Differential D33389

fusefs: implement VOP_ALLOCATE
ClosedPublic
Actions

Authored by asomers on Dec 11 2021, 8:31 PM.

Details

Reviewers

pfg

Commits

rGeccfee8330e8: fusefs: implement VOP_ALLOCATE
rG398c88c7582a: fusefs: implement VOP_ALLOCATE

Summary

Now posix_fallocate will be correctly forwarded to fuse file system
servers, for those that support it.

MFC after: 2 weeks

Test Plan

Test cases added

Diff Detail

Repository

rG FreeBSD src repository

Lint

Lint Not Applicable

Unit

Tests Not Applicable

Event Timeline

asomers created this revision.Dec 11 2021, 8:31 PM

Herald added a subscriber: imp. · View Herald TranscriptDec 11 2021, 8:31 PM

asomers requested review of this revision.Dec 11 2021, 8:31 PM

Harbormaster completed remote builds in B43295: Diff 99858.Dec 11 2021, 8:32 PM

I don't know if fuse does this by default, but if IO_SYNC is
specified in the flags argument to VOP_ALLOCATE(), the
allocation must be committed to stable storage (disk or ???)
before VOP_ALLOCATE() returns.

Just fyi, in case you were not aware of it.

In D33389#754891, @rmacklem wrote:

I don't know if fuse does this by default, but if IO_SYNC is
specified in the flags argument to VOP_ALLOCATE(), the
allocation must be committed to stable storage (disk or ???)
before VOP_ALLOCATE() returns.

Just fyi, in case you were not aware of it.

I was not aware of that. It looks like the NFS server will set IO_SYNC? Is there any other way call it with that flag set? FUSE has no equivalent, but I could follow the FUSE_ALLOCATE with a call to FUSE_FSYNC.

I am not aware of any other way that IO_SYNC can get
set, except for the NFSv4.2 server.
Doing FUSE_FSYNC after FUSE_ALLCOATE for that
case sounds fine to me.

In D33389#754893, @rmacklem wrote:

I am not aware of any other way that IO_SYNC can get
set, except for the NFSv4.2 server.
Doing FUSE_FSYNC after FUSE_ALLCOATE for that
case sounds fine to me.

This is going to take me awhile. I can't use my usual test suite, so I need to test with an actual fuse file system exported over NFS. But to do that, I first need to write a FUSE file system that works with NFS 4.2 and posix_fallocate.

Sure, I don't see this as anywhere near urgent.
1 - Few use NFSv4.2 (as you noted).
2 - The VOP_FSYNC() only matters if the server

crashes just after doing the VOP_ALLOCATE()
and loses the changes.

Oh, and because this (plus the cred argument) are not in stable/13,
the FreeBSD NFSv4.2 Allocate is broken in FreeBSD 13. I should
probably do a commit to completely disable it instead of only
having it disabled by default.