当前位置：首页 > news >正文

ftrace使用实战

news 2025/8/4 1:43:20

诉求：遇到一个问题 echo blocked > /sys/class/block/sdb/device/state 报非法参数，想要知道根因，但是对这块内核代码不熟悉，不知道从哪里下手，那就先用ftrace看看内核调用栈，如下所示。

root@rlk:/home/rlk/rlk# echo blocked > /sys/class/block/sdb/device/state
bash: echo: write error: Invalid argument

cd /sys/kernel/debug/tracing

设定跟踪的进程pid
set_ftrace_pid
查看可以设定的tracer

root@rlk:/home/rlk/rlk# cat /sys/kernel/debug/tracing/available_tracers
hwlat blk mmiotrace function_graph wakeup_dl wakeup_rt wakeup function nop

设定tracer类型

root@rlk:/sys/kernel/debug/tracing# echo function_graph > current_tracer

开启tracer

root@rlk:/sys/kernel/debug/tracing# echo 1 > tracing_on

写文件/sys/class/block/sdb/device/state是一个很快的过程，没有办法精准的知道什么时候把trace打开，所以只能采集相对来说比较泛的数据，为了减少一些无用的数据，可以通过写两个脚本来实现，先抓取echo running > /sys/class/block/sdb/device/state的调用栈。

write.sh

#!/bin/bash
sleep 0.3
echo running > /sys/class/block/sdb/device/state

查找进程pid命令花费的时间为0.109s

root@rlk:/home/rlk/rlk# time ps aux | grep write.sh | grep -v grep | awk '{print $2}'
5442real    0m0.109s
user    0m0.019s
sys     0m0.121s

#!/bin/bash
./write.sh &
pid=`ps aux | grep write.sh | grep -v grep | awk '{print $2}'`
echo $pidecho $pid > /sys/kernel/debug/tracing/set_ftrace_pid
echo 1 > /sys/kernel/debug/tracing/tracing_on

echo 其实就是一个打开文件，然后写文件的过程，用户态写文件会调用vfs_write，在trace日志中抓出如下调用信息。

 1)               |  do_syscall_64() {1)               |    __x64_sys_write() {1)               |      ksys_write() {1)               |        __fdget_pos() {1)   0.100 us    |          __fget_light();1)   0.291 us    |        }1)               |        vfs_write() {1)               |          rw_verify_area() {1)               |            security_file_permission() {1)               |              apparmor_file_permission() {1)               |                common_file_perm() {1)   0.100 us    |                  aa_file_perm();1)   0.290 us    |                }1)   0.481 us    |              }1)   0.682 us    |            }1)   0.872 us    |          }1)               |          __sb_start_write() {1)               |            _cond_resched() {1)   0.090 us    |              rcu_all_qs();1)   0.271 us    |            }1)   0.461 us    |          }1)               |          __vfs_write() {1)               |            kernfs_fop_write() {1)               |              __kmalloc() {1)   0.090 us    |                kmalloc_slab();1)               |                _cond_resched() {1)   0.091 us    |                  rcu_all_qs();1)   0.291 us    |                }1)   0.090 us    |                should_failslab();1)   0.330 us    |                memcg_kmem_put_cache();1)   1.513 us    |              }1)               |              __check_object_size() {1)   0.100 us    |                check_stack_object();1)   0.090 us    |                __virt_addr_valid();1)   0.491 us    |                __check_heap_object();1)   1.202 us    |              }1)               |              mutex_lock() {1)               |                _cond_resched() {1)   0.090 us    |                  rcu_all_qs();1)   0.261 us    |                }1)   0.451 us    |              }1)   0.090 us    |              kernfs_get_active();1)               |              sysfs_kf_write() {	//调用sfsfs写接口1)               |                dev_attr_store() {1)               |                  store_state_field() {1)               |                    mutex_lock() {1)               |                      _cond_resched() {1)   0.100 us    |                        rcu_all_qs();1)   0.271 us    |                      }1)   0.461 us    |                    }1)   0.200 us    |                    scsi_device_set_state(); //在这里设置了device的状态1)               |                    blk_mq_run_hw_queues() {1)               |                      blk_mq_run_hw_queue() {1)   0.390 us    |                        dd_has_work();1)   2.084 us    |                      }1)   2.775 us    |                    }1)   0.090 us    |                    mutex_unlock();1)   4.628 us    |                  }1)   5.210 us    |                }1)   6.191 us    |              }1)   0.101 us    |              kernfs_put_active();1)   0.090 us    |              mutex_unlock();1)   0.091 us    |              kfree();1) + 10.841 us   |            }1) + 11.071 us   |          }1)   0.091 us    |              kfree();1) + 10.841 us   |            }1) + 11.071 us   |          }1)   0.090 us    |          __fsnotify_parent();1)   0.090 us    |          fsnotify();1)   0.080 us    |          __sb_end_write();1) + 13.415 us   |        }1) + 14.006 us   |      }1) + 14.187 us   |    }1)   0.090 us    |    fpregs_assert_state_consistent();1) + 14.627 us   |  }

drivers/scsi/scsi_sysfs.c定义了 /sys/class/block/sdx/device/state操作函数，所有块设备相关（ll /sys/class/block/sdx/device/）DEVICE_ATTR都可以在这个文件中找到，测试的内核版本为5.4.0-26-generic


// 
static const struct {enum scsi_host_state	value;char			*name;
} shost_states[] = {{ SHOST_CREATED, "created" },{ SHOST_RUNNING, "running" },{ SHOST_CANCEL, "cancel" },{ SHOST_DEL, "deleted" },{ SHOST_RECOVERY, "recovery" },{ SHOST_CANCEL_RECOVERY, "cancel/recovery" },{ SHOST_DEL_RECOVERY, "deleted/recovery", },
};
const char *scsi_host_state_name(enum scsi_host_state state)
{int i;char *name = NULL;for (i = 0; i < ARRAY_SIZE(shost_states); i++) {if (shost_states[i].value == state) {name = shost_states[i].name;break;}}return name;
}static ssize_t
store_state_field(struct device *dev, struct device_attribute *attr,const char *buf, size_t count)
{int i, ret;struct scsi_device *sdev = to_scsi_device(dev);enum scsi_device_state state = 0;for (i = 0; i < ARRAY_SIZE(sdev_states); i++) {const int len = strlen(sdev_states[i].name);if (strncmp(sdev_states[i].name, buf, len) == 0 &&buf[len] == '\n') {state = sdev_states[i].value;break;}}switch (state) {case SDEV_RUNNING:case SDEV_OFFLINE:break;default:return -EINVAL;}mutex_lock(&sdev->state_mutex);ret = scsi_device_set_state(sdev, state);/** If the device state changes to SDEV_RUNNING, we need to run* the queue to avoid I/O hang.*/if (ret == 0 && state == SDEV_RUNNING)blk_mq_run_hw_queues(sdev->request_queue, true);mutex_unlock(&sdev->state_mutex);return ret == 0 ? count : -EINVAL;
}static ssize_t
show_state_field(struct device *dev, struct device_attribute *attr, char *buf)
{struct scsi_device *sdev = to_scsi_device(dev);const char *name = scsi_device_state_name(sdev->sdev_state);if (!name)return -EINVAL;return snprintf(buf, 20, "%s\n", name);
}static DEVICE_ATTR(state, S_IRUGO | S_IWUSR, show_state_field, store_state_field);

以上问题可以在上面的代码找到答案，当前仅支持设置SDEV_RUNNING,SDEV_OFFLINE两种状态。

DEVICE_ATTR 是一个宏定义，用于在 Linux 设备驱动程序中定义设备属性。它定义了一个名为 dev_attr_ 的静态结构体变量，其中是属性的名称。该结构体包含了属性的名称、读取和写入函数的指针，以及一些其他属性。

使用 DEVICE_ATTR 宏可以方便地定义设备属性，而无需手动编写结构体和函数。例如，以下代码定义了一个名为 my_attribute 的设备属性：

static ssize_t my_attribute_show(struct device *dev, struct device_attribute *attr, char *buf)
{// 读取属性值并将其写入缓冲区return sprintf(buf, "Hello, world!\n");
}static ssize_t my_attribute_store(struct device *dev, struct device_attribute *attr, const char *buf, size_t count)
{// 将缓冲区中的值写入属性return count;
}DEVICE_ATTR(my_attribute, 0644, my_attribute_show, my_attribute_store);