1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 3 * 4 * Copyright (c) 2020 Alexander V. Chernikov 5 * 6 * Redistribution and use in source and binary forms, with or without 7 * modification, are permitted provided that the following conditions 8 * are met: 9 * 1. Redistributions of source code must retain the above copyright 10 * notice, this list of conditions and the following disclaimer. 11 * 2. Redistributions in binary form must reproduce the above copyright 12 * notice, this list of conditions and the following disclaimer in the 13 * documentation and/or other materials provided with the distribution. 14 * 15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 16 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 17 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 18 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 19 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 20 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 21 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 22 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 23 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 24 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 25 * SUCH DAMAGE. 26 * 27 * $FreeBSD$ 28 */ 29 #include "opt_inet.h" 30 #include "opt_route.h" 31 32 #include <sys/cdefs.h> 33 #include <sys/param.h> 34 #include <sys/systm.h> 35 #include <sys/lock.h> 36 #include <sys/rmlock.h> 37 #include <sys/malloc.h> 38 #include <sys/mbuf.h> 39 #include <sys/refcount.h> 40 #include <sys/socket.h> 41 #include <sys/sysctl.h> 42 #include <sys/kernel.h> 43 #include <sys/epoch.h> 44 45 #include <net/if.h> 46 #include <net/if_var.h> 47 #include <net/route.h> 48 #include <net/route/route_ctl.h> 49 #include <net/route/route_var.h> 50 #include <net/vnet.h> 51 52 #include <netinet/in.h> 53 #include <netinet/in_var.h> 54 #include <netinet/in_fib.h> 55 56 #include <net/route/nhop_utils.h> 57 #include <net/route/nhop.h> 58 #include <net/route/nhop_var.h> 59 #include <net/route/nhgrp_var.h> 60 61 #define DEBUG_MOD_NAME nhgrp_ctl 62 #define DEBUG_MAX_LEVEL LOG_DEBUG 63 #include <net/route/route_debug.h> 64 _DECLARE_DEBUG(LOG_INFO); 65 66 /* 67 * This file contains the supporting functions for creating multipath groups 68 * and compiling their dataplane parts. 69 */ 70 71 /* MPF_MULTIPATH must be the same as NHF_MULTIPATH for nhop selection to work */ 72 _Static_assert(MPF_MULTIPATH == NHF_MULTIPATH, 73 "MPF_MULTIPATH must be the same as NHF_MULTIPATH"); 74 /* Offset and size of flags field has to be the same for nhop/nhop groups */ 75 CHK_STRUCT_FIELD_GENERIC(struct nhop_object, nh_flags, struct nhgrp_object, nhg_flags); 76 /* Cap multipath to 64, as the larger values would break rib_cmd_info bmasks */ 77 CTASSERT(RIB_MAX_MPATH_WIDTH <= 64); 78 79 static int wn_cmp_idx(const void *a, const void *b); 80 static void sort_weightened_nhops(struct weightened_nhop *wn, int num_nhops); 81 82 static struct nhgrp_priv *get_nhgrp(struct nh_control *ctl, 83 struct weightened_nhop *wn, int num_nhops, uint32_t uidx, int *perror); 84 static void destroy_nhgrp(struct nhgrp_priv *nhg_priv); 85 static void destroy_nhgrp_epoch(epoch_context_t ctx); 86 static void free_nhgrp_nhops(struct nhgrp_priv *nhg_priv); 87 88 static int 89 wn_cmp_idx(const void *a, const void *b) 90 { 91 const struct weightened_nhop *w_a = a; 92 const struct weightened_nhop *w_b = b; 93 uint32_t a_idx = w_a->nh->nh_priv->nh_idx; 94 uint32_t b_idx = w_b->nh->nh_priv->nh_idx; 95 96 if (a_idx < b_idx) 97 return (-1); 98 else if (a_idx > b_idx) 99 return (1); 100 else 101 return (0); 102 } 103 104 /* 105 * Perform in-place sorting for array of nexthops in @wn. 106 * Sort by nexthop index ascending. 107 */ 108 static void 109 sort_weightened_nhops(struct weightened_nhop *wn, int num_nhops) 110 { 111 112 qsort(wn, num_nhops, sizeof(struct weightened_nhop), wn_cmp_idx); 113 } 114 115 /* 116 * In order to determine the minimum weight difference in the array 117 * of weights, create a sorted array of weights, using spare "storage" 118 * field in the `struct weightened_nhop`. 119 * Assume weights to be (mostly) the same and use insertion sort to 120 * make it sorted. 121 */ 122 static void 123 sort_weightened_nhops_weights(struct weightened_nhop *wn, int num_items) 124 { 125 wn[0].storage = wn[0].weight; 126 for (int i = 1, j = 0; i < num_items; i++) { 127 uint32_t weight = wn[i].weight; // read from 'weight' as it's not reordered 128 /* Move all weights > weight 1 position right */ 129 for (j = i - 1; j >= 0 && wn[j].storage > weight; j--) 130 wn[j + 1].storage = wn[j].storage; 131 wn[j + 1].storage = weight; 132 } 133 } 134 135 /* 136 * Calculate minimum number of slots required to fit the existing 137 * set of weights in the common use case where weights are "easily" 138 * comparable. 139 * Assumes @wn is sorted by weight ascending and each weight is > 0. 140 * Returns number of slots or 0 if precise calculation failed. 141 * 142 * Some examples: 143 * note: (i, X) pair means (nhop=i, weight=X): 144 * (1, 1) (2, 2) -> 3 slots [1, 2, 2] 145 * (1, 100), (2, 200) -> 3 slots [1, 2, 2] 146 * (1, 100), (2, 200), (3, 400) -> 7 slots [1, 2, 2, 3, 3, 3] 147 */ 148 static uint32_t 149 calc_min_mpath_slots_fast(struct weightened_nhop *wn, size_t num_items, 150 uint64_t *ptotal) 151 { 152 uint32_t i, last, xmin; 153 uint64_t total = 0; 154 155 // Get sorted array of weights in .storage field 156 sort_weightened_nhops_weights(wn, num_items); 157 158 last = 0; 159 xmin = wn[0].storage; 160 for (i = 0; i < num_items; i++) { 161 total += wn[i].storage; 162 if ((wn[i].storage != last) && 163 ((wn[i].storage - last < xmin) || xmin == 0)) { 164 xmin = wn[i].storage - last; 165 } 166 last = wn[i].storage; 167 } 168 *ptotal = total; 169 /* xmin is the minimum unit of desired capacity */ 170 if ((total % xmin) != 0) 171 return (0); 172 for (i = 0; i < num_items; i++) { 173 if ((wn[i].weight % xmin) != 0) 174 return (0); 175 } 176 177 return ((uint32_t)(total / xmin)); 178 } 179 180 /* 181 * Calculate minimum number of slots required to fit the existing 182 * set of weights while maintaining weight coefficients. 183 * 184 * Assume @wn is sorted by weight ascending and each weight is > 0. 185 * 186 * Tries to find simple precise solution first and falls back to 187 * RIB_MAX_MPATH_WIDTH in case of any failure. 188 */ 189 static uint32_t 190 calc_min_mpath_slots(struct weightened_nhop *wn, size_t num_items) 191 { 192 uint32_t v; 193 uint64_t total; 194 195 v = calc_min_mpath_slots_fast(wn, num_items, &total); 196 if (total == 0) 197 return (0); 198 if ((v == 0) || (v > RIB_MAX_MPATH_WIDTH)) 199 v = RIB_MAX_MPATH_WIDTH; 200 201 return (v); 202 } 203 204 /* 205 * Nexthop group data consists of 206 * 1) dataplane part, with nhgrp_object as a header followed by an 207 * arbitrary number of nexthop pointers. 208 * 2) control plane part, with nhgrp_priv as a header, followed by 209 * an arbirtrary number of 'struct weightened_nhop' object. 210 * 211 * Given nexthop groups are (mostly) immutable, allocate all data 212 * in one go. 213 * 214 */ 215 __noinline static size_t 216 get_nhgrp_alloc_size(uint32_t nhg_size, uint32_t num_nhops) 217 { 218 size_t sz; 219 220 sz = sizeof(struct nhgrp_object); 221 sz += nhg_size * sizeof(struct nhop_object *); 222 sz += sizeof(struct nhgrp_priv); 223 sz += num_nhops * sizeof(struct weightened_nhop); 224 return (sz); 225 } 226 227 /* 228 * Compile actual list of nexthops to be used by datapath from 229 * the nexthop group @dst. 230 * 231 * For example, compiling control plane list of 2 nexthops 232 * [(200, A), (100, B)] would result in the datapath array 233 * [A, A, B] 234 */ 235 static void 236 compile_nhgrp(struct nhgrp_priv *dst_priv, const struct weightened_nhop *x, 237 uint32_t num_slots) 238 { 239 struct nhgrp_object *dst; 240 int i, slot_idx, remaining_slots; 241 uint64_t remaining_sum, nh_weight, nh_slots; 242 243 slot_idx = 0; 244 dst = dst_priv->nhg; 245 /* Calculate sum of all weights */ 246 remaining_sum = 0; 247 for (i = 0; i < dst_priv->nhg_nh_count; i++) 248 remaining_sum += x[i].weight; 249 remaining_slots = num_slots; 250 FIB_NH_LOG(LOG_DEBUG3, x[0].nh, "sum: %lu, slots: %d", 251 remaining_sum, remaining_slots); 252 for (i = 0; i < dst_priv->nhg_nh_count; i++) { 253 /* Calculate number of slots for the current nexthop */ 254 if (remaining_sum > 0) { 255 nh_weight = (uint64_t)x[i].weight; 256 nh_slots = (nh_weight * remaining_slots / remaining_sum); 257 } else 258 nh_slots = 0; 259 260 remaining_sum -= x[i].weight; 261 remaining_slots -= nh_slots; 262 263 FIB_NH_LOG(LOG_DEBUG3, x[0].nh, 264 " rem_sum: %lu, rem_slots: %d nh_slots: %d, slot_idx: %d", 265 remaining_sum, remaining_slots, (int)nh_slots, slot_idx); 266 267 KASSERT((slot_idx + nh_slots <= num_slots), 268 ("index overflow during nhg compilation")); 269 while (nh_slots-- > 0) 270 dst->nhops[slot_idx++] = x[i].nh; 271 } 272 } 273 274 /* 275 * Allocates new nexthop group for the list of weightened nexthops. 276 * Assume sorted list. 277 * Does NOT reference any nexthops in the group. 278 * Returns group with refcount=1 or NULL. 279 */ 280 static struct nhgrp_priv * 281 alloc_nhgrp(struct weightened_nhop *wn, int num_nhops) 282 { 283 uint32_t nhgrp_size; 284 struct nhgrp_object *nhg; 285 struct nhgrp_priv *nhg_priv; 286 287 nhgrp_size = calc_min_mpath_slots(wn, num_nhops); 288 if (nhgrp_size == 0) { 289 /* Zero weights, abort */ 290 return (NULL); 291 } 292 293 size_t sz = get_nhgrp_alloc_size(nhgrp_size, num_nhops); 294 nhg = malloc(sz, M_NHOP, M_NOWAIT | M_ZERO); 295 if (nhg == NULL) { 296 FIB_NH_LOG(LOG_INFO, wn[0].nh, 297 "unable to allocate group with num_nhops %d (compiled %u)", 298 num_nhops, nhgrp_size); 299 return (NULL); 300 } 301 302 /* Has to be the first to make NHGRP_PRIV() work */ 303 nhg->nhg_size = nhgrp_size; 304 nhg->nhg_flags = MPF_MULTIPATH; 305 306 nhg_priv = NHGRP_PRIV(nhg); 307 nhg_priv->nhg_nh_count = num_nhops; 308 refcount_init(&nhg_priv->nhg_refcount, 1); 309 310 /* Please see nhgrp_free() comments on the initial value */ 311 refcount_init(&nhg_priv->nhg_linked, 2); 312 313 nhg_priv->nhg = nhg; 314 memcpy(&nhg_priv->nhg_nh_weights[0], wn, 315 num_nhops * sizeof(struct weightened_nhop)); 316 317 FIB_NH_LOG(LOG_DEBUG, wn[0].nh, "num_nhops: %d, compiled_nhop: %u", 318 num_nhops, nhgrp_size); 319 320 compile_nhgrp(nhg_priv, wn, nhg->nhg_size); 321 322 return (nhg_priv); 323 } 324 325 void 326 nhgrp_ref_object(struct nhgrp_object *nhg) 327 { 328 struct nhgrp_priv *nhg_priv; 329 u_int old __diagused; 330 331 nhg_priv = NHGRP_PRIV(nhg); 332 old = refcount_acquire(&nhg_priv->nhg_refcount); 333 KASSERT(old > 0, ("%s: nhgrp object %p has 0 refs", __func__, nhg)); 334 } 335 336 void 337 nhgrp_free(struct nhgrp_object *nhg) 338 { 339 struct nhgrp_priv *nhg_priv; 340 struct nh_control *ctl; 341 struct epoch_tracker et; 342 343 nhg_priv = NHGRP_PRIV(nhg); 344 345 if (!refcount_release(&nhg_priv->nhg_refcount)) 346 return; 347 348 /* 349 * group objects don't have an explicit lock attached to it. 350 * As groups are reclaimed based on reference count, it is possible 351 * that some groups will persist after vnet destruction callback 352 * called. Given that, handle scenario with nhgrp_free_group() being 353 * called either after or simultaneously with nhgrp_ctl_unlink_all() 354 * by using another reference counter: nhg_linked. 355 * 356 * There are only 2 places, where nhg_linked can be decreased: 357 * rib destroy (nhgrp_ctl_unlink_all) and this function. 358 * nhg_link can never be increased. 359 * 360 * Hence, use initial value of 2 to make use of 361 * refcount_release_if_not_last(). 362 * 363 * There can be two scenarious when calling this function: 364 * 365 * 1) nhg_linked value is 2. This means that either 366 * nhgrp_ctl_unlink_all() has not been called OR it is running, 367 * but we are guaranteed that nh_control won't be freed in 368 * this epoch. Hence, nexthop can be safely unlinked. 369 * 370 * 2) nh_linked value is 1. In that case, nhgrp_ctl_unlink_all() 371 * has been called and nhgrp unlink can be skipped. 372 */ 373 374 NET_EPOCH_ENTER(et); 375 if (refcount_release_if_not_last(&nhg_priv->nhg_linked)) { 376 ctl = nhg_priv->nh_control; 377 if (unlink_nhgrp(ctl, nhg_priv) == NULL) { 378 /* Do not try to reclaim */ 379 RT_LOG(LOG_INFO, "Failed to unlink nexhop group %p", 380 nhg_priv); 381 NET_EPOCH_EXIT(et); 382 return; 383 } 384 } 385 NET_EPOCH_EXIT(et); 386 387 KASSERT((nhg_priv->nhg_idx == 0), ("gr_idx != 0")); 388 epoch_call(net_epoch_preempt, destroy_nhgrp_epoch, 389 &nhg_priv->nhg_epoch_ctx); 390 } 391 392 /* 393 * Destroys all local resources belonging to @nhg_priv. 394 */ 395 __noinline static void 396 destroy_nhgrp_int(struct nhgrp_priv *nhg_priv) 397 { 398 399 free(nhg_priv->nhg, M_NHOP); 400 } 401 402 __noinline static void 403 destroy_nhgrp(struct nhgrp_priv *nhg_priv) 404 { 405 406 KASSERT((nhg_priv->nhg_refcount == 0), ("nhg_refcount != 0")); 407 KASSERT((nhg_priv->nhg_idx == 0), ("gr_idx != 0")); 408 409 IF_DEBUG_LEVEL(LOG_DEBUG2) { 410 char nhgbuf[NHOP_PRINT_BUFSIZE] __unused; 411 FIB_NH_LOG(LOG_DEBUG2, nhg_priv->nhg_nh_weights[0].nh, 412 "destroying %s", nhgrp_print_buf(nhg_priv->nhg, 413 nhgbuf, sizeof(nhgbuf))); 414 } 415 416 free_nhgrp_nhops(nhg_priv); 417 destroy_nhgrp_int(nhg_priv); 418 } 419 420 /* 421 * Epoch callback indicating group is safe to destroy 422 */ 423 static void 424 destroy_nhgrp_epoch(epoch_context_t ctx) 425 { 426 struct nhgrp_priv *nhg_priv; 427 428 nhg_priv = __containerof(ctx, struct nhgrp_priv, nhg_epoch_ctx); 429 430 destroy_nhgrp(nhg_priv); 431 } 432 433 static bool 434 ref_nhgrp_nhops(struct nhgrp_priv *nhg_priv) 435 { 436 437 for (int i = 0; i < nhg_priv->nhg_nh_count; i++) { 438 if (nhop_try_ref_object(nhg_priv->nhg_nh_weights[i].nh) != 0) 439 continue; 440 441 /* 442 * Failed to ref the nexthop, b/c it's deleted. 443 * Need to rollback references back. 444 */ 445 for (int j = 0; j < i; j++) 446 nhop_free(nhg_priv->nhg_nh_weights[j].nh); 447 return (false); 448 } 449 450 return (true); 451 } 452 453 static void 454 free_nhgrp_nhops(struct nhgrp_priv *nhg_priv) 455 { 456 457 for (int i = 0; i < nhg_priv->nhg_nh_count; i++) 458 nhop_free(nhg_priv->nhg_nh_weights[i].nh); 459 } 460 461 /* 462 * Creates or looks up an existing nexthop group based on @wn and @num_nhops. 463 * 464 * Returns referenced nhop group or NULL, passing error code in @perror. 465 */ 466 struct nhgrp_priv * 467 get_nhgrp(struct nh_control *ctl, struct weightened_nhop *wn, int num_nhops, 468 uint32_t uidx, int *perror) 469 { 470 struct nhgrp_priv *key, *nhg_priv; 471 472 if (num_nhops > RIB_MAX_MPATH_WIDTH) { 473 *perror = E2BIG; 474 return (NULL); 475 } 476 477 if (ctl->gr_head.hash_size == 0) { 478 /* First multipath request. Bootstrap mpath datastructures. */ 479 if (nhgrp_ctl_alloc_default(ctl, M_NOWAIT) == 0) { 480 *perror = ENOMEM; 481 return (NULL); 482 } 483 } 484 485 /* Sort nexthops & check there are no duplicates */ 486 sort_weightened_nhops(wn, num_nhops); 487 uint32_t last_id = 0; 488 for (int i = 0; i < num_nhops; i++) { 489 if (wn[i].nh->nh_priv->nh_idx == last_id) { 490 *perror = EEXIST; 491 return (NULL); 492 } 493 last_id = wn[i].nh->nh_priv->nh_idx; 494 } 495 496 if ((key = alloc_nhgrp(wn, num_nhops)) == NULL) { 497 *perror = ENOMEM; 498 return (NULL); 499 } 500 key->nhg_uidx = uidx; 501 502 nhg_priv = find_nhgrp(ctl, key); 503 if (nhg_priv != NULL) { 504 /* 505 * Free originally-created group. As it hasn't been linked 506 * and the dependent nexhops haven't been referenced, just free 507 * the group. 508 */ 509 destroy_nhgrp_int(key); 510 *perror = 0; 511 return (nhg_priv); 512 } else { 513 /* No existing group, try to link the new one */ 514 if (!ref_nhgrp_nhops(key)) { 515 /* 516 * Some of the nexthops have been scheduled for deletion. 517 * As the group hasn't been linked / no nexhops have been 518 * referenced, call the final destructor immediately. 519 */ 520 destroy_nhgrp_int(key); 521 *perror = EAGAIN; 522 return (NULL); 523 } 524 if (link_nhgrp(ctl, key) == 0) { 525 /* Unable to allocate index? */ 526 *perror = EAGAIN; 527 free_nhgrp_nhops(key); 528 destroy_nhgrp_int(key); 529 return (NULL); 530 } 531 *perror = 0; 532 return (key); 533 } 534 535 /* NOTREACHED */ 536 } 537 538 /* 539 * Appends one or more nexthops denoted by @wm to the nexthop group @gr_orig. 540 * 541 * Returns referenced nexthop group or NULL. In the latter case, @perror is 542 * filled with an error code. 543 * Note that function does NOT care if the next nexthops already exists 544 * in the @gr_orig. As a result, they will be added, resulting in the 545 * same nexthop being present multiple times in the new group. 546 */ 547 static struct nhgrp_priv * 548 append_nhops(struct nh_control *ctl, const struct nhgrp_object *gr_orig, 549 struct weightened_nhop *wn, int num_nhops, int *perror) 550 { 551 char storage[64]; 552 struct weightened_nhop *pnhops; 553 struct nhgrp_priv *nhg_priv; 554 const struct nhgrp_priv *src_priv; 555 size_t sz; 556 int curr_nhops; 557 558 src_priv = NHGRP_PRIV_CONST(gr_orig); 559 curr_nhops = src_priv->nhg_nh_count; 560 561 *perror = 0; 562 563 sz = (src_priv->nhg_nh_count + num_nhops) * (sizeof(struct weightened_nhop)); 564 /* optimize for <= 4 paths, each path=16 bytes */ 565 if (sz <= sizeof(storage)) 566 pnhops = (struct weightened_nhop *)&storage[0]; 567 else { 568 pnhops = malloc(sz, M_TEMP, M_NOWAIT); 569 if (pnhops == NULL) { 570 *perror = ENOMEM; 571 return (NULL); 572 } 573 } 574 575 /* Copy nhops from original group first */ 576 memcpy(pnhops, src_priv->nhg_nh_weights, 577 curr_nhops * sizeof(struct weightened_nhop)); 578 memcpy(&pnhops[curr_nhops], wn, num_nhops * sizeof(struct weightened_nhop)); 579 curr_nhops += num_nhops; 580 581 nhg_priv = get_nhgrp(ctl, pnhops, curr_nhops, 0, perror); 582 583 if (pnhops != (struct weightened_nhop *)&storage[0]) 584 free(pnhops, M_TEMP); 585 586 if (nhg_priv == NULL) 587 return (NULL); 588 589 return (nhg_priv); 590 } 591 592 593 /* 594 * Creates/finds nexthop group based on @wn and @num_nhops. 595 * Returns 0 on success with referenced group in @rnd, or 596 * errno. 597 * 598 * If the error is EAGAIN, then the operation can be retried. 599 */ 600 int 601 nhgrp_get_group(struct rib_head *rh, struct weightened_nhop *wn, int num_nhops, 602 uint32_t uidx, struct nhgrp_object **pnhg) 603 { 604 struct nh_control *ctl = rh->nh_control; 605 struct nhgrp_priv *nhg_priv; 606 int error; 607 608 nhg_priv = get_nhgrp(ctl, wn, num_nhops, uidx, &error); 609 if (nhg_priv != NULL) 610 *pnhg = nhg_priv->nhg; 611 612 return (error); 613 } 614 615 /* 616 * Creates new nexthop group based on @src group without the nexthops 617 * chosen by @flt_func. 618 * Returns 0 on success, storring the reference nhop group/object in @rnd. 619 */ 620 int 621 nhgrp_get_filtered_group(struct rib_head *rh, const struct rtentry *rt, 622 const struct nhgrp_object *src, rib_filter_f_t flt_func, void *flt_data, 623 struct route_nhop_data *rnd) 624 { 625 char storage[64]; 626 struct nh_control *ctl = rh->nh_control; 627 struct weightened_nhop *pnhops; 628 const struct nhgrp_priv *mp_priv, *src_priv; 629 size_t sz; 630 int error, i, num_nhops; 631 632 src_priv = NHGRP_PRIV_CONST(src); 633 634 sz = src_priv->nhg_nh_count * (sizeof(struct weightened_nhop)); 635 /* optimize for <= 4 paths, each path=16 bytes */ 636 if (sz <= sizeof(storage)) 637 pnhops = (struct weightened_nhop *)&storage[0]; 638 else { 639 if ((pnhops = malloc(sz, M_TEMP, M_NOWAIT)) == NULL) 640 return (ENOMEM); 641 } 642 643 /* Filter nexthops */ 644 error = 0; 645 num_nhops = 0; 646 for (i = 0; i < src_priv->nhg_nh_count; i++) { 647 if (flt_func(rt, src_priv->nhg_nh_weights[i].nh, flt_data)) 648 continue; 649 memcpy(&pnhops[num_nhops++], &src_priv->nhg_nh_weights[i], 650 sizeof(struct weightened_nhop)); 651 } 652 653 if (num_nhops == 0) { 654 rnd->rnd_nhgrp = NULL; 655 rnd->rnd_weight = 0; 656 } else if (num_nhops == 1) { 657 rnd->rnd_nhop = pnhops[0].nh; 658 rnd->rnd_weight = pnhops[0].weight; 659 if (nhop_try_ref_object(rnd->rnd_nhop) == 0) 660 error = EAGAIN; 661 } else { 662 mp_priv = get_nhgrp(ctl, pnhops, num_nhops, 0, &error); 663 if (mp_priv != NULL) 664 rnd->rnd_nhgrp = mp_priv->nhg; 665 rnd->rnd_weight = 0; 666 } 667 668 if (pnhops != (struct weightened_nhop *)&storage[0]) 669 free(pnhops, M_TEMP); 670 671 return (error); 672 } 673 674 /* 675 * Creates new multipath group based on existing group/nhop in @rnd_orig and 676 * to-be-added nhop @wn_add. 677 * Returns 0 on success and stores result in @rnd_new. 678 */ 679 int 680 nhgrp_get_addition_group(struct rib_head *rh, struct route_nhop_data *rnd_orig, 681 struct route_nhop_data *rnd_add, struct route_nhop_data *rnd_new) 682 { 683 struct nh_control *ctl = rh->nh_control; 684 struct nhgrp_priv *nhg_priv; 685 struct weightened_nhop wn[2] = {}; 686 int error; 687 688 if (rnd_orig->rnd_nhop == NULL) { 689 /* No paths to add to, just reference current nhop */ 690 *rnd_new = *rnd_add; 691 if (nhop_try_ref_object(rnd_new->rnd_nhop) == 0) 692 return (EAGAIN); 693 return (0); 694 } 695 696 wn[0].nh = rnd_add->rnd_nhop; 697 wn[0].weight = rnd_add->rnd_weight; 698 699 if (!NH_IS_NHGRP(rnd_orig->rnd_nhop)) { 700 /* Simple merge of 2 non-multipath nexthops */ 701 wn[1].nh = rnd_orig->rnd_nhop; 702 wn[1].weight = rnd_orig->rnd_weight; 703 nhg_priv = get_nhgrp(ctl, wn, 2, 0, &error); 704 } else { 705 /* Get new nhop group with @rt->rt_nhop as an additional nhop */ 706 nhg_priv = append_nhops(ctl, rnd_orig->rnd_nhgrp, &wn[0], 1, 707 &error); 708 } 709 710 if (nhg_priv == NULL) 711 return (error); 712 rnd_new->rnd_nhgrp = nhg_priv->nhg; 713 rnd_new->rnd_weight = 0; 714 715 return (0); 716 } 717 718 /* 719 * Returns pointer to array of nexthops with weights for 720 * given @nhg. Stores number of items in the array into @pnum_nhops. 721 */ 722 const struct weightened_nhop * 723 nhgrp_get_nhops(const struct nhgrp_object *nhg, uint32_t *pnum_nhops) 724 { 725 const struct nhgrp_priv *nhg_priv; 726 727 KASSERT(((nhg->nhg_flags & MPF_MULTIPATH) != 0), ("nhop is not mpath")); 728 729 nhg_priv = NHGRP_PRIV_CONST(nhg); 730 *pnum_nhops = nhg_priv->nhg_nh_count; 731 732 return (nhg_priv->nhg_nh_weights); 733 } 734 735 uint32_t 736 nhgrp_get_uidx(const struct nhgrp_object *nhg) 737 { 738 const struct nhgrp_priv *nhg_priv; 739 740 KASSERT(((nhg->nhg_flags & MPF_MULTIPATH) != 0), ("nhop is not mpath")); 741 742 nhg_priv = NHGRP_PRIV_CONST(nhg); 743 return (nhg_priv->nhg_uidx); 744 } 745 746 /* 747 * Prints nexhop group @nhg data in the provided @buf. 748 * Example: nhg#33/sz=3:[#1:100,#2:100,#3:100] 749 * Example: nhg#33/sz=5:[#1:100,#2:100,..] 750 */ 751 char * 752 nhgrp_print_buf(const struct nhgrp_object *nhg, char *buf, size_t bufsize) 753 { 754 const struct nhgrp_priv *nhg_priv = NHGRP_PRIV_CONST(nhg); 755 756 int off = snprintf(buf, bufsize, "nhg#%u/sz=%u:[", nhg_priv->nhg_idx, 757 nhg_priv->nhg_nh_count); 758 759 for (int i = 0; i < nhg_priv->nhg_nh_count; i++) { 760 const struct weightened_nhop *wn = &nhg_priv->nhg_nh_weights[i]; 761 int len = snprintf(&buf[off], bufsize - off, "#%u:%u,", 762 wn->nh->nh_priv->nh_idx, wn->weight); 763 if (len + off + 3 >= bufsize) { 764 int len = snprintf(&buf[off], bufsize - off, "..."); 765 off += len; 766 break; 767 } 768 off += len; 769 } 770 if (off > 0) 771 off--; // remove last "," 772 if (off + 1 < bufsize) 773 snprintf(&buf[off], bufsize - off, "]"); 774 return buf; 775 } 776 777 __noinline static int 778 dump_nhgrp_entry(struct rib_head *rh, const struct nhgrp_priv *nhg_priv, 779 char *buffer, size_t buffer_size, struct sysctl_req *w) 780 { 781 struct rt_msghdr *rtm; 782 struct nhgrp_external *nhge; 783 struct nhgrp_container *nhgc; 784 const struct nhgrp_object *nhg; 785 struct nhgrp_nhop_external *ext; 786 int error; 787 size_t sz; 788 789 nhg = nhg_priv->nhg; 790 791 sz = sizeof(struct rt_msghdr) + sizeof(struct nhgrp_external); 792 /* controlplane nexthops */ 793 sz += sizeof(struct nhgrp_container); 794 sz += sizeof(struct nhgrp_nhop_external) * nhg_priv->nhg_nh_count; 795 /* dataplane nexthops */ 796 sz += sizeof(struct nhgrp_container); 797 sz += sizeof(struct nhgrp_nhop_external) * nhg->nhg_size; 798 799 KASSERT(sz <= buffer_size, ("increase nhgrp buffer size")); 800 801 bzero(buffer, sz); 802 803 rtm = (struct rt_msghdr *)buffer; 804 rtm->rtm_msglen = sz; 805 rtm->rtm_version = RTM_VERSION; 806 rtm->rtm_type = RTM_GET; 807 808 nhge = (struct nhgrp_external *)(rtm + 1); 809 810 nhge->nhg_idx = nhg_priv->nhg_idx; 811 nhge->nhg_refcount = nhg_priv->nhg_refcount; 812 813 /* fill in control plane nexthops firs */ 814 nhgc = (struct nhgrp_container *)(nhge + 1); 815 nhgc->nhgc_type = NHG_C_TYPE_CNHOPS; 816 nhgc->nhgc_subtype = 0; 817 nhgc->nhgc_len = sizeof(struct nhgrp_container); 818 nhgc->nhgc_len += sizeof(struct nhgrp_nhop_external) * nhg_priv->nhg_nh_count; 819 nhgc->nhgc_count = nhg_priv->nhg_nh_count; 820 821 ext = (struct nhgrp_nhop_external *)(nhgc + 1); 822 for (int i = 0; i < nhg_priv->nhg_nh_count; i++) { 823 ext[i].nh_idx = nhg_priv->nhg_nh_weights[i].nh->nh_priv->nh_idx; 824 ext[i].nh_weight = nhg_priv->nhg_nh_weights[i].weight; 825 } 826 827 /* fill in dataplane nexthops */ 828 nhgc = (struct nhgrp_container *)(&ext[nhg_priv->nhg_nh_count]); 829 nhgc->nhgc_type = NHG_C_TYPE_DNHOPS; 830 nhgc->nhgc_subtype = 0; 831 nhgc->nhgc_len = sizeof(struct nhgrp_container); 832 nhgc->nhgc_len += sizeof(struct nhgrp_nhop_external) * nhg->nhg_size; 833 nhgc->nhgc_count = nhg->nhg_size; 834 835 ext = (struct nhgrp_nhop_external *)(nhgc + 1); 836 for (int i = 0; i < nhg->nhg_size; i++) { 837 ext[i].nh_idx = nhg->nhops[i]->nh_priv->nh_idx; 838 ext[i].nh_weight = 0; 839 } 840 841 error = SYSCTL_OUT(w, buffer, sz); 842 843 return (error); 844 } 845 846 uint32_t 847 nhgrp_get_idx(const struct nhgrp_object *nhg) 848 { 849 const struct nhgrp_priv *nhg_priv; 850 851 nhg_priv = NHGRP_PRIV_CONST(nhg); 852 return (nhg_priv->nhg_idx); 853 } 854 855 uint8_t 856 nhgrp_get_origin(const struct nhgrp_object *nhg) 857 { 858 return (NHGRP_PRIV_CONST(nhg)->nhg_origin); 859 } 860 861 void 862 nhgrp_set_origin(struct nhgrp_object *nhg, uint8_t origin) 863 { 864 NHGRP_PRIV(nhg)->nhg_origin = origin; 865 } 866 867 uint32_t 868 nhgrp_get_count(struct rib_head *rh) 869 { 870 struct nh_control *ctl; 871 uint32_t count; 872 873 ctl = rh->nh_control; 874 875 NHOPS_RLOCK(ctl); 876 count = ctl->gr_head.items_count; 877 NHOPS_RUNLOCK(ctl); 878 879 return (count); 880 } 881 882 int 883 nhgrp_dump_sysctl(struct rib_head *rh, struct sysctl_req *w) 884 { 885 struct nh_control *ctl = rh->nh_control; 886 struct epoch_tracker et; 887 struct nhgrp_priv *nhg_priv; 888 char *buffer; 889 size_t sz; 890 int error = 0; 891 892 if (ctl->gr_head.items_count == 0) 893 return (0); 894 895 /* Calculate the maximum nhop group size in bytes */ 896 sz = sizeof(struct rt_msghdr) + sizeof(struct nhgrp_external); 897 sz += 2 * sizeof(struct nhgrp_container); 898 sz += 2 * sizeof(struct nhgrp_nhop_external) * RIB_MAX_MPATH_WIDTH; 899 buffer = malloc(sz, M_TEMP, M_NOWAIT); 900 if (buffer == NULL) 901 return (ENOMEM); 902 903 NET_EPOCH_ENTER(et); 904 NHOPS_RLOCK(ctl); 905 CHT_SLIST_FOREACH(&ctl->gr_head, mpath, nhg_priv) { 906 error = dump_nhgrp_entry(rh, nhg_priv, buffer, sz, w); 907 if (error != 0) 908 break; 909 } CHT_SLIST_FOREACH_END; 910 NHOPS_RUNLOCK(ctl); 911 NET_EPOCH_EXIT(et); 912 913 free(buffer, M_TEMP); 914 915 return (error); 916 } 917