diff options
Diffstat (limited to 'drivers/net/xen-netback/netback.c')
-rw-r--r-- | drivers/net/xen-netback/netback.c | 537 |
1 files changed, 108 insertions, 429 deletions
diff --git a/drivers/net/xen-netback/netback.c b/drivers/net/xen-netback/netback.c index 78425554a537..438d0c09b7e6 100644 --- a/drivers/net/xen-netback/netback.c +++ b/drivers/net/xen-netback/netback.c | |||
@@ -39,7 +39,6 @@ | |||
39 | #include <linux/udp.h> | 39 | #include <linux/udp.h> |
40 | 40 | ||
41 | #include <net/tcp.h> | 41 | #include <net/tcp.h> |
42 | #include <net/ip6_checksum.h> | ||
43 | 42 | ||
44 | #include <xen/xen.h> | 43 | #include <xen/xen.h> |
45 | #include <xen/events.h> | 44 | #include <xen/events.h> |
@@ -138,36 +137,26 @@ static inline pending_ring_idx_t nr_pending_reqs(struct xenvif *vif) | |||
138 | vif->pending_prod + vif->pending_cons; | 137 | vif->pending_prod + vif->pending_cons; |
139 | } | 138 | } |
140 | 139 | ||
141 | static int max_required_rx_slots(struct xenvif *vif) | 140 | bool xenvif_rx_ring_slots_available(struct xenvif *vif, int needed) |
142 | { | 141 | { |
143 | int max = DIV_ROUND_UP(vif->dev->mtu, PAGE_SIZE); | 142 | RING_IDX prod, cons; |
144 | 143 | ||
145 | /* XXX FIXME: RX path dependent on MAX_SKB_FRAGS */ | 144 | do { |
146 | if (vif->can_sg || vif->gso_mask || vif->gso_prefix_mask) | 145 | prod = vif->rx.sring->req_prod; |
147 | max += MAX_SKB_FRAGS + 1; /* extra_info + frags */ | 146 | cons = vif->rx.req_cons; |
148 | |||
149 | return max; | ||
150 | } | ||
151 | 147 | ||
152 | int xenvif_rx_ring_full(struct xenvif *vif) | 148 | if (prod - cons >= needed) |
153 | { | 149 | return true; |
154 | RING_IDX peek = vif->rx_req_cons_peek; | ||
155 | RING_IDX needed = max_required_rx_slots(vif); | ||
156 | 150 | ||
157 | return ((vif->rx.sring->req_prod - peek) < needed) || | 151 | vif->rx.sring->req_event = prod + 1; |
158 | ((vif->rx.rsp_prod_pvt + XEN_NETIF_RX_RING_SIZE - peek) < needed); | ||
159 | } | ||
160 | 152 | ||
161 | int xenvif_must_stop_queue(struct xenvif *vif) | 153 | /* Make sure event is visible before we check prod |
162 | { | 154 | * again. |
163 | if (!xenvif_rx_ring_full(vif)) | 155 | */ |
164 | return 0; | 156 | mb(); |
165 | 157 | } while (vif->rx.sring->req_prod != prod); | |
166 | vif->rx.sring->req_event = vif->rx_req_cons_peek + | ||
167 | max_required_rx_slots(vif); | ||
168 | mb(); /* request notification /then/ check the queue */ | ||
169 | 158 | ||
170 | return xenvif_rx_ring_full(vif); | 159 | return false; |
171 | } | 160 | } |
172 | 161 | ||
173 | /* | 162 | /* |
@@ -210,93 +199,6 @@ static bool start_new_rx_buffer(int offset, unsigned long size, int head) | |||
210 | return false; | 199 | return false; |
211 | } | 200 | } |
212 | 201 | ||
213 | struct xenvif_count_slot_state { | ||
214 | unsigned long copy_off; | ||
215 | bool head; | ||
216 | }; | ||
217 | |||
218 | unsigned int xenvif_count_frag_slots(struct xenvif *vif, | ||
219 | unsigned long offset, unsigned long size, | ||
220 | struct xenvif_count_slot_state *state) | ||
221 | { | ||
222 | unsigned count = 0; | ||
223 | |||
224 | offset &= ~PAGE_MASK; | ||
225 | |||
226 | while (size > 0) { | ||
227 | unsigned long bytes; | ||
228 | |||
229 | bytes = PAGE_SIZE - offset; | ||
230 | |||
231 | if (bytes > size) | ||
232 | bytes = size; | ||
233 | |||
234 | if (start_new_rx_buffer(state->copy_off, bytes, state->head)) { | ||
235 | count++; | ||
236 | state->copy_off = 0; | ||
237 | } | ||
238 | |||
239 | if (state->copy_off + bytes > MAX_BUFFER_OFFSET) | ||
240 | bytes = MAX_BUFFER_OFFSET - state->copy_off; | ||
241 | |||
242 | state->copy_off += bytes; | ||
243 | |||
244 | offset += bytes; | ||
245 | size -= bytes; | ||
246 | |||
247 | if (offset == PAGE_SIZE) | ||
248 | offset = 0; | ||
249 | |||
250 | state->head = false; | ||
251 | } | ||
252 | |||
253 | return count; | ||
254 | } | ||
255 | |||
256 | /* | ||
257 | * Figure out how many ring slots we're going to need to send @skb to | ||
258 | * the guest. This function is essentially a dry run of | ||
259 | * xenvif_gop_frag_copy. | ||
260 | */ | ||
261 | unsigned int xenvif_count_skb_slots(struct xenvif *vif, struct sk_buff *skb) | ||
262 | { | ||
263 | struct xenvif_count_slot_state state; | ||
264 | unsigned int count; | ||
265 | unsigned char *data; | ||
266 | unsigned i; | ||
267 | |||
268 | state.head = true; | ||
269 | state.copy_off = 0; | ||
270 | |||
271 | /* Slot for the first (partial) page of data. */ | ||
272 | count = 1; | ||
273 | |||
274 | /* Need a slot for the GSO prefix for GSO extra data? */ | ||
275 | if (skb_shinfo(skb)->gso_size) | ||
276 | count++; | ||
277 | |||
278 | data = skb->data; | ||
279 | while (data < skb_tail_pointer(skb)) { | ||
280 | unsigned long offset = offset_in_page(data); | ||
281 | unsigned long size = PAGE_SIZE - offset; | ||
282 | |||
283 | if (data + size > skb_tail_pointer(skb)) | ||
284 | size = skb_tail_pointer(skb) - data; | ||
285 | |||
286 | count += xenvif_count_frag_slots(vif, offset, size, &state); | ||
287 | |||
288 | data += size; | ||
289 | } | ||
290 | |||
291 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { | ||
292 | unsigned long size = skb_frag_size(&skb_shinfo(skb)->frags[i]); | ||
293 | unsigned long offset = skb_shinfo(skb)->frags[i].page_offset; | ||
294 | |||
295 | count += xenvif_count_frag_slots(vif, offset, size, &state); | ||
296 | } | ||
297 | return count; | ||
298 | } | ||
299 | |||
300 | struct netrx_pending_operations { | 202 | struct netrx_pending_operations { |
301 | unsigned copy_prod, copy_cons; | 203 | unsigned copy_prod, copy_cons; |
302 | unsigned meta_prod, meta_cons; | 204 | unsigned meta_prod, meta_cons; |
@@ -338,7 +240,7 @@ static void xenvif_gop_frag_copy(struct xenvif *vif, struct sk_buff *skb, | |||
338 | struct gnttab_copy *copy_gop; | 240 | struct gnttab_copy *copy_gop; |
339 | struct xenvif_rx_meta *meta; | 241 | struct xenvif_rx_meta *meta; |
340 | unsigned long bytes; | 242 | unsigned long bytes; |
341 | int gso_type; | 243 | int gso_type = XEN_NETIF_GSO_TYPE_NONE; |
342 | 244 | ||
343 | /* Data must not cross a page boundary. */ | 245 | /* Data must not cross a page boundary. */ |
344 | BUG_ON(size + offset > PAGE_SIZE<<compound_order(page)); | 246 | BUG_ON(size + offset > PAGE_SIZE<<compound_order(page)); |
@@ -397,12 +299,12 @@ static void xenvif_gop_frag_copy(struct xenvif *vif, struct sk_buff *skb, | |||
397 | } | 299 | } |
398 | 300 | ||
399 | /* Leave a gap for the GSO descriptor. */ | 301 | /* Leave a gap for the GSO descriptor. */ |
400 | if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) | 302 | if (skb_is_gso(skb)) { |
401 | gso_type = XEN_NETIF_GSO_TYPE_TCPV4; | 303 | if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) |
402 | else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) | 304 | gso_type = XEN_NETIF_GSO_TYPE_TCPV4; |
403 | gso_type = XEN_NETIF_GSO_TYPE_TCPV6; | 305 | else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) |
404 | else | 306 | gso_type = XEN_NETIF_GSO_TYPE_TCPV6; |
405 | gso_type = XEN_NETIF_GSO_TYPE_NONE; | 307 | } |
406 | 308 | ||
407 | if (*head && ((1 << gso_type) & vif->gso_mask)) | 309 | if (*head && ((1 << gso_type) & vif->gso_mask)) |
408 | vif->rx.req_cons++; | 310 | vif->rx.req_cons++; |
@@ -436,19 +338,15 @@ static int xenvif_gop_skb(struct sk_buff *skb, | |||
436 | int head = 1; | 338 | int head = 1; |
437 | int old_meta_prod; | 339 | int old_meta_prod; |
438 | int gso_type; | 340 | int gso_type; |
439 | int gso_size; | ||
440 | 341 | ||
441 | old_meta_prod = npo->meta_prod; | 342 | old_meta_prod = npo->meta_prod; |
442 | 343 | ||
443 | if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) { | 344 | gso_type = XEN_NETIF_GSO_TYPE_NONE; |
444 | gso_type = XEN_NETIF_GSO_TYPE_TCPV4; | 345 | if (skb_is_gso(skb)) { |
445 | gso_size = skb_shinfo(skb)->gso_size; | 346 | if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) |
446 | } else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) { | 347 | gso_type = XEN_NETIF_GSO_TYPE_TCPV4; |
447 | gso_type = XEN_NETIF_GSO_TYPE_TCPV6; | 348 | else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) |
448 | gso_size = skb_shinfo(skb)->gso_size; | 349 | gso_type = XEN_NETIF_GSO_TYPE_TCPV6; |
449 | } else { | ||
450 | gso_type = XEN_NETIF_GSO_TYPE_NONE; | ||
451 | gso_size = 0; | ||
452 | } | 350 | } |
453 | 351 | ||
454 | /* Set up a GSO prefix descriptor, if necessary */ | 352 | /* Set up a GSO prefix descriptor, if necessary */ |
@@ -456,7 +354,7 @@ static int xenvif_gop_skb(struct sk_buff *skb, | |||
456 | req = RING_GET_REQUEST(&vif->rx, vif->rx.req_cons++); | 354 | req = RING_GET_REQUEST(&vif->rx, vif->rx.req_cons++); |
457 | meta = npo->meta + npo->meta_prod++; | 355 | meta = npo->meta + npo->meta_prod++; |
458 | meta->gso_type = gso_type; | 356 | meta->gso_type = gso_type; |
459 | meta->gso_size = gso_size; | 357 | meta->gso_size = skb_shinfo(skb)->gso_size; |
460 | meta->size = 0; | 358 | meta->size = 0; |
461 | meta->id = req->id; | 359 | meta->id = req->id; |
462 | } | 360 | } |
@@ -466,7 +364,7 @@ static int xenvif_gop_skb(struct sk_buff *skb, | |||
466 | 364 | ||
467 | if ((1 << gso_type) & vif->gso_mask) { | 365 | if ((1 << gso_type) & vif->gso_mask) { |
468 | meta->gso_type = gso_type; | 366 | meta->gso_type = gso_type; |
469 | meta->gso_size = gso_size; | 367 | meta->gso_size = skb_shinfo(skb)->gso_size; |
470 | } else { | 368 | } else { |
471 | meta->gso_type = XEN_NETIF_GSO_TYPE_NONE; | 369 | meta->gso_type = XEN_NETIF_GSO_TYPE_NONE; |
472 | meta->gso_size = 0; | 370 | meta->gso_size = 0; |
@@ -557,12 +455,12 @@ struct skb_cb_overlay { | |||
557 | int meta_slots_used; | 455 | int meta_slots_used; |
558 | }; | 456 | }; |
559 | 457 | ||
560 | static void xenvif_kick_thread(struct xenvif *vif) | 458 | void xenvif_kick_thread(struct xenvif *vif) |
561 | { | 459 | { |
562 | wake_up(&vif->wq); | 460 | wake_up(&vif->wq); |
563 | } | 461 | } |
564 | 462 | ||
565 | void xenvif_rx_action(struct xenvif *vif) | 463 | static void xenvif_rx_action(struct xenvif *vif) |
566 | { | 464 | { |
567 | s8 status; | 465 | s8 status; |
568 | u16 flags; | 466 | u16 flags; |
@@ -571,11 +469,9 @@ void xenvif_rx_action(struct xenvif *vif) | |||
571 | struct sk_buff *skb; | 469 | struct sk_buff *skb; |
572 | LIST_HEAD(notify); | 470 | LIST_HEAD(notify); |
573 | int ret; | 471 | int ret; |
574 | int nr_frags; | ||
575 | int count; | ||
576 | unsigned long offset; | 472 | unsigned long offset; |
577 | struct skb_cb_overlay *sco; | 473 | struct skb_cb_overlay *sco; |
578 | int need_to_notify = 0; | 474 | bool need_to_notify = false; |
579 | 475 | ||
580 | struct netrx_pending_operations npo = { | 476 | struct netrx_pending_operations npo = { |
581 | .copy = vif->grant_copy_op, | 477 | .copy = vif->grant_copy_op, |
@@ -584,29 +480,47 @@ void xenvif_rx_action(struct xenvif *vif) | |||
584 | 480 | ||
585 | skb_queue_head_init(&rxq); | 481 | skb_queue_head_init(&rxq); |
586 | 482 | ||
587 | count = 0; | ||
588 | |||
589 | while ((skb = skb_dequeue(&vif->rx_queue)) != NULL) { | 483 | while ((skb = skb_dequeue(&vif->rx_queue)) != NULL) { |
590 | vif = netdev_priv(skb->dev); | 484 | RING_IDX max_slots_needed; |
591 | nr_frags = skb_shinfo(skb)->nr_frags; | 485 | int i; |
486 | |||
487 | /* We need a cheap worse case estimate for the number of | ||
488 | * slots we'll use. | ||
489 | */ | ||
490 | |||
491 | max_slots_needed = DIV_ROUND_UP(offset_in_page(skb->data) + | ||
492 | skb_headlen(skb), | ||
493 | PAGE_SIZE); | ||
494 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { | ||
495 | unsigned int size; | ||
496 | size = skb_frag_size(&skb_shinfo(skb)->frags[i]); | ||
497 | max_slots_needed += DIV_ROUND_UP(size, PAGE_SIZE); | ||
498 | } | ||
499 | if (skb_is_gso(skb) && | ||
500 | (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4 || | ||
501 | skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)) | ||
502 | max_slots_needed++; | ||
503 | |||
504 | /* If the skb may not fit then bail out now */ | ||
505 | if (!xenvif_rx_ring_slots_available(vif, max_slots_needed)) { | ||
506 | skb_queue_head(&vif->rx_queue, skb); | ||
507 | need_to_notify = true; | ||
508 | vif->rx_last_skb_slots = max_slots_needed; | ||
509 | break; | ||
510 | } else | ||
511 | vif->rx_last_skb_slots = 0; | ||
592 | 512 | ||
593 | sco = (struct skb_cb_overlay *)skb->cb; | 513 | sco = (struct skb_cb_overlay *)skb->cb; |
594 | sco->meta_slots_used = xenvif_gop_skb(skb, &npo); | 514 | sco->meta_slots_used = xenvif_gop_skb(skb, &npo); |
595 | 515 | BUG_ON(sco->meta_slots_used > max_slots_needed); | |
596 | count += nr_frags + 1; | ||
597 | 516 | ||
598 | __skb_queue_tail(&rxq, skb); | 517 | __skb_queue_tail(&rxq, skb); |
599 | |||
600 | /* Filled the batch queue? */ | ||
601 | /* XXX FIXME: RX path dependent on MAX_SKB_FRAGS */ | ||
602 | if (count + MAX_SKB_FRAGS >= XEN_NETIF_RX_RING_SIZE) | ||
603 | break; | ||
604 | } | 518 | } |
605 | 519 | ||
606 | BUG_ON(npo.meta_prod > ARRAY_SIZE(vif->meta)); | 520 | BUG_ON(npo.meta_prod > ARRAY_SIZE(vif->meta)); |
607 | 521 | ||
608 | if (!npo.copy_prod) | 522 | if (!npo.copy_prod) |
609 | return; | 523 | goto done; |
610 | 524 | ||
611 | BUG_ON(npo.copy_prod > MAX_GRANT_COPY_OPS); | 525 | BUG_ON(npo.copy_prod > MAX_GRANT_COPY_OPS); |
612 | gnttab_batch_copy(vif->grant_copy_op, npo.copy_prod); | 526 | gnttab_batch_copy(vif->grant_copy_op, npo.copy_prod); |
@@ -614,8 +528,6 @@ void xenvif_rx_action(struct xenvif *vif) | |||
614 | while ((skb = __skb_dequeue(&rxq)) != NULL) { | 528 | while ((skb = __skb_dequeue(&rxq)) != NULL) { |
615 | sco = (struct skb_cb_overlay *)skb->cb; | 529 | sco = (struct skb_cb_overlay *)skb->cb; |
616 | 530 | ||
617 | vif = netdev_priv(skb->dev); | ||
618 | |||
619 | if ((1 << vif->meta[npo.meta_cons].gso_type) & | 531 | if ((1 << vif->meta[npo.meta_cons].gso_type) & |
620 | vif->gso_prefix_mask) { | 532 | vif->gso_prefix_mask) { |
621 | resp = RING_GET_RESPONSE(&vif->rx, | 533 | resp = RING_GET_RESPONSE(&vif->rx, |
@@ -678,28 +590,15 @@ void xenvif_rx_action(struct xenvif *vif) | |||
678 | 590 | ||
679 | RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&vif->rx, ret); | 591 | RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&vif->rx, ret); |
680 | 592 | ||
681 | if (ret) | 593 | need_to_notify |= !!ret; |
682 | need_to_notify = 1; | ||
683 | |||
684 | xenvif_notify_tx_completion(vif); | ||
685 | 594 | ||
686 | npo.meta_cons += sco->meta_slots_used; | 595 | npo.meta_cons += sco->meta_slots_used; |
687 | dev_kfree_skb(skb); | 596 | dev_kfree_skb(skb); |
688 | } | 597 | } |
689 | 598 | ||
599 | done: | ||
690 | if (need_to_notify) | 600 | if (need_to_notify) |
691 | notify_remote_via_irq(vif->rx_irq); | 601 | notify_remote_via_irq(vif->rx_irq); |
692 | |||
693 | /* More work to do? */ | ||
694 | if (!skb_queue_empty(&vif->rx_queue)) | ||
695 | xenvif_kick_thread(vif); | ||
696 | } | ||
697 | |||
698 | void xenvif_queue_tx_skb(struct xenvif *vif, struct sk_buff *skb) | ||
699 | { | ||
700 | skb_queue_tail(&vif->rx_queue, skb); | ||
701 | |||
702 | xenvif_kick_thread(vif); | ||
703 | } | 602 | } |
704 | 603 | ||
705 | void xenvif_check_rx_xenvif(struct xenvif *vif) | 604 | void xenvif_check_rx_xenvif(struct xenvif *vif) |
@@ -1141,265 +1040,14 @@ static int xenvif_set_skb_gso(struct xenvif *vif, | |||
1141 | } | 1040 | } |
1142 | 1041 | ||
1143 | skb_shinfo(skb)->gso_size = gso->u.gso.size; | 1042 | skb_shinfo(skb)->gso_size = gso->u.gso.size; |
1144 | 1043 | /* gso_segs will be calculated later */ | |
1145 | /* Header must be checked, and gso_segs computed. */ | ||
1146 | skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; | ||
1147 | skb_shinfo(skb)->gso_segs = 0; | ||
1148 | 1044 | ||
1149 | return 0; | 1045 | return 0; |
1150 | } | 1046 | } |
1151 | 1047 | ||
1152 | static inline int maybe_pull_tail(struct sk_buff *skb, unsigned int len, | ||
1153 | unsigned int max) | ||
1154 | { | ||
1155 | if (skb_headlen(skb) >= len) | ||
1156 | return 0; | ||
1157 | |||
1158 | /* If we need to pullup then pullup to the max, so we | ||
1159 | * won't need to do it again. | ||
1160 | */ | ||
1161 | if (max > skb->len) | ||
1162 | max = skb->len; | ||
1163 | |||
1164 | if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL) | ||
1165 | return -ENOMEM; | ||
1166 | |||
1167 | if (skb_headlen(skb) < len) | ||
1168 | return -EPROTO; | ||
1169 | |||
1170 | return 0; | ||
1171 | } | ||
1172 | |||
1173 | /* This value should be large enough to cover a tagged ethernet header plus | ||
1174 | * maximally sized IP and TCP or UDP headers. | ||
1175 | */ | ||
1176 | #define MAX_IP_HDR_LEN 128 | ||
1177 | |||
1178 | static int checksum_setup_ip(struct xenvif *vif, struct sk_buff *skb, | ||
1179 | int recalculate_partial_csum) | ||
1180 | { | ||
1181 | unsigned int off; | ||
1182 | bool fragment; | ||
1183 | int err; | ||
1184 | |||
1185 | fragment = false; | ||
1186 | |||
1187 | err = maybe_pull_tail(skb, | ||
1188 | sizeof(struct iphdr), | ||
1189 | MAX_IP_HDR_LEN); | ||
1190 | if (err < 0) | ||
1191 | goto out; | ||
1192 | |||
1193 | if (ip_hdr(skb)->frag_off & htons(IP_OFFSET | IP_MF)) | ||
1194 | fragment = true; | ||
1195 | |||
1196 | off = ip_hdrlen(skb); | ||
1197 | |||
1198 | err = -EPROTO; | ||
1199 | |||
1200 | if (fragment) | ||
1201 | goto out; | ||
1202 | |||
1203 | switch (ip_hdr(skb)->protocol) { | ||
1204 | case IPPROTO_TCP: | ||
1205 | err = maybe_pull_tail(skb, | ||
1206 | off + sizeof(struct tcphdr), | ||
1207 | MAX_IP_HDR_LEN); | ||
1208 | if (err < 0) | ||
1209 | goto out; | ||
1210 | |||
1211 | if (!skb_partial_csum_set(skb, off, | ||
1212 | offsetof(struct tcphdr, check))) { | ||
1213 | err = -EPROTO; | ||
1214 | goto out; | ||
1215 | } | ||
1216 | |||
1217 | if (recalculate_partial_csum) | ||
1218 | tcp_hdr(skb)->check = | ||
1219 | ~csum_tcpudp_magic(ip_hdr(skb)->saddr, | ||
1220 | ip_hdr(skb)->daddr, | ||
1221 | skb->len - off, | ||
1222 | IPPROTO_TCP, 0); | ||
1223 | break; | ||
1224 | case IPPROTO_UDP: | ||
1225 | err = maybe_pull_tail(skb, | ||
1226 | off + sizeof(struct udphdr), | ||
1227 | MAX_IP_HDR_LEN); | ||
1228 | if (err < 0) | ||
1229 | goto out; | ||
1230 | |||
1231 | if (!skb_partial_csum_set(skb, off, | ||
1232 | offsetof(struct udphdr, check))) { | ||
1233 | err = -EPROTO; | ||
1234 | goto out; | ||
1235 | } | ||
1236 | |||
1237 | if (recalculate_partial_csum) | ||
1238 | udp_hdr(skb)->check = | ||
1239 | ~csum_tcpudp_magic(ip_hdr(skb)->saddr, | ||
1240 | ip_hdr(skb)->daddr, | ||
1241 | skb->len - off, | ||
1242 | IPPROTO_UDP, 0); | ||
1243 | break; | ||
1244 | default: | ||
1245 | goto out; | ||
1246 | } | ||
1247 | |||
1248 | err = 0; | ||
1249 | |||
1250 | out: | ||
1251 | return err; | ||
1252 | } | ||
1253 | |||
1254 | /* This value should be large enough to cover a tagged ethernet header plus | ||
1255 | * an IPv6 header, all options, and a maximal TCP or UDP header. | ||
1256 | */ | ||
1257 | #define MAX_IPV6_HDR_LEN 256 | ||
1258 | |||
1259 | #define OPT_HDR(type, skb, off) \ | ||
1260 | (type *)(skb_network_header(skb) + (off)) | ||
1261 | |||
1262 | static int checksum_setup_ipv6(struct xenvif *vif, struct sk_buff *skb, | ||
1263 | int recalculate_partial_csum) | ||
1264 | { | ||
1265 | int err; | ||
1266 | u8 nexthdr; | ||
1267 | unsigned int off; | ||
1268 | unsigned int len; | ||
1269 | bool fragment; | ||
1270 | bool done; | ||
1271 | |||
1272 | fragment = false; | ||
1273 | done = false; | ||
1274 | |||
1275 | off = sizeof(struct ipv6hdr); | ||
1276 | |||
1277 | err = maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN); | ||
1278 | if (err < 0) | ||
1279 | goto out; | ||
1280 | |||
1281 | nexthdr = ipv6_hdr(skb)->nexthdr; | ||
1282 | |||
1283 | len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len); | ||
1284 | while (off <= len && !done) { | ||
1285 | switch (nexthdr) { | ||
1286 | case IPPROTO_DSTOPTS: | ||
1287 | case IPPROTO_HOPOPTS: | ||
1288 | case IPPROTO_ROUTING: { | ||
1289 | struct ipv6_opt_hdr *hp; | ||
1290 | |||
1291 | err = maybe_pull_tail(skb, | ||
1292 | off + | ||
1293 | sizeof(struct ipv6_opt_hdr), | ||
1294 | MAX_IPV6_HDR_LEN); | ||
1295 | if (err < 0) | ||
1296 | goto out; | ||
1297 | |||
1298 | hp = OPT_HDR(struct ipv6_opt_hdr, skb, off); | ||
1299 | nexthdr = hp->nexthdr; | ||
1300 | off += ipv6_optlen(hp); | ||
1301 | break; | ||
1302 | } | ||
1303 | case IPPROTO_AH: { | ||
1304 | struct ip_auth_hdr *hp; | ||
1305 | |||
1306 | err = maybe_pull_tail(skb, | ||
1307 | off + | ||
1308 | sizeof(struct ip_auth_hdr), | ||
1309 | MAX_IPV6_HDR_LEN); | ||
1310 | if (err < 0) | ||
1311 | goto out; | ||
1312 | |||
1313 | hp = OPT_HDR(struct ip_auth_hdr, skb, off); | ||
1314 | nexthdr = hp->nexthdr; | ||
1315 | off += ipv6_authlen(hp); | ||
1316 | break; | ||
1317 | } | ||
1318 | case IPPROTO_FRAGMENT: { | ||
1319 | struct frag_hdr *hp; | ||
1320 | |||
1321 | err = maybe_pull_tail(skb, | ||
1322 | off + | ||
1323 | sizeof(struct frag_hdr), | ||
1324 | MAX_IPV6_HDR_LEN); | ||
1325 | if (err < 0) | ||
1326 | goto out; | ||
1327 | |||
1328 | hp = OPT_HDR(struct frag_hdr, skb, off); | ||
1329 | |||
1330 | if (hp->frag_off & htons(IP6_OFFSET | IP6_MF)) | ||
1331 | fragment = true; | ||
1332 | |||
1333 | nexthdr = hp->nexthdr; | ||
1334 | off += sizeof(struct frag_hdr); | ||
1335 | break; | ||
1336 | } | ||
1337 | default: | ||
1338 | done = true; | ||
1339 | break; | ||
1340 | } | ||
1341 | } | ||
1342 | |||
1343 | err = -EPROTO; | ||
1344 | |||
1345 | if (!done || fragment) | ||
1346 | goto out; | ||
1347 | |||
1348 | switch (nexthdr) { | ||
1349 | case IPPROTO_TCP: | ||
1350 | err = maybe_pull_tail(skb, | ||
1351 | off + sizeof(struct tcphdr), | ||
1352 | MAX_IPV6_HDR_LEN); | ||
1353 | if (err < 0) | ||
1354 | goto out; | ||
1355 | |||
1356 | if (!skb_partial_csum_set(skb, off, | ||
1357 | offsetof(struct tcphdr, check))) { | ||
1358 | err = -EPROTO; | ||
1359 | goto out; | ||
1360 | } | ||
1361 | |||
1362 | if (recalculate_partial_csum) | ||
1363 | tcp_hdr(skb)->check = | ||
1364 | ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr, | ||
1365 | &ipv6_hdr(skb)->daddr, | ||
1366 | skb->len - off, | ||
1367 | IPPROTO_TCP, 0); | ||
1368 | break; | ||
1369 | case IPPROTO_UDP: | ||
1370 | err = maybe_pull_tail(skb, | ||
1371 | off + sizeof(struct udphdr), | ||
1372 | MAX_IPV6_HDR_LEN); | ||
1373 | if (err < 0) | ||
1374 | goto out; | ||
1375 | |||
1376 | if (!skb_partial_csum_set(skb, off, | ||
1377 | offsetof(struct udphdr, check))) { | ||
1378 | err = -EPROTO; | ||
1379 | goto out; | ||
1380 | } | ||
1381 | |||
1382 | if (recalculate_partial_csum) | ||
1383 | udp_hdr(skb)->check = | ||
1384 | ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr, | ||
1385 | &ipv6_hdr(skb)->daddr, | ||
1386 | skb->len - off, | ||
1387 | IPPROTO_UDP, 0); | ||
1388 | break; | ||
1389 | default: | ||
1390 | goto out; | ||
1391 | } | ||
1392 | |||
1393 | err = 0; | ||
1394 | |||
1395 | out: | ||
1396 | return err; | ||
1397 | } | ||
1398 | |||
1399 | static int checksum_setup(struct xenvif *vif, struct sk_buff *skb) | 1048 | static int checksum_setup(struct xenvif *vif, struct sk_buff *skb) |
1400 | { | 1049 | { |
1401 | int err = -EPROTO; | 1050 | bool recalculate_partial_csum = false; |
1402 | int recalculate_partial_csum = 0; | ||
1403 | 1051 | ||
1404 | /* A GSO SKB must be CHECKSUM_PARTIAL. However some buggy | 1052 | /* A GSO SKB must be CHECKSUM_PARTIAL. However some buggy |
1405 | * peers can fail to set NETRXF_csum_blank when sending a GSO | 1053 | * peers can fail to set NETRXF_csum_blank when sending a GSO |
@@ -1409,19 +1057,14 @@ static int checksum_setup(struct xenvif *vif, struct sk_buff *skb) | |||
1409 | if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) { | 1057 | if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) { |
1410 | vif->rx_gso_checksum_fixup++; | 1058 | vif->rx_gso_checksum_fixup++; |
1411 | skb->ip_summed = CHECKSUM_PARTIAL; | 1059 | skb->ip_summed = CHECKSUM_PARTIAL; |
1412 | recalculate_partial_csum = 1; | 1060 | recalculate_partial_csum = true; |
1413 | } | 1061 | } |
1414 | 1062 | ||
1415 | /* A non-CHECKSUM_PARTIAL SKB does not require setup. */ | 1063 | /* A non-CHECKSUM_PARTIAL SKB does not require setup. */ |
1416 | if (skb->ip_summed != CHECKSUM_PARTIAL) | 1064 | if (skb->ip_summed != CHECKSUM_PARTIAL) |
1417 | return 0; | 1065 | return 0; |
1418 | 1066 | ||
1419 | if (skb->protocol == htons(ETH_P_IP)) | 1067 | return skb_checksum_setup(skb, recalculate_partial_csum); |
1420 | err = checksum_setup_ip(vif, skb, recalculate_partial_csum); | ||
1421 | else if (skb->protocol == htons(ETH_P_IPV6)) | ||
1422 | err = checksum_setup_ipv6(vif, skb, recalculate_partial_csum); | ||
1423 | |||
1424 | return err; | ||
1425 | } | 1068 | } |
1426 | 1069 | ||
1427 | static bool tx_credit_exceeded(struct xenvif *vif, unsigned size) | 1070 | static bool tx_credit_exceeded(struct xenvif *vif, unsigned size) |
@@ -1687,6 +1330,20 @@ static int xenvif_tx_submit(struct xenvif *vif) | |||
1687 | 1330 | ||
1688 | skb_probe_transport_header(skb, 0); | 1331 | skb_probe_transport_header(skb, 0); |
1689 | 1332 | ||
1333 | /* If the packet is GSO then we will have just set up the | ||
1334 | * transport header offset in checksum_setup so it's now | ||
1335 | * straightforward to calculate gso_segs. | ||
1336 | */ | ||
1337 | if (skb_is_gso(skb)) { | ||
1338 | int mss = skb_shinfo(skb)->gso_size; | ||
1339 | int hdrlen = skb_transport_header(skb) - | ||
1340 | skb_mac_header(skb) + | ||
1341 | tcp_hdrlen(skb); | ||
1342 | |||
1343 | skb_shinfo(skb)->gso_segs = | ||
1344 | DIV_ROUND_UP(skb->len - hdrlen, mss); | ||
1345 | } | ||
1346 | |||
1690 | vif->dev->stats.rx_bytes += skb->len; | 1347 | vif->dev->stats.rx_bytes += skb->len; |
1691 | vif->dev->stats.rx_packets++; | 1348 | vif->dev->stats.rx_packets++; |
1692 | 1349 | ||
@@ -1811,7 +1468,8 @@ static struct xen_netif_rx_response *make_rx_response(struct xenvif *vif, | |||
1811 | 1468 | ||
1812 | static inline int rx_work_todo(struct xenvif *vif) | 1469 | static inline int rx_work_todo(struct xenvif *vif) |
1813 | { | 1470 | { |
1814 | return !skb_queue_empty(&vif->rx_queue); | 1471 | return !skb_queue_empty(&vif->rx_queue) && |
1472 | xenvif_rx_ring_slots_available(vif, vif->rx_last_skb_slots); | ||
1815 | } | 1473 | } |
1816 | 1474 | ||
1817 | static inline int tx_work_todo(struct xenvif *vif) | 1475 | static inline int tx_work_todo(struct xenvif *vif) |
@@ -1861,8 +1519,6 @@ int xenvif_map_frontend_rings(struct xenvif *vif, | |||
1861 | rxs = (struct xen_netif_rx_sring *)addr; | 1519 | rxs = (struct xen_netif_rx_sring *)addr; |
1862 | BACK_RING_INIT(&vif->rx, rxs, PAGE_SIZE); | 1520 | BACK_RING_INIT(&vif->rx, rxs, PAGE_SIZE); |
1863 | 1521 | ||
1864 | vif->rx_req_cons_peek = 0; | ||
1865 | |||
1866 | return 0; | 1522 | return 0; |
1867 | 1523 | ||
1868 | err: | 1524 | err: |
@@ -1870,9 +1526,24 @@ err: | |||
1870 | return err; | 1526 | return err; |
1871 | } | 1527 | } |
1872 | 1528 | ||
1529 | void xenvif_stop_queue(struct xenvif *vif) | ||
1530 | { | ||
1531 | if (!vif->can_queue) | ||
1532 | return; | ||
1533 | |||
1534 | netif_stop_queue(vif->dev); | ||
1535 | } | ||
1536 | |||
1537 | static void xenvif_start_queue(struct xenvif *vif) | ||
1538 | { | ||
1539 | if (xenvif_schedulable(vif)) | ||
1540 | netif_wake_queue(vif->dev); | ||
1541 | } | ||
1542 | |||
1873 | int xenvif_kthread(void *data) | 1543 | int xenvif_kthread(void *data) |
1874 | { | 1544 | { |
1875 | struct xenvif *vif = data; | 1545 | struct xenvif *vif = data; |
1546 | struct sk_buff *skb; | ||
1876 | 1547 | ||
1877 | while (!kthread_should_stop()) { | 1548 | while (!kthread_should_stop()) { |
1878 | wait_event_interruptible(vif->wq, | 1549 | wait_event_interruptible(vif->wq, |
@@ -1881,12 +1552,20 @@ int xenvif_kthread(void *data) | |||
1881 | if (kthread_should_stop()) | 1552 | if (kthread_should_stop()) |
1882 | break; | 1553 | break; |
1883 | 1554 | ||
1884 | if (rx_work_todo(vif)) | 1555 | if (!skb_queue_empty(&vif->rx_queue)) |
1885 | xenvif_rx_action(vif); | 1556 | xenvif_rx_action(vif); |
1886 | 1557 | ||
1558 | if (skb_queue_empty(&vif->rx_queue) && | ||
1559 | netif_queue_stopped(vif->dev)) | ||
1560 | xenvif_start_queue(vif); | ||
1561 | |||
1887 | cond_resched(); | 1562 | cond_resched(); |
1888 | } | 1563 | } |
1889 | 1564 | ||
1565 | /* Bin any remaining skbs */ | ||
1566 | while ((skb = skb_dequeue(&vif->rx_queue)) != NULL) | ||
1567 | dev_kfree_skb(skb); | ||
1568 | |||
1890 | return 0; | 1569 | return 0; |
1891 | } | 1570 | } |
1892 | 1571 | ||