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 #include <sys/param.h> 35 #include <sys/eventhandler.h> 36 #include <sys/systm.h> 37 #include <sys/kernel.h> 38 #include <sys/module.h> 39 #include <sys/bus.h> 40 #include <sys/lock.h> 41 #include <sys/mutex.h> 42 #include <sys/rwlock.h> 43 #include <sys/socket.h> 44 #include <sys/sbuf.h> 45 #include <netinet/in.h> 46 47 #include "common/common.h" 48 #include "common/t4_msg.h" 49 #include "t4_l2t.h" 50 51 /* 52 * Module locking notes: There is a RW lock protecting the L2 table as a 53 * whole plus a spinlock per L2T entry. Entry lookups and allocations happen 54 * under the protection of the table lock, individual entry changes happen 55 * while holding that entry's spinlock. The table lock nests outside the 56 * entry locks. Allocations of new entries take the table lock as writers so 57 * no other lookups can happen while allocating new entries. Entry updates 58 * take the table lock as readers so multiple entries can be updated in 59 * parallel. An L2T entry can be dropped by decrementing its reference count 60 * and therefore can happen in parallel with entry allocation but no entry 61 * can change state or increment its ref count during allocation as both of 62 * these perform lookups. 63 * 64 * Note: We do not take references to ifnets in this module because both 65 * the TOE and the sockets already hold references to the interfaces and the 66 * lifetime of an L2T entry is fully contained in the lifetime of the TOE. 67 */ 68 69 /* 70 * Allocate a free L2T entry. Must be called with l2t_data.lock held. 71 */ 72 struct l2t_entry * 73 t4_alloc_l2e(struct l2t_data *d) 74 { 75 struct l2t_entry *end, *e, **p; 76 77 rw_assert(&d->lock, RA_WLOCKED); 78 79 if (!atomic_load_acq_int(&d->nfree)) 80 return (NULL); 81 82 /* there's definitely a free entry */ 83 for (e = d->rover, end = &d->l2tab[d->l2t_size]; e != end; ++e) 84 if (atomic_load_acq_int(&e->refcnt) == 0) 85 goto found; 86 87 for (e = d->l2tab; atomic_load_acq_int(&e->refcnt); ++e) 88 continue; 89 found: 90 d->rover = e + 1; 91 atomic_subtract_int(&d->nfree, 1); 92 93 /* 94 * The entry we found may be an inactive entry that is 95 * presently in the hash table. We need to remove it. 96 */ 97 if (e->state < L2T_STATE_SWITCHING) { 98 for (p = &d->l2tab[e->hash].first; *p; p = &(*p)->next) { 99 if (*p == e) { 100 *p = e->next; 101 e->next = NULL; 102 break; 103 } 104 } 105 } 106 107 e->state = L2T_STATE_UNUSED; 108 return (e); 109 } 110 111 static struct l2t_entry * 112 find_or_alloc_l2e(struct l2t_data *d, uint16_t vlan, uint8_t port, uint8_t *dmac) 113 { 114 struct l2t_entry *end, *e, **p; 115 struct l2t_entry *first_free = NULL; 116 117 for (e = &d->l2tab[0], end = &d->l2tab[d->l2t_size]; e != end; ++e) { 118 if (atomic_load_acq_int(&e->refcnt) == 0) { 119 if (!first_free) 120 first_free = e; 121 } else if (e->state == L2T_STATE_SWITCHING && 122 memcmp(e->dmac, dmac, ETHER_ADDR_LEN) == 0 && 123 e->vlan == vlan && e->lport == port) 124 return (e); /* Found existing entry that matches. */ 125 } 126 127 if (first_free == NULL) 128 return (NULL); /* No match and no room for a new entry. */ 129 130 /* 131 * The entry we found may be an inactive entry that is 132 * presently in the hash table. We need to remove it. 133 */ 134 e = first_free; 135 if (e->state < L2T_STATE_SWITCHING) { 136 for (p = &d->l2tab[e->hash].first; *p; p = &(*p)->next) { 137 if (*p == e) { 138 *p = e->next; 139 e->next = NULL; 140 break; 141 } 142 } 143 } 144 e->state = L2T_STATE_UNUSED; 145 return (e); 146 } 147 148 149 /* 150 * Write an L2T entry. Must be called with the entry locked. 151 * The write may be synchronous or asynchronous. 152 */ 153 int 154 t4_write_l2e(struct l2t_entry *e, int sync) 155 { 156 struct sge_wrq *wrq; 157 struct adapter *sc; 158 struct wrq_cookie cookie; 159 struct cpl_l2t_write_req *req; 160 int idx; 161 162 mtx_assert(&e->lock, MA_OWNED); 163 MPASS(e->wrq != NULL); 164 165 wrq = e->wrq; 166 sc = wrq->adapter; 167 168 req = start_wrq_wr(wrq, howmany(sizeof(*req), 16), &cookie); 169 if (req == NULL) 170 return (ENOMEM); 171 172 idx = e->idx + sc->vres.l2t.start; 173 INIT_TP_WR(req, 0); 174 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_L2T_WRITE_REQ, idx | 175 V_SYNC_WR(sync) | V_TID_QID(e->iqid))); 176 req->params = htons(V_L2T_W_PORT(e->lport) | V_L2T_W_NOREPLY(!sync)); 177 req->l2t_idx = htons(idx); 178 req->vlan = htons(e->vlan); 179 memcpy(req->dst_mac, e->dmac, sizeof(req->dst_mac)); 180 181 commit_wrq_wr(wrq, req, &cookie); 182 183 if (sync && e->state != L2T_STATE_SWITCHING) 184 e->state = L2T_STATE_SYNC_WRITE; 185 186 return (0); 187 } 188 189 /* 190 * Allocate an L2T entry for use by a switching rule. Such need to be 191 * explicitly freed and while busy they are not on any hash chain, so normal 192 * address resolution updates do not see them. 193 */ 194 struct l2t_entry * 195 t4_l2t_alloc_switching(struct adapter *sc, uint16_t vlan, uint8_t port, 196 uint8_t *eth_addr) 197 { 198 struct l2t_data *d = sc->l2t; 199 struct l2t_entry *e; 200 int rc; 201 202 rw_wlock(&d->lock); 203 e = find_or_alloc_l2e(d, vlan, port, eth_addr); 204 if (e) { 205 if (atomic_load_acq_int(&e->refcnt) == 0) { 206 mtx_lock(&e->lock); /* avoid race with t4_l2t_free */ 207 e->wrq = &sc->sge.ctrlq[0]; 208 e->iqid = sc->sge.fwq.abs_id; 209 e->state = L2T_STATE_SWITCHING; 210 e->vlan = vlan; 211 e->lport = port; 212 memcpy(e->dmac, eth_addr, ETHER_ADDR_LEN); 213 atomic_store_rel_int(&e->refcnt, 1); 214 atomic_subtract_int(&d->nfree, 1); 215 rc = t4_write_l2e(e, 0); 216 mtx_unlock(&e->lock); 217 if (rc != 0) 218 e = NULL; 219 } else { 220 MPASS(e->vlan == vlan); 221 MPASS(e->lport == port); 222 atomic_add_int(&e->refcnt, 1); 223 } 224 } 225 rw_wunlock(&d->lock); 226 return (e); 227 } 228 229 int 230 t4_init_l2t(struct adapter *sc, int flags) 231 { 232 int i, l2t_size; 233 struct l2t_data *d; 234 235 l2t_size = sc->vres.l2t.size; 236 if (l2t_size < 2) /* At least 1 bucket for IP and 1 for IPv6 */ 237 return (EINVAL); 238 239 d = malloc(sizeof(*d) + l2t_size * sizeof (struct l2t_entry), M_CXGBE, 240 M_ZERO | flags); 241 if (!d) 242 return (ENOMEM); 243 244 d->l2t_size = l2t_size; 245 d->rover = d->l2tab; 246 atomic_store_rel_int(&d->nfree, l2t_size); 247 rw_init(&d->lock, "L2T"); 248 249 for (i = 0; i < l2t_size; i++) { 250 struct l2t_entry *e = &d->l2tab[i]; 251 252 e->idx = i; 253 e->state = L2T_STATE_UNUSED; 254 mtx_init(&e->lock, "L2T_E", NULL, MTX_DEF); 255 STAILQ_INIT(&e->wr_list); 256 atomic_store_rel_int(&e->refcnt, 0); 257 } 258 259 sc->l2t = d; 260 261 return (0); 262 } 263 264 int 265 t4_free_l2t(struct l2t_data *d) 266 { 267 int i; 268 269 for (i = 0; i < d->l2t_size; i++) 270 mtx_destroy(&d->l2tab[i].lock); 271 rw_destroy(&d->lock); 272 free(d, M_CXGBE); 273 274 return (0); 275 } 276 277 int 278 do_l2t_write_rpl(struct sge_iq *iq, const struct rss_header *rss, 279 struct mbuf *m) 280 { 281 const struct cpl_l2t_write_rpl *rpl = (const void *)(rss + 1); 282 unsigned int tid = GET_TID(rpl); 283 unsigned int idx = tid % L2T_SIZE; 284 285 if (__predict_false(rpl->status != CPL_ERR_NONE)) { 286 log(LOG_ERR, 287 "Unexpected L2T_WRITE_RPL (%u) for entry at hw_idx %u\n", 288 rpl->status, idx); 289 return (EINVAL); 290 } 291 292 return (0); 293 } 294 295 static inline unsigned int 296 vlan_prio(const struct l2t_entry *e) 297 { 298 return e->vlan >> 13; 299 } 300 301 static char 302 l2e_state(const struct l2t_entry *e) 303 { 304 switch (e->state) { 305 case L2T_STATE_VALID: return 'V'; /* valid, fast-path entry */ 306 case L2T_STATE_STALE: return 'S'; /* needs revalidation, but usable */ 307 case L2T_STATE_SYNC_WRITE: return 'W'; 308 case L2T_STATE_RESOLVING: return STAILQ_EMPTY(&e->wr_list) ? 'R' : 'A'; 309 case L2T_STATE_SWITCHING: return 'X'; 310 default: return 'U'; 311 } 312 } 313 314 int 315 sysctl_l2t(SYSCTL_HANDLER_ARGS) 316 { 317 struct adapter *sc = arg1; 318 struct l2t_data *l2t = sc->l2t; 319 struct l2t_entry *e; 320 struct sbuf *sb; 321 int rc, i, header = 0; 322 char ip[INET6_ADDRSTRLEN]; 323 324 if (l2t == NULL) 325 return (ENXIO); 326 327 rc = sysctl_wire_old_buffer(req, 0); 328 if (rc != 0) 329 return (rc); 330 331 sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req); 332 if (sb == NULL) 333 return (ENOMEM); 334 335 e = &l2t->l2tab[0]; 336 for (i = 0; i < l2t->l2t_size; i++, e++) { 337 mtx_lock(&e->lock); 338 if (e->state == L2T_STATE_UNUSED) 339 goto skip; 340 341 if (header == 0) { 342 sbuf_printf(sb, " Idx IP address " 343 "Ethernet address VLAN/P LP State Users Port"); 344 header = 1; 345 } 346 if (e->state == L2T_STATE_SWITCHING) 347 ip[0] = 0; 348 else { 349 inet_ntop(e->ipv6 ? AF_INET6 : AF_INET, &e->addr[0], 350 &ip[0], sizeof(ip)); 351 } 352 353 /* 354 * XXX: IPv6 addresses may not align properly in the output. 355 */ 356 sbuf_printf(sb, "\n%4u %-15s %02x:%02x:%02x:%02x:%02x:%02x %4d" 357 " %u %2u %c %5u %s", 358 e->idx, ip, e->dmac[0], e->dmac[1], e->dmac[2], 359 e->dmac[3], e->dmac[4], e->dmac[5], 360 e->vlan & 0xfff, vlan_prio(e), e->lport, 361 l2e_state(e), atomic_load_acq_int(&e->refcnt), 362 e->ifp ? e->ifp->if_xname : "-"); 363 skip: 364 mtx_unlock(&e->lock); 365 } 366 367 rc = sbuf_finish(sb); 368 sbuf_delete(sb); 369 370 return (rc); 371 } 372