[linux-lvm] Using Oracle with lvm AND rawio: read(512) from

[Michael Ju. Tokarev]

"Michael Ju. Tokarev" wrote:
> 
> The only way I have to try is to use stock kernel with clean lvm
> patch, that I'll do at monday.

Ok, installed now:
  stock 2.2.17 kernel with
    rawio patch (from lvm source)
    lvm-0.9-2.2.17-stock.patch
  lvmtools 0.9

 # raw /dev/raw/raw100 /dev/vg0/lv0
 /dev/raw/raw100:        bound to major 58, minor 0
 # dd if=/dev/zero of=/dev/raw/raw100 bs=512
 dd: /dev/raw/raw100: Invalid argument
 1+0 records in
 0+0 records out
 # _

 # raw /dev/raw/raw100 /dev/sda4
 dev/raw/raw100:        bound to major 8, minor 4
 # dd if=/dev/zero of=/dev/raw/raw100 bs=512
 dd: /dev/raw/raw100: No such device or address
 131073+0 records in
 131072+0 records out
 # _

So, as I can see, "stock lvm" also won't work with rawio
with 512-byte block i/o, while disk partition with rawio
is pretty happy with that block size.

Question: *why* lvm+rawio won't work with 512-bytes i/o?
And this is not an oracle question, oracle here is just
an example application that really uses 512-byte i/o.

I see in linux/include/linux/lvm.h :

 [...]
 #ifdef BLOCK_SIZE
 #undef BLOCK_SIZE
 #endif

 #ifdef CONFIG_ARCH_S390 
 #define BLOCK_SIZE     4096
 #else
 #define BLOCK_SIZE     1024  <<<<<<<<<========
 #endif

 #ifndef        SECTOR_SIZE
 #define SECTOR_SIZE    512
 #endif
 [...]

I'm not shure if it is legal to change this value...

Next 2 questions: can I change this value to 512?
And will this cure the problem?  For now I can't
experiment as I have no physical access to that
machine (only over dialup modem), and I wan't to
risk rebooting it again with unknown kernel...

BTW, why it ever defined?  I see that this value used
to initialize lvm_blocksizes[] static array, and that
array isn't changed anywhere, only read.  And only
in lvm_snap.c, if #defined DEBUG_SNAPSHOT (it will not
compile if this #defined, since that array referenced
here is static in other source) it is checked if underlying
device has different block size, and that value 
instead in further calculations.
Why lvm_blocksize[] is used at all while it is constant
array ?! ;)

Little *strong* suggestion.  Please rename BLOCK_SIZE
and SECTOR_SIZE to LVM_BLOCK_SIZE and LVM_SECTOR_SIZE
in lvm.h -- currently used names are very common to
be a source of namespace collisions.

Regards,
 Michael.