kernel/fs/smb/client/cifsencrypt.c, branch linux-6.18.y

smb: client: Compare MACs in constant time

2026-03-19T15:08:50Z

commit 26bc83b88bbbf054f0980a4a42047a8d1e210e4c upstream. To prevent timing attacks, MAC comparisons need to be constant-time. Replace the memcmp() with the correct function, crypto_memneq(). Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2") Cc: stable@vger.kernel.org Acked-by: Paulo Alcantara (Red Hat) Signed-off-by: Eric Biggers Signed-off-by: Steve French Signed-off-by: Greg Kroah-Hartman

smb: client: Remove obsolete crypto_shash allocations

2025-10-16T03:10:28Z

Now that the SMB client accesses MD5, HMAC-MD5, HMAC-SHA256, and SHA-512 only via the library API and not via crypto_shash, allocating crypto_shash objects for these algorithms is no longer necessary. Remove all these allocations, their corresponding kconfig selections, and their corresponding module soft dependencies. Reviewed-by: Stefan Metzmacher Acked-by: Ard Biesheuvel Signed-off-by: Eric Biggers Signed-off-by: Steve French

smb: client: Use HMAC-MD5 library for NTLMv2

2025-10-16T03:10:28Z

For the HMAC-MD5 computations in NTLMv2, use the HMAC-MD5 library instead of a "hmac(md5)" crypto_shash. This is simpler and faster. With the library there's no need to allocate memory, no need to handle errors, and the HMAC-MD5 code is accessed directly without inefficient indirect calls and other unnecessary API overhead. To preserve the existing behavior of NTLMv2 support being disabled when the kernel is booted with "fips=1", make setup_ntlmv2_rsp() check fips_enabled itself. Previously it relied on the error from cifs_alloc_hash("hmac(md5)", &hmacmd5). Reviewed-by: Stefan Metzmacher Acked-by: Ard Biesheuvel Signed-off-by: Eric Biggers Signed-off-by: Steve French

smb: client: Use MD5 library for SMB1 signature calculation

2025-10-16T03:10:28Z

Convert cifs_calc_signature() to use the MD5 library instead of a "md5" crypto_shash. This is simpler and faster. With the library there's no need to allocate memory, no need to handle errors, and the MD5 code is accessed directly without inefficient indirect calls and other unnecessary API overhead. To preserve the existing behavior of MD5 signature support being disabled when the kernel is booted with "fips=1", make cifs_calc_signature() check fips_enabled itself. Previously it relied on the error from cifs_alloc_hash("md5", &server->secmech.md5). Reviewed-by: Stefan Metzmacher Acked-by: Ard Biesheuvel Signed-off-by: Eric Biggers Signed-off-by: Steve French

smb: client: Use HMAC-SHA256 library for SMB2 signature calculation

2025-10-16T03:10:28Z

Convert smb2_calc_signature() to use the HMAC-SHA256 library instead of a "hmac(sha256)" crypto_shash. This is simpler and faster. With the library there's no need to allocate memory, no need to handle errors, and the HMAC-SHA256 code is accessed directly without inefficient indirect calls and other unnecessary API overhead. To make this possible, make __cifs_calc_signature() support both the HMAC-SHA256 library and crypto_shash. (crypto_shash is still needed for HMAC-MD5 and AES-CMAC. A later commit will switch HMAC-MD5 from shash to the library. I'd like to eventually do the same for AES-CMAC, but it doesn't have a library API yet. So for now, shash is still needed.) Also remove the unnecessary 'sigptr' variable. For now smb3_crypto_shash_allocate() still allocates a "hmac(sha256)" crypto_shash. It will be removed in a later commit. Reviewed-by: Stefan Metzmacher Acked-by: Ard Biesheuvel Signed-off-by: Eric Biggers Signed-off-by: Steve French

smb: Use arc4 library instead of duplicate arc4 code

2025-09-30T02:21:23Z

fs/smb/common/cifs_arc4.c has an implementation of ARC4, but a copy of this same code is also present in lib/crypto/arc4.c to serve the other users of this legacy algorithm in the kernel. Remove the duplicate implementation in fs/smb/, which seems to have been added because of a misunderstanding, and just use the lib/crypto/ one. Signed-off-by: Eric Biggers Reviewed-by: Paulo Alcantara (Red Hat) Signed-off-by: Steve French

smb: client: fix session setup against servers that require SPN

2025-07-27T21:59:59Z

Some servers might enforce the SPN to be set in the target info blob (AV pairs) when sending NTLMSSP_AUTH message. In Windows Server, this could be enforced with SmbServerNameHardeningLevel set to 2. Fix this by always appending SPN (cifs/) to the existing list of target infos when setting up NTLMv2 response blob. Cc: linux-cifs@vger.kernel.org Cc: David Howells Reported-by: Pierguido Lambri Signed-off-by: Paulo Alcantara (Red Hat) Signed-off-by: Steve French

smb: client: allow parsing zero-length AV pairs

2025-07-27T21:59:59Z

Zero-length AV pairs should be considered as valid target infos. Don't skip the next AV pairs that follow them. Cc: linux-cifs@vger.kernel.org Cc: David Howells Fixes: 0e8ae9b953bc ("smb: client: parse av pair type 4 in CHALLENGE_MESSAGE") Signed-off-by: Paulo Alcantara (Red Hat) Signed-off-by: Steve French

smb: client: fix UAF in decryption with multichannel

2025-04-09T15:24:55Z

After commit f7025d861694 ("smb: client: allocate crypto only for primary server") and commit b0abcd65ec54 ("smb: client: fix UAF in async decryption"), the channels started reusing AEAD TFM from primary channel to perform synchronous decryption, but that can't done as there could be multiple cifsd threads (one per channel) simultaneously accessing it to perform decryption. This fixes the following KASAN splat when running fstest generic/249 with 'vers=3.1.1,multichannel,max_channels=4,seal' against Windows Server 2022: BUG: KASAN: slab-use-after-free in gf128mul_4k_lle+0xba/0x110 Read of size 8 at addr ffff8881046c18a0 by task cifsd/986 CPU: 3 UID: 0 PID: 986 Comm: cifsd Not tainted 6.15.0-rc1 #1 PREEMPT(voluntary) Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 1.16.3-3.fc41 04/01/2014 Call Trace: dump_stack_lvl+0x5d/0x80 print_report+0x156/0x528 ? gf128mul_4k_lle+0xba/0x110 ? __virt_addr_valid+0x145/0x300 ? __phys_addr+0x46/0x90 ? gf128mul_4k_lle+0xba/0x110 kasan_report+0xdf/0x1a0 ? gf128mul_4k_lle+0xba/0x110 gf128mul_4k_lle+0xba/0x110 ghash_update+0x189/0x210 shash_ahash_update+0x295/0x370 ? __pfx_shash_ahash_update+0x10/0x10 ? __pfx_shash_ahash_update+0x10/0x10 ? __pfx_extract_iter_to_sg+0x10/0x10 ? ___kmalloc_large_node+0x10e/0x180 ? __asan_memset+0x23/0x50 crypto_ahash_update+0x3c/0xc0 gcm_hash_assoc_remain_continue+0x93/0xc0 crypt_message+0xe09/0xec0 [cifs] ? __pfx_crypt_message+0x10/0x10 [cifs] ? _raw_spin_unlock+0x23/0x40 ? __pfx_cifs_readv_from_socket+0x10/0x10 [cifs] decrypt_raw_data+0x229/0x380 [cifs] ? __pfx_decrypt_raw_data+0x10/0x10 [cifs] ? __pfx_cifs_read_iter_from_socket+0x10/0x10 [cifs] smb3_receive_transform+0x837/0xc80 [cifs] ? __pfx_smb3_receive_transform+0x10/0x10 [cifs] ? __pfx___might_resched+0x10/0x10 ? __pfx_smb3_is_transform_hdr+0x10/0x10 [cifs] cifs_demultiplex_thread+0x692/0x1570 [cifs] ? __pfx_cifs_demultiplex_thread+0x10/0x10 [cifs] ? rcu_is_watching+0x20/0x50 ? rcu_lockdep_current_cpu_online+0x62/0xb0 ? find_held_lock+0x32/0x90 ? kvm_sched_clock_read+0x11/0x20 ? local_clock_noinstr+0xd/0xd0 ? trace_irq_enable.constprop.0+0xa8/0xe0 ? __pfx_cifs_demultiplex_thread+0x10/0x10 [cifs] kthread+0x1fe/0x380 ? kthread+0x10f/0x380 ? __pfx_kthread+0x10/0x10 ? local_clock_noinstr+0xd/0xd0 ? ret_from_fork+0x1b/0x60 ? local_clock+0x15/0x30 ? lock_release+0x29b/0x390 ? rcu_is_watching+0x20/0x50 ? __pfx_kthread+0x10/0x10 ret_from_fork+0x31/0x60 ? __pfx_kthread+0x10/0x10 ret_from_fork_asm+0x1a/0x30 Tested-by: David Howells Reported-by: Steve French Closes: https://lore.kernel.org/r/CAH2r5mu6Yc0-RJXM3kFyBYUB09XmXBrNodOiCVR4EDrmxq5Szg@mail.gmail.com Fixes: f7025d861694 ("smb: client: allocate crypto only for primary server") Fixes: b0abcd65ec54 ("smb: client: fix UAF in async decryption") Signed-off-by: Paulo Alcantara (Red Hat) Signed-off-by: Steve French

smb: client: parse av pair type 4 in CHALLENGE_MESSAGE

2025-01-20T01:33:59Z

Parse FQDN of the domain in CHALLENGE_MESSAGE message as it's gonna be useful when mounting DFS shares against old Windows Servers (2012 R2 or earlier) that return not fully qualified hostnames for DFS targets by default. Signed-off-by: Paulo Alcantara (Red Hat) Signed-off-by: Steve French