Complete support for IO_APPEND flag in fuse
ClosedPublic
Actions

Authored by cem on Jun 25 2017, 2:27 AM.

Details

Reviewers

bapt
rmacklem

Commits

rS320451: Complete support for IO_APPEND flag in fuse

Summary

This finishes what r245164 started and makes open(..., O_APPEND) work again
after r299753.

Pass ioflags, incl. IO_APPEND, down to the direct write backend (r245164 added it to only the bio backend).
(r299753 changed the WRONLY backend from bio to direct.)

PR: 220185
Reported by: Ben RUBSON <ben.rubson at gmail.com>

Diff Detail

Lint

Lint Passed

Unit

No Test Coverage

Build Status

Buildable 10112
Build 10535: arc lint + arc unit

Event Timeline

cem created this revision.Jun 25 2017, 2:27 AM

Harbormaster completed remote builds in B10112: Diff 30049.Jun 25 2017, 2:27 AM

Fix looks good to me. I didn't realize that direct I/O could be happening on
multiple descriptors concurrently, so this would be needed.

Good catch.

This revision is now accepted and ready to land.Jun 25 2017, 12:35 PM

In D11348#234893, @rmacklem wrote:

direct I/O could be happening on
multiple descriptors concurrently, so this would be needed.

It's not a race condition; it happens even with a single descriptor. Just a logic bug where IO_APPEND wasn't getting passed into and handled by the direct write path.

Ok, yes. I was assuming that direct I/O of a single thread would be sequentially
increasing writes that would remain in order (ie. all would grow/append to the
file).

bapt accepted this revision.Jun 28 2017, 8:26 AM

Closed by commit rS320451: Complete support for IO_APPEND flag in fuse (authored by cem). · Explain WhyJun 28 2017, 1:56 PM

This revision was automatically updated to reflect the committed changes.