Linux 4.14 Kernel Officially Released

[lutel]

In 4.14 it is not possible to disable frame pointers like in previous kernels. You have to compile in ORC unwinder if you try to disable frame pointer. Why can't we remove the tracing code from kernel 4.14 now? This shouldn't happen in any stable kernel, and now this code landed in LTS and is enforced on all users.

[hiryu]

A bit of warning... If you're using the very rare case of btrfs+bcache, this kernel will damage your filesystem.

I ran into the same issue with rc5 only to find out that my backups weren't backing up everything. Fortunately, I was able to recover by the skin of my teeth and fix my backup script.

Like I said, this is a rare setup so it was my mistake for not reporting this issue. No idea if the issue is with bcache or btrfs. Maybe I'll sign up to the btrfs mailing list just to report it where it will have some visibility.

The root filesystem on that box is btrfs, but without bcache, so the bug doesn't seem to affect btrfs alone.

Here's how I corrected the issue for anyone else affected:
btrfs rescue chunk-recover /dev/bcache0

The the first time this happened, I ran chunk-recover for many hours (more than 8). I was about to just give up so I ran control-c, tried to mount it, and it worked!

This time, I ran it for 15.5 minutes, control-c'ed and it mounted.

This time, I first tried several times, running it for ~1 minute, ~3 minutes ,etc, but it failed. So at least for me, you need to run it for some minimum amount of time.

It's an 11TB hardware RAID10 volume on a 3ware 9750 controller. 6.1 TB used. The caching device is a 1TB RAID1 volume running on the same controller.

The first time I recovered, I immediately ran a full backup and then a filesystem scrub to be sure. Because this time my backups are solid, I mounted and then just rebooted. Came up fine. However, If you encounter this issue yourself, just be careful.

The previous time this happened, I set the cache to writethrough, and never set it back to writeback. Which means this issue probably wasn't caused by losing my cache.

Hope this helps someone else.

[aceman]

In 4.14 it is not possible to disable frame pointers like in previous kernels. You have to compile in ORC unwinder if you try to disable frame pointer.

Actually the Kernel Newbies page seems to imply ORC is the one to work without frame pointer, see https://kernelnewbies.org/Linux_4.14...88678b19a9ef33 . Maybe the option was removed as no longer needed? Or can you see that frame pointers are really inserted in the binary?