Intended for search (Ctrl-F) and reference. For tutorials, start with tutorial.md.
This guide is incomplete. If something feels missing, check the bcc and kernel source. And if you confirm we're missing something, please send a pull request to fix it, and help out everyone.
This section describes the C part of a bcc program.
Syntax: kprobe__kernel_function_name
kprobe__ is a special prefix that creates a kprobe (dynamic tracing of a kernel function call) for the kernel function name provided as the remainder. You can also use kprobes by declaring a normal C function, then using the Python BPF.attach_kprobe() (covered later) to associate it with a kernel function.
Arguments are specified on the function declaration: kprobe__kernel_function_name(struct pt_regs *ctx [, argument1 ...])
For example:
int kprobe__tcp_v4_connect(struct pt_regs *ctx, struct sock *sk) [...] }
This instruments the tcp_v4_connect() kernel function using a kprobe, with the following arguments:
struct pt_regs *ctx: Registers and BPF context.struct sock *sk: First argument to tcp_v4_connect().The first argument is always struct pt_regs *, the remainder are the arguments to the function (they don‘t need to be specified, if you don’t intend to use them).
Examples in situ: code (output), code (output)
Syntax: kretprobe__kernel_function_name
kretprobe__ is a special prefix that creates a kretprobe (dynamic tracing of a kernel function return) for the kernel function name provided as the remainder. You can also use kretprobes by declaring a normal C function, then using the Python BPF.attach_kretprobe() (covered later) to associate it with a kernel function.
Return value is available as PT_REGS_RC(ctx), given a function declaration of: kretprobe__kernel_function_name(struct pt_regs *ctx)
For example:
int kretprobe__tcp_v4_connect(struct pt_regs *ctx) { int ret = PT_REGS_RC(ctx); [...] }
This instruments the return of the tcp_v4_connect() kernel function using a kretprobe, and stores the return value in ret.
Examples in situ: code (output)
Syntax: TRACEPOINT_PROBE(category, event)
This is a macro that instruments the tracepoint defined by category:event.
Arguments are available in an args struct, which are the tracepoint arguments. One way to list these is to cat the relevant format file under /sys/kernel/debug/tracing/events/category/event/format.
The args struct can be used in place of ctx in each functions requiring a context as an argument. This includes notably perf_submit().
For example:
TRACEPOINT_PROBE(random, urandom_read) { // args is from /sys/kernel/debug/tracing/events/random/urandom_read/format bpf_trace_printk("%d\\n", args->got_bits); return 0; }
This instruments the random:urandom_read tracepoint, and prints the tracepoint argument got_bits.
Examples in situ: code (output), search /examples, search /tools
These are instrumented by declaring a normal function in C, then associating it as a uprobe probe in Python via BPF.attach_uprobe() (covered later).
Arguments can be examined using PT_REGS_PARM macros.
For example:
int count(struct pt_regs *ctx) { char buf[64]; bpf_probe_read(&buf, sizeof(buf), (void *)PT_REGS_PARM1(ctx)); bpf_trace_printk("%s %d", buf, PT_REGS_PARM2(ctx)); return(0); }
This reads the first argument as a string, and then prints it with the second argument as an integer.
Examples in situ: code
These are instrumented by declaring a normal function in C, then associating it as a uretprobe probe in Python via BPF.attach_uretprobe() (covered later).
Return value is available as PT_REGS_RC(ctx), given a function declaration of: function_name(struct pt_regs *ctx)
For example:
BPF_HISTOGRAM(dist); int count(struct pt_regs *ctx) { dist.increment(PT_REGS_RC(ctx)); return 0; }
This increments the bucket in the dist histogram that is indexed by the return value.
Examples in situ: code (output), code (output)
These are User Statically-Defined Tracing (USDT) probes, which may be placed in some applications or libraries to provide a user-level equivalent of tracepoints. The primary BPF method provided for USDT support method is enable_probe(). USDT probes are instrumented by declaring a normal function in C, then associating it as a USDT probe in Python via USDT.enable_probe().
Arguments can be read via: bpf_usdt_readarg(index, ctx, &addr)
For example:
int do_trace(struct pt_regs *ctx) { uint64_t addr; char path[128]; bpf_usdt_readarg(6, ctx, &addr); bpf_probe_read(&path, sizeof(path), (void *)addr); bpf_trace_printk("path:%s\\n", path); return 0; };
This reads the sixth USDT argument, and then pulls it in as a string to path.
Examples in situ: code, search /examples, search /tools
Syntax: RAW_TRACEPOINT_PROBE(event)
This is a macro that instruments the raw tracepoint defined by event.
The argument is a pointer to struct bpf_raw_tracepoint_args, which is defined in bpf.h. The struct field args contains all parameters of the raw tracepoint where you can found at linux tree include/trace/events directory.
For example:
RAW_TRACEPOINT_PROBE(sched_switch) { // TP_PROTO(bool preempt, struct task_struct *prev, struct task_struct *next) struct task_struct *prev = (struct task_struct *)ctx->args[1]; struct task_struct *next= (struct task_struct *)ctx->args[2]; s32 prev_tgid, next_tgid; bpf_probe_read(&prev_tgid, sizeof(prev->tgid), &prev->tgid); bpf_probe_read(&next_tgid, sizeof(next->tgid), &next->tgid); bpf_trace_printk("%d -> %d\\n", prev_tgid, next_tgid); }
This instruments the sched:sched_switch tracepoint, and prints the prev and next tgid.
Examples in situ: search /tools
Syntax: int bpf_probe_read(void *dst, int size, const void *src)
Return: 0 on success
This copies a memory location to the BPF stack, so that BPF can later operate on it. For safety, all memory reads must pass through bpf_probe_read(). This happens automatically in some cases, such as dereferencing kernel variables, as bcc will rewrite the BPF program to include the necessary bpf_probe_reads().
Examples in situ: search /examples, search /tools
Syntax: int bpf_probe_read_str(void *dst, int size, const void *src)
Return:
This copies a NULL terminated string from memory location to BPF stack, so that BPF can later operate on it. In case the string length is smaller than size, the target is not padded with further NULL bytes. In case the string length is larger than size, just size - 1 bytes are copied and the last byte is set to NULL.
Examples in situ: search /examples, search /tools
Syntax: u64 bpf_ktime_get_ns(void)
Return: current time in nanoseconds
Examples in situ: search /examples, search /tools
Syntax: u64 bpf_get_current_pid_tgid(void)
Return: current->tgid << 32 | current->pid
Returns the process ID in the lower 32 bits (kernel's view of the PID, which in user space is usually presented as the thread ID), and the thread group ID in the upper 32 bits (what user space often thinks of as the PID). By directly setting this to a u32, we discard the upper 32 bits.
Examples in situ: search /examples, search /tools
Syntax: u64 bpf_get_current_uid_gid(void)
Return: current_gid << 32 | current_uid
Returns the user ID and group IDs.
Examples in situ: search /examples, search /tools
Syntax: bpf_get_current_comm(char *buf, int size_of_buf)
Return: 0 on success
Populates the first argument address with the current process name. It should be a pointer to a char array of at least size TASK_COMM_LEN, which is defined in linux/sched.h. For example:
#include <linux/sched.h> int do_trace(struct pt_regs *ctx) { char comm[TASK_COMM_LEN]; bpf_get_current_comm(&comm, sizeof(comm)); [...]
Examples in situ: search /examples, search /tools
Syntax: bpf_get_current_task()
Return: current task as a pointer to struct task_struct.
Returns a pointer to the current task‘s task_struct object. This helper can be used to compute the on-CPU time for a process, identify kernel threads, get the current CPU’s run queue, or retrieve many other pieces of information.
With Linux 4.13, due to issues with field randomization, you may need two #define directives before the includes:
#define randomized_struct_fields_start struct { #define randomized_struct_fields_end }; #include <linux/sched.h> int do_trace(void *ctx) { struct task_struct *t = (struct task_struct *)bpf_get_current_task(); [...]
Examples in situ: search /examples, search /tools
Syntax: unsigned int bpf_log2l(unsigned long v)
Returns the log-2 of the provided value. This is often used to create indexes for histograms, to construct power-of-2 histograms.
Examples in situ: search /examples, search /tools
Syntax: u32 bpf_get_prandom_u32()
Returns a pseudo-random u32.
Example in situ: search /examples, search /tools
Syntax: int bpf_override_return(struct pt_regs *, unsigned long rc)
Return: 0 on success
When used in a program attached to a function entry kprobe, causes the execution of the function to be skipped, immediately returning rc instead. This is used for targeted error injection.
bpf_override_return will only work when the kprobed function is whitelisted to allow error injections. Whitelisting entails tagging a function with BPF_ALLOW_ERROR_INJECTION() in the kernel source tree; see io_ctl_init for an example. If the kprobed function is not whitelisted, the bpf program will fail to attach with ioctl(PERF_EVENT_IOC_SET_BPF): Invalid argument
int kprobe__io_ctl_init(void *ctx) { bpf_override_return(ctx, -ENOMEM); return 0; }
Syntax: int bpf_trace_printk(const char *fmt, ...)
Return: 0 on success
A simple kernel facility for printf() to the common trace_pipe (/sys/kernel/debug/tracing/trace_pipe). This is ok for some quick examples, but has limitations: 3 args max, 1 %s only, and trace_pipe is globally shared, so concurrent programs will have clashing output. A better interface is via BPF_PERF_OUTPUT(). Note that calling this helper is made simpler than the original kernel version, which has fmt_size as the second parameter.
Examples in situ: search /examples, search /tools
Syntax: BPF_PERF_OUTPUT(name)
Creates a BPF table for pushing out custom event data to user space via a perf ring buffer. This is the preferred method for pushing per-event data to user space.
For example:
struct data_t { u32 pid; u64 ts; char comm[TASK_COMM_LEN]; }; BPF_PERF_OUTPUT(events); int hello(struct pt_regs *ctx) { struct data_t data = {}; data.pid = bpf_get_current_pid_tgid(); data.ts = bpf_ktime_get_ns(); bpf_get_current_comm(&data.comm, sizeof(data.comm)); events.perf_submit(ctx, &data, sizeof(data)); return 0; }
The output table is named events, and data is pushed to it via events.perf_submit().
Examples in situ: search /examples, search /tools
Syntax: int perf_submit((void *)ctx, (void *)data, u32 data_size)
Return: 0 on success
A method of a BPF_PERF_OUTPUT table, for submitting custom event data to user space. See the BPF_PERF_OUTPUT entry. (This ultimately calls bpf_perf_event_output().)
Examples in situ: search /examples, search /tools
Maps are BPF data stores, and are the basis for higher level object types including tables, hashes, and histograms.
Syntax: BPF_TABLE(_table_type, _key_type, _leaf_type, _name, _max_entries)
Creates a map named _name. Most of the time this will be used via higher-level macros, like BPF_HASH, BPF_HIST, etc.
Methods (covered later): map.lookup(), map.lookup_or_init(), map.delete(), map.update(), map.insert(), map.increment().
Examples in situ: search /examples, search /tools
Syntax: BPF_HASH(name [, key_type [, leaf_type [, size]]])
Creates a hash map (associative array) named name, with optional parameters.
Defaults: BPF_HASH(name, key_type=u64, leaf_type=u64, size=10240)
For example:
BPF_HASH(start, struct request *);
This creates a hash named start where the key is a struct request *, and the value defaults to u64. This hash is used by the disksnoop.py example for saving timestamps for each I/O request, where the key is the pointer to struct request, and the value is the timestamp.
Methods (covered later): map.lookup(), map.lookup_or_init(), map.delete(), map.update(), map.insert(), map.increment().
Examples in situ: search /examples, search /tools
Syntax: BPF_ARRAY(name [, leaf_type [, size]])
Creates an int-indexed array which is optimized for fastest lookup and update, named name, with optional parameters.
Defaults: BPF_ARRAY(name, leaf_type=u64, size=10240)
For example:
BPF_ARRAY(counts, u64, 32);
This creates an array named counts where with 32 buckets and 64-bit integer values. This array is used by the funccount.py example for saving call count of each function.
Methods (covered later): map.lookup(), map.update(), map.increment(). Note that all array elements are pre-allocated with zero values and can not be deleted.
Examples in situ: search /examples, search /tools
Syntax: BPF_HISTOGRAM(name [, key_type [, size ]])
Creates a histogram map named name, with optional parameters.
Defaults: BPF_HISTOGRAM(name, key_type=int, size=64)
For example:
BPF_HISTOGRAM(dist);
This creates a histogram named dist, which defaults to 64 buckets indexed by keys of type int.
Methods (covered later): map.increment().
Examples in situ: search /examples, search /tools
Syntax: BPF_STACK_TRACE(name, max_entries)
Creates stack trace map named name, with a maximum entry count provided. These maps are used to store stack traces.
For example:
BPF_STACK_TRACE(stack_traces, 1024);
This creates stack trace map named stack_traces, with a maximum number of stack trace entries of 1024.
Methods (covered later): map.get_stackid().
Examples in situ: search /examples, search /tools
Syntax: BPF_PERF_ARRAY(name, max_entries)
Creates perf array named name, with a maximum entry count provided, which must be equal to the number of system cpus. These maps are used to fetch hardware performance counters.
For example:
text=""" BPF_PERF_ARRAY(cpu_cycles, NUM_CPUS); """ b = bcc.BPF(text=text, cflags=["-DNUM_CPUS=%d" % multiprocessing.cpu_count()]) b["cpu_cycles"].open_perf_event(b["cpu_cycles"].HW_CPU_CYCLES)
This creates a perf array named cpu_cycles, with number of entries equal to the number of cpus/cores. The array is configured so that later calling map.perf_read() will return a hardware-calculated counter of the number of cycles elapsed from some point in the past. Only one type of hardware counter may be configured per table at a time.
Methods (covered later): map.perf_read().
Examples in situ: search /tests
Syntax: BPF_PERCPU_ARRAY(name [, leaf_type [, size]])
Creates NUM_CPU int-indexed arrays which are optimized for fastest lookup and update, named name, with optional parameters. Each CPU will have a separate copy of this array. The copies are not kept synchronized in any way.
Defaults: BPF_PERCPU_ARRAY(name, leaf_type=u64, size=10240)
For example:
BPF_PERCPU_ARRAY(counts, u64, 32);
This creates NUM_CPU arrays named counts where with 32 buckets and 64-bit integer values.
Methods (covered later): map.lookup(), map.update(), map.increment(). Note that all array elements are pre-allocated with zero values and can not be deleted.
Examples in situ: search /examples, search /tools
Syntax: BPF_LPM_TRIE(name [, key_type [, leaf_type [, size]]])
Creates a longest prefix match trie map named name, with optional parameters.
Defaults: BPF_LPM_TRIE(name, key_type=u64, leaf_type=u64, size=10240)
For example:
BPF_LPM_TRIE(trie, struct key_v6);
This creates an LPM trie map named trie where the key is a struct key_v6, and the value defaults to u64.
Methods (covered later): map.lookup(), map.lookup_or_init(), map.delete(), map.update(), map.insert(), map.increment().
Examples in situ: search /examples, search /tools
Syntax: BPF_PROG_ARRAY(name, size)
This creates a program array named name with size entries. Each entry of the array is either a file descriptor to a bpf program or NULL. The array acts as a jump table so that bpf programs can “tail-call” other bpf programs.
Methods (covered later): map.call().
Examples in situ: search /examples, search /tests, assign fd
Syntax: BPF_DEVMAP(name, size)
This creates a device map named name with size entries. Each entry of the map is an ifindex to a network interface. This map is only used in XDP.
For example:
BPF_DEVMAP(devmap, 10);
Methods (covered later): map.redirect_map().
Examples in situ: search /examples,
Syntax: BPF_CPUMAP(name, size)
This creates a cpu map named name with size entries. The index of the map represents the CPU id and each entry is the size of the ring buffer allocated for the CPU. This map is only used in XDP.
For example:
BPF_CPUMAP(cpumap, 16);
Methods (covered later): map.redirect_map().
Examples in situ: search /examples,
Syntax: *val map.lookup(&key)
Lookup the key in the map, and return a pointer to its value if it exists, else NULL. We pass the key in as an address to a pointer.
Examples in situ: search /examples, search /tools
Syntax: *val map.lookup_or_init(&key, &zero)
Lookup the key in the map, and return a pointer to its value if it exists, else initialize the key's value to the second argument. This is often used to initialize values to zero.
Examples in situ: search /examples, search /tools
Syntax: map.delete(&key)
Delete the key from the hash.
Examples in situ: search /examples, search /tools
Syntax: map.update(&key, &val)
Associate the value in the second argument to the key, overwriting any previous value.
Examples in situ: search /examples, search /tools
Syntax: map.insert(&key, &val)
Associate the value in the second argument to the key, only if there was no previous value.
Examples in situ: search /examples
Syntax: map.increment(key[, increment_amount])
Increments the key's value by increment_amount, which defaults to 1. Used for histograms.
Examples in situ: search /examples, search /tools
Syntax: int map.get_stackid(void *ctx, u64 flags)
This walks the stack found via the struct pt_regs in ctx, saves it in the stack trace map, and returns a unique ID for the stack trace.
Examples in situ: search /examples, search /tools
Syntax: u64 map.perf_read(u32 cpu)
This returns the hardware performance counter as configured in 5. BPF_PERF_ARRAY
Examples in situ: search /tests
Syntax: void map.call(void *ctx, int index)
This invokes bpf_tail_call() to tail-call the bpf program which the index entry in 9. BPF_PROG_ARRAY points to. A tail-call is different from the normal call. It reuses the current stack frame after jumping to another bpf program and never goes back. If the index entry is empty, it won't jump anywhere and the program execution continues as normal.
For example:
BPF_PROG_ARRAY(prog_array, 10); int tail_call(void *ctx) { bpf_trace_printk("Tail-call\n"); return 0; } int do_tail_call(void *ctx) { bpf_trace_printk("Original program\n"); prog_array.call(ctx, 2); return 0; }
b = BPF(src_file="example.c") tail_fn = b.load_func("tail_call", BPF.KPROBE) prog_array = b.get_table("prog_array") prog_array[c_int(2)] = c_int(tail_fn.fd) b.attach_kprobe(event="some_kprobe_event", fn_name="do_tail_call")
This assigns tail_call() to prog_array[2]. In the end of do_tail_call(), prog_array.call(ctx, 2) tail-calls tail_call() and executes it.
NOTE: To prevent infinite loop, the maximum number of tail-calls is 32 (MAX_TAIL_CALL_CNT).
Examples in situ: search /examples, search /tests
Syntax: int map.redirect_map(int index, int flags)
This redirects the incoming packets based on the index entry. If the map is 10. BPF_DEVMAP, the packet will be sent to the transmit queue of the network interface that the entry points to. If the map is 11. BPF_CPUMAP, the packet will be sent to the ring buffer of the index CPU and be processed by the CPU later.
If the packet is redirected successfully, the function will return XDP_REDIRECT. Otherwise, it will return XDP_ABORTED to discard the packet.
For example:
BPF_DEVMAP(devmap, 1); int redirect_example(struct xdp_md *ctx) { return devmap.redirect_map(0, 0); } int xdp_dummy(struct xdp_md *ctx) { return XDP_PASS; }
ip = pyroute2.IPRoute() idx = ip.link_lookup(ifname="eth1")[0] b = bcc.BPF(src_file="example.c") devmap = b.get_table("devmap") devmap[c_uint32(0)] = c_int(idx) in_fn = b.load_func("redirect_example", BPF.XDP) out_fn = b.load_func("xdp_dummy", BPF.XDP) b.attach_xdp("eth0", in_fn, 0) b.attach_xdp("eth1", out_fn, 0)
Examples in situ: search /examples,
Depending on which BPF helpers are used, a GPL-compatible license is required.
The special BCC macro BPF_LICENSE specifies the license of the BPF program. You can set the license as a comment in your source code, but the kernel has a special interface to specify it programmatically. If you need to use GPL-only helpers, it is recommended to specify the macro in your C code so that the kernel can understand it:
// SPDX-License-Identifier: GPL-2.0+ #define BPF_LICENSE GPL
Otherwise, the kernel may reject loading your program (see the error description below). Note that it supports multiple words and quotes are not necessary:
// SPDX-License-Identifier: GPL-2.0+ OR BSD-2-Clause #define BPF_LICENSE Dual BSD/GPL
Check the BPF helpers reference to see which helpers are GPL-only and what the kernel understands as GPL-compatible.
If the macro is not specified, BCC will automatically define the license of the program as GPL.
Constructors.
Syntax: BPF({text=BPF_program | src_file=filename} [, usdt_contexts=[USDT_object, ...]] [, cflags=[arg1, ...]] [, debug=int])
Creates a BPF object. This is the main object for defining a BPF program, and interacting with its output.
Exactly one of text or src_file must be supplied (not both).
The cflags specifies additional arguments to be passed to the compiler, for example -DMACRO_NAME=value or -I/include/path. The arguments are passed as an array, with each element being an additional argument. Note that strings are not split on whitespace, so each argument must be a different element of the array, e.g. ["-include", "header.h"].
The debug flags control debug output, and can be or'ed together:
DEBUG_LLVM_IR = 0x1 compiled LLVM IRDEBUG_BPF = 0x2 loaded BPF bytecode and register state on branchesDEBUG_PREPROCESSOR = 0x4 pre-processor resultDEBUG_SOURCE = 0x8 ASM instructions embedded with sourceDEBUG_BPF_REGISTER_STATE = 0x10 register state on all instructions in addition to DEBUG_BPFExamples:
# define entire BPF program in one line: BPF(text='int do_trace(void *ctx) { bpf_trace_printk("hit!\\n"); return 0; }'); # define program as a variable: prog = """ int hello(void *ctx) { bpf_trace_printk("Hello, World!\\n"); return 0; } """ b = BPF(text=prog) # source a file: b = BPF(src_file = "vfsreadlat.c") # include a USDT object: u = USDT(pid=int(pid)) [...] b = BPF(text=bpf_text, usdt_contexts=[u]) # add include paths: u = BPF(text=prog, cflags=["-I/path/to/include"])
Examples in situ: search /examples, search /tools
Syntax: USDT({pid=pid | path=path})
Creates an object to instrument User Statically-Defined Tracing (USDT) probes. Its primary method is enable_probe().
Arguments:
Examples:
# include a USDT object: u = USDT(pid=int(pid)) [...] b = BPF(text=bpf_text, usdt_contexts=[u])
Examples in situ: search /examples, search /tools
Syntax: BPF.attach_kprobe(event="event", fn_name="name")
Instruments the kernel function event() using kernel dynamic tracing of the function entry, and attaches our C defined function name() to be called when the kernel function is called.
For example:
b.attach_kprobe(event="sys_clone", fn_name="do_trace")
This will instrument the kernel sys_clone() function, which will then run our BPF defined do_trace() function each time it is called.
You can call attach_kprobe() more than once, and attach your BPF function to multiple kernel functions.
See the previous kprobes section for how to instrument arguments from BPF.
Examples in situ: search /examples, search /tools
Syntax: BPF.attach_kretprobe(event="event", fn_name="name")
Instruments the return of the kernel function event() using kernel dynamic tracing of the function return, and attaches our C defined function name() to be called when the kernel function returns.
For example:
b.attach_kretprobe(event="vfs_read", fn_name="do_return")
This will instrument the kernel vfs_read() function, which will then run our BPF defined do_return() function each time it is called.
You can call attach_kretprobe() more than once, and attach your BPF function to multiple kernel function returns.
See the previous kretprobes section for how to instrument the return value from BPF.
Examples in situ: search /examples, search /tools
Syntax: BPF.attach_tracepoint(tp="tracepoint", fn_name="name")
Instruments the kernel tracepoint described by tracepoint, and when hit, runs the BPF function name().
This is an explicit way to instrument tracepoints. The TRACEPOINT_PROBE syntax, covered in the earlier tracepoints section, is an alternate method with the advantage of auto-declaring an args struct containing the tracepoint arguments. With attach_tracepoint(), the tracepoint arguments need to be declared in the BPF program.
For example:
# define BPF program bpf_text = """ #include <uapi/linux/ptrace.h> struct urandom_read_args { // from /sys/kernel/debug/tracing/events/random/urandom_read/format u64 __unused__; u32 got_bits; u32 pool_left; u32 input_left; }; int printarg(struct urandom_read_args *args) { bpf_trace_printk("%d\\n", args->got_bits); return 0; }; """ # load BPF program b = BPF(text=bpf_text) b.attach_tracepoint("random:urandom_read", "printarg")
Notice how the first argument to printarg() is now our defined struct.
Examples in situ: code
Syntax: BPF.attach_uprobe(name="location", sym="symbol", fn_name="name")
Instruments the user-level function symbol() from either the library or binary named by location using user-level dynamic tracing of the function entry, and attach our C defined function name() to be called whenever the user-level function is called.
Libraries can be given in the name argument without the lib prefix, or with the full path (/usr/lib/...). Binaries can be given only with the full path (/bin/sh).
For example:
b.attach_uprobe(name="c", sym="strlen", fn_name="count")
This will instrument strlen() function from libc, and call our BPF function count() when it is called. Note how the “lib” in “libc” is not necessary to specify.
Other examples:
b.attach_uprobe(name="c", sym="getaddrinfo", fn_name="do_entry") b.attach_uprobe(name="/usr/bin/python", sym="main", fn_name="do_main")
You can call attach_uprobe() more than once, and attach your BPF function to multiple user-level functions.
See the previous uprobes section for how to instrument arguments from BPF.
Examples in situ: search /examples, search /tools
Syntax: BPF.attach_uretprobe(name="location", sym="symbol", fn_name="name")
Instruments the return of the user-level function symbol() from either the library or binary named by location using user-level dynamic tracing of the function return, and attach our C defined function name() to be called whenever the user-level function returns.
For example:
b.attach_uretprobe(name="c", sym="strlen", fn_name="count")
This will instrument strlen() function from libc, and call our BPF function count() when it returns.
Other examples:
b.attach_uprobe(name="c", sym="getaddrinfo", fn_name="do_entry") b.attach_uprobe(name="/usr/bin/python", sym="main", fn_name="do_main")
You can call attach_uretprobe() more than once, and attach your BPF function to multiple user-level functions.
See the previous uretprobes section for how to instrument the return value from BPF.
Examples in situ: search /examples, search /tools
Syntax: USDT.enable_probe(probe=probe, fn_name=name)
Attaches a BPF C function name to the USDT probe probe.
Example:
# enable USDT probe from given PID u = USDT(pid=int(pid)) u.enable_probe(probe="http__server__request", fn_name="do_trace")
To check if your binary has USDT probes, and what they are, you can run readelf -n binary and check the stap debug section.
Examples in situ: search /examples, search /tools
Syntax: BPF.attach_raw_tracepoint(tp="tracepoint", fn_name="name")
Instruments the kernel raw tracepoint described by tracepoint (event only, no category), and when hit, runs the BPF function name().
This is an explicit way to instrument tracepoints. The RAW_TRACEPOINT_PROBE syntax, covered in the earlier raw tracepoints section, is an alternate method.
For example:
b.attach_raw_tracepoint("sched_swtich", "do_trace")
Examples in situ: search /tools
Syntax: BPF.trace_print(fmt="fields")
This method continually reads the globally shared /sys/kernel/debug/tracing/trace_pipe file and prints its contents. This file can be written to via BPF and the bpf_trace_printk() function, however, that method has limitations, including a lack of concurrent tracing support. The BPF_PERF_OUTPUT mechanism, covered earlier, is preferred.
Arguments:
fmt: optional, and can contain a field formatting string. It defaults to None.Examples:
# print trace_pipe output as-is: b.trace_print() # print PID and message: b.trace_print(fmt="{1} {5}")
Examples in situ: search /examples, search /tools
Syntax: BPF.trace_fields(nonblocking=False)
This method reads one line from the globally shared /sys/kernel/debug/tracing/trace_pipe file and returns it as fields. This file can be written to via BPF and the bpf_trace_printk() function, however, that method has limitations, including a lack of concurrent tracing support. The BPF_PERF_OUTPUT mechanism, covered earlier, is preferred.
Arguments:
nonblocking: optional, defaults to False. When set to True, the program will not block waiting for input.Examples:
while 1: try: (task, pid, cpu, flags, ts, msg) = b.trace_fields() except ValueError: continue [...]
Examples in situ: search /examples, search /tools
Normal output from a BPF program is either:
Syntax: BPF.perf_buffer_poll()
This polls from all open perf ring buffers, calling the callback function that was provided when calling open_perf_buffer for each entry.
Example:
# loop with callback to print_event b["events"].open_perf_buffer(print_event) while 1: b.perf_buffer_poll()
Examples in situ: code, search /examples, search /tools
Maps are BPF data stores, and are used in bcc to implement a table, and then higher level objects on top of tables, including hashes and histograms.
Syntax: BPF.get_table(name)
Returns a table object. This is no longer used, as tables can now be read as items from BPF. Eg: BPF[name].
Examples:
counts = b.get_table("counts") counts = b["counts"]
These are equivalent.
Syntax: table.open_perf_buffers(callback, page_cnt=N, lost_cb=None)
This operates on a table as defined in BPF as BPF_PERF_OUTPUT(), and associates the callback Python function callback to be called when data is available in the perf ring buffer. This is part of the recommended mechanism for transferring per-event data from kernel to user space. The size of the perf ring buffer can be specified via the page_cnt parameter, which must be a power of two number of pages and defaults to 8. If the callback is not processing data fast enough, some submitted data may be lost. lost_cb will be called to log / monitor the lost count. If lost_cb is the default None value, it will just print a line of message to stderr.
Example:
# process event def print_event(cpu, data, size): event = ct.cast(data, ct.POINTER(Data)).contents [...] # loop with callback to print_event b["events"].open_perf_buffer(print_event) while 1: b.perf_buffer_poll()
Note that the data structure transferred will need to be declared in C in the BPF program, and in Python. For example:
// define output data structure in C struct data_t { u32 pid; u64 ts; char comm[TASK_COMM_LEN]; };
# define output data structure in Python TASK_COMM_LEN = 16 # linux/sched.h class Data(ct.Structure): _fields_ = [("pid", ct.c_ulonglong), ("ts", ct.c_ulonglong), ("comm", ct.c_char * TASK_COMM_LEN)]
Perhaps in a future bcc version, the Python data structure will be automatically generated from the C declaration.
Examples in situ: code, search /examples, search /tools
Syntax: table.items()
Returns an array of the keys in a table. This can be used with BPF_HASH maps to fetch, and iterate, over the keys.
Example:
# print output print("%10s %s" % ("COUNT", "STRING")) counts = b.get_table("counts") for k, v in sorted(counts.items(), key=lambda counts: counts[1].value): print("%10d \"%s\"" % (v.value, k.c.encode('string-escape')))
This example also uses the sorted() method to sort by value.
Examples in situ: search /examples, search /tools
Syntax: table.values()
Returns an array of the values in a table.
Syntax: table.clear()
Clears the table: deletes all entries.
Example:
# print map summary every second: while True: time.sleep(1) print("%-8s\n" % time.strftime("%H:%M:%S"), end="") dist.print_log2_hist(sym + " return:") dist.clear()
Examples in situ: search /examples, search /tools
Syntax: table.print_log2_hist(val_type="value", section_header="Bucket ptr", section_print_fn=None)
Prints a table as a log2 histogram in ASCII. The table must be stored as log2, which can be done using the BPF function bpf_log2l().
Arguments:
Example:
b = BPF(text=""" BPF_HISTOGRAM(dist); int kprobe__blk_account_io_completion(struct pt_regs *ctx, struct request *req) { dist.increment(bpf_log2l(req->__data_len / 1024)); return 0; } """) [...] b["dist"].print_log2_hist("kbytes")
Output:
kbytes : count distribution
0 -> 1 : 3 | |
2 -> 3 : 0 | |
4 -> 7 : 211 |********** |
8 -> 15 : 0 | |
16 -> 31 : 0 | |
32 -> 63 : 0 | |
64 -> 127 : 1 | |
128 -> 255 : 800 |**************************************|
This output shows a multi-modal distribution, with the largest mode of 128->255 kbytes and a count of 800.
This is an efficient way to summarize data, as the summarization is performed in-kernel, and only the count column is passed to user space.
Examples in situ: search /examples, search /tools
Syntax: table.print_linear_hist(val_type="value", section_header="Bucket ptr", section_print_fn=None)
Prints a table as a linear histogram in ASCII. This is intended to visualize small integer ranges, eg, 0 to 100.
Arguments:
Example:
b = BPF(text=""" BPF_HISTOGRAM(dist); int kprobe__blk_account_io_completion(struct pt_regs *ctx, struct request *req) { dist.increment(req->__data_len / 1024); return 0; } """) [...] b["dist"].print_linear_hist("kbytes")
Output:
kbytes : count distribution
0 : 3 |****** |
1 : 0 | |
2 : 0 | |
3 : 0 | |
4 : 19 |****************************************|
5 : 0 | |
6 : 0 | |
7 : 0 | |
8 : 4 |******** |
9 : 0 | |
10 : 0 | |
11 : 0 | |
12 : 0 | |
13 : 0 | |
14 : 0 | |
15 : 0 | |
16 : 2 |**** |
[...]
This is an efficient way to summarize data, as the summarization is performed in-kernel, and only the values in the count column are passed to user space.
Examples in situ: search /examples, search /tools
Some helper methods provided by bcc. Note that since we're in Python, we can import any Python library and their methods, including, for example, the libraries: argparse, collections, ctypes, datetime, re, socket, struct, subprocess, sys, and time.
Syntax: BPF.ksym(addr)
Translate a kernel memory address into a kernel function name, which is returned.
Example:
print("kernel function: " + b.ksym(addr))
Examples in situ: search /examples, search /tools
Syntax: BPF.ksymname(name)
Translate a kernel name into an address. This is the reverse of ksym. Returns -1 when the function name is unknown.
Example:
print("kernel address: %x" % b.ksymname("vfs_read"))
Examples in situ: search /examples, search /tools
Syntax: BPF.sym(addr, pid, show_module=False, show_offset=False)
Translate a memory address into a function name for a pid, which is returned. A pid of less than zero will access the kernel symbol cache. The show_module and show_offset parameters control whether the module in which the symbol lies should be displayed, and whether the instruction offset from the beginning of the symbol should be displayed. These extra parameters default to False.
Example:
print("function: " + b.sym(addr, pid))
Examples in situ: search /examples, search /tools
Syntax: BPF.num_open_kprobes()
Returns the number of open k[ret]probes. Can be useful for scenarios where event_re is used while attaching and detaching probes. Excludes perf_events readers.
Example:
b.attach_kprobe(event_re=pattern, fn_name="trace_count") matched = b.num_open_kprobes() if matched == 0: print("0 functions matched by \"%s\". Exiting." % args.pattern) exit()
Examples in situ: search /examples, search /tools
See the “Understanding eBPF verifier messages” section in the kernel source under Documentation/networking/filter.txt.
This can be due to trying to read memory directly, instead of operating on memory on the BPF stack. All memory reads must be passed via bpf_probe_read() to copy memory into the BPF stack, which can be automatic by the bcc rewriter in some cases of simple dereferencing. bpf_probe_read() does all the required checks.
Example:
bpf: Permission denied
0: (bf) r6 = r1
1: (79) r7 = *(u64 *)(r6 +80)
2: (85) call 14
3: (bf) r8 = r0
[...]
23: (69) r1 = *(u16 *)(r7 +16)
R7 invalid mem access 'inv'
Traceback (most recent call last):
File "./tcpaccept", line 179, in <module>
b = BPF(text=bpf_text)
File "/usr/lib/python2.7/dist-packages/bcc/__init__.py", line 172, in __init__
self._trace_autoload()
File "/usr/lib/python2.7/dist-packages/bcc/__init__.py", line 612, in _trace_autoload
fn = self.load_func(func_name, BPF.KPROBE)
File "/usr/lib/python2.7/dist-packages/bcc/__init__.py", line 212, in load_func
raise Exception("Failed to load BPF program %s" % func_name)
Exception: Failed to load BPF program kretprobe__inet_csk_accept
This error happens when a GPL-only helper is called from a non-GPL BPF program. To fix this error, do not use GPL-only helpers from a proprietary BPF program, or relicense the BPF program under a GPL-compatible license. Check which BPF helpers are GPL-only, and what licenses are considered GPL-compatible.
Example calling bpf_get_stackid(), a GPL-only BPF helper, from a proprietary program (#define BPF_LICENSE Proprietary):
bpf: Failed to load program: Invalid argument [...] 8: (85) call bpf_get_stackid#27 cannot call GPL only function from proprietary program
eBPF program compilation needs kernel sources or kernel headers with headers compiled. In case your kernel sources are at a non-standard location where BCC cannot find then, its possible to provide BCC the absolute path of the location by setting BCC_KERNEL_SOURCE to it.
By default, BCC stores the LINUX_VERSION_CODE in the generated eBPF object which is then passed along to the kernel when the eBPF program is loaded. Sometimes this is quite inconvenient especially when the kernel is slightly updated such as an LTS kernel release. Its extremely unlikely the slight mismatch would cause any issues with the loaded eBPF program. By setting BCC_LINUX_VERSION_CODE to the version of the kernel that's running, the check for verifying the kernel version can be bypassed. This is needed for programs that use kprobes. This needs to be encoded in the format: (VERSION * 65536) + (PATCHLEVEL * 256) + SUBLEVEL. For example, if the running kernel is 4.9.10, then can set export BCC_LINUX_VERSION_CODE=264458 to override the kernel version check successfully.