netdev - [PATCH v5 11/11] net/9p: allocate appropriate reduced message buffers

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite for Android: free password hash cracker in your pocket

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Tue, 12 Jul 2022 16:31:36 +0200
From:   Christian Schoenebeck <linux_oss@...debyte.com>
To:     v9fs-developer@...ts.sourceforge.net
Cc:     linux-kernel@...r.kernel.org, netdev@...r.kernel.org,
        Dominique Martinet <asmadeus@...ewreck.org>,
        Eric Van Hensbergen <ericvh@...il.com>,
        Latchesar Ionkov <lucho@...kov.net>,
        Nikolay Kichukov <nikolay@...um.net>
Subject: [PATCH v5 11/11] net/9p: allocate appropriate reduced message buffers

So far 'msize' was simply used for all 9p message types, which is far
too much and slowed down performance tremendously with large values
for user configurable 'msize' option.

Let's stop this waste by using the new p9_msg_buf_size() function for
allocating more appropriate, smaller buffers according to what is
actually sent over the wire.

Only exception: RDMA transport is currently excluded from this, as
it would not cope with it. [1]

Link: https://lore.kernel.org/all/YkmVI6pqTuMD8dVi@codewreck.org/ [1]
Signed-off-by: Christian Schoenebeck <linux_oss@...debyte.com>
---

Is the !strcmp(c->trans_mod->name, "rdma") check in this patch maybe a bit
too hack-ish? Should there rather be transport API extension instead?

 net/9p/client.c | 39 ++++++++++++++++++++++++++++++++-------
 1 file changed, 32 insertions(+), 7 deletions(-)

diff --git a/net/9p/client.c b/net/9p/client.c
index 56be1658870d..dc1a7b26fab4 100644
--- a/net/9p/client.c
+++ b/net/9p/client.c
@@ -255,19 +255,35 @@ static struct kmem_cache *p9_req_cache;
  * p9_tag_alloc - Allocate a new request.
  * @c: Client session.
  * @type: Transaction type.
- * @t_size: Buffer size for holding this request.
- * @r_size: Buffer size for holding server's reply on 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, uint t_size, uint r_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_tsize = min(c->msize, t_size);
-	int alloc_rsize = min(c->msize, r_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);
@@ -685,6 +701,7 @@ static struct p9_req_t *p9_client_prepare_req(struct p9_client *c,
 {
 	int err;
 	struct p9_req_t *req;
+	va_list apc;
 
 	p9_debug(P9_DEBUG_MUX, "client %p op %d\n", c, type);
 
@@ -696,7 +713,9 @@ 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, t_size, r_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;
 
@@ -731,9 +750,15 @@ 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 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 message
+	 * size optimization, as it would not be able to cope with it.
+	 */
+	const uint max_size = !strcmp(c->trans_mod->name, "rdma") ? c->msize : 0;
 
 	va_start(ap, fmt);
-	req = p9_client_prepare_req(c, type, c->msize, c->msize, fmt, ap);
+	req = p9_client_prepare_req(c, type, max_size, max_size, fmt, ap);
 	va_end(ap);
 	if (IS_ERR(req))
 		return req;
-- 
2.30.2