projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
net: ethtool: Add destination MAC address to flow steering API
[deliverable/linux.git] / 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
41 #include <net/tcp.h>
42
43 #include <xen/xen.h>
44 #include <xen/events.h>
45 #include <xen/interface/memory.h>
46
47 #include <asm/xen/hypercall.h>
48 #include <asm/xen/page.h>
49
50 struct pending_tx_info {
51 struct xen_netif_tx_request req;
52 struct xenvif *vif;
53 };
54 typedef unsigned int pending_ring_idx_t;
55
56 struct netbk_rx_meta {
57 int id;
58 int size;
59 int gso_size;
60 };
61
62 #define MAX_PENDING_REQS 256
63
64 /* Discriminate from any valid pending_idx value. */
65 #define INVALID_PENDING_IDX 0xFFFF
66
67 #define MAX_BUFFER_OFFSET PAGE_SIZE
68
69 /* extra field used in struct page */
70 union page_ext {
71 struct {
72 #if BITS_PER_LONG < 64
73 #define IDX_WIDTH 8
74 #define GROUP_WIDTH (BITS_PER_LONG - IDX_WIDTH)
75 unsigned int group:GROUP_WIDTH;
76 unsigned int idx:IDX_WIDTH;
77 #else
78 unsigned int group, idx;
79 #endif
80 } e;
81 void *mapping;
82 };
83
84 struct xen_netbk {
85 wait_queue_head_t wq;
86 struct task_struct *task;
87
88 struct sk_buff_head rx_queue;
89 struct sk_buff_head tx_queue;
90
91 struct timer_list net_timer;
92
93 struct page *mmap_pages[MAX_PENDING_REQS];
94
95 pending_ring_idx_t pending_prod;
96 pending_ring_idx_t pending_cons;
97 struct list_head net_schedule_list;
98
99 /* Protect the net_schedule_list in netif. */
100 spinlock_t net_schedule_list_lock;
101
102 atomic_t netfront_count;
103
104 struct pending_tx_info pending_tx_info[MAX_PENDING_REQS];
105 struct gnttab_copy tx_copy_ops[MAX_PENDING_REQS];
106
107 u16 pending_ring[MAX_PENDING_REQS];
108
109 /*
110 * Given MAX_BUFFER_OFFSET of 4096 the worst case is that each
111 * head/fragment page uses 2 copy operations because it
112 * straddles two buffers in the frontend.
113 */
114 struct gnttab_copy grant_copy_op[2*XEN_NETIF_RX_RING_SIZE];
115 struct netbk_rx_meta meta[2*XEN_NETIF_RX_RING_SIZE];
116 };
117
118 static struct xen_netbk *xen_netbk;
119 static int xen_netbk_group_nr;
120
121 void xen_netbk_add_xenvif(struct xenvif *vif)
122 {
123 int i;
124 int min_netfront_count;
125 int min_group = 0;
126 struct xen_netbk *netbk;
127
128 min_netfront_count = atomic_read(&xen_netbk[0].netfront_count);
129 for (i = 0; i < xen_netbk_group_nr; i++) {
130 int netfront_count = atomic_read(&xen_netbk[i].netfront_count);
131 if (netfront_count < min_netfront_count) {
132 min_group = i;
133 min_netfront_count = netfront_count;
134 }
135 }
136
137 netbk = &xen_netbk[min_group];
138
139 vif->netbk = netbk;
140 atomic_inc(&netbk->netfront_count);
141 }
142
143 void xen_netbk_remove_xenvif(struct xenvif *vif)
144 {
145 struct xen_netbk *netbk = vif->netbk;
146 vif->netbk = NULL;
147 atomic_dec(&netbk->netfront_count);
148 }
149
150 static void xen_netbk_idx_release(struct xen_netbk *netbk, u16 pending_idx);
151 static void make_tx_response(struct xenvif *vif,
152 struct xen_netif_tx_request *txp,
153 s8 st);
154 static struct xen_netif_rx_response *make_rx_response(struct xenvif *vif,
155 u16 id,
156 s8 st,
157 u16 offset,
158 u16 size,
159 u16 flags);
160
161 static inline unsigned long idx_to_pfn(struct xen_netbk *netbk,
162 u16 idx)
163 {
164 return page_to_pfn(netbk->mmap_pages[idx]);
165 }
166
167 static inline unsigned long idx_to_kaddr(struct xen_netbk *netbk,
168 u16 idx)
169 {
170 return (unsigned long)pfn_to_kaddr(idx_to_pfn(netbk, idx));
171 }
172
173 /* extra field used in struct page */
174 static inline void set_page_ext(struct page *pg, struct xen_netbk *netbk,
175 unsigned int idx)
176 {
177 unsigned int group = netbk - xen_netbk;
178 union page_ext ext = { .e = { .group = group + 1, .idx = idx } };
179
180 BUILD_BUG_ON(sizeof(ext) > sizeof(ext.mapping));
181 pg->mapping = ext.mapping;
182 }
183
184 static int get_page_ext(struct page *pg,
185 unsigned int *pgroup, unsigned int *pidx)
186 {
187 union page_ext ext = { .mapping = pg->mapping };
188 struct xen_netbk *netbk;
189 unsigned int group, idx;
190
191 group = ext.e.group - 1;
192
193 if (group < 0 || group >= xen_netbk_group_nr)
194 return 0;
195
196 netbk = &xen_netbk[group];
197
198 idx = ext.e.idx;
199
200 if ((idx < 0) || (idx >= MAX_PENDING_REQS))
201 return 0;
202
203 if (netbk->mmap_pages[idx] != pg)
204 return 0;
205
206 *pgroup = group;
207 *pidx = idx;
208
209 return 1;
210 }
211
212 /*
213 * This is the amount of packet we copy rather than map, so that the
214 * guest can't fiddle with the contents of the headers while we do
215 * packet processing on them (netfilter, routing, etc).
216 */
217 #define PKT_PROT_LEN (ETH_HLEN + \
218 VLAN_HLEN + \
219 sizeof(struct iphdr) + MAX_IPOPTLEN + \
220 sizeof(struct tcphdr) + MAX_TCP_OPTION_SPACE)
221
222 static u16 frag_get_pending_idx(skb_frag_t *frag)
223 {
224 return (u16)frag->page_offset;
225 }
226
227 static void frag_set_pending_idx(skb_frag_t *frag, u16 pending_idx)
228 {
229 frag->page_offset = pending_idx;
230 }
231
232 static inline pending_ring_idx_t pending_index(unsigned i)
233 {
234 return i & (MAX_PENDING_REQS-1);
235 }
236
237 static inline pending_ring_idx_t nr_pending_reqs(struct xen_netbk *netbk)
238 {
239 return MAX_PENDING_REQS -
240 netbk->pending_prod + netbk->pending_cons;
241 }
242
243 static void xen_netbk_kick_thread(struct xen_netbk *netbk)
244 {
245 wake_up(&netbk->wq);
246 }
247
248 static int max_required_rx_slots(struct xenvif *vif)
249 {
250 int max = DIV_ROUND_UP(vif->dev->mtu, PAGE_SIZE);
251
252 if (vif->can_sg || vif->gso || vif->gso_prefix)
253 max += MAX_SKB_FRAGS + 1; /* extra_info + frags */
254
255 return max;
256 }
257
258 int xen_netbk_rx_ring_full(struct xenvif *vif)
259 {
260 RING_IDX peek = vif->rx_req_cons_peek;
261 RING_IDX needed = max_required_rx_slots(vif);
262
263 return ((vif->rx.sring->req_prod - peek) < needed) ||
264 ((vif->rx.rsp_prod_pvt + XEN_NETIF_RX_RING_SIZE - peek) < needed);
265 }
266
267 int xen_netbk_must_stop_queue(struct xenvif *vif)
268 {
269 if (!xen_netbk_rx_ring_full(vif))
270 return 0;
271
272 vif->rx.sring->req_event = vif->rx_req_cons_peek +
273 max_required_rx_slots(vif);
274 mb(); /* request notification /then/ check the queue */
275
276 return xen_netbk_rx_ring_full(vif);
277 }
278
279 /*
280 * Returns true if we should start a new receive buffer instead of
281 * adding 'size' bytes to a buffer which currently contains 'offset'
282 * bytes.
283 */
284 static bool start_new_rx_buffer(int offset, unsigned long size, int head)
285 {
286 /* simple case: we have completely filled the current buffer. */
287 if (offset == MAX_BUFFER_OFFSET)
288 return true;
289
290 /*
291 * complex case: start a fresh buffer if the current frag
292 * would overflow the current buffer but only if:
293 * (i) this frag would fit completely in the next buffer
294 * and (ii) there is already some data in the current buffer
295 * and (iii) this is not the head buffer.
296 *
297 * Where:
298 * - (i) stops us splitting a frag into two copies
299 * unless the frag is too large for a single buffer.
300 * - (ii) stops us from leaving a buffer pointlessly empty.
301 * - (iii) stops us leaving the first buffer
302 * empty. Strictly speaking this is already covered
303 * by (ii) but is explicitly checked because
304 * netfront relies on the first buffer being
305 * non-empty and can crash otherwise.
306 *
307 * This means we will effectively linearise small
308 * frags but do not needlessly split large buffers
309 * into multiple copies tend to give large frags their
310 * own buffers as before.
311 */
312 if ((offset + size > MAX_BUFFER_OFFSET) &&
313 (size <= MAX_BUFFER_OFFSET) && offset && !head)
314 return true;
315
316 return false;
317 }
318
319 /*
320 * Figure out how many ring slots we're going to need to send @skb to
321 * the guest. This function is essentially a dry run of
322 * netbk_gop_frag_copy.
323 */
324 unsigned int xen_netbk_count_skb_slots(struct xenvif *vif, struct sk_buff *skb)
325 {
326 unsigned int count;
327 int i, copy_off;
328
329 count = DIV_ROUND_UP(skb_headlen(skb), PAGE_SIZE);
330
331 copy_off = skb_headlen(skb) % PAGE_SIZE;
332
333 if (skb_shinfo(skb)->gso_size)
334 count++;
335
336 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
337 unsigned long size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
338 unsigned long offset = skb_shinfo(skb)->frags[i].page_offset;
339 unsigned long bytes;
340
341 offset &= ~PAGE_MASK;
342
343 while (size > 0) {
344 BUG_ON(offset >= PAGE_SIZE);
345 BUG_ON(copy_off > MAX_BUFFER_OFFSET);
346
347 bytes = PAGE_SIZE - offset;
348
349 if (bytes > size)
350 bytes = size;
351
352 if (start_new_rx_buffer(copy_off, bytes, 0)) {
353 count++;
354 copy_off = 0;
355 }
356
357 if (copy_off + bytes > MAX_BUFFER_OFFSET)
358 bytes = MAX_BUFFER_OFFSET - copy_off;
359
360 copy_off += bytes;
361
362 offset += bytes;
363 size -= bytes;
364
365 if (offset == PAGE_SIZE)
366 offset = 0;
367 }
368 }
369 return count;
370 }
371
372 struct netrx_pending_operations {
373 unsigned copy_prod, copy_cons;
374 unsigned meta_prod, meta_cons;
375 struct gnttab_copy *copy;
376 struct netbk_rx_meta *meta;
377 int copy_off;
378 grant_ref_t copy_gref;
379 };
380
381 static struct netbk_rx_meta *get_next_rx_buffer(struct xenvif *vif,
382 struct netrx_pending_operations *npo)
383 {
384 struct netbk_rx_meta *meta;
385 struct xen_netif_rx_request *req;
386
387 req = RING_GET_REQUEST(&vif->rx, vif->rx.req_cons++);
388
389 meta = npo->meta + npo->meta_prod++;
390 meta->gso_size = 0;
391 meta->size = 0;
392 meta->id = req->id;
393
394 npo->copy_off = 0;
395 npo->copy_gref = req->gref;
396
397 return meta;
398 }
399
400 /*
401 * Set up the grant operations for this fragment. If it's a flipping
402 * interface, we also set up the unmap request from here.
403 */
404 static void netbk_gop_frag_copy(struct xenvif *vif, struct sk_buff *skb,
405 struct netrx_pending_operations *npo,
406 struct page *page, unsigned long size,
407 unsigned long offset, int *head)
408 {
409 struct gnttab_copy *copy_gop;
410 struct netbk_rx_meta *meta;
411 /*
412 * These variables are used iff get_page_ext returns true,
413 * in which case they are guaranteed to be initialized.
414 */
415 unsigned int uninitialized_var(group), uninitialized_var(idx);
416 int foreign = get_page_ext(page, &group, &idx);
417 unsigned long bytes;
418
419 /* Data must not cross a page boundary. */
420 BUG_ON(size + offset > PAGE_SIZE<<compound_order(page));
421
422 meta = npo->meta + npo->meta_prod - 1;
423
424 /* Skip unused frames from start of page */
425 page += offset >> PAGE_SHIFT;
426 offset &= ~PAGE_MASK;
427
428 while (size > 0) {
429 BUG_ON(offset >= PAGE_SIZE);
430 BUG_ON(npo->copy_off > MAX_BUFFER_OFFSET);
431
432 bytes = PAGE_SIZE - offset;
433
434 if (bytes > size)
435 bytes = size;
436
437 if (start_new_rx_buffer(npo->copy_off, bytes, *head)) {
438 /*
439 * Netfront requires there to be some data in the head
440 * buffer.
441 */
442 BUG_ON(*head);
443
444 meta = get_next_rx_buffer(vif, npo);
445 }
446
447 if (npo->copy_off + bytes > MAX_BUFFER_OFFSET)
448 bytes = MAX_BUFFER_OFFSET - npo->copy_off;
449
450 copy_gop = npo->copy + npo->copy_prod++;
451 copy_gop->flags = GNTCOPY_dest_gref;
452 if (foreign) {
453 struct xen_netbk *netbk = &xen_netbk[group];
454 struct pending_tx_info *src_pend;
455
456 src_pend = &netbk->pending_tx_info[idx];
457
458 copy_gop->source.domid = src_pend->vif->domid;
459 copy_gop->source.u.ref = src_pend->req.gref;
460 copy_gop->flags |= GNTCOPY_source_gref;
461 } else {
462 void *vaddr = page_address(page);
463 copy_gop->source.domid = DOMID_SELF;
464 copy_gop->source.u.gmfn = virt_to_mfn(vaddr);
465 }
466 copy_gop->source.offset = offset;
467 copy_gop->dest.domid = vif->domid;
468
469 copy_gop->dest.offset = npo->copy_off;
470 copy_gop->dest.u.ref = npo->copy_gref;
471 copy_gop->len = bytes;
472
473 npo->copy_off += bytes;
474 meta->size += bytes;
475
476 offset += bytes;
477 size -= bytes;
478
479 /* Next frame */
480 if (offset == PAGE_SIZE && size) {
481 BUG_ON(!PageCompound(page));
482 page++;
483 offset = 0;
484 }
485
486 /* Leave a gap for the GSO descriptor. */
487 if (*head && skb_shinfo(skb)->gso_size && !vif->gso_prefix)
488 vif->rx.req_cons++;
489
490 *head = 0; /* There must be something in this buffer now. */
491
492 }
493 }
494
495 /*
496 * Prepare an SKB to be transmitted to the frontend.
497 *
498 * This function is responsible for allocating grant operations, meta
499 * structures, etc.
500 *
501 * It returns the number of meta structures consumed. The number of
502 * ring slots used is always equal to the number of meta slots used
503 * plus the number of GSO descriptors used. Currently, we use either
504 * zero GSO descriptors (for non-GSO packets) or one descriptor (for
505 * frontend-side LRO).
506 */
507 static int netbk_gop_skb(struct sk_buff *skb,
508 struct netrx_pending_operations *npo)
509 {
510 struct xenvif *vif = netdev_priv(skb->dev);
511 int nr_frags = skb_shinfo(skb)->nr_frags;
512 int i;
513 struct xen_netif_rx_request *req;
514 struct netbk_rx_meta *meta;
515 unsigned char *data;
516 int head = 1;
517 int old_meta_prod;
518
519 old_meta_prod = npo->meta_prod;
520
521 /* Set up a GSO prefix descriptor, if necessary */
522 if (skb_shinfo(skb)->gso_size && vif->gso_prefix) {
523 req = RING_GET_REQUEST(&vif->rx, vif->rx.req_cons++);
524 meta = npo->meta + npo->meta_prod++;
525 meta->gso_size = skb_shinfo(skb)->gso_size;
526 meta->size = 0;
527 meta->id = req->id;
528 }
529
530 req = RING_GET_REQUEST(&vif->rx, vif->rx.req_cons++);
531 meta = npo->meta + npo->meta_prod++;
532
533 if (!vif->gso_prefix)
534 meta->gso_size = skb_shinfo(skb)->gso_size;
535 else
536 meta->gso_size = 0;
537
538 meta->size = 0;
539 meta->id = req->id;
540 npo->copy_off = 0;
541 npo->copy_gref = req->gref;
542
543 data = skb->data;
544 while (data < skb_tail_pointer(skb)) {
545 unsigned int offset = offset_in_page(data);
546 unsigned int len = PAGE_SIZE - offset;
547
548 if (data + len > skb_tail_pointer(skb))
549 len = skb_tail_pointer(skb) - data;
550
551 netbk_gop_frag_copy(vif, skb, npo,
552 virt_to_page(data), len, offset, &head);
553 data += len;
554 }
555
556 for (i = 0; i < nr_frags; i++) {
557 netbk_gop_frag_copy(vif, skb, npo,
558 skb_frag_page(&skb_shinfo(skb)->frags[i]),
559 skb_frag_size(&skb_shinfo(skb)->frags[i]),
560 skb_shinfo(skb)->frags[i].page_offset,
561 &head);
562 }
563
564 return npo->meta_prod - old_meta_prod;
565 }
566
567 /*
568 * This is a twin to netbk_gop_skb. Assume that netbk_gop_skb was
569 * used to set up the operations on the top of
570 * netrx_pending_operations, which have since been done. Check that
571 * they didn't give any errors and advance over them.
572 */
573 static int netbk_check_gop(struct xenvif *vif, int nr_meta_slots,
574 struct netrx_pending_operations *npo)
575 {
576 struct gnttab_copy *copy_op;
577 int status = XEN_NETIF_RSP_OKAY;
578 int i;
579
580 for (i = 0; i < nr_meta_slots; i++) {
581 copy_op = npo->copy + npo->copy_cons++;
582 if (copy_op->status != GNTST_okay) {
583 netdev_dbg(vif->dev,
584 "Bad status %d from copy to DOM%d.\n",
585 copy_op->status, vif->domid);
586 status = XEN_NETIF_RSP_ERROR;
587 }
588 }
589
590 return status;
591 }
592
593 static void netbk_add_frag_responses(struct xenvif *vif, int status,
594 struct netbk_rx_meta *meta,
595 int nr_meta_slots)
596 {
597 int i;
598 unsigned long offset;
599
600 /* No fragments used */
601 if (nr_meta_slots <= 1)
602 return;
603
604 nr_meta_slots--;
605
606 for (i = 0; i < nr_meta_slots; i++) {
607 int flags;
608 if (i == nr_meta_slots - 1)
609 flags = 0;
610 else
611 flags = XEN_NETRXF_more_data;
612
613 offset = 0;
614 make_rx_response(vif, meta[i].id, status, offset,
615 meta[i].size, flags);
616 }
617 }
618
619 struct skb_cb_overlay {
620 int meta_slots_used;
621 };
622
623 static void xen_netbk_rx_action(struct xen_netbk *netbk)
624 {
625 struct xenvif *vif = NULL, *tmp;
626 s8 status;
627 u16 irq, flags;
628 struct xen_netif_rx_response *resp;
629 struct sk_buff_head rxq;
630 struct sk_buff *skb;
631 LIST_HEAD(notify);
632 int ret;
633 int nr_frags;
634 int count;
635 unsigned long offset;
636 struct skb_cb_overlay *sco;
637
638 struct netrx_pending_operations npo = {
639 .copy = netbk->grant_copy_op,
640 .meta = netbk->meta,
641 };
642
643 skb_queue_head_init(&rxq);
644
645 count = 0;
646
647 while ((skb = skb_dequeue(&netbk->rx_queue)) != NULL) {
648 vif = netdev_priv(skb->dev);
649 nr_frags = skb_shinfo(skb)->nr_frags;
650
651 sco = (struct skb_cb_overlay *)skb->cb;
652 sco->meta_slots_used = netbk_gop_skb(skb, &npo);
653
654 count += nr_frags + 1;
655
656 __skb_queue_tail(&rxq, skb);
657
658 /* Filled the batch queue? */
659 if (count + MAX_SKB_FRAGS >= XEN_NETIF_RX_RING_SIZE)
660 break;
661 }
662
663 BUG_ON(npo.meta_prod > ARRAY_SIZE(netbk->meta));
664
665 if (!npo.copy_prod)
666 return;
667
668 BUG_ON(npo.copy_prod > ARRAY_SIZE(netbk->grant_copy_op));
669 gnttab_batch_copy(netbk->grant_copy_op, npo.copy_prod);
670
671 while ((skb = __skb_dequeue(&rxq)) != NULL) {
672 sco = (struct skb_cb_overlay *)skb->cb;
673
674 vif = netdev_priv(skb->dev);
675
676 if (netbk->meta[npo.meta_cons].gso_size && vif->gso_prefix) {
677 resp = RING_GET_RESPONSE(&vif->rx,
678 vif->rx.rsp_prod_pvt++);
679
680 resp->flags = XEN_NETRXF_gso_prefix | XEN_NETRXF_more_data;
681
682 resp->offset = netbk->meta[npo.meta_cons].gso_size;
683 resp->id = netbk->meta[npo.meta_cons].id;
684 resp->status = sco->meta_slots_used;
685
686 npo.meta_cons++;
687 sco->meta_slots_used--;
688 }
689
690
691 vif->dev->stats.tx_bytes += skb->len;
692 vif->dev->stats.tx_packets++;
693
694 status = netbk_check_gop(vif, sco->meta_slots_used, &npo);
695
696 if (sco->meta_slots_used == 1)
697 flags = 0;
698 else
699 flags = XEN_NETRXF_more_data;
700
701 if (skb->ip_summed == CHECKSUM_PARTIAL) /* local packet? */
702 flags |= XEN_NETRXF_csum_blank | XEN_NETRXF_data_validated;
703 else if (skb->ip_summed == CHECKSUM_UNNECESSARY)
704 /* remote but checksummed. */
705 flags |= XEN_NETRXF_data_validated;
706
707 offset = 0;
708 resp = make_rx_response(vif, netbk->meta[npo.meta_cons].id,
709 status, offset,
710 netbk->meta[npo.meta_cons].size,
711 flags);
712
713 if (netbk->meta[npo.meta_cons].gso_size && !vif->gso_prefix) {
714 struct xen_netif_extra_info *gso =
715 (struct xen_netif_extra_info *)
716 RING_GET_RESPONSE(&vif->rx,
717 vif->rx.rsp_prod_pvt++);
718
719 resp->flags |= XEN_NETRXF_extra_info;
720
721 gso->u.gso.size = netbk->meta[npo.meta_cons].gso_size;
722 gso->u.gso.type = XEN_NETIF_GSO_TYPE_TCPV4;
723 gso->u.gso.pad = 0;
724 gso->u.gso.features = 0;
725
726 gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
727 gso->flags = 0;
728 }
729
730 netbk_add_frag_responses(vif, status,
731 netbk->meta + npo.meta_cons + 1,
732 sco->meta_slots_used);
733
734 RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&vif->rx, ret);
735 irq = vif->irq;
736 if (ret && list_empty(&vif->notify_list))
737 list_add_tail(&vif->notify_list, &notify);
738
739 xenvif_notify_tx_completion(vif);
740
741 xenvif_put(vif);
742 npo.meta_cons += sco->meta_slots_used;
743 dev_kfree_skb(skb);
744 }
745
746 list_for_each_entry_safe(vif, tmp, &notify, notify_list) {
747 notify_remote_via_irq(vif->irq);
748 list_del_init(&vif->notify_list);
749 }
750
751 /* More work to do? */
752 if (!skb_queue_empty(&netbk->rx_queue) &&
753 !timer_pending(&netbk->net_timer))
754 xen_netbk_kick_thread(netbk);
755 }
756
757 void xen_netbk_queue_tx_skb(struct xenvif *vif, struct sk_buff *skb)
758 {
759 struct xen_netbk *netbk = vif->netbk;
760
761 skb_queue_tail(&netbk->rx_queue, skb);
762
763 xen_netbk_kick_thread(netbk);
764 }
765
766 static void xen_netbk_alarm(unsigned long data)
767 {
768 struct xen_netbk *netbk = (struct xen_netbk *)data;
769 xen_netbk_kick_thread(netbk);
770 }
771
772 static int __on_net_schedule_list(struct xenvif *vif)
773 {
774 return !list_empty(&vif->schedule_list);
775 }
776
777 /* Must be called with net_schedule_list_lock held */
778 static void remove_from_net_schedule_list(struct xenvif *vif)
779 {
780 if (likely(__on_net_schedule_list(vif))) {
781 list_del_init(&vif->schedule_list);
782 xenvif_put(vif);
783 }
784 }
785
786 static struct xenvif *poll_net_schedule_list(struct xen_netbk *netbk)
787 {
788 struct xenvif *vif = NULL;
789
790 spin_lock_irq(&netbk->net_schedule_list_lock);
791 if (list_empty(&netbk->net_schedule_list))
792 goto out;
793
794 vif = list_first_entry(&netbk->net_schedule_list,
795 struct xenvif, schedule_list);
796 if (!vif)
797 goto out;
798
799 xenvif_get(vif);
800
801 remove_from_net_schedule_list(vif);
802 out:
803 spin_unlock_irq(&netbk->net_schedule_list_lock);
804 return vif;
805 }
806
807 void xen_netbk_schedule_xenvif(struct xenvif *vif)
808 {
809 unsigned long flags;
810 struct xen_netbk *netbk = vif->netbk;
811
812 if (__on_net_schedule_list(vif))
813 goto kick;
814
815 spin_lock_irqsave(&netbk->net_schedule_list_lock, flags);
816 if (!__on_net_schedule_list(vif) &&
817 likely(xenvif_schedulable(vif))) {
818 list_add_tail(&vif->schedule_list, &netbk->net_schedule_list);
819 xenvif_get(vif);
820 }
821 spin_unlock_irqrestore(&netbk->net_schedule_list_lock, flags);
822
823 kick:
824 smp_mb();
825 if ((nr_pending_reqs(netbk) < (MAX_PENDING_REQS/2)) &&
826 !list_empty(&netbk->net_schedule_list))
827 xen_netbk_kick_thread(netbk);
828 }
829
830 void xen_netbk_deschedule_xenvif(struct xenvif *vif)
831 {
832 struct xen_netbk *netbk = vif->netbk;
833 spin_lock_irq(&netbk->net_schedule_list_lock);
834 remove_from_net_schedule_list(vif);
835 spin_unlock_irq(&netbk->net_schedule_list_lock);
836 }
837
838 void xen_netbk_check_rx_xenvif(struct xenvif *vif)
839 {
840 int more_to_do;
841
842 RING_FINAL_CHECK_FOR_REQUESTS(&vif->tx, more_to_do);
843
844 if (more_to_do)
845 xen_netbk_schedule_xenvif(vif);
846 }
847
848 static void tx_add_credit(struct xenvif *vif)
849 {
850 unsigned long max_burst, max_credit;
851
852 /*
853 * Allow a burst big enough to transmit a jumbo packet of up to 128kB.
854 * Otherwise the interface can seize up due to insufficient credit.
855 */
856 max_burst = RING_GET_REQUEST(&vif->tx, vif->tx.req_cons)->size;
857 max_burst = min(max_burst, 131072UL);
858 max_burst = max(max_burst, vif->credit_bytes);
859
860 /* Take care that adding a new chunk of credit doesn't wrap to zero. */
861 max_credit = vif->remaining_credit + vif->credit_bytes;
862 if (max_credit < vif->remaining_credit)
863 max_credit = ULONG_MAX; /* wrapped: clamp to ULONG_MAX */
864
865 vif->remaining_credit = min(max_credit, max_burst);
866 }
867
868 static void tx_credit_callback(unsigned long data)
869 {
870 struct xenvif *vif = (struct xenvif *)data;
871 tx_add_credit(vif);
872 xen_netbk_check_rx_xenvif(vif);
873 }
874
875 static void netbk_tx_err(struct xenvif *vif,
876 struct xen_netif_tx_request *txp, RING_IDX end)
877 {
878 RING_IDX cons = vif->tx.req_cons;
879
880 do {
881 make_tx_response(vif, txp, XEN_NETIF_RSP_ERROR);
882 if (cons >= end)
883 break;
884 txp = RING_GET_REQUEST(&vif->tx, cons++);
885 } while (1);
886 vif->tx.req_cons = cons;
887 xen_netbk_check_rx_xenvif(vif);
888 xenvif_put(vif);
889 }
890
891 static int netbk_count_requests(struct xenvif *vif,
892 struct xen_netif_tx_request *first,
893 struct xen_netif_tx_request *txp,
894 int work_to_do)
895 {
896 RING_IDX cons = vif->tx.req_cons;
897 int frags = 0;
898
899 if (!(first->flags & XEN_NETTXF_more_data))
900 return 0;
901
902 do {
903 if (frags >= work_to_do) {
904 netdev_dbg(vif->dev, "Need more frags\n");
905 return -frags;
906 }
907
908 if (unlikely(frags >= MAX_SKB_FRAGS)) {
909 netdev_dbg(vif->dev, "Too many frags\n");
910 return -frags;
911 }
912
913 memcpy(txp, RING_GET_REQUEST(&vif->tx, cons + frags),
914 sizeof(*txp));
915 if (txp->size > first->size) {
916 netdev_dbg(vif->dev, "Frags galore\n");
917 return -frags;
918 }
919
920 first->size -= txp->size;
921 frags++;
922
923 if (unlikely((txp->offset + txp->size) > PAGE_SIZE)) {
924 netdev_dbg(vif->dev, "txp->offset: %x, size: %u\n",
925 txp->offset, txp->size);
926 return -frags;
927 }
928 } while ((txp++)->flags & XEN_NETTXF_more_data);
929 return frags;
930 }
931
932 static struct page *xen_netbk_alloc_page(struct xen_netbk *netbk,
933 struct sk_buff *skb,
934 u16 pending_idx)
935 {
936 struct page *page;
937 page = alloc_page(GFP_KERNEL|__GFP_COLD);
938 if (!page)
939 return NULL;
940 set_page_ext(page, netbk, pending_idx);
941 netbk->mmap_pages[pending_idx] = page;
942 return page;
943 }
944
945 static struct gnttab_copy *xen_netbk_get_requests(struct xen_netbk *netbk,
946 struct xenvif *vif,
947 struct sk_buff *skb,
948 struct xen_netif_tx_request *txp,
949 struct gnttab_copy *gop)
950 {
951 struct skb_shared_info *shinfo = skb_shinfo(skb);
952 skb_frag_t *frags = shinfo->frags;
953 u16 pending_idx = *((u16 *)skb->data);
954 int i, start;
955
956 /* Skip first skb fragment if it is on same page as header fragment. */
957 start = (frag_get_pending_idx(&shinfo->frags[0]) == pending_idx);
958
959 for (i = start; i < shinfo->nr_frags; i++, txp++) {
960 struct page *page;
961 pending_ring_idx_t index;
962 struct pending_tx_info *pending_tx_info =
963 netbk->pending_tx_info;
964
965 index = pending_index(netbk->pending_cons++);
966 pending_idx = netbk->pending_ring[index];
967 page = xen_netbk_alloc_page(netbk, skb, pending_idx);
968 if (!page)
969 return NULL;
970
971 gop->source.u.ref = txp->gref;
972 gop->source.domid = vif->domid;
973 gop->source.offset = txp->offset;
974
975 gop->dest.u.gmfn = virt_to_mfn(page_address(page));
976 gop->dest.domid = DOMID_SELF;
977 gop->dest.offset = txp->offset;
978
979 gop->len = txp->size;
980 gop->flags = GNTCOPY_source_gref;
981
982 gop++;
983
984 memcpy(&pending_tx_info[pending_idx].req, txp, sizeof(*txp));
985 xenvif_get(vif);
986 pending_tx_info[pending_idx].vif = vif;
987 frag_set_pending_idx(&frags[i], pending_idx);
988 }
989
990 return gop;
991 }
992
993 static int xen_netbk_tx_check_gop(struct xen_netbk *netbk,
994 struct sk_buff *skb,
995 struct gnttab_copy **gopp)
996 {
997 struct gnttab_copy *gop = *gopp;
998 u16 pending_idx = *((u16 *)skb->data);
999 struct pending_tx_info *pending_tx_info = netbk->pending_tx_info;
1000 struct xenvif *vif = pending_tx_info[pending_idx].vif;
1001 struct xen_netif_tx_request *txp;
1002 struct skb_shared_info *shinfo = skb_shinfo(skb);
1003 int nr_frags = shinfo->nr_frags;
1004 int i, err, start;
1005
1006 /* Check status of header. */
1007 err = gop->status;
1008 if (unlikely(err)) {
1009 pending_ring_idx_t index;
1010 index = pending_index(netbk->pending_prod++);
1011 txp = &pending_tx_info[pending_idx].req;
1012 make_tx_response(vif, txp, XEN_NETIF_RSP_ERROR);
1013 netbk->pending_ring[index] = pending_idx;
1014 xenvif_put(vif);
1015 }
1016
1017 /* Skip first skb fragment if it is on same page as header fragment. */
1018 start = (frag_get_pending_idx(&shinfo->frags[0]) == pending_idx);
1019
1020 for (i = start; i < nr_frags; i++) {
1021 int j, newerr;
1022 pending_ring_idx_t index;
1023
1024 pending_idx = frag_get_pending_idx(&shinfo->frags[i]);
1025
1026 /* Check error status: if okay then remember grant handle. */
1027 newerr = (++gop)->status;
1028 if (likely(!newerr)) {
1029 /* Had a previous error? Invalidate this fragment. */
1030 if (unlikely(err))
1031 xen_netbk_idx_release(netbk, pending_idx);
1032 continue;
1033 }
1034
1035 /* Error on this fragment: respond to client with an error. */
1036 txp = &netbk->pending_tx_info[pending_idx].req;
1037 make_tx_response(vif, txp, XEN_NETIF_RSP_ERROR);
1038 index = pending_index(netbk->pending_prod++);
1039 netbk->pending_ring[index] = pending_idx;
1040 xenvif_put(vif);
1041
1042 /* Not the first error? Preceding frags already invalidated. */
1043 if (err)
1044 continue;
1045
1046 /* First error: invalidate header and preceding fragments. */
1047 pending_idx = *((u16 *)skb->data);
1048 xen_netbk_idx_release(netbk, pending_idx);
1049 for (j = start; j < i; j++) {
1050 pending_idx = frag_get_pending_idx(&shinfo->frags[j]);
1051 xen_netbk_idx_release(netbk, pending_idx);
1052 }
1053
1054 /* Remember the error: invalidate all subsequent fragments. */
1055 err = newerr;
1056 }
1057
1058 *gopp = gop + 1;
1059 return err;
1060 }
1061
1062 static void xen_netbk_fill_frags(struct xen_netbk *netbk, struct sk_buff *skb)
1063 {
1064 struct skb_shared_info *shinfo = skb_shinfo(skb);
1065 int nr_frags = shinfo->nr_frags;
1066 int i;
1067
1068 for (i = 0; i < nr_frags; i++) {
1069 skb_frag_t *frag = shinfo->frags + i;
1070 struct xen_netif_tx_request *txp;
1071 struct page *page;
1072 u16 pending_idx;
1073
1074 pending_idx = frag_get_pending_idx(frag);
1075
1076 txp = &netbk->pending_tx_info[pending_idx].req;
1077 page = virt_to_page(idx_to_kaddr(netbk, pending_idx));
1078 __skb_fill_page_desc(skb, i, page, txp->offset, txp->size);
1079 skb->len += txp->size;
1080 skb->data_len += txp->size;
1081 skb->truesize += txp->size;
1082
1083 /* Take an extra reference to offset xen_netbk_idx_release */
1084 get_page(netbk->mmap_pages[pending_idx]);
1085 xen_netbk_idx_release(netbk, pending_idx);
1086 }
1087 }
1088
1089 static int xen_netbk_get_extras(struct xenvif *vif,
1090 struct xen_netif_extra_info *extras,
1091 int work_to_do)
1092 {
1093 struct xen_netif_extra_info extra;
1094 RING_IDX cons = vif->tx.req_cons;
1095
1096 do {
1097 if (unlikely(work_to_do-- <= 0)) {
1098 netdev_dbg(vif->dev, "Missing extra info\n");
1099 return -EBADR;
1100 }
1101
1102 memcpy(&extra, RING_GET_REQUEST(&vif->tx, cons),
1103 sizeof(extra));
1104 if (unlikely(!extra.type ||
1105 extra.type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
1106 vif->tx.req_cons = ++cons;
1107 netdev_dbg(vif->dev,
1108 "Invalid extra type: %d\n", extra.type);
1109 return -EINVAL;
1110 }
1111
1112 memcpy(&extras[extra.type - 1], &extra, sizeof(extra));
1113 vif->tx.req_cons = ++cons;
1114 } while (extra.flags & XEN_NETIF_EXTRA_FLAG_MORE);
1115
1116 return work_to_do;
1117 }
1118
1119 static int netbk_set_skb_gso(struct xenvif *vif,
1120 struct sk_buff *skb,
1121 struct xen_netif_extra_info *gso)
1122 {
1123 if (!gso->u.gso.size) {
1124 netdev_dbg(vif->dev, "GSO size must not be zero.\n");
1125 return -EINVAL;
1126 }
1127
1128 /* Currently only TCPv4 S.O. is supported. */
1129 if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4) {
1130 netdev_dbg(vif->dev, "Bad GSO type %d.\n", gso->u.gso.type);
1131 return -EINVAL;
1132 }
1133
1134 skb_shinfo(skb)->gso_size = gso->u.gso.size;
1135 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1136
1137 /* Header must be checked, and gso_segs computed. */
1138 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1139 skb_shinfo(skb)->gso_segs = 0;
1140
1141 return 0;
1142 }
1143
1144 static int checksum_setup(struct xenvif *vif, struct sk_buff *skb)
1145 {
1146 struct iphdr *iph;
1147 unsigned char *th;
1148 int err = -EPROTO;
1149 int recalculate_partial_csum = 0;
1150
1151 /*
1152 * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy
1153 * peers can fail to set NETRXF_csum_blank when sending a GSO
1154 * frame. In this case force the SKB to CHECKSUM_PARTIAL and
1155 * recalculate the partial checksum.
1156 */
1157 if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) {
1158 vif->rx_gso_checksum_fixup++;
1159 skb->ip_summed = CHECKSUM_PARTIAL;
1160 recalculate_partial_csum = 1;
1161 }
1162
1163 /* A non-CHECKSUM_PARTIAL SKB does not require setup. */
1164 if (skb->ip_summed != CHECKSUM_PARTIAL)
1165 return 0;
1166
1167 if (skb->protocol != htons(ETH_P_IP))
1168 goto out;
1169
1170 iph = (void *)skb->data;
1171 th = skb->data + 4 * iph->ihl;
1172 if (th >= skb_tail_pointer(skb))
1173 goto out;
1174
1175 skb->csum_start = th - skb->head;
1176 switch (iph->protocol) {
1177 case IPPROTO_TCP:
1178 skb->csum_offset = offsetof(struct tcphdr, check);
1179
1180 if (recalculate_partial_csum) {
1181 struct tcphdr *tcph = (struct tcphdr *)th;
1182 tcph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
1183 skb->len - iph->ihl*4,
1184 IPPROTO_TCP, 0);
1185 }
1186 break;
1187 case IPPROTO_UDP:
1188 skb->csum_offset = offsetof(struct udphdr, check);
1189
1190 if (recalculate_partial_csum) {
1191 struct udphdr *udph = (struct udphdr *)th;
1192 udph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
1193 skb->len - iph->ihl*4,
1194 IPPROTO_UDP, 0);
1195 }
1196 break;
1197 default:
1198 if (net_ratelimit())
1199 netdev_err(vif->dev,
1200 "Attempting to checksum a non-TCP/UDP packet, dropping a protocol %d packet\n",
1201 iph->protocol);
1202 goto out;
1203 }
1204
1205 if ((th + skb->csum_offset + 2) > skb_tail_pointer(skb))
1206 goto out;
1207
1208 err = 0;
1209
1210 out:
1211 return err;
1212 }
1213
1214 static bool tx_credit_exceeded(struct xenvif *vif, unsigned size)
1215 {
1216 unsigned long now = jiffies;
1217 unsigned long next_credit =
1218 vif->credit_timeout.expires +
1219 msecs_to_jiffies(vif->credit_usec / 1000);
1220
1221 /* Timer could already be pending in rare cases. */
1222 if (timer_pending(&vif->credit_timeout))
1223 return true;
1224
1225 /* Passed the point where we can replenish credit? */
1226 if (time_after_eq(now, next_credit)) {
1227 vif->credit_timeout.expires = now;
1228 tx_add_credit(vif);
1229 }
1230
1231 /* Still too big to send right now? Set a callback. */
1232 if (size > vif->remaining_credit) {
1233 vif->credit_timeout.data =
1234 (unsigned long)vif;
1235 vif->credit_timeout.function =
1236 tx_credit_callback;
1237 mod_timer(&vif->credit_timeout,
1238 next_credit);
1239
1240 return true;
1241 }
1242
1243 return false;
1244 }
1245
1246 static unsigned xen_netbk_tx_build_gops(struct xen_netbk *netbk)
1247 {
1248 struct gnttab_copy *gop = netbk->tx_copy_ops, *request_gop;
1249 struct sk_buff *skb;
1250 int ret;
1251
1252 while (((nr_pending_reqs(netbk) + MAX_SKB_FRAGS) < MAX_PENDING_REQS) &&
1253 !list_empty(&netbk->net_schedule_list)) {
1254 struct xenvif *vif;
1255 struct xen_netif_tx_request txreq;
1256 struct xen_netif_tx_request txfrags[MAX_SKB_FRAGS];
1257 struct page *page;
1258 struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX-1];
1259 u16 pending_idx;
1260 RING_IDX idx;
1261 int work_to_do;
1262 unsigned int data_len;
1263 pending_ring_idx_t index;
1264
1265 /* Get a netif from the list with work to do. */
1266 vif = poll_net_schedule_list(netbk);
1267 if (!vif)
1268 continue;
1269
1270 RING_FINAL_CHECK_FOR_REQUESTS(&vif->tx, work_to_do);
1271 if (!work_to_do) {
1272 xenvif_put(vif);
1273 continue;
1274 }
1275
1276 idx = vif->tx.req_cons;
1277 rmb(); /* Ensure that we see the request before we copy it. */
1278 memcpy(&txreq, RING_GET_REQUEST(&vif->tx, idx), sizeof(txreq));
1279
1280 /* Credit-based scheduling. */
1281 if (txreq.size > vif->remaining_credit &&
1282 tx_credit_exceeded(vif, txreq.size)) {
1283 xenvif_put(vif);
1284 continue;
1285 }
1286
1287 vif->remaining_credit -= txreq.size;
1288
1289 work_to_do--;
1290 vif->tx.req_cons = ++idx;
1291
1292 memset(extras, 0, sizeof(extras));
1293 if (txreq.flags & XEN_NETTXF_extra_info) {
1294 work_to_do = xen_netbk_get_extras(vif, extras,
1295 work_to_do);
1296 idx = vif->tx.req_cons;
1297 if (unlikely(work_to_do < 0)) {
1298 netbk_tx_err(vif, &txreq, idx);
1299 continue;
1300 }
1301 }
1302
1303 ret = netbk_count_requests(vif, &txreq, txfrags, work_to_do);
1304 if (unlikely(ret < 0)) {
1305 netbk_tx_err(vif, &txreq, idx - ret);
1306 continue;
1307 }
1308 idx += ret;
1309
1310 if (unlikely(txreq.size < ETH_HLEN)) {
1311 netdev_dbg(vif->dev,
1312 "Bad packet size: %d\n", txreq.size);
1313 netbk_tx_err(vif, &txreq, idx);
1314 continue;
1315 }
1316
1317 /* No crossing a page as the payload mustn't fragment. */
1318 if (unlikely((txreq.offset + txreq.size) > PAGE_SIZE)) {
1319 netdev_dbg(vif->dev,
1320 "txreq.offset: %x, size: %u, end: %lu\n",
1321 txreq.offset, txreq.size,
1322 (txreq.offset&~PAGE_MASK) + txreq.size);
1323 netbk_tx_err(vif, &txreq, idx);
1324 continue;
1325 }
1326
1327 index = pending_index(netbk->pending_cons);
1328 pending_idx = netbk->pending_ring[index];
1329
1330 data_len = (txreq.size > PKT_PROT_LEN &&
1331 ret < MAX_SKB_FRAGS) ?
1332 PKT_PROT_LEN : txreq.size;
1333
1334 skb = alloc_skb(data_len + NET_SKB_PAD + NET_IP_ALIGN,
1335 GFP_ATOMIC | __GFP_NOWARN);
1336 if (unlikely(skb == NULL)) {
1337 netdev_dbg(vif->dev,
1338 "Can't allocate a skb in start_xmit.\n");
1339 netbk_tx_err(vif, &txreq, idx);
1340 break;
1341 }
1342
1343 /* Packets passed to netif_rx() must have some headroom. */
1344 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
1345
1346 if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) {
1347 struct xen_netif_extra_info *gso;
1348 gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1];
1349
1350 if (netbk_set_skb_gso(vif, skb, gso)) {
1351 kfree_skb(skb);
1352 netbk_tx_err(vif, &txreq, idx);
1353 continue;
1354 }
1355 }
1356
1357 /* XXX could copy straight to head */
1358 page = xen_netbk_alloc_page(netbk, skb, pending_idx);
1359 if (!page) {
1360 kfree_skb(skb);
1361 netbk_tx_err(vif, &txreq, idx);
1362 continue;
1363 }
1364
1365 gop->source.u.ref = txreq.gref;
1366 gop->source.domid = vif->domid;
1367 gop->source.offset = txreq.offset;
1368
1369 gop->dest.u.gmfn = virt_to_mfn(page_address(page));
1370 gop->dest.domid = DOMID_SELF;
1371 gop->dest.offset = txreq.offset;
1372
1373 gop->len = txreq.size;
1374 gop->flags = GNTCOPY_source_gref;
1375
1376 gop++;
1377
1378 memcpy(&netbk->pending_tx_info[pending_idx].req,
1379 &txreq, sizeof(txreq));
1380 netbk->pending_tx_info[pending_idx].vif = vif;
1381 *((u16 *)skb->data) = pending_idx;
1382
1383 __skb_put(skb, data_len);
1384
1385 skb_shinfo(skb)->nr_frags = ret;
1386 if (data_len < txreq.size) {
1387 skb_shinfo(skb)->nr_frags++;
1388 frag_set_pending_idx(&skb_shinfo(skb)->frags[0],
1389 pending_idx);
1390 } else {
1391 frag_set_pending_idx(&skb_shinfo(skb)->frags[0],
1392 INVALID_PENDING_IDX);
1393 }
1394
1395 netbk->pending_cons++;
1396
1397 request_gop = xen_netbk_get_requests(netbk, vif,
1398 skb, txfrags, gop);
1399 if (request_gop == NULL) {
1400 kfree_skb(skb);
1401 netbk_tx_err(vif, &txreq, idx);
1402 continue;
1403 }
1404 gop = request_gop;
1405
1406 __skb_queue_tail(&netbk->tx_queue, skb);
1407
1408 vif->tx.req_cons = idx;
1409 xen_netbk_check_rx_xenvif(vif);
1410
1411 if ((gop-netbk->tx_copy_ops) >= ARRAY_SIZE(netbk->tx_copy_ops))
1412 break;
1413 }
1414
1415 return gop - netbk->tx_copy_ops;
1416 }
1417
1418 static void xen_netbk_tx_submit(struct xen_netbk *netbk)
1419 {
1420 struct gnttab_copy *gop = netbk->tx_copy_ops;
1421 struct sk_buff *skb;
1422
1423 while ((skb = __skb_dequeue(&netbk->tx_queue)) != NULL) {
1424 struct xen_netif_tx_request *txp;
1425 struct xenvif *vif;
1426 u16 pending_idx;
1427 unsigned data_len;
1428
1429 pending_idx = *((u16 *)skb->data);
1430 vif = netbk->pending_tx_info[pending_idx].vif;
1431 txp = &netbk->pending_tx_info[pending_idx].req;
1432
1433 /* Check the remap error code. */
1434 if (unlikely(xen_netbk_tx_check_gop(netbk, skb, &gop))) {
1435 netdev_dbg(vif->dev, "netback grant failed.\n");
1436 skb_shinfo(skb)->nr_frags = 0;
1437 kfree_skb(skb);
1438 continue;
1439 }
1440
1441 data_len = skb->len;
1442 memcpy(skb->data,
1443 (void *)(idx_to_kaddr(netbk, pending_idx)|txp->offset),
1444 data_len);
1445 if (data_len < txp->size) {
1446 /* Append the packet payload as a fragment. */
1447 txp->offset += data_len;
1448 txp->size -= data_len;
1449 } else {
1450 /* Schedule a response immediately. */
1451 xen_netbk_idx_release(netbk, pending_idx);
1452 }
1453
1454 if (txp->flags & XEN_NETTXF_csum_blank)
1455 skb->ip_summed = CHECKSUM_PARTIAL;
1456 else if (txp->flags & XEN_NETTXF_data_validated)
1457 skb->ip_summed = CHECKSUM_UNNECESSARY;
1458
1459 xen_netbk_fill_frags(netbk, skb);
1460
1461 /*
1462 * If the initial fragment was < PKT_PROT_LEN then
1463 * pull through some bytes from the other fragments to
1464 * increase the linear region to PKT_PROT_LEN bytes.
1465 */
1466 if (skb_headlen(skb) < PKT_PROT_LEN && skb_is_nonlinear(skb)) {
1467 int target = min_t(int, skb->len, PKT_PROT_LEN);
1468 __pskb_pull_tail(skb, target - skb_headlen(skb));
1469 }
1470
1471 skb->dev = vif->dev;
1472 skb->protocol = eth_type_trans(skb, skb->dev);
1473
1474 if (checksum_setup(vif, skb)) {
1475 netdev_dbg(vif->dev,
1476 "Can't setup checksum in net_tx_action\n");
1477 kfree_skb(skb);
1478 continue;
1479 }
1480
1481 vif->dev->stats.rx_bytes += skb->len;
1482 vif->dev->stats.rx_packets++;
1483
1484 xenvif_receive_skb(vif, skb);
1485 }
1486 }
1487
1488 /* Called after netfront has transmitted */
1489 static void xen_netbk_tx_action(struct xen_netbk *netbk)
1490 {
1491 unsigned nr_gops;
1492
1493 nr_gops = xen_netbk_tx_build_gops(netbk);
1494
1495 if (nr_gops == 0)
1496 return;
1497
1498 gnttab_batch_copy(netbk->tx_copy_ops, nr_gops);
1499
1500 xen_netbk_tx_submit(netbk);
1501 }
1502
1503 static void xen_netbk_idx_release(struct xen_netbk *netbk, u16 pending_idx)
1504 {
1505 struct xenvif *vif;
1506 struct pending_tx_info *pending_tx_info;
1507 pending_ring_idx_t index;
1508
1509 /* Already complete? */
1510 if (netbk->mmap_pages[pending_idx] == NULL)
1511 return;
1512
1513 pending_tx_info = &netbk->pending_tx_info[pending_idx];
1514
1515 vif = pending_tx_info->vif;
1516
1517 make_tx_response(vif, &pending_tx_info->req, XEN_NETIF_RSP_OKAY);
1518
1519 index = pending_index(netbk->pending_prod++);
1520 netbk->pending_ring[index] = pending_idx;
1521
1522 xenvif_put(vif);
1523
1524 netbk->mmap_pages[pending_idx]->mapping = 0;
1525 put_page(netbk->mmap_pages[pending_idx]);
1526 netbk->mmap_pages[pending_idx] = NULL;
1527 }
1528
1529 static void make_tx_response(struct xenvif *vif,
1530 struct xen_netif_tx_request *txp,
1531 s8 st)
1532 {
1533 RING_IDX i = vif->tx.rsp_prod_pvt;
1534 struct xen_netif_tx_response *resp;
1535 int notify;
1536
1537 resp = RING_GET_RESPONSE(&vif->tx, i);
1538 resp->id = txp->id;
1539 resp->status = st;
1540
1541 if (txp->flags & XEN_NETTXF_extra_info)
1542 RING_GET_RESPONSE(&vif->tx, ++i)->status = XEN_NETIF_RSP_NULL;
1543
1544 vif->tx.rsp_prod_pvt = ++i;
1545 RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&vif->tx, notify);
1546 if (notify)
1547 notify_remote_via_irq(vif->irq);
1548 }
1549
1550 static struct xen_netif_rx_response *make_rx_response(struct xenvif *vif,
1551 u16 id,
1552 s8 st,
1553 u16 offset,
1554 u16 size,
1555 u16 flags)
1556 {
1557 RING_IDX i = vif->rx.rsp_prod_pvt;
1558 struct xen_netif_rx_response *resp;
1559
1560 resp = RING_GET_RESPONSE(&vif->rx, i);
1561 resp->offset = offset;
1562 resp->flags = flags;
1563 resp->id = id;
1564 resp->status = (s16)size;
1565 if (st < 0)
1566 resp->status = (s16)st;
1567
1568 vif->rx.rsp_prod_pvt = ++i;
1569
1570 return resp;
1571 }
1572
1573 static inline int rx_work_todo(struct xen_netbk *netbk)
1574 {
1575 return !skb_queue_empty(&netbk->rx_queue);
1576 }
1577
1578 static inline int tx_work_todo(struct xen_netbk *netbk)
1579 {
1580
1581 if (((nr_pending_reqs(netbk) + MAX_SKB_FRAGS) < MAX_PENDING_REQS) &&
1582 !list_empty(&netbk->net_schedule_list))
1583 return 1;
1584
1585 return 0;
1586 }
1587
1588 static int xen_netbk_kthread(void *data)
1589 {
1590 struct xen_netbk *netbk = data;
1591 while (!kthread_should_stop()) {
1592 wait_event_interruptible(netbk->wq,
1593 rx_work_todo(netbk) ||
1594 tx_work_todo(netbk) ||
1595 kthread_should_stop());
1596 cond_resched();
1597
1598 if (kthread_should_stop())
1599 break;
1600
1601 if (rx_work_todo(netbk))
1602 xen_netbk_rx_action(netbk);
1603
1604 if (tx_work_todo(netbk))
1605 xen_netbk_tx_action(netbk);
1606 }
1607
1608 return 0;
1609 }
1610
1611 void xen_netbk_unmap_frontend_rings(struct xenvif *vif)
1612 {
1613 if (vif->tx.sring)
1614 xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif),
1615 vif->tx.sring);
1616 if (vif->rx.sring)
1617 xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif),
1618 vif->rx.sring);
1619 }
1620
1621 int xen_netbk_map_frontend_rings(struct xenvif *vif,
1622 grant_ref_t tx_ring_ref,
1623 grant_ref_t rx_ring_ref)
1624 {
1625 void *addr;
1626 struct xen_netif_tx_sring *txs;
1627 struct xen_netif_rx_sring *rxs;
1628
1629 int err = -ENOMEM;
1630
1631 err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(vif),
1632 tx_ring_ref, &addr);
1633 if (err)
1634 goto err;
1635
1636 txs = (struct xen_netif_tx_sring *)addr;
1637 BACK_RING_INIT(&vif->tx, txs, PAGE_SIZE);
1638
1639 err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(vif),
1640 rx_ring_ref, &addr);
1641 if (err)
1642 goto err;
1643
1644 rxs = (struct xen_netif_rx_sring *)addr;
1645 BACK_RING_INIT(&vif->rx, rxs, PAGE_SIZE);
1646
1647 vif->rx_req_cons_peek = 0;
1648
1649 return 0;
1650
1651 err:
1652 xen_netbk_unmap_frontend_rings(vif);
1653 return err;
1654 }
1655
1656 static int __init netback_init(void)
1657 {
1658 int i;
1659 int rc = 0;
1660 int group;
1661
1662 if (!xen_domain())
1663 return -ENODEV;
1664
1665 xen_netbk_group_nr = num_online_cpus();
1666 xen_netbk = vzalloc(sizeof(struct xen_netbk) * xen_netbk_group_nr);
1667 if (!xen_netbk)
1668 return -ENOMEM;
1669
1670 for (group = 0; group < xen_netbk_group_nr; group++) {
1671 struct xen_netbk *netbk = &xen_netbk[group];
1672 skb_queue_head_init(&netbk->rx_queue);
1673 skb_queue_head_init(&netbk->tx_queue);
1674
1675 init_timer(&netbk->net_timer);
1676 netbk->net_timer.data = (unsigned long)netbk;
1677 netbk->net_timer.function = xen_netbk_alarm;
1678
1679 netbk->pending_cons = 0;
1680 netbk->pending_prod = MAX_PENDING_REQS;
1681 for (i = 0; i < MAX_PENDING_REQS; i++)
1682 netbk->pending_ring[i] = i;
1683
1684 init_waitqueue_head(&netbk->wq);
1685 netbk->task = kthread_create(xen_netbk_kthread,
1686 (void *)netbk,
1687 "netback/%u", group);
1688
1689 if (IS_ERR(netbk->task)) {
1690 printk(KERN_ALERT "kthread_create() fails at netback\n");
1691 del_timer(&netbk->net_timer);
1692 rc = PTR_ERR(netbk->task);
1693 goto failed_init;
1694 }
1695
1696 kthread_bind(netbk->task, group);
1697
1698 INIT_LIST_HEAD(&netbk->net_schedule_list);
1699
1700 spin_lock_init(&netbk->net_schedule_list_lock);
1701
1702 atomic_set(&netbk->netfront_count, 0);
1703
1704 wake_up_process(netbk->task);
1705 }
1706
1707 rc = xenvif_xenbus_init();
1708 if (rc)
1709 goto failed_init;
1710
1711 return 0;
1712
1713 failed_init:
1714 while (--group >= 0) {
1715 struct xen_netbk *netbk = &xen_netbk[group];
1716 for (i = 0; i < MAX_PENDING_REQS; i++) {
1717 if (netbk->mmap_pages[i])
1718 __free_page(netbk->mmap_pages[i]);
1719 }
1720 del_timer(&netbk->net_timer);
1721 kthread_stop(netbk->task);
1722 }
1723 vfree(xen_netbk);
1724 return rc;
1725
1726 }
1727
1728 module_init(netback_init);
1729
1730 MODULE_LICENSE("Dual BSD/GPL");
1731 MODULE_ALIAS("xen-backend:vif");
Linux kernel - Deliverables
This page took 0.070088 seconds and 5 git commands to generate.