mem_watcher:添加对slab分配速率和总字节数的监控

eBPF_Supermarket/Memory_Subsystem/mem_watcher/Makefile

@@ -24,7 +24,7 @@ INCLUDES := -I$(OUTPUT) -I../../libbpf/include/uapi -I$(dir $(VMLINUX)) -I$(LIBB
 CFLAGS := -g -Wall
 ALL_LDFLAGS := $(LDFLAGS) $(EXTRA_LDFLAGS)
 
-APPS = paf pr procstat sysstat memleak fraginfo vmasnap drsnoop oomkiller numafraginfo
+APPS = paf pr procstat sysstat memleak fraginfo vmasnap drsnoop oomkiller numafraginfo slabrate
 
 TARGETS= mem_watcher
 CARGO ?= $(shell which cargo)

eBPF_Supermarket/Memory_Subsystem/mem_watcher/README.md

@@ -9,6 +9,10 @@
 **eBPF 提供了一种高效的机制来监控和追踪系统级别的事件，包括内存的分配和释放。通过 eBPF，可以跟踪内存分配和释放的请求，并收集每次分配的调用堆栈。然后，分析这些信息，找出执行了内存分配但未执行释放操作的调用堆栈，这有助于程序员找出导致内存泄漏的源头。**
 
 ---
+## TODO list
+
+- [x] 监控SLAB分配器的内存使用情况
+- [ ] 跟踪共享内存的用量信息
 
 ## 背景意义
 
@@ -58,7 +62,7 @@
    make
    ```
    make后没有编译生成任何的二进制文件，只打印了logo，效果如下：
-   ![alt text](/docs/image/15.png)
+   ![alt text](../docs/image/15.png)
 
 打开makefile，检查makefile逻辑，代码如下：
 ```
@@ -82,7 +86,7 @@ else
 BZS_APPS := 
 ```
 再次执行make，发现报错为 "vmlinux.h file not find"，如下：
-   ![alt text](/docs/image/16.png)
+   ![alt text](../docs/image/16.png)
 
 执行以下命令,生成vmlinux.h文件
 ```
@@ -125,15 +129,15 @@ registry = "git://crates.rustcc.cn/crates.io-index"
 ```
    重新安装还是会报错：
 
-   ![alt text](/docs/image/17.png)
+   ![alt text](../docs/image/17.png)
 
    在 `~/.cargo/config` 文件中添加以下内容，即可解决：
    ```
    [net]
 git-fetch-with-cli = true
 ```
 再次make编译完成，生成二进制文件 mem_watcher，并能正常运行。
-   ![alt text](/docs/image/18.png)
+   ![alt text](../docs/image/18.png)
 
 # 工具的使用方法说明
 

eBPF_Supermarket/Memory_Subsystem/mem_watcher/bpf/slabrate.bpf.c

@@ -0,0 +1,51 @@
+#include <vmlinux.h>
+#include <bpf/bpf_helpers.h>
+#include <bpf/bpf_core_read.h>
+#include <bpf/bpf_tracing.h>
+#include "mem_watcher.h"
+
+#define MAX_ENTRIES	10240
+
+const volatile pid_t target_pid = 0;
+
+static struct slabrate_info slab_zero_value = {};
+
+struct {
+	__uint(type, BPF_MAP_TYPE_HASH);
+	__uint(max_entries, MAX_ENTRIES);
+	__type(key, char *);
+	__type(value, struct slabrate_info);
+} slab_entries SEC(".maps");
+
+static int probe_entry(struct kmem_cache *cachep)
+{
+	__u64 pid_tgid = bpf_get_current_pid_tgid();
+	__u32 pid = pid_tgid >> 32;
+	struct slabrate_info *valuep;
+	const char *name = BPF_CORE_READ(cachep, name);
+
+	if (target_pid && target_pid != pid)
+		return 0;
+
+	valuep = bpf_map_lookup_elem(&slab_entries, &name);
+	if (!valuep) {
+		bpf_map_update_elem(&slab_entries, &name, &slab_zero_value, BPF_ANY);
+		valuep = bpf_map_lookup_elem(&slab_entries, &name);
+		if (!valuep)
+			return 0;
+		bpf_probe_read_kernel(&valuep->name, sizeof(valuep->name), name);
+	}
+
+	valuep->count++;
+	valuep->size += BPF_CORE_READ(cachep, size);
+
+	return 0;
+}
+
+SEC("kprobe/kmem_cache_alloc")
+int BPF_KPROBE(kmem_cache_alloc, struct kmem_cache *cachep)
+{
+	return probe_entry(cachep);
+}
+
+char LICENSE[] SEC("license") = "Dual BSD/GPL";

eBPF_Supermarket/Memory_Subsystem/mem_watcher/include/mem_watcher.h

@@ -200,4 +200,13 @@ struct event {
     char comm[TASK_COMM_LEN];  // 被杀死进程的命令名
 };
 
+/* slabrate.h */
+#define CACHE_NAME_SIZE 32
+
+struct slabrate_info {
+	char name[CACHE_NAME_SIZE];
+	__u64 count;
+	__u64 size;
+};
+
 #endif /* __MEM_WATCHER_H */

eBPF_Supermarket/Memory_Subsystem/mem_watcher/mem_watcher.c

@@ -37,6 +37,7 @@
 #include "memleak.skel.h"
 #include "vmasnap.skel.h"
 #include "drsnoop.skel.h"
+#include "slabrate.skel.h"
 
 #include "mem_watcher.h"
 #include "fraginfo.h"
@@ -122,6 +123,13 @@ struct order_entry
 	struct ctg_info oinfo;
 };
 
+#define OUTPUT_ROWS_LIMIT 10240
+
+static pid_t target_pid = 0;
+static bool clear_screen = true;
+static int output_rows = 20;
+static int count = 99999999;
+
 int compare_entries(const void *a, const void *b)
 {
 	struct order_entry *entryA = (struct order_entry *)a;
@@ -204,7 +212,6 @@ static volatile bool exiting = false;
 		}                                                   \
 	}
 
-// 为 oomkiller 使用的宏，指定 map_name
 #define LOAD_AND_ATTACH_SKELETON_WITH_MAP(skel, event, map_name)                                \
 	do                                                                                         \
 	{                                                                                          \
@@ -230,7 +237,6 @@ static volatile bool exiting = false;
 		}                                                                                      \
 	} while (0)
 
-// 保留原有逻辑的宏
 #define LOAD_AND_ATTACH_SKELETON(skel, event)                                                   \
 	do                                                                                         \
 	{                                                                                          \
@@ -273,7 +279,8 @@ static struct env
     int interval;        // 打印间隔，单位为秒
     int duration;        // 运行时长，单位为秒
     bool part2;          // 是否启用系统内存状态报告的扩展部分
-    bool oomkiller;      // 是否启用oomkiller事件处理
+	bool oomkiller;      // 是否启用oomkiller事件处理
+	bool slabrate;
 
     long choose_pid;     // 选择的进程号
     bool rss;            // 是否打印进程页面信息
@@ -292,8 +299,9 @@ static struct env
     .print_time = false,   // 默认不打印地址申请时间
     .rss = false,          // 默认不打印进程页面信息
     .part2 = false,        // 默认关闭系统内存状态报告的扩展部分
-    .oomkiller = false,    // 默认关闭oomkiller事件处理
-    .choose_pid = 0,       // 默认不选择特定进程
+	.oomkiller = false,    // 默认关闭oomkiller事件处理
+	.slabrate = false,
+	.choose_pid = 0,       // 默认不选择特定进程
     .interval = 1,         // 默认打印间隔为1秒
     .duration = 10,        // 默认持续运行10秒
 };
@@ -343,7 +351,9 @@ static const struct argp_option opts[] = {
 	{"oomkiller", 'o', 0, 0, "print oomkiller (内存不足时被杀死的进程信息)"},
 	{0, 0, 0, 0, "numafraginfo:", 16},
 	{"numafraginfo", 'N', 0, 0, "print numafraginfo"},
-
+
+	{0, 0, 0, 0, "slabrate:", 17},
+	{"slabrate", 'e', 0, 0, "print slabrate"},
 
 	{NULL, 'h', NULL, OPTION_HIDDEN, "show the full help"},
 	{0},
@@ -409,6 +419,9 @@ static error_t parse_arg(int key, char *arg, struct argp_state *state)
 	case 'N':
 		env.numafraginfo = true;
 		break;
+	case 'e':
+		env.slabrate = true;
+		break;
 	default:
 		return ARGP_ERR_UNKNOWN;
 	}
@@ -449,6 +462,7 @@ static int process_numafraginfo(struct numafraginfo_bpf *skel_numafraginfo);
 static int process_vmasnap(struct vmasnap_bpf *skel_vmasnap);
 static int process_drsnoop(struct drsnoop_bpf *skel_drsnoop);
 static int process_oomkiller(struct oomkiller_bpf *skel_oomkiller);  // 新增的oomkiller处理函数原型
+static int process_slabrate(struct slabrate_bpf *skel_slabrate);
 static int handle_event_oomkiller(void *ctx, void *data, size_t data_sz);  // 新增的oomkiller事件处理函数
 static __u64 adjust_time_to_program_start_time(__u64 first_query_time);
 static int update_addr_times(struct memleak_bpf *skel_memleak);
@@ -469,7 +483,8 @@ int main(int argc, char **argv)
 	struct numafraginfo_bpf *skel_numafraginfo;
     struct vmasnap_bpf *skel_vmasnap;
     struct oomkiller_bpf *skel_oomkiller;
-    struct drsnoop_bpf *skel_drsnoop;
+	struct drsnoop_bpf *skel_drsnoop;
+	struct slabrate_bpf *skel_slabrate;
 
     err = argp_parse(&argp, argc, argv, 0, NULL, NULL);
     if (err)
@@ -538,6 +553,51 @@ int main(int argc, char **argv)
 	{
 		PROCESS_SKEL(skel_drsnoop, drsnoop);
 	}
+	else if (env.slabrate)
+	{
+		PROCESS_SKEL(skel_slabrate, slabrate);
+	}
+	else
+	{
+		fprintf(stderr, "No valid option was given\n");
+		return 1;
+	}
+
+	if (env.paf)
+	{
+		err = process_paf(skel_paf);
+	}
+	else if (env.pr)
+	{
+		err = process_pr(skel_pr);
+	}
+	else if (env.procstat)
+	{
+		err = process_procstat(skel_procstat);
+	}
+	else if (env.fraginfo)
+	{
+		err = process_fraginfo(skel_fraginfo);
+	}
+	else if(env.numafraginfo){
+		err = process_numafraginfo(skel_numafraginfo);
+	}
+	else if (env.vmasnap)
+	{
+		err = process_vmasnap(skel_vmasnap);
+	}
+	else if (env.sysstat)
+	{
+		err = process_sysstat(skel_sysstat);
+	}
+	else if (env.memleak)
+	{
+		err = process_memleak(skel_memleak, env);
+	}
+	else if (env.oomkiller)  // 处理 oomkiller
+	{
+		err = process_oomkiller(skel_oomkiller);  // 使用处理 oom
+	}
 
     return 0;
 }
@@ -1706,3 +1766,106 @@ static int process_drsnoop(struct drsnoop_bpf *skel_drsnoop) {
 
     return err < 0 ? -err : 0;
 }
+
+static int sort_column(const void *obj1, const void *obj2)
+{
+	struct slabrate_info *s1 = (struct slabrate_info *)obj1;
+	struct slabrate_info *s2 = (struct slabrate_info *)obj2;
+
+	return s2->size - s1->size;
+}
+
+static int print_stat(struct slabrate_bpf *obj)
+{
+	char *key, **prev_key = NULL;
+	static struct slabrate_info values[OUTPUT_ROWS_LIMIT];
+	int i, err = 0, rows = 0;
+	int fd = bpf_map__fd(obj->maps.slab_entries);
+
+	printf("%-32s %6s %10s\n", "CACHE", "ALLOCS", "BYTES");
+
+	while (1) {
+		err = bpf_map_get_next_key(fd, prev_key, &key);
+		if (err) {
+			if (errno == ENOENT) {
+				err = 0;
+				break;
+			}
+			return err;
+		}
+		err = bpf_map_lookup_elem(fd, &key, &values[rows++]);
+		if (err) {
+			return err;
+		}
+		prev_key = &key;
+	}
+
+	qsort(values, rows, sizeof(struct slabrate_info), sort_column);
+	rows = rows < output_rows ? rows : output_rows;
+	for (i = 0; i < rows; i++)
+		printf("%-32s %6lld %10lld\n",
+		       values[i].name, values[i].count, values[i].size);
+
+	printf("\n");
+	prev_key = NULL;
+
+	while (1) {
+		err = bpf_map_get_next_key(fd, prev_key, &key);
+		if (err) {
+			if (errno == ENOENT) {
+				err = 0;
+				break;
+			}
+			return err;
+		}
+		err = bpf_map_delete_elem(fd, &key);
+		if (err) {
+			return err;
+		}
+		prev_key = &key;
+	}
+	return err;
+}
+
+static int process_slabrate(struct slabrate_bpf *skel_slabrate) {
+	int err;
+
+	bpf_program__set_autoload(skel_slabrate->progs.kmem_cache_alloc, true);
+
+	skel_slabrate->rodata->target_pid = target_pid;
+
+	err = slabrate_bpf__load(skel_slabrate);
+    if (err) {
+        fprintf(stderr, "Failed to load and verify BPF skeleton\n");
+        return 1;
+    }
+
+    err = slabrate_bpf__attach(skel_slabrate);
+    if (err) {
+        fprintf(stderr, "Failed to attach BPF skeleton\n");
+        goto slabrate_cleanup;
+    }
+
+	while (1) {
+		sleep(1);
+
+		if (clear_screen) {
+			err = system("clear");
+			if (err)
+				goto slabrate_cleanup;
+		}
+
+		err = print_stat(skel_slabrate);
+		if (err)
+			goto slabrate_cleanup;
+
+		count--;
+		if (exiting || !count)
+			goto slabrate_cleanup;
+	}
+
+slabrate_cleanup:
+	slabrate_bpf__destroy(skel_slabrate);
+
+	return err != 0;
+}