summaryrefslogtreecommitdiffstats
path: root/rpc/rpc-transport/rdma
Commit message (Collapse)AuthorAgeFilesLines
* spelling errors fixed across the code baseAmar Tumballi2011-03-221-3/+3
| | | | | | | | | | | | | recieve -> receive maintainence -> maintenance verison -> version commited -> committed Signed-off-by: Amar Tumballi <amar@gluster.com> Signed-off-by: Vijay Bellur <vijay@dev.gluster.com> BUG: 2262 (Spelling errors in source) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=2262
* rpc-transport: log enhancementAmar Tumballi2011-03-172-66/+48
| | | | | | | | Signed-off-by: Amar Tumballi <amar@gluster.com> Signed-off-by: Vijay Bellur <vijay@dev.gluster.com> BUG: 2346 (Log message enhancements in GlusterFS - phase 1) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=2346
* rpc-transport: whitespace cleanupAmar Tumballi2011-03-173-291/+291
| | | | | | | | Signed-off-by: Amar Tumballi <amar@gluster.com> Signed-off-by: Vijay Bellur <vijay@dev.gluster.com> BUG: 2346 (Log message enhancements in GlusterFS - phase 1) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=2346
* rpc-transport/rdma: handle cases where peer is NULL in send_completion.Raghavendra G2011-03-121-1/+1
| | | | | | | | | | | | | | - There can be valid cases where peer is NULL - For eg., race b/w pollin received on handshake socket and as a result transport getting disconnected and handling of a send/recv completion on the same qp for some previous read/write - in send/recv_completion_process. Hence we should be checking for a valid peer pointer before accessing it. Signed-off-by: Raghavendra G <raghavendra@gluster.com> Signed-off-by: Vijay Bellur <vijay@dev.gluster.com> BUG: 2514 () URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=2514
* rdma : prevent confusing logs when both 'poll_in' and 'poll_err' are received.Amar Tumballi2010-12-271-1/+1
| | | | | | | | Signed-off-by: Amar Tumballi <amar@gluster.com> Signed-off-by: Anand V. Avati <avati@dev.gluster.com> BUG: 875 (Implement a new protocol to provide proper backward/forward compatibility) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=875
* rpc-transport/rdma: QP configuration changes.Raghavendra G2010-12-121-29/+46
| | | | | | | | | | | | - set send and recv count to 4K. - set max_send_sge of QP to 2 since for msgs of type RDMA_NOMSG, there can be two vectors - rpc header and message from program. Signed-off-by: Raghavendra G <raghavendra@gluster.com> Signed-off-by: Anand V. Avati <avati@dev.gluster.com> BUG: 2197 (applications running on large number of clients simultaneously result in ibv_post_send errors on server.) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=2197
* rpc-transport/rdma: add informative debug messages when posting of send ↵Raghavendra G2010-12-121-13/+33
| | | | | | | | | | requests fail. Signed-off-by: Raghavendra G <raghavendra@gluster.com> Signed-off-by: Anand V. Avati <avati@dev.gluster.com> BUG: 2197 (applications running on large number of clients simultaneously result in ibv_post_send errors on server.) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=2197
* rpc-transport: don't merge payload iobuf and iobuf which holds program ↵Raghavendra G2010-11-252-29/+25
| | | | | | | | | | | | | header into single iobref. - io-cache holds a reference on iobref passed from transport layer. Hence, two iobufs are accounted instead of one in calculated used cache size. Signed-off-by: Raghavendra G <raghavendra@gluster.com> Signed-off-by: Anand V. Avati <avati@dev.gluster.com> BUG: 2135 () URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=2135
* rpc-transport/rdma: set the default quota of recv buffers to 128.Raghavendra G2010-11-091-2/+2
| | | | | | | | | | - since quota is returned once the reply is recieved, 32 is too less value. Signed-off-by: Raghavendra G <raghavendra@gluster.com> Signed-off-by: Anand V. Avati <avati@dev.gluster.com> BUG: 2060 (Infiniteband errors on large file find) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=2060
* rpc-transport/rdma: create request_context only if message contains read ↵Raghavendra G2010-11-071-18/+28
| | | | | | | | | | chunks or write chunks. Signed-off-by: Raghavendra G <raghavendra@gluster.com> Signed-off-by: Anand V. Avati <avati@dev.gluster.com> BUG: 1992 (glusterd crashed while submitting a callback.) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=1992
* rpc-transport/rdma: Fix 2KB as threshold size for msgs that can be ↵Raghavendra G2010-11-072-13/+23
| | | | | | | | | | | | | transferred inline. - Any fop/mop that can result a reply whose size is greater than this threshold allocates and submits a buffer along with the request to receive reply. Signed-off-by: Raghavendra G <raghavendra@gluster.com> Signed-off-by: Anand V. Avati <avati@dev.gluster.com> BUG: 513 (Introduce 0 copy rdma) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=513
* rpc-transport/rdma: increment post->ctx.count in a loop doint rdma_read.Raghavendra G2010-11-071-5/+5
| | | | | | | | | | | - incrementing by the number of read chunks even before doing rdma read will cause issues since post->ctx.count is used while posting rdma read. Signed-off-by: Raghavendra G <raghavendra@gluster.com> Signed-off-by: Anand V. Avati <avati@dev.gluster.com> BUG: 1877 (data corruption while running arequal.) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=1877
* rpc-transport: fix race-condition between rdma-read completion and updating ↵Raghavendra G2010-10-291-1/+1
| | | | | | | | | | | | | | the count of number of vectors to be passed to rpc. - If rdma read completes before incrementing the vector count, the count value sent to rpc will be improper. For fops like write, this may result in missing out a vector to be written, thereby causing data corruption. Signed-off-by: Raghavendra G <raghavendra@gluster.com> Signed-off-by: Anand V. Avati <avati@dev.gluster.com> BUG: 1877 (data corruption while running arequal.) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=1877
* Revert "rpc-transport/rdma: create request_context only if message contains ↵Anand V. Avati2010-10-291-26/+18
| | | | | | read chunks or write chunks." This reverts commit b66f4598f4dafb6479232b20c5c521ab0da4e84a.
* rpc-transport/rdma: create request_context only if message contains read ↵Raghavendra G2010-10-261-18/+26
| | | | | | | | | | chunks or write chunks. Signed-off-by: Raghavendra G <raghavendra@gluster.com> Signed-off-by: Anand V. Avati <avati@dev.gluster.com> BUG: 1992 (glusterd crashed while submitting a callback.) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=1992
* rpc-transport/rdma: associate mem-pools with device instead of transport.Raghavendra G2010-10-252-69/+102
| | | | | | | | | | - this will help to reduce memory usage of glusterfs process. Signed-off-by: Raghavendra G <raghavendra@gluster.com> Signed-off-by: Anand V. Avati <avati@dev.gluster.com> BUG: 513 (Introduce 0 copy rdma) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=513
* Change assert to GF_ASSERTVijay Bellur2010-10-121-1/+1
| | | | | | | | Signed-off-by: Vijay Bellur <vijay@gluster.com> Signed-off-by: Vijay Bellur <vijay@dev.gluster.com> BUG: 971 (dynamic volume management) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=971
* rpc-transport/rdma: by default use infiniband port 1Raghavendra G2010-10-121-1/+1
| | | | | | | | Signed-off-by: Raghavendra G <raghavendra@gluster.com> Signed-off-by: Vijay Bellur <vijay@dev.gluster.com> BUG: 513 (Introduce 0 copy rdma) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=513
* Copyright changesVijay Bellur2010-10-114-4/+4
| | | | | | | | Signed-off-by: Vijay Bellur <vijay@gluster.com> Signed-off-by: Vijay Bellur <vijay@dev.gluster.com> BUG: 971 (dynamic volume management) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=971
* rpc-transport/rdma: fix the bug which was causing an ioq-entry being mem_put ↵Raghavendra G2010-10-091-1/+3
| | | | | | | | | | | | | | | | | | | twice. - in case of failure to send message, ioq-entry would be freed, but it used to get added to the ioq-list. Failure in sending a message (for reasons other than insufficient quota), would result in transport being disconnected. In the process of handling pollerr, when ioq-list is flushed, the same ioq-entry which would've already been freed, would be freed for second time. - this patch also fixes spurious disconnects happening when quota was not available while sending a request. Signed-off-by: Raghavendra G <raghavendra@gluster.com> Signed-off-by: Vijay Bellur <vijay@dev.gluster.com> BUG: 1854 (GlusterFS 3.1 beta crashes over Infiniband with rdma issue) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=1854
* migrate from port 69* to 24007Amar Tumballi2010-10-081-2/+2
| | | | | | | | Signed-off-by: Amar Tumballi <amar@gluster.com> Signed-off-by: Vijay Bellur <vijay@dev.gluster.com> BUG: 1870 (Change the port from 69* to some other ports.) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=1870
* Change GNU GPL to GNU AGPLPranith K2010-10-044-12/+12
| | | | | | | | Signed-off-by: Pranith Kumar K <pranithk@gluster.com> Signed-off-by: Vijay Bellur <vijay@dev.gluster.com> BUG: 1388 () URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=1388
* rpc-transport/rdma: send disconnect event only if the transport is already ↵Raghavendra G2010-10-021-3/+7
| | | | | | | | | | | | | | connected. - A disconnect event can be sent before a connect event to rpc, if some error happens during rdma handshake. - Also call rdma_quota_put only if peer is not NULL. Signed-off-by: Raghavendra G <raghavendra@gluster.com> Signed-off-by: Vijay Bellur <vijay@dev.gluster.com> BUG: 513 (Introduce 0 copy rdma) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=513
* rpc/rpc-transport: fix frame-loss during rdma-reads.v3.1.0qa31Raghavendra G2010-09-242-3/+18
| | | | | | | | | | | | | | | | | | - total no of rdma reads in progress was tracked using refcount of post (refcount = total rdma reads + 1). However doing rdma_post_unref _after_ rdma_do_reads, can result in race condition between execution of rdma_post_unref and procedure handling rdma read completion. This makes it impossible to find out whether the current rdma read was the last rdma-read being done in procedure handling rdma read-completion. The fix to this should either do rdma_post_unref before posting an rdma read or use another variable to track the number of rdma-reads. This patch implements second method. Signed-off-by: Raghavendra G <raghavendra@gluster.com> Signed-off-by: Vijay Bellur <vijay@dev.gluster.com> BUG: 1700 (write fops are bailing out on a distribute setup) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=1700
* rdma: do event_unregister() in fini() too.Amar Tumballi2010-09-171-12/+19
| | | | | | | | Signed-off-by: Amar Tumballi <amar@gluster.com> Signed-off-by: Vijay Bellur <vijay@dev.gluster.com> BUG: 1611 (crash in gf_free) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=1611
* rpc-transport/rdma: while notifying a reply, pickup the correct iobref from ↵Raghavendra G2010-09-081-3/+3
| | | | | | | | | | request context. Signed-off-by: Raghavendra G <raghavendra@gluster.com> Signed-off-by: Vijay Bellur <vijay@dev.gluster.com> BUG: 1567 ([rdma] data corruption due to read-ahead) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=1567
* rpc-transport/rdma: change the default listening port to 6968.Raghavendra G2010-09-081-1/+1
| | | | | | | | Signed-off-by: Raghavendra G <raghavendra@gluster.com> Signed-off-by: Vijay Bellur <vijay@dev.gluster.com> BUG: 513 (Introduce 0 copy rdma) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=513
* rpc-transport/rdma: honour port argument sent in rdma_connect.v3.1.0qa17Raghavendra G2010-09-073-21/+26
| | | | | | | | Signed-off-by: Raghavendra G <raghavendra@gluster.com> Signed-off-by: Vijay Bellur <vijay@dev.gluster.com> BUG: 1555 (rdma remote port defaults to 6997) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=1555
* rpc-transport/rdma: fix memory corruptions caused by rdma.Raghavendra G2010-09-071-2/+4
| | | | | | | | | | | | | - disconnect notification should be sent before unrefing transport, since transport might be freed when unrefed. - set trans->listener. This member is used by rpcsvc to decide whether a listener or a new connection is dead. Signed-off-by: Raghavendra G <raghavendra@gluster.com> Signed-off-by: Vijay Bellur <vijay@dev.gluster.com> BUG: 1558 (glusterd hung) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=1558
* rpc-transport/rdma: notify rpc when pollerr happens, so that client side can ↵Raghavendra G2010-09-031-2/+2
| | | | | | | | | | try to reconnect. Signed-off-by: Raghavendra G <raghavendra@gluster.com> Signed-off-by: Vijay Bellur <vijay@dev.gluster.com> BUG: 513 (Introduce 0 copy rdma) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=513
* rpc-transport/rdma: when rdma msg of type RDMA_ERROR is recieved, disconnect ↵Raghavendra G2010-09-021-6/+13
| | | | | | | | | | transport. Signed-off-by: Raghavendra G <raghavendra@gluster.com> Signed-off-by: Vijay Bellur <vijay@dev.gluster.com> BUG: 1462 () URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=1462
* rpc-transport/rdma: set whether pollin data is for request/reply.Raghavendra G2010-09-011-0/+2
| | | | | | | | | | | - with implementation of call backs client and server can both recieve request/reply. Signed-off-by: Raghavendra G <raghavendra@gluster.com> Signed-off-by: Vijay Bellur <vijay@dev.gluster.com> BUG: 1488 (rdma client fails to connect to the servers) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=1488
* rpc-transport/rdma: free ioq entry even when churning of message fails.Raghavendra G2010-08-311-1/+1
| | | | | | | | | | | - An ioq entry is added to ioq-list only when it is not processed due to lack of quota. For all other cases, it should be freed to avoid memory leak. Signed-off-by: Raghavendra G <raghavendra@gluster.com> Signed-off-by: Vijay Bellur <vijay@dev.gluster.com> BUG: 513 (Introduce 0 copy rdma) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=513
* Bring in new transport rdma.Raghavendra G2010-08-306-0/+5971
- rdma is new transport and improvement over current ib-verbs. It uses rdma-read and rdma-write for efficiently transferring large buffers. For more details please refer to rfc-5666 and rfc-5667. Signed-off-by: Raghavendra G <raghavendra@gluster.com> Signed-off-by: Vijay Bellur <vijay@dev.gluster.com> BUG: 513 (Introduce 0 copy rdma) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=513