Loading...
1/*
2 * Virtual network driver for conversing with remote driver backends.
3 *
4 * Copyright (c) 2002-2005, K A Fraser
5 * Copyright (c) 2005, XenSource Ltd
6 *
7 * This program is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU General Public License version 2
9 * as published by the Free Software Foundation; or, when distributed
10 * separately from the Linux kernel or incorporated into other
11 * software packages, subject to the following license:
12 *
13 * Permission is hereby granted, free of charge, to any person obtaining a copy
14 * of this source file (the "Software"), to deal in the Software without
15 * restriction, including without limitation the rights to use, copy, modify,
16 * merge, publish, distribute, sublicense, and/or sell copies of the Software,
17 * and to permit persons to whom the Software is furnished to do so, subject to
18 * the following conditions:
19 *
20 * The above copyright notice and this permission notice shall be included in
21 * all copies or substantial portions of the Software.
22 *
23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
24 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
25 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
26 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
27 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
28 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
29 * IN THE SOFTWARE.
30 */
31
32#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
33
34#include <linux/module.h>
35#include <linux/kernel.h>
36#include <linux/netdevice.h>
37#include <linux/etherdevice.h>
38#include <linux/skbuff.h>
39#include <linux/ethtool.h>
40#include <linux/if_ether.h>
41#include <net/tcp.h>
42#include <linux/udp.h>
43#include <linux/moduleparam.h>
44#include <linux/mm.h>
45#include <linux/slab.h>
46#include <net/ip.h>
47
48#include <asm/xen/page.h>
49#include <xen/xen.h>
50#include <xen/xenbus.h>
51#include <xen/events.h>
52#include <xen/page.h>
53#include <xen/platform_pci.h>
54#include <xen/grant_table.h>
55
56#include <xen/interface/io/netif.h>
57#include <xen/interface/memory.h>
58#include <xen/interface/grant_table.h>
59
60static const struct ethtool_ops xennet_ethtool_ops;
61
62struct netfront_cb {
63 int pull_to;
64};
65
66#define NETFRONT_SKB_CB(skb) ((struct netfront_cb *)((skb)->cb))
67
68#define RX_COPY_THRESHOLD 256
69
70#define GRANT_INVALID_REF 0
71
72#define NET_TX_RING_SIZE __CONST_RING_SIZE(xen_netif_tx, PAGE_SIZE)
73#define NET_RX_RING_SIZE __CONST_RING_SIZE(xen_netif_rx, PAGE_SIZE)
74#define TX_MAX_TARGET min_t(int, NET_TX_RING_SIZE, 256)
75
76struct netfront_stats {
77 u64 rx_packets;
78 u64 tx_packets;
79 u64 rx_bytes;
80 u64 tx_bytes;
81 struct u64_stats_sync syncp;
82};
83
84struct netfront_info {
85 struct list_head list;
86 struct net_device *netdev;
87
88 struct napi_struct napi;
89
90 /* Split event channels support, tx_* == rx_* when using
91 * single event channel.
92 */
93 unsigned int tx_evtchn, rx_evtchn;
94 unsigned int tx_irq, rx_irq;
95 /* Only used when split event channels support is enabled */
96 char tx_irq_name[IFNAMSIZ+4]; /* DEVNAME-tx */
97 char rx_irq_name[IFNAMSIZ+4]; /* DEVNAME-rx */
98
99 struct xenbus_device *xbdev;
100
101 spinlock_t tx_lock;
102 struct xen_netif_tx_front_ring tx;
103 int tx_ring_ref;
104
105 /*
106 * {tx,rx}_skbs store outstanding skbuffs. Free tx_skb entries
107 * are linked from tx_skb_freelist through skb_entry.link.
108 *
109 * NB. Freelist index entries are always going to be less than
110 * PAGE_OFFSET, whereas pointers to skbs will always be equal or
111 * greater than PAGE_OFFSET: we use this property to distinguish
112 * them.
113 */
114 union skb_entry {
115 struct sk_buff *skb;
116 unsigned long link;
117 } tx_skbs[NET_TX_RING_SIZE];
118 grant_ref_t gref_tx_head;
119 grant_ref_t grant_tx_ref[NET_TX_RING_SIZE];
120 struct page *grant_tx_page[NET_TX_RING_SIZE];
121 unsigned tx_skb_freelist;
122
123 spinlock_t rx_lock ____cacheline_aligned_in_smp;
124 struct xen_netif_rx_front_ring rx;
125 int rx_ring_ref;
126
127 /* Receive-ring batched refills. */
128#define RX_MIN_TARGET 8
129#define RX_DFL_MIN_TARGET 64
130#define RX_MAX_TARGET min_t(int, NET_RX_RING_SIZE, 256)
131 unsigned rx_min_target, rx_max_target, rx_target;
132 struct sk_buff_head rx_batch;
133
134 struct timer_list rx_refill_timer;
135
136 struct sk_buff *rx_skbs[NET_RX_RING_SIZE];
137 grant_ref_t gref_rx_head;
138 grant_ref_t grant_rx_ref[NET_RX_RING_SIZE];
139
140 unsigned long rx_pfn_array[NET_RX_RING_SIZE];
141 struct multicall_entry rx_mcl[NET_RX_RING_SIZE+1];
142 struct mmu_update rx_mmu[NET_RX_RING_SIZE];
143
144 /* Statistics */
145 struct netfront_stats __percpu *stats;
146
147 unsigned long rx_gso_checksum_fixup;
148};
149
150struct netfront_rx_info {
151 struct xen_netif_rx_response rx;
152 struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX - 1];
153};
154
155static void skb_entry_set_link(union skb_entry *list, unsigned short id)
156{
157 list->link = id;
158}
159
160static int skb_entry_is_link(const union skb_entry *list)
161{
162 BUILD_BUG_ON(sizeof(list->skb) != sizeof(list->link));
163 return (unsigned long)list->skb < PAGE_OFFSET;
164}
165
166/*
167 * Access macros for acquiring freeing slots in tx_skbs[].
168 */
169
170static void add_id_to_freelist(unsigned *head, union skb_entry *list,
171 unsigned short id)
172{
173 skb_entry_set_link(&list[id], *head);
174 *head = id;
175}
176
177static unsigned short get_id_from_freelist(unsigned *head,
178 union skb_entry *list)
179{
180 unsigned int id = *head;
181 *head = list[id].link;
182 return id;
183}
184
185static int xennet_rxidx(RING_IDX idx)
186{
187 return idx & (NET_RX_RING_SIZE - 1);
188}
189
190static struct sk_buff *xennet_get_rx_skb(struct netfront_info *np,
191 RING_IDX ri)
192{
193 int i = xennet_rxidx(ri);
194 struct sk_buff *skb = np->rx_skbs[i];
195 np->rx_skbs[i] = NULL;
196 return skb;
197}
198
199static grant_ref_t xennet_get_rx_ref(struct netfront_info *np,
200 RING_IDX ri)
201{
202 int i = xennet_rxidx(ri);
203 grant_ref_t ref = np->grant_rx_ref[i];
204 np->grant_rx_ref[i] = GRANT_INVALID_REF;
205 return ref;
206}
207
208#ifdef CONFIG_SYSFS
209static int xennet_sysfs_addif(struct net_device *netdev);
210static void xennet_sysfs_delif(struct net_device *netdev);
211#else /* !CONFIG_SYSFS */
212#define xennet_sysfs_addif(dev) (0)
213#define xennet_sysfs_delif(dev) do { } while (0)
214#endif
215
216static bool xennet_can_sg(struct net_device *dev)
217{
218 return dev->features & NETIF_F_SG;
219}
220
221
222static void rx_refill_timeout(unsigned long data)
223{
224 struct net_device *dev = (struct net_device *)data;
225 struct netfront_info *np = netdev_priv(dev);
226 napi_schedule(&np->napi);
227}
228
229static int netfront_tx_slot_available(struct netfront_info *np)
230{
231 return (np->tx.req_prod_pvt - np->tx.rsp_cons) <
232 (TX_MAX_TARGET - MAX_SKB_FRAGS - 2);
233}
234
235static void xennet_maybe_wake_tx(struct net_device *dev)
236{
237 struct netfront_info *np = netdev_priv(dev);
238
239 if (unlikely(netif_queue_stopped(dev)) &&
240 netfront_tx_slot_available(np) &&
241 likely(netif_running(dev)))
242 netif_wake_queue(dev);
243}
244
245static void xennet_alloc_rx_buffers(struct net_device *dev)
246{
247 unsigned short id;
248 struct netfront_info *np = netdev_priv(dev);
249 struct sk_buff *skb;
250 struct page *page;
251 int i, batch_target, notify;
252 RING_IDX req_prod = np->rx.req_prod_pvt;
253 grant_ref_t ref;
254 unsigned long pfn;
255 void *vaddr;
256 struct xen_netif_rx_request *req;
257
258 if (unlikely(!netif_carrier_ok(dev)))
259 return;
260
261 /*
262 * Allocate skbuffs greedily, even though we batch updates to the
263 * receive ring. This creates a less bursty demand on the memory
264 * allocator, so should reduce the chance of failed allocation requests
265 * both for ourself and for other kernel subsystems.
266 */
267 batch_target = np->rx_target - (req_prod - np->rx.rsp_cons);
268 for (i = skb_queue_len(&np->rx_batch); i < batch_target; i++) {
269 skb = __netdev_alloc_skb(dev, RX_COPY_THRESHOLD + NET_IP_ALIGN,
270 GFP_ATOMIC | __GFP_NOWARN);
271 if (unlikely(!skb))
272 goto no_skb;
273
274 /* Align ip header to a 16 bytes boundary */
275 skb_reserve(skb, NET_IP_ALIGN);
276
277 page = alloc_page(GFP_ATOMIC | __GFP_NOWARN);
278 if (!page) {
279 kfree_skb(skb);
280no_skb:
281 /* Could not allocate any skbuffs. Try again later. */
282 mod_timer(&np->rx_refill_timer,
283 jiffies + (HZ/10));
284
285 /* Any skbuffs queued for refill? Force them out. */
286 if (i != 0)
287 goto refill;
288 break;
289 }
290
291 skb_add_rx_frag(skb, 0, page, 0, 0, PAGE_SIZE);
292 __skb_queue_tail(&np->rx_batch, skb);
293 }
294
295 /* Is the batch large enough to be worthwhile? */
296 if (i < (np->rx_target/2)) {
297 if (req_prod > np->rx.sring->req_prod)
298 goto push;
299 return;
300 }
301
302 /* Adjust our fill target if we risked running out of buffers. */
303 if (((req_prod - np->rx.sring->rsp_prod) < (np->rx_target / 4)) &&
304 ((np->rx_target *= 2) > np->rx_max_target))
305 np->rx_target = np->rx_max_target;
306
307 refill:
308 for (i = 0; ; i++) {
309 skb = __skb_dequeue(&np->rx_batch);
310 if (skb == NULL)
311 break;
312
313 skb->dev = dev;
314
315 id = xennet_rxidx(req_prod + i);
316
317 BUG_ON(np->rx_skbs[id]);
318 np->rx_skbs[id] = skb;
319
320 ref = gnttab_claim_grant_reference(&np->gref_rx_head);
321 BUG_ON((signed short)ref < 0);
322 np->grant_rx_ref[id] = ref;
323
324 pfn = page_to_pfn(skb_frag_page(&skb_shinfo(skb)->frags[0]));
325 vaddr = page_address(skb_frag_page(&skb_shinfo(skb)->frags[0]));
326
327 req = RING_GET_REQUEST(&np->rx, req_prod + i);
328 gnttab_grant_foreign_access_ref(ref,
329 np->xbdev->otherend_id,
330 pfn_to_mfn(pfn),
331 0);
332
333 req->id = id;
334 req->gref = ref;
335 }
336
337 wmb(); /* barrier so backend seens requests */
338
339 /* Above is a suitable barrier to ensure backend will see requests. */
340 np->rx.req_prod_pvt = req_prod + i;
341 push:
342 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&np->rx, notify);
343 if (notify)
344 notify_remote_via_irq(np->rx_irq);
345}
346
347static int xennet_open(struct net_device *dev)
348{
349 struct netfront_info *np = netdev_priv(dev);
350
351 napi_enable(&np->napi);
352
353 spin_lock_bh(&np->rx_lock);
354 if (netif_carrier_ok(dev)) {
355 xennet_alloc_rx_buffers(dev);
356 np->rx.sring->rsp_event = np->rx.rsp_cons + 1;
357 if (RING_HAS_UNCONSUMED_RESPONSES(&np->rx))
358 napi_schedule(&np->napi);
359 }
360 spin_unlock_bh(&np->rx_lock);
361
362 netif_start_queue(dev);
363
364 return 0;
365}
366
367static void xennet_tx_buf_gc(struct net_device *dev)
368{
369 RING_IDX cons, prod;
370 unsigned short id;
371 struct netfront_info *np = netdev_priv(dev);
372 struct sk_buff *skb;
373
374 BUG_ON(!netif_carrier_ok(dev));
375
376 do {
377 prod = np->tx.sring->rsp_prod;
378 rmb(); /* Ensure we see responses up to 'rp'. */
379
380 for (cons = np->tx.rsp_cons; cons != prod; cons++) {
381 struct xen_netif_tx_response *txrsp;
382
383 txrsp = RING_GET_RESPONSE(&np->tx, cons);
384 if (txrsp->status == XEN_NETIF_RSP_NULL)
385 continue;
386
387 id = txrsp->id;
388 skb = np->tx_skbs[id].skb;
389 if (unlikely(gnttab_query_foreign_access(
390 np->grant_tx_ref[id]) != 0)) {
391 pr_alert("%s: warning -- grant still in use by backend domain\n",
392 __func__);
393 BUG();
394 }
395 gnttab_end_foreign_access_ref(
396 np->grant_tx_ref[id], GNTMAP_readonly);
397 gnttab_release_grant_reference(
398 &np->gref_tx_head, np->grant_tx_ref[id]);
399 np->grant_tx_ref[id] = GRANT_INVALID_REF;
400 np->grant_tx_page[id] = NULL;
401 add_id_to_freelist(&np->tx_skb_freelist, np->tx_skbs, id);
402 dev_kfree_skb_irq(skb);
403 }
404
405 np->tx.rsp_cons = prod;
406
407 /*
408 * Set a new event, then check for race with update of tx_cons.
409 * Note that it is essential to schedule a callback, no matter
410 * how few buffers are pending. Even if there is space in the
411 * transmit ring, higher layers may be blocked because too much
412 * data is outstanding: in such cases notification from Xen is
413 * likely to be the only kick that we'll get.
414 */
415 np->tx.sring->rsp_event =
416 prod + ((np->tx.sring->req_prod - prod) >> 1) + 1;
417 mb(); /* update shared area */
418 } while ((cons == prod) && (prod != np->tx.sring->rsp_prod));
419
420 xennet_maybe_wake_tx(dev);
421}
422
423static void xennet_make_frags(struct sk_buff *skb, struct net_device *dev,
424 struct xen_netif_tx_request *tx)
425{
426 struct netfront_info *np = netdev_priv(dev);
427 char *data = skb->data;
428 unsigned long mfn;
429 RING_IDX prod = np->tx.req_prod_pvt;
430 int frags = skb_shinfo(skb)->nr_frags;
431 unsigned int offset = offset_in_page(data);
432 unsigned int len = skb_headlen(skb);
433 unsigned int id;
434 grant_ref_t ref;
435 int i;
436
437 /* While the header overlaps a page boundary (including being
438 larger than a page), split it it into page-sized chunks. */
439 while (len > PAGE_SIZE - offset) {
440 tx->size = PAGE_SIZE - offset;
441 tx->flags |= XEN_NETTXF_more_data;
442 len -= tx->size;
443 data += tx->size;
444 offset = 0;
445
446 id = get_id_from_freelist(&np->tx_skb_freelist, np->tx_skbs);
447 np->tx_skbs[id].skb = skb_get(skb);
448 tx = RING_GET_REQUEST(&np->tx, prod++);
449 tx->id = id;
450 ref = gnttab_claim_grant_reference(&np->gref_tx_head);
451 BUG_ON((signed short)ref < 0);
452
453 mfn = virt_to_mfn(data);
454 gnttab_grant_foreign_access_ref(ref, np->xbdev->otherend_id,
455 mfn, GNTMAP_readonly);
456
457 np->grant_tx_page[id] = virt_to_page(data);
458 tx->gref = np->grant_tx_ref[id] = ref;
459 tx->offset = offset;
460 tx->size = len;
461 tx->flags = 0;
462 }
463
464 /* Grant backend access to each skb fragment page. */
465 for (i = 0; i < frags; i++) {
466 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
467 struct page *page = skb_frag_page(frag);
468
469 len = skb_frag_size(frag);
470 offset = frag->page_offset;
471
472 /* Data must not cross a page boundary. */
473 BUG_ON(len + offset > PAGE_SIZE<<compound_order(page));
474
475 /* Skip unused frames from start of page */
476 page += offset >> PAGE_SHIFT;
477 offset &= ~PAGE_MASK;
478
479 while (len > 0) {
480 unsigned long bytes;
481
482 BUG_ON(offset >= PAGE_SIZE);
483
484 bytes = PAGE_SIZE - offset;
485 if (bytes > len)
486 bytes = len;
487
488 tx->flags |= XEN_NETTXF_more_data;
489
490 id = get_id_from_freelist(&np->tx_skb_freelist,
491 np->tx_skbs);
492 np->tx_skbs[id].skb = skb_get(skb);
493 tx = RING_GET_REQUEST(&np->tx, prod++);
494 tx->id = id;
495 ref = gnttab_claim_grant_reference(&np->gref_tx_head);
496 BUG_ON((signed short)ref < 0);
497
498 mfn = pfn_to_mfn(page_to_pfn(page));
499 gnttab_grant_foreign_access_ref(ref,
500 np->xbdev->otherend_id,
501 mfn, GNTMAP_readonly);
502
503 np->grant_tx_page[id] = page;
504 tx->gref = np->grant_tx_ref[id] = ref;
505 tx->offset = offset;
506 tx->size = bytes;
507 tx->flags = 0;
508
509 offset += bytes;
510 len -= bytes;
511
512 /* Next frame */
513 if (offset == PAGE_SIZE && len) {
514 BUG_ON(!PageCompound(page));
515 page++;
516 offset = 0;
517 }
518 }
519 }
520
521 np->tx.req_prod_pvt = prod;
522}
523
524/*
525 * Count how many ring slots are required to send the frags of this
526 * skb. Each frag might be a compound page.
527 */
528static int xennet_count_skb_frag_slots(struct sk_buff *skb)
529{
530 int i, frags = skb_shinfo(skb)->nr_frags;
531 int pages = 0;
532
533 for (i = 0; i < frags; i++) {
534 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
535 unsigned long size = skb_frag_size(frag);
536 unsigned long offset = frag->page_offset;
537
538 /* Skip unused frames from start of page */
539 offset &= ~PAGE_MASK;
540
541 pages += PFN_UP(offset + size);
542 }
543
544 return pages;
545}
546
547static int xennet_start_xmit(struct sk_buff *skb, struct net_device *dev)
548{
549 unsigned short id;
550 struct netfront_info *np = netdev_priv(dev);
551 struct netfront_stats *stats = this_cpu_ptr(np->stats);
552 struct xen_netif_tx_request *tx;
553 char *data = skb->data;
554 RING_IDX i;
555 grant_ref_t ref;
556 unsigned long mfn;
557 int notify;
558 int slots;
559 unsigned int offset = offset_in_page(data);
560 unsigned int len = skb_headlen(skb);
561 unsigned long flags;
562
563 /* If skb->len is too big for wire format, drop skb and alert
564 * user about misconfiguration.
565 */
566 if (unlikely(skb->len > XEN_NETIF_MAX_TX_SIZE)) {
567 net_alert_ratelimited(
568 "xennet: skb->len = %u, too big for wire format\n",
569 skb->len);
570 goto drop;
571 }
572
573 slots = DIV_ROUND_UP(offset + len, PAGE_SIZE) +
574 xennet_count_skb_frag_slots(skb);
575 if (unlikely(slots > MAX_SKB_FRAGS + 1)) {
576 net_alert_ratelimited(
577 "xennet: skb rides the rocket: %d slots\n", slots);
578 goto drop;
579 }
580
581 spin_lock_irqsave(&np->tx_lock, flags);
582
583 if (unlikely(!netif_carrier_ok(dev) ||
584 (slots > 1 && !xennet_can_sg(dev)) ||
585 netif_needs_gso(skb, netif_skb_features(skb)))) {
586 spin_unlock_irqrestore(&np->tx_lock, flags);
587 goto drop;
588 }
589
590 i = np->tx.req_prod_pvt;
591
592 id = get_id_from_freelist(&np->tx_skb_freelist, np->tx_skbs);
593 np->tx_skbs[id].skb = skb;
594
595 tx = RING_GET_REQUEST(&np->tx, i);
596
597 tx->id = id;
598 ref = gnttab_claim_grant_reference(&np->gref_tx_head);
599 BUG_ON((signed short)ref < 0);
600 mfn = virt_to_mfn(data);
601 gnttab_grant_foreign_access_ref(
602 ref, np->xbdev->otherend_id, mfn, GNTMAP_readonly);
603 np->grant_tx_page[id] = virt_to_page(data);
604 tx->gref = np->grant_tx_ref[id] = ref;
605 tx->offset = offset;
606 tx->size = len;
607
608 tx->flags = 0;
609 if (skb->ip_summed == CHECKSUM_PARTIAL)
610 /* local packet? */
611 tx->flags |= XEN_NETTXF_csum_blank | XEN_NETTXF_data_validated;
612 else if (skb->ip_summed == CHECKSUM_UNNECESSARY)
613 /* remote but checksummed. */
614 tx->flags |= XEN_NETTXF_data_validated;
615
616 if (skb_shinfo(skb)->gso_size) {
617 struct xen_netif_extra_info *gso;
618
619 gso = (struct xen_netif_extra_info *)
620 RING_GET_REQUEST(&np->tx, ++i);
621
622 tx->flags |= XEN_NETTXF_extra_info;
623
624 gso->u.gso.size = skb_shinfo(skb)->gso_size;
625 gso->u.gso.type = (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) ?
626 XEN_NETIF_GSO_TYPE_TCPV6 :
627 XEN_NETIF_GSO_TYPE_TCPV4;
628 gso->u.gso.pad = 0;
629 gso->u.gso.features = 0;
630
631 gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
632 gso->flags = 0;
633 }
634
635 np->tx.req_prod_pvt = i + 1;
636
637 xennet_make_frags(skb, dev, tx);
638 tx->size = skb->len;
639
640 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&np->tx, notify);
641 if (notify)
642 notify_remote_via_irq(np->tx_irq);
643
644 u64_stats_update_begin(&stats->syncp);
645 stats->tx_bytes += skb->len;
646 stats->tx_packets++;
647 u64_stats_update_end(&stats->syncp);
648
649 /* Note: It is not safe to access skb after xennet_tx_buf_gc()! */
650 xennet_tx_buf_gc(dev);
651
652 if (!netfront_tx_slot_available(np))
653 netif_stop_queue(dev);
654
655 spin_unlock_irqrestore(&np->tx_lock, flags);
656
657 return NETDEV_TX_OK;
658
659 drop:
660 dev->stats.tx_dropped++;
661 dev_kfree_skb_any(skb);
662 return NETDEV_TX_OK;
663}
664
665static int xennet_close(struct net_device *dev)
666{
667 struct netfront_info *np = netdev_priv(dev);
668 netif_stop_queue(np->netdev);
669 napi_disable(&np->napi);
670 return 0;
671}
672
673static void xennet_move_rx_slot(struct netfront_info *np, struct sk_buff *skb,
674 grant_ref_t ref)
675{
676 int new = xennet_rxidx(np->rx.req_prod_pvt);
677
678 BUG_ON(np->rx_skbs[new]);
679 np->rx_skbs[new] = skb;
680 np->grant_rx_ref[new] = ref;
681 RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->id = new;
682 RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->gref = ref;
683 np->rx.req_prod_pvt++;
684}
685
686static int xennet_get_extras(struct netfront_info *np,
687 struct xen_netif_extra_info *extras,
688 RING_IDX rp)
689
690{
691 struct xen_netif_extra_info *extra;
692 struct device *dev = &np->netdev->dev;
693 RING_IDX cons = np->rx.rsp_cons;
694 int err = 0;
695
696 do {
697 struct sk_buff *skb;
698 grant_ref_t ref;
699
700 if (unlikely(cons + 1 == rp)) {
701 if (net_ratelimit())
702 dev_warn(dev, "Missing extra info\n");
703 err = -EBADR;
704 break;
705 }
706
707 extra = (struct xen_netif_extra_info *)
708 RING_GET_RESPONSE(&np->rx, ++cons);
709
710 if (unlikely(!extra->type ||
711 extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
712 if (net_ratelimit())
713 dev_warn(dev, "Invalid extra type: %d\n",
714 extra->type);
715 err = -EINVAL;
716 } else {
717 memcpy(&extras[extra->type - 1], extra,
718 sizeof(*extra));
719 }
720
721 skb = xennet_get_rx_skb(np, cons);
722 ref = xennet_get_rx_ref(np, cons);
723 xennet_move_rx_slot(np, skb, ref);
724 } while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE);
725
726 np->rx.rsp_cons = cons;
727 return err;
728}
729
730static int xennet_get_responses(struct netfront_info *np,
731 struct netfront_rx_info *rinfo, RING_IDX rp,
732 struct sk_buff_head *list)
733{
734 struct xen_netif_rx_response *rx = &rinfo->rx;
735 struct xen_netif_extra_info *extras = rinfo->extras;
736 struct device *dev = &np->netdev->dev;
737 RING_IDX cons = np->rx.rsp_cons;
738 struct sk_buff *skb = xennet_get_rx_skb(np, cons);
739 grant_ref_t ref = xennet_get_rx_ref(np, cons);
740 int max = MAX_SKB_FRAGS + (rx->status <= RX_COPY_THRESHOLD);
741 int slots = 1;
742 int err = 0;
743 unsigned long ret;
744
745 if (rx->flags & XEN_NETRXF_extra_info) {
746 err = xennet_get_extras(np, extras, rp);
747 cons = np->rx.rsp_cons;
748 }
749
750 for (;;) {
751 if (unlikely(rx->status < 0 ||
752 rx->offset + rx->status > PAGE_SIZE)) {
753 if (net_ratelimit())
754 dev_warn(dev, "rx->offset: %x, size: %u\n",
755 rx->offset, rx->status);
756 xennet_move_rx_slot(np, skb, ref);
757 err = -EINVAL;
758 goto next;
759 }
760
761 /*
762 * This definitely indicates a bug, either in this driver or in
763 * the backend driver. In future this should flag the bad
764 * situation to the system controller to reboot the backend.
765 */
766 if (ref == GRANT_INVALID_REF) {
767 if (net_ratelimit())
768 dev_warn(dev, "Bad rx response id %d.\n",
769 rx->id);
770 err = -EINVAL;
771 goto next;
772 }
773
774 ret = gnttab_end_foreign_access_ref(ref, 0);
775 BUG_ON(!ret);
776
777 gnttab_release_grant_reference(&np->gref_rx_head, ref);
778
779 __skb_queue_tail(list, skb);
780
781next:
782 if (!(rx->flags & XEN_NETRXF_more_data))
783 break;
784
785 if (cons + slots == rp) {
786 if (net_ratelimit())
787 dev_warn(dev, "Need more slots\n");
788 err = -ENOENT;
789 break;
790 }
791
792 rx = RING_GET_RESPONSE(&np->rx, cons + slots);
793 skb = xennet_get_rx_skb(np, cons + slots);
794 ref = xennet_get_rx_ref(np, cons + slots);
795 slots++;
796 }
797
798 if (unlikely(slots > max)) {
799 if (net_ratelimit())
800 dev_warn(dev, "Too many slots\n");
801 err = -E2BIG;
802 }
803
804 if (unlikely(err))
805 np->rx.rsp_cons = cons + slots;
806
807 return err;
808}
809
810static int xennet_set_skb_gso(struct sk_buff *skb,
811 struct xen_netif_extra_info *gso)
812{
813 if (!gso->u.gso.size) {
814 if (net_ratelimit())
815 pr_warn("GSO size must not be zero\n");
816 return -EINVAL;
817 }
818
819 if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4 &&
820 gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV6) {
821 if (net_ratelimit())
822 pr_warn("Bad GSO type %d\n", gso->u.gso.type);
823 return -EINVAL;
824 }
825
826 skb_shinfo(skb)->gso_size = gso->u.gso.size;
827 skb_shinfo(skb)->gso_type =
828 (gso->u.gso.type == XEN_NETIF_GSO_TYPE_TCPV4) ?
829 SKB_GSO_TCPV4 :
830 SKB_GSO_TCPV6;
831
832 /* Header must be checked, and gso_segs computed. */
833 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
834 skb_shinfo(skb)->gso_segs = 0;
835
836 return 0;
837}
838
839static RING_IDX xennet_fill_frags(struct netfront_info *np,
840 struct sk_buff *skb,
841 struct sk_buff_head *list)
842{
843 struct skb_shared_info *shinfo = skb_shinfo(skb);
844 RING_IDX cons = np->rx.rsp_cons;
845 struct sk_buff *nskb;
846
847 while ((nskb = __skb_dequeue(list))) {
848 struct xen_netif_rx_response *rx =
849 RING_GET_RESPONSE(&np->rx, ++cons);
850 skb_frag_t *nfrag = &skb_shinfo(nskb)->frags[0];
851
852 if (shinfo->nr_frags == MAX_SKB_FRAGS) {
853 unsigned int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
854
855 BUG_ON(pull_to <= skb_headlen(skb));
856 __pskb_pull_tail(skb, pull_to - skb_headlen(skb));
857 }
858 BUG_ON(shinfo->nr_frags >= MAX_SKB_FRAGS);
859
860 skb_add_rx_frag(skb, shinfo->nr_frags, skb_frag_page(nfrag),
861 rx->offset, rx->status, PAGE_SIZE);
862
863 skb_shinfo(nskb)->nr_frags = 0;
864 kfree_skb(nskb);
865 }
866
867 return cons;
868}
869
870static int checksum_setup(struct net_device *dev, struct sk_buff *skb)
871{
872 bool recalculate_partial_csum = false;
873
874 /*
875 * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy
876 * peers can fail to set NETRXF_csum_blank when sending a GSO
877 * frame. In this case force the SKB to CHECKSUM_PARTIAL and
878 * recalculate the partial checksum.
879 */
880 if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) {
881 struct netfront_info *np = netdev_priv(dev);
882 np->rx_gso_checksum_fixup++;
883 skb->ip_summed = CHECKSUM_PARTIAL;
884 recalculate_partial_csum = true;
885 }
886
887 /* A non-CHECKSUM_PARTIAL SKB does not require setup. */
888 if (skb->ip_summed != CHECKSUM_PARTIAL)
889 return 0;
890
891 return skb_checksum_setup(skb, recalculate_partial_csum);
892}
893
894static int handle_incoming_queue(struct net_device *dev,
895 struct sk_buff_head *rxq)
896{
897 struct netfront_info *np = netdev_priv(dev);
898 struct netfront_stats *stats = this_cpu_ptr(np->stats);
899 int packets_dropped = 0;
900 struct sk_buff *skb;
901
902 while ((skb = __skb_dequeue(rxq)) != NULL) {
903 int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
904
905 if (pull_to > skb_headlen(skb))
906 __pskb_pull_tail(skb, pull_to - skb_headlen(skb));
907
908 /* Ethernet work: Delayed to here as it peeks the header. */
909 skb->protocol = eth_type_trans(skb, dev);
910 skb_reset_network_header(skb);
911
912 if (checksum_setup(dev, skb)) {
913 kfree_skb(skb);
914 packets_dropped++;
915 dev->stats.rx_errors++;
916 continue;
917 }
918
919 u64_stats_update_begin(&stats->syncp);
920 stats->rx_packets++;
921 stats->rx_bytes += skb->len;
922 u64_stats_update_end(&stats->syncp);
923
924 /* Pass it up. */
925 napi_gro_receive(&np->napi, skb);
926 }
927
928 return packets_dropped;
929}
930
931static int xennet_poll(struct napi_struct *napi, int budget)
932{
933 struct netfront_info *np = container_of(napi, struct netfront_info, napi);
934 struct net_device *dev = np->netdev;
935 struct sk_buff *skb;
936 struct netfront_rx_info rinfo;
937 struct xen_netif_rx_response *rx = &rinfo.rx;
938 struct xen_netif_extra_info *extras = rinfo.extras;
939 RING_IDX i, rp;
940 int work_done;
941 struct sk_buff_head rxq;
942 struct sk_buff_head errq;
943 struct sk_buff_head tmpq;
944 unsigned long flags;
945 int err;
946
947 spin_lock(&np->rx_lock);
948
949 skb_queue_head_init(&rxq);
950 skb_queue_head_init(&errq);
951 skb_queue_head_init(&tmpq);
952
953 rp = np->rx.sring->rsp_prod;
954 rmb(); /* Ensure we see queued responses up to 'rp'. */
955
956 i = np->rx.rsp_cons;
957 work_done = 0;
958 while ((i != rp) && (work_done < budget)) {
959 memcpy(rx, RING_GET_RESPONSE(&np->rx, i), sizeof(*rx));
960 memset(extras, 0, sizeof(rinfo.extras));
961
962 err = xennet_get_responses(np, &rinfo, rp, &tmpq);
963
964 if (unlikely(err)) {
965err:
966 while ((skb = __skb_dequeue(&tmpq)))
967 __skb_queue_tail(&errq, skb);
968 dev->stats.rx_errors++;
969 i = np->rx.rsp_cons;
970 continue;
971 }
972
973 skb = __skb_dequeue(&tmpq);
974
975 if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) {
976 struct xen_netif_extra_info *gso;
977 gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1];
978
979 if (unlikely(xennet_set_skb_gso(skb, gso))) {
980 __skb_queue_head(&tmpq, skb);
981 np->rx.rsp_cons += skb_queue_len(&tmpq);
982 goto err;
983 }
984 }
985
986 NETFRONT_SKB_CB(skb)->pull_to = rx->status;
987 if (NETFRONT_SKB_CB(skb)->pull_to > RX_COPY_THRESHOLD)
988 NETFRONT_SKB_CB(skb)->pull_to = RX_COPY_THRESHOLD;
989
990 skb_shinfo(skb)->frags[0].page_offset = rx->offset;
991 skb_frag_size_set(&skb_shinfo(skb)->frags[0], rx->status);
992 skb->data_len = rx->status;
993 skb->len += rx->status;
994
995 i = xennet_fill_frags(np, skb, &tmpq);
996
997 if (rx->flags & XEN_NETRXF_csum_blank)
998 skb->ip_summed = CHECKSUM_PARTIAL;
999 else if (rx->flags & XEN_NETRXF_data_validated)
1000 skb->ip_summed = CHECKSUM_UNNECESSARY;
1001
1002 __skb_queue_tail(&rxq, skb);
1003
1004 np->rx.rsp_cons = ++i;
1005 work_done++;
1006 }
1007
1008 __skb_queue_purge(&errq);
1009
1010 work_done -= handle_incoming_queue(dev, &rxq);
1011
1012 /* If we get a callback with very few responses, reduce fill target. */
1013 /* NB. Note exponential increase, linear decrease. */
1014 if (((np->rx.req_prod_pvt - np->rx.sring->rsp_prod) >
1015 ((3*np->rx_target) / 4)) &&
1016 (--np->rx_target < np->rx_min_target))
1017 np->rx_target = np->rx_min_target;
1018
1019 xennet_alloc_rx_buffers(dev);
1020
1021 if (work_done < budget) {
1022 int more_to_do = 0;
1023
1024 napi_gro_flush(napi, false);
1025
1026 local_irq_save(flags);
1027
1028 RING_FINAL_CHECK_FOR_RESPONSES(&np->rx, more_to_do);
1029 if (!more_to_do)
1030 __napi_complete(napi);
1031
1032 local_irq_restore(flags);
1033 }
1034
1035 spin_unlock(&np->rx_lock);
1036
1037 return work_done;
1038}
1039
1040static int xennet_change_mtu(struct net_device *dev, int mtu)
1041{
1042 int max = xennet_can_sg(dev) ?
1043 XEN_NETIF_MAX_TX_SIZE - MAX_TCP_HEADER : ETH_DATA_LEN;
1044
1045 if (mtu > max)
1046 return -EINVAL;
1047 dev->mtu = mtu;
1048 return 0;
1049}
1050
1051static struct rtnl_link_stats64 *xennet_get_stats64(struct net_device *dev,
1052 struct rtnl_link_stats64 *tot)
1053{
1054 struct netfront_info *np = netdev_priv(dev);
1055 int cpu;
1056
1057 for_each_possible_cpu(cpu) {
1058 struct netfront_stats *stats = per_cpu_ptr(np->stats, cpu);
1059 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1060 unsigned int start;
1061
1062 do {
1063 start = u64_stats_fetch_begin_irq(&stats->syncp);
1064
1065 rx_packets = stats->rx_packets;
1066 tx_packets = stats->tx_packets;
1067 rx_bytes = stats->rx_bytes;
1068 tx_bytes = stats->tx_bytes;
1069 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1070
1071 tot->rx_packets += rx_packets;
1072 tot->tx_packets += tx_packets;
1073 tot->rx_bytes += rx_bytes;
1074 tot->tx_bytes += tx_bytes;
1075 }
1076
1077 tot->rx_errors = dev->stats.rx_errors;
1078 tot->tx_dropped = dev->stats.tx_dropped;
1079
1080 return tot;
1081}
1082
1083static void xennet_release_tx_bufs(struct netfront_info *np)
1084{
1085 struct sk_buff *skb;
1086 int i;
1087
1088 for (i = 0; i < NET_TX_RING_SIZE; i++) {
1089 /* Skip over entries which are actually freelist references */
1090 if (skb_entry_is_link(&np->tx_skbs[i]))
1091 continue;
1092
1093 skb = np->tx_skbs[i].skb;
1094 get_page(np->grant_tx_page[i]);
1095 gnttab_end_foreign_access(np->grant_tx_ref[i],
1096 GNTMAP_readonly,
1097 (unsigned long)page_address(np->grant_tx_page[i]));
1098 np->grant_tx_page[i] = NULL;
1099 np->grant_tx_ref[i] = GRANT_INVALID_REF;
1100 add_id_to_freelist(&np->tx_skb_freelist, np->tx_skbs, i);
1101 dev_kfree_skb_irq(skb);
1102 }
1103}
1104
1105static void xennet_release_rx_bufs(struct netfront_info *np)
1106{
1107 int id, ref;
1108
1109 spin_lock_bh(&np->rx_lock);
1110
1111 for (id = 0; id < NET_RX_RING_SIZE; id++) {
1112 struct sk_buff *skb;
1113 struct page *page;
1114
1115 skb = np->rx_skbs[id];
1116 if (!skb)
1117 continue;
1118
1119 ref = np->grant_rx_ref[id];
1120 if (ref == GRANT_INVALID_REF)
1121 continue;
1122
1123 page = skb_frag_page(&skb_shinfo(skb)->frags[0]);
1124
1125 /* gnttab_end_foreign_access() needs a page ref until
1126 * foreign access is ended (which may be deferred).
1127 */
1128 get_page(page);
1129 gnttab_end_foreign_access(ref, 0,
1130 (unsigned long)page_address(page));
1131 np->grant_rx_ref[id] = GRANT_INVALID_REF;
1132
1133 kfree_skb(skb);
1134 }
1135
1136 spin_unlock_bh(&np->rx_lock);
1137}
1138
1139static void xennet_uninit(struct net_device *dev)
1140{
1141 struct netfront_info *np = netdev_priv(dev);
1142 xennet_release_tx_bufs(np);
1143 xennet_release_rx_bufs(np);
1144 gnttab_free_grant_references(np->gref_tx_head);
1145 gnttab_free_grant_references(np->gref_rx_head);
1146}
1147
1148static netdev_features_t xennet_fix_features(struct net_device *dev,
1149 netdev_features_t features)
1150{
1151 struct netfront_info *np = netdev_priv(dev);
1152 int val;
1153
1154 if (features & NETIF_F_SG) {
1155 if (xenbus_scanf(XBT_NIL, np->xbdev->otherend, "feature-sg",
1156 "%d", &val) < 0)
1157 val = 0;
1158
1159 if (!val)
1160 features &= ~NETIF_F_SG;
1161 }
1162
1163 if (features & NETIF_F_IPV6_CSUM) {
1164 if (xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1165 "feature-ipv6-csum-offload", "%d", &val) < 0)
1166 val = 0;
1167
1168 if (!val)
1169 features &= ~NETIF_F_IPV6_CSUM;
1170 }
1171
1172 if (features & NETIF_F_TSO) {
1173 if (xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1174 "feature-gso-tcpv4", "%d", &val) < 0)
1175 val = 0;
1176
1177 if (!val)
1178 features &= ~NETIF_F_TSO;
1179 }
1180
1181 if (features & NETIF_F_TSO6) {
1182 if (xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1183 "feature-gso-tcpv6", "%d", &val) < 0)
1184 val = 0;
1185
1186 if (!val)
1187 features &= ~NETIF_F_TSO6;
1188 }
1189
1190 return features;
1191}
1192
1193static int xennet_set_features(struct net_device *dev,
1194 netdev_features_t features)
1195{
1196 if (!(features & NETIF_F_SG) && dev->mtu > ETH_DATA_LEN) {
1197 netdev_info(dev, "Reducing MTU because no SG offload");
1198 dev->mtu = ETH_DATA_LEN;
1199 }
1200
1201 return 0;
1202}
1203
1204static irqreturn_t xennet_tx_interrupt(int irq, void *dev_id)
1205{
1206 struct netfront_info *np = dev_id;
1207 struct net_device *dev = np->netdev;
1208 unsigned long flags;
1209
1210 spin_lock_irqsave(&np->tx_lock, flags);
1211 xennet_tx_buf_gc(dev);
1212 spin_unlock_irqrestore(&np->tx_lock, flags);
1213
1214 return IRQ_HANDLED;
1215}
1216
1217static irqreturn_t xennet_rx_interrupt(int irq, void *dev_id)
1218{
1219 struct netfront_info *np = dev_id;
1220 struct net_device *dev = np->netdev;
1221
1222 if (likely(netif_carrier_ok(dev) &&
1223 RING_HAS_UNCONSUMED_RESPONSES(&np->rx)))
1224 napi_schedule(&np->napi);
1225
1226 return IRQ_HANDLED;
1227}
1228
1229static irqreturn_t xennet_interrupt(int irq, void *dev_id)
1230{
1231 xennet_tx_interrupt(irq, dev_id);
1232 xennet_rx_interrupt(irq, dev_id);
1233 return IRQ_HANDLED;
1234}
1235
1236#ifdef CONFIG_NET_POLL_CONTROLLER
1237static void xennet_poll_controller(struct net_device *dev)
1238{
1239 xennet_interrupt(0, dev);
1240}
1241#endif
1242
1243static const struct net_device_ops xennet_netdev_ops = {
1244 .ndo_open = xennet_open,
1245 .ndo_uninit = xennet_uninit,
1246 .ndo_stop = xennet_close,
1247 .ndo_start_xmit = xennet_start_xmit,
1248 .ndo_change_mtu = xennet_change_mtu,
1249 .ndo_get_stats64 = xennet_get_stats64,
1250 .ndo_set_mac_address = eth_mac_addr,
1251 .ndo_validate_addr = eth_validate_addr,
1252 .ndo_fix_features = xennet_fix_features,
1253 .ndo_set_features = xennet_set_features,
1254#ifdef CONFIG_NET_POLL_CONTROLLER
1255 .ndo_poll_controller = xennet_poll_controller,
1256#endif
1257};
1258
1259static struct net_device *xennet_create_dev(struct xenbus_device *dev)
1260{
1261 int i, err;
1262 struct net_device *netdev;
1263 struct netfront_info *np;
1264
1265 netdev = alloc_etherdev(sizeof(struct netfront_info));
1266 if (!netdev)
1267 return ERR_PTR(-ENOMEM);
1268
1269 np = netdev_priv(netdev);
1270 np->xbdev = dev;
1271
1272 spin_lock_init(&np->tx_lock);
1273 spin_lock_init(&np->rx_lock);
1274
1275 skb_queue_head_init(&np->rx_batch);
1276 np->rx_target = RX_DFL_MIN_TARGET;
1277 np->rx_min_target = RX_DFL_MIN_TARGET;
1278 np->rx_max_target = RX_MAX_TARGET;
1279
1280 init_timer(&np->rx_refill_timer);
1281 np->rx_refill_timer.data = (unsigned long)netdev;
1282 np->rx_refill_timer.function = rx_refill_timeout;
1283
1284 err = -ENOMEM;
1285 np->stats = netdev_alloc_pcpu_stats(struct netfront_stats);
1286 if (np->stats == NULL)
1287 goto exit;
1288
1289 /* Initialise tx_skbs as a free chain containing every entry. */
1290 np->tx_skb_freelist = 0;
1291 for (i = 0; i < NET_TX_RING_SIZE; i++) {
1292 skb_entry_set_link(&np->tx_skbs[i], i+1);
1293 np->grant_tx_ref[i] = GRANT_INVALID_REF;
1294 np->grant_tx_page[i] = NULL;
1295 }
1296
1297 /* Clear out rx_skbs */
1298 for (i = 0; i < NET_RX_RING_SIZE; i++) {
1299 np->rx_skbs[i] = NULL;
1300 np->grant_rx_ref[i] = GRANT_INVALID_REF;
1301 }
1302
1303 /* A grant for every tx ring slot */
1304 if (gnttab_alloc_grant_references(TX_MAX_TARGET,
1305 &np->gref_tx_head) < 0) {
1306 pr_alert("can't alloc tx grant refs\n");
1307 err = -ENOMEM;
1308 goto exit_free_stats;
1309 }
1310 /* A grant for every rx ring slot */
1311 if (gnttab_alloc_grant_references(RX_MAX_TARGET,
1312 &np->gref_rx_head) < 0) {
1313 pr_alert("can't alloc rx grant refs\n");
1314 err = -ENOMEM;
1315 goto exit_free_tx;
1316 }
1317
1318 netdev->netdev_ops = &xennet_netdev_ops;
1319
1320 netif_napi_add(netdev, &np->napi, xennet_poll, 64);
1321 netdev->features = NETIF_F_IP_CSUM | NETIF_F_RXCSUM |
1322 NETIF_F_GSO_ROBUST;
1323 netdev->hw_features = NETIF_F_SG |
1324 NETIF_F_IPV6_CSUM |
1325 NETIF_F_TSO | NETIF_F_TSO6;
1326
1327 /*
1328 * Assume that all hw features are available for now. This set
1329 * will be adjusted by the call to netdev_update_features() in
1330 * xennet_connect() which is the earliest point where we can
1331 * negotiate with the backend regarding supported features.
1332 */
1333 netdev->features |= netdev->hw_features;
1334
1335 SET_ETHTOOL_OPS(netdev, &xennet_ethtool_ops);
1336 SET_NETDEV_DEV(netdev, &dev->dev);
1337
1338 netif_set_gso_max_size(netdev, XEN_NETIF_MAX_TX_SIZE - MAX_TCP_HEADER);
1339
1340 np->netdev = netdev;
1341
1342 netif_carrier_off(netdev);
1343
1344 return netdev;
1345
1346 exit_free_tx:
1347 gnttab_free_grant_references(np->gref_tx_head);
1348 exit_free_stats:
1349 free_percpu(np->stats);
1350 exit:
1351 free_netdev(netdev);
1352 return ERR_PTR(err);
1353}
1354
1355/**
1356 * Entry point to this code when a new device is created. Allocate the basic
1357 * structures and the ring buffers for communication with the backend, and
1358 * inform the backend of the appropriate details for those.
1359 */
1360static int netfront_probe(struct xenbus_device *dev,
1361 const struct xenbus_device_id *id)
1362{
1363 int err;
1364 struct net_device *netdev;
1365 struct netfront_info *info;
1366
1367 netdev = xennet_create_dev(dev);
1368 if (IS_ERR(netdev)) {
1369 err = PTR_ERR(netdev);
1370 xenbus_dev_fatal(dev, err, "creating netdev");
1371 return err;
1372 }
1373
1374 info = netdev_priv(netdev);
1375 dev_set_drvdata(&dev->dev, info);
1376
1377 err = register_netdev(info->netdev);
1378 if (err) {
1379 pr_warn("%s: register_netdev err=%d\n", __func__, err);
1380 goto fail;
1381 }
1382
1383 err = xennet_sysfs_addif(info->netdev);
1384 if (err) {
1385 unregister_netdev(info->netdev);
1386 pr_warn("%s: add sysfs failed err=%d\n", __func__, err);
1387 goto fail;
1388 }
1389
1390 return 0;
1391
1392 fail:
1393 free_netdev(netdev);
1394 dev_set_drvdata(&dev->dev, NULL);
1395 return err;
1396}
1397
1398static void xennet_end_access(int ref, void *page)
1399{
1400 /* This frees the page as a side-effect */
1401 if (ref != GRANT_INVALID_REF)
1402 gnttab_end_foreign_access(ref, 0, (unsigned long)page);
1403}
1404
1405static void xennet_disconnect_backend(struct netfront_info *info)
1406{
1407 /* Stop old i/f to prevent errors whilst we rebuild the state. */
1408 spin_lock_bh(&info->rx_lock);
1409 spin_lock_irq(&info->tx_lock);
1410 netif_carrier_off(info->netdev);
1411 spin_unlock_irq(&info->tx_lock);
1412 spin_unlock_bh(&info->rx_lock);
1413
1414 if (info->tx_irq && (info->tx_irq == info->rx_irq))
1415 unbind_from_irqhandler(info->tx_irq, info);
1416 if (info->tx_irq && (info->tx_irq != info->rx_irq)) {
1417 unbind_from_irqhandler(info->tx_irq, info);
1418 unbind_from_irqhandler(info->rx_irq, info);
1419 }
1420 info->tx_evtchn = info->rx_evtchn = 0;
1421 info->tx_irq = info->rx_irq = 0;
1422
1423 /* End access and free the pages */
1424 xennet_end_access(info->tx_ring_ref, info->tx.sring);
1425 xennet_end_access(info->rx_ring_ref, info->rx.sring);
1426
1427 info->tx_ring_ref = GRANT_INVALID_REF;
1428 info->rx_ring_ref = GRANT_INVALID_REF;
1429 info->tx.sring = NULL;
1430 info->rx.sring = NULL;
1431}
1432
1433/**
1434 * We are reconnecting to the backend, due to a suspend/resume, or a backend
1435 * driver restart. We tear down our netif structure and recreate it, but
1436 * leave the device-layer structures intact so that this is transparent to the
1437 * rest of the kernel.
1438 */
1439static int netfront_resume(struct xenbus_device *dev)
1440{
1441 struct netfront_info *info = dev_get_drvdata(&dev->dev);
1442
1443 dev_dbg(&dev->dev, "%s\n", dev->nodename);
1444
1445 xennet_disconnect_backend(info);
1446 return 0;
1447}
1448
1449static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[])
1450{
1451 char *s, *e, *macstr;
1452 int i;
1453
1454 macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL);
1455 if (IS_ERR(macstr))
1456 return PTR_ERR(macstr);
1457
1458 for (i = 0; i < ETH_ALEN; i++) {
1459 mac[i] = simple_strtoul(s, &e, 16);
1460 if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) {
1461 kfree(macstr);
1462 return -ENOENT;
1463 }
1464 s = e+1;
1465 }
1466
1467 kfree(macstr);
1468 return 0;
1469}
1470
1471static int setup_netfront_single(struct netfront_info *info)
1472{
1473 int err;
1474
1475 err = xenbus_alloc_evtchn(info->xbdev, &info->tx_evtchn);
1476 if (err < 0)
1477 goto fail;
1478
1479 err = bind_evtchn_to_irqhandler(info->tx_evtchn,
1480 xennet_interrupt,
1481 0, info->netdev->name, info);
1482 if (err < 0)
1483 goto bind_fail;
1484 info->rx_evtchn = info->tx_evtchn;
1485 info->rx_irq = info->tx_irq = err;
1486
1487 return 0;
1488
1489bind_fail:
1490 xenbus_free_evtchn(info->xbdev, info->tx_evtchn);
1491 info->tx_evtchn = 0;
1492fail:
1493 return err;
1494}
1495
1496static int setup_netfront_split(struct netfront_info *info)
1497{
1498 int err;
1499
1500 err = xenbus_alloc_evtchn(info->xbdev, &info->tx_evtchn);
1501 if (err < 0)
1502 goto fail;
1503 err = xenbus_alloc_evtchn(info->xbdev, &info->rx_evtchn);
1504 if (err < 0)
1505 goto alloc_rx_evtchn_fail;
1506
1507 snprintf(info->tx_irq_name, sizeof(info->tx_irq_name),
1508 "%s-tx", info->netdev->name);
1509 err = bind_evtchn_to_irqhandler(info->tx_evtchn,
1510 xennet_tx_interrupt,
1511 0, info->tx_irq_name, info);
1512 if (err < 0)
1513 goto bind_tx_fail;
1514 info->tx_irq = err;
1515
1516 snprintf(info->rx_irq_name, sizeof(info->rx_irq_name),
1517 "%s-rx", info->netdev->name);
1518 err = bind_evtchn_to_irqhandler(info->rx_evtchn,
1519 xennet_rx_interrupt,
1520 0, info->rx_irq_name, info);
1521 if (err < 0)
1522 goto bind_rx_fail;
1523 info->rx_irq = err;
1524
1525 return 0;
1526
1527bind_rx_fail:
1528 unbind_from_irqhandler(info->tx_irq, info);
1529 info->tx_irq = 0;
1530bind_tx_fail:
1531 xenbus_free_evtchn(info->xbdev, info->rx_evtchn);
1532 info->rx_evtchn = 0;
1533alloc_rx_evtchn_fail:
1534 xenbus_free_evtchn(info->xbdev, info->tx_evtchn);
1535 info->tx_evtchn = 0;
1536fail:
1537 return err;
1538}
1539
1540static int setup_netfront(struct xenbus_device *dev, struct netfront_info *info)
1541{
1542 struct xen_netif_tx_sring *txs;
1543 struct xen_netif_rx_sring *rxs;
1544 int err;
1545 struct net_device *netdev = info->netdev;
1546 unsigned int feature_split_evtchn;
1547
1548 info->tx_ring_ref = GRANT_INVALID_REF;
1549 info->rx_ring_ref = GRANT_INVALID_REF;
1550 info->rx.sring = NULL;
1551 info->tx.sring = NULL;
1552 netdev->irq = 0;
1553
1554 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1555 "feature-split-event-channels", "%u",
1556 &feature_split_evtchn);
1557 if (err < 0)
1558 feature_split_evtchn = 0;
1559
1560 err = xen_net_read_mac(dev, netdev->dev_addr);
1561 if (err) {
1562 xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename);
1563 goto fail;
1564 }
1565
1566 txs = (struct xen_netif_tx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1567 if (!txs) {
1568 err = -ENOMEM;
1569 xenbus_dev_fatal(dev, err, "allocating tx ring page");
1570 goto fail;
1571 }
1572 SHARED_RING_INIT(txs);
1573 FRONT_RING_INIT(&info->tx, txs, PAGE_SIZE);
1574
1575 err = xenbus_grant_ring(dev, virt_to_mfn(txs));
1576 if (err < 0)
1577 goto grant_tx_ring_fail;
1578
1579 info->tx_ring_ref = err;
1580 rxs = (struct xen_netif_rx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1581 if (!rxs) {
1582 err = -ENOMEM;
1583 xenbus_dev_fatal(dev, err, "allocating rx ring page");
1584 goto alloc_rx_ring_fail;
1585 }
1586 SHARED_RING_INIT(rxs);
1587 FRONT_RING_INIT(&info->rx, rxs, PAGE_SIZE);
1588
1589 err = xenbus_grant_ring(dev, virt_to_mfn(rxs));
1590 if (err < 0)
1591 goto grant_rx_ring_fail;
1592 info->rx_ring_ref = err;
1593
1594 if (feature_split_evtchn)
1595 err = setup_netfront_split(info);
1596 /* setup single event channel if
1597 * a) feature-split-event-channels == 0
1598 * b) feature-split-event-channels == 1 but failed to setup
1599 */
1600 if (!feature_split_evtchn || (feature_split_evtchn && err))
1601 err = setup_netfront_single(info);
1602
1603 if (err)
1604 goto alloc_evtchn_fail;
1605
1606 return 0;
1607
1608 /* If we fail to setup netfront, it is safe to just revoke access to
1609 * granted pages because backend is not accessing it at this point.
1610 */
1611alloc_evtchn_fail:
1612 gnttab_end_foreign_access_ref(info->rx_ring_ref, 0);
1613grant_rx_ring_fail:
1614 free_page((unsigned long)rxs);
1615alloc_rx_ring_fail:
1616 gnttab_end_foreign_access_ref(info->tx_ring_ref, 0);
1617grant_tx_ring_fail:
1618 free_page((unsigned long)txs);
1619fail:
1620 return err;
1621}
1622
1623/* Common code used when first setting up, and when resuming. */
1624static int talk_to_netback(struct xenbus_device *dev,
1625 struct netfront_info *info)
1626{
1627 const char *message;
1628 struct xenbus_transaction xbt;
1629 int err;
1630
1631 /* Create shared ring, alloc event channel. */
1632 err = setup_netfront(dev, info);
1633 if (err)
1634 goto out;
1635
1636again:
1637 err = xenbus_transaction_start(&xbt);
1638 if (err) {
1639 xenbus_dev_fatal(dev, err, "starting transaction");
1640 goto destroy_ring;
1641 }
1642
1643 err = xenbus_printf(xbt, dev->nodename, "tx-ring-ref", "%u",
1644 info->tx_ring_ref);
1645 if (err) {
1646 message = "writing tx ring-ref";
1647 goto abort_transaction;
1648 }
1649 err = xenbus_printf(xbt, dev->nodename, "rx-ring-ref", "%u",
1650 info->rx_ring_ref);
1651 if (err) {
1652 message = "writing rx ring-ref";
1653 goto abort_transaction;
1654 }
1655
1656 if (info->tx_evtchn == info->rx_evtchn) {
1657 err = xenbus_printf(xbt, dev->nodename,
1658 "event-channel", "%u", info->tx_evtchn);
1659 if (err) {
1660 message = "writing event-channel";
1661 goto abort_transaction;
1662 }
1663 } else {
1664 err = xenbus_printf(xbt, dev->nodename,
1665 "event-channel-tx", "%u", info->tx_evtchn);
1666 if (err) {
1667 message = "writing event-channel-tx";
1668 goto abort_transaction;
1669 }
1670 err = xenbus_printf(xbt, dev->nodename,
1671 "event-channel-rx", "%u", info->rx_evtchn);
1672 if (err) {
1673 message = "writing event-channel-rx";
1674 goto abort_transaction;
1675 }
1676 }
1677
1678 err = xenbus_printf(xbt, dev->nodename, "request-rx-copy", "%u",
1679 1);
1680 if (err) {
1681 message = "writing request-rx-copy";
1682 goto abort_transaction;
1683 }
1684
1685 err = xenbus_printf(xbt, dev->nodename, "feature-rx-notify", "%d", 1);
1686 if (err) {
1687 message = "writing feature-rx-notify";
1688 goto abort_transaction;
1689 }
1690
1691 err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", 1);
1692 if (err) {
1693 message = "writing feature-sg";
1694 goto abort_transaction;
1695 }
1696
1697 err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", "%d", 1);
1698 if (err) {
1699 message = "writing feature-gso-tcpv4";
1700 goto abort_transaction;
1701 }
1702
1703 err = xenbus_write(xbt, dev->nodename, "feature-gso-tcpv6", "1");
1704 if (err) {
1705 message = "writing feature-gso-tcpv6";
1706 goto abort_transaction;
1707 }
1708
1709 err = xenbus_write(xbt, dev->nodename, "feature-ipv6-csum-offload",
1710 "1");
1711 if (err) {
1712 message = "writing feature-ipv6-csum-offload";
1713 goto abort_transaction;
1714 }
1715
1716 err = xenbus_transaction_end(xbt, 0);
1717 if (err) {
1718 if (err == -EAGAIN)
1719 goto again;
1720 xenbus_dev_fatal(dev, err, "completing transaction");
1721 goto destroy_ring;
1722 }
1723
1724 return 0;
1725
1726 abort_transaction:
1727 xenbus_transaction_end(xbt, 1);
1728 xenbus_dev_fatal(dev, err, "%s", message);
1729 destroy_ring:
1730 xennet_disconnect_backend(info);
1731 out:
1732 return err;
1733}
1734
1735static int xennet_connect(struct net_device *dev)
1736{
1737 struct netfront_info *np = netdev_priv(dev);
1738 int i, requeue_idx, err;
1739 struct sk_buff *skb;
1740 grant_ref_t ref;
1741 struct xen_netif_rx_request *req;
1742 unsigned int feature_rx_copy;
1743
1744 err = xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1745 "feature-rx-copy", "%u", &feature_rx_copy);
1746 if (err != 1)
1747 feature_rx_copy = 0;
1748
1749 if (!feature_rx_copy) {
1750 dev_info(&dev->dev,
1751 "backend does not support copying receive path\n");
1752 return -ENODEV;
1753 }
1754
1755 err = talk_to_netback(np->xbdev, np);
1756 if (err)
1757 return err;
1758
1759 rtnl_lock();
1760 netdev_update_features(dev);
1761 rtnl_unlock();
1762
1763 spin_lock_bh(&np->rx_lock);
1764 spin_lock_irq(&np->tx_lock);
1765
1766 /* Step 1: Discard all pending TX packet fragments. */
1767 xennet_release_tx_bufs(np);
1768
1769 /* Step 2: Rebuild the RX buffer freelist and the RX ring itself. */
1770 for (requeue_idx = 0, i = 0; i < NET_RX_RING_SIZE; i++) {
1771 skb_frag_t *frag;
1772 const struct page *page;
1773 if (!np->rx_skbs[i])
1774 continue;
1775
1776 skb = np->rx_skbs[requeue_idx] = xennet_get_rx_skb(np, i);
1777 ref = np->grant_rx_ref[requeue_idx] = xennet_get_rx_ref(np, i);
1778 req = RING_GET_REQUEST(&np->rx, requeue_idx);
1779
1780 frag = &skb_shinfo(skb)->frags[0];
1781 page = skb_frag_page(frag);
1782 gnttab_grant_foreign_access_ref(
1783 ref, np->xbdev->otherend_id,
1784 pfn_to_mfn(page_to_pfn(page)),
1785 0);
1786 req->gref = ref;
1787 req->id = requeue_idx;
1788
1789 requeue_idx++;
1790 }
1791
1792 np->rx.req_prod_pvt = requeue_idx;
1793
1794 /*
1795 * Step 3: All public and private state should now be sane. Get
1796 * ready to start sending and receiving packets and give the driver
1797 * domain a kick because we've probably just requeued some
1798 * packets.
1799 */
1800 netif_carrier_on(np->netdev);
1801 notify_remote_via_irq(np->tx_irq);
1802 if (np->tx_irq != np->rx_irq)
1803 notify_remote_via_irq(np->rx_irq);
1804 xennet_tx_buf_gc(dev);
1805 xennet_alloc_rx_buffers(dev);
1806
1807 spin_unlock_irq(&np->tx_lock);
1808 spin_unlock_bh(&np->rx_lock);
1809
1810 return 0;
1811}
1812
1813/**
1814 * Callback received when the backend's state changes.
1815 */
1816static void netback_changed(struct xenbus_device *dev,
1817 enum xenbus_state backend_state)
1818{
1819 struct netfront_info *np = dev_get_drvdata(&dev->dev);
1820 struct net_device *netdev = np->netdev;
1821
1822 dev_dbg(&dev->dev, "%s\n", xenbus_strstate(backend_state));
1823
1824 switch (backend_state) {
1825 case XenbusStateInitialising:
1826 case XenbusStateInitialised:
1827 case XenbusStateReconfiguring:
1828 case XenbusStateReconfigured:
1829 case XenbusStateUnknown:
1830 break;
1831
1832 case XenbusStateInitWait:
1833 if (dev->state != XenbusStateInitialising)
1834 break;
1835 if (xennet_connect(netdev) != 0)
1836 break;
1837 xenbus_switch_state(dev, XenbusStateConnected);
1838 break;
1839
1840 case XenbusStateConnected:
1841 netdev_notify_peers(netdev);
1842 break;
1843
1844 case XenbusStateClosed:
1845 if (dev->state == XenbusStateClosed)
1846 break;
1847 /* Missed the backend's CLOSING state -- fallthrough */
1848 case XenbusStateClosing:
1849 xenbus_frontend_closed(dev);
1850 break;
1851 }
1852}
1853
1854static const struct xennet_stat {
1855 char name[ETH_GSTRING_LEN];
1856 u16 offset;
1857} xennet_stats[] = {
1858 {
1859 "rx_gso_checksum_fixup",
1860 offsetof(struct netfront_info, rx_gso_checksum_fixup)
1861 },
1862};
1863
1864static int xennet_get_sset_count(struct net_device *dev, int string_set)
1865{
1866 switch (string_set) {
1867 case ETH_SS_STATS:
1868 return ARRAY_SIZE(xennet_stats);
1869 default:
1870 return -EINVAL;
1871 }
1872}
1873
1874static void xennet_get_ethtool_stats(struct net_device *dev,
1875 struct ethtool_stats *stats, u64 * data)
1876{
1877 void *np = netdev_priv(dev);
1878 int i;
1879
1880 for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
1881 data[i] = *(unsigned long *)(np + xennet_stats[i].offset);
1882}
1883
1884static void xennet_get_strings(struct net_device *dev, u32 stringset, u8 * data)
1885{
1886 int i;
1887
1888 switch (stringset) {
1889 case ETH_SS_STATS:
1890 for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
1891 memcpy(data + i * ETH_GSTRING_LEN,
1892 xennet_stats[i].name, ETH_GSTRING_LEN);
1893 break;
1894 }
1895}
1896
1897static const struct ethtool_ops xennet_ethtool_ops =
1898{
1899 .get_link = ethtool_op_get_link,
1900
1901 .get_sset_count = xennet_get_sset_count,
1902 .get_ethtool_stats = xennet_get_ethtool_stats,
1903 .get_strings = xennet_get_strings,
1904};
1905
1906#ifdef CONFIG_SYSFS
1907static ssize_t show_rxbuf_min(struct device *dev,
1908 struct device_attribute *attr, char *buf)
1909{
1910 struct net_device *netdev = to_net_dev(dev);
1911 struct netfront_info *info = netdev_priv(netdev);
1912
1913 return sprintf(buf, "%u\n", info->rx_min_target);
1914}
1915
1916static ssize_t store_rxbuf_min(struct device *dev,
1917 struct device_attribute *attr,
1918 const char *buf, size_t len)
1919{
1920 struct net_device *netdev = to_net_dev(dev);
1921 struct netfront_info *np = netdev_priv(netdev);
1922 char *endp;
1923 unsigned long target;
1924
1925 if (!capable(CAP_NET_ADMIN))
1926 return -EPERM;
1927
1928 target = simple_strtoul(buf, &endp, 0);
1929 if (endp == buf)
1930 return -EBADMSG;
1931
1932 if (target < RX_MIN_TARGET)
1933 target = RX_MIN_TARGET;
1934 if (target > RX_MAX_TARGET)
1935 target = RX_MAX_TARGET;
1936
1937 spin_lock_bh(&np->rx_lock);
1938 if (target > np->rx_max_target)
1939 np->rx_max_target = target;
1940 np->rx_min_target = target;
1941 if (target > np->rx_target)
1942 np->rx_target = target;
1943
1944 xennet_alloc_rx_buffers(netdev);
1945
1946 spin_unlock_bh(&np->rx_lock);
1947 return len;
1948}
1949
1950static ssize_t show_rxbuf_max(struct device *dev,
1951 struct device_attribute *attr, char *buf)
1952{
1953 struct net_device *netdev = to_net_dev(dev);
1954 struct netfront_info *info = netdev_priv(netdev);
1955
1956 return sprintf(buf, "%u\n", info->rx_max_target);
1957}
1958
1959static ssize_t store_rxbuf_max(struct device *dev,
1960 struct device_attribute *attr,
1961 const char *buf, size_t len)
1962{
1963 struct net_device *netdev = to_net_dev(dev);
1964 struct netfront_info *np = netdev_priv(netdev);
1965 char *endp;
1966 unsigned long target;
1967
1968 if (!capable(CAP_NET_ADMIN))
1969 return -EPERM;
1970
1971 target = simple_strtoul(buf, &endp, 0);
1972 if (endp == buf)
1973 return -EBADMSG;
1974
1975 if (target < RX_MIN_TARGET)
1976 target = RX_MIN_TARGET;
1977 if (target > RX_MAX_TARGET)
1978 target = RX_MAX_TARGET;
1979
1980 spin_lock_bh(&np->rx_lock);
1981 if (target < np->rx_min_target)
1982 np->rx_min_target = target;
1983 np->rx_max_target = target;
1984 if (target < np->rx_target)
1985 np->rx_target = target;
1986
1987 xennet_alloc_rx_buffers(netdev);
1988
1989 spin_unlock_bh(&np->rx_lock);
1990 return len;
1991}
1992
1993static ssize_t show_rxbuf_cur(struct device *dev,
1994 struct device_attribute *attr, char *buf)
1995{
1996 struct net_device *netdev = to_net_dev(dev);
1997 struct netfront_info *info = netdev_priv(netdev);
1998
1999 return sprintf(buf, "%u\n", info->rx_target);
2000}
2001
2002static struct device_attribute xennet_attrs[] = {
2003 __ATTR(rxbuf_min, S_IRUGO|S_IWUSR, show_rxbuf_min, store_rxbuf_min),
2004 __ATTR(rxbuf_max, S_IRUGO|S_IWUSR, show_rxbuf_max, store_rxbuf_max),
2005 __ATTR(rxbuf_cur, S_IRUGO, show_rxbuf_cur, NULL),
2006};
2007
2008static int xennet_sysfs_addif(struct net_device *netdev)
2009{
2010 int i;
2011 int err;
2012
2013 for (i = 0; i < ARRAY_SIZE(xennet_attrs); i++) {
2014 err = device_create_file(&netdev->dev,
2015 &xennet_attrs[i]);
2016 if (err)
2017 goto fail;
2018 }
2019 return 0;
2020
2021 fail:
2022 while (--i >= 0)
2023 device_remove_file(&netdev->dev, &xennet_attrs[i]);
2024 return err;
2025}
2026
2027static void xennet_sysfs_delif(struct net_device *netdev)
2028{
2029 int i;
2030
2031 for (i = 0; i < ARRAY_SIZE(xennet_attrs); i++)
2032 device_remove_file(&netdev->dev, &xennet_attrs[i]);
2033}
2034
2035#endif /* CONFIG_SYSFS */
2036
2037static const struct xenbus_device_id netfront_ids[] = {
2038 { "vif" },
2039 { "" }
2040};
2041
2042
2043static int xennet_remove(struct xenbus_device *dev)
2044{
2045 struct netfront_info *info = dev_get_drvdata(&dev->dev);
2046
2047 dev_dbg(&dev->dev, "%s\n", dev->nodename);
2048
2049 xennet_disconnect_backend(info);
2050
2051 xennet_sysfs_delif(info->netdev);
2052
2053 unregister_netdev(info->netdev);
2054
2055 del_timer_sync(&info->rx_refill_timer);
2056
2057 free_percpu(info->stats);
2058
2059 free_netdev(info->netdev);
2060
2061 return 0;
2062}
2063
2064static DEFINE_XENBUS_DRIVER(netfront, ,
2065 .probe = netfront_probe,
2066 .remove = xennet_remove,
2067 .resume = netfront_resume,
2068 .otherend_changed = netback_changed,
2069);
2070
2071static int __init netif_init(void)
2072{
2073 if (!xen_domain())
2074 return -ENODEV;
2075
2076 if (!xen_has_pv_nic_devices())
2077 return -ENODEV;
2078
2079 pr_info("Initialising Xen virtual ethernet driver\n");
2080
2081 return xenbus_register_frontend(&netfront_driver);
2082}
2083module_init(netif_init);
2084
2085
2086static void __exit netif_exit(void)
2087{
2088 xenbus_unregister_driver(&netfront_driver);
2089}
2090module_exit(netif_exit);
2091
2092MODULE_DESCRIPTION("Xen virtual network device frontend");
2093MODULE_LICENSE("GPL");
2094MODULE_ALIAS("xen:vif");
2095MODULE_ALIAS("xennet");
1/*
2 * Virtual network driver for conversing with remote driver backends.
3 *
4 * Copyright (c) 2002-2005, K A Fraser
5 * Copyright (c) 2005, XenSource Ltd
6 *
7 * This program is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU General Public License version 2
9 * as published by the Free Software Foundation; or, when distributed
10 * separately from the Linux kernel or incorporated into other
11 * software packages, subject to the following license:
12 *
13 * Permission is hereby granted, free of charge, to any person obtaining a copy
14 * of this source file (the "Software"), to deal in the Software without
15 * restriction, including without limitation the rights to use, copy, modify,
16 * merge, publish, distribute, sublicense, and/or sell copies of the Software,
17 * and to permit persons to whom the Software is furnished to do so, subject to
18 * the following conditions:
19 *
20 * The above copyright notice and this permission notice shall be included in
21 * all copies or substantial portions of the Software.
22 *
23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
24 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
25 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
26 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
27 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
28 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
29 * IN THE SOFTWARE.
30 */
31
32#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
33
34#include <linux/module.h>
35#include <linux/kernel.h>
36#include <linux/netdevice.h>
37#include <linux/etherdevice.h>
38#include <linux/skbuff.h>
39#include <linux/ethtool.h>
40#include <linux/if_ether.h>
41#include <net/tcp.h>
42#include <linux/udp.h>
43#include <linux/moduleparam.h>
44#include <linux/mm.h>
45#include <linux/slab.h>
46#include <net/ip.h>
47
48#include <xen/xen.h>
49#include <xen/xenbus.h>
50#include <xen/events.h>
51#include <xen/page.h>
52#include <xen/platform_pci.h>
53#include <xen/grant_table.h>
54
55#include <xen/interface/io/netif.h>
56#include <xen/interface/memory.h>
57#include <xen/interface/grant_table.h>
58
59/* Module parameters */
60#define MAX_QUEUES_DEFAULT 8
61static unsigned int xennet_max_queues;
62module_param_named(max_queues, xennet_max_queues, uint, 0644);
63MODULE_PARM_DESC(max_queues,
64 "Maximum number of queues per virtual interface");
65
66static const struct ethtool_ops xennet_ethtool_ops;
67
68struct netfront_cb {
69 int pull_to;
70};
71
72#define NETFRONT_SKB_CB(skb) ((struct netfront_cb *)((skb)->cb))
73
74#define RX_COPY_THRESHOLD 256
75
76#define GRANT_INVALID_REF 0
77
78#define NET_TX_RING_SIZE __CONST_RING_SIZE(xen_netif_tx, XEN_PAGE_SIZE)
79#define NET_RX_RING_SIZE __CONST_RING_SIZE(xen_netif_rx, XEN_PAGE_SIZE)
80
81/* Minimum number of Rx slots (includes slot for GSO metadata). */
82#define NET_RX_SLOTS_MIN (XEN_NETIF_NR_SLOTS_MIN + 1)
83
84/* Queue name is interface name with "-qNNN" appended */
85#define QUEUE_NAME_SIZE (IFNAMSIZ + 6)
86
87/* IRQ name is queue name with "-tx" or "-rx" appended */
88#define IRQ_NAME_SIZE (QUEUE_NAME_SIZE + 3)
89
90static DECLARE_WAIT_QUEUE_HEAD(module_wq);
91
92struct netfront_stats {
93 u64 packets;
94 u64 bytes;
95 struct u64_stats_sync syncp;
96};
97
98struct netfront_info;
99
100struct netfront_queue {
101 unsigned int id; /* Queue ID, 0-based */
102 char name[QUEUE_NAME_SIZE]; /* DEVNAME-qN */
103 struct netfront_info *info;
104
105 struct napi_struct napi;
106
107 /* Split event channels support, tx_* == rx_* when using
108 * single event channel.
109 */
110 unsigned int tx_evtchn, rx_evtchn;
111 unsigned int tx_irq, rx_irq;
112 /* Only used when split event channels support is enabled */
113 char tx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-tx */
114 char rx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-rx */
115
116 spinlock_t tx_lock;
117 struct xen_netif_tx_front_ring tx;
118 int tx_ring_ref;
119
120 /*
121 * {tx,rx}_skbs store outstanding skbuffs. Free tx_skb entries
122 * are linked from tx_skb_freelist through skb_entry.link.
123 *
124 * NB. Freelist index entries are always going to be less than
125 * PAGE_OFFSET, whereas pointers to skbs will always be equal or
126 * greater than PAGE_OFFSET: we use this property to distinguish
127 * them.
128 */
129 union skb_entry {
130 struct sk_buff *skb;
131 unsigned long link;
132 } tx_skbs[NET_TX_RING_SIZE];
133 grant_ref_t gref_tx_head;
134 grant_ref_t grant_tx_ref[NET_TX_RING_SIZE];
135 struct page *grant_tx_page[NET_TX_RING_SIZE];
136 unsigned tx_skb_freelist;
137
138 spinlock_t rx_lock ____cacheline_aligned_in_smp;
139 struct xen_netif_rx_front_ring rx;
140 int rx_ring_ref;
141
142 struct timer_list rx_refill_timer;
143
144 struct sk_buff *rx_skbs[NET_RX_RING_SIZE];
145 grant_ref_t gref_rx_head;
146 grant_ref_t grant_rx_ref[NET_RX_RING_SIZE];
147};
148
149struct netfront_info {
150 struct list_head list;
151 struct net_device *netdev;
152
153 struct xenbus_device *xbdev;
154
155 /* Multi-queue support */
156 struct netfront_queue *queues;
157
158 /* Statistics */
159 struct netfront_stats __percpu *rx_stats;
160 struct netfront_stats __percpu *tx_stats;
161
162 atomic_t rx_gso_checksum_fixup;
163};
164
165struct netfront_rx_info {
166 struct xen_netif_rx_response rx;
167 struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX - 1];
168};
169
170static void skb_entry_set_link(union skb_entry *list, unsigned short id)
171{
172 list->link = id;
173}
174
175static int skb_entry_is_link(const union skb_entry *list)
176{
177 BUILD_BUG_ON(sizeof(list->skb) != sizeof(list->link));
178 return (unsigned long)list->skb < PAGE_OFFSET;
179}
180
181/*
182 * Access macros for acquiring freeing slots in tx_skbs[].
183 */
184
185static void add_id_to_freelist(unsigned *head, union skb_entry *list,
186 unsigned short id)
187{
188 skb_entry_set_link(&list[id], *head);
189 *head = id;
190}
191
192static unsigned short get_id_from_freelist(unsigned *head,
193 union skb_entry *list)
194{
195 unsigned int id = *head;
196 *head = list[id].link;
197 return id;
198}
199
200static int xennet_rxidx(RING_IDX idx)
201{
202 return idx & (NET_RX_RING_SIZE - 1);
203}
204
205static struct sk_buff *xennet_get_rx_skb(struct netfront_queue *queue,
206 RING_IDX ri)
207{
208 int i = xennet_rxidx(ri);
209 struct sk_buff *skb = queue->rx_skbs[i];
210 queue->rx_skbs[i] = NULL;
211 return skb;
212}
213
214static grant_ref_t xennet_get_rx_ref(struct netfront_queue *queue,
215 RING_IDX ri)
216{
217 int i = xennet_rxidx(ri);
218 grant_ref_t ref = queue->grant_rx_ref[i];
219 queue->grant_rx_ref[i] = GRANT_INVALID_REF;
220 return ref;
221}
222
223#ifdef CONFIG_SYSFS
224static const struct attribute_group xennet_dev_group;
225#endif
226
227static bool xennet_can_sg(struct net_device *dev)
228{
229 return dev->features & NETIF_F_SG;
230}
231
232
233static void rx_refill_timeout(struct timer_list *t)
234{
235 struct netfront_queue *queue = from_timer(queue, t, rx_refill_timer);
236 napi_schedule(&queue->napi);
237}
238
239static int netfront_tx_slot_available(struct netfront_queue *queue)
240{
241 return (queue->tx.req_prod_pvt - queue->tx.rsp_cons) <
242 (NET_TX_RING_SIZE - XEN_NETIF_NR_SLOTS_MIN - 1);
243}
244
245static void xennet_maybe_wake_tx(struct netfront_queue *queue)
246{
247 struct net_device *dev = queue->info->netdev;
248 struct netdev_queue *dev_queue = netdev_get_tx_queue(dev, queue->id);
249
250 if (unlikely(netif_tx_queue_stopped(dev_queue)) &&
251 netfront_tx_slot_available(queue) &&
252 likely(netif_running(dev)))
253 netif_tx_wake_queue(netdev_get_tx_queue(dev, queue->id));
254}
255
256
257static struct sk_buff *xennet_alloc_one_rx_buffer(struct netfront_queue *queue)
258{
259 struct sk_buff *skb;
260 struct page *page;
261
262 skb = __netdev_alloc_skb(queue->info->netdev,
263 RX_COPY_THRESHOLD + NET_IP_ALIGN,
264 GFP_ATOMIC | __GFP_NOWARN);
265 if (unlikely(!skb))
266 return NULL;
267
268 page = alloc_page(GFP_ATOMIC | __GFP_NOWARN);
269 if (!page) {
270 kfree_skb(skb);
271 return NULL;
272 }
273 skb_add_rx_frag(skb, 0, page, 0, 0, PAGE_SIZE);
274
275 /* Align ip header to a 16 bytes boundary */
276 skb_reserve(skb, NET_IP_ALIGN);
277 skb->dev = queue->info->netdev;
278
279 return skb;
280}
281
282
283static void xennet_alloc_rx_buffers(struct netfront_queue *queue)
284{
285 RING_IDX req_prod = queue->rx.req_prod_pvt;
286 int notify;
287 int err = 0;
288
289 if (unlikely(!netif_carrier_ok(queue->info->netdev)))
290 return;
291
292 for (req_prod = queue->rx.req_prod_pvt;
293 req_prod - queue->rx.rsp_cons < NET_RX_RING_SIZE;
294 req_prod++) {
295 struct sk_buff *skb;
296 unsigned short id;
297 grant_ref_t ref;
298 struct page *page;
299 struct xen_netif_rx_request *req;
300
301 skb = xennet_alloc_one_rx_buffer(queue);
302 if (!skb) {
303 err = -ENOMEM;
304 break;
305 }
306
307 id = xennet_rxidx(req_prod);
308
309 BUG_ON(queue->rx_skbs[id]);
310 queue->rx_skbs[id] = skb;
311
312 ref = gnttab_claim_grant_reference(&queue->gref_rx_head);
313 WARN_ON_ONCE(IS_ERR_VALUE((unsigned long)(int)ref));
314 queue->grant_rx_ref[id] = ref;
315
316 page = skb_frag_page(&skb_shinfo(skb)->frags[0]);
317
318 req = RING_GET_REQUEST(&queue->rx, req_prod);
319 gnttab_page_grant_foreign_access_ref_one(ref,
320 queue->info->xbdev->otherend_id,
321 page,
322 0);
323 req->id = id;
324 req->gref = ref;
325 }
326
327 queue->rx.req_prod_pvt = req_prod;
328
329 /* Try again later if there are not enough requests or skb allocation
330 * failed.
331 * Enough requests is quantified as the sum of newly created slots and
332 * the unconsumed slots at the backend.
333 */
334 if (req_prod - queue->rx.rsp_cons < NET_RX_SLOTS_MIN ||
335 unlikely(err)) {
336 mod_timer(&queue->rx_refill_timer, jiffies + (HZ/10));
337 return;
338 }
339
340 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->rx, notify);
341 if (notify)
342 notify_remote_via_irq(queue->rx_irq);
343}
344
345static int xennet_open(struct net_device *dev)
346{
347 struct netfront_info *np = netdev_priv(dev);
348 unsigned int num_queues = dev->real_num_tx_queues;
349 unsigned int i = 0;
350 struct netfront_queue *queue = NULL;
351
352 if (!np->queues)
353 return -ENODEV;
354
355 for (i = 0; i < num_queues; ++i) {
356 queue = &np->queues[i];
357 napi_enable(&queue->napi);
358
359 spin_lock_bh(&queue->rx_lock);
360 if (netif_carrier_ok(dev)) {
361 xennet_alloc_rx_buffers(queue);
362 queue->rx.sring->rsp_event = queue->rx.rsp_cons + 1;
363 if (RING_HAS_UNCONSUMED_RESPONSES(&queue->rx))
364 napi_schedule(&queue->napi);
365 }
366 spin_unlock_bh(&queue->rx_lock);
367 }
368
369 netif_tx_start_all_queues(dev);
370
371 return 0;
372}
373
374static void xennet_tx_buf_gc(struct netfront_queue *queue)
375{
376 RING_IDX cons, prod;
377 unsigned short id;
378 struct sk_buff *skb;
379 bool more_to_do;
380
381 BUG_ON(!netif_carrier_ok(queue->info->netdev));
382
383 do {
384 prod = queue->tx.sring->rsp_prod;
385 rmb(); /* Ensure we see responses up to 'rp'. */
386
387 for (cons = queue->tx.rsp_cons; cons != prod; cons++) {
388 struct xen_netif_tx_response *txrsp;
389
390 txrsp = RING_GET_RESPONSE(&queue->tx, cons);
391 if (txrsp->status == XEN_NETIF_RSP_NULL)
392 continue;
393
394 id = txrsp->id;
395 skb = queue->tx_skbs[id].skb;
396 if (unlikely(gnttab_query_foreign_access(
397 queue->grant_tx_ref[id]) != 0)) {
398 pr_alert("%s: warning -- grant still in use by backend domain\n",
399 __func__);
400 BUG();
401 }
402 gnttab_end_foreign_access_ref(
403 queue->grant_tx_ref[id], GNTMAP_readonly);
404 gnttab_release_grant_reference(
405 &queue->gref_tx_head, queue->grant_tx_ref[id]);
406 queue->grant_tx_ref[id] = GRANT_INVALID_REF;
407 queue->grant_tx_page[id] = NULL;
408 add_id_to_freelist(&queue->tx_skb_freelist, queue->tx_skbs, id);
409 dev_kfree_skb_irq(skb);
410 }
411
412 queue->tx.rsp_cons = prod;
413
414 RING_FINAL_CHECK_FOR_RESPONSES(&queue->tx, more_to_do);
415 } while (more_to_do);
416
417 xennet_maybe_wake_tx(queue);
418}
419
420struct xennet_gnttab_make_txreq {
421 struct netfront_queue *queue;
422 struct sk_buff *skb;
423 struct page *page;
424 struct xen_netif_tx_request *tx; /* Last request */
425 unsigned int size;
426};
427
428static void xennet_tx_setup_grant(unsigned long gfn, unsigned int offset,
429 unsigned int len, void *data)
430{
431 struct xennet_gnttab_make_txreq *info = data;
432 unsigned int id;
433 struct xen_netif_tx_request *tx;
434 grant_ref_t ref;
435 /* convenient aliases */
436 struct page *page = info->page;
437 struct netfront_queue *queue = info->queue;
438 struct sk_buff *skb = info->skb;
439
440 id = get_id_from_freelist(&queue->tx_skb_freelist, queue->tx_skbs);
441 tx = RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++);
442 ref = gnttab_claim_grant_reference(&queue->gref_tx_head);
443 WARN_ON_ONCE(IS_ERR_VALUE((unsigned long)(int)ref));
444
445 gnttab_grant_foreign_access_ref(ref, queue->info->xbdev->otherend_id,
446 gfn, GNTMAP_readonly);
447
448 queue->tx_skbs[id].skb = skb;
449 queue->grant_tx_page[id] = page;
450 queue->grant_tx_ref[id] = ref;
451
452 tx->id = id;
453 tx->gref = ref;
454 tx->offset = offset;
455 tx->size = len;
456 tx->flags = 0;
457
458 info->tx = tx;
459 info->size += tx->size;
460}
461
462static struct xen_netif_tx_request *xennet_make_first_txreq(
463 struct netfront_queue *queue, struct sk_buff *skb,
464 struct page *page, unsigned int offset, unsigned int len)
465{
466 struct xennet_gnttab_make_txreq info = {
467 .queue = queue,
468 .skb = skb,
469 .page = page,
470 .size = 0,
471 };
472
473 gnttab_for_one_grant(page, offset, len, xennet_tx_setup_grant, &info);
474
475 return info.tx;
476}
477
478static void xennet_make_one_txreq(unsigned long gfn, unsigned int offset,
479 unsigned int len, void *data)
480{
481 struct xennet_gnttab_make_txreq *info = data;
482
483 info->tx->flags |= XEN_NETTXF_more_data;
484 skb_get(info->skb);
485 xennet_tx_setup_grant(gfn, offset, len, data);
486}
487
488static struct xen_netif_tx_request *xennet_make_txreqs(
489 struct netfront_queue *queue, struct xen_netif_tx_request *tx,
490 struct sk_buff *skb, struct page *page,
491 unsigned int offset, unsigned int len)
492{
493 struct xennet_gnttab_make_txreq info = {
494 .queue = queue,
495 .skb = skb,
496 .tx = tx,
497 };
498
499 /* Skip unused frames from start of page */
500 page += offset >> PAGE_SHIFT;
501 offset &= ~PAGE_MASK;
502
503 while (len) {
504 info.page = page;
505 info.size = 0;
506
507 gnttab_foreach_grant_in_range(page, offset, len,
508 xennet_make_one_txreq,
509 &info);
510
511 page++;
512 offset = 0;
513 len -= info.size;
514 }
515
516 return info.tx;
517}
518
519/*
520 * Count how many ring slots are required to send this skb. Each frag
521 * might be a compound page.
522 */
523static int xennet_count_skb_slots(struct sk_buff *skb)
524{
525 int i, frags = skb_shinfo(skb)->nr_frags;
526 int slots;
527
528 slots = gnttab_count_grant(offset_in_page(skb->data),
529 skb_headlen(skb));
530
531 for (i = 0; i < frags; i++) {
532 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
533 unsigned long size = skb_frag_size(frag);
534 unsigned long offset = skb_frag_off(frag);
535
536 /* Skip unused frames from start of page */
537 offset &= ~PAGE_MASK;
538
539 slots += gnttab_count_grant(offset, size);
540 }
541
542 return slots;
543}
544
545static u16 xennet_select_queue(struct net_device *dev, struct sk_buff *skb,
546 struct net_device *sb_dev)
547{
548 unsigned int num_queues = dev->real_num_tx_queues;
549 u32 hash;
550 u16 queue_idx;
551
552 /* First, check if there is only one queue */
553 if (num_queues == 1) {
554 queue_idx = 0;
555 } else {
556 hash = skb_get_hash(skb);
557 queue_idx = hash % num_queues;
558 }
559
560 return queue_idx;
561}
562
563#define MAX_XEN_SKB_FRAGS (65536 / XEN_PAGE_SIZE + 1)
564
565static netdev_tx_t xennet_start_xmit(struct sk_buff *skb, struct net_device *dev)
566{
567 struct netfront_info *np = netdev_priv(dev);
568 struct netfront_stats *tx_stats = this_cpu_ptr(np->tx_stats);
569 struct xen_netif_tx_request *tx, *first_tx;
570 unsigned int i;
571 int notify;
572 int slots;
573 struct page *page;
574 unsigned int offset;
575 unsigned int len;
576 unsigned long flags;
577 struct netfront_queue *queue = NULL;
578 unsigned int num_queues = dev->real_num_tx_queues;
579 u16 queue_index;
580 struct sk_buff *nskb;
581
582 /* Drop the packet if no queues are set up */
583 if (num_queues < 1)
584 goto drop;
585 /* Determine which queue to transmit this SKB on */
586 queue_index = skb_get_queue_mapping(skb);
587 queue = &np->queues[queue_index];
588
589 /* If skb->len is too big for wire format, drop skb and alert
590 * user about misconfiguration.
591 */
592 if (unlikely(skb->len > XEN_NETIF_MAX_TX_SIZE)) {
593 net_alert_ratelimited(
594 "xennet: skb->len = %u, too big for wire format\n",
595 skb->len);
596 goto drop;
597 }
598
599 slots = xennet_count_skb_slots(skb);
600 if (unlikely(slots > MAX_XEN_SKB_FRAGS + 1)) {
601 net_dbg_ratelimited("xennet: skb rides the rocket: %d slots, %d bytes\n",
602 slots, skb->len);
603 if (skb_linearize(skb))
604 goto drop;
605 }
606
607 page = virt_to_page(skb->data);
608 offset = offset_in_page(skb->data);
609
610 /* The first req should be at least ETH_HLEN size or the packet will be
611 * dropped by netback.
612 */
613 if (unlikely(PAGE_SIZE - offset < ETH_HLEN)) {
614 nskb = skb_copy(skb, GFP_ATOMIC);
615 if (!nskb)
616 goto drop;
617 dev_consume_skb_any(skb);
618 skb = nskb;
619 page = virt_to_page(skb->data);
620 offset = offset_in_page(skb->data);
621 }
622
623 len = skb_headlen(skb);
624
625 spin_lock_irqsave(&queue->tx_lock, flags);
626
627 if (unlikely(!netif_carrier_ok(dev) ||
628 (slots > 1 && !xennet_can_sg(dev)) ||
629 netif_needs_gso(skb, netif_skb_features(skb)))) {
630 spin_unlock_irqrestore(&queue->tx_lock, flags);
631 goto drop;
632 }
633
634 /* First request for the linear area. */
635 first_tx = tx = xennet_make_first_txreq(queue, skb,
636 page, offset, len);
637 offset += tx->size;
638 if (offset == PAGE_SIZE) {
639 page++;
640 offset = 0;
641 }
642 len -= tx->size;
643
644 if (skb->ip_summed == CHECKSUM_PARTIAL)
645 /* local packet? */
646 tx->flags |= XEN_NETTXF_csum_blank | XEN_NETTXF_data_validated;
647 else if (skb->ip_summed == CHECKSUM_UNNECESSARY)
648 /* remote but checksummed. */
649 tx->flags |= XEN_NETTXF_data_validated;
650
651 /* Optional extra info after the first request. */
652 if (skb_shinfo(skb)->gso_size) {
653 struct xen_netif_extra_info *gso;
654
655 gso = (struct xen_netif_extra_info *)
656 RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++);
657
658 tx->flags |= XEN_NETTXF_extra_info;
659
660 gso->u.gso.size = skb_shinfo(skb)->gso_size;
661 gso->u.gso.type = (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) ?
662 XEN_NETIF_GSO_TYPE_TCPV6 :
663 XEN_NETIF_GSO_TYPE_TCPV4;
664 gso->u.gso.pad = 0;
665 gso->u.gso.features = 0;
666
667 gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
668 gso->flags = 0;
669 }
670
671 /* Requests for the rest of the linear area. */
672 tx = xennet_make_txreqs(queue, tx, skb, page, offset, len);
673
674 /* Requests for all the frags. */
675 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
676 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
677 tx = xennet_make_txreqs(queue, tx, skb, skb_frag_page(frag),
678 skb_frag_off(frag),
679 skb_frag_size(frag));
680 }
681
682 /* First request has the packet length. */
683 first_tx->size = skb->len;
684
685 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->tx, notify);
686 if (notify)
687 notify_remote_via_irq(queue->tx_irq);
688
689 u64_stats_update_begin(&tx_stats->syncp);
690 tx_stats->bytes += skb->len;
691 tx_stats->packets++;
692 u64_stats_update_end(&tx_stats->syncp);
693
694 /* Note: It is not safe to access skb after xennet_tx_buf_gc()! */
695 xennet_tx_buf_gc(queue);
696
697 if (!netfront_tx_slot_available(queue))
698 netif_tx_stop_queue(netdev_get_tx_queue(dev, queue->id));
699
700 spin_unlock_irqrestore(&queue->tx_lock, flags);
701
702 return NETDEV_TX_OK;
703
704 drop:
705 dev->stats.tx_dropped++;
706 dev_kfree_skb_any(skb);
707 return NETDEV_TX_OK;
708}
709
710static int xennet_close(struct net_device *dev)
711{
712 struct netfront_info *np = netdev_priv(dev);
713 unsigned int num_queues = dev->real_num_tx_queues;
714 unsigned int i;
715 struct netfront_queue *queue;
716 netif_tx_stop_all_queues(np->netdev);
717 for (i = 0; i < num_queues; ++i) {
718 queue = &np->queues[i];
719 napi_disable(&queue->napi);
720 }
721 return 0;
722}
723
724static void xennet_move_rx_slot(struct netfront_queue *queue, struct sk_buff *skb,
725 grant_ref_t ref)
726{
727 int new = xennet_rxidx(queue->rx.req_prod_pvt);
728
729 BUG_ON(queue->rx_skbs[new]);
730 queue->rx_skbs[new] = skb;
731 queue->grant_rx_ref[new] = ref;
732 RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->id = new;
733 RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->gref = ref;
734 queue->rx.req_prod_pvt++;
735}
736
737static int xennet_get_extras(struct netfront_queue *queue,
738 struct xen_netif_extra_info *extras,
739 RING_IDX rp)
740
741{
742 struct xen_netif_extra_info *extra;
743 struct device *dev = &queue->info->netdev->dev;
744 RING_IDX cons = queue->rx.rsp_cons;
745 int err = 0;
746
747 do {
748 struct sk_buff *skb;
749 grant_ref_t ref;
750
751 if (unlikely(cons + 1 == rp)) {
752 if (net_ratelimit())
753 dev_warn(dev, "Missing extra info\n");
754 err = -EBADR;
755 break;
756 }
757
758 extra = (struct xen_netif_extra_info *)
759 RING_GET_RESPONSE(&queue->rx, ++cons);
760
761 if (unlikely(!extra->type ||
762 extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
763 if (net_ratelimit())
764 dev_warn(dev, "Invalid extra type: %d\n",
765 extra->type);
766 err = -EINVAL;
767 } else {
768 memcpy(&extras[extra->type - 1], extra,
769 sizeof(*extra));
770 }
771
772 skb = xennet_get_rx_skb(queue, cons);
773 ref = xennet_get_rx_ref(queue, cons);
774 xennet_move_rx_slot(queue, skb, ref);
775 } while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE);
776
777 queue->rx.rsp_cons = cons;
778 return err;
779}
780
781static int xennet_get_responses(struct netfront_queue *queue,
782 struct netfront_rx_info *rinfo, RING_IDX rp,
783 struct sk_buff_head *list)
784{
785 struct xen_netif_rx_response *rx = &rinfo->rx;
786 struct xen_netif_extra_info *extras = rinfo->extras;
787 struct device *dev = &queue->info->netdev->dev;
788 RING_IDX cons = queue->rx.rsp_cons;
789 struct sk_buff *skb = xennet_get_rx_skb(queue, cons);
790 grant_ref_t ref = xennet_get_rx_ref(queue, cons);
791 int max = XEN_NETIF_NR_SLOTS_MIN + (rx->status <= RX_COPY_THRESHOLD);
792 int slots = 1;
793 int err = 0;
794 unsigned long ret;
795
796 if (rx->flags & XEN_NETRXF_extra_info) {
797 err = xennet_get_extras(queue, extras, rp);
798 cons = queue->rx.rsp_cons;
799 }
800
801 for (;;) {
802 if (unlikely(rx->status < 0 ||
803 rx->offset + rx->status > XEN_PAGE_SIZE)) {
804 if (net_ratelimit())
805 dev_warn(dev, "rx->offset: %u, size: %d\n",
806 rx->offset, rx->status);
807 xennet_move_rx_slot(queue, skb, ref);
808 err = -EINVAL;
809 goto next;
810 }
811
812 /*
813 * This definitely indicates a bug, either in this driver or in
814 * the backend driver. In future this should flag the bad
815 * situation to the system controller to reboot the backend.
816 */
817 if (ref == GRANT_INVALID_REF) {
818 if (net_ratelimit())
819 dev_warn(dev, "Bad rx response id %d.\n",
820 rx->id);
821 err = -EINVAL;
822 goto next;
823 }
824
825 ret = gnttab_end_foreign_access_ref(ref, 0);
826 BUG_ON(!ret);
827
828 gnttab_release_grant_reference(&queue->gref_rx_head, ref);
829
830 __skb_queue_tail(list, skb);
831
832next:
833 if (!(rx->flags & XEN_NETRXF_more_data))
834 break;
835
836 if (cons + slots == rp) {
837 if (net_ratelimit())
838 dev_warn(dev, "Need more slots\n");
839 err = -ENOENT;
840 break;
841 }
842
843 rx = RING_GET_RESPONSE(&queue->rx, cons + slots);
844 skb = xennet_get_rx_skb(queue, cons + slots);
845 ref = xennet_get_rx_ref(queue, cons + slots);
846 slots++;
847 }
848
849 if (unlikely(slots > max)) {
850 if (net_ratelimit())
851 dev_warn(dev, "Too many slots\n");
852 err = -E2BIG;
853 }
854
855 if (unlikely(err))
856 queue->rx.rsp_cons = cons + slots;
857
858 return err;
859}
860
861static int xennet_set_skb_gso(struct sk_buff *skb,
862 struct xen_netif_extra_info *gso)
863{
864 if (!gso->u.gso.size) {
865 if (net_ratelimit())
866 pr_warn("GSO size must not be zero\n");
867 return -EINVAL;
868 }
869
870 if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4 &&
871 gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV6) {
872 if (net_ratelimit())
873 pr_warn("Bad GSO type %d\n", gso->u.gso.type);
874 return -EINVAL;
875 }
876
877 skb_shinfo(skb)->gso_size = gso->u.gso.size;
878 skb_shinfo(skb)->gso_type =
879 (gso->u.gso.type == XEN_NETIF_GSO_TYPE_TCPV4) ?
880 SKB_GSO_TCPV4 :
881 SKB_GSO_TCPV6;
882
883 /* Header must be checked, and gso_segs computed. */
884 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
885 skb_shinfo(skb)->gso_segs = 0;
886
887 return 0;
888}
889
890static int xennet_fill_frags(struct netfront_queue *queue,
891 struct sk_buff *skb,
892 struct sk_buff_head *list)
893{
894 RING_IDX cons = queue->rx.rsp_cons;
895 struct sk_buff *nskb;
896
897 while ((nskb = __skb_dequeue(list))) {
898 struct xen_netif_rx_response *rx =
899 RING_GET_RESPONSE(&queue->rx, ++cons);
900 skb_frag_t *nfrag = &skb_shinfo(nskb)->frags[0];
901
902 if (skb_shinfo(skb)->nr_frags == MAX_SKB_FRAGS) {
903 unsigned int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
904
905 BUG_ON(pull_to < skb_headlen(skb));
906 __pskb_pull_tail(skb, pull_to - skb_headlen(skb));
907 }
908 if (unlikely(skb_shinfo(skb)->nr_frags >= MAX_SKB_FRAGS)) {
909 queue->rx.rsp_cons = ++cons + skb_queue_len(list);
910 kfree_skb(nskb);
911 return -ENOENT;
912 }
913
914 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags,
915 skb_frag_page(nfrag),
916 rx->offset, rx->status, PAGE_SIZE);
917
918 skb_shinfo(nskb)->nr_frags = 0;
919 kfree_skb(nskb);
920 }
921
922 queue->rx.rsp_cons = cons;
923
924 return 0;
925}
926
927static int checksum_setup(struct net_device *dev, struct sk_buff *skb)
928{
929 bool recalculate_partial_csum = false;
930
931 /*
932 * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy
933 * peers can fail to set NETRXF_csum_blank when sending a GSO
934 * frame. In this case force the SKB to CHECKSUM_PARTIAL and
935 * recalculate the partial checksum.
936 */
937 if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) {
938 struct netfront_info *np = netdev_priv(dev);
939 atomic_inc(&np->rx_gso_checksum_fixup);
940 skb->ip_summed = CHECKSUM_PARTIAL;
941 recalculate_partial_csum = true;
942 }
943
944 /* A non-CHECKSUM_PARTIAL SKB does not require setup. */
945 if (skb->ip_summed != CHECKSUM_PARTIAL)
946 return 0;
947
948 return skb_checksum_setup(skb, recalculate_partial_csum);
949}
950
951static int handle_incoming_queue(struct netfront_queue *queue,
952 struct sk_buff_head *rxq)
953{
954 struct netfront_stats *rx_stats = this_cpu_ptr(queue->info->rx_stats);
955 int packets_dropped = 0;
956 struct sk_buff *skb;
957
958 while ((skb = __skb_dequeue(rxq)) != NULL) {
959 int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
960
961 if (pull_to > skb_headlen(skb))
962 __pskb_pull_tail(skb, pull_to - skb_headlen(skb));
963
964 /* Ethernet work: Delayed to here as it peeks the header. */
965 skb->protocol = eth_type_trans(skb, queue->info->netdev);
966 skb_reset_network_header(skb);
967
968 if (checksum_setup(queue->info->netdev, skb)) {
969 kfree_skb(skb);
970 packets_dropped++;
971 queue->info->netdev->stats.rx_errors++;
972 continue;
973 }
974
975 u64_stats_update_begin(&rx_stats->syncp);
976 rx_stats->packets++;
977 rx_stats->bytes += skb->len;
978 u64_stats_update_end(&rx_stats->syncp);
979
980 /* Pass it up. */
981 napi_gro_receive(&queue->napi, skb);
982 }
983
984 return packets_dropped;
985}
986
987static int xennet_poll(struct napi_struct *napi, int budget)
988{
989 struct netfront_queue *queue = container_of(napi, struct netfront_queue, napi);
990 struct net_device *dev = queue->info->netdev;
991 struct sk_buff *skb;
992 struct netfront_rx_info rinfo;
993 struct xen_netif_rx_response *rx = &rinfo.rx;
994 struct xen_netif_extra_info *extras = rinfo.extras;
995 RING_IDX i, rp;
996 int work_done;
997 struct sk_buff_head rxq;
998 struct sk_buff_head errq;
999 struct sk_buff_head tmpq;
1000 int err;
1001
1002 spin_lock(&queue->rx_lock);
1003
1004 skb_queue_head_init(&rxq);
1005 skb_queue_head_init(&errq);
1006 skb_queue_head_init(&tmpq);
1007
1008 rp = queue->rx.sring->rsp_prod;
1009 rmb(); /* Ensure we see queued responses up to 'rp'. */
1010
1011 i = queue->rx.rsp_cons;
1012 work_done = 0;
1013 while ((i != rp) && (work_done < budget)) {
1014 memcpy(rx, RING_GET_RESPONSE(&queue->rx, i), sizeof(*rx));
1015 memset(extras, 0, sizeof(rinfo.extras));
1016
1017 err = xennet_get_responses(queue, &rinfo, rp, &tmpq);
1018
1019 if (unlikely(err)) {
1020err:
1021 while ((skb = __skb_dequeue(&tmpq)))
1022 __skb_queue_tail(&errq, skb);
1023 dev->stats.rx_errors++;
1024 i = queue->rx.rsp_cons;
1025 continue;
1026 }
1027
1028 skb = __skb_dequeue(&tmpq);
1029
1030 if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) {
1031 struct xen_netif_extra_info *gso;
1032 gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1];
1033
1034 if (unlikely(xennet_set_skb_gso(skb, gso))) {
1035 __skb_queue_head(&tmpq, skb);
1036 queue->rx.rsp_cons += skb_queue_len(&tmpq);
1037 goto err;
1038 }
1039 }
1040
1041 NETFRONT_SKB_CB(skb)->pull_to = rx->status;
1042 if (NETFRONT_SKB_CB(skb)->pull_to > RX_COPY_THRESHOLD)
1043 NETFRONT_SKB_CB(skb)->pull_to = RX_COPY_THRESHOLD;
1044
1045 skb_frag_off_set(&skb_shinfo(skb)->frags[0], rx->offset);
1046 skb_frag_size_set(&skb_shinfo(skb)->frags[0], rx->status);
1047 skb->data_len = rx->status;
1048 skb->len += rx->status;
1049
1050 if (unlikely(xennet_fill_frags(queue, skb, &tmpq)))
1051 goto err;
1052
1053 if (rx->flags & XEN_NETRXF_csum_blank)
1054 skb->ip_summed = CHECKSUM_PARTIAL;
1055 else if (rx->flags & XEN_NETRXF_data_validated)
1056 skb->ip_summed = CHECKSUM_UNNECESSARY;
1057
1058 __skb_queue_tail(&rxq, skb);
1059
1060 i = ++queue->rx.rsp_cons;
1061 work_done++;
1062 }
1063
1064 __skb_queue_purge(&errq);
1065
1066 work_done -= handle_incoming_queue(queue, &rxq);
1067
1068 xennet_alloc_rx_buffers(queue);
1069
1070 if (work_done < budget) {
1071 int more_to_do = 0;
1072
1073 napi_complete_done(napi, work_done);
1074
1075 RING_FINAL_CHECK_FOR_RESPONSES(&queue->rx, more_to_do);
1076 if (more_to_do)
1077 napi_schedule(napi);
1078 }
1079
1080 spin_unlock(&queue->rx_lock);
1081
1082 return work_done;
1083}
1084
1085static int xennet_change_mtu(struct net_device *dev, int mtu)
1086{
1087 int max = xennet_can_sg(dev) ? XEN_NETIF_MAX_TX_SIZE : ETH_DATA_LEN;
1088
1089 if (mtu > max)
1090 return -EINVAL;
1091 dev->mtu = mtu;
1092 return 0;
1093}
1094
1095static void xennet_get_stats64(struct net_device *dev,
1096 struct rtnl_link_stats64 *tot)
1097{
1098 struct netfront_info *np = netdev_priv(dev);
1099 int cpu;
1100
1101 for_each_possible_cpu(cpu) {
1102 struct netfront_stats *rx_stats = per_cpu_ptr(np->rx_stats, cpu);
1103 struct netfront_stats *tx_stats = per_cpu_ptr(np->tx_stats, cpu);
1104 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1105 unsigned int start;
1106
1107 do {
1108 start = u64_stats_fetch_begin_irq(&tx_stats->syncp);
1109 tx_packets = tx_stats->packets;
1110 tx_bytes = tx_stats->bytes;
1111 } while (u64_stats_fetch_retry_irq(&tx_stats->syncp, start));
1112
1113 do {
1114 start = u64_stats_fetch_begin_irq(&rx_stats->syncp);
1115 rx_packets = rx_stats->packets;
1116 rx_bytes = rx_stats->bytes;
1117 } while (u64_stats_fetch_retry_irq(&rx_stats->syncp, start));
1118
1119 tot->rx_packets += rx_packets;
1120 tot->tx_packets += tx_packets;
1121 tot->rx_bytes += rx_bytes;
1122 tot->tx_bytes += tx_bytes;
1123 }
1124
1125 tot->rx_errors = dev->stats.rx_errors;
1126 tot->tx_dropped = dev->stats.tx_dropped;
1127}
1128
1129static void xennet_release_tx_bufs(struct netfront_queue *queue)
1130{
1131 struct sk_buff *skb;
1132 int i;
1133
1134 for (i = 0; i < NET_TX_RING_SIZE; i++) {
1135 /* Skip over entries which are actually freelist references */
1136 if (skb_entry_is_link(&queue->tx_skbs[i]))
1137 continue;
1138
1139 skb = queue->tx_skbs[i].skb;
1140 get_page(queue->grant_tx_page[i]);
1141 gnttab_end_foreign_access(queue->grant_tx_ref[i],
1142 GNTMAP_readonly,
1143 (unsigned long)page_address(queue->grant_tx_page[i]));
1144 queue->grant_tx_page[i] = NULL;
1145 queue->grant_tx_ref[i] = GRANT_INVALID_REF;
1146 add_id_to_freelist(&queue->tx_skb_freelist, queue->tx_skbs, i);
1147 dev_kfree_skb_irq(skb);
1148 }
1149}
1150
1151static void xennet_release_rx_bufs(struct netfront_queue *queue)
1152{
1153 int id, ref;
1154
1155 spin_lock_bh(&queue->rx_lock);
1156
1157 for (id = 0; id < NET_RX_RING_SIZE; id++) {
1158 struct sk_buff *skb;
1159 struct page *page;
1160
1161 skb = queue->rx_skbs[id];
1162 if (!skb)
1163 continue;
1164
1165 ref = queue->grant_rx_ref[id];
1166 if (ref == GRANT_INVALID_REF)
1167 continue;
1168
1169 page = skb_frag_page(&skb_shinfo(skb)->frags[0]);
1170
1171 /* gnttab_end_foreign_access() needs a page ref until
1172 * foreign access is ended (which may be deferred).
1173 */
1174 get_page(page);
1175 gnttab_end_foreign_access(ref, 0,
1176 (unsigned long)page_address(page));
1177 queue->grant_rx_ref[id] = GRANT_INVALID_REF;
1178
1179 kfree_skb(skb);
1180 }
1181
1182 spin_unlock_bh(&queue->rx_lock);
1183}
1184
1185static netdev_features_t xennet_fix_features(struct net_device *dev,
1186 netdev_features_t features)
1187{
1188 struct netfront_info *np = netdev_priv(dev);
1189
1190 if (features & NETIF_F_SG &&
1191 !xenbus_read_unsigned(np->xbdev->otherend, "feature-sg", 0))
1192 features &= ~NETIF_F_SG;
1193
1194 if (features & NETIF_F_IPV6_CSUM &&
1195 !xenbus_read_unsigned(np->xbdev->otherend,
1196 "feature-ipv6-csum-offload", 0))
1197 features &= ~NETIF_F_IPV6_CSUM;
1198
1199 if (features & NETIF_F_TSO &&
1200 !xenbus_read_unsigned(np->xbdev->otherend, "feature-gso-tcpv4", 0))
1201 features &= ~NETIF_F_TSO;
1202
1203 if (features & NETIF_F_TSO6 &&
1204 !xenbus_read_unsigned(np->xbdev->otherend, "feature-gso-tcpv6", 0))
1205 features &= ~NETIF_F_TSO6;
1206
1207 return features;
1208}
1209
1210static int xennet_set_features(struct net_device *dev,
1211 netdev_features_t features)
1212{
1213 if (!(features & NETIF_F_SG) && dev->mtu > ETH_DATA_LEN) {
1214 netdev_info(dev, "Reducing MTU because no SG offload");
1215 dev->mtu = ETH_DATA_LEN;
1216 }
1217
1218 return 0;
1219}
1220
1221static irqreturn_t xennet_tx_interrupt(int irq, void *dev_id)
1222{
1223 struct netfront_queue *queue = dev_id;
1224 unsigned long flags;
1225
1226 spin_lock_irqsave(&queue->tx_lock, flags);
1227 xennet_tx_buf_gc(queue);
1228 spin_unlock_irqrestore(&queue->tx_lock, flags);
1229
1230 return IRQ_HANDLED;
1231}
1232
1233static irqreturn_t xennet_rx_interrupt(int irq, void *dev_id)
1234{
1235 struct netfront_queue *queue = dev_id;
1236 struct net_device *dev = queue->info->netdev;
1237
1238 if (likely(netif_carrier_ok(dev) &&
1239 RING_HAS_UNCONSUMED_RESPONSES(&queue->rx)))
1240 napi_schedule(&queue->napi);
1241
1242 return IRQ_HANDLED;
1243}
1244
1245static irqreturn_t xennet_interrupt(int irq, void *dev_id)
1246{
1247 xennet_tx_interrupt(irq, dev_id);
1248 xennet_rx_interrupt(irq, dev_id);
1249 return IRQ_HANDLED;
1250}
1251
1252#ifdef CONFIG_NET_POLL_CONTROLLER
1253static void xennet_poll_controller(struct net_device *dev)
1254{
1255 /* Poll each queue */
1256 struct netfront_info *info = netdev_priv(dev);
1257 unsigned int num_queues = dev->real_num_tx_queues;
1258 unsigned int i;
1259 for (i = 0; i < num_queues; ++i)
1260 xennet_interrupt(0, &info->queues[i]);
1261}
1262#endif
1263
1264static const struct net_device_ops xennet_netdev_ops = {
1265 .ndo_open = xennet_open,
1266 .ndo_stop = xennet_close,
1267 .ndo_start_xmit = xennet_start_xmit,
1268 .ndo_change_mtu = xennet_change_mtu,
1269 .ndo_get_stats64 = xennet_get_stats64,
1270 .ndo_set_mac_address = eth_mac_addr,
1271 .ndo_validate_addr = eth_validate_addr,
1272 .ndo_fix_features = xennet_fix_features,
1273 .ndo_set_features = xennet_set_features,
1274 .ndo_select_queue = xennet_select_queue,
1275#ifdef CONFIG_NET_POLL_CONTROLLER
1276 .ndo_poll_controller = xennet_poll_controller,
1277#endif
1278};
1279
1280static void xennet_free_netdev(struct net_device *netdev)
1281{
1282 struct netfront_info *np = netdev_priv(netdev);
1283
1284 free_percpu(np->rx_stats);
1285 free_percpu(np->tx_stats);
1286 free_netdev(netdev);
1287}
1288
1289static struct net_device *xennet_create_dev(struct xenbus_device *dev)
1290{
1291 int err;
1292 struct net_device *netdev;
1293 struct netfront_info *np;
1294
1295 netdev = alloc_etherdev_mq(sizeof(struct netfront_info), xennet_max_queues);
1296 if (!netdev)
1297 return ERR_PTR(-ENOMEM);
1298
1299 np = netdev_priv(netdev);
1300 np->xbdev = dev;
1301
1302 np->queues = NULL;
1303
1304 err = -ENOMEM;
1305 np->rx_stats = netdev_alloc_pcpu_stats(struct netfront_stats);
1306 if (np->rx_stats == NULL)
1307 goto exit;
1308 np->tx_stats = netdev_alloc_pcpu_stats(struct netfront_stats);
1309 if (np->tx_stats == NULL)
1310 goto exit;
1311
1312 netdev->netdev_ops = &xennet_netdev_ops;
1313
1314 netdev->features = NETIF_F_IP_CSUM | NETIF_F_RXCSUM |
1315 NETIF_F_GSO_ROBUST;
1316 netdev->hw_features = NETIF_F_SG |
1317 NETIF_F_IPV6_CSUM |
1318 NETIF_F_TSO | NETIF_F_TSO6;
1319
1320 /*
1321 * Assume that all hw features are available for now. This set
1322 * will be adjusted by the call to netdev_update_features() in
1323 * xennet_connect() which is the earliest point where we can
1324 * negotiate with the backend regarding supported features.
1325 */
1326 netdev->features |= netdev->hw_features;
1327
1328 netdev->ethtool_ops = &xennet_ethtool_ops;
1329 netdev->min_mtu = ETH_MIN_MTU;
1330 netdev->max_mtu = XEN_NETIF_MAX_TX_SIZE;
1331 SET_NETDEV_DEV(netdev, &dev->dev);
1332
1333 np->netdev = netdev;
1334
1335 netif_carrier_off(netdev);
1336
1337 xenbus_switch_state(dev, XenbusStateInitialising);
1338 wait_event(module_wq,
1339 xenbus_read_driver_state(dev->otherend) !=
1340 XenbusStateClosed &&
1341 xenbus_read_driver_state(dev->otherend) !=
1342 XenbusStateUnknown);
1343 return netdev;
1344
1345 exit:
1346 xennet_free_netdev(netdev);
1347 return ERR_PTR(err);
1348}
1349
1350/**
1351 * Entry point to this code when a new device is created. Allocate the basic
1352 * structures and the ring buffers for communication with the backend, and
1353 * inform the backend of the appropriate details for those.
1354 */
1355static int netfront_probe(struct xenbus_device *dev,
1356 const struct xenbus_device_id *id)
1357{
1358 int err;
1359 struct net_device *netdev;
1360 struct netfront_info *info;
1361
1362 netdev = xennet_create_dev(dev);
1363 if (IS_ERR(netdev)) {
1364 err = PTR_ERR(netdev);
1365 xenbus_dev_fatal(dev, err, "creating netdev");
1366 return err;
1367 }
1368
1369 info = netdev_priv(netdev);
1370 dev_set_drvdata(&dev->dev, info);
1371#ifdef CONFIG_SYSFS
1372 info->netdev->sysfs_groups[0] = &xennet_dev_group;
1373#endif
1374
1375 return 0;
1376}
1377
1378static void xennet_end_access(int ref, void *page)
1379{
1380 /* This frees the page as a side-effect */
1381 if (ref != GRANT_INVALID_REF)
1382 gnttab_end_foreign_access(ref, 0, (unsigned long)page);
1383}
1384
1385static void xennet_disconnect_backend(struct netfront_info *info)
1386{
1387 unsigned int i = 0;
1388 unsigned int num_queues = info->netdev->real_num_tx_queues;
1389
1390 netif_carrier_off(info->netdev);
1391
1392 for (i = 0; i < num_queues && info->queues; ++i) {
1393 struct netfront_queue *queue = &info->queues[i];
1394
1395 del_timer_sync(&queue->rx_refill_timer);
1396
1397 if (queue->tx_irq && (queue->tx_irq == queue->rx_irq))
1398 unbind_from_irqhandler(queue->tx_irq, queue);
1399 if (queue->tx_irq && (queue->tx_irq != queue->rx_irq)) {
1400 unbind_from_irqhandler(queue->tx_irq, queue);
1401 unbind_from_irqhandler(queue->rx_irq, queue);
1402 }
1403 queue->tx_evtchn = queue->rx_evtchn = 0;
1404 queue->tx_irq = queue->rx_irq = 0;
1405
1406 if (netif_running(info->netdev))
1407 napi_synchronize(&queue->napi);
1408
1409 xennet_release_tx_bufs(queue);
1410 xennet_release_rx_bufs(queue);
1411 gnttab_free_grant_references(queue->gref_tx_head);
1412 gnttab_free_grant_references(queue->gref_rx_head);
1413
1414 /* End access and free the pages */
1415 xennet_end_access(queue->tx_ring_ref, queue->tx.sring);
1416 xennet_end_access(queue->rx_ring_ref, queue->rx.sring);
1417
1418 queue->tx_ring_ref = GRANT_INVALID_REF;
1419 queue->rx_ring_ref = GRANT_INVALID_REF;
1420 queue->tx.sring = NULL;
1421 queue->rx.sring = NULL;
1422 }
1423}
1424
1425/**
1426 * We are reconnecting to the backend, due to a suspend/resume, or a backend
1427 * driver restart. We tear down our netif structure and recreate it, but
1428 * leave the device-layer structures intact so that this is transparent to the
1429 * rest of the kernel.
1430 */
1431static int netfront_resume(struct xenbus_device *dev)
1432{
1433 struct netfront_info *info = dev_get_drvdata(&dev->dev);
1434
1435 dev_dbg(&dev->dev, "%s\n", dev->nodename);
1436
1437 xennet_disconnect_backend(info);
1438 return 0;
1439}
1440
1441static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[])
1442{
1443 char *s, *e, *macstr;
1444 int i;
1445
1446 macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL);
1447 if (IS_ERR(macstr))
1448 return PTR_ERR(macstr);
1449
1450 for (i = 0; i < ETH_ALEN; i++) {
1451 mac[i] = simple_strtoul(s, &e, 16);
1452 if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) {
1453 kfree(macstr);
1454 return -ENOENT;
1455 }
1456 s = e+1;
1457 }
1458
1459 kfree(macstr);
1460 return 0;
1461}
1462
1463static int setup_netfront_single(struct netfront_queue *queue)
1464{
1465 int err;
1466
1467 err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn);
1468 if (err < 0)
1469 goto fail;
1470
1471 err = bind_evtchn_to_irqhandler(queue->tx_evtchn,
1472 xennet_interrupt,
1473 0, queue->info->netdev->name, queue);
1474 if (err < 0)
1475 goto bind_fail;
1476 queue->rx_evtchn = queue->tx_evtchn;
1477 queue->rx_irq = queue->tx_irq = err;
1478
1479 return 0;
1480
1481bind_fail:
1482 xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn);
1483 queue->tx_evtchn = 0;
1484fail:
1485 return err;
1486}
1487
1488static int setup_netfront_split(struct netfront_queue *queue)
1489{
1490 int err;
1491
1492 err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn);
1493 if (err < 0)
1494 goto fail;
1495 err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->rx_evtchn);
1496 if (err < 0)
1497 goto alloc_rx_evtchn_fail;
1498
1499 snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name),
1500 "%s-tx", queue->name);
1501 err = bind_evtchn_to_irqhandler(queue->tx_evtchn,
1502 xennet_tx_interrupt,
1503 0, queue->tx_irq_name, queue);
1504 if (err < 0)
1505 goto bind_tx_fail;
1506 queue->tx_irq = err;
1507
1508 snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name),
1509 "%s-rx", queue->name);
1510 err = bind_evtchn_to_irqhandler(queue->rx_evtchn,
1511 xennet_rx_interrupt,
1512 0, queue->rx_irq_name, queue);
1513 if (err < 0)
1514 goto bind_rx_fail;
1515 queue->rx_irq = err;
1516
1517 return 0;
1518
1519bind_rx_fail:
1520 unbind_from_irqhandler(queue->tx_irq, queue);
1521 queue->tx_irq = 0;
1522bind_tx_fail:
1523 xenbus_free_evtchn(queue->info->xbdev, queue->rx_evtchn);
1524 queue->rx_evtchn = 0;
1525alloc_rx_evtchn_fail:
1526 xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn);
1527 queue->tx_evtchn = 0;
1528fail:
1529 return err;
1530}
1531
1532static int setup_netfront(struct xenbus_device *dev,
1533 struct netfront_queue *queue, unsigned int feature_split_evtchn)
1534{
1535 struct xen_netif_tx_sring *txs;
1536 struct xen_netif_rx_sring *rxs;
1537 grant_ref_t gref;
1538 int err;
1539
1540 queue->tx_ring_ref = GRANT_INVALID_REF;
1541 queue->rx_ring_ref = GRANT_INVALID_REF;
1542 queue->rx.sring = NULL;
1543 queue->tx.sring = NULL;
1544
1545 txs = (struct xen_netif_tx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1546 if (!txs) {
1547 err = -ENOMEM;
1548 xenbus_dev_fatal(dev, err, "allocating tx ring page");
1549 goto fail;
1550 }
1551 SHARED_RING_INIT(txs);
1552 FRONT_RING_INIT(&queue->tx, txs, XEN_PAGE_SIZE);
1553
1554 err = xenbus_grant_ring(dev, txs, 1, &gref);
1555 if (err < 0)
1556 goto grant_tx_ring_fail;
1557 queue->tx_ring_ref = gref;
1558
1559 rxs = (struct xen_netif_rx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1560 if (!rxs) {
1561 err = -ENOMEM;
1562 xenbus_dev_fatal(dev, err, "allocating rx ring page");
1563 goto alloc_rx_ring_fail;
1564 }
1565 SHARED_RING_INIT(rxs);
1566 FRONT_RING_INIT(&queue->rx, rxs, XEN_PAGE_SIZE);
1567
1568 err = xenbus_grant_ring(dev, rxs, 1, &gref);
1569 if (err < 0)
1570 goto grant_rx_ring_fail;
1571 queue->rx_ring_ref = gref;
1572
1573 if (feature_split_evtchn)
1574 err = setup_netfront_split(queue);
1575 /* setup single event channel if
1576 * a) feature-split-event-channels == 0
1577 * b) feature-split-event-channels == 1 but failed to setup
1578 */
1579 if (!feature_split_evtchn || (feature_split_evtchn && err))
1580 err = setup_netfront_single(queue);
1581
1582 if (err)
1583 goto alloc_evtchn_fail;
1584
1585 return 0;
1586
1587 /* If we fail to setup netfront, it is safe to just revoke access to
1588 * granted pages because backend is not accessing it at this point.
1589 */
1590alloc_evtchn_fail:
1591 gnttab_end_foreign_access_ref(queue->rx_ring_ref, 0);
1592grant_rx_ring_fail:
1593 free_page((unsigned long)rxs);
1594alloc_rx_ring_fail:
1595 gnttab_end_foreign_access_ref(queue->tx_ring_ref, 0);
1596grant_tx_ring_fail:
1597 free_page((unsigned long)txs);
1598fail:
1599 return err;
1600}
1601
1602/* Queue-specific initialisation
1603 * This used to be done in xennet_create_dev() but must now
1604 * be run per-queue.
1605 */
1606static int xennet_init_queue(struct netfront_queue *queue)
1607{
1608 unsigned short i;
1609 int err = 0;
1610 char *devid;
1611
1612 spin_lock_init(&queue->tx_lock);
1613 spin_lock_init(&queue->rx_lock);
1614
1615 timer_setup(&queue->rx_refill_timer, rx_refill_timeout, 0);
1616
1617 devid = strrchr(queue->info->xbdev->nodename, '/') + 1;
1618 snprintf(queue->name, sizeof(queue->name), "vif%s-q%u",
1619 devid, queue->id);
1620
1621 /* Initialise tx_skbs as a free chain containing every entry. */
1622 queue->tx_skb_freelist = 0;
1623 for (i = 0; i < NET_TX_RING_SIZE; i++) {
1624 skb_entry_set_link(&queue->tx_skbs[i], i+1);
1625 queue->grant_tx_ref[i] = GRANT_INVALID_REF;
1626 queue->grant_tx_page[i] = NULL;
1627 }
1628
1629 /* Clear out rx_skbs */
1630 for (i = 0; i < NET_RX_RING_SIZE; i++) {
1631 queue->rx_skbs[i] = NULL;
1632 queue->grant_rx_ref[i] = GRANT_INVALID_REF;
1633 }
1634
1635 /* A grant for every tx ring slot */
1636 if (gnttab_alloc_grant_references(NET_TX_RING_SIZE,
1637 &queue->gref_tx_head) < 0) {
1638 pr_alert("can't alloc tx grant refs\n");
1639 err = -ENOMEM;
1640 goto exit;
1641 }
1642
1643 /* A grant for every rx ring slot */
1644 if (gnttab_alloc_grant_references(NET_RX_RING_SIZE,
1645 &queue->gref_rx_head) < 0) {
1646 pr_alert("can't alloc rx grant refs\n");
1647 err = -ENOMEM;
1648 goto exit_free_tx;
1649 }
1650
1651 return 0;
1652
1653 exit_free_tx:
1654 gnttab_free_grant_references(queue->gref_tx_head);
1655 exit:
1656 return err;
1657}
1658
1659static int write_queue_xenstore_keys(struct netfront_queue *queue,
1660 struct xenbus_transaction *xbt, int write_hierarchical)
1661{
1662 /* Write the queue-specific keys into XenStore in the traditional
1663 * way for a single queue, or in a queue subkeys for multiple
1664 * queues.
1665 */
1666 struct xenbus_device *dev = queue->info->xbdev;
1667 int err;
1668 const char *message;
1669 char *path;
1670 size_t pathsize;
1671
1672 /* Choose the correct place to write the keys */
1673 if (write_hierarchical) {
1674 pathsize = strlen(dev->nodename) + 10;
1675 path = kzalloc(pathsize, GFP_KERNEL);
1676 if (!path) {
1677 err = -ENOMEM;
1678 message = "out of memory while writing ring references";
1679 goto error;
1680 }
1681 snprintf(path, pathsize, "%s/queue-%u",
1682 dev->nodename, queue->id);
1683 } else {
1684 path = (char *)dev->nodename;
1685 }
1686
1687 /* Write ring references */
1688 err = xenbus_printf(*xbt, path, "tx-ring-ref", "%u",
1689 queue->tx_ring_ref);
1690 if (err) {
1691 message = "writing tx-ring-ref";
1692 goto error;
1693 }
1694
1695 err = xenbus_printf(*xbt, path, "rx-ring-ref", "%u",
1696 queue->rx_ring_ref);
1697 if (err) {
1698 message = "writing rx-ring-ref";
1699 goto error;
1700 }
1701
1702 /* Write event channels; taking into account both shared
1703 * and split event channel scenarios.
1704 */
1705 if (queue->tx_evtchn == queue->rx_evtchn) {
1706 /* Shared event channel */
1707 err = xenbus_printf(*xbt, path,
1708 "event-channel", "%u", queue->tx_evtchn);
1709 if (err) {
1710 message = "writing event-channel";
1711 goto error;
1712 }
1713 } else {
1714 /* Split event channels */
1715 err = xenbus_printf(*xbt, path,
1716 "event-channel-tx", "%u", queue->tx_evtchn);
1717 if (err) {
1718 message = "writing event-channel-tx";
1719 goto error;
1720 }
1721
1722 err = xenbus_printf(*xbt, path,
1723 "event-channel-rx", "%u", queue->rx_evtchn);
1724 if (err) {
1725 message = "writing event-channel-rx";
1726 goto error;
1727 }
1728 }
1729
1730 if (write_hierarchical)
1731 kfree(path);
1732 return 0;
1733
1734error:
1735 if (write_hierarchical)
1736 kfree(path);
1737 xenbus_dev_fatal(dev, err, "%s", message);
1738 return err;
1739}
1740
1741static void xennet_destroy_queues(struct netfront_info *info)
1742{
1743 unsigned int i;
1744
1745 for (i = 0; i < info->netdev->real_num_tx_queues; i++) {
1746 struct netfront_queue *queue = &info->queues[i];
1747
1748 if (netif_running(info->netdev))
1749 napi_disable(&queue->napi);
1750 netif_napi_del(&queue->napi);
1751 }
1752
1753 kfree(info->queues);
1754 info->queues = NULL;
1755}
1756
1757static int xennet_create_queues(struct netfront_info *info,
1758 unsigned int *num_queues)
1759{
1760 unsigned int i;
1761 int ret;
1762
1763 info->queues = kcalloc(*num_queues, sizeof(struct netfront_queue),
1764 GFP_KERNEL);
1765 if (!info->queues)
1766 return -ENOMEM;
1767
1768 for (i = 0; i < *num_queues; i++) {
1769 struct netfront_queue *queue = &info->queues[i];
1770
1771 queue->id = i;
1772 queue->info = info;
1773
1774 ret = xennet_init_queue(queue);
1775 if (ret < 0) {
1776 dev_warn(&info->xbdev->dev,
1777 "only created %d queues\n", i);
1778 *num_queues = i;
1779 break;
1780 }
1781
1782 netif_napi_add(queue->info->netdev, &queue->napi,
1783 xennet_poll, 64);
1784 if (netif_running(info->netdev))
1785 napi_enable(&queue->napi);
1786 }
1787
1788 netif_set_real_num_tx_queues(info->netdev, *num_queues);
1789
1790 if (*num_queues == 0) {
1791 dev_err(&info->xbdev->dev, "no queues\n");
1792 return -EINVAL;
1793 }
1794 return 0;
1795}
1796
1797/* Common code used when first setting up, and when resuming. */
1798static int talk_to_netback(struct xenbus_device *dev,
1799 struct netfront_info *info)
1800{
1801 const char *message;
1802 struct xenbus_transaction xbt;
1803 int err;
1804 unsigned int feature_split_evtchn;
1805 unsigned int i = 0;
1806 unsigned int max_queues = 0;
1807 struct netfront_queue *queue = NULL;
1808 unsigned int num_queues = 1;
1809
1810 info->netdev->irq = 0;
1811
1812 /* Check if backend supports multiple queues */
1813 max_queues = xenbus_read_unsigned(info->xbdev->otherend,
1814 "multi-queue-max-queues", 1);
1815 num_queues = min(max_queues, xennet_max_queues);
1816
1817 /* Check feature-split-event-channels */
1818 feature_split_evtchn = xenbus_read_unsigned(info->xbdev->otherend,
1819 "feature-split-event-channels", 0);
1820
1821 /* Read mac addr. */
1822 err = xen_net_read_mac(dev, info->netdev->dev_addr);
1823 if (err) {
1824 xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename);
1825 goto out_unlocked;
1826 }
1827
1828 rtnl_lock();
1829 if (info->queues)
1830 xennet_destroy_queues(info);
1831
1832 err = xennet_create_queues(info, &num_queues);
1833 if (err < 0) {
1834 xenbus_dev_fatal(dev, err, "creating queues");
1835 kfree(info->queues);
1836 info->queues = NULL;
1837 goto out;
1838 }
1839 rtnl_unlock();
1840
1841 /* Create shared ring, alloc event channel -- for each queue */
1842 for (i = 0; i < num_queues; ++i) {
1843 queue = &info->queues[i];
1844 err = setup_netfront(dev, queue, feature_split_evtchn);
1845 if (err)
1846 goto destroy_ring;
1847 }
1848
1849again:
1850 err = xenbus_transaction_start(&xbt);
1851 if (err) {
1852 xenbus_dev_fatal(dev, err, "starting transaction");
1853 goto destroy_ring;
1854 }
1855
1856 if (xenbus_exists(XBT_NIL,
1857 info->xbdev->otherend, "multi-queue-max-queues")) {
1858 /* Write the number of queues */
1859 err = xenbus_printf(xbt, dev->nodename,
1860 "multi-queue-num-queues", "%u", num_queues);
1861 if (err) {
1862 message = "writing multi-queue-num-queues";
1863 goto abort_transaction_no_dev_fatal;
1864 }
1865 }
1866
1867 if (num_queues == 1) {
1868 err = write_queue_xenstore_keys(&info->queues[0], &xbt, 0); /* flat */
1869 if (err)
1870 goto abort_transaction_no_dev_fatal;
1871 } else {
1872 /* Write the keys for each queue */
1873 for (i = 0; i < num_queues; ++i) {
1874 queue = &info->queues[i];
1875 err = write_queue_xenstore_keys(queue, &xbt, 1); /* hierarchical */
1876 if (err)
1877 goto abort_transaction_no_dev_fatal;
1878 }
1879 }
1880
1881 /* The remaining keys are not queue-specific */
1882 err = xenbus_printf(xbt, dev->nodename, "request-rx-copy", "%u",
1883 1);
1884 if (err) {
1885 message = "writing request-rx-copy";
1886 goto abort_transaction;
1887 }
1888
1889 err = xenbus_printf(xbt, dev->nodename, "feature-rx-notify", "%d", 1);
1890 if (err) {
1891 message = "writing feature-rx-notify";
1892 goto abort_transaction;
1893 }
1894
1895 err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", 1);
1896 if (err) {
1897 message = "writing feature-sg";
1898 goto abort_transaction;
1899 }
1900
1901 err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", "%d", 1);
1902 if (err) {
1903 message = "writing feature-gso-tcpv4";
1904 goto abort_transaction;
1905 }
1906
1907 err = xenbus_write(xbt, dev->nodename, "feature-gso-tcpv6", "1");
1908 if (err) {
1909 message = "writing feature-gso-tcpv6";
1910 goto abort_transaction;
1911 }
1912
1913 err = xenbus_write(xbt, dev->nodename, "feature-ipv6-csum-offload",
1914 "1");
1915 if (err) {
1916 message = "writing feature-ipv6-csum-offload";
1917 goto abort_transaction;
1918 }
1919
1920 err = xenbus_transaction_end(xbt, 0);
1921 if (err) {
1922 if (err == -EAGAIN)
1923 goto again;
1924 xenbus_dev_fatal(dev, err, "completing transaction");
1925 goto destroy_ring;
1926 }
1927
1928 return 0;
1929
1930 abort_transaction:
1931 xenbus_dev_fatal(dev, err, "%s", message);
1932abort_transaction_no_dev_fatal:
1933 xenbus_transaction_end(xbt, 1);
1934 destroy_ring:
1935 xennet_disconnect_backend(info);
1936 rtnl_lock();
1937 xennet_destroy_queues(info);
1938 out:
1939 rtnl_unlock();
1940out_unlocked:
1941 device_unregister(&dev->dev);
1942 return err;
1943}
1944
1945static int xennet_connect(struct net_device *dev)
1946{
1947 struct netfront_info *np = netdev_priv(dev);
1948 unsigned int num_queues = 0;
1949 int err;
1950 unsigned int j = 0;
1951 struct netfront_queue *queue = NULL;
1952
1953 if (!xenbus_read_unsigned(np->xbdev->otherend, "feature-rx-copy", 0)) {
1954 dev_info(&dev->dev,
1955 "backend does not support copying receive path\n");
1956 return -ENODEV;
1957 }
1958
1959 err = talk_to_netback(np->xbdev, np);
1960 if (err)
1961 return err;
1962
1963 /* talk_to_netback() sets the correct number of queues */
1964 num_queues = dev->real_num_tx_queues;
1965
1966 if (dev->reg_state == NETREG_UNINITIALIZED) {
1967 err = register_netdev(dev);
1968 if (err) {
1969 pr_warn("%s: register_netdev err=%d\n", __func__, err);
1970 device_unregister(&np->xbdev->dev);
1971 return err;
1972 }
1973 }
1974
1975 rtnl_lock();
1976 netdev_update_features(dev);
1977 rtnl_unlock();
1978
1979 /*
1980 * All public and private state should now be sane. Get
1981 * ready to start sending and receiving packets and give the driver
1982 * domain a kick because we've probably just requeued some
1983 * packets.
1984 */
1985 netif_carrier_on(np->netdev);
1986 for (j = 0; j < num_queues; ++j) {
1987 queue = &np->queues[j];
1988
1989 notify_remote_via_irq(queue->tx_irq);
1990 if (queue->tx_irq != queue->rx_irq)
1991 notify_remote_via_irq(queue->rx_irq);
1992
1993 spin_lock_irq(&queue->tx_lock);
1994 xennet_tx_buf_gc(queue);
1995 spin_unlock_irq(&queue->tx_lock);
1996
1997 spin_lock_bh(&queue->rx_lock);
1998 xennet_alloc_rx_buffers(queue);
1999 spin_unlock_bh(&queue->rx_lock);
2000 }
2001
2002 return 0;
2003}
2004
2005/**
2006 * Callback received when the backend's state changes.
2007 */
2008static void netback_changed(struct xenbus_device *dev,
2009 enum xenbus_state backend_state)
2010{
2011 struct netfront_info *np = dev_get_drvdata(&dev->dev);
2012 struct net_device *netdev = np->netdev;
2013
2014 dev_dbg(&dev->dev, "%s\n", xenbus_strstate(backend_state));
2015
2016 wake_up_all(&module_wq);
2017
2018 switch (backend_state) {
2019 case XenbusStateInitialising:
2020 case XenbusStateInitialised:
2021 case XenbusStateReconfiguring:
2022 case XenbusStateReconfigured:
2023 case XenbusStateUnknown:
2024 break;
2025
2026 case XenbusStateInitWait:
2027 if (dev->state != XenbusStateInitialising)
2028 break;
2029 if (xennet_connect(netdev) != 0)
2030 break;
2031 xenbus_switch_state(dev, XenbusStateConnected);
2032 break;
2033
2034 case XenbusStateConnected:
2035 netdev_notify_peers(netdev);
2036 break;
2037
2038 case XenbusStateClosed:
2039 if (dev->state == XenbusStateClosed)
2040 break;
2041 /* Fall through - Missed the backend's CLOSING state. */
2042 case XenbusStateClosing:
2043 xenbus_frontend_closed(dev);
2044 break;
2045 }
2046}
2047
2048static const struct xennet_stat {
2049 char name[ETH_GSTRING_LEN];
2050 u16 offset;
2051} xennet_stats[] = {
2052 {
2053 "rx_gso_checksum_fixup",
2054 offsetof(struct netfront_info, rx_gso_checksum_fixup)
2055 },
2056};
2057
2058static int xennet_get_sset_count(struct net_device *dev, int string_set)
2059{
2060 switch (string_set) {
2061 case ETH_SS_STATS:
2062 return ARRAY_SIZE(xennet_stats);
2063 default:
2064 return -EINVAL;
2065 }
2066}
2067
2068static void xennet_get_ethtool_stats(struct net_device *dev,
2069 struct ethtool_stats *stats, u64 * data)
2070{
2071 void *np = netdev_priv(dev);
2072 int i;
2073
2074 for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
2075 data[i] = atomic_read((atomic_t *)(np + xennet_stats[i].offset));
2076}
2077
2078static void xennet_get_strings(struct net_device *dev, u32 stringset, u8 * data)
2079{
2080 int i;
2081
2082 switch (stringset) {
2083 case ETH_SS_STATS:
2084 for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
2085 memcpy(data + i * ETH_GSTRING_LEN,
2086 xennet_stats[i].name, ETH_GSTRING_LEN);
2087 break;
2088 }
2089}
2090
2091static const struct ethtool_ops xennet_ethtool_ops =
2092{
2093 .get_link = ethtool_op_get_link,
2094
2095 .get_sset_count = xennet_get_sset_count,
2096 .get_ethtool_stats = xennet_get_ethtool_stats,
2097 .get_strings = xennet_get_strings,
2098};
2099
2100#ifdef CONFIG_SYSFS
2101static ssize_t show_rxbuf(struct device *dev,
2102 struct device_attribute *attr, char *buf)
2103{
2104 return sprintf(buf, "%lu\n", NET_RX_RING_SIZE);
2105}
2106
2107static ssize_t store_rxbuf(struct device *dev,
2108 struct device_attribute *attr,
2109 const char *buf, size_t len)
2110{
2111 char *endp;
2112 unsigned long target;
2113
2114 if (!capable(CAP_NET_ADMIN))
2115 return -EPERM;
2116
2117 target = simple_strtoul(buf, &endp, 0);
2118 if (endp == buf)
2119 return -EBADMSG;
2120
2121 /* rxbuf_min and rxbuf_max are no longer configurable. */
2122
2123 return len;
2124}
2125
2126static DEVICE_ATTR(rxbuf_min, 0644, show_rxbuf, store_rxbuf);
2127static DEVICE_ATTR(rxbuf_max, 0644, show_rxbuf, store_rxbuf);
2128static DEVICE_ATTR(rxbuf_cur, 0444, show_rxbuf, NULL);
2129
2130static struct attribute *xennet_dev_attrs[] = {
2131 &dev_attr_rxbuf_min.attr,
2132 &dev_attr_rxbuf_max.attr,
2133 &dev_attr_rxbuf_cur.attr,
2134 NULL
2135};
2136
2137static const struct attribute_group xennet_dev_group = {
2138 .attrs = xennet_dev_attrs
2139};
2140#endif /* CONFIG_SYSFS */
2141
2142static int xennet_remove(struct xenbus_device *dev)
2143{
2144 struct netfront_info *info = dev_get_drvdata(&dev->dev);
2145
2146 dev_dbg(&dev->dev, "%s\n", dev->nodename);
2147
2148 if (xenbus_read_driver_state(dev->otherend) != XenbusStateClosed) {
2149 xenbus_switch_state(dev, XenbusStateClosing);
2150 wait_event(module_wq,
2151 xenbus_read_driver_state(dev->otherend) ==
2152 XenbusStateClosing ||
2153 xenbus_read_driver_state(dev->otherend) ==
2154 XenbusStateUnknown);
2155
2156 xenbus_switch_state(dev, XenbusStateClosed);
2157 wait_event(module_wq,
2158 xenbus_read_driver_state(dev->otherend) ==
2159 XenbusStateClosed ||
2160 xenbus_read_driver_state(dev->otherend) ==
2161 XenbusStateUnknown);
2162 }
2163
2164 xennet_disconnect_backend(info);
2165
2166 if (info->netdev->reg_state == NETREG_REGISTERED)
2167 unregister_netdev(info->netdev);
2168
2169 if (info->queues) {
2170 rtnl_lock();
2171 xennet_destroy_queues(info);
2172 rtnl_unlock();
2173 }
2174 xennet_free_netdev(info->netdev);
2175
2176 return 0;
2177}
2178
2179static const struct xenbus_device_id netfront_ids[] = {
2180 { "vif" },
2181 { "" }
2182};
2183
2184static struct xenbus_driver netfront_driver = {
2185 .ids = netfront_ids,
2186 .probe = netfront_probe,
2187 .remove = xennet_remove,
2188 .resume = netfront_resume,
2189 .otherend_changed = netback_changed,
2190};
2191
2192static int __init netif_init(void)
2193{
2194 if (!xen_domain())
2195 return -ENODEV;
2196
2197 if (!xen_has_pv_nic_devices())
2198 return -ENODEV;
2199
2200 pr_info("Initialising Xen virtual ethernet driver\n");
2201
2202 /* Allow as many queues as there are CPUs inut max. 8 if user has not
2203 * specified a value.
2204 */
2205 if (xennet_max_queues == 0)
2206 xennet_max_queues = min_t(unsigned int, MAX_QUEUES_DEFAULT,
2207 num_online_cpus());
2208
2209 return xenbus_register_frontend(&netfront_driver);
2210}
2211module_init(netif_init);
2212
2213
2214static void __exit netif_exit(void)
2215{
2216 xenbus_unregister_driver(&netfront_driver);
2217}
2218module_exit(netif_exit);
2219
2220MODULE_DESCRIPTION("Xen virtual network device frontend");
2221MODULE_LICENSE("GPL");
2222MODULE_ALIAS("xen:vif");
2223MODULE_ALIAS("xennet");