The current fork+close fds sequence has deficiencies which are hard to
work around:
  - logging can start logging before we fork (e.g. if we need to emit
    messages related to master checking), and thus use FDs which we
    can't track nicely
  - the queue locks the queue file, and again this fd needs to be kept
    open which is hard from the main loop (and this error is currently
    hidden by the fact that we don't log it)

Given the above, it's much simpler, in case we will fork later, to close
file descriptors right at the beginning of the program, and in Daemonize
only close/reopen the stdin/out/err fds.

In addition, we also close() the handlers we remove in SetupLogging so
that the cleanup is more thorough.

Reviewed-by: imsnah

In 1.2, due to the md+drbd7 legacy, we had a complex choice of replace
modes, and the new drbd8 modes where forced into this syntax, with some
complicated rules of transition from one mode to another (if REPLACE_ALL
but not new node passed, switch to REPLACE_SEC, etc.).

This patch cleans this situation by making a clear separation between
the two main modes:
  - replace on current nodes (with the two sub-cases on primary and on
    secondary)
  - change to a new node (either via manually specified node or via
    iallocator)

Reviewed-by: imsnah

For offline nodes, we shouldn't add them to the NV_NODELIST and
NV_NODENETTEST tests since they most likely won't succeed.

The patch makes gnt-cluster verify happy again in such cases.

Reviewed-by: imsnah

The patch adds a new method to the rpc.RpcResult class called
"RemoteFailMsg" which is useful for the RPC calls which return a
(status, payload) style result.

Reviewed-by: imsnah

This is a forward-port of commit 1194 on the 1.2 branch:

  This call will check whether an instance is up on its primary, and that
  it has been started with symlinks. We currently have no on-secondary
  checks, nor any hypervisor specific call.

  Reviewed-by: iustinp

The difference from the original patch is that we don't include the
cmdlib changes, since those will come as a copy from the 1.2 cmdlib.py,
and not as individual patches.

Original-Author: ultrotter

This is plain copy of the 1.2 ReAttachNet and DisconnectNet methods on
the DRBD8 device, with the logger to logging module changes and the
ReAttachNet method renamed to AttachNet.

These methods are not used anywhere right now, but will be used for
migration and a simpler disk-replace.

The code was originally committed on the 1.2 branch as revision numbers
1165 and 1204.

Originally-Reviewed-by: imsnah, ultrotter

This is a modified forward-port of commit 1184 on the 1.2 branch:

  backend: Remove symlinks by disk name, not using a wildcard

  Reviewed-by: ultrotter

The changes to the original patch are related to the docstring style and
iv_name to index switch.

Original-Author: imsnah

This is an extract of commit 1166 on the 1.2 branch (Add a rpc call for
drbd network reconfiguration), but only the blockdev_close part.

The patch changes the blockdev_close call to take the instance so that
it can remove the symlinks of the instance.

Originally-Reviewed-by: imsnah

This is a forward-port of commit 1163 on the 1.2 branch:
  This fixes the removal of the instance symlinks (probably breakage from
  the glob changes).

  Reviewed-by: imsnah

This is a forward-port of commits 1150 and 1151 on the 1.2 branch:
  Add _RemoveBlockDevLinks auxiliary function, called when an instance
  fails to start and when it is shut down.

  Reviewed-by: iustinp

and:
  Fix cut&paste error when removing symlinks

  It's just whitespace... isn't it? uhm... :) Anyway, fixing an error made
  when reformatting the code for the new "safer" behaviour.

  Reviewed-by: iustinp

Original-Author: ultrotter

This is a forward-port of commit 1149 on the 1.2 branch:
  _GatherAndLinkBlockDevs used to raise the errors.BlockDeviceError
  exception when it failed to create a block device, and with this patch
  set it does so also when it fails to create a symlink to it.

  With this patch we move the call to this function into a pre-existing
  try-except block in the code, and catch the BlockDeviceError exception,
  logging a message and returning a failure state if it happens.

  Reviewed-by: iustinp

The changes are related to the new hypervisor and logging syntax.

Original-Author: ultrotter

This is a forward-port of commit 1148 on the 1.2 branch:
  Change the _GatherBlockDevs private function, called only one time by
  StartInstance, to _GatherAndLinkBlockDevs, and make it transform the
  device returned even more by calling the new _SimlinkBlockDev auxiliary
  function.

  This makes sure that every time an instance is started symlinks to its
  block devices are created, and the instance is started off them, rather
  than the underlying block devices.

  Reviewed-by: iustinp

The changes we make to the patch is related to newer function signatures
in 2.0, and to the fact that iv_name is deprecated and we use instead
disk%d based on the disk index.

Original-Author: ultrotter

This is a partial forward-port of commit 1136 on the 1.2 branch:

  The hypervisor doesn't need to be passed the whole block device
  structure, so we'll just give it the block device name on the local
  node, and the name as seen by the instance. This will make it easier to
  manipulate it later without messing with the block devices (eg. by
  changing the system name to a symlink to the name itself).

  Since the HVM hypervisor changes the "virtual" name a note is added
  calling for a redesign that doesn't need this change, as different
  hypervisors and emulation types will anyway have different names for
  exported devices.

  Reviewed-by: iustinp

The changes in this patch compared to the original are:
  - we keep passing the original disk object, not for its iv_name, but
    for it's physical_id which is needed by the file driver (this could
    be fixed maybe)
  - we don't use the iv_name anymore, since in 2.0 we already use the
    index of the device

Original-Author: ultrotter

Commit 2117 changed _AssembleInstanceDisks to correctly parse the
failure status of the new RpcResult structure, but it didn't fix the
storing of only the result payload. Since RpcResult is not JSON
serializable, LUActivateInstanceDisks is failing.

Reviewed-by: ultrotter

Two bad indentation cases and a missing variable.

Reviewed-by: imsnah

Reviewed-by: amishchenko

Reviewed-by: amishchenko

Reviewed-by: amishchenko

Reviewed-by: amishchenko

Previously, we set the permissions only after writing the key. This
gave other users on the system a small window during which they could
read the key.

Reviewed-by: amishchenko

Reviewed-by: amishchenko

Reviewed-by: amishchenko

Passwords are stored in "$localstatedir/lib/ganeti/rapi_users". User
options specify the access permissions of a user (see docstring for
ganeti.http.ReadPasswordFile), for which only "write" is supported
to grant write access. Every other user has read-only access.

Reviewed-by: amishchenko

This will be used to evaluate access permissions to resources.

Reviewed-by: amishchenko

Lines in the password file are of the following format:

  <username> <password> [options]

Fields are separated by whitespace. Username and password are
mandatory, options are optional and separated by comma (",").
Empty lines and comments ("#") are ignored.

Reviewed-by: amishchenko

Reviewed-by: amishchenko

As per RFC2617.

Reviewed-by: amishchenko

The authentication class will override PreHandleRequest.

Reviewed-by: amishchenko

Reviewed-by: ultrotter

Reducing the actual number of RPC calls will come in another patch.

Reviewed-by: ultrotter

With a large job queue, auto-archiving jobs can take a very long time,
causing timeouts on the luxi RPC layer. With this change, auto-
archive returns after half of the RPC timeout has passed. The user
will see how many jobs are left unchecked.

Reviewed-by: ultrotter

This is done by passing the job object to _ArchiveJobUnlocked instead
of only the job ID. Also return whether job was actually archived.

Reviewed-by: ultrotter

As it turned out, having many files in a single directory can be
very painful. With this patch, only 10'000 files are stored in a
directory for the job queue archive. With 10'000 directries, this
allows for up to 100 million jobs be archived without having large
numbers of files in a single directories. Not that it is realistic,
anyway.

Reviewed-by: ultrotter

Unfortunately, os.makedirs in Python 2.4 is not safe against multiple
processes creating the same directory tree at the same time. This is
only fixed in Python 2.5 and up. Adding more checks in our code doesn't
make it any better.

Reviewed-by: iustinp

They're cheap to instantiate and doing this changes makes the code
a bit simpler.

Reviewed-by: ultrotter

All other exceptions are named after the error name in RFC2616 (HTTP/1.1).

Reviewed-by: amishchenko

Reviewed-by: amishchenko

Reviewed-by: amishchenko

Reviewed-by: amishchenko

More complex unittests will need some refactoring in the HTTP code.

Reviewed-by: amishchenko