fprobe, samples: Add use_trace option and show hit/missed counter
authorMasami Hiramatsu (Google) <mhiramat@kernel.org>
Tue, 7 Jun 2022 16:11:02 +0000 (01:11 +0900)
committerDaniel Borkmann <daniel@iogearbox.net>
Fri, 17 Jun 2022 19:53:29 +0000 (21:53 +0200)
Add use_trace option to use trace_printk() instead of pr_info()
so that the handler doesn't involve the RCU operations.
And show the hit and missed counter so that the user can check
how many times the probe handler hit and missed.

Signed-off-by: Masami Hiramatsu (Google) <mhiramat@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Link: https://lore.kernel.org/bpf/165461826247.280167.11939123218334322352.stgit@devnote2
samples/fprobe/fprobe_example.c

index 24d3cf1..01ee6c8 100644 (file)
@@ -21,6 +21,7 @@
 #define BACKTRACE_DEPTH 16
 #define MAX_SYMBOL_LEN 4096
 struct fprobe sample_probe;
+static unsigned long nhit;
 
 static char symbol[MAX_SYMBOL_LEN] = "kernel_clone";
 module_param_string(symbol, symbol, sizeof(symbol), 0644);
@@ -28,6 +29,8 @@ static char nosymbol[MAX_SYMBOL_LEN] = "";
 module_param_string(nosymbol, nosymbol, sizeof(nosymbol), 0644);
 static bool stackdump = true;
 module_param(stackdump, bool, 0644);
+static bool use_trace = false;
+module_param(use_trace, bool, 0644);
 
 static void show_backtrace(void)
 {
@@ -40,7 +43,15 @@ static void show_backtrace(void)
 
 static void sample_entry_handler(struct fprobe *fp, unsigned long ip, struct pt_regs *regs)
 {
-       pr_info("Enter <%pS> ip = 0x%p\n", (void *)ip, (void *)ip);
+       if (use_trace)
+               /*
+                * This is just an example, no kernel code should call
+                * trace_printk() except when actively debugging.
+                */
+               trace_printk("Enter <%pS> ip = 0x%p\n", (void *)ip, (void *)ip);
+       else
+               pr_info("Enter <%pS> ip = 0x%p\n", (void *)ip, (void *)ip);
+       nhit++;
        if (stackdump)
                show_backtrace();
 }
@@ -49,8 +60,17 @@ static void sample_exit_handler(struct fprobe *fp, unsigned long ip, struct pt_r
 {
        unsigned long rip = instruction_pointer(regs);
 
-       pr_info("Return from <%pS> ip = 0x%p to rip = 0x%p (%pS)\n",
-               (void *)ip, (void *)ip, (void *)rip, (void *)rip);
+       if (use_trace)
+               /*
+                * This is just an example, no kernel code should call
+                * trace_printk() except when actively debugging.
+                */
+               trace_printk("Return from <%pS> ip = 0x%p to rip = 0x%p (%pS)\n",
+                       (void *)ip, (void *)ip, (void *)rip, (void *)rip);
+       else
+               pr_info("Return from <%pS> ip = 0x%p to rip = 0x%p (%pS)\n",
+                       (void *)ip, (void *)ip, (void *)rip, (void *)rip);
+       nhit++;
        if (stackdump)
                show_backtrace();
 }
@@ -112,7 +132,8 @@ static void __exit fprobe_exit(void)
 {
        unregister_fprobe(&sample_probe);
 
-       pr_info("fprobe at %s unregistered\n", symbol);
+       pr_info("fprobe at %s unregistered. %ld times hit, %ld times missed\n",
+               symbol, nhit, sample_probe.nmissed);
 }
 
 module_init(fprobe_init)