So everyone probably knows that glibc's /lib/libc.so.6 can be executed in the shell like a normal executable in which cases it prints its version information and exits. This is done via defining an entry point in the .so. For some cases it could be interesting to use this for other projects too. Unfortunately, the low-level entry point you can set by ld's -e option is a bit too low-level: the dynamic loader is not available so you cannot call any proper library functions. glibc for this reason implements the write() system call via a naked system call in this entry point.
My question now is, can anyone think of a nice way how one could bootstrap a full dynamic linker from that entry point so that one could access functions from other .so's?
Update 2: see Andrew G Morgan's slightly more complicated solution which does work for any GLIBC (that solution is also used in libc.so.6 itself (since forever), which is why you can run it as ./libc.so.6 (it prints version info when invoked that way)).
Update 1: this no longer works with newer GLIBC versions:
./a.out: error while loading shared libraries: ./pie.so: cannot dynamically load position-independent executable
Original answer from 2009:
Building your shared library with -pie option appears to give you everything you want:
/* pie.c */
#include <stdio.h>
int foo()
{
printf("in %s %s:%d\n", __func__, __FILE__, __LINE__);
return 42;
}
int main()
{
printf("in %s %s:%d\n", __func__, __FILE__, __LINE__);
return foo();
}
/* main.c */
#include <stdio.h>
extern int foo(void);
int main()
{
printf("in %s %s:%d\n", __func__, __FILE__, __LINE__);
return foo();
}
$ gcc -fPIC -pie -o pie.so pie.c -Wl,-E
$ gcc main.c ./pie.so
$ ./pie.so
in main pie.c:9
in foo pie.c:4
$ ./a.out
in main main.c:6
in foo pie.c:4
$
P.S. glibc implements write(3) via system call because it doesn't have anywhere else to call (it is the lowest level already). This has nothing to do with being able to execute libc.so.6.
I have been looking to add support for this to pam_cap.so, and found this question. As #EmployedRussian notes in a follow-up to their own post, the accepted answer stopped working at some point. It took a while to figure out how to make this work again, so here is a worked example.
This worked example involves 5 files to show how things work with some corresponding tests.
First, consider this trivial program (call it empty.c):
int main(int argc, char **argv) { return 0; }
Compiling it, we can see how it resolves the dynamic symbols on my system as follows:
$ gcc -o empty empty.c
$ objcopy --dump-section .interp=/dev/stdout empty ; echo
/lib64/ld-linux-x86-64.so.2
$ DL_LOADER=/lib64/ld-linux-x86-64.so.2
That last line sets a shell variable for use later.
Here are the two files that build my example shared library:
/* multi.h */
void multi_main(void);
void multi(const char *caller);
and
/* multi.c */
#include <stdio.h>
#include <stdlib.h>
#include "multi.h"
void multi(const char *caller) {
printf("called from %s\n", caller);
}
__attribute__((force_align_arg_pointer))
void multi_main(void) {
multi(__FILE__);
exit(42);
}
const char dl_loader[] __attribute__((section(".interp"))) =
DL_LOADER ;
(Update 2021-11-13: The forced alignment is to help __i386__ code be SSE compatible - without it we get hard to debug glibc SIGSEGV crashes.)
We can compile and run it as follows:
$ gcc -fPIC -shared -o multi.so -DDL_LOADER="\"${DL_LOADER}\"" multi.c -Wl,-e,multi_main
$ ./multi.so
called from multi.c
$ echo $?
42
So, this is a .so that can be executed as a stand alone binary. Next, we validate that it can be loaded as shared object.
/* opener.c */
#include <dlfcn.h>
#include <stdio.h>
#include <stdlib.h>
int main(int argc, char **argv) {
void *handle = dlopen("./multi.so", RTLD_NOW);
if (handle == NULL) {
perror("no multi.so load");
exit(1);
}
void (*multi)(const char *) = dlsym(handle, "multi");
multi(__FILE__);
}
That is we dynamically load the shared-object and run a function from it:
$ gcc -o opener opener.c -ldl
$ ./opener
called from opener.c
Finally, we link against this shared object:
/* main.c */
#include "multi.h"
int main(int argc, char **argv) {
multi(__FILE__);
}
Where we compile and run it as follows:
$ gcc main.c -o main multi.so
$ LD_LIBRARY_PATH=./ ./main
called from main.c
(Note, because multi.so isn't in a standard system library location, we need to override where the runtime looks for the shared object file with the LD_LIBRARY_PATH environment variable.)
I suppose you'd have your ld -e point to an entry point which would then use the dlopen() family of functions to find and bootstrap the rest of the dynamic linker. Of course you'd have to ensure that dlopen() itself was either statically linked or you might have to implement enough of your own linker stub to get at it (using system call interfaces such as mmap() just as libc itself is doing.
None of that sounds "nice" to me. In fact just the thought of reading the glibc sources (and the ld-linux source code, as one example) enough to assess the size of the job sounds pretty hoary to me. It might also be a portability nightmare. There may be major differences between how Linux implements ld-linux and how the linkages are done under OpenSolaris, FreeBSD, and so on. (I don't know).
Is there any way to dump the call stack in a running process in C or C++ every time a certain function is called? What I have in mind is something like this:
void foo()
{
print_stack_trace();
// foo's body
return
}
Where print_stack_trace works similarly to caller in Perl.
Or something like this:
int main (void)
{
// will print out debug info every time foo() is called
register_stack_trace_function(foo);
// etc...
}
where register_stack_trace_function puts some sort of internal breakpoint that will cause a stack trace to be printed whenever foo is called.
Does anything like this exist in some standard C library?
I am working on Linux, using GCC.
Background
I have a test run that behaves differently based on some commandline switches that shouldn't affect this behavior. My code has a pseudo-random number generator that I assume is being called differently based on these switches. I want to be able to run the test with each set of switches and see if the random number generator is called differently for each one.
Survey of C/C++ backtrace methods
In this answer I will try to run a single benchmark for a bunch of solutions to see which one runs faster, while also considering other points such as features and portability.
Tool
Time / call
Line number
Function name
C++ demangling
Recompile
Signal safe
As string
C
C++23 <stacktrace> GCC 12.1
7 us
y
y
y
y
n
y
n
Boost 1.74 stacktrace()
5 us
y
y
y
y
n
y
n
Boost 1.74 stacktrace::safe_dump_to
y
n
n
glibc backtrace_symbols_fd
25 us
n
-rdynamic
hacks
y
y
n
y
glibc backtrace_symbols
21 us
n
-rdynamic
hacks
y
n
y
y
GDB scripting
600 us
y
y
y
n
y
n
y
GDB code injection
n
n
y
libunwind
y
libdwfl
4 ms
n
y
libbacktrace
y
Empty cells mean "TODO", not "no".
us: microsecond
Line number: shows actual line number, not just function name + a memory address.
It is usually possible to recover the line number from an address manually after the fact with addr2line. But it is a pain.
Recompile: requires recompiling the program to get your traces. Not recompiling is better!
Signal safe: crucial for the important uses case of "getting a stack trace in case of segfault": How to automatically generate a stacktrace when my program crashes
As string: you get the stack trace as a string in the program itself, as opposed to e.g. just printing to stdout. Usually implies not signal safe, as we don't know the size of the stack trace string size in advance, and therefore requires malloc which is not async signal safe.
C: does it work on a plain-C project (yes, there are still poor souls out there), or is C++ required?
Test setup
All benchmarks will run the following
main.cpp
#include <cstdlib> // strtoul
#include <mystacktrace.h>
void my_func_2(void) {
print_stacktrace(); // line 6
}
void my_func_1(double f) {
(void)f;
my_func_2();
}
void my_func_1(int i) {
(void)i;
my_func_2(); // line 16
}
int main(int argc, char **argv) {
long long unsigned int n;
if (argc > 1) {
n = std::strtoul(argv[1], NULL, 0);
} else {
n = 1;
}
for (long long unsigned int i = 0; i < n; ++i) {
my_func_1(1); // line 27
}
}
This input is designed to test C++ name demangling since my_func_1(int) and my_func_1(float) are necessarily mangled as a way to implement C++ function overload.
We differentiate between the benchmarks by using different -I includes to point to different implementations of print_stacktrace().
Each benchmark is done with a command of form:
time ./stacktrace.out 100000 &>/dev/null
The number of iterations is adjusted for each implementation to produce a total runtime of the order of 1s for that benchmark.
-O0 is used on all tests below unless noted. Stack traces may be irreparably mutilated by certain optimizations. Tail call optimization is a notable example of that: What is tail call optimization? There's nothing we can do about it.
C++23 <stacktrace>
This method was previously mentioned at: https://stackoverflow.com/a/69384663/895245 please consider upvoting that answer.
This is the best solution... it's portable, fast, shows line numbers and demangles C++ symbols. This option will displace every other alternative as soon as it becomes more widely available, with the exception perhaps only of GDB for one-offs without the need or recompilation.
cpp20_stacktrace/mystacktrace.h
#include <iostream>
#include <stacktrace>
void print_stacktrace() {
std::cout << std::stacktrace::current();
}
GCC 12.1.0 from Ubuntu 22.04 does not have support compiled in, so for now I built it from source as per: How to edit and re-build the GCC libstdc++ C++ standard library source? and set --enable-libstdcxx-backtrace=yes, and it worked!
Compile with:
g++ -O0 -ggdb3 -Wall -Wextra -pedantic -std=c++23 -o cpp20_stacktrace.out main.cpp -lstdc++_libbacktrace
Sample output:
0# print_stacktrace() at cpp20_stacktrace/mystacktrace.h:5
1# my_func_2() at /home/ciro/main.cpp:6
2# my_func_1(int) at /home/ciro/main.cpp:16
3# at /home/ciro/main.cpp:27
4# at :0
5# at :0
6# at :0
7#
If we try to use GCC 12.1.0 from Ubuntu 22.04:
sudo apt install g++-12
g++-12 -ggdb3 -O2 -std=c++23 -Wall -Wextra -pedantic -o stacktrace.out stacktrace.cpp -lstdc++_libbacktrace
It fails with:
stacktrace.cpp: In function ‘void my_func_2()’:
stacktrace.cpp:6:23: error: ‘std::stacktrace’ has not been declared
6 | std::cout << std::stacktrace::current();
| ^~~~~~~~~~
Checking build options with:
g++-12 -v
does not show:
--enable-libstdcxx-backtrace=yes
so it wasn't compiled in. Bibliography:
How to use <stacktrace> in GCC trunk?
How can I generate a C++23 stacktrace with GCC 12.1?
It does not fail on the include because the header file:
/usr/include/c++/12
has a feature check:
#if __cplusplus > 202002L && _GLIBCXX_HAVE_STACKTRACE
Boost stacktrace
The library has changed quite a lot around Ubuntu 22.04, so make sure your version matches: Boost stack-trace not showing function names and line numbers
The library is pretty much superseded by the more portable C++23 implementation, but remains a very good option for those that are not at that standard version yet, but already have a "Boost clearance".
Documented at: https://www.boost.org/doc/libs/1_66_0/doc/html/stacktrace/getting_started.html#stacktrace.getting_started.how_to_print_current_call_stack
Tested on Ubuntu 22.04, boost 1.74.0, you should do:
boost_stacktrace/mystacktrace.h
#include <iostream>
#define BOOST_STACKTRACE_LINK
#include <boost/stacktrace.hpp>
void print_stacktrace(void) {
std::cout << boost::stacktrace::stacktrace();
}
On Ubuntu 19.10 boost 1.67.0 to get the line numbers we had to instead:
#include <iostream>
#define BOOST_STACKTRACE_USE_ADDR2LINE
#include <boost/stacktrace.hpp>
void print_stacktrace(void) {
std::cout << boost::stacktrace::stacktrace();
}
which would call out to the addr2line executable and be 1000x slower than the newer Boost version.
The package libboost-stacktrace-dev did not exist at all on Ubuntu 16.04.
The rest of this section considers only the Ubuntu 22.04, boost 1.74 behaviour.
Compile:
sudo apt-get install libboost-stacktrace-dev
g++ -O0 -ggdb3 -Wall -Wextra -pedantic -std=c++11 -o boost_stacktrace.out main.cpp -lboost_stacktrace_backtrace
Sample output:
0# print_stacktrace() at boost_stacktrace/mystacktrace.h:7
1# my_func_2() at /home/ciro/main.cpp:7
2# my_func_1(int) at /home/ciro/main.cpp:17
3# main at /home/ciro/main.cpp:26
4# __libc_start_call_main at ../sysdeps/nptl/libc_start_call_main.h:58
5# __libc_start_main at ../csu/libc-start.c:379
6# _start in ./boost_stacktrace.out
Note that the lines are off by one line. It was suggested in the comments that this is because the following instruction address is being considered.
Boost stacktrace header only
What the BOOST_STACKTRACE_LINK does is to require -lboost_stacktrace_backtrace at link time, so we imagine without that it will just work. This would be a good option for devs who don't have the "Boost clearance" to just add as one offs to debug.
TODO unfortunately it didn't so well for me:
#include <iostream>
#include <boost/stacktrace.hpp>
void print_stacktrace(void) {
std::cout << boost::stacktrace::stacktrace();
}
then:
g++ -O0 -ggdb3 -Wall -Wextra -pedantic -std=c++11 -o boost_stacktrace_header_only.out main.cpp
contains the overly short output:
0# 0x000055FF74AFB601 in ./boost_stacktrace_header_only.out
1# 0x000055FF74AFB66C in ./boost_stacktrace_header_only.out
2# 0x000055FF74AFB69C in ./boost_stacktrace_header_only.out
3# 0x000055FF74AFB6F7 in ./boost_stacktrace_header_only.out
4# 0x00007F0176E7BD90 in /lib/x86_64-linux-gnu/libc.so.6
5# __libc_start_main in /lib/x86_64-linux-gnu/libc.so.6
6# 0x000055FF74AFB4E5 in ./boost_stacktrace_header_only.out
which we can't even use with addr2line. Maybe we have to pass some other define from: https://www.boost.org/doc/libs/1_80_0/doc/html/stacktrace/configuration_and_build.html ?
Tested on Ubuntu 22.04. boost 1.74.
Boost boost::stacktrace::safe_dump_to
This is an interesting alternative to boost::stacktrace::stacktrace as it writes the stack trace in a async signal safe manner to a file, which makes it a good option for automatically dumping stack traces on segfaults which is a super common use case: How to automatically generate a stacktrace when my program crashes
Documented at: https://www.boost.org/doc/libs/1_70_0/doc/html/boost/stacktrace/safe_dump_1_3_38_7_6_2_1_6.html
TODO get it to work. All I see each time is a bunch of random bytes. My attempt:
boost_stacktrace_safe/mystacktrace.h
#include <unistd.h>
#define BOOST_STACKTRACE_LINK
#include <boost/stacktrace.hpp>
void print_stacktrace(void) {
boost::stacktrace::safe_dump_to(0, 1024, STDOUT_FILENO);
}
Sample output:
1[FU1[FU"2[FU}2[FUm1#n10[FU
Changes drastically each time, suggesting it is random memory addresses.
Tested on Ubuntu 22.04, boost 1.74.0.
glibc backtrace
This method is quite portable as it comes with glibc itself. Documented at: https://www.gnu.org/software/libc/manual/html_node/Backtraces.html
Tested on Ubuntu 22.04, glibc 2.35.
glibc_backtrace_symbols_fd/mystacktrace.h
#include <execinfo.h> /* backtrace, backtrace_symbols_fd */
#include <unistd.h> /* STDOUT_FILENO */
void print_stacktrace(void) {
size_t size;
enum Constexpr { MAX_SIZE = 1024 };
void *array[MAX_SIZE];
size = backtrace(array, MAX_SIZE);
backtrace_symbols_fd(array, size, STDOUT_FILENO);
}
Compile with:
g++ -O0 -ggdb3 -Wall -Wextra -pedantic -rdynamic -std=c++11 -o glibc_backtrace_symbols_fd.out main.cpp
Sample output with -rdynamic:
./glibc_backtrace_symbols.out(_Z16print_stacktracev+0x47) [0x556e6a131230]
./glibc_backtrace_symbols.out(_Z9my_func_2v+0xd) [0x556e6a1312d6]
./glibc_backtrace_symbols.out(_Z9my_func_1i+0x14) [0x556e6a131306]
./glibc_backtrace_symbols.out(main+0x58) [0x556e6a131361]
/lib/x86_64-linux-gnu/libc.so.6(+0x29d90) [0x7f175e7bdd90]
/lib/x86_64-linux-gnu/libc.so.6(__libc_start_main+0x80) [0x7f175e7bde40]
./glibc_backtrace_symbols.out(_start+0x25) [0x556e6a131125]
Sample output without -rdynamic:
./glibc_backtrace_symbols_fd_no_rdynamic.out(+0x11f0)[0x556bd40461f0]
./glibc_backtrace_symbols_fd_no_rdynamic.out(+0x123c)[0x556bd404623c]
./glibc_backtrace_symbols_fd_no_rdynamic.out(+0x126c)[0x556bd404626c]
./glibc_backtrace_symbols_fd_no_rdynamic.out(+0x12c7)[0x556bd40462c7]
/lib/x86_64-linux-gnu/libc.so.6(+0x29d90)[0x7f0da2b70d90]
/lib/x86_64-linux-gnu/libc.so.6(__libc_start_main+0x80)[0x7f0da2b70e40]
./glibc_backtrace_symbols_fd_no_rdynamic.out(+0x10e5)[0x556bd40460e5]
To get the line numbers without -rdynamic we can use addr2line:
addr2line -C -e glibc_backtrace_symbols_fd_no_rdynamic.out 0x11f0 0x123c 0x126c 0x12c7
addr2line cannot unfortunately handle the function name + offset in function format of when we are not using -rdynamic, e.g. _Z9my_func_2v+0xd.
GDB can however:
gdb -nh -batch -ex 'info line *(_Z9my_func_2v+0xd)' -ex 'info line *(_Z9my_func_1i+0x14)' glibc_backtrace_symbols.out
Line 7 of "main.cpp" starts at address 0x12d6 <_Z9my_func_2v+13> and ends at 0x12d9 <_Z9my_func_1d>.
Line 17 of "main.cpp" starts at address 0x1306 <_Z9my_func_1i+20> and ends at 0x1309 <main(int, char**)>.
A helper to make it more bearable:
addr2lines() (
perl -ne '$m = s/(.*).*\(([^)]*)\).*/gdb -nh -q -batch -ex "info line *\2" \1/;print $_ if $m' | bash
)
Usage:
xsel -b | addr2lines
glibc backtrace_symbols
A version of backtrace_symbols_fd that returns a string rather than printing to a file handle.
glibc_backtrace_symbols/mystacktrace.h
#include <execinfo.h> /* backtrace, backtrace_symbols */
#include <stdio.h> /* printf */
void print_stacktrace(void) {
char **strings;
size_t i, size;
enum Constexpr { MAX_SIZE = 1024 };
void *array[MAX_SIZE];
size = backtrace(array, MAX_SIZE);
strings = backtrace_symbols(array, size);
for (i = 0; i < size; i++)
printf("%s\n", strings[i]);
free(strings);
}
glibc backtrace with C++ demangling hack 1: -export-dynamic + dladdr
I couldn't find a simple way to automatically demangle C++ symbols with glibc backtrace.
https://panthema.net/2008/0901-stacktrace-demangled/
https://gist.github.com/fmela/591333/c64f4eb86037bb237862a8283df70cdfc25f01d3
Adapted from: https://gist.github.com/fmela/591333/c64f4eb86037bb237862a8283df70cdfc25f01d3
This is a "hack" because it requires changing the ELF with -export-dynamic.
glibc_ldl.cpp
#include <dlfcn.h> // for dladdr
#include <cxxabi.h> // for __cxa_demangle
#include <cstdio>
#include <string>
#include <sstream>
#include <iostream>
// This function produces a stack backtrace with demangled function & method names.
std::string backtrace(int skip = 1)
{
void *callstack[128];
const int nMaxFrames = sizeof(callstack) / sizeof(callstack[0]);
char buf[1024];
int nFrames = backtrace(callstack, nMaxFrames);
char **symbols = backtrace_symbols(callstack, nFrames);
std::ostringstream trace_buf;
for (int i = skip; i < nFrames; i++) {
Dl_info info;
if (dladdr(callstack[i], &info)) {
char *demangled = NULL;
int status;
demangled = abi::__cxa_demangle(info.dli_sname, NULL, 0, &status);
std::snprintf(
buf,
sizeof(buf),
"%-3d %*p %s + %zd\n",
i,
(int)(2 + sizeof(void*) * 2),
callstack[i],
status == 0 ? demangled : info.dli_sname,
(char *)callstack[i] - (char *)info.dli_saddr
);
free(demangled);
} else {
std::snprintf(buf, sizeof(buf), "%-3d %*p\n",
i, (int)(2 + sizeof(void*) * 2), callstack[i]);
}
trace_buf << buf;
std::snprintf(buf, sizeof(buf), "%s\n", symbols[i]);
trace_buf << buf;
}
free(symbols);
if (nFrames == nMaxFrames)
trace_buf << "[truncated]\n";
return trace_buf.str();
}
void my_func_2(void) {
std::cout << backtrace() << std::endl;
}
void my_func_1(double f) {
(void)f;
my_func_2();
}
void my_func_1(int i) {
(void)i;
my_func_2();
}
int main() {
my_func_1(1);
my_func_1(2.0);
}
Compile and run:
g++ -fno-pie -ggdb3 -O0 -no-pie -o glibc_ldl.out -std=c++11 -Wall -Wextra \
-pedantic-errors -fpic glibc_ldl.cpp -export-dynamic -ldl
./glibc_ldl.out
output:
1 0x40130a my_func_2() + 41
./glibc_ldl.out(_Z9my_func_2v+0x29) [0x40130a]
2 0x40139e my_func_1(int) + 16
./glibc_ldl.out(_Z9my_func_1i+0x10) [0x40139e]
3 0x4013b3 main + 18
./glibc_ldl.out(main+0x12) [0x4013b3]
4 0x7f7594552b97 __libc_start_main + 231
/lib/x86_64-linux-gnu/libc.so.6(__libc_start_main+0xe7) [0x7f7594552b97]
5 0x400f3a _start + 42
./glibc_ldl.out(_start+0x2a) [0x400f3a]
1 0x40130a my_func_2() + 41
./glibc_ldl.out(_Z9my_func_2v+0x29) [0x40130a]
2 0x40138b my_func_1(double) + 18
./glibc_ldl.out(_Z9my_func_1d+0x12) [0x40138b]
3 0x4013c8 main + 39
./glibc_ldl.out(main+0x27) [0x4013c8]
4 0x7f7594552b97 __libc_start_main + 231
/lib/x86_64-linux-gnu/libc.so.6(__libc_start_main+0xe7) [0x7f7594552b97]
5 0x400f3a _start + 42
./glibc_ldl.out(_start+0x2a) [0x400f3a]
Tested on Ubuntu 18.04.
glibc backtrace with C++ demangling hack 2: parse backtrace output
Shown at: https://panthema.net/2008/0901-stacktrace-demangled/
This is a hack because it requires parsing.
TODO get it to compile and show it here.
GDB scripting
We can also do this with GDB without recompiling by using: How to do an specific action when a certain breakpoint is hit in GDB?
We setup an empty backtrace function for our testing:
gdb/mystacktrace.h
void print_stacktrace(void) {}
and then with:
main.gdb
start
break print_stacktrace
commands
silent
backtrace
printf "\n"
continue
end
continue
we can run:
gdb -nh -batch -x main.gdb --args gdb.out
Sample output:
Temporary breakpoint 1 at 0x11a7: file main.cpp, line 21.
[Thread debugging using libthread_db enabled]
Using host libthread_db library "/lib/x86_64-linux-gnu/libthread_db.so.1".
Temporary breakpoint 1, main (argc=1, argv=0x7fffffffc3e8) at main.cpp:21
warning: Source file is more recent than executable.
21 if (argc > 1) {
Breakpoint 2 at 0x555555555151: file gdb/mystacktrace.h, line 1.
#0 print_stacktrace () at gdb/mystacktrace.h:1
#1 0x0000555555555161 in my_func_2 () at main.cpp:6
#2 0x0000555555555191 in my_func_1 (i=1) at main.cpp:16
#3 0x00005555555551ec in main (argc=1, argv=0x7fffffffc3e8) at main.cpp:27
[Inferior 1 (process 165453) exited normally]
The above can be made more usable with the following Bash function:
gdbbt() (
tmpfile=$(mktemp /tmp/gdbbt.XXXXXX)
fn="$1"
shift
printf '%s' "
start
break $fn
commands
silent
backtrace
printf \"\n\"
continue
end
continue
" > "$tmpfile"
gdb -nh -batch -x "$tmpfile" -args "$#"
rm -f "$tmpfile"
)
Usage:
gdbbt print_stacktrace gdb.out 2
I don't know how to make commands with -ex without the temporary file: Problems adding a breakpoint with commands from command line with ex command
Tested in Ubuntu 22.04, GDB 12.0.90.
GDB code injection
TODO this is the dream! It might allow for both compiled-liked speeds, but without the need to recompile! Either:
with compile code + one of the other options, ideally C++23 <stacktrace>: How to call assembly in gdb? Might already be possible. But compile code is mega-quirky so I'm lazy to even try
a built-in dbt command analogous to dprintf dynamic printf: How to do an specific action when a certain breakpoint is hit in GDB?
libunwind
TODO does this have any advantage over glibc backtrace? Very similar output, also requires modifying the build command, but not part of glibc so requires an extra package installation.
Code adapted from: https://eli.thegreenplace.net/2015/programmatic-access-to-the-call-stack-in-c/
main.c
/* This must be on top. */
#define _XOPEN_SOURCE 700
#include <stdio.h>
#include <stdlib.h>
/* Paste this on the file you want to debug. */
#define UNW_LOCAL_ONLY
#include <libunwind.h>
#include <stdio.h>
void print_trace() {
char sym[256];
unw_context_t context;
unw_cursor_t cursor;
unw_getcontext(&context);
unw_init_local(&cursor, &context);
while (unw_step(&cursor) > 0) {
unw_word_t offset, pc;
unw_get_reg(&cursor, UNW_REG_IP, &pc);
if (pc == 0) {
break;
}
printf("0x%lx:", pc);
if (unw_get_proc_name(&cursor, sym, sizeof(sym), &offset) == 0) {
printf(" (%s+0x%lx)\n", sym, offset);
} else {
printf(" -- error: unable to obtain symbol name for this frame\n");
}
}
puts("");
}
void my_func_3(void) {
print_trace();
}
void my_func_2(void) {
my_func_3();
}
void my_func_1(void) {
my_func_3();
}
int main(void) {
my_func_1(); /* line 46 */
my_func_2(); /* line 47 */
return 0;
}
Compile and run:
sudo apt-get install libunwind-dev
gcc -fno-pie -ggdb3 -O3 -no-pie -o main.out -std=c99 \
-Wall -Wextra -pedantic-errors main.c -lunwind
Either #define _XOPEN_SOURCE 700 must be on top, or we must use -std=gnu99:
Is the type `stack_t` no longer defined on linux?
Glibc - error in ucontext.h, but only with -std=c11
Run:
./main.out
Output:
0x4007db: (main+0xb)
0x7f4ff50aa830: (__libc_start_main+0xf0)
0x400819: (_start+0x29)
0x4007e2: (main+0x12)
0x7f4ff50aa830: (__libc_start_main+0xf0)
0x400819: (_start+0x29)
and:
addr2line -e main.out 0x4007db 0x4007e2
gives:
/home/ciro/main.c:34
/home/ciro/main.c:49
With -O0:
0x4009cf: (my_func_3+0xe)
0x4009e7: (my_func_1+0x9)
0x4009f3: (main+0x9)
0x7f7b84ad7830: (__libc_start_main+0xf0)
0x4007d9: (_start+0x29)
0x4009cf: (my_func_3+0xe)
0x4009db: (my_func_2+0x9)
0x4009f8: (main+0xe)
0x7f7b84ad7830: (__libc_start_main+0xf0)
0x4007d9: (_start+0x29)
and:
addr2line -e main.out 0x4009f3 0x4009f8
gives:
/home/ciro/main.c:47
/home/ciro/main.c:48
Tested on Ubuntu 16.04, GCC 6.4.0, libunwind 1.1.
libunwind with C++ name demangling
Code adapted from: https://eli.thegreenplace.net/2015/programmatic-access-to-the-call-stack-in-c/
unwind.cpp
#define UNW_LOCAL_ONLY
#include <cxxabi.h>
#include <libunwind.h>
#include <cstdio>
#include <cstdlib>
#include <iostream>
void backtrace() {
unw_cursor_t cursor;
unw_context_t context;
// Initialize cursor to current frame for local unwinding.
unw_getcontext(&context);
unw_init_local(&cursor, &context);
// Unwind frames one by one, going up the frame stack.
while (unw_step(&cursor) > 0) {
unw_word_t offset, pc;
unw_get_reg(&cursor, UNW_REG_IP, &pc);
if (pc == 0) {
break;
}
std::printf("0x%lx:", pc);
char sym[256];
if (unw_get_proc_name(&cursor, sym, sizeof(sym), &offset) == 0) {
char* nameptr = sym;
int status;
char* demangled = abi::__cxa_demangle(sym, nullptr, nullptr, &status);
if (status == 0) {
nameptr = demangled;
}
std::printf(" (%s+0x%lx)\n", nameptr, offset);
std::free(demangled);
} else {
std::printf(" -- error: unable to obtain symbol name for this frame\n");
}
}
}
void my_func_2(void) {
backtrace();
std::cout << std::endl; // line 43
}
void my_func_1(double f) {
(void)f;
my_func_2();
}
void my_func_1(int i) {
(void)i;
my_func_2();
} // line 54
int main() {
my_func_1(1);
my_func_1(2.0);
}
Compile and run:
sudo apt-get install libunwind-dev
g++ -fno-pie -ggdb3 -O0 -no-pie -o unwind.out -std=c++11 \
-Wall -Wextra -pedantic-errors unwind.cpp -lunwind -pthread
./unwind.out
Output:
0x400c80: (my_func_2()+0x9)
0x400cb7: (my_func_1(int)+0x10)
0x400ccc: (main+0x12)
0x7f4c68926b97: (__libc_start_main+0xe7)
0x400a3a: (_start+0x2a)
0x400c80: (my_func_2()+0x9)
0x400ca4: (my_func_1(double)+0x12)
0x400ce1: (main+0x27)
0x7f4c68926b97: (__libc_start_main+0xe7)
0x400a3a: (_start+0x2a)
and then we can find the lines of my_func_2 and my_func_1(int) with:
addr2line -e unwind.out 0x400c80 0x400cb7
which gives:
/home/ciro/test/unwind.cpp:43
/home/ciro/test/unwind.cpp:54
TODO: why are the lines off by one?
Tested on Ubuntu 18.04, GCC 7.4.0, libunwind 1.2.1.
Linux kernel
How to print the current thread stack trace inside the Linux kernel?
libdwfl
This was originally mentioned at: https://stackoverflow.com/a/60713161/895245 and it might be the best method, but I have to benchmark a bit more, but please go upvote that answer.
TODO: I tried to minimize the code in that answer, which was working, to a single function, but it is segfaulting, let me know if anyone can find why.
dwfl.cpp: answer reached 30k chars and this was the easiest cut: https://gist.github.com/cirosantilli/f1dd3ee5d324b9d24e40f855723544ac
Compile and run:
sudo apt install libdw-dev libunwind-dev
g++ -fno-pie -ggdb3 -O0 -no-pie -o dwfl.out -std=c++11 -Wall -Wextra -pedantic-errors dwfl.cpp -ldw -lunwind
./dwfl.out
We also need libunwind as that makes results more correct. If you do without it, it runs, but you will see that some of the lines are a bit wrong.
Output:
0: 0x402b72 stacktrace[abi:cxx11]() at /home/ciro/test/dwfl.cpp:65
1: 0x402cda my_func_2() at /home/ciro/test/dwfl.cpp:100
2: 0x402d76 my_func_1(int) at /home/ciro/test/dwfl.cpp:111
3: 0x402dd1 main at /home/ciro/test/dwfl.cpp:122
4: 0x7ff227ea0d8f __libc_start_call_main at ../sysdeps/nptl/libc_start_call_main.h:58
5: 0x7ff227ea0e3f __libc_start_main##GLIBC_2.34 at ../csu/libc-start.c:392
6: 0x402534 _start at ../csu/libc-start.c:-1
0: 0x402b72 stacktrace[abi:cxx11]() at /home/ciro/test/dwfl.cpp:65
1: 0x402cda my_func_2() at /home/ciro/test/dwfl.cpp:100
2: 0x402d5f my_func_1(double) at /home/ciro/test/dwfl.cpp:106
3: 0x402de2 main at /home/ciro/test/dwfl.cpp:123
4: 0x7ff227ea0d8f __libc_start_call_main at ../sysdeps/nptl/libc_start_call_main.h:58
5: 0x7ff227ea0e3f __libc_start_main##GLIBC_2.34 at ../csu/libc-start.c:392
6: 0x402534 _start at ../csu/libc-start.c:-1
Benchmark run:
g++ -fno-pie -ggdb3 -O3 -no-pie -o dwfl.out -std=c++11 -Wall -Wextra -pedantic-errors dwfl.cpp -ldw
time ./dwfl.out 1000 >/dev/null
Output:
real 0m3.751s
user 0m2.822s
sys 0m0.928s
So we see that this method is 10x faster than Boost's stacktrace, and might therefore be applicable to more use cases.
Tested in Ubuntu 22.04 amd64, libdw-dev 0.186, libunwind 1.3.2.
libbacktrace
https://github.com/ianlancetaylor/libbacktrace
Considering the harcore library author, it is worth trying this out, maybe it is The One. TODO check it out.
A C library that may be linked into a C/C++ program to produce symbolic backtraces
As of October 2020, libbacktrace supports ELF, PE/COFF, Mach-O, and XCOFF executables with DWARF debugging information. In other words, it supports GNU/Linux, *BSD, macOS, Windows, and AIX. The library is written to make it straightforward to add support for other object file and debugging formats.
The library relies on the C++ unwind API defined at https://itanium-cxx-abi.github.io/cxx-abi/abi-eh.html This API is provided by GCC and clang.
See also
How can one grab a stack trace in C?
How to make backtrace()/backtrace_symbols() print the function names?
Is there a portable/standard-compliant way to get filenames and linenumbers in a stack trace?
Best way to invoke gdb from inside program to print its stacktrace?
automatic stack trace on failure:
on C++ exception: C++ display stack trace on exception
generic: How to automatically generate a stacktrace when my program crashes
For a linux-only solution you can use backtrace(3) that simply returns an array of void * (in fact each of these point to the return address from the corresponding stack frame). To translate these to something of use, there's backtrace_symbols(3).
Pay attention to the notes section in backtrace(3):
The symbol names may be unavailable
without the use of special linker
options.
For systems using the GNU linker, it is necessary to use the
-rdynamic linker
option. Note that names of "static" functions are not exposed,
and won't be
available in the backtrace.
In C++23, there will be <stacktrace>, and then you can do:
#include <stacktrace>
/* ... */
std::cout << std::stacktrace::current();
Further details:
• https://en.cppreference.com/w/cpp/header/stacktrace
• https://en.cppreference.com/w/cpp/utility/basic_stacktrace/operator_ltlt
Is there any way to dump the call stack in a running process in C or C++ every time a certain function is called?
You can use a macro function instead of return statement in the specific function.
For example, instead of using return,
int foo(...)
{
if (error happened)
return -1;
... do something ...
return 0
}
You can use a macro function.
#include "c-callstack.h"
int foo(...)
{
if (error happened)
NL_RETURN(-1);
... do something ...
NL_RETURN(0);
}
Whenever an error happens in a function, you will see Java-style call stack as shown below.
Error(code:-1) at : so_topless_ranking_server (sample.c:23)
Error(code:-1) at : nanolat_database (sample.c:31)
Error(code:-1) at : nanolat_message_queue (sample.c:39)
Error(code:-1) at : main (sample.c:47)
Full source code is available here.
c-callstack at https://github.com/Nanolat
Linux specific, TLDR:
backtrace in glibc produces accurate stacktraces only when -lunwind is linked (undocumented platform-specific feature).
To output function name, source file and line number use #include <elfutils/libdwfl.h> (this library is documented only in its header file). backtrace_symbols and backtrace_symbolsd_fd are least informative.
On modern Linux your can get the stacktrace addresses using function backtrace. The undocumented way to make backtrace produce more accurate addresses on popular platforms is to link with -lunwind (libunwind-dev on Ubuntu 18.04) (see the example output below). backtrace uses function _Unwind_Backtrace and by default the latter comes from libgcc_s.so.1 and that implementation is most portable. When -lunwind is linked it provides a more accurate version of _Unwind_Backtrace but this library is less portable (see supported architectures in libunwind/src).
Unfortunately, the companion backtrace_symbolsd and backtrace_symbols_fd functions have not been able to resolve the stacktrace addresses to function names with source file name and line number for probably a decade now (see the example output below).
However, there is another method to resolve addresses to symbols and it produces the most useful traces with function name, source file and line number. The method is to #include <elfutils/libdwfl.h>and link with -ldw (libdw-dev on Ubuntu 18.04).
Working C++ example (test.cc):
#include <stdexcept>
#include <iostream>
#include <cassert>
#include <cstdlib>
#include <string>
#include <boost/core/demangle.hpp>
#include <execinfo.h>
#include <elfutils/libdwfl.h>
struct DebugInfoSession {
Dwfl_Callbacks callbacks = {};
char* debuginfo_path = nullptr;
Dwfl* dwfl = nullptr;
DebugInfoSession() {
callbacks.find_elf = dwfl_linux_proc_find_elf;
callbacks.find_debuginfo = dwfl_standard_find_debuginfo;
callbacks.debuginfo_path = &debuginfo_path;
dwfl = dwfl_begin(&callbacks);
assert(dwfl);
int r;
r = dwfl_linux_proc_report(dwfl, getpid());
assert(!r);
r = dwfl_report_end(dwfl, nullptr, nullptr);
assert(!r);
static_cast<void>(r);
}
~DebugInfoSession() {
dwfl_end(dwfl);
}
DebugInfoSession(DebugInfoSession const&) = delete;
DebugInfoSession& operator=(DebugInfoSession const&) = delete;
};
struct DebugInfo {
void* ip;
std::string function;
char const* file;
int line;
DebugInfo(DebugInfoSession const& dis, void* ip)
: ip(ip)
, file()
, line(-1)
{
// Get function name.
uintptr_t ip2 = reinterpret_cast<uintptr_t>(ip);
Dwfl_Module* module = dwfl_addrmodule(dis.dwfl, ip2);
char const* name = dwfl_module_addrname(module, ip2);
function = name ? boost::core::demangle(name) : "<unknown>";
// Get source filename and line number.
if(Dwfl_Line* dwfl_line = dwfl_module_getsrc(module, ip2)) {
Dwarf_Addr addr;
file = dwfl_lineinfo(dwfl_line, &addr, &line, nullptr, nullptr, nullptr);
}
}
};
std::ostream& operator<<(std::ostream& s, DebugInfo const& di) {
s << di.ip << ' ' << di.function;
if(di.file)
s << " at " << di.file << ':' << di.line;
return s;
}
void terminate_with_stacktrace() {
void* stack[512];
int stack_size = ::backtrace(stack, sizeof stack / sizeof *stack);
// Print the exception info, if any.
if(auto ex = std::current_exception()) {
try {
std::rethrow_exception(ex);
}
catch(std::exception& e) {
std::cerr << "Fatal exception " << boost::core::demangle(typeid(e).name()) << ": " << e.what() << ".\n";
}
catch(...) {
std::cerr << "Fatal unknown exception.\n";
}
}
DebugInfoSession dis;
std::cerr << "Stacktrace of " << stack_size << " frames:\n";
for(int i = 0; i < stack_size; ++i) {
std::cerr << i << ": " << DebugInfo(dis, stack[i]) << '\n';
}
std::cerr.flush();
std::_Exit(EXIT_FAILURE);
}
int main() {
std::set_terminate(terminate_with_stacktrace);
throw std::runtime_error("test exception");
}
Compiled on Ubuntu 18.04.4 LTS with gcc-8.3:
g++ -o test.o -c -m{arch,tune}=native -std=gnu++17 -W{all,extra,error} -g -Og -fstack-protector-all test.cc
g++ -o test -g test.o -ldw -lunwind
Outputs:
Fatal exception std::runtime_error: test exception.
Stacktrace of 7 frames:
0: 0x55f3837c1a8c terminate_with_stacktrace() at /home/max/src/test/test.cc:76
1: 0x7fbc1c845ae5 <unknown>
2: 0x7fbc1c845b20 std::terminate()
3: 0x7fbc1c845d53 __cxa_throw
4: 0x55f3837c1a43 main at /home/max/src/test/test.cc:103
5: 0x7fbc1c3e3b96 __libc_start_main at ../csu/libc-start.c:310
6: 0x55f3837c17e9 _start
When no -lunwind is linked, it produces a less accurate stacktrace:
0: 0x5591dd9d1a4d terminate_with_stacktrace() at /home/max/src/test/test.cc:76
1: 0x7f3c18ad6ae6 <unknown>
2: 0x7f3c18ad6b21 <unknown>
3: 0x7f3c18ad6d54 <unknown>
4: 0x5591dd9d1a04 main at /home/max/src/test/test.cc:103
5: 0x7f3c1845cb97 __libc_start_main at ../csu/libc-start.c:344
6: 0x5591dd9d17aa _start
For comparison, backtrace_symbols_fd output for the same stacktrace is least informative:
/home/max/src/test/debug/gcc/test(+0x192f)[0x5601c5a2092f]
/usr/lib/x86_64-linux-gnu/libstdc++.so.6(+0x92ae5)[0x7f95184f5ae5]
/usr/lib/x86_64-linux-gnu/libstdc++.so.6(_ZSt9terminatev+0x10)[0x7f95184f5b20]
/usr/lib/x86_64-linux-gnu/libstdc++.so.6(__cxa_throw+0x43)[0x7f95184f5d53]
/home/max/src/test/debug/gcc/test(+0x1ae7)[0x5601c5a20ae7]
/lib/x86_64-linux-gnu/libc.so.6(__libc_start_main+0xe6)[0x7f9518093b96]
/home/max/src/test/debug/gcc/test(+0x1849)[0x5601c5a20849]
In a production version (as well as C language version) you may like to make this code extra robust by replacing boost::core::demangle, std::string and std::cout with their underlying calls.
You can also override __cxa_throw to capture the stacktrace when an exception is thrown and print it when the exception is caught. By the time it enters catch block the stack has been unwound, so it is too late to call backtrace, and this is why the stack must be captured on throw which is implemented by function __cxa_throw. Note that in a multi-threaded program __cxa_throw can be called simultaneously by multiple threads, so that if it captures the stacktrace into a global array that must be thread_local.
You can also make the stack trace printing function async-signal safe, so that you can invoke it directly from your SIGSEGV, SIGBUS signal handlers (which should use their own stacks for robustness). Obtaining function name, source file and line number using libdwfl from a signal handler may fail because it is not async-signal safe or if the address space of the process has been substantially corrupted, but in practice it succeeds 99% of the time (I haven't seen it fail).
To summarize, a complete production-ready library for automatic stacktrace output should:
Capture the stacktrace on throw into thread-specific storage.
Automatically print the stacktrace on unhandled exceptions.
Print the stacktrace in async-signal-safe manner.
Provide a robust signal handler function which uses its own stack that prints the stacktrace in a async-signal-safe manner. The user can install this function as a signal handler for SIGSEGV, SIGBUS, SIGFPE, etc..
The signal handler may as well print the values of all CPU registers at the point of the fault from ucontext_t signal function argument (may be excluding vector registers), a-la Linux kernel oops log messages.
Another answer to an old thread.
When I need to do this, I usually just use system() and pstack
So something like this:
#include <sys/types.h>
#include <unistd.h>
#include <string>
#include <sstream>
#include <cstdlib>
void f()
{
pid_t myPid = getpid();
std::string pstackCommand = "pstack ";
std::stringstream ss;
ss << myPid;
pstackCommand += ss.str();
system(pstackCommand.c_str());
}
void g()
{
f();
}
void h()
{
g();
}
int main()
{
h();
}
This outputs
#0 0x00002aaaab62d61e in waitpid () from /lib64/libc.so.6
#1 0x00002aaaab5bf609 in do_system () from /lib64/libc.so.6
#2 0x0000000000400c3c in f() ()
#3 0x0000000000400cc5 in g() ()
#4 0x0000000000400cd1 in h() ()
#5 0x0000000000400cdd in main ()
This should work on Linux, FreeBSD and Solaris. I don't think that macOS has pstack or a simple equivalent, but this thread seems to have an alternative.
If you are using C, then you will need to use C string functions.
#include <sys/types.h>
#include <unistd.h>
#include <stdlib.h>
#include <stdio.h>
void f()
{
pid_t myPid = getpid();
/*
length of command 7 for 'pstack ', 7 for the PID, 1 for nul
*/
char pstackCommand[7+7+1];
sprintf(pstackCommand, "pstack %d", (int)myPid);
system(pstackCommand);
}
I've used 7 for the max number of digits in the PID, based on this post.
There is no standardized way to do that. For windows the functionality is provided in the DbgHelp library
You can use the Boost libraries to print the current callstack.
#include <boost/stacktrace.hpp>
// ... somewhere inside the `bar(int)` function that is called recursively:
std::cout << boost::stacktrace::stacktrace();
Man here: https://www.boost.org/doc/libs/1_65_1/doc/html/stacktrace.html
I know this thread is old, but I think it can be useful for other people. If you are using gcc, you can use its instrument features (-finstrument-functions option) to log any function call (entry and exit). Have a look at this for more information: http://hacktalks.blogspot.fr/2013/08/gcc-instrument-functions.html
You can thus for instance push and pop every calls into a stack, and when you want to print it, you just look at what you have in your stack.
I've tested it, it works perfectly and is very handy
UPDATE: you can also find information about the -finstrument-functions compile option in the GCC doc concerning the Instrumentation options: https://gcc.gnu.org/onlinedocs/gcc/Instrumentation-Options.html
You can implement the functionality yourself:
Use a global (string)stack and at start of each function push the function name and such other values (eg parameters) onto this stack; at exit of function pop it again.
Write a function that will printout the stack content when it is called, and use this in the function where you want to see the callstack.
This may sound like a lot of work but is quite useful.
Of course the next question is: will this be enough ?
The main disadvantage of stack-traces is that why you have the precise function being called you do not have anything else, like the value of its arguments, which is very useful for debugging.
If you have access to gcc and gdb, I would suggest using assert to check for a specific condition, and produce a memory dump if it is not met. Of course this means the process will stop, but you'll have a full fledged report instead of a mere stack-trace.
If you wish for a less obtrusive way, you can always use logging. There are very efficient logging facilities out there, like Pantheios for example. Which once again could give you a much more accurate image of what is going on.
You can use Poppy for this. It is normally used to gather the stack trace during a crash but it can also output it for a running program as well.
Now here's the good part: it can output the actual parameter values for each function on the stack, and even local variables, loop counters, etc.
You can use the GNU profiler. It shows the call-graph as well! the command is gprof and you need to compile your code with some option.
Is there any way to dump the call stack in a running process in C or C++ every time a certain function is called?
No there is not, although platform-dependent solutions might exist.
I'll explain:
Let's say I'm interested in replacing the rand() function used by a certain application.
So I attach gdb to this process and make it load my custom shared library (which has a customized rand() function):
call (int) dlopen("path_to_library/asdf.so")
This would place the customized rand() function inside the process' memory. However, at this point the symbol rand will still point to the default rand() function. Is there a way to make gdb point the symbol to the new rand() function, forcing the process to use my version?
I must say I'm also not allowed to use the LD_PRELOAD (linux) nor DYLD_INSERT_LIBRARIES (mac os x) methods for this, because they allow code injection only in the beginning of the program execution.
The application that I would like to replace rand(), starts several threads and some of them start new processes, and I'm interested in injecting code on one of these new processes. As I mentioned above, GDB is great for this purpose because it allows code injection into a specific process.
I followed this post and this presentation and came up with the following set of gdb commands for OSX with x86-64 executable, which can be loaded with -x option when attaching to the process:
set $s = dyld_stub_rand
set $p = ($s+6+*(int*)($s+2))
call (void*)dlsym((void*)dlopen("myrand.dylib"), "my_rand")
set *(void**)$p = my_rand
c
The magic is in set $p = ... command. dyld_stub_rand is a 6-byte jump instruction. Jump offset is at dyld_stub_rand+2 (4 bytes). This is a $rip-relative jump, so add offset to what $rip would be at this point (right after the instruction, dyld_stub_rand+6).
This points to a symbol table entry, which should be either real rand or dynamic linker routine to load it (if it was never called). It is then replaced by my_rand.
Sometimes gdb will pick up dyld_stub_rand from libSystem or another shared library, if that happens, unload them first with remove-symbol-file before running other commands.
This question intrigued me, so I did a little research. What you are looking for is a 'dll injection'. You write a function to replace some library function, put it in a .so, and tell ld to preload your dll. I just tried it out and it worked great! I realize this doesn't really answer your question in relation to gdb, but I think it offers a viable workaround.
For a gdb-only solution, see my other solution.
// -*- compile-command: "gcc -Wall -ggdb -o test test.c"; -*-
// test.c
#include "stdio.h"
#include "stdlib.h"
int main(int argc, char** argv)
{
//should print a fairly random number...
printf("Super random number: %d\n", rand());
return 0;
}
/ -*- compile-command: "gcc -Wall -fPIC -shared my_rand.c -o my_rand.so"; -*-
//my_rand.c
int rand(void)
{
return 42;
}
compile both files, then run:
LD_PRELOAD="./my_rand.so" ./test
Super random number: 42
I have a new solution, based on the new original constraints. (I am not deleting my first answer, as others may find it useful.)
I have been doing a bunch of research, and I think it would work with a bit more fiddling.
In your .so rename your replacement rand function, e.g my_rand
Compile everything and load up gdb
Use info functions to find the address of rand in the symbol table
Use dlopen then dlsym to load the function into memory and get its address
call (int) dlopen("my_rand.so", 1) -> -val-
call (unsigned int) dlsym(-val-, "my_rand") -> my_rand_addr
-the tricky part- Find the hex code of a jumpq 0x*my_rand_addr* instruction
Use set {int}*rand_addr* = *my_rand_addr* to change symbol table instruction
Continue execution: now whenever rand is called, it will jump to my_rand instead
This is a bit complicated, and very round-about, but I'm pretty sure it would work. The only thing I haven't accomplished yet is creating the jumpq instruction code. Everything up until that point works fine.
I'm not sure how to do this in a running program, but perhaps LD_PRELOAD will work for you. If you set this environment variable to a list of shared objects, the runtime loader will load the shared object early in the process and allow the functions in it to take precedence over others.
LD_PRELOAD=path_to_library/asdf.so path/to/prog
You do have to do this before you start the process but you don't have to rebuild the program.
Several of the answers here and the code injection article you linked to in your answer cover chunks of what I consider the optimal gdb-oriented solution, but none of them pull it all together or cover all the points. The code-expression of the solution is a bit long, so here's a summary of the important steps:
Load the code to inject. Most of the answers posted here use what I consider the best approach -- call dlopen() in the inferior process to link in a shared library containing the injected code. In the article you linked to the author instead loaded a relocatable object file and hand-linked it against the inferior. This is quite frankly insane -- relocatable objects are not "ready-to-run" and include relocations even for internal references. And hand-linking is tedious and error-prone -- far simpler to let the real runtime dynamic linker do the work. This does mean getting libdl into the process in the first place, but there are many options for doing that.
Create a detour. Most of the answers posted here so far have involved locating the PLT entry for the function of interest, using that to find the matching GOT entry, then modifying the GOT entry to point to your injected function. This is fine up to a point, but certain linker features -- e.g., use of dlsym -- can circumvent the GOT and provide direct access to the function of interest. The only way to be certain of intercepting all calls to a particular function is overwrite the initial instructions of that function's code in-memory to create a "detour" redirecting execution to your injected function.
Create a trampoline (optional). Frequently when doing this sort of injection you'll want to call the original function whose invocation you are intercepting. The way to allow this with a function detour is to create a small code "trampoline" which includes the overwritten instructions of the original function then a jump to the remainder of the original. This can be complex, because any IP-relative instructions in the copied set need to be modified to account for their new addresses.
Automate it all. These steps can be tedious, even if doing some of the simpler solutions posted in other answers. The best way to ensure that the steps are done correctly every time with variable parameters (injecting different functions, etc) is to automate their execution. Starting with the 7.0 series, gdb has included the ability to write new commands in Python. This support can be used to implement a turn-key solution for injecting and detouring code in/to the inferior process.
Here's an example. I have the same a and b executables as before and an inject2.so created from the following code:
#include <unistd.h>
#include <stdio.h>
int (*rand__)(void) = NULL;
int
rand(void)
{
int result = rand__();
printf("rand invoked! result = %d\n", result);
return result % 47;
}
I can then place my Python detour command in detour.py and have the following gdb session:
(gdb) source detour.py
(gdb) exec-file a
(gdb) set follow-fork-mode child
(gdb) catch exec
Catchpoint 1 (exec)
(gdb) run
Starting program: /home/llasram/ws/detour/a
a: 1933263113
a: 831502921
[New process 8500]
b: 918844931
process 8500 is executing new program: /home/llasram/ws/detour/b
[Switching to process 8500]
Catchpoint 1 (exec'd /home/llasram/ws/detour/b), 0x00007ffff7ddfaf0 in _start ()
from /lib64/ld-linux-x86-64.so.2
(gdb) break main
Breakpoint 2 at 0x4005d0: file b.c, line 7.
(gdb) cont
Continuing.
Breakpoint 2, main (argc=1, argv=0x7fffffffdd68) at b.c:7
7 {
(gdb) detour libc.so.6:rand inject2.so:rand inject2.so:rand__
(gdb) cont
Continuing.
rand invoked! result = 392103444
b: 22
Program exited normally.
In the child process, I create a detour from the rand() function in libc.so.6 to the rand() function in inject2.so and store a pointer to a trampoline for the original rand() in the rand__ variable of inject2.so. And as expected, the injected code calls the original, displays the full result, and returns that result modulo 47.
Due to length, I'm just linking to a pastie containing the code for my detour command. This is a fairly superficial implementation (especially in terms of the trampoline generation), but it should work well in a large percentage of cases. I've tested it with gdb 7.2 (most recently released version) on Linux with both 32-bit and 64-bit executables. I haven't tested it on OS X, but any differences should be relatively minor.
For executables you can easily find the address where the function pointer is stored by using objdump. For example:
objdump -R /bin/bash | grep write
00000000006db558 R_X86_64_JUMP_SLOT fwrite
00000000006db5a0 R_X86_64_JUMP_SLOT write
Therefore, 0x6db5a0 is the adress of the pointer for write. If you change it, calls to write will be redirected to your chosen function. Loading new libraries in gdb and getting function pointers has been covered in earlier posts. The executable and every library have their own pointers. Replacing affects only the module whose pointer was changed.
For libraries, you need to find the base address of the library and add it to the address given by objdump. In Linux, /proc/<pid>/maps gives it out. I don't know whether position-independent executables with address randomization would work. maps-information might be unavailable in such cases.
As long as the function you want to replace is in a shared library, you can redirect calls to that function at runtime (during debugging) by poking at the PLT. Here is an article that might be helpful:
Shared library call redirection using ELF PLT infection
It's written from the standpoint of malware modifying a program, but a much easier procedure is adaptable to live use in the debugger. Basically you just need to find the function's entry in the PLT and overwrite the address with the address of the function you want to replace it with.
Googling for "PLT" along with terms like "ELF", "shared library", "dynamic linking", "PIC", etc. might find you more details on the subject.
You can still us LD_PRELOAD if you make the preloaded function understand the situations it's getting used in. Here is an example that will use the rand() as normal, except inside a forked process when it will always return 42. I use the dl routines to load the standard library's rand() function into a function pointer for use by the hijacked rand().
// -*- compile-command: "gcc -Wall -fPIC -shared my_rand.c -o my_rand.so -ldl"; -*-
//my_rand.c
#include <sys/types.h>
#include <unistd.h>
#include <dlfcn.h>
int pid = 0;
int (*real_rand)(void) = NULL;
void f(void) __attribute__ ((constructor));
void f(void) {
pid = getpid();
void* dl = dlopen("libc.so.6", RTLD_LAZY);
if(dl) {
real_rand = dlsym(dl, "rand");
}
}
int rand(void)
{
if(pid == getpid() && real_rand)
return real_rand();
else
return 42;
}
//test.c
#include <dlfcn.h>
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
int main(int argc, char** argv)
{
printf("Super random number: %d\n", rand());
if(fork()) {
printf("original process rand: %d\n", rand());
} else {
printf("forked process rand: %d\n", rand());
}
return 0;
}
jdizzle#pudding:~$ ./test
Super random number: 1804289383
original process rand: 846930886
forked process rand: 846930886
jdizzle#pudding:~$ LD_PRELOAD="/lib/ld-linux.so.2 ./my_rand.so" ./test
Super random number: 1804289383
original process rand: 846930886
forked process rand: 42
I found this tutorial incredibly useful, and so far its the only way I managed to achieve what I was looking with GDB: Code Injection into Running Linux Application: http://www.codeproject.com/KB/DLL/code_injection.aspx
There is also a good Q&A on code injection for Mac here: http://www.mikeash.com/pyblog/friday-qa-2009-01-30-code-injection.html
I frequently use code injection as a method of mocking for automated testing of C code. If that's the sort of situation you're in -- if your use of GDB is simply because you're not interested in the parent processes, and not because you want to interactively select the processes which are of interest -- then you can still use LD_PRELOAD to achieve your solution. Your injected code just needs to determine whether it is in the parent or child processes. There are several ways you could do this, but on Linux, since your child processes exec(), the simplest is probably to look at the active executable image.
I produced two executables, one named a and the other b. Executable a prints the result of calling rand() twice, then fork()s and exec()s b twice. Executable b print the result of calling rand() once. I use LD_PRELOAD to inject the result of compiling the following code into the executables:
// -*- compile-command: "gcc -D_GNU_SOURCE=1 -Wall -std=gnu99 -O2 -pipe -fPIC -shared -o inject.so inject.c"; -*-
#include <sys/types.h>
#include <unistd.h>
#include <limits.h>
#include <stdio.h>
#include <dlfcn.h>
#define constructor __attribute__((__constructor__))
typedef int (*rand_t)(void);
typedef enum {
UNKNOWN,
PARENT,
CHILD
} state_t;
state_t state = UNKNOWN;
rand_t rand__ = NULL;
state_t
determine_state(void)
{
pid_t pid = getpid();
char linkpath[PATH_MAX] = { 0, };
char exepath[PATH_MAX] = { 0, };
ssize_t exesz = 0;
snprintf(linkpath, PATH_MAX, "/proc/%d/exe", pid);
exesz = readlink(linkpath, exepath, PATH_MAX);
if (exesz < 0)
return UNKNOWN;
switch (exepath[exesz - 1]) {
case 'a':
return PARENT;
case 'b':
return CHILD;
}
return UNKNOWN;
}
int
rand(void)
{
if (state == CHILD)
return 47;
return rand__();
}
constructor static void
inject_init(void)
{
rand__ = dlsym(RTLD_NEXT, "rand");
state = determine_state();
}
The result of running a with and without injection:
$ ./a
a: 644034683
a: 2011954203
b: 375870504
b: 1222326746
$ LD_PRELOAD=$PWD/inject.so ./a
a: 1023059566
a: 986551064
b: 47
b: 47
I'll post a gdb-oriented solution later.