* subsequent patch.
*/
-#include <linux/config.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/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);
-static void nfs4_recover_state(void *);
-
-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)
-{
- if (server->mnt_path) {
- kfree(server->mnt_path);
- server->mnt_path = NULL;
- }
- if (server->nfs4_state) {
- nfs4_put_client(server->nfs4_state);
- server->nfs4_state = NULL;
- }
-}
-
-/*
- * 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 nfs4_client *
-nfs4_alloc_client(struct in_addr *addr)
+static int nfs4_init_client(struct nfs_client *clp, struct rpc_cred *cred)
{
- struct nfs4_client *clp;
-
- if (nfs_callback_up() < 0)
- return NULL;
- if ((clp = kmalloc(sizeof(*clp), GFP_KERNEL)) == NULL) {
- nfs_callback_down();
- return NULL;
- }
- memset(clp, 0, sizeof(*clp));
- 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_recoverd, nfs4_recover_state, clp);
- INIT_WORK(&clp->cl_renewd, nfs4_renew_state, clp);
- INIT_LIST_HEAD(&clp->cl_superblocks);
- init_waitqueue_head(&clp->cl_waitq);
- 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_OK;
- return clp;
-}
-
-static void
-nfs4_free_client(struct nfs4_client *clp)
-{
- struct nfs4_state_owner *sp;
-
- 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));
- if (clp->cl_cred)
- put_rpccred(clp->cl_cred);
- nfs_idmap_delete(clp);
- if (!IS_ERR(clp->cl_rpcclient))
- rpc_shutdown_client(clp->cl_rpcclient);
- kfree(clp);
- nfs_callback_down();
-}
-
-static struct nfs4_client *__nfs4_find_client(struct in_addr *addr)
-{
- struct nfs4_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;
- }
- }
- return NULL;
-}
-
-struct nfs4_client *nfs4_find_client(struct in_addr *addr)
-{
- struct nfs4_client *clp;
- spin_lock(&state_spinlock);
- clp = __nfs4_find_client(addr);
- spin_unlock(&state_spinlock);
- return clp;
-}
-
-struct nfs4_client *
-nfs4_get_client(struct in_addr *addr)
-{
- struct nfs4_client *clp, *new = NULL;
-
- spin_lock(&state_spinlock);
- 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;
- break;
- }
- spin_unlock(&state_spinlock);
- new = nfs4_alloc_client(addr);
- spin_lock(&state_spinlock);
- if (new == NULL)
- break;
- }
- spin_unlock(&state_spinlock);
- if (new)
- nfs4_free_client(new);
- return clp;
-}
-
-void
-nfs4_put_client(struct nfs4_client *clp)
-{
- 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));
- wake_up_all(&clp->cl_waitq);
- rpc_wake_up(&clp->cl_rpcwaitq);
- nfs4_kill_renewd(clp);
- nfs4_free_client(clp);
-}
-
-static int __nfs4_init_client(struct nfs4_client *clp)
-{
- int status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, nfs_callback_tcpport);
+ int status = nfs4_proc_setclientid(clp, NFS4_CALLBACK,
+ nfs_callback_tcpport, cred);
if (status == 0)
- status = nfs4_proc_setclientid_confirm(clp);
+ status = nfs4_proc_setclientid_confirm(clp, cred);
if (status == 0)
nfs4_schedule_state_renewal(clp);
return status;
}
-int nfs4_init_client(struct nfs4_client *clp)
-{
- return nfs4_map_errors(__nfs4_init_client(clp));
-}
-
u32
-nfs4_alloc_lockowner_id(struct nfs4_client *clp)
+nfs4_alloc_lockowner_id(struct nfs_client *clp)
{
return clp->cl_lockowner_id ++;
}
static struct nfs4_state_owner *
-nfs4_client_grab_unused(struct nfs4_client *clp, struct rpc_cred *cred)
+nfs4_client_grab_unused(struct nfs_client *clp, struct rpc_cred *cred)
{
struct nfs4_state_owner *sp = NULL;
return sp;
}
+struct rpc_cred *nfs4_get_renew_cred(struct nfs_client *clp)
+{
+ struct nfs4_state_owner *sp;
+ struct rpc_cred *cred = NULL;
+
+ list_for_each_entry(sp, &clp->cl_state_owners, so_list) {
+ if (list_empty(&sp->so_states))
+ continue;
+ cred = get_rpccred(sp->so_cred);
+ break;
+ }
+ return cred;
+}
+
+struct rpc_cred *nfs4_get_setclientid_cred(struct nfs_client *clp)
+{
+ 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);
+ }
+ return NULL;
+}
+
static struct nfs4_state_owner *
-nfs4_find_state_owner(struct nfs4_client *clp, struct rpc_cred *cred)
+nfs4_find_state_owner(struct nfs_client *clp, struct rpc_cred *cred)
{
struct nfs4_state_owner *sp, *res = NULL;
void
nfs4_drop_state_owner(struct nfs4_state_owner *sp)
{
- struct nfs4_client *clp = sp->so_client;
+ struct nfs_client *clp = sp->so_client;
spin_lock(&clp->cl_lock);
list_del_init(&sp->so_list);
spin_unlock(&clp->cl_lock);
*/
struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server, struct rpc_cred *cred)
{
- struct nfs4_client *clp = server->nfs4_state;
+ struct nfs_client *clp = server->nfs_client;
struct nfs4_state_owner *sp, *new;
get_rpccred(cred);
new = NULL;
}
spin_unlock(&clp->cl_lock);
- if (new)
- kfree(new);
+ kfree(new);
if (sp != NULL)
return sp;
put_rpccred(cred);
*/
void nfs4_put_state_owner(struct nfs4_state_owner *sp)
{
- struct nfs4_client *clp = sp->so_client;
+ struct nfs_client *clp = sp->so_client;
struct rpc_cred *cred = sp->so_cred;
if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
{
struct nfs4_state *state;
- state = kmalloc(sizeof(*state), GFP_KERNEL);
+ state = kzalloc(sizeof(*state), GFP_KERNEL);
if (!state)
return NULL;
- state->state = 0;
- state->nreaders = 0;
- state->nwriters = 0;
- state->flags = 0;
- memset(state->stateid.data, 0, sizeof(state->stateid.data));
atomic_set(&state->count, 1);
INIT_LIST_HEAD(&state->lock_states);
spin_lock_init(&state->state_lock);
return state;
}
-static struct nfs4_state *
-__nfs4_find_state(struct inode *inode, struct rpc_cred *cred, mode_t mode)
+void
+nfs4_state_set_mode_locked(struct nfs4_state *state, mode_t mode)
{
- struct nfs_inode *nfsi = NFS_I(inode);
- struct nfs4_state *state;
-
- mode &= (FMODE_READ|FMODE_WRITE);
- list_for_each_entry(state, &nfsi->open_states, inode_states) {
- if (state->owner->so_cred != cred)
- continue;
- if ((mode & FMODE_READ) != 0 && state->nreaders == 0)
- continue;
- if ((mode & FMODE_WRITE) != 0 && state->nwriters == 0)
- continue;
- if ((state->state & mode) != mode)
- continue;
- atomic_inc(&state->count);
- if (mode & FMODE_READ)
- state->nreaders++;
+ if (state->state == mode)
+ return;
+ /* NB! List reordering - see the reclaim code for why. */
+ if ((mode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
if (mode & FMODE_WRITE)
- state->nwriters++;
- return state;
+ list_move(&state->open_states, &state->owner->so_states);
+ else
+ list_move_tail(&state->open_states, &state->owner->so_states);
}
- return NULL;
+ if (mode == 0)
+ list_del_init(&state->inode_states);
+ state->state = mode;
}
static struct nfs4_state *
list_for_each_entry(state, &nfsi->open_states, inode_states) {
/* Is this in the process of being freed? */
- if (state->nreaders == 0 && state->nwriters == 0)
+ if (state->state == 0)
continue;
if (state->owner == owner) {
atomic_inc(&state->count);
return NULL;
}
-struct nfs4_state *
-nfs4_find_state(struct inode *inode, struct rpc_cred *cred, mode_t mode)
-{
- struct nfs4_state *state;
-
- spin_lock(&inode->i_lock);
- state = __nfs4_find_state(inode, cred, mode);
- spin_unlock(&inode->i_lock);
- return state;
-}
-
static void
nfs4_free_open_state(struct nfs4_state *state)
{
spin_unlock(&inode->i_lock);
spin_unlock(&owner->so_lock);
iput(inode);
- BUG_ON (state->state != 0);
nfs4_free_open_state(state);
nfs4_put_state_owner(owner);
}
{
struct inode *inode = state->inode;
struct nfs4_state_owner *owner = state->owner;
- int newstate;
+ int oldstate, newstate = 0;
atomic_inc(&owner->so_count);
/* Protect against nfs4_find_state() */
spin_lock(&owner->so_lock);
spin_lock(&inode->i_lock);
- if (mode & FMODE_READ)
- state->nreaders--;
- if (mode & FMODE_WRITE)
- state->nwriters--;
- if (state->nwriters == 0) {
- if (state->nreaders == 0)
- list_del_init(&state->inode_states);
- /* See reclaim code */
- list_move_tail(&state->open_states, &owner->so_states);
+ switch (mode & (FMODE_READ | FMODE_WRITE)) {
+ case FMODE_READ:
+ state->n_rdonly--;
+ break;
+ case FMODE_WRITE:
+ state->n_wronly--;
+ break;
+ case FMODE_READ|FMODE_WRITE:
+ state->n_rdwr--;
+ }
+ oldstate = newstate = state->state;
+ if (state->n_rdwr == 0) {
+ if (state->n_rdonly == 0)
+ newstate &= ~FMODE_READ;
+ if (state->n_wronly == 0)
+ newstate &= ~FMODE_WRITE;
+ }
+ if (test_bit(NFS_DELEGATED_STATE, &state->flags)) {
+ nfs4_state_set_mode_locked(state, newstate);
+ oldstate = newstate;
}
spin_unlock(&inode->i_lock);
spin_unlock(&owner->so_lock);
- newstate = 0;
- if (state->state != 0) {
- if (state->nreaders)
- newstate |= FMODE_READ;
- if (state->nwriters)
- newstate |= FMODE_WRITE;
- if (state->state == newstate)
- goto out;
- if (test_bit(NFS_DELEGATED_STATE, &state->flags)) {
- state->state = newstate;
- goto out;
- }
- if (nfs4_do_close(inode, state, newstate) == 0)
- return;
- }
-out:
+
+ if (oldstate != newstate && nfs4_do_close(inode, state) == 0)
+ return;
nfs4_put_open_state(state);
nfs4_put_state_owner(owner);
}
static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
{
struct nfs4_lock_state *lsp;
- struct nfs4_client *clp = state->owner->so_client;
+ struct nfs_client *clp = state->owner->so_client;
lsp = kzalloc(sizeof(*lsp), GFP_KERNEL);
if (lsp == NULL)
struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter)
{
+ struct rpc_sequence *sequence = counter->sequence;
struct nfs_seqid *new;
new = kmalloc(sizeof(*new), GFP_KERNEL);
if (new != NULL) {
new->sequence = counter;
- INIT_LIST_HEAD(&new->list);
+ spin_lock(&sequence->lock);
+ list_add_tail(&new->list, &sequence->list);
+ spin_unlock(&sequence->lock);
}
return new;
}
{
struct rpc_sequence *sequence = seqid->sequence->sequence;
- if (!list_empty(&seqid->list)) {
- spin_lock(&sequence->lock);
- list_del(&seqid->list);
- spin_unlock(&sequence->lock);
- }
- rpc_wake_up_next(&sequence->wait);
+ spin_lock(&sequence->lock);
+ list_del(&seqid->list);
+ spin_unlock(&sequence->lock);
+ rpc_wake_up(&sequence->wait);
kfree(seqid);
}
if (sequence->list.next == &seqid->list)
goto out;
spin_lock(&sequence->lock);
- if (!list_empty(&sequence->list)) {
+ if (sequence->list.next != &seqid->list) {
rpc_sleep_on(&sequence->wait, task, NULL, NULL);
status = -EAGAIN;
- } else
- list_add(&seqid->list, &sequence->list);
+ }
spin_unlock(&sequence->lock);
out:
return status;
}
static int reclaimer(void *);
-struct reclaimer_args {
- struct nfs4_client *clp;
- struct completion complete;
-};
+
+static inline void nfs4_clear_recover_bit(struct nfs_client *clp)
+{
+ smp_mb__before_clear_bit();
+ clear_bit(NFS4CLNT_STATE_RECOVER, &clp->cl_state);
+ smp_mb__after_clear_bit();
+ wake_up_bit(&clp->cl_state, NFS4CLNT_STATE_RECOVER);
+ rpc_wake_up(&clp->cl_rpcwaitq);
+}
/*
* State recovery routine
*/
-void
-nfs4_recover_state(void *data)
+static void nfs4_recover_state(struct nfs_client *clp)
{
- struct nfs4_client *clp = (struct nfs4_client *)data;
- struct reclaimer_args args = {
- .clp = clp,
- };
- might_sleep();
-
- init_completion(&args.complete);
+ struct task_struct *task;
- if (kernel_thread(reclaimer, &args, CLONE_KERNEL) < 0)
- goto out_failed_clear;
- wait_for_completion(&args.complete);
- return;
-out_failed_clear:
- set_bit(NFS4CLNT_OK, &clp->cl_state);
- wake_up_all(&clp->cl_waitq);
- rpc_wake_up(&clp->cl_rpcwaitq);
+ __module_get(THIS_MODULE);
+ atomic_inc(&clp->cl_count);
+ task = kthread_run(reclaimer, clp, "%u.%u.%u.%u-reclaim",
+ NIPQUAD(clp->cl_addr.sin_addr));
+ if (!IS_ERR(task))
+ return;
+ nfs4_clear_recover_bit(clp);
+ nfs_put_client(clp);
+ module_put(THIS_MODULE);
}
/*
* Schedule a state recovery attempt
*/
-void
-nfs4_schedule_state_recovery(struct nfs4_client *clp)
+void nfs4_schedule_state_recovery(struct nfs_client *clp)
{
if (!clp)
return;
- if (test_and_clear_bit(NFS4CLNT_OK, &clp->cl_state))
- schedule_work(&clp->cl_recoverd);
+ if (test_and_set_bit(NFS4CLNT_STATE_RECOVER, &clp->cl_state) == 0)
+ nfs4_recover_state(clp);
}
static int nfs4_reclaim_locks(struct nfs4_state_recovery_ops *ops, struct nfs4_state *state)
int status = 0;
for (fl = inode->i_flock; fl != 0; fl = fl->fl_next) {
- if (!(fl->fl_flags & FL_POSIX))
+ if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
continue;
if (((struct nfs_open_context *)fl->fl_file->private_data)->state != state)
continue;
case -NFS4ERR_NO_GRACE:
case -NFS4ERR_RECLAIM_BAD:
case -NFS4ERR_RECLAIM_CONFLICT:
- /* kill_proc(fl->fl_owner, SIGLOST, 1); */
+ /* kill_proc(fl->fl_pid, SIGLOST, 1); */
break;
case -NFS4ERR_STALE_CLIENTID:
goto out_err;
return status;
}
-static void nfs4_state_mark_reclaim(struct nfs4_client *clp)
+static void nfs4_state_mark_reclaim(struct nfs_client *clp)
{
struct nfs4_state_owner *sp;
struct nfs4_state *state;
static int reclaimer(void *ptr)
{
- struct reclaimer_args *args = (struct reclaimer_args *)ptr;
- struct nfs4_client *clp = args->clp;
+ struct nfs_client *clp = ptr;
struct nfs4_state_owner *sp;
struct nfs4_state_recovery_ops *ops;
+ struct rpc_cred *cred;
int status = 0;
- daemonize("%u.%u.%u.%u-reclaim", NIPQUAD(clp->cl_addr));
allow_signal(SIGKILL);
- atomic_inc(&clp->cl_count);
- complete(&args->complete);
-
/* Ensure exclusive access to NFSv4 state */
lock_kernel();
down_write(&clp->cl_sem);
if (list_empty(&clp->cl_superblocks))
goto out;
restart_loop:
- status = nfs4_proc_renew(clp);
- switch (status) {
- case 0:
- case -NFS4ERR_CB_PATH_DOWN:
- goto out;
- case -NFS4ERR_STALE_CLIENTID:
- case -NFS4ERR_LEASE_MOVED:
- ops = &nfs4_reboot_recovery_ops;
- break;
- default:
- ops = &nfs4_network_partition_recovery_ops;
- };
+ ops = &nfs4_network_partition_recovery_ops;
+ /* Are there any open files on this volume? */
+ cred = nfs4_get_renew_cred(clp);
+ if (cred != NULL) {
+ /* Yes there are: try to renew the old lease */
+ status = nfs4_proc_renew(clp, cred);
+ switch (status) {
+ case 0:
+ case -NFS4ERR_CB_PATH_DOWN:
+ put_rpccred(cred);
+ goto out;
+ case -NFS4ERR_STALE_CLIENTID:
+ case -NFS4ERR_LEASE_MOVED:
+ ops = &nfs4_reboot_recovery_ops;
+ }
+ } else {
+ /* "reboot" to ensure we clear all state on the server */
+ clp->cl_boot_time = CURRENT_TIME;
+ cred = nfs4_get_setclientid_cred(clp);
+ }
+ /* We're going to have to re-establish a clientid */
nfs4_state_mark_reclaim(clp);
- status = __nfs4_init_client(clp);
+ status = -ENOENT;
+ if (cred != NULL) {
+ status = nfs4_init_client(clp, cred);
+ put_rpccred(cred);
+ }
if (status)
goto out_error;
/* Mark all delegations for reclaim */
}
nfs_delegation_reap_unclaimed(clp);
out:
- set_bit(NFS4CLNT_OK, &clp->cl_state);
up_write(&clp->cl_sem);
unlock_kernel();
- wake_up_all(&clp->cl_waitq);
- rpc_wake_up(&clp->cl_rpcwaitq);
if (status == -NFS4ERR_CB_PATH_DOWN)
nfs_handle_cb_pathdown(clp);
- nfs4_put_client(clp);
+ nfs4_clear_recover_bit(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;
}