- 05 Dec, 2009 12 commits
-
-
Ricardo Labiaga authored
Signed-off-by: Ricardo Labiaga <Ricardo.Labiaga@netapp.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
-
Trond Myklebust authored
Otherwise we have no guarantees that other processes won't start another RPC call while we're resetting the session. Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
-
Alexandros Batsakis authored
in NFSv4.1 the seqid part of a stateid in CB_RECALL must be 0 Signed-off-by: Alexandros Batsakis <batsakis@netapp.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
-
Alexandros Batsakis authored
the server can indicate a number of error conditions by setting the appropriate bits in the SEQUENCE operation. The client re-establishes state with the server when it receives one of those, with the action depending on the specific case. Signed-off-by: Alexandros Batsakis <batsakis@netapp.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
-
Alexandros Batsakis authored
The v4.1 client should take into account the desired rsize, wsize when negotiating the max size in CREATE_SESSION. Accordingly, it should use rsize, wsize that are smaller than the session negotiated values. Signed-off-by: Alexandros Batsakis <batsakis@netapp.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
-
Alexandros Batsakis authored
Signed-off-by: Alexandros Batsakis <batsakis@netapp.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
-
Alexandros Batsakis authored
In v4.1 the client MUST/SHOULD use the EXCLUSIVE4_1 flag instead of EXCLUSIVE4, and GUARDED when the server supports persistent sessions. For now (and until we support suppattr_exclcreat), we don't send any attributes with EXCLUSIVE4_1 relying in the subsequent SETATTR as in v4.0 Signed-off-by: Alexandros Batsakis <batsakis@netapp.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
-
Alexandros Batsakis authored
Signed-off-by: Alexandros Batsakis <batsakis@netapp.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
-
Alexandros Batsakis authored
For now the clients returns _all_ the delegations of the specificed type it holds Signed-off-by: Alexandros Batsakis <batsakis@netapp.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
-
Alexandros Batsakis authored
Signed-off-by: Alexandros Batsakis <batsakis@netapp.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
-
Alexandros Batsakis authored
Signed-off-by: Alexandros Batsakis <batsakis@netapp.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
-
Alexandros Batsakis authored
The NFSv4.1 spec-29 (18.36.3) says that the server MUST use an ONC RPC (program) version number equal to 4 in callbacks sent to the client. For now we allow both versions 1 and 4. Signed-off-by: Alexandros Batsakis <batsakis@netapp.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
-
- 04 Dec, 2009 14 commits
-
-
Andy Adamson authored
Replace sync and async handlers setting of the NFS4CLNT_SESSION_SETUP bit with setting NFS4CLNT_CHECK_LEASE, and let the state manager decide to reset the session. Signed-off-by: Andy Adamson <andros@netapp.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
-
Andy Adamson authored
Do not wake up the next slot_tbl_waitq task in nfs4_free_slot because we may be draining the slot. Either signal the state manager that the session is drained (the state manager wakes up tasks) OR wake up the next task. In nfs41_sequence_done, the slot dereference is only needed in the sequence operation success case. Signed-off-by: Andy Adamson <andros@netapp.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
-
Andy Adamson authored
If the session is reset during state recovery, the state manager thread can sleep on the slot_tbl_waitq causing a deadlock. Add a completion framework to the session. Have the state manager thread set a new session state (NFS4CLNT_SESSION_DRAINING) and wait for the session slot table to drain. Signal the state manager thread in nfs41_sequence_free_slot when the NFS4CLNT_SESSION_DRAINING bit is set and the session is drained. Reported-by: Trond Myklebust <trond@netapp.com> Signed-off-by: Andy Adamson <andros@netapp.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
-
Andy Adamson authored
nfs4_recover_session can put rpciod to sleep. Just use nfs4_schedule_recovery. Reported-by: Trond Myklebust <trond.myklebust@netapp.com> Signed-off-by: Andy Adamson <andros@netapp.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
-
Andy Adamson authored
Signed-off-by: Andy Adamson <andros@netapp.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
-
Andy Adamson authored
Do not fall through and set NFS4CLNT_SESSION_RESET bit on NFS4ERR_EXPIRED Signed-off-by: Andy Adamson <andros@netapp.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
-
Andy Adamson authored
Do not fall through and call nfs4_delay on session error handling. Signed-off-by: Andy Adamson <andros@netapp.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
-
Andy Adamson authored
nfs4_read_done returns zero on unhandled errors. nfs_readpage_result will return on a negative tk_status without freeing the slot. Call nfs4_sequence_free_slot on unhandled errors in nfs4_read_done. Signed-off-by: Andy Adamson <andros@netapp.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
-
Andy Adamson authored
nfs41_sequence_free_slot can be called multiple times on SEQUENCE operation errors. No reason to inline nfs4_restart_rpc Reported-by: Trond Myklebust <trond.myklebust@netapp.com> nfs_writeback_done and nfs_readpage_retry call nfs4_restart_rpc outside the error handler, and the slot is not freed prior to restarting in the rpc_prepare state during session reset. Fix this by moving the call to nfs41_sequence_free_slot from the error path of nfs41_sequence_done into nfs4_restart_rpc, and by removing the test for NFS4CLNT_SESSION_SETUP. Always free slot and goto the rpc prepare state on async errors. Signed-off-by: Andy Adamson <andros@netapp.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
-
Andy Adamson authored
Make this clear by calling rpc_restart-call. Prepare for nfs4_restart_rpc() to free slots. Signed-off-by: Andy Adamson <andros@netapp.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
-
Andy Adamson authored
The bit is no longer used for session setup, only for session reset. Signed-off-by: Andy Adamson <andros@netapp.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
-
Andy Adamson authored
Reported-by: Trond Myklebust <trond.myklebust@netapp.com> Resetting the clientid from the state manager could result in not confirming the clientid due to create session not being called. Move the create session call from the NFS4CLNT_SESSION_SETUP state manager initialize session case into the NFS4CLNT_LEASE_EXPIRED case establish_clid call. Signed-off-by: Andy Adamson <andros@netapp.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
-
Trond Myklebust authored
-
NeilBrown authored
NFS4ERR_FILE_OPEN is return by the server when an operation cannot be performed because the file is currently open and local (to the server) semantics prohibit the operation while the file is open. A typical case is a RENAME operation on an MS-Windows platform, which prevents rename while the file is open. While it is possible that such a condition is transitory, it is also very possible that the file will be held open for an extended period of time thus preventing the operation. The current behaviour of Linux/NFS is to retry the operation indefinitely. This is not appropriate - we do not expect a rename to take an arbitrary amount of time to complete. Rather, and error should be returned. The most obvious error code would be EBUSY, which is a legal at least for 'rename' and 'unlink', and accurately captures the reason for the error. This patch allows a few retries until about 2 seconds have elapsed, then returns EBUSY. Signed-off-by: NeilBrown <neilb@suse.de> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
-
- 03 Dec, 2009 14 commits
-
-
Trond Myklebust authored
-
Miklos Szeredi authored
The d_instantiate(new_dentry, NULL) is superfluous, the dentry is already negative. Rehashing this dummy dentry isn't needed either, d_move() works fine on an unhashed target. The re-checking for busy after a failed nfs_sillyrename() is bogus too: new_dentry->d_count < 2 would be a bug here. Signed-off-by: Miklos Szeredi <mszeredi@suse.cz> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
-
Miklos Szeredi authored
Move unhashing the target to after the check for existence and being a non-directory. If renaming a directory then the VFS already unhashes the target if it is not busy. If it's busy then acquiring more references during the rename makes no difference. Signed-off-by: Miklos Szeredi <mszeredi@suse.cz> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
-
Miklos Szeredi authored
Comments are wrong or out of date. In particular d_drop() doesn't free the inode it just unhashes the dentry. And if target is a directory then it is not checked for being busy. Signed-off-by: Miklos Szeredi <mszeredi@suse.cz> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
-
Miklos Szeredi authored
VFS already checks if both source and target are directories. Signed-off-by: Miklos Szeredi <mszeredi@suse.cz> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
-
Trond Myklebust authored
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
-
Chuck Lever authored
Introduce soft connect behavior for UDP transports. In this case, a major timeout returns ETIMEDOUT instead of EIO. Signed-off-by: Chuck Lever <chuck.lever@oracle.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
-
Chuck Lever authored
Currently, if a remote RPC service is unreachable, an RPC ping will hang until the underlying transport connect attempt times out. A more desirable behavior might be to have the ping fail immediately so upper layers can recover appropriately. In the case of an NFS mount, for instance, this would mean the mount(2) system call could fail immediately if the server isn't listening, rather than hanging uninterruptibly for more than 3 minutes. Change rpc_ping() so that it fails immediately for connection-oriented transports. rpc_create() will then fail immediately for such transports if an RPC ping was requested. Signed-off-by: Chuck Lever <chuck.lever@oracle.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
-
Chuck Lever authored
Autobinding is handled by the rpciod process, not in user processes that are generating regular RPC requests. Thus autobinding is usually not affected by signals targetting user processes, such as KILL or timer expiration events. In addition, an RPC request generated by a user process that has RPC_TASK_SOFTCONN set and needs to perform an autobind will hang if the remote rpcbind service is not available. For rpcbind queries on connection-oriented transports, let's use the new soft connect semantic to return control to the user's process quickly, if the kernel's rpcbind client can't connect to the remote rpcbind service. Logic is introduced in call_bind_status() to handle connection errors that occurred during an asynchronous rpcbind query. The logic abandons the rpcbind query if the RPC request has SOFTCONN set, and retries after a few seconds in the normal case. Signed-off-by: Chuck Lever <chuck.lever@oracle.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
-
Chuck Lever authored
Use TCP with the soft connect semantic for local rpcbind upcalls so the kernel can detect immediately if the local rpcbind daemon is not running. Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
-
Chuck Lever authored
The kernel's rpcbind client creates and deletes an rpc_clnt and its underlying transport socket for every upcall to the local rpcbind daemon. When starting a typical NFS server on IPv4 and IPv6, the NFS service itself does three upcalls (one per version) times two upcalls (one per transport) times two upcalls (one per address family), making 12, plus another one for the initial call to unregister previous NFS services. Starting the NLM service adds an additional 13 upcalls, for similar reasons. (Currently the NFS service doesn't start IPv6 listeners, but it will soon enough). Instead, let's create an rpc_clnt for rpcbind upcalls during the first local rpcbind query, and cache it. This saves the overhead of creating and destroying an rpc_clnt and a socket for every upcall. The new logic also prevents the kernel from attempting an RPCB_SET or RPCB_UNSET if it knows from the start that the local portmapper does not support rpcbind protocol version 4. This will cut down on the number of rpcbind upcalls in legacy environments. Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
-
Chuck Lever authored
Clean up: At one point, rpcb_local_clnt() handled IPv6 loopback addresses too, but it doesn't any more; only IPv4 loopback is used now. Get rid of the @addr and @addrlen arguments to rpcb_local_clnt(). Signed-off-by: Chuck Lever <chuck.lever@oracle.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
-
Chuck Lever authored
The kernel sometimes makes RPC calls to services that aren't running. Because the kernel's RPC client always assumes the hard retry semantic when reconnecting a connection-oriented RPC transport, the underlying reconnect logic takes a long while to time out, even though the remote may have responded immediately with ECONNREFUSED. In certain cases, like upcalls to our local rpcbind daemon, or for NFS mount requests, we'd like the kernel to fail immediately if the remote service isn't reachable. This allows another transport to be tried immediately, or the pending request can be abandoned quickly. Introduce a per-request flag which controls how call_transmit_status() behaves when request transmission fails because the server cannot be reached. We don't want soft connection semantics to apply to other errors. The default case of the switch statement in call_transmit_status() no longer falls through; the fall through code is copied to the default case, and a "break;" is added. The transport's connection re-establishment timeout is also ignored for such requests. We want the request to fail immediately, so the reconnect delay is skipped. Additionally, we don't want a connect failure here to further increase the reconnect timeout value, since this request will not be retried. Signed-off-by: Chuck Lever <chuck.lever@oracle.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
-
Chuck Lever authored
The success case, where task->tk_status == 0, is by far the most frequent case in call_transmit_status(). The default: arm of the switch statement in call_transmit_status() handles the 0 case. default: was moved close to the top of the switch statement in call_transmit_status() under the theory that the compiler places object code for the earliest arms of a switch statement first, making the CPU do less work. The default: arm of a switch statement, however, is executed only after all the other cases have been checked. Even if the compiler rearranges the object code, the default: arm is the "last resort", meaning all of the other cases have been explicitly exhausted. That makes the current arrangement about as inefficient as it gets for the common case. To fix this, add an explicit check for zero before the switch statement. That forces the compiler to do the zero check first, no matter what optimizations it might try to do to the switch statement. Signed-off-by: Chuck Lever <chuck.lever@oracle.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
-