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#include <linux/bpf.h>
48#include <net/page_pool.h>
49#include <linux/bpf_trace.h>
50
51#include <xen/xen.h>
52#include <xen/xenbus.h>
53#include <xen/events.h>
54#include <xen/page.h>
55#include <xen/platform_pci.h>
56#include <xen/grant_table.h>
57
58#include <xen/interface/io/netif.h>
59#include <xen/interface/memory.h>
60#include <xen/interface/grant_table.h>
61
62/* Module parameters */
63#define MAX_QUEUES_DEFAULT 8
64static unsigned int xennet_max_queues;
65module_param_named(max_queues, xennet_max_queues, uint, 0644);
66MODULE_PARM_DESC(max_queues,
67 "Maximum number of queues per virtual interface");
68
69static bool __read_mostly xennet_trusted = true;
70module_param_named(trusted, xennet_trusted, bool, 0644);
71MODULE_PARM_DESC(trusted, "Is the backend trusted");
72
73#define XENNET_TIMEOUT (5 * HZ)
74
75static const struct ethtool_ops xennet_ethtool_ops;
76
77struct netfront_cb {
78 int pull_to;
79};
80
81#define NETFRONT_SKB_CB(skb) ((struct netfront_cb *)((skb)->cb))
82
83#define RX_COPY_THRESHOLD 256
84
85#define NET_TX_RING_SIZE __CONST_RING_SIZE(xen_netif_tx, XEN_PAGE_SIZE)
86#define NET_RX_RING_SIZE __CONST_RING_SIZE(xen_netif_rx, XEN_PAGE_SIZE)
87
88/* Minimum number of Rx slots (includes slot for GSO metadata). */
89#define NET_RX_SLOTS_MIN (XEN_NETIF_NR_SLOTS_MIN + 1)
90
91/* Queue name is interface name with "-qNNN" appended */
92#define QUEUE_NAME_SIZE (IFNAMSIZ + 6)
93
94/* IRQ name is queue name with "-tx" or "-rx" appended */
95#define IRQ_NAME_SIZE (QUEUE_NAME_SIZE + 3)
96
97static DECLARE_WAIT_QUEUE_HEAD(module_wq);
98
99struct netfront_stats {
100 u64 packets;
101 u64 bytes;
102 struct u64_stats_sync syncp;
103};
104
105struct netfront_info;
106
107struct netfront_queue {
108 unsigned int id; /* Queue ID, 0-based */
109 char name[QUEUE_NAME_SIZE]; /* DEVNAME-qN */
110 struct netfront_info *info;
111
112 struct bpf_prog __rcu *xdp_prog;
113
114 struct napi_struct napi;
115
116 /* Split event channels support, tx_* == rx_* when using
117 * single event channel.
118 */
119 unsigned int tx_evtchn, rx_evtchn;
120 unsigned int tx_irq, rx_irq;
121 /* Only used when split event channels support is enabled */
122 char tx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-tx */
123 char rx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-rx */
124
125 spinlock_t tx_lock;
126 struct xen_netif_tx_front_ring tx;
127 int tx_ring_ref;
128
129 /*
130 * {tx,rx}_skbs store outstanding skbuffs. Free tx_skb entries
131 * are linked from tx_skb_freelist through tx_link.
132 */
133 struct sk_buff *tx_skbs[NET_TX_RING_SIZE];
134 unsigned short tx_link[NET_TX_RING_SIZE];
135#define TX_LINK_NONE 0xffff
136#define TX_PENDING 0xfffe
137 grant_ref_t gref_tx_head;
138 grant_ref_t grant_tx_ref[NET_TX_RING_SIZE];
139 struct page *grant_tx_page[NET_TX_RING_SIZE];
140 unsigned tx_skb_freelist;
141 unsigned int tx_pend_queue;
142
143 spinlock_t rx_lock ____cacheline_aligned_in_smp;
144 struct xen_netif_rx_front_ring rx;
145 int rx_ring_ref;
146
147 struct timer_list rx_refill_timer;
148
149 struct sk_buff *rx_skbs[NET_RX_RING_SIZE];
150 grant_ref_t gref_rx_head;
151 grant_ref_t grant_rx_ref[NET_RX_RING_SIZE];
152
153 unsigned int rx_rsp_unconsumed;
154 spinlock_t rx_cons_lock;
155
156 struct page_pool *page_pool;
157 struct xdp_rxq_info xdp_rxq;
158};
159
160struct netfront_info {
161 struct list_head list;
162 struct net_device *netdev;
163
164 struct xenbus_device *xbdev;
165
166 /* Multi-queue support */
167 struct netfront_queue *queues;
168
169 /* Statistics */
170 struct netfront_stats __percpu *rx_stats;
171 struct netfront_stats __percpu *tx_stats;
172
173 /* XDP state */
174 bool netback_has_xdp_headroom;
175 bool netfront_xdp_enabled;
176
177 /* Is device behaving sane? */
178 bool broken;
179
180 /* Should skbs be bounced into a zeroed buffer? */
181 bool bounce;
182
183 atomic_t rx_gso_checksum_fixup;
184};
185
186struct netfront_rx_info {
187 struct xen_netif_rx_response rx;
188 struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX - 1];
189};
190
191/*
192 * Access macros for acquiring freeing slots in tx_skbs[].
193 */
194
195static void add_id_to_list(unsigned *head, unsigned short *list,
196 unsigned short id)
197{
198 list[id] = *head;
199 *head = id;
200}
201
202static unsigned short get_id_from_list(unsigned *head, unsigned short *list)
203{
204 unsigned int id = *head;
205
206 if (id != TX_LINK_NONE) {
207 *head = list[id];
208 list[id] = TX_LINK_NONE;
209 }
210 return id;
211}
212
213static int xennet_rxidx(RING_IDX idx)
214{
215 return idx & (NET_RX_RING_SIZE - 1);
216}
217
218static struct sk_buff *xennet_get_rx_skb(struct netfront_queue *queue,
219 RING_IDX ri)
220{
221 int i = xennet_rxidx(ri);
222 struct sk_buff *skb = queue->rx_skbs[i];
223 queue->rx_skbs[i] = NULL;
224 return skb;
225}
226
227static grant_ref_t xennet_get_rx_ref(struct netfront_queue *queue,
228 RING_IDX ri)
229{
230 int i = xennet_rxidx(ri);
231 grant_ref_t ref = queue->grant_rx_ref[i];
232 queue->grant_rx_ref[i] = INVALID_GRANT_REF;
233 return ref;
234}
235
236#ifdef CONFIG_SYSFS
237static const struct attribute_group xennet_dev_group;
238#endif
239
240static bool xennet_can_sg(struct net_device *dev)
241{
242 return dev->features & NETIF_F_SG;
243}
244
245
246static void rx_refill_timeout(struct timer_list *t)
247{
248 struct netfront_queue *queue = from_timer(queue, t, rx_refill_timer);
249 napi_schedule(&queue->napi);
250}
251
252static int netfront_tx_slot_available(struct netfront_queue *queue)
253{
254 return (queue->tx.req_prod_pvt - queue->tx.rsp_cons) <
255 (NET_TX_RING_SIZE - XEN_NETIF_NR_SLOTS_MIN - 1);
256}
257
258static void xennet_maybe_wake_tx(struct netfront_queue *queue)
259{
260 struct net_device *dev = queue->info->netdev;
261 struct netdev_queue *dev_queue = netdev_get_tx_queue(dev, queue->id);
262
263 if (unlikely(netif_tx_queue_stopped(dev_queue)) &&
264 netfront_tx_slot_available(queue) &&
265 likely(netif_running(dev)))
266 netif_tx_wake_queue(netdev_get_tx_queue(dev, queue->id));
267}
268
269
270static struct sk_buff *xennet_alloc_one_rx_buffer(struct netfront_queue *queue)
271{
272 struct sk_buff *skb;
273 struct page *page;
274
275 skb = __netdev_alloc_skb(queue->info->netdev,
276 RX_COPY_THRESHOLD + NET_IP_ALIGN,
277 GFP_ATOMIC | __GFP_NOWARN);
278 if (unlikely(!skb))
279 return NULL;
280
281 page = page_pool_alloc_pages(queue->page_pool,
282 GFP_ATOMIC | __GFP_NOWARN | __GFP_ZERO);
283 if (unlikely(!page)) {
284 kfree_skb(skb);
285 return NULL;
286 }
287 skb_add_rx_frag(skb, 0, page, 0, 0, PAGE_SIZE);
288
289 /* Align ip header to a 16 bytes boundary */
290 skb_reserve(skb, NET_IP_ALIGN);
291 skb->dev = queue->info->netdev;
292
293 return skb;
294}
295
296
297static void xennet_alloc_rx_buffers(struct netfront_queue *queue)
298{
299 RING_IDX req_prod = queue->rx.req_prod_pvt;
300 int notify;
301 int err = 0;
302
303 if (unlikely(!netif_carrier_ok(queue->info->netdev)))
304 return;
305
306 for (req_prod = queue->rx.req_prod_pvt;
307 req_prod - queue->rx.rsp_cons < NET_RX_RING_SIZE;
308 req_prod++) {
309 struct sk_buff *skb;
310 unsigned short id;
311 grant_ref_t ref;
312 struct page *page;
313 struct xen_netif_rx_request *req;
314
315 skb = xennet_alloc_one_rx_buffer(queue);
316 if (!skb) {
317 err = -ENOMEM;
318 break;
319 }
320
321 id = xennet_rxidx(req_prod);
322
323 BUG_ON(queue->rx_skbs[id]);
324 queue->rx_skbs[id] = skb;
325
326 ref = gnttab_claim_grant_reference(&queue->gref_rx_head);
327 WARN_ON_ONCE(IS_ERR_VALUE((unsigned long)(int)ref));
328 queue->grant_rx_ref[id] = ref;
329
330 page = skb_frag_page(&skb_shinfo(skb)->frags[0]);
331
332 req = RING_GET_REQUEST(&queue->rx, req_prod);
333 gnttab_page_grant_foreign_access_ref_one(ref,
334 queue->info->xbdev->otherend_id,
335 page,
336 0);
337 req->id = id;
338 req->gref = ref;
339 }
340
341 queue->rx.req_prod_pvt = req_prod;
342
343 /* Try again later if there are not enough requests or skb allocation
344 * failed.
345 * Enough requests is quantified as the sum of newly created slots and
346 * the unconsumed slots at the backend.
347 */
348 if (req_prod - queue->rx.rsp_cons < NET_RX_SLOTS_MIN ||
349 unlikely(err)) {
350 mod_timer(&queue->rx_refill_timer, jiffies + (HZ/10));
351 return;
352 }
353
354 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->rx, notify);
355 if (notify)
356 notify_remote_via_irq(queue->rx_irq);
357}
358
359static int xennet_open(struct net_device *dev)
360{
361 struct netfront_info *np = netdev_priv(dev);
362 unsigned int num_queues = dev->real_num_tx_queues;
363 unsigned int i = 0;
364 struct netfront_queue *queue = NULL;
365
366 if (!np->queues || np->broken)
367 return -ENODEV;
368
369 for (i = 0; i < num_queues; ++i) {
370 queue = &np->queues[i];
371 napi_enable(&queue->napi);
372
373 spin_lock_bh(&queue->rx_lock);
374 if (netif_carrier_ok(dev)) {
375 xennet_alloc_rx_buffers(queue);
376 queue->rx.sring->rsp_event = queue->rx.rsp_cons + 1;
377 if (RING_HAS_UNCONSUMED_RESPONSES(&queue->rx))
378 napi_schedule(&queue->napi);
379 }
380 spin_unlock_bh(&queue->rx_lock);
381 }
382
383 netif_tx_start_all_queues(dev);
384
385 return 0;
386}
387
388static bool xennet_tx_buf_gc(struct netfront_queue *queue)
389{
390 RING_IDX cons, prod;
391 unsigned short id;
392 struct sk_buff *skb;
393 bool more_to_do;
394 bool work_done = false;
395 const struct device *dev = &queue->info->netdev->dev;
396
397 BUG_ON(!netif_carrier_ok(queue->info->netdev));
398
399 do {
400 prod = queue->tx.sring->rsp_prod;
401 if (RING_RESPONSE_PROD_OVERFLOW(&queue->tx, prod)) {
402 dev_alert(dev, "Illegal number of responses %u\n",
403 prod - queue->tx.rsp_cons);
404 goto err;
405 }
406 rmb(); /* Ensure we see responses up to 'rp'. */
407
408 for (cons = queue->tx.rsp_cons; cons != prod; cons++) {
409 struct xen_netif_tx_response txrsp;
410
411 work_done = true;
412
413 RING_COPY_RESPONSE(&queue->tx, cons, &txrsp);
414 if (txrsp.status == XEN_NETIF_RSP_NULL)
415 continue;
416
417 id = txrsp.id;
418 if (id >= RING_SIZE(&queue->tx)) {
419 dev_alert(dev,
420 "Response has incorrect id (%u)\n",
421 id);
422 goto err;
423 }
424 if (queue->tx_link[id] != TX_PENDING) {
425 dev_alert(dev,
426 "Response for inactive request\n");
427 goto err;
428 }
429
430 queue->tx_link[id] = TX_LINK_NONE;
431 skb = queue->tx_skbs[id];
432 queue->tx_skbs[id] = NULL;
433 if (unlikely(!gnttab_end_foreign_access_ref(
434 queue->grant_tx_ref[id]))) {
435 dev_alert(dev,
436 "Grant still in use by backend domain\n");
437 goto err;
438 }
439 gnttab_release_grant_reference(
440 &queue->gref_tx_head, queue->grant_tx_ref[id]);
441 queue->grant_tx_ref[id] = INVALID_GRANT_REF;
442 queue->grant_tx_page[id] = NULL;
443 add_id_to_list(&queue->tx_skb_freelist, queue->tx_link, id);
444 dev_kfree_skb_irq(skb);
445 }
446
447 queue->tx.rsp_cons = prod;
448
449 RING_FINAL_CHECK_FOR_RESPONSES(&queue->tx, more_to_do);
450 } while (more_to_do);
451
452 xennet_maybe_wake_tx(queue);
453
454 return work_done;
455
456 err:
457 queue->info->broken = true;
458 dev_alert(dev, "Disabled for further use\n");
459
460 return work_done;
461}
462
463struct xennet_gnttab_make_txreq {
464 struct netfront_queue *queue;
465 struct sk_buff *skb;
466 struct page *page;
467 struct xen_netif_tx_request *tx; /* Last request on ring page */
468 struct xen_netif_tx_request tx_local; /* Last request local copy*/
469 unsigned int size;
470};
471
472static void xennet_tx_setup_grant(unsigned long gfn, unsigned int offset,
473 unsigned int len, void *data)
474{
475 struct xennet_gnttab_make_txreq *info = data;
476 unsigned int id;
477 struct xen_netif_tx_request *tx;
478 grant_ref_t ref;
479 /* convenient aliases */
480 struct page *page = info->page;
481 struct netfront_queue *queue = info->queue;
482 struct sk_buff *skb = info->skb;
483
484 id = get_id_from_list(&queue->tx_skb_freelist, queue->tx_link);
485 tx = RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++);
486 ref = gnttab_claim_grant_reference(&queue->gref_tx_head);
487 WARN_ON_ONCE(IS_ERR_VALUE((unsigned long)(int)ref));
488
489 gnttab_grant_foreign_access_ref(ref, queue->info->xbdev->otherend_id,
490 gfn, GNTMAP_readonly);
491
492 queue->tx_skbs[id] = skb;
493 queue->grant_tx_page[id] = page;
494 queue->grant_tx_ref[id] = ref;
495
496 info->tx_local.id = id;
497 info->tx_local.gref = ref;
498 info->tx_local.offset = offset;
499 info->tx_local.size = len;
500 info->tx_local.flags = 0;
501
502 *tx = info->tx_local;
503
504 /*
505 * Put the request in the pending queue, it will be set to be pending
506 * when the producer index is about to be raised.
507 */
508 add_id_to_list(&queue->tx_pend_queue, queue->tx_link, id);
509
510 info->tx = tx;
511 info->size += info->tx_local.size;
512}
513
514static struct xen_netif_tx_request *xennet_make_first_txreq(
515 struct xennet_gnttab_make_txreq *info,
516 unsigned int offset, unsigned int len)
517{
518 info->size = 0;
519
520 gnttab_for_one_grant(info->page, offset, len, xennet_tx_setup_grant, info);
521
522 return info->tx;
523}
524
525static void xennet_make_one_txreq(unsigned long gfn, unsigned int offset,
526 unsigned int len, void *data)
527{
528 struct xennet_gnttab_make_txreq *info = data;
529
530 info->tx->flags |= XEN_NETTXF_more_data;
531 skb_get(info->skb);
532 xennet_tx_setup_grant(gfn, offset, len, data);
533}
534
535static void xennet_make_txreqs(
536 struct xennet_gnttab_make_txreq *info,
537 struct page *page,
538 unsigned int offset, unsigned int len)
539{
540 /* Skip unused frames from start of page */
541 page += offset >> PAGE_SHIFT;
542 offset &= ~PAGE_MASK;
543
544 while (len) {
545 info->page = page;
546 info->size = 0;
547
548 gnttab_foreach_grant_in_range(page, offset, len,
549 xennet_make_one_txreq,
550 info);
551
552 page++;
553 offset = 0;
554 len -= info->size;
555 }
556}
557
558/*
559 * Count how many ring slots are required to send this skb. Each frag
560 * might be a compound page.
561 */
562static int xennet_count_skb_slots(struct sk_buff *skb)
563{
564 int i, frags = skb_shinfo(skb)->nr_frags;
565 int slots;
566
567 slots = gnttab_count_grant(offset_in_page(skb->data),
568 skb_headlen(skb));
569
570 for (i = 0; i < frags; i++) {
571 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
572 unsigned long size = skb_frag_size(frag);
573 unsigned long offset = skb_frag_off(frag);
574
575 /* Skip unused frames from start of page */
576 offset &= ~PAGE_MASK;
577
578 slots += gnttab_count_grant(offset, size);
579 }
580
581 return slots;
582}
583
584static u16 xennet_select_queue(struct net_device *dev, struct sk_buff *skb,
585 struct net_device *sb_dev)
586{
587 unsigned int num_queues = dev->real_num_tx_queues;
588 u32 hash;
589 u16 queue_idx;
590
591 /* First, check if there is only one queue */
592 if (num_queues == 1) {
593 queue_idx = 0;
594 } else {
595 hash = skb_get_hash(skb);
596 queue_idx = hash % num_queues;
597 }
598
599 return queue_idx;
600}
601
602static void xennet_mark_tx_pending(struct netfront_queue *queue)
603{
604 unsigned int i;
605
606 while ((i = get_id_from_list(&queue->tx_pend_queue, queue->tx_link)) !=
607 TX_LINK_NONE)
608 queue->tx_link[i] = TX_PENDING;
609}
610
611static int xennet_xdp_xmit_one(struct net_device *dev,
612 struct netfront_queue *queue,
613 struct xdp_frame *xdpf)
614{
615 struct netfront_info *np = netdev_priv(dev);
616 struct netfront_stats *tx_stats = this_cpu_ptr(np->tx_stats);
617 struct xennet_gnttab_make_txreq info = {
618 .queue = queue,
619 .skb = NULL,
620 .page = virt_to_page(xdpf->data),
621 };
622 int notify;
623
624 xennet_make_first_txreq(&info,
625 offset_in_page(xdpf->data),
626 xdpf->len);
627
628 xennet_mark_tx_pending(queue);
629
630 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->tx, notify);
631 if (notify)
632 notify_remote_via_irq(queue->tx_irq);
633
634 u64_stats_update_begin(&tx_stats->syncp);
635 tx_stats->bytes += xdpf->len;
636 tx_stats->packets++;
637 u64_stats_update_end(&tx_stats->syncp);
638
639 xennet_tx_buf_gc(queue);
640
641 return 0;
642}
643
644static int xennet_xdp_xmit(struct net_device *dev, int n,
645 struct xdp_frame **frames, u32 flags)
646{
647 unsigned int num_queues = dev->real_num_tx_queues;
648 struct netfront_info *np = netdev_priv(dev);
649 struct netfront_queue *queue = NULL;
650 unsigned long irq_flags;
651 int nxmit = 0;
652 int i;
653
654 if (unlikely(np->broken))
655 return -ENODEV;
656 if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
657 return -EINVAL;
658
659 queue = &np->queues[smp_processor_id() % num_queues];
660
661 spin_lock_irqsave(&queue->tx_lock, irq_flags);
662 for (i = 0; i < n; i++) {
663 struct xdp_frame *xdpf = frames[i];
664
665 if (!xdpf)
666 continue;
667 if (xennet_xdp_xmit_one(dev, queue, xdpf))
668 break;
669 nxmit++;
670 }
671 spin_unlock_irqrestore(&queue->tx_lock, irq_flags);
672
673 return nxmit;
674}
675
676static struct sk_buff *bounce_skb(const struct sk_buff *skb)
677{
678 unsigned int headerlen = skb_headroom(skb);
679 /* Align size to allocate full pages and avoid contiguous data leaks */
680 unsigned int size = ALIGN(skb_end_offset(skb) + skb->data_len,
681 XEN_PAGE_SIZE);
682 struct sk_buff *n = alloc_skb(size, GFP_ATOMIC | __GFP_ZERO);
683
684 if (!n)
685 return NULL;
686
687 if (!IS_ALIGNED((uintptr_t)n->head, XEN_PAGE_SIZE)) {
688 WARN_ONCE(1, "misaligned skb allocated\n");
689 kfree_skb(n);
690 return NULL;
691 }
692
693 /* Set the data pointer */
694 skb_reserve(n, headerlen);
695 /* Set the tail pointer and length */
696 skb_put(n, skb->len);
697
698 BUG_ON(skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len));
699
700 skb_copy_header(n, skb);
701 return n;
702}
703
704#define MAX_XEN_SKB_FRAGS (65536 / XEN_PAGE_SIZE + 1)
705
706static netdev_tx_t xennet_start_xmit(struct sk_buff *skb, struct net_device *dev)
707{
708 struct netfront_info *np = netdev_priv(dev);
709 struct netfront_stats *tx_stats = this_cpu_ptr(np->tx_stats);
710 struct xen_netif_tx_request *first_tx;
711 unsigned int i;
712 int notify;
713 int slots;
714 struct page *page;
715 unsigned int offset;
716 unsigned int len;
717 unsigned long flags;
718 struct netfront_queue *queue = NULL;
719 struct xennet_gnttab_make_txreq info = { };
720 unsigned int num_queues = dev->real_num_tx_queues;
721 u16 queue_index;
722 struct sk_buff *nskb;
723
724 /* Drop the packet if no queues are set up */
725 if (num_queues < 1)
726 goto drop;
727 if (unlikely(np->broken))
728 goto drop;
729 /* Determine which queue to transmit this SKB on */
730 queue_index = skb_get_queue_mapping(skb);
731 queue = &np->queues[queue_index];
732
733 /* If skb->len is too big for wire format, drop skb and alert
734 * user about misconfiguration.
735 */
736 if (unlikely(skb->len > XEN_NETIF_MAX_TX_SIZE)) {
737 net_alert_ratelimited(
738 "xennet: skb->len = %u, too big for wire format\n",
739 skb->len);
740 goto drop;
741 }
742
743 slots = xennet_count_skb_slots(skb);
744 if (unlikely(slots > MAX_XEN_SKB_FRAGS + 1)) {
745 net_dbg_ratelimited("xennet: skb rides the rocket: %d slots, %d bytes\n",
746 slots, skb->len);
747 if (skb_linearize(skb))
748 goto drop;
749 }
750
751 page = virt_to_page(skb->data);
752 offset = offset_in_page(skb->data);
753
754 /* The first req should be at least ETH_HLEN size or the packet will be
755 * dropped by netback.
756 *
757 * If the backend is not trusted bounce all data to zeroed pages to
758 * avoid exposing contiguous data on the granted page not belonging to
759 * the skb.
760 */
761 if (np->bounce || unlikely(PAGE_SIZE - offset < ETH_HLEN)) {
762 nskb = bounce_skb(skb);
763 if (!nskb)
764 goto drop;
765 dev_consume_skb_any(skb);
766 skb = nskb;
767 page = virt_to_page(skb->data);
768 offset = offset_in_page(skb->data);
769 }
770
771 len = skb_headlen(skb);
772
773 spin_lock_irqsave(&queue->tx_lock, flags);
774
775 if (unlikely(!netif_carrier_ok(dev) ||
776 (slots > 1 && !xennet_can_sg(dev)) ||
777 netif_needs_gso(skb, netif_skb_features(skb)))) {
778 spin_unlock_irqrestore(&queue->tx_lock, flags);
779 goto drop;
780 }
781
782 /* First request for the linear area. */
783 info.queue = queue;
784 info.skb = skb;
785 info.page = page;
786 first_tx = xennet_make_first_txreq(&info, offset, len);
787 offset += info.tx_local.size;
788 if (offset == PAGE_SIZE) {
789 page++;
790 offset = 0;
791 }
792 len -= info.tx_local.size;
793
794 if (skb->ip_summed == CHECKSUM_PARTIAL)
795 /* local packet? */
796 first_tx->flags |= XEN_NETTXF_csum_blank |
797 XEN_NETTXF_data_validated;
798 else if (skb->ip_summed == CHECKSUM_UNNECESSARY)
799 /* remote but checksummed. */
800 first_tx->flags |= XEN_NETTXF_data_validated;
801
802 /* Optional extra info after the first request. */
803 if (skb_shinfo(skb)->gso_size) {
804 struct xen_netif_extra_info *gso;
805
806 gso = (struct xen_netif_extra_info *)
807 RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++);
808
809 first_tx->flags |= XEN_NETTXF_extra_info;
810
811 gso->u.gso.size = skb_shinfo(skb)->gso_size;
812 gso->u.gso.type = (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) ?
813 XEN_NETIF_GSO_TYPE_TCPV6 :
814 XEN_NETIF_GSO_TYPE_TCPV4;
815 gso->u.gso.pad = 0;
816 gso->u.gso.features = 0;
817
818 gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
819 gso->flags = 0;
820 }
821
822 /* Requests for the rest of the linear area. */
823 xennet_make_txreqs(&info, page, offset, len);
824
825 /* Requests for all the frags. */
826 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
827 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
828 xennet_make_txreqs(&info, skb_frag_page(frag),
829 skb_frag_off(frag),
830 skb_frag_size(frag));
831 }
832
833 /* First request has the packet length. */
834 first_tx->size = skb->len;
835
836 /* timestamp packet in software */
837 skb_tx_timestamp(skb);
838
839 xennet_mark_tx_pending(queue);
840
841 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->tx, notify);
842 if (notify)
843 notify_remote_via_irq(queue->tx_irq);
844
845 u64_stats_update_begin(&tx_stats->syncp);
846 tx_stats->bytes += skb->len;
847 tx_stats->packets++;
848 u64_stats_update_end(&tx_stats->syncp);
849
850 /* Note: It is not safe to access skb after xennet_tx_buf_gc()! */
851 xennet_tx_buf_gc(queue);
852
853 if (!netfront_tx_slot_available(queue))
854 netif_tx_stop_queue(netdev_get_tx_queue(dev, queue->id));
855
856 spin_unlock_irqrestore(&queue->tx_lock, flags);
857
858 return NETDEV_TX_OK;
859
860 drop:
861 dev->stats.tx_dropped++;
862 dev_kfree_skb_any(skb);
863 return NETDEV_TX_OK;
864}
865
866static int xennet_close(struct net_device *dev)
867{
868 struct netfront_info *np = netdev_priv(dev);
869 unsigned int num_queues = dev->real_num_tx_queues;
870 unsigned int i;
871 struct netfront_queue *queue;
872 netif_tx_stop_all_queues(np->netdev);
873 for (i = 0; i < num_queues; ++i) {
874 queue = &np->queues[i];
875 napi_disable(&queue->napi);
876 }
877 return 0;
878}
879
880static void xennet_destroy_queues(struct netfront_info *info)
881{
882 unsigned int i;
883
884 for (i = 0; i < info->netdev->real_num_tx_queues; i++) {
885 struct netfront_queue *queue = &info->queues[i];
886
887 if (netif_running(info->netdev))
888 napi_disable(&queue->napi);
889 netif_napi_del(&queue->napi);
890 }
891
892 kfree(info->queues);
893 info->queues = NULL;
894}
895
896static void xennet_uninit(struct net_device *dev)
897{
898 struct netfront_info *np = netdev_priv(dev);
899 xennet_destroy_queues(np);
900}
901
902static void xennet_set_rx_rsp_cons(struct netfront_queue *queue, RING_IDX val)
903{
904 unsigned long flags;
905
906 spin_lock_irqsave(&queue->rx_cons_lock, flags);
907 queue->rx.rsp_cons = val;
908 queue->rx_rsp_unconsumed = XEN_RING_NR_UNCONSUMED_RESPONSES(&queue->rx);
909 spin_unlock_irqrestore(&queue->rx_cons_lock, flags);
910}
911
912static void xennet_move_rx_slot(struct netfront_queue *queue, struct sk_buff *skb,
913 grant_ref_t ref)
914{
915 int new = xennet_rxidx(queue->rx.req_prod_pvt);
916
917 BUG_ON(queue->rx_skbs[new]);
918 queue->rx_skbs[new] = skb;
919 queue->grant_rx_ref[new] = ref;
920 RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->id = new;
921 RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->gref = ref;
922 queue->rx.req_prod_pvt++;
923}
924
925static int xennet_get_extras(struct netfront_queue *queue,
926 struct xen_netif_extra_info *extras,
927 RING_IDX rp)
928
929{
930 struct xen_netif_extra_info extra;
931 struct device *dev = &queue->info->netdev->dev;
932 RING_IDX cons = queue->rx.rsp_cons;
933 int err = 0;
934
935 do {
936 struct sk_buff *skb;
937 grant_ref_t ref;
938
939 if (unlikely(cons + 1 == rp)) {
940 if (net_ratelimit())
941 dev_warn(dev, "Missing extra info\n");
942 err = -EBADR;
943 break;
944 }
945
946 RING_COPY_RESPONSE(&queue->rx, ++cons, &extra);
947
948 if (unlikely(!extra.type ||
949 extra.type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
950 if (net_ratelimit())
951 dev_warn(dev, "Invalid extra type: %d\n",
952 extra.type);
953 err = -EINVAL;
954 } else {
955 extras[extra.type - 1] = extra;
956 }
957
958 skb = xennet_get_rx_skb(queue, cons);
959 ref = xennet_get_rx_ref(queue, cons);
960 xennet_move_rx_slot(queue, skb, ref);
961 } while (extra.flags & XEN_NETIF_EXTRA_FLAG_MORE);
962
963 xennet_set_rx_rsp_cons(queue, cons);
964 return err;
965}
966
967static u32 xennet_run_xdp(struct netfront_queue *queue, struct page *pdata,
968 struct xen_netif_rx_response *rx, struct bpf_prog *prog,
969 struct xdp_buff *xdp, bool *need_xdp_flush)
970{
971 struct xdp_frame *xdpf;
972 u32 len = rx->status;
973 u32 act;
974 int err;
975
976 xdp_init_buff(xdp, XEN_PAGE_SIZE - XDP_PACKET_HEADROOM,
977 &queue->xdp_rxq);
978 xdp_prepare_buff(xdp, page_address(pdata), XDP_PACKET_HEADROOM,
979 len, false);
980
981 act = bpf_prog_run_xdp(prog, xdp);
982 switch (act) {
983 case XDP_TX:
984 get_page(pdata);
985 xdpf = xdp_convert_buff_to_frame(xdp);
986 err = xennet_xdp_xmit(queue->info->netdev, 1, &xdpf, 0);
987 if (unlikely(!err))
988 xdp_return_frame_rx_napi(xdpf);
989 else if (unlikely(err < 0))
990 trace_xdp_exception(queue->info->netdev, prog, act);
991 break;
992 case XDP_REDIRECT:
993 get_page(pdata);
994 err = xdp_do_redirect(queue->info->netdev, xdp, prog);
995 *need_xdp_flush = true;
996 if (unlikely(err))
997 trace_xdp_exception(queue->info->netdev, prog, act);
998 break;
999 case XDP_PASS:
1000 case XDP_DROP:
1001 break;
1002
1003 case XDP_ABORTED:
1004 trace_xdp_exception(queue->info->netdev, prog, act);
1005 break;
1006
1007 default:
1008 bpf_warn_invalid_xdp_action(queue->info->netdev, prog, act);
1009 }
1010
1011 return act;
1012}
1013
1014static int xennet_get_responses(struct netfront_queue *queue,
1015 struct netfront_rx_info *rinfo, RING_IDX rp,
1016 struct sk_buff_head *list,
1017 bool *need_xdp_flush)
1018{
1019 struct xen_netif_rx_response *rx = &rinfo->rx, rx_local;
1020 int max = XEN_NETIF_NR_SLOTS_MIN + (rx->status <= RX_COPY_THRESHOLD);
1021 RING_IDX cons = queue->rx.rsp_cons;
1022 struct sk_buff *skb = xennet_get_rx_skb(queue, cons);
1023 struct xen_netif_extra_info *extras = rinfo->extras;
1024 grant_ref_t ref = xennet_get_rx_ref(queue, cons);
1025 struct device *dev = &queue->info->netdev->dev;
1026 struct bpf_prog *xdp_prog;
1027 struct xdp_buff xdp;
1028 int slots = 1;
1029 int err = 0;
1030 u32 verdict;
1031
1032 if (rx->flags & XEN_NETRXF_extra_info) {
1033 err = xennet_get_extras(queue, extras, rp);
1034 if (!err) {
1035 if (extras[XEN_NETIF_EXTRA_TYPE_XDP - 1].type) {
1036 struct xen_netif_extra_info *xdp;
1037
1038 xdp = &extras[XEN_NETIF_EXTRA_TYPE_XDP - 1];
1039 rx->offset = xdp->u.xdp.headroom;
1040 }
1041 }
1042 cons = queue->rx.rsp_cons;
1043 }
1044
1045 for (;;) {
1046 /*
1047 * This definitely indicates a bug, either in this driver or in
1048 * the backend driver. In future this should flag the bad
1049 * situation to the system controller to reboot the backend.
1050 */
1051 if (ref == INVALID_GRANT_REF) {
1052 if (net_ratelimit())
1053 dev_warn(dev, "Bad rx response id %d.\n",
1054 rx->id);
1055 err = -EINVAL;
1056 goto next;
1057 }
1058
1059 if (unlikely(rx->status < 0 ||
1060 rx->offset + rx->status > XEN_PAGE_SIZE)) {
1061 if (net_ratelimit())
1062 dev_warn(dev, "rx->offset: %u, size: %d\n",
1063 rx->offset, rx->status);
1064 xennet_move_rx_slot(queue, skb, ref);
1065 err = -EINVAL;
1066 goto next;
1067 }
1068
1069 if (!gnttab_end_foreign_access_ref(ref)) {
1070 dev_alert(dev,
1071 "Grant still in use by backend domain\n");
1072 queue->info->broken = true;
1073 dev_alert(dev, "Disabled for further use\n");
1074 return -EINVAL;
1075 }
1076
1077 gnttab_release_grant_reference(&queue->gref_rx_head, ref);
1078
1079 rcu_read_lock();
1080 xdp_prog = rcu_dereference(queue->xdp_prog);
1081 if (xdp_prog) {
1082 if (!(rx->flags & XEN_NETRXF_more_data)) {
1083 /* currently only a single page contains data */
1084 verdict = xennet_run_xdp(queue,
1085 skb_frag_page(&skb_shinfo(skb)->frags[0]),
1086 rx, xdp_prog, &xdp, need_xdp_flush);
1087 if (verdict != XDP_PASS)
1088 err = -EINVAL;
1089 } else {
1090 /* drop the frame */
1091 err = -EINVAL;
1092 }
1093 }
1094 rcu_read_unlock();
1095
1096 __skb_queue_tail(list, skb);
1097
1098next:
1099 if (!(rx->flags & XEN_NETRXF_more_data))
1100 break;
1101
1102 if (cons + slots == rp) {
1103 if (net_ratelimit())
1104 dev_warn(dev, "Need more slots\n");
1105 err = -ENOENT;
1106 break;
1107 }
1108
1109 RING_COPY_RESPONSE(&queue->rx, cons + slots, &rx_local);
1110 rx = &rx_local;
1111 skb = xennet_get_rx_skb(queue, cons + slots);
1112 ref = xennet_get_rx_ref(queue, cons + slots);
1113 slots++;
1114 }
1115
1116 if (unlikely(slots > max)) {
1117 if (net_ratelimit())
1118 dev_warn(dev, "Too many slots\n");
1119 err = -E2BIG;
1120 }
1121
1122 if (unlikely(err))
1123 xennet_set_rx_rsp_cons(queue, cons + slots);
1124
1125 return err;
1126}
1127
1128static int xennet_set_skb_gso(struct sk_buff *skb,
1129 struct xen_netif_extra_info *gso)
1130{
1131 if (!gso->u.gso.size) {
1132 if (net_ratelimit())
1133 pr_warn("GSO size must not be zero\n");
1134 return -EINVAL;
1135 }
1136
1137 if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4 &&
1138 gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV6) {
1139 if (net_ratelimit())
1140 pr_warn("Bad GSO type %d\n", gso->u.gso.type);
1141 return -EINVAL;
1142 }
1143
1144 skb_shinfo(skb)->gso_size = gso->u.gso.size;
1145 skb_shinfo(skb)->gso_type =
1146 (gso->u.gso.type == XEN_NETIF_GSO_TYPE_TCPV4) ?
1147 SKB_GSO_TCPV4 :
1148 SKB_GSO_TCPV6;
1149
1150 /* Header must be checked, and gso_segs computed. */
1151 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1152 skb_shinfo(skb)->gso_segs = 0;
1153
1154 return 0;
1155}
1156
1157static int xennet_fill_frags(struct netfront_queue *queue,
1158 struct sk_buff *skb,
1159 struct sk_buff_head *list)
1160{
1161 RING_IDX cons = queue->rx.rsp_cons;
1162 struct sk_buff *nskb;
1163
1164 while ((nskb = __skb_dequeue(list))) {
1165 struct xen_netif_rx_response rx;
1166 skb_frag_t *nfrag = &skb_shinfo(nskb)->frags[0];
1167
1168 RING_COPY_RESPONSE(&queue->rx, ++cons, &rx);
1169
1170 if (skb_shinfo(skb)->nr_frags == MAX_SKB_FRAGS) {
1171 unsigned int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
1172
1173 BUG_ON(pull_to < skb_headlen(skb));
1174 __pskb_pull_tail(skb, pull_to - skb_headlen(skb));
1175 }
1176 if (unlikely(skb_shinfo(skb)->nr_frags >= MAX_SKB_FRAGS)) {
1177 xennet_set_rx_rsp_cons(queue,
1178 ++cons + skb_queue_len(list));
1179 kfree_skb(nskb);
1180 return -ENOENT;
1181 }
1182
1183 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags,
1184 skb_frag_page(nfrag),
1185 rx.offset, rx.status, PAGE_SIZE);
1186
1187 skb_shinfo(nskb)->nr_frags = 0;
1188 kfree_skb(nskb);
1189 }
1190
1191 xennet_set_rx_rsp_cons(queue, cons);
1192
1193 return 0;
1194}
1195
1196static int checksum_setup(struct net_device *dev, struct sk_buff *skb)
1197{
1198 bool recalculate_partial_csum = false;
1199
1200 /*
1201 * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy
1202 * peers can fail to set NETRXF_csum_blank when sending a GSO
1203 * frame. In this case force the SKB to CHECKSUM_PARTIAL and
1204 * recalculate the partial checksum.
1205 */
1206 if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) {
1207 struct netfront_info *np = netdev_priv(dev);
1208 atomic_inc(&np->rx_gso_checksum_fixup);
1209 skb->ip_summed = CHECKSUM_PARTIAL;
1210 recalculate_partial_csum = true;
1211 }
1212
1213 /* A non-CHECKSUM_PARTIAL SKB does not require setup. */
1214 if (skb->ip_summed != CHECKSUM_PARTIAL)
1215 return 0;
1216
1217 return skb_checksum_setup(skb, recalculate_partial_csum);
1218}
1219
1220static int handle_incoming_queue(struct netfront_queue *queue,
1221 struct sk_buff_head *rxq)
1222{
1223 struct netfront_stats *rx_stats = this_cpu_ptr(queue->info->rx_stats);
1224 int packets_dropped = 0;
1225 struct sk_buff *skb;
1226
1227 while ((skb = __skb_dequeue(rxq)) != NULL) {
1228 int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
1229
1230 if (pull_to > skb_headlen(skb))
1231 __pskb_pull_tail(skb, pull_to - skb_headlen(skb));
1232
1233 /* Ethernet work: Delayed to here as it peeks the header. */
1234 skb->protocol = eth_type_trans(skb, queue->info->netdev);
1235 skb_reset_network_header(skb);
1236
1237 if (checksum_setup(queue->info->netdev, skb)) {
1238 kfree_skb(skb);
1239 packets_dropped++;
1240 queue->info->netdev->stats.rx_errors++;
1241 continue;
1242 }
1243
1244 u64_stats_update_begin(&rx_stats->syncp);
1245 rx_stats->packets++;
1246 rx_stats->bytes += skb->len;
1247 u64_stats_update_end(&rx_stats->syncp);
1248
1249 /* Pass it up. */
1250 napi_gro_receive(&queue->napi, skb);
1251 }
1252
1253 return packets_dropped;
1254}
1255
1256static int xennet_poll(struct napi_struct *napi, int budget)
1257{
1258 struct netfront_queue *queue = container_of(napi, struct netfront_queue, napi);
1259 struct net_device *dev = queue->info->netdev;
1260 struct sk_buff *skb;
1261 struct netfront_rx_info rinfo;
1262 struct xen_netif_rx_response *rx = &rinfo.rx;
1263 struct xen_netif_extra_info *extras = rinfo.extras;
1264 RING_IDX i, rp;
1265 int work_done;
1266 struct sk_buff_head rxq;
1267 struct sk_buff_head errq;
1268 struct sk_buff_head tmpq;
1269 int err;
1270 bool need_xdp_flush = false;
1271
1272 spin_lock(&queue->rx_lock);
1273
1274 skb_queue_head_init(&rxq);
1275 skb_queue_head_init(&errq);
1276 skb_queue_head_init(&tmpq);
1277
1278 rp = queue->rx.sring->rsp_prod;
1279 if (RING_RESPONSE_PROD_OVERFLOW(&queue->rx, rp)) {
1280 dev_alert(&dev->dev, "Illegal number of responses %u\n",
1281 rp - queue->rx.rsp_cons);
1282 queue->info->broken = true;
1283 spin_unlock(&queue->rx_lock);
1284 return 0;
1285 }
1286 rmb(); /* Ensure we see queued responses up to 'rp'. */
1287
1288 i = queue->rx.rsp_cons;
1289 work_done = 0;
1290 while ((i != rp) && (work_done < budget)) {
1291 RING_COPY_RESPONSE(&queue->rx, i, rx);
1292 memset(extras, 0, sizeof(rinfo.extras));
1293
1294 err = xennet_get_responses(queue, &rinfo, rp, &tmpq,
1295 &need_xdp_flush);
1296
1297 if (unlikely(err)) {
1298 if (queue->info->broken) {
1299 spin_unlock(&queue->rx_lock);
1300 return 0;
1301 }
1302err:
1303 while ((skb = __skb_dequeue(&tmpq)))
1304 __skb_queue_tail(&errq, skb);
1305 dev->stats.rx_errors++;
1306 i = queue->rx.rsp_cons;
1307 continue;
1308 }
1309
1310 skb = __skb_dequeue(&tmpq);
1311
1312 if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) {
1313 struct xen_netif_extra_info *gso;
1314 gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1];
1315
1316 if (unlikely(xennet_set_skb_gso(skb, gso))) {
1317 __skb_queue_head(&tmpq, skb);
1318 xennet_set_rx_rsp_cons(queue,
1319 queue->rx.rsp_cons +
1320 skb_queue_len(&tmpq));
1321 goto err;
1322 }
1323 }
1324
1325 NETFRONT_SKB_CB(skb)->pull_to = rx->status;
1326 if (NETFRONT_SKB_CB(skb)->pull_to > RX_COPY_THRESHOLD)
1327 NETFRONT_SKB_CB(skb)->pull_to = RX_COPY_THRESHOLD;
1328
1329 skb_frag_off_set(&skb_shinfo(skb)->frags[0], rx->offset);
1330 skb_frag_size_set(&skb_shinfo(skb)->frags[0], rx->status);
1331 skb->data_len = rx->status;
1332 skb->len += rx->status;
1333
1334 if (unlikely(xennet_fill_frags(queue, skb, &tmpq)))
1335 goto err;
1336
1337 if (rx->flags & XEN_NETRXF_csum_blank)
1338 skb->ip_summed = CHECKSUM_PARTIAL;
1339 else if (rx->flags & XEN_NETRXF_data_validated)
1340 skb->ip_summed = CHECKSUM_UNNECESSARY;
1341
1342 __skb_queue_tail(&rxq, skb);
1343
1344 i = queue->rx.rsp_cons + 1;
1345 xennet_set_rx_rsp_cons(queue, i);
1346 work_done++;
1347 }
1348 if (need_xdp_flush)
1349 xdp_do_flush();
1350
1351 __skb_queue_purge(&errq);
1352
1353 work_done -= handle_incoming_queue(queue, &rxq);
1354
1355 xennet_alloc_rx_buffers(queue);
1356
1357 if (work_done < budget) {
1358 int more_to_do = 0;
1359
1360 napi_complete_done(napi, work_done);
1361
1362 RING_FINAL_CHECK_FOR_RESPONSES(&queue->rx, more_to_do);
1363 if (more_to_do)
1364 napi_schedule(napi);
1365 }
1366
1367 spin_unlock(&queue->rx_lock);
1368
1369 return work_done;
1370}
1371
1372static int xennet_change_mtu(struct net_device *dev, int mtu)
1373{
1374 int max = xennet_can_sg(dev) ? XEN_NETIF_MAX_TX_SIZE : ETH_DATA_LEN;
1375
1376 if (mtu > max)
1377 return -EINVAL;
1378 dev->mtu = mtu;
1379 return 0;
1380}
1381
1382static void xennet_get_stats64(struct net_device *dev,
1383 struct rtnl_link_stats64 *tot)
1384{
1385 struct netfront_info *np = netdev_priv(dev);
1386 int cpu;
1387
1388 for_each_possible_cpu(cpu) {
1389 struct netfront_stats *rx_stats = per_cpu_ptr(np->rx_stats, cpu);
1390 struct netfront_stats *tx_stats = per_cpu_ptr(np->tx_stats, cpu);
1391 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1392 unsigned int start;
1393
1394 do {
1395 start = u64_stats_fetch_begin(&tx_stats->syncp);
1396 tx_packets = tx_stats->packets;
1397 tx_bytes = tx_stats->bytes;
1398 } while (u64_stats_fetch_retry(&tx_stats->syncp, start));
1399
1400 do {
1401 start = u64_stats_fetch_begin(&rx_stats->syncp);
1402 rx_packets = rx_stats->packets;
1403 rx_bytes = rx_stats->bytes;
1404 } while (u64_stats_fetch_retry(&rx_stats->syncp, start));
1405
1406 tot->rx_packets += rx_packets;
1407 tot->tx_packets += tx_packets;
1408 tot->rx_bytes += rx_bytes;
1409 tot->tx_bytes += tx_bytes;
1410 }
1411
1412 tot->rx_errors = dev->stats.rx_errors;
1413 tot->tx_dropped = dev->stats.tx_dropped;
1414}
1415
1416static void xennet_release_tx_bufs(struct netfront_queue *queue)
1417{
1418 struct sk_buff *skb;
1419 int i;
1420
1421 for (i = 0; i < NET_TX_RING_SIZE; i++) {
1422 /* Skip over entries which are actually freelist references */
1423 if (!queue->tx_skbs[i])
1424 continue;
1425
1426 skb = queue->tx_skbs[i];
1427 queue->tx_skbs[i] = NULL;
1428 get_page(queue->grant_tx_page[i]);
1429 gnttab_end_foreign_access(queue->grant_tx_ref[i],
1430 queue->grant_tx_page[i]);
1431 queue->grant_tx_page[i] = NULL;
1432 queue->grant_tx_ref[i] = INVALID_GRANT_REF;
1433 add_id_to_list(&queue->tx_skb_freelist, queue->tx_link, i);
1434 dev_kfree_skb_irq(skb);
1435 }
1436}
1437
1438static void xennet_release_rx_bufs(struct netfront_queue *queue)
1439{
1440 int id, ref;
1441
1442 spin_lock_bh(&queue->rx_lock);
1443
1444 for (id = 0; id < NET_RX_RING_SIZE; id++) {
1445 struct sk_buff *skb;
1446 struct page *page;
1447
1448 skb = queue->rx_skbs[id];
1449 if (!skb)
1450 continue;
1451
1452 ref = queue->grant_rx_ref[id];
1453 if (ref == INVALID_GRANT_REF)
1454 continue;
1455
1456 page = skb_frag_page(&skb_shinfo(skb)->frags[0]);
1457
1458 /* gnttab_end_foreign_access() needs a page ref until
1459 * foreign access is ended (which may be deferred).
1460 */
1461 get_page(page);
1462 gnttab_end_foreign_access(ref, page);
1463 queue->grant_rx_ref[id] = INVALID_GRANT_REF;
1464
1465 kfree_skb(skb);
1466 }
1467
1468 spin_unlock_bh(&queue->rx_lock);
1469}
1470
1471static netdev_features_t xennet_fix_features(struct net_device *dev,
1472 netdev_features_t features)
1473{
1474 struct netfront_info *np = netdev_priv(dev);
1475
1476 if (features & NETIF_F_SG &&
1477 !xenbus_read_unsigned(np->xbdev->otherend, "feature-sg", 0))
1478 features &= ~NETIF_F_SG;
1479
1480 if (features & NETIF_F_IPV6_CSUM &&
1481 !xenbus_read_unsigned(np->xbdev->otherend,
1482 "feature-ipv6-csum-offload", 0))
1483 features &= ~NETIF_F_IPV6_CSUM;
1484
1485 if (features & NETIF_F_TSO &&
1486 !xenbus_read_unsigned(np->xbdev->otherend, "feature-gso-tcpv4", 0))
1487 features &= ~NETIF_F_TSO;
1488
1489 if (features & NETIF_F_TSO6 &&
1490 !xenbus_read_unsigned(np->xbdev->otherend, "feature-gso-tcpv6", 0))
1491 features &= ~NETIF_F_TSO6;
1492
1493 return features;
1494}
1495
1496static int xennet_set_features(struct net_device *dev,
1497 netdev_features_t features)
1498{
1499 if (!(features & NETIF_F_SG) && dev->mtu > ETH_DATA_LEN) {
1500 netdev_info(dev, "Reducing MTU because no SG offload");
1501 dev->mtu = ETH_DATA_LEN;
1502 }
1503
1504 return 0;
1505}
1506
1507static bool xennet_handle_tx(struct netfront_queue *queue, unsigned int *eoi)
1508{
1509 unsigned long flags;
1510
1511 if (unlikely(queue->info->broken))
1512 return false;
1513
1514 spin_lock_irqsave(&queue->tx_lock, flags);
1515 if (xennet_tx_buf_gc(queue))
1516 *eoi = 0;
1517 spin_unlock_irqrestore(&queue->tx_lock, flags);
1518
1519 return true;
1520}
1521
1522static irqreturn_t xennet_tx_interrupt(int irq, void *dev_id)
1523{
1524 unsigned int eoiflag = XEN_EOI_FLAG_SPURIOUS;
1525
1526 if (likely(xennet_handle_tx(dev_id, &eoiflag)))
1527 xen_irq_lateeoi(irq, eoiflag);
1528
1529 return IRQ_HANDLED;
1530}
1531
1532static bool xennet_handle_rx(struct netfront_queue *queue, unsigned int *eoi)
1533{
1534 unsigned int work_queued;
1535 unsigned long flags;
1536
1537 if (unlikely(queue->info->broken))
1538 return false;
1539
1540 spin_lock_irqsave(&queue->rx_cons_lock, flags);
1541 work_queued = XEN_RING_NR_UNCONSUMED_RESPONSES(&queue->rx);
1542 if (work_queued > queue->rx_rsp_unconsumed) {
1543 queue->rx_rsp_unconsumed = work_queued;
1544 *eoi = 0;
1545 } else if (unlikely(work_queued < queue->rx_rsp_unconsumed)) {
1546 const struct device *dev = &queue->info->netdev->dev;
1547
1548 spin_unlock_irqrestore(&queue->rx_cons_lock, flags);
1549 dev_alert(dev, "RX producer index going backwards\n");
1550 dev_alert(dev, "Disabled for further use\n");
1551 queue->info->broken = true;
1552 return false;
1553 }
1554 spin_unlock_irqrestore(&queue->rx_cons_lock, flags);
1555
1556 if (likely(netif_carrier_ok(queue->info->netdev) && work_queued))
1557 napi_schedule(&queue->napi);
1558
1559 return true;
1560}
1561
1562static irqreturn_t xennet_rx_interrupt(int irq, void *dev_id)
1563{
1564 unsigned int eoiflag = XEN_EOI_FLAG_SPURIOUS;
1565
1566 if (likely(xennet_handle_rx(dev_id, &eoiflag)))
1567 xen_irq_lateeoi(irq, eoiflag);
1568
1569 return IRQ_HANDLED;
1570}
1571
1572static irqreturn_t xennet_interrupt(int irq, void *dev_id)
1573{
1574 unsigned int eoiflag = XEN_EOI_FLAG_SPURIOUS;
1575
1576 if (xennet_handle_tx(dev_id, &eoiflag) &&
1577 xennet_handle_rx(dev_id, &eoiflag))
1578 xen_irq_lateeoi(irq, eoiflag);
1579
1580 return IRQ_HANDLED;
1581}
1582
1583#ifdef CONFIG_NET_POLL_CONTROLLER
1584static void xennet_poll_controller(struct net_device *dev)
1585{
1586 /* Poll each queue */
1587 struct netfront_info *info = netdev_priv(dev);
1588 unsigned int num_queues = dev->real_num_tx_queues;
1589 unsigned int i;
1590
1591 if (info->broken)
1592 return;
1593
1594 for (i = 0; i < num_queues; ++i)
1595 xennet_interrupt(0, &info->queues[i]);
1596}
1597#endif
1598
1599#define NETBACK_XDP_HEADROOM_DISABLE 0
1600#define NETBACK_XDP_HEADROOM_ENABLE 1
1601
1602static int talk_to_netback_xdp(struct netfront_info *np, int xdp)
1603{
1604 int err;
1605 unsigned short headroom;
1606
1607 headroom = xdp ? XDP_PACKET_HEADROOM : 0;
1608 err = xenbus_printf(XBT_NIL, np->xbdev->nodename,
1609 "xdp-headroom", "%hu",
1610 headroom);
1611 if (err)
1612 pr_warn("Error writing xdp-headroom\n");
1613
1614 return err;
1615}
1616
1617static int xennet_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1618 struct netlink_ext_ack *extack)
1619{
1620 unsigned long max_mtu = XEN_PAGE_SIZE - XDP_PACKET_HEADROOM;
1621 struct netfront_info *np = netdev_priv(dev);
1622 struct bpf_prog *old_prog;
1623 unsigned int i, err;
1624
1625 if (dev->mtu > max_mtu) {
1626 netdev_warn(dev, "XDP requires MTU less than %lu\n", max_mtu);
1627 return -EINVAL;
1628 }
1629
1630 if (!np->netback_has_xdp_headroom)
1631 return 0;
1632
1633 xenbus_switch_state(np->xbdev, XenbusStateReconfiguring);
1634
1635 err = talk_to_netback_xdp(np, prog ? NETBACK_XDP_HEADROOM_ENABLE :
1636 NETBACK_XDP_HEADROOM_DISABLE);
1637 if (err)
1638 return err;
1639
1640 /* avoid the race with XDP headroom adjustment */
1641 wait_event(module_wq,
1642 xenbus_read_driver_state(np->xbdev->otherend) ==
1643 XenbusStateReconfigured);
1644 np->netfront_xdp_enabled = true;
1645
1646 old_prog = rtnl_dereference(np->queues[0].xdp_prog);
1647
1648 if (prog)
1649 bpf_prog_add(prog, dev->real_num_tx_queues);
1650
1651 for (i = 0; i < dev->real_num_tx_queues; ++i)
1652 rcu_assign_pointer(np->queues[i].xdp_prog, prog);
1653
1654 if (old_prog)
1655 for (i = 0; i < dev->real_num_tx_queues; ++i)
1656 bpf_prog_put(old_prog);
1657
1658 xenbus_switch_state(np->xbdev, XenbusStateConnected);
1659
1660 return 0;
1661}
1662
1663static int xennet_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1664{
1665 struct netfront_info *np = netdev_priv(dev);
1666
1667 if (np->broken)
1668 return -ENODEV;
1669
1670 switch (xdp->command) {
1671 case XDP_SETUP_PROG:
1672 return xennet_xdp_set(dev, xdp->prog, xdp->extack);
1673 default:
1674 return -EINVAL;
1675 }
1676}
1677
1678static const struct net_device_ops xennet_netdev_ops = {
1679 .ndo_uninit = xennet_uninit,
1680 .ndo_open = xennet_open,
1681 .ndo_stop = xennet_close,
1682 .ndo_start_xmit = xennet_start_xmit,
1683 .ndo_change_mtu = xennet_change_mtu,
1684 .ndo_get_stats64 = xennet_get_stats64,
1685 .ndo_set_mac_address = eth_mac_addr,
1686 .ndo_validate_addr = eth_validate_addr,
1687 .ndo_fix_features = xennet_fix_features,
1688 .ndo_set_features = xennet_set_features,
1689 .ndo_select_queue = xennet_select_queue,
1690 .ndo_bpf = xennet_xdp,
1691 .ndo_xdp_xmit = xennet_xdp_xmit,
1692#ifdef CONFIG_NET_POLL_CONTROLLER
1693 .ndo_poll_controller = xennet_poll_controller,
1694#endif
1695};
1696
1697static void xennet_free_netdev(struct net_device *netdev)
1698{
1699 struct netfront_info *np = netdev_priv(netdev);
1700
1701 free_percpu(np->rx_stats);
1702 free_percpu(np->tx_stats);
1703 free_netdev(netdev);
1704}
1705
1706static struct net_device *xennet_create_dev(struct xenbus_device *dev)
1707{
1708 int err;
1709 struct net_device *netdev;
1710 struct netfront_info *np;
1711
1712 netdev = alloc_etherdev_mq(sizeof(struct netfront_info), xennet_max_queues);
1713 if (!netdev)
1714 return ERR_PTR(-ENOMEM);
1715
1716 np = netdev_priv(netdev);
1717 np->xbdev = dev;
1718
1719 np->queues = NULL;
1720
1721 err = -ENOMEM;
1722 np->rx_stats = netdev_alloc_pcpu_stats(struct netfront_stats);
1723 if (np->rx_stats == NULL)
1724 goto exit;
1725 np->tx_stats = netdev_alloc_pcpu_stats(struct netfront_stats);
1726 if (np->tx_stats == NULL)
1727 goto exit;
1728
1729 netdev->netdev_ops = &xennet_netdev_ops;
1730
1731 netdev->features = NETIF_F_IP_CSUM | NETIF_F_RXCSUM |
1732 NETIF_F_GSO_ROBUST;
1733 netdev->hw_features = NETIF_F_SG |
1734 NETIF_F_IPV6_CSUM |
1735 NETIF_F_TSO | NETIF_F_TSO6;
1736
1737 /*
1738 * Assume that all hw features are available for now. This set
1739 * will be adjusted by the call to netdev_update_features() in
1740 * xennet_connect() which is the earliest point where we can
1741 * negotiate with the backend regarding supported features.
1742 */
1743 netdev->features |= netdev->hw_features;
1744
1745 netdev->ethtool_ops = &xennet_ethtool_ops;
1746 netdev->min_mtu = ETH_MIN_MTU;
1747 netdev->max_mtu = XEN_NETIF_MAX_TX_SIZE;
1748 SET_NETDEV_DEV(netdev, &dev->dev);
1749
1750 np->netdev = netdev;
1751 np->netfront_xdp_enabled = false;
1752
1753 netif_carrier_off(netdev);
1754
1755 do {
1756 xenbus_switch_state(dev, XenbusStateInitialising);
1757 err = wait_event_timeout(module_wq,
1758 xenbus_read_driver_state(dev->otherend) !=
1759 XenbusStateClosed &&
1760 xenbus_read_driver_state(dev->otherend) !=
1761 XenbusStateUnknown, XENNET_TIMEOUT);
1762 } while (!err);
1763
1764 return netdev;
1765
1766 exit:
1767 xennet_free_netdev(netdev);
1768 return ERR_PTR(err);
1769}
1770
1771/*
1772 * Entry point to this code when a new device is created. Allocate the basic
1773 * structures and the ring buffers for communication with the backend, and
1774 * inform the backend of the appropriate details for those.
1775 */
1776static int netfront_probe(struct xenbus_device *dev,
1777 const struct xenbus_device_id *id)
1778{
1779 int err;
1780 struct net_device *netdev;
1781 struct netfront_info *info;
1782
1783 netdev = xennet_create_dev(dev);
1784 if (IS_ERR(netdev)) {
1785 err = PTR_ERR(netdev);
1786 xenbus_dev_fatal(dev, err, "creating netdev");
1787 return err;
1788 }
1789
1790 info = netdev_priv(netdev);
1791 dev_set_drvdata(&dev->dev, info);
1792#ifdef CONFIG_SYSFS
1793 info->netdev->sysfs_groups[0] = &xennet_dev_group;
1794#endif
1795
1796 return 0;
1797}
1798
1799static void xennet_end_access(int ref, void *page)
1800{
1801 /* This frees the page as a side-effect */
1802 if (ref != INVALID_GRANT_REF)
1803 gnttab_end_foreign_access(ref, virt_to_page(page));
1804}
1805
1806static void xennet_disconnect_backend(struct netfront_info *info)
1807{
1808 unsigned int i = 0;
1809 unsigned int num_queues = info->netdev->real_num_tx_queues;
1810
1811 netif_carrier_off(info->netdev);
1812
1813 for (i = 0; i < num_queues && info->queues; ++i) {
1814 struct netfront_queue *queue = &info->queues[i];
1815
1816 del_timer_sync(&queue->rx_refill_timer);
1817
1818 if (queue->tx_irq && (queue->tx_irq == queue->rx_irq))
1819 unbind_from_irqhandler(queue->tx_irq, queue);
1820 if (queue->tx_irq && (queue->tx_irq != queue->rx_irq)) {
1821 unbind_from_irqhandler(queue->tx_irq, queue);
1822 unbind_from_irqhandler(queue->rx_irq, queue);
1823 }
1824 queue->tx_evtchn = queue->rx_evtchn = 0;
1825 queue->tx_irq = queue->rx_irq = 0;
1826
1827 if (netif_running(info->netdev))
1828 napi_synchronize(&queue->napi);
1829
1830 xennet_release_tx_bufs(queue);
1831 xennet_release_rx_bufs(queue);
1832 gnttab_free_grant_references(queue->gref_tx_head);
1833 gnttab_free_grant_references(queue->gref_rx_head);
1834
1835 /* End access and free the pages */
1836 xennet_end_access(queue->tx_ring_ref, queue->tx.sring);
1837 xennet_end_access(queue->rx_ring_ref, queue->rx.sring);
1838
1839 queue->tx_ring_ref = INVALID_GRANT_REF;
1840 queue->rx_ring_ref = INVALID_GRANT_REF;
1841 queue->tx.sring = NULL;
1842 queue->rx.sring = NULL;
1843
1844 page_pool_destroy(queue->page_pool);
1845 }
1846}
1847
1848/*
1849 * We are reconnecting to the backend, due to a suspend/resume, or a backend
1850 * driver restart. We tear down our netif structure and recreate it, but
1851 * leave the device-layer structures intact so that this is transparent to the
1852 * rest of the kernel.
1853 */
1854static int netfront_resume(struct xenbus_device *dev)
1855{
1856 struct netfront_info *info = dev_get_drvdata(&dev->dev);
1857
1858 dev_dbg(&dev->dev, "%s\n", dev->nodename);
1859
1860 netif_tx_lock_bh(info->netdev);
1861 netif_device_detach(info->netdev);
1862 netif_tx_unlock_bh(info->netdev);
1863
1864 xennet_disconnect_backend(info);
1865
1866 rtnl_lock();
1867 if (info->queues)
1868 xennet_destroy_queues(info);
1869 rtnl_unlock();
1870
1871 return 0;
1872}
1873
1874static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[])
1875{
1876 char *s, *e, *macstr;
1877 int i;
1878
1879 macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL);
1880 if (IS_ERR(macstr))
1881 return PTR_ERR(macstr);
1882
1883 for (i = 0; i < ETH_ALEN; i++) {
1884 mac[i] = simple_strtoul(s, &e, 16);
1885 if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) {
1886 kfree(macstr);
1887 return -ENOENT;
1888 }
1889 s = e+1;
1890 }
1891
1892 kfree(macstr);
1893 return 0;
1894}
1895
1896static int setup_netfront_single(struct netfront_queue *queue)
1897{
1898 int err;
1899
1900 err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn);
1901 if (err < 0)
1902 goto fail;
1903
1904 err = bind_evtchn_to_irqhandler_lateeoi(queue->tx_evtchn,
1905 xennet_interrupt, 0,
1906 queue->info->netdev->name,
1907 queue);
1908 if (err < 0)
1909 goto bind_fail;
1910 queue->rx_evtchn = queue->tx_evtchn;
1911 queue->rx_irq = queue->tx_irq = err;
1912
1913 return 0;
1914
1915bind_fail:
1916 xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn);
1917 queue->tx_evtchn = 0;
1918fail:
1919 return err;
1920}
1921
1922static int setup_netfront_split(struct netfront_queue *queue)
1923{
1924 int err;
1925
1926 err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn);
1927 if (err < 0)
1928 goto fail;
1929 err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->rx_evtchn);
1930 if (err < 0)
1931 goto alloc_rx_evtchn_fail;
1932
1933 snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name),
1934 "%s-tx", queue->name);
1935 err = bind_evtchn_to_irqhandler_lateeoi(queue->tx_evtchn,
1936 xennet_tx_interrupt, 0,
1937 queue->tx_irq_name, queue);
1938 if (err < 0)
1939 goto bind_tx_fail;
1940 queue->tx_irq = err;
1941
1942 snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name),
1943 "%s-rx", queue->name);
1944 err = bind_evtchn_to_irqhandler_lateeoi(queue->rx_evtchn,
1945 xennet_rx_interrupt, 0,
1946 queue->rx_irq_name, queue);
1947 if (err < 0)
1948 goto bind_rx_fail;
1949 queue->rx_irq = err;
1950
1951 return 0;
1952
1953bind_rx_fail:
1954 unbind_from_irqhandler(queue->tx_irq, queue);
1955 queue->tx_irq = 0;
1956bind_tx_fail:
1957 xenbus_free_evtchn(queue->info->xbdev, queue->rx_evtchn);
1958 queue->rx_evtchn = 0;
1959alloc_rx_evtchn_fail:
1960 xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn);
1961 queue->tx_evtchn = 0;
1962fail:
1963 return err;
1964}
1965
1966static int setup_netfront(struct xenbus_device *dev,
1967 struct netfront_queue *queue, unsigned int feature_split_evtchn)
1968{
1969 struct xen_netif_tx_sring *txs;
1970 struct xen_netif_rx_sring *rxs;
1971 int err;
1972
1973 queue->tx_ring_ref = INVALID_GRANT_REF;
1974 queue->rx_ring_ref = INVALID_GRANT_REF;
1975 queue->rx.sring = NULL;
1976 queue->tx.sring = NULL;
1977
1978 err = xenbus_setup_ring(dev, GFP_NOIO | __GFP_HIGH, (void **)&txs,
1979 1, &queue->tx_ring_ref);
1980 if (err)
1981 goto fail;
1982
1983 XEN_FRONT_RING_INIT(&queue->tx, txs, XEN_PAGE_SIZE);
1984
1985 err = xenbus_setup_ring(dev, GFP_NOIO | __GFP_HIGH, (void **)&rxs,
1986 1, &queue->rx_ring_ref);
1987 if (err)
1988 goto fail;
1989
1990 XEN_FRONT_RING_INIT(&queue->rx, rxs, XEN_PAGE_SIZE);
1991
1992 if (feature_split_evtchn)
1993 err = setup_netfront_split(queue);
1994 /* setup single event channel if
1995 * a) feature-split-event-channels == 0
1996 * b) feature-split-event-channels == 1 but failed to setup
1997 */
1998 if (!feature_split_evtchn || err)
1999 err = setup_netfront_single(queue);
2000
2001 if (err)
2002 goto fail;
2003
2004 return 0;
2005
2006 fail:
2007 xenbus_teardown_ring((void **)&queue->rx.sring, 1, &queue->rx_ring_ref);
2008 xenbus_teardown_ring((void **)&queue->tx.sring, 1, &queue->tx_ring_ref);
2009
2010 return err;
2011}
2012
2013/* Queue-specific initialisation
2014 * This used to be done in xennet_create_dev() but must now
2015 * be run per-queue.
2016 */
2017static int xennet_init_queue(struct netfront_queue *queue)
2018{
2019 unsigned short i;
2020 int err = 0;
2021 char *devid;
2022
2023 spin_lock_init(&queue->tx_lock);
2024 spin_lock_init(&queue->rx_lock);
2025 spin_lock_init(&queue->rx_cons_lock);
2026
2027 timer_setup(&queue->rx_refill_timer, rx_refill_timeout, 0);
2028
2029 devid = strrchr(queue->info->xbdev->nodename, '/') + 1;
2030 snprintf(queue->name, sizeof(queue->name), "vif%s-q%u",
2031 devid, queue->id);
2032
2033 /* Initialise tx_skb_freelist as a free chain containing every entry. */
2034 queue->tx_skb_freelist = 0;
2035 queue->tx_pend_queue = TX_LINK_NONE;
2036 for (i = 0; i < NET_TX_RING_SIZE; i++) {
2037 queue->tx_link[i] = i + 1;
2038 queue->grant_tx_ref[i] = INVALID_GRANT_REF;
2039 queue->grant_tx_page[i] = NULL;
2040 }
2041 queue->tx_link[NET_TX_RING_SIZE - 1] = TX_LINK_NONE;
2042
2043 /* Clear out rx_skbs */
2044 for (i = 0; i < NET_RX_RING_SIZE; i++) {
2045 queue->rx_skbs[i] = NULL;
2046 queue->grant_rx_ref[i] = INVALID_GRANT_REF;
2047 }
2048
2049 /* A grant for every tx ring slot */
2050 if (gnttab_alloc_grant_references(NET_TX_RING_SIZE,
2051 &queue->gref_tx_head) < 0) {
2052 pr_alert("can't alloc tx grant refs\n");
2053 err = -ENOMEM;
2054 goto exit;
2055 }
2056
2057 /* A grant for every rx ring slot */
2058 if (gnttab_alloc_grant_references(NET_RX_RING_SIZE,
2059 &queue->gref_rx_head) < 0) {
2060 pr_alert("can't alloc rx grant refs\n");
2061 err = -ENOMEM;
2062 goto exit_free_tx;
2063 }
2064
2065 return 0;
2066
2067 exit_free_tx:
2068 gnttab_free_grant_references(queue->gref_tx_head);
2069 exit:
2070 return err;
2071}
2072
2073static int write_queue_xenstore_keys(struct netfront_queue *queue,
2074 struct xenbus_transaction *xbt, int write_hierarchical)
2075{
2076 /* Write the queue-specific keys into XenStore in the traditional
2077 * way for a single queue, or in a queue subkeys for multiple
2078 * queues.
2079 */
2080 struct xenbus_device *dev = queue->info->xbdev;
2081 int err;
2082 const char *message;
2083 char *path;
2084 size_t pathsize;
2085
2086 /* Choose the correct place to write the keys */
2087 if (write_hierarchical) {
2088 pathsize = strlen(dev->nodename) + 10;
2089 path = kzalloc(pathsize, GFP_KERNEL);
2090 if (!path) {
2091 err = -ENOMEM;
2092 message = "out of memory while writing ring references";
2093 goto error;
2094 }
2095 snprintf(path, pathsize, "%s/queue-%u",
2096 dev->nodename, queue->id);
2097 } else {
2098 path = (char *)dev->nodename;
2099 }
2100
2101 /* Write ring references */
2102 err = xenbus_printf(*xbt, path, "tx-ring-ref", "%u",
2103 queue->tx_ring_ref);
2104 if (err) {
2105 message = "writing tx-ring-ref";
2106 goto error;
2107 }
2108
2109 err = xenbus_printf(*xbt, path, "rx-ring-ref", "%u",
2110 queue->rx_ring_ref);
2111 if (err) {
2112 message = "writing rx-ring-ref";
2113 goto error;
2114 }
2115
2116 /* Write event channels; taking into account both shared
2117 * and split event channel scenarios.
2118 */
2119 if (queue->tx_evtchn == queue->rx_evtchn) {
2120 /* Shared event channel */
2121 err = xenbus_printf(*xbt, path,
2122 "event-channel", "%u", queue->tx_evtchn);
2123 if (err) {
2124 message = "writing event-channel";
2125 goto error;
2126 }
2127 } else {
2128 /* Split event channels */
2129 err = xenbus_printf(*xbt, path,
2130 "event-channel-tx", "%u", queue->tx_evtchn);
2131 if (err) {
2132 message = "writing event-channel-tx";
2133 goto error;
2134 }
2135
2136 err = xenbus_printf(*xbt, path,
2137 "event-channel-rx", "%u", queue->rx_evtchn);
2138 if (err) {
2139 message = "writing event-channel-rx";
2140 goto error;
2141 }
2142 }
2143
2144 if (write_hierarchical)
2145 kfree(path);
2146 return 0;
2147
2148error:
2149 if (write_hierarchical)
2150 kfree(path);
2151 xenbus_dev_fatal(dev, err, "%s", message);
2152 return err;
2153}
2154
2155
2156
2157static int xennet_create_page_pool(struct netfront_queue *queue)
2158{
2159 int err;
2160 struct page_pool_params pp_params = {
2161 .order = 0,
2162 .flags = 0,
2163 .pool_size = NET_RX_RING_SIZE,
2164 .nid = NUMA_NO_NODE,
2165 .dev = &queue->info->netdev->dev,
2166 .offset = XDP_PACKET_HEADROOM,
2167 .max_len = XEN_PAGE_SIZE - XDP_PACKET_HEADROOM,
2168 };
2169
2170 queue->page_pool = page_pool_create(&pp_params);
2171 if (IS_ERR(queue->page_pool)) {
2172 err = PTR_ERR(queue->page_pool);
2173 queue->page_pool = NULL;
2174 return err;
2175 }
2176
2177 err = xdp_rxq_info_reg(&queue->xdp_rxq, queue->info->netdev,
2178 queue->id, 0);
2179 if (err) {
2180 netdev_err(queue->info->netdev, "xdp_rxq_info_reg failed\n");
2181 goto err_free_pp;
2182 }
2183
2184 err = xdp_rxq_info_reg_mem_model(&queue->xdp_rxq,
2185 MEM_TYPE_PAGE_POOL, queue->page_pool);
2186 if (err) {
2187 netdev_err(queue->info->netdev, "xdp_rxq_info_reg_mem_model failed\n");
2188 goto err_unregister_rxq;
2189 }
2190 return 0;
2191
2192err_unregister_rxq:
2193 xdp_rxq_info_unreg(&queue->xdp_rxq);
2194err_free_pp:
2195 page_pool_destroy(queue->page_pool);
2196 queue->page_pool = NULL;
2197 return err;
2198}
2199
2200static int xennet_create_queues(struct netfront_info *info,
2201 unsigned int *num_queues)
2202{
2203 unsigned int i;
2204 int ret;
2205
2206 info->queues = kcalloc(*num_queues, sizeof(struct netfront_queue),
2207 GFP_KERNEL);
2208 if (!info->queues)
2209 return -ENOMEM;
2210
2211 for (i = 0; i < *num_queues; i++) {
2212 struct netfront_queue *queue = &info->queues[i];
2213
2214 queue->id = i;
2215 queue->info = info;
2216
2217 ret = xennet_init_queue(queue);
2218 if (ret < 0) {
2219 dev_warn(&info->xbdev->dev,
2220 "only created %d queues\n", i);
2221 *num_queues = i;
2222 break;
2223 }
2224
2225 /* use page pool recycling instead of buddy allocator */
2226 ret = xennet_create_page_pool(queue);
2227 if (ret < 0) {
2228 dev_err(&info->xbdev->dev, "can't allocate page pool\n");
2229 *num_queues = i;
2230 return ret;
2231 }
2232
2233 netif_napi_add(queue->info->netdev, &queue->napi, xennet_poll);
2234 if (netif_running(info->netdev))
2235 napi_enable(&queue->napi);
2236 }
2237
2238 netif_set_real_num_tx_queues(info->netdev, *num_queues);
2239
2240 if (*num_queues == 0) {
2241 dev_err(&info->xbdev->dev, "no queues\n");
2242 return -EINVAL;
2243 }
2244 return 0;
2245}
2246
2247/* Common code used when first setting up, and when resuming. */
2248static int talk_to_netback(struct xenbus_device *dev,
2249 struct netfront_info *info)
2250{
2251 const char *message;
2252 struct xenbus_transaction xbt;
2253 int err;
2254 unsigned int feature_split_evtchn;
2255 unsigned int i = 0;
2256 unsigned int max_queues = 0;
2257 struct netfront_queue *queue = NULL;
2258 unsigned int num_queues = 1;
2259 u8 addr[ETH_ALEN];
2260
2261 info->netdev->irq = 0;
2262
2263 /* Check if backend is trusted. */
2264 info->bounce = !xennet_trusted ||
2265 !xenbus_read_unsigned(dev->nodename, "trusted", 1);
2266
2267 /* Check if backend supports multiple queues */
2268 max_queues = xenbus_read_unsigned(info->xbdev->otherend,
2269 "multi-queue-max-queues", 1);
2270 num_queues = min(max_queues, xennet_max_queues);
2271
2272 /* Check feature-split-event-channels */
2273 feature_split_evtchn = xenbus_read_unsigned(info->xbdev->otherend,
2274 "feature-split-event-channels", 0);
2275
2276 /* Read mac addr. */
2277 err = xen_net_read_mac(dev, addr);
2278 if (err) {
2279 xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename);
2280 goto out_unlocked;
2281 }
2282 eth_hw_addr_set(info->netdev, addr);
2283
2284 info->netback_has_xdp_headroom = xenbus_read_unsigned(info->xbdev->otherend,
2285 "feature-xdp-headroom", 0);
2286 if (info->netback_has_xdp_headroom) {
2287 /* set the current xen-netfront xdp state */
2288 err = talk_to_netback_xdp(info, info->netfront_xdp_enabled ?
2289 NETBACK_XDP_HEADROOM_ENABLE :
2290 NETBACK_XDP_HEADROOM_DISABLE);
2291 if (err)
2292 goto out_unlocked;
2293 }
2294
2295 rtnl_lock();
2296 if (info->queues)
2297 xennet_destroy_queues(info);
2298
2299 /* For the case of a reconnect reset the "broken" indicator. */
2300 info->broken = false;
2301
2302 err = xennet_create_queues(info, &num_queues);
2303 if (err < 0) {
2304 xenbus_dev_fatal(dev, err, "creating queues");
2305 kfree(info->queues);
2306 info->queues = NULL;
2307 goto out;
2308 }
2309 rtnl_unlock();
2310
2311 /* Create shared ring, alloc event channel -- for each queue */
2312 for (i = 0; i < num_queues; ++i) {
2313 queue = &info->queues[i];
2314 err = setup_netfront(dev, queue, feature_split_evtchn);
2315 if (err)
2316 goto destroy_ring;
2317 }
2318
2319again:
2320 err = xenbus_transaction_start(&xbt);
2321 if (err) {
2322 xenbus_dev_fatal(dev, err, "starting transaction");
2323 goto destroy_ring;
2324 }
2325
2326 if (xenbus_exists(XBT_NIL,
2327 info->xbdev->otherend, "multi-queue-max-queues")) {
2328 /* Write the number of queues */
2329 err = xenbus_printf(xbt, dev->nodename,
2330 "multi-queue-num-queues", "%u", num_queues);
2331 if (err) {
2332 message = "writing multi-queue-num-queues";
2333 goto abort_transaction_no_dev_fatal;
2334 }
2335 }
2336
2337 if (num_queues == 1) {
2338 err = write_queue_xenstore_keys(&info->queues[0], &xbt, 0); /* flat */
2339 if (err)
2340 goto abort_transaction_no_dev_fatal;
2341 } else {
2342 /* Write the keys for each queue */
2343 for (i = 0; i < num_queues; ++i) {
2344 queue = &info->queues[i];
2345 err = write_queue_xenstore_keys(queue, &xbt, 1); /* hierarchical */
2346 if (err)
2347 goto abort_transaction_no_dev_fatal;
2348 }
2349 }
2350
2351 /* The remaining keys are not queue-specific */
2352 err = xenbus_printf(xbt, dev->nodename, "request-rx-copy", "%u",
2353 1);
2354 if (err) {
2355 message = "writing request-rx-copy";
2356 goto abort_transaction;
2357 }
2358
2359 err = xenbus_printf(xbt, dev->nodename, "feature-rx-notify", "%d", 1);
2360 if (err) {
2361 message = "writing feature-rx-notify";
2362 goto abort_transaction;
2363 }
2364
2365 err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", 1);
2366 if (err) {
2367 message = "writing feature-sg";
2368 goto abort_transaction;
2369 }
2370
2371 err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", "%d", 1);
2372 if (err) {
2373 message = "writing feature-gso-tcpv4";
2374 goto abort_transaction;
2375 }
2376
2377 err = xenbus_write(xbt, dev->nodename, "feature-gso-tcpv6", "1");
2378 if (err) {
2379 message = "writing feature-gso-tcpv6";
2380 goto abort_transaction;
2381 }
2382
2383 err = xenbus_write(xbt, dev->nodename, "feature-ipv6-csum-offload",
2384 "1");
2385 if (err) {
2386 message = "writing feature-ipv6-csum-offload";
2387 goto abort_transaction;
2388 }
2389
2390 err = xenbus_transaction_end(xbt, 0);
2391 if (err) {
2392 if (err == -EAGAIN)
2393 goto again;
2394 xenbus_dev_fatal(dev, err, "completing transaction");
2395 goto destroy_ring;
2396 }
2397
2398 return 0;
2399
2400 abort_transaction:
2401 xenbus_dev_fatal(dev, err, "%s", message);
2402abort_transaction_no_dev_fatal:
2403 xenbus_transaction_end(xbt, 1);
2404 destroy_ring:
2405 xennet_disconnect_backend(info);
2406 rtnl_lock();
2407 xennet_destroy_queues(info);
2408 out:
2409 rtnl_unlock();
2410out_unlocked:
2411 device_unregister(&dev->dev);
2412 return err;
2413}
2414
2415static int xennet_connect(struct net_device *dev)
2416{
2417 struct netfront_info *np = netdev_priv(dev);
2418 unsigned int num_queues = 0;
2419 int err;
2420 unsigned int j = 0;
2421 struct netfront_queue *queue = NULL;
2422
2423 if (!xenbus_read_unsigned(np->xbdev->otherend, "feature-rx-copy", 0)) {
2424 dev_info(&dev->dev,
2425 "backend does not support copying receive path\n");
2426 return -ENODEV;
2427 }
2428
2429 err = talk_to_netback(np->xbdev, np);
2430 if (err)
2431 return err;
2432 if (np->netback_has_xdp_headroom)
2433 pr_info("backend supports XDP headroom\n");
2434 if (np->bounce)
2435 dev_info(&np->xbdev->dev,
2436 "bouncing transmitted data to zeroed pages\n");
2437
2438 /* talk_to_netback() sets the correct number of queues */
2439 num_queues = dev->real_num_tx_queues;
2440
2441 if (dev->reg_state == NETREG_UNINITIALIZED) {
2442 err = register_netdev(dev);
2443 if (err) {
2444 pr_warn("%s: register_netdev err=%d\n", __func__, err);
2445 device_unregister(&np->xbdev->dev);
2446 return err;
2447 }
2448 }
2449
2450 rtnl_lock();
2451 netdev_update_features(dev);
2452 rtnl_unlock();
2453
2454 /*
2455 * All public and private state should now be sane. Get
2456 * ready to start sending and receiving packets and give the driver
2457 * domain a kick because we've probably just requeued some
2458 * packets.
2459 */
2460 netif_tx_lock_bh(np->netdev);
2461 netif_device_attach(np->netdev);
2462 netif_tx_unlock_bh(np->netdev);
2463
2464 netif_carrier_on(np->netdev);
2465 for (j = 0; j < num_queues; ++j) {
2466 queue = &np->queues[j];
2467
2468 notify_remote_via_irq(queue->tx_irq);
2469 if (queue->tx_irq != queue->rx_irq)
2470 notify_remote_via_irq(queue->rx_irq);
2471
2472 spin_lock_bh(&queue->rx_lock);
2473 xennet_alloc_rx_buffers(queue);
2474 spin_unlock_bh(&queue->rx_lock);
2475 }
2476
2477 return 0;
2478}
2479
2480/*
2481 * Callback received when the backend's state changes.
2482 */
2483static void netback_changed(struct xenbus_device *dev,
2484 enum xenbus_state backend_state)
2485{
2486 struct netfront_info *np = dev_get_drvdata(&dev->dev);
2487 struct net_device *netdev = np->netdev;
2488
2489 dev_dbg(&dev->dev, "%s\n", xenbus_strstate(backend_state));
2490
2491 wake_up_all(&module_wq);
2492
2493 switch (backend_state) {
2494 case XenbusStateInitialising:
2495 case XenbusStateInitialised:
2496 case XenbusStateReconfiguring:
2497 case XenbusStateReconfigured:
2498 case XenbusStateUnknown:
2499 break;
2500
2501 case XenbusStateInitWait:
2502 if (dev->state != XenbusStateInitialising)
2503 break;
2504 if (xennet_connect(netdev) != 0)
2505 break;
2506 xenbus_switch_state(dev, XenbusStateConnected);
2507 break;
2508
2509 case XenbusStateConnected:
2510 netdev_notify_peers(netdev);
2511 break;
2512
2513 case XenbusStateClosed:
2514 if (dev->state == XenbusStateClosed)
2515 break;
2516 fallthrough; /* Missed the backend's CLOSING state */
2517 case XenbusStateClosing:
2518 xenbus_frontend_closed(dev);
2519 break;
2520 }
2521}
2522
2523static const struct xennet_stat {
2524 char name[ETH_GSTRING_LEN];
2525 u16 offset;
2526} xennet_stats[] = {
2527 {
2528 "rx_gso_checksum_fixup",
2529 offsetof(struct netfront_info, rx_gso_checksum_fixup)
2530 },
2531};
2532
2533static int xennet_get_sset_count(struct net_device *dev, int string_set)
2534{
2535 switch (string_set) {
2536 case ETH_SS_STATS:
2537 return ARRAY_SIZE(xennet_stats);
2538 default:
2539 return -EINVAL;
2540 }
2541}
2542
2543static void xennet_get_ethtool_stats(struct net_device *dev,
2544 struct ethtool_stats *stats, u64 * data)
2545{
2546 void *np = netdev_priv(dev);
2547 int i;
2548
2549 for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
2550 data[i] = atomic_read((atomic_t *)(np + xennet_stats[i].offset));
2551}
2552
2553static void xennet_get_strings(struct net_device *dev, u32 stringset, u8 * data)
2554{
2555 int i;
2556
2557 switch (stringset) {
2558 case ETH_SS_STATS:
2559 for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
2560 memcpy(data + i * ETH_GSTRING_LEN,
2561 xennet_stats[i].name, ETH_GSTRING_LEN);
2562 break;
2563 }
2564}
2565
2566static const struct ethtool_ops xennet_ethtool_ops =
2567{
2568 .get_link = ethtool_op_get_link,
2569
2570 .get_sset_count = xennet_get_sset_count,
2571 .get_ethtool_stats = xennet_get_ethtool_stats,
2572 .get_strings = xennet_get_strings,
2573 .get_ts_info = ethtool_op_get_ts_info,
2574};
2575
2576#ifdef CONFIG_SYSFS
2577static ssize_t show_rxbuf(struct device *dev,
2578 struct device_attribute *attr, char *buf)
2579{
2580 return sprintf(buf, "%lu\n", NET_RX_RING_SIZE);
2581}
2582
2583static ssize_t store_rxbuf(struct device *dev,
2584 struct device_attribute *attr,
2585 const char *buf, size_t len)
2586{
2587 char *endp;
2588
2589 if (!capable(CAP_NET_ADMIN))
2590 return -EPERM;
2591
2592 simple_strtoul(buf, &endp, 0);
2593 if (endp == buf)
2594 return -EBADMSG;
2595
2596 /* rxbuf_min and rxbuf_max are no longer configurable. */
2597
2598 return len;
2599}
2600
2601static DEVICE_ATTR(rxbuf_min, 0644, show_rxbuf, store_rxbuf);
2602static DEVICE_ATTR(rxbuf_max, 0644, show_rxbuf, store_rxbuf);
2603static DEVICE_ATTR(rxbuf_cur, 0444, show_rxbuf, NULL);
2604
2605static struct attribute *xennet_dev_attrs[] = {
2606 &dev_attr_rxbuf_min.attr,
2607 &dev_attr_rxbuf_max.attr,
2608 &dev_attr_rxbuf_cur.attr,
2609 NULL
2610};
2611
2612static const struct attribute_group xennet_dev_group = {
2613 .attrs = xennet_dev_attrs
2614};
2615#endif /* CONFIG_SYSFS */
2616
2617static void xennet_bus_close(struct xenbus_device *dev)
2618{
2619 int ret;
2620
2621 if (xenbus_read_driver_state(dev->otherend) == XenbusStateClosed)
2622 return;
2623 do {
2624 xenbus_switch_state(dev, XenbusStateClosing);
2625 ret = wait_event_timeout(module_wq,
2626 xenbus_read_driver_state(dev->otherend) ==
2627 XenbusStateClosing ||
2628 xenbus_read_driver_state(dev->otherend) ==
2629 XenbusStateClosed ||
2630 xenbus_read_driver_state(dev->otherend) ==
2631 XenbusStateUnknown,
2632 XENNET_TIMEOUT);
2633 } while (!ret);
2634
2635 if (xenbus_read_driver_state(dev->otherend) == XenbusStateClosed)
2636 return;
2637
2638 do {
2639 xenbus_switch_state(dev, XenbusStateClosed);
2640 ret = wait_event_timeout(module_wq,
2641 xenbus_read_driver_state(dev->otherend) ==
2642 XenbusStateClosed ||
2643 xenbus_read_driver_state(dev->otherend) ==
2644 XenbusStateUnknown,
2645 XENNET_TIMEOUT);
2646 } while (!ret);
2647}
2648
2649static void xennet_remove(struct xenbus_device *dev)
2650{
2651 struct netfront_info *info = dev_get_drvdata(&dev->dev);
2652
2653 xennet_bus_close(dev);
2654 xennet_disconnect_backend(info);
2655
2656 if (info->netdev->reg_state == NETREG_REGISTERED)
2657 unregister_netdev(info->netdev);
2658
2659 if (info->queues) {
2660 rtnl_lock();
2661 xennet_destroy_queues(info);
2662 rtnl_unlock();
2663 }
2664 xennet_free_netdev(info->netdev);
2665}
2666
2667static const struct xenbus_device_id netfront_ids[] = {
2668 { "vif" },
2669 { "" }
2670};
2671
2672static struct xenbus_driver netfront_driver = {
2673 .ids = netfront_ids,
2674 .probe = netfront_probe,
2675 .remove = xennet_remove,
2676 .resume = netfront_resume,
2677 .otherend_changed = netback_changed,
2678};
2679
2680static int __init netif_init(void)
2681{
2682 if (!xen_domain())
2683 return -ENODEV;
2684
2685 if (!xen_has_pv_nic_devices())
2686 return -ENODEV;
2687
2688 pr_info("Initialising Xen virtual ethernet driver\n");
2689
2690 /* Allow as many queues as there are CPUs inut max. 8 if user has not
2691 * specified a value.
2692 */
2693 if (xennet_max_queues == 0)
2694 xennet_max_queues = min_t(unsigned int, MAX_QUEUES_DEFAULT,
2695 num_online_cpus());
2696
2697 return xenbus_register_frontend(&netfront_driver);
2698}
2699module_init(netif_init);
2700
2701
2702static void __exit netif_exit(void)
2703{
2704 xenbus_unregister_driver(&netfront_driver);
2705}
2706module_exit(netif_exit);
2707
2708MODULE_DESCRIPTION("Xen virtual network device frontend");
2709MODULE_LICENSE("GPL");
2710MODULE_ALIAS("xen:vif");
2711MODULE_ALIAS("xennet");