client: ensure that completion occurs exactly once

This commit is contained in:
Thomas Schoebel-Theuer 2021-04-21 10:04:14 +02:00
parent aef14d7011
commit f98dd17aa4
2 changed files with 39 additions and 2 deletions

View File

@ -526,7 +526,9 @@ static void client_ref_io(struct client_output *output, struct mref_object *mref
atomic_inc(&brick->fly_count);
_mref_get(mref);
mref_a->has_completed = false;
mref_a->submit_jiffies = jiffies;
_hash_insert(output, mref_a);
MARS_IO("added request id = %d pos = %lld len = %d flags = %ux (flying = %d)\n",
@ -539,6 +541,7 @@ static void client_ref_io(struct client_output *output, struct mref_object *mref
error:
MARS_ERR("IO submission on dead instance\n");
mref_a->has_completed = true;
error = -ESHUTDOWN;
SIMPLE_CALLBACK(mref, error);
return;
@ -604,6 +607,7 @@ int receiver_thread(void *data)
struct client_mref_aspect *mref_a = NULL;
unsigned long id = READ_ONCE(cmd.cmd_int1);
unsigned int hash_index = CLIENT_HASH_FN(id);
bool had_completed = false;
mutex_lock(&output->mutex);
for (tmp = output->hash_table[hash_index].next; tmp != &output->hash_table[hash_index]; tmp = tmp->next) {
@ -626,6 +630,11 @@ int receiver_thread(void *data)
mref = tmp_mref;
list_del_init(&mref_a->hash_head);
list_del_init(&mref_a->io_head);
/* Networking produces inherent races between re-submission and
* completion. Compensate them here.
*/
had_completed = mref_a->has_completed;
mref_a->has_completed = true;
break;
err:
@ -660,6 +669,8 @@ int receiver_thread(void *data)
output->bundle.path,
output->bundle.host,
status);
if (had_completed)
goto has_finished;
_hash_insert(output, mref_a);
goto done;
}
@ -667,8 +678,11 @@ int receiver_thread(void *data)
if (mref->_object_cb.cb_error < 0) {
MARS_DBG("ERROR %d\n", mref->_object_cb.cb_error);
}
SIMPLE_CALLBACK(mref, mref->_object_cb.cb_error);
if (!had_completed) {
SIMPLE_CALLBACK(mref, mref->_object_cb.cb_error);
}
has_finished:
client_ref_put(output, mref);
atomic_dec(&output->brick->fly_count);
@ -729,6 +743,7 @@ void _do_timeout(struct client_output *output, struct list_head *anchor, int *ro
struct list_head *tmp;
struct list_head *prev;
LIST_HEAD(tmp_list);
LIST_HEAD(completed_list);
long io_timeout = _compute_timeout(brick, false);
int i;
@ -770,7 +785,12 @@ void _do_timeout(struct client_output *output, struct list_head *anchor, int *ro
list_del_init(&mref_a->hash_head);
list_del_init(&mref_a->io_head);
list_add_tail(&mref_a->tmp_head, &tmp_list);
if (mref_a->has_completed) {
list_add_tail(&mref_a->tmp_head, &completed_list);
} else {
mref_a->has_completed = true;
list_add_tail(&mref_a->tmp_head, &tmp_list);
}
}
mutex_unlock(&output->mutex);
@ -801,6 +821,22 @@ void _do_timeout(struct client_output *output, struct list_head *anchor, int *ro
atomic_dec(&brick->fly_count);
atomic_dec(&mars_global_io_flying);
}
while (!list_empty(&completed_list)) {
struct client_mref_aspect *mref_a;
struct mref_object *mref;
tmp = completed_list.next;
list_del_init(tmp);
mref_a = container_of(tmp, struct client_mref_aspect, tmp_head);
mref = mref_a->object;
client_ref_put(output, mref);
atomic_dec(&brick->fly_count);
atomic_dec(&mars_global_io_flying);
}
}
static

View File

@ -44,6 +44,7 @@ struct client_mref_aspect {
unsigned long submit_jiffies;
int alloc_len;
bool do_dealloc;
bool has_completed;
};
struct client_brick {