oom-killer disable for iscsi/lvm2/multipath userland critical sections
Hello,
some private discussion (that was continuing some kernel-summit-discuss thread) ended in the below patch. I also liked a textual "disable" instead of value "-17" (internally to the kernel it could be represented the same way, but the /proc parsing would be more complicated). If you prefer textual "disable" we can change this of course.
Comments welcome.
From: Andrea Arcangeli <[email protected]>
Subject: oom killer protection
iscsi/lvm2/multipath needs guaranteed protection from the oom-killer.
fs/proc/base.c renamed from fs/proc/base.c.~1~ +1 -1
@@ -751,7 +751,7 @@ static ssize_t oom_adjust_write(struct fif (copy_from_user (buffer, buf, count))return -EFAULT ; oom_adjust =simple_strtol (buffer, &end,0 ); -if (oom_adjust <-16 || oom_adjust >15 ) +if ((oom_adjust <-16 || oom_adjust >15 ) && oom_adjust !=OOM_DISABLE )return -EINVAL ;if (*end =='\n' ) end++;
include/linux/mm.h renamed from include/linux/mm.h.~1~ +3 -0
@@ -856,5 +856,8 @@ int in_gate_area_no_task(unsigned long a#define in_gate_area (task, addr) ({(void)task; in_gate_area_no_task(addr);})#endif /* __HAVE_ARCH_GATE_AREA */ +/* /proc/<pid>/oom_adj set to -17 protects from the oom-killer */ +#define OOM_DISABLE -17 +#endif /* __KERNEL__ */ #endif /* _LINUX_MM_H */
mm/oom_kill.c renamed from mm/oom_kill.c.~1~ +1 -1
@@ -145,7 +145,7 @@ static struct task_struct * select_bad_pdo_posix_clock_monotonic_gettime (&uptime);do_each_thread (g, p) /* skip the init task with pid == 1 */ -if (p->pid >1 ) { +if (p->pid >1 && p->oomkilladj !=OOM_DISABLE ) {unsigned long points;/*
Andrea,
I just successfully tested the patch on my environment. It actually resolved OOM-killer problem for my iscsid.
Important note: daemon's parent must be init.
In my test, OOM-killer killed everything around but iscsid, and iscsid successfully finished registration of new SCSI host in the middle of crazy OOM-killer :)
Thanks!
Dima
Quoted 55 lines · Andrea Arcangeli
Hello,
some private discussion (that was continuing some kernel-summit-discuss thread) ended in the below patch. I also liked a textual "disable" instead of value "-17" (internally to the kernel it could be represented the same way, but the /proc parsing would be more complicated). If you prefer textual "disable" we can change this of course.
Comments welcome.
From: Andrea Arcangeli <[email protected]>
Subject: oom killer protectioniscsi/lvm2/multipath needs guaranteed protection from the oom-killer.
Signed-off-by Andrea Arcangeli <[email protected]>fs/proc/base.c renamed from fs/proc/base.c.~1~ +1
-1@@ -751,7 +751,7 @@ static ssize_t oom_adjust_write(struct fif (copy_from_user (buffer, buf, count))return -EFAULT ; oom_adjust =simple_strtol (buffer, &end,0 ); -if (oom_adjust <-16 || oom_adjust >15 ) +if ((oom_adjust <-16 || oom_adjust >15 ) && oom_adjust !=OOM_DISABLE )return -EINVAL ;if (*end =='\n' ) end++;include/linux/mm.h renamed from include/linux/mm.h.~1~ +3
-0@@ -856,5 +856,8 @@ int in_gate_area_no_task(unsigned long a#define in_gate_area (task, addr) ({(void)task; in_gate_area_no_task(addr);})#endif /* __HAVE_ARCH_GATE_AREA */ +/* /proc/<pid>/oom_adj set to -17 protects from the oom-killer */ +#define OOM_DISABLE -17 +#endif /* __KERNEL__ */ #endif /* _LINUX_MM_H */ mm/oom_kill.c renamed from mm/oom_kill.c.~1~ +1
-2@@ -145,7 +145,7 @@ static struct task_struct * select_bad_pdo_posix_clock_monotonic_gettime (&uptime);do_each_thread (g, p) /* skip the init task with pid == 1 */ -if (p->pid >1 ) { +if (p->pid >1 && p->oomkilladj !=OOM_DISABLE ) {unsigned long points;/* -To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/