packages/sysSentry
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
sysSentry/fix-bug-of-ebpf-and-ai_block_io.patch
zhuofeng 95cc3d6f1c fix bug of ebpf and ai_block_io 
Signed-off-by: zhuofeng <1107893276@qq.com>
2025-02-14 14:20:04 +08:00

398 lines
14 KiB
Diff
Raw Permalink Blame History

This file contains ambiguous Unicode characters

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

From 480c0fc479ec882786cdb58d699cf84ce5995531 Mon Sep 17 00:00:00 2001
From: zhuofeng <zhuofeng2@huawei.com>
Date: Fri, 14 Feb 2025 09:42:27 +0800
Subject: [PATCH] fix bug of ebpf and ai_block_io
---
src/c/ebpf_collector/ebpf_collector.bpf.c | 357 ------------------
.../sentryPlugins/ai_block_io/detector.py | 3 +-
2 files changed, 2 insertions(+), 358 deletions(-)
diff --git a/src/c/ebpf_collector/ebpf_collector.bpf.c b/src/c/ebpf_collector/ebpf_collector.bpf.c
index 417618d..7a2f481 100644
--- a/src/c/ebpf_collector/ebpf_collector.bpf.c
+++ b/src/c/ebpf_collector/ebpf_collector.bpf.c
@@ -590,361 +590,4 @@ int kprobe_bio_endio(struct pt_regs *regs)
return 0;
}
-// start get_tag
-SEC("kprobe/blk_mq_get_tag")
-int kprobe_blk_mq_get_tag(struct pt_regs *regs)
-{
- u64 tagkey = bpf_get_current_task();
- u64 value = (u64)PT_REGS_PARM1(regs);
- (void)bpf_map_update_elem(&tag_args, &tagkey, &value, BPF_ANY);
-
- struct blk_mq_alloc_data *bd;
- struct request_queue *q;
- struct backing_dev_info *backing_dev_info;
- struct device *owner;
- dev_t devt;
- unsigned int cmd_flags = 0;
-
- bd = (struct blk_mq_alloc_data *)value;
- bpf_core_read(&q, sizeof(q), &bd->q);
- bpf_core_read(&backing_dev_info, sizeof(backing_dev_info), &q->backing_dev_info);
- bpf_core_read(&owner, sizeof(owner), &backing_dev_info->owner);
- bpf_core_read(&devt, sizeof(devt), &owner->devt);
- int major = MAJOR(devt);
- int first_minor = MINOR(devt);
-
- if (major == 0) {
- log_event(STAGE_GET_TAG, PERIOD_START, ERROR_MAJOR_ZERO);
- return 0;
- }
-
- u32 key = find_matching_key_get_tag(major, first_minor);
- if (key >= MAP_SIZE) {
- return 0;
- }
-
- struct io_counter *counterp, zero = {};
- init_io_counter(&zero, major, first_minor);
- counterp = bpf_map_lookup_elem(&tag_map, &tagkey);
- if (counterp) {
- return 0;
- }
- long err = bpf_map_update_elem(&tag_map, &tagkey, &zero, BPF_NOEXIST);
- if (err) {
- log_event(STAGE_GET_TAG, PERIOD_START, ERROR_UPDATE_FAIL);
- return 0;
- }
-
- u64 curr_start_range = zero.start_time / THRESHOLD;
-
- struct update_params params = {
- .major = major,
- .first_minor = first_minor,
- .cmd_flags = cmd_flags,
- .curr_start_range = curr_start_range,
- };
-
- struct stage_data *curr_data;
- curr_data = bpf_map_lookup_elem(&tag_res, &key);
- if (!curr_data) {
- struct stage_data new_data = {
- .start_count = 1,
- .finish_count = 0,
- .finish_over_time = 0,
- .duration = 0,
- .major = major,
- .first_minor = first_minor,
- .io_type = "",
- };
- blk_fill_rwbs(new_data.io_type, cmd_flags);
- bpf_map_update_elem(&tag_res, &key, &new_data, 0);
- } else {
- update_curr_data_in_start(curr_data, &params);
- }
-
- struct time_range_io_count *curr_data_time_range;
- curr_data_time_range = bpf_map_lookup_elem(&tag_res_2, &curr_start_range);
- if (curr_data_time_range == NULL) {
- struct time_range_io_count new_data = { .count = {0} };
- bpf_map_update_elem(&tag_res_2, &curr_start_range, &new_data, 0);
- } else {
- if (key < MAP_SIZE && key >= 0) {
- __sync_fetch_and_add(&curr_data_time_range->count[key], 1);
- }
- }
- return 0;
-}
-
-// finish get_tag
-SEC("kretprobe/blk_mq_get_tag")
-int kretprobe_blk_mq_get_tag(struct pt_regs *regs)
-{
- u64 tagkey = bpf_get_current_task();
- u64 *tagargs = (u64 *)bpf_map_lookup_elem(&tag_args, &tagkey);
- if (tagargs == NULL) {
- bpf_map_delete_elem(&tag_args, &tagkey);
- return 0;
- }
-
- struct blk_mq_alloc_data *bd;
- struct request_queue *q;
- struct backing_dev_info *backing_dev_info;
- struct device *owner;
- dev_t devt;
- unsigned int cmd_flags = 0;
-
- bd = (struct blk_mq_alloc_data *)*tagargs;
- bpf_core_read(&q, sizeof(q), &bd->q);
- bpf_core_read(&backing_dev_info, sizeof(backing_dev_info), &q->backing_dev_info);
- bpf_core_read(&owner, sizeof(owner), &backing_dev_info->owner);
- bpf_core_read(&devt, sizeof(devt), &owner->devt);
- int major = MAJOR(devt);
- int first_minor = MINOR(devt);
-
- if (major == 0) {
- log_event(STAGE_GET_TAG, PERIOD_END, ERROR_MAJOR_ZERO);
- return 0;
- }
-
- u32 key = find_matching_key_get_tag(major, first_minor);
- if (key >= MAP_SIZE) {
- return 0;
- }
-
- struct io_counter *counterp = bpf_map_lookup_elem(&tag_map, &tagkey);
- if (!counterp) {
- return 0;
- }
-
- u64 duration = bpf_ktime_get_ns() - counterp->start_time;
- u64 curr_start_range = counterp->start_time / THRESHOLD;
-
- struct update_params params = {
- .major = major,
- .first_minor = first_minor,
- .cmd_flags = cmd_flags,
- .curr_start_range = curr_start_range,
- };
-
- struct stage_data *curr_data;
- curr_data = bpf_map_lookup_elem(&tag_res, &key);
- if (curr_data == NULL && duration > DURATION_THRESHOLD) {
- struct stage_data new_data = {
- .start_count = 1,
- .finish_count = 1,
- .finish_over_time = 1,
- .duration = 0,
- .major = major,
- .first_minor = first_minor,
- .io_type = "",
- };
- blk_fill_rwbs(new_data.io_type, cmd_flags);
- bpf_map_update_elem(&tag_res, &key, &new_data, 0);
- } else if (curr_data == NULL) {
- struct stage_data new_data = {
- .start_count = 1,
- .finish_count = 1,
- .finish_over_time = 0,
- .duration = 0,
- .major = major,
- .first_minor = first_minor,
- .io_type = "",
- };
- blk_fill_rwbs(new_data.io_type, cmd_flags);
- bpf_map_update_elem(&tag_res, &key, &new_data, 0);
- } else {
- curr_data->duration += duration;
- update_curr_data_in_finish(curr_data, &params, duration);
- }
-
- struct time_range_io_count *curr_data_time_range;
- curr_data_time_range = bpf_map_lookup_elem(&tag_res_2, &curr_start_range);
- if (curr_data_time_range == NULL) {
- struct time_range_io_count new_data = { .count = {0} };
- bpf_map_update_elem(&tag_res_2, &curr_start_range, &new_data, 0);
- } else {
- if (key < MAP_SIZE && curr_data_time_range->count[key] > 0) {
- __sync_fetch_and_add(&curr_data_time_range->count[key], -1);
- }
- }
-
- bpf_map_delete_elem(&tag_map, &tagkey);
- bpf_map_delete_elem(&tag_args, &tagkey);
- return 0;
-}
-
-// start wbt
-SEC("kprobe/wbt_wait")
-int kprobe_wbt_wait(struct pt_regs *regs)
-{
- u64 wbtkey = bpf_get_current_task();
- u64 value = (u64)PT_REGS_PARM2(regs);
- (void)bpf_map_update_elem(&wbt_args, &wbtkey, &value, BPF_ANY);
-
- struct bio *bio;
- struct gendisk *curr_rq_disk;
- int major, first_minor;
- unsigned int cmd_flags;
-
- bio = (struct bio *)value;
- bpf_core_read(&curr_rq_disk, sizeof(curr_rq_disk), &bio->bi_disk);
- bpf_core_read(&major, sizeof(major), &curr_rq_disk->major);
- bpf_core_read(&first_minor, sizeof(first_minor), &curr_rq_disk->first_minor);
- bpf_core_read(&cmd_flags, sizeof(cmd_flags), &bio->bi_opf);
-
- if (major == 0) {
- log_event(STAGE_WBT, PERIOD_START, ERROR_MAJOR_ZERO);
- return 0;
- }
-
- u32 key = find_matching_key_wbt(major, first_minor);
- if (key >= MAP_SIZE) {
- return 0;
- }
-
- struct io_counter *counterp, zero = {};
- init_io_counter(&zero, major, first_minor);
- counterp = bpf_map_lookup_elem(&wbt_map, &wbtkey);
- if (counterp) {
- return 0;
- }
- long err = bpf_map_update_elem(&wbt_map, &wbtkey, &zero, BPF_NOEXIST);
- if (err) {
- log_event(STAGE_WBT, PERIOD_START, ERROR_UPDATE_FAIL);
- return 0;
- }
-
- u64 curr_start_range = zero.start_time / THRESHOLD;
-
- struct update_params params = {
- .major = major,
- .first_minor = first_minor,
- .cmd_flags = cmd_flags,
- .curr_start_range = curr_start_range,
- };
-
- struct stage_data *curr_data;
- curr_data = bpf_map_lookup_elem(&wbt_res, &key);
- if (!curr_data) {
- struct stage_data new_data = {
- .start_count = 1,
- .finish_count = 0,
- .finish_over_time = 0,
- .duration = 0,
- .major = major,
- .first_minor = first_minor,
- .io_type = "",
- };
- blk_fill_rwbs(new_data.io_type, cmd_flags);
- bpf_map_update_elem(&wbt_res, &key, &new_data, 0);
- } else {
- update_curr_data_in_start(curr_data, &params);
- }
-
- struct time_range_io_count *curr_data_time_range;
- curr_data_time_range = bpf_map_lookup_elem(&wbt_res_2, &curr_start_range);
- if (curr_data_time_range == NULL) {
- struct time_range_io_count new_data = { .count = {0} };
- bpf_map_update_elem(&wbt_res_2, &curr_start_range, &new_data, 0);
- } else {
- if (key < MAP_SIZE && key >= 0) {
- __sync_fetch_and_add(&curr_data_time_range->count[key], 1);
- }
- }
- return 0;
-}
-
-// finish wbt
-SEC("kretprobe/wbt_wait")
-int kretprobe_wbt_wait(struct pt_regs *regs)
-{
- u64 wbtkey = bpf_get_current_task();
- u64 *wbtargs = (u64 *)bpf_map_lookup_elem(&wbt_args, &wbtkey);
- if (wbtargs == NULL) {
- bpf_map_delete_elem(&wbt_args, &wbtkey);
- return 0;
- }
-
- struct bio *bio;
- struct gendisk *curr_rq_disk;
- int major, first_minor;
- unsigned int cmd_flags;
-
- bio = (struct bio *)(*wbtargs);
- bpf_core_read(&curr_rq_disk, sizeof(curr_rq_disk), &bio->bi_disk);
- bpf_core_read(&major, sizeof(major), &curr_rq_disk->major);
- bpf_core_read(&first_minor, sizeof(first_minor), &curr_rq_disk->first_minor);
- bpf_core_read(&cmd_flags, sizeof(cmd_flags), &bio->bi_opf);
-
- if (major == 0) {
- log_event(STAGE_WBT, PERIOD_END, ERROR_MAJOR_ZERO);
- return 0;
- }
-
- u32 key = find_matching_key_wbt(major, first_minor);
- if (key >= MAP_SIZE) {
- return 0;
- }
-
- struct io_counter *counterp = bpf_map_lookup_elem(&wbt_map, &wbtkey);
- if (!counterp) {
- return 0;
- }
-
- u64 duration = bpf_ktime_get_ns() - counterp->start_time;
- u64 curr_start_range = counterp->start_time / THRESHOLD;
-
- struct update_params params = {
- .major = major,
- .first_minor = first_minor,
- .cmd_flags = cmd_flags,
- .curr_start_range = curr_start_range,
- };
-
- struct stage_data *curr_data;
- curr_data = bpf_map_lookup_elem(&wbt_res, &key);
- if (curr_data == NULL && duration > DURATION_THRESHOLD) {
- struct stage_data new_data = {
- .start_count = 1,
- .finish_count = 1,
- .finish_over_time = 1,
- .duration = 0,
- .major = major,
- .first_minor = first_minor,
- .io_type = "",
- };
- blk_fill_rwbs(new_data.io_type, cmd_flags);
- bpf_map_update_elem(&wbt_res, &key, &new_data, 0);
- } else if (curr_data == NULL) {
- struct stage_data new_data = {
- .start_count = 1,
- .finish_count = 1,
- .finish_over_time = 0,
- .duration = 0,
- .io_type = "",
- .major = major,
- .first_minor = first_minor,
- };
- blk_fill_rwbs(new_data.io_type, cmd_flags);
- bpf_map_update_elem(&wbt_res, &key, &new_data, 0);
- } else {
- curr_data->duration += duration;
- update_curr_data_in_finish(curr_data, &params, duration);
- }
-
- struct time_range_io_count *curr_data_time_range;
- curr_data_time_range = bpf_map_lookup_elem(&wbt_res_2, &curr_start_range);
- if (curr_data_time_range == NULL) {
- struct time_range_io_count new_data = { .count = {0} };
- bpf_map_update_elem(&wbt_res_2, &curr_start_range, &new_data, 0);
- } else {
- if (key < MAP_SIZE && curr_data_time_range->count[key] > 0) {
- __sync_fetch_and_add(&curr_data_time_range->count[key], -1);
- }
- }
- bpf_map_delete_elem(&wbt_map, &wbtkey);
- bpf_map_delete_elem(&wbt_args, &wbtkey);
- return 0;
-}
-
char _license[] SEC("license") = "GPL";
diff --git a/src/python/sentryPlugins/ai_block_io/detector.py b/src/python/sentryPlugins/ai_block_io/detector.py
index 27fb7f7..2688cb1 100644
--- a/src/python/sentryPlugins/ai_block_io/detector.py
+++ b/src/python/sentryPlugins/ai_block_io/detector.py
@@ -55,11 +55,12 @@ class Detector:
detection_result = self._slidingWindow.is_slow_io_event(metric_value)
# 检测到慢周期，由Detector负责打印info级别日志
if detection_result[0][1]:
+ ai_threshold = "None" if detection_result[2] is None else round(detection_result[2], 3)
logging.info(f'[abnormal_period]: disk: {self._metric_name.disk_name}, '
f'stage: {self._metric_name.stage_name}, '
f'iotype: {self._metric_name.io_access_type_name}, '
f'type: {self._metric_name.metric_name}, '
- f'ai_threshold: {round(detection_result[2], 3)}, '
+ f'ai_threshold: {ai_threshold}, '
f'curr_val: {metric_value}')
else:
logging.debug(f'Detection result: {str(detection_result)}')
--
2.33.0
Reference in New Issue View Git Blame Copy Permalink