summaryrefslogtreecommitdiff
path: root/net
diff options
context:
space:
mode:
authorChuck Lever <chuck.lever@oracle.com>2007-03-30 00:47:58 +0400
committerTrond Myklebust <Trond.Myklebust@netapp.com>2007-05-01 09:17:11 +0400
commitc5a4dd8b7c15927a8fbff83171b57cad675a79b9 (patch)
tree2d3b1930449b31f69dc70a6e1d4e0f0532f3f118 /net
parent2bea90d43a050bbc4021d44e59beb34f384438db (diff)
downloadlinux-c5a4dd8b7c15927a8fbff83171b57cad675a79b9.tar.xz
SUNRPC: Eliminate side effects from rpc_malloc
Currently rpc_malloc sets req->rq_buffer internally. Make this a more generic interface: return a pointer to the new buffer (or NULL) and make the caller set req->rq_buffer and req->rq_bufsize. This looks much more like kmalloc and eliminates the side effects. To fix a potential deadlock, this patch also replaces GFP_NOFS with GFP_NOWAIT in rpc_malloc. This prevents async RPCs from sleeping outside the RPC's task scheduler while allocating their buffer. Signed-off-by: Chuck Lever <chuck.lever@oracle.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Diffstat (limited to 'net')
-rw-r--r--net/sunrpc/clnt.c3
-rw-r--r--net/sunrpc/sched.c65
-rw-r--r--net/sunrpc/xprt.c2
3 files changed, 37 insertions, 33 deletions
diff --git a/net/sunrpc/clnt.c b/net/sunrpc/clnt.c
index 12487aafaab5..e7dc09ecc470 100644
--- a/net/sunrpc/clnt.c
+++ b/net/sunrpc/clnt.c
@@ -774,7 +774,8 @@ call_allocate(struct rpc_task *task)
req->rq_rcvsize = RPC_REPHDRSIZE + slack + proc->p_replen;
req->rq_rcvsize <<= 2;
- xprt->ops->buf_alloc(task, req->rq_callsize + req->rq_rcvsize);
+ req->rq_buffer = xprt->ops->buf_alloc(task,
+ req->rq_callsize + req->rq_rcvsize);
if (req->rq_buffer != NULL)
return;
diff --git a/net/sunrpc/sched.c b/net/sunrpc/sched.c
index 6d87320074b1..4a53e94f8134 100644
--- a/net/sunrpc/sched.c
+++ b/net/sunrpc/sched.c
@@ -741,50 +741,53 @@ static void rpc_async_schedule(struct work_struct *work)
* @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)
+void *rpc_malloc(struct rpc_task *task, size_t size)
{
- struct rpc_rqst *req = task->tk_rqstp;
- gfp_t gfp;
+ size_t *buf;
+ gfp_t gfp = RPC_IS_SWAPPER(task) ? GFP_ATOMIC : GFP_NOWAIT;
- if (task->tk_flags & RPC_TASK_SWAPPER)
- gfp = GFP_ATOMIC;
+ size += sizeof(size_t);
+ if (size <= RPC_BUFFER_MAXSIZE)
+ buf = mempool_alloc(rpc_buffer_mempool, gfp);
else
- gfp = GFP_NOFS;
-
- if (size > RPC_BUFFER_MAXSIZE) {
- req->rq_buffer = kmalloc(size, gfp);
- if (req->rq_buffer)
- req->rq_bufsize = size;
- } else {
- req->rq_buffer = mempool_alloc(rpc_buffer_mempool, gfp);
- if (req->rq_buffer)
- req->rq_bufsize = RPC_BUFFER_MAXSIZE;
- }
- return req->rq_buffer;
+ buf = kmalloc(size, gfp);
+ *buf = size;
+ dprintk("RPC: %5u allocated buffer of size %u at %p\n",
+ task->tk_pid, size, buf);
+ return (void *) ++buf;
}
/**
* rpc_free - free buffer allocated via rpc_malloc
- * @task: RPC task with a buffer to be freed
+ * @buffer: buffer to free
*
*/
-void rpc_free(struct rpc_task *task)
+void rpc_free(void *buffer)
{
- struct rpc_rqst *req = task->tk_rqstp;
+ size_t size, *buf = (size_t *) buffer;
- if (req->rq_buffer) {
- if (req->rq_bufsize == RPC_BUFFER_MAXSIZE)
- mempool_free(req->rq_buffer, rpc_buffer_mempool);
- else
- kfree(req->rq_buffer);
- req->rq_buffer = NULL;
- req->rq_bufsize = 0;
- }
+ if (!buffer)
+ return;
+ size = *buf;
+ buf--;
+
+ dprintk("RPC: freeing buffer of size %u at %p\n",
+ size, buf);
+ if (size <= RPC_BUFFER_MAXSIZE)
+ mempool_free(buf, rpc_buffer_mempool);
+ else
+ kfree(buf);
}
/*
diff --git a/net/sunrpc/xprt.c b/net/sunrpc/xprt.c
index 432ee92cf262..81fe830da8aa 100644
--- a/net/sunrpc/xprt.c
+++ b/net/sunrpc/xprt.c
@@ -854,7 +854,7 @@ void xprt_release(struct rpc_task *task)
mod_timer(&xprt->timer,
xprt->last_used + xprt->idle_timeout);
spin_unlock_bh(&xprt->transport_lock);
- xprt->ops->buf_free(task);
+ xprt->ops->buf_free(req->rq_buffer);
task->tk_rqstp = NULL;
if (req->rq_release_snd_buf)
req->rq_release_snd_buf(req);