Attachment 311138 Details for Bug 451317 – proposed patch -- don't allow rpc_malloc to block

[patch] proposed patch -- don't allow rpc_malloc to block

bz-451317-sunrpc-prevent-deadl (text/plain), 2.25 KB, created by Jeff Layton on 2008-07-07 11:24:21 UTC

(hide)

Description:

Filename:

MIME Type:

Creator: Jeff Layton

Created: 2008-07-07 11:24:21 UTC

Size: 2.25 KB

patch

obsolete

>BZ#451317: sunrpc: prevent deadlock by not allowing rpc_malloc to sleep
>
>From: Jeff Layton <jlayton@redhat.com>
>
>We don't want RPC's sleeping outside of the task scheduler while
>allocating their buffer. Change rpc_malloc from using GFP_NOFS
>to GFP_NOWAIT when allocating memory.
>    
>Also, change the printk in call_allocate() to a dprintk since it'll
>be somewhat more likely that it could fire.
>---
>
> net/sunrpc/clnt.c  |    2 +-
> net/sunrpc/sched.c |   14 ++++++++++----
> 2 files changed, 11 insertions(+), 5 deletions(-)
>
>
>diff --git a/net/sunrpc/clnt.c b/net/sunrpc/clnt.c
>index 2fbf037..63a7a20 100644
>--- a/net/sunrpc/clnt.c
>+++ b/net/sunrpc/clnt.c
>@@ -734,7 +734,7 @@ call_allocate(struct rpc_task *task)
> 
> 	if (xprt->ops->buf_alloc(task, bufsiz << 1) != NULL)
> 		return;
>-	printk(KERN_INFO "RPC: buffer allocation failed for task %p\n", task); 
>+	dprintk("RPC: buffer allocation failed for task %p\n", task); 
> 
> 	if (RPC_IS_ASYNC(task) || !signalled()) {
> 		xprt_release(task);
>diff --git a/net/sunrpc/sched.c b/net/sunrpc/sched.c
>index 7e870ed..68f7e36 100644
>--- a/net/sunrpc/sched.c
>+++ b/net/sunrpc/sched.c
>@@ -758,10 +758,16 @@ static void rpc_async_schedule(void *arg)
>  * @task: RPC task that will use this buffer
>  * @size: requested byte size
>  *
>- * We try to ensure that some NFS reads and writes can always proceed
>- * by using a mempool when allocating 'small' buffers.
>+ * To prevent rpciod from hanging, this allocator never sleeps,
>+ * returning NULL if the request cannot be serviced immediately.
>+ * The caller can arrange to sleep in a way that is safe for rpciod.
>+ *
>+ * Most requests are 'small' (under 2KiB) and can be serviced from a
>+ * mempool, ensuring that NFS reads and writes can always proceed,
>+ * and that there is good locality of reference for these buffers.
>+ *
>  * In order to avoid memory starvation triggering more writebacks of
>- * NFS requests, we use GFP_NOFS rather than GFP_KERNEL.
>+ * NFS requests, we avoid using GFP_KERNEL.
>  */
> void * rpc_malloc(struct rpc_task *task, size_t size)
> {
>@@ -771,7 +777,7 @@ void * rpc_malloc(struct rpc_task *task, size_t size)
> 	if (task->tk_flags & RPC_TASK_SWAPPER)
> 		gfp = GFP_ATOMIC;
> 	else
>-		gfp = GFP_NOFS;
>+		gfp = GFP_NOWAIT;
> 
> 	if (size > RPC_BUFFER_MAXSIZE) {
> 		req->rq_buffer = kmalloc(size, gfp);

Actions: View | Diff

Attachments on bug 451317: 311138