1// SPDX-License-Identifier: GPL-2.0 2/* XDP sockets 3 * 4 * AF_XDP sockets allows a channel between XDP programs and userspace 5 * applications. 6 * Copyright(c) 2018 Intel Corporation. 7 * 8 * Author(s): Björn Töpel <bjorn.topel@intel.com> 9 * Magnus Karlsson <magnus.karlsson@intel.com> 10 */ 11 12#define pr_fmt(fmt) "AF_XDP: %s: " fmt, __func__ 13 14#include <linux/if_xdp.h> 15#include <linux/init.h> 16#include <linux/sched/mm.h> 17#include <linux/sched/signal.h> 18#include <linux/sched/task.h> 19#include <linux/socket.h> 20#include <linux/file.h> 21#include <linux/uaccess.h> 22#include <linux/net.h> 23#include <linux/netdevice.h> 24#include <linux/rculist.h> 25#include <linux/vmalloc.h> 26#include <net/xdp_sock_drv.h> 27#include <net/busy_poll.h> 28#include <net/netdev_rx_queue.h> 29#include <net/xdp.h> 30 31#include "xsk_queue.h" 32#include "xdp_umem.h" 33#include "xsk.h" 34 35#define TX_BATCH_SIZE 32 36 37static DEFINE_PER_CPU(struct list_head, xskmap_flush_list); 38 39void xsk_set_rx_need_wakeup(struct xsk_buff_pool *pool) 40{ 41 if (pool->cached_need_wakeup & XDP_WAKEUP_RX) 42 return; 43 44 pool->fq->ring->flags |= XDP_RING_NEED_WAKEUP; 45 pool->cached_need_wakeup |= XDP_WAKEUP_RX; 46} 47EXPORT_SYMBOL(xsk_set_rx_need_wakeup); 48 49void xsk_set_tx_need_wakeup(struct xsk_buff_pool *pool) 50{ 51 struct xdp_sock *xs; 52 53 if (pool->cached_need_wakeup & XDP_WAKEUP_TX) 54 return; 55 56 rcu_read_lock(); 57 list_for_each_entry_rcu(xs, &pool->xsk_tx_list, tx_list) { 58 xs->tx->ring->flags |= XDP_RING_NEED_WAKEUP; 59 } 60 rcu_read_unlock(); 61 62 pool->cached_need_wakeup |= XDP_WAKEUP_TX; 63} 64EXPORT_SYMBOL(xsk_set_tx_need_wakeup); 65 66void xsk_clear_rx_need_wakeup(struct xsk_buff_pool *pool) 67{ 68 if (!(pool->cached_need_wakeup & XDP_WAKEUP_RX)) 69 return; 70 71 pool->fq->ring->flags &= ~XDP_RING_NEED_WAKEUP; 72 pool->cached_need_wakeup &= ~XDP_WAKEUP_RX; 73} 74EXPORT_SYMBOL(xsk_clear_rx_need_wakeup); 75 76void xsk_clear_tx_need_wakeup(struct xsk_buff_pool *pool) 77{ 78 struct xdp_sock *xs; 79 80 if (!(pool->cached_need_wakeup & XDP_WAKEUP_TX)) 81 return; 82 83 rcu_read_lock(); 84 list_for_each_entry_rcu(xs, &pool->xsk_tx_list, tx_list) { 85 xs->tx->ring->flags &= ~XDP_RING_NEED_WAKEUP; 86 } 87 rcu_read_unlock(); 88 89 pool->cached_need_wakeup &= ~XDP_WAKEUP_TX; 90} 91EXPORT_SYMBOL(xsk_clear_tx_need_wakeup); 92 93bool xsk_uses_need_wakeup(struct xsk_buff_pool *pool) 94{ 95 return pool->uses_need_wakeup; 96} 97EXPORT_SYMBOL(xsk_uses_need_wakeup); 98 99struct xsk_buff_pool *xsk_get_pool_from_qid(struct net_device *dev, 100 u16 queue_id) 101{ 102 if (queue_id < dev->real_num_rx_queues) 103 return dev->_rx[queue_id].pool; 104 if (queue_id < dev->real_num_tx_queues) 105 return dev->_tx[queue_id].pool; 106 107 return NULL; 108} 109EXPORT_SYMBOL(xsk_get_pool_from_qid); 110 111void xsk_clear_pool_at_qid(struct net_device *dev, u16 queue_id) 112{ 113 if (queue_id < dev->num_rx_queues) 114 dev->_rx[queue_id].pool = NULL; 115 if (queue_id < dev->num_tx_queues) 116 dev->_tx[queue_id].pool = NULL; 117} 118 119/* The buffer pool is stored both in the _rx struct and the _tx struct as we do 120 * not know if the device has more tx queues than rx, or the opposite. 121 * This might also change during run time. 122 */ 123int xsk_reg_pool_at_qid(struct net_device *dev, struct xsk_buff_pool *pool, 124 u16 queue_id) 125{ 126 if (queue_id >= max_t(unsigned int, 127 dev->real_num_rx_queues, 128 dev->real_num_tx_queues)) 129 return -EINVAL; 130 131 if (queue_id < dev->real_num_rx_queues) 132 dev->_rx[queue_id].pool = pool; 133 if (queue_id < dev->real_num_tx_queues) 134 dev->_tx[queue_id].pool = pool; 135 136 return 0; 137} 138 139static int __xsk_rcv_zc(struct xdp_sock *xs, struct xdp_buff_xsk *xskb, u32 len, 140 u32 flags) 141{ 142 u64 addr; 143 int err; 144 145 addr = xp_get_handle(xskb); 146 err = xskq_prod_reserve_desc(xs->rx, addr, len, flags); 147 if (err) { 148 xs->rx_queue_full++; 149 return err; 150 } 151 152 xp_release(xskb); 153 return 0; 154} 155 156static int xsk_rcv_zc(struct xdp_sock *xs, struct xdp_buff *xdp, u32 len) 157{ 158 struct xdp_buff_xsk *xskb = container_of(xdp, struct xdp_buff_xsk, xdp); 159 u32 frags = xdp_buff_has_frags(xdp); 160 struct xdp_buff_xsk *pos, *tmp; 161 struct list_head *xskb_list; 162 u32 contd = 0; 163 int err; 164 165 if (frags) 166 contd = XDP_PKT_CONTD; 167 168 err = __xsk_rcv_zc(xs, xskb, len, contd); 169 if (err) 170 goto err; 171 if (likely(!frags)) 172 return 0; 173 174 xskb_list = &xskb->pool->xskb_list; 175 list_for_each_entry_safe(pos, tmp, xskb_list, xskb_list_node) { 176 if (list_is_singular(xskb_list)) 177 contd = 0; 178 len = pos->xdp.data_end - pos->xdp.data; 179 err = __xsk_rcv_zc(xs, pos, len, contd); 180 if (err) 181 goto err; 182 list_del(&pos->xskb_list_node); 183 } 184 185 return 0; 186err: 187 xsk_buff_free(xdp); 188 return err; 189} 190 191static void *xsk_copy_xdp_start(struct xdp_buff *from) 192{ 193 if (unlikely(xdp_data_meta_unsupported(from))) 194 return from->data; 195 else 196 return from->data_meta; 197} 198 199static u32 xsk_copy_xdp(void *to, void **from, u32 to_len, 200 u32 *from_len, skb_frag_t **frag, u32 rem) 201{ 202 u32 copied = 0; 203 204 while (1) { 205 u32 copy_len = min_t(u32, *from_len, to_len); 206 207 memcpy(to, *from, copy_len); 208 copied += copy_len; 209 if (rem == copied) 210 return copied; 211 212 if (*from_len == copy_len) { 213 *from = skb_frag_address(*frag); 214 *from_len = skb_frag_size((*frag)++); 215 } else { 216 *from += copy_len; 217 *from_len -= copy_len; 218 } 219 if (to_len == copy_len) 220 return copied; 221 222 to_len -= copy_len; 223 to += copy_len; 224 } 225} 226 227static int __xsk_rcv(struct xdp_sock *xs, struct xdp_buff *xdp, u32 len) 228{ 229 u32 frame_size = xsk_pool_get_rx_frame_size(xs->pool); 230 void *copy_from = xsk_copy_xdp_start(xdp), *copy_to; 231 u32 from_len, meta_len, rem, num_desc; 232 struct xdp_buff_xsk *xskb; 233 struct xdp_buff *xsk_xdp; 234 skb_frag_t *frag; 235 236 from_len = xdp->data_end - copy_from; 237 meta_len = xdp->data - copy_from; 238 rem = len + meta_len; 239 240 if (len <= frame_size && !xdp_buff_has_frags(xdp)) { 241 int err; 242 243 xsk_xdp = xsk_buff_alloc(xs->pool); 244 if (!xsk_xdp) { 245 xs->rx_dropped++; 246 return -ENOMEM; 247 } 248 memcpy(xsk_xdp->data - meta_len, copy_from, rem); 249 xskb = container_of(xsk_xdp, struct xdp_buff_xsk, xdp); 250 err = __xsk_rcv_zc(xs, xskb, len, 0); 251 if (err) { 252 xsk_buff_free(xsk_xdp); 253 return err; 254 } 255 256 return 0; 257 } 258 259 num_desc = (len - 1) / frame_size + 1; 260 261 if (!xsk_buff_can_alloc(xs->pool, num_desc)) { 262 xs->rx_dropped++; 263 return -ENOMEM; 264 } 265 if (xskq_prod_nb_free(xs->rx, num_desc) < num_desc) { 266 xs->rx_queue_full++; 267 return -ENOBUFS; 268 } 269 270 if (xdp_buff_has_frags(xdp)) { 271 struct skb_shared_info *sinfo; 272 273 sinfo = xdp_get_shared_info_from_buff(xdp); 274 frag = &sinfo->frags[0]; 275 } 276 277 do { 278 u32 to_len = frame_size + meta_len; 279 u32 copied; 280 281 xsk_xdp = xsk_buff_alloc(xs->pool); 282 copy_to = xsk_xdp->data - meta_len; 283 284 copied = xsk_copy_xdp(copy_to, ©_from, to_len, &from_len, &frag, rem); 285 rem -= copied; 286 287 xskb = container_of(xsk_xdp, struct xdp_buff_xsk, xdp); 288 __xsk_rcv_zc(xs, xskb, copied - meta_len, rem ? XDP_PKT_CONTD : 0); 289 meta_len = 0; 290 } while (rem); 291 292 return 0; 293} 294 295static bool xsk_tx_writeable(struct xdp_sock *xs) 296{ 297 if (xskq_cons_present_entries(xs->tx) > xs->tx->nentries / 2) 298 return false; 299 300 return true; 301} 302 303static bool xsk_is_bound(struct xdp_sock *xs) 304{ 305 if (READ_ONCE(xs->state) == XSK_BOUND) { 306 /* Matches smp_wmb() in bind(). */ 307 smp_rmb(); 308 return true; 309 } 310 return false; 311} 312 313static int xsk_rcv_check(struct xdp_sock *xs, struct xdp_buff *xdp, u32 len) 314{ 315 if (!xsk_is_bound(xs)) 316 return -ENXIO; 317 318 if (xs->dev != xdp->rxq->dev || xs->queue_id != xdp->rxq->queue_index) 319 return -EINVAL; 320 321 if (len > xsk_pool_get_rx_frame_size(xs->pool) && !xs->sg) { 322 xs->rx_dropped++; 323 return -ENOSPC; 324 } 325 326 sk_mark_napi_id_once_xdp(&xs->sk, xdp); 327 return 0; 328} 329 330static void xsk_flush(struct xdp_sock *xs) 331{ 332 xskq_prod_submit(xs->rx); 333 __xskq_cons_release(xs->pool->fq); 334 sock_def_readable(&xs->sk); 335} 336 337int xsk_generic_rcv(struct xdp_sock *xs, struct xdp_buff *xdp) 338{ 339 u32 len = xdp_get_buff_len(xdp); 340 int err; 341 342 spin_lock_bh(&xs->rx_lock); 343 err = xsk_rcv_check(xs, xdp, len); 344 if (!err) { 345 err = __xsk_rcv(xs, xdp, len); 346 xsk_flush(xs); 347 } 348 spin_unlock_bh(&xs->rx_lock); 349 return err; 350} 351 352static int xsk_rcv(struct xdp_sock *xs, struct xdp_buff *xdp) 353{ 354 u32 len = xdp_get_buff_len(xdp); 355 int err; 356 357 err = xsk_rcv_check(xs, xdp, len); 358 if (err) 359 return err; 360 361 if (xdp->rxq->mem.type == MEM_TYPE_XSK_BUFF_POOL) { 362 len = xdp->data_end - xdp->data; 363 return xsk_rcv_zc(xs, xdp, len); 364 } 365 366 err = __xsk_rcv(xs, xdp, len); 367 if (!err) 368 xdp_return_buff(xdp); 369 return err; 370} 371 372int __xsk_map_redirect(struct xdp_sock *xs, struct xdp_buff *xdp) 373{ 374 struct list_head *flush_list = this_cpu_ptr(&xskmap_flush_list); 375 int err; 376 377 err = xsk_rcv(xs, xdp); 378 if (err) 379 return err; 380 381 if (!xs->flush_node.prev) 382 list_add(&xs->flush_node, flush_list); 383 384 return 0; 385} 386 387void __xsk_map_flush(void) 388{ 389 struct list_head *flush_list = this_cpu_ptr(&xskmap_flush_list); 390 struct xdp_sock *xs, *tmp; 391 392 list_for_each_entry_safe(xs, tmp, flush_list, flush_node) { 393 xsk_flush(xs); 394 __list_del_clearprev(&xs->flush_node); 395 } 396} 397 398void xsk_tx_completed(struct xsk_buff_pool *pool, u32 nb_entries) 399{ 400 xskq_prod_submit_n(pool->cq, nb_entries); 401} 402EXPORT_SYMBOL(xsk_tx_completed); 403 404void xsk_tx_release(struct xsk_buff_pool *pool) 405{ 406 struct xdp_sock *xs; 407 408 rcu_read_lock(); 409 list_for_each_entry_rcu(xs, &pool->xsk_tx_list, tx_list) { 410 __xskq_cons_release(xs->tx); 411 if (xsk_tx_writeable(xs)) 412 xs->sk.sk_write_space(&xs->sk); 413 } 414 rcu_read_unlock(); 415} 416EXPORT_SYMBOL(xsk_tx_release); 417 418bool xsk_tx_peek_desc(struct xsk_buff_pool *pool, struct xdp_desc *desc) 419{ 420 struct xdp_sock *xs; 421 422 rcu_read_lock(); 423 list_for_each_entry_rcu(xs, &pool->xsk_tx_list, tx_list) { 424 if (!xskq_cons_peek_desc(xs->tx, desc, pool)) { 425 if (xskq_has_descs(xs->tx)) 426 xskq_cons_release(xs->tx); 427 continue; 428 } 429 430 /* This is the backpressure mechanism for the Tx path. 431 * Reserve space in the completion queue and only proceed 432 * if there is space in it. This avoids having to implement 433 * any buffering in the Tx path. 434 */ 435 if (xskq_prod_reserve_addr(pool->cq, desc->addr)) 436 goto out; 437 438 xskq_cons_release(xs->tx); 439 rcu_read_unlock(); 440 return true; 441 } 442 443out: 444 rcu_read_unlock(); 445 return false; 446} 447EXPORT_SYMBOL(xsk_tx_peek_desc); 448 449static u32 xsk_tx_peek_release_fallback(struct xsk_buff_pool *pool, u32 max_entries) 450{ 451 struct xdp_desc *descs = pool->tx_descs; 452 u32 nb_pkts = 0; 453 454 while (nb_pkts < max_entries && xsk_tx_peek_desc(pool, &descs[nb_pkts])) 455 nb_pkts++; 456 457 xsk_tx_release(pool); 458 return nb_pkts; 459} 460 461u32 xsk_tx_peek_release_desc_batch(struct xsk_buff_pool *pool, u32 nb_pkts) 462{ 463 struct xdp_sock *xs; 464 465 rcu_read_lock(); 466 if (!list_is_singular(&pool->xsk_tx_list)) { 467 /* Fallback to the non-batched version */ 468 rcu_read_unlock(); 469 return xsk_tx_peek_release_fallback(pool, nb_pkts); 470 } 471 472 xs = list_first_or_null_rcu(&pool->xsk_tx_list, struct xdp_sock, tx_list); 473 if (!xs) { 474 nb_pkts = 0; 475 goto out; 476 } 477 478 nb_pkts = xskq_cons_nb_entries(xs->tx, nb_pkts); 479 480 /* This is the backpressure mechanism for the Tx path. Try to 481 * reserve space in the completion queue for all packets, but 482 * if there are fewer slots available, just process that many 483 * packets. This avoids having to implement any buffering in 484 * the Tx path. 485 */ 486 nb_pkts = xskq_prod_nb_free(pool->cq, nb_pkts); 487 if (!nb_pkts) 488 goto out; 489 490 nb_pkts = xskq_cons_read_desc_batch(xs->tx, pool, nb_pkts); 491 if (!nb_pkts) { 492 xs->tx->queue_empty_descs++; 493 goto out; 494 } 495 496 __xskq_cons_release(xs->tx); 497 xskq_prod_write_addr_batch(pool->cq, pool->tx_descs, nb_pkts); 498 xs->sk.sk_write_space(&xs->sk); 499 500out: 501 rcu_read_unlock(); 502 return nb_pkts; 503} 504EXPORT_SYMBOL(xsk_tx_peek_release_desc_batch); 505 506static int xsk_wakeup(struct xdp_sock *xs, u8 flags) 507{ 508 struct net_device *dev = xs->dev; 509 510 return dev->netdev_ops->ndo_xsk_wakeup(dev, xs->queue_id, flags); 511} 512 513static int xsk_cq_reserve_addr_locked(struct xdp_sock *xs, u64 addr) 514{ 515 unsigned long flags; 516 int ret; 517 518 spin_lock_irqsave(&xs->pool->cq_lock, flags); 519 ret = xskq_prod_reserve_addr(xs->pool->cq, addr); 520 spin_unlock_irqrestore(&xs->pool->cq_lock, flags); 521 522 return ret; 523} 524 525static void xsk_cq_submit_locked(struct xdp_sock *xs, u32 n) 526{ 527 unsigned long flags; 528 529 spin_lock_irqsave(&xs->pool->cq_lock, flags); 530 xskq_prod_submit_n(xs->pool->cq, n); 531 spin_unlock_irqrestore(&xs->pool->cq_lock, flags); 532} 533 534static void xsk_cq_cancel_locked(struct xdp_sock *xs, u32 n) 535{ 536 unsigned long flags; 537 538 spin_lock_irqsave(&xs->pool->cq_lock, flags); 539 xskq_prod_cancel_n(xs->pool->cq, n); 540 spin_unlock_irqrestore(&xs->pool->cq_lock, flags); 541} 542 543static u32 xsk_get_num_desc(struct sk_buff *skb) 544{ 545 return skb ? (long)skb_shinfo(skb)->destructor_arg : 0; 546} 547 548static void xsk_destruct_skb(struct sk_buff *skb) 549{ 550 xsk_cq_submit_locked(xdp_sk(skb->sk), xsk_get_num_desc(skb)); 551 sock_wfree(skb); 552} 553 554static void xsk_set_destructor_arg(struct sk_buff *skb) 555{ 556 long num = xsk_get_num_desc(xdp_sk(skb->sk)->skb) + 1; 557 558 skb_shinfo(skb)->destructor_arg = (void *)num; 559} 560 561static void xsk_consume_skb(struct sk_buff *skb) 562{ 563 struct xdp_sock *xs = xdp_sk(skb->sk); 564 565 skb->destructor = sock_wfree; 566 xsk_cq_cancel_locked(xs, xsk_get_num_desc(skb)); 567 /* Free skb without triggering the perf drop trace */ 568 consume_skb(skb); 569 xs->skb = NULL; 570} 571 572static void xsk_drop_skb(struct sk_buff *skb) 573{ 574 xdp_sk(skb->sk)->tx->invalid_descs += xsk_get_num_desc(skb); 575 xsk_consume_skb(skb); 576} 577 578static struct sk_buff *xsk_build_skb_zerocopy(struct xdp_sock *xs, 579 struct xdp_desc *desc) 580{ 581 struct xsk_buff_pool *pool = xs->pool; 582 u32 hr, len, ts, offset, copy, copied; 583 struct sk_buff *skb = xs->skb; 584 struct page *page; 585 void *buffer; 586 int err, i; 587 u64 addr; 588 589 if (!skb) { 590 hr = max(NET_SKB_PAD, L1_CACHE_ALIGN(xs->dev->needed_headroom)); 591 592 skb = sock_alloc_send_skb(&xs->sk, hr, 1, &err); 593 if (unlikely(!skb)) 594 return ERR_PTR(err); 595 596 skb_reserve(skb, hr); 597 } 598 599 addr = desc->addr; 600 len = desc->len; 601 ts = pool->unaligned ? len : pool->chunk_size; 602 603 buffer = xsk_buff_raw_get_data(pool, addr); 604 offset = offset_in_page(buffer); 605 addr = buffer - pool->addrs; 606 607 for (copied = 0, i = skb_shinfo(skb)->nr_frags; copied < len; i++) { 608 if (unlikely(i >= MAX_SKB_FRAGS)) 609 return ERR_PTR(-EOVERFLOW); 610 611 page = pool->umem->pgs[addr >> PAGE_SHIFT]; 612 get_page(page); 613 614 copy = min_t(u32, PAGE_SIZE - offset, len - copied); 615 skb_fill_page_desc(skb, i, page, offset, copy); 616 617 copied += copy; 618 addr += copy; 619 offset = 0; 620 } 621 622 skb->len += len; 623 skb->data_len += len; 624 skb->truesize += ts; 625 626 refcount_add(ts, &xs->sk.sk_wmem_alloc); 627 628 return skb; 629} 630 631static struct sk_buff *xsk_build_skb(struct xdp_sock *xs, 632 struct xdp_desc *desc) 633{ 634 struct net_device *dev = xs->dev; 635 struct sk_buff *skb = xs->skb; 636 int err; 637 638 if (dev->priv_flags & IFF_TX_SKB_NO_LINEAR) { 639 skb = xsk_build_skb_zerocopy(xs, desc); 640 if (IS_ERR(skb)) { 641 err = PTR_ERR(skb); 642 goto free_err; 643 } 644 } else { 645 u32 hr, tr, len; 646 void *buffer; 647 648 buffer = xsk_buff_raw_get_data(xs->pool, desc->addr); 649 len = desc->len; 650 651 if (!skb) { 652 hr = max(NET_SKB_PAD, L1_CACHE_ALIGN(dev->needed_headroom)); 653 tr = dev->needed_tailroom; 654 skb = sock_alloc_send_skb(&xs->sk, hr + len + tr, 1, &err); 655 if (unlikely(!skb)) 656 goto free_err; 657 658 skb_reserve(skb, hr); 659 skb_put(skb, len); 660 661 err = skb_store_bits(skb, 0, buffer, len); 662 if (unlikely(err)) { 663 kfree_skb(skb); 664 goto free_err; 665 } 666 } else { 667 int nr_frags = skb_shinfo(skb)->nr_frags; 668 struct page *page; 669 u8 *vaddr; 670 671 if (unlikely(nr_frags == (MAX_SKB_FRAGS - 1) && xp_mb_desc(desc))) { 672 err = -EOVERFLOW; 673 goto free_err; 674 } 675 676 page = alloc_page(xs->sk.sk_allocation); 677 if (unlikely(!page)) { 678 err = -EAGAIN; 679 goto free_err; 680 } 681 682 vaddr = kmap_local_page(page); 683 memcpy(vaddr, buffer, len); 684 kunmap_local(vaddr); 685 686 skb_add_rx_frag(skb, nr_frags, page, 0, len, PAGE_SIZE); 687 refcount_add(PAGE_SIZE, &xs->sk.sk_wmem_alloc); 688 } 689 } 690 691 skb->dev = dev; 692 skb->priority = xs->sk.sk_priority; 693 skb->mark = READ_ONCE(xs->sk.sk_mark); 694 skb->destructor = xsk_destruct_skb; 695 xsk_set_destructor_arg(skb); 696 697 return skb; 698 699free_err: 700 if (err == -EOVERFLOW) { 701 /* Drop the packet */ 702 xsk_set_destructor_arg(xs->skb); 703 xsk_drop_skb(xs->skb); 704 xskq_cons_release(xs->tx); 705 } else { 706 /* Let application retry */ 707 xsk_cq_cancel_locked(xs, 1); 708 } 709 710 return ERR_PTR(err); 711} 712 713static int __xsk_generic_xmit(struct sock *sk) 714{ 715 struct xdp_sock *xs = xdp_sk(sk); 716 u32 max_batch = TX_BATCH_SIZE; 717 bool sent_frame = false; 718 struct xdp_desc desc; 719 struct sk_buff *skb; 720 int err = 0; 721 722 mutex_lock(&xs->mutex); 723 724 /* Since we dropped the RCU read lock, the socket state might have changed. */ 725 if (unlikely(!xsk_is_bound(xs))) { 726 err = -ENXIO; 727 goto out; 728 } 729 730 if (xs->queue_id >= xs->dev->real_num_tx_queues) 731 goto out; 732 733 while (xskq_cons_peek_desc(xs->tx, &desc, xs->pool)) { 734 if (max_batch-- == 0) { 735 err = -EAGAIN; 736 goto out; 737 } 738 739 /* This is the backpressure mechanism for the Tx path. 740 * Reserve space in the completion queue and only proceed 741 * if there is space in it. This avoids having to implement 742 * any buffering in the Tx path. 743 */ 744 if (xsk_cq_reserve_addr_locked(xs, desc.addr)) 745 goto out; 746 747 skb = xsk_build_skb(xs, &desc); 748 if (IS_ERR(skb)) { 749 err = PTR_ERR(skb); 750 if (err != -EOVERFLOW) 751 goto out; 752 err = 0; 753 continue; 754 } 755 756 xskq_cons_release(xs->tx); 757 758 if (xp_mb_desc(&desc)) { 759 xs->skb = skb; 760 continue; 761 } 762 763 err = __dev_direct_xmit(skb, xs->queue_id); 764 if (err == NETDEV_TX_BUSY) { 765 /* Tell user-space to retry the send */ 766 xskq_cons_cancel_n(xs->tx, xsk_get_num_desc(skb)); 767 xsk_consume_skb(skb); 768 err = -EAGAIN; 769 goto out; 770 } 771 772 /* Ignore NET_XMIT_CN as packet might have been sent */ 773 if (err == NET_XMIT_DROP) { 774 /* SKB completed but not sent */ 775 err = -EBUSY; 776 xs->skb = NULL; 777 goto out; 778 } 779 780 sent_frame = true; 781 xs->skb = NULL; 782 } 783 784 if (xskq_has_descs(xs->tx)) { 785 if (xs->skb) 786 xsk_drop_skb(xs->skb); 787 xskq_cons_release(xs->tx); 788 } 789 790out: 791 if (sent_frame) 792 if (xsk_tx_writeable(xs)) 793 sk->sk_write_space(sk); 794 795 mutex_unlock(&xs->mutex); 796 return err; 797} 798 799static int xsk_generic_xmit(struct sock *sk) 800{ 801 int ret; 802 803 /* Drop the RCU lock since the SKB path might sleep. */ 804 rcu_read_unlock(); 805 ret = __xsk_generic_xmit(sk); 806 /* Reaquire RCU lock before going into common code. */ 807 rcu_read_lock(); 808 809 return ret; 810} 811 812static bool xsk_no_wakeup(struct sock *sk) 813{ 814#ifdef CONFIG_NET_RX_BUSY_POLL 815 /* Prefer busy-polling, skip the wakeup. */ 816 return READ_ONCE(sk->sk_prefer_busy_poll) && READ_ONCE(sk->sk_ll_usec) && 817 READ_ONCE(sk->sk_napi_id) >= MIN_NAPI_ID; 818#else 819 return false; 820#endif 821} 822 823static int xsk_check_common(struct xdp_sock *xs) 824{ 825 if (unlikely(!xsk_is_bound(xs))) 826 return -ENXIO; 827 if (unlikely(!(xs->dev->flags & IFF_UP))) 828 return -ENETDOWN; 829 830 return 0; 831} 832 833static int __xsk_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len) 834{ 835 bool need_wait = !(m->msg_flags & MSG_DONTWAIT); 836 struct sock *sk = sock->sk; 837 struct xdp_sock *xs = xdp_sk(sk); 838 struct xsk_buff_pool *pool; 839 int err; 840 841 err = xsk_check_common(xs); 842 if (err) 843 return err; 844 if (unlikely(need_wait)) 845 return -EOPNOTSUPP; 846 if (unlikely(!xs->tx)) 847 return -ENOBUFS; 848 849 if (sk_can_busy_loop(sk)) { 850 if (xs->zc) 851 __sk_mark_napi_id_once(sk, xsk_pool_get_napi_id(xs->pool)); 852 sk_busy_loop(sk, 1); /* only support non-blocking sockets */ 853 } 854 855 if (xs->zc && xsk_no_wakeup(sk)) 856 return 0; 857 858 pool = xs->pool; 859 if (pool->cached_need_wakeup & XDP_WAKEUP_TX) { 860 if (xs->zc) 861 return xsk_wakeup(xs, XDP_WAKEUP_TX); 862 return xsk_generic_xmit(sk); 863 } 864 return 0; 865} 866 867static int xsk_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len) 868{ 869 int ret; 870 871 rcu_read_lock(); 872 ret = __xsk_sendmsg(sock, m, total_len); 873 rcu_read_unlock(); 874 875 return ret; 876} 877 878static int __xsk_recvmsg(struct socket *sock, struct msghdr *m, size_t len, int flags) 879{ 880 bool need_wait = !(flags & MSG_DONTWAIT); 881 struct sock *sk = sock->sk; 882 struct xdp_sock *xs = xdp_sk(sk); 883 int err; 884 885 err = xsk_check_common(xs); 886 if (err) 887 return err; 888 if (unlikely(!xs->rx)) 889 return -ENOBUFS; 890 if (unlikely(need_wait)) 891 return -EOPNOTSUPP; 892 893 if (sk_can_busy_loop(sk)) 894 sk_busy_loop(sk, 1); /* only support non-blocking sockets */ 895 896 if (xsk_no_wakeup(sk)) 897 return 0; 898 899 if (xs->pool->cached_need_wakeup & XDP_WAKEUP_RX && xs->zc) 900 return xsk_wakeup(xs, XDP_WAKEUP_RX); 901 return 0; 902} 903 904static int xsk_recvmsg(struct socket *sock, struct msghdr *m, size_t len, int flags) 905{ 906 int ret; 907 908 rcu_read_lock(); 909 ret = __xsk_recvmsg(sock, m, len, flags); 910 rcu_read_unlock(); 911 912 return ret; 913} 914 915static __poll_t xsk_poll(struct file *file, struct socket *sock, 916 struct poll_table_struct *wait) 917{ 918 __poll_t mask = 0; 919 struct sock *sk = sock->sk; 920 struct xdp_sock *xs = xdp_sk(sk); 921 struct xsk_buff_pool *pool; 922 923 sock_poll_wait(file, sock, wait); 924 925 rcu_read_lock(); 926 if (xsk_check_common(xs)) 927 goto out; 928 929 pool = xs->pool; 930 931 if (pool->cached_need_wakeup) { 932 if (xs->zc) 933 xsk_wakeup(xs, pool->cached_need_wakeup); 934 else if (xs->tx) 935 /* Poll needs to drive Tx also in copy mode */ 936 xsk_generic_xmit(sk); 937 } 938 939 if (xs->rx && !xskq_prod_is_empty(xs->rx)) 940 mask |= EPOLLIN | EPOLLRDNORM; 941 if (xs->tx && xsk_tx_writeable(xs)) 942 mask |= EPOLLOUT | EPOLLWRNORM; 943out: 944 rcu_read_unlock(); 945 return mask; 946} 947 948static int xsk_init_queue(u32 entries, struct xsk_queue **queue, 949 bool umem_queue) 950{ 951 struct xsk_queue *q; 952 953 if (entries == 0 || *queue || !is_power_of_2(entries)) 954 return -EINVAL; 955 956 q = xskq_create(entries, umem_queue); 957 if (!q) 958 return -ENOMEM; 959 960 /* Make sure queue is ready before it can be seen by others */ 961 smp_wmb(); 962 WRITE_ONCE(*queue, q); 963 return 0; 964} 965 966static void xsk_unbind_dev(struct xdp_sock *xs) 967{ 968 struct net_device *dev = xs->dev; 969 970 if (xs->state != XSK_BOUND) 971 return; 972 WRITE_ONCE(xs->state, XSK_UNBOUND); 973 974 /* Wait for driver to stop using the xdp socket. */ 975 xp_del_xsk(xs->pool, xs); 976 synchronize_net(); 977 dev_put(dev); 978} 979 980static struct xsk_map *xsk_get_map_list_entry(struct xdp_sock *xs, 981 struct xdp_sock __rcu ***map_entry) 982{ 983 struct xsk_map *map = NULL; 984 struct xsk_map_node *node; 985 986 *map_entry = NULL; 987 988 spin_lock_bh(&xs->map_list_lock); 989 node = list_first_entry_or_null(&xs->map_list, struct xsk_map_node, 990 node); 991 if (node) { 992 bpf_map_inc(&node->map->map); 993 map = node->map; 994 *map_entry = node->map_entry; 995 } 996 spin_unlock_bh(&xs->map_list_lock); 997 return map; 998} 999 1000static void xsk_delete_from_maps(struct xdp_sock *xs) 1001{ 1002 /* This function removes the current XDP socket from all the 1003 * maps it resides in. We need to take extra care here, due to 1004 * the two locks involved. Each map has a lock synchronizing 1005 * updates to the entries, and each socket has a lock that 1006 * synchronizes access to the list of maps (map_list). For 1007 * deadlock avoidance the locks need to be taken in the order 1008 * "map lock"->"socket map list lock". We start off by 1009 * accessing the socket map list, and take a reference to the 1010 * map to guarantee existence between the 1011 * xsk_get_map_list_entry() and xsk_map_try_sock_delete() 1012 * calls. Then we ask the map to remove the socket, which 1013 * tries to remove the socket from the map. Note that there 1014 * might be updates to the map between 1015 * xsk_get_map_list_entry() and xsk_map_try_sock_delete(). 1016 */ 1017 struct xdp_sock __rcu **map_entry = NULL; 1018 struct xsk_map *map; 1019 1020 while ((map = xsk_get_map_list_entry(xs, &map_entry))) { 1021 xsk_map_try_sock_delete(map, xs, map_entry); 1022 bpf_map_put(&map->map); 1023 } 1024} 1025 1026static int xsk_release(struct socket *sock) 1027{ 1028 struct sock *sk = sock->sk; 1029 struct xdp_sock *xs = xdp_sk(sk); 1030 struct net *net; 1031 1032 if (!sk) 1033 return 0; 1034 1035 net = sock_net(sk); 1036 1037 if (xs->skb) 1038 xsk_drop_skb(xs->skb); 1039 1040 mutex_lock(&net->xdp.lock); 1041 sk_del_node_init_rcu(sk); 1042 mutex_unlock(&net->xdp.lock); 1043 1044 sock_prot_inuse_add(net, sk->sk_prot, -1); 1045 1046 xsk_delete_from_maps(xs); 1047 mutex_lock(&xs->mutex); 1048 xsk_unbind_dev(xs); 1049 mutex_unlock(&xs->mutex); 1050 1051 xskq_destroy(xs->rx); 1052 xskq_destroy(xs->tx); 1053 xskq_destroy(xs->fq_tmp); 1054 xskq_destroy(xs->cq_tmp); 1055 1056 sock_orphan(sk); 1057 sock->sk = NULL; 1058 1059 sock_put(sk); 1060 1061 return 0; 1062} 1063 1064static struct socket *xsk_lookup_xsk_from_fd(int fd) 1065{ 1066 struct socket *sock; 1067 int err; 1068 1069 sock = sockfd_lookup(fd, &err); 1070 if (!sock) 1071 return ERR_PTR(-ENOTSOCK); 1072 1073 if (sock->sk->sk_family != PF_XDP) { 1074 sockfd_put(sock); 1075 return ERR_PTR(-ENOPROTOOPT); 1076 } 1077 1078 return sock; 1079} 1080 1081static bool xsk_validate_queues(struct xdp_sock *xs) 1082{ 1083 return xs->fq_tmp && xs->cq_tmp; 1084} 1085 1086static int xsk_bind(struct socket *sock, struct sockaddr *addr, int addr_len) 1087{ 1088 struct sockaddr_xdp *sxdp = (struct sockaddr_xdp *)addr; 1089 struct sock *sk = sock->sk; 1090 struct xdp_sock *xs = xdp_sk(sk); 1091 struct net_device *dev; 1092 int bound_dev_if; 1093 u32 flags, qid; 1094 int err = 0; 1095 1096 if (addr_len < sizeof(struct sockaddr_xdp)) 1097 return -EINVAL; 1098 if (sxdp->sxdp_family != AF_XDP) 1099 return -EINVAL; 1100 1101 flags = sxdp->sxdp_flags; 1102 if (flags & ~(XDP_SHARED_UMEM | XDP_COPY | XDP_ZEROCOPY | 1103 XDP_USE_NEED_WAKEUP | XDP_USE_SG)) 1104 return -EINVAL; 1105 1106 bound_dev_if = READ_ONCE(sk->sk_bound_dev_if); 1107 if (bound_dev_if && bound_dev_if != sxdp->sxdp_ifindex) 1108 return -EINVAL; 1109 1110 rtnl_lock(); 1111 mutex_lock(&xs->mutex); 1112 if (xs->state != XSK_READY) { 1113 err = -EBUSY; 1114 goto out_release; 1115 } 1116 1117 dev = dev_get_by_index(sock_net(sk), sxdp->sxdp_ifindex); 1118 if (!dev) { 1119 err = -ENODEV; 1120 goto out_release; 1121 } 1122 1123 if (!xs->rx && !xs->tx) { 1124 err = -EINVAL; 1125 goto out_unlock; 1126 } 1127 1128 qid = sxdp->sxdp_queue_id; 1129 1130 if (flags & XDP_SHARED_UMEM) { 1131 struct xdp_sock *umem_xs; 1132 struct socket *sock; 1133 1134 if ((flags & XDP_COPY) || (flags & XDP_ZEROCOPY) || 1135 (flags & XDP_USE_NEED_WAKEUP) || (flags & XDP_USE_SG)) { 1136 /* Cannot specify flags for shared sockets. */ 1137 err = -EINVAL; 1138 goto out_unlock; 1139 } 1140 1141 if (xs->umem) { 1142 /* We have already our own. */ 1143 err = -EINVAL; 1144 goto out_unlock; 1145 } 1146 1147 sock = xsk_lookup_xsk_from_fd(sxdp->sxdp_shared_umem_fd); 1148 if (IS_ERR(sock)) { 1149 err = PTR_ERR(sock); 1150 goto out_unlock; 1151 } 1152 1153 umem_xs = xdp_sk(sock->sk); 1154 if (!xsk_is_bound(umem_xs)) { 1155 err = -EBADF; 1156 sockfd_put(sock); 1157 goto out_unlock; 1158 } 1159 1160 if (umem_xs->queue_id != qid || umem_xs->dev != dev) { 1161 /* Share the umem with another socket on another qid 1162 * and/or device. 1163 */ 1164 xs->pool = xp_create_and_assign_umem(xs, 1165 umem_xs->umem); 1166 if (!xs->pool) { 1167 err = -ENOMEM; 1168 sockfd_put(sock); 1169 goto out_unlock; 1170 } 1171 1172 err = xp_assign_dev_shared(xs->pool, umem_xs, dev, 1173 qid); 1174 if (err) { 1175 xp_destroy(xs->pool); 1176 xs->pool = NULL; 1177 sockfd_put(sock); 1178 goto out_unlock; 1179 } 1180 } else { 1181 /* Share the buffer pool with the other socket. */ 1182 if (xs->fq_tmp || xs->cq_tmp) { 1183 /* Do not allow setting your own fq or cq. */ 1184 err = -EINVAL; 1185 sockfd_put(sock); 1186 goto out_unlock; 1187 } 1188 1189 xp_get_pool(umem_xs->pool); 1190 xs->pool = umem_xs->pool; 1191 1192 /* If underlying shared umem was created without Tx 1193 * ring, allocate Tx descs array that Tx batching API 1194 * utilizes 1195 */ 1196 if (xs->tx && !xs->pool->tx_descs) { 1197 err = xp_alloc_tx_descs(xs->pool, xs); 1198 if (err) { 1199 xp_put_pool(xs->pool); 1200 xs->pool = NULL; 1201 sockfd_put(sock); 1202 goto out_unlock; 1203 } 1204 } 1205 } 1206 1207 xdp_get_umem(umem_xs->umem); 1208 WRITE_ONCE(xs->umem, umem_xs->umem); 1209 sockfd_put(sock); 1210 } else if (!xs->umem || !xsk_validate_queues(xs)) { 1211 err = -EINVAL; 1212 goto out_unlock; 1213 } else { 1214 /* This xsk has its own umem. */ 1215 xs->pool = xp_create_and_assign_umem(xs, xs->umem); 1216 if (!xs->pool) { 1217 err = -ENOMEM; 1218 goto out_unlock; 1219 } 1220 1221 err = xp_assign_dev(xs->pool, dev, qid, flags); 1222 if (err) { 1223 xp_destroy(xs->pool); 1224 xs->pool = NULL; 1225 goto out_unlock; 1226 } 1227 } 1228 1229 /* FQ and CQ are now owned by the buffer pool and cleaned up with it. */ 1230 xs->fq_tmp = NULL; 1231 xs->cq_tmp = NULL; 1232 1233 xs->dev = dev; 1234 xs->zc = xs->umem->zc; 1235 xs->sg = !!(xs->umem->flags & XDP_UMEM_SG_FLAG); 1236 xs->queue_id = qid; 1237 xp_add_xsk(xs->pool, xs); 1238 1239out_unlock: 1240 if (err) { 1241 dev_put(dev); 1242 } else { 1243 /* Matches smp_rmb() in bind() for shared umem 1244 * sockets, and xsk_is_bound(). 1245 */ 1246 smp_wmb(); 1247 WRITE_ONCE(xs->state, XSK_BOUND); 1248 } 1249out_release: 1250 mutex_unlock(&xs->mutex); 1251 rtnl_unlock(); 1252 return err; 1253} 1254 1255struct xdp_umem_reg_v1 { 1256 __u64 addr; /* Start of packet data area */ 1257 __u64 len; /* Length of packet data area */ 1258 __u32 chunk_size; 1259 __u32 headroom; 1260}; 1261 1262static int xsk_setsockopt(struct socket *sock, int level, int optname, 1263 sockptr_t optval, unsigned int optlen) 1264{ 1265 struct sock *sk = sock->sk; 1266 struct xdp_sock *xs = xdp_sk(sk); 1267 int err; 1268 1269 if (level != SOL_XDP) 1270 return -ENOPROTOOPT; 1271 1272 switch (optname) { 1273 case XDP_RX_RING: 1274 case XDP_TX_RING: 1275 { 1276 struct xsk_queue **q; 1277 int entries; 1278 1279 if (optlen < sizeof(entries)) 1280 return -EINVAL; 1281 if (copy_from_sockptr(&entries, optval, sizeof(entries))) 1282 return -EFAULT; 1283 1284 mutex_lock(&xs->mutex); 1285 if (xs->state != XSK_READY) { 1286 mutex_unlock(&xs->mutex); 1287 return -EBUSY; 1288 } 1289 q = (optname == XDP_TX_RING) ? &xs->tx : &xs->rx; 1290 err = xsk_init_queue(entries, q, false); 1291 if (!err && optname == XDP_TX_RING) 1292 /* Tx needs to be explicitly woken up the first time */ 1293 xs->tx->ring->flags |= XDP_RING_NEED_WAKEUP; 1294 mutex_unlock(&xs->mutex); 1295 return err; 1296 } 1297 case XDP_UMEM_REG: 1298 { 1299 size_t mr_size = sizeof(struct xdp_umem_reg); 1300 struct xdp_umem_reg mr = {}; 1301 struct xdp_umem *umem; 1302 1303 if (optlen < sizeof(struct xdp_umem_reg_v1)) 1304 return -EINVAL; 1305 else if (optlen < sizeof(mr)) 1306 mr_size = sizeof(struct xdp_umem_reg_v1); 1307 1308 if (copy_from_sockptr(&mr, optval, mr_size)) 1309 return -EFAULT; 1310 1311 mutex_lock(&xs->mutex); 1312 if (xs->state != XSK_READY || xs->umem) { 1313 mutex_unlock(&xs->mutex); 1314 return -EBUSY; 1315 } 1316 1317 umem = xdp_umem_create(&mr); 1318 if (IS_ERR(umem)) { 1319 mutex_unlock(&xs->mutex); 1320 return PTR_ERR(umem); 1321 } 1322 1323 /* Make sure umem is ready before it can be seen by others */ 1324 smp_wmb(); 1325 WRITE_ONCE(xs->umem, umem); 1326 mutex_unlock(&xs->mutex); 1327 return 0; 1328 } 1329 case XDP_UMEM_FILL_RING: 1330 case XDP_UMEM_COMPLETION_RING: 1331 { 1332 struct xsk_queue **q; 1333 int entries; 1334 1335 if (copy_from_sockptr(&entries, optval, sizeof(entries))) 1336 return -EFAULT; 1337 1338 mutex_lock(&xs->mutex); 1339 if (xs->state != XSK_READY) { 1340 mutex_unlock(&xs->mutex); 1341 return -EBUSY; 1342 } 1343 1344 q = (optname == XDP_UMEM_FILL_RING) ? &xs->fq_tmp : 1345 &xs->cq_tmp; 1346 err = xsk_init_queue(entries, q, true); 1347 mutex_unlock(&xs->mutex); 1348 return err; 1349 } 1350 default: 1351 break; 1352 } 1353 1354 return -ENOPROTOOPT; 1355} 1356 1357static void xsk_enter_rxtx_offsets(struct xdp_ring_offset_v1 *ring) 1358{ 1359 ring->producer = offsetof(struct xdp_rxtx_ring, ptrs.producer); 1360 ring->consumer = offsetof(struct xdp_rxtx_ring, ptrs.consumer); 1361 ring->desc = offsetof(struct xdp_rxtx_ring, desc); 1362} 1363 1364static void xsk_enter_umem_offsets(struct xdp_ring_offset_v1 *ring) 1365{ 1366 ring->producer = offsetof(struct xdp_umem_ring, ptrs.producer); 1367 ring->consumer = offsetof(struct xdp_umem_ring, ptrs.consumer); 1368 ring->desc = offsetof(struct xdp_umem_ring, desc); 1369} 1370 1371struct xdp_statistics_v1 { 1372 __u64 rx_dropped; 1373 __u64 rx_invalid_descs; 1374 __u64 tx_invalid_descs; 1375}; 1376 1377static int xsk_getsockopt(struct socket *sock, int level, int optname, 1378 char __user *optval, int __user *optlen) 1379{ 1380 struct sock *sk = sock->sk; 1381 struct xdp_sock *xs = xdp_sk(sk); 1382 int len; 1383 1384 if (level != SOL_XDP) 1385 return -ENOPROTOOPT; 1386 1387 if (get_user(len, optlen)) 1388 return -EFAULT; 1389 if (len < 0) 1390 return -EINVAL; 1391 1392 switch (optname) { 1393 case XDP_STATISTICS: 1394 { 1395 struct xdp_statistics stats = {}; 1396 bool extra_stats = true; 1397 size_t stats_size; 1398 1399 if (len < sizeof(struct xdp_statistics_v1)) { 1400 return -EINVAL; 1401 } else if (len < sizeof(stats)) { 1402 extra_stats = false; 1403 stats_size = sizeof(struct xdp_statistics_v1); 1404 } else { 1405 stats_size = sizeof(stats); 1406 } 1407 1408 mutex_lock(&xs->mutex); 1409 stats.rx_dropped = xs->rx_dropped; 1410 if (extra_stats) { 1411 stats.rx_ring_full = xs->rx_queue_full; 1412 stats.rx_fill_ring_empty_descs = 1413 xs->pool ? xskq_nb_queue_empty_descs(xs->pool->fq) : 0; 1414 stats.tx_ring_empty_descs = xskq_nb_queue_empty_descs(xs->tx); 1415 } else { 1416 stats.rx_dropped += xs->rx_queue_full; 1417 } 1418 stats.rx_invalid_descs = xskq_nb_invalid_descs(xs->rx); 1419 stats.tx_invalid_descs = xskq_nb_invalid_descs(xs->tx); 1420 mutex_unlock(&xs->mutex); 1421 1422 if (copy_to_user(optval, &stats, stats_size)) 1423 return -EFAULT; 1424 if (put_user(stats_size, optlen)) 1425 return -EFAULT; 1426 1427 return 0; 1428 } 1429 case XDP_MMAP_OFFSETS: 1430 { 1431 struct xdp_mmap_offsets off; 1432 struct xdp_mmap_offsets_v1 off_v1; 1433 bool flags_supported = true; 1434 void *to_copy; 1435 1436 if (len < sizeof(off_v1)) 1437 return -EINVAL; 1438 else if (len < sizeof(off)) 1439 flags_supported = false; 1440 1441 if (flags_supported) { 1442 /* xdp_ring_offset is identical to xdp_ring_offset_v1 1443 * except for the flags field added to the end. 1444 */ 1445 xsk_enter_rxtx_offsets((struct xdp_ring_offset_v1 *) 1446 &off.rx); 1447 xsk_enter_rxtx_offsets((struct xdp_ring_offset_v1 *) 1448 &off.tx); 1449 xsk_enter_umem_offsets((struct xdp_ring_offset_v1 *) 1450 &off.fr); 1451 xsk_enter_umem_offsets((struct xdp_ring_offset_v1 *) 1452 &off.cr); 1453 off.rx.flags = offsetof(struct xdp_rxtx_ring, 1454 ptrs.flags); 1455 off.tx.flags = offsetof(struct xdp_rxtx_ring, 1456 ptrs.flags); 1457 off.fr.flags = offsetof(struct xdp_umem_ring, 1458 ptrs.flags); 1459 off.cr.flags = offsetof(struct xdp_umem_ring, 1460 ptrs.flags); 1461 1462 len = sizeof(off); 1463 to_copy = &off; 1464 } else { 1465 xsk_enter_rxtx_offsets(&off_v1.rx); 1466 xsk_enter_rxtx_offsets(&off_v1.tx); 1467 xsk_enter_umem_offsets(&off_v1.fr); 1468 xsk_enter_umem_offsets(&off_v1.cr); 1469 1470 len = sizeof(off_v1); 1471 to_copy = &off_v1; 1472 } 1473 1474 if (copy_to_user(optval, to_copy, len)) 1475 return -EFAULT; 1476 if (put_user(len, optlen)) 1477 return -EFAULT; 1478 1479 return 0; 1480 } 1481 case XDP_OPTIONS: 1482 { 1483 struct xdp_options opts = {}; 1484 1485 if (len < sizeof(opts)) 1486 return -EINVAL; 1487 1488 mutex_lock(&xs->mutex); 1489 if (xs->zc) 1490 opts.flags |= XDP_OPTIONS_ZEROCOPY; 1491 mutex_unlock(&xs->mutex); 1492 1493 len = sizeof(opts); 1494 if (copy_to_user(optval, &opts, len)) 1495 return -EFAULT; 1496 if (put_user(len, optlen)) 1497 return -EFAULT; 1498 1499 return 0; 1500 } 1501 default: 1502 break; 1503 } 1504 1505 return -EOPNOTSUPP; 1506} 1507 1508static int xsk_mmap(struct file *file, struct socket *sock, 1509 struct vm_area_struct *vma) 1510{ 1511 loff_t offset = (loff_t)vma->vm_pgoff << PAGE_SHIFT; 1512 unsigned long size = vma->vm_end - vma->vm_start; 1513 struct xdp_sock *xs = xdp_sk(sock->sk); 1514 int state = READ_ONCE(xs->state); 1515 struct xsk_queue *q = NULL; 1516 1517 if (state != XSK_READY && state != XSK_BOUND) 1518 return -EBUSY; 1519 1520 if (offset == XDP_PGOFF_RX_RING) { 1521 q = READ_ONCE(xs->rx); 1522 } else if (offset == XDP_PGOFF_TX_RING) { 1523 q = READ_ONCE(xs->tx); 1524 } else { 1525 /* Matches the smp_wmb() in XDP_UMEM_REG */ 1526 smp_rmb(); 1527 if (offset == XDP_UMEM_PGOFF_FILL_RING) 1528 q = state == XSK_READY ? READ_ONCE(xs->fq_tmp) : 1529 READ_ONCE(xs->pool->fq); 1530 else if (offset == XDP_UMEM_PGOFF_COMPLETION_RING) 1531 q = state == XSK_READY ? READ_ONCE(xs->cq_tmp) : 1532 READ_ONCE(xs->pool->cq); 1533 } 1534 1535 if (!q) 1536 return -EINVAL; 1537 1538 /* Matches the smp_wmb() in xsk_init_queue */ 1539 smp_rmb(); 1540 if (size > q->ring_vmalloc_size) 1541 return -EINVAL; 1542 1543 return remap_vmalloc_range(vma, q->ring, 0); 1544} 1545 1546static int xsk_notifier(struct notifier_block *this, 1547 unsigned long msg, void *ptr) 1548{ 1549 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 1550 struct net *net = dev_net(dev); 1551 struct sock *sk; 1552 1553 switch (msg) { 1554 case NETDEV_UNREGISTER: 1555 mutex_lock(&net->xdp.lock); 1556 sk_for_each(sk, &net->xdp.list) { 1557 struct xdp_sock *xs = xdp_sk(sk); 1558 1559 mutex_lock(&xs->mutex); 1560 if (xs->dev == dev) { 1561 sk->sk_err = ENETDOWN; 1562 if (!sock_flag(sk, SOCK_DEAD)) 1563 sk_error_report(sk); 1564 1565 xsk_unbind_dev(xs); 1566 1567 /* Clear device references. */ 1568 xp_clear_dev(xs->pool); 1569 } 1570 mutex_unlock(&xs->mutex); 1571 } 1572 mutex_unlock(&net->xdp.lock); 1573 break; 1574 } 1575 return NOTIFY_DONE; 1576} 1577 1578static struct proto xsk_proto = { 1579 .name = "XDP", 1580 .owner = THIS_MODULE, 1581 .obj_size = sizeof(struct xdp_sock), 1582}; 1583 1584static const struct proto_ops xsk_proto_ops = { 1585 .family = PF_XDP, 1586 .owner = THIS_MODULE, 1587 .release = xsk_release, 1588 .bind = xsk_bind, 1589 .connect = sock_no_connect, 1590 .socketpair = sock_no_socketpair, 1591 .accept = sock_no_accept, 1592 .getname = sock_no_getname, 1593 .poll = xsk_poll, 1594 .ioctl = sock_no_ioctl, 1595 .listen = sock_no_listen, 1596 .shutdown = sock_no_shutdown, 1597 .setsockopt = xsk_setsockopt, 1598 .getsockopt = xsk_getsockopt, 1599 .sendmsg = xsk_sendmsg, 1600 .recvmsg = xsk_recvmsg, 1601 .mmap = xsk_mmap, 1602}; 1603 1604static void xsk_destruct(struct sock *sk) 1605{ 1606 struct xdp_sock *xs = xdp_sk(sk); 1607 1608 if (!sock_flag(sk, SOCK_DEAD)) 1609 return; 1610 1611 if (!xp_put_pool(xs->pool)) 1612 xdp_put_umem(xs->umem, !xs->pool); 1613} 1614 1615static int xsk_create(struct net *net, struct socket *sock, int protocol, 1616 int kern) 1617{ 1618 struct xdp_sock *xs; 1619 struct sock *sk; 1620 1621 if (!ns_capable(net->user_ns, CAP_NET_RAW)) 1622 return -EPERM; 1623 if (sock->type != SOCK_RAW) 1624 return -ESOCKTNOSUPPORT; 1625 1626 if (protocol) 1627 return -EPROTONOSUPPORT; 1628 1629 sock->state = SS_UNCONNECTED; 1630 1631 sk = sk_alloc(net, PF_XDP, GFP_KERNEL, &xsk_proto, kern); 1632 if (!sk) 1633 return -ENOBUFS; 1634 1635 sock->ops = &xsk_proto_ops; 1636 1637 sock_init_data(sock, sk); 1638 1639 sk->sk_family = PF_XDP; 1640 1641 sk->sk_destruct = xsk_destruct; 1642 1643 sock_set_flag(sk, SOCK_RCU_FREE); 1644 1645 xs = xdp_sk(sk); 1646 xs->state = XSK_READY; 1647 mutex_init(&xs->mutex); 1648 spin_lock_init(&xs->rx_lock); 1649 1650 INIT_LIST_HEAD(&xs->map_list); 1651 spin_lock_init(&xs->map_list_lock); 1652 1653 mutex_lock(&net->xdp.lock); 1654 sk_add_node_rcu(sk, &net->xdp.list); 1655 mutex_unlock(&net->xdp.lock); 1656 1657 sock_prot_inuse_add(net, &xsk_proto, 1); 1658 1659 return 0; 1660} 1661 1662static const struct net_proto_family xsk_family_ops = { 1663 .family = PF_XDP, 1664 .create = xsk_create, 1665 .owner = THIS_MODULE, 1666}; 1667 1668static struct notifier_block xsk_netdev_notifier = { 1669 .notifier_call = xsk_notifier, 1670}; 1671 1672static int __net_init xsk_net_init(struct net *net) 1673{ 1674 mutex_init(&net->xdp.lock); 1675 INIT_HLIST_HEAD(&net->xdp.list); 1676 return 0; 1677} 1678 1679static void __net_exit xsk_net_exit(struct net *net) 1680{ 1681 WARN_ON_ONCE(!hlist_empty(&net->xdp.list)); 1682} 1683 1684static struct pernet_operations xsk_net_ops = { 1685 .init = xsk_net_init, 1686 .exit = xsk_net_exit, 1687}; 1688 1689static int __init xsk_init(void) 1690{ 1691 int err, cpu; 1692 1693 err = proto_register(&xsk_proto, 0 /* no slab */); 1694 if (err) 1695 goto out; 1696 1697 err = sock_register(&xsk_family_ops); 1698 if (err) 1699 goto out_proto; 1700 1701 err = register_pernet_subsys(&xsk_net_ops); 1702 if (err) 1703 goto out_sk; 1704 1705 err = register_netdevice_notifier(&xsk_netdev_notifier); 1706 if (err) 1707 goto out_pernet; 1708 1709 for_each_possible_cpu(cpu) 1710 INIT_LIST_HEAD(&per_cpu(xskmap_flush_list, cpu)); 1711 return 0; 1712 1713out_pernet: 1714 unregister_pernet_subsys(&xsk_net_ops); 1715out_sk: 1716 sock_unregister(PF_XDP); 1717out_proto: 1718 proto_unregister(&xsk_proto); 1719out: 1720 return err; 1721} 1722 1723fs_initcall(xsk_init); 1724