cosmetic: recovery.c for client state recovery
recovery.h exposes the following functions for nfs41_compound.c: nfs41_recovery_start_or_wait() nfs41_recovery_finish() nfs41_recover_client_state() nfs41_recover_stateid() Signed-off-by: Casey Bodley <cbodley@citi.umich.edu>
This commit is contained in:
Casey Bodley
unknown
parent
316dfe568a
commit
8ef3ec9247
5 changed files with 312 additions and 228 deletions
|
|
@ -27,7 +27,7 @@
|
|||
#include "nfs41_compound.h"
|
||||
#include "nfs41_xdr.h"
|
||||
#include "nfs41_ops.h"
|
||||
#include "nfs41_callback.h"
|
||||
#include "recovery.h"
|
||||
#include "name_cache.h"
|
||||
#include "daemon_debug.h"
|
||||
#include "rpc/rpc.h"
|
||||
|
|
@ -92,215 +92,6 @@ static void set_expected_res(
|
|||
compound->res.resarray[i].op = compound->args.argarray[i].op;
|
||||
}
|
||||
|
||||
/* session/client recovery uses a lock and condition variable in nfs41_client
|
||||
* to prevent multiple threads from attempting to recover at the same time */
|
||||
static bool_t recovery_start_or_wait(
|
||||
IN nfs41_client *client)
|
||||
{
|
||||
bool_t status = TRUE;
|
||||
|
||||
EnterCriticalSection(&client->recovery.lock);
|
||||
|
||||
if (!client->recovery.in_recovery) {
|
||||
dprintf(1, "Entering recovery mode for client %llu\n", client->clnt_id);
|
||||
client->recovery.in_recovery = TRUE;
|
||||
} else {
|
||||
status = FALSE;
|
||||
dprintf(1, "Waiting for recovery of client %llu\n", client->clnt_id);
|
||||
while (client->recovery.in_recovery)
|
||||
SleepConditionVariableCS(&client->recovery.cond,
|
||||
&client->recovery.lock, INFINITE);
|
||||
dprintf(1, "Woke up after recovery of client %llu\n", client->clnt_id);
|
||||
}
|
||||
|
||||
LeaveCriticalSection(&client->recovery.lock);
|
||||
return status;
|
||||
}
|
||||
|
||||
static void recovery_finish(
|
||||
IN nfs41_client *client)
|
||||
{
|
||||
EnterCriticalSection(&client->recovery.lock);
|
||||
dprintf(1, "Finished recovery for client %llu\n", client->clnt_id);
|
||||
client->recovery.in_recovery = FALSE;
|
||||
WakeAllConditionVariable(&client->recovery.cond);
|
||||
LeaveCriticalSection(&client->recovery.lock);
|
||||
}
|
||||
|
||||
static int recover_open(
|
||||
IN nfs41_session *session,
|
||||
IN nfs41_open_state *open)
|
||||
{
|
||||
open_claim4 claim;
|
||||
open_delegation4 delegation;
|
||||
stateid_arg stateid;
|
||||
struct list_entry *entry;
|
||||
nfs41_lock_state *lock;
|
||||
int status;
|
||||
|
||||
/* reclaim the open stateid */
|
||||
claim.claim = CLAIM_PREVIOUS;
|
||||
claim.u.prev.delegate_type = OPEN_DELEGATE_NONE;
|
||||
|
||||
status = nfs41_rpc_open(session, &open->parent, &open->file,
|
||||
&open->owner, &claim, open->share_access, open->share_deny,
|
||||
OPEN4_NOCREATE, 0, 0, FALSE, &stateid.stateid, &delegation, NULL);
|
||||
|
||||
if (status == NFS4_OK) {
|
||||
/* update the open stateid on success */
|
||||
memcpy(&open->stateid, &stateid.stateid, sizeof(stateid4));
|
||||
|
||||
} else if (status == NFS4ERR_NO_GRACE) {
|
||||
dprintf(1, "not in grace period, retrying a normal open\n");
|
||||
status = nfs41_open(open, OPEN4_NOCREATE, 0, 0, FALSE, NULL);
|
||||
|
||||
/* update the stateid arg with the new open->stateid */
|
||||
memcpy(&stateid.stateid, &open->stateid, sizeof(stateid4));
|
||||
}
|
||||
if (status)
|
||||
goto out;
|
||||
|
||||
AcquireSRWLockExclusive(&open->lock);
|
||||
|
||||
open->layout = NULL;
|
||||
stateid.type = STATEID_OPEN;
|
||||
stateid.open = open;
|
||||
|
||||
/* recover any locks for this open */
|
||||
list_for_each(entry, &open->locks.list) {
|
||||
lock = list_container(entry, nfs41_lock_state, open_entry);
|
||||
status = nfs41_lock(session, &open->file, &open->owner,
|
||||
lock->type, lock->offset, lock->length, TRUE, FALSE, &stateid);
|
||||
if (status == NFS4ERR_NO_GRACE) {
|
||||
dprintf(1, "not in grace period, retrying a normal lock\n");
|
||||
status = nfs41_lock(session, &open->file, &open->owner,
|
||||
lock->type, lock->offset, lock->length, FALSE, FALSE, &stateid);
|
||||
}
|
||||
if (status == NFS4ERR_BADSESSION)
|
||||
break;
|
||||
}
|
||||
|
||||
if (status != NFS4ERR_BADSESSION) {
|
||||
/* if we got a lock stateid back, save the lock with the open */
|
||||
if (stateid.type == STATEID_LOCK)
|
||||
memcpy(&open->locks.stateid, &stateid.stateid, sizeof(stateid4));
|
||||
else
|
||||
open->locks.stateid.seqid = 0;
|
||||
}
|
||||
|
||||
ReleaseSRWLockExclusive(&open->lock);
|
||||
out:
|
||||
return status;
|
||||
}
|
||||
|
||||
static int recover_client_state(
|
||||
IN nfs41_session *session,
|
||||
IN nfs41_client *client)
|
||||
{
|
||||
const struct cb_layoutrecall_args recall = { PNFS_LAYOUTTYPE_FILE,
|
||||
PNFS_IOMODE_ANY, TRUE, { PNFS_RETURN_ALL } };
|
||||
struct client_state *state = &session->client->state;
|
||||
struct list_entry *entry;
|
||||
nfs41_open_state *open;
|
||||
int status = NFS4_OK;
|
||||
|
||||
/* recover each of the client's opens */
|
||||
EnterCriticalSection(&state->lock);
|
||||
list_for_each(entry, &state->opens) {
|
||||
open = list_container(entry, nfs41_open_state, client_entry);
|
||||
status = recover_open(session, open);
|
||||
if (status == NFS4ERR_BADSESSION)
|
||||
break;
|
||||
}
|
||||
LeaveCriticalSection(&state->lock);
|
||||
|
||||
/* revoke all of the client's layouts */
|
||||
pnfs_file_layout_recall(client, &recall);
|
||||
|
||||
if (status != NFS4ERR_BADSESSION) {
|
||||
/* send reclaim_complete, but don't fail on errors */
|
||||
status = nfs41_reclaim_complete(session);
|
||||
if (status && status == NFS4ERR_NOTSUPP)
|
||||
eprintf("nfs41_reclaim_complete() failed with %s\n",
|
||||
nfs_error_string(status));
|
||||
}
|
||||
return status;
|
||||
}
|
||||
|
||||
static bool_t recover_stateid(nfs_argop4 *argop, nfs41_session *session)
|
||||
{
|
||||
stateid_arg *stateid = NULL;
|
||||
stateid4 *source = NULL;
|
||||
bool_t retry = FALSE;
|
||||
|
||||
if (argop->op == OP_CLOSE) {
|
||||
nfs41_op_close_args *close = (nfs41_op_close_args*)argop->arg;
|
||||
stateid = close->stateid;
|
||||
} else if (argop->op == OP_READ) {
|
||||
nfs41_read_args *read = (nfs41_read_args*)argop->arg;
|
||||
stateid = read->stateid;
|
||||
} else if (argop->op == OP_WRITE) {
|
||||
nfs41_write_args *write = (nfs41_write_args*)argop->arg;
|
||||
stateid = write->stateid;
|
||||
} else if (argop->op == OP_LOCK) {
|
||||
nfs41_lock_args *lock = (nfs41_lock_args*)argop->arg;
|
||||
if (lock->locker.new_lock_owner)
|
||||
stateid = lock->locker.u.open_owner.open_stateid;
|
||||
else
|
||||
stateid = lock->locker.u.lock_owner.lock_stateid;
|
||||
} else if (argop->op == OP_LOCKU) {
|
||||
nfs41_locku_args *locku = (nfs41_locku_args*)argop->arg;
|
||||
stateid = locku->lock_stateid;
|
||||
} else if (argop->op == OP_SETATTR) {
|
||||
nfs41_setattr_args *setattr = (nfs41_setattr_args*)argop->arg;
|
||||
stateid = setattr->stateid;
|
||||
} else if (argop->op == OP_LAYOUTGET) {
|
||||
pnfs_layoutget_args *lget = (pnfs_layoutget_args*)argop->arg;
|
||||
stateid = lget->stateid;
|
||||
}
|
||||
|
||||
if (stateid) {
|
||||
switch (stateid->type) {
|
||||
case STATEID_OPEN:
|
||||
case STATEID_LOCK:
|
||||
/* if there's recovery in progress, wait for it to finish */
|
||||
EnterCriticalSection(&session->client->recovery.lock);
|
||||
while (session->client->recovery.in_recovery)
|
||||
SleepConditionVariableCS(&session->client->recovery.cond,
|
||||
&session->client->recovery.lock, INFINITE);
|
||||
LeaveCriticalSection(&session->client->recovery.lock);
|
||||
|
||||
if (stateid->type == STATEID_OPEN)
|
||||
source = &stateid->open->stateid;
|
||||
else
|
||||
source = &stateid->open->locks.stateid;
|
||||
|
||||
/* if the source stateid is different, update and retry */
|
||||
AcquireSRWLockShared(&stateid->open->lock);
|
||||
if (memcmp(&stateid->stateid, source, sizeof(stateid4))) {
|
||||
/* if it was a lock stateid that was cleared, resend it with an open stateid */
|
||||
if (argop->op == OP_LOCK && stateid->type == STATEID_LOCK && source->seqid == 0) {
|
||||
nfs41_lock_args *lock = (nfs41_lock_args*)argop->arg;
|
||||
lock->locker.new_lock_owner = 1;
|
||||
lock->locker.u.open_owner.open_stateid = stateid;
|
||||
lock->locker.u.open_owner.lock_owner = &stateid->open->owner;
|
||||
source = &stateid->open->stateid;
|
||||
}
|
||||
|
||||
memcpy(&stateid->stateid, source, sizeof(stateid4));
|
||||
retry = TRUE;
|
||||
}
|
||||
ReleaseSRWLockShared(&stateid->open->lock);
|
||||
break;
|
||||
|
||||
default:
|
||||
eprintf("%s can't recover stateid type %u\n",
|
||||
nfs_opnum_to_string(argop->op), stateid->type);
|
||||
break;
|
||||
}
|
||||
}
|
||||
return retry;
|
||||
}
|
||||
|
||||
static int create_new_rpc_auth(nfs41_session *session, uint32_t op,
|
||||
nfs41_secinfo_info *secinfo)
|
||||
|
|
@ -416,12 +207,12 @@ retry:
|
|||
case NFS4ERR_STALE_CLIENTID:
|
||||
if (!try_recovery)
|
||||
goto out;
|
||||
if (!recovery_start_or_wait(session->client))
|
||||
if (!nfs41_recovery_start_or_wait(session->client))
|
||||
goto do_retry;
|
||||
//try to create a new client
|
||||
status = nfs41_client_renew(session->client);
|
||||
|
||||
recovery_finish(session->client);
|
||||
nfs41_recovery_finish(session->client);
|
||||
if (status) {
|
||||
eprintf("nfs41_exchange_id() failed with %d\n", status);
|
||||
status = ERROR_BAD_NET_RESP;
|
||||
|
|
@ -440,7 +231,7 @@ retry:
|
|||
case NFS4ERR_BADSESSION:
|
||||
if (!try_recovery)
|
||||
goto out;
|
||||
if (!recovery_start_or_wait(session->client))
|
||||
if (!nfs41_recovery_start_or_wait(session->client))
|
||||
goto do_retry;
|
||||
restart_recovery:
|
||||
//try to create a new session
|
||||
|
|
@ -451,22 +242,22 @@ restart_recovery:
|
|||
if (status) {
|
||||
eprintf("nfs41_exchange_id() failed with %d\n", status);
|
||||
status = ERROR_BAD_NET_RESP;
|
||||
recovery_finish(session->client);
|
||||
nfs41_recovery_finish(session->client);
|
||||
goto out;
|
||||
}
|
||||
goto restart_recovery;
|
||||
} else if (status) {
|
||||
eprintf("nfs41_session_renew: failed with %d\n", status);
|
||||
recovery_finish(session->client);
|
||||
nfs41_recovery_finish(session->client);
|
||||
goto out;
|
||||
}
|
||||
if (client_state_lost) {
|
||||
/* do client state recovery */
|
||||
status = recover_client_state(session, session->client);
|
||||
status = nfs41_recover_client_state(session, session->client);
|
||||
if (status == NFS4ERR_BADSESSION)
|
||||
goto restart_recovery;
|
||||
}
|
||||
recovery_finish(session->client);
|
||||
nfs41_recovery_finish(session->client);
|
||||
goto do_retry;
|
||||
|
||||
case NFS4ERR_EXPIRED: /* revoked by lease expiration */
|
||||
|
|
@ -477,11 +268,8 @@ restart_recovery:
|
|||
compound->args.argarray[0].arg;
|
||||
nfs41_session_free_slot(session, seq->sa_slotid);
|
||||
}
|
||||
if (!try_recovery)
|
||||
goto out;
|
||||
|
||||
if (recover_stateid(&compound->args.argarray[compound->res.resarray_count-1],
|
||||
session))
|
||||
if (try_recovery && nfs41_recover_stateid(session,
|
||||
&compound->args.argarray[compound->res.resarray_count-1]))
|
||||
goto do_retry;
|
||||
goto out;
|
||||
|
||||
|
|
@ -583,7 +371,7 @@ restart_recovery:
|
|||
}
|
||||
if (!try_recovery)
|
||||
goto out;
|
||||
if (!recovery_start_or_wait(session->client))
|
||||
if (!nfs41_recovery_start_or_wait(session->client))
|
||||
goto do_retry;
|
||||
ZeroMemory(secinfo, sizeof(nfs41_secinfo_info)*MAX_SECINFOS);
|
||||
saved_sec_flavor = session->client->rpc->sec_flavor;
|
||||
|
|
@ -604,7 +392,7 @@ restart_recovery:
|
|||
file = &tmp;
|
||||
}
|
||||
else {
|
||||
recovery_finish(session->client);
|
||||
nfs41_recovery_finish(session->client);
|
||||
goto out;
|
||||
}
|
||||
|
||||
|
|
@ -620,7 +408,7 @@ restart_recovery:
|
|||
secinfo_status = nfs41_secinfo(session, file, name, secinfo);
|
||||
if (secinfo_status) {
|
||||
eprintf("nfs41_secinfo failed with %d\n", secinfo_status);
|
||||
recovery_finish(session->client);
|
||||
nfs41_recovery_finish(session->client);
|
||||
if (secinfo_status == NFS4ERR_BADSESSION) {
|
||||
if (compound->args.argarray[0].op == OP_SEQUENCE) {
|
||||
nfs41_sequence_args *seq =
|
||||
|
|
@ -642,7 +430,7 @@ restart_recovery:
|
|||
secinfo_status = nfs41_secinfo_noname(session, file, secinfo);
|
||||
if (secinfo_status) {
|
||||
eprintf("nfs41_secinfo_noname failed with %d\n", secinfo_status);
|
||||
recovery_finish(session->client);
|
||||
nfs41_recovery_finish(session->client);
|
||||
if (compound->args.argarray[0].op == OP_SEQUENCE) {
|
||||
nfs41_sequence_args *seq =
|
||||
(nfs41_sequence_args *)compound->args.argarray[0].arg;
|
||||
|
|
@ -654,14 +442,14 @@ restart_recovery:
|
|||
secinfo_status = create_new_rpc_auth(session, op, secinfo);
|
||||
if (!secinfo_status) {
|
||||
auth_destroy(saved_auth);
|
||||
recovery_finish(session->client);
|
||||
nfs41_recovery_finish(session->client);
|
||||
// Need to retry only
|
||||
goto do_retry;
|
||||
} else {
|
||||
AcquireSRWLockExclusive(&session->client->rpc->lock);
|
||||
session->client->rpc->rpc->cl_auth = saved_auth;
|
||||
ReleaseSRWLockExclusive(&session->client->rpc->lock);
|
||||
recovery_finish(session->client);
|
||||
nfs41_recovery_finish(session->client);
|
||||
}
|
||||
break;
|
||||
}
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue