All drivers have copy + pasted inadequate error reporting which wraps
util.c:virGetHostname. Move all error reporting to this function, and improve
what we report.

Changes from v1:
  Drop the driver wrappers around virGetHostname. This means we still need
  to keep the new conn argument to virGetHostname, but I think it's worth
  it.

* src/conf/interface_conf.c: the code was erronously returning -1
  in the two functions if <dhcp> is not provided

* src/conf/interface_conf.c: lack of one of these in the live xml output
  was causing the parse in virInterfaceDefParseBond() to fail

* src/conf/interface_conf.c: forgot to free the structure itself

This patch updates the xml parsing and formatting, and the associated
virInterfaceDef data structure to support IPv6, along the way adding
support for multiple protocols per interface, and multiple IP
addresses per protocol.
* src/conf/interface_conf.[ch]: update the structures, code for parsing
  and serialization

This patch adds the flag VIR_INTERFACE_XML_INACTIVE to
virInterfaceGetXMLDesc's flags. When it is*not* set (the default), the
live interface info will be returned in the XML (in particular, the IP
address(es) and netmask(s) will be retrieved by querying the interface
directly, rather than  reporting what's in the config file). The
backend of this is in netcf's ncf_if_xml_state() function.

* configure.in libvirt.spec.in: requires netcf >= 0.1.3
* include/libvirt/libvirt.h.in: adds flag VIR_INTERFACE_XML_INACTIVE
* src/conf/interface_conf.c src/interface/netcf_driver.c src/libvirt.c:
  update the parsing and backend routines accordingly
* tools/virsh.c: change interface edit to inactive definition and
  adds the inactive flag for interface dump

The minimal XML returned from ncf_if_xml_state() doesn't contain this
attribute (which makes no sense in the case of reporting current
status of the interface), and it was preventing it from passing
through the parse/format step.

* src/conf/interface_conf.[ch]: add a new virInterfaceStartMode value
  and modify loading/saving accordingly

The virStateInitialize() call for starting up stateful drivers
may require that the event loop is running already. This it is
neccessary to start the event loop before this call. At the
same time, network clients must not be processed until afte
virStateInitialize has completed.

The qemudListenUnix() and remoteListenTCP() methods must
therefore not register file handle watches, merely open the
network sockets & listen() on them. This means clients can
connected and are queued, pending completion of initialization

The qemudRunLoop() method is moved into a background thread
that is started early to allow access to the event loop during
driver initialization. The main process thread leader pretty
much does nothing once the daemon is running, merely waits
for the event loop thread to quit

* daemon/libvirtd.c, daemon/libvirtd.h: Move event loop into
  a background thread
* daemon/THREADING.txt: Rewrite docs to better reflect reality

suggested by danpb on irc, patch by danken fixed for proper C syntax

* python/libvirt-override.c: on event callback release the python
  interpreter lock and take it again when coming back so that the
  callback can reinvoke libvirt.

There are places where an interface will not have a mac address, and netcf
returns this as a NULL pointer rather than a pointer to an empty string.
Rather than checking for this all over the place in libvirt, just save it
in the virInterface object as an empty string.

* src/datatypes.c: allow NULL mac in virGetInterface()

introduced on commit 9231aa7d
* src/qemu/qemu_driver.c: in qemudRemoveDomainStatus fix a reference
  to an undefined variable buf and free up an allocated string

When building with --disable-nls, I got a few messages like this:

storage/storage_backend.c: In function 'virStorageBackendCreateQemuImg':
storage/storage_backend.c:571: warning: format not a string literal and no format arguments

Fix these up.
Signed-off-by: NChris Lalancette <clalance@redhat.com>

qemudShutdownVMDaemon() calls qemudRemoveDomainStatus(), which
then calls virFileDeletePID().  qemudShutdownVMDaemon() then
unnecessarily calls virFileDeletePID() again.  Remove this second
usage of it, and also slightly refactor qemudRemoveDomainStatus()
to VIR_WARN appropriate error messages.
Signed-off-by: NChris Lalancette <clalance@redhat.com>

The LXC driver was mistakenly returning -1 for lxcStartup()
in scenarios that are not an error. This caused the libvirtd
to quit for unprivileged users. This fixes the return code
of LXC driver, and also adds a "name" field to the virStateDriver
struct and logging to make it easier to find these problems
in the future

* src/driver.h: Add a 'name' field to state driver to allow
  easy identification during failures
* src/libvirt.c: Log name of failed driver for virStateInit
  failures
* src/lxc/lxc_driver.c: Don't return a failure code for
  lxcStartup() if LXC is not available on this host, simply
  disable the driver.
* src/network/bridge_driver.c, src/node_device/node_device_devkit.c,
  src/node_device/node_device_hal.c, src/opennebula/one_driver.c,
  src/qemu/qemu_driver.c, src/remote/remote_driver.c,
  src/secret/secret_driver.c, src/storage/storage_driver.c,
  src/uml/uml_driver.c, src/xen/xen_driver.c: Fill in name
  field in virStateDriver struct

If an error occurs between the allocation of an item and appending it
to the list, the item leaks. Free such orphaned items in error cases.

* src/esx/esx_vi.c: free orphaned items in error cases

The default transport for the VI API is HTTPS. If the server redirects
from HTTPS to HTTP the driver would silently follow that redirection.
The user assumes to communicate with the server over a secure transport
but isn't.

This patch disables automatical redirection following. The driver reports
an error if the server tries to redirect.

* src/esx/esx_vi.c: refactor the call to curl_easy_perform() into a
  function and do error handling there, disable automatical redirection
  following for curl
* src/esx/esx_vi.h: change the type of responseCode to int

Unified function naming scheme:
- 'lookup' functions query the ESX or vCenter for information
- 'get' functions return information from a local object

* src/esx/esx_driver.c, src/esx/esx_vi.[ch]: unify function naming

In order to register a new virtual machine the ESX driver needs to upload
a VMX file to a datastore. Try to put this file beside the main VMDK file
of the virtual machine. Change the disk selection for datastore detection
to choose the first file-based harddisk instead of just the first disk.
The first disk may be a CDROM disk and ISO images are normaly not located
in the virtual machine's directory.

* src/esx/esx_driver.c: change disk selection for datastore detection

This allows to use domain-xml-from-native with VMX files that reference
unavailable datastores.

* src/esx/esx_vmx.c: fallback to the preliminary name if the datastore
  cannot be found

Rename virDomainIsActive to virDomainObjIsActive, and
virInterfaceIsActive to virInterfaceObjIsActive and finally
virNetworkIsActive to virNetworkObjIsActive.

* src/conf/domain_conf.c, src/conf/domain_conf.h,
  src/conf/interface_conf.h, src/conf/network_conf.c,
  src/conf/network_conf.h, src/lxc/lxc_driver.c,
  src/network/bridge_driver.c, src/opennebula/one_driver.c,
  src/openvz/openvz_driver.c, src/qemu/qemu_driver.c,
  src/test/test_driver.c, src/uml/uml_driver.c: Update for
  renamed APIs.

* configure.in daemon/Makefile.am: the --with-init-script configure
  option was broken, and always defaulted based on the existence of
  /etc/redhat-release. This was a systematic typo based on
  mixed use of init-script and init-scripts.

The daemonizing code lets the parent exit almost immediately. This
means that it may think it has successfully started even when
important failures occur like not being able to acquire the PID
file. It also means network sockets are not yet open.

To address this when daemonizing the parent passes an open pipe
file descriptor to the child. The child does its basic initialization
and then writes a status code to the pipe indicating either success,
or failure. This ensures that when daemonizing, the parent does not
exit until the pidfile is acquired & basic network sockets are open.

Initialization of the libvirt drivers is still done asynchronously
since this may take a very long time.

* daemon/libvirtd.c: Force parent to stay around until basic config
  file, pidfile & network socket init is completed

* daemon/libvirtd.c: Introduce a daemonSetupSignals() method
  and put all signal handling code there
* daemon/libvirtd.h: Add sigread/sigwrite to qemud_server type

* src/openvz/openvz_driver.c: add a new openvzDomainSetMemoryInternal()
  function, used to initialize the KMEMSIZE parameter of a container.

The libvirt default error handling callback will print all errors
to stderr. The libvirtd default logging callback will do the same.
Set a no-op error handling callback in libvirtd to prevent this
duplication

* daemon/libvirtd.c: Register a no-op error handling function

virInitialize must be the first libvirt function called to ensure
threads, error handling & random number generator are all setup.

Move UNIX socket directory permissions change to place of use

* daemon/libvirtd.c: Change qemudNetworkInit() so that it doesn't try
to free its argument, leaving the caller todo cleanup as is normal
practice. Add missing policykit cleanup to qemudCleanup, and remove
server watch if set. Remove duplicated call to listen() on TCP sockets

Nearly all of the methods in src/util/util.h have error codes that
must be checked by the caller to correct detect & report failure.
Add ATTRIBUTE_RETURN_CHECK to ensure compile time validation of
this

* daemon/libvirtd.c: Add explicit check on return value of virAsprintf
* src/conf/domain_conf.c: Add missing check on virParseMacAddr return
  value status & report error
* src/network/bridge_driver.c: Add missing OOM check on virAsprintf
  and report error
* src/qemu/qemu_conf.c: Add missing check on virParseMacAddr return
  value status & report error
* src/security/security_selinux.c: Remove call to virRandomInitialize
  that's done in libvirt.c already
* src/storage/storage_backend_logical.c: Add check & log on virRun
  return status
* src/util/util.c: Add missing checks on virAsprintf/Run status
* src/util/util.h: Annotate all methods with ATTRIBUTE_RETURN_CHECK
  if they return an error status code
* src/vbox/vbox_tmpl.c: Add missing check on virParseMacAddr
* src/xen/xm_internal.c: Add missing checks on virAsprintf
* tests/qemuargv2xmltest.c: Remove bogus call to virRandomInitialize()

The virDomainObjPtr object stores state about a running domain.
This object is shared across all drivers so it is not appropriate
to include driver specific state here. This patch adds the ability
to request a blob of private data per domain object instance. The
driver must provide a allocator & deallocator for this purpose

THis patch abuses the virCapabilitiesPtr structure for storing the
allocator/deallocator callbacks, since it is already being abused
for other internal things relating to parsing. This should be moved
out into a separate object at some point.

* src/conf/capabilities.h: Add privateDataAllocFunc and
  privateDataFreeFunc fields
* src/conf/domain_conf.c: Invoke the driver allocators / deallocators
  when creating/freeing virDomainObjPtr instances.
* src/conf/domain_conf.h: Pass virCapsPtr into virDomainAssignDef
  to allow access to the driver specific allocator function
* src/lxc/lxc_driver.c, src/opennebula/one_driver.c,
  src/openvz/openvz_driver.c, src/qemu/qemu_driver.c,
  src/test/test_driver.c, src/uml/uml_driver.c: Update for
  change in virDomainAssignDef contract

__in6_u.__u6_addr16 is the private name for this struct member,
s6_addr16 is the public one
* src/util/network.c: dont use the private field, but the public one.

* src/util/network.[ch] Add functions for address->text and get/set
  port number
* src/libvirt_private.syms: add new entry points

* src/util/network.c: fix a few typo

* src/libvirt_private.syms: keep all symbols from network.h private

* src/util/network.h src/util/network.c: utilities to parse network
  addresses, check netmask and compute ranges

* configure.in: the PKG_CHECK_MODULES() for it would not work on
  Ubuntu and SuSE, so add a direct check fallback

John Levon raised the issue that remoteIOEventLoop() poll call was
reissued after EINTR was caught making it uninterruptible.
* src/remote/remote_driver.c: catch EAGAIN instead as suggested by
  Richard Jones

The current virDomainObjListPtr object stores domain objects in
an array. This means that to find a particular objects requires
O(n) time, and more critically acquiring O(n) mutex locks.

The new impl replaces the array with a virHashTable, keyed off
UUID. Finding a object based on UUID is now O(1) time, and only
requires a single mutex lock. Finding by name/id is unchanged
in complexity.

In changing this, all code which iterates over the array had
to be updated to use a hash table iterator function callback.
Several of the functions which were identically duplicating
across all drivers were pulled into domain_conf.c

* src/conf/domain_conf.h, src/conf/domain_conf.c: Change
  virDomainObjListPtr to use virHashTable. Add a initializer
  method virDomainObjListInit, and rename virDomainObjListFree
  to virDomainObjListDeinit, since its not actually freeing
  the container, only its contents. Also add some convenient
  methods virDomainObjListGetInactiveNames,
  virDomainObjListGetActiveIDs and virDomainObjListNumOfDomains
  which can be used to implement the correspondingly named
  public API entry points in drivers
* src/libvirt_private.syms: Export new methods from domain_conf.h
* src/lxc/lxc_driver.c, src/opennebula/one_driver.c,
  src/openvz/openvz_conf.c, src/openvz/openvz_driver.c,
  src/qemu/qemu_driver.c, src/test/test_driver.c,
  src/uml/uml_driver.c, src/vbox/vbox_tmpl.c: Update all code
  to deal with hash tables instead of arrays for domains

There's a long known issue where if you install libvirt in a guest,
the default virtual network will conflict with the default virtual
network in the host.

That's one of the reasons we have the libvirt-client RPM - it allows
you to install the client library without having the host side
installed.

Rich Jones points out that if you install libvirt-devel in a guest,
then you get libvirtd installed and the network conflict:

  https://bugzilla.redhat.com/531200

libvirt-devel should only require libvirt-client - e.g. nothing in
the devel package pertains to anything in the libvirt RPM. The Fedora
packaging guidelines say:

  https://fedoraproject.org/wiki/Packaging/Guidelines#Devel_Packages

  Devel packages must require the base package using a fully versioned
  dependency ...

But for all intents and purposes, libvirt-client is our base RPM.

* libvirt.spec.in: make libvirt-devel require libvirt-client

The xen driver will generate a migration port if only a hostname is passed
in the optional migrate URI, so let's do the same in qemu.