refactor: Remove old profiler regex

Author: rvql

agent/src/ebpf/mod.rs

@@ -615,45 +615,6 @@ extern "C" {
      */
     pub fn continuous_profiler_running() -> bool;
 
-    /*
-     * To set the regex matching for the profiler.
-     *
-     * Perform regular expression matching on process names.
-     * Processes that successfully match the regular expression are
-     * aggregated using the key:
-     *     `{pid + stime + u_stack_id + k_stack_id + tid + cpu}`.
-     *
-     * For processes that do not match, they are aggregated using the
-     * key:
-     *     `<process name + u_stack_id + k_stack_id + cpu>`.
-     *
-     * Using regular expressions, we match process names to establish
-     * symbol table caching for specific processes, rather than enabling
-     * symbol caching for all processes. This approach aims to reduce
-     * memory usage.
-     *
-     * For example:
-     *
-     *  "^(java|nginx|profiler|telegraf|mysqld|socket_tracer|.*deepflow.*)$"
-     *
-     * This regex pattern matches the process names: "java", "nginx", "profiler",
-     * "telegraf", "mysqld", "socket_tracer", and any process containing the
-     * substring "deepflow".
-     *
-     * By using this interface, you can customize the regular expression to
-     * match specific process names that you want to establish symbol table
-     * caching for, providing flexibility and control over which processes will
-     * have symbol caching enabled. This allows you to finetune the memory usage
-     * and profiling behavior of the profiler according to your requirements.
-     *
-     * The default expression is empty (''), indicating that no profile data will
-     * be generated.
-     *
-     * @pattern : Regular expression pattern. e.g. "^(java|nginx|.*ser.*)$"
-     * @returns 0 on success, < 0 on error
-     */
-    pub fn set_profiler_regex(pattern: *const c_char) -> c_int;
-
     /*
      * This interface is used to set whether CPUID should be included in the
      * aggregation of stack trace data.
@@ -744,8 +705,6 @@ extern "C" {
 
     cfg_if::cfg_if! {
         if #[cfg(feature = "extended_profile")] {
-            pub fn set_offcpu_profiler_regex(pattern: *const c_char) -> c_int;
-
             pub fn enable_offcpu_profiler() -> c_int;
 
             pub fn disable_offcpu_profiler() -> c_int;
@@ -756,8 +715,6 @@ extern "C" {
                 block_time: c_uint,
             ) -> c_int;
 
-            pub fn set_memory_profiler_regex(pattern: *const c_char) -> c_int;
-
             pub fn enable_memory_profiler() -> c_int;
 
             pub fn disable_memory_profiler() -> c_int;

agent/src/ebpf/samples/rust/socket-tracer/src/main.rs

@@ -637,7 +637,8 @@ fn main() {
         //    ::std::process::exit(1);
         //}
 
-        //set_profiler_regex(
+        //set_feature_regex(
+        //    ebpf::FEATURE_PROFILE_ONCPU,
         //    CString::new(
         //        "^(java|nginx|profiler|telegraf|mysqld|.*deepflow.*|socket_tracer)$".as_bytes(),
         //    )

agent/src/ebpf/user/extended/extended.c

@@ -59,7 +59,7 @@ void __attribute__ ((weak)) extended_process_exit(int pid) {}
 
 void __attribute__ ((weak)) extended_match_pid_handle(int feat, int pid, enum match_pids_act act) {}
 
-bool __attribute__ ((weak)) extended_require_dwarf(int pid, const char *name)
+bool __attribute__ ((weak)) extended_require_dwarf(int pid, const char *path)
 {
 	return false;
 }

agent/src/ebpf/user/extended/extended.h

@@ -83,8 +83,8 @@ void extended_match_pid_handle(int feat, int pid, enum match_pids_act act);
 /**
  * @brief **extended_requires_dwarf()** whether extended profilers require DWARF unwinding
  * @param pid Process ID
- * @param name Process executable name
+ * @param name Process executable path
  */
-bool extended_require_dwarf(int pid, const char *name);
+bool extended_require_dwarf(int pid, const char *path);
 
 #endif /* DF_EXTENDED_H */

agent/src/ebpf/user/profile/perf_profiler.c

@@ -523,10 +523,6 @@ int stop_continuous_profiler(void *cb_ctx[PROFILER_CTX_NUM])
 
 	u64 alloc_b, free_b;
 	get_mem_stat(&alloc_b, &free_b);
-	if (oncpu_ctx.regex_existed) {
-		regfree(&oncpu_ctx.profiler_regex);
-		oncpu_ctx.regex_existed = false;
-	}
 
 	ebpf_info(LOG_CP_TAG "== alloc_b %lu bytes, free_b %lu bytes, "
 		  "use %lu bytes ==\n", alloc_b, free_b, alloc_b - free_b);
@@ -876,48 +872,6 @@ void release_flame_graph_hash(void)
 		  flame_graph_start_time, flame_graph_end_time);
 }
 
-/*
- * To set the regex matching for the profiler. 
- *
- * @pattern : Regular expression pattern. e.g. "^(java|nginx|.*ser.*)$"
- * @returns 0 on success, < 0 on error
- */
-int set_profiler_regex(const char *pattern)
-{
-	if (profiler_tracer == NULL) {
-		ebpf_warning(LOG_CP_TAG
-			     "The 'profiler_tracer' has not been created yet."
-			     " Please use start_continuous_profiler() to create it first.\n");
-		return (-1);
-	}
-
-	if (!g_enable_oncpu) {
-		ebpf_warning(LOG_CP_TAG
-			     "'profiler_regex' cannot be set while on-CPU is currently disabled.\n");
-		return (-1);
-	}
-
-	/*
-	 * During the data processing, the thread responsible for matching reads the
-	 * regular expression, while the thread handling the regular expression upd-
-	 * ates is different. Synchronization is implemented to ensure protection and
-	 * coordination between these two threads.
-	 */
-	profile_regex_lock(&oncpu_ctx);
-	do_profiler_regex_config(pattern, &oncpu_ctx);
-	profile_regex_unlock(&oncpu_ctx);
-
-	unwind_process_reload();
-
-	ebpf_info(LOG_CP_TAG "Set 'profiler_regex' successful, pattern : '%s'",
-		  pattern);
-	return (0);
-}
-
-bool check_oncpu_profiler_regex(const char *name) {
-	return check_profiler_regex(&oncpu_ctx, name);
-}
-
 int set_profiler_cpu_aggregation(int flag)
 {
 	if (flag != 0 && flag != 1) {
@@ -1026,20 +980,11 @@ void release_flame_graph_hash(void)
 	return;
 }
 
-int set_profiler_regex(const char *pattern)
-{
-	return (-1);
-}
-
 int set_profiler_cpu_aggregation(int flag)
 {
 	return (-1);
 }
 
-bool check_oncpu_profiler_regex(const char *name) {
-	return false;
-}
-
 struct bpf_tracer *get_profiler_tracer(void)
 {
 	return NULL;

agent/src/ebpf/user/profile/perf_profiler.h

@@ -200,9 +200,7 @@ int start_continuous_profiler(int freq, int java_syms_update_delay,
 			      tracer_callback_t callback, void *cb_ctx[PROFILER_CTX_NUM]);
 void process_stack_trace_data_for_flame_graph(stack_trace_msg_t * val);
 void release_flame_graph_hash(void);
-int set_profiler_regex(const char *pattern);
 int set_profiler_cpu_aggregation(int flag);
-bool check_oncpu_profiler_regex(const char *name);
 struct bpf_tracer *get_profiler_tracer(void);
 void set_enable_perf_sample(struct bpf_tracer *t, u64 enable_flag);
 void cpdbg_process(stack_trace_msg_t * msg);

agent/src/ebpf/user/profile/profile_common.c

@@ -131,7 +131,6 @@ int profiler_context_init(struct profiler_context *ctx,
 	snprintf(ctx->custom_stack_map_b.name, sizeof(ctx->custom_stack_map_b.name), "%s",
 		 custom_stack_map_name_b);
 
-	ctx->regex_existed = false;
 	ctx->only_matched_data = only_matched;
 	ctx->use_delta_time = use_delta_time;
 	ctx->type = type;
@@ -162,37 +161,6 @@ void set_bpf_run_enabled(struct bpf_tracer *t, struct profiler_context *ctx,
 		  enable_flag);
 }
 
-int do_profiler_regex_config(const char *pattern, struct profiler_context *ctx)
-{
-	if (*pattern == '\0') {
-		ctx->regex_existed = false;
-		ebpf_warning("%sSet 'profiler_regex' pattern : '', an empty"
-			     " regular expression will not generate any stack data."
-			     "Please configure the regular expression for profiler.\n",
-			     ctx->tag);
-		return (0);
-	}
-
-	if (ctx->regex_existed) {
-		regfree(&ctx->profiler_regex);
-	}
-
-	int ret = regcomp(&ctx->profiler_regex, pattern, REG_EXTENDED);
-	if (ret != 0) {
-		char error_buffer[100];
-		regerror(ret, &ctx->profiler_regex, error_buffer,
-			 sizeof(error_buffer));
-		ebpf_warning("%sPattern %s failed to compile the regular "
-			     "expression: %s\n", ctx->tag, pattern,
-			     error_buffer);
-		ctx->regex_existed = false;
-		return (-1);
-	}
-
-	ctx->regex_existed = true;
-	return 0;
-}
-
 static bool check_kallsyms_addr_is_zero(void)
 {
 	const int check_num = 100;
@@ -1268,34 +1236,6 @@ void process_bpf_stacktraces(struct profiler_context *ctx, struct bpf_tracer *t)
 	push_and_release_stack_trace_msg(ctx, &ctx->msg_hash, false);
 }
 
-bool check_profiler_regex(struct profiler_context *ctx, const char *name)
-{
-	bool matched = false;
-	for (int i = 0; i < ARRAY_SIZE(g_ctx_array); i++) {
-		if (g_ctx_array[i] == NULL)
-			continue;
-
-		if (ctx != NULL && ctx->name != NULL) {
-			if (strcmp(g_ctx_array[i]->name, ctx->name))
-				continue;
-		}
-
-		if (g_ctx_array[i]->regex_existed) {
-			profile_regex_lock(g_ctx_array[i]);
-			matched =
-			    (regexec
-			     (&g_ctx_array[i]->profiler_regex, name, 0, NULL, 0)
-			     == 0);
-			profile_regex_unlock(g_ctx_array[i]);
-			if (matched) {
-				return true;
-			}
-		}
-	}
-
-	return false;
-}
-
 bool profiler_is_running(void)
 {
 	for (int i = 0; i < ARRAY_SIZE(g_ctx_array); i++) {

agent/src/ebpf/user/profile/profile_common.h

@@ -66,21 +66,6 @@ struct profiler_context {
 	 */
 	volatile u64 cpu_aggregation_flag;
 
-	/*
-	 * To perform regular expression matching on process names,
-	 * the 'profiler_regex' is set using the 'set_profiler_regex()'
-	 * interface. Processes that successfully match the regular
-	 * expression are aggregated using the key:
-	 * `{pid + stime + u_stack_id + k_stack_id + tid + cpu}`.
-	 *
-	 * For processes that do not match, they are aggregated using the
-	 * key:
-	 * `<process name + u_stack_id + k_stack_id + cpu>`.
-	 */
-	regex_t profiler_regex;
-	bool regex_existed;
-	volatile u32 regex_lock;
-
 	/*
 	 * The profiler stop flag, with 1 indicating stop and
 	 * 0 indicating running status.
@@ -147,20 +132,8 @@ struct profiler_context {
 	void *callback_ctx;
 };
 
-static inline void profile_regex_lock(struct profiler_context *c)
-{
-	while (__atomic_test_and_set(&c->regex_lock, __ATOMIC_ACQUIRE))
-		CLIB_PAUSE();
-}
-
-static inline void profile_regex_unlock(struct profiler_context *c)
-{
-	__atomic_clear(&c->regex_lock, __ATOMIC_RELEASE);
-}
-
 void process_bpf_stacktraces(struct profiler_context *ctx,
 			     struct bpf_tracer *t);
-int do_profiler_regex_config(const char *pattern, struct profiler_context *ctx);
 void set_bpf_run_enabled(struct bpf_tracer *t, struct profiler_context *ctx,
 			 u64 enable_flag);
 int profiler_context_init(struct profiler_context *ctx, const char *name,
@@ -177,16 +150,6 @@ int java_libs_and_tools_install(void);
 void push_and_release_stack_trace_msg(struct profiler_context *ctx,
 				      stack_trace_msg_hash_t * h,
 				      bool is_force);
-/**
- * @brief **check_profiler_regex()** check if the process name matches
- * the regular expression.
- *
- * @param ctx Profiler context(runtime environment status & configuration)
- * address
- * @param name Process name to be checked
- * @return true successfully passed the check, false otherwise
- */
-bool check_profiler_regex(struct profiler_context *ctx, const char *name);
 // Check if the profiler is currently running.
 bool profiler_is_running(void);
 #endif /*DF_USER_PROFILE_COMMON_H */

agent/src/ebpf/user/tracer.c

@@ -38,6 +38,7 @@
 #include "elf.h"
 #include "load.h"
 #include "mem.h"
+#include "unwind_tracer.h"
 #include "extended/extended.h"
 
 uint32_t k_version;
@@ -1707,6 +1708,10 @@ int set_feature_regex(int feature, const char *pattern)
 	}
 
 	cfg_feature_regex_array[feature].ok = true;
+
+	if (feature == FEATURE_PROFILE_ONCPU || feature == FEATURE_PROFILE_OFFCPU || feature == FEATURE_PROFILE_MEMORY) {
+		unwind_process_reload();
+	}
 	return 0;
 }
 

agent/src/ebpf/user/unwind_tracer.c

@@ -174,16 +174,17 @@ static bool requires_dwarf_unwind_table(int pid) {
     if (path == NULL) {
         return false;
     }
-    char *exe_name = basename(path);
-    // Java has JIT compiled code without DWARF info, not supported at the moment
-    if (strcmp(exe_name, "java") == 0) {
+
+    // TBD: offcpu
+    bool need_unwind = is_feature_matched(FEATURE_PROFILE_ONCPU, pid, path) || extended_require_dwarf(pid, path);
+    if (!need_unwind) {
         free(path);
         return false;
     }
 
-    // TBD: offcpu
-    bool need_unwind = (oncpu_profiler_enabled() && check_oncpu_profiler_regex(exe_name)) || extended_require_dwarf(pid, exe_name);
-    if (!need_unwind) {
+    char *exe_name = basename(path);
+    // Java has JIT compiled code without DWARF info, not supported at the moment
+    if (strcmp(exe_name, "java") == 0) {
         free(path);
         return false;
     }

agent/src/ebpf_dispatcher.rs

@@ -849,7 +849,8 @@ impl EbpfCollector {
             }
 
             if !on_cpu.disabled {
-                ebpf::set_profiler_regex(
+                ebpf::set_feature_regex(
+                    ebpf::FEATURE_PROFILE_ONCPU,
                     CString::new(on_cpu.regex.as_bytes())
                         .unwrap()
                         .as_c_str()
@@ -863,7 +864,8 @@ impl EbpfCollector {
             #[cfg(feature = "extended_profile")]
             {
                 if !off_cpu.disabled {
-                    ebpf::set_offcpu_profiler_regex(
+                    ebpf::set_feature_regex(
+                        ebpf::FEATURE_PROFILE_OFFCPU,
                         CString::new(off_cpu.regex.as_bytes())
                             .unwrap()
                             .as_c_str()
@@ -875,7 +877,8 @@ impl EbpfCollector {
                 }
 
                 if !memory.disabled {
-                    ebpf::set_memory_profiler_regex(
+                    ebpf::set_feature_regex(
+                        ebpf::FEATURE_PROFILE_MEMORY,
                         CString::new(memory.regex.as_bytes())
                             .unwrap()
                             .as_c_str()