When the adist_free list is empty and we lose connection to the receiver we

move all elements from the adist_send and adist_recv lists back onto the
adist_free list, but we don't wake consumers waitings for the adist_free list
to become non-empty. This can lead to the sender process stopping audit trail
files distribution and waiting forever.

Fix the problem by adding the missing wakeup.

While here slow down spinning on CPU in case of a short race in
sender_disconnect() and add an explaination when it can occur.

PR:		201953
Reported by:	peter
Approved by:	re (kib)
This commit is contained in:
pjd 2018-10-04 05:54:57 +00:00
parent 55f081d363
commit dd4a41d654
2 changed files with 21 additions and 9 deletions

View File

@ -248,6 +248,21 @@ struct adrep {
if (_wakeup) \
cv_signal(list##_cond); \
} while (0)
#define QUEUE_CONCAT2(tolist, fromlist1, fromlist2) do { \
bool _wakeup; \
\
mtx_lock(tolist##_lock); \
_wakeup = TAILQ_EMPTY(tolist); \
mtx_lock(fromlist1##_lock); \
TAILQ_CONCAT((tolist), (fromlist1), adr_next); \
mtx_unlock(fromlist1##_lock); \
mtx_lock(fromlist2##_lock); \
TAILQ_CONCAT((tolist), (fromlist2), adr_next); \
mtx_unlock(fromlist2##_lock); \
mtx_unlock(tolist##_lock); \
if (_wakeup) \
cv_signal(tolist##_cond); \
} while (0)
#define QUEUE_WAIT(list) do { \
mtx_lock(list##_lock); \
while (TAILQ_EMPTY(list)) \

View File

@ -342,14 +342,7 @@ sender_disconnect(void)
pjdlog_warning("Disconnected from %s.", adhost->adh_remoteaddr);
/* Move all in-flight requests back onto free list. */
mtx_lock(&adist_free_list_lock);
mtx_lock(&adist_send_list_lock);
TAILQ_CONCAT(&adist_free_list, &adist_send_list, adr_next);
mtx_unlock(&adist_send_list_lock);
mtx_lock(&adist_recv_list_lock);
TAILQ_CONCAT(&adist_free_list, &adist_recv_list, adr_next);
mtx_unlock(&adist_recv_list_lock);
mtx_unlock(&adist_free_list_lock);
QUEUE_CONCAT2(&adist_free_list, &adist_send_list, &adist_recv_list);
}
static void
@ -609,9 +602,13 @@ recv_thread(void *arg __unused)
if (adhost->adh_remote == NULL) {
/*
* Connection is dead.
* XXX: We shouldn't be here.
* There is a short race in sender_disconnect() between
* setting adh_remote to NULL and removing entries from
* the recv list, which can result in us being here.
* To avoid just spinning, wait for 0.1s.
*/
rw_unlock(&adist_remote_lock);
usleep(100000);
continue;
}
if (proto_recv(adhost->adh_remote, &adrep,