linux/drivers/net/tap.c
<<
>>
Prefs
   1// SPDX-License-Identifier: GPL-2.0-only
   2#include <linux/etherdevice.h>
   3#include <linux/if_tap.h>
   4#include <linux/if_vlan.h>
   5#include <linux/interrupt.h>
   6#include <linux/nsproxy.h>
   7#include <linux/compat.h>
   8#include <linux/if_tun.h>
   9#include <linux/module.h>
  10#include <linux/skbuff.h>
  11#include <linux/cache.h>
  12#include <linux/sched/signal.h>
  13#include <linux/types.h>
  14#include <linux/slab.h>
  15#include <linux/wait.h>
  16#include <linux/cdev.h>
  17#include <linux/idr.h>
  18#include <linux/fs.h>
  19#include <linux/uio.h>
  20
  21#include <net/net_namespace.h>
  22#include <net/rtnetlink.h>
  23#include <net/sock.h>
  24#include <linux/virtio_net.h>
  25#include <linux/skb_array.h>
  26
  27#define TAP_IFFEATURES (IFF_VNET_HDR | IFF_MULTI_QUEUE)
  28
  29#define TAP_VNET_LE 0x80000000
  30#define TAP_VNET_BE 0x40000000
  31
  32#ifdef CONFIG_TUN_VNET_CROSS_LE
  33static inline bool tap_legacy_is_little_endian(struct tap_queue *q)
  34{
  35        return q->flags & TAP_VNET_BE ? false :
  36                virtio_legacy_is_little_endian();
  37}
  38
  39static long tap_get_vnet_be(struct tap_queue *q, int __user *sp)
  40{
  41        int s = !!(q->flags & TAP_VNET_BE);
  42
  43        if (put_user(s, sp))
  44                return -EFAULT;
  45
  46        return 0;
  47}
  48
  49static long tap_set_vnet_be(struct tap_queue *q, int __user *sp)
  50{
  51        int s;
  52
  53        if (get_user(s, sp))
  54                return -EFAULT;
  55
  56        if (s)
  57                q->flags |= TAP_VNET_BE;
  58        else
  59                q->flags &= ~TAP_VNET_BE;
  60
  61        return 0;
  62}
  63#else
  64static inline bool tap_legacy_is_little_endian(struct tap_queue *q)
  65{
  66        return virtio_legacy_is_little_endian();
  67}
  68
  69static long tap_get_vnet_be(struct tap_queue *q, int __user *argp)
  70{
  71        return -EINVAL;
  72}
  73
  74static long tap_set_vnet_be(struct tap_queue *q, int __user *argp)
  75{
  76        return -EINVAL;
  77}
  78#endif /* CONFIG_TUN_VNET_CROSS_LE */
  79
  80static inline bool tap_is_little_endian(struct tap_queue *q)
  81{
  82        return q->flags & TAP_VNET_LE ||
  83                tap_legacy_is_little_endian(q);
  84}
  85
  86static inline u16 tap16_to_cpu(struct tap_queue *q, __virtio16 val)
  87{
  88        return __virtio16_to_cpu(tap_is_little_endian(q), val);
  89}
  90
  91static inline __virtio16 cpu_to_tap16(struct tap_queue *q, u16 val)
  92{
  93        return __cpu_to_virtio16(tap_is_little_endian(q), val);
  94}
  95
  96static struct proto tap_proto = {
  97        .name = "tap",
  98        .owner = THIS_MODULE,
  99        .obj_size = sizeof(struct tap_queue),
 100};
 101
 102#define TAP_NUM_DEVS (1U << MINORBITS)
 103
 104static LIST_HEAD(major_list);
 105
 106struct major_info {
 107        struct rcu_head rcu;
 108        dev_t major;
 109        struct idr minor_idr;
 110        spinlock_t minor_lock;
 111        const char *device_name;
 112        struct list_head next;
 113};
 114
 115#define GOODCOPY_LEN 128
 116
 117static const struct proto_ops tap_socket_ops;
 118
 119#define RX_OFFLOADS (NETIF_F_GRO | NETIF_F_LRO)
 120#define TAP_FEATURES (NETIF_F_GSO | NETIF_F_SG | NETIF_F_FRAGLIST)
 121
 122static struct tap_dev *tap_dev_get_rcu(const struct net_device *dev)
 123{
 124        return rcu_dereference(dev->rx_handler_data);
 125}
 126
 127/*
 128 * RCU usage:
 129 * The tap_queue and the macvlan_dev are loosely coupled, the
 130 * pointers from one to the other can only be read while rcu_read_lock
 131 * or rtnl is held.
 132 *
 133 * Both the file and the macvlan_dev hold a reference on the tap_queue
 134 * through sock_hold(&q->sk). When the macvlan_dev goes away first,
 135 * q->vlan becomes inaccessible. When the files gets closed,
 136 * tap_get_queue() fails.
 137 *
 138 * There may still be references to the struct sock inside of the
 139 * queue from outbound SKBs, but these never reference back to the
 140 * file or the dev. The data structure is freed through __sk_free
 141 * when both our references and any pending SKBs are gone.
 142 */
 143
 144static int tap_enable_queue(struct tap_dev *tap, struct file *file,
 145                            struct tap_queue *q)
 146{
 147        int err = -EINVAL;
 148
 149        ASSERT_RTNL();
 150
 151        if (q->enabled)
 152                goto out;
 153
 154        err = 0;
 155        rcu_assign_pointer(tap->taps[tap->numvtaps], q);
 156        q->queue_index = tap->numvtaps;
 157        q->enabled = true;
 158
 159        tap->numvtaps++;
 160out:
 161        return err;
 162}
 163
 164/* Requires RTNL */
 165static int tap_set_queue(struct tap_dev *tap, struct file *file,
 166                         struct tap_queue *q)
 167{
 168        if (tap->numqueues == MAX_TAP_QUEUES)
 169                return -EBUSY;
 170
 171        rcu_assign_pointer(q->tap, tap);
 172        rcu_assign_pointer(tap->taps[tap->numvtaps], q);
 173        sock_hold(&q->sk);
 174
 175        q->file = file;
 176        q->queue_index = tap->numvtaps;
 177        q->enabled = true;
 178        file->private_data = q;
 179        list_add_tail(&q->next, &tap->queue_list);
 180
 181        tap->numvtaps++;
 182        tap->numqueues++;
 183
 184        return 0;
 185}
 186
 187static int tap_disable_queue(struct tap_queue *q)
 188{
 189        struct tap_dev *tap;
 190        struct tap_queue *nq;
 191
 192        ASSERT_RTNL();
 193        if (!q->enabled)
 194                return -EINVAL;
 195
 196        tap = rtnl_dereference(q->tap);
 197
 198        if (tap) {
 199                int index = q->queue_index;
 200                BUG_ON(index >= tap->numvtaps);
 201                nq = rtnl_dereference(tap->taps[tap->numvtaps - 1]);
 202                nq->queue_index = index;
 203
 204                rcu_assign_pointer(tap->taps[index], nq);
 205                RCU_INIT_POINTER(tap->taps[tap->numvtaps - 1], NULL);
 206                q->enabled = false;
 207
 208                tap->numvtaps--;
 209        }
 210
 211        return 0;
 212}
 213
 214/*
 215 * The file owning the queue got closed, give up both
 216 * the reference that the files holds as well as the
 217 * one from the macvlan_dev if that still exists.
 218 *
 219 * Using the spinlock makes sure that we don't get
 220 * to the queue again after destroying it.
 221 */
 222static void tap_put_queue(struct tap_queue *q)
 223{
 224        struct tap_dev *tap;
 225
 226        rtnl_lock();
 227        tap = rtnl_dereference(q->tap);
 228
 229        if (tap) {
 230                if (q->enabled)
 231                        BUG_ON(tap_disable_queue(q));
 232
 233                tap->numqueues--;
 234                RCU_INIT_POINTER(q->tap, NULL);
 235                sock_put(&q->sk);
 236                list_del_init(&q->next);
 237        }
 238
 239        rtnl_unlock();
 240
 241        synchronize_rcu();
 242        sock_put(&q->sk);
 243}
 244
 245/*
 246 * Select a queue based on the rxq of the device on which this packet
 247 * arrived. If the incoming device is not mq, calculate a flow hash
 248 * to select a queue. If all fails, find the first available queue.
 249 * Cache vlan->numvtaps since it can become zero during the execution
 250 * of this function.
 251 */
 252static struct tap_queue *tap_get_queue(struct tap_dev *tap,
 253                                       struct sk_buff *skb)
 254{
 255        struct tap_queue *queue = NULL;
 256        /* Access to taps array is protected by rcu, but access to numvtaps
 257         * isn't. Below we use it to lookup a queue, but treat it as a hint
 258         * and validate that the result isn't NULL - in case we are
 259         * racing against queue removal.
 260         */
 261        int numvtaps = READ_ONCE(tap->numvtaps);
 262        __u32 rxq;
 263
 264        if (!numvtaps)
 265                goto out;
 266
 267        if (numvtaps == 1)
 268                goto single;
 269
 270        /* Check if we can use flow to select a queue */
 271        rxq = skb_get_hash(skb);
 272        if (rxq) {
 273                queue = rcu_dereference(tap->taps[rxq % numvtaps]);
 274                goto out;
 275        }
 276
 277        if (likely(skb_rx_queue_recorded(skb))) {
 278                rxq = skb_get_rx_queue(skb);
 279
 280                while (unlikely(rxq >= numvtaps))
 281                        rxq -= numvtaps;
 282
 283                queue = rcu_dereference(tap->taps[rxq]);
 284                goto out;
 285        }
 286
 287single:
 288        queue = rcu_dereference(tap->taps[0]);
 289out:
 290        return queue;
 291}
 292
 293/*
 294 * The net_device is going away, give up the reference
 295 * that it holds on all queues and safely set the pointer
 296 * from the queues to NULL.
 297 */
 298void tap_del_queues(struct tap_dev *tap)
 299{
 300        struct tap_queue *q, *tmp;
 301
 302        ASSERT_RTNL();
 303        list_for_each_entry_safe(q, tmp, &tap->queue_list, next) {
 304                list_del_init(&q->next);
 305                RCU_INIT_POINTER(q->tap, NULL);
 306                if (q->enabled)
 307                        tap->numvtaps--;
 308                tap->numqueues--;
 309                sock_put(&q->sk);
 310        }
 311        BUG_ON(tap->numvtaps);
 312        BUG_ON(tap->numqueues);
 313        /* guarantee that any future tap_set_queue will fail */
 314        tap->numvtaps = MAX_TAP_QUEUES;
 315}
 316EXPORT_SYMBOL_GPL(tap_del_queues);
 317
 318rx_handler_result_t tap_handle_frame(struct sk_buff **pskb)
 319{
 320        struct sk_buff *skb = *pskb;
 321        struct net_device *dev = skb->dev;
 322        struct tap_dev *tap;
 323        struct tap_queue *q;
 324        netdev_features_t features = TAP_FEATURES;
 325
 326        tap = tap_dev_get_rcu(dev);
 327        if (!tap)
 328                return RX_HANDLER_PASS;
 329
 330        q = tap_get_queue(tap, skb);
 331        if (!q)
 332                return RX_HANDLER_PASS;
 333
 334        skb_push(skb, ETH_HLEN);
 335
 336        /* Apply the forward feature mask so that we perform segmentation
 337         * according to users wishes.  This only works if VNET_HDR is
 338         * enabled.
 339         */
 340        if (q->flags & IFF_VNET_HDR)
 341                features |= tap->tap_features;
 342        if (netif_needs_gso(skb, features)) {
 343                struct sk_buff *segs = __skb_gso_segment(skb, features, false);
 344
 345                if (IS_ERR(segs))
 346                        goto drop;
 347
 348                if (!segs) {
 349                        if (ptr_ring_produce(&q->ring, skb))
 350                                goto drop;
 351                        goto wake_up;
 352                }
 353
 354                consume_skb(skb);
 355                while (segs) {
 356                        struct sk_buff *nskb = segs->next;
 357
 358                        segs->next = NULL;
 359                        if (ptr_ring_produce(&q->ring, segs)) {
 360                                kfree_skb(segs);
 361                                kfree_skb_list(nskb);
 362                                break;
 363                        }
 364                        segs = nskb;
 365                }
 366        } else {
 367                /* If we receive a partial checksum and the tap side
 368                 * doesn't support checksum offload, compute the checksum.
 369                 * Note: it doesn't matter which checksum feature to
 370                 *        check, we either support them all or none.
 371                 */
 372                if (skb->ip_summed == CHECKSUM_PARTIAL &&
 373                    !(features & NETIF_F_CSUM_MASK) &&
 374                    skb_checksum_help(skb))
 375                        goto drop;
 376                if (ptr_ring_produce(&q->ring, skb))
 377                        goto drop;
 378        }
 379
 380wake_up:
 381        wake_up_interruptible_poll(sk_sleep(&q->sk), EPOLLIN | EPOLLRDNORM | EPOLLRDBAND);
 382        return RX_HANDLER_CONSUMED;
 383
 384drop:
 385        /* Count errors/drops only here, thus don't care about args. */
 386        if (tap->count_rx_dropped)
 387                tap->count_rx_dropped(tap);
 388        kfree_skb(skb);
 389        return RX_HANDLER_CONSUMED;
 390}
 391EXPORT_SYMBOL_GPL(tap_handle_frame);
 392
 393static struct major_info *tap_get_major(int major)
 394{
 395        struct major_info *tap_major;
 396
 397        list_for_each_entry_rcu(tap_major, &major_list, next) {
 398                if (tap_major->major == major)
 399                        return tap_major;
 400        }
 401
 402        return NULL;
 403}
 404
 405int tap_get_minor(dev_t major, struct tap_dev *tap)
 406{
 407        int retval = -ENOMEM;
 408        struct major_info *tap_major;
 409
 410        rcu_read_lock();
 411        tap_major = tap_get_major(MAJOR(major));
 412        if (!tap_major) {
 413                retval = -EINVAL;
 414                goto unlock;
 415        }
 416
 417        spin_lock(&tap_major->minor_lock);
 418        retval = idr_alloc(&tap_major->minor_idr, tap, 1, TAP_NUM_DEVS, GFP_ATOMIC);
 419        if (retval >= 0) {
 420                tap->minor = retval;
 421        } else if (retval == -ENOSPC) {
 422                netdev_err(tap->dev, "Too many tap devices\n");
 423                retval = -EINVAL;
 424        }
 425        spin_unlock(&tap_major->minor_lock);
 426
 427unlock:
 428        rcu_read_unlock();
 429        return retval < 0 ? retval : 0;
 430}
 431EXPORT_SYMBOL_GPL(tap_get_minor);
 432
 433void tap_free_minor(dev_t major, struct tap_dev *tap)
 434{
 435        struct major_info *tap_major;
 436
 437        rcu_read_lock();
 438        tap_major = tap_get_major(MAJOR(major));
 439        if (!tap_major) {
 440                goto unlock;
 441        }
 442
 443        spin_lock(&tap_major->minor_lock);
 444        if (tap->minor) {
 445                idr_remove(&tap_major->minor_idr, tap->minor);
 446                tap->minor = 0;
 447        }
 448        spin_unlock(&tap_major->minor_lock);
 449
 450unlock:
 451        rcu_read_unlock();
 452}
 453EXPORT_SYMBOL_GPL(tap_free_minor);
 454
 455static struct tap_dev *dev_get_by_tap_file(int major, int minor)
 456{
 457        struct net_device *dev = NULL;
 458        struct tap_dev *tap;
 459        struct major_info *tap_major;
 460
 461        rcu_read_lock();
 462        tap_major = tap_get_major(major);
 463        if (!tap_major) {
 464                tap = NULL;
 465                goto unlock;
 466        }
 467
 468        spin_lock(&tap_major->minor_lock);
 469        tap = idr_find(&tap_major->minor_idr, minor);
 470        if (tap) {
 471                dev = tap->dev;
 472                dev_hold(dev);
 473        }
 474        spin_unlock(&tap_major->minor_lock);
 475
 476unlock:
 477        rcu_read_unlock();
 478        return tap;
 479}
 480
 481static void tap_sock_write_space(struct sock *sk)
 482{
 483        wait_queue_head_t *wqueue;
 484
 485        if (!sock_writeable(sk) ||
 486            !test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
 487                return;
 488
 489        wqueue = sk_sleep(sk);
 490        if (wqueue && waitqueue_active(wqueue))
 491                wake_up_interruptible_poll(wqueue, EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND);
 492}
 493
 494static void tap_sock_destruct(struct sock *sk)
 495{
 496        struct tap_queue *q = container_of(sk, struct tap_queue, sk);
 497
 498        ptr_ring_cleanup(&q->ring, __skb_array_destroy_skb);
 499}
 500
 501static int tap_open(struct inode *inode, struct file *file)
 502{
 503        struct net *net = current->nsproxy->net_ns;
 504        struct tap_dev *tap;
 505        struct tap_queue *q;
 506        int err = -ENODEV;
 507
 508        rtnl_lock();
 509        tap = dev_get_by_tap_file(imajor(inode), iminor(inode));
 510        if (!tap)
 511                goto err;
 512
 513        err = -ENOMEM;
 514        q = (struct tap_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
 515                                             &tap_proto, 0);
 516        if (!q)
 517                goto err;
 518        if (ptr_ring_init(&q->ring, tap->dev->tx_queue_len, GFP_KERNEL)) {
 519                sk_free(&q->sk);
 520                goto err;
 521        }
 522
 523        init_waitqueue_head(&q->sock.wq.wait);
 524        q->sock.type = SOCK_RAW;
 525        q->sock.state = SS_CONNECTED;
 526        q->sock.file = file;
 527        q->sock.ops = &tap_socket_ops;
 528        sock_init_data(&q->sock, &q->sk);
 529        q->sk.sk_write_space = tap_sock_write_space;
 530        q->sk.sk_destruct = tap_sock_destruct;
 531        q->flags = IFF_VNET_HDR | IFF_NO_PI | IFF_TAP;
 532        q->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
 533
 534        /*
 535         * so far only KVM virtio_net uses tap, enable zero copy between
 536         * guest kernel and host kernel when lower device supports zerocopy
 537         *
 538         * The macvlan supports zerocopy iff the lower device supports zero
 539         * copy so we don't have to look at the lower device directly.
 540         */
 541        if ((tap->dev->features & NETIF_F_HIGHDMA) && (tap->dev->features & NETIF_F_SG))
 542                sock_set_flag(&q->sk, SOCK_ZEROCOPY);
 543
 544        err = tap_set_queue(tap, file, q);
 545        if (err) {
 546                /* tap_sock_destruct() will take care of freeing ptr_ring */
 547                goto err_put;
 548        }
 549
 550        dev_put(tap->dev);
 551
 552        rtnl_unlock();
 553        return err;
 554
 555err_put:
 556        sock_put(&q->sk);
 557err:
 558        if (tap)
 559                dev_put(tap->dev);
 560
 561        rtnl_unlock();
 562        return err;
 563}
 564
 565static int tap_release(struct inode *inode, struct file *file)
 566{
 567        struct tap_queue *q = file->private_data;
 568        tap_put_queue(q);
 569        return 0;
 570}
 571
 572static __poll_t tap_poll(struct file *file, poll_table *wait)
 573{
 574        struct tap_queue *q = file->private_data;
 575        __poll_t mask = EPOLLERR;
 576
 577        if (!q)
 578                goto out;
 579
 580        mask = 0;
 581        poll_wait(file, &q->sock.wq.wait, wait);
 582
 583        if (!ptr_ring_empty(&q->ring))
 584                mask |= EPOLLIN | EPOLLRDNORM;
 585
 586        if (sock_writeable(&q->sk) ||
 587            (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &q->sock.flags) &&
 588             sock_writeable(&q->sk)))
 589                mask |= EPOLLOUT | EPOLLWRNORM;
 590
 591out:
 592        return mask;
 593}
 594
 595static inline struct sk_buff *tap_alloc_skb(struct sock *sk, size_t prepad,
 596                                            size_t len, size_t linear,
 597                                                int noblock, int *err)
 598{
 599        struct sk_buff *skb;
 600
 601        /* Under a page?  Don't bother with paged skb. */
 602        if (prepad + len < PAGE_SIZE || !linear)
 603                linear = len;
 604
 605        skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
 606                                   err, 0);
 607        if (!skb)
 608                return NULL;
 609
 610        skb_reserve(skb, prepad);
 611        skb_put(skb, linear);
 612        skb->data_len = len - linear;
 613        skb->len += len - linear;
 614
 615        return skb;
 616}
 617
 618/* Neighbour code has some assumptions on HH_DATA_MOD alignment */
 619#define TAP_RESERVE HH_DATA_OFF(ETH_HLEN)
 620
 621/* Get packet from user space buffer */
 622static ssize_t tap_get_user(struct tap_queue *q, void *msg_control,
 623                            struct iov_iter *from, int noblock)
 624{
 625        int good_linear = SKB_MAX_HEAD(TAP_RESERVE);
 626        struct sk_buff *skb;
 627        struct tap_dev *tap;
 628        unsigned long total_len = iov_iter_count(from);
 629        unsigned long len = total_len;
 630        int err;
 631        struct virtio_net_hdr vnet_hdr = { 0 };
 632        int vnet_hdr_len = 0;
 633        int copylen = 0;
 634        int depth;
 635        bool zerocopy = false;
 636        size_t linear;
 637
 638        if (q->flags & IFF_VNET_HDR) {
 639                vnet_hdr_len = READ_ONCE(q->vnet_hdr_sz);
 640
 641                err = -EINVAL;
 642                if (len < vnet_hdr_len)
 643                        goto err;
 644                len -= vnet_hdr_len;
 645
 646                err = -EFAULT;
 647                if (!copy_from_iter_full(&vnet_hdr, sizeof(vnet_hdr), from))
 648                        goto err;
 649                iov_iter_advance(from, vnet_hdr_len - sizeof(vnet_hdr));
 650                if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
 651                     tap16_to_cpu(q, vnet_hdr.csum_start) +
 652                     tap16_to_cpu(q, vnet_hdr.csum_offset) + 2 >
 653                             tap16_to_cpu(q, vnet_hdr.hdr_len))
 654                        vnet_hdr.hdr_len = cpu_to_tap16(q,
 655                                 tap16_to_cpu(q, vnet_hdr.csum_start) +
 656                                 tap16_to_cpu(q, vnet_hdr.csum_offset) + 2);
 657                err = -EINVAL;
 658                if (tap16_to_cpu(q, vnet_hdr.hdr_len) > len)
 659                        goto err;
 660        }
 661
 662        err = -EINVAL;
 663        if (unlikely(len < ETH_HLEN))
 664                goto err;
 665
 666        if (msg_control && sock_flag(&q->sk, SOCK_ZEROCOPY)) {
 667                struct iov_iter i;
 668
 669                copylen = vnet_hdr.hdr_len ?
 670                        tap16_to_cpu(q, vnet_hdr.hdr_len) : GOODCOPY_LEN;
 671                if (copylen > good_linear)
 672                        copylen = good_linear;
 673                else if (copylen < ETH_HLEN)
 674                        copylen = ETH_HLEN;
 675                linear = copylen;
 676                i = *from;
 677                iov_iter_advance(&i, copylen);
 678                if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
 679                        zerocopy = true;
 680        }
 681
 682        if (!zerocopy) {
 683                copylen = len;
 684                linear = tap16_to_cpu(q, vnet_hdr.hdr_len);
 685                if (linear > good_linear)
 686                        linear = good_linear;
 687                else if (linear < ETH_HLEN)
 688                        linear = ETH_HLEN;
 689        }
 690
 691        skb = tap_alloc_skb(&q->sk, TAP_RESERVE, copylen,
 692                            linear, noblock, &err);
 693        if (!skb)
 694                goto err;
 695
 696        if (zerocopy)
 697                err = zerocopy_sg_from_iter(skb, from);
 698        else
 699                err = skb_copy_datagram_from_iter(skb, 0, from, len);
 700
 701        if (err)
 702                goto err_kfree;
 703
 704        skb_set_network_header(skb, ETH_HLEN);
 705        skb_reset_mac_header(skb);
 706        skb->protocol = eth_hdr(skb)->h_proto;
 707
 708        if (vnet_hdr_len) {
 709                err = virtio_net_hdr_to_skb(skb, &vnet_hdr,
 710                                            tap_is_little_endian(q));
 711                if (err)
 712                        goto err_kfree;
 713        }
 714
 715        skb_probe_transport_header(skb);
 716
 717        /* Move network header to the right position for VLAN tagged packets */
 718        if ((skb->protocol == htons(ETH_P_8021Q) ||
 719             skb->protocol == htons(ETH_P_8021AD)) &&
 720            __vlan_get_protocol(skb, skb->protocol, &depth) != 0)
 721                skb_set_network_header(skb, depth);
 722
 723        rcu_read_lock();
 724        tap = rcu_dereference(q->tap);
 725        /* copy skb_ubuf_info for callback when skb has no error */
 726        if (zerocopy) {
 727                skb_shinfo(skb)->destructor_arg = msg_control;
 728                skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
 729                skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
 730        } else if (msg_control) {
 731                struct ubuf_info *uarg = msg_control;
 732                uarg->callback(uarg, false);
 733        }
 734
 735        if (tap) {
 736                skb->dev = tap->dev;
 737                dev_queue_xmit(skb);
 738        } else {
 739                kfree_skb(skb);
 740        }
 741        rcu_read_unlock();
 742
 743        return total_len;
 744
 745err_kfree:
 746        kfree_skb(skb);
 747
 748err:
 749        rcu_read_lock();
 750        tap = rcu_dereference(q->tap);
 751        if (tap && tap->count_tx_dropped)
 752                tap->count_tx_dropped(tap);
 753        rcu_read_unlock();
 754
 755        return err;
 756}
 757
 758static ssize_t tap_write_iter(struct kiocb *iocb, struct iov_iter *from)
 759{
 760        struct file *file = iocb->ki_filp;
 761        struct tap_queue *q = file->private_data;
 762
 763        return tap_get_user(q, NULL, from, file->f_flags & O_NONBLOCK);
 764}
 765
 766/* Put packet to the user space buffer */
 767static ssize_t tap_put_user(struct tap_queue *q,
 768                            const struct sk_buff *skb,
 769                            struct iov_iter *iter)
 770{
 771        int ret;
 772        int vnet_hdr_len = 0;
 773        int vlan_offset = 0;
 774        int total;
 775
 776        if (q->flags & IFF_VNET_HDR) {
 777                int vlan_hlen = skb_vlan_tag_present(skb) ? VLAN_HLEN : 0;
 778                struct virtio_net_hdr vnet_hdr;
 779
 780                vnet_hdr_len = READ_ONCE(q->vnet_hdr_sz);
 781                if (iov_iter_count(iter) < vnet_hdr_len)
 782                        return -EINVAL;
 783
 784                if (virtio_net_hdr_from_skb(skb, &vnet_hdr,
 785                                            tap_is_little_endian(q), true,
 786                                            vlan_hlen))
 787                        BUG();
 788
 789                if (copy_to_iter(&vnet_hdr, sizeof(vnet_hdr), iter) !=
 790                    sizeof(vnet_hdr))
 791                        return -EFAULT;
 792
 793                iov_iter_advance(iter, vnet_hdr_len - sizeof(vnet_hdr));
 794        }
 795        total = vnet_hdr_len;
 796        total += skb->len;
 797
 798        if (skb_vlan_tag_present(skb)) {
 799                struct {
 800                        __be16 h_vlan_proto;
 801                        __be16 h_vlan_TCI;
 802                } veth;
 803                veth.h_vlan_proto = skb->vlan_proto;
 804                veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
 805
 806                vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
 807                total += VLAN_HLEN;
 808
 809                ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
 810                if (ret || !iov_iter_count(iter))
 811                        goto done;
 812
 813                ret = copy_to_iter(&veth, sizeof(veth), iter);
 814                if (ret != sizeof(veth) || !iov_iter_count(iter))
 815                        goto done;
 816        }
 817
 818        ret = skb_copy_datagram_iter(skb, vlan_offset, iter,
 819                                     skb->len - vlan_offset);
 820
 821done:
 822        return ret ? ret : total;
 823}
 824
 825static ssize_t tap_do_read(struct tap_queue *q,
 826                           struct iov_iter *to,
 827                           int noblock, struct sk_buff *skb)
 828{
 829        DEFINE_WAIT(wait);
 830        ssize_t ret = 0;
 831
 832        if (!iov_iter_count(to)) {
 833                kfree_skb(skb);
 834                return 0;
 835        }
 836
 837        if (skb)
 838                goto put;
 839
 840        while (1) {
 841                if (!noblock)
 842                        prepare_to_wait(sk_sleep(&q->sk), &wait,
 843                                        TASK_INTERRUPTIBLE);
 844
 845                /* Read frames from the queue */
 846                skb = ptr_ring_consume(&q->ring);
 847                if (skb)
 848                        break;
 849                if (noblock) {
 850                        ret = -EAGAIN;
 851                        break;
 852                }
 853                if (signal_pending(current)) {
 854                        ret = -ERESTARTSYS;
 855                        break;
 856                }
 857                /* Nothing to read, let's sleep */
 858                schedule();
 859        }
 860        if (!noblock)
 861                finish_wait(sk_sleep(&q->sk), &wait);
 862
 863put:
 864        if (skb) {
 865                ret = tap_put_user(q, skb, to);
 866                if (unlikely(ret < 0))
 867                        kfree_skb(skb);
 868                else
 869                        consume_skb(skb);
 870        }
 871        return ret;
 872}
 873
 874static ssize_t tap_read_iter(struct kiocb *iocb, struct iov_iter *to)
 875{
 876        struct file *file = iocb->ki_filp;
 877        struct tap_queue *q = file->private_data;
 878        ssize_t len = iov_iter_count(to), ret;
 879
 880        ret = tap_do_read(q, to, file->f_flags & O_NONBLOCK, NULL);
 881        ret = min_t(ssize_t, ret, len);
 882        if (ret > 0)
 883                iocb->ki_pos = ret;
 884        return ret;
 885}
 886
 887static struct tap_dev *tap_get_tap_dev(struct tap_queue *q)
 888{
 889        struct tap_dev *tap;
 890
 891        ASSERT_RTNL();
 892        tap = rtnl_dereference(q->tap);
 893        if (tap)
 894                dev_hold(tap->dev);
 895
 896        return tap;
 897}
 898
 899static void tap_put_tap_dev(struct tap_dev *tap)
 900{
 901        dev_put(tap->dev);
 902}
 903
 904static int tap_ioctl_set_queue(struct file *file, unsigned int flags)
 905{
 906        struct tap_queue *q = file->private_data;
 907        struct tap_dev *tap;
 908        int ret;
 909
 910        tap = tap_get_tap_dev(q);
 911        if (!tap)
 912                return -EINVAL;
 913
 914        if (flags & IFF_ATTACH_QUEUE)
 915                ret = tap_enable_queue(tap, file, q);
 916        else if (flags & IFF_DETACH_QUEUE)
 917                ret = tap_disable_queue(q);
 918        else
 919                ret = -EINVAL;
 920
 921        tap_put_tap_dev(tap);
 922        return ret;
 923}
 924
 925static int set_offload(struct tap_queue *q, unsigned long arg)
 926{
 927        struct tap_dev *tap;
 928        netdev_features_t features;
 929        netdev_features_t feature_mask = 0;
 930
 931        tap = rtnl_dereference(q->tap);
 932        if (!tap)
 933                return -ENOLINK;
 934
 935        features = tap->dev->features;
 936
 937        if (arg & TUN_F_CSUM) {
 938                feature_mask = NETIF_F_HW_CSUM;
 939
 940                if (arg & (TUN_F_TSO4 | TUN_F_TSO6)) {
 941                        if (arg & TUN_F_TSO_ECN)
 942                                feature_mask |= NETIF_F_TSO_ECN;
 943                        if (arg & TUN_F_TSO4)
 944                                feature_mask |= NETIF_F_TSO;
 945                        if (arg & TUN_F_TSO6)
 946                                feature_mask |= NETIF_F_TSO6;
 947                }
 948        }
 949
 950        /* tun/tap driver inverts the usage for TSO offloads, where
 951         * setting the TSO bit means that the userspace wants to
 952         * accept TSO frames and turning it off means that user space
 953         * does not support TSO.
 954         * For tap, we have to invert it to mean the same thing.
 955         * When user space turns off TSO, we turn off GSO/LRO so that
 956         * user-space will not receive TSO frames.
 957         */
 958        if (feature_mask & (NETIF_F_TSO | NETIF_F_TSO6))
 959                features |= RX_OFFLOADS;
 960        else
 961                features &= ~RX_OFFLOADS;
 962
 963        /* tap_features are the same as features on tun/tap and
 964         * reflect user expectations.
 965         */
 966        tap->tap_features = feature_mask;
 967        if (tap->update_features)
 968                tap->update_features(tap, features);
 969
 970        return 0;
 971}
 972
 973/*
 974 * provide compatibility with generic tun/tap interface
 975 */
 976static long tap_ioctl(struct file *file, unsigned int cmd,
 977                      unsigned long arg)
 978{
 979        struct tap_queue *q = file->private_data;
 980        struct tap_dev *tap;
 981        void __user *argp = (void __user *)arg;
 982        struct ifreq __user *ifr = argp;
 983        unsigned int __user *up = argp;
 984        unsigned short u;
 985        int __user *sp = argp;
 986        struct sockaddr sa;
 987        int s;
 988        int ret;
 989
 990        switch (cmd) {
 991        case TUNSETIFF:
 992                /* ignore the name, just look at flags */
 993                if (get_user(u, &ifr->ifr_flags))
 994                        return -EFAULT;
 995
 996                ret = 0;
 997                if ((u & ~TAP_IFFEATURES) != (IFF_NO_PI | IFF_TAP))
 998                        ret = -EINVAL;
 999                else
1000                        q->flags = (q->flags & ~TAP_IFFEATURES) | u;
1001
1002                return ret;
1003
1004        case TUNGETIFF:
1005                rtnl_lock();
1006                tap = tap_get_tap_dev(q);
1007                if (!tap) {
1008                        rtnl_unlock();
1009                        return -ENOLINK;
1010                }
1011
1012                ret = 0;
1013                u = q->flags;
1014                if (copy_to_user(&ifr->ifr_name, tap->dev->name, IFNAMSIZ) ||
1015                    put_user(u, &ifr->ifr_flags))
1016                        ret = -EFAULT;
1017                tap_put_tap_dev(tap);
1018                rtnl_unlock();
1019                return ret;
1020
1021        case TUNSETQUEUE:
1022                if (get_user(u, &ifr->ifr_flags))
1023                        return -EFAULT;
1024                rtnl_lock();
1025                ret = tap_ioctl_set_queue(file, u);
1026                rtnl_unlock();
1027                return ret;
1028
1029        case TUNGETFEATURES:
1030                if (put_user(IFF_TAP | IFF_NO_PI | TAP_IFFEATURES, up))
1031                        return -EFAULT;
1032                return 0;
1033
1034        case TUNSETSNDBUF:
1035                if (get_user(s, sp))
1036                        return -EFAULT;
1037                if (s <= 0)
1038                        return -EINVAL;
1039
1040                q->sk.sk_sndbuf = s;
1041                return 0;
1042
1043        case TUNGETVNETHDRSZ:
1044                s = q->vnet_hdr_sz;
1045                if (put_user(s, sp))
1046                        return -EFAULT;
1047                return 0;
1048
1049        case TUNSETVNETHDRSZ:
1050                if (get_user(s, sp))
1051                        return -EFAULT;
1052                if (s < (int)sizeof(struct virtio_net_hdr))
1053                        return -EINVAL;
1054
1055                q->vnet_hdr_sz = s;
1056                return 0;
1057
1058        case TUNGETVNETLE:
1059                s = !!(q->flags & TAP_VNET_LE);
1060                if (put_user(s, sp))
1061                        return -EFAULT;
1062                return 0;
1063
1064        case TUNSETVNETLE:
1065                if (get_user(s, sp))
1066                        return -EFAULT;
1067                if (s)
1068                        q->flags |= TAP_VNET_LE;
1069                else
1070                        q->flags &= ~TAP_VNET_LE;
1071                return 0;
1072
1073        case TUNGETVNETBE:
1074                return tap_get_vnet_be(q, sp);
1075
1076        case TUNSETVNETBE:
1077                return tap_set_vnet_be(q, sp);
1078
1079        case TUNSETOFFLOAD:
1080                /* let the user check for future flags */
1081                if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 |
1082                            TUN_F_TSO_ECN | TUN_F_UFO))
1083                        return -EINVAL;
1084
1085                rtnl_lock();
1086                ret = set_offload(q, arg);
1087                rtnl_unlock();
1088                return ret;
1089
1090        case SIOCGIFHWADDR:
1091                rtnl_lock();
1092                tap = tap_get_tap_dev(q);
1093                if (!tap) {
1094                        rtnl_unlock();
1095                        return -ENOLINK;
1096                }
1097                ret = 0;
1098                u = tap->dev->type;
1099                if (copy_to_user(&ifr->ifr_name, tap->dev->name, IFNAMSIZ) ||
1100                    copy_to_user(&ifr->ifr_hwaddr.sa_data, tap->dev->dev_addr, ETH_ALEN) ||
1101                    put_user(u, &ifr->ifr_hwaddr.sa_family))
1102                        ret = -EFAULT;
1103                tap_put_tap_dev(tap);
1104                rtnl_unlock();
1105                return ret;
1106
1107        case SIOCSIFHWADDR:
1108                if (copy_from_user(&sa, &ifr->ifr_hwaddr, sizeof(sa)))
1109                        return -EFAULT;
1110                rtnl_lock();
1111                tap = tap_get_tap_dev(q);
1112                if (!tap) {
1113                        rtnl_unlock();
1114                        return -ENOLINK;
1115                }
1116                ret = dev_set_mac_address(tap->dev, &sa, NULL);
1117                tap_put_tap_dev(tap);
1118                rtnl_unlock();
1119                return ret;
1120
1121        default:
1122                return -EINVAL;
1123        }
1124}
1125
1126static const struct file_operations tap_fops = {
1127        .owner          = THIS_MODULE,
1128        .open           = tap_open,
1129        .release        = tap_release,
1130        .read_iter      = tap_read_iter,
1131        .write_iter     = tap_write_iter,
1132        .poll           = tap_poll,
1133        .llseek         = no_llseek,
1134        .unlocked_ioctl = tap_ioctl,
1135        .compat_ioctl   = compat_ptr_ioctl,
1136};
1137
1138static int tap_get_user_xdp(struct tap_queue *q, struct xdp_buff *xdp)
1139{
1140        struct tun_xdp_hdr *hdr = xdp->data_hard_start;
1141        struct virtio_net_hdr *gso = &hdr->gso;
1142        int buflen = hdr->buflen;
1143        int vnet_hdr_len = 0;
1144        struct tap_dev *tap;
1145        struct sk_buff *skb;
1146        int err, depth;
1147
1148        if (q->flags & IFF_VNET_HDR)
1149                vnet_hdr_len = READ_ONCE(q->vnet_hdr_sz);
1150
1151        skb = build_skb(xdp->data_hard_start, buflen);
1152        if (!skb) {
1153                err = -ENOMEM;
1154                goto err;
1155        }
1156
1157        skb_reserve(skb, xdp->data - xdp->data_hard_start);
1158        skb_put(skb, xdp->data_end - xdp->data);
1159
1160        skb_set_network_header(skb, ETH_HLEN);
1161        skb_reset_mac_header(skb);
1162        skb->protocol = eth_hdr(skb)->h_proto;
1163
1164        if (vnet_hdr_len) {
1165                err = virtio_net_hdr_to_skb(skb, gso, tap_is_little_endian(q));
1166                if (err)
1167                        goto err_kfree;
1168        }
1169
1170        /* Move network header to the right position for VLAN tagged packets */
1171        if ((skb->protocol == htons(ETH_P_8021Q) ||
1172             skb->protocol == htons(ETH_P_8021AD)) &&
1173            __vlan_get_protocol(skb, skb->protocol, &depth) != 0)
1174                skb_set_network_header(skb, depth);
1175
1176        rcu_read_lock();
1177        tap = rcu_dereference(q->tap);
1178        if (tap) {
1179                skb->dev = tap->dev;
1180                skb_probe_transport_header(skb);
1181                dev_queue_xmit(skb);
1182        } else {
1183                kfree_skb(skb);
1184        }
1185        rcu_read_unlock();
1186
1187        return 0;
1188
1189err_kfree:
1190        kfree_skb(skb);
1191err:
1192        rcu_read_lock();
1193        tap = rcu_dereference(q->tap);
1194        if (tap && tap->count_tx_dropped)
1195                tap->count_tx_dropped(tap);
1196        rcu_read_unlock();
1197        return err;
1198}
1199
1200static int tap_sendmsg(struct socket *sock, struct msghdr *m,
1201                       size_t total_len)
1202{
1203        struct tap_queue *q = container_of(sock, struct tap_queue, sock);
1204        struct tun_msg_ctl *ctl = m->msg_control;
1205        struct xdp_buff *xdp;
1206        int i;
1207
1208        if (ctl && (ctl->type == TUN_MSG_PTR)) {
1209                for (i = 0; i < ctl->num; i++) {
1210                        xdp = &((struct xdp_buff *)ctl->ptr)[i];
1211                        tap_get_user_xdp(q, xdp);
1212                }
1213                return 0;
1214        }
1215
1216        return tap_get_user(q, ctl ? ctl->ptr : NULL, &m->msg_iter,
1217                            m->msg_flags & MSG_DONTWAIT);
1218}
1219
1220static int tap_recvmsg(struct socket *sock, struct msghdr *m,
1221                       size_t total_len, int flags)
1222{
1223        struct tap_queue *q = container_of(sock, struct tap_queue, sock);
1224        struct sk_buff *skb = m->msg_control;
1225        int ret;
1226        if (flags & ~(MSG_DONTWAIT|MSG_TRUNC)) {
1227                kfree_skb(skb);
1228                return -EINVAL;
1229        }
1230        ret = tap_do_read(q, &m->msg_iter, flags & MSG_DONTWAIT, skb);
1231        if (ret > total_len) {
1232                m->msg_flags |= MSG_TRUNC;
1233                ret = flags & MSG_TRUNC ? ret : total_len;
1234        }
1235        return ret;
1236}
1237
1238static int tap_peek_len(struct socket *sock)
1239{
1240        struct tap_queue *q = container_of(sock, struct tap_queue,
1241                                               sock);
1242        return PTR_RING_PEEK_CALL(&q->ring, __skb_array_len_with_tag);
1243}
1244
1245/* Ops structure to mimic raw sockets with tun */
1246static const struct proto_ops tap_socket_ops = {
1247        .sendmsg = tap_sendmsg,
1248        .recvmsg = tap_recvmsg,
1249        .peek_len = tap_peek_len,
1250};
1251
1252/* Get an underlying socket object from tun file.  Returns error unless file is
1253 * attached to a device.  The returned object works like a packet socket, it
1254 * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
1255 * holding a reference to the file for as long as the socket is in use. */
1256struct socket *tap_get_socket(struct file *file)
1257{
1258        struct tap_queue *q;
1259        if (file->f_op != &tap_fops)
1260                return ERR_PTR(-EINVAL);
1261        q = file->private_data;
1262        if (!q)
1263                return ERR_PTR(-EBADFD);
1264        return &q->sock;
1265}
1266EXPORT_SYMBOL_GPL(tap_get_socket);
1267
1268struct ptr_ring *tap_get_ptr_ring(struct file *file)
1269{
1270        struct tap_queue *q;
1271
1272        if (file->f_op != &tap_fops)
1273                return ERR_PTR(-EINVAL);
1274        q = file->private_data;
1275        if (!q)
1276                return ERR_PTR(-EBADFD);
1277        return &q->ring;
1278}
1279EXPORT_SYMBOL_GPL(tap_get_ptr_ring);
1280
1281int tap_queue_resize(struct tap_dev *tap)
1282{
1283        struct net_device *dev = tap->dev;
1284        struct tap_queue *q;
1285        struct ptr_ring **rings;
1286        int n = tap->numqueues;
1287        int ret, i = 0;
1288
1289        rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
1290        if (!rings)
1291                return -ENOMEM;
1292
1293        list_for_each_entry(q, &tap->queue_list, next)
1294                rings[i++] = &q->ring;
1295
1296        ret = ptr_ring_resize_multiple(rings, n,
1297                                       dev->tx_queue_len, GFP_KERNEL,
1298                                       __skb_array_destroy_skb);
1299
1300        kfree(rings);
1301        return ret;
1302}
1303EXPORT_SYMBOL_GPL(tap_queue_resize);
1304
1305static int tap_list_add(dev_t major, const char *device_name)
1306{
1307        struct major_info *tap_major;
1308
1309        tap_major = kzalloc(sizeof(*tap_major), GFP_ATOMIC);
1310        if (!tap_major)
1311                return -ENOMEM;
1312
1313        tap_major->major = MAJOR(major);
1314
1315        idr_init(&tap_major->minor_idr);
1316        spin_lock_init(&tap_major->minor_lock);
1317
1318        tap_major->device_name = device_name;
1319
1320        list_add_tail_rcu(&tap_major->next, &major_list);
1321        return 0;
1322}
1323
1324int tap_create_cdev(struct cdev *tap_cdev, dev_t *tap_major,
1325                    const char *device_name, struct module *module)
1326{
1327        int err;
1328
1329        err = alloc_chrdev_region(tap_major, 0, TAP_NUM_DEVS, device_name);
1330        if (err)
1331                goto out1;
1332
1333        cdev_init(tap_cdev, &tap_fops);
1334        tap_cdev->owner = module;
1335        err = cdev_add(tap_cdev, *tap_major, TAP_NUM_DEVS);
1336        if (err)
1337                goto out2;
1338
1339        err =  tap_list_add(*tap_major, device_name);
1340        if (err)
1341                goto out3;
1342
1343        return 0;
1344
1345out3:
1346        cdev_del(tap_cdev);
1347out2:
1348        unregister_chrdev_region(*tap_major, TAP_NUM_DEVS);
1349out1:
1350        return err;
1351}
1352EXPORT_SYMBOL_GPL(tap_create_cdev);
1353
1354void tap_destroy_cdev(dev_t major, struct cdev *tap_cdev)
1355{
1356        struct major_info *tap_major, *tmp;
1357
1358        cdev_del(tap_cdev);
1359        unregister_chrdev_region(major, TAP_NUM_DEVS);
1360        list_for_each_entry_safe(tap_major, tmp, &major_list, next) {
1361                if (tap_major->major == MAJOR(major)) {
1362                        idr_destroy(&tap_major->minor_idr);
1363                        list_del_rcu(&tap_major->next);
1364                        kfree_rcu(tap_major, rcu);
1365                }
1366        }
1367}
1368EXPORT_SYMBOL_GPL(tap_destroy_cdev);
1369
1370MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>");
1371MODULE_AUTHOR("Sainath Grandhi <sainath.grandhi@intel.com>");
1372MODULE_LICENSE("GPL");
1373