[sanlock] branch master updated: sanlock: retry release on EIO

This is an automated email from the git hooks/post-receive script.

teigland pushed a commit to branch master
in repository sanlock.

The following commit(s) were added to refs/heads/master by this push:
     new 4c32c1d  sanlock: retry release on EIO
4c32c1d is described below

commit 4c32c1d763727552cf984502827e22d795a61c10
Author: David Teigland 
AuthorDate: Mon Mar 3 13:43:50 2025 -0600

sanlock: retry release on EIO

Releasing a lease on disk would be retried if it failed
    from an io timeout.  It should also retry if it fails with
    EIO, in case the i/o errors are transient.
---
 src/resource.c | 26 +++++++++++++-------------
 1 file changed, 13 insertions(+), 13 deletions(-)

diff --git a/src/resource.c b/src/resource.c
index 5fb5183..ee8b9dc 100644
--- a/src/resource.c
+++ b/src/resource.c
@@ -1185,7 +1185,7 @@ static int _release_token(struct task *task, struct token *token,
    		ret = rv;
    	}

-		if (rv == SANLK_AIO_TIMEOUT)
+		if (rv == SANLK_AIO_TIMEOUT || rv == -EIO)
    		retry_async = 1;

/* Even when acquire did not get far enough to get a copy of the
@@ -1200,7 +1200,7 @@ static int _release_token(struct task *task, struct token *token,
    	if (rv < 0)
    		ret = rv;

-		if (rv == SANLK_AIO_TIMEOUT)
+		if (rv == SANLK_AIO_TIMEOUT || rv == -EIO)
    		retry_async = 1;

/* want to see this result in sanlock.log but not worry people with error */
@@ -1216,7 +1216,7 @@ static int _release_token(struct task *task, struct token *token,
    		ret = rv;
    	}

-		if (rv == SANLK_AIO_TIMEOUT)
+		if (rv == SANLK_AIO_TIMEOUT || rv == -EIO)
    		retry_async = 1;

rv = release_disk(task, token, resrename, &r->leader);
@@ -1225,7 +1225,7 @@ static int _release_token(struct task *task, struct token *token,
    		ret = rv;
    	}

-		if (rv == SANLK_AIO_TIMEOUT)
+		if (rv == SANLK_AIO_TIMEOUT || rv == -EIO)
    		retry_async = 1;

} else if (r_flags & R_SHARED) {
@@ -1237,7 +1237,7 @@ static int _release_token(struct task *task, struct token *token,
    		ret = rv;
    	}

-		if (rv == SANLK_AIO_TIMEOUT)
+		if (rv == SANLK_AIO_TIMEOUT || rv == -EIO)
    		retry_async = 1;

} else {
@@ -1271,7 +1271,7 @@ static int _release_token(struct task *task, struct token *token,
    		ret = rv;
    	}

-		if (rv == SANLK_AIO_TIMEOUT)
+		if (rv == SANLK_AIO_TIMEOUT || rv == -EIO)
    		retry_async = 1;
    }

@@ -1301,7 +1301,7 @@ static int _release_token(struct task *task, struct token *token,
    pthread_mutex_lock(&resource_mutex);
    r->flags |= R_THREAD_RELEASE;
    pthread_mutex_unlock(&resource_mutex);
-	return SANLK_AIO_TIMEOUT;
+	return ret;
 }

static int release_token_nodisk_opened(struct task *task, struct token *token)
@@ -2465,7 +2465,7 @@ static void resource_thread_release(struct task *task, struct resource *r, struc
    	if (rv < 0)
    		log_errot(token, "release async erase all write_host_block %d", rv);

-		if (rv == SANLK_AIO_TIMEOUT)
+		if (rv == SANLK_AIO_TIMEOUT || rv == -EIO)
    		retry_async = 1;

/* Even when acquire did not get far enough to get a copy of the
@@ -2477,7 +2477,7 @@ static void resource_thread_release(struct task *task, struct resource *r, struc
    	else
    		rv = paxos_lease_release(task, token, NULL, &r->leader, &leader);

-		if (rv == SANLK_AIO_TIMEOUT)
+		if (rv == SANLK_AIO_TIMEOUT || rv == -EIO)
    		retry_async = 1;

/* want to see this result in sanlock.log but not worry people with error */
@@ -2489,14 +2489,14 @@ static void resource_thread_release(struct task *task, struct resource *r, struc
    	if (rv < 0)
    		log_errot(token, "release async undo shared write_host_block %d", rv);

-		if (rv == SANLK_AIO_TIMEOUT)
+		if (rv == SANLK_AIO_TIMEOUT || rv == -EIO)
    		retry_async = 1;

rv = release_disk(task, token, NULL, &r->leader);
    	if (rv < 0)
    		log_errot(token, "release async undo shared release leader %d", rv);

-		if (rv == SANLK_AIO_TIMEOUT)
+		if (rv == SANLK_AIO_TIMEOUT || rv == -EIO)
    		retry_async = 1;

} else if (r_flags & R_SHARED) {
@@ -2506,7 +2506,7 @@ static void resource_thread_release(struct task *task, struct resource *r, struc
    	if (rv < 0)
    		log_errot(token, "release async shared write_host_block %d", rv);

-		if (rv == SANLK_AIO_TIMEOUT)
+		if (rv == SANLK_AIO_TIMEOUT || rv == -EIO)
    		retry_async = 1;
    } else {
    	/* normal release of ex lease */
@@ -2529,7 +2529,7 @@ static void resource_thread_release(struct task *task, struct resource *r, struc
    	if (rv < 0)
    		log_errot(token, "release async release leader %d", rv);

-		if (rv == SANLK_AIO_TIMEOUT)
+		if (rv == SANLK_AIO_TIMEOUT || rv == -EIO)
    		retry_async = 1;
    }