blob: a2865f17c6673837ea982b0ac2a983c455a62cad [file] [log] [blame]
Ian Campbellf942dc22011-03-15 00:06:18 +00001/*
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
Stefano Stabellinica981632012-08-08 17:21:23 +000043#include <xen/xen.h>
Ian Campbellf942dc22011-03-15 00:06:18 +000044#include <xen/events.h>
45#include <xen/interface/memory.h>
46
47#include <asm/xen/hypercall.h>
48#include <asm/xen/page.h>
49
Wei Liu2810e5b2013-04-22 02:20:42 +000050/*
51 * This is the maximum slots a skb can have. If a guest sends a skb
52 * which exceeds this limit it is considered malicious.
53 */
54#define MAX_SKB_SLOTS_DEFAULT 20
55static unsigned int max_skb_slots = MAX_SKB_SLOTS_DEFAULT;
56module_param(max_skb_slots, uint, 0444);
57
Ian Campbellf942dc22011-03-15 00:06:18 +000058typedef unsigned int pending_ring_idx_t;
Wei Liu2810e5b2013-04-22 02:20:42 +000059#define INVALID_PENDING_RING_IDX (~0U)
60
61struct pending_tx_info {
62 struct xen_netif_tx_request req; /* coalesced tx request */
63 struct xenvif *vif;
64 pending_ring_idx_t head; /* head != INVALID_PENDING_RING_IDX
65 * if it is head of one or more tx
66 * reqs
67 */
68};
Ian Campbellf942dc22011-03-15 00:06:18 +000069
70struct netbk_rx_meta {
71 int id;
72 int size;
73 int gso_size;
74};
75
76#define MAX_PENDING_REQS 256
77
Ian Campbellea066ad2011-10-05 00:28:46 +000078/* Discriminate from any valid pending_idx value. */
79#define INVALID_PENDING_IDX 0xFFFF
80
Ian Campbellf942dc22011-03-15 00:06:18 +000081#define MAX_BUFFER_OFFSET PAGE_SIZE
82
83/* extra field used in struct page */
84union page_ext {
85 struct {
86#if BITS_PER_LONG < 64
87#define IDX_WIDTH 8
88#define GROUP_WIDTH (BITS_PER_LONG - IDX_WIDTH)
89 unsigned int group:GROUP_WIDTH;
90 unsigned int idx:IDX_WIDTH;
91#else
92 unsigned int group, idx;
93#endif
94 } e;
95 void *mapping;
96};
97
98struct xen_netbk {
99 wait_queue_head_t wq;
100 struct task_struct *task;
101
102 struct sk_buff_head rx_queue;
103 struct sk_buff_head tx_queue;
104
105 struct timer_list net_timer;
106
107 struct page *mmap_pages[MAX_PENDING_REQS];
108
109 pending_ring_idx_t pending_prod;
110 pending_ring_idx_t pending_cons;
111 struct list_head net_schedule_list;
112
113 /* Protect the net_schedule_list in netif. */
114 spinlock_t net_schedule_list_lock;
115
116 atomic_t netfront_count;
117
118 struct pending_tx_info pending_tx_info[MAX_PENDING_REQS];
Wei Liu2810e5b2013-04-22 02:20:42 +0000119 /* Coalescing tx requests before copying makes number of grant
120 * copy ops greater or equal to number of slots required. In
121 * worst case a tx request consumes 2 gnttab_copy.
122 */
123 struct gnttab_copy tx_copy_ops[2*MAX_PENDING_REQS];
Ian Campbellf942dc22011-03-15 00:06:18 +0000124
125 u16 pending_ring[MAX_PENDING_REQS];
126
127 /*
128 * Given MAX_BUFFER_OFFSET of 4096 the worst case is that each
129 * head/fragment page uses 2 copy operations because it
130 * straddles two buffers in the frontend.
131 */
132 struct gnttab_copy grant_copy_op[2*XEN_NETIF_RX_RING_SIZE];
133 struct netbk_rx_meta meta[2*XEN_NETIF_RX_RING_SIZE];
134};
135
136static struct xen_netbk *xen_netbk;
137static int xen_netbk_group_nr;
138
Wei Liu2810e5b2013-04-22 02:20:42 +0000139/*
140 * If head != INVALID_PENDING_RING_IDX, it means this tx request is head of
141 * one or more merged tx requests, otherwise it is the continuation of
142 * previous tx request.
143 */
144static inline int pending_tx_is_head(struct xen_netbk *netbk, RING_IDX idx)
145{
146 return netbk->pending_tx_info[idx].head != INVALID_PENDING_RING_IDX;
147}
148
Ian Campbellf942dc22011-03-15 00:06:18 +0000149void xen_netbk_add_xenvif(struct xenvif *vif)
150{
151 int i;
152 int min_netfront_count;
153 int min_group = 0;
154 struct xen_netbk *netbk;
155
156 min_netfront_count = atomic_read(&xen_netbk[0].netfront_count);
157 for (i = 0; i < xen_netbk_group_nr; i++) {
158 int netfront_count = atomic_read(&xen_netbk[i].netfront_count);
159 if (netfront_count < min_netfront_count) {
160 min_group = i;
161 min_netfront_count = netfront_count;
162 }
163 }
164
165 netbk = &xen_netbk[min_group];
166
167 vif->netbk = netbk;
168 atomic_inc(&netbk->netfront_count);
169}
170
171void xen_netbk_remove_xenvif(struct xenvif *vif)
172{
173 struct xen_netbk *netbk = vif->netbk;
174 vif->netbk = NULL;
175 atomic_dec(&netbk->netfront_count);
176}
177
Matthew Daley7d5145d2013-02-06 23:41:36 +0000178static void xen_netbk_idx_release(struct xen_netbk *netbk, u16 pending_idx,
179 u8 status);
Ian Campbellf942dc22011-03-15 00:06:18 +0000180static void make_tx_response(struct xenvif *vif,
181 struct xen_netif_tx_request *txp,
182 s8 st);
183static struct xen_netif_rx_response *make_rx_response(struct xenvif *vif,
184 u16 id,
185 s8 st,
186 u16 offset,
187 u16 size,
188 u16 flags);
189
190static inline unsigned long idx_to_pfn(struct xen_netbk *netbk,
Ian Campbellea066ad2011-10-05 00:28:46 +0000191 u16 idx)
Ian Campbellf942dc22011-03-15 00:06:18 +0000192{
193 return page_to_pfn(netbk->mmap_pages[idx]);
194}
195
196static inline unsigned long idx_to_kaddr(struct xen_netbk *netbk,
Ian Campbellea066ad2011-10-05 00:28:46 +0000197 u16 idx)
Ian Campbellf942dc22011-03-15 00:06:18 +0000198{
199 return (unsigned long)pfn_to_kaddr(idx_to_pfn(netbk, idx));
200}
201
202/* extra field used in struct page */
203static inline void set_page_ext(struct page *pg, struct xen_netbk *netbk,
204 unsigned int idx)
205{
206 unsigned int group = netbk - xen_netbk;
207 union page_ext ext = { .e = { .group = group + 1, .idx = idx } };
208
209 BUILD_BUG_ON(sizeof(ext) > sizeof(ext.mapping));
210 pg->mapping = ext.mapping;
211}
212
213static int get_page_ext(struct page *pg,
214 unsigned int *pgroup, unsigned int *pidx)
215{
216 union page_ext ext = { .mapping = pg->mapping };
217 struct xen_netbk *netbk;
218 unsigned int group, idx;
219
220 group = ext.e.group - 1;
221
222 if (group < 0 || group >= xen_netbk_group_nr)
223 return 0;
224
225 netbk = &xen_netbk[group];
226
227 idx = ext.e.idx;
228
229 if ((idx < 0) || (idx >= MAX_PENDING_REQS))
230 return 0;
231
232 if (netbk->mmap_pages[idx] != pg)
233 return 0;
234
235 *pgroup = group;
236 *pidx = idx;
237
238 return 1;
239}
240
241/*
242 * This is the amount of packet we copy rather than map, so that the
243 * guest can't fiddle with the contents of the headers while we do
244 * packet processing on them (netfilter, routing, etc).
245 */
246#define PKT_PROT_LEN (ETH_HLEN + \
247 VLAN_HLEN + \
248 sizeof(struct iphdr) + MAX_IPOPTLEN + \
249 sizeof(struct tcphdr) + MAX_TCP_OPTION_SPACE)
250
Ian Campbellea066ad2011-10-05 00:28:46 +0000251static u16 frag_get_pending_idx(skb_frag_t *frag)
252{
253 return (u16)frag->page_offset;
254}
255
256static void frag_set_pending_idx(skb_frag_t *frag, u16 pending_idx)
257{
258 frag->page_offset = pending_idx;
259}
260
Ian Campbellf942dc22011-03-15 00:06:18 +0000261static inline pending_ring_idx_t pending_index(unsigned i)
262{
263 return i & (MAX_PENDING_REQS-1);
264}
265
266static inline pending_ring_idx_t nr_pending_reqs(struct xen_netbk *netbk)
267{
268 return MAX_PENDING_REQS -
269 netbk->pending_prod + netbk->pending_cons;
270}
271
272static void xen_netbk_kick_thread(struct xen_netbk *netbk)
273{
274 wake_up(&netbk->wq);
275}
276
277static int max_required_rx_slots(struct xenvif *vif)
278{
279 int max = DIV_ROUND_UP(vif->dev->mtu, PAGE_SIZE);
280
Wei Liu2810e5b2013-04-22 02:20:42 +0000281 /* XXX FIXME: RX path dependent on MAX_SKB_FRAGS */
Ian Campbellf942dc22011-03-15 00:06:18 +0000282 if (vif->can_sg || vif->gso || vif->gso_prefix)
283 max += MAX_SKB_FRAGS + 1; /* extra_info + frags */
284
285 return max;
286}
287
288int xen_netbk_rx_ring_full(struct xenvif *vif)
289{
290 RING_IDX peek = vif->rx_req_cons_peek;
291 RING_IDX needed = max_required_rx_slots(vif);
292
293 return ((vif->rx.sring->req_prod - peek) < needed) ||
294 ((vif->rx.rsp_prod_pvt + XEN_NETIF_RX_RING_SIZE - peek) < needed);
295}
296
297int xen_netbk_must_stop_queue(struct xenvif *vif)
298{
299 if (!xen_netbk_rx_ring_full(vif))
300 return 0;
301
302 vif->rx.sring->req_event = vif->rx_req_cons_peek +
303 max_required_rx_slots(vif);
304 mb(); /* request notification /then/ check the queue */
305
306 return xen_netbk_rx_ring_full(vif);
307}
308
309/*
310 * Returns true if we should start a new receive buffer instead of
311 * adding 'size' bytes to a buffer which currently contains 'offset'
312 * bytes.
313 */
314static bool start_new_rx_buffer(int offset, unsigned long size, int head)
315{
316 /* simple case: we have completely filled the current buffer. */
317 if (offset == MAX_BUFFER_OFFSET)
318 return true;
319
320 /*
321 * complex case: start a fresh buffer if the current frag
322 * would overflow the current buffer but only if:
323 * (i) this frag would fit completely in the next buffer
324 * and (ii) there is already some data in the current buffer
325 * and (iii) this is not the head buffer.
326 *
327 * Where:
328 * - (i) stops us splitting a frag into two copies
329 * unless the frag is too large for a single buffer.
330 * - (ii) stops us from leaving a buffer pointlessly empty.
331 * - (iii) stops us leaving the first buffer
332 * empty. Strictly speaking this is already covered
333 * by (ii) but is explicitly checked because
334 * netfront relies on the first buffer being
335 * non-empty and can crash otherwise.
336 *
337 * This means we will effectively linearise small
338 * frags but do not needlessly split large buffers
339 * into multiple copies tend to give large frags their
340 * own buffers as before.
341 */
342 if ((offset + size > MAX_BUFFER_OFFSET) &&
343 (size <= MAX_BUFFER_OFFSET) && offset && !head)
344 return true;
345
346 return false;
347}
348
349/*
350 * Figure out how many ring slots we're going to need to send @skb to
351 * the guest. This function is essentially a dry run of
352 * netbk_gop_frag_copy.
353 */
354unsigned int xen_netbk_count_skb_slots(struct xenvif *vif, struct sk_buff *skb)
355{
356 unsigned int count;
357 int i, copy_off;
358
Simon Grahame26b2032012-05-24 06:26:07 +0000359 count = DIV_ROUND_UP(skb_headlen(skb), PAGE_SIZE);
Ian Campbellf942dc22011-03-15 00:06:18 +0000360
361 copy_off = skb_headlen(skb) % PAGE_SIZE;
362
363 if (skb_shinfo(skb)->gso_size)
364 count++;
365
366 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
Eric Dumazet9e903e02011-10-18 21:00:24 +0000367 unsigned long size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
Ian Campbell6a8ed462012-10-10 03:48:42 +0000368 unsigned long offset = skb_shinfo(skb)->frags[i].page_offset;
Ian Campbellf942dc22011-03-15 00:06:18 +0000369 unsigned long bytes;
Ian Campbell6a8ed462012-10-10 03:48:42 +0000370
371 offset &= ~PAGE_MASK;
372
Ian Campbellf942dc22011-03-15 00:06:18 +0000373 while (size > 0) {
Ian Campbell6a8ed462012-10-10 03:48:42 +0000374 BUG_ON(offset >= PAGE_SIZE);
Ian Campbellf942dc22011-03-15 00:06:18 +0000375 BUG_ON(copy_off > MAX_BUFFER_OFFSET);
376
Ian Campbell6a8ed462012-10-10 03:48:42 +0000377 bytes = PAGE_SIZE - offset;
378
379 if (bytes > size)
380 bytes = size;
381
382 if (start_new_rx_buffer(copy_off, bytes, 0)) {
Ian Campbellf942dc22011-03-15 00:06:18 +0000383 count++;
384 copy_off = 0;
385 }
386
Ian Campbellf942dc22011-03-15 00:06:18 +0000387 if (copy_off + bytes > MAX_BUFFER_OFFSET)
388 bytes = MAX_BUFFER_OFFSET - copy_off;
389
390 copy_off += bytes;
Ian Campbell6a8ed462012-10-10 03:48:42 +0000391
392 offset += bytes;
Ian Campbellf942dc22011-03-15 00:06:18 +0000393 size -= bytes;
Ian Campbell6a8ed462012-10-10 03:48:42 +0000394
395 if (offset == PAGE_SIZE)
396 offset = 0;
Ian Campbellf942dc22011-03-15 00:06:18 +0000397 }
398 }
399 return count;
400}
401
402struct netrx_pending_operations {
403 unsigned copy_prod, copy_cons;
404 unsigned meta_prod, meta_cons;
405 struct gnttab_copy *copy;
406 struct netbk_rx_meta *meta;
407 int copy_off;
408 grant_ref_t copy_gref;
409};
410
411static struct netbk_rx_meta *get_next_rx_buffer(struct xenvif *vif,
412 struct netrx_pending_operations *npo)
413{
414 struct netbk_rx_meta *meta;
415 struct xen_netif_rx_request *req;
416
417 req = RING_GET_REQUEST(&vif->rx, vif->rx.req_cons++);
418
419 meta = npo->meta + npo->meta_prod++;
420 meta->gso_size = 0;
421 meta->size = 0;
422 meta->id = req->id;
423
424 npo->copy_off = 0;
425 npo->copy_gref = req->gref;
426
427 return meta;
428}
429
430/*
431 * Set up the grant operations for this fragment. If it's a flipping
432 * interface, we also set up the unmap request from here.
433 */
434static void netbk_gop_frag_copy(struct xenvif *vif, struct sk_buff *skb,
435 struct netrx_pending_operations *npo,
436 struct page *page, unsigned long size,
437 unsigned long offset, int *head)
438{
439 struct gnttab_copy *copy_gop;
440 struct netbk_rx_meta *meta;
441 /*
Wei Liue34c0242011-12-06 02:04:50 +0000442 * These variables are used iff get_page_ext returns true,
Ian Campbellf942dc22011-03-15 00:06:18 +0000443 * in which case they are guaranteed to be initialized.
444 */
445 unsigned int uninitialized_var(group), uninitialized_var(idx);
446 int foreign = get_page_ext(page, &group, &idx);
447 unsigned long bytes;
448
449 /* Data must not cross a page boundary. */
Ian Campbell6a8ed462012-10-10 03:48:42 +0000450 BUG_ON(size + offset > PAGE_SIZE<<compound_order(page));
Ian Campbellf942dc22011-03-15 00:06:18 +0000451
452 meta = npo->meta + npo->meta_prod - 1;
453
Ian Campbell6a8ed462012-10-10 03:48:42 +0000454 /* Skip unused frames from start of page */
455 page += offset >> PAGE_SHIFT;
456 offset &= ~PAGE_MASK;
457
Ian Campbellf942dc22011-03-15 00:06:18 +0000458 while (size > 0) {
Ian Campbell6a8ed462012-10-10 03:48:42 +0000459 BUG_ON(offset >= PAGE_SIZE);
Ian Campbellf942dc22011-03-15 00:06:18 +0000460 BUG_ON(npo->copy_off > MAX_BUFFER_OFFSET);
461
Ian Campbell6a8ed462012-10-10 03:48:42 +0000462 bytes = PAGE_SIZE - offset;
463
464 if (bytes > size)
465 bytes = size;
466
467 if (start_new_rx_buffer(npo->copy_off, bytes, *head)) {
Ian Campbellf942dc22011-03-15 00:06:18 +0000468 /*
469 * Netfront requires there to be some data in the head
470 * buffer.
471 */
472 BUG_ON(*head);
473
474 meta = get_next_rx_buffer(vif, npo);
475 }
476
Ian Campbellf942dc22011-03-15 00:06:18 +0000477 if (npo->copy_off + bytes > MAX_BUFFER_OFFSET)
478 bytes = MAX_BUFFER_OFFSET - npo->copy_off;
479
480 copy_gop = npo->copy + npo->copy_prod++;
481 copy_gop->flags = GNTCOPY_dest_gref;
482 if (foreign) {
483 struct xen_netbk *netbk = &xen_netbk[group];
484 struct pending_tx_info *src_pend;
485
486 src_pend = &netbk->pending_tx_info[idx];
487
488 copy_gop->source.domid = src_pend->vif->domid;
489 copy_gop->source.u.ref = src_pend->req.gref;
490 copy_gop->flags |= GNTCOPY_source_gref;
491 } else {
492 void *vaddr = page_address(page);
493 copy_gop->source.domid = DOMID_SELF;
494 copy_gop->source.u.gmfn = virt_to_mfn(vaddr);
495 }
496 copy_gop->source.offset = offset;
497 copy_gop->dest.domid = vif->domid;
498
499 copy_gop->dest.offset = npo->copy_off;
500 copy_gop->dest.u.ref = npo->copy_gref;
501 copy_gop->len = bytes;
502
503 npo->copy_off += bytes;
504 meta->size += bytes;
505
506 offset += bytes;
507 size -= bytes;
508
Ian Campbell6a8ed462012-10-10 03:48:42 +0000509 /* Next frame */
510 if (offset == PAGE_SIZE && size) {
511 BUG_ON(!PageCompound(page));
512 page++;
513 offset = 0;
514 }
515
Ian Campbellf942dc22011-03-15 00:06:18 +0000516 /* Leave a gap for the GSO descriptor. */
517 if (*head && skb_shinfo(skb)->gso_size && !vif->gso_prefix)
518 vif->rx.req_cons++;
519
520 *head = 0; /* There must be something in this buffer now. */
521
522 }
523}
524
525/*
526 * Prepare an SKB to be transmitted to the frontend.
527 *
528 * This function is responsible for allocating grant operations, meta
529 * structures, etc.
530 *
531 * It returns the number of meta structures consumed. The number of
532 * ring slots used is always equal to the number of meta slots used
533 * plus the number of GSO descriptors used. Currently, we use either
534 * zero GSO descriptors (for non-GSO packets) or one descriptor (for
535 * frontend-side LRO).
536 */
537static int netbk_gop_skb(struct sk_buff *skb,
538 struct netrx_pending_operations *npo)
539{
540 struct xenvif *vif = netdev_priv(skb->dev);
541 int nr_frags = skb_shinfo(skb)->nr_frags;
542 int i;
543 struct xen_netif_rx_request *req;
544 struct netbk_rx_meta *meta;
545 unsigned char *data;
546 int head = 1;
547 int old_meta_prod;
548
549 old_meta_prod = npo->meta_prod;
550
551 /* Set up a GSO prefix descriptor, if necessary */
552 if (skb_shinfo(skb)->gso_size && vif->gso_prefix) {
553 req = RING_GET_REQUEST(&vif->rx, vif->rx.req_cons++);
554 meta = npo->meta + npo->meta_prod++;
555 meta->gso_size = skb_shinfo(skb)->gso_size;
556 meta->size = 0;
557 meta->id = req->id;
558 }
559
560 req = RING_GET_REQUEST(&vif->rx, vif->rx.req_cons++);
561 meta = npo->meta + npo->meta_prod++;
562
563 if (!vif->gso_prefix)
564 meta->gso_size = skb_shinfo(skb)->gso_size;
565 else
566 meta->gso_size = 0;
567
568 meta->size = 0;
569 meta->id = req->id;
570 npo->copy_off = 0;
571 npo->copy_gref = req->gref;
572
573 data = skb->data;
574 while (data < skb_tail_pointer(skb)) {
575 unsigned int offset = offset_in_page(data);
576 unsigned int len = PAGE_SIZE - offset;
577
578 if (data + len > skb_tail_pointer(skb))
579 len = skb_tail_pointer(skb) - data;
580
581 netbk_gop_frag_copy(vif, skb, npo,
582 virt_to_page(data), len, offset, &head);
583 data += len;
584 }
585
586 for (i = 0; i < nr_frags; i++) {
587 netbk_gop_frag_copy(vif, skb, npo,
Ian Campbellea066ad2011-10-05 00:28:46 +0000588 skb_frag_page(&skb_shinfo(skb)->frags[i]),
Eric Dumazet9e903e02011-10-18 21:00:24 +0000589 skb_frag_size(&skb_shinfo(skb)->frags[i]),
Ian Campbellf942dc22011-03-15 00:06:18 +0000590 skb_shinfo(skb)->frags[i].page_offset,
591 &head);
592 }
593
594 return npo->meta_prod - old_meta_prod;
595}
596
597/*
598 * This is a twin to netbk_gop_skb. Assume that netbk_gop_skb was
599 * used to set up the operations on the top of
600 * netrx_pending_operations, which have since been done. Check that
601 * they didn't give any errors and advance over them.
602 */
603static int netbk_check_gop(struct xenvif *vif, int nr_meta_slots,
604 struct netrx_pending_operations *npo)
605{
606 struct gnttab_copy *copy_op;
607 int status = XEN_NETIF_RSP_OKAY;
608 int i;
609
610 for (i = 0; i < nr_meta_slots; i++) {
611 copy_op = npo->copy + npo->copy_cons++;
612 if (copy_op->status != GNTST_okay) {
613 netdev_dbg(vif->dev,
614 "Bad status %d from copy to DOM%d.\n",
615 copy_op->status, vif->domid);
616 status = XEN_NETIF_RSP_ERROR;
617 }
618 }
619
620 return status;
621}
622
623static void netbk_add_frag_responses(struct xenvif *vif, int status,
624 struct netbk_rx_meta *meta,
625 int nr_meta_slots)
626{
627 int i;
628 unsigned long offset;
629
630 /* No fragments used */
631 if (nr_meta_slots <= 1)
632 return;
633
634 nr_meta_slots--;
635
636 for (i = 0; i < nr_meta_slots; i++) {
637 int flags;
638 if (i == nr_meta_slots - 1)
639 flags = 0;
640 else
641 flags = XEN_NETRXF_more_data;
642
643 offset = 0;
644 make_rx_response(vif, meta[i].id, status, offset,
645 meta[i].size, flags);
646 }
647}
648
649struct skb_cb_overlay {
650 int meta_slots_used;
651};
652
653static void xen_netbk_rx_action(struct xen_netbk *netbk)
654{
655 struct xenvif *vif = NULL, *tmp;
656 s8 status;
657 u16 irq, flags;
658 struct xen_netif_rx_response *resp;
659 struct sk_buff_head rxq;
660 struct sk_buff *skb;
661 LIST_HEAD(notify);
662 int ret;
663 int nr_frags;
664 int count;
665 unsigned long offset;
666 struct skb_cb_overlay *sco;
667
668 struct netrx_pending_operations npo = {
669 .copy = netbk->grant_copy_op,
670 .meta = netbk->meta,
671 };
672
673 skb_queue_head_init(&rxq);
674
675 count = 0;
676
677 while ((skb = skb_dequeue(&netbk->rx_queue)) != NULL) {
678 vif = netdev_priv(skb->dev);
679 nr_frags = skb_shinfo(skb)->nr_frags;
680
681 sco = (struct skb_cb_overlay *)skb->cb;
682 sco->meta_slots_used = netbk_gop_skb(skb, &npo);
683
684 count += nr_frags + 1;
685
686 __skb_queue_tail(&rxq, skb);
687
688 /* Filled the batch queue? */
Wei Liu2810e5b2013-04-22 02:20:42 +0000689 /* XXX FIXME: RX path dependent on MAX_SKB_FRAGS */
Ian Campbellf942dc22011-03-15 00:06:18 +0000690 if (count + MAX_SKB_FRAGS >= XEN_NETIF_RX_RING_SIZE)
691 break;
692 }
693
694 BUG_ON(npo.meta_prod > ARRAY_SIZE(netbk->meta));
695
696 if (!npo.copy_prod)
697 return;
698
699 BUG_ON(npo.copy_prod > ARRAY_SIZE(netbk->grant_copy_op));
Andres Lagar-Cavillac5718982012-09-14 14:26:59 +0000700 gnttab_batch_copy(netbk->grant_copy_op, npo.copy_prod);
Ian Campbellf942dc22011-03-15 00:06:18 +0000701
702 while ((skb = __skb_dequeue(&rxq)) != NULL) {
703 sco = (struct skb_cb_overlay *)skb->cb;
704
705 vif = netdev_priv(skb->dev);
706
707 if (netbk->meta[npo.meta_cons].gso_size && vif->gso_prefix) {
708 resp = RING_GET_RESPONSE(&vif->rx,
709 vif->rx.rsp_prod_pvt++);
710
711 resp->flags = XEN_NETRXF_gso_prefix | XEN_NETRXF_more_data;
712
713 resp->offset = netbk->meta[npo.meta_cons].gso_size;
714 resp->id = netbk->meta[npo.meta_cons].id;
715 resp->status = sco->meta_slots_used;
716
717 npo.meta_cons++;
718 sco->meta_slots_used--;
719 }
720
721
722 vif->dev->stats.tx_bytes += skb->len;
723 vif->dev->stats.tx_packets++;
724
725 status = netbk_check_gop(vif, sco->meta_slots_used, &npo);
726
727 if (sco->meta_slots_used == 1)
728 flags = 0;
729 else
730 flags = XEN_NETRXF_more_data;
731
732 if (skb->ip_summed == CHECKSUM_PARTIAL) /* local packet? */
733 flags |= XEN_NETRXF_csum_blank | XEN_NETRXF_data_validated;
734 else if (skb->ip_summed == CHECKSUM_UNNECESSARY)
735 /* remote but checksummed. */
736 flags |= XEN_NETRXF_data_validated;
737
738 offset = 0;
739 resp = make_rx_response(vif, netbk->meta[npo.meta_cons].id,
740 status, offset,
741 netbk->meta[npo.meta_cons].size,
742 flags);
743
744 if (netbk->meta[npo.meta_cons].gso_size && !vif->gso_prefix) {
745 struct xen_netif_extra_info *gso =
746 (struct xen_netif_extra_info *)
747 RING_GET_RESPONSE(&vif->rx,
748 vif->rx.rsp_prod_pvt++);
749
750 resp->flags |= XEN_NETRXF_extra_info;
751
752 gso->u.gso.size = netbk->meta[npo.meta_cons].gso_size;
753 gso->u.gso.type = XEN_NETIF_GSO_TYPE_TCPV4;
754 gso->u.gso.pad = 0;
755 gso->u.gso.features = 0;
756
757 gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
758 gso->flags = 0;
759 }
760
761 netbk_add_frag_responses(vif, status,
762 netbk->meta + npo.meta_cons + 1,
763 sco->meta_slots_used);
764
765 RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&vif->rx, ret);
766 irq = vif->irq;
767 if (ret && list_empty(&vif->notify_list))
768 list_add_tail(&vif->notify_list, &notify);
769
770 xenvif_notify_tx_completion(vif);
771
772 xenvif_put(vif);
773 npo.meta_cons += sco->meta_slots_used;
774 dev_kfree_skb(skb);
775 }
776
777 list_for_each_entry_safe(vif, tmp, &notify, notify_list) {
778 notify_remote_via_irq(vif->irq);
779 list_del_init(&vif->notify_list);
780 }
781
782 /* More work to do? */
783 if (!skb_queue_empty(&netbk->rx_queue) &&
784 !timer_pending(&netbk->net_timer))
785 xen_netbk_kick_thread(netbk);
786}
787
788void xen_netbk_queue_tx_skb(struct xenvif *vif, struct sk_buff *skb)
789{
790 struct xen_netbk *netbk = vif->netbk;
791
792 skb_queue_tail(&netbk->rx_queue, skb);
793
794 xen_netbk_kick_thread(netbk);
795}
796
797static void xen_netbk_alarm(unsigned long data)
798{
799 struct xen_netbk *netbk = (struct xen_netbk *)data;
800 xen_netbk_kick_thread(netbk);
801}
802
803static int __on_net_schedule_list(struct xenvif *vif)
804{
805 return !list_empty(&vif->schedule_list);
806}
807
808/* Must be called with net_schedule_list_lock held */
809static void remove_from_net_schedule_list(struct xenvif *vif)
810{
811 if (likely(__on_net_schedule_list(vif))) {
812 list_del_init(&vif->schedule_list);
813 xenvif_put(vif);
814 }
815}
816
817static struct xenvif *poll_net_schedule_list(struct xen_netbk *netbk)
818{
819 struct xenvif *vif = NULL;
820
821 spin_lock_irq(&netbk->net_schedule_list_lock);
822 if (list_empty(&netbk->net_schedule_list))
823 goto out;
824
825 vif = list_first_entry(&netbk->net_schedule_list,
826 struct xenvif, schedule_list);
827 if (!vif)
828 goto out;
829
830 xenvif_get(vif);
831
832 remove_from_net_schedule_list(vif);
833out:
834 spin_unlock_irq(&netbk->net_schedule_list_lock);
835 return vif;
836}
837
838void xen_netbk_schedule_xenvif(struct xenvif *vif)
839{
840 unsigned long flags;
841 struct xen_netbk *netbk = vif->netbk;
842
843 if (__on_net_schedule_list(vif))
844 goto kick;
845
846 spin_lock_irqsave(&netbk->net_schedule_list_lock, flags);
847 if (!__on_net_schedule_list(vif) &&
848 likely(xenvif_schedulable(vif))) {
849 list_add_tail(&vif->schedule_list, &netbk->net_schedule_list);
850 xenvif_get(vif);
851 }
852 spin_unlock_irqrestore(&netbk->net_schedule_list_lock, flags);
853
854kick:
855 smp_mb();
856 if ((nr_pending_reqs(netbk) < (MAX_PENDING_REQS/2)) &&
857 !list_empty(&netbk->net_schedule_list))
858 xen_netbk_kick_thread(netbk);
859}
860
861void xen_netbk_deschedule_xenvif(struct xenvif *vif)
862{
863 struct xen_netbk *netbk = vif->netbk;
864 spin_lock_irq(&netbk->net_schedule_list_lock);
865 remove_from_net_schedule_list(vif);
866 spin_unlock_irq(&netbk->net_schedule_list_lock);
867}
868
869void xen_netbk_check_rx_xenvif(struct xenvif *vif)
870{
871 int more_to_do;
872
873 RING_FINAL_CHECK_FOR_REQUESTS(&vif->tx, more_to_do);
874
875 if (more_to_do)
876 xen_netbk_schedule_xenvif(vif);
877}
878
879static void tx_add_credit(struct xenvif *vif)
880{
881 unsigned long max_burst, max_credit;
882
883 /*
884 * Allow a burst big enough to transmit a jumbo packet of up to 128kB.
885 * Otherwise the interface can seize up due to insufficient credit.
886 */
887 max_burst = RING_GET_REQUEST(&vif->tx, vif->tx.req_cons)->size;
888 max_burst = min(max_burst, 131072UL);
889 max_burst = max(max_burst, vif->credit_bytes);
890
891 /* Take care that adding a new chunk of credit doesn't wrap to zero. */
892 max_credit = vif->remaining_credit + vif->credit_bytes;
893 if (max_credit < vif->remaining_credit)
894 max_credit = ULONG_MAX; /* wrapped: clamp to ULONG_MAX */
895
896 vif->remaining_credit = min(max_credit, max_burst);
897}
898
899static void tx_credit_callback(unsigned long data)
900{
901 struct xenvif *vif = (struct xenvif *)data;
902 tx_add_credit(vif);
903 xen_netbk_check_rx_xenvif(vif);
904}
905
906static void netbk_tx_err(struct xenvif *vif,
907 struct xen_netif_tx_request *txp, RING_IDX end)
908{
909 RING_IDX cons = vif->tx.req_cons;
910
911 do {
912 make_tx_response(vif, txp, XEN_NETIF_RSP_ERROR);
Ian Campbellb9149722013-02-06 23:41:38 +0000913 if (cons == end)
Ian Campbellf942dc22011-03-15 00:06:18 +0000914 break;
915 txp = RING_GET_REQUEST(&vif->tx, cons++);
916 } while (1);
917 vif->tx.req_cons = cons;
918 xen_netbk_check_rx_xenvif(vif);
919 xenvif_put(vif);
920}
921
Ian Campbell488562862013-02-06 23:41:35 +0000922static void netbk_fatal_tx_err(struct xenvif *vif)
923{
924 netdev_err(vif->dev, "fatal error; disabling device\n");
925 xenvif_carrier_off(vif);
David S. Miller629821d2013-02-19 13:04:34 -0500926 xenvif_put(vif);
Ian Campbell488562862013-02-06 23:41:35 +0000927}
928
Ian Campbellf942dc22011-03-15 00:06:18 +0000929static int netbk_count_requests(struct xenvif *vif,
930 struct xen_netif_tx_request *first,
Wei Liu2810e5b2013-04-22 02:20:42 +0000931 RING_IDX first_idx,
Ian Campbellf942dc22011-03-15 00:06:18 +0000932 struct xen_netif_tx_request *txp,
933 int work_to_do)
934{
935 RING_IDX cons = vif->tx.req_cons;
Wei Liu2810e5b2013-04-22 02:20:42 +0000936 int slots = 0;
937 int drop_err = 0;
Ian Campbellf942dc22011-03-15 00:06:18 +0000938
939 if (!(first->flags & XEN_NETTXF_more_data))
940 return 0;
941
942 do {
Wei Liu2810e5b2013-04-22 02:20:42 +0000943 if (slots >= work_to_do) {
944 netdev_err(vif->dev,
945 "Asked for %d slots but exceeds this limit\n",
946 work_to_do);
Ian Campbell488562862013-02-06 23:41:35 +0000947 netbk_fatal_tx_err(vif);
David Vrabel35876b52013-02-14 03:18:57 +0000948 return -ENODATA;
Ian Campbellf942dc22011-03-15 00:06:18 +0000949 }
950
Wei Liu2810e5b2013-04-22 02:20:42 +0000951 /* This guest is really using too many slots and
952 * considered malicious.
953 */
954 if (unlikely(slots >= max_skb_slots)) {
955 netdev_err(vif->dev,
956 "Malicious frontend using %d slots, threshold %u\n",
957 slots, max_skb_slots);
Ian Campbell488562862013-02-06 23:41:35 +0000958 netbk_fatal_tx_err(vif);
David Vrabel35876b52013-02-14 03:18:57 +0000959 return -E2BIG;
Ian Campbellf942dc22011-03-15 00:06:18 +0000960 }
961
Wei Liu2810e5b2013-04-22 02:20:42 +0000962 /* Xen network protocol had implicit dependency on
963 * MAX_SKB_FRAGS. XEN_NETIF_NR_SLOTS_MIN is set to the
964 * historical MAX_SKB_FRAGS value 18 to honor the same
965 * behavior as before. Any packet using more than 18
966 * slots but less than max_skb_slots slots is dropped
967 */
968 if (!drop_err && slots >= XEN_NETIF_NR_SLOTS_MIN) {
969 if (net_ratelimit())
970 netdev_dbg(vif->dev,
971 "Too many slots (%d) exceeding limit (%d), dropping packet\n",
972 slots, XEN_NETIF_NR_SLOTS_MIN);
973 drop_err = -E2BIG;
974 }
975
976 memcpy(txp, RING_GET_REQUEST(&vif->tx, cons + slots),
Ian Campbellf942dc22011-03-15 00:06:18 +0000977 sizeof(*txp));
Wei Liu03393fd52013-04-22 02:20:43 +0000978
979 /* If the guest submitted a frame >= 64 KiB then
980 * first->size overflowed and following slots will
981 * appear to be larger than the frame.
982 *
983 * This cannot be fatal error as there are buggy
984 * frontends that do this.
985 *
986 * Consume all slots and drop the packet.
987 */
988 if (!drop_err && txp->size > first->size) {
989 if (net_ratelimit())
990 netdev_dbg(vif->dev,
991 "Invalid tx request, slot size %u > remaining size %u\n",
992 txp->size, first->size);
993 drop_err = -EIO;
Ian Campbellf942dc22011-03-15 00:06:18 +0000994 }
995
996 first->size -= txp->size;
Wei Liu2810e5b2013-04-22 02:20:42 +0000997 slots++;
Ian Campbellf942dc22011-03-15 00:06:18 +0000998
999 if (unlikely((txp->offset + txp->size) > PAGE_SIZE)) {
Wei Liu2810e5b2013-04-22 02:20:42 +00001000 netdev_err(vif->dev, "Cross page boundary, txp->offset: %x, size: %u\n",
Ian Campbellf942dc22011-03-15 00:06:18 +00001001 txp->offset, txp->size);
Ian Campbell488562862013-02-06 23:41:35 +00001002 netbk_fatal_tx_err(vif);
David Vrabel35876b52013-02-14 03:18:57 +00001003 return -EINVAL;
Ian Campbellf942dc22011-03-15 00:06:18 +00001004 }
1005 } while ((txp++)->flags & XEN_NETTXF_more_data);
Wei Liu2810e5b2013-04-22 02:20:42 +00001006
1007 if (drop_err) {
1008 netbk_tx_err(vif, first, first_idx + slots);
1009 return drop_err;
1010 }
1011
1012 return slots;
Ian Campbellf942dc22011-03-15 00:06:18 +00001013}
1014
1015static struct page *xen_netbk_alloc_page(struct xen_netbk *netbk,
Ian Campbellea066ad2011-10-05 00:28:46 +00001016 u16 pending_idx)
Ian Campbellf942dc22011-03-15 00:06:18 +00001017{
1018 struct page *page;
1019 page = alloc_page(GFP_KERNEL|__GFP_COLD);
1020 if (!page)
1021 return NULL;
1022 set_page_ext(page, netbk, pending_idx);
1023 netbk->mmap_pages[pending_idx] = page;
1024 return page;
1025}
1026
1027static struct gnttab_copy *xen_netbk_get_requests(struct xen_netbk *netbk,
1028 struct xenvif *vif,
1029 struct sk_buff *skb,
1030 struct xen_netif_tx_request *txp,
1031 struct gnttab_copy *gop)
1032{
1033 struct skb_shared_info *shinfo = skb_shinfo(skb);
1034 skb_frag_t *frags = shinfo->frags;
Ian Campbellea066ad2011-10-05 00:28:46 +00001035 u16 pending_idx = *((u16 *)skb->data);
Wei Liu2810e5b2013-04-22 02:20:42 +00001036 u16 head_idx = 0;
1037 int slot, start;
1038 struct page *page;
1039 pending_ring_idx_t index, start_idx = 0;
1040 uint16_t dst_offset;
1041 unsigned int nr_slots;
1042 struct pending_tx_info *first = NULL;
1043
1044 /* At this point shinfo->nr_frags is in fact the number of
1045 * slots, which can be as large as XEN_NETIF_NR_SLOTS_MIN.
1046 */
1047 nr_slots = shinfo->nr_frags;
Ian Campbellf942dc22011-03-15 00:06:18 +00001048
1049 /* Skip first skb fragment if it is on same page as header fragment. */
Ian Campbellea066ad2011-10-05 00:28:46 +00001050 start = (frag_get_pending_idx(&shinfo->frags[0]) == pending_idx);
Ian Campbellf942dc22011-03-15 00:06:18 +00001051
Wei Liu2810e5b2013-04-22 02:20:42 +00001052 /* Coalesce tx requests, at this point the packet passed in
1053 * should be <= 64K. Any packets larger than 64K have been
1054 * handled in netbk_count_requests().
1055 */
1056 for (shinfo->nr_frags = slot = start; slot < nr_slots;
1057 shinfo->nr_frags++) {
Ian Campbellf942dc22011-03-15 00:06:18 +00001058 struct pending_tx_info *pending_tx_info =
1059 netbk->pending_tx_info;
1060
Wei Liu2810e5b2013-04-22 02:20:42 +00001061 page = alloc_page(GFP_KERNEL|__GFP_COLD);
Ian Campbellf942dc22011-03-15 00:06:18 +00001062 if (!page)
Ian Campbell4cc7c1c2013-02-06 23:41:37 +00001063 goto err;
Ian Campbellf942dc22011-03-15 00:06:18 +00001064
Wei Liu2810e5b2013-04-22 02:20:42 +00001065 dst_offset = 0;
1066 first = NULL;
1067 while (dst_offset < PAGE_SIZE && slot < nr_slots) {
1068 gop->flags = GNTCOPY_source_gref;
Ian Campbellf942dc22011-03-15 00:06:18 +00001069
Wei Liu2810e5b2013-04-22 02:20:42 +00001070 gop->source.u.ref = txp->gref;
1071 gop->source.domid = vif->domid;
1072 gop->source.offset = txp->offset;
Ian Campbellf942dc22011-03-15 00:06:18 +00001073
Wei Liu2810e5b2013-04-22 02:20:42 +00001074 gop->dest.domid = DOMID_SELF;
Ian Campbellf942dc22011-03-15 00:06:18 +00001075
Wei Liu2810e5b2013-04-22 02:20:42 +00001076 gop->dest.offset = dst_offset;
1077 gop->dest.u.gmfn = virt_to_mfn(page_address(page));
Ian Campbellf942dc22011-03-15 00:06:18 +00001078
Wei Liu2810e5b2013-04-22 02:20:42 +00001079 if (dst_offset + txp->size > PAGE_SIZE) {
1080 /* This page can only merge a portion
1081 * of tx request. Do not increment any
1082 * pointer / counter here. The txp
1083 * will be dealt with in future
1084 * rounds, eventually hitting the
1085 * `else` branch.
1086 */
1087 gop->len = PAGE_SIZE - dst_offset;
1088 txp->offset += gop->len;
1089 txp->size -= gop->len;
1090 dst_offset += gop->len; /* quit loop */
1091 } else {
1092 /* This tx request can be merged in the page */
1093 gop->len = txp->size;
1094 dst_offset += gop->len;
1095
1096 index = pending_index(netbk->pending_cons++);
1097
1098 pending_idx = netbk->pending_ring[index];
1099
1100 memcpy(&pending_tx_info[pending_idx].req, txp,
1101 sizeof(*txp));
1102 xenvif_get(vif);
1103
1104 pending_tx_info[pending_idx].vif = vif;
1105
1106 /* Poison these fields, corresponding
1107 * fields for head tx req will be set
1108 * to correct values after the loop.
1109 */
1110 netbk->mmap_pages[pending_idx] = (void *)(~0UL);
1111 pending_tx_info[pending_idx].head =
1112 INVALID_PENDING_RING_IDX;
1113
1114 if (!first) {
1115 first = &pending_tx_info[pending_idx];
1116 start_idx = index;
1117 head_idx = pending_idx;
1118 }
1119
1120 txp++;
1121 slot++;
1122 }
1123
1124 gop++;
1125 }
1126
1127 first->req.offset = 0;
1128 first->req.size = dst_offset;
1129 first->head = start_idx;
1130 set_page_ext(page, netbk, head_idx);
1131 netbk->mmap_pages[head_idx] = page;
1132 frag_set_pending_idx(&frags[shinfo->nr_frags], head_idx);
Ian Campbellf942dc22011-03-15 00:06:18 +00001133 }
1134
Wei Liu2810e5b2013-04-22 02:20:42 +00001135 BUG_ON(shinfo->nr_frags > MAX_SKB_FRAGS);
1136
Ian Campbellf942dc22011-03-15 00:06:18 +00001137 return gop;
Ian Campbell4cc7c1c2013-02-06 23:41:37 +00001138err:
1139 /* Unwind, freeing all pages and sending error responses. */
Wei Liu2810e5b2013-04-22 02:20:42 +00001140 while (shinfo->nr_frags-- > start) {
1141 xen_netbk_idx_release(netbk,
1142 frag_get_pending_idx(&frags[shinfo->nr_frags]),
1143 XEN_NETIF_RSP_ERROR);
Ian Campbell4cc7c1c2013-02-06 23:41:37 +00001144 }
1145 /* The head too, if necessary. */
1146 if (start)
1147 xen_netbk_idx_release(netbk, pending_idx, XEN_NETIF_RSP_ERROR);
1148
1149 return NULL;
Ian Campbellf942dc22011-03-15 00:06:18 +00001150}
1151
1152static int xen_netbk_tx_check_gop(struct xen_netbk *netbk,
1153 struct sk_buff *skb,
1154 struct gnttab_copy **gopp)
1155{
1156 struct gnttab_copy *gop = *gopp;
Ian Campbellea066ad2011-10-05 00:28:46 +00001157 u16 pending_idx = *((u16 *)skb->data);
Ian Campbellf942dc22011-03-15 00:06:18 +00001158 struct skb_shared_info *shinfo = skb_shinfo(skb);
Wei Liu2810e5b2013-04-22 02:20:42 +00001159 struct pending_tx_info *tx_info;
Ian Campbellf942dc22011-03-15 00:06:18 +00001160 int nr_frags = shinfo->nr_frags;
1161 int i, err, start;
Wei Liu2810e5b2013-04-22 02:20:42 +00001162 u16 peek; /* peek into next tx request */
Ian Campbellf942dc22011-03-15 00:06:18 +00001163
1164 /* Check status of header. */
1165 err = gop->status;
Matthew Daley7d5145d2013-02-06 23:41:36 +00001166 if (unlikely(err))
1167 xen_netbk_idx_release(netbk, pending_idx, XEN_NETIF_RSP_ERROR);
Ian Campbellf942dc22011-03-15 00:06:18 +00001168
1169 /* Skip first skb fragment if it is on same page as header fragment. */
Ian Campbellea066ad2011-10-05 00:28:46 +00001170 start = (frag_get_pending_idx(&shinfo->frags[0]) == pending_idx);
Ian Campbellf942dc22011-03-15 00:06:18 +00001171
1172 for (i = start; i < nr_frags; i++) {
1173 int j, newerr;
Wei Liu2810e5b2013-04-22 02:20:42 +00001174 pending_ring_idx_t head;
Ian Campbellf942dc22011-03-15 00:06:18 +00001175
Ian Campbellea066ad2011-10-05 00:28:46 +00001176 pending_idx = frag_get_pending_idx(&shinfo->frags[i]);
Wei Liu2810e5b2013-04-22 02:20:42 +00001177 tx_info = &netbk->pending_tx_info[pending_idx];
1178 head = tx_info->head;
Ian Campbellf942dc22011-03-15 00:06:18 +00001179
1180 /* Check error status: if okay then remember grant handle. */
Wei Liu2810e5b2013-04-22 02:20:42 +00001181 do {
1182 newerr = (++gop)->status;
1183 if (newerr)
1184 break;
1185 peek = netbk->pending_ring[pending_index(++head)];
1186 } while (!pending_tx_is_head(netbk, peek));
1187
Ian Campbellf942dc22011-03-15 00:06:18 +00001188 if (likely(!newerr)) {
1189 /* Had a previous error? Invalidate this fragment. */
1190 if (unlikely(err))
Matthew Daley7d5145d2013-02-06 23:41:36 +00001191 xen_netbk_idx_release(netbk, pending_idx, XEN_NETIF_RSP_OKAY);
Ian Campbellf942dc22011-03-15 00:06:18 +00001192 continue;
1193 }
1194
1195 /* Error on this fragment: respond to client with an error. */
Matthew Daley7d5145d2013-02-06 23:41:36 +00001196 xen_netbk_idx_release(netbk, pending_idx, XEN_NETIF_RSP_ERROR);
Ian Campbellf942dc22011-03-15 00:06:18 +00001197
1198 /* Not the first error? Preceding frags already invalidated. */
1199 if (err)
1200 continue;
1201
1202 /* First error: invalidate header and preceding fragments. */
1203 pending_idx = *((u16 *)skb->data);
Matthew Daley7d5145d2013-02-06 23:41:36 +00001204 xen_netbk_idx_release(netbk, pending_idx, XEN_NETIF_RSP_OKAY);
Ian Campbellf942dc22011-03-15 00:06:18 +00001205 for (j = start; j < i; j++) {
Jan Beulich5ccb3ea2011-11-18 05:42:05 +00001206 pending_idx = frag_get_pending_idx(&shinfo->frags[j]);
Matthew Daley7d5145d2013-02-06 23:41:36 +00001207 xen_netbk_idx_release(netbk, pending_idx, XEN_NETIF_RSP_OKAY);
Ian Campbellf942dc22011-03-15 00:06:18 +00001208 }
1209
1210 /* Remember the error: invalidate all subsequent fragments. */
1211 err = newerr;
1212 }
1213
1214 *gopp = gop + 1;
1215 return err;
1216}
1217
1218static void xen_netbk_fill_frags(struct xen_netbk *netbk, struct sk_buff *skb)
1219{
1220 struct skb_shared_info *shinfo = skb_shinfo(skb);
1221 int nr_frags = shinfo->nr_frags;
1222 int i;
1223
1224 for (i = 0; i < nr_frags; i++) {
1225 skb_frag_t *frag = shinfo->frags + i;
1226 struct xen_netif_tx_request *txp;
Ian Campbellea066ad2011-10-05 00:28:46 +00001227 struct page *page;
1228 u16 pending_idx;
Ian Campbellf942dc22011-03-15 00:06:18 +00001229
Ian Campbellea066ad2011-10-05 00:28:46 +00001230 pending_idx = frag_get_pending_idx(frag);
Ian Campbellf942dc22011-03-15 00:06:18 +00001231
1232 txp = &netbk->pending_tx_info[pending_idx].req;
Ian Campbellea066ad2011-10-05 00:28:46 +00001233 page = virt_to_page(idx_to_kaddr(netbk, pending_idx));
1234 __skb_fill_page_desc(skb, i, page, txp->offset, txp->size);
Ian Campbellf942dc22011-03-15 00:06:18 +00001235 skb->len += txp->size;
1236 skb->data_len += txp->size;
1237 skb->truesize += txp->size;
1238
1239 /* Take an extra reference to offset xen_netbk_idx_release */
1240 get_page(netbk->mmap_pages[pending_idx]);
Matthew Daley7d5145d2013-02-06 23:41:36 +00001241 xen_netbk_idx_release(netbk, pending_idx, XEN_NETIF_RSP_OKAY);
Ian Campbellf942dc22011-03-15 00:06:18 +00001242 }
1243}
1244
1245static int xen_netbk_get_extras(struct xenvif *vif,
1246 struct xen_netif_extra_info *extras,
1247 int work_to_do)
1248{
1249 struct xen_netif_extra_info extra;
1250 RING_IDX cons = vif->tx.req_cons;
1251
1252 do {
1253 if (unlikely(work_to_do-- <= 0)) {
Ian Campbell488562862013-02-06 23:41:35 +00001254 netdev_err(vif->dev, "Missing extra info\n");
1255 netbk_fatal_tx_err(vif);
Ian Campbellf942dc22011-03-15 00:06:18 +00001256 return -EBADR;
1257 }
1258
1259 memcpy(&extra, RING_GET_REQUEST(&vif->tx, cons),
1260 sizeof(extra));
1261 if (unlikely(!extra.type ||
1262 extra.type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
1263 vif->tx.req_cons = ++cons;
Ian Campbell488562862013-02-06 23:41:35 +00001264 netdev_err(vif->dev,
Ian Campbellf942dc22011-03-15 00:06:18 +00001265 "Invalid extra type: %d\n", extra.type);
Ian Campbell488562862013-02-06 23:41:35 +00001266 netbk_fatal_tx_err(vif);
Ian Campbellf942dc22011-03-15 00:06:18 +00001267 return -EINVAL;
1268 }
1269
1270 memcpy(&extras[extra.type - 1], &extra, sizeof(extra));
1271 vif->tx.req_cons = ++cons;
1272 } while (extra.flags & XEN_NETIF_EXTRA_FLAG_MORE);
1273
1274 return work_to_do;
1275}
1276
1277static int netbk_set_skb_gso(struct xenvif *vif,
1278 struct sk_buff *skb,
1279 struct xen_netif_extra_info *gso)
1280{
1281 if (!gso->u.gso.size) {
Ian Campbell488562862013-02-06 23:41:35 +00001282 netdev_err(vif->dev, "GSO size must not be zero.\n");
1283 netbk_fatal_tx_err(vif);
Ian Campbellf942dc22011-03-15 00:06:18 +00001284 return -EINVAL;
1285 }
1286
1287 /* Currently only TCPv4 S.O. is supported. */
1288 if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4) {
Ian Campbell488562862013-02-06 23:41:35 +00001289 netdev_err(vif->dev, "Bad GSO type %d.\n", gso->u.gso.type);
1290 netbk_fatal_tx_err(vif);
Ian Campbellf942dc22011-03-15 00:06:18 +00001291 return -EINVAL;
1292 }
1293
1294 skb_shinfo(skb)->gso_size = gso->u.gso.size;
1295 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1296
1297 /* Header must be checked, and gso_segs computed. */
1298 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1299 skb_shinfo(skb)->gso_segs = 0;
1300
1301 return 0;
1302}
1303
1304static int checksum_setup(struct xenvif *vif, struct sk_buff *skb)
1305{
1306 struct iphdr *iph;
Ian Campbellf942dc22011-03-15 00:06:18 +00001307 int err = -EPROTO;
1308 int recalculate_partial_csum = 0;
1309
1310 /*
1311 * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy
1312 * peers can fail to set NETRXF_csum_blank when sending a GSO
1313 * frame. In this case force the SKB to CHECKSUM_PARTIAL and
1314 * recalculate the partial checksum.
1315 */
1316 if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) {
1317 vif->rx_gso_checksum_fixup++;
1318 skb->ip_summed = CHECKSUM_PARTIAL;
1319 recalculate_partial_csum = 1;
1320 }
1321
1322 /* A non-CHECKSUM_PARTIAL SKB does not require setup. */
1323 if (skb->ip_summed != CHECKSUM_PARTIAL)
1324 return 0;
1325
1326 if (skb->protocol != htons(ETH_P_IP))
1327 goto out;
1328
1329 iph = (void *)skb->data;
Ian Campbellf942dc22011-03-15 00:06:18 +00001330 switch (iph->protocol) {
1331 case IPPROTO_TCP:
Jason Wangbea89332013-04-10 20:35:29 +00001332 if (!skb_partial_csum_set(skb, 4 * iph->ihl,
1333 offsetof(struct tcphdr, check)))
1334 goto out;
Ian Campbellf942dc22011-03-15 00:06:18 +00001335
1336 if (recalculate_partial_csum) {
Jason Wangbea89332013-04-10 20:35:29 +00001337 struct tcphdr *tcph = tcp_hdr(skb);
Ian Campbellf942dc22011-03-15 00:06:18 +00001338 tcph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
1339 skb->len - iph->ihl*4,
1340 IPPROTO_TCP, 0);
1341 }
1342 break;
1343 case IPPROTO_UDP:
Jason Wangbea89332013-04-10 20:35:29 +00001344 if (!skb_partial_csum_set(skb, 4 * iph->ihl,
1345 offsetof(struct udphdr, check)))
1346 goto out;
Ian Campbellf942dc22011-03-15 00:06:18 +00001347
1348 if (recalculate_partial_csum) {
Jason Wangbea89332013-04-10 20:35:29 +00001349 struct udphdr *udph = udp_hdr(skb);
Ian Campbellf942dc22011-03-15 00:06:18 +00001350 udph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
1351 skb->len - iph->ihl*4,
1352 IPPROTO_UDP, 0);
1353 }
1354 break;
1355 default:
1356 if (net_ratelimit())
1357 netdev_err(vif->dev,
1358 "Attempting to checksum a non-TCP/UDP packet, dropping a protocol %d packet\n",
1359 iph->protocol);
1360 goto out;
1361 }
1362
Ian Campbellf942dc22011-03-15 00:06:18 +00001363 err = 0;
1364
1365out:
1366 return err;
1367}
1368
1369static bool tx_credit_exceeded(struct xenvif *vif, unsigned size)
1370{
1371 unsigned long now = jiffies;
1372 unsigned long next_credit =
1373 vif->credit_timeout.expires +
1374 msecs_to_jiffies(vif->credit_usec / 1000);
1375
1376 /* Timer could already be pending in rare cases. */
1377 if (timer_pending(&vif->credit_timeout))
1378 return true;
1379
1380 /* Passed the point where we can replenish credit? */
1381 if (time_after_eq(now, next_credit)) {
1382 vif->credit_timeout.expires = now;
1383 tx_add_credit(vif);
1384 }
1385
1386 /* Still too big to send right now? Set a callback. */
1387 if (size > vif->remaining_credit) {
1388 vif->credit_timeout.data =
1389 (unsigned long)vif;
1390 vif->credit_timeout.function =
1391 tx_credit_callback;
1392 mod_timer(&vif->credit_timeout,
1393 next_credit);
1394
1395 return true;
1396 }
1397
1398 return false;
1399}
1400
1401static unsigned xen_netbk_tx_build_gops(struct xen_netbk *netbk)
1402{
1403 struct gnttab_copy *gop = netbk->tx_copy_ops, *request_gop;
1404 struct sk_buff *skb;
1405 int ret;
1406
Wei Liu2810e5b2013-04-22 02:20:42 +00001407 while ((nr_pending_reqs(netbk) + XEN_NETIF_NR_SLOTS_MIN
1408 < MAX_PENDING_REQS) &&
Ian Campbellf942dc22011-03-15 00:06:18 +00001409 !list_empty(&netbk->net_schedule_list)) {
1410 struct xenvif *vif;
1411 struct xen_netif_tx_request txreq;
Wei Liu2810e5b2013-04-22 02:20:42 +00001412 struct xen_netif_tx_request txfrags[max_skb_slots];
Ian Campbellf942dc22011-03-15 00:06:18 +00001413 struct page *page;
1414 struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX-1];
1415 u16 pending_idx;
1416 RING_IDX idx;
1417 int work_to_do;
1418 unsigned int data_len;
1419 pending_ring_idx_t index;
1420
1421 /* Get a netif from the list with work to do. */
1422 vif = poll_net_schedule_list(netbk);
Ian Campbell488562862013-02-06 23:41:35 +00001423 /* This can sometimes happen because the test of
1424 * list_empty(net_schedule_list) at the top of the
1425 * loop is unlocked. Just go back and have another
1426 * look.
1427 */
Ian Campbellf942dc22011-03-15 00:06:18 +00001428 if (!vif)
1429 continue;
1430
Ian Campbell488562862013-02-06 23:41:35 +00001431 if (vif->tx.sring->req_prod - vif->tx.req_cons >
1432 XEN_NETIF_TX_RING_SIZE) {
1433 netdev_err(vif->dev,
1434 "Impossible number of requests. "
1435 "req_prod %d, req_cons %d, size %ld\n",
1436 vif->tx.sring->req_prod, vif->tx.req_cons,
1437 XEN_NETIF_TX_RING_SIZE);
1438 netbk_fatal_tx_err(vif);
1439 continue;
1440 }
1441
Ian Campbellf942dc22011-03-15 00:06:18 +00001442 RING_FINAL_CHECK_FOR_REQUESTS(&vif->tx, work_to_do);
1443 if (!work_to_do) {
1444 xenvif_put(vif);
1445 continue;
1446 }
1447
1448 idx = vif->tx.req_cons;
1449 rmb(); /* Ensure that we see the request before we copy it. */
1450 memcpy(&txreq, RING_GET_REQUEST(&vif->tx, idx), sizeof(txreq));
1451
1452 /* Credit-based scheduling. */
1453 if (txreq.size > vif->remaining_credit &&
1454 tx_credit_exceeded(vif, txreq.size)) {
1455 xenvif_put(vif);
1456 continue;
1457 }
1458
1459 vif->remaining_credit -= txreq.size;
1460
1461 work_to_do--;
1462 vif->tx.req_cons = ++idx;
1463
1464 memset(extras, 0, sizeof(extras));
1465 if (txreq.flags & XEN_NETTXF_extra_info) {
1466 work_to_do = xen_netbk_get_extras(vif, extras,
1467 work_to_do);
1468 idx = vif->tx.req_cons;
Ian Campbell488562862013-02-06 23:41:35 +00001469 if (unlikely(work_to_do < 0))
Ian Campbellf942dc22011-03-15 00:06:18 +00001470 continue;
Ian Campbellf942dc22011-03-15 00:06:18 +00001471 }
1472
Wei Liu2810e5b2013-04-22 02:20:42 +00001473 ret = netbk_count_requests(vif, &txreq, idx,
1474 txfrags, work_to_do);
Ian Campbell488562862013-02-06 23:41:35 +00001475 if (unlikely(ret < 0))
Ian Campbellf942dc22011-03-15 00:06:18 +00001476 continue;
Ian Campbell488562862013-02-06 23:41:35 +00001477
Ian Campbellf942dc22011-03-15 00:06:18 +00001478 idx += ret;
1479
1480 if (unlikely(txreq.size < ETH_HLEN)) {
1481 netdev_dbg(vif->dev,
1482 "Bad packet size: %d\n", txreq.size);
1483 netbk_tx_err(vif, &txreq, idx);
1484 continue;
1485 }
1486
1487 /* No crossing a page as the payload mustn't fragment. */
1488 if (unlikely((txreq.offset + txreq.size) > PAGE_SIZE)) {
Ian Campbell488562862013-02-06 23:41:35 +00001489 netdev_err(vif->dev,
Ian Campbellf942dc22011-03-15 00:06:18 +00001490 "txreq.offset: %x, size: %u, end: %lu\n",
1491 txreq.offset, txreq.size,
1492 (txreq.offset&~PAGE_MASK) + txreq.size);
Ian Campbell488562862013-02-06 23:41:35 +00001493 netbk_fatal_tx_err(vif);
Ian Campbellf942dc22011-03-15 00:06:18 +00001494 continue;
1495 }
1496
1497 index = pending_index(netbk->pending_cons);
1498 pending_idx = netbk->pending_ring[index];
1499
1500 data_len = (txreq.size > PKT_PROT_LEN &&
Wei Liu2810e5b2013-04-22 02:20:42 +00001501 ret < XEN_NETIF_NR_SLOTS_MIN) ?
Ian Campbellf942dc22011-03-15 00:06:18 +00001502 PKT_PROT_LEN : txreq.size;
1503
1504 skb = alloc_skb(data_len + NET_SKB_PAD + NET_IP_ALIGN,
1505 GFP_ATOMIC | __GFP_NOWARN);
1506 if (unlikely(skb == NULL)) {
1507 netdev_dbg(vif->dev,
1508 "Can't allocate a skb in start_xmit.\n");
1509 netbk_tx_err(vif, &txreq, idx);
1510 break;
1511 }
1512
1513 /* Packets passed to netif_rx() must have some headroom. */
1514 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
1515
1516 if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) {
1517 struct xen_netif_extra_info *gso;
1518 gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1];
1519
1520 if (netbk_set_skb_gso(vif, skb, gso)) {
Ian Campbell488562862013-02-06 23:41:35 +00001521 /* Failure in netbk_set_skb_gso is fatal. */
Ian Campbellf942dc22011-03-15 00:06:18 +00001522 kfree_skb(skb);
Ian Campbellf942dc22011-03-15 00:06:18 +00001523 continue;
1524 }
1525 }
1526
1527 /* XXX could copy straight to head */
Wei Liu27f85222013-03-25 01:08:20 +00001528 page = xen_netbk_alloc_page(netbk, pending_idx);
Ian Campbellf942dc22011-03-15 00:06:18 +00001529 if (!page) {
1530 kfree_skb(skb);
1531 netbk_tx_err(vif, &txreq, idx);
1532 continue;
1533 }
1534
Ian Campbellf942dc22011-03-15 00:06:18 +00001535 gop->source.u.ref = txreq.gref;
1536 gop->source.domid = vif->domid;
1537 gop->source.offset = txreq.offset;
1538
1539 gop->dest.u.gmfn = virt_to_mfn(page_address(page));
1540 gop->dest.domid = DOMID_SELF;
1541 gop->dest.offset = txreq.offset;
1542
1543 gop->len = txreq.size;
1544 gop->flags = GNTCOPY_source_gref;
1545
1546 gop++;
1547
1548 memcpy(&netbk->pending_tx_info[pending_idx].req,
1549 &txreq, sizeof(txreq));
1550 netbk->pending_tx_info[pending_idx].vif = vif;
Wei Liu2810e5b2013-04-22 02:20:42 +00001551 netbk->pending_tx_info[pending_idx].head = index;
Ian Campbellf942dc22011-03-15 00:06:18 +00001552 *((u16 *)skb->data) = pending_idx;
1553
1554 __skb_put(skb, data_len);
1555
1556 skb_shinfo(skb)->nr_frags = ret;
1557 if (data_len < txreq.size) {
1558 skb_shinfo(skb)->nr_frags++;
Ian Campbellea066ad2011-10-05 00:28:46 +00001559 frag_set_pending_idx(&skb_shinfo(skb)->frags[0],
1560 pending_idx);
Ian Campbellf942dc22011-03-15 00:06:18 +00001561 } else {
Ian Campbellea066ad2011-10-05 00:28:46 +00001562 frag_set_pending_idx(&skb_shinfo(skb)->frags[0],
1563 INVALID_PENDING_IDX);
Ian Campbellf942dc22011-03-15 00:06:18 +00001564 }
1565
Ian Campbellf942dc22011-03-15 00:06:18 +00001566 netbk->pending_cons++;
1567
1568 request_gop = xen_netbk_get_requests(netbk, vif,
1569 skb, txfrags, gop);
1570 if (request_gop == NULL) {
1571 kfree_skb(skb);
1572 netbk_tx_err(vif, &txreq, idx);
1573 continue;
1574 }
1575 gop = request_gop;
1576
Annie Li1e0b6ea2012-06-27 00:46:58 +00001577 __skb_queue_tail(&netbk->tx_queue, skb);
1578
Ian Campbellf942dc22011-03-15 00:06:18 +00001579 vif->tx.req_cons = idx;
1580 xen_netbk_check_rx_xenvif(vif);
1581
1582 if ((gop-netbk->tx_copy_ops) >= ARRAY_SIZE(netbk->tx_copy_ops))
1583 break;
1584 }
1585
1586 return gop - netbk->tx_copy_ops;
1587}
1588
1589static void xen_netbk_tx_submit(struct xen_netbk *netbk)
1590{
1591 struct gnttab_copy *gop = netbk->tx_copy_ops;
1592 struct sk_buff *skb;
1593
1594 while ((skb = __skb_dequeue(&netbk->tx_queue)) != NULL) {
1595 struct xen_netif_tx_request *txp;
1596 struct xenvif *vif;
1597 u16 pending_idx;
1598 unsigned data_len;
1599
1600 pending_idx = *((u16 *)skb->data);
1601 vif = netbk->pending_tx_info[pending_idx].vif;
1602 txp = &netbk->pending_tx_info[pending_idx].req;
1603
1604 /* Check the remap error code. */
1605 if (unlikely(xen_netbk_tx_check_gop(netbk, skb, &gop))) {
1606 netdev_dbg(vif->dev, "netback grant failed.\n");
1607 skb_shinfo(skb)->nr_frags = 0;
1608 kfree_skb(skb);
1609 continue;
1610 }
1611
1612 data_len = skb->len;
1613 memcpy(skb->data,
1614 (void *)(idx_to_kaddr(netbk, pending_idx)|txp->offset),
1615 data_len);
1616 if (data_len < txp->size) {
1617 /* Append the packet payload as a fragment. */
1618 txp->offset += data_len;
1619 txp->size -= data_len;
1620 } else {
1621 /* Schedule a response immediately. */
Matthew Daley7d5145d2013-02-06 23:41:36 +00001622 xen_netbk_idx_release(netbk, pending_idx, XEN_NETIF_RSP_OKAY);
Ian Campbellf942dc22011-03-15 00:06:18 +00001623 }
1624
1625 if (txp->flags & XEN_NETTXF_csum_blank)
1626 skb->ip_summed = CHECKSUM_PARTIAL;
1627 else if (txp->flags & XEN_NETTXF_data_validated)
1628 skb->ip_summed = CHECKSUM_UNNECESSARY;
1629
1630 xen_netbk_fill_frags(netbk, skb);
1631
1632 /*
1633 * If the initial fragment was < PKT_PROT_LEN then
1634 * pull through some bytes from the other fragments to
1635 * increase the linear region to PKT_PROT_LEN bytes.
1636 */
1637 if (skb_headlen(skb) < PKT_PROT_LEN && skb_is_nonlinear(skb)) {
1638 int target = min_t(int, skb->len, PKT_PROT_LEN);
1639 __pskb_pull_tail(skb, target - skb_headlen(skb));
1640 }
1641
1642 skb->dev = vif->dev;
1643 skb->protocol = eth_type_trans(skb, skb->dev);
Jason Wangf9ca8f72013-03-25 20:19:58 +00001644 skb_reset_network_header(skb);
Ian Campbellf942dc22011-03-15 00:06:18 +00001645
1646 if (checksum_setup(vif, skb)) {
1647 netdev_dbg(vif->dev,
1648 "Can't setup checksum in net_tx_action\n");
1649 kfree_skb(skb);
1650 continue;
1651 }
1652
Jason Wang40893fd2013-03-26 23:11:22 +00001653 skb_probe_transport_header(skb, 0);
Jason Wangf9ca8f72013-03-25 20:19:58 +00001654
Ian Campbellf942dc22011-03-15 00:06:18 +00001655 vif->dev->stats.rx_bytes += skb->len;
1656 vif->dev->stats.rx_packets++;
1657
1658 xenvif_receive_skb(vif, skb);
1659 }
1660}
1661
1662/* Called after netfront has transmitted */
1663static void xen_netbk_tx_action(struct xen_netbk *netbk)
1664{
1665 unsigned nr_gops;
Ian Campbellf942dc22011-03-15 00:06:18 +00001666
1667 nr_gops = xen_netbk_tx_build_gops(netbk);
1668
1669 if (nr_gops == 0)
1670 return;
Andres Lagar-Cavillac5718982012-09-14 14:26:59 +00001671
1672 gnttab_batch_copy(netbk->tx_copy_ops, nr_gops);
Ian Campbellf942dc22011-03-15 00:06:18 +00001673
1674 xen_netbk_tx_submit(netbk);
Ian Campbellf942dc22011-03-15 00:06:18 +00001675}
1676
Matthew Daley7d5145d2013-02-06 23:41:36 +00001677static void xen_netbk_idx_release(struct xen_netbk *netbk, u16 pending_idx,
1678 u8 status)
Ian Campbellf942dc22011-03-15 00:06:18 +00001679{
1680 struct xenvif *vif;
1681 struct pending_tx_info *pending_tx_info;
Wei Liu2810e5b2013-04-22 02:20:42 +00001682 pending_ring_idx_t head;
1683 u16 peek; /* peek into next tx request */
1684
1685 BUG_ON(netbk->mmap_pages[pending_idx] == (void *)(~0UL));
Ian Campbellf942dc22011-03-15 00:06:18 +00001686
1687 /* Already complete? */
1688 if (netbk->mmap_pages[pending_idx] == NULL)
1689 return;
1690
1691 pending_tx_info = &netbk->pending_tx_info[pending_idx];
1692
1693 vif = pending_tx_info->vif;
Wei Liu2810e5b2013-04-22 02:20:42 +00001694 head = pending_tx_info->head;
Ian Campbellf942dc22011-03-15 00:06:18 +00001695
Wei Liu2810e5b2013-04-22 02:20:42 +00001696 BUG_ON(!pending_tx_is_head(netbk, head));
1697 BUG_ON(netbk->pending_ring[pending_index(head)] != pending_idx);
Ian Campbellf942dc22011-03-15 00:06:18 +00001698
Wei Liu2810e5b2013-04-22 02:20:42 +00001699 do {
1700 pending_ring_idx_t index;
1701 pending_ring_idx_t idx = pending_index(head);
1702 u16 info_idx = netbk->pending_ring[idx];
Ian Campbellf942dc22011-03-15 00:06:18 +00001703
Wei Liu2810e5b2013-04-22 02:20:42 +00001704 pending_tx_info = &netbk->pending_tx_info[info_idx];
1705 make_tx_response(vif, &pending_tx_info->req, status);
Ian Campbellf942dc22011-03-15 00:06:18 +00001706
Wei Liu2810e5b2013-04-22 02:20:42 +00001707 /* Setting any number other than
1708 * INVALID_PENDING_RING_IDX indicates this slot is
1709 * starting a new packet / ending a previous packet.
1710 */
1711 pending_tx_info->head = 0;
1712
1713 index = pending_index(netbk->pending_prod++);
1714 netbk->pending_ring[index] = netbk->pending_ring[info_idx];
1715
1716 xenvif_put(vif);
1717
1718 peek = netbk->pending_ring[pending_index(++head)];
1719
1720 } while (!pending_tx_is_head(netbk, peek));
1721
1722 netbk->mmap_pages[pending_idx]->mapping = 0;
Ian Campbellf942dc22011-03-15 00:06:18 +00001723 put_page(netbk->mmap_pages[pending_idx]);
1724 netbk->mmap_pages[pending_idx] = NULL;
1725}
1726
Wei Liu2810e5b2013-04-22 02:20:42 +00001727
Ian Campbellf942dc22011-03-15 00:06:18 +00001728static void make_tx_response(struct xenvif *vif,
1729 struct xen_netif_tx_request *txp,
1730 s8 st)
1731{
1732 RING_IDX i = vif->tx.rsp_prod_pvt;
1733 struct xen_netif_tx_response *resp;
1734 int notify;
1735
1736 resp = RING_GET_RESPONSE(&vif->tx, i);
1737 resp->id = txp->id;
1738 resp->status = st;
1739
1740 if (txp->flags & XEN_NETTXF_extra_info)
1741 RING_GET_RESPONSE(&vif->tx, ++i)->status = XEN_NETIF_RSP_NULL;
1742
1743 vif->tx.rsp_prod_pvt = ++i;
1744 RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&vif->tx, notify);
1745 if (notify)
1746 notify_remote_via_irq(vif->irq);
1747}
1748
1749static struct xen_netif_rx_response *make_rx_response(struct xenvif *vif,
1750 u16 id,
1751 s8 st,
1752 u16 offset,
1753 u16 size,
1754 u16 flags)
1755{
1756 RING_IDX i = vif->rx.rsp_prod_pvt;
1757 struct xen_netif_rx_response *resp;
1758
1759 resp = RING_GET_RESPONSE(&vif->rx, i);
1760 resp->offset = offset;
1761 resp->flags = flags;
1762 resp->id = id;
1763 resp->status = (s16)size;
1764 if (st < 0)
1765 resp->status = (s16)st;
1766
1767 vif->rx.rsp_prod_pvt = ++i;
1768
1769 return resp;
1770}
1771
1772static inline int rx_work_todo(struct xen_netbk *netbk)
1773{
1774 return !skb_queue_empty(&netbk->rx_queue);
1775}
1776
1777static inline int tx_work_todo(struct xen_netbk *netbk)
1778{
1779
Wei Liu2810e5b2013-04-22 02:20:42 +00001780 if ((nr_pending_reqs(netbk) + XEN_NETIF_NR_SLOTS_MIN
1781 < MAX_PENDING_REQS) &&
1782 !list_empty(&netbk->net_schedule_list))
Ian Campbellf942dc22011-03-15 00:06:18 +00001783 return 1;
1784
1785 return 0;
1786}
1787
1788static int xen_netbk_kthread(void *data)
1789{
1790 struct xen_netbk *netbk = data;
1791 while (!kthread_should_stop()) {
1792 wait_event_interruptible(netbk->wq,
1793 rx_work_todo(netbk) ||
1794 tx_work_todo(netbk) ||
1795 kthread_should_stop());
1796 cond_resched();
1797
1798 if (kthread_should_stop())
1799 break;
1800
1801 if (rx_work_todo(netbk))
1802 xen_netbk_rx_action(netbk);
1803
1804 if (tx_work_todo(netbk))
1805 xen_netbk_tx_action(netbk);
1806 }
1807
1808 return 0;
1809}
1810
1811void xen_netbk_unmap_frontend_rings(struct xenvif *vif)
1812{
David Vrabelc9d63692011-09-29 16:53:31 +01001813 if (vif->tx.sring)
1814 xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif),
1815 vif->tx.sring);
1816 if (vif->rx.sring)
1817 xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif),
1818 vif->rx.sring);
Ian Campbellf942dc22011-03-15 00:06:18 +00001819}
1820
1821int xen_netbk_map_frontend_rings(struct xenvif *vif,
1822 grant_ref_t tx_ring_ref,
1823 grant_ref_t rx_ring_ref)
1824{
David Vrabelc9d63692011-09-29 16:53:31 +01001825 void *addr;
Ian Campbellf942dc22011-03-15 00:06:18 +00001826 struct xen_netif_tx_sring *txs;
1827 struct xen_netif_rx_sring *rxs;
1828
1829 int err = -ENOMEM;
1830
David Vrabelc9d63692011-09-29 16:53:31 +01001831 err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(vif),
1832 tx_ring_ref, &addr);
1833 if (err)
Ian Campbellf942dc22011-03-15 00:06:18 +00001834 goto err;
1835
David Vrabelc9d63692011-09-29 16:53:31 +01001836 txs = (struct xen_netif_tx_sring *)addr;
Ian Campbellf942dc22011-03-15 00:06:18 +00001837 BACK_RING_INIT(&vif->tx, txs, PAGE_SIZE);
1838
David Vrabelc9d63692011-09-29 16:53:31 +01001839 err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(vif),
1840 rx_ring_ref, &addr);
1841 if (err)
Ian Campbellf942dc22011-03-15 00:06:18 +00001842 goto err;
Ian Campbellf942dc22011-03-15 00:06:18 +00001843
David Vrabelc9d63692011-09-29 16:53:31 +01001844 rxs = (struct xen_netif_rx_sring *)addr;
Ian Campbellf942dc22011-03-15 00:06:18 +00001845 BACK_RING_INIT(&vif->rx, rxs, PAGE_SIZE);
1846
David Vrabelc9d63692011-09-29 16:53:31 +01001847 vif->rx_req_cons_peek = 0;
1848
Ian Campbellf942dc22011-03-15 00:06:18 +00001849 return 0;
1850
1851err:
1852 xen_netbk_unmap_frontend_rings(vif);
1853 return err;
1854}
1855
1856static int __init netback_init(void)
1857{
1858 int i;
1859 int rc = 0;
1860 int group;
1861
Daniel De Graaf2a14b2442011-12-14 15:12:13 -05001862 if (!xen_domain())
Ian Campbellf942dc22011-03-15 00:06:18 +00001863 return -ENODEV;
1864
Wei Liu2810e5b2013-04-22 02:20:42 +00001865 if (max_skb_slots < XEN_NETIF_NR_SLOTS_MIN) {
1866 printk(KERN_INFO
1867 "xen-netback: max_skb_slots too small (%d), bump it to XEN_NETIF_NR_SLOTS_MIN (%d)\n",
1868 max_skb_slots, XEN_NETIF_NR_SLOTS_MIN);
1869 max_skb_slots = XEN_NETIF_NR_SLOTS_MIN;
1870 }
1871
Ian Campbellf942dc22011-03-15 00:06:18 +00001872 xen_netbk_group_nr = num_online_cpus();
1873 xen_netbk = vzalloc(sizeof(struct xen_netbk) * xen_netbk_group_nr);
Joe Perchese404dec2012-01-29 12:56:23 +00001874 if (!xen_netbk)
Ian Campbellf942dc22011-03-15 00:06:18 +00001875 return -ENOMEM;
Ian Campbellf942dc22011-03-15 00:06:18 +00001876
1877 for (group = 0; group < xen_netbk_group_nr; group++) {
1878 struct xen_netbk *netbk = &xen_netbk[group];
1879 skb_queue_head_init(&netbk->rx_queue);
1880 skb_queue_head_init(&netbk->tx_queue);
1881
1882 init_timer(&netbk->net_timer);
1883 netbk->net_timer.data = (unsigned long)netbk;
1884 netbk->net_timer.function = xen_netbk_alarm;
1885
1886 netbk->pending_cons = 0;
1887 netbk->pending_prod = MAX_PENDING_REQS;
1888 for (i = 0; i < MAX_PENDING_REQS; i++)
1889 netbk->pending_ring[i] = i;
1890
1891 init_waitqueue_head(&netbk->wq);
1892 netbk->task = kthread_create(xen_netbk_kthread,
1893 (void *)netbk,
1894 "netback/%u", group);
1895
1896 if (IS_ERR(netbk->task)) {
Wei Liu6b84bd12011-12-05 06:57:44 +00001897 printk(KERN_ALERT "kthread_create() fails at netback\n");
Ian Campbellf942dc22011-03-15 00:06:18 +00001898 del_timer(&netbk->net_timer);
1899 rc = PTR_ERR(netbk->task);
1900 goto failed_init;
1901 }
1902
1903 kthread_bind(netbk->task, group);
1904
1905 INIT_LIST_HEAD(&netbk->net_schedule_list);
1906
1907 spin_lock_init(&netbk->net_schedule_list_lock);
1908
1909 atomic_set(&netbk->netfront_count, 0);
1910
1911 wake_up_process(netbk->task);
1912 }
1913
1914 rc = xenvif_xenbus_init();
1915 if (rc)
1916 goto failed_init;
1917
1918 return 0;
1919
1920failed_init:
1921 while (--group >= 0) {
1922 struct xen_netbk *netbk = &xen_netbk[group];
1923 for (i = 0; i < MAX_PENDING_REQS; i++) {
1924 if (netbk->mmap_pages[i])
1925 __free_page(netbk->mmap_pages[i]);
1926 }
1927 del_timer(&netbk->net_timer);
1928 kthread_stop(netbk->task);
1929 }
1930 vfree(xen_netbk);
1931 return rc;
1932
1933}
1934
1935module_init(netback_init);
1936
1937MODULE_LICENSE("Dual BSD/GPL");
Bastian Blankf984cec2011-06-30 11:19:09 -07001938MODULE_ALIAS("xen-backend:vif");