From 399f11c3d872bd748e1575574de265a6304c7c43 Mon Sep 17 00:00:00 2001 From: Bryan Schumaker Date: Tue, 30 Oct 2012 16:06:35 -0400 Subject: NFS: Wait for session recovery to finish before returning Currently, we will schedule session recovery and then return to the caller of nfs4_handle_exception. This works for most cases, but causes a hang on the following test case: Client Server ------ ------ Open file over NFS v4.1 Write to file Expire client Try to lock file The server will return NFS4ERR_BADSESSION, prompting the client to schedule recovery. However, the client will continue placing lock attempts and the open recovery never seems to be scheduled. The simplest solution is to wait for session recovery to run before retrying the lock. Signed-off-by: Bryan Schumaker Signed-off-by: Trond Myklebust Cc: stable@vger.kernel.org --- fs/nfs/nfs4proc.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-) (limited to 'fs/nfs/nfs4proc.c') diff --git a/fs/nfs/nfs4proc.c b/fs/nfs/nfs4proc.c index 68b21d81b7ac..d5fbf1f49d5f 100644 --- a/fs/nfs/nfs4proc.c +++ b/fs/nfs/nfs4proc.c @@ -339,8 +339,7 @@ static int nfs4_handle_exception(struct nfs_server *server, int errorcode, struc dprintk("%s ERROR: %d Reset session\n", __func__, errorcode); nfs4_schedule_session_recovery(clp->cl_session, errorcode); - exception->retry = 1; - break; + goto wait_on_recovery; #endif /* defined(CONFIG_NFS_V4_1) */ case -NFS4ERR_FILE_OPEN: if (exception->timeout > HZ) { -- cgit v1.2.3 From 2240a9e2d013d8269ea425b73e1d7a54c7bc141f Mon Sep 17 00:00:00 2001 From: Trond Myklebust Date: Mon, 29 Oct 2012 18:37:40 -0400 Subject: NFSv4.1: We must release the sequence id when we fail to get a session slot If we do not release the sequence id in cases where we fail to get a session slot, then we can deadlock if we hit a recovery scenario. Signed-off-by: Trond Myklebust Cc: stable@vger.kernel.org --- fs/nfs/nfs4proc.c | 36 +++++++++++++++++++++++------------- 1 file changed, 23 insertions(+), 13 deletions(-) (limited to 'fs/nfs/nfs4proc.c') diff --git a/fs/nfs/nfs4proc.c b/fs/nfs/nfs4proc.c index d5fbf1f49d5f..e0423bb5a880 100644 --- a/fs/nfs/nfs4proc.c +++ b/fs/nfs/nfs4proc.c @@ -1571,9 +1571,11 @@ static void nfs4_open_prepare(struct rpc_task *task, void *calldata) data->timestamp = jiffies; if (nfs4_setup_sequence(data->o_arg.server, &data->o_arg.seq_args, - &data->o_res.seq_res, task)) - return; - rpc_call_start(task); + &data->o_res.seq_res, + task) != 0) + nfs_release_seqid(data->o_arg.seqid); + else + rpc_call_start(task); return; unlock_no_action: rcu_read_unlock(); @@ -2295,9 +2297,10 @@ static void nfs4_close_prepare(struct rpc_task *task, void *data) if (nfs4_setup_sequence(NFS_SERVER(inode), &calldata->arg.seq_args, &calldata->res.seq_res, - task)) - goto out; - rpc_call_start(task); + task) != 0) + nfs_release_seqid(calldata->arg.seqid); + else + rpc_call_start(task); out: dprintk("%s: done!\n", __func__); } @@ -4544,9 +4547,11 @@ static void nfs4_locku_prepare(struct rpc_task *task, void *data) calldata->timestamp = jiffies; if (nfs4_setup_sequence(calldata->server, &calldata->arg.seq_args, - &calldata->res.seq_res, task)) - return; - rpc_call_start(task); + &calldata->res.seq_res, + task) != 0) + nfs_release_seqid(calldata->arg.seqid); + else + rpc_call_start(task); } static const struct rpc_call_ops nfs4_locku_ops = { @@ -4691,7 +4696,7 @@ static void nfs4_lock_prepare(struct rpc_task *task, void *calldata) /* Do we need to do an open_to_lock_owner? */ if (!(data->arg.lock_seqid->sequence->flags & NFS_SEQID_CONFIRMED)) { if (nfs_wait_on_sequence(data->arg.open_seqid, task) != 0) - return; + goto out_release_lock_seqid; data->arg.open_stateid = &state->stateid; data->arg.new_lock_owner = 1; data->res.open_seqid = data->arg.open_seqid; @@ -4700,10 +4705,15 @@ static void nfs4_lock_prepare(struct rpc_task *task, void *calldata) data->timestamp = jiffies; if (nfs4_setup_sequence(data->server, &data->arg.seq_args, - &data->res.seq_res, task)) + &data->res.seq_res, + task) == 0) { + rpc_call_start(task); return; - rpc_call_start(task); - dprintk("%s: done!, ret = %d\n", __func__, data->rpc_status); + } + nfs_release_seqid(data->arg.open_seqid); +out_release_lock_seqid: + nfs_release_seqid(data->arg.lock_seqid); + dprintk("%s: done!, ret = %d\n", __func__, task->tk_status); } static void nfs4_recover_lock_prepare(struct rpc_task *task, void *calldata) -- cgit v1.2.3 From 2b1bc308f492589f7d49012ed24561534ea2be8c Mon Sep 17 00:00:00 2001 From: Trond Myklebust Date: Mon, 29 Oct 2012 18:53:23 -0400 Subject: NFSv4: nfs4_locku_done must release the sequence id If the state recovery machinery is triggered by the call to nfs4_async_handle_error() then we can deadlock. Signed-off-by: Trond Myklebust Cc: stable@vger.kernel.org --- fs/nfs/nfs4proc.c | 1 + 1 file changed, 1 insertion(+) (limited to 'fs/nfs/nfs4proc.c') diff --git a/fs/nfs/nfs4proc.c b/fs/nfs/nfs4proc.c index e0423bb5a880..1465364501ba 100644 --- a/fs/nfs/nfs4proc.c +++ b/fs/nfs/nfs4proc.c @@ -4531,6 +4531,7 @@ static void nfs4_locku_done(struct rpc_task *task, void *data) if (nfs4_async_handle_error(task, calldata->server, NULL) == -EAGAIN) rpc_restart_call_prepare(task); } + nfs_release_seqid(calldata->arg.seqid); } static void nfs4_locku_prepare(struct rpc_task *task, void *data) -- cgit v1.2.3 From 324d003b0cd82151adbaecefef57b73f7959a469 Mon Sep 17 00:00:00 2001 From: Weston Andros Adamson Date: Tue, 30 Oct 2012 17:01:39 -0400 Subject: NFS: add nfs_sb_deactive_async to avoid deadlock Use nfs_sb_deactive_async instead of nfs_sb_deactive when in a workqueue context. This avoids a deadlock where rpc_shutdown_client loops forever in a workqueue kworker context, trying to kill all RPC tasks associated with the client, while one or more of these tasks have already been assigned to the same kworker (and will never run rpc_exit_task). This approach is needed because RPC tasks that have already been assigned to a kworker by queue_work cannot be canceled, as explained in the comment for workqueue.c:insert_wq_barrier. Signed-off-by: Weston Andros Adamson [Trond: add module_get/put.] Signed-off-by: Trond Myklebust --- fs/nfs/inode.c | 5 ++++- fs/nfs/internal.h | 1 + fs/nfs/nfs4proc.c | 2 +- fs/nfs/super.c | 49 +++++++++++++++++++++++++++++++++++++++++++++++++ fs/nfs/unlink.c | 2 +- 5 files changed, 56 insertions(+), 3 deletions(-) (limited to 'fs/nfs/nfs4proc.c') diff --git a/fs/nfs/inode.c b/fs/nfs/inode.c index 5c7325c5c5e6..6fa01aea2488 100644 --- a/fs/nfs/inode.c +++ b/fs/nfs/inode.c @@ -685,7 +685,10 @@ static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync) if (ctx->cred != NULL) put_rpccred(ctx->cred); dput(ctx->dentry); - nfs_sb_deactive(sb); + if (is_sync) + nfs_sb_deactive(sb); + else + nfs_sb_deactive_async(sb); kfree(ctx->mdsthreshold); kfree(ctx); } diff --git a/fs/nfs/internal.h b/fs/nfs/internal.h index a54fe51c1dfb..05521cadac2e 100644 --- a/fs/nfs/internal.h +++ b/fs/nfs/internal.h @@ -351,6 +351,7 @@ extern int __init register_nfs_fs(void); extern void __exit unregister_nfs_fs(void); extern void nfs_sb_active(struct super_block *sb); extern void nfs_sb_deactive(struct super_block *sb); +extern void nfs_sb_deactive_async(struct super_block *sb); /* namespace.c */ #define NFS_PATH_CANONICAL 1 diff --git a/fs/nfs/nfs4proc.c b/fs/nfs/nfs4proc.c index 1465364501ba..8cfbac1a8d5e 100644 --- a/fs/nfs/nfs4proc.c +++ b/fs/nfs/nfs4proc.c @@ -2197,7 +2197,7 @@ static void nfs4_free_closedata(void *data) nfs4_put_open_state(calldata->state); nfs_free_seqid(calldata->arg.seqid); nfs4_put_state_owner(sp); - nfs_sb_deactive(sb); + nfs_sb_deactive_async(sb); kfree(calldata); } diff --git a/fs/nfs/super.c b/fs/nfs/super.c index 13c2a5be4765..652d3f7176a9 100644 --- a/fs/nfs/super.c +++ b/fs/nfs/super.c @@ -54,6 +54,7 @@ #include #include #include +#include #include @@ -415,6 +416,54 @@ void nfs_sb_deactive(struct super_block *sb) } EXPORT_SYMBOL_GPL(nfs_sb_deactive); +static int nfs_deactivate_super_async_work(void *ptr) +{ + struct super_block *sb = ptr; + + deactivate_super(sb); + module_put_and_exit(0); + return 0; +} + +/* + * same effect as deactivate_super, but will do final unmount in kthread + * context + */ +static void nfs_deactivate_super_async(struct super_block *sb) +{ + struct task_struct *task; + char buf[INET6_ADDRSTRLEN + 1]; + struct nfs_server *server = NFS_SB(sb); + struct nfs_client *clp = server->nfs_client; + + if (!atomic_add_unless(&sb->s_active, -1, 1)) { + rcu_read_lock(); + snprintf(buf, sizeof(buf), + rpc_peeraddr2str(clp->cl_rpcclient, RPC_DISPLAY_ADDR)); + rcu_read_unlock(); + + __module_get(THIS_MODULE); + task = kthread_run(nfs_deactivate_super_async_work, sb, + "%s-deactivate-super", buf); + if (IS_ERR(task)) { + pr_err("%s: kthread_run: %ld\n", + __func__, PTR_ERR(task)); + /* make synchronous call and hope for the best */ + deactivate_super(sb); + module_put(THIS_MODULE); + } + } +} + +void nfs_sb_deactive_async(struct super_block *sb) +{ + struct nfs_server *server = NFS_SB(sb); + + if (atomic_dec_and_test(&server->active)) + nfs_deactivate_super_async(sb); +} +EXPORT_SYMBOL_GPL(nfs_sb_deactive_async); + /* * Deliver file system statistics to userspace */ diff --git a/fs/nfs/unlink.c b/fs/nfs/unlink.c index 13cea637eff8..3f79c77153b8 100644 --- a/fs/nfs/unlink.c +++ b/fs/nfs/unlink.c @@ -95,7 +95,7 @@ static void nfs_async_unlink_release(void *calldata) nfs_dec_sillycount(data->dir); nfs_free_unlinkdata(data); - nfs_sb_deactive(sb); + nfs_sb_deactive_async(sb); } static void nfs_unlink_prepare(struct rpc_task *task, void *calldata) -- cgit v1.2.3 From f9b1ef5f06d65a01952169b67d474f7f0dcb0206 Mon Sep 17 00:00:00 2001 From: Trond Myklebust Date: Mon, 29 Oct 2012 16:48:40 -0400 Subject: NFSv4: Initialise the NFSv4.1 slot table highest_used_slotid correctly Signed-off-by: Trond Myklebust --- fs/nfs/nfs4proc.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'fs/nfs/nfs4proc.c') diff --git a/fs/nfs/nfs4proc.c b/fs/nfs/nfs4proc.c index 8cfbac1a8d5e..091baab3eccf 100644 --- a/fs/nfs/nfs4proc.c +++ b/fs/nfs/nfs4proc.c @@ -5677,7 +5677,7 @@ static void nfs4_add_and_init_slots(struct nfs4_slot_table *tbl, tbl->slots = new; tbl->max_slots = max_slots; } - tbl->highest_used_slotid = -1; /* no slot is currently used */ + tbl->highest_used_slotid = NFS4_NO_SLOT; for (i = 0; i < tbl->max_slots; i++) tbl->slots[i].seq_nr = ivalue; spin_unlock(&tbl->slot_tbl_lock); -- cgit v1.2.3 From 998f40b550f257e436485291802fa938e4cf580f Mon Sep 17 00:00:00 2001 From: Weston Andros Adamson Date: Fri, 2 Nov 2012 18:00:56 -0400 Subject: NFS4: nfs4_opendata_access should return errno Return errno - not an NFS4ERR_. This worked because NFS4ERR_ACCESS == EACCES. Signed-off-by: Weston Andros Adamson Signed-off-by: Trond Myklebust --- fs/nfs/nfs4proc.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'fs/nfs/nfs4proc.c') diff --git a/fs/nfs/nfs4proc.c b/fs/nfs/nfs4proc.c index 091baab3eccf..5eec4429970c 100644 --- a/fs/nfs/nfs4proc.c +++ b/fs/nfs/nfs4proc.c @@ -1749,7 +1749,7 @@ static int nfs4_opendata_access(struct rpc_cred *cred, /* even though OPEN succeeded, access is denied. Close the file */ nfs4_close_state(state, fmode); - return -NFS4ERR_ACCESS; + return -EACCES; } /* -- cgit v1.2.3