Abort ImportantFileWriter::WriteFileAtomically() on Flush() error.

Abort ImportantFileWriter::WriteFileAtomically() on Flush() error.

As discussed on chromium-os-dev [1], if there is no fsync() after writing to the
temp file, atomicity is not guaranteed.  Thus, to be able to guarantee
atomicity, we must check the return value of Flush() and bail in case Flush()
wasn't successful.

I'll update the UMA histogram enum in a separate CL.

[1] https://groups.google.com/a/chromium.org/forum/?hl=en#!topic/chromium-os-dev/Eef8gNIRwjc

BUG=none

Committed: https://crrev.com/604e5c185a4f6a9c1a59525a45243e714983262a
Cr-Commit-Position: refs/heads/master@{#300230}

[Thiemo Nagel, 2014-10-16 09:13:38]

tnagel@chromium.org changed reviewers:
+ mnissler@chromium.org, rvargas@chromium.org

[Thiemo Nagel, 2014-10-16 09:13:38]

Hi Ricardo,

could you please take a look?

Thank you!
Thiemo

[rvargas (doing something else), 2014-10-17 23:03:24]

I have a question: The CL looks strictly speaking correct, but I wonder if it is
meant to fix any real issue that we are seeing.

I guess we'll find out if flush actually fails sometimes.

lgtm

[rvargas (doing something else), 2014-10-17 23:05:46]

On 2014/10/17 23:03:24, rvargas wrote:
> I have a question: The CL looks strictly speaking correct, but I wonder if it
is
> meant to fix any real issue that we are seeing.
> 
> I guess we'll find out if flush actually fails sometimes.
> 
> lgtm

(it's usually better to update histogram definitions with the code)

[Thiemo Nagel, 2014-10-20 06:49:39]

The CQ bit was checked by tnagel@chromium.org

[commit-bot: I haz the power, 2014-10-20 06:50:24]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/663463002/20001

[Thiemo Nagel, 2014-10-20 06:55:45]

On 2014/10/17 23:05:46, rvargas wrote:
> I have a question: The CL looks strictly speaking correct, but I wonder if it
is
> meant to fix any real issue that we are seeing.

When reading the code, I noticed that in case fsync() fails, atomicity
cannot be guaranteed (at least on ext4), thus I thought it would be more
consistent if the function returned false instead of running the risk
of being non-atomic.  It's not meant to fix a specific issue.

> I guess we'll find out if flush actually fails sometimes.

Yes, that will be interesting.

> (it's usually better to update histogram definitions with the code)

Agreed, however this wasn't possible because the definitions reside in a
different
source tree.

[commit-bot: I haz the power, 2014-10-20 09:31:15]

Committed patchset #2 (id:20001)

[commit-bot: I haz the power, 2014-10-20 09:32:24]

Patchset 2 (id:??) landed as
https://crrev.com/604e5c185a4f6a9c1a59525a45243e714983262a
Cr-Commit-Position: refs/heads/master@{#300230}