Avoid unaligned loads/stores in USE=hpn cipher code #498632 by Bruno.

(Portage version: 2.2.8-r1/cvs/Linux x86_64, signed Manifest commit with key D2E96200)

3 files changed, 111 insertions, 2 deletions

diff --git a/net-misc/openssh/ChangeLog b/net-misc/openssh/ChangeLog
index f8ee609c46b8..5282f4533820 100644
--- a/net-misc/openssh/ChangeLog
+++ b/net-misc/openssh/ChangeLog
@@ -1,6 +1,13 @@
 # ChangeLog for net-misc/openssh
 # Copyright 1999-2014 Gentoo Foundation; Distributed under the GPL v2
-# $Header: /var/cvsroot/gentoo-x86/net-misc/openssh/ChangeLog,v 1.501 2014/03/14 06:56:57 vapier Exp $
+# $Header: /var/cvsroot/gentoo-x86/net-misc/openssh/ChangeLog,v 1.502 2014/03/14 12:12:13 vapier Exp $
+
+*openssh-6.5_p1-r1 (14 Mar 2014)
+
+  14 Mar 2014; Mike Frysinger <vapier@gentoo.org>
+  +files/openssh-6.5_p1-hpn-cipher-align.patch, +openssh-6.5_p1-r1.ebuild,
+  -openssh-6.5_p1.ebuild:
+  Avoid unaligned loads/stores in USE=hpn cipher code #498632 by Bruno.
 
 *openssh-6.5_p1 (14 Mar 2014)
 
diff --git a/net-misc/openssh/files/openssh-6.5_p1-hpn-cipher-align.patch b/net-misc/openssh/files/openssh-6.5_p1-hpn-cipher-align.patch
new file mode 100644
index 000000000000..6610f07c35b9
--- /dev/null
+++ b/net-misc/openssh/files/openssh-6.5_p1-hpn-cipher-align.patch
@@ -0,0 +1,101 @@
+https://bugs.gentoo.org/498632
+
+make sure we do not use unaligned loads/stores as some arches really hate that.
+
+--- a/cipher-ctr-mt.c
++++ b/cipher-ctr-mt.c
+@@ -58,8 +58,10 @@
+ /* Collect thread stats and print at cancellation when in debug mode */
+ /* #define CIPHER_THREAD_STATS */
+ 
+-/* Use single-byte XOR instead of 8-byte XOR */
+-/* #define CIPHER_BYTE_XOR */
++/* Can the system do unaligned loads natively? */
++#if defined(__x86_64__) || defined(__i386__)
++# define CIPHER_UNALIGNED_OK
++#endif
+ /*-------------------- END TUNABLES --------------------*/
+ 
+ 
+@@ -285,8 +286,18 @@ thread_loop(void *x)
+ 
+ static int
+ ssh_aes_ctr(EVP_CIPHER_CTX *ctx, u_char *dest, const u_char *src,
+-    u_int len)
++    size_t len)
+ {
++	typedef union {
++		__uint128_t *u128;
++		uint64_t *u64;
++		uint32_t *u32;
++		uint8_t *u8;
++		const uint8_t *cu8;
++		uintptr_t u;
++	} ptrs_t;
++	ptrs_t destp, srcp, bufp;
++	uintptr_t align;
+ 	struct ssh_aes_ctr_ctx *c;
+ 	struct kq *q, *oldq;
+ 	int ridx;
+@@ -301,35 +312,38 @@ ssh_aes_ctr(EVP_CIPHER_CTX *ctx, u_char *dest, const u_char *src,
+ 	ridx = c->ridx;
+ 
+ 	/* src already padded to block multiple */
++	srcp.cu8 = src;
++	destp.u8 = dest;
+ 	while (len > 0) {
+ 		buf = q->keys[ridx];
++		bufp.u8 = buf;
+ 
+-#ifdef CIPHER_BYTE_XOR
+-		dest[0] = src[0] ^ buf[0];
+-		dest[1] = src[1] ^ buf[1];
+-		dest[2] = src[2] ^ buf[2];
+-		dest[3] = src[3] ^ buf[3];
+-		dest[4] = src[4] ^ buf[4];
+-		dest[5] = src[5] ^ buf[5];
+-		dest[6] = src[6] ^ buf[6];
+-		dest[7] = src[7] ^ buf[7];
+-		dest[8] = src[8] ^ buf[8];
+-		dest[9] = src[9] ^ buf[9];
+-		dest[10] = src[10] ^ buf[10];
+-		dest[11] = src[11] ^ buf[11];
+-		dest[12] = src[12] ^ buf[12];
+-		dest[13] = src[13] ^ buf[13];
+-		dest[14] = src[14] ^ buf[14];
+-		dest[15] = src[15] ^ buf[15];
++#ifdef CIPHER_UNALIGNED_OK
++		destp.u128[0] = srcp.u128[0] ^ bufp.u128[0];
+ #else
+-		*(uint64_t *)dest = *(uint64_t *)src ^ *(uint64_t *)buf;
+-		*(uint64_t *)(dest + 8) = *(uint64_t *)(src + 8) ^
+-						*(uint64_t *)(buf + 8);
++		/* figure out the alignment on the fly */
++		align = destp.u | srcp.u | bufp.u;
++
++		if ((align & 0xf) == 0) {
++			destp.u128[0] = srcp.u128[0] ^ bufp.u128[0];
++		} else if ((align & 0x7) == 0) {
++			destp.u64[0] = srcp.u64[0] ^ bufp.u64[0];
++			destp.u64[1] = srcp.u64[1] ^ bufp.u64[1];
++		} else if ((align & 0x3) == 0) {
++			destp.u32[0] = srcp.u32[0] ^ bufp.u32[0];
++			destp.u32[1] = srcp.u32[1] ^ bufp.u32[1];
++			destp.u32[2] = srcp.u32[2] ^ bufp.u32[2];
++			destp.u32[3] = srcp.u32[3] ^ bufp.u32[3];
++		} else {
++			size_t i;
++			for (i = 0; i < AES_BLOCK_SIZE; ++i)
++				dest[i] = src[i] ^ buf[i];
++		}
+ #endif
+ 
+-		dest += 16;
+-		src += 16;
+-		len -= 16;
++		destp.u += AES_BLOCK_SIZE;
++		srcp.u += AES_BLOCK_SIZE;
++		len -= AES_BLOCK_SIZE;
+ 		ssh_ctr_inc(ctx->iv, AES_BLOCK_SIZE);
+ 
+ 		/* Increment read index, switch queues on rollover */
diff --git a/net-misc/openssh/openssh-6.5_p1.ebuild b/net-misc/openssh/openssh-6.5_p1-r1.ebuild
index 200856eff0db..082e94a5abc2 100644
--- a/net-misc/openssh/openssh-6.5_p1.ebuild
+++ b/net-misc/openssh/openssh-6.5_p1-r1.ebuild
@@ -1,6 +1,6 @@
 # Copyright 1999-2014 Gentoo Foundation
 # Distributed under the terms of the GNU General Public License v2
-# $Header: /var/cvsroot/gentoo-x86/net-misc/openssh/openssh-6.5_p1.ebuild,v 1.1 2014/03/14 06:56:57 vapier Exp $
+# $Header: /var/cvsroot/gentoo-x86/net-misc/openssh/openssh-6.5_p1-r1.ebuild,v 1.1 2014/03/14 12:12:13 vapier Exp $
 
 EAPI="4"
 inherit eutils user flag-o-matic multilib autotools pam systemd versionator
@@ -119,6 +119,7 @@ src_prepare() {
 	epatch "${FILESDIR}"/${PN}-4.7_p1-GSSAPI-dns.patch #165444 integrated into gsskex
 	if [[ -n ${HPN_PATCH} ]] && use hpn; then
 		epatch "${WORKDIR}"/${HPN_PATCH%.*}
+		epatch "${FILESDIR}"/${PN}-6.5_p1-hpn-cipher-align.patch #498632
 		save_version HPN
 	fi