platform/kernel/linux-rpi.git
2 years agoselftests/bpf: Add CO-RE relos into linked_funcs selftests
Andrii Nakryiko [Fri, 8 Apr 2022 18:14:25 +0000 (11:14 -0700)]
selftests/bpf: Add CO-RE relos into linked_funcs selftests

Add CO-RE relocations into __weak subprogs for multi-file linked_funcs
selftest to make sure libbpf handles such combination well.

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20220408181425.2287230-4-andrii@kernel.org
2 years agolibbpf: Use weak hidden modifier for USDT BPF-side API functions
Andrii Nakryiko [Fri, 8 Apr 2022 18:14:24 +0000 (11:14 -0700)]
libbpf: Use weak hidden modifier for USDT BPF-side API functions

Use __weak __hidden for bpf_usdt_xxx() APIs instead of much more
confusing `static inline __noinline`. This was previously impossible due
to libbpf erroring out on CO-RE relocations pointing to eliminated weak
subprogs. Now that previous patch fixed this issue, switch back to
__weak __hidden as it's a more direct way of specifying the desired
behavior.

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20220408181425.2287230-3-andrii@kernel.org
2 years agolibbpf: Don't error out on CO-RE relos for overriden weak subprogs
Andrii Nakryiko [Fri, 8 Apr 2022 18:14:23 +0000 (11:14 -0700)]
libbpf: Don't error out on CO-RE relos for overriden weak subprogs

During BPF static linking, all the ELF relocations and .BTF.ext
information (including CO-RE relocations) are preserved for __weak
subprograms that were logically overriden by either previous weak
subprogram instance or by corresponding "strong" (non-weak) subprogram.
This is just how native user-space linkers work, nothing new.

But libbpf is over-zealous when processing CO-RE relocation to error out
when CO-RE relocation belonging to such eliminated weak subprogram is
encountered. Instead of erroring out on this expected situation, log
debug-level message and skip the relocation.

Fixes: db2b8b06423c ("libbpf: Support CO-RE relocations for multi-prog sections")
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20220408181425.2287230-2-andrii@kernel.org
2 years agosamples, bpf: Move routes monitor in xdp_router_ipv4 in a dedicated thread
Lorenzo Bianconi [Tue, 5 Apr 2022 14:15:14 +0000 (16:15 +0200)]
samples, bpf: Move routes monitor in xdp_router_ipv4 in a dedicated thread

In order to not miss any netlink message from the kernel, move routes
monitor to a dedicated thread.

Fixes: 85bf1f51691c ("samples: bpf: Convert xdp_router_ipv4 to XDP samples helper")
Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/e364b817c69ded73be24b677ab47a157f7c21b64.1649167911.git.lorenzo@kernel.org
2 years agolibbpf: Allow WEAK and GLOBAL bindings during BTF fixup
Andrii Nakryiko [Thu, 7 Apr 2022 23:04:46 +0000 (16:04 -0700)]
libbpf: Allow WEAK and GLOBAL bindings during BTF fixup

During BTF fix up for global variables, global variable can be global
weak and will have STB_WEAK binding in ELF. Support such global
variables in addition to non-weak ones.

This is not the problem when using BPF static linking, as BPF static
linker "fixes up" BTF during generation so that libbpf doesn't have to
do it anymore during bpf_object__open(), which led to this not being
noticed for a while, along with a pretty rare (currently) use of __weak
variables and maps.

Reported-by: Hengqi Chen <hengqi.chen@gmail.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20220407230446.3980075-2-andrii@kernel.org
2 years agolibbpf: Use strlcpy() in path resolution fallback logic
Andrii Nakryiko [Thu, 7 Apr 2022 23:04:45 +0000 (16:04 -0700)]
libbpf: Use strlcpy() in path resolution fallback logic

Coverity static analyzer complains that strcpy() can cause buffer
overflow. Use libbpf_strlcpy() instead to be 100% sure this doesn't
happen.

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20220407230446.3980075-1-andrii@kernel.org
2 years agoMerge branch 'Add USDT support for s390'
Andrii Nakryiko [Fri, 8 Apr 2022 03:59:12 +0000 (20:59 -0700)]
Merge branch 'Add USDT support for s390'

Ilya Leoshkevich says:

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

This series adds USDT support for s390, making the "usdt" test pass
there. Patch 1 is a collection of minor cleanups, patch 2 adds
BPF-side support, patch 3 adds userspace-side support.
====================

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
2 years agolibbpf: Add s390-specific USDT arg spec parsing logic
Ilya Leoshkevich [Thu, 7 Apr 2022 21:44:11 +0000 (23:44 +0200)]
libbpf: Add s390-specific USDT arg spec parsing logic

The logic is superficially similar to that of x86, but the small
differences (no need for register table and dynamic allocation of
register names, no $ sign before constants) make maintaining a common
implementation too burdensome. Therefore simply add a s390x-specific
version of parse_usdt_arg().

Note that while bcc supports index registers, this patch does not. This
should not be a problem in most cases, since s390 uses a default value
"nor" for STAP_SDT_ARG_CONSTRAINT.

Signed-off-by: Ilya Leoshkevich <iii@linux.ibm.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20220407214411.257260-4-iii@linux.ibm.com
2 years agolibbpf: Make BPF-side of USDT support work on big-endian machines
Ilya Leoshkevich [Thu, 7 Apr 2022 21:44:10 +0000 (23:44 +0200)]
libbpf: Make BPF-side of USDT support work on big-endian machines

BPF_USDT_ARG_REG_DEREF handling always reads 8 bytes, regardless of
the actual argument size. On little-endian the relevant argument bits
end up in the lower bits of val, and later on the code that handles
all the argument types expects them to be there.

On big-endian they end up in the upper bits of val, breaking that
expectation. Fix by right-shifting val on big-endian.

Signed-off-by: Ilya Leoshkevich <iii@linux.ibm.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20220407214411.257260-3-iii@linux.ibm.com
2 years agolibbpf: Minor style improvements in USDT code
Ilya Leoshkevich [Thu, 7 Apr 2022 21:44:09 +0000 (23:44 +0200)]
libbpf: Minor style improvements in USDT code

Fix several typos and references to non-existing headers.
Also use __BYTE_ORDER__ instead of __BYTE_ORDER for consistency with
the rest of the bpf code - see commit 45f2bebc8079 ("libbpf: Fix
endianness detection in BPF_CORE_READ_BITFIELD_PROBED()") for
rationale).

Signed-off-by: Ilya Leoshkevich <iii@linux.ibm.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20220407214411.257260-2-iii@linux.ibm.com
2 years agolibbpf: Fix use #ifdef instead of #if to avoid compiler warning
Andrii Nakryiko [Thu, 7 Apr 2022 20:38:42 +0000 (13:38 -0700)]
libbpf: Fix use #ifdef instead of #if to avoid compiler warning

As reported by Naresh:

  perf build errors on i386 [1] on Linux next-20220407 [2]

  usdt.c:1181:5: error: "__x86_64__" is not defined, evaluates to 0
  [-Werror=undef]
   1181 | #if __x86_64__
        |     ^~~~~~~~~~
  usdt.c:1196:5: error: "__x86_64__" is not defined, evaluates to 0
  [-Werror=undef]
   1196 | #if __x86_64__
        |     ^~~~~~~~~~
  cc1: all warnings being treated as errors

Use #ifdef instead of #if to avoid this.

Fixes: 4c59e584d158 ("libbpf: Add x86-specific USDT arg spec parsing logic")
Reported-by: Naresh Kamboju <naresh.kamboju@linaro.org>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20220407203842.3019904-1-andrii@kernel.org
2 years agolibbpf: Potential NULL dereference in usdt_manager_attach_usdt()
Haowen Bai [Thu, 7 Apr 2022 02:38:17 +0000 (10:38 +0800)]
libbpf: Potential NULL dereference in usdt_manager_attach_usdt()

link could be null but still dereference bpf_link__destroy(&link->link)
and it will lead to a null pointer access.

Signed-off-by: Haowen Bai <baihaowen@meizu.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/1649299098-2069-1-git-send-email-baihaowen@meizu.com
2 years agoMerge branch 'libbpf: uprobe name-based attach followups'
Andrii Nakryiko [Thu, 7 Apr 2022 18:42:51 +0000 (11:42 -0700)]
Merge branch 'libbpf: uprobe name-based attach followups'

Alan Maguire says:

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

Follow-up series to [1] to address some suggestions from Andrii to
improve parsing and make it more robust (patches 1, 2) and to improve
validation of u[ret]probe firing by validating expected argument
and return values (patch 3).

[1] https://lore.kernel.org/bpf/164903521182.13106.12656654142629368774.git-patchwork-notify@kernel.org/

Changes since v1:
- split library name, auto-attach parsing into separate patches (Andrii, patches 1, 2)
- made str_has_sfx() static inline, avoided repeated strlen()s by storing lengths,
  used strlen() instead of strnlen() (Andrii, patch 1)
- fixed sscanf() arg to use %li, switched logging to use "prog '%s'" format,
  used direct strcmp() on probe_type instead of prefix check (Andrii, patch 2)
- switched auto-attach tests to log parameter/return values to be checked by
  user-space side of tests. Needed to add pid filtering to avoid capturing
  stray malloc()s (Andrii, patch 3)
====================

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
2 years agoselftests/bpf: Uprobe tests should verify param/return values
Alan Maguire [Wed, 6 Apr 2022 11:43:51 +0000 (12:43 +0100)]
selftests/bpf: Uprobe tests should verify param/return values

uprobe/uretprobe tests don't do any validation of arguments/return values,
and without this we can't be sure we are attached to the right function,
or that we are indeed attached to a uprobe or uretprobe.  To fix this
record argument and return value for auto-attached functions and ensure
these match expectations.  Also need to filter by pid to ensure we do
not pick up stray malloc()s since auto-attach traces libc system-wide.

Suggested-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alan Maguire <alan.maguire@oracle.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/1649245431-29956-4-git-send-email-alan.maguire@oracle.com
2 years agolibbpf: Improve string parsing for uprobe auto-attach
Alan Maguire [Wed, 6 Apr 2022 11:43:50 +0000 (12:43 +0100)]
libbpf: Improve string parsing for uprobe auto-attach

For uprobe auto-attach, the parsing can be simplified for the SEC()
name to a single sscanf(); the return value of the sscanf can then
be used to distinguish between sections that simply specify
"u[ret]probe" (and thus cannot auto-attach), those that specify
"u[ret]probe/binary_path:function+offset" etc.

Suggested-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alan Maguire <alan.maguire@oracle.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/1649245431-29956-3-git-send-email-alan.maguire@oracle.com
2 years agolibbpf: Improve library identification for uprobe binary path resolution
Alan Maguire [Wed, 6 Apr 2022 11:43:49 +0000 (12:43 +0100)]
libbpf: Improve library identification for uprobe binary path resolution

In the process of doing path resolution for uprobe attach, libraries are
identified by matching a ".so" substring in the binary_path.
This matches a lot of patterns that do not conform to library.so[.version]
format, so instead match a ".so" _suffix_, and if that fails match a
".so." substring for the versioned library case.

Suggested-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alan Maguire <alan.maguire@oracle.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/1649245431-29956-2-git-send-email-alan.maguire@oracle.com
2 years agoselftests/bpf: Test for writes to map key from BPF helpers
Kumar Kartikeya Dwivedi [Sat, 19 Mar 2022 08:08:27 +0000 (13:38 +0530)]
selftests/bpf: Test for writes to map key from BPF helpers

When invoking bpf_for_each_map_elem callback, we are passed a
PTR_TO_MAP_KEY, previously writes to this through helper may be allowed,
but the fix in previous patches is meant to prevent that case. The test
case tries to pass it as writable memory to helper, and fails test if it
succeeds to pass the verifier.

Signed-off-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Link: https://lore.kernel.org/r/20220319080827.73251-6-memxor@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2 years agoselftests/bpf: Test passing rdonly mem to global func
Kumar Kartikeya Dwivedi [Sat, 19 Mar 2022 08:08:26 +0000 (13:38 +0530)]
selftests/bpf: Test passing rdonly mem to global func

Add two test cases, one pass read only map value pointer to global
func, which should be rejected. The same code checks it for kfunc, so
that is covered as well. Second one tries to use the missing check for
PTR_TO_MEM's MEM_RDONLY flag and tries to write to a read only memory
pointer. Without prior patches, both of these tests fail.

Reviewed-by: Hao Luo <haoluo@google.com>
Signed-off-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Link: https://lore.kernel.org/r/20220319080827.73251-5-memxor@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2 years agobpf: Reject writes for PTR_TO_MAP_KEY in check_helper_mem_access
Kumar Kartikeya Dwivedi [Sat, 19 Mar 2022 08:08:25 +0000 (13:38 +0530)]
bpf: Reject writes for PTR_TO_MAP_KEY in check_helper_mem_access

It is not permitted to write to PTR_TO_MAP_KEY, but the current code in
check_helper_mem_access would allow for it, reject this case as well, as
helpers taking ARG_PTR_TO_UNINIT_MEM also take PTR_TO_MAP_KEY.

Fixes: 69c087ba6225 ("bpf: Add bpf_for_each_map_elem() helper")
Signed-off-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Link: https://lore.kernel.org/r/20220319080827.73251-4-memxor@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2 years agobpf: Check PTR_TO_MEM | MEM_RDONLY in check_helper_mem_access
Kumar Kartikeya Dwivedi [Sat, 19 Mar 2022 08:08:24 +0000 (13:38 +0530)]
bpf: Check PTR_TO_MEM | MEM_RDONLY in check_helper_mem_access

The commit being fixed was aiming to disallow users from incorrectly
obtaining writable pointer to memory that is only meant to be read. This
is enforced now using a MEM_RDONLY flag.

For instance, in case of global percpu variables, when the BTF type is
not struct (e.g. bpf_prog_active), the verifier marks register type as
PTR_TO_MEM | MEM_RDONLY from bpf_this_cpu_ptr or bpf_per_cpu_ptr
helpers. However, when passing such pointer to kfunc, global funcs, or
BPF helpers, in check_helper_mem_access, there is no expectation
MEM_RDONLY flag will be set, hence it is checked as pointer to writable
memory. Later, verifier sets up argument type of global func as
PTR_TO_MEM | PTR_MAYBE_NULL, so user can use a global func to get around
the limitations imposed by this flag.

This check will also cover global non-percpu variables that may be
introduced in kernel BTF in future.

Also, we update the log message for PTR_TO_BUF case to be similar to
PTR_TO_MEM case, so that the reason for error is clear to user.

Fixes: 34d3a78c681e ("bpf: Make per_cpu_ptr return rdonly PTR_TO_MEM.")
Reviewed-by: Hao Luo <haoluo@google.com>
Signed-off-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Link: https://lore.kernel.org/r/20220319080827.73251-3-memxor@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2 years agobpf: Do write access check for kfunc and global func
Kumar Kartikeya Dwivedi [Sat, 19 Mar 2022 08:08:23 +0000 (13:38 +0530)]
bpf: Do write access check for kfunc and global func

When passing pointer to some map value to kfunc or global func, in
verifier we are passing meta as NULL to various functions, which uses
meta->raw_mode to check whether memory is being written to. Since some
kfunc or global funcs may also write to memory pointers they receive as
arguments, we must check for write access to memory. E.g. in some case
map may be read only and this will be missed by current checks.

However meta->raw_mode allows for uninitialized memory (e.g. on stack),
since there is not enough info available through BTF, we must perform
one call for read access (raw_mode = false), and one for write access
(raw_mode = true).

Fixes: e5069b9c23b3 ("bpf: Support pointers in global func args")
Fixes: d583691c47dc ("bpf: Introduce mem, size argument pair support for kfunc")
Signed-off-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Link: https://lore.kernel.org/r/20220319080827.73251-2-memxor@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2 years agoselftests/bpf: Use bpf_num_possible_cpus() in per-cpu map allocations
Artem Savkov [Wed, 6 Apr 2022 08:54:08 +0000 (10:54 +0200)]
selftests/bpf: Use bpf_num_possible_cpus() in per-cpu map allocations

bpf_map_value_size() uses num_possible_cpus() to determine map size, but
some of the tests only allocate enough memory for online cpus. This
results in out-of-bound writes in userspace during bpf(BPF_MAP_LOOKUP_ELEM)
syscalls in cases when number of online cpus is lower than the number of
possible cpus. Fix by switching from get_nprocs_conf() to
bpf_num_possible_cpus() when determining the number of processors in
these tests (test_progs/netcnt and test_cgroup_storage).

Signed-off-by: Artem Savkov <asavkov@redhat.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20220406085408.339336-1-asavkov@redhat.com
2 years agolibbpf: Fix spelling mistake "libaries" -> "libraries"
Colin Ian King [Wed, 6 Apr 2022 08:08:35 +0000 (09:08 +0100)]
libbpf: Fix spelling mistake "libaries" -> "libraries"

There is a spelling mistake in a pr_warn message. Fix it.

Signed-off-by: Colin Ian King <colin.i.king@gmail.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20220406080835.14879-1-colin.i.king@gmail.com
2 years agoselftests/bpf: Fix issues in parse_num_list()
Yuntao Wang [Wed, 6 Apr 2022 00:36:22 +0000 (08:36 +0800)]
selftests/bpf: Fix issues in parse_num_list()

The function does not check that parsing_end is false after parsing
argument. Thus, if the final part of the argument is something like '4-',
which is invalid, parse_num_list() will discard it instead of returning
-EINVAL.

Before:

 $ ./test_progs -n 2,4-
 #2 atomic_bounds:OK
 Summary: 1/0 PASSED, 0 SKIPPED, 0 FAILED

After:

 $ ./test_progs -n 2,4-
 Failed to parse test numbers.

Signed-off-by: Yuntao Wang <ytcoode@gmail.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20220406003622.73539-1-ytcoode@gmail.com
2 years agonet: netfilter: Reports ct direction in CT lookup helpers for XDP and TC-BPF
Lorenzo Bianconi [Sat, 2 Apr 2022 14:19:14 +0000 (16:19 +0200)]
net: netfilter: Reports ct direction in CT lookup helpers for XDP and TC-BPF

Report connection tracking tuple direction in
bpf_skb_ct_lookup/bpf_xdp_ct_lookup helpers. Direction will be used to
implement snat/dnat through xdp ebpf program.

Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/aa1aaac89191cfc64078ecef36c0a48c302321b9.1648908601.git.lorenzo@kernel.org
2 years agoselftests/bpf: Fix file descriptor leak in load_kallsyms()
Yuntao Wang [Tue, 5 Apr 2022 14:57:11 +0000 (22:57 +0800)]
selftests/bpf: Fix file descriptor leak in load_kallsyms()

Currently, if sym_cnt > 0, it just returns and does not close file, fix it.

Signed-off-by: Yuntao Wang <ytcoode@gmail.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20220405145711.49543-1-ytcoode@gmail.com
2 years agobpf, arm64: Sign return address for JITed code
Xu Kuohai [Sat, 2 Apr 2022 07:39:42 +0000 (03:39 -0400)]
bpf, arm64: Sign return address for JITed code

Sign return address for JITed code when the kernel is built with pointer
authentication enabled:

1. Sign LR with paciasp instruction before LR is pushed to stack. Since
   paciasp acts like landing pads for function entry, no need to insert
   bti instruction before paciasp.

2. Authenticate LR with autiasp instruction after LR is popped from stack.

For BPF tail call, the stack frame constructed by the caller is reused by
the callee. That is, the stack frame is constructed by the caller and
destructed by the callee. Thus LR is signed and pushed to the stack in the
caller's prologue, and poped from the stack and authenticated in the
callee's epilogue.

For BPF2BPF call, the caller and callee construct their own stack frames,
and sign and authenticate their own LRs.

Signed-off-by: Xu Kuohai <xukuohai@huawei.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://events.static.linuxfound.org/sites/events/files/slides/slides_23.pdf
Link: https://lore.kernel.org/bpf/20220402073942.3782529-1-xukuohai@huawei.com
2 years agoMerge branch 'Add libbpf support for USDTs'
Alexei Starovoitov [Tue, 5 Apr 2022 20:16:08 +0000 (13:16 -0700)]
Merge branch 'Add libbpf support for USDTs'

Andrii Nakryiko says:

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

Add libbpf support for USDT (User Statically-Defined Tracing) probes.
USDTs is important part of tracing, and BPF, ecosystem, widely used in
mission-critical production applications for observability, performance
analysis, and debugging.

And while USDTs themselves are pretty complicated abstraction built on top of
uprobes, for end-users USDT is as natural a primitive as uprobes themselves.
And thus it's important for libbpf to provide best possible user experience
when it comes to build tracing applications relying on USDTs.

USDTs historically presented a lot of challenges for libbpf's no
compilation-on-the-fly general approach to BPF tracing. BCC utilizes power of
on-the-fly source code generation and compilation using its embedded Clang
toolchain, which was impractical for more lightweight and thus more rigid
libbpf-based approach. But still, with enough diligence and BPF cookies it's
possible to implement USDT support that feels as natural as tracing any
uprobe.

This patch set is the culmination of such effort to add libbpf USDT support
following the spirit and philosophy of BPF CO-RE (even though it's not
inherently relying on BPF CO-RE much, see patch #1 for some notes regarding
this). Each respective patch has enough details and explanations, so I won't
go into details here.

In the end, I think the overall usability of libbpf's USDT support *exceeds*
the status quo set by BCC due to the elimination of awkward runtime USDT
supporting code generation. It also exceeds BCC's capabilities due to the use
of BPF cookie. This eliminates the need to determine a USDT call site (and
thus specifics about how exactly to fetch arguments) based on its *absolute IP
address*, which is impossible with shared libraries if no PID is specified (as
we then just *can't* know absolute IP at which shared library is loaded,
because it might be different for each process). With BPF cookie this is not
a problem as we record "call site ID" directly in a BPF cookie value. This
makes it possible to do a system-wide tracing of a USDT defined in a shared
library. Think about tracing some USDT in libc across any process in the
system, both running at the time of attachment and all the new processes
started *afterwards*. This is a very powerful capability that allows more
efficient observability and tracing tooling.

Once this functionality lands, the plan is to extend libbpf-bootstrap ([0])
with an USDT example. It will also become possible to start converting BCC
tools that rely on USDTs to their libbpf-based counterparts ([1]).

It's worth noting that preliminary version of this code was currently used and
tested in production code running fleet-wide observability toolkit.

Libbpf functionality is broken down into 5 mostly logically independent parts,
for ease of reviewing:
  - patch #1 adds BPF-side implementation;
  - patch #2 adds user-space APIs and wires bpf_link for USDTs;
  - patch #3 adds the most mundate pieces: handling ELF, parsing USDT notes,
    dealing with memory segments, relative vs absolute addresses, etc;
  - patch #4 adds internal ID allocation and setting up/tearing down of
    BPF-side state (spec and IP-to-ID mapping);
  - patch #5 implements x86/x86-64-specific logic of parsing USDT argument
    specifications;
  - patch #6 adds testing of various basic aspects of handling of USDT;
  - patch #7 extends the set of tests with more combinations of semaphore,
    executable vs shared library, and PID filter options.

  [0] https://github.com/libbpf/libbpf-bootstrap
  [1] https://github.com/iovisor/bcc/tree/master/libbpf-tools

v2->v3:
  - fix typos, leave link to systemtap doc, acks, etc (Dave);
  - include sys/sdt.h to avoid extra system-wide package dependencies;
v1->v2:
  - huge high-level comment describing how all the moving parts fit together
    (Alan, Alexei);
  - switched from `__hidden __weak` to `static inline __noinline` for now, as
    there is a bug in BPF linker breaking final BPF object file due to invalid
    .BTF.ext data; I want to fix it separately at which point I'll switch back
    to __hidden __weak again. The fix isn't trivial, so I don't want to block
    on that. Same for __weak variable lookup bug that Henqi reported.
  - various fixes and improvements, addressing other feedback (Alan, Hengqi);

Cc: Alan Maguire <alan.maguire@oracle.com>
Cc: Dave Marchevsky <davemarchevsky@fb.com>
Cc: Hengqi Chen <hengqi.chen@gmail.com>
====================

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2 years agoselftests/bpf: Add urandom_read shared lib and USDTs
Andrii Nakryiko [Mon, 4 Apr 2022 23:42:02 +0000 (16:42 -0700)]
selftests/bpf: Add urandom_read shared lib and USDTs

Extend urandom_read helper binary to include USDTs of 4 combinations:
semaphore/semaphoreless (refcounted and non-refcounted) and based in
executable or shared library. We also extend urandom_read with ability
to report it's own PID to parent process and wait for parent process to
ready itself up for tracing urandom_read. We utilize popen() and
underlying pipe properties for proper signaling.

Once urandom_read is ready, we add few tests to validate that libbpf's
USDT attachment handles all the above combinations of semaphore (or lack
of it) and static or shared library USDTs. Also, we validate that libbpf
handles shared libraries both with PID filter and without one (i.e., -1
for PID argument).

Having the shared library case tested with and without PID is important
because internal logic differs on kernels that don't support BPF
cookies. On such older kernels, attaching to USDTs in shared libraries
without specifying concrete PID doesn't work in principle, because it's
impossible to determine shared library's load address to derive absolute
IPs for uprobe attachments. Without absolute IPs, it's impossible to
perform correct look up of USDT spec based on uprobe's absolute IP (the
only kind available from BPF at runtime). This is not the problem on
newer kernels with BPF cookie as we don't need IP-to-ID lookup because
BPF cookie value *is* spec ID.

So having those two situations as separate subtests is good because
libbpf CI is able to test latest selftests against old kernels (e.g.,
4.9 and 5.5), so we'll be able to disable PID-less shared lib attachment
for old kernels, but will still leave PID-specific one enabled to validate
this legacy logic is working correctly.

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Dave Marchevsky <davemarchevsky@fb.com>
Link: https://lore.kernel.org/bpf/20220404234202.331384-8-andrii@kernel.org
2 years agoselftests/bpf: Add basic USDT selftests
Andrii Nakryiko [Mon, 4 Apr 2022 23:42:01 +0000 (16:42 -0700)]
selftests/bpf: Add basic USDT selftests

Add semaphore-based USDT to test_progs itself and write basic tests to
valicate both auto-attachment and manual attachment logic, as well as
BPF-side functionality.

Also add subtests to validate that libbpf properly deduplicates USDT
specs and handles spec overflow situations correctly, as well as proper
"rollback" of partially-attached multi-spec USDT.

BPF-side of selftest intentionally consists of two files to validate
that usdt.bpf.h header can be included from multiple source code files
that are subsequently linked into final BPF object file without causing
any symbol duplication or other issues. We are validating that __weak
maps and bpf_usdt_xxx() API functions defined in usdt.bpf.h do work as
intended.

USDT selftests utilize sys/sdt.h header that on Ubuntu systems comes
from systemtap-sdt-devel package. But to simplify everyone's life,
including CI but especially casual contributors to bpf/bpf-next that
are trying to build selftests, I've checked in sys/sdt.h header from [0]
directly. This way it will work on all architectures and distros without
having to figure it out for every relevant combination and adding any
extra implicit package dependencies.

  [0] https://sourceware.org/git?p=systemtap.git;a=blob_plain;f=includes/sys/sdt.h;h=ca0162b4dc57520b96638c8ae79ad547eb1dd3a1;hb=HEAD

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Reviewed-by: Alan Maguire <alan.maguire@oracle.com>
Acked-by: Dave Marchevsky <davemarchevsky@fb.com>
Link: https://lore.kernel.org/bpf/20220404234202.331384-7-andrii@kernel.org
2 years agolibbpf: Add x86-specific USDT arg spec parsing logic
Andrii Nakryiko [Mon, 4 Apr 2022 23:42:00 +0000 (16:42 -0700)]
libbpf: Add x86-specific USDT arg spec parsing logic

Add x86/x86_64-specific USDT argument specification parsing. Each
architecture will require their own logic, as all this is arch-specific
assembly-based notation. Architectures that libbpf doesn't support for
USDTs will pr_warn() with specific error and return -ENOTSUP.

We use sscanf() as a very powerful and easy to use string parser. Those
spaces in sscanf's format string mean "skip any whitespaces", which is
pretty nifty (and somewhat little known) feature.

All this was tested on little-endian architecture, so bit shifts are
probably off on big-endian, which our CI will hopefully prove.

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Reviewed-by: Alan Maguire <alan.maguire@oracle.com>
Reviewed-by: Dave Marchevsky <davemarchevsky@fb.com>
Link: https://lore.kernel.org/bpf/20220404234202.331384-6-andrii@kernel.org
2 years agolibbpf: Wire up spec management and other arch-independent USDT logic
Andrii Nakryiko [Mon, 4 Apr 2022 23:41:59 +0000 (16:41 -0700)]
libbpf: Wire up spec management and other arch-independent USDT logic

Last part of architecture-agnostic user-space USDT handling logic is to
set up BPF spec and, optionally, IP-to-ID maps from user-space.
usdt_manager performs a compact spec ID allocation to utilize
fixed-sized BPF maps as efficiently as possible. We also use hashmap to
deduplicate USDT arg spec strings and map identical strings to single
USDT spec, minimizing the necessary BPF map size. usdt_manager supports
arbitrary sequences of attachment and detachment, both of the same USDT
and multiple different USDTs and internally maintains a free list of
unused spec IDs. bpf_link_usdt's logic is extended with proper setup and
teardown of this spec ID free list and supporting BPF maps.

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Reviewed-by: Alan Maguire <alan.maguire@oracle.com>
Reviewed-by: Dave Marchevsky <davemarchevsky@fb.com>
Link: https://lore.kernel.org/bpf/20220404234202.331384-5-andrii@kernel.org
2 years agolibbpf: Add USDT notes parsing and resolution logic
Andrii Nakryiko [Mon, 4 Apr 2022 23:41:58 +0000 (16:41 -0700)]
libbpf: Add USDT notes parsing and resolution logic

Implement architecture-agnostic parts of USDT parsing logic. The code is
the documentation in this case, it's futile to try to succinctly
describe how USDT parsing is done in any sort of concreteness. But
still, USDTs are recorded in special ELF notes section (.note.stapsdt),
where each USDT call site is described separately. Along with USDT
provider and USDT name, each such note contains USDT argument
specification, which uses assembly-like syntax to describe how to fetch
value of USDT argument. USDT arg spec could be just a constant, or
a register, or a register dereference (most common cases in x86_64), but
it technically can be much more complicated cases, like offset relative
to global symbol and stuff like that. One of the later patches will
implement most common subset of this for x86 and x86-64 architectures,
which seems to handle a lot of real-world production application.

USDT arg spec contains a compact encoding allowing usdt.bpf.h from
previous patch to handle the above 3 cases. Instead of recording which
register might be needed, we encode register's offset within struct
pt_regs to simplify BPF-side implementation. USDT argument can be of
different byte sizes (1, 2, 4, and 8) and signed or unsigned. To handle
this, libbpf pre-calculates necessary bit shifts to do proper casting
and sign-extension in a short sequences of left and right shifts.

The rest is in the code with sometimes extensive comments and references
to external "documentation" for USDTs.

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Reviewed-by: Alan Maguire <alan.maguire@oracle.com>
Reviewed-by: Dave Marchevsky <davemarchevsky@fb.com>
Link: https://lore.kernel.org/bpf/20220404234202.331384-4-andrii@kernel.org
2 years agolibbpf: Wire up USDT API and bpf_link integration
Andrii Nakryiko [Mon, 4 Apr 2022 23:41:57 +0000 (16:41 -0700)]
libbpf: Wire up USDT API and bpf_link integration

Wire up libbpf USDT support APIs without yet implementing all the
nitty-gritty details of USDT discovery, spec parsing, and BPF map
initialization.

User-visible user-space API is simple and is conceptually very similar
to uprobe API.

bpf_program__attach_usdt() API allows to programmatically attach given
BPF program to a USDT, specified through binary path (executable or
shared lib), USDT provider and name. Also, just like in uprobe case, PID
filter is specified (0 - self, -1 - any process, or specific PID).
Optionally, USDT cookie value can be specified. Such single API
invocation will try to discover given USDT in specified binary and will
use (potentially many) BPF uprobes to attach this program in correct
locations.

Just like any bpf_program__attach_xxx() APIs, bpf_link is returned that
represents this attachment. It is a virtual BPF link that doesn't have
direct kernel object, as it can consist of multiple underlying BPF
uprobe links. As such, attachment is not atomic operation and there can
be brief moment when some USDT call sites are attached while others are
still in the process of attaching. This should be taken into
consideration by user. But bpf_program__attach_usdt() guarantees that
in the case of success all USDT call sites are successfully attached, or
all the successfuly attachments will be detached as soon as some USDT
call sites failed to be attached. So, in theory, there could be cases of
failed bpf_program__attach_usdt() call which did trigger few USDT
program invocations. This is unavoidable due to multi-uprobe nature of
USDT and has to be handled by user, if it's important to create an
illusion of atomicity.

USDT BPF programs themselves are marked in BPF source code as either
SEC("usdt"), in which case they won't be auto-attached through
skeleton's <skel>__attach() method, or it can have a full definition,
which follows the spirit of fully-specified uprobes:
SEC("usdt/<path>:<provider>:<name>"). In the latter case skeleton's
attach method will attempt auto-attachment. Similarly, generic
bpf_program__attach() will have enought information to go off of for
parameterless attachment.

USDT BPF programs are actually uprobes, and as such for kernel they are
marked as BPF_PROG_TYPE_KPROBE.

Another part of this patch is USDT-related feature probing:
  - BPF cookie support detection from user-space;
  - detection of kernel support for auto-refcounting of USDT semaphore.

The latter is optional. If kernel doesn't support such feature and USDT
doesn't rely on USDT semaphores, no error is returned. But if libbpf
detects that USDT requires setting semaphores and kernel doesn't support
this, libbpf errors out with explicit pr_warn() message. Libbpf doesn't
support poking process's memory directly to increment semaphore value,
like BCC does on legacy kernels, due to inherent raciness and danger of
such process memory manipulation. Libbpf let's kernel take care of this
properly or gives up.

Logistically, all the extra USDT-related infrastructure of libbpf is put
into a separate usdt.c file and abstracted behind struct usdt_manager.
Each bpf_object has lazily-initialized usdt_manager pointer, which is
only instantiated if USDT programs are attempted to be attached. Closing
BPF object frees up usdt_manager resources. usdt_manager keeps track of
USDT spec ID assignment and few other small things.

Subsequent patches will fill out remaining missing pieces of USDT
initialization and setup logic.

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Reviewed-by: Alan Maguire <alan.maguire@oracle.com>
Link: https://lore.kernel.org/bpf/20220404234202.331384-3-andrii@kernel.org
2 years agolibbpf: Add BPF-side of USDT support
Andrii Nakryiko [Mon, 4 Apr 2022 23:41:56 +0000 (16:41 -0700)]
libbpf: Add BPF-side of USDT support

Add BPF-side implementation of libbpf-provided USDT support. This
consists of single header library, usdt.bpf.h, which is meant to be used
from user's BPF-side source code. This header is added to the list of
installed libbpf header, along bpf_helpers.h and others.

BPF-side implementation consists of two BPF maps:
  - spec map, which contains "a USDT spec" which encodes information
    necessary to be able to fetch USDT arguments and other information
    (argument count, user-provided cookie value, etc) at runtime;
  - IP-to-spec-ID map, which is only used on kernels that don't support
    BPF cookie feature. It allows to lookup spec ID based on the place
    in user application that triggers USDT program.

These maps have default sizes, 256 and 1024, which are chosen
conservatively to not waste a lot of space, but handling a lot of common
cases. But there could be cases when user application needs to either
trace a lot of different USDTs, or USDTs are heavily inlined and their
arguments are located in a lot of differing locations. For such cases it
might be necessary to size those maps up, which libbpf allows to do by
overriding BPF_USDT_MAX_SPEC_CNT and BPF_USDT_MAX_IP_CNT macros.

It is an important aspect to keep in mind. Single USDT (user-space
equivalent of kernel tracepoint) can have multiple USDT "call sites".
That is, single logical USDT is triggered from multiple places in user
application. This can happen due to function inlining. Each such inlined
instance of USDT invocation can have its own unique USDT argument
specification (instructions about the location of the value of each of
USDT arguments). So while USDT looks very similar to usual uprobe or
kernel tracepoint, under the hood it's actually a collection of uprobes,
each potentially needing different spec to know how to fetch arguments.

User-visible API consists of three helper functions:
  - bpf_usdt_arg_cnt(), which returns number of arguments of current USDT;
  - bpf_usdt_arg(), which reads value of specified USDT argument (by
    it's zero-indexed position) and returns it as 64-bit value;
  - bpf_usdt_cookie(), which functions like BPF cookie for USDT
    programs; this is necessary as libbpf doesn't allow specifying actual
    BPF cookie and utilizes it internally for USDT support implementation.

Each bpf_usdt_xxx() APIs expect struct pt_regs * context, passed into
BPF program. On kernels that don't support BPF cookie it is used to
fetch absolute IP address of the underlying uprobe.

usdt.bpf.h also provides BPF_USDT() macro, which functions like
BPF_PROG() and BPF_KPROBE() and allows much more user-friendly way to
get access to USDT arguments, if USDT definition is static and known to
the user. It is expected that majority of use cases won't have to use
bpf_usdt_arg_cnt() and bpf_usdt_arg() directly and BPF_USDT() will cover
all their needs.

Last, usdt.bpf.h is utilizing BPF CO-RE for one single purpose: to
detect kernel support for BPF cookie. If BPF CO-RE dependency is
undesirable, user application can redefine BPF_USDT_HAS_BPF_COOKIE to
either a boolean constant (or equivalently zero and non-zero), or even
point it to its own .rodata variable that can be specified from user's
application user-space code. It is important that
BPF_USDT_HAS_BPF_COOKIE is known to BPF verifier as static value (thus
.rodata and not just .data), as otherwise BPF code will still contain
bpf_get_attach_cookie() BPF helper call and will fail validation at
runtime, if not dead-code eliminated.

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Reviewed-by: Alan Maguire <alan.maguire@oracle.com>
Link: https://lore.kernel.org/bpf/20220404234202.331384-2-andrii@kernel.org
2 years agolibbpf: Support Debian in resolve_full_path()
Ilya Leoshkevich [Mon, 4 Apr 2022 22:50:20 +0000 (00:50 +0200)]
libbpf: Support Debian in resolve_full_path()

attach_probe selftest fails on Debian-based distros with `failed to
resolve full path for 'libc.so.6'`. The reason is that these distros
embraced multiarch to the point where even for the "main" architecture
they store libc in /lib/<triple>.

This is configured in /etc/ld.so.conf and in theory it's possible to
replicate the loader's parsing and processing logic in libbpf, however
a much simpler solution is to just enumerate the known library paths.

Signed-off-by: Ilya Leoshkevich <iii@linux.ibm.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20220404225020.51029-1-iii@linux.ibm.com
2 years agoselftests/bpf: Define SYS_NANOSLEEP_KPROBE_NAME for aarch64
Ilya Leoshkevich [Mon, 4 Apr 2022 14:21:01 +0000 (16:21 +0200)]
selftests/bpf: Define SYS_NANOSLEEP_KPROBE_NAME for aarch64

attach_probe selftest fails on aarch64 with `failed to create kprobe
'sys_nanosleep+0x0' perf event: No such file or directory`. This is
because, like on several other architectures, nanosleep has a prefix.

Signed-off-by: Ilya Leoshkevich <iii@linux.ibm.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Tested-by: Alan Maguire <alan.maguire@oracle.com>
Link: https://lore.kernel.org/bpf/20220404142101.27900-1-iii@linux.ibm.com
2 years agoMerge branch 'bpf/bpftool: add program & link type names'
Andrii Nakryiko [Mon, 4 Apr 2022 21:51:48 +0000 (14:51 -0700)]
Merge branch 'bpf/bpftool: add program & link type names'

Milan Landaverde says:

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

With the addition of the syscall prog type we should now
be able to see feature probe info for that prog type:

    $ bpftool feature probe kernel
    ...
    eBPF program_type syscall is available
    ...
    eBPF helpers supported for program type syscall:
        ...
        - bpf_sys_bpf
        - bpf_sys_close

And for the link types, their names should aid in
the output.

Before:
    $ bpftool link show
    50: type 7  prog 5042
    bpf_cookie 0
    pids vfsstat(394433)

After:
    $ bpftool link show
    57: perf_event  prog 5058
    bpf_cookie 0
    pids vfsstat(394725)
====================

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
2 years agobpftool: Handle libbpf_probe_prog_type errors
Milan Landaverde [Thu, 31 Mar 2022 15:45:55 +0000 (11:45 -0400)]
bpftool: Handle libbpf_probe_prog_type errors

Previously [1], we were using bpf_probe_prog_type which returned a
bool, but the new libbpf_probe_bpf_prog_type can return a negative
error code on failure. This change decides for bpftool to declare
a program type is not available on probe failure.

[1] https://lore.kernel.org/bpf/20220202225916.3313522-3-andrii@kernel.org/

Signed-off-by: Milan Landaverde <milan@mdaverde.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Reviewed-by: Quentin Monnet <quentin@isovalent.com>
Link: https://lore.kernel.org/bpf/20220331154555.422506-4-milan@mdaverde.com
2 years agobpftool: Add missing link types
Milan Landaverde [Thu, 31 Mar 2022 15:45:54 +0000 (11:45 -0400)]
bpftool: Add missing link types

Will display the link type names in bpftool link show output

Signed-off-by: Milan Landaverde <milan@mdaverde.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20220331154555.422506-3-milan@mdaverde.com
2 years agobpftool: Add syscall prog type
Milan Landaverde [Thu, 31 Mar 2022 15:45:53 +0000 (11:45 -0400)]
bpftool: Add syscall prog type

In addition to displaying the program type in bpftool prog show
this enables us to be able to query bpf_prog_type_syscall
availability through feature probe as well as see
which helpers are available in those programs (such as
bpf_sys_bpf and bpf_sys_close)

Signed-off-by: Milan Landaverde <milan@mdaverde.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Reviewed-by: Quentin Monnet <quentin@isovalent.com>
Link: https://lore.kernel.org/bpf/20220331154555.422506-2-milan@mdaverde.com
2 years agoselftests/bpf: Fix parsing of prog types in UAPI hdr for bpftool sync
Quentin Monnet [Mon, 4 Apr 2022 14:09:44 +0000 (15:09 +0100)]
selftests/bpf: Fix parsing of prog types in UAPI hdr for bpftool sync

The script for checking that various lists of types in bpftool remain in
sync with the UAPI BPF header uses a regex to parse enum bpf_prog_type.
If this enum contains a set of values different from the list of program
types in bpftool, it complains.

This script should have reported the addition, some time ago, of the new
BPF_PROG_TYPE_SYSCALL, which was not reported to bpftool's program types
list. It failed to do so, because it failed to parse that new type from
the enum. This is because the new value, in the BPF header, has an
explicative comment on the same line, and the regex does not support
that.

Let's update the script to support parsing enum values when they have
comments on the same line.

Signed-off-by: Quentin Monnet <quentin@isovalent.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20220404140944.64744-1-quentin@isovalent.com
2 years agosamples: bpf: Fix linking xdp_router_ipv4 after migration
Alexander Lobakin [Mon, 4 Apr 2022 11:54:51 +0000 (13:54 +0200)]
samples: bpf: Fix linking xdp_router_ipv4 after migration

Users of the xdp_sample_user infra should be explicitly linked
with the standard math library (`-lm`). Otherwise, the following
happens:

/usr/bin/ld: xdp_sample_user.c:(.text+0x59fc): undefined reference to `ceil'
/usr/bin/ld: xdp_sample_user.c:(.text+0x5a0d): undefined reference to `ceil'
/usr/bin/ld: xdp_sample_user.c:(.text+0x5adc): undefined reference to `floor'
/usr/bin/ld: xdp_sample_user.c:(.text+0x5b01): undefined reference to `ceil'
/usr/bin/ld: xdp_sample_user.c:(.text+0x5c1e): undefined reference to `floor'
/usr/bin/ld: xdp_sample_user.c:(.text+0x5c43): undefined reference to `ceil
[...]

That happened previously, so there's a block of linkage flags in the
Makefile. xdp_router_ipv4 has been transferred to this infra quite
recently, but hasn't been added to it. Fix.

Fixes: 85bf1f51691c ("samples: bpf: Convert xdp_router_ipv4 to XDP samples helper")
Signed-off-by: Alexander Lobakin <alexandr.lobakin@intel.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20220404115451.1116478-1-alexandr.lobakin@intel.com
2 years agosample: bpf: syscall_tp_user: Print result of verify_map
Song Chen [Sat, 2 Apr 2022 08:57:08 +0000 (16:57 +0800)]
sample: bpf: syscall_tp_user: Print result of verify_map

At the end of the test, we already print out
    prog <prog number>: map ids <...> <...>
Value is the number read from kernel through bpf map, further print out
    verify map:<map id> val:<...>
will help users to understand the program runs successfully.

Signed-off-by: Song Chen <chensong_2000@189.cn>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/1648889828-12417-1-git-send-email-chensong_2000@189.cn
2 years agolibbpf: Don't return -EINVAL if hdr_len < offsetofend(core_relo_len)
Yuntao Wang [Mon, 4 Apr 2022 00:53:20 +0000 (08:53 +0800)]
libbpf: Don't return -EINVAL if hdr_len < offsetofend(core_relo_len)

Since core relos is an optional part of the .BTF.ext ELF section, we should
skip parsing it instead of returning -EINVAL if header size is less than
offsetofend(struct btf_ext_header, core_relo_len).

Signed-off-by: Yuntao Wang <ytcoode@gmail.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20220404005320.1723055-1-ytcoode@gmail.com
2 years agoMerge branch 'libbpf: name-based u[ret]probe attach'
Andrii Nakryiko [Mon, 4 Apr 2022 00:55:46 +0000 (17:55 -0700)]
Merge branch 'libbpf: name-based u[ret]probe attach'

Alan Maguire says:

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

This patch series focuses on supporting name-based attach - similar
to that supported for kprobes - for uprobe BPF programs.

Currently attach for such probes is done by determining the offset
manually, so the aim is to try and mimic the simplicity of kprobe
attach, making use of uprobe opts to specify a name string.
Patch 1 supports expansion of the binary_path argument used for
bpf_program__attach_uprobe_opts(), allowing it to determine paths
for programs and shared objects automatically, allowing for
specification of "libc.so.6" rather than the full path
"/usr/lib64/libc.so.6".

Patch 2 adds the "func_name" option to allow uprobe attach by
name; the mechanics are described there.

Having name-based support allows us to support auto-attach for
uprobes; patch 3 adds auto-attach support while attempting
to handle backwards-compatibility issues that arise.  The format
supported is

u[ret]probe/binary_path:[raw_offset|function[+offset]]

For example, to attach to libc malloc:

SEC("uprobe//usr/lib64/libc.so.6:malloc")

..or, making use of the path computation mechanisms introduced in patch 1

SEC("uprobe/libc.so.6:malloc")

Finally patch 4 add tests to the attach_probe selftests covering
attach by name, with patch 5 covering skeleton auto-attach.

Changes since v4 [1]:
- replaced strtok_r() usage with copying segments from static char *; avoids
  unneeded string allocation (Andrii, patch 1)
- switched to using access() instead of stat() when checking path-resolved
  binary (Andrii, patch 1)
- removed computation of .plt offset for instrumenting shared library calls
  within binaries.  Firstly it proved too brittle, and secondly it was somewhat
  unintuitive in that this form of instrumentation did not support function+offset
  as the "local function in binary" and "shared library function in shared library"
  cases did.  We can still instrument library calls, just need to do it in the
  library .so (patch 2)
- added binary path logging in cases where it was missing (Andrii, patch 2)
- avoid strlen() calcuation in checking name match (Andrii, patch 2)
- reword comments for func_name option (Andrii, patch 2)
- tightened SEC() name validation to support "u[ret]probe" and fail on other
  permutations that do not support auto-attach (i.e. have u[ret]probe/binary_path:func
  format (Andrii, patch 3)
- fixed selftests to fail independently rather than skip remainder on failure
  (Andrii, patches 4,5)
Changes since v3 [2]:
- reworked variable naming to fit better with libbpf conventions
  (Andrii, patch 2)
- use quoted binary path in log messages (Andrii, patch 2)
- added path determination mechanisms using LD_LIBRARY_PATH/PATH and
  standard locations (patch 1, Andrii)
- changed section lookup to be type+name (if name is specified) to
  simplify use cases (patch 2, Andrii)
- fixed .plt lookup scheme to match symbol table entries with .plt
  index via the .rela.plt table; also fix the incorrect assumption
  that the code in the .plt that does library linking is the same
  size as .plt entries (it just happens to be on x86_64)
- aligned with pluggable section support such that uprobe SEC() names
  that do not conform to auto-attach format do not cause skeleton load
  failure (patch 3, Andrii)
- no longer need to look up absolute path to libraries used by test_progs
  since we have mechanism to determine path automatically
- replaced CHECK()s with ASSERT*()s for attach_probe test (Andrii, patch 4)
- added auto-attach selftests also (Andrii, patch 5)
Changes since RFC [3]:
- used "long" for addresses instead of ssize_t (Andrii, patch 1).
- used gelf_ interfaces to avoid assumptions about 64-bit
  binaries (Andrii, patch 1)
- clarified string matching in symbol table lookups
  (Andrii, patch 1)
- added support for specification of shared object functions
  in a non-shared object binary.  This approach instruments
  the Procedure Linking Table (PLT) - malloc@PLT.
- changed logic in symbol search to check dynamic symbol table
  first, then fall back to symbol table (Andrii, patch 1).
- modified auto-attach string to require "/" separator prior
  to path prefix i.e. uprobe//path/to/binary (Andrii, patch 2)
- modified auto-attach string to use ':' separator (Andrii,
  patch 2)
- modified auto-attach to support raw offset (Andrii, patch 2)
- modified skeleton attach to interpret -ESRCH errors as
  a non-fatal "unable to auto-attach" (Andrii suggested
  -EOPNOTSUPP but my concern was it might collide with other
  instances where that value is returned and reflects a
  failure to attach a to-be-expected attachment rather than
  skip a program that does not present an auto-attachable
  section name. Admittedly -EOPNOTSUPP seems a more natural
  value here).
- moved library path retrieval code to trace_helpers (Andrii,
  patch 3)

[1] https://lore.kernel.org/bpf/1647000658-16149-1-git-send-email-alan.maguire@oracle.com/
[2] https://lore.kernel.org/bpf/1643645554-28723-1-git-send-email-alan.maguire@oracle.com/
[3] https://lore.kernel.org/bpf/1642678950-19584-1-git-send-email-alan.maguire@oracle.com/
====================

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
2 years agoselftests/bpf: Add tests for uprobe auto-attach via skeleton
Alan Maguire [Wed, 30 Mar 2022 15:26:40 +0000 (16:26 +0100)]
selftests/bpf: Add tests for uprobe auto-attach via skeleton

tests that verify auto-attach works for function entry/return for
local functions in program and library functions in a library.

Signed-off-by: Alan Maguire <alan.maguire@oracle.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/1648654000-21758-6-git-send-email-alan.maguire@oracle.com
2 years agoselftests/bpf: Add tests for u[ret]probe attach by name
Alan Maguire [Wed, 30 Mar 2022 15:26:39 +0000 (16:26 +0100)]
selftests/bpf: Add tests for u[ret]probe attach by name

add tests that verify attaching by name for

1. local functions in a program
2. library functions in a shared object

...succeed for uprobe and uretprobes using new "func_name"
option for bpf_program__attach_uprobe_opts().  Also verify
auto-attach works where uprobe, path to binary and function
name are specified, but fails with -EOPNOTSUPP with a SEC
name that does not specify binary path/function.

Signed-off-by: Alan Maguire <alan.maguire@oracle.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/1648654000-21758-5-git-send-email-alan.maguire@oracle.com
2 years agolibbpf: Add auto-attach for uprobes based on section name
Alan Maguire [Wed, 30 Mar 2022 15:26:38 +0000 (16:26 +0100)]
libbpf: Add auto-attach for uprobes based on section name

Now that u[ret]probes can use name-based specification, it makes
sense to add support for auto-attach based on SEC() definition.
The format proposed is

        SEC("u[ret]probe/binary:[raw_offset|[function_name[+offset]]")

For example, to trace malloc() in libc:

        SEC("uprobe/libc.so.6:malloc")

...or to trace function foo2 in /usr/bin/foo:

        SEC("uprobe//usr/bin/foo:foo2")

Auto-attach is done for all tasks (pid -1).  prog can be an absolute
path or simply a program/library name; in the latter case, we use
PATH/LD_LIBRARY_PATH to resolve the full path, falling back to
standard locations (/usr/bin:/usr/sbin or /usr/lib64:/usr/lib) if
the file is not found via environment-variable specified locations.

Signed-off-by: Alan Maguire <alan.maguire@oracle.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/1648654000-21758-4-git-send-email-alan.maguire@oracle.com
2 years agolibbpf: Support function name-based attach uprobes
Alan Maguire [Wed, 30 Mar 2022 15:26:37 +0000 (16:26 +0100)]
libbpf: Support function name-based attach uprobes

kprobe attach is name-based, using lookups of kallsyms to translate
a function name to an address.  Currently uprobe attach is done
via an offset value as described in [1].  Extend uprobe opts
for attach to include a function name which can then be converted
into a uprobe-friendly offset.  The calcualation is done in
several steps:

1. First, determine the symbol address using libelf; this gives us
   the offset as reported by objdump
2. If the function is a shared library function - and the binary
   provided is a shared library - no further work is required;
   the address found is the required address
3. Finally, if the function is local, subtract the base address
   associated with the object, retrieved from ELF program headers.

The resultant value is then added to the func_offset value passed
in to specify the uprobe attach address.  So specifying a func_offset
of 0 along with a function name "printf" will attach to printf entry.

The modes of operation supported are then

1. to attach to a local function in a binary; function "foo1" in
   "/usr/bin/foo"
2. to attach to a shared library function in a shared library -
   function "malloc" in libc.

[1] https://www.kernel.org/doc/html/latest/trace/uprobetracer.html

Signed-off-by: Alan Maguire <alan.maguire@oracle.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/1648654000-21758-3-git-send-email-alan.maguire@oracle.com
2 years agolibbpf: auto-resolve programs/libraries when necessary for uprobes
Alan Maguire [Wed, 30 Mar 2022 15:26:36 +0000 (16:26 +0100)]
libbpf: auto-resolve programs/libraries when necessary for uprobes

bpf_program__attach_uprobe_opts() requires a binary_path argument
specifying binary to instrument.  Supporting simply specifying
"libc.so.6" or "foo" should be possible too.

Library search checks LD_LIBRARY_PATH, then /usr/lib64, /usr/lib.
This allows users to run BPF programs prefixed with
LD_LIBRARY_PATH=/path2/lib while still searching standard locations.
Similarly for non .so files, we check PATH and /usr/bin, /usr/sbin.

Path determination will be useful for auto-attach of BPF uprobe programs
using SEC() definition.

Signed-off-by: Alan Maguire <alan.maguire@oracle.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/1648654000-21758-2-git-send-email-alan.maguire@oracle.com
2 years agosamples: bpf: Convert xdp_router_ipv4 to XDP samples helper
Lorenzo Bianconi [Wed, 16 Mar 2022 07:13:23 +0000 (08:13 +0100)]
samples: bpf: Convert xdp_router_ipv4 to XDP samples helper

Rely on the libbpf skeleton facility and other utilities provided by XDP
sample helpers in xdp_router_ipv4 sample.

Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/7f4d98ee2c13c04d5eb924eebf79ced32fee8418.1647414711.git.lorenzo@kernel.org
2 years agobpf: Correct the comment for BTF kind bitfield
Haiyue Wang [Sun, 3 Apr 2022 11:53:26 +0000 (19:53 +0800)]
bpf: Correct the comment for BTF kind bitfield

The commit 8fd886911a6a ("bpf: Add BTF_KIND_FLOAT to uapi") has extended
the BTF kind bitfield from 4 to 5 bits, correct the comment.

Signed-off-by: Haiyue Wang <haiyue.wang@intel.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20220403115327.205964-1-haiyue.wang@intel.com
2 years agoselftests/bpf: Fix cd_flavor_subdir() of test_progs
Yuntao Wang [Sun, 3 Apr 2022 13:52:45 +0000 (21:52 +0800)]
selftests/bpf: Fix cd_flavor_subdir() of test_progs

Currently, when we run test_progs with just executable file name, for
example 'PATH=. test_progs-no_alu32', cd_flavor_subdir() will not check
if test_progs is running as a flavored test runner and switch into
corresponding sub-directory.

This will cause test_progs-no_alu32 executed by the
'PATH=. test_progs-no_alu32' command to run in the wrong directory and
load the wrong BPF objects.

Signed-off-by: Yuntao Wang <ytcoode@gmail.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20220403135245.1713283-1-ytcoode@gmail.com
2 years agoselftests/bpf: Return true/false (not 1/0) from bool functions
Haowen Bai [Fri, 1 Apr 2022 02:15:54 +0000 (10:15 +0800)]
selftests/bpf: Return true/false (not 1/0) from bool functions

Return boolean values ("true" or "false") instead of 1 or 0 from bool
functions.  This fixes the following warnings from coccicheck:

./tools/testing/selftests/bpf/progs/test_xdp_noinline.c:567:9-10: WARNING:
return of 0/1 in function 'get_packet_dst' with return type bool
./tools/testing/selftests/bpf/progs/test_l4lb_noinline.c:221:9-10: WARNING:
return of 0/1 in function 'get_packet_dst' with return type bool

Signed-off-by: Haowen Bai <baihaowen@meizu.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Reviewed-by: Shuah Khan <skhan@linuxfoundation.org>
Acked-by: Yonghong Song <yhs@fb.com>
Link: https://lore.kernel.org/bpf/1648779354-14700-1-git-send-email-baihaowen@meizu.com
2 years agoselftests/bpf: Fix vfs_link kprobe definition
Nikolay Borisov [Thu, 31 Mar 2022 14:09:49 +0000 (17:09 +0300)]
selftests/bpf: Fix vfs_link kprobe definition

Since commit 6521f8917082 ("namei: prepare for idmapped mounts")
vfs_link's prototype was changed, the kprobe definition in
profiler selftest in turn wasn't updated. The result is that all
argument after the first are now stored in different registers. This
means that self-test has been broken ever since. Fix it by updating the
kprobe definition accordingly.

Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20220331140949.1410056-1-nborisov@suse.com
2 years agobpf: Replace usage of supported with dedicated list iterator variable
Jakob Koschel [Thu, 31 Mar 2022 09:19:29 +0000 (11:19 +0200)]
bpf: Replace usage of supported with dedicated list iterator variable

To move the list iterator variable into the list_for_each_entry_*()
macro in the future it should be avoided to use the list iterator
variable after the loop body.

To *never* use the list iterator variable after the loop it was
concluded to use a separate iterator variable instead of a
found boolean [1].

This removes the need to use the found variable (existed & supported)
and simply checking if the variable was set, can determine if the
break/goto was hit.

[1] https://lore.kernel.org/all/CAHk-=wgRr_D8CB-D9Kg-c=EHreAsk5SqXPwr9Y7k9sA6cWXJ6w@mail.gmail.com/

Signed-off-by: Jakob Koschel <jakobkoschel@gmail.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Acked-by: Yonghong Song <yhs@fb.com>
Link: https://lore.kernel.org/bpf/20220331091929.647057-1-jakobkoschel@gmail.com
2 years agobpf, test_offload.py: Skip base maps without names
Yauheni Kaliuta [Tue, 29 Mar 2022 08:11:00 +0000 (11:11 +0300)]
bpf, test_offload.py: Skip base maps without names

The test fails:

  # ./test_offload.py
  [...]
  Test bpftool bound info reporting (own ns)...
  FAIL: 3 BPF maps loaded, expected 2
    File "/root/bpf-next/tools/testing/selftests/bpf/./test_offload.py", line 1177, in <module>
      check_dev_info(False, "")
    File "/root/bpf-next/tools/testing/selftests/bpf/./test_offload.py", line 645, in check_dev_info
      maps = bpftool_map_list(expected=2, ns=ns)
    File "/root/bpf-next/tools/testing/selftests/bpf/./test_offload.py", line 190, in bpftool_map_list
      fail(True, "%d BPF maps loaded, expected %d" %
    File "/root/bpf-next/tools/testing/selftests/bpf/./test_offload.py", line 86, in fail
      tb = "".join(traceback.extract_stack().format())

Some base maps do not have names and they cannot be added due to compatibility
with older kernels, see [0]. So, just skip the unnamed maps.

  [0] https://lore.kernel.org/bpf/CAEf4BzY66WPKQbDe74AKZ6nFtZjq5e+G3Ji2egcVytB9R6_sGQ@mail.gmail.com/

Signed-off-by: Yauheni Kaliuta <ykaliuta@redhat.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Quentin Monnet <quentin@isovalent.com>
Link: https://lore.kernel.org/bpf/20220329081100.9705-1-ykaliuta@redhat.com
2 years agobpf: Remove redundant assignment to smap->map.value_size
Yuntao Wang [Wed, 23 Mar 2022 07:36:26 +0000 (15:36 +0800)]
bpf: Remove redundant assignment to smap->map.value_size

The attr->value_size is already assigned to smap->map.value_size
in bpf_map_init_from_attr(), there is no need to do it again in
stack_map_alloc().

Signed-off-by: Yuntao Wang <ytcoode@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Joanne Koong <joannelkoong@gmail.com>
Link: https://lore.kernel.org/bpf/20220323073626.958652-1-ytcoode@gmail.com
2 years agoselftests/bpf: Remove unused variable from bpf_sk_assign test
Eyal Birger [Tue, 29 Mar 2022 15:49:14 +0000 (18:49 +0300)]
selftests/bpf: Remove unused variable from bpf_sk_assign test

Was never used in bpf_sk_assign_test(), and was removed from handle_{tcp,udp}()
in commit 0b9ad56b1ea6 ("selftests/bpf: Use SOCKMAP for server sockets in
bpf_sk_assign test").

Fixes: 0b9ad56b1ea6 ("selftests/bpf: Use SOCKMAP for server sockets in bpf_sk_assign test")
Signed-off-by: Eyal Birger <eyal.birger@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20220329154914.3718658-1-eyal.birger@gmail.com
2 years agobpf: Use swap() instead of open coding it
Jiapeng Chong [Tue, 22 Mar 2022 06:21:49 +0000 (14:21 +0800)]
bpf: Use swap() instead of open coding it

Clean the following coccicheck warning:

./kernel/trace/bpf_trace.c:2263:34-35: WARNING opportunity for swap().
./kernel/trace/bpf_trace.c:2264:40-41: WARNING opportunity for swap().

Reported-by: Abaci Robot <abaci@linux.alibaba.com>
Signed-off-by: Jiapeng Chong <jiapeng.chong@linux.alibaba.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20220322062149.109180-1-jiapeng.chong@linux.alibaba.com
2 years agobpf, tests: Add load store test case for tail call
Xu Kuohai [Mon, 21 Mar 2022 15:28:52 +0000 (11:28 -0400)]
bpf, tests: Add load store test case for tail call

Add test case to enusre that the caller and callee's fp offsets are
correct during tail call (mainly asserting for arm64 JIT).

Tested on both big-endian and little-endian arm64 qemu, result:

 test_bpf: Summary: 1026 PASSED, 0 FAILED, [1014/1014 JIT'ed]
 test_bpf: test_tail_calls: Summary: 10 PASSED, 0 FAILED, [10/10 JIT'ed]
 test_bpf: test_skb_segment: Summary: 2 PASSED, 0 FAILED

Signed-off-by: Xu Kuohai <xukuohai@huawei.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20220321152852.2334294-6-xukuohai@huawei.com
2 years agobpf, tests: Add tests for BPF_LDX/BPF_STX with different offsets
Xu Kuohai [Mon, 21 Mar 2022 15:28:51 +0000 (11:28 -0400)]
bpf, tests: Add tests for BPF_LDX/BPF_STX with different offsets

This patch adds tests to verify the behavior of BPF_LDX/BPF_STX +
BPF_B/BPF_H/BPF_W/BPF_DW with negative offset, small positive offset,
large positive offset, and misaligned offset.

Tested on both big-endian and little-endian arm64 qemu, result:

 test_bpf: Summary: 1026 PASSED, 0 FAILED, [1014/1014 JIT'ed]']
 test_bpf: test_tail_calls: Summary: 8 PASSED, 0 FAILED, [8/8 JIT'ed]
 test_bpf: test_skb_segment: Summary: 2 PASSED, 0 FAILED

Signed-off-by: Xu Kuohai <xukuohai@huawei.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20220321152852.2334294-5-xukuohai@huawei.com
2 years agobpf, arm64: Adjust the offset of str/ldr(immediate) to positive number
Xu Kuohai [Mon, 21 Mar 2022 15:28:50 +0000 (11:28 -0400)]
bpf, arm64: Adjust the offset of str/ldr(immediate) to positive number

The BPF STX/LDX instruction uses offset relative to the FP to address
stack space. Since the BPF_FP locates at the top of the frame, the offset
is usually a negative number. However, arm64 str/ldr immediate instruction
requires that offset be a positive number.  Therefore, this patch tries to
convert the offsets.

The method is to find the negative offset furthest from the FP firstly.
Then add it to the FP, calculate a bottom position, called FPB, and then
adjust the offsets in other STR/LDX instructions relative to FPB.

FPB is saved using the callee-saved register x27 of arm64 which is not
used yet.

Before adjusting the offset, the patch checks every instruction to ensure
that the FP does not change in run-time. If the FP may change, no offset
is adjusted.

For example, for the following bpftrace command:

  bpftrace -e 'kprobe:do_sys_open { printf("opening: %s\n", str(arg1)); }'

Without this patch, jited code(fragment):

   0:   bti     c
   4:   stp     x29, x30, [sp, #-16]!
   8:   mov     x29, sp
   c:   stp     x19, x20, [sp, #-16]!
  10:   stp     x21, x22, [sp, #-16]!
  14:   stp     x25, x26, [sp, #-16]!
  18:   mov     x25, sp
  1c:   mov     x26, #0x0                       // #0
  20:   bti     j
  24:   sub     sp, sp, #0x90
  28:   add     x19, x0, #0x0
  2c:   mov     x0, #0x0                        // #0
  30:   mov     x10, #0xffffffffffffff78        // #-136
  34:   str     x0, [x25, x10]
  38:   mov     x10, #0xffffffffffffff80        // #-128
  3c:   str     x0, [x25, x10]
  40:   mov     x10, #0xffffffffffffff88        // #-120
  44:   str     x0, [x25, x10]
  48:   mov     x10, #0xffffffffffffff90        // #-112
  4c:   str     x0, [x25, x10]
  50:   mov     x10, #0xffffffffffffff98        // #-104
  54:   str     x0, [x25, x10]
  58:   mov     x10, #0xffffffffffffffa0        // #-96
  5c:   str     x0, [x25, x10]
  60:   mov     x10, #0xffffffffffffffa8        // #-88
  64:   str     x0, [x25, x10]
  68:   mov     x10, #0xffffffffffffffb0        // #-80
  6c:   str     x0, [x25, x10]
  70:   mov     x10, #0xffffffffffffffb8        // #-72
  74:   str     x0, [x25, x10]
  78:   mov     x10, #0xffffffffffffffc0        // #-64
  7c:   str     x0, [x25, x10]
  80:   mov     x10, #0xffffffffffffffc8        // #-56
  84:   str     x0, [x25, x10]
  88:   mov     x10, #0xffffffffffffffd0        // #-48
  8c:   str     x0, [x25, x10]
  90:   mov     x10, #0xffffffffffffffd8        // #-40
  94:   str     x0, [x25, x10]
  98:   mov     x10, #0xffffffffffffffe0        // #-32
  9c:   str     x0, [x25, x10]
  a0:   mov     x10, #0xffffffffffffffe8        // #-24
  a4:   str     x0, [x25, x10]
  a8:   mov     x10, #0xfffffffffffffff0        // #-16
  ac:   str     x0, [x25, x10]
  b0:   mov     x10, #0xfffffffffffffff8        // #-8
  b4:   str     x0, [x25, x10]
  b8:   mov     x10, #0x8                       // #8
  bc:   ldr     x2, [x19, x10]
  [...]

With this patch, jited code(fragment):

   0:   bti     c
   4:   stp     x29, x30, [sp, #-16]!
   8:   mov     x29, sp
   c:   stp     x19, x20, [sp, #-16]!
  10:   stp     x21, x22, [sp, #-16]!
  14:   stp     x25, x26, [sp, #-16]!
  18:   stp     x27, x28, [sp, #-16]!
  1c:   mov     x25, sp
  20:   sub     x27, x25, #0x88
  24:   mov     x26, #0x0                       // #0
  28:   bti     j
  2c:   sub     sp, sp, #0x90
  30:   add     x19, x0, #0x0
  34:   mov     x0, #0x0                        // #0
  38:   str     x0, [x27]
  3c:   str     x0, [x27, #8]
  40:   str     x0, [x27, #16]
  44:   str     x0, [x27, #24]
  48:   str     x0, [x27, #32]
  4c:   str     x0, [x27, #40]
  50:   str     x0, [x27, #48]
  54:   str     x0, [x27, #56]
  58:   str     x0, [x27, #64]
  5c:   str     x0, [x27, #72]
  60:   str     x0, [x27, #80]
  64:   str     x0, [x27, #88]
  68:   str     x0, [x27, #96]
  6c:   str     x0, [x27, #104]
  70:   str     x0, [x27, #112]
  74:   str     x0, [x27, #120]
  78:   str     x0, [x27, #128]
  7c:   ldr     x2, [x19, #8]
  [...]

Signed-off-by: Xu Kuohai <xukuohai@huawei.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20220321152852.2334294-4-xukuohai@huawei.com
2 years agobpf, arm64: Optimize BPF store/load using arm64 str/ldr(immediate offset)
Xu Kuohai [Mon, 21 Mar 2022 15:28:49 +0000 (11:28 -0400)]
bpf, arm64: Optimize BPF store/load using arm64 str/ldr(immediate offset)

The current BPF store/load instruction is translated by the JIT into two
instructions. The first instruction moves the immediate offset into a
temporary register. The second instruction uses this temporary register
to do the real store/load.

In fact, arm64 supports addressing with immediate offsets. So This patch
introduces optimization that uses arm64 str/ldr instruction with immediate
offset when the offset fits.

Example of generated instuction for r2 = *(u64 *)(r1 + 0):

without optimization:
mov x10, 0
ldr x1, [x0, x10]

with optimization:
ldr x1, [x0, 0]

If the offset is negative, or is not aligned correctly, or exceeds max
value, rollback to the use of temporary register.

Signed-off-by: Xu Kuohai <xukuohai@huawei.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20220321152852.2334294-3-xukuohai@huawei.com
2 years agoarm64, insn: Add ldr/str with immediate offset
Xu Kuohai [Mon, 21 Mar 2022 15:28:48 +0000 (11:28 -0400)]
arm64, insn: Add ldr/str with immediate offset

This patch introduces ldr/str with immediate offset support to simplify
the JIT implementation of BPF LDX/STX instructions on arm64. Although
arm64 ldr/str immediate is available in pre-index, post-index and
unsigned offset forms, the unsigned offset form is sufficient for BPF,
so this patch only adds this type.

Signed-off-by: Xu Kuohai <xukuohai@huawei.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20220321152852.2334294-2-xukuohai@huawei.com
2 years agoMerge tag 'net-5.18-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net
Linus Torvalds [Thu, 31 Mar 2022 18:23:31 +0000 (11:23 -0700)]
Merge tag 'net-5.18-rc1' of git://git./linux/kernel/git/netdev/net

Pull more networking updates from Jakub Kicinski:
 "Networking fixes and rethook patches.

  Features:

   - kprobes: rethook: x86: replace kretprobe trampoline with rethook

  Current release - regressions:

   - sfc: avoid null-deref on systems without NUMA awareness in the new
     queue sizing code

  Current release - new code bugs:

   - vxlan: do not feed vxlan_vnifilter_dump_dev with non-vxlan devices

   - eth: lan966x: fix null-deref on PHY pointer in timestamp ioctl when
     interface is down

  Previous releases - always broken:

   - openvswitch: correct neighbor discovery target mask field in the
     flow dump

   - wireguard: ignore v6 endpoints when ipv6 is disabled and fix a leak

   - rxrpc: fix call timer start racing with call destruction

   - rxrpc: fix null-deref when security type is rxrpc_no_security

   - can: fix UAF bugs around echo skbs in multiple drivers

  Misc:

   - docs: move netdev-FAQ to the 'process' section of the
     documentation"

* tag 'net-5.18-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net: (57 commits)
  vxlan: do not feed vxlan_vnifilter_dump_dev with non vxlan devices
  openvswitch: Add recirc_id to recirc warning
  rxrpc: fix some null-ptr-deref bugs in server_key.c
  rxrpc: Fix call timer start racing with call destruction
  net: hns3: fix software vlan talbe of vlan 0 inconsistent with hardware
  net: hns3: fix the concurrency between functions reading debugfs
  docs: netdev: move the netdev-FAQ to the process pages
  docs: netdev: broaden the new vs old code formatting guidelines
  docs: netdev: call out the merge window in tag checking
  docs: netdev: add missing back ticks
  docs: netdev: make the testing requirement more stringent
  docs: netdev: add a question about re-posting frequency
  docs: netdev: rephrase the 'should I update patchwork' question
  docs: netdev: rephrase the 'Under review' question
  docs: netdev: shorten the name and mention msgid for patch status
  docs: netdev: note that RFC postings are allowed any time
  docs: netdev: turn the net-next closed into a Warning
  docs: netdev: move the patch marking section up
  docs: netdev: minor reword
  docs: netdev: replace references to old archives
  ...

2 years agoMerge tag 'v5.18-p1' of git://git.kernel.org/pub/scm/linux/kernel/git/herbert/crypto-2.6
Linus Torvalds [Thu, 31 Mar 2022 18:17:39 +0000 (11:17 -0700)]
Merge tag 'v5.18-p1' of git://git./linux/kernel/git/herbert/crypto-2.6

Pull crypto fixes from Herbert Xu:

 - Missing Kconfig dependency on arm that leads to boot failure

 - x86 SLS fixes

 - Reference leak in the stm32 driver

* tag 'v5.18-p1' of git://git.kernel.org/pub/scm/linux/kernel/git/herbert/crypto-2.6:
  crypto: x86/sm3 - Fixup SLS
  crypto: x86/poly1305 - Fixup SLS
  crypto: x86/chacha20 - Avoid spurious jumps to other functions
  crypto: stm32 - fix reference leak in stm32_crc_remove
  crypto: arm/aes-neonbs-cbc - Select generic cbc and aes

2 years agovxlan: do not feed vxlan_vnifilter_dump_dev with non vxlan devices
Eric Dumazet [Wed, 30 Mar 2022 19:46:43 +0000 (12:46 -0700)]
vxlan: do not feed vxlan_vnifilter_dump_dev with non vxlan devices

vxlan_vnifilter_dump_dev() assumes it is called only
for vxlan devices. Make sure it is the case.

BUG: KASAN: slab-out-of-bounds in vxlan_vnifilter_dump_dev+0x9a0/0xb40 drivers/net/vxlan/vxlan_vnifilter.c:349
Read of size 4 at addr ffff888060d1ce70 by task syz-executor.3/17662

CPU: 0 PID: 17662 Comm: syz-executor.3 Tainted: G        W         5.17.0-syzkaller-12888-g77c9387c0c5b #0
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011
Call Trace:
 <TASK>
 __dump_stack lib/dump_stack.c:88 [inline]
 dump_stack_lvl+0xcd/0x134 lib/dump_stack.c:106
 print_address_description.constprop.0.cold+0xeb/0x495 mm/kasan/report.c:313
 print_report mm/kasan/report.c:429 [inline]
 kasan_report.cold+0xf4/0x1c6 mm/kasan/report.c:491
 vxlan_vnifilter_dump_dev+0x9a0/0xb40 drivers/net/vxlan/vxlan_vnifilter.c:349
 vxlan_vnifilter_dump+0x3ff/0x650 drivers/net/vxlan/vxlan_vnifilter.c:428
 netlink_dump+0x4b5/0xb70 net/netlink/af_netlink.c:2270
 __netlink_dump_start+0x647/0x900 net/netlink/af_netlink.c:2375
 netlink_dump_start include/linux/netlink.h:245 [inline]
 rtnetlink_rcv_msg+0x70c/0xb80 net/core/rtnetlink.c:5953
 netlink_rcv_skb+0x153/0x420 net/netlink/af_netlink.c:2496
 netlink_unicast_kernel net/netlink/af_netlink.c:1319 [inline]
 netlink_unicast+0x543/0x7f0 net/netlink/af_netlink.c:1345
 netlink_sendmsg+0x904/0xe00 net/netlink/af_netlink.c:1921
 sock_sendmsg_nosec net/socket.c:705 [inline]
 sock_sendmsg+0xcf/0x120 net/socket.c:725
 ____sys_sendmsg+0x6e2/0x800 net/socket.c:2413
 ___sys_sendmsg+0xf3/0x170 net/socket.c:2467
 __sys_sendmsg+0xe5/0x1b0 net/socket.c:2496
 do_syscall_x64 arch/x86/entry/common.c:50 [inline]
 do_syscall_64+0x35/0x80 arch/x86/entry/common.c:80
 entry_SYSCALL_64_after_hwframe+0x44/0xae
RIP: 0033:0x7f87b8e89049

Fixes: f9c4bb0b245c ("vxlan: vni filtering support on collect metadata device")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: syzbot <syzkaller@googlegroups.com>
Acked-by: Roopa Prabhu <roopa@nvidia.com>
Link: https://lore.kernel.org/r/20220330194643.2706132-1-eric.dumazet@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 years agoopenvswitch: Add recirc_id to recirc warning
Stéphane Graber [Wed, 30 Mar 2022 19:42:45 +0000 (15:42 -0400)]
openvswitch: Add recirc_id to recirc warning

When hitting the recirculation limit, the kernel would currently log
something like this:

[   58.586597] openvswitch: ovs-system: deferred action limit reached, drop recirc action

Which isn't all that useful to debug as we only have the interface name
to go on but can't track it down to a specific flow.

With this change, we now instead get:

[   58.586597] openvswitch: ovs-system: deferred action limit reached, drop recirc action (recirc_id=0x9e)

Which can now be correlated with the flow entries from OVS.

Suggested-by: Frode Nordahl <frode.nordahl@canonical.com>
Signed-off-by: Stéphane Graber <stgraber@ubuntu.com>
Tested-by: Stephane Graber <stgraber@ubuntu.com>
Acked-by: Eelco Chaudron <echaudro@redhat.com>
Link: https://lore.kernel.org/r/20220330194244.3476544-1-stgraber@ubuntu.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 years agoMerge tag 'linux-can-fixes-for-5.18-20220331' of git://git.kernel.org/pub/scm/linux...
Jakub Kicinski [Thu, 31 Mar 2022 15:36:17 +0000 (08:36 -0700)]
Merge tag 'linux-can-fixes-for-5.18-20220331' of git://git./linux/kernel/git/mkl/linux-can

Marc Kleine-Budde says:

====================
pull-request: can 2022-03-31

The first patch is by Oliver Hartkopp and fixes MSG_PEEK feature in
the CAN ISOTP protocol (broken in net-next for v5.18 only).

Tom Rix's patch for the mcp251xfd driver fixes the propagation of an
error value in case of an error.

A patch by me for the m_can driver fixes a use-after-free in the xmit
handler for m_can IP cores v3.0.x.

Hangyu Hua contributes 3 patches fixing the same double free in the
error path of the xmit handler in the ems_usb, usb_8dev and mcba_usb
USB CAN driver.

Pavel Skripkin contributes a patch for the mcba_usb driver to properly
check the endpoint type.

The last patch is by me and fixes a mem leak in the gs_usb, which was
introduced in net-next for v5.18.

* tag 'linux-can-fixes-for-5.18-20220331' of git://git.kernel.org/pub/scm/linux/kernel/git/mkl/linux-can:
  can: gs_usb: gs_make_candev(): fix memory leak for devices with extended bit timing configuration
  can: mcba_usb: properly check endpoint type
  can: mcba_usb: mcba_usb_start_xmit(): fix double dev_kfree_skb in error path
  can: usb_8dev: usb_8dev_start_xmit(): fix double dev_kfree_skb() in error path
  can: ems_usb: ems_usb_start_xmit(): fix double dev_kfree_skb() in error path
  can: m_can: m_can_tx_handler(): fix use after free of skb
  can: mcp251xfd: mcp251xfd_register_get_dev_id(): fix return of error value
  can: isotp: restore accidentally removed MSG_PEEK feature
====================

Link: https://lore.kernel.org/r/
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 years agorxrpc: fix some null-ptr-deref bugs in server_key.c
Xiaolong Huang [Wed, 30 Mar 2022 14:22:14 +0000 (15:22 +0100)]
rxrpc: fix some null-ptr-deref bugs in server_key.c

Some function calls are not implemented in rxrpc_no_security, there are
preparse_server_key, free_preparse_server_key and destroy_server_key.
When rxrpc security type is rxrpc_no_security, user can easily trigger a
null-ptr-deref bug via ioctl. So judgment should be added to prevent it

The crash log:
user@syzkaller:~$ ./rxrpc_preparse_s
[   37.956878][T15626] BUG: kernel NULL pointer dereference, address: 0000000000000000
[   37.957645][T15626] #PF: supervisor instruction fetch in kernel mode
[   37.958229][T15626] #PF: error_code(0x0010) - not-present page
[   37.958762][T15626] PGD 4aadf067 P4D 4aadf067 PUD 4aade067 PMD 0
[   37.959321][T15626] Oops: 0010 [#1] PREEMPT SMP
[   37.959739][T15626] CPU: 0 PID: 15626 Comm: rxrpc_preparse_ Not tainted 5.17.0-01442-gb47d5a4f6b8d #43
[   37.960588][T15626] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.13.0-1ubuntu1 04/01/2014
[   37.961474][T15626] RIP: 0010:0x0
[   37.961787][T15626] Code: Unable to access opcode bytes at RIP 0xffffffffffffffd6.
[   37.962480][T15626] RSP: 0018:ffffc9000d9abdc0 EFLAGS: 00010286
[   37.963018][T15626] RAX: ffffffff84335200 RBX: ffff888012a1ce80 RCX: 0000000000000000
[   37.963727][T15626] RDX: 0000000000000000 RSI: ffffffff84a736dc RDI: ffffc9000d9abe48
[   37.964425][T15626] RBP: ffffc9000d9abe48 R08: 0000000000000000 R09: 0000000000000002
[   37.965118][T15626] R10: 000000000000000a R11: f000000000000000 R12: ffff888013145680
[   37.965836][T15626] R13: 0000000000000000 R14: ffffffffffffffec R15: ffff8880432aba80
[   37.966441][T15626] FS:  00007f2177907700(0000) GS:ffff88803ec00000(0000) knlGS:0000000000000000
[   37.966979][T15626] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[   37.967384][T15626] CR2: ffffffffffffffd6 CR3: 000000004aaf1000 CR4: 00000000000006f0
[   37.967864][T15626] Call Trace:
[   37.968062][T15626]  <TASK>
[   37.968240][T15626]  rxrpc_preparse_s+0x59/0x90
[   37.968541][T15626]  key_create_or_update+0x174/0x510
[   37.968863][T15626]  __x64_sys_add_key+0x139/0x1d0
[   37.969165][T15626]  do_syscall_64+0x35/0xb0
[   37.969451][T15626]  entry_SYSCALL_64_after_hwframe+0x44/0xae
[   37.969824][T15626] RIP: 0033:0x43a1f9

Signed-off-by: Xiaolong Huang <butterflyhuangxx@gmail.com>
Tested-by: Xiaolong Huang <butterflyhuangxx@gmail.com>
Signed-off-by: David Howells <dhowells@redhat.com>
Acked-by: Marc Dionne <marc.dionne@auristor.com>
cc: linux-afs@lists.infradead.org
Link: http://lists.infradead.org/pipermail/linux-afs/2022-March/005069.html
Fixes: 12da59fcab5a ("rxrpc: Hand server key parsing off to the security class")
Link: https://lore.kernel.org/r/164865013439.2941502.8966285221215590921.stgit@warthog.procyon.org.uk
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2 years agorxrpc: Fix call timer start racing with call destruction
David Howells [Wed, 30 Mar 2022 14:39:16 +0000 (15:39 +0100)]
rxrpc: Fix call timer start racing with call destruction

The rxrpc_call struct has a timer used to handle various timed events
relating to a call.  This timer can get started from the packet input
routines that are run in softirq mode with just the RCU read lock held.
Unfortunately, because only the RCU read lock is held - and neither ref or
other lock is taken - the call can start getting destroyed at the same time
a packet comes in addressed to that call.  This causes the timer - which
was already stopped - to get restarted.  Later, the timer dispatch code may
then oops if the timer got deallocated first.

Fix this by trying to take a ref on the rxrpc_call struct and, if
successful, passing that ref along to the timer.  If the timer was already
running, the ref is discarded.

The timer completion routine can then pass the ref along to the call's work
item when it queues it.  If the timer or work item where already
queued/running, the extra ref is discarded.

Fixes: a158bdd3247b ("rxrpc: Fix call timeouts")
Reported-by: Marc Dionne <marc.dionne@auristor.com>
Signed-off-by: David Howells <dhowells@redhat.com>
Reviewed-by: Marc Dionne <marc.dionne@auristor.com>
Tested-by: Marc Dionne <marc.dionne@auristor.com>
cc: linux-afs@lists.infradead.org
Link: http://lists.infradead.org/pipermail/linux-afs/2022-March/005073.html
Link: https://lore.kernel.org/r/164865115696.2943015.11097991776647323586.stgit@warthog.procyon.org.uk
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2 years agoMerge branch 'net-hns3-add-two-fixes-for-net'
Paolo Abeni [Thu, 31 Mar 2022 09:40:02 +0000 (11:40 +0200)]
Merge branch 'net-hns3-add-two-fixes-for-net'

Guangbin Huang says:

====================
net: hns3: add two fixes for -net

This series adds two fixes for the HNS3 ethernet driver.
====================

Link: https://lore.kernel.org/r/20220330134506.36635-1-huangguangbin2@huawei.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2 years agonet: hns3: fix software vlan talbe of vlan 0 inconsistent with hardware
Guangbin Huang [Wed, 30 Mar 2022 13:45:06 +0000 (21:45 +0800)]
net: hns3: fix software vlan talbe of vlan 0 inconsistent with hardware

When user delete vlan 0, as driver will not delete vlan 0 for hardware in
function hclge_set_vlan_filter_hw(), so vlan 0 in software vlan talbe should
not be deleted.

Fixes: fe4144d47eef ("net: hns3: sync VLAN filter entries when kill VLAN ID failed")
Signed-off-by: Guangbin Huang <huangguangbin2@huawei.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2 years agonet: hns3: fix the concurrency between functions reading debugfs
Yufeng Mo [Wed, 30 Mar 2022 13:45:05 +0000 (21:45 +0800)]
net: hns3: fix the concurrency between functions reading debugfs

Currently, the debugfs mechanism is that all functions share a
global variable to save the pointer for obtaining data. When
different functions concurrently access the same file node,
repeated release exceptions occur. Therefore, the granularity
of the pointer for storing the obtained data is adjusted to be
private for each function.

Fixes: 5e69ea7ee2a6 ("net: hns3: refactor the debugfs process")
Signed-off-by: Yufeng Mo <moyufeng@huawei.com>
Signed-off-by: Guangbin Huang <huangguangbin2@huawei.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2 years agoMerge branch 'docs-update-and-move-the-netdev-faq'
Paolo Abeni [Thu, 31 Mar 2022 08:49:42 +0000 (10:49 +0200)]
Merge branch 'docs-update-and-move-the-netdev-faq'

Jakub Kicinski says:

====================
docs: update and move the netdev-FAQ

A section of documentation for tree-specific process quirks had
been created a while back. There's only one tree in it, so far,
the tip tree, but the contents seem to answer similar questions
as we answer in the netdev-FAQ. Move the netdev-FAQ.

Take this opportunity to touch up and update a few sections.

v3: remove some confrontational? language from patch 7
v2: remove non-git in patch 3
    add patch 5
====================

Link: https://lore.kernel.org/r/20220330042505.2902770-1-kuba@kernel.org
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2 years agodocs: netdev: move the netdev-FAQ to the process pages
Jakub Kicinski [Wed, 30 Mar 2022 04:25:05 +0000 (21:25 -0700)]
docs: netdev: move the netdev-FAQ to the process pages

The documentation for the tip tree is really in quite a similar
spirit to the netdev-FAQ. Move the netdev-FAQ to the process docs
as well.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2 years agodocs: netdev: broaden the new vs old code formatting guidelines
Jakub Kicinski [Wed, 30 Mar 2022 04:25:04 +0000 (21:25 -0700)]
docs: netdev: broaden the new vs old code formatting guidelines

Convert the "should I use new or old comment formatting" to cover
all formatting. This makes the question itself shorter.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2 years agodocs: netdev: call out the merge window in tag checking
Jakub Kicinski [Wed, 30 Mar 2022 04:25:03 +0000 (21:25 -0700)]
docs: netdev: call out the merge window in tag checking

Add the most important case to the question about "where are we
in the cycle" - the case of net-next being closed.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2 years agodocs: netdev: add missing back ticks
Jakub Kicinski [Wed, 30 Mar 2022 04:25:02 +0000 (21:25 -0700)]
docs: netdev: add missing back ticks

I think double back ticks are more correct. Add where they are missing.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2 years agodocs: netdev: make the testing requirement more stringent
Jakub Kicinski [Wed, 30 Mar 2022 04:25:01 +0000 (21:25 -0700)]
docs: netdev: make the testing requirement more stringent

These days we often ask for selftests so let's update our
testing requirements.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2 years agodocs: netdev: add a question about re-posting frequency
Jakub Kicinski [Wed, 30 Mar 2022 04:25:00 +0000 (21:25 -0700)]
docs: netdev: add a question about re-posting frequency

We have to tell people to stop reposting to often lately,
or not to repost while the discussion is ongoing.
Document this.

Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2 years agodocs: netdev: rephrase the 'should I update patchwork' question
Jakub Kicinski [Wed, 30 Mar 2022 04:24:59 +0000 (21:24 -0700)]
docs: netdev: rephrase the 'should I update patchwork' question

Make the question shorter and adjust the start of the answer accordingly.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2 years agodocs: netdev: rephrase the 'Under review' question
Jakub Kicinski [Wed, 30 Mar 2022 04:24:58 +0000 (21:24 -0700)]
docs: netdev: rephrase the 'Under review' question

The semantics of "Under review" have shifted. Reword the question
about it a bit and focus it on the response time.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2 years agodocs: netdev: shorten the name and mention msgid for patch status
Jakub Kicinski [Wed, 30 Mar 2022 04:24:57 +0000 (21:24 -0700)]
docs: netdev: shorten the name and mention msgid for patch status

Cut down the length of the question so it renders better in docs.
Mention that Message-ID can be used to search patchwork.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2 years agodocs: netdev: note that RFC postings are allowed any time
Jakub Kicinski [Wed, 30 Mar 2022 04:24:56 +0000 (21:24 -0700)]
docs: netdev: note that RFC postings are allowed any time

Document that RFCs are allowed during the merge window.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2 years agodocs: netdev: turn the net-next closed into a Warning
Jakub Kicinski [Wed, 30 Mar 2022 04:24:55 +0000 (21:24 -0700)]
docs: netdev: turn the net-next closed into a Warning

Use the sphinx Warning box to make the net-next being closed
stand out more.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2 years agodocs: netdev: move the patch marking section up
Jakub Kicinski [Wed, 30 Mar 2022 04:24:54 +0000 (21:24 -0700)]
docs: netdev: move the patch marking section up

We want people to mark their patches with net and net-next in the subject.
Many miss doing that. Move the FAQ section which points that out up, and
place it after the section which enumerates the trees, that seems like
a pretty logical place for it. Since the two sections are together we
can remove a little bit (not too much) of the repetition.

v2: also remove the text for non-git setups, we want people to use git.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2 years agodocs: netdev: minor reword
Jakub Kicinski [Wed, 30 Mar 2022 04:24:53 +0000 (21:24 -0700)]
docs: netdev: minor reword

that -> those

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2 years agodocs: netdev: replace references to old archives
Jakub Kicinski [Wed, 30 Mar 2022 04:24:52 +0000 (21:24 -0700)]
docs: netdev: replace references to old archives

Most people use (or should use) lore at this point.
Replace the pointers to older archiving systems.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2 years agocan: gs_usb: gs_make_candev(): fix memory leak for devices with extended bit timing...
Marc Kleine-Budde [Tue, 29 Mar 2022 19:29:43 +0000 (21:29 +0200)]
can: gs_usb: gs_make_candev(): fix memory leak for devices with extended bit timing configuration

Some CAN-FD capable devices offer extended bit timing information for
the data bit timing. The information must be read with an USB control
message. The memory for this message is allocated but not free()ed (in
the non error case). This patch adds the missing free.

Fixes: 6679f4c5e5a6 ("can: gs_usb: add extended bt_const feature")
Link: https://lore.kernel.org/all/20220329193450.659726-1-mkl@pengutronix.de
Reported-by: syzbot+4d0ae90a195b269f102d@syzkaller.appspotmail.com
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
2 years agocan: mcba_usb: properly check endpoint type
Pavel Skripkin [Sun, 13 Mar 2022 10:09:03 +0000 (13:09 +0300)]
can: mcba_usb: properly check endpoint type

Syzbot reported warning in usb_submit_urb() which is caused by wrong
endpoint type. We should check that in endpoint is actually present to
prevent this warning.

Found pipes are now saved to struct mcba_priv and code uses them
directly instead of making pipes in place.

Fail log:

| usb 5-1: BOGUS urb xfer, pipe 3 != type 1
| WARNING: CPU: 1 PID: 49 at drivers/usb/core/urb.c:502 usb_submit_urb+0xed2/0x18a0 drivers/usb/core/urb.c:502
| Modules linked in:
| CPU: 1 PID: 49 Comm: kworker/1:2 Not tainted 5.17.0-rc6-syzkaller-00184-g38f80f42147f #0
| Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 1.14.0-2 04/01/2014
| Workqueue: usb_hub_wq hub_event
| RIP: 0010:usb_submit_urb+0xed2/0x18a0 drivers/usb/core/urb.c:502
| ...
| Call Trace:
|  <TASK>
|  mcba_usb_start drivers/net/can/usb/mcba_usb.c:662 [inline]
|  mcba_usb_probe+0x8a3/0xc50 drivers/net/can/usb/mcba_usb.c:858
|  usb_probe_interface+0x315/0x7f0 drivers/usb/core/driver.c:396
|  call_driver_probe drivers/base/dd.c:517 [inline]

Fixes: 51f3baad7de9 ("can: mcba_usb: Add support for Microchip CAN BUS Analyzer")
Link: https://lore.kernel.org/all/20220313100903.10868-1-paskripkin@gmail.com
Reported-and-tested-by: syzbot+3bc1dce0cc0052d60fde@syzkaller.appspotmail.com
Signed-off-by: Pavel Skripkin <paskripkin@gmail.com>
Reviewed-by: Vincent Mailhol <mailhol.vincent@wanadoo.fr>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
2 years agocan: mcba_usb: mcba_usb_start_xmit(): fix double dev_kfree_skb in error path
Hangyu Hua [Fri, 11 Mar 2022 08:02:08 +0000 (16:02 +0800)]
can: mcba_usb: mcba_usb_start_xmit(): fix double dev_kfree_skb in error path

There is no need to call dev_kfree_skb() when usb_submit_urb() fails
because can_put_echo_skb() deletes original skb and
can_free_echo_skb() deletes the cloned skb.

Fixes: 51f3baad7de9 ("can: mcba_usb: Add support for Microchip CAN BUS Analyzer")
Link: https://lore.kernel.org/all/20220311080208.45047-1-hbh25y@gmail.com
Signed-off-by: Hangyu Hua <hbh25y@gmail.com>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
2 years agocan: usb_8dev: usb_8dev_start_xmit(): fix double dev_kfree_skb() in error path
Hangyu Hua [Fri, 11 Mar 2022 08:06:14 +0000 (16:06 +0800)]
can: usb_8dev: usb_8dev_start_xmit(): fix double dev_kfree_skb() in error path

There is no need to call dev_kfree_skb() when usb_submit_urb() fails
because can_put_echo_skb() deletes original skb and
can_free_echo_skb() deletes the cloned skb.

Fixes: 0024d8ad1639 ("can: usb_8dev: Add support for USB2CAN interface from 8 devices")
Link: https://lore.kernel.org/all/20220311080614.45229-1-hbh25y@gmail.com
Cc: stable@vger.kernel.org
Signed-off-by: Hangyu Hua <hbh25y@gmail.com>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
2 years agocan: ems_usb: ems_usb_start_xmit(): fix double dev_kfree_skb() in error path
Hangyu Hua [Mon, 28 Feb 2022 08:36:39 +0000 (16:36 +0800)]
can: ems_usb: ems_usb_start_xmit(): fix double dev_kfree_skb() in error path

There is no need to call dev_kfree_skb() when usb_submit_urb() fails
beacause can_put_echo_skb() deletes the original skb and
can_free_echo_skb() deletes the cloned skb.

Link: https://lore.kernel.org/all/20220228083639.38183-1-hbh25y@gmail.com
Fixes: 702171adeed3 ("ems_usb: Added support for EMS CPC-USB/ARM7 CAN/USB interface")
Cc: stable@vger.kernel.org
Cc: Sebastian Haas <haas@ems-wuensche.com>
Signed-off-by: Hangyu Hua <hbh25y@gmail.com>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
2 years agocan: m_can: m_can_tx_handler(): fix use after free of skb
Marc Kleine-Budde [Thu, 17 Mar 2022 07:57:35 +0000 (08:57 +0100)]
can: m_can: m_can_tx_handler(): fix use after free of skb

can_put_echo_skb() will clone skb then free the skb. Move the
can_put_echo_skb() for the m_can version 3.0.x directly before the
start of the xmit in hardware, similar to the 3.1.x branch.

Fixes: 80646733f11c ("can: m_can: update to support CAN FD features")
Link: https://lore.kernel.org/all/20220317081305.739554-1-mkl@pengutronix.de
Cc: stable@vger.kernel.org
Reported-by: Hangyu Hua <hbh25y@gmail.com>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
2 years agocan: mcp251xfd: mcp251xfd_register_get_dev_id(): fix return of error value
Tom Rix [Sat, 19 Mar 2022 15:31:28 +0000 (08:31 -0700)]
can: mcp251xfd: mcp251xfd_register_get_dev_id(): fix return of error value

Clang static analysis reports this issue:

| mcp251xfd-core.c:1813:7: warning: The left operand
|   of '&' is a garbage value
|   FIELD_GET(MCP251XFD_REG_DEVID_ID_MASK, dev_id),
|   ^                                      ~~~~~~

dev_id is set in a successful call to mcp251xfd_register_get_dev_id().
Though the status of calls made by mcp251xfd_register_get_dev_id() are
checked and handled, their status' are not returned. So return err.

Fixes: 55e5b97f003e ("can: mcp25xxfd: add driver for Microchip MCP25xxFD SPI CAN")
Link: https://lore.kernel.org/all/20220319153128.2164120-1-trix@redhat.com
Signed-off-by: Tom Rix <trix@redhat.com>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
2 years agocan: isotp: restore accidentally removed MSG_PEEK feature
Oliver Hartkopp [Mon, 28 Mar 2022 11:36:11 +0000 (13:36 +0200)]
can: isotp: restore accidentally removed MSG_PEEK feature

In commit 42bf50a1795a ("can: isotp: support MSG_TRUNC flag when
reading from socket") a new check for recvmsg flags has been
introduced that only checked for the flags that are handled in
isotp_recvmsg() itself.

This accidentally removed the MSG_PEEK feature flag which is processed
later in the call chain in __skb_try_recv_from_queue().

Add MSG_PEEK to the set of valid flags to restore the feature.

Fixes: 42bf50a1795a ("can: isotp: support MSG_TRUNC flag when reading from socket")
Link: https://github.com/linux-can/can-utils/issues/347#issuecomment-1079554254
Link: https://lore.kernel.org/all/20220328113611.3691-1-socketcan@hartkopp.net
Reported-by: Derek Will <derekrobertwill@gmail.com>
Suggested-by: Derek Will <derekrobertwill@gmail.com>
Tested-by: Derek Will <derekrobertwill@gmail.com>
Signed-off-by: Oliver Hartkopp <socketcan@hartkopp.net>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
2 years agonet: sparx5: uses, depends on BRIDGE or !BRIDGE
Randy Dunlap [Wed, 30 Mar 2022 01:20:25 +0000 (18:20 -0700)]
net: sparx5: uses, depends on BRIDGE or !BRIDGE

Fix build errors when BRIDGE=m and SPARX5_SWITCH=y:

riscv64-linux-ld: drivers/net/ethernet/microchip/sparx5/sparx5_switchdev.o: in function `.L305':
sparx5_switchdev.c:(.text+0xdb0): undefined reference to `br_vlan_enabled'
riscv64-linux-ld: drivers/net/ethernet/microchip/sparx5/sparx5_switchdev.o: in function `.L283':
sparx5_switchdev.c:(.text+0xee0): undefined reference to `br_vlan_enabled'

Fixes: 3cfa11bac9bb ("net: sparx5: add the basic sparx5 driver")
Signed-off-by: Randy Dunlap <rdunlap@infradead.org>
Reported-by: kernel test robot <lkp@intel.com>
Cc: Horatiu Vultur <horatiu.vultur@microchip.com>
Cc: Lars Povlsen <lars.povlsen@microchip.com>
Cc: Steen Hegelund <Steen.Hegelund@microchip.com>
Cc: UNGLinuxDriver@microchip.com
Cc: Paolo Abeni <pabeni@redhat.com>
Link: https://lore.kernel.org/r/20220330012025.29560-1-rdunlap@infradead.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>