static void check_pids(void *arg)
{
int i;
size_t nr;
struct local_node lnodes[SD_MAX_NODES];
struct local_event *ev;
shm_queue_lock();
nr = get_nodes(lnodes);
for (i = 0; i < nr; i++)
if (!process_exists(lnodes[i].pid)) {
add_event(EVENT_LEAVE, lnodes + i, NULL, 0);
/* unblock blocking event if sender has gone */
ev = shm_queue_peek_block_event();
if (lnode_eq(lnodes + i, &ev->sender)) {
ev->removed = true;
msync(ev, sizeof(*ev), MS_SYNC);
}
}
shm_queue_unlock();
add_timer(arg, PROCESS_CHECK_INTERVAL);
}
static struct local_node *find_lnode(struct local_node *key, size_t nr_lnodes,
struct local_node *lnodes)
{
int i;
for (i = 0; i < nr_lnodes; i++)
if (lnode_eq(key, lnodes + i))
return lnodes + i;
panic("internal error");
}
/* Returns true if an event is processed */
static bool local_process_event(void)
{
struct local_event *ev;
enum cluster_join_result res;
int i;
struct sd_node nodes[SD_MAX_NODES];
size_t nr_nodes;
ev = shm_queue_peek();
if (!ev)
return false;
sd_dprintf("type = %d, sender = %s", ev->type,
lnode_to_str(&ev->sender));
sd_dprintf("callbacked = %d, removed = %d", ev->callbacked,
ev->removed);
nr_nodes = 0;
for (i = 0; i < ev->nr_lnodes; i++) {
sd_dprintf("%d: %s", i, lnode_to_str(ev->lnodes + i));
if (!ev->lnodes[i].gateway)
nodes[nr_nodes++] = ev->lnodes[i].node;
}
if (ev->removed)
goto out;
if (ev->callbacked)
return false; /* wait for unblock event */
if (ev->type == EVENT_JOIN_RESPONSE &&
lnode_eq(&this_node, &ev->sender)) {
sd_dprintf("join Sheepdog");
joined = true;
}
if (!joined) {
if (ev->type == EVENT_JOIN_REQUEST &&
lnode_eq(&this_node, &ev->sender)) {
struct local_node lnodes[SD_MAX_NODES];
get_nodes(lnodes);
if (!lnode_eq(&this_node, &lnodes[0])) {
sd_dprintf("wait for another node"
" to accept this node");
return false;
}
} else
goto out;
}
switch (ev->type) {
case EVENT_JOIN_REQUEST:
res = sd_check_join_cb(&ev->sender.node, ev->buf);
ev->join_result = res;
ev->type = EVENT_JOIN_RESPONSE;
msync(ev, sizeof(*ev), MS_SYNC);
shm_queue_notify();
if (res == CJ_RES_MASTER_TRANSFER) {
sd_eprintf("failed to join sheepdog cluster: "
"please retry when master is up");
shm_queue_unlock();
exit(1);
}
return false;
case EVENT_JOIN_RESPONSE:
if (ev->join_result == CJ_RES_MASTER_TRANSFER) {
/* FIXME: This code is tricky, but Sheepdog assumes that */
/* nr_nodes = 1 when join_result = MASTER_TRANSFER... */
ev->nr_lnodes = 1;
ev->lnodes[0] = this_node;
nr_nodes = 1;
nodes[0] = this_node.node;
msync(ev, sizeof(*ev), MS_SYNC);
}
sd_join_handler(&ev->sender.node, nodes, nr_nodes,
ev->join_result, ev->buf);
break;
case EVENT_LEAVE:
case EVENT_GATEWAY:
sd_leave_handler(&ev->sender.node, nodes, nr_nodes);
break;
case EVENT_BLOCK:
ev->callbacked = sd_block_handler(&ev->sender.node);
msync(ev, sizeof(*ev), MS_SYNC);
return false;
case EVENT_NOTIFY:
sd_notify_handler(&ev->sender.node, ev->buf, ev->buf_len);
break;
}
out:
shm_queue_remove(ev);
return true;
}
/* Returns true if an event is processed */
static bool local_process_event(void)
{
struct local_event *ev;
int i;
struct sd_node nodes[SD_MAX_NODES];
size_t nr_nodes;
ev = shm_queue_peek();
if (!ev)
return false;
sd_debug("type = %d, sender = %s", ev->type, lnode_to_str(&ev->sender));
sd_debug("callbacked = %d, removed = %d", ev->callbacked, ev->removed);
nr_nodes = 0;
for (i = 0; i < ev->nr_lnodes; i++) {
sd_debug("%d: %s", i, lnode_to_str(ev->lnodes + i));
if (!ev->lnodes[i].gateway)
nodes[nr_nodes++] = ev->lnodes[i].node;
}
if (ev->removed)
goto out;
if (ev->callbacked)
return false; /* wait for unblock event */
if (!joined) {
if (!lnode_eq(&this_node, &ev->sender))
goto out;
switch (ev->type) {
case EVENT_JOIN:
break;
case EVENT_ACCEPT:
sd_debug("join Sheepdog");
joined = true;
break;
default:
goto out;
}
}
switch (ev->type) {
case EVENT_JOIN:
/* nodes[nr_nodes - 1] is a sender, so don't include it */
assert(node_eq(&ev->sender.node, &nodes[nr_nodes - 1]));
if (sd_join_handler(&ev->sender.node, nodes, nr_nodes - 1,
ev->buf)) {
ev->type = EVENT_ACCEPT;
msync(ev, sizeof(*ev), MS_SYNC);
shm_queue_notify();
}
return false;
case EVENT_ACCEPT:
sd_accept_handler(&ev->sender.node, nodes, nr_nodes, ev->buf);
break;
case EVENT_LEAVE:
if (ev->sender.gateway) {
sd_debug("gateway %s left sheepdog",
lnode_to_str(&ev->sender));
break;
}
/* fall through */
case EVENT_GATEWAY:
sd_leave_handler(&ev->sender.node, nodes, nr_nodes);
break;
case EVENT_BLOCK:
ev->callbacked = sd_block_handler(&ev->sender.node);
msync(ev, sizeof(*ev), MS_SYNC);
return false;
case EVENT_NOTIFY:
sd_notify_handler(&ev->sender.node, ev->buf, ev->buf_len);
break;
case EVENT_UPDATE_NODE:
if (lnode_eq(&ev->sender, &this_node))
this_node = ev->sender;
sd_update_node_handler(&ev->sender.node);
break;
}
out:
shm_queue_remove(ev);
return true;
}
