- 25 Oct, 2010 4 commits
-
-
J. Bruce Fields authored
If any xprt marked DEAD is also left BUSY for the rest of its life, then the XPT_DEAD check here is superfluous--we'll get the same result from the XPT_BUSY check just after. Signed-off-by:J. Bruce Fields <bfields@redhat.com>
-
J. Bruce Fields authored
As long as DEAD exports are left BUSY, and svc_delete_xprt is called only with BUSY held, then svc_delete_xprt() will never be called on an xprt that is already DEAD. Signed-off-by: -
J. Bruce Fields authored
Once an xprt has been deleted, there's no reason to allow it to be enqueued--at worst, that might cause the xprt to be re-added to some global list, resulting in later corruption. Also, note this leaves us with no need for the reference-count manipulation here. Reviewed-by:
NeilBrown <neilb@suse.de> Signed-off-by:
-
J. Bruce Fields authored
We're doing an allocation under a spinlock, and ignoring the possibility of allocation failure. A better fix wouldn't require an unnecessary allocation in the common case, but we'll leave that for later. Signed-off-by:
-
- 21 Oct, 2010 14 commits
-
-
J. Bruce Fields authored
In the sessions backchannel case, we don't need a krb5 principal name for the client; we use the already-created forechannel credentials instead. Some cleanup, while we're there: make it clearer which code here is 4.0- or sessions- specific. Signed-off-by: -
J. Bruce Fields authored
The minorversion seems more a property of the client than the callback channel. Some time we should probably also enforce consistent minorversion usage from the client; for now, this is just a cosmetic change. Signed-off-by: -
J. Bruce Fields authored
Have unhash_client_locked() remove client and associated sessions from global hashes, but delay further dismantling till free_client(). (After unhash_client_locked(), the only remaining references outside the destroying thread are from any connections which have xpt_user callbacks registered.) This will simplify locking on session destruction. Signed-off-by: -
J. Bruce Fields authored
Only one of the nfsd4_callback_probe callers actually cares about changing the callback information. Signed-off-by: -
J. Bruce Fields authored
The callback program is allowed to depend on the session which the callback is going over. No change in behavior yet, while we still only do callbacks over a single session for the lifetime of the client. Signed-off-by: -
J. Bruce Fields authored
We need to keep track of which connections are available for use with the backchannel, which for the forechannel, and which for both. Signed-off-by:J. Bruce Fields <bfields@citi.umich.edu>
-
J. Bruce Fields authored
Following rfc 5661, section 18.36.4: "If the session is not successfully created, then no changes are made to any client records on the server." We shouldn't be confirming or incrementing the sequence id in this case. Signed-off-by: -
J. Bruce Fields authored
Currently we don't deal well with a client that has multiple sessions associated with it (even simultaneously, or serially over the lifetime of the client). In particular, we don't attempt to keep the backchannel running after the original session diseappears. We will fix that soon. Once we do that, we need the slot sequence number to be per-session; otherwise, for example, we cannot correctly handle a case like this: - All session 1 connections are lost. - The client creates session 2. We use it for the backchannel (since it's the only working choice). - The client gives us a new connection to use with session 1. - The client destroys session 2. At this point our only choice is to go back to using session 1. When we do so we must use the sequence number that is next for session 1. We therefore need to maintain multiple sequence number streams. Signed-off-by: -
J. Bruce Fields authored
Instead of copying the sessionid, use the new cl_cb_session pointer, which indicates which session we're using for the backchannel. Signed-off-by: -
J. Bruce Fields authored
The backchannel should be associated with a session, it isn't really global to the client. We do, however, want a pointer global to the client which tracks which session we're currently using for client-based callbacks. This is a first step in that direction; for now, just reshuffling of code with no significant change in behavior. Signed-off-by: -
J. Bruce Fields authored
Signed-off-by: -
Chuck Lever authored
The source address field in the transport's sock_xprt is initialized ONLY IF the RPC application passed a pointer to a source address during the call to rpc_create(). However, xs_bind() subsequently uses the value of this field without regard to whether the source address was initialized during transport creation or not. So far we've been lucky: the uninitialized value of this field is zeroes. xs_bind(), until recently, used only the sin[6]_addr field in this sockaddr, and all zeroes is a valid value for this: it means ANYADDR. This is a happy coincidence. However, xs_bind() now wants to use the sa_family field as well, and expects it to be initialized to something other than zero. Therefore, the source address sockaddr field should be fully initialized at transport create time in _every_ case, not just when the RPC application wants to use a specific bind address. Bruce added a workaround for this missing initialization by adjusting commit 6bc9638a, but the "right" way to do this is to ensure that the source address sockaddr is always correctly initialized from the get-go. This patch doesn't introduce a behavior change. It's simply a clean-up of Bruce's fix, to prevent future problems of this kind. It may look like overkill, but a) it clearly documents the default initial value of this field, b) it doesn't assume that the sockaddr_storage memory is first initialized to any particular value, and c) it will fail verbosely if some unknown address family is passed in Originally introduced by commit d3bc9a1d. Signed-off-by:
Chuck Lever <chuck.lever@oracle.com> Signed-off-by:
-
Chuck Lever authored
Clean up. Defensive coding: If "family" is ever something that is neither AF_INET nor AF_INET6, xs_reclassify_socket6() is not the appropriate default action. Choose to do nothing in that case. Introduced by commit 6bc9638a. Signed-off-by:
-
Tejun Heo authored
* Create and use svc_rdma_wq instead of using the system workqueue and flush_scheduled_work(). This workqueue is necessary to serve as flushing domain for rdma->sc_work which is used to destroy itself and thus can't be flushed explicitly. * Replace cancel_delayed_work() + flush_scheduled_work() with cancel_delayed_work_sync(). * Implement synchronous connect in xprt_rdma_connect() using flush_delayed_work() on the rdma_connect work instead of using flush_scheduled_work(). This is to prepare for the deprecation and removal of flush_scheduled_work(). Signed-off-by:
Tejun Heo <tj@kernel.org> Signed-off-by:
-
- 19 Oct, 2010 17 commits
-
-
Pavel Emelyanov authored
Saves some lines of code and some branticks when reading one. Signed-off-by:
Pavel Emelyanov <xemul@openvz.org> Reviewed-by:
-
Pavel Emelyanov authored
Since the xprt in question is forcibly set to be bound the else branch of this check is unneeded. Signed-off-by:
-
Pavel Emelyanov authored
> The reason for this is in the future, we may want to support additional > address family types. We should, therefore, ensure that every piece of > code that is sensitive to address families fail in some orderly manner > to let developers know where a change is needed. Makes sense. I was under impression, that AF-s other than INET are not cared about at all :( Here's a fixed version of the patch. Log: Its callers check for ERR_PTR. Signed-off-by:
-
Pavel Emelyanov authored
The task in question is dereferenced above (and is actually never NULL). Signed-off-by:
-
Pavel Emelyanov authored
Same for UDP sockets creation paths. Signed-off-by:
-
Pavel Emelyanov authored
The v4 and the v6 wrappers only pass the respective family to the xs_tcp_setup_socket. This family can be taken from the xprt's sockaddr. Signed-off-by:
-
Pavel Emelyanov authored
Now we have a single socket creation routine and can call it directly from the setup_socket routines. Signed-off-by:
-
Pavel Emelyanov authored
After xs_bind is merged it's easy to merge its callers. Signed-off-by:
-
Pavel Emelyanov authored
There's the only difference betseen the xs_bind4 and the xs_bind6 - the size of sockaddr structure they use. Fortunatelly its size can be indirectly get from the transport. Change since v1: * use sockaddr_storage instead of sockaddr * use rpc_set_port instead of manual port assigning Signed-off-by:
-
Pavel Emelyanov authored
Remove now unneeded wrappers that just add type and protocol to socket creation callback. Signed-off-by:
-
Pavel Emelyanov authored
Same patch for v6 protocols. Signed-off-by:
-
Pavel Emelyanov authored
The UDPv4 and TCPv4 socket creation callbacks now look very similar. Signed-off-by:
-
Pavel Emelyanov authored
Make it look like the TCP sockets creation. Unfortunately the git diff made the patch look messy :( Signed-off-by:
-
Pavel Emelyanov authored
The xs_tcp_reuse_connection takes the xprt only to pass it down to the xs_abort_connection. The later one can get it from the given transport itself. Signed-off-by:
-
Pavel Emelyanov authored
Signed-off-by:
-
Pavel Emelyanov authored
Signed-off-by:
-
Pavel Emelyanov authored
Signed-off-by:
-
- 18 Oct, 2010 2 commits
-
-
Tom Tucker authored
There are several error paths in the code that do not unmap DMA. This patch adds calls to svc_rdma_unmap_dma to free these DMA contexts. Signed-off-by:
Tom Tucker <tom@opengridcomputing.com> Signed-off-by:
-
Tom Tucker authored
There was logic in the send path that assumed that a page containing data to send to the client has a KVA. This is not always the case and can result in data corruption when page_address returns zero and we end up DMA mapping zero. This patch changes the bus mapping logic to avoid page_address() where necessary and converts all calls from ib_dma_map_single to ib_dma_map_page in order to keep the map/unmap calls symmetric. Signed-off-by:
Tom Tucker <tom@ogc.us> Signed-off-by:
-
- 12 Oct, 2010 2 commits
-
-
J. Bruce Fields authored
Expire clients more promptly, at the expense of possibly running the laundromat thread more frequently. Though it's not the default, I'd like it to be feasible to run with a lease time of just a few seconds, at which point a minimum 10 second wait between laundromat runs seems a little much. Signed-off-by: -
Pavel Emelyanov authored
Signed-off-by:
-
- 11 Oct, 2010 1 commit
-
-
NeilBrown authored
We limit the number of 'defer' requests to DFR_MAX. The imposition of this limit is spread about a bit - sometime we don't add new things to the list, sometimes we remove old things. Also it is currently applied to requests which we are 'waiting' for rather than 'deferring'. This doesn't seem ideal as 'waiting' requests are naturally limited by the number of threads. So gather the DFR_MAX handling code to one place and only apply it to requests that are actually being deferred. This means that not all 'cache_deferred_req' structures go on the 'cache_defer_list, so we need to be careful when adding and removing things. Signed-off-by:
-
