argdist: print process name COMM (#4633)

* Print expression to print process name
* Update documentation to add usage of COMM agrument

tools/argdist.py

@@ -21,7 +21,7 @@
 class Probe(object):
         next_probe_index = 0
         streq_index = 0
-        aliases = {"$PID": "(bpf_get_current_pid_tgid() >> 32)"}
+        aliases = {"$PID": "(bpf_get_current_pid_tgid() >> 32)", "$COMM": "&val.name"}
 
         def _substitute_aliases(self, expr):
                 if expr is None:
@@ -125,6 +125,17 @@ def _generate_retprobe_prefix(self):
                         text += "if (%s == 0) { return 0 ; }\n" % val_name
                         self.param_val_names[pname] = val_name
                 return text
+
+        def _generate_comm_prefix(self):
+                text = """
+struct val_t {
+        u32 pid;
+        char name[sizeof(struct __string_t)];
+};
+struct val_t val = {.pid = (bpf_get_current_pid_tgid() >> 32) };
+bpf_get_current_comm(&val.name, sizeof(val.name));
+        """
+                return text
 
         def _replace_entry_exprs(self):
                 for pname, vname in self.param_val_names.items():
@@ -397,6 +408,10 @@ def generate_text(self):
                         # signatures. Other probes force it to ().
                         signature = ", " + self.signature
 
+                # If COMM is specified prefix with code to get process name
+                if self.exprs.count(self.aliases['$COMM']):
+                        prefix += self._generate_comm_prefix()
+
                 program += probe_text.replace("PROBENAME",
                                               self.probe_func_name)
                 program = program.replace("SIGNATURE", signature)
@@ -568,6 +583,9 @@ class Tool(object):
 argdist -C 'r::__vfs_read():u32:$PID:$latency > 100000'
         Print frequency of reads by process where the latency was >0.1ms
 
+argdist -C 'r::__vfs_read():u32:$COMM:$latency > 100000'
+        Print frequency of reads by process name where the latency was >0.1ms
+
 argdist -H 'r::__vfs_read(void *file, void *buf, size_t count):size_t:
             $entry(count):$latency > 1000000'
         Print a histogram of read sizes that were longer than 1ms

tools/argdist_example.txt

@@ -203,6 +203,24 @@ r::__vfs_read():u32:$PID:$latency > 100000
 
 It looks like process 2780 performed 21 slow reads.
 
+You can print the name of the process. This is helpful for short lived processes
+and for easier identification of processes response. For example, we can identify
+the process using the epoll I/O multiplexing system call
+
+# ./argdist -C 't:syscalls:sys_exit_epoll_wait():char*:$COMM'
+[19:57:56]
+t:syscalls:sys_exit_epoll_wait():char*:$COMM
+	COUNT      EVENT
+	4          $COMM = b'node'
+[19:57:57]
+t:syscalls:sys_exit_epoll_wait():char*:$COMM
+	COUNT      EVENT
+	2          $COMM = b'open5gs-sgwud'
+	3          $COMM = b'open5gs-sgwcd'
+	3          $COMM = b'open5gs-nrfd'
+	3          $COMM = b'open5gs-udmd'
+	4          $COMM = b'open5gs-scpd'
+
 Occasionally, entry parameter values are also interesting. For example, you
 might be curious how long it takes malloc() to allocate memory -- nanoseconds
 per byte allocated. Let's go:
@@ -413,6 +431,9 @@ argdist -p 1005 -H 'r:c:read()'
 argdist -C 'r::__vfs_read():u32:$PID:$latency > 100000'
         Print frequency of reads by process where the latency was >0.1ms
 
+argdist -C 'r::__vfs_read():u32:$COMM:$latency > 100000'
+        Print frequency of reads by process name where the latency was >0.1ms
+
 argdist -H 'r::__vfs_read(void *file, void *buf, size_t count):size_t:$entry(count):$latency > 1000000' 
         Print a histogram of read sizes that were longer than 1ms