If watchdog was triggered, that means that userspace is unreliable and we can't expect kernel messages are logged by userspace logger, thus we need ourselfs write info about watchdog triggered to special file configured through sysfs /sys/kernel/watchdog_log_path, so that users can see that watchdog was triggered but not some other event.
Note: we schedule do_halt_or_reboot to workqueue, which consists of pool of kernel threads. Manager thread adds more worker threads when all existing workers sleep, restricted by number of max_active. For our system_wq, max_active == 256. So we can not fear that waiting for fsync in our work will deadlock with writeback work. v2: use kernel_write helper https://jira.sw.ru/browse/PSBM-54747 Signed-off-by: Pavel Tikhomirov <[email protected]> --- kernel/fence-watchdog.c | 70 +++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 70 insertions(+) diff --git a/kernel/fence-watchdog.c b/kernel/fence-watchdog.c index f5c2347f..c555d21 100644 --- a/kernel/fence-watchdog.c +++ b/kernel/fence-watchdog.c @@ -23,6 +23,8 @@ #include <linux/fence-watchdog.h> #include <linux/device.h> #include <linux/kmsg_dump.h> +#include <linux/fs.h> +#include <linux/string.h> #define MAX_U64 (~(u64)0) #define MAX_JIFFIES_DELTA (10 * 365UL * 24UL * 3600UL * HZ) @@ -41,11 +43,60 @@ const char *action_names[] = {"crash", "reboot", "halt", "netfilter", NULL}; DEFINE_VVAR(volatile unsigned long, fence_wdog_jiffies64) = MAX_U64; static int fence_wdog_action = FENCE_WDOG_CRASH; static atomic_t not_fenced = ATOMIC_INIT(-1); +static char fence_wdog_log_path[PATH_MAX] = "/fence_wdog.log"; + +#define MSG_LEN 32 + +void fence_wdog_log(void) +{ + char msg[MSG_LEN] = "fence-watchdog: "; + struct file *file; + int ret; + + ret = sprintf(msg+strlen(msg), "%s\n", action_names[fence_wdog_action]); + if (ret != strlen(action_names[fence_wdog_action]) + 1) { + printk(KERN_EMERG"fence-watchdog: Failed to sprintf msg\n"); + return; + } + + file = filp_open(fence_wdog_log_path, + O_CREAT | O_WRONLY | O_APPEND | O_NOFOLLOW | O_LARGEFILE, + 0600); + if (IS_ERR(file)) { + printk(KERN_EMERG"fence-watchdog: Failed to open log path\n"); + return; + } + + if (!S_ISREG(file_inode(file)->i_mode)) { + printk(KERN_EMERG"fence-watchdog: Wrong type of log file\n"); + goto close; + } + + ret = kernel_write(file, msg, strlen(msg), file->f_pos); + if (ret < 0) { + printk(KERN_EMERG"fence-watchdog: Failed to write msg, ret=%d\n", ret); + goto close; + } + + ret = vfs_fsync(file, 0); + if (ret < 0) + printk(KERN_EMERG"fence-watchdog: Failed to fsync log file ret=%d\n", ret); + +close: + ret = filp_close(file, NULL); + if (ret < 0) + printk(KERN_EMERG"fence-watchdog: Failed to close log file ret=%d\n", ret); + + return; +} static void do_halt_or_reboot(struct work_struct *dummy) { printk(KERN_EMERG"fence-watchdog: %s\n", action_names[fence_wdog_action]); + + fence_wdog_log(); + switch (fence_wdog_action) { case FENCE_WDOG_REBOOT: emergency_restart(); @@ -166,6 +217,20 @@ static ssize_t fence_wdog_available_actions_show(struct kobject *kobj, return ret; } +static ssize_t fence_wdog_log_path_show(struct kobject *kobj, + struct kobj_attribute *attr, char *buf) +{ + return sprintf(buf, "%s\n", fence_wdog_log_path); +} + +static ssize_t fence_wdog_log_path_store(struct kobject *kobj, + struct kobj_attribute *attr, const char *buf, size_t count) +{ + if (sscanf(buf, "%s", fence_wdog_log_path) != 1) + return -EINVAL; + return 0; +} + static struct kobj_attribute fence_wdog_timer_attr = __ATTR(watchdog_timer, 0644, fence_wdog_timer_show, fence_wdog_timer_store); @@ -178,10 +243,15 @@ static struct kobj_attribute fence_wdog_available_actions_attr = __ATTR(watchdog_available_actions, 0644, fence_wdog_available_actions_show, NULL); +static struct kobj_attribute fence_wdog_log_path_attr = + __ATTR(watchdog_log_path, 0644, + fence_wdog_log_path_show, fence_wdog_log_path_store); + static struct attribute *fence_wdog_attrs[] = { &fence_wdog_timer_attr.attr, &fence_wdog_action_attr.attr, &fence_wdog_available_actions_attr.attr, + &fence_wdog_log_path_attr.attr, NULL, }; -- 2.9.4 _______________________________________________ Devel mailing list [email protected] https://lists.openvz.org/mailman/listinfo/devel
