Re: dm bug: hibernate to swap located on dm-integrity doesn't work (how to get data redundancy for swap?)

[Askar Safin <safinaskar@gmail.com>]

Milan Broz <gmazyland@gmail.com>:
> Hi,

That patch doesn't fix the problem. I will send more details within some
days, hopefully today.

Also, I just found that for reliable reproduction you need to do
"swapoff /dev/mapper/swap; cat /dev/mapper/swap > /dev/null" after resume
(assuming you were able to resume, of course).

So here is updated script for reproduction in Qemu:

https://zerobin.net/?0aa379bae218cf92#DDVFMvfi6S3ydCQLSrL+us1lHjXQJIZasW55JI7gEfU=

This script is very easy to use!

Try 1-2 times to reproduce.

Here is result of that script:

https://zerobin.net/?3d9447900052f9ce#ng0htJDAdSsvqVunL+BnoLHXszM6ardt9R3wkbO9L28=

This results are on 43e9ad0c55a3, which is current master, without Mikulas Patocka's
patch, but with this Mario's patch:
https://lore.kernel.org/linux-pm/20251026033115.436448-1-superm1@kernel.org/ .
Mario's patch is needed, otherwise we get WARNING when we try to hibernate.

Again: these logs are without Mikulas Patocka's patch. I will send results
of testing his patch later, hopefully today. But don't expect much. As I said,
his patch doesn't work.

"log-def-1" is output of first Qemu invocation (i. e. first boot) with
default integritysetup options. "log-def-2" is second Qemu invocation
(i. e. when we try to resume).

log-bit-{1,2} is same thing, but with "--integrity-bitmap-mode" added to
"integritysetup format" and "integritysetup open".

log-no-{1,2} is same, but with "--integrity-no-journal".

log-nodm-{1,2} is same, but without dm-integrity at all, i. e. we create
swap directly on partition.

As you can see in logs, hibernate with dm-integrity never works perfectly.
We either unable to resume, either we are able to resume, but then get
integrity errors when we do "cat /dev/mapper/swap > /dev/null".

Swap directly on partition works.

Again: you may need 1-2 attempts to reproduce using this script.

Also: the bug is reproducible even if we do "echo test_resume > /sys/power/disk".

> Are you sure you used --integrity-no-journal both in activation before
> hibernation and also in resume? If not, please try it.

I'm totally sure. (You can see script above and "set -x" output in logs.)

> You can verify it with "integritysetup status <device>" - it should say "journal: not active".

I just checked. It indeed says so.

> And if it does not work, could you try to use -integrity-recovery-mode the same
> way (both before hibernation and later in resume)? This will effectively ignore checksums

So I should pass it to both "integritysetup open" invocations, but
not to "integritysetup format" invocation. Right? Okay, I did so.

I. e. I did this:

integritysetup format --batch-mode --integrity xxhash64 /dev/sda # when formatting
integritysetup open --integrity-recovery-mode --integrity xxhash64 /dev/sda swap # before hibernate
integritysetup open --integrity-recovery-mode --integrity xxhash64 /dev/sda early-swap # when resuming

And something completely unexpected happened!
"swapon" failed immediately after "mkswap"!!! I. e. "swapon" was
unable to read swap signature right after "mkswap".
Here is log:
https://zerobin.net/?ebe34fc9ce94be45#6DWKSXvgUDKIrF4299th0ylhQNEcdqeeBfxzSJjROpA=
This seems like another dm-integrity bug.

> You can verify it with "integritysetup status <device>" - it should say "mode: read/write recovery".

Yes, it says so in logs above.

> You can also try to decrease journal commit time with --journal-commit-time option,

I just put "--journal-commit-time 1" to format and both opens. I got the same
result I get with default options: i. e. blkid returns "swap" instead of
"swsuspend", when I try to resume. Here are logs:
https://zerobin.net/?c5f8320eb89b1cfb#drrxRgnGk817oEDUA8idhn+WEQgocWjtbsAYuEHF5rI= .

> Redundancy? You mean data integrity protection? There is no redundancy, only additional authentication tag
> (detecting integrity error but not correcting it).

Yes, I meant integrity protection.

-- 
Askar Safin