- 08 Nov, 2012 40 commits
-
-
Philipp Reisner authored
Signed-off-by:
Philipp Reisner <philipp.reisner@linbit.com> Signed-off-by:
Lars Ellenberg <lars.ellenberg@linbit.com>
-
Lars Ellenberg authored
...and drop explicit typecasts (int)meta_dev_idx < 0. Signed-off-by:
-
Philipp Reisner authored
Preparing RCU for disk_conf Signed-off-by:
-
Philipp Reisner authored
Preparing RCU for disk_conf Signed-off-by:
-
Philipp Reisner authored
Preparing RCU for disk_conf Signed-off-by:
-
Philipp Reisner authored
Preparing to use the same mutex for disk_conf updates Signed-off-by:
-
Philipp Reisner authored
Signed-off-by:
-
Andreas Gruenbacher authored
Signed-off-by:
-
Andreas Gruenbacher authored
Signed-off-by:
-
Lars Ellenberg authored
An administrative detach used to request a state change directly to D_DISKLESS, first suspending IO to avoid the last put_ldev() occuring from an endio handler, potentially in irq context. This is not enough on the receiving side (typically secondary), we may miss some peer_req on the way to local disk, which then may do the last put_ldev() from their drbd_peer_request_endio(). This patch makes the detach always go through the intermediate D_FAILED state. We may consider to rename it D_DETACHING. Alternative approach would be to create yet an other work item to be scheduled on the worker, do the destructor work from there, and get the timing right. Signed-off-by:
-
Lars Ellenberg authored
There are races where the receiver may be exiting, but still need the worker to process some stuff. Do not wait for the receiver to die from an exiting worker. The receiver must already be dead in case the worker decides to exit. If the receiver was still alive, it may still want to queue work, and do drbd_flush_workqueue() from it's disconnect cleanup code, which would no longer be processed by an exiting worker. This also would deadlock, if the worker was to synchornously wait for the receiver to die. Do not implicitly stop the worker. The worker will only be stopped from configuration context, from conn_reconfig_done(), drbd_adm_down() or drbd_adm_delete_connection(), after making sure the receiver is already stopped. Signed-off-by:
-
Lars Ellenberg authored
If a forced disconnect hits a restarting receiver right after it passed its final "if (C_DISCONNECTING)" test in drbdd_init(), but before it was actually restarted by drbd_thread_setup, we could be left with a connection stuck in C_DISCONNECTING, never reaching C_STANDALONE, which would be necessary to take it down or reconfigure it. Move the last cleanup into w_after_conn_state_ch(), and do an additional state change request in conn_try_disconnect(), just in case. Signed-off-by:
-
Andreas Gruenbacher authored
The main purpose of this is to allow to turn data integrity checking on and off on demand without causing interruptions. Implemented by allocating tconn->peer_integrity_tfm only when receiving a P_PROTOCOL message. l accesses to tconn->peer_integrity_tf happen in worker context, and no further synchronization is necessary. On the sender side, tconn->integrity_tfm is modified under tconn->data.mutex, and a P_PROTOCOL message is sent whenever. All accesses to tconn->integrity_tfm already happen under this mutex. Signed-off-by:
-
Andreas Gruenbacher authored
Signed-off-by:
-
Andreas Gruenbacher authored
We allocate hash transformations with crypto_alloc_hash() which will only return hash algorithms. It is not necessary to reconfirm that we actually got a hash algorithm. Signed-off-by:
-
Andreas Gruenbacher authored
Signed-off-by:
-
Andreas Gruenbacher authored
Signed-off-by:
-
Andreas Gruenbacher authored
It is not enough to grab net_conf->integrity_alg under rcu_read_lock() and access it outside of it; the entire net_conf object may be gone by then. Signed-off-by:
-
Lars Ellenberg authored
Signed-off-by:
-
Lars Ellenberg authored
sc was short for syncer conf, which does not exist anymore anyways. Signed-off-by:
-
Lars Ellenberg authored
Signed-off-by:
-
Lars Ellenberg authored
The DRBD_GENL_F_SET_DEFAULTS flag was ignored for drbd_adm_disk_opts() and drbd_adm_net_opts(). Factor out drbd_set_*_defaults() helper functions, and call them appropriately. Signed-off-by:
-
Philipp Reisner authored
So for this was simply not considered after the options have been re-arranged. Signed-off-by:
-
Lars Ellenberg authored
If an admin requests disconnect at a time when the state handling already disconnects/reconnects, there have been some races. Make sure to always really stop the network threads before returning success for disconnect. Do not pretend successfull forced disconnect, if the state handling returned an error. Return success from drbd_adm_down() only after all threads are finished. Signed-off-by:
-
Lars Ellenberg authored
Calling kobject_uevent, which may sleep, from within rcu_read_lock() protected regions is not possible. This particular kobject_uevent also is also wrong. It was supposed to trigger a udev run, just in case something relevant to udev symlink magic has changed, when adjusting runtime re-configurable settings while we still had the "syncer conf". It was improperly placed in connect when we dropped the "syncer conf". The right thing to do is probably to call "udevadm trigger" directly in those cases where drbdadm thinks there was a need to trigger extra udev runs. Signed-off-by:
-
Philipp Reisner authored
superseded by refcounting Signed-off-by:
-
Philipp Reisner authored
Rule #3 of kref.txt Signed-off-by:
-
Philipp Reisner authored
References hold by: * Each (running) drbd thread has a reference on tconn * Each mdev has a referenc on tconn * Beeing in the all_tconn list counts for one reference * Each after_conn_state_chg_work has a reference to tconn Signed-off-by:
-
Philipp Reisner authored
Signed-off-by:
-
Lars Ellenberg authored
When the last volume of a replication group is unconfigured, the worker thread exits. To not interfere with cleanup of other threads, before the the last cleanups run, we need to make sure the receiver has already exited. The commend explaining that clearly belongs above drbd_thread_stop(&tconn->receiver), not in the cleanup loop below. Signed-off-by:
-
Lars Ellenberg authored
Signed-off-by:
-
Lars Ellenberg authored
We use our own copy of kernel_setsockopt, and did not mess around with get_fs/set_fs, since we thought we knew we would always be KERNEL_DS anyways. Apparently not so for at least user mode linux, so put the set_fs(KERNEL_DS) in there. Signed-off-by:
-
Lars Ellenberg authored
We had drbd_adm_get_status (one single volume), and drbd_adm_get_status_all (dump of all volumes of all resources). This enhances the latter to be able to dump all volumes of just one specific resource. Signed-off-by:
-
Philipp Reisner authored
Now since it is possible to change the two_primaries config flag while the connection is up, make sure we treat a peer_req in a consistent way if the config flag changes while the peer_req is under IO. Signed-off-by:
-
Philipp Reisner authored
Removing the get_net_conf()/put_net_conf() functions Signed-off-by:
-
Philipp Reisner authored
Removing the get_net_conf()/put_net_conf() calls Signed-off-by:
-
Philipp Reisner authored
Signed-off-by:
-
Philipp Reisner authored
Signed-off-by:
-
Philipp Reisner authored
The wire protocol is no longer a property that is negotiated between the two peers. It is now expressed with two bits (DP_SEND_WRITE_ACK and DP_SEND_RECEIVE_ACK) in each data packet. Therefore the primary node is free to change the wire protocol at any time without disconnect/reconnect. Signed-off-by:
-
Philipp Reisner authored
With this commit the locking for all accesses to IDRs is complete: * Non sleeping read accesses are protected by RCU * sleeping read accesses are protocted by a read lock on drbd_cfg_rwsem * accesses that add anything are protected by a write lock * accesses that remove an object are protoected by a write lock and a call to synchronize_rcu() after it is removed from the IDR and before the object is actually free()ed. Signed-off-by:
-
