Skip to contentSkip to conversation

NFS: Delay Retry Proposed Changes

PATCH In review Wenle Chen 3 participants 7 messages Nov 2, 2020 – Nov 4, 2020

Subsystems (2)
lore
Share & tools
lore

Wenle Chen

author
#0 lore raw

cover NFS: Delay Retry Proposed Changes

Little modify for nfs4_lock_delegation_recall delay retry

Wenle Chen (2):
NFS: Reduce redundant comparison
NFS: Limit the number of retries

 fs/nfs/nfs4proc.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)
Signature · 1 line
2.29.1
author
#1 lore raw

PATCH 1/2 NFS: Reduce redundant comparison

The err wouldn't be change after sleep. There is no need
to compare err and -NFS4ERR_DELAY

---
 fs/nfs/nfs4proc.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
fs/nfs/nfs4proc.c +1 -1
index 9e0ca9b2b210..f6b5dc792b33 100644
@@ -7399,7 +7399,7 @@ int nfs4_lock_delegation_recall(struct file_lock *fl, struct nfs4_state *state,
 		if (err != -NFS4ERR_DELAY)
 			break;
 		ssleep(1);
-	} while (err == -NFS4ERR_DELAY);
+	} while (1);
 	return nfs4_handle_delegation_recall_error(server, state, stateid, fl, err);
 }
 
Signature · 1 line
2.29.1
author
#2 lore raw

PATCH 2/2 NFS: Limit the number of retries

In reply to Wenle Chen

We can't wait forever, even if the state
is always delayed.

---
 fs/nfs/nfs4proc.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)
fs/nfs/nfs4proc.c +3 -1
index f6b5dc792b33..bb2316bf13f6 100644
@@ -7390,15 +7390,17 @@ int nfs4_lock_delegation_recall(struct file_lock *fl, struct nfs4_state *state,
 {
 	struct nfs_server *server = NFS_SERVER(state->inode);
 	int err;
+	int retry = 3;
 
 	err = nfs4_set_lock_state(state, fl);
 	if (err != 0)
 		return err;
 	do {
 		err = _nfs4_do_setlk(state, F_SETLK, fl, NFS_LOCK_NEW);
-		if (err != -NFS4ERR_DELAY)
+		if (err != -NFS4ERR_DELAY || retry == 0)
 			break;
 		ssleep(1);
+		--retry;
 	} while (1);
 	return nfs4_handle_delegation_recall_error(server, state, stateid, fl, err);
 }
Signature · 1 line
2.29.1
#3 lore raw

Re: [PATCH 2/2] NFS: Limit the number of retries

Quoted 26 lines · Wenle Chen

  We can't wait forever, even if the state
is always delayed.

---
 fs/nfs/nfs4proc.c | 4 +++-

 1 file changed, 3 insertions(+), 1 deletion(-)
fs/nfs/nfs4proc.c +1 -0
index f6b5dc792b33..bb2316bf13f6 100644
@@ -7390,15 +7390,17 @@ int nfs4_lock_delegation_recall(struct
 file_lock *fl, struct nfs4_state *state,
  {
         struct nfs_server *server = NFS_SERVER(state->inode);
         int err;
+       int retry = 3;


        err = nfs4_set_lock_state(state, fl); if (err != 0)
                return err;
        do {
                err = _nfs4_do_setlk(state, F_SETLK, fl, NFS_LOCK_NEW);
-               if (err != -NFS4ERR_DELAY)

Quoted 8 lines

+               if (err != -NFS4ERR_DELAY || retry == 0) break;
                ssleep(1);
+               --retry;
        } while (1);
        return nfs4_handle_delegation_recall_error(server, state, stateid, fl, err);
 }

This patch will just cause the locks to be silently lost, no?

Signature · 3 lines
Trond Myklebust
Linux NFS client maintainer, Hammerspace
[email protected]
author
#4 lore raw

Re: [PATCH 2/2] NFS: Limit the number of retries

Trond Myklebust 於 2020/11/3 上午1:45 寫道:

Quoted 29 lines · Wenle Chen
Quoted 28 lines

  We can't wait forever, even if the state
is always delayed.

---
  fs/nfs/nfs4proc.c | 4 +++-
  1 file changed, 3 insertions(+), 1 deletion(-)
fs/nfs/nfs4proc.c +2 -1
index f6b5dc792b33..bb2316bf13f6 100644
@@ -7390,15 +7390,17 @@ int nfs4_lock_delegation_recall(struct
 file_lock *fl, struct nfs4_state *state,
  {
         struct nfs_server *server = NFS_SERVER(state->inode);
         int err;
+       int retry = 3;
 
         err = nfs4_set_lock_state(state, fl);
         if (err != 0)
                 return err;
         do {
                 err = _nfs4_do_setlk(state, F_SETLK, fl,
 NFS_LOCK_NEW);
-               if (err != -NFS4ERR_DELAY)
+               if (err != -NFS4ERR_DELAY || retry == 0)
                         break;
Quoted 9 lines
Quoted 6 lines
fs/nfs/nfs4proc.c +1 -0
                 ssleep(1);
+               --retry;
         } while (1);

        return nfs4_handle_delegation_recall_error(server, state,
stateid, fl, err);
 }

This patch will just cause the locks to be silently lost, no?

This loop was introduced in commit 3d7a9520f0c3e to simplify the delay retry loop. Before this, the function nfs4_lock_delegation_recall would return a -EAGAIN to do a whole retry loop.

When we retried three times and waited three seconds, it was still in delay. I think we can get a whole loop and check the other points if it was changed or not. It is just a proposal.

#5 lore raw

Re: [PATCH 2/2] NFS: Limit the number of retries

Quoted 42 lines · Wenle Chen

Trond Myklebust 於 2020/11/3 上午1:45 寫道:

Quoted 29 lines · Wenle Chen
Quoted 28 lines

We can't wait forever, even if the state
is always delayed.

---
  fs/nfs/nfs4proc.c | 4 +++-
  1 file changed, 3 insertions(+), 1 deletion(-)
fs/nfs/nfs4proc.c +2 -1
index f6b5dc792b33..bb2316bf13f6 100644
@@ -7390,15 +7390,17 @@ int nfs4_lock_delegation_recall(struct
 file_lock *fl, struct nfs4_state *state,
  {
         struct nfs_server *server = NFS_SERVER(state->inode);
         int err;
+       int retry = 3;
 
         err = nfs4_set_lock_state(state, fl);
         if (err != 0)
                 return err;
         do {
                 err = _nfs4_do_setlk(state, F_SETLK, fl,
 NFS_LOCK_NEW);
-               if (err != -NFS4ERR_DELAY)
+               if (err != -NFS4ERR_DELAY || retry == 0)
                         break;
Quoted 9 lines
Quoted 6 lines
fs/nfs/nfs4proc.c +1 -0
                 ssleep(1);
+               --retry;
         } while (1);

return nfs4_handle_delegation_recall_error(server, state,
stateid, fl, err);
}

This patch will just cause the locks to be silently lost, no?

Quoted 3 lines

This loop was introduced in commit 3d7a9520f0c3e to simplify the delay retry loop. Before this, the function nfs4_lock_delegation_recall would return a -EAGAIN to do a whole retry loop.

This commit was not simplifying retry but actually handling the error. Without it the error isn't handled and client falsely thinks it holds the lock. Limiting the number of retries as Trond points out would lead to the same problem which in the end is data corruption. Alternative would be to fail the application. However ERR_DELAY is a transient error and the server would, when ready, return something else. If server is broken and continues to do so then the server needs to be fix (client isn't coded to the broken server). I don't see a good argument for limiting the number of re-tries.

Quoted 3 lines

When we retried three times and waited three seconds, it was still in delay. I think we can get a whole loop and check the other points if it was changed or not. It is just a proposal.

author
#6 lore raw

Re: [PATCH 2/2] NFS: Limit the number of retries

Olga Kornievskaia 於 2020/11/4 下午9:22 寫道:

Quoted 47 lines · Wenle Chen
Quoted 42 lines

Trond Myklebust 於 2020/11/3 上午1:45 寫道:

Quoted 29 lines · Wenle Chen
Quoted 28 lines

We can't wait forever, even if the state
is always delayed.

---
   fs/nfs/nfs4proc.c | 4 +++-
   1 file changed, 3 insertions(+), 1 deletion(-)
fs/nfs/nfs4proc.c +2 -1
index f6b5dc792b33..bb2316bf13f6 100644
@@ -7390,15 +7390,17 @@ int nfs4_lock_delegation_recall(struct
 file_lock *fl, struct nfs4_state *state,
   {
          struct nfs_server *server = NFS_SERVER(state->inode);
          int err;
+       int retry = 3;
 
          err = nfs4_set_lock_state(state, fl);
          if (err != 0)
                  return err;
          do {
                  err = _nfs4_do_setlk(state, F_SETLK, fl,
 NFS_LOCK_NEW);
-               if (err != -NFS4ERR_DELAY)
+               if (err != -NFS4ERR_DELAY || retry == 0)
                          break;
Quoted 9 lines
Quoted 6 lines
fs/nfs/nfs4proc.c +1 -0
                  ssleep(1);
+               --retry;
          } while (1);

return nfs4_handle_delegation_recall_error(server, state,

stateid, fl, err);
   }

This patch will just cause the locks to be silently lost, no?

Quoted 3 lines

This loop was introduced in commit 3d7a9520f0c3e to simplify the delay retry loop. Before this, the function nfs4_lock_delegation_recall would return a -EAGAIN to do a whole retry loop.

Quoted 13 lines

This commit was not simplifying retry but actually handling the error. Without it the error isn't handled and client falsely thinks it holds the lock. Limiting the number of retries as Trond points out would lead to the same problem which in the end is data corruption. Alternative would be to fail the application. However ERR_DELAY is a transient error and the server would, when ready, return something else. If server is broken and continues to do so then the server needs to be fix (client isn't coded to the broken server). I don't see a good argument for limiting the number of re-tries.

Quoted 3 lines

When we retried three times and waited three seconds, it was still in delay. I think we can get a whole loop and check the other points if it was changed or not. It is just a proposal.

In the function nfs_end_delegation_return, it would get the return err=-EAGAIN and check the client is active and get a retry. I has so thought. Maybe I think wrong. I will understand more carefully. Thinks.