Retrieving number of free unused eraseblocks in a UBI filesystem

[Boris Brezillon]

On Wed, 7 Dec 2016 14:47:13 +0000
Martin Townsend <mtownsend1973 at gmail.com> wrote:

> On Wed, Dec 7, 2016 at 2:17 PM, Boris Brezillon
> <boris.brezillon at free-electrons.com> wrote:
> > On Wed, 7 Dec 2016 13:55:42 +0000
> > Martin Townsend <mtownsend1973 at gmail.com> wrote:
> >  
> >> On Wed, Dec 7, 2016 at 1:14 PM, Boris Brezillon
> >> <boris.brezillon at free-electrons.com> wrote:  
> >> > On Wed, 7 Dec 2016 13:07:25 +0000
> >> > Martin Townsend <mtownsend1973 at gmail.com> wrote:
> >> >  
> >> >> On Wed, Dec 7, 2016 at 12:39 PM, Boris Brezillon
> >> >> <boris.brezillon at free-electrons.com> wrote:  
> >> >> > On Wed, 7 Dec 2016 11:59:13 +0000
> >> >> > Martin Townsend <mtownsend1973 at gmail.com> wrote:
> >> >> >  
> >> >> >> Hi,
> >> >> >>
> >> >> >> I'm running a 4.1 Kernel and have a UBI Filesystem with 2 volumes
> >> >> >> taking up all the NAND.
> >> >> >> ubinfo -d 0
> >> >> >> ubi0
> >> >> >> Volumes count:                           2
> >> >> >> Logical eraseblock size:                 126976 bytes, 124.0 KiB
> >> >> >> Total amount of logical eraseblocks:     4016 (509935616 bytes, 486.3 MiB)
> >> >> >> Amount of available logical eraseblocks: 0 (0 bytes)
> >> >> >> Maximum count of volumes                 128
> >> >> >> Count of bad physical eraseblocks:       56
> >> >> >> Count of reserved physical eraseblocks:  24
> >> >> >> Current maximum erase counter value:     15
> >> >> >> Minimum input/output unit size:          2048 bytes
> >> >> >> Character device major/minor:            249:0
> >> >> >> Present volumes:                         0, 1
> >> >> >>
> >> >> >> So I'm guessing that the Total amount of logical erase blocks is 0 as
> >> >> >> this is because I have 2 volumes of a fixed size that take up all the
> >> >> >> available eraseblocks of the /dev/ubi0 device.
> >> >> >>
> >> >> >> Is there a way of getting, per volume preferably, the number of
> >> >> >> eraseblocks that aren't being used? Or conversely get the number of
> >> >> >> eraseblocks that are used as I can work it out from the total amount
> >> >> >> of logical eraseblocks.  
> >> >> >
> >> >> > cat /sys/class/ubi/ubiX_Y/reserved_ebs
> >> >> >
> >> >> > where X is the UBI device id and Y is the volume id.
> >> >> >  
> >> >>
> >> >> Thanks for the swift reply, this seems to give me the total number of
> >> >> eraseblocks:
> >> >>
> >> >> # cat /sys/class/ubi/ubi0_1/reserved_ebs
> >> >> 3196
> >> >>
> >> >> # ubinfo -d0 -n1
> >> >> Volume ID:   1 (on ubi0)
> >> >> Type:        dynamic
> >> >> Alignment:   1
> >> >> Size:        3196 LEBs (405815296 bytes, 387.0 MiB)
> >> >> State:       OK
> >> >> Name:        rootfs
> >> >>
> >> >> # df -h .
> >> >> Filesystem      Size  Used Avail Use% Mounted on
> >> >> ubi0:rootfs     357M  135M  222M  38% /
> >> >>
> >> >> What I would like to know is how many of them are being used and how
> >> >> many are not.  
> >> >
> >> > Well, for static volumes, you could extract this information from the
> >> > data_bytes sysfs file. It doesn't work for dynamic volumes though, but
> >> > even if you could extract this information from the number of mapped
> >> > LEBs, not sure it would have any meaning, because unmapped LEBs can
> >> > still be reserved by the upper layer, and just because you have a lot
> >> > of unmapped LEBs does not mean you can shrink the volume (which I guess
> >> > is your ultimate goal).
> >> >  
> >>
> >> My goal is for failover, the SOM we use has 2 flash devices, the
> >> primary is UBI on NAND, which we keep in sync with the other flash
> >> device which is eMMC.  I'm basically monitoring both erase counts
> >> using Richard's ubihealthd patch and bad blocks as an indication of
> >> when to failover.  Maybe naively I'm thinking if I use the number of
> >> unused eraseblocks it could serve 2 purposes failover due to badblocks
> >> which will decrease available blocks naturally and also fail over to
> >> the larger secondary flash device if it becomes near full.
> >>
> >> If I stick to using bad blocks as a metric can I safely assume that
> >> every time a block is marked bad it will decrease the overall size of
> >> the filesystem as seen by utilities such as 'df'?  
> >
> > That's not the case. UBI reserves some amount of blocks to deal with
> > bad blocks, which means your FS available size will not decrease, or at
> > least, not until you are in a situation where you run out of erase
> > blocks reserved for bad block handling. And even in that case, assuming
> > the UBI device still has some blocks available (i.e. not reserved for a
> > volume or for internal UBI usage),  those blocks will be used when a
> > new bad block is discovered.  
> 
> Thanks for this it clears up a misconception I had, I was assuming it
> would take blocks from those allocated to the volumes which is not the
> case.  So if I reserve say 50 blocks for bad block management and then
> split the remaining blocks between the 2 volumes and then once 50
> blocks have gone bad, the next bad block will result in a corrupt
> filesystem?

Hm, I'm not sure it will directly result in a FS corruption, since some
blocks might actually be available at the time the FS tries to map a
LEB. But it's likely to result in a UBI/UBIFS error at some point (when
your FS is full and UBI fails to pick a PEB for a new LEB).

> 
> >  
> >> If so maybe it's
> >> best if I stick to a combination of number of bad blocks from MTD in
> >> sysfs and free space in filesystem.  
> >
> > Not sure free FS space is a good metric either. Maybe you should just
> > monitor the number of available blocks and the number of blocks
> > reserved for bad block handling at the UBI level (not sure those
> > numbers are exposed through sysfs though). When these numbers are too
> > low, you should consider switching to the other storage.  
> 
> I'm having a dig through the code and if I expose used_ebs from struct
> ubi_volume to sysfs would this give me what I want, the comment say
> "how many logical eraseblocks in this volume contain data" which looks
> like exactly what I want.

->used_ebs is always set to ->reserved_pebs for dynamic volumes, so no.

> I could then monitor the bad blocks and if
> they get to the reserved bad block amount fail over and then use
> used_ebs to see if it gets within a % of the "reserved_ebs" value from
> sysfs?

IMO, you should just rely on information exposed at the UBI device
level (/sys/class/ubi/ubiX/).

You already have 'reserved_for_bad' and 'avail_eraseblocks' exposed,
and this should be enough. When reserved_for_bad approaches 0, then you
should be near the EOL of your UBI device.