diff --git a/sys/kern/uipc_ktls.c b/sys/kern/uipc_ktls.c
index b4b169c4daf2..294a196db60d 100644
--- a/sys/kern/uipc_ktls.c
+++ b/sys/kern/uipc_ktls.c
@@ -1,3334 +1,3353 @@
/*-
* SPDX-License-Identifier: BSD-2-Clause
*
* Copyright (c) 2014-2019 Netflix Inc.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*/
#include <sys/cdefs.h>
#include "opt_inet.h"
#include "opt_inet6.h"
#include "opt_kern_tls.h"
#include "opt_ratelimit.h"
#include "opt_rss.h"
#include <sys/param.h>
#include <sys/kernel.h>
#include <sys/domainset.h>
#include <sys/endian.h>
#include <sys/ktls.h>
#include <sys/lock.h>
#include <sys/mbuf.h>
#include <sys/mutex.h>
#include <sys/rmlock.h>
#include <sys/proc.h>
#include <sys/protosw.h>
#include <sys/refcount.h>
#include <sys/smp.h>
#include <sys/socket.h>
#include <sys/socketvar.h>
#include <sys/sysctl.h>
#include <sys/taskqueue.h>
#include <sys/kthread.h>
#include <sys/uio.h>
#include <sys/vmmeter.h>
#if defined(__aarch64__) || defined(__amd64__) || defined(__i386__)
#include <machine/pcb.h>
#endif
#include <machine/vmparam.h>
#include <net/if.h>
#include <net/if_var.h>
#ifdef RSS
#include <net/netisr.h>
#include <net/rss_config.h>
#endif
#include <net/route.h>
#include <net/route/nhop.h>
#include <netinet/in.h>
#include <netinet/in_pcb.h>
#include <netinet/tcp_var.h>
#ifdef TCP_OFFLOAD
#include <netinet/tcp_offload.h>
#endif
#include <opencrypto/cryptodev.h>
#include <opencrypto/ktls.h>
#include <vm/vm.h>
#include <vm/vm_pageout.h>
#include <vm/vm_page.h>
#include <vm/vm_pagequeue.h>
struct ktls_wq {
struct mtx mtx;
STAILQ_HEAD(, mbuf) m_head;
STAILQ_HEAD(, socket) so_head;
bool running;
int lastallocfail;
} __aligned(CACHE_LINE_SIZE);
struct ktls_reclaim_thread {
uint64_t wakeups;
uint64_t reclaims;
struct thread *td;
int running;
};
struct ktls_domain_info {
int count;
int cpu[MAXCPU];
struct ktls_reclaim_thread reclaim_td;
};
struct ktls_domain_info ktls_domains[MAXMEMDOM];
static struct ktls_wq *ktls_wq;
static struct proc *ktls_proc;
static uma_zone_t ktls_session_zone;
static uma_zone_t ktls_buffer_zone;
static uint16_t ktls_cpuid_lookup[MAXCPU];
static int ktls_init_state;
static struct sx ktls_init_lock;
SX_SYSINIT(ktls_init_lock, &ktls_init_lock, "ktls init");
SYSCTL_NODE(_kern_ipc, OID_AUTO, tls, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
"Kernel TLS offload");
SYSCTL_NODE(_kern_ipc_tls, OID_AUTO, stats, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
"Kernel TLS offload stats");
#ifdef RSS
static int ktls_bind_threads = 1;
#else
static int ktls_bind_threads;
#endif
SYSCTL_INT(_kern_ipc_tls, OID_AUTO, bind_threads, CTLFLAG_RDTUN,
&ktls_bind_threads, 0,
"Bind crypto threads to cores (1) or cores and domains (2) at boot");
static u_int ktls_maxlen = 16384;
SYSCTL_UINT(_kern_ipc_tls, OID_AUTO, maxlen, CTLFLAG_RDTUN,
&ktls_maxlen, 0, "Maximum TLS record size");
static int ktls_number_threads;
SYSCTL_INT(_kern_ipc_tls_stats, OID_AUTO, threads, CTLFLAG_RD,
&ktls_number_threads, 0,
"Number of TLS threads in thread-pool");
unsigned int ktls_ifnet_max_rexmit_pct = 2;
SYSCTL_UINT(_kern_ipc_tls, OID_AUTO, ifnet_max_rexmit_pct, CTLFLAG_RWTUN,
&ktls_ifnet_max_rexmit_pct, 2,
"Max percent bytes retransmitted before ifnet TLS is disabled");
static bool ktls_offload_enable;
SYSCTL_BOOL(_kern_ipc_tls, OID_AUTO, enable, CTLFLAG_RWTUN,
&ktls_offload_enable, 0,
"Enable support for kernel TLS offload");
static bool ktls_cbc_enable = true;
SYSCTL_BOOL(_kern_ipc_tls, OID_AUTO, cbc_enable, CTLFLAG_RWTUN,
&ktls_cbc_enable, 1,
"Enable support of AES-CBC crypto for kernel TLS");
static bool ktls_sw_buffer_cache = true;
SYSCTL_BOOL(_kern_ipc_tls, OID_AUTO, sw_buffer_cache, CTLFLAG_RDTUN,
&ktls_sw_buffer_cache, 1,
"Enable caching of output buffers for SW encryption");
static int ktls_max_reclaim = 1024;
SYSCTL_INT(_kern_ipc_tls, OID_AUTO, max_reclaim, CTLFLAG_RWTUN,
&ktls_max_reclaim, 128,
"Max number of 16k buffers to reclaim in thread context");
static COUNTER_U64_DEFINE_EARLY(ktls_tasks_active);
SYSCTL_COUNTER_U64(_kern_ipc_tls, OID_AUTO, tasks_active, CTLFLAG_RD,
&ktls_tasks_active, "Number of active tasks");
static COUNTER_U64_DEFINE_EARLY(ktls_cnt_tx_pending);
SYSCTL_COUNTER_U64(_kern_ipc_tls_stats, OID_AUTO, sw_tx_pending, CTLFLAG_RD,
&ktls_cnt_tx_pending,
"Number of TLS 1.0 records waiting for earlier TLS records");
static COUNTER_U64_DEFINE_EARLY(ktls_cnt_tx_queued);
SYSCTL_COUNTER_U64(_kern_ipc_tls_stats, OID_AUTO, sw_tx_inqueue, CTLFLAG_RD,
&ktls_cnt_tx_queued,
"Number of TLS records in queue to tasks for SW encryption");
static COUNTER_U64_DEFINE_EARLY(ktls_cnt_rx_queued);
SYSCTL_COUNTER_U64(_kern_ipc_tls_stats, OID_AUTO, sw_rx_inqueue, CTLFLAG_RD,
&ktls_cnt_rx_queued,
"Number of TLS sockets in queue to tasks for SW decryption");
static COUNTER_U64_DEFINE_EARLY(ktls_offload_total);
SYSCTL_COUNTER_U64(_kern_ipc_tls_stats, OID_AUTO, offload_total,
CTLFLAG_RD, &ktls_offload_total,
"Total successful TLS setups (parameters set)");
static COUNTER_U64_DEFINE_EARLY(ktls_offload_enable_calls);
SYSCTL_COUNTER_U64(_kern_ipc_tls_stats, OID_AUTO, enable_calls,
CTLFLAG_RD, &ktls_offload_enable_calls,
"Total number of TLS enable calls made");
static COUNTER_U64_DEFINE_EARLY(ktls_offload_active);
SYSCTL_COUNTER_U64(_kern_ipc_tls_stats, OID_AUTO, active, CTLFLAG_RD,
&ktls_offload_active, "Total Active TLS sessions");
static COUNTER_U64_DEFINE_EARLY(ktls_offload_corrupted_records);
SYSCTL_COUNTER_U64(_kern_ipc_tls_stats, OID_AUTO, corrupted_records, CTLFLAG_RD,
&ktls_offload_corrupted_records, "Total corrupted TLS records received");
static COUNTER_U64_DEFINE_EARLY(ktls_offload_failed_crypto);
SYSCTL_COUNTER_U64(_kern_ipc_tls_stats, OID_AUTO, failed_crypto, CTLFLAG_RD,
&ktls_offload_failed_crypto, "Total TLS crypto failures");
static COUNTER_U64_DEFINE_EARLY(ktls_switch_to_ifnet);
SYSCTL_COUNTER_U64(_kern_ipc_tls_stats, OID_AUTO, switch_to_ifnet, CTLFLAG_RD,
&ktls_switch_to_ifnet, "TLS sessions switched from SW to ifnet");
static COUNTER_U64_DEFINE_EARLY(ktls_switch_to_sw);
SYSCTL_COUNTER_U64(_kern_ipc_tls_stats, OID_AUTO, switch_to_sw, CTLFLAG_RD,
&ktls_switch_to_sw, "TLS sessions switched from ifnet to SW");
static COUNTER_U64_DEFINE_EARLY(ktls_switch_failed);
SYSCTL_COUNTER_U64(_kern_ipc_tls_stats, OID_AUTO, switch_failed, CTLFLAG_RD,
&ktls_switch_failed, "TLS sessions unable to switch between SW and ifnet");
static COUNTER_U64_DEFINE_EARLY(ktls_ifnet_disable_fail);
SYSCTL_COUNTER_U64(_kern_ipc_tls_stats, OID_AUTO, ifnet_disable_failed, CTLFLAG_RD,
&ktls_ifnet_disable_fail, "TLS sessions unable to switch to SW from ifnet");
static COUNTER_U64_DEFINE_EARLY(ktls_ifnet_disable_ok);
SYSCTL_COUNTER_U64(_kern_ipc_tls_stats, OID_AUTO, ifnet_disable_ok, CTLFLAG_RD,
&ktls_ifnet_disable_ok, "TLS sessions able to switch to SW from ifnet");
static COUNTER_U64_DEFINE_EARLY(ktls_destroy_task);
SYSCTL_COUNTER_U64(_kern_ipc_tls_stats, OID_AUTO, destroy_task, CTLFLAG_RD,
&ktls_destroy_task,
"Number of times ktls session was destroyed via taskqueue");
SYSCTL_NODE(_kern_ipc_tls, OID_AUTO, sw, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
"Software TLS session stats");
SYSCTL_NODE(_kern_ipc_tls, OID_AUTO, ifnet, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
"Hardware (ifnet) TLS session stats");
#ifdef TCP_OFFLOAD
SYSCTL_NODE(_kern_ipc_tls, OID_AUTO, toe, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
"TOE TLS session stats");
#endif
static COUNTER_U64_DEFINE_EARLY(ktls_sw_cbc);
SYSCTL_COUNTER_U64(_kern_ipc_tls_sw, OID_AUTO, cbc, CTLFLAG_RD, &ktls_sw_cbc,
"Active number of software TLS sessions using AES-CBC");
static COUNTER_U64_DEFINE_EARLY(ktls_sw_gcm);
SYSCTL_COUNTER_U64(_kern_ipc_tls_sw, OID_AUTO, gcm, CTLFLAG_RD, &ktls_sw_gcm,
"Active number of software TLS sessions using AES-GCM");
static COUNTER_U64_DEFINE_EARLY(ktls_sw_chacha20);
SYSCTL_COUNTER_U64(_kern_ipc_tls_sw, OID_AUTO, chacha20, CTLFLAG_RD,
&ktls_sw_chacha20,
"Active number of software TLS sessions using Chacha20-Poly1305");
static COUNTER_U64_DEFINE_EARLY(ktls_ifnet_cbc);
SYSCTL_COUNTER_U64(_kern_ipc_tls_ifnet, OID_AUTO, cbc, CTLFLAG_RD,
&ktls_ifnet_cbc,
"Active number of ifnet TLS sessions using AES-CBC");
static COUNTER_U64_DEFINE_EARLY(ktls_ifnet_gcm);
SYSCTL_COUNTER_U64(_kern_ipc_tls_ifnet, OID_AUTO, gcm, CTLFLAG_RD,
&ktls_ifnet_gcm,
"Active number of ifnet TLS sessions using AES-GCM");
static COUNTER_U64_DEFINE_EARLY(ktls_ifnet_chacha20);
SYSCTL_COUNTER_U64(_kern_ipc_tls_ifnet, OID_AUTO, chacha20, CTLFLAG_RD,
&ktls_ifnet_chacha20,
"Active number of ifnet TLS sessions using Chacha20-Poly1305");
static COUNTER_U64_DEFINE_EARLY(ktls_ifnet_reset);
SYSCTL_COUNTER_U64(_kern_ipc_tls_ifnet, OID_AUTO, reset, CTLFLAG_RD,
&ktls_ifnet_reset, "TLS sessions updated to a new ifnet send tag");
static COUNTER_U64_DEFINE_EARLY(ktls_ifnet_reset_dropped);
SYSCTL_COUNTER_U64(_kern_ipc_tls_ifnet, OID_AUTO, reset_dropped, CTLFLAG_RD,
&ktls_ifnet_reset_dropped,
"TLS sessions dropped after failing to update ifnet send tag");
static COUNTER_U64_DEFINE_EARLY(ktls_ifnet_reset_failed);
SYSCTL_COUNTER_U64(_kern_ipc_tls_ifnet, OID_AUTO, reset_failed, CTLFLAG_RD,
&ktls_ifnet_reset_failed,
"TLS sessions that failed to allocate a new ifnet send tag");
static int ktls_ifnet_permitted;
SYSCTL_UINT(_kern_ipc_tls_ifnet, OID_AUTO, permitted, CTLFLAG_RWTUN,
&ktls_ifnet_permitted, 1,
"Whether to permit hardware (ifnet) TLS sessions");
#ifdef TCP_OFFLOAD
static COUNTER_U64_DEFINE_EARLY(ktls_toe_cbc);
SYSCTL_COUNTER_U64(_kern_ipc_tls_toe, OID_AUTO, cbc, CTLFLAG_RD,
&ktls_toe_cbc,
"Active number of TOE TLS sessions using AES-CBC");
static COUNTER_U64_DEFINE_EARLY(ktls_toe_gcm);
SYSCTL_COUNTER_U64(_kern_ipc_tls_toe, OID_AUTO, gcm, CTLFLAG_RD,
&ktls_toe_gcm,
"Active number of TOE TLS sessions using AES-GCM");
static COUNTER_U64_DEFINE_EARLY(ktls_toe_chacha20);
SYSCTL_COUNTER_U64(_kern_ipc_tls_toe, OID_AUTO, chacha20, CTLFLAG_RD,
&ktls_toe_chacha20,
"Active number of TOE TLS sessions using Chacha20-Poly1305");
#endif
static MALLOC_DEFINE(M_KTLS, "ktls", "Kernel TLS");
static void ktls_reset_receive_tag(void *context, int pending);
static void ktls_reset_send_tag(void *context, int pending);
static void ktls_work_thread(void *ctx);
static void ktls_reclaim_thread(void *ctx);
static u_int
ktls_get_cpu(struct socket *so)
{
struct inpcb *inp;
#ifdef NUMA
struct ktls_domain_info *di;
#endif
u_int cpuid;
inp = sotoinpcb(so);
#ifdef RSS
cpuid = rss_hash2cpuid(inp->inp_flowid, inp->inp_flowtype);
if (cpuid != NETISR_CPUID_NONE)
return (cpuid);
#endif
/*
* Just use the flowid to shard connections in a repeatable
* fashion. Note that TLS 1.0 sessions rely on the
* serialization provided by having the same connection use
* the same queue.
*/
#ifdef NUMA
if (ktls_bind_threads > 1 && inp->inp_numa_domain != M_NODOM) {
di = &ktls_domains[inp->inp_numa_domain];
cpuid = di->cpu[inp->inp_flowid % di->count];
} else
#endif
cpuid = ktls_cpuid_lookup[inp->inp_flowid % ktls_number_threads];
return (cpuid);
}
static int
ktls_buffer_import(void *arg, void **store, int count, int domain, int flags)
{
vm_page_t m;
int i, req;
KASSERT((ktls_maxlen & PAGE_MASK) == 0,
("%s: ktls max length %d is not page size-aligned",
__func__, ktls_maxlen));
req = VM_ALLOC_WIRED | VM_ALLOC_NODUMP | malloc2vm_flags(flags);
for (i = 0; i < count; i++) {
m = vm_page_alloc_noobj_contig_domain(domain, req,
atop(ktls_maxlen), 0, ~0ul, PAGE_SIZE, 0,
VM_MEMATTR_DEFAULT);
if (m == NULL)
break;
store[i] = (void *)PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m));
}
return (i);
}
static void
ktls_buffer_release(void *arg __unused, void **store, int count)
{
vm_page_t m;
int i, j;
for (i = 0; i < count; i++) {
m = PHYS_TO_VM_PAGE(DMAP_TO_PHYS((vm_offset_t)store[i]));
for (j = 0; j < atop(ktls_maxlen); j++) {
(void)vm_page_unwire_noq(m + j);
vm_page_free(m + j);
}
}
}
static void
ktls_free_mext_contig(struct mbuf *m)
{
M_ASSERTEXTPG(m);
uma_zfree(ktls_buffer_zone, (void *)PHYS_TO_DMAP(m->m_epg_pa[0]));
}
static int
ktls_init(void)
{
struct thread *td;
struct pcpu *pc;
int count, domain, error, i;
ktls_wq = malloc(sizeof(*ktls_wq) * (mp_maxid + 1), M_KTLS,
M_WAITOK | M_ZERO);
ktls_session_zone = uma_zcreate("ktls_session",
sizeof(struct ktls_session),
NULL, NULL, NULL, NULL,
UMA_ALIGN_CACHE, 0);
if (ktls_sw_buffer_cache) {
ktls_buffer_zone = uma_zcache_create("ktls_buffers",
roundup2(ktls_maxlen, PAGE_SIZE), NULL, NULL, NULL, NULL,
ktls_buffer_import, ktls_buffer_release, NULL,
UMA_ZONE_FIRSTTOUCH);
}
/*
* Initialize the workqueues to run the TLS work. We create a
* work queue for each CPU.
*/
CPU_FOREACH(i) {
STAILQ_INIT(&ktls_wq[i].m_head);
STAILQ_INIT(&ktls_wq[i].so_head);
mtx_init(&ktls_wq[i].mtx, "ktls work queue", NULL, MTX_DEF);
if (ktls_bind_threads > 1) {
pc = pcpu_find(i);
domain = pc->pc_domain;
count = ktls_domains[domain].count;
ktls_domains[domain].cpu[count] = i;
ktls_domains[domain].count++;
}
ktls_cpuid_lookup[ktls_number_threads] = i;
ktls_number_threads++;
}
/*
* If we somehow have an empty domain, fall back to choosing
* among all KTLS threads.
*/
if (ktls_bind_threads > 1) {
for (i = 0; i < vm_ndomains; i++) {
if (ktls_domains[i].count == 0) {
ktls_bind_threads = 1;
break;
}
}
}
/* Start kthreads for each workqueue. */
CPU_FOREACH(i) {
error = kproc_kthread_add(ktls_work_thread, &ktls_wq[i],
&ktls_proc, &td, 0, 0, "KTLS", "thr_%d", i);
if (error) {
printf("Can't add KTLS thread %d error %d\n", i, error);
return (error);
}
}
/*
* Start an allocation thread per-domain to perform blocking allocations
* of 16k physically contiguous TLS crypto destination buffers.
*/
if (ktls_sw_buffer_cache) {
for (domain = 0; domain < vm_ndomains; domain++) {
if (VM_DOMAIN_EMPTY(domain))
continue;
if (CPU_EMPTY(&cpuset_domain[domain]))
continue;
error = kproc_kthread_add(ktls_reclaim_thread,
&ktls_domains[domain], &ktls_proc,
&ktls_domains[domain].reclaim_td.td,
0, 0, "KTLS", "reclaim_%d", domain);
if (error) {
printf("Can't add KTLS reclaim thread %d error %d\n",
domain, error);
return (error);
}
}
}
if (bootverbose)
printf("KTLS: Initialized %d threads\n", ktls_number_threads);
return (0);
}
static int
ktls_start_kthreads(void)
{
int error, state;
start:
state = atomic_load_acq_int(&ktls_init_state);
if (__predict_true(state > 0))
return (0);
if (state < 0)
return (ENXIO);
sx_xlock(&ktls_init_lock);
if (ktls_init_state != 0) {
sx_xunlock(&ktls_init_lock);
goto start;
}
error = ktls_init();
if (error == 0)
state = 1;
else
state = -1;
atomic_store_rel_int(&ktls_init_state, state);
sx_xunlock(&ktls_init_lock);
return (error);
}
static int
ktls_create_session(struct socket *so, struct tls_enable *en,
struct ktls_session **tlsp, int direction)
{
struct ktls_session *tls;
int error;
/* Only TLS 1.0 - 1.3 are supported. */
if (en->tls_vmajor != TLS_MAJOR_VER_ONE)
return (EINVAL);
if (en->tls_vminor < TLS_MINOR_VER_ZERO ||
en->tls_vminor > TLS_MINOR_VER_THREE)
return (EINVAL);
if (en->auth_key_len < 0 || en->auth_key_len > TLS_MAX_PARAM_SIZE)
return (EINVAL);
if (en->cipher_key_len < 0 || en->cipher_key_len > TLS_MAX_PARAM_SIZE)
return (EINVAL);
if (en->iv_len < 0 || en->iv_len > sizeof(tls->params.iv))
return (EINVAL);
/* All supported algorithms require a cipher key. */
if (en->cipher_key_len == 0)
return (EINVAL);
/* No flags are currently supported. */
if (en->flags != 0)
return (EINVAL);
/* Common checks for supported algorithms. */
switch (en->cipher_algorithm) {
case CRYPTO_AES_NIST_GCM_16:
/*
* auth_algorithm isn't used, but permit GMAC values
* for compatibility.
*/
switch (en->auth_algorithm) {
case 0:
#ifdef COMPAT_FREEBSD12
/* XXX: Really 13.0-current COMPAT. */
case CRYPTO_AES_128_NIST_GMAC:
case CRYPTO_AES_192_NIST_GMAC:
case CRYPTO_AES_256_NIST_GMAC:
#endif
break;
default:
return (EINVAL);
}
if (en->auth_key_len != 0)
return (EINVAL);
switch (en->tls_vminor) {
case TLS_MINOR_VER_TWO:
if (en->iv_len != TLS_AEAD_GCM_LEN)
return (EINVAL);
break;
case TLS_MINOR_VER_THREE:
if (en->iv_len != TLS_1_3_GCM_IV_LEN)
return (EINVAL);
break;
default:
return (EINVAL);
}
break;
case CRYPTO_AES_CBC:
switch (en->auth_algorithm) {
case CRYPTO_SHA1_HMAC:
break;
case CRYPTO_SHA2_256_HMAC:
case CRYPTO_SHA2_384_HMAC:
if (en->tls_vminor != TLS_MINOR_VER_TWO)
return (EINVAL);
break;
default:
return (EINVAL);
}
if (en->auth_key_len == 0)
return (EINVAL);
/*
* TLS 1.0 requires an implicit IV. TLS 1.1 and 1.2
* use explicit IVs.
*/
switch (en->tls_vminor) {
case TLS_MINOR_VER_ZERO:
if (en->iv_len != TLS_CBC_IMPLICIT_IV_LEN)
return (EINVAL);
break;
case TLS_MINOR_VER_ONE:
case TLS_MINOR_VER_TWO:
/* Ignore any supplied IV. */
en->iv_len = 0;
break;
default:
return (EINVAL);
}
break;
case CRYPTO_CHACHA20_POLY1305:
if (en->auth_algorithm != 0 || en->auth_key_len != 0)
return (EINVAL);
if (en->tls_vminor != TLS_MINOR_VER_TWO &&
en->tls_vminor != TLS_MINOR_VER_THREE)
return (EINVAL);
if (en->iv_len != TLS_CHACHA20_IV_LEN)
return (EINVAL);
break;
default:
return (EINVAL);
}
error = ktls_start_kthreads();
if (error != 0)
return (error);
tls = uma_zalloc(ktls_session_zone, M_WAITOK | M_ZERO);
counter_u64_add(ktls_offload_active, 1);
refcount_init(&tls->refcount, 1);
if (direction == KTLS_RX) {
TASK_INIT(&tls->reset_tag_task, 0, ktls_reset_receive_tag, tls);
} else {
TASK_INIT(&tls->reset_tag_task, 0, ktls_reset_send_tag, tls);
tls->inp = so->so_pcb;
in_pcbref(tls->inp);
tls->tx = true;
}
tls->wq_index = ktls_get_cpu(so);
tls->params.cipher_algorithm = en->cipher_algorithm;
tls->params.auth_algorithm = en->auth_algorithm;
tls->params.tls_vmajor = en->tls_vmajor;
tls->params.tls_vminor = en->tls_vminor;
tls->params.flags = en->flags;
tls->params.max_frame_len = min(TLS_MAX_MSG_SIZE_V10_2, ktls_maxlen);
/* Set the header and trailer lengths. */
tls->params.tls_hlen = sizeof(struct tls_record_layer);
switch (en->cipher_algorithm) {
case CRYPTO_AES_NIST_GCM_16:
/*
* TLS 1.2 uses a 4 byte implicit IV with an explicit 8 byte
* nonce. TLS 1.3 uses a 12 byte implicit IV.
*/
if (en->tls_vminor < TLS_MINOR_VER_THREE)
tls->params.tls_hlen += sizeof(uint64_t);
tls->params.tls_tlen = AES_GMAC_HASH_LEN;
tls->params.tls_bs = 1;
break;
case CRYPTO_AES_CBC:
switch (en->auth_algorithm) {
case CRYPTO_SHA1_HMAC:
if (en->tls_vminor == TLS_MINOR_VER_ZERO) {
/* Implicit IV, no nonce. */
tls->sequential_records = true;
tls->next_seqno = be64dec(en->rec_seq);
STAILQ_INIT(&tls->pending_records);
} else {
tls->params.tls_hlen += AES_BLOCK_LEN;
}
tls->params.tls_tlen = AES_BLOCK_LEN +
SHA1_HASH_LEN;
break;
case CRYPTO_SHA2_256_HMAC:
tls->params.tls_hlen += AES_BLOCK_LEN;
tls->params.tls_tlen = AES_BLOCK_LEN +
SHA2_256_HASH_LEN;
break;
case CRYPTO_SHA2_384_HMAC:
tls->params.tls_hlen += AES_BLOCK_LEN;
tls->params.tls_tlen = AES_BLOCK_LEN +
SHA2_384_HASH_LEN;
break;
default:
panic("invalid hmac");
}
tls->params.tls_bs = AES_BLOCK_LEN;
break;
case CRYPTO_CHACHA20_POLY1305:
/*
* Chacha20 uses a 12 byte implicit IV.
*/
tls->params.tls_tlen = POLY1305_HASH_LEN;
tls->params.tls_bs = 1;
break;
default:
panic("invalid cipher");
}
/*
* TLS 1.3 includes optional padding which we do not support,
* and also puts the "real" record type at the end of the
* encrypted data.
*/
if (en->tls_vminor == TLS_MINOR_VER_THREE)
tls->params.tls_tlen += sizeof(uint8_t);
KASSERT(tls->params.tls_hlen <= MBUF_PEXT_HDR_LEN,
("TLS header length too long: %d", tls->params.tls_hlen));
KASSERT(tls->params.tls_tlen <= MBUF_PEXT_TRAIL_LEN,
("TLS trailer length too long: %d", tls->params.tls_tlen));
if (en->auth_key_len != 0) {
tls->params.auth_key_len = en->auth_key_len;
tls->params.auth_key = malloc(en->auth_key_len, M_KTLS,
M_WAITOK);
error = copyin(en->auth_key, tls->params.auth_key,
en->auth_key_len);
if (error)
goto out;
}
tls->params.cipher_key_len = en->cipher_key_len;
tls->params.cipher_key = malloc(en->cipher_key_len, M_KTLS, M_WAITOK);
error = copyin(en->cipher_key, tls->params.cipher_key,
en->cipher_key_len);
if (error)
goto out;
/*
* This holds the implicit portion of the nonce for AEAD
* ciphers and the initial implicit IV for TLS 1.0. The
* explicit portions of the IV are generated in ktls_frame().
*/
if (en->iv_len != 0) {
tls->params.iv_len = en->iv_len;
error = copyin(en->iv, tls->params.iv, en->iv_len);
if (error)
goto out;
/*
* For TLS 1.2 with GCM, generate an 8-byte nonce as a
* counter to generate unique explicit IVs.
*
* Store this counter in the last 8 bytes of the IV
* array so that it is 8-byte aligned.
*/
if (en->cipher_algorithm == CRYPTO_AES_NIST_GCM_16 &&
en->tls_vminor == TLS_MINOR_VER_TWO)
arc4rand(tls->params.iv + 8, sizeof(uint64_t), 0);
}
*tlsp = tls;
return (0);
out:
ktls_free(tls);
return (error);
}
static struct ktls_session *
ktls_clone_session(struct ktls_session *tls, int direction)
{
struct ktls_session *tls_new;
tls_new = uma_zalloc(ktls_session_zone, M_WAITOK | M_ZERO);
counter_u64_add(ktls_offload_active, 1);
refcount_init(&tls_new->refcount, 1);
if (direction == KTLS_RX) {
TASK_INIT(&tls_new->reset_tag_task, 0, ktls_reset_receive_tag,
tls_new);
} else {
TASK_INIT(&tls_new->reset_tag_task, 0, ktls_reset_send_tag,
tls_new);
tls_new->inp = tls->inp;
tls_new->tx = true;
in_pcbref(tls_new->inp);
}
/* Copy fields from existing session. */
tls_new->params = tls->params;
tls_new->wq_index = tls->wq_index;
/* Deep copy keys. */
if (tls_new->params.auth_key != NULL) {
tls_new->params.auth_key = malloc(tls->params.auth_key_len,
M_KTLS, M_WAITOK);
memcpy(tls_new->params.auth_key, tls->params.auth_key,
tls->params.auth_key_len);
}
tls_new->params.cipher_key = malloc(tls->params.cipher_key_len, M_KTLS,
M_WAITOK);
memcpy(tls_new->params.cipher_key, tls->params.cipher_key,
tls->params.cipher_key_len);
return (tls_new);
}
#ifdef TCP_OFFLOAD
static int
ktls_try_toe(struct socket *so, struct ktls_session *tls, int direction)
{
struct inpcb *inp;
struct tcpcb *tp;
int error;
inp = so->so_pcb;
INP_WLOCK(inp);
if (inp->inp_flags & INP_DROPPED) {
INP_WUNLOCK(inp);
return (ECONNRESET);
}
if (inp->inp_socket == NULL) {
INP_WUNLOCK(inp);
return (ECONNRESET);
}
tp = intotcpcb(inp);
if (!(tp->t_flags & TF_TOE)) {
INP_WUNLOCK(inp);
return (EOPNOTSUPP);
}
error = tcp_offload_alloc_tls_session(tp, tls, direction);
INP_WUNLOCK(inp);
if (error == 0) {
tls->mode = TCP_TLS_MODE_TOE;
switch (tls->params.cipher_algorithm) {
case CRYPTO_AES_CBC:
counter_u64_add(ktls_toe_cbc, 1);
break;
case CRYPTO_AES_NIST_GCM_16:
counter_u64_add(ktls_toe_gcm, 1);
break;
case CRYPTO_CHACHA20_POLY1305:
counter_u64_add(ktls_toe_chacha20, 1);
break;
}
}
return (error);
}
#endif
/*
* Common code used when first enabling ifnet TLS on a connection or
* when allocating a new ifnet TLS session due to a routing change.
* This function allocates a new TLS send tag on whatever interface
* the connection is currently routed over.
*/
static int
ktls_alloc_snd_tag(struct inpcb *inp, struct ktls_session *tls, bool force,
struct m_snd_tag **mstp)
{
union if_snd_tag_alloc_params params;
struct ifnet *ifp;
struct nhop_object *nh;
struct tcpcb *tp;
int error;
INP_RLOCK(inp);
if (inp->inp_flags & INP_DROPPED) {
INP_RUNLOCK(inp);
return (ECONNRESET);
}
if (inp->inp_socket == NULL) {
INP_RUNLOCK(inp);
return (ECONNRESET);
}
tp = intotcpcb(inp);
/*
* Check administrative controls on ifnet TLS to determine if
* ifnet TLS should be denied.
*
* - Always permit 'force' requests.
* - ktls_ifnet_permitted == 0: always deny.
*/
if (!force && ktls_ifnet_permitted == 0) {
INP_RUNLOCK(inp);
return (ENXIO);
}
/*
* XXX: Use the cached route in the inpcb to find the
* interface. This should perhaps instead use
* rtalloc1_fib(dst, 0, 0, fibnum). Since KTLS is only
* enabled after a connection has completed key negotiation in
* userland, the cached route will be present in practice.
*/
nh = inp->inp_route.ro_nh;
if (nh == NULL) {
INP_RUNLOCK(inp);
return (ENXIO);
}
ifp = nh->nh_ifp;
if_ref(ifp);
/*
* Allocate a TLS + ratelimit tag if the connection has an
* existing pacing rate.
*/
if (tp->t_pacing_rate != -1 &&
(if_getcapenable(ifp) & IFCAP_TXTLS_RTLMT) != 0) {
params.hdr.type = IF_SND_TAG_TYPE_TLS_RATE_LIMIT;
params.tls_rate_limit.inp = inp;
params.tls_rate_limit.tls = tls;
params.tls_rate_limit.max_rate = tp->t_pacing_rate;
} else {
params.hdr.type = IF_SND_TAG_TYPE_TLS;
params.tls.inp = inp;
params.tls.tls = tls;
}
params.hdr.flowid = inp->inp_flowid;
params.hdr.flowtype = inp->inp_flowtype;
params.hdr.numa_domain = inp->inp_numa_domain;
INP_RUNLOCK(inp);
if ((if_getcapenable(ifp) & IFCAP_MEXTPG) == 0) {
error = EOPNOTSUPP;
goto out;
}
if (inp->inp_vflag & INP_IPV6) {
if ((if_getcapenable(ifp) & IFCAP_TXTLS6) == 0) {
error = EOPNOTSUPP;
goto out;
}
} else {
if ((if_getcapenable(ifp) & IFCAP_TXTLS4) == 0) {
error = EOPNOTSUPP;
goto out;
}
}
error = m_snd_tag_alloc(ifp, ¶ms, mstp);
out:
if_rele(ifp);
return (error);
}
/*
* Allocate an initial TLS receive tag for doing HW decryption of TLS
* data.
*
* This function allocates a new TLS receive tag on whatever interface
* the connection is currently routed over. If the connection ends up
* using a different interface for receive this will get fixed up via
* ktls_input_ifp_mismatch as future packets arrive.
*/
static int
ktls_alloc_rcv_tag(struct inpcb *inp, struct ktls_session *tls,
struct m_snd_tag **mstp)
{
union if_snd_tag_alloc_params params;
struct ifnet *ifp;
struct nhop_object *nh;
int error;
if (!ktls_ocf_recrypt_supported(tls))
return (ENXIO);
INP_RLOCK(inp);
if (inp->inp_flags & INP_DROPPED) {
INP_RUNLOCK(inp);
return (ECONNRESET);
}
if (inp->inp_socket == NULL) {
INP_RUNLOCK(inp);
return (ECONNRESET);
}
/*
* Check administrative controls on ifnet TLS to determine if
* ifnet TLS should be denied.
*/
if (ktls_ifnet_permitted == 0) {
INP_RUNLOCK(inp);
return (ENXIO);
}
/*
* XXX: As with ktls_alloc_snd_tag, use the cached route in
* the inpcb to find the interface.
*/
nh = inp->inp_route.ro_nh;
if (nh == NULL) {
INP_RUNLOCK(inp);
return (ENXIO);
}
ifp = nh->nh_ifp;
if_ref(ifp);
tls->rx_ifp = ifp;
params.hdr.type = IF_SND_TAG_TYPE_TLS_RX;
params.hdr.flowid = inp->inp_flowid;
params.hdr.flowtype = inp->inp_flowtype;
params.hdr.numa_domain = inp->inp_numa_domain;
params.tls_rx.inp = inp;
params.tls_rx.tls = tls;
params.tls_rx.vlan_id = 0;
INP_RUNLOCK(inp);
if (inp->inp_vflag & INP_IPV6) {
if ((if_getcapenable2(ifp) & IFCAP2_BIT(IFCAP2_RXTLS6)) == 0) {
error = EOPNOTSUPP;
goto out;
}
} else {
if ((if_getcapenable2(ifp) & IFCAP2_BIT(IFCAP2_RXTLS4)) == 0) {
error = EOPNOTSUPP;
goto out;
}
}
error = m_snd_tag_alloc(ifp, ¶ms, mstp);
/*
* If this connection is over a vlan, vlan_snd_tag_alloc
* rewrites vlan_id with the saved interface. Save the VLAN
* ID for use in ktls_reset_receive_tag which allocates new
* receive tags directly from the leaf interface bypassing
* if_vlan.
*/
if (error == 0)
tls->rx_vlan_id = params.tls_rx.vlan_id;
out:
return (error);
}
static int
ktls_try_ifnet(struct socket *so, struct ktls_session *tls, int direction,
bool force)
{
struct m_snd_tag *mst;
int error;
switch (direction) {
case KTLS_TX:
error = ktls_alloc_snd_tag(so->so_pcb, tls, force, &mst);
if (__predict_false(error != 0))
goto done;
break;
case KTLS_RX:
KASSERT(!force, ("%s: forced receive tag", __func__));
error = ktls_alloc_rcv_tag(so->so_pcb, tls, &mst);
if (__predict_false(error != 0))
goto done;
break;
default:
__assert_unreachable();
}
tls->mode = TCP_TLS_MODE_IFNET;
tls->snd_tag = mst;
switch (tls->params.cipher_algorithm) {
case CRYPTO_AES_CBC:
counter_u64_add(ktls_ifnet_cbc, 1);
break;
case CRYPTO_AES_NIST_GCM_16:
counter_u64_add(ktls_ifnet_gcm, 1);
break;
case CRYPTO_CHACHA20_POLY1305:
counter_u64_add(ktls_ifnet_chacha20, 1);
break;
default:
break;
}
done:
return (error);
}
static void
ktls_use_sw(struct ktls_session *tls)
{
tls->mode = TCP_TLS_MODE_SW;
switch (tls->params.cipher_algorithm) {
case CRYPTO_AES_CBC:
counter_u64_add(ktls_sw_cbc, 1);
break;
case CRYPTO_AES_NIST_GCM_16:
counter_u64_add(ktls_sw_gcm, 1);
break;
case CRYPTO_CHACHA20_POLY1305:
counter_u64_add(ktls_sw_chacha20, 1);
break;
}
}
static int
ktls_try_sw(struct ktls_session *tls, int direction)
{
int error;
error = ktls_ocf_try(tls, direction);
if (error)
return (error);
ktls_use_sw(tls);
return (0);
}
/*
* KTLS RX stores data in the socket buffer as a list of TLS records,
* where each record is stored as a control message containg the TLS
* header followed by data mbufs containing the decrypted data. This
* is different from KTLS TX which always uses an mb_ext_pgs mbuf for
* both encrypted and decrypted data. TLS records decrypted by a NIC
* should be queued to the socket buffer as records, but encrypted
* data which needs to be decrypted by software arrives as a stream of
* regular mbufs which need to be converted. In addition, there may
* already be pending encrypted data in the socket buffer when KTLS RX
* is enabled.
*
* To manage not-yet-decrypted data for KTLS RX, the following scheme
* is used:
*
* - A single chain of NOTREADY mbufs is hung off of sb_mtls.
*
* - ktls_check_rx checks this chain of mbufs reading the TLS header
* from the first mbuf. Once all of the data for that TLS record is
* queued, the socket is queued to a worker thread.
*
* - The worker thread calls ktls_decrypt to decrypt TLS records in
* the TLS chain. Each TLS record is detached from the TLS chain,
* decrypted, and inserted into the regular socket buffer chain as
* record starting with a control message holding the TLS header and
* a chain of mbufs holding the encrypted data.
*/
static void
sb_mark_notready(struct sockbuf *sb)
{
struct mbuf *m;
m = sb->sb_mb;
sb->sb_mtls = m;
sb->sb_mb = NULL;
sb->sb_mbtail = NULL;
sb->sb_lastrecord = NULL;
for (; m != NULL; m = m->m_next) {
KASSERT(m->m_nextpkt == NULL, ("%s: m_nextpkt != NULL",
__func__));
KASSERT((m->m_flags & M_NOTAVAIL) == 0, ("%s: mbuf not avail",
__func__));
KASSERT(sb->sb_acc >= m->m_len, ("%s: sb_acc < m->m_len",
__func__));
m->m_flags |= M_NOTREADY;
sb->sb_acc -= m->m_len;
sb->sb_tlscc += m->m_len;
sb->sb_mtlstail = m;
}
KASSERT(sb->sb_acc == 0 && sb->sb_tlscc == sb->sb_ccc,
("%s: acc %u tlscc %u ccc %u", __func__, sb->sb_acc, sb->sb_tlscc,
sb->sb_ccc));
}
/*
* Return information about the pending TLS data in a socket
* buffer. On return, 'seqno' is set to the sequence number
* of the next TLS record to be received, 'resid' is set to
* the amount of bytes still needed for the last pending
* record. The function returns 'false' if the last pending
* record contains a partial TLS header. In that case, 'resid'
* is the number of bytes needed to complete the TLS header.
*/
bool
ktls_pending_rx_info(struct sockbuf *sb, uint64_t *seqnop, size_t *residp)
{
struct tls_record_layer hdr;
struct mbuf *m;
uint64_t seqno;
size_t resid;
u_int offset, record_len;
SOCKBUF_LOCK_ASSERT(sb);
MPASS(sb->sb_flags & SB_TLS_RX);
seqno = sb->sb_tls_seqno;
resid = sb->sb_tlscc;
m = sb->sb_mtls;
offset = 0;
if (resid == 0) {
*seqnop = seqno;
*residp = 0;
return (true);
}
for (;;) {
seqno++;
if (resid < sizeof(hdr)) {
*seqnop = seqno;
*residp = sizeof(hdr) - resid;
return (false);
}
m_copydata(m, offset, sizeof(hdr), (void *)&hdr);
record_len = sizeof(hdr) + ntohs(hdr.tls_length);
if (resid <= record_len) {
*seqnop = seqno;
*residp = record_len - resid;
return (true);
}
resid -= record_len;
while (record_len != 0) {
if (m->m_len - offset > record_len) {
offset += record_len;
break;
}
record_len -= (m->m_len - offset);
offset = 0;
m = m->m_next;
}
}
}
int
ktls_enable_rx(struct socket *so, struct tls_enable *en)
{
struct ktls_session *tls;
int error;
if (!ktls_offload_enable)
return (ENOTSUP);
counter_u64_add(ktls_offload_enable_calls, 1);
/*
* This should always be true since only the TCP socket option
* invokes this function.
*/
if (so->so_proto->pr_protocol != IPPROTO_TCP)
return (EINVAL);
/*
* XXX: Don't overwrite existing sessions. We should permit
* this to support rekeying in the future.
*/
if (so->so_rcv.sb_tls_info != NULL)
return (EALREADY);
if (en->cipher_algorithm == CRYPTO_AES_CBC && !ktls_cbc_enable)
return (ENOTSUP);
error = ktls_create_session(so, en, &tls, KTLS_RX);
if (error)
return (error);
error = ktls_ocf_try(tls, KTLS_RX);
if (error) {
ktls_free(tls);
return (error);
}
+ /*
+ * Serialize with soreceive_generic() and make sure that we're not
+ * operating on a listening socket.
+ */
+ error = SOCK_IO_RECV_LOCK(so, SBL_WAIT);
+ if (error) {
+ ktls_free(tls);
+ return (error);
+ }
+
/* Mark the socket as using TLS offload. */
SOCK_RECVBUF_LOCK(so);
- if (SOLISTENING(so)) {
+ if (__predict_false(so->so_rcv.sb_tls_info != NULL)) {
SOCK_RECVBUF_UNLOCK(so);
+ SOCK_IO_RECV_UNLOCK(so);
ktls_free(tls);
- return (EINVAL);
+ return (EALREADY);
}
so->so_rcv.sb_tls_seqno = be64dec(en->rec_seq);
so->so_rcv.sb_tls_info = tls;
so->so_rcv.sb_flags |= SB_TLS_RX;
/* Mark existing data as not ready until it can be decrypted. */
sb_mark_notready(&so->so_rcv);
ktls_check_rx(&so->so_rcv);
SOCK_RECVBUF_UNLOCK(so);
+ SOCK_IO_RECV_UNLOCK(so);
/* Prefer TOE -> ifnet TLS -> software TLS. */
#ifdef TCP_OFFLOAD
error = ktls_try_toe(so, tls, KTLS_RX);
if (error)
#endif
error = ktls_try_ifnet(so, tls, KTLS_RX, false);
if (error)
ktls_use_sw(tls);
counter_u64_add(ktls_offload_total, 1);
return (0);
}
int
ktls_enable_tx(struct socket *so, struct tls_enable *en)
{
struct ktls_session *tls;
struct inpcb *inp;
struct tcpcb *tp;
int error;
if (!ktls_offload_enable)
return (ENOTSUP);
counter_u64_add(ktls_offload_enable_calls, 1);
/*
* This should always be true since only the TCP socket option
* invokes this function.
*/
if (so->so_proto->pr_protocol != IPPROTO_TCP)
return (EINVAL);
/*
* XXX: Don't overwrite existing sessions. We should permit
* this to support rekeying in the future.
*/
if (so->so_snd.sb_tls_info != NULL)
return (EALREADY);
if (en->cipher_algorithm == CRYPTO_AES_CBC && !ktls_cbc_enable)
return (ENOTSUP);
/* TLS requires ext pgs */
if (mb_use_ext_pgs == 0)
return (ENXIO);
error = ktls_create_session(so, en, &tls, KTLS_TX);
if (error)
return (error);
/* Prefer TOE -> ifnet TLS -> software TLS. */
#ifdef TCP_OFFLOAD
error = ktls_try_toe(so, tls, KTLS_TX);
if (error)
#endif
error = ktls_try_ifnet(so, tls, KTLS_TX, false);
if (error)
error = ktls_try_sw(tls, KTLS_TX);
if (error) {
ktls_free(tls);
return (error);
}
/*
* Serialize with sosend_generic() and make sure that we're not
* operating on a listening socket.
*/
error = SOCK_IO_SEND_LOCK(so, SBL_WAIT);
if (error) {
ktls_free(tls);
return (error);
}
/*
* Write lock the INP when setting sb_tls_info so that
* routines in tcp_ratelimit.c can read sb_tls_info while
* holding the INP lock.
*/
inp = so->so_pcb;
INP_WLOCK(inp);
SOCK_SENDBUF_LOCK(so);
+ if (__predict_false(so->so_snd.sb_tls_info != NULL)) {
+ SOCK_SENDBUF_UNLOCK(so);
+ INP_WUNLOCK(inp);
+ SOCK_IO_SEND_UNLOCK(so);
+ ktls_free(tls);
+ return (EALREADY);
+ }
so->so_snd.sb_tls_seqno = be64dec(en->rec_seq);
so->so_snd.sb_tls_info = tls;
if (tls->mode != TCP_TLS_MODE_SW) {
tp = intotcpcb(inp);
MPASS(tp->t_nic_ktls_xmit == 0);
tp->t_nic_ktls_xmit = 1;
if (tp->t_fb->tfb_hwtls_change != NULL)
(*tp->t_fb->tfb_hwtls_change)(tp, 1);
}
SOCK_SENDBUF_UNLOCK(so);
INP_WUNLOCK(inp);
SOCK_IO_SEND_UNLOCK(so);
counter_u64_add(ktls_offload_total, 1);
return (0);
}
int
ktls_get_rx_mode(struct socket *so, int *modep)
{
struct ktls_session *tls;
struct inpcb *inp __diagused;
if (SOLISTENING(so))
return (EINVAL);
inp = so->so_pcb;
INP_WLOCK_ASSERT(inp);
SOCK_RECVBUF_LOCK(so);
tls = so->so_rcv.sb_tls_info;
if (tls == NULL)
*modep = TCP_TLS_MODE_NONE;
else
*modep = tls->mode;
SOCK_RECVBUF_UNLOCK(so);
return (0);
}
/*
* ktls_get_rx_sequence - get the next TCP- and TLS- sequence number.
*
* This function gets information about the next TCP- and TLS-
* sequence number to be processed by the TLS receive worker
* thread. The information is extracted from the given "inpcb"
* structure. The values are stored in host endian format at the two
* given output pointer locations. The TCP sequence number points to
* the beginning of the TLS header.
*
* This function returns zero on success, else a non-zero error code
* is returned.
*/
int
ktls_get_rx_sequence(struct inpcb *inp, uint32_t *tcpseq, uint64_t *tlsseq)
{
struct socket *so;
struct tcpcb *tp;
INP_RLOCK(inp);
so = inp->inp_socket;
if (__predict_false(so == NULL)) {
INP_RUNLOCK(inp);
return (EINVAL);
}
if (inp->inp_flags & INP_DROPPED) {
INP_RUNLOCK(inp);
return (ECONNRESET);
}
tp = intotcpcb(inp);
MPASS(tp != NULL);
SOCKBUF_LOCK(&so->so_rcv);
*tcpseq = tp->rcv_nxt - so->so_rcv.sb_tlscc;
*tlsseq = so->so_rcv.sb_tls_seqno;
SOCKBUF_UNLOCK(&so->so_rcv);
INP_RUNLOCK(inp);
return (0);
}
int
ktls_get_tx_mode(struct socket *so, int *modep)
{
struct ktls_session *tls;
struct inpcb *inp __diagused;
if (SOLISTENING(so))
return (EINVAL);
inp = so->so_pcb;
INP_WLOCK_ASSERT(inp);
SOCK_SENDBUF_LOCK(so);
tls = so->so_snd.sb_tls_info;
if (tls == NULL)
*modep = TCP_TLS_MODE_NONE;
else
*modep = tls->mode;
SOCK_SENDBUF_UNLOCK(so);
return (0);
}
/*
* Switch between SW and ifnet TLS sessions as requested.
*/
int
ktls_set_tx_mode(struct socket *so, int mode)
{
struct ktls_session *tls, *tls_new;
struct inpcb *inp;
struct tcpcb *tp;
int error;
if (SOLISTENING(so))
return (EINVAL);
switch (mode) {
case TCP_TLS_MODE_SW:
case TCP_TLS_MODE_IFNET:
break;
default:
return (EINVAL);
}
inp = so->so_pcb;
INP_WLOCK_ASSERT(inp);
tp = intotcpcb(inp);
if (mode == TCP_TLS_MODE_IFNET) {
/* Don't allow enabling ifnet ktls multiple times */
if (tp->t_nic_ktls_xmit)
return (EALREADY);
/*
* Don't enable ifnet ktls if we disabled it due to an
* excessive retransmission rate
*/
if (tp->t_nic_ktls_xmit_dis)
return (ENXIO);
}
SOCKBUF_LOCK(&so->so_snd);
tls = so->so_snd.sb_tls_info;
if (tls == NULL) {
SOCKBUF_UNLOCK(&so->so_snd);
return (0);
}
if (tls->mode == mode) {
SOCKBUF_UNLOCK(&so->so_snd);
return (0);
}
tls = ktls_hold(tls);
SOCKBUF_UNLOCK(&so->so_snd);
INP_WUNLOCK(inp);
tls_new = ktls_clone_session(tls, KTLS_TX);
if (mode == TCP_TLS_MODE_IFNET)
error = ktls_try_ifnet(so, tls_new, KTLS_TX, true);
else
error = ktls_try_sw(tls_new, KTLS_TX);
if (error) {
counter_u64_add(ktls_switch_failed, 1);
ktls_free(tls_new);
ktls_free(tls);
INP_WLOCK(inp);
return (error);
}
error = SOCK_IO_SEND_LOCK(so, SBL_WAIT);
if (error) {
counter_u64_add(ktls_switch_failed, 1);
ktls_free(tls_new);
ktls_free(tls);
INP_WLOCK(inp);
return (error);
}
/*
* If we raced with another session change, keep the existing
* session.
*/
if (tls != so->so_snd.sb_tls_info) {
counter_u64_add(ktls_switch_failed, 1);
SOCK_IO_SEND_UNLOCK(so);
ktls_free(tls_new);
ktls_free(tls);
INP_WLOCK(inp);
return (EBUSY);
}
INP_WLOCK(inp);
SOCKBUF_LOCK(&so->so_snd);
so->so_snd.sb_tls_info = tls_new;
if (tls_new->mode != TCP_TLS_MODE_SW) {
MPASS(tp->t_nic_ktls_xmit == 0);
tp->t_nic_ktls_xmit = 1;
if (tp->t_fb->tfb_hwtls_change != NULL)
(*tp->t_fb->tfb_hwtls_change)(tp, 1);
}
SOCKBUF_UNLOCK(&so->so_snd);
SOCK_IO_SEND_UNLOCK(so);
/*
* Drop two references on 'tls'. The first is for the
* ktls_hold() above. The second drops the reference from the
* socket buffer.
*/
KASSERT(tls->refcount >= 2, ("too few references on old session"));
ktls_free(tls);
ktls_free(tls);
if (mode == TCP_TLS_MODE_IFNET)
counter_u64_add(ktls_switch_to_ifnet, 1);
else
counter_u64_add(ktls_switch_to_sw, 1);
return (0);
}
/*
* Try to allocate a new TLS receive tag. This task is scheduled when
* sbappend_ktls_rx detects an input path change. If a new tag is
* allocated, replace the tag in the TLS session. If a new tag cannot
* be allocated, let the session fall back to software decryption.
*/
static void
ktls_reset_receive_tag(void *context, int pending)
{
union if_snd_tag_alloc_params params;
struct ktls_session *tls;
struct m_snd_tag *mst;
struct inpcb *inp;
struct ifnet *ifp;
struct socket *so;
int error;
MPASS(pending == 1);
tls = context;
so = tls->so;
inp = so->so_pcb;
ifp = NULL;
INP_RLOCK(inp);
if (inp->inp_flags & INP_DROPPED) {
INP_RUNLOCK(inp);
goto out;
}
SOCKBUF_LOCK(&so->so_rcv);
mst = tls->snd_tag;
tls->snd_tag = NULL;
if (mst != NULL)
m_snd_tag_rele(mst);
ifp = tls->rx_ifp;
if_ref(ifp);
SOCKBUF_UNLOCK(&so->so_rcv);
params.hdr.type = IF_SND_TAG_TYPE_TLS_RX;
params.hdr.flowid = inp->inp_flowid;
params.hdr.flowtype = inp->inp_flowtype;
params.hdr.numa_domain = inp->inp_numa_domain;
params.tls_rx.inp = inp;
params.tls_rx.tls = tls;
params.tls_rx.vlan_id = tls->rx_vlan_id;
INP_RUNLOCK(inp);
if (inp->inp_vflag & INP_IPV6) {
if ((if_getcapenable2(ifp) & IFCAP2_RXTLS6) == 0)
goto out;
} else {
if ((if_getcapenable2(ifp) & IFCAP2_RXTLS4) == 0)
goto out;
}
error = m_snd_tag_alloc(ifp, ¶ms, &mst);
if (error == 0) {
SOCKBUF_LOCK(&so->so_rcv);
tls->snd_tag = mst;
SOCKBUF_UNLOCK(&so->so_rcv);
counter_u64_add(ktls_ifnet_reset, 1);
} else {
/*
* Just fall back to software decryption if a tag
* cannot be allocated leaving the connection intact.
* If a future input path change switches to another
* interface this connection will resume ifnet TLS.
*/
counter_u64_add(ktls_ifnet_reset_failed, 1);
}
out:
mtx_pool_lock(mtxpool_sleep, tls);
tls->reset_pending = false;
mtx_pool_unlock(mtxpool_sleep, tls);
if (ifp != NULL)
if_rele(ifp);
CURVNET_SET(so->so_vnet);
sorele(so);
CURVNET_RESTORE();
ktls_free(tls);
}
/*
* Try to allocate a new TLS send tag. This task is scheduled when
* ip_output detects a route change while trying to transmit a packet
* holding a TLS record. If a new tag is allocated, replace the tag
* in the TLS session. Subsequent packets on the connection will use
* the new tag. If a new tag cannot be allocated, drop the
* connection.
*/
static void
ktls_reset_send_tag(void *context, int pending)
{
struct epoch_tracker et;
struct ktls_session *tls;
struct m_snd_tag *old, *new;
struct inpcb *inp;
struct tcpcb *tp;
int error;
MPASS(pending == 1);
tls = context;
inp = tls->inp;
/*
* Free the old tag first before allocating a new one.
* ip[6]_output_send() will treat a NULL send tag the same as
* an ifp mismatch and drop packets until a new tag is
* allocated.
*
* Write-lock the INP when changing tls->snd_tag since
* ip[6]_output_send() holds a read-lock when reading the
* pointer.
*/
INP_WLOCK(inp);
old = tls->snd_tag;
tls->snd_tag = NULL;
INP_WUNLOCK(inp);
if (old != NULL)
m_snd_tag_rele(old);
error = ktls_alloc_snd_tag(inp, tls, true, &new);
if (error == 0) {
INP_WLOCK(inp);
tls->snd_tag = new;
mtx_pool_lock(mtxpool_sleep, tls);
tls->reset_pending = false;
mtx_pool_unlock(mtxpool_sleep, tls);
INP_WUNLOCK(inp);
counter_u64_add(ktls_ifnet_reset, 1);
/*
* XXX: Should we kick tcp_output explicitly now that
* the send tag is fixed or just rely on timers?
*/
} else {
NET_EPOCH_ENTER(et);
INP_WLOCK(inp);
if (!(inp->inp_flags & INP_DROPPED)) {
tp = intotcpcb(inp);
CURVNET_SET(inp->inp_vnet);
tp = tcp_drop(tp, ECONNABORTED);
CURVNET_RESTORE();
if (tp != NULL) {
counter_u64_add(ktls_ifnet_reset_dropped, 1);
INP_WUNLOCK(inp);
}
} else
INP_WUNLOCK(inp);
NET_EPOCH_EXIT(et);
counter_u64_add(ktls_ifnet_reset_failed, 1);
/*
* Leave reset_pending true to avoid future tasks while
* the socket goes away.
*/
}
ktls_free(tls);
}
void
ktls_input_ifp_mismatch(struct sockbuf *sb, struct ifnet *ifp)
{
struct ktls_session *tls;
struct socket *so;
SOCKBUF_LOCK_ASSERT(sb);
KASSERT(sb->sb_flags & SB_TLS_RX, ("%s: sockbuf %p isn't TLS RX",
__func__, sb));
so = __containerof(sb, struct socket, so_rcv);
tls = sb->sb_tls_info;
if_rele(tls->rx_ifp);
if_ref(ifp);
tls->rx_ifp = ifp;
/*
* See if we should schedule a task to update the receive tag for
* this session.
*/
mtx_pool_lock(mtxpool_sleep, tls);
if (!tls->reset_pending) {
(void) ktls_hold(tls);
soref(so);
tls->so = so;
tls->reset_pending = true;
taskqueue_enqueue(taskqueue_thread, &tls->reset_tag_task);
}
mtx_pool_unlock(mtxpool_sleep, tls);
}
int
ktls_output_eagain(struct inpcb *inp, struct ktls_session *tls)
{
if (inp == NULL)
return (ENOBUFS);
INP_LOCK_ASSERT(inp);
/*
* See if we should schedule a task to update the send tag for
* this session.
*/
mtx_pool_lock(mtxpool_sleep, tls);
if (!tls->reset_pending) {
(void) ktls_hold(tls);
tls->reset_pending = true;
taskqueue_enqueue(taskqueue_thread, &tls->reset_tag_task);
}
mtx_pool_unlock(mtxpool_sleep, tls);
return (ENOBUFS);
}
#ifdef RATELIMIT
int
ktls_modify_txrtlmt(struct ktls_session *tls, uint64_t max_pacing_rate)
{
union if_snd_tag_modify_params params = {
.rate_limit.max_rate = max_pacing_rate,
.rate_limit.flags = M_NOWAIT,
};
struct m_snd_tag *mst;
/* Can't get to the inp, but it should be locked. */
/* INP_LOCK_ASSERT(inp); */
MPASS(tls->mode == TCP_TLS_MODE_IFNET);
if (tls->snd_tag == NULL) {
/*
* Resetting send tag, ignore this change. The
* pending reset may or may not see this updated rate
* in the tcpcb. If it doesn't, we will just lose
* this rate change.
*/
return (0);
}
mst = tls->snd_tag;
MPASS(mst != NULL);
MPASS(mst->sw->type == IF_SND_TAG_TYPE_TLS_RATE_LIMIT);
return (mst->sw->snd_tag_modify(mst, ¶ms));
}
#endif
static void
ktls_destroy_help(void *context, int pending __unused)
{
ktls_destroy(context);
}
void
ktls_destroy(struct ktls_session *tls)
{
struct inpcb *inp;
struct tcpcb *tp;
bool wlocked;
MPASS(tls->refcount == 0);
inp = tls->inp;
if (tls->tx) {
wlocked = INP_WLOCKED(inp);
if (!wlocked && !INP_TRY_WLOCK(inp)) {
/*
* rwlocks read locks are anonymous, and there
* is no way to know if our current thread
* holds an rlock on the inp. As a rough
* estimate, check to see if the thread holds
* *any* rlocks at all. If it does not, then we
* know that we don't hold the inp rlock, and
* can safely take the wlock
*/
if (curthread->td_rw_rlocks == 0) {
INP_WLOCK(inp);
} else {
/*
* We might hold the rlock, so let's
* do the destroy in a taskqueue
* context to avoid a potential
* deadlock. This should be very
* rare.
*/
counter_u64_add(ktls_destroy_task, 1);
TASK_INIT(&tls->destroy_task, 0,
ktls_destroy_help, tls);
(void)taskqueue_enqueue(taskqueue_thread,
&tls->destroy_task);
return;
}
}
}
if (tls->sequential_records) {
struct mbuf *m, *n;
int page_count;
STAILQ_FOREACH_SAFE(m, &tls->pending_records, m_epg_stailq, n) {
page_count = m->m_epg_enc_cnt;
while (page_count > 0) {
KASSERT(page_count >= m->m_epg_nrdy,
("%s: too few pages", __func__));
page_count -= m->m_epg_nrdy;
m = m_free(m);
}
}
}
counter_u64_add(ktls_offload_active, -1);
switch (tls->mode) {
case TCP_TLS_MODE_SW:
switch (tls->params.cipher_algorithm) {
case CRYPTO_AES_CBC:
counter_u64_add(ktls_sw_cbc, -1);
break;
case CRYPTO_AES_NIST_GCM_16:
counter_u64_add(ktls_sw_gcm, -1);
break;
case CRYPTO_CHACHA20_POLY1305:
counter_u64_add(ktls_sw_chacha20, -1);
break;
}
break;
case TCP_TLS_MODE_IFNET:
switch (tls->params.cipher_algorithm) {
case CRYPTO_AES_CBC:
counter_u64_add(ktls_ifnet_cbc, -1);
break;
case CRYPTO_AES_NIST_GCM_16:
counter_u64_add(ktls_ifnet_gcm, -1);
break;
case CRYPTO_CHACHA20_POLY1305:
counter_u64_add(ktls_ifnet_chacha20, -1);
break;
}
if (tls->snd_tag != NULL)
m_snd_tag_rele(tls->snd_tag);
if (tls->rx_ifp != NULL)
if_rele(tls->rx_ifp);
if (tls->tx) {
INP_WLOCK_ASSERT(inp);
tp = intotcpcb(inp);
MPASS(tp->t_nic_ktls_xmit == 1);
tp->t_nic_ktls_xmit = 0;
}
break;
#ifdef TCP_OFFLOAD
case TCP_TLS_MODE_TOE:
switch (tls->params.cipher_algorithm) {
case CRYPTO_AES_CBC:
counter_u64_add(ktls_toe_cbc, -1);
break;
case CRYPTO_AES_NIST_GCM_16:
counter_u64_add(ktls_toe_gcm, -1);
break;
case CRYPTO_CHACHA20_POLY1305:
counter_u64_add(ktls_toe_chacha20, -1);
break;
}
break;
#endif
}
if (tls->ocf_session != NULL)
ktls_ocf_free(tls);
if (tls->params.auth_key != NULL) {
zfree(tls->params.auth_key, M_KTLS);
tls->params.auth_key = NULL;
tls->params.auth_key_len = 0;
}
if (tls->params.cipher_key != NULL) {
zfree(tls->params.cipher_key, M_KTLS);
tls->params.cipher_key = NULL;
tls->params.cipher_key_len = 0;
}
if (tls->tx) {
INP_WLOCK_ASSERT(inp);
if (!in_pcbrele_wlocked(inp) && !wlocked)
INP_WUNLOCK(inp);
}
explicit_bzero(tls->params.iv, sizeof(tls->params.iv));
uma_zfree(ktls_session_zone, tls);
}
void
ktls_seq(struct sockbuf *sb, struct mbuf *m)
{
for (; m != NULL; m = m->m_next) {
KASSERT((m->m_flags & M_EXTPG) != 0,
("ktls_seq: mapped mbuf %p", m));
m->m_epg_seqno = sb->sb_tls_seqno;
sb->sb_tls_seqno++;
}
}
/*
* Add TLS framing (headers and trailers) to a chain of mbufs. Each
* mbuf in the chain must be an unmapped mbuf. The payload of the
* mbuf must be populated with the payload of each TLS record.
*
* The record_type argument specifies the TLS record type used when
* populating the TLS header.
*
* The enq_count argument on return is set to the number of pages of
* payload data for this entire chain that need to be encrypted via SW
* encryption. The returned value should be passed to ktls_enqueue
* when scheduling encryption of this chain of mbufs. To handle the
* special case of empty fragments for TLS 1.0 sessions, an empty
* fragment counts as one page.
*/
void
ktls_frame(struct mbuf *top, struct ktls_session *tls, int *enq_cnt,
uint8_t record_type)
{
struct tls_record_layer *tlshdr;
struct mbuf *m;
uint64_t *noncep;
uint16_t tls_len;
int maxlen __diagused;
maxlen = tls->params.max_frame_len;
*enq_cnt = 0;
for (m = top; m != NULL; m = m->m_next) {
/*
* All mbufs in the chain should be TLS records whose
* payload does not exceed the maximum frame length.
*
* Empty TLS 1.0 records are permitted when using CBC.
*/
KASSERT(m->m_len <= maxlen && m->m_len >= 0 &&
(m->m_len > 0 || ktls_permit_empty_frames(tls)),
("ktls_frame: m %p len %d", m, m->m_len));
/*
* TLS frames require unmapped mbufs to store session
* info.
*/
KASSERT((m->m_flags & M_EXTPG) != 0,
("ktls_frame: mapped mbuf %p (top = %p)", m, top));
tls_len = m->m_len;
/* Save a reference to the session. */
m->m_epg_tls = ktls_hold(tls);
m->m_epg_hdrlen = tls->params.tls_hlen;
m->m_epg_trllen = tls->params.tls_tlen;
if (tls->params.cipher_algorithm == CRYPTO_AES_CBC) {
int bs, delta;
/*
* AES-CBC pads messages to a multiple of the
* block size. Note that the padding is
* applied after the digest and the encryption
* is done on the "plaintext || mac || padding".
* At least one byte of padding is always
* present.
*
* Compute the final trailer length assuming
* at most one block of padding.
* tls->params.tls_tlen is the maximum
* possible trailer length (padding + digest).
* delta holds the number of excess padding
* bytes if the maximum were used. Those
* extra bytes are removed.
*/
bs = tls->params.tls_bs;
delta = (tls_len + tls->params.tls_tlen) & (bs - 1);
m->m_epg_trllen -= delta;
}
m->m_len += m->m_epg_hdrlen + m->m_epg_trllen;
/* Populate the TLS header. */
tlshdr = (void *)m->m_epg_hdr;
tlshdr->tls_vmajor = tls->params.tls_vmajor;
/*
* TLS 1.3 masquarades as TLS 1.2 with a record type
* of TLS_RLTYPE_APP.
*/
if (tls->params.tls_vminor == TLS_MINOR_VER_THREE &&
tls->params.tls_vmajor == TLS_MAJOR_VER_ONE) {
tlshdr->tls_vminor = TLS_MINOR_VER_TWO;
tlshdr->tls_type = TLS_RLTYPE_APP;
/* save the real record type for later */
m->m_epg_record_type = record_type;
m->m_epg_trail[0] = record_type;
} else {
tlshdr->tls_vminor = tls->params.tls_vminor;
tlshdr->tls_type = record_type;
}
tlshdr->tls_length = htons(m->m_len - sizeof(*tlshdr));
/*
* Store nonces / explicit IVs after the end of the
* TLS header.
*
* For GCM with TLS 1.2, an 8 byte nonce is copied
* from the end of the IV. The nonce is then
* incremented for use by the next record.
*
* For CBC, a random nonce is inserted for TLS 1.1+.
*/
if (tls->params.cipher_algorithm == CRYPTO_AES_NIST_GCM_16 &&
tls->params.tls_vminor == TLS_MINOR_VER_TWO) {
noncep = (uint64_t *)(tls->params.iv + 8);
be64enc(tlshdr + 1, *noncep);
(*noncep)++;
} else if (tls->params.cipher_algorithm == CRYPTO_AES_CBC &&
tls->params.tls_vminor >= TLS_MINOR_VER_ONE)
arc4rand(tlshdr + 1, AES_BLOCK_LEN, 0);
/*
* When using SW encryption, mark the mbuf not ready.
* It will be marked ready via sbready() after the
* record has been encrypted.
*
* When using ifnet TLS, unencrypted TLS records are
* sent down the stack to the NIC.
*/
if (tls->mode == TCP_TLS_MODE_SW) {
m->m_flags |= M_NOTREADY;
if (__predict_false(tls_len == 0)) {
/* TLS 1.0 empty fragment. */
m->m_epg_nrdy = 1;
} else
m->m_epg_nrdy = m->m_epg_npgs;
*enq_cnt += m->m_epg_nrdy;
}
}
}
bool
ktls_permit_empty_frames(struct ktls_session *tls)
{
return (tls->params.cipher_algorithm == CRYPTO_AES_CBC &&
tls->params.tls_vminor == TLS_MINOR_VER_ZERO);
}
void
ktls_check_rx(struct sockbuf *sb)
{
struct tls_record_layer hdr;
struct ktls_wq *wq;
struct socket *so;
bool running;
SOCKBUF_LOCK_ASSERT(sb);
KASSERT(sb->sb_flags & SB_TLS_RX, ("%s: sockbuf %p isn't TLS RX",
__func__, sb));
so = __containerof(sb, struct socket, so_rcv);
if (sb->sb_flags & SB_TLS_RX_RUNNING)
return;
/* Is there enough queued for a TLS header? */
if (sb->sb_tlscc < sizeof(hdr)) {
if ((sb->sb_state & SBS_CANTRCVMORE) != 0 && sb->sb_tlscc != 0)
so->so_error = EMSGSIZE;
return;
}
m_copydata(sb->sb_mtls, 0, sizeof(hdr), (void *)&hdr);
/* Is the entire record queued? */
if (sb->sb_tlscc < sizeof(hdr) + ntohs(hdr.tls_length)) {
if ((sb->sb_state & SBS_CANTRCVMORE) != 0)
so->so_error = EMSGSIZE;
return;
}
sb->sb_flags |= SB_TLS_RX_RUNNING;
soref(so);
wq = &ktls_wq[so->so_rcv.sb_tls_info->wq_index];
mtx_lock(&wq->mtx);
STAILQ_INSERT_TAIL(&wq->so_head, so, so_ktls_rx_list);
running = wq->running;
mtx_unlock(&wq->mtx);
if (!running)
wakeup(wq);
counter_u64_add(ktls_cnt_rx_queued, 1);
}
static struct mbuf *
ktls_detach_record(struct sockbuf *sb, int len)
{
struct mbuf *m, *n, *top;
int remain;
SOCKBUF_LOCK_ASSERT(sb);
MPASS(len <= sb->sb_tlscc);
/*
* If TLS chain is the exact size of the record,
* just grab the whole record.
*/
top = sb->sb_mtls;
if (sb->sb_tlscc == len) {
sb->sb_mtls = NULL;
sb->sb_mtlstail = NULL;
goto out;
}
/*
* While it would be nice to use m_split() here, we need
* to know exactly what m_split() allocates to update the
* accounting, so do it inline instead.
*/
remain = len;
for (m = top; remain > m->m_len; m = m->m_next)
remain -= m->m_len;
/* Easy case: don't have to split 'm'. */
if (remain == m->m_len) {
sb->sb_mtls = m->m_next;
if (sb->sb_mtls == NULL)
sb->sb_mtlstail = NULL;
m->m_next = NULL;
goto out;
}
/*
* Need to allocate an mbuf to hold the remainder of 'm'. Try
* with M_NOWAIT first.
*/
n = m_get(M_NOWAIT, MT_DATA);
if (n == NULL) {
/*
* Use M_WAITOK with socket buffer unlocked. If
* 'sb_mtls' changes while the lock is dropped, return
* NULL to force the caller to retry.
*/
SOCKBUF_UNLOCK(sb);
n = m_get(M_WAITOK, MT_DATA);
SOCKBUF_LOCK(sb);
if (sb->sb_mtls != top) {
m_free(n);
return (NULL);
}
}
n->m_flags |= (m->m_flags & (M_NOTREADY | M_DECRYPTED));
/* Store remainder in 'n'. */
n->m_len = m->m_len - remain;
if (m->m_flags & M_EXT) {
n->m_data = m->m_data + remain;
mb_dupcl(n, m);
} else {
bcopy(mtod(m, caddr_t) + remain, mtod(n, caddr_t), n->m_len);
}
/* Trim 'm' and update accounting. */
m->m_len -= n->m_len;
sb->sb_tlscc -= n->m_len;
sb->sb_ccc -= n->m_len;
/* Account for 'n'. */
sballoc_ktls_rx(sb, n);
/* Insert 'n' into the TLS chain. */
sb->sb_mtls = n;
n->m_next = m->m_next;
if (sb->sb_mtlstail == m)
sb->sb_mtlstail = n;
/* Detach the record from the TLS chain. */
m->m_next = NULL;
out:
MPASS(m_length(top, NULL) == len);
for (m = top; m != NULL; m = m->m_next)
sbfree_ktls_rx(sb, m);
sb->sb_tlsdcc = len;
sb->sb_ccc += len;
SBCHECK(sb);
return (top);
}
/*
* Determine the length of the trailing zero padding and find the real
* record type in the byte before the padding.
*
* Walking the mbuf chain backwards is clumsy, so another option would
* be to scan forwards remembering the last non-zero byte before the
* trailer. However, it would be expensive to scan the entire record.
* Instead, find the last non-zero byte of each mbuf in the chain
* keeping track of the relative offset of that nonzero byte.
*
* trail_len is the size of the MAC/tag on input and is set to the
* size of the full trailer including padding and the record type on
* return.
*/
static int
tls13_find_record_type(struct ktls_session *tls, struct mbuf *m, int tls_len,
int *trailer_len, uint8_t *record_typep)
{
char *cp;
u_int digest_start, last_offset, m_len, offset;
uint8_t record_type;
digest_start = tls_len - *trailer_len;
last_offset = 0;
offset = 0;
for (; m != NULL && offset < digest_start;
offset += m->m_len, m = m->m_next) {
/* Don't look for padding in the tag. */
m_len = min(digest_start - offset, m->m_len);
cp = mtod(m, char *);
/* Find last non-zero byte in this mbuf. */
while (m_len > 0 && cp[m_len - 1] == 0)
m_len--;
if (m_len > 0) {
record_type = cp[m_len - 1];
last_offset = offset + m_len;
}
}
if (last_offset < tls->params.tls_hlen)
return (EBADMSG);
*record_typep = record_type;
*trailer_len = tls_len - last_offset + 1;
return (0);
}
/*
* Check if a mbuf chain is fully decrypted at the given offset and
* length. Returns KTLS_MBUF_CRYPTO_ST_DECRYPTED if all data is
* decrypted. KTLS_MBUF_CRYPTO_ST_MIXED if there is a mix of encrypted
* and decrypted data. Else KTLS_MBUF_CRYPTO_ST_ENCRYPTED if all data
* is encrypted.
*/
ktls_mbuf_crypto_st_t
ktls_mbuf_crypto_state(struct mbuf *mb, int offset, int len)
{
int m_flags_ored = 0;
int m_flags_anded = -1;
for (; mb != NULL; mb = mb->m_next) {
if (offset < mb->m_len)
break;
offset -= mb->m_len;
}
offset += len;
for (; mb != NULL; mb = mb->m_next) {
m_flags_ored |= mb->m_flags;
m_flags_anded &= mb->m_flags;
if (offset <= mb->m_len)
break;
offset -= mb->m_len;
}
MPASS(mb != NULL || offset == 0);
if ((m_flags_ored ^ m_flags_anded) & M_DECRYPTED)
return (KTLS_MBUF_CRYPTO_ST_MIXED);
else
return ((m_flags_ored & M_DECRYPTED) ?
KTLS_MBUF_CRYPTO_ST_DECRYPTED :
KTLS_MBUF_CRYPTO_ST_ENCRYPTED);
}
/*
* ktls_resync_ifnet - get HW TLS RX back on track after packet loss
*/
static int
ktls_resync_ifnet(struct socket *so, uint32_t tls_len, uint64_t tls_rcd_num)
{
union if_snd_tag_modify_params params;
struct m_snd_tag *mst;
struct inpcb *inp;
struct tcpcb *tp;
mst = so->so_rcv.sb_tls_info->snd_tag;
if (__predict_false(mst == NULL))
return (EINVAL);
inp = sotoinpcb(so);
if (__predict_false(inp == NULL))
return (EINVAL);
INP_RLOCK(inp);
if (inp->inp_flags & INP_DROPPED) {
INP_RUNLOCK(inp);
return (ECONNRESET);
}
tp = intotcpcb(inp);
MPASS(tp != NULL);
/* Get the TCP sequence number of the next valid TLS header. */
SOCKBUF_LOCK(&so->so_rcv);
params.tls_rx.tls_hdr_tcp_sn =
tp->rcv_nxt - so->so_rcv.sb_tlscc - tls_len;
params.tls_rx.tls_rec_length = tls_len;
params.tls_rx.tls_seq_number = tls_rcd_num;
SOCKBUF_UNLOCK(&so->so_rcv);
INP_RUNLOCK(inp);
MPASS(mst->sw->type == IF_SND_TAG_TYPE_TLS_RX);
return (mst->sw->snd_tag_modify(mst, ¶ms));
}
static void
ktls_drop(struct socket *so, int error)
{
struct epoch_tracker et;
struct inpcb *inp = sotoinpcb(so);
struct tcpcb *tp;
NET_EPOCH_ENTER(et);
INP_WLOCK(inp);
if (!(inp->inp_flags & INP_DROPPED)) {
tp = intotcpcb(inp);
CURVNET_SET(inp->inp_vnet);
tp = tcp_drop(tp, error);
CURVNET_RESTORE();
if (tp != NULL)
INP_WUNLOCK(inp);
} else {
so->so_error = error;
SOCK_RECVBUF_LOCK(so);
sorwakeup_locked(so);
INP_WUNLOCK(inp);
}
NET_EPOCH_EXIT(et);
}
static void
ktls_decrypt(struct socket *so)
{
char tls_header[MBUF_PEXT_HDR_LEN];
struct ktls_session *tls;
struct sockbuf *sb;
struct tls_record_layer *hdr;
struct tls_get_record tgr;
struct mbuf *control, *data, *m;
ktls_mbuf_crypto_st_t state;
uint64_t seqno;
int error, remain, tls_len, trail_len;
bool tls13;
uint8_t vminor, record_type;
hdr = (struct tls_record_layer *)tls_header;
sb = &so->so_rcv;
SOCKBUF_LOCK(sb);
KASSERT(sb->sb_flags & SB_TLS_RX_RUNNING,
("%s: socket %p not running", __func__, so));
tls = sb->sb_tls_info;
MPASS(tls != NULL);
tls13 = (tls->params.tls_vminor == TLS_MINOR_VER_THREE);
if (tls13)
vminor = TLS_MINOR_VER_TWO;
else
vminor = tls->params.tls_vminor;
for (;;) {
/* Is there enough queued for a TLS header? */
if (sb->sb_tlscc < tls->params.tls_hlen)
break;
m_copydata(sb->sb_mtls, 0, tls->params.tls_hlen, tls_header);
tls_len = sizeof(*hdr) + ntohs(hdr->tls_length);
if (hdr->tls_vmajor != tls->params.tls_vmajor ||
hdr->tls_vminor != vminor)
error = EINVAL;
else if (tls13 && hdr->tls_type != TLS_RLTYPE_APP)
error = EINVAL;
else if (tls_len < tls->params.tls_hlen || tls_len >
tls->params.tls_hlen + TLS_MAX_MSG_SIZE_V10_2 +
tls->params.tls_tlen)
error = EMSGSIZE;
else
error = 0;
if (__predict_false(error != 0)) {
/*
* We have a corrupted record and are likely
* out of sync. The connection isn't
* recoverable at this point, so abort it.
*/
SOCKBUF_UNLOCK(sb);
counter_u64_add(ktls_offload_corrupted_records, 1);
ktls_drop(so, error);
goto deref;
}
/* Is the entire record queued? */
if (sb->sb_tlscc < tls_len)
break;
/*
* Split out the portion of the mbuf chain containing
* this TLS record.
*/
data = ktls_detach_record(sb, tls_len);
if (data == NULL)
continue;
MPASS(sb->sb_tlsdcc == tls_len);
seqno = sb->sb_tls_seqno;
sb->sb_tls_seqno++;
SBCHECK(sb);
SOCKBUF_UNLOCK(sb);
/* get crypto state for this TLS record */
state = ktls_mbuf_crypto_state(data, 0, tls_len);
switch (state) {
case KTLS_MBUF_CRYPTO_ST_MIXED:
error = ktls_ocf_recrypt(tls, hdr, data, seqno);
if (error)
break;
/* FALLTHROUGH */
case KTLS_MBUF_CRYPTO_ST_ENCRYPTED:
error = ktls_ocf_decrypt(tls, hdr, data, seqno,
&trail_len);
if (__predict_true(error == 0)) {
if (tls13) {
error = tls13_find_record_type(tls, data,
tls_len, &trail_len, &record_type);
} else {
record_type = hdr->tls_type;
}
}
break;
case KTLS_MBUF_CRYPTO_ST_DECRYPTED:
/*
* NIC TLS is only supported for AEAD
* ciphersuites which used a fixed sized
* trailer.
*/
if (tls13) {
trail_len = tls->params.tls_tlen - 1;
error = tls13_find_record_type(tls, data,
tls_len, &trail_len, &record_type);
} else {
trail_len = tls->params.tls_tlen;
error = 0;
record_type = hdr->tls_type;
}
break;
default:
error = EINVAL;
break;
}
if (error) {
counter_u64_add(ktls_offload_failed_crypto, 1);
SOCKBUF_LOCK(sb);
if (sb->sb_tlsdcc == 0) {
/*
* sbcut/drop/flush discarded these
* mbufs.
*/
m_freem(data);
break;
}
/*
* Drop this TLS record's data, but keep
* decrypting subsequent records.
*/
sb->sb_ccc -= tls_len;
sb->sb_tlsdcc = 0;
if (error != EMSGSIZE)
error = EBADMSG;
CURVNET_SET(so->so_vnet);
so->so_error = error;
sorwakeup_locked(so);
CURVNET_RESTORE();
m_freem(data);
SOCKBUF_LOCK(sb);
continue;
}
/* Allocate the control mbuf. */
memset(&tgr, 0, sizeof(tgr));
tgr.tls_type = record_type;
tgr.tls_vmajor = hdr->tls_vmajor;
tgr.tls_vminor = hdr->tls_vminor;
tgr.tls_length = htobe16(tls_len - tls->params.tls_hlen -
trail_len);
control = sbcreatecontrol(&tgr, sizeof(tgr),
TLS_GET_RECORD, IPPROTO_TCP, M_WAITOK);
SOCKBUF_LOCK(sb);
if (sb->sb_tlsdcc == 0) {
/* sbcut/drop/flush discarded these mbufs. */
MPASS(sb->sb_tlscc == 0);
m_freem(data);
m_freem(control);
break;
}
/*
* Clear the 'dcc' accounting in preparation for
* adding the decrypted record.
*/
sb->sb_ccc -= tls_len;
sb->sb_tlsdcc = 0;
SBCHECK(sb);
/* If there is no payload, drop all of the data. */
if (tgr.tls_length == htobe16(0)) {
m_freem(data);
data = NULL;
} else {
/* Trim header. */
remain = tls->params.tls_hlen;
while (remain > 0) {
if (data->m_len > remain) {
data->m_data += remain;
data->m_len -= remain;
break;
}
remain -= data->m_len;
data = m_free(data);
}
/* Trim trailer and clear M_NOTREADY. */
remain = be16toh(tgr.tls_length);
m = data;
for (m = data; remain > m->m_len; m = m->m_next) {
m->m_flags &= ~(M_NOTREADY | M_DECRYPTED);
remain -= m->m_len;
}
m->m_len = remain;
m_freem(m->m_next);
m->m_next = NULL;
m->m_flags &= ~(M_NOTREADY | M_DECRYPTED);
/* Set EOR on the final mbuf. */
m->m_flags |= M_EOR;
}
sbappendcontrol_locked(sb, data, control, 0);
if (__predict_false(state != KTLS_MBUF_CRYPTO_ST_DECRYPTED)) {
sb->sb_flags |= SB_TLS_RX_RESYNC;
SOCKBUF_UNLOCK(sb);
ktls_resync_ifnet(so, tls_len, seqno);
SOCKBUF_LOCK(sb);
} else if (__predict_false(sb->sb_flags & SB_TLS_RX_RESYNC)) {
sb->sb_flags &= ~SB_TLS_RX_RESYNC;
SOCKBUF_UNLOCK(sb);
ktls_resync_ifnet(so, 0, seqno);
SOCKBUF_LOCK(sb);
}
}
sb->sb_flags &= ~SB_TLS_RX_RUNNING;
if ((sb->sb_state & SBS_CANTRCVMORE) != 0 && sb->sb_tlscc > 0)
so->so_error = EMSGSIZE;
sorwakeup_locked(so);
deref:
SOCKBUF_UNLOCK_ASSERT(sb);
CURVNET_SET(so->so_vnet);
sorele(so);
CURVNET_RESTORE();
}
void
ktls_enqueue_to_free(struct mbuf *m)
{
struct ktls_wq *wq;
bool running;
/* Mark it for freeing. */
m->m_epg_flags |= EPG_FLAG_2FREE;
wq = &ktls_wq[m->m_epg_tls->wq_index];
mtx_lock(&wq->mtx);
STAILQ_INSERT_TAIL(&wq->m_head, m, m_epg_stailq);
running = wq->running;
mtx_unlock(&wq->mtx);
if (!running)
wakeup(wq);
}
static void *
ktls_buffer_alloc(struct ktls_wq *wq, struct mbuf *m)
{
void *buf;
int domain, running;
if (m->m_epg_npgs <= 2)
return (NULL);
if (ktls_buffer_zone == NULL)
return (NULL);
if ((u_int)(ticks - wq->lastallocfail) < hz) {
/*
* Rate-limit allocation attempts after a failure.
* ktls_buffer_import() will acquire a per-domain mutex to check
* the free page queues and may fail consistently if memory is
* fragmented.
*/
return (NULL);
}
buf = uma_zalloc(ktls_buffer_zone, M_NOWAIT | M_NORECLAIM);
if (buf == NULL) {
domain = PCPU_GET(domain);
wq->lastallocfail = ticks;
/*
* Note that this check is "racy", but the races are
* harmless, and are either a spurious wakeup if
* multiple threads fail allocations before the alloc
* thread wakes, or waiting an extra second in case we
* see an old value of running == true.
*/
if (!VM_DOMAIN_EMPTY(domain)) {
running = atomic_load_int(&ktls_domains[domain].reclaim_td.running);
if (!running)
wakeup(&ktls_domains[domain].reclaim_td);
}
}
return (buf);
}
static int
ktls_encrypt_record(struct ktls_wq *wq, struct mbuf *m,
struct ktls_session *tls, struct ktls_ocf_encrypt_state *state)
{
vm_page_t pg;
int error, i, len, off;
KASSERT((m->m_flags & (M_EXTPG | M_NOTREADY)) == (M_EXTPG | M_NOTREADY),
("%p not unready & nomap mbuf\n", m));
KASSERT(ptoa(m->m_epg_npgs) <= ktls_maxlen,
("page count %d larger than maximum frame length %d", m->m_epg_npgs,
ktls_maxlen));
/* Anonymous mbufs are encrypted in place. */
if ((m->m_epg_flags & EPG_FLAG_ANON) != 0)
return (ktls_ocf_encrypt(state, tls, m, NULL, 0));
/*
* For file-backed mbufs (from sendfile), anonymous wired
* pages are allocated and used as the encryption destination.
*/
if ((state->cbuf = ktls_buffer_alloc(wq, m)) != NULL) {
len = ptoa(m->m_epg_npgs - 1) + m->m_epg_last_len -
m->m_epg_1st_off;
state->dst_iov[0].iov_base = (char *)state->cbuf +
m->m_epg_1st_off;
state->dst_iov[0].iov_len = len;
state->parray[0] = DMAP_TO_PHYS((vm_offset_t)state->cbuf);
i = 1;
} else {
off = m->m_epg_1st_off;
for (i = 0; i < m->m_epg_npgs; i++, off = 0) {
pg = vm_page_alloc_noobj(VM_ALLOC_NODUMP |
VM_ALLOC_WIRED | VM_ALLOC_WAITOK);
len = m_epg_pagelen(m, i, off);
state->parray[i] = VM_PAGE_TO_PHYS(pg);
state->dst_iov[i].iov_base =
(char *)PHYS_TO_DMAP(state->parray[i]) + off;
state->dst_iov[i].iov_len = len;
}
}
KASSERT(i + 1 <= nitems(state->dst_iov), ("dst_iov is too small"));
state->dst_iov[i].iov_base = m->m_epg_trail;
state->dst_iov[i].iov_len = m->m_epg_trllen;
error = ktls_ocf_encrypt(state, tls, m, state->dst_iov, i + 1);
if (__predict_false(error != 0)) {
/* Free the anonymous pages. */
if (state->cbuf != NULL)
uma_zfree(ktls_buffer_zone, state->cbuf);
else {
for (i = 0; i < m->m_epg_npgs; i++) {
pg = PHYS_TO_VM_PAGE(state->parray[i]);
(void)vm_page_unwire_noq(pg);
vm_page_free(pg);
}
}
}
return (error);
}
/* Number of TLS records in a batch passed to ktls_enqueue(). */
static u_int
ktls_batched_records(struct mbuf *m)
{
int page_count, records;
records = 0;
page_count = m->m_epg_enc_cnt;
while (page_count > 0) {
records++;
page_count -= m->m_epg_nrdy;
m = m->m_next;
}
KASSERT(page_count == 0, ("%s: mismatched page count", __func__));
return (records);
}
void
ktls_enqueue(struct mbuf *m, struct socket *so, int page_count)
{
struct ktls_session *tls;
struct ktls_wq *wq;
int queued;
bool running;
KASSERT(((m->m_flags & (M_EXTPG | M_NOTREADY)) ==
(M_EXTPG | M_NOTREADY)),
("ktls_enqueue: %p not unready & nomap mbuf\n", m));
KASSERT(page_count != 0, ("enqueueing TLS mbuf with zero page count"));
KASSERT(m->m_epg_tls->mode == TCP_TLS_MODE_SW, ("!SW TLS mbuf"));
m->m_epg_enc_cnt = page_count;
/*
* Save a pointer to the socket. The caller is responsible
* for taking an additional reference via soref().
*/
m->m_epg_so = so;
queued = 1;
tls = m->m_epg_tls;
wq = &ktls_wq[tls->wq_index];
mtx_lock(&wq->mtx);
if (__predict_false(tls->sequential_records)) {
/*
* For TLS 1.0, records must be encrypted
* sequentially. For a given connection, all records
* queued to the associated work queue are processed
* sequentially. However, sendfile(2) might complete
* I/O requests spanning multiple TLS records out of
* order. Here we ensure TLS records are enqueued to
* the work queue in FIFO order.
*
* tls->next_seqno holds the sequence number of the
* next TLS record that should be enqueued to the work
* queue. If this next record is not tls->next_seqno,
* it must be a future record, so insert it, sorted by
* TLS sequence number, into tls->pending_records and
* return.
*
* If this TLS record matches tls->next_seqno, place
* it in the work queue and then check
* tls->pending_records to see if any
* previously-queued records are now ready for
* encryption.
*/
if (m->m_epg_seqno != tls->next_seqno) {
struct mbuf *n, *p;
p = NULL;
STAILQ_FOREACH(n, &tls->pending_records, m_epg_stailq) {
if (n->m_epg_seqno > m->m_epg_seqno)
break;
p = n;
}
if (n == NULL)
STAILQ_INSERT_TAIL(&tls->pending_records, m,
m_epg_stailq);
else if (p == NULL)
STAILQ_INSERT_HEAD(&tls->pending_records, m,
m_epg_stailq);
else
STAILQ_INSERT_AFTER(&tls->pending_records, p, m,
m_epg_stailq);
mtx_unlock(&wq->mtx);
counter_u64_add(ktls_cnt_tx_pending, 1);
return;
}
tls->next_seqno += ktls_batched_records(m);
STAILQ_INSERT_TAIL(&wq->m_head, m, m_epg_stailq);
while (!STAILQ_EMPTY(&tls->pending_records)) {
struct mbuf *n;
n = STAILQ_FIRST(&tls->pending_records);
if (n->m_epg_seqno != tls->next_seqno)
break;
queued++;
STAILQ_REMOVE_HEAD(&tls->pending_records, m_epg_stailq);
tls->next_seqno += ktls_batched_records(n);
STAILQ_INSERT_TAIL(&wq->m_head, n, m_epg_stailq);
}
counter_u64_add(ktls_cnt_tx_pending, -(queued - 1));
} else
STAILQ_INSERT_TAIL(&wq->m_head, m, m_epg_stailq);
running = wq->running;
mtx_unlock(&wq->mtx);
if (!running)
wakeup(wq);
counter_u64_add(ktls_cnt_tx_queued, queued);
}
/*
* Once a file-backed mbuf (from sendfile) has been encrypted, free
* the pages from the file and replace them with the anonymous pages
* allocated in ktls_encrypt_record().
*/
static void
ktls_finish_nonanon(struct mbuf *m, struct ktls_ocf_encrypt_state *state)
{
int i;
MPASS((m->m_epg_flags & EPG_FLAG_ANON) == 0);
/* Free the old pages. */
m->m_ext.ext_free(m);
/* Replace them with the new pages. */
if (state->cbuf != NULL) {
for (i = 0; i < m->m_epg_npgs; i++)
m->m_epg_pa[i] = state->parray[0] + ptoa(i);
/* Contig pages should go back to the cache. */
m->m_ext.ext_free = ktls_free_mext_contig;
} else {
for (i = 0; i < m->m_epg_npgs; i++)
m->m_epg_pa[i] = state->parray[i];
/* Use the basic free routine. */
m->m_ext.ext_free = mb_free_mext_pgs;
}
/* Pages are now writable. */
m->m_epg_flags |= EPG_FLAG_ANON;
}
static __noinline void
ktls_encrypt(struct ktls_wq *wq, struct mbuf *top)
{
struct ktls_ocf_encrypt_state state;
struct ktls_session *tls;
struct socket *so;
struct mbuf *m;
int error, npages, total_pages;
so = top->m_epg_so;
tls = top->m_epg_tls;
KASSERT(tls != NULL, ("tls = NULL, top = %p\n", top));
KASSERT(so != NULL, ("so = NULL, top = %p\n", top));
#ifdef INVARIANTS
top->m_epg_so = NULL;
#endif
total_pages = top->m_epg_enc_cnt;
npages = 0;
/*
* Encrypt the TLS records in the chain of mbufs starting with
* 'top'. 'total_pages' gives us a total count of pages and is
* used to know when we have finished encrypting the TLS
* records originally queued with 'top'.
*
* NB: These mbufs are queued in the socket buffer and
* 'm_next' is traversing the mbufs in the socket buffer. The
* socket buffer lock is not held while traversing this chain.
* Since the mbufs are all marked M_NOTREADY their 'm_next'
* pointers should be stable. However, the 'm_next' of the
* last mbuf encrypted is not necessarily NULL. It can point
* to other mbufs appended while 'top' was on the TLS work
* queue.
*
* Each mbuf holds an entire TLS record.
*/
error = 0;
for (m = top; npages != total_pages; m = m->m_next) {
KASSERT(m->m_epg_tls == tls,
("different TLS sessions in a single mbuf chain: %p vs %p",
tls, m->m_epg_tls));
KASSERT(npages + m->m_epg_npgs <= total_pages,
("page count mismatch: top %p, total_pages %d, m %p", top,
total_pages, m));
error = ktls_encrypt_record(wq, m, tls, &state);
if (error) {
counter_u64_add(ktls_offload_failed_crypto, 1);
break;
}
if ((m->m_epg_flags & EPG_FLAG_ANON) == 0)
ktls_finish_nonanon(m, &state);
npages += m->m_epg_nrdy;
/*
* Drop a reference to the session now that it is no
* longer needed. Existing code depends on encrypted
* records having no associated session vs
* yet-to-be-encrypted records having an associated
* session.
*/
m->m_epg_tls = NULL;
ktls_free(tls);
}
CURVNET_SET(so->so_vnet);
if (error == 0) {
(void)so->so_proto->pr_ready(so, top, npages);
} else {
ktls_drop(so, EIO);
mb_free_notready(top, total_pages);
}
sorele(so);
CURVNET_RESTORE();
}
void
ktls_encrypt_cb(struct ktls_ocf_encrypt_state *state, int error)
{
struct ktls_session *tls;
struct socket *so;
struct mbuf *m;
int npages;
m = state->m;
if ((m->m_epg_flags & EPG_FLAG_ANON) == 0)
ktls_finish_nonanon(m, state);
so = state->so;
free(state, M_KTLS);
/*
* Drop a reference to the session now that it is no longer
* needed. Existing code depends on encrypted records having
* no associated session vs yet-to-be-encrypted records having
* an associated session.
*/
tls = m->m_epg_tls;
m->m_epg_tls = NULL;
ktls_free(tls);
if (error != 0)
counter_u64_add(ktls_offload_failed_crypto, 1);
CURVNET_SET(so->so_vnet);
npages = m->m_epg_nrdy;
if (error == 0) {
(void)so->so_proto->pr_ready(so, m, npages);
} else {
ktls_drop(so, EIO);
mb_free_notready(m, npages);
}
sorele(so);
CURVNET_RESTORE();
}
/*
* Similar to ktls_encrypt, but used with asynchronous OCF backends
* (coprocessors) where encryption does not use host CPU resources and
* it can be beneficial to queue more requests than CPUs.
*/
static __noinline void
ktls_encrypt_async(struct ktls_wq *wq, struct mbuf *top)
{
struct ktls_ocf_encrypt_state *state;
struct ktls_session *tls;
struct socket *so;
struct mbuf *m, *n;
int error, mpages, npages, total_pages;
so = top->m_epg_so;
tls = top->m_epg_tls;
KASSERT(tls != NULL, ("tls = NULL, top = %p\n", top));
KASSERT(so != NULL, ("so = NULL, top = %p\n", top));
#ifdef INVARIANTS
top->m_epg_so = NULL;
#endif
total_pages = top->m_epg_enc_cnt;
npages = 0;
error = 0;
for (m = top; npages != total_pages; m = n) {
KASSERT(m->m_epg_tls == tls,
("different TLS sessions in a single mbuf chain: %p vs %p",
tls, m->m_epg_tls));
KASSERT(npages + m->m_epg_npgs <= total_pages,
("page count mismatch: top %p, total_pages %d, m %p", top,
total_pages, m));
state = malloc(sizeof(*state), M_KTLS, M_WAITOK | M_ZERO);
soref(so);
state->so = so;
state->m = m;
mpages = m->m_epg_nrdy;
n = m->m_next;
error = ktls_encrypt_record(wq, m, tls, state);
if (error) {
counter_u64_add(ktls_offload_failed_crypto, 1);
free(state, M_KTLS);
CURVNET_SET(so->so_vnet);
sorele(so);
CURVNET_RESTORE();
break;
}
npages += mpages;
}
CURVNET_SET(so->so_vnet);
if (error != 0) {
ktls_drop(so, EIO);
mb_free_notready(m, total_pages - npages);
}
sorele(so);
CURVNET_RESTORE();
}
static int
ktls_bind_domain(int domain)
{
int error;
error = cpuset_setthread(curthread->td_tid, &cpuset_domain[domain]);
if (error != 0)
return (error);
curthread->td_domain.dr_policy = DOMAINSET_PREF(domain);
return (0);
}
static void
ktls_reclaim_thread(void *ctx)
{
struct ktls_domain_info *ktls_domain = ctx;
struct ktls_reclaim_thread *sc = &ktls_domain->reclaim_td;
struct sysctl_oid *oid;
char name[80];
int error, domain;
domain = ktls_domain - ktls_domains;
if (bootverbose)
printf("Starting KTLS reclaim thread for domain %d\n", domain);
error = ktls_bind_domain(domain);
if (error)
printf("Unable to bind KTLS reclaim thread for domain %d: error %d\n",
domain, error);
snprintf(name, sizeof(name), "domain%d", domain);
oid = SYSCTL_ADD_NODE(NULL, SYSCTL_STATIC_CHILDREN(_kern_ipc_tls), OID_AUTO,
name, CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "");
SYSCTL_ADD_U64(NULL, SYSCTL_CHILDREN(oid), OID_AUTO, "reclaims",
CTLFLAG_RD, &sc->reclaims, 0, "buffers reclaimed");
SYSCTL_ADD_U64(NULL, SYSCTL_CHILDREN(oid), OID_AUTO, "wakeups",
CTLFLAG_RD, &sc->wakeups, 0, "thread wakeups");
SYSCTL_ADD_INT(NULL, SYSCTL_CHILDREN(oid), OID_AUTO, "running",
CTLFLAG_RD, &sc->running, 0, "thread running");
for (;;) {
atomic_store_int(&sc->running, 0);
tsleep(sc, PZERO | PNOLOCK, "-", 0);
atomic_store_int(&sc->running, 1);
sc->wakeups++;
/*
* Below we attempt to reclaim ktls_max_reclaim
* buffers using vm_page_reclaim_contig_domain_ext().
* We do this here, as this function can take several
* seconds to scan all of memory and it does not
* matter if this thread pauses for a while. If we
* block a ktls worker thread, we risk developing
* backlogs of buffers to be encrypted, leading to
* surges of traffic and potential NIC output drops.
*/
if (!vm_page_reclaim_contig_domain_ext(domain, VM_ALLOC_NORMAL,
atop(ktls_maxlen), 0, ~0ul, PAGE_SIZE, 0, ktls_max_reclaim)) {
vm_wait_domain(domain);
} else {
sc->reclaims += ktls_max_reclaim;
}
}
}
static void
ktls_work_thread(void *ctx)
{
struct ktls_wq *wq = ctx;
struct mbuf *m, *n;
struct socket *so, *son;
STAILQ_HEAD(, mbuf) local_m_head;
STAILQ_HEAD(, socket) local_so_head;
int cpu;
cpu = wq - ktls_wq;
if (bootverbose)
printf("Starting KTLS worker thread for CPU %d\n", cpu);
/*
* Bind to a core. If ktls_bind_threads is > 1, then
* we bind to the NUMA domain instead.
*/
if (ktls_bind_threads) {
int error;
if (ktls_bind_threads > 1) {
struct pcpu *pc = pcpu_find(cpu);
error = ktls_bind_domain(pc->pc_domain);
} else {
cpuset_t mask;
CPU_SETOF(cpu, &mask);
error = cpuset_setthread(curthread->td_tid, &mask);
}
if (error)
printf("Unable to bind KTLS worker thread for CPU %d: error %d\n",
cpu, error);
}
#if defined(__aarch64__) || defined(__amd64__) || defined(__i386__)
fpu_kern_thread(0);
#endif
for (;;) {
mtx_lock(&wq->mtx);
while (STAILQ_EMPTY(&wq->m_head) &&
STAILQ_EMPTY(&wq->so_head)) {
wq->running = false;
mtx_sleep(wq, &wq->mtx, 0, "-", 0);
wq->running = true;
}
STAILQ_INIT(&local_m_head);
STAILQ_CONCAT(&local_m_head, &wq->m_head);
STAILQ_INIT(&local_so_head);
STAILQ_CONCAT(&local_so_head, &wq->so_head);
mtx_unlock(&wq->mtx);
STAILQ_FOREACH_SAFE(m, &local_m_head, m_epg_stailq, n) {
if (m->m_epg_flags & EPG_FLAG_2FREE) {
ktls_free(m->m_epg_tls);
m_free_raw(m);
} else {
if (m->m_epg_tls->sync_dispatch)
ktls_encrypt(wq, m);
else
ktls_encrypt_async(wq, m);
counter_u64_add(ktls_cnt_tx_queued, -1);
}
}
STAILQ_FOREACH_SAFE(so, &local_so_head, so_ktls_rx_list, son) {
ktls_decrypt(so);
counter_u64_add(ktls_cnt_rx_queued, -1);
}
}
}
static void
ktls_disable_ifnet_help(void *context, int pending __unused)
{
struct ktls_session *tls;
struct inpcb *inp;
struct tcpcb *tp;
struct socket *so;
int err;
tls = context;
inp = tls->inp;
if (inp == NULL)
return;
INP_WLOCK(inp);
so = inp->inp_socket;
MPASS(so != NULL);
if (inp->inp_flags & INP_DROPPED) {
goto out;
}
if (so->so_snd.sb_tls_info != NULL)
err = ktls_set_tx_mode(so, TCP_TLS_MODE_SW);
else
err = ENXIO;
if (err == 0) {
counter_u64_add(ktls_ifnet_disable_ok, 1);
/* ktls_set_tx_mode() drops inp wlock, so recheck flags */
if ((inp->inp_flags & INP_DROPPED) == 0 &&
(tp = intotcpcb(inp)) != NULL &&
tp->t_fb->tfb_hwtls_change != NULL)
(*tp->t_fb->tfb_hwtls_change)(tp, 0);
} else {
counter_u64_add(ktls_ifnet_disable_fail, 1);
}
out:
CURVNET_SET(so->so_vnet);
sorele(so);
CURVNET_RESTORE();
INP_WUNLOCK(inp);
ktls_free(tls);
}
/*
* Called when re-transmits are becoming a substantial portion of the
* sends on this connection. When this happens, we transition the
* connection to software TLS. This is needed because most inline TLS
* NICs keep crypto state only for in-order transmits. This means
* that to handle a TCP rexmit (which is out-of-order), the NIC must
* re-DMA the entire TLS record up to and including the current
* segment. This means that when re-transmitting the last ~1448 byte
* segment of a 16KB TLS record, we could wind up re-DMA'ing an order
* of magnitude more data than we are sending. This can cause the
* PCIe link to saturate well before the network, which can cause
* output drops, and a general loss of capacity.
*/
void
ktls_disable_ifnet(void *arg)
{
struct tcpcb *tp;
struct inpcb *inp;
struct socket *so;
struct ktls_session *tls;
tp = arg;
inp = tptoinpcb(tp);
INP_WLOCK_ASSERT(inp);
so = inp->inp_socket;
SOCK_LOCK(so);
tls = so->so_snd.sb_tls_info;
if (tp->t_nic_ktls_xmit_dis == 1) {
SOCK_UNLOCK(so);
return;
}
/*
* note that t_nic_ktls_xmit_dis is never cleared; disabling
* ifnet can only be done once per connection, so we never want
* to do it again
*/
(void)ktls_hold(tls);
soref(so);
tp->t_nic_ktls_xmit_dis = 1;
SOCK_UNLOCK(so);
TASK_INIT(&tls->disable_ifnet_task, 0, ktls_disable_ifnet_help, tls);
(void)taskqueue_enqueue(taskqueue_thread, &tls->disable_ifnet_task);
}
diff --git a/sys/sys/sockbuf.h b/sys/sys/sockbuf.h
index 14107f5b2a10..a6ec72975252 100644
--- a/sys/sys/sockbuf.h
+++ b/sys/sys/sockbuf.h
@@ -1,319 +1,320 @@
/*-
* SPDX-License-Identifier: BSD-3-Clause
*
* Copyright (c) 1982, 1986, 1990, 1993
* The Regents of the University of California. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* 3. Neither the name of the University nor the names of its contributors
* may be used to endorse or promote products derived from this software
* without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*
* @(#)socketvar.h 8.3 (Berkeley) 2/19/95
*/
#ifndef _SYS_SOCKBUF_H_
#define _SYS_SOCKBUF_H_
/*
* Constants for sb_flags field of struct sockbuf/xsockbuf.
*/
#define SB_TLS_RX 0x01 /* using KTLS on RX */
#define SB_TLS_RX_RUNNING 0x02 /* KTLS RX operation running */
#define SB_WAIT 0x04 /* someone is waiting for data/space */
#define SB_SEL 0x08 /* someone is selecting */
#define SB_ASYNC 0x10 /* ASYNC I/O, need signals */
#define SB_UPCALL 0x20 /* someone wants an upcall */
#define SB_NOINTR 0x40 /* operations not interruptible */
#define SB_AIO 0x80 /* AIO operations queued */
#define SB_KNOTE 0x100 /* kernel note attached */
#define SB_NOCOALESCE 0x200 /* don't coalesce new data into existing mbufs */
#define SB_IN_TOE 0x400 /* socket buffer is in the middle of an operation */
#define SB_AUTOSIZE 0x800 /* automatically size socket buffer */
#define SB_STOP 0x1000 /* backpressure indicator */
#define SB_AIO_RUNNING 0x2000 /* AIO operation running */
#define SB_UNUSED 0x4000 /* previously used for SB_TLS_IFNET */
#define SB_TLS_RX_RESYNC 0x8000 /* KTLS RX lost HW sync */
#define SBS_CANTSENDMORE 0x0010 /* can't send more data to peer */
#define SBS_CANTRCVMORE 0x0020 /* can't receive more data from peer */
#define SBS_RCVATMARK 0x0040 /* at mark on input */
#if defined(_KERNEL) || defined(_WANT_SOCKET)
#include <sys/_lock.h>
#include <sys/_mutex.h>
#include <sys/_sx.h>
#include <sys/_task.h>
#define SB_MAX (2*1024*1024) /* default for max chars in sockbuf */
struct ktls_session;
struct mbuf;
struct sockaddr;
struct socket;
struct sockopt;
struct thread;
struct selinfo;
/*
* Socket buffer
*
* A buffer starts with the fields that are accessed by I/O multiplexing
* APIs like select(2), kevent(2) or AIO and thus are shared between different
* buffer implementations. They are protected by the SOCK_RECVBUF_LOCK()
* or SOCK_SENDBUF_LOCK() of the owning socket.
*
* XXX: sb_acc, sb_ccc and sb_mbcnt shall become implementation specific
* methods.
*
* Protocol specific implementations follow in a union.
*/
struct sockbuf {
struct selinfo *sb_sel; /* process selecting read/write */
short sb_state; /* socket state on sockbuf */
short sb_flags; /* flags, see above */
u_int sb_acc; /* available chars in buffer */
u_int sb_ccc; /* claimed chars in buffer */
u_int sb_mbcnt; /* chars of mbufs used */
u_int sb_ctl; /* non-data chars in buffer */
u_int sb_hiwat; /* max actual char count */
u_int sb_lowat; /* low water mark */
u_int sb_mbmax; /* max chars of mbufs to use */
sbintime_t sb_timeo; /* timeout for read/write */
int (*sb_upcall)(struct socket *, void *, int);
void *sb_upcallarg;
TAILQ_HEAD(, kaiocb) sb_aiojobq; /* pending AIO ops */
struct task sb_aiotask; /* AIO task */
union {
/*
* Classic BSD one-size-fits-all socket buffer, capable of
* doing streams and datagrams. The stream part is able
* to perform special features:
* - not ready data (sendfile)
* - TLS
*/
struct {
/* compat: sockbuf lock pointer */
struct mtx *sb_mtx;
/* first and last mbufs in the chain */
struct mbuf *sb_mb;
struct mbuf *sb_mbtail;
/* first mbuf of last record in socket buffer */
struct mbuf *sb_lastrecord;
/* pointer to data to send next (TCP */
struct mbuf *sb_sndptr;
/* pointer to first not ready buffer */
struct mbuf *sb_fnrdy;
/* byte offset of ptr into chain, used with sb_sndptr */
u_int sb_sndptroff;
/* TLS */
u_int sb_tlscc; /* TLS chain characters */
u_int sb_tlsdcc; /* characters being decrypted */
struct mbuf *sb_mtls; /* TLS mbuf chain */
struct mbuf *sb_mtlstail; /* last mbuf in TLS chain */
uint64_t sb_tls_seqno; /* TLS seqno */
- struct ktls_session *sb_tls_info; /* TLS state */
+ /* TLS state, locked by sockbuf and sock I/O mutexes. */
+ struct ktls_session *sb_tls_info;
};
/*
* PF_UNIX/SOCK_DGRAM
*
* Local protocol, thus we should buffer on the receive side
* only. However, in one to many configuration we don't want
* a single receive buffer to be shared. So we would link
* send buffers onto receive buffer. All the fields are locked
* by the receive buffer lock.
*/
struct {
/*
* For receive buffer: own queue of this buffer for
* unconnected sends. For send buffer: queue lended
* to the peer receive buffer, to isolate ourselves
* from other senders.
*/
STAILQ_HEAD(, mbuf) uxdg_mb;
/* For receive buffer: datagram seen via MSG_PEEK. */
struct mbuf *uxdg_peeked;
/*
* For receive buffer: queue of send buffers of
* connected peers. For send buffer: linkage on
* connected peer receive buffer queue.
*/
union {
TAILQ_HEAD(, sockbuf) uxdg_conns;
TAILQ_ENTRY(sockbuf) uxdg_clist;
};
/* Counters for this buffer uxdg_mb chain + peeked. */
u_int uxdg_cc;
u_int uxdg_ctl;
u_int uxdg_mbcnt;
};
};
};
#endif /* defined(_KERNEL) || defined(_WANT_SOCKET) */
#ifdef _KERNEL
/* 'which' values for KPIs that operate on one buffer of a socket. */
typedef enum { SO_RCV, SO_SND } sb_which;
/*
* Per-socket buffer mutex used to protect most fields in the socket buffer.
* These make use of the mutex pointer embedded in struct sockbuf, which
* currently just references mutexes in the containing socket. The
* SOCK_SENDBUF_LOCK() etc. macros can be used instead of or in combination with
* these locking macros.
*/
#define SOCKBUF_MTX(_sb) ((_sb)->sb_mtx)
#define SOCKBUF_LOCK(_sb) mtx_lock(SOCKBUF_MTX(_sb))
#define SOCKBUF_OWNED(_sb) mtx_owned(SOCKBUF_MTX(_sb))
#define SOCKBUF_UNLOCK(_sb) mtx_unlock(SOCKBUF_MTX(_sb))
#define SOCKBUF_LOCK_ASSERT(_sb) mtx_assert(SOCKBUF_MTX(_sb), MA_OWNED)
#define SOCKBUF_UNLOCK_ASSERT(_sb) mtx_assert(SOCKBUF_MTX(_sb), MA_NOTOWNED)
/*
* Socket buffer private mbuf(9) flags.
*/
#define M_NOTREADY M_PROTO1 /* m_data not populated yet */
#define M_BLOCKED M_PROTO2 /* M_NOTREADY in front of m */
#define M_NOTAVAIL (M_NOTREADY | M_BLOCKED)
void sbappend(struct sockbuf *sb, struct mbuf *m, int flags);
void sbappend_locked(struct sockbuf *sb, struct mbuf *m, int flags);
void sbappendstream(struct sockbuf *sb, struct mbuf *m, int flags);
void sbappendstream_locked(struct sockbuf *sb, struct mbuf *m, int flags);
int sbappendaddr(struct sockbuf *sb, const struct sockaddr *asa,
struct mbuf *m0, struct mbuf *control);
int sbappendaddr_locked(struct sockbuf *sb, const struct sockaddr *asa,
struct mbuf *m0, struct mbuf *control);
int sbappendaddr_nospacecheck_locked(struct sockbuf *sb,
const struct sockaddr *asa, struct mbuf *m0, struct mbuf *control);
void sbappendcontrol(struct sockbuf *sb, struct mbuf *m0,
struct mbuf *control, int flags);
void sbappendcontrol_locked(struct sockbuf *sb, struct mbuf *m0,
struct mbuf *control, int flags);
void sbappendrecord(struct sockbuf *sb, struct mbuf *m0);
void sbappendrecord_locked(struct sockbuf *sb, struct mbuf *m0);
void sbcompress(struct sockbuf *sb, struct mbuf *m, struct mbuf *n);
struct mbuf *
sbcreatecontrol(const void *p, u_int size, int type, int level,
int wait);
void sbdestroy(struct socket *, sb_which);
void sbdrop(struct sockbuf *sb, int len);
void sbdrop_locked(struct sockbuf *sb, int len);
struct mbuf *
sbcut_locked(struct sockbuf *sb, int len);
void sbdroprecord(struct sockbuf *sb);
void sbdroprecord_locked(struct sockbuf *sb);
void sbflush(struct sockbuf *sb);
void sbflush_locked(struct sockbuf *sb);
void sbrelease(struct socket *, sb_which);
void sbrelease_locked(struct socket *, sb_which);
int sbsetopt(struct socket *so, struct sockopt *);
bool sbreserve_locked(struct socket *so, sb_which which, u_long cc,
struct thread *td);
bool sbreserve_locked_limit(struct socket *so, sb_which which, u_long cc,
u_long buf_max, struct thread *td);
void sbsndptr_adv(struct sockbuf *sb, struct mbuf *mb, u_int len);
struct mbuf *
sbsndptr_noadv(struct sockbuf *sb, u_int off, u_int *moff);
struct mbuf *
sbsndmbuf(struct sockbuf *sb, u_int off, u_int *moff);
int sbwait(struct socket *, sb_which);
void sballoc(struct sockbuf *, struct mbuf *);
void sbfree(struct sockbuf *, struct mbuf *);
void sballoc_ktls_rx(struct sockbuf *sb, struct mbuf *m);
void sbfree_ktls_rx(struct sockbuf *sb, struct mbuf *m);
int sbready(struct sockbuf *, struct mbuf *, int);
/*
* Return how much data is available to be taken out of socket
* buffer right now.
*/
static inline u_int
sbavail(struct sockbuf *sb)
{
#if 0
SOCKBUF_LOCK_ASSERT(sb);
#endif
return (sb->sb_acc);
}
/*
* Return how much data sits there in the socket buffer
* It might be that some data is not yet ready to be read.
*/
static inline u_int
sbused(struct sockbuf *sb)
{
#if 0
SOCKBUF_LOCK_ASSERT(sb);
#endif
return (sb->sb_ccc);
}
/*
* How much space is there in a socket buffer (so->so_snd or so->so_rcv)?
* This is problematical if the fields are unsigned, as the space might
* still be negative (ccc > hiwat or mbcnt > mbmax).
*/
static inline long
sbspace(struct sockbuf *sb)
{
int bleft, mleft; /* size should match sockbuf fields */
#if 0
SOCKBUF_LOCK_ASSERT(sb);
#endif
if (sb->sb_flags & SB_STOP)
return(0);
bleft = sb->sb_hiwat - sb->sb_ccc;
mleft = sb->sb_mbmax - sb->sb_mbcnt;
return ((bleft < mleft) ? bleft : mleft);
}
#define SB_EMPTY_FIXUP(sb) do { \
if ((sb)->sb_mb == NULL) { \
(sb)->sb_mbtail = NULL; \
(sb)->sb_lastrecord = NULL; \
} \
} while (/*CONSTCOND*/0)
#ifdef SOCKBUF_DEBUG
void sblastrecordchk(struct sockbuf *, const char *, int);
void sblastmbufchk(struct sockbuf *, const char *, int);
void sbcheck(struct sockbuf *, const char *, int);
#define SBLASTRECORDCHK(sb) sblastrecordchk((sb), __FILE__, __LINE__)
#define SBLASTMBUFCHK(sb) sblastmbufchk((sb), __FILE__, __LINE__)
#define SBCHECK(sb) sbcheck((sb), __FILE__, __LINE__)
#else
#define SBLASTRECORDCHK(sb) do {} while (0)
#define SBLASTMBUFCHK(sb) do {} while (0)
#define SBCHECK(sb) do {} while (0)
#endif /* SOCKBUF_DEBUG */
#endif /* _KERNEL */
#endif /* _SYS_SOCKBUF_H_ */
diff --git a/tests/sys/kern/ktls_test.c b/tests/sys/kern/ktls_test.c
index f57ae74112a2..72497196b945 100644
--- a/tests/sys/kern/ktls_test.c
+++ b/tests/sys/kern/ktls_test.c
@@ -1,2848 +1,2848 @@
/*-
* SPDX-License-Identifier: BSD-2-Clause
*
* Copyright (c) 2021 Netflix Inc.
* Written by: John Baldwin <jhb@FreeBSD.org>
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*/
#include <sys/param.h>
#include <sys/endian.h>
#include <sys/event.h>
#include <sys/ktls.h>
#include <sys/socket.h>
#include <sys/sysctl.h>
#include <netinet/in.h>
#include <netinet/tcp.h>
#include <crypto/cryptodev.h>
#include <assert.h>
#include <err.h>
#include <fcntl.h>
#include <libutil.h>
#include <netdb.h>
#include <poll.h>
#include <stdbool.h>
#include <stdlib.h>
#include <atf-c.h>
#include <openssl/err.h>
#include <openssl/evp.h>
#include <openssl/hmac.h>
static void
require_ktls(void)
{
size_t len;
bool enable;
len = sizeof(enable);
if (sysctlbyname("kern.ipc.tls.enable", &enable, &len, NULL, 0) == -1) {
if (errno == ENOENT)
atf_tc_skip("kernel does not support TLS offload");
atf_libc_error(errno, "Failed to read kern.ipc.tls.enable");
}
if (!enable)
atf_tc_skip("Kernel TLS is disabled");
}
#define ATF_REQUIRE_KTLS() require_ktls()
static void
check_tls_mode(const atf_tc_t *tc, int s, int sockopt)
{
if (atf_tc_get_config_var_as_bool_wd(tc, "ktls.require_ifnet", false)) {
socklen_t len;
int mode;
len = sizeof(mode);
if (getsockopt(s, IPPROTO_TCP, sockopt, &mode, &len) == -1)
atf_libc_error(errno, "Failed to fetch TLS mode");
if (mode != TCP_TLS_MODE_IFNET)
atf_tc_skip("connection did not use ifnet TLS");
}
if (atf_tc_get_config_var_as_bool_wd(tc, "ktls.require_toe", false)) {
socklen_t len;
int mode;
len = sizeof(mode);
if (getsockopt(s, IPPROTO_TCP, sockopt, &mode, &len) == -1)
atf_libc_error(errno, "Failed to fetch TLS mode");
if (mode != TCP_TLS_MODE_TOE)
atf_tc_skip("connection did not use TOE TLS");
}
}
static void __printflike(2, 3)
debug(const atf_tc_t *tc, const char *fmt, ...)
{
if (!atf_tc_get_config_var_as_bool_wd(tc, "ktls.debug", false))
return;
va_list ap;
va_start(ap, fmt);
vprintf(fmt, ap);
va_end(ap);
}
static void
debug_hexdump(const atf_tc_t *tc, const void *buf, int length,
const char *label)
{
if (!atf_tc_get_config_var_as_bool_wd(tc, "ktls.debug", false))
return;
if (label != NULL)
printf("%s:\n", label);
hexdump(buf, length, NULL, 0);
}
static char
rdigit(void)
{
/* ASCII printable values between 0x20 and 0x7e */
return (0x20 + random() % (0x7f - 0x20));
}
static char *
alloc_buffer(size_t len)
{
char *buf;
size_t i;
if (len == 0)
return (NULL);
buf = malloc(len);
for (i = 0; i < len; i++)
buf[i] = rdigit();
return (buf);
}
static bool
socketpair_tcp(int sv[2])
{
struct pollfd pfd;
struct sockaddr_in sin;
socklen_t len;
int as, cs, ls;
ls = socket(PF_INET, SOCK_STREAM, 0);
if (ls == -1) {
warn("socket() for listen");
return (false);
}
memset(&sin, 0, sizeof(sin));
sin.sin_len = sizeof(sin);
sin.sin_family = AF_INET;
sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
if (bind(ls, (struct sockaddr *)&sin, sizeof(sin)) == -1) {
warn("bind");
close(ls);
return (false);
}
if (listen(ls, 1) == -1) {
warn("listen");
close(ls);
return (false);
}
len = sizeof(sin);
if (getsockname(ls, (struct sockaddr *)&sin, &len) == -1) {
warn("getsockname");
close(ls);
return (false);
}
cs = socket(PF_INET, SOCK_STREAM | SOCK_NONBLOCK, 0);
if (cs == -1) {
warn("socket() for connect");
close(ls);
return (false);
}
if (connect(cs, (struct sockaddr *)&sin, sizeof(sin)) == -1) {
if (errno != EINPROGRESS) {
warn("connect");
close(ls);
close(cs);
return (false);
}
}
as = accept4(ls, NULL, NULL, SOCK_NONBLOCK);
if (as == -1) {
warn("accept4");
close(ls);
close(cs);
return (false);
}
close(ls);
pfd.fd = cs;
pfd.events = POLLOUT;
pfd.revents = 0;
ATF_REQUIRE_INTEQ(1, poll(&pfd, 1, INFTIM));
ATF_REQUIRE_INTEQ(POLLOUT, pfd.revents);
sv[0] = cs;
sv[1] = as;
return (true);
}
static bool
echo_socket(const atf_tc_t *tc, int sv[2])
{
const char *cause, *host, *port;
struct addrinfo hints, *ai, *tofree;
int error, flags, s;
host = atf_tc_get_config_var(tc, "ktls.host");
port = atf_tc_get_config_var_wd(tc, "ktls.port", "echo");
memset(&hints, 0, sizeof(hints));
hints.ai_family = AF_UNSPEC;
hints.ai_socktype = SOCK_STREAM;
hints.ai_protocol = IPPROTO_TCP;
error = getaddrinfo(host, port, &hints, &tofree);
if (error != 0) {
warnx("getaddrinfo(%s:%s) failed: %s", host, port,
gai_strerror(error));
return (false);
}
cause = NULL;
for (ai = tofree; ai != NULL; ai = ai->ai_next) {
s = socket(ai->ai_family, ai->ai_socktype, ai->ai_protocol);
if (s == -1) {
cause = "socket";
error = errno;
continue;
}
if (connect(s, ai->ai_addr, ai->ai_addrlen) == -1) {
cause = "connect";
error = errno;
close(s);
continue;
}
freeaddrinfo(tofree);
ATF_REQUIRE((flags = fcntl(s, F_GETFL)) != -1);
flags |= O_NONBLOCK;
ATF_REQUIRE(fcntl(s, F_SETFL, flags) != -1);
sv[0] = s;
sv[1] = s;
return (true);
}
warnc(error, "%s", cause);
freeaddrinfo(tofree);
return (false);
}
static bool
open_sockets(const atf_tc_t *tc, int sv[2])
{
if (atf_tc_has_config_var(tc, "ktls.host"))
return (echo_socket(tc, sv));
else
return (socketpair_tcp(sv));
}
static void
close_sockets(int sv[2])
{
if (sv[0] != sv[1])
ATF_REQUIRE(close(sv[1]) == 0);
ATF_REQUIRE(close(sv[0]) == 0);
}
static void
close_sockets_ignore_errors(int sv[2])
{
if (sv[0] != sv[1])
close(sv[1]);
close(sv[0]);
}
static void
fd_set_blocking(int fd)
{
int flags;
ATF_REQUIRE((flags = fcntl(fd, F_GETFL)) != -1);
flags &= ~O_NONBLOCK;
ATF_REQUIRE(fcntl(fd, F_SETFL, flags) != -1);
}
static bool
cbc_crypt(const EVP_CIPHER *cipher, const char *key, const char *iv,
const char *input, char *output, size_t size, int enc)
{
EVP_CIPHER_CTX *ctx;
int outl, total;
ctx = EVP_CIPHER_CTX_new();
if (ctx == NULL) {
warnx("EVP_CIPHER_CTX_new failed: %s",
ERR_error_string(ERR_get_error(), NULL));
return (false);
}
if (EVP_CipherInit_ex(ctx, cipher, NULL, (const u_char *)key,
(const u_char *)iv, enc) != 1) {
warnx("EVP_CipherInit_ex failed: %s",
ERR_error_string(ERR_get_error(), NULL));
EVP_CIPHER_CTX_free(ctx);
return (false);
}
EVP_CIPHER_CTX_set_padding(ctx, 0);
if (EVP_CipherUpdate(ctx, (u_char *)output, &outl,
(const u_char *)input, size) != 1) {
warnx("EVP_CipherUpdate failed: %s",
ERR_error_string(ERR_get_error(), NULL));
EVP_CIPHER_CTX_free(ctx);
return (false);
}
total = outl;
if (EVP_CipherFinal_ex(ctx, (u_char *)output + outl, &outl) != 1) {
warnx("EVP_CipherFinal_ex failed: %s",
ERR_error_string(ERR_get_error(), NULL));
EVP_CIPHER_CTX_free(ctx);
return (false);
}
total += outl;
if ((size_t)total != size) {
warnx("decrypt size mismatch: %zu vs %d", size, total);
EVP_CIPHER_CTX_free(ctx);
return (false);
}
EVP_CIPHER_CTX_free(ctx);
return (true);
}
static bool
cbc_encrypt(const EVP_CIPHER *cipher, const char *key, const char *iv,
const char *input, char *output, size_t size)
{
return (cbc_crypt(cipher, key, iv, input, output, size, 1));
}
static bool
cbc_decrypt(const EVP_CIPHER *cipher, const char *key, const char *iv,
const char *input, char *output, size_t size)
{
return (cbc_crypt(cipher, key, iv, input, output, size, 0));
}
static bool
compute_hash(const EVP_MD *md, const void *key, size_t key_len, const void *aad,
size_t aad_len, const void *buffer, size_t len, void *digest,
u_int *digest_len)
{
HMAC_CTX *ctx;
ctx = HMAC_CTX_new();
if (ctx == NULL) {
warnx("HMAC_CTX_new failed: %s",
ERR_error_string(ERR_get_error(), NULL));
return (false);
}
if (HMAC_Init_ex(ctx, key, key_len, md, NULL) != 1) {
warnx("HMAC_Init_ex failed: %s",
ERR_error_string(ERR_get_error(), NULL));
HMAC_CTX_free(ctx);
return (false);
}
if (HMAC_Update(ctx, aad, aad_len) != 1) {
warnx("HMAC_Update (aad) failed: %s",
ERR_error_string(ERR_get_error(), NULL));
HMAC_CTX_free(ctx);
return (false);
}
if (HMAC_Update(ctx, buffer, len) != 1) {
warnx("HMAC_Update (payload) failed: %s",
ERR_error_string(ERR_get_error(), NULL));
HMAC_CTX_free(ctx);
return (false);
}
if (HMAC_Final(ctx, digest, digest_len) != 1) {
warnx("HMAC_Final failed: %s",
ERR_error_string(ERR_get_error(), NULL));
HMAC_CTX_free(ctx);
return (false);
}
HMAC_CTX_free(ctx);
return (true);
}
static bool
verify_hash(const EVP_MD *md, const void *key, size_t key_len, const void *aad,
size_t aad_len, const void *buffer, size_t len, const void *digest)
{
unsigned char digest2[EVP_MAX_MD_SIZE];
u_int digest_len;
if (!compute_hash(md, key, key_len, aad, aad_len, buffer, len, digest2,
&digest_len))
return (false);
if (memcmp(digest, digest2, digest_len) != 0) {
warnx("HMAC mismatch");
return (false);
}
return (true);
}
static bool
aead_encrypt(const EVP_CIPHER *cipher, const char *key, const char *nonce,
const void *aad, size_t aad_len, const char *input, char *output,
size_t size, char *tag, size_t tag_len)
{
EVP_CIPHER_CTX *ctx;
int outl, total;
ctx = EVP_CIPHER_CTX_new();
if (ctx == NULL) {
warnx("EVP_CIPHER_CTX_new failed: %s",
ERR_error_string(ERR_get_error(), NULL));
return (false);
}
if (EVP_EncryptInit_ex(ctx, cipher, NULL, (const u_char *)key,
(const u_char *)nonce) != 1) {
warnx("EVP_EncryptInit_ex failed: %s",
ERR_error_string(ERR_get_error(), NULL));
EVP_CIPHER_CTX_free(ctx);
return (false);
}
EVP_CIPHER_CTX_set_padding(ctx, 0);
if (aad != NULL) {
if (EVP_EncryptUpdate(ctx, NULL, &outl, (const u_char *)aad,
aad_len) != 1) {
warnx("EVP_EncryptUpdate for AAD failed: %s",
ERR_error_string(ERR_get_error(), NULL));
EVP_CIPHER_CTX_free(ctx);
return (false);
}
}
if (EVP_EncryptUpdate(ctx, (u_char *)output, &outl,
(const u_char *)input, size) != 1) {
warnx("EVP_EncryptUpdate failed: %s",
ERR_error_string(ERR_get_error(), NULL));
EVP_CIPHER_CTX_free(ctx);
return (false);
}
total = outl;
if (EVP_EncryptFinal_ex(ctx, (u_char *)output + outl, &outl) != 1) {
warnx("EVP_EncryptFinal_ex failed: %s",
ERR_error_string(ERR_get_error(), NULL));
EVP_CIPHER_CTX_free(ctx);
return (false);
}
total += outl;
if ((size_t)total != size) {
warnx("encrypt size mismatch: %zu vs %d", size, total);
EVP_CIPHER_CTX_free(ctx);
return (false);
}
if (EVP_CIPHER_CTX_ctrl(ctx, EVP_CTRL_AEAD_GET_TAG, tag_len, tag) !=
1) {
warnx("EVP_CIPHER_CTX_ctrl(EVP_CTRL_AEAD_GET_TAG) failed: %s",
ERR_error_string(ERR_get_error(), NULL));
EVP_CIPHER_CTX_free(ctx);
return (false);
}
EVP_CIPHER_CTX_free(ctx);
return (true);
}
static bool
aead_decrypt(const EVP_CIPHER *cipher, const char *key, const char *nonce,
const void *aad, size_t aad_len, const char *input, char *output,
size_t size, const char *tag, size_t tag_len)
{
EVP_CIPHER_CTX *ctx;
int outl, total;
bool valid;
ctx = EVP_CIPHER_CTX_new();
if (ctx == NULL) {
warnx("EVP_CIPHER_CTX_new failed: %s",
ERR_error_string(ERR_get_error(), NULL));
return (false);
}
if (EVP_DecryptInit_ex(ctx, cipher, NULL, (const u_char *)key,
(const u_char *)nonce) != 1) {
warnx("EVP_DecryptInit_ex failed: %s",
ERR_error_string(ERR_get_error(), NULL));
EVP_CIPHER_CTX_free(ctx);
return (false);
}
EVP_CIPHER_CTX_set_padding(ctx, 0);
if (aad != NULL) {
if (EVP_DecryptUpdate(ctx, NULL, &outl, (const u_char *)aad,
aad_len) != 1) {
warnx("EVP_DecryptUpdate for AAD failed: %s",
ERR_error_string(ERR_get_error(), NULL));
EVP_CIPHER_CTX_free(ctx);
return (false);
}
}
if (EVP_DecryptUpdate(ctx, (u_char *)output, &outl,
(const u_char *)input, size) != 1) {
warnx("EVP_DecryptUpdate failed: %s",
ERR_error_string(ERR_get_error(), NULL));
EVP_CIPHER_CTX_free(ctx);
return (false);
}
total = outl;
if (EVP_CIPHER_CTX_ctrl(ctx, EVP_CTRL_AEAD_SET_TAG, tag_len,
__DECONST(char *, tag)) != 1) {
warnx("EVP_CIPHER_CTX_ctrl(EVP_CTRL_AEAD_SET_TAG) failed: %s",
ERR_error_string(ERR_get_error(), NULL));
EVP_CIPHER_CTX_free(ctx);
return (false);
}
valid = (EVP_DecryptFinal_ex(ctx, (u_char *)output + outl, &outl) == 1);
total += outl;
if ((size_t)total != size) {
warnx("decrypt size mismatch: %zu vs %d", size, total);
EVP_CIPHER_CTX_free(ctx);
return (false);
}
if (!valid)
warnx("tag mismatch");
EVP_CIPHER_CTX_free(ctx);
return (valid);
}
static void
build_tls_enable(const atf_tc_t *tc, int cipher_alg, size_t cipher_key_len,
int auth_alg, int minor, uint64_t seqno, struct tls_enable *en)
{
u_int auth_key_len, iv_len;
memset(en, 0, sizeof(*en));
switch (cipher_alg) {
case CRYPTO_AES_CBC:
if (minor == TLS_MINOR_VER_ZERO)
iv_len = AES_BLOCK_LEN;
else
iv_len = 0;
break;
case CRYPTO_AES_NIST_GCM_16:
if (minor == TLS_MINOR_VER_TWO)
iv_len = TLS_AEAD_GCM_LEN;
else
iv_len = TLS_1_3_GCM_IV_LEN;
break;
case CRYPTO_CHACHA20_POLY1305:
iv_len = TLS_CHACHA20_IV_LEN;
break;
default:
iv_len = 0;
break;
}
switch (auth_alg) {
case CRYPTO_SHA1_HMAC:
auth_key_len = SHA1_HASH_LEN;
break;
case CRYPTO_SHA2_256_HMAC:
auth_key_len = SHA2_256_HASH_LEN;
break;
case CRYPTO_SHA2_384_HMAC:
auth_key_len = SHA2_384_HASH_LEN;
break;
default:
auth_key_len = 0;
break;
}
en->cipher_key = alloc_buffer(cipher_key_len);
debug_hexdump(tc, en->cipher_key, cipher_key_len, "cipher key");
en->iv = alloc_buffer(iv_len);
if (iv_len != 0)
debug_hexdump(tc, en->iv, iv_len, "iv");
en->auth_key = alloc_buffer(auth_key_len);
if (auth_key_len != 0)
debug_hexdump(tc, en->auth_key, auth_key_len, "auth key");
en->cipher_algorithm = cipher_alg;
en->cipher_key_len = cipher_key_len;
en->iv_len = iv_len;
en->auth_algorithm = auth_alg;
en->auth_key_len = auth_key_len;
en->tls_vmajor = TLS_MAJOR_VER_ONE;
en->tls_vminor = minor;
be64enc(en->rec_seq, seqno);
debug(tc, "seqno: %ju\n", (uintmax_t)seqno);
}
static void
free_tls_enable(struct tls_enable *en)
{
free(__DECONST(void *, en->cipher_key));
free(__DECONST(void *, en->iv));
free(__DECONST(void *, en->auth_key));
}
static const EVP_CIPHER *
tls_EVP_CIPHER(const struct tls_enable *en)
{
switch (en->cipher_algorithm) {
case CRYPTO_AES_CBC:
switch (en->cipher_key_len) {
case 128 / 8:
return (EVP_aes_128_cbc());
case 256 / 8:
return (EVP_aes_256_cbc());
default:
return (NULL);
}
break;
case CRYPTO_AES_NIST_GCM_16:
switch (en->cipher_key_len) {
case 128 / 8:
return (EVP_aes_128_gcm());
case 256 / 8:
return (EVP_aes_256_gcm());
default:
return (NULL);
}
break;
case CRYPTO_CHACHA20_POLY1305:
return (EVP_chacha20_poly1305());
default:
return (NULL);
}
}
static const EVP_MD *
tls_EVP_MD(const struct tls_enable *en)
{
switch (en->auth_algorithm) {
case CRYPTO_SHA1_HMAC:
return (EVP_sha1());
case CRYPTO_SHA2_256_HMAC:
return (EVP_sha256());
case CRYPTO_SHA2_384_HMAC:
return (EVP_sha384());
default:
return (NULL);
}
}
static size_t
tls_header_len(struct tls_enable *en)
{
size_t len;
len = sizeof(struct tls_record_layer);
switch (en->cipher_algorithm) {
case CRYPTO_AES_CBC:
if (en->tls_vminor != TLS_MINOR_VER_ZERO)
len += AES_BLOCK_LEN;
return (len);
case CRYPTO_AES_NIST_GCM_16:
if (en->tls_vminor == TLS_MINOR_VER_TWO)
len += sizeof(uint64_t);
return (len);
case CRYPTO_CHACHA20_POLY1305:
return (len);
default:
return (0);
}
}
static size_t
tls_mac_len(struct tls_enable *en)
{
switch (en->cipher_algorithm) {
case CRYPTO_AES_CBC:
switch (en->auth_algorithm) {
case CRYPTO_SHA1_HMAC:
return (SHA1_HASH_LEN);
case CRYPTO_SHA2_256_HMAC:
return (SHA2_256_HASH_LEN);
case CRYPTO_SHA2_384_HMAC:
return (SHA2_384_HASH_LEN);
default:
return (0);
}
case CRYPTO_AES_NIST_GCM_16:
return (AES_GMAC_HASH_LEN);
case CRYPTO_CHACHA20_POLY1305:
return (POLY1305_HASH_LEN);
default:
return (0);
}
}
/* Includes maximum padding for MTE. */
static size_t
tls_trailer_len(struct tls_enable *en)
{
size_t len;
len = tls_mac_len(en);
if (en->cipher_algorithm == CRYPTO_AES_CBC)
len += AES_BLOCK_LEN;
if (en->tls_vminor == TLS_MINOR_VER_THREE)
len++;
return (len);
}
/* Minimum valid record payload size for a given cipher suite. */
static size_t
tls_minimum_record_payload(struct tls_enable *en)
{
size_t len;
len = tls_header_len(en);
if (en->cipher_algorithm == CRYPTO_AES_CBC)
len += roundup2(tls_mac_len(en) + 1, AES_BLOCK_LEN);
else
len += tls_mac_len(en);
if (en->tls_vminor == TLS_MINOR_VER_THREE)
len++;
return (len - sizeof(struct tls_record_layer));
}
/* 'len' is the length of the payload application data. */
static void
tls_mte_aad(struct tls_enable *en, size_t len,
const struct tls_record_layer *hdr, uint64_t seqno, struct tls_mac_data *ad)
{
ad->seq = htobe64(seqno);
ad->type = hdr->tls_type;
ad->tls_vmajor = hdr->tls_vmajor;
ad->tls_vminor = hdr->tls_vminor;
ad->tls_length = htons(len);
}
static void
tls_12_aead_aad(struct tls_enable *en, size_t len,
const struct tls_record_layer *hdr, uint64_t seqno,
struct tls_aead_data *ad)
{
ad->seq = htobe64(seqno);
ad->type = hdr->tls_type;
ad->tls_vmajor = hdr->tls_vmajor;
ad->tls_vminor = hdr->tls_vminor;
ad->tls_length = htons(len);
}
static void
tls_13_aad(struct tls_enable *en, const struct tls_record_layer *hdr,
uint64_t seqno, struct tls_aead_data_13 *ad)
{
ad->type = hdr->tls_type;
ad->tls_vmajor = hdr->tls_vmajor;
ad->tls_vminor = hdr->tls_vminor;
ad->tls_length = hdr->tls_length;
}
static void
tls_12_gcm_nonce(struct tls_enable *en, const struct tls_record_layer *hdr,
char *nonce)
{
memcpy(nonce, en->iv, TLS_AEAD_GCM_LEN);
memcpy(nonce + TLS_AEAD_GCM_LEN, hdr + 1, sizeof(uint64_t));
}
static void
tls_13_nonce(struct tls_enable *en, uint64_t seqno, char *nonce)
{
static_assert(TLS_1_3_GCM_IV_LEN == TLS_CHACHA20_IV_LEN,
"TLS 1.3 nonce length mismatch");
memcpy(nonce, en->iv, TLS_1_3_GCM_IV_LEN);
*(uint64_t *)(nonce + 4) ^= htobe64(seqno);
}
/*
* Decrypt a TLS record 'len' bytes long at 'src' and store the result at
* 'dst'. If the TLS record header length doesn't match or 'dst' doesn't
* have sufficient room ('avail'), fail the test.
*/
static size_t
decrypt_tls_aes_cbc_mte(const atf_tc_t *tc, struct tls_enable *en,
uint64_t seqno, const void *src, size_t len, void *dst, size_t avail,
uint8_t *record_type)
{
const struct tls_record_layer *hdr;
struct tls_mac_data aad;
const char *iv;
char *buf;
size_t hdr_len, mac_len, payload_len;
int padding;
hdr = src;
hdr_len = tls_header_len(en);
mac_len = tls_mac_len(en);
ATF_REQUIRE_INTEQ(TLS_MAJOR_VER_ONE, hdr->tls_vmajor);
ATF_REQUIRE_INTEQ(en->tls_vminor, hdr->tls_vminor);
debug(tc, "decrypting MTE record seqno %ju:\n", (uintmax_t)seqno);
debug_hexdump(tc, src, len, NULL);
/* First, decrypt the outer payload into a temporary buffer. */
payload_len = len - hdr_len;
buf = malloc(payload_len);
if (en->tls_vminor == TLS_MINOR_VER_ZERO)
iv = en->iv;
else
iv = (void *)(hdr + 1);
debug_hexdump(tc, iv, AES_BLOCK_LEN, "iv");
ATF_REQUIRE(cbc_decrypt(tls_EVP_CIPHER(en), en->cipher_key, iv,
(const u_char *)src + hdr_len, buf, payload_len));
debug_hexdump(tc, buf, payload_len, "decrypted buffer");
/*
* Copy the last encrypted block to use as the IV for the next
* record for TLS 1.0.
*/
if (en->tls_vminor == TLS_MINOR_VER_ZERO)
memcpy(__DECONST(uint8_t *, en->iv), (const u_char *)src +
(len - AES_BLOCK_LEN), AES_BLOCK_LEN);
/*
* Verify trailing padding and strip.
*
* The kernel always generates the smallest amount of padding.
*/
padding = buf[payload_len - 1] + 1;
ATF_REQUIRE_MSG(padding > 0 && padding <= AES_BLOCK_LEN,
"invalid padding %d", padding);
ATF_REQUIRE_MSG(payload_len >= mac_len + padding,
"payload_len (%zu) < mac_len (%zu) + padding (%d)", payload_len,
mac_len, padding);
payload_len -= padding;
/* Verify HMAC. */
payload_len -= mac_len;
tls_mte_aad(en, payload_len, hdr, seqno, &aad);
debug_hexdump(tc, &aad, sizeof(aad), "aad");
ATF_REQUIRE(verify_hash(tls_EVP_MD(en), en->auth_key, en->auth_key_len,
&aad, sizeof(aad), buf, payload_len, buf + payload_len));
ATF_REQUIRE_MSG(payload_len <= avail, "payload_len (%zu) < avail (%zu)",
payload_len, avail);
memcpy(dst, buf, payload_len);
*record_type = hdr->tls_type;
return (payload_len);
}
static size_t
decrypt_tls_12_aead(const atf_tc_t *tc, struct tls_enable *en, uint64_t seqno,
const void *src, size_t len, void *dst, uint8_t *record_type)
{
const struct tls_record_layer *hdr;
struct tls_aead_data aad;
char nonce[12];
size_t hdr_len, mac_len, payload_len;
hdr = src;
hdr_len = tls_header_len(en);
mac_len = tls_mac_len(en);
payload_len = len - (hdr_len + mac_len);
ATF_REQUIRE_INTEQ(TLS_MAJOR_VER_ONE, hdr->tls_vmajor);
ATF_REQUIRE_INTEQ(TLS_MINOR_VER_TWO, hdr->tls_vminor);
debug(tc, "decrypting TLS 1.2 record seqno %ju:\n", (uintmax_t)seqno);
debug_hexdump(tc, src, len, NULL);
tls_12_aead_aad(en, payload_len, hdr, seqno, &aad);
debug_hexdump(tc, &aad, sizeof(aad), "aad");
if (en->cipher_algorithm == CRYPTO_AES_NIST_GCM_16)
tls_12_gcm_nonce(en, hdr, nonce);
else
tls_13_nonce(en, seqno, nonce);
debug_hexdump(tc, nonce, sizeof(nonce), "nonce");
ATF_REQUIRE(aead_decrypt(tls_EVP_CIPHER(en), en->cipher_key, nonce,
&aad, sizeof(aad), (const char *)src + hdr_len, dst, payload_len,
(const char *)src + hdr_len + payload_len, mac_len));
*record_type = hdr->tls_type;
return (payload_len);
}
static size_t
decrypt_tls_13_aead(const atf_tc_t *tc, struct tls_enable *en, uint64_t seqno,
const void *src, size_t len, void *dst, uint8_t *record_type)
{
const struct tls_record_layer *hdr;
struct tls_aead_data_13 aad;
char nonce[12];
char *buf;
size_t hdr_len, mac_len, payload_len;
hdr = src;
hdr_len = tls_header_len(en);
mac_len = tls_mac_len(en);
payload_len = len - (hdr_len + mac_len);
ATF_REQUIRE_MSG(payload_len >= 1,
"payload_len (%zu) too short: len %zu hdr_len %zu mac_len %zu",
payload_len, len, hdr_len, mac_len);
ATF_REQUIRE_INTEQ(TLS_RLTYPE_APP, hdr->tls_type);
ATF_REQUIRE_INTEQ(TLS_MAJOR_VER_ONE, hdr->tls_vmajor);
ATF_REQUIRE_INTEQ(TLS_MINOR_VER_TWO, hdr->tls_vminor);
debug(tc, "decrypting TLS 1.3 record seqno %ju:\n", (uintmax_t)seqno);
debug_hexdump(tc, src, len, NULL);
tls_13_aad(en, hdr, seqno, &aad);
debug_hexdump(tc, &aad, sizeof(aad), "aad");
tls_13_nonce(en, seqno, nonce);
debug_hexdump(tc, nonce, sizeof(nonce), "nonce");
/*
* Have to use a temporary buffer for the output due to the
* record type as the last byte of the trailer.
*/
buf = malloc(payload_len);
ATF_REQUIRE(aead_decrypt(tls_EVP_CIPHER(en), en->cipher_key, nonce,
&aad, sizeof(aad), (const char *)src + hdr_len, buf, payload_len,
(const char *)src + hdr_len + payload_len, mac_len));
debug_hexdump(tc, buf, payload_len, "decrypted buffer");
/* Trim record type. */
*record_type = buf[payload_len - 1];
payload_len--;
memcpy(dst, buf, payload_len);
free(buf);
return (payload_len);
}
static size_t
decrypt_tls_aead(const atf_tc_t *tc, struct tls_enable *en, uint64_t seqno,
const void *src, size_t len, void *dst, size_t avail, uint8_t *record_type)
{
const struct tls_record_layer *hdr;
size_t payload_len;
hdr = src;
ATF_REQUIRE_INTEQ(len, ntohs(hdr->tls_length) + sizeof(*hdr));
payload_len = len - (tls_header_len(en) + tls_trailer_len(en));
ATF_REQUIRE_MSG(payload_len <= avail, "payload_len (%zu) > avail (%zu)",
payload_len, avail);
if (en->tls_vminor == TLS_MINOR_VER_TWO) {
ATF_REQUIRE_INTEQ(payload_len, decrypt_tls_12_aead(tc, en,
seqno, src, len, dst, record_type));
} else {
ATF_REQUIRE_INTEQ(payload_len, decrypt_tls_13_aead(tc, en,
seqno, src, len, dst, record_type));
}
return (payload_len);
}
static size_t
decrypt_tls_record(const atf_tc_t *tc, struct tls_enable *en, uint64_t seqno,
const void *src, size_t len, void *dst, size_t avail, uint8_t *record_type)
{
if (en->cipher_algorithm == CRYPTO_AES_CBC)
return (decrypt_tls_aes_cbc_mte(tc, en, seqno, src, len, dst,
avail, record_type));
else
return (decrypt_tls_aead(tc, en, seqno, src, len, dst, avail,
record_type));
}
/*
* Encrypt a TLS record of type 'record_type' with payload 'len' bytes
* long at 'src' and store the result at 'dst'. If 'dst' doesn't have
* sufficient room ('avail'), fail the test. 'padding' is the amount
* of additional padding to include beyond any amount mandated by the
* cipher suite.
*/
static size_t
encrypt_tls_aes_cbc_mte(const atf_tc_t *tc, struct tls_enable *en,
uint8_t record_type, uint64_t seqno, const void *src, size_t len, void *dst,
size_t avail, size_t padding)
{
struct tls_record_layer *hdr;
struct tls_mac_data aad;
char *buf, *iv;
size_t hdr_len, mac_len, record_len;
u_int digest_len, i;
ATF_REQUIRE_INTEQ(0, padding % 16);
hdr = dst;
buf = dst;
debug(tc, "encrypting MTE record seqno %ju:\n", (uintmax_t)seqno);
hdr_len = tls_header_len(en);
mac_len = tls_mac_len(en);
padding += (AES_BLOCK_LEN - (len + mac_len) % AES_BLOCK_LEN);
ATF_REQUIRE_MSG(padding > 0 && padding <= 255, "invalid padding (%zu)",
padding);
record_len = hdr_len + len + mac_len + padding;
ATF_REQUIRE_MSG(record_len <= avail, "record_len (%zu) > avail (%zu): "
"hdr_len %zu, len %zu, mac_len %zu, padding %zu", record_len,
avail, hdr_len, len, mac_len, padding);
hdr->tls_type = record_type;
hdr->tls_vmajor = TLS_MAJOR_VER_ONE;
hdr->tls_vminor = en->tls_vminor;
hdr->tls_length = htons(record_len - sizeof(*hdr));
iv = (char *)(hdr + 1);
for (i = 0; i < AES_BLOCK_LEN; i++)
iv[i] = rdigit();
debug_hexdump(tc, iv, AES_BLOCK_LEN, "explicit IV");
/* Copy plaintext to ciphertext region. */
memcpy(buf + hdr_len, src, len);
/* Compute HMAC. */
tls_mte_aad(en, len, hdr, seqno, &aad);
debug_hexdump(tc, &aad, sizeof(aad), "aad");
debug_hexdump(tc, src, len, "plaintext");
ATF_REQUIRE(compute_hash(tls_EVP_MD(en), en->auth_key, en->auth_key_len,
&aad, sizeof(aad), src, len, buf + hdr_len + len, &digest_len));
ATF_REQUIRE_INTEQ(mac_len, digest_len);
/* Store padding. */
for (i = 0; i < padding; i++)
buf[hdr_len + len + mac_len + i] = padding - 1;
debug_hexdump(tc, buf + hdr_len + len, mac_len + padding,
"MAC and padding");
/* Encrypt the record. */
ATF_REQUIRE(cbc_encrypt(tls_EVP_CIPHER(en), en->cipher_key, iv,
buf + hdr_len, buf + hdr_len, len + mac_len + padding));
debug_hexdump(tc, dst, record_len, "encrypted record");
return (record_len);
}
static size_t
encrypt_tls_12_aead(const atf_tc_t *tc, struct tls_enable *en,
uint8_t record_type, uint64_t seqno, const void *src, size_t len, void *dst)
{
struct tls_record_layer *hdr;
struct tls_aead_data aad;
char nonce[12];
size_t hdr_len, mac_len, record_len;
hdr = dst;
debug(tc, "encrypting TLS 1.2 record seqno %ju:\n", (uintmax_t)seqno);
hdr_len = tls_header_len(en);
mac_len = tls_mac_len(en);
record_len = hdr_len + len + mac_len;
hdr->tls_type = record_type;
hdr->tls_vmajor = TLS_MAJOR_VER_ONE;
hdr->tls_vminor = TLS_MINOR_VER_TWO;
hdr->tls_length = htons(record_len - sizeof(*hdr));
if (en->cipher_algorithm == CRYPTO_AES_NIST_GCM_16)
memcpy(hdr + 1, &seqno, sizeof(seqno));
tls_12_aead_aad(en, len, hdr, seqno, &aad);
debug_hexdump(tc, &aad, sizeof(aad), "aad");
if (en->cipher_algorithm == CRYPTO_AES_NIST_GCM_16)
tls_12_gcm_nonce(en, hdr, nonce);
else
tls_13_nonce(en, seqno, nonce);
debug_hexdump(tc, nonce, sizeof(nonce), "nonce");
debug_hexdump(tc, src, len, "plaintext");
ATF_REQUIRE(aead_encrypt(tls_EVP_CIPHER(en), en->cipher_key, nonce,
&aad, sizeof(aad), src, (char *)dst + hdr_len, len,
(char *)dst + hdr_len + len, mac_len));
debug_hexdump(tc, dst, record_len, "encrypted record");
return (record_len);
}
static size_t
encrypt_tls_13_aead(const atf_tc_t *tc, struct tls_enable *en,
uint8_t record_type, uint64_t seqno, const void *src, size_t len, void *dst,
size_t padding)
{
struct tls_record_layer *hdr;
struct tls_aead_data_13 aad;
char nonce[12];
char *buf;
size_t hdr_len, mac_len, record_len;
hdr = dst;
debug(tc, "encrypting TLS 1.3 record seqno %ju:\n", (uintmax_t)seqno);
hdr_len = tls_header_len(en);
mac_len = tls_mac_len(en);
record_len = hdr_len + len + 1 + padding + mac_len;
hdr->tls_type = TLS_RLTYPE_APP;
hdr->tls_vmajor = TLS_MAJOR_VER_ONE;
hdr->tls_vminor = TLS_MINOR_VER_TWO;
hdr->tls_length = htons(record_len - sizeof(*hdr));
tls_13_aad(en, hdr, seqno, &aad);
debug_hexdump(tc, &aad, sizeof(aad), "aad");
tls_13_nonce(en, seqno, nonce);
debug_hexdump(tc, nonce, sizeof(nonce), "nonce");
/*
* Have to use a temporary buffer for the input so that the record
* type can be appended.
*/
buf = malloc(len + 1 + padding);
memcpy(buf, src, len);
buf[len] = record_type;
memset(buf + len + 1, 0, padding);
debug_hexdump(tc, buf, len + 1 + padding, "plaintext + type + padding");
ATF_REQUIRE(aead_encrypt(tls_EVP_CIPHER(en), en->cipher_key, nonce,
&aad, sizeof(aad), buf, (char *)dst + hdr_len, len + 1 + padding,
(char *)dst + hdr_len + len + 1 + padding, mac_len));
debug_hexdump(tc, dst, record_len, "encrypted record");
free(buf);
return (record_len);
}
static size_t
encrypt_tls_aead(const atf_tc_t *tc, struct tls_enable *en,
uint8_t record_type, uint64_t seqno, const void *src, size_t len, void *dst,
size_t avail, size_t padding)
{
size_t record_len;
record_len = tls_header_len(en) + len + padding + tls_trailer_len(en);
ATF_REQUIRE_MSG(record_len <= avail, "record_len (%zu) > avail (%zu): "
"header %zu len %zu padding %zu trailer %zu", record_len, avail,
tls_header_len(en), len, padding, tls_trailer_len(en));
if (en->tls_vminor == TLS_MINOR_VER_TWO) {
ATF_REQUIRE_INTEQ(0, padding);
ATF_REQUIRE_INTEQ(record_len, encrypt_tls_12_aead(tc, en,
record_type, seqno, src, len, dst));
} else
ATF_REQUIRE_INTEQ(record_len, encrypt_tls_13_aead(tc, en,
record_type, seqno, src, len, dst, padding));
return (record_len);
}
static size_t
encrypt_tls_record(const atf_tc_t *tc, struct tls_enable *en,
uint8_t record_type, uint64_t seqno, const void *src, size_t len, void *dst,
size_t avail, size_t padding)
{
if (en->cipher_algorithm == CRYPTO_AES_CBC)
return (encrypt_tls_aes_cbc_mte(tc, en, record_type, seqno, src,
len, dst, avail, padding));
else
return (encrypt_tls_aead(tc, en, record_type, seqno, src, len,
dst, avail, padding));
}
static void
test_ktls_transmit_app_data(const atf_tc_t *tc, struct tls_enable *en,
uint64_t seqno, size_t len)
{
struct kevent ev;
struct tls_record_layer *hdr;
char *plaintext, *decrypted, *outbuf;
size_t decrypted_len, outbuf_len, outbuf_cap, record_len, written;
ssize_t rv;
int kq, sockets[2];
uint8_t record_type;
plaintext = alloc_buffer(len);
debug_hexdump(tc, plaintext, len, "plaintext");
decrypted = malloc(len);
outbuf_cap = tls_header_len(en) + TLS_MAX_MSG_SIZE_V10_2 +
tls_trailer_len(en);
outbuf = malloc(outbuf_cap);
hdr = (struct tls_record_layer *)outbuf;
ATF_REQUIRE((kq = kqueue()) != -1);
ATF_REQUIRE_MSG(open_sockets(tc, sockets), "failed to create sockets");
ATF_REQUIRE(setsockopt(sockets[1], IPPROTO_TCP, TCP_TXTLS_ENABLE, en,
sizeof(*en)) == 0);
check_tls_mode(tc, sockets[1], TCP_TXTLS_MODE);
EV_SET(&ev, sockets[0], EVFILT_READ, EV_ADD, 0, 0, NULL);
ATF_REQUIRE(kevent(kq, &ev, 1, NULL, 0, NULL) == 0);
EV_SET(&ev, sockets[1], EVFILT_WRITE, EV_ADD, 0, 0, NULL);
ATF_REQUIRE(kevent(kq, &ev, 1, NULL, 0, NULL) == 0);
decrypted_len = 0;
outbuf_len = 0;
written = 0;
while (decrypted_len != len) {
ATF_REQUIRE(kevent(kq, NULL, 0, &ev, 1, NULL) == 1);
switch (ev.filter) {
case EVFILT_WRITE:
/* Try to write any remaining data. */
rv = write(ev.ident, plaintext + written,
len - written);
ATF_REQUIRE_MSG(rv > 0,
"failed to write to socket");
written += rv;
if (written == len) {
ev.flags = EV_DISABLE;
ATF_REQUIRE(kevent(kq, &ev, 1, NULL, 0,
NULL) == 0);
}
break;
case EVFILT_READ:
ATF_REQUIRE((ev.flags & EV_EOF) == 0);
/*
* Try to read data for the next TLS record
* into outbuf. Start by reading the header
* to determine how much additional data to
* read.
*/
if (outbuf_len < sizeof(struct tls_record_layer)) {
rv = read(ev.ident, outbuf + outbuf_len,
sizeof(struct tls_record_layer) -
outbuf_len);
ATF_REQUIRE_MSG(rv > 0,
"failed to read from socket");
outbuf_len += rv;
if (outbuf_len ==
sizeof(struct tls_record_layer)) {
debug(tc, "TLS header for seqno %ju:\n",
(uintmax_t)seqno);
debug_hexdump(tc, outbuf, outbuf_len,
NULL);
}
}
if (outbuf_len < sizeof(struct tls_record_layer))
break;
record_len = sizeof(struct tls_record_layer) +
ntohs(hdr->tls_length);
debug(tc, "record_len %zu outbuf_cap %zu\n",
record_len, outbuf_cap);
ATF_REQUIRE(record_len <= outbuf_cap);
ATF_REQUIRE(record_len > outbuf_len);
rv = read(ev.ident, outbuf + outbuf_len,
record_len - outbuf_len);
if (rv == -1 && errno == EAGAIN)
break;
ATF_REQUIRE_MSG(rv > 0,
"failed to read from socket: %s", strerror(errno));
outbuf_len += rv;
if (outbuf_len == record_len) {
decrypted_len += decrypt_tls_record(tc, en,
seqno, outbuf, outbuf_len,
decrypted + decrypted_len,
len - decrypted_len, &record_type);
ATF_REQUIRE_INTEQ(TLS_RLTYPE_APP, record_type);
seqno++;
outbuf_len = 0;
}
break;
}
}
ATF_REQUIRE_MSG(written == decrypted_len,
"read %zu decrypted bytes, but wrote %zu", decrypted_len, written);
ATF_REQUIRE(memcmp(plaintext, decrypted, len) == 0);
free(outbuf);
free(decrypted);
free(plaintext);
close_sockets(sockets);
ATF_REQUIRE(close(kq) == 0);
}
static void
ktls_send_control_message(int fd, uint8_t type, void *data, size_t len)
{
struct msghdr msg;
struct cmsghdr *cmsg;
char cbuf[CMSG_SPACE(sizeof(type))];
struct iovec iov;
memset(&msg, 0, sizeof(msg));
msg.msg_control = cbuf;
msg.msg_controllen = sizeof(cbuf);
cmsg = CMSG_FIRSTHDR(&msg);
cmsg->cmsg_level = IPPROTO_TCP;
cmsg->cmsg_type = TLS_SET_RECORD_TYPE;
cmsg->cmsg_len = CMSG_LEN(sizeof(type));
*(uint8_t *)CMSG_DATA(cmsg) = type;
iov.iov_base = data;
iov.iov_len = len;
msg.msg_iov = &iov;
msg.msg_iovlen = 1;
ATF_REQUIRE_INTEQ((ssize_t)len, sendmsg(fd, &msg, 0));
}
static void
test_ktls_transmit_control(const atf_tc_t *tc, struct tls_enable *en,
uint64_t seqno, uint8_t type, size_t len)
{
struct tls_record_layer *hdr;
char *plaintext, *decrypted, *outbuf;
size_t outbuf_cap, payload_len, record_len;
ssize_t rv;
int sockets[2];
uint8_t record_type;
ATF_REQUIRE(len <= TLS_MAX_MSG_SIZE_V10_2);
plaintext = alloc_buffer(len);
decrypted = malloc(len);
outbuf_cap = tls_header_len(en) + len + tls_trailer_len(en);
outbuf = malloc(outbuf_cap);
hdr = (struct tls_record_layer *)outbuf;
ATF_REQUIRE_MSG(open_sockets(tc, sockets), "failed to create sockets");
ATF_REQUIRE(setsockopt(sockets[1], IPPROTO_TCP, TCP_TXTLS_ENABLE, en,
sizeof(*en)) == 0);
check_tls_mode(tc, sockets[1], TCP_TXTLS_MODE);
fd_set_blocking(sockets[0]);
fd_set_blocking(sockets[1]);
ktls_send_control_message(sockets[1], type, plaintext, len);
/*
* First read the header to determine how much additional data
* to read.
*/
rv = read(sockets[0], outbuf, sizeof(struct tls_record_layer));
ATF_REQUIRE_INTEQ(sizeof(struct tls_record_layer), rv);
payload_len = ntohs(hdr->tls_length);
record_len = payload_len + sizeof(struct tls_record_layer);
ATF_REQUIRE_MSG(record_len <= outbuf_cap,
"record_len (%zu) > outbuf_cap (%zu)", record_len, outbuf_cap);
rv = read(sockets[0], outbuf + sizeof(struct tls_record_layer),
payload_len);
ATF_REQUIRE_INTEQ((ssize_t)payload_len, rv);
rv = decrypt_tls_record(tc, en, seqno, outbuf, record_len, decrypted,
len, &record_type);
ATF_REQUIRE_MSG((ssize_t)len == rv,
"read %zd decrypted bytes, but wrote %zu", rv, len);
ATF_REQUIRE_INTEQ(type, record_type);
ATF_REQUIRE(memcmp(plaintext, decrypted, len) == 0);
free(outbuf);
free(decrypted);
free(plaintext);
close_sockets(sockets);
}
static void
test_ktls_transmit_empty_fragment(const atf_tc_t *tc, struct tls_enable *en,
uint64_t seqno)
{
struct tls_record_layer *hdr;
char *outbuf;
size_t outbuf_cap, payload_len, record_len;
ssize_t rv;
int sockets[2];
uint8_t record_type;
outbuf_cap = tls_header_len(en) + tls_trailer_len(en);
outbuf = malloc(outbuf_cap);
hdr = (struct tls_record_layer *)outbuf;
ATF_REQUIRE_MSG(open_sockets(tc, sockets), "failed to create sockets");
ATF_REQUIRE(setsockopt(sockets[1], IPPROTO_TCP, TCP_TXTLS_ENABLE, en,
sizeof(*en)) == 0);
check_tls_mode(tc, sockets[1], TCP_TXTLS_MODE);
fd_set_blocking(sockets[0]);
fd_set_blocking(sockets[1]);
/*
* A write of zero bytes should send an empty fragment only for
* TLS 1.0, otherwise an error should be raised.
*/
rv = write(sockets[1], NULL, 0);
if (rv == 0) {
ATF_REQUIRE_INTEQ(CRYPTO_AES_CBC, en->cipher_algorithm);
ATF_REQUIRE_INTEQ(TLS_MINOR_VER_ZERO, en->tls_vminor);
} else {
ATF_REQUIRE_INTEQ(-1, rv);
ATF_REQUIRE_ERRNO(EINVAL, true);
goto out;
}
/*
* First read the header to determine how much additional data
* to read.
*/
rv = read(sockets[0], outbuf, sizeof(struct tls_record_layer));
ATF_REQUIRE_INTEQ(sizeof(struct tls_record_layer), rv);
payload_len = ntohs(hdr->tls_length);
record_len = payload_len + sizeof(struct tls_record_layer);
ATF_REQUIRE_MSG(record_len <= outbuf_cap,
"record_len (%zu) > outbuf_cap (%zu)", record_len, outbuf_cap);
rv = read(sockets[0], outbuf + sizeof(struct tls_record_layer),
payload_len);
ATF_REQUIRE_INTEQ((ssize_t)payload_len, rv);
rv = decrypt_tls_record(tc, en, seqno, outbuf, record_len, NULL, 0,
&record_type);
ATF_REQUIRE_MSG(rv == 0,
"read %zd decrypted bytes for an empty fragment", rv);
ATF_REQUIRE_INTEQ(TLS_RLTYPE_APP, record_type);
out:
free(outbuf);
close_sockets(sockets);
}
static size_t
ktls_receive_tls_record(struct tls_enable *en, int fd, uint8_t record_type,
void *data, size_t len)
{
struct msghdr msg;
struct cmsghdr *cmsg;
struct tls_get_record *tgr;
char cbuf[CMSG_SPACE(sizeof(*tgr))];
struct iovec iov;
ssize_t rv;
memset(&msg, 0, sizeof(msg));
msg.msg_control = cbuf;
msg.msg_controllen = sizeof(cbuf);
iov.iov_base = data;
iov.iov_len = len;
msg.msg_iov = &iov;
msg.msg_iovlen = 1;
ATF_REQUIRE((rv = recvmsg(fd, &msg, 0)) > 0);
ATF_REQUIRE((msg.msg_flags & (MSG_EOR | MSG_CTRUNC)) == MSG_EOR);
cmsg = CMSG_FIRSTHDR(&msg);
ATF_REQUIRE(cmsg != NULL);
ATF_REQUIRE_INTEQ(IPPROTO_TCP, cmsg->cmsg_level);
ATF_REQUIRE_INTEQ(TLS_GET_RECORD, cmsg->cmsg_type);
ATF_REQUIRE_INTEQ(CMSG_LEN(sizeof(*tgr)), cmsg->cmsg_len);
tgr = (struct tls_get_record *)CMSG_DATA(cmsg);
ATF_REQUIRE_INTEQ(record_type, tgr->tls_type);
ATF_REQUIRE_INTEQ(en->tls_vmajor, tgr->tls_vmajor);
/* XXX: Not sure if this is what OpenSSL expects? */
if (en->tls_vminor == TLS_MINOR_VER_THREE)
ATF_REQUIRE_INTEQ(TLS_MINOR_VER_TWO, tgr->tls_vminor);
else
ATF_REQUIRE_INTEQ(en->tls_vminor, tgr->tls_vminor);
ATF_REQUIRE_INTEQ(htons(rv), tgr->tls_length);
return (rv);
}
static void
test_ktls_receive_app_data(const atf_tc_t *tc, struct tls_enable *en,
uint64_t seqno, size_t len, size_t padding)
{
struct kevent ev;
char *plaintext, *received, *outbuf;
size_t outbuf_cap, outbuf_len, outbuf_sent, received_len, todo, written;
ssize_t rv;
int kq, sockets[2];
plaintext = alloc_buffer(len);
received = malloc(len);
outbuf_cap = tls_header_len(en) + TLS_MAX_MSG_SIZE_V10_2 +
tls_trailer_len(en);
outbuf = malloc(outbuf_cap);
ATF_REQUIRE((kq = kqueue()) != -1);
ATF_REQUIRE_MSG(open_sockets(tc, sockets), "failed to create sockets");
ATF_REQUIRE(setsockopt(sockets[0], IPPROTO_TCP, TCP_RXTLS_ENABLE, en,
sizeof(*en)) == 0);
check_tls_mode(tc, sockets[0], TCP_RXTLS_MODE);
EV_SET(&ev, sockets[0], EVFILT_READ, EV_ADD, 0, 0, NULL);
ATF_REQUIRE(kevent(kq, &ev, 1, NULL, 0, NULL) == 0);
EV_SET(&ev, sockets[1], EVFILT_WRITE, EV_ADD, 0, 0, NULL);
ATF_REQUIRE(kevent(kq, &ev, 1, NULL, 0, NULL) == 0);
received_len = 0;
outbuf_len = 0;
written = 0;
while (received_len != len) {
ATF_REQUIRE(kevent(kq, NULL, 0, &ev, 1, NULL) == 1);
switch (ev.filter) {
case EVFILT_WRITE:
/*
* Compose the next TLS record to send.
*/
if (outbuf_len == 0) {
ATF_REQUIRE(written < len);
todo = len - written;
if (todo > TLS_MAX_MSG_SIZE_V10_2 - padding)
todo = TLS_MAX_MSG_SIZE_V10_2 - padding;
outbuf_len = encrypt_tls_record(tc, en,
TLS_RLTYPE_APP, seqno, plaintext + written,
todo, outbuf, outbuf_cap, padding);
outbuf_sent = 0;
written += todo;
seqno++;
}
/*
* Try to write the remainder of the current
* TLS record.
*/
rv = write(ev.ident, outbuf + outbuf_sent,
outbuf_len - outbuf_sent);
ATF_REQUIRE_MSG(rv > 0,
"failed to write to socket: %s", strerror(errno));
outbuf_sent += rv;
if (outbuf_sent == outbuf_len) {
outbuf_len = 0;
if (written == len) {
ev.flags = EV_DISABLE;
ATF_REQUIRE(kevent(kq, &ev, 1, NULL, 0,
NULL) == 0);
}
}
break;
case EVFILT_READ:
ATF_REQUIRE((ev.flags & EV_EOF) == 0);
rv = ktls_receive_tls_record(en, ev.ident,
TLS_RLTYPE_APP, received + received_len,
len - received_len);
received_len += rv;
break;
}
}
ATF_REQUIRE_MSG(written == received_len,
"read %zu decrypted bytes, but wrote %zu", received_len, written);
ATF_REQUIRE(memcmp(plaintext, received, len) == 0);
free(outbuf);
free(received);
free(plaintext);
close_sockets(sockets);
ATF_REQUIRE(close(kq) == 0);
}
static void
ktls_receive_tls_error(int fd, int expected_error)
{
struct msghdr msg;
struct tls_get_record *tgr;
char cbuf[CMSG_SPACE(sizeof(*tgr))];
char buf[64];
struct iovec iov;
memset(&msg, 0, sizeof(msg));
msg.msg_control = cbuf;
msg.msg_controllen = sizeof(cbuf);
iov.iov_base = buf;
iov.iov_len = sizeof(buf);
msg.msg_iov = &iov;
msg.msg_iovlen = 1;
ATF_REQUIRE(recvmsg(fd, &msg, 0) == -1);
if (expected_error != 0)
ATF_REQUIRE_ERRNO(expected_error, true);
}
static void
test_ktls_receive_corrupted_record(const atf_tc_t *tc, struct tls_enable *en,
uint64_t seqno, size_t len, ssize_t offset)
{
char *plaintext, *outbuf;
size_t outbuf_cap, outbuf_len;
ssize_t rv;
int sockets[2];
ATF_REQUIRE(len <= TLS_MAX_MSG_SIZE_V10_2);
plaintext = alloc_buffer(len);
outbuf_cap = tls_header_len(en) + len + tls_trailer_len(en);
outbuf = malloc(outbuf_cap);
ATF_REQUIRE_MSG(open_sockets(tc, sockets), "failed to create sockets");
ATF_REQUIRE(setsockopt(sockets[0], IPPROTO_TCP, TCP_RXTLS_ENABLE, en,
sizeof(*en)) == 0);
check_tls_mode(tc, sockets[0], TCP_RXTLS_MODE);
fd_set_blocking(sockets[0]);
fd_set_blocking(sockets[1]);
outbuf_len = encrypt_tls_record(tc, en, TLS_RLTYPE_APP, seqno,
plaintext, len, outbuf, outbuf_cap, 0);
/* A negative offset is an offset from the end. */
if (offset < 0)
offset += outbuf_len;
outbuf[offset] ^= 0x01;
rv = write(sockets[1], outbuf, outbuf_len);
ATF_REQUIRE_INTEQ((ssize_t)outbuf_len, rv);
ktls_receive_tls_error(sockets[0], EBADMSG);
free(outbuf);
free(plaintext);
close_sockets_ignore_errors(sockets);
}
static void
test_ktls_receive_corrupted_iv(const atf_tc_t *tc, struct tls_enable *en,
uint64_t seqno, size_t len)
{
ATF_REQUIRE(tls_header_len(en) > sizeof(struct tls_record_layer));
/* Corrupt the first byte of the explicit IV after the header. */
test_ktls_receive_corrupted_record(tc, en, seqno, len,
sizeof(struct tls_record_layer));
}
static void
test_ktls_receive_corrupted_data(const atf_tc_t *tc, struct tls_enable *en,
uint64_t seqno, size_t len)
{
ATF_REQUIRE(len > 0);
/* Corrupt the first ciphertext byte after the header. */
test_ktls_receive_corrupted_record(tc, en, seqno, len,
tls_header_len(en));
}
static void
test_ktls_receive_corrupted_mac(const atf_tc_t *tc, struct tls_enable *en,
uint64_t seqno, size_t len)
{
size_t offset;
/* Corrupt the first byte of the MAC. */
if (en->cipher_algorithm == CRYPTO_AES_CBC)
offset = tls_header_len(en) + len;
else
offset = -tls_mac_len(en);
test_ktls_receive_corrupted_record(tc, en, seqno, len, offset);
}
static void
test_ktls_receive_corrupted_padding(const atf_tc_t *tc, struct tls_enable *en,
uint64_t seqno, size_t len)
{
ATF_REQUIRE_INTEQ(CRYPTO_AES_CBC, en->cipher_algorithm);
/* Corrupt the last byte of the padding. */
test_ktls_receive_corrupted_record(tc, en, seqno, len, -1);
}
static void
test_ktls_receive_truncated_record(const atf_tc_t *tc, struct tls_enable *en,
uint64_t seqno, size_t len)
{
char *plaintext, *outbuf;
size_t outbuf_cap, outbuf_len;
ssize_t rv;
int sockets[2];
ATF_REQUIRE(len <= TLS_MAX_MSG_SIZE_V10_2);
plaintext = alloc_buffer(len);
outbuf_cap = tls_header_len(en) + len + tls_trailer_len(en);
outbuf = malloc(outbuf_cap);
ATF_REQUIRE_MSG(open_sockets(tc, sockets), "failed to create sockets");
ATF_REQUIRE(setsockopt(sockets[0], IPPROTO_TCP, TCP_RXTLS_ENABLE, en,
sizeof(*en)) == 0);
check_tls_mode(tc, sockets[0], TCP_RXTLS_MODE);
fd_set_blocking(sockets[0]);
fd_set_blocking(sockets[1]);
outbuf_len = encrypt_tls_record(tc, en, TLS_RLTYPE_APP, seqno,
plaintext, len, outbuf, outbuf_cap, 0);
rv = write(sockets[1], outbuf, outbuf_len / 2);
ATF_REQUIRE_INTEQ((ssize_t)(outbuf_len / 2), rv);
ATF_REQUIRE(shutdown(sockets[1], SHUT_WR) == 0);
ktls_receive_tls_error(sockets[0], EMSGSIZE);
free(outbuf);
free(plaintext);
close_sockets_ignore_errors(sockets);
}
static void
test_ktls_receive_bad_major(const atf_tc_t *tc, struct tls_enable *en,
uint64_t seqno, size_t len)
{
struct tls_record_layer *hdr;
char *plaintext, *outbuf;
size_t outbuf_cap, outbuf_len;
ssize_t rv;
int sockets[2];
ATF_REQUIRE(len <= TLS_MAX_MSG_SIZE_V10_2);
plaintext = alloc_buffer(len);
outbuf_cap = tls_header_len(en) + len + tls_trailer_len(en);
outbuf = malloc(outbuf_cap);
ATF_REQUIRE_MSG(open_sockets(tc, sockets), "failed to create sockets");
ATF_REQUIRE(setsockopt(sockets[0], IPPROTO_TCP, TCP_RXTLS_ENABLE, en,
sizeof(*en)) == 0);
check_tls_mode(tc, sockets[0], TCP_RXTLS_MODE);
fd_set_blocking(sockets[0]);
fd_set_blocking(sockets[1]);
outbuf_len = encrypt_tls_record(tc, en, TLS_RLTYPE_APP, seqno,
plaintext, len, outbuf, outbuf_cap, 0);
hdr = (void *)outbuf;
hdr->tls_vmajor++;
rv = write(sockets[1], outbuf, outbuf_len);
ATF_REQUIRE_INTEQ((ssize_t)outbuf_len, rv);
ktls_receive_tls_error(sockets[0], EINVAL);
free(outbuf);
free(plaintext);
close_sockets_ignore_errors(sockets);
}
static void
test_ktls_receive_bad_minor(const atf_tc_t *tc, struct tls_enable *en,
uint64_t seqno, size_t len)
{
struct tls_record_layer *hdr;
char *plaintext, *outbuf;
size_t outbuf_cap, outbuf_len;
ssize_t rv;
int sockets[2];
ATF_REQUIRE(len <= TLS_MAX_MSG_SIZE_V10_2);
plaintext = alloc_buffer(len);
outbuf_cap = tls_header_len(en) + len + tls_trailer_len(en);
outbuf = malloc(outbuf_cap);
ATF_REQUIRE_MSG(open_sockets(tc, sockets), "failed to create sockets");
ATF_REQUIRE(setsockopt(sockets[0], IPPROTO_TCP, TCP_RXTLS_ENABLE, en,
sizeof(*en)) == 0);
check_tls_mode(tc, sockets[0], TCP_RXTLS_MODE);
fd_set_blocking(sockets[0]);
fd_set_blocking(sockets[1]);
outbuf_len = encrypt_tls_record(tc, en, TLS_RLTYPE_APP, seqno,
plaintext, len, outbuf, outbuf_cap, 0);
hdr = (void *)outbuf;
hdr->tls_vminor++;
rv = write(sockets[1], outbuf, outbuf_len);
ATF_REQUIRE_INTEQ((ssize_t)outbuf_len, rv);
ktls_receive_tls_error(sockets[0], EINVAL);
free(outbuf);
free(plaintext);
close_sockets_ignore_errors(sockets);
}
static void
test_ktls_receive_bad_type(const atf_tc_t *tc, struct tls_enable *en,
uint64_t seqno, size_t len)
{
struct tls_record_layer *hdr;
char *plaintext, *outbuf;
size_t outbuf_cap, outbuf_len;
ssize_t rv;
int sockets[2];
ATF_REQUIRE(len <= TLS_MAX_MSG_SIZE_V10_2);
ATF_REQUIRE_INTEQ(TLS_MINOR_VER_THREE, en->tls_vminor);
plaintext = alloc_buffer(len);
outbuf_cap = tls_header_len(en) + len + tls_trailer_len(en);
outbuf = malloc(outbuf_cap);
ATF_REQUIRE_MSG(open_sockets(tc, sockets), "failed to create sockets");
ATF_REQUIRE(setsockopt(sockets[0], IPPROTO_TCP, TCP_RXTLS_ENABLE, en,
sizeof(*en)) == 0);
check_tls_mode(tc, sockets[0], TCP_RXTLS_MODE);
fd_set_blocking(sockets[0]);
fd_set_blocking(sockets[1]);
outbuf_len = encrypt_tls_record(tc, en, 0x21 /* Alert */, seqno,
plaintext, len, outbuf, outbuf_cap, 0);
hdr = (void *)outbuf;
hdr->tls_type = TLS_RLTYPE_APP + 1;
rv = write(sockets[1], outbuf, outbuf_len);
ATF_REQUIRE_INTEQ((ssize_t)outbuf_len, rv);
ktls_receive_tls_error(sockets[0], EINVAL);
free(outbuf);
free(plaintext);
close_sockets_ignore_errors(sockets);
}
static void
test_ktls_receive_bad_size(const atf_tc_t *tc, struct tls_enable *en,
uint64_t seqno, size_t len)
{
struct tls_record_layer *hdr;
char *outbuf;
size_t outbuf_len;
ssize_t rv;
int sockets[2];
outbuf_len = sizeof(*hdr) + len;
outbuf = calloc(1, outbuf_len);
ATF_REQUIRE_MSG(open_sockets(tc, sockets), "failed to create sockets");
ATF_REQUIRE(setsockopt(sockets[0], IPPROTO_TCP, TCP_RXTLS_ENABLE, en,
sizeof(*en)) == 0);
check_tls_mode(tc, sockets[0], TCP_RXTLS_MODE);
fd_set_blocking(sockets[0]);
fd_set_blocking(sockets[1]);
hdr = (void *)outbuf;
hdr->tls_vmajor = en->tls_vmajor;
if (en->tls_vminor == TLS_MINOR_VER_THREE)
hdr->tls_vminor = TLS_MINOR_VER_TWO;
else
hdr->tls_vminor = en->tls_vminor;
hdr->tls_type = TLS_RLTYPE_APP;
hdr->tls_length = htons(len);
rv = write(sockets[1], outbuf, outbuf_len);
ATF_REQUIRE_INTEQ((ssize_t)outbuf_len, rv);
/*
* The other end may notice the error and drop the connection
* before this executes resulting in shutdown() failing with
* either ENOTCONN or ECONNRESET. Ignore this error if it
* occurs.
*/
if (shutdown(sockets[1], SHUT_WR) != 0) {
ATF_REQUIRE_MSG(errno == ENOTCONN || errno == ECONNRESET,
"shutdown() failed: %s", strerror(errno));
}
ktls_receive_tls_error(sockets[0], EMSGSIZE);
free(outbuf);
close_sockets_ignore_errors(sockets);
}
#define TLS_10_TESTS(M) \
M(aes128_cbc_1_0_sha1, CRYPTO_AES_CBC, 128 / 8, \
CRYPTO_SHA1_HMAC, TLS_MINOR_VER_ZERO) \
M(aes256_cbc_1_0_sha1, CRYPTO_AES_CBC, 256 / 8, \
CRYPTO_SHA1_HMAC, TLS_MINOR_VER_ZERO)
#define TLS_13_TESTS(M) \
M(aes128_gcm_1_3, CRYPTO_AES_NIST_GCM_16, 128 / 8, 0, \
TLS_MINOR_VER_THREE) \
M(aes256_gcm_1_3, CRYPTO_AES_NIST_GCM_16, 256 / 8, 0, \
TLS_MINOR_VER_THREE) \
M(chacha20_poly1305_1_3, CRYPTO_CHACHA20_POLY1305, 256 / 8, 0, \
TLS_MINOR_VER_THREE)
#define AES_CBC_NONZERO_TESTS(M) \
M(aes128_cbc_1_1_sha1, CRYPTO_AES_CBC, 128 / 8, \
CRYPTO_SHA1_HMAC, TLS_MINOR_VER_ONE) \
M(aes256_cbc_1_1_sha1, CRYPTO_AES_CBC, 256 / 8, \
CRYPTO_SHA1_HMAC, TLS_MINOR_VER_ONE) \
M(aes128_cbc_1_2_sha1, CRYPTO_AES_CBC, 128 / 8, \
CRYPTO_SHA1_HMAC, TLS_MINOR_VER_TWO) \
M(aes256_cbc_1_2_sha1, CRYPTO_AES_CBC, 256 / 8, \
CRYPTO_SHA1_HMAC, TLS_MINOR_VER_TWO) \
M(aes128_cbc_1_2_sha256, CRYPTO_AES_CBC, 128 / 8, \
CRYPTO_SHA2_256_HMAC, TLS_MINOR_VER_TWO) \
M(aes256_cbc_1_2_sha256, CRYPTO_AES_CBC, 256 / 8, \
CRYPTO_SHA2_256_HMAC, TLS_MINOR_VER_TWO) \
M(aes128_cbc_1_2_sha384, CRYPTO_AES_CBC, 128 / 8, \
CRYPTO_SHA2_384_HMAC, TLS_MINOR_VER_TWO) \
M(aes256_cbc_1_2_sha384, CRYPTO_AES_CBC, 256 / 8, \
CRYPTO_SHA2_384_HMAC, TLS_MINOR_VER_TWO) \
#define AES_CBC_TESTS(M) \
TLS_10_TESTS(M) \
AES_CBC_NONZERO_TESTS(M)
#define AES_GCM_12_TESTS(M) \
M(aes128_gcm_1_2, CRYPTO_AES_NIST_GCM_16, 128 / 8, 0, \
TLS_MINOR_VER_TWO) \
M(aes256_gcm_1_2, CRYPTO_AES_NIST_GCM_16, 256 / 8, 0, \
TLS_MINOR_VER_TWO)
#define AES_GCM_TESTS(M) \
AES_GCM_12_TESTS(M) \
M(aes128_gcm_1_3, CRYPTO_AES_NIST_GCM_16, 128 / 8, 0, \
TLS_MINOR_VER_THREE) \
M(aes256_gcm_1_3, CRYPTO_AES_NIST_GCM_16, 256 / 8, 0, \
TLS_MINOR_VER_THREE)
#define CHACHA20_TESTS(M) \
M(chacha20_poly1305_1_2, CRYPTO_CHACHA20_POLY1305, 256 / 8, 0, \
TLS_MINOR_VER_TWO) \
M(chacha20_poly1305_1_3, CRYPTO_CHACHA20_POLY1305, 256 / 8, 0, \
TLS_MINOR_VER_THREE)
#define GEN_TRANSMIT_APP_DATA_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, name, len) \
ATF_TC_WITHOUT_HEAD(ktls_transmit_##cipher_name##_##name); \
ATF_TC_BODY(ktls_transmit_##cipher_name##_##name, tc) \
{ \
struct tls_enable en; \
uint64_t seqno; \
\
ATF_REQUIRE_KTLS(); \
seqno = random(); \
build_tls_enable(tc, cipher_alg, key_size, auth_alg, minor, \
seqno, &en); \
test_ktls_transmit_app_data(tc, &en, seqno, len); \
free_tls_enable(&en); \
}
#define ADD_TRANSMIT_APP_DATA_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, name) \
ATF_TP_ADD_TC(tp, ktls_transmit_##cipher_name##_##name);
#define GEN_TRANSMIT_CONTROL_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, name, type, len) \
ATF_TC_WITHOUT_HEAD(ktls_transmit_##cipher_name##_##name); \
ATF_TC_BODY(ktls_transmit_##cipher_name##_##name, tc) \
{ \
struct tls_enable en; \
uint64_t seqno; \
\
ATF_REQUIRE_KTLS(); \
seqno = random(); \
build_tls_enable(tc, cipher_alg, key_size, auth_alg, minor, \
seqno, &en); \
test_ktls_transmit_control(tc, &en, seqno, type, len); \
free_tls_enable(&en); \
}
#define ADD_TRANSMIT_CONTROL_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, name) \
ATF_TP_ADD_TC(tp, ktls_transmit_##cipher_name##_##name);
#define GEN_TRANSMIT_EMPTY_FRAGMENT_TEST(cipher_name, cipher_alg, \
key_size, auth_alg, minor) \
ATF_TC_WITHOUT_HEAD(ktls_transmit_##cipher_name##_empty_fragment); \
ATF_TC_BODY(ktls_transmit_##cipher_name##_empty_fragment, tc) \
{ \
struct tls_enable en; \
uint64_t seqno; \
\
ATF_REQUIRE_KTLS(); \
seqno = random(); \
build_tls_enable(tc, cipher_alg, key_size, auth_alg, minor, \
seqno, &en); \
test_ktls_transmit_empty_fragment(tc, &en, seqno); \
free_tls_enable(&en); \
}
#define ADD_TRANSMIT_EMPTY_FRAGMENT_TEST(cipher_name, cipher_alg, \
key_size, auth_alg, minor) \
ATF_TP_ADD_TC(tp, ktls_transmit_##cipher_name##_empty_fragment);
#define GEN_TRANSMIT_TESTS(cipher_name, cipher_alg, key_size, auth_alg, \
minor) \
GEN_TRANSMIT_APP_DATA_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, short, 64) \
GEN_TRANSMIT_APP_DATA_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, long, 64 * 1024) \
GEN_TRANSMIT_CONTROL_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, control, 0x21 /* Alert */, 32)
#define ADD_TRANSMIT_TESTS(cipher_name, cipher_alg, key_size, auth_alg, \
minor) \
ADD_TRANSMIT_APP_DATA_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, short) \
ADD_TRANSMIT_APP_DATA_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, long) \
ADD_TRANSMIT_CONTROL_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, control)
/*
* For each supported cipher suite, run three transmit tests:
*
* - a short test which sends 64 bytes of application data (likely as
* a single TLS record)
*
* - a long test which sends 64KB of application data (split across
* multiple TLS records)
*
* - a control test which sends a single record with a specific
* content type via sendmsg()
*/
AES_CBC_TESTS(GEN_TRANSMIT_TESTS);
AES_GCM_TESTS(GEN_TRANSMIT_TESTS);
CHACHA20_TESTS(GEN_TRANSMIT_TESTS);
#define GEN_TRANSMIT_PADDING_TESTS(cipher_name, cipher_alg, key_size, \
auth_alg, minor) \
GEN_TRANSMIT_CONTROL_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_1, 0x21 /* Alert */, 1) \
GEN_TRANSMIT_CONTROL_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_2, 0x21 /* Alert */, 2) \
GEN_TRANSMIT_CONTROL_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_3, 0x21 /* Alert */, 3) \
GEN_TRANSMIT_CONTROL_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_4, 0x21 /* Alert */, 4) \
GEN_TRANSMIT_CONTROL_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_5, 0x21 /* Alert */, 5) \
GEN_TRANSMIT_CONTROL_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_6, 0x21 /* Alert */, 6) \
GEN_TRANSMIT_CONTROL_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_7, 0x21 /* Alert */, 7) \
GEN_TRANSMIT_CONTROL_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_8, 0x21 /* Alert */, 8) \
GEN_TRANSMIT_CONTROL_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_9, 0x21 /* Alert */, 9) \
GEN_TRANSMIT_CONTROL_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_10, 0x21 /* Alert */, 10) \
GEN_TRANSMIT_CONTROL_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_11, 0x21 /* Alert */, 11) \
GEN_TRANSMIT_CONTROL_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_12, 0x21 /* Alert */, 12) \
GEN_TRANSMIT_CONTROL_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_13, 0x21 /* Alert */, 13) \
GEN_TRANSMIT_CONTROL_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_14, 0x21 /* Alert */, 14) \
GEN_TRANSMIT_CONTROL_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_15, 0x21 /* Alert */, 15) \
GEN_TRANSMIT_CONTROL_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_16, 0x21 /* Alert */, 16)
#define ADD_TRANSMIT_PADDING_TESTS(cipher_name, cipher_alg, key_size, \
auth_alg, minor) \
ADD_TRANSMIT_CONTROL_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_1) \
ADD_TRANSMIT_CONTROL_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_2) \
ADD_TRANSMIT_CONTROL_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_3) \
ADD_TRANSMIT_CONTROL_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_4) \
ADD_TRANSMIT_CONTROL_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_5) \
ADD_TRANSMIT_CONTROL_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_6) \
ADD_TRANSMIT_CONTROL_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_7) \
ADD_TRANSMIT_CONTROL_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_8) \
ADD_TRANSMIT_CONTROL_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_9) \
ADD_TRANSMIT_CONTROL_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_10) \
ADD_TRANSMIT_CONTROL_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_11) \
ADD_TRANSMIT_CONTROL_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_12) \
ADD_TRANSMIT_CONTROL_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_13) \
ADD_TRANSMIT_CONTROL_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_14) \
ADD_TRANSMIT_CONTROL_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_15) \
ADD_TRANSMIT_CONTROL_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_16)
/*
* For AES-CBC MTE cipher suites using padding, add tests of messages
* with each possible padding size. Note that the padding_<N> tests
* do not necessarily test <N> bytes of padding as the padding is a
* function of the cipher suite's MAC length. However, cycling
* through all of the payload sizes from 1 to 16 should exercise all
* of the possible padding lengths for each suite.
*/
AES_CBC_TESTS(GEN_TRANSMIT_PADDING_TESTS);
/*
* Test "empty fragments" which are TLS records with no payload that
* OpenSSL can send for TLS 1.0 connections.
*/
AES_CBC_TESTS(GEN_TRANSMIT_EMPTY_FRAGMENT_TEST);
AES_GCM_TESTS(GEN_TRANSMIT_EMPTY_FRAGMENT_TEST);
CHACHA20_TESTS(GEN_TRANSMIT_EMPTY_FRAGMENT_TEST);
static void
test_ktls_invalid_transmit_cipher_suite(const atf_tc_t *tc,
struct tls_enable *en)
{
int sockets[2];
ATF_REQUIRE_MSG(open_sockets(tc, sockets), "failed to create sockets");
ATF_REQUIRE_ERRNO(EINVAL, setsockopt(sockets[1], IPPROTO_TCP,
TCP_TXTLS_ENABLE, en, sizeof(*en)) == -1);
close_sockets(sockets);
}
#define GEN_INVALID_TRANSMIT_TEST(name, cipher_alg, key_size, auth_alg, \
minor) \
ATF_TC_WITHOUT_HEAD(ktls_transmit_invalid_##name); \
ATF_TC_BODY(ktls_transmit_invalid_##name, tc) \
{ \
struct tls_enable en; \
uint64_t seqno; \
\
ATF_REQUIRE_KTLS(); \
seqno = random(); \
build_tls_enable(tc, cipher_alg, key_size, auth_alg, minor, \
seqno, &en); \
test_ktls_invalid_transmit_cipher_suite(tc, &en); \
free_tls_enable(&en); \
}
#define ADD_INVALID_TRANSMIT_TEST(name, cipher_alg, key_size, auth_alg, \
minor) \
ATF_TP_ADD_TC(tp, ktls_transmit_invalid_##name);
#define INVALID_CIPHER_SUITES(M) \
M(aes128_cbc_1_0_sha256, CRYPTO_AES_CBC, 128 / 8, \
CRYPTO_SHA2_256_HMAC, TLS_MINOR_VER_ZERO) \
M(aes128_cbc_1_0_sha384, CRYPTO_AES_CBC, 128 / 8, \
CRYPTO_SHA2_384_HMAC, TLS_MINOR_VER_ZERO) \
M(aes128_gcm_1_0, CRYPTO_AES_NIST_GCM_16, 128 / 8, 0, \
TLS_MINOR_VER_ZERO) \
M(chacha20_poly1305_1_0, CRYPTO_CHACHA20_POLY1305, 256 / 8, 0, \
TLS_MINOR_VER_ZERO) \
M(aes128_cbc_1_1_sha256, CRYPTO_AES_CBC, 128 / 8, \
CRYPTO_SHA2_256_HMAC, TLS_MINOR_VER_ONE) \
M(aes128_cbc_1_1_sha384, CRYPTO_AES_CBC, 128 / 8, \
CRYPTO_SHA2_384_HMAC, TLS_MINOR_VER_ONE) \
M(aes128_gcm_1_1, CRYPTO_AES_NIST_GCM_16, 128 / 8, 0, \
TLS_MINOR_VER_ONE) \
M(chacha20_poly1305_1_1, CRYPTO_CHACHA20_POLY1305, 256 / 8, 0, \
TLS_MINOR_VER_ONE) \
M(aes128_cbc_1_3_sha1, CRYPTO_AES_CBC, 128 / 8, \
CRYPTO_SHA1_HMAC, TLS_MINOR_VER_THREE) \
M(aes128_cbc_1_3_sha256, CRYPTO_AES_CBC, 128 / 8, \
CRYPTO_SHA2_256_HMAC, TLS_MINOR_VER_THREE) \
M(aes128_cbc_1_3_sha384, CRYPTO_AES_CBC, 128 / 8, \
CRYPTO_SHA2_384_HMAC, TLS_MINOR_VER_THREE)
/*
* Ensure that invalid cipher suites are rejected for transmit.
*/
INVALID_CIPHER_SUITES(GEN_INVALID_TRANSMIT_TEST);
#define GEN_RECEIVE_APP_DATA_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, name, len, padding) \
ATF_TC_WITHOUT_HEAD(ktls_receive_##cipher_name##_##name); \
ATF_TC_BODY(ktls_receive_##cipher_name##_##name, tc) \
{ \
struct tls_enable en; \
uint64_t seqno; \
\
ATF_REQUIRE_KTLS(); \
seqno = random(); \
build_tls_enable(tc, cipher_alg, key_size, auth_alg, minor, \
seqno, &en); \
test_ktls_receive_app_data(tc, &en, seqno, len, padding); \
free_tls_enable(&en); \
}
#define ADD_RECEIVE_APP_DATA_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, name) \
ATF_TP_ADD_TC(tp, ktls_receive_##cipher_name##_##name);
#define GEN_RECEIVE_BAD_DATA_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, len) \
ATF_TC_WITHOUT_HEAD(ktls_receive_##cipher_name##_bad_data); \
ATF_TC_BODY(ktls_receive_##cipher_name##_bad_data, tc) \
{ \
struct tls_enable en; \
uint64_t seqno; \
\
ATF_REQUIRE_KTLS(); \
seqno = random(); \
build_tls_enable(tc, cipher_alg, key_size, auth_alg, minor, \
seqno, &en); \
test_ktls_receive_corrupted_data(tc, &en, seqno, len); \
free_tls_enable(&en); \
}
#define ADD_RECEIVE_BAD_DATA_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor) \
ATF_TP_ADD_TC(tp, ktls_receive_##cipher_name##_bad_data);
#define GEN_RECEIVE_BAD_MAC_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, len) \
ATF_TC_WITHOUT_HEAD(ktls_receive_##cipher_name##_bad_mac); \
ATF_TC_BODY(ktls_receive_##cipher_name##_bad_mac, tc) \
{ \
struct tls_enable en; \
uint64_t seqno; \
\
ATF_REQUIRE_KTLS(); \
seqno = random(); \
build_tls_enable(tc, cipher_alg, key_size, auth_alg, minor, \
seqno, &en); \
test_ktls_receive_corrupted_mac(tc, &en, seqno, len); \
free_tls_enable(&en); \
}
#define ADD_RECEIVE_BAD_MAC_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor) \
ATF_TP_ADD_TC(tp, ktls_receive_##cipher_name##_bad_mac);
#define GEN_RECEIVE_TRUNCATED_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, len) \
ATF_TC_WITHOUT_HEAD(ktls_receive_##cipher_name##_truncated_record); \
ATF_TC_BODY(ktls_receive_##cipher_name##_truncated_record, tc) \
{ \
struct tls_enable en; \
uint64_t seqno; \
\
ATF_REQUIRE_KTLS(); \
seqno = random(); \
build_tls_enable(tc, cipher_alg, key_size, auth_alg, minor, \
seqno, &en); \
test_ktls_receive_truncated_record(tc, &en, seqno, len); \
free_tls_enable(&en); \
}
#define ADD_RECEIVE_TRUNCATED_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor) \
ATF_TP_ADD_TC(tp, ktls_receive_##cipher_name##_truncated_record);
#define GEN_RECEIVE_BAD_MAJOR_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, len) \
ATF_TC_WITHOUT_HEAD(ktls_receive_##cipher_name##_bad_major); \
ATF_TC_BODY(ktls_receive_##cipher_name##_bad_major, tc) \
{ \
struct tls_enable en; \
uint64_t seqno; \
\
ATF_REQUIRE_KTLS(); \
seqno = random(); \
build_tls_enable(tc, cipher_alg, key_size, auth_alg, minor, \
seqno, &en); \
test_ktls_receive_bad_major(tc, &en, seqno, len); \
free_tls_enable(&en); \
}
#define ADD_RECEIVE_BAD_MAJOR_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor) \
ATF_TP_ADD_TC(tp, ktls_receive_##cipher_name##_bad_major);
#define GEN_RECEIVE_BAD_MINOR_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, len) \
ATF_TC_WITHOUT_HEAD(ktls_receive_##cipher_name##_bad_minor); \
ATF_TC_BODY(ktls_receive_##cipher_name##_bad_minor, tc) \
{ \
struct tls_enable en; \
uint64_t seqno; \
\
ATF_REQUIRE_KTLS(); \
seqno = random(); \
build_tls_enable(tc, cipher_alg, key_size, auth_alg, minor, \
seqno, &en); \
test_ktls_receive_bad_minor(tc, &en, seqno, len); \
free_tls_enable(&en); \
}
#define ADD_RECEIVE_BAD_MINOR_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor) \
ATF_TP_ADD_TC(tp, ktls_receive_##cipher_name##_bad_minor);
#define GEN_RECEIVE_BAD_SIZE_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, name, len) \
ATF_TC_WITHOUT_HEAD(ktls_receive_##cipher_name##_##name); \
ATF_TC_BODY(ktls_receive_##cipher_name##_##name, tc) \
{ \
struct tls_enable en; \
uint64_t seqno; \
\
ATF_REQUIRE_KTLS(); \
seqno = random(); \
build_tls_enable(tc, cipher_alg, key_size, auth_alg, minor, \
seqno, &en); \
test_ktls_receive_bad_size(tc, &en, seqno, (len)); \
free_tls_enable(&en); \
}
#define ADD_RECEIVE_BAD_SIZE_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, name) \
ATF_TP_ADD_TC(tp, ktls_receive_##cipher_name##_##name);
#define GEN_RECEIVE_TESTS(cipher_name, cipher_alg, key_size, auth_alg, \
minor) \
GEN_RECEIVE_APP_DATA_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, short, 64, 0) \
GEN_RECEIVE_APP_DATA_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, long, 64 * 1024, 0) \
GEN_RECEIVE_BAD_DATA_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, 64) \
GEN_RECEIVE_BAD_MAC_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, 64) \
GEN_RECEIVE_TRUNCATED_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, 64) \
GEN_RECEIVE_BAD_MAJOR_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, 64) \
GEN_RECEIVE_BAD_MINOR_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, 64) \
GEN_RECEIVE_BAD_SIZE_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, small_record, \
tls_minimum_record_payload(&en) - 1) \
GEN_RECEIVE_BAD_SIZE_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, oversized_record, \
TLS_MAX_MSG_SIZE_V10_2 * 2)
#define ADD_RECEIVE_TESTS(cipher_name, cipher_alg, key_size, auth_alg, \
minor) \
ADD_RECEIVE_APP_DATA_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, short) \
ADD_RECEIVE_APP_DATA_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, long) \
ADD_RECEIVE_BAD_DATA_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor) \
ADD_RECEIVE_BAD_MAC_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor) \
ADD_RECEIVE_TRUNCATED_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor) \
ADD_RECEIVE_BAD_MAJOR_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor) \
ADD_RECEIVE_BAD_MINOR_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor) \
ADD_RECEIVE_BAD_SIZE_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, small_record) \
ADD_RECEIVE_BAD_SIZE_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, oversized_record)
/*
* For each supported cipher suite, run several receive tests:
*
* - a short test which sends 64 bytes of application data (likely as
* a single TLS record)
*
* - a long test which sends 64KB of application data (split across
* multiple TLS records)
*
* - a test with corrupted payload data in a single TLS record
*
* - a test with a corrupted MAC in a single TLS record
*
* - a test with a truncated TLS record
*
* - tests with invalid TLS major and minor versions
*
* - a tests with a record whose is one less than the smallest valid
* size
*
* - a test with an oversized TLS record
*/
AES_CBC_NONZERO_TESTS(GEN_RECEIVE_TESTS);
AES_GCM_TESTS(GEN_RECEIVE_TESTS);
CHACHA20_TESTS(GEN_RECEIVE_TESTS);
#define GEN_RECEIVE_MTE_PADDING_TESTS(cipher_name, cipher_alg, \
key_size, auth_alg, minor) \
GEN_RECEIVE_APP_DATA_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_1, 1, 0) \
GEN_RECEIVE_APP_DATA_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_2, 2, 0) \
GEN_RECEIVE_APP_DATA_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_3, 3, 0) \
GEN_RECEIVE_APP_DATA_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_4, 4, 0) \
GEN_RECEIVE_APP_DATA_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_5, 5, 0) \
GEN_RECEIVE_APP_DATA_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_6, 6, 0) \
GEN_RECEIVE_APP_DATA_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_7, 7, 0) \
GEN_RECEIVE_APP_DATA_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_8, 8, 0) \
GEN_RECEIVE_APP_DATA_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_9, 9, 0) \
GEN_RECEIVE_APP_DATA_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_10, 10, 0) \
GEN_RECEIVE_APP_DATA_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_11, 11, 0) \
GEN_RECEIVE_APP_DATA_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_12, 12, 0) \
GEN_RECEIVE_APP_DATA_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_13, 13, 0) \
GEN_RECEIVE_APP_DATA_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_14, 14, 0) \
GEN_RECEIVE_APP_DATA_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_15, 15, 0) \
GEN_RECEIVE_APP_DATA_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_16, 16, 0) \
GEN_RECEIVE_APP_DATA_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_16_extra, 16, 16) \
GEN_RECEIVE_APP_DATA_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_32_extra, 16, 32)
#define ADD_RECEIVE_MTE_PADDING_TESTS(cipher_name, cipher_alg, \
key_size, auth_alg, minor) \
ADD_RECEIVE_APP_DATA_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_1) \
ADD_RECEIVE_APP_DATA_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_2) \
ADD_RECEIVE_APP_DATA_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_3) \
ADD_RECEIVE_APP_DATA_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_4) \
ADD_RECEIVE_APP_DATA_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_5) \
ADD_RECEIVE_APP_DATA_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_6) \
ADD_RECEIVE_APP_DATA_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_7) \
ADD_RECEIVE_APP_DATA_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_8) \
ADD_RECEIVE_APP_DATA_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_9) \
ADD_RECEIVE_APP_DATA_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_10) \
ADD_RECEIVE_APP_DATA_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_11) \
ADD_RECEIVE_APP_DATA_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_12) \
ADD_RECEIVE_APP_DATA_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_13) \
ADD_RECEIVE_APP_DATA_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_14) \
ADD_RECEIVE_APP_DATA_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_15) \
ADD_RECEIVE_APP_DATA_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_16) \
ADD_RECEIVE_APP_DATA_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_16_extra) \
ADD_RECEIVE_APP_DATA_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, padding_32_extra)
#define GEN_RECEIVE_BAD_PADDING_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, len) \
ATF_TC_WITHOUT_HEAD(ktls_receive_##cipher_name##_bad_padding); \
ATF_TC_BODY(ktls_receive_##cipher_name##_bad_padding, tc) \
{ \
struct tls_enable en; \
uint64_t seqno; \
\
ATF_REQUIRE_KTLS(); \
seqno = random(); \
build_tls_enable(tc, cipher_alg, key_size, auth_alg, minor, \
seqno, &en); \
test_ktls_receive_corrupted_padding(tc, &en, seqno, len); \
free_tls_enable(&en); \
}
#define ADD_RECEIVE_BAD_PADDING_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor) \
ATF_TP_ADD_TC(tp, ktls_receive_##cipher_name##_bad_padding);
#define GEN_RECEIVE_MTE_TESTS(cipher_name, cipher_alg, key_size, \
auth_alg, minor) \
GEN_RECEIVE_MTE_PADDING_TESTS(cipher_name, cipher_alg, \
key_size, auth_alg, minor) \
GEN_RECEIVE_BAD_PADDING_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, 64) \
GEN_RECEIVE_BAD_SIZE_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, non_block_size, \
tls_minimum_record_payload(&en) + 1)
#define ADD_RECEIVE_MTE_TESTS(cipher_name, cipher_alg, key_size, \
auth_alg, minor) \
ADD_RECEIVE_MTE_PADDING_TESTS(cipher_name, cipher_alg, \
key_size, auth_alg, minor) \
ADD_RECEIVE_BAD_PADDING_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor) \
ADD_RECEIVE_BAD_SIZE_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, non_block_size)
/*
* For AES-CBC MTE cipher suites using padding, add tests of messages
* with each possible padding size. Note that the padding_<N> tests
* do not necessarily test <N> bytes of padding as the padding is a
* function of the cipher suite's MAC length. However, cycling
* through all of the payload sizes from 1 to 16 should exercise all
* of the possible padding lengths for each suite.
*
* Two additional tests check for additional padding with an extra
* 16 or 32 bytes beyond the normal padding.
*
* Another test checks for corrupted padding.
*
* Another test checks for a record whose payload is not a multiple of
* the AES block size.
*/
AES_CBC_NONZERO_TESTS(GEN_RECEIVE_MTE_TESTS);
#define GEN_RECEIVE_BAD_IV_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor) \
ATF_TC_WITHOUT_HEAD(ktls_receive_##cipher_name##_bad_iv); \
ATF_TC_BODY(ktls_receive_##cipher_name##_bad_iv, tc) \
{ \
struct tls_enable en; \
uint64_t seqno; \
\
ATF_REQUIRE_KTLS(); \
seqno = random(); \
build_tls_enable(tc, cipher_alg, key_size, auth_alg, minor, \
seqno, &en); \
test_ktls_receive_corrupted_iv(tc, &en, seqno, 64); \
free_tls_enable(&en); \
}
#define ADD_RECEIVE_BAD_IV_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor) \
ATF_TP_ADD_TC(tp, ktls_receive_##cipher_name##_bad_iv);
#define GEN_RECEIVE_EXPLICIT_IV_TESTS(cipher_name, cipher_alg, \
key_size, auth_alg, minor) \
GEN_RECEIVE_BAD_IV_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor) \
GEN_RECEIVE_BAD_SIZE_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, short_header, \
sizeof(struct tls_record_layer) + 1)
#define ADD_RECEIVE_EXPLICIT_IV_TESTS(cipher_name, cipher_alg, \
key_size, auth_alg, minor) \
ADD_RECEIVE_BAD_IV_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor) \
ADD_RECEIVE_BAD_SIZE_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, short_header)
/*
* For cipher suites with an explicit IV, run a receive test where the
* explicit IV has been corrupted. Also run a receive test that sends
* a short record without a complete IV.
*/
AES_CBC_NONZERO_TESTS(GEN_RECEIVE_EXPLICIT_IV_TESTS);
AES_GCM_12_TESTS(GEN_RECEIVE_EXPLICIT_IV_TESTS);
#define GEN_RECEIVE_BAD_TYPE_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, len) \
ATF_TC_WITHOUT_HEAD(ktls_receive_##cipher_name##_bad_type); \
ATF_TC_BODY(ktls_receive_##cipher_name##_bad_type, tc) \
{ \
struct tls_enable en; \
uint64_t seqno; \
\
ATF_REQUIRE_KTLS(); \
seqno = random(); \
build_tls_enable(tc, cipher_alg, key_size, auth_alg, minor, \
seqno, &en); \
test_ktls_receive_bad_type(tc, &en, seqno, len); \
free_tls_enable(&en); \
}
#define ADD_RECEIVE_BAD_TYPE_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor) \
ATF_TP_ADD_TC(tp, ktls_receive_##cipher_name##_bad_type);
#define GEN_RECEIVE_TLS13_TESTS(cipher_name, cipher_alg, key_size, \
auth_alg, minor) \
GEN_RECEIVE_APP_DATA_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, short_padded, 64, 16) \
GEN_RECEIVE_APP_DATA_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, long_padded, 64 * 1024, 15) \
GEN_RECEIVE_BAD_TYPE_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, 64)
#define ADD_RECEIVE_TLS13_TESTS(cipher_name, cipher_alg, key_size, \
auth_alg, minor) \
ADD_RECEIVE_APP_DATA_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, short_padded) \
ADD_RECEIVE_APP_DATA_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor, long_padded) \
ADD_RECEIVE_BAD_TYPE_TEST(cipher_name, cipher_alg, key_size, \
auth_alg, minor)
/*
* For TLS 1.3 cipher suites, run two additional receive tests which
* use add padding to each record. Also run a test that uses an
* invalid "outer" record type.
*/
TLS_13_TESTS(GEN_RECEIVE_TLS13_TESTS);
static void
test_ktls_invalid_receive_cipher_suite(const atf_tc_t *tc,
struct tls_enable *en)
{
int sockets[2];
ATF_REQUIRE_MSG(open_sockets(tc, sockets), "failed to create sockets");
ATF_REQUIRE_ERRNO(EINVAL, setsockopt(sockets[1], IPPROTO_TCP,
TCP_RXTLS_ENABLE, en, sizeof(*en)) == -1);
close_sockets(sockets);
}
#define GEN_INVALID_RECEIVE_TEST(name, cipher_alg, key_size, auth_alg, \
minor) \
ATF_TC_WITHOUT_HEAD(ktls_receive_invalid_##name); \
ATF_TC_BODY(ktls_receive_invalid_##name, tc) \
{ \
struct tls_enable en; \
uint64_t seqno; \
\
ATF_REQUIRE_KTLS(); \
seqno = random(); \
build_tls_enable(tc, cipher_alg, key_size, auth_alg, minor, \
seqno, &en); \
test_ktls_invalid_receive_cipher_suite(tc, &en); \
free_tls_enable(&en); \
}
#define ADD_INVALID_RECEIVE_TEST(name, cipher_alg, key_size, auth_alg, \
minor) \
ATF_TP_ADD_TC(tp, ktls_receive_invalid_##name);
/*
* Ensure that invalid cipher suites are rejected for receive.
*/
INVALID_CIPHER_SUITES(GEN_INVALID_RECEIVE_TEST);
static void
test_ktls_unsupported_receive_cipher_suite(const atf_tc_t *tc,
struct tls_enable *en)
{
int sockets[2];
ATF_REQUIRE_MSG(open_sockets(tc, sockets), "failed to create sockets");
ATF_REQUIRE_ERRNO(EPROTONOSUPPORT, setsockopt(sockets[1], IPPROTO_TCP,
TCP_RXTLS_ENABLE, en, sizeof(*en)) == -1);
close_sockets(sockets);
}
#define GEN_UNSUPPORTED_RECEIVE_TEST(name, cipher_alg, key_size, \
auth_alg, minor) \
ATF_TC_WITHOUT_HEAD(ktls_receive_unsupported_##name); \
ATF_TC_BODY(ktls_receive_unsupported_##name, tc) \
{ \
struct tls_enable en; \
uint64_t seqno; \
\
ATF_REQUIRE_KTLS(); \
seqno = random(); \
build_tls_enable(tc, cipher_alg, key_size, auth_alg, minor, \
seqno, &en); \
test_ktls_unsupported_receive_cipher_suite(tc, &en); \
free_tls_enable(&en); \
}
#define ADD_UNSUPPORTED_RECEIVE_TEST(name, cipher_alg, key_size, \
auth_alg, minor) \
ATF_TP_ADD_TC(tp, ktls_receive_unsupported_##name);
/*
* Ensure that valid cipher suites not supported for receive are
* rejected.
*/
TLS_10_TESTS(GEN_UNSUPPORTED_RECEIVE_TEST);
/*
* Try to perform an invalid sendto(2) on a TXTLS-enabled socket, to exercise
* KTLS error handling in the socket layer.
*/
ATF_TC_WITHOUT_HEAD(ktls_sendto_baddst);
ATF_TC_BODY(ktls_sendto_baddst, tc)
{
char buf[32];
struct sockaddr_in dst;
struct tls_enable en;
ssize_t n;
int s;
ATF_REQUIRE_KTLS();
s = socket(AF_INET, SOCK_STREAM, IPPROTO_TCP);
ATF_REQUIRE(s >= 0);
build_tls_enable(tc, CRYPTO_AES_NIST_GCM_16, 128 / 8, 0,
TLS_MINOR_VER_THREE, (uint64_t)random(), &en);
ATF_REQUIRE(setsockopt(s, IPPROTO_TCP, TCP_TXTLS_ENABLE, &en,
sizeof(en)) == 0);
memset(&dst, 0, sizeof(dst));
dst.sin_family = AF_INET;
dst.sin_len = sizeof(dst);
dst.sin_addr.s_addr = htonl(INADDR_BROADCAST);
dst.sin_port = htons(12345);
memset(buf, 0, sizeof(buf));
n = sendto(s, buf, sizeof(buf), 0, (struct sockaddr *)&dst,
sizeof(dst));
/* Can't transmit to the broadcast address over TCP. */
ATF_REQUIRE_ERRNO(EACCES, n == -1);
ATF_REQUIRE(close(s) == 0);
}
/*
* Make sure that listen(2) returns an error for KTLS-enabled sockets, and
* verify that an attempt to enable KTLS on a listening socket fails.
*/
ATF_TC_WITHOUT_HEAD(ktls_listening_socket);
ATF_TC_BODY(ktls_listening_socket, tc)
{
struct tls_enable en;
struct sockaddr_in sin;
int s;
ATF_REQUIRE_KTLS();
s = socket(AF_INET, SOCK_STREAM, IPPROTO_TCP);
ATF_REQUIRE(s >= 0);
build_tls_enable(tc, CRYPTO_AES_NIST_GCM_16, 128 / 8, 0,
TLS_MINOR_VER_THREE, (uint64_t)random(), &en);
ATF_REQUIRE(setsockopt(s, IPPROTO_TCP, TCP_TXTLS_ENABLE, &en,
sizeof(en)) == 0);
ATF_REQUIRE_ERRNO(EINVAL, listen(s, 1) == -1);
ATF_REQUIRE(close(s) == 0);
s = socket(AF_INET, SOCK_STREAM, IPPROTO_TCP);
ATF_REQUIRE(s >= 0);
build_tls_enable(tc, CRYPTO_AES_NIST_GCM_16, 128 / 8, 0,
TLS_MINOR_VER_THREE, (uint64_t)random(), &en);
ATF_REQUIRE(setsockopt(s, IPPROTO_TCP, TCP_RXTLS_ENABLE, &en,
sizeof(en)) == 0);
ATF_REQUIRE_ERRNO(EINVAL, listen(s, 1) == -1);
ATF_REQUIRE(close(s) == 0);
s = socket(AF_INET, SOCK_STREAM, IPPROTO_TCP);
ATF_REQUIRE(s >= 0);
memset(&sin, 0, sizeof(sin));
sin.sin_family = AF_INET;
sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
ATF_REQUIRE(bind(s, (struct sockaddr *)&sin, sizeof(sin)) == 0);
ATF_REQUIRE(listen(s, 1) == 0);
build_tls_enable(tc, CRYPTO_AES_NIST_GCM_16, 128 / 8, 0,
TLS_MINOR_VER_THREE, (uint64_t)random(), &en);
ATF_REQUIRE_ERRNO(ENOTCONN,
setsockopt(s, IPPROTO_TCP, TCP_TXTLS_ENABLE, &en, sizeof(en)) != 0);
- ATF_REQUIRE_ERRNO(EINVAL,
+ ATF_REQUIRE_ERRNO(ENOTCONN,
setsockopt(s, IPPROTO_TCP, TCP_RXTLS_ENABLE, &en, sizeof(en)) != 0);
ATF_REQUIRE(close(s) == 0);
}
ATF_TP_ADD_TCS(tp)
{
/* Transmit tests */
AES_CBC_TESTS(ADD_TRANSMIT_TESTS);
AES_GCM_TESTS(ADD_TRANSMIT_TESTS);
CHACHA20_TESTS(ADD_TRANSMIT_TESTS);
AES_CBC_TESTS(ADD_TRANSMIT_PADDING_TESTS);
AES_CBC_TESTS(ADD_TRANSMIT_EMPTY_FRAGMENT_TEST);
AES_GCM_TESTS(ADD_TRANSMIT_EMPTY_FRAGMENT_TEST);
CHACHA20_TESTS(ADD_TRANSMIT_EMPTY_FRAGMENT_TEST);
INVALID_CIPHER_SUITES(ADD_INVALID_TRANSMIT_TEST);
/* Receive tests */
TLS_10_TESTS(ADD_UNSUPPORTED_RECEIVE_TEST);
AES_CBC_NONZERO_TESTS(ADD_RECEIVE_TESTS);
AES_GCM_TESTS(ADD_RECEIVE_TESTS);
CHACHA20_TESTS(ADD_RECEIVE_TESTS);
AES_CBC_NONZERO_TESTS(ADD_RECEIVE_MTE_TESTS);
AES_CBC_NONZERO_TESTS(ADD_RECEIVE_EXPLICIT_IV_TESTS);
AES_GCM_12_TESTS(ADD_RECEIVE_EXPLICIT_IV_TESTS);
TLS_13_TESTS(ADD_RECEIVE_TLS13_TESTS);
INVALID_CIPHER_SUITES(ADD_INVALID_RECEIVE_TEST);
/* Miscellaneous */
ATF_TP_ADD_TC(tp, ktls_sendto_baddst);
ATF_TP_ADD_TC(tp, ktls_listening_socket);
return (atf_no_error());
}