feat:Prevent segmentation fault in probe_read & probe_write #367
Conversation
runtime/src/bpf_helper.cpp
Outdated
| sigemptyset(&sa.sa_mask); // clear signal set | ||
| sa.sa_flags = 0; | ||
|
|
||
| if (sigaction(SIGSEGV, &sa, &old_sa) == -1) { |
There was a problem hiding this comment.
Do we have any method to optimize these syscalls? For now, each call to bpftime_probe_read may lead to a series syscalls, which are relatively slow.
Maybe we can setup signal handlers only once, and in signal handler, walk stack trace to dertermine whether a call to bpftime_probe_read triggers this signal, and determine to let the call fail or crash.
Will this be too hard?
There was a problem hiding this comment.
I will make a try for this idea later.
There was a problem hiding this comment.
You can try to use https://github.com/eunomia-bpf/bpftime/blob/master/benchmark%2FREADME.md to see how it performs.
runtime/src/bpf_helper.cpp
Outdated
| } | ||
|
|
||
| // restore the origin signal handler | ||
| if (sigaction(SIGSEGV, &old_sa, NULL) == -1) { |
There was a problem hiding this comment.
Use nullptr in C++, don't use NULL
runtime/unit-test/test_probe.cpp
Outdated
| uint64_t, uint64_t); | ||
|
|
||
| // prepare for future use | ||
| long bpftime_strncmp(const char *s1, uint64_t s1_sz, const char *s2); |
There was a problem hiding this comment.
Maybe you don't need to copy these unused declarations now, I'll sync my fork with your once yours is merged, because my PR still has a long way to go
Description
We avoid core dump when accessing an illegal address (such as NULL). Now we will get different error code for such situation.
Type of change