Re: [PATCH 5.15 00/23] 5.15.160-rc1 review

On Thu, May 30, 2024 at 01:11:47PM +0100, Jon Hunter wrote:

On 29/05/2024 21:59, NeilBrown wrote:

...

...

Thanks for testing. I can only guess that you had an active NFSv4.1 mount and that the callback thread was causing problems. Please try this. I also changed to use freezable_schedule* which seems like a better interface to do the same thing.

If this doesn't fix it, we'll probably need to ask someone who remembers the old freezer code.

Thanks, NeilBrown

From 518f0c1150f988b3fe8e5e0d053a25c3aa6c7d44 Mon Sep 17 00:00:00 2001 From: NeilBrown neilb@suse.de Date: Wed, 29 May 2024 09:38:22 +1000 Subject: [PATCH] sunrpc: exclude from freezer when waiting for requests:

Prior to v6.1, the freezer will only wake a kernel thread from an uninterruptible sleep. Since we changed svc_get_next_xprt() to use and IDLE sleep the freezer cannot wake it. we need to tell the freezer to ignore it instead.

To make this work with only upstream requests we would need Commit f5d39b020809 ("freezer,sched: Rewrite core freezer logic") which allows non-interruptible sleeps to be woken by the freezer.

Fixes: 9b8a8e5e8129 ("nfsd: don't allow nfsd threads to be signalled.") Signed-off-by: NeilBrown neilb@suse.de

---

fs/nfs/callback.c | 2 +- fs/nfsd/nfs4proc.c | 3 ++- net/sunrpc/svc_xprt.c | 4 ++-- 3 files changed, 5 insertions(+), 4 deletions(-)

diff --git a/fs/nfs/callback.c b/fs/nfs/callback.c index 46a0a2d6962e..8fe143cad4a2 100644 --- a/fs/nfs/callback.c +++ b/fs/nfs/callback.c @@ -124,7 +124,7 @@ nfs41_callback_svc(void *vrqstp) } else { spin_unlock_bh(&serv->sv_cb_lock); if (!kthread_should_stop())

• 		schedule();
  

• 		freezable_schedule();
  finish_wait(&serv->sv_cb_waitq, &wq);
  

  } }

diff --git a/fs/nfsd/nfs4proc.c b/fs/nfsd/nfs4proc.c index 6779291efca9..e0ff2212866a 100644 --- a/fs/nfsd/nfs4proc.c +++ b/fs/nfsd/nfs4proc.c @@ -38,6 +38,7 @@ #include <linux/slab.h> #include <linux/kthread.h> #include <linux/namei.h> +#include <linux/freezer.h> #include <linux/sunrpc/addr.h> #include <linux/nfs_ssc.h> @@ -1322,7 +1323,7 @@ static __be32 nfsd4_ssc_setup_dul(struct nfsd_net *nn, char *ipaddr, /* allow 20secs for mount/unmount for now - revisit */ if (kthread_should_stop() ||

• 			(schedule_timeout(20*HZ) == 0)) {
  

• 			(freezable_schedule_timeout(20*HZ) == 0)) {
  	finish_wait(&nn->nfsd_ssc_waitq, &wait);
  	kfree(work);
  	return nfserr_eagain;
  

diff --git a/net/sunrpc/svc_xprt.c b/net/sunrpc/svc_xprt.c index b19592673eef..3cf53e3140a5 100644 --- a/net/sunrpc/svc_xprt.c +++ b/net/sunrpc/svc_xprt.c @@ -705,7 +705,7 @@ static int svc_alloc_arg(struct svc_rqst *rqstp) set_current_state(TASK_RUNNING); return -EINTR; }

• schedule_timeout(msecs_to_jiffies(500));
  

• freezable_schedule_timeout(msecs_to_jiffies(500));
  

  } rqstp->rq_page_end = &rqstp->rq_pages[pages]; rqstp->rq_pages[pages] = NULL; /* this might be seen in nfsd_splice_actor() */

@@ -765,7 +765,7 @@ static struct svc_xprt *svc_get_next_xprt(struct svc_rqst *rqstp, long timeout) smp_mb__after_atomic(); if (likely(rqst_should_sleep(rqstp)))

• time_left = schedule_timeout(timeout);
  

• time_left = freezable_schedule_timeout(timeout);
  

  else __set_current_state(TASK_RUNNING);

That did the trick! Suspend is now working again on top of v5.15.160-rc1 with this change.

Feel free to add my ...

Tested-by: Jon Hunter jonathanh@nvidia.com

Since I've heard no objections, I've added this to nfsd-5.15.y for testing. I plan to send it along to stable@ when testing completes.