1
0
mirror of https://github.com/samba-team/samba.git synced 2024-12-23 17:34:34 +03:00
Commit Graph

2112 Commits

Author SHA1 Message Date
Martin Schwenke
2650f37018 ctdb-logging: Drop enum debug_level
We are switching to Samba-style integer debug levels.

Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>
2016-12-05 08:09:21 +01:00
Amitay Isaacs
9b7308b202 ctdb-daemon: Remove tevent debug logging
Signed-off-by: Amitay Isaacs <amitay@gmail.com>
Reviewed-by: Martin Schwenke <martin@meltin.net>
2016-12-05 08:09:21 +01:00
Martin Schwenke
429377a242 ctdb-ipalloc: Optimise check to see if IP is available on a node
Use a "bitmap" of available IPs for each IP address instead of walking
the list of available IP addresses.

For ctdb/tests/takeover/lcp2.030.sh, this improves the time taken on
my laptop from:

  real	0m11.997s
  user	0m11.960s
  sys	0m0.000s

to

  real	0m8.571s
  user	0m8.544s
  sys	0m0.000s

So, when assigning all 900 IP addresses the improvement is about 25%.

For the no-op case (where all IPs are already assigned to nodes), the
extra setup adds a small fraction of a second for 900 IPs.
Intermediate cases result in intermediate improvements.

Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>
2016-12-02 00:24:28 +01:00
Martin Schwenke
24db43839f ctdb-daemon: Exit early if there are trailing command-line arguments
Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>
2016-12-02 00:24:28 +01:00
Martin Schwenke
df2d6518e7 ctdb-daemon: Don't call ctdb_local_node_got_banned() on flag changes
This function is currently called twice each time a node is banned.

ctdb_local_node_got_banned() is already called from the banning code,
either due to a received banning control or a node banning itself.
Given that other nodes can't set a node's BANNED flag, a node can only
be banned via the above mechanisms, so drop the redundant call.

Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>
2016-12-02 00:24:28 +01:00
Martin Schwenke
bdc049dfce ctdb-common: Drop CTDB's copy of sys_read() and sys_write()
Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>

Autobuild-User(master): Amitay Isaacs <amitay@samba.org>
Autobuild-Date(master): Tue Nov 29 11:22:40 CET 2016 on sn-devel-144
2016-11-29 11:22:40 +01:00
Martin Schwenke
dcde6f1619 ctdb-lock-helper: Drop include of ctdb_private.h
Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>
2016-11-29 07:25:19 +01:00
Amitay Isaacs
f2414841f2 ctdb-daemon: Mark RecoverPDBBySeqNum tunable deprecated
Persistent databases are now always recovered by sequence number, so
there is no need for this tunable.

Signed-off-by: Amitay Isaacs <amitay@gmail.com>
Reviewed-by: Martin Schwenke <martin@meltin.net>

Autobuild-User(master): Martin Schwenke <martins@samba.org>
Autobuild-Date(master): Fri Nov 25 08:13:59 CET 2016 on sn-devel-144
2016-11-25 08:13:59 +01:00
Amitay Isaacs
2a9584dc0a ctdb-daemon: Remove unused code cmdline.[ch]
Signed-off-by: Amitay Isaacs <amitay@gmail.com>
Reviewed-by: Martin Schwenke <martin@meltin.net>
2016-11-25 04:19:23 +01:00
Amitay Isaacs
569d787a1d ctdb-daemon: Consolidate command line options to ctdbd
This inserts the code from ctdb_cmdline_init() function directly in
main(), so common/cmdline.[ch] can be removed.

Signed-off-by: Amitay Isaacs <amitay@gmail.com>
Reviewed-by: Martin Schwenke <martin@meltin.net>
2016-11-25 04:19:23 +01:00
Amitay Isaacs
54e392b385 ctdb-recovery: Avoid NULL dereference in failure case
BUG: https://bugzilla.samba.org/show_bug.cgi?id=12434

Signed-off-by: Amitay Isaacs <amitay@gmail.com>
Reviewed-by: Volker Lendecke <vl@samba.org>

Autobuild-User(master): Volker Lendecke <vl@samba.org>
Autobuild-Date(master): Mon Nov 21 12:26:04 CET 2016 on sn-devel-144
2016-11-21 12:26:04 +01:00
Amitay Isaacs
6c6d63c044 ctdb-locking: Reset real-time priority in lock helper
BUG: https://bugzilla.samba.org/show_bug.cgi?id=12392

Earlier we were relying on SCHED_RESET_ON_FORK to reset the priority of lock
helper processes.  Since SCHED_RESET_ON_FORK support has been removed, the
scheduling priority of child processes created using vfork() need to be reset
explicitly in the helper processes.

Signed-off-by: Amitay Isaacs <amitay@gmail.com>
Reviewed-by: Martin Schwenke <martin@meltin.net>

Autobuild-User(master): Martin Schwenke <martins@samba.org>
Autobuild-Date(master): Fri Nov 18 10:18:27 CET 2016 on sn-devel-144
2016-11-18 10:18:27 +01:00
Amitay Isaacs
3c03754921 ctdb-daemon: Simplify code using tdb_storev
Signed-off-by: Amitay Isaacs <amitay@gmail.com>
2016-10-27 23:53:12 +02:00
Amitay Isaacs
12fd2ddc01 ctdb-daemon: Simplify code using local variable
Signed-off-by: Amitay Isaacs <amitay@gmail.com>
2016-10-27 23:53:12 +02:00
Volker Lendecke
6c95148f85 ctdb: Fix format errors for time_t!=long
Signed-off-by: Volker Lendecke <vl@samba.org>
Reviewed-by: Ralph Boehme <slow@samba.org>
2016-10-17 22:34:20 +02:00
Martin Schwenke
0ec01826d3 ctdb-daemon: Log when removing stale Unix domain socket
BUG: https://bugzilla.samba.org/show_bug.cgi?id=12287

Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>

Autobuild-User(master): Amitay Isaacs <amitay@samba.org>
Autobuild-Date(master): Thu Sep 22 12:28:12 CEST 2016 on sn-devel-144
2016-09-22 12:28:12 +02:00
Martin Schwenke
8eff9e9603 ctdb-daemon: Drop attempt to connect to Unix domain socket
This was a weak attempt at exclusivity.  PID file creation now does
that properly.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=12287

Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>
2016-09-22 08:34:20 +02:00
Martin Schwenke
d719a87fe0 ctdb-daemon: Don't try to reopen TDB files
There aren't any open at this stage.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=12287

Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>
2016-09-22 08:34:20 +02:00
Martin Schwenke
1e501c7749 ctdb-daemon: Bind to Unix domain socket after PID file creation
No use touching the socket if PID file creation fails.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=12287

Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>
2016-09-22 08:34:20 +02:00
Martin Schwenke
5148e02adb ctdb-daemon: Use PID file abstraction
BUG: https://bugzilla.samba.org/show_bug.cgi?id=12287

Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>
2016-09-22 08:34:20 +02:00
Amitay Isaacs
8b979c729b ctdb-locking: Restrict lock debugging to once per second
Signed-off-by: Amitay Isaacs <amitay@gmail.com>
Reviewed-by: Martin Schwenke <martin@meltin.net>
2016-09-22 08:34:20 +02:00
Amitay Isaacs
cdc46ef9d7 ctdb-locking: Log if ctdb is unable to take db locks in INACTIVE state
This is useful information if ctdb is unable to freeze any of the
databases on banning or stopping.

Signed-off-by: Amitay Isaacs <amitay@gmail.com>
Reviewed-by: Martin Schwenke <martin@meltin.net>
2016-09-22 08:34:20 +02:00
Amitay Isaacs
28fad1c59d ctdb-daemon: Log a message when vfork() takes long time
Signed-off-by: Amitay Isaacs <amitay@gmail.com>
Reviewed-by: Martin Schwenke <martin@meltin.net>

Autobuild-User(master): Martin Schwenke <martins@samba.org>
Autobuild-Date(master): Mon Sep 19 12:18:15 CEST 2016 on sn-devel-144
2016-09-19 12:18:15 +02:00
Amitay Isaacs
06171961d6 ctdb-daemon: Log a message when fork() takes long time
Signed-off-by: Amitay Isaacs <amitay@gmail.com>
Reviewed-by: Martin Schwenke <martin@meltin.net>
2016-09-19 08:23:22 +02:00
Amitay Isaacs
a69d1e202a ctdb-daemon: Avoid extra condition in tevent trace callback
Signed-off-by: Amitay Isaacs <amitay@gmail.com>
Reviewed-by: Martin Schwenke <martin@meltin.net>
2016-09-19 08:23:22 +02:00
Amitay Isaacs
6b93b57921 ctdb-recovery-helper: Add missing initialisation of ban_credits
BUG: https://bugzilla.samba.org/show_bug.cgi?id=12275

Signed-off-by: Amitay Isaacs <amitay@gmail.com>
Reviewed-by: Martin Schwenke <martin@meltin.net>
2016-09-19 08:23:22 +02:00
Martin Schwenke
7ec7d4f3c0 ctdb-ipalloc: ipalloc_set_public_ips() can't fail
So make it a void function.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=12254

Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>
2016-09-14 08:39:29 +02:00
Martin Schwenke
7522a7aee8 ctdb-ipalloc: Move merged IP list creation to ipalloc()
BUG: https://bugzilla.samba.org/show_bug.cgi?id=12254

Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>
2016-09-14 08:39:29 +02:00
Martin Schwenke
c1efb801a4 ctdb-ipalloc: Drop known_ips argument from merged IP list creation
This is available in the IP allocation state.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=12254

Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>
2016-09-14 08:39:29 +02:00
Martin Schwenke
fed251726f ctdb-ipalloc: Optimise check to see if IPs can be hosted
Add an early return if there are no known IP addresses.

Also add an extra comment for clarification.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=12254

Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>
2016-09-14 08:39:29 +02:00
Martin Schwenke
dfc3b8855d ctdb-ipalloc: Whether IPs can be hosted need not depend on merged IP list
Merged IP list won't be available here...

BUG: https://bugzilla.samba.org/show_bug.cgi?id=12254

Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>
2016-09-14 08:39:29 +02:00
Martin Schwenke
0e5c62d8fc ctdb-ipalloc: Store known public IPs in IP allocation state
This was dropped because it wasn't used, but it will be needed again.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=12254

Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>
2016-09-14 08:39:28 +02:00
Amitay Isaacs
67351e61ee ctdb-recoverd: Drop code to freeze databases from set_recovery_mode()
This function is called only once from force_election() and does not
require freezing of databases.

Signed-off-by: Amitay Isaacs <amitay@gmail.com>
Reviewed-by: Martin Schwenke <martin@meltin.net>
2016-09-14 08:39:28 +02:00
Martin Schwenke
a2abc78c87 ctdb-daemon: Compare interface pointers instead of using strcmp(3)
If the interfaces have different names then they are different
interfaces.

Also, move assignment of new_name just above where is is first used.

Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>
2016-09-08 06:57:21 +02:00
Martin Schwenke
a66072efdf ctdb-daemon: Use ctdb_find_iface() instead of duplicating logic
Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>
2016-09-08 06:57:21 +02:00
Martin Schwenke
2f46056228 ctdb-deamon: Rename vnn_has_interface_with_name() to vnn_has_interface()
Now takes a pointer to an interface structure and does direct pointer
comparisons.

Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>
2016-09-08 06:57:21 +02:00
Martin Schwenke
b129c288f5 ctdb-daemon: Drop redundant uses of ctdb_find_iface()
Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>
2016-09-08 06:57:20 +02:00
Martin Schwenke
29787cb3c8 ctdb-daemon: Make vnn->iface a list of new struct vnn_interface
To keep this change small, this leaves behind some redundant calls to
ctdb_find_iface() and similar.  They will be cleaned up later.

Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>
2016-09-08 06:57:20 +02:00
Martin Schwenke
e6258acf1b ctdb-daemon: Change ctdb_add_local_iface() to return struct ctdb_interface
This will allow a change to the way interfaces are handled in a VNN.

Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>
2016-09-08 06:57:20 +02:00
Martin Schwenke
ff4b452c6a ctdb-daemon: Move interface addition into interface parsing
Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>
2016-09-08 06:57:20 +02:00
Martin Schwenke
1790f9f754 ctdb-daemon: Drop some uses of CTDB_NO_MEMORY{,_FATAL}()
Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>
2016-09-08 06:57:20 +02:00
Martin Schwenke
c3502cf9f9 ctdb-daemon: Consolidate interface checking with interface parsing
There's no point parsing the interfaces twice, especially since it
doesn't improve error handling.

This also removes a use of strdup(3)/free(3), which is not generally
used in our code.

Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>
2016-09-08 06:57:20 +02:00
Martin Schwenke
05665d9ede ctdb-daemon: Replace some uses of CTDB_NO_MEMORY_FATAL()
Also add a missing out-of-memory check for vnn->ifaces.

Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>
2016-09-08 06:57:20 +02:00
Martin Schwenke
d658d4e701 ctdb-daemon: Move and improve public IP duplicate checking
This also moves the interface validation down, making more obvious
that it can be consolidated.

Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>
2016-09-08 06:57:20 +02:00
Martin Schwenke
f63fe7c3b9 ctdb-daemon: Drop use of strdup(3) and free(3) when releasing IP
If anything should be used here it should be talloc functions.
However, this is a remnant from when ctdb_sys_find_ifname() was used
here and, for some reason, it used strdup(3).

In this case the interface string doesn't actually need to be copied.
The only use of it is when ctdb_event_script_callback_v() uses it with
the format string in a call to talloc_vasprintf().  In the same
context the IP address isn't copied.

Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>
2016-09-08 06:57:20 +02:00
Martin Schwenke
9d975b860d ctdb-daemon: Don't steal control structure before synchronous reply
If *async_reply isn't set then the calling code will reply to the
control and free the control structure.  In some places the control
structure pointer is stolen onto state before a synchronous exit due
to an error condition.  The error handling then frees state and
returns an error.  The calling code will access-after-free when trying
to reply to the control.

To make this easier to understand, the convention is that any
(immediate) error results in a synchronous reply to the control via an
error return code AND *async_reply not being set.  In this case the
control structure pointer should never be stolen onto state.  State is
never used for a synchronous reply, it is only ever used by a
callback.

Also initialise state->c to NULL so that any premature call to a
callback (e.g. in an immediate error path) is more obvious.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=12180

Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>
2016-09-01 13:30:10 +02:00
Martin Schwenke
582518c7e8 ctdb-daemon: Handle failure immediately, do housekeeping later
The callback should never be called before an immediate return.  The
callback might reply to a control and the caller of
ctdb_event_script_callback_v() may not have assigned/stolen the
pointer to control structure into the private data.  Therefore,
calling the callback can dereference an uninitialised pointer to the
control structure when attempting to reply.

An event script isn't being run until the child has been forked.  So
update relevant state and set the destructor after this.

If the child can't be forked then free the state and return with an
error.  The callback will not be called and the caller will process
the error correctly.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=12180

Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>
2016-09-01 13:30:10 +02:00
Martin Schwenke
9076c44f35 ctdb-daemon: Schedule running of callback if there are no event scripts
The callback should never be called before an immediate return.  The
callback might reply to a control and the caller of
ctdb_event_script_callback_v() may not have assigned/stolen the
pointer to control structure into the private data.  Therefore,
calling the callback can dereference an uninitialised pointer to the
control structure when attempting to reply.

ctdb_event_script_callback_v() must succeed when there are no event
scripts.  On success the caller will mark the call as asynchronous and
expect the callback to be called.  Given that it can't be called
before return then it needs to be scheduled.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=12180

Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>
2016-09-01 13:30:10 +02:00
Martin Schwenke
6dc75c7d24 ctdb-daemon: When releasing an IP, update PNN in callback
When an error occurs so an IP address is not released then the PNN in
the VNN is currently incorrectly updated.

Instead, update the PNN in the callback when the release is
successful.  Also, explicitly update the PNN on redundant releases.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=12158

Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>

Autobuild-User(master): Amitay Isaacs <amitay@samba.org>
Autobuild-Date(master): Sun Aug 21 22:45:33 CEST 2016 on sn-devel-144
2016-08-21 22:45:32 +02:00
Martin Schwenke
976a50af6f ctdb-daemon: Rename takeover_callback_state -> release_ip_callback_state
Many years ago takeover_callback_state was used for both IP takeover
and release.  Now it is only used when releasing an IP so rename it to
improve clarity.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=12158

Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>
2016-08-21 18:52:05 +02:00