tried to send this but forgot to set Gmail's Plaintext Mode, so it bounced back. meanwhile I created this bug report: https://bugzilla.kernel.org/show_bug.cgi?id=151621 -- Hello. During unmounting(umount command, umount2 syscall), I've noticed that it takes a few seconds(sometimes up to 127 seconds) for "Unmounting /oldroot" to finish. During which 1 core is under 100% CPU usage. It looks like this happens during cleanup_mnt which is in kernel's fs/namespace.c and more specifically during deactivate_super(mnt->mnt.mnt_sb);
I did a sysrq+c to crash while this was happening and this is the screenshot of the stacktrace: https://i.imgur.com/tdteOjB.jpg What I've also noticed that Sysrq+m to show memory info shows the about 2 million "total pagecache pages" which decrease as time passes while unmounting is doing this heavy work(of freeing the pages - my guess) until it gets to about 771 when it finishes. This started happening maybe 6 or more months ago, or maybe it was something that I did(other than update kernel to latest git all the time), but it didn't use to hang at all during umount. Now it's just taking a lot of time stalling apparently taking its time to free each pagecache - my guess. It doesn't seem to be flushing anything because there's no SSD led blinking and also plenty of sync commands were issued beforehand (and rootfs / is also in remount,ro if that even matters). The amount of time that it hangs depends on the uptime(and/or maybe on the amount that was written during that uptime) - it's now apparent to me that the more pagecache pages there are, the more time is spent during umounting the btrfs partition. I remember umount-ing btrfs filesystems while the system was up(so not during shutdown) and I think they might've even been mounted ro only but the stalling and hight CPU usage was present even then. Or my memory fails me(about the ro part!) Some examples of right now "pagecache pages" via echo m | sudo tee /proc/sysrq-trigger then checking dmesg: [ 598.226377] 288309 total pagecache pages [ 1463.371632] 294881 total pagecache pages [ 2050.372096] 338460 total pagecache pages These increase a lot in time(depending on how much I write? my guess - but regardless of how much I execute "sync") When there are over 2 million, it takes like 2 minutes to unmount, so that's how extra long I've to wait for a reboot/shutdown to complete. Thanks. Using kernel: Linux myzee 4.7.0-ga157b3a #7 SMP PREEMPT Sat Aug 6 15:22:34 EEST 2016 x86_64 GNU/Linux I've tried commit=300 and commit=3 in fstab, no change. I've tried many other things I can't remember:D I can try patches, no problem. Please? ;-) Thanks my friends. -- To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
