1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 3 * 4 * Copyright (c) 2012 Chelsio Communications, Inc. 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 19 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 26 * SUCH DAMAGE. 27 */ 28 #include <sys/cdefs.h> 29 __FBSDID("$FreeBSD$"); 30 31 #include "opt_inet.h" 32 #include "opt_inet6.h" 33 34 #ifdef TCP_OFFLOAD 35 #include <sys/param.h> 36 #include <sys/systm.h> 37 #include <sys/kernel.h> 38 #include <sys/module.h> 39 #include <sys/bus.h> 40 #include <sys/fnv_hash.h> 41 #include <sys/lock.h> 42 #include <sys/mutex.h> 43 #include <sys/rwlock.h> 44 #include <sys/socket.h> 45 #include <sys/socketvar.h> 46 #include <sys/sbuf.h> 47 #include <sys/taskqueue.h> 48 #include <net/if.h> 49 #include <net/if_types.h> 50 #include <net/ethernet.h> 51 #include <net/if_vlan_var.h> 52 #include <net/route.h> 53 #include <netinet/in.h> 54 #include <netinet/tcp_var.h> 55 #include <netinet/toecore.h> 56 57 #include "common/common.h" 58 #include "common/t4_msg.h" 59 #include "tom/t4_tom_l2t.h" 60 #include "tom/t4_tom.h" 61 62 #define VLAN_NONE 0xfff 63 64 static inline void 65 l2t_hold(struct l2t_data *d, struct l2t_entry *e) 66 { 67 68 if (atomic_fetchadd_int(&e->refcnt, 1) == 0) /* 0 -> 1 transition */ 69 atomic_subtract_int(&d->nfree, 1); 70 } 71 72 static inline u_int 73 l2_hash(struct l2t_data *d, const struct sockaddr *sa, int ifindex) 74 { 75 u_int hash, half = d->l2t_size / 2, start = 0; 76 const void *key; 77 size_t len; 78 79 KASSERT(sa->sa_family == AF_INET || sa->sa_family == AF_INET6, 80 ("%s: sa %p has unexpected sa_family %d", __func__, sa, 81 sa->sa_family)); 82 83 if (sa->sa_family == AF_INET) { 84 const struct sockaddr_in *sin = (const void *)sa; 85 86 key = &sin->sin_addr; 87 len = sizeof(sin->sin_addr); 88 } else { 89 const struct sockaddr_in6 *sin6 = (const void *)sa; 90 91 key = &sin6->sin6_addr; 92 len = sizeof(sin6->sin6_addr); 93 start = half; 94 } 95 96 hash = fnv_32_buf(key, len, FNV1_32_INIT); 97 hash = fnv_32_buf(&ifindex, sizeof(ifindex), hash); 98 hash %= half; 99 100 return (hash + start); 101 } 102 103 static inline int 104 l2_cmp(const struct sockaddr *sa, struct l2t_entry *e) 105 { 106 107 KASSERT(sa->sa_family == AF_INET || sa->sa_family == AF_INET6, 108 ("%s: sa %p has unexpected sa_family %d", __func__, sa, 109 sa->sa_family)); 110 111 if (sa->sa_family == AF_INET) { 112 const struct sockaddr_in *sin = (const void *)sa; 113 114 return (e->addr[0] != sin->sin_addr.s_addr); 115 } else { 116 const struct sockaddr_in6 *sin6 = (const void *)sa; 117 118 return (memcmp(&e->addr[0], &sin6->sin6_addr, sizeof(e->addr))); 119 } 120 } 121 122 static inline void 123 l2_store(const struct sockaddr *sa, struct l2t_entry *e) 124 { 125 126 KASSERT(sa->sa_family == AF_INET || sa->sa_family == AF_INET6, 127 ("%s: sa %p has unexpected sa_family %d", __func__, sa, 128 sa->sa_family)); 129 130 if (sa->sa_family == AF_INET) { 131 const struct sockaddr_in *sin = (const void *)sa; 132 133 e->addr[0] = sin->sin_addr.s_addr; 134 e->ipv6 = 0; 135 } else { 136 const struct sockaddr_in6 *sin6 = (const void *)sa; 137 138 memcpy(&e->addr[0], &sin6->sin6_addr, sizeof(e->addr)); 139 e->ipv6 = 1; 140 } 141 } 142 143 /* 144 * Add a WR to an L2T entry's queue of work requests awaiting resolution. 145 * Must be called with the entry's lock held. 146 */ 147 static inline void 148 arpq_enqueue(struct l2t_entry *e, struct wrqe *wr) 149 { 150 mtx_assert(&e->lock, MA_OWNED); 151 152 STAILQ_INSERT_TAIL(&e->wr_list, wr, link); 153 } 154 155 static inline void 156 send_pending(struct adapter *sc, struct l2t_entry *e) 157 { 158 struct wrqe *wr; 159 160 mtx_assert(&e->lock, MA_OWNED); 161 162 while ((wr = STAILQ_FIRST(&e->wr_list)) != NULL) { 163 STAILQ_REMOVE_HEAD(&e->wr_list, link); 164 t4_wrq_tx(sc, wr); 165 } 166 } 167 168 static void 169 resolution_failed(struct adapter *sc, struct l2t_entry *e) 170 { 171 struct tom_data *td = sc->tom_softc; 172 173 mtx_assert(&e->lock, MA_OWNED); 174 175 mtx_lock(&td->unsent_wr_lock); 176 STAILQ_CONCAT(&td->unsent_wr_list, &e->wr_list); 177 mtx_unlock(&td->unsent_wr_lock); 178 179 taskqueue_enqueue(taskqueue_thread, &td->reclaim_wr_resources); 180 } 181 182 static void 183 update_entry(struct adapter *sc, struct l2t_entry *e, uint8_t *lladdr, 184 uint16_t vtag) 185 { 186 187 mtx_assert(&e->lock, MA_OWNED); 188 189 /* 190 * The entry may be in active use (e->refcount > 0) or not. We update 191 * it even when it's not as this simplifies the case where we decide to 192 * reuse the entry later. 193 */ 194 195 if (lladdr == NULL && 196 (e->state == L2T_STATE_RESOLVING || e->state == L2T_STATE_FAILED)) { 197 /* 198 * Never got a valid L2 address for this one. Just mark it as 199 * failed instead of removing it from the hash (for which we'd 200 * need to wlock the table). 201 */ 202 e->state = L2T_STATE_FAILED; 203 resolution_failed(sc, e); 204 return; 205 206 } else if (lladdr == NULL) { 207 208 /* Valid or already-stale entry was deleted (or expired) */ 209 210 KASSERT(e->state == L2T_STATE_VALID || 211 e->state == L2T_STATE_STALE, 212 ("%s: lladdr NULL, state %d", __func__, e->state)); 213 214 e->state = L2T_STATE_STALE; 215 216 } else { 217 218 if (e->state == L2T_STATE_RESOLVING || 219 e->state == L2T_STATE_FAILED || 220 memcmp(e->dmac, lladdr, ETHER_ADDR_LEN)) { 221 222 /* unresolved -> resolved; or dmac changed */ 223 224 memcpy(e->dmac, lladdr, ETHER_ADDR_LEN); 225 e->vlan = vtag; 226 t4_write_l2e(e, 1); 227 } 228 e->state = L2T_STATE_VALID; 229 } 230 } 231 232 static int 233 resolve_entry(struct adapter *sc, struct l2t_entry *e) 234 { 235 struct tom_data *td = sc->tom_softc; 236 struct toedev *tod = &td->tod; 237 struct sockaddr_in sin = {0}; 238 struct sockaddr_in6 sin6 = {0}; 239 struct sockaddr *sa; 240 uint8_t dmac[ETHER_HDR_LEN]; 241 uint16_t vtag; 242 int rc; 243 244 if (e->ipv6 == 0) { 245 sin.sin_family = AF_INET; 246 sin.sin_len = sizeof(struct sockaddr_in); 247 sin.sin_addr.s_addr = e->addr[0]; 248 sa = (void *)&sin; 249 } else { 250 sin6.sin6_family = AF_INET6; 251 sin6.sin6_len = sizeof(struct sockaddr_in6); 252 memcpy(&sin6.sin6_addr, &e->addr[0], sizeof(e->addr)); 253 sa = (void *)&sin6; 254 } 255 256 vtag = EVL_MAKETAG(VLAN_NONE, 0, 0); 257 rc = toe_l2_resolve(tod, e->ifp, sa, dmac, &vtag); 258 if (rc == EWOULDBLOCK) 259 return (rc); 260 261 mtx_lock(&e->lock); 262 update_entry(sc, e, rc == 0 ? dmac : NULL, vtag); 263 mtx_unlock(&e->lock); 264 265 return (rc); 266 } 267 268 int 269 t4_l2t_send_slow(struct adapter *sc, struct wrqe *wr, struct l2t_entry *e) 270 { 271 272 again: 273 switch (e->state) { 274 case L2T_STATE_STALE: /* entry is stale, kick off revalidation */ 275 276 resolve_entry(sc, e); 277 278 /* Fall through */ 279 280 case L2T_STATE_VALID: /* fast-path, send the packet on */ 281 282 t4_wrq_tx(sc, wr); 283 return (0); 284 285 case L2T_STATE_RESOLVING: 286 case L2T_STATE_SYNC_WRITE: 287 288 mtx_lock(&e->lock); 289 if (e->state != L2T_STATE_SYNC_WRITE && 290 e->state != L2T_STATE_RESOLVING) { 291 /* state changed by the time we got here */ 292 mtx_unlock(&e->lock); 293 goto again; 294 } 295 arpq_enqueue(e, wr); 296 mtx_unlock(&e->lock); 297 298 if (resolve_entry(sc, e) == EWOULDBLOCK) 299 break; 300 301 mtx_lock(&e->lock); 302 if (e->state == L2T_STATE_VALID && !STAILQ_EMPTY(&e->wr_list)) 303 send_pending(sc, e); 304 if (e->state == L2T_STATE_FAILED) 305 resolution_failed(sc, e); 306 mtx_unlock(&e->lock); 307 break; 308 309 case L2T_STATE_FAILED: 310 return (EHOSTUNREACH); 311 } 312 313 return (0); 314 } 315 316 int 317 do_l2t_write_rpl2(struct sge_iq *iq, const struct rss_header *rss, 318 struct mbuf *m) 319 { 320 struct adapter *sc = iq->adapter; 321 const struct cpl_l2t_write_rpl *rpl = (const void *)(rss + 1); 322 unsigned int tid = GET_TID(rpl); 323 unsigned int idx = tid % L2T_SIZE; 324 325 if (__predict_false(rpl->status != CPL_ERR_NONE)) { 326 log(LOG_ERR, 327 "Unexpected L2T_WRITE_RPL (%u) for entry at hw_idx %u\n", 328 rpl->status, idx); 329 return (EINVAL); 330 } 331 332 if (tid & F_SYNC_WR) { 333 struct l2t_entry *e = &sc->l2t->l2tab[idx - sc->vres.l2t.start]; 334 335 mtx_lock(&e->lock); 336 if (e->state != L2T_STATE_SWITCHING) { 337 send_pending(sc, e); 338 e->state = L2T_STATE_VALID; 339 } 340 mtx_unlock(&e->lock); 341 } 342 343 return (0); 344 } 345 346 /* 347 * The TOE wants an L2 table entry that it can use to reach the next hop over 348 * the specified port. Produce such an entry - create one if needed. 349 * 350 * Note that the ifnet could be a pseudo-device like if_vlan, if_lagg, etc. on 351 * top of the real cxgbe interface. 352 */ 353 struct l2t_entry * 354 t4_l2t_get(struct port_info *pi, struct ifnet *ifp, struct sockaddr *sa) 355 { 356 struct l2t_entry *e; 357 struct adapter *sc = pi->adapter; 358 struct l2t_data *d = sc->l2t; 359 u_int hash, smt_idx = pi->port_id; 360 uint16_t vid, pcp, vtag; 361 362 KASSERT(sa->sa_family == AF_INET || sa->sa_family == AF_INET6, 363 ("%s: sa %p has unexpected sa_family %d", __func__, sa, 364 sa->sa_family)); 365 366 vid = VLAN_NONE; 367 pcp = 0; 368 if (ifp->if_type == IFT_L2VLAN) { 369 VLAN_TAG(ifp, &vid); 370 VLAN_PCP(ifp, &pcp); 371 } else if (ifp->if_pcp != IFNET_PCP_NONE) { 372 vid = 0; 373 pcp = ifp->if_pcp; 374 } 375 vtag = EVL_MAKETAG(vid, pcp, 0); 376 377 hash = l2_hash(d, sa, ifp->if_index); 378 rw_wlock(&d->lock); 379 for (e = d->l2tab[hash].first; e; e = e->next) { 380 if (l2_cmp(sa, e) == 0 && e->ifp == ifp && e->vlan == vtag && 381 e->smt_idx == smt_idx) { 382 l2t_hold(d, e); 383 goto done; 384 } 385 } 386 387 /* Need to allocate a new entry */ 388 e = t4_alloc_l2e(d); 389 if (e) { 390 mtx_lock(&e->lock); /* avoid race with t4_l2t_free */ 391 e->next = d->l2tab[hash].first; 392 d->l2tab[hash].first = e; 393 394 e->state = L2T_STATE_RESOLVING; 395 l2_store(sa, e); 396 e->ifp = ifp; 397 e->smt_idx = smt_idx; 398 e->hash = hash; 399 e->lport = pi->lport; 400 e->wrq = &sc->sge.ctrlq[pi->port_id]; 401 e->iqid = sc->sge.ofld_rxq[pi->vi[0].first_ofld_rxq].iq.abs_id; 402 atomic_store_rel_int(&e->refcnt, 1); 403 e->vlan = vtag; 404 mtx_unlock(&e->lock); 405 } 406 done: 407 rw_wunlock(&d->lock); 408 return e; 409 } 410 411 /* 412 * Called when the host's ARP layer makes a change to some entry that is loaded 413 * into the HW L2 table. 414 */ 415 void 416 t4_l2_update(struct toedev *tod, struct ifnet *ifp, struct sockaddr *sa, 417 uint8_t *lladdr, uint16_t vtag) 418 { 419 struct adapter *sc = tod->tod_softc; 420 struct l2t_entry *e; 421 struct l2t_data *d = sc->l2t; 422 u_int hash; 423 424 KASSERT(d != NULL, ("%s: no L2 table", __func__)); 425 426 hash = l2_hash(d, sa, ifp->if_index); 427 rw_rlock(&d->lock); 428 for (e = d->l2tab[hash].first; e; e = e->next) { 429 if (l2_cmp(sa, e) == 0 && e->ifp == ifp) { 430 mtx_lock(&e->lock); 431 if (atomic_load_acq_int(&e->refcnt)) 432 goto found; 433 e->state = L2T_STATE_STALE; 434 mtx_unlock(&e->lock); 435 break; 436 } 437 } 438 rw_runlock(&d->lock); 439 440 /* 441 * This is of no interest to us. We've never had an offloaded 442 * connection to this destination, and we aren't attempting one right 443 * now. 444 */ 445 return; 446 447 found: 448 rw_runlock(&d->lock); 449 450 KASSERT(e->state != L2T_STATE_UNUSED, 451 ("%s: unused entry in the hash.", __func__)); 452 453 update_entry(sc, e, lladdr, vtag); 454 mtx_unlock(&e->lock); 455 } 456 #endif 457