return ret;
}
-/*
+/*
* ocfs2_dio_end_io is called by the dio core when a dio is finished. We're
* particularly interested in the aio/dio case. Like the core uses
* i_alloc_sem, we use the rw_lock DLM lock to protect io on one node from
ret = blockdev_direct_IO_no_locking(rw, iocb, inode,
inode->i_sb->s_bdev, iov, offset,
- nr_segs,
+ nr_segs,
ocfs2_direct_IO_get_blocks,
ocfs2_dio_end_io);
}
ocfs2_metadata_cache_io_unlock(ci);
- mlog(ML_BH_IO, "block=(%llu), nr=(%d), cached=%s, flags=0x%x\n",
+ mlog(ML_BH_IO, "block=(%llu), nr=(%d), cached=%s, flags=0x%x\n",
(unsigned long long)block, nr,
((flags & OCFS2_BH_IGNORE_CACHE) || ignore_cache) ? "no" : "yes",
flags);
unsigned int o2hb_dead_threshold = O2HB_DEFAULT_DEAD_THRESHOLD;
-/* Only sets a new threshold if there are no active regions.
+/* Only sets a new threshold if there are no active regions.
*
* No locking or otherwise interesting code is required for reading
* o2hb_dead_threshold as it can't change once regions are active and
mlog(ML_ERROR, "Heartbeat write timeout to device %s after %u "
"milliseconds\n", reg->hr_dev_name,
- jiffies_to_msecs(jiffies - reg->hr_last_timeout_start));
+ jiffies_to_msecs(jiffies - reg->hr_last_timeout_start));
o2quo_disk_timeout();
}
"seq %llu last %llu changed %u equal %u\n",
slot->ds_node_num, (long long)slot->ds_last_generation,
le32_to_cpu(hb_block->hb_cksum),
- (unsigned long long)le64_to_cpu(hb_block->hb_seq),
+ (unsigned long long)le64_to_cpu(hb_block->hb_seq),
(unsigned long long)slot->ds_last_time, slot->ds_changed_samples,
slot->ds_equal_samples);
}
if (was_valid && !valid) {
- printk(KERN_INFO "o2net: no longer connected to "
+ printk(KERN_NOTICE "o2net: no longer connected to "
SC_NODEF_FMT "\n", SC_NODEF_ARGS(old_sc));
o2net_complete_nodes_nsw(nn);
}
if (!was_valid && valid) {
o2quo_conn_up(o2net_num_from_nn(nn));
cancel_delayed_work(&nn->nn_connect_expired);
- printk(KERN_INFO "o2net: %s " SC_NODEF_FMT "\n",
+ printk(KERN_NOTICE "o2net: %s " SC_NODEF_FMT "\n",
o2nm_this_node() > sc->sc_node->nd_num ?
"connected to" : "accepted connection from",
SC_NODEF_ARGS(sc));
cond_resched();
continue;
}
- mlog(ML_ERROR, "sendpage of size %zu to " SC_NODEF_FMT
+ mlog(ML_ERROR, "sendpage of size %zu to " SC_NODEF_FMT
" failed with %zd\n", size, SC_NODEF_ARGS(sc), ret);
o2net_ensure_shutdown(nn, sc, 0);
break;
do_gettimeofday(&now);
- printk(KERN_INFO "o2net: connection to " SC_NODEF_FMT " has been idle for %u.%u "
+ printk(KERN_NOTICE "o2net: connection to " SC_NODEF_FMT " has been idle for %u.%u "
"seconds, shutting it down.\n", SC_NODEF_ARGS(sc),
o2net_idle_timeout() / 1000,
o2net_idle_timeout() % 1000);
mlog(ML_NOTICE, "here are some times that might help debug the "
"situation: (tmr %ld.%ld now %ld.%ld dr %ld.%ld adv "
"%ld.%ld:%ld.%ld func (%08x:%u) %ld.%ld:%ld.%ld)\n",
- sc->sc_tv_timer.tv_sec, (long) sc->sc_tv_timer.tv_usec,
+ sc->sc_tv_timer.tv_sec, (long) sc->sc_tv_timer.tv_usec,
now.tv_sec, (long) now.tv_usec,
sc->sc_tv_data_ready.tv_sec, (long) sc->sc_tv_data_ready.tv_usec,
sc->sc_tv_advance_start.tv_sec,
* on their number */
#define O2NET_QUORUM_DELAY_MS ((o2hb_dead_threshold + 2) * O2HB_REGION_TIMEOUT_MS)
-/*
+/*
* This version number represents quite a lot, unfortunately. It not
* only represents the raw network message protocol on the wire but also
- * locking semantics of the file system using the protocol. It should
+ * locking semantics of the file system using the protocol. It should
* be somewhere else, I'm sure, but right now it isn't.
*
* With version 11, we separate out the filesystem locking portion. The
mlog(ML_ERROR, "dlm status = %s\n", dlm_errname((st))); \
} while (0)
-#define DLM_LKSB_UNUSED1 0x01
+#define DLM_LKSB_UNUSED1 0x01
#define DLM_LKSB_PUT_LVB 0x02
#define DLM_LKSB_GET_LVB 0x04
#define DLM_LKSB_UNUSED2 0x08
dlm_lock_put(lock);
/* free up the reserved bast that we are cancelling.
* guaranteed that this will not be the last reserved
- * ast because *both* an ast and a bast were reserved
+ * ast because *both* an ast and a bast were reserved
* to get to this point. the res->spinlock will not be
* taken here */
dlm_lockres_release_ast(dlm, res);
/* instead of logging the same network error over
* and over, sleep here and wait for the heartbeat
* to notice the node is dead. times out after 5s. */
- dlm_wait_for_node_death(dlm, res->owner,
+ dlm_wait_for_node_death(dlm, res->owner,
DLM_NODE_DEATH_WAIT_MAX);
ret = DLM_RECOVERING;
mlog(0, "node %u died so returning DLM_RECOVERING "
assert_spin_locked(&res->spinlock);
stringify_lockname(res->lockname.name, res->lockname.len,
- buf, sizeof(buf) - 1);
+ buf, sizeof(buf));
printk("lockres: %s, owner=%u, state=%u\n",
buf, res->owner, res->state);
printk(" last used: %lu, refcnt: %u, on purge list: %s\n",
}
/* Once the dlm ctxt is marked as leaving then we don't want
- * to be put in someone's domain map.
+ * to be put in someone's domain map.
* Also, explicitly disallow joining at certain troublesome
* times (ie. during recovery). */
if (dlm && dlm->dlm_state != DLM_CTXT_LEAVING) {
}
dlm_revert_pending_lock(res, lock);
dlm_lock_put(lock);
- } else if (dlm_is_recovery_lock(res->lockname.name,
+ } else if (dlm_is_recovery_lock(res->lockname.name,
res->lockname.len)) {
/* special case for the $RECOVERY lock.
* there will never be an AST delivered to put
struct dlm_master_list_entry *mle;
assert_spin_locked(&dlm->spinlock);
-
+
list_for_each_entry(mle, &dlm->mle_hb_events, hb_events) {
if (node_up)
dlm_mle_node_up(dlm, mle, NULL, idx);
__dlm_insert_mle(dlm, mle);
/* still holding the dlm spinlock, check the recovery map
- * to see if there are any nodes that still need to be
+ * to see if there are any nodes that still need to be
* considered. these will not appear in the mle nodemap
* but they might own this lockres. wait on them. */
bit = find_next_bit(dlm->recovery_map, O2NM_MAX_NODES, 0);
msleep(500);
}
continue;
- }
+ }
dlm_kick_recovery_thread(dlm);
msleep(1000);
res->lockname.name, blocked);
if (++tries > 20) {
mlog(ML_ERROR, "%s:%.*s: spinning on "
- "dlm_wait_for_lock_mastery, blocked=%d\n",
- dlm->name, res->lockname.len,
+ "dlm_wait_for_lock_mastery, blocked=%d\n",
+ dlm->name, res->lockname.len,
res->lockname.name, blocked);
dlm_print_one_lock_resource(res);
dlm_print_one_mle(mle);
ret = dlm_restart_lock_mastery(dlm, res, mle, *blocked);
b = (mle->type == DLM_MLE_BLOCK);
if ((*blocked && !b) || (!*blocked && b)) {
- mlog(0, "%s:%.*s: status change: old=%d new=%d\n",
+ mlog(0, "%s:%.*s: status change: old=%d new=%d\n",
dlm->name, res->lockname.len, res->lockname.name,
*blocked, b);
*blocked = b;
}
mlog(0, "%u is the owner of %.*s, cleaning everyone else\n",
dlm->node_num, res->lockname.len, res->lockname.name);
- ret = dlm_dispatch_assert_master(dlm, res, 0, request->node_idx,
+ ret = dlm_dispatch_assert_master(dlm, res, 0, request->node_idx,
DLM_ASSERT_MASTER_MLE_CLEANUP);
if (ret < 0) {
mlog(ML_ERROR, "failed to dispatch assert master work\n");
if (r & DLM_ASSERT_RESPONSE_REASSERT) {
mlog(0, "%.*s: node %u create mles on other "
- "nodes and requests a re-assert\n",
+ "nodes and requests a re-assert\n",
namelen, lockname, to);
reassert = 1;
}
spin_unlock(&dlm->master_lock);
spin_unlock(&dlm->spinlock);
goto done;
- }
+ }
}
}
spin_unlock(&dlm->master_lock);
int extra_ref = 0;
int nn = -1;
int rr, err = 0;
-
+
spin_lock(&mle->spinlock);
if (mle->type == DLM_MLE_BLOCK || mle->type == DLM_MLE_MIGRATION)
extra_ref = 1;
/* MASTER mle: if any bits set in the response map
* then the calling node needs to re-assert to clear
* up nodes that this node contacted */
- while ((nn = find_next_bit (mle->response_map, O2NM_MAX_NODES,
+ while ((nn = find_next_bit (mle->response_map, O2NM_MAX_NODES,
nn+1)) < O2NM_MAX_NODES) {
if (nn != dlm->node_num && nn != assert->node_idx)
master_request = 1;
__dlm_print_one_lock_resource(res);
spin_unlock(&res->spinlock);
spin_unlock(&dlm->spinlock);
- *ret_data = (void *)res;
+ *ret_data = (void *)res;
dlm_put(dlm);
return -EINVAL;
}
item->u.am.request_from = request_from;
item->u.am.flags = flags;
- if (ignore_higher)
- mlog(0, "IGNORE HIGHER: %.*s\n", res->lockname.len,
+ if (ignore_higher)
+ mlog(0, "IGNORE HIGHER: %.*s\n", res->lockname.len,
res->lockname.name);
-
+
spin_lock(&dlm->work_lock);
list_add_tail(&item->list, &dlm->work_list);
spin_unlock(&dlm->work_lock);
* think that $RECOVERY is currently mastered by a dead node. If so,
* we wait a short time to allow that node to get notified by its own
* heartbeat stack, then check again. All $RECOVERY lock resources
- * mastered by dead nodes are purged when the hearbeat callback is
+ * mastered by dead nodes are purged when the hearbeat callback is
* fired, so we can know for sure that it is safe to continue once
* the node returns a live node or no node. */
static int dlm_pre_master_reco_lockres(struct dlm_ctxt *dlm,
ret = -EAGAIN;
}
spin_unlock(&dlm->spinlock);
- mlog(0, "%s: reco lock master is %u\n", dlm->name,
+ mlog(0, "%s: reco lock master is %u\n", dlm->name,
master);
break;
}
mlog(0, "%s:%.*s: timed out during migration\n",
dlm->name, res->lockname.len, res->lockname.name);
- /* avoid hang during shutdown when migrating lockres
+ /* avoid hang during shutdown when migrating lockres
* to a node which also goes down */
if (dlm_is_node_dead(dlm, target)) {
mlog(0, "%s:%.*s: expected migration "
can_proceed = !!(res->state & DLM_LOCK_RES_MIGRATING);
spin_unlock(&res->spinlock);
- /* target has died, so make the caller break out of the
+ /* target has died, so make the caller break out of the
* wait_event, but caller must recheck the domain_map */
spin_lock(&dlm->spinlock);
if (!test_bit(mig_target, dlm->domain_map))
if (lock->ml.node == dead_node) {
mlog(0, "AHA! there was "
"a $RECOVERY lock for dead "
- "node %u (%s)!\n",
+ "node %u (%s)!\n",
dead_node, dlm->name);
list_del_init(&lock->list);
dlm_lock_put(lock);
mres->master = master;
}
+static void dlm_prepare_lvb_for_migration(struct dlm_lock *lock,
+ struct dlm_migratable_lockres *mres,
+ int queue)
+{
+ if (!lock->lksb)
+ return;
+
+ /* Ignore lvb in all locks in the blocked list */
+ if (queue == DLM_BLOCKED_LIST)
+ return;
+
+ /* Only consider lvbs in locks with granted EX or PR lock levels */
+ if (lock->ml.type != LKM_EXMODE && lock->ml.type != LKM_PRMODE)
+ return;
+
+ if (dlm_lvb_is_empty(mres->lvb)) {
+ memcpy(mres->lvb, lock->lksb->lvb, DLM_LVB_LEN);
+ return;
+ }
+
+ /* Ensure the lvb copied for migration matches in other valid locks */
+ if (!memcmp(mres->lvb, lock->lksb->lvb, DLM_LVB_LEN))
+ return;
+
+ mlog(ML_ERROR, "Mismatched lvb in lock cookie=%u:%llu, name=%.*s, "
+ "node=%u\n",
+ dlm_get_lock_cookie_node(be64_to_cpu(lock->ml.cookie)),
+ dlm_get_lock_cookie_seq(be64_to_cpu(lock->ml.cookie)),
+ lock->lockres->lockname.len, lock->lockres->lockname.name,
+ lock->ml.node);
+ dlm_print_one_lock_resource(lock->lockres);
+ BUG();
+}
/* returns 1 if this lock fills the network structure,
* 0 otherwise */
ml->list = queue;
if (lock->lksb) {
ml->flags = lock->lksb->flags;
- /* send our current lvb */
- if (ml->type == LKM_EXMODE ||
- ml->type == LKM_PRMODE) {
- /* if it is already set, this had better be a PR
- * and it has to match */
- if (!dlm_lvb_is_empty(mres->lvb) &&
- (ml->type == LKM_EXMODE ||
- memcmp(mres->lvb, lock->lksb->lvb, DLM_LVB_LEN))) {
- mlog(ML_ERROR, "mismatched lvbs!\n");
- dlm_print_one_lock_resource(lock->lockres);
- BUG();
- }
- memcpy(mres->lvb, lock->lksb->lvb, DLM_LVB_LEN);
- }
+ dlm_prepare_lvb_for_migration(lock, mres, queue);
}
ml->node = lock->ml.node;
mres->num_locks++;
struct dlm_lock *lock = NULL;
u8 from = O2NM_MAX_NODES;
unsigned int added = 0;
+ __be64 c;
mlog(0, "running %d locks for this lockres\n", mres->num_locks);
for (i=0; i<mres->num_locks; i++) {
/* lock is always created locally first, and
* destroyed locally last. it must be on the list */
if (!lock) {
- __be64 c = ml->cookie;
- mlog(ML_ERROR, "could not find local lock "
- "with cookie %u:%llu!\n",
+ c = ml->cookie;
+ mlog(ML_ERROR, "Could not find local lock "
+ "with cookie %u:%llu, node %u, "
+ "list %u, flags 0x%x, type %d, "
+ "conv %d, highest blocked %d\n",
dlm_get_lock_cookie_node(be64_to_cpu(c)),
- dlm_get_lock_cookie_seq(be64_to_cpu(c)));
+ dlm_get_lock_cookie_seq(be64_to_cpu(c)),
+ ml->node, ml->list, ml->flags, ml->type,
+ ml->convert_type, ml->highest_blocked);
+ __dlm_print_one_lock_resource(res);
+ BUG();
+ }
+
+ if (lock->ml.node != ml->node) {
+ c = lock->ml.cookie;
+ mlog(ML_ERROR, "Mismatched node# in lock "
+ "cookie %u:%llu, name %.*s, node %u\n",
+ dlm_get_lock_cookie_node(be64_to_cpu(c)),
+ dlm_get_lock_cookie_seq(be64_to_cpu(c)),
+ res->lockname.len, res->lockname.name,
+ lock->ml.node);
+ c = ml->cookie;
+ mlog(ML_ERROR, "Migrate lock cookie %u:%llu, "
+ "node %u, list %u, flags 0x%x, type %d, "
+ "conv %d, highest blocked %d\n",
+ dlm_get_lock_cookie_node(be64_to_cpu(c)),
+ dlm_get_lock_cookie_seq(be64_to_cpu(c)),
+ ml->node, ml->list, ml->flags, ml->type,
+ ml->convert_type, ml->highest_blocked);
__dlm_print_one_lock_resource(res);
BUG();
}
- BUG_ON(lock->ml.node != ml->node);
if (tmpq != queue) {
- mlog(0, "lock was on %u instead of %u for %.*s\n",
- j, ml->list, res->lockname.len, res->lockname.name);
+ c = ml->cookie;
+ mlog(0, "Lock cookie %u:%llu was on list %u "
+ "instead of list %u for %.*s\n",
+ dlm_get_lock_cookie_node(be64_to_cpu(c)),
+ dlm_get_lock_cookie_seq(be64_to_cpu(c)),
+ j, ml->list, res->lockname.len,
+ res->lockname.name);
+ __dlm_print_one_lock_resource(res);
spin_unlock(&res->spinlock);
continue;
}
* the lvb. */
memcpy(res->lvb, mres->lvb, DLM_LVB_LEN);
} else {
- /* otherwise, the node is sending its
+ /* otherwise, the node is sending its
* most recent valid lvb info */
BUG_ON(ml->type != LKM_EXMODE &&
ml->type != LKM_PRMODE);
spin_lock(&res->spinlock);
list_for_each_entry(lock, queue, list) {
if (lock->ml.cookie == ml->cookie) {
- __be64 c = lock->ml.cookie;
+ c = lock->ml.cookie;
mlog(ML_ERROR, "%s:%.*s: %u:%llu: lock already "
"exists on this lockres!\n", dlm->name,
res->lockname.len, res->lockname.name,
assert_spin_locked(&res->spinlock);
if (res->owner == dlm->node_num)
- /* if this node owned the lockres, and if the dead node
+ /* if this node owned the lockres, and if the dead node
* had an EX when he died, blank out the lvb */
search_node = dead_node;
else {
/* this node is the lockres master:
* 1) remove any stale locks for the dead node
- * 2) if the dead node had an EX when he died, blank out the lvb
+ * 2) if the dead node had an EX when he died, blank out the lvb
*/
assert_spin_locked(&dlm->spinlock);
assert_spin_locked(&res->spinlock);
mlog(0, "%s:%.*s: freed %u locks for dead node %u, "
"dropping ref from lockres\n", dlm->name,
res->lockname.len, res->lockname.name, freed, dead_node);
- BUG_ON(!test_bit(dead_node, res->refmap));
+ if(!test_bit(dead_node, res->refmap)) {
+ mlog(ML_ERROR, "%s:%.*s: freed %u locks for dead node %u, "
+ "but ref was not set\n", dlm->name,
+ res->lockname.len, res->lockname.name, freed, dead_node);
+ __dlm_print_one_lock_resource(res);
+ }
dlm_lockres_clear_refmap_bit(dead_node, res);
} else if (test_bit(dead_node, res->refmap)) {
mlog(0, "%s:%.*s: dead node %u had a ref, but had "
}
spin_unlock(&res->spinlock);
continue;
- }
+ }
spin_lock(&res->spinlock);
/* zero the lvb if necessary */
dlm_revalidate_lvb(dlm, res, dead_node);
* this function on each node racing to become the recovery
* master will not stop attempting this until either:
* a) this node gets the EX (and becomes the recovery master),
- * or b) dlm->reco.new_master gets set to some nodenum
+ * or b) dlm->reco.new_master gets set to some nodenum
* != O2NM_INVALID_NODE_NUM (another node will do the reco).
* so each time a recovery master is needed, the entire cluster
* will sync at this point. if the new master dies, that will
mlog(0, "starting recovery of %s at %lu, dead=%u, this=%u\n",
dlm->name, jiffies, dlm->reco.dead_node, dlm->node_num);
-again:
+again:
memset(&lksb, 0, sizeof(lksb));
ret = dlmlock(dlm, LKM_EXMODE, &lksb, LKM_NOQUEUE|LKM_RECOVERY,
if (ret == DLM_NORMAL) {
mlog(0, "dlm=%s dlmlock says I got it (this=%u)\n",
dlm->name, dlm->node_num);
-
- /* got the EX lock. check to see if another node
+
+ /* got the EX lock. check to see if another node
* just became the reco master */
if (dlm_reco_master_ready(dlm)) {
mlog(0, "%s: got reco EX lock, but %u will "
/* see if recovery was already finished elsewhere */
spin_lock(&dlm->spinlock);
if (dlm->reco.dead_node == O2NM_INVALID_NODE_NUM) {
- status = -EINVAL;
+ status = -EINVAL;
mlog(0, "%s: got reco EX lock, but "
"node got recovered already\n", dlm->name);
if (dlm->reco.new_master != O2NM_INVALID_NODE_NUM) {
mlog(ML_ERROR, "%s: new master is %u "
- "but no dead node!\n",
+ "but no dead node!\n",
dlm->name, dlm->reco.new_master);
BUG();
}
* set the master and send the messages to begin recovery */
if (!status) {
mlog(0, "%s: dead=%u, this=%u, sending "
- "begin_reco now\n", dlm->name,
+ "begin_reco now\n", dlm->name,
dlm->reco.dead_node, dlm->node_num);
status = dlm_send_begin_reco_message(dlm,
dlm->reco.dead_node);
mlog(0, "dlm=%s dlmlock says another node got it (this=%u)\n",
dlm->name, dlm->node_num);
/* another node is master. wait on
- * reco.new_master != O2NM_INVALID_NODE_NUM
+ * reco.new_master != O2NM_INVALID_NODE_NUM
* for at most one second */
wait_event_timeout(dlm->dlm_reco_thread_wq,
dlm_reco_master_ready(dlm),
"begin reco msg (%d)\n", dlm->name, nodenum, ret);
ret = 0;
}
- if (ret == -EAGAIN) {
+
+ /*
+ * Prior to commit aad1b15310b9bcd59fa81ab8f2b1513b59553ea8,
+ * dlm_begin_reco_handler() returned EAGAIN and not -EAGAIN.
+ * We are handling both for compatibility reasons.
+ */
+ if (ret == -EAGAIN || ret == EAGAIN) {
mlog(0, "%s: trying to start recovery of node "
"%u, but node %u is waiting for last recovery "
"to complete, backoff for a bit\n", dlm->name,
}
if (ret < 0) {
struct dlm_lock_resource *res;
- /* this is now a serious problem, possibly ENOMEM
+ /* this is now a serious problem, possibly ENOMEM
* in the network stack. must retry */
mlog_errno(ret);
mlog(ML_ERROR, "begin reco of dlm %s to node %u "
} else {
mlog(ML_ERROR, "recovery lock not found\n");
}
- /* sleep for a bit in hopes that we can avoid
+ /* sleep for a bit in hopes that we can avoid
* another ENOMEM */
msleep(100);
goto retry;
}
if (dlm->reco.dead_node != O2NM_INVALID_NODE_NUM) {
mlog(ML_NOTICE, "%s: dead_node previously set to %u, "
- "node %u changing it to %u\n", dlm->name,
+ "node %u changing it to %u\n", dlm->name,
dlm->reco.dead_node, br->node_idx, br->dead_node);
}
dlm_set_reco_master(dlm, br->node_idx);
if (ret < 0) {
mlog_errno(ret);
if (dlm_is_host_down(ret)) {
- /* this has no effect on this recovery
- * session, so set the status to zero to
+ /* this has no effect on this recovery
+ * session, so set the status to zero to
* finish out the last recovery */
mlog(ML_ERROR, "node %u went down after this "
"node finished recovery.\n", nodenum);
mlog(0, "%s: node %u finalizing recovery stage%d of "
"node %u (%u:%u)\n", dlm->name, fr->node_idx, stage,
fr->dead_node, dlm->reco.dead_node, dlm->reco.new_master);
-
+
spin_lock(&dlm->spinlock);
if (dlm->reco.new_master != fr->node_idx) {
actions &= ~(DLM_UNLOCK_REMOVE_LOCK|
DLM_UNLOCK_REGRANT_LOCK|
DLM_UNLOCK_CLEAR_CONVERT_TYPE);
- } else if (status == DLM_RECOVERING ||
- status == DLM_MIGRATING ||
+ } else if (status == DLM_RECOVERING ||
+ status == DLM_MIGRATING ||
status == DLM_FORWARD) {
/* must clear the actions because this unlock
* is about to be retried. cannot free or do
if (call_ast) {
mlog(0, "calling unlockast(%p, %d)\n", data, status);
if (is_master) {
- /* it is possible that there is one last bast
+ /* it is possible that there is one last bast
* pending. make sure it is flushed, then
* call the unlockast.
* not an issue if this is a mastered remotely,
* since this lock has been removed from the
* lockres queues and cannot be found. */
dlm_kick_thread(dlm, NULL);
- wait_event(dlm->ast_wq,
+ wait_event(dlm->ast_wq,
dlm_lock_basts_flushed(dlm, lock));
}
(*unlockast)(data, status);
lockres_or_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH);
lockres->l_level = lockres->l_requested;
+
+ /*
+ * We set the OCFS2_LOCK_UPCONVERT_FINISHING flag before clearing
+ * the OCFS2_LOCK_BUSY flag to prevent the dc thread from
+ * downconverting the lock before the upconvert has fully completed.
+ */
+ lockres_or_flags(lockres, OCFS2_LOCK_UPCONVERT_FINISHING);
+
lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
mlog_exit_void();
assert_spin_locked(&lockres->l_lock);
- lockres_or_flags(lockres, OCFS2_LOCK_BLOCKED);
-
if (level > lockres->l_blocking) {
/* only schedule a downconvert if we haven't already scheduled
* one that goes low enough to satisfy the level we're
lockres->l_blocking = level;
}
+ if (needs_downconvert)
+ lockres_or_flags(lockres, OCFS2_LOCK_BLOCKED);
+
mlog_exit(needs_downconvert);
return needs_downconvert;
}
mlog_entry_void();
spin_lock_irqsave(&lockres->l_lock, flags);
lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
+ lockres_clear_flags(lockres, OCFS2_LOCK_UPCONVERT_FINISHING);
if (convert)
lockres->l_action = OCFS2_AST_INVALID;
else
again:
wait = 0;
+ spin_lock_irqsave(&lockres->l_lock, flags);
+
if (catch_signals && signal_pending(current)) {
ret = -ERESTARTSYS;
- goto out;
+ goto unlock;
}
- spin_lock_irqsave(&lockres->l_lock, flags);
-
mlog_bug_on_msg(lockres->l_flags & OCFS2_LOCK_FREEING,
"Cluster lock called on freeing lockres %s! flags "
"0x%lx\n", lockres->l_name, lockres->l_flags);
goto unlock;
}
+ if (lockres->l_flags & OCFS2_LOCK_UPCONVERT_FINISHING) {
+ /*
+ * We've upconverted. If the lock now has a level we can
+ * work with, we take it. If, however, the lock is not at the
+ * required level, we go thru the full cycle. One way this could
+ * happen is if a process requesting an upconvert to PR is
+ * closely followed by another requesting upconvert to an EX.
+ * If the process requesting EX lands here, we want it to
+ * continue attempting to upconvert and let the process
+ * requesting PR take the lock.
+ * If multiple processes request upconvert to PR, the first one
+ * here will take the lock. The others will have to go thru the
+ * OCFS2_LOCK_BLOCKED check to ensure that there is no pending
+ * downconvert request.
+ */
+ if (level <= lockres->l_level)
+ goto update_holders;
+ }
+
if (lockres->l_flags & OCFS2_LOCK_BLOCKED &&
!ocfs2_may_continue_on_blocked_lock(lockres, level)) {
/* is the lock is currently blocked on behalf of
goto again;
}
+update_holders:
/* Ok, if we get here then we're good to go. */
ocfs2_inc_holders(lockres, level);
ret = 0;
unlock:
+ lockres_clear_flags(lockres, OCFS2_LOCK_UPCONVERT_FINISHING);
+
spin_unlock_irqrestore(&lockres->l_lock, flags);
out:
/*
/* Mark the lockres as being dropped. It will no longer be
* queued if blocking, but we still may have to wait on it
* being dequeued from the downconvert thread before we can consider
- * it safe to drop.
+ * it safe to drop.
*
* You can *not* attempt to call cluster_lock on this lockres anymore. */
void ocfs2_mark_lockres_freeing(struct ocfs2_lock_res *lockres)
unsigned long flags;
int blocking;
int new_level;
+ int level;
int ret = 0;
int set_lvb = 0;
unsigned int gen;
spin_lock_irqsave(&lockres->l_lock, flags);
- BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BLOCKED));
-
recheck:
+ /*
+ * Is it still blocking? If not, we have no more work to do.
+ */
+ if (!(lockres->l_flags & OCFS2_LOCK_BLOCKED)) {
+ BUG_ON(lockres->l_blocking != DLM_LOCK_NL);
+ spin_unlock_irqrestore(&lockres->l_lock, flags);
+ ret = 0;
+ goto leave;
+ }
+
if (lockres->l_flags & OCFS2_LOCK_BUSY) {
/* XXX
* This is a *big* race. The OCFS2_LOCK_PENDING flag
goto leave;
}
+ /*
+ * This prevents livelocks. OCFS2_LOCK_UPCONVERT_FINISHING flag is
+ * set when the ast is received for an upconvert just before the
+ * OCFS2_LOCK_BUSY flag is cleared. Now if the fs received a bast
+ * on the heels of the ast, we want to delay the downconvert just
+ * enough to allow the up requestor to do its task. Because this
+ * lock is in the blocked queue, the lock will be downconverted
+ * as soon as the requestor is done with the lock.
+ */
+ if (lockres->l_flags & OCFS2_LOCK_UPCONVERT_FINISHING)
+ goto leave_requeue;
+
+ /*
+ * How can we block and yet be at NL? We were trying to upconvert
+ * from NL and got canceled. The code comes back here, and now
+ * we notice and clear BLOCKING.
+ */
+ if (lockres->l_level == DLM_LOCK_NL) {
+ BUG_ON(lockres->l_ex_holders || lockres->l_ro_holders);
+ lockres->l_blocking = DLM_LOCK_NL;
+ lockres_clear_flags(lockres, OCFS2_LOCK_BLOCKED);
+ spin_unlock_irqrestore(&lockres->l_lock, flags);
+ goto leave;
+ }
+
/* if we're blocking an exclusive and we have *any* holders,
* then requeue. */
if ((lockres->l_blocking == DLM_LOCK_EX)
* may sleep, so we save off a copy of what we're blocking as
* it may change while we're not holding the spin lock. */
blocking = lockres->l_blocking;
+ level = lockres->l_level;
spin_unlock_irqrestore(&lockres->l_lock, flags);
ctl->unblock_action = lockres->l_ops->downconvert_worker(lockres, blocking);
goto leave;
spin_lock_irqsave(&lockres->l_lock, flags);
- if (blocking != lockres->l_blocking) {
+ if ((blocking != lockres->l_blocking) || (level != lockres->l_level)) {
/* If this changed underneath us, then we can't drop
* it just yet. */
goto recheck;
mlog(0, "Encoding parent: blkno: %llu, generation: %u\n",
(unsigned long long)blkno, generation);
}
-
+
*max_len = len;
bail:
emi->ei_clusters += ins->ei_clusters;
return 1;
} else if ((ins->ei_phys + ins->ei_clusters) == emi->ei_phys &&
- (ins->ei_cpos + ins->ei_clusters) == emi->ei_phys &&
+ (ins->ei_cpos + ins->ei_clusters) == emi->ei_cpos &&
ins->ei_flags == emi->ei_flags) {
emi->ei_phys = ins->ei_phys;
emi->ei_cpos = ins->ei_cpos;
int ret;
offset = (size & (PAGE_CACHE_SIZE-1)); /* Within page */
- /* ugh. in prepare/commit_write, if from==to==start of block, we
+ /* ugh. in prepare/commit_write, if from==to==start of block, we
** skip the prepare. make sure we never send an offset for the start
** of a block
*/
struct inode *inode = dentry->d_inode;
loff_t saved_pos, end;
- /*
+ /*
* We start with a read level meta lock and only jump to an ex
* if we need to make modifications here.
*/
/* buffered aio wouldn't have proper lock coverage today */
BUG_ON(ret == -EIOCBQUEUED && !(file->f_flags & O_DIRECT));
- if ((file->f_flags & O_DSYNC && !direct_io) || IS_SYNC(inode) ||
- (file->f_flags & O_DIRECT && has_refcount)) {
+ if (((file->f_flags & O_DSYNC) && !direct_io) || IS_SYNC(inode) ||
+ ((file->f_flags & O_DIRECT) && has_refcount)) {
ret = filemap_fdatawrite_range(file->f_mapping, pos,
pos + count - 1);
if (ret < 0)
pos + count - 1);
}
- /*
+ /*
* deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
* function pointer which is called when o_direct io completes so that
* it can unlock our rw lock. (it's the clustered equivalent of
goto bail;
}
- /*
+ /*
* buffered reads protect themselves in ->readpage(). O_DIRECT reads
* need locks to protect pending reads from racing with truncate.
*/
* We're fine letting folks race truncates and extending
* writes with read across the cluster, just like they can
* locally. Hence no rw_lock during read.
- *
+ *
* Take and drop the meta data lock to update inode fields
* like i_size. This allows the checks down below
- * generic_file_aio_read() a chance of actually working.
+ * generic_file_aio_read() a chance of actually working.
*/
ret = ocfs2_inode_lock_atime(inode, filp->f_vfsmnt, &lock_level);
if (ret < 0) {
bail:
if (have_alloc_sem)
up_read(&inode->i_alloc_sem);
- if (rw_level != -1)
+ if (rw_level != -1)
ocfs2_rw_unlock(inode, rw_level);
mlog_exit(ret);
if (args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY) {
status = ocfs2_try_open_lock(inode, 0);
if (status) {
- make_bad_inode(inode);
+ make_bad_inode(inode);
return status;
}
}
return status;
}
-/*
+/*
* Serialize with orphan dir recovery. If the process doing
* recovery on this orphan dir does an iget() with the dir
* i_mutex held, we'll deadlock here. Instead we detect this
#include <linux/fs.h>
#include <linux/mount.h>
+#include <linux/compat.h>
#define MLOG_MASK_PREFIX ML_INODE
#include <cluster/masklog.h>
#ifdef CONFIG_COMPAT
long ocfs2_compat_ioctl(struct file *file, unsigned cmd, unsigned long arg)
{
+ bool preserve;
+ struct reflink_arguments args;
+ struct inode *inode = file->f_path.dentry->d_inode;
+
switch (cmd) {
case OCFS2_IOC32_GETFLAGS:
cmd = OCFS2_IOC_GETFLAGS;
case OCFS2_IOC_GROUP_EXTEND:
case OCFS2_IOC_GROUP_ADD:
case OCFS2_IOC_GROUP_ADD64:
- case OCFS2_IOC_REFLINK:
break;
+ case OCFS2_IOC_REFLINK:
+ if (copy_from_user(&args, (struct reflink_arguments *)arg,
+ sizeof(args)))
+ return -EFAULT;
+ preserve = (args.preserve != 0);
+
+ return ocfs2_reflink_ioctl(inode, compat_ptr(args.old_path),
+ compat_ptr(args.new_path), preserve);
default:
return -ENOIOCTLCMD;
}
status = -ENOENT;
mlog_errno(status);
return status;
- }
+ }
mutex_lock(&orphan_dir_inode->i_mutex);
status = ocfs2_inode_lock(orphan_dir_inode, NULL, 0);
#define OCFS2_LOCK_PENDING (0x00000400) /* This lockres is pending a
call to dlm_lock. Only
exists with BUSY set. */
+#define OCFS2_LOCK_UPCONVERT_FINISHING (0x00000800) /* blocks the dc thread
+ * from downconverting
+ * before the upconvert
+ * has completed */
struct ocfs2_lock_res_ops;
return blocksize - offsetof(struct ocfs2_dinode, id2.i_symlink);
}
-static inline int ocfs2_max_inline_data(int blocksize)
+static inline int ocfs2_max_inline_data_with_xattr(int blocksize,
+ struct ocfs2_dinode *di)
{
- return blocksize - offsetof(struct ocfs2_dinode, id2.i_data.id_data);
+ if (di && (di->i_dyn_features & OCFS2_INLINE_XATTR_FL))
+ return blocksize -
+ offsetof(struct ocfs2_dinode, id2.i_data.id_data) -
+ di->i_xattr_inline_size;
+ else
+ return blocksize -
+ offsetof(struct ocfs2_dinode, id2.i_data.id_data);
}
static inline int ocfs2_extent_recs_per_inode(int blocksize)
while (offset < end) {
page_index = offset >> PAGE_CACHE_SHIFT;
- map_end = (page_index + 1) << PAGE_CACHE_SHIFT;
+ map_end = ((loff_t)page_index + 1) << PAGE_CACHE_SHIFT;
if (map_end > end)
map_end = end;
page = grab_cache_page(mapping, page_index);
- /* This page can't be dirtied before we CoW it out. */
- BUG_ON(PageDirty(page));
+ /*
+ * In case PAGE_CACHE_SIZE <= CLUSTER_SIZE, This page
+ * can't be dirtied before we CoW it out.
+ */
+ if (PAGE_CACHE_SIZE <= OCFS2_SB(sb)->s_clustersize)
+ BUG_ON(PageDirty(page));
if (!PageUptodate(page)) {
ret = block_read_full_page(page, ocfs2_get_block);
while (offset < end) {
page_index = offset >> PAGE_CACHE_SHIFT;
- map_end = (page_index + 1) << PAGE_CACHE_SHIFT;
+ map_end = ((loff_t)page_index + 1) << PAGE_CACHE_SHIFT;
if (map_end > end)
map_end = end;
u32 dlm_key;
struct dlm_ctxt *dlm;
struct o2dlm_private *priv;
- struct dlm_protocol_version dlm_version;
+ struct dlm_protocol_version fs_version;
BUG_ON(conn == NULL);
BUG_ON(o2cb_stack.sp_proto == NULL);
/* used by the dlm code to make message headers unique, each
* node in this domain must agree on this. */
dlm_key = crc32_le(0, conn->cc_name, conn->cc_namelen);
- dlm_version.pv_major = conn->cc_version.pv_major;
- dlm_version.pv_minor = conn->cc_version.pv_minor;
+ fs_version.pv_major = conn->cc_version.pv_major;
+ fs_version.pv_minor = conn->cc_version.pv_minor;
- dlm = dlm_register_domain(conn->cc_name, dlm_key, &dlm_version);
+ dlm = dlm_register_domain(conn->cc_name, dlm_key, &fs_version);
if (IS_ERR(dlm)) {
rc = PTR_ERR(dlm);
mlog_errno(rc);
goto out_free;
}
- conn->cc_version.pv_major = dlm_version.pv_major;
- conn->cc_version.pv_minor = dlm_version.pv_minor;
+ conn->cc_version.pv_major = fs_version.pv_major;
+ conn->cc_version.pv_minor = fs_version.pv_minor;
conn->cc_lockspace = dlm;
dlm_register_eviction_cb(dlm, &priv->op_eviction_cb);
"file system, but write access is "
"unavailable.\n");
else
- mlog_errno(status);
+ mlog_errno(status);
goto read_super_error;
}
}
memcpy(link, target, len);
- nd_set_link(nd, link);
bail:
+ nd_set_link(nd, status ? ERR_PTR(status) : link);
brelse(bh);
mlog_exit(status);
- return status ? ERR_PTR(status) : link;
+ return NULL;
}
static void ocfs2_fast_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
{
- char *link = cookie;
-
- kfree(link);
+ char *link = nd_get_link(nd);
+ if (!IS_ERR(link))
+ kfree(link);
}
const struct inode_operations ocfs2_symlink_inode_operations = {
}
/* Warning: even if it returns true, this does *not* guarantee that
- * the block is stored in our inode metadata cache.
- *
+ * the block is stored in our inode metadata cache.
+ *
* This can be called under lock_buffer()
*/
int ocfs2_buffer_uptodate(struct ocfs2_caching_info *ci,
projects / platform / adaptation / renesas_rcar / renesas_kernel.git / commitdiff