From 4a35bd41aff5714deb41c8f14766df3871e2e8f7 Mon Sep 17 00:00:00 2001 From: Trond Myklebust Date: Tue, 5 Jun 2007 10:31:33 -0400 Subject: NFSv4: Ensure that nfs4_do_close() doesn't race with umount nfs4_do_close() does not currently have any way to ensure that the user won't attempt to unmount the partition while the asynchronous RPC call is completing. This again may cause Oopses in nfs_update_inode(). Add a vfsmount argument to nfs4_close_state to ensure that the partition remains mounted while we're closing the file. Signed-off-by: Trond Myklebust --- fs/nfs/nfs4state.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'fs/nfs/nfs4state.c') diff --git a/fs/nfs/nfs4state.c b/fs/nfs/nfs4state.c index 8ed79d5c54f9..a85138ef67ad 100644 --- a/fs/nfs/nfs4state.c +++ b/fs/nfs/nfs4state.c @@ -341,7 +341,7 @@ void nfs4_put_open_state(struct nfs4_state *state) /* * Close the current file. */ -void nfs4_close_state(struct nfs4_state *state, mode_t mode) +void nfs4_close_state(struct path *path, struct nfs4_state *state, mode_t mode) { struct inode *inode = state->inode; struct nfs4_state_owner *owner = state->owner; @@ -375,7 +375,7 @@ void nfs4_close_state(struct nfs4_state *state, mode_t mode) spin_unlock(&inode->i_lock); spin_unlock(&owner->so_lock); - if (oldstate != newstate && nfs4_do_close(inode, state) == 0) + if (oldstate != newstate && nfs4_do_close(path, state) == 0) return; nfs4_put_open_state(state); nfs4_put_state_owner(owner); -- cgit v1.2.3 From b39e625b6e75aa70e26c13f9378756bb5f2af032 Mon Sep 17 00:00:00 2001 From: Trond Myklebust Date: Mon, 11 Jun 2007 23:05:07 -0400 Subject: NFSv4: Clean up nfs4_call_async() Use rpc_run_task() instead of doing it ourselves. Signed-off-by: Trond Myklebust --- fs/nfs/nfs4proc.c | 26 +++++++++----------------- fs/nfs/nfs4state.c | 9 +++++---- net/sunrpc/sunrpc_syms.c | 1 - 3 files changed, 14 insertions(+), 22 deletions(-) (limited to 'fs/nfs/nfs4state.c') diff --git a/fs/nfs/nfs4proc.c b/fs/nfs/nfs4proc.c index 8feaf232f2e4..3cc75445a68d 100644 --- a/fs/nfs/nfs4proc.c +++ b/fs/nfs/nfs4proc.c @@ -295,18 +295,6 @@ static void nfs4_opendata_free(struct nfs4_opendata *p) } } -/* Helper for asynchronous RPC calls */ -static int nfs4_call_async(struct rpc_clnt *clnt, - const struct rpc_call_ops *tk_ops, void *calldata) -{ - struct rpc_task *task; - - if (!(task = rpc_new_task(clnt, RPC_TASK_ASYNC, tk_ops, calldata))) - return -ENOMEM; - rpc_execute(task); - return 0; -} - static int nfs4_wait_for_completion_rpc_task(struct rpc_task *task) { sigset_t oldset; @@ -1218,6 +1206,8 @@ int nfs4_do_close(struct path *path, struct nfs4_state *state) { struct nfs_server *server = NFS_SERVER(state->inode); struct nfs4_closedata *calldata; + struct nfs4_state_owner *sp = state->owner; + struct rpc_task *task; int status = -ENOMEM; calldata = kmalloc(sizeof(*calldata), GFP_KERNEL); @@ -1237,14 +1227,16 @@ int nfs4_do_close(struct path *path, struct nfs4_state *state) calldata->path.mnt = mntget(path->mnt); calldata->path.dentry = dget(path->dentry); - status = nfs4_call_async(server->client, &nfs4_close_ops, calldata); - if (status == 0) - goto out; - - nfs_free_seqid(calldata->arg.seqid); + task = rpc_run_task(server->client, RPC_TASK_ASYNC, &nfs4_close_ops, calldata); + if (IS_ERR(task)) + return PTR_ERR(task); + rpc_put_task(task); + return 0; out_free_calldata: kfree(calldata); out: + nfs4_put_open_state(state); + nfs4_put_state_owner(sp); return status; } diff --git a/fs/nfs/nfs4state.c b/fs/nfs/nfs4state.c index a85138ef67ad..5d7ffbfc3483 100644 --- a/fs/nfs/nfs4state.c +++ b/fs/nfs/nfs4state.c @@ -375,10 +375,11 @@ void nfs4_close_state(struct path *path, struct nfs4_state *state, mode_t mode) spin_unlock(&inode->i_lock); spin_unlock(&owner->so_lock); - if (oldstate != newstate && nfs4_do_close(path, state) == 0) - return; - nfs4_put_open_state(state); - nfs4_put_state_owner(owner); + if (oldstate == newstate) { + nfs4_put_open_state(state); + nfs4_put_state_owner(owner); + } else + nfs4_do_close(path, state); } /* diff --git a/net/sunrpc/sunrpc_syms.c b/net/sunrpc/sunrpc_syms.c index 73075dec83c0..c46d31ca307b 100644 --- a/net/sunrpc/sunrpc_syms.c +++ b/net/sunrpc/sunrpc_syms.c @@ -30,7 +30,6 @@ EXPORT_SYMBOL(rpc_wake_up_next); EXPORT_SYMBOL(rpc_wake_up_task); EXPORT_SYMBOL(rpciod_down); EXPORT_SYMBOL(rpciod_up); -EXPORT_SYMBOL(rpc_new_task); EXPORT_SYMBOL(rpc_wake_up_status); /* RPC client functions */ -- cgit v1.2.3 From 27b3f949b769a208e2849d28e7ad64cadac5d0e3 Mon Sep 17 00:00:00 2001 From: Trond Myklebust Date: Mon, 2 Jul 2007 10:24:56 -0400 Subject: NFSv4: Fix a credential reference leak in nfs4_get_state_owner() Signed-off-by: Trond Myklebust --- fs/nfs/nfs4state.c | 6 ++---- 1 file changed, 2 insertions(+), 4 deletions(-) (limited to 'fs/nfs/nfs4state.c') diff --git a/fs/nfs/nfs4state.c b/fs/nfs/nfs4state.c index 5d7ffbfc3483..0030248d63ec 100644 --- a/fs/nfs/nfs4state.c +++ b/fs/nfs/nfs4state.c @@ -83,7 +83,7 @@ nfs4_client_grab_unused(struct nfs_client *clp, struct rpc_cred *cred) if (!list_empty(&clp->cl_unused)) { sp = list_entry(clp->cl_unused.next, struct nfs4_state_owner, so_list); atomic_inc(&sp->so_count); - sp->so_cred = cred; + sp->so_cred = get_rpccred(cred); list_move(&sp->so_list, &clp->cl_state_owners); clp->cl_nunused--; } @@ -175,7 +175,6 @@ struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server, struct struct nfs_client *clp = server->nfs_client; struct nfs4_state_owner *sp, *new; - get_rpccred(cred); new = nfs4_alloc_state_owner(); spin_lock(&clp->cl_lock); sp = nfs4_find_state_owner(clp, cred); @@ -185,7 +184,7 @@ struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server, struct list_add(&new->so_list, &clp->cl_state_owners); new->so_client = clp; new->so_id = nfs4_alloc_lockowner_id(clp); - new->so_cred = cred; + new->so_cred = get_rpccred(cred); sp = new; new = NULL; } @@ -193,7 +192,6 @@ struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server, struct kfree(new); if (sp != NULL) return sp; - put_rpccred(cred); return NULL; } -- cgit v1.2.3 From 7af654f8d1b7460415af5d1d326233478dd0f563 Mon Sep 17 00:00:00 2001 From: Trond Myklebust Date: Mon, 2 Jul 2007 12:49:23 -0400 Subject: NFSv4: Don't reuse expired nfs4_state_owner structs That just confuses certain NFSv4 servers. Signed-off-by: Trond Myklebust --- fs/nfs/client.c | 10 ---------- fs/nfs/nfs4state.c | 28 ---------------------------- include/linux/nfs_fs_sb.h | 2 -- 3 files changed, 40 deletions(-) (limited to 'fs/nfs/nfs4state.c') diff --git a/fs/nfs/client.c b/fs/nfs/client.c index 71d4c4cdac52..6b424407d631 100644 --- a/fs/nfs/client.c +++ b/fs/nfs/client.c @@ -131,7 +131,6 @@ static struct nfs_client *nfs_alloc_client(const char *hostname, init_rwsem(&clp->cl_sem); INIT_LIST_HEAD(&clp->cl_delegations); INIT_LIST_HEAD(&clp->cl_state_owners); - INIT_LIST_HEAD(&clp->cl_unused); spin_lock_init(&clp->cl_lock); INIT_DELAYED_WORK(&clp->cl_renewd, nfs4_renew_state); rpc_init_wait_queue(&clp->cl_rpcwaitq, "NFS client"); @@ -155,15 +154,6 @@ static void nfs4_shutdown_client(struct nfs_client *clp) #ifdef CONFIG_NFS_V4 if (__test_and_clear_bit(NFS_CS_RENEWD, &clp->cl_res_state)) nfs4_kill_renewd(clp); - while (!list_empty(&clp->cl_unused)) { - struct nfs4_state_owner *sp; - - sp = list_entry(clp->cl_unused.next, - struct nfs4_state_owner, - so_list); - list_del(&sp->so_list); - kfree(sp); - } BUG_ON(!list_empty(&clp->cl_state_owners)); if (__test_and_clear_bit(NFS_CS_IDMAP, &clp->cl_res_state)) nfs_idmap_delete(clp); diff --git a/fs/nfs/nfs4state.c b/fs/nfs/nfs4state.c index 0030248d63ec..2b00c45aebea 100644 --- a/fs/nfs/nfs4state.c +++ b/fs/nfs/nfs4state.c @@ -75,21 +75,6 @@ nfs4_alloc_lockowner_id(struct nfs_client *clp) return clp->cl_lockowner_id ++; } -static struct nfs4_state_owner * -nfs4_client_grab_unused(struct nfs_client *clp, struct rpc_cred *cred) -{ - struct nfs4_state_owner *sp = NULL; - - if (!list_empty(&clp->cl_unused)) { - sp = list_entry(clp->cl_unused.next, struct nfs4_state_owner, so_list); - atomic_inc(&sp->so_count); - sp->so_cred = get_rpccred(cred); - list_move(&sp->so_list, &clp->cl_state_owners); - clp->cl_nunused--; - } - return sp; -} - struct rpc_cred *nfs4_get_renew_cred(struct nfs_client *clp) { struct nfs4_state_owner *sp; @@ -178,8 +163,6 @@ struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server, struct new = nfs4_alloc_state_owner(); spin_lock(&clp->cl_lock); sp = nfs4_find_state_owner(clp, cred); - if (sp == NULL) - sp = nfs4_client_grab_unused(clp, cred); if (sp == NULL && new != NULL) { list_add(&new->so_list, &clp->cl_state_owners); new->so_client = clp; @@ -206,17 +189,6 @@ void nfs4_put_state_owner(struct nfs4_state_owner *sp) if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock)) return; - if (clp->cl_nunused >= OPENOWNER_POOL_SIZE) - goto out_free; - if (list_empty(&sp->so_list)) - goto out_free; - list_move(&sp->so_list, &clp->cl_unused); - clp->cl_nunused++; - spin_unlock(&clp->cl_lock); - put_rpccred(cred); - cred = NULL; - return; -out_free: list_del(&sp->so_list); spin_unlock(&clp->cl_lock); put_rpccred(cred); diff --git a/include/linux/nfs_fs_sb.h b/include/linux/nfs_fs_sb.h index 144d955dc46a..2cef0a68aa77 100644 --- a/include/linux/nfs_fs_sb.h +++ b/include/linux/nfs_fs_sb.h @@ -44,8 +44,6 @@ struct nfs_client { struct list_head cl_delegations; struct list_head cl_state_owners; - struct list_head cl_unused; - int cl_nunused; spinlock_t cl_lock; unsigned long cl_lease_time; -- cgit v1.2.3 From 88d9093997e1c73ca98db41b5605dbde7783845f Mon Sep 17 00:00:00 2001 From: Trond Myklebust Date: Mon, 2 Jul 2007 14:03:03 -0400 Subject: NFSv4: nfs_increment_open_seqid should not return a value It is a void function... Signed-off-by: Trond Myklebust --- fs/nfs/nfs4state.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) (limited to 'fs/nfs/nfs4state.c') diff --git a/fs/nfs/nfs4state.c b/fs/nfs/nfs4state.c index 2b00c45aebea..0f79d56e97f0 100644 --- a/fs/nfs/nfs4state.c +++ b/fs/nfs/nfs4state.c @@ -528,7 +528,7 @@ void nfs_free_seqid(struct nfs_seqid *seqid) * failed with a seqid incrementing error - * see comments nfs_fs.h:seqid_mutating_error() */ -static inline void nfs_increment_seqid(int status, struct nfs_seqid *seqid) +static void nfs_increment_seqid(int status, struct nfs_seqid *seqid) { switch (status) { case 0: @@ -557,7 +557,7 @@ void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid) struct nfs4_state_owner, so_seqid); nfs4_drop_state_owner(sp); } - return nfs_increment_seqid(status, seqid); + nfs_increment_seqid(status, seqid); } /* @@ -567,7 +567,7 @@ void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid) */ void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid) { - return nfs_increment_seqid(status, seqid); + nfs_increment_seqid(status, seqid); } int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task) -- cgit v1.2.3 From 9f958ab8858c75df800e0121b1920182820cbc39 Mon Sep 17 00:00:00 2001 From: Trond Myklebust Date: Mon, 2 Jul 2007 13:58:33 -0400 Subject: NFSv4: Reduce the chances of an open_owner identifier collision Currently we just use a 32-bit counter. Signed-off-by: Trond Myklebust --- fs/nfs/client.c | 3 +- fs/nfs/nfs4_fs.h | 17 ++-- fs/nfs/nfs4proc.c | 8 +- fs/nfs/nfs4state.c | 193 ++++++++++++++++++++++++++++++++++++---------- fs/nfs/nfs4xdr.c | 27 ++++--- include/linux/nfs_fs_sb.h | 5 +- include/linux/nfs_xdr.h | 4 +- 7 files changed, 190 insertions(+), 67 deletions(-) (limited to 'fs/nfs/nfs4state.c') diff --git a/fs/nfs/client.c b/fs/nfs/client.c index 6b424407d631..ccb455053ee4 100644 --- a/fs/nfs/client.c +++ b/fs/nfs/client.c @@ -130,7 +130,6 @@ static struct nfs_client *nfs_alloc_client(const char *hostname, #ifdef CONFIG_NFS_V4 init_rwsem(&clp->cl_sem); INIT_LIST_HEAD(&clp->cl_delegations); - INIT_LIST_HEAD(&clp->cl_state_owners); spin_lock_init(&clp->cl_lock); INIT_DELAYED_WORK(&clp->cl_renewd, nfs4_renew_state); rpc_init_wait_queue(&clp->cl_rpcwaitq, "NFS client"); @@ -154,7 +153,7 @@ static void nfs4_shutdown_client(struct nfs_client *clp) #ifdef CONFIG_NFS_V4 if (__test_and_clear_bit(NFS_CS_RENEWD, &clp->cl_res_state)) nfs4_kill_renewd(clp); - BUG_ON(!list_empty(&clp->cl_state_owners)); + BUG_ON(!RB_EMPTY_ROOT(&clp->cl_state_owners)); if (__test_and_clear_bit(NFS_CS_IDMAP, &clp->cl_res_state)) nfs_idmap_delete(clp); #endif diff --git a/fs/nfs/nfs4_fs.h b/fs/nfs/nfs4_fs.h index c97a0ad8430e..44b56c915f72 100644 --- a/fs/nfs/nfs4_fs.h +++ b/fs/nfs/nfs4_fs.h @@ -70,19 +70,25 @@ static inline void nfs_confirm_seqid(struct nfs_seqid_counter *seqid, int status seqid->flags |= NFS_SEQID_CONFIRMED; } +struct nfs_unique_id { + struct rb_node rb_node; + __u64 id; +}; + /* * NFS4 state_owners and lock_owners are simply labels for ordered * sequences of RPC calls. Their sole purpose is to provide once-only * semantics by allowing the server to identify replayed requests. */ struct nfs4_state_owner { - spinlock_t so_lock; - struct list_head so_list; /* per-clientid list of state_owners */ + struct nfs_unique_id so_owner_id; struct nfs_client *so_client; - u32 so_id; /* 32-bit identifier, unique */ - atomic_t so_count; + struct rb_node so_client_node; struct rpc_cred *so_cred; /* Associated cred */ + + spinlock_t so_lock; + atomic_t so_count; struct list_head so_states; struct list_head so_delegations; struct nfs_seqid_counter so_seqid; @@ -108,7 +114,7 @@ struct nfs4_lock_state { #define NFS_LOCK_INITIALIZED 1 int ls_flags; struct nfs_seqid_counter ls_seqid; - u32 ls_id; + struct nfs_unique_id ls_id; nfs4_stateid ls_stateid; atomic_t ls_count; }; @@ -189,7 +195,6 @@ extern void nfs4_renew_state(struct work_struct *); /* nfs4state.c */ struct rpc_cred *nfs4_get_renew_cred(struct nfs_client *clp); -extern u32 nfs4_alloc_lockowner_id(struct nfs_client *); extern struct nfs4_state_owner * nfs4_get_state_owner(struct nfs_server *, struct rpc_cred *); extern void nfs4_put_state_owner(struct nfs4_state_owner *); diff --git a/fs/nfs/nfs4proc.c b/fs/nfs/nfs4proc.c index 84d0b7e0dd67..1840ebc78fd3 100644 --- a/fs/nfs/nfs4proc.c +++ b/fs/nfs/nfs4proc.c @@ -253,7 +253,7 @@ static struct nfs4_opendata *nfs4_opendata_alloc(struct path *path, p->o_arg.fh = NFS_FH(dir); p->o_arg.open_flags = flags, p->o_arg.clientid = server->nfs_client->cl_clientid; - p->o_arg.id = sp->so_id; + p->o_arg.id = sp->so_owner_id.id; p->o_arg.name = &p->path.dentry->d_name; p->o_arg.server = server; p->o_arg.bitmask = server->attr_bitmask; @@ -651,7 +651,7 @@ static void nfs4_open_prepare(struct rpc_task *task, void *calldata) if (nfs_wait_on_sequence(data->o_arg.seqid, task) != 0) return; /* Update sequence id. */ - data->o_arg.id = sp->so_id; + data->o_arg.id = sp->so_owner_id.id; data->o_arg.clientid = sp->so_client->cl_clientid; if (data->o_arg.claim == NFS4_OPEN_CLAIM_PREVIOUS) msg.rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_OPEN_NOATTR]; @@ -3029,7 +3029,7 @@ static int _nfs4_proc_getlk(struct nfs4_state *state, int cmd, struct file_lock if (status != 0) goto out; lsp = request->fl_u.nfs4_fl.owner; - arg.lock_owner.id = lsp->ls_id; + arg.lock_owner.id = lsp->ls_id.id; status = rpc_call_sync(server->client, &msg, 0); switch (status) { case 0: @@ -3243,7 +3243,7 @@ static struct nfs4_lockdata *nfs4_alloc_lockdata(struct file_lock *fl, goto out_free; p->arg.lock_stateid = &lsp->ls_stateid; p->arg.lock_owner.clientid = server->nfs_client->cl_clientid; - p->arg.lock_owner.id = lsp->ls_id; + p->arg.lock_owner.id = lsp->ls_id.id; p->lsp = lsp; atomic_inc(&lsp->ls_count); p->ctx = get_nfs_open_context(ctx); diff --git a/fs/nfs/nfs4state.c b/fs/nfs/nfs4state.c index 0f79d56e97f0..ab0b5ab60e60 100644 --- a/fs/nfs/nfs4state.c +++ b/fs/nfs/nfs4state.c @@ -44,6 +44,7 @@ #include #include #include +#include #include #include @@ -69,18 +70,14 @@ static int nfs4_init_client(struct nfs_client *clp, struct rpc_cred *cred) return status; } -u32 -nfs4_alloc_lockowner_id(struct nfs_client *clp) -{ - return clp->cl_lockowner_id ++; -} - struct rpc_cred *nfs4_get_renew_cred(struct nfs_client *clp) { struct nfs4_state_owner *sp; + struct rb_node *pos; struct rpc_cred *cred = NULL; - list_for_each_entry(sp, &clp->cl_state_owners, so_list) { + for (pos = rb_first(&clp->cl_state_owners); pos != NULL; pos = rb_next(pos)) { + sp = rb_entry(pos, struct nfs4_state_owner, so_client_node); if (list_empty(&sp->so_states)) continue; cred = get_rpccred(sp->so_cred); @@ -92,32 +89,129 @@ struct rpc_cred *nfs4_get_renew_cred(struct nfs_client *clp) static struct rpc_cred *nfs4_get_setclientid_cred(struct nfs_client *clp) { struct nfs4_state_owner *sp; + struct rb_node *pos; - if (!list_empty(&clp->cl_state_owners)) { - sp = list_entry(clp->cl_state_owners.next, - struct nfs4_state_owner, so_list); + pos = rb_first(&clp->cl_state_owners); + if (pos != NULL) { + sp = rb_entry(pos, struct nfs4_state_owner, so_client_node); return get_rpccred(sp->so_cred); } return NULL; } +static void nfs_alloc_unique_id(struct rb_root *root, struct nfs_unique_id *new, + __u64 minval, int maxbits) +{ + struct rb_node **p, *parent; + struct nfs_unique_id *pos; + __u64 mask = ~0ULL; + + if (maxbits < 64) + mask = (1ULL << maxbits) - 1ULL; + + /* Ensure distribution is more or less flat */ + get_random_bytes(&new->id, sizeof(new->id)); + new->id &= mask; + if (new->id < minval) + new->id += minval; +retry: + p = &root->rb_node; + parent = NULL; + + while (*p != NULL) { + parent = *p; + pos = rb_entry(parent, struct nfs_unique_id, rb_node); + + if (new->id < pos->id) + p = &(*p)->rb_left; + else if (new->id > pos->id) + p = &(*p)->rb_right; + else + goto id_exists; + } + rb_link_node(&new->rb_node, parent, p); + rb_insert_color(&new->rb_node, root); + return; +id_exists: + for (;;) { + new->id++; + if (new->id < minval || (new->id & mask) != new->id) { + new->id = minval; + break; + } + parent = rb_next(parent); + if (parent == NULL) + break; + pos = rb_entry(parent, struct nfs_unique_id, rb_node); + if (new->id < pos->id) + break; + } + goto retry; +} + +static void nfs_free_unique_id(struct rb_root *root, struct nfs_unique_id *id) +{ + rb_erase(&id->rb_node, root); +} + static struct nfs4_state_owner * nfs4_find_state_owner(struct nfs_client *clp, struct rpc_cred *cred) { + struct rb_node **p = &clp->cl_state_owners.rb_node, + *parent = NULL; struct nfs4_state_owner *sp, *res = NULL; - list_for_each_entry(sp, &clp->cl_state_owners, so_list) { - if (sp->so_cred != cred) - continue; - atomic_inc(&sp->so_count); - /* Move to the head of the list */ - list_move(&sp->so_list, &clp->cl_state_owners); - res = sp; - break; + while (*p != NULL) { + parent = *p; + sp = rb_entry(parent, struct nfs4_state_owner, so_client_node); + + if (cred < sp->so_cred) + p = &parent->rb_left; + else if (cred > sp->so_cred) + p = &parent->rb_right; + else { + atomic_inc(&sp->so_count); + res = sp; + break; + } } return res; } +static struct nfs4_state_owner * +nfs4_insert_state_owner(struct nfs_client *clp, struct nfs4_state_owner *new) +{ + struct rb_node **p = &clp->cl_state_owners.rb_node, + *parent = NULL; + struct nfs4_state_owner *sp; + + while (*p != NULL) { + parent = *p; + sp = rb_entry(parent, struct nfs4_state_owner, so_client_node); + + if (new->so_cred < sp->so_cred) + p = &parent->rb_left; + else if (new->so_cred > sp->so_cred) + p = &parent->rb_right; + else { + atomic_inc(&sp->so_count); + return sp; + } + } + nfs_alloc_unique_id(&clp->cl_openowner_id, &new->so_owner_id, 1, 64); + rb_link_node(&new->so_client_node, parent, p); + rb_insert_color(&new->so_client_node, &clp->cl_state_owners); + return new; +} + +static void +nfs4_remove_state_owner(struct nfs_client *clp, struct nfs4_state_owner *sp) +{ + if (!RB_EMPTY_NODE(&sp->so_client_node)) + rb_erase(&sp->so_client_node, &clp->cl_state_owners); + nfs_free_unique_id(&clp->cl_openowner_id, &sp->so_owner_id); +} + /* * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to * create a new state_owner. @@ -145,10 +239,14 @@ nfs4_alloc_state_owner(void) void nfs4_drop_state_owner(struct nfs4_state_owner *sp) { - struct nfs_client *clp = sp->so_client; - spin_lock(&clp->cl_lock); - list_del_init(&sp->so_list); - spin_unlock(&clp->cl_lock); + if (!RB_EMPTY_NODE(&sp->so_client_node)) { + struct nfs_client *clp = sp->so_client; + + spin_lock(&clp->cl_lock); + rb_erase(&sp->so_client_node, &clp->cl_state_owners); + RB_CLEAR_NODE(&sp->so_client_node); + spin_unlock(&clp->cl_lock); + } } /* @@ -160,22 +258,24 @@ struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server, struct struct nfs_client *clp = server->nfs_client; struct nfs4_state_owner *sp, *new; - new = nfs4_alloc_state_owner(); spin_lock(&clp->cl_lock); sp = nfs4_find_state_owner(clp, cred); - if (sp == NULL && new != NULL) { - list_add(&new->so_list, &clp->cl_state_owners); - new->so_client = clp; - new->so_id = nfs4_alloc_lockowner_id(clp); - new->so_cred = get_rpccred(cred); - sp = new; - new = NULL; - } spin_unlock(&clp->cl_lock); - kfree(new); if (sp != NULL) return sp; - return NULL; + new = nfs4_alloc_state_owner(); + if (new == NULL) + return NULL; + new->so_client = clp; + new->so_cred = cred; + spin_lock(&clp->cl_lock); + sp = nfs4_insert_state_owner(clp, new); + spin_unlock(&clp->cl_lock); + if (sp == new) + get_rpccred(cred); + else + kfree(new); + return sp; } /* @@ -189,7 +289,7 @@ void nfs4_put_state_owner(struct nfs4_state_owner *sp) if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock)) return; - list_del(&sp->so_list); + nfs4_remove_state_owner(clp, sp); spin_unlock(&clp->cl_lock); put_rpccred(cred); kfree(sp); @@ -386,12 +486,22 @@ static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, f atomic_set(&lsp->ls_count, 1); lsp->ls_owner = fl_owner; spin_lock(&clp->cl_lock); - lsp->ls_id = nfs4_alloc_lockowner_id(clp); + nfs_alloc_unique_id(&clp->cl_lockowner_id, &lsp->ls_id, 1, 64); spin_unlock(&clp->cl_lock); INIT_LIST_HEAD(&lsp->ls_locks); return lsp; } +static void nfs4_free_lock_state(struct nfs4_lock_state *lsp) +{ + struct nfs_client *clp = lsp->ls_state->owner->so_client; + + spin_lock(&clp->cl_lock); + nfs_free_unique_id(&clp->cl_lockowner_id, &lsp->ls_id); + spin_unlock(&clp->cl_lock); + kfree(lsp); +} + /* * Return a compatible lock_state. If no initialized lock_state structure * exists, return an uninitialized one. @@ -421,7 +531,8 @@ static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_ return NULL; } spin_unlock(&state->state_lock); - kfree(new); + if (new != NULL) + nfs4_free_lock_state(new); return lsp; } @@ -442,7 +553,7 @@ void nfs4_put_lock_state(struct nfs4_lock_state *lsp) if (list_empty(&state->lock_states)) clear_bit(LK_STATE_IN_USE, &state->flags); spin_unlock(&state->state_lock); - kfree(lsp); + nfs4_free_lock_state(lsp); } static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src) @@ -719,11 +830,13 @@ out_err: static void nfs4_state_mark_reclaim(struct nfs_client *clp) { struct nfs4_state_owner *sp; + struct rb_node *pos; struct nfs4_state *state; struct nfs4_lock_state *lock; /* Reset all sequence ids to zero */ - list_for_each_entry(sp, &clp->cl_state_owners, so_list) { + for (pos = rb_first(&clp->cl_state_owners); pos != NULL; pos = rb_next(pos)) { + sp = rb_entry(pos, struct nfs4_state_owner, so_client_node); sp->so_seqid.counter = 0; sp->so_seqid.flags = 0; spin_lock(&sp->so_lock); @@ -742,6 +855,7 @@ static int reclaimer(void *ptr) { struct nfs_client *clp = ptr; struct nfs4_state_owner *sp; + struct rb_node *pos; struct nfs4_state_recovery_ops *ops; struct rpc_cred *cred; int status = 0; @@ -787,7 +901,8 @@ restart_loop: /* Mark all delegations for reclaim */ nfs_delegation_mark_reclaim(clp); /* Note: list is protected by exclusive lock on cl->cl_sem */ - list_for_each_entry(sp, &clp->cl_state_owners, so_list) { + for (pos = rb_first(&clp->cl_state_owners); pos != NULL; pos = rb_next(pos)) { + sp = rb_entry(pos, struct nfs4_state_owner, so_client_node); status = nfs4_reclaim_open_state(ops, sp); if (status < 0) { if (status == -NFS4ERR_NO_GRACE) { diff --git a/fs/nfs/nfs4xdr.c b/fs/nfs/nfs4xdr.c index 4c8f67d47523..c08738441f73 100644 --- a/fs/nfs/nfs4xdr.c +++ b/fs/nfs/nfs4xdr.c @@ -68,10 +68,10 @@ static int nfs4_stat_to_errno(int); #endif /* lock,open owner id: - * we currently use size 1 (u32) out of (NFS4_OPAQUE_LIMIT >> 2) + * we currently use size 2 (u64) out of (NFS4_OPAQUE_LIMIT >> 2) */ -#define open_owner_id_maxsz (1 + 1) -#define lock_owner_id_maxsz (1 + 1) +#define open_owner_id_maxsz (1 + 4) +#define lock_owner_id_maxsz (1 + 4) #define compound_encode_hdr_maxsz (3 + (NFS4_MAXTAGLEN >> 2)) #define compound_decode_hdr_maxsz (3 + (NFS4_MAXTAGLEN >> 2)) #define op_encode_hdr_maxsz (1) @@ -827,13 +827,14 @@ static int encode_lock(struct xdr_stream *xdr, const struct nfs_lock_args *args) WRITE64(nfs4_lock_length(args->fl)); WRITE32(args->new_lock_owner); if (args->new_lock_owner){ - RESERVE_SPACE(4+NFS4_STATEID_SIZE+20); + RESERVE_SPACE(4+NFS4_STATEID_SIZE+32); WRITE32(args->open_seqid->sequence->counter); WRITEMEM(args->open_stateid->data, NFS4_STATEID_SIZE); WRITE32(args->lock_seqid->sequence->counter); WRITE64(args->lock_owner.clientid); - WRITE32(4); - WRITE32(args->lock_owner.id); + WRITE32(16); + WRITEMEM("lock id:", 8); + WRITE64(args->lock_owner.id); } else { RESERVE_SPACE(NFS4_STATEID_SIZE+4); @@ -848,14 +849,15 @@ static int encode_lockt(struct xdr_stream *xdr, const struct nfs_lockt_args *arg { __be32 *p; - RESERVE_SPACE(40); + RESERVE_SPACE(52); WRITE32(OP_LOCKT); WRITE32(nfs4_lock_type(args->fl, 0)); WRITE64(args->fl->fl_start); WRITE64(nfs4_lock_length(args->fl)); WRITE64(args->lock_owner.clientid); - WRITE32(4); - WRITE32(args->lock_owner.id); + WRITE32(16); + WRITEMEM("lock id:", 8); + WRITE64(args->lock_owner.id); return 0; } @@ -920,10 +922,11 @@ static inline void encode_openhdr(struct xdr_stream *xdr, const struct nfs_opena WRITE32(OP_OPEN); WRITE32(arg->seqid->sequence->counter); encode_share_access(xdr, arg->open_flags); - RESERVE_SPACE(16); + RESERVE_SPACE(28); WRITE64(arg->clientid); - WRITE32(4); - WRITE32(arg->id); + WRITE32(16); + WRITEMEM("open id:", 8); + WRITE64(arg->id); } static inline void encode_createmode(struct xdr_stream *xdr, const struct nfs_openargs *arg) diff --git a/include/linux/nfs_fs_sb.h b/include/linux/nfs_fs_sb.h index 2cef0a68aa77..0cac49bc0955 100644 --- a/include/linux/nfs_fs_sb.h +++ b/include/linux/nfs_fs_sb.h @@ -34,7 +34,8 @@ struct nfs_client { nfs4_verifier cl_confirm; unsigned long cl_state; - u32 cl_lockowner_id; + struct rb_root cl_openowner_id; + struct rb_root cl_lockowner_id; /* * The following rwsem ensures exclusive access to the server @@ -43,7 +44,7 @@ struct nfs_client { struct rw_semaphore cl_sem; struct list_head cl_delegations; - struct list_head cl_state_owners; + struct rb_root cl_state_owners; spinlock_t cl_lock; unsigned long cl_lease_time; diff --git a/include/linux/nfs_xdr.h b/include/linux/nfs_xdr.h index f7100df3a690..38d77681cf27 100644 --- a/include/linux/nfs_xdr.h +++ b/include/linux/nfs_xdr.h @@ -119,7 +119,7 @@ struct nfs_openargs { struct nfs_seqid * seqid; int open_flags; __u64 clientid; - __u32 id; + __u64 id; union { struct iattr * attrs; /* UNCHECKED, GUARDED */ nfs4_verifier verifier; /* EXCLUSIVE */ @@ -181,7 +181,7 @@ struct nfs_closeres { * */ struct nfs_lowner { __u64 clientid; - u32 id; + __u64 id; }; struct nfs_lock_args { -- cgit v1.2.3 From 1c816efa245111c52858fbe55d99474f3c149dd3 Mon Sep 17 00:00:00 2001 From: Trond Myklebust Date: Tue, 3 Jul 2007 14:41:19 -0400 Subject: NFSv4: Fix a bug in __nfs4_find_state_byowner The test for state->state == 0 does not tell you that the stateid is in the process of being freed. It really tells you that the stateid is not yet initialised... Signed-off-by: Trond Myklebust --- fs/nfs/nfs4state.c | 7 ++----- 1 file changed, 2 insertions(+), 5 deletions(-) (limited to 'fs/nfs/nfs4state.c') diff --git a/fs/nfs/nfs4state.c b/fs/nfs/nfs4state.c index ab0b5ab60e60..ac816b303f3a 100644 --- a/fs/nfs/nfs4state.c +++ b/fs/nfs/nfs4state.c @@ -333,13 +333,10 @@ __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner) struct nfs4_state *state; list_for_each_entry(state, &nfsi->open_states, inode_states) { - /* Is this in the process of being freed? */ - if (state->state == 0) + if (state->owner != owner) continue; - if (state->owner == owner) { - atomic_inc(&state->count); + if (atomic_inc_not_zero(&state->count)) return state; - } } return NULL; } -- cgit v1.2.3 From 003707c7225dbd4bf879b6c204743554de0a08d6 Mon Sep 17 00:00:00 2001 From: Trond Myklebust Date: Thu, 5 Jul 2007 18:07:55 -0400 Subject: NFSv4: Always use the delegation if we have one Signed-off-by: Trond Myklebust --- fs/nfs/nfs4_fs.h | 14 +++++---- fs/nfs/nfs4proc.c | 88 ++++++++++++++++++++++++++++++++++++++++-------------- fs/nfs/nfs4state.c | 28 +++++++++++------ 3 files changed, 94 insertions(+), 36 deletions(-) (limited to 'fs/nfs/nfs4state.c') diff --git a/fs/nfs/nfs4_fs.h b/fs/nfs/nfs4_fs.h index 44b56c915f72..4a1c4d80a577 100644 --- a/fs/nfs/nfs4_fs.h +++ b/fs/nfs/nfs4_fs.h @@ -122,7 +122,10 @@ struct nfs4_lock_state { /* bits for nfs4_state->flags */ enum { LK_STATE_IN_USE, - NFS_DELEGATED_STATE, + NFS_DELEGATED_STATE, /* Current stateid is delegation */ + NFS_O_RDONLY_STATE, /* OPEN stateid has read-only state */ + NFS_O_WRONLY_STATE, /* OPEN stateid has write-only state */ + NFS_O_RDWR_STATE, /* OPEN stateid has read/write state */ }; struct nfs4_state { @@ -136,11 +139,12 @@ struct nfs4_state { unsigned long flags; /* Do we hold any locks? */ spinlock_t state_lock; /* Protects the lock_states list */ - nfs4_stateid stateid; + nfs4_stateid stateid; /* Current stateid: may be delegation */ + nfs4_stateid open_stateid; /* OPEN stateid */ - unsigned int n_rdonly; - unsigned int n_wronly; - unsigned int n_rdwr; + unsigned int n_rdonly; /* Number of read-only references */ + unsigned int n_wronly; /* Number of write-only references */ + unsigned int n_rdwr; /* Number of read/write references */ int state; /* State on the server (R,W, or RW) */ atomic_t count; }; diff --git a/fs/nfs/nfs4proc.c b/fs/nfs/nfs4proc.c index 61ba32af4d2f..128fe23d3f1e 100644 --- a/fs/nfs/nfs4proc.c +++ b/fs/nfs/nfs4proc.c @@ -319,7 +319,7 @@ static int nfs4_wait_for_completion_rpc_task(struct rpc_task *task) return ret; } -static inline void update_open_stateflags(struct nfs4_state *state, mode_t open_flags) +static void update_open_stateflags(struct nfs4_state *state, mode_t open_flags) { switch (open_flags) { case FMODE_WRITE: @@ -331,9 +331,36 @@ static inline void update_open_stateflags(struct nfs4_state *state, mode_t open_ case FMODE_READ|FMODE_WRITE: state->n_rdwr++; } + nfs4_state_set_mode_locked(state, state->state | open_flags); +} + +static void nfs_set_open_stateid_locked(struct nfs4_state *state, nfs4_stateid *stateid, int open_flags) +{ + if (test_bit(NFS_DELEGATED_STATE, &state->flags) == 0) + memcpy(state->stateid.data, stateid->data, sizeof(state->stateid.data)); + memcpy(state->open_stateid.data, stateid->data, sizeof(state->open_stateid.data)); + switch (open_flags) { + case FMODE_READ: + set_bit(NFS_O_RDONLY_STATE, &state->flags); + break; + case FMODE_WRITE: + set_bit(NFS_O_WRONLY_STATE, &state->flags); + break; + case FMODE_READ|FMODE_WRITE: + set_bit(NFS_O_RDWR_STATE, &state->flags); + } } -static void update_open_stateid(struct nfs4_state *state, nfs4_stateid *stateid, int open_flags) +static void nfs_set_open_stateid(struct nfs4_state *state, nfs4_stateid *stateid, int open_flags) +{ + spin_lock(&state->owner->so_lock); + spin_lock(&state->inode->i_lock); + nfs_set_open_stateid_locked(state, stateid, open_flags); + spin_unlock(&state->inode->i_lock); + spin_unlock(&state->owner->so_lock); +} + +static void update_open_stateid(struct nfs4_state *state, nfs4_stateid *open_stateid, nfs4_stateid *deleg_stateid, int open_flags) { struct inode *inode = state->inode; @@ -341,9 +368,13 @@ static void update_open_stateid(struct nfs4_state *state, nfs4_stateid *stateid, /* Protect against nfs4_find_state_byowner() */ spin_lock(&state->owner->so_lock); spin_lock(&inode->i_lock); - memcpy(&state->stateid, stateid, sizeof(state->stateid)); + if (deleg_stateid != NULL) { + memcpy(state->stateid.data, deleg_stateid->data, sizeof(state->stateid.data)); + set_bit(NFS_DELEGATED_STATE, &state->flags); + } + if (open_stateid != NULL) + nfs_set_open_stateid_locked(state, open_stateid, open_flags); update_open_stateflags(state, open_flags); - nfs4_state_set_mode_locked(state, state->state | open_flags); spin_unlock(&inode->i_lock); spin_unlock(&state->owner->so_lock); } @@ -352,6 +383,8 @@ static struct nfs4_state *nfs4_opendata_to_nfs4_state(struct nfs4_opendata *data { struct inode *inode; struct nfs4_state *state = NULL; + struct nfs_delegation *delegation; + nfs4_stateid *deleg_stateid = NULL; if (!(data->f_attr.valid & NFS_ATTR_FATTR)) goto out; @@ -361,13 +394,14 @@ static struct nfs4_state *nfs4_opendata_to_nfs4_state(struct nfs4_opendata *data state = nfs4_get_open_state(inode, data->owner); if (state == NULL) goto put_inode; - update_open_stateid(state, &data->o_res.stateid, data->o_arg.open_flags); if (data->o_res.delegation_type != 0) { - struct nfs_inode *nfsi = NFS_I(inode); int delegation_flags = 0; - if (nfsi->delegation) - delegation_flags = nfsi->delegation->flags; + rcu_read_lock(); + delegation = rcu_dereference(NFS_I(inode)->delegation); + if (delegation) + delegation_flags = delegation->flags; + rcu_read_unlock(); if (!(delegation_flags & NFS_DELEGATION_NEED_RECLAIM)) nfs_inode_set_delegation(state->inode, data->owner->so_cred, @@ -377,6 +411,12 @@ static struct nfs4_state *nfs4_opendata_to_nfs4_state(struct nfs4_opendata *data data->owner->so_cred, &data->o_res); } + rcu_read_lock(); + delegation = rcu_dereference(NFS_I(inode)->delegation); + if (delegation != NULL) + deleg_stateid = &delegation->stateid; + update_open_stateid(state, &data->o_res.stateid, deleg_stateid, data->o_arg.open_flags); + rcu_read_unlock(); put_inode: iput(inode); out: @@ -911,8 +951,7 @@ static int _nfs4_open_delegated(struct inode *inode, int flags, struct rpc_cred unlock_kernel(); if (err != 0) goto out_put_open_state; - set_bit(NFS_DELEGATED_STATE, &state->flags); - update_open_stateid(state, &delegation->stateid, open_flags); + update_open_stateid(state, NULL, &delegation->stateid, open_flags); out_ok: nfs4_put_state_owner(sp); up_read(&nfsi->rwsem); @@ -1149,8 +1188,7 @@ static void nfs4_close_done(struct rpc_task *task, void *data) nfs_increment_open_seqid(task->tk_status, calldata->arg.seqid); switch (task->tk_status) { case 0: - memcpy(&state->stateid, &calldata->res.stateid, - sizeof(state->stateid)); + nfs_set_open_stateid(state, &calldata->res.stateid, calldata->arg.open_flags); renew_lease(server, calldata->timestamp); break; case -NFS4ERR_STALE_STATEID: @@ -1175,26 +1213,32 @@ static void nfs4_close_prepare(struct rpc_task *task, void *data) .rpc_resp = &calldata->res, .rpc_cred = state->owner->so_cred, }; - int mode = 0, old_mode; + int clear_rd, clear_wr, clear_rdwr; + int mode; if (nfs_wait_on_sequence(calldata->arg.seqid, task) != 0) return; - /* Recalculate the new open mode in case someone reopened the file - * while we were waiting in line to be scheduled. - */ + + mode = FMODE_READ|FMODE_WRITE; + clear_rd = clear_wr = clear_rdwr = 0; spin_lock(&state->owner->so_lock); spin_lock(&calldata->inode->i_lock); - mode = old_mode = state->state; + /* Calculate the change in open mode */ if (state->n_rdwr == 0) { - if (state->n_rdonly == 0) + if (state->n_rdonly == 0) { mode &= ~FMODE_READ; - if (state->n_wronly == 0) + clear_rd |= test_and_clear_bit(NFS_O_RDONLY_STATE, &state->flags); + clear_rdwr |= test_and_clear_bit(NFS_O_RDWR_STATE, &state->flags); + } + if (state->n_wronly == 0) { mode &= ~FMODE_WRITE; + clear_wr |= test_and_clear_bit(NFS_O_WRONLY_STATE, &state->flags); + clear_rdwr |= test_and_clear_bit(NFS_O_RDWR_STATE, &state->flags); + } } - nfs4_state_set_mode_locked(state, mode); spin_unlock(&calldata->inode->i_lock); spin_unlock(&state->owner->so_lock); - if (mode == old_mode || test_bit(NFS_DELEGATED_STATE, &state->flags)) { + if (!clear_rd && !clear_wr && !clear_rdwr) { /* Note: exit _without_ calling nfs4_close_done */ task->tk_action = NULL; return; @@ -1238,7 +1282,7 @@ int nfs4_do_close(struct path *path, struct nfs4_state *state) calldata->inode = state->inode; calldata->state = state; calldata->arg.fh = NFS_FH(state->inode); - calldata->arg.stateid = &state->stateid; + calldata->arg.stateid = &state->open_stateid; /* Serialization for the sequence id */ calldata->arg.seqid = nfs_alloc_seqid(&state->owner->so_seqid); if (calldata->arg.seqid == NULL) diff --git a/fs/nfs/nfs4state.c b/fs/nfs/nfs4state.c index ac816b303f3a..4f78c0d1eab5 100644 --- a/fs/nfs/nfs4state.c +++ b/fs/nfs/nfs4state.c @@ -412,7 +412,8 @@ void nfs4_close_state(struct path *path, struct nfs4_state *state, mode_t mode) { struct inode *inode = state->inode; struct nfs4_state_owner *owner = state->owner; - int oldstate, newstate = 0; + int call_close = 0; + int newstate; atomic_inc(&owner->so_count); /* Protect against nfs4_find_state() */ @@ -428,21 +429,26 @@ void nfs4_close_state(struct path *path, struct nfs4_state *state, mode_t mode) case FMODE_READ|FMODE_WRITE: state->n_rdwr--; } - oldstate = newstate = state->state; + newstate = FMODE_READ|FMODE_WRITE; if (state->n_rdwr == 0) { - if (state->n_rdonly == 0) + if (state->n_rdonly == 0) { newstate &= ~FMODE_READ; - if (state->n_wronly == 0) + call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags); + call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags); + } + if (state->n_wronly == 0) { newstate &= ~FMODE_WRITE; + call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags); + call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags); + } + if (newstate == 0) + clear_bit(NFS_DELEGATED_STATE, &state->flags); } - if (test_bit(NFS_DELEGATED_STATE, &state->flags)) { - nfs4_state_set_mode_locked(state, newstate); - oldstate = newstate; - } + nfs4_state_set_mode_locked(state, newstate); spin_unlock(&inode->i_lock); spin_unlock(&owner->so_lock); - if (oldstate == newstate) { + if (!call_close) { nfs4_put_open_state(state); nfs4_put_state_owner(owner); } else @@ -838,6 +844,10 @@ static void nfs4_state_mark_reclaim(struct nfs_client *clp) sp->so_seqid.flags = 0; spin_lock(&sp->so_lock); list_for_each_entry(state, &sp->so_states, open_states) { + clear_bit(NFS_DELEGATED_STATE, &state->flags); + clear_bit(NFS_O_RDONLY_STATE, &state->flags); + clear_bit(NFS_O_WRONLY_STATE, &state->flags); + clear_bit(NFS_O_RDWR_STATE, &state->flags); list_for_each_entry(lock, &state->lock_states, ls_locks) { lock->ls_seqid.counter = 0; lock->ls_seqid.flags = 0; -- cgit v1.2.3 From 6f43ddccb31b5bd2297878f6f3735d45fd4dfce3 Mon Sep 17 00:00:00 2001 From: Trond Myklebust Date: Sun, 8 Jul 2007 16:49:11 -0400 Subject: NFSv4: Improve the debugging of bad sequence id errors... Signed-off-by: Trond Myklebust --- fs/nfs/nfs4proc.c | 4 +++- fs/nfs/nfs4state.c | 7 +++++++ 2 files changed, 10 insertions(+), 1 deletion(-) (limited to 'fs/nfs/nfs4state.c') diff --git a/fs/nfs/nfs4proc.c b/fs/nfs/nfs4proc.c index 128fe23d3f1e..3b59c5ded3fb 100644 --- a/fs/nfs/nfs4proc.c +++ b/fs/nfs/nfs4proc.c @@ -1079,7 +1079,9 @@ static struct nfs4_state *nfs4_do_open(struct inode *dir, struct path *path, int * the user though... */ if (status == -NFS4ERR_BAD_SEQID) { - printk(KERN_WARNING "NFS: v4 server returned a bad sequence-id error!\n"); + printk(KERN_WARNING "NFS: v4 server %s " + " returned a bad sequence-id error!\n", + NFS_SERVER(dir)->nfs_client->cl_hostname); exception.retry = 1; continue; } diff --git a/fs/nfs/nfs4state.c b/fs/nfs/nfs4state.c index 4f78c0d1eab5..4fa4054cdf34 100644 --- a/fs/nfs/nfs4state.c +++ b/fs/nfs/nfs4state.c @@ -38,6 +38,7 @@ * subsequent patch. */ +#include #include #include #include @@ -648,6 +649,12 @@ static void nfs_increment_seqid(int status, struct nfs_seqid *seqid) case 0: break; case -NFS4ERR_BAD_SEQID: + if (seqid->sequence->flags & NFS_SEQID_CONFIRMED) + return; + printk(KERN_WARNING "NFS: v4 server returned a bad" + "sequence-id error on an" + "unconfirmed sequence %p!\n", + seqid->sequence); case -NFS4ERR_STALE_CLIENTID: case -NFS4ERR_STALE_STATEID: case -NFS4ERR_BAD_STATEID: -- cgit v1.2.3 From 8bda4e4c98d14566fc1a354c62fb59d70cc49b97 Mon Sep 17 00:00:00 2001 From: Trond Myklebust Date: Mon, 9 Jul 2007 10:45:42 -0400 Subject: NFSv4: Fix up stateid locking... We really don't need to grab both the state->so_owner and the inode->i_lock. Signed-off-by: Trond Myklebust --- fs/nfs/nfs4_fs.h | 2 ++ fs/nfs/nfs4proc.c | 27 +++++++++++---------------- fs/nfs/nfs4state.c | 10 ++++++---- 3 files changed, 19 insertions(+), 20 deletions(-) (limited to 'fs/nfs/nfs4state.c') diff --git a/fs/nfs/nfs4_fs.h b/fs/nfs/nfs4_fs.h index 4a1c4d80a577..dd1aa2b598ce 100644 --- a/fs/nfs/nfs4_fs.h +++ b/fs/nfs/nfs4_fs.h @@ -139,9 +139,11 @@ struct nfs4_state { unsigned long flags; /* Do we hold any locks? */ spinlock_t state_lock; /* Protects the lock_states list */ + seqlock_t seqlock; /* Protects the stateid/open_stateid */ nfs4_stateid stateid; /* Current stateid: may be delegation */ nfs4_stateid open_stateid; /* OPEN stateid */ + /* The following 3 fields are protected by owner->so_lock */ unsigned int n_rdonly; /* Number of read-only references */ unsigned int n_wronly; /* Number of write-only references */ unsigned int n_rdwr; /* Number of read/write references */ diff --git a/fs/nfs/nfs4proc.c b/fs/nfs/nfs4proc.c index 3a2af8053767..ba86ec654c2e 100644 --- a/fs/nfs/nfs4proc.c +++ b/fs/nfs/nfs4proc.c @@ -385,29 +385,28 @@ static void nfs_set_open_stateid_locked(struct nfs4_state *state, nfs4_stateid * static void nfs_set_open_stateid(struct nfs4_state *state, nfs4_stateid *stateid, int open_flags) { - spin_lock(&state->owner->so_lock); - spin_lock(&state->inode->i_lock); + write_seqlock(&state->seqlock); nfs_set_open_stateid_locked(state, stateid, open_flags); - spin_unlock(&state->inode->i_lock); - spin_unlock(&state->owner->so_lock); + write_sequnlock(&state->seqlock); } static void update_open_stateid(struct nfs4_state *state, nfs4_stateid *open_stateid, nfs4_stateid *deleg_stateid, int open_flags) { - struct inode *inode = state->inode; - open_flags &= (FMODE_READ|FMODE_WRITE); - /* Protect against nfs4_find_state_byowner() */ - spin_lock(&state->owner->so_lock); - spin_lock(&inode->i_lock); + /* + * Protect the call to nfs4_state_set_mode_locked and + * serialise the stateid update + */ + write_seqlock(&state->seqlock); if (deleg_stateid != NULL) { memcpy(state->stateid.data, deleg_stateid->data, sizeof(state->stateid.data)); set_bit(NFS_DELEGATED_STATE, &state->flags); } if (open_stateid != NULL) nfs_set_open_stateid_locked(state, open_stateid, open_flags); + write_sequnlock(&state->seqlock); + spin_lock(&state->owner->so_lock); update_open_stateflags(state, open_flags); - spin_unlock(&inode->i_lock); spin_unlock(&state->owner->so_lock); } @@ -608,12 +607,10 @@ static int nfs4_open_recover(struct nfs4_opendata *opendata, struct nfs4_state * */ if (test_bit(NFS_DELEGATED_STATE, &state->flags) == 0 && memcmp(state->stateid.data, state->open_stateid.data, sizeof(state->stateid.data)) != 0) { - spin_lock(&state->owner->so_lock); - spin_lock(&state->inode->i_lock); + write_seqlock(&state->seqlock); if (test_bit(NFS_DELEGATED_STATE, &state->flags) == 0) memcpy(state->stateid.data, state->open_stateid.data, sizeof(state->stateid.data)); - spin_unlock(&state->inode->i_lock); - spin_unlock(&state->owner->so_lock); + write_sequnlock(&state->seqlock); } return 0; } @@ -1280,7 +1277,6 @@ static void nfs4_close_prepare(struct rpc_task *task, void *data) mode = FMODE_READ|FMODE_WRITE; clear_rd = clear_wr = clear_rdwr = 0; spin_lock(&state->owner->so_lock); - spin_lock(&calldata->inode->i_lock); /* Calculate the change in open mode */ if (state->n_rdwr == 0) { if (state->n_rdonly == 0) { @@ -1294,7 +1290,6 @@ static void nfs4_close_prepare(struct rpc_task *task, void *data) clear_rdwr |= test_and_clear_bit(NFS_O_RDWR_STATE, &state->flags); } } - spin_unlock(&calldata->inode->i_lock); spin_unlock(&state->owner->so_lock); if (!clear_rd && !clear_wr && !clear_rdwr) { /* Note: exit _without_ calling nfs4_close_done */ diff --git a/fs/nfs/nfs4state.c b/fs/nfs/nfs4state.c index 4fa4054cdf34..523cc2cbb5e1 100644 --- a/fs/nfs/nfs4state.c +++ b/fs/nfs/nfs4state.c @@ -307,6 +307,7 @@ nfs4_alloc_open_state(void) atomic_set(&state->count, 1); INIT_LIST_HEAD(&state->lock_states); spin_lock_init(&state->state_lock); + seqlock_init(&state->seqlock); return state; } @@ -411,7 +412,6 @@ void nfs4_put_open_state(struct nfs4_state *state) */ void nfs4_close_state(struct path *path, struct nfs4_state *state, mode_t mode) { - struct inode *inode = state->inode; struct nfs4_state_owner *owner = state->owner; int call_close = 0; int newstate; @@ -419,7 +419,6 @@ void nfs4_close_state(struct path *path, struct nfs4_state *state, mode_t mode) atomic_inc(&owner->so_count); /* Protect against nfs4_find_state() */ spin_lock(&owner->so_lock); - spin_lock(&inode->i_lock); switch (mode & (FMODE_READ | FMODE_WRITE)) { case FMODE_READ: state->n_rdonly--; @@ -446,7 +445,6 @@ void nfs4_close_state(struct path *path, struct nfs4_state *state, mode_t mode) clear_bit(NFS_DELEGATED_STATE, &state->flags); } nfs4_state_set_mode_locked(state, newstate); - spin_unlock(&inode->i_lock); spin_unlock(&owner->so_lock); if (!call_close) { @@ -599,8 +597,12 @@ int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl) void nfs4_copy_stateid(nfs4_stateid *dst, struct nfs4_state *state, fl_owner_t fl_owner) { struct nfs4_lock_state *lsp; + int seq; - memcpy(dst, &state->stateid, sizeof(*dst)); + do { + seq = read_seqbegin(&state->seqlock); + memcpy(dst, &state->stateid, sizeof(*dst)); + } while (read_seqretry(&state->seqlock, seq)); if (test_bit(LK_STATE_IN_USE, &state->flags) == 0) return; -- cgit v1.2.3 From 6f2e64d3e1f661095e274c9d9d47e3f39a6cf1c0 Mon Sep 17 00:00:00 2001 From: Trond Myklebust Date: Fri, 6 Jul 2007 10:53:21 -0400 Subject: NFSv4: Make the NFS state model work with the nosharedcache mount option Consider the case where the user has mounted the remote filesystem server:/foo on the two local directories /bar and /baz using the nosharedcache mount option. The files /bar/file and /baz/file are represented by different inodes in the local namespace, but refer to the same file /foo/file on the server. Consider the case where a process opens both /bar/file and /baz/file, then closes /bar/file: because the nfs4_state is not shared between /bar/file and /baz/file, the kernel will see that the nfs4_state for /bar/file is no longer referenced, so it will send off a CLOSE rpc call. Unless the open_owners differ, then that CLOSE call will invalidate the open state on /baz/file too. Conclusion: we cannot share open state owners between two different non-shared mount instances of the same filesystem. Signed-off-by: Trond Myklebust --- fs/nfs/nfs4_fs.h | 1 + fs/nfs/nfs4state.c | 22 ++++++++++++++++++++-- 2 files changed, 21 insertions(+), 2 deletions(-) (limited to 'fs/nfs/nfs4state.c') diff --git a/fs/nfs/nfs4_fs.h b/fs/nfs/nfs4_fs.h index dd1aa2b598ce..6c028e734fe6 100644 --- a/fs/nfs/nfs4_fs.h +++ b/fs/nfs/nfs4_fs.h @@ -83,6 +83,7 @@ struct nfs_unique_id { struct nfs4_state_owner { struct nfs_unique_id so_owner_id; struct nfs_client *so_client; + struct nfs_server *so_server; struct rb_node so_client_node; struct rpc_cred *so_cred; /* Associated cred */ diff --git a/fs/nfs/nfs4state.c b/fs/nfs/nfs4state.c index 523cc2cbb5e1..e9662ba81d86 100644 --- a/fs/nfs/nfs4state.c +++ b/fs/nfs/nfs4state.c @@ -156,8 +156,9 @@ static void nfs_free_unique_id(struct rb_root *root, struct nfs_unique_id *id) } static struct nfs4_state_owner * -nfs4_find_state_owner(struct nfs_client *clp, struct rpc_cred *cred) +nfs4_find_state_owner(struct nfs_server *server, struct rpc_cred *cred) { + struct nfs_client *clp = server->nfs_client; struct rb_node **p = &clp->cl_state_owners.rb_node, *parent = NULL; struct nfs4_state_owner *sp, *res = NULL; @@ -166,6 +167,14 @@ nfs4_find_state_owner(struct nfs_client *clp, struct rpc_cred *cred) parent = *p; sp = rb_entry(parent, struct nfs4_state_owner, so_client_node); + if (server < sp->so_server) { + p = &parent->rb_left; + continue; + } + if (server > sp->so_server) { + p = &parent->rb_right; + continue; + } if (cred < sp->so_cred) p = &parent->rb_left; else if (cred > sp->so_cred) @@ -190,6 +199,14 @@ nfs4_insert_state_owner(struct nfs_client *clp, struct nfs4_state_owner *new) parent = *p; sp = rb_entry(parent, struct nfs4_state_owner, so_client_node); + if (new->so_server < sp->so_server) { + p = &parent->rb_left; + continue; + } + if (new->so_server > sp->so_server) { + p = &parent->rb_right; + continue; + } if (new->so_cred < sp->so_cred) p = &parent->rb_left; else if (new->so_cred > sp->so_cred) @@ -260,7 +277,7 @@ struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server, struct struct nfs4_state_owner *sp, *new; spin_lock(&clp->cl_lock); - sp = nfs4_find_state_owner(clp, cred); + sp = nfs4_find_state_owner(server, cred); spin_unlock(&clp->cl_lock); if (sp != NULL) return sp; @@ -268,6 +285,7 @@ struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server, struct if (new == NULL) return NULL; new->so_client = clp; + new->so_server = server; new->so_cred = cred; spin_lock(&clp->cl_lock); sp = nfs4_insert_state_owner(clp, new); -- cgit v1.2.3