Re: [PATCH v3 00/10] liveupdate: Preserve tmpfs files across Live Update

[David Matlack]

On Tue Nov 10, 2026 at 7:45 AM PST, Pratyush Yadav wrote:
> On Thu, Nov 05 2026, David Matlack wrote:
>> On 2026-11-05 04:42 PM, Christian Brauner wrote:
>>> The checks on the target are trivial: a writable fd to an empty shmem
>>> file with the inode lock held while you fill it. A tmpfs file can't be
>>> sealed, it only ever has F_SEAL_SEAL, so there's nothing to restore but
>>> contents and size anyway.
>>
>> Sounds good. The one other check I plan to add is rejecting a target
>> that LUO is already managing, e.g. a file that is preserved in an
>> outgoing session and whose size is frozen.
>
> I have been playing with this idea too, and I also stumbled onto this
> problem.
>
> You also need to make sure a preserve doesn't race with a retrieve.
> Since userspace has access to the file, it can do both at the same time.
> I think LUO should expose a "register file" function that registers the
> file in luo_preserved_files. This should be the very first thing
> retrieve does.

Agreed, that's what I have queued up for v4, which I'm hoping to send by
tomorrow.

LUO gets two new functions for .retrieve() callbacks:

 - liveupdate_claim_file() inserts the target into luo_preserved_files,
   and fails with -EBUSY if it's already there, e.g. because it's
   preserved in an outgoing session or was retrieved before.

 - liveupdate_unclaim_file() undoes it. If .retrieve() claimed the file
   it returns, LUO doesn't insert it again after the callback, and a
   callback that fails after claiming must unclaim.

On the preserve side, LUO inserts the file into luo_preserved_files
before calling the file handler .preserve(), so whichever of the two
inserts first wins and the other fails with -EBUSY.

> Also, today LUO core expects a fresh file from file handlers, so it
> always installs a FD to that file in the task. When doing the "retrieve
> into FD" thing, that will result in two FDs to the same file. While it
> isn't a strict problem, it seems kinda ugly from a uAPI prespective.
>
> Perhaps we should let retrieve signal to LUO core that "I already
> installed the file", or a "I have no file to install", or something
> similar?

Makes sense, I'll do that in v4. The file handler will tell LUO that it
restored into an existing fd, and LIVEUPDATE_SESSION_RETRIEVE_FD will
return that fd instead of installing a new one. Retrieving the token
again later still returns a new fd for the same file, like it does today
for any file that was already retrieved.