A bundle is a list of all persistent connections to the same host.
The connection cache consists of a hash of bundles, with the
hostname as the key.
The benefits may not be obvious, but they are two:

1) Faster search for connections to reuse, since the hash
   lookup only finds connections to the host in question.
2) It lays out the groundworks for an upcoming patch,
   which will introduce multiple HTTP pipelines.

This patch also removes the awkward list of "closure handles",
which were needed to send QUIT commands to the FTP server
when closing a connection.
Now we allocate a separate closure handle and use that
one to close all connections.

This has been tested in a live system for a few weeks, and of
course passes the test suite.

DragonFly uses milliseconds, while our API and Linux use full seconds.

Reported by: John Marino
Bug: http://curl.haxx.se/bug/view.cgi?id=3546257

This reverts commit 9c94236e.

It didn't server its purpose, so lets go back to long-time working code.

Fixed warning: dereferencing pointer does break strict-aliasing rules
by using a union instead of separate pointer variables.
Internal union sockaddr_u could probably be moved to generic header.
Thanks to Paul Howarth for the hint about using unions for this.

Important for winbuild: Separate declaration of sockaddr_u pointer.
The pointer variable *sock cannot be declared and initialized right
after the union declaration. Therefore it has to be a separate statement.

Commit 9109cdec brought this regression (shipped since 7.24.0).

The singleipconnect() function must not return an error if Curl_socket()
returns an error. It should then simply return OK and pass a SOCKET_BAD
back simply because that is how the user of this function expects it to
work and something else is not fine.

Reported by: Blaise Potard
Bug: http://curl.haxx.se/bug/view.cgi?id=3516508

Curl_socket returns CURLE_COULDNT_CONNECT when the opensocket callback
returns CURL_SOCKET_BAD. Previous return value CURLE_FAILED_INIT
conveys incorrect information to the user.

This adds three new options to control the behavior of TCP keepalives:

- CURLOPT_TCP_KEEPALIVE: enable/disable probes
- CURLOPT_TCP_KEEPIDLE: idle time before sending first probe
- CURLOPT_TCP_KEEPINTVL: delay between successive probes

While not all operating systems support the TCP_KEEPIDLE and
TCP_KEEPINTVL knobs, the library will still allow these options to be
set by clients, silently ignoring the values.

When connecting to a domain with multiple IP addresses, allow different,
decreasing connection timeout values. This should guarantee some
connections attempts with sufficiently long timeouts, while still
providing fallback.

First off the timeout for accepting a server connect back must of course
respect a global timeout. Then the timeleft function is only used by ftp
code so it was moved to ftp.c and made static.

1- Two new error codes are introduced.

CURLE_FTP_ACCEPT_FAILED to be set whenever ACCEPTing fails because of
FTP server connected.

CURLE_FTP_ACCEPT_TIMEOUT to be set whenever ACCEPTing timeouts.

Neither of these errors are considered fatal and control connection
remains OK because it could just be a firewall blocking server to
connect to the client.

2- One new setopt option was introduced.

CURLOPT_ACCEPTTIMEOUT_MS

It sets the maximum amount of time FTP client is going to wait for a
server to connect. Internal default accept timeout is 60 seconds.

Do not try to resolve interfaces names via DNS by recognizing interface
names in a few ways.  If the interface option argument has a prefix of
"if!" then treat the argument as only an interface.  Similarly, if the
interface argument is the name of an interface (even if it does not have
an IP address assigned), treat it as an interface name.  Finally, if the
interface argument is prefixed by "host!" treat it as a hostname that
must be resolved by /etc/hosts or DNS.

These changes allow a client using the multi interfaces to avoid
blocking on name resolution if the interface loses its IP address or
disappears.

As a follow-up from commit d5b5f64b, clear the sock_accepted status
when such a socket is closed to avoid a re-used connection to retain the
state wrongly.

Bug: http://curl.haxx.se/mail/lib-2011-12/0079.html

Keep track of which sockets that are the result of accept() calls and
refuse to call the closesocket callback for those sockets. Test case 596
now verifies that the open socket callback is called the same number of
times as the closed socket callback for active FTP connections.

Bug: http://curl.haxx.se/mail/lib-2011-12/0018.html
Reported by: Gokhan Sengun

When the new socket is created for an active connection, it is now done
using the open socket callback.

Test case 596 was modified to run fine, although it hides the fact that
the close callback is still called too many times, as it also gets
called for closing sockets that were created with accept().

Moved out into a separate function to work as a "generic" socket()
replacement.

Previously the bit was set before the connection was found working so if
it would first fail to an ipv6 address and then connect fine to a IPv4
address the variable would still be TRUE.

Reported by: Thomas L. Shinnick
Bug: http://curl.haxx.se/bug/view.cgi?id=3421912

Save the errno value immediately after a connect() failure so that it
won't get reset to something else before we read it.

Bug: http://curl.haxx.se/mail/lib-2011-10/0066.html
Reported by: Frank Van Uffelen and Fabian Hiernaux

Renamed the variable from 'proto' to 'level' simply because it is not
protocol you set but level and that is the name of the argument used in
man pages and the POSIX documentation of the setsockopt function.

When connecting, make sure TIMER_CONNECT is set.

When using the multi interface, a SOCKS proxy, and a connection that
wouldn't immediately consider itself connected (which my Linux tests do
by default), libcurl would be tricked into doing _two_ connects to the
SOCKS proxy when it setup the data connection and then of course the
second attempt would fail miserably and cause error.

This problem is a regression that was introduced by commit
4a42e5cd that was introduced in the 7.21.7 release.

Bug: http://curl.haxx.se/mail/lib-2011-08/0199.html
Reported by: Fabian Keil

With "@unittest: [num]" in the header comment for each tested function.
Shows we have a log way to go still...

It was mostly typecasted to int all over the code so switching to long
instead all over should be a net gain.

Fix compiler warning: conversion may lose significant bits

Fix the return type of the callback to match close() and make use of it.

Introduced the initial setup to allow closesocket callbacks by making
sure sclose() is only ever called from one place in the libcurl source
and still run all test cases fine.

When connecting to a socks or similar proxy we do the proxy handshake at
once when we know the TCP connect is completed and we only consider the
"connection" complete after the proxy handshake. This fixes test 564
which is now no longer considered disabled.

Reported by: Dmitri Shubin
Bug: http://curl.haxx.se/mail/lib-2011-04/0127.html

"if(a)" is our style, not "if( a )"

By the use of a the new lib/checksrc.pl script that checks that our
basic source style rules are followed.

asyn-ares.c and asyn-thread.c are two separate backends that implement
the same (internal) async resolver API for libcurl to use. Backend is
specified at build time.

The internal resolver API is defined in asyn.h for asynch resolvers.