Skip to contentSkip to conversation

SUNRPC/cache: Allow garbage collection of invalid cache entries

PATCH In review Trond Myklebust 5 participants 11 messages Jan 14, 2020 – Mar 27, 2020

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 SUNRPC/cache: Allow garbage collection of invalid cache entries

If the cache entry never gets initialised, we want the garbage collector to be able to evict it. Otherwise if the upcall daemon fails to initialise the entry, we end up never expiring it.

Fixes d6fc8821c2d2 ("SUNRPC/Cache: Always treat the invalid cache as unexpired")
---
 include/linux/sunrpc/cache.h |  3 ---
 net/sunrpc/cache.c           | 36 +++++++++++++++++++-----------------
 2 files changed, 19 insertions(+), 20 deletions(-)
include/linux/sunrpc/cache.h +0 -3
index f8603724fbee..0428dd23fd79 100644
@@ -202,9 +202,6 @@ static inline void cache_put(struct cache_head *h, struct cache_detail *cd)
 
 static inline bool cache_is_expired(struct cache_detail *detail, struct cache_head *h)
 {
-	if (!test_bit(CACHE_VALID, &h->flags))
-		return false;
-
 	return  (h->expiry_time < seconds_since_boot()) ||
 		(detail->flush_time >= h->last_refresh);
 }
net/sunrpc/cache.c +19 -17
index 52d927210d32..99d630150af6 100644
@@ -65,13 +65,14 @@ static struct cache_head *sunrpc_cache_find_rcu(struct cache_detail *detail,
 
 	rcu_read_lock();
 	hlist_for_each_entry_rcu(tmp, head, cache_list) {
-		if (detail->match(tmp, key)) {
-			if (cache_is_expired(detail, tmp))
-				continue;
-			tmp = cache_get_rcu(tmp);
-			rcu_read_unlock();
-			return tmp;
-		}
+		if (!detail->match(tmp, key))
+			continue;
+		if (test_bit(CACHE_VALID, &tmp->flags) &&
+		    cache_is_expired(detail, tmp))
+			continue;
+		tmp = cache_get_rcu(tmp);
+		rcu_read_unlock();
+		return tmp;
 	}
 	rcu_read_unlock();
 	return NULL;
@@ -114,17 +115,18 @@ static struct cache_head *sunrpc_cache_add_entry(struct cache_detail *detail,
 
 	/* check if entry appeared while we slept */
 	hlist_for_each_entry_rcu(tmp, head, cache_list) {
-		if (detail->match(tmp, key)) {
-			if (cache_is_expired(detail, tmp)) {
-				sunrpc_begin_cache_remove_entry(tmp, detail);
-				freeme = tmp;
-				break;
-			}
-			cache_get(tmp);
-			spin_unlock(&detail->hash_lock);
-			cache_put(new, detail);
-			return tmp;
+		if (!detail->match(tmp, key))
+			continue;
+		if (test_bit(CACHE_VALID, &tmp->flags) &&
+		    cache_is_expired(detail, tmp)) {
+			sunrpc_begin_cache_remove_entry(tmp, detail);
+			freeme = tmp;
+			break;
 		}
+		cache_get(tmp);
+		spin_unlock(&detail->hash_lock);
+		cache_put(new, detail);
+		return tmp;
 	}
 
 	hlist_add_head_rcu(&new->cache_list, head);
Signature · 1 line
2.24.1
#1 lore raw
Quoted 3 lines · Trond Myklebust

If the cache entry never gets initialised, we want the garbage collector to be able to evict it. Otherwise if the upcall daemon fails to initialise the entry, we end up never expiring it.

Could you tell us more about what motivated this?

It's causing failures on pynfs server-reboot tests. I haven't pinned down the cause yet, but it looks like it could be a regression to the behavior Kinglong Mee describes in detail in his original patch.

Dropping for the time being.

--b.

Quoted 79 lines
Fixes d6fc8821c2d2 ("SUNRPC/Cache: Always treat the invalid cache as unexpired")
---
 include/linux/sunrpc/cache.h |  3 ---
 net/sunrpc/cache.c           | 36 +++++++++++++++++++-----------------
 2 files changed, 19 insertions(+), 20 deletions(-)
include/linux/sunrpc/cache.h +0 -3
index f8603724fbee..0428dd23fd79 100644
@@ -202,9 +202,6 @@ static inline void cache_put(struct cache_head *h, struct cache_detail *cd)
 
 static inline bool cache_is_expired(struct cache_detail *detail, struct cache_head *h)
 {
-	if (!test_bit(CACHE_VALID, &h->flags))
-		return false;
-
 	return  (h->expiry_time < seconds_since_boot()) ||
 		(detail->flush_time >= h->last_refresh);
 }
net/sunrpc/cache.c +19 -17
index 52d927210d32..99d630150af6 100644
@@ -65,13 +65,14 @@ static struct cache_head *sunrpc_cache_find_rcu(struct cache_detail *detail,
 
 	rcu_read_lock();
 	hlist_for_each_entry_rcu(tmp, head, cache_list) {
-		if (detail->match(tmp, key)) {
-			if (cache_is_expired(detail, tmp))
-				continue;
-			tmp = cache_get_rcu(tmp);
-			rcu_read_unlock();
-			return tmp;
-		}
+		if (!detail->match(tmp, key))
+			continue;
+		if (test_bit(CACHE_VALID, &tmp->flags) &&
+		    cache_is_expired(detail, tmp))
+			continue;
+		tmp = cache_get_rcu(tmp);
+		rcu_read_unlock();
+		return tmp;
 	}
 	rcu_read_unlock();
 	return NULL;
@@ -114,17 +115,18 @@ static struct cache_head *sunrpc_cache_add_entry(struct cache_detail *detail,
 
 	/* check if entry appeared while we slept */
 	hlist_for_each_entry_rcu(tmp, head, cache_list) {
-		if (detail->match(tmp, key)) {
-			if (cache_is_expired(detail, tmp)) {
-				sunrpc_begin_cache_remove_entry(tmp, detail);
-				freeme = tmp;
-				break;
-			}
-			cache_get(tmp);
-			spin_unlock(&detail->hash_lock);
-			cache_put(new, detail);
-			return tmp;
+		if (!detail->match(tmp, key))
+			continue;
+		if (test_bit(CACHE_VALID, &tmp->flags) &&
+		    cache_is_expired(detail, tmp)) {
+			sunrpc_begin_cache_remove_entry(tmp, detail);
+			freeme = tmp;
+			break;
 		}
+		cache_get(tmp);
+		spin_unlock(&detail->hash_lock);
+		cache_put(new, detail);
+		return tmp;
 	}
 
 	hlist_add_head_rcu(&new->cache_list, head);
Signature · 1 line
2.24.1
#2 lore raw
Quoted 11 lines · J. Bruce Fields
Quoted 3 lines · Trond Myklebust

If the cache entry never gets initialised, we want the garbage collector to be able to evict it. Otherwise if the upcall daemon fails to initialise the entry, we end up never expiring it.

Could you tell us more about what motivated this?

It's causing failures on pynfs server-reboot tests. I haven't pinned down the cause yet, but it looks like it could be a regression to the behavior Kinglong Mee describes in detail in his original patch.

Can you point me to the tests that are failing?

The motivation here is to allow the garbage collector to do its job of evicting cache entries after they are supposed to have timed out. The fact that uninitialised cache entries are given an infinite lifetime, and are never evicted is a de facto memory leak if, for instance, the mountd daemon ignores the cache request, or the downcall in expkey_parse() or svc_export_parse() fails without being able to update the request.

The threads that are waiting for the cache replies already have a mechanism for dealing with timeouts (with cache_wait_req() and deferred requests), so the question is what is so special about uninitialised requests that we have to leak them in order to avoid a problem with reboot?

Quoted 90 lines

Dropping for the time being.

--b.

Quoted 86 lines
Fixes d6fc8821c2d2 ("SUNRPC/Cache: Always treat the invalid cache

as unexpired")

---
 include/linux/sunrpc/cache.h |  3 ---
 net/sunrpc/cache.c           | 36 +++++++++++++++++++-------------
----
 2 files changed, 19 insertions(+), 20 deletions(-)
include/linux/sunrpc/cache.h +0 -3
index f8603724fbee..0428dd23fd79 100644
@@ -202,9 +202,6 @@ static inline void cache_put(struct cache_head
 *h, struct cache_detail *cd)
 
 static inline bool cache_is_expired(struct cache_detail *detail,
 struct cache_head *h)
 {
-	if (!test_bit(CACHE_VALID, &h->flags))
-		return false;
-
 	return  (h->expiry_time < seconds_since_boot()) ||
include/linux/sunrpc/cache.h +0 -0
 		(detail->flush_time >= h->last_refresh);
 }
net/sunrpc/cache.c +8 -7
index 52d927210d32..99d630150af6 100644
@@ -65,13 +65,14 @@ static struct cache_head
 *sunrpc_cache_find_rcu(struct cache_detail *detail,
 
 	rcu_read_lock();
 	hlist_for_each_entry_rcu(tmp, head, cache_list) {
-		if (detail->match(tmp, key)) {
-			if (cache_is_expired(detail, tmp))
-				continue;
-			tmp = cache_get_rcu(tmp);
-			rcu_read_unlock();
-			return tmp;
-		}
+		if (!detail->match(tmp, key))
+			continue;
+		if (test_bit(CACHE_VALID, &tmp->flags) &&
+		    cache_is_expired(detail, tmp))
+			continue;
+		tmp = cache_get_rcu(tmp);
+		rcu_read_unlock();
+		return tmp;
 	}
 	rcu_read_unlock();
net/sunrpc/cache.c +11 -10
 	return NULL;
@@ -114,17 +115,18 @@ static struct cache_head
 *sunrpc_cache_add_entry(struct cache_detail *detail,
 
 	/* check if entry appeared while we slept */
 	hlist_for_each_entry_rcu(tmp, head, cache_list) {
-		if (detail->match(tmp, key)) {
-			if (cache_is_expired(detail, tmp)) {
-				sunrpc_begin_cache_remove_entry(tmp,
 detail);
-				freeme = tmp;
-				break;
-			}
-			cache_get(tmp);
-			spin_unlock(&detail->hash_lock);
-			cache_put(new, detail);
-			return tmp;
+		if (!detail->match(tmp, key))
+			continue;
+		if (test_bit(CACHE_VALID, &tmp->flags) &&
+		    cache_is_expired(detail, tmp)) {
+			sunrpc_begin_cache_remove_entry(tmp, detail);
+			freeme = tmp;
+			break;
 		}
+		cache_get(tmp);
+		spin_unlock(&detail->hash_lock);
+		cache_put(new, detail);
+		return tmp;
 	}

hlist_add_head_rcu(&new->cache_list, head);

Signature · 1 line
2.24.1
Signature · 3 lines
Trond Myklebust
Linux NFS client maintainer, Hammerspace
[email protected]
#3 lore raw
Quoted 6 lines · Trond Myklebust
Quoted 5 lines · J. Bruce Fields
Quoted 3 lines · Trond Myklebust

If the cache entry never gets initialised, we want the garbage collector to be able to evict it. Otherwise if the upcall daemon fails to initialise the entry, we end up never expiring it.

Quoted 8 lines
Quoted 6 lines

Could you tell us more about what motivated this?

It's causing failures on pynfs server-reboot tests. I haven't pinned down the cause yet, but it looks like it could be a regression to the behavior Kinglong Mee describes in detail in his original patch.

Can you point me to the tests that are failing?

I'm basically doing

./nfs4.1/testserver.py myserver:/path reboot
--serverhelper=examples/server_helper.sh --serverhelperarg=myserver

For all I know at this point, the change could be exposing a pynfs-side bug.

Quoted 2 lines

The motivation here is to allow the garbage collector to do its job of evicting cache entries after they are supposed to have timed out.

Understood. I was curious whether this was found by code inspection or because you'd run across a case where the leak was causing a practical problem.

--b.

Quoted 11 lines

The fact that uninitialised cache entries are given an infinite lifetime, and are never evicted is a de facto memory leak if, for instance, the mountd daemon ignores the cache request, or the downcall in expkey_parse() or svc_export_parse() fails without being able to update the request.

The threads that are waiting for the cache replies already have a mechanism for dealing with timeouts (with cache_wait_req() and deferred requests), so the question is what is so special about uninitialised requests that we have to leak them in order to avoid a problem with reboot?

#4 lore raw
Quoted 33 lines · On Fri, 2020-02-07 at 13:18 -0500, [email protected] wrote:
Quoted 9 lines · Trond Myklebust
Quoted 8 lines · J. Bruce Fields
Quoted 4 lines · Trond Myklebust

If the cache entry never gets initialised, we want the garbage collector to be able to evict it. Otherwise if the upcall daemon
fails to initialise the entry, we end up never expiring it.

Could you tell us more about what motivated this?

Quoted 8 lines
Quoted 6 lines

It's causing failures on pynfs server-reboot tests. I haven't pinned
down the cause yet, but it looks like it could be a regression to the
behavior Kinglong Mee describes in detail in his original patch.

Can you point me to the tests that are failing?

I'm basically doing

./nfs4.1/testserver.py myserver:/path reboot
--serverhelper=examples/server_helper.sh --serverhelperarg=myserver

For all I know at this point, the change could be exposing a pynfs- side
bug.

Quoted 3 lines

The motivation here is to allow the garbage collector to do its job of
evicting cache entries after they are supposed to have timed out.

Quoted 5 lines

Understood. I was curious whether this was found by code inspection or
because you'd run across a case where the leak was causing a practical
problem.

I was seeing hangs in some cases where the path lookup was temporarily failing in the export path downcall. In that case, the mount daemon just ignores the returned error, and leaves it up to the kernel to fix what is happening.

The problem is that since the garbage collector is currently told to ignore stuff if CACHE_VALID is not set, and since cache_check() won't do a second upcall, then any RPC call trying to access data on the path that was affected just hangs and eventually times out. Nothing prompts for this condition to be fixed.

Quoted 16 lines

--b.

Quoted 13 lines

The fact that uninitialised cache entries are given an infinite lifetime, and are never evicted is a de facto memory leak if, for instance, the mountd daemon ignores the cache request, or the downcall
in expkey_parse() or svc_export_parse() fails without being able to update the request.

The threads that are waiting for the cache replies already have a mechanism for dealing with timeouts (with cache_wait_req() and deferred requests), so the question is what is so special about uninitialised requests that we have to leak them in order to avoid a
problem with reboot?

Trond Myklebust
CTO, Hammerspace Inc
4300 El Camino Real, Suite 105
Los Altos, CA 94022
www.hammer.space

Signature · 3 lines
Trond Myklebust
Linux NFS client maintainer, Hammerspace
[email protected]
#5 lore raw
In reply to [email protected]

Sorry, just getting back to this:

Quoted 16 lines · On Fri, Feb 07, 2020 at 01:18:17PM -0500, [email protected] wrote:
Quoted 6 lines · Trond Myklebust
Quoted 5 lines · J. Bruce Fields
Quoted 3 lines · Trond Myklebust

If the cache entry never gets initialised, we want the garbage collector to be able to evict it. Otherwise if the upcall daemon fails to initialise the entry, we end up never expiring it.

Quoted 8 lines
Quoted 6 lines

Could you tell us more about what motivated this?

It's causing failures on pynfs server-reboot tests. I haven't pinned down the cause yet, but it looks like it could be a regression to the behavior Kinglong Mee describes in detail in his original patch.

Can you point me to the tests that are failing?

Quoted 8 lines

I'm basically doing

./nfs4.1/testserver.py myserver:/path reboot
--serverhelper=examples/server_helper.sh --serverhelperarg=myserver

For all I know at this point, the change could be exposing a pynfs-side bug.

From a trace, it's clear that the server is actually becoming unresponsive, so it's not a pynfs bug.

Quoted 6 lines
Quoted 2 lines

The motivation here is to allow the garbage collector to do its job of evicting cache entries after they are supposed to have timed out.

Understood. I was curious whether this was found by code inspection or because you'd run across a case where the leak was causing a practical problem.

I'm still curious.

Quoted 5 lines
Quoted 5 lines

The fact that uninitialised cache entries are given an infinite lifetime, and are never evicted is a de facto memory leak if, for instance, the mountd daemon ignores the cache request, or the downcall in expkey_parse() or svc_export_parse() fails without being able to update the request.

If mountd ignores cache requests, or downcalls fail, then the server's broken anyway. The server can't do anything without mountd.

Quoted 5 lines
Quoted 5 lines

The threads that are waiting for the cache replies already have a mechanism for dealing with timeouts (with cache_wait_req() and deferred requests), so the question is what is so special about uninitialised requests that we have to leak them in order to avoid a problem with reboot?

I'm not sure I have this right yet. I'm just staring at the code and at Kinglong Mee's description on d6fc8821c2d2. I think the way it works is that a cash flush from mountd results in all cache entries (including invalid entries that nfsd threads are waiting on) being considered expired. So cache_check() returns an immediate ETIMEDOUT without waiting.

Maybe the cache_is_expired() logic should be something more like:

	if (h->expiry_time < seconds_since_boot())
		return true;
	if (!test_bit(CACHE_VALID, &h->flags))
		return false;
	return h->expiry_time < seconds_since_boot();

So invalid cache entries (which are waiting for a reply from mountd) can expire, but they can't be flushed. If that makes sense.

As a stopgap we may want to revert or drop the "Allow garbage collection" patch, as the (preexisting) memory leak seems lower impact than the server hang.

--b.

#6 lore raw

Hi Bruce,

Quoted 86 lines · On Thu, 2020-03-26 at 16:40 -0400, [email protected] wrote:

Sorry, just getting back to this:

Quoted 24 lines · On Fri, Feb 07, 2020 at 01:18:17PM -0500, [email protected] wrote:
Quoted 19 lines · Trond Myklebust
Quoted 8 lines · J. Bruce Fields
Quoted 5 lines · Trond Myklebust

If the cache entry never gets initialised, we want the garbage
collector to be able to evict it. Otherwise if the upcall daemon
fails to initialise the entry, we end up never expiring it.

Quoted 9 lines

Could you tell us more about what motivated this?

It's causing failures on pynfs server-reboot tests. I haven't pinned
down the cause yet, but it looks like it could be a regression to the
behavior Kinglong Mee describes in detail in his original patch.

Quoted 1 line

Can you point me to the tests that are failing?

I'm basically doing

Quoted 7 lines

./nfs4.1/testserver.py myserver:/path reboot
--serverhelper=examples/server_helper.sh --serverhelperarg=myserver

For all I know at this point, the change could be exposing a pynfs- side
bug.

From a trace, it's clear that the server is actually becoming unresponsive, so it's not a pynfs bug.

Quoted 9 lines
Quoted 3 lines

The motivation here is to allow the garbage collector to do its job of
evicting cache entries after they are supposed to have timed out.

Understood. I was curious whether this was found by code inspection or
because you'd run across a case where the leak was causing a practical
problem.

I'm still curious.

Quoted 7 lines
Quoted 7 lines

The fact that uninitialised cache entries are given an infinite lifetime, and are never evicted is a de facto memory leak if, for instance, the mountd daemon ignores the cache request, or the downcall
in expkey_parse() or svc_export_parse() fails without being able to
update the request.

If mountd ignores cache requests, or downcalls fail, then the server's
broken anyway. The server can't do anything without mountd.

Quoted 6 lines
Quoted 6 lines

The threads that are waiting for the cache replies already have a mechanism for dealing with timeouts (with cache_wait_req() and deferred requests), so the question is what is so special about uninitialised requests that we have to leak them in order to avoid a
problem with reboot?

I'm not sure I have this right yet. I'm just staring at the code and at
Kinglong Mee's description on d6fc8821c2d2. I think the way it works is
that a cash flush from mountd results in all cache entries (including invalid entries that nfsd threads are waiting on) being considered expired. So cache_check() returns an immediate ETIMEDOUT without waiting.

Maybe the cache_is_expired() logic should be something more like:

	if (h->expiry_time < seconds_since_boot())
		return true;
	if (!test_bit(CACHE_VALID, &h->flags))
		return false;
	return h->expiry_time < seconds_since_boot();
Quoted 9 lines

So invalid cache entries (which are waiting for a reply from mountd) can
expire, but they can't be flushed. If that makes sense.

As a stopgap we may want to revert or drop the "Allow garbage collection" patch, as the (preexisting) memory leak seems lower impact
than the server hang.

I believe you were probably seeing the effect of the cache_listeners_exist() test, which is just wrong for all cache upcall users except idmapper and svcauth_gss. We should not be creating negative cache entries just because the rpc.mountd daemon happens to be slow to connect to the upcall pipes when starting up, or because it crashes and fails to restart correctly.

That's why, when I resubmitted this patch, I included
https://git.linux-nfs.org/?p=cel/cel-2.6.git;a=commitdiff;h=b840228cd6096bebe16b3e4eb5d93597d0e02c6d

which turns off that particular test for all the upcalls to rpc.mountd.

Note that the patch series I sent includes a bunch of kernel tracepoints that we can enable to help debugging if we see anything similar happening in the future.

Cheers
Trond

Signature · 3 lines
Trond Myklebust
Linux NFS client maintainer, Hammerspace
[email protected]
#7 lore raw
Quoted 25 lines · Trond Myklebust
Quoted 8 lines · On Thu, 2020-03-26 at 16:40 -0400, [email protected] wrote:

Maybe the cache_is_expired() logic should be something more like:

	if (h->expiry_time < seconds_since_boot())
		return true;
	if (!test_bit(CACHE_VALID, &h->flags))
		return false;
	return h->expiry_time < seconds_since_boot();
Quoted 8 lines

So invalid cache entries (which are waiting for a reply from mountd) can
expire, but they can't be flushed. If that makes sense.

As a stopgap we may want to revert or drop the "Allow garbage collection" patch, as the (preexisting) memory leak seems lower impact
than the server hang.

I believe you were probably seeing the effect of the cache_listeners_exist() test, which is just wrong for all cache upcall users except idmapper and svcauth_gss. We should not be creating negative cache entries just because the rpc.mountd daemon happens to be slow to connect to the upcall pipes when starting up, or because it crashes and fails to restart correctly.

Quoted 4 lines

That's why, when I resubmitted this patch, I included
https://git.linux-nfs.org/?p=cel/cel-2.6.git;a=commitdiff;h=b840228cd6096bebe16b3e4eb5d93597d0e02c6d

which turns off that particular test for all the upcalls to rpc.mountd.

The hangs persist with that patch, but go away with the change to the cache_is_expired() logic above.

--b.

#8 lore raw
Quoted 9 lines · J. Bruce Fields
Quoted 8 lines · Trond Myklebust
Quoted 7 lines · On Thu, 2020-03-26 at 16:40 -0400, [email protected] wrote:

Maybe the cache_is_expired() logic should be something more like:

	if (h->expiry_time < seconds_since_boot())
		return true;
	if (!test_bit(CACHE_VALID, &h->flags))
		return false;
	return h->expiry_time < seconds_since_boot();

Did you mean

return detail->flush_time >= h->last_refresh;

instead of repeating the h->expiry_time check?

Quoted 20 lines
Quoted 20 lines
Quoted 10 lines

So invalid cache entries (which are waiting for a reply from mountd)
can
expire, but they can't be flushed. If that makes sense.

As a stopgap we may want to revert or drop the "Allow garbage collection" patch, as the (preexisting) memory leak seems lower impact
than the server hang.

I believe you were probably seeing the effect of the cache_listeners_exist() test, which is just wrong for all cache upcall
users except idmapper and svcauth_gss. We should not be creating negative cache entries just because the rpc.mountd daemon happens to be
slow to connect to the upcall pipes when starting up, or because it crashes and fails to restart correctly.

Quoted 8 lines
Quoted 5 lines

That's why, when I resubmitted this patch, I included
https://git.linux-nfs.org/?p=cel/cel-2.6.git;a=commitdiff;h=b840228cd6096bebe16b3e4eb5d93597d0e02c6d

which turns off that particular test for all the upcalls to rpc.mountd.

The hangs persist with that patch, but go away with the change to the cache_is_expired() logic above.

Fair enough. Do you want to send Chuck a fix?

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

PATCH 1/1 SUNRPC/cache: don't allow invalid entries to be flushed

From: "J. Bruce Fields" <[email protected]>

Trond points out in 277f27e2f277 that we allow invalid cache entries to persist indefinitely. That fix, however, reintroduces the problem fixed by Kinglong Mee's d6fc8821c2d2 "SUNRPC/Cache: Always treat the invalid cache as unexpired", where an invalid cache entry is immediately removed by a flush before mountd responds to it. The result is that the server thread that should be waiting for mountd to fill in that entry instead gets an -ETIMEDOUT return from cache_check(). Symptoms are the server becoming unresponsive after a restart, reproduceable by running pynfs 4.1 test REBT5.

Instead, take a compromise approach: allow invalid cache entries to be removed after they expire, but not to be removed by a cache flush.

Fixes 277f27e2f277 "SUNRPC/cache: Allow garbage collection..."
---
 include/linux/sunrpc/cache.h | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)
include/linux/sunrpc/cache.h +5 -2
index 532cdbda43da..10891b70fc7b 100644
@@ -209,8 +209,11 @@ static inline void cache_put(struct cache_head *h, struct cache_detail *cd)
 
 static inline bool cache_is_expired(struct cache_detail *detail, struct cache_head *h)
 {
-	return  (h->expiry_time < seconds_since_boot()) ||
-		(detail->flush_time >= h->last_refresh);
+	if (h->expiry_time < seconds_since_boot())
+		return true;
+	if (!test_bit(CACHE_VALID, &h->flags))
+		return false;
+	return detail->flush_time >= h->last_refresh;
 }
 
 extern int cache_check(struct cache_detail *detail,
Signature · 1 line
2.25.1
#10 lore raw

Re: [PATCH] SUNRPC/cache: don't allow invalid entries to be flushed

Quoted 16 lines · J. Bruce Fields

From: "J. Bruce Fields" <[email protected]>

Trond points out in 277f27e2f277 that we allow invalid cache entries to persist indefinitely. That fix, however, reintroduces the problem fixed by Kinglong Mee's d6fc8821c2d2 "SUNRPC/Cache: Always treat the invalid cache as unexpired", where an invalid cache entry is immediately removed by a flush before mountd responds to it. The result is that the server thread that should be waiting for mountd to fill in that entry instead gets an -ETIMEDOUT return from cache_check(). Symptoms are the server becoming unresponsive after a restart, reproduceable by running pynfs 4.1 test REBT5.

Quoted 5 lines

Instead, take a compromise approach: allow invalid cache entries to be removed after they expire, but not to be removed by a cache flush.

Fixes 277f27e2f277 "SUNRPC/cache: Allow garbage collection..."

Thanks, Bruce. checkpatch.pl is complaining loudly about the style of the short commit descriptions. I'll fix those up before applying it to nfsd-5.7.

Quoted 25 lines

---
include/linux/sunrpc/cache.h | 7 +++++--

1 file changed, 5 insertions(+), 2 deletions(-)
include/linux/sunrpc/cache.h +5 -2
index 532cdbda43da..10891b70fc7b 100644
@@ -209,8 +209,11 @@ static inline void cache_put(struct cache_head *h, struct cache_detail *cd)
 
 static inline bool cache_is_expired(struct cache_detail *detail, struct cache_head *h)
 {
-	return  (h->expiry_time < seconds_since_boot()) ||
-		(detail->flush_time >= h->last_refresh);
+	if (h->expiry_time < seconds_since_boot())
+		return true;
+	if (!test_bit(CACHE_VALID, &h->flags))
+		return false;
+	return detail->flush_time >= h->last_refresh;
 }
 
 extern int cache_check(struct cache_detail *detail,
Signature · 1 line
2.25.1

--
Chuck Lever