|
Re: Extracting data from tracepoints (and anything else)
Andrii Nakryiko
On Mon, Mar 23, 2020 at 9:38 AM <mayfieldtristan@...> wrote:
nit: this is a legacy syntax of specifying BPF maps, please see [0] for some newer examples [0] https://github.com/iovisor/bcc/tree/master/libbpf-tools you don't need to bpf_probe_read() ctx here, you can just access its members directly. if(!res) {But here if you want to read filename contents itself, you'll need to use bpf_probe_read_str(). Having data_t definition would be also helpful. }this error from verifier is quite misleading, but what verifier complains about here is that you try to read uninitialized pointer (arg) and pass it as a first parameter into bpf_probe_read(). But see above, you don't need to bpf_probe_read() anything, and even if you wanted to it would have to be done very differently: struct sys_enter_openat_args args; /* notice no pointer here */ bpf_probe_read(&args, sizeof(args), ctx); /* taking address of args, taking size of args, not its pointer */ The kernel didn't load the BPF programIt's a mix. If your data should/can be pre-aggregated in kernel, using map might benefit you in that you will be sending much less data to user-space. But if you want to send every piece of information than perf_buffer is faster and more convenient than having user-space query BPF maps all the time. You're welcome. Check libbpf-tools in BCC repo, it should give you some examples to work off of.
|
||
|
|
||
|
Array brace-enclosed initialization
Federico Parola <fede.parola@...>
Hello everybody, #include <linux/bpf.h>
#ifndef __section
# define __section(NAME) \
__attribute__((section(NAME), used))
#endif
#ifndef BPF_FUNC
# define BPF_FUNC(NAME, ...) \
(*NAME)(__VA_ARGS__) = (void *)BPF_FUNC_##NAME
#endif
#ifndef printk
# define printk(fmt, ...) \
({ \
char ____fmt[] = fmt; \
trace_printk(____fmt, sizeof(____fmt), ##__VA_ARGS__); \
})
#endif
static void BPF_FUNC(trace_printk, const char *fmt, int fmt_size, ...);
__section("prog")
int xdp_prog(struct xdp_md *ctx) {
int i;
int array[10] = {0, 1, 2, 3, 4, 5, 6, 7, 8, 9};
#pragma nounroll
for (i = 0; i < 10; i++) {
printk("%d", array[i]);
}
return XDP_PASS;
}
char __license[] __section("license") = "GPL"
This is the error reported by the verifier:
0: (b7) r6 = 0 1: (b7) r7 = 25637 2: (b7) r8 = 0 3: (73) *(u8 *)(r10 -2) = r6 last_idx 3 first_idx 0 regs=40 stack=0 before 2: (b7) r8 = 0 regs=40 stack=0 before 1: (b7) r7 = 25637 regs=40 stack=0 before 0: (b7) r6 = 0 4: (6b) *(u16 *)(r10 -4) = r7 5: (18) r1 = 0x0 7: (0f) r1 += r8 8: (61) r3 = *(u32 *)(r1 +0) R1 invalid mem access 'inv' processed 8 insns (limit 1000000) max_states_per_insn 0 total_states 0 peak_states 0 mark_read 0I tried compiling with clang-6 and clang-9 with optimizaton set to O1 and O2, but I get this error in all cases. If I initialize the array in another way (e.g. with a loop) the program works correctly. The eBPF bytecode generated by clang is the following: .text
.file "test.c"
.section prog,"ax",@progbits
.globl xdp_prog # -- Begin function xdp_prog
.p2align 3
.type xdp_prog,@function
xdp_prog: # @xdp_prog
# %bb.0:
r6 = 0
r7 = 25637
r8 = 0
LBB0_1: # =>This Inner Loop Header: Depth=1
*(u8 *)(r10 - 2) = r6
*(u16 *)(r10 - 4) = r7
r1 = .L__const.xdp_prog.array ll
r1 += r8
r3 = *(u32 *)(r1 + 0)
r1 = r10
r1 += -4
r2 = 3
call 6
r8 += 4
if r8 != 24 goto LBB0_1
# %bb.2:
r0 = 2
exit
.Lfunc_end0:
.size xdp_prog, .Lfunc_end0-xdp_prog
# -- End function
.type .L__const.xdp_prog.array,@object # @__const.xdp_prog.array
.section .rodata,"a",@progbits
.p2align 2
.L__const.xdp_prog.array:
.long 0 # 0x0
.long 1 # 0x1
.long 2 # 0x2
.long 3 # 0x3
.long 4 # 0x4
.long 5 # 0x5
.long 6 # 0x6
.long 7 # 0x7
.long 8 # 0x8
.long 9 # 0x9
.size .L__const.xdp_prog.array, 40
.type .L__const.xdp_prog.____fmt,@object # @__const.xdp_prog.____fmt
.section .rodata.str1.1,"aMS",@progbits,1
.L__const.xdp_prog.____fmt:
.asciz "%d"
.size .L__const.xdp_prog.____fmt, 3
.type __license,@object # @__license
.section license,"aw",@progbits
.globl __license
__license:
.asciz "GPL"
.size __license, 4
.addrsig
.addrsig_sym xdp_prog
.addrsig_sym __license
It seems like the array in the stack is not initialized in the code. With some declarations the code works, for example decalring the array in the following way:int array[10] = {0, 1, 2, 3};everything works, the generated bytecode is the following: .text
.file "test.c"
.section prog,"ax",@progbits
.globl xdp_prog # -- Begin function xdp_prog
.p2align 3
.type xdp_prog,@function
xdp_prog: # @xdp_prog
# %bb.0:
r1 = 2
*(u32 *)(r10 - 36) = r1
r6 = 0
*(u32 *)(r10 - 8) = r6
*(u32 *)(r10 - 12) = r6
*(u32 *)(r10 - 16) = r6
*(u32 *)(r10 - 20) = r6
*(u32 *)(r10 - 24) = r6
*(u32 *)(r10 - 28) = r6
*(u32 *)(r10 - 4) = r6
r1 = 3
*(u32 *)(r10 - 32) = r1
r1 = 1
*(u32 *)(r10 - 40) = r1
*(u8 *)(r10 - 42) = r6
r7 = 25637
*(u16 *)(r10 - 44) = r7
r1 = r10
r1 += -44
r2 = 3
r3 = 1
call 6
r8 = 4
LBB0_1: # =>This Inner Loop Header: Depth=1
r1 = r10
r1 += -40
r1 += r8
r3 = *(u32 *)(r1 + 0)
*(u8 *)(r10 - 42) = r6
*(u16 *)(r10 - 44) = r7
r1 = r10
r1 += -44
r2 = 3
call 6
r8 += 4
if r8 != 24 goto LBB0_1
# %bb.2:
r0 = 2
exit
.Lfunc_end0:
.size xdp_prog, .Lfunc_end0-xdp_prog
# -- End function
.type .L__const.xdp_prog.____fmt,@object # @__const.xdp_prog.____fmt
.section .rodata.str1.1,"aMS",@progbits,1
.L__const.xdp_prog.____fmt:
.asciz "%d"
.size .L__const.xdp_prog.____fmt, 3
.type __license,@object # @__license
.section license,"aw",@progbits
.globl __license
__license:
.asciz "GPL"
.size __license, 4
.addrsig
.addrsig_sym xdp_prog
.addrsig_sym __license
This time the array is correctly initialized into code.Is this a clang bug? |
||
|
|
||
|
Extracting data from tracepoints (and anything else)
Tristan Mayfield
I've been exploring the libbpf library for different versions of the Linux kernel, and trying to rewrite some of the BCC tools. I would like to do more work with CO-RE eventually, but I'm trying to understand the entire model of how BPF programs work and how data flows between the kernel, the VM, and userspace. I just started using perf buffers instead of bpf_trace_printk and came across an issue that has me scratching my head. In the below code, I'm not able to access the const char * arg in the tracepoint sys_enter_openat (kernel 4.15). For some reason the verifier rejects this code. I think it's valid C (although I'm a little bit rusty still) and I think I followed the correct flow where data must be copied from the kernel to the VM before being able to use.
If anyone has insight to share, I'd much appreciate it. Conversely, if anyone can point me in the direction of how to debug BPF programs that would be extremely helpful too. Should I just dig into learning the basics of BPF asm? Highlights of the code: struct bpf_map_def SEC("maps") events = { .type = BPF_MAP_TYPE_PERF_EVENT_ARRAY, .key_size = sizeof(int), .value_size = sizeof(u32), .max_entries = MAX_CPUS, }; struct sys_enter_openat_args { u16 common_type; u8 common_flags; u8 common_preempt_count; int common_pid; int __syscall_nr; int dfd; char *filename; int flags; __mode_t mode; }; SEC("tracepoint/syscalls/sys_enter_openat") int bpf_prog(struct sys_enter_openat_args *ctx) { struct data_t data; struct sys_enter_openat_args *args; int res = bpf_probe_read(args, sizeof(ctx), ctx); if(!res) { data.file = "couldn't get file"; } else { data.file = args->filename; } Error Message: bpf_load_program() err=13 0: (bf) r6 = r1 1: (b7) r2 = 8 2: (bf) r3 = r6 3: (85) call bpf_probe_read#4 R1 type=ctx expected=fp The kernel didn't load the BPF program data.pid = bpf_get_current_pid_tgid(); // use fn from libbpf.h to get pid_tgid bpf_get_current_comm(data.program_name, sizeof(data.program_name)); // puts current comm into char array bpf_perf_event_output(ctx, &events, 0, &data, sizeof(data)); return 0; } If more code would be helpful, I'm happy to share. I recognize that libbpf and CO-RE in later kernels provides an easier API for dealing with char * (bpf_probe_read_str() I believe) but I'm trying to understand what needs to be done to target different kernels and not just the most cutting edge. As a second question, how much should I learn about perf(1) and its overlap with BPF? Finally, for long-term monitoring solutions and passing readable data, do most programs rely on pinning maps to the vfs instead of using perf buffers or passing directly to a userspace process? Thanks for the patience and goodwill with a new systems dev. I've enjoyed my interactions with the BPF community. Tristan |
||
|
|
||
|
Study on annotation of design and implementation choices, and of technical debt
a.serebrenik@...
Dear all,
-- As software engineering research teams at the University of Sannio (Italy) and Eindhoven University of Technology (The Netherlands) we are interested in investigating the protocol used by developers while they have to annotate implementation and design
choices during their normal development activities. More specifically, we are looking at whether, where and what kind of annotations developers usually use trying to be focused more on those annotations mainly aimed at highlighting that the code is not in
the right shape (e.g., comments for annotating delayed or intended work activities such as TODO, FIXME, hack, workaround, etc). In the latter case, we are looking at what is the content of the above annotations, as well as how they usually behave while evolving
the code that has been previously annotated.
When answering the survey, in case your annotation practices are different in different open source projects you may contribute, please refer to how you behave for the projects where you have been contacted.
Filling out the survey will take about 5 minutes.
Please note that your identity and personal data will not be disclosed, while we plan to use the aggregated results and anonymized responses as part of a scientific publication.
If you have any questions about the questionnaire or our research, please do not hesitate to contact us.
You can find the survey link here:
Thanks and regards,
Gianmarco Fucci (gianmarcofucci94@...)
Fiorella Zampetti (fzampetti@...)
Alexander Serebrenik (a.serebrenik@...)
Massimiliano Di Penta (dipenta@...)
|
||
|
|
||
|
Re: is BCC tools safe to enable root privilegies in production?
Cristian Spinetta
Thanks for your fast reply! In our infrastructure the owners of the app can logging into the production VMs that are running their apps and execute a restricted list of command with sudo (e.g. tcpdump, netstat, ...). The idea is to give root access to each script of bcc tool (all within /usr/share/bcc/tools/*). We are concerned if there are some bcc scripts that can run another command like in the example above or if there are other security concerns to be aware of. Best, Cristian Spinetta On Fri, Mar 13, 2020 at 1:23 PM Brendan Gregg <brendan.d.gregg@...> wrote: On Fri, Mar 13, 2020 at 7:59 AM Cristian Spinetta <cebspinetta@...> wrote: |
||
|
|
||
|
Re: is BCC tools safe to enable root privilegies in production?
Brendan Gregg
On Fri, Mar 13, 2020 at 7:59 AM Cristian Spinetta <cebspinetta@...> wrote:
^^^^ sudo isn't safe. If you remove the BCC tool from this one-liner, you'll find it still destroys your disk. In practice the production concern I have is for the overhead of each tool, hence the overhead section in each tool's man page. Brendan
|
||
|
|
||
|
is BCC tools safe to enable root privilegies in production?
Cristian Spinetta
Hi all! I am curious about whether it is safe to enable root access to BCC scripts on production machines. In my company, each team has access to their instances via ssh, and we are thinking to allow them to use bcc in production. For this purpose we need to allow root access to any BCC tool. Do you think it would be safe? for example, is there some tool that can receive a command to execute? in that case it would be unsafe because someone could execute any command thought a bcc tool. e.g.: sudo /usr/share/bcc/tools/some-great-tool.sh dd if=/dev/zero of=/dev/sda bs=512 count=1 conv=notrunc Best, Cristian Spinetta |
||
|
|
||
|
Re: clang 10 for BPF CO-RE
On Wed, 11 Mar 2020 10:36:47 -0700
"Andrii Nakryiko" <andrii.nakryiko@...> wrote: On Wed, Mar 11, 2020 at 10:33 AM <tmayfield@...> wrote:I assume this is the blog post[1]:Thanks! Glad it was useful. [1] https://facebookmicrosites.github.io/bpf/blog/2020/02/20/bcc-to-libbpf-howto-guide.html Thanks for writing that Andrii! For using libbpf directly from C, we also have the XDP-tutorial[2], but doesn't contain BPF CO-RE examples. And it uses the old style map defines. We are planning to update/fix that, once LLVM 10 gets more widely available in distros. [2] https://github.com/xdp-project/xdp-tutorial Let me give you the manual compile recipe (that I got from Eelco):For kernel/libbpf development we do build Clang from sources, but you git clone https://github.com/llvm/llvm-project.git cd llvm-project mkdir -p llvm/build/install cd llvm/build cmake -G "Ninja" -DLLVM_TARGETS_TO_BUILD="BPF;X86" \ -DLLVM_ENABLE_PROJECTS="clang" \ -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=$PWD/install .. ninja && ninja install export PATH=$PWD/install/bin:$PATH -- Best regards, Jesper Dangaard Brouer MSc.CS, Principal Kernel Engineer at Red Hat LinkedIn: http://www.linkedin.com/in/brouer |
||
|
|
||
|
Re: clang 10 for BPF CO-RE
Andrii Nakryiko
On Wed, Mar 11, 2020 at 10:33 AM <tmayfield@...> wrote:
Thanks! Glad it was useful. For kernel/libbpf development we do build Clang from sources, but you can install it from packages as well. See https://apt.llvm.org/, there are packages for Clang 10 and even Clang 11 and they are updated frequently. Great, please do!
|
||
|
|
||
|
clang 10 for BPF CO-RE
Tristan Mayfield
Hi all,
Finally found the BPF blog and it's been nice to get more information on using libbpf directly since I don't have a lot of systems or kernel experience.
Scanning through the "BCC to libbpf" post, I noticed Andrii calls for using clang 10. I went to look at llvm releases and only saw clang/llvm 9 (as of September 2019).
Is clang 10 just built from source? Looking forward to building with CO-RE and move some of my BCC tooling to libbpf.
-Tristan |
||
|
|
||
|
Re: Getting function's address from BPF_TRACE_FENTRY BPF program
Yutaro Hayakawa
I see, so this means the fentry program
toggle quoted message
Show quoted text
needs to load and verify the program for every functions to attach right? In my (maybe very specific) case, the tool may attaches programs to more than 1000 functions. So it is important to reduce the programs to reduce the attach time. I will continue to use kprobe. Thank you very much for your help. Yutaro On Mar 8, 2020, at 4:19, Alexei Starovoitov <alexei.starovoitov@...> wrote: |
||
|
|
||
|
Re: Getting function's address from BPF_TRACE_FENTRY BPF program
Alexei Starovoitov
On Fri, Mar 6, 2020 at 11:19 PM Yutaro Hayakawa <yhayakawa3720@...> wrote:
I think this approach won't quite work with fentry because the same fenty type prog cannot be attached to multiple kernel functions. At load time the kernel verifier needs to hold target kernel function, check that arguments match, etc. So at that point the target function address is fixed and when fentry prog is called it will see only one 'faddr' == regs_ip. |
||
|
|
||
|
Getting function's address from BPF_TRACE_FENTRY BPF program
Yutaro Hayakawa
Hello, Is there any way to get the address of the function in fentry type programs like kprobe type programs does by PT_REGS_IP(pt_regs)? I'd like to migrate my kprobe based tool[1] to fentry based one, but only this feature is missing right now. Since the tool attaches single BPF program to the multiple kernel functions, it needs to have function's address to identify which function the trace data comes from. Regards, Yutaro |
||
|
|
||
|
why bpf output wakeup_events and sample_period is 1?
Hayden Livingston
wakeup_events and sample_period is set to 1. what is the reason for this?
Isn't it better if this number is higher so the polling doesn't happen all the time? what is "sample_period" if wakeup_events tells kernel to wake up. |
||
|
|
||
|
Re: BCC integration into Buildroot
Jugurtha BELKALEM
Unfortunately, no; I have not used libbpf directly but I was thinking of doing it. My goal for having BCC integrated into buildroot is is that embedded systems are not so limited as they were before. THis brings the following advantages : - We can reuse BCC scripts made for desktops and run them on embedded devices. BCC can fit smoothly to provide us with a clear, easy simple and ease script maintenance (easy even for non C developers who can understand quickly). - Having python parsing returned results opens a bunch of endless possibilities like drawing graphs, saving to a remote database or even hand it them to an IA engine to understand system's behaviour over time (without having to develop another application for that). People used SystemTap in the past for some embedded systems, some others are using LTTng for debugging. So why not BCC (though, It's right that we need more space compared to hard coded ebpf). Thanks for your response, I'm going to try it out. Regards. |
||
|
|
||
|
Re: BCC integration into Buildroot
Andrii Nakryiko
On Mon, Jun 3, 2019 at 4:52 AM Jugurtha BELKALEM <jugurtha.belkalem@...> wrote:
Have you looked at using libbpf and BPF CO-RE for such use cases? The difference is that you won't have any additional runtime dependencies (no Clang/LLVM, etc), which makes this more suitable for embedded applications. The main requirement for running BPF CO-RE programs would be to compile kernel with CONFIG_DEBUG_INFO_BTF=y for BTF type information. Check out also https://github.com/iovisor/bcc/pull/2755 that adds first BPF CO-RE converted tool to BCC. See few links below for more details.
|
||
|
|
||
|
Dale Hamel
https://android.googlesource.com/platform/external/adeb/+/master/BCC.md is a way to already use bcc on android. I also have a WIP branch of bpftrace that supports bionic libc, for Android. On Tue, Feb 25, 2020 at 07:03 Mingo <novelinuxer@...> wrote: Does bcc have an adaptation plan for the Android platform? |
||
|
|
||
|
Does bcc have an adaptation plan for the Android platform?
|
||
|
|
||
|
Re: Run CO-RE version's runqslower failed
Andrii Nakryiko
On Sun, Feb 23, 2020 at 7:39 PM Andrii Nakryiko via Lists.Iovisor.Org
<andrii.nakryiko=gmail.com@...> wrote: Discussion has been moved to https://github.com/iovisor/bcc/issues/2770 |
||
|
|
||
|
Re: Run CO-RE version's runqslower failed
Andrii Nakryiko
On Sun, Feb 23, 2020 at 6:52 PM <ethercflow@...> wrote:
Hi! runqslower expects that kernel was built with BTF type info (which is enabled by CONFIG_DEBUG_INFO_BTF=y Kconfig option). Can you please re-build your kernel with BTF enabled and try again? |
||
|
|
Pour la planète, n'imprimez ce mail que si c'est nécessaire