The synthesize benchmark, run on a single process and thread, shows
perf_event__synthesize_mmap_events as the hottest function with fgets
and sscanf taking the majority of execution time.
fscanf performs similarly well. Replace the scanf call with manual
reading of each field of the /proc/pid/maps line, and remove some
unnecessary buffering.
This change also addresses potential, but unlikely, buffer overruns for
the string values read by scanf.
Performance before is:
$ sudo perf bench internals synthesize -m 16 -M 16 -s -t
\# Running 'internals/synthesize' benchmark:
Computing performance of single threaded perf event synthesis by
synthesizing events on the perf process itself:
Average synthesis took: 102.810 usec (+- 0.027 usec)
Average num. events: 17.000 (+- 0.000)
Average time per event 6.048 usec
Average data synthesis took: 106.325 usec (+- 0.018 usec)
Average num. events: 89.000 (+- 0.000)
Average time per event 1.195 usec
Computing performance of multi threaded perf event synthesis by
synthesizing events on CPU 0:
Number of synthesis threads: 16
Average synthesis took: 68103.100 usec (+- 441.234 usec)
Average num. events: 30703.000 (+- 0.730)
Average time per event 2.218 usec
And after is:
$ sudo perf bench internals synthesize -m 16 -M 16 -s -t
\# Running 'internals/synthesize' benchmark:
Computing performance of single threaded perf event synthesis by
synthesizing events on the perf process itself:
Average synthesis took: 50.388 usec (+- 0.031 usec)
Average num. events: 17.000 (+- 0.000)
Average time per event 2.964 usec
Average data synthesis took: 52.693 usec (+- 0.020 usec)
Average num. events: 89.000 (+- 0.000)
Average time per event 0.592 usec
Computing performance of multi threaded perf event synthesis by
synthesizing events on CPU 0:
Number of synthesis threads: 16
Average synthesis took: 45022.400 usec (+- 552.740 usec)
Average num. events: 30624.200 (+- 10.037)
Average time per event 1.470 usec
On a Intel Xeon 6154 compiling with Debian gcc 9.2.1.
Committer testing:
On a AMD Ryzen 5 3600X 6-Core Processor:
Before:
# perf bench internals synthesize --min-threads 12 --max-threads 12 --st --mt
# Running 'internals/synthesize' benchmark:
Computing performance of single threaded perf event synthesis by
synthesizing events on the perf process itself:
Average synthesis took: 267.491 usec (+- 0.176 usec)
Average num. events: 56.000 (+- 0.000)
Average time per event 4.777 usec
Average data synthesis took: 277.257 usec (+- 0.169 usec)
Average num. events: 287.000 (+- 0.000)
Average time per event 0.966 usec
Computing performance of multi threaded perf event synthesis by
synthesizing events on CPU 0:
Number of synthesis threads: 12
Average synthesis took: 81599.500 usec (+- 346.315 usec)
Average num. events: 36096.100 (+- 2.523)
Average time per event 2.261 usec
#
After:
# perf bench internals synthesize --min-threads 12 --max-threads 12 --st --mt
# Running 'internals/synthesize' benchmark:
Computing performance of single threaded perf event synthesis by
synthesizing events on the perf process itself:
Average synthesis took: 110.125 usec (+- 0.080 usec)
Average num. events: 56.000 (+- 0.000)
Average time per event 1.967 usec
Average data synthesis took: 118.518 usec (+- 0.057 usec)
Average num. events: 287.000 (+- 0.000)
Average time per event 0.413 usec
Computing performance of multi threaded perf event synthesis by
synthesizing events on CPU 0:
Number of synthesis threads: 12
Average synthesis took: 43490.700 usec (+- 284.527 usec)
Average num. events: 37028.500 (+- 0.563)
Average time per event 1.175 usec
#
Signed-off-by: Ian Rogers <irogers@google.com>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Acked-by: Jiri Olsa <jolsa@redhat.com>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Andrey Zhizhikin <andrey.z@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Kan Liang <kan.liang@linux.intel.com>
Cc: Kefeng Wang <wangkefeng.wang@huawei.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Petr Mladek <pmladek@suse.com>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lore.kernel.org/lkml/20200415054050.31645-4-irogers@google.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
By default this isn't run as it reads /proc and may not have access.
For consistency, modify the single threaded benchmark to compute an
average time per event.
Committer testing:
$ grep -m1 "model name" /proc/cpuinfo
model name : Intel(R) Core(TM) i7-8650U CPU @ 1.90GHz
$ grep "model name" /proc/cpuinfo | wc -l
8
$
$ perf bench internals synthesize -h
# Running 'internals/synthesize' benchmark:
Usage: perf bench internals synthesize <options>
-I, --multi-iterations <n>
Number of iterations used to compute multi-threaded average
-i, --single-iterations <n>
Number of iterations used to compute single-threaded average
-M, --max-threads <n>
Maximum number of threads in multithreaded bench
-m, --min-threads <n>
Minimum number of threads in multithreaded bench
-s, --st Run single threaded benchmark
-t, --mt Run multi-threaded benchmark
$
$ perf bench internals synthesize -t
# Running 'internals/synthesize' benchmark:
Computing performance of multi threaded perf event synthesis by
synthesizing events on CPU 0:
Number of synthesis threads: 1
Average synthesis took: 65449.000 usec (+- 586.442 usec)
Average num. events: 9405.400 (+- 0.306)
Average time per event 6.959 usec
Number of synthesis threads: 2
Average synthesis took: 37838.300 usec (+- 130.259 usec)
Average num. events: 9501.800 (+- 20.469)
Average time per event 3.982 usec
Number of synthesis threads: 3
Average synthesis took: 48551.400 usec (+- 225.686 usec)
Average num. events: 9544.000 (+- 0.000)
Average time per event 5.087 usec
Number of synthesis threads: 4
Average synthesis took: 29632.500 usec (+- 50.808 usec)
Average num. events: 9544.000 (+- 0.000)
Average time per event 3.105 usec
Number of synthesis threads: 5
Average synthesis took: 33920.400 usec (+- 284.509 usec)
Average num. events: 9544.000 (+- 0.000)
Average time per event 3.554 usec
Number of synthesis threads: 6
Average synthesis took: 27604.100 usec (+- 72.344 usec)
Average num. events: 9548.000 (+- 0.000)
Average time per event 2.891 usec
Number of synthesis threads: 7
Average synthesis took: 25406.300 usec (+- 933.371 usec)
Average num. events: 9545.500 (+- 0.167)
Average time per event 2.662 usec
Number of synthesis threads: 8
Average synthesis took: 24110.400 usec (+- 73.229 usec)
Average num. events: 9551.000 (+- 0.000)
Average time per event 2.524 usec
$
Signed-off-by: Ian Rogers <irogers@google.com>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Acked-by: Jiri Olsa <jolsa@redhat.com>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Andrey Zhizhikin <andrey.z@gmail.com>
Cc: Kan Liang <kan.liang@linux.intel.com>
Cc: Kefeng Wang <wangkefeng.wang@huawei.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Petr Mladek <pmladek@suse.com>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lore.kernel.org/lkml/20200415054050.31645-2-irogers@google.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
There is demand for a way to identify what BPF helper functions are
available to unprivileged users. To do so, allow unprivileged users to
run "bpftool feature probe" to list BPF-related features. This will only
show features accessible to those users, and may not reflect the full
list of features available (to administrators) on the system.
To avoid the case where bpftool is inadvertently run as non-root and
would list only a subset of the features supported by the system when it
would be expected to list all of them, running as unprivileged is gated
behind the "unprivileged" keyword passed to the command line. When used
by a privileged user, this keyword allows to drop the CAP_SYS_ADMIN and
to list the features available to unprivileged users. Note that this
addsd a dependency on libpcap for compiling bpftool.
Note that there is no particular reason why the probes were restricted
to root, other than the fact I did not need them for unprivileged and
did not bother with the additional checks at the time probes were added.
Signed-off-by: Quentin Monnet <quentin@isovalent.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Link: https://lore.kernel.org/bpf/20200429144506.8999-3-quentin@isovalent.com
The "full_mode" variable used for switching between full or partial
feature probing (i.e. with or without probing helpers that will log
warnings in kernel logs) was piped from the main do_probe() function
down to probe_helpers_for_progtype(), where it is needed.
Define it as a global variable: the calls will be more readable, and if
other similar flags were to be used in the future, we could use global
variables as well instead of extending again the list of arguments with
new flags.
Signed-off-by: Quentin Monnet <quentin@isovalent.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Link: https://lore.kernel.org/bpf/20200429144506.8999-2-quentin@isovalent.com
AddressSanitizer assumes that all memory dereferences are done against memory
allocated by sanitizer's malloc()/free() code and not touched by anyone else.
Seems like this doesn't hold for perf buffer memory. Disable instrumentation
on perf buffer callback function.
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20200429012111.277390-10-andriin@fb.com
Fix memory leak in hashmap_clear() not freeing hashmap_entry structs for each
of the remaining entries. Also NULL-out bucket list to prevent possible
double-free between hashmap__clear() and hashmap__free().
Running test_progs-asan flavor clearly showed this problem.
Reported-by: Alston Tang <alston64@fb.com>
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20200429012111.277390-5-andriin@fb.com
Add ability to specify extra compiler flags with SAN_CFLAGS for compilation of
all user-space C files. This allows to build all of selftest programs with,
e.g., custom sanitizer flags, without requiring support for such sanitizers
from anyone compiling selftest/bpf.
As an example, to compile everything with AddressSanitizer, one would do:
$ make clean && make SAN_CFLAGS="-fsanitize=address"
For AddressSanitizer to work, one needs appropriate libasan shared library
installed in the system, with version of libasan matching what GCC links
against. E.g., GCC8 needs libasan5, while GCC7 uses libasan4.
For CentOS 7, to build everything successfully one would need to:
$ sudo yum install devtoolset-8-gcc devtoolset-libasan-devel
$ scl enable devtoolset-8 bash # set up environment
For Arch Linux to run selftests, one would need to install gcc-libs package to
get libasan.so.5:
$ sudo pacman -S gcc-libs
N.B. EXTRA_CFLAGS name wasn't used, because it's also used by libbpf's
Makefile and this causes few issues:
1. default "-g -Wall" flags are overriden;
2. compiling shared library with AddressSanitizer generates a bunch of symbols
like: "_GLOBAL__sub_D_00099_0_btf_dump.c", "_GLOBAL__sub_D_00099_0_bpf.c",
etc, which screws up versioned symbols check.
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Cc: Julia Kartseva <hex@fb.com>
Link: https://lore.kernel.org/bpf/20200429012111.277390-3-andriin@fb.com
Ensure that test runner flavors include their own skeletons from <flavor>/
directory. Previously, skeletons generated for no-flavor test_progs were used.
Apart from fixing correctness, this also makes it possible to compile only
flavors individually:
$ make clean && make test_progs-no_alu32
... now succeeds ...
Fixes: 74b5a5968f ("selftests/bpf: Replace test_progs and test_maps w/ general rule")
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20200429012111.277390-2-andriin@fb.com
As discussed at LPC 2019 ([0]), this patch brings (a quite belated) support
for declarative BTF-defined map-in-map support in libbpf. It allows to define
ARRAY_OF_MAPS and HASH_OF_MAPS BPF maps without any user-space initialization
code involved.
Additionally, it allows to initialize outer map's slots with references to
respective inner maps at load time, also completely declaratively.
Despite a weak type system of C, the way BTF-defined map-in-map definition
works, it's actually quite hard to accidentally initialize outer map with
incompatible inner maps. This being C, of course, it's still possible, but
even that would be caught at load time and error returned with helpful debug
log pointing exactly to the slot that failed to be initialized.
As an example, here's a rather advanced HASH_OF_MAPS declaration and
initialization example, filling slots #0 and #4 with two inner maps:
#include <bpf/bpf_helpers.h>
struct inner_map {
__uint(type, BPF_MAP_TYPE_ARRAY);
__uint(max_entries, 1);
__type(key, int);
__type(value, int);
} inner_map1 SEC(".maps"),
inner_map2 SEC(".maps");
struct outer_hash {
__uint(type, BPF_MAP_TYPE_HASH_OF_MAPS);
__uint(max_entries, 5);
__uint(key_size, sizeof(int));
__array(values, struct inner_map);
} outer_hash SEC(".maps") = {
.values = {
[0] = &inner_map2,
[4] = &inner_map1,
},
};
Here's the relevant part of libbpf debug log showing pretty clearly of what's
going on with map-in-map initialization:
libbpf: .maps relo #0: for 6 value 0 rel.r_offset 96 name 260 ('inner_map1')
libbpf: .maps relo #0: map 'outer_arr' slot [0] points to map 'inner_map1'
libbpf: .maps relo #1: for 7 value 32 rel.r_offset 112 name 249 ('inner_map2')
libbpf: .maps relo #1: map 'outer_arr' slot [2] points to map 'inner_map2'
libbpf: .maps relo #2: for 7 value 32 rel.r_offset 144 name 249 ('inner_map2')
libbpf: .maps relo #2: map 'outer_hash' slot [0] points to map 'inner_map2'
libbpf: .maps relo #3: for 6 value 0 rel.r_offset 176 name 260 ('inner_map1')
libbpf: .maps relo #3: map 'outer_hash' slot [4] points to map 'inner_map1'
libbpf: map 'inner_map1': created successfully, fd=4
libbpf: map 'inner_map2': created successfully, fd=5
libbpf: map 'outer_hash': created successfully, fd=7
libbpf: map 'outer_hash': slot [0] set to map 'inner_map2' fd=5
libbpf: map 'outer_hash': slot [4] set to map 'inner_map1' fd=4
Notice from the log above that fd=6 (not logged explicitly) is used for inner
"prototype" map, necessary for creation of outer map. It is destroyed
immediately after outer map is created.
See also included selftest with some extra comments explaining extra details
of usage. Additionally, similar initialization syntax and libbpf functionality
can be used to do initialization of BPF_PROG_ARRAY with references to BPF
sub-programs. This can be done in follow up patches, if there will be a demand
for this.
[0] https://linuxplumbersconf.org/event/4/contributions/448/
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Toke Høiland-Jørgensen <toke@redhat.com>
Link: https://lore.kernel.org/bpf/20200429002739.48006-4-andriin@fb.com
Factor out map creation and destruction logic to simplify code and especially
error handling. Also fix map FD leak in case of partially successful map
creation during bpf_object load operation.
Fixes: 57a00f4164 ("libbpf: Add auto-pinning of maps when loading BPF objects")
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Toke Høiland-Jørgensen <toke@redhat.com>
Link: https://lore.kernel.org/bpf/20200429002739.48006-3-andriin@fb.com
Add ability to fetch bpf_link details through BPF_OBJ_GET_INFO_BY_FD command.
Also enhance show_fdinfo to potentially include bpf_link type-specific
information (similarly to obj_info).
Also introduce enum bpf_link_type stored in bpf_link itself and expose it in
UAPI. bpf_link_tracing also now will store and return bpf_attach_type.
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20200429001614.1544-5-andriin@fb.com
Similar to commit b7a0d65d80 ("bpf, testing: Workaround a verifier failure for test_progs")
fix test_sysctl_prog.c as well.
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
- add SPDX header;
- adjust title markup;
- mark code blocks and literals as such;
- use footnote markup;
- mark tables as such;
- adjust identation, whitespaces and blank lines;
- add to networking/index.rst.
Signed-off-by: Mauro Carvalho Chehab <mchehab+huawei@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
TLS 1.2 and TLS 1.3 differ in the implementation.
Use fixture parameters to run all tests for both
versions, and remove the one-off TLS 1.2 test.
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: Kees Cook <keescook@chromium.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Allow users to build parameterized variants of fixtures.
If fixtures want variants, they call FIXTURE_VARIANT() to declare
the structure to fill for each variant. Each fixture will be re-run
for each of the variants defined by calling FIXTURE_VARIANT_ADD()
with the differing parameters initializing the structure.
Since tests are being re-run, additional initialization (steps,
no_print) is also added.
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Acked-by: Kees Cook <keescook@chromium.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Now that all tests have a fixture object move from a global
list of tests to a list of tests per fixture.
Order of tests may change as we will now group and run test
fixture by fixture, rather than in declaration order.
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Acked-by: Kees Cook <keescook@chromium.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Grouping tests by fixture will allow us to parametrize
test runs. Create full objects for fixtures.
Add a "global" fixture for tests without a fixture.
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Acked-by: Kees Cook <keescook@chromium.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Kees suggest to factor out the list append code to a macro,
since following commits need it, which leads to code duplication.
Suggested-by: Kees Cook <keescook@chromium.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Acked-by: Kees Cook <keescook@chromium.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
New tests to check route dump and notifications with
net.ipv4.nexthop_compat_mode on and off.
Signed-off-by: Roopa Prabhu <roopa@cumulusnetworks.com>
Reviewed-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
$(OUTPUT)/runqslower makefile target doesn't actually create runqslower
binary in the $(OUTPUT) directory. As lib.mk expects all
TEST_GEN_PROGS_EXTENDED (which runqslower is a part of) to be present in
the OUTPUT directory, this results in an error when running e.g. `make
install`:
rsync: link_stat "tools/testing/selftests/bpf/runqslower" failed: No
such file or directory (2)
Copy the binary into the OUTPUT directory after building it to fix the
error.
Fixes: 3a0d3092a4 ("selftests/bpf: Build runqslower from selftests")
Signed-off-by: Veronika Kabatova <vkabatov@redhat.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Link: https://lore.kernel.org/bpf/20200428173742.2988395-1-vkabatov@redhat.com
This patch adds a new port attribute, IFLA_BRPORT_MRP_RING_OPEN, which allows
to notify the userspace when the port lost the continuite of MRP frames.
This attribute is set by kernel whenever the SW or HW detects that the ring is
being open or closed.
Reviewed-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: Horatiu Vultur <horatiu.vultur@microchip.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
System hangs or killed rcutorture guest OSes can result in truncated
"Reader Pipe:" lines, which can in turn result in false-positive
reader-batch near-miss warnings. This commit therefore adjusts the
reader-batch checks to account for possible line truncation.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
This commit adds a TRACE02 scenario which enables preemption and RCU
Tasks Trace IPIs, more specifically, disabling heavyweight readers.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
cls_redirect is a TC clsact based replacement for the glb-redirect iptables
module available at [1]. It enables what GitHub calls "second chance"
flows [2], similarly proposed by the Beamer paper [3]. In contrast to
glb-redirect, it also supports migrating UDP flows as long as connected
sockets are used. cls_redirect is in production at Cloudflare, as part of
our own L4 load balancer.
We have modified the encapsulation format slightly from glb-redirect:
glbgue_chained_routing.private_data_type has been repurposed to form a
version field and several flags. Both have been arranged in a way that
a private_data_type value of zero matches the current glb-redirect
behaviour. This means that cls_redirect will understand packets in
glb-redirect format, but not vice versa.
The test suite only covers basic features. For example, cls_redirect will
correctly forward path MTU discovery packets, but this is not exercised.
It is also possible to switch the encapsulation format to GRE on the last
hop, which is also not tested.
There are two major distinctions from glb-redirect: first, cls_redirect
relies on receiving encapsulated packets directly from a router. This is
because we don't have access to the neighbour tables from BPF, yet. See
forward_to_next_hop for details. Second, cls_redirect performs decapsulation
instead of using separate ipip and sit tunnel devices. This
avoids issues with the sit tunnel [4] and makes deploying the classifier
easier: decapsulated packets appear on the same interface, so existing
firewall rules continue to work as expected.
The code base started it's life on v4.19, so there are most likely still
hold overs from old workarounds. In no particular order:
- The function buf_off is required to defeat a clang optimization
that leads to the verifier rejecting the program due to pointer
arithmetic in the wrong order.
- The function pkt_parse_ipv6 is force inlined, because it would
otherwise be rejected due to returning a pointer to stack memory.
- The functions fill_tuple and classify_tcp contain kludges, because
we've run out of function arguments.
- The logic in general is rather nested, due to verifier restrictions.
I think this is either because the verifier loses track of constants
on the stack, or because it can't track enum like variables.
1: https://github.com/github/glb-director/tree/master/src/glb-redirect
2: https://github.com/github/glb-director/blob/master/docs/development/second-chance-design.md
3: https://www.usenix.org/conference/nsdi18/presentation/olteanu
4: https://github.com/github/glb-director/issues/64
Signed-off-by: Lorenz Bauer <lmb@cloudflare.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20200424185556.7358-2-lmb@cloudflare.com
To make BPF verifier verbose log more releavant and easier to use to debug
verification failures, "pop" parts of log that were successfully verified.
This has effect of leaving only verifier logs that correspond to code branches
that lead to verification failure, which in practice should result in much
shorter and more relevant verifier log dumps. This behavior is made the
default behavior and can be overriden to do exhaustive logging by specifying
BPF_LOG_LEVEL2 log level.
Using BPF_LOG_LEVEL2 to disable this behavior is not ideal, because in some
cases it's good to have BPF_LOG_LEVEL2 per-instruction register dump
verbosity, but still have only relevant verifier branches logged. But for this
patch, I didn't want to add any new flags. It might be worth-while to just
rethink how BPF verifier logging is performed and requested and streamline it
a bit. But this trimming of successfully verified branches seems to be useful
and a good default behavior.
To test this, I modified runqslower slightly to introduce read of
uninitialized stack variable. Log (**truncated in the middle** to save many
lines out of this commit message) BEFORE this change:
; int handle__sched_switch(u64 *ctx)
0: (bf) r6 = r1
; struct task_struct *prev = (struct task_struct *)ctx[1];
1: (79) r1 = *(u64 *)(r6 +8)
func 'sched_switch' arg1 has btf_id 151 type STRUCT 'task_struct'
2: (b7) r2 = 0
; struct event event = {};
3: (7b) *(u64 *)(r10 -24) = r2
last_idx 3 first_idx 0
regs=4 stack=0 before 2: (b7) r2 = 0
4: (7b) *(u64 *)(r10 -32) = r2
5: (7b) *(u64 *)(r10 -40) = r2
6: (7b) *(u64 *)(r10 -48) = r2
; if (prev->state == TASK_RUNNING)
[ ... instruction dump from insn #7 through #50 are cut out ... ]
51: (b7) r2 = 16
52: (85) call bpf_get_current_comm#16
last_idx 52 first_idx 42
regs=4 stack=0 before 51: (b7) r2 = 16
; bpf_perf_event_output(ctx, &events, BPF_F_CURRENT_CPU,
53: (bf) r1 = r6
54: (18) r2 = 0xffff8881f3868800
56: (18) r3 = 0xffffffff
58: (bf) r4 = r7
59: (b7) r5 = 32
60: (85) call bpf_perf_event_output#25
last_idx 60 first_idx 53
regs=20 stack=0 before 59: (b7) r5 = 32
61: (bf) r2 = r10
; event.pid = pid;
62: (07) r2 += -16
; bpf_map_delete_elem(&start, &pid);
63: (18) r1 = 0xffff8881f3868000
65: (85) call bpf_map_delete_elem#3
; }
66: (b7) r0 = 0
67: (95) exit
from 44 to 66: safe
from 34 to 66: safe
from 11 to 28: R1_w=inv0 R2_w=inv0 R6_w=ctx(id=0,off=0,imm=0) R10=fp0 fp-8=mmmm???? fp-24_w=00000000 fp-32_w=00000000 fp-40_w=00000000 fp-48_w=00000000
; bpf_map_update_elem(&start, &pid, &ts, 0);
28: (bf) r2 = r10
;
29: (07) r2 += -16
; tsp = bpf_map_lookup_elem(&start, &pid);
30: (18) r1 = 0xffff8881f3868000
32: (85) call bpf_map_lookup_elem#1
invalid indirect read from stack off -16+0 size 4
processed 65 insns (limit 1000000) max_states_per_insn 1 total_states 5 peak_states 5 mark_read 4
Notice how there is a successful code path from instruction 0 through 67, few
successfully verified jumps (44->66, 34->66), and only after that 11->28 jump
plus error on instruction #32.
AFTER this change (full verifier log, **no truncation**):
; int handle__sched_switch(u64 *ctx)
0: (bf) r6 = r1
; struct task_struct *prev = (struct task_struct *)ctx[1];
1: (79) r1 = *(u64 *)(r6 +8)
func 'sched_switch' arg1 has btf_id 151 type STRUCT 'task_struct'
2: (b7) r2 = 0
; struct event event = {};
3: (7b) *(u64 *)(r10 -24) = r2
last_idx 3 first_idx 0
regs=4 stack=0 before 2: (b7) r2 = 0
4: (7b) *(u64 *)(r10 -32) = r2
5: (7b) *(u64 *)(r10 -40) = r2
6: (7b) *(u64 *)(r10 -48) = r2
; if (prev->state == TASK_RUNNING)
7: (79) r2 = *(u64 *)(r1 +16)
; if (prev->state == TASK_RUNNING)
8: (55) if r2 != 0x0 goto pc+19
R1_w=ptr_task_struct(id=0,off=0,imm=0) R2_w=inv0 R6_w=ctx(id=0,off=0,imm=0) R10=fp0 fp-24_w=00000000 fp-32_w=00000000 fp-40_w=00000000 fp-48_w=00000000
; trace_enqueue(prev->tgid, prev->pid);
9: (61) r1 = *(u32 *)(r1 +1184)
10: (63) *(u32 *)(r10 -4) = r1
; if (!pid || (targ_pid && targ_pid != pid))
11: (15) if r1 == 0x0 goto pc+16
from 11 to 28: R1_w=inv0 R2_w=inv0 R6_w=ctx(id=0,off=0,imm=0) R10=fp0 fp-8=mmmm???? fp-24_w=00000000 fp-32_w=00000000 fp-40_w=00000000 fp-48_w=00000000
; bpf_map_update_elem(&start, &pid, &ts, 0);
28: (bf) r2 = r10
;
29: (07) r2 += -16
; tsp = bpf_map_lookup_elem(&start, &pid);
30: (18) r1 = 0xffff8881db3ce800
32: (85) call bpf_map_lookup_elem#1
invalid indirect read from stack off -16+0 size 4
processed 65 insns (limit 1000000) max_states_per_insn 1 total_states 5 peak_states 5 mark_read 4
Notice how in this case, there are 0-11 instructions + jump from 11 to
28 is recorded + 28-32 instructions with error on insn #32.
test_verifier test runner was updated to specify BPF_LOG_LEVEL2 for
VERBOSE_ACCEPT expected result due to potentially "incomplete" success verbose
log at BPF_LOG_LEVEL1.
On success, verbose log will only have a summary of number of processed
instructions, etc, but no branch tracing log. Having just a last succesful
branch tracing seemed weird and confusing. Having small and clean summary log
in success case seems quite logical and nice, though.
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20200423195850.1259827-1-andriin@fb.com
