recovery: remember byte-range locks and reclaim during recovery

nfs41_open_state maintains a list of outstanding byte-range locks by calling open_lock_add() and open_lock_remove() in lock.c

during client state recovery, after reclaiming each OPEN stateid, send LOCK requests with reclaim=TRUE for each lock it owns, and update the open's lock stateid with the result

added 'bool_t reclaim' argument to nfs41_lock(); when set, compound_encode_send_decode() is called with try_recovery=FALSE to avoid recursive recovery

Signed-off-by: Casey Bodley <cbodley@citi.umich.edu>
This commit is contained in:
Casey Bodley 2010-11-18 15:11:36 -05:00 committed by unknown
parent 1906610544
commit 222c1bf020
7 changed files with 141 additions and 56 deletions

View file

@ -37,51 +37,82 @@ void nfs41_lock_stateid_arg(
IN nfs41_open_state *state,
OUT stateid_arg *arg)
{
AcquireSRWLockShared(&state->last_lock.lock);
if (state->last_lock.initialized) {
AcquireSRWLockShared(&state->lock);
if (state->locks.stateid.seqid) {
/* use lock stateid where available */
memcpy(&arg->stateid, &state->last_lock.stateid, sizeof(stateid4));
memcpy(&arg->stateid, &state->locks.stateid, sizeof(stateid4));
arg->type = STATEID_LOCK;
ReleaseSRWLockShared(&state->last_lock.lock);
arg->open = state;
} else {
ReleaseSRWLockShared(&state->last_lock.lock);
/* fall back on open stateid */
nfs41_open_stateid_arg(state, arg);
}
ReleaseSRWLockShared(&state->lock);
}
static void update_last_lock_state(
OUT nfs41_lock_state *lock_state,
IN stateid4 *stateid)
/* expects the caller to hold an exclusive lock on nfs41_open_state.lock */
static void update_lock_state(
OUT nfs41_open_state *state,
IN const stateid4 *stateid)
{
/* update the lock state if the seqid is more recent */
AcquireSRWLockShared(&lock_state->lock);
if (stateid->seqid > lock_state->stateid.seqid) {
ReleaseSRWLockShared(&lock_state->lock);
if (state->locks.stateid.seqid == 0) {
/* if it's a new lock stateid, copy it in */
memcpy(&state->locks.stateid, stateid, sizeof(stateid4));
} else if (stateid->seqid > state->locks.stateid.seqid) {
/* update the seqid if it's more recent */
state->locks.stateid.seqid = stateid->seqid;
}
}
AcquireSRWLockExclusive(&lock_state->lock);
if (stateid->seqid > lock_state->stateid.seqid) {
if (lock_state->initialized) {
/* if the lock state already existed, update the seqid only;
* assume that stateid->other remains unchanged */
dprintf(LKLVL, "update_last_lock_state: setting seqid=%u "
"(was %u)\n", stateid->seqid, lock_state->stateid.seqid);
lock_state->stateid.seqid = stateid->seqid;
} else {
/* copy the entire stateid and mark as initialized */
dprintf(LKLVL, "update_last_lock_state: stateid "
"initialized with seqid=%u\n", stateid->seqid);
memcpy(&lock_state->stateid, stateid, sizeof(stateid4));
lock_state->initialized = 1;
static int open_lock_add(
IN nfs41_open_state *state,
IN const stateid4 *stateid,
IN uint64_t offset,
IN uint64_t length,
IN uint32_t type)
{
nfs41_lock_state *lock;
int status = NO_ERROR;
AcquireSRWLockExclusive(&state->lock);
update_lock_state(state, stateid);
lock = malloc(sizeof(nfs41_lock_state));
if (lock == NULL) {
status = GetLastError();
goto out;
}
lock->offset = offset;
lock->length = length;
lock->type = type;
list_add_tail(&state->locks.list, &lock->open_entry);
out:
ReleaseSRWLockExclusive(&state->lock);
return status;
}
static void open_lock_remove(
IN nfs41_open_state *state,
IN const stateid4 *stateid,
IN uint64_t offset,
IN uint64_t length)
{
struct list_entry *entry;
nfs41_lock_state *lock;
AcquireSRWLockExclusive(&state->lock);
update_lock_state(state, stateid);
list_for_each(entry, &state->locks.list) {
lock = list_container(entry, nfs41_lock_state, open_entry);
if (lock->offset == offset && lock->length == length) {
list_remove(entry);
free(lock);
break;
}
}
ReleaseSRWLockExclusive(&lock_state->lock);
} else {
dprintf(LKLVL, "update_last_lock_state: discarding seqid=%u "
"(already %u)\n", stateid->seqid, lock_state->stateid.seqid);
ReleaseSRWLockShared(&lock_state->lock);
}
ReleaseSRWLockExclusive(&state->lock);
}
@ -132,7 +163,7 @@ static int handle_lock(nfs41_upcall *upcall)
nfs41_lock_stateid_arg(state, &stateid);
status = nfs41_lock(state->session, &state->file, &state->owner,
type, args->offset, args->length, &stateid);
type, args->offset, args->length, 0, &stateid);
if (status) {
dprintf(LKLVL, "nfs41_lock failed with %s\n",
nfs_error_string(status));
@ -140,7 +171,9 @@ static int handle_lock(nfs41_upcall *upcall)
goto out;
}
update_last_lock_state(&state->last_lock, &stateid.stateid);
/* ignore errors from open_lock_add(); they just mean we
* won't be able to recover the lock after reboot */
open_lock_add(state, &stateid.stateid, args->offset, args->length, type);
out:
return status;
}
@ -168,7 +201,7 @@ static void cancel_lock(IN nfs41_upcall *upcall)
goto out;
}
update_last_lock_state(&state->last_lock, &stateid.stateid);
open_lock_remove(state, &stateid.stateid, args->offset, args->length);
out:
dprintf(1, "<-- cancel_lock() returning %d\n", status);
}
@ -208,7 +241,7 @@ static int handle_unlock(nfs41_upcall *upcall)
uint32_t buf_len = args->buf_len;
uint64_t offset;
uint64_t length;
int status;
int status = NO_ERROR;
nfs41_lock_stateid_arg(state, &stateid);
if (stateid.type != STATEID_LOCK) {
@ -217,7 +250,6 @@ static int handle_unlock(nfs41_upcall *upcall)
goto out;
}
status = NO_ERROR;
for (i = 0; i < args->count; i++) {
if (safe_read(&buf, &buf_len, &offset, sizeof(LONGLONG))) break;
if (safe_read(&buf, &buf_len, &length, sizeof(LONGLONG))) break;
@ -225,6 +257,7 @@ static int handle_unlock(nfs41_upcall *upcall)
status = nfs41_unlock(state->session,
&state->file, offset, length, &stateid);
if (status == NFS4_OK) {
open_lock_remove(state, &stateid.stateid, offset, length);
nsuccess++;
} else {
dprintf(LKLVL, "nfs41_unlock failed with %s\n",
@ -234,7 +267,7 @@ static int handle_unlock(nfs41_upcall *upcall)
}
if (nsuccess) {
update_last_lock_state(&state->last_lock, &stateid.stateid);
update_lock_state(state, &stateid.stateid);
status = NO_ERROR;
}
out:

View file

@ -68,9 +68,10 @@ typedef struct __nfs41_server {
} nfs41_server;
typedef struct __nfs41_lock_state {
bool_t initialized;
stateid4 stateid;
SRWLOCK lock;
struct list_entry open_entry; /* entry in nfs41_open_state.locks */
uint64_t offset;
uint64_t length;
uint32_t type;
} nfs41_lock_state;
/* nfs41_open_state reference counting:
@ -89,13 +90,17 @@ typedef struct __nfs41_open_state {
bool_t do_close;
stateid4 stateid;
state_owner4 owner;
nfs41_lock_state last_lock;
struct __pnfs_file_layout *layout;
struct list_entry client_entry; /* entry in nfs41_client.opens */
SRWLOCK lock;
LONG ref_count;
uint32_t share_access;
uint32_t share_deny;
struct { /* list of open lock state for recovery */
stateid4 stateid;
struct list_entry list;
} locks;
} nfs41_open_state;
typedef struct __nfs41_rpc_clnt {

View file

@ -127,18 +127,39 @@ static int recover_open(
IN nfs41_session *session,
IN nfs41_open_state *open)
{
stateid4 stateid;
stateid_arg stateid;
struct list_entry *entry;
nfs41_lock_state *lock;
int status;
/* reclaim the open stateid */
status = nfs41_open_reclaim(session, &open->parent, &open->file,
&open->owner, open->share_access, open->share_deny, &stateid);
&open->owner, open->share_access, open->share_deny, &stateid.stateid);
if (status)
goto out;
AcquireSRWLockExclusive(&open->lock);
/* update the open stateid */
memcpy(&open->stateid, &stateid, sizeof(stateid4));
memcpy(&open->stateid, &stateid.stateid, sizeof(stateid4));
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, &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));
}
ReleaseSRWLockExclusive(&open->lock);
out:
return status;
@ -295,6 +316,7 @@ restart_recovery:
nfs_argop4 *argop = &compound->args.argarray[
compound->res.resarray_count-1];
stateid_arg *stateid = NULL;
stateid4 *source = NULL;
if (argop->op == OP_CLOSE) {
nfs41_op_close_args *close = (nfs41_op_close_args*)argop->arg;
@ -327,6 +349,7 @@ restart_recovery:
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)
@ -334,14 +357,20 @@ restart_recovery:
&session->client->recovery.lock, INFINITE);
LeaveCriticalSection(&session->client->recovery.lock);
/* if the open stateid is different, update and retry */
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, &stateid->open->stateid, sizeof(stateid4))) {
memcpy(&stateid->stateid, &stateid->open->stateid, sizeof(stateid4));
if (memcmp(&stateid->stateid, source, sizeof(stateid4))) {
memcpy(&stateid->stateid, source, sizeof(stateid4));
retry = TRUE;
}
ReleaseSRWLockShared(&stateid->open->lock);
break;
default:
eprintf("%s returned %s: can't recover stateid type %u\n",
nfs_opnum_to_string(argop->op),

View file

@ -866,6 +866,7 @@ int nfs41_lock(
IN uint32_t type,
IN uint64_t offset,
IN uint64_t length,
IN bool_t reclaim,
IN OUT stateid_arg *stateid)
{
int status;
@ -892,7 +893,7 @@ int nfs41_lock(
compound_add_op(&compound, OP_LOCK, &lock_args, &lock_res);
lock_args.locktype = type;
lock_args.reclaim = 0;
lock_args.reclaim = reclaim;
lock_args.offset = offset;
lock_args.length = length;
if (stateid->type == STATEID_LOCK) {
@ -909,11 +910,14 @@ int nfs41_lock(
lock_res.u.resok4.lock_stateid = &stateid->stateid;
lock_res.u.denied.owner.owner_len = NFS4_OPAQUE_LIMIT;
status = compound_encode_send_decode(session, &compound, TRUE);
status = compound_encode_send_decode(session, &compound, !reclaim);
if (status)
goto out;
compound_error(status = compound.res.status);
if (compound_error(status = compound.res.status))
goto out;
stateid->type = STATEID_LOCK; /* returning a lock stateid */
out:
return status;
}

View file

@ -975,6 +975,7 @@ int nfs41_lock(
IN uint32_t type,
IN uint64_t offset,
IN uint64_t length,
IN bool_t reclaim,
IN OUT stateid_arg *stateid);
int nfs41_unlock(

View file

@ -62,6 +62,7 @@ static int create_open_state(
state->owner.owner_len = (uint32_t)strlen(
(const char*)state->owner.owner);
state->ref_count = 1;
list_init(&state->locks.list);
*state_out = state;
status = NO_ERROR;
@ -73,6 +74,18 @@ out_free:
goto out;
}
static void open_state_free(
IN nfs41_open_state *state)
{
struct list_entry *entry, *tmp;
/* free associated lock state */
list_for_each_tmp(entry, tmp, &state->locks.list)
free(list_container(entry, nfs41_lock_state, open_entry));
free(state);
}
/* open state reference counting */
void nfs41_open_state_ref(
IN nfs41_open_state *state)
@ -91,7 +104,7 @@ void nfs41_open_state_deref(
dprintf(2, "nfs41_open_state_deref(%s) count %d\n",
state->path.path, count);
if (count == 0)
free(state);
open_state_free(state);
}
void nfs41_open_stateid_arg(

View file

@ -551,7 +551,7 @@ enum pnfs_status pnfs_open_state_layout(
goto out;
}
nfs41_open_stateid_arg(state, &stateid);
nfs41_lock_stateid_arg(state, &stateid);
/* make sure the layout can satisfy this request */
status = file_layout_get(layout, session, &state->file,