[NaCl SDK] nacl_io: Always create directory node in http filesystem root.

[NaCl SDK] nacl_io: Always create directory node in http filesystem root.

The root of all filesystems should be a directory node
except in the special case of blob URL where we maintain
the legacy behavior of having the mount point be a file
containing the blob content.

Also, fix the Access() call for http filesystem so that
it honors the access bits of the node itself.

This fixes an issue in naclports where a mount() of
an http filsystem followed by a chdir(http_root) would
always fail with ENOTDIR.

TEST=nacl_io_test

Committed: https://crrev.com/eed34ebc773dc3372565ac0b3a0ecea1a94a2631
Cr-Commit-Position: refs/heads/master@{#294000}

[Sam Clegg, 2014-09-05 18:39:16]

sbc@chromium.org changed reviewers:
+ binji@chromium.org

[binji, 2014-09-08 18:25:34]

What is the bug in naclports? If you mount a blob, then in a sense you are
mounting just that file. Why should you be able to able to chdir to it?

I suppose it is inconsistent with the traditional idea of mounting, but this
change will break some users (specifically the one who requested this change).

[Sam Clegg, 2014-09-08 18:31:00]

On 2014/09/08 18:25:34, binji wrote:
> What is the bug in naclports? If you mount a blob, then in a sense you are
> mounting just that file. Why should you be able to able to chdir to it?
> 
> I suppose it is inconsistent with the traditional idea of mounting, but this
> change will break some users (specifically the one who requested this change).

Yes, this is bug that is currently blocking the switch to ppapi_simple of sdl.

The problem is that even for non-blobs you can never currently do this:

mount(http)
chdir(httpdir)

Because http_fs.c will try to fetch the root URL, and even if it can (in the
case of httpd.py is can) it will never report this as a directory.  So chdir()
will always fail with "No a directory".

I do understand that this will break some use cases.  I would modify this change
to keep the old blob behaviour I guess, despite it going againt the normal
semantics of mount().  Maybe that would be best.   What do you think?

[binji, 2014-09-08 18:52:52]

On 2014/09/08 18:31:00, Sam Clegg wrote:
> On 2014/09/08 18:25:34, binji wrote:
> > What is the bug in naclports? If you mount a blob, then in a sense you are
> > mounting just that file. Why should you be able to able to chdir to it?
> > 
> > I suppose it is inconsistent with the traditional idea of mounting, but this
> > change will break some users (specifically the one who requested this
change).
> 
> Yes, this is bug that is currently blocking the switch to ppapi_simple of sdl.
> 
> The problem is that even for non-blobs you can never currently do this:
> 
> mount(http)
> chdir(httpdir)
> 
> Because http_fs.c will try to fetch the root URL, and even if it can (in the
> case of httpd.py is can) it will never report this as a directory.  So chdir()
> will always fail with "No a directory".
> 
> I do understand that this will break some use cases.  I would modify this
change
> to keep the old blob behaviour I guess, despite it going againt the normal
> semantics of mount().  Maybe that would be best.   What do you think?

I'd rather keep the old behavior for now, and consider changing it in another
CL.

[Sam Clegg, 2014-09-08 23:31:02]

Done.

[binji, 2014-09-09 17:08:01]

lgtm

[Sam Clegg, 2014-09-09 17:50:33]

Tracked down an uninitialized member variable (is_blob_url_).

In doing so, factored out FindExistingNode().

PTAL.

[binji, 2014-09-09 19:02:27]

On 2014/09/09 17:50:33, Sam Clegg wrote:
> Tracked down an uninitialized member variable (is_blob_url_).
> 
> In doing so, factored out FindExistingNode().
> 
> PTAL.

nice, thanks. slgtm

[Sam Clegg, 2014-09-09 19:14:40]

The CQ bit was checked by sbc@chromium.org

[commit-bot: I haz the power, 2014-09-09 19:16:16]

CQ is trying da patch. Follow status at
 https://chromium-status.appspot.com/cq/sbc@chromium.org/547713002/40001

[commit-bot: I haz the power, 2014-09-09 20:32:06]

Committed patchset #3 (id:40001) as 8705bc9f7a4803563ad712ef0dd76795dd47dd11

[commit-bot: I haz the power, 2014-09-10 03:55:18]

Patchset 3 (id:??) landed as
https://crrev.com/eed34ebc773dc3372565ac0b3a0ecea1a94a2631
Cr-Commit-Position: refs/heads/master@{#294000}