- 19 Sep, 2016 40 commits
-
-
David Vrabel authored
Write space becoming available may race with putting the task to sleep in xprt_wait_for_buffer_space(). The existing mechanism to avoid the race does not work. This (edited) partial trace illustrates the problem: [1] rpc_task_run_action: task:43546@5 ... action=call_transmit [2] xs_write_space <-xs_tcp_write_space [3] xprt_write_space <-xs_write_space [4] rpc_task_sleep: task:43546@5 ... [5] xs_write_space <-xs_tcp_write_space [1] Task 43546 runs but is out of write space. [2] Space becomes available, xs_write_space() clears the SOCKWQ_ASYNC_NOSPACE bit. [3] xprt_write_space() attemts to wake xprt->snd_task (== 43546), but this has not yet been queued and the wake up is lost. [4] xs_nospace() is called which calls xprt_wait_for_buffer_space() which queues task 43546. [5] The call to sk->sk_write_space() at the end of xs_nospace() (which is supposed to handle the above race) does not call xprt_write_space() as the SOCKWQ_ASYNC_NOSPACE bit is clear and thus the task is not woken. Fix the race by resetting the SOCKWQ_ASYNC_NOSPACE bit in xs_nospace() so the second call to sk->sk_write_space() calls xprt_write_space(). Suggested-by:Trond Myklebust <trondmy@primarydata.com> Signed-off-by:
David Vrabel <david.vrabel@citrix.com> cc: stable@vger.kernel.org # 4.4 Signed-off-by:
Anna Schumaker <Anna.Schumaker@Netapp.com>
-
Jeff Layton authored
Currently, the layout driver selection code always chooses the first one from the list. That's not really ideal however, as the server can send the list of layout types in any order that it likes. It's up to the client to select the best one for its needs. This patch adds an ordered list of preferred driver types and has the selection code sort the list of available layout drivers according to it. Any unrecognized layout type is sorted to the end of the list. For now, the order of preference is hardcoded, but it should be possible to make this configurable in the future. Signed-off-by:
Jeff Layton <jlayton@redhat.com> Reviewed-by:
J. Bruce Fields <bfields@fieldses.org> Signed-off-by:
-
Chuck Lever authored
Clean up: the extra layer of indirection doesn't add value. Signed-off-by:
Chuck Lever <chuck.lever@oracle.com> Signed-off-by:
-
Chuck Lever authored
Clean up: When converting xprtrdma to use the new CQ API, I missed a spot. The naming convention elsewhere is: {svc_rdma,rpcrdma}_wc_{operation} Signed-off-by: -
Chuck Lever authored
Tie frwr debugging messages together by always reporting the address of the frwr. Signed-off-by:
-
Chuck Lever authored
The Version One default inline threshold is still 1KB. But allow testing with thresholds up to 64KB. This maximum is somewhat arbitrary. There's no fundamental architectural limit I'm aware of, but it's good to keep the size of Receive buffers reasonable. Now that Send can use a s/g list, a Send buffer is only as large as each RPC requires. Receive buffers are always the size of the inline threshold, however. Signed-off-by:
-
Chuck Lever authored
An RPC Call message that is sent inline but that has a data payload (ie, one or more items in rq_snd_buf's page list) must be "pulled up:" - call_allocate has to reserve enough RPC Call buffer space to accommodate the data payload - call_transmit has to memcopy the rq_snd_buf's page list and tail into its head iovec before it is sent As the inline threshold is increased beyond its current 1KB default, however, this means data payloads of more than a few KB are copied by the host CPU. For example, if the inline threshold is increased just to 4KB, then NFS WRITE requests up to 4KB would involve a memcpy of the NFS WRITE's payload data into the RPC Call buffer. This is an undesirable amount of participation by the host CPU. The inline threshold may be much larger than 4KB in the future, after negotiation with a peer server. Instead of copying the components of rq_snd_buf into its head iovec, construct a gather list of these components, and send them all in place. The same approach is already used in the Linux server's RPC-over-RDMA reply path. This mechanism also eliminates the need for rpcrdma_tail_pullup, which is used to manage the XDR pad and trailing inline content when a Read list is present. This requires that the pages in rq_snd_buf's page list be DMA-mapped during marshaling, and unmapped when a data-bearing RPC is completed. This is slightly less efficient for very small I/O payloads, but significantly more efficient as data payload size and inline threshold increase past a kilobyte. Signed-off-by:
-
Chuck Lever authored
Have frwr's ro_unmap_sync recognize an invalidated rkey that appears as part of a Receive completion. Local invalidation can be skipped for that rkey. Use an out-of-band signaling mechanism to indicate to the server that the client is prepared to receive RDMA Send With Invalidate. Signed-off-by:
-
Chuck Lever authored
Send an RDMA-CM private message on connect, and look for one during a connection-established event. Both sides can communicate their various implementation limits. Implementations that don't support this sideband protocol ignore it. Once the client knows the server's inline threshold maxima, it can adjust the use of Reply chunks, and eliminate most use of Position Zero Read chunks. Moderately-sized I/O can be done using a pure inline RDMA Send instead of RDMA operations that require memory registration. Signed-off-by:
-
Chuck Lever authored
Introduce data structure used by both client and server to exchange implementation details during RDMA/CM connection establishment. This is an experimental out-of-band exchange between Linux RPC-over-RDMA Version One implementations, replacing the deprecated CCP (see RFC 5666bis). The purpose of this extension is to enable prototyping of features that might be introduced in a subsequent version of RPC-over-RDMA. Suggested by Christoph Hellwig and Devesh Sharma. Signed-off-by:
Sagi Grimberg <sagi@grimberg.me> Signed-off-by:
-
Chuck Lever authored
Clean up: The fields in the recv_wr do not vary. There is no need to initialize them before each ib_post_recv(). This removes a large-ish data structure from the stack. Signed-off-by:
-
Chuck Lever authored
Clean up: Most of the fields in each send_wr do not vary. There is no need to initialize them before each ib_post_send(). This removes a large-ish data structure from the stack. Signed-off-by:
-
Chuck Lever authored
Clean up. Since commit fc664485 ("xprtrdma: Split the completion queue"), rpcrdma_ep_post_recv() no longer uses the "ep" argument. Signed-off-by:
-
Chuck Lever authored
Clean up. The "ia" argument is no longer used. Signed-off-by:
-
Chuck Lever authored
Currently, each regbuf is allocated and DMA mapped at the same time. This is done during transport creation. When a device driver is unloaded, every DMA-mapped buffer in use by a transport has to be unmapped, and then remapped to the new device if the driver is loaded again. Remapping will have to be done _after_ the connect worker has set up the new device. But there's an ordering problem: call_allocate, which invokes xprt_rdma_allocate which calls rpcrdma_alloc_regbuf to allocate Send buffers, happens _before_ the connect worker can run to set up the new device. Instead, at transport creation, allocate each buffer, but leave it unmapped. Once the RPC carries these buffers into ->send_request, by which time a transport connection should have been established, check to see that the RPC's buffers have been DMA mapped. If not, map them there. When device driver unplug support is added, it will simply unmap all the transport's regbufs, but it doesn't have to deallocate the underlying memory. Signed-off-by:
-
Chuck Lever authored
The use of DMA_BIDIRECTIONAL is discouraged by DMA-API.txt. Fortunately, xprtrdma now knows which direction I/O is going as soon as it allocates each regbuf. The RPC Call and Reply buffers are no longer the same regbuf. They can each be labeled correctly now. The RPC Reply buffer is never part of either a Send or Receive WR, but it can be part of Reply chunk, which is mapped and registered via ->ro_map . So it is not DMA mapped when it is allocated (DMA_NONE), to avoid a double- mapping. Since Receive buffers are no longer DMA_BIDIRECTIONAL and their contents are never modified by the host CPU, DMA-API-HOWTO.txt suggests that a DMA sync before posting each buffer should be unnecessary. (See my_card_interrupt_handler). Signed-off-by:
-
Chuck Lever authored
Commit 94931746 ("xprtrdma: Limit number of RDMA segments in RPC-over-RDMA headers") capped the number of chunks that may appear in RPC-over-RDMA headers. The maximum header size can be estimated and fixed to avoid allocating buffer space that is never used. Signed-off-by:
-
Chuck Lever authored
RPC-over-RDMA needs to separate its RPC call and reply buffers. o When an RPC Call is sent, rq_snd_buf is DMA mapped for an RDMA Send operation using DMA_TO_DEVICE o If the client expects a large RPC reply, it DMA maps rq_rcv_buf as part of a Reply chunk using DMA_FROM_DEVICE The two mappings are for data movement in opposite directions. DMA-API.txt suggests that if these mappings share a DMA cacheline, bad things can happen. This could occur in the final bytes of rq_snd_buf and the first bytes of rq_rcv_buf if the two buffers happen to share a DMA cacheline. On x86_64 the cacheline size is typically 8 bytes, and RPC call messages are usually much smaller than the send buffer, so this hasn't been a noticeable problem. But the DMA cacheline size can be larger on other platforms. Also, often rq_rcv_buf starts most of the way into a page, thus an additional RDMA segment is needed to map and register the end of that buffer. Try to avoid that scenario to reduce the cost of registering and invalidating Reply chunks. Instead of carrying a single regbuf that covers both rq_snd_buf and rq_rcv_buf, each struct rpcrdma_req now carries one regbuf for rq_snd_buf and one regbuf for rq_rcv_buf. Some incidental changes worth noting: - To clear out some spaghetti, refactor xprt_rdma_allocate. - The value stored in rg_size is the same as the value stored in the iov.length field, so eliminate rg_size Signed-off-by:
-
Chuck Lever authored
Currently there's a hidden and indirect mechanism for finding the rpcrdma_req that goes with an rpc_rqst. It depends on getting from the rq_buffer pointer in struct rpc_rqst to the struct rpcrdma_regbuf that controls that buffer, and then to the struct rpcrdma_req it goes with. This was done back in the day to avoid the need to add a per-rqst pointer or to alter the buf_free API when support for RPC-over-RDMA was introduced. I'm about to change the way regbuf's work to support larger inline thresholds. Now is a good time to replace this indirect mechanism with something that is more straightforward. I guess this should be considered a clean up. Signed-off-by:
-
Chuck Lever authored
For xprtrdma, the RPC Call and Reply buffers are involved in real I/O operations. To start with, the DMA direction of the I/O for a Call is opposite that of a Reply. In the current arrangement, the Reply buffer address is on a four-byte alignment just past the call buffer. Would be friendlier on some platforms if that was at a DMA cache alignment instead. Because the current arrangement allocates a single memory region which contains both buffers, the RPC Reply buffer often contains a page boundary in it when the Call buffer is large enough (which is frequent). It would be a little nicer for setting up DMA operations (and possible registration of the Reply buffer) if the two buffers were separated, well-aligned, and contained as few page boundaries as possible. Now, I could just pad out the single memory region used for the pair of buffers. But frequently that would mean a lot of unused space to ensure the Reply buffer did not have a page boundary. Add a separate pointer to rpc_rqst that points right to the RPC Reply buffer. This makes no difference to xprtsock, but it will help xprtrdma in subsequent patches. Signed-off-by:
-
Chuck Lever authored
xprtrdma needs to allocate the Call and Reply buffers separately. TBH, the reliance on using a single buffer for the pair of XDR buffers is transport implementation-specific. Instead of passing just the rq_buffer into the buf_free method, pass the task structure and let buf_free take care of freeing both XDR buffers at once. There's a micro-optimization here. In the common case, both xprt_release and the transport's buf_free method were checking if rq_buffer was NULL. Now the check is done only once per RPC. Signed-off-by:
-
Chuck Lever authored
xprtrdma needs to allocate the Call and Reply buffers separately. TBH, the reliance on using a single buffer for the pair of XDR buffers is transport implementation-specific. Transports that want to allocate separate Call and Reply buffers will ignore the "size" argument anyway. Don't bother passing it. The buf_alloc method can't return two pointers. Instead, make the method's return value an error code, and set the rq_buffer pointer in the method itself. This gives call_allocate an opportunity to terminate an RPC instead of looping forever when a permanent problem occurs. If a request is just bogus, or the transport is in a state where it can't allocate resources for any request, there needs to be a way to kill the RPC right there and not loop. This immediately fixes a rare problem in the backchannel send path, which loops if the server happens to send a CB request whose call+reply size is larger than a page (which it shouldn't do yet). One more issue: looks like xprt_inject_disconnect was incorrectly placed in the failure path in call_allocate. It needs to be in the success path, as it is for other call-sites. Signed-off-by:
-
Chuck Lever authored
Clean up: there is some XDR initialization logic that is common to the forward channel and backchannel. Move it to an XDR header so it can be shared. rpc_rqst::rq_buffer points to a buffer containing big-endian data. Update its annotation as part of the clean up. Signed-off-by:
-
Chuck Lever authored
Clean up: r_xprt is already available everywhere these macros are invoked, so just dereference that directly. RPCRDMA_INLINE_PAD_VALUE is no longer used, so it can simply be removed. Signed-off-by:
-
Andy Adamson authored
Try all multipath addresses for a data server. The first address that successfully connects and creates a session is the DS mount address. All subsequent addresses are tested for session trunking and added as aliases. Signed-off-by:
Andy Adamson <andros@netapp.com> Signed-off-by:
-
Andy Adamson authored
Use an async exchange id call to test for session trunking To conform with RFC 5661 section 18.35.4, the Non-Update on Existing Clientid case, save the exchange id verifier in cl_confirm and use it for the session trunking exhange id test. Signed-off-by:
-
Andy Adamson authored
Signed-off-by:
-
Andy Adamson authored
Use a setup function to call into the NFS layer to test an rpc_xprt for session trunking so as to not leak the rpc_xprt_switch into the nfs layer. Search for the address in the rpc_xprt_switch first so as not to put an unnecessary EXCHANGE_ID on the wire. Signed-off-by:
-
Andy Adamson authored
Signed-off-by:
-
Andy Adamson authored
Give the NFS layer access to the rpc_xprt_switch_add_xprt function Signed-off-by:
-
Andy Adamson authored
Give the NFS layer access to the xprt_switch_put function Signed-off-by:
-
Andy Adamson authored
rpc_task_set_client is only called from rpc_run_task after rpc_new_task and rpc_task_release_client is not needed as the task is new. When called from rpc_new_task, rpc_task_set_client also removed the assigned rpc_xprt which is not desired. Signed-off-by:
-
Andy Adamson authored
Signed-off-by:
-
Andy Adamson authored
For session trunking, to compare nfs41_exchange_id_res with existing nfs_client Signed-off-by:
-
Andy Adamson authored
For session trunking, to compare nfs41_exchange_id_res with exiting nfs_client. Signed-off-by:
-
Andy Adamson authored
Testing an rpc_xprt for session trunking should not delay application progress over already established transports. Setup exchange_id to be able to be an async call to test an rpc_xprt for session trunking use. Signed-off-by:
-
Trond Myklebust authored
Add support for the kernel parameter nfs.callback_nr_threads to set the number of threads that will be assigned to the callback channel. Add support for the kernel parameter nfs.nfs.max_session_cb_slots to set the maximum size of the callback channel slot table. Signed-off-by:
Trond Myklebust <trond.myklebust@primarydata.com> Signed-off-by:
-
Trond Myklebust authored
This will allow us to bump the number of callback threads at will. Signed-off-by:
-
Trond Myklebust authored
Ensure that the nfs_callback_info[] array correctly tracks the struct svc_serv. Signed-off-by:
-
Trond Myklebust authored
Clean up. Signed-off-by:
-
