projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
xen-netfront: fix resource leak in netfront
[deliverable/linux.git] / drivers / net / xen-netfront.c
CommitLineData
0d160211
JF		 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
383eda32
JP		 32#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
33
0d160211
JF		 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>
9ecd1a75 41#include <net/tcp.h>
0d160211
JF		 42#include <linux/udp.h>
43#include <linux/moduleparam.h>
44#include <linux/mm.h>
5a0e3ad6 45#include <linux/slab.h>
0d160211
JF		 46#include <net/ip.h>
47
ca981633 48#include <asm/xen/page.h>
1ccbf534 49#include <xen/xen.h>
0d160211
JF		 50#include <xen/xenbus.h>
51#include <xen/events.h>
52#include <xen/page.h>
b9136d20 53#include <xen/platform_pci.h>
0d160211
JF		 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
0fc0b732 60static const struct ethtool_ops xennet_ethtool_ops;
0d160211
JF		 61
62struct netfront_cb {
3683243b 63 int pull_to;
0d160211
JF		 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
667c78af
JF		 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)
40206dd9 74#define TX_MAX_TARGET min_t(int, NET_TX_RING_SIZE, 256)
0d160211 75
e00f85be 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
0d160211
JF		 84struct netfront_info {
85 struct list_head list;
86 struct net_device *netdev;
87
bea3348e 88 struct napi_struct napi;
0d160211 89
d634bf2c
WL		 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
84284d3c 99 struct xenbus_device *xbdev;
0d160211 100
84284d3c
JF		 101 spinlock_t tx_lock;
102 struct xen_netif_tx_front_ring tx;
103 int tx_ring_ref;
0d160211
JF		 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;
1ffb40b8 116 unsigned long link;
0d160211
JF		 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];
cefe0078 120 struct page *grant_tx_page[NET_TX_RING_SIZE];
0d160211
JF		 121 unsigned tx_skb_freelist;
122
84284d3c
JF		 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
0d160211
JF		 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
0d160211
JF		 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];
e0ce4af9
IC		 143
144 /* Statistics */
e00f85be 145 struct netfront_stats __percpu *stats;
146
e9a799ea 147 unsigned long rx_gso_checksum_fixup;
0d160211
JF		 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
1ffb40b8
IY		 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));
807540ba 163 return (unsigned long)list->skb < PAGE_OFFSET;
1ffb40b8
IY		 164}
165
0d160211
JF		 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{
1ffb40b8 173 skb_entry_set_link(&list[id], *head);
0d160211
JF		 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
3ad9b358 216static bool xennet_can_sg(struct net_device *dev)
0d160211 217{
3ad9b358 218 return dev->features & NETIF_F_SG;
0d160211
JF		 219}
220
221
222static void rx_refill_timeout(unsigned long data)
223{
224 struct net_device *dev = (struct net_device *)data;
bea3348e 225 struct netfront_info *np = netdev_priv(dev);
288379f0 226 napi_schedule(&np->napi);
0d160211
JF		 227}
228
229static int netfront_tx_slot_available(struct netfront_info *np)
230{
807540ba
ED		 231 return (np->tx.req_prod_pvt - np->tx.rsp_cons) <
232 (TX_MAX_TARGET - MAX_SKB_FRAGS - 2);
0d160211
JF		 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;
0d160211
JF		 253 grant_ref_t ref;
254 unsigned long pfn;
255 void *vaddr;
0d160211
JF		 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++) {
617a20bb 269 skb = __netdev_alloc_skb(dev, RX_COPY_THRESHOLD + NET_IP_ALIGN,
0d160211
JF		 270 GFP_ATOMIC | __GFP_NOWARN);
271 if (unlikely(!skb))
272 goto no_skb;
273
617a20bb
IY		 274 /* Align ip header to a 16 bytes boundary */
275 skb_reserve(skb, NET_IP_ALIGN);
276
0d160211
JF		 277 page = alloc_page(GFP_ATOMIC | __GFP_NOWARN);
278 if (!page) {
279 kfree_skb(skb);
280no_skb:
0d160211
JF		 281 /* Could not allocate any skbuffs. Try again later. */
282 mod_timer(&np->rx_refill_timer,
283 jiffies + (HZ/10));
fdcf7765
MJ		 284
285 /* Any skbuffs queued for refill? Force them out. */
286 if (i != 0)
287 goto refill;
0d160211
JF		 288 break;
289 }
290
093b9c71 291 skb_add_rx_frag(skb, 0, page, 0, 0, PAGE_SIZE);
0d160211
JF		 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:
5dcddfae 308 for (i = 0; ; i++) {
0d160211
JF		 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
01c68026
IC		 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]));
0d160211
JF		 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
5dcddfae 337 wmb(); /* barrier so backend seens requests */
0d160211
JF		 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)
d634bf2c 344 notify_remote_via_irq(np->rx_irq);
0d160211
JF		 345}
346
347static int xennet_open(struct net_device *dev)
348{
349 struct netfront_info *np = netdev_priv(dev);
350
bea3348e
SH		 351 napi_enable(&np->napi);
352
0d160211
JF		 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))
288379f0 358 napi_schedule(&np->napi);
0d160211
JF		 359 }
360 spin_unlock_bh(&np->rx_lock);
361
0b1ab1b8 362 netif_start_queue(dev);
0d160211
JF		 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);
f942dc25 384 if (txrsp->status == XEN_NETIF_RSP_NULL)
0d160211
JF		 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)) {
383eda32
JP		 391 pr_alert("%s: warning -- grant still in use by backend domain\n",
392 __func__);
0d160211
JF		 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;
cefe0078 400 np->grant_tx_page[id] = NULL;
0d160211
JF		 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;
f942dc25 441 tx->flags |= XEN_NETTXF_more_data;
0d160211
JF		 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
cefe0078 457 np->grant_tx_page[id] = virt_to_page(data);
0d160211
JF		 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;
f36c3747 467 struct page *page = skb_frag_page(frag);
0d160211 468
f36c3747
IC		 469 len = skb_frag_size(frag);
470 offset = frag->page_offset;
0d160211 471
f36c3747
IC		 472 /* Data must not cross a page boundary. */
473 BUG_ON(len + offset > PAGE_SIZE<<compound_order(page));
0d160211 474
f36c3747
IC		 475 /* Skip unused frames from start of page */
476 page += offset >> PAGE_SHIFT;
477 offset &= ~PAGE_MASK;
0d160211 478
f36c3747
IC		 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
cefe0078 503 np->grant_tx_page[id] = page;
f36c3747
IC		 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 }
0d160211
JF		 519 }
520
521 np->tx.req_prod_pvt = prod;
522}
523
f36c3747
IC		 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
0d160211
JF		 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);
e00f85be 551 struct netfront_stats *stats = this_cpu_ptr(np->stats);
0d160211 552 struct xen_netif_tx_request *tx;
0d160211
JF		 553 char *data = skb->data;
554 RING_IDX i;
555 grant_ref_t ref;
556 unsigned long mfn;
557 int notify;
f36c3747 558 int slots;
0d160211
JF		 559 unsigned int offset = offset_in_page(data);
560 unsigned int len = skb_headlen(skb);
cf66f9d4 561 unsigned long flags;
0d160211 562
9ecd1a75
WL		 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
f36c3747
IC		 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);
0d160211
JF		 578 goto drop;
579 }
580
cf66f9d4 581 spin_lock_irqsave(&np->tx_lock, flags);
0d160211
JF		 582
583 if (unlikely(!netif_carrier_ok(dev) ||
f36c3747 584 (slots > 1 && !xennet_can_sg(dev)) ||
fc741216 585 netif_needs_gso(skb, netif_skb_features(skb)))) {
cf66f9d4 586 spin_unlock_irqrestore(&np->tx_lock, flags);
0d160211
JF		 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);
cefe0078 603 np->grant_tx_page[id] = virt_to_page(data);
0d160211
JF		 604 tx->gref = np->grant_tx_ref[id] = ref;
605 tx->offset = offset;
606 tx->size = len;
0d160211
JF		 607
608 tx->flags = 0;
609 if (skb->ip_summed == CHECKSUM_PARTIAL)
610 /* local packet? */
f942dc25 611 tx->flags |= XEN_NETTXF_csum_blank | XEN_NETTXF_data_validated;
0d160211
JF		 612 else if (skb->ip_summed == CHECKSUM_UNNECESSARY)
613 /* remote but checksummed. */
f942dc25 614 tx->flags |= XEN_NETTXF_data_validated;
0d160211
JF		 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
e2d617c0 622 tx->flags |= XEN_NETTXF_extra_info;
0d160211
JF		 623
624 gso->u.gso.size = skb_shinfo(skb)->gso_size;
2c0057de
PD		 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;
0d160211
JF		 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;
0d160211
JF		 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)
d634bf2c 642 notify_remote_via_irq(np->tx_irq);
0d160211 643
e00f85be 644 u64_stats_update_begin(&stats->syncp);
645 stats->tx_bytes += skb->len;
646 stats->tx_packets++;
647 u64_stats_update_end(&stats->syncp);
10a273a6
JF		 648
649 /* Note: It is not safe to access skb after xennet_tx_buf_gc()! */
0d160211
JF		 650 xennet_tx_buf_gc(dev);
651
652 if (!netfront_tx_slot_available(np))
653 netif_stop_queue(dev);
654
cf66f9d4 655 spin_unlock_irqrestore(&np->tx_lock, flags);
0d160211 656
6ed10654 657 return NETDEV_TX_OK;
0d160211
JF		 658
659 drop:
09f75cd7 660 dev->stats.tx_dropped++;
0d160211 661 dev_kfree_skb(skb);
6ed10654 662 return NETDEV_TX_OK;
0d160211
JF		 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);
bea3348e 669 napi_disable(&np->napi);
0d160211
JF		 670 return 0;
671}
672
0d160211
JF		 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);
7158ff6d 741 int slots = 1;
0d160211
JF		 742 int err = 0;
743 unsigned long ret;
744
f942dc25 745 if (rx->flags & XEN_NETRXF_extra_info) {
0d160211
JF		 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
697089dc 764 * situation to the system controller to reboot the backend.
0d160211
JF		 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:
f942dc25 782 if (!(rx->flags & XEN_NETRXF_more_data))
0d160211
JF		 783 break;
784
7158ff6d 785 if (cons + slots == rp) {
0d160211 786 if (net_ratelimit())
7158ff6d 787 dev_warn(dev, "Need more slots\n");
0d160211
JF		 788 err = -ENOENT;
789 break;
790 }
791
7158ff6d
WL		 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++;
0d160211
JF		 796 }
797
7158ff6d 798 if (unlikely(slots > max)) {
0d160211 799 if (net_ratelimit())
697089dc 800 dev_warn(dev, "Too many slots\n");
0d160211
JF		 801 err = -E2BIG;
802 }
803
804 if (unlikely(err))
7158ff6d 805 np->rx.rsp_cons = cons + slots;
0d160211
JF		 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())
383eda32 815 pr_warn("GSO size must not be zero\n");
0d160211
JF		 816 return -EINVAL;
817 }
818
2c0057de
PD		 819 if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4 &&
820 gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV6) {
0d160211 821 if (net_ratelimit())
383eda32 822 pr_warn("Bad GSO type %d\n", gso->u.gso.type);
0d160211
JF		 823 return -EINVAL;
824 }
825
826 skb_shinfo(skb)->gso_size = gso->u.gso.size;
2c0057de
PD		 827 skb_shinfo(skb)->gso_type =
828 (gso->u.gso.type == XEN_NETIF_GSO_TYPE_TCPV4) ?
829 SKB_GSO_TCPV4 :
830 SKB_GSO_TCPV6;
0d160211
JF		 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);
0d160211 844 RING_IDX cons = np->rx.rsp_cons;
0d160211
JF		 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);
01c68026 850 skb_frag_t *nfrag = &skb_shinfo(nskb)->frags[0];
0d160211 851
093b9c71
JB		 852 if (shinfo->nr_frags == MAX_SKB_FRAGS) {
853 unsigned int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
0d160211 854
093b9c71
JB		 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);
0d160211
JF		 862
863 skb_shinfo(nskb)->nr_frags = 0;
864 kfree_skb(nskb);
0d160211
JF		 865 }
866
0d160211
JF		 867 return cons;
868}
869
e0ce4af9 870static int checksum_setup(struct net_device *dev, struct sk_buff *skb)
0d160211 871{
b5cf66cd 872 bool recalculate_partial_csum = false;
e0ce4af9
IC		 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;
b5cf66cd 884 recalculate_partial_csum = true;
e0ce4af9
IC		 885 }
886
887 /* A non-CHECKSUM_PARTIAL SKB does not require setup. */
888 if (skb->ip_summed != CHECKSUM_PARTIAL)
889 return 0;
0d160211 890
b5cf66cd 891 return skb_checksum_setup(skb, recalculate_partial_csum);
0d160211
JF		 892}
893
894static int handle_incoming_queue(struct net_device *dev,
09f75cd7 895 struct sk_buff_head *rxq)
0d160211 896{
e00f85be 897 struct netfront_info *np = netdev_priv(dev);
898 struct netfront_stats *stats = this_cpu_ptr(np->stats);
0d160211
JF		 899 int packets_dropped = 0;
900 struct sk_buff *skb;
901
902 while ((skb = __skb_dequeue(rxq)) != NULL) {
3683243b 903 int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
0d160211 904
093b9c71
JB		 905 if (pull_to > skb_headlen(skb))
906 __pskb_pull_tail(skb, pull_to - skb_headlen(skb));
0d160211
JF		 907
908 /* Ethernet work: Delayed to here as it peeks the header. */
909 skb->protocol = eth_type_trans(skb, dev);
910
e0ce4af9
IC		 911 if (checksum_setup(dev, skb)) {
912 kfree_skb(skb);
913 packets_dropped++;
914 dev->stats.rx_errors++;
915 continue;
0d160211
JF		 916 }
917
e00f85be 918 u64_stats_update_begin(&stats->syncp);
919 stats->rx_packets++;
920 stats->rx_bytes += skb->len;
921 u64_stats_update_end(&stats->syncp);
0d160211
JF		 922
923 /* Pass it up. */
99d3d587 924 napi_gro_receive(&np->napi, skb);
0d160211
JF		 925 }
926
927 return packets_dropped;
928}
929
bea3348e 930static int xennet_poll(struct napi_struct *napi, int budget)
0d160211 931{
bea3348e
SH		 932 struct netfront_info *np = container_of(napi, struct netfront_info, napi);
933 struct net_device *dev = np->netdev;
0d160211
JF		 934 struct sk_buff *skb;
935 struct netfront_rx_info rinfo;
936 struct xen_netif_rx_response *rx = &rinfo.rx;
937 struct xen_netif_extra_info *extras = rinfo.extras;
938 RING_IDX i, rp;
bea3348e 939 int work_done;
0d160211
JF		 940 struct sk_buff_head rxq;
941 struct sk_buff_head errq;
942 struct sk_buff_head tmpq;
943 unsigned long flags;
0d160211
JF		 944 int err;
945
946 spin_lock(&np->rx_lock);
947
0d160211
JF		 948 skb_queue_head_init(&rxq);
949 skb_queue_head_init(&errq);
950 skb_queue_head_init(&tmpq);
951
0d160211
JF		 952 rp = np->rx.sring->rsp_prod;
953 rmb(); /* Ensure we see queued responses up to 'rp'. */
954
955 i = np->rx.rsp_cons;
956 work_done = 0;
957 while ((i != rp) && (work_done < budget)) {
958 memcpy(rx, RING_GET_RESPONSE(&np->rx, i), sizeof(*rx));
959 memset(extras, 0, sizeof(rinfo.extras));
960
961 err = xennet_get_responses(np, &rinfo, rp, &tmpq);
962
963 if (unlikely(err)) {
964err:
965 while ((skb = __skb_dequeue(&tmpq)))
966 __skb_queue_tail(&errq, skb);
09f75cd7 967 dev->stats.rx_errors++;
0d160211
JF		 968 i = np->rx.rsp_cons;
969 continue;
970 }
971
972 skb = __skb_dequeue(&tmpq);
973
974 if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) {
975 struct xen_netif_extra_info *gso;
976 gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1];
977
978 if (unlikely(xennet_set_skb_gso(skb, gso))) {
979 __skb_queue_head(&tmpq, skb);
980 np->rx.rsp_cons += skb_queue_len(&tmpq);
981 goto err;
982 }
983 }
984
3683243b
IC		 985 NETFRONT_SKB_CB(skb)->pull_to = rx->status;
986 if (NETFRONT_SKB_CB(skb)->pull_to > RX_COPY_THRESHOLD)
987 NETFRONT_SKB_CB(skb)->pull_to = RX_COPY_THRESHOLD;
0d160211 988
3683243b
IC		 989 skb_shinfo(skb)->frags[0].page_offset = rx->offset;
990 skb_frag_size_set(&skb_shinfo(skb)->frags[0], rx->status);
991 skb->data_len = rx->status;
093b9c71 992 skb->len += rx->status;
0d160211
JF		 993
994 i = xennet_fill_frags(np, skb, &tmpq);
995
f942dc25 996 if (rx->flags & XEN_NETRXF_csum_blank)
0d160211 997 skb->ip_summed = CHECKSUM_PARTIAL;
f942dc25 998 else if (rx->flags & XEN_NETRXF_data_validated)
0d160211
JF		 999 skb->ip_summed = CHECKSUM_UNNECESSARY;
1000
1001 __skb_queue_tail(&rxq, skb);
1002
1003 np->rx.rsp_cons = ++i;
1004 work_done++;
1005 }
1006
56cfe5d0 1007 __skb_queue_purge(&errq);
0d160211
JF		 1008
1009 work_done -= handle_incoming_queue(dev, &rxq);
1010
1011 /* If we get a callback with very few responses, reduce fill target. */
1012 /* NB. Note exponential increase, linear decrease. */
1013 if (((np->rx.req_prod_pvt - np->rx.sring->rsp_prod) >
1014 ((3*np->rx_target) / 4)) &&
1015 (--np->rx_target < np->rx_min_target))
1016 np->rx_target = np->rx_min_target;
1017
1018 xennet_alloc_rx_buffers(dev);
1019
0d160211 1020 if (work_done < budget) {
bea3348e
SH		 1021 int more_to_do = 0;
1022
99d3d587
WL		 1023 napi_gro_flush(napi, false);
1024
0d160211
JF		 1025 local_irq_save(flags);
1026
1027 RING_FINAL_CHECK_FOR_RESPONSES(&np->rx, more_to_do);
1028 if (!more_to_do)
288379f0 1029 __napi_complete(napi);
0d160211
JF		 1030
1031 local_irq_restore(flags);
1032 }
1033
1034 spin_unlock(&np->rx_lock);
1035
bea3348e 1036 return work_done;
0d160211
JF		 1037}
1038
1039static int xennet_change_mtu(struct net_device *dev, int mtu)
1040{
9ecd1a75
WL		 1041 int max = xennet_can_sg(dev) ?
1042 XEN_NETIF_MAX_TX_SIZE - MAX_TCP_HEADER : ETH_DATA_LEN;
0d160211
JF		 1043
1044 if (mtu > max)
1045 return -EINVAL;
1046 dev->mtu = mtu;
1047 return 0;
1048}
1049
e00f85be 1050static struct rtnl_link_stats64 *xennet_get_stats64(struct net_device *dev,
1051 struct rtnl_link_stats64 *tot)
1052{
1053 struct netfront_info *np = netdev_priv(dev);
1054 int cpu;
1055
1056 for_each_possible_cpu(cpu) {
1057 struct netfront_stats *stats = per_cpu_ptr(np->stats, cpu);
1058 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1059 unsigned int start;
1060
1061 do {
1062 start = u64_stats_fetch_begin_bh(&stats->syncp);
1063
1064 rx_packets = stats->rx_packets;
1065 tx_packets = stats->tx_packets;
1066 rx_bytes = stats->rx_bytes;
1067 tx_bytes = stats->tx_bytes;
1068 } while (u64_stats_fetch_retry_bh(&stats->syncp, start));
1069
1070 tot->rx_packets += rx_packets;
1071 tot->tx_packets += tx_packets;
1072 tot->rx_bytes += rx_bytes;
1073 tot->tx_bytes += tx_bytes;
1074 }
1075
1076 tot->rx_errors = dev->stats.rx_errors;
1077 tot->tx_dropped = dev->stats.tx_dropped;
1078
1079 return tot;
1080}
1081
0d160211
JF		 1082static void xennet_release_tx_bufs(struct netfront_info *np)
1083{
1084 struct sk_buff *skb;
1085 int i;
1086
1087 for (i = 0; i < NET_TX_RING_SIZE; i++) {
1088 /* Skip over entries which are actually freelist references */
1ffb40b8 1089 if (skb_entry_is_link(&np->tx_skbs[i]))
0d160211
JF		 1090 continue;
1091
1092 skb = np->tx_skbs[i].skb;
cefe0078
AL		 1093 get_page(np->grant_tx_page[i]);
1094 gnttab_end_foreign_access(np->grant_tx_ref[i],
1095 GNTMAP_readonly,
1096 (unsigned long)page_address(np->grant_tx_page[i]));
1097 np->grant_tx_page[i] = NULL;
0d160211
JF		 1098 np->grant_tx_ref[i] = GRANT_INVALID_REF;
1099 add_id_to_freelist(&np->tx_skb_freelist, np->tx_skbs, i);
1100 dev_kfree_skb_irq(skb);
1101 }
1102}
1103
1104static void xennet_release_rx_bufs(struct netfront_info *np)
1105{
0d160211
JF		 1106 int id, ref;
1107
0d160211
JF		 1108 spin_lock_bh(&np->rx_lock);
1109
1110 for (id = 0; id < NET_RX_RING_SIZE; id++) {
cefe0078
AL		 1111 struct sk_buff *skb;
1112 struct page *page;
0d160211
JF		 1113
1114 skb = np->rx_skbs[id];
cefe0078 1115 if (!skb)
0d160211 1116 continue;
0d160211 1117
cefe0078
AL		 1118 ref = np->grant_rx_ref[id];
1119 if (ref == GRANT_INVALID_REF)
1120 continue;
0d160211 1121
cefe0078 1122 page = skb_frag_page(&skb_shinfo(skb)->frags[0]);
0d160211 1123
cefe0078
AL		 1124 /* gnttab_end_foreign_access() needs a page ref until
1125 * foreign access is ended (which may be deferred).
1126 */
1127 get_page(page);
1128 gnttab_end_foreign_access(ref, 0,
1129 (unsigned long)page_address(page));
1130 np->grant_rx_ref[id] = GRANT_INVALID_REF;
0d160211 1131
cefe0078 1132 kfree_skb(skb);
0d160211
JF		 1133 }
1134
0d160211
JF		 1135 spin_unlock_bh(&np->rx_lock);
1136}
1137
1138static void xennet_uninit(struct net_device *dev)
1139{
1140 struct netfront_info *np = netdev_priv(dev);
1141 xennet_release_tx_bufs(np);
1142 xennet_release_rx_bufs(np);
1143 gnttab_free_grant_references(np->gref_tx_head);
1144 gnttab_free_grant_references(np->gref_rx_head);
1145}
1146
c8f44aff
MM		 1147static netdev_features_t xennet_fix_features(struct net_device *dev,
1148 netdev_features_t features)
8f7b01a1
ED		 1149{
1150 struct netfront_info *np = netdev_priv(dev);
1151 int val;
1152
1153 if (features & NETIF_F_SG) {
1154 if (xenbus_scanf(XBT_NIL, np->xbdev->otherend, "feature-sg",
1155 "%d", &val) < 0)
1156 val = 0;
1157
1158 if (!val)
1159 features &= ~NETIF_F_SG;
1160 }
1161
2c0057de
PD		 1162 if (features & NETIF_F_IPV6_CSUM) {
1163 if (xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1164 "feature-ipv6-csum-offload", "%d", &val) < 0)
1165 val = 0;
1166
1167 if (!val)
1168 features &= ~NETIF_F_IPV6_CSUM;
1169 }
1170
8f7b01a1
ED		 1171 if (features & NETIF_F_TSO) {
1172 if (xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1173 "feature-gso-tcpv4", "%d", &val) < 0)
1174 val = 0;
1175
1176 if (!val)
1177 features &= ~NETIF_F_TSO;
1178 }
1179
2c0057de
PD		 1180 if (features & NETIF_F_TSO6) {
1181 if (xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1182 "feature-gso-tcpv6", "%d", &val) < 0)
1183 val = 0;
1184
1185 if (!val)
1186 features &= ~NETIF_F_TSO6;
1187 }
1188
8f7b01a1
ED		 1189 return features;
1190}
1191
c8f44aff
MM		 1192static int xennet_set_features(struct net_device *dev,
1193 netdev_features_t features)
8f7b01a1
ED		 1194{
1195 if (!(features & NETIF_F_SG) && dev->mtu > ETH_DATA_LEN) {
1196 netdev_info(dev, "Reducing MTU because no SG offload");
1197 dev->mtu = ETH_DATA_LEN;
1198 }
1199
1200 return 0;
1201}
1202
d634bf2c 1203static irqreturn_t xennet_tx_interrupt(int irq, void *dev_id)
cf66f9d4 1204{
d634bf2c
WL		 1205 struct netfront_info *np = dev_id;
1206 struct net_device *dev = np->netdev;
cf66f9d4
KRW		 1207 unsigned long flags;
1208
1209 spin_lock_irqsave(&np->tx_lock, flags);
d634bf2c
WL		 1210 xennet_tx_buf_gc(dev);
1211 spin_unlock_irqrestore(&np->tx_lock, flags);
cf66f9d4 1212
d634bf2c
WL		 1213 return IRQ_HANDLED;
1214}
1215
1216static irqreturn_t xennet_rx_interrupt(int irq, void *dev_id)
1217{
1218 struct netfront_info *np = dev_id;
1219 struct net_device *dev = np->netdev;
1220
1221 if (likely(netif_carrier_ok(dev) &&
1222 RING_HAS_UNCONSUMED_RESPONSES(&np->rx)))
cf66f9d4 1223 napi_schedule(&np->napi);
cf66f9d4 1224
d634bf2c
WL		 1225 return IRQ_HANDLED;
1226}
cf66f9d4 1227
d634bf2c
WL		 1228static irqreturn_t xennet_interrupt(int irq, void *dev_id)
1229{
1230 xennet_tx_interrupt(irq, dev_id);
1231 xennet_rx_interrupt(irq, dev_id);
cf66f9d4
KRW		 1232 return IRQ_HANDLED;
1233}
1234
1235#ifdef CONFIG_NET_POLL_CONTROLLER
1236static void xennet_poll_controller(struct net_device *dev)
1237{
1238 xennet_interrupt(0, dev);
1239}
1240#endif
1241
0a0b9d2e
SH		 1242static const struct net_device_ops xennet_netdev_ops = {
1243 .ndo_open = xennet_open,
1244 .ndo_uninit = xennet_uninit,
1245 .ndo_stop = xennet_close,
1246 .ndo_start_xmit = xennet_start_xmit,
1247 .ndo_change_mtu = xennet_change_mtu,
e00f85be 1248 .ndo_get_stats64 = xennet_get_stats64,
0a0b9d2e
SH		 1249 .ndo_set_mac_address = eth_mac_addr,
1250 .ndo_validate_addr = eth_validate_addr,
fb507934
MM		 1251 .ndo_fix_features = xennet_fix_features,
1252 .ndo_set_features = xennet_set_features,
cf66f9d4
KRW		 1253#ifdef CONFIG_NET_POLL_CONTROLLER
1254 .ndo_poll_controller = xennet_poll_controller,
1255#endif
0a0b9d2e
SH		 1256};
1257
8e0e46bb 1258static struct net_device *xennet_create_dev(struct xenbus_device *dev)
0d160211
JF		 1259{
1260 int i, err;
1261 struct net_device *netdev;
1262 struct netfront_info *np;
1263
1264 netdev = alloc_etherdev(sizeof(struct netfront_info));
41de8d4c 1265 if (!netdev)
0d160211 1266 return ERR_PTR(-ENOMEM);
0d160211
JF		 1267
1268 np = netdev_priv(netdev);
1269 np->xbdev = dev;
1270
1271 spin_lock_init(&np->tx_lock);
1272 spin_lock_init(&np->rx_lock);
1273
1274 skb_queue_head_init(&np->rx_batch);
1275 np->rx_target = RX_DFL_MIN_TARGET;
1276 np->rx_min_target = RX_DFL_MIN_TARGET;
1277 np->rx_max_target = RX_MAX_TARGET;
1278
1279 init_timer(&np->rx_refill_timer);
1280 np->rx_refill_timer.data = (unsigned long)netdev;
1281 np->rx_refill_timer.function = rx_refill_timeout;
1282
e00f85be 1283 err = -ENOMEM;
1284 np->stats = alloc_percpu(struct netfront_stats);
1285 if (np->stats == NULL)
1286 goto exit;
1287
827da44c
JS		 1288 for_each_possible_cpu(i) {
1289 struct netfront_stats *xen_nf_stats;
1290 xen_nf_stats = per_cpu_ptr(np->stats, i);
1291 u64_stats_init(&xen_nf_stats->syncp);
1292 }
1293
0d160211
JF		 1294 /* Initialise tx_skbs as a free chain containing every entry. */
1295 np->tx_skb_freelist = 0;
1296 for (i = 0; i < NET_TX_RING_SIZE; i++) {
1ffb40b8 1297 skb_entry_set_link(&np->tx_skbs[i], i+1);
0d160211
JF		 1298 np->grant_tx_ref[i] = GRANT_INVALID_REF;
1299 }
1300
1301 /* Clear out rx_skbs */
1302 for (i = 0; i < NET_RX_RING_SIZE; i++) {
1303 np->rx_skbs[i] = NULL;
1304 np->grant_rx_ref[i] = GRANT_INVALID_REF;
cefe0078 1305 np->grant_tx_page[i] = NULL;
0d160211
JF		 1306 }
1307
1308 /* A grant for every tx ring slot */
1309 if (gnttab_alloc_grant_references(TX_MAX_TARGET,
1310 &np->gref_tx_head) < 0) {
383eda32 1311 pr_alert("can't alloc tx grant refs\n");
0d160211 1312 err = -ENOMEM;
e00f85be 1313 goto exit_free_stats;
0d160211
JF		 1314 }
1315 /* A grant for every rx ring slot */
1316 if (gnttab_alloc_grant_references(RX_MAX_TARGET,
1317 &np->gref_rx_head) < 0) {
383eda32 1318 pr_alert("can't alloc rx grant refs\n");
0d160211
JF		 1319 err = -ENOMEM;
1320 goto exit_free_tx;
1321 }
1322
0a0b9d2e
SH		 1323 netdev->netdev_ops = &xennet_netdev_ops;
1324
bea3348e 1325 netif_napi_add(netdev, &np->napi, xennet_poll, 64);
fb507934
MM		 1326 netdev->features = NETIF_F_IP_CSUM | NETIF_F_RXCSUM |
1327 NETIF_F_GSO_ROBUST;
2c0057de
PD		 1328 netdev->hw_features = NETIF_F_SG |
1329 NETIF_F_IPV6_CSUM |
1330 NETIF_F_TSO | NETIF_F_TSO6;
0d160211 1331
fc3e5941
IC		 1332 /*
1333 * Assume that all hw features are available for now. This set
1334 * will be adjusted by the call to netdev_update_features() in
1335 * xennet_connect() which is the earliest point where we can
1336 * negotiate with the backend regarding supported features.
1337 */
1338 netdev->features |= netdev->hw_features;
1339
0d160211 1340 SET_ETHTOOL_OPS(netdev, &xennet_ethtool_ops);
0d160211
JF		 1341 SET_NETDEV_DEV(netdev, &dev->dev);
1342
9ecd1a75
WL		 1343 netif_set_gso_max_size(netdev, XEN_NETIF_MAX_TX_SIZE - MAX_TCP_HEADER);
1344
0d160211
JF		 1345 np->netdev = netdev;
1346
1347 netif_carrier_off(netdev);
1348
1349 return netdev;
1350
1351 exit_free_tx:
1352 gnttab_free_grant_references(np->gref_tx_head);
e00f85be 1353 exit_free_stats:
1354 free_percpu(np->stats);
0d160211
JF		 1355 exit:
1356 free_netdev(netdev);
1357 return ERR_PTR(err);
1358}
1359
1360/**
1361 * Entry point to this code when a new device is created. Allocate the basic
1362 * structures and the ring buffers for communication with the backend, and
1363 * inform the backend of the appropriate details for those.
1364 */
8e0e46bb 1365static int netfront_probe(struct xenbus_device *dev,
1dd06ae8 1366 const struct xenbus_device_id *id)
0d160211
JF		 1367{
1368 int err;
1369 struct net_device *netdev;
1370 struct netfront_info *info;
1371
1372 netdev = xennet_create_dev(dev);
1373 if (IS_ERR(netdev)) {
1374 err = PTR_ERR(netdev);
1375 xenbus_dev_fatal(dev, err, "creating netdev");
1376 return err;
1377 }
1378
1379 info = netdev_priv(netdev);
1b713e00 1380 dev_set_drvdata(&dev->dev, info);
0d160211
JF		 1381
1382 err = register_netdev(info->netdev);
1383 if (err) {
383eda32 1384 pr_warn("%s: register_netdev err=%d\n", __func__, err);
0d160211
JF		 1385 goto fail;
1386 }
1387
1388 err = xennet_sysfs_addif(info->netdev);
1389 if (err) {
1390 unregister_netdev(info->netdev);
383eda32 1391 pr_warn("%s: add sysfs failed err=%d\n", __func__, err);
0d160211
JF		 1392 goto fail;
1393 }
1394
1395 return 0;
1396
1397 fail:
1398 free_netdev(netdev);
1b713e00 1399 dev_set_drvdata(&dev->dev, NULL);
0d160211
JF		 1400 return err;
1401}
1402
1403static void xennet_end_access(int ref, void *page)
1404{
1405 /* This frees the page as a side-effect */
1406 if (ref != GRANT_INVALID_REF)
1407 gnttab_end_foreign_access(ref, 0, (unsigned long)page);
1408}
1409
1410static void xennet_disconnect_backend(struct netfront_info *info)
1411{
1412 /* Stop old i/f to prevent errors whilst we rebuild the state. */
1413 spin_lock_bh(&info->rx_lock);
1414 spin_lock_irq(&info->tx_lock);
1415 netif_carrier_off(info->netdev);
1416 spin_unlock_irq(&info->tx_lock);
1417 spin_unlock_bh(&info->rx_lock);
1418
d634bf2c
WL		 1419 if (info->tx_irq && (info->tx_irq == info->rx_irq))
1420 unbind_from_irqhandler(info->tx_irq, info);
1421 if (info->tx_irq && (info->tx_irq != info->rx_irq)) {
1422 unbind_from_irqhandler(info->tx_irq, info);
1423 unbind_from_irqhandler(info->rx_irq, info);
1424 }
1425 info->tx_evtchn = info->rx_evtchn = 0;
1426 info->tx_irq = info->rx_irq = 0;
0d160211
JF		 1427
1428 /* End access and free the pages */
1429 xennet_end_access(info->tx_ring_ref, info->tx.sring);
1430 xennet_end_access(info->rx_ring_ref, info->rx.sring);
1431
1432 info->tx_ring_ref = GRANT_INVALID_REF;
1433 info->rx_ring_ref = GRANT_INVALID_REF;
1434 info->tx.sring = NULL;
1435 info->rx.sring = NULL;
1436}
1437
1438/**
1439 * We are reconnecting to the backend, due to a suspend/resume, or a backend
1440 * driver restart. We tear down our netif structure and recreate it, but
1441 * leave the device-layer structures intact so that this is transparent to the
1442 * rest of the kernel.
1443 */
1444static int netfront_resume(struct xenbus_device *dev)
1445{
1b713e00 1446 struct netfront_info *info = dev_get_drvdata(&dev->dev);
0d160211
JF		 1447
1448 dev_dbg(&dev->dev, "%s\n", dev->nodename);
1449
1450 xennet_disconnect_backend(info);
1451 return 0;
1452}
1453
1454static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[])
1455{
1456 char *s, *e, *macstr;
1457 int i;
1458
1459 macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL);
1460 if (IS_ERR(macstr))
1461 return PTR_ERR(macstr);
1462
1463 for (i = 0; i < ETH_ALEN; i++) {
1464 mac[i] = simple_strtoul(s, &e, 16);
1465 if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) {
1466 kfree(macstr);
1467 return -ENOENT;
1468 }
1469 s = e+1;
1470 }
1471
1472 kfree(macstr);
1473 return 0;
1474}
1475
d634bf2c
WL		 1476static int setup_netfront_single(struct netfront_info *info)
1477{
1478 int err;
1479
1480 err = xenbus_alloc_evtchn(info->xbdev, &info->tx_evtchn);
1481 if (err < 0)
1482 goto fail;
1483
1484 err = bind_evtchn_to_irqhandler(info->tx_evtchn,
1485 xennet_interrupt,
1486 0, info->netdev->name, info);
1487 if (err < 0)
1488 goto bind_fail;
1489 info->rx_evtchn = info->tx_evtchn;
1490 info->rx_irq = info->tx_irq = err;
1491
1492 return 0;
1493
1494bind_fail:
1495 xenbus_free_evtchn(info->xbdev, info->tx_evtchn);
1496 info->tx_evtchn = 0;
1497fail:
1498 return err;
1499}
1500
1501static int setup_netfront_split(struct netfront_info *info)
1502{
1503 int err;
1504
1505 err = xenbus_alloc_evtchn(info->xbdev, &info->tx_evtchn);
1506 if (err < 0)
1507 goto fail;
1508 err = xenbus_alloc_evtchn(info->xbdev, &info->rx_evtchn);
1509 if (err < 0)
1510 goto alloc_rx_evtchn_fail;
1511
1512 snprintf(info->tx_irq_name, sizeof(info->tx_irq_name),
1513 "%s-tx", info->netdev->name);
1514 err = bind_evtchn_to_irqhandler(info->tx_evtchn,
1515 xennet_tx_interrupt,
1516 0, info->tx_irq_name, info);
1517 if (err < 0)
1518 goto bind_tx_fail;
1519 info->tx_irq = err;
1520
1521 snprintf(info->rx_irq_name, sizeof(info->rx_irq_name),
1522 "%s-rx", info->netdev->name);
1523 err = bind_evtchn_to_irqhandler(info->rx_evtchn,
1524 xennet_rx_interrupt,
1525 0, info->rx_irq_name, info);
1526 if (err < 0)
1527 goto bind_rx_fail;
1528 info->rx_irq = err;
1529
1530 return 0;
1531
1532bind_rx_fail:
1533 unbind_from_irqhandler(info->tx_irq, info);
1534 info->tx_irq = 0;
1535bind_tx_fail:
1536 xenbus_free_evtchn(info->xbdev, info->rx_evtchn);
1537 info->rx_evtchn = 0;
1538alloc_rx_evtchn_fail:
1539 xenbus_free_evtchn(info->xbdev, info->tx_evtchn);
1540 info->tx_evtchn = 0;
1541fail:
1542 return err;
1543}
1544
0d160211
JF		 1545static int setup_netfront(struct xenbus_device *dev, struct netfront_info *info)
1546{
1547 struct xen_netif_tx_sring *txs;
1548 struct xen_netif_rx_sring *rxs;
1549 int err;
1550 struct net_device *netdev = info->netdev;
d634bf2c 1551 unsigned int feature_split_evtchn;
0d160211
JF		 1552
1553 info->tx_ring_ref = GRANT_INVALID_REF;
1554 info->rx_ring_ref = GRANT_INVALID_REF;
1555 info->rx.sring = NULL;
1556 info->tx.sring = NULL;
1557 netdev->irq = 0;
1558
d634bf2c
WL		 1559 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1560 "feature-split-event-channels", "%u",
1561 &feature_split_evtchn);
1562 if (err < 0)
1563 feature_split_evtchn = 0;
1564
0d160211
JF		 1565 err = xen_net_read_mac(dev, netdev->dev_addr);
1566 if (err) {
1567 xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename);
1568 goto fail;
1569 }
1570
a144ff09 1571 txs = (struct xen_netif_tx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
0d160211
JF		 1572 if (!txs) {
1573 err = -ENOMEM;
1574 xenbus_dev_fatal(dev, err, "allocating tx ring page");
1575 goto fail;
1576 }
1577 SHARED_RING_INIT(txs);
1578 FRONT_RING_INIT(&info->tx, txs, PAGE_SIZE);
1579
1580 err = xenbus_grant_ring(dev, virt_to_mfn(txs));
1ca2983a
WL		 1581 if (err < 0)
1582 goto grant_tx_ring_fail;
0d160211
JF		 1583
1584 info->tx_ring_ref = err;
a144ff09 1585 rxs = (struct xen_netif_rx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
0d160211
JF		 1586 if (!rxs) {
1587 err = -ENOMEM;
1588 xenbus_dev_fatal(dev, err, "allocating rx ring page");
1ca2983a 1589 goto alloc_rx_ring_fail;
0d160211
JF		 1590 }
1591 SHARED_RING_INIT(rxs);
1592 FRONT_RING_INIT(&info->rx, rxs, PAGE_SIZE);
1593
1594 err = xenbus_grant_ring(dev, virt_to_mfn(rxs));
1ca2983a
WL		 1595 if (err < 0)
1596 goto grant_rx_ring_fail;
0d160211
JF		 1597 info->rx_ring_ref = err;
1598
d634bf2c
WL		 1599 if (feature_split_evtchn)
1600 err = setup_netfront_split(info);
1601 /* setup single event channel if
1602 * a) feature-split-event-channels == 0
1603 * b) feature-split-event-channels == 1 but failed to setup
1604 */
1605 if (!feature_split_evtchn || (feature_split_evtchn && err))
1606 err = setup_netfront_single(info);
1607
0d160211 1608 if (err)
1ca2983a 1609 goto alloc_evtchn_fail;
0d160211 1610
0d160211
JF		 1611 return 0;
1612
1ca2983a
WL		 1613 /* If we fail to setup netfront, it is safe to just revoke access to
1614 * granted pages because backend is not accessing it at this point.
1615 */
1ca2983a
WL		 1616alloc_evtchn_fail:
1617 gnttab_end_foreign_access_ref(info->rx_ring_ref, 0);
1618grant_rx_ring_fail:
1619 free_page((unsigned long)rxs);
1620alloc_rx_ring_fail:
1621 gnttab_end_foreign_access_ref(info->tx_ring_ref, 0);
1622grant_tx_ring_fail:
1623 free_page((unsigned long)txs);
1624fail:
0d160211
JF		 1625 return err;
1626}
1627
1628/* Common code used when first setting up, and when resuming. */
f502bf2b 1629static int talk_to_netback(struct xenbus_device *dev,
0d160211
JF		 1630 struct netfront_info *info)
1631{
1632 const char *message;
1633 struct xenbus_transaction xbt;
1634 int err;
1635
1636 /* Create shared ring, alloc event channel. */
1637 err = setup_netfront(dev, info);
1638 if (err)
1639 goto out;
1640
1641again:
1642 err = xenbus_transaction_start(&xbt);
1643 if (err) {
1644 xenbus_dev_fatal(dev, err, "starting transaction");
1645 goto destroy_ring;
1646 }
1647
1648 err = xenbus_printf(xbt, dev->nodename, "tx-ring-ref", "%u",
1649 info->tx_ring_ref);
1650 if (err) {
1651 message = "writing tx ring-ref";
1652 goto abort_transaction;
1653 }
1654 err = xenbus_printf(xbt, dev->nodename, "rx-ring-ref", "%u",
1655 info->rx_ring_ref);
1656 if (err) {
1657 message = "writing rx ring-ref";
1658 goto abort_transaction;
1659 }
d634bf2c
WL		 1660
1661 if (info->tx_evtchn == info->rx_evtchn) {
1662 err = xenbus_printf(xbt, dev->nodename,
1663 "event-channel", "%u", info->tx_evtchn);
1664 if (err) {
1665 message = "writing event-channel";
1666 goto abort_transaction;
1667 }
1668 } else {
1669 err = xenbus_printf(xbt, dev->nodename,
1670 "event-channel-tx", "%u", info->tx_evtchn);
1671 if (err) {
1672 message = "writing event-channel-tx";
1673 goto abort_transaction;
1674 }
1675 err = xenbus_printf(xbt, dev->nodename,
1676 "event-channel-rx", "%u", info->rx_evtchn);
1677 if (err) {
1678 message = "writing event-channel-rx";
1679 goto abort_transaction;
1680 }
0d160211
JF		 1681 }
1682
1683 err = xenbus_printf(xbt, dev->nodename, "request-rx-copy", "%u",
1684 1);
1685 if (err) {
1686 message = "writing request-rx-copy";
1687 goto abort_transaction;
1688 }
1689
1690 err = xenbus_printf(xbt, dev->nodename, "feature-rx-notify", "%d", 1);
1691 if (err) {
1692 message = "writing feature-rx-notify";
1693 goto abort_transaction;
1694 }
1695
1696 err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", 1);
1697 if (err) {
1698 message = "writing feature-sg";
1699 goto abort_transaction;
1700 }
1701
1702 err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", "%d", 1);
1703 if (err) {
1704 message = "writing feature-gso-tcpv4";
1705 goto abort_transaction;
1706 }
1707
2c0057de
PD		 1708 err = xenbus_write(xbt, dev->nodename, "feature-gso-tcpv6", "1");
1709 if (err) {
1710 message = "writing feature-gso-tcpv6";
1711 goto abort_transaction;
1712 }
1713
1714 err = xenbus_write(xbt, dev->nodename, "feature-ipv6-csum-offload",
1715 "1");
1716 if (err) {
1717 message = "writing feature-ipv6-csum-offload";
1718 goto abort_transaction;
1719 }
1720
0d160211
JF		 1721 err = xenbus_transaction_end(xbt, 0);
1722 if (err) {
1723 if (err == -EAGAIN)
1724 goto again;
1725 xenbus_dev_fatal(dev, err, "completing transaction");
1726 goto destroy_ring;
1727 }
1728
1729 return 0;
1730
1731 abort_transaction:
1732 xenbus_transaction_end(xbt, 1);
1733 xenbus_dev_fatal(dev, err, "%s", message);
1734 destroy_ring:
1735 xennet_disconnect_backend(info);
1736 out:
1737 return err;
1738}
1739
0d160211
JF		 1740static int xennet_connect(struct net_device *dev)
1741{
1742 struct netfront_info *np = netdev_priv(dev);
1743 int i, requeue_idx, err;
1744 struct sk_buff *skb;
1745 grant_ref_t ref;
1746 struct xen_netif_rx_request *req;
1747 unsigned int feature_rx_copy;
1748
1749 err = xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1750 "feature-rx-copy", "%u", &feature_rx_copy);
1751 if (err != 1)
1752 feature_rx_copy = 0;
1753
1754 if (!feature_rx_copy) {
1755 dev_info(&dev->dev,
898eb71c 1756 "backend does not support copying receive path\n");
0d160211
JF		 1757 return -ENODEV;
1758 }
1759
f502bf2b 1760 err = talk_to_netback(np->xbdev, np);
0d160211
JF		 1761 if (err)
1762 return err;
1763
1ba37c51 1764 rtnl_lock();
fb507934 1765 netdev_update_features(dev);
1ba37c51 1766 rtnl_unlock();
0d160211
JF		 1767
1768 spin_lock_bh(&np->rx_lock);
1769 spin_lock_irq(&np->tx_lock);
1770
1771 /* Step 1: Discard all pending TX packet fragments. */
1772 xennet_release_tx_bufs(np);
1773
1774 /* Step 2: Rebuild the RX buffer freelist and the RX ring itself. */
1775 for (requeue_idx = 0, i = 0; i < NET_RX_RING_SIZE; i++) {
01c68026
IC		 1776 skb_frag_t *frag;
1777 const struct page *page;
0d160211
JF		 1778 if (!np->rx_skbs[i])
1779 continue;
1780
1781 skb = np->rx_skbs[requeue_idx] = xennet_get_rx_skb(np, i);
1782 ref = np->grant_rx_ref[requeue_idx] = xennet_get_rx_ref(np, i);
1783 req = RING_GET_REQUEST(&np->rx, requeue_idx);
1784
01c68026
IC		 1785 frag = &skb_shinfo(skb)->frags[0];
1786 page = skb_frag_page(frag);
0d160211
JF		 1787 gnttab_grant_foreign_access_ref(
1788 ref, np->xbdev->otherend_id,
01c68026 1789 pfn_to_mfn(page_to_pfn(page)),
0d160211
JF		 1790 0);
1791 req->gref = ref;
1792 req->id = requeue_idx;
1793
1794 requeue_idx++;
1795 }
1796
1797 np->rx.req_prod_pvt = requeue_idx;
1798
1799 /*
1800 * Step 3: All public and private state should now be sane. Get
1801 * ready to start sending and receiving packets and give the driver
1802 * domain a kick because we've probably just requeued some
1803 * packets.
1804 */
1805 netif_carrier_on(np->netdev);
d634bf2c
WL		 1806 notify_remote_via_irq(np->tx_irq);
1807 if (np->tx_irq != np->rx_irq)
1808 notify_remote_via_irq(np->rx_irq);
0d160211
JF		 1809 xennet_tx_buf_gc(dev);
1810 xennet_alloc_rx_buffers(dev);
1811
1812 spin_unlock_irq(&np->tx_lock);
1813 spin_unlock_bh(&np->rx_lock);
1814
1815 return 0;
1816}
1817
1818/**
1819 * Callback received when the backend's state changes.
1820 */
f502bf2b 1821static void netback_changed(struct xenbus_device *dev,
0d160211
JF		 1822 enum xenbus_state backend_state)
1823{
1b713e00 1824 struct netfront_info *np = dev_get_drvdata(&dev->dev);
0d160211
JF		 1825 struct net_device *netdev = np->netdev;
1826
1827 dev_dbg(&dev->dev, "%s\n", xenbus_strstate(backend_state));
1828
1829 switch (backend_state) {
1830 case XenbusStateInitialising:
1831 case XenbusStateInitialised:
b78c9512
NI		 1832 case XenbusStateReconfiguring:
1833 case XenbusStateReconfigured:
0d160211
JF		 1834 case XenbusStateUnknown:
1835 case XenbusStateClosed:
1836 break;
1837
1838 case XenbusStateInitWait:
1839 if (dev->state != XenbusStateInitialising)
1840 break;
1841 if (xennet_connect(netdev) != 0)
1842 break;
1843 xenbus_switch_state(dev, XenbusStateConnected);
08e34eb1
LE		 1844 break;
1845
1846 case XenbusStateConnected:
ee89bab1 1847 netdev_notify_peers(netdev);
0d160211
JF		 1848 break;
1849
1850 case XenbusStateClosing:
1851 xenbus_frontend_closed(dev);
1852 break;
1853 }
1854}
1855
e0ce4af9
IC		 1856static const struct xennet_stat {
1857 char name[ETH_GSTRING_LEN];
1858 u16 offset;
1859} xennet_stats[] = {
1860 {
1861 "rx_gso_checksum_fixup",
1862 offsetof(struct netfront_info, rx_gso_checksum_fixup)
1863 },
1864};
1865
1866static int xennet_get_sset_count(struct net_device *dev, int string_set)
1867{
1868 switch (string_set) {
1869 case ETH_SS_STATS:
1870 return ARRAY_SIZE(xennet_stats);
1871 default:
1872 return -EINVAL;
1873 }
1874}
1875
1876static void xennet_get_ethtool_stats(struct net_device *dev,
1877 struct ethtool_stats *stats, u64 * data)
1878{
1879 void *np = netdev_priv(dev);
1880 int i;
1881
1882 for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
48f26d51 1883 data[i] = *(unsigned long *)(np + xennet_stats[i].offset);
e0ce4af9
IC		 1884}
1885
1886static void xennet_get_strings(struct net_device *dev, u32 stringset, u8 * data)
1887{
1888 int i;
1889
1890 switch (stringset) {
1891 case ETH_SS_STATS:
1892 for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
1893 memcpy(data + i * ETH_GSTRING_LEN,
1894 xennet_stats[i].name, ETH_GSTRING_LEN);
1895 break;
1896 }
1897}
1898
0fc0b732 1899static const struct ethtool_ops xennet_ethtool_ops =
0d160211 1900{
0d160211 1901 .get_link = ethtool_op_get_link,
e0ce4af9
IC		 1902
1903 .get_sset_count = xennet_get_sset_count,
1904 .get_ethtool_stats = xennet_get_ethtool_stats,
1905 .get_strings = xennet_get_strings,
0d160211
JF		 1906};
1907
1908#ifdef CONFIG_SYSFS
1909static ssize_t show_rxbuf_min(struct device *dev,
1910 struct device_attribute *attr, char *buf)
1911{
1912 struct net_device *netdev = to_net_dev(dev);
1913 struct netfront_info *info = netdev_priv(netdev);
1914
1915 return sprintf(buf, "%u\n", info->rx_min_target);
1916}
1917
1918static ssize_t store_rxbuf_min(struct device *dev,
1919 struct device_attribute *attr,
1920 const char *buf, size_t len)
1921{
1922 struct net_device *netdev = to_net_dev(dev);
1923 struct netfront_info *np = netdev_priv(netdev);
1924 char *endp;
1925 unsigned long target;
1926
1927 if (!capable(CAP_NET_ADMIN))
1928 return -EPERM;
1929
1930 target = simple_strtoul(buf, &endp, 0);
1931 if (endp == buf)
1932 return -EBADMSG;
1933
1934 if (target < RX_MIN_TARGET)
1935 target = RX_MIN_TARGET;
1936 if (target > RX_MAX_TARGET)
1937 target = RX_MAX_TARGET;
1938
1939 spin_lock_bh(&np->rx_lock);
1940 if (target > np->rx_max_target)
1941 np->rx_max_target = target;
1942 np->rx_min_target = target;
1943 if (target > np->rx_target)
1944 np->rx_target = target;
1945
1946 xennet_alloc_rx_buffers(netdev);
1947
1948 spin_unlock_bh(&np->rx_lock);
1949 return len;
1950}
1951
1952static ssize_t show_rxbuf_max(struct device *dev,
1953 struct device_attribute *attr, char *buf)
1954{
1955 struct net_device *netdev = to_net_dev(dev);
1956 struct netfront_info *info = netdev_priv(netdev);
1957
1958 return sprintf(buf, "%u\n", info->rx_max_target);
1959}
1960
1961static ssize_t store_rxbuf_max(struct device *dev,
1962 struct device_attribute *attr,
1963 const char *buf, size_t len)
1964{
1965 struct net_device *netdev = to_net_dev(dev);
1966 struct netfront_info *np = netdev_priv(netdev);
1967 char *endp;
1968 unsigned long target;
1969
1970 if (!capable(CAP_NET_ADMIN))
1971 return -EPERM;
1972
1973 target = simple_strtoul(buf, &endp, 0);
1974 if (endp == buf)
1975 return -EBADMSG;
1976
1977 if (target < RX_MIN_TARGET)
1978 target = RX_MIN_TARGET;
1979 if (target > RX_MAX_TARGET)
1980 target = RX_MAX_TARGET;
1981
1982 spin_lock_bh(&np->rx_lock);
1983 if (target < np->rx_min_target)
1984 np->rx_min_target = target;
1985 np->rx_max_target = target;
1986 if (target < np->rx_target)
1987 np->rx_target = target;
1988
1989 xennet_alloc_rx_buffers(netdev);
1990
1991 spin_unlock_bh(&np->rx_lock);
1992 return len;
1993}
1994
1995static ssize_t show_rxbuf_cur(struct device *dev,
1996 struct device_attribute *attr, char *buf)
1997{
1998 struct net_device *netdev = to_net_dev(dev);
1999 struct netfront_info *info = netdev_priv(netdev);
2000
2001 return sprintf(buf, "%u\n", info->rx_target);
2002}
2003
2004static struct device_attribute xennet_attrs[] = {
2005 __ATTR(rxbuf_min, S_IRUGO|S_IWUSR, show_rxbuf_min, store_rxbuf_min),
2006 __ATTR(rxbuf_max, S_IRUGO|S_IWUSR, show_rxbuf_max, store_rxbuf_max),
2007 __ATTR(rxbuf_cur, S_IRUGO, show_rxbuf_cur, NULL),
2008};
2009
2010static int xennet_sysfs_addif(struct net_device *netdev)
2011{
2012 int i;
2013 int err;
2014
2015 for (i = 0; i < ARRAY_SIZE(xennet_attrs); i++) {
2016 err = device_create_file(&netdev->dev,
2017 &xennet_attrs[i]);
2018 if (err)
2019 goto fail;
2020 }
2021 return 0;
2022
2023 fail:
2024 while (--i >= 0)
2025 device_remove_file(&netdev->dev, &xennet_attrs[i]);
2026 return err;
2027}
2028
2029static void xennet_sysfs_delif(struct net_device *netdev)
2030{
2031 int i;
2032
2033 for (i = 0; i < ARRAY_SIZE(xennet_attrs); i++)
2034 device_remove_file(&netdev->dev, &xennet_attrs[i]);
2035}
2036
2037#endif /* CONFIG_SYSFS */
2038
73db144b 2039static const struct xenbus_device_id netfront_ids[] = {
0d160211
JF		 2040 { "vif" },
2041 { "" }
2042};
2043
2044
8e0e46bb 2045static int xennet_remove(struct xenbus_device *dev)
0d160211 2046{
1b713e00 2047 struct netfront_info *info = dev_get_drvdata(&dev->dev);
0d160211
JF		 2048
2049 dev_dbg(&dev->dev, "%s\n", dev->nodename);
2050
0d160211
JF		 2051 xennet_disconnect_backend(info);
2052
0d160211
JF		 2053 xennet_sysfs_delif(info->netdev);
2054
6bc96d04
IC		 2055 unregister_netdev(info->netdev);
2056
2057 del_timer_sync(&info->rx_refill_timer);
2058
e00f85be 2059 free_percpu(info->stats);
2060
0d160211
JF		 2061 free_netdev(info->netdev);
2062
2063 return 0;
2064}
2065
73db144b 2066static DEFINE_XENBUS_DRIVER(netfront, ,
0d160211 2067 .probe = netfront_probe,
8e0e46bb 2068 .remove = xennet_remove,
0d160211 2069 .resume = netfront_resume,
f502bf2b 2070 .otherend_changed = netback_changed,
73db144b 2071);
0d160211
JF		 2072
2073static int __init netif_init(void)
2074{
6e833587 2075 if (!xen_domain())
0d160211
JF		 2076 return -ENODEV;
2077
51c71a3b 2078 if (!xen_has_pv_nic_devices())
b9136d20
IM		 2079 return -ENODEV;
2080
383eda32 2081 pr_info("Initialising Xen virtual ethernet driver\n");
0d160211 2082
ffb78a26 2083 return xenbus_register_frontend(&netfront_driver);
0d160211
JF		 2084}
2085module_init(netif_init);
2086
2087
2088static void __exit netif_exit(void)
2089{
ffb78a26 2090 xenbus_unregister_driver(&netfront_driver);
0d160211
JF		 2091}
2092module_exit(netif_exit);
2093
2094MODULE_DESCRIPTION("Xen virtual network device frontend");
2095MODULE_LICENSE("GPL");
d2f0c52b 2096MODULE_ALIAS("xen:vif");
4f93f09b 2097MODULE_ALIAS("xennet");
Linux kernel - Deliverables
This page took 1.386278 seconds and 5 git commands to generate.