OSDN Git Service

tomoyo/tomoyo-test1.git
4 years agobpf: Add BPF_FUNC_jiffies64
Martin KaFai Lau [Wed, 22 Jan 2020 23:36:46 +0000 (15:36 -0800)]
bpf: Add BPF_FUNC_jiffies64

This patch adds a helper to read the 64bit jiffies.  It will be used
in a later patch to implement the bpf_cubic.c.

The helper is inlined for jit_requested and 64 BITS_PER_LONG
as the map_gen_lookup().  Other cases could be considered together
with map_gen_lookup() if needed.

Signed-off-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20200122233646.903260-1-kafai@fb.com
4 years agoMerge branch 'bpf-dynamic-relinking'
Daniel Borkmann [Wed, 22 Jan 2020 22:04:53 +0000 (23:04 +0100)]
Merge branch 'bpf-dynamic-relinking'

Alexei Starovoitov says:

====================
The last few month BPF community has been discussing an approach to call
chaining, since exiting bpt_tail_call() mechanism used in production XDP
programs has plenty of downsides. The outcome of these discussion was a
conclusion to implement dynamic re-linking of BPF programs. Where rootlet XDP
program attached to a netdevice can programmatically define a policy of
execution of other XDP programs. Such rootlet would be compiled as normal XDP
program and provide a number of placeholder global functions which later can be
replaced with future XDP programs. BPF trampoline, function by function
verification were building blocks towards that goal. The patch 1 is a final
building block. It introduces dynamic program extensions. A number of
improvements like more flexible function by function verification and better
libbpf api will be implemented in future patches.

v1->v2:
- addressed Andrii's comments
- rebase
====================

Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
4 years agoselftests/bpf: Add tests for program extensions
Alexei Starovoitov [Tue, 21 Jan 2020 00:53:48 +0000 (16:53 -0800)]
selftests/bpf: Add tests for program extensions

Add program extension tests that build on top of fexit_bpf2bpf tests.
Replace three global functions in previously loaded test_pkt_access.c program
with three new implementations:
int get_skb_len(struct __sk_buff *skb);
int get_constant(long val);
int get_skb_ifindex(int val, struct __sk_buff *skb, int var);
New function return the same results as original only if arguments match.

new_get_skb_ifindex() demonstrates that 'skb' argument doesn't have to be first
and only argument of BPF program. All normal skb based accesses are available.

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Toke Høiland-Jørgensen <toke@redhat.com>
Link: https://lore.kernel.org/bpf/20200121005348.2769920-4-ast@kernel.org
4 years agolibbpf: Add support for program extensions
Alexei Starovoitov [Tue, 21 Jan 2020 00:53:47 +0000 (16:53 -0800)]
libbpf: Add support for program extensions

Add minimal support for program extensions. bpf_object_open_opts() needs to be
called with attach_prog_fd = target_prog_fd and BPF program extension needs to
have in .c file section definition like SEC("freplace/func_to_be_replaced").
libbpf will search for "func_to_be_replaced" in the target_prog_fd's BTF and
will pass it in attach_btf_id to the kernel. This approach works for tests, but
more compex use case may need to request function name (and attach_btf_id that
kernel sees) to be more dynamic. Such API will be added in future patches.

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Toke Høiland-Jørgensen <toke@redhat.com>
Link: https://lore.kernel.org/bpf/20200121005348.2769920-3-ast@kernel.org
4 years agobpf: Introduce dynamic program extensions
Alexei Starovoitov [Tue, 21 Jan 2020 00:53:46 +0000 (16:53 -0800)]
bpf: Introduce dynamic program extensions

Introduce dynamic program extensions. The users can load additional BPF
functions and replace global functions in previously loaded BPF programs while
these programs are executing.

Global functions are verified individually by the verifier based on their types only.
Hence the global function in the new program which types match older function can
safely replace that corresponding function.

This new function/program is called 'an extension' of old program. At load time
the verifier uses (attach_prog_fd, attach_btf_id) pair to identify the function
to be replaced. The BPF program type is derived from the target program into
extension program. Technically bpf_verifier_ops is copied from target program.
The BPF_PROG_TYPE_EXT program type is a placeholder. It has empty verifier_ops.
The extension program can call the same bpf helper functions as target program.
Single BPF_PROG_TYPE_EXT type is used to extend XDP, SKB and all other program
types. The verifier allows only one level of replacement. Meaning that the
extension program cannot recursively extend an extension. That also means that
the maximum stack size is increasing from 512 to 1024 bytes and maximum
function nesting level from 8 to 16. The programs don't always consume that
much. The stack usage is determined by the number of on-stack variables used by
the program. The verifier could have enforced 512 limit for combined original
plus extension program, but it makes for difficult user experience. The main
use case for extensions is to provide generic mechanism to plug external
programs into policy program or function call chaining.

BPF trampoline is used to track both fentry/fexit and program extensions
because both are using the same nop slot at the beginning of every BPF
function. Attaching fentry/fexit to a function that was replaced is not
allowed. The opposite is true as well. Replacing a function that currently
being analyzed with fentry/fexit is not allowed. The executable page allocated
by BPF trampoline is not used by program extensions. This inefficiency will be
optimized in future patches.

Function by function verification of global function supports scalars and
pointer to context only. Hence program extensions are supported for such class
of global functions only. In the future the verifier will be extended with
support to pointers to structures, arrays with sizes, etc.

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Toke Høiland-Jørgensen <toke@redhat.com>
Link: https://lore.kernel.org/bpf/20200121005348.2769920-2-ast@kernel.org
4 years agobpf, btf: Always output invariant hit in pahole DWARF to BTF transform
Chris Down [Wed, 22 Jan 2020 00:01:10 +0000 (00:01 +0000)]
bpf, btf: Always output invariant hit in pahole DWARF to BTF transform

When trying to compile with CONFIG_DEBUG_INFO_BTF enabled, I got this
error:

    % make -s
    Failed to generate BTF for vmlinux
    Try to disable CONFIG_DEBUG_INFO_BTF
    make[3]: *** [vmlinux] Error 1

Compiling again without -s shows the true error (that pahole is
missing), but since this is fatal, we should show the error
unconditionally on stderr as well, not silence it using the `info`
function. With this patch:

    % make -s
    BTF: .tmp_vmlinux.btf: pahole (pahole) is not available
    Failed to generate BTF for vmlinux
    Try to disable CONFIG_DEBUG_INFO_BTF
    make[3]: *** [vmlinux] Error 1

Signed-off-by: Chris Down <chris@chrisdown.name>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Link: https://lore.kernel.org/bpf/20200122000110.GA310073@chrisdown.name
4 years agoselftests/bpf: Build urandom_read with LDFLAGS and LDLIBS
Daniel Díaz [Wed, 22 Jan 2020 16:44:24 +0000 (17:44 +0100)]
selftests/bpf: Build urandom_read with LDFLAGS and LDLIBS

During cross-compilation, it was discovered that LDFLAGS and
LDLIBS were not being used while building binaries, leading
to defaults which were not necessarily correct.

OpenEmbedded reported this kind of problem:

  ERROR: QA Issue: No GNU_HASH in the ELF binary [...], didn't pass LDFLAGS?

Signed-off-by: Daniel Díaz <daniel.diaz@linaro.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: John Fastabend <john.fastabend@gmail.com>
4 years agobpf: Fix error path under memory pressure
Alexei Starovoitov [Wed, 22 Jan 2020 02:41:38 +0000 (18:41 -0800)]
bpf: Fix error path under memory pressure

Restore the 'if (env->cur_state)' check that was incorrectly removed during
code move. Under memory pressure env->cur_state can be freed and zeroed inside
do_check(). Hence the check is necessary.

Fixes: 51c39bb1d5d1 ("bpf: Introduce function-by-function verification")
Reported-by: syzbot+b296579ba5015704d9fa@syzkaller.appspotmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Song Liu <songliubraving@fb.com>
Link: https://lore.kernel.org/bpf/20200122024138.3385590-1-ast@kernel.org
4 years agobpf: Fix trampoline usage in preempt
Alexei Starovoitov [Tue, 21 Jan 2020 03:22:31 +0000 (19:22 -0800)]
bpf: Fix trampoline usage in preempt

Though the second half of trampoline page is unused a task could be
preempted in the middle of the first half of trampoline and two
updates to trampoline would change the code from underneath the
preempted task. Hence wait for tasks to voluntarily schedule or go
to userspace. Add similar wait before freeing the trampoline.

Fixes: fec56f5890d9 ("bpf: Introduce BPF trampoline")
Reported-by: Jann Horn <jannh@google.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Paul E. McKenney <paulmck@kernel.org>
Link: https://lore.kernel.org/bpf/20200121032231.3292185-1-ast@kernel.org
4 years agoxsk, net: Make sock_def_readable() have external linkage
Björn Töpel [Mon, 20 Jan 2020 09:29:17 +0000 (10:29 +0100)]
xsk, net: Make sock_def_readable() have external linkage

XDP sockets use the default implementation of struct sock's
sk_data_ready callback, which is sock_def_readable(). This function
is called in the XDP socket fast-path, and involves a retpoline. By
letting sock_def_readable() have external linkage, and being called
directly, the retpoline can be avoided.

Signed-off-by: Björn Töpel <bjorn.topel@intel.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20200120092917.13949-1-bjorn.topel@gmail.com
4 years agobpf: don't bother with getname/kern_path - use user_path_at
Al Viro [Mon, 20 Jan 2020 23:28:58 +0000 (23:28 +0000)]
bpf: don't bother with getname/kern_path - use user_path_at

kernel/bpf/inode.c misuses kern_path...() - it's much simpler (and
more efficient, on top of that) to use user_path...() counterparts
rather than bothering with doing getname() manually.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20200120232858.GF8904@ZenIV.linux.org.uk
4 years agoMerge branch 'libbpf-include-path'
Alexei Starovoitov [Tue, 21 Jan 2020 00:37:46 +0000 (16:37 -0800)]
Merge branch 'libbpf-include-path'

Toke Høiland-Jørgensen says:

====================
We are currently being somewhat inconsistent with the libbpf include paths,
which makes it difficult to move files from the kernel into an external
libbpf-using project without adjusting include paths.

Having the bpf/ subdir of $INCLUDEDIR in the include path has never been a
requirement for building against libbpf before, and indeed the libbpf pkg-config
file doesn't include it. So let's make all libbpf includes across the kernel
tree use the bpf/ prefix in their includes. Since bpftool skeleton generation
emits code with a libbpf include, this also ensures that those can be used in
existing external projects using the regular pkg-config include path.

This turns out to be a somewhat invasive change in the number of files touched;
however, the actual changes to files are fairly trivial (most of them are simply
made with 'sed'). The series is split to make the change for one tool subdir at
a time, while trying not to break the build along the way. It is structured like
this:

- Patch 1-3: Trivial fixes to Makefiles for issues I discovered while changing
  the include paths.

- Patch 4-8: Change the include directives to use the bpf/ prefix, and updates
  Makefiles to make sure tools/lib/ is part of the include path, but without
  removing tools/lib/bpf

- Patch 9-11: Remove tools/lib/bpf from include paths to make sure we don't
  inadvertently re-introduce includes without the bpf/ prefix.

Changelog:

v5:
  - Combine the libbpf build rules in selftests Makefile (using Andrii's
    suggestion for a make rule).
  - Re-use self-tests libbpf build for runqslower (new patch 10)
  - Formatting fixes

v4:
  - Move runqslower error on missing BTF into make rule
  - Make sure we don't always force a rebuild selftests
  - Rebase on latest bpf-next (dropping patch 11)

v3:
  - Don't add the kernel build dir to the runqslower Makefile, pass it in from
    selftests instead.
  - Use libbpf's 'make install_headers' in selftests instead of trying to
    generate bpf_helper_defs.h in-place (to also work on read-only filesystems).
  - Use a scratch builddir for both libbpf and bpftool when building in selftests.
  - Revert bpf_helpers.h to quoted include instead of angled include with a bpf/
    prefix.
  - Fix a few style nits from Andrii

v2:
  - Do a full cleanup of libbpf includes instead of just changing the
    bpf_helper_defs.h include.
====================

Acked-by: Andrii Nakryiko <andriin@fb.com>
Tested-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
4 years agoselftests: Refactor build to remove tools/lib/bpf from include path
Toke Høiland-Jørgensen [Mon, 20 Jan 2020 13:06:52 +0000 (14:06 +0100)]
selftests: Refactor build to remove tools/lib/bpf from include path

To make sure no new files are introduced that doesn't include the bpf/
prefix in its #include, remove tools/lib/bpf from the include path
entirely.

Instead, we introduce a new header files directory under the scratch tools/
dir, and add a rule to run the 'install_headers' rule from libbpf to have a
full set of consistent libbpf headers in $(OUTPUT)/tools/include/bpf, and
then use $(OUTPUT)/tools/include as the include path for selftests.

For consistency we also make sure we put all the scratch build files from
other bpftool and libbpf into tools/build/, so everything stays within
selftests/.

Signed-off-by: Toke Høiland-Jørgensen <toke@redhat.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/157952561246.1683545.2762245552022369203.stgit@toke.dk
4 years agorunsqslower: Support user-specified libbpf include and object paths
Toke Høiland-Jørgensen [Mon, 20 Jan 2020 13:06:51 +0000 (14:06 +0100)]
runsqslower: Support user-specified libbpf include and object paths

This adds support for specifying the libbpf include and object paths as
arguments to the runqslower Makefile, to support reusing the libbpf version
built as part of the selftests.

Signed-off-by: Toke Høiland-Jørgensen <toke@redhat.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/157952561135.1683545.5660339645093141381.stgit@toke.dk
4 years agotools/runqslower: Remove tools/lib/bpf from include path
Toke Høiland-Jørgensen [Mon, 20 Jan 2020 13:06:50 +0000 (14:06 +0100)]
tools/runqslower: Remove tools/lib/bpf from include path

Since we are now consistently using the bpf/ prefix on #include directives,
we don't need to include tools/lib/bpf in the include path. Remove it to
make sure we don't inadvertently introduce new includes without the prefix.

Signed-off-by: Toke Høiland-Jørgensen <toke@redhat.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Link: https://lore.kernel.org/bpf/157952561027.1683545.1976265477926794138.stgit@toke.dk
4 years agosamples/bpf: Use consistent include paths for libbpf
Toke Høiland-Jørgensen [Mon, 20 Jan 2020 13:06:49 +0000 (14:06 +0100)]
samples/bpf: Use consistent include paths for libbpf

Fix all files in samples/bpf to include libbpf header files with the bpf/
prefix, to be consistent with external users of the library. Also ensure
that all includes of exported libbpf header files (those that are exported
on 'make install' of the library) use bracketed includes instead of quoted.

To make sure no new files are introduced that doesn't include the bpf/
prefix in its include, remove tools/lib/bpf from the include path entirely,
and use tools/lib instead.

Fixes: 6910d7d3867a ("selftests/bpf: Ensure bpf_helper_defs.h are taken from selftests dir")
Signed-off-by: Toke Høiland-Jørgensen <toke@redhat.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Jesper Dangaard Brouer <brouer@redhat.com>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Link: https://lore.kernel.org/bpf/157952560911.1683545.8795966751309534150.stgit@toke.dk
4 years agoperf: Use consistent include paths for libbpf
Toke Høiland-Jørgensen [Mon, 20 Jan 2020 13:06:48 +0000 (14:06 +0100)]
perf: Use consistent include paths for libbpf

Fix perf to include libbpf header files with the bpf/ prefix, to
be consistent with external users of the library.

Signed-off-by: Toke Høiland-Jørgensen <toke@redhat.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Link: https://lore.kernel.org/bpf/157952560797.1683545.7685921032671386301.stgit@toke.dk
4 years agobpftool: Use consistent include paths for libbpf
Toke Høiland-Jørgensen [Mon, 20 Jan 2020 13:06:46 +0000 (14:06 +0100)]
bpftool: Use consistent include paths for libbpf

Fix bpftool to include libbpf header files with the bpf/ prefix, to be
consistent with external users of the library. Also ensure that all
includes of exported libbpf header files (those that are exported on 'make
install' of the library) use bracketed includes instead of quoted.

To make sure no new files are introduced that doesn't include the bpf/
prefix in its include, remove tools/lib/bpf from the include path entirely,
and use tools/lib instead.

Fixes: 6910d7d3867a ("selftests/bpf: Ensure bpf_helper_defs.h are taken from selftests dir")
Signed-off-by: Toke Høiland-Jørgensen <toke@redhat.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Link: https://lore.kernel.org/bpf/157952560684.1683545.4765181397974997027.stgit@toke.dk
4 years agoselftests: Use consistent include paths for libbpf
Toke Høiland-Jørgensen [Mon, 20 Jan 2020 13:06:45 +0000 (14:06 +0100)]
selftests: Use consistent include paths for libbpf

Fix all selftests to include libbpf header files with the bpf/ prefix, to
be consistent with external users of the library. Also ensure that all
includes of exported libbpf header files (those that are exported on 'make
install' of the library) use bracketed includes instead of quoted.

To not break the build, keep the old include path until everything has been
changed to the new one; a subsequent patch will remove that.

Fixes: 6910d7d3867a ("selftests/bpf: Ensure bpf_helper_defs.h are taken from selftests dir")
Signed-off-by: Toke Høiland-Jørgensen <toke@redhat.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Link: https://lore.kernel.org/bpf/157952560568.1683545.9649335788846513446.stgit@toke.dk
4 years agotools/runqslower: Use consistent include paths for libbpf
Toke Høiland-Jørgensen [Mon, 20 Jan 2020 13:06:44 +0000 (14:06 +0100)]
tools/runqslower: Use consistent include paths for libbpf

Fix the runqslower tool to include libbpf header files with the bpf/
prefix, to be consistent with external users of the library. Also ensure
that all includes of exported libbpf header files (those that are exported
on 'make install' of the library) use bracketed includes instead of quoted.

To not break the build, keep the old include path until everything has been
changed to the new one; a subsequent patch will remove that.

Fixes: 6910d7d3867a ("selftests/bpf: Ensure bpf_helper_defs.h are taken from selftests dir")
Signed-off-by: Toke Høiland-Jørgensen <toke@redhat.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Link: https://lore.kernel.org/bpf/157952560457.1683545.9913736511685743625.stgit@toke.dk
4 years agoselftests: Pass VMLINUX_BTF to runqslower Makefile
Toke Høiland-Jørgensen [Mon, 20 Jan 2020 13:06:43 +0000 (14:06 +0100)]
selftests: Pass VMLINUX_BTF to runqslower Makefile

Add a VMLINUX_BTF variable with the locally-built path when calling the
runqslower Makefile from selftests. This makes sure a simple 'make'
invocation in the selftests dir works even when there is no BTF information
for the running kernel. Do a wildcard expansion and include the same paths
for BTF for the running kernel as in the runqslower Makefile, to make it
possible to build selftests without having a vmlinux in the local tree.

Also fix the make invocation to use $(OUTPUT)/tools as the destination
directory instead of $(CURDIR)/tools.

Fixes: 3a0d3092a4ed ("selftests/bpf: Build runqslower from selftests")
Signed-off-by: Toke Høiland-Jørgensen <toke@redhat.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Link: https://lore.kernel.org/bpf/157952560344.1683545.2723631988771664417.stgit@toke.dk
4 years agotools/bpf/runqslower: Fix override option for VMLINUX_BTF
Toke Høiland-Jørgensen [Mon, 20 Jan 2020 13:06:42 +0000 (14:06 +0100)]
tools/bpf/runqslower: Fix override option for VMLINUX_BTF

The runqslower tool refuses to build without a file to read vmlinux BTF
from. The build fails with an error message to override the location by
setting the VMLINUX_BTF variable if autodetection fails. However, the
Makefile doesn't actually work with that override - the error message is
still emitted.

Fix this by including the value of VMLINUX_BTF in the expansion, and only
emitting the error message if the *result* is empty. Also permit running
'make clean' even though no VMLINUX_BTF is set.

Fixes: 9c01546d26d2 ("tools/bpf: Add runqslower tool to tools/bpf")
Signed-off-by: Toke Høiland-Jørgensen <toke@redhat.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Link: https://lore.kernel.org/bpf/157952560237.1683545.17771785178857224877.stgit@toke.dk
4 years agosamples/bpf: Don't try to remove user's homedir on clean
Toke Høiland-Jørgensen [Mon, 20 Jan 2020 13:06:41 +0000 (14:06 +0100)]
samples/bpf: Don't try to remove user's homedir on clean

The 'clean' rule in the samples/bpf Makefile tries to remove backup
files (ending in ~). However, if no such files exist, it will instead try
to remove the user's home directory. While the attempt is mostly harmless,
it does lead to a somewhat scary warning like this:

rm: cannot remove '~': Is a directory

Fix this by using find instead of shell expansion to locate any actual
backup files that need to be removed.

Fixes: b62a796c109c ("samples/bpf: allow make to be run from samples/bpf/ directory")
Signed-off-by: Toke Høiland-Jørgensen <toke@redhat.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Jesper Dangaard Brouer <brouer@redhat.com>
Link: https://lore.kernel.org/bpf/157952560126.1683545.7273054725976032511.stgit@toke.dk
4 years agoselftests/bpf: Skip perf hw events test if the setup disabled it
Hangbin Liu [Fri, 17 Jan 2020 10:06:56 +0000 (18:06 +0800)]
selftests/bpf: Skip perf hw events test if the setup disabled it

The same with commit 4e59afbbed96 ("selftests/bpf: skip nmi test when perf
hw events are disabled"), it would make more sense to skip the
test_stacktrace_build_id_nmi test if the setup (e.g. virtual machines) has
disabled hardware perf events.

Fixes: 13790d1cc72c ("bpf: add selftest for stackmap with build_id in NMI context")
Signed-off-by: Hangbin Liu <liuhangbin@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Link: https://lore.kernel.org/bpf/20200117100656.10359-1-liuhangbin@gmail.com
4 years agoselftests/bpf: Don't check for btf fd in test_btf
Stanislav Fomichev [Sat, 18 Jan 2020 01:05:46 +0000 (17:05 -0800)]
selftests/bpf: Don't check for btf fd in test_btf

After commit 0d13bfce023a ("libbpf: Don't require root for
bpf_object__open()") we no longer load BTF during bpf_object__open(),
so let's remove the expectation from test_btf that the fd is not -1.
The test currently fails.

Before:
BTF libbpf test[1] (test_btf_haskv.o): do_test_file:4152:FAIL bpf_object__btf_fd: -1
BTF libbpf test[2] (test_btf_newkv.o): do_test_file:4152:FAIL bpf_object__btf_fd: -1
BTF libbpf test[3] (test_btf_nokv.o): do_test_file:4152:FAIL bpf_object__btf_fd: -1

After:
BTF libbpf test[1] (test_btf_haskv.o): OK
BTF libbpf test[2] (test_btf_newkv.o): OK
BTF libbpf test[3] (test_btf_nokv.o): OK

Fixes: 0d13bfce023a ("libbpf: Don't require root for bpf_object__open()")
Signed-off-by: Stanislav Fomichev <sdf@google.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Link: https://lore.kernel.org/bpf/20200118010546.74279-1-sdf@google.com
4 years agobpf: Fix memory leaks in generic update/delete batch ops
Brian Vazquez [Sun, 19 Jan 2020 19:40:40 +0000 (11:40 -0800)]
bpf: Fix memory leaks in generic update/delete batch ops

Generic update/delete batch ops functions were using __bpf_copy_key
without properly freeing the memory. Handle the memory allocation and
copy_from_user separately.

Fixes: aa2e93b8e58e ("bpf: Add generic support for update and delete batch ops")
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Brian Vazquez <brianvv@google.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Yonghong Song <yhs@fb.com>
Link: https://lore.kernel.org/bpf/20200119194040.128369-1-brianvv@google.com
4 years agolibbpf: Load btf_vmlinux only once per object.
KP Singh [Fri, 17 Jan 2020 21:28:25 +0000 (22:28 +0100)]
libbpf: Load btf_vmlinux only once per object.

As more programs (TRACING, STRUCT_OPS, and upcoming LSM) use vmlinux
BTF information, loading the BTF vmlinux information for every program
in an object is sub-optimal. The fix was originally proposed in:

   https://lore.kernel.org/bpf/CAEf4BzZodr3LKJuM7QwD38BiEH02Cc1UbtnGpVkCJ00Mf+V_Qg@mail.gmail.com/

The btf_vmlinux is populated in the object if any of the programs in
the object requires it just before the programs are loaded and freed
after the programs finish loading.

Reported-by: Andrii Nakryiko <andrii.nakryiko@gmail.com>
Signed-off-by: KP Singh <kpsingh@google.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Reviewed-by: Brendan Jackman <jackmanb@chromium.org>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Link: https://lore.kernel.org/bpf/20200117212825.11755-1-kpsingh@chromium.org
4 years agolibbpf: Fix potential multiplication overflow in mmap() size calculation
Andrii Nakryiko [Fri, 17 Jan 2020 06:08:00 +0000 (22:08 -0800)]
libbpf: Fix potential multiplication overflow in mmap() size calculation

Prevent potential overflow performed in 32-bit integers, before assigning
result to size_t. Reported by LGTM static analysis.

Fixes: eba9c5f498a1 ("libbpf: Refactor global data map initialization")
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20200117060801.1311525-4-andriin@fb.com
4 years agolibbpf: Simplify BTF initialization logic
Andrii Nakryiko [Fri, 17 Jan 2020 06:07:59 +0000 (22:07 -0800)]
libbpf: Simplify BTF initialization logic

Current implementation of bpf_object's BTF initialization is very convoluted
and thus prone to errors. It doesn't have to be like that. This patch
simplifies it significantly.

This code also triggered static analysis issues over logically dead code due
to redundant error checks. This simplification should fix that as well.

Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20200117060801.1311525-3-andriin@fb.com
4 years agolibbpf: Fix error handling bug in btf_dump__new
Andrii Nakryiko [Fri, 17 Jan 2020 06:07:58 +0000 (22:07 -0800)]
libbpf: Fix error handling bug in btf_dump__new

Fix missing jump to error handling in btf_dump__new, found by Coverity static
code analysis.

Fixes: 9f81654eebe8 ("libbpf: Expose BTF-to-C type declaration emitting API")
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20200117060801.1311525-2-andriin@fb.com
4 years agobpf: Remove set but not used variable 'first_key'
YueHaibing [Thu, 16 Jan 2020 14:53:00 +0000 (22:53 +0800)]
bpf: Remove set but not used variable 'first_key'

kernel/bpf/syscall.c: In function generic_map_lookup_batch:
kernel/bpf/syscall.c:1339:7: warning: variable first_key set but not used [-Wunused-but-set-variable]

It is never used, so remove it.

Reported-by: Hulk Robot <hulkci@huawei.com>
Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Brian Vazquez <brianvv@google.com>
Link: https://lore.kernel.org/bpf/20200116145300.59056-1-yuehaibing@huawei.com
4 years agoMerge branch 'xdp_redirect-bulking'
Alexei Starovoitov [Fri, 17 Jan 2020 04:03:35 +0000 (20:03 -0800)]
Merge branch 'xdp_redirect-bulking'

Toke Høiland-Jørgensen says:

====================
Since commit 96360004b862 ("xdp: Make devmap flush_list common for all map
instances"), devmap flushing is a global operation instead of tied to a
particular map. This means that with a bit of refactoring, we can finally fix
the performance delta between the bpf_redirect_map() and bpf_redirect() helper
functions, by introducing bulking for the latter as well.

This series makes this change by moving the data structure used for the bulking
into struct net_device itself, so we can access it even when there is not
devmap. Once this is done, moving the bpf_redirect() helper to use the bulking
mechanism becomes quite trivial, and brings bpf_redirect() up to the same as
bpf_redirect_map():

                       Before:   After:
1 CPU:
bpf_redirect_map:      8.4 Mpps  8.4 Mpps  (no change)
bpf_redirect:          5.0 Mpps  8.4 Mpps  (+68%)
2 CPUs:
bpf_redirect_map:     15.9 Mpps  16.1 Mpps  (+1% or ~no change)
bpf_redirect:          9.5 Mpps  15.9 Mpps  (+67%)

After this patch series, the only semantics different between the two variants
of the bpf() helper (apart from the absence of a map argument, obviously) is
that the _map() variant will return an error if passed an invalid map index,
whereas the bpf_redirect() helper will succeed, but drop packets on
xdp_do_redirect(). This is because the helper has no reference to the calling
netdev, so unfortunately we can't do the ifindex lookup directly in the helper.

Changelog:

v3:
  - Switch two more fields to avoid a list_head spanning two cache lines
  - Include Jesper's tracepoint patch
  - Also rename xdp_do_flush_map()
  - Fix a few nits from Maciej

v2:
  - Consolidate code paths and tracepoints for map and non-map redirect variants
    (Björn)
  - Add performance data for 2-CPU test (Jesper)
  - Move fields to avoid shifting cache lines in struct net_device (Eric)
====================

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
4 years agodevmap: Adjust tracepoint for map-less queue flush
Jesper Dangaard Brouer [Thu, 16 Jan 2020 15:14:46 +0000 (16:14 +0100)]
devmap: Adjust tracepoint for map-less queue flush

Now that we don't have a reference to a devmap when flushing the device
bulk queue, let's change the the devmap_xmit tracepoint to remote the
map_id and map_index fields entirely. Rearrange the fields so 'drops' and
'sent' stay in the same position in the tracepoint struct, to make it
possible for the xdp_monitor utility to read both the old and the new
format.

Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Toke Høiland-Jørgensen <toke@redhat.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/157918768613.1458396.9165902403373826572.stgit@toke.dk
4 years agoxdp: Use bulking for non-map XDP_REDIRECT and consolidate code paths
Toke Høiland-Jørgensen [Thu, 16 Jan 2020 15:14:45 +0000 (16:14 +0100)]
xdp: Use bulking for non-map XDP_REDIRECT and consolidate code paths

Since the bulk queue used by XDP_REDIRECT now lives in struct net_device,
we can re-use the bulking for the non-map version of the bpf_redirect()
helper. This is a simple matter of having xdp_do_redirect_slow() queue the
frame on the bulk queue instead of sending it out with __bpf_tx_xdp().

Unfortunately we can't make the bpf_redirect() helper return an error if
the ifindex doesn't exit (as bpf_redirect_map() does), because we don't
have a reference to the network namespace of the ingress device at the time
the helper is called. So we have to leave it as-is and keep the device
lookup in xdp_do_redirect_slow().

Since this leaves less reason to have the non-map redirect code in a
separate function, so we get rid of the xdp_do_redirect_slow() function
entirely. This does lose us the tracepoint disambiguation, but fortunately
the xdp_redirect and xdp_redirect_map tracepoints use the same tracepoint
entry structures. This means both can contain a map index, so we can just
amend the tracepoint definitions so we always emit the xdp_redirect(_err)
tracepoints, but with the map ID only populated if a map is present. This
means we retire the xdp_redirect_map(_err) tracepoints entirely, but keep
the definitions around in case someone is still listening for them.

With this change, the performance of the xdp_redirect sample program goes
from 5Mpps to 8.4Mpps (a 68% increase).

Since the flush functions are no longer map-specific, rename the flush()
functions to drop _map from their names. One of the renamed functions is
the xdp_do_flush_map() callback used in all the xdp-enabled drivers. To
keep from having to update all drivers, use a #define to keep the old name
working, and only update the virtual drivers in this patch.

Signed-off-by: Toke Høiland-Jørgensen <toke@redhat.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Link: https://lore.kernel.org/bpf/157918768505.1458396.17518057312953572912.stgit@toke.dk
4 years agoxdp: Move devmap bulk queue into struct net_device
Toke Høiland-Jørgensen [Thu, 16 Jan 2020 15:14:44 +0000 (16:14 +0100)]
xdp: Move devmap bulk queue into struct net_device

Commit 96360004b862 ("xdp: Make devmap flush_list common for all map
instances"), changed devmap flushing to be a global operation instead of a
per-map operation. However, the queue structure used for bulking was still
allocated as part of the containing map.

This patch moves the devmap bulk queue into struct net_device. The
motivation for this is reusing it for the non-map variant of XDP_REDIRECT,
which will be changed in a subsequent commit.  To avoid other fields of
struct net_device moving to different cache lines, we also move a couple of
other members around.

We defer the actual allocation of the bulk queue structure until the
NETDEV_REGISTER notification devmap.c. This makes it possible to check for
ndo_xdp_xmit support before allocating the structure, which is not possible
at the time struct net_device is allocated. However, we keep the freeing in
free_netdev() to avoid adding another RCU callback on NETDEV_UNREGISTER.

Because of this change, we lose the reference back to the map that
originated the redirect, so change the tracepoint to always return 0 as the
map ID and index. Otherwise no functional change is intended with this
patch.

After this patch, the relevant part of struct net_device looks like this,
according to pahole:

/* --- cacheline 14 boundary (896 bytes) --- */
struct netdev_queue *      _tx __attribute__((__aligned__(64))); /*   896     8 */
unsigned int               num_tx_queues;        /*   904     4 */
unsigned int               real_num_tx_queues;   /*   908     4 */
struct Qdisc *             qdisc;                /*   912     8 */
unsigned int               tx_queue_len;         /*   920     4 */
spinlock_t                 tx_global_lock;       /*   924     4 */
struct xdp_dev_bulk_queue * xdp_bulkq;           /*   928     8 */
struct xps_dev_maps *      xps_cpus_map;         /*   936     8 */
struct xps_dev_maps *      xps_rxqs_map;         /*   944     8 */
struct mini_Qdisc *        miniq_egress;         /*   952     8 */
/* --- cacheline 15 boundary (960 bytes) --- */
struct hlist_head  qdisc_hash[16];               /*   960   128 */
/* --- cacheline 17 boundary (1088 bytes) --- */
struct timer_list  watchdog_timer;               /*  1088    40 */

/* XXX last struct has 4 bytes of padding */

int                        watchdog_timeo;       /*  1128     4 */

/* XXX 4 bytes hole, try to pack */

struct list_head   todo_list;                    /*  1136    16 */
/* --- cacheline 18 boundary (1152 bytes) --- */

Signed-off-by: Toke Høiland-Jørgensen <toke@redhat.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Björn Töpel <bjorn.topel@intel.com>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Link: https://lore.kernel.org/bpf/157918768397.1458396.12673224324627072349.stgit@toke.dk
4 years agolibbpf: Revert bpf_helper_defs.h inclusion regression
Andrii Nakryiko [Fri, 17 Jan 2020 00:41:02 +0000 (16:41 -0800)]
libbpf: Revert bpf_helper_defs.h inclusion regression

Revert bpf_helpers.h's change to include auto-generated bpf_helper_defs.h
through <> instead of "", which causes it to be searched in include path. This
can break existing applications that don't have their include path pointing
directly to where libbpf installs its headers.

There is ongoing work to make all (not just bpf_helper_defs.h) includes more
consistent across libbpf and its consumers, but this unbreaks user code as is
right now without any regressions. Selftests still behave sub-optimally
(taking bpf_helper_defs.h from libbpf's source directory, if it's present
there), which will be fixed in subsequent patches.

Fixes: 6910d7d3867a ("selftests/bpf: Ensure bpf_helper_defs.h are taken from selftests dir")
Reported-by: Toke Høiland-Jørgensen <toke@redhat.com>
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20200117004103.148068-1-andriin@fb.com
4 years agoselftests/bpf: Fix test_progs send_signal flakiness with nmi mode
Yonghong Song [Thu, 16 Jan 2020 17:40:04 +0000 (09:40 -0800)]
selftests/bpf: Fix test_progs send_signal flakiness with nmi mode

Alexei observed that test_progs send_signal may fail if run
with command line "./test_progs" and the tests will pass
if just run "./test_progs -n 40".

I observed similar issue with nmi subtest failure
and added a delay 100 us in Commit ab8b7f0cb358
("tools/bpf: Add self tests for bpf_send_signal_thread()")
and the problem is gone for me. But the issue still exists
in Alexei's testing environment.

The current code uses sample_freq = 50 (50 events/second), which
may not be enough. But if the sample_freq value is larger than
sysctl kernel/perf_event_max_sample_rate, the perf_event_open
syscall will fail.

This patch changed nmi perf testing to use sample_period = 1,
which means trying to sampling every event. This seems fixing
the issue.

Fixes: ab8b7f0cb358 ("tools/bpf: Add self tests for bpf_send_signal_thread()")
Signed-off-by: Yonghong Song <yhs@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Link: https://lore.kernel.org/bpf/20200116174004.1522812-1-yhs@fb.com
4 years agolibbpf: Fix unneeded extra initialization in bpf_map_batch_common
Brian Vazquez [Thu, 16 Jan 2020 04:59:18 +0000 (20:59 -0800)]
libbpf: Fix unneeded extra initialization in bpf_map_batch_common

bpf_attr doesn't required to be declared with '= {}' as memset is used
in the code.

Fixes: 2ab3d86ea1859 ("libbpf: Add libbpf support to batch ops")
Reported-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Brian Vazquez <brianvv@google.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20200116045918.75597-1-brianvv@google.com
4 years agoselftests/bpf: Add whitelist/blacklist of test names to test_progs
Andrii Nakryiko [Thu, 16 Jan 2020 00:55:49 +0000 (16:55 -0800)]
selftests/bpf: Add whitelist/blacklist of test names to test_progs

Add ability to specify a list of test name substrings for selecting which
tests to run. So now -t is accepting a comma-separated list of strings,
similarly to how -n accepts a comma-separated list of test numbers.

Additionally, add ability to blacklist tests by name. Blacklist takes
precedence over whitelist. Blacklisting is important for cases where it's
known that some tests can't pass (e.g., due to perf hardware events that are
not available within VM). This is going to be used for libbpf testing in
Travis CI in its Github repo.

Example runs with just whitelist and whitelist + blacklist:

  $ sudo ./test_progs -tattach,core/existence
  #1 attach_probe:OK
  #6 cgroup_attach_autodetach:OK
  #7 cgroup_attach_multi:OK
  #8 cgroup_attach_override:OK
  #9 core_extern:OK
  #10/44 existence:OK
  #10/45 existence___minimal:OK
  #10/46 existence__err_int_sz:OK
  #10/47 existence__err_int_type:OK
  #10/48 existence__err_int_kind:OK
  #10/49 existence__err_arr_kind:OK
  #10/50 existence__err_arr_value_type:OK
  #10/51 existence__err_struct_type:OK
  #10 core_reloc:OK
  #19 flow_dissector_reattach:OK
  #60 tp_attach_query:OK
  Summary: 8/8 PASSED, 0 SKIPPED, 0 FAILED

  $ sudo ./test_progs -tattach,core/existence -bcgroup,flow/arr
  #1 attach_probe:OK
  #9 core_extern:OK
  #10/44 existence:OK
  #10/45 existence___minimal:OK
  #10/46 existence__err_int_sz:OK
  #10/47 existence__err_int_type:OK
  #10/48 existence__err_int_kind:OK
  #10/51 existence__err_struct_type:OK
  #10 core_reloc:OK
  #60 tp_attach_query:OK
  Summary: 4/6 PASSED, 0 SKIPPED, 0 FAILED

Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Cc: Julia Kartseva <hex@fb.com>
Link: https://lore.kernel.org/bpf/20200116005549.3644118-1-andriin@fb.com
4 years agoMerge branch 'bpftool-improvements'
Alexei Starovoitov [Wed, 15 Jan 2020 23:23:28 +0000 (15:23 -0800)]
Merge branch 'bpftool-improvements'

Martin Lau says:

====================
When a map is storing a kernel's struct, its
map_info->btf_vmlinux_value_type_id is set.  The first map type
supporting it is BPF_MAP_TYPE_STRUCT_OPS.

This series adds support to dump this kind of map with BTF.
The first two patches are bug fixes which are only applicable to
bpf-next.

Please see individual patches for details.

v3:
- Remove unnecessary #include "libbpf_internal.h" from patch 5

v2:
- Expose bpf_find_kernel_btf() as a LIBBPF_API in patch 3 (Andrii)
- Cache btf_vmlinux in bpftool/map.c (Andrii)
====================

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
4 years agobpftool: Support dumping a map with btf_vmlinux_value_type_id
Martin KaFai Lau [Wed, 15 Jan 2020 23:00:44 +0000 (15:00 -0800)]
bpftool: Support dumping a map with btf_vmlinux_value_type_id

This patch makes bpftool support dumping a map's value properly
when the map's value type is a type of the running kernel's btf.
(i.e. map_info.btf_vmlinux_value_type_id is set instead of
map_info.btf_value_type_id).  The first usecase is for the
BPF_MAP_TYPE_STRUCT_OPS.

Signed-off-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Link: https://lore.kernel.org/bpf/20200115230044.1103008-1-kafai@fb.com
4 years agobpftool: Add struct_ops map name
Martin KaFai Lau [Wed, 15 Jan 2020 23:00:37 +0000 (15:00 -0800)]
bpftool: Add struct_ops map name

This patch adds BPF_MAP_TYPE_STRUCT_OPS to "struct_ops" name mapping
so that "bpftool map show" can print the "struct_ops" map type
properly.

[root@arch-fb-vm1 bpf]# ~/devshare/fb-kernel/linux/tools/bpf/bpftool/bpftool map show id 8
8: struct_ops  name dctcp  flags 0x0
key 4B  value 256B  max_entries 1  memlock 4096B
btf_id 7

Signed-off-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Link: https://lore.kernel.org/bpf/20200115230037.1102674-1-kafai@fb.com
4 years agolibbpf: Expose bpf_find_kernel_btf as a LIBBPF_API
Martin KaFai Lau [Wed, 15 Jan 2020 23:00:31 +0000 (15:00 -0800)]
libbpf: Expose bpf_find_kernel_btf as a LIBBPF_API

This patch exposes bpf_find_kernel_btf() as a LIBBPF_API.
It will be used in 'bpftool map dump' in a following patch
to dump a map with btf_vmlinux_value_type_id set.

bpf_find_kernel_btf() is renamed to libbpf_find_kernel_btf()
and moved to btf.c.  As <linux/kernel.h> is included,
some of the max/min type casting needs to be fixed.

Signed-off-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Link: https://lore.kernel.org/bpf/20200115230031.1102305-1-kafai@fb.com
4 years agobpftool: Fix missing BTF output for json during map dump
Martin KaFai Lau [Wed, 15 Jan 2020 23:00:25 +0000 (15:00 -0800)]
bpftool: Fix missing BTF output for json during map dump

The btf availability check is only done for plain text output.
It causes the whole BTF output went missing when json_output
is used.

This patch simplifies the logic a little by avoiding passing "int btf" to
map_dump().

For plain text output, the btf_wtr is only created when the map has
BTF (i.e. info->btf_id != 0).  The nullness of "json_writer_t *wtr"
in map_dump() alone can decide if dumping BTF output is needed.
As long as wtr is not NULL, map_dump() will print out the BTF-described
data whenever a map has BTF available (i.e. info->btf_id != 0)
regardless of json or plain-text output.

In do_dump(), the "int btf" is also renamed to "int do_plain_btf".

Fixes: 99f9863a0c45 ("bpftool: Match maps by name")
Signed-off-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Cc: Paul Chaignon <paul.chaignon@orange.com>
Link: https://lore.kernel.org/bpf/20200115230025.1101828-1-kafai@fb.com
4 years agobpftool: Fix a leak of btf object
Martin KaFai Lau [Wed, 15 Jan 2020 23:00:19 +0000 (15:00 -0800)]
bpftool: Fix a leak of btf object

When testing a map has btf or not, maps_have_btf() tests it by actually
getting a btf_fd from sys_bpf(BPF_BTF_GET_FD_BY_ID). However, it
forgot to btf__free() it.

In maps_have_btf() stage, there is no need to test it by really
calling sys_bpf(BPF_BTF_GET_FD_BY_ID). Testing non zero
info.btf_id is good enough.

Also, the err_close case is unnecessary, and also causes double
close() because the calling func do_dump() will close() all fds again.

Fixes: 99f9863a0c45 ("bpftool: Match maps by name")
Signed-off-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Cc: Paul Chaignon <paul.chaignon@orange.com>
Link: https://lore.kernel.org/bpf/20200115230019.1101352-1-kafai@fb.com
4 years agoMerge branch 'bpf-batch-ops'
Alexei Starovoitov [Wed, 15 Jan 2020 22:00:36 +0000 (14:00 -0800)]
Merge branch 'bpf-batch-ops'

Brian Vazquez says:

====================
This patch series introduce batch ops that can be added to bpf maps to
lookup/lookup_and_delete/update/delete more than 1 element at the time,
this is specially useful when syscall overhead is a problem and in case
of hmap it will provide a reliable way of traversing them.

The implementation inclues a generic approach that could potentially be
used by any bpf map and adds it to arraymap, it also includes the specific
implementation of hashmaps which are traversed using buckets instead
of keys.

The bpf syscall subcommands introduced are:

  BPF_MAP_LOOKUP_BATCH
  BPF_MAP_LOOKUP_AND_DELETE_BATCH
  BPF_MAP_UPDATE_BATCH
  BPF_MAP_DELETE_BATCH

The UAPI attribute is:

  struct { /* struct used by BPF_MAP_*_BATCH commands */
         __aligned_u64   in_batch;       /* start batch,
                                          * NULL to start from beginning
                                          */
         __aligned_u64   out_batch;      /* output: next start batch */
         __aligned_u64   keys;
         __aligned_u64   values;
         __u32           count;          /* input/output:
                                          * input: # of key/value
                                          * elements
                                          * output: # of filled elements
                                          */
         __u32           map_fd;
         __u64           elem_flags;
         __u64           flags;
  } batch;

in_batch and out_batch are only used for lookup and lookup_and_delete since
those are the only two operations that attempt to traverse the map.

update/delete batch ops should provide the keys/values that user wants
to modify.

Here are the previous discussions on the batch processing:
 - https://lore.kernel.org/bpf/20190724165803.87470-1-brianvv@google.com/
 - https://lore.kernel.org/bpf/20190829064502.2750303-1-yhs@fb.com/
 - https://lore.kernel.org/bpf/20190906225434.3635421-1-yhs@fb.com/

Changelog sinve v4:
 - Remove unnecessary checks from libbpf API (Andrii Nakryiko)
 - Move DECLARE_LIBBPF_OPTS with all var declarations (Andrii Nakryiko)
 - Change bucket internal buffer size to 5 entries (Yonghong Song)
 - Fix some minor bugs in hashtab batch ops implementation (Yonghong Song)

Changelog sinve v3:
 - Do not use copy_to_user inside atomic region (Yonghong Song)
 - Use _opts approach on libbpf APIs (Andrii Nakryiko)
 - Drop generic_map_lookup_and_delete_batch support
 - Free malloc-ed memory in tests (Yonghong Song)
 - Reverse christmas tree (Yonghong Song)
 - Add acked labels

Changelog sinve v2:
 - Add generic batch support for lpm_trie and test it (Yonghong Song)
 - Use define MAP_LOOKUP_RETRIES for retries (John Fastabend)
 - Return errors directly and remove labels (Yonghong Song)
 - Insert new API functions into libbpf alphabetically (Yonghong Song)
 - Change hlist_nulls_for_each_entry_rcu to
   hlist_nulls_for_each_entry_safe in htab batch ops (Yonghong Song)

Changelog since v1:
 - Fix SOB ordering and remove Co-authored-by tag (Alexei Starovoitov)

Changelog since RFC:
 - Change batch to in_batch and out_batch to support more flexible opaque
   values to iterate the bpf maps.
 - Remove update/delete specific batch ops for htab and use the generic
   implementations instead.
====================

Acked-by: Yonghong Song <yhs@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
4 years agoselftests/bpf: Add batch ops testing to array bpf map
Brian Vazquez [Wed, 15 Jan 2020 18:43:08 +0000 (10:43 -0800)]
selftests/bpf: Add batch ops testing to array bpf map

Tested bpf_map_lookup_batch() and bpf_map_update_batch()
functionality.

  $ ./test_maps
      ...
        test_array_map_batch_ops:PASS
      ...

Signed-off-by: Brian Vazquez <brianvv@google.com>
Signed-off-by: Yonghong Song <yhs@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20200115184308.162644-10-brianvv@google.com
4 years agoselftests/bpf: Add batch ops testing for htab and htab_percpu map
Yonghong Song [Wed, 15 Jan 2020 18:43:07 +0000 (10:43 -0800)]
selftests/bpf: Add batch ops testing for htab and htab_percpu map

Tested bpf_map_lookup_batch(), bpf_map_lookup_and_delete_batch(),
bpf_map_update_batch(), and bpf_map_delete_batch() functionality.
  $ ./test_maps
    ...
      test_htab_map_batch_ops:PASS
      test_htab_percpu_map_batch_ops:PASS
    ...

Signed-off-by: Yonghong Song <yhs@fb.com>
Signed-off-by: Brian Vazquez <brianvv@google.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20200115184308.162644-9-brianvv@google.com
4 years agolibbpf: Add libbpf support to batch ops
Yonghong Song [Wed, 15 Jan 2020 18:43:06 +0000 (10:43 -0800)]
libbpf: Add libbpf support to batch ops

Added four libbpf API functions to support map batch operations:
  . int bpf_map_delete_batch( ... )
  . int bpf_map_lookup_batch( ... )
  . int bpf_map_lookup_and_delete_batch( ... )
  . int bpf_map_update_batch( ... )

Signed-off-by: Yonghong Song <yhs@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20200115184308.162644-8-brianvv@google.com
4 years agotools/bpf: Sync uapi header bpf.h
Yonghong Song [Wed, 15 Jan 2020 18:43:05 +0000 (10:43 -0800)]
tools/bpf: Sync uapi header bpf.h

sync uapi header include/uapi/linux/bpf.h to
tools/include/uapi/linux/bpf.h

Signed-off-by: Yonghong Song <yhs@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20200115184308.162644-7-brianvv@google.com
4 years agobpf: Add batch ops to all htab bpf map
Yonghong Song [Wed, 15 Jan 2020 18:43:04 +0000 (10:43 -0800)]
bpf: Add batch ops to all htab bpf map

htab can't use generic batch support due some problematic behaviours
inherent to the data structre, i.e. while iterating the bpf map  a
concurrent program might delete the next entry that batch was about to
use, in that case there's no easy solution to retrieve the next entry,
the issue has been discussed multiple times (see [1] and [2]).

The only way hmap can be traversed without the problem previously
exposed is by making sure that the map is traversing entire buckets.
This commit implements those strict requirements for hmap, the
implementation follows the same interaction that generic support with
some exceptions:

 - If keys/values buffer are not big enough to traverse a bucket,
   ENOSPC will be returned.
 - out_batch contains the value of the next bucket in the iteration, not
   the next key, but this is transparent for the user since the user
   should never use out_batch for other than bpf batch syscalls.

This commits implements BPF_MAP_LOOKUP_BATCH and adds support for new
command BPF_MAP_LOOKUP_AND_DELETE_BATCH. Note that for update/delete
batch ops it is possible to use the generic implementations.

[1] https://lore.kernel.org/bpf/20190724165803.87470-1-brianvv@google.com/
[2] https://lore.kernel.org/bpf/20190906225434.3635421-1-yhs@fb.com/

Signed-off-by: Yonghong Song <yhs@fb.com>
Signed-off-by: Brian Vazquez <brianvv@google.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20200115184308.162644-6-brianvv@google.com
4 years agobpf: Add lookup and update batch ops to arraymap
Brian Vazquez [Wed, 15 Jan 2020 18:43:03 +0000 (10:43 -0800)]
bpf: Add lookup and update batch ops to arraymap

This adds the generic batch ops functionality to bpf arraymap, note that
since deletion is not a valid operation for arraymap, only batch and
lookup are added.

Signed-off-by: Brian Vazquez <brianvv@google.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Yonghong Song <yhs@fb.com>
Link: https://lore.kernel.org/bpf/20200115184308.162644-5-brianvv@google.com
4 years agobpf: Add generic support for update and delete batch ops
Brian Vazquez [Wed, 15 Jan 2020 18:43:02 +0000 (10:43 -0800)]
bpf: Add generic support for update and delete batch ops

This commit adds generic support for update and delete batch ops that
can be used for almost all the bpf maps. These commands share the same
UAPI attr that lookup and lookup_and_delete batch ops use and the
syscall commands are:

  BPF_MAP_UPDATE_BATCH
  BPF_MAP_DELETE_BATCH

The main difference between update/delete and lookup batch ops is that
for update/delete keys/values must be specified for userspace and
because of that, neither in_batch nor out_batch are used.

Suggested-by: Stanislav Fomichev <sdf@google.com>
Signed-off-by: Brian Vazquez <brianvv@google.com>
Signed-off-by: Yonghong Song <yhs@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20200115184308.162644-4-brianvv@google.com
4 years agobpf: Add generic support for lookup batch op
Brian Vazquez [Wed, 15 Jan 2020 18:43:01 +0000 (10:43 -0800)]
bpf: Add generic support for lookup batch op

This commit introduces generic support for the bpf_map_lookup_batch.
This implementation can be used by almost all the bpf maps since its core
implementation is relying on the existing map_get_next_key and
map_lookup_elem. The bpf syscall subcommand introduced is:

  BPF_MAP_LOOKUP_BATCH

The UAPI attribute is:

  struct { /* struct used by BPF_MAP_*_BATCH commands */
         __aligned_u64   in_batch;       /* start batch,
                                          * NULL to start from beginning
                                          */
         __aligned_u64   out_batch;      /* output: next start batch */
         __aligned_u64   keys;
         __aligned_u64   values;
         __u32           count;          /* input/output:
                                          * input: # of key/value
                                          * elements
                                          * output: # of filled elements
                                          */
         __u32           map_fd;
         __u64           elem_flags;
         __u64           flags;
  } batch;

in_batch/out_batch are opaque values use to communicate between
user/kernel space, in_batch/out_batch must be of key_size length.

To start iterating from the beginning in_batch must be null,
count is the # of key/value elements to retrieve. Note that the 'keys'
buffer must be a buffer of key_size * count size and the 'values' buffer
must be value_size * count, where value_size must be aligned to 8 bytes
by userspace if it's dealing with percpu maps. 'count' will contain the
number of keys/values successfully retrieved. Note that 'count' is an
input/output variable and it can contain a lower value after a call.

If there's no more entries to retrieve, ENOENT will be returned. If error
is ENOENT, count might be > 0 in case it copied some values but there were
no more entries to retrieve.

Note that if the return code is an error and not -EFAULT,
count indicates the number of elements successfully processed.

Suggested-by: Stanislav Fomichev <sdf@google.com>
Signed-off-by: Brian Vazquez <brianvv@google.com>
Signed-off-by: Yonghong Song <yhs@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20200115184308.162644-3-brianvv@google.com
4 years agobpf: Add bpf_map_{value_size, update_value, map_copy_value} functions
Brian Vazquez [Wed, 15 Jan 2020 18:43:00 +0000 (10:43 -0800)]
bpf: Add bpf_map_{value_size, update_value, map_copy_value} functions

This commit moves reusable code from map_lookup_elem and map_update_elem
to avoid code duplication in kernel/bpf/syscall.c.

Signed-off-by: Brian Vazquez <brianvv@google.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Acked-by: Yonghong Song <yhs@fb.com>
Link: https://lore.kernel.org/bpf/20200115184308.162644-2-brianvv@google.com
4 years agoselftests/bpf: Add a test for attaching a bpf fentry/fexit trace to an XDP program
Eelco Chaudron [Wed, 15 Jan 2020 13:15:39 +0000 (13:15 +0000)]
selftests/bpf: Add a test for attaching a bpf fentry/fexit trace to an XDP program

Add a test that will attach a FENTRY and FEXIT program to the XDP test
program. It will also verify data from the XDP context on FENTRY and
verifies the return code on exit.

Signed-off-by: Eelco Chaudron <echaudro@redhat.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Link: https://lore.kernel.org/bpf/157909410480.47481.11202505690938004673.stgit@xdp-tutorial
4 years agolibbpf: Support .text sub-calls relocations
Andrii Nakryiko [Wed, 15 Jan 2020 19:08:56 +0000 (11:08 -0800)]
libbpf: Support .text sub-calls relocations

The LLVM patch https://reviews.llvm.org/D72197 makes LLVM emit function call
relocations within the same section. This includes a default .text section,
which contains any BPF sub-programs. This wasn't the case before and so libbpf
was able to get a way with slightly simpler handling of subprogram call
relocations.

This patch adds support for .text section relocations. It needs to ensure
correct order of relocations, so does two passes:
- first, relocate .text instructions, if there are any relocations in it;
- then process all the other programs and copy over patched .text instructions
for all sub-program calls.

v1->v2:
- break early once .text program is processed.

Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Yonghong Song <yhs@fb.com>
Cc: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20200115190856.2391325-1-andriin@fb.com
4 years agoMerge branch 'bpf_send_signal_thread'
Alexei Starovoitov [Wed, 15 Jan 2020 19:44:51 +0000 (11:44 -0800)]
Merge branch 'bpf_send_signal_thread'

Yonghong Song says:

====================
Commit 8b401f9ed244 ("bpf: implement bpf_send_signal() helper")
added helper bpf_send_signal() which permits bpf program to
send a signal to the current process. The signal may send to
any thread of the process.

This patch implemented a new helper bpf_send_signal_thread()
to send a signal to the thread corresponding to the kernel current task.
This helper can simplify user space code if the thread context of
bpf sending signal is needed in user space. Please see Patch #1 for
details of use case and kernel implementation.

Patch #2 added some bpf self tests for the new helper.

Changelogs:
  v2 -> v3:
    - More simplification for skeleton codes by removing not-needed
      mmap code and redundantly created tracepoint link.
  v1 -> v2:
    - More description for the difference between bpf_send_signal()
      and bpf_send_signal_thread() in the uapi header bpf.h.
    - Use skeleton and mmap for send_signal test.
====================

Acked-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
4 years agotools/bpf: Add self tests for bpf_send_signal_thread()
Yonghong Song [Wed, 15 Jan 2020 03:50:03 +0000 (19:50 -0800)]
tools/bpf: Add self tests for bpf_send_signal_thread()

The test_progs send_signal() is amended to test
bpf_send_signal_thread() as well.

   $ ./test_progs -n 40
   #40/1 send_signal_tracepoint:OK
   #40/2 send_signal_perf:OK
   #40/3 send_signal_nmi:OK
   #40/4 send_signal_tracepoint_thread:OK
   #40/5 send_signal_perf_thread:OK
   #40/6 send_signal_nmi_thread:OK
   #40 send_signal:OK
   Summary: 1/6 PASSED, 0 SKIPPED, 0 FAILED

Also took this opportunity to rewrite the send_signal test
using skeleton framework and array mmap to make code
simpler and more readable.

Signed-off-by: Yonghong Song <yhs@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20200115035003.602425-1-yhs@fb.com
4 years agobpf: Add bpf_send_signal_thread() helper
Yonghong Song [Wed, 15 Jan 2020 03:50:02 +0000 (19:50 -0800)]
bpf: Add bpf_send_signal_thread() helper

Commit 8b401f9ed244 ("bpf: implement bpf_send_signal() helper")
added helper bpf_send_signal() which permits bpf program to
send a signal to the current process. The signal may be
delivered to any threads in the process.

We found a use case where sending the signal to the current
thread is more preferable.
  - A bpf program will collect the stack trace and then
    send signal to the user application.
  - The user application will add some thread specific
    information to the just collected stack trace for
    later analysis.

If bpf_send_signal() is used, user application will need
to check whether the thread receiving the signal matches
the thread collecting the stack by checking thread id.
If not, it will need to send signal to another thread
through pthread_kill().

This patch proposed a new helper bpf_send_signal_thread(),
which sends the signal to the thread corresponding to
the current kernel task. This way, user space is guaranteed that
bpf_program execution context and user space signal handling
context are the same thread.

Signed-off-by: Yonghong Song <yhs@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20200115035002.602336-1-yhs@fb.com
4 years agoxsk: Support allocations of large umems
Magnus Karlsson [Tue, 14 Jan 2020 09:49:25 +0000 (10:49 +0100)]
xsk: Support allocations of large umems

When registering a umem area that is sufficiently large (>1G on an
x86), kmalloc cannot be used to allocate one of the internal data
structures, as the size requested gets too large. Use kvmalloc instead
that falls back on vmalloc if the allocation is too large for kmalloc.

Also add accounting for this structure as it is triggered by a user
space action (the XDP_UMEM_REG setsockopt) and it is by far the
largest structure of kernel allocated memory in xsk.

Reported-by: Ryan Goodfellow <rgoodfel@isi.edu>
Signed-off-by: Magnus Karlsson <magnus.karlsson@intel.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Jonathan Lemon <jonathan.lemon@gmail.com>
Link: https://lore.kernel.org/bpf/1578995365-7050-1-git-send-email-magnus.karlsson@intel.com
4 years agobpf: Return -EBADRQC for invalid map type in __bpf_tx_xdp_map
Li RongQing [Fri, 10 Jan 2020 01:04:37 +0000 (09:04 +0800)]
bpf: Return -EBADRQC for invalid map type in __bpf_tx_xdp_map

A negative value should be returned if map->map_type is invalid
although that is impossible now, but if we run into such situation
in future, then xdpbuff could be leaked.

Daniel Borkmann suggested:

-EBADRQC should be returned to stay consistent with generic XDP
for the tracepoint output and not to be confused with -EOPNOTSUPP
from other locations like dev_map_enqueue() when ndo_xdp_xmit is
missing and such.

Suggested-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Li RongQing <lirongqing@baidu.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/1578618277-18085-1-git-send-email-lirongqing@baidu.com
4 years agobpf: Fix seq_show for BPF_MAP_TYPE_STRUCT_OPS
Martin KaFai Lau [Tue, 14 Jan 2020 07:26:47 +0000 (23:26 -0800)]
bpf: Fix seq_show for BPF_MAP_TYPE_STRUCT_OPS

Instead of using bpf_struct_ops_map_lookup_elem() which is
not implemented, bpf_struct_ops_map_seq_show_elem() should
also use bpf_struct_ops_map_sys_lookup_elem() which does
an inplace update to the value.  The change allocates
a value to pass to bpf_struct_ops_map_sys_lookup_elem().

[root@arch-fb-vm1 bpf]# cat /sys/fs/bpf/dctcp
{{{1}},BPF_STRUCT_OPS_STATE_INUSE,{{00000000df93eebc,00000000df93eebc},0,2, ...

Fixes: 85d33df357b6 ("bpf: Introduce BPF_MAP_TYPE_STRUCT_OPS")
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20200114072647.3188298-1-kafai@fb.com
4 years agoMerge branch 'runqslower'
Alexei Starovoitov [Tue, 14 Jan 2020 01:48:13 +0000 (17:48 -0800)]
Merge branch 'runqslower'

Andrii Nakryiko says:

====================
Based on recent BPF CO-RE, tp_btf, and BPF skeleton changes, re-implement
BCC-based runqslower tool as a portable pre-compiled BPF CO-RE-based tool.
Make sure it's built as part of selftests to ensure it doesn't bit rot.

As part of this patch set, augment `format c` output of `bpftool btf dump`
sub-command with applying `preserve_access_index` attribute to all structs and
unions. This makes all such structs and unions automatically relocatable under
BPF CO-RE, which improves user experience of writing TRACING programs with
direct kernel memory read access.

Also, further clean up selftest/bpf Makefile output and make it conforming to
libbpf and bpftool succinct output format.

v1->v2:
- build in-tree bpftool for runqslower (Yonghong);
- drop `format core` and augment `format c` instead (Alexei);
- move runqslower under tools/bpf (Daniel).
====================

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
4 years agoselftests/bpf: Build runqslower from selftests
Andrii Nakryiko [Mon, 13 Jan 2020 07:31:43 +0000 (23:31 -0800)]
selftests/bpf: Build runqslower from selftests

Ensure runqslower tool is built as part of selftests to prevent it from bit
rotting.

Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20200113073143.1779940-7-andriin@fb.com
4 years agotools/bpf: Add runqslower tool to tools/bpf
Andrii Nakryiko [Mon, 13 Jan 2020 07:31:42 +0000 (23:31 -0800)]
tools/bpf: Add runqslower tool to tools/bpf

Convert one of BCC tools (runqslower [0]) to BPF CO-RE + libbpf. It matches
its BCC-based counterpart 1-to-1, supporting all the same parameters and
functionality.

runqslower tool utilizes BPF skeleton, auto-generated from BPF object file,
as well as memory-mapped interface to global (read-only, in this case) data.
Its Makefile also ensures auto-generation of "relocatable" vmlinux.h, which is
necessary for BTF-typed raw tracepoints with direct memory access.

  [0] https://github.com/iovisor/bcc/blob/11bf5d02c895df9646c117c713082eb192825293/tools/runqslower.py

Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20200113073143.1779940-6-andriin@fb.com
4 years agobpftool: Apply preserve_access_index attribute to all types in BTF dump
Andrii Nakryiko [Mon, 13 Jan 2020 07:31:41 +0000 (23:31 -0800)]
bpftool: Apply preserve_access_index attribute to all types in BTF dump

This patch makes structs and unions, emitted through BTF dump, automatically
CO-RE-relocatable (unless disabled with `#define BPF_NO_PRESERVE_ACCESS_INDEX`,
specified before including generated header file).

This effectivaly turns usual bpf_probe_read() call into equivalent of
bpf_core_read(), by automatically applying builtin_preserve_access_index to
any field accesses of types in generated C types header.

This is especially useful for tp_btf/fentry/fexit BPF program types. They
allow direct memory access, so BPF C code just uses straightfoward a->b->c
access pattern to read data from kernel. But without kernel structs marked as
CO-RE relocatable through preserve_access_index attribute, one has to enclose
all the data reads into a special __builtin_preserve_access_index code block,
like so:

__builtin_preserve_access_index(({
    x = p->pid; /* where p is struct task_struct *, for example */
}));

This is very inconvenient and obscures the logic quite a bit. By marking all
auto-generated types with preserve_access_index attribute the above code is
reduced to just a clean and natural `x = p->pid;`.

Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20200113073143.1779940-5-andriin@fb.com
4 years agoselftests/bpf: Conform selftests/bpf Makefile output to libbpf and bpftool
Andrii Nakryiko [Mon, 13 Jan 2020 07:31:40 +0000 (23:31 -0800)]
selftests/bpf: Conform selftests/bpf Makefile output to libbpf and bpftool

Bring selftest/bpf's Makefile output to the same format used by libbpf and
bpftool: 2 spaces of padding on the left + 8-character left-aligned build step
identifier.

Also, hide feature detection output by default. Can be enabled back by setting
V=1.

Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20200113073143.1779940-4-andriin@fb.com
4 years agolibbpf: Clean up bpf_helper_defs.h generation output
Andrii Nakryiko [Mon, 13 Jan 2020 07:31:39 +0000 (23:31 -0800)]
libbpf: Clean up bpf_helper_defs.h generation output

bpf_helpers_doc.py script, used to generate bpf_helper_defs.h, unconditionally
emits one informational message to stderr. Remove it and preserve stderr to
contain only relevant errors. Also make sure script invocations command is
muted by default in libbpf's Makefile.

Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20200113073143.1779940-3-andriin@fb.com
4 years agotools: Sync uapi/linux/if_link.h
Andrii Nakryiko [Mon, 13 Jan 2020 07:31:38 +0000 (23:31 -0800)]
tools: Sync uapi/linux/if_link.h

Sync uapi/linux/if_link.h into tools to avoid out of sync warnings during
libbpf build.

Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20200113073143.1779940-2-andriin@fb.com
4 years agoselftests/bpf: Add BPF_PROG, BPF_KPROBE, and BPF_KRETPROBE macros
Andrii Nakryiko [Fri, 10 Jan 2020 21:16:34 +0000 (13:16 -0800)]
selftests/bpf: Add BPF_PROG, BPF_KPROBE, and BPF_KRETPROBE macros

Streamline BPF_TRACE_x macro by moving out return type and section attribute
definition out of macro itself. That makes those function look in source code
similar to other BPF programs. Additionally, simplify its usage by determining
number of arguments automatically (so just single BPF_TRACE vs a family of
BPF_TRACE_1, BPF_TRACE_2, etc). Also, allow more natural function argument
syntax without commas inbetween argument type and name.

Given this helper is useful not only for tracing tp_btf/fenty/fexit programs,
but could be used for LSM programs and others following the same pattern,
rename BPF_TRACE macro into more generic BPF_PROG. Existing BPF_TRACE_x
usages in selftests are converted to new BPF_PROG macro.

Following the same pattern, define BPF_KPROBE and BPF_KRETPROBE macros for
nicer usage of kprobe/kretprobe arguments, respectively. BPF_KRETPROBE, adopts
same convention used by fexit programs, that last defined argument is probed
function's return result.

v4->v5:
- fix test_overhead test (__set_task_comm is void) (Alexei);

v3->v4:
- rebased and fixed one more BPF_TRACE_x occurence (Alexei);

v2->v3:
- rename to shorter and as generic BPF_PROG (Alexei);

v1->v2:
- verified GCC handles pragmas as expected;
- added descriptions to macros;
- converted new STRUCT_OPS selftest to BPF_HANDLER (worked as expected);
- added original context as 'ctx' parameter, for cases where it has to be
  passed into BPF helpers. This might cause an accidental naming collision,
  unfortunately, but at least it's easy to work around. Fortunately, this
  situation produces quite legible compilation error:

progs/bpf_dctcp.c:46:6: error: redefinition of 'ctx' with a different type: 'int' vs 'unsigned long long *'
        int ctx = 123;
            ^
progs/bpf_dctcp.c:42:6: note: previous definition is here
void BPF_HANDLER(dctcp_init, struct sock *sk)
     ^
./bpf_trace_helpers.h:58:32: note: expanded from macro 'BPF_HANDLER'
____##name(unsigned long long *ctx, ##args)

Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Link: https://lore.kernel.org/bpf/20200110211634.1614739-1-andriin@fb.com
4 years agolibbpf: Poison kernel-only integer types
Andrii Nakryiko [Fri, 10 Jan 2020 18:19:16 +0000 (10:19 -0800)]
libbpf: Poison kernel-only integer types

It's been a recurring issue with types like u32 slipping into libbpf source
code accidentally. This is not detected during builds inside kernel source
tree, but becomes a compilation error in libbpf's Github repo. Libbpf is
supposed to use only __{s,u}{8,16,32,64} typedefs, so poison {s,u}{8,16,32,64}
explicitly in every .c file. Doing that in a bit more centralized way, e.g.,
inside libbpf_internal.h breaks selftests, which are both using kernel u32 and
libbpf_internal.h.

This patch also fixes a new u32 occurence in libbpf.c, added recently.

Fixes: 590a00888250 ("bpf: libbpf: Add STRUCT_OPS support")
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Link: https://lore.kernel.org/bpf/20200110181916.271446-1-andriin@fb.com
4 years agoMerge branch 'bpf-global-funcs'
Daniel Borkmann [Fri, 10 Jan 2020 16:20:07 +0000 (17:20 +0100)]
Merge branch 'bpf-global-funcs'

Alexei Starovoitov says:

====================
Introduce static vs global functions and function by function verification.
This is another step toward dynamic re-linking (or replacement) of global
functions. See patch 2 for details.

v2->v3:
- cleaned up a check spotted by Song.
- rebased and dropped patch 2 that was trying to improve BTF based on ELF.
- added one more unit test for scalar return value from global func.

v1->v2:
- addressed review comments from Song, Andrii, Yonghong
- fixed memory leak in error path
- added modified ctx check
- added more tests in patch 7
====================

Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
4 years agoselftests/bpf: Add unit tests for global functions
Alexei Starovoitov [Fri, 10 Jan 2020 06:41:24 +0000 (22:41 -0800)]
selftests/bpf: Add unit tests for global functions

test_global_func[12] - check 512 stack limit.
test_global_func[34] - check 8 frame call chain limit.
test_global_func5    - check that non-ctx pointer cannot be passed into
                       a function that expects context.
test_global_func6    - check that ctx pointer is unmodified.
test_global_func7    - check that global function returns scalar.

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Song Liu <songliubraving@fb.com>
Link: https://lore.kernel.org/bpf/20200110064124.1760511-7-ast@kernel.org
4 years agoselftests/bpf: Modify a test to check global functions
Alexei Starovoitov [Fri, 10 Jan 2020 06:41:23 +0000 (22:41 -0800)]
selftests/bpf: Modify a test to check global functions

Make two static functions in test_xdp_noinline.c global:
before: processed 2790 insns
after: processed 2598 insns

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Song Liu <songliubraving@fb.com>
Link: https://lore.kernel.org/bpf/20200110064124.1760511-6-ast@kernel.org
4 years agoselftests/bpf: Add a test for a large global function
Alexei Starovoitov [Fri, 10 Jan 2020 06:41:22 +0000 (22:41 -0800)]
selftests/bpf: Add a test for a large global function

test results:
pyperf50 with always_inlined the same function five times: processed 46378 insns
pyperf50 with global function: processed 6102 insns

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Song Liu <songliubraving@fb.com>
Link: https://lore.kernel.org/bpf/20200110064124.1760511-5-ast@kernel.org
4 years agoselftests/bpf: Add fexit-to-skb test for global funcs
Alexei Starovoitov [Fri, 10 Jan 2020 06:41:21 +0000 (22:41 -0800)]
selftests/bpf: Add fexit-to-skb test for global funcs

Add simple fexit prog type to skb prog type test when subprogram is a global
function.

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Song Liu <songliubraving@fb.com>
Link: https://lore.kernel.org/bpf/20200110064124.1760511-4-ast@kernel.org
4 years agobpf: Introduce function-by-function verification
Alexei Starovoitov [Fri, 10 Jan 2020 06:41:20 +0000 (22:41 -0800)]
bpf: Introduce function-by-function verification

New llvm and old llvm with libbpf help produce BTF that distinguish global and
static functions. Unlike arguments of static function the arguments of global
functions cannot be removed or optimized away by llvm. The compiler has to use
exactly the arguments specified in a function prototype. The argument type
information allows the verifier validate each global function independently.
For now only supported argument types are pointer to context and scalars. In
the future pointers to structures, sizes, pointer to packet data can be
supported as well. Consider the following example:

static int f1(int ...)
{
  ...
}

int f3(int b);

int f2(int a)
{
  f1(a) + f3(a);
}

int f3(int b)
{
  ...
}

int main(...)
{
  f1(...) + f2(...) + f3(...);
}

The verifier will start its safety checks from the first global function f2().
It will recursively descend into f1() because it's static. Then it will check
that arguments match for the f3() invocation inside f2(). It will not descend
into f3(). It will finish f2() that has to be successfully verified for all
possible values of 'a'. Then it will proceed with f3(). That function also has
to be safe for all possible values of 'b'. Then it will start subprog 0 (which
is main() function). It will recursively descend into f1() and will skip full
check of f2() and f3(), since they are global. The order of processing global
functions doesn't affect safety, since all global functions must be proven safe
based on their arguments only.

Such function by function verification can drastically improve speed of the
verification and reduce complexity.

Note that the stack limit of 512 still applies to the call chain regardless whether
functions were static or global. The nested level of 8 also still applies. The
same recursion prevention checks are in place as well.

The type information and static/global kind is preserved after the verification
hence in the above example global function f2() and f3() can be replaced later
by equivalent functions with the same types that are loaded and verified later
without affecting safety of this main() program. Such replacement (re-linking)
of global functions is a subject of future patches.

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Song Liu <songliubraving@fb.com>
Link: https://lore.kernel.org/bpf/20200110064124.1760511-3-ast@kernel.org
4 years agolibbpf: Sanitize global functions
Alexei Starovoitov [Fri, 10 Jan 2020 06:41:19 +0000 (22:41 -0800)]
libbpf: Sanitize global functions

In case the kernel doesn't support BTF_FUNC_GLOBAL sanitize BTF produced by the
compiler for global functions.

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Song Liu <songliubraving@fb.com>
Link: https://lore.kernel.org/bpf/20200110064124.1760511-2-ast@kernel.org
4 years agoMerge branch 'selftest-makefile-cleanup'
Alexei Starovoitov [Fri, 10 Jan 2020 05:55:08 +0000 (21:55 -0800)]
Merge branch 'selftest-makefile-cleanup'

Andrii Nakryiko says:

====================
Fix issues with bpf_helper_defs.h usage in selftests/bpf. As part of that, fix
the way clean up is performed for libbpf and selftests/bpf. Some for Makefile
output clean ups as well.
====================

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
4 years agoselftests/bpf: Further clean up Makefile output
Andrii Nakryiko [Fri, 10 Jan 2020 05:17:16 +0000 (21:17 -0800)]
selftests/bpf: Further clean up Makefile output

Further clean up Makefile output:
- hide "entering directory" messages;
- silvence sub-Make command echoing;
- succinct MKDIR messages.

Also remove few test binaries that are not produced anymore from .gitignore.

Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20200110051716.1591485-4-andriin@fb.com
4 years agoselftests/bpf: Ensure bpf_helper_defs.h are taken from selftests dir
Andrii Nakryiko [Fri, 10 Jan 2020 05:17:15 +0000 (21:17 -0800)]
selftests/bpf: Ensure bpf_helper_defs.h are taken from selftests dir

Reorder includes search path to ensure $(OUTPUT) and $(CURDIR) go before
libbpf's directory. Also fix bpf_helpers.h to include bpf_helper_defs.h in
such a way as to leverage includes search path. This allows selftests to not
use libbpf's local and potentially stale bpf_helper_defs.h. It's important
because selftests/bpf's Makefile only re-generates bpf_helper_defs.h in
seltests' output directory, not the one in libbpf's directory.

Also force regeneration of bpf_helper_defs.h when libbpf.a is updated to
reduce staleness.

Fixes: fa633a0f8919 ("libbpf: Fix build on read-only filesystems")
Reported-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20200110051716.1591485-3-andriin@fb.com
4 years agolibbpf,selftests/bpf: Fix clean targets
Andrii Nakryiko [Fri, 10 Jan 2020 05:17:14 +0000 (21:17 -0800)]
libbpf,selftests/bpf: Fix clean targets

Libbpf's clean target should clean out generated files in $(OUTPUT) directory
and not make assumption that $(OUTPUT) directory is current working directory.

Selftest's Makefile should delegate cleaning of libbpf-generated files to
libbpf's Makefile. This ensures more robust clean up.

Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20200110051716.1591485-2-andriin@fb.com
4 years agolibbpf: Make bpf_map order and indices stable
Andrii Nakryiko [Fri, 10 Jan 2020 03:42:46 +0000 (19:42 -0800)]
libbpf: Make bpf_map order and indices stable

Currently, libbpf re-sorts bpf_map structs after all the maps are added and
initialized, which might change their relative order and invalidate any
bpf_map pointer or index taken before that. This is inconvenient and
error-prone. For instance, it can cause .kconfig map index to point to a wrong
map.

Furthermore, libbpf itself doesn't rely on any specific ordering of bpf_maps,
so it's just an unnecessary complication right now. This patch drops sorting
of maps and makes their relative positions fixed. If efficient index is ever
needed, it's better to have a separate array of pointers as a search index,
instead of reordering bpf_map struct in-place. This will be less error-prone
and will allow multiple independent orderings, if necessary (e.g., either by
section index or by name).

Fixes: 166750bc1dd2 ("libbpf: Support libbpf-provided extern variables")
Reported-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20200110034247.1220142-1-andriin@fb.com
4 years agobpf: Document BPF_F_QUERY_EFFECTIVE flag
Andrey Ignatov [Wed, 8 Jan 2020 01:40:06 +0000 (17:40 -0800)]
bpf: Document BPF_F_QUERY_EFFECTIVE flag

Document BPF_F_QUERY_EFFECTIVE flag, mostly to clarify how it affects
attach_flags what may not be obvious and what may lead to confision.

Specifically attach_flags is returned only for target_fd but if programs
are inherited from an ancestor cgroup then returned attach_flags for
current cgroup may be confusing. For example, two effective programs of
same attach_type can be returned but w/o BPF_F_ALLOW_MULTI in
attach_flags.

Simple repro:
  # bpftool c s /sys/fs/cgroup/path/to/task
  ID       AttachType      AttachFlags     Name
  # bpftool c s /sys/fs/cgroup/path/to/task effective
  ID       AttachType      AttachFlags     Name
  95043    ingress                         tw_ipt_ingress
  95048    ingress                         tw_ingress

Signed-off-by: Andrey Ignatov <rdna@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Song Liu <songliubraving@fb.com>
Link: https://lore.kernel.org/bpf/20200108014006.938363-1-rdna@fb.com
4 years agoMerge branch 'tcp-bpf-cc'
Alexei Starovoitov [Thu, 9 Jan 2020 16:46:19 +0000 (08:46 -0800)]
Merge branch 'tcp-bpf-cc'

Martin Lau says:

====================
This series introduces BPF STRUCT_OPS.  It is an infra to allow
implementing some specific kernel's function pointers in BPF.
The first use case included in this series is to implement
TCP congestion control algorithm in BPF  (i.e. implement
struct tcp_congestion_ops in BPF).

There has been attempt to move the TCP CC to the user space
(e.g. CCP in TCP).   The common arguments are faster turn around,
get away from long-tail kernel versions in production...etc,
which are legit points.

BPF has been the continuous effort to join both kernel and
userspace upsides together (e.g. XDP to gain the performance
advantage without bypassing the kernel).  The recent BPF
advancements (in particular BTF-aware verifier, BPF trampoline,
BPF CO-RE...) made implementing kernel struct ops (e.g. tcp cc)
possible in BPF.

The idea is to allow implementing tcp_congestion_ops in bpf.
It allows a faster turnaround for testing algorithm in the
production while leveraging the existing (and continue growing) BPF
feature/framework instead of building one specifically for
userspace TCP CC.

Please see individual patch for details.

The bpftool support will be posted in follow-up patches.

v4:
- Expose tcp_ca_find() to tcp.h in patch 7.
  It is used to check the same bpf-tcp-cc
  does not exist to guarantee the register()
  will succeed.
- set_memory_ro() and then set_memory_x() only after all
  trampolines are written to the image in patch 6. (Daniel)
  spinlock is replaced by mutex because set_memory_*
  requires sleepable context.

v3:
- Fix kbuild error by considering CONFIG_BPF_SYSCALL (kbuild)
- Support anonymous bitfield in patch 4 (Andrii, Yonghong)
- Push boundary safety check to a specific arch's trampoline function
  (in patch 6) (Yonghong).
  Reuse the WANR_ON_ONCE check in arch_prepare_bpf_trampoline() in x86.
- Check module field is 0 in udata in patch 6 (Yonghong)
- Check zero holes in patch 6 (Andrii)
- s/_btf_vmlinux/btf/ in patch 5 and 7 (Andrii)
- s/check_xxx/is_xxx/ in patch 7 (Andrii)
- Use "struct_ops/" convention in patch 11 (Andrii)
- Use the skel instead of bpf_object in patch 11 (Andrii)
- libbpf: Decide BPF_PROG_TYPE_STRUCT_OPS at open phase by using
          find_sec_def()
- libbpf: Avoid a debug message at open phase (Andrii)
- libbpf: Add bpf_program__(is|set)_struct_ops() for consistency (Andrii)
- libbpf: Add "struct_ops" to section_defs (Andrii)
- libbpf: Some code shuffling in init_kern_struct_ops() (Andrii)
- libbpf: A few safety checks (Andrii)

v2:
- Dropped cubic for now.  They will be reposted
  once there are more clarity in "jiffies" on both
  bpf side (about the helper) and
  tcp_cubic side (some of jiffies usages are being replaced
  by tp->tcp_mstamp)
- Remove unnecssary check on bitfield support from btf_struct_access()
  (Yonghong)
- BTF_TYPE_EMIT macro (Yonghong, Andrii)
- value_name's length check to avoid an unlikely
  type match during truncation case (Yonghong)
- BUILD_BUG_ON to ensure no trampoline-image overrun
  in the future (Yonghong)
- Simplify get_next_key() (Yonghong)
- Added comment to explain how to check mandatory
  func ptr in net/ipv4/bpf_tcp_ca.c (Yonghong)
- Rename "__bpf_" to "bpf_struct_ops_" for value prefix (Andrii)
- Add comment to highlight the bpf_dctcp.c is not necessarily
  the same as tcp_dctcp.c. (Alexei, Eric)
- libbpf: Renmae "struct_ops" to ".struct_ops" for elf sec (Andrii)
- libbpf: Expose struct_ops as a bpf_map (Andrii)
- libbpf: Support multiple struct_ops in SEC(".struct_ops") (Andrii)
- libbpf: Add bpf_map__attach_struct_ops()  (Andrii)
====================

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
4 years agobpf: Add bpf_dctcp example
Martin KaFai Lau [Thu, 9 Jan 2020 00:35:17 +0000 (16:35 -0800)]
bpf: Add bpf_dctcp example

This patch adds a bpf_dctcp example.  It currently does not do
no-ECN fallback but the same could be done through the cgrp2-bpf.

Signed-off-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20200109003517.3856825-1-kafai@fb.com
4 years agobpf: libbpf: Add STRUCT_OPS support
Martin KaFai Lau [Thu, 9 Jan 2020 00:35:14 +0000 (16:35 -0800)]
bpf: libbpf: Add STRUCT_OPS support

This patch adds BPF STRUCT_OPS support to libbpf.

The only sec_name convention is SEC(".struct_ops") to identify the
struct_ops implemented in BPF,
e.g. To implement a tcp_congestion_ops:

SEC(".struct_ops")
struct tcp_congestion_ops dctcp = {
.init           = (void *)dctcp_init,  /* <-- a bpf_prog */
/* ... some more func prts ... */
.name           = "bpf_dctcp",
};

Each struct_ops is defined as a global variable under SEC(".struct_ops")
as above.  libbpf creates a map for each variable and the variable name
is the map's name.  Multiple struct_ops is supported under
SEC(".struct_ops").

In the bpf_object__open phase, libbpf will look for the SEC(".struct_ops")
section and find out what is the btf-type the struct_ops is
implementing.  Note that the btf-type here is referring to
a type in the bpf_prog.o's btf.  A "struct bpf_map" is added
by bpf_object__add_map() as other maps do.  It will then
collect (through SHT_REL) where are the bpf progs that the
func ptrs are referring to.  No btf_vmlinux is needed in
the open phase.

In the bpf_object__load phase, the map-fields, which depend
on the btf_vmlinux, are initialized (in bpf_map__init_kern_struct_ops()).
It will also set the prog->type, prog->attach_btf_id, and
prog->expected_attach_type.  Thus, the prog's properties do
not rely on its section name.
[ Currently, the bpf_prog's btf-type ==> btf_vmlinux's btf-type matching
  process is as simple as: member-name match + btf-kind match + size match.
  If these matching conditions fail, libbpf will reject.
  The current targeting support is "struct tcp_congestion_ops" which
  most of its members are function pointers.
  The member ordering of the bpf_prog's btf-type can be different from
  the btf_vmlinux's btf-type. ]

Then, all obj->maps are created as usual (in bpf_object__create_maps()).

Once the maps are created and prog's properties are all set,
the libbpf will proceed to load all the progs.

bpf_map__attach_struct_ops() is added to register a struct_ops
map to a kernel subsystem.

Signed-off-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20200109003514.3856730-1-kafai@fb.com
4 years agobpf: Synch uapi bpf.h to tools/
Martin KaFai Lau [Thu, 9 Jan 2020 00:35:12 +0000 (16:35 -0800)]
bpf: Synch uapi bpf.h to tools/

This patch sync uapi bpf.h to tools/

Signed-off-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20200109003512.3856559-1-kafai@fb.com
4 years agobpf: Add BPF_FUNC_tcp_send_ack helper
Martin KaFai Lau [Thu, 9 Jan 2020 00:45:51 +0000 (16:45 -0800)]
bpf: Add BPF_FUNC_tcp_send_ack helper

Add a helper to send out a tcp-ack.  It will be used in the later
bpf_dctcp implementation that requires to send out an ack
when the CE state changed.

Signed-off-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Yonghong Song <yhs@fb.com>
Link: https://lore.kernel.org/bpf/20200109004551.3900448-1-kafai@fb.com
4 years agobpf: tcp: Support tcp_congestion_ops in bpf
Martin KaFai Lau [Thu, 9 Jan 2020 00:35:08 +0000 (16:35 -0800)]
bpf: tcp: Support tcp_congestion_ops in bpf

This patch makes "struct tcp_congestion_ops" to be the first user
of BPF STRUCT_OPS.  It allows implementing a tcp_congestion_ops
in bpf.

The BPF implemented tcp_congestion_ops can be used like
regular kernel tcp-cc through sysctl and setsockopt.  e.g.
[root@arch-fb-vm1 bpf]# sysctl -a | egrep congestion
net.ipv4.tcp_allowed_congestion_control = reno cubic bpf_cubic
net.ipv4.tcp_available_congestion_control = reno bic cubic bpf_cubic
net.ipv4.tcp_congestion_control = bpf_cubic

There has been attempt to move the TCP CC to the user space
(e.g. CCP in TCP).   The common arguments are faster turn around,
get away from long-tail kernel versions in production...etc,
which are legit points.

BPF has been the continuous effort to join both kernel and
userspace upsides together (e.g. XDP to gain the performance
advantage without bypassing the kernel).  The recent BPF
advancements (in particular BTF-aware verifier, BPF trampoline,
BPF CO-RE...) made implementing kernel struct ops (e.g. tcp cc)
possible in BPF.  It allows a faster turnaround for testing algorithm
in the production while leveraging the existing (and continue growing)
BPF feature/framework instead of building one specifically for
userspace TCP CC.

This patch allows write access to a few fields in tcp-sock
(in bpf_tcp_ca_btf_struct_access()).

The optional "get_info" is unsupported now.  It can be added
later.  One possible way is to output the info with a btf-id
to describe the content.

Signed-off-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Yonghong Song <yhs@fb.com>
Link: https://lore.kernel.org/bpf/20200109003508.3856115-1-kafai@fb.com
4 years agobpf: Introduce BPF_MAP_TYPE_STRUCT_OPS
Martin KaFai Lau [Thu, 9 Jan 2020 00:35:05 +0000 (16:35 -0800)]
bpf: Introduce BPF_MAP_TYPE_STRUCT_OPS

The patch introduces BPF_MAP_TYPE_STRUCT_OPS.  The map value
is a kernel struct with its func ptr implemented in bpf prog.
This new map is the interface to register/unregister/introspect
a bpf implemented kernel struct.

The kernel struct is actually embedded inside another new struct
(or called the "value" struct in the code).  For example,
"struct tcp_congestion_ops" is embbeded in:
struct bpf_struct_ops_tcp_congestion_ops {
refcount_t refcnt;
enum bpf_struct_ops_state state;
struct tcp_congestion_ops data;  /* <-- kernel subsystem struct here */
}
The map value is "struct bpf_struct_ops_tcp_congestion_ops".
The "bpftool map dump" will then be able to show the
state ("inuse"/"tobefree") and the number of subsystem's refcnt (e.g.
number of tcp_sock in the tcp_congestion_ops case).  This "value" struct
is created automatically by a macro.  Having a separate "value" struct
will also make extending "struct bpf_struct_ops_XYZ" easier (e.g. adding
"void (*init)(void)" to "struct bpf_struct_ops_XYZ" to do some
initialization works before registering the struct_ops to the kernel
subsystem).  The libbpf will take care of finding and populating the
"struct bpf_struct_ops_XYZ" from "struct XYZ".

Register a struct_ops to a kernel subsystem:
1. Load all needed BPF_PROG_TYPE_STRUCT_OPS prog(s)
2. Create a BPF_MAP_TYPE_STRUCT_OPS with attr->btf_vmlinux_value_type_id
   set to the btf id "struct bpf_struct_ops_tcp_congestion_ops" of the
   running kernel.
   Instead of reusing the attr->btf_value_type_id,
   btf_vmlinux_value_type_id s added such that attr->btf_fd can still be
   used as the "user" btf which could store other useful sysadmin/debug
   info that may be introduced in the furture,
   e.g. creation-date/compiler-details/map-creator...etc.
3. Create a "struct bpf_struct_ops_tcp_congestion_ops" object as described
   in the running kernel btf.  Populate the value of this object.
   The function ptr should be populated with the prog fds.
4. Call BPF_MAP_UPDATE with the object created in (3) as
   the map value.  The key is always "0".

During BPF_MAP_UPDATE, the code that saves the kernel-func-ptr's
args as an array of u64 is generated.  BPF_MAP_UPDATE also allows
the specific struct_ops to do some final checks in "st_ops->init_member()"
(e.g. ensure all mandatory func ptrs are implemented).
If everything looks good, it will register this kernel struct
to the kernel subsystem.  The map will not allow further update
from this point.

Unregister a struct_ops from the kernel subsystem:
BPF_MAP_DELETE with key "0".

Introspect a struct_ops:
BPF_MAP_LOOKUP_ELEM with key "0".  The map value returned will
have the prog _id_ populated as the func ptr.

The map value state (enum bpf_struct_ops_state) will transit from:
INIT (map created) =>
INUSE (map updated, i.e. reg) =>
TOBEFREE (map value deleted, i.e. unreg)

The kernel subsystem needs to call bpf_struct_ops_get() and
bpf_struct_ops_put() to manage the "refcnt" in the
"struct bpf_struct_ops_XYZ".  This patch uses a separate refcnt
for the purose of tracking the subsystem usage.  Another approach
is to reuse the map->refcnt and then "show" (i.e. during map_lookup)
the subsystem's usage by doing map->refcnt - map->usercnt to filter out
the map-fd/pinned-map usage.  However, that will also tie down the
future semantics of map->refcnt and map->usercnt.

The very first subsystem's refcnt (during reg()) holds one
count to map->refcnt.  When the very last subsystem's refcnt
is gone, it will also release the map->refcnt.  All bpf_prog will be
freed when the map->refcnt reaches 0 (i.e. during map_free()).

Here is how the bpftool map command will look like:
[root@arch-fb-vm1 bpf]# bpftool map show
6: struct_ops  name dctcp  flags 0x0
key 4B  value 256B  max_entries 1  memlock 4096B
btf_id 6
[root@arch-fb-vm1 bpf]# bpftool map dump id 6
[{
        "value": {
            "refcnt": {
                "refs": {
                    "counter": 1
                }
            },
            "state": 1,
            "data": {
                "list": {
                    "next": 0,
                    "prev": 0
                },
                "key": 0,
                "flags": 2,
                "init": 24,
                "release": 0,
                "ssthresh": 25,
                "cong_avoid": 30,
                "set_state": 27,
                "cwnd_event": 28,
                "in_ack_event": 26,
                "undo_cwnd": 29,
                "pkts_acked": 0,
                "min_tso_segs": 0,
                "sndbuf_expand": 0,
                "cong_control": 0,
                "get_info": 0,
                "name": [98,112,102,95,100,99,116,99,112,0,0,0,0,0,0,0
                ],
                "owner": 0
            }
        }
    }
]

Misc Notes:
* bpf_struct_ops_map_sys_lookup_elem() is added for syscall lookup.
  It does an inplace update on "*value" instead returning a pointer
  to syscall.c.  Otherwise, it needs a separate copy of "zero" value
  for the BPF_STRUCT_OPS_STATE_INIT to avoid races.

* The bpf_struct_ops_map_delete_elem() is also called without
  preempt_disable() from map_delete_elem().  It is because
  the "->unreg()" may requires sleepable context, e.g.
  the "tcp_unregister_congestion_control()".

* "const" is added to some of the existing "struct btf_func_model *"
  function arg to avoid a compiler warning caused by this patch.

Signed-off-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Yonghong Song <yhs@fb.com>
Link: https://lore.kernel.org/bpf/20200109003505.3855919-1-kafai@fb.com
4 years agobpf: Introduce BPF_PROG_TYPE_STRUCT_OPS
Martin KaFai Lau [Thu, 9 Jan 2020 00:35:03 +0000 (16:35 -0800)]
bpf: Introduce BPF_PROG_TYPE_STRUCT_OPS

This patch allows the kernel's struct ops (i.e. func ptr) to be
implemented in BPF.  The first use case in this series is the
"struct tcp_congestion_ops" which will be introduced in a
latter patch.

This patch introduces a new prog type BPF_PROG_TYPE_STRUCT_OPS.
The BPF_PROG_TYPE_STRUCT_OPS prog is verified against a particular
func ptr of a kernel struct.  The attr->attach_btf_id is the btf id
of a kernel struct.  The attr->expected_attach_type is the member
"index" of that kernel struct.  The first member of a struct starts
with member index 0.  That will avoid ambiguity when a kernel struct
has multiple func ptrs with the same func signature.

For example, a BPF_PROG_TYPE_STRUCT_OPS prog is written
to implement the "init" func ptr of the "struct tcp_congestion_ops".
The attr->attach_btf_id is the btf id of the "struct tcp_congestion_ops"
of the _running_ kernel.  The attr->expected_attach_type is 3.

The ctx of BPF_PROG_TYPE_STRUCT_OPS is an array of u64 args saved
by arch_prepare_bpf_trampoline that will be done in the next
patch when introducing BPF_MAP_TYPE_STRUCT_OPS.

"struct bpf_struct_ops" is introduced as a common interface for the kernel
struct that supports BPF_PROG_TYPE_STRUCT_OPS prog.  The supporting kernel
struct will need to implement an instance of the "struct bpf_struct_ops".

The supporting kernel struct also needs to implement a bpf_verifier_ops.
During BPF_PROG_LOAD, bpf_struct_ops_find() will find the right
bpf_verifier_ops by searching the attr->attach_btf_id.

A new "btf_struct_access" is also added to the bpf_verifier_ops such
that the supporting kernel struct can optionally provide its own specific
check on accessing the func arg (e.g. provide limited write access).

After btf_vmlinux is parsed, the new bpf_struct_ops_init() is called
to initialize some values (e.g. the btf id of the supporting kernel
struct) and it can only be done once the btf_vmlinux is available.

The R0 checks at BPF_EXIT is excluded for the BPF_PROG_TYPE_STRUCT_OPS prog
if the return type of the prog->aux->attach_func_proto is "void".

Signed-off-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Yonghong Song <yhs@fb.com>
Link: https://lore.kernel.org/bpf/20200109003503.3855825-1-kafai@fb.com
4 years agobpf: Support bitfield read access in btf_struct_access
Martin KaFai Lau [Thu, 9 Jan 2020 00:35:01 +0000 (16:35 -0800)]
bpf: Support bitfield read access in btf_struct_access

This patch allows bitfield access as a scalar.

It checks "off + size > t->size" to avoid accessing bitfield
end up accessing beyond the struct.  This check is done
outside of the loop since it is applicable to all access.

It also takes this chance to break early on the "off < moff" case.

Signed-off-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Yonghong Song <yhs@fb.com>
Link: https://lore.kernel.org/bpf/20200109003501.3855427-1-kafai@fb.com
4 years agobpf: Add enum support to btf_ctx_access()
Martin KaFai Lau [Thu, 9 Jan 2020 00:34:59 +0000 (16:34 -0800)]
bpf: Add enum support to btf_ctx_access()

It allows bpf prog (e.g. tracing) to attach
to a kernel function that takes enum argument.

Signed-off-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Yonghong Song <yhs@fb.com>
Link: https://lore.kernel.org/bpf/20200109003459.3855366-1-kafai@fb.com
4 years agobpf: Avoid storing modifier to info->btf_id
Martin KaFai Lau [Thu, 9 Jan 2020 00:34:56 +0000 (16:34 -0800)]
bpf: Avoid storing modifier to info->btf_id

info->btf_id expects the btf_id of a struct, so it should
store the final result after skipping modifiers (if any).

It also takes this chanace to add a missing newline in one of the
bpf_log() messages.

Signed-off-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Yonghong Song <yhs@fb.com>
Link: https://lore.kernel.org/bpf/20200109003456.3855176-1-kafai@fb.com
4 years agobpf: Save PTR_TO_BTF_ID register state when spilling to stack
Martin KaFai Lau [Thu, 9 Jan 2020 00:34:54 +0000 (16:34 -0800)]
bpf: Save PTR_TO_BTF_ID register state when spilling to stack

This patch makes the verifier save the PTR_TO_BTF_ID register state when
spilling to the stack.

Signed-off-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Yonghong Song <yhs@fb.com>
Link: https://lore.kernel.org/bpf/20200109003454.3854870-1-kafai@fb.com
4 years agoselftests/bpf: Restore original comm in test_overhead
Stanislav Fomichev [Wed, 8 Jan 2020 19:21:32 +0000 (11:21 -0800)]
selftests/bpf: Restore original comm in test_overhead

test_overhead changes task comm in order to estimate BPF trampoline
overhead but never sets the comm back to the original one.
We have the tests (like core_reloc.c) that have 'test_progs'
as hard-coded expected comm, so let's try to preserve the
original comm.

Currently, everything works because the order of execution is:
first core_recloc, then test_overhead; but let's make it a bit
future-proof.

Other related changes: use 'test_overhead' as new comm instead of
'test' to make it easy to debug and drop '\n' at the end.

Signed-off-by: Stanislav Fomichev <sdf@google.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Petar Penkov <ppenkov@google.com>
Acked-by: Song Liu <songliubraving@fb.com>
Link: https://lore.kernel.org/bpf/20200108192132.189221-1-sdf@google.com
4 years agobpftool: Add misc section and probe for large INSN limit
Michal Rostecki [Wed, 8 Jan 2020 16:23:53 +0000 (17:23 +0100)]
bpftool: Add misc section and probe for large INSN limit

Introduce a new probe section (misc) for probes not related to concrete
map types, program types, functions or kernel configuration. Introduce a
probe for large INSN limit as the first one in that section.

Example outputs:

  # bpftool feature probe
  [...]
  Scanning miscellaneous eBPF features...
  Large program size limit is available

  # bpftool feature probe macros
  [...]
  /*** eBPF misc features ***/
  #define HAVE_HAVE_LARGE_INSN_LIMIT

  # bpftool feature probe -j | jq '.["misc"]'
  {
    "have_large_insn_limit": true
  }

Signed-off-by: Michal Rostecki <mrostecki@opensuse.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: Quentin Monnet <quentin.monnet@netronome.com>
Link: https://lore.kernel.org/bpf/20200108162428.25014-3-mrostecki@opensuse.org
4 years agolibbpf: Add probe for large INSN limit
Michal Rostecki [Wed, 8 Jan 2020 16:23:52 +0000 (17:23 +0100)]
libbpf: Add probe for large INSN limit

Introduce a new probe which checks whether kernel has large maximum
program size which was increased in the following commit:

c04c0d2b968a ("bpf: increase complexity limit and maximum program size")

Based on the similar check in Cilium[0], authored by Daniel Borkmann.

  [0] https://github.com/cilium/cilium/commit/657d0f585afd26232cfa5d4e70b6f64d2ea91596

Co-authored-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Michal Rostecki <mrostecki@opensuse.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: Quentin Monnet <quentin.monnet@netronome.com>
Link: https://lore.kernel.org/bpf/20200108162428.25014-2-mrostecki@opensuse.org