3 commits - src/host_id.c src/log.c src/main.c src/sanlock_internal.h src/sanlock_rv.h

[David Teigland]

 src/host_id.c          |   36 ++++++++++++++++++++++++++++++++++++
 src/log.c              |    2 +-
 src/main.c             |    2 +-
 src/sanlock_internal.h |    1 +
 src/sanlock_rv.h       |    3 ++-
 5 files changed, 41 insertions(+), 3 deletions(-)

New commits:
commit d852c084b7d46915dab20ab641945a6aa2c166b7
Author: David Teigland <teigland(a)redhat.com&gt;
Date:   Thu Jun 30 12:37:58 2011 -0500

    sanlock: fail host_id when corrupted
    
    If we read our host_id for renewal and find it's corrupted,
    there's no point in retrying for a while, we should treat it
    as failed right away.

diff --git a/src/host_id.c b/src/host_id.c
index b2c64c2..142f61b 100644
--- a/src/host_id.c
+++ b/src/host_id.c
@@ -53,6 +53,7 @@ int print_space_state(struct space *sp, char *str)
 		 "host_generation=%llu "
 		 "space_dead=%d "
 		 "killing_pids=%d "
+		 "corrupt_result=%d "
 		 "acquire_last_result=%d "
 		 "renewal_last_result=%d "
 		 "acquire_last_attempt=%llu "
@@ -63,6 +64,7 @@ int print_space_state(struct space *sp, char *str)
 		 (unsigned long long)sp->host_generation,
 		 sp->space_dead,
 		 sp->killing_pids,
+		 sp->lease_status.corrupt_result,
 		 sp->lease_status.acquire_last_result,
 		 sp->lease_status.renewal_last_result,
 		 (unsigned long long)sp->lease_status.acquire_last_attempt,
@@ -171,12 +173,19 @@ int host_id_disk_info(char *name, struct sync_disk *disk)
 int host_id_check(struct task *task, struct space *sp)
 {
 	uint64_t last_success;
+	int corrupt_result;
 	int gap;
 
 	pthread_mutex_lock(&sp->mutex);
 	last_success = sp->lease_status.renewal_last_success;
+	corrupt_result = sp->lease_status.corrupt_result;
 	pthread_mutex_unlock(&sp->mutex);
 
+	if (corrupt_result) {
+		log_erros(sp, "host_id_check corrupt %d", corrupt_result);
+		return 0;
+	}
+
 	gap = time(NULL) - last_success;
 
 	if (gap >= task->id_renewal_fail_seconds) {
@@ -197,6 +206,27 @@ int host_id_check(struct task *task, struct space *sp)
 	return 1;
 }
 
+/* If a renewal result is one of the listed errors, it means our
+   delta lease has been corrupted/overwritten/reinitialized out from
+   under us, and we should stop using it immediately.  There's no
+   point in retrying the renewal. */
+
+static int corrupt_result(int result)
+{
+	switch (result) {
+	case SANLK_RENEW_OWNER:
+	case SANLK_RENEW_DIFF:
+	case SANLK_LEADER_MAGIC:
+	case SANLK_LEADER_VERSION:
+	case SANLK_LEADER_SECTORSIZE:
+	case SANLK_LEADER_LOCKSPACE:
+	case SANLK_LEADER_CHECKSUM:
+		return result;
+	default:
+		return 0;
+	}
+}
+
 static void *lockspace_thread(void *arg_in)
 {
 	struct task task;
@@ -316,6 +346,12 @@ static void *lockspace_thread(void *arg_in)
 			log_erros(sp, "renewal error %d delta_length %d last_success %llu",
 				  result, delta_length,
 				  (unsigned long long)sp->lease_status.renewal_last_success);
+
+			if (!sp->lease_status.corrupt_result) {
+				sp->lease_status.corrupt_result = corrupt_result(result);
+				log_erros(sp, "renewal error %d is corruption",
+					  sp->lease_status.corrupt_result);
+			}
 		}
 		stop = sp->thread_stop;
 		pthread_mutex_unlock(&sp->mutex);
diff --git a/src/sanlock_internal.h b/src/sanlock_internal.h
index c38ebf8..cb7cf60 100644
--- a/src/sanlock_internal.h
+++ b/src/sanlock_internal.h
@@ -94,6 +94,7 @@ struct token {
 };
 
 struct lease_status {
+	int corrupt_result;
 	int acquire_last_result;
 	int renewal_last_result;
 	uint64_t acquire_last_attempt;
diff --git a/src/sanlock_rv.h b/src/sanlock_rv.h
index 9683477..8133efd 100644
--- a/src/sanlock_rv.h
+++ b/src/sanlock_rv.h
@@ -22,7 +22,8 @@
 #define SANLK_DBLOCK_MBAL	-213
 #define SANLK_DBLOCK_CHECKSUM	-214
 
-/* verify_leader, leader_read, leader_write (paxos or delta) */
+/* verify_leader, leader_read, leader_write (paxos or delta)
+   (when adding to list, check if it should be a corrupt_result()) */
 
 #define SANLK_LEADER_READ	-220
 #define SANLK_LEADER_WRITE	-221


commit c8130f398bdb1aa6bab088e1c9a4e5d745e0e9b7
Author: David Teigland <teigland(a)redhat.com&gt;
Date:   Tue Jun 28 17:33:31 2011 -0500

    sanlock: change a log_error to log_debug
    
    for a non error case we shouldn't need to worry about

diff --git a/src/main.c b/src/main.c
index 5918c6c..70e2630 100644
--- a/src/main.c
+++ b/src/main.c
@@ -652,7 +652,7 @@ static void client_recv_all(int ci, struct sm_header *h_recv, int
pos)
 			break;
 	}
 
-	log_error("recv_all %d,%d,%d pos %d rv %d error %d rem %d total %d",
+	log_debug("recv_all %d,%d,%d pos %d rv %d error %d rem %d total %d",
 		  ci, client[ci].fd, client[ci].pid, pos, rv, error, rem, total);
 }
 


commit 06674b31739d84bf6bc8a484552f4e1de85cab90
Author: David Teigland <teigland(a)redhat.com&gt;
Date:   Tue Jun 28 17:31:39 2011 -0500

    sanlock: increase log line to 512 bytes

diff --git a/src/log.c b/src/log.c
index 884507e..e753ee6 100644
--- a/src/log.c
+++ b/src/log.c
@@ -25,7 +25,7 @@
 #include "sanlock_internal.h"
 #include "log.h"
 
-#define LOG_STR_LEN 256
+#define LOG_STR_LEN 512
 static char log_str[LOG_STR_LEN];
 
 static pthread_t thread_handle;