packages/gala-gopher
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
gala-gopher/fix-bugs-for-ksliprobe-pgsliprobe-stackprobe.patch
wo_cow a430f297a0 fix issues raised during 22.03 test
2022-11-28 11:43:21 +08:00

350 lines
13 KiB
Diff
Raw Blame History

From 72946403263a0ac056a65f85fd19736d200c8aa4 Mon Sep 17 00:00:00 2001
From: wo_cow <niuqianqian@huawei.com>
Date: Tue, 22 Nov 2022 15:40:11 +0800
Subject: [PATCH] fix bugs for ksliprobe pgsliprobe stackprobe
---
src/common/common.h | 1 +
src/common/util.c | 19 +++++
.../ebpf.probe/src/ksliprobe/ksliprobe.bpf.c | 4 +
.../ebpf.probe/src/pgsliprobe/pgsliprobe.c | 74 +++++++++----------
.../src/pgsliprobe/pgsliprobe_bpf.h | 7 +-
.../src/stackprobe/conf/stackprobe.conf | 2 +-
.../src/stackprobe/stack_bpf/memleak.bpf.c | 11 +--
.../extends/ebpf.probe/src/stackprobe/svg.c | 31 +++++++-
.../pg_stat.probe/pg_stat_probe.py | 8 +-
src/probes/system_infos.probe/system_procs.c | 2 +-
10 files changed, 105 insertions(+), 54 deletions(-)
diff --git a/src/common/common.h b/src/common/common.h
index 4ce6a81..49496c4 100644
--- a/src/common/common.h
+++ b/src/common/common.h
@@ -168,6 +168,7 @@ unsigned short ntohs(unsigned short netshort);
/* get uprobe func offset */
int get_func_offset(char *proc_name, char *func_name, char *bin_file_path);
+char *get_cur_date(void);
char *get_cur_time(void);
void ip_str(unsigned int family, unsigned char *ip, unsigned char *ip_str, unsigned int ip_str_size);
diff --git a/src/common/util.c b/src/common/util.c
index 65e5d4d..dbe5f75 100644
--- a/src/common/util.c
+++ b/src/common/util.c
@@ -21,6 +21,25 @@
#include <stdarg.h>
#include "common.h"
+char *get_cur_date(void)
+{
+ /* return date str, ex: 2021/5/17 */
+ static char tm[TM_STR_LEN] = {0};
+ struct tm *tmp_ptr = NULL;
+ time_t t;
+
+ (void)time(&t);
+
+ tmp_ptr = localtime(&t);
+ (void)snprintf(tm,
+ TM_STR_LEN,
+ "%d-%d-%d",
+ (1900 + tmp_ptr->tm_year),
+ (1 + tmp_ptr->tm_mon),
+ tmp_ptr->tm_mday);
+ return tm;
+}
+
char *get_cur_time(void)
{
/* return time str, ex: 2021/5/17 19:56:03 */
diff --git a/src/probes/extends/ebpf.probe/src/ksliprobe/ksliprobe.bpf.c b/src/probes/extends/ebpf.probe/src/ksliprobe/ksliprobe.bpf.c
index bf653f0..6fa64ab 100644
--- a/src/probes/extends/ebpf.probe/src/ksliprobe/ksliprobe.bpf.c
+++ b/src/probes/extends/ebpf.probe/src/ksliprobe/ksliprobe.bpf.c
@@ -374,6 +374,10 @@ static __always_inline void process_rd_msg(u32 tgid, int fd, const char *buf, co
#ifndef KERNEL_SUPPORT_TSTAMP
csd->start_ts_nsec = ts_nsec;
+#else
+ if (csd->start_ts_nsec == 0) {
+ csd->start_ts_nsec = ts_nsec;
+ }
#endif
csd->status = SAMP_READ_READY;
diff --git a/src/probes/extends/ebpf.probe/src/pgsliprobe/pgsliprobe.c b/src/probes/extends/ebpf.probe/src/pgsliprobe/pgsliprobe.c
index 9171985..6b75f06 100644
--- a/src/probes/extends/ebpf.probe/src/pgsliprobe/pgsliprobe.c
+++ b/src/probes/extends/ebpf.probe/src/pgsliprobe/pgsliprobe.c
@@ -180,7 +180,6 @@ static void *msg_event_receiver(void *arg)
}
poll_pb(pb, params.period * 1000);
-
stop = 1;
return NULL;
}
@@ -380,6 +379,7 @@ int main(int argc, char **argv)
{
int err, ret;
FILE *fp = NULL;
+ int init = 0;
struct bpf_link_hash_t *item, *tmp;
err = args_parse(argc, argv, &params);
@@ -409,15 +409,6 @@ int main(int argc, char **argv)
goto init_err;
}
- load_args(GET_MAP_FD(pgsli_kprobe, args_map), &params);
- err = init_conn_mgt_process(GET_MAP_FD(pgsli_kprobe, output));
- if (err != 0) {
- fprintf(stderr, "Init connection management process failed.\n");
- goto init_err;
- }
-
- printf("pgsliprobe probe successfully started!\n");
-
while (!stop) {
sleep(params.period);
if (noDependLibssl) {
@@ -426,37 +417,46 @@ int main(int argc, char **argv)
set_bpf_link_inactive();
if (add_bpf_link_by_search_pids() != SLI_OK) {
- if (noDependLibssl) {
- continue;
+ if (!noDependLibssl) {
+ goto init_err;
}
- goto init_err;
- }
-
- // attach to libssl
- H_ITER(head, item, tmp) {
- if (item->v.pid_state == PID_ELF_TOBE_ATTACHED) {
- UBPF_ATTACH_ONELINK(pgsli_uprobe, SSL_read, item->v.elf_path, SSL_read,
- item->v.bpf_link_read, ret);
- if (ret <= 0) {
- fprintf(stderr, "Can't attach function SSL_read at elf_path %s.\n", item->v.elf_path);
- goto init_err;
- }
- UBPF_RET_ATTACH_ONELINK(pgsli_uprobe, SSL_read, item->v.elf_path, SSL_read,
- item->v.bpf_link_read_ret, ret);
- if (ret <= 0) {
- fprintf(stderr, "Can't attach ret function SSL_read at elf_path %s.\n", item->v.elf_path);
- goto init_err;
+ } else {
+ // attach to libssl
+ H_ITER(head, item, tmp) {
+ if (item->v.pid_state == PID_ELF_TOBE_ATTACHED) {
+ UBPF_ATTACH_ONELINK(pgsli_uprobe, SSL_read, item->v.elf_path, SSL_read,
+ item->v.bpf_link_read, ret);
+ if (ret <= 0) {
+ fprintf(stderr, "Can't attach function SSL_read at elf_path %s.\n", item->v.elf_path);
+ goto init_err;
+ }
+ UBPF_RET_ATTACH_ONELINK(pgsli_uprobe, SSL_read, item->v.elf_path, SSL_read,
+ item->v.bpf_link_read_ret, ret);
+ if (ret <= 0) {
+ fprintf(stderr, "Can't attach ret function SSL_read at elf_path %s.\n", item->v.elf_path);
+ goto init_err;
+ }
+ UBPF_ATTACH_ONELINK(pgsli_uprobe, SSL_write, item->v.elf_path, SSL_write,
+ item->v.bpf_link_write, ret);
+ if (ret <= 0) {
+ fprintf(stderr, "Can't attach function SSL_write at elf_path %s.\n", item->v.elf_path);
+ goto init_err;
+ }
+ item->v.pid_state = PID_ELF_ATTACHED;
}
- UBPF_ATTACH_ONELINK(pgsli_uprobe, SSL_write, item->v.elf_path, SSL_write,
- item->v.bpf_link_write, ret);
- if (ret <= 0) {
- fprintf(stderr, "Can't attach function SSL_write at elf_path %s.\n", item->v.elf_path);
- goto init_err;
- }
- item->v.pid_state = PID_ELF_ATTACHED;
}
+ clear_invalid_bpf_link();
+ }
+ if (init == 0) {
+ load_args(GET_MAP_FD(pgsli_kprobe, args_map), &params);
+ err = init_conn_mgt_process(GET_MAP_FD(pgsli_kprobe, output));
+ if (err != 0) {
+ fprintf(stderr, "Init connection management process failed.\n");
+ goto init_err;
+ }
+ printf("pgsliprobe probe successfully started!\n");
+ init = 1;
}
- clear_invalid_bpf_link();
}
init_err:
diff --git a/src/probes/extends/ebpf.probe/src/pgsliprobe/pgsliprobe_bpf.h b/src/probes/extends/ebpf.probe/src/pgsliprobe/pgsliprobe_bpf.h
index a2e85b2..a3dfc5d 100644
--- a/src/probes/extends/ebpf.probe/src/pgsliprobe/pgsliprobe_bpf.h
+++ b/src/probes/extends/ebpf.probe/src/pgsliprobe/pgsliprobe_bpf.h
@@ -156,7 +156,7 @@ static __always_inline void periodic_report(u64 ts_nsec, struct conn_data_t *con
if (ts_nsec > conn_data->last_report_ts_nsec &&
ts_nsec - conn_data->last_report_ts_nsec >= period) {
// rtt larger than period is considered an invalid value
- if (conn_data->latency.rtt_nsec < period && conn_data->max.rtt_nsec < period) {
+ if (conn_data->latency.rtt_nsec < period * 2 && conn_data->max.rtt_nsec < period * 2) {
struct msg_event_data_t msg_evt_data = {0};
msg_evt_data.tgid = conn_key->tgid;
msg_evt_data.fd = conn_key->fd;
@@ -224,10 +224,13 @@ static __always_inline void process_rdwr_msg(int fd, const char *buf, int count,
}
csd->req_cmd = cmd;
+#ifndef KERNEL_SUPPORT_TSTAMP
+ csd->start_ts_nsec = ts_nsec;
+#else
if (csd->start_ts_nsec == 0) {
csd->start_ts_nsec = ts_nsec;
}
-
+#endif
csd->status = SAMP_READ_READY;
} else { // MSG_WRITE
if (csd->status == SAMP_READ_READY) {
diff --git a/src/probes/extends/ebpf.probe/src/stackprobe/conf/stackprobe.conf b/src/probes/extends/ebpf.probe/src/stackprobe/conf/stackprobe.conf
index f0b3c29..a2edd5d 100644
--- a/src/probes/extends/ebpf.probe/src/stackprobe/conf/stackprobe.conf
+++ b/src/probes/extends/ebpf.probe/src/stackprobe/conf/stackprobe.conf
@@ -12,7 +12,7 @@ flame_name =
oncpu = true;
offcpu = false;
io = false;
- memleak = true;
+ memleak = false;
};
application =
diff --git a/src/probes/extends/ebpf.probe/src/stackprobe/stack_bpf/memleak.bpf.c b/src/probes/extends/ebpf.probe/src/stackprobe/stack_bpf/memleak.bpf.c
index e855401..fe2f22b 100644
--- a/src/probes/extends/ebpf.probe/src/stackprobe/stack_bpf/memleak.bpf.c
+++ b/src/probes/extends/ebpf.probe/src/stackprobe/stack_bpf/memleak.bpf.c
@@ -109,20 +109,12 @@ struct bpf_map_def SEC("maps") mmap_allocs = {
};
struct bpf_map_def SEC("maps") brk_allocs = {
- .type = BPF_MAP_TYPE_HASH,
+ .type = BPF_MAP_TYPE_LRU_HASH,
.key_size = sizeof(u32), // tgid
.value_size = sizeof(struct brk_info_t),
.max_entries = 1000000,
};
-// allocated memory for the process
-struct bpf_map_def SEC("maps") combined_allocs = {
- .type = BPF_MAP_TYPE_HASH,
- .key_size = sizeof(struct stack_id_s),
- .value_size = sizeof(struct combined_alloc_info_t),
- .max_entries = 1000,
-};
-
static __always_inline u64 get_real_start_time()
{
struct task_struct* task = (struct task_struct*)bpf_get_current_task();
@@ -185,7 +177,6 @@ static inline void update_statistics(void *ctx, char stackmap_cur, s64 count, st
}
}
-
static inline int alloc_exit(void *ctx, u64 addr) {
u32 tgid = bpf_get_current_pid_tgid() >> INT_LEN;
struct pid_addr_t pa = {0};
diff --git a/src/probes/extends/ebpf.probe/src/stackprobe/svg.c b/src/probes/extends/ebpf.probe/src/stackprobe/svg.c
index db43aab..78a1cb0 100644
--- a/src/probes/extends/ebpf.probe/src/stackprobe/svg.c
+++ b/src/probes/extends/ebpf.probe/src/stackprobe/svg.c
@@ -156,10 +156,33 @@ static int __create_svg_files(struct stack_svg_s* svg_files, u32 period)
return 0;
}
+int __mkdir_with_svg_date(const char *svg_dir, char *svg_date_dir, size_t size)
+{
+ size_t len = strlen(svg_dir);
+ const char *day = get_cur_date();
+ if (len <= 1 || len + strlen(day) + 1 >= size) {
+ return -1;
+ }
+
+ (void)snprintf(svg_date_dir, size, "%s/%s", svg_dir, day);
+ if (access(svg_date_dir, F_OK) != 0) {
+ FILE *fp;
+ char command[COMMAND_LEN] = {0};
+ (void)snprintf(command, COMMAND_LEN, "/usr/bin/mkdir -p %s", svg_date_dir);
+ fp = popen(command, "r");
+ if (fp != NULL) {
+ (void)pclose(fp);
+ }
+ }
+
+ return 0;
+}
+
static int stack_get_next_svg_file(struct stack_svgs_s* svgs, char svg_file[], size_t size, int en_type)
{
int next;
char svg_name[PATH_LEN];
+ char svg_date_dir[PATH_LEN] = {0};
if (svgs->svg_files.files == NULL) {
return -1;
@@ -176,11 +199,15 @@ static int stack_get_next_svg_file(struct stack_svgs_s* svgs, char svg_file[], s
svgs->svg_files.files[next] = NULL;
}
+ if (__mkdir_with_svg_date(svgs->svg_dir, svg_date_dir, PATH_LEN) < 0) {
+ return -1;
+ }
+
svg_name[0] = 0;
- (void)snprintf(svg_name, PATH_LEN, "%s_%s.svg", svg_params[en_type].file_name, get_cur_time());
+ (void)snprintf(svg_name, PATH_LEN, "%s.svg", get_cur_time());
svg_file[0] = 0;
- (void)snprintf(svg_file, size, "%s/%s", svgs->svg_dir, svg_name);
+ (void)snprintf(svg_file, size, "%s/%s", svg_date_dir, svg_name);
__rm_svg(svg_file);
svgs->svg_files.files[next] = strdup(svg_file);
diff --git a/src/probes/extends/python.probe/pg_stat.probe/pg_stat_probe.py b/src/probes/extends/python.probe/pg_stat.probe/pg_stat_probe.py
index 45f80b5..fe8f955 100644
--- a/src/probes/extends/python.probe/pg_stat.probe/pg_stat_probe.py
+++ b/src/probes/extends/python.probe/pg_stat.probe/pg_stat_probe.py
@@ -121,5 +121,11 @@ if __name__ == "__main__":
while True:
time.sleep(g_period)
- get_metrics()
+ try:
+ get_metrics()
+ except Exception as e:
+ print("[pg_stat_probe]get metrics failed. Err:" + str(e))
+ stop_conns()
+
+
diff --git a/src/probes/system_infos.probe/system_procs.c b/src/probes/system_infos.probe/system_procs.c
index 6812bbc..0e1f8a4 100644
--- a/src/probes/system_infos.probe/system_procs.c
+++ b/src/probes/system_infos.probe/system_procs.c
@@ -31,7 +31,7 @@
#define PROC_START_TIME_CMD "/usr/bin/cat /proc/%s/stat | awk '{print $22}'"
#define PROC_CMDLINE_CMD "/proc/%s/cmdline"
#define PROC_FD "/proc/%s/fd"
-#define PROC_FD_CNT_CMD "/usr/bin/ls -l /proc/%s/fd | wc -l"
+#define PROC_FD_CNT_CMD "/usr/bin/ls -l /proc/%s/fd 2>/dev/null | wc -l 2>/dev/null"
#define PROC_IO "/proc/%s/io"
#define PROC_IO_CMD "/usr/bin/cat /proc/%s/io"
#define PROC_SMAPS "/proc/%s/smaps_rollup"
--
2.28.0.windows.1
Reference in New Issue View Git Blame Copy Permalink