Skip to contentSkip to conversation

nvme: implement secure concatenation

PATCH Superseded Hannes Reinecke 3 participants 21 messages Jul 22, 2024 – Jul 26, 2024

Subsystems (5)

v8 of 15 Superseded Latest is v15 →

lore
Share & tools
lore
#0 lore raw

v8 cover nvme: implement secure concatenation

From: Hannes Reinecke <[email protected]>

Hi all,

here's my attempt to implement secure concatenation for NVMe-of TCP as outlined in TP8018.
The original (v5) patchset had been split in two, and this is the second part based on top of the patchset 'nvme: fixes for secure concatenation' sent earlier to the mailinglist.

Secure concatenation means that a TLS PSK is generated from the key material negotiated by the DH-HMAC-CHAP protocol, and the TLS PSK is then used for a subsequent TLS connection.
The difference between the original definition of secure concatenation and the method outlined in TP8018 is that with TP8018 the connection is reset after DH-HMAC-CHAP negotiation, and a new connection is setup with the generated TLS PSK.

To implement that Sagi came up with the idea to directly reset the admin queue once the DH-CHAP negotiation has completed; that way it will be transparent to the upper layers and we don't have to worry about exposing queues which should not be used.

As usual, comments and reviews are welcome.

Patchset can be found at git.kernel.org:/pub/scm/linux/kernel/git/hare/nvme.git branch secure-concat.v8

Changes to v7:
- Add patch to display nvme target TLS status in debugfs
- Include reviews from Sagi

Changes to v6:
- Rebase to nvme-6.11

Changes to v5:
- Include reviews from Sagi
- Split patchset in two parts

Changes to v4:
- Rework reset admin queue functionality based on an idea from Sagi (thanks!)
- kbuild robot fixes
- Fixup dhchap negotiation with non-empty C2 value

Changes to v3:
- Include reviews from Sagi
- Do not start I/O queues after DH-HMAC-CHAP negotiation
- Use bool to indicate TLS has been enabled on a queue
- Add 'tls_keyring' sysfs attribute
- Add 'tls_configured_key' sysfs attribute

Changes to v2:
- Fixup reset after dhchap negotiation
- Disable namespace scanning on I/O queues after
dhchap negotiation
- Reworked TLS key handling (again)

Changes to the original submission:
- Sanitize TLS key handling
- Fixup modconfig compilation

Hannes Reinecke (9):
crypto,fs: Separate out hkdf_extract() and hkdf_expand()
nvme: add nvme_auth_generate_psk()
nvme: add nvme_auth_generate_digest()
nvme: add nvme_auth_derive_tls_psk()
nvme-keyring: add nvme_tls_psk_refresh()
nvme-tcp: request secure channel concatenation
nvme-fabrics: reset admin connection for secure concatenation
nvmet-tcp: support secure channel concatenation
nvmet: add tls_concat and tls_key debugfs entries

 crypto/Makefile                        |   1 +
 crypto/hkdf.c                          | 112 +++++++++
 drivers/nvme/common/auth.c             | 303 +++++++++++++++++++++++++
 drivers/nvme/common/keyring.c          |  50 ++++
 drivers/nvme/host/auth.c               | 108 ++++++++-
 drivers/nvme/host/fabrics.c            |  34 ++-
 drivers/nvme/host/fabrics.h            |   3 +
 drivers/nvme/host/nvme.h               |   2 +
 drivers/nvme/host/sysfs.c              |   4 +-
 drivers/nvme/host/tcp.c                |  56 ++++-
 drivers/nvme/target/auth.c             |  72 +++++-
 drivers/nvme/target/debugfs.c          |  27 +++
 drivers/nvme/target/fabrics-cmd-auth.c |  49 +++-
 drivers/nvme/target/fabrics-cmd.c      |  33 ++-
 drivers/nvme/target/nvmet.h            |  38 +++-
 drivers/nvme/target/tcp.c              |  23 +-
 fs/crypto/hkdf.c                       |  68 +-----
 include/crypto/hkdf.h                  |  18 ++
 include/linux/nvme-auth.h              |   7 +
 include/linux/nvme-keyring.h           |   7 +
 include/linux/nvme.h                   |   7 +
 21 files changed, 926 insertions(+), 96 deletions(-)
 create mode 100644 crypto/hkdf.c
 create mode 100644 include/crypto/hkdf.h
Signature · 1 line
2.35.3
#1 lore raw

PATCH v8 1/9 crypto,fs: Separate out hkdf_extract() and hkdf_expand()

Separate out the HKDF functions into a separate file to make them available to other callers.

---
 crypto/Makefile       |   1 +
 crypto/hkdf.c         | 112 ++++++++++++++++++++++++++++++++++++++++++
 fs/crypto/hkdf.c      |  68 ++++---------------------
 include/crypto/hkdf.h |  18 +++++++
 4 files changed, 140 insertions(+), 59 deletions(-)
 create mode 100644 crypto/hkdf.c
 create mode 100644 include/crypto/hkdf.h
crypto/Makefile +1 -0
index edbbaa3ffef5..b77fc360f0ff 100644
@@ -29,6 +29,7 @@ obj-$(CONFIG_CRYPTO_ECHAINIV) += echainiv.o
 
 crypto_hash-y += ahash.o
 crypto_hash-y += shash.o
+crypto_hash-y += hkdf.o
 obj-$(CONFIG_CRYPTO_HASH2) += crypto_hash.o
 
 obj-$(CONFIG_CRYPTO_AKCIPHER2) += akcipher.o
crypto/hkdf.c new file +112 -0

112 lines not on this page · raw message

fs/crypto/hkdf.c +9 -59

96 lines not on this page · raw message

include/crypto/hkdf.h new file +18 -0
new file mode 100644
index 000000000000..bf06c080d7ed
@@ -0,0 +1,18 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+/*
+ * HKDF: HMAC-based Key Derivation Function (HKDF), RFC 5869
+ *
+ * Extracted from fs/crypto/hkdf.c, which has
+ * Copyright 2019 Google LLC
+ */
+
+#ifndef _CRYPTO_HKDF_H
+#define _CRYPTO_HKDF_H
+
+int hkdf_extract(struct crypto_shash *hmac_tfm, const u8 *ikm,
+		 unsigned int ikmlen, u8 *prk);
+int hkdf_expand(struct crypto_shash *hmac_tfm,
+		const u8 *info, unsigned int infolen,
+		u8 *okm, unsigned int okmlen);
+
+#endif
Signature · 1 line
2.35.3
#4 lore raw

PATCH v8 2/9 nvme: add nvme_auth_generate_psk()

In reply to Hannes Reinecke

Add a function to generate a NVMe PSK from the shared credentials negotiated by DH-HMAC-CHAP.

---
 drivers/nvme/common/auth.c | 87 ++++++++++++++++++++++++++++++++++++++
 include/linux/nvme-auth.h  |  3 ++
 2 files changed, 90 insertions(+)
drivers/nvme/common/auth.c +87 -0
index a3455f1d67fa..f6d21960b140 100644
@@ -11,6 +11,7 @@
 #include <asm/unaligned.h>
 #include <crypto/hash.h>
 #include <crypto/dh.h>
+#include <crypto/hkdf.h>
 #include <linux/nvme.h>
 #include <linux/nvme-auth.h>
 
@@ -471,5 +472,91 @@ int nvme_auth_generate_key(u8 *secret, struct nvme_dhchap_key **ret_key)
 }
 EXPORT_SYMBOL_GPL(nvme_auth_generate_key);
 
+/*
+ * nvme_auth_generate_psk - Generate a PSK for TLS
+ *
+ * Generate a PSK for TLS as specified in NVMe base specification, section 8.13.5.9:
+ *    Generated PSK for TLS
+ *
+ * The generated PSK for TLS shall be computed applying the HMAC function using the
+ * hash function H( ) selected by the HashID parameter in the DH-HMAC-CHAP_Challenge
+ * message with the session key KS as key to the concatenation of the two challenges
+ * C1 and C2 (i.e., generated PSK = HMAC(KS, C1 || C2)).
+ */
+int nvme_auth_generate_psk(u8 hmac_id, u8 *skey, size_t skey_len,
+		u8 *c1, u8 *c2, size_t hash_len, u8 **ret_psk,size_t *ret_len)
+{
+	struct crypto_shash *tfm;
+	struct shash_desc *shash;
+	u8 *psk;
+	const char *hmac_name;
+	int ret, psk_len;
+
+	if (!c1 || !c2) {
+		pr_warn("%s: invalid parameter\n", __func__);
+		return -EINVAL;
+	}
+
+	hmac_name = nvme_auth_hmac_name(hmac_id);
+	if (!hmac_name) {
+		pr_warn("%s: invalid hash algoritm %d\n",
+			__func__, hmac_id);
+		return -EINVAL;
+	}
+
+	tfm = crypto_alloc_shash(hmac_name, 0, 0);
+	if (IS_ERR(tfm))
+		return PTR_ERR(tfm);
+
+	psk_len = crypto_shash_digestsize(tfm);
+	psk = kzalloc(psk_len, GFP_KERNEL);
+	if (!psk) {
+		ret = -ENOMEM;
+		goto out_free_tfm;
+	}
+
+	shash = kmalloc(sizeof(struct shash_desc) +
+			crypto_shash_descsize(tfm),
+			GFP_KERNEL);
+	if (!shash) {
+		ret = -ENOMEM;
+		goto out_free_psk;
+	}
+
+	shash->tfm = tfm;
+	ret = crypto_shash_setkey(tfm, skey, skey_len);
+	if (ret)
+		goto out_free_shash;
+
+	ret = crypto_shash_init(shash);
+	if (ret)
+		goto out_free_shash;
+
+	ret = crypto_shash_update(shash, c1, hash_len);
+	if (ret)
+		goto out_free_shash;
+
+	ret = crypto_shash_update(shash, c2, hash_len);
+	if (ret)
+		goto out_free_shash;
+
+	ret = crypto_shash_final(shash, psk);
+	if (!ret) {
+		*ret_psk = psk;
+		*ret_len = psk_len;
+	}
+
+out_free_shash:
+	kfree_sensitive(shash);
+out_free_psk:
+	if (ret)
+		kfree_sensitive(psk);
+out_free_tfm:
+	crypto_free_shash(tfm);
+
+	return ret;
+}
+EXPORT_SYMBOL_GPL(nvme_auth_generate_psk);
+
 MODULE_DESCRIPTION("NVMe Authentication framework");
 MODULE_LICENSE("GPL v2");
include/linux/nvme-auth.h +3 -0
index c1d0bc5d9624..b13884b04dfd 100644
@@ -40,5 +40,8 @@ int nvme_auth_gen_pubkey(struct crypto_kpp *dh_tfm,
 int nvme_auth_gen_shared_secret(struct crypto_kpp *dh_tfm,
 				u8 *ctrl_key, size_t ctrl_key_len,
 				u8 *sess_key, size_t sess_key_len);
+int nvme_auth_generate_psk(u8 hmac_id, u8 *skey, size_t skey_len,
+			   u8 *c1, u8 *c2, size_t hash_len,
+			   u8 **ret_psk, size_t *ret_len);
 
 #endif /* _NVME_AUTH_H */
Signature · 1 line
2.35.3
#5 lore raw

PATCH v8 3/9 nvme: add nvme_auth_generate_digest()

In reply to Hannes Reinecke

Add a function to calculate the PSK digest as specified in TP8018.

---
 drivers/nvme/common/auth.c | 126 +++++++++++++++++++++++++++++++++++++
 include/linux/nvme-auth.h  |   2 +
 2 files changed, 128 insertions(+)
drivers/nvme/common/auth.c +126 -0
index f6d21960b140..7e40f205d3e4 100644
@@ -558,5 +558,131 @@ int nvme_auth_generate_psk(u8 hmac_id, u8 *skey, size_t skey_len,
 }
 EXPORT_SYMBOL_GPL(nvme_auth_generate_psk);
 
+/*
+ * Generate a TLS PSK digest as specified in TP8018 Section 3.6.1.3:
+ *   TLS PSK and PSK identity Derivation
+ *
+ * The PSK digest shall be computed by encoding in Base64 (refer to RFC 4648)
+ * the result of the application of the HMAC function using the hash function
+ * specified in item 4 above (ie the hash function of the cipher suite associated
+ * with the PSK identity) with the PSK as HMAC key to the concatenation of:
+ * - the NQN of the host (i.e., NQNh) not including the null terminator;
+ * - a space character;
+ * - the NQN of the NVM subsystem (i.e., NQNc) not including the null terminator;
+ * - a space character; and
+ * - the seventeen ASCII characters "NVMe-over-Fabrics"
+ * (i.e., <PSK digest> = Base64(HMAC(PSK, NQNh || " " || NQNc || " " || "NVMe-over-Fabrics"))).
+ * The length of the PSK digest depends on the hash function used to compute
+ * it as follows:
+ * - If the SHA-256 hash function is used, the resulting PSK digest is 44 characters long; or
+ * - If the SHA-384 hash function is used, the resulting PSK digest is 64 characters long.
+ */
+int nvme_auth_generate_digest(u8 hmac_id, u8 *psk, size_t psk_len,
+		char *subsysnqn, char *hostnqn, u8 **ret_digest)
+{
+	struct crypto_shash *tfm;
+	struct shash_desc *shash;
+	u8 *digest, *hmac;
+	const char *hmac_name;
+	size_t digest_len, hmac_len;
+	int ret;
+
+	if (WARN_ON(!subsysnqn || !hostnqn))
+		return -EINVAL;
+
+	hmac_name = nvme_auth_hmac_name(hmac_id);
+	if (!hmac_name) {
+		pr_warn("%s: invalid hash algoritm %d\n",
+			__func__, hmac_id);
+		return -EINVAL;
+	}
+
+	switch (nvme_auth_hmac_hash_len(hmac_id)) {
+	case 32:
+		hmac_len = 44;
+		break;
+	case 48:
+		hmac_len = 64;
+		break;
+	default:
+		pr_warn("%s: invalid hash algorithm '%s'\n",
+			__func__, hmac_name);
+		return -EINVAL;
+	}
+
+	hmac = kzalloc(hmac_len + 1, GFP_KERNEL);
+	if (!hmac)
+		return -ENOMEM;
+
+	tfm = crypto_alloc_shash(hmac_name, 0, 0);
+	if (IS_ERR(tfm)) {
+		ret = PTR_ERR(tfm);
+		goto out_free_hmac;
+	}
+
+	digest_len = crypto_shash_digestsize(tfm);
+	digest = kzalloc(digest_len, GFP_KERNEL);
+	if (!digest) {
+		ret = -ENOMEM;
+		goto out_free_tfm;
+	}
+
+	shash = kmalloc(sizeof(struct shash_desc) +
+			crypto_shash_descsize(tfm),
+			GFP_KERNEL);
+	if (!shash) {
+		ret = -ENOMEM;
+		goto out_free_digest;
+	}
+
+	shash->tfm = tfm;
+	ret = crypto_shash_setkey(tfm, psk, psk_len);
+	if (ret)
+		goto out_free_shash;
+
+	ret = crypto_shash_init(shash);
+	if (ret)
+		goto out_free_shash;
+
+	ret = crypto_shash_update(shash, hostnqn, strlen(hostnqn));
+	if (ret)
+		goto out_free_shash;
+
+	ret = crypto_shash_update(shash, " ", 1);
+	if (ret)
+		goto out_free_shash;
+
+	ret = crypto_shash_update(shash, subsysnqn, strlen(subsysnqn));
+	if (ret)
+		goto out_free_shash;
+
+	ret = crypto_shash_update(shash, " NVMe-over-Fabrics", 18);
+	if (ret)
+		goto out_free_shash;
+
+	ret = crypto_shash_final(shash, digest);
+	if (ret)
+		goto out_free_shash;
+
+	ret = base64_encode(digest, digest_len, hmac);
+	if (ret < hmac_len)
+		ret = -ENOKEY;
+	*ret_digest = hmac;
+	ret = 0;
+
+out_free_shash:
+	kfree_sensitive(shash);
+out_free_digest:
+	kfree_sensitive(digest);
+out_free_tfm:
+	crypto_free_shash(tfm);
+out_free_hmac:
+	if (ret)
+		kfree_sensitive(hmac);
+
+	return ret;
+}
+EXPORT_SYMBOL_GPL(nvme_auth_generate_digest);
+
 MODULE_DESCRIPTION("NVMe Authentication framework");
 MODULE_LICENSE("GPL v2");
include/linux/nvme-auth.h +2 -0
index b13884b04dfd..998f06bf10fd 100644
@@ -43,5 +43,7 @@ int nvme_auth_gen_shared_secret(struct crypto_kpp *dh_tfm,
 int nvme_auth_generate_psk(u8 hmac_id, u8 *skey, size_t skey_len,
 			   u8 *c1, u8 *c2, size_t hash_len,
 			   u8 **ret_psk, size_t *ret_len);
+int nvme_auth_generate_digest(u8 hmac_id, u8 *psk, size_t psk_len,
+		char *subsysnqn, char *hostnqn, u8 **ret_digest);
 
 #endif /* _NVME_AUTH_H */
Signature · 1 line
2.35.3
#6 lore raw

PATCH v8 4/9 nvme: add nvme_auth_derive_tls_psk()

In reply to Hannes Reinecke

Add a function to derive the TLS PSK as specified TP8018.

---
 drivers/nvme/common/auth.c | 90 ++++++++++++++++++++++++++++++++++++++
 include/linux/nvme-auth.h  |  2 +
 2 files changed, 92 insertions(+)
drivers/nvme/common/auth.c +90 -0
index 7e40f205d3e4..0b000a562c0f 100644
@@ -684,5 +684,95 @@ int nvme_auth_generate_digest(u8 hmac_id, u8 *psk, size_t psk_len,
 }
 EXPORT_SYMBOL_GPL(nvme_auth_generate_digest);
 
+/*
+ * Derive a TLS PSK as specified in TP8018 Section 3.6.1.3:
+ *   TLS PSK and PSK identity Derivation
+ *
+ * The TLS PSK shall be derived as follows from an input PSK
+ * (i.e., either a retained PSK or a generated PSK) and a PSK
+ * identity using the HKDF-Extract and HKDF-Expand-Label operations
+ * (refer to RFC 5869 and RFC 8446) where the hash function is the
+ * one specified by the hash specifier of the PSK identity:
+ * 1. PRK = HKDF-Extract(0, Input PSK); and
+ * 2. TLS PSK = HKDF-Expand-Label(PRK, "nvme-tls-psk", PskIdentityContext, L),
+ * where PskIdentityContext is the hash identifier indicated in
+ * the PSK identity concatenated to a space character and to the
+ * Base64 PSK digest (i.e., "<hash> <PSK digest>") and L is the
+ * output size in bytes of the hash function (i.e., 32 for SHA-256
+ * and 48 for SHA-384).
+ */
+int nvme_auth_derive_tls_psk(int hmac_id, u8 *psk, size_t psk_len,
+		u8 *psk_digest, u8 **ret_psk)
+{
+	struct crypto_shash *hmac_tfm;
+	const char *hmac_name;
+	const char *psk_prefix = "tls13 nvme-tls-psk";
+	size_t info_len, prk_len;
+	char *info;
+	unsigned char *prk, *tls_key;
+	int ret;
+
+	hmac_name = nvme_auth_hmac_name(hmac_id);
+	if (!hmac_name) {
+		pr_warn("%s: invalid hash algoritm %d\n",
+			__func__, hmac_id);
+		return -EINVAL;
+	}
+	if (hmac_id == NVME_AUTH_HASH_SHA512) {
+		pr_warn("%s: unsupported hash algorithm %s\n",
+			__func__, hmac_name);
+		return -EINVAL;
+	}
+
+	hmac_tfm = crypto_alloc_shash(hmac_name, 0, 0);
+	if (IS_ERR(hmac_tfm))
+		return PTR_ERR(hmac_tfm);
+
+	prk_len = crypto_shash_digestsize(hmac_tfm);
+	prk = kzalloc(prk_len, GFP_KERNEL);
+	if (!prk) {
+		ret = -ENOMEM;
+		goto out_free_shash;
+	}
+
+	ret = hkdf_extract(hmac_tfm, psk, psk_len, prk);
+	if (ret)
+		goto out_free_prk;
+
+	ret = crypto_shash_setkey(hmac_tfm, prk, prk_len);
+	if (ret)
+		goto out_free_prk;
+
+	info_len = strlen(psk_digest) + strlen(psk_prefix) + 1;
+	info = kzalloc(info_len, GFP_KERNEL);
+	if (!info)
+		goto out_free_prk;
+
+	memcpy(info, psk_prefix, strlen(psk_prefix));
+	memcpy(info + strlen(psk_prefix), psk_digest, strlen(psk_digest));
+
+	tls_key = kzalloc(psk_len, GFP_KERNEL);
+	if (!tls_key) {
+		ret = -ENOMEM;
+		goto out_free_info;
+	}
+	ret = hkdf_expand(hmac_tfm, info, strlen(info), tls_key, psk_len);
+	if (ret) {
+		kfree(tls_key);
+		goto out_free_info;
+	}
+	*ret_psk = tls_key;
+
+out_free_info:
+	kfree(info);
+out_free_prk:
+	kfree(prk);
+out_free_shash:
+	crypto_free_shash(hmac_tfm);
+
+	return ret;
+}
+EXPORT_SYMBOL_GPL(nvme_auth_derive_tls_psk);
+
 MODULE_DESCRIPTION("NVMe Authentication framework");
 MODULE_LICENSE("GPL v2");
include/linux/nvme-auth.h +2 -0
index 998f06bf10fd..60e069a6757f 100644
@@ -45,5 +45,7 @@ int nvme_auth_generate_psk(u8 hmac_id, u8 *skey, size_t skey_len,
 			   u8 **ret_psk, size_t *ret_len);
 int nvme_auth_generate_digest(u8 hmac_id, u8 *psk, size_t psk_len,
 		char *subsysnqn, char *hostnqn, u8 **ret_digest);
+int nvme_auth_derive_tls_psk(int hmac_id, u8 *psk, size_t psk_len,
+		u8 *psk_digest, u8 **ret_psk);
 
 #endif /* _NVME_AUTH_H */
Signature · 1 line
2.35.3
#9 lore raw

PATCH v8 5/9 nvme-keyring: add nvme_tls_psk_refresh()

In reply to Hannes Reinecke

Add a function to refresh a generated PSK in the specified keyring.

---
 drivers/nvme/common/keyring.c | 50 +++++++++++++++++++++++++++++++++++
 include/linux/nvme-keyring.h  |  7 +++++
 2 files changed, 57 insertions(+)
drivers/nvme/common/keyring.c +50 -0
index ed5167f942d8..a8d378f41fcc 100644
@@ -124,6 +124,56 @@ static struct key *nvme_tls_psk_lookup(struct key *keyring,
 	return key_ref_to_ptr(keyref);
 }
 
+struct key *nvme_tls_psk_refresh(struct key *keyring, char *hostnqn, char *subnqn,
+		u8 hmac_id, bool generated, u8 *data, size_t data_len, char *digest)
+{
+	key_perm_t keyperm =
+		KEY_POS_SEARCH | KEY_POS_VIEW | KEY_POS_READ |
+		KEY_POS_WRITE | KEY_POS_LINK | KEY_POS_SETATTR |
+		KEY_USR_SEARCH | KEY_USR_VIEW | KEY_USR_READ;
+	char *identity;
+	size_t identity_len = (NVMF_NQN_SIZE) * 2 + 77;
+	key_ref_t keyref;
+	key_serial_t keyring_id;
+	struct key *key;
+
+	if (!hostnqn || !subnqn || !data || !data_len)
+		return ERR_PTR(-EINVAL);
+
+	identity = kzalloc(identity_len, GFP_KERNEL);
+	if (!identity)
+		return ERR_PTR(-ENOMEM);
+
+	snprintf(identity, identity_len, "NVMe1%c%02d %s %s %s",
+		 generated ? 'G' : 'R', hmac_id, hostnqn, subnqn, digest);
+
+	if (!keyring)
+		keyring = nvme_keyring;
+	keyring_id = key_serial(keyring);
+	pr_debug("keyring %x refresh tls psk '%s'\n",
+		 keyring_id, identity);
+	keyref = key_create_or_update(make_key_ref(keyring, true),
+				"psk", identity, data, data_len,
+				keyperm, KEY_ALLOC_NOT_IN_QUOTA |
+				      KEY_ALLOC_BUILT_IN |
+				      KEY_ALLOC_BYPASS_RESTRICTION);
+	if (IS_ERR(keyref)) {
+		pr_debug("refresh tls psk '%s' failed, error %ld\n",
+			 identity, PTR_ERR(keyref));
+		kfree(identity);
+		return ERR_PTR(-ENOKEY);
+	}
+	kfree(identity);
+	/*
+	 * Set the default timeout to 1 hour
+	 * as suggested in TP8018.
+	 */
+	key = key_ref_to_ptr(keyref);
+	key_set_timeout(key, 3600);
+	return key;
+}
+EXPORT_SYMBOL_GPL(nvme_tls_psk_refresh);
+
 /*
  * NVMe PSK priority list
  *
include/linux/nvme-keyring.h +7 -0
index 19d2b256180f..84824facc1f8 100644
@@ -8,6 +8,8 @@
 
 #if IS_ENABLED(CONFIG_NVME_KEYRING)
 
+struct key *nvme_tls_psk_refresh(struct key *keyring, char *hostnqn, char *subnqn,
+				 u8 hmac_id, bool generated, u8 *data, size_t data_len, char *digest);
 key_serial_t nvme_tls_psk_default(struct key *keyring,
 		const char *hostnqn, const char *subnqn);
 
@@ -15,6 +17,11 @@ key_serial_t nvme_keyring_id(void);
 struct key *nvme_tls_key_lookup(key_serial_t key_id);
 #else
 
+static struct key *nvme_tls_psk_refresh(struct key *keyring, char *hostnqn, char *subnqn,
+		u8 hmac_id, bool generated, u8 *data, size_t data_len, char *digest)
+{
+	return ERR_PTR(-ENOTSUPP);
+}
 static inline key_serial_t nvme_tls_psk_default(struct key *keyring,
 		const char *hostnqn, const char *subnqn)
 {
Signature · 1 line
2.35.3
#11 lore raw

PATCH v8 6/9 nvme-tcp: request secure channel concatenation

In reply to Hannes Reinecke

Add a fabrics option 'concat' to request secure channel concatenation. When secure channel concatenation is enabled a 'generated PSK' is inserted into the keyring such that it's available after reset.

---
 drivers/nvme/host/auth.c    | 108 +++++++++++++++++++++++++++++++++++-
 drivers/nvme/host/fabrics.c |  34 +++++++++++-
 drivers/nvme/host/fabrics.h |   3 +
 drivers/nvme/host/nvme.h    |   2 +
 drivers/nvme/host/sysfs.c   |   4 +-
 drivers/nvme/host/tcp.c     |  47 ++++++++++++++--
 include/linux/nvme.h        |   7 +++
 7 files changed, 191 insertions(+), 14 deletions(-)
drivers/nvme/host/auth.c +105 -3
index 371e14f0a203..f7292c49e66c 100644
@@ -12,6 +12,7 @@
 #include "nvme.h"
 #include "fabrics.h"
 #include <linux/nvme-auth.h>
+#include <linux/nvme-keyring.h>
 
 #define CHAP_BUF_SIZE 4096
 static struct kmem_cache *nvme_chap_buf_cache;
@@ -131,7 +132,12 @@ static int nvme_auth_set_dhchap_negotiate_data(struct nvme_ctrl *ctrl,
 	data->auth_type = NVME_AUTH_COMMON_MESSAGES;
 	data->auth_id = NVME_AUTH_DHCHAP_MESSAGE_NEGOTIATE;
 	data->t_id = cpu_to_le16(chap->transaction);
-	data->sc_c = 0; /* No secure channel concatenation */
+	if (!ctrl->opts->concat || chap->qid != 0)
+		data->sc_c = NVME_AUTH_SECP_NOSC;
+	else if (ctrl->opts->tls_key)
+		data->sc_c = NVME_AUTH_SECP_REPLACETLSPSK;
+	else
+		data->sc_c = NVME_AUTH_SECP_NEWTLSPSK;
 	data->napd = 1;
 	data->auth_protocol[0].dhchap.authid = NVME_AUTH_DHCHAP_AUTH_ID;
 	data->auth_protocol[0].dhchap.halen = 3;
@@ -311,8 +317,9 @@ static int nvme_auth_set_dhchap_reply_data(struct nvme_ctrl *ctrl,
 	data->hl = chap->hash_len;
 	data->dhvlen = cpu_to_le16(chap->host_key_len);
 	memcpy(data->rval, chap->response, chap->hash_len);
-	if (ctrl->ctrl_key) {
+	if (ctrl->ctrl_key)
 		chap->bi_directional = true;
+	if (ctrl->ctrl_key || ctrl->opts->concat) {
 		get_random_bytes(chap->c2, chap->hash_len);
 		data->cvalid = 1;
 		memcpy(data->rval + chap->hash_len, chap->c2,
@@ -322,7 +329,10 @@ static int nvme_auth_set_dhchap_reply_data(struct nvme_ctrl *ctrl,
 	} else {
 		memset(chap->c2, 0, chap->hash_len);
 	}
-	chap->s2 = nvme_auth_get_seqnum();
+	if (ctrl->opts->concat)
+		chap->s2 = 0;
+	else
+		chap->s2 = nvme_auth_get_seqnum();
 	data->seqnum = cpu_to_le32(chap->s2);
 	if (chap->host_key_len) {
 		dev_dbg(ctrl->device, "%s: qid %d host public key %*ph\n",
@@ -677,6 +687,84 @@ static void nvme_auth_free_dhchap(struct nvme_dhchap_queue_context *chap)
 		crypto_free_kpp(chap->dh_tfm);
 }
 
+void nvme_auth_revoke_tls_key(struct nvme_ctrl *ctrl)
+{
+	dev_dbg(ctrl->device, "Wipe generated TLS PSK %08x\n",
+		key_serial(ctrl->opts->tls_key));
+	key_revoke(ctrl->opts->tls_key);
+	key_put(ctrl->opts->tls_key);
+	ctrl->opts->tls_key = NULL;
+}
+EXPORT_SYMBOL_GPL(nvme_auth_revoke_tls_key);
+
+static int nvme_auth_secure_concat(struct nvme_ctrl *ctrl,
+				   struct nvme_dhchap_queue_context *chap)
+{
+	u8 *psk, *digest, *tls_psk;
+	struct key *tls_key;
+	size_t psk_len;
+	int ret = 0;
+
+	if (!chap->sess_key) {
+		dev_warn(ctrl->device,
+			 "%s: qid %d no session key negotiated\n",
+			 __func__, chap->qid);
+		return -ENOKEY;
+	}
+
+	ret = nvme_auth_generate_psk(chap->hash_id, chap->sess_key,
+				     chap->sess_key_len,
+				     chap->c1, chap->c2,
+				     chap->hash_len, &psk, &psk_len);
+	if (ret) {
+		dev_warn(ctrl->device,
+			 "%s: qid %d failed to generate PSK, error %d\n",
+			 __func__, chap->qid, ret);
+		return ret;
+	}
+	dev_dbg(ctrl->device,
+		  "%s: generated psk %*ph\n", __func__, (int)psk_len, psk);
+
+	ret = nvme_auth_generate_digest(chap->hash_id, psk, psk_len,
+					ctrl->opts->subsysnqn,
+					ctrl->opts->host->nqn, &digest);
+	if (ret) {
+		dev_warn(ctrl->device,
+			 "%s: qid %d failed to generate digest, error %d\n",
+			 __func__, chap->qid, ret);
+		goto out_free_psk;
+	};
+	dev_dbg(ctrl->device, "%s: generated digest %s\n",
+		 __func__, digest);
+	ret = nvme_auth_derive_tls_psk(chap->hash_id, psk, psk_len, digest, &tls_psk);
+	if (ret) {
+		dev_warn(ctrl->device,
+			 "%s: qid %d failed to derive TLS psk, error %d\n",
+			 __func__, chap->qid, ret);
+		goto out_free_digest;
+	};
+
+	tls_key = nvme_tls_psk_refresh(ctrl->opts->keyring, ctrl->opts->host->nqn,
+				       ctrl->opts->subsysnqn, chap->hash_id,
+				       true, tls_psk, psk_len, digest);
+	if (IS_ERR(tls_key)) {
+		ret = PTR_ERR(tls_key);
+		dev_warn(ctrl->device,
+			 "%s: qid %d failed to insert generated key, error %d\n",
+			 __func__, chap->qid, ret);
+		tls_key = NULL;
+		kfree_sensitive(tls_psk);
+	}
+	if (ctrl->opts->tls_key)
+		nvme_auth_revoke_tls_key(ctrl);
+	ctrl->opts->tls_key = tls_key;
+out_free_digest:
+	kfree_sensitive(digest);
+out_free_psk:
+	kfree_sensitive(psk);
+	return ret;
+}
+
 static void nvme_queue_auth_work(struct work_struct *work)
 {
 	struct nvme_dhchap_queue_context *chap =
@@ -833,6 +921,14 @@ static void nvme_queue_auth_work(struct work_struct *work)
 	}
 	if (!ret) {
 		chap->error = 0;
+		/* Secure concatenation can only be enabled on the admin queue */
+		if (!chap->qid && ctrl->opts->concat &&
+		    (ret = nvme_auth_secure_concat(ctrl, chap))) {
+			dev_warn(ctrl->device,
+				 "%s: qid %d failed to enable secure concatenation\n",
+				 __func__, chap->qid);
+			chap->error = ret;
+		}
 		return;
 	}
 
@@ -912,6 +1008,12 @@ static void nvme_ctrl_auth_work(struct work_struct *work)
 			 "qid 0: authentication failed\n");
 		return;
 	}
+	/*
+	 * Only run authentication on the admin queue for
+	 * secure concatenation
+	 */
+	if (ctrl->opts->concat)
+		return;
 
 	for (q = 1; q < ctrl->queue_count; q++) {
 		ret = nvme_auth_negotiate(ctrl, q);
drivers/nvme/host/fabrics.c +31 -3
index 432efcbf9e2f..93e9041b9657 100644
@@ -472,8 +472,9 @@ int nvmf_connect_admin_queue(struct nvme_ctrl *ctrl)
 	result = le32_to_cpu(res.u32);
 	ctrl->cntlid = result & 0xFFFF;
 	if (result & (NVME_CONNECT_AUTHREQ_ATR | NVME_CONNECT_AUTHREQ_ASCR)) {
-		/* Secure concatenation is not implemented */
-		if (result & NVME_CONNECT_AUTHREQ_ASCR) {
+		/* Check for secure concatenation */
+		if ((result & NVME_CONNECT_AUTHREQ_ASCR) &&
+		    !ctrl->opts->concat) {
 			dev_warn(ctrl->device,
 				 "qid 0: secure concatenation is not supported\n");
 			ret = -EOPNOTSUPP;
@@ -550,7 +551,7 @@ int nvmf_connect_io_queue(struct nvme_ctrl *ctrl, u16 qid)
 		/* Secure concatenation is not implemented */
 		if (result & NVME_CONNECT_AUTHREQ_ASCR) {
 			dev_warn(ctrl->device,
-				 "qid 0: secure concatenation is not supported\n");
+				 "qid %d: secure concatenation is not supported\n", qid);
 			ret = -EOPNOTSUPP;
 			goto out_free_data;
 		}
@@ -706,6 +707,7 @@ static const match_table_t opt_tokens = {
 #endif
 #ifdef CONFIG_NVME_TCP_TLS
 	{ NVMF_OPT_TLS,			"tls"			},
+	{ NVMF_OPT_CONCAT,		"concat"		},
 #endif
 	{ NVMF_OPT_ERR,			NULL			}
 };
@@ -735,6 +737,7 @@ static int nvmf_parse_options(struct nvmf_ctrl_options *opts,
 	opts->tls = false;
 	opts->tls_key = NULL;
 	opts->keyring = NULL;
+	opts->concat = false;
 
 	options = o = kstrdup(buf, GFP_KERNEL);
 	if (!options)
@@ -1053,6 +1056,14 @@ static int nvmf_parse_options(struct nvmf_ctrl_options *opts,
 			}
 			opts->tls = true;
 			break;
+		case NVMF_OPT_CONCAT:
+			if (!IS_ENABLED(CONFIG_NVME_TCP_TLS)) {
+				pr_err("TLS is not supported\n");
+				ret = -EINVAL;
+				goto out;
+			}
+			opts->concat = true;
+			break;
 		default:
 			pr_warn("unknown parameter or missing value '%s' in ctrl creation request\n",
 				p);
@@ -1079,6 +1090,23 @@ static int nvmf_parse_options(struct nvmf_ctrl_options *opts,
 			pr_warn("failfast tmo (%d) larger than controller loss tmo (%d)\n",
 				opts->fast_io_fail_tmo, ctrl_loss_tmo);
 	}
+	if (opts->concat) {
+		if (opts->tls) {
+			pr_err("Secure concatenation over TLS is not supported\n");
+			ret = -EINVAL;
+			goto out;
+		}
+		if (opts->tls_key) {
+			pr_err("Cannot specify a TLS key for secure concatenation\n");
+			ret = -EINVAL;
+			goto out;
+		}
+		if (!opts->dhchap_secret) {
+			pr_err("Need to enable DH-CHAP for secure concatenation\n");
+			ret = -EINVAL;
+			goto out;
+		}
+	}
 
 	opts->host = nvmf_host_add(hostnqn, &hostid);
 	if (IS_ERR(opts->host)) {
drivers/nvme/host/fabrics.h +3 -0
index 21d75dc4a3a0..9cf5b020adba 100644
@@ -66,6 +66,7 @@ enum {
 	NVMF_OPT_TLS		= 1 << 25,
 	NVMF_OPT_KEYRING	= 1 << 26,
 	NVMF_OPT_TLS_KEY	= 1 << 27,
+	NVMF_OPT_CONCAT		= 1 << 28,
 };
 
 /**
@@ -101,6 +102,7 @@ enum {
  * @keyring:    Keyring to use for key lookups
  * @tls_key:    TLS key for encrypted connections (TCP)
  * @tls:        Start TLS encrypted connections (TCP)
+ * @concat:     Enabled Secure channel concatenation (TCP)
  * @disable_sqflow: disable controller sq flow control
  * @hdr_digest: generate/verify header digest (TCP)
  * @data_digest: generate/verify data digest (TCP)
@@ -130,6 +132,7 @@ struct nvmf_ctrl_options {
 	struct key		*keyring;
 	struct key		*tls_key;
 	bool			tls;
+	bool			concat;
 	bool			disable_sqflow;
 	bool			hdr_digest;
 	bool			data_digest;
drivers/nvme/host/nvme.h +2 -0
index 185fb986a5cc..08436c8915d8 100644
@@ -1129,6 +1129,7 @@ void nvme_auth_stop(struct nvme_ctrl *ctrl);
 int nvme_auth_negotiate(struct nvme_ctrl *ctrl, int qid);
 int nvme_auth_wait(struct nvme_ctrl *ctrl, int qid);
 void nvme_auth_free(struct nvme_ctrl *ctrl);
+void nvme_auth_revoke_tls_key(struct nvme_ctrl *ctrl);
 #else
 static inline int nvme_auth_init_ctrl(struct nvme_ctrl *ctrl)
 {
@@ -1151,6 +1152,7 @@ static inline int nvme_auth_wait(struct nvme_ctrl *ctrl, int qid)
 	return -EPROTONOSUPPORT;
 }
 static inline void nvme_auth_free(struct nvme_ctrl *ctrl) {};
+static void nvme_auth_revoke_tls_key(struct nvme_ctrl *ctrl) {};
 #endif
 
 u32 nvme_command_effects(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
drivers/nvme/host/sysfs.c +2 -2
index eb345551d6fe..32a621f75821 100644
@@ -779,10 +779,10 @@ static umode_t nvme_tls_attrs_are_visible(struct kobject *kobj,
 		return 0;
 
 	if (a == &dev_attr_tls_key.attr &&
-	    !ctrl->opts->tls)
+	    !ctrl->opts->tls && !ctrl->opts->concat)
 		return 0;
 	if (a == &dev_attr_tls_configured_key.attr &&
-	    !ctrl->opts->tls_key)
+	    (!ctrl->opts->tls_key || ctrl->opts->concat))
 		return 0;
 	if (a == &dev_attr_tls_keyring.attr &&
 	    !ctrl->opts->keyring)
drivers/nvme/host/tcp.c +41 -6

89 lines not on this page · raw message

include/linux/nvme.h +7 -0
index 5c2290f0d5af..5626894d7700 100644
@@ -1676,6 +1676,13 @@ enum {
 	NVME_AUTH_DHGROUP_INVALID	= 0xff,
 };
 
+enum {
+	NVME_AUTH_SECP_NOSC		= 0x00,
+	NVME_AUTH_SECP_SC		= 0x01,
+	NVME_AUTH_SECP_NEWTLSPSK	= 0x02,
+	NVME_AUTH_SECP_REPLACETLSPSK	= 0x03,
+};
+
 union nvmf_auth_protocol {
 	struct nvmf_auth_dhchap_protocol_descriptor dhchap;
 };
Signature · 1 line
2.35.3
#12 lore raw

PATCH v8 7/9 nvme-fabrics: reset admin connection for secure concatenation

In reply to Hannes Reinecke

When secure concatenation is requested the connection needs to be reset to enable TLS encryption on the new cnnection. That implies that the original connection used for the DH-CHAP negotiation really shouldn't be used, and we should reset as soon as the DH-CHAP negotiation has succeeded on the admin queue.

Based on an idea from Sagi.

---
 drivers/nvme/host/tcp.c | 9 +++++++++
 1 file changed, 9 insertions(+)
drivers/nvme/host/tcp.c +9 -0
index b229c4402482..76c8e01d8f08 100644
@@ -2251,6 +2251,15 @@ static int nvme_tcp_setup_ctrl(struct nvme_ctrl *ctrl, bool new)
 	if (ret)
 		return ret;
 
+	if (new && ctrl->opts && ctrl->opts->concat && !ctrl->tls_pskid) {
+		dev_dbg(ctrl->device, "restart admin queue for secure concatenation\n");
+		nvme_stop_keep_alive(ctrl);
+		nvme_tcp_teardown_admin_queue(ctrl, false);
+		ret = nvme_tcp_configure_admin_queue(ctrl, false);
+		if (ret)
+			return ret;
+	}
+
 	if (ctrl->icdoff) {
 		ret = -EOPNOTSUPP;
 		dev_err(ctrl->device, "icdoff is not supported!\n");
Signature · 1 line
2.35.3
#13 lore raw

PATCH v8 8/9 nvmet-tcp: support secure channel concatenation

In reply to Hannes Reinecke

Evaluate the SC_C flag during DH-CHAP-HMAC negotiation and insert the generated PSK once negotiation has finished.

---
 drivers/nvme/target/auth.c             | 72 +++++++++++++++++++++++++-
 drivers/nvme/target/fabrics-cmd-auth.c | 49 +++++++++++++++---
 drivers/nvme/target/fabrics-cmd.c      | 33 +++++++++---
 drivers/nvme/target/nvmet.h            | 38 +++++++++++---
 drivers/nvme/target/tcp.c              | 23 +++++++-
 5 files changed, 192 insertions(+), 23 deletions(-)
drivers/nvme/target/auth.c +71 -1
index 7897d02c681d..560321df5bf6 100644
@@ -15,6 +15,7 @@
 #include <linux/ctype.h>
 #include <linux/random.h>
 #include <linux/nvme-auth.h>
+#include <linux/nvme-keyring.h>
 #include <asm/unaligned.h>
 
 #include "nvmet.h"
@@ -138,7 +139,7 @@ int nvmet_setup_dhgroup(struct nvmet_ctrl *ctrl, u8 dhgroup_id)
 	return ret;
 }
 
-u8 nvmet_setup_auth(struct nvmet_ctrl *ctrl)
+u8 nvmet_setup_auth(struct nvmet_ctrl *ctrl, struct nvmet_req *req)
 {
 	int ret = 0;
 	struct nvmet_host_link *p;
@@ -164,6 +165,11 @@ u8 nvmet_setup_auth(struct nvmet_ctrl *ctrl)
 		goto out_unlock;
 	}
 
+	if (nvmet_queue_tls_keyid(req->sq)) {
+		pr_debug("host %s tls enabled\n", ctrl->hostnqn);
+		goto out_unlock;
+	}
+
 	ret = nvmet_setup_dhgroup(ctrl, host->dhchap_dhgroup_id);
 	if (ret < 0) {
 		pr_warn("Failed to setup DH group");
@@ -232,6 +238,9 @@ u8 nvmet_setup_auth(struct nvmet_ctrl *ctrl)
 void nvmet_auth_sq_free(struct nvmet_sq *sq)
 {
 	cancel_delayed_work(&sq->auth_expired_work);
+#ifdef CONFIG_NVME_TARGET_TCP_TLS
+	sq->tls_key = 0;
+#endif
 	kfree(sq->dhchap_c1);
 	sq->dhchap_c1 = NULL;
 	kfree(sq->dhchap_c2);
@@ -260,6 +269,12 @@ void nvmet_destroy_auth(struct nvmet_ctrl *ctrl)
 		nvme_auth_free_key(ctrl->ctrl_key);
 		ctrl->ctrl_key = NULL;
 	}
+#ifdef CONFIG_NVME_TARGET_TCP_TLS
+	if (ctrl->tls_key) {
+		key_put(ctrl->tls_key);
+		ctrl->tls_key = NULL;
+	}
+#endif
 }
 
 bool nvmet_check_auth_status(struct nvmet_req *req)
@@ -541,3 +556,58 @@ int nvmet_auth_ctrl_sesskey(struct nvmet_req *req,
 
 	return ret;
 }
+
+void nvmet_auth_insert_psk(struct nvmet_sq *sq)
+{
+	int hash_len = nvme_auth_hmac_hash_len(sq->ctrl->shash_id);
+	u8 *psk, *digest, *tls_psk;
+	size_t psk_len;
+	int ret;
+#ifdef CONFIG_NVME_TARGET_TCP_TLS
+	struct key *tls_key = NULL;
+#endif
+
+	ret = nvme_auth_generate_psk(sq->ctrl->shash_id,
+				     sq->dhchap_skey,
+				     sq->dhchap_skey_len,
+				     sq->dhchap_c1, sq->dhchap_c2,
+				     hash_len, &psk, &psk_len);
+	if (ret) {
+		pr_warn("%s: ctrl %d qid %d failed to generate PSK, error %d\n",
+			__func__, sq->ctrl->cntlid, sq->qid, ret);
+		return;
+	}
+	ret = nvme_auth_generate_digest(sq->ctrl->shash_id, psk, psk_len,
+					sq->ctrl->subsysnqn,
+					sq->ctrl->hostnqn, &digest);
+	if (ret) {
+		pr_warn("%s: ctrl %d qid %d failed to generate digest, error %d\n",
+			__func__, sq->ctrl->cntlid, sq->qid, ret);
+		goto out_free_psk;
+	}
+	ret = nvme_auth_derive_tls_psk(sq->ctrl->shash_id, psk, psk_len,
+				       digest, &tls_psk);
+	if (ret) {
+		pr_warn("%s: ctrl %d qid %d failed to derive TLS PSK, error %d\n",
+			__func__, sq->ctrl->cntlid, sq->qid, ret);
+		goto out_free_digest;
+	}
+#ifdef CONFIG_NVME_TARGET_TCP_TLS
+	tls_key = nvme_tls_psk_refresh(NULL, sq->ctrl->hostnqn, sq->ctrl->subsysnqn,
+				       sq->ctrl->shash_id, true, tls_psk, psk_len, digest);
+	if (IS_ERR(tls_key)) {
+		pr_warn("%s: ctrl %d qid %d failed to refresh key, error %ld\n",
+			__func__, sq->ctrl->cntlid, sq->qid, PTR_ERR(tls_key));
+		tls_key = NULL;
+		kfree_sensitive(tls_psk);
+	}
+	if (sq->ctrl->tls_key)
+		key_put(sq->ctrl->tls_key);
+	sq->ctrl->tls_key = tls_key;
+#endif
+
+out_free_digest:
+	kfree_sensitive(digest);
+out_free_psk:
+	kfree_sensitive(psk);
+}
drivers/nvme/target/fabrics-cmd-auth.c +43 -6

87 lines not on this page · raw message

drivers/nvme/target/fabrics-cmd.c +25 -8
index 344ab75d8864..306cbd6cf394 100644
@@ -199,10 +199,26 @@ static u16 nvmet_install_queue(struct nvmet_ctrl *ctrl, struct nvmet_req *req)
 	return ret;
 }
 
-static u32 nvmet_connect_result(struct nvmet_ctrl *ctrl)
+static u32 nvmet_connect_result(struct nvmet_ctrl *ctrl, struct nvmet_req *req)
 {
+	bool needs_auth = nvmet_has_auth(ctrl, req);
+	key_serial_t keyid = nvmet_queue_tls_keyid(req->sq);
+
+	/* Do not authenticate I/O queues for secure concatenation */
+	if (ctrl->concat && req->sq->qid)
+		needs_auth = false;
+
+	if (keyid)
+		pr_debug("%s: ctrl %d qid %d should %sauthenticate, tls psk %08x\n",
+			 __func__, ctrl->cntlid, req->sq->qid,
+			 needs_auth ? "" : "not ", keyid);
+	else
+		pr_debug("%s: ctrl %d qid %d should %sauthenticate%s\n",
+			 __func__, ctrl->cntlid, req->sq->qid,
+			 needs_auth ? "" : "not ",
+			 ctrl->concat ? ", secure concatenation" : "");
 	return (u32)ctrl->cntlid |
-		(nvmet_has_auth(ctrl) ? NVME_CONNECT_AUTHREQ_ATR : 0);
+		(needs_auth ? NVME_CONNECT_AUTHREQ_ATR : 0);
 }
 
 static void nvmet_execute_admin_connect(struct nvmet_req *req)
@@ -254,7 +270,7 @@ static void nvmet_execute_admin_connect(struct nvmet_req *req)
 
 	uuid_copy(&ctrl->hostid, &d->hostid);
 
-	dhchap_status = nvmet_setup_auth(ctrl);
+	dhchap_status = nvmet_setup_auth(ctrl, req);
 	if (dhchap_status) {
 		pr_err("Failed to setup authentication, dhchap status %u\n",
 		       dhchap_status);
@@ -272,12 +288,13 @@ static void nvmet_execute_admin_connect(struct nvmet_req *req)
 		goto out;
 	}
 
-	pr_info("creating %s controller %d for subsystem %s for NQN %s%s%s.\n",
+	pr_info("creating %s controller %d for subsystem %s for NQN %s%s%s%s.\n",
 		nvmet_is_disc_subsys(ctrl->subsys) ? "discovery" : "nvm",
 		ctrl->cntlid, ctrl->subsys->subsysnqn, ctrl->hostnqn,
-		ctrl->pi_support ? " T10-PI is enabled" : "",
-		nvmet_has_auth(ctrl) ? " with DH-HMAC-CHAP" : "");
-	req->cqe->result.u32 = cpu_to_le32(nvmet_connect_result(ctrl));
+		ctrl->pi_support ? ", T10-PI" : "",
+		nvmet_has_auth(ctrl, req) ? ", DH-HMAC-CHAP" : "",
+		nvmet_queue_tls_keyid(req->sq) ? ", TLS" : "");
+	req->cqe->result.u32 = cpu_to_le32(nvmet_connect_result(ctrl, req));
 out:
 	kfree(d);
 complete:
@@ -336,7 +353,7 @@ static void nvmet_execute_io_connect(struct nvmet_req *req)
 		goto out_ctrl_put;
 
 	pr_debug("adding queue %d to ctrl %d.\n", qid, ctrl->cntlid);
-	req->cqe->result.u32 = cpu_to_le32(nvmet_connect_result(ctrl));
+	req->cqe->result.u32 = cpu_to_le32(nvmet_connect_result(ctrl, req));
 out:
 	kfree(d);
 complete:
drivers/nvme/target/nvmet.h +32 -6
index 190f55e6d753..c2e17201c757 100644
@@ -121,6 +121,9 @@ struct nvmet_sq {
 	u32			dhchap_s2;
 	u8			*dhchap_skey;
 	int			dhchap_skey_len;
+#endif
+#ifdef CONFIG_NVME_TARGET_TCP_TLS
+	struct key		*tls_key;
 #endif
 	struct completion	free_done;
 	struct completion	confirm_done;
@@ -237,6 +240,7 @@ struct nvmet_ctrl {
 	u64			err_counter;
 	struct nvme_error_slot	slots[NVMET_ERROR_LOG_SLOTS];
 	bool			pi_support;
+	bool			concat;
 #ifdef CONFIG_NVME_TARGET_AUTH
 	struct nvme_dhchap_key	*host_key;
 	struct nvme_dhchap_key	*ctrl_key;
@@ -246,6 +250,9 @@ struct nvmet_ctrl {
 	u8			*dh_key;
 	size_t			dh_keysize;
 #endif
+#ifdef CONFIG_NVME_TARGET_TCP_TLS
+	struct key		*tls_key;
+#endif
 };
 
 struct nvmet_subsys {
@@ -716,13 +723,29 @@ static inline void nvmet_req_bio_put(struct nvmet_req *req, struct bio *bio)
 		bio_put(bio);
 }
 
+#ifdef CONFIG_NVME_TARGET_TCP_TLS
+static inline key_serial_t nvmet_queue_tls_keyid(struct nvmet_sq *sq)
+{
+	return sq->tls_key ? key_serial(sq->tls_key) : 0;
+}
+static inline void nvmet_sq_put_tls_key(struct nvmet_sq *sq)
+{
+	if (sq->tls_key) {
+		key_put(sq->tls_key);
+		sq->tls_key = NULL;
+	}
+}
+#else
+static inline key_serial_t nvmet_queue_tls_keyid(struct nvmet_sq *sq) { return 0; }
+static inline void nvmet_sq_put_tls_key(struct nvmet_sq *sq) {}
+#endif
 #ifdef CONFIG_NVME_TARGET_AUTH
 void nvmet_execute_auth_send(struct nvmet_req *req);
 void nvmet_execute_auth_receive(struct nvmet_req *req);
 int nvmet_auth_set_key(struct nvmet_host *host, const char *secret,
 		       bool set_ctrl);
 int nvmet_auth_set_host_hash(struct nvmet_host *host, const char *hash);
-u8 nvmet_setup_auth(struct nvmet_ctrl *ctrl);
+u8 nvmet_setup_auth(struct nvmet_ctrl *ctrl, struct nvmet_req *req);
 void nvmet_auth_sq_init(struct nvmet_sq *sq);
 void nvmet_destroy_auth(struct nvmet_ctrl *ctrl);
 void nvmet_auth_sq_free(struct nvmet_sq *sq);
@@ -732,16 +755,18 @@ int nvmet_auth_host_hash(struct nvmet_req *req, u8 *response,
 			 unsigned int hash_len);
 int nvmet_auth_ctrl_hash(struct nvmet_req *req, u8 *response,
 			 unsigned int hash_len);
-static inline bool nvmet_has_auth(struct nvmet_ctrl *ctrl)
+static inline bool nvmet_has_auth(struct nvmet_ctrl *ctrl, struct nvmet_req *req)
 {
-	return ctrl->host_key != NULL;
+	return ctrl->host_key != NULL && !nvmet_queue_tls_keyid(req->sq);
 }
 int nvmet_auth_ctrl_exponential(struct nvmet_req *req,
 				u8 *buf, int buf_size);
 int nvmet_auth_ctrl_sesskey(struct nvmet_req *req,
 			    u8 *buf, int buf_size);
+void nvmet_auth_insert_psk(struct nvmet_sq *sq);
 #else
-static inline u8 nvmet_setup_auth(struct nvmet_ctrl *ctrl)
+static inline u8 nvmet_setup_auth(struct nvmet_ctrl *ctrl,
+				  struct nvmet_req *req)
 {
 	return 0;
 }
@@ -754,11 +779,12 @@ static inline bool nvmet_check_auth_status(struct nvmet_req *req)
 {
 	return true;
 }
-static inline bool nvmet_has_auth(struct nvmet_ctrl *ctrl)
+static inline bool nvmet_has_auth(struct nvmet_ctrl *ctrl,
+				  struct nvmet_req *req)
 {
 	return false;
 }
 static inline const char *nvmet_dhchap_dhgroup_name(u8 dhgid) { return NULL; }
+static inline void nvmet_auth_insert_psk(struct nvmet_sq *sq) {};
 #endif
-
 #endif /* _NVMET_H */
drivers/nvme/target/tcp.c +21 -2
index 5bff0d5464d1..49849f028966 100644
@@ -1073,10 +1073,11 @@ static int nvmet_tcp_done_recv_pdu(struct nvmet_tcp_queue *queue)
 
 	if (unlikely(!nvmet_req_init(req, &queue->nvme_cq,
 			&queue->nvme_sq, &nvmet_tcp_ops))) {
-		pr_err("failed cmd %p id %d opcode %d, data_len: %d\n",
+		pr_err("failed cmd %p id %d opcode %d, data_len: %d, status: %04x\n",
 			req->cmd, req->cmd->common.command_id,
 			req->cmd->common.opcode,
-			le32_to_cpu(req->cmd->common.dptr.sgl.length));
+		       le32_to_cpu(req->cmd->common.dptr.sgl.length),
+		       le16_to_cpu(req->cqe->status));
 
 		nvmet_tcp_handle_req_failure(queue, queue->cmd, req);
 		return 0;
@@ -1602,6 +1603,7 @@ static void nvmet_tcp_release_queue_work(struct work_struct *w)
 	/* stop accepting incoming data */
 	queue->rcv_state = NVMET_TCP_RECV_ERR;
 
+	nvmet_sq_put_tls_key(&queue->nvme_sq);
 	nvmet_tcp_uninit_data_in_cmds(queue);
 	nvmet_sq_destroy(&queue->nvme_sq);
 	cancel_work_sync(&queue->io_work);
@@ -1807,6 +1809,23 @@ static void nvmet_tcp_tls_handshake_done(void *data, int status,
 	spin_unlock_bh(&queue->state_lock);
 
 	cancel_delayed_work_sync(&queue->tls_handshake_tmo_work);
+
+	if (!status) {
+		struct key *tls_key = nvme_tls_key_lookup(peerid);
+
+		if (IS_ERR(tls_key)) {
+			pr_warn("%s: queue %d failed to lookup key %x\n",
+				__func__, queue->idx, peerid);
+			spin_lock_bh(&queue->state_lock);
+			queue->state = NVMET_TCP_Q_FAILED;
+			spin_unlock_bh(&queue->state_lock);
+			status = PTR_ERR(tls_key);
+		} else {
+			pr_debug("%s: queue %d using TLS PSK %x\n",
+				 __func__, queue->idx, peerid);
+			queue->nvme_sq.tls_key = tls_key;
+		}
+	}
 	if (status)
 		nvmet_tcp_schedule_release_queue(queue);
 	else
Signature · 1 line
2.35.3
#18 lore raw

PATCH v8 9/9 nvmet: add tls_concat and tls_key debugfs entries

In reply to Hannes Reinecke

Add debugfs entries to display the 'concat' and 'tls_key' controller attributes.

---
 drivers/nvme/target/debugfs.c | 27 +++++++++++++++++++++++++++
 1 file changed, 27 insertions(+)
drivers/nvme/target/debugfs.c +27 -0
index cb2befc8619e..40e7b834a5eb 100644
@@ -132,6 +132,27 @@ static int nvmet_ctrl_host_traddr_show(struct seq_file *m, void *p)
 }
 NVMET_DEBUGFS_ATTR(nvmet_ctrl_host_traddr);
 
+#ifdef CONFIG_NVME_TARGET_TCP_TLS
+static int nvmet_ctrl_tls_key_show(struct seq_file *m, void *p)
+{
+	struct nvmet_ctrl *ctrl = m->private;
+	key_serial_t keyid = nvmet_queue_tls_keyid(ctrl->sqs[0]);
+
+	seq_printf(m, "%08x\n", keyid);
+	return 0;
+}
+NVMET_DEBUGFS_ATTR(nvmet_ctrl_tls_key);
+
+static int nvmet_ctrl_tls_concat_show(struct seq_file *m, void *p)
+{
+	struct nvmet_ctrl *ctrl = m->private;
+
+	seq_printf(m, "%d\n", ctrl->concat);
+	return 0;
+}
+NVMET_DEBUGFS_ATTR(nvmet_ctrl_tls_concat);
+#endif
+
 int nvmet_debugfs_ctrl_setup(struct nvmet_ctrl *ctrl)
 {
 	char name[32];
@@ -157,6 +178,12 @@ int nvmet_debugfs_ctrl_setup(struct nvmet_ctrl *ctrl)
 			    &nvmet_ctrl_state_fops);
 	debugfs_create_file("host_traddr", S_IRUSR, ctrl->debugfs_dir, ctrl,
 			    &nvmet_ctrl_host_traddr_fops);
+#ifdef CONFIG_NVME_TARGET_TCP_TLS
+	debugfs_create_file("tls_concat", S_IRUSR, ctrl->debugfs_dir, ctrl,
+			    &nvmet_ctrl_tls_concat_fops);
+	debugfs_create_file("tls_key", S_IRUSR, ctrl->debugfs_dir, ctrl,
+			    &nvmet_ctrl_tls_key_fops);
+#endif
 	return 0;
 }
 
Signature · 1 line
2.35.3
maintainer
#19 lore raw
In reply to Hannes Reinecke
Quoted 4 lines · Hannes Reinecke

Patchset can be found at git.kernel.org:/pub/scm/linux/kernel/git/hare/nvme.git branch secure-concat.v8

Not a valid git URL.

Also no base commit given and this doesn't apply to upstream.

- Eric

maintainer
#2 lore raw

Re: [PATCH 1/9] crypto,fs: Separate out hkdf_extract() and hkdf_expand()

In reply to Hannes Reinecke

crypto/Makefile obj-$(CONFIG_CRYPTO_ECHAINIV) += echainiv.o · line 33

Quoted 10 lines · Hannes Reinecke
crypto/Makefile +1 -0
index edbbaa3ffef5..b77fc360f0ff 100644
@@ -29,6 +29,7 @@ obj-$(CONFIG_CRYPTO_ECHAINIV) += echainiv.o
 
 crypto_hash-y += ahash.o
 crypto_hash-y += shash.o
+crypto_hash-y += hkdf.o
 obj-$(CONFIG_CRYPTO_HASH2) += crypto_hash.o

This should go under a kconfig option CONFIG_CRYPTO_HKDF that is selected by the users that need it. That way the code will be built only when needed.

Including a self-test would also be desirable.

crypto/hkdf.c line 7

Quoted 13 lines
crypto/hkdf.c new file +7 -0
new file mode 100644
index 000000000000..22e343851c0b
@@ -0,0 +1,112 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Implementation of HKDF ("HMAC-based Extract-and-Expand Key Derivation
+ * Function"), aka RFC 5869.  See also the original paper (Krawczyk 2010):
+ * "Cryptographic Extraction and Key Derivation: The HKDF Scheme".
+ *
+ * This is used to derive keys from the fscrypt master keys.

This is no longer in fs/crypto/, so the part about fscrypt should be removed.

crypto/hkdf.c line 20

Quoted 5 lines
crypto/hkdf.c new file +5 -0
+/*
+ * HKDF consists of two steps:
+ *
+ * 1. HKDF-Extract: extract a pseudorandom key of length HKDF_HASHLEN bytes from
+ *    the input keying material and optional salt.

It doesn't make sense to refer to HKDF_HASHLEN here since it is specific to fs/crypto/.

crypto/hkdf.c line 28

Quoted 3 lines
crypto/hkdf.c new file +3 -0
+/* HKDF-Extract (RFC 5869 section 2.2), unsalted */
+int hkdf_extract(struct crypto_shash *hmac_tfm, const u8 *ikm,
+		 unsigned int ikmlen, u8 *prk)

Needs kerneldoc now that this is a library interface.

crypto/hkdf.c line 36

Quoted 8 lines
crypto/hkdf.c new file +8 -0
+{
+	unsigned int prklen = crypto_shash_digestsize(hmac_tfm);
+	u8 *default_salt;
+	int err;
+
+	default_salt = kzalloc(prklen, GFP_KERNEL);
+	if (!default_salt)
+		return -ENOMEM;

Now that this is a library interface, it should take the salt as a parameter, and the users who want the default salt should explicitly specify that. If we only provide support for unsalted use, that might inadventently discourage the use of a salt in future code. As the function is named hkdf_extract(), people might also overlook that it's unsalted and doesn't actually match the RFC's definition of HKDF-Extract.

The use of kzalloc here is also inefficient, as the maximum length of a digest is known (HKDF_HASHLEN in fs/crypto/ case, HASH_MAX_DIGESTSIZE in general).

Quoted 9 lines
crypto/hkdf.c new file +9 -0
+	err = crypto_shash_setkey(hmac_tfm, default_salt, prklen);
+	if (!err)
+		err = crypto_shash_tfm_digest(hmac_tfm, ikm, ikmlen, prk);
+
+	kfree(default_salt);
+	return err;
+}
+EXPORT_SYMBOL_GPL(hkdf_extract);
+

crypto/hkdf.c line 55

Quoted 10 lines
crypto/hkdf.c new file +10 -0
+/*
+ * HKDF-Expand (RFC 5869 section 2.3).
+ * This expands the pseudorandom key, which was already keyed into @hmac_tfm,
+ * into @okmlen bytes of output keying material parameterized by the
+ * application-specific @info of length @infolen bytes.
+ * This is thread-safe and may be called by multiple threads in parallel.
+ */
+int hkdf_expand(struct crypto_shash *hmac_tfm,
+		const u8 *info, unsigned int infolen,
+		u8 *okm, unsigned int okmlen)

Needs kerneldoc now that this is a library interface.

Quoted 6 lines
fs/crypto/hkdf.c +0 -0
index 5a384dad2c72..9c2f9aca9412 100644

// SPDX-License-Identifier: GPL-2.0
/*

Quoted 8 lines

* Implementation of HKDF ("HMAC-based Extract-and-Expand Key Derivation
* Function"), aka RFC 5869. See also the original paper (Krawczyk 2010):
* "Cryptographic Extraction and Key Derivation: The HKDF Scheme". *
* This is used to derive keys from the fscrypt master keys.
*
* Copyright 2019 Google LLC
*/

The above file comment should be adjusted now that this file doesn't contain the actual HKDF implementation.

Quoted 13 lines
fs/crypto/hkdf.c +1 -5
@@ -118,61 +105,24 @@ int fscrypt_hkdf_expand(const struct fscrypt_hkdf *hkdf, u8 context,
 			u8 *okm, unsigned int okmlen)
 {
 	SHASH_DESC_ON_STACK(desc, hkdf->hmac_tfm);
-	u8 prefix[9];
-	unsigned int i;
+	u8 *prefix;
 	int err;
-	const u8 *prev = NULL;
-	u8 counter = 1;
-	u8 tmp[HKDF_HASHLEN];
 
 	if (WARN_ON_ONCE(okmlen > 255 * HKDF_HASHLEN))

fs/crypto/hkdf.c int fscrypt_hkdf_expand(const struct fscrypt_hkdf *hkdf, u8 context, · line 121

Quoted 10 lines
fs/crypto/hkdf.c +4 -0
 		return -EINVAL;
 
+	prefix = kzalloc(okmlen + 9, GFP_KERNEL);
+	if (!prefix)
+		return -ENOMEM;
 	desc->tfm = hkdf->hmac_tfm;
 
 	memcpy(prefix, "fscrypt\0", 8);
 	prefix[8] = context;
+	memcpy(prefix + 9, info, infolen);

This makes the variable called 'prefix' no longer be the prefix, but rather the full info string. A better name for it would be 'full_info'.

Also, it's being allocated with the wrong length. It should be 9 + infolen.

fs/crypto/hkdf.c int fscrypt_hkdf_expand(const struct fscrypt_hkdf *hkdf, u8 context, · line 125

Quoted 3 lines
fs/crypto/hkdf.c +3 -0
+	err = hkdf_expand(hkdf->hmac_tfm, prefix, infolen + 8,
+			  okm, okmlen);
+	kfree(prefix);

kfree_sensitive()

include/crypto/hkdf.h line 7

Quoted 13 lines
include/crypto/hkdf.h new file +7 -0
new file mode 100644
index 000000000000..bf06c080d7ed
@@ -0,0 +1,18 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+/*
+ * HKDF: HMAC-based Key Derivation Function (HKDF), RFC 5869
+ *
+ * Extracted from fs/crypto/hkdf.c, which has
+ * Copyright 2019 Google LLC
+ */

If this is keeping the copyright of fs/crypto/hkdf.c, the license needs to stay the same (GPL-2.0, not GPL-2.0-or-later).

- Eric

maintainer
#7 lore raw

Re: [PATCH 4/9] nvme: add nvme_auth_derive_tls_psk()

In reply to Hannes Reinecke
Quoted 56 lines · Hannes Reinecke
drivers/nvme/common/auth.c +56 -0
+/*
+ * Derive a TLS PSK as specified in TP8018 Section 3.6.1.3:
+ *   TLS PSK and PSK identity Derivation
+ *
+ * The TLS PSK shall be derived as follows from an input PSK
+ * (i.e., either a retained PSK or a generated PSK) and a PSK
+ * identity using the HKDF-Extract and HKDF-Expand-Label operations
+ * (refer to RFC 5869 and RFC 8446) where the hash function is the
+ * one specified by the hash specifier of the PSK identity:
+ * 1. PRK = HKDF-Extract(0, Input PSK); and
+ * 2. TLS PSK = HKDF-Expand-Label(PRK, "nvme-tls-psk", PskIdentityContext, L),
+ * where PskIdentityContext is the hash identifier indicated in
+ * the PSK identity concatenated to a space character and to the
+ * Base64 PSK digest (i.e., "<hash> <PSK digest>") and L is the
+ * output size in bytes of the hash function (i.e., 32 for SHA-256
+ * and 48 for SHA-384).
+ */
+int nvme_auth_derive_tls_psk(int hmac_id, u8 *psk, size_t psk_len,
+		u8 *psk_digest, u8 **ret_psk)
+{
+	struct crypto_shash *hmac_tfm;
+	const char *hmac_name;
+	const char *psk_prefix = "tls13 nvme-tls-psk";
+	size_t info_len, prk_len;
+	char *info;
+	unsigned char *prk, *tls_key;
+	int ret;
+
+	hmac_name = nvme_auth_hmac_name(hmac_id);
+	if (!hmac_name) {
+		pr_warn("%s: invalid hash algoritm %d\n",
+			__func__, hmac_id);
+		return -EINVAL;
+	}
+	if (hmac_id == NVME_AUTH_HASH_SHA512) {
+		pr_warn("%s: unsupported hash algorithm %s\n",
+			__func__, hmac_name);
+		return -EINVAL;
+	}
+
+	hmac_tfm = crypto_alloc_shash(hmac_name, 0, 0);
+	if (IS_ERR(hmac_tfm))
+		return PTR_ERR(hmac_tfm);
+
+	prk_len = crypto_shash_digestsize(hmac_tfm);
+	prk = kzalloc(prk_len, GFP_KERNEL);
+	if (!prk) {
+		ret = -ENOMEM;
+		goto out_free_shash;
+	}
+
+	ret = hkdf_extract(hmac_tfm, psk, psk_len, prk);
+	if (ret)
+		goto out_free_prk;
+
+	ret = crypto_shash_setkey(hmac_tfm, prk, prk_len);

drivers/nvme/common/auth.c int nvme_auth_generate_digest(u8 hmac_id, u8 *psk, size_t psk_len, · line 752

Quoted 10 lines
drivers/nvme/common/auth.c +10 -0
+	if (ret)
+		goto out_free_prk;
+
+	info_len = strlen(psk_digest) + strlen(psk_prefix) + 1;
+	info = kzalloc(info_len, GFP_KERNEL);
+	if (!info)
+		goto out_free_prk;
+
+	memcpy(info, psk_prefix, strlen(psk_prefix));
+	memcpy(info + strlen(psk_prefix), psk_digest, strlen(psk_digest));

The code doesn't match the description given in the function comment (which looks like it was quoted from a specification).

The code does HKDF-Expand with info="tls13 nvme-tls-psk<PSK digest>".

The description does HKDF-Expand-Label with Label="nvme-tls-psk", Context="<hash identifier> <PSK digest>", Length=digest_size.

Not only does the code not actually use <hash identifier>, but it doesn't follow the definition of HKDF-Expand-Label from RFC8446
(https://datatracker.ietf.org/doc/html/rfc8446#section-7.1) in that it's missing all the length fields. So the info string used by the actual code ends up being quite different from the one specified.

- Eric

maintainer
#14 lore raw

Re: [PATCH 8/9] nvmet-tcp: support secure channel concatenation

In reply to Hannes Reinecke
Quoted 5 lines · Hannes Reinecke
drivers/nvme/target/auth.c +5 -0
+	ret = nvme_auth_generate_digest(sq->ctrl->shash_id, psk, psk_len,
+					sq->ctrl->subsysnqn,
+					sq->ctrl->hostnqn, &digest);
+	if (ret) {
+		pr_warn("%s: ctrl %d qid %d failed to generate digest, error %d\n",

drivers/nvme/target/auth.c int nvmet_auth_ctrl_sesskey(struct nvmet_req *req, · line 594

Quoted 10 lines
drivers/nvme/target/auth.c +10 -0
+			__func__, sq->ctrl->cntlid, sq->qid, ret);
+		goto out_free_psk;
+	}
+	ret = nvme_auth_derive_tls_psk(sq->ctrl->shash_id, psk, psk_len,
+				       digest, &tls_psk);
+	if (ret) {
+		pr_warn("%s: ctrl %d qid %d failed to derive TLS PSK, error %d\n",
+			__func__, sq->ctrl->cntlid, sq->qid, ret);
+		goto out_free_digest;
+	}

This reuses 'psk' as both an HMAC key and as input keying material for HKDF. It's *probably* still secure, but this violates cryptographic best practices in that it reuses a key for multiple purposes. Is this a defect in the spec?

- Eric

maintainer
#10 lore raw

Re: [PATCH 5/9] nvme-keyring: add nvme_tls_psk_refresh()

In reply to Hannes Reinecke

drivers/nvme/common/keyring.c static struct key *nvme_tls_psk_lookup(struct key *keyring, · line 128

Quoted 2 lines · Hannes Reinecke
drivers/nvme/common/keyring.c +2 -0
+struct key *nvme_tls_psk_refresh(struct key *keyring, char *hostnqn, char *subnqn,
+		u8 hmac_id, bool generated, u8 *data, size_t data_len, char *digest)

For inputs use pointer to const, e.g. 'const char *hostnqn'. This applies to the whole patchset.

drivers/nvme/common/keyring.c static struct key *nvme_tls_psk_lookup(struct key *keyring, · line 148

Quoted 6 lines
drivers/nvme/common/keyring.c +6 -0
+	identity = kzalloc(identity_len, GFP_KERNEL);
+	if (!identity)
+		return ERR_PTR(-ENOMEM);
+
+	snprintf(identity, identity_len, "NVMe1%c%02d %s %s %s",
+		 generated ? 'G' : 'R', hmac_id, hostnqn, subnqn, digest);

This is reinventing kasprintf().

- Eric

maintainer
#20 lore raw
In reply to Eric Biggers
Quoted 8 lines · Eric Biggers
Quoted 4 lines · Hannes Reinecke

Patchset can be found at git.kernel.org:/pub/scm/linux/kernel/git/hare/nvme.git branch secure-concat.v8

Not a valid git URL.

git://git.kernel.org/pub/scm/linux/kernel/git/hare/nvme.git
branch secure-concat.v8
base commit f178e8f9eaa900a353145179fc0164a12acfbb2c

Sorry for the inconvenience.

Cheers,

Hannes

Signature · 4 lines
Dr. Hannes Reinecke                  Kernel Storage Architect
[email protected]                                +49 911 74053 688
SUSE Software Solutions GmbH, Frankenstr. 146, 90461 Nürnberg
HRB 36809 (AG Nürnberg), GF: I. Totev, A. McDonald, W. Knoblich
maintainer
#3 lore raw

Re: [PATCH 1/9] crypto,fs: Separate out hkdf_extract() and hkdf_expand()

In reply to Eric Biggers
Quoted 7 lines · Eric Biggers
Quoted 6 lines · Hannes Reinecke
crypto/Makefile +0 -0
index edbbaa3ffef5..b77fc360f0ff 100644
@@ -29,6 +29,7 @@ obj-$(CONFIG_CRYPTO_ECHAINIV) += echainiv.o
 
Quoted 8 lines

crypto/Makefile obj-$(CONFIG_CRYPTO_ECHAINIV) += echainiv.o · line 33

Quoted 4 lines
crypto/Makefile +1 -0
  crypto_hash-y += ahash.o
  crypto_hash-y += shash.o
+crypto_hash-y += hkdf.o
  obj-$(CONFIG_CRYPTO_HASH2) += crypto_hash.o

This should go under a kconfig option CONFIG_CRYPTO_HKDF that is selected by the users that need it. That way the code will be built only when needed.

Okay.

Quoted 2 lines

Including a self-test would also be desirable.

First time for me, but ok.

Quoted 7 lines
Quoted 7 lines
crypto/hkdf.c new file +1 -0
new file mode 100644
index 000000000000..22e343851c0b
@@ -0,0 +1,112 @@
+// SPDX-License-Identifier: GPL-2.0
Quoted 9 lines

crypto/hkdf.c line 7

Quoted 6 lines
crypto/hkdf.c new file +6 -0
+/*
+ * Implementation of HKDF ("HMAC-based Extract-and-Expand Key Derivation
+ * Function"), aka RFC 5869.  See also the original paper (Krawczyk 2010):
+ * "Cryptographic Extraction and Key Derivation: The HKDF Scheme".
+ *
+ * This is used to derive keys from the fscrypt master keys.

This is no longer in fs/crypto/, so the part about fscrypt should be removed.

Will be removing this line.

Quoted 9 lines

crypto/hkdf.c line 20

Quoted 5 lines
crypto/hkdf.c new file +5 -0
+/*
+ * HKDF consists of two steps:
+ *
+ * 1. HKDF-Extract: extract a pseudorandom key of length HKDF_HASHLEN bytes from
+ *    the input keying material and optional salt.

It doesn't make sense to refer to HKDF_HASHLEN here since it is specific to fs/crypto/.

Ok.

Quoted 6 lines

crypto/hkdf.c line 28

Quoted 3 lines
crypto/hkdf.c new file +3 -0
+/* HKDF-Extract (RFC 5869 section 2.2), unsalted */
+int hkdf_extract(struct crypto_shash *hmac_tfm, const u8 *ikm,
+		 unsigned int ikmlen, u8 *prk)

Needs kerneldoc now that this is a library interface.

Indeed, will be updating the comment.

Quoted 16 lines

crypto/hkdf.c line 36

Quoted 8 lines
crypto/hkdf.c new file +8 -0
+{
+	unsigned int prklen = crypto_shash_digestsize(hmac_tfm);
+	u8 *default_salt;
+	int err;
+
+	default_salt = kzalloc(prklen, GFP_KERNEL);
+	if (!default_salt)
+		return -ENOMEM;

Now that this is a library interface, it should take the salt as a parameter, and the users who want the default salt should explicitly specify that. If we only provide support for unsalted use, that might inadventently discourage the use of a salt in future code. As the function is named hkdf_extract(), people might also overlook that it's unsalted and doesn't actually match the RFC's definition of HKDF-Extract.

Quoted 3 lines

The use of kzalloc here is also inefficient, as the maximum length of a digest is known (HKDF_HASHLEN in fs/crypto/ case, HASH_MAX_DIGESTSIZE in general).

I was trying to keep the changes to the actual code to a minimum, so I didn't modify the calling sequence of the original code. But sure, passing in the 'salt' as a parameter is sensible.

Quoted 13 lines
Quoted 13 lines
crypto/hkdf.c new file +13 -0
+	err = crypto_shash_setkey(hmac_tfm, default_salt, prklen);
+	if (!err)
+		err = crypto_shash_tfm_digest(hmac_tfm, ikm, ikmlen, prk);
+
+	kfree(default_salt);
+	return err;
+}
+EXPORT_SYMBOL_GPL(hkdf_extract);
+
+/*
+ * HKDF-Expand (RFC 5869 section 2.3).
+ * This expands the pseudorandom key, which was already keyed into @hmac_tfm,
+ * into @okmlen bytes of output keying material parameterized by the
Quoted 9 lines

crypto/hkdf.c line 55

Quoted 6 lines
crypto/hkdf.c new file +6 -0
+ * application-specific @info of length @infolen bytes.
+ * This is thread-safe and may be called by multiple threads in parallel.
+ */
+int hkdf_expand(struct crypto_shash *hmac_tfm,
+		const u8 *info, unsigned int infolen,
+		u8 *okm, unsigned int okmlen)

Needs kerneldoc now that this is a library interface.

Ok.

Quoted 9 lines
Quoted 9 lines
fs/crypto/hkdf.c +0 -0
index 5a384dad2c72..9c2f9aca9412 100644

// SPDX-License-Identifier: GPL-2.0
/*
* Implementation of HKDF ("HMAC-based Extract-and-Expand Key Derivation
* Function"), aka RFC 5869. See also the original paper (Krawczyk 2010):
* "Cryptographic Extraction and Key Derivation: The HKDF Scheme".

Quoted 9 lines
Quoted 5 lines

*
* This is used to derive keys from the fscrypt master keys. *
* Copyright 2019 Google LLC
*/

The above file comment should be adjusted now that this file doesn't contain the actual HKDF implementation.

Ok.

Quoted 20 lines
Quoted 20 lines
fs/crypto/hkdf.c +4 -5
@@ -118,61 +105,24 @@ int fscrypt_hkdf_expand(const struct fscrypt_hkdf *hkdf, u8 context,
  			u8 *okm, unsigned int okmlen)
  {
  	SHASH_DESC_ON_STACK(desc, hkdf->hmac_tfm);
-	u8 prefix[9];
-	unsigned int i;
+	u8 *prefix;
  	int err;
-	const u8 *prev = NULL;
-	u8 counter = 1;
-	u8 tmp[HKDF_HASHLEN];
 
  	if (WARN_ON_ONCE(okmlen > 255 * HKDF_HASHLEN))
  		return -EINVAL;
 
+	prefix = kzalloc(okmlen + 9, GFP_KERNEL);
+	if (!prefix)
+		return -ENOMEM;
  	desc->tfm = hkdf->hmac_tfm;
 
Quoted 9 lines

fs/crypto/hkdf.c int fscrypt_hkdf_expand(const struct fscrypt_hkdf *hkdf, u8 context, · line 121

Quoted 3 lines
fs/crypto/hkdf.c +1 -0
  	memcpy(prefix, "fscrypt\0", 8);
  	prefix[8] = context;
+	memcpy(prefix + 9, info, infolen);

This makes the variable called 'prefix' no longer be the prefix, but rather the full info string. A better name for it would be 'full_info'.

Also, it's being allocated with the wrong length. It should be 9 + infolen.

Will fix it up.

Quoted 6 lines

fs/crypto/hkdf.c int fscrypt_hkdf_expand(const struct fscrypt_hkdf *hkdf, u8 context, · line 125

Quoted 3 lines
fs/crypto/hkdf.c +3 -0
+	err = hkdf_expand(hkdf->hmac_tfm, prefix, infolen + 8,
+			  okm, okmlen);
+	kfree(prefix);

kfree_sensitive()

Quoted 17 lines

include/crypto/hkdf.h line 7

Quoted 13 lines
include/crypto/hkdf.h new file +7 -0
new file mode 100644
index 000000000000..bf06c080d7ed
@@ -0,0 +1,18 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+/*
+ * HKDF: HMAC-based Key Derivation Function (HKDF), RFC 5869
+ *
+ * Extracted from fs/crypto/hkdf.c, which has
+ * Copyright 2019 Google LLC
+ */

If this is keeping the copyright of fs/crypto/hkdf.c, the license needs to stay the same (GPL-2.0, not GPL-2.0-or-later).

Will be modifying it.

Thanks a lot for your review!

Cheers,

Hannes

Signature · 4 lines
Dr. Hannes Reinecke                  Kernel Storage Architect
[email protected]                                +49 911 74053 688
SUSE Software Solutions GmbH, Frankenstr. 146, 90461 Nürnberg
HRB 36809 (AG Nürnberg), GF: I. Totev, A. McDonald, W. Knoblich
maintainer
#8 lore raw

Re: [PATCH 4/9] nvme: add nvme_auth_derive_tls_psk()

In reply to Eric Biggers
Quoted 73 lines · Eric Biggers
Quoted 56 lines · Hannes Reinecke
drivers/nvme/common/auth.c +56 -0
+/*
+ * Derive a TLS PSK as specified in TP8018 Section 3.6.1.3:
+ *   TLS PSK and PSK identity Derivation
+ *
+ * The TLS PSK shall be derived as follows from an input PSK
+ * (i.e., either a retained PSK or a generated PSK) and a PSK
+ * identity using the HKDF-Extract and HKDF-Expand-Label operations
+ * (refer to RFC 5869 and RFC 8446) where the hash function is the
+ * one specified by the hash specifier of the PSK identity:
+ * 1. PRK = HKDF-Extract(0, Input PSK); and
+ * 2. TLS PSK = HKDF-Expand-Label(PRK, "nvme-tls-psk", PskIdentityContext, L),
+ * where PskIdentityContext is the hash identifier indicated in
+ * the PSK identity concatenated to a space character and to the
+ * Base64 PSK digest (i.e., "<hash> <PSK digest>") and L is the
+ * output size in bytes of the hash function (i.e., 32 for SHA-256
+ * and 48 for SHA-384).
+ */
+int nvme_auth_derive_tls_psk(int hmac_id, u8 *psk, size_t psk_len,
+		u8 *psk_digest, u8 **ret_psk)
+{
+	struct crypto_shash *hmac_tfm;
+	const char *hmac_name;
+	const char *psk_prefix = "tls13 nvme-tls-psk";
+	size_t info_len, prk_len;
+	char *info;
+	unsigned char *prk, *tls_key;
+	int ret;
+
+	hmac_name = nvme_auth_hmac_name(hmac_id);
+	if (!hmac_name) {
+		pr_warn("%s: invalid hash algoritm %d\n",
+			__func__, hmac_id);
+		return -EINVAL;
+	}
+	if (hmac_id == NVME_AUTH_HASH_SHA512) {
+		pr_warn("%s: unsupported hash algorithm %s\n",
+			__func__, hmac_name);
+		return -EINVAL;
+	}
+
+	hmac_tfm = crypto_alloc_shash(hmac_name, 0, 0);
+	if (IS_ERR(hmac_tfm))
+		return PTR_ERR(hmac_tfm);
+
+	prk_len = crypto_shash_digestsize(hmac_tfm);
+	prk = kzalloc(prk_len, GFP_KERNEL);
+	if (!prk) {
+		ret = -ENOMEM;
+		goto out_free_shash;
+	}
+
+	ret = hkdf_extract(hmac_tfm, psk, psk_len, prk);
+	if (ret)
+		goto out_free_prk;
+
+	ret = crypto_shash_setkey(hmac_tfm, prk, prk_len);

drivers/nvme/common/auth.c int nvme_auth_generate_digest(u8 hmac_id, u8 *psk, size_t psk_len, · line 752

Quoted 10 lines
drivers/nvme/common/auth.c +10 -0
+	if (ret)
+		goto out_free_prk;
+
+	info_len = strlen(psk_digest) + strlen(psk_prefix) + 1;
+	info = kzalloc(info_len, GFP_KERNEL);
+	if (!info)
+		goto out_free_prk;
+
+	memcpy(info, psk_prefix, strlen(psk_prefix));
+	memcpy(info + strlen(psk_prefix), psk_digest, strlen(psk_digest));

The code doesn't match the description given in the function comment (which looks like it was quoted from a specification).

The code does HKDF-Expand with info="tls13 nvme-tls-psk<PSK digest>".

Quoted 9 lines

The description does HKDF-Expand-Label with Label="nvme-tls-psk", Context="<hash identifier> <PSK digest>", Length=digest_size.

Not only does the code not actually use <hash identifier>, but it doesn't follow the definition of HKDF-Expand-Label from RFC8446
(https://datatracker.ietf.org/doc/html/rfc8446#section-7.1) in that it's missing all the length fields. So the info string used by the actual code ends up being quite different from the one specified.

Aw. Guess you are right. Will be fixing it up.

Cheers,

Hannes

Signature · 4 lines
Dr. Hannes Reinecke                  Kernel Storage Architect
[email protected]                                +49 911 74053 688
SUSE Software Solutions GmbH, Frankenstr. 146, 90461 Nürnberg
HRB 36809 (AG Nürnberg), GF: I. Totev, A. McDonald, W. Knoblich
maintainer
#15 lore raw

Re: [PATCH 8/9] nvmet-tcp: support secure channel concatenation

In reply to Eric Biggers
Quoted 12 lines · Eric Biggers
Quoted 11 lines · Hannes Reinecke
drivers/nvme/target/auth.c +11 -0
+	ret = nvme_auth_generate_digest(sq->ctrl->shash_id, psk, psk_len,
+					sq->ctrl->subsysnqn,
+					sq->ctrl->hostnqn, &digest);
+	if (ret) {
+		pr_warn("%s: ctrl %d qid %d failed to generate digest, error %d\n",
+			__func__, sq->ctrl->cntlid, sq->qid, ret);
+		goto out_free_psk;
+	}
+	ret = nvme_auth_derive_tls_psk(sq->ctrl->shash_id, psk, psk_len,
+				       digest, &tls_psk);
+	if (ret) {
Quoted 9 lines

drivers/nvme/target/auth.c int nvmet_auth_ctrl_sesskey(struct nvmet_req *req, · line 594

Quoted 4 lines
drivers/nvme/target/auth.c +4 -0
+		pr_warn("%s: ctrl %d qid %d failed to derive TLS PSK, error %d\n",
+			__func__, sq->ctrl->cntlid, sq->qid, ret);
+		goto out_free_digest;
+	}

This reuses 'psk' as both an HMAC key and as input keying material for HKDF. It's *probably* still secure, but this violates cryptographic best practices in that it reuses a key for multiple purposes. Is this a defect in the spec?

This is using a digest calculated from the actual PSK key material, true. You are right that with that we probably impact cryptographic reliability, but that that is what the spec mandates.

Actual reason for this modification was the need to identify the TLS PSKs for each connection, _and_ support key refresh.

We identify TLS PSKs by the combination of '<hash> <hostnqn> <subsysnqn>', where '<hostnqn>' is the identification of the initiator (source), and '<subsynqn>' the identification of the
target. But as we regenerate the PSK for each reset we are having
a hard time identifying the newly generated PSK by the original
'<hash> <hostnqn> <subsysnqn>' tuple only.
We cannot delete the original TLS PSK directly as it might be used
by other connections, so there will be a time where both PSKs
are valid and have to be stored in the keyring.

And keeping a global 'revision' counter is horrible, the alternative of having a per-connection instance counter is similarly unpleasant.

Not ideal, true, so if you have better ideas I'd like to hear them.

But thanks for your consideration. I'll be bringing them up.

Cheers,

Hannes

Signature · 4 lines
Dr. Hannes Reinecke                  Kernel Storage Architect
[email protected]                                +49 911 74053 688
SUSE Software Solutions GmbH, Frankenstr. 146, 90461 Nürnberg
HRB 36809 (AG Nürnberg), GF: I. Totev, A. McDonald, W. Knoblich
maintainer
#16 lore raw

Re: [PATCH 8/9] nvmet-tcp: support secure channel concatenation

Quoted 22 lines · Hannes Reinecke
Quoted 12 lines · Eric Biggers
Quoted 11 lines · Hannes Reinecke
drivers/nvme/target/auth.c +11 -0
+	ret = nvme_auth_generate_digest(sq->ctrl->shash_id, psk, psk_len,
+					sq->ctrl->subsysnqn,
+					sq->ctrl->hostnqn, &digest);
+	if (ret) {
+		pr_warn("%s: ctrl %d qid %d failed to generate digest, error %d\n",
+			__func__, sq->ctrl->cntlid, sq->qid, ret);
+		goto out_free_psk;
+	}
+	ret = nvme_auth_derive_tls_psk(sq->ctrl->shash_id, psk, psk_len,
+				       digest, &tls_psk);
+	if (ret) {
Quoted 9 lines

drivers/nvme/target/auth.c int nvmet_auth_ctrl_sesskey(struct nvmet_req *req, · line 594

Quoted 4 lines
drivers/nvme/target/auth.c +4 -0
+		pr_warn("%s: ctrl %d qid %d failed to derive TLS PSK, error %d\n",
+			__func__, sq->ctrl->cntlid, sq->qid, ret);
+		goto out_free_digest;
+	}

This reuses 'psk' as both an HMAC key and as input keying material for HKDF. It's *probably* still secure, but this violates cryptographic best practices in that it reuses a key for multiple purposes. Is this a defect in the spec?

Quoted 3 lines

This is using a digest calculated from the actual PSK key material, true. You are right that with that we probably impact cryptographic reliability, but that that is what the spec mandates.

How set in stone is this specification? Is it finalized and has it been implemented by anyone else? Your code didn't correctly implement the spec anyway, so at least you must not have done any interoperability testing.

Quoted 14 lines

Actual reason for this modification was the need to identify the TLS PSKs for each connection, _and_ support key refresh.

We identify TLS PSKs by the combination of '<hash> <hostnqn> <subsysnqn>', where '<hostnqn>' is the identification of the
initiator (source), and '<subsynqn>' the identification of the target. But as we regenerate the PSK for each reset we are having a hard time identifying the newly generated PSK by the original '<hash> <hostnqn> <subsysnqn>' tuple only.
We cannot delete the original TLS PSK directly as it might be used by other connections, so there will be a time where both PSKs are valid and have to be stored in the keyring.

Quoted 7 lines

And keeping a global 'revision' counter is horrible, the alternative of having a per-connection instance counter is similarly unpleasant.

Not ideal, true, so if you have better ideas I'd like to hear them.

But thanks for your consideration. I'll be bringing them up.

You are already using HKDF, so you could just use that, similar to how fscrypt uses HKDF to derive both its key identifiers and its actual subkeys. HKDF can be used to derive both public and non-public values; there's no need to use a separate algorithm such as plain HMAC just because you need a public value.

- Eric

maintainer
#17 lore raw

Re: [PATCH 8/9] nvmet-tcp: support secure channel concatenation

Quoted 23 lines · Eric Biggers
Quoted 22 lines · Hannes Reinecke
Quoted 21 lines · Eric Biggers
Quoted 5 lines · Hannes Reinecke
drivers/nvme/target/auth.c +5 -0
+	ret = nvme_auth_generate_digest(sq->ctrl->shash_id, psk, psk_len,
+					sq->ctrl->subsysnqn,
+					sq->ctrl->hostnqn, &digest);
+	if (ret) {
+		pr_warn("%s: ctrl %d qid %d failed to generate digest, error %d\n",

drivers/nvme/target/auth.c int nvmet_auth_ctrl_sesskey(struct nvmet_req *req, · line 594

Quoted 10 lines
drivers/nvme/target/auth.c +10 -0
+			__func__, sq->ctrl->cntlid, sq->qid, ret);
+		goto out_free_psk;
+	}
+	ret = nvme_auth_derive_tls_psk(sq->ctrl->shash_id, psk, psk_len,
+				       digest, &tls_psk);
+	if (ret) {
+		pr_warn("%s: ctrl %d qid %d failed to derive TLS PSK, error %d\n",
+			__func__, sq->ctrl->cntlid, sq->qid, ret);
+		goto out_free_digest;
+	}

This reuses 'psk' as both an HMAC key and as input keying material for HKDF. It's *probably* still secure, but this violates cryptographic best practices in that it reuses a key for multiple purposes. Is this a defect in the spec?

Quoted 8 lines
Quoted 3 lines

This is using a digest calculated from the actual PSK key material, true. You are right that with that we probably impact cryptographic reliability, but that that is what the spec mandates.

How set in stone is this specification? Is it finalized and has it been implemented by anyone else? Your code didn't correctly implement the spec anyway, so at least you must not have done any interoperability testing.

Well ... thing is, this _is_ the first implementation. Anyone else does interop testing against us.
The spec is pretty much set in stone here; sure we can update the spec, but that takes time. I can ask the primary author if he's willing to engage in a conversation about the cryptographic implications if you are up to it.

Quoted 19 lines
Quoted 19 lines

Actual reason for this modification was the need to identify the TLS PSKs for each connection, _and_ support key refresh.

We identify TLS PSKs by the combination of '<hash> <hostnqn> <subsysnqn>', where '<hostnqn>' is the identification of the
initiator (source), and '<subsynqn>' the identification of the target. But as we regenerate the PSK for each reset we are having a hard time identifying the newly generated PSK by the original '<hash> <hostnqn> <subsysnqn>' tuple only.
We cannot delete the original TLS PSK directly as it might be used by other connections, so there will be a time where both PSKs are valid and have to be stored in the keyring.

And keeping a global 'revision' counter is horrible, the alternative of having a per-connection instance counter is similarly unpleasant.

Not ideal, true, so if you have better ideas I'd like to hear them.

Quoted 8 lines
Quoted 2 lines

But thanks for your consideration. I'll be bringing them up.

You are already using HKDF, so you could just use that, similar to how fscrypt uses HKDF to derive both its key identifiers and its actual subkeys. HKDF can be used to derive both public and non-public values; there's no need to use a separate algorithm such as plain HMAC just because you need a public value.

I will check. But at this time I fear we have to stick with this implementation because that's what the spec mandates.

But thanks for the review, very much appreciated.

Cheers,

Hannes

Signature · 4 lines
Dr. Hannes Reinecke                  Kernel Storage Architect
[email protected]                                +49 911 74053 688
SUSE Software Solutions GmbH, Frankenstr. 146, 90461 Nürnberg
HRB 36809 (AG Nürnberg), GF: I. Totev, A. McDonald, W. Knoblich