BUG REPORT: XFS LOG FORWARD COMPATIBILITY PROBLEM

[Eric Sandeen]

Eddy Zhao wrote:
> Hello Eric:
> 
>>
>> Well, the fact that you have 2 arches would have been highly relevant in
>> your first email.  ;)  Is this the old abi or the new?  Guessing old,
>> based on the alignment issues.
>>
> 
> ARM compiler is producing eabi. The problem lies in: "long long" on arm
> is 8 bytes aligned, on x86 is 4 bytes aligned. So 4 more bytes are
> padded before xfs_inode_log_format::ilf_ino on arm.

A tool that can be very helpful here is "pahole" found in the dwarves
package:

# pahole -C xfs_inode_log_format xfs-arm-eabi.ko
struct xfs_inode_log_format {
	__uint16_t                 ilf_type;          /*     0     2 */
	__uint16_t                 ilf_size;          /*     2     2 */
	__uint32_t                 ilf_fields;        /*     4     4 */
	__uint16_t                 ilf_asize;         /*     8     2 */
	__uint16_t                 ilf_dsize;         /*    10     2 */

	/* XXX 4 bytes hole, try to pack */

	__uint64_t                 ilf_ino;           /*    16     8 */
	union {
		__uint32_t         ilfu_rdev;         /*           4 */
		uuid_t             ilfu_uuid;         /*          16 */
	} ilf_u;                                      /*    24    16 */
	__int64_t                  ilf_blkno;         /*    40     8 */
	__int32_t                  ilf_len;           /*    48     4 */
	__int32_t                  ilf_boffset;       /*    52     4 */

	/* size: 56, cachelines: 1, members: 10 */
	/* sum members: 52, holes: 1, sum holes: 4 */
	/* last cacheline: 56 bytes */
};

That was for an arm eabi xfs.ko from 2.6.25.  Don't worry about the "try
to pack" message; as long as holes match everywhere it's ok ;)

...


>> To be honest, I am not certain that this is "worth fixing" - we already
>> have the requirement that log replay can only be done on an architecture
>> of the same endianness, and a further requirement that it can only be
>> done on architectures with similar alignment rules doesn't seem so bad.
>>
> 
> Where is the official description of such requirment? The requirment is
> not obvious to new XFS users, and they should be warned :)

For the endian requirement ... not sure where it's doc'd to be honest.
if you don't find it in Documentation/filesystems/xfs.txt feel free to
send a patch ;)

> Why is such requirment? For backward compatibility?

The xfs log was done native-endian for performance reasons, I think.

> I think the requirment is bad. It inhibit user from exchanging data
> freely between systems, which is an obvious requirment for file system
> design.

No, it just means that you have to cleanly unmount on one filesystem
before you try to mount it on another different system.

>> The problem is, if you go off and "fix" this on arm now, you'll find
>> that logs created under older arm kernels can't be replayed on newer arm
>> kernels and vice versa...
> 
> Compared with the above restriction("requirment"), we can afford this :)
> 
> Do I miss any other log/metadata needs pack on 2.6.10?

I can't really speak to 2.6.10, I don't want to go back that far ;)  I'm
afraid  you're on your own there.

So just to be clear, on 2.6.28, arm eabi logs replay fine on x86_64?

-Eric