linux-kernel - [PATCH 4.4 072/112] cifs: Limit the overall credit acquired

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: Windows password security audit tool. GUI, reports in PDF.

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

Message-Id: <20161026122307.834659388@linuxfoundation.org>
Date:   Wed, 26 Oct 2016 14:22:55 +0200
From:   Greg Kroah-Hartman <gregkh@...uxfoundation.org>
To:     linux-kernel@...r.kernel.org
Cc:     Greg Kroah-Hartman <gregkh@...uxfoundation.org>,
        stable@...r.kernel.org, Ross Lagerwall <ross.lagerwall@...rix.com>,
        Steve French <smfrench@...il.com>
Subject: [PATCH 4.4 072/112] cifs: Limit the overall credit acquired

4.4-stable review patch.  If anyone has any objections, please let me know.

------------------

From: Ross Lagerwall <ross.lagerwall@...rix.com>

commit 7d414f396c91a3382e51cf628c1cf0709ad0188b upstream.

The kernel client requests 2 credits for many operations even though
they only use 1 credit (presumably to build up a buffer of credit).
Some servers seem to give the client as much credit as is requested.  In
this case, the amount of credit the client has continues increasing to
the point where (server->credits * MAX_BUFFER_SIZE) overflows in
smb2_wait_mtu_credits().

Fix this by throttling the credit requests if an set limit is reached.
For async requests where the credit charge may be > 1, request as much
credit as what is charged.
The limit is chosen somewhat arbitrarily. The Windows client
defaults to 128 credits, the Windows server allows clients up to
512 credits (or 8192 for Windows 2016), and the NetApp server
(and at least one other) does not limit clients at all.
Choose a high enough value such that the client shouldn't limit
performance.

This behavior was seen with a NetApp filer (NetApp Release 9.0RC2).

Signed-off-by: Ross Lagerwall <ross.lagerwall@...rix.com>
Signed-off-by: Steve French <smfrench@...il.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@...uxfoundation.org>

---
 fs/cifs/smb2glob.h |   10 ++++++++++
 fs/cifs/smb2pdu.c  |   18 +++++++++++++++++-
 2 files changed, 27 insertions(+), 1 deletion(-)

--- a/fs/cifs/smb2glob.h
+++ b/fs/cifs/smb2glob.h
@@ -61,4 +61,14 @@
 /* Maximum buffer size value we can send with 1 credit */
 #define SMB2_MAX_BUFFER_SIZE 65536
 
+/*
+ * Maximum number of credits to keep available.
+ * This value is chosen somewhat arbitrarily. The Windows client
+ * defaults to 128 credits, the Windows server allows clients up to
+ * 512 credits, and the NetApp server does not limit clients at all.
+ * Choose a high enough value such that the client shouldn't limit
+ * performance.
+ */
+#define SMB2_MAX_CREDITS_AVAILABLE 32000
+
 #endif	/* _SMB2_GLOB_H */
--- a/fs/cifs/smb2pdu.c
+++ b/fs/cifs/smb2pdu.c
@@ -103,7 +103,21 @@ smb2_hdr_assemble(struct smb2_hdr *hdr,
 	hdr->ProtocolId[3] = 'B';
 	hdr->StructureSize = cpu_to_le16(64);
 	hdr->Command = smb2_cmd;
-	hdr->CreditRequest = cpu_to_le16(2); /* BB make this dynamic */
+	if (tcon && tcon->ses && tcon->ses->server) {
+		struct TCP_Server_Info *server = tcon->ses->server;
+
+		spin_lock(&server->req_lock);
+		/* Request up to 2 credits but don't go over the limit. */
+		if (server->credits >= SMB2_MAX_CREDITS_AVAILABLE)
+			hdr->CreditRequest = cpu_to_le16(0);
+		else
+			hdr->CreditRequest = cpu_to_le16(
+				min_t(int, SMB2_MAX_CREDITS_AVAILABLE -
+						server->credits, 2));
+		spin_unlock(&server->req_lock);
+	} else {
+		hdr->CreditRequest = cpu_to_le16(2);
+	}
 	hdr->ProcessId = cpu_to_le32((__u16)current->tgid);
 
 	if (!tcon)
@@ -2059,6 +2073,7 @@ smb2_async_readv(struct cifs_readdata *r
 	if (rdata->credits) {
 		buf->CreditCharge = cpu_to_le16(DIV_ROUND_UP(rdata->bytes,
 						SMB2_MAX_BUFFER_SIZE));
+		buf->CreditRequest = buf->CreditCharge;
 		spin_lock(&server->req_lock);
 		server->credits += rdata->credits -
 						le16_to_cpu(buf->CreditCharge);
@@ -2245,6 +2260,7 @@ smb2_async_writev(struct cifs_writedata
 	if (wdata->credits) {
 		req->hdr.CreditCharge = cpu_to_le16(DIV_ROUND_UP(wdata->bytes,
 						    SMB2_MAX_BUFFER_SIZE));
+		req->hdr.CreditRequest = req->hdr.CreditCharge;
 		spin_lock(&server->req_lock);
 		server->credits += wdata->credits -
 					le16_to_cpu(req->hdr.CreditCharge);