net/xdp/xsk.c

   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 <net/xdp_sock.h>
  26 #include <net/xdp.h>
  27
  28 #include "xsk_queue.h"
  29 #include "xdp_umem.h"
  30
  31 #define TX_BATCH_SIZE 16
  32
  33 static struct xdp_sock *xdp_sk(struct sock *sk)
  34 {
  35         return (struct xdp_sock *)sk;
  36 }
  37
  38 bool xsk_is_setup_for_bpf_map(struct xdp_sock *xs)
  39 {
  40         return READ_ONCE(xs->rx) &&  READ_ONCE(xs->umem) &&
  41                 READ_ONCE(xs->umem->fq);
  42 }
  43
  44 u64 *xsk_umem_peek_addr(struct xdp_umem *umem, u64 *addr)
  45 {
  46         return xskq_peek_addr(umem->fq, addr);
  47 }
  48 EXPORT_SYMBOL(xsk_umem_peek_addr);
  49
  50 void xsk_umem_discard_addr(struct xdp_umem *umem)
  51 {
  52         xskq_discard_addr(umem->fq);
  53 }
  54 EXPORT_SYMBOL(xsk_umem_discard_addr);
  55
  56 static int __xsk_rcv(struct xdp_sock *xs, struct xdp_buff *xdp, u32 len)
  57 {
  58         void *to_buf, *from_buf;
  59         u32 metalen;
  60         u64 addr;
  61         int err;
  62
  63         if (!xskq_peek_addr(xs->umem->fq, &addr) ||
  64             len > xs->umem->chunk_size_nohr - XDP_PACKET_HEADROOM) {
  65                 xs->rx_dropped++;
  66                 return -ENOSPC;
  67         }
  68
  69         addr += xs->umem->headroom;
  70
  71         if (unlikely(xdp_data_meta_unsupported(xdp))) {
  72                 from_buf = xdp->data;
  73                 metalen = 0;
  74         } else {
  75                 from_buf = xdp->data_meta;
  76                 metalen = xdp->data - xdp->data_meta;
  77         }
  78
  79         to_buf = xdp_umem_get_data(xs->umem, addr);
  80         memcpy(to_buf, from_buf, len + metalen);
  81         addr += metalen;
  82         err = xskq_produce_batch_desc(xs->rx, addr, len);
  83         if (!err) {
  84                 xskq_discard_addr(xs->umem->fq);
  85                 xdp_return_buff(xdp);
  86                 return 0;
  87         }
  88
  89         xs->rx_dropped++;
  90         return err;
  91 }
  92
  93 static int __xsk_rcv_zc(struct xdp_sock *xs, struct xdp_buff *xdp, u32 len)
  94 {
  95         int err = xskq_produce_batch_desc(xs->rx, (u64)xdp->handle, len);
  96
  97         if (err)
  98                 xs->rx_dropped++;
  99
 100         return err;
 101 }
 102
 103 int xsk_rcv(struct xdp_sock *xs, struct xdp_buff *xdp)
 104 {
 105         u32 len;
 106
 107         if (xs->dev != xdp->rxq->dev || xs->queue_id != xdp->rxq->queue_index)
 108                 return -EINVAL;
 109
 110         len = xdp->data_end - xdp->data;
 111
 112         return (xdp->rxq->mem.type == MEM_TYPE_ZERO_COPY) ?
 113                 __xsk_rcv_zc(xs, xdp, len) : __xsk_rcv(xs, xdp, len);
 114 }
 115
 116 void xsk_flush(struct xdp_sock *xs)
 117 {
 118         xskq_produce_flush_desc(xs->rx);
 119         xs->sk.sk_data_ready(&xs->sk);
 120 }
 121
 122 int xsk_generic_rcv(struct xdp_sock *xs, struct xdp_buff *xdp)
 123 {
 124         u32 metalen = xdp->data - xdp->data_meta;
 125         u32 len = xdp->data_end - xdp->data;
 126         void *buffer;
 127         u64 addr;
 128         int err;
 129
 130         if (xs->dev != xdp->rxq->dev || xs->queue_id != xdp->rxq->queue_index)
 131                 return -EINVAL;
 132
 133         if (!xskq_peek_addr(xs->umem->fq, &addr) ||
 134             len > xs->umem->chunk_size_nohr - XDP_PACKET_HEADROOM) {
 135                 xs->rx_dropped++;
 136                 return -ENOSPC;
 137         }
 138
 139         addr += xs->umem->headroom;
 140
 141         buffer = xdp_umem_get_data(xs->umem, addr);
 142         memcpy(buffer, xdp->data_meta, len + metalen);
 143         addr += metalen;
 144         err = xskq_produce_batch_desc(xs->rx, addr, len);
 145         if (!err) {
 146                 xskq_discard_addr(xs->umem->fq);
 147                 xsk_flush(xs);
 148                 return 0;
 149         }
 150
 151         xs->rx_dropped++;
 152         return err;
 153 }
 154
 155 void xsk_umem_complete_tx(struct xdp_umem *umem, u32 nb_entries)
 156 {
 157         xskq_produce_flush_addr_n(umem->cq, nb_entries);
 158 }
 159 EXPORT_SYMBOL(xsk_umem_complete_tx);
 160
 161 void xsk_umem_consume_tx_done(struct xdp_umem *umem)
 162 {
 163         struct xdp_sock *xs;
 164
 165         rcu_read_lock();
 166         list_for_each_entry_rcu(xs, &umem->xsk_list, list) {
 167                 xs->sk.sk_write_space(&xs->sk);
 168         }
 169         rcu_read_unlock();
 170 }
 171 EXPORT_SYMBOL(xsk_umem_consume_tx_done);
 172
 173 bool xsk_umem_consume_tx(struct xdp_umem *umem, dma_addr_t *dma, u32 *len)
 174 {
 175         struct xdp_desc desc;
 176         struct xdp_sock *xs;
 177
 178         rcu_read_lock();
 179         list_for_each_entry_rcu(xs, &umem->xsk_list, list) {
 180                 if (!xskq_peek_desc(xs->tx, &desc))
 181                         continue;
 182
 183                 if (xskq_produce_addr_lazy(umem->cq, desc.addr))
 184                         goto out;
 185
 186                 *dma = xdp_umem_get_dma(umem, desc.addr);
 187                 *len = desc.len;
 188
 189                 xskq_discard_desc(xs->tx);
 190                 rcu_read_unlock();
 191                 return true;
 192         }
 193
 194 out:
 195         rcu_read_unlock();
 196         return false;
 197 }
 198 EXPORT_SYMBOL(xsk_umem_consume_tx);
 199
 200 static int xsk_zc_xmit(struct sock *sk)
 201 {
 202         struct xdp_sock *xs = xdp_sk(sk);
 203         struct net_device *dev = xs->dev;
 204
 205         return dev->netdev_ops->ndo_xsk_async_xmit(dev, xs->queue_id);
 206 }
 207
 208 static void xsk_destruct_skb(struct sk_buff *skb)
 209 {
 210         u64 addr = (u64)(long)skb_shinfo(skb)->destructor_arg;
 211         struct xdp_sock *xs = xdp_sk(skb->sk);
 212         unsigned long flags;
 213
 214         spin_lock_irqsave(&xs->tx_completion_lock, flags);
 215         WARN_ON_ONCE(xskq_produce_addr(xs->umem->cq, addr));
 216         spin_unlock_irqrestore(&xs->tx_completion_lock, flags);
 217
 218         sock_wfree(skb);
 219 }
 220
 221 static int xsk_generic_xmit(struct sock *sk, struct msghdr *m,
 222                             size_t total_len)
 223 {
 224         u32 max_batch = TX_BATCH_SIZE;
 225         struct xdp_sock *xs = xdp_sk(sk);
 226         bool sent_frame = false;
 227         struct xdp_desc desc;
 228         struct sk_buff *skb;
 229         int err = 0;
 230
 231         mutex_lock(&xs->mutex);
 232
 233         while (xskq_peek_desc(xs->tx, &desc)) {
 234                 char *buffer;
 235                 u64 addr;
 236                 u32 len;
 237
 238                 if (max_batch-- == 0) {
 239                         err = -EAGAIN;
 240                         goto out;
 241                 }
 242
 243                 if (xskq_reserve_addr(xs->umem->cq))
 244                         goto out;
 245
 246                 if (xs->queue_id >= xs->dev->real_num_tx_queues)
 247                         goto out;
 248
 249                 len = desc.len;
 250                 skb = sock_alloc_send_skb(sk, len, 1, &err);
 251                 if (unlikely(!skb)) {
 252                         err = -EAGAIN;
 253                         goto out;
 254                 }
 255
 256                 skb_put(skb, len);
 257                 addr = desc.addr;
 258                 buffer = xdp_umem_get_data(xs->umem, addr);
 259                 err = skb_store_bits(skb, 0, buffer, len);
 260                 if (unlikely(err)) {
 261                         kfree_skb(skb);
 262                         goto out;
 263                 }
 264
 265                 skb->dev = xs->dev;
 266                 skb->priority = sk->sk_priority;
 267                 skb->mark = sk->sk_mark;
 268                 skb_shinfo(skb)->destructor_arg = (void *)(long)addr;
 269                 skb->destructor = xsk_destruct_skb;
 270
 271                 err = dev_direct_xmit(skb, xs->queue_id);
 272                 xskq_discard_desc(xs->tx);
 273                 /* Ignore NET_XMIT_CN as packet might have been sent */
 274                 if (err == NET_XMIT_DROP || err == NETDEV_TX_BUSY) {
 275                         /* SKB completed but not sent */
 276                         err = -EBUSY;
 277                         goto out;
 278                 }
 279
 280                 sent_frame = true;
 281         }
 282
 283 out:
 284         if (sent_frame)
 285                 sk->sk_write_space(sk);
 286
 287         mutex_unlock(&xs->mutex);
 288         return err;
 289 }
 290
 291 static int xsk_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
 292 {
 293         bool need_wait = !(m->msg_flags & MSG_DONTWAIT);
 294         struct sock *sk = sock->sk;
 295         struct xdp_sock *xs = xdp_sk(sk);
 296
 297         if (unlikely(!xs->dev))
 298                 return -ENXIO;
 299         if (unlikely(!(xs->dev->flags & IFF_UP)))
 300                 return -ENETDOWN;
 301         if (unlikely(!xs->tx))
 302                 return -ENOBUFS;
 303         if (need_wait)
 304                 return -EOPNOTSUPP;
 305
 306         return (xs->zc) ? xsk_zc_xmit(sk) : xsk_generic_xmit(sk, m, total_len);
 307 }
 308
 309 static unsigned int xsk_poll(struct file *file, struct socket *sock,
 310                              struct poll_table_struct *wait)
 311 {
 312         unsigned int mask = datagram_poll(file, sock, wait);
 313         struct sock *sk = sock->sk;
 314         struct xdp_sock *xs = xdp_sk(sk);
 315
 316         if (xs->rx && !xskq_empty_desc(xs->rx))
 317                 mask |= POLLIN | POLLRDNORM;
 318         if (xs->tx && !xskq_full_desc(xs->tx))
 319                 mask |= POLLOUT | POLLWRNORM;
 320
 321         return mask;
 322 }
 323
 324 static int xsk_init_queue(u32 entries, struct xsk_queue **queue,
 325                           bool umem_queue)
 326 {
 327         struct xsk_queue *q;
 328
 329         if (entries == 0 || *queue || !is_power_of_2(entries))
 330                 return -EINVAL;
 331
 332         q = xskq_create(entries, umem_queue);
 333         if (!q)
 334                 return -ENOMEM;
 335
 336         /* Make sure queue is ready before it can be seen by others */
 337         smp_wmb();
 338         *queue = q;
 339         return 0;
 340 }
 341
 342 static int xsk_release(struct socket *sock)
 343 {
 344         struct sock *sk = sock->sk;
 345         struct xdp_sock *xs = xdp_sk(sk);
 346         struct net *net;
 347
 348         if (!sk)
 349                 return 0;
 350
 351         net = sock_net(sk);
 352
 353         local_bh_disable();
 354         sock_prot_inuse_add(net, sk->sk_prot, -1);
 355         local_bh_enable();
 356
 357         if (xs->dev) {
 358                 struct net_device *dev = xs->dev;
 359
 360                 /* Wait for driver to stop using the xdp socket. */
 361                 xdp_del_sk_umem(xs->umem, xs);
 362                 xs->dev = NULL;
 363                 synchronize_net();
 364                 dev_put(dev);
 365         }
 366
 367         xskq_destroy(xs->rx);
 368         xskq_destroy(xs->tx);
 369
 370         sock_orphan(sk);
 371         sock->sk = NULL;
 372
 373         sk_refcnt_debug_release(sk);
 374         sock_put(sk);
 375
 376         return 0;
 377 }
 378
 379 static struct socket *xsk_lookup_xsk_from_fd(int fd)
 380 {
 381         struct socket *sock;
 382         int err;
 383
 384         sock = sockfd_lookup(fd, &err);
 385         if (!sock)
 386                 return ERR_PTR(-ENOTSOCK);
 387
 388         if (sock->sk->sk_family != PF_XDP) {
 389                 sockfd_put(sock);
 390                 return ERR_PTR(-ENOPROTOOPT);
 391         }
 392
 393         return sock;
 394 }
 395
 396 static int xsk_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
 397 {
 398         struct sockaddr_xdp *sxdp = (struct sockaddr_xdp *)addr;
 399         struct sock *sk = sock->sk;
 400         struct xdp_sock *xs = xdp_sk(sk);
 401         struct net_device *dev;
 402         u32 flags, qid;
 403         int err = 0;
 404
 405         if (addr_len < sizeof(struct sockaddr_xdp))
 406                 return -EINVAL;
 407         if (sxdp->sxdp_family != AF_XDP)
 408                 return -EINVAL;
 409
 410         mutex_lock(&xs->mutex);
 411         if (xs->dev) {
 412                 err = -EBUSY;
 413                 goto out_release;
 414         }
 415
 416         dev = dev_get_by_index(sock_net(sk), sxdp->sxdp_ifindex);
 417         if (!dev) {
 418                 err = -ENODEV;
 419                 goto out_release;
 420         }
 421
 422         if (!xs->rx && !xs->tx) {
 423                 err = -EINVAL;
 424                 goto out_unlock;
 425         }
 426
 427         qid = sxdp->sxdp_queue_id;
 428         flags = sxdp->sxdp_flags;
 429
 430         if (flags & XDP_SHARED_UMEM) {
 431                 struct xdp_sock *umem_xs;
 432                 struct socket *sock;
 433
 434                 if ((flags & XDP_COPY) || (flags & XDP_ZEROCOPY)) {
 435                         /* Cannot specify flags for shared sockets. */
 436                         err = -EINVAL;
 437                         goto out_unlock;
 438                 }
 439
 440                 if (xs->umem) {
 441                         /* We have already our own. */
 442                         err = -EINVAL;
 443                         goto out_unlock;
 444                 }
 445
 446                 sock = xsk_lookup_xsk_from_fd(sxdp->sxdp_shared_umem_fd);
 447                 if (IS_ERR(sock)) {
 448                         err = PTR_ERR(sock);
 449                         goto out_unlock;
 450                 }
 451
 452                 umem_xs = xdp_sk(sock->sk);
 453                 if (!umem_xs->umem) {
 454                         /* No umem to inherit. */
 455                         err = -EBADF;
 456                         sockfd_put(sock);
 457                         goto out_unlock;
 458                 } else if (umem_xs->dev != dev || umem_xs->queue_id != qid) {
 459                         err = -EINVAL;
 460                         sockfd_put(sock);
 461                         goto out_unlock;
 462                 }
 463
 464                 xdp_get_umem(umem_xs->umem);
 465                 xs->umem = umem_xs->umem;
 466                 sockfd_put(sock);
 467         } else if (!xs->umem || !xdp_umem_validate_queues(xs->umem)) {
 468                 err = -EINVAL;
 469                 goto out_unlock;
 470         } else {
 471                 /* This xsk has its own umem. */
 472                 xskq_set_umem(xs->umem->fq, xs->umem->size,
 473                               xs->umem->chunk_mask);
 474                 xskq_set_umem(xs->umem->cq, xs->umem->size,
 475                               xs->umem->chunk_mask);
 476
 477                 err = xdp_umem_assign_dev(xs->umem, dev, qid, flags);
 478                 if (err)
 479                         goto out_unlock;
 480         }
 481
 482         xs->dev = dev;
 483         xs->zc = xs->umem->zc;
 484         xs->queue_id = qid;
 485         xskq_set_umem(xs->rx, xs->umem->size, xs->umem->chunk_mask);
 486         xskq_set_umem(xs->tx, xs->umem->size, xs->umem->chunk_mask);
 487         xdp_add_sk_umem(xs->umem, xs);
 488
 489 out_unlock:
 490         if (err)
 491                 dev_put(dev);
 492 out_release:
 493         mutex_unlock(&xs->mutex);
 494         return err;
 495 }
 496
 497 static int xsk_setsockopt(struct socket *sock, int level, int optname,
 498                           char __user *optval, unsigned int optlen)
 499 {
 500         struct sock *sk = sock->sk;
 501         struct xdp_sock *xs = xdp_sk(sk);
 502         int err;
 503
 504         if (level != SOL_XDP)
 505                 return -ENOPROTOOPT;
 506
 507         switch (optname) {
 508         case XDP_RX_RING:
 509         case XDP_TX_RING:
 510         {
 511                 struct xsk_queue **q;
 512                 int entries;
 513
 514                 if (optlen < sizeof(entries))
 515                         return -EINVAL;
 516                 if (copy_from_user(&entries, optval, sizeof(entries)))
 517                         return -EFAULT;
 518
 519                 mutex_lock(&xs->mutex);
 520                 q = (optname == XDP_TX_RING) ? &xs->tx : &xs->rx;
 521                 err = xsk_init_queue(entries, q, false);
 522                 mutex_unlock(&xs->mutex);
 523                 return err;
 524         }
 525         case XDP_UMEM_REG:
 526         {
 527                 struct xdp_umem_reg mr;
 528                 struct xdp_umem *umem;
 529
 530                 if (copy_from_user(&mr, optval, sizeof(mr)))
 531                         return -EFAULT;
 532
 533                 mutex_lock(&xs->mutex);
 534                 if (xs->umem) {
 535                         mutex_unlock(&xs->mutex);
 536                         return -EBUSY;
 537                 }
 538
 539                 umem = xdp_umem_create(&mr);
 540                 if (IS_ERR(umem)) {
 541                         mutex_unlock(&xs->mutex);
 542                         return PTR_ERR(umem);
 543                 }
 544
 545                 /* Make sure umem is ready before it can be seen by others */
 546                 smp_wmb();
 547                 xs->umem = umem;
 548                 mutex_unlock(&xs->mutex);
 549                 return 0;
 550         }
 551         case XDP_UMEM_FILL_RING:
 552         case XDP_UMEM_COMPLETION_RING:
 553         {
 554                 struct xsk_queue **q;
 555                 int entries;
 556
 557                 if (copy_from_user(&entries, optval, sizeof(entries)))
 558                         return -EFAULT;
 559
 560                 mutex_lock(&xs->mutex);
 561                 if (!xs->umem) {
 562                         mutex_unlock(&xs->mutex);
 563                         return -EINVAL;
 564                 }
 565
 566                 q = (optname == XDP_UMEM_FILL_RING) ? &xs->umem->fq :
 567                         &xs->umem->cq;
 568                 err = xsk_init_queue(entries, q, true);
 569                 mutex_unlock(&xs->mutex);
 570                 return err;
 571         }
 572         default:
 573                 break;
 574         }
 575
 576         return -ENOPROTOOPT;
 577 }
 578
 579 static int xsk_getsockopt(struct socket *sock, int level, int optname,
 580                           char __user *optval, int __user *optlen)
 581 {
 582         struct sock *sk = sock->sk;
 583         struct xdp_sock *xs = xdp_sk(sk);
 584         int len;
 585
 586         if (level != SOL_XDP)
 587                 return -ENOPROTOOPT;
 588
 589         if (get_user(len, optlen))
 590                 return -EFAULT;
 591         if (len < 0)
 592                 return -EINVAL;
 593
 594         switch (optname) {
 595         case XDP_STATISTICS:
 596         {
 597                 struct xdp_statistics stats;
 598
 599                 if (len < sizeof(stats))
 600                         return -EINVAL;
 601
 602                 mutex_lock(&xs->mutex);
 603                 stats.rx_dropped = xs->rx_dropped;
 604                 stats.rx_invalid_descs = xskq_nb_invalid_descs(xs->rx);
 605                 stats.tx_invalid_descs = xskq_nb_invalid_descs(xs->tx);
 606                 mutex_unlock(&xs->mutex);
 607
 608                 if (copy_to_user(optval, &stats, sizeof(stats)))
 609                         return -EFAULT;
 610                 if (put_user(sizeof(stats), optlen))
 611                         return -EFAULT;
 612
 613                 return 0;
 614         }
 615         case XDP_MMAP_OFFSETS:
 616         {
 617                 struct xdp_mmap_offsets off;
 618
 619                 if (len < sizeof(off))
 620                         return -EINVAL;
 621
 622                 off.rx.producer = offsetof(struct xdp_rxtx_ring, ptrs.producer);
 623                 off.rx.consumer = offsetof(struct xdp_rxtx_ring, ptrs.consumer);
 624                 off.rx.desc     = offsetof(struct xdp_rxtx_ring, desc);
 625                 off.tx.producer = offsetof(struct xdp_rxtx_ring, ptrs.producer);
 626                 off.tx.consumer = offsetof(struct xdp_rxtx_ring, ptrs.consumer);
 627                 off.tx.desc     = offsetof(struct xdp_rxtx_ring, desc);
 628
 629                 off.fr.producer = offsetof(struct xdp_umem_ring, ptrs.producer);
 630                 off.fr.consumer = offsetof(struct xdp_umem_ring, ptrs.consumer);
 631                 off.fr.desc     = offsetof(struct xdp_umem_ring, desc);
 632                 off.cr.producer = offsetof(struct xdp_umem_ring, ptrs.producer);
 633                 off.cr.consumer = offsetof(struct xdp_umem_ring, ptrs.consumer);
 634                 off.cr.desc     = offsetof(struct xdp_umem_ring, desc);
 635
 636                 len = sizeof(off);
 637                 if (copy_to_user(optval, &off, len))
 638                         return -EFAULT;
 639                 if (put_user(len, optlen))
 640                         return -EFAULT;
 641
 642                 return 0;
 643         }
 644         default:
 645                 break;
 646         }
 647
 648         return -EOPNOTSUPP;
 649 }
 650
 651 static int xsk_mmap(struct file *file, struct socket *sock,
 652                     struct vm_area_struct *vma)
 653 {
 654         loff_t offset = (loff_t)vma->vm_pgoff << PAGE_SHIFT;
 655         unsigned long size = vma->vm_end - vma->vm_start;
 656         struct xdp_sock *xs = xdp_sk(sock->sk);
 657         struct xsk_queue *q = NULL;
 658         struct xdp_umem *umem;
 659         unsigned long pfn;
 660         struct page *qpg;
 661
 662         if (offset == XDP_PGOFF_RX_RING) {
 663                 q = READ_ONCE(xs->rx);
 664         } else if (offset == XDP_PGOFF_TX_RING) {
 665                 q = READ_ONCE(xs->tx);
 666         } else {
 667                 umem = READ_ONCE(xs->umem);
 668                 if (!umem)
 669                         return -EINVAL;
 670
 671                 if (offset == XDP_UMEM_PGOFF_FILL_RING)
 672                         q = READ_ONCE(umem->fq);
 673                 else if (offset == XDP_UMEM_PGOFF_COMPLETION_RING)
 674                         q = READ_ONCE(umem->cq);
 675         }
 676
 677         if (!q)
 678                 return -EINVAL;
 679
 680         qpg = virt_to_head_page(q->ring);
 681         if (size > (PAGE_SIZE << compound_order(qpg)))
 682                 return -EINVAL;
 683
 684         pfn = virt_to_phys(q->ring) >> PAGE_SHIFT;
 685         return remap_pfn_range(vma, vma->vm_start, pfn,
 686                                size, vma->vm_page_prot);
 687 }
 688
 689 static struct proto xsk_proto = {
 690         .name =         "XDP",
 691         .owner =        THIS_MODULE,
 692         .obj_size =     sizeof(struct xdp_sock),
 693 };
 694
 695 static const struct proto_ops xsk_proto_ops = {
 696         .family         = PF_XDP,
 697         .owner          = THIS_MODULE,
 698         .release        = xsk_release,
 699         .bind           = xsk_bind,
 700         .connect        = sock_no_connect,
 701         .socketpair     = sock_no_socketpair,
 702         .accept         = sock_no_accept,
 703         .getname        = sock_no_getname,
 704         .poll           = xsk_poll,
 705         .ioctl          = sock_no_ioctl,
 706         .listen         = sock_no_listen,
 707         .shutdown       = sock_no_shutdown,
 708         .setsockopt     = xsk_setsockopt,
 709         .getsockopt     = xsk_getsockopt,
 710         .sendmsg        = xsk_sendmsg,
 711         .recvmsg        = sock_no_recvmsg,
 712         .mmap           = xsk_mmap,
 713         .sendpage       = sock_no_sendpage,
 714 };
 715
 716 static void xsk_destruct(struct sock *sk)
 717 {
 718         struct xdp_sock *xs = xdp_sk(sk);
 719
 720         if (!sock_flag(sk, SOCK_DEAD))
 721                 return;
 722
 723         xdp_put_umem(xs->umem);
 724
 725         sk_refcnt_debug_dec(sk);
 726 }
 727
 728 static int xsk_create(struct net *net, struct socket *sock, int protocol,
 729                       int kern)
 730 {
 731         struct sock *sk;
 732         struct xdp_sock *xs;
 733
 734         if (!ns_capable(net->user_ns, CAP_NET_RAW))
 735                 return -EPERM;
 736         if (sock->type != SOCK_RAW)
 737                 return -ESOCKTNOSUPPORT;
 738
 739         if (protocol)
 740                 return -EPROTONOSUPPORT;
 741
 742         sock->state = SS_UNCONNECTED;
 743
 744         sk = sk_alloc(net, PF_XDP, GFP_KERNEL, &xsk_proto, kern);
 745         if (!sk)
 746                 return -ENOBUFS;
 747
 748         sock->ops = &xsk_proto_ops;
 749
 750         sock_init_data(sock, sk);
 751
 752         sk->sk_family = PF_XDP;
 753
 754         sk->sk_destruct = xsk_destruct;
 755         sk_refcnt_debug_inc(sk);
 756
 757         sock_set_flag(sk, SOCK_RCU_FREE);
 758
 759         xs = xdp_sk(sk);
 760         mutex_init(&xs->mutex);
 761         spin_lock_init(&xs->tx_completion_lock);
 762
 763         local_bh_disable();
 764         sock_prot_inuse_add(net, &xsk_proto, 1);
 765         local_bh_enable();
 766
 767         return 0;
 768 }
 769
 770 static const struct net_proto_family xsk_family_ops = {
 771         .family = PF_XDP,
 772         .create = xsk_create,
 773         .owner  = THIS_MODULE,
 774 };
 775
 776 static int __init xsk_init(void)
 777 {
 778         int err;
 779
 780         err = proto_register(&xsk_proto, 0 /* no slab */);
 781         if (err)
 782                 goto out;
 783
 784         err = sock_register(&xsk_family_ops);
 785         if (err)
 786                 goto out_proto;
 787
 788         return 0;
 789
 790 out_proto:
 791         proto_unregister(&xsk_proto);
 792 out:
 793         return err;
 794 }
 795
 796 fs_initcall(xsk_init);