Hi Alex.

The difference is that for MD devices, both types
of metadata are on the same block device. You're
prioritizing which *type of metadata* is checked

how? 1.0 in 1.1 is the same as 1.1 in 1.0...
The only difference would be that one is smaller
than the other, which can hint which is first
and which is second.

for first in that case. For bcache, you'd have to
scan /dev/sdz before /dev/sda if sdz is the cache
and sda is the backing device. Now consider a
few things:

Again, you scan *all* and check *only* for
cache devices.
After that, if none found, you've your list of
devices, if someone found, you activate these
first and then the corresponding backing device.

1.) SCSI/SATA devices may be probed in parallel

And this does not make any difference, in
this context. Probed does not mean necessarily
activated. Maybe you mean probed as activated.
For me it is different.

2.) udev gets events when each device is probed,
*not* after all devices have been probed

This is a udev issue, which can be fixed... :-)

3.) The bcache device may not even be attached
to the system at the time

Good, so the persistency is not needed, I guess,
in that case...
Or, the backing device cannot be activated,
which might be an option, in the current
architecture, but, maybe a bit borderline.

4.) Even in the MD case, there is still *some*
change to the backing device, there is still some
sort of data there that says "hey, there's more."

If you mean the 1.1 in 1.0 (or the other way around),
there is no information telling you there's more,
except, as mentioned, the size, which is not directly
related to device probing.

Otherwise, I do not understand what do you mean.

Even if it doesn't invalidate the other metadata, it
still tells the kernel that it's not enough - think of
it as invalidating it at the logical rather than the
physical level

3 and 4 are the really critical ones. If the cable
that connects the SSD to the computer is flaky,

In this case you've much more serious problems,
I guess, this is not a use case.
The cable can be flaky also after the probing
and activation, and result in a disaster.

Also, you say that the cache must be scanned
before the backing device - but how do you know
it's a cache or a backing device until you've probed it?

The cache has ad "header" with enough information,
namely the UUID(s) of the backing device(s)
So you probe (I use "scan") all devices, sort out
caches, sort out backing and the rest.
Then you activate in proper order.
There are many other alternatives.

You could delay sending any uevents untill all
devices are probed, except there are some devices
that take 30sec timeouts and fail, or iscsi, or devices
that get plugged in at runtime, or...

Those are *all* solvable problems. Some of
them are even too generic. That is, they're
problems in any case.

As I wrote few posts ago, it is clear why it is
like it is. It is *complex* to implement all the
required changes in order to have the backing
device unformatted. Which has, in the end,
limited advantage.

No problem with that, very fine for me, but
telling the it is not possible, it is just,
well, let's say funny.

And since you can't do that, you have a chicken
and egg problem. You can't probe the backing
device before the cache, but you don't know which
is the cache until you probe it. And there may be
more than one of each. You can have one cache
and 200 backing devices, in theory. Want to take
the odds that the cache gets probed first at random?
Because the kernel doesn't have enough information
for it to be anything other than random.

The kernel, again, has to separate the probing
process, from the activation process.

Furthermore, it could always be possible to
configure the booting process to do so, in
an *explicit* way, like md does usually, i.e.
with a configuration file (in initramfs).

bye,

-- 

piergiorgio