[Linux-cachefs] 3.0.3 64-bit Crash running fscache/cachefilesd

[David Howells]

Mark Moseley <moseleymark at gmail.com> wrote:

> I thought I'd be extra-helpful by getting that trace with a 3.0.4
> kernel but got a completely different error this time (there was
> nothing logged above this though). There was a
> '__fscache_read_or_alloc_pages' crash for the previous boot too,
> though it went for about 2.5 hours that time (with an empty cache
> partition though).

I'm fairly certain I know what the cause of this one is: Invalidation upon
server change isn't handled correctly.  NFS tries to invalidate a file by
discarding that file's attachment to the cache - without first clearing up the
operations it has outstanding on the cache for that file.

I'm working on adding formal invalidation at the moment.

The attached patch may get you more precise information.  The first hunk is the
main catcher.

David
---
diff --git a/fs/fscache/cookie.c b/fs/fscache/cookie.c
index 9905350..48c63b8 100644
--- a/fs/fscache/cookie.c
+++ b/fs/fscache/cookie.c
@@ -452,6 +452,13 @@ void __fscache_relinquish_cookie(struct fscache_cookie *cookie, int retire)
 
 		_debug("RELEASE OBJ%x", object->debug_id);
 
+		if (atomic_read(&object->n_reads)) {
+			spin_unlock(&cookie->lock);
+			printk(KERN_ERR "FS-Cache: Cookie '%s' still has outstanding reads\n",
+			       cookie->def->name);
+			BUG();
+		}
+
 		/* detach each cache object from the object cookie */
 		spin_lock(&object->lock);
 		hlist_del_init(&object->cookie_link);
diff --git a/fs/fscache/page.c b/fs/fscache/page.c
index b8b62f4..f087051 100644
--- a/fs/fscache/page.c
+++ b/fs/fscache/page.c
@@ -496,6 +496,7 @@ int __fscache_read_or_alloc_pages(struct fscache_cookie *cookie,
 	if (fscache_submit_op(object, &op->op) < 0)
 		goto nobufs_unlock;
 	spin_unlock(&cookie->lock);
+	ASSERTCMP(object->cookie, ==, cookie);
 
 	fscache_stat(&fscache_n_retrieval_ops);
 
@@ -513,6 +514,26 @@ int __fscache_read_or_alloc_pages(struct fscache_cookie *cookie,
 		goto error;
 
 	/* ask the cache to honour the operation */
+	if (!object->cookie) {
+		const char prefix[] = "fs-";
+		printk(KERN_ERR "%sobject: OBJ%x\n",
+		       prefix, object->debug_id);
+		printk(KERN_ERR "%sobjstate=%s fl=%lx wbusy=%x ev=%lx[%lx]\n",
+		       prefix, fscache_object_states[object->state],
+		       object->flags, work_busy(&object->work),
+		       object->events,
+		       object->event_mask & FSCACHE_OBJECT_EVENTS_MASK);
+		printk(KERN_ERR "%sops=%u inp=%u exc=%u\n",
+		       prefix, object->n_ops, object->n_in_progress,
+		       object->n_exclusive);
+		printk(KERN_ERR "%sparent=%p\n",
+		       prefix, object->parent);
+		printk(KERN_ERR "%scookie=%p [pr=%p nd=%p fl=%lx]\n",
+		       prefix, object->cookie,
+		       cookie->parent, cookie->netfs_data, cookie->flags);
+	}
+	ASSERTCMP(object->cookie, ==, cookie);
+
 	if (test_bit(FSCACHE_COOKIE_NO_DATA_YET, &object->cookie->flags)) {
 		fscache_stat(&fscache_n_cop_allocate_pages);
 		ret = object->cache->ops->allocate_pages(