Merge tag '9p-for-6.1' of https://github.com/martinetd/linux

Pull 9p updates from Dominique Martinet:
 "Smaller buffers for small messages and fixes.

  The highlight of this is Christian's patch to allocate smaller buffers
  for most metadata requests: 9p with a big msize would try to allocate
  large buffers when just 4 or 8k would be more than enough; this brings
  in nice performance improvements.

  There's also a few fixes for problems reported by syzkaller (thanks to
  Schspa Shi, Tetsuo Handa for tests and feedback/patches) as well as
  some minor cleanup"

* tag '9p-for-6.1' of https://github.com/martinetd/linux:
  net/9p: clarify trans_fd parse_opt failure handling
  net/9p: add __init/__exit annotations to module init/exit funcs
  net/9p: use a dedicated spinlock for trans_fd
  9p/trans_fd: always use O_NONBLOCK read/write
  net/9p: allocate appropriate reduced message buffers
  net/9p: add 'pooled_rbuffers' flag to struct p9_trans_module
  net/9p: add p9_msg_buf_size()
  9p: add P9_ERRMAX for 9p2000 and 9p2000.u
  net/9p: split message size argument into 't_size' and 'r_size' pair
  9p: trans_fd/p9_conn_cancel: drop client lock earlier

1 files changed, 39 insertions, 9 deletions

diff --git a/net/9p/client.c b/net/9p/client.c
index 0a6110e15d0f..aaa37b07e30a 100644
--- a/net/9p/client.c
+++ b/net/9p/client.c
@@ -255,24 +255,42 @@ static struct kmem_cache *p9_req_cache;
  * p9_tag_alloc - Allocate a new request.
  * @c: Client session.
  * @type: Transaction type.
- * @max_size: Maximum packet size for this request.
+ * @t_size: Buffer size for holding this request
+ * (automatic calculation by format template if 0).
+ * @r_size: Buffer size for holding server's reply on this request
+ * (automatic calculation by format template if 0).
+ * @fmt: Format template for assembling 9p request message
+ * (see p9pdu_vwritef).
+ * @ap: Variable arguments to be fed to passed format template
+ * (see p9pdu_vwritef).
  *
  * Context: Process context.
  * Return: Pointer to new request.
  */
 static struct p9_req_t *
-p9_tag_alloc(struct p9_client *c, int8_t type, unsigned int max_size)
+p9_tag_alloc(struct p9_client *c, int8_t type, uint t_size, uint r_size,
+	      const char *fmt, va_list ap)
 {
 	struct p9_req_t *req = kmem_cache_alloc(p9_req_cache, GFP_NOFS);
-	int alloc_msize = min(c->msize, max_size);
+	int alloc_tsize;
+	int alloc_rsize;
 	int tag;
+	va_list apc;
+
+	va_copy(apc, ap);
+	alloc_tsize = min_t(size_t, c->msize,
+			    t_size ?: p9_msg_buf_size(c, type, fmt, apc));
+	va_end(apc);
+
+	alloc_rsize = min_t(size_t, c->msize,
+			    r_size ?: p9_msg_buf_size(c, type + 1, fmt, ap));
 
 	if (!req)
 		return ERR_PTR(-ENOMEM);
 
-	if (p9_fcall_init(c, &req->tc, alloc_msize))
+	if (p9_fcall_init(c, &req->tc, alloc_tsize))
 		goto free_req;
-	if (p9_fcall_init(c, &req->rc, alloc_msize))
+	if (p9_fcall_init(c, &req->rc, alloc_rsize))
 		goto free;
 
 	p9pdu_reset(&req->tc);
@@ -592,11 +610,12 @@ static int p9_client_flush(struct p9_client *c, struct p9_req_t *oldreq)
 }
 
 static struct p9_req_t *p9_client_prepare_req(struct p9_client *c,
-					      int8_t type, int req_size,
+					      int8_t type, uint t_size, uint r_size,
 					      const char *fmt, va_list ap)
 {
 	int err;
 	struct p9_req_t *req;
+	va_list apc;
 
 	p9_debug(P9_DEBUG_MUX, "client %p op %d\n", c, type);
 
@@ -608,7 +627,9 @@ static struct p9_req_t *p9_client_prepare_req(struct p9_client *c,
 	if (c->status == BeginDisconnect && type != P9_TCLUNK)
 		return ERR_PTR(-EIO);
 
-	req = p9_tag_alloc(c, type, req_size);
+	va_copy(apc, ap);
+	req = p9_tag_alloc(c, type, t_size, r_size, fmt, apc);
+	va_end(apc);
 	if (IS_ERR(req))
 		return req;
 
@@ -643,9 +664,18 @@ p9_client_rpc(struct p9_client *c, int8_t type, const char *fmt, ...)
 	int sigpending, err;
 	unsigned long flags;
 	struct p9_req_t *req;
+	/* Passing zero for tsize/rsize to p9_client_prepare_req() tells it to
+	 * auto determine an appropriate (small) request/response size
+	 * according to actual message data being sent. Currently RDMA
+	 * transport is excluded from this response message size optimization,
+	 * as it would not cope with it, due to its pooled response buffers
+	 * (using an optimized request size for RDMA as well though).
+	 */
+	const uint tsize = 0;
+	const uint rsize = c->trans_mod->pooled_rbuffers ? c->msize : 0;
 
 	va_start(ap, fmt);
-	req = p9_client_prepare_req(c, type, c->msize, fmt, ap);
+	req = p9_client_prepare_req(c, type, tsize, rsize, fmt, ap);
 	va_end(ap);
 	if (IS_ERR(req))
 		return req;
@@ -743,7 +773,7 @@ static struct p9_req_t *p9_client_zc_rpc(struct p9_client *c, int8_t type,
 	/* We allocate a inline protocol data of only 4k bytes.
 	 * The actual content is passed in zero-copy fashion.
 	 */
-	req = p9_client_prepare_req(c, type, P9_ZC_HDR_SZ, fmt, ap);
+	req = p9_client_prepare_req(c, type, P9_ZC_HDR_SZ, P9_ZC_HDR_SZ, fmt, ap);
 	va_end(ap);
 	if (IS_ERR(req))
 		return req;