It is allocated with kvmalloc(), the corresponding release function
should not be kfree(), use kvfree() instead.

Generated by: scripts/coccinelle/api/kfree_mismatch.cocci

Fixes: b530e9e1

 ("bpf: Add "live packet" mode for XDP in BPF_PROG_RUN")
Signed-off-by: Yihao Han <hanyihao@vivo.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Cc: Toke Høiland-Jørgensen <toke@redhat.com>
Link: https://lore.kernel.org/bpf/20220310092828.13405-1-hanyihao@vivo.com

The kernel test robot pointed out that the newly added
bpf_test_run_xdp_live() runner doesn't set the retval in the caller (by
design), which means that the variable can be passed unitialised to
bpf_test_finish(). Fix this by initialising the variable properly.

Fixes: b530e9e1

 ("bpf: Add "live packet" mode for XDP in BPF_PROG_RUN")
Reported-by: kernel test robot <lkp@intel.com>
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Toke Høiland-Jørgensen <toke@redhat.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20220310110228.161869-1-toke@redhat.com

Commit 1a56c18e

 ("bpftool: Stop supporting BPF offload-enabled
feature probing") removed the support to probe for BPF offload features.
This is still something that is useful for NFP NIC that can support
offloading of BPF programs.

The reason for the dropped support was that libbpf starting with v1.0
would drop support for passing the ifindex to the BPF prog/map/helper
feature probing APIs. In order to keep this useful feature for NFP
restore the functionality by moving it directly into bpftool.

The code restored is a simplified version of the code that existed in
libbpf which supposed passing the ifindex. The simplification is that it
only targets the cases where ifindex is given and call into libbpf for
the cases where it's not.

Before restoring support for probing offload features:

  # bpftool feature probe dev ens4np0
  Scanning system call availability...
  bpf() syscall is available

  Scanning eBPF program types...

  Scanning eBPF map types...

  Scanning eBPF helper functions...
  eBPF helpers supported for program type sched_cls:
  eBPF helpers supported for program type xdp:

  Scanning miscellaneous eBPF features...
  Large program size limit is NOT available
  Bounded loop support is NOT available
  ISA extension v2 is NOT available
  ISA extension v3 is NOT available

With support for probing offload features restored:

  # bpftool feature probe dev ens4np0
  Scanning system call availability...
  bpf() syscall is available

  Scanning eBPF program types...
  eBPF program_type sched_cls is available
  eBPF program_type xdp is available

  Scanning eBPF map types...
  eBPF map_type hash is available
  eBPF map_type array is available

  Scanning eBPF helper functions...
  eBPF helpers supported for program type sched_cls:
  	- bpf_map_lookup_elem
  	- bpf_get_prandom_u32
  	- bpf_perf_event_output
  eBPF helpers supported for program type xdp:
  	- bpf_map_lookup_elem
  	- bpf_get_prandom_u32
  	- bpf_perf_event_output
  	- bpf_xdp_adjust_head
  	- bpf_xdp_adjust_tail

  Scanning miscellaneous eBPF features...
  Large program size limit is NOT available
  Bounded loop support is NOT available
  ISA extension v2 is NOT available
  ISA extension v3 is NOT available

Signed-off-by: Niklas Söderlund <niklas.soderlund@corigine.com>
Signed-off-by: Simon Horman <simon.horman@corigine.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: Quentin Monnet <quentin@isovalent.com>
Link: https://lore.kernel.org/bpf/20220310121846.921256-1-niklas.soderlund@corigine.com

Toke Høiland-Jørgensen says:

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

This series adds support for transmitting packets using XDP in
bpf_prog_run(), by enabling a new mode "live packet" mode which will handle
the XDP program return codes and redirect the packets to the stack or other
devices.

The primary use case for this is testing the redirect map types and the
ndo_xdp_xmit driver operation without an external traffic generator. But it
turns out to also be useful for creating a programmable traffic generator
in XDP, as well as injecting frames into the stack. A sample traffic
generator, which was included in previous versions of the series, but now
moved to xdp-tools, transmits up to 9 Mpps/core on my test machine.

To transmit the frames, the new mode instantiates a page_pool structure in
bpf_prog_run() and initialises the pages to contain XDP frames with the
data passed in by userspace. These frames can then be handled as though
they came from the hardware XDP path, and the existing page_pool code takes
care of returning and recycling them. The setup is optimised for high
performance with a high number of repetitions to support stress testing and
the traffic generator use case; see patch 1 for details.

v11:
- Fix override of return code in xdp_test_run_batch()
- Add Martin's ACKs to remaining patches

v10:
- Only propagate memory allocation errors from xdp_test_run_batch()
- Get rid of BPF_F_TEST_XDP_RESERVED; batch_size can be used to probe
- Check that batch_size is unset in non-XDP test_run funcs
- Lower the number of repetitions in the selftest to 10k
- Count number of recycled pages in the selftest
- Fix a few other nits from Martin, carry forward ACKs

v9:
- XDP_DROP packets in the selftest to ensure pages are recycled
- Fix a few issues reported by the kernel test robot
- Rewrite the documentation of the batch size to make it a bit clearer
- Rebase to newest bpf-next

v8:
- Make the batch size configurable from userspace
- Don't interrupt the packet loop on errors in do_redirect (this can be
  caught from the tracepoint)
- Add documentation of the feature
- Add reserved flag userspace can use to probe for support (kernel didn't
  check flags previously)
- Rebase to newest bpf-next, disallow live mode for jumbo frames

v7:
- Extend the local_bh_disable() to cover the full test run loop, to prevent
  running concurrently with the softirq. Fixes a deadlock with veth xmit.
- Reinstate the forwarding sysctl setting in the selftest, and bump up the
  number of packets being transmitted to trigger the above bug.
- Update commit message to make it clear that user space can select the
  ingress interface.

v6:
- Fix meta vs data pointer setting and add a selftest for it
- Add local_bh_disable() around code passing packets up the stack
- Create a new netns for the selftest and use a TC program instead of the
  forwarding hack to count packets being XDP_PASS'ed from the test prog.
- Check for the correct ingress ifindex in the selftest
- Rebase and drop patches 1-5 that were already merged

v5:
- Rebase to current bpf-next

v4:
- Fix a few code style issues (Alexei)
- Also handle the other return codes: XDP_PASS builds skbs and injects them
  into the stack, and XDP_TX is turned into a redirect out the same
  interface (Alexei).
- Drop the last patch adding an xdp_trafficgen program to samples/bpf; this
  will live in xdp-tools instead (Alexei).
- Add a separate bpf_test_run_xdp_live() function to test_run.c instead of
  entangling the new mode in the existing bpf_test_run().

v3:
- Reorder patches to make sure they all build individually (Patchwork)
- Remove a couple of unused variables (Patchwork)
- Remove unlikely() annotation in slow path and add back John's ACK that I
  accidentally dropped for v2 (John)

v2:
- Split up up __xdp_do_redirect to avoid passing two pointers to it (John)
- Always reset context pointers before each test run (John)
- Use get_mac_addr() from xdp_sample_user.h instead of rolling our own (Kumar)
- Fix wrong offset for metadata pointer
====================

Signed-off-by: Alexei Starovoitov <ast@kernel.org>

This adds a selftest for the XDP_REDIRECT facility in BPF_PROG_RUN, that
redirects packets into a veth and counts them using an XDP program on the
other side of the veth pair and a TC program on the local side of the veth.

Signed-off-by: Toke Høiland-Jørgensen <toke@redhat.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Link: https://lore.kernel.org/bpf/20220309105346.100053-6-toke@redhat.com

These will also be used by the xdp_do_redirect test being added in the next
commit.

Signed-off-by: Toke Høiland-Jørgensen <toke@redhat.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Link: https://lore.kernel.org/bpf/20220309105346.100053-5-toke@redhat.com

Add support for setting the new batch_size parameter to BPF_PROG_TEST_RUN
to libbpf; just add it as an option and pass it through to the kernel.

Signed-off-by: Toke Høiland-Jørgensen <toke@redhat.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Link: https://lore.kernel.org/bpf/20220309105346.100053-4-toke@redhat.com

This adds documentation for the BPF_PROG_RUN command; a short overview of
the command itself, and a more verbose description of the "live packet"
mode for XDP introduced in the previous commit.

Signed-off-by: Toke Høiland-Jørgensen <toke@redhat.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Link: https://lore.kernel.org/bpf/20220309105346.100053-3-toke@redhat.com

This adds support for running XDP programs through BPF_PROG_RUN in a mode
that enables live packet processing of the resulting frames. Previous uses
of BPF_PROG_RUN for XDP returned the XDP program return code and the
modified packet data to userspace, which is useful for unit testing of XDP
programs.

The existing BPF_PROG_RUN for XDP allows userspace to set the ingress
ifindex and RXQ number as part of the context object being passed to the
kernel. This patch reuses that code, but adds a new mode with different
semantics, which can be selected with the new BPF_F_TEST_XDP_LIVE_FRAMES
flag.

When running BPF_PROG_RUN in this mode, the XDP program return codes will
be honoured: returning XDP_PASS will result in the frame being injected
into the networking stack as if it came from the selected networking
interface, while returning XDP_TX and XDP_REDIRECT will result in the frame
being transmitted out that interface. XDP_TX is translated into an
XDP_REDIRECT operation to the same interface, since the real XDP_TX action
is only possible from within the network drivers themselves, not from the
process context where BPF_PROG_RUN is executed.

Internally, this new mode of operation creates a page pool instance while
setting up the test run, and feeds pages from that into the XDP program.
The setup cost of this is amortised over the number of repetitions
specified by userspace.

To support the performance testing use case, we further optimise the setup
step so that all pages in the pool are pre-initialised with the packet
data, and pre-computed context and xdp_frame objects stored at the start of
each page. This makes it possible to entirely avoid touching the page
content on each XDP program invocation, and enables sending up to 9
Mpps/core on my test box.

Because the data pages are recycled by the page pool, and the test runner
doesn't re-initialise them for each run, subsequent invocations of the XDP
program will see the packet data in the state it was after the last time it
ran on that particular page. This means that an XDP program that modifies
the packet before redirecting it has to be careful about which assumptions
it makes about the packet content, but that is only an issue for the most
naively written programs.

Enabling the new flag is only allowed when not setting ctx_out and data_out
in the test specification, since using it means frames will be redirected
somewhere else, so they can't be returned.

Signed-off-by: Toke Høiland-Jørgensen <toke@redhat.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Link: https://lore.kernel.org/bpf/20220309105346.100053-2-toke@redhat.com

Mykola Lysenko says:

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

First patch reduces the sample_freq to 1000 to ensure test will
work even when kernel.perf_event_max_sample_rate was reduced to 1000.

Patches for send_signal and find_vma tune the test implementation to
make sure needed thread is scheduled. Also, both tests will finish as
soon as possible after the test condition is met.
====================

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>

Remove unneeded spleep and increase length of dummy CPU
intensive computation to guarantee test process execution.
Also, complete aforemention computation as soon as
test success criteria is met

Signed-off-by: Mykola Lysenko <mykolal@fb.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Acked-by: Yonghong Song <yhs@fb.com>
Link: https://lore.kernel.org/bpf/20220308200449.1757478-4-mykolal@fb.com

Substitute sleep with dummy CPU intensive computation.
Finish aforemention computation as soon as signal was
delivered to the test process. Make the BPF code to
only execute when PID global variable is set

Signed-off-by: Mykola Lysenko <mykolal@fb.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Acked-by: Yonghong Song <yhs@fb.com>
Link: https://lore.kernel.org/bpf/20220308200449.1757478-3-mykolal@fb.com

Linux kernel may automatically reduce kernel.perf_event_max_sample_rate
value when running tests in parallel on slow systems. Linux kernel checks
against this limit when opening perf event with freq=1 parameter set.
The lower bound is 1000. This patch reduces sample_freq value to 1000
in all BPF tests that use sample_freq to ensure they always can open
perf event.

Signed-off-by: Mykola Lysenko <mykolal@fb.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Acked-by: Yonghong Song <yhs@fb.com>
Link: https://lore.kernel.org/bpf/20220308200449.1757478-2-mykolal@fb.com

In ChromeOS and Gentoo we catch any unwanted mixed Clang/LLVM
and GCC/binutils usage via toolchain wrappers which fail builds.
This has revealed that GCC is called unconditionally in Clang
configured builds to populate GCC_TOOLCHAIN_DIR.

Allow the user to override CLANG_CROSS_FLAGS to avoid the GCC
call - in our case we set the var directly in the ebuild recipe.

In theory Clang could be able to autodetect these settings so
this logic could be removed entirely, but in practice as the
commit cebdb737

 ("tools: Help cross-building with clang")
mentions, this does not always work, so giving distributions
more control to specify their flags & sysroot is beneficial.

Suggested-by: Manoj Gupta <manojgupta@chromium.com>
Suggested-by: Nathan Chancellor <nathan@kernel.org>
Signed-off-by: Adrian Ratiu <adrian.ratiu@collabora.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Nathan Chancellor <nathan@kernel.org>
Cc: Nick Desaulniers <ndesaulniers@google.com>
Link: https://lore.kernel.org/lkml/87czjk4osi.fsf@ryzen9.i-did-not-set--mail-host-address--so-tickle-me
Link: https://lore.kernel.org/bpf/20220308121428.81735-1-adrian.ratiu@collabora.com

In test_lwt_ip_encap, the ingress IPv6 encap test failed from time to
time. The failure occured when an IPv4 ping through the IPv6 GRE
encapsulation did not receive a reply within the timeout. The IPv4 ping
and the IPv6 ping in the test used different timeouts (1 sec for IPv4
and 6 sec for IPv6), probably taking into account that IPv6 might need
longer to successfully complete. However, when IPv4 pings (with the
short timeout) are encapsulated into the IPv6 tunnel, the delays of IPv6
apply.

The actual reason for the long delays with IPv6 was that the IPv6
neighbor discovery sometimes did not complete in time. This was caused
by the outgoing interface only having a tentative link local address,
i.e., not having completed DAD for that lladdr. The ND was successfully
retried after 1 sec but that was too late for the ping timeout.

The IPv6 addresses for the test were already added with nodad. However,
for the lladdrs, DAD was still performed. We now disable DAD in the test
netns completely and just assume that the two lladdrs on each veth pair
do not collide. This removes all the delays for IPv6 traffic in the
test.

Without the delays, we can now also reduce the delay of the IPv6 ping to
1 sec. This makes the whole test complete faster because we don't need
to wait for the excessive timeout for each IPv6 ping that is supposed
to fail.

Fixes: 0fde56e4

 ("selftests: bpf: add test_lwt_ip_encap selftest")
Signed-off-by: Felix Maurer <fmaurer@redhat.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/4987d549d48b4e316cd5b3936de69c8d4bc75a4f.1646305899.git.fmaurer@redhat.com

Instead of determining buf_info string in the caller of check_buffer_access(),
we can determine whether the register type is read-only through
type_is_rdonly_mem() helper inside check_buffer_access() and construct
buf_info, making the code slightly cleaner.

Signed-off-by: Shung-Hsi Yu <shung-hsi.yu@suse.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Yonghong Song <yhs@fb.com>
Link: https://lore.kernel.org/bpf/YiWYLnAkEZXBP/gH@syu-laptop

vmtest.sh also supports s390x now.

Signed-off-by: KP Singh <kpsingh@kernel.org>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20220307133048.1287644-2-kpsingh@kernel.org

Dynamic linking when compiling on the host can cause issues when the
libc version does not match the one in the VM image. Update the
docs to explain how to do this.

Before:
  ./vmtest.sh -- ./test_progs -t test_ima
  ./test_progs: /usr/lib/libc.so.6: version `GLIBC_2.33' not found (required by ./test_progs)

After:

  LDLIBS=-static ./vmtest.sh -- ./test_progs -t test_ima
  test_ima:OK
  Summary: 1/0 PASSED, 0 SKIPPED, 0 FAILED

Reported-by: "Geyslan G. Bem" <geyslan@gmail.com>
Signed-off-by: KP Singh <kpsingh@kernel.org>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20220307133048.1287644-1-kpsingh@kernel.org

The trailing slash of LIBBPF_SRCS is redundant, remove it. Also inline
it as its only used in LIBBPF_INCLUDE.

Signed-off-by: Yuntao Wang <ytcoode@gmail.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20220305161013.361646-1-ytcoode@gmail.com

Fix the following coccicheck warning:
tools/lib/bpf/bpf.c:114:31-32: WARNING: Use ARRAY_SIZE
tools/lib/bpf/xsk.c:484:34-35: WARNING: Use ARRAY_SIZE
tools/lib/bpf/xsk.c:485:35-36: WARNING: Use ARRAY_SIZE

It has been tested with gcc (Debian 8.3.0-6) 8.3.0 on x86_64.

Signed-off-by: Guo Zhengkui <guozhengkui@vivo.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20220306023426.19324-1-guozhengkui@vivo.com

Using strncpy() on NUL-terminated strings is considered deprecated[1].
Moreover, if the length of 'task->comm' is less than the destination buffer
size, strncpy() will NUL-pad the destination buffer, which is a needless
performance penalty.

Replacing strncpy() with strscpy() fixes all these issues.

[1] https://www.kernel.org/doc/html/latest/process/deprecated.html#strncpy-on-nul-terminated-strings



Signed-off-by: Yuntao Wang <ytcoode@gmail.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Acked-by: Yonghong Song <yhs@fb.com>
Link: https://lore.kernel.org/bpf/20220304070408.233658-1-ytcoode@gmail.com

xsk_umem__create() does mmap for fill/comp rings, but xsk_umem__delete()
doesn't do the unmap. This works fine for regular cases, because
xsk_socket__delete() does unmap for the rings. But for the case that
xsk_socket__create_shared() fails, umem rings are not unmapped.

fill_save/comp_save are checked to determine if rings have already be
unmapped by xsk. If fill_save and comp_save are NULL, it means that the
rings have already been used by xsk. Then they are supposed to be
unmapped by xsk_socket__delete(). Otherwise, xsk_umem__delete() does the
unmap.

Fixes: 2f6324a3

 ("libbpf: Support shared umems between queues and devices")
Signed-off-by: Cheng Li <lic121@chinatelecom.cn>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20220301132623.GA19995@vscode.7~

Hao Luo says:

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

This patchset is very much similar to Yonghong's patchset on adding
__user tagging [1], where a "user" btf_type_tag was introduced to
describe __user memory pointers. Similar approach can be applied on
__percpu pointers. The __percpu attribute in kernel is used to identify
pointers that point to memory allocated in percpu region. Normally,
accessing __percpu memory requires using special functions like
per_cpu_ptr() etc. Directly accessing __percpu pointer is meaningless.

Currently vmlinux BTF does not have a way to differentiate a __percpu
pointer from a regular pointer. So BPF programs are allowed to load
__percpu memory directly, which is an incorrect behavior.

With the previous work that encodes __user information in BTF, a nice
framework has been set up to allow us to encode __percpu information in
BTF and let the verifier to reject programs that try to directly access
percpu pointer. Previously, there is a PTR_TO_PERCPU_BTF_ID reg type which
is used to represent those percpu static variables in the kernel. Pahole
is able to collect variables that are stored in ".data..percpu" section
in the kernel image and emit BTF information for those variables. The
bpf_per_cpu_ptr() and bpf_this_cpu_ptr() helper functions were added to
access these variables. Now with __percpu information, we can tag those
__percpu fields in a struct (such as cgroup->rstat_cpu) and allow the
pair of bpf percpu helpers to access them as well.

In addition to adding __percpu tagging, this patchset also fixes a
harmless bug in the previous patch that introduced __user. Patch 01/04
is for that. Patch 02/04 adds the new attribute "percpu". Patch 03/04
adds MEM_PERCPU tag for PTR_TO_BTF_ID and replaces PTR_TO_PERCPU_BTF_ID
with (BTF_ID | MEM_PERCPU). Patch 04/04 refactors the btf_tag test a bit
and adds tests for percpu tag.

Like [1], the minimal requirements for btf_type_tag is
clang (>= clang14) and pahole (>= 1.23).

[1] https://lore.kernel.org/bpf/20211220015110.3rqxk5qwub3pa2gh@ast-mbp.dhcp.thefacebook.com/t/


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

Signed-off-by: Alexei Starovoitov <ast@kernel.org>

Add test for percpu btf_type_tag. Similar to the "user" tag, we test
the following cases:

 1. __percpu struct field.
 2. __percpu as function parameter.
 3. per_cpu_ptr() accepts dynamically allocated __percpu memory.

Because the test for "user" and the test for "percpu" are very similar,
a little bit of refactoring has been done in btf_tag.c. Basically, both
tests share the same function for loading vmlinux and module btf.

Example output from log:

 > ./test_progs -v -t btf_tag

 libbpf: prog 'test_percpu1': BPF program load failed: Permission denied
 libbpf: prog 'test_percpu1': -- BEGIN PROG LOAD LOG --
 ...
 ; g = arg->a;
 1: (61) r1 = *(u32 *)(r1 +0)
 R1 is ptr_bpf_testmod_btf_type_tag_1 access percpu memory: off=0
 ...
 test_btf_type_tag_mod_percpu:PASS:btf_type_tag_percpu 0 nsec
 #26/6 btf_tag/btf_type_tag_percpu_mod1:OK

 libbpf: prog 'test_percpu2': BPF program load failed: Permission denied
 libbpf: prog 'test_percpu2': -- BEGIN PROG LOAD LOG --
 ...
 ; g = arg->p->a;
 2: (61) r1 = *(u32 *)(r1 +0)
 R1 is ptr_bpf_testmod_btf_type_tag_1 access percpu memory: off=0
 ...
 test_btf_type_tag_mod_percpu:PASS:btf_type_tag_percpu 0 nsec
 #26/7 btf_tag/btf_type_tag_percpu_mod2:OK

 libbpf: prog 'test_percpu_load': BPF program load failed: Permission denied
 libbpf: prog 'test_percpu_load': -- BEGIN PROG LOAD LOG --
 ...
 ; g = (__u64)cgrp->rstat_cpu->updated_children;
 2: (79) r1 = *(u64 *)(r1 +48)
 R1 is ptr_cgroup_rstat_cpu access percpu memory: off=48
 ...
 test_btf_type_tag_vmlinux_percpu:PASS:btf_type_tag_percpu_load 0 nsec
 #26/8 btf_tag/btf_type_tag_percpu_vmlinux_load:OK

 load_btfs:PASS:could not load vmlinux BTF 0 nsec
 test_btf_type_tag_vmlinux_percpu:PASS:btf_type_tag_percpu 0 nsec
 test_btf_type_tag_vmlinux_percpu:PASS:btf_type_tag_percpu_helper 0 nsec
 #26/9 btf_tag/btf_type_tag_percpu_vmlinux_helper:OK

Signed-off-by: Hao Luo <haoluo@google.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Yonghong Song <yhs@fb.com>
Link: https://lore.kernel.org/bpf/20220304191657.981240-5-haoluo@google.com

With the introduction of the btf_type_tag "percpu", we can add a
MEM_PERCPU to identify those pointers that point to percpu memory.
The ability of differetiating percpu pointers from regular memory
pointers have two benefits:

 1. It forbids unexpected use of percpu pointers, such as direct loads.
    In kernel, there are special functions used for accessing percpu
    memory. Directly loading percpu memory is meaningless. We already
    have BPF helpers like bpf_per_cpu_ptr() and bpf_this_cpu_ptr() that
    wrap the kernel percpu functions. So we can now convert percpu
    pointers into regular pointers in a safe way.

 2. Previously, bpf_per_cpu_ptr() and bpf_this_cpu_ptr() only work on
    PTR_TO_PERCPU_BTF_ID, a special reg_type which describes static
    percpu variables in kernel (we rely on pahole to encode them into
    vmlinux BTF). Now, since we can identify __percpu tagged pointers,
    we can also identify dynamically allocated percpu memory as well.
    It means we can use bpf_xxx_cpu_ptr() on dynamic percpu memory.
    This would be very convenient when accessing fields like
    "cgroup->rstat_cpu".

Signed-off-by: Hao Luo <haoluo@google.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Yonghong Song <yhs@fb.com>
Link: https://lore.kernel.org/bpf/20220304191657.981240-4-haoluo@google.com

This is similar to commit 7472d5a6 ("compiler_types: define __user as
__attribute__((btf_type_tag("user")))"), where a type tag "user" was
introduced to identify the pointers that point to user memory. With that
change, the newest compile toolchain can encode __user information into
vmlinux BTF, which can be used by the BPF verifier to enforce safe
program behaviors.

Similarly, we have __percpu attribute, which is mainly used to indicate
memory is allocated in percpu region. The __percpu pointers in kernel
are supposed to be used together with functions like per_cpu_ptr() and
this_cpu_ptr(), which perform necessary calculation on the pointer's
base address. Without the btf_type_tag introduced in this patch,
__percpu pointers will be treated as regular memory pointers in vmlinux
BTF and BPF programs are allowed to directly dereference them, generating
incorrect behaviors. Now with "percpu" btf_type_tag, the BPF verifier is
able to differentiate __percpu pointers from regular pointers and forbids
unexpected behaviors like direct load.

The following is an example similar to the one given in commit
7472d5a6:

  [$ ~] cat test.c
  #define __percpu __attribute__((btf_type_tag("percpu")))
  int foo(int __percpu *arg) {
  	return *arg;
  }
  [$ ~] clang -O2 -g -c test.c
  [$ ~] pahole -JV test.o
  ...
  File test.o:
  [1] INT int size=4 nr_bits=32 encoding=SIGNED
  [2] TYPE_TAG percpu type_id=1
  [3] PTR (anon) type_id=2
  [4] FUNC_PROTO (anon) return=1 args=(3 arg)
  [5] FUNC foo type_id=4
  [$ ~]

for the function argument "int __percpu *arg", its type is described as
	PTR -> TYPE_TAG(percpu) -> INT
The kernel can use this information for bpf verification or other
use cases.

Like commit 7472d5a6

, this feature requires clang (>= clang14) and
pahole (>= 1.23).

Signed-off-by: Hao Luo <haoluo@google.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Yonghong Song <yhs@fb.com>
Link: https://lore.kernel.org/bpf/20220304191657.981240-3-haoluo@google.com

With the introduction of MEM_USER in

 commit c6f1bfe8 ("bpf: reject program if a __user tagged memory accessed in kernel way")

PTR_TO_BTF_ID can be combined with a MEM_USER tag. Therefore, most
likely, when we compare reg_type against PTR_TO_BTF_ID, we want to use
the reg's base_type. Previously the check in check_mem_access() wants
to say: if the reg is BTF_ID but not NULL, the execution flow falls
into the 'then' branch. But now a reg of (BTF_ID | MEM_USER), which
should go into the 'then' branch, goes into the 'else'.

The end results before and after this patch are the same: regs tagged
with MEM_USER get rejected, but not in a way we intended. So fix the
condition, the error message now is correct.

Before (log from commit 696c3901):

  $ ./test_progs -v -n 22/3
  ...
  libbpf: prog 'test_user1': BPF program load failed: Permission denied
  libbpf: prog 'test_user1': -- BEGIN PROG LOAD LOG --
  R1 type=ctx expected=fp
  0: R1=ctx(id=0,off=0,imm=0) R10=fp0
  ; int BPF_PROG(test_user1, struct bpf_testmod_btf_type_tag_1 *arg)
  0: (79) r1 = *(u64 *)(r1 +0)
  func 'bpf_testmod_test_btf_type_tag_user_1' arg0 has btf_id 136561 type STRUCT 'bpf_testmod_btf_type_tag_1'
  1: R1_w=user_ptr_bpf_testmod_btf_type_tag_1(id=0,off=0,imm=0)
  ; g = arg->a;
  1: (61) r1 = *(u32 *)(r1 +0)
  R1 invalid mem access 'user_ptr_'

Now:

  libbpf: prog 'test_user1': BPF program load failed: Permission denied
  libbpf: prog 'test_user1': -- BEGIN PROG LOAD LOG --
  R1 type=ctx expected=fp
  0: R1=ctx(id=0,off=0,imm=0) R10=fp0
  ; int BPF_PROG(test_user1, struct bpf_testmod_btf_type_tag_1 *arg)
  0: (79) r1 = *(u64 *)(r1 +0)
  func 'bpf_testmod_test_btf_type_tag_user_1' arg0 has btf_id 104036 type STRUCT 'bpf_testmod_btf_type_tag_1'
  1: R1_w=user_ptr_bpf_testmod_btf_type_tag_1(id=0,ref_obj_id=0,off=0,imm=0)
  ; g = arg->a;
  1: (61) r1 = *(u32 *)(r1 +0)
  R1 is ptr_bpf_testmod_btf_type_tag_1 access user memory: off=0

Note the error message for the reason of rejection.

Fixes: c6f1bfe8

 ("bpf: reject program if a __user tagged memory accessed in kernel way")
Signed-off-by: Hao Luo <haoluo@google.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Yonghong Song <yhs@fb.com>
Link: https://lore.kernel.org/bpf/20220304191657.981240-2-haoluo@google.com

Kumar Kartikeya Dwivedi says:

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

This set fixes a bug related to bad var_off being permitted for kfunc call in
case of PTR_TO_BTF_ID, consolidates offset checks for all register types allowed
as helper or kfunc arguments into a common shared helper, and introduces a
couple of other checks to harden the kfunc release logic and prevent future
bugs. Some selftests are also included that fail in absence of these fixes,
serving as demonstration of the issues being fixed.

Changelog:
----------
v3 -> v4:
v3: https://lore.kernel.org/bpf/20220304000508.2904128-1-memxor@gmail.com

 * Update commit message for __diag patch to say clang instead of LLVM (Nathan)
 * Address nits for check_func_arg_reg_off (Martin)
 * Add comment for fixed_off_ok case, remove is_kfunc check (Martin)

v2 -> v3:
v2: https://lore.kernel.org/bpf/20220303045029.2645297-1-memxor@gmail.com

 * Add my SoB to __diag for clang patch (Nathan)

v1 -> v2:
v1: https://lore.kernel.org/bpf/20220301065745.1634848-1-memxor@gmail.com

 * Put reg->off check for release kfunc inside check_func_arg_reg_off,
   make the check a bit more readable
 * Squash verifier selftests errstr update into patch 3 for bisect (Alexei)
 * Include fix from Nathan for clang warning about missing prototypes
 * Add unified __diag_ingore_all that works for both GCC/LLVM (Alexei)

Older discussion:
Link: https://lore.kernel.org/bpf/20220219113744.1852259-1-memxor@gmail.com



Kumar Kartikeya Dwivedi (7):
  bpf: Add check_func_arg_reg_off function
  bpf: Fix PTR_TO_BTF_ID var_off check
  bpf: Disallow negative offset in check_ptr_off_reg
  bpf: Harden register offset checks for release helpers and kfuncs
  compiler_types.h: Add unified __diag_ignore_all for GCC/LLVM
  bpf: Replace __diag_ignore with unified __diag_ignore_all
  selftests/bpf: Add tests for kfunc register offset checks
====================

Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>

Include a few verifier selftests that test against the problems being
fixed by previous commits, i.e. release kfunc always require
PTR_TO_BTF_ID fixed and var_off to be 0, and negative offset is not
permitted and returns a helpful error message.

Signed-off-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20220304224645.3677453-9-memxor@gmail.com

Currently, -Wmissing-prototypes warning is ignored for GCC, but not
clang. This leads to clang build warning in W=1 mode. Since the flag
used by both compilers is same, we can use the unified __diag_ignore_all
macro that works for all supported versions and compilers which have
__diag macro support (currently GCC >= 8.0, and Clang >= 11.0).

Also add nf_conntrack_bpf.h include to prevent missing prototype warning
for register_nf_conntrack_bpf.

Signed-off-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20220304224645.3677453-8-memxor@gmail.com

Add a __diag_ignore_all macro, to ignore warnings for both GCC and LLVM,
without having to specify the compiler type and version. By default, GCC
8 and clang 11 are used. This will be used by bpf subsystem to ignore
-Wmissing-prototypes warning for functions that are meant to be global
functions so that they are in vmlinux BTF, but don't have a prototype.

Signed-off-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20220304224645.3677453-7-memxor@gmail.com

Add __diag macros similar to those in compiler-gcc.h, so that warnings
that need to be adjusted for specific cases but not globally can be
ignored when building with clang.

Signed-off-by: Nathan Chancellor <nathan@kernel.org>
Signed-off-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20220304224645.3677453-6-memxor@gmail.com

[ Kartikeya: wrote commit message ]

Let's ensure that the PTR_TO_BTF_ID reg being passed in to release BPF
helpers and kfuncs always has its offset set to 0. While not a real
problem now, there's a very real possibility this will become a problem
when more and more kfuncs are exposed, and more BPF helpers are added
which can release PTR_TO_BTF_ID.

Previous commits already protected against non-zero var_off. One of the
case we are concerned about now is when we have a type that can be
returned by e.g. an acquire kfunc:

struct foo {
	int a;
	int b;
	struct bar b;
};

... and struct bar is also a type that can be returned by another
acquire kfunc.

Then, doing the following sequence:

	struct foo *f = bpf_get_foo(); // acquire kfunc
	if (!f)
		return 0;
	bpf_put_bar(&f->b); // release kfunc

... would work with the current code, since the btf_struct_ids_match
takes reg->off into account for matching pointer type with release kfunc
argument type, but would obviously be incorrect, and most likely lead to
a kernel crash. A test has been included later to prevent regressions in
this area.

Signed-off-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20220304224645.3677453-5-memxor@gmail.com

check_ptr_off_reg only allows fixed offset to be set for PTR_TO_BTF_ID,
where reg->off < 0 doesn't make sense. This would shift the pointer
backwards, and fails later in btf_struct_ids_match or btf_struct_walk
due to out of bounds access (since offset is interpreted as unsigned).

Improve the verifier by rejecting this case by using a better error
message for BPF helpers and kfunc, by putting a check inside the
check_func_arg_reg_off function.

Also, update existing verifier selftests to work with new error string.

Signed-off-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20220304224645.3677453-4-memxor@gmail.com

When kfunc support was added, check_ctx_reg was called for PTR_TO_CTX
register, but no offset checks were made for PTR_TO_BTF_ID. Only
reg->off was taken into account by btf_struct_ids_match, which protected
against type mismatch due to non-zero reg->off, but when reg->off was
zero, a user could set the variable offset of the register and allow it
to be passed to kfunc, leading to bad pointer being passed into the
kernel.

Fix this by reusing the extracted helper check_func_arg_reg_off from
previous commit, and make one call before checking all supported
register types. Since the list is maintained, any future changes will be
taken into account by updating check_func_arg_reg_off. This function
prevents non-zero var_off to be set for PTR_TO_BTF_ID, but still allows
a fixed non-zero reg->off, which is needed for type matching to work
correctly when using pointer arithmetic.

ARG_DONTCARE is passed as arg_type, since kfunc doesn't support
accepting a ARG_PTR_TO_ALLOC_MEM without relying on size of parameter
type from BTF (in case of pointer), or using a mem, len pair. The
forcing of offset check for ARG_PTR_TO_ALLOC_MEM is done because ringbuf
helpers obtain the size from the header located at the beginning of the
memory region, hence any changes to the original pointer shouldn't be
allowed. In case of kfunc, size is always known, either at verification
time, or using the length parameter, hence this forcing is not required.

Since this check will happen once already for PTR_TO_CTX, remove the
check_ptr_off_reg call inside its block.

Fixes: e6ac2450

 ("bpf: Support bpf program calling kernel function")
Signed-off-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20220304224645.3677453-3-memxor@gmail.com

Lift the list of register types allowed for having fixed and variable
offsets when passed as helper function arguments into a common helper,
so that they can be reused for kfunc checks in later commits. Keeping a
common helper aids maintainability and allows us to follow the same
consistent rules across helpers and kfuncs. Also, convert check_func_arg
to use this function.

Signed-off-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20220304224645.3677453-2-memxor@gmail.com

Andrii Nakryiko says:

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

Add ability for user applications and libraries to register custom BPF program
SEC() handlers. See patch #2 for examples where this is useful.

Patch #1 does some preliminary refactoring to allow exponsing program
init, preload, and attach callbacks as public API. It also establishes
a protocol to allow optional auto-attach behavior. This will also help the
case of sometimes auto-attachable uprobes.

v4->v5:
  - API documentation improvements (Daniel);
v3->v4:
  - init_fn -> prog_setup_fn, preload_fn -> prog_prepare_load_fn (Alexei);
v2->v3:
  - moved callbacks and cookie into OPTS struct (Alan);
  - added more test scenarios (Alan);
  - address most of Alan's feedback, but kept API name;
v1->v2:
  - resubmitting due to git send-email screw up.

Cc: Alan Maguire <alan.maguire@oracle.com>
====================

Signed-off-by: Alexei Starovoitov <ast@kernel.org>

Add a selftest validating various aspects of libbpf's handling of custom
SEC() handlers. It also demonstrates how libraries can ensure very early
callbacks registration and unregistration using
__attribute__((constructor))/__attribute__((destructor)) functions.

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Tested-by: Alan Maguire <alan.maguire@oracle.com>
Reviewed-by: Alan Maguire <alan.maguire@oracle.com>
Link: https://lore.kernel.org/bpf/20220305010129.1549719-4-andrii@kernel.org

Allow registering and unregistering custom handlers for BPF program.
This allows user applications and libraries to plug into libbpf's
declarative SEC() definition handling logic. This allows to offload
complex and intricate custom logic into external libraries, but still
provide a great user experience.

One such example is USDT handling library, which has a lot of code and
complexity which doesn't make sense to put into libbpf directly, but it
would be really great for users to be able to specify BPF programs with
something like SEC("usdt/<path-to-binary>:<usdt_provider>:<usdt_name>")
and have correct BPF program type set (BPF_PROGRAM_TYPE_KPROBE, as it is
uprobe) and even support BPF skeleton's auto-attach logic.

In some cases, it might be even good idea to override libbpf's default
handling, like for SEC("perf_event") programs. With custom library, it's
possible to extend logic to support specifying perf event specification
right there in SEC() definition without burdening libbpf with lots of
custom logic or extra library dependecies (e.g., libpfm4). With current
patch it's possible to override libbpf's SEC("perf_event") handling and
specify a completely custom ones.

Further, it's possible to specify a generic fallback handling for any
SEC() that doesn't match any other custom or standard libbpf handlers.
This allows to accommodate whatever legacy use cases there might be, if
necessary.

See doc comments for libbpf_register_prog_handler() and
libbpf_unregister_prog_handler() for detailed semantics.

This patch also bumps libbpf development version to v0.8 and adds new
APIs there.

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Tested-by: Alan Maguire <alan.maguire@oracle.com>
Reviewed-by: Alan Maguire <alan.maguire@oracle.com>
Link: https://lore.kernel.org/bpf/20220305010129.1549719-3-andrii@kernel.org

Allow some BPF program types to support auto-attach only in subste of
cases. Currently, if some BPF program type specifies attach callback, it
is assumed that during skeleton attach operation all such programs
either successfully attach or entire skeleton attachment fails. If some
program doesn't support auto-attachment from skeleton, such BPF program
types shouldn't have attach callback specified.

This is limiting for cases when, depending on how full the SEC("")
definition is, there could either be enough details to support
auto-attach or there might not be and user has to use some specific API
to provide more details at runtime.

One specific example of such desired behavior might be SEC("uprobe"). If
it's specified as just uprobe auto-attach isn't possible. But if it's
SEC("uprobe/<some_binary>:<some_func>") then there are enough details to
support auto-attach. Note that there is a somewhat subtle difference
between auto-attach behavior of BPF skeleton and using "generic"
bpf_program__attach(prog) (which uses the same attach handlers under the
cover). Skeleton allow some programs within bpf_object to not have
auto-attach implemented and doesn't treat that as an error. Instead such
BPF programs are just skipped during skeleton's (optional) attach step.
bpf_program__attach(), on the other hand, is called when user *expects*
auto-attach to work, so if specified program doesn't implement or
doesn't support auto-attach functionality, that will be treated as an
error.

Another improvement to the way libbpf is handling SEC()s would be to not
require providing dummy kernel function name for kprobe. Currently,
SEC("kprobe/whatever") is necessary even if actual kernel function is
determined by user at runtime and bpf_program__attach_kprobe() is used
to specify it. With changes in this patch, it's possible to support both
SEC("kprobe") and SEC("kprobe/<actual_kernel_function"), while only in
the latter case auto-attach will be performed. In the former one, such
kprobe will be skipped during skeleton attach operation.

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Tested-by: Alan Maguire <alan.maguire@oracle.com>
Reviewed-by: Alan Maguire <alan.maguire@oracle.com>
Link: https://lore.kernel.org/bpf/20220305010129.1549719-2-andrii@kernel.org