Re: 5.11.0: open ctree failed: devide total_bytes should be at most X but found Y

[David Sterba <dsterba@suse.cz>]

On Tue, Feb 23, 2021 at 09:43:04AM +0000, Johannes Thumshirn wrote:
> On 23/02/2021 10:13, Johannes Thumshirn wrote:
> > On 22/02/2021 21:07, Steven Davies wrote:
> > 
> > [+CC Anand ]
> > 
> >> Booted my system with kernel 5.11.0 vanilla with the first time and received this:
> >>
> >> BTRFS info (device nvme0n1p2): has skinny extents
> >> BTRFS error (device nvme0n1p2): device total_bytes should be at most 964757028864 but found 
> >> 964770336768
> >> BTRFS error (device nvme0n1p2): failed to read chunk tree: -22
> >>
> >> Booting with 5.10.12 has no issues.
> >>
> >> # btrfs filesystem usage /
> >> Overall:
> >>      Device size:                 898.51GiB
> >>      Device allocated:            620.06GiB
> >>      Device unallocated:          278.45GiB
> >>      Device missing:                  0.00B
> >>      Used:                        616.58GiB
> >>      Free (estimated):            279.94GiB      (min: 140.72GiB)
> >>      Data ratio:                       1.00
> >>      Metadata ratio:                   2.00
> >>      Global reserve:              512.00MiB      (used: 0.00B)
> >>
> >> Data,single: Size:568.00GiB, Used:566.51GiB (99.74%)
> >>     /dev/nvme0n1p2        568.00GiB
> >>
> >> Metadata,DUP: Size:26.00GiB, Used:25.03GiB (96.29%)
> >>     /dev/nvme0n1p2         52.00GiB
> >>
> >> System,DUP: Size:32.00MiB, Used:80.00KiB (0.24%)
> >>     /dev/nvme0n1p2         64.00MiB
> >>
> >> Unallocated:
> >>     /dev/nvme0n1p2        278.45GiB
> >>
> >> # parted -l
> >> Model: Sabrent Rocket Q (nvme)
> >> Disk /dev/nvme0n1: 1000GB
> >> Sector size (logical/physical): 512B/512B
> >> Partition Table: gpt
> >> Disk Flags:
> >>
> >> Number  Start   End     Size    File system     Name  Flags
> >>   1      1049kB  1075MB  1074MB  fat32                 boot, esp
> >>   2      1075MB  966GB   965GB   btrfs
> >>   3      966GB   1000GB  34.4GB  linux-swap(v1)        swap
> >>
> >> What has changed in 5.11 which might cause this?
> >>
> >>
> > 
> > This line:
> >> BTRFS info (device nvme0n1p2): has skinny extents
> >> BTRFS error (device nvme0n1p2): device total_bytes should be at most 964757028864 but found 
> >> 964770336768
> >> BTRFS error (device nvme0n1p2): failed to read chunk tree: -22
> > 
> > comes from 3a160a933111 ("btrfs: drop never met disk total bytes check in verify_one_dev_extent")
> > which went into v5.11-rc1.
> > 
> > IIUIC the device item's total_bytes and the block device inode's size are off by 12M, so the check
> > introduced in the above commit refuses to mount the FS.
> > 
> > Anand any idea?
> 
> OK this is getting interesting:
> btrfs-porgs sets the device's total_bytes at mkfs time and obtains it from ioctl(..., BLKGETSIZE64, ...); 
> 
> BLKGETSIZE64 does:
> return put_u64(argp, i_size_read(bdev->bd_inode));
> 
> The new check in read_one_dev() does:
> 
>                u64 max_total_bytes = i_size_read(device->bdev->bd_inode);
> 
>                if (device->total_bytes > max_total_bytes) {
>                        btrfs_err(fs_info,
>                        "device total_bytes should be at most %llu but found %llu",
>                                  max_total_bytes, device->total_bytes);
>                        return -EINVAL;
> 
> 
> So the bdev inode's i_size must have changed between mkfs and mount.

The kernel side verifies that the physical device size is not smaller
than the size recorded in the device item, so that makes sense. I was a
bit doubtful about the check but it can detect real problems or point
out some weirdness.

The 12M delta is not big, but I'd expect that for a physical device it
should not change. Another possibility would be some kind of rounding to
a reasonable number, like 16M.