projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge branch 'master' of git://1984.lsi.us.es/nf
[linux-2.6-block.git] / drivers / net / xen-netback / netback.c
CommitLineData
f942dc25
IC		 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
ca981633 43#include <xen/xen.h>
f942dc25
IC		 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
50struct pending_tx_info {
51 struct xen_netif_tx_request req;
52 struct xenvif *vif;
53};
54typedef unsigned int pending_ring_idx_t;
55
56struct netbk_rx_meta {
57 int id;
58 int size;
59 int gso_size;
60};
61
62#define MAX_PENDING_REQS 256
63
ea066ad1
IC		 64/* Discriminate from any valid pending_idx value. */
65#define INVALID_PENDING_IDX 0xFFFF
66
f942dc25
IC		 67#define MAX_BUFFER_OFFSET PAGE_SIZE
68
69/* extra field used in struct page */
70union 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
84struct 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
118static struct xen_netbk *xen_netbk;
119static int xen_netbk_group_nr;
120
121void 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
143void 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
7d5145d8
MD		 150static void xen_netbk_idx_release(struct xen_netbk *netbk, u16 pending_idx,
151 u8 status);
f942dc25
IC		 152static void make_tx_response(struct xenvif *vif,
153 struct xen_netif_tx_request *txp,
154 s8 st);
155static struct xen_netif_rx_response *make_rx_response(struct xenvif *vif,
156 u16 id,
157 s8 st,
158 u16 offset,
159 u16 size,
160 u16 flags);
161
162static inline unsigned long idx_to_pfn(struct xen_netbk *netbk,
ea066ad1 163 u16 idx)
f942dc25
IC		 164{
165 return page_to_pfn(netbk->mmap_pages[idx]);
166}
167
168static inline unsigned long idx_to_kaddr(struct xen_netbk *netbk,
ea066ad1 169 u16 idx)
f942dc25
IC		 170{
171 return (unsigned long)pfn_to_kaddr(idx_to_pfn(netbk, idx));
172}
173
174/* extra field used in struct page */
175static inline void set_page_ext(struct page *pg, struct xen_netbk *netbk,
176 unsigned int idx)
177{
178 unsigned int group = netbk - xen_netbk;
179 union page_ext ext = { .e = { .group = group + 1, .idx = idx } };
180
181 BUILD_BUG_ON(sizeof(ext) > sizeof(ext.mapping));
182 pg->mapping = ext.mapping;
183}
184
185static int get_page_ext(struct page *pg,
186 unsigned int *pgroup, unsigned int *pidx)
187{
188 union page_ext ext = { .mapping = pg->mapping };
189 struct xen_netbk *netbk;
190 unsigned int group, idx;
191
192 group = ext.e.group - 1;
193
194 if (group < 0 || group >= xen_netbk_group_nr)
195 return 0;
196
197 netbk = &xen_netbk[group];
198
199 idx = ext.e.idx;
200
201 if ((idx < 0) || (idx >= MAX_PENDING_REQS))
202 return 0;
203
204 if (netbk->mmap_pages[idx] != pg)
205 return 0;
206
207 *pgroup = group;
208 *pidx = idx;
209
210 return 1;
211}
212
213/*
214 * This is the amount of packet we copy rather than map, so that the
215 * guest can't fiddle with the contents of the headers while we do
216 * packet processing on them (netfilter, routing, etc).
217 */
218#define PKT_PROT_LEN (ETH_HLEN + \
219 VLAN_HLEN + \
220 sizeof(struct iphdr) + MAX_IPOPTLEN + \
221 sizeof(struct tcphdr) + MAX_TCP_OPTION_SPACE)
222
ea066ad1
IC		 223static u16 frag_get_pending_idx(skb_frag_t *frag)
224{
225 return (u16)frag->page_offset;
226}
227
228static void frag_set_pending_idx(skb_frag_t *frag, u16 pending_idx)
229{
230 frag->page_offset = pending_idx;
231}
232
f942dc25
IC		 233static inline pending_ring_idx_t pending_index(unsigned i)
234{
235 return i & (MAX_PENDING_REQS-1);
236}
237
238static inline pending_ring_idx_t nr_pending_reqs(struct xen_netbk *netbk)
239{
240 return MAX_PENDING_REQS -
241 netbk->pending_prod + netbk->pending_cons;
242}
243
244static void xen_netbk_kick_thread(struct xen_netbk *netbk)
245{
246 wake_up(&netbk->wq);
247}
248
249static int max_required_rx_slots(struct xenvif *vif)
250{
251 int max = DIV_ROUND_UP(vif->dev->mtu, PAGE_SIZE);
252
253 if (vif->can_sg || vif->gso || vif->gso_prefix)
254 max += MAX_SKB_FRAGS + 1; /* extra_info + frags */
255
256 return max;
257}
258
259int xen_netbk_rx_ring_full(struct xenvif *vif)
260{
261 RING_IDX peek = vif->rx_req_cons_peek;
262 RING_IDX needed = max_required_rx_slots(vif);
263
264 return ((vif->rx.sring->req_prod - peek) < needed) ||
265 ((vif->rx.rsp_prod_pvt + XEN_NETIF_RX_RING_SIZE - peek) < needed);
266}
267
268int xen_netbk_must_stop_queue(struct xenvif *vif)
269{
270 if (!xen_netbk_rx_ring_full(vif))
271 return 0;
272
273 vif->rx.sring->req_event = vif->rx_req_cons_peek +
274 max_required_rx_slots(vif);
275 mb(); /* request notification /then/ check the queue */
276
277 return xen_netbk_rx_ring_full(vif);
278}
279
280/*
281 * Returns true if we should start a new receive buffer instead of
282 * adding 'size' bytes to a buffer which currently contains 'offset'
283 * bytes.
284 */
285static bool start_new_rx_buffer(int offset, unsigned long size, int head)
286{
287 /* simple case: we have completely filled the current buffer. */
288 if (offset == MAX_BUFFER_OFFSET)
289 return true;
290
291 /*
292 * complex case: start a fresh buffer if the current frag
293 * would overflow the current buffer but only if:
294 * (i) this frag would fit completely in the next buffer
295 * and (ii) there is already some data in the current buffer
296 * and (iii) this is not the head buffer.
297 *
298 * Where:
299 * - (i) stops us splitting a frag into two copies
300 * unless the frag is too large for a single buffer.
301 * - (ii) stops us from leaving a buffer pointlessly empty.
302 * - (iii) stops us leaving the first buffer
303 * empty. Strictly speaking this is already covered
304 * by (ii) but is explicitly checked because
305 * netfront relies on the first buffer being
306 * non-empty and can crash otherwise.
307 *
308 * This means we will effectively linearise small
309 * frags but do not needlessly split large buffers
310 * into multiple copies tend to give large frags their
311 * own buffers as before.
312 */
313 if ((offset + size > MAX_BUFFER_OFFSET) &&
314 (size <= MAX_BUFFER_OFFSET) && offset && !head)
315 return true;
316
317 return false;
318}
319
320/*
321 * Figure out how many ring slots we're going to need to send @skb to
322 * the guest. This function is essentially a dry run of
323 * netbk_gop_frag_copy.
324 */
325unsigned int xen_netbk_count_skb_slots(struct xenvif *vif, struct sk_buff *skb)
326{
327 unsigned int count;
328 int i, copy_off;
329
e26b203e 330 count = DIV_ROUND_UP(skb_headlen(skb), PAGE_SIZE);
f942dc25
IC		 331
332 copy_off = skb_headlen(skb) % PAGE_SIZE;
333
334 if (skb_shinfo(skb)->gso_size)
335 count++;
336
337 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
9e903e08 338 unsigned long size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
6a8ed462 339 unsigned long offset = skb_shinfo(skb)->frags[i].page_offset;
f942dc25 340 unsigned long bytes;
6a8ed462
IC		 341
342 offset &= ~PAGE_MASK;
343
f942dc25 344 while (size > 0) {
6a8ed462 345 BUG_ON(offset >= PAGE_SIZE);
f942dc25
IC		 346 BUG_ON(copy_off > MAX_BUFFER_OFFSET);
347
6a8ed462
IC		 348 bytes = PAGE_SIZE - offset;
349
350 if (bytes > size)
351 bytes = size;
352
353 if (start_new_rx_buffer(copy_off, bytes, 0)) {
f942dc25
IC		 354 count++;
355 copy_off = 0;
356 }
357
f942dc25
IC		 358 if (copy_off + bytes > MAX_BUFFER_OFFSET)
359 bytes = MAX_BUFFER_OFFSET - copy_off;
360
361 copy_off += bytes;
6a8ed462
IC		 362
363 offset += bytes;
f942dc25 364 size -= bytes;
6a8ed462
IC		 365
366 if (offset == PAGE_SIZE)
367 offset = 0;
f942dc25
IC		 368 }
369 }
370 return count;
371}
372
373struct netrx_pending_operations {
374 unsigned copy_prod, copy_cons;
375 unsigned meta_prod, meta_cons;
376 struct gnttab_copy *copy;
377 struct netbk_rx_meta *meta;
378 int copy_off;
379 grant_ref_t copy_gref;
380};
381
382static struct netbk_rx_meta *get_next_rx_buffer(struct xenvif *vif,
383 struct netrx_pending_operations *npo)
384{
385 struct netbk_rx_meta *meta;
386 struct xen_netif_rx_request *req;
387
388 req = RING_GET_REQUEST(&vif->rx, vif->rx.req_cons++);
389
390 meta = npo->meta + npo->meta_prod++;
391 meta->gso_size = 0;
392 meta->size = 0;
393 meta->id = req->id;
394
395 npo->copy_off = 0;
396 npo->copy_gref = req->gref;
397
398 return meta;
399}
400
401/*
402 * Set up the grant operations for this fragment. If it's a flipping
403 * interface, we also set up the unmap request from here.
404 */
405static void netbk_gop_frag_copy(struct xenvif *vif, struct sk_buff *skb,
406 struct netrx_pending_operations *npo,
407 struct page *page, unsigned long size,
408 unsigned long offset, int *head)
409{
410 struct gnttab_copy *copy_gop;
411 struct netbk_rx_meta *meta;
412 /*
e34c0246 413 * These variables are used iff get_page_ext returns true,
f942dc25
IC		 414 * in which case they are guaranteed to be initialized.
415 */
416 unsigned int uninitialized_var(group), uninitialized_var(idx);
417 int foreign = get_page_ext(page, &group, &idx);
418 unsigned long bytes;
419
420 /* Data must not cross a page boundary. */
6a8ed462 421 BUG_ON(size + offset > PAGE_SIZE<<compound_order(page));
f942dc25
IC		 422
423 meta = npo->meta + npo->meta_prod - 1;
424
6a8ed462
IC		 425 /* Skip unused frames from start of page */
426 page += offset >> PAGE_SHIFT;
427 offset &= ~PAGE_MASK;
428
f942dc25 429 while (size > 0) {
6a8ed462 430 BUG_ON(offset >= PAGE_SIZE);
f942dc25
IC		 431 BUG_ON(npo->copy_off > MAX_BUFFER_OFFSET);
432
6a8ed462
IC		 433 bytes = PAGE_SIZE - offset;
434
435 if (bytes > size)
436 bytes = size;
437
438 if (start_new_rx_buffer(npo->copy_off, bytes, *head)) {
f942dc25
IC		 439 /*
440 * Netfront requires there to be some data in the head
441 * buffer.
442 */
443 BUG_ON(*head);
444
445 meta = get_next_rx_buffer(vif, npo);
446 }
447
f942dc25
IC		 448 if (npo->copy_off + bytes > MAX_BUFFER_OFFSET)
449 bytes = MAX_BUFFER_OFFSET - npo->copy_off;
450
451 copy_gop = npo->copy + npo->copy_prod++;
452 copy_gop->flags = GNTCOPY_dest_gref;
453 if (foreign) {
454 struct xen_netbk *netbk = &xen_netbk[group];
455 struct pending_tx_info *src_pend;
456
457 src_pend = &netbk->pending_tx_info[idx];
458
459 copy_gop->source.domid = src_pend->vif->domid;
460 copy_gop->source.u.ref = src_pend->req.gref;
461 copy_gop->flags |= GNTCOPY_source_gref;
462 } else {
463 void *vaddr = page_address(page);
464 copy_gop->source.domid = DOMID_SELF;
465 copy_gop->source.u.gmfn = virt_to_mfn(vaddr);
466 }
467 copy_gop->source.offset = offset;
468 copy_gop->dest.domid = vif->domid;
469
470 copy_gop->dest.offset = npo->copy_off;
471 copy_gop->dest.u.ref = npo->copy_gref;
472 copy_gop->len = bytes;
473
474 npo->copy_off += bytes;
475 meta->size += bytes;
476
477 offset += bytes;
478 size -= bytes;
479
6a8ed462
IC		 480 /* Next frame */
481 if (offset == PAGE_SIZE && size) {
482 BUG_ON(!PageCompound(page));
483 page++;
484 offset = 0;
485 }
486
f942dc25
IC		 487 /* Leave a gap for the GSO descriptor. */
488 if (*head && skb_shinfo(skb)->gso_size && !vif->gso_prefix)
489 vif->rx.req_cons++;
490
491 *head = 0; /* There must be something in this buffer now. */
492
493 }
494}
495
496/*
497 * Prepare an SKB to be transmitted to the frontend.
498 *
499 * This function is responsible for allocating grant operations, meta
500 * structures, etc.
501 *
502 * It returns the number of meta structures consumed. The number of
503 * ring slots used is always equal to the number of meta slots used
504 * plus the number of GSO descriptors used. Currently, we use either
505 * zero GSO descriptors (for non-GSO packets) or one descriptor (for
506 * frontend-side LRO).
507 */
508static int netbk_gop_skb(struct sk_buff *skb,
509 struct netrx_pending_operations *npo)
510{
511 struct xenvif *vif = netdev_priv(skb->dev);
512 int nr_frags = skb_shinfo(skb)->nr_frags;
513 int i;
514 struct xen_netif_rx_request *req;
515 struct netbk_rx_meta *meta;
516 unsigned char *data;
517 int head = 1;
518 int old_meta_prod;
519
520 old_meta_prod = npo->meta_prod;
521
522 /* Set up a GSO prefix descriptor, if necessary */
523 if (skb_shinfo(skb)->gso_size && vif->gso_prefix) {
524 req = RING_GET_REQUEST(&vif->rx, vif->rx.req_cons++);
525 meta = npo->meta + npo->meta_prod++;
526 meta->gso_size = skb_shinfo(skb)->gso_size;
527 meta->size = 0;
528 meta->id = req->id;
529 }
530
531 req = RING_GET_REQUEST(&vif->rx, vif->rx.req_cons++);
532 meta = npo->meta + npo->meta_prod++;
533
534 if (!vif->gso_prefix)
535 meta->gso_size = skb_shinfo(skb)->gso_size;
536 else
537 meta->gso_size = 0;
538
539 meta->size = 0;
540 meta->id = req->id;
541 npo->copy_off = 0;
542 npo->copy_gref = req->gref;
543
544 data = skb->data;
545 while (data < skb_tail_pointer(skb)) {
546 unsigned int offset = offset_in_page(data);
547 unsigned int len = PAGE_SIZE - offset;
548
549 if (data + len > skb_tail_pointer(skb))
550 len = skb_tail_pointer(skb) - data;
551
552 netbk_gop_frag_copy(vif, skb, npo,
553 virt_to_page(data), len, offset, &head);
554 data += len;
555 }
556
557 for (i = 0; i < nr_frags; i++) {
558 netbk_gop_frag_copy(vif, skb, npo,
ea066ad1 559 skb_frag_page(&skb_shinfo(skb)->frags[i]),
9e903e08 560 skb_frag_size(&skb_shinfo(skb)->frags[i]),
f942dc25
IC		 561 skb_shinfo(skb)->frags[i].page_offset,
562 &head);
563 }
564
565 return npo->meta_prod - old_meta_prod;
566}
567
568/*
569 * This is a twin to netbk_gop_skb. Assume that netbk_gop_skb was
570 * used to set up the operations on the top of
571 * netrx_pending_operations, which have since been done. Check that
572 * they didn't give any errors and advance over them.
573 */
574static int netbk_check_gop(struct xenvif *vif, int nr_meta_slots,
575 struct netrx_pending_operations *npo)
576{
577 struct gnttab_copy *copy_op;
578 int status = XEN_NETIF_RSP_OKAY;
579 int i;
580
581 for (i = 0; i < nr_meta_slots; i++) {
582 copy_op = npo->copy + npo->copy_cons++;
583 if (copy_op->status != GNTST_okay) {
584 netdev_dbg(vif->dev,
585 "Bad status %d from copy to DOM%d.\n",
586 copy_op->status, vif->domid);
587 status = XEN_NETIF_RSP_ERROR;
588 }
589 }
590
591 return status;
592}
593
594static void netbk_add_frag_responses(struct xenvif *vif, int status,
595 struct netbk_rx_meta *meta,
596 int nr_meta_slots)
597{
598 int i;
599 unsigned long offset;
600
601 /* No fragments used */
602 if (nr_meta_slots <= 1)
603 return;
604
605 nr_meta_slots--;
606
607 for (i = 0; i < nr_meta_slots; i++) {
608 int flags;
609 if (i == nr_meta_slots - 1)
610 flags = 0;
611 else
612 flags = XEN_NETRXF_more_data;
613
614 offset = 0;
615 make_rx_response(vif, meta[i].id, status, offset,
616 meta[i].size, flags);
617 }
618}
619
620struct skb_cb_overlay {
621 int meta_slots_used;
622};
623
624static void xen_netbk_rx_action(struct xen_netbk *netbk)
625{
626 struct xenvif *vif = NULL, *tmp;
627 s8 status;
628 u16 irq, flags;
629 struct xen_netif_rx_response *resp;
630 struct sk_buff_head rxq;
631 struct sk_buff *skb;
632 LIST_HEAD(notify);
633 int ret;
634 int nr_frags;
635 int count;
636 unsigned long offset;
637 struct skb_cb_overlay *sco;
638
639 struct netrx_pending_operations npo = {
640 .copy = netbk->grant_copy_op,
641 .meta = netbk->meta,
642 };
643
644 skb_queue_head_init(&rxq);
645
646 count = 0;
647
648 while ((skb = skb_dequeue(&netbk->rx_queue)) != NULL) {
649 vif = netdev_priv(skb->dev);
650 nr_frags = skb_shinfo(skb)->nr_frags;
651
652 sco = (struct skb_cb_overlay *)skb->cb;
653 sco->meta_slots_used = netbk_gop_skb(skb, &npo);
654
655 count += nr_frags + 1;
656
657 __skb_queue_tail(&rxq, skb);
658
659 /* Filled the batch queue? */
660 if (count + MAX_SKB_FRAGS >= XEN_NETIF_RX_RING_SIZE)
661 break;
662 }
663
664 BUG_ON(npo.meta_prod > ARRAY_SIZE(netbk->meta));
665
666 if (!npo.copy_prod)
667 return;
668
669 BUG_ON(npo.copy_prod > ARRAY_SIZE(netbk->grant_copy_op));
c571898f 670 gnttab_batch_copy(netbk->grant_copy_op, npo.copy_prod);
f942dc25
IC		 671
672 while ((skb = __skb_dequeue(&rxq)) != NULL) {
673 sco = (struct skb_cb_overlay *)skb->cb;
674
675 vif = netdev_priv(skb->dev);
676
677 if (netbk->meta[npo.meta_cons].gso_size && vif->gso_prefix) {
678 resp = RING_GET_RESPONSE(&vif->rx,
679 vif->rx.rsp_prod_pvt++);
680
681 resp->flags = XEN_NETRXF_gso_prefix | XEN_NETRXF_more_data;
682
683 resp->offset = netbk->meta[npo.meta_cons].gso_size;
684 resp->id = netbk->meta[npo.meta_cons].id;
685 resp->status = sco->meta_slots_used;
686
687 npo.meta_cons++;
688 sco->meta_slots_used--;
689 }
690
691
692 vif->dev->stats.tx_bytes += skb->len;
693 vif->dev->stats.tx_packets++;
694
695 status = netbk_check_gop(vif, sco->meta_slots_used, &npo);
696
697 if (sco->meta_slots_used == 1)
698 flags = 0;
699 else
700 flags = XEN_NETRXF_more_data;
701
702 if (skb->ip_summed == CHECKSUM_PARTIAL) /* local packet? */
703 flags |= XEN_NETRXF_csum_blank | XEN_NETRXF_data_validated;
704 else if (skb->ip_summed == CHECKSUM_UNNECESSARY)
705 /* remote but checksummed. */
706 flags |= XEN_NETRXF_data_validated;
707
708 offset = 0;
709 resp = make_rx_response(vif, netbk->meta[npo.meta_cons].id,
710 status, offset,
711 netbk->meta[npo.meta_cons].size,
712 flags);
713
714 if (netbk->meta[npo.meta_cons].gso_size && !vif->gso_prefix) {
715 struct xen_netif_extra_info *gso =
716 (struct xen_netif_extra_info *)
717 RING_GET_RESPONSE(&vif->rx,
718 vif->rx.rsp_prod_pvt++);
719
720 resp->flags |= XEN_NETRXF_extra_info;
721
722 gso->u.gso.size = netbk->meta[npo.meta_cons].gso_size;
723 gso->u.gso.type = XEN_NETIF_GSO_TYPE_TCPV4;
724 gso->u.gso.pad = 0;
725 gso->u.gso.features = 0;
726
727 gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
728 gso->flags = 0;
729 }
730
731 netbk_add_frag_responses(vif, status,
732 netbk->meta + npo.meta_cons + 1,
733 sco->meta_slots_used);
734
735 RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&vif->rx, ret);
736 irq = vif->irq;
737 if (ret && list_empty(&vif->notify_list))
738 list_add_tail(&vif->notify_list, &notify);
739
740 xenvif_notify_tx_completion(vif);
741
742 xenvif_put(vif);
743 npo.meta_cons += sco->meta_slots_used;
744 dev_kfree_skb(skb);
745 }
746
747 list_for_each_entry_safe(vif, tmp, &notify, notify_list) {
748 notify_remote_via_irq(vif->irq);
749 list_del_init(&vif->notify_list);
750 }
751
752 /* More work to do? */
753 if (!skb_queue_empty(&netbk->rx_queue) &&
754 !timer_pending(&netbk->net_timer))
755 xen_netbk_kick_thread(netbk);
756}
757
758void xen_netbk_queue_tx_skb(struct xenvif *vif, struct sk_buff *skb)
759{
760 struct xen_netbk *netbk = vif->netbk;
761
762 skb_queue_tail(&netbk->rx_queue, skb);
763
764 xen_netbk_kick_thread(netbk);
765}
766
767static void xen_netbk_alarm(unsigned long data)
768{
769 struct xen_netbk *netbk = (struct xen_netbk *)data;
770 xen_netbk_kick_thread(netbk);
771}
772
773static int __on_net_schedule_list(struct xenvif *vif)
774{
775 return !list_empty(&vif->schedule_list);
776}
777
778/* Must be called with net_schedule_list_lock held */
779static void remove_from_net_schedule_list(struct xenvif *vif)
780{
781 if (likely(__on_net_schedule_list(vif))) {
782 list_del_init(&vif->schedule_list);
783 xenvif_put(vif);
784 }
785}
786
787static struct xenvif *poll_net_schedule_list(struct xen_netbk *netbk)
788{
789 struct xenvif *vif = NULL;
790
791 spin_lock_irq(&netbk->net_schedule_list_lock);
792 if (list_empty(&netbk->net_schedule_list))
793 goto out;
794
795 vif = list_first_entry(&netbk->net_schedule_list,
796 struct xenvif, schedule_list);
797 if (!vif)
798 goto out;
799
800 xenvif_get(vif);
801
802 remove_from_net_schedule_list(vif);
803out:
804 spin_unlock_irq(&netbk->net_schedule_list_lock);
805 return vif;
806}
807
808void xen_netbk_schedule_xenvif(struct xenvif *vif)
809{
810 unsigned long flags;
811 struct xen_netbk *netbk = vif->netbk;
812
813 if (__on_net_schedule_list(vif))
814 goto kick;
815
816 spin_lock_irqsave(&netbk->net_schedule_list_lock, flags);
817 if (!__on_net_schedule_list(vif) &&
818 likely(xenvif_schedulable(vif))) {
819 list_add_tail(&vif->schedule_list, &netbk->net_schedule_list);
820 xenvif_get(vif);
821 }
822 spin_unlock_irqrestore(&netbk->net_schedule_list_lock, flags);
823
824kick:
825 smp_mb();
826 if ((nr_pending_reqs(netbk) < (MAX_PENDING_REQS/2)) &&
827 !list_empty(&netbk->net_schedule_list))
828 xen_netbk_kick_thread(netbk);
829}
830
831void xen_netbk_deschedule_xenvif(struct xenvif *vif)
832{
833 struct xen_netbk *netbk = vif->netbk;
834 spin_lock_irq(&netbk->net_schedule_list_lock);
835 remove_from_net_schedule_list(vif);
836 spin_unlock_irq(&netbk->net_schedule_list_lock);
837}
838
839void xen_netbk_check_rx_xenvif(struct xenvif *vif)
840{
841 int more_to_do;
842
843 RING_FINAL_CHECK_FOR_REQUESTS(&vif->tx, more_to_do);
844
845 if (more_to_do)
846 xen_netbk_schedule_xenvif(vif);
847}
848
849static void tx_add_credit(struct xenvif *vif)
850{
851 unsigned long max_burst, max_credit;
852
853 /*
854 * Allow a burst big enough to transmit a jumbo packet of up to 128kB.
855 * Otherwise the interface can seize up due to insufficient credit.
856 */
857 max_burst = RING_GET_REQUEST(&vif->tx, vif->tx.req_cons)->size;
858 max_burst = min(max_burst, 131072UL);
859 max_burst = max(max_burst, vif->credit_bytes);
860
861 /* Take care that adding a new chunk of credit doesn't wrap to zero. */
862 max_credit = vif->remaining_credit + vif->credit_bytes;
863 if (max_credit < vif->remaining_credit)
864 max_credit = ULONG_MAX; /* wrapped: clamp to ULONG_MAX */
865
866 vif->remaining_credit = min(max_credit, max_burst);
867}
868
869static void tx_credit_callback(unsigned long data)
870{
871 struct xenvif *vif = (struct xenvif *)data;
872 tx_add_credit(vif);
873 xen_netbk_check_rx_xenvif(vif);
874}
875
876static void netbk_tx_err(struct xenvif *vif,
877 struct xen_netif_tx_request *txp, RING_IDX end)
878{
879 RING_IDX cons = vif->tx.req_cons;
880
881 do {
882 make_tx_response(vif, txp, XEN_NETIF_RSP_ERROR);
b9149729 883 if (cons == end)
f942dc25
IC		 884 break;
885 txp = RING_GET_REQUEST(&vif->tx, cons++);
886 } while (1);
887 vif->tx.req_cons = cons;
888 xen_netbk_check_rx_xenvif(vif);
889 xenvif_put(vif);
890}
891
48856286
IC		 892static void netbk_fatal_tx_err(struct xenvif *vif)
893{
894 netdev_err(vif->dev, "fatal error; disabling device\n");
895 xenvif_carrier_off(vif);
896 xenvif_put(vif);
897}
898
f942dc25
IC		 899static int netbk_count_requests(struct xenvif *vif,
900 struct xen_netif_tx_request *first,
901 struct xen_netif_tx_request *txp,
902 int work_to_do)
903{
904 RING_IDX cons = vif->tx.req_cons;
905 int frags = 0;
906
907 if (!(first->flags & XEN_NETTXF_more_data))
908 return 0;
909
910 do {
911 if (frags >= work_to_do) {
48856286
IC		 912 netdev_err(vif->dev, "Need more frags\n");
913 netbk_fatal_tx_err(vif);
f942dc25
IC		 914 return -frags;
915 }
916
917 if (unlikely(frags >= MAX_SKB_FRAGS)) {
48856286
IC		 918 netdev_err(vif->dev, "Too many frags\n");
919 netbk_fatal_tx_err(vif);
f942dc25
IC		 920 return -frags;
921 }
922
923 memcpy(txp, RING_GET_REQUEST(&vif->tx, cons + frags),
924 sizeof(*txp));
925 if (txp->size > first->size) {
48856286
IC		 926 netdev_err(vif->dev, "Frag is bigger than frame.\n");
927 netbk_fatal_tx_err(vif);
f942dc25
IC		 928 return -frags;
929 }
930
931 first->size -= txp->size;
932 frags++;
933
934 if (unlikely((txp->offset + txp->size) > PAGE_SIZE)) {
48856286 935 netdev_err(vif->dev, "txp->offset: %x, size: %u\n",
f942dc25 936 txp->offset, txp->size);
48856286 937 netbk_fatal_tx_err(vif);
f942dc25
IC		 938 return -frags;
939 }
940 } while ((txp++)->flags & XEN_NETTXF_more_data);
941 return frags;
942}
943
944static struct page *xen_netbk_alloc_page(struct xen_netbk *netbk,
945 struct sk_buff *skb,
ea066ad1 946 u16 pending_idx)
f942dc25
IC		 947{
948 struct page *page;
949 page = alloc_page(GFP_KERNEL|__GFP_COLD);
950 if (!page)
951 return NULL;
952 set_page_ext(page, netbk, pending_idx);
953 netbk->mmap_pages[pending_idx] = page;
954 return page;
955}
956
957static struct gnttab_copy *xen_netbk_get_requests(struct xen_netbk *netbk,
958 struct xenvif *vif,
959 struct sk_buff *skb,
960 struct xen_netif_tx_request *txp,
961 struct gnttab_copy *gop)
962{
963 struct skb_shared_info *shinfo = skb_shinfo(skb);
964 skb_frag_t *frags = shinfo->frags;
ea066ad1 965 u16 pending_idx = *((u16 *)skb->data);
f942dc25
IC		 966 int i, start;
967
968 /* Skip first skb fragment if it is on same page as header fragment. */
ea066ad1 969 start = (frag_get_pending_idx(&shinfo->frags[0]) == pending_idx);
f942dc25
IC		 970
971 for (i = start; i < shinfo->nr_frags; i++, txp++) {
972 struct page *page;
973 pending_ring_idx_t index;
974 struct pending_tx_info *pending_tx_info =
975 netbk->pending_tx_info;
976
977 index = pending_index(netbk->pending_cons++);
978 pending_idx = netbk->pending_ring[index];
979 page = xen_netbk_alloc_page(netbk, skb, pending_idx);
980 if (!page)
4cc7c1cb 981 goto err;
f942dc25 982
f942dc25
IC		 983 gop->source.u.ref = txp->gref;
984 gop->source.domid = vif->domid;
985 gop->source.offset = txp->offset;
986
987 gop->dest.u.gmfn = virt_to_mfn(page_address(page));
988 gop->dest.domid = DOMID_SELF;
989 gop->dest.offset = txp->offset;
990
991 gop->len = txp->size;
992 gop->flags = GNTCOPY_source_gref;
993
994 gop++;
995
996 memcpy(&pending_tx_info[pending_idx].req, txp, sizeof(*txp));
997 xenvif_get(vif);
998 pending_tx_info[pending_idx].vif = vif;
ea066ad1 999 frag_set_pending_idx(&frags[i], pending_idx);
f942dc25
IC		 1000 }
1001
1002 return gop;
4cc7c1cb
IC		 1003err:
1004 /* Unwind, freeing all pages and sending error responses. */
1005 while (i-- > start) {
1006 xen_netbk_idx_release(netbk, frag_get_pending_idx(&frags[i]),
1007 XEN_NETIF_RSP_ERROR);
1008 }
1009 /* The head too, if necessary. */
1010 if (start)
1011 xen_netbk_idx_release(netbk, pending_idx, XEN_NETIF_RSP_ERROR);
1012
1013 return NULL;
f942dc25
IC		 1014}
1015
1016static int xen_netbk_tx_check_gop(struct xen_netbk *netbk,
1017 struct sk_buff *skb,
1018 struct gnttab_copy **gopp)
1019{
1020 struct gnttab_copy *gop = *gopp;
ea066ad1 1021 u16 pending_idx = *((u16 *)skb->data);
f942dc25
IC		 1022 struct skb_shared_info *shinfo = skb_shinfo(skb);
1023 int nr_frags = shinfo->nr_frags;
1024 int i, err, start;
1025
1026 /* Check status of header. */
1027 err = gop->status;
7d5145d8
MD		 1028 if (unlikely(err))
1029 xen_netbk_idx_release(netbk, pending_idx, XEN_NETIF_RSP_ERROR);
f942dc25
IC		 1030
1031 /* Skip first skb fragment if it is on same page as header fragment. */
ea066ad1 1032 start = (frag_get_pending_idx(&shinfo->frags[0]) == pending_idx);
f942dc25
IC		 1033
1034 for (i = start; i < nr_frags; i++) {
1035 int j, newerr;
f942dc25 1036
ea066ad1 1037 pending_idx = frag_get_pending_idx(&shinfo->frags[i]);
f942dc25
IC		 1038
1039 /* Check error status: if okay then remember grant handle. */
1040 newerr = (++gop)->status;
1041 if (likely(!newerr)) {
1042 /* Had a previous error? Invalidate this fragment. */
1043 if (unlikely(err))
7d5145d8 1044 xen_netbk_idx_release(netbk, pending_idx, XEN_NETIF_RSP_OKAY);
f942dc25
IC		 1045 continue;
1046 }
1047
1048 /* Error on this fragment: respond to client with an error. */
7d5145d8 1049 xen_netbk_idx_release(netbk, pending_idx, XEN_NETIF_RSP_ERROR);
f942dc25
IC		 1050
1051 /* Not the first error? Preceding frags already invalidated. */
1052 if (err)
1053 continue;
1054
1055 /* First error: invalidate header and preceding fragments. */
1056 pending_idx = *((u16 *)skb->data);
7d5145d8 1057 xen_netbk_idx_release(netbk, pending_idx, XEN_NETIF_RSP_OKAY);
f942dc25 1058 for (j = start; j < i; j++) {
5ccb3ea7 1059 pending_idx = frag_get_pending_idx(&shinfo->frags[j]);
7d5145d8 1060 xen_netbk_idx_release(netbk, pending_idx, XEN_NETIF_RSP_OKAY);
f942dc25
IC		 1061 }
1062
1063 /* Remember the error: invalidate all subsequent fragments. */
1064 err = newerr;
1065 }
1066
1067 *gopp = gop + 1;
1068 return err;
1069}
1070
1071static void xen_netbk_fill_frags(struct xen_netbk *netbk, struct sk_buff *skb)
1072{
1073 struct skb_shared_info *shinfo = skb_shinfo(skb);
1074 int nr_frags = shinfo->nr_frags;
1075 int i;
1076
1077 for (i = 0; i < nr_frags; i++) {
1078 skb_frag_t *frag = shinfo->frags + i;
1079 struct xen_netif_tx_request *txp;
ea066ad1
IC		 1080 struct page *page;
1081 u16 pending_idx;
f942dc25 1082
ea066ad1 1083 pending_idx = frag_get_pending_idx(frag);
f942dc25
IC		 1084
1085 txp = &netbk->pending_tx_info[pending_idx].req;
ea066ad1
IC		 1086 page = virt_to_page(idx_to_kaddr(netbk, pending_idx));
1087 __skb_fill_page_desc(skb, i, page, txp->offset, txp->size);
f942dc25
IC		 1088 skb->len += txp->size;
1089 skb->data_len += txp->size;
1090 skb->truesize += txp->size;
1091
1092 /* Take an extra reference to offset xen_netbk_idx_release */
1093 get_page(netbk->mmap_pages[pending_idx]);
7d5145d8 1094 xen_netbk_idx_release(netbk, pending_idx, XEN_NETIF_RSP_OKAY);
f942dc25
IC		 1095 }
1096}
1097
1098static int xen_netbk_get_extras(struct xenvif *vif,
1099 struct xen_netif_extra_info *extras,
1100 int work_to_do)
1101{
1102 struct xen_netif_extra_info extra;
1103 RING_IDX cons = vif->tx.req_cons;
1104
1105 do {
1106 if (unlikely(work_to_do-- <= 0)) {
48856286
IC		 1107 netdev_err(vif->dev, "Missing extra info\n");
1108 netbk_fatal_tx_err(vif);
f942dc25
IC		 1109 return -EBADR;
1110 }
1111
1112 memcpy(&extra, RING_GET_REQUEST(&vif->tx, cons),
1113 sizeof(extra));
1114 if (unlikely(!extra.type ||
1115 extra.type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
1116 vif->tx.req_cons = ++cons;
48856286 1117 netdev_err(vif->dev,
f942dc25 1118 "Invalid extra type: %d\n", extra.type);
48856286 1119 netbk_fatal_tx_err(vif);
f942dc25
IC		 1120 return -EINVAL;
1121 }
1122
1123 memcpy(&extras[extra.type - 1], &extra, sizeof(extra));
1124 vif->tx.req_cons = ++cons;
1125 } while (extra.flags & XEN_NETIF_EXTRA_FLAG_MORE);
1126
1127 return work_to_do;
1128}
1129
1130static int netbk_set_skb_gso(struct xenvif *vif,
1131 struct sk_buff *skb,
1132 struct xen_netif_extra_info *gso)
1133{
1134 if (!gso->u.gso.size) {
48856286
IC		 1135 netdev_err(vif->dev, "GSO size must not be zero.\n");
1136 netbk_fatal_tx_err(vif);
f942dc25
IC		 1137 return -EINVAL;
1138 }
1139
1140 /* Currently only TCPv4 S.O. is supported. */
1141 if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4) {
48856286
IC		 1142 netdev_err(vif->dev, "Bad GSO type %d.\n", gso->u.gso.type);
1143 netbk_fatal_tx_err(vif);
f942dc25
IC		 1144 return -EINVAL;
1145 }
1146
1147 skb_shinfo(skb)->gso_size = gso->u.gso.size;
1148 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1149
1150 /* Header must be checked, and gso_segs computed. */
1151 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1152 skb_shinfo(skb)->gso_segs = 0;
1153
1154 return 0;
1155}
1156
1157static int checksum_setup(struct xenvif *vif, struct sk_buff *skb)
1158{
1159 struct iphdr *iph;
1160 unsigned char *th;
1161 int err = -EPROTO;
1162 int recalculate_partial_csum = 0;
1163
1164 /*
1165 * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy
1166 * peers can fail to set NETRXF_csum_blank when sending a GSO
1167 * frame. In this case force the SKB to CHECKSUM_PARTIAL and
1168 * recalculate the partial checksum.
1169 */
1170 if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) {
1171 vif->rx_gso_checksum_fixup++;
1172 skb->ip_summed = CHECKSUM_PARTIAL;
1173 recalculate_partial_csum = 1;
1174 }
1175
1176 /* A non-CHECKSUM_PARTIAL SKB does not require setup. */
1177 if (skb->ip_summed != CHECKSUM_PARTIAL)
1178 return 0;
1179
1180 if (skb->protocol != htons(ETH_P_IP))
1181 goto out;
1182
1183 iph = (void *)skb->data;
1184 th = skb->data + 4 * iph->ihl;
1185 if (th >= skb_tail_pointer(skb))
1186 goto out;
1187
1188 skb->csum_start = th - skb->head;
1189 switch (iph->protocol) {
1190 case IPPROTO_TCP:
1191 skb->csum_offset = offsetof(struct tcphdr, check);
1192
1193 if (recalculate_partial_csum) {
1194 struct tcphdr *tcph = (struct tcphdr *)th;
1195 tcph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
1196 skb->len - iph->ihl*4,
1197 IPPROTO_TCP, 0);
1198 }
1199 break;
1200 case IPPROTO_UDP:
1201 skb->csum_offset = offsetof(struct udphdr, check);
1202
1203 if (recalculate_partial_csum) {
1204 struct udphdr *udph = (struct udphdr *)th;
1205 udph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
1206 skb->len - iph->ihl*4,
1207 IPPROTO_UDP, 0);
1208 }
1209 break;
1210 default:
1211 if (net_ratelimit())
1212 netdev_err(vif->dev,
1213 "Attempting to checksum a non-TCP/UDP packet, dropping a protocol %d packet\n",
1214 iph->protocol);
1215 goto out;
1216 }
1217
1218 if ((th + skb->csum_offset + 2) > skb_tail_pointer(skb))
1219 goto out;
1220
1221 err = 0;
1222
1223out:
1224 return err;
1225}
1226
1227static bool tx_credit_exceeded(struct xenvif *vif, unsigned size)
1228{
1229 unsigned long now = jiffies;
1230 unsigned long next_credit =
1231 vif->credit_timeout.expires +
1232 msecs_to_jiffies(vif->credit_usec / 1000);
1233
1234 /* Timer could already be pending in rare cases. */
1235 if (timer_pending(&vif->credit_timeout))
1236 return true;
1237
1238 /* Passed the point where we can replenish credit? */
1239 if (time_after_eq(now, next_credit)) {
1240 vif->credit_timeout.expires = now;
1241 tx_add_credit(vif);
1242 }
1243
1244 /* Still too big to send right now? Set a callback. */
1245 if (size > vif->remaining_credit) {
1246 vif->credit_timeout.data =
1247 (unsigned long)vif;
1248 vif->credit_timeout.function =
1249 tx_credit_callback;
1250 mod_timer(&vif->credit_timeout,
1251 next_credit);
1252
1253 return true;
1254 }
1255
1256 return false;
1257}
1258
1259static unsigned xen_netbk_tx_build_gops(struct xen_netbk *netbk)
1260{
1261 struct gnttab_copy *gop = netbk->tx_copy_ops, *request_gop;
1262 struct sk_buff *skb;
1263 int ret;
1264
1265 while (((nr_pending_reqs(netbk) + MAX_SKB_FRAGS) < MAX_PENDING_REQS) &&
1266 !list_empty(&netbk->net_schedule_list)) {
1267 struct xenvif *vif;
1268 struct xen_netif_tx_request txreq;
1269 struct xen_netif_tx_request txfrags[MAX_SKB_FRAGS];
1270 struct page *page;
1271 struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX-1];
1272 u16 pending_idx;
1273 RING_IDX idx;
1274 int work_to_do;
1275 unsigned int data_len;
1276 pending_ring_idx_t index;
1277
1278 /* Get a netif from the list with work to do. */
1279 vif = poll_net_schedule_list(netbk);
48856286
IC		 1280 /* This can sometimes happen because the test of
1281 * list_empty(net_schedule_list) at the top of the
1282 * loop is unlocked. Just go back and have another
1283 * look.
1284 */
f942dc25
IC		 1285 if (!vif)
1286 continue;
1287
48856286
IC		 1288 if (vif->tx.sring->req_prod - vif->tx.req_cons >
1289 XEN_NETIF_TX_RING_SIZE) {
1290 netdev_err(vif->dev,
1291 "Impossible number of requests. "
1292 "req_prod %d, req_cons %d, size %ld\n",
1293 vif->tx.sring->req_prod, vif->tx.req_cons,
1294 XEN_NETIF_TX_RING_SIZE);
1295 netbk_fatal_tx_err(vif);
1296 continue;
1297 }
1298
f942dc25
IC		 1299 RING_FINAL_CHECK_FOR_REQUESTS(&vif->tx, work_to_do);
1300 if (!work_to_do) {
1301 xenvif_put(vif);
1302 continue;
1303 }
1304
1305 idx = vif->tx.req_cons;
1306 rmb(); /* Ensure that we see the request before we copy it. */
1307 memcpy(&txreq, RING_GET_REQUEST(&vif->tx, idx), sizeof(txreq));
1308
1309 /* Credit-based scheduling. */
1310 if (txreq.size > vif->remaining_credit &&
1311 tx_credit_exceeded(vif, txreq.size)) {
1312 xenvif_put(vif);
1313 continue;
1314 }
1315
1316 vif->remaining_credit -= txreq.size;
1317
1318 work_to_do--;
1319 vif->tx.req_cons = ++idx;
1320
1321 memset(extras, 0, sizeof(extras));
1322 if (txreq.flags & XEN_NETTXF_extra_info) {
1323 work_to_do = xen_netbk_get_extras(vif, extras,
1324 work_to_do);
1325 idx = vif->tx.req_cons;
48856286 1326 if (unlikely(work_to_do < 0))
f942dc25 1327 continue;
f942dc25
IC		 1328 }
1329
1330 ret = netbk_count_requests(vif, &txreq, txfrags, work_to_do);
48856286 1331 if (unlikely(ret < 0))
f942dc25 1332 continue;
48856286 1333
f942dc25
IC		 1334 idx += ret;
1335
1336 if (unlikely(txreq.size < ETH_HLEN)) {
1337 netdev_dbg(vif->dev,
1338 "Bad packet size: %d\n", txreq.size);
1339 netbk_tx_err(vif, &txreq, idx);
1340 continue;
1341 }
1342
1343 /* No crossing a page as the payload mustn't fragment. */
1344 if (unlikely((txreq.offset + txreq.size) > PAGE_SIZE)) {
48856286 1345 netdev_err(vif->dev,
f942dc25
IC		 1346 "txreq.offset: %x, size: %u, end: %lu\n",
1347 txreq.offset, txreq.size,
1348 (txreq.offset&~PAGE_MASK) + txreq.size);
48856286 1349 netbk_fatal_tx_err(vif);
f942dc25
IC		 1350 continue;
1351 }
1352
1353 index = pending_index(netbk->pending_cons);
1354 pending_idx = netbk->pending_ring[index];
1355
1356 data_len = (txreq.size > PKT_PROT_LEN &&
1357 ret < MAX_SKB_FRAGS) ?
1358 PKT_PROT_LEN : txreq.size;
1359
1360 skb = alloc_skb(data_len + NET_SKB_PAD + NET_IP_ALIGN,
1361 GFP_ATOMIC | __GFP_NOWARN);
1362 if (unlikely(skb == NULL)) {
1363 netdev_dbg(vif->dev,
1364 "Can't allocate a skb in start_xmit.\n");
1365 netbk_tx_err(vif, &txreq, idx);
1366 break;
1367 }
1368
1369 /* Packets passed to netif_rx() must have some headroom. */
1370 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
1371
1372 if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) {
1373 struct xen_netif_extra_info *gso;
1374 gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1];
1375
1376 if (netbk_set_skb_gso(vif, skb, gso)) {
48856286 1377 /* Failure in netbk_set_skb_gso is fatal. */
f942dc25 1378 kfree_skb(skb);
f942dc25
IC		 1379 continue;
1380 }
1381 }
1382
1383 /* XXX could copy straight to head */
1384 page = xen_netbk_alloc_page(netbk, skb, pending_idx);
1385 if (!page) {
1386 kfree_skb(skb);
1387 netbk_tx_err(vif, &txreq, idx);
1388 continue;
1389 }
1390
f942dc25
IC		 1391 gop->source.u.ref = txreq.gref;
1392 gop->source.domid = vif->domid;
1393 gop->source.offset = txreq.offset;
1394
1395 gop->dest.u.gmfn = virt_to_mfn(page_address(page));
1396 gop->dest.domid = DOMID_SELF;
1397 gop->dest.offset = txreq.offset;
1398
1399 gop->len = txreq.size;
1400 gop->flags = GNTCOPY_source_gref;
1401
1402 gop++;
1403
1404 memcpy(&netbk->pending_tx_info[pending_idx].req,
1405 &txreq, sizeof(txreq));
1406 netbk->pending_tx_info[pending_idx].vif = vif;
1407 *((u16 *)skb->data) = pending_idx;
1408
1409 __skb_put(skb, data_len);
1410
1411 skb_shinfo(skb)->nr_frags = ret;
1412 if (data_len < txreq.size) {
1413 skb_shinfo(skb)->nr_frags++;
ea066ad1
IC		 1414 frag_set_pending_idx(&skb_shinfo(skb)->frags[0],
1415 pending_idx);
f942dc25 1416 } else {
ea066ad1
IC		 1417 frag_set_pending_idx(&skb_shinfo(skb)->frags[0],
1418 INVALID_PENDING_IDX);
f942dc25
IC		 1419 }
1420
f942dc25
IC		 1421 netbk->pending_cons++;
1422
1423 request_gop = xen_netbk_get_requests(netbk, vif,
1424 skb, txfrags, gop);
1425 if (request_gop == NULL) {
1426 kfree_skb(skb);
1427 netbk_tx_err(vif, &txreq, idx);
1428 continue;
1429 }
1430 gop = request_gop;
1431
1e0b6eac
AL		 1432 __skb_queue_tail(&netbk->tx_queue, skb);
1433
f942dc25
IC		 1434 vif->tx.req_cons = idx;
1435 xen_netbk_check_rx_xenvif(vif);
1436
1437 if ((gop-netbk->tx_copy_ops) >= ARRAY_SIZE(netbk->tx_copy_ops))
1438 break;
1439 }
1440
1441 return gop - netbk->tx_copy_ops;
1442}
1443
1444static void xen_netbk_tx_submit(struct xen_netbk *netbk)
1445{
1446 struct gnttab_copy *gop = netbk->tx_copy_ops;
1447 struct sk_buff *skb;
1448
1449 while ((skb = __skb_dequeue(&netbk->tx_queue)) != NULL) {
1450 struct xen_netif_tx_request *txp;
1451 struct xenvif *vif;
1452 u16 pending_idx;
1453 unsigned data_len;
1454
1455 pending_idx = *((u16 *)skb->data);
1456 vif = netbk->pending_tx_info[pending_idx].vif;
1457 txp = &netbk->pending_tx_info[pending_idx].req;
1458
1459 /* Check the remap error code. */
1460 if (unlikely(xen_netbk_tx_check_gop(netbk, skb, &gop))) {
1461 netdev_dbg(vif->dev, "netback grant failed.\n");
1462 skb_shinfo(skb)->nr_frags = 0;
1463 kfree_skb(skb);
1464 continue;
1465 }
1466
1467 data_len = skb->len;
1468 memcpy(skb->data,
1469 (void *)(idx_to_kaddr(netbk, pending_idx)|txp->offset),
1470 data_len);
1471 if (data_len < txp->size) {
1472 /* Append the packet payload as a fragment. */
1473 txp->offset += data_len;
1474 txp->size -= data_len;
1475 } else {
1476 /* Schedule a response immediately. */
7d5145d8 1477 xen_netbk_idx_release(netbk, pending_idx, XEN_NETIF_RSP_OKAY);
f942dc25
IC		 1478 }
1479
1480 if (txp->flags & XEN_NETTXF_csum_blank)
1481 skb->ip_summed = CHECKSUM_PARTIAL;
1482 else if (txp->flags & XEN_NETTXF_data_validated)
1483 skb->ip_summed = CHECKSUM_UNNECESSARY;
1484
1485 xen_netbk_fill_frags(netbk, skb);
1486
1487 /*
1488 * If the initial fragment was < PKT_PROT_LEN then
1489 * pull through some bytes from the other fragments to
1490 * increase the linear region to PKT_PROT_LEN bytes.
1491 */
1492 if (skb_headlen(skb) < PKT_PROT_LEN && skb_is_nonlinear(skb)) {
1493 int target = min_t(int, skb->len, PKT_PROT_LEN);
1494 __pskb_pull_tail(skb, target - skb_headlen(skb));
1495 }
1496
1497 skb->dev = vif->dev;
1498 skb->protocol = eth_type_trans(skb, skb->dev);
1499
1500 if (checksum_setup(vif, skb)) {
1501 netdev_dbg(vif->dev,
1502 "Can't setup checksum in net_tx_action\n");
1503 kfree_skb(skb);
1504 continue;
1505 }
1506
1507 vif->dev->stats.rx_bytes += skb->len;
1508 vif->dev->stats.rx_packets++;
1509
1510 xenvif_receive_skb(vif, skb);
1511 }
1512}
1513
1514/* Called after netfront has transmitted */
1515static void xen_netbk_tx_action(struct xen_netbk *netbk)
1516{
1517 unsigned nr_gops;
f942dc25
IC		 1518
1519 nr_gops = xen_netbk_tx_build_gops(netbk);
1520
1521 if (nr_gops == 0)
1522 return;
f942dc25 1523
c571898f 1524 gnttab_batch_copy(netbk->tx_copy_ops, nr_gops);
f942dc25 1525
c571898f 1526 xen_netbk_tx_submit(netbk);
f942dc25
IC		 1527}
1528
7d5145d8
MD		 1529static void xen_netbk_idx_release(struct xen_netbk *netbk, u16 pending_idx,
1530 u8 status)
f942dc25
IC		 1531{
1532 struct xenvif *vif;
1533 struct pending_tx_info *pending_tx_info;
1534 pending_ring_idx_t index;
1535
1536 /* Already complete? */
1537 if (netbk->mmap_pages[pending_idx] == NULL)
1538 return;
1539
1540 pending_tx_info = &netbk->pending_tx_info[pending_idx];
1541
1542 vif = pending_tx_info->vif;
1543
7d5145d8 1544 make_tx_response(vif, &pending_tx_info->req, status);
f942dc25
IC		 1545
1546 index = pending_index(netbk->pending_prod++);
1547 netbk->pending_ring[index] = pending_idx;
1548
1549 xenvif_put(vif);
1550
1551 netbk->mmap_pages[pending_idx]->mapping = 0;
1552 put_page(netbk->mmap_pages[pending_idx]);
1553 netbk->mmap_pages[pending_idx] = NULL;
1554}
1555
1556static void make_tx_response(struct xenvif *vif,
1557 struct xen_netif_tx_request *txp,
1558 s8 st)
1559{
1560 RING_IDX i = vif->tx.rsp_prod_pvt;
1561 struct xen_netif_tx_response *resp;
1562 int notify;
1563
1564 resp = RING_GET_RESPONSE(&vif->tx, i);
1565 resp->id = txp->id;
1566 resp->status = st;
1567
1568 if (txp->flags & XEN_NETTXF_extra_info)
1569 RING_GET_RESPONSE(&vif->tx, ++i)->status = XEN_NETIF_RSP_NULL;
1570
1571 vif->tx.rsp_prod_pvt = ++i;
1572 RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&vif->tx, notify);
1573 if (notify)
1574 notify_remote_via_irq(vif->irq);
1575}
1576
1577static struct xen_netif_rx_response *make_rx_response(struct xenvif *vif,
1578 u16 id,
1579 s8 st,
1580 u16 offset,
1581 u16 size,
1582 u16 flags)
1583{
1584 RING_IDX i = vif->rx.rsp_prod_pvt;
1585 struct xen_netif_rx_response *resp;
1586
1587 resp = RING_GET_RESPONSE(&vif->rx, i);
1588 resp->offset = offset;
1589 resp->flags = flags;
1590 resp->id = id;
1591 resp->status = (s16)size;
1592 if (st < 0)
1593 resp->status = (s16)st;
1594
1595 vif->rx.rsp_prod_pvt = ++i;
1596
1597 return resp;
1598}
1599
1600static inline int rx_work_todo(struct xen_netbk *netbk)
1601{
1602 return !skb_queue_empty(&netbk->rx_queue);
1603}
1604
1605static inline int tx_work_todo(struct xen_netbk *netbk)
1606{
1607
1608 if (((nr_pending_reqs(netbk) + MAX_SKB_FRAGS) < MAX_PENDING_REQS) &&
1609 !list_empty(&netbk->net_schedule_list))
1610 return 1;
1611
1612 return 0;
1613}
1614
1615static int xen_netbk_kthread(void *data)
1616{
1617 struct xen_netbk *netbk = data;
1618 while (!kthread_should_stop()) {
1619 wait_event_interruptible(netbk->wq,
1620 rx_work_todo(netbk) ||
1621 tx_work_todo(netbk) ||
1622 kthread_should_stop());
1623 cond_resched();
1624
1625 if (kthread_should_stop())
1626 break;
1627
1628 if (rx_work_todo(netbk))
1629 xen_netbk_rx_action(netbk);
1630
1631 if (tx_work_todo(netbk))
1632 xen_netbk_tx_action(netbk);
1633 }
1634
1635 return 0;
1636}
1637
1638void xen_netbk_unmap_frontend_rings(struct xenvif *vif)
1639{
c9d63699
DV		 1640 if (vif->tx.sring)
1641 xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif),
1642 vif->tx.sring);
1643 if (vif->rx.sring)
1644 xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif),
1645 vif->rx.sring);
f942dc25
IC		 1646}
1647
1648int xen_netbk_map_frontend_rings(struct xenvif *vif,
1649 grant_ref_t tx_ring_ref,
1650 grant_ref_t rx_ring_ref)
1651{
c9d63699 1652 void *addr;
f942dc25
IC		 1653 struct xen_netif_tx_sring *txs;
1654 struct xen_netif_rx_sring *rxs;
1655
1656 int err = -ENOMEM;
1657
c9d63699
DV		 1658 err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(vif),
1659 tx_ring_ref, &addr);
1660 if (err)
f942dc25
IC		 1661 goto err;
1662
c9d63699 1663 txs = (struct xen_netif_tx_sring *)addr;
f942dc25
IC		 1664 BACK_RING_INIT(&vif->tx, txs, PAGE_SIZE);
1665
c9d63699
DV		 1666 err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(vif),
1667 rx_ring_ref, &addr);
1668 if (err)
f942dc25 1669 goto err;
f942dc25 1670
c9d63699 1671 rxs = (struct xen_netif_rx_sring *)addr;
f942dc25
IC		 1672 BACK_RING_INIT(&vif->rx, rxs, PAGE_SIZE);
1673
c9d63699
DV		 1674 vif->rx_req_cons_peek = 0;
1675
f942dc25
IC		 1676 return 0;
1677
1678err:
1679 xen_netbk_unmap_frontend_rings(vif);
1680 return err;
1681}
1682
1683static int __init netback_init(void)
1684{
1685 int i;
1686 int rc = 0;
1687 int group;
1688
2a14b244 1689 if (!xen_domain())
f942dc25
IC		 1690 return -ENODEV;
1691
1692 xen_netbk_group_nr = num_online_cpus();
1693 xen_netbk = vzalloc(sizeof(struct xen_netbk) * xen_netbk_group_nr);
e404decb 1694 if (!xen_netbk)
f942dc25 1695 return -ENOMEM;
f942dc25
IC		 1696
1697 for (group = 0; group < xen_netbk_group_nr; group++) {
1698 struct xen_netbk *netbk = &xen_netbk[group];
1699 skb_queue_head_init(&netbk->rx_queue);
1700 skb_queue_head_init(&netbk->tx_queue);
1701
1702 init_timer(&netbk->net_timer);
1703 netbk->net_timer.data = (unsigned long)netbk;
1704 netbk->net_timer.function = xen_netbk_alarm;
1705
1706 netbk->pending_cons = 0;
1707 netbk->pending_prod = MAX_PENDING_REQS;
1708 for (i = 0; i < MAX_PENDING_REQS; i++)
1709 netbk->pending_ring[i] = i;
1710
1711 init_waitqueue_head(&netbk->wq);
1712 netbk->task = kthread_create(xen_netbk_kthread,
1713 (void *)netbk,
1714 "netback/%u", group);
1715
1716 if (IS_ERR(netbk->task)) {
6b84bd16 1717 printk(KERN_ALERT "kthread_create() fails at netback\n");
f942dc25
IC		 1718 del_timer(&netbk->net_timer);
1719 rc = PTR_ERR(netbk->task);
1720 goto failed_init;
1721 }
1722
1723 kthread_bind(netbk->task, group);
1724
1725 INIT_LIST_HEAD(&netbk->net_schedule_list);
1726
1727 spin_lock_init(&netbk->net_schedule_list_lock);
1728
1729 atomic_set(&netbk->netfront_count, 0);
1730
1731 wake_up_process(netbk->task);
1732 }
1733
1734 rc = xenvif_xenbus_init();
1735 if (rc)
1736 goto failed_init;
1737
1738 return 0;
1739
1740failed_init:
1741 while (--group >= 0) {
1742 struct xen_netbk *netbk = &xen_netbk[group];
1743 for (i = 0; i < MAX_PENDING_REQS; i++) {
1744 if (netbk->mmap_pages[i])
1745 __free_page(netbk->mmap_pages[i]);
1746 }
1747 del_timer(&netbk->net_timer);
1748 kthread_stop(netbk->task);
1749 }
1750 vfree(xen_netbk);
1751 return rc;
1752
1753}
1754
1755module_init(netback_init);
1756
1757MODULE_LICENSE("Dual BSD/GPL");
f984cec6 1758MODULE_ALIAS("xen-backend:vif");
Linux 6.x block layer and io_uring tree(s)