drivers/net/xen-netback/netback.c

   1 /*
   2  * Back-end of the driver for virtual network devices. This portion of the
   3  * driver exports a 'unified' network-device interface that can be accessed
   4  * by any operating system that implements a compatible front end. A
   5  * reference front-end implementation can be found in:
   6  *  drivers/net/xen-netfront.c
   7  *
   8  * Copyright (c) 2002-2005, K A Fraser
   9  *
  10  * This program is free software; you can redistribute it and/or
  11  * modify it under the terms of the GNU General Public License version 2
  12  * as published by the Free Software Foundation; or, when distributed
  13  * separately from the Linux kernel or incorporated into other
  14  * software packages, subject to the following license:
  15  *
  16  * Permission is hereby granted, free of charge, to any person obtaining a copy
  17  * of this source file (the "Software"), to deal in the Software without
  18  * restriction, including without limitation the rights to use, copy, modify,
  19  * merge, publish, distribute, sublicense, and/or sell copies of the Software,
  20  * and to permit persons to whom the Software is furnished to do so, subject to
  21  * the following conditions:
  22  *
  23  * The above copyright notice and this permission notice shall be included in
  24  * all copies or substantial portions of the Software.
  25  *
  26  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  27  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  28  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  29  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  30  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
  31  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
  32  * IN THE SOFTWARE.
  33  */
  34
  35 #include "common.h"
  36
  37 #include <linux/kthread.h>
  38 #include <linux/if_vlan.h>
  39 #include <linux/udp.h>
  40 #include <linux/highmem.h>
  41
  42 #include <net/tcp.h>
  43
  44 #include <xen/xen.h>
  45 #include <xen/events.h>
  46 #include <xen/interface/memory.h>
  47 #include <xen/page.h>
  48
  49 #include <asm/xen/hypercall.h>
  50
  51 /* Provide an option to disable split event channels at load time as
  52  * event channels are limited resource. Split event channels are
  53  * enabled by default.
  54  */
  55 bool separate_tx_rx_irq = true;
  56 module_param(separate_tx_rx_irq, bool, 0644);
  57
  58 /* The time that packets can stay on the guest Rx internal queue
  59  * before they are dropped.
  60  */
  61 unsigned int rx_drain_timeout_msecs = 10000;
  62 module_param(rx_drain_timeout_msecs, uint, 0444);
  63
  64 /* The length of time before the frontend is considered unresponsive
  65  * because it isn't providing Rx slots.
  66  */
  67 unsigned int rx_stall_timeout_msecs = 60000;
  68 module_param(rx_stall_timeout_msecs, uint, 0444);
  69
  70 #define MAX_QUEUES_DEFAULT 8
  71 unsigned int xenvif_max_queues;
  72 module_param_named(max_queues, xenvif_max_queues, uint, 0644);
  73 MODULE_PARM_DESC(max_queues,
  74                  "Maximum number of queues per virtual interface");
  75
  76 /*
  77  * This is the maximum slots a skb can have. If a guest sends a skb
  78  * which exceeds this limit it is considered malicious.
  79  */
  80 #define FATAL_SKB_SLOTS_DEFAULT 20
  81 static unsigned int fatal_skb_slots = FATAL_SKB_SLOTS_DEFAULT;
  82 module_param(fatal_skb_slots, uint, 0444);
  83
  84 /* The amount to copy out of the first guest Tx slot into the skb's
  85  * linear area.  If the first slot has more data, it will be mapped
  86  * and put into the first frag.
  87  *
  88  * This is sized to avoid pulling headers from the frags for most
  89  * TCP/IP packets.
  90  */
  91 #define XEN_NETBACK_TX_COPY_LEN 128
  92
  93 /* This is the maximum number of flows in the hash cache. */
  94 #define XENVIF_HASH_CACHE_SIZE_DEFAULT 64
  95 unsigned int xenvif_hash_cache_size = XENVIF_HASH_CACHE_SIZE_DEFAULT;
  96 module_param_named(hash_cache_size, xenvif_hash_cache_size, uint, 0644);
  97 MODULE_PARM_DESC(hash_cache_size, "Number of flows in the hash cache");
  98
  99 /* The module parameter tells that we have to put data
 100  * for xen-netfront with the XDP_PACKET_HEADROOM offset
 101  * needed for XDP processing
 102  */
 103 bool provides_xdp_headroom = true;
 104 module_param(provides_xdp_headroom, bool, 0644);
 105
 106 static void xenvif_idx_release(struct xenvif_queue *queue, u16 pending_idx,
 107                                u8 status);
 108
 109 static void make_tx_response(struct xenvif_queue *queue,
 110                              struct xen_netif_tx_request *txp,
 111                              unsigned int extra_count,
 112                              s8       st);
 113 static void push_tx_responses(struct xenvif_queue *queue);
 114
 115 static void xenvif_idx_unmap(struct xenvif_queue *queue, u16 pending_idx);
 116
 117 static inline int tx_work_todo(struct xenvif_queue *queue);
 118
 119 static inline unsigned long idx_to_pfn(struct xenvif_queue *queue,
 120                                        u16 idx)
 121 {
 122         return page_to_pfn(queue->mmap_pages[idx]);
 123 }
 124
 125 static inline unsigned long idx_to_kaddr(struct xenvif_queue *queue,
 126                                          u16 idx)
 127 {
 128         return (unsigned long)pfn_to_kaddr(idx_to_pfn(queue, idx));
 129 }
 130
 131 #define callback_param(vif, pending_idx) \
 132         (vif->pending_tx_info[pending_idx].callback_struct)
 133
 134 /* Find the containing VIF's structure from a pointer in pending_tx_info array
 135  */
 136 static inline struct xenvif_queue *ubuf_to_queue(const struct ubuf_info_msgzc *ubuf)
 137 {
 138         u16 pending_idx = ubuf->desc;
 139         struct pending_tx_info *temp =
 140                 container_of(ubuf, struct pending_tx_info, callback_struct);
 141         return container_of(temp - pending_idx,
 142                             struct xenvif_queue,
 143                             pending_tx_info[0]);
 144 }
 145
 146 static u16 frag_get_pending_idx(skb_frag_t *frag)
 147 {
 148         return (u16)skb_frag_off(frag);
 149 }
 150
 151 static void frag_set_pending_idx(skb_frag_t *frag, u16 pending_idx)
 152 {
 153         skb_frag_off_set(frag, pending_idx);
 154 }
 155
 156 static inline pending_ring_idx_t pending_index(unsigned i)
 157 {
 158         return i & (MAX_PENDING_REQS-1);
 159 }
 160
 161 void xenvif_kick_thread(struct xenvif_queue *queue)
 162 {
 163         wake_up(&queue->wq);
 164 }
 165
 166 void xenvif_napi_schedule_or_enable_events(struct xenvif_queue *queue)
 167 {
 168         int more_to_do;
 169
 170         RING_FINAL_CHECK_FOR_REQUESTS(&queue->tx, more_to_do);
 171
 172         if (more_to_do)
 173                 napi_schedule(&queue->napi);
 174         else if (atomic_fetch_andnot(NETBK_TX_EOI | NETBK_COMMON_EOI,
 175                                      &queue->eoi_pending) &
 176                  (NETBK_TX_EOI | NETBK_COMMON_EOI))
 177                 xen_irq_lateeoi(queue->tx_irq, 0);
 178 }
 179
 180 static void tx_add_credit(struct xenvif_queue *queue)
 181 {
 182         unsigned long max_burst, max_credit;
 183
 184         /*
 185          * Allow a burst big enough to transmit a jumbo packet of up to 128kB.
 186          * Otherwise the interface can seize up due to insufficient credit.
 187          */
 188         max_burst = max(131072UL, queue->credit_bytes);
 189
 190         /* Take care that adding a new chunk of credit doesn't wrap to zero. */
 191         max_credit = queue->remaining_credit + queue->credit_bytes;
 192         if (max_credit < queue->remaining_credit)
 193                 max_credit = ULONG_MAX; /* wrapped: clamp to ULONG_MAX */
 194
 195         queue->remaining_credit = min(max_credit, max_burst);
 196         queue->rate_limited = false;
 197 }
 198
 199 void xenvif_tx_credit_callback(struct timer_list *t)
 200 {
 201         struct xenvif_queue *queue = from_timer(queue, t, credit_timeout);
 202         tx_add_credit(queue);
 203         xenvif_napi_schedule_or_enable_events(queue);
 204 }
 205
 206 static void xenvif_tx_err(struct xenvif_queue *queue,
 207                           struct xen_netif_tx_request *txp,
 208                           unsigned int extra_count, RING_IDX end)
 209 {
 210         RING_IDX cons = queue->tx.req_cons;
 211         unsigned long flags;
 212
 213         do {
 214                 spin_lock_irqsave(&queue->response_lock, flags);
 215                 make_tx_response(queue, txp, extra_count, XEN_NETIF_RSP_ERROR);
 216                 push_tx_responses(queue);
 217                 spin_unlock_irqrestore(&queue->response_lock, flags);
 218                 if (cons == end)
 219                         break;
 220                 RING_COPY_REQUEST(&queue->tx, cons++, txp);
 221                 extra_count = 0; /* only the first frag can have extras */
 222         } while (1);
 223         queue->tx.req_cons = cons;
 224 }
 225
 226 static void xenvif_fatal_tx_err(struct xenvif *vif)
 227 {
 228         netdev_err(vif->dev, "fatal error; disabling device\n");
 229         vif->disabled = true;
 230         /* Disable the vif from queue 0's kthread */
 231         if (vif->num_queues)
 232                 xenvif_kick_thread(&vif->queues[0]);
 233 }
 234
 235 static int xenvif_count_requests(struct xenvif_queue *queue,
 236                                  struct xen_netif_tx_request *first,
 237                                  unsigned int extra_count,
 238                                  struct xen_netif_tx_request *txp,
 239                                  int work_to_do)
 240 {
 241         RING_IDX cons = queue->tx.req_cons;
 242         int slots = 0;
 243         int drop_err = 0;
 244         int more_data;
 245
 246         if (!(first->flags & XEN_NETTXF_more_data))
 247                 return 0;
 248
 249         do {
 250                 struct xen_netif_tx_request dropped_tx = { 0 };
 251
 252                 if (slots >= work_to_do) {
 253                         netdev_err(queue->vif->dev,
 254                                    "Asked for %d slots but exceeds this limit\n",
 255                                    work_to_do);
 256                         xenvif_fatal_tx_err(queue->vif);
 257                         return -ENODATA;
 258                 }
 259
 260                 /* This guest is really using too many slots and
 261                  * considered malicious.
 262                  */
 263                 if (unlikely(slots >= fatal_skb_slots)) {
 264                         netdev_err(queue->vif->dev,
 265                                    "Malicious frontend using %d slots, threshold %u\n",
 266                                    slots, fatal_skb_slots);
 267                         xenvif_fatal_tx_err(queue->vif);
 268                         return -E2BIG;
 269                 }
 270
 271                 /* Xen network protocol had implicit dependency on
 272                  * MAX_SKB_FRAGS. XEN_NETBK_LEGACY_SLOTS_MAX is set to
 273                  * the historical MAX_SKB_FRAGS value 18 to honor the
 274                  * same behavior as before. Any packet using more than
 275                  * 18 slots but less than fatal_skb_slots slots is
 276                  * dropped
 277                  */
 278                 if (!drop_err && slots >= XEN_NETBK_LEGACY_SLOTS_MAX) {
 279                         if (net_ratelimit())
 280                                 netdev_dbg(queue->vif->dev,
 281                                            "Too many slots (%d) exceeding limit (%d), dropping packet\n",
 282                                            slots, XEN_NETBK_LEGACY_SLOTS_MAX);
 283                         drop_err = -E2BIG;
 284                 }
 285
 286                 if (drop_err)
 287                         txp = &dropped_tx;
 288
 289                 RING_COPY_REQUEST(&queue->tx, cons + slots, txp);
 290
 291                 /* If the guest submitted a frame >= 64 KiB then
 292                  * first->size overflowed and following slots will
 293                  * appear to be larger than the frame.
 294                  *
 295                  * This cannot be fatal error as there are buggy
 296                  * frontends that do this.
 297                  *
 298                  * Consume all slots and drop the packet.
 299                  */
 300                 if (!drop_err && txp->size > first->size) {
 301                         if (net_ratelimit())
 302                                 netdev_dbg(queue->vif->dev,
 303                                            "Invalid tx request, slot size %u > remaining size %u\n",
 304                                            txp->size, first->size);
 305                         drop_err = -EIO;
 306                 }
 307
 308                 first->size -= txp->size;
 309                 slots++;
 310
 311                 if (unlikely((txp->offset + txp->size) > XEN_PAGE_SIZE)) {
 312                         netdev_err(queue->vif->dev, "Cross page boundary, txp->offset: %u, size: %u\n",
 313                                  txp->offset, txp->size);
 314                         xenvif_fatal_tx_err(queue->vif);
 315                         return -EINVAL;
 316                 }
 317
 318                 more_data = txp->flags & XEN_NETTXF_more_data;
 319
 320                 if (!drop_err)
 321                         txp++;
 322
 323         } while (more_data);
 324
 325         if (drop_err) {
 326                 xenvif_tx_err(queue, first, extra_count, cons + slots);
 327                 return drop_err;
 328         }
 329
 330         return slots;
 331 }
 332
 333
 334 struct xenvif_tx_cb {
 335         u16 copy_pending_idx[XEN_NETBK_LEGACY_SLOTS_MAX + 1];
 336         u8 copy_count;
 337         u32 split_mask;
 338 };
 339
 340 #define XENVIF_TX_CB(skb) ((struct xenvif_tx_cb *)(skb)->cb)
 341 #define copy_pending_idx(skb, i) (XENVIF_TX_CB(skb)->copy_pending_idx[i])
 342 #define copy_count(skb) (XENVIF_TX_CB(skb)->copy_count)
 343
 344 static inline void xenvif_tx_create_map_op(struct xenvif_queue *queue,
 345                                            u16 pending_idx,
 346                                            struct xen_netif_tx_request *txp,
 347                                            unsigned int extra_count,
 348                                            struct gnttab_map_grant_ref *mop)
 349 {
 350         queue->pages_to_map[mop-queue->tx_map_ops] = queue->mmap_pages[pending_idx];
 351         gnttab_set_map_op(mop, idx_to_kaddr(queue, pending_idx),
 352                           GNTMAP_host_map | GNTMAP_readonly,
 353                           txp->gref, queue->vif->domid);
 354
 355         memcpy(&queue->pending_tx_info[pending_idx].req, txp,
 356                sizeof(*txp));
 357         queue->pending_tx_info[pending_idx].extra_count = extra_count;
 358 }
 359
 360 static inline struct sk_buff *xenvif_alloc_skb(unsigned int size)
 361 {
 362         struct sk_buff *skb =
 363                 alloc_skb(size + NET_SKB_PAD + NET_IP_ALIGN,
 364                           GFP_ATOMIC | __GFP_NOWARN);
 365
 366         BUILD_BUG_ON(sizeof(*XENVIF_TX_CB(skb)) > sizeof(skb->cb));
 367         if (unlikely(skb == NULL))
 368                 return NULL;
 369
 370         /* Packets passed to netif_rx() must have some headroom. */
 371         skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
 372
 373         /* Initialize it here to avoid later surprises */
 374         skb_shinfo(skb)->destructor_arg = NULL;
 375
 376         return skb;
 377 }
 378
 379 static void xenvif_get_requests(struct xenvif_queue *queue,
 380                                 struct sk_buff *skb,
 381                                 struct xen_netif_tx_request *first,
 382                                 struct xen_netif_tx_request *txfrags,
 383                                 unsigned *copy_ops,
 384                                 unsigned *map_ops,
 385                                 unsigned int frag_overflow,
 386                                 struct sk_buff *nskb,
 387                                 unsigned int extra_count,
 388                                 unsigned int data_len)
 389 {
 390         struct skb_shared_info *shinfo = skb_shinfo(skb);
 391         skb_frag_t *frags = shinfo->frags;
 392         u16 pending_idx;
 393         pending_ring_idx_t index;
 394         unsigned int nr_slots;
 395         struct gnttab_copy *cop = queue->tx_copy_ops + *copy_ops;
 396         struct gnttab_map_grant_ref *gop = queue->tx_map_ops + *map_ops;
 397         struct xen_netif_tx_request *txp = first;
 398
 399         nr_slots = shinfo->nr_frags + 1;
 400
 401         copy_count(skb) = 0;
 402         XENVIF_TX_CB(skb)->split_mask = 0;
 403
 404         /* Create copy ops for exactly data_len bytes into the skb head. */
 405         __skb_put(skb, data_len);
 406         while (data_len > 0) {
 407                 int amount = data_len > txp->size ? txp->size : data_len;
 408                 bool split = false;
 409
 410                 cop->source.u.ref = txp->gref;
 411                 cop->source.domid = queue->vif->domid;
 412                 cop->source.offset = txp->offset;
 413
 414                 cop->dest.domid = DOMID_SELF;
 415                 cop->dest.offset = (offset_in_page(skb->data +
 416                                                    skb_headlen(skb) -
 417                                                    data_len)) & ~XEN_PAGE_MASK;
 418                 cop->dest.u.gmfn = virt_to_gfn(skb->data + skb_headlen(skb)
 419                                                - data_len);
 420
 421                 /* Don't cross local page boundary! */
 422                 if (cop->dest.offset + amount > XEN_PAGE_SIZE) {
 423                         amount = XEN_PAGE_SIZE - cop->dest.offset;
 424                         XENVIF_TX_CB(skb)->split_mask |= 1U << copy_count(skb);
 425                         split = true;
 426                 }
 427
 428                 cop->len = amount;
 429                 cop->flags = GNTCOPY_source_gref;
 430
 431                 index = pending_index(queue->pending_cons);
 432                 pending_idx = queue->pending_ring[index];
 433                 callback_param(queue, pending_idx).ctx = NULL;
 434                 copy_pending_idx(skb, copy_count(skb)) = pending_idx;
 435                 if (!split)
 436                         copy_count(skb)++;
 437
 438                 cop++;
 439                 data_len -= amount;
 440
 441                 if (amount == txp->size) {
 442                         /* The copy op covered the full tx_request */
 443
 444                         memcpy(&queue->pending_tx_info[pending_idx].req,
 445                                txp, sizeof(*txp));
 446                         queue->pending_tx_info[pending_idx].extra_count =
 447                                 (txp == first) ? extra_count : 0;
 448
 449                         if (txp == first)
 450                                 txp = txfrags;
 451                         else
 452                                 txp++;
 453                         queue->pending_cons++;
 454                         nr_slots--;
 455                 } else {
 456                         /* The copy op partially covered the tx_request.
 457                          * The remainder will be mapped or copied in the next
 458                          * iteration.
 459                          */
 460                         txp->offset += amount;
 461                         txp->size -= amount;
 462                 }
 463         }
 464
 465         for (shinfo->nr_frags = 0; shinfo->nr_frags < nr_slots;
 466              shinfo->nr_frags++, gop++) {
 467                 index = pending_index(queue->pending_cons++);
 468                 pending_idx = queue->pending_ring[index];
 469                 xenvif_tx_create_map_op(queue, pending_idx, txp,
 470                                         txp == first ? extra_count : 0, gop);
 471                 frag_set_pending_idx(&frags[shinfo->nr_frags], pending_idx);
 472
 473                 if (txp == first)
 474                         txp = txfrags;
 475                 else
 476                         txp++;
 477         }
 478
 479         if (frag_overflow) {
 480
 481                 shinfo = skb_shinfo(nskb);
 482                 frags = shinfo->frags;
 483
 484                 for (shinfo->nr_frags = 0; shinfo->nr_frags < frag_overflow;
 485                      shinfo->nr_frags++, txp++, gop++) {
 486                         index = pending_index(queue->pending_cons++);
 487                         pending_idx = queue->pending_ring[index];
 488                         xenvif_tx_create_map_op(queue, pending_idx, txp, 0,
 489                                                 gop);
 490                         frag_set_pending_idx(&frags[shinfo->nr_frags],
 491                                              pending_idx);
 492                 }
 493
 494                 skb_shinfo(skb)->frag_list = nskb;
 495         }
 496
 497         (*copy_ops) = cop - queue->tx_copy_ops;
 498         (*map_ops) = gop - queue->tx_map_ops;
 499 }
 500
 501 static inline void xenvif_grant_handle_set(struct xenvif_queue *queue,
 502                                            u16 pending_idx,
 503                                            grant_handle_t handle)
 504 {
 505         if (unlikely(queue->grant_tx_handle[pending_idx] !=
 506                      NETBACK_INVALID_HANDLE)) {
 507                 netdev_err(queue->vif->dev,
 508                            "Trying to overwrite active handle! pending_idx: 0x%x\n",
 509                            pending_idx);
 510                 BUG();
 511         }
 512         queue->grant_tx_handle[pending_idx] = handle;
 513 }
 514
 515 static inline void xenvif_grant_handle_reset(struct xenvif_queue *queue,
 516                                              u16 pending_idx)
 517 {
 518         if (unlikely(queue->grant_tx_handle[pending_idx] ==
 519                      NETBACK_INVALID_HANDLE)) {
 520                 netdev_err(queue->vif->dev,
 521                            "Trying to unmap invalid handle! pending_idx: 0x%x\n",
 522                            pending_idx);
 523                 BUG();
 524         }
 525         queue->grant_tx_handle[pending_idx] = NETBACK_INVALID_HANDLE;
 526 }
 527
 528 static int xenvif_tx_check_gop(struct xenvif_queue *queue,
 529                                struct sk_buff *skb,
 530                                struct gnttab_map_grant_ref **gopp_map,
 531                                struct gnttab_copy **gopp_copy)
 532 {
 533         struct gnttab_map_grant_ref *gop_map = *gopp_map;
 534         u16 pending_idx;
 535         /* This always points to the shinfo of the skb being checked, which
 536          * could be either the first or the one on the frag_list
 537          */
 538         struct skb_shared_info *shinfo = skb_shinfo(skb);
 539         /* If this is non-NULL, we are currently checking the frag_list skb, and
 540          * this points to the shinfo of the first one
 541          */
 542         struct skb_shared_info *first_shinfo = NULL;
 543         int nr_frags = shinfo->nr_frags;
 544         const bool sharedslot = nr_frags &&
 545                                 frag_get_pending_idx(&shinfo->frags[0]) ==
 546                                     copy_pending_idx(skb, copy_count(skb) - 1);
 547         int i, err = 0;
 548
 549         for (i = 0; i < copy_count(skb); i++) {
 550                 int newerr;
 551
 552                 /* Check status of header. */
 553                 pending_idx = copy_pending_idx(skb, i);
 554
 555                 newerr = (*gopp_copy)->status;
 556
 557                 /* Split copies need to be handled together. */
 558                 if (XENVIF_TX_CB(skb)->split_mask & (1U << i)) {
 559                         (*gopp_copy)++;
 560                         if (!newerr)
 561                                 newerr = (*gopp_copy)->status;
 562                 }
 563                 if (likely(!newerr)) {
 564                         /* The first frag might still have this slot mapped */
 565                         if (i < copy_count(skb) - 1 || !sharedslot)
 566                                 xenvif_idx_release(queue, pending_idx,
 567                                                    XEN_NETIF_RSP_OKAY);
 568                 } else {
 569                         err = newerr;
 570                         if (net_ratelimit())
 571                                 netdev_dbg(queue->vif->dev,
 572                                            "Grant copy of header failed! status: %d pending_idx: %u ref: %u\n",
 573                                            (*gopp_copy)->status,
 574                                            pending_idx,
 575                                            (*gopp_copy)->source.u.ref);
 576                         /* The first frag might still have this slot mapped */
 577                         if (i < copy_count(skb) - 1 || !sharedslot)
 578                                 xenvif_idx_release(queue, pending_idx,
 579                                                    XEN_NETIF_RSP_ERROR);
 580                 }
 581                 (*gopp_copy)++;
 582         }
 583
 584 check_frags:
 585         for (i = 0; i < nr_frags; i++, gop_map++) {
 586                 int j, newerr;
 587
 588                 pending_idx = frag_get_pending_idx(&shinfo->frags[i]);
 589
 590                 /* Check error status: if okay then remember grant handle. */
 591                 newerr = gop_map->status;
 592
 593                 if (likely(!newerr)) {
 594                         xenvif_grant_handle_set(queue,
 595                                                 pending_idx,
 596                                                 gop_map->handle);
 597                         /* Had a previous error? Invalidate this fragment. */
 598                         if (unlikely(err)) {
 599                                 xenvif_idx_unmap(queue, pending_idx);
 600                                 /* If the mapping of the first frag was OK, but
 601                                  * the header's copy failed, and they are
 602                                  * sharing a slot, send an error
 603                                  */
 604                                 if (i == 0 && !first_shinfo && sharedslot)
 605                                         xenvif_idx_release(queue, pending_idx,
 606                                                            XEN_NETIF_RSP_ERROR);
 607                                 else
 608                                         xenvif_idx_release(queue, pending_idx,
 609                                                            XEN_NETIF_RSP_OKAY);
 610                         }
 611                         continue;
 612                 }
 613
 614                 /* Error on this fragment: respond to client with an error. */
 615                 if (net_ratelimit())
 616                         netdev_dbg(queue->vif->dev,
 617                                    "Grant map of %d. frag failed! status: %d pending_idx: %u ref: %u\n",
 618                                    i,
 619                                    gop_map->status,
 620                                    pending_idx,
 621                                    gop_map->ref);
 622
 623                 xenvif_idx_release(queue, pending_idx, XEN_NETIF_RSP_ERROR);
 624
 625                 /* Not the first error? Preceding frags already invalidated. */
 626                 if (err)
 627                         continue;
 628
 629                 /* Invalidate preceding fragments of this skb. */
 630                 for (j = 0; j < i; j++) {
 631                         pending_idx = frag_get_pending_idx(&shinfo->frags[j]);
 632                         xenvif_idx_unmap(queue, pending_idx);
 633                         xenvif_idx_release(queue, pending_idx,
 634                                            XEN_NETIF_RSP_OKAY);
 635                 }
 636
 637                 /* And if we found the error while checking the frag_list, unmap
 638                  * the first skb's frags
 639                  */
 640                 if (first_shinfo) {
 641                         for (j = 0; j < first_shinfo->nr_frags; j++) {
 642                                 pending_idx = frag_get_pending_idx(&first_shinfo->frags[j]);
 643                                 xenvif_idx_unmap(queue, pending_idx);
 644                                 xenvif_idx_release(queue, pending_idx,
 645                                                    XEN_NETIF_RSP_OKAY);
 646                         }
 647                 }
 648
 649                 /* Remember the error: invalidate all subsequent fragments. */
 650                 err = newerr;
 651         }
 652
 653         if (skb_has_frag_list(skb) && !first_shinfo) {
 654                 first_shinfo = shinfo;
 655                 shinfo = skb_shinfo(shinfo->frag_list);
 656                 nr_frags = shinfo->nr_frags;
 657
 658                 goto check_frags;
 659         }
 660
 661         *gopp_map = gop_map;
 662         return err;
 663 }
 664
 665 static void xenvif_fill_frags(struct xenvif_queue *queue, struct sk_buff *skb)
 666 {
 667         struct skb_shared_info *shinfo = skb_shinfo(skb);
 668         int nr_frags = shinfo->nr_frags;
 669         int i;
 670         u16 prev_pending_idx = INVALID_PENDING_IDX;
 671
 672         for (i = 0; i < nr_frags; i++) {
 673                 skb_frag_t *frag = shinfo->frags + i;
 674                 struct xen_netif_tx_request *txp;
 675                 struct page *page;
 676                 u16 pending_idx;
 677
 678                 pending_idx = frag_get_pending_idx(frag);
 679
 680                 /* If this is not the first frag, chain it to the previous*/
 681                 if (prev_pending_idx == INVALID_PENDING_IDX)
 682                         skb_shinfo(skb)->destructor_arg =
 683                                 &callback_param(queue, pending_idx);
 684                 else
 685                         callback_param(queue, prev_pending_idx).ctx =
 686                                 &callback_param(queue, pending_idx);
 687
 688                 callback_param(queue, pending_idx).ctx = NULL;
 689                 prev_pending_idx = pending_idx;
 690
 691                 txp = &queue->pending_tx_info[pending_idx].req;
 692                 page = virt_to_page(idx_to_kaddr(queue, pending_idx));
 693                 __skb_fill_page_desc(skb, i, page, txp->offset, txp->size);
 694                 skb->len += txp->size;
 695                 skb->data_len += txp->size;
 696                 skb->truesize += txp->size;
 697
 698                 /* Take an extra reference to offset network stack's put_page */
 699                 get_page(queue->mmap_pages[pending_idx]);
 700         }
 701 }
 702
 703 static int xenvif_get_extras(struct xenvif_queue *queue,
 704                              struct xen_netif_extra_info *extras,
 705                              unsigned int *extra_count,
 706                              int work_to_do)
 707 {
 708         struct xen_netif_extra_info extra;
 709         RING_IDX cons = queue->tx.req_cons;
 710
 711         do {
 712                 if (unlikely(work_to_do-- <= 0)) {
 713                         netdev_err(queue->vif->dev, "Missing extra info\n");
 714                         xenvif_fatal_tx_err(queue->vif);
 715                         return -EBADR;
 716                 }
 717
 718                 RING_COPY_REQUEST(&queue->tx, cons, &extra);
 719
 720                 queue->tx.req_cons = ++cons;
 721                 (*extra_count)++;
 722
 723                 if (unlikely(!extra.type ||
 724                              extra.type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
 725                         netdev_err(queue->vif->dev,
 726                                    "Invalid extra type: %d\n", extra.type);
 727                         xenvif_fatal_tx_err(queue->vif);
 728                         return -EINVAL;
 729                 }
 730
 731                 memcpy(&extras[extra.type - 1], &extra, sizeof(extra));
 732         } while (extra.flags & XEN_NETIF_EXTRA_FLAG_MORE);
 733
 734         return work_to_do;
 735 }
 736
 737 static int xenvif_set_skb_gso(struct xenvif *vif,
 738                               struct sk_buff *skb,
 739                               struct xen_netif_extra_info *gso)
 740 {
 741         if (!gso->u.gso.size) {
 742                 netdev_err(vif->dev, "GSO size must not be zero.\n");
 743                 xenvif_fatal_tx_err(vif);
 744                 return -EINVAL;
 745         }
 746
 747         switch (gso->u.gso.type) {
 748         case XEN_NETIF_GSO_TYPE_TCPV4:
 749                 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
 750                 break;
 751         case XEN_NETIF_GSO_TYPE_TCPV6:
 752                 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
 753                 break;
 754         default:
 755                 netdev_err(vif->dev, "Bad GSO type %d.\n", gso->u.gso.type);
 756                 xenvif_fatal_tx_err(vif);
 757                 return -EINVAL;
 758         }
 759
 760         skb_shinfo(skb)->gso_size = gso->u.gso.size;
 761         /* gso_segs will be calculated later */
 762
 763         return 0;
 764 }
 765
 766 static int checksum_setup(struct xenvif_queue *queue, struct sk_buff *skb)
 767 {
 768         bool recalculate_partial_csum = false;
 769
 770         /* A GSO SKB must be CHECKSUM_PARTIAL. However some buggy
 771          * peers can fail to set NETRXF_csum_blank when sending a GSO
 772          * frame. In this case force the SKB to CHECKSUM_PARTIAL and
 773          * recalculate the partial checksum.
 774          */
 775         if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) {
 776                 queue->stats.rx_gso_checksum_fixup++;
 777                 skb->ip_summed = CHECKSUM_PARTIAL;
 778                 recalculate_partial_csum = true;
 779         }
 780
 781         /* A non-CHECKSUM_PARTIAL SKB does not require setup. */
 782         if (skb->ip_summed != CHECKSUM_PARTIAL)
 783                 return 0;
 784
 785         return skb_checksum_setup(skb, recalculate_partial_csum);
 786 }
 787
 788 static bool tx_credit_exceeded(struct xenvif_queue *queue, unsigned size)
 789 {
 790         u64 now = get_jiffies_64();
 791         u64 next_credit = queue->credit_window_start +
 792                 msecs_to_jiffies(queue->credit_usec / 1000);
 793
 794         /* Timer could already be pending in rare cases. */
 795         if (timer_pending(&queue->credit_timeout)) {
 796                 queue->rate_limited = true;
 797                 return true;
 798         }
 799
 800         /* Passed the point where we can replenish credit? */
 801         if (time_after_eq64(now, next_credit)) {
 802                 queue->credit_window_start = now;
 803                 tx_add_credit(queue);
 804         }
 805
 806         /* Still too big to send right now? Set a callback. */
 807         if (size > queue->remaining_credit) {
 808                 mod_timer(&queue->credit_timeout,
 809                           next_credit);
 810                 queue->credit_window_start = next_credit;
 811                 queue->rate_limited = true;
 812
 813                 return true;
 814         }
 815
 816         return false;
 817 }
 818
 819 /* No locking is required in xenvif_mcast_add/del() as they are
 820  * only ever invoked from NAPI poll. An RCU list is used because
 821  * xenvif_mcast_match() is called asynchronously, during start_xmit.
 822  */
 823
 824 static int xenvif_mcast_add(struct xenvif *vif, const u8 *addr)
 825 {
 826         struct xenvif_mcast_addr *mcast;
 827
 828         if (vif->fe_mcast_count == XEN_NETBK_MCAST_MAX) {
 829                 if (net_ratelimit())
 830                         netdev_err(vif->dev,
 831                                    "Too many multicast addresses\n");
 832                 return -ENOSPC;
 833         }
 834
 835         mcast = kzalloc(sizeof(*mcast), GFP_ATOMIC);
 836         if (!mcast)
 837                 return -ENOMEM;
 838
 839         ether_addr_copy(mcast->addr, addr);
 840         list_add_tail_rcu(&mcast->entry, &vif->fe_mcast_addr);
 841         vif->fe_mcast_count++;
 842
 843         return 0;
 844 }
 845
 846 static void xenvif_mcast_del(struct xenvif *vif, const u8 *addr)
 847 {
 848         struct xenvif_mcast_addr *mcast;
 849
 850         list_for_each_entry_rcu(mcast, &vif->fe_mcast_addr, entry) {
 851                 if (ether_addr_equal(addr, mcast->addr)) {
 852                         --vif->fe_mcast_count;
 853                         list_del_rcu(&mcast->entry);
 854                         kfree_rcu(mcast, rcu);
 855                         break;
 856                 }
 857         }
 858 }
 859
 860 bool xenvif_mcast_match(struct xenvif *vif, const u8 *addr)
 861 {
 862         struct xenvif_mcast_addr *mcast;
 863
 864         rcu_read_lock();
 865         list_for_each_entry_rcu(mcast, &vif->fe_mcast_addr, entry) {
 866                 if (ether_addr_equal(addr, mcast->addr)) {
 867                         rcu_read_unlock();
 868                         return true;
 869                 }
 870         }
 871         rcu_read_unlock();
 872
 873         return false;
 874 }
 875
 876 void xenvif_mcast_addr_list_free(struct xenvif *vif)
 877 {
 878         /* No need for locking or RCU here. NAPI poll and TX queue
 879          * are stopped.
 880          */
 881         while (!list_empty(&vif->fe_mcast_addr)) {
 882                 struct xenvif_mcast_addr *mcast;
 883
 884                 mcast = list_first_entry(&vif->fe_mcast_addr,
 885                                          struct xenvif_mcast_addr,
 886                                          entry);
 887                 --vif->fe_mcast_count;
 888                 list_del(&mcast->entry);
 889                 kfree(mcast);
 890         }
 891 }
 892
 893 static void xenvif_tx_build_gops(struct xenvif_queue *queue,
 894                                      int budget,
 895                                      unsigned *copy_ops,
 896                                      unsigned *map_ops)
 897 {
 898         struct sk_buff *skb, *nskb;
 899         int ret;
 900         unsigned int frag_overflow;
 901
 902         while (skb_queue_len(&queue->tx_queue) < budget) {
 903                 struct xen_netif_tx_request txreq;
 904                 struct xen_netif_tx_request txfrags[XEN_NETBK_LEGACY_SLOTS_MAX];
 905                 struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX-1];
 906                 unsigned int extra_count;
 907                 RING_IDX idx;
 908                 int work_to_do;
 909                 unsigned int data_len;
 910
 911                 if (queue->tx.sring->req_prod - queue->tx.req_cons >
 912                     XEN_NETIF_TX_RING_SIZE) {
 913                         netdev_err(queue->vif->dev,
 914                                    "Impossible number of requests. "
 915                                    "req_prod %d, req_cons %d, size %ld\n",
 916                                    queue->tx.sring->req_prod, queue->tx.req_cons,
 917                                    XEN_NETIF_TX_RING_SIZE);
 918                         xenvif_fatal_tx_err(queue->vif);
 919                         break;
 920                 }
 921
 922                 work_to_do = XEN_RING_NR_UNCONSUMED_REQUESTS(&queue->tx);
 923                 if (!work_to_do)
 924                         break;
 925
 926                 idx = queue->tx.req_cons;
 927                 rmb(); /* Ensure that we see the request before we copy it. */
 928                 RING_COPY_REQUEST(&queue->tx, idx, &txreq);
 929
 930                 /* Credit-based scheduling. */
 931                 if (txreq.size > queue->remaining_credit &&
 932                     tx_credit_exceeded(queue, txreq.size))
 933                         break;
 934
 935                 queue->remaining_credit -= txreq.size;
 936
 937                 work_to_do--;
 938                 queue->tx.req_cons = ++idx;
 939
 940                 memset(extras, 0, sizeof(extras));
 941                 extra_count = 0;
 942                 if (txreq.flags & XEN_NETTXF_extra_info) {
 943                         work_to_do = xenvif_get_extras(queue, extras,
 944                                                        &extra_count,
 945                                                        work_to_do);
 946                         idx = queue->tx.req_cons;
 947                         if (unlikely(work_to_do < 0))
 948                                 break;
 949                 }
 950
 951                 if (extras[XEN_NETIF_EXTRA_TYPE_MCAST_ADD - 1].type) {
 952                         struct xen_netif_extra_info *extra;
 953
 954                         extra = &extras[XEN_NETIF_EXTRA_TYPE_MCAST_ADD - 1];
 955                         ret = xenvif_mcast_add(queue->vif, extra->u.mcast.addr);
 956
 957                         make_tx_response(queue, &txreq, extra_count,
 958                                          (ret == 0) ?
 959                                          XEN_NETIF_RSP_OKAY :
 960                                          XEN_NETIF_RSP_ERROR);
 961                         push_tx_responses(queue);
 962                         continue;
 963                 }
 964
 965                 if (extras[XEN_NETIF_EXTRA_TYPE_MCAST_DEL - 1].type) {
 966                         struct xen_netif_extra_info *extra;
 967
 968                         extra = &extras[XEN_NETIF_EXTRA_TYPE_MCAST_DEL - 1];
 969                         xenvif_mcast_del(queue->vif, extra->u.mcast.addr);
 970
 971                         make_tx_response(queue, &txreq, extra_count,
 972                                          XEN_NETIF_RSP_OKAY);
 973                         push_tx_responses(queue);
 974                         continue;
 975                 }
 976
 977                 data_len = (txreq.size > XEN_NETBACK_TX_COPY_LEN) ?
 978                         XEN_NETBACK_TX_COPY_LEN : txreq.size;
 979
 980                 ret = xenvif_count_requests(queue, &txreq, extra_count,
 981                                             txfrags, work_to_do);
 982
 983                 if (unlikely(ret < 0))
 984                         break;
 985
 986                 idx += ret;
 987
 988                 if (unlikely(txreq.size < ETH_HLEN)) {
 989                         netdev_dbg(queue->vif->dev,
 990                                    "Bad packet size: %d\n", txreq.size);
 991                         xenvif_tx_err(queue, &txreq, extra_count, idx);
 992                         break;
 993                 }
 994
 995                 /* No crossing a page as the payload mustn't fragment. */
 996                 if (unlikely((txreq.offset + txreq.size) > XEN_PAGE_SIZE)) {
 997                         netdev_err(queue->vif->dev, "Cross page boundary, txreq.offset: %u, size: %u\n",
 998                                    txreq.offset, txreq.size);
 999                         xenvif_fatal_tx_err(queue->vif);
1000                         break;
1001                 }
1002
1003                 if (ret >= XEN_NETBK_LEGACY_SLOTS_MAX - 1 && data_len < txreq.size)
1004                         data_len = txreq.size;
1005
1006                 skb = xenvif_alloc_skb(data_len);
1007                 if (unlikely(skb == NULL)) {
1008                         netdev_dbg(queue->vif->dev,
1009                                    "Can't allocate a skb in start_xmit.\n");
1010                         xenvif_tx_err(queue, &txreq, extra_count, idx);
1011                         break;
1012                 }
1013
1014                 skb_shinfo(skb)->nr_frags = ret;
1015                 /* At this point shinfo->nr_frags is in fact the number of
1016                  * slots, which can be as large as XEN_NETBK_LEGACY_SLOTS_MAX.
1017                  */
1018                 frag_overflow = 0;
1019                 nskb = NULL;
1020                 if (skb_shinfo(skb)->nr_frags > MAX_SKB_FRAGS) {
1021                         frag_overflow = skb_shinfo(skb)->nr_frags - MAX_SKB_FRAGS;
1022                         BUG_ON(frag_overflow > MAX_SKB_FRAGS);
1023                         skb_shinfo(skb)->nr_frags = MAX_SKB_FRAGS;
1024                         nskb = xenvif_alloc_skb(0);
1025                         if (unlikely(nskb == NULL)) {
1026                                 skb_shinfo(skb)->nr_frags = 0;
1027                                 kfree_skb(skb);
1028                                 xenvif_tx_err(queue, &txreq, extra_count, idx);
1029                                 if (net_ratelimit())
1030                                         netdev_err(queue->vif->dev,
1031                                                    "Can't allocate the frag_list skb.\n");
1032                                 break;
1033                         }
1034                 }
1035
1036                 if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) {
1037                         struct xen_netif_extra_info *gso;
1038                         gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1];
1039
1040                         if (xenvif_set_skb_gso(queue->vif, skb, gso)) {
1041                                 /* Failure in xenvif_set_skb_gso is fatal. */
1042                                 skb_shinfo(skb)->nr_frags = 0;
1043                                 kfree_skb(skb);
1044                                 kfree_skb(nskb);
1045                                 break;
1046                         }
1047                 }
1048
1049                 if (extras[XEN_NETIF_EXTRA_TYPE_HASH - 1].type) {
1050                         struct xen_netif_extra_info *extra;
1051                         enum pkt_hash_types type = PKT_HASH_TYPE_NONE;
1052
1053                         extra = &extras[XEN_NETIF_EXTRA_TYPE_HASH - 1];
1054
1055                         switch (extra->u.hash.type) {
1056                         case _XEN_NETIF_CTRL_HASH_TYPE_IPV4:
1057                         case _XEN_NETIF_CTRL_HASH_TYPE_IPV6:
1058                                 type = PKT_HASH_TYPE_L3;
1059                                 break;
1060
1061                         case _XEN_NETIF_CTRL_HASH_TYPE_IPV4_TCP:
1062                         case _XEN_NETIF_CTRL_HASH_TYPE_IPV6_TCP:
1063                                 type = PKT_HASH_TYPE_L4;
1064                                 break;
1065
1066                         default:
1067                                 break;
1068                         }
1069
1070                         if (type != PKT_HASH_TYPE_NONE)
1071                                 skb_set_hash(skb,
1072                                              *(u32 *)extra->u.hash.value,
1073                                              type);
1074                 }
1075
1076                 xenvif_get_requests(queue, skb, &txreq, txfrags, copy_ops,
1077                                     map_ops, frag_overflow, nskb, extra_count,
1078                                     data_len);
1079
1080                 __skb_queue_tail(&queue->tx_queue, skb);
1081
1082                 queue->tx.req_cons = idx;
1083         }
1084
1085         return;
1086 }
1087
1088 /* Consolidate skb with a frag_list into a brand new one with local pages on
1089  * frags. Returns 0 or -ENOMEM if can't allocate new pages.
1090  */
1091 static int xenvif_handle_frag_list(struct xenvif_queue *queue, struct sk_buff *skb)
1092 {
1093         unsigned int offset = skb_headlen(skb);
1094         skb_frag_t frags[MAX_SKB_FRAGS];
1095         int i, f;
1096         struct ubuf_info *uarg;
1097         struct sk_buff *nskb = skb_shinfo(skb)->frag_list;
1098
1099         queue->stats.tx_zerocopy_sent += 2;
1100         queue->stats.tx_frag_overflow++;
1101
1102         xenvif_fill_frags(queue, nskb);
1103         /* Subtract frags size, we will correct it later */
1104         skb->truesize -= skb->data_len;
1105         skb->len += nskb->len;
1106         skb->data_len += nskb->len;
1107
1108         /* create a brand new frags array and coalesce there */
1109         for (i = 0; offset < skb->len; i++) {
1110                 struct page *page;
1111                 unsigned int len;
1112
1113                 BUG_ON(i >= MAX_SKB_FRAGS);
1114                 page = alloc_page(GFP_ATOMIC);
1115                 if (!page) {
1116                         int j;
1117                         skb->truesize += skb->data_len;
1118                         for (j = 0; j < i; j++)
1119                                 put_page(skb_frag_page(&frags[j]));
1120                         return -ENOMEM;
1121                 }
1122
1123                 if (offset + PAGE_SIZE < skb->len)
1124                         len = PAGE_SIZE;
1125                 else
1126                         len = skb->len - offset;
1127                 if (skb_copy_bits(skb, offset, page_address(page), len))
1128                         BUG();
1129
1130                 offset += len;
1131                 __skb_frag_set_page(&frags[i], page);
1132                 skb_frag_off_set(&frags[i], 0);
1133                 skb_frag_size_set(&frags[i], len);
1134         }
1135
1136         /* Release all the original (foreign) frags. */
1137         for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
1138                 skb_frag_unref(skb, f);
1139         uarg = skb_shinfo(skb)->destructor_arg;
1140         /* increase inflight counter to offset decrement in callback */
1141         atomic_inc(&queue->inflight_packets);
1142         uarg->callback(NULL, uarg, true);
1143         skb_shinfo(skb)->destructor_arg = NULL;
1144
1145         /* Fill the skb with the new (local) frags. */
1146         memcpy(skb_shinfo(skb)->frags, frags, i * sizeof(skb_frag_t));
1147         skb_shinfo(skb)->nr_frags = i;
1148         skb->truesize += i * PAGE_SIZE;
1149
1150         return 0;
1151 }
1152
1153 static int xenvif_tx_submit(struct xenvif_queue *queue)
1154 {
1155         struct gnttab_map_grant_ref *gop_map = queue->tx_map_ops;
1156         struct gnttab_copy *gop_copy = queue->tx_copy_ops;
1157         struct sk_buff *skb;
1158         int work_done = 0;
1159
1160         while ((skb = __skb_dequeue(&queue->tx_queue)) != NULL) {
1161                 struct xen_netif_tx_request *txp;
1162                 u16 pending_idx;
1163
1164                 pending_idx = copy_pending_idx(skb, 0);
1165                 txp = &queue->pending_tx_info[pending_idx].req;
1166
1167                 /* Check the remap error code. */
1168                 if (unlikely(xenvif_tx_check_gop(queue, skb, &gop_map, &gop_copy))) {
1169                         /* If there was an error, xenvif_tx_check_gop is
1170                          * expected to release all the frags which were mapped,
1171                          * so kfree_skb shouldn't do it again
1172                          */
1173                         skb_shinfo(skb)->nr_frags = 0;
1174                         if (skb_has_frag_list(skb)) {
1175                                 struct sk_buff *nskb =
1176                                                 skb_shinfo(skb)->frag_list;
1177                                 skb_shinfo(nskb)->nr_frags = 0;
1178                         }
1179                         kfree_skb(skb);
1180                         continue;
1181                 }
1182
1183                 if (txp->flags & XEN_NETTXF_csum_blank)
1184                         skb->ip_summed = CHECKSUM_PARTIAL;
1185                 else if (txp->flags & XEN_NETTXF_data_validated)
1186                         skb->ip_summed = CHECKSUM_UNNECESSARY;
1187
1188                 xenvif_fill_frags(queue, skb);
1189
1190                 if (unlikely(skb_has_frag_list(skb))) {
1191                         struct sk_buff *nskb = skb_shinfo(skb)->frag_list;
1192                         xenvif_skb_zerocopy_prepare(queue, nskb);
1193                         if (xenvif_handle_frag_list(queue, skb)) {
1194                                 if (net_ratelimit())
1195                                         netdev_err(queue->vif->dev,
1196                                                    "Not enough memory to consolidate frag_list!\n");
1197                                 xenvif_skb_zerocopy_prepare(queue, skb);
1198                                 kfree_skb(skb);
1199                                 continue;
1200                         }
1201                         /* Copied all the bits from the frag list -- free it. */
1202                         skb_frag_list_init(skb);
1203                         kfree_skb(nskb);
1204                 }
1205
1206                 skb->dev      = queue->vif->dev;
1207                 skb->protocol = eth_type_trans(skb, skb->dev);
1208                 skb_reset_network_header(skb);
1209
1210                 if (checksum_setup(queue, skb)) {
1211                         netdev_dbg(queue->vif->dev,
1212                                    "Can't setup checksum in net_tx_action\n");
1213                         /* We have to set this flag to trigger the callback */
1214                         if (skb_shinfo(skb)->destructor_arg)
1215                                 xenvif_skb_zerocopy_prepare(queue, skb);
1216                         kfree_skb(skb);
1217                         continue;
1218                 }
1219
1220                 skb_probe_transport_header(skb);
1221
1222                 /* If the packet is GSO then we will have just set up the
1223                  * transport header offset in checksum_setup so it's now
1224                  * straightforward to calculate gso_segs.
1225                  */
1226                 if (skb_is_gso(skb)) {
1227                         int mss, hdrlen;
1228
1229                         /* GSO implies having the L4 header. */
1230                         WARN_ON_ONCE(!skb_transport_header_was_set(skb));
1231                         if (unlikely(!skb_transport_header_was_set(skb))) {
1232                                 kfree_skb(skb);
1233                                 continue;
1234                         }
1235
1236                         mss = skb_shinfo(skb)->gso_size;
1237                         hdrlen = skb_tcp_all_headers(skb);
1238
1239                         skb_shinfo(skb)->gso_segs =
1240                                 DIV_ROUND_UP(skb->len - hdrlen, mss);
1241                 }
1242
1243                 queue->stats.rx_bytes += skb->len;
1244                 queue->stats.rx_packets++;
1245
1246                 work_done++;
1247
1248                 /* Set this flag right before netif_receive_skb, otherwise
1249                  * someone might think this packet already left netback, and
1250                  * do a skb_copy_ubufs while we are still in control of the
1251                  * skb. E.g. the __pskb_pull_tail earlier can do such thing.
1252                  */
1253                 if (skb_shinfo(skb)->destructor_arg) {
1254                         xenvif_skb_zerocopy_prepare(queue, skb);
1255                         queue->stats.tx_zerocopy_sent++;
1256                 }
1257
1258                 netif_receive_skb(skb);
1259         }
1260
1261         return work_done;
1262 }
1263
1264 void xenvif_zerocopy_callback(struct sk_buff *skb, struct ubuf_info *ubuf_base,
1265                               bool zerocopy_success)
1266 {
1267         unsigned long flags;
1268         pending_ring_idx_t index;
1269         struct ubuf_info_msgzc *ubuf = uarg_to_msgzc(ubuf_base);
1270         struct xenvif_queue *queue = ubuf_to_queue(ubuf);
1271
1272         /* This is the only place where we grab this lock, to protect callbacks
1273          * from each other.
1274          */
1275         spin_lock_irqsave(&queue->callback_lock, flags);
1276         do {
1277                 u16 pending_idx = ubuf->desc;
1278                 ubuf = (struct ubuf_info_msgzc *) ubuf->ctx;
1279                 BUG_ON(queue->dealloc_prod - queue->dealloc_cons >=
1280                         MAX_PENDING_REQS);
1281                 index = pending_index(queue->dealloc_prod);
1282                 queue->dealloc_ring[index] = pending_idx;
1283                 /* Sync with xenvif_tx_dealloc_action:
1284                  * insert idx then incr producer.
1285                  */
1286                 smp_wmb();
1287                 queue->dealloc_prod++;
1288         } while (ubuf);
1289         spin_unlock_irqrestore(&queue->callback_lock, flags);
1290
1291         if (likely(zerocopy_success))
1292                 queue->stats.tx_zerocopy_success++;
1293         else
1294                 queue->stats.tx_zerocopy_fail++;
1295         xenvif_skb_zerocopy_complete(queue);
1296 }
1297
1298 static inline void xenvif_tx_dealloc_action(struct xenvif_queue *queue)
1299 {
1300         struct gnttab_unmap_grant_ref *gop;
1301         pending_ring_idx_t dc, dp;
1302         u16 pending_idx, pending_idx_release[MAX_PENDING_REQS];
1303         unsigned int i = 0;
1304
1305         dc = queue->dealloc_cons;
1306         gop = queue->tx_unmap_ops;
1307
1308         /* Free up any grants we have finished using */
1309         do {
1310                 dp = queue->dealloc_prod;
1311
1312                 /* Ensure we see all indices enqueued by all
1313                  * xenvif_zerocopy_callback().
1314                  */
1315                 smp_rmb();
1316
1317                 while (dc != dp) {
1318                         BUG_ON(gop - queue->tx_unmap_ops >= MAX_PENDING_REQS);
1319                         pending_idx =
1320                                 queue->dealloc_ring[pending_index(dc++)];
1321
1322                         pending_idx_release[gop - queue->tx_unmap_ops] =
1323                                 pending_idx;
1324                         queue->pages_to_unmap[gop - queue->tx_unmap_ops] =
1325                                 queue->mmap_pages[pending_idx];
1326                         gnttab_set_unmap_op(gop,
1327                                             idx_to_kaddr(queue, pending_idx),
1328                                             GNTMAP_host_map,
1329                                             queue->grant_tx_handle[pending_idx]);
1330                         xenvif_grant_handle_reset(queue, pending_idx);
1331                         ++gop;
1332                 }
1333
1334         } while (dp != queue->dealloc_prod);
1335
1336         queue->dealloc_cons = dc;
1337
1338         if (gop - queue->tx_unmap_ops > 0) {
1339                 int ret;
1340                 ret = gnttab_unmap_refs(queue->tx_unmap_ops,
1341                                         NULL,
1342                                         queue->pages_to_unmap,
1343                                         gop - queue->tx_unmap_ops);
1344                 if (ret) {
1345                         netdev_err(queue->vif->dev, "Unmap fail: nr_ops %tu ret %d\n",
1346                                    gop - queue->tx_unmap_ops, ret);
1347                         for (i = 0; i < gop - queue->tx_unmap_ops; ++i) {
1348                                 if (gop[i].status != GNTST_okay)
1349                                         netdev_err(queue->vif->dev,
1350                                                    " host_addr: 0x%llx handle: 0x%x status: %d\n",
1351                                                    gop[i].host_addr,
1352                                                    gop[i].handle,
1353                                                    gop[i].status);
1354                         }
1355                         BUG();
1356                 }
1357         }
1358
1359         for (i = 0; i < gop - queue->tx_unmap_ops; ++i)
1360                 xenvif_idx_release(queue, pending_idx_release[i],
1361                                    XEN_NETIF_RSP_OKAY);
1362 }
1363
1364
1365 /* Called after netfront has transmitted */
1366 int xenvif_tx_action(struct xenvif_queue *queue, int budget)
1367 {
1368         unsigned nr_mops = 0, nr_cops = 0;
1369         int work_done, ret;
1370
1371         if (unlikely(!tx_work_todo(queue)))
1372                 return 0;
1373
1374         xenvif_tx_build_gops(queue, budget, &nr_cops, &nr_mops);
1375
1376         if (nr_cops == 0)
1377                 return 0;
1378
1379         gnttab_batch_copy(queue->tx_copy_ops, nr_cops);
1380         if (nr_mops != 0) {
1381                 ret = gnttab_map_refs(queue->tx_map_ops,
1382                                       NULL,
1383                                       queue->pages_to_map,
1384                                       nr_mops);
1385                 if (ret) {
1386                         unsigned int i;
1387
1388                         netdev_err(queue->vif->dev, "Map fail: nr %u ret %d\n",
1389                                    nr_mops, ret);
1390                         for (i = 0; i < nr_mops; ++i)
1391                                 WARN_ON_ONCE(queue->tx_map_ops[i].status ==
1392                                              GNTST_okay);
1393                 }
1394         }
1395
1396         work_done = xenvif_tx_submit(queue);
1397
1398         return work_done;
1399 }
1400
1401 static void xenvif_idx_release(struct xenvif_queue *queue, u16 pending_idx,
1402                                u8 status)
1403 {
1404         struct pending_tx_info *pending_tx_info;
1405         pending_ring_idx_t index;
1406         unsigned long flags;
1407
1408         pending_tx_info = &queue->pending_tx_info[pending_idx];
1409
1410         spin_lock_irqsave(&queue->response_lock, flags);
1411
1412         make_tx_response(queue, &pending_tx_info->req,
1413                          pending_tx_info->extra_count, status);
1414
1415         /* Release the pending index before pusing the Tx response so
1416          * its available before a new Tx request is pushed by the
1417          * frontend.
1418          */
1419         index = pending_index(queue->pending_prod++);
1420         queue->pending_ring[index] = pending_idx;
1421
1422         push_tx_responses(queue);
1423
1424         spin_unlock_irqrestore(&queue->response_lock, flags);
1425 }
1426
1427
1428 static void make_tx_response(struct xenvif_queue *queue,
1429                              struct xen_netif_tx_request *txp,
1430                              unsigned int extra_count,
1431                              s8       st)
1432 {
1433         RING_IDX i = queue->tx.rsp_prod_pvt;
1434         struct xen_netif_tx_response *resp;
1435
1436         resp = RING_GET_RESPONSE(&queue->tx, i);
1437         resp->id     = txp->id;
1438         resp->status = st;
1439
1440         while (extra_count-- != 0)
1441                 RING_GET_RESPONSE(&queue->tx, ++i)->status = XEN_NETIF_RSP_NULL;
1442
1443         queue->tx.rsp_prod_pvt = ++i;
1444 }
1445
1446 static void push_tx_responses(struct xenvif_queue *queue)
1447 {
1448         int notify;
1449
1450         RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&queue->tx, notify);
1451         if (notify)
1452                 notify_remote_via_irq(queue->tx_irq);
1453 }
1454
1455 static void xenvif_idx_unmap(struct xenvif_queue *queue, u16 pending_idx)
1456 {
1457         int ret;
1458         struct gnttab_unmap_grant_ref tx_unmap_op;
1459
1460         gnttab_set_unmap_op(&tx_unmap_op,
1461                             idx_to_kaddr(queue, pending_idx),
1462                             GNTMAP_host_map,
1463                             queue->grant_tx_handle[pending_idx]);
1464         xenvif_grant_handle_reset(queue, pending_idx);
1465
1466         ret = gnttab_unmap_refs(&tx_unmap_op, NULL,
1467                                 &queue->mmap_pages[pending_idx], 1);
1468         if (ret) {
1469                 netdev_err(queue->vif->dev,
1470                            "Unmap fail: ret: %d pending_idx: %d host_addr: %llx handle: 0x%x status: %d\n",
1471                            ret,
1472                            pending_idx,
1473                            tx_unmap_op.host_addr,
1474                            tx_unmap_op.handle,
1475                            tx_unmap_op.status);
1476                 BUG();
1477         }
1478 }
1479
1480 static inline int tx_work_todo(struct xenvif_queue *queue)
1481 {
1482         if (likely(RING_HAS_UNCONSUMED_REQUESTS(&queue->tx)))
1483                 return 1;
1484
1485         return 0;
1486 }
1487
1488 static inline bool tx_dealloc_work_todo(struct xenvif_queue *queue)
1489 {
1490         return queue->dealloc_cons != queue->dealloc_prod;
1491 }
1492
1493 void xenvif_unmap_frontend_data_rings(struct xenvif_queue *queue)
1494 {
1495         if (queue->tx.sring)
1496                 xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(queue->vif),
1497                                         queue->tx.sring);
1498         if (queue->rx.sring)
1499                 xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(queue->vif),
1500                                         queue->rx.sring);
1501 }
1502
1503 int xenvif_map_frontend_data_rings(struct xenvif_queue *queue,
1504                                    grant_ref_t tx_ring_ref,
1505                                    grant_ref_t rx_ring_ref)
1506 {
1507         void *addr;
1508         struct xen_netif_tx_sring *txs;
1509         struct xen_netif_rx_sring *rxs;
1510         RING_IDX rsp_prod, req_prod;
1511         int err;
1512
1513         err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(queue->vif),
1514                                      &tx_ring_ref, 1, &addr);
1515         if (err)
1516                 goto err;
1517
1518         txs = (struct xen_netif_tx_sring *)addr;
1519         rsp_prod = READ_ONCE(txs->rsp_prod);
1520         req_prod = READ_ONCE(txs->req_prod);
1521
1522         BACK_RING_ATTACH(&queue->tx, txs, rsp_prod, XEN_PAGE_SIZE);
1523
1524         err = -EIO;
1525         if (req_prod - rsp_prod > RING_SIZE(&queue->tx))
1526                 goto err;
1527
1528         err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(queue->vif),
1529                                      &rx_ring_ref, 1, &addr);
1530         if (err)
1531                 goto err;
1532
1533         rxs = (struct xen_netif_rx_sring *)addr;
1534         rsp_prod = READ_ONCE(rxs->rsp_prod);
1535         req_prod = READ_ONCE(rxs->req_prod);
1536
1537         BACK_RING_ATTACH(&queue->rx, rxs, rsp_prod, XEN_PAGE_SIZE);
1538
1539         err = -EIO;
1540         if (req_prod - rsp_prod > RING_SIZE(&queue->rx))
1541                 goto err;
1542
1543         return 0;
1544
1545 err:
1546         xenvif_unmap_frontend_data_rings(queue);
1547         return err;
1548 }
1549
1550 static bool xenvif_dealloc_kthread_should_stop(struct xenvif_queue *queue)
1551 {
1552         /* Dealloc thread must remain running until all inflight
1553          * packets complete.
1554          */
1555         return kthread_should_stop() &&
1556                 !atomic_read(&queue->inflight_packets);
1557 }
1558
1559 int xenvif_dealloc_kthread(void *data)
1560 {
1561         struct xenvif_queue *queue = data;
1562
1563         for (;;) {
1564                 wait_event_interruptible(queue->dealloc_wq,
1565                                          tx_dealloc_work_todo(queue) ||
1566                                          xenvif_dealloc_kthread_should_stop(queue));
1567                 if (xenvif_dealloc_kthread_should_stop(queue))
1568                         break;
1569
1570                 xenvif_tx_dealloc_action(queue);
1571                 cond_resched();
1572         }
1573
1574         /* Unmap anything remaining*/
1575         if (tx_dealloc_work_todo(queue))
1576                 xenvif_tx_dealloc_action(queue);
1577
1578         return 0;
1579 }
1580
1581 static void make_ctrl_response(struct xenvif *vif,
1582                                const struct xen_netif_ctrl_request *req,
1583                                u32 status, u32 data)
1584 {
1585         RING_IDX idx = vif->ctrl.rsp_prod_pvt;
1586         struct xen_netif_ctrl_response rsp = {
1587                 .id = req->id,
1588                 .type = req->type,
1589                 .status = status,
1590                 .data = data,
1591         };
1592
1593         *RING_GET_RESPONSE(&vif->ctrl, idx) = rsp;
1594         vif->ctrl.rsp_prod_pvt = ++idx;
1595 }
1596
1597 static void push_ctrl_response(struct xenvif *vif)
1598 {
1599         int notify;
1600
1601         RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&vif->ctrl, notify);
1602         if (notify)
1603                 notify_remote_via_irq(vif->ctrl_irq);
1604 }
1605
1606 static void process_ctrl_request(struct xenvif *vif,
1607                                  const struct xen_netif_ctrl_request *req)
1608 {
1609         u32 status = XEN_NETIF_CTRL_STATUS_NOT_SUPPORTED;
1610         u32 data = 0;
1611
1612         switch (req->type) {
1613         case XEN_NETIF_CTRL_TYPE_SET_HASH_ALGORITHM:
1614                 status = xenvif_set_hash_alg(vif, req->data[0]);
1615                 break;
1616
1617         case XEN_NETIF_CTRL_TYPE_GET_HASH_FLAGS:
1618                 status = xenvif_get_hash_flags(vif, &data);
1619                 break;
1620
1621         case XEN_NETIF_CTRL_TYPE_SET_HASH_FLAGS:
1622                 status = xenvif_set_hash_flags(vif, req->data[0]);
1623                 break;
1624
1625         case XEN_NETIF_CTRL_TYPE_SET_HASH_KEY:
1626                 status = xenvif_set_hash_key(vif, req->data[0],
1627                                              req->data[1]);
1628                 break;
1629
1630         case XEN_NETIF_CTRL_TYPE_GET_HASH_MAPPING_SIZE:
1631                 status = XEN_NETIF_CTRL_STATUS_SUCCESS;
1632                 data = XEN_NETBK_MAX_HASH_MAPPING_SIZE;
1633                 break;
1634
1635         case XEN_NETIF_CTRL_TYPE_SET_HASH_MAPPING_SIZE:
1636                 status = xenvif_set_hash_mapping_size(vif,
1637                                                       req->data[0]);
1638                 break;
1639
1640         case XEN_NETIF_CTRL_TYPE_SET_HASH_MAPPING:
1641                 status = xenvif_set_hash_mapping(vif, req->data[0],
1642                                                  req->data[1],
1643                                                  req->data[2]);
1644                 break;
1645
1646         default:
1647                 break;
1648         }
1649
1650         make_ctrl_response(vif, req, status, data);
1651         push_ctrl_response(vif);
1652 }
1653
1654 static void xenvif_ctrl_action(struct xenvif *vif)
1655 {
1656         for (;;) {
1657                 RING_IDX req_prod, req_cons;
1658
1659                 req_prod = vif->ctrl.sring->req_prod;
1660                 req_cons = vif->ctrl.req_cons;
1661
1662                 /* Make sure we can see requests before we process them. */
1663                 rmb();
1664
1665                 if (req_cons == req_prod)
1666                         break;
1667
1668                 while (req_cons != req_prod) {
1669                         struct xen_netif_ctrl_request req;
1670
1671                         RING_COPY_REQUEST(&vif->ctrl, req_cons, &req);
1672                         req_cons++;
1673
1674                         process_ctrl_request(vif, &req);
1675                 }
1676
1677                 vif->ctrl.req_cons = req_cons;
1678                 vif->ctrl.sring->req_event = req_cons + 1;
1679         }
1680 }
1681
1682 static bool xenvif_ctrl_work_todo(struct xenvif *vif)
1683 {
1684         if (likely(RING_HAS_UNCONSUMED_REQUESTS(&vif->ctrl)))
1685                 return true;
1686
1687         return false;
1688 }
1689
1690 irqreturn_t xenvif_ctrl_irq_fn(int irq, void *data)
1691 {
1692         struct xenvif *vif = data;
1693         unsigned int eoi_flag = XEN_EOI_FLAG_SPURIOUS;
1694
1695         while (xenvif_ctrl_work_todo(vif)) {
1696                 xenvif_ctrl_action(vif);
1697                 eoi_flag = 0;
1698         }
1699
1700         xen_irq_lateeoi(irq, eoi_flag);
1701
1702         return IRQ_HANDLED;
1703 }
1704
1705 static int __init netback_init(void)
1706 {
1707         int rc = 0;
1708
1709         if (!xen_domain())
1710                 return -ENODEV;
1711
1712         /* Allow as many queues as there are CPUs but max. 8 if user has not
1713          * specified a value.
1714          */
1715         if (xenvif_max_queues == 0)
1716                 xenvif_max_queues = min_t(unsigned int, MAX_QUEUES_DEFAULT,
1717                                           num_online_cpus());
1718
1719         if (fatal_skb_slots < XEN_NETBK_LEGACY_SLOTS_MAX) {
1720                 pr_info("fatal_skb_slots too small (%d), bump it to XEN_NETBK_LEGACY_SLOTS_MAX (%d)\n",
1721                         fatal_skb_slots, XEN_NETBK_LEGACY_SLOTS_MAX);
1722                 fatal_skb_slots = XEN_NETBK_LEGACY_SLOTS_MAX;
1723         }
1724
1725         rc = xenvif_xenbus_init();
1726         if (rc)
1727                 goto failed_init;
1728
1729 #ifdef CONFIG_DEBUG_FS
1730         xen_netback_dbg_root = debugfs_create_dir("xen-netback", NULL);
1731 #endif /* CONFIG_DEBUG_FS */
1732
1733         return 0;
1734
1735 failed_init:
1736         return rc;
1737 }
1738
1739 module_init(netback_init);
1740
1741 static void __exit netback_fini(void)
1742 {
1743 #ifdef CONFIG_DEBUG_FS
1744         debugfs_remove_recursive(xen_netback_dbg_root);
1745 #endif /* CONFIG_DEBUG_FS */
1746         xenvif_xenbus_fini();
1747 }
1748 module_exit(netback_fini);
1749
1750 MODULE_LICENSE("Dual BSD/GPL");
1751 MODULE_ALIAS("xen-backend:vif");