[U-Boot] ext4 delete file fails when ext4 extents enabled in filesystem

["Brüns, Stefan" <Stefan.Bruens@rwth-aachen.de>]

On Donnerstag, 1. September 2016 19:25:30 CEST you wrote:
> On Donnerstag, 1. September 2016 16:08:51 CEST you wrote:
> > Hi Stefan,
> > 
> > applying patch [U-Boot,v4,06/13]ext4 and Michael Walles patch
> > [U-Boot,v4,3/4]ext4, I'm now able to write into directories on ext4 fs
> > from
> > u-boot. However, when deleting a given file (i.e. when writing to an
> > existing filename), u-boot crashes when ext4 extents are enabled.
> > 
> > Some debugging showd that blknr from 'read_allocated_block' function
> > returns negative value. I can only guess, maybe its due to 64 bit values
> > calculated from ee_start_hi and ee_start_lo entries in the ext4_extent
> > structure.
> > 
> > When disabling extents in the ext4 fs, deleting a given file is working.
> 
> Hi Thomas,
> 
> U-boots ext4 implementation currently does not support 64bit or even 48bit
> block numbers, so this may be the cause.
> 
> Can you provide some information about your test setup?
> 
> You can use the debugsfs ext tool to gather some information about the
> problematic file. Just access the filesystem with:
> 
> /sbin/debugfs /dev/sda1  ;  (or whatever your partion name is)
> 
>   or
> 
> /sbin/debugfs /path/to/imagefile
> 
> debugfs supports commands like cd, stat, ls. stat gives you the block number
> list.
> 
> Kind regards,
> 
> Stefan

Hi Thomas,

short followup:

read_allocated_blocks returns either 0 or -1 in case of an error. 
Unfortunately, the return value is only checked for 0 equality in most/all? 
cases, and seemingly my patch series introduced some more occasions.


Now, what *should* read_allocated_blocks return in case of an error? Either:

- 0: a file block can never be allocated as block 0, as that is always in use 
by the superblock and/or the bootsector block.

- <0: Extents allow 48 bit block numbers. "Limiting" the return value to the 
positive half of int64_t for valid block numbers and and reserving negative 
values for error codes is fine.

I would go for negative error codes, as these are more expressive. Comments/
opinions welcome!


I will update the patch series for correct checking of read_allocated_blocks 
return values and fix all the other block number checks.

Anyway, it would be good to know why *exactly* read_allocated_blocks returns 
an error code in your case. Do you remember the exact negative value returned 
(there are -EINVAL and -ENOMEN, and many several unspecific uses of 0 and -1). 

Can you provide a disk image of the failing file system?

Kind regards,

Stefan