platform/kernel/linux-rpi.git
2 years agobpf/selftests: Update local storage selftest for sleepable programs
KP Singh [Fri, 24 Dec 2021 15:29:16 +0000 (15:29 +0000)]
bpf/selftests: Update local storage selftest for sleepable programs

Remove the spin lock logic and update the selftests to use sleepable
programs to use a mix of sleepable and non-sleepable programs. It's more
useful to test the sleepable programs since the tests don't really need
spinlocks.

Signed-off-by: KP Singh <kpsingh@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Link: https://lore.kernel.org/bpf/20211224152916.1550677-3-kpsingh@kernel.org
2 years agobpf: Allow bpf_local_storage to be used by sleepable programs
KP Singh [Fri, 24 Dec 2021 15:29:15 +0000 (15:29 +0000)]
bpf: Allow bpf_local_storage to be used by sleepable programs

Other maps like hashmaps are already available to sleepable programs.
Sleepable BPF programs run under trace RCU. Allow task, sk and inode
storage to be used from sleepable programs. This allows sleepable and
non-sleepable programs to provide shareable annotations on kernel
objects.

Sleepable programs run in trace RCU where as non-sleepable programs run
in a normal RCU critical section i.e.  __bpf_prog_enter{_sleepable}
and __bpf_prog_exit{_sleepable}) (rcu_read_lock or rcu_read_lock_trace).

In order to make the local storage maps accessible to both sleepable
and non-sleepable programs, one needs to call both
call_rcu_tasks_trace and call_rcu to wait for both trace and classical
RCU grace periods to expire before freeing memory.

Paul's work on call_rcu_tasks_trace allows us to have per CPU queueing
for call_rcu_tasks_trace. This behaviour can be achieved by setting
rcupdate.rcu_task_enqueue_lim=<num_cpus> boot parameter.

In light of these new performance changes and to keep the local storage
code simple, avoid adding a new flag for sleepable maps / local storage
to select the RCU synchronization (trace / classical).

Also, update the dereferencing of the pointers to use
rcu_derference_check (with either the trace or normal RCU locks held)
with a common bpf_rcu_lock_held helper method.

Signed-off-by: KP Singh <kpsingh@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Link: https://lore.kernel.org/bpf/20211224152916.1550677-2-kpsingh@kernel.org
2 years agobpf: Add missing map_get_next_key method to bloom filter map.
Haimin Zhang [Wed, 29 Dec 2021 11:20:02 +0000 (19:20 +0800)]
bpf: Add missing map_get_next_key method to bloom filter map.

Without it, kernel crashes in map_get_next_key().

Fixes: 9330986c0300 ("bpf: Add bloom filter map implementation")
Reported-by: TCS Robot <tcs_robot@tencent.com>
Signed-off-by: Haimin Zhang <tcs_kernel@tencent.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Joanne Koong <joannekoong@fb.com>
Link: https://lore.kernel.org/bpf/1640776802-22421-1-git-send-email-tcs.kernel@gmail.com
2 years agonet: Don't include filter.h from net/sock.h
Jakub Kicinski [Wed, 29 Dec 2021 00:49:13 +0000 (16:49 -0800)]
net: Don't include filter.h from net/sock.h

sock.h is pretty heavily used (5k objects rebuilt on x86 after
it's touched). We can drop the include of filter.h from it and
add a forward declaration of struct sk_filter instead.
This decreases the number of rebuilt objects when bpf.h
is touched from ~5k to ~1k.

There's a lot of missing includes this was masking. Primarily
in networking tho, this time.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Marc Kleine-Budde <mkl@pengutronix.de>
Acked-by: Florian Fainelli <f.fainelli@gmail.com>
Acked-by: Nikolay Aleksandrov <nikolay@nvidia.com>
Acked-by: Stefano Garzarella <sgarzare@redhat.com>
Link: https://lore.kernel.org/bpf/20211229004913.513372-1-kuba@kernel.org
2 years agolibbpf: Improve LINUX_VERSION_CODE detection
Andrii Nakryiko [Wed, 22 Dec 2021 23:10:03 +0000 (15:10 -0800)]
libbpf: Improve LINUX_VERSION_CODE detection

Ubuntu reports incorrect kernel version through uname(), which on older
kernels leads to kprobe BPF programs failing to load due to the version
check mismatch.

Accommodate Ubuntu's quirks with LINUX_VERSION_CODE by using
Ubuntu-specific /proc/version_code to fetch major/minor/patch versions
to form LINUX_VERSION_CODE.

While at it, consolide libbpf's kernel version detection code between
libbpf.c and libbpf_probes.c.

  [0] Closes: https://github.com/libbpf/libbpf/issues/421

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Yonghong Song <yhs@fb.com>
Link: https://lore.kernel.org/bpf/20211222231003.2334940-1-andrii@kernel.org
2 years agolibbpf: Use 100-character limit to make bpf_tracing.h easier to read
Andrii Nakryiko [Wed, 22 Dec 2021 21:39:24 +0000 (13:39 -0800)]
libbpf: Use 100-character limit to make bpf_tracing.h easier to read

Improve bpf_tracing.h's macro definition readability by keeping them
single-line and better aligned. This makes it easier to follow all those
variadic patterns.

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Yonghong Song <yhs@fb.com>
Link: https://lore.kernel.org/bpf/20211222213924.1869758-2-andrii@kernel.org
2 years agolibbpf: Normalize PT_REGS_xxx() macro definitions
Andrii Nakryiko [Wed, 22 Dec 2021 21:39:23 +0000 (13:39 -0800)]
libbpf: Normalize PT_REGS_xxx() macro definitions

Refactor PT_REGS macros definitions in  bpf_tracing.h to avoid excessive
duplication. We currently have classic PT_REGS_xxx() and CO-RE-enabled
PT_REGS_xxx_CORE(). We are about to add also _SYSCALL variants, which
would require excessive copying of all the per-architecture definitions.

Instead, separate architecture-specific field/register names from the
final macro that utilize them. That way for upcoming _SYSCALL variants
we'll be able to just define x86_64 exception and otherwise have one
common set of _SYSCALL macro definitions common for all architectures.

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Tested-by: Ilya Leoshkevich <iii@linux.ibm.com>
Acked-by: Yonghong Song <yhs@fb.com>
Acked-by: Ilya Leoshkevich <iii@linux.ibm.com>
Link: https://lore.kernel.org/bpf/20211222213924.1869758-1-andrii@kernel.org
2 years agoselftests/bpf: Add btf_dump__new to test_cpp
Jiri Olsa [Thu, 23 Dec 2021 13:17:36 +0000 (14:17 +0100)]
selftests/bpf: Add btf_dump__new to test_cpp

Adding btf_dump__new call to test_cpp, so we can
test C++ compilation with that.

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Acked-by: Yonghong Song <yhs@fb.com>
Link: https://lore.kernel.org/bpf/20211223131736.483956-2-jolsa@kernel.org
2 years agolibbpf: Do not use btf_dump__new() macro in C++ mode
Jiri Olsa [Thu, 23 Dec 2021 13:17:35 +0000 (14:17 +0100)]
libbpf: Do not use btf_dump__new() macro in C++ mode

As reported in here [0], C++ compilers don't support
__builtin_types_compatible_p(), so at least don't screw up compilation
for them and let C++ users pick btf_dump__new vs
btf_dump__new_deprecated explicitly.

  [0] https://github.com/libbpf/libbpf/issues/283#issuecomment-986100727

Fixes: 6084f5dc928f ("libbpf: Ensure btf_dump__new() and btf_dump_opts are future-proof")
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Acked-by: Yonghong Song <yhs@fb.com>
Link: https://lore.kernel.org/bpf/20211223131736.483956-1-jolsa@kernel.org
2 years agobpftool: Enable line buffering for stdout
Paul Chaignon [Mon, 20 Dec 2021 21:45:28 +0000 (22:45 +0100)]
bpftool: Enable line buffering for stdout

The output of bpftool prog tracelog is currently buffered, which is
inconvenient when piping the output into other commands. A simple
tracelog | grep will typically not display anything. This patch fixes it
by enabling line buffering on stdout for the whole bpftool binary.

Fixes: 30da46b5dc3a ("tools: bpftool: add a command to dump the trace pipe")
Signed-off-by: Quentin Monnet <quentin@isovalent.com>
Signed-off-by: Paul Chaignon <paul@isovalent.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Acked-by: Yonghong Song <yhs@fb.com>
Link: https://lore.kernel.org/bpf/20211220214528.GA11706@Mem
2 years agobpf: Use struct_size() helper
Xiu Jianfeng [Mon, 20 Dec 2021 11:30:48 +0000 (19:30 +0800)]
bpf: Use struct_size() helper

In an effort to avoid open-coded arithmetic in the kernel, use the
struct_size() helper instead of open-coded calculation.

Signed-off-by: Xiu Jianfeng <xiujianfeng@huawei.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Acked-by: Yonghong Song <yhs@fb.com>
Link: https://github.com/KSPP/linux/issues/160
Link: https://lore.kernel.org/bpf/20211220113048.2859-1-xiujianfeng@huawei.com
2 years agoselftests/bpf: Correct the INDEX address in vmtest.sh
Pu Lehui [Mon, 20 Dec 2021 05:08:03 +0000 (05:08 +0000)]
selftests/bpf: Correct the INDEX address in vmtest.sh

Migration of vmtest to libbpf/ci will change the address
of INDEX in vmtest.sh, which will cause vmtest.sh to not
work due to the failure of rootfs fetching.

Signed-off-by: Pu Lehui <pulehui@huawei.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Tested-by: Lorenzo Fontana <lorenzo.fontana@elastic.co>
Link: https://lore.kernel.org/bpf/20211220050803.2670677-1-pulehui@huawei.com
2 years agobpf: Extend kfunc with PTR_TO_CTX, PTR_TO_MEM argument support
Kumar Kartikeya Dwivedi [Fri, 17 Dec 2021 01:50:24 +0000 (07:20 +0530)]
bpf: Extend kfunc with PTR_TO_CTX, PTR_TO_MEM argument support

Allow passing PTR_TO_CTX, if the kfunc expects a matching struct type,
and punt to PTR_TO_MEM block if reg->type does not fall in one of
PTR_TO_BTF_ID or PTR_TO_SOCK* types. This will be used by future commits
to get access to XDP and TC PTR_TO_CTX, and pass various data (flags,
l4proto, netns_id, etc.) encoded in opts struct passed as pointer to
kfunc.

For PTR_TO_MEM support, arguments are currently limited to pointer to
scalar, or pointer to struct composed of scalars. This is done so that
unsafe scenarios (like passing PTR_TO_MEM where PTR_TO_BTF_ID of
in-kernel valid structure is expected, which may have pointers) are
avoided. Since the argument checking happens basd on argument register
type, it is not easy to ascertain what the expected type is. In the
future, support for PTR_TO_MEM for kfunc can be extended to serve other
usecases. The struct type whose pointer is passed in may have maximum
nesting depth of 4, all recursively composed of scalars or struct with
scalars.

Future commits will add negative tests that check whether these
restrictions imposed for kfunc arguments are duly rejected by BPF
verifier or not.

Signed-off-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20211217015031.1278167-4-memxor@gmail.com
2 years agoMerge branch 'Introduce composable bpf types'
Alexei Starovoitov [Sat, 18 Dec 2021 21:40:42 +0000 (13:40 -0800)]
Merge branch 'Introduce composable bpf types'

Hao Luo says:

====================
This patch set consists of two changes:

 - a cleanup of arg_type, ret_type and reg_type which try to make those
   types composable. (patch 1/9 - patch 6/9)
 - a bug fix that prevents bpf programs from writing kernel memory.
   (patch 7/9 - patch 9/9)

The purpose of the cleanup is to find a scalable way to express type
nullness and read-onliness. This patchset introduces two flags that
can be applied on all three types: PTR_MAYBE_NULL and MEM_RDONLY.
Previous types such as ARG_XXX_OR_NULL can now be written as

 ARG_XXX | PTR_MAYBE_NULL

Similarly, PTR_TO_RDONLY_BUF is now "PTR_TO_BUF | MEM_RDONLY".

Flags can be composed, as ARGs can be both MEM_RDONLY and MAYBE_NULL.

 ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY

Based on this new composable types, patch 7/9 applies MEM_RDONLY on
PTR_TO_MEM, in order to tag the returned memory from per_cpu_ptr as
read-only. Therefore fixing a previous bug that one can leverage
per_cpu_ptr to modify kernel memory within BPF programs.

Patch 8/9 generalizes the use of MEM_RDONLY further by tagging a set of
helper arguments ARG_PTR_TO_MEM with MEM_RDONLY. Some helper functions
may override their arguments, such as bpf_d_path, bpf_snprintf. In this
patch, we narrow the ARG_PTR_TO_MEM to be compatible with only a subset
of memory types. This prevents these helpers from writing read-only
memories. For the helpers that do not write its arguments, we add tag
MEM_RDONLY to allow taking a RDONLY memory as argument.

Changes since v1:
 - use %u to print base_type(type) instead of %lu. (Andrii, patch 3/9)
 - improve reg_type_str() by appending '_or_null' and prepending 'rdonly_'.
   use preallocated buffer in 'bpf_env'.
 - unified handling of the previous XXX_OR_NULL in adjust_ptr_min_max_vals
   (Andrii, patch 4/9)
 - move PTR_TO_MAP_KEY up to PTR_TO_MAP_VALUE so that we don't have
   to change to drivers that assume the numeric values of bpf_reg.
   (patch 4/9)
 - reintroduce the typo from previous commits in fixes tags (Andrii, patch 7/9)
 - extensive comments on the reason behind folding flags in
   check_reg_type (Andrii, patch 8/9)

Changes since RFC v2:
 - renamed BPF_BASE_TYPE to a more succinct name base_type and move its
   definition to bpf_verifier.h. Same for BPF_TYPE_FLAG. (Alexei)
 - made checking MEM_RDONLY in check_reg_type() universal (Alexei)
 - ran through majority of test_progs and fixed bugs in RFC v2:
   - fixed incorrect BPF_BASE_TYPE_MASK. The high bit of GENMASK should
     be BITS - 1, rather than BITS. patch 1/9.
   - fixed incorrect conditions when checking ARG_PTR_TO_MAP_VALUE in
     check_func_arg(). See patch 2/9.
   - fixed a bug where PTR_TO_BTF_ID may be combined with MEM_RDONLY,
     causing the check in check_mem_access() to fall through to the
     'else' branch. See check_helper_call() in patch 7/9.
 - fixed build failure on netronome driver. Entries in bpf_reg_type have
   been ordered. patch 4/9.
 - fixed build warnings of using '%d' to print base_type. patch 4/9
 - unify arg_type_may_be_null() and reg_type_may_be_null() into a single
   type_may_be_null().

Previous versions:

 v1:
   https://lwn.net/Articles/877938/

 RFC v2:
   https://lwn.net/Articles/877171/

 RFC v1:
   https://lore.kernel.org/bpf/20211109003052.3499225-1-haoluo@google.com/T/
   https://lore.kernel.org/bpf/20211109021624.1140446-8-haoluo@google.com/T/
====================

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2 years agobpf/selftests: Test PTR_TO_RDONLY_MEM
Hao Luo [Fri, 17 Dec 2021 00:31:52 +0000 (16:31 -0800)]
bpf/selftests: Test PTR_TO_RDONLY_MEM

This test verifies that a ksym of non-struct can not be directly
updated.

Signed-off-by: Hao Luo <haoluo@google.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20211217003152.48334-10-haoluo@google.com
2 years agobpf: Add MEM_RDONLY for helper args that are pointers to rdonly mem.
Hao Luo [Fri, 17 Dec 2021 00:31:51 +0000 (16:31 -0800)]
bpf: Add MEM_RDONLY for helper args that are pointers to rdonly mem.

Some helper functions may modify its arguments, for example,
bpf_d_path, bpf_get_stack etc. Previously, their argument types
were marked as ARG_PTR_TO_MEM, which is compatible with read-only
mem types, such as PTR_TO_RDONLY_BUF. Therefore it's legitimate,
but technically incorrect, to modify a read-only memory by passing
it into one of such helper functions.

This patch tags the bpf_args compatible with immutable memory with
MEM_RDONLY flag. The arguments that don't have this flag will be
only compatible with mutable memory types, preventing the helper
from modifying a read-only memory. The bpf_args that have
MEM_RDONLY are compatible with both mutable memory and immutable
memory.

Signed-off-by: Hao Luo <haoluo@google.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20211217003152.48334-9-haoluo@google.com
2 years agobpf: Make per_cpu_ptr return rdonly PTR_TO_MEM.
Hao Luo [Fri, 17 Dec 2021 00:31:50 +0000 (16:31 -0800)]
bpf: Make per_cpu_ptr return rdonly PTR_TO_MEM.

Tag the return type of {per, this}_cpu_ptr with RDONLY_MEM. The
returned value of this pair of helpers is kernel object, which
can not be updated by bpf programs. Previously these two helpers
return PTR_OT_MEM for kernel objects of scalar type, which allows
one to directly modify the memory. Now with RDONLY_MEM tagging,
the verifier will reject programs that write into RDONLY_MEM.

Fixes: 63d9b80dcf2c ("bpf: Introducte bpf_this_cpu_ptr()")
Fixes: eaa6bcb71ef6 ("bpf: Introduce bpf_per_cpu_ptr()")
Fixes: 4976b718c355 ("bpf: Introduce pseudo_btf_id")
Signed-off-by: Hao Luo <haoluo@google.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20211217003152.48334-8-haoluo@google.com
2 years agobpf: Convert PTR_TO_MEM_OR_NULL to composable types.
Hao Luo [Fri, 17 Dec 2021 00:31:49 +0000 (16:31 -0800)]
bpf: Convert PTR_TO_MEM_OR_NULL to composable types.

Remove PTR_TO_MEM_OR_NULL and replace it with PTR_TO_MEM combined with
flag PTR_MAYBE_NULL.

Signed-off-by: Hao Luo <haoluo@google.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20211217003152.48334-7-haoluo@google.com
2 years agobpf: Introduce MEM_RDONLY flag
Hao Luo [Fri, 17 Dec 2021 00:31:48 +0000 (16:31 -0800)]
bpf: Introduce MEM_RDONLY flag

This patch introduce a flag MEM_RDONLY to tag a reg value
pointing to read-only memory. It makes the following changes:

1. PTR_TO_RDWR_BUF -> PTR_TO_BUF
2. PTR_TO_RDONLY_BUF -> PTR_TO_BUF | MEM_RDONLY

Signed-off-by: Hao Luo <haoluo@google.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20211217003152.48334-6-haoluo@google.com
2 years agobpf: Replace PTR_TO_XXX_OR_NULL with PTR_TO_XXX | PTR_MAYBE_NULL
Hao Luo [Fri, 17 Dec 2021 00:31:47 +0000 (16:31 -0800)]
bpf: Replace PTR_TO_XXX_OR_NULL with PTR_TO_XXX | PTR_MAYBE_NULL

We have introduced a new type to make bpf_reg composable, by
allocating bits in the type to represent flags.

One of the flags is PTR_MAYBE_NULL which indicates a pointer
may be NULL. This patch switches the qualified reg_types to
use this flag. The reg_types changed in this patch include:

1. PTR_TO_MAP_VALUE_OR_NULL
2. PTR_TO_SOCKET_OR_NULL
3. PTR_TO_SOCK_COMMON_OR_NULL
4. PTR_TO_TCP_SOCK_OR_NULL
5. PTR_TO_BTF_ID_OR_NULL
6. PTR_TO_MEM_OR_NULL
7. PTR_TO_RDONLY_BUF_OR_NULL
8. PTR_TO_RDWR_BUF_OR_NULL

Signed-off-by: Hao Luo <haoluo@google.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/r/20211217003152.48334-5-haoluo@google.com
2 years agobpf: Replace RET_XXX_OR_NULL with RET_XXX | PTR_MAYBE_NULL
Hao Luo [Fri, 17 Dec 2021 00:31:46 +0000 (16:31 -0800)]
bpf: Replace RET_XXX_OR_NULL with RET_XXX | PTR_MAYBE_NULL

We have introduced a new type to make bpf_ret composable, by
reserving high bits to represent flags.

One of the flag is PTR_MAYBE_NULL, which indicates a pointer
may be NULL. When applying this flag to ret_types, it means
the returned value could be a NULL pointer. This patch
switches the qualified arg_types to use this flag.
The ret_types changed in this patch include:

1. RET_PTR_TO_MAP_VALUE_OR_NULL
2. RET_PTR_TO_SOCKET_OR_NULL
3. RET_PTR_TO_TCP_SOCK_OR_NULL
4. RET_PTR_TO_SOCK_COMMON_OR_NULL
5. RET_PTR_TO_ALLOC_MEM_OR_NULL
6. RET_PTR_TO_MEM_OR_BTF_ID_OR_NULL
7. RET_PTR_TO_BTF_ID_OR_NULL

This patch doesn't eliminate the use of these names, instead
it makes them aliases to 'RET_PTR_TO_XXX | PTR_MAYBE_NULL'.

Signed-off-by: Hao Luo <haoluo@google.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20211217003152.48334-4-haoluo@google.com
2 years agobpf: Replace ARG_XXX_OR_NULL with ARG_XXX | PTR_MAYBE_NULL
Hao Luo [Fri, 17 Dec 2021 00:31:45 +0000 (16:31 -0800)]
bpf: Replace ARG_XXX_OR_NULL with ARG_XXX | PTR_MAYBE_NULL

We have introduced a new type to make bpf_arg composable, by
reserving high bits of bpf_arg to represent flags of a type.

One of the flags is PTR_MAYBE_NULL which indicates a pointer
may be NULL. When applying this flag to an arg_type, it means
the arg can take NULL pointer. This patch switches the
qualified arg_types to use this flag. The arg_types changed
in this patch include:

1. ARG_PTR_TO_MAP_VALUE_OR_NULL
2. ARG_PTR_TO_MEM_OR_NULL
3. ARG_PTR_TO_CTX_OR_NULL
4. ARG_PTR_TO_SOCKET_OR_NULL
5. ARG_PTR_TO_ALLOC_MEM_OR_NULL
6. ARG_PTR_TO_STACK_OR_NULL

This patch does not eliminate the use of these arg_types, instead
it makes them an alias to the 'ARG_XXX | PTR_MAYBE_NULL'.

Signed-off-by: Hao Luo <haoluo@google.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20211217003152.48334-3-haoluo@google.com
2 years agobpf: Introduce composable reg, ret and arg types.
Hao Luo [Fri, 17 Dec 2021 00:31:44 +0000 (16:31 -0800)]
bpf: Introduce composable reg, ret and arg types.

There are some common properties shared between bpf reg, ret and arg
values. For instance, a value may be a NULL pointer, or a pointer to
a read-only memory. Previously, to express these properties, enumeration
was used. For example, in order to test whether a reg value can be NULL,
reg_type_may_be_null() simply enumerates all types that are possibly
NULL. The problem of this approach is that it's not scalable and causes
a lot of duplication. These properties can be combined, for example, a
type could be either MAYBE_NULL or RDONLY, or both.

This patch series rewrites the layout of reg_type, arg_type and
ret_type, so that common properties can be extracted and represented as
composable flag. For example, one can write

 ARG_PTR_TO_MEM | PTR_MAYBE_NULL

which is equivalent to the previous

 ARG_PTR_TO_MEM_OR_NULL

The type ARG_PTR_TO_MEM are called "base type" in this patch. Base
types can be extended with flags. A flag occupies the higher bits while
base types sits in the lower bits.

This patch in particular sets up a set of macro for this purpose. The
following patches will rewrite arg_types, ret_types and reg_types
respectively.

Signed-off-by: Hao Luo <haoluo@google.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20211217003152.48334-2-haoluo@google.com
2 years agobpftool: Reimplement large insn size limit feature probing
Andrii Nakryiko [Fri, 17 Dec 2021 17:12:02 +0000 (09:12 -0800)]
bpftool: Reimplement large insn size limit feature probing

Reimplement bpf_probe_large_insn_limit() in bpftool, as that libbpf API
is scheduled for deprecation in v0.8.

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Dave Marchevsky <davemarchevsky@fb.com>
Link: https://lore.kernel.org/bpf/20211217171202.3352835-4-andrii@kernel.org
2 years agoselftests/bpf: Add libbpf feature-probing API selftests
Andrii Nakryiko [Fri, 17 Dec 2021 17:12:01 +0000 (09:12 -0800)]
selftests/bpf: Add libbpf feature-probing API selftests

Add selftests for prog/map/prog+helper feature probing APIs. Prog and
map selftests are designed in such a way that they will always test all
the possible prog/map types, based on running kernel's vmlinux BTF enum
definition. This way we'll always be sure that when adding new BPF
program types or map types, libbpf will be always updated accordingly to
be able to feature-detect them.

BPF prog_helper selftest will have to be manually extended with
interesting and important prog+helper combinations, it's easy, but can't
be completely automated.

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Dave Marchevsky <davemarchevsky@fb.com>
Link: https://lore.kernel.org/bpf/20211217171202.3352835-3-andrii@kernel.org
2 years agolibbpf: Rework feature-probing APIs
Andrii Nakryiko [Fri, 17 Dec 2021 17:12:00 +0000 (09:12 -0800)]
libbpf: Rework feature-probing APIs

Create three extensible alternatives to inconsistently named
feature-probing APIs:

  - libbpf_probe_bpf_prog_type() instead of bpf_probe_prog_type();
  - libbpf_probe_bpf_map_type() instead of bpf_probe_map_type();
  - libbpf_probe_bpf_helper() instead of bpf_probe_helper().

Set up return values such that libbpf can report errors (e.g., if some
combination of input arguments isn't possible to validate, etc), in
addition to whether the feature is supported (return value 1) or not
supported (return value 0).

Also schedule deprecation of those three APIs. Also schedule deprecation
of bpf_probe_large_insn_limit().

Also fix all the existing detection logic for various program and map
types that never worked:

  - BPF_PROG_TYPE_LIRC_MODE2;
  - BPF_PROG_TYPE_TRACING;
  - BPF_PROG_TYPE_LSM;
  - BPF_PROG_TYPE_EXT;
  - BPF_PROG_TYPE_SYSCALL;
  - BPF_PROG_TYPE_STRUCT_OPS;
  - BPF_MAP_TYPE_STRUCT_OPS;
  - BPF_MAP_TYPE_BLOOM_FILTER.

Above prog/map types needed special setups and detection logic to work.
Subsequent patch adds selftests that will make sure that all the
detection logic keeps working for all current and future program and map
types, avoiding otherwise inevitable bit rot.

  [0] Closes: https://github.com/libbpf/libbpf/issues/312

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Dave Marchevsky <davemarchevsky@fb.com>
Cc: Julia Kartseva <hex@fb.com>
Link: https://lore.kernel.org/bpf/20211217171202.3352835-2-andrii@kernel.org
2 years agoOnly output backtracking information in log level 2
Christy Lee [Thu, 16 Dec 2021 21:33:58 +0000 (13:33 -0800)]
Only output backtracking information in log level 2

Backtracking information is very verbose, don't print it in log
level 1 to improve readability.

Signed-off-by: Christy Lee <christylee@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20211216213358.3374427-4-christylee@fb.com
2 years agobpf: Right align verifier states in verifier logs.
Christy Lee [Fri, 17 Dec 2021 03:42:45 +0000 (19:42 -0800)]
bpf: Right align verifier states in verifier logs.

Make the verifier logs more readable, print the verifier states
on the corresponding instruction line. If the previous line was
not a bpf instruction, then print the verifier states on its own
line.

Before:

Validating test_pkt_access_subprog3() func#3...
86: R1=invP(id=0) R2=ctx(id=0,off=0,imm=0) R10=fp0
; int test_pkt_access_subprog3(int val, struct __sk_buff *skb)
86: (bf) r6 = r2
87: R2=ctx(id=0,off=0,imm=0) R6_w=ctx(id=0,off=0,imm=0)
87: (bc) w7 = w1
88: R1=invP(id=0) R7_w=invP(id=0,umax_value=4294967295,var_off=(0x0; 0xffffffff))
; return get_skb_len(skb) * get_skb_ifindex(val, skb, get_constant(123));
88: (bf) r1 = r6
89: R1_w=ctx(id=0,off=0,imm=0) R6_w=ctx(id=0,off=0,imm=0)
89: (85) call pc+9
Func#4 is global and valid. Skipping.
90: R0_w=invP(id=0)
90: (bc) w8 = w0
91: R0_w=invP(id=0) R8_w=invP(id=0,umax_value=4294967295,var_off=(0x0; 0xffffffff))
; return get_skb_len(skb) * get_skb_ifindex(val, skb, get_constant(123));
91: (b7) r1 = 123
92: R1_w=invP123
92: (85) call pc+65
Func#5 is global and valid. Skipping.
93: R0=invP(id=0)

After:

86: R1=invP(id=0) R2=ctx(id=0,off=0,imm=0) R10=fp0
; int test_pkt_access_subprog3(int val, struct __sk_buff *skb)
86: (bf) r6 = r2                      ; R2=ctx(id=0,off=0,imm=0) R6_w=ctx(id=0,off=0,imm=0)
87: (bc) w7 = w1                      ; R1=invP(id=0) R7_w=invP(id=0,umax_value=4294967295,var_off=(0x0; 0xffffffff))
; return get_skb_len(skb) * get_skb_ifindex(val, skb, get_constant(123));
88: (bf) r1 = r6                      ; R1_w=ctx(id=0,off=0,imm=0) R6_w=ctx(id=0,off=0,imm=0)
89: (85) call pc+9
Func#4 is global and valid. Skipping.
90: R0_w=invP(id=0)
90: (bc) w8 = w0                      ; R0_w=invP(id=0) R8_w=invP(id=0,umax_value=4294967295,var_off=(0x0; 0xffffffff))
; return get_skb_len(skb) * get_skb_ifindex(val, skb, get_constant(123));
91: (b7) r1 = 123                     ; R1_w=invP123
92: (85) call pc+65
Func#5 is global and valid. Skipping.
93: R0=invP(id=0)

Signed-off-by: Christy Lee <christylee@fb.com>
Acked-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2 years agobpf: Only print scratched registers and stack slots to verifier logs.
Christy Lee [Thu, 16 Dec 2021 21:33:56 +0000 (13:33 -0800)]
bpf: Only print scratched registers and stack slots to verifier logs.

When printing verifier state for any log level, print full verifier
state only on function calls or on errors. Otherwise, only print the
registers and stack slots that were accessed.

Log size differences:

verif_scale_loop6 before: 234566564
verif_scale_loop6 after: 72143943
69% size reduction

kfree_skb before: 166406
kfree_skb after: 55386
69% size reduction

Before:

156: (61) r0 = *(u32 *)(r1 +0)
157: R0_w=invP(id=0,umax_value=4294967295,var_off=(0x0; 0xffffffff)) R1=ctx(id=0,off=0,imm=0) R2_w=invP0 R10=fp0 fp-8_w=00000000 fp-16_w=00\
000000 fp-24_w=00000000 fp-32_w=00000000 fp-40_w=00000000 fp-48_w=00000000 fp-56_w=00000000 fp-64_w=00000000 fp-72_w=00000000 fp-80_w=00000\
000 fp-88_w=00000000 fp-96_w=00000000 fp-104_w=00000000 fp-112_w=00000000 fp-120_w=00000000 fp-128_w=00000000 fp-136_w=00000000 fp-144_w=00\
000000 fp-152_w=00000000 fp-160_w=00000000 fp-168_w=00000000 fp-176_w=00000000 fp-184_w=00000000 fp-192_w=00000000 fp-200_w=00000000 fp-208\
_w=00000000 fp-216_w=00000000 fp-224_w=00000000 fp-232_w=00000000 fp-240_w=00000000 fp-248_w=00000000 fp-256_w=00000000 fp-264_w=00000000 f\
p-272_w=00000000 fp-280_w=00000000 fp-288_w=00000000 fp-296_w=00000000 fp-304_w=00000000 fp-312_w=00000000 fp-320_w=00000000 fp-328_w=00000\
000 fp-336_w=00000000 fp-344_w=00000000 fp-352_w=00000000 fp-360_w=00000000 fp-368_w=00000000 fp-376_w=00000000 fp-384_w=00000000 fp-392_w=\
00000000 fp-400_w=00000000 fp-408_w=00000000 fp-416_w=00000000 fp-424_w=00000000 fp-432_w=00000000 fp-440_w=00000000 fp-448_w=00000000
; return skb->len;
157: (95) exit
Func#4 is safe for any args that match its prototype
Validating get_constant() func#5...
158: R1=invP(id=0) R10=fp0
; int get_constant(long val)
158: (bf) r0 = r1
159: R0_w=invP(id=1) R1=invP(id=1) R10=fp0
; return val - 122;
159: (04) w0 += -122
160: R0_w=invP(id=0,umax_value=4294967295,var_off=(0x0; 0xffffffff)) R1=invP(id=1) R10=fp0
; return val - 122;
160: (95) exit
Func#5 is safe for any args that match its prototype
Validating get_skb_ifindex() func#6...
161: R1=invP(id=0) R2=ctx(id=0,off=0,imm=0) R3=invP(id=0) R10=fp0
; int get_skb_ifindex(int val, struct __sk_buff *skb, int var)
161: (bc) w0 = w3
162: R0_w=invP(id=0,umax_value=4294967295,var_off=(0x0; 0xffffffff)) R1=invP(id=0) R2=ctx(id=0,off=0,imm=0) R3=invP(id=0) R10=fp0

After:

156: (61) r0 = *(u32 *)(r1 +0)
157: R0_w=invP(id=0,umax_value=4294967295,var_off=(0x0; 0xffffffff)) R1=ctx(id=0,off=0,imm=0)
; return skb->len;
157: (95) exit
Func#4 is safe for any args that match its prototype
Validating get_constant() func#5...
158: R1=invP(id=0) R10=fp0
; int get_constant(long val)
158: (bf) r0 = r1
159: R0_w=invP(id=1) R1=invP(id=1)
; return val - 122;
159: (04) w0 += -122
160: R0_w=invP(id=0,umax_value=4294967295,var_off=(0x0; 0xffffffff))
; return val - 122;
160: (95) exit
Func#5 is safe for any args that match its prototype
Validating get_skb_ifindex() func#6...
161: R1=invP(id=0) R2=ctx(id=0,off=0,imm=0) R3=invP(id=0) R10=fp0
; int get_skb_ifindex(int val, struct __sk_buff *skb, int var)
161: (bc) w0 = w3
162: R0_w=invP(id=0,umax_value=4294967295,var_off=(0x0; 0xffffffff)) R3=invP(id=0)

Signed-off-by: Christy Lee <christylee@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20211216213358.3374427-2-christylee@fb.com
2 years agoMerge branch 'bpf: remove the cgroup -> bpf header dependecy'
Alexei Starovoitov [Thu, 16 Dec 2021 22:57:10 +0000 (14:57 -0800)]
Merge branch 'bpf: remove the cgroup -> bpf header dependecy'

Jakub Kicinski says:

====================

Changes to bpf.h tend to clog up our build systems. The netdev/bpf
build bot does incremental builds to save time (reusing the build
directory to only rebuild changed objects).

This is the rough breakdown of how many objects needs to be rebuilt
based on file touched:

kernel.h      40633
bpf.h         17881
bpf-cgroup.h  17875
skbuff.h      10696
bpf-netns.h    7604
netdevice.h    7452
filter.h       5003
sock.h         4959
tcp.h          4048

As the stats show touching bpf.h is _very_ expensive.

Bulk of the objects get rebuilt because MM includes cgroup headers.
Luckily bpf-cgroup.h does not fundamentally depend on bpf.h so we
can break that dependency and reduce the number of objects.

With the patches applied touching bpf.h causes 5019 objects to be rebuilt
(17881 / 5019 = 3.56x). That's pretty much down to filter.h plus noise.

v2:
Try to make the new headers wider in scope. Collapse bpf-link and
bpf-cgroup-types into one header, which may serve as "BPF kernel
API" header in the future if needed. Rename bpf-cgroup-storage.h
to bpf-inlines.h.

Add a fix for the s390 build issue.

v3: https://lore.kernel.org/all/20211215061916.715513-1-kuba@kernel.org/
Merge bpf-includes.h into bpf.h.

v4: https://lore.kernel.org/all/20211215181231.1053479-1-kuba@kernel.org/
Change course - break off cgroup instead of breaking off bpf.

v5:
Add forward declaration of struct bpf_prog to perf_event.h
when !CONFIG_BPF_SYSCALL (kbuild bot).
====================

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2 years agobpf: Remove the cgroup -> bpf header dependecy
Jakub Kicinski [Thu, 16 Dec 2021 02:55:38 +0000 (18:55 -0800)]
bpf: Remove the cgroup -> bpf header dependecy

Remove the dependency from cgroup-defs.h to bpf-cgroup.h and bpf.h.
This reduces the incremental build size of x86 allmodconfig after
bpf.h was touched from ~17k objects rebuilt to ~5k objects.
bpf.h is 2.2kLoC and is modified relatively often.

We need a new header with just the definition of struct cgroup_bpf
and enum cgroup_bpf_attach_type, this is akin to cgroup-defs.h.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Tejun Heo <tj@kernel.org>
Link: https://lore.kernel.org/bpf/20211216025538.1649516-4-kuba@kernel.org
2 years agoadd missing bpf-cgroup.h includes
Jakub Kicinski [Thu, 16 Dec 2021 02:55:37 +0000 (18:55 -0800)]
add missing bpf-cgroup.h includes

We're about to break the cgroup-defs.h -> bpf-cgroup.h dependency,
make sure those who actually need more than the definition of
struct cgroup_bpf include bpf-cgroup.h explicitly.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Tejun Heo <tj@kernel.org>
Link: https://lore.kernel.org/bpf/20211216025538.1649516-3-kuba@kernel.org
2 years agoadd includes masked by cgroup -> bpf dependency
Jakub Kicinski [Thu, 16 Dec 2021 02:55:36 +0000 (18:55 -0800)]
add includes masked by cgroup -> bpf dependency

cgroup pulls in BPF which pulls in a lot of includes.
We're about to break that chain so fix those who were
depending on it.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20211216025538.1649516-2-kuba@kernel.org
2 years agoMerge branch 'tools/bpf: Enable cross-building with clang'
Andrii Nakryiko [Thu, 16 Dec 2021 20:15:05 +0000 (12:15 -0800)]
Merge branch 'tools/bpf: Enable cross-building with clang'

Jean-Philippe Brucker says:

====================

Since v1 [1], I added Quentin's acks and applied Andrii's suggestions:
* Pass CFLAGS to libbpf link in patch 3
* Substitute CLANG_CROSS_FLAGS whole in HOST_CFLAGS to avoid accidents,
  patch 4

Add support for cross-building BPF tools and selftests with clang, by
passing LLVM=1 or CC=clang to make, as well as CROSS_COMPILE. A single
clang toolchain can generate binaries for multiple architectures, so
instead of having prefixes such as aarch64-linux-gnu-gcc, clang uses the
-target parameter: `clang -target aarch64-linux-gnu'.

Patch 1 adds the parameter in Makefile.include so tools can easily
support this. Patch 2 prepares for the libbpf change from patch 3 (keep
building resolve_btfids's libbpf in the host arch, when cross-building
the kernel with clang). Patches 3-6 enable cross-building BPF tools with
clang.

[1] https://lore.kernel.org/bpf/20211122192019.1277299-1-jean-philippe@linaro.org/
====================

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
2 years agoselftests/bpf: Enable cross-building with clang
Jean-Philippe Brucker [Thu, 16 Dec 2021 16:38:43 +0000 (16:38 +0000)]
selftests/bpf: Enable cross-building with clang

Cross building using clang requires passing the "-target" flag rather
than using the CROSS_COMPILE prefix. Makefile.include transforms
CROSS_COMPILE into CLANG_CROSS_FLAGS. Clear CROSS_COMPILE for bpftool
and the host libbpf, and use the clang flags for urandom_read and bench.

Signed-off-by: Jean-Philippe Brucker <jean-philippe@linaro.org>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Acked-by: Quentin Monnet <quentin@isovalent.com>
Link: https://lore.kernel.org/bpf/20211216163842.829836-7-jean-philippe@linaro.org
2 years agotools/runqslower: Enable cross-building with clang
Jean-Philippe Brucker [Thu, 16 Dec 2021 16:38:42 +0000 (16:38 +0000)]
tools/runqslower: Enable cross-building with clang

Cross-building using clang requires passing the "-target" flag rather
than using the CROSS_COMPILE prefix. Makefile.include transforms
CROSS_COMPILE into CLANG_CROSS_FLAGS. Add them to CFLAGS, and erase
CROSS_COMPILE for the bpftool build, since it needs to be executed on
the host.

Signed-off-by: Jean-Philippe Brucker <jean-philippe@linaro.org>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Acked-by: Quentin Monnet <quentin@isovalent.com>
Link: https://lore.kernel.org/bpf/20211216163842.829836-6-jean-philippe@linaro.org
2 years agobpftool: Enable cross-building with clang
Jean-Philippe Brucker [Thu, 16 Dec 2021 16:38:41 +0000 (16:38 +0000)]
bpftool: Enable cross-building with clang

Cross-building using clang requires passing the "-target" flag rather
than using the CROSS_COMPILE prefix. Makefile.include transforms
CROSS_COMPILE into CLANG_CROSS_FLAGS, and adds that to CFLAGS. Remove
the cross flags for the bootstrap bpftool, and erase the CROSS_COMPILE
flag for the bootstrap libbpf.

Signed-off-by: Jean-Philippe Brucker <jean-philippe@linaro.org>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Acked-by: Quentin Monnet <quentin@isovalent.com>
Link: https://lore.kernel.org/bpf/20211216163842.829836-5-jean-philippe@linaro.org
2 years agotools/libbpf: Enable cross-building with clang
Jean-Philippe Brucker [Thu, 16 Dec 2021 16:38:40 +0000 (16:38 +0000)]
tools/libbpf: Enable cross-building with clang

Cross-building using clang requires passing the "-target" flag rather
than using the CROSS_COMPILE prefix. Makefile.include transforms
CROSS_COMPILE into CLANG_CROSS_FLAGS. Add them to the CFLAGS.

Signed-off-by: Jean-Philippe Brucker <jean-philippe@linaro.org>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Acked-by: Quentin Monnet <quentin@isovalent.com>
Link: https://lore.kernel.org/bpf/20211216163842.829836-4-jean-philippe@linaro.org
2 years agotools/resolve_btfids: Support cross-building the kernel with clang
Jean-Philippe Brucker [Thu, 16 Dec 2021 16:38:39 +0000 (16:38 +0000)]
tools/resolve_btfids: Support cross-building the kernel with clang

The CROSS_COMPILE variable may be present during resolve_btfids build if
the kernel is being cross-built. Since resolve_btfids is always executed
on the host, we set CC to HOSTCC in order to use the host toolchain when
cross-building with GCC. But instead of a toolchain prefix, cross-build
with clang uses a "-target" parameter, which Makefile.include deduces
from the CROSS_COMPILE variable. In order to avoid cross-building
libbpf, clear CROSS_COMPILE before building resolve_btfids.

Signed-off-by: Jean-Philippe Brucker <jean-philippe@linaro.org>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Acked-by: Quentin Monnet <quentin@isovalent.com>
Link: https://lore.kernel.org/bpf/20211216163842.829836-3-jean-philippe@linaro.org
2 years agotools: Help cross-building with clang
Jean-Philippe Brucker [Thu, 16 Dec 2021 16:38:38 +0000 (16:38 +0000)]
tools: Help cross-building with clang

Cross-compilation with clang uses the -target parameter rather than a
toolchain prefix. Just like the kernel Makefile, add that parameter to
CFLAGS when CROSS_COMPILE is set.

Unlike the kernel Makefile, we use the --sysroot and --gcc-toolchain
options because unlike the kernel, tools require standard libraries.
Commit c91d4e47e10e ("Makefile: Remove '--gcc-toolchain' flag") provides
some background about --gcc-toolchain. Normally clang finds on its own
the additional utilities and libraries that it needs (for example GNU ld
or glibc). On some systems however, this autodetection doesn't work.
There, our only recourse is asking GCC directly, and pass the result to
--sysroot and --gcc-toolchain. Of course that only works when a cross
GCC is available.

Autodetection worked fine on Debian, but to use the aarch64-linux-gnu
toolchain from Archlinux I needed both --sysroot (for crt1.o) and
--gcc-toolchain (for crtbegin.o, -lgcc). The --prefix parameter wasn't
needed there, but it might be useful on other distributions.

Use the CLANG_CROSS_FLAGS variable instead of CLANG_FLAGS because it
allows tools such as bpftool, that need to build both host and target
binaries, to easily filter out the cross-build flags from CFLAGS.

Signed-off-by: Jean-Philippe Brucker <jean-philippe@linaro.org>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Acked-by: Quentin Monnet <quentin@isovalent.com>
Acked-by: Nick Desaulniers <ndesaulniers@google.com>
Link: https://lore.kernel.org/bpf/20211216163842.829836-2-jean-philippe@linaro.org
2 years agolibbpf: Avoid reading past ELF data section end when copying license
Andrii Nakryiko [Tue, 14 Dec 2021 23:20:54 +0000 (15:20 -0800)]
libbpf: Avoid reading past ELF data section end when copying license

Fix possible read beyond ELF "license" data section if the license
string is not properly zero-terminated. Use the fact that libbpf_strlcpy
never accesses the (N-1)st byte of the source string because it's
replaced with '\0' anyways.

If this happens, it's a violation of contract between libbpf and a user,
but not handling this more robustly upsets CIFuzz, so given the fix is
trivial, let's fix the potential issue.

Fixes: 9fc205b413b3 ("libbpf: Add sane strncpy alternative and use it internally")
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20211214232054.3458774-1-andrii@kernel.org
2 years agoMerge branch 'Stop using bpf_object__find_program_by_title API'
Andrii Nakryiko [Tue, 14 Dec 2021 22:38:06 +0000 (14:38 -0800)]
Merge branch 'Stop using bpf_object__find_program_by_title API'

Kui-Feng Lee says:

====================

bpf_object__find_program_by_title is going to be deprecated since
v0.7.  Replace all use cases with bpf_object__find_program_by_name if
possible, or use bpf_object__for_each_program to iterate over
programs, matching section names.

V3 fixes a broken test case, fexit_bpf2bpf, in selftests/bpf, using
bpf_obj__for_each_program API instead.

[v2] https://lore.kernel.org/bpf/20211211003608.2764928-1-kuifeng@fb.com/
[v1] https://lore.kernel.org/bpf/20211210190855.1369060-1-kuifeng@fb.com/T/
====================

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
2 years agolibbpf: Mark bpf_object__find_program_by_title API deprecated.
Kui-Feng Lee [Tue, 14 Dec 2021 03:59:31 +0000 (19:59 -0800)]
libbpf: Mark bpf_object__find_program_by_title API deprecated.

Deprecate this API since v0.7.  All callers should move to
bpf_object__find_program_by_name if possible, otherwise use
bpf_object__for_each_program to find a program out from a given
section.

[0] Closes: https://github.com/libbpf/libbpf/issues/292

Signed-off-by: Kui-Feng Lee <kuifeng@fb.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20211214035931.1148209-5-kuifeng@fb.com
2 years agotools/perf: Stop using bpf_object__find_program_by_title API.
Kui-Feng Lee [Tue, 14 Dec 2021 03:59:30 +0000 (19:59 -0800)]
tools/perf: Stop using bpf_object__find_program_by_title API.

bpf_obj__find_program_by_title() in libbpf is going to be deprecated.
Call bpf_object_for_each_program to find a program in the section with
a given name instead.

Signed-off-by: Kui-Feng Lee <kuifeng@fb.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20211214035931.1148209-4-kuifeng@fb.com
2 years agosamples/bpf: Stop using bpf_object__find_program_by_title API.
Kui-Feng Lee [Tue, 14 Dec 2021 03:59:29 +0000 (19:59 -0800)]
samples/bpf: Stop using bpf_object__find_program_by_title API.

bpf_object__find_program_by_title is going to be deprecated.
Replace use cases of bpf_object__find_program_by_title in samples/bpf/
with bpf_object__for_each_program.

Signed-off-by: Kui-Feng Lee <kuifeng@fb.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20211214035931.1148209-3-kuifeng@fb.com
2 years agoselftests/bpf: Stop using bpf_object__find_program_by_title API.
Kui-Feng Lee [Tue, 14 Dec 2021 03:59:28 +0000 (19:59 -0800)]
selftests/bpf: Stop using bpf_object__find_program_by_title API.

bpf_object__find_program_by_title is going to be deprecated.  Replace
all use cases in tools/testing/selftests/bpf with
bpf_object__find_program_by_name or bpf_object__for_each_program.

Signed-off-by: Kui-Feng Lee <kuifeng@fb.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20211214035931.1148209-2-kuifeng@fb.com
2 years agoselftests/bpf: Remove explicit setrlimit(RLIMIT_MEMLOCK) in main selftests
Andrii Nakryiko [Tue, 14 Dec 2021 19:59:04 +0000 (11:59 -0800)]
selftests/bpf: Remove explicit setrlimit(RLIMIT_MEMLOCK) in main selftests

As libbpf now is able to automatically take care of RLIMIT_MEMLOCK
increase (or skip it altogether on recent enough kernels), remove
explicit setrlimit() invocations in bench, test_maps, test_verifier, and
test_progs.

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20211214195904.1785155-3-andrii@kernel.org
2 years agolibbpf: Auto-bump RLIMIT_MEMLOCK if kernel needs it for BPF
Andrii Nakryiko [Tue, 14 Dec 2021 19:59:03 +0000 (11:59 -0800)]
libbpf: Auto-bump RLIMIT_MEMLOCK if kernel needs it for BPF

The need to increase RLIMIT_MEMLOCK to do anything useful with BPF is
one of the first extremely frustrating gotchas that all new BPF users go
through and in some cases have to learn it a very hard way.

Luckily, starting with upstream Linux kernel version 5.11, BPF subsystem
dropped the dependency on memlock and uses memcg-based memory accounting
instead. Unfortunately, detecting memcg-based BPF memory accounting is
far from trivial (as can be evidenced by this patch), so in practice
most BPF applications still do unconditional RLIMIT_MEMLOCK increase.

As we move towards libbpf 1.0, it would be good to allow users to forget
about RLIMIT_MEMLOCK vs memcg and let libbpf do the sensible adjustment
automatically. This patch paves the way forward in this matter. Libbpf
will do feature detection of memcg-based accounting, and if detected,
will do nothing. But if the kernel is too old, just like BCC, libbpf
will automatically increase RLIMIT_MEMLOCK on behalf of user
application ([0]).

As this is technically a breaking change, during the transition period
applications have to opt into libbpf 1.0 mode by setting
LIBBPF_STRICT_AUTO_RLIMIT_MEMLOCK bit when calling
libbpf_set_strict_mode().

Libbpf allows to control the exact amount of set RLIMIT_MEMLOCK limit
with libbpf_set_memlock_rlim_max() API. Passing 0 will make libbpf do
nothing with RLIMIT_MEMLOCK. libbpf_set_memlock_rlim_max() has to be
called before the first bpf_prog_load(), bpf_btf_load(), or
bpf_object__load() call, otherwise it has no effect and will return
-EBUSY.

  [0] Closes: https://github.com/libbpf/libbpf/issues/369

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20211214195904.1785155-2-andrii@kernel.org
2 years agolibbpf: Add sane strncpy alternative and use it internally
Andrii Nakryiko [Sat, 11 Dec 2021 00:40:43 +0000 (16:40 -0800)]
libbpf: Add sane strncpy alternative and use it internally

strncpy() has a notoriously error-prone semantics which makes GCC
complain about it a lot (and quite often completely completely falsely
at that). Instead of pleasing GCC all the time (-Wno-stringop-truncation
is unfortunately only supported by GCC, so it's a bit too messy to just
enable it in Makefile), add libbpf-internal libbpf_strlcpy() helper
which follows what FreeBSD's strlcpy() does and what most people would
expect from strncpy(): copies up to N-1 first bytes from source string
into destination string and ensures zero-termination afterwards.

Replace all the relevant uses of strncpy/strncat/memcpy in libbpf with
libbpf_strlcpy().

This also fixes the issue reported by Emmanuel Deloget in xsk.c where
memcpy() could access source string beyond its end.

Fixes: 2f6324a3937f8 (libbpf: Support shared umems between queues and devices)
Reported-by: Emmanuel Deloget <emmanuel.deloget@eho.link>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20211211004043.2374068-1-andrii@kernel.org
2 years agolibbpf: Fix potential uninit memory read
Andrii Nakryiko [Tue, 14 Dec 2021 01:00:32 +0000 (17:00 -0800)]
libbpf: Fix potential uninit memory read

In case of BPF_CORE_TYPE_ID_LOCAL we fill out target result explicitly.
But targ_res itself isn't initialized in such a case, and subsequent
call to bpf_core_patch_insn() might read uninitialized field (like
fail_memsz_adjust in this case). So ensure that targ_res is
zero-initialized for BPF_CORE_TYPE_ID_LOCAL case.

This was reported by Coverity static analyzer.

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20211214010032.3843804-1-andrii@kernel.org
2 years agoxsk: Wipe out dead zero_copy_allocator declarations
Maciej Fijalkowski [Fri, 10 Dec 2021 17:15:11 +0000 (18:15 +0100)]
xsk: Wipe out dead zero_copy_allocator declarations

zero_copy_allocator has been removed back when Bjorn Topel introduced
xsk_buff_pool. Remove references to it that were dangling in the tree.

Signed-off-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Magnus Karlsson <magnus.karlsson@intel.com>
Link: https://lore.kernel.org/bpf/20211210171511.11574-1-maciej.fijalkowski@intel.com
2 years agolibbpf: Add doc comments for bpf_program__(un)pin()
Grant Seltzer [Thu, 9 Dec 2021 23:22:22 +0000 (18:22 -0500)]
libbpf: Add doc comments for bpf_program__(un)pin()

This adds doc comments for the two bpf_program pinning functions,
bpf_program__pin() and bpf_program__unpin()

Signed-off-by: Grant Seltzer <grantseltzer@gmail.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20211209232222.541733-1-grantseltzer@gmail.com
2 years agoselftests/bpf: Fix segfault in bpf_tcp_ca
Jean-Philippe Brucker [Mon, 13 Dec 2021 18:30:59 +0000 (18:30 +0000)]
selftests/bpf: Fix segfault in bpf_tcp_ca

Since commit ad9a7f96445b ("libbpf: Improve logging around BPF program
loading"), libbpf_debug_print() gets an additional prog_name parameter
but doesn't pass it to printf(). Since the format string now expects two
arguments, printf() may read uninitialized data and segfault. Pass
prog_name through.

Fixes: ad9a7f96445b ("libbpf: Improve logging around BPF program loading")
Signed-off-by: Jean-Philippe Brucker <jean-philippe@linaro.org>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20211213183058.346066-1-jean-philippe@linaro.org
2 years agobpf: Let bpf_warn_invalid_xdp_action() report more info
Paolo Abeni [Tue, 30 Nov 2021 10:08:07 +0000 (11:08 +0100)]
bpf: Let bpf_warn_invalid_xdp_action() report more info

In non trivial scenarios, the action id alone is not sufficient to
identify the program causing the warning. Before the previous patch,
the generated stack-trace pointed out at least the involved device
driver.

Let's additionally include the program name and id, and the relevant
device name.

If the user needs additional infos, he can fetch them via a kernel
probe, leveraging the arguments added here.

Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Toke Høiland-Jørgensen <toke@redhat.com>
Link: https://lore.kernel.org/bpf/ddb96bb975cbfddb1546cf5da60e77d5100b533c.1638189075.git.pabeni@redhat.com
2 years agobpf: Do not WARN in bpf_warn_invalid_xdp_action()
Paolo Abeni [Tue, 30 Nov 2021 10:08:06 +0000 (11:08 +0100)]
bpf: Do not WARN in bpf_warn_invalid_xdp_action()

The WARN_ONCE() in bpf_warn_invalid_xdp_action() can be triggered by
any bugged program, and even attaching a correct program to a NIC
not supporting the given action.

The resulting splat, beyond polluting the logs, fouls automated tools:
e.g. a syzkaller reproducers using an XDP program returning an
unsupported action will never pass validation.

Replace the WARN_ONCE with a less intrusive pr_warn_once().

Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Toke Høiland-Jørgensen <toke@redhat.com>
Link: https://lore.kernel.org/bpf/016ceec56e4817ebb2a9e35ce794d5c917df572c.1638189075.git.pabeni@redhat.com
2 years agoMerge branch 'bpf: Add helpers to access traced function arguments'
Alexei Starovoitov [Mon, 13 Dec 2021 17:26:50 +0000 (09:26 -0800)]
Merge branch 'bpf: Add helpers to access traced function arguments'

Jiri Olsa says:

====================
Add new helpers to access traced function arguments that
came out of the trampoline batch changes [1].

  Get n-th argument of the traced function:
    long bpf_get_func_arg(void *ctx, u32 n, u64 *value)

  Get return value of the traced function:
    long bpf_get_func_ret(void *ctx, u64 *value)

  Get arguments count of the traced funtion:
    long bpf_get_func_arg_cnt(void *ctx)

v2 changes:
  - added acks
  - updated stack diagram
  - return -EOPNOTSUPP instead of -EINVAL in bpf_get_func_ret
  - removed gpl_only for all helpers
  - added verifier fix to allow proper arguments checks,
    Andrii asked for checking also 'int *b' argument in
    bpf_modify_return_test programs and it turned out that it's currently
    not supported by verifier - we can't read argument that is int pointer,
    so I had to add verifier change to allow that + adding verifier selftest
  - checking all arguments in bpf_modify_return_test test programs
  - moved helpers proto gets in tracing_prog_func_proto with attach type check

thanks,
jirka

[1] https://lore.kernel.org/bpf/20211118112455.475349-1-jolsa@kernel.org/
====================

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2 years agoselftests/bpf: Add tests for get_func_[arg|ret|arg_cnt] helpers
Jiri Olsa [Wed, 8 Dec 2021 19:32:45 +0000 (20:32 +0100)]
selftests/bpf: Add tests for get_func_[arg|ret|arg_cnt] helpers

Adding tests for get_func_[arg|ret|arg_cnt] helpers.
Using these helpers in fentry/fexit/fmod_ret programs.

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20211208193245.172141-6-jolsa@kernel.org
2 years agobpf: Add get_func_[arg|ret|arg_cnt] helpers
Jiri Olsa [Wed, 8 Dec 2021 19:32:44 +0000 (20:32 +0100)]
bpf: Add get_func_[arg|ret|arg_cnt] helpers

Adding following helpers for tracing programs:

Get n-th argument of the traced function:
  long bpf_get_func_arg(void *ctx, u32 n, u64 *value)

Get return value of the traced function:
  long bpf_get_func_ret(void *ctx, u64 *value)

Get arguments count of the traced function:
  long bpf_get_func_arg_cnt(void *ctx)

The trampoline now stores number of arguments on ctx-8
address, so it's easy to verify argument index and find
return value argument's position.

Moving function ip address on the trampoline stack behind
the number of functions arguments, so it's now stored on
ctx-16 address if it's needed.

All helpers above are inlined by verifier.

Also bit unrelated small change - using newly added function
bpf_prog_has_trampoline in check_get_func_ip.

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20211208193245.172141-5-jolsa@kernel.org
2 years agobpf, x64: Replace some stack_size usage with offset variables
Jiri Olsa [Wed, 8 Dec 2021 19:32:43 +0000 (20:32 +0100)]
bpf, x64: Replace some stack_size usage with offset variables

As suggested by Andrii, adding variables for registers and ip
address offsets, which makes the code more clear, rather than
abusing single stack_size variable for everything.

Also describing the stack layout in the comment.

There is no function change.

Suggested-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Acked-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20211208193245.172141-4-jolsa@kernel.org
2 years agoselftests/bpf: Add test to access int ptr argument in tracing program
Jiri Olsa [Wed, 8 Dec 2021 19:32:42 +0000 (20:32 +0100)]
selftests/bpf: Add test to access int ptr argument in tracing program

Adding verifier test for accessing int pointer argument in
tracing programs.

The test program loads 2nd argument of bpf_modify_return_test
function which is int pointer and checks that verifier allows
that.

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20211208193245.172141-3-jolsa@kernel.org
2 years agobpf: Allow access to int pointer arguments in tracing programs
Jiri Olsa [Wed, 8 Dec 2021 19:32:41 +0000 (20:32 +0100)]
bpf: Allow access to int pointer arguments in tracing programs

Adding support to access arguments with int pointer arguments
in tracing programs.

Currently we allow tracing programs to access only pointers to
string (char pointer), void pointers and pointers to structs.

If we try to access argument which is pointer to int, verifier
will fail to load the program with;

  R1 type=ctx expected=fp
  ; int BPF_PROG(fmod_ret_test, int _a, __u64 _b, int _ret)
  0: (bf) r6 = r1
  ; int BPF_PROG(fmod_ret_test, int _a, __u64 _b, int _ret)
  1: (79) r9 = *(u64 *)(r6 +8)
  func 'bpf_modify_return_test' arg1 type INT is not a struct

There is no harm for the program to access int pointer argument.
We are already doing that for string pointer, which is pointer
to int with 1 byte size.

Changing the is_string_ptr to generic integer check and renaming
it to btf_type_is_int.

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20211208193245.172141-2-jolsa@kernel.org
2 years agolibbpf: Don't validate TYPE_ID relo's original imm value
Andrii Nakryiko [Mon, 13 Dec 2021 01:07:06 +0000 (17:07 -0800)]
libbpf: Don't validate TYPE_ID relo's original imm value

During linking, type IDs in the resulting linked BPF object file can
change, and so ldimm64 instructions corresponding to
BPF_CORE_TYPE_ID_TARGET and BPF_CORE_TYPE_ID_LOCAL CO-RE relos can get
their imm value out of sync with actual CO-RE relocation information
that's updated by BPF linker properly during linking process.

We could teach BPF linker to adjust such instructions, but it feels
a bit too much for linker to re-implement good chunk of
bpf_core_patch_insns logic just for this. This is a redundant safety
check for TYPE_ID relocations, as the real validation is in matching
CO-RE specs, so if that works fine, it's very unlikely that there is
something wrong with the instruction itself.

So, instead, teach libbpf (and kernel) to ignore insn->imm for
BPF_CORE_TYPE_ID_TARGET and BPF_CORE_TYPE_ID_LOCAL relos.

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20211213010706.100231-1-andrii@kernel.org
2 years agoselftests/bpf: Remove last bpf_create_map_xattr from test_verifier
Andrii Nakryiko [Sun, 12 Dec 2021 19:13:41 +0000 (11:13 -0800)]
selftests/bpf: Remove last bpf_create_map_xattr from test_verifier

bpf_create_map_xattr() call was reintroduced after merging bpf tree into
bpf-next tree. Convert the last instance into bpf_map_create() call.

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20211212191341.2529573-1-andrii@kernel.org
2 years agobpf: Silence coverity false positive warning.
Alexei Starovoitov [Sun, 12 Dec 2021 02:08:19 +0000 (18:08 -0800)]
bpf: Silence coverity false positive warning.

Coverity issued the following warning:
6685            cands = bpf_core_add_cands(cands, main_btf, 1);
6686            if (IS_ERR(cands))
>>>     CID 1510300:    (RETURN_LOCAL)
>>>     Returning pointer "cands" which points to local variable "local_cand".
6687                    return cands;

It's a false positive.
Add ERR_CAST() to silence it.

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2 years agobpf: Use kmemdup() to replace kmalloc + memcpy
Jiapeng Chong [Thu, 9 Dec 2021 06:21:22 +0000 (14:21 +0800)]
bpf: Use kmemdup() to replace kmalloc + memcpy

Eliminate the follow coccicheck warning:

./kernel/bpf/btf.c:6537:13-20: WARNING opportunity for kmemdup.

Reported-by: Abaci Robot <abaci@linux.alibaba.com>
Signed-off-by: Jiapeng Chong <jiapeng.chong@linux.alibaba.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/1639030882-92383-1-git-send-email-jiapeng.chong@linux.alibaba.com
2 years agoMerge branch 'introduce bpf_strncmp() helper'
Alexei Starovoitov [Sun, 12 Dec 2021 01:40:23 +0000 (17:40 -0800)]
Merge branch 'introduce bpf_strncmp() helper'

Hou Tao says:

====================

Hi,

The motivation for introducing bpf_strncmp() helper comes from
two aspects:

(1) clang doesn't always replace strncmp() automatically
In tracing program, sometimes we need to using a home-made
strncmp() to check whether or not the file name is expected.

(2) the performance of home-made strncmp is not so good
As shown in the benchmark in patch #4, the performance of
bpf_strncmp() helper is 18% or 33% better than home-made strncmp()
under x86-64 or arm64 when the compared string length is 64. When
the string length grows to 4095, the performance win will be
179% or 600% under x86-64 or arm64.

Any comments are welcome.
Regards,
Tao

Change Log:
v2:
 * rebased on bpf-next
 * drop patch "selftests/bpf: factor out common helpers for benchmarks"
   (suggested by Andrii)
 * remove unnecessary inline functions and add comments for programs which
   will be rejected by verifier in patch 4 (suggested by Andrii)
 * rename variables used in will-fail programs to clarify the purposes.

v1: https://lore.kernel.org/bpf/20211130142215.1237217-1-houtao1@huawei.com
 * change API to bpf_strncmp(const char *s1, u32 s1_sz, const char *s2)
 * add benchmark refactor and benchmark between bpf_strncmp() and strncmp()

RFC: https://lore.kernel.org/bpf/20211106132822.1396621-1-houtao1@huawei.com/
====================

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2 years agoselftests/bpf: Add test cases for bpf_strncmp()
Hou Tao [Fri, 10 Dec 2021 14:16:52 +0000 (22:16 +0800)]
selftests/bpf: Add test cases for bpf_strncmp()

Four test cases are added:
(1) ensure the return value is expected
(2) ensure no const string size is rejected
(3) ensure writable target is rejected
(4) ensure no null-terminated target is rejected

Signed-off-by: Hou Tao <houtao1@huawei.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20211210141652.877186-5-houtao1@huawei.com
2 years agoselftests/bpf: Add benchmark for bpf_strncmp() helper
Hou Tao [Fri, 10 Dec 2021 14:16:51 +0000 (22:16 +0800)]
selftests/bpf: Add benchmark for bpf_strncmp() helper

Add benchmark to compare the performance between home-made strncmp()
in bpf program and bpf_strncmp() helper. In summary, the performance
win of bpf_strncmp() under x86-64 is greater than 18% when the compared
string length is greater than 64, and is 179% when the length is 4095.
Under arm64 the performance win is even bigger: 33% when the length
is greater than 64 and 600% when the length is 4095.

The following is the details:

no-helper-X: use home-made strncmp() to compare X-sized string
helper-Y: use bpf_strncmp() to compare Y-sized string

Under x86-64:

no-helper-1          3.504 ± 0.000M/s (drops 0.000 ± 0.000M/s)
helper-1             3.347 ± 0.001M/s (drops 0.000 ± 0.000M/s)

no-helper-8          3.357 ± 0.001M/s (drops 0.000 ± 0.000M/s)
helper-8             3.307 ± 0.001M/s (drops 0.000 ± 0.000M/s)

no-helper-32         3.064 ± 0.000M/s (drops 0.000 ± 0.000M/s)
helper-32            3.253 ± 0.001M/s (drops 0.000 ± 0.000M/s)

no-helper-64         2.563 ± 0.001M/s (drops 0.000 ± 0.000M/s)
helper-64            3.040 ± 0.001M/s (drops 0.000 ± 0.000M/s)

no-helper-128        1.975 ± 0.000M/s (drops 0.000 ± 0.000M/s)
helper-128           2.641 ± 0.000M/s (drops 0.000 ± 0.000M/s)

no-helper-512        0.759 ± 0.000M/s (drops 0.000 ± 0.000M/s)
helper-512           1.574 ± 0.000M/s (drops 0.000 ± 0.000M/s)

no-helper-2048       0.329 ± 0.000M/s (drops 0.000 ± 0.000M/s)
helper-2048          0.602 ± 0.000M/s (drops 0.000 ± 0.000M/s)

no-helper-4095       0.117 ± 0.000M/s (drops 0.000 ± 0.000M/s)
helper-4095          0.327 ± 0.000M/s (drops 0.000 ± 0.000M/s)

Under arm64:

no-helper-1          2.806 ± 0.004M/s (drops 0.000 ± 0.000M/s)
helper-1             2.819 ± 0.002M/s (drops 0.000 ± 0.000M/s)

no-helper-8          2.797 ± 0.109M/s (drops 0.000 ± 0.000M/s)
helper-8             2.786 ± 0.025M/s (drops 0.000 ± 0.000M/s)

no-helper-32         2.399 ± 0.011M/s (drops 0.000 ± 0.000M/s)
helper-32            2.703 ± 0.002M/s (drops 0.000 ± 0.000M/s)

no-helper-64         2.020 ± 0.015M/s (drops 0.000 ± 0.000M/s)
helper-64            2.702 ± 0.073M/s (drops 0.000 ± 0.000M/s)

no-helper-128        1.604 ± 0.001M/s (drops 0.000 ± 0.000M/s)
helper-128           2.516 ± 0.002M/s (drops 0.000 ± 0.000M/s)

no-helper-512        0.699 ± 0.000M/s (drops 0.000 ± 0.000M/s)
helper-512           2.106 ± 0.003M/s (drops 0.000 ± 0.000M/s)

no-helper-2048       0.215 ± 0.000M/s (drops 0.000 ± 0.000M/s)
helper-2048          1.223 ± 0.003M/s (drops 0.000 ± 0.000M/s)

no-helper-4095       0.112 ± 0.000M/s (drops 0.000 ± 0.000M/s)
helper-4095          0.796 ± 0.000M/s (drops 0.000 ± 0.000M/s)

Signed-off-by: Hou Tao <houtao1@huawei.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20211210141652.877186-4-houtao1@huawei.com
2 years agoselftests/bpf: Fix checkpatch error on empty function parameter
Hou Tao [Fri, 10 Dec 2021 14:16:50 +0000 (22:16 +0800)]
selftests/bpf: Fix checkpatch error on empty function parameter

Fix checkpatch error: "ERROR: Bad function definition - void foo()
should probably be void foo(void)". Most replacements are done by
the following command:

  sed -i 's#\([a-z]\)()$#\1(void)#g' testing/selftests/bpf/benchs/*.c

Signed-off-by: Hou Tao <houtao1@huawei.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20211210141652.877186-3-houtao1@huawei.com
2 years agobpf: Add bpf_strncmp helper
Hou Tao [Fri, 10 Dec 2021 14:16:49 +0000 (22:16 +0800)]
bpf: Add bpf_strncmp helper

The helper compares two strings: one string is a null-terminated
read-only string, and another string has const max storage size
but doesn't need to be null-terminated. It can be used to compare
file name in tracing or LSM program.

Signed-off-by: Hou Tao <houtao1@huawei.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20211210141652.877186-2-houtao1@huawei.com
2 years agolibbpf: Fix gen_loader assumption on number of programs.
Alexei Starovoitov [Sun, 12 Dec 2021 01:16:19 +0000 (17:16 -0800)]
libbpf: Fix gen_loader assumption on number of programs.

libbpf's obj->nr_programs includes static and global functions. That number
could be higher than the actual number of bpf programs going be loaded by
gen_loader. Passing larger nr_programs to bpf_gen__init() doesn't hurt. Those
exra stack slots will stay as zero. bpf_gen__finish() needs to check that
actual number of progs that gen_loader saw is less than or equal to
obj->nr_programs.

Fixes: ba05fd36b851 ("libbpf: Perform map fd cleanup for gen_loader in case of error")
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2 years agoMerge branch 'add-fdma-support-on-ocelot-switch-driver'
Jakub Kicinski [Sat, 11 Dec 2021 04:56:59 +0000 (20:56 -0800)]
Merge branch 'add-fdma-support-on-ocelot-switch-driver'

Clément Léger says:

====================
Add FDMA support on ocelot switch driver

This series adds support for the Frame DMA present on the VSC7514
switch. The FDMA is able to extract and inject packets on the various
ethernet interfaces present on the switch.
====================

Link: https://lore.kernel.org/r/20211209154911.3152830-1-clement.leger@bootlin.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 years agonet: ocelot: add FDMA support
Clément Léger [Thu, 9 Dec 2021 15:49:11 +0000 (16:49 +0100)]
net: ocelot: add FDMA support

Ethernet frames can be extracted or injected autonomously to or from
the device’s DDR3/DDR3L memory and/or PCIe memory space. Linked list
data structures in memory are used for injecting or extracting Ethernet
frames. The FDMA generates interrupts when frame extraction or
injection is done and when the linked lists need updating.

The FDMA is shared between all the ethernet ports of the switch and
uses a linked list of descriptors (DCB) to inject and extract packets.
Before adding descriptors, the FDMA channels must be stopped. It would
be inefficient to do that each time a descriptor would be added so the
channels are restarted only once they stopped.

Both channels uses ring-like structure to feed the DCBs to the FDMA.
head and tail are never touched by hardware and are completely handled
by the driver. On top of that, page recycling has been added and is
mostly taken from gianfar driver.

Reviewed-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Co-developed-by: Alexandre Belloni <alexandre.belloni@bootlin.com>
Signed-off-by: Alexandre Belloni <alexandre.belloni@bootlin.com>
Signed-off-by: Clément Léger <clement.leger@bootlin.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 years agonet: ocelot: add support for ndo_change_mtu
Clément Léger [Thu, 9 Dec 2021 15:49:10 +0000 (16:49 +0100)]
net: ocelot: add support for ndo_change_mtu

This commit adds support for changing MTU for the ocelot register based
interface. For ocelot, JUMBO frame size can be set up to 25000 bytes
but has been set to 9000 which is a saner value and allows for maximum
gain of performance with FDMA.

Reviewed-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: Clément Léger <clement.leger@bootlin.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 years agonet: ocelot: add and export ocelot_ptp_rx_timestamp()
Clément Léger [Thu, 9 Dec 2021 15:49:09 +0000 (16:49 +0100)]
net: ocelot: add and export ocelot_ptp_rx_timestamp()

In order to support PTP in FDMA, PTP handling code is needed. Since
this is the same as for register-based extraction, export it with
a new ocelot_ptp_rx_timestamp() function.

Reviewed-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: Clément Léger <clement.leger@bootlin.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 years agonet: ocelot: export ocelot_ifh_port_set() to setup IFH
Clément Léger [Thu, 9 Dec 2021 15:49:08 +0000 (16:49 +0100)]
net: ocelot: export ocelot_ifh_port_set() to setup IFH

FDMA will need this code to prepare the injection frame header when
sending SKBs. Move this code into ocelot_ifh_port_set() and add
conditional IFH setting for vlan and rew op if they are not set.

Reviewed-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: Clément Léger <clement.leger@bootlin.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 years agoMerge branch 'net-wwan-iosm-improvements'
Jakub Kicinski [Sat, 11 Dec 2021 04:53:58 +0000 (20:53 -0800)]
Merge branch 'net-wwan-iosm-improvements'

M Chetan Kumar says:

====================
net: wwan: iosm: improvements

This patch series brings in IOSM driver improvments.

PATCH1: Set tx queue len.
PATCH2: Release data channel if there is no active IP session.
PATCH3: Removes dead code.
PATCH4: Correct open parenthesis alignment.
====================

Link: https://lore.kernel.org/r/20211209143230.3054755-1-m.chetan.kumar@linux.intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 years agonet: wwan: iosm: correct open parenthesis alignment
M Chetan Kumar [Thu, 9 Dec 2021 14:32:30 +0000 (20:02 +0530)]
net: wwan: iosm: correct open parenthesis alignment

Fix checkpatch warning in iosm_ipc_mmio.c
- Alignment should match open parenthesis

Signed-off-by: M Chetan Kumar <m.chetan.kumar@linux.intel.com>
Reviewed-by: Sergey Ryazanov <ryazanov.s.a@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 years agonet: wwan: iosm: removed unused function decl
M Chetan Kumar [Thu, 9 Dec 2021 14:32:29 +0000 (20:02 +0530)]
net: wwan: iosm: removed unused function decl

ipc_wwan_tx_flowctrl() is declared in iosm_ipc_wwan.h but is
not defined.

Removed the dead code.

Signed-off-by: M Chetan Kumar <m.chetan.kumar@linux.intel.com>
Reviewed-by: Sergey Ryazanov <ryazanov.s.a@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 years agonet: wwan: iosm: release data channel in case no active IP session
M Chetan Kumar [Thu, 9 Dec 2021 14:32:28 +0000 (20:02 +0530)]
net: wwan: iosm: release data channel in case no active IP session

If there is no active IP session (interface up & running) then
release the data channel.

Use nr_sessions variable to track current active IP sessions.
If the count drops to 0, then send pipe close ctrl message to
release the data channel.

Signed-off-by: M Chetan Kumar <m.chetan.kumar@linux.intel.com>
Reviewed-by: Sergey Ryazanov <ryazanov.s.a@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 years agonet: wwan: iosm: set tx queue len
M Chetan Kumar [Thu, 9 Dec 2021 14:32:27 +0000 (20:02 +0530)]
net: wwan: iosm: set tx queue len

Set wwan net dev tx queue len to DEFAULT_TX_QUEUE_LEN.

Signed-off-by: M Chetan Kumar <m.chetan.kumar@linux.intel.com>
Reviewed-by: Sergey Ryazanov <ryazanov.s.a@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 years agonet: ocelot: fix missed include in the vsc7514_regs.h file
Colin Foster [Thu, 9 Dec 2021 07:40:10 +0000 (23:40 -0800)]
net: ocelot: fix missed include in the vsc7514_regs.h file

commit 32ecd22ba60b ("net: mscc: ocelot: split register definitions to a
separate file") left out an include for <soc/mscc/ocelot_vcap.h>. It was
missed because the only consumer was ocelot_vsc7514.h, which already
included ocelot_vcap.

Fixes: 32ecd22ba60b ("net: mscc: ocelot: split register definitions to a separate file")
Signed-off-by: Colin Foster <colin.foster@in-advantage.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Link: https://lore.kernel.org/r/20211209074010.1813010-1-colin.foster@in-advantage.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 years agonet: bna: Update supported link modes
Erik Ekman [Wed, 8 Dec 2021 23:00:22 +0000 (00:00 +0100)]
net: bna: Update supported link modes

The BR-series installation guide from https://driverdownloads.qlogic.com/
mentions the cards support 10Gbase-SR/LR as well as direct attach cables.

The cards only have SFP+ ports, so 10000baseT is not the right mode.
Switch to using more specific link modes added in commit 5711a98221443
("net: ethtool: add support for 1000BaseX and missing 10G link modes").

Only compile tested.

Signed-off-by: Erik Ekman <erik@kryo.se>
Link: https://lore.kernel.org/r/20211208230022.153496-1-erik@kryo.se
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 years agosock: Use sock_owned_by_user_nocheck() instead of sk_lock.owned.
Kuniyuki Iwashima [Wed, 8 Dec 2021 06:21:58 +0000 (15:21 +0900)]
sock: Use sock_owned_by_user_nocheck() instead of sk_lock.owned.

This patch moves sock_release_ownership() down in include/net/sock.h and
replaces some sk_lock.owned tests with sock_owned_by_user_nocheck().

Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.co.jp>
Link: https://lore.kernel.org/r/20211208062158.54132-1-kuniyu@amazon.co.jp
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 years agoMerge https://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf-next
Jakub Kicinski [Fri, 10 Dec 2021 23:56:10 +0000 (15:56 -0800)]
Merge https://git./linux/kernel/git/bpf/bpf-next

Andrii Nakryiko says:

====================
bpf-next 2021-12-10 v2

We've added 115 non-merge commits during the last 26 day(s) which contain
a total of 182 files changed, 5747 insertions(+), 2564 deletions(-).

The main changes are:

1) Various samples fixes, from Alexander Lobakin.

2) BPF CO-RE support in kernel and light skeleton, from Alexei Starovoitov.

3) A batch of new unified APIs for libbpf, logging improvements, version
   querying, etc. Also a batch of old deprecations for old APIs and various
   bug fixes, in preparation for libbpf 1.0, from Andrii Nakryiko.

4) BPF documentation reorganization and improvements, from Christoph Hellwig
   and Dave Tucker.

5) Support for declarative initialization of BPF_MAP_TYPE_PROG_ARRAY in
   libbpf, from Hengqi Chen.

6) Verifier log fixes, from Hou Tao.

7) Runtime-bounded loops support with bpf_loop() helper, from Joanne Koong.

8) Extend branch record capturing to all platforms that support it,
   from Kajol Jain.

9) Light skeleton codegen improvements, from Kumar Kartikeya Dwivedi.

10) bpftool doc-generating script improvements, from Quentin Monnet.

11) Two libbpf v0.6 bug fixes, from Shuyi Cheng and Vincent Minet.

12) Deprecation warning fix for perf/bpf_counter, from Song Liu.

13) MAX_TAIL_CALL_CNT unification and MIPS build fix for libbpf,
    from Tiezhu Yang.

14) BTF_KING_TYPE_TAG follow-up fixes, from Yonghong Song.

15) Selftests fixes and improvements, from Ilya Leoshkevich, Jean-Philippe
    Brucker, Jiri Olsa, Maxim Mikityanskiy, Tirthendu Sarkar, Yucong Sun,
    and others.

* https://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf-next: (115 commits)
  libbpf: Add "bool skipped" to struct bpf_map
  libbpf: Fix typo in btf__dedup@LIBBPF_0.0.2 definition
  bpftool: Switch bpf_object__load_xattr() to bpf_object__load()
  selftests/bpf: Remove the only use of deprecated bpf_object__load_xattr()
  selftests/bpf: Add test for libbpf's custom log_buf behavior
  selftests/bpf: Replace all uses of bpf_load_btf() with bpf_btf_load()
  libbpf: Deprecate bpf_object__load_xattr()
  libbpf: Add per-program log buffer setter and getter
  libbpf: Preserve kernel error code and remove kprobe prog type guessing
  libbpf: Improve logging around BPF program loading
  libbpf: Allow passing user log setting through bpf_object_open_opts
  libbpf: Allow passing preallocated log_buf when loading BTF into kernel
  libbpf: Add OPTS-based bpf_btf_load() API
  libbpf: Fix bpf_prog_load() log_buf logic for log_level 0
  samples/bpf: Remove unneeded variable
  bpf: Remove redundant assignment to pointer t
  selftests/bpf: Fix a compilation warning
  perf/bpf_counter: Use bpf_map_create instead of bpf_create_map
  samples: bpf: Fix 'unknown warning group' build warning on Clang
  samples: bpf: Fix xdp_sample_user.o linking with Clang
  ...
====================

Link: https://lore.kernel.org/r/20211210234746.2100561-1-andrii@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 years agolibbpf: Add "bool skipped" to struct bpf_map
Shuyi Cheng [Fri, 10 Dec 2021 09:39:57 +0000 (17:39 +0800)]
libbpf: Add "bool skipped" to struct bpf_map

Fix error: "failed to pin map: Bad file descriptor, path:
/sys/fs/bpf/_rodata_str1_1."

In the old kernel, the global data map will not be created, see [0]. So
we should skip the pinning of the global data map to avoid
bpf_object__pin_maps returning error. Therefore, when the map is not
created, we mark “map->skipped" as true and then check during relocation
and during pinning.

Fixes: 16e0c35c6f7a ("libbpf: Load global data maps lazily on legacy kernels")
Signed-off-by: Shuyi Cheng <chengshuyi@linux.alibaba.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
2 years agolibbpf: Fix typo in btf__dedup@LIBBPF_0.0.2 definition
Vincent Minet [Fri, 10 Dec 2021 06:31:12 +0000 (07:31 +0100)]
libbpf: Fix typo in btf__dedup@LIBBPF_0.0.2 definition

The btf__dedup_deprecated name was misspelled in the definition of the
compat symbol for btf__dedup. This leads it to be missing from the
shared library.

This fixes it.

Fixes: 957d350a8b94 ("libbpf: Turn btf_dedup_opts into OPTS-based struct")
Signed-off-by: Vincent Minet <vincent@vincent-minet.net>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20211210063112.80047-1-vincent@vincent-minet.net
2 years agoMerge branch 'Enhance and rework logging controls in libbpf'
Alexei Starovoitov [Thu, 9 Dec 2021 23:40:34 +0000 (15:40 -0800)]
Merge branch 'Enhance and rework logging controls in libbpf'

Andrii Nakryiko says:

====================

Add new open options and per-program setters to control BTF and program
loading log verboseness and allow providing custom log buffers to capture logs
of interest. Note how custom log_buf and log_level are orthogonal, which
matches previous (alas less customizable) behavior of libbpf, even though it
sort of worked by accident: if someone specified log_level = 1 in
bpf_object__load_xattr(), first attempt to load any BPF program resulted in
wasted bpf() syscall with -EINVAL due to !!log_buf != !!log_level. Then on
retry libbpf would allocated log_buffer and try again, after which prog
loading would succeed and libbpf would print verbose program loading log
through its print callback.

This behavior is now documented and made more efficient, not wasting
unnecessary syscall. But additionally, log_level can be controlled globally on
a per-bpf_object level through bpf_object_open_opts, as well as on
a per-program basis with bpf_program__set_log_buf() and
bpf_program__set_log_level() APIs.

Now that we have a more future-proof way to set log_level, deprecate
bpf_object__load_xattr().

v2->v3:
  - added log_buf selftests for bpf_prog_load() and bpf_btf_load();
  - fix !log_buf in bpf_prog_load (John);
  - fix log_level==0 in bpf_btf_load (thanks selftest!);

v1->v2:
  - fix log_level == 0 handling of bpf_prog_load, add as patch #1 (Alexei);
  - add comments explaining log_buf_size overflow prevention (Alexei).
====================

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
2 years agobpftool: Switch bpf_object__load_xattr() to bpf_object__load()
Andrii Nakryiko [Thu, 9 Dec 2021 19:38:40 +0000 (11:38 -0800)]
bpftool: Switch bpf_object__load_xattr() to bpf_object__load()

Switch all the uses of to-be-deprecated bpf_object__load_xattr() into
a simple bpf_object__load() calls with optional log_level passed through
open_opts.kernel_log_level, if -d option is specified.

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20211209193840.1248570-13-andrii@kernel.org
2 years agoselftests/bpf: Remove the only use of deprecated bpf_object__load_xattr()
Andrii Nakryiko [Thu, 9 Dec 2021 19:38:39 +0000 (11:38 -0800)]
selftests/bpf: Remove the only use of deprecated bpf_object__load_xattr()

Switch from bpf_object__load_xattr() to bpf_object__load() and
kernel_log_level in bpf_object_open_opts.

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20211209193840.1248570-12-andrii@kernel.org
2 years agoselftests/bpf: Add test for libbpf's custom log_buf behavior
Andrii Nakryiko [Thu, 9 Dec 2021 19:38:38 +0000 (11:38 -0800)]
selftests/bpf: Add test for libbpf's custom log_buf behavior

Add a selftest that validates that per-program and per-object log_buf
overrides work as expected. Also test same logic for low-level
bpf_prog_load() and bpf_btf_load() APIs.

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20211209193840.1248570-11-andrii@kernel.org
2 years agoselftests/bpf: Replace all uses of bpf_load_btf() with bpf_btf_load()
Andrii Nakryiko [Thu, 9 Dec 2021 19:38:37 +0000 (11:38 -0800)]
selftests/bpf: Replace all uses of bpf_load_btf() with bpf_btf_load()

Switch all selftests uses of to-be-deprecated bpf_load_btf() with
equivalent bpf_btf_load() calls.

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20211209193840.1248570-10-andrii@kernel.org
2 years agolibbpf: Deprecate bpf_object__load_xattr()
Andrii Nakryiko [Thu, 9 Dec 2021 19:38:36 +0000 (11:38 -0800)]
libbpf: Deprecate bpf_object__load_xattr()

Deprecate non-extensible bpf_object__load_xattr() in v0.8 ([0]).

With log_level control through bpf_object_open_opts or
bpf_program__set_log_level(), we are finally at the point where
bpf_object__load_xattr() doesn't provide any functionality that can't be
accessed through other (better) ways. The other feature,
target_btf_path, is also controllable through bpf_object_open_opts.

  [0] Closes: https://github.com/libbpf/libbpf/issues/289

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20211209193840.1248570-9-andrii@kernel.org
2 years agolibbpf: Add per-program log buffer setter and getter
Andrii Nakryiko [Thu, 9 Dec 2021 19:38:35 +0000 (11:38 -0800)]
libbpf: Add per-program log buffer setter and getter

Allow to set user-provided log buffer on a per-program basis ([0]). This
gives great deal of flexibility in terms of which programs are loaded
with logging enabled and where corresponding logs go.

Log buffer set with bpf_program__set_log_buf() overrides kernel_log_buf
and kernel_log_size settings set at bpf_object open time through
bpf_object_open_opts, if any.

Adjust bpf_object_load_prog_instance() logic to not perform own log buf
allocation and load retry if custom log buffer is provided by the user.

  [0] Closes: https://github.com/libbpf/libbpf/issues/418

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20211209193840.1248570-8-andrii@kernel.org
2 years agolibbpf: Preserve kernel error code and remove kprobe prog type guessing
Andrii Nakryiko [Thu, 9 Dec 2021 19:38:34 +0000 (11:38 -0800)]
libbpf: Preserve kernel error code and remove kprobe prog type guessing

Instead of rewriting error code returned by the kernel of prog load with
libbpf-sepcific variants pass through the original error.

There is now also no need to have a backup generic -LIBBPF_ERRNO__LOAD
fallback error as bpf_prog_load() guarantees that errno will be properly
set no matter what.

Also drop a completely outdated and pretty useless BPF_PROG_TYPE_KPROBE
guess logic. It's not necessary and neither it's helpful in modern BPF
applications.

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20211209193840.1248570-7-andrii@kernel.org
2 years agolibbpf: Improve logging around BPF program loading
Andrii Nakryiko [Thu, 9 Dec 2021 19:38:33 +0000 (11:38 -0800)]
libbpf: Improve logging around BPF program loading

Add missing "prog '%s': " prefixes in few places and use consistently
markers for beginning and end of program load logs. Here's an example of
log output:

libbpf: prog 'handler': BPF program load failed: Permission denied
libbpf: -- BEGIN PROG LOAD LOG ---
arg#0 reference type('UNKNOWN ') size cannot be determined: -22
; out1 = in1;
0: (18) r1 = 0xffffc9000cdcc000
2: (61) r1 = *(u32 *)(r1 +0)

...

81: (63) *(u32 *)(r4 +0) = r5
 R1_w=map_value(id=0,off=16,ks=4,vs=20,imm=0) R4=map_value(id=0,off=400,ks=4,vs=16,imm=0)
invalid access to map value, value_size=16 off=400 size=4
R4 min value is outside of the allowed memory range
processed 63 insns (limit 1000000) max_states_per_insn 0 total_states 0 peak_states 0 mark_read 0
 -- END PROG LOAD LOG --
libbpf: failed to load program 'handler'
libbpf: failed to load object 'test_skeleton'

The entire verifier log, including BEGIN and END markers are now always
youtput during a single print callback call. This should make it much
easier to post-process or parse it, if necessary. It's not an explicit
API guarantee, but it can be reasonably expected to stay like that.

Also __bpf_object__open is renamed to bpf_object_open() as it's always
an adventure to find the exact function that implements bpf_object's
open phase, so drop the double underscored and use internal libbpf
naming convention.

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20211209193840.1248570-6-andrii@kernel.org
2 years agolibbpf: Allow passing user log setting through bpf_object_open_opts
Andrii Nakryiko [Thu, 9 Dec 2021 19:38:32 +0000 (11:38 -0800)]
libbpf: Allow passing user log setting through bpf_object_open_opts

Allow users to provide their own custom log_buf, log_size, and log_level
at bpf_object level through bpf_object_open_opts. This log_buf will be
used during BTF loading. Subsequent patch will use same log_buf during
BPF program loading, unless overriden at per-bpf_program level.

When such custom log_buf is provided, libbpf won't be attempting
retrying loading of BTF to try to provide its own log buffer to capture
kernel's error log output. User is responsible to provide big enough
buffer, otherwise they run a risk of getting -ENOSPC error from the
bpf() syscall.

See also comments in bpf_object_open_opts regarding log_level and
log_buf interactions.

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20211209193840.1248570-5-andrii@kernel.org
2 years agolibbpf: Allow passing preallocated log_buf when loading BTF into kernel
Andrii Nakryiko [Thu, 9 Dec 2021 19:38:31 +0000 (11:38 -0800)]
libbpf: Allow passing preallocated log_buf when loading BTF into kernel

Add libbpf-internal btf_load_into_kernel() that allows to pass
preallocated log_buf and custom log_level to be passed into kernel
during BPF_BTF_LOAD call. When custom log_buf is provided,
btf_load_into_kernel() won't attempt an retry with automatically
allocated internal temporary buffer to capture BTF validation log.

It's important to note the relation between log_buf and log_level, which
slightly deviates from stricter kernel logic. From kernel's POV, if
log_buf is specified, log_level has to be > 0, and vice versa. While
kernel has good reasons to request such "sanity, this, in practice, is
a bit unconvenient and restrictive for libbpf's high-level bpf_object APIs.

So libbpf will allow to set non-NULL log_buf and log_level == 0. This is
fine and means to attempt to load BTF without logging requested, but if
it failes, retry the load with custom log_buf and log_level 1. Similar
logic will be implemented for program loading. In practice this means
that users can provide custom log buffer just in case error happens, but
not really request slower verbose logging all the time. This is also
consistent with libbpf behavior when custom log_buf is not set: libbpf
first tries to load everything with log_level=0, and only if error
happens allocates internal log buffer and retries with log_level=1.

Also, while at it, make BTF validation log more obvious and follow the log
pattern libbpf is using for dumping BPF verifier log during
BPF_PROG_LOAD. BTF loading resulting in an error will look like this:

libbpf: BTF loading error: -22
libbpf: -- BEGIN BTF LOAD LOG ---
magic: 0xeb9f
version: 1
flags: 0x0
hdr_len: 24
type_off: 0
type_len: 1040
str_off: 1040
str_len: 2063598257
btf_total_size: 1753
Total section length too long
-- END BTF LOAD LOG --
libbpf: Error loading .BTF into kernel: -22. BTF is optional, ignoring.

This makes it much easier to find relevant parts in libbpf log output.

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20211209193840.1248570-4-andrii@kernel.org
2 years agolibbpf: Add OPTS-based bpf_btf_load() API
Andrii Nakryiko [Thu, 9 Dec 2021 19:38:30 +0000 (11:38 -0800)]
libbpf: Add OPTS-based bpf_btf_load() API

Similar to previous bpf_prog_load() and bpf_map_create() APIs, add
bpf_btf_load() API which is taking optional OPTS struct. Schedule
bpf_load_btf() for deprecation in v0.8 ([0]).

This makes naming consistent with BPF_BTF_LOAD command, sets up an API
for extensibility in the future, moves options parameters (log-related
fields) into optional options, and also allows to pass log_level
directly.

It also removes log buffer auto-allocation logic from low-level API
(consistent with bpf_prog_load() behavior), but preserves a special
treatment of log_level == 0 with non-NULL log_buf, which matches
low-level bpf_prog_load() and high-level libbpf APIs for BTF and program
loading behaviors.

  [0] Closes: https://github.com/libbpf/libbpf/issues/419

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20211209193840.1248570-3-andrii@kernel.org
2 years agolibbpf: Fix bpf_prog_load() log_buf logic for log_level 0
Andrii Nakryiko [Thu, 9 Dec 2021 19:38:29 +0000 (11:38 -0800)]
libbpf: Fix bpf_prog_load() log_buf logic for log_level 0

To unify libbpf APIs behavior w.r.t. log_buf and log_level, fix
bpf_prog_load() to follow the same logic as bpf_btf_load() and
high-level bpf_object__load() API will follow in the subsequent patches:
  - if log_level is 0 and non-NULL log_buf is provided by a user, attempt
    load operation initially with no log_buf and log_level set;
  - if successful, we are done, return new FD;
  - on error, retry the load operation with log_level bumped to 1 and
    log_buf set; this way verbose logging will be requested only when we
    are sure that there is a failure, but will be fast in the
    common/expected success case.

Of course, user can still specify log_level > 0 from the very beginning
to force log collection.

Suggested-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20211209193840.1248570-2-andrii@kernel.org