Change to use a unix socket instead of TCP.

Change to use a unix socket instead of TCP.

This change makes Trousers use a unix domain socket instead of a TCP socket.  This allows tcsd to start without waiting on the network.

Change-Id: Ic66830a03d35bbe503e9a91ca2ebbbd96924fb42

BUG=7892
TEST=install, test cryptohome tpm ownership, crypto, and status

Committed: http://chrome-svn/viewvc/chromeos?view=rev&revision=987b8de

[Luigi Semenzato, 2010-10-20 19:36:26]

Generally looks fine after you address a couple of issues below.

Strategy-wise, if you put this code under control of a flag (-u for unix
domain), I think we have good chances of being able to upstream it.  But since
it's good to commit this quickly, you are welcome to leave it for later.

http://codereview.chromium.org/3958001/diff/1/3
File src/tcsd/svrside.c (right):

http://codereview.chromium.org/3958001/diff/1/3#newcode256
src/tcsd/svrside.c:256: strcpy(serv_addr.sun_path, "/var/run/trousers");
Should /var/run/trousers be defined as a string in a single place? There are two
uses.  (I know, I know, it's a pain.)

http://codereview.chromium.org/3958001/diff/1/3#newcode257
src/tcsd/svrside.c:257: unlink(serv_addr.sun_path);
Check for errors here?

http://codereview.chromium.org/3958001/diff/1/4
File src/tspi/rpc/tcstp/rpc.c (right):

http://codereview.chromium.org/3958001/diff/1/4#newcode362
src/tspi/rpc/tcstp/rpc.c:362: LogDebug("Sending TSP packet to host %s.",
hte->hostname);
Is hte->hostname still meaningful?  Could it be garbage?

[fes, 2010-10-20 20:16:21]

Fixed, PTAL.

The -u option is a good idea, but would complicate use of the client library.  I
agree that for now, it's best to get this fix in without an upstream-able fix.

http://codereview.chromium.org/3958001/diff/1/3
File src/tcsd/svrside.c (right):

http://codereview.chromium.org/3958001/diff/1/3#newcode256
src/tcsd/svrside.c:256: strcpy(serv_addr.sun_path, "/var/run/trousers");
On 2010/10/20 19:36:26, Luigi Semenzato wrote:
> Should /var/run/trousers be defined as a string in a single place? There are
two
> uses.  (I know, I know, it's a pain.)

Done.

http://codereview.chromium.org/3958001/diff/1/3#newcode257
src/tcsd/svrside.c:257: unlink(serv_addr.sun_path);
On 2010/10/20 19:36:26, Luigi Semenzato wrote:
> Check for errors here?

I could, but this is implicitly checked by calling bind(), no?  If the
filesystem socket exists, it should return EADDRINUSE, unless I'm mistaken.

http://codereview.chromium.org/3958001/diff/1/4
File src/tspi/rpc/tcstp/rpc.c (right):

http://codereview.chromium.org/3958001/diff/1/4#newcode362
src/tspi/rpc/tcstp/rpc.c:362: LogDebug("Sending TSP packet to host %s.",
hte->hostname);
On 2010/10/20 19:36:26, Luigi Semenzato wrote:
> Is hte->hostname still meaningful?  Could it be garbage?

Done.

[Luigi Semenzato, 2010-10-20 20:21:31]

LGTM