[PATCH 0/2] nfsd: fix handling of error from unshare_fs_struct()

[PATCH 0/2] nfsd: fix handling of error from unshare_fs_struct()

[NeilBrown]

These two patches replace my previous patch:
 [PATCH 07/14] Change unshare_fs_struct() to never fail.

I had explored ways to change kthread_create() to avoid the need for
GFP_NOFAIL and concluded that we can do everything we need in the sunrpc
layer.  So the first patch here is a simple cleanup, and the second adds
simple infrastructure for an svc thread to confirm that it has started
up and to report if it was successful in that.

Thanks,
NeilBrown



 [PATCH 1/2] sunrpc: merge svc_rqst_alloc() into svc_prepare_thread()
 [PATCH 2/2] sunrpc: allow svc threads to fail initialisation cleanly

[PATCH 1/2] sunrpc: merge svc_rqst_alloc() into svc_prepare_thread()

[NeilBrown]

From: NeilBrown <neil@brown.name>

The only caller of svc_rqst_alloc() is svc_prepare_thread().  So merge
the one into the other and simplify.

Signed-off-by: NeilBrown <neil@brown.name>
---
 include/linux/sunrpc/svc.h |  2 --
 net/sunrpc/svc.c           | 28 ++++++++--------------------
 2 files changed, 8 insertions(+), 22 deletions(-)

diff --git a/include/linux/sunrpc/svc.h b/include/linux/sunrpc/svc.h
index 23617da0e565..01383ea077ad 100644
--- a/include/linux/sunrpc/svc.h
+++ b/include/linux/sunrpc/svc.h
@@ -403,8 +403,6 @@ void svc_rpcb_cleanup(struct svc_serv *serv, struct net *net);
 int svc_bind(struct svc_serv *serv, struct net *net);
 struct svc_serv *svc_create(struct svc_program *, unsigned int,
 			    int (*threadfn)(void *data));
-struct svc_rqst *svc_rqst_alloc(struct svc_serv *serv,
-					struct svc_pool *pool, int node);
 bool		   svc_rqst_replace_page(struct svc_rqst *rqstp,
 					 struct page *page);
 void		   svc_rqst_release_pages(struct svc_rqst *rqstp);
diff --git a/net/sunrpc/svc.c b/net/sunrpc/svc.c
index d9cda1e53a01..c0e91bea72e7 100644
--- a/net/sunrpc/svc.c
+++ b/net/sunrpc/svc.c
@@ -633,8 +633,8 @@ svc_release_buffer(struct svc_rqst *rqstp)
 			put_page(rqstp->rq_pages[i]);
 }
 
-struct svc_rqst *
-svc_rqst_alloc(struct svc_serv *serv, struct svc_pool *pool, int node)
+static struct svc_rqst *
+svc_prepare_thread(struct svc_serv *serv, struct svc_pool *pool, int node)
 {
 	struct svc_rqst	*rqstp;
 
@@ -662,22 +662,6 @@ svc_rqst_alloc(struct svc_serv *serv, struct svc_pool *pool, int node)
 	if (!svc_init_buffer(rqstp, serv->sv_max_mesg, node))
 		goto out_enomem;
 
-	return rqstp;
-out_enomem:
-	svc_rqst_free(rqstp);
-	return NULL;
-}
-EXPORT_SYMBOL_GPL(svc_rqst_alloc);
-
-static struct svc_rqst *
-svc_prepare_thread(struct svc_serv *serv, struct svc_pool *pool, int node)
-{
-	struct svc_rqst	*rqstp;
-
-	rqstp = svc_rqst_alloc(serv, pool, node);
-	if (!rqstp)
-		return ERR_PTR(-ENOMEM);
-
 	spin_lock_bh(&serv->sv_lock);
 	serv->sv_nrthreads += 1;
 	spin_unlock_bh(&serv->sv_lock);
@@ -690,6 +674,10 @@ svc_prepare_thread(struct svc_serv *serv, struct svc_pool *pool, int node)
 	list_add_rcu(&rqstp->rq_all, &pool->sp_all_threads);
 
 	return rqstp;
+
+out_enomem:
+	svc_rqst_free(rqstp);
+	return NULL;
 }
 
 /**
@@ -779,8 +767,8 @@ svc_start_kthreads(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
 		node = svc_pool_map_get_node(chosen_pool->sp_id);
 
 		rqstp = svc_prepare_thread(serv, chosen_pool, node);
-		if (IS_ERR(rqstp))
-			return PTR_ERR(rqstp);
+		if (!rqstp)
+			return -ENOMEM;
 		task = kthread_create_on_node(serv->sv_threadfn, rqstp,
 					      node, "%s", serv->sv_name);
 		if (IS_ERR(task)) {
-- 
2.44.0

[PATCH 2/2] sunrpc: allow svc threads to fail initialisation cleanly

[NeilBrown]

If an svc thread needs to perform some initialisation that might fail,
it has no good way to handle the failure.

Before the thread can exit it must call svc_exit_thread(), but that
requires the service mutex to be held.  The thread cannot simply take
the mutex as that could deadlock if there is a concurrent attempt to
shut down all threads (which is unlikely, but not impossible).

nfsd currently call svc_exit_thread() unprotected in the unlikely event
that unshare_fs_struct() fails.

We can clean this up by introducing svc_thread_init_status() by which an
svc thread can report whether initialisation has succeeded.  If it has,
it continues normally into the action loop.  If it has not,
svc_thread_init_status() immediately aborts the thread.
svc_start_kthread() waits for either of these to happen, and calls
svc_exit_thread() (under the mutex) if the thread aborted.

Signed-off-by: NeilBrown <neilb@suse.de>
---
 fs/lockd/svc.c             |  2 ++
 fs/nfs/callback.c          |  2 ++
 fs/nfsd/nfssvc.c           |  9 +++------
 include/linux/sunrpc/svc.h | 28 ++++++++++++++++++++++++++++
 net/sunrpc/svc.c           | 11 +++++++++++
 5 files changed, 46 insertions(+), 6 deletions(-)

diff --git a/fs/lockd/svc.c b/fs/lockd/svc.c
index ab8042a5b895..7d74858e1abe 100644
--- a/fs/lockd/svc.c
+++ b/fs/lockd/svc.c
@@ -125,6 +125,8 @@ lockd(void *vrqstp)
 	struct net *net = &init_net;
 	struct lockd_net *ln = net_generic(net, lockd_net_id);
 
+	svc_thread_init_status(rqstp, 0);
+
 	/* try_to_freeze() is called from svc_recv() */
 	set_freezable();
 
diff --git a/fs/nfs/callback.c b/fs/nfs/callback.c
index 8adfcd4c8c1a..6cf92498a5ac 100644
--- a/fs/nfs/callback.c
+++ b/fs/nfs/callback.c
@@ -76,6 +76,8 @@ nfs4_callback_svc(void *vrqstp)
 {
 	struct svc_rqst *rqstp = vrqstp;
 
+	svc_thread_init_status(rqstp, 0);
+
 	set_freezable();
 
 	while (!svc_thread_should_stop(rqstp))
diff --git a/fs/nfsd/nfssvc.c b/fs/nfsd/nfssvc.c
index 89d7918de7b1..2a99787d6a86 100644
--- a/fs/nfsd/nfssvc.c
+++ b/fs/nfsd/nfssvc.c
@@ -915,11 +915,9 @@ nfsd(void *vrqstp)
 
 	/* At this point, the thread shares current->fs
 	 * with the init process. We need to create files with the
-	 * umask as defined by the client instead of init's umask. */
-	if (unshare_fs_struct() < 0) {
-		printk("Unable to start nfsd thread: out of memory\n");
-		goto out;
-	}
+	 * umask as defined by the client instead of init's umask.
+	 */
+	svc_thread_init_status(rqstp, unshare_fs_struct());
 
 	current->fs->umask = 0;
 
@@ -941,7 +939,6 @@ nfsd(void *vrqstp)
 
 	atomic_dec(&nfsd_th_cnt);
 
-out:
 	/* Release the thread */
 	svc_exit_thread(rqstp);
 	return 0;
diff --git a/include/linux/sunrpc/svc.h b/include/linux/sunrpc/svc.h
index 01383ea077ad..2c4366afc9bd 100644
--- a/include/linux/sunrpc/svc.h
+++ b/include/linux/sunrpc/svc.h
@@ -21,6 +21,7 @@
 #include <linux/wait.h>
 #include <linux/mm.h>
 #include <linux/pagevec.h>
+#include <linux/kthread.h>
 
 /*
  *
@@ -231,6 +232,11 @@ struct svc_rqst {
 	struct net		*rq_bc_net;	/* pointer to backchannel's
 						 * net namespace
 						 */
+
+	int			rq_err;		/* Thread sets this to inidicate
+						 * initialisation success.
+						 */
+
 	unsigned long	bc_to_initval;
 	unsigned int	bc_to_retries;
 	void **			rq_lease_breaker; /* The v4 client breaking a lease */
@@ -304,6 +310,28 @@ static inline bool svc_thread_should_stop(struct svc_rqst *rqstp)
 	return test_bit(RQ_VICTIM, &rqstp->rq_flags);
 }
 
+/**
+ * svc_thread_init_status - report whether thread has initialised successfully
+ * @rqstp: the thread in question
+ * @err: errno code
+ *
+ * After performing any initialisation that could fail, and before starting
+ * normal work, each sunrpc svc_thread must call svc_thread_init_status()
+ * with an appropriate error, or zero.
+ *
+ * If zero is passed, the thread is ready and must continue until
+ * svc_thread_should_stop() returns true.  If a non-zero error is passed
+ * the call will not return - the thread will exit.
+ */
+static inline void svc_thread_init_status(struct svc_rqst *rqstp, int err)
+{
+	/* store_release ensures svc_start_kthreads() sees the error */
+	smp_store_release(&rqstp->rq_err, err);
+	wake_up_var(&rqstp->rq_err);
+	if (err)
+		kthread_exit(1);
+}
+
 struct svc_deferred_req {
 	u32			prot;	/* protocol (UDP or TCP) */
 	struct svc_xprt		*xprt;
diff --git a/net/sunrpc/svc.c b/net/sunrpc/svc.c
index c0e91bea72e7..0fc9c97ae419 100644
--- a/net/sunrpc/svc.c
+++ b/net/sunrpc/svc.c
@@ -662,6 +662,8 @@ svc_prepare_thread(struct svc_serv *serv, struct svc_pool *pool, int node)
 	if (!svc_init_buffer(rqstp, serv->sv_max_mesg, node))
 		goto out_enomem;
 
+	rqstp->rq_err = -EAGAIN; /* No error yet */
+
 	spin_lock_bh(&serv->sv_lock);
 	serv->sv_nrthreads += 1;
 	spin_unlock_bh(&serv->sv_lock);
@@ -760,6 +762,7 @@ svc_start_kthreads(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
 	struct svc_pool *chosen_pool;
 	unsigned int state = serv->sv_nrthreads-1;
 	int node;
+	int err;
 
 	do {
 		nrservs--;
@@ -782,6 +785,14 @@ svc_start_kthreads(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
 
 		svc_sock_update_bufs(serv);
 		wake_up_process(task);
+
+		/* load_acquire ensures we get value stored in svc_thread_init_status() */
+		wait_var_event(&rqstp->rq_err, smp_load_acquire(&rqstp->rq_err) != -EAGAIN);
+		err = rqstp->rq_err;
+		if (err) {
+			svc_exit_thread(rqstp);
+			return err;
+		}
 	} while (nrservs > 0);
 
 	return 0;
-- 
2.44.0

Re: [PATCH 0/2] nfsd: fix handling of error from unshare_fs_struct()

[Jeff Layton]

On Mon, 2024-07-29 at 12:18 +1000, NeilBrown wrote:
> These two patches replace my previous patch:
>  [PATCH 07/14] Change unshare_fs_struct() to never fail.
> 
> I had explored ways to change kthread_create() to avoid the need for
> GFP_NOFAIL and concluded that we can do everything we need in the
> sunrpc
> layer.  So the first patch here is a simple cleanup, and the second
> adds
> simple infrastructure for an svc thread to confirm that it has
> started
> up and to report if it was successful in that.
> 
> Thanks,
> NeilBrown
> 
> 
> 
>  [PATCH 1/2] sunrpc: merge svc_rqst_alloc() into svc_prepare_thread()
>  [PATCH 2/2] sunrpc: allow svc threads to fail initialisation cleanly

I like this solution better:

Reviewed-by: Jeff Layton <jlayton@kernel.org>

Re: [PATCH 0/2] nfsd: fix handling of error from unshare_fs_struct()

[Chuck Lever]

On Mon, Jul 29, 2024 at 12:18:19PM +1000, NeilBrown wrote:
> These two patches replace my previous patch:
>  [PATCH 07/14] Change unshare_fs_struct() to never fail.
> 
> I had explored ways to change kthread_create() to avoid the need for
> GFP_NOFAIL and concluded that we can do everything we need in the sunrpc
> layer.  So the first patch here is a simple cleanup, and the second adds
> simple infrastructure for an svc thread to confirm that it has started
> up and to report if it was successful in that.
> 
> Thanks,
> NeilBrown
> 
> 
> 
>  [PATCH 1/2] sunrpc: merge svc_rqst_alloc() into svc_prepare_thread()
>  [PATCH 2/2] sunrpc: allow svc threads to fail initialisation cleanly

This series does not apply to nfsd-next. It looks like 1/2 expects
to see the EXPORT_SYMBOL after svc_rqst_alloc() that you already
removed in "SUNRPC: make various functions static, or not exported."

Also, 1/2 is From: your brown.name account, but the SoB is your
SuSE email. (Maybe that doesn't matter).

Can you rebase and resend?

In 2/2, what is the reason to make svc_thread_init_status() a static
inline rather than an exported function? I don't think this is going
to be a performance hot path, but maybe it becomes one in a future
patch?


-- 
Chuck Lever

Re: [PATCH 0/2] nfsd: fix handling of error from unshare_fs_struct()

[NeilBrown]

On Tue, 30 Jul 2024, Chuck Lever wrote:
> On Mon, Jul 29, 2024 at 12:18:19PM +1000, NeilBrown wrote:
> > These two patches replace my previous patch:
> >  [PATCH 07/14] Change unshare_fs_struct() to never fail.
> > 
> > I had explored ways to change kthread_create() to avoid the need for
> > GFP_NOFAIL and concluded that we can do everything we need in the sunrpc
> > layer.  So the first patch here is a simple cleanup, and the second adds
> > simple infrastructure for an svc thread to confirm that it has started
> > up and to report if it was successful in that.
> > 
> > Thanks,
> > NeilBrown
> > 
> > 
> > 
> >  [PATCH 1/2] sunrpc: merge svc_rqst_alloc() into svc_prepare_thread()
> >  [PATCH 2/2] sunrpc: allow svc threads to fail initialisation cleanly
> 
> This series does not apply to nfsd-next. It looks like 1/2 expects
> to see the EXPORT_SYMBOL after svc_rqst_alloc() that you already
> removed in "SUNRPC: make various functions static, or not exported."
> 
> Also, 1/2 is From: your brown.name account, but the SoB is your
> SuSE email. (Maybe that doesn't matter).

Probably don't matter.  That happened because I wrote that patch on my
notebook instead of my desktop and they have different defaults.  I'll
try to remember that for next time.

> 
> Can you rebase and resend?

I can't see "SUNRPC: make various functions static, or not exported." in
    git://git.kernel.org/pub/scm/linux/kernel/git/cel/linux 
Am I looking in the wrong place? If not, can you push out nfsd-next?
(and if you could delete "for-next" from there, and possibly other old
cruft, that might help too)

Thanks,
NeilBrown


> 
> In 2/2, what is the reason to make svc_thread_init_status() a static
> inline rather than an exported function? I don't think this is going
> to be a performance hot path, but maybe it becomes one in a future
> patch?
> 
> 
> -- 
> Chuck Lever
> 

Re: [PATCH 0/2] nfsd: fix handling of error from unshare_fs_struct()

[Chuck Lever III]

> On Jul 29, 2024, at 5:07 PM, NeilBrown <neilb@suse.de> wrote:
> 
> On Tue, 30 Jul 2024, Chuck Lever wrote:
>> On Mon, Jul 29, 2024 at 12:18:19PM +1000, NeilBrown wrote:
>>> These two patches replace my previous patch:
>>> [PATCH 07/14] Change unshare_fs_struct() to never fail.
>>> 
>>> I had explored ways to change kthread_create() to avoid the need for
>>> GFP_NOFAIL and concluded that we can do everything we need in the sunrpc
>>> layer.  So the first patch here is a simple cleanup, and the second adds
>>> simple infrastructure for an svc thread to confirm that it has started
>>> up and to report if it was successful in that.
>>> 
>>> Thanks,
>>> NeilBrown
>>> 
>>> 
>>> 
>>> [PATCH 1/2] sunrpc: merge svc_rqst_alloc() into svc_prepare_thread()
>>> [PATCH 2/2] sunrpc: allow svc threads to fail initialisation cleanly
>> 
>> This series does not apply to nfsd-next. It looks like 1/2 expects
>> to see the EXPORT_SYMBOL after svc_rqst_alloc() that you already
>> removed in "SUNRPC: make various functions static, or not exported."
>> 
>> Also, 1/2 is From: your brown.name account, but the SoB is your
>> SuSE email. (Maybe that doesn't matter).
> 
> Probably don't matter.  That happened because I wrote that patch on my
> notebook instead of my desktop and they have different defaults.  I'll
> try to remember that for next time.

Some patch linting tools like the From: to match the SoB tag.
But again, I'm not sure that matters except to the legally
pedantic.


>> Can you rebase and resend?
> 
> I can't see "SUNRPC: make various functions static, or not exported." in
>    git://git.kernel.org/pub/scm/linux/kernel/git/cel/linux
> Am I looking in the wrong place? If not, can you push out nfsd-next?

For some reason, git keeps pushing my local nfsd-next branch
to cel's "master" branch. I've pushed again to cel's nfsd-next,
and it seems to be showing up now on the web UI as I intended.


> (and if you could delete "for-next" from there, and possibly other old
> cruft, that might help too)

I'm looking at the list of branches, and I don't see a for-next
in the cel git repo. The only branches I can see are:

nfsd-next
nfsd-6.1.y
nfsd-testing
master
nfsd-5.10.y
nfsd-5.15.y
nfsd-fixes
exportfs-next
topic-xdr-tracepoints
topic-rdma-fault-injection


I've been deleting old tags as I notice them, but "git push :branch"
doesn't always seem to work against git.kernel.org.


> Thanks,
> NeilBrown
> 
> 
>> 
>> In 2/2, what is the reason to make svc_thread_init_status() a static
>> inline rather than an exported function? I don't think this is going
>> to be a performance hot path, but maybe it becomes one in a future
>> patch?
>> 
>> 
>> -- 
>> Chuck Lever


--
Chuck Lever

Re: [PATCH 0/2] nfsd: fix handling of error from unshare_fs_struct()

[Christoph Hellwig]

On Tue, Jul 30, 2024 at 07:07:29AM +1000, NeilBrown wrote:
> > Also, 1/2 is From: your brown.name account, but the SoB is your
> > SuSE email. (Maybe that doesn't matter).
> 
> Probably don't matter.  That happened because I wrote that patch on my
> notebook instead of my desktop and they have different defaults.  I'll
> try to remember that for next time.

The signoff is supposed to match the author address.  If you send it
out using a different email account you need to add a separate From:
line to the mail body.  git-send-email will do that automatically for
you.

Re: [PATCH 0/2] nfsd: fix handling of error from unshare_fs_struct()

[NeilBrown]

On Tue, 30 Jul 2024, Chuck Lever III wrote:
> 
> 
> > On Jul 29, 2024, at 5:07 PM, NeilBrown <neilb@suse.de> wrote:
> > 
> 
> > (and if you could delete "for-next" from there, and possibly other old
> > cruft, that might help too)
> 
> I'm looking at the list of branches, and I don't see a for-next
> in the cel git repo. The only branches I can see are:
> 
> nfsd-next
> nfsd-6.1.y
> nfsd-testing
> master
> nfsd-5.10.y
> nfsd-5.15.y
> nfsd-fixes
> exportfs-next
> topic-xdr-tracepoints
> topic-rdma-fault-injection
> 
> 
> I've been deleting old tags as I notice them, but "git push :branch"
> doesn't always seem to work against git.kernel.org.


Ahh...  "git fetch" does discard branched that have been deleted
remotely.
I need
   git remote prune $remote
or to set remote.$remote.prune to true

Thanks,
NeilBrown

Re: [PATCH 0/2] nfsd: fix handling of error from unshare_fs_struct()

[NeilBrown]

On Tue, 30 Jul 2024, Chuck Lever wrote:
> 
> In 2/2, what is the reason to make svc_thread_init_status() a static
> inline rather than an exported function? I don't think this is going
> to be a performance hot path, but maybe it becomes one in a future
> patch?
> 

Sorry - I missed this question the first time.

My reasoning was that it was a tiny function which would be optimised
even smaller in two out of three call sites where the err number is a
constant zero.  Also my original draft didn't even have this as a
function but was open coded.

I don't think it matters much either way.

Thanks,
NeilBrown