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