| @@ -0,0 +1,821 @@ |
| | 1 | // SPDX-License-Identifier: GPL-2.0-only |
| | 2 | // Copyright (c) 2020 Facebook Inc. |
| | 3 | |
| | 4 | #include <linux/netdevice.h> |
| | 5 | #include <linux/slab.h> |
| | 6 | #include <linux/types.h> |
| | 7 | #include <linux/workqueue.h> |
| | 8 | #include <net/udp_tunnel.h> |
| | 9 | |
| | 10 | enum udp_tunnel_nic_table_entry_flags { |
| | 11 | UDP_TUNNEL_NIC_ENTRY_ADD = BIT(0), |
| | 12 | UDP_TUNNEL_NIC_ENTRY_DEL = BIT(1), |
| | 13 | UDP_TUNNEL_NIC_ENTRY_OP_FAIL = BIT(2), |
| | 14 | UDP_TUNNEL_NIC_ENTRY_FROZEN = BIT(3), |
| | 15 | }; |
| | 16 | |
| | 17 | struct udp_tunnel_nic_table_entry { |
| | 18 | __be16 port; |
| | 19 | u8 type; |
| | 20 | u8 use_cnt; |
| | 21 | u8 flags; |
| | 22 | u8 hw_priv; |
| | 23 | }; |
| | 24 | |
| | 25 | /** |
| | 26 | * struct udp_tunnel_nic - UDP tunnel port offload state |
| | 27 | * @work: async work for talking to hardware from process context |
| | 28 | * @dev: netdev pointer |
| | 29 | * @need_sync: at least one port start changed |
| | 30 | * @need_replay: space was freed, we need a replay of all ports |
| | 31 | * @work_pending: @work is currently scheduled |
| | 32 | * @n_tables: number of tables under @entries |
| | 33 | * @missed: bitmap of tables which overflown |
| | 34 | * @entries: table of tables of ports currently offloaded |
| | 35 | */ |
| | 36 | struct udp_tunnel_nic { |
| | 37 | struct work_struct work; |
| | 38 | |
| | 39 | struct net_device *dev; |
| | 40 | |
| | 41 | u8 need_sync:1; |
| | 42 | u8 need_replay:1; |
| | 43 | u8 work_pending:1; |
| | 44 | |
| | 45 | unsigned int n_tables; |
| | 46 | unsigned long missed; |
| | 47 | struct udp_tunnel_nic_table_entry **entries; |
| | 48 | }; |
| | 49 | |
| | 50 | /* We ensure all work structs are done using driver state, but not the code. |
| | 51 | * We need a workqueue we can flush before module gets removed. |
| | 52 | */ |
| | 53 | static struct workqueue_struct *udp_tunnel_nic_workqueue; |
| | 54 | |
| | 55 | static const char *udp_tunnel_nic_tunnel_type_name(unsigned int type) |
| | 56 | { |
| | 57 | switch (type) { |
| | 58 | case UDP_TUNNEL_TYPE_VXLAN: |
| | 59 | return "vxlan"; |
| | 60 | case UDP_TUNNEL_TYPE_GENEVE: |
| | 61 | return "geneve"; |
| | 62 | case UDP_TUNNEL_TYPE_VXLAN_GPE: |
| | 63 | return "vxlan-gpe"; |
| | 64 | default: |
| | 65 | return "unknown"; |
| | 66 | } |
| | 67 | } |
| | 68 | |
| | 69 | static bool |
| | 70 | udp_tunnel_nic_entry_is_free(struct udp_tunnel_nic_table_entry *entry) |
| | 71 | { |
| | 72 | return entry->use_cnt == 0 && !entry->flags; |
| | 73 | } |
| | 74 | |
| | 75 | static bool |
| | 76 | udp_tunnel_nic_entry_is_frozen(struct udp_tunnel_nic_table_entry *entry) |
| | 77 | { |
| | 78 | return entry->flags & UDP_TUNNEL_NIC_ENTRY_FROZEN; |
| | 79 | } |
| | 80 | |
| | 81 | static void |
| | 82 | udp_tunnel_nic_entry_freeze_used(struct udp_tunnel_nic_table_entry *entry) |
| | 83 | { |
| | 84 | if (!udp_tunnel_nic_entry_is_free(entry)) |
| | 85 | entry->flags |= UDP_TUNNEL_NIC_ENTRY_FROZEN; |
| | 86 | } |
| | 87 | |
| | 88 | static void |
| | 89 | udp_tunnel_nic_entry_unfreeze(struct udp_tunnel_nic_table_entry *entry) |
| | 90 | { |
| | 91 | entry->flags &= ~UDP_TUNNEL_NIC_ENTRY_FROZEN; |
| | 92 | } |
| | 93 | |
| | 94 | static bool |
| | 95 | udp_tunnel_nic_entry_is_queued(struct udp_tunnel_nic_table_entry *entry) |
| | 96 | { |
| | 97 | return entry->flags & (UDP_TUNNEL_NIC_ENTRY_ADD | |
| | 98 | UDP_TUNNEL_NIC_ENTRY_DEL); |
| | 99 | } |
| | 100 | |
| | 101 | static void |
| | 102 | udp_tunnel_nic_entry_queue(struct udp_tunnel_nic *utn, |
| | 103 | struct udp_tunnel_nic_table_entry *entry, |
| | 104 | unsigned int flag) |
| | 105 | { |
| | 106 | entry->flags |= flag; |
| | 107 | utn->need_sync = 1; |
| | 108 | } |
| | 109 | |
| | 110 | static void |
| | 111 | udp_tunnel_nic_ti_from_entry(struct udp_tunnel_nic_table_entry *entry, |
| | 112 | struct udp_tunnel_info *ti) |
| | 113 | { |
| | 114 | memset(ti, 0, sizeof(*ti)); |
| | 115 | ti->port = entry->port; |
| | 116 | ti->type = entry->type; |
| | 117 | ti->hw_priv = entry->hw_priv; |
| | 118 | } |
| | 119 | |
| | 120 | static bool |
| | 121 | udp_tunnel_nic_is_empty(struct net_device *dev, struct udp_tunnel_nic *utn) |
| | 122 | { |
| | 123 | const struct udp_tunnel_nic_info *info = dev->udp_tunnel_nic_info; |
| | 124 | unsigned int i, j; |
| | 125 | |
| | 126 | for (i = 0; i < utn->n_tables; i++) |
| | 127 | for (j = 0; j < info->tables[i].n_entries; j++) |
| | 128 | if (!udp_tunnel_nic_entry_is_free(&utn->entries[i][j])) |
| | 129 | return false; |
| | 130 | return true; |
| | 131 | } |
| | 132 | |
| | 133 | static bool |
| | 134 | udp_tunnel_nic_should_replay(struct net_device *dev, struct udp_tunnel_nic *utn) |
| | 135 | { |
| | 136 | const struct udp_tunnel_nic_table_info *table; |
| | 137 | unsigned int i, j; |
| | 138 | |
| | 139 | if (!utn->missed) |
| | 140 | return false; |
| | 141 | |
| | 142 | for (i = 0; i < utn->n_tables; i++) { |
| | 143 | table = &dev->udp_tunnel_nic_info->tables[i]; |
| | 144 | if (!test_bit(i, &utn->missed)) |
| | 145 | continue; |
| | 146 | |
| | 147 | for (j = 0; j < table->n_entries; j++) |
| | 148 | if (udp_tunnel_nic_entry_is_free(&utn->entries[i][j])) |
| | 149 | return true; |
| | 150 | } |
| | 151 | |
| | 152 | return false; |
| | 153 | } |
| | 154 | |
| | 155 | static void |
| | 156 | __udp_tunnel_nic_get_port(struct net_device *dev, unsigned int table, |
| | 157 | unsigned int idx, struct udp_tunnel_info *ti) |
| | 158 | { |
| | 159 | struct udp_tunnel_nic_table_entry *entry; |
| | 160 | struct udp_tunnel_nic *utn; |
| | 161 | |
| | 162 | utn = dev->udp_tunnel_nic; |
| | 163 | entry = &utn->entries[table][idx]; |
| | 164 | |
| | 165 | if (entry->use_cnt) |
| | 166 | udp_tunnel_nic_ti_from_entry(entry, ti); |
| | 167 | } |
| | 168 | |
| | 169 | static void |
| | 170 | __udp_tunnel_nic_set_port_priv(struct net_device *dev, unsigned int table, |
| | 171 | unsigned int idx, u8 priv) |
| | 172 | { |
| | 173 | dev->udp_tunnel_nic->entries[table][idx].hw_priv = priv; |
| | 174 | } |
| | 175 | |
| | 176 | static void |
| | 177 | udp_tunnel_nic_entry_update_done(struct udp_tunnel_nic_table_entry *entry, |
| | 178 | int err) |
| | 179 | { |
| | 180 | bool dodgy = entry->flags & UDP_TUNNEL_NIC_ENTRY_OP_FAIL; |
| | 181 | |
| | 182 | WARN_ON_ONCE(entry->flags & UDP_TUNNEL_NIC_ENTRY_ADD && |
| | 183 | entry->flags & UDP_TUNNEL_NIC_ENTRY_DEL); |
| | 184 | |
| | 185 | if (entry->flags & UDP_TUNNEL_NIC_ENTRY_ADD && |
| | 186 | (!err || (err == -EEXIST && dodgy))) |
| | 187 | entry->flags &= ~UDP_TUNNEL_NIC_ENTRY_ADD; |
| | 188 | |
| | 189 | if (entry->flags & UDP_TUNNEL_NIC_ENTRY_DEL && |
| | 190 | (!err || (err == -ENOENT && dodgy))) |
| | 191 | entry->flags &= ~UDP_TUNNEL_NIC_ENTRY_DEL; |
| | 192 | |
| | 193 | if (!err) |
| | 194 | entry->flags &= ~UDP_TUNNEL_NIC_ENTRY_OP_FAIL; |
| | 195 | else |
| | 196 | entry->flags |= UDP_TUNNEL_NIC_ENTRY_OP_FAIL; |
| | 197 | } |
| | 198 | |
| | 199 | static void |
| | 200 | udp_tunnel_nic_device_sync_one(struct net_device *dev, |
| | 201 | struct udp_tunnel_nic *utn, |
| | 202 | unsigned int table, unsigned int idx) |
| | 203 | { |
| | 204 | struct udp_tunnel_nic_table_entry *entry; |
| | 205 | struct udp_tunnel_info ti; |
| | 206 | int err; |
| | 207 | |
| | 208 | entry = &utn->entries[table][idx]; |
| | 209 | if (!udp_tunnel_nic_entry_is_queued(entry)) |
| | 210 | return; |
| | 211 | |
| | 212 | udp_tunnel_nic_ti_from_entry(entry, &ti); |
| | 213 | if (entry->flags & UDP_TUNNEL_NIC_ENTRY_ADD) |
| | 214 | err = dev->udp_tunnel_nic_info->set_port(dev, table, idx, &ti); |
| | 215 | else |
| | 216 | err = dev->udp_tunnel_nic_info->unset_port(dev, table, idx, |
| | 217 | &ti); |
| | 218 | udp_tunnel_nic_entry_update_done(entry, err); |
| | 219 | |
| | 220 | if (err) |
| | 221 | netdev_warn(dev, |
| | 222 | "UDP tunnel port sync failed port %d type %s: %d\n", |
| | 223 | be16_to_cpu(entry->port), |
| | 224 | udp_tunnel_nic_tunnel_type_name(entry->type), |
| | 225 | err); |
| | 226 | } |
| | 227 | |
| | 228 | static void |
| | 229 | udp_tunnel_nic_device_sync_by_port(struct net_device *dev, |
| | 230 | struct udp_tunnel_nic *utn) |
| | 231 | { |
| | 232 | const struct udp_tunnel_nic_info *info = dev->udp_tunnel_nic_info; |
| | 233 | unsigned int i, j; |
| | 234 | |
| | 235 | for (i = 0; i < utn->n_tables; i++) |
| | 236 | for (j = 0; j < info->tables[i].n_entries; j++) |
| | 237 | udp_tunnel_nic_device_sync_one(dev, utn, i, j); |
| | 238 | } |
| | 239 | |
| | 240 | static void |
| | 241 | udp_tunnel_nic_device_sync_by_table(struct net_device *dev, |
| | 242 | struct udp_tunnel_nic *utn) |
| | 243 | { |
| | 244 | const struct udp_tunnel_nic_info *info = dev->udp_tunnel_nic_info; |
| | 245 | unsigned int i, j; |
| | 246 | int err; |
| | 247 | |
| | 248 | for (i = 0; i < utn->n_tables; i++) { |
| | 249 | /* Find something that needs sync in this table */ |
| | 250 | for (j = 0; j < info->tables[i].n_entries; j++) |
| | 251 | if (udp_tunnel_nic_entry_is_queued(&utn->entries[i][j])) |
| | 252 | break; |
| | 253 | if (j == info->tables[i].n_entries) |
| | 254 | continue; |
| | 255 | |
| | 256 | err = info->sync_table(dev, i); |
| | 257 | if (err) |
| | 258 | netdev_warn(dev, "UDP tunnel port sync failed for table %d: %d\n", |
| | 259 | i, err); |
| | 260 | |
| | 261 | for (j = 0; j < info->tables[i].n_entries; j++) { |
| | 262 | struct udp_tunnel_nic_table_entry *entry; |
| | 263 | |
| | 264 | entry = &utn->entries[i][j]; |
| | 265 | if (udp_tunnel_nic_entry_is_queued(entry)) |
| | 266 | udp_tunnel_nic_entry_update_done(entry, err); |
| | 267 | } |
| | 268 | } |
| | 269 | } |
| | 270 | |
| | 271 | static void |
| | 272 | __udp_tunnel_nic_device_sync(struct net_device *dev, struct udp_tunnel_nic *utn) |
| | 273 | { |
| | 274 | if (!utn->need_sync) |
| | 275 | return; |
| | 276 | |
| | 277 | if (dev->udp_tunnel_nic_info->sync_table) |
| | 278 | udp_tunnel_nic_device_sync_by_table(dev, utn); |
| | 279 | else |
| | 280 | udp_tunnel_nic_device_sync_by_port(dev, utn); |
| | 281 | |
| | 282 | utn->need_sync = 0; |
| | 283 | /* Can't replay directly here, in case we come from the tunnel driver's |
| | 284 | * notification - trying to replay may deadlock inside tunnel driver. |
| | 285 | */ |
| | 286 | utn->need_replay = udp_tunnel_nic_should_replay(dev, utn); |
| | 287 | } |
| | 288 | |
| | 289 | static void |
| | 290 | udp_tunnel_nic_device_sync(struct net_device *dev, struct udp_tunnel_nic *utn) |
| | 291 | { |
| | 292 | const struct udp_tunnel_nic_info *info = dev->udp_tunnel_nic_info; |
| | 293 | bool may_sleep; |
| | 294 | |
| | 295 | if (!utn->need_sync) |
| | 296 | return; |
| | 297 | |
| | 298 | /* Drivers which sleep in the callback need to update from |
| | 299 | * the workqueue, if we come from the tunnel driver's notification. |
| | 300 | */ |
| | 301 | may_sleep = info->flags & UDP_TUNNEL_NIC_INFO_MAY_SLEEP; |
| | 302 | if (!may_sleep) |
| | 303 | __udp_tunnel_nic_device_sync(dev, utn); |
| | 304 | if (may_sleep || utn->need_replay) { |
| | 305 | queue_work(udp_tunnel_nic_workqueue, &utn->work); |
| | 306 | utn->work_pending = 1; |
| | 307 | } |
| | 308 | } |
| | 309 | |
| | 310 | static bool |
| | 311 | udp_tunnel_nic_table_is_capable(const struct udp_tunnel_nic_table_info *table, |
| | 312 | struct udp_tunnel_info *ti) |
| | 313 | { |
| | 314 | return table->tunnel_types & ti->type; |
| | 315 | } |
| | 316 | |
| | 317 | static bool |
| | 318 | udp_tunnel_nic_is_capable(struct net_device *dev, struct udp_tunnel_nic *utn, |
| | 319 | struct udp_tunnel_info *ti) |
| | 320 | { |
| | 321 | const struct udp_tunnel_nic_info *info = dev->udp_tunnel_nic_info; |
| | 322 | unsigned int i; |
| | 323 | |
| | 324 | /* Special case IPv4-only NICs */ |
| | 325 | if (info->flags & UDP_TUNNEL_NIC_INFO_IPV4_ONLY && |
| | 326 | ti->sa_family != AF_INET) |
| | 327 | return false; |
| | 328 | |
| | 329 | for (i = 0; i < utn->n_tables; i++) |
| | 330 | if (udp_tunnel_nic_table_is_capable(&info->tables[i], ti)) |
| | 331 | return true; |
| | 332 | return false; |
| | 333 | } |
| | 334 | |
| | 335 | static int |
| | 336 | udp_tunnel_nic_has_collision(struct net_device *dev, struct udp_tunnel_nic *utn, |
| | 337 | struct udp_tunnel_info *ti) |
| | 338 | { |
| | 339 | const struct udp_tunnel_nic_info *info = dev->udp_tunnel_nic_info; |
| | 340 | struct udp_tunnel_nic_table_entry *entry; |
| | 341 | unsigned int i, j; |
| | 342 | |
| | 343 | for (i = 0; i < utn->n_tables; i++) |
| | 344 | for (j = 0; j < info->tables[i].n_entries; j++) { |
| | 345 | entry = &utn->entries[i][j]; |
| | 346 | |
| | 347 | if (!udp_tunnel_nic_entry_is_free(entry) && |
| | 348 | entry->port == ti->port && |
| | 349 | entry->type != ti->type) { |
| | 350 | __set_bit(i, &utn->missed); |
| | 351 | return true; |
| | 352 | } |
| | 353 | } |
| | 354 | return false; |
| | 355 | } |
| | 356 | |
| | 357 | static void |
| | 358 | udp_tunnel_nic_entry_adj(struct udp_tunnel_nic *utn, |
| | 359 | unsigned int table, unsigned int idx, int use_cnt_adj) |
| | 360 | { |
| | 361 | struct udp_tunnel_nic_table_entry *entry = &utn->entries[table][idx]; |
| | 362 | bool dodgy = entry->flags & UDP_TUNNEL_NIC_ENTRY_OP_FAIL; |
| | 363 | unsigned int from, to; |
| | 364 | |
| | 365 | /* If not going from used to unused or vice versa - all done. |
| | 366 | * For dodgy entries make sure we try to sync again (queue the entry). |
| | 367 | */ |
| | 368 | entry->use_cnt += use_cnt_adj; |
| | 369 | if (!dodgy && !entry->use_cnt == !(entry->use_cnt - use_cnt_adj)) |
| | 370 | return; |
| | 371 | |
| | 372 | /* Cancel the op before it was sent to the device, if possible, |
| | 373 | * otherwise we'd need to take special care to issue commands |
| | 374 | * in the same order the ports arrived. |
| | 375 | */ |
| | 376 | if (use_cnt_adj < 0) { |
| | 377 | from = UDP_TUNNEL_NIC_ENTRY_ADD; |
| | 378 | to = UDP_TUNNEL_NIC_ENTRY_DEL; |
| | 379 | } else { |
| | 380 | from = UDP_TUNNEL_NIC_ENTRY_DEL; |
| | 381 | to = UDP_TUNNEL_NIC_ENTRY_ADD; |
| | 382 | } |
| | 383 | |
| | 384 | if (entry->flags & from) { |
| | 385 | entry->flags &= ~from; |
| | 386 | if (!dodgy) |
| | 387 | return; |
| | 388 | } |
| | 389 | |
| | 390 | udp_tunnel_nic_entry_queue(utn, entry, to); |
| | 391 | } |
| | 392 | |
| | 393 | static bool |
| | 394 | udp_tunnel_nic_entry_try_adj(struct udp_tunnel_nic *utn, |
| | 395 | unsigned int table, unsigned int idx, |
| | 396 | struct udp_tunnel_info *ti, int use_cnt_adj) |
| | 397 | { |
| | 398 | struct udp_tunnel_nic_table_entry *entry = &utn->entries[table][idx]; |
| | 399 | |
| | 400 | if (udp_tunnel_nic_entry_is_free(entry) || |
| | 401 | entry->port != ti->port || |
| | 402 | entry->type != ti->type) |
| | 403 | return false; |
| | 404 | |
| | 405 | if (udp_tunnel_nic_entry_is_frozen(entry)) |
| | 406 | return true; |
| | 407 | |
| | 408 | udp_tunnel_nic_entry_adj(utn, table, idx, use_cnt_adj); |
| | 409 | return true; |
| | 410 | } |
| | 411 | |
| | 412 | /* Try to find existing matching entry and adjust its use count, instead of |
| | 413 | * adding a new one. Returns true if entry was found. In case of delete the |
| | 414 | * entry may have gotten removed in the process, in which case it will be |
| | 415 | * queued for removal. |
| | 416 | */ |
| | 417 | static bool |
| | 418 | udp_tunnel_nic_try_existing(struct net_device *dev, struct udp_tunnel_nic *utn, |
| | 419 | struct udp_tunnel_info *ti, int use_cnt_adj) |
| | 420 | { |
| | 421 | const struct udp_tunnel_nic_table_info *table; |
| | 422 | unsigned int i, j; |
| | 423 | |
| | 424 | for (i = 0; i < utn->n_tables; i++) { |
| | 425 | table = &dev->udp_tunnel_nic_info->tables[i]; |
| | 426 | if (!udp_tunnel_nic_table_is_capable(table, ti)) |
| | 427 | continue; |
| | 428 | |
| | 429 | for (j = 0; j < table->n_entries; j++) |
| | 430 | if (udp_tunnel_nic_entry_try_adj(utn, i, j, ti, |
| | 431 | use_cnt_adj)) |
| | 432 | return true; |
| | 433 | } |
| | 434 | |
| | 435 | return false; |
| | 436 | } |
| | 437 | |
| | 438 | static bool |
| | 439 | udp_tunnel_nic_add_existing(struct net_device *dev, struct udp_tunnel_nic *utn, |
| | 440 | struct udp_tunnel_info *ti) |
| | 441 | { |
| | 442 | return udp_tunnel_nic_try_existing(dev, utn, ti, +1); |
| | 443 | } |
| | 444 | |
| | 445 | static bool |
| | 446 | udp_tunnel_nic_del_existing(struct net_device *dev, struct udp_tunnel_nic *utn, |
| | 447 | struct udp_tunnel_info *ti) |
| | 448 | { |
| | 449 | return udp_tunnel_nic_try_existing(dev, utn, ti, -1); |
| | 450 | } |
| | 451 | |
| | 452 | static bool |
| | 453 | udp_tunnel_nic_add_new(struct net_device *dev, struct udp_tunnel_nic *utn, |
| | 454 | struct udp_tunnel_info *ti) |
| | 455 | { |
| | 456 | const struct udp_tunnel_nic_table_info *table; |
| | 457 | unsigned int i, j; |
| | 458 | |
| | 459 | for (i = 0; i < utn->n_tables; i++) { |
| | 460 | table = &dev->udp_tunnel_nic_info->tables[i]; |
| | 461 | if (!udp_tunnel_nic_table_is_capable(table, ti)) |
| | 462 | continue; |
| | 463 | |
| | 464 | for (j = 0; j < table->n_entries; j++) { |
| | 465 | struct udp_tunnel_nic_table_entry *entry; |
| | 466 | |
| | 467 | entry = &utn->entries[i][j]; |
| | 468 | if (!udp_tunnel_nic_entry_is_free(entry)) |
| | 469 | continue; |
| | 470 | |
| | 471 | entry->port = ti->port; |
| | 472 | entry->type = ti->type; |
| | 473 | entry->use_cnt = 1; |
| | 474 | udp_tunnel_nic_entry_queue(utn, entry, |
| | 475 | UDP_TUNNEL_NIC_ENTRY_ADD); |
| | 476 | return true; |
| | 477 | } |
| | 478 | |
| | 479 | /* The different table may still fit this port in, but there |
| | 480 | * are no devices currently which have multiple tables accepting |
| | 481 | * the same tunnel type, and false positives are okay. |
| | 482 | */ |
| | 483 | __set_bit(i, &utn->missed); |
| | 484 | } |
| | 485 | |
| | 486 | return false; |
| | 487 | } |
| | 488 | |
| | 489 | static void |
| | 490 | __udp_tunnel_nic_add_port(struct net_device *dev, struct udp_tunnel_info *ti) |
| | 491 | { |
| | 492 | const struct udp_tunnel_nic_info *info = dev->udp_tunnel_nic_info; |
| | 493 | struct udp_tunnel_nic *utn; |
| | 494 | |
| | 495 | utn = dev->udp_tunnel_nic; |
| | 496 | if (!utn) |
| | 497 | return; |
| | 498 | if (!netif_running(dev) && info->flags & UDP_TUNNEL_NIC_INFO_OPEN_ONLY) |
| | 499 | return; |
| | 500 | |
| | 501 | if (!udp_tunnel_nic_is_capable(dev, utn, ti)) |
| | 502 | return; |
| | 503 | |
| | 504 | /* It may happen that a tunnel of one type is removed and different |
| | 505 | * tunnel type tries to reuse its port before the device was informed. |
| | 506 | * Rely on utn->missed to re-add this port later. |
| | 507 | */ |
| | 508 | if (udp_tunnel_nic_has_collision(dev, utn, ti)) |
| | 509 | return; |
| | 510 | |
| | 511 | if (!udp_tunnel_nic_add_existing(dev, utn, ti)) |
| | 512 | udp_tunnel_nic_add_new(dev, utn, ti); |
| | 513 | |
| | 514 | udp_tunnel_nic_device_sync(dev, utn); |
| | 515 | } |
| | 516 | |
| | 517 | static void |
| | 518 | __udp_tunnel_nic_del_port(struct net_device *dev, struct udp_tunnel_info *ti) |
| | 519 | { |
| | 520 | struct udp_tunnel_nic *utn; |
| | 521 | |
| | 522 | utn = dev->udp_tunnel_nic; |
| | 523 | if (!utn) |
| | 524 | return; |
| | 525 | |
| | 526 | if (!udp_tunnel_nic_is_capable(dev, utn, ti)) |
| | 527 | return; |
| | 528 | |
| | 529 | udp_tunnel_nic_del_existing(dev, utn, ti); |
| | 530 | |
| | 531 | udp_tunnel_nic_device_sync(dev, utn); |
| | 532 | } |
| | 533 | |
| | 534 | static void __udp_tunnel_nic_reset_ntf(struct net_device *dev) |
| | 535 | { |
| | 536 | const struct udp_tunnel_nic_info *info = dev->udp_tunnel_nic_info; |
| | 537 | struct udp_tunnel_nic *utn; |
| | 538 | unsigned int i, j; |
| | 539 | |
| | 540 | ASSERT_RTNL(); |
| | 541 | |
| | 542 | utn = dev->udp_tunnel_nic; |
| | 543 | if (!utn) |
| | 544 | return; |
| | 545 | |
| | 546 | utn->need_sync = false; |
| | 547 | for (i = 0; i < utn->n_tables; i++) |
| | 548 | for (j = 0; j < info->tables[i].n_entries; j++) { |
| | 549 | struct udp_tunnel_nic_table_entry *entry; |
| | 550 | |
| | 551 | entry = &utn->entries[i][j]; |
| | 552 | |
| | 553 | entry->flags &= ~(UDP_TUNNEL_NIC_ENTRY_DEL | |
| | 554 | UDP_TUNNEL_NIC_ENTRY_OP_FAIL); |
| | 555 | /* We don't release rtnl across ops */ |
| | 556 | WARN_ON(entry->flags & UDP_TUNNEL_NIC_ENTRY_FROZEN); |
| | 557 | if (!entry->use_cnt) |
| | 558 | continue; |
| | 559 | |
| | 560 | udp_tunnel_nic_entry_queue(utn, entry, |
| | 561 | UDP_TUNNEL_NIC_ENTRY_ADD); |
| | 562 | } |
| | 563 | |
| | 564 | __udp_tunnel_nic_device_sync(dev, utn); |
| | 565 | } |
| | 566 | |
| | 567 | static const struct udp_tunnel_nic_ops __udp_tunnel_nic_ops = { |
| | 568 | .get_port = __udp_tunnel_nic_get_port, |
| | 569 | .set_port_priv = __udp_tunnel_nic_set_port_priv, |
| | 570 | .add_port = __udp_tunnel_nic_add_port, |
| | 571 | .del_port = __udp_tunnel_nic_del_port, |
| | 572 | .reset_ntf = __udp_tunnel_nic_reset_ntf, |
| | 573 | }; |
| | 574 | |
| | 575 | static void |
| | 576 | udp_tunnel_nic_flush(struct net_device *dev, struct udp_tunnel_nic *utn) |
| | 577 | { |
| | 578 | const struct udp_tunnel_nic_info *info = dev->udp_tunnel_nic_info; |
| | 579 | unsigned int i, j; |
| | 580 | |
| | 581 | for (i = 0; i < utn->n_tables; i++) |
| | 582 | for (j = 0; j < info->tables[i].n_entries; j++) { |
| | 583 | int adj_cnt = -utn->entries[i][j].use_cnt; |
| | 584 | |
| | 585 | if (adj_cnt) |
| | 586 | udp_tunnel_nic_entry_adj(utn, i, j, adj_cnt); |
| | 587 | } |
| | 588 | |
| | 589 | __udp_tunnel_nic_device_sync(dev, utn); |
| | 590 | |
| | 591 | for (i = 0; i < utn->n_tables; i++) |
| | 592 | memset(utn->entries[i], 0, array_size(info->tables[i].n_entries, |
| | 593 | sizeof(**utn->entries))); |
| | 594 | WARN_ON(utn->need_sync); |
| | 595 | utn->need_replay = 0; |
| | 596 | } |
| | 597 | |
| | 598 | static void |
| | 599 | udp_tunnel_nic_replay(struct net_device *dev, struct udp_tunnel_nic *utn) |
| | 600 | { |
| | 601 | const struct udp_tunnel_nic_info *info = dev->udp_tunnel_nic_info; |
| | 602 | unsigned int i, j; |
| | 603 | |
| | 604 | /* Freeze all the ports we are already tracking so that the replay |
| | 605 | * does not double up the refcount. |
| | 606 | */ |
| | 607 | for (i = 0; i < utn->n_tables; i++) |
| | 608 | for (j = 0; j < info->tables[i].n_entries; j++) |
| | 609 | udp_tunnel_nic_entry_freeze_used(&utn->entries[i][j]); |
| | 610 | utn->missed = 0; |
| | 611 | utn->need_replay = 0; |
| | 612 | |
| | 613 | udp_tunnel_get_rx_info(dev); |
| | 614 | |
| | 615 | for (i = 0; i < utn->n_tables; i++) |
| | 616 | for (j = 0; j < info->tables[i].n_entries; j++) |
| | 617 | udp_tunnel_nic_entry_unfreeze(&utn->entries[i][j]); |
| | 618 | } |
| | 619 | |
| | 620 | static void udp_tunnel_nic_device_sync_work(struct work_struct *work) |
| | 621 | { |
| | 622 | struct udp_tunnel_nic *utn = |
| | 623 | container_of(work, struct udp_tunnel_nic, work); |
| | 624 | |
| | 625 | rtnl_lock(); |
| | 626 | utn->work_pending = 0; |
| | 627 | __udp_tunnel_nic_device_sync(utn->dev, utn); |
| | 628 | |
| | 629 | if (utn->need_replay) |
| | 630 | udp_tunnel_nic_replay(utn->dev, utn); |
| | 631 | rtnl_unlock(); |
| | 632 | } |
| | 633 | |
| | 634 | static struct udp_tunnel_nic * |
| | 635 | udp_tunnel_nic_alloc(const struct udp_tunnel_nic_info *info, |
| | 636 | unsigned int n_tables) |
| | 637 | { |
| | 638 | struct udp_tunnel_nic *utn; |
| | 639 | unsigned int i; |
| | 640 | |
| | 641 | utn = kzalloc(sizeof(*utn), GFP_KERNEL); |
| | 642 | if (!utn) |
| | 643 | return NULL; |
| | 644 | utn->n_tables = n_tables; |
| | 645 | INIT_WORK(&utn->work, udp_tunnel_nic_device_sync_work); |
| | 646 | |
| | 647 | utn->entries = kmalloc_array(n_tables, sizeof(void *), GFP_KERNEL); |
| | 648 | if (!utn->entries) |
| | 649 | goto err_free_utn; |
| | 650 | |
| | 651 | for (i = 0; i < n_tables; i++) { |
| | 652 | utn->entries[i] = kcalloc(info->tables[i].n_entries, |
| | 653 | sizeof(*utn->entries[i]), GFP_KERNEL); |
| | 654 | if (!utn->entries[i]) |
| | 655 | goto err_free_prev_entries; |
| | 656 | } |
| | 657 | |
| | 658 | return utn; |
| | 659 | |
| | 660 | err_free_prev_entries: |
| | 661 | while (i--) |
| | 662 | kfree(utn->entries[i]); |
| | 663 | kfree(utn->entries); |
| | 664 | err_free_utn: |
| | 665 | kfree(utn); |
| | 666 | return NULL; |
| | 667 | } |
| | 668 | |
| | 669 | static int udp_tunnel_nic_register(struct net_device *dev) |
| | 670 | { |
| | 671 | const struct udp_tunnel_nic_info *info = dev->udp_tunnel_nic_info; |
| | 672 | struct udp_tunnel_nic *utn; |
| | 673 | unsigned int n_tables, i; |
| | 674 | |
| | 675 | BUILD_BUG_ON(sizeof(utn->missed) * BITS_PER_BYTE < |
| | 676 | UDP_TUNNEL_NIC_MAX_TABLES); |
| | 677 | |
| | 678 | if (WARN_ON(!info->set_port != !info->unset_port) || |
| | 679 | WARN_ON(!info->set_port == !info->sync_table) || |
| | 680 | WARN_ON(!info->tables[0].n_entries)) |
| | 681 | return -EINVAL; |
| | 682 | |
| | 683 | n_tables = 1; |
| | 684 | for (i = 1; i < UDP_TUNNEL_NIC_MAX_TABLES; i++) { |
| | 685 | if (!info->tables[i].n_entries) |
| | 686 | continue; |
| | 687 | |
| | 688 | n_tables++; |
| | 689 | if (WARN_ON(!info->tables[i - 1].n_entries)) |
| | 690 | return -EINVAL; |
| | 691 | } |
| | 692 | |
| | 693 | utn = udp_tunnel_nic_alloc(info, n_tables); |
| | 694 | if (!utn) |
| | 695 | return -ENOMEM; |
| | 696 | |
| | 697 | utn->dev = dev; |
| | 698 | dev_hold(dev); |
| | 699 | dev->udp_tunnel_nic = utn; |
| | 700 | |
| | 701 | if (!(info->flags & UDP_TUNNEL_NIC_INFO_OPEN_ONLY)) |
| | 702 | udp_tunnel_get_rx_info(dev); |
| | 703 | |
| | 704 | return 0; |
| | 705 | } |
| | 706 | |
| | 707 | static void |
| | 708 | udp_tunnel_nic_unregister(struct net_device *dev, struct udp_tunnel_nic *utn) |
| | 709 | { |
| | 710 | unsigned int i; |
| | 711 | |
| | 712 | /* Flush before we check work, so we don't waste time adding entries |
| | 713 | * from the work which we will boot immediately. |
| | 714 | */ |
| | 715 | udp_tunnel_nic_flush(dev, utn); |
| | 716 | |
| | 717 | /* Wait for the work to be done using the state, netdev core will |
| | 718 | * retry unregister until we give up our reference on this device. |
| | 719 | */ |
| | 720 | if (utn->work_pending) |
| | 721 | return; |
| | 722 | |
| | 723 | for (i = 0; i < utn->n_tables; i++) |
| | 724 | kfree(utn->entries[i]); |
| | 725 | kfree(utn->entries); |
| | 726 | kfree(utn); |
| | 727 | dev->udp_tunnel_nic = NULL; |
| | 728 | dev_put(dev); |
| | 729 | } |
| | 730 | |
| | 731 | static int |
| | 732 | udp_tunnel_nic_netdevice_event(struct notifier_block *unused, |
| | 733 | unsigned long event, void *ptr) |
| | 734 | { |
| | 735 | struct net_device *dev = netdev_notifier_info_to_dev(ptr); |
| | 736 | const struct udp_tunnel_nic_info *info; |
| | 737 | struct udp_tunnel_nic *utn; |
| | 738 | |
| | 739 | info = dev->udp_tunnel_nic_info; |
| | 740 | if (!info) |
| | 741 | return NOTIFY_DONE; |
| | 742 | |
| | 743 | if (event == NETDEV_REGISTER) { |
| | 744 | int err; |
| | 745 | |
| | 746 | err = udp_tunnel_nic_register(dev); |
| | 747 | if (err) |
| | 748 | netdev_WARN(dev, "failed to register for UDP tunnel offloads: %d", err); |
| | 749 | return notifier_from_errno(err); |
| | 750 | } |
| | 751 | /* All other events will need the udp_tunnel_nic state */ |
| | 752 | utn = dev->udp_tunnel_nic; |
| | 753 | if (!utn) |
| | 754 | return NOTIFY_DONE; |
| | 755 | |
| | 756 | if (event == NETDEV_UNREGISTER) { |
| | 757 | udp_tunnel_nic_unregister(dev, utn); |
| | 758 | return NOTIFY_OK; |
| | 759 | } |
| | 760 | |
| | 761 | /* All other events only matter if NIC has to be programmed open */ |
| | 762 | if (!(info->flags & UDP_TUNNEL_NIC_INFO_OPEN_ONLY)) |
| | 763 | return NOTIFY_DONE; |
| | 764 | |
| | 765 | if (event == NETDEV_UP) { |
| | 766 | WARN_ON(!udp_tunnel_nic_is_empty(dev, utn)); |
| | 767 | udp_tunnel_get_rx_info(dev); |
| | 768 | return NOTIFY_OK; |
| | 769 | } |
| | 770 | if (event == NETDEV_GOING_DOWN) { |
| | 771 | udp_tunnel_nic_flush(dev, utn); |
| | 772 | return NOTIFY_OK; |
| | 773 | } |
| | 774 | |
| | 775 | return NOTIFY_DONE; |
| | 776 | } |
| | 777 | |
| | 778 | static struct notifier_block udp_tunnel_nic_notifier_block __read_mostly = { |
| | 779 | .notifier_call = udp_tunnel_nic_netdevice_event, |
| | 780 | }; |
| | 781 | |
| | 782 | static int __init udp_tunnel_nic_init_module(void) |
| | 783 | { |
| | 784 | int err; |
| | 785 | |
| | 786 | udp_tunnel_nic_workqueue = alloc_workqueue("udp_tunnel_nic", 0, 0); |
| | 787 | if (!udp_tunnel_nic_workqueue) |
| | 788 | return -ENOMEM; |
| | 789 | |
| | 790 | rtnl_lock(); |
| | 791 | udp_tunnel_nic_ops = &__udp_tunnel_nic_ops; |
| | 792 | rtnl_unlock(); |
| | 793 | |
| | 794 | err = register_netdevice_notifier(&udp_tunnel_nic_notifier_block); |
| | 795 | if (err) |
| | 796 | goto err_unset_ops; |
| | 797 | |
| | 798 | return 0; |
| | 799 | |
| | 800 | err_unset_ops: |
| | 801 | rtnl_lock(); |
| | 802 | udp_tunnel_nic_ops = NULL; |
| | 803 | rtnl_unlock(); |
| | 804 | destroy_workqueue(udp_tunnel_nic_workqueue); |
| | 805 | return err; |
| | 806 | } |
| | 807 | late_initcall(udp_tunnel_nic_init_module); |
| | 808 | |
| | 809 | static void __exit udp_tunnel_nic_cleanup_module(void) |
| | 810 | { |
| | 811 | unregister_netdevice_notifier(&udp_tunnel_nic_notifier_block); |
| | 812 | |
| | 813 | rtnl_lock(); |
| | 814 | udp_tunnel_nic_ops = NULL; |
| | 815 | rtnl_unlock(); |
| | 816 | |
| | 817 | destroy_workqueue(udp_tunnel_nic_workqueue); |
| | 818 | } |
| | 819 | module_exit(udp_tunnel_nic_cleanup_module); |
| | 820 | |
| | 821 | MODULE_LICENSE("GPL"); |