* [PATCH v3 0/5] LivePatch signing support
@ 2025-06-02 13:36 Ross Lagerwall
2025-06-02 13:36 ` [PATCH v3 1/5] docs: Introduce live patch signing Ross Lagerwall
` (4 more replies)
0 siblings, 5 replies; 20+ messages in thread
From: Ross Lagerwall @ 2025-06-02 13:36 UTC (permalink / raw)
To: xen-devel
Cc: Ross Lagerwall, Roger Pau Monné, Andrew Cooper,
Anthony PERARD, Michal Orzel, Jan Beulich, Julien Grall,
Stefano Stabellini
Live patch signing support was mentioned as future work in the design
document several years ago. This series finally implements support for
it since it is a requirement of Secure Boot to prevent loading unsigned
code into Xen.
See the individual patches for what has changed in v3.
Jennifer Herbert (1):
livepatch: Verify livepatch signatures
Kevin Lampis (1):
livepatch: Embed public key in Xen
Ross Lagerwall (3):
docs: Introduce live patch signing
crypto: Add RSA support
livepatch: Load built-in key during boot
docs/misc/livepatch.pandoc | 106 +-
xen/common/Kconfig | 18 +
xen/common/livepatch.c | 139 +++
xen/common/livepatch_elf.c | 55 +
xen/crypto/Makefile | 12 +
xen/crypto/rsa.c | 196 ++++
xen/include/xen/livepatch.h | 15 +
xen/include/xen/livepatch_elf.h | 18 +
xen/include/xen/mpi.h | 68 ++
xen/include/xen/rsa.h | 74 ++
xen/lib/Makefile | 1 +
xen/lib/mpi.c | 1729 +++++++++++++++++++++++++++++++
xen/tools/extract-key.py | 40 +
13 files changed, 2419 insertions(+), 52 deletions(-)
create mode 100644 xen/crypto/rsa.c
create mode 100644 xen/include/xen/mpi.h
create mode 100644 xen/include/xen/rsa.h
create mode 100644 xen/lib/mpi.c
create mode 100755 xen/tools/extract-key.py
--
2.49.0
^ permalink raw reply [flat|nested] 20+ messages in thread
* [PATCH v3 1/5] docs: Introduce live patch signing
2025-06-02 13:36 [PATCH v3 0/5] LivePatch signing support Ross Lagerwall
@ 2025-06-02 13:36 ` Ross Lagerwall
2025-06-02 13:36 ` [PATCH v3 2/5] livepatch: Embed public key in Xen Ross Lagerwall
` (3 subsequent siblings)
4 siblings, 0 replies; 20+ messages in thread
From: Ross Lagerwall @ 2025-06-02 13:36 UTC (permalink / raw)
To: xen-devel; +Cc: Ross Lagerwall, Roger Pau Monné
Remove a never-implemented description of live patch signing from the
TODO section and document signing as implemented by the following
patches.
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
---
No change in v3.
docs/misc/livepatch.pandoc | 106 +++++++++++++++++++------------------
1 file changed, 54 insertions(+), 52 deletions(-)
diff --git a/docs/misc/livepatch.pandoc b/docs/misc/livepatch.pandoc
index 04dd5ed7b271..f36de449e992 100644
--- a/docs/misc/livepatch.pandoc
+++ b/docs/misc/livepatch.pandoc
@@ -917,6 +917,60 @@ The normal sequence of events is to:
3. *XEN_SYSCTL_LIVEPATCH_ACTION* with *LIVEPATCH_ACTION_APPLY* to apply the patch.
4. *XEN_SYSCTL_LIVEPATCH_GET* to check the `->rc`. If in *-XEN_EAGAIN* spin. If zero exit with success.
+## Signature Checking
+
+While loading live patches would generally be restricted to a privileged
+process in dom0, in certain cases signature checking in Xen may be required.
+For example, when Secure Boot is enabled live patches need to be verified
+before being loaded.
+
+Xen live patches are ELF binaries but there is no standardized mechanism for
+signing ELF binaries. One approach used by Linux is to append a signature to
+the end of the binary, outside of the ELF container. While this works, it tends
+to be fragile since tools that handle ELF binaries do not correctly handle the
+signature. Instead, the approach taken here is to use an ELF note for the
+signature.
+
+The ELF note section name shall be `.note.Xen.signature` with note name `Xen`.
+The note type shall encode the signature version, algorithm, and hash:
+
+* version - uint16_t, bits 0-15
+* algorithm - uint8_t, bits 16-23
+* hash - uint8_t, bits 24-31
+
+All other bits of the note type shall be zero.
+
+The known values for the above fields are:
+
+ #define LIVEPATCH_SIGNATURE_VERSION 1
+ #define SIGNATURE_ALGORITHM_RSA 0
+ #define SIGNATURE_HASH_SHA256 0
+
+The note descriptor length defines the length of the signature.
+
+To sign a live patch:
+
+1) Add a new note section with a populated payload signature and zeroed out
+ signature.
+2) Generate a detached signature over the entire binary.
+3) Fill in the signature in the note section.
+
+During live patch load, Xen shall verify the signature using the following
+steps:
+
+1) Copy the signature out of the note section.
+2) Zero the signature.
+3) Generate a detached signature over the entire binary.
+4) Compare against the signature from (1).
+
+Initially, to avoid including DER / X.509 parsing of certificates, handling
+chains, etc. Xen shall verify signatures against a compiled in RSA key in
+exponent/modulus form. However, it may be extended in future to support other
+types of signatures and key types.
+
+Support of signatures in Xen and in live patches is optional. However, certain
+features such as Secure Boot may require live patches to be signed.
+
## Addendum
@@ -1178,58 +1232,6 @@ the function itself.
Similar considerations are true to a lesser extent for \__FILE__, but it
could be argued that file renaming should be done outside of hotpatches.
-## Signature checking requirements.
-
-The signature checking requires that the layout of the data in memory
-**MUST** be same for signature to be verified. This means that the payload
-data layout in ELF format **MUST** match what the hypervisor would be
-expecting such that it can properly do signature verification.
-
-The signature is based on the all of the payloads continuously laid out
-in memory. The signature is to be appended at the end of the ELF payload
-prefixed with the string '`~Module signature appended~\n`', followed by
-an signature header then followed by the signature, key identifier, and signers
-name.
-
-Specifically the signature header would be:
-
- #define PKEY_ALGO_DSA 0
- #define PKEY_ALGO_RSA 1
-
- #define PKEY_ID_PGP 0 /* OpenPGP generated key ID */
- #define PKEY_ID_X509 1 /* X.509 arbitrary subjectKeyIdentifier */
-
- #define HASH_ALGO_MD4 0
- #define HASH_ALGO_MD5 1
- #define HASH_ALGO_SHA1 2
- #define HASH_ALGO_RIPE_MD_160 3
- #define HASH_ALGO_SHA256 4
- #define HASH_ALGO_SHA384 5
- #define HASH_ALGO_SHA512 6
- #define HASH_ALGO_SHA224 7
- #define HASH_ALGO_RIPE_MD_128 8
- #define HASH_ALGO_RIPE_MD_256 9
- #define HASH_ALGO_RIPE_MD_320 10
- #define HASH_ALGO_WP_256 11
- #define HASH_ALGO_WP_384 12
- #define HASH_ALGO_WP_512 13
- #define HASH_ALGO_TGR_128 14
- #define HASH_ALGO_TGR_160 15
- #define HASH_ALGO_TGR_192 16
-
- struct elf_payload_signature {
- u8 algo; /* Public-key crypto algorithm PKEY_ALGO_*. */
- u8 hash; /* Digest algorithm: HASH_ALGO_*. */
- u8 id_type; /* Key identifier type PKEY_ID*. */
- u8 signer_len; /* Length of signer's name */
- u8 key_id_len; /* Length of key identifier */
- u8 __pad[3];
- __be32 sig_len; /* Length of signature data */
- };
-
-(Note that this has been borrowed from Linux module signature code.).
-
-
### .bss and .data sections.
In place patching writable data is not suitable as it is unclear what should be done
--
2.49.0
^ permalink raw reply related [flat|nested] 20+ messages in thread
* [PATCH v3 2/5] livepatch: Embed public key in Xen
2025-06-02 13:36 [PATCH v3 0/5] LivePatch signing support Ross Lagerwall
2025-06-02 13:36 ` [PATCH v3 1/5] docs: Introduce live patch signing Ross Lagerwall
@ 2025-06-02 13:36 ` Ross Lagerwall
2025-06-05 11:02 ` Jan Beulich
` (2 more replies)
2025-06-02 13:36 ` [PATCH v3 3/5] crypto: Add RSA support Ross Lagerwall
` (2 subsequent siblings)
4 siblings, 3 replies; 20+ messages in thread
From: Ross Lagerwall @ 2025-06-02 13:36 UTC (permalink / raw)
To: xen-devel
Cc: Kevin Lampis, Andrew Cooper, Anthony PERARD, Michal Orzel,
Jan Beulich, Julien Grall, Roger Pau Monné,
Stefano Stabellini, Ross Lagerwall
From: Kevin Lampis <kevin.lampis@cloud.com>
Make it possible to embed a public key in Xen to be used when verifying
live patch payloads. Inclusion of the public key is optional.
To avoid needing to include a DER / X.509 parser in the hypervisor, the
public key is unpacked at build time and included in a form that is
convenient for the hypervisor to consume. This is different approach
from that used by Linux which embeds the entire X.509 certificate and
builds in a parser for it.
A suitable key can be created using openssl:
openssl req -x509 -newkey rsa:2048 -keyout priv.pem -out pub.pem \
-sha256 -days 3650 -nodes \
-subj "/C=XX/ST=StateName/L=CityName/O=CompanyName/OU=CompanySectionName/CN=CommonNameOrHostname"
openssl x509 -inform PEM -in pub.pem -outform PEM -pubkey -nocert -out verify_key.pem
Signed-off-by: Kevin Lampis <kevin.lampis@cloud.com>
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
---
In v3:
* Drop unnecessary condition in Makefile
* Use dashes instead of underscores
* Drop section placement annotation on declaration
* Clarify endianness of embedded key
xen/common/Kconfig | 18 +++++++++++++++++
xen/crypto/Makefile | 11 ++++++++++
xen/include/xen/livepatch.h | 5 +++++
| 40 +++++++++++++++++++++++++++++++++++++
4 files changed, 74 insertions(+)
create mode 100755 xen/tools/extract-key.py
diff --git a/xen/common/Kconfig b/xen/common/Kconfig
index 0951d4c2f286..74673078202a 100644
--- a/xen/common/Kconfig
+++ b/xen/common/Kconfig
@@ -472,6 +472,24 @@ config LIVEPATCH
If unsure, say Y.
+config PAYLOAD_VERIFY
+ bool "Verify signed LivePatch payloads"
+ depends on LIVEPATCH
+ select CRYPTO
+ help
+ Verify signed LivePatch payloads using an RSA public key built
+ into the Xen hypervisor. Selecting this option requires a
+ public key in PEM format to be available for embedding during
+ the build.
+
+config PAYLOAD_VERIFY_KEY
+ string "File name of public key used to verify payloads"
+ default "verify_key.pem"
+ depends on PAYLOAD_VERIFY
+ help
+ The file name of an RSA public key in PEM format to be used for
+ verifying signed LivePatch payloads.
+
config FAST_SYMBOL_LOOKUP
bool "Fast symbol lookup (bigger binary)"
default y
diff --git a/xen/crypto/Makefile b/xen/crypto/Makefile
index db29655333a3..3d17232b78bc 100644
--- a/xen/crypto/Makefile
+++ b/xen/crypto/Makefile
@@ -1,2 +1,13 @@
obj-y += rijndael.o
obj-y += vmac.o
+
+obj-$(CONFIG_PAYLOAD_VERIFY) += builtin-payload-key.o
+
+key-path := $(objtree)/$(patsubst "%",%,$(CONFIG_PAYLOAD_VERIFY_KEY))
+$(obj)/builtin-payload-key.bin: $(key-path) $(srctree)/tools/extract-key.py
+ $(srctree)/tools/extract-key.py < $< > $@.new
+ $(call move-if-changed,$@.new,$@)
+
+$(obj)/builtin-payload-key.S: BINFILE_FLAGS := -i
+$(obj)/builtin-payload-key.S: $(srctree)/tools/binfile $(obj)/builtin-payload-key.bin FORCE
+ $(call if_changed,binfile,$(obj)/builtin-payload-key.bin xen_livepatch_key_data)
diff --git a/xen/include/xen/livepatch.h b/xen/include/xen/livepatch.h
index d074a5bebecc..52f90cbed45b 100644
--- a/xen/include/xen/livepatch.h
+++ b/xen/include/xen/livepatch.h
@@ -143,6 +143,11 @@ struct payload;
int revert_payload(struct payload *data);
void revert_payload_tail(struct payload *data);
+#ifdef CONFIG_PAYLOAD_VERIFY
+/* The public key data contained with Xen used to verify payload signatures. */
+extern const uint8_t xen_livepatch_key_data[];
+#endif
+
#else
/*
--git a/xen/tools/extract-key.py b/xen/tools/extract-key.py
new file mode 100755
index 000000000000..05c6bc261b3f
--- /dev/null
+++ b/xen/tools/extract-key.py
@@ -0,0 +1,40 @@
+#!/usr/bin/env python3
+
+# SPDX-License-Identifier: GPL-2.0
+
+import binascii
+import struct
+import sys
+import subprocess
+import re
+
+# Decode a certificate into a format suitable for embedding in Xen.
+
+out = subprocess.check_output(['openssl', 'rsa', '-pubin', '-inform', 'PEM',
+ '-noout', '-text'], stdin=sys.stdin,
+ universal_newlines=True)
+combined = ''
+for line in out.split('\n'):
+ line = line.rstrip()
+ if line.startswith(' '):
+ combined += line.strip().replace(':', '')
+ match = re.match(r'Exponent: .* \(0x(.*)\)', line)
+ if match:
+ e = match.group(1)
+
+n = combined.lstrip('0')
+if len(n) % 2 == 1:
+ n = '0' + n
+n = binascii.unhexlify(n)
+e = e.lstrip('0')
+if len(e) % 2 == 1:
+ e = '0' + e
+e = binascii.unhexlify(e)
+
+# Use little-endian for lengths.
+# Use big-endian for byte sequences (as openssl does).
+
+sys.stdout.buffer.write(struct.pack('<I', len(n)))
+sys.stdout.buffer.write(n)
+sys.stdout.buffer.write(struct.pack('<I', len(e)))
+sys.stdout.buffer.write(e)
--
2.49.0
^ permalink raw reply related [flat|nested] 20+ messages in thread
* [PATCH v3 3/5] crypto: Add RSA support
2025-06-02 13:36 [PATCH v3 0/5] LivePatch signing support Ross Lagerwall
2025-06-02 13:36 ` [PATCH v3 1/5] docs: Introduce live patch signing Ross Lagerwall
2025-06-02 13:36 ` [PATCH v3 2/5] livepatch: Embed public key in Xen Ross Lagerwall
@ 2025-06-02 13:36 ` Ross Lagerwall
2025-06-05 11:06 ` Jan Beulich
2025-06-20 9:53 ` Roger Pau Monné
2025-06-02 13:36 ` [PATCH v3 4/5] livepatch: Load built-in key during boot Ross Lagerwall
2025-06-02 13:36 ` [PATCH v3 5/5] livepatch: Verify livepatch signatures Ross Lagerwall
4 siblings, 2 replies; 20+ messages in thread
From: Ross Lagerwall @ 2025-06-02 13:36 UTC (permalink / raw)
To: xen-devel
Cc: Ross Lagerwall, Andrew Cooper, Anthony PERARD, Michal Orzel,
Jan Beulich, Julien Grall, Roger Pau Monné,
Stefano Stabellini
In preparation for adding support for livepatch signing, add support for
RSA crypto.
The RSA code is extracted from Nettle at tag nettle_3.2_release_20160128
(https://git.lysator.liu.se/nettle/nettle).
The MPI code is extracted from Linux at commit eef0df6a5953 (lib/mpi/*).
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
---
In v3:
* Move mpi.c to lib
* Fix header guard name
xen/crypto/Makefile | 1 +
xen/crypto/rsa.c | 196 +++++
xen/include/xen/mpi.h | 68 ++
xen/include/xen/rsa.h | 74 ++
xen/lib/Makefile | 1 +
xen/lib/mpi.c | 1729 +++++++++++++++++++++++++++++++++++++++++
6 files changed, 2069 insertions(+)
create mode 100644 xen/crypto/rsa.c
create mode 100644 xen/include/xen/mpi.h
create mode 100644 xen/include/xen/rsa.h
create mode 100644 xen/lib/mpi.c
diff --git a/xen/crypto/Makefile b/xen/crypto/Makefile
index 3d17232b78bc..55a045a7f0c0 100644
--- a/xen/crypto/Makefile
+++ b/xen/crypto/Makefile
@@ -1,4 +1,5 @@
obj-y += rijndael.o
+obj-$(CONFIG_PAYLOAD_VERIFY) += rsa.o
obj-y += vmac.o
obj-$(CONFIG_PAYLOAD_VERIFY) += builtin-payload-key.o
diff --git a/xen/crypto/rsa.c b/xen/crypto/rsa.c
new file mode 100644
index 000000000000..bd78c65f7393
--- /dev/null
+++ b/xen/crypto/rsa.c
@@ -0,0 +1,196 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later OR LGPL-3.0-or-later */
+/* rsa.c
+
+ The RSA publickey algorithm.
+
+ Copyright (C) 2001 Niels Möller
+
+ This file is part of GNU Nettle.
+
+ GNU Nettle is free software: you can redistribute it and/or
+ modify it under the terms of either:
+
+ * the GNU Lesser General Public License as published by the Free
+ Software Foundation; either version 3 of the License, or (at your
+ option) any later version.
+
+ or
+
+ * the GNU General Public License as published by the Free
+ Software Foundation; either version 2 of the License, or (at your
+ option) any later version.
+
+ or both in parallel, as here.
+
+ GNU Nettle is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ General Public License for more details.
+
+ You should have received copies of the GNU General Public License and
+ the GNU Lesser General Public License along with this program. If
+ not, see http://www.gnu.org/licenses/.
+*/
+
+#include <xen/bug.h>
+#include <xen/err.h>
+#include <xen/lib.h>
+#include <xen/rsa.h>
+#include <xen/sha2.h>
+#include <xen/string.h>
+
+void rsa_public_key_init(struct rsa_public_key *key)
+{
+ key->n = NULL;
+ key->e = NULL;
+ key->size = 0;
+}
+
+/*
+ * Computes the size, in octets, of the modulo. Returns 0 if the
+ * modulo is too small to be useful, or otherwise appears invalid.
+ */
+static size_t rsa_check_size(MPI n)
+{
+ /* Round upwards */
+ size_t size;
+
+ /* Even moduli are invalid */
+ if ( mpi_test_bit(n, 0) == 0 )
+ return 0;
+
+ size = (mpi_get_nbits(n) + 7) / 8;
+
+ if ( size < RSA_MINIMUM_N_OCTETS )
+ return 0;
+
+ return size;
+}
+
+int rsa_public_key_prepare(struct rsa_public_key *key)
+{
+ if ( !key->n || !key->e || key->size )
+ return -EINVAL;
+
+ key->size = rsa_check_size(key->n);
+
+ return key->size > 0 ? 0 : -EINVAL;
+}
+
+/*
+ * Formats the PKCS#1 padding, of the form
+ *
+ * 0x00 0x01 0xff ... 0xff 0x00 id ...digest...
+ *
+ * where the 0xff ... 0xff part consists of at least 8 octets. The
+ * total size equals the octet size of n.
+ */
+static uint8_t *pkcs1_signature_prefix(unsigned int key_size, uint8_t *buffer,
+ unsigned int id_size, const uint8_t *id,
+ unsigned int digest_size)
+{
+ unsigned int j;
+
+ if ( key_size < 11 + id_size + digest_size )
+ return NULL;
+
+ j = key_size - digest_size - id_size;
+
+ memcpy(buffer + j, id, id_size);
+ buffer[0] = 0;
+ buffer[1] = 1;
+ buffer[j - 1] = 0;
+
+ ASSERT(j >= 11);
+ memset(buffer + 2, 0xff, j - 3);
+
+ return buffer + j + id_size;
+}
+
+/*
+ * From RFC 3447, Public-Key Cryptography Standards (PKCS) #1: RSA
+ * Cryptography Specifications Version 2.1.
+ *
+ * id-sha256 OBJECT IDENTIFIER ::=
+ * {joint-iso-itu-t(2) country(16) us(840) organization(1)
+ * gov(101) csor(3) nistalgorithm(4) hashalgs(2) 1}
+ */
+static const uint8_t
+sha256_prefix[] =
+{
+ /* 19 octets prefix, 32 octets hash, total 51 */
+ 0x30, 49, /* SEQUENCE */
+ 0x30, 13, /* SEQUENCE */
+ 0x06, 9, /* OBJECT IDENTIFIER */
+ 0x60, 0x86, 0x48, 0x01, 0x65, 0x03, 0x04, 0x02, 0x01,
+ 0x05, 0, /* NULL */
+ 0x04, 32 /* OCTET STRING */
+ /* Here comes the raw hash value */
+};
+
+static int pkcs1_rsa_sha256_encode(MPI *m, size_t key_size,
+ struct sha2_256_state *hash)
+{
+ uint8_t *ptr;
+ uint8_t *buf;
+
+ buf = xmalloc_bytes(key_size);
+ if ( !buf )
+ return -ENOMEM;
+
+ ptr = pkcs1_signature_prefix(key_size, buf,
+ sizeof(sha256_prefix), sha256_prefix,
+ SHA2_256_DIGEST_SIZE);
+ if ( !ptr )
+ {
+ xfree(buf);
+ return -EINVAL;
+ }
+
+ sha2_256_final(hash, ptr);
+ *m = mpi_read_raw_data(buf, key_size);
+ xfree(buf);
+ return 0;
+}
+
+static int rsa_verify(const struct rsa_public_key *key, MPI m, MPI s)
+{
+ int ret;
+ MPI m1;
+
+ /* (1) Validate 0 <= s < n */
+ if ( mpi_cmp_ui(s, 0) < 0 || mpi_cmp(s, key->n) >= 0 )
+ return -EINVAL;
+
+ m1 = mpi_alloc(key->size / BYTES_PER_MPI_LIMB);
+ if ( !m1 )
+ return -ENOMEM;
+
+ /* (2) m = s^e mod n */
+ ret = mpi_powm(m1, s, key->e, key->n);
+ if ( ret )
+ goto out;
+
+ ret = mpi_cmp(m, m1) ? -EINVAL : 0;
+
+ out:
+ mpi_free(m1);
+ return ret;
+}
+
+int rsa_sha256_verify(const struct rsa_public_key *key,
+ struct sha2_256_state *hash, MPI s)
+{
+ int ret;
+ MPI m;
+
+ ret = pkcs1_rsa_sha256_encode(&m, key->size, hash);
+ if ( ret )
+ return ret;
+
+ ret = rsa_verify(key, m, s);
+
+ mpi_free(m);
+
+ return ret;
+}
diff --git a/xen/include/xen/mpi.h b/xen/include/xen/mpi.h
new file mode 100644
index 000000000000..2d6adb8d8d0a
--- /dev/null
+++ b/xen/include/xen/mpi.h
@@ -0,0 +1,68 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+/* mpi.h - Multi Precision Integers
+ * Copyright (C) 1994, 1996, 1998, 1999,
+ * 2000, 2001 Free Software Foundation, Inc.
+ *
+ * This file is part of GNUPG.
+ *
+ * GNUPG is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * GNUPG is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA
+ *
+ * Note: This code is heavily based on the GNU MP Library.
+ * Actually it's the same code with only minor changes in the
+ * way the data is stored; this is to support the abstraction
+ * of an optional secure memory allocation which may be used
+ * to avoid revealing of sensitive data due to paging etc.
+ * The GNU MP Library itself is published under the LGPL;
+ * however I decided to publish this code under the plain GPL.
+ */
+
+#ifndef XEN_MPI_H
+#define XEN_MPI_H
+
+#include <xen/types.h>
+
+#define BYTES_PER_MPI_LIMB (BITS_PER_LONG / 8)
+#define BITS_PER_MPI_LIMB BITS_PER_LONG
+
+typedef unsigned long int mpi_limb_t;
+typedef signed long int mpi_limb_signed_t;
+
+struct mpi {
+ int alloced; /* array size (# of allocated limbs) */
+ int nlimbs; /* number of valid limbs */
+ int nbits; /* the real number of valid bits (info only) */
+ int sign; /* indicates a negative number */
+#define MPI_FLAG_SECURE_MEM (1 << 0)
+#define MPI_FLAG_UNUSED (1 << 1)
+#define MPI_FLAG_PTR_ALLOC (1 << 4)
+#define MPI_FLAG_MASK 7
+ unsigned int flags; /* bit 0: array must be allocated in secure memory space */
+ /* bit 1: not used */
+ /* bit 2: the limb is a pointer to some m_alloced data */
+ mpi_limb_t *d; /* array with the limbs */
+};
+
+typedef struct mpi *MPI;
+
+MPI mpi_alloc(unsigned nlimbs);
+void mpi_free(MPI a);
+MPI mpi_read_raw_data(const void *xbuffer, size_t nbytes);
+int mpi_powm(MPI res, MPI base, MPI exp, MPI mod);
+int mpi_cmp_ui(MPI u, unsigned long v);
+int mpi_cmp(MPI u, MPI v);
+unsigned mpi_get_nbits(MPI a);
+int mpi_test_bit(MPI a, unsigned int n);
+
+#endif /* XEN_MPI_H */
diff --git a/xen/include/xen/rsa.h b/xen/include/xen/rsa.h
new file mode 100644
index 000000000000..be60383b18d5
--- /dev/null
+++ b/xen/include/xen/rsa.h
@@ -0,0 +1,74 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later OR LGPL-3.0-or-later */
+/* rsa.h
+
+ The RSA publickey algorithm.
+
+ Copyright (C) 2001, 2002 Niels Möller
+
+ This file is part of GNU Nettle.
+
+ GNU Nettle is free software: you can redistribute it and/or
+ modify it under the terms of either:
+
+ * the GNU Lesser General Public License as published by the Free
+ Software Foundation; either version 3 of the License, or (at your
+ option) any later version.
+
+ or
+
+ * the GNU General Public License as published by the Free
+ Software Foundation; either version 2 of the License, or (at your
+ option) any later version.
+
+ or both in parallel, as here.
+
+ GNU Nettle is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ General Public License for more details.
+
+ You should have received copies of the GNU General Public License and
+ the GNU Lesser General Public License along with this program. If
+ not, see http://www.gnu.org/licenses/.
+*/
+
+#ifndef XEN_RSA_H
+#define XEN_RSA_H
+
+#include <xen/mpi.h>
+#include <xen/types.h>
+
+struct sha2_256_state;
+
+/*
+ * This limit is somewhat arbitrary. Technically, the smallest modulo
+ * which makes sense at all is 15 = 3*5, phi(15) = 8, size 4 bits. But
+ * for ridiculously small keys, not all odd e are possible (e.g., for
+ * 5 bits, the only possible modulo is 3*7 = 21, phi(21) = 12, and e =
+ * 3 don't work). The smallest size that makes sense with pkcs#1, and
+ * which allows RSA encryption of one byte messages, is 12 octets, 89
+ * bits.
+ */
+#define RSA_MINIMUM_N_OCTETS 12
+#define RSA_MINIMUM_N_BITS (8 * RSA_MINIMUM_N_OCTETS - 7)
+
+struct rsa_public_key
+{
+ /*
+ * Size of the modulo, in octets. This is also the size of all
+ * signatures that are created or verified with this key.
+ */
+ size_t size;
+ MPI n; /* Modulo */
+ MPI e; /* Public exponent */
+};
+
+void rsa_public_key_init(struct rsa_public_key *key);
+
+int rsa_public_key_prepare(struct rsa_public_key *key);
+
+int rsa_sha256_verify(const struct rsa_public_key *key,
+ struct sha2_256_state *hash,
+ MPI signature);
+
+#endif /* XEN_RSA_H */
diff --git a/xen/lib/Makefile b/xen/lib/Makefile
index 5ccb1e5241c5..f37367277ede 100644
--- a/xen/lib/Makefile
+++ b/xen/lib/Makefile
@@ -14,6 +14,7 @@ lib-y += memcmp.o
lib-y += memcpy.o
lib-y += memmove.o
lib-y += memset.o
+lib-y += mpi.o
lib-y += muldiv64.o
lib-y += parse-size.o
lib-y += rbtree.o
diff --git a/xen/lib/mpi.c b/xen/lib/mpi.c
new file mode 100644
index 000000000000..94010d14b3d0
--- /dev/null
+++ b/xen/lib/mpi.c
@@ -0,0 +1,1729 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+/* mpi.c - MPI functions
+ * Copyright (C) 1994, 1996, 1998, 2000 Free Software Foundation, Inc.
+ *
+ * This file is part of GnuPG.
+ *
+ * GnuPG is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * GnuPG is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA
+ *
+ * Note: This code is heavily based on the GNU MP Library.
+ * Actually it's the same code with only minor changes in the
+ * way the data is stored; this is to support the abstraction
+ * of an optional secure memory allocation which may be used
+ * to avoid revealing of sensitive data due to paging etc.
+ * The GNU MP Library itself is published under the LGPL;
+ * however I decided to publish this code under the plain GPL.
+ */
+
+#include <xen/bitops.h>
+#include <xen/bug.h>
+#include <xen/err.h>
+#include <xen/lib.h>
+#include <xen/mpi.h>
+#include <xen/string.h>
+#include <xen/xmalloc.h>
+
+#define MAX_EXTERN_MPI_BITS 16384
+
+/* Define it to a value which is good on most machines.
+ * tested 4, 16, 32 and 64, where 16 gave the best performance when
+ * checking a 768 and a 1024 bit ElGamal signature.
+ * (wk 22.12.97) */
+#define KARATSUBA_THRESHOLD 16
+
+typedef mpi_limb_t *mpi_ptr_t; /* pointer to a limb */
+typedef int mpi_size_t; /* (must be a signed type) */
+
+/* Copy N limbs from S to D. */
+#define MPN_COPY(d, s, n) \
+ do { \
+ mpi_size_t _i; \
+ for (_i = 0; _i < (n); _i++) \
+ (d)[_i] = (s)[_i]; \
+ } while (0)
+
+#define MPN_COPY_DECR(d, s, n) \
+ do { \
+ mpi_size_t _i; \
+ for (_i = (n)-1; _i >= 0; _i--) \
+ (d)[_i] = (s)[_i]; \
+ } while (0)
+
+/* Zero N limbs at D */
+#define MPN_ZERO(d, n) \
+ do { \
+ mpi_size_t _i; \
+ for (_i = 0; _i < (n); _i++) \
+ (d)[_i] = 0; \
+ } while (0)
+
+#define MPN_NORMALIZE(d, n) \
+ do { \
+ while ((n) > 0) { \
+ if ((d)[(n)-1]) \
+ break; \
+ (n)--; \
+ } \
+ } while (0)
+
+#define MPN_MUL_N_RECURSE(prodp, up, vp, size, tspace) \
+ do { \
+ if ((size) < KARATSUBA_THRESHOLD) \
+ mul_n_basecase(prodp, up, vp, size); \
+ else \
+ mul_n(prodp, up, vp, size, tspace); \
+ } while (0);
+
+#define MPN_SQR_N_RECURSE(prodp, up, size, tspace) \
+ do { \
+ if ((size) < KARATSUBA_THRESHOLD) \
+ mpih_sqr_n_basecase(prodp, up, size); \
+ else \
+ mpih_sqr_n(prodp, up, size, tspace); \
+ } while (0);
+
+#define add_ssaaaa(sh, sl, ah, al, bh, bl) \
+do { \
+ mpi_limb_t __x; \
+ __x = (al) + (bl); \
+ (sh) = (ah) + (bh) + (__x < (al)); \
+ (sl) = __x; \
+} while (0)
+
+#define sub_ddmmss(sh, sl, ah, al, bh, bl) \
+do { \
+ mpi_limb_t __x; \
+ __x = (al) - (bl); \
+ (sh) = (ah) - (bh) - (__x > (al)); \
+ (sl) = __x; \
+} while (0)
+
+#define __ll_B ((mpi_limb_t) 1 << (BITS_PER_MPI_LIMB / 2))
+#define __ll_lowpart(t) ((mpi_limb_t) (t) & (__ll_B - 1))
+#define __ll_highpart(t) ((mpi_limb_t) (t) >> (BITS_PER_MPI_LIMB / 2))
+
+#define umul_ppmm(w1, w0, u, v) \
+do { \
+ mpi_limb_t __x0, __x1, __x2, __x3; \
+ unsigned int __ul, __vl, __uh, __vh; \
+ mpi_limb_t __u = (u), __v = (v); \
+ \
+ __ul = __ll_lowpart(__u); \
+ __uh = __ll_highpart(__u); \
+ __vl = __ll_lowpart(__v); \
+ __vh = __ll_highpart(__v); \
+ \
+ __x0 = (mpi_limb_t) __ul * __vl; \
+ __x1 = (mpi_limb_t) __ul * __vh; \
+ __x2 = (mpi_limb_t) __uh * __vl; \
+ __x3 = (mpi_limb_t) __uh * __vh; \
+ \
+ __x1 += __ll_highpart(__x0);/* this can't give carry */ \
+ __x1 += __x2; /* but this indeed can */ \
+ if (__x1 < __x2) /* did we get it? */ \
+ __x3 += __ll_B; /* yes, add it in the proper pos. */ \
+ \
+ (w1) = __x3 + __ll_highpart(__x1); \
+ (w0) = (__ll_lowpart(__x1) << BITS_PER_MPI_LIMB/2) + __ll_lowpart(__x0); \
+} while (0)
+
+#define udiv_qrnnd(q, r, n1, n0, d) \
+do { \
+ mpi_limb_t __d1, __d0, __q1, __q0, __r1, __r0, __m; \
+ __d1 = __ll_highpart(d); \
+ __d0 = __ll_lowpart(d); \
+ \
+ __r1 = (n1) % __d1; \
+ __q1 = (n1) / __d1; \
+ __m = (mpi_limb_t) __q1 * __d0; \
+ __r1 = __r1 * __ll_B | __ll_highpart(n0); \
+ if (__r1 < __m) { \
+ __q1--, __r1 += (d); \
+ if (__r1 >= (d)) /* i.e. we didn't get carry when adding to __r1 */ \
+ if (__r1 < __m) \
+ __q1--, __r1 += (d); \
+ } \
+ __r1 -= __m; \
+ \
+ __r0 = __r1 % __d1; \
+ __q0 = __r1 / __d1; \
+ __m = (mpi_limb_t) __q0 * __d0; \
+ __r0 = __r0 * __ll_B | __ll_lowpart(n0); \
+ if (__r0 < __m) { \
+ __q0--, __r0 += (d); \
+ if (__r0 >= (d)) \
+ if (__r0 < __m) \
+ __q0--, __r0 += (d); \
+ } \
+ __r0 -= __m; \
+ \
+ (q) = (mpi_limb_t) __q1 * __ll_B | __q0; \
+ (r) = __r0; \
+} while (0)
+
+struct karatsuba_ctx {
+ struct karatsuba_ctx *next;
+ mpi_ptr_t tspace;
+ mpi_size_t tspace_size;
+ mpi_ptr_t tp;
+ mpi_size_t tp_size;
+};
+
+static void mpi_normalize(MPI a);
+static mpi_limb_t mpihelp_submul_1(mpi_ptr_t res_ptr, mpi_ptr_t s1_ptr,
+ mpi_size_t s1_size, mpi_limb_t s2_limb);
+static mpi_limb_t mpihelp_divrem(mpi_ptr_t qp, mpi_size_t qextra_limbs,
+ mpi_ptr_t np, mpi_size_t nsize,
+ mpi_ptr_t dp, mpi_size_t dsize);
+static mpi_limb_t mpihelp_rshift(mpi_ptr_t wp, mpi_ptr_t up, mpi_size_t usize,
+ unsigned cnt);
+static void mpi_assign_limb_space(MPI a, mpi_ptr_t ap, unsigned nlimbs);
+static mpi_ptr_t mpi_alloc_limb_space(unsigned nlimbs);
+static void mpi_free_limb_space(mpi_ptr_t a);
+static mpi_limb_t mpihelp_addmul_1(mpi_ptr_t res_ptr, mpi_ptr_t s1_ptr,
+ mpi_size_t s1_size, mpi_limb_t s2_limb);
+static int mpihelp_mul(mpi_ptr_t prodp, mpi_ptr_t up, mpi_size_t usize,
+ mpi_ptr_t vp, mpi_size_t vsize, mpi_limb_t *_result);
+static mpi_limb_t mpihelp_lshift(mpi_ptr_t wp, mpi_ptr_t up, mpi_size_t usize,
+ unsigned cnt);
+static int mpihelp_cmp(mpi_ptr_t op1_ptr, mpi_ptr_t op2_ptr, mpi_size_t size);
+static void mpih_sqr_n(mpi_ptr_t prodp, mpi_ptr_t up, mpi_size_t size,
+ mpi_ptr_t tspace);
+static mpi_limb_t mpihelp_add_n(mpi_ptr_t res_ptr, mpi_ptr_t s1_ptr,
+ mpi_ptr_t s2_ptr, mpi_size_t size);
+static mpi_limb_t mpihelp_sub_n(mpi_ptr_t res_ptr, mpi_ptr_t s1_ptr,
+ mpi_ptr_t s2_ptr, mpi_size_t size);
+static mpi_limb_t mpihelp_mul_1(mpi_ptr_t res_ptr, mpi_ptr_t s1_ptr,
+ mpi_size_t s1_size, mpi_limb_t s2_limb);
+static void mpihelp_release_karatsuba_ctx(struct karatsuba_ctx *ctx);
+static void mpih_sqr_n_basecase(mpi_ptr_t prodp, mpi_ptr_t up, mpi_size_t size);
+static int mpihelp_mul_karatsuba_case(mpi_ptr_t prodp,
+ mpi_ptr_t up, mpi_size_t usize,
+ mpi_ptr_t vp, mpi_size_t vsize,
+ struct karatsuba_ctx *ctx);
+static int mpi_resize(MPI a, unsigned nlimbs);
+
+/**
+ * count_leading_zeros - Count the number of zeros from the MSB back
+ * @x: The value
+ *
+ * Count the number of leading zeros from the MSB going towards the LSB in @x.
+ *
+ * If the MSB of @x is set, the result is 0.
+ * If only the LSB of @x is set, then the result is BITS_PER_LONG-1.
+ * If @x is 0 then the result is BITS_PER_LONG.
+ */
+static inline int count_leading_zeros(unsigned long x)
+{
+ if (sizeof(x) == 4)
+ return BITS_PER_LONG - fls(x);
+ else
+ return BITS_PER_LONG - fls64(x);
+}
+
+static mpi_limb_t
+mpihelp_add_1(mpi_ptr_t res_ptr, mpi_ptr_t s1_ptr,
+ mpi_size_t s1_size, mpi_limb_t s2_limb)
+{
+ mpi_limb_t x;
+
+ x = *s1_ptr++;
+ s2_limb += x;
+ *res_ptr++ = s2_limb;
+ if (s2_limb < x) { /* sum is less than the left operand: handle carry */
+ while (--s1_size) {
+ x = *s1_ptr++ + 1; /* add carry */
+ *res_ptr++ = x; /* and store */
+ if (x) /* not 0 (no overflow): we can stop */
+ goto leave;
+ }
+ return 1; /* return carry (size of s1 to small) */
+ }
+
+ leave:
+ if (res_ptr != s1_ptr) { /* not the same variable */
+ mpi_size_t i; /* copy the rest */
+ for (i = 0; i < s1_size - 1; i++)
+ res_ptr[i] = s1_ptr[i];
+ }
+ return 0; /* no carry */
+}
+
+static mpi_limb_t
+mpihelp_sub_1(mpi_ptr_t res_ptr, mpi_ptr_t s1_ptr,
+ mpi_size_t s1_size, mpi_limb_t s2_limb)
+{
+ mpi_limb_t x;
+
+ x = *s1_ptr++;
+ s2_limb = x - s2_limb;
+ *res_ptr++ = s2_limb;
+ if (s2_limb > x) {
+ while (--s1_size) {
+ x = *s1_ptr++;
+ *res_ptr++ = x - 1;
+ if (x)
+ goto leave;
+ }
+ return 1;
+ }
+
+ leave:
+ if (res_ptr != s1_ptr) {
+ mpi_size_t i;
+ for (i = 0; i < s1_size - 1; i++)
+ res_ptr[i] = s1_ptr[i];
+ }
+ return 0;
+}
+
+static mpi_limb_t
+mpihelp_sub(mpi_ptr_t res_ptr, mpi_ptr_t s1_ptr, mpi_size_t s1_size,
+ mpi_ptr_t s2_ptr, mpi_size_t s2_size)
+{
+ mpi_limb_t cy = 0;
+
+ if (s2_size)
+ cy = mpihelp_sub_n(res_ptr, s1_ptr, s2_ptr, s2_size);
+
+ if (s1_size - s2_size)
+ cy = mpihelp_sub_1(res_ptr + s2_size, s1_ptr + s2_size,
+ s1_size - s2_size, cy);
+ return cy;
+}
+
+static mpi_ptr_t mpi_alloc_limb_space(unsigned nlimbs)
+{
+ size_t len = nlimbs * sizeof(mpi_limb_t);
+
+ if (!len)
+ return NULL;
+
+ return xmalloc_bytes(len);
+}
+
+static void mpi_free_limb_space(mpi_ptr_t a)
+{
+ if (!a)
+ return;
+
+ xfree(a);
+}
+
+/****************
+ * Resize the array of A to NLIMBS. the additional space is cleared
+ * (set to 0) [done by m_realloc()]
+ */
+static int mpi_resize(MPI a, unsigned nlimbs)
+{
+ void *p;
+
+ if (nlimbs <= a->alloced)
+ return 0; /* no need to do it */
+
+ if (a->d) {
+ p = xmalloc_array(mpi_limb_t, nlimbs);
+ if (!p)
+ return -ENOMEM;
+ memcpy(p, a->d, a->alloced * sizeof(mpi_limb_t));
+ xfree(a->d);
+ a->d = p;
+ } else {
+ a->d = xzalloc_array(mpi_limb_t, nlimbs);
+ if (!a->d)
+ return -ENOMEM;
+ }
+ a->alloced = nlimbs;
+ return 0;
+}
+
+/****************
+ * RES = BASE ^ EXP mod MOD
+ */
+int mpi_powm(MPI res, MPI base, MPI exp, MPI mod)
+{
+ mpi_ptr_t mp_marker = NULL, bp_marker = NULL, ep_marker = NULL;
+ mpi_ptr_t xp_marker = NULL;
+ mpi_ptr_t tspace = NULL;
+ mpi_ptr_t rp, ep, mp, bp;
+ mpi_size_t esize, msize, bsize, rsize;
+ int msign, bsign, rsign;
+ mpi_size_t size;
+ int mod_shift_cnt;
+ int negative_result;
+ int assign_rp = 0;
+ mpi_size_t tsize = 0; /* to avoid compiler warning */
+ /* fixme: we should check that the warning is void */
+ int rc = -ENOMEM;
+
+ esize = exp->nlimbs;
+ msize = mod->nlimbs;
+ size = 2 * msize;
+ msign = mod->sign;
+
+ rp = res->d;
+ ep = exp->d;
+
+ if (!msize)
+ return -EINVAL;
+
+ if (!esize) {
+ /* Exponent is zero, result is 1 mod MOD, i.e., 1 or 0
+ * depending on if MOD equals 1. */
+ res->nlimbs = (msize == 1 && mod->d[0] == 1) ? 0 : 1;
+ if (res->nlimbs) {
+ if (mpi_resize(res, 1) < 0)
+ goto enomem;
+ rp = res->d;
+ rp[0] = 1;
+ }
+ res->sign = 0;
+ goto leave;
+ }
+
+ /* Normalize MOD (i.e. make its most significant bit set) as required by
+ * mpn_divrem. This will make the intermediate values in the calculation
+ * slightly larger, but the correct result is obtained after a final
+ * reduction using the original MOD value. */
+ mp = mp_marker = mpi_alloc_limb_space(msize);
+ if (!mp)
+ goto enomem;
+ mod_shift_cnt = count_leading_zeros(mod->d[msize - 1]);
+ if (mod_shift_cnt)
+ mpihelp_lshift(mp, mod->d, msize, mod_shift_cnt);
+ else
+ MPN_COPY(mp, mod->d, msize);
+
+ bsize = base->nlimbs;
+ bsign = base->sign;
+ if (bsize > msize) { /* The base is larger than the module. Reduce it. */
+ /* Allocate (BSIZE + 1) with space for remainder and quotient.
+ * (The quotient is (bsize - msize + 1) limbs.) */
+ bp = bp_marker = mpi_alloc_limb_space(bsize + 1);
+ if (!bp)
+ goto enomem;
+ MPN_COPY(bp, base->d, bsize);
+ /* We don't care about the quotient, store it above the remainder,
+ * at BP + MSIZE. */
+ mpihelp_divrem(bp + msize, 0, bp, bsize, mp, msize);
+ bsize = msize;
+ /* Canonicalize the base, since we are going to multiply with it
+ * quite a few times. */
+ MPN_NORMALIZE(bp, bsize);
+ } else
+ bp = base->d;
+
+ if (!bsize) {
+ res->nlimbs = 0;
+ res->sign = 0;
+ goto leave;
+ }
+
+ if (res->alloced < size) {
+ /* We have to allocate more space for RES. If any of the input
+ * parameters are identical to RES, defer deallocation of the old
+ * space. */
+ if (rp == ep || rp == mp || rp == bp) {
+ rp = mpi_alloc_limb_space(size);
+ if (!rp)
+ goto enomem;
+ assign_rp = 1;
+ } else {
+ if (mpi_resize(res, size) < 0)
+ goto enomem;
+ rp = res->d;
+ }
+ } else { /* Make BASE, EXP and MOD not overlap with RES. */
+ if (rp == bp) {
+ /* RES and BASE are identical. Allocate temp. space for BASE. */
+ BUG_ON(bp_marker);
+ bp = bp_marker = mpi_alloc_limb_space(bsize);
+ if (!bp)
+ goto enomem;
+ MPN_COPY(bp, rp, bsize);
+ }
+ if (rp == ep) {
+ /* RES and EXP are identical. Allocate temp. space for EXP. */
+ ep = ep_marker = mpi_alloc_limb_space(esize);
+ if (!ep)
+ goto enomem;
+ MPN_COPY(ep, rp, esize);
+ }
+ if (rp == mp) {
+ /* RES and MOD are identical. Allocate temporary space for MOD. */
+ BUG_ON(mp_marker);
+ mp = mp_marker = mpi_alloc_limb_space(msize);
+ if (!mp)
+ goto enomem;
+ MPN_COPY(mp, rp, msize);
+ }
+ }
+
+ MPN_COPY(rp, bp, bsize);
+ rsize = bsize;
+ rsign = bsign;
+
+ {
+ mpi_size_t i;
+ mpi_ptr_t xp;
+ int c;
+ mpi_limb_t e;
+ mpi_limb_t carry_limb;
+ struct karatsuba_ctx karactx;
+
+ xp = xp_marker = mpi_alloc_limb_space(2 * (msize + 1));
+ if (!xp)
+ goto enomem;
+
+ memset(&karactx, 0, sizeof karactx);
+ negative_result = (ep[0] & 1) && base->sign;
+
+ i = esize - 1;
+ e = ep[i];
+ c = count_leading_zeros(e);
+ e = (e << c) << 1; /* shift the exp bits to the left, lose msb */
+ c = BITS_PER_MPI_LIMB - 1 - c;
+
+ /* Main loop.
+ *
+ * Make the result be pointed to alternately by XP and RP. This
+ * helps us avoid block copying, which would otherwise be necessary
+ * with the overlap restrictions of mpihelp_divmod. With 50% probability
+ * the result after this loop will be in the area originally pointed
+ * by RP (==RES->d), and with 50% probability in the area originally
+ * pointed to by XP.
+ */
+
+ for (;;) {
+ while (c) {
+ mpi_ptr_t tp;
+ mpi_size_t xsize;
+
+ /*if (mpihelp_mul_n(xp, rp, rp, rsize) < 0) goto enomem */
+ if (rsize < KARATSUBA_THRESHOLD)
+ mpih_sqr_n_basecase(xp, rp, rsize);
+ else {
+ if (!tspace) {
+ tsize = 2 * rsize;
+ tspace =
+ mpi_alloc_limb_space(tsize);
+ if (!tspace)
+ goto enomem;
+ } else if (tsize < (2 * rsize)) {
+ mpi_free_limb_space(tspace);
+ tsize = 2 * rsize;
+ tspace =
+ mpi_alloc_limb_space(tsize);
+ if (!tspace)
+ goto enomem;
+ }
+ mpih_sqr_n(xp, rp, rsize, tspace);
+ }
+
+ xsize = 2 * rsize;
+ if (xsize > msize) {
+ mpihelp_divrem(xp + msize, 0, xp, xsize,
+ mp, msize);
+ xsize = msize;
+ }
+
+ tp = rp;
+ rp = xp;
+ xp = tp;
+ rsize = xsize;
+
+ if ((mpi_limb_signed_t) e < 0) {
+ /*mpihelp_mul( xp, rp, rsize, bp, bsize ); */
+ if (bsize < KARATSUBA_THRESHOLD) {
+ mpi_limb_t tmp;
+ if (mpihelp_mul
+ (xp, rp, rsize, bp, bsize,
+ &tmp) < 0)
+ goto enomem;
+ } else {
+ if (mpihelp_mul_karatsuba_case
+ (xp, rp, rsize, bp, bsize,
+ &karactx) < 0)
+ goto enomem;
+ }
+
+ xsize = rsize + bsize;
+ if (xsize > msize) {
+ mpihelp_divrem(xp + msize, 0,
+ xp, xsize, mp,
+ msize);
+ xsize = msize;
+ }
+
+ tp = rp;
+ rp = xp;
+ xp = tp;
+ rsize = xsize;
+ }
+ e <<= 1;
+ c--;
+ }
+
+ i--;
+ if (i < 0)
+ break;
+ e = ep[i];
+ c = BITS_PER_MPI_LIMB;
+ }
+
+ /* We shifted MOD, the modulo reduction argument, left MOD_SHIFT_CNT
+ * steps. Adjust the result by reducing it with the original MOD.
+ *
+ * Also make sure the result is put in RES->d (where it already
+ * might be, see above).
+ */
+ if (mod_shift_cnt) {
+ carry_limb =
+ mpihelp_lshift(res->d, rp, rsize, mod_shift_cnt);
+ rp = res->d;
+ if (carry_limb) {
+ rp[rsize] = carry_limb;
+ rsize++;
+ }
+ } else {
+ MPN_COPY(res->d, rp, rsize);
+ rp = res->d;
+ }
+
+ if (rsize >= msize) {
+ mpihelp_divrem(rp + msize, 0, rp, rsize, mp, msize);
+ rsize = msize;
+ }
+
+ /* Remove any leading zero words from the result. */
+ if (mod_shift_cnt)
+ mpihelp_rshift(rp, rp, rsize, mod_shift_cnt);
+ MPN_NORMALIZE(rp, rsize);
+
+ mpihelp_release_karatsuba_ctx(&karactx);
+ }
+
+ if (negative_result && rsize) {
+ if (mod_shift_cnt)
+ mpihelp_rshift(mp, mp, msize, mod_shift_cnt);
+ mpihelp_sub(rp, mp, msize, rp, rsize);
+ rsize = msize;
+ rsign = msign;
+ MPN_NORMALIZE(rp, rsize);
+ }
+ res->nlimbs = rsize;
+ res->sign = rsign;
+
+ leave:
+ rc = 0;
+ enomem:
+ if (assign_rp)
+ mpi_assign_limb_space(res, rp, size);
+ if (mp_marker)
+ mpi_free_limb_space(mp_marker);
+ if (bp_marker)
+ mpi_free_limb_space(bp_marker);
+ if (ep_marker)
+ mpi_free_limb_space(ep_marker);
+ if (xp_marker)
+ mpi_free_limb_space(xp_marker);
+ if (tspace)
+ mpi_free_limb_space(tspace);
+ return rc;
+}
+
+/* Multiply the natural numbers u (pointed to by UP) and v (pointed to by VP),
+ * both with SIZE limbs, and store the result at PRODP. 2 * SIZE limbs are
+ * always stored. Return the most significant limb.
+ *
+ * Argument constraints:
+ * 1. PRODP != UP and PRODP != VP, i.e. the destination
+ * must be distinct from the multiplier and the multiplicand.
+ *
+ *
+ * Handle simple cases with traditional multiplication.
+ *
+ * This is the most critical code of multiplication. All multiplies rely
+ * on this, both small and huge. Small ones arrive here immediately. Huge
+ * ones arrive here as this is the base case for Karatsuba's recursive
+ * algorithm below.
+ */
+
+static mpi_limb_t
+mul_n_basecase(mpi_ptr_t prodp, mpi_ptr_t up, mpi_ptr_t vp, mpi_size_t size)
+{
+ mpi_size_t i;
+ mpi_limb_t cy;
+ mpi_limb_t v_limb;
+
+ /* Multiply by the first limb in V separately, as the result can be
+ * stored (not added) to PROD. We also avoid a loop for zeroing. */
+ v_limb = vp[0];
+ if (v_limb <= 1) {
+ if (v_limb == 1)
+ MPN_COPY(prodp, up, size);
+ else
+ MPN_ZERO(prodp, size);
+ cy = 0;
+ } else
+ cy = mpihelp_mul_1(prodp, up, size, v_limb);
+
+ prodp[size] = cy;
+ prodp++;
+
+ /* For each iteration in the outer loop, multiply one limb from
+ * U with one limb from V, and add it to PROD. */
+ for (i = 1; i < size; i++) {
+ v_limb = vp[i];
+ if (v_limb <= 1) {
+ cy = 0;
+ if (v_limb == 1)
+ cy = mpihelp_add_n(prodp, prodp, up, size);
+ } else
+ cy = mpihelp_addmul_1(prodp, up, size, v_limb);
+
+ prodp[size] = cy;
+ prodp++;
+ }
+
+ return cy;
+}
+
+static void
+mul_n(mpi_ptr_t prodp, mpi_ptr_t up, mpi_ptr_t vp,
+ mpi_size_t size, mpi_ptr_t tspace)
+{
+ if (size & 1) {
+ /* The size is odd, and the code below doesn't handle that.
+ * Multiply the least significant (size - 1) limbs with a recursive
+ * call, and handle the most significant limb of S1 and S2
+ * separately.
+ * A slightly faster way to do this would be to make the Karatsuba
+ * code below behave as if the size were even, and let it check for
+ * odd size in the end. I.e., in essence move this code to the end.
+ * Doing so would save us a recursive call, and potentially make the
+ * stack grow a lot less.
+ */
+ mpi_size_t esize = size - 1; /* even size */
+ mpi_limb_t cy_limb;
+
+ MPN_MUL_N_RECURSE(prodp, up, vp, esize, tspace);
+ cy_limb = mpihelp_addmul_1(prodp + esize, up, esize, vp[esize]);
+ prodp[esize + esize] = cy_limb;
+ cy_limb = mpihelp_addmul_1(prodp + esize, vp, size, up[esize]);
+ prodp[esize + size] = cy_limb;
+ } else {
+ /* Anatolij Alekseevich Karatsuba's divide-and-conquer algorithm.
+ *
+ * Split U in two pieces, U1 and U0, such that
+ * U = U0 + U1*(B**n),
+ * and V in V1 and V0, such that
+ * V = V0 + V1*(B**n).
+ *
+ * UV is then computed recursively using the identity
+ *
+ * 2n n n n
+ * UV = (B + B )U V + B (U -U )(V -V ) + (B + 1)U V
+ * 1 1 1 0 0 1 0 0
+ *
+ * Where B = 2**BITS_PER_MP_LIMB.
+ */
+ mpi_size_t hsize = size >> 1;
+ mpi_limb_t cy;
+ int negflg;
+
+ /* Product H. ________________ ________________
+ * |_____U1 x V1____||____U0 x V0_____|
+ * Put result in upper part of PROD and pass low part of TSPACE
+ * as new TSPACE.
+ */
+ MPN_MUL_N_RECURSE(prodp + size, up + hsize, vp + hsize, hsize,
+ tspace);
+
+ /* Product M. ________________
+ * |_(U1-U0)(V0-V1)_|
+ */
+ if (mpihelp_cmp(up + hsize, up, hsize) >= 0) {
+ mpihelp_sub_n(prodp, up + hsize, up, hsize);
+ negflg = 0;
+ } else {
+ mpihelp_sub_n(prodp, up, up + hsize, hsize);
+ negflg = 1;
+ }
+ if (mpihelp_cmp(vp + hsize, vp, hsize) >= 0) {
+ mpihelp_sub_n(prodp + hsize, vp + hsize, vp, hsize);
+ negflg ^= 1;
+ } else {
+ mpihelp_sub_n(prodp + hsize, vp, vp + hsize, hsize);
+ /* No change of NEGFLG. */
+ }
+ /* Read temporary operands from low part of PROD.
+ * Put result in low part of TSPACE using upper part of TSPACE
+ * as new TSPACE.
+ */
+ MPN_MUL_N_RECURSE(tspace, prodp, prodp + hsize, hsize,
+ tspace + size);
+
+ /* Add/copy product H. */
+ MPN_COPY(prodp + hsize, prodp + size, hsize);
+ cy = mpihelp_add_n(prodp + size, prodp + size,
+ prodp + size + hsize, hsize);
+
+ /* Add product M (if NEGFLG M is a negative number) */
+ if (negflg)
+ cy -=
+ mpihelp_sub_n(prodp + hsize, prodp + hsize, tspace,
+ size);
+ else
+ cy +=
+ mpihelp_add_n(prodp + hsize, prodp + hsize, tspace,
+ size);
+
+ /* Product L. ________________ ________________
+ * |________________||____U0 x V0_____|
+ * Read temporary operands from low part of PROD.
+ * Put result in low part of TSPACE using upper part of TSPACE
+ * as new TSPACE.
+ */
+ MPN_MUL_N_RECURSE(tspace, up, vp, hsize, tspace + size);
+
+ /* Add/copy Product L (twice) */
+
+ cy += mpihelp_add_n(prodp + hsize, prodp + hsize, tspace, size);
+ if (cy)
+ mpihelp_add_1(prodp + hsize + size,
+ prodp + hsize + size, hsize, cy);
+
+ MPN_COPY(prodp, tspace, hsize);
+ cy = mpihelp_add_n(prodp + hsize, prodp + hsize, tspace + hsize,
+ hsize);
+ if (cy)
+ mpihelp_add_1(prodp + size, prodp + size, size, 1);
+ }
+}
+
+static void mpih_sqr_n_basecase(mpi_ptr_t prodp, mpi_ptr_t up, mpi_size_t size)
+{
+ mpi_size_t i;
+ mpi_limb_t cy_limb;
+ mpi_limb_t v_limb;
+
+ /* Multiply by the first limb in V separately, as the result can be
+ * stored (not added) to PROD. We also avoid a loop for zeroing. */
+ v_limb = up[0];
+ if (v_limb <= 1) {
+ if (v_limb == 1)
+ MPN_COPY(prodp, up, size);
+ else
+ MPN_ZERO(prodp, size);
+ cy_limb = 0;
+ } else
+ cy_limb = mpihelp_mul_1(prodp, up, size, v_limb);
+
+ prodp[size] = cy_limb;
+ prodp++;
+
+ /* For each iteration in the outer loop, multiply one limb from
+ * U with one limb from V, and add it to PROD. */
+ for (i = 1; i < size; i++) {
+ v_limb = up[i];
+ if (v_limb <= 1) {
+ cy_limb = 0;
+ if (v_limb == 1)
+ cy_limb = mpihelp_add_n(prodp, prodp, up, size);
+ } else
+ cy_limb = mpihelp_addmul_1(prodp, up, size, v_limb);
+
+ prodp[size] = cy_limb;
+ prodp++;
+ }
+}
+
+static void
+mpih_sqr_n(mpi_ptr_t prodp, mpi_ptr_t up, mpi_size_t size, mpi_ptr_t tspace)
+{
+ if (size & 1) {
+ /* The size is odd, and the code below doesn't handle that.
+ * Multiply the least significant (size - 1) limbs with a recursive
+ * call, and handle the most significant limb of S1 and S2
+ * separately.
+ * A slightly faster way to do this would be to make the Karatsuba
+ * code below behave as if the size were even, and let it check for
+ * odd size in the end. I.e., in essence move this code to the end.
+ * Doing so would save us a recursive call, and potentially make the
+ * stack grow a lot less.
+ */
+ mpi_size_t esize = size - 1; /* even size */
+ mpi_limb_t cy_limb;
+
+ MPN_SQR_N_RECURSE(prodp, up, esize, tspace);
+ cy_limb = mpihelp_addmul_1(prodp + esize, up, esize, up[esize]);
+ prodp[esize + esize] = cy_limb;
+ cy_limb = mpihelp_addmul_1(prodp + esize, up, size, up[esize]);
+
+ prodp[esize + size] = cy_limb;
+ } else {
+ mpi_size_t hsize = size >> 1;
+ mpi_limb_t cy;
+
+ /* Product H. ________________ ________________
+ * |_____U1 x U1____||____U0 x U0_____|
+ * Put result in upper part of PROD and pass low part of TSPACE
+ * as new TSPACE.
+ */
+ MPN_SQR_N_RECURSE(prodp + size, up + hsize, hsize, tspace);
+
+ /* Product M. ________________
+ * |_(U1-U0)(U0-U1)_|
+ */
+ if (mpihelp_cmp(up + hsize, up, hsize) >= 0)
+ mpihelp_sub_n(prodp, up + hsize, up, hsize);
+ else
+ mpihelp_sub_n(prodp, up, up + hsize, hsize);
+
+ /* Read temporary operands from low part of PROD.
+ * Put result in low part of TSPACE using upper part of TSPACE
+ * as new TSPACE. */
+ MPN_SQR_N_RECURSE(tspace, prodp, hsize, tspace + size);
+
+ /* Add/copy product H */
+ MPN_COPY(prodp + hsize, prodp + size, hsize);
+ cy = mpihelp_add_n(prodp + size, prodp + size,
+ prodp + size + hsize, hsize);
+
+ /* Add product M (if NEGFLG M is a negative number). */
+ cy -= mpihelp_sub_n(prodp + hsize, prodp + hsize, tspace, size);
+
+ /* Product L. ________________ ________________
+ * |________________||____U0 x U0_____|
+ * Read temporary operands from low part of PROD.
+ * Put result in low part of TSPACE using upper part of TSPACE
+ * as new TSPACE. */
+ MPN_SQR_N_RECURSE(tspace, up, hsize, tspace + size);
+
+ /* Add/copy Product L (twice). */
+ cy += mpihelp_add_n(prodp + hsize, prodp + hsize, tspace, size);
+ if (cy)
+ mpihelp_add_1(prodp + hsize + size,
+ prodp + hsize + size, hsize, cy);
+
+ MPN_COPY(prodp, tspace, hsize);
+ cy = mpihelp_add_n(prodp + hsize, prodp + hsize, tspace + hsize,
+ hsize);
+ if (cy)
+ mpihelp_add_1(prodp + size, prodp + size, size, 1);
+ }
+}
+
+static int
+mpihelp_mul_karatsuba_case(mpi_ptr_t prodp,
+ mpi_ptr_t up, mpi_size_t usize,
+ mpi_ptr_t vp, mpi_size_t vsize,
+ struct karatsuba_ctx *ctx)
+{
+ mpi_limb_t cy;
+
+ if (!ctx->tspace || ctx->tspace_size < vsize) {
+ if (ctx->tspace)
+ mpi_free_limb_space(ctx->tspace);
+ ctx->tspace = mpi_alloc_limb_space(2 * vsize);
+ if (!ctx->tspace)
+ return -ENOMEM;
+ ctx->tspace_size = vsize;
+ }
+
+ MPN_MUL_N_RECURSE(prodp, up, vp, vsize, ctx->tspace);
+
+ prodp += vsize;
+ up += vsize;
+ usize -= vsize;
+ if (usize >= vsize) {
+ if (!ctx->tp || ctx->tp_size < vsize) {
+ if (ctx->tp)
+ mpi_free_limb_space(ctx->tp);
+ ctx->tp = mpi_alloc_limb_space(2 * vsize);
+ if (!ctx->tp) {
+ if (ctx->tspace)
+ mpi_free_limb_space(ctx->tspace);
+ ctx->tspace = NULL;
+ return -ENOMEM;
+ }
+ ctx->tp_size = vsize;
+ }
+
+ do {
+ MPN_MUL_N_RECURSE(ctx->tp, up, vp, vsize, ctx->tspace);
+ cy = mpihelp_add_n(prodp, prodp, ctx->tp, vsize);
+ mpihelp_add_1(prodp + vsize, ctx->tp + vsize, vsize,
+ cy);
+ prodp += vsize;
+ up += vsize;
+ usize -= vsize;
+ } while (usize >= vsize);
+ }
+
+ if (usize) {
+ if (usize < KARATSUBA_THRESHOLD) {
+ mpi_limb_t tmp;
+ if (mpihelp_mul(ctx->tspace, vp, vsize, up, usize, &tmp)
+ < 0)
+ return -ENOMEM;
+ } else {
+ if (!ctx->next) {
+ ctx->next = xzalloc(struct karatsuba_ctx);
+ if (!ctx->next)
+ return -ENOMEM;
+ }
+ if (mpihelp_mul_karatsuba_case(ctx->tspace,
+ vp, vsize,
+ up, usize,
+ ctx->next) < 0)
+ return -ENOMEM;
+ }
+
+ cy = mpihelp_add_n(prodp, prodp, ctx->tspace, vsize);
+ mpihelp_add_1(prodp + vsize, ctx->tspace + vsize, usize, cy);
+ }
+
+ return 0;
+}
+
+static void mpihelp_release_karatsuba_ctx(struct karatsuba_ctx *ctx)
+{
+ struct karatsuba_ctx *ctx2;
+
+ if (ctx->tp)
+ mpi_free_limb_space(ctx->tp);
+ if (ctx->tspace)
+ mpi_free_limb_space(ctx->tspace);
+ for (ctx = ctx->next; ctx; ctx = ctx2) {
+ ctx2 = ctx->next;
+ if (ctx->tp)
+ mpi_free_limb_space(ctx->tp);
+ if (ctx->tspace)
+ mpi_free_limb_space(ctx->tspace);
+ xfree(ctx);
+ }
+}
+
+/* Multiply the natural numbers u (pointed to by UP, with USIZE limbs)
+ * and v (pointed to by VP, with VSIZE limbs), and store the result at
+ * PRODP. USIZE + VSIZE limbs are always stored, but if the input
+ * operands are normalized. Return the most significant limb of the
+ * result.
+ *
+ * NOTE: The space pointed to by PRODP is overwritten before finished
+ * with U and V, so overlap is an error.
+ *
+ * Argument constraints:
+ * 1. USIZE >= VSIZE.
+ * 2. PRODP != UP and PRODP != VP, i.e. the destination
+ * must be distinct from the multiplier and the multiplicand.
+ */
+
+static int
+mpihelp_mul(mpi_ptr_t prodp, mpi_ptr_t up, mpi_size_t usize,
+ mpi_ptr_t vp, mpi_size_t vsize, mpi_limb_t *_result)
+{
+ mpi_ptr_t prod_endp = prodp + usize + vsize - 1;
+ mpi_limb_t cy;
+ struct karatsuba_ctx ctx;
+
+ if (vsize < KARATSUBA_THRESHOLD) {
+ mpi_size_t i;
+ mpi_limb_t v_limb;
+
+ if (!vsize) {
+ *_result = 0;
+ return 0;
+ }
+
+ /* Multiply by the first limb in V separately, as the result can be
+ * stored (not added) to PROD. We also avoid a loop for zeroing. */
+ v_limb = vp[0];
+ if (v_limb <= 1) {
+ if (v_limb == 1)
+ MPN_COPY(prodp, up, usize);
+ else
+ MPN_ZERO(prodp, usize);
+ cy = 0;
+ } else
+ cy = mpihelp_mul_1(prodp, up, usize, v_limb);
+
+ prodp[usize] = cy;
+ prodp++;
+
+ /* For each iteration in the outer loop, multiply one limb from
+ * U with one limb from V, and add it to PROD. */
+ for (i = 1; i < vsize; i++) {
+ v_limb = vp[i];
+ if (v_limb <= 1) {
+ cy = 0;
+ if (v_limb == 1)
+ cy = mpihelp_add_n(prodp, prodp, up,
+ usize);
+ } else
+ cy = mpihelp_addmul_1(prodp, up, usize, v_limb);
+
+ prodp[usize] = cy;
+ prodp++;
+ }
+
+ *_result = cy;
+ return 0;
+ }
+
+ memset(&ctx, 0, sizeof ctx);
+ if (mpihelp_mul_karatsuba_case(prodp, up, usize, vp, vsize, &ctx) < 0)
+ return -ENOMEM;
+ mpihelp_release_karatsuba_ctx(&ctx);
+ *_result = *prod_endp;
+ return 0;
+}
+
+static mpi_limb_t
+mpihelp_mul_1(mpi_ptr_t res_ptr, mpi_ptr_t s1_ptr, mpi_size_t s1_size,
+ mpi_limb_t s2_limb)
+{
+ mpi_limb_t cy_limb;
+ mpi_size_t j;
+ mpi_limb_t prod_high, prod_low;
+
+ /* The loop counter and index J goes from -S1_SIZE to -1. This way
+ * the loop becomes faster. */
+ j = -s1_size;
+
+ /* Offset the base pointers to compensate for the negative indices. */
+ s1_ptr -= j;
+ res_ptr -= j;
+
+ cy_limb = 0;
+ do {
+ umul_ppmm(prod_high, prod_low, s1_ptr[j], s2_limb);
+ prod_low += cy_limb;
+ cy_limb = (prod_low < cy_limb ? 1 : 0) + prod_high;
+ res_ptr[j] = prod_low;
+ } while (++j);
+
+ return cy_limb;
+}
+
+static mpi_limb_t
+mpihelp_add_n(mpi_ptr_t res_ptr, mpi_ptr_t s1_ptr,
+ mpi_ptr_t s2_ptr, mpi_size_t size)
+{
+ mpi_limb_t x, y, cy;
+ mpi_size_t j;
+
+ /* The loop counter and index J goes from -SIZE to -1. This way
+ the loop becomes faster. */
+ j = -size;
+
+ /* Offset the base pointers to compensate for the negative indices. */
+ s1_ptr -= j;
+ s2_ptr -= j;
+ res_ptr -= j;
+
+ cy = 0;
+ do {
+ y = s2_ptr[j];
+ x = s1_ptr[j];
+ y += cy; /* add previous carry to one addend */
+ cy = y < cy; /* get out carry from that addition */
+ y += x; /* add other addend */
+ cy += y < x; /* get out carry from that add, combine */
+ res_ptr[j] = y;
+ } while (++j);
+
+ return cy;
+}
+
+/* Shift U (pointed to by UP and USIZE digits long) CNT bits to the left
+ * and store the USIZE least significant digits of the result at WP.
+ * Return the bits shifted out from the most significant digit.
+ *
+ * Argument constraints:
+ * 1. 0 < CNT < BITS_PER_MP_LIMB
+ * 2. If the result is to be written over the input, WP must be >= UP.
+ */
+
+static mpi_limb_t
+mpihelp_lshift(mpi_ptr_t wp, mpi_ptr_t up, mpi_size_t usize, unsigned int cnt)
+{
+ mpi_limb_t high_limb, low_limb;
+ unsigned sh_1, sh_2;
+ mpi_size_t i;
+ mpi_limb_t retval;
+
+ sh_1 = cnt;
+ wp += 1;
+ sh_2 = BITS_PER_MPI_LIMB - sh_1;
+ i = usize - 1;
+ low_limb = up[i];
+ retval = low_limb >> sh_2;
+ high_limb = low_limb;
+ while (--i >= 0) {
+ low_limb = up[i];
+ wp[i] = (high_limb << sh_1) | (low_limb >> sh_2);
+ high_limb = low_limb;
+ }
+ wp[i] = high_limb << sh_1;
+
+ return retval;
+}
+
+static mpi_limb_t
+mpihelp_addmul_1(mpi_ptr_t res_ptr, mpi_ptr_t s1_ptr,
+ mpi_size_t s1_size, mpi_limb_t s2_limb)
+{
+ mpi_limb_t cy_limb;
+ mpi_size_t j;
+ mpi_limb_t prod_high, prod_low;
+ mpi_limb_t x;
+
+ /* The loop counter and index J goes from -SIZE to -1. This way
+ * the loop becomes faster. */
+ j = -s1_size;
+ res_ptr -= j;
+ s1_ptr -= j;
+
+ cy_limb = 0;
+ do {
+ umul_ppmm(prod_high, prod_low, s1_ptr[j], s2_limb);
+
+ prod_low += cy_limb;
+ cy_limb = (prod_low < cy_limb ? 1 : 0) + prod_high;
+
+ x = res_ptr[j];
+ prod_low = x + prod_low;
+ cy_limb += prod_low < x ? 1 : 0;
+ res_ptr[j] = prod_low;
+ } while (++j);
+ return cy_limb;
+}
+
+static mpi_limb_t
+mpihelp_sub_n(mpi_ptr_t res_ptr, mpi_ptr_t s1_ptr,
+ mpi_ptr_t s2_ptr, mpi_size_t size)
+{
+ mpi_limb_t x, y, cy;
+ mpi_size_t j;
+
+ /* The loop counter and index J goes from -SIZE to -1. This way
+ the loop becomes faster. */
+ j = -size;
+
+ /* Offset the base pointers to compensate for the negative indices. */
+ s1_ptr -= j;
+ s2_ptr -= j;
+ res_ptr -= j;
+
+ cy = 0;
+ do {
+ y = s2_ptr[j];
+ x = s1_ptr[j];
+ y += cy; /* add previous carry to subtrahend */
+ cy = y < cy; /* get out carry from that addition */
+ y = x - y; /* main subtract */
+ cy += y > x; /* get out carry from the subtract, combine */
+ res_ptr[j] = y;
+ } while (++j);
+
+ return cy;
+}
+
+/****************
+ * Compare OP1_PTR/OP1_SIZE with OP2_PTR/OP2_SIZE.
+ * There are no restrictions on the relative sizes of
+ * the two arguments.
+ * Return 1 if OP1 > OP2, 0 if they are equal, and -1 if OP1 < OP2.
+ */
+static int mpihelp_cmp(mpi_ptr_t op1_ptr, mpi_ptr_t op2_ptr, mpi_size_t size)
+{
+ mpi_size_t i;
+ mpi_limb_t op1_word, op2_word;
+
+ for (i = size - 1; i >= 0; i--) {
+ op1_word = op1_ptr[i];
+ op2_word = op2_ptr[i];
+ if (op1_word != op2_word)
+ goto diff;
+ }
+ return 0;
+
+ diff:
+ /* This can *not* be simplified to
+ * op2_word - op2_word
+ * since that expression might give signed overflow. */
+ return (op1_word > op2_word) ? 1 : -1;
+}
+
+static void mpi_assign_limb_space(MPI a, mpi_ptr_t ap, unsigned nlimbs)
+{
+ mpi_free_limb_space(a->d);
+ a->d = ap;
+ a->alloced = nlimbs;
+}
+
+/* Shift U (pointed to by UP and USIZE limbs long) CNT bits to the right
+ * and store the USIZE least significant limbs of the result at WP.
+ * The bits shifted out to the right are returned.
+ *
+ * Argument constraints:
+ * 1. 0 < CNT < BITS_PER_MP_LIMB
+ * 2. If the result is to be written over the input, WP must be <= UP.
+ */
+
+static mpi_limb_t
+mpihelp_rshift(mpi_ptr_t wp, mpi_ptr_t up, mpi_size_t usize, unsigned cnt)
+{
+ mpi_limb_t high_limb, low_limb;
+ unsigned sh_1, sh_2;
+ mpi_size_t i;
+ mpi_limb_t retval;
+
+ sh_1 = cnt;
+ wp -= 1;
+ sh_2 = BITS_PER_MPI_LIMB - sh_1;
+ high_limb = up[0];
+ retval = high_limb << sh_2;
+ low_limb = high_limb;
+ for (i = 1; i < usize; i++) {
+ high_limb = up[i];
+ wp[i] = (low_limb >> sh_1) | (high_limb << sh_2);
+ low_limb = high_limb;
+ }
+ wp[i] = low_limb >> sh_1;
+
+ return retval;
+}
+
+/* Divide num (NP/NSIZE) by den (DP/DSIZE) and write
+ * the NSIZE-DSIZE least significant quotient limbs at QP
+ * and the DSIZE long remainder at NP. If QEXTRA_LIMBS is
+ * non-zero, generate that many fraction bits and append them after the
+ * other quotient limbs.
+ * Return the most significant limb of the quotient, this is always 0 or 1.
+ *
+ * Preconditions:
+ * 0. NSIZE >= DSIZE.
+ * 1. The most significant bit of the divisor must be set.
+ * 2. QP must either not overlap with the input operands at all, or
+ * QP + DSIZE >= NP must hold true. (This means that it's
+ * possible to put the quotient in the high part of NUM, right after the
+ * remainder in NUM.
+ * 3. NSIZE >= DSIZE, even if QEXTRA_LIMBS is non-zero.
+ */
+
+static mpi_limb_t
+mpihelp_divrem(mpi_ptr_t qp, mpi_size_t qextra_limbs,
+ mpi_ptr_t np, mpi_size_t nsize, mpi_ptr_t dp, mpi_size_t dsize)
+{
+ mpi_limb_t most_significant_q_limb = 0;
+
+ switch (dsize) {
+ case 0:
+ /* We are asked to divide by zero, so go ahead and do it! (To make
+ the compiler not remove this statement, return the value.) */
+ /*
+ * existing clients of this function have been modified
+ * not to call it with dsize == 0, so this should not happen
+ */
+ return 1 / dsize;
+
+ case 1:
+ {
+ mpi_size_t i;
+ mpi_limb_t n1;
+ mpi_limb_t d;
+
+ d = dp[0];
+ n1 = np[nsize - 1];
+
+ if (n1 >= d) {
+ n1 -= d;
+ most_significant_q_limb = 1;
+ }
+
+ qp += qextra_limbs;
+ for (i = nsize - 2; i >= 0; i--)
+ udiv_qrnnd(qp[i], n1, n1, np[i], d);
+ qp -= qextra_limbs;
+
+ for (i = qextra_limbs - 1; i >= 0; i--)
+ udiv_qrnnd(qp[i], n1, n1, 0, d);
+
+ np[0] = n1;
+ }
+ break;
+
+ case 2:
+ {
+ mpi_size_t i;
+ mpi_limb_t n1, n0, n2;
+ mpi_limb_t d1, d0;
+
+ np += nsize - 2;
+ d1 = dp[1];
+ d0 = dp[0];
+ n1 = np[1];
+ n0 = np[0];
+
+ if (n1 >= d1 && (n1 > d1 || n0 >= d0)) {
+ sub_ddmmss(n1, n0, n1, n0, d1, d0);
+ most_significant_q_limb = 1;
+ }
+
+ for (i = qextra_limbs + nsize - 2 - 1; i >= 0; i--) {
+ mpi_limb_t q;
+ mpi_limb_t r;
+
+ if (i >= qextra_limbs)
+ np--;
+ else
+ np[0] = 0;
+
+ if (n1 == d1) {
+ /* Q should be either 111..111 or 111..110. Need special
+ * treatment of this rare case as normal division would
+ * give overflow. */
+ q = ~(mpi_limb_t) 0;
+
+ r = n0 + d1;
+ if (r < d1) { /* Carry in the addition? */
+ add_ssaaaa(n1, n0, r - d0,
+ np[0], 0, d0);
+ qp[i] = q;
+ continue;
+ }
+ n1 = d0 - (d0 != 0 ? 1 : 0);
+ n0 = -d0;
+ } else {
+ udiv_qrnnd(q, r, n1, n0, d1);
+ umul_ppmm(n1, n0, d0, q);
+ }
+
+ n2 = np[0];
+ q_test:
+ if (n1 > r || (n1 == r && n0 > n2)) {
+ /* The estimated Q was too large. */
+ q--;
+ sub_ddmmss(n1, n0, n1, n0, 0, d0);
+ r += d1;
+ if (r >= d1) /* If not carry, test Q again. */
+ goto q_test;
+ }
+
+ qp[i] = q;
+ sub_ddmmss(n1, n0, r, n2, n1, n0);
+ }
+ np[1] = n1;
+ np[0] = n0;
+ }
+ break;
+
+ default:
+ {
+ mpi_size_t i;
+ mpi_limb_t dX, d1, n0;
+
+ np += nsize - dsize;
+ dX = dp[dsize - 1];
+ d1 = dp[dsize - 2];
+ n0 = np[dsize - 1];
+
+ if (n0 >= dX) {
+ if (n0 > dX
+ || mpihelp_cmp(np, dp, dsize - 1) >= 0) {
+ mpihelp_sub_n(np, np, dp, dsize);
+ n0 = np[dsize - 1];
+ most_significant_q_limb = 1;
+ }
+ }
+
+ for (i = qextra_limbs + nsize - dsize - 1; i >= 0; i--) {
+ mpi_limb_t q;
+ mpi_limb_t n1, n2;
+ mpi_limb_t cy_limb;
+
+ if (i >= qextra_limbs) {
+ np--;
+ n2 = np[dsize];
+ } else {
+ n2 = np[dsize - 1];
+ MPN_COPY_DECR(np + 1, np, dsize - 1);
+ np[0] = 0;
+ }
+
+ if (n0 == dX) {
+ /* This might over-estimate q, but it's probably not worth
+ * the extra code here to find out. */
+ q = ~(mpi_limb_t) 0;
+ } else {
+ mpi_limb_t r;
+
+ udiv_qrnnd(q, r, n0, np[dsize - 1], dX);
+ umul_ppmm(n1, n0, d1, q);
+
+ while (n1 > r
+ || (n1 == r
+ && n0 > np[dsize - 2])) {
+ q--;
+ r += dX;
+ if (r < dX) /* I.e. "carry in previous addition?" */
+ break;
+ n1 -= n0 < d1;
+ n0 -= d1;
+ }
+ }
+
+ /* Possible optimization: We already have (q * n0) and (1 * n1)
+ * after the calculation of q. Taking advantage of that, we
+ * could make this loop make two iterations less. */
+ cy_limb = mpihelp_submul_1(np, dp, dsize, q);
+
+ if (n2 != cy_limb) {
+ mpihelp_add_n(np, np, dp, dsize);
+ q--;
+ }
+
+ qp[i] = q;
+ n0 = np[dsize - 1];
+ }
+ }
+ break;
+ }
+
+ return most_significant_q_limb;
+}
+
+static mpi_limb_t
+mpihelp_submul_1(mpi_ptr_t res_ptr, mpi_ptr_t s1_ptr,
+ mpi_size_t s1_size, mpi_limb_t s2_limb)
+{
+ mpi_limb_t cy_limb;
+ mpi_size_t j;
+ mpi_limb_t prod_high, prod_low;
+ mpi_limb_t x;
+
+ /* The loop counter and index J goes from -SIZE to -1. This way
+ * the loop becomes faster. */
+ j = -s1_size;
+ res_ptr -= j;
+ s1_ptr -= j;
+
+ cy_limb = 0;
+ do {
+ umul_ppmm(prod_high, prod_low, s1_ptr[j], s2_limb);
+
+ prod_low += cy_limb;
+ cy_limb = (prod_low < cy_limb ? 1 : 0) + prod_high;
+
+ x = res_ptr[j];
+ prod_low = x - prod_low;
+ cy_limb += prod_low > x ? 1 : 0;
+ res_ptr[j] = prod_low;
+ } while (++j);
+
+ return cy_limb;
+}
+
+/**
+ * mpi_read_raw_data - Read a raw byte stream as a positive integer
+ * @xbuffer: The data to read
+ * @nbytes: The amount of data to read
+ */
+MPI mpi_read_raw_data(const void *xbuffer, size_t nbytes)
+{
+ const uint8_t *buffer = xbuffer;
+ int i, j;
+ unsigned nbits, nlimbs;
+ mpi_limb_t a;
+ MPI val = NULL;
+
+ while (nbytes > 0 && buffer[0] == 0) {
+ buffer++;
+ nbytes--;
+ }
+
+ nbits = nbytes * 8;
+ if (nbits > MAX_EXTERN_MPI_BITS) {
+ printk("MPI: mpi too large (%u bits)\n", nbits);
+ return NULL;
+ }
+ if (nbytes > 0)
+ nbits -= count_leading_zeros(buffer[0]) - (BITS_PER_LONG - 8);
+
+ nlimbs = DIV_ROUND_UP(nbytes, BYTES_PER_MPI_LIMB);
+ val = mpi_alloc(nlimbs);
+ if (!val)
+ return NULL;
+ val->nbits = nbits;
+ val->sign = 0;
+ val->nlimbs = nlimbs;
+
+ if (nbytes > 0) {
+ i = BYTES_PER_MPI_LIMB - nbytes % BYTES_PER_MPI_LIMB;
+ i %= BYTES_PER_MPI_LIMB;
+ for (j = nlimbs; j > 0; j--) {
+ a = 0;
+ for (; i < BYTES_PER_MPI_LIMB; i++) {
+ a <<= 8;
+ a |= *buffer++;
+ }
+ i = 0;
+ val->d[j - 1] = a;
+ }
+ }
+ return val;
+}
+
+/****************
+ * Note: It was a bad idea to use the number of limbs to allocate
+ * because on a alpha the limbs are large but we normally need
+ * integers of n bits - So we should chnage this to bits (or bytes).
+ *
+ * But mpi_alloc is used in a lot of places :-)
+ */
+MPI mpi_alloc(unsigned nlimbs)
+{
+ MPI a;
+
+ a = xmalloc(struct mpi);
+ if (!a)
+ return a;
+
+ if (nlimbs) {
+ a->d = mpi_alloc_limb_space(nlimbs);
+ if (!a->d) {
+ xfree(a);
+ return NULL;
+ }
+ } else {
+ a->d = NULL;
+ }
+
+ a->alloced = nlimbs;
+ a->nlimbs = 0;
+ a->sign = 0;
+ a->flags = 0;
+ a->nbits = 0;
+ return a;
+}
+
+void mpi_free(MPI a)
+{
+ if (!a)
+ return;
+
+ if (a->flags & MPI_FLAG_PTR_ALLOC)
+ xfree(a->d);
+ else
+ mpi_free_limb_space(a->d);
+
+ if (a->flags & ~MPI_FLAG_MASK)
+ printk("invalid flag value in mpi\n");
+ xfree(a);
+}
+
+int mpi_cmp_ui(MPI u, unsigned long v)
+{
+ mpi_limb_t limb = v;
+
+ mpi_normalize(u);
+ if (!u->nlimbs && !limb)
+ return 0;
+ if (u->sign)
+ return -1;
+ if (u->nlimbs > 1)
+ return 1;
+
+ if (u->d[0] == limb)
+ return 0;
+ else if (u->d[0] > limb)
+ return 1;
+ else
+ return -1;
+}
+
+int mpi_cmp(MPI u, MPI v)
+{
+ mpi_size_t usize, vsize;
+ int cmp;
+
+ mpi_normalize(u);
+ mpi_normalize(v);
+ usize = u->nlimbs;
+ vsize = v->nlimbs;
+ if (!u->sign && v->sign)
+ return 1;
+ if (u->sign && !v->sign)
+ return -1;
+ if (usize != vsize && !u->sign && !v->sign)
+ return usize - vsize;
+ if (usize != vsize && u->sign && v->sign)
+ return vsize - usize;
+ if (!usize)
+ return 0;
+ cmp = mpihelp_cmp(u->d, v->d, usize);
+ if (u->sign)
+ return -cmp;
+ return cmp;
+}
+
+/****************
+ * Sometimes we have MSL (most significant limbs) which are 0;
+ * this is for some reasons not good, so this function removes them.
+ */
+static void mpi_normalize(MPI a)
+{
+ for (; a->nlimbs && !a->d[a->nlimbs - 1]; a->nlimbs--)
+ ;
+}
+
+/****************
+ * Return the number of bits in A.
+ */
+unsigned mpi_get_nbits(MPI a)
+{
+ unsigned n;
+
+ mpi_normalize(a);
+
+ if (a->nlimbs) {
+ mpi_limb_t alimb = a->d[a->nlimbs - 1];
+ if (alimb)
+ n = count_leading_zeros(alimb);
+ else
+ n = BITS_PER_MPI_LIMB;
+ n = BITS_PER_MPI_LIMB - n + (a->nlimbs - 1) * BITS_PER_MPI_LIMB;
+ } else
+ n = 0;
+ return n;
+}
+
+int mpi_test_bit(MPI a, unsigned int n)
+{
+ unsigned int limbno, bitno;
+ mpi_limb_t limb;
+
+ limbno = n / BITS_PER_MPI_LIMB;
+ bitno = n % BITS_PER_MPI_LIMB;
+
+ if (limbno >= a->nlimbs)
+ return 0; /* too far left: this is a 0 */
+ limb = a->d[limbno];
+ return (limb & (((mpi_limb_t)1) << bitno))? 1: 0;
+}
--
2.49.0
^ permalink raw reply related [flat|nested] 20+ messages in thread
* [PATCH v3 4/5] livepatch: Load built-in key during boot
2025-06-02 13:36 [PATCH v3 0/5] LivePatch signing support Ross Lagerwall
` (2 preceding siblings ...)
2025-06-02 13:36 ` [PATCH v3 3/5] crypto: Add RSA support Ross Lagerwall
@ 2025-06-02 13:36 ` Ross Lagerwall
2025-06-05 11:17 ` Jan Beulich
2025-06-02 13:36 ` [PATCH v3 5/5] livepatch: Verify livepatch signatures Ross Lagerwall
4 siblings, 1 reply; 20+ messages in thread
From: Ross Lagerwall @ 2025-06-02 13:36 UTC (permalink / raw)
To: xen-devel; +Cc: Ross Lagerwall, Roger Pau Monné
Parse the raw data of the embedded RSA key into a form that can be later
used for verifying live patch signatures.
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
---
* Fix endianness of builtin key lengths
* Set builtin key to __ro_after_init
xen/common/livepatch.c | 36 ++++++++++++++++++++++++++++++++++++
1 file changed, 36 insertions(+)
diff --git a/xen/common/livepatch.c b/xen/common/livepatch.c
index 9a0df5363b59..92d1d342d872 100644
--- a/xen/common/livepatch.c
+++ b/xen/common/livepatch.c
@@ -11,6 +11,8 @@
#include <xen/lib.h>
#include <xen/list.h>
#include <xen/mm.h>
+#include <xen/mpi.h>
+#include <xen/rsa.h>
#include <xen/sched.h>
#include <xen/smp.h>
#include <xen/softirq.h>
@@ -73,6 +75,10 @@ static struct livepatch_work livepatch_work;
static DEFINE_PER_CPU(bool, work_to_do);
static DEFINE_PER_CPU(struct tasklet, livepatch_tasklet);
+#ifdef CONFIG_PAYLOAD_VERIFY
+static struct rsa_public_key __ro_after_init builtin_payload_key;
+#endif
+
static int get_name(const struct xen_livepatch_name *name, char *n)
{
if ( !name->size || name->size > XEN_LIVEPATCH_NAME_SIZE )
@@ -2300,6 +2306,31 @@ static void cf_check livepatch_printall(unsigned char key)
spin_unlock(&payload_lock);
}
+static int load_builtin_payload_key(void)
+{
+#ifdef CONFIG_PAYLOAD_VERIFY
+ const uint8_t *ptr;
+ uint32_t len;
+
+ rsa_public_key_init(&builtin_payload_key);
+
+ ptr = xen_livepatch_key_data;
+
+ memcpy(&len, ptr, sizeof(len));
+ ptr += sizeof(len);
+ builtin_payload_key.n = mpi_read_raw_data(ptr, le32_to_cpu(len));
+ ptr += len;
+
+ memcpy(&len, ptr, sizeof(len));
+ ptr += sizeof(len);
+ builtin_payload_key.e = mpi_read_raw_data(ptr, le32_to_cpu(len));
+
+ return rsa_public_key_prepare(&builtin_payload_key);
+#else
+ return 0;
+#endif
+}
+
static int cf_check cpu_callback(
struct notifier_block *nfb, unsigned long action, void *hcpu)
{
@@ -2318,6 +2349,11 @@ static struct notifier_block cpu_nfb = {
static int __init cf_check livepatch_init(void)
{
unsigned int cpu;
+ int err;
+
+ err = load_builtin_payload_key();
+ if ( err )
+ return err;
for_each_online_cpu ( cpu )
{
--
2.49.0
^ permalink raw reply related [flat|nested] 20+ messages in thread
* [PATCH v3 5/5] livepatch: Verify livepatch signatures
2025-06-02 13:36 [PATCH v3 0/5] LivePatch signing support Ross Lagerwall
` (3 preceding siblings ...)
2025-06-02 13:36 ` [PATCH v3 4/5] livepatch: Load built-in key during boot Ross Lagerwall
@ 2025-06-02 13:36 ` Ross Lagerwall
2025-06-05 11:52 ` Jan Beulich
2025-06-20 10:31 ` Roger Pau Monné
4 siblings, 2 replies; 20+ messages in thread
From: Ross Lagerwall @ 2025-06-02 13:36 UTC (permalink / raw)
To: xen-devel; +Cc: Jennifer Herbert, Roger Pau Monné, Ross Lagerwall
From: Jennifer Herbert <jennifer.herbert@cloud.com>
Verify livepatch signatures against the embedded public key in Xen.
Failing to verify does not prevent the livepatch from being loaded.
In future, this will be changed for certain cases (e.g. when Secure Boot
is enabled).
Signed-off-by: Jennifer Herbert <jennifer.herbert@cloud.com>
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
---
In v3:
* Minor style fixes
xen/common/livepatch.c | 103 ++++++++++++++++++++++++++++++++
xen/common/livepatch_elf.c | 55 +++++++++++++++++
xen/include/xen/livepatch.h | 10 ++++
xen/include/xen/livepatch_elf.h | 18 ++++++
4 files changed, 186 insertions(+)
diff --git a/xen/common/livepatch.c b/xen/common/livepatch.c
index 92d1d342d872..56a3d125483f 100644
--- a/xen/common/livepatch.c
+++ b/xen/common/livepatch.c
@@ -14,6 +14,7 @@
#include <xen/mpi.h>
#include <xen/rsa.h>
#include <xen/sched.h>
+#include <xen/sha2.h>
#include <xen/smp.h>
#include <xen/softirq.h>
#include <xen/spinlock.h>
@@ -525,6 +526,106 @@ static int check_xen_buildid(const struct livepatch_elf *elf)
return 0;
}
+#ifdef CONFIG_PAYLOAD_VERIFY
+static int check_rsa_sha256_signature(void *data, size_t datalen,
+ void *sig, uint32_t siglen)
+{
+ struct sha2_256_state hash;
+ MPI s;
+ int rc;
+
+ s = mpi_read_raw_data(sig, siglen);
+ if ( !s )
+ {
+ printk(XENLOG_ERR LIVEPATCH "Failed to mpi_read_raw_data\n");
+ return -ENOMEM;
+ }
+
+ sha2_256_init(&hash);
+ sha2_256_update(&hash, data, datalen);
+
+ rc = rsa_sha256_verify(&builtin_payload_key, &hash, s);
+ if ( rc )
+ printk(XENLOG_ERR LIVEPATCH "rsa_sha256_verify failed: %d\n", rc);
+
+ mpi_free(s);
+
+ return rc;
+}
+#endif
+
+static int check_signature(const struct livepatch_elf *elf, void *raw,
+ size_t size)
+{
+#ifdef CONFIG_PAYLOAD_VERIFY
+#define MAX_SIG_NOTE_SIZE 1024
+ static const char notename[] = "Xen";
+ void *sig;
+ livepatch_elf_note note;
+ int rc;
+
+ rc = livepatch_elf_note_by_names(elf, ELF_XEN_SIGNATURE, notename, -1,
+ ¬e);
+ if ( rc )
+ {
+ dprintk(XENLOG_DEBUG, LIVEPATCH "%s: Signature not present\n",
+ elf->name);
+ return rc;
+ }
+
+ /* We expect only one signature, find a second is an error! */
+ rc = livepatch_elf_next_note_by_name(notename, -1, ¬e);
+ if ( rc != -ENOENT )
+ {
+ if ( rc )
+ {
+ printk(XENLOG_ERR LIVEPATCH
+ "Error while checking for notes! err = %d\n", rc);
+ return rc;
+ }
+ else
+ {
+ printk(XENLOG_ERR LIVEPATCH
+ "Error, found second signature note! There can be only one!\n");
+ return -EINVAL;
+ }
+ }
+
+ if ( SIGNATURE_VERSION(note.type) != LIVEPATCH_SIGNATURE_VERSION ||
+ SIGNATURE_ALGORITHM(note.type) != SIGNATURE_ALGORITHM_RSA ||
+ SIGNATURE_HASH(note.type) != SIGNATURE_HASH_SHA256 )
+ {
+ printk(XENLOG_ERR LIVEPATCH
+ "Unsupported signature type: v:%u, a:%u, h:%u\n",
+ SIGNATURE_VERSION(note.type), SIGNATURE_ALGORITHM(note.type),
+ SIGNATURE_HASH(note.type));
+ return -EINVAL;
+ }
+
+ if ( note.size == 0 || note.size >= MAX_SIG_NOTE_SIZE )
+ {
+ printk(XENLOG_ERR LIVEPATCH "Invalid signature note size: %u\n",
+ note.size);
+ return -EINVAL;
+ }
+
+ sig = xmalloc_bytes(note.size);
+ if ( !sig )
+ return -ENOMEM;
+
+ memcpy(sig, note.data, note.size);
+
+ /* Remove signature from data, as can't be verified with it. */
+ memset((void *)note.data, 0, note.size);
+ rc = check_rsa_sha256_signature(raw, size, sig, note.size);
+
+ xfree(sig);
+ return rc;
+#else
+ return -EINVAL;
+#endif
+}
+
static int check_special_sections(const struct livepatch_elf *elf)
{
unsigned int i;
@@ -1162,6 +1263,8 @@ static int load_payload_data(struct payload *payload, void *raw, size_t len)
if ( rc )
goto out;
+ check_signature(&elf, raw, len);
+
rc = move_payload(payload, &elf);
if ( rc )
goto out;
diff --git a/xen/common/livepatch_elf.c b/xen/common/livepatch_elf.c
index 25ce1bd5a0ad..b27c4524611d 100644
--- a/xen/common/livepatch_elf.c
+++ b/xen/common/livepatch_elf.c
@@ -23,6 +23,61 @@ livepatch_elf_sec_by_name(const struct livepatch_elf *elf,
return NULL;
}
+int livepatch_elf_note_by_names(const struct livepatch_elf *elf,
+ const char *sec_name, const char *note_name,
+ const unsigned int type,
+ livepatch_elf_note *note)
+{
+ const struct livepatch_elf_sec *sec = livepatch_elf_sec_by_name(elf,
+ sec_name);
+ if ( !sec )
+ return -ENOENT;
+
+ note->end = sec->addr + sec->sec->sh_size;
+ note->next = sec->addr;
+
+ return livepatch_elf_next_note_by_name(note_name, type, note);
+}
+
+int livepatch_elf_next_note_by_name(const char *note_name,
+ const unsigned int type,
+ livepatch_elf_note *note)
+{
+ const Elf_Note *pkd_note = note->next;
+ size_t notenamelen = strlen(note_name) + 1;
+ size_t note_hd_size;
+ size_t note_size;
+ size_t remaining;
+
+ while ( (void *)pkd_note < note->end )
+ {
+
+ remaining = note->end - (void *)pkd_note;
+ if ( remaining < sizeof(livepatch_elf_note) )
+ return -EINVAL;
+
+ note_hd_size = sizeof(Elf_Note) + ((pkd_note->namesz + 3) & ~0x3);
+ note_size = note_hd_size + ((pkd_note->descsz + 3) & ~0x3);
+ if ( remaining < note_size )
+ return -EINVAL;
+
+ if ( notenamelen == pkd_note->namesz &&
+ !memcmp(note_name, (const void *) pkd_note + sizeof(Elf_Note),
+ notenamelen) &&
+ (type == -1 || pkd_note->type == type) )
+ {
+ note->size = pkd_note->descsz;
+ note->type = pkd_note->type;
+ note->data = (void *)pkd_note + note_hd_size;
+ note->next = (void *)pkd_note + note_size;
+ return 0;
+ }
+ pkd_note = (void *)pkd_note + note_size;
+ }
+
+ return -ENOENT;
+}
+
static int elf_verify_strtab(const struct livepatch_elf_sec *sec)
{
const Elf_Shdr *s;
diff --git a/xen/include/xen/livepatch.h b/xen/include/xen/livepatch.h
index 52f90cbed45b..12206ce3b2b8 100644
--- a/xen/include/xen/livepatch.h
+++ b/xen/include/xen/livepatch.h
@@ -38,6 +38,7 @@ struct xen_sysctl_livepatch_op;
#define ELF_LIVEPATCH_DEPENDS ".livepatch.depends"
#define ELF_LIVEPATCH_XEN_DEPENDS ".livepatch.xen_depends"
#define ELF_BUILD_ID_NOTE ".note.gnu.build-id"
+#define ELF_XEN_SIGNATURE ".note.Xen.signature"
#define ELF_LIVEPATCH_LOAD_HOOKS ".livepatch.hooks.load"
#define ELF_LIVEPATCH_UNLOAD_HOOKS ".livepatch.hooks.unload"
#define ELF_LIVEPATCH_PREAPPLY_HOOK ".livepatch.hooks.preapply"
@@ -49,6 +50,15 @@ struct xen_sysctl_livepatch_op;
/* Arbitrary limit for payload size and .bss section size. */
#define LIVEPATCH_MAX_SIZE MB(2)
+#define SIGNATURE_VERSION(type) ((type) & 0xffff)
+#define LIVEPATCH_SIGNATURE_VERSION 1
+
+#define SIGNATURE_ALGORITHM(type) (((type) >> 16) & 0xff)
+#define SIGNATURE_ALGORITHM_RSA 0
+
+#define SIGNATURE_HASH(type) (((type) >> 24) & 0xff)
+#define SIGNATURE_HASH_SHA256 0
+
struct livepatch_symbol {
const char *name;
unsigned long value;
diff --git a/xen/include/xen/livepatch_elf.h b/xen/include/xen/livepatch_elf.h
index 842111e14518..04611bac410e 100644
--- a/xen/include/xen/livepatch_elf.h
+++ b/xen/include/xen/livepatch_elf.h
@@ -39,6 +39,16 @@ struct livepatch_elf {
unsigned int symtab_idx;
};
+typedef struct {
+ uint32_t size; /* Note size */
+ uint32_t type; /* Note type */
+ const void *data; /* Pointer to note */
+
+ /* Private */
+ const Elf_Note *next;
+ const void *end;
+} livepatch_elf_note;
+
const struct livepatch_elf_sec *
livepatch_elf_sec_by_name(const struct livepatch_elf *elf,
const char *name);
@@ -48,6 +58,14 @@ void livepatch_elf_free(struct livepatch_elf *elf);
int livepatch_elf_resolve_symbols(struct livepatch_elf *elf);
int livepatch_elf_perform_relocs(struct livepatch_elf *elf);
+int livepatch_elf_note_by_names(const struct livepatch_elf *elf,
+ const char *sec_name, const char *note_name,
+ const unsigned int type,
+ livepatch_elf_note *note);
+int livepatch_elf_next_note_by_name(const char *note_name,
+ const unsigned int type,
+ livepatch_elf_note *note);
+
static inline bool livepatch_elf_ignore_section(const Elf_Shdr *sec)
{
return !(sec->sh_flags & SHF_ALLOC);
--
2.49.0
^ permalink raw reply related [flat|nested] 20+ messages in thread
* Re: [PATCH v3 2/5] livepatch: Embed public key in Xen
2025-06-02 13:36 ` [PATCH v3 2/5] livepatch: Embed public key in Xen Ross Lagerwall
@ 2025-06-05 11:02 ` Jan Beulich
2025-06-05 11:19 ` Jan Beulich
2025-06-20 9:39 ` Roger Pau Monné
2 siblings, 0 replies; 20+ messages in thread
From: Jan Beulich @ 2025-06-05 11:02 UTC (permalink / raw)
To: Ross Lagerwall, Stefano Stabellini, consulting@bugseng.com
Cc: Kevin Lampis, Andrew Cooper, Anthony PERARD, Michal Orzel,
Julien Grall, Roger Pau Monné, xen-devel
On 02.06.2025 15:36, Ross Lagerwall wrote:
> From: Kevin Lampis <kevin.lampis@cloud.com>
>
> Make it possible to embed a public key in Xen to be used when verifying
> live patch payloads. Inclusion of the public key is optional.
>
> To avoid needing to include a DER / X.509 parser in the hypervisor, the
> public key is unpacked at build time and included in a form that is
> convenient for the hypervisor to consume. This is different approach
> from that used by Linux which embeds the entire X.509 certificate and
> builds in a parser for it.
>
> A suitable key can be created using openssl:
>
> openssl req -x509 -newkey rsa:2048 -keyout priv.pem -out pub.pem \
> -sha256 -days 3650 -nodes \
> -subj "/C=XX/ST=StateName/L=CityName/O=CompanyName/OU=CompanySectionName/CN=CommonNameOrHostname"
> openssl x509 -inform PEM -in pub.pem -outform PEM -pubkey -nocert -out verify_key.pem
>
> Signed-off-by: Kevin Lampis <kevin.lampis@cloud.com>
> Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
> ---
>
> In v3:
>
> * Drop unnecessary condition in Makefile
> * Use dashes instead of underscores
> * Drop section placement annotation on declaration
> * Clarify endianness of embedded key
>
> xen/common/Kconfig | 18 +++++++++++++++++
> xen/crypto/Makefile | 11 ++++++++++
> xen/include/xen/livepatch.h | 5 +++++
> xen/tools/extract-key.py | 40 +++++++++++++++++++++++++++++++++++++
> 4 files changed, 74 insertions(+)
> create mode 100755 xen/tools/extract-key.py
>
> diff --git a/xen/common/Kconfig b/xen/common/Kconfig
> index 0951d4c2f286..74673078202a 100644
> --- a/xen/common/Kconfig
> +++ b/xen/common/Kconfig
> @@ -472,6 +472,24 @@ config LIVEPATCH
>
> If unsure, say Y.
>
> +config PAYLOAD_VERIFY
> + bool "Verify signed LivePatch payloads"
> + depends on LIVEPATCH
> + select CRYPTO
> + help
> + Verify signed LivePatch payloads using an RSA public key built
> + into the Xen hypervisor. Selecting this option requires a
> + public key in PEM format to be available for embedding during
> + the build.
> +
> +config PAYLOAD_VERIFY_KEY
This identifier to me mean "verify the key of payloads", when really ...
> + string "File name of public key used to verify payloads"
... you mean the key used for verification. To express this in the name,
how about PAYLOAD_VERIFICATION_KEY?
> + default "verify_key.pem"
The revlog says you replaced underscores by dashes.
> + depends on PAYLOAD_VERIFY
> + help
> + The file name of an RSA public key in PEM format to be used for
> + verifying signed LivePatch payloads.
Perhaps the fact that it needs to be an RSA key may also want expressing
in the Kconfig identifier?
Should the help text perhaps further clarify where the file is going to be
looked for in case it's a relative path (as in particular the default is)?
Seeing that ...
> --- a/xen/crypto/Makefile
> +++ b/xen/crypto/Makefile
> @@ -1,2 +1,13 @@
> obj-y += rijndael.o
> obj-y += vmac.o
> +
> +obj-$(CONFIG_PAYLOAD_VERIFY) += builtin-payload-key.o
> +
> +key-path := $(objtree)/$(patsubst "%",%,$(CONFIG_PAYLOAD_VERIFY_KEY))
... it's in the object tree now (which will break if an absolute path was
specified), how do you intend the key to make it there? In the default
case, for an out-of-tree build the build tree doesn't exist ahead of the
build being invoked.
> +$(obj)/builtin-payload-key.bin: $(key-path) $(srctree)/tools/extract-key.py
> + $(srctree)/tools/extract-key.py < $< > $@.new
> + $(call move-if-changed,$@.new,$@)
> +
> +$(obj)/builtin-payload-key.S: BINFILE_FLAGS := -i
> +$(obj)/builtin-payload-key.S: $(srctree)/tools/binfile $(obj)/builtin-payload-key.bin FORCE
> + $(call if_changed,binfile,$(obj)/builtin-payload-key.bin xen_livepatch_key_data)
Does the builtin- prefix have any real significance here? Anything that's
part of the hypervisor build is "built in", after all.
> --- a/xen/include/xen/livepatch.h
> +++ b/xen/include/xen/livepatch.h
> @@ -143,6 +143,11 @@ struct payload;
> int revert_payload(struct payload *data);
> void revert_payload_tail(struct payload *data);
>
> +#ifdef CONFIG_PAYLOAD_VERIFY
> +/* The public key data contained with Xen used to verify payload signatures. */
> +extern const uint8_t xen_livepatch_key_data[];
> +#endif
What's our position towards such declarations without (C) definition? The
8.6 deviation talks about cases "when the definition is compiled-out or
optimized-out by the compiler". Shouldn't we follow the 8.4 model here and
annotate them asmlinkage as well?
Jan
^ permalink raw reply [flat|nested] 20+ messages in thread
* Re: [PATCH v3 3/5] crypto: Add RSA support
2025-06-02 13:36 ` [PATCH v3 3/5] crypto: Add RSA support Ross Lagerwall
@ 2025-06-05 11:06 ` Jan Beulich
2025-06-20 9:53 ` Roger Pau Monné
1 sibling, 0 replies; 20+ messages in thread
From: Jan Beulich @ 2025-06-05 11:06 UTC (permalink / raw)
To: Ross Lagerwall
Cc: Andrew Cooper, Anthony PERARD, Michal Orzel, Julien Grall,
Roger Pau Monné, Stefano Stabellini, xen-devel
On 02.06.2025 15:36, Ross Lagerwall wrote:
> In preparation for adding support for livepatch signing, add support for
> RSA crypto.
>
> The RSA code is extracted from Nettle at tag nettle_3.2_release_20160128
> (https://git.lysator.liu.se/nettle/nettle).
>
> The MPI code is extracted from Linux at commit eef0df6a5953 (lib/mpi/*).
>
> Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Acked-by: Jan Beulich <jbeulich@suse.com>
This looks like it's independent of the earlier patches, and hence could go
in right away. Then again maybe doing so wouldn't make much sense prior to ...
> --- a/xen/crypto/Makefile
> +++ b/xen/crypto/Makefile
> @@ -1,4 +1,5 @@
> obj-y += rijndael.o
> +obj-$(CONFIG_PAYLOAD_VERIFY) += rsa.o
... this new Kconfig setting appearing (eve though the lack thereof wouldn't
break anything, afaict).
Jan
^ permalink raw reply [flat|nested] 20+ messages in thread
* Re: [PATCH v3 4/5] livepatch: Load built-in key during boot
2025-06-02 13:36 ` [PATCH v3 4/5] livepatch: Load built-in key during boot Ross Lagerwall
@ 2025-06-05 11:17 ` Jan Beulich
0 siblings, 0 replies; 20+ messages in thread
From: Jan Beulich @ 2025-06-05 11:17 UTC (permalink / raw)
To: Ross Lagerwall; +Cc: Roger Pau Monné, xen-devel
On 02.06.2025 15:36, Ross Lagerwall wrote:
> Parse the raw data of the embedded RSA key into a form that can be later
> used for verifying live patch signatures.
>
> Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
> ---
>
> * Fix endianness of builtin key lengths
> * Set builtin key to __ro_after_init
>
> xen/common/livepatch.c | 36 ++++++++++++++++++++++++++++++++++++
> 1 file changed, 36 insertions(+)
>
> diff --git a/xen/common/livepatch.c b/xen/common/livepatch.c
> index 9a0df5363b59..92d1d342d872 100644
> --- a/xen/common/livepatch.c
> +++ b/xen/common/livepatch.c
> @@ -11,6 +11,8 @@
> #include <xen/lib.h>
> #include <xen/list.h>
> #include <xen/mm.h>
> +#include <xen/mpi.h>
> +#include <xen/rsa.h>
> #include <xen/sched.h>
> #include <xen/smp.h>
> #include <xen/softirq.h>
> @@ -73,6 +75,10 @@ static struct livepatch_work livepatch_work;
> static DEFINE_PER_CPU(bool, work_to_do);
> static DEFINE_PER_CPU(struct tasklet, livepatch_tasklet);
>
> +#ifdef CONFIG_PAYLOAD_VERIFY
> +static struct rsa_public_key __ro_after_init builtin_payload_key;
> +#endif
> +
> static int get_name(const struct xen_livepatch_name *name, char *n)
> {
> if ( !name->size || name->size > XEN_LIVEPATCH_NAME_SIZE )
> @@ -2300,6 +2306,31 @@ static void cf_check livepatch_printall(unsigned char key)
> spin_unlock(&payload_lock);
> }
>
> +static int load_builtin_payload_key(void)
__init
> +{
> +#ifdef CONFIG_PAYLOAD_VERIFY
> + const uint8_t *ptr;
> + uint32_t len;
> +
> + rsa_public_key_init(&builtin_payload_key);
> +
> + ptr = xen_livepatch_key_data;
> +
> + memcpy(&len, ptr, sizeof(len));
How do you know enough data is available?
> + ptr += sizeof(len);
> + builtin_payload_key.n = mpi_read_raw_data(ptr, le32_to_cpu(len));
Even more so here. And then again below.
> + ptr += len;
> +
> + memcpy(&len, ptr, sizeof(len));
> + ptr += sizeof(len);
> + builtin_payload_key.e = mpi_read_raw_data(ptr, le32_to_cpu(len));
> +
> + return rsa_public_key_prepare(&builtin_payload_key);
> +#else
> + return 0;
> +#endif
> +}
> +
> static int cf_check cpu_callback(
> struct notifier_block *nfb, unsigned long action, void *hcpu)
> {
> @@ -2318,6 +2349,11 @@ static struct notifier_block cpu_nfb = {
> static int __init cf_check livepatch_init(void)
> {
> unsigned int cpu;
> + int err;
> +
> + err = load_builtin_payload_key();
> + if ( err )
> + return err;
>
> for_each_online_cpu ( cpu )
> {
So far this function can't fail. Therefore it makes sense that I can't find any
provisions for this case elsewhere in the file. Such provisions will need
adding as a prereq, I expect. Or alternatively it would want clarifying
explicitly that (why) none are required.
Jan
^ permalink raw reply [flat|nested] 20+ messages in thread
* Re: [PATCH v3 2/5] livepatch: Embed public key in Xen
2025-06-02 13:36 ` [PATCH v3 2/5] livepatch: Embed public key in Xen Ross Lagerwall
2025-06-05 11:02 ` Jan Beulich
@ 2025-06-05 11:19 ` Jan Beulich
2025-06-20 9:42 ` Roger Pau Monné
2025-06-20 9:39 ` Roger Pau Monné
2 siblings, 1 reply; 20+ messages in thread
From: Jan Beulich @ 2025-06-05 11:19 UTC (permalink / raw)
To: Ross Lagerwall
Cc: Kevin Lampis, Andrew Cooper, Anthony PERARD, Michal Orzel,
Julien Grall, Roger Pau Monné, Stefano Stabellini, xen-devel
On 02.06.2025 15:36, Ross Lagerwall wrote:
> From: Kevin Lampis <kevin.lampis@cloud.com>
>
> Make it possible to embed a public key in Xen to be used when verifying
> live patch payloads. Inclusion of the public key is optional.
>
> To avoid needing to include a DER / X.509 parser in the hypervisor, the
> public key is unpacked at build time and included in a form that is
> convenient for the hypervisor to consume. This is different approach
> from that used by Linux which embeds the entire X.509 certificate and
> builds in a parser for it.
>
> A suitable key can be created using openssl:
>
> openssl req -x509 -newkey rsa:2048 -keyout priv.pem -out pub.pem \
> -sha256 -days 3650 -nodes \
> -subj "/C=XX/ST=StateName/L=CityName/O=CompanyName/OU=CompanySectionName/CN=CommonNameOrHostname"
> openssl x509 -inform PEM -in pub.pem -outform PEM -pubkey -nocert -out verify_key.pem
>
> Signed-off-by: Kevin Lampis <kevin.lampis@cloud.com>
> Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
While reviewing patch 4 it occurred to me: Why embed the key? Can't this
be specified as (another) boot module?
Jan
^ permalink raw reply [flat|nested] 20+ messages in thread
* Re: [PATCH v3 5/5] livepatch: Verify livepatch signatures
2025-06-02 13:36 ` [PATCH v3 5/5] livepatch: Verify livepatch signatures Ross Lagerwall
@ 2025-06-05 11:52 ` Jan Beulich
2025-06-20 10:31 ` Roger Pau Monné
1 sibling, 0 replies; 20+ messages in thread
From: Jan Beulich @ 2025-06-05 11:52 UTC (permalink / raw)
To: Ross Lagerwall; +Cc: Roger Pau Monné, xen-devel
On 02.06.2025 15:36, Ross Lagerwall wrote:
> @@ -525,6 +526,106 @@ static int check_xen_buildid(const struct livepatch_elf *elf)
> return 0;
> }
>
> +#ifdef CONFIG_PAYLOAD_VERIFY
> +static int check_rsa_sha256_signature(void *data, size_t datalen,
> + void *sig, uint32_t siglen)
> +{
> + struct sha2_256_state hash;
> + MPI s;
> + int rc;
> +
> + s = mpi_read_raw_data(sig, siglen);
sig apparently being an input, can't the function parameter be pointer-to-const?
> + if ( !s )
> + {
> + printk(XENLOG_ERR LIVEPATCH "Failed to mpi_read_raw_data\n");
Both there and ...
> + return -ENOMEM;
> + }
> +
> + sha2_256_init(&hash);
> + sha2_256_update(&hash, data, datalen);
> +
> + rc = rsa_sha256_verify(&builtin_payload_key, &hash, s);
> + if ( rc )
> + printk(XENLOG_ERR LIVEPATCH "rsa_sha256_verify failed: %d\n", rc);
... here: Instead of mentioning function names, perhaps better to say in normal
word what failed. E.g. here "RSA/SHA256 verification failed".
> + mpi_free(s);
> +
> + return rc;
> +}
> +#endif
> +
> +static int check_signature(const struct livepatch_elf *elf, void *raw,
"raw" is an input only, too, isn't it? As such it ought to be const void *.
> + size_t size)
> +{
> +#ifdef CONFIG_PAYLOAD_VERIFY
> +#define MAX_SIG_NOTE_SIZE 1024
> + static const char notename[] = "Xen";
> + void *sig;
> + livepatch_elf_note note;
> + int rc;
> +
> + rc = livepatch_elf_note_by_names(elf, ELF_XEN_SIGNATURE, notename, -1,
> + ¬e);
> + if ( rc )
> + {
> + dprintk(XENLOG_DEBUG, LIVEPATCH "%s: Signature not present\n",
> + elf->name);
> + return rc;
> + }
> +
> + /* We expect only one signature, find a second is an error! */
> + rc = livepatch_elf_next_note_by_name(notename, -1, ¬e);
I can see why finding another signature of the same type might be a problem,
but one of a different type should be fine (i.e. unambiguous), I suppose.
Irrespective I find it slightly odd that you pass in a pointer to the same
local variable. The function, after all, is free to alter that variable
ahead of encountering the next matching note.
> + if ( rc != -ENOENT )
> + {
> + if ( rc )
> + {
> + printk(XENLOG_ERR LIVEPATCH
> + "Error while checking for notes! err = %d\n", rc);
> + return rc;
> + }
> + else
> + {
> + printk(XENLOG_ERR LIVEPATCH
> + "Error, found second signature note! There can be only one!\n");
> + return -EINVAL;
> + }
> + }
> +
> + if ( SIGNATURE_VERSION(note.type) != LIVEPATCH_SIGNATURE_VERSION ||
> + SIGNATURE_ALGORITHM(note.type) != SIGNATURE_ALGORITHM_RSA ||
> + SIGNATURE_HASH(note.type) != SIGNATURE_HASH_SHA256 )
> + {
> + printk(XENLOG_ERR LIVEPATCH
> + "Unsupported signature type: v:%u, a:%u, h:%u\n",
> + SIGNATURE_VERSION(note.type), SIGNATURE_ALGORITHM(note.type),
> + SIGNATURE_HASH(note.type));
> + return -EINVAL;
> + }
> +
> + if ( note.size == 0 || note.size >= MAX_SIG_NOTE_SIZE )
> + {
> + printk(XENLOG_ERR LIVEPATCH "Invalid signature note size: %u\n",
> + note.size);
> + return -EINVAL;
> + }
> +
> + sig = xmalloc_bytes(note.size);
In principle new code is supposed to be using functions from the xvmalloc()
family. (That family deliberately does not include xvmalloc_bytes(). I think
you mean xmalloc_array() here anyway.)
> @@ -1162,6 +1263,8 @@ static int load_payload_data(struct payload *payload, void *raw, size_t len)
> if ( rc )
> goto out;
>
> + check_signature(&elf, raw, len);
You entirely ignore the return value? (I was first wondering why the stub variant
of the function would return -EINVAL rather than 0.)
> --- a/xen/common/livepatch_elf.c
> +++ b/xen/common/livepatch_elf.c
> @@ -23,6 +23,61 @@ livepatch_elf_sec_by_name(const struct livepatch_elf *elf,
> return NULL;
> }
>
> +int livepatch_elf_note_by_names(const struct livepatch_elf *elf,
> + const char *sec_name, const char *note_name,
So the plural in the function name is because it's both a section and a note
name, whereas ...
> + const unsigned int type,
> + livepatch_elf_note *note)
> +{
> + const struct livepatch_elf_sec *sec = livepatch_elf_sec_by_name(elf,
> + sec_name);
> + if ( !sec )
> + return -ENOENT;
(Nit: Too deep indentation.)
> + note->end = sec->addr + sec->sec->sh_size;
> + note->next = sec->addr;
> +
> + return livepatch_elf_next_note_by_name(note_name, type, note);
> +}
> +
> +int livepatch_elf_next_note_by_name(const char *note_name,
> + const unsigned int type,
> + livepatch_elf_note *note)
... here you only look for further instances in the previously found section.
What if there's a 2nd section of the given name?
livepatch_elf_sec_by_name() also doesn't look to be making sure the section is
of the correct type.
Why is it necessary o pass in the note name again? A typical "find next" would
find the same kind as the "find first" did. Whereas here by passing in a
different name one can achieve "interesting" results. (Getting this correct
may not be necessary, but a comment would then want putting ahead of the
function.)
> +{
> + const Elf_Note *pkd_note = note->next;
> + size_t notenamelen = strlen(note_name) + 1;
> + size_t note_hd_size;
> + size_t note_size;
> + size_t remaining;
> +
> + while ( (void *)pkd_note < note->end )
Misra objects to the casting away of type qualifiers (and I do, too). More such
further down (and at least one instance also further up).
> + {
> +
Nit: Stray blank line. Perhaps you really meant to move some of the variable
declarations here?
> + remaining = note->end - (void *)pkd_note;
> + if ( remaining < sizeof(livepatch_elf_note) )
> + return -EINVAL;
> +
> + note_hd_size = sizeof(Elf_Note) + ((pkd_note->namesz + 3) & ~0x3);
> + note_size = note_hd_size + ((pkd_note->descsz + 3) & ~0x3);
What use are the 0x prefixes here (and then only on 2 of the three instances)?
> + if ( remaining < note_size )
> + return -EINVAL;
> +
> + if ( notenamelen == pkd_note->namesz &&
> + !memcmp(note_name, (const void *) pkd_note + sizeof(Elf_Note),
Nit: Stray blank between cast and cast value.
Jan
^ permalink raw reply [flat|nested] 20+ messages in thread
* Re: [PATCH v3 2/5] livepatch: Embed public key in Xen
2025-06-02 13:36 ` [PATCH v3 2/5] livepatch: Embed public key in Xen Ross Lagerwall
2025-06-05 11:02 ` Jan Beulich
2025-06-05 11:19 ` Jan Beulich
@ 2025-06-20 9:39 ` Roger Pau Monné
2025-06-20 10:09 ` Jan Beulich
2 siblings, 1 reply; 20+ messages in thread
From: Roger Pau Monné @ 2025-06-20 9:39 UTC (permalink / raw)
To: Ross Lagerwall
Cc: xen-devel, Kevin Lampis, Andrew Cooper, Anthony PERARD,
Michal Orzel, Jan Beulich, Julien Grall, Stefano Stabellini
On Mon, Jun 02, 2025 at 02:36:34PM +0100, Ross Lagerwall wrote:
> From: Kevin Lampis <kevin.lampis@cloud.com>
>
> Make it possible to embed a public key in Xen to be used when verifying
> live patch payloads. Inclusion of the public key is optional.
>
> To avoid needing to include a DER / X.509 parser in the hypervisor, the
> public key is unpacked at build time and included in a form that is
> convenient for the hypervisor to consume. This is different approach
> from that used by Linux which embeds the entire X.509 certificate and
> builds in a parser for it.
>
> A suitable key can be created using openssl:
>
> openssl req -x509 -newkey rsa:2048 -keyout priv.pem -out pub.pem \
> -sha256 -days 3650 -nodes \
> -subj "/C=XX/ST=StateName/L=CityName/O=CompanyName/OU=CompanySectionName/CN=CommonNameOrHostname"
> openssl x509 -inform PEM -in pub.pem -outform PEM -pubkey -nocert -out verify_key.pem
>
> Signed-off-by: Kevin Lampis <kevin.lampis@cloud.com>
> Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
> ---
>
> In v3:
>
> * Drop unnecessary condition in Makefile
> * Use dashes instead of underscores
> * Drop section placement annotation on declaration
> * Clarify endianness of embedded key
>
> xen/common/Kconfig | 18 +++++++++++++++++
> xen/crypto/Makefile | 11 ++++++++++
> xen/include/xen/livepatch.h | 5 +++++
> xen/tools/extract-key.py | 40 +++++++++++++++++++++++++++++++++++++
> 4 files changed, 74 insertions(+)
> create mode 100755 xen/tools/extract-key.py
>
> diff --git a/xen/common/Kconfig b/xen/common/Kconfig
> index 0951d4c2f286..74673078202a 100644
> --- a/xen/common/Kconfig
> +++ b/xen/common/Kconfig
> @@ -472,6 +472,24 @@ config LIVEPATCH
>
> If unsure, say Y.
>
> +config PAYLOAD_VERIFY
> + bool "Verify signed LivePatch payloads"
> + depends on LIVEPATCH
> + select CRYPTO
> + help
> + Verify signed LivePatch payloads using an RSA public key built
> + into the Xen hypervisor. Selecting this option requires a
> + public key in PEM format to be available for embedding during
> + the build.
> +
> +config PAYLOAD_VERIFY_KEY
> + string "File name of public key used to verify payloads"
> + default "verify_key.pem"
> + depends on PAYLOAD_VERIFY
> + help
> + The file name of an RSA public key in PEM format to be used for
> + verifying signed LivePatch payloads.
I think this is likely to break the randconfig testing that we do in
Gitlab CI, as randconfig could select PAYLOAD_VERIFY, but there will
be no key included, and hence the build will fail?
Ideally Gitlab CI would need to be adjusted to provide such key so the
build doesn't fail. I think it could be provided unconditionally to
simplify the logic, if the option is not selected the file will simply
be ignored.
Thanks, Roger.
^ permalink raw reply [flat|nested] 20+ messages in thread
* Re: [PATCH v3 2/5] livepatch: Embed public key in Xen
2025-06-05 11:19 ` Jan Beulich
@ 2025-06-20 9:42 ` Roger Pau Monné
0 siblings, 0 replies; 20+ messages in thread
From: Roger Pau Monné @ 2025-06-20 9:42 UTC (permalink / raw)
To: Jan Beulich
Cc: Ross Lagerwall, Kevin Lampis, Andrew Cooper, Anthony PERARD,
Michal Orzel, Julien Grall, Stefano Stabellini, xen-devel
On Thu, Jun 05, 2025 at 01:19:00PM +0200, Jan Beulich wrote:
> On 02.06.2025 15:36, Ross Lagerwall wrote:
> > From: Kevin Lampis <kevin.lampis@cloud.com>
> >
> > Make it possible to embed a public key in Xen to be used when verifying
> > live patch payloads. Inclusion of the public key is optional.
> >
> > To avoid needing to include a DER / X.509 parser in the hypervisor, the
> > public key is unpacked at build time and included in a form that is
> > convenient for the hypervisor to consume. This is different approach
> > from that used by Linux which embeds the entire X.509 certificate and
> > builds in a parser for it.
> >
> > A suitable key can be created using openssl:
> >
> > openssl req -x509 -newkey rsa:2048 -keyout priv.pem -out pub.pem \
> > -sha256 -days 3650 -nodes \
> > -subj "/C=XX/ST=StateName/L=CityName/O=CompanyName/OU=CompanySectionName/CN=CommonNameOrHostname"
> > openssl x509 -inform PEM -in pub.pem -outform PEM -pubkey -nocert -out verify_key.pem
> >
> > Signed-off-by: Kevin Lampis <kevin.lampis@cloud.com>
> > Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
>
> While reviewing patch 4 it occurred to me: Why embed the key? Can't this
> be specified as (another) boot module?
Then the key itself will need to be signed, and it's extra churn that
we would need to verify at boot. I'm not opposed to being able to
load the key as a module, but it seems reasonable to also bundle one
in Xen. If there's interest in passing one as a module it could
always be implemented as a separate feature. IMO: I don't see both
approaches as being incompatible with each other.
Thanks, Roger.
^ permalink raw reply [flat|nested] 20+ messages in thread
* Re: [PATCH v3 3/5] crypto: Add RSA support
2025-06-02 13:36 ` [PATCH v3 3/5] crypto: Add RSA support Ross Lagerwall
2025-06-05 11:06 ` Jan Beulich
@ 2025-06-20 9:53 ` Roger Pau Monné
2025-06-20 16:11 ` Ross Lagerwall
1 sibling, 1 reply; 20+ messages in thread
From: Roger Pau Monné @ 2025-06-20 9:53 UTC (permalink / raw)
To: Ross Lagerwall
Cc: xen-devel, Andrew Cooper, Anthony PERARD, Michal Orzel,
Jan Beulich, Julien Grall, Stefano Stabellini
On Mon, Jun 02, 2025 at 02:36:35PM +0100, Ross Lagerwall wrote:
> In preparation for adding support for livepatch signing, add support for
> RSA crypto.
>
> The RSA code is extracted from Nettle at tag nettle_3.2_release_20160128
> (https://git.lysator.liu.se/nettle/nettle).
>
> The MPI code is extracted from Linux at commit eef0df6a5953 (lib/mpi/*).
>
> Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
> ---
>
> In v3:
>
> * Move mpi.c to lib
> * Fix header guard name
>
> xen/crypto/Makefile | 1 +
> xen/crypto/rsa.c | 196 +++++
> xen/include/xen/mpi.h | 68 ++
> xen/include/xen/rsa.h | 74 ++
> xen/lib/Makefile | 1 +
> xen/lib/mpi.c | 1729 +++++++++++++++++++++++++++++++++++++++++
Just FTAOD: all the functions imported in mpi.c are used I think? So
that we don't introduce unreachable code.
Thanks, Roger.
^ permalink raw reply [flat|nested] 20+ messages in thread
* Re: [PATCH v3 2/5] livepatch: Embed public key in Xen
2025-06-20 9:39 ` Roger Pau Monné
@ 2025-06-20 10:09 ` Jan Beulich
2025-06-20 10:33 ` Roger Pau Monné
0 siblings, 1 reply; 20+ messages in thread
From: Jan Beulich @ 2025-06-20 10:09 UTC (permalink / raw)
To: Roger Pau Monné, Ross Lagerwall
Cc: xen-devel, Kevin Lampis, Andrew Cooper, Anthony PERARD,
Michal Orzel, Julien Grall, Stefano Stabellini
On 20.06.2025 11:39, Roger Pau Monné wrote:
> On Mon, Jun 02, 2025 at 02:36:34PM +0100, Ross Lagerwall wrote:
>> From: Kevin Lampis <kevin.lampis@cloud.com>
>>
>> Make it possible to embed a public key in Xen to be used when verifying
>> live patch payloads. Inclusion of the public key is optional.
>>
>> To avoid needing to include a DER / X.509 parser in the hypervisor, the
>> public key is unpacked at build time and included in a form that is
>> convenient for the hypervisor to consume. This is different approach
>> from that used by Linux which embeds the entire X.509 certificate and
>> builds in a parser for it.
>>
>> A suitable key can be created using openssl:
>>
>> openssl req -x509 -newkey rsa:2048 -keyout priv.pem -out pub.pem \
>> -sha256 -days 3650 -nodes \
>> -subj "/C=XX/ST=StateName/L=CityName/O=CompanyName/OU=CompanySectionName/CN=CommonNameOrHostname"
>> openssl x509 -inform PEM -in pub.pem -outform PEM -pubkey -nocert -out verify_key.pem
>>
>> Signed-off-by: Kevin Lampis <kevin.lampis@cloud.com>
>> Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
>> ---
>>
>> In v3:
>>
>> * Drop unnecessary condition in Makefile
>> * Use dashes instead of underscores
>> * Drop section placement annotation on declaration
>> * Clarify endianness of embedded key
>>
>> xen/common/Kconfig | 18 +++++++++++++++++
>> xen/crypto/Makefile | 11 ++++++++++
>> xen/include/xen/livepatch.h | 5 +++++
>> xen/tools/extract-key.py | 40 +++++++++++++++++++++++++++++++++++++
>> 4 files changed, 74 insertions(+)
>> create mode 100755 xen/tools/extract-key.py
>>
>> diff --git a/xen/common/Kconfig b/xen/common/Kconfig
>> index 0951d4c2f286..74673078202a 100644
>> --- a/xen/common/Kconfig
>> +++ b/xen/common/Kconfig
>> @@ -472,6 +472,24 @@ config LIVEPATCH
>>
>> If unsure, say Y.
>>
>> +config PAYLOAD_VERIFY
>> + bool "Verify signed LivePatch payloads"
>> + depends on LIVEPATCH
>> + select CRYPTO
>> + help
>> + Verify signed LivePatch payloads using an RSA public key built
>> + into the Xen hypervisor. Selecting this option requires a
>> + public key in PEM format to be available for embedding during
>> + the build.
>> +
>> +config PAYLOAD_VERIFY_KEY
>> + string "File name of public key used to verify payloads"
>> + default "verify_key.pem"
>> + depends on PAYLOAD_VERIFY
>> + help
>> + The file name of an RSA public key in PEM format to be used for
>> + verifying signed LivePatch payloads.
>
> I think this is likely to break the randconfig testing that we do in
> Gitlab CI, as randconfig could select PAYLOAD_VERIFY, but there will
> be no key included, and hence the build will fail?
>
> Ideally Gitlab CI would need to be adjusted to provide such key so the
> build doesn't fail. I think it could be provided unconditionally to
> simplify the logic, if the option is not selected the file will simply
> be ignored.
Alternatively the two options could be folded, the default being the
empty string meaning "no payload verification". I don't think randconfig
can sensibly make up random strings ...
Jan
^ permalink raw reply [flat|nested] 20+ messages in thread
* Re: [PATCH v3 5/5] livepatch: Verify livepatch signatures
2025-06-02 13:36 ` [PATCH v3 5/5] livepatch: Verify livepatch signatures Ross Lagerwall
2025-06-05 11:52 ` Jan Beulich
@ 2025-06-20 10:31 ` Roger Pau Monné
2025-06-20 16:50 ` Ross Lagerwall
1 sibling, 1 reply; 20+ messages in thread
From: Roger Pau Monné @ 2025-06-20 10:31 UTC (permalink / raw)
To: Ross Lagerwall; +Cc: xen-devel, Jennifer Herbert
On Mon, Jun 02, 2025 at 02:36:37PM +0100, Ross Lagerwall wrote:
> From: Jennifer Herbert <jennifer.herbert@cloud.com>
>
> Verify livepatch signatures against the embedded public key in Xen.
> Failing to verify does not prevent the livepatch from being loaded.
> In future, this will be changed for certain cases (e.g. when Secure Boot
> is enabled).
>
> Signed-off-by: Jennifer Herbert <jennifer.herbert@cloud.com>
> Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
> ---
>
> In v3:
>
> * Minor style fixes
>
> xen/common/livepatch.c | 103 ++++++++++++++++++++++++++++++++
> xen/common/livepatch_elf.c | 55 +++++++++++++++++
> xen/include/xen/livepatch.h | 10 ++++
> xen/include/xen/livepatch_elf.h | 18 ++++++
> 4 files changed, 186 insertions(+)
>
> diff --git a/xen/common/livepatch.c b/xen/common/livepatch.c
> index 92d1d342d872..56a3d125483f 100644
> --- a/xen/common/livepatch.c
> +++ b/xen/common/livepatch.c
> @@ -14,6 +14,7 @@
> #include <xen/mpi.h>
> #include <xen/rsa.h>
> #include <xen/sched.h>
> +#include <xen/sha2.h>
> #include <xen/smp.h>
> #include <xen/softirq.h>
> #include <xen/spinlock.h>
> @@ -525,6 +526,106 @@ static int check_xen_buildid(const struct livepatch_elf *elf)
> return 0;
> }
>
> +#ifdef CONFIG_PAYLOAD_VERIFY
> +static int check_rsa_sha256_signature(void *data, size_t datalen,
> + void *sig, uint32_t siglen)
I think both data and sig could be const here?
> +{
> + struct sha2_256_state hash;
> + MPI s;
> + int rc;
> +
> + s = mpi_read_raw_data(sig, siglen);
> + if ( !s )
> + {
> + printk(XENLOG_ERR LIVEPATCH "Failed to mpi_read_raw_data\n");
> + return -ENOMEM;
> + }
> +
> + sha2_256_init(&hash);
> + sha2_256_update(&hash, data, datalen);
> +
> + rc = rsa_sha256_verify(&builtin_payload_key, &hash, s);
> + if ( rc )
> + printk(XENLOG_ERR LIVEPATCH "rsa_sha256_verify failed: %d\n", rc);
> +
> + mpi_free(s);
> +
> + return rc;
> +}
> +#endif
> +
> +static int check_signature(const struct livepatch_elf *elf, void *raw,
> + size_t size)
> +{
> +#ifdef CONFIG_PAYLOAD_VERIFY
> +#define MAX_SIG_NOTE_SIZE 1024
> + static const char notename[] = "Xen";
> + void *sig;
> + livepatch_elf_note note;
> + int rc;
> +
> + rc = livepatch_elf_note_by_names(elf, ELF_XEN_SIGNATURE, notename, -1,
> + ¬e);
> + if ( rc )
> + {
> + dprintk(XENLOG_DEBUG, LIVEPATCH "%s: Signature not present\n",
> + elf->name);
> + return rc;
> + }
> +
> + /* We expect only one signature, find a second is an error! */
> + rc = livepatch_elf_next_note_by_name(notename, -1, ¬e);
> + if ( rc != -ENOENT )
> + {
> + if ( rc )
> + {
> + printk(XENLOG_ERR LIVEPATCH
> + "Error while checking for notes! err = %d\n", rc);
> + return rc;
> + }
> + else
> + {
> + printk(XENLOG_ERR LIVEPATCH
> + "Error, found second signature note! There can be only one!\n");
FWIW, it's a nit, but I think there are too many exclamation marks in
the error message above:
"Error, multiple signatures found\n"
Would be better.
I'm also wondering, would it make sense to allow multiple signatures
to be present? I guess not because livepatches are built for specific
Xen binaries, and then we know exactly which key is embedded there. A
livepatch would never apply to two different builds with different
keys.
> + return -EINVAL;
> + }
> + }
> +
> + if ( SIGNATURE_VERSION(note.type) != LIVEPATCH_SIGNATURE_VERSION ||
> + SIGNATURE_ALGORITHM(note.type) != SIGNATURE_ALGORITHM_RSA ||
> + SIGNATURE_HASH(note.type) != SIGNATURE_HASH_SHA256 )
> + {
> + printk(XENLOG_ERR LIVEPATCH
> + "Unsupported signature type: v:%u, a:%u, h:%u\n",
> + SIGNATURE_VERSION(note.type), SIGNATURE_ALGORITHM(note.type),
> + SIGNATURE_HASH(note.type));
> + return -EINVAL;
> + }
> +
> + if ( note.size == 0 || note.size >= MAX_SIG_NOTE_SIZE )
> + {
> + printk(XENLOG_ERR LIVEPATCH "Invalid signature note size: %u\n",
> + note.size);
> + return -EINVAL;
> + }
> +
> + sig = xmalloc_bytes(note.size);
> + if ( !sig )
> + return -ENOMEM;
> +
> + memcpy(sig, note.data, note.size);
> +
> + /* Remove signature from data, as can't be verified with it. */
> + memset((void *)note.data, 0, note.size);
> + rc = check_rsa_sha256_signature(raw, size, sig, note.size);
> +
> + xfree(sig);
> + return rc;
> +#else
> + return -EINVAL;
EOPNOTSUPP would be more accurate here.
> +#endif
> +}
> +
> static int check_special_sections(const struct livepatch_elf *elf)
> {
> unsigned int i;
> @@ -1162,6 +1263,8 @@ static int load_payload_data(struct payload *payload, void *raw, size_t len)
> if ( rc )
> goto out;
>
> + check_signature(&elf, raw, len);
Wouldn't it make more sense to unconditionally fail here if
CONFIG_PAYLOAD_VERIFY is enabled and signature verification fails?
IOW: if you built an hypervisor with PAYLOAD_VERIFY enabled signature
verification needs to be enforced.
> +
> rc = move_payload(payload, &elf);
> if ( rc )
> goto out;
> diff --git a/xen/common/livepatch_elf.c b/xen/common/livepatch_elf.c
> index 25ce1bd5a0ad..b27c4524611d 100644
> --- a/xen/common/livepatch_elf.c
> +++ b/xen/common/livepatch_elf.c
> @@ -23,6 +23,61 @@ livepatch_elf_sec_by_name(const struct livepatch_elf *elf,
> return NULL;
> }
>
> +int livepatch_elf_note_by_names(const struct livepatch_elf *elf,
> + const char *sec_name, const char *note_name,
> + const unsigned int type,
> + livepatch_elf_note *note)
> +{
> + const struct livepatch_elf_sec *sec = livepatch_elf_sec_by_name(elf,
> + sec_name);
> + if ( !sec )
> + return -ENOENT;
> +
> + note->end = sec->addr + sec->sec->sh_size;
> + note->next = sec->addr;
> +
> + return livepatch_elf_next_note_by_name(note_name, type, note);
> +}
> +
> +int livepatch_elf_next_note_by_name(const char *note_name,
> + const unsigned int type,
> + livepatch_elf_note *note)
> +{
> + const Elf_Note *pkd_note = note->next;
> + size_t notenamelen = strlen(note_name) + 1;
> + size_t note_hd_size;
> + size_t note_size;
> + size_t remaining;
> +
> + while ( (void *)pkd_note < note->end )
> + {
> +
Stray newline?
remaining and note_hd_size can be defined inside of the loop to reduce
the scope.
> + remaining = note->end - (void *)pkd_note;
> + if ( remaining < sizeof(livepatch_elf_note) )
> + return -EINVAL;
> +
> + note_hd_size = sizeof(Elf_Note) + ((pkd_note->namesz + 3) & ~0x3);
> + note_size = note_hd_size + ((pkd_note->descsz + 3) & ~0x3);
Could be have a macro or similar for this ((x + 3) & ~0x3)
manipulation?
> + if ( remaining < note_size )
> + return -EINVAL;
ENOSPC might be less ambiguous than EINVAL for both the above?
> +
> + if ( notenamelen == pkd_note->namesz &&
> + !memcmp(note_name, (const void *) pkd_note + sizeof(Elf_Note),
> + notenamelen) &&
> + (type == -1 || pkd_note->type == type) )
> + {
> + note->size = pkd_note->descsz;
> + note->type = pkd_note->type;
> + note->data = (void *)pkd_note + note_hd_size;
> + note->next = (void *)pkd_note + note_size;
> + return 0;
> + }
> + pkd_note = (void *)pkd_note + note_size;
> + }
It's a nit, but I would write this as a for loop, as it's clearer then
the index advance:
for ( pkd_note = note->next; (void *)pkd_note < note->end;
pkd_note = (void *)pkd_note + note_size )
{
...
> +
> + return -ENOENT;
> +}
> +
> static int elf_verify_strtab(const struct livepatch_elf_sec *sec)
> {
> const Elf_Shdr *s;
> diff --git a/xen/include/xen/livepatch.h b/xen/include/xen/livepatch.h
> index 52f90cbed45b..12206ce3b2b8 100644
> --- a/xen/include/xen/livepatch.h
> +++ b/xen/include/xen/livepatch.h
> @@ -38,6 +38,7 @@ struct xen_sysctl_livepatch_op;
> #define ELF_LIVEPATCH_DEPENDS ".livepatch.depends"
> #define ELF_LIVEPATCH_XEN_DEPENDS ".livepatch.xen_depends"
> #define ELF_BUILD_ID_NOTE ".note.gnu.build-id"
> +#define ELF_XEN_SIGNATURE ".note.Xen.signature"
> #define ELF_LIVEPATCH_LOAD_HOOKS ".livepatch.hooks.load"
> #define ELF_LIVEPATCH_UNLOAD_HOOKS ".livepatch.hooks.unload"
> #define ELF_LIVEPATCH_PREAPPLY_HOOK ".livepatch.hooks.preapply"
> @@ -49,6 +50,15 @@ struct xen_sysctl_livepatch_op;
> /* Arbitrary limit for payload size and .bss section size. */
> #define LIVEPATCH_MAX_SIZE MB(2)
>
> +#define SIGNATURE_VERSION(type) ((type) & 0xffff)
> +#define LIVEPATCH_SIGNATURE_VERSION 1
> +
> +#define SIGNATURE_ALGORITHM(type) (((type) >> 16) & 0xff)
> +#define SIGNATURE_ALGORITHM_RSA 0
> +
> +#define SIGNATURE_HASH(type) (((type) >> 24) & 0xff)
> +#define SIGNATURE_HASH_SHA256 0
You could possibly use MASK_EXTR() for the macros above?
Thanks, Roger.
^ permalink raw reply [flat|nested] 20+ messages in thread
* Re: [PATCH v3 2/5] livepatch: Embed public key in Xen
2025-06-20 10:09 ` Jan Beulich
@ 2025-06-20 10:33 ` Roger Pau Monné
0 siblings, 0 replies; 20+ messages in thread
From: Roger Pau Monné @ 2025-06-20 10:33 UTC (permalink / raw)
To: Jan Beulich
Cc: Ross Lagerwall, xen-devel, Kevin Lampis, Andrew Cooper,
Anthony PERARD, Michal Orzel, Julien Grall, Stefano Stabellini
On Fri, Jun 20, 2025 at 12:09:21PM +0200, Jan Beulich wrote:
> On 20.06.2025 11:39, Roger Pau Monné wrote:
> > On Mon, Jun 02, 2025 at 02:36:34PM +0100, Ross Lagerwall wrote:
> >> From: Kevin Lampis <kevin.lampis@cloud.com>
> >>
> >> Make it possible to embed a public key in Xen to be used when verifying
> >> live patch payloads. Inclusion of the public key is optional.
> >>
> >> To avoid needing to include a DER / X.509 parser in the hypervisor, the
> >> public key is unpacked at build time and included in a form that is
> >> convenient for the hypervisor to consume. This is different approach
> >> from that used by Linux which embeds the entire X.509 certificate and
> >> builds in a parser for it.
> >>
> >> A suitable key can be created using openssl:
> >>
> >> openssl req -x509 -newkey rsa:2048 -keyout priv.pem -out pub.pem \
> >> -sha256 -days 3650 -nodes \
> >> -subj "/C=XX/ST=StateName/L=CityName/O=CompanyName/OU=CompanySectionName/CN=CommonNameOrHostname"
> >> openssl x509 -inform PEM -in pub.pem -outform PEM -pubkey -nocert -out verify_key.pem
> >>
> >> Signed-off-by: Kevin Lampis <kevin.lampis@cloud.com>
> >> Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
> >> ---
> >>
> >> In v3:
> >>
> >> * Drop unnecessary condition in Makefile
> >> * Use dashes instead of underscores
> >> * Drop section placement annotation on declaration
> >> * Clarify endianness of embedded key
> >>
> >> xen/common/Kconfig | 18 +++++++++++++++++
> >> xen/crypto/Makefile | 11 ++++++++++
> >> xen/include/xen/livepatch.h | 5 +++++
> >> xen/tools/extract-key.py | 40 +++++++++++++++++++++++++++++++++++++
> >> 4 files changed, 74 insertions(+)
> >> create mode 100755 xen/tools/extract-key.py
> >>
> >> diff --git a/xen/common/Kconfig b/xen/common/Kconfig
> >> index 0951d4c2f286..74673078202a 100644
> >> --- a/xen/common/Kconfig
> >> +++ b/xen/common/Kconfig
> >> @@ -472,6 +472,24 @@ config LIVEPATCH
> >>
> >> If unsure, say Y.
> >>
> >> +config PAYLOAD_VERIFY
> >> + bool "Verify signed LivePatch payloads"
> >> + depends on LIVEPATCH
> >> + select CRYPTO
> >> + help
> >> + Verify signed LivePatch payloads using an RSA public key built
> >> + into the Xen hypervisor. Selecting this option requires a
> >> + public key in PEM format to be available for embedding during
> >> + the build.
> >> +
> >> +config PAYLOAD_VERIFY_KEY
> >> + string "File name of public key used to verify payloads"
> >> + default "verify_key.pem"
> >> + depends on PAYLOAD_VERIFY
> >> + help
> >> + The file name of an RSA public key in PEM format to be used for
> >> + verifying signed LivePatch payloads.
> >
> > I think this is likely to break the randconfig testing that we do in
> > Gitlab CI, as randconfig could select PAYLOAD_VERIFY, but there will
> > be no key included, and hence the build will fail?
> >
> > Ideally Gitlab CI would need to be adjusted to provide such key so the
> > build doesn't fail. I think it could be provided unconditionally to
> > simplify the logic, if the option is not selected the file will simply
> > be ignored.
>
> Alternatively the two options could be folded, the default being the
> empty string meaning "no payload verification". I don't think randconfig
> can sensibly make up random strings ...
Could be an option. Not sure if the menu interface would then look a
bit weird. IMO it's a nicer UI to enable the option and then get
asked for the cert to use rather than bundling everything in a single
string-like option.
Thanks, Roger.
^ permalink raw reply [flat|nested] 20+ messages in thread
* Re: [PATCH v3 3/5] crypto: Add RSA support
2025-06-20 9:53 ` Roger Pau Monné
@ 2025-06-20 16:11 ` Ross Lagerwall
2025-06-23 7:28 ` Jan Beulich
0 siblings, 1 reply; 20+ messages in thread
From: Ross Lagerwall @ 2025-06-20 16:11 UTC (permalink / raw)
To: Roger Pau Monné
Cc: xen-devel, Andrew Cooper, Anthony PERARD, Michal Orzel,
Jan Beulich, Julien Grall, Stefano Stabellini
On Fri, Jun 20, 2025 at 10:53 AM Roger Pau Monné <roger.pau@citrix.com> wrote:
>
> On Mon, Jun 02, 2025 at 02:36:35PM +0100, Ross Lagerwall wrote:
> > In preparation for adding support for livepatch signing, add support for
> > RSA crypto.
> >
> > The RSA code is extracted from Nettle at tag nettle_3.2_release_20160128
> > (https://git.lysator.liu.se/nettle/nettle).
> >
> > The MPI code is extracted from Linux at commit eef0df6a5953 (lib/mpi/*).
> >
> > Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
> > ---
> >
> > In v3:
> >
> > * Move mpi.c to lib
> > * Fix header guard name
> >
> > xen/crypto/Makefile | 1 +
> > xen/crypto/rsa.c | 196 +++++
> > xen/include/xen/mpi.h | 68 ++
> > xen/include/xen/rsa.h | 74 ++
> > xen/lib/Makefile | 1 +
> > xen/lib/mpi.c | 1729 +++++++++++++++++++++++++++++++++++++++++
>
> Just FTAOD: all the functions imported in mpi.c are used I think? So
> that we don't introduce unreachable code.
>
Yes, everything included here is used by the end of the patch series.
Ross
^ permalink raw reply [flat|nested] 20+ messages in thread
* Re: [PATCH v3 5/5] livepatch: Verify livepatch signatures
2025-06-20 10:31 ` Roger Pau Monné
@ 2025-06-20 16:50 ` Ross Lagerwall
0 siblings, 0 replies; 20+ messages in thread
From: Ross Lagerwall @ 2025-06-20 16:50 UTC (permalink / raw)
To: Roger Pau Monné; +Cc: xen-devel
On Fri, Jun 20, 2025 at 11:31 AM Roger Pau Monné <roger.pau@citrix.com> wrote:
>
> On Mon, Jun 02, 2025 at 02:36:37PM +0100, Ross Lagerwall wrote:
> > From: Jennifer Herbert <jennifer.herbert@cloud.com>
> >
> > Verify livepatch signatures against the embedded public key in Xen.
> > Failing to verify does not prevent the livepatch from being loaded.
> > In future, this will be changed for certain cases (e.g. when Secure Boot
> > is enabled).
> >
> > Signed-off-by: Jennifer Herbert <jennifer.herbert@cloud.com>
> > Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
> > ---
> >
> > In v3:
> >
> > * Minor style fixes
> >
> > xen/common/livepatch.c | 103 ++++++++++++++++++++++++++++++++
> > xen/common/livepatch_elf.c | 55 +++++++++++++++++
> > xen/include/xen/livepatch.h | 10 ++++
> > xen/include/xen/livepatch_elf.h | 18 ++++++
> > 4 files changed, 186 insertions(+)
> >
> > diff --git a/xen/common/livepatch.c b/xen/common/livepatch.c
> > index 92d1d342d872..56a3d125483f 100644
> > --- a/xen/common/livepatch.c
> > +++ b/xen/common/livepatch.c
> > @@ -14,6 +14,7 @@
> > #include <xen/mpi.h>
> > #include <xen/rsa.h>
> > #include <xen/sched.h>
> > +#include <xen/sha2.h>
> > #include <xen/smp.h>
> > #include <xen/softirq.h>
> > #include <xen/spinlock.h>
> > @@ -525,6 +526,106 @@ static int check_xen_buildid(const struct livepatch_elf *elf)
> > return 0;
> > }
> >
> > +#ifdef CONFIG_PAYLOAD_VERIFY
> > +static int check_rsa_sha256_signature(void *data, size_t datalen,
> > + void *sig, uint32_t siglen)
>
> I think both data and sig could be const here?
>
> > +{
> > + struct sha2_256_state hash;
> > + MPI s;
> > + int rc;
> > +
> > + s = mpi_read_raw_data(sig, siglen);
> > + if ( !s )
> > + {
> > + printk(XENLOG_ERR LIVEPATCH "Failed to mpi_read_raw_data\n");
> > + return -ENOMEM;
> > + }
> > +
> > + sha2_256_init(&hash);
> > + sha2_256_update(&hash, data, datalen);
> > +
> > + rc = rsa_sha256_verify(&builtin_payload_key, &hash, s);
> > + if ( rc )
> > + printk(XENLOG_ERR LIVEPATCH "rsa_sha256_verify failed: %d\n", rc);
> > +
> > + mpi_free(s);
> > +
> > + return rc;
> > +}
> > +#endif
> > +
> > +static int check_signature(const struct livepatch_elf *elf, void *raw,
> > + size_t size)
> > +{
> > +#ifdef CONFIG_PAYLOAD_VERIFY
> > +#define MAX_SIG_NOTE_SIZE 1024
> > + static const char notename[] = "Xen";
> > + void *sig;
> > + livepatch_elf_note note;
> > + int rc;
> > +
> > + rc = livepatch_elf_note_by_names(elf, ELF_XEN_SIGNATURE, notename, -1,
> > + ¬e);
> > + if ( rc )
> > + {
> > + dprintk(XENLOG_DEBUG, LIVEPATCH "%s: Signature not present\n",
> > + elf->name);
> > + return rc;
> > + }
> > +
> > + /* We expect only one signature, find a second is an error! */
> > + rc = livepatch_elf_next_note_by_name(notename, -1, ¬e);
> > + if ( rc != -ENOENT )
> > + {
> > + if ( rc )
> > + {
> > + printk(XENLOG_ERR LIVEPATCH
> > + "Error while checking for notes! err = %d\n", rc);
> > + return rc;
> > + }
> > + else
> > + {
> > + printk(XENLOG_ERR LIVEPATCH
> > + "Error, found second signature note! There can be only one!\n");
>
> FWIW, it's a nit, but I think there are too many exclamation marks in
> the error message above:
>
> "Error, multiple signatures found\n"
>
> Would be better.
>
> I'm also wondering, would it make sense to allow multiple signatures
> to be present? I guess not because livepatches are built for specific
> Xen binaries, and then we know exactly which key is embedded there. A
> livepatch would never apply to two different builds with different
> keys.
I agree that while it could be possible, I don't really see a valid use
for it. This restriction could be relaxed in future if it is deemed
useful.
>
> > + return -EINVAL;
> > + }
> > + }
> > +
> > + if ( SIGNATURE_VERSION(note.type) != LIVEPATCH_SIGNATURE_VERSION ||
> > + SIGNATURE_ALGORITHM(note.type) != SIGNATURE_ALGORITHM_RSA ||
> > + SIGNATURE_HASH(note.type) != SIGNATURE_HASH_SHA256 )
> > + {
> > + printk(XENLOG_ERR LIVEPATCH
> > + "Unsupported signature type: v:%u, a:%u, h:%u\n",
> > + SIGNATURE_VERSION(note.type), SIGNATURE_ALGORITHM(note.type),
> > + SIGNATURE_HASH(note.type));
> > + return -EINVAL;
> > + }
> > +
> > + if ( note.size == 0 || note.size >= MAX_SIG_NOTE_SIZE )
> > + {
> > + printk(XENLOG_ERR LIVEPATCH "Invalid signature note size: %u\n",
> > + note.size);
> > + return -EINVAL;
> > + }
> > +
> > + sig = xmalloc_bytes(note.size);
> > + if ( !sig )
> > + return -ENOMEM;
> > +
> > + memcpy(sig, note.data, note.size);
> > +
> > + /* Remove signature from data, as can't be verified with it. */
> > + memset((void *)note.data, 0, note.size);
> > + rc = check_rsa_sha256_signature(raw, size, sig, note.size);
> > +
> > + xfree(sig);
> > + return rc;
> > +#else
> > + return -EINVAL;
>
> EOPNOTSUPP would be more accurate here.
>
> > +#endif
> > +}
> > +
> > static int check_special_sections(const struct livepatch_elf *elf)
> > {
> > unsigned int i;
> > @@ -1162,6 +1263,8 @@ static int load_payload_data(struct payload *payload, void *raw, size_t len)
> > if ( rc )
> > goto out;
> >
> > + check_signature(&elf, raw, len);
>
> Wouldn't it make more sense to unconditionally fail here if
> CONFIG_PAYLOAD_VERIFY is enabled and signature verification fails?
>
> IOW: if you built an hypervisor with PAYLOAD_VERIFY enabled signature
> verification needs to be enforced.
As per the commit message, this doesn't actually enforce anything yet.
The intention is to tie signature enforcement to lockdown mode once
those patches have been merged.
Ross
^ permalink raw reply [flat|nested] 20+ messages in thread
* Re: [PATCH v3 3/5] crypto: Add RSA support
2025-06-20 16:11 ` Ross Lagerwall
@ 2025-06-23 7:28 ` Jan Beulich
0 siblings, 0 replies; 20+ messages in thread
From: Jan Beulich @ 2025-06-23 7:28 UTC (permalink / raw)
To: Ross Lagerwall
Cc: xen-devel, Andrew Cooper, Anthony PERARD, Michal Orzel,
Julien Grall, Stefano Stabellini, Roger Pau Monné
On 20.06.2025 18:11, Ross Lagerwall wrote:
> On Fri, Jun 20, 2025 at 10:53 AM Roger Pau Monné <roger.pau@citrix.com> wrote:
>>
>> On Mon, Jun 02, 2025 at 02:36:35PM +0100, Ross Lagerwall wrote:
>>> In preparation for adding support for livepatch signing, add support for
>>> RSA crypto.
>>>
>>> The RSA code is extracted from Nettle at tag nettle_3.2_release_20160128
>>> (https://git.lysator.liu.se/nettle/nettle).
>>>
>>> The MPI code is extracted from Linux at commit eef0df6a5953 (lib/mpi/*).
>>>
>>> Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
>>> ---
>>>
>>> In v3:
>>>
>>> * Move mpi.c to lib
>>> * Fix header guard name
>>>
>>> xen/crypto/Makefile | 1 +
>>> xen/crypto/rsa.c | 196 +++++
>>> xen/include/xen/mpi.h | 68 ++
>>> xen/include/xen/rsa.h | 74 ++
>>> xen/lib/Makefile | 1 +
>>> xen/lib/mpi.c | 1729 +++++++++++++++++++++++++++++++++++++++++
>>
>> Just FTAOD: all the functions imported in mpi.c are used I think? So
>> that we don't introduce unreachable code.
>
> Yes, everything included here is used by the end of the patch series.
What about intermediate state? It's conceivable that such a series might go
in piecemeal ...
Jan
^ permalink raw reply [flat|nested] 20+ messages in thread
end of thread, other threads:[~2025-06-23 7:29 UTC | newest]
Thread overview: 20+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2025-06-02 13:36 [PATCH v3 0/5] LivePatch signing support Ross Lagerwall
2025-06-02 13:36 ` [PATCH v3 1/5] docs: Introduce live patch signing Ross Lagerwall
2025-06-02 13:36 ` [PATCH v3 2/5] livepatch: Embed public key in Xen Ross Lagerwall
2025-06-05 11:02 ` Jan Beulich
2025-06-05 11:19 ` Jan Beulich
2025-06-20 9:42 ` Roger Pau Monné
2025-06-20 9:39 ` Roger Pau Monné
2025-06-20 10:09 ` Jan Beulich
2025-06-20 10:33 ` Roger Pau Monné
2025-06-02 13:36 ` [PATCH v3 3/5] crypto: Add RSA support Ross Lagerwall
2025-06-05 11:06 ` Jan Beulich
2025-06-20 9:53 ` Roger Pau Monné
2025-06-20 16:11 ` Ross Lagerwall
2025-06-23 7:28 ` Jan Beulich
2025-06-02 13:36 ` [PATCH v3 4/5] livepatch: Load built-in key during boot Ross Lagerwall
2025-06-05 11:17 ` Jan Beulich
2025-06-02 13:36 ` [PATCH v3 5/5] livepatch: Verify livepatch signatures Ross Lagerwall
2025-06-05 11:52 ` Jan Beulich
2025-06-20 10:31 ` Roger Pau Monné
2025-06-20 16:50 ` Ross Lagerwall
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.