* subsequent patch.
*/
+#include <linux/kernel.h>
#include <linux/slab.h>
#include <linux/smp_lock.h>
#include <linux/nfs_fs.h>
#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>
#include "nfs4_fs.h"
#include "callback.h"
#include "delegation.h"
+#include "internal.h"
#define OPENOWNER_POOL_SIZE 8
const nfs4_stateid zero_stateid;
-static DEFINE_SPINLOCK(state_spinlock);
static LIST_HEAD(nfs4_clientid_list);
-void
-init_nfsv4_state(struct nfs_server *server)
-{
- server->nfs4_state = NULL;
- INIT_LIST_HEAD(&server->nfs4_siblings);
-}
-
-void
-destroy_nfsv4_state(struct nfs_server *server)
+static int nfs4_init_client(struct nfs_client *clp, struct rpc_cred *cred)
{
- kfree(server->mnt_path);
- server->mnt_path = NULL;
- if (server->nfs4_state) {
- nfs4_put_client(server->nfs4_state);
- server->nfs4_state = NULL;
- }
+ int status = nfs4_proc_setclientid(clp, NFS4_CALLBACK,
+ nfs_callback_tcpport, cred);
+ if (status == 0)
+ status = nfs4_proc_setclientid_confirm(clp, cred);
+ if (status == 0)
+ nfs4_schedule_state_renewal(clp);
+ return status;
}
-/*
- * nfs4_get_client(): returns an empty client structure
- * nfs4_put_client(): drops reference to client structure
- *
- * Since these are allocated/deallocated very rarely, we don't
- * bother putting them in a slab cache...
- */
-static struct nfs_client *
-nfs4_alloc_client(struct in_addr *addr)
+struct rpc_cred *nfs4_get_renew_cred(struct nfs_client *clp)
{
- struct nfs_client *clp;
+ struct nfs4_state_owner *sp;
+ struct rb_node *pos;
+ struct rpc_cred *cred = NULL;
- if (nfs_callback_up() < 0)
- return NULL;
- if ((clp = kzalloc(sizeof(*clp), GFP_KERNEL)) == NULL) {
- nfs_callback_down();
- return NULL;
+ 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);
+ break;
}
- memcpy(&clp->cl_addr, addr, sizeof(clp->cl_addr));
- 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);
- atomic_set(&clp->cl_count, 1);
- INIT_WORK(&clp->cl_renewd, nfs4_renew_state, clp);
- INIT_LIST_HEAD(&clp->cl_superblocks);
- rpc_init_wait_queue(&clp->cl_rpcwaitq, "NFS4 client");
- clp->cl_rpcclient = ERR_PTR(-EINVAL);
- clp->cl_boot_time = CURRENT_TIME;
- clp->cl_state = 1 << NFS4CLNT_LEASE_EXPIRED;
- return clp;
+ return cred;
}
-static void
-nfs4_free_client(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;
- while (!list_empty(&clp->cl_unused)) {
- 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));
- nfs_idmap_delete(clp);
- if (!IS_ERR(clp->cl_rpcclient))
- rpc_shutdown_client(clp->cl_rpcclient);
- kfree(clp);
- nfs_callback_down();
-}
-
-static struct nfs_client *__nfs4_find_client(struct in_addr *addr)
-{
- struct nfs_client *clp;
- list_for_each_entry(clp, &nfs4_clientid_list, cl_servers) {
- if (memcmp(&clp->cl_addr, addr, sizeof(clp->cl_addr)) == 0) {
- atomic_inc(&clp->cl_count);
- return clp;
- }
+ 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;
}
-struct nfs_client *nfs4_find_client(struct in_addr *addr)
+static void nfs_alloc_unique_id(struct rb_root *root, struct nfs_unique_id *new,
+ __u64 minval, int maxbits)
{
- struct nfs_client *clp;
- spin_lock(&state_spinlock);
- clp = __nfs4_find_client(addr);
- spin_unlock(&state_spinlock);
- return clp;
-}
+ struct rb_node **p, *parent;
+ struct nfs_unique_id *pos;
+ __u64 mask = ~0ULL;
-struct nfs_client *
-nfs4_get_client(struct in_addr *addr)
-{
- struct nfs_client *clp, *new = NULL;
+ 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;
- spin_lock(&state_spinlock);
+ 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 (;;) {
- clp = __nfs4_find_client(addr);
- if (clp != NULL)
- break;
- clp = new;
- if (clp != NULL) {
- list_add(&clp->cl_servers, &nfs4_clientid_list);
- new = NULL;
+ new->id++;
+ if (new->id < minval || (new->id & mask) != new->id) {
+ new->id = minval;
break;
}
- spin_unlock(&state_spinlock);
- new = nfs4_alloc_client(addr);
- spin_lock(&state_spinlock);
- if (new == NULL)
+ parent = rb_next(parent);
+ if (parent == NULL)
+ break;
+ pos = rb_entry(parent, struct nfs_unique_id, rb_node);
+ if (new->id < pos->id)
break;
}
- spin_unlock(&state_spinlock);
- if (new)
- nfs4_free_client(new);
- return clp;
+ goto retry;
}
-void
-nfs4_put_client(struct nfs_client *clp)
+static void nfs_free_unique_id(struct rb_root *root, struct nfs_unique_id *id)
{
- if (!atomic_dec_and_lock(&clp->cl_count, &state_spinlock))
- return;
- list_del(&clp->cl_servers);
- spin_unlock(&state_spinlock);
- BUG_ON(!list_empty(&clp->cl_superblocks));
- rpc_wake_up(&clp->cl_rpcwaitq);
- nfs4_kill_renewd(clp);
- nfs4_free_client(clp);
-}
-
-static int nfs4_init_client(struct nfs_client *clp, struct rpc_cred *cred)
-{
- int status = nfs4_proc_setclientid(clp, NFS4_CALLBACK,
- nfs_callback_tcpport, cred);
- if (status == 0)
- status = nfs4_proc_setclientid_confirm(clp, cred);
- if (status == 0)
- nfs4_schedule_state_renewal(clp);
- return status;
-}
-
-u32
-nfs4_alloc_lockowner_id(struct nfs_client *clp)
-{
- return clp->cl_lockowner_id ++;
+ rb_erase(&id->rb_node, root);
}
static struct nfs4_state_owner *
-nfs4_client_grab_unused(struct nfs_client *clp, struct rpc_cred *cred)
+nfs4_find_state_owner(struct nfs_server *server, 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 = cred;
- list_move(&sp->so_list, &clp->cl_state_owners);
- clp->cl_nunused--;
- }
- return sp;
-}
+ 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;
-struct rpc_cred *nfs4_get_renew_cred(struct nfs_client *clp)
-{
- struct nfs4_state_owner *sp;
- struct rpc_cred *cred = NULL;
+ while (*p != NULL) {
+ parent = *p;
+ sp = rb_entry(parent, struct nfs4_state_owner, so_client_node);
- list_for_each_entry(sp, &clp->cl_state_owners, so_list) {
- if (list_empty(&sp->so_states))
+ if (server < sp->so_server) {
+ p = &parent->rb_left;
continue;
- cred = get_rpccred(sp->so_cred);
- break;
+ }
+ 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)
+ p = &parent->rb_right;
+ else {
+ atomic_inc(&sp->so_count);
+ res = sp;
+ break;
+ }
}
- return cred;
+ return res;
}
-struct rpc_cred *nfs4_get_setclientid_cred(struct nfs_client *clp)
+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;
- if (!list_empty(&clp->cl_state_owners)) {
- sp = list_entry(clp->cl_state_owners.next,
- struct nfs4_state_owner, so_list);
- return get_rpccred(sp->so_cred);
+ while (*p != NULL) {
+ 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)
+ p = &parent->rb_right;
+ else {
+ atomic_inc(&sp->so_count);
+ return sp;
+ }
}
- return NULL;
+ 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 struct nfs4_state_owner *
-nfs4_find_state_owner(struct nfs_client *clp, struct rpc_cred *cred)
+static void
+nfs4_remove_state_owner(struct nfs_client *clp, struct nfs4_state_owner *sp)
{
- 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;
- }
- return res;
+ 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);
}
/*
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);
+ }
}
/*
*/
struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server, struct rpc_cred *cred)
{
- struct nfs_client *clp = server->nfs4_state;
+ 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);
- 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;
- new->so_id = nfs4_alloc_lockowner_id(clp);
- new->so_cred = cred;
- sp = new;
- new = NULL;
- }
+ sp = nfs4_find_state_owner(server, cred);
spin_unlock(&clp->cl_lock);
- kfree(new);
if (sp != NULL)
return sp;
- put_rpccred(cred);
- return NULL;
+ new = nfs4_alloc_state_owner();
+ 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);
+ spin_unlock(&clp->cl_lock);
+ if (sp == new)
+ get_rpccred(cred);
+ else
+ kfree(new);
+ return 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);
+ nfs4_remove_state_owner(clp, sp);
spin_unlock(&clp->cl_lock);
put_rpccred(cred);
kfree(sp);
atomic_set(&state->count, 1);
INIT_LIST_HEAD(&state->lock_states);
spin_lock_init(&state->state_lock);
+ seqlock_init(&state->seqlock);
return state;
}
else
list_move_tail(&state->open_states, &state->owner->so_states);
}
- if (mode == 0)
- list_del_init(&state->inode_states);
state->state = mode;
}
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;
}
if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
return;
spin_lock(&inode->i_lock);
- if (!list_empty(&state->inode_states))
- list_del(&state->inode_states);
+ list_del(&state->inode_states);
list_del(&state->open_states);
spin_unlock(&inode->i_lock);
spin_unlock(&owner->so_lock);
/*
* 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;
- int oldstate, newstate = 0;
+ int call_close = 0;
+ int newstate;
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--;
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;
- }
- spin_unlock(&inode->i_lock);
+ nfs4_state_set_mode_locked(state, newstate);
spin_unlock(&owner->so_lock);
- if (oldstate != newstate && nfs4_do_close(inode, state) == 0)
- return;
- nfs4_put_open_state(state);
- nfs4_put_state_owner(owner);
+ if (!call_close) {
+ nfs4_put_open_state(state);
+ nfs4_put_state_owner(owner);
+ } else
+ nfs4_do_close(path, state);
}
/*
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.
return NULL;
}
spin_unlock(&state->state_lock);
- kfree(new);
+ if (new != NULL)
+ nfs4_free_lock_state(new);
return 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)
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;
* 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:
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:
struct nfs4_state_owner, so_seqid);
nfs4_drop_state_owner(sp);
}
- return nfs_increment_seqid(status, seqid);
+ nfs_increment_seqid(status, 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)
__module_get(THIS_MODULE);
atomic_inc(&clp->cl_count);
task = kthread_run(reclaimer, clp, "%u.%u.%u.%u-reclaim",
- NIPQUAD(clp->cl_addr));
+ NIPQUAD(clp->cl_addr.sin_addr));
if (!IS_ERR(task))
return;
nfs4_clear_recover_bit(clp);
- nfs4_put_client(clp);
+ nfs_put_client(clp);
module_put(THIS_MODULE);
}
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);
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;
{
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;
/* 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) {
if (status == -NFS4ERR_CB_PATH_DOWN)
nfs_handle_cb_pathdown(clp);
nfs4_clear_recover_bit(clp);
- nfs4_put_client(clp);
+ nfs_put_client(clp);
module_put_and_exit(0);
return 0;
out_error:
printk(KERN_WARNING "Error: state recovery failed on NFSv4 server %u.%u.%u.%u with error %d\n",
- NIPQUAD(clp->cl_addr.s_addr), -status);
+ NIPQUAD(clp->cl_addr.sin_addr), -status);
set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
goto out;
}
git://ftp.safe.ca / safe / jmp / linux-2.6 / blobdiff