UBI: Cannot create static volumes when FTL is enabled

UBI: Cannot create static volumes when FTL is enabled

[John Smith]

Hi,

Probably not a major bug, but here it is....

I have a slightly muddled kernel using MTD and UBI from a recent snapshot
from kernel.org, and the rest of the kernel from a derivative of 2.6.18.

I accidentally enabled the CONFIG_FTL as well as MTD and UBI.

When trying to create a static UBI volume with the command

  # ubimkvol -d 0 -n 1 -N ubi-volume --type=static -s 123000

I trigger a kernel stack dump. It arises from the ubi_assert

   ubi_assert(vol->vol_type != UBI_STATIC_VOLUME);

in function ubi_eba_read_leb() of file  drivers/mtd/ubi/eba.c

The rest of the call trace reads:
Call Trace:
 [<801d2e04>] gluebi_read+0x144/0x1bc        drivers/mtd/ubi/gluebi.c:132
 [<801aa2b4>] scan_header+0x84/0x248         drivers/mtd/ftl.c:178
 [<801aba18>] ftl_add_mtd+0x2c/0xcc          drivers/mtd/ftl.c:1045
 [<801a961c>] blktrans_notify_add+0x50/0x7c  drivers/mtd/mtd_blkdevs.c:336
 [<801a5d74>] add_mtd_device+0xd4/0x148      drivers/mtd/mtdcore.c:75
 [<801d32f0>] ubi_create_gluebi+0xd8/0x170   drivers/mtd/ubi/gluebi.c:297
 [<801c459c>] ubi_create_volume+0x43c/0x854  drivers/mtd/ubi/vmt.c:316
 [<801c7f60>] ubi_cdev_ioctl+0x1d0/0x2f0     drivers/mtd/ubi/cdev.c:630
 [<8008b6d4>] do_ioctl+0x54/0x7c             fs/ioctl.c:34
 [<8008b8e0>] vfs_ioctl+0x78/0x220           fs/ioctl.c:153
 [<8008bad8>] sys_ioctl+0x50/0x98            fs/ioctl.c:173

Regards
John

John Smith
Blackburn,
England.

Re: UBI: Cannot create static volumes when FTL is enabled

[Artem Bityutskiy]

On Fri, 2007-05-04 at 23:22 +0100, John Smith wrote:
> Probably not a major bug, but here it is....
> 
> I have a slightly muddled kernel using MTD and UBI from a recent snapshot
> from kernel.org, and the rest of the kernel from a derivative of 2.6.18.
> 
> I accidentally enabled the CONFIG_FTL as well as MTD and UBI.
> 
> When trying to create a static UBI volume with the command
> 
>   # ubimkvol -d 0 -n 1 -N ubi-volume --type=static -s 123000
> 
> I trigger a kernel stack dump. It arises from the ubi_assert
> 
>    ubi_assert(vol->vol_type != UBI_STATIC_VOLUME);
> 
> in function ubi_eba_read_leb() of file  drivers/mtd/ubi/eba.c

Yeah, you have found a flaw. We actually assume that mtd devices are
emulated _only_ for dynamic volumes, but we do not disable the MTD
device emulation for static volumes. This should we fixed.

The reason why only dynamic volumes may have emulated MTD devices is
that they have "stable" size. I mean, if you create dynamic volume of
size N, you can always read N bytes from it. And it does not depend on
how many bytes of data you have on them.

In case of static volumes the situation is different. When you create an
empty static volume, it contains no data, and a read operation returns
EOF immediately. Then you write put some data to it using the "volume
update" operation. In this case you'll be able to read that amount of
data back.

Here is an example:
1. Create static volume A of size Sa.
2. Create dynamic volume B of size Sb.
3. Read whole A - 0 bytes are read.
4. Read whole B - Sb bytes are read, all of them contain 0xFFs.
5. Update A with Xa bytes of data (Xa <= Sa).
6. Update B with Xb bytes of data (Xb <= Sb).
7. Read whole A - Sa bytes are read.
8. Read whole B - Xb bytes are read.

See the difference in semantics? In case of dynamic volumes we always
read Sb bytes. If there is only Xb bytes of data, the rest will be 0xFF
bytes.

So you see that dynamic volumes are mapped to the MTD device model quite
nicely: mtd device size = volume size.

Size of static volumes depends on how much data you put there. And since
MTD infrastructure assumes that MTD device sizes are stable, there is no
good and nice way to emulate MTD devices on top of static volumes,
unless MTD subsystem is somehow re-designed.

But anyway, you have found a bug. I am going to disable MTD device
emulation for static volumes.

Thanks.

-- 
Best regards,
Artem Bityutskiy (Битюцкий Артём)

Re: UBI: Cannot create static volumes when FTL is enabled

[Artem Bityutskiy]

On Sat, 2007-05-05 at 14:39 +0300, Artem Bityutskiy wrote:
> So you see that dynamic volumes are mapped to the MTD device model quite
> nicely: mtd device size = volume size.
> 
> Size of static volumes depends on how much data you put there. And since
> MTD infrastructure assumes that MTD device sizes are stable, there is no
> good and nice way to emulate MTD devices on top of static volumes,
> unless MTD subsystem is somehow re-designed.
> 
> But anyway, you have found a bug. I am going to disable MTD device
> emulation for static volumes.

Although it still would be nice to have MTD device emulation for static
volumes. May be we have the same semantics as dynamic volumes if they
are used via gluebi. It needs more thinking.

-- 
Best regards,
Artem Bityutskiy (Битюцкий Артём)

Re: UBI: Cannot create static volumes when FTL is enabled

[Artem Bityutskiy]

On Sat, 2007-05-05 at 14:54 +0300, Artem Bityutskiy wrote:
> On Sat, 2007-05-05 at 14:39 +0300, Artem Bityutskiy wrote:
> > So you see that dynamic volumes are mapped to the MTD device model quite
> > nicely: mtd device size = volume size.
> > 
> > Size of static volumes depends on how much data you put there. And since
> > MTD infrastructure assumes that MTD device sizes are stable, there is no
> > good and nice way to emulate MTD devices on top of static volumes,
> > unless MTD subsystem is somehow re-designed.
> > 
> > But anyway, you have found a bug. I am going to disable MTD device
> > emulation for static volumes.
> 
> Although it still would be nice to have MTD device emulation for static
> volumes. May be we have the same semantics as dynamic volumes if they
> are used via gluebi. It needs more thinking.

Well, after thinking more I have fixed this. Now we set emulated mtd
device size to data size if the volume is static.

I.e., when you create a static volume, its size is zero:

# ubimkvol -d0 -n0 -Nvol1 -s60MiB -t static
# cat /sys/class/ubi/ubi0_0/data_bytes
# 0

# cat /proc/mtd
# dev:    size   erasesize  name
# mtd0: 08000000 00004000 "NAND simulator partition 0"
# mtd1: 00000000 00003e00 "vol1"

# ubiupdatevol -d 0 -n 0 small_file_72_bytes
# cat /sys/class/ubi/ubi0_0/data_bytes
# 72

# cat /proc/mtd
# dev:    size   erasesize  name
# mtd0: 08000000 00004000 "NAND simulator partition 0"
# mtd1: 00000048 00003e00 "vol1"

I've committed the fix to git://git.infradead.org/~dedekind/ubi-2.6.git.
Or see
http://git.infradead.org/?p=users/dedekind/ubi-2.6.git;a=commit;h=962f24709f91fe0634b3faa3cac915c66357a239

-- 
Best regards,
Artem Bityutskiy (Битюцкий Артём)