = ev.u.get_report.id; + answer.u.get_report_reply.err = ev.u.get_report.rnum == 1 ? 0 : -EIO; + answer.u.get_report_reply.size = sizeof(feature_data); + memcpy(answer.u.get_report_reply.data, feature_data, sizeof(feature_data)); + + uhid_write(fd, &answer); + break; case UHID_SET_REPORT: fprintf(stderr, "UHID_SET_REPORT from uhid-dev\n"); @@ -516,6 +529,59 @@ static int test_hid_user_call(struct hid *hid_skel, int uhid_fd, int sysfs_fd) return ret; }
+/* + * Attach hid_user_raw_request to the given uhid device, + * call the bpf program from userspace + * check that the program is called and does the expected. + */ +static int test_hid_user_raw_request_call(struct hid *hid_skel, int uhid_fd, int sysfs_fd) +{ + int err, prog_fd; + u8 buf[10] = {0}; + int ret = -1; + + LIBBPF_OPTS(bpf_test_run_opts, run_attrs, + .repeat = 1, + .ctx_in = &sysfs_fd, + .ctx_size_in = sizeof(sysfs_fd), + .data_in = buf, + .data_size_in = sizeof(buf), + .data_out = buf, + .data_size_out = sizeof(buf), + ); + + /* attach hid_user_raw_request program */ + hid_skel->links.hid_user_raw_request = + bpf_program__attach_hid(hid_skel->progs.hid_user_raw_request, sysfs_fd); + if (!ASSERT_OK_PTR(hid_skel->links.hid_user_raw_request, + "attach_hid(hid_user_raw_request)")) + return PTR_ERR(hid_skel->links.hid_user_raw_request); + + buf[0] = 2; /* HID_FEATURE_REPORT */ + buf[1] = 1; /* HID_REQ_GET_REPORT */ + buf[2] = 1; /* report ID */ + + prog_fd = bpf_program__fd(hid_skel->progs.hid_user_raw_request); + + err = bpf_prog_test_run_opts(prog_fd, &run_attrs); + if (!ASSERT_EQ(err, 0, "bpf_prog_test_run_xattr")) + goto cleanup; + + if (!ASSERT_EQ(run_attrs.retval, 2, "bpf_prog_test_run_xattr_retval")) + goto cleanup; + + if (!ASSERT_EQ(buf[3], 2, "hid_user_raw_request_check_in")) + goto cleanup; + + ret = 0; + +cleanup: + + hid__detach(hid_skel); + + return ret; +} + /* * Attach hid_rdesc_fixup to the given uhid device, * retrieve and open the matching hidraw node, @@ -626,6 +692,9 @@ void serial_test_hid_bpf(void) err = test_hid_user_call(hid_skel, uhid_fd, sysfs_fd); ASSERT_OK(err, "hid_user");
+ err = test_hid_user_raw_request_call(hid_skel, uhid_fd, sysfs_fd); + ASSERT_OK(err, "hid_user_raw_request"); + err = test_rdesc_fixup(hid_skel, uhid_fd, sysfs_fd); ASSERT_OK(err, "hid_rdesc_fixup");
diff --git a/tools/testing/selftests/bpf/progs/hid.c b/tools/testing/selftests/bpf/progs/hid.c index fabfaf0f2526..ea011ff9e752 100644 --- a/tools/testing/selftests/bpf/progs/hid.c +++ b/tools/testing/selftests/bpf/progs/hid.c @@ -9,6 +9,11 @@ char _license[] SEC("license") = "GPL"; __u64 callback_check = 52; __u64 callback2_check = 52;
+struct { + __uint(type, BPF_MAP_TYPE_RINGBUF); + __uint(max_entries, 4096 * 64); +} ringbuf SEC(".maps"); + SEC("hid/device_event") int hid_first_event(struct hid_bpf_ctx *ctx) { @@ -121,3 +126,55 @@ int hid_user(struct hid_bpf_ctx *ctx)
return 0; } + +SEC("hid/user_event") +int hid_user_raw_request(struct hid_bpf_ctx *ctx) +{ + const unsigned int buflen = 256; + const unsigned int _buflen = buflen * sizeof(__u8); + __u8 *buf; + int ret; + __u32 size; + __u8 rtype, reqtype; + + buf = bpf_ringbuf_reserve(&ringbuf, _buflen, 0); + if (!buf) + return -12; /* -ENOMEM */ + + __builtin_memcpy(buf, ctx->data, _buflen); + + /* + * build up a custom API for our needs: + * offset 0, size 1: report type + * offset 1, size 1: request type + * offset 2+: data + */ + rtype = buf[0]; + reqtype = buf[1]; + size = ctx->size - 2; + + if (size < _buflen - 2) { + ret = bpf_hid_raw_request(ctx, + &buf[2], + size, + rtype, + reqtype); + if (ret < 0) + goto discard; + } else { + ret = -7; /* -E2BIG */ + goto discard; + } + + __builtin_memcpy(&ctx->data[2], &buf[2], _buflen - 2); + + ctx->size = ret + 2; + ctx->u.user.retval = ret; + + ret = 0; + + discard: + bpf_ringbuf_discard(buf, 0); + + return ret; +}
This is just to hammer the obvious because I suspect you can not already load a bpf HID program which is not GPL because all of the useful functions are GPL only.
Anyway, this ensures that users are not tempted to bypass this requirement and will allow us to ship tested BPF programs in the kernel without having to aorry about the license.
Signed-off-by: Benjamin Tissoires benjamin.tissoires@redhat.com
---
new in v2: - Note: I placed this statement in check_attach_btf_id() to be local to other similar checks (regarding LSM), however, I have no idea if this is the correct place. Please shout at me if it isn't. --- include/linux/bpf-hid.h | 8 ++++++++ kernel/bpf/hid.c | 12 ++++++++++++ kernel/bpf/verifier.c | 7 +++++++ 3 files changed, 27 insertions(+)
diff --git a/include/linux/bpf-hid.h b/include/linux/bpf-hid.h index bd548f6a4a26..3da1d0ecb9be 100644 --- a/include/linux/bpf-hid.h +++ b/include/linux/bpf-hid.h @@ -2,6 +2,7 @@ #ifndef _BPF_HID_H #define _BPF_HID_H
+#include <linux/bpf_verifier.h> #include <linux/mutex.h> #include <uapi/linux/bpf.h> #include <uapi/linux/bpf_hid.h> @@ -71,6 +72,8 @@ int bpf_hid_prog_query(const union bpf_attr *attr, union bpf_attr __user *uattr); int bpf_hid_link_create(const union bpf_attr *attr, struct bpf_prog *prog); +int bpf_hid_verify_prog(struct bpf_verifier_log *vlog, + const struct bpf_prog *prog); #else static inline int bpf_hid_prog_query(const union bpf_attr *attr, union bpf_attr __user *uattr) @@ -83,6 +86,11 @@ static inline int bpf_hid_link_create(const union bpf_attr *attr, { return -EOPNOTSUPP; } +static inline int bpf_hid_verify_prog(struct bpf_verifier_log *vlog, + const struct bpf_prog *prog) +{ + return -EOPNOTSUPP; +} #endif
static inline bool bpf_hid_link_empty(struct bpf_hid *bpf, diff --git a/kernel/bpf/hid.c b/kernel/bpf/hid.c index 653d10c0f4e6..b3dc1cd37a3e 100644 --- a/kernel/bpf/hid.c +++ b/kernel/bpf/hid.c @@ -37,6 +37,18 @@ void bpf_hid_set_hooks(struct bpf_hid_hooks *hooks) } EXPORT_SYMBOL_GPL(bpf_hid_set_hooks);
+int bpf_hid_verify_prog(struct bpf_verifier_log *vlog, + const struct bpf_prog *prog) +{ + if (!prog->gpl_compatible) { + bpf_log(vlog, + "HID programs must have a GPL compatible license\n"); + return -EINVAL; + } + + return 0; +} + BPF_CALL_5(bpf_hid_get_data, void*, ctx, u64, offset, u32, n, void*, data, u64, size) { struct hid_bpf_ctx *bpf_ctx = ctx; diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c index a57db4b2803c..afec8fa1d674 100644 --- a/kernel/bpf/verifier.c +++ b/kernel/bpf/verifier.c @@ -21,6 +21,7 @@ #include <linux/perf_event.h> #include <linux/ctype.h> #include <linux/error-injection.h> +#include <linux/bpf-hid.h> #include <linux/bpf_lsm.h> #include <linux/btf_ids.h>
@@ -14235,6 +14236,12 @@ static int check_attach_btf_id(struct bpf_verifier_env *env) if (prog->type == BPF_PROG_TYPE_STRUCT_OPS) return check_struct_ops_btf_id(env);
+ if (prog->type == BPF_PROG_TYPE_HID) { + ret = bpf_hid_verify_prog(&env->log, prog); + if (ret < 0) + return ret; + } + if (prog->type != BPF_PROG_TYPE_TRACING && prog->type != BPF_PROG_TYPE_LSM && prog->type != BPF_PROG_TYPE_EXT)
There is no point in using 16 kB of memory if the device needs less for all of its reports (uwhich is usually the case).
Signed-off-by: Benjamin Tissoires benjamin.tissoires@redhat.com
---
new in v2 --- drivers/hid/hid-bpf.c | 15 ++++++++++++++- 1 file changed, 14 insertions(+), 1 deletion(-)
diff --git a/drivers/hid/hid-bpf.c b/drivers/hid/hid-bpf.c index b8c0060f3180..d56fbad990ed 100644 --- a/drivers/hid/hid-bpf.c +++ b/drivers/hid/hid-bpf.c @@ -61,11 +61,24 @@ static int hid_reconnect(struct hid_device *hdev) static int hid_bpf_link_attach(struct hid_device *hdev, enum bpf_hid_attach_type type) { int err = 0; + unsigned int i, j, max_report_len = 0; + + /* compute the maximum report length for this device */ + for (i = 0; i < HID_REPORT_TYPES; i++) { + struct hid_report_enum *report_enum = hdev->report_enum + i; + + for (j = 0; j < HID_MAX_IDS; j++) { + struct hid_report *report = report_enum->report_id_hash[j]; + + if (report) + max_report_len = max(max_report_len, hid_report_len(report)); + } + }
switch (type) { case BPF_HID_ATTACH_DEVICE_EVENT: if (!hdev->bpf.ctx) { - hdev->bpf.ctx = bpf_hid_allocate_ctx(hdev, HID_BPF_MAX_BUFFER_SIZE); + hdev->bpf.ctx = bpf_hid_allocate_ctx(hdev, max_report_len); if (IS_ERR(hdev->bpf.ctx)) { err = PTR_ERR(hdev->bpf.ctx); hdev->bpf.ctx = NULL;
the context is allocated the first time a program of type DEVICE_EVENT is attached to the device. To not add too much jumps in the code for the general device handling, call hid_bpf_raw_event() only if we know that a program has been attached once during the life of the device.
Signed-off-by: Benjamin Tissoires benjamin.tissoires@redhat.com
---
new in v2 --- drivers/hid/hid-core.c | 11 +++++++---- 1 file changed, 7 insertions(+), 4 deletions(-)
diff --git a/drivers/hid/hid-core.c b/drivers/hid/hid-core.c index d0e015986e17..2b49f6064a40 100644 --- a/drivers/hid/hid-core.c +++ b/drivers/hid/hid-core.c @@ -1751,10 +1751,13 @@ int hid_report_raw_event(struct hid_device *hid, int type, u8 *data, u32 size, u8 *cdata; int ret = 0;
- data = hid_bpf_raw_event(hid, data, &size); - if (IS_ERR(data)) { - ret = PTR_ERR(data); - goto out; + /* we pre-test if ctx is available here to cut the calls at the earliest */ + if (hid->bpf.ctx) { + data = hid_bpf_raw_event(hid, data, &size); + if (IS_ERR(data)) { + ret = PTR_ERR(data); + goto out; + } }
report = hid_get_report(report_enum, data);
When tracing the incoming events, if a bpf program is already loaded, the next bpf program will see the potentially changed data.
Add a flag to BPF_LINK_CREATE that allows to chose the position of the inserted program: at the beginning or at the end.
This way, we can have a tracing program that compare the raw event from the device and the transformed stream from all the other bpf programs.
Signed-off-by: Benjamin Tissoires benjamin.tissoires@redhat.com
---
new in v2 --- include/uapi/linux/bpf.h | 10 ++++++++++ kernel/bpf/hid.c | 11 +++++++---- 2 files changed, 17 insertions(+), 4 deletions(-)
diff --git a/include/uapi/linux/bpf.h b/include/uapi/linux/bpf.h index 417cf1c31579..23ebe5e96d69 100644 --- a/include/uapi/linux/bpf.h +++ b/include/uapi/linux/bpf.h @@ -1123,6 +1123,16 @@ enum bpf_link_type { */ #define BPF_F_XDP_HAS_FRAGS (1U << 5)
+/* HID flag used in BPF_LINK_CREATE command + * + * NONE(default): The bpf program will be added at the tail of the list + * of existing bpf program for this type. + * + * BPF_F_INSERT_HEAD: The bpf program will be added at the beginning + * of the list of existing bpf program for this type.. + */ +#define BPF_F_INSERT_HEAD (1U << 0) + /* When BPF ldimm64's insn[0].src_reg != 0 then this can have * the following extensions: * diff --git a/kernel/bpf/hid.c b/kernel/bpf/hid.c index b3dc1cd37a3e..141eb4169079 100644 --- a/kernel/bpf/hid.c +++ b/kernel/bpf/hid.c @@ -416,7 +416,7 @@ static int bpf_hid_max_progs(enum bpf_hid_attach_type type) }
static int bpf_hid_link_attach(struct hid_device *hdev, struct bpf_link *link, - enum bpf_hid_attach_type type) + enum bpf_hid_attach_type type, u32 flags) { struct bpf_hid_link *hid_link = container_of(link, struct bpf_hid_link, link); @@ -443,7 +443,10 @@ static int bpf_hid_link_attach(struct hid_device *hdev, struct bpf_link *link, goto out_unlock; }
- list_add_tail(&hid_link->node, &hdev->bpf.links[type]); + if (flags & BPF_F_INSERT_HEAD) + list_add(&hid_link->node, &hdev->bpf.links[type]); + else + list_add_tail(&hid_link->node, &hdev->bpf.links[type]);
fill_prog_array(hdev, type, run_array); run_array = rcu_replace_pointer(hdev->bpf.run_array[type], run_array, @@ -467,7 +470,7 @@ int bpf_hid_link_create(const union bpf_attr *attr, struct bpf_prog *prog) struct hid_device *hdev; int err;
- if (attr->link_create.flags || !hid_hooks.hdev_from_fd) + if ((attr->link_create.flags & ~BPF_F_INSERT_HEAD) || !hid_hooks.hdev_from_fd) return -EINVAL;
type = attr->link_create.attach_type; @@ -495,7 +498,7 @@ int bpf_hid_link_create(const union bpf_attr *attr, struct bpf_prog *prog) return err; }
- err = bpf_hid_link_attach(hdev, &hid_link->link, hid_type); + err = bpf_hid_link_attach(hdev, &hid_link->link, hid_type, attr->link_create.flags); if (err) { bpf_link_cleanup(&link_primer); return err;
Export the newly created flag to libbpf.
Signed-off-by: Benjamin Tissoires benjamin.tissoires@redhat.com
---
new in v2 --- tools/include/uapi/linux/bpf.h | 10 ++++++++++ tools/lib/bpf/libbpf.c | 17 +++++++++-------- tools/lib/bpf/libbpf.h | 2 +- 3 files changed, 20 insertions(+), 9 deletions(-)
diff --git a/tools/include/uapi/linux/bpf.h b/tools/include/uapi/linux/bpf.h index 417cf1c31579..23ebe5e96d69 100644 --- a/tools/include/uapi/linux/bpf.h +++ b/tools/include/uapi/linux/bpf.h @@ -1123,6 +1123,16 @@ enum bpf_link_type { */ #define BPF_F_XDP_HAS_FRAGS (1U << 5)
+/* HID flag used in BPF_LINK_CREATE command + * + * NONE(default): The bpf program will be added at the tail of the list + * of existing bpf program for this type. + * + * BPF_F_INSERT_HEAD: The bpf program will be added at the beginning + * of the list of existing bpf program for this type.. + */ +#define BPF_F_INSERT_HEAD (1U << 0) + /* When BPF ldimm64's insn[0].src_reg != 0 then this can have * the following extensions: * diff --git a/tools/lib/bpf/libbpf.c b/tools/lib/bpf/libbpf.c index d1e305f760bb..6f0cb6717207 100644 --- a/tools/lib/bpf/libbpf.c +++ b/tools/lib/bpf/libbpf.c @@ -10533,10 +10533,11 @@ static struct bpf_link *attach_lsm(const struct bpf_program *prog, long cookie)
static struct bpf_link * bpf_program__attach_fd(const struct bpf_program *prog, int target_fd, int btf_id, - const char *target_name) + const char *target_name, __u32 flags) { DECLARE_LIBBPF_OPTS(bpf_link_create_opts, opts, - .target_btf_id = btf_id); + .target_btf_id = btf_id, + .flags = flags); enum bpf_attach_type attach_type; char errmsg[STRERR_BUFSIZE]; struct bpf_link *link; @@ -10570,19 +10571,19 @@ bpf_program__attach_fd(const struct bpf_program *prog, int target_fd, int btf_id struct bpf_link * bpf_program__attach_cgroup(const struct bpf_program *prog, int cgroup_fd) { - return bpf_program__attach_fd(prog, cgroup_fd, 0, "cgroup"); + return bpf_program__attach_fd(prog, cgroup_fd, 0, "cgroup", 0); }
struct bpf_link * bpf_program__attach_netns(const struct bpf_program *prog, int netns_fd) { - return bpf_program__attach_fd(prog, netns_fd, 0, "netns"); + return bpf_program__attach_fd(prog, netns_fd, 0, "netns", 0); }
struct bpf_link *bpf_program__attach_xdp(const struct bpf_program *prog, int ifindex) { /* target_fd/target_ifindex use the same field in LINK_CREATE */ - return bpf_program__attach_fd(prog, ifindex, 0, "xdp"); + return bpf_program__attach_fd(prog, ifindex, 0, "xdp", 0); }
struct bpf_link *bpf_program__attach_freplace(const struct bpf_program *prog, @@ -10608,7 +10609,7 @@ struct bpf_link *bpf_program__attach_freplace(const struct bpf_program *prog, if (btf_id < 0) return libbpf_err_ptr(btf_id);
- return bpf_program__attach_fd(prog, target_fd, btf_id, "freplace"); + return bpf_program__attach_fd(prog, target_fd, btf_id, "freplace", 0); } else { /* no target, so use raw_tracepoint_open for compatibility * with old kernels @@ -10663,9 +10664,9 @@ static struct bpf_link *attach_iter(const struct bpf_program *prog, long cookie) }
struct bpf_link * -bpf_program__attach_hid(const struct bpf_program *prog, int hid_fd) +bpf_program__attach_hid(const struct bpf_program *prog, int hid_fd, __u32 flags) { - return bpf_program__attach_fd(prog, hid_fd, 0, "hid"); + return bpf_program__attach_fd(prog, hid_fd, 0, "hid", flags); }
struct bpf_link *bpf_program__attach(const struct bpf_program *prog) diff --git a/tools/lib/bpf/libbpf.h b/tools/lib/bpf/libbpf.h index f677ac0a9ede..65be3e2ec62d 100644 --- a/tools/lib/bpf/libbpf.h +++ b/tools/lib/bpf/libbpf.h @@ -530,7 +530,7 @@ LIBBPF_API struct bpf_link * bpf_program__attach_iter(const struct bpf_program *prog, const struct bpf_iter_attach_opts *opts); LIBBPF_API struct bpf_link * -bpf_program__attach_hid(const struct bpf_program *prog, int hid_fd); +bpf_program__attach_hid(const struct bpf_program *prog, int hid_fd, __u32 flags);
/* * Libbpf allows callers to adjust BPF programs before being loaded
Insert 3 programs to check that we are doing the correct thing: '2', '1', '3' are inserted, but '1' is supposed to be executed first.
Signed-off-by: Benjamin Tissoires benjamin.tissoires@redhat.com
---
new in v2 --- tools/testing/selftests/bpf/prog_tests/hid.c | 90 ++++++++++++++++++-- tools/testing/selftests/bpf/progs/hid.c | 36 ++++++++ 2 files changed, 121 insertions(+), 5 deletions(-)
diff --git a/tools/testing/selftests/bpf/prog_tests/hid.c b/tools/testing/selftests/bpf/prog_tests/hid.c index a2bab6a799a1..f7fec748a0b9 100644 --- a/tools/testing/selftests/bpf/prog_tests/hid.c +++ b/tools/testing/selftests/bpf/prog_tests/hid.c @@ -353,7 +353,7 @@ static int test_hid_raw_event(struct hid *hid_skel, int uhid_fd, int sysfs_fd)
/* attach the first program */ hid_skel->links.hid_first_event = - bpf_program__attach_hid(hid_skel->progs.hid_first_event, sysfs_fd); + bpf_program__attach_hid(hid_skel->progs.hid_first_event, sysfs_fd, 0); if (!ASSERT_OK_PTR(hid_skel->links.hid_first_event, "attach_hid(hid_first_event)")) return PTR_ERR(hid_skel->links.hid_first_event); @@ -429,7 +429,7 @@ static int test_hid_set_get_data(struct hid *hid_skel, int uhid_fd, int sysfs_fd
/* attach hid_set_get_data program */ hid_skel->links.hid_set_get_data = - bpf_program__attach_hid(hid_skel->progs.hid_set_get_data, sysfs_fd); + bpf_program__attach_hid(hid_skel->progs.hid_set_get_data, sysfs_fd, 0); if (!ASSERT_OK_PTR(hid_skel->links.hid_set_get_data, "attach_hid(hid_set_get_data)")) return PTR_ERR(hid_skel->links.hid_set_get_data); @@ -498,7 +498,7 @@ static int test_hid_user_call(struct hid *hid_skel, int uhid_fd, int sysfs_fd) );
/* attach hid_user program */ - hid_skel->links.hid_user = bpf_program__attach_hid(hid_skel->progs.hid_user, sysfs_fd); + hid_skel->links.hid_user = bpf_program__attach_hid(hid_skel->progs.hid_user, sysfs_fd, 0); if (!ASSERT_OK_PTR(hid_skel->links.hid_user, "attach_hid(hid_user)")) return PTR_ERR(hid_skel->links.hid_user); @@ -552,7 +552,7 @@ static int test_hid_user_raw_request_call(struct hid *hid_skel, int uhid_fd, int
/* attach hid_user_raw_request program */ hid_skel->links.hid_user_raw_request = - bpf_program__attach_hid(hid_skel->progs.hid_user_raw_request, sysfs_fd); + bpf_program__attach_hid(hid_skel->progs.hid_user_raw_request, sysfs_fd, 0); if (!ASSERT_OK_PTR(hid_skel->links.hid_user_raw_request, "attach_hid(hid_user_raw_request)")) return PTR_ERR(hid_skel->links.hid_user_raw_request); @@ -582,6 +582,83 @@ static int test_hid_user_raw_request_call(struct hid *hid_skel, int uhid_fd, int return ret; }
+/* + * Attach hid_insert{0,1,2} to the given uhid device, + * retrieve and open the matching hidraw node, + * inject one event in the uhid device, + * check that the programs have been inserted in the correct order. + */ +static int test_hid_attach_flags(struct hid *hid_skel, int uhid_fd, int sysfs_fd) +{ + int err, hidraw_ino, hidraw_fd = -1; + char hidraw_path[64] = {0}; + u8 buf[10] = {0}; + int ret = -1; + + /* attach hid_test_insert2 program */ + hid_skel->links.hid_test_insert2 = + bpf_program__attach_hid(hid_skel->progs.hid_test_insert2, sysfs_fd, 0); + if (!ASSERT_OK_PTR(hid_skel->links.hid_test_insert2, + "attach_hid(hid_test_insert2)")) + return PTR_ERR(hid_skel->links.hid_test_insert2); + + /* then attach hid_test_insert1 program before the previous*/ + hid_skel->links.hid_test_insert1 = + bpf_program__attach_hid(hid_skel->progs.hid_test_insert1, + sysfs_fd, + BPF_F_INSERT_HEAD); + if (!ASSERT_OK_PTR(hid_skel->links.hid_test_insert1, + "attach_hid(hid_test_insert1)")) + return PTR_ERR(hid_skel->links.hid_test_insert1); + + /* finally attach hid_test_insert3 at the end */ + hid_skel->links.hid_test_insert3 = + bpf_program__attach_hid(hid_skel->progs.hid_test_insert3, sysfs_fd, 0); + if (!ASSERT_OK_PTR(hid_skel->links.hid_test_insert3, + "attach_hid(hid_test_insert3)")) + return PTR_ERR(hid_skel->links.hid_test_insert3); + + hidraw_ino = get_hidraw(hid_skel->links.hid_test_insert1); + if (!ASSERT_GE(hidraw_ino, 0, "get_hidraw")) + goto cleanup; + + /* open hidraw node to check the other side of the pipe */ + sprintf(hidraw_path, "/dev/hidraw%d", hidraw_ino); + hidraw_fd = open(hidraw_path, O_RDWR | O_NONBLOCK); + + if (!ASSERT_GE(hidraw_fd, 0, "open_hidraw")) + goto cleanup; + + /* inject one event */ + buf[0] = 1; + send_event(uhid_fd, buf, 6); + + /* read the data from hidraw */ + memset(buf, 0, sizeof(buf)); + err = read(hidraw_fd, buf, sizeof(buf)); + if (!ASSERT_EQ(err, 6, "read_hidraw")) + goto cleanup; + + if (!ASSERT_EQ(buf[1], 1, "hid_test_insert1")) + goto cleanup; + + if (!ASSERT_EQ(buf[2], 2, "hid_test_insert2")) + goto cleanup; + + if (!ASSERT_EQ(buf[3], 3, "hid_test_insert3")) + goto cleanup; + + ret = 0; + +cleanup: + if (hidraw_fd >= 0) + close(hidraw_fd); + + hid__detach(hid_skel); + + return ret; +} + /* * Attach hid_rdesc_fixup to the given uhid device, * retrieve and open the matching hidraw node, @@ -598,7 +675,7 @@ static int test_rdesc_fixup(struct hid *hid_skel, int uhid_fd, int sysfs_fd)
/* attach the program */ hid_skel->links.hid_rdesc_fixup = - bpf_program__attach_hid(hid_skel->progs.hid_rdesc_fixup, sysfs_fd); + bpf_program__attach_hid(hid_skel->progs.hid_rdesc_fixup, sysfs_fd, 0); if (!ASSERT_OK_PTR(hid_skel->links.hid_rdesc_fixup, "attach_hid(hid_rdesc_fixup)")) return PTR_ERR(hid_skel->links.hid_rdesc_fixup); @@ -695,6 +772,9 @@ void serial_test_hid_bpf(void) err = test_hid_user_raw_request_call(hid_skel, uhid_fd, sysfs_fd); ASSERT_OK(err, "hid_user_raw_request");
+ err = test_hid_attach_flags(hid_skel, uhid_fd, sysfs_fd); + ASSERT_OK(err, "hid_user_raw_request"); + err = test_rdesc_fixup(hid_skel, uhid_fd, sysfs_fd); ASSERT_OK(err, "hid_rdesc_fixup");
diff --git a/tools/testing/selftests/bpf/progs/hid.c b/tools/testing/selftests/bpf/progs/hid.c index ea011ff9e752..40161aa8cb6e 100644 --- a/tools/testing/selftests/bpf/progs/hid.c +++ b/tools/testing/selftests/bpf/progs/hid.c @@ -178,3 +178,39 @@ int hid_user_raw_request(struct hid_bpf_ctx *ctx)
return ret; } + +SEC("hid/device_event") +int hid_test_insert1(struct hid_bpf_ctx *ctx) +{ + /* we need to be run first */ + if (ctx->data[2] || ctx->data[3]) + return -1; + + ctx->data[1] = 1; + + return 0; +} + +SEC("hid/device_event") +int hid_test_insert2(struct hid_bpf_ctx *ctx) +{ + /* after insert0 and before insert2 */ + if (!ctx->data[1] || ctx->data[3]) + return -1; + + ctx->data[2] = 2; + + return 0; +} + +SEC("hid/device_event") +int hid_test_insert3(struct hid_bpf_ctx *ctx) +{ + /* at the end */ + if (!ctx->data[1] || !ctx->data[2]) + return -1; + + ctx->data[3] = 3; + + return 0; +}
Add the new flag parameter.
Signed-off-by: Benjamin Tissoires benjamin.tissoires@redhat.com
---
new in v2 --- samples/bpf/hid_mouse_user.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/samples/bpf/hid_mouse_user.c b/samples/bpf/hid_mouse_user.c index d4f37caca2fa..bfae25e7b659 100644 --- a/samples/bpf/hid_mouse_user.c +++ b/samples/bpf/hid_mouse_user.c @@ -98,7 +98,7 @@ int main(int argc, char **argv) bpf_object__for_each_program(prog, obj) { progs[prog_count].fd = bpf_program__fd(prog); progs[prog_count].type = bpf_program__get_expected_attach_type(prog); - progs[prog_count].link = bpf_program__attach_hid(prog, sysfs_fd); + progs[prog_count].link = bpf_program__attach_hid(prog, sysfs_fd, 0); if (libbpf_get_error(progs[prog_count].link)) { fprintf(stderr, "bpf_prog_attach: err=%m\n"); progs[prog_count].fd = 0;
On Fri, Mar 4, 2022 at 9:29 AM Benjamin Tissoires benjamin.tissoires@redhat.com wrote:
Hi,
This is a followup of my v1 at [0].
The short summary of the previous cover letter and discussions is that HID could benefit from BPF for the following use cases:
simple fixup of report descriptor: benefits are faster development time and testing, with the produced bpf program being shipped in the kernel directly (the shipping part is *not* addressed here).
Universal Stylus Interface: allows a user-space program to define its own kernel interface
Surface Dial: somehow similar to the previous one except that userspace can decide to change the shape of the exported device
firewall: still partly missing there, there is not yet interception of hidraw calls, but it's coming in a followup series, I promise
tracing: well, tracing.
I tried to address as many comments as I could and here is the short log of changes:
v2:
split the series by subsystem (bpf, HID, libbpf, selftests and samples)
Added an extra patch at the beginning to not require CAP_NET_ADMIN for BPF_PROG_TYPE_LIRC_MODE2 (please shout if this is wrong)
made the bpf context attached to HID program of dynamic size:
- the first 1 kB will be able to be addressed directly
- the rest can be retrieved through bpf_hid_{set|get}_data (note that I am definitivey not happy with that API, because there is part of it in bits and other in bytes. ouch)
added an extra patch to prevent non GPL HID bpf programs to be loaded of type BPF_PROG_TYPE_HID
- same here, not really happy but I don't know where to put that check in verifier.c
added a new flag BPF_F_INSERT_HEAD for BPF_LINK_CREATE syscall when in used with HID program types.
- this flag is used for tracing, to be able to load a program before any others that might already have been inserted and that might change the data stream.
Cheers, Benjamin
The set looks good so far. I will review the rest later.
[...]
A quick note about how we organize these patches. Maybe we can merge some of these patches like:
bpf: introduce hid program type bpf/hid: add a new attach type to change the report descriptor bpf/hid: add new BPF type to trigger commands from userspace
I guess the three can merge into one.
HID: hook up with bpf HID: allow to change the report descriptor from an eBPF program HID: bpf: compute only the required buffer size for the device HID: bpf: only call hid_bpf_raw_event() if a ctx is available
I haven't read through all of them, but I guess they can probably merge as well.
libbpf: add HID program type and API libbpf: add new attach type BPF_HID_RDESC_FIXUP libbpf: add new attach type BPF_HID_USER_EVENT
There 3 can merge, and maybe also the one below
libbpf: add handling for BPF_F_INSERT_HEAD in HID programs
samples/bpf: add new hid_mouse example samples/bpf: add a report descriptor fixup samples/bpf: fix bpf_program__attach_hid() api change
Maybe it makes sense to merge these 3?
bpf/hid: add hid_{get|set}_data helpers HID: bpf: implement hid_bpf_get|set_data bpf/hid: add bpf_hid_raw_request helper function HID: add implementation of bpf_hid_raw_request
We can have 1 or 2 patches for these helpers
selftests/bpf: add tests for the HID-bpf initial implementation selftests/bpf: add report descriptor fixup tests selftests/bpf: add tests for hid_{get|set}_data helpers selftests/bpf: add test for user call of HID bpf programs selftests/bpf: hid: rely on uhid event to know if a test device is ready selftests/bpf: add tests for bpf_hid_hw_request selftests/bpf: Add a test for BPF_F_INSERT_HEAD
These selftests could also merge into 1 or 2 patches I guess.
I understand rearranging these patches may take quite some effort. But I do feel that's a cleaner approach (from someone doesn't know much about HID). If you really hate it that way, we can discuss...
Thanks, Song
Hi Song,
Thanks a lot for the review.
I'll comment on the review in more details next week, but I have a quick question here:
On Sat, Mar 5, 2022 at 2:14 AM Song Liu song@kernel.org wrote:
On Fri, Mar 4, 2022 at 9:29 AM Benjamin Tissoires benjamin.tissoires@redhat.com wrote:
Hi,
This is a followup of my v1 at [0].
The short summary of the previous cover letter and discussions is that HID could benefit from BPF for the following use cases:
simple fixup of report descriptor: benefits are faster development time and testing, with the produced bpf program being shipped in the kernel directly (the shipping part is *not* addressed here).
Universal Stylus Interface: allows a user-space program to define its own kernel interface
Surface Dial: somehow similar to the previous one except that userspace can decide to change the shape of the exported device
firewall: still partly missing there, there is not yet interception of hidraw calls, but it's coming in a followup series, I promise
tracing: well, tracing.
I tried to address as many comments as I could and here is the short log of changes:
v2:
split the series by subsystem (bpf, HID, libbpf, selftests and samples)
Added an extra patch at the beginning to not require CAP_NET_ADMIN for BPF_PROG_TYPE_LIRC_MODE2 (please shout if this is wrong)
made the bpf context attached to HID program of dynamic size:
- the first 1 kB will be able to be addressed directly
- the rest can be retrieved through bpf_hid_{set|get}_data (note that I am definitivey not happy with that API, because there is part of it in bits and other in bytes. ouch)
added an extra patch to prevent non GPL HID bpf programs to be loaded of type BPF_PROG_TYPE_HID
- same here, not really happy but I don't know where to put that check in verifier.c
added a new flag BPF_F_INSERT_HEAD for BPF_LINK_CREATE syscall when in used with HID program types.
- this flag is used for tracing, to be able to load a program before any others that might already have been inserted and that might change the data stream.
Cheers, Benjamin
The set looks good so far. I will review the rest later.
[...]
A quick note about how we organize these patches. Maybe we can merge some of these patches like:
Just to be sure we are talking about the same thing: you mean squash the patch together?
bpf: introduce hid program type bpf/hid: add a new attach type to change the report descriptor bpf/hid: add new BPF type to trigger commands from userspace
I guess the three can merge into one.
HID: hook up with bpf HID: allow to change the report descriptor from an eBPF program HID: bpf: compute only the required buffer size for the device HID: bpf: only call hid_bpf_raw_event() if a ctx is available
I haven't read through all of them, but I guess they can probably merge as well.
There are certainly patches that we could squash together (3 and 4 from this list into the previous ones), but I'd like to keep some sort of granularity here to not have a patch bomb that gets harder to come back later.
libbpf: add HID program type and API libbpf: add new attach type BPF_HID_RDESC_FIXUP libbpf: add new attach type BPF_HID_USER_EVENT
There 3 can merge, and maybe also the one below
libbpf: add handling for BPF_F_INSERT_HEAD in HID programs
Yeah, the libbpf changes are small enough to not really justify separate patches.
samples/bpf: add new hid_mouse example samples/bpf: add a report descriptor fixup samples/bpf: fix bpf_program__attach_hid() api change
Maybe it makes sense to merge these 3?
Sure, why not.
bpf/hid: add hid_{get|set}_data helpers HID: bpf: implement hid_bpf_get|set_data bpf/hid: add bpf_hid_raw_request helper function HID: add implementation of bpf_hid_raw_request
We can have 1 or 2 patches for these helpers
OK, the patches should be self-contained enough.
selftests/bpf: add tests for the HID-bpf initial implementation selftests/bpf: add report descriptor fixup tests selftests/bpf: add tests for hid_{get|set}_data helpers selftests/bpf: add test for user call of HID bpf programs selftests/bpf: hid: rely on uhid event to know if a test device is ready selftests/bpf: add tests for bpf_hid_hw_request selftests/bpf: Add a test for BPF_F_INSERT_HEAD
These selftests could also merge into 1 or 2 patches I guess.
I'd still like to link them to the granularity of the bpf changes, so I can refer a selftest change to a specific commit/functionality added. But that's just my personal taste, and I can be convinced otherwise. This should give us maybe 4 patches instead of 7.
I understand rearranging these patches may take quite some effort. But I do feel that's a cleaner approach (from someone doesn't know much about HID). If you really hate it that way, we can discuss...
No worries. I don't mind iterating on the series. IIRC I already rewrote it twice from scratch, and that's when the selftests I introduced in the second rewrite were tremendously helpful :) And honestly I don't think it'll be too much effort to reorder/squash the patches given that the v2 is *very* granular.
Anyway, I prefer having the reviewers happy so we can have a solid rock API from day 1 than keeping it obscure for everyone and having to deal with design issues forever. So if it takes 10 or 20 revisions to have everybody on the same page, that's fine with me (not that I want to have that many revisions, just that I won't be afraid of the bikeshedding we might have at some point).
Cheers, Benjamin
On Sat, Mar 5, 2022 at 2:23 AM Benjamin Tissoires benjamin.tissoires@redhat.com wrote:
The set looks good so far. I will review the rest later.
[...]
A quick note about how we organize these patches. Maybe we can merge some of these patches like:
Just to be sure we are talking about the same thing: you mean squash the patch together?
Right, squash some patches together.
bpf: introduce hid program type bpf/hid: add a new attach type to change the report descriptor bpf/hid: add new BPF type to trigger commands from userspace
I guess the three can merge into one.
HID: hook up with bpf HID: allow to change the report descriptor from an eBPF program HID: bpf: compute only the required buffer size for the device HID: bpf: only call hid_bpf_raw_event() if a ctx is available
I haven't read through all of them, but I guess they can probably merge as well.
There are certainly patches that we could squash together (3 and 4 from this list into the previous ones), but I'd like to keep some sort of granularity here to not have a patch bomb that gets harder to come back later.
Totally agreed with the granularity of patches. I am not a big fan of patch bombs either. :)
I guess the problem I have with the current version is that I don't have a big picture of the design while reading through relatively big patches. A overview with the following information in the cover letter would be really help here: 1. How different types of programs are triggered (IRQ, user input, etc.); 2. What are the operations and/or outcomes of these programs; 3. How would programs of different types (or attach types) interact with each other (via bpf maps? chaining?) 4. What's the new uapi; 5. New helpers and other logistics
Sometimes, I find the changes to uapi are the key for me to understand the patches, and I would like to see one or two patches with all the UAPI changes (i.e. bpf_hid_attach_type). However, that may or may not apply to this set due to granularity concerns.
Does this make sense?
Thanks, Song
libbpf: add HID program type and API libbpf: add new attach type BPF_HID_RDESC_FIXUP libbpf: add new attach type BPF_HID_USER_EVENT
There 3 can merge, and maybe also the one below
libbpf: add handling for BPF_F_INSERT_HEAD in HID programs
Yeah, the libbpf changes are small enough to not really justify separate patches.
samples/bpf: add new hid_mouse example samples/bpf: add a report descriptor fixup samples/bpf: fix bpf_program__attach_hid() api change
Maybe it makes sense to merge these 3?
Sure, why not.
bpf/hid: add hid_{get|set}_data helpers HID: bpf: implement hid_bpf_get|set_data bpf/hid: add bpf_hid_raw_request helper function HID: add implementation of bpf_hid_raw_request
We can have 1 or 2 patches for these helpers
OK, the patches should be self-contained enough.
selftests/bpf: add tests for the HID-bpf initial implementation selftests/bpf: add report descriptor fixup tests selftests/bpf: add tests for hid_{get|set}_data helpers selftests/bpf: add test for user call of HID bpf programs selftests/bpf: hid: rely on uhid event to know if a test device is ready selftests/bpf: add tests for bpf_hid_hw_request selftests/bpf: Add a test for BPF_F_INSERT_HEAD
These selftests could also merge into 1 or 2 patches I guess.
I'd still like to link them to the granularity of the bpf changes, so I can refer a selftest change to a specific commit/functionality added. But that's just my personal taste, and I can be convinced otherwise. This should give us maybe 4 patches instead of 7.
I understand rearranging these patches may take quite some effort. But I do feel that's a cleaner approach (from someone doesn't know much about HID). If you really hate it that way, we can discuss...
No worries. I don't mind iterating on the series. IIRC I already rewrote it twice from scratch, and that's when the selftests I introduced in the second rewrite were tremendously helpful :) And honestly I don't think it'll be too much effort to reorder/squash the patches given that the v2 is *very* granular.
Anyway, I prefer having the reviewers happy so we can have a solid rock API from day 1 than keeping it obscure for everyone and having to deal with design issues forever. So if it takes 10 or 20 revisions to have everybody on the same page, that's fine with me (not that I want to have that many revisions, just that I won't be afraid of the bikeshedding we might have at some point).
Cheers, Benjamin
On Mon, Mar 7, 2022 at 7:12 PM Song Liu song@kernel.org wrote:
On Sat, Mar 5, 2022 at 2:23 AM Benjamin Tissoires benjamin.tissoires@redhat.com wrote:
The set looks good so far. I will review the rest later.
[...]
A quick note about how we organize these patches. Maybe we can merge some of these patches like:
Just to be sure we are talking about the same thing: you mean squash the patch together?
Right, squash some patches together.
bpf: introduce hid program type bpf/hid: add a new attach type to change the report descriptor bpf/hid: add new BPF type to trigger commands from userspace
I guess the three can merge into one.
HID: hook up with bpf HID: allow to change the report descriptor from an eBPF program HID: bpf: compute only the required buffer size for the device HID: bpf: only call hid_bpf_raw_event() if a ctx is available
I haven't read through all of them, but I guess they can probably merge as well.
There are certainly patches that we could squash together (3 and 4 from this list into the previous ones), but I'd like to keep some sort of granularity here to not have a patch bomb that gets harder to come back later.
Totally agreed with the granularity of patches. I am not a big fan of patch bombs either. :)
I guess the problem I have with the current version is that I don't have a big picture of the design while reading through relatively big patches. A overview with the following information in the cover letter would be really help here:
- How different types of programs are triggered (IRQ, user input, etc.);
- What are the operations and/or outcomes of these programs;
- How would programs of different types (or attach types) interact
with each other (via bpf maps? chaining?) 4. What's the new uapi; 5. New helpers and other logistics
Sometimes, I find the changes to uapi are the key for me to understand the patches, and I would like to see one or two patches with all the UAPI changes (i.e. bpf_hid_attach_type). However, that may or may not apply to this set due to granularity concerns.
Does this make sense?
It definitely does. And as I read that, I realized that if I manage to get such a clear depiction of what HID-BPF is, it would certainly be a good idea to paste that in a file into the Documentation directory as well :)
I think you have a slightly better picture now with the exchanges we are having on the individual patches, but I'll try to come out with that description in the cover letter for v3.
Cheers, Benjamin
Hi Benjamin,
On 04/03/2022 19:28, Benjamin Tissoires wrote:
Hi,
This is a followup of my v1 at [0].
The short summary of the previous cover letter and discussions is that HID could benefit from BPF for the following use cases:
simple fixup of report descriptor: benefits are faster development time and testing, with the produced bpf program being shipped in the kernel directly (the shipping part is *not* addressed here).
Universal Stylus Interface: allows a user-space program to define its own kernel interface
Surface Dial: somehow similar to the previous one except that userspace can decide to change the shape of the exported device
firewall: still partly missing there, there is not yet interception of hidraw calls, but it's coming in a followup series, I promise
tracing: well, tracing.
I tried to address as many comments as I could and here is the short log of changes:
v2:
split the series by subsystem (bpf, HID, libbpf, selftests and samples)
Added an extra patch at the beginning to not require CAP_NET_ADMIN for BPF_PROG_TYPE_LIRC_MODE2 (please shout if this is wrong)
made the bpf context attached to HID program of dynamic size:
- the first 1 kB will be able to be addressed directly
- the rest can be retrieved through bpf_hid_{set|get}_data (note that I am definitivey not happy with that API, because there is part of it in bits and other in bytes. ouch)
added an extra patch to prevent non GPL HID bpf programs to be loaded of type BPF_PROG_TYPE_HID
- same here, not really happy but I don't know where to put that check in verifier.c
added a new flag BPF_F_INSERT_HEAD for BPF_LINK_CREATE syscall when in used with HID program types.
- this flag is used for tracing, to be able to load a program before any others that might already have been inserted and that might change the data stream.
Cheers, Benjamin
I posted a couple of comments to the series, but other than that for the whole series you can use:
Reviewed-by: Tero Kristo tero.kristo@linux.intel.com
Tested-by: Tero Kristo tero.kristo@linux.intel.com
I did test this with my USI-BPF program + userspace code, they work with few minor updates compared to previous version.
-Tero
[0] https://lore.kernel.org/linux-input/20220224110828.2168231-1-benjamin.tissoi...
Benjamin Tissoires (28): bpf: add new is_sys_admin_prog_type() helper bpf: introduce hid program type HID: hook up with bpf libbpf: add HID program type and API selftests/bpf: add tests for the HID-bpf initial implementation samples/bpf: add new hid_mouse example bpf/hid: add a new attach type to change the report descriptor HID: allow to change the report descriptor from an eBPF program libbpf: add new attach type BPF_HID_RDESC_FIXUP selftests/bpf: add report descriptor fixup tests samples/bpf: add a report descriptor fixup bpf/hid: add hid_{get|set}_data helpers HID: bpf: implement hid_bpf_get|set_data selftests/bpf: add tests for hid_{get|set}_data helpers bpf/hid: add new BPF type to trigger commands from userspace libbpf: add new attach type BPF_HID_USER_EVENT selftests/bpf: add test for user call of HID bpf programs selftests/bpf: hid: rely on uhid event to know if a test device is ready bpf/hid: add bpf_hid_raw_request helper function HID: add implementation of bpf_hid_raw_request selftests/bpf: add tests for bpf_hid_hw_request bpf/verifier: prevent non GPL programs to be loaded against HID HID: bpf: compute only the required buffer size for the device HID: bpf: only call hid_bpf_raw_event() if a ctx is available bpf/hid: Add a flag to add the program at the beginning of the list libbpf: add handling for BPF_F_INSERT_HEAD in HID programs selftests/bpf: Add a test for BPF_F_INSERT_HEAD samples/bpf: fix bpf_program__attach_hid() api change
drivers/hid/Makefile | 1 + drivers/hid/hid-bpf.c | 361 +++++++++ drivers/hid/hid-core.c | 34 +- include/linux/bpf-hid.h | 129 +++ include/linux/bpf_types.h | 4 + include/linux/hid.h | 25 + include/uapi/linux/bpf.h | 59 ++ include/uapi/linux/bpf_hid.h | 50 ++ kernel/bpf/Makefile | 3 + kernel/bpf/hid.c | 652 +++++++++++++++ kernel/bpf/syscall.c | 26 +- kernel/bpf/verifier.c | 7 + samples/bpf/.gitignore | 1 + samples/bpf/Makefile | 4 + samples/bpf/hid_mouse_kern.c | 91 +++ samples/bpf/hid_mouse_user.c | 129 +++ tools/include/uapi/linux/bpf.h | 59 ++ tools/lib/bpf/libbpf.c | 22 +- tools/lib/bpf/libbpf.h | 2 + tools/lib/bpf/libbpf.map | 1 + tools/testing/selftests/bpf/prog_tests/hid.c | 788 +++++++++++++++++++ tools/testing/selftests/bpf/progs/hid.c | 216 +++++ 22 files changed, 2649 insertions(+), 15 deletions(-) create mode 100644 drivers/hid/hid-bpf.c create mode 100644 include/linux/bpf-hid.h create mode 100644 include/uapi/linux/bpf_hid.h create mode 100644 kernel/bpf/hid.c create mode 100644 samples/bpf/hid_mouse_kern.c create mode 100644 samples/bpf/hid_mouse_user.c create mode 100644 tools/testing/selftests/bpf/prog_tests/hid.c create mode 100644 tools/testing/selftests/bpf/progs/hid.c
linux-kselftest-mirror@lists.linaro.org
-
Andrii Nakryiko -
Benjamin Tissoires -
Greg KH -
Sean Young -
Song Liu -
Song Liu -
Tero Kristo