]> err.no Git - linux-2.6/commitdiff
NFSv4: Reduce the chances of an open_owner identifier collision
authorTrond Myklebust <Trond.Myklebust@netapp.com>
Mon, 2 Jul 2007 17:58:33 +0000 (13:58 -0400)
committerTrond Myklebust <Trond.Myklebust@netapp.com>
Wed, 11 Jul 2007 03:40:39 +0000 (23:40 -0400)
Currently we just use a 32-bit counter.

Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
fs/nfs/client.c
fs/nfs/nfs4_fs.h
fs/nfs/nfs4proc.c
fs/nfs/nfs4state.c
fs/nfs/nfs4xdr.c
include/linux/nfs_fs_sb.h
include/linux/nfs_xdr.h

index 6b424407d631c0750595b1e8fbcc822959dbebb2..ccb455053ee48090ab21b4bdd55294d3f4694010 100644 (file)
@@ -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
index c97a0ad8430ee23cd93fd9f0320e12804f52cd88..44b56c915f722a5deadba122d4e9e533cbc1cb8c 100644 (file)
@@ -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 *);
index 84d0b7e0dd678e44d953da97f59f75466a9ebb55..1840ebc78fd3dc505171bfcd6ba08ff4338e4ec7 100644 (file)
@@ -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);
index 0f79d56e97f0507b377fa5b0a6edd4c8536921f1..ab0b5ab60e60dd7e192c9f0be85af64e1a5e2c53 100644 (file)
@@ -44,6 +44,7 @@
 #include <linux/nfs_idmap.h>
 #include <linux/kthread.h>
 #include <linux/module.h>
+#include <linux/random.h>
 #include <linux/workqueue.h>
 #include <linux/bitops.h>
 
@@ -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) {
index 4c8f67d4752338d1eb67a58f3b2f30c92d998a8a..c08738441f732216f7b785fa3805860475a40917 100644 (file)
@@ -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)
index 2cef0a68aa77ad2a189650379c3f7206e21f061e..0cac49bc09553ef33b572593f038bb0e198fb17d 100644 (file)
@@ -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;
index f7100df3a690ecda87580bd2473382a91ac9905c..38d77681cf2796b5343f7e419ed769374ca368f4 100644 (file)
@@ -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 {