This will all come from userspace, but to test the changes to make 'perf
trace' output similar to strace's, do this one more now manually.
To update the precompiled augmented_raw_syscalls.o binary I just run:
# perf record -e ~acme/git/perf/tools/perf/examples/bpf/augmented_raw_syscalls.c sleep 1
LLVM: dumping /home/acme/git/perf/tools/perf/examples/bpf/augmented_raw_syscalls.o
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.022 MB perf.data ]
#
Because to have augmented_raw_syscalls to be always used and a fast
startup and remove the need to have the llvm toolchain installed, I'm
using:
# perf config | grep add_events
trace.add_events=/home/acme/git/perf/tools/perf/examples/bpf/augmented_raw_syscalls.o
#
So when doing changes to augmented_raw_syscals.c one needs to rebuild
the .o file.
This will be done automagically later, i.e. have a 'make' behaviour of
recompiling when the .c gets changed.
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Luis Cláudio Gonçalves <lclaudio@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: https://lkml.kernel.org/n/tip-lw3i2atyq8549fpqwmszn3qp@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
We're not using that puts() thing, and thus we don't need to define the
__bpf_stdout__ map, reducing the setup time.
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Luis Cláudio Gonçalves <lclaudio@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: https://lkml.kernel.org/n/tip-3452xgatncpil7v22minkwbo@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
We'll start adding more perf-syscall stuff, so lets do this prep step so
that the next ones are just about adding more fields.
Run it with the .c file once to cache the .o file:
# trace --filter-pids 2834,2199 -e openat,augmented_raw_syscalls.c
LLVM: dumping augmented_raw_syscalls.o
0.000 ( 0.021 ms): tmux: server/4952 openat(dfd: CWD, filename: /proc/5691/cmdline ) = 11
349.807 ( 0.040 ms): DNS Res~er #39/11082 openat(dfd: CWD, filename: /etc/hosts, flags: CLOEXEC ) = 44
4988.759 ( 0.052 ms): gsd-color/2431 openat(dfd: CWD, filename: /etc/localtime ) = 18
4988.976 ( 0.029 ms): gsd-color/2431 openat(dfd: CWD, filename: /etc/localtime ) = 18
^C[root@quaco bpf]#
From now on, we can use just the newly built .o file, skipping the
compilation step for a faster startup:
# trace --filter-pids 2834,2199 -e openat,augmented_raw_syscalls.o
0.000 ( 0.046 ms): DNS Res~er #39/11088 openat(dfd: CWD, filename: /etc/hosts, flags: CLOEXEC ) = 44
1946.408 ( 0.190 ms): systemd/1 openat(dfd: CWD, filename: /proc/1071/cgroup, flags: CLOEXEC ) = 20
1946.792 ( 0.215 ms): systemd/1 openat(dfd: CWD, filename: /proc/954/cgroup, flags: CLOEXEC ) = 20
^C#
Now on to do the same in the builtin-trace.c side of things.
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Link: https://lkml.kernel.org/n/tip-k8mwu04l8es29rje5loq9vg7@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Just another map, this time an BPF_MAP_TYPE_ARRAY, stating with
one bool per syscall, stating if it should be filtered or not.
So, with a pre-built augmented_raw_syscalls.o file, we use:
# perf trace -e open*,augmented_raw_syscalls.o
0.000 ( 0.016 ms): DNS Res~er #37/29652 openat(dfd: CWD, filename: /etc/hosts, flags: CLOEXEC ) = 138
187.039 ( 0.048 ms): gsd-housekeepi/2436 openat(dfd: CWD, filename: /etc/fstab, flags: CLOEXEC ) = 11
187.348 ( 0.041 ms): gsd-housekeepi/2436 openat(dfd: CWD, filename: /proc/self/mountinfo, flags: CLOEXEC ) = 11
188.793 ( 0.036 ms): gsd-housekeepi/2436 openat(dfd: CWD, filename: /proc/self/mountinfo, flags: CLOEXEC ) = 11
189.803 ( 0.029 ms): gsd-housekeepi/2436 openat(dfd: CWD, filename: /proc/self/mountinfo, flags: CLOEXEC ) = 11
190.774 ( 0.027 ms): gsd-housekeepi/2436 openat(dfd: CWD, filename: /proc/self/mountinfo, flags: CLOEXEC ) = 11
284.620 ( 0.149 ms): DataStorage/3076 openat(dfd: CWD, filename: /home/acme/.mozilla/firefox/ina67tev.default/SiteSecurityServiceState.txt, flags: CREAT|TRUNC|WRONLY, mode: IRUGO|IWUSR|IWGRP) = 167
^C#
What is it that this gsd-housekeeping thingy needs to open
/proc/self/mountinfo four times periodically? :-)
This map will be extended to tell per-syscall parameters, i.e. how many
bytes to copy per arg, using the function signature to get the types and
then the size of those types, via BTF.
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: https://lkml.kernel.org/n/tip-cy222g9ucvnym3raqvxp0hpg@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Now that we have the "filtered_pids" logic in place, no need to do this
rough filter to avoid the feedback loop from 'perf trace's own syscalls,
revert it.
This reverts commit 7ed71f124284359676b6496ae7db724fee9da753.
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: https://lkml.kernel.org/n/tip-88vh02cnkam0vv5f9vp02o3h@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Just to test filtering a bunch of pids, now its time to go and get that
hooked up in 'perf trace', right after we load the bpf program, if we
find a "pids_filtered" map defined, we'll populate it with the filtered
pids.
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: https://lkml.kernel.org/n/tip-1i9s27wqqdhafk3fappow84x@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
When testing system wide tracing without filtering the syscalls called
by 'perf trace' itself we get into a feedback loop, drop for now those
two syscalls, that are the ones that 'perf trace' does in its loop for
writing the syscalls it intercepts, to help with testing till we get
that filtering in place.
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: https://lkml.kernel.org/n/tip-rkbu536af66dbsfx51sr8yof@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Leftover from when we started augmented_raw_syscalls.c from
tools/perf/examples/bpf/augmented_syscalls.
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Fixes: e58a0322dbac ("perf examples bpf: Start augmenting raw_syscalls:sys_{start,exit}")
Link: https://lkml.kernel.org/n/tip-pmts9ls2skh8n3zisb4txudd@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Just to show where we'll hook pid based filters, and what we use to
obtain the current pid, using a BPF getpid() equivalent.
Now we need to remove that hardcoded PID with a BPF hash map, so that we
start by filtering 'perf trace's own PID, implement the --filter-pid
functionality, etc.
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: https://lkml.kernel.org/n/tip-oshrcgcekiyhd0whwisxfvtv@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
This is the start of having the raw_syscalls:sys_enter BPF handler
collecting pointer arguments, namely pathnames, and with two syscalls
that have that pointer in different arguments, "open" as it as its first
argument, "openat" as the second.
With this in place the existing beautifiers in 'perf trace' works, those
args are shown instead of just the pointer that comes with the syscalls
tracepoints.
This also serves to show and document pitfalls in the process of using
just that place in the kernel (raw_syscalls:sys_enter) plus tables
provided by userspace to collect syscall pointer arguments.
One is the need to use a barrier, as suggested by Edward, to avoid clang
optimizations that make the kernel BPF verifier to refuse loading our
pointer contents collector.
The end result should be a generic eBPF program that works in all
architectures, with the differences amongst archs resolved by the
userspace component, 'perf trace', that should get all its tables
created automatically from the kernel components where they are defined,
via string table constructors for things not expressed in BTF/DWARF
(enums, structs, etc), and otherwise using those observability files
(BTF).
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alexei Starovoitov <ast@kernel.org>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: David Ahern <dsahern@gmail.com>
Cc: Edward Cree <ecree@solarflare.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Martin KaFai Lau <kafai@fb.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Cc: Yonghong Song <yhs@fb.com>
Link: https://lkml.kernel.org/n/tip-37dz54pmotgpnwg9tb6zuk9j@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Using a value returned from probe_read_str() to tell how many bytes to
copy using perf_event_output() has issues in some older kernels, like
4.17.17-100.fc27.x86_64, so separate the bounds checking done on how
many bytes to copy to a separate variable, so that the next patch has
only what is being done to make the test pass on older BPF validators.
For reference, see the discussion in this thread:
https://www.spinics.net/lists/netdev/msg480099.html
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Cc: Yonghong Song <yhs@fb.com>
Link: https://lkml.kernel.org/n/tip-jtsapwibyxrnv1xjfsgzp0fj@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Hook the pair enter/exit when using augmented_{filename,sockaddr,etc}_syscall(),
this way we'll be able to see what entries are in the ELF sections generated
from augmented_syscalls.c and filter them out from the main raw_syscalls:*
tracepoints used by 'perf trace'.
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: https://lkml.kernel.org/n/tip-cyav42qj5yylolw4attcw99z@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
As we'll also hook into the syscalls:sys_exit_SYSCALL for which there
are enter hooks.
This way we'll be able to iterate the ELF file for the eBPF program,
find the syscalls that have hooks and filter them out from the general
raw_syscalls:sys_{enter,exit} tracepoint for not-yet-augmented (the ones
with pointer arguments not yet being attached to the usual syscalls
tracepoint payload) and non augmentable syscalls (syscalls without
pointer arguments).
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: https://lkml.kernel.org/n/tip-cl1xyghwb1usp500354mv37h@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Reflecting the fact that it now augments more than syscalls:sys_enter_SYSCALL
tracepoints that have filename strings as args. Also mention how the
extra data is handled by the by now modified 'perf trace' beautifiers,
that will use special "augmented" beautifiers when extra data is found
after the expected syscall enter/exit tracepoints.
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: https://lkml.kernel.org/n/tip-ybskanehmdilj5fs7080nz1g@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Its a 'struct sockaddr' pointer, augment it with the same beautifier as
for 'connect' and 'bind', that all receive from userspace that pointer.
Doing it in the other direction remains to be done, hooking at the
syscalls:sys_exit_{accept4?,recvmsg} tracepoints somehow.
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: https://lkml.kernel.org/n/tip-k2eu68lsphnm2fthc32gq76c@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
One more, to reuse the augmented_sockaddr_syscall_enter() macro
introduced from the augmentation of connect's sockaddr arg, also to get
a subset of the struct arg augmentations done using the manual method,
before switching to something automatic, using tracefs's format file or,
even better, BTF containing the syscall args structs.
# perf trace -e tools/perf/examples/bpf/augmented_syscalls.c
0.000 sshd/11479 bind(fd: 3<socket:[170336]>, umyaddr: { .family: NETLINK }, addrlen: 12)
1.752 sshd/11479 bind(fd: 3<socket:[170336]>, umyaddr: { .family: INET, port: 22, addr: 0.0.0.0 }, addrlen: 16)
1.924 sshd/11479 bind(fd: 4<socket:[170338]>, umyaddr: { .family: INET6, port: 22, addr: :: }, addrlen: 28)
^C#
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: https://lkml.kernel.org/n/tip-a2drqpahpmc7uwb3n3gj2plu@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
From the one for 'connect', so that we can use it with sendto and others
that receive a 'struct sockaddr'.
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: https://lkml.kernel.org/n/tip-8bdqv1q0ndcjl1nqns5r5je2@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Helping with tons of boilerplate for syscalls that only want to augment
a filename. Now supporting one such syscall is just a matter of
declaring its arguments struct + using:
augmented_filename_syscall_enter(openat);
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: https://lkml.kernel.org/n/tip-ls7ojdseu8fxw7fvj77ejpao@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
This is the final touch in showing how a syscall argument beautifier can
access the augmented args put in place by the
tools/perf/examples/bpf/augmented_syscalls.c eBPF script, right after
the regular raw syscall args, i.e. the up to 6 long integer values in
the syscall interface.
With this we are able to show the 'openat' syscall arg, now with up to
64 bytes, but in time this will be configurable, just like with the
'strace -s strsize' argument, from 'strace''s man page:
-s strsize Specify the maximum string size to print (the default is 32).
This actually is the maximum string to _collect_ and store in the ring
buffer, not just print.
Before:
# perf trace -e tools/perf/examples/bpf/augmented_syscalls.c,openat cat /etc/passwd > /dev/null
0.000 ( ): cat/9658 openat(dfd: CWD, filename: 0x6626eda8, flags: CLOEXEC)
0.017 ( 0.007 ms): cat/9658 openat(dfd: CWD, filename: 0x6626eda8, flags: CLOEXEC) = 3
0.049 ( ): cat/9658 openat(dfd: CWD, filename: 0x66476ce0, flags: CLOEXEC)
0.051 ( 0.007 ms): cat/9658 openat(dfd: CWD, filename: 0x66476ce0, flags: CLOEXEC) = 3
0.377 ( ): cat/9658 openat(dfd: CWD, filename: 0x1e8f806b)
0.379 ( 0.005 ms): cat/9658 openat(dfd: CWD, filename: 0x1e8f806b) = 3
#
After:
# perf trace -e tools/perf/examples/bpf/augmented_syscalls.c,openat cat /etc/passwd > /dev/null
0.000 ( ): cat/11966 openat(dfd: CWD, filename: /etc/ld.so.cache, flags: CLOEXEC)
0.006 ( 0.006 ms): cat/11966 openat(dfd: CWD, filename: 0x4bfdcda8, flags: CLOEXEC) = 3
0.034 ( ): cat/11966 openat(dfd: CWD, filename: /lib64/libc.so.6, flags: CLOEXEC)
0.036 ( 0.008 ms): cat/11966 openat(dfd: CWD, filename: 0x4c1e4ce0, flags: CLOEXEC) = 3
0.375 ( ): cat/11966 openat(dfd: CWD, filename: /etc/passwd)
0.377 ( 0.005 ms): cat/11966 openat(dfd: CWD, filename: 0xe87906b) = 3
#
This cset should show all the aspects of establishing a protocol between
an eBPF syscall arg augmenter program, tools/perf/examples/bpf/augmented_syscalls.c and
a 'perf trace' beautifier, the one associated with all 'char *' point
syscall args with names that can heuristically be associated with
filenames.
Now to wire up 'open' to show a second syscall using this scheme, all we
have to do now is to change tools/perf/examples/bpf/augmented_syscalls.c,
as 'perf trace' will notice that the perf_sample.raw_size is more than
what is expected for a particular syscall payload as defined by its
tracefs format file and will then use the augmented payload in the
'filename' syscall arg beautifier.
The same protocol will be used for structs such as 'struct sockaddr *',
'struct pollfd', etc, with additions for handling arrays.
This will all be done under the hood when 'perf trace' realizes the
system has the necessary components, and also can be done by providing
a precompiled augmented_syscalls.c eBPF ELF object.
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: https://lkml.kernel.org/n/tip-gj9kqb61wo7m3shtpzercbcr@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
When we attach a eBPF object to a tracepoint, if we return 1, then that
tracepoint will be stored in the perf's ring buffer. In the
augmented_syscalls.c case we want to just attach and _override_ the
tracepoint payload with an augmented, extended one.
In this example, tools/perf/examples/bpf/augmented_syscalls.c, we are
attaching to the 'openat' syscall, and adding, after the
syscalls:sys_enter_openat usual payload as defined by
/sys/kernel/debug/tracing/events/syscalls/sys_enter_openat/format, a
snapshot of its sole pointer arg:
# grep 'field:.*\*' /sys/kernel/debug/tracing/events/syscalls/sys_enter_openat/format
field:const char * filename; offset:24; size:8; signed:0;
#
For now this is not being considered, the next csets will make use of
it, but as this is overriding the syscall tracepoint enter, we don't
want that event appearing on the ring buffer, just our synthesized one.
Before:
# perf trace -e ~acme/git/perf/tools/perf/examples/bpf/augmented_syscalls.c,openat cat /etc/passwd > /dev/null
0.000 ( ): __augmented_syscalls__:dfd: CWD, filename: /etc/ld.so.cache, flags: CLOEXEC
0.006 ( ): syscalls:sys_enter_openat:dfd: CWD, filename: , flags: CLOEXEC
0.007 ( 0.004 ms): cat/24044 openat(dfd: CWD, filename: 0x216dda8, flags: CLOEXEC ) = 3
0.028 ( ): __augmented_syscalls__:dfd: CWD, filename: /lib64/libc.so.6, flags: CLOEXEC
0.030 ( ): syscalls:sys_enter_openat:dfd: CWD, filename: , flags: CLOEXEC
0.031 ( 0.006 ms): cat/24044 openat(dfd: CWD, filename: 0x2375ce0, flags: CLOEXEC ) = 3
0.291 ( ): __augmented_syscalls__:dfd: CWD, filename: /etc/passwd
0.293 ( ): syscalls:sys_enter_openat:dfd: CWD, filename:
0.294 ( 0.004 ms): cat/24044 openat(dfd: CWD, filename: 0x637db06b ) = 3
#
After:
# perf trace -e ~acme/git/perf/tools/perf/examples/bpf/augmented_syscalls.c,openat cat /etc/passwd > /dev/null
0.000 ( ): __augmented_syscalls__:dfd: CWD, filename: 0x9c6a1da8, flags: CLOEXEC
0.005 ( 0.015 ms): cat/27341 openat(dfd: CWD, filename: 0x9c6a1da8, flags: CLOEXEC ) = 3
0.040 ( ): __augmented_syscalls__:dfd: CWD, filename: 0x9c8a9ce0, flags: CLOEXEC
0.041 ( 0.006 ms): cat/27341 openat(dfd: CWD, filename: 0x9c8a9ce0, flags: CLOEXEC ) = 3
0.294 ( ): __augmented_syscalls__:dfd: CWD, filename: 0x482a706b
0.296 ( 0.067 ms): cat/27341 openat(dfd: CWD, filename: 0x482a706b ) = 3
#
Now lets replace that __augmented_syscalls__ name with the syscall name,
using:
# grep 'field:.*syscall_nr' /sys/kernel/debug/tracing/events/syscalls/sys_enter_openat/format
field:int __syscall_nr; offset:8; size:4; signed:1;
#
That the synthesized payload has exactly where the syscall enter
tracepoint puts it.
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: https://lkml.kernel.org/n/tip-og4r9k87mzp9hv7el046idmd@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Add an example BPF script that writes syscalls:sys_enter_openat raw
tracepoint payloads augmented with the first 64 bytes of the "filename"
syscall pointer arg.
Then catch it and print it just like with things written to the
"__bpf_stdout__" map associated with a PERF_COUNT_SW_BPF_OUTPUT software
event, by just letting the default tracepoint handler in 'perf trace',
trace__event_handler(), to use bpf_output__fprintf(trace, sample), just
like it does with all other PERF_COUNT_SW_BPF_OUTPUT events, i.e. just
do a dump on the payload, so that we can check if what is being printed
has at least the first 64 bytes of the "filename" arg:
The augmented_syscalls.c eBPF script:
# cat tools/perf/examples/bpf/augmented_syscalls.c
// SPDX-License-Identifier: GPL-2.0
#include <stdio.h>
struct bpf_map SEC("maps") __augmented_syscalls__ = {
.type = BPF_MAP_TYPE_PERF_EVENT_ARRAY,
.key_size = sizeof(int),
.value_size = sizeof(u32),
.max_entries = __NR_CPUS__,
};
struct syscall_enter_openat_args {
unsigned long long common_tp_fields;
long syscall_nr;
long dfd;
char *filename_ptr;
long flags;
long mode;
};
struct augmented_enter_openat_args {
struct syscall_enter_openat_args args;
char filename[64];
};
int syscall_enter(openat)(struct syscall_enter_openat_args *args)
{
struct augmented_enter_openat_args augmented_args;
probe_read(&augmented_args.args, sizeof(augmented_args.args), args);
probe_read_str(&augmented_args.filename, sizeof(augmented_args.filename), args->filename_ptr);
perf_event_output(args, &__augmented_syscalls__, BPF_F_CURRENT_CPU,
&augmented_args, sizeof(augmented_args));
return 1;
}
license(GPL);
#
So it will just prepare a raw_syscalls:sys_enter payload for the
"openat" syscall.
This will eventually be done for all syscalls with pointer args,
globally or just when the user asks, using some spec, which args of
which syscalls it wants "expanded" this way, we'll probably start with
just all the syscalls that have char * pointers with familiar names, the
ones we already handle with the probe:vfs_getname kprobe if it is in
place hooking the kernel getname_flags() function used to copy from user
the paths.
Running it we get:
# perf trace -e perf/tools/perf/examples/bpf/augmented_syscalls.c,openat cat /etc/passwd > /dev/null
0.000 ( ): __augmented_syscalls__:X?.C......................`\..................../etc/ld.so.cache..#......,....ao.k...............k......1.".........
0.006 ( ): syscalls:sys_enter_openat:dfd: CWD, filename: 0x5c600da8, flags: CLOEXEC
0.008 ( 0.005 ms): cat/31292 openat(dfd: CWD, filename: 0x5c600da8, flags: CLOEXEC ) = 3
0.036 ( ): __augmented_syscalls__:X?.C.......................\..................../lib64/libc.so.6......... .\....#........?.......=.C..../.".........
0.037 ( ): syscalls:sys_enter_openat:dfd: CWD, filename: 0x5c808ce0, flags: CLOEXEC
0.039 ( 0.007 ms): cat/31292 openat(dfd: CWD, filename: 0x5c808ce0, flags: CLOEXEC ) = 3
0.323 ( ): __augmented_syscalls__:X?.C.....................P....................../etc/passwd......>.C....@................>.C.....,....ao.>.C........
0.325 ( ): syscalls:sys_enter_openat:dfd: CWD, filename: 0xe8be50d6
0.327 ( 0.004 ms): cat/31292 openat(dfd: CWD, filename: 0xe8be50d6 ) = 3
#
We need to go on optimizing this to avoid seding trash or zeroes in the
pointer content payload, using the return from bpf_probe_read_str(), but
to keep things simple at this stage and make incremental progress, lets
leave it at that for now.
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: https://lkml.kernel.org/n/tip-g360n1zbj6bkbk6q0qo11c28@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
That, together with the map __bpf_output__ that is already handled by
'perf trace' to print that event's contents as strings provides a
debugging facility, to show it in use, print a simple string everytime
the syscalls:sys_enter_openat() syscall tracepoint is hit:
# cat tools/perf/examples/bpf/hello.c
#include <stdio.h>
int syscall_enter(openat)(void *args)
{
puts("Hello, world\n");
return 0;
}
license(GPL);
#
# perf trace -e openat,tools/perf/examples/bpf/hello.c cat /etc/passwd > /dev/null
0.016 ( ): __bpf_stdout__:Hello, world
0.018 ( 0.010 ms): cat/9079 openat(dfd: CWD, filename: /etc/ld.so.cache, flags: CLOEXEC) = 3
0.057 ( ): __bpf_stdout__:Hello, world
0.059 ( 0.011 ms): cat/9079 openat(dfd: CWD, filename: /lib64/libc.so.6, flags: CLOEXEC) = 3
0.417 ( ): __bpf_stdout__:Hello, world
0.419 ( 0.009 ms): cat/9079 openat(dfd: CWD, filename: /etc/passwd) = 3
#
This is part of an ongoing experimentation on making eBPF scripts as
consumed by perf to be as concise as possible and using familiar
concepts such as stdio.h functions, that end up just wrapping the
existing BPF functions, trying to hide as much boilerplate as possible
while using just conventions and C preprocessor tricks.
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: https://lkml.kernel.org/n/tip-4tiaqlx5crf0fwpe7a6j84x7@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Allowing one to hook into the syscalls:sys_enter_NAME tracepoints,
an example is provided that hooks into the 'openat' syscall.
Using it with the probe:vfs_getname probe into getname_flags to get the
filename args as it is copied from userspace:
# perf probe -l
probe:vfs_getname (on getname_flags:73@acme/git/linux/fs/namei.c with pathname)
# perf trace -e probe:*getname,tools/perf/examples/bpf/sys_enter_openat.c cat /etc/passwd > /dev/null
0.000 probe:vfs_getname:(ffffffffbd2a8983) pathname="/etc/ld.so.preload"
0.022 syscalls:sys_enter_openat:dfd: CWD, filename: 0xafbe8da8, flags: CLOEXEC
0.027 probe:vfs_getname:(ffffffffbd2a8983) pathname="/etc/ld.so.cache"
0.054 syscalls:sys_enter_openat:dfd: CWD, filename: 0xafdf0ce0, flags: CLOEXEC
0.057 probe:vfs_getname:(ffffffffbd2a8983) pathname="/lib64/libc.so.6"
0.316 probe:vfs_getname:(ffffffffbd2a8983) pathname="/usr/lib/locale/locale-archive"
0.375 syscalls:sys_enter_openat:dfd: CWD, filename: 0xe2b2b0b4
0.379 probe:vfs_getname:(ffffffffbd2a8983) pathname="/etc/passwd"
#
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: https://lkml.kernel.org/n/tip-2po9jcqv1qgj0koxlg8kkg30@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
So that kprobe definitions become:
int probe(function, variables)(void *ctx, int err, var1, var2, ...)
The existing 5sec.c, got converted and goes from:
SEC("func=hrtimer_nanosleep rqtp->tv_sec")
int func(void *ctx, int err, long sec)
{
}
To:
int probe(hrtimer_nanosleep, rqtp->tv_sec)(void *ctx, int err, long sec)
{
}
If we decide to add tv_nsec as well, then it becomes:
$ cat tools/perf/examples/bpf/5sec.c
#include <bpf.h>
int probe(hrtimer_nanosleep, rqtp->tv_sec rqtp->tv_nsec)(void *ctx, int err, long sec, long nsec)
{
return sec == 5;
}
license(GPL);
$
And if we run it, system wide as before and run some 'sleep' with values
for the tv_nsec field, we get:
# perf trace --no-syscalls -e tools/perf/examples/bpf/5sec.c
0.000 perf_bpf_probe:hrtimer_nanosleep:(ffffffff9811b5f0) tv_sec=5 tv_nsec=100000000
9641.650 perf_bpf_probe:hrtimer_nanosleep:(ffffffff9811b5f0) tv_sec=5 tv_nsec=123450001
^C#
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: https://lkml.kernel.org/n/tip-1v9r8f6ds5av0w9pcwpeknyl@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
To further reduce boilerplate.
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: https://lkml.kernel.org/n/tip-vst6hj335s0ebxzqltes3nsc@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Description:
. Disable strace like syscall tracing (--no-syscalls), or try tracing
just some (-e *sleep).
. Attach a filter function to a kernel function, returning when it should
be considered, i.e. appear on the output:
$ cat tools/perf/examples/bpf/5sec.c
#include <bpf.h>
SEC("func=hrtimer_nanosleep rqtp->tv_sec")
int func(void *ctx, int err, long sec)
{
return sec == 5;
}
char _license[] SEC("license") = "GPL";
int _version SEC("version") = LINUX_VERSION_CODE;
$
. Run it system wide, so that any sleep of >= 5 seconds and < than 6
seconds gets caught.
. Ask for callgraphs using DWARF info, so that userspace can be unwound
. While this is running, run something like "sleep 5s".
# perf trace --no-syscalls -e tools/perf/examples/bpf/5sec.c/call-graph=dwarf/
0.000 perf_bpf_probe:func:(ffffffff9811b5f0) tv_sec=5
hrtimer_nanosleep ([kernel.kallsyms])
__x64_sys_nanosleep ([kernel.kallsyms])
do_syscall_64 ([kernel.kallsyms])
entry_SYSCALL_64 ([kernel.kallsyms])
__GI___nanosleep (/usr/lib64/libc-2.26.so)
rpl_nanosleep (/usr/bin/sleep)
xnanosleep (/usr/bin/sleep)
main (/usr/bin/sleep)
__libc_start_main (/usr/lib64/libc-2.26.so)
_start (/usr/bin/sleep)
^C#
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: https://lkml.kernel.org/n/tip-2nmxth2l2h09f9gy85lyexcq@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
So, the first helper is the one shortening a variable/function section
attribute, from, for instance:
char _license[] __attribute__((section("license"), used)) = "GPL";
to:
char _license[] SEC("license") = "GPL";
Convert empty.c to that and it becomes:
# cat ~acme/lib/examples/perf/bpf/empty.c
#include <bpf.h>
char _license[] SEC("license") = "GPL";
int _version SEC("version") = LINUX_VERSION_CODE;
#
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: https://lkml.kernel.org/n/tip-zmeg52dlvy51rdlhyumfl5yf@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
The first one is the bare minimum that bpf infrastructure accepts before
it expects actual events to be set up:
$ cat tools/perf/examples/bpf/empty.c
char _license[] __attribute__((section("license"), used)) = "GPL";
int _version __attribute__((section("version"), used)) = LINUX_VERSION_CODE;
$
If you remove that "version" line, then it will be refused with:
# perf trace -e tools/perf/examples/bpf/empty.c
event syntax error: 'tools/perf/examples/bpf/empty.c'
\___ Failed to load tools/perf/examples/bpf/empty.c from source: 'version' section incorrect or lost
(add -v to see detail)
Run 'perf list' for a list of valid events
Usage: perf trace [<options>] [<command>]
or: perf trace [<options>] -- <command> [<options>]
or: perf trace record [<options>] [<command>]
or: perf trace record [<options>] -- <command> [<options>]
-e, --event <event> event/syscall selector. use 'perf list' to list available events
#
The next ones will, step by step, show simple filters, then the needs
for headers will be made clear, it will be put in place and tested with
new examples, rinse, repeat.
Back to using this first one to test the perf+bpf infrastructure:
If we run it will fail, as no functions are present connecting with,
say, a tracepoint or a function using the kprobes or uprobes
infrastructure:
# perf trace -e tools/perf/examples/bpf/empty.c
WARNING: event parser found nothing
invalid or unsupported event: 'tools/perf/examples/bpf/empty.c'
Run 'perf list' for a list of valid events
Usage: perf trace [<options>] [<command>]
or: perf trace [<options>] -- <command> [<options>]
or: perf trace record [<options>] [<command>]
or: perf trace record [<options>] -- <command> [<options>]
-e, --event <event> event/syscall selector. use 'perf list' to list available events
#
But, if we set things up to dump the generated object file to a file,
and do this after having run 'make install', still on the developer's
$HOME directory:
# cat ~/.perfconfig
[llvm]
dump-obj = true
#
# perf trace -e ~acme/lib/examples/perf/bpf/empty.c
LLVM: dumping /home/acme/lib/examples/perf/bpf/empty.o
WARNING: event parser found nothing
invalid or unsupported event: '/home/acme/lib/examples/perf/bpf/empty.c'
<SNIP>
#
We can look at the dumped object file:
# ls -la ~acme/lib/examples/perf/bpf/empty.o
-rw-r--r--. 1 root root 576 May 4 12:10 /home/acme/lib/examples/perf/bpf/empty.o
# file ~acme/lib/examples/perf/bpf/empty.o
/home/acme/lib/examples/perf/bpf/empty.o: ELF 64-bit LSB relocatable, *unknown arch 0xf7* version 1 (SYSV), not stripped
# readelf -sw ~acme/lib/examples/perf/bpf/empty.o
Symbol table '.symtab' contains 3 entries:
Num: Value Size Type Bind Vis Ndx Name
0: 0000000000000000 0 NOTYPE LOCAL DEFAULT UND
1: 0000000000000000 0 NOTYPE GLOBAL DEFAULT 3 _license
2: 0000000000000000 0 NOTYPE GLOBAL DEFAULT 4 _version
#
# tools/bpf/bpftool/bpftool --pretty ~acme/lib/examples/perf/bpf/empty.o
null
#
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: https://lkml.kernel.org/n/tip-y7dkhakejz3013o0w21n98xd@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Arnaldo Carvalho de Melo