[NaCl SDK] nacl_io: Add support for non-blocking connect/accept

[NaCl SDK] nacl_io: Add support for non-blocking connect/accept

Move the socket calls that depend on the handle state
into kernel_handle.h/cc.

TEST=new unittest + libcurl and thttpd from naclports
BUG=
R=noelallen@chromium.org

Committed: https://src.chromium.org/viewvc/chrome?view=rev&revision=228290

[Sam Clegg, 2013-10-11 18:09:31]

Still have some dbgprintfs in here to remove, but
thought I'd send this out for early comments.

Works with libcurl and httpd from naclports as well
as passing the new tests.

I discovered a new bug while testing thttpd but thats
for a different CL:
https://code.google.com/p/chromium/issues/detail?id=306592

[noelallen1, 2013-10-11 19:20:24]

Quick look comments.

I'm getting the feeling that we may want to slit more stuff out of MountNode.   
At the moment a Mount node is a combination if the inode and the data or stream.

If we separate out the data/stream portion, driving it as needed from emitters,
it may make the socket stuff cleaner and provide a better place for the lock.

https://codereview.chromium.org/26703008/diff/31001/native_client_sdk/src/lib...
File native_client_sdk/src/libraries/nacl_io/event_emitter_stream.h (right):

https://codereview.chromium.org/26703008/diff/31001/native_client_sdk/src/lib...
native_client_sdk/src/libraries/nacl_io/event_emitter_stream.h:30: uint32_t
BytesInOutputFIFO();
Isn't this part of FIFOInterface?

https://codereview.chromium.org/26703008/diff/31001/native_client_sdk/src/lib...
native_client_sdk/src/libraries/nacl_io/event_emitter_stream.h:31: uint32_t
SpaceInInputFIFO();

Is the idea to get rid of the in_fifo() and out_fifo() bellow?
Maybe you should have BytesInInputFIFO, SapceInOutputFIFO as well.

[Sam Clegg, 2013-10-11 20:56:00]

I'd really like to land this before any major refactor since
its the final piece that is needed to completely replace 
nacl-mounts (we will finally have to all the features of
nacl-mounts and can hopefully kill it!).

Happy to be part of the refactor effort, but hopefully this
change should not be gated on it.

https://codereview.chromium.org/26703008/diff/31001/native_client_sdk/src/lib...
File native_client_sdk/src/libraries/nacl_io/event_emitter_stream.h (right):

https://codereview.chromium.org/26703008/diff/31001/native_client_sdk/src/lib...
native_client_sdk/src/libraries/nacl_io/event_emitter_stream.h:30: uint32_t
BytesInOutputFIFO();
On 2013/10/11 19:20:24, noelallen1 wrote:
> Isn't this part of FIFOInterface?

Yes, but we wanted to remove direct access to the FIFO from
outside the emitter.

https://codereview.chromium.org/26703008/diff/31001/native_client_sdk/src/lib...
native_client_sdk/src/libraries/nacl_io/event_emitter_stream.h:31: uint32_t
SpaceInInputFIFO();
On 2013/10/11 19:20:24, noelallen1 wrote:
> 
> Is the idea to get rid of the in_fifo() and out_fifo() bellow?
> Maybe you should have BytesInInputFIFO, SapceInOutputFIFO as well.

We don't need those at this point, so they would have no
callers.

[Sam Clegg, 2013-10-11 22:02:35]

Removed debugging statements.

[noelallen1, 2013-10-12 00:13:29]

LGTM

https://codereview.chromium.org/26703008/diff/53001/native_client_sdk/src/lib...
File native_client_sdk/src/libraries/nacl_io/event_emitter_pipe.h (right):

https://codereview.chromium.org/26703008/diff/53001/native_client_sdk/src/lib...
native_client_sdk/src/libraries/nacl_io/event_emitter_pipe.h:35: FIFOChar fifo_;
Should remove in_fifo, out_fifo if they are private.

https://codereview.chromium.org/26703008/diff/53001/native_client_sdk/src/lib...
File native_client_sdk/src/libraries/nacl_io/event_emitter_tcp.h (right):

https://codereview.chromium.org/26703008/diff/53001/native_client_sdk/src/lib...
native_client_sdk/src/libraries/nacl_io/event_emitter_tcp.h:39: virtual
FIFOChar* in_fifo() { return &in_fifo_; }
same

https://codereview.chromium.org/26703008/diff/53001/native_client_sdk/src/lib...
File native_client_sdk/src/libraries/nacl_io/event_emitter_tty.h (right):

https://codereview.chromium.org/26703008/diff/53001/native_client_sdk/src/lib...
native_client_sdk/src/libraries/nacl_io/event_emitter_tty.h:32: virtual
FIFOChar* in_fifo() { return &fifo_; }
same

https://codereview.chromium.org/26703008/diff/53001/native_client_sdk/src/lib...
File native_client_sdk/src/libraries/nacl_io/event_emitter_udp.h (right):

https://codereview.chromium.org/26703008/diff/53001/native_client_sdk/src/lib...
native_client_sdk/src/libraries/nacl_io/event_emitter_udp.h:31: virtual
FIFOPacket* in_fifo() { return &in_fifo_; }
same

[Sam Clegg, 2013-10-12 00:24:11]

https://codereview.chromium.org/26703008/diff/53001/native_client_sdk/src/lib...
File native_client_sdk/src/libraries/nacl_io/event_emitter_pipe.h (right):

https://codereview.chromium.org/26703008/diff/53001/native_client_sdk/src/lib...
native_client_sdk/src/libraries/nacl_io/event_emitter_pipe.h:35: FIFOChar fifo_;
On 2013/10/12 00:13:29, noelallen1 wrote:
> Should remove in_fifo, out_fifo if they are private.

Turns out these are pure virtuals that are used by the
the parent class (EventEmitterStream) to access the fifos
declared in the child classes such as this one, so 
we still need them for now.

[Sam Clegg, 2013-10-12 00:26:51]

Committed patchset #6 manually as r228290 (presubmit successful).