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/types.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 |
64 | static unsigned int xennet_max_queues; |
65 | module_param_named(max_queues, xennet_max_queues, uint, 0644); |
66 | MODULE_PARM_DESC(max_queues, |
67 | "Maximum number of queues per virtual interface" ); |
68 | |
69 | static bool __read_mostly xennet_trusted = true; |
70 | module_param_named(trusted, xennet_trusted, bool, 0644); |
71 | MODULE_PARM_DESC(trusted, "Is the backend trusted" ); |
72 | |
73 | #define XENNET_TIMEOUT (5 * HZ) |
74 | |
75 | static const struct ethtool_ops xennet_ethtool_ops; |
76 | |
77 | struct 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 | |
97 | static DECLARE_WAIT_QUEUE_HEAD(module_wq); |
98 | |
99 | struct netfront_stats { |
100 | u64 packets; |
101 | u64 bytes; |
102 | struct u64_stats_sync syncp; |
103 | }; |
104 | |
105 | struct netfront_info; |
106 | |
107 | struct 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 | |
160 | struct 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 | |
186 | struct netfront_rx_info { |
187 | struct xen_netif_rx_response rx; |
188 | struct xen_netif_extra_info [XEN_NETIF_EXTRA_TYPE_MAX - 1]; |
189 | }; |
190 | |
191 | /* |
192 | * Access macros for acquiring freeing slots in tx_skbs[]. |
193 | */ |
194 | |
195 | static void add_id_to_list(unsigned *head, unsigned short *list, |
196 | unsigned short id) |
197 | { |
198 | list[id] = *head; |
199 | *head = id; |
200 | } |
201 | |
202 | static 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 | |
213 | static int xennet_rxidx(RING_IDX idx) |
214 | { |
215 | return idx & (NET_RX_RING_SIZE - 1); |
216 | } |
217 | |
218 | static struct sk_buff *xennet_get_rx_skb(struct netfront_queue *queue, |
219 | RING_IDX ri) |
220 | { |
221 | int i = xennet_rxidx(idx: ri); |
222 | struct sk_buff *skb = queue->rx_skbs[i]; |
223 | queue->rx_skbs[i] = NULL; |
224 | return skb; |
225 | } |
226 | |
227 | static grant_ref_t xennet_get_rx_ref(struct netfront_queue *queue, |
228 | RING_IDX ri) |
229 | { |
230 | int i = xennet_rxidx(idx: 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 |
237 | static const struct attribute_group xennet_dev_group; |
238 | #endif |
239 | |
240 | static bool xennet_can_sg(struct net_device *dev) |
241 | { |
242 | return dev->features & NETIF_F_SG; |
243 | } |
244 | |
245 | |
246 | static void rx_refill_timeout(struct timer_list *t) |
247 | { |
248 | struct netfront_queue *queue = from_timer(queue, t, rx_refill_timer); |
249 | napi_schedule(n: &queue->napi); |
250 | } |
251 | |
252 | static 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 | |
258 | static 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, index: 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(dev_queue: netdev_get_tx_queue(dev, index: queue->id)); |
267 | } |
268 | |
269 | |
270 | static 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(dev: 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(pool: 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, i: 0, page, off: 0, size: 0, PAGE_SIZE); |
288 | skb_mark_for_recycle(skb); |
289 | |
290 | /* Align ip header to a 16 bytes boundary */ |
291 | skb_reserve(skb, NET_IP_ALIGN); |
292 | skb->dev = queue->info->netdev; |
293 | |
294 | return skb; |
295 | } |
296 | |
297 | |
298 | static void xennet_alloc_rx_buffers(struct netfront_queue *queue) |
299 | { |
300 | RING_IDX req_prod = queue->rx.req_prod_pvt; |
301 | int notify; |
302 | int err = 0; |
303 | |
304 | if (unlikely(!netif_carrier_ok(queue->info->netdev))) |
305 | return; |
306 | |
307 | for (req_prod = queue->rx.req_prod_pvt; |
308 | req_prod - queue->rx.rsp_cons < NET_RX_RING_SIZE; |
309 | req_prod++) { |
310 | struct sk_buff *skb; |
311 | unsigned short id; |
312 | grant_ref_t ref; |
313 | struct page *page; |
314 | struct xen_netif_rx_request *req; |
315 | |
316 | skb = xennet_alloc_one_rx_buffer(queue); |
317 | if (!skb) { |
318 | err = -ENOMEM; |
319 | break; |
320 | } |
321 | |
322 | id = xennet_rxidx(idx: req_prod); |
323 | |
324 | BUG_ON(queue->rx_skbs[id]); |
325 | queue->rx_skbs[id] = skb; |
326 | |
327 | ref = gnttab_claim_grant_reference(pprivate_head: &queue->gref_rx_head); |
328 | WARN_ON_ONCE(IS_ERR_VALUE((unsigned long)(int)ref)); |
329 | queue->grant_rx_ref[id] = ref; |
330 | |
331 | page = skb_frag_page(frag: &skb_shinfo(skb)->frags[0]); |
332 | |
333 | req = RING_GET_REQUEST(&queue->rx, req_prod); |
334 | gnttab_page_grant_foreign_access_ref_one(ref, |
335 | domid: queue->info->xbdev->otherend_id, |
336 | page, |
337 | readonly: 0); |
338 | req->id = id; |
339 | req->gref = ref; |
340 | } |
341 | |
342 | queue->rx.req_prod_pvt = req_prod; |
343 | |
344 | /* Try again later if there are not enough requests or skb allocation |
345 | * failed. |
346 | * Enough requests is quantified as the sum of newly created slots and |
347 | * the unconsumed slots at the backend. |
348 | */ |
349 | if (req_prod - queue->rx.rsp_cons < NET_RX_SLOTS_MIN || |
350 | unlikely(err)) { |
351 | mod_timer(timer: &queue->rx_refill_timer, expires: jiffies + (HZ/10)); |
352 | return; |
353 | } |
354 | |
355 | RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->rx, notify); |
356 | if (notify) |
357 | notify_remote_via_irq(irq: queue->rx_irq); |
358 | } |
359 | |
360 | static int xennet_open(struct net_device *dev) |
361 | { |
362 | struct netfront_info *np = netdev_priv(dev); |
363 | unsigned int num_queues = dev->real_num_tx_queues; |
364 | unsigned int i = 0; |
365 | struct netfront_queue *queue = NULL; |
366 | |
367 | if (!np->queues || np->broken) |
368 | return -ENODEV; |
369 | |
370 | for (i = 0; i < num_queues; ++i) { |
371 | queue = &np->queues[i]; |
372 | napi_enable(n: &queue->napi); |
373 | |
374 | spin_lock_bh(lock: &queue->rx_lock); |
375 | if (netif_carrier_ok(dev)) { |
376 | xennet_alloc_rx_buffers(queue); |
377 | queue->rx.sring->rsp_event = queue->rx.rsp_cons + 1; |
378 | if (RING_HAS_UNCONSUMED_RESPONSES(&queue->rx)) |
379 | napi_schedule(n: &queue->napi); |
380 | } |
381 | spin_unlock_bh(lock: &queue->rx_lock); |
382 | } |
383 | |
384 | netif_tx_start_all_queues(dev); |
385 | |
386 | return 0; |
387 | } |
388 | |
389 | static bool xennet_tx_buf_gc(struct netfront_queue *queue) |
390 | { |
391 | RING_IDX cons, prod; |
392 | unsigned short id; |
393 | struct sk_buff *skb; |
394 | bool more_to_do; |
395 | bool work_done = false; |
396 | const struct device *dev = &queue->info->netdev->dev; |
397 | |
398 | BUG_ON(!netif_carrier_ok(queue->info->netdev)); |
399 | |
400 | do { |
401 | prod = queue->tx.sring->rsp_prod; |
402 | if (RING_RESPONSE_PROD_OVERFLOW(&queue->tx, prod)) { |
403 | dev_alert(dev, "Illegal number of responses %u\n" , |
404 | prod - queue->tx.rsp_cons); |
405 | goto err; |
406 | } |
407 | rmb(); /* Ensure we see responses up to 'rp'. */ |
408 | |
409 | for (cons = queue->tx.rsp_cons; cons != prod; cons++) { |
410 | struct xen_netif_tx_response txrsp; |
411 | |
412 | work_done = true; |
413 | |
414 | RING_COPY_RESPONSE(&queue->tx, cons, &txrsp); |
415 | if (txrsp.status == XEN_NETIF_RSP_NULL) |
416 | continue; |
417 | |
418 | id = txrsp.id; |
419 | if (id >= RING_SIZE(&queue->tx)) { |
420 | dev_alert(dev, |
421 | "Response has incorrect id (%u)\n" , |
422 | id); |
423 | goto err; |
424 | } |
425 | if (queue->tx_link[id] != TX_PENDING) { |
426 | dev_alert(dev, |
427 | "Response for inactive request\n" ); |
428 | goto err; |
429 | } |
430 | |
431 | queue->tx_link[id] = TX_LINK_NONE; |
432 | skb = queue->tx_skbs[id]; |
433 | queue->tx_skbs[id] = NULL; |
434 | if (unlikely(!gnttab_end_foreign_access_ref( |
435 | queue->grant_tx_ref[id]))) { |
436 | dev_alert(dev, |
437 | "Grant still in use by backend domain\n" ); |
438 | goto err; |
439 | } |
440 | gnttab_release_grant_reference( |
441 | private_head: &queue->gref_tx_head, release: queue->grant_tx_ref[id]); |
442 | queue->grant_tx_ref[id] = INVALID_GRANT_REF; |
443 | queue->grant_tx_page[id] = NULL; |
444 | add_id_to_list(head: &queue->tx_skb_freelist, list: queue->tx_link, id); |
445 | dev_kfree_skb_irq(skb); |
446 | } |
447 | |
448 | queue->tx.rsp_cons = prod; |
449 | |
450 | RING_FINAL_CHECK_FOR_RESPONSES(&queue->tx, more_to_do); |
451 | } while (more_to_do); |
452 | |
453 | xennet_maybe_wake_tx(queue); |
454 | |
455 | return work_done; |
456 | |
457 | err: |
458 | queue->info->broken = true; |
459 | dev_alert(dev, "Disabled for further use\n" ); |
460 | |
461 | return work_done; |
462 | } |
463 | |
464 | struct xennet_gnttab_make_txreq { |
465 | struct netfront_queue *queue; |
466 | struct sk_buff *skb; |
467 | struct page *page; |
468 | struct xen_netif_tx_request *tx; /* Last request on ring page */ |
469 | struct xen_netif_tx_request tx_local; /* Last request local copy*/ |
470 | unsigned int size; |
471 | }; |
472 | |
473 | static void xennet_tx_setup_grant(unsigned long gfn, unsigned int offset, |
474 | unsigned int len, void *data) |
475 | { |
476 | struct xennet_gnttab_make_txreq *info = data; |
477 | unsigned int id; |
478 | struct xen_netif_tx_request *tx; |
479 | grant_ref_t ref; |
480 | /* convenient aliases */ |
481 | struct page *page = info->page; |
482 | struct netfront_queue *queue = info->queue; |
483 | struct sk_buff *skb = info->skb; |
484 | |
485 | id = get_id_from_list(head: &queue->tx_skb_freelist, list: queue->tx_link); |
486 | tx = RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++); |
487 | ref = gnttab_claim_grant_reference(pprivate_head: &queue->gref_tx_head); |
488 | WARN_ON_ONCE(IS_ERR_VALUE((unsigned long)(int)ref)); |
489 | |
490 | gnttab_grant_foreign_access_ref(ref, domid: queue->info->xbdev->otherend_id, |
491 | frame: gfn, GNTMAP_readonly); |
492 | |
493 | queue->tx_skbs[id] = skb; |
494 | queue->grant_tx_page[id] = page; |
495 | queue->grant_tx_ref[id] = ref; |
496 | |
497 | info->tx_local.id = id; |
498 | info->tx_local.gref = ref; |
499 | info->tx_local.offset = offset; |
500 | info->tx_local.size = len; |
501 | info->tx_local.flags = 0; |
502 | |
503 | *tx = info->tx_local; |
504 | |
505 | /* |
506 | * Put the request in the pending queue, it will be set to be pending |
507 | * when the producer index is about to be raised. |
508 | */ |
509 | add_id_to_list(head: &queue->tx_pend_queue, list: queue->tx_link, id); |
510 | |
511 | info->tx = tx; |
512 | info->size += info->tx_local.size; |
513 | } |
514 | |
515 | static struct xen_netif_tx_request *xennet_make_first_txreq( |
516 | struct xennet_gnttab_make_txreq *info, |
517 | unsigned int offset, unsigned int len) |
518 | { |
519 | info->size = 0; |
520 | |
521 | gnttab_for_one_grant(page: info->page, offset, len, fn: xennet_tx_setup_grant, data: info); |
522 | |
523 | return info->tx; |
524 | } |
525 | |
526 | static void xennet_make_one_txreq(unsigned long gfn, unsigned int offset, |
527 | unsigned int len, void *data) |
528 | { |
529 | struct xennet_gnttab_make_txreq *info = data; |
530 | |
531 | info->tx->flags |= XEN_NETTXF_more_data; |
532 | skb_get(skb: info->skb); |
533 | xennet_tx_setup_grant(gfn, offset, len, data); |
534 | } |
535 | |
536 | static void xennet_make_txreqs( |
537 | struct xennet_gnttab_make_txreq *info, |
538 | struct page *page, |
539 | unsigned int offset, unsigned int len) |
540 | { |
541 | /* Skip unused frames from start of page */ |
542 | page += offset >> PAGE_SHIFT; |
543 | offset &= ~PAGE_MASK; |
544 | |
545 | while (len) { |
546 | info->page = page; |
547 | info->size = 0; |
548 | |
549 | gnttab_foreach_grant_in_range(page, offset, len, |
550 | fn: xennet_make_one_txreq, |
551 | data: info); |
552 | |
553 | page++; |
554 | offset = 0; |
555 | len -= info->size; |
556 | } |
557 | } |
558 | |
559 | /* |
560 | * Count how many ring slots are required to send this skb. Each frag |
561 | * might be a compound page. |
562 | */ |
563 | static int xennet_count_skb_slots(struct sk_buff *skb) |
564 | { |
565 | int i, frags = skb_shinfo(skb)->nr_frags; |
566 | int slots; |
567 | |
568 | slots = gnttab_count_grant(offset_in_page(skb->data), |
569 | len: skb_headlen(skb)); |
570 | |
571 | for (i = 0; i < frags; i++) { |
572 | skb_frag_t *frag = skb_shinfo(skb)->frags + i; |
573 | unsigned long size = skb_frag_size(frag); |
574 | unsigned long offset = skb_frag_off(frag); |
575 | |
576 | /* Skip unused frames from start of page */ |
577 | offset &= ~PAGE_MASK; |
578 | |
579 | slots += gnttab_count_grant(start: offset, len: size); |
580 | } |
581 | |
582 | return slots; |
583 | } |
584 | |
585 | static u16 xennet_select_queue(struct net_device *dev, struct sk_buff *skb, |
586 | struct net_device *sb_dev) |
587 | { |
588 | unsigned int num_queues = dev->real_num_tx_queues; |
589 | u32 hash; |
590 | u16 queue_idx; |
591 | |
592 | /* First, check if there is only one queue */ |
593 | if (num_queues == 1) { |
594 | queue_idx = 0; |
595 | } else { |
596 | hash = skb_get_hash(skb); |
597 | queue_idx = hash % num_queues; |
598 | } |
599 | |
600 | return queue_idx; |
601 | } |
602 | |
603 | static void xennet_mark_tx_pending(struct netfront_queue *queue) |
604 | { |
605 | unsigned int i; |
606 | |
607 | while ((i = get_id_from_list(head: &queue->tx_pend_queue, list: queue->tx_link)) != |
608 | TX_LINK_NONE) |
609 | queue->tx_link[i] = TX_PENDING; |
610 | } |
611 | |
612 | static int xennet_xdp_xmit_one(struct net_device *dev, |
613 | struct netfront_queue *queue, |
614 | struct xdp_frame *xdpf) |
615 | { |
616 | struct netfront_info *np = netdev_priv(dev); |
617 | struct netfront_stats *tx_stats = this_cpu_ptr(np->tx_stats); |
618 | struct xennet_gnttab_make_txreq info = { |
619 | .queue = queue, |
620 | .skb = NULL, |
621 | .page = virt_to_page(xdpf->data), |
622 | }; |
623 | int notify; |
624 | |
625 | xennet_make_first_txreq(info: &info, |
626 | offset_in_page(xdpf->data), |
627 | len: xdpf->len); |
628 | |
629 | xennet_mark_tx_pending(queue); |
630 | |
631 | RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->tx, notify); |
632 | if (notify) |
633 | notify_remote_via_irq(irq: queue->tx_irq); |
634 | |
635 | u64_stats_update_begin(syncp: &tx_stats->syncp); |
636 | tx_stats->bytes += xdpf->len; |
637 | tx_stats->packets++; |
638 | u64_stats_update_end(syncp: &tx_stats->syncp); |
639 | |
640 | xennet_tx_buf_gc(queue); |
641 | |
642 | return 0; |
643 | } |
644 | |
645 | static int xennet_xdp_xmit(struct net_device *dev, int n, |
646 | struct xdp_frame **frames, u32 flags) |
647 | { |
648 | unsigned int num_queues = dev->real_num_tx_queues; |
649 | struct netfront_info *np = netdev_priv(dev); |
650 | struct netfront_queue *queue = NULL; |
651 | unsigned long irq_flags; |
652 | int nxmit = 0; |
653 | int i; |
654 | |
655 | if (unlikely(np->broken)) |
656 | return -ENODEV; |
657 | if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK)) |
658 | return -EINVAL; |
659 | |
660 | queue = &np->queues[smp_processor_id() % num_queues]; |
661 | |
662 | spin_lock_irqsave(&queue->tx_lock, irq_flags); |
663 | for (i = 0; i < n; i++) { |
664 | struct xdp_frame *xdpf = frames[i]; |
665 | |
666 | if (!xdpf) |
667 | continue; |
668 | if (xennet_xdp_xmit_one(dev, queue, xdpf)) |
669 | break; |
670 | nxmit++; |
671 | } |
672 | spin_unlock_irqrestore(lock: &queue->tx_lock, flags: irq_flags); |
673 | |
674 | return nxmit; |
675 | } |
676 | |
677 | static struct sk_buff *bounce_skb(const struct sk_buff *skb) |
678 | { |
679 | unsigned int = skb_headroom(skb); |
680 | /* Align size to allocate full pages and avoid contiguous data leaks */ |
681 | unsigned int size = ALIGN(skb_end_offset(skb) + skb->data_len, |
682 | XEN_PAGE_SIZE); |
683 | struct sk_buff *n = alloc_skb(size, GFP_ATOMIC | __GFP_ZERO); |
684 | |
685 | if (!n) |
686 | return NULL; |
687 | |
688 | if (!IS_ALIGNED((uintptr_t)n->head, XEN_PAGE_SIZE)) { |
689 | WARN_ONCE(1, "misaligned skb allocated\n" ); |
690 | kfree_skb(skb: n); |
691 | return NULL; |
692 | } |
693 | |
694 | /* Set the data pointer */ |
695 | skb_reserve(skb: n, len: headerlen); |
696 | /* Set the tail pointer and length */ |
697 | skb_put(skb: n, len: skb->len); |
698 | |
699 | BUG_ON(skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len)); |
700 | |
701 | skb_copy_header(new: n, old: skb); |
702 | return n; |
703 | } |
704 | |
705 | #define MAX_XEN_SKB_FRAGS (65536 / XEN_PAGE_SIZE + 1) |
706 | |
707 | static netdev_tx_t xennet_start_xmit(struct sk_buff *skb, struct net_device *dev) |
708 | { |
709 | struct netfront_info *np = netdev_priv(dev); |
710 | struct netfront_stats *tx_stats = this_cpu_ptr(np->tx_stats); |
711 | struct xen_netif_tx_request *first_tx; |
712 | unsigned int i; |
713 | int notify; |
714 | int slots; |
715 | struct page *page; |
716 | unsigned int offset; |
717 | unsigned int len; |
718 | unsigned long flags; |
719 | struct netfront_queue *queue = NULL; |
720 | struct xennet_gnttab_make_txreq info = { }; |
721 | unsigned int num_queues = dev->real_num_tx_queues; |
722 | u16 queue_index; |
723 | struct sk_buff *nskb; |
724 | |
725 | /* Drop the packet if no queues are set up */ |
726 | if (num_queues < 1) |
727 | goto drop; |
728 | if (unlikely(np->broken)) |
729 | goto drop; |
730 | /* Determine which queue to transmit this SKB on */ |
731 | queue_index = skb_get_queue_mapping(skb); |
732 | queue = &np->queues[queue_index]; |
733 | |
734 | /* If skb->len is too big for wire format, drop skb and alert |
735 | * user about misconfiguration. |
736 | */ |
737 | if (unlikely(skb->len > XEN_NETIF_MAX_TX_SIZE)) { |
738 | net_alert_ratelimited( |
739 | "xennet: skb->len = %u, too big for wire format\n" , |
740 | skb->len); |
741 | goto drop; |
742 | } |
743 | |
744 | slots = xennet_count_skb_slots(skb); |
745 | if (unlikely(slots > MAX_XEN_SKB_FRAGS + 1)) { |
746 | net_dbg_ratelimited("xennet: skb rides the rocket: %d slots, %d bytes\n" , |
747 | slots, skb->len); |
748 | if (skb_linearize(skb)) |
749 | goto drop; |
750 | } |
751 | |
752 | page = virt_to_page(skb->data); |
753 | offset = offset_in_page(skb->data); |
754 | |
755 | /* The first req should be at least ETH_HLEN size or the packet will be |
756 | * dropped by netback. |
757 | * |
758 | * If the backend is not trusted bounce all data to zeroed pages to |
759 | * avoid exposing contiguous data on the granted page not belonging to |
760 | * the skb. |
761 | */ |
762 | if (np->bounce || unlikely(PAGE_SIZE - offset < ETH_HLEN)) { |
763 | nskb = bounce_skb(skb); |
764 | if (!nskb) |
765 | goto drop; |
766 | dev_consume_skb_any(skb); |
767 | skb = nskb; |
768 | page = virt_to_page(skb->data); |
769 | offset = offset_in_page(skb->data); |
770 | } |
771 | |
772 | len = skb_headlen(skb); |
773 | |
774 | spin_lock_irqsave(&queue->tx_lock, flags); |
775 | |
776 | if (unlikely(!netif_carrier_ok(dev) || |
777 | (slots > 1 && !xennet_can_sg(dev)) || |
778 | netif_needs_gso(skb, netif_skb_features(skb)))) { |
779 | spin_unlock_irqrestore(lock: &queue->tx_lock, flags); |
780 | goto drop; |
781 | } |
782 | |
783 | /* First request for the linear area. */ |
784 | info.queue = queue; |
785 | info.skb = skb; |
786 | info.page = page; |
787 | first_tx = xennet_make_first_txreq(info: &info, offset, len); |
788 | offset += info.tx_local.size; |
789 | if (offset == PAGE_SIZE) { |
790 | page++; |
791 | offset = 0; |
792 | } |
793 | len -= info.tx_local.size; |
794 | |
795 | if (skb->ip_summed == CHECKSUM_PARTIAL) |
796 | /* local packet? */ |
797 | first_tx->flags |= XEN_NETTXF_csum_blank | |
798 | XEN_NETTXF_data_validated; |
799 | else if (skb->ip_summed == CHECKSUM_UNNECESSARY) |
800 | /* remote but checksummed. */ |
801 | first_tx->flags |= XEN_NETTXF_data_validated; |
802 | |
803 | /* Optional extra info after the first request. */ |
804 | if (skb_shinfo(skb)->gso_size) { |
805 | struct xen_netif_extra_info *gso; |
806 | |
807 | gso = (struct xen_netif_extra_info *) |
808 | RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++); |
809 | |
810 | first_tx->flags |= XEN_NETTXF_extra_info; |
811 | |
812 | gso->u.gso.size = skb_shinfo(skb)->gso_size; |
813 | gso->u.gso.type = (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) ? |
814 | XEN_NETIF_GSO_TYPE_TCPV6 : |
815 | XEN_NETIF_GSO_TYPE_TCPV4; |
816 | gso->u.gso.pad = 0; |
817 | gso->u.gso.features = 0; |
818 | |
819 | gso->type = XEN_NETIF_EXTRA_TYPE_GSO; |
820 | gso->flags = 0; |
821 | } |
822 | |
823 | /* Requests for the rest of the linear area. */ |
824 | xennet_make_txreqs(info: &info, page, offset, len); |
825 | |
826 | /* Requests for all the frags. */ |
827 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
828 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; |
829 | xennet_make_txreqs(info: &info, page: skb_frag_page(frag), |
830 | offset: skb_frag_off(frag), |
831 | len: skb_frag_size(frag)); |
832 | } |
833 | |
834 | /* First request has the packet length. */ |
835 | first_tx->size = skb->len; |
836 | |
837 | /* timestamp packet in software */ |
838 | skb_tx_timestamp(skb); |
839 | |
840 | xennet_mark_tx_pending(queue); |
841 | |
842 | RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->tx, notify); |
843 | if (notify) |
844 | notify_remote_via_irq(irq: queue->tx_irq); |
845 | |
846 | u64_stats_update_begin(syncp: &tx_stats->syncp); |
847 | tx_stats->bytes += skb->len; |
848 | tx_stats->packets++; |
849 | u64_stats_update_end(syncp: &tx_stats->syncp); |
850 | |
851 | /* Note: It is not safe to access skb after xennet_tx_buf_gc()! */ |
852 | xennet_tx_buf_gc(queue); |
853 | |
854 | if (!netfront_tx_slot_available(queue)) |
855 | netif_tx_stop_queue(dev_queue: netdev_get_tx_queue(dev, index: queue->id)); |
856 | |
857 | spin_unlock_irqrestore(lock: &queue->tx_lock, flags); |
858 | |
859 | return NETDEV_TX_OK; |
860 | |
861 | drop: |
862 | dev->stats.tx_dropped++; |
863 | dev_kfree_skb_any(skb); |
864 | return NETDEV_TX_OK; |
865 | } |
866 | |
867 | static int xennet_close(struct net_device *dev) |
868 | { |
869 | struct netfront_info *np = netdev_priv(dev); |
870 | unsigned int num_queues = dev->real_num_tx_queues; |
871 | unsigned int i; |
872 | struct netfront_queue *queue; |
873 | netif_tx_stop_all_queues(dev: np->netdev); |
874 | for (i = 0; i < num_queues; ++i) { |
875 | queue = &np->queues[i]; |
876 | napi_disable(n: &queue->napi); |
877 | } |
878 | return 0; |
879 | } |
880 | |
881 | static void xennet_destroy_queues(struct netfront_info *info) |
882 | { |
883 | unsigned int i; |
884 | |
885 | for (i = 0; i < info->netdev->real_num_tx_queues; i++) { |
886 | struct netfront_queue *queue = &info->queues[i]; |
887 | |
888 | if (netif_running(dev: info->netdev)) |
889 | napi_disable(n: &queue->napi); |
890 | netif_napi_del(napi: &queue->napi); |
891 | } |
892 | |
893 | kfree(objp: info->queues); |
894 | info->queues = NULL; |
895 | } |
896 | |
897 | static void xennet_uninit(struct net_device *dev) |
898 | { |
899 | struct netfront_info *np = netdev_priv(dev); |
900 | xennet_destroy_queues(info: np); |
901 | } |
902 | |
903 | static void xennet_set_rx_rsp_cons(struct netfront_queue *queue, RING_IDX val) |
904 | { |
905 | unsigned long flags; |
906 | |
907 | spin_lock_irqsave(&queue->rx_cons_lock, flags); |
908 | queue->rx.rsp_cons = val; |
909 | queue->rx_rsp_unconsumed = XEN_RING_NR_UNCONSUMED_RESPONSES(&queue->rx); |
910 | spin_unlock_irqrestore(lock: &queue->rx_cons_lock, flags); |
911 | } |
912 | |
913 | static void xennet_move_rx_slot(struct netfront_queue *queue, struct sk_buff *skb, |
914 | grant_ref_t ref) |
915 | { |
916 | int new = xennet_rxidx(idx: queue->rx.req_prod_pvt); |
917 | |
918 | BUG_ON(queue->rx_skbs[new]); |
919 | queue->rx_skbs[new] = skb; |
920 | queue->grant_rx_ref[new] = ref; |
921 | RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->id = new; |
922 | RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->gref = ref; |
923 | queue->rx.req_prod_pvt++; |
924 | } |
925 | |
926 | static int (struct netfront_queue *queue, |
927 | struct xen_netif_extra_info *, |
928 | RING_IDX rp) |
929 | |
930 | { |
931 | struct xen_netif_extra_info ; |
932 | struct device *dev = &queue->info->netdev->dev; |
933 | RING_IDX cons = queue->rx.rsp_cons; |
934 | int err = 0; |
935 | |
936 | do { |
937 | struct sk_buff *skb; |
938 | grant_ref_t ref; |
939 | |
940 | if (unlikely(cons + 1 == rp)) { |
941 | if (net_ratelimit()) |
942 | dev_warn(dev, "Missing extra info\n" ); |
943 | err = -EBADR; |
944 | break; |
945 | } |
946 | |
947 | RING_COPY_RESPONSE(&queue->rx, ++cons, &extra); |
948 | |
949 | if (unlikely(!extra.type || |
950 | extra.type >= XEN_NETIF_EXTRA_TYPE_MAX)) { |
951 | if (net_ratelimit()) |
952 | dev_warn(dev, "Invalid extra type: %d\n" , |
953 | extra.type); |
954 | err = -EINVAL; |
955 | } else { |
956 | extras[extra.type - 1] = extra; |
957 | } |
958 | |
959 | skb = xennet_get_rx_skb(queue, ri: cons); |
960 | ref = xennet_get_rx_ref(queue, ri: cons); |
961 | xennet_move_rx_slot(queue, skb, ref); |
962 | } while (extra.flags & XEN_NETIF_EXTRA_FLAG_MORE); |
963 | |
964 | xennet_set_rx_rsp_cons(queue, val: cons); |
965 | return err; |
966 | } |
967 | |
968 | static u32 xennet_run_xdp(struct netfront_queue *queue, struct page *pdata, |
969 | struct xen_netif_rx_response *rx, struct bpf_prog *prog, |
970 | struct xdp_buff *xdp, bool *need_xdp_flush) |
971 | { |
972 | struct xdp_frame *xdpf; |
973 | u32 len = rx->status; |
974 | u32 act; |
975 | int err; |
976 | |
977 | xdp_init_buff(xdp, XEN_PAGE_SIZE - XDP_PACKET_HEADROOM, |
978 | rxq: &queue->xdp_rxq); |
979 | xdp_prepare_buff(xdp, page_address(pdata), XDP_PACKET_HEADROOM, |
980 | data_len: len, meta_valid: false); |
981 | |
982 | act = bpf_prog_run_xdp(prog, xdp); |
983 | switch (act) { |
984 | case XDP_TX: |
985 | get_page(page: pdata); |
986 | xdpf = xdp_convert_buff_to_frame(xdp); |
987 | err = xennet_xdp_xmit(dev: queue->info->netdev, n: 1, frames: &xdpf, flags: 0); |
988 | if (unlikely(!err)) |
989 | xdp_return_frame_rx_napi(xdpf); |
990 | else if (unlikely(err < 0)) |
991 | trace_xdp_exception(dev: queue->info->netdev, xdp: prog, act); |
992 | break; |
993 | case XDP_REDIRECT: |
994 | get_page(page: pdata); |
995 | err = xdp_do_redirect(dev: queue->info->netdev, xdp, prog); |
996 | *need_xdp_flush = true; |
997 | if (unlikely(err)) |
998 | trace_xdp_exception(dev: queue->info->netdev, xdp: prog, act); |
999 | break; |
1000 | case XDP_PASS: |
1001 | case XDP_DROP: |
1002 | break; |
1003 | |
1004 | case XDP_ABORTED: |
1005 | trace_xdp_exception(dev: queue->info->netdev, xdp: prog, act); |
1006 | break; |
1007 | |
1008 | default: |
1009 | bpf_warn_invalid_xdp_action(dev: queue->info->netdev, prog, act); |
1010 | } |
1011 | |
1012 | return act; |
1013 | } |
1014 | |
1015 | static int xennet_get_responses(struct netfront_queue *queue, |
1016 | struct netfront_rx_info *rinfo, RING_IDX rp, |
1017 | struct sk_buff_head *list, |
1018 | bool *need_xdp_flush) |
1019 | { |
1020 | struct xen_netif_rx_response *rx = &rinfo->rx, rx_local; |
1021 | int max = XEN_NETIF_NR_SLOTS_MIN + (rx->status <= RX_COPY_THRESHOLD); |
1022 | RING_IDX cons = queue->rx.rsp_cons; |
1023 | struct sk_buff *skb = xennet_get_rx_skb(queue, ri: cons); |
1024 | struct xen_netif_extra_info * = rinfo->extras; |
1025 | grant_ref_t ref = xennet_get_rx_ref(queue, ri: cons); |
1026 | struct device *dev = &queue->info->netdev->dev; |
1027 | struct bpf_prog *xdp_prog; |
1028 | struct xdp_buff xdp; |
1029 | int slots = 1; |
1030 | int err = 0; |
1031 | u32 verdict; |
1032 | |
1033 | if (rx->flags & XEN_NETRXF_extra_info) { |
1034 | err = xennet_get_extras(queue, extras, rp); |
1035 | if (!err) { |
1036 | if (extras[XEN_NETIF_EXTRA_TYPE_XDP - 1].type) { |
1037 | struct xen_netif_extra_info *xdp; |
1038 | |
1039 | xdp = &extras[XEN_NETIF_EXTRA_TYPE_XDP - 1]; |
1040 | rx->offset = xdp->u.xdp.headroom; |
1041 | } |
1042 | } |
1043 | cons = queue->rx.rsp_cons; |
1044 | } |
1045 | |
1046 | for (;;) { |
1047 | /* |
1048 | * This definitely indicates a bug, either in this driver or in |
1049 | * the backend driver. In future this should flag the bad |
1050 | * situation to the system controller to reboot the backend. |
1051 | */ |
1052 | if (ref == INVALID_GRANT_REF) { |
1053 | if (net_ratelimit()) |
1054 | dev_warn(dev, "Bad rx response id %d.\n" , |
1055 | rx->id); |
1056 | err = -EINVAL; |
1057 | goto next; |
1058 | } |
1059 | |
1060 | if (unlikely(rx->status < 0 || |
1061 | rx->offset + rx->status > XEN_PAGE_SIZE)) { |
1062 | if (net_ratelimit()) |
1063 | dev_warn(dev, "rx->offset: %u, size: %d\n" , |
1064 | rx->offset, rx->status); |
1065 | xennet_move_rx_slot(queue, skb, ref); |
1066 | err = -EINVAL; |
1067 | goto next; |
1068 | } |
1069 | |
1070 | if (!gnttab_end_foreign_access_ref(ref)) { |
1071 | dev_alert(dev, |
1072 | "Grant still in use by backend domain\n" ); |
1073 | queue->info->broken = true; |
1074 | dev_alert(dev, "Disabled for further use\n" ); |
1075 | return -EINVAL; |
1076 | } |
1077 | |
1078 | gnttab_release_grant_reference(private_head: &queue->gref_rx_head, release: ref); |
1079 | |
1080 | rcu_read_lock(); |
1081 | xdp_prog = rcu_dereference(queue->xdp_prog); |
1082 | if (xdp_prog) { |
1083 | if (!(rx->flags & XEN_NETRXF_more_data)) { |
1084 | /* currently only a single page contains data */ |
1085 | verdict = xennet_run_xdp(queue, |
1086 | pdata: skb_frag_page(frag: &skb_shinfo(skb)->frags[0]), |
1087 | rx, prog: xdp_prog, xdp: &xdp, need_xdp_flush); |
1088 | if (verdict != XDP_PASS) |
1089 | err = -EINVAL; |
1090 | } else { |
1091 | /* drop the frame */ |
1092 | err = -EINVAL; |
1093 | } |
1094 | } |
1095 | rcu_read_unlock(); |
1096 | |
1097 | __skb_queue_tail(list, newsk: skb); |
1098 | |
1099 | next: |
1100 | if (!(rx->flags & XEN_NETRXF_more_data)) |
1101 | break; |
1102 | |
1103 | if (cons + slots == rp) { |
1104 | if (net_ratelimit()) |
1105 | dev_warn(dev, "Need more slots\n" ); |
1106 | err = -ENOENT; |
1107 | break; |
1108 | } |
1109 | |
1110 | RING_COPY_RESPONSE(&queue->rx, cons + slots, &rx_local); |
1111 | rx = &rx_local; |
1112 | skb = xennet_get_rx_skb(queue, ri: cons + slots); |
1113 | ref = xennet_get_rx_ref(queue, ri: cons + slots); |
1114 | slots++; |
1115 | } |
1116 | |
1117 | if (unlikely(slots > max)) { |
1118 | if (net_ratelimit()) |
1119 | dev_warn(dev, "Too many slots\n" ); |
1120 | err = -E2BIG; |
1121 | } |
1122 | |
1123 | if (unlikely(err)) |
1124 | xennet_set_rx_rsp_cons(queue, val: cons + slots); |
1125 | |
1126 | return err; |
1127 | } |
1128 | |
1129 | static int xennet_set_skb_gso(struct sk_buff *skb, |
1130 | struct xen_netif_extra_info *gso) |
1131 | { |
1132 | if (!gso->u.gso.size) { |
1133 | if (net_ratelimit()) |
1134 | pr_warn("GSO size must not be zero\n" ); |
1135 | return -EINVAL; |
1136 | } |
1137 | |
1138 | if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4 && |
1139 | gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV6) { |
1140 | if (net_ratelimit()) |
1141 | pr_warn("Bad GSO type %d\n" , gso->u.gso.type); |
1142 | return -EINVAL; |
1143 | } |
1144 | |
1145 | skb_shinfo(skb)->gso_size = gso->u.gso.size; |
1146 | skb_shinfo(skb)->gso_type = |
1147 | (gso->u.gso.type == XEN_NETIF_GSO_TYPE_TCPV4) ? |
1148 | SKB_GSO_TCPV4 : |
1149 | SKB_GSO_TCPV6; |
1150 | |
1151 | /* Header must be checked, and gso_segs computed. */ |
1152 | skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; |
1153 | skb_shinfo(skb)->gso_segs = 0; |
1154 | |
1155 | return 0; |
1156 | } |
1157 | |
1158 | static int xennet_fill_frags(struct netfront_queue *queue, |
1159 | struct sk_buff *skb, |
1160 | struct sk_buff_head *list) |
1161 | { |
1162 | RING_IDX cons = queue->rx.rsp_cons; |
1163 | struct sk_buff *nskb; |
1164 | |
1165 | while ((nskb = __skb_dequeue(list))) { |
1166 | struct xen_netif_rx_response rx; |
1167 | skb_frag_t *nfrag = &skb_shinfo(nskb)->frags[0]; |
1168 | |
1169 | RING_COPY_RESPONSE(&queue->rx, ++cons, &rx); |
1170 | |
1171 | if (skb_shinfo(skb)->nr_frags == MAX_SKB_FRAGS) { |
1172 | unsigned int pull_to = NETFRONT_SKB_CB(skb)->pull_to; |
1173 | |
1174 | BUG_ON(pull_to < skb_headlen(skb)); |
1175 | __pskb_pull_tail(skb, delta: pull_to - skb_headlen(skb)); |
1176 | } |
1177 | if (unlikely(skb_shinfo(skb)->nr_frags >= MAX_SKB_FRAGS)) { |
1178 | xennet_set_rx_rsp_cons(queue, |
1179 | val: ++cons + skb_queue_len(list_: list)); |
1180 | kfree_skb(skb: nskb); |
1181 | return -ENOENT; |
1182 | } |
1183 | |
1184 | skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, |
1185 | page: skb_frag_page(frag: nfrag), |
1186 | off: rx.offset, size: rx.status, PAGE_SIZE); |
1187 | |
1188 | skb_shinfo(nskb)->nr_frags = 0; |
1189 | kfree_skb(skb: nskb); |
1190 | } |
1191 | |
1192 | xennet_set_rx_rsp_cons(queue, val: cons); |
1193 | |
1194 | return 0; |
1195 | } |
1196 | |
1197 | static int checksum_setup(struct net_device *dev, struct sk_buff *skb) |
1198 | { |
1199 | bool recalculate_partial_csum = false; |
1200 | |
1201 | /* |
1202 | * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy |
1203 | * peers can fail to set NETRXF_csum_blank when sending a GSO |
1204 | * frame. In this case force the SKB to CHECKSUM_PARTIAL and |
1205 | * recalculate the partial checksum. |
1206 | */ |
1207 | if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) { |
1208 | struct netfront_info *np = netdev_priv(dev); |
1209 | atomic_inc(v: &np->rx_gso_checksum_fixup); |
1210 | skb->ip_summed = CHECKSUM_PARTIAL; |
1211 | recalculate_partial_csum = true; |
1212 | } |
1213 | |
1214 | /* A non-CHECKSUM_PARTIAL SKB does not require setup. */ |
1215 | if (skb->ip_summed != CHECKSUM_PARTIAL) |
1216 | return 0; |
1217 | |
1218 | return skb_checksum_setup(skb, recalculate: recalculate_partial_csum); |
1219 | } |
1220 | |
1221 | static int handle_incoming_queue(struct netfront_queue *queue, |
1222 | struct sk_buff_head *rxq) |
1223 | { |
1224 | struct netfront_stats *rx_stats = this_cpu_ptr(queue->info->rx_stats); |
1225 | int packets_dropped = 0; |
1226 | struct sk_buff *skb; |
1227 | |
1228 | while ((skb = __skb_dequeue(list: rxq)) != NULL) { |
1229 | int pull_to = NETFRONT_SKB_CB(skb)->pull_to; |
1230 | |
1231 | if (pull_to > skb_headlen(skb)) |
1232 | __pskb_pull_tail(skb, delta: pull_to - skb_headlen(skb)); |
1233 | |
1234 | /* Ethernet work: Delayed to here as it peeks the header. */ |
1235 | skb->protocol = eth_type_trans(skb, dev: queue->info->netdev); |
1236 | skb_reset_network_header(skb); |
1237 | |
1238 | if (checksum_setup(dev: queue->info->netdev, skb)) { |
1239 | kfree_skb(skb); |
1240 | packets_dropped++; |
1241 | queue->info->netdev->stats.rx_errors++; |
1242 | continue; |
1243 | } |
1244 | |
1245 | u64_stats_update_begin(syncp: &rx_stats->syncp); |
1246 | rx_stats->packets++; |
1247 | rx_stats->bytes += skb->len; |
1248 | u64_stats_update_end(syncp: &rx_stats->syncp); |
1249 | |
1250 | /* Pass it up. */ |
1251 | napi_gro_receive(napi: &queue->napi, skb); |
1252 | } |
1253 | |
1254 | return packets_dropped; |
1255 | } |
1256 | |
1257 | static int xennet_poll(struct napi_struct *napi, int budget) |
1258 | { |
1259 | struct netfront_queue *queue = container_of(napi, struct netfront_queue, napi); |
1260 | struct net_device *dev = queue->info->netdev; |
1261 | struct sk_buff *skb; |
1262 | struct netfront_rx_info rinfo; |
1263 | struct xen_netif_rx_response *rx = &rinfo.rx; |
1264 | struct xen_netif_extra_info * = rinfo.extras; |
1265 | RING_IDX i, rp; |
1266 | int work_done; |
1267 | struct sk_buff_head rxq; |
1268 | struct sk_buff_head errq; |
1269 | struct sk_buff_head tmpq; |
1270 | int err; |
1271 | bool need_xdp_flush = false; |
1272 | |
1273 | spin_lock(lock: &queue->rx_lock); |
1274 | |
1275 | skb_queue_head_init(list: &rxq); |
1276 | skb_queue_head_init(list: &errq); |
1277 | skb_queue_head_init(list: &tmpq); |
1278 | |
1279 | rp = queue->rx.sring->rsp_prod; |
1280 | if (RING_RESPONSE_PROD_OVERFLOW(&queue->rx, rp)) { |
1281 | dev_alert(&dev->dev, "Illegal number of responses %u\n" , |
1282 | rp - queue->rx.rsp_cons); |
1283 | queue->info->broken = true; |
1284 | spin_unlock(lock: &queue->rx_lock); |
1285 | return 0; |
1286 | } |
1287 | rmb(); /* Ensure we see queued responses up to 'rp'. */ |
1288 | |
1289 | i = queue->rx.rsp_cons; |
1290 | work_done = 0; |
1291 | while ((i != rp) && (work_done < budget)) { |
1292 | RING_COPY_RESPONSE(&queue->rx, i, rx); |
1293 | memset(extras, 0, sizeof(rinfo.extras)); |
1294 | |
1295 | err = xennet_get_responses(queue, rinfo: &rinfo, rp, list: &tmpq, |
1296 | need_xdp_flush: &need_xdp_flush); |
1297 | |
1298 | if (unlikely(err)) { |
1299 | if (queue->info->broken) { |
1300 | spin_unlock(lock: &queue->rx_lock); |
1301 | return 0; |
1302 | } |
1303 | err: |
1304 | while ((skb = __skb_dequeue(list: &tmpq))) |
1305 | __skb_queue_tail(list: &errq, newsk: skb); |
1306 | dev->stats.rx_errors++; |
1307 | i = queue->rx.rsp_cons; |
1308 | continue; |
1309 | } |
1310 | |
1311 | skb = __skb_dequeue(list: &tmpq); |
1312 | |
1313 | if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) { |
1314 | struct xen_netif_extra_info *gso; |
1315 | gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1]; |
1316 | |
1317 | if (unlikely(xennet_set_skb_gso(skb, gso))) { |
1318 | __skb_queue_head(list: &tmpq, newsk: skb); |
1319 | xennet_set_rx_rsp_cons(queue, |
1320 | val: queue->rx.rsp_cons + |
1321 | skb_queue_len(list_: &tmpq)); |
1322 | goto err; |
1323 | } |
1324 | } |
1325 | |
1326 | NETFRONT_SKB_CB(skb)->pull_to = rx->status; |
1327 | if (NETFRONT_SKB_CB(skb)->pull_to > RX_COPY_THRESHOLD) |
1328 | NETFRONT_SKB_CB(skb)->pull_to = RX_COPY_THRESHOLD; |
1329 | |
1330 | skb_frag_off_set(frag: &skb_shinfo(skb)->frags[0], offset: rx->offset); |
1331 | skb_frag_size_set(frag: &skb_shinfo(skb)->frags[0], size: rx->status); |
1332 | skb->data_len = rx->status; |
1333 | skb->len += rx->status; |
1334 | |
1335 | if (unlikely(xennet_fill_frags(queue, skb, &tmpq))) |
1336 | goto err; |
1337 | |
1338 | if (rx->flags & XEN_NETRXF_csum_blank) |
1339 | skb->ip_summed = CHECKSUM_PARTIAL; |
1340 | else if (rx->flags & XEN_NETRXF_data_validated) |
1341 | skb->ip_summed = CHECKSUM_UNNECESSARY; |
1342 | |
1343 | __skb_queue_tail(list: &rxq, newsk: skb); |
1344 | |
1345 | i = queue->rx.rsp_cons + 1; |
1346 | xennet_set_rx_rsp_cons(queue, val: i); |
1347 | work_done++; |
1348 | } |
1349 | if (need_xdp_flush) |
1350 | xdp_do_flush(); |
1351 | |
1352 | __skb_queue_purge(list: &errq); |
1353 | |
1354 | work_done -= handle_incoming_queue(queue, rxq: &rxq); |
1355 | |
1356 | xennet_alloc_rx_buffers(queue); |
1357 | |
1358 | if (work_done < budget) { |
1359 | int more_to_do = 0; |
1360 | |
1361 | napi_complete_done(n: napi, work_done); |
1362 | |
1363 | RING_FINAL_CHECK_FOR_RESPONSES(&queue->rx, more_to_do); |
1364 | if (more_to_do) |
1365 | napi_schedule(n: napi); |
1366 | } |
1367 | |
1368 | spin_unlock(lock: &queue->rx_lock); |
1369 | |
1370 | return work_done; |
1371 | } |
1372 | |
1373 | static int xennet_change_mtu(struct net_device *dev, int mtu) |
1374 | { |
1375 | int max = xennet_can_sg(dev) ? XEN_NETIF_MAX_TX_SIZE : ETH_DATA_LEN; |
1376 | |
1377 | if (mtu > max) |
1378 | return -EINVAL; |
1379 | dev->mtu = mtu; |
1380 | return 0; |
1381 | } |
1382 | |
1383 | static void xennet_get_stats64(struct net_device *dev, |
1384 | struct rtnl_link_stats64 *tot) |
1385 | { |
1386 | struct netfront_info *np = netdev_priv(dev); |
1387 | int cpu; |
1388 | |
1389 | for_each_possible_cpu(cpu) { |
1390 | struct netfront_stats *rx_stats = per_cpu_ptr(np->rx_stats, cpu); |
1391 | struct netfront_stats *tx_stats = per_cpu_ptr(np->tx_stats, cpu); |
1392 | u64 rx_packets, rx_bytes, tx_packets, tx_bytes; |
1393 | unsigned int start; |
1394 | |
1395 | do { |
1396 | start = u64_stats_fetch_begin(syncp: &tx_stats->syncp); |
1397 | tx_packets = tx_stats->packets; |
1398 | tx_bytes = tx_stats->bytes; |
1399 | } while (u64_stats_fetch_retry(syncp: &tx_stats->syncp, start)); |
1400 | |
1401 | do { |
1402 | start = u64_stats_fetch_begin(syncp: &rx_stats->syncp); |
1403 | rx_packets = rx_stats->packets; |
1404 | rx_bytes = rx_stats->bytes; |
1405 | } while (u64_stats_fetch_retry(syncp: &rx_stats->syncp, start)); |
1406 | |
1407 | tot->rx_packets += rx_packets; |
1408 | tot->tx_packets += tx_packets; |
1409 | tot->rx_bytes += rx_bytes; |
1410 | tot->tx_bytes += tx_bytes; |
1411 | } |
1412 | |
1413 | tot->rx_errors = dev->stats.rx_errors; |
1414 | tot->tx_dropped = dev->stats.tx_dropped; |
1415 | } |
1416 | |
1417 | static void xennet_release_tx_bufs(struct netfront_queue *queue) |
1418 | { |
1419 | struct sk_buff *skb; |
1420 | int i; |
1421 | |
1422 | for (i = 0; i < NET_TX_RING_SIZE; i++) { |
1423 | /* Skip over entries which are actually freelist references */ |
1424 | if (!queue->tx_skbs[i]) |
1425 | continue; |
1426 | |
1427 | skb = queue->tx_skbs[i]; |
1428 | queue->tx_skbs[i] = NULL; |
1429 | get_page(page: queue->grant_tx_page[i]); |
1430 | gnttab_end_foreign_access(ref: queue->grant_tx_ref[i], |
1431 | page: queue->grant_tx_page[i]); |
1432 | queue->grant_tx_page[i] = NULL; |
1433 | queue->grant_tx_ref[i] = INVALID_GRANT_REF; |
1434 | add_id_to_list(head: &queue->tx_skb_freelist, list: queue->tx_link, id: i); |
1435 | dev_kfree_skb_irq(skb); |
1436 | } |
1437 | } |
1438 | |
1439 | static void xennet_release_rx_bufs(struct netfront_queue *queue) |
1440 | { |
1441 | int id, ref; |
1442 | |
1443 | spin_lock_bh(lock: &queue->rx_lock); |
1444 | |
1445 | for (id = 0; id < NET_RX_RING_SIZE; id++) { |
1446 | struct sk_buff *skb; |
1447 | struct page *page; |
1448 | |
1449 | skb = queue->rx_skbs[id]; |
1450 | if (!skb) |
1451 | continue; |
1452 | |
1453 | ref = queue->grant_rx_ref[id]; |
1454 | if (ref == INVALID_GRANT_REF) |
1455 | continue; |
1456 | |
1457 | page = skb_frag_page(frag: &skb_shinfo(skb)->frags[0]); |
1458 | |
1459 | /* gnttab_end_foreign_access() needs a page ref until |
1460 | * foreign access is ended (which may be deferred). |
1461 | */ |
1462 | get_page(page); |
1463 | gnttab_end_foreign_access(ref, page); |
1464 | queue->grant_rx_ref[id] = INVALID_GRANT_REF; |
1465 | |
1466 | kfree_skb(skb); |
1467 | } |
1468 | |
1469 | spin_unlock_bh(lock: &queue->rx_lock); |
1470 | } |
1471 | |
1472 | static netdev_features_t xennet_fix_features(struct net_device *dev, |
1473 | netdev_features_t features) |
1474 | { |
1475 | struct netfront_info *np = netdev_priv(dev); |
1476 | |
1477 | if (features & NETIF_F_SG && |
1478 | !xenbus_read_unsigned(dir: np->xbdev->otherend, node: "feature-sg" , default_val: 0)) |
1479 | features &= ~NETIF_F_SG; |
1480 | |
1481 | if (features & NETIF_F_IPV6_CSUM && |
1482 | !xenbus_read_unsigned(dir: np->xbdev->otherend, |
1483 | node: "feature-ipv6-csum-offload" , default_val: 0)) |
1484 | features &= ~NETIF_F_IPV6_CSUM; |
1485 | |
1486 | if (features & NETIF_F_TSO && |
1487 | !xenbus_read_unsigned(dir: np->xbdev->otherend, node: "feature-gso-tcpv4" , default_val: 0)) |
1488 | features &= ~NETIF_F_TSO; |
1489 | |
1490 | if (features & NETIF_F_TSO6 && |
1491 | !xenbus_read_unsigned(dir: np->xbdev->otherend, node: "feature-gso-tcpv6" , default_val: 0)) |
1492 | features &= ~NETIF_F_TSO6; |
1493 | |
1494 | return features; |
1495 | } |
1496 | |
1497 | static int xennet_set_features(struct net_device *dev, |
1498 | netdev_features_t features) |
1499 | { |
1500 | if (!(features & NETIF_F_SG) && dev->mtu > ETH_DATA_LEN) { |
1501 | netdev_info(dev, format: "Reducing MTU because no SG offload" ); |
1502 | dev->mtu = ETH_DATA_LEN; |
1503 | } |
1504 | |
1505 | return 0; |
1506 | } |
1507 | |
1508 | static bool xennet_handle_tx(struct netfront_queue *queue, unsigned int *eoi) |
1509 | { |
1510 | unsigned long flags; |
1511 | |
1512 | if (unlikely(queue->info->broken)) |
1513 | return false; |
1514 | |
1515 | spin_lock_irqsave(&queue->tx_lock, flags); |
1516 | if (xennet_tx_buf_gc(queue)) |
1517 | *eoi = 0; |
1518 | spin_unlock_irqrestore(lock: &queue->tx_lock, flags); |
1519 | |
1520 | return true; |
1521 | } |
1522 | |
1523 | static irqreturn_t xennet_tx_interrupt(int irq, void *dev_id) |
1524 | { |
1525 | unsigned int eoiflag = XEN_EOI_FLAG_SPURIOUS; |
1526 | |
1527 | if (likely(xennet_handle_tx(dev_id, &eoiflag))) |
1528 | xen_irq_lateeoi(irq, eoi_flags: eoiflag); |
1529 | |
1530 | return IRQ_HANDLED; |
1531 | } |
1532 | |
1533 | static bool xennet_handle_rx(struct netfront_queue *queue, unsigned int *eoi) |
1534 | { |
1535 | unsigned int work_queued; |
1536 | unsigned long flags; |
1537 | |
1538 | if (unlikely(queue->info->broken)) |
1539 | return false; |
1540 | |
1541 | spin_lock_irqsave(&queue->rx_cons_lock, flags); |
1542 | work_queued = XEN_RING_NR_UNCONSUMED_RESPONSES(&queue->rx); |
1543 | if (work_queued > queue->rx_rsp_unconsumed) { |
1544 | queue->rx_rsp_unconsumed = work_queued; |
1545 | *eoi = 0; |
1546 | } else if (unlikely(work_queued < queue->rx_rsp_unconsumed)) { |
1547 | const struct device *dev = &queue->info->netdev->dev; |
1548 | |
1549 | spin_unlock_irqrestore(lock: &queue->rx_cons_lock, flags); |
1550 | dev_alert(dev, "RX producer index going backwards\n" ); |
1551 | dev_alert(dev, "Disabled for further use\n" ); |
1552 | queue->info->broken = true; |
1553 | return false; |
1554 | } |
1555 | spin_unlock_irqrestore(lock: &queue->rx_cons_lock, flags); |
1556 | |
1557 | if (likely(netif_carrier_ok(queue->info->netdev) && work_queued)) |
1558 | napi_schedule(n: &queue->napi); |
1559 | |
1560 | return true; |
1561 | } |
1562 | |
1563 | static irqreturn_t xennet_rx_interrupt(int irq, void *dev_id) |
1564 | { |
1565 | unsigned int eoiflag = XEN_EOI_FLAG_SPURIOUS; |
1566 | |
1567 | if (likely(xennet_handle_rx(dev_id, &eoiflag))) |
1568 | xen_irq_lateeoi(irq, eoi_flags: eoiflag); |
1569 | |
1570 | return IRQ_HANDLED; |
1571 | } |
1572 | |
1573 | static irqreturn_t xennet_interrupt(int irq, void *dev_id) |
1574 | { |
1575 | unsigned int eoiflag = XEN_EOI_FLAG_SPURIOUS; |
1576 | |
1577 | if (xennet_handle_tx(queue: dev_id, eoi: &eoiflag) && |
1578 | xennet_handle_rx(queue: dev_id, eoi: &eoiflag)) |
1579 | xen_irq_lateeoi(irq, eoi_flags: eoiflag); |
1580 | |
1581 | return IRQ_HANDLED; |
1582 | } |
1583 | |
1584 | #ifdef CONFIG_NET_POLL_CONTROLLER |
1585 | static void xennet_poll_controller(struct net_device *dev) |
1586 | { |
1587 | /* Poll each queue */ |
1588 | struct netfront_info *info = netdev_priv(dev); |
1589 | unsigned int num_queues = dev->real_num_tx_queues; |
1590 | unsigned int i; |
1591 | |
1592 | if (info->broken) |
1593 | return; |
1594 | |
1595 | for (i = 0; i < num_queues; ++i) |
1596 | xennet_interrupt(irq: 0, dev_id: &info->queues[i]); |
1597 | } |
1598 | #endif |
1599 | |
1600 | #define NETBACK_XDP_HEADROOM_DISABLE 0 |
1601 | #define NETBACK_XDP_HEADROOM_ENABLE 1 |
1602 | |
1603 | static int talk_to_netback_xdp(struct netfront_info *np, int xdp) |
1604 | { |
1605 | int err; |
1606 | unsigned short headroom; |
1607 | |
1608 | headroom = xdp ? XDP_PACKET_HEADROOM : 0; |
1609 | err = xenbus_printf(XBT_NIL, dir: np->xbdev->nodename, |
1610 | node: "xdp-headroom" , fmt: "%hu" , |
1611 | headroom); |
1612 | if (err) |
1613 | pr_warn("Error writing xdp-headroom\n" ); |
1614 | |
1615 | return err; |
1616 | } |
1617 | |
1618 | static int xennet_xdp_set(struct net_device *dev, struct bpf_prog *prog, |
1619 | struct netlink_ext_ack *extack) |
1620 | { |
1621 | unsigned long max_mtu = XEN_PAGE_SIZE - XDP_PACKET_HEADROOM; |
1622 | struct netfront_info *np = netdev_priv(dev); |
1623 | struct bpf_prog *old_prog; |
1624 | unsigned int i, err; |
1625 | |
1626 | if (dev->mtu > max_mtu) { |
1627 | netdev_warn(dev, format: "XDP requires MTU less than %lu\n" , max_mtu); |
1628 | return -EINVAL; |
1629 | } |
1630 | |
1631 | if (!np->netback_has_xdp_headroom) |
1632 | return 0; |
1633 | |
1634 | xenbus_switch_state(dev: np->xbdev, new_state: XenbusStateReconfiguring); |
1635 | |
1636 | err = talk_to_netback_xdp(np, xdp: prog ? NETBACK_XDP_HEADROOM_ENABLE : |
1637 | NETBACK_XDP_HEADROOM_DISABLE); |
1638 | if (err) |
1639 | return err; |
1640 | |
1641 | /* avoid the race with XDP headroom adjustment */ |
1642 | wait_event(module_wq, |
1643 | xenbus_read_driver_state(np->xbdev->otherend) == |
1644 | XenbusStateReconfigured); |
1645 | np->netfront_xdp_enabled = true; |
1646 | |
1647 | old_prog = rtnl_dereference(np->queues[0].xdp_prog); |
1648 | |
1649 | if (prog) |
1650 | bpf_prog_add(prog, i: dev->real_num_tx_queues); |
1651 | |
1652 | for (i = 0; i < dev->real_num_tx_queues; ++i) |
1653 | rcu_assign_pointer(np->queues[i].xdp_prog, prog); |
1654 | |
1655 | if (old_prog) |
1656 | for (i = 0; i < dev->real_num_tx_queues; ++i) |
1657 | bpf_prog_put(prog: old_prog); |
1658 | |
1659 | xenbus_switch_state(dev: np->xbdev, new_state: XenbusStateConnected); |
1660 | |
1661 | return 0; |
1662 | } |
1663 | |
1664 | static int xennet_xdp(struct net_device *dev, struct netdev_bpf *xdp) |
1665 | { |
1666 | struct netfront_info *np = netdev_priv(dev); |
1667 | |
1668 | if (np->broken) |
1669 | return -ENODEV; |
1670 | |
1671 | switch (xdp->command) { |
1672 | case XDP_SETUP_PROG: |
1673 | return xennet_xdp_set(dev, prog: xdp->prog, extack: xdp->extack); |
1674 | default: |
1675 | return -EINVAL; |
1676 | } |
1677 | } |
1678 | |
1679 | static const struct net_device_ops xennet_netdev_ops = { |
1680 | .ndo_uninit = xennet_uninit, |
1681 | .ndo_open = xennet_open, |
1682 | .ndo_stop = xennet_close, |
1683 | .ndo_start_xmit = xennet_start_xmit, |
1684 | .ndo_change_mtu = xennet_change_mtu, |
1685 | .ndo_get_stats64 = xennet_get_stats64, |
1686 | .ndo_set_mac_address = eth_mac_addr, |
1687 | .ndo_validate_addr = eth_validate_addr, |
1688 | .ndo_fix_features = xennet_fix_features, |
1689 | .ndo_set_features = xennet_set_features, |
1690 | .ndo_select_queue = xennet_select_queue, |
1691 | .ndo_bpf = xennet_xdp, |
1692 | .ndo_xdp_xmit = xennet_xdp_xmit, |
1693 | #ifdef CONFIG_NET_POLL_CONTROLLER |
1694 | .ndo_poll_controller = xennet_poll_controller, |
1695 | #endif |
1696 | }; |
1697 | |
1698 | static void xennet_free_netdev(struct net_device *netdev) |
1699 | { |
1700 | struct netfront_info *np = netdev_priv(dev: netdev); |
1701 | |
1702 | free_percpu(pdata: np->rx_stats); |
1703 | free_percpu(pdata: np->tx_stats); |
1704 | free_netdev(dev: netdev); |
1705 | } |
1706 | |
1707 | static struct net_device *xennet_create_dev(struct xenbus_device *dev) |
1708 | { |
1709 | int err; |
1710 | struct net_device *netdev; |
1711 | struct netfront_info *np; |
1712 | |
1713 | netdev = alloc_etherdev_mq(sizeof(struct netfront_info), xennet_max_queues); |
1714 | if (!netdev) |
1715 | return ERR_PTR(error: -ENOMEM); |
1716 | |
1717 | np = netdev_priv(dev: netdev); |
1718 | np->xbdev = dev; |
1719 | |
1720 | np->queues = NULL; |
1721 | |
1722 | err = -ENOMEM; |
1723 | np->rx_stats = netdev_alloc_pcpu_stats(struct netfront_stats); |
1724 | if (np->rx_stats == NULL) |
1725 | goto exit; |
1726 | np->tx_stats = netdev_alloc_pcpu_stats(struct netfront_stats); |
1727 | if (np->tx_stats == NULL) |
1728 | goto exit; |
1729 | |
1730 | netdev->netdev_ops = &xennet_netdev_ops; |
1731 | |
1732 | netdev->features = NETIF_F_IP_CSUM | NETIF_F_RXCSUM | |
1733 | NETIF_F_GSO_ROBUST; |
1734 | netdev->hw_features = NETIF_F_SG | |
1735 | NETIF_F_IPV6_CSUM | |
1736 | NETIF_F_TSO | NETIF_F_TSO6; |
1737 | |
1738 | /* |
1739 | * Assume that all hw features are available for now. This set |
1740 | * will be adjusted by the call to netdev_update_features() in |
1741 | * xennet_connect() which is the earliest point where we can |
1742 | * negotiate with the backend regarding supported features. |
1743 | */ |
1744 | netdev->features |= netdev->hw_features; |
1745 | netdev->xdp_features = NETDEV_XDP_ACT_BASIC | NETDEV_XDP_ACT_REDIRECT | |
1746 | NETDEV_XDP_ACT_NDO_XMIT; |
1747 | |
1748 | netdev->ethtool_ops = &xennet_ethtool_ops; |
1749 | netdev->min_mtu = ETH_MIN_MTU; |
1750 | netdev->max_mtu = XEN_NETIF_MAX_TX_SIZE; |
1751 | SET_NETDEV_DEV(netdev, &dev->dev); |
1752 | |
1753 | np->netdev = netdev; |
1754 | np->netfront_xdp_enabled = false; |
1755 | |
1756 | netif_carrier_off(dev: netdev); |
1757 | |
1758 | do { |
1759 | xenbus_switch_state(dev, new_state: XenbusStateInitialising); |
1760 | err = wait_event_timeout(module_wq, |
1761 | xenbus_read_driver_state(dev->otherend) != |
1762 | XenbusStateClosed && |
1763 | xenbus_read_driver_state(dev->otherend) != |
1764 | XenbusStateUnknown, XENNET_TIMEOUT); |
1765 | } while (!err); |
1766 | |
1767 | return netdev; |
1768 | |
1769 | exit: |
1770 | xennet_free_netdev(netdev); |
1771 | return ERR_PTR(error: err); |
1772 | } |
1773 | |
1774 | /* |
1775 | * Entry point to this code when a new device is created. Allocate the basic |
1776 | * structures and the ring buffers for communication with the backend, and |
1777 | * inform the backend of the appropriate details for those. |
1778 | */ |
1779 | static int netfront_probe(struct xenbus_device *dev, |
1780 | const struct xenbus_device_id *id) |
1781 | { |
1782 | int err; |
1783 | struct net_device *netdev; |
1784 | struct netfront_info *info; |
1785 | |
1786 | netdev = xennet_create_dev(dev); |
1787 | if (IS_ERR(ptr: netdev)) { |
1788 | err = PTR_ERR(ptr: netdev); |
1789 | xenbus_dev_fatal(dev, err, fmt: "creating netdev" ); |
1790 | return err; |
1791 | } |
1792 | |
1793 | info = netdev_priv(dev: netdev); |
1794 | dev_set_drvdata(dev: &dev->dev, data: info); |
1795 | #ifdef CONFIG_SYSFS |
1796 | info->netdev->sysfs_groups[0] = &xennet_dev_group; |
1797 | #endif |
1798 | |
1799 | return 0; |
1800 | } |
1801 | |
1802 | static void xennet_end_access(int ref, void *page) |
1803 | { |
1804 | /* This frees the page as a side-effect */ |
1805 | if (ref != INVALID_GRANT_REF) |
1806 | gnttab_end_foreign_access(ref, virt_to_page(page)); |
1807 | } |
1808 | |
1809 | static void xennet_disconnect_backend(struct netfront_info *info) |
1810 | { |
1811 | unsigned int i = 0; |
1812 | unsigned int num_queues = info->netdev->real_num_tx_queues; |
1813 | |
1814 | netif_carrier_off(dev: info->netdev); |
1815 | |
1816 | for (i = 0; i < num_queues && info->queues; ++i) { |
1817 | struct netfront_queue *queue = &info->queues[i]; |
1818 | |
1819 | del_timer_sync(timer: &queue->rx_refill_timer); |
1820 | |
1821 | if (queue->tx_irq && (queue->tx_irq == queue->rx_irq)) |
1822 | unbind_from_irqhandler(irq: queue->tx_irq, dev_id: queue); |
1823 | if (queue->tx_irq && (queue->tx_irq != queue->rx_irq)) { |
1824 | unbind_from_irqhandler(irq: queue->tx_irq, dev_id: queue); |
1825 | unbind_from_irqhandler(irq: queue->rx_irq, dev_id: queue); |
1826 | } |
1827 | queue->tx_evtchn = queue->rx_evtchn = 0; |
1828 | queue->tx_irq = queue->rx_irq = 0; |
1829 | |
1830 | if (netif_running(dev: info->netdev)) |
1831 | napi_synchronize(n: &queue->napi); |
1832 | |
1833 | xennet_release_tx_bufs(queue); |
1834 | xennet_release_rx_bufs(queue); |
1835 | gnttab_free_grant_references(head: queue->gref_tx_head); |
1836 | gnttab_free_grant_references(head: queue->gref_rx_head); |
1837 | |
1838 | /* End access and free the pages */ |
1839 | xennet_end_access(ref: queue->tx_ring_ref, page: queue->tx.sring); |
1840 | xennet_end_access(ref: queue->rx_ring_ref, page: queue->rx.sring); |
1841 | |
1842 | queue->tx_ring_ref = INVALID_GRANT_REF; |
1843 | queue->rx_ring_ref = INVALID_GRANT_REF; |
1844 | queue->tx.sring = NULL; |
1845 | queue->rx.sring = NULL; |
1846 | |
1847 | page_pool_destroy(pool: queue->page_pool); |
1848 | } |
1849 | } |
1850 | |
1851 | /* |
1852 | * We are reconnecting to the backend, due to a suspend/resume, or a backend |
1853 | * driver restart. We tear down our netif structure and recreate it, but |
1854 | * leave the device-layer structures intact so that this is transparent to the |
1855 | * rest of the kernel. |
1856 | */ |
1857 | static int netfront_resume(struct xenbus_device *dev) |
1858 | { |
1859 | struct netfront_info *info = dev_get_drvdata(dev: &dev->dev); |
1860 | |
1861 | dev_dbg(&dev->dev, "%s\n" , dev->nodename); |
1862 | |
1863 | netif_tx_lock_bh(dev: info->netdev); |
1864 | netif_device_detach(dev: info->netdev); |
1865 | netif_tx_unlock_bh(dev: info->netdev); |
1866 | |
1867 | xennet_disconnect_backend(info); |
1868 | |
1869 | rtnl_lock(); |
1870 | if (info->queues) |
1871 | xennet_destroy_queues(info); |
1872 | rtnl_unlock(); |
1873 | |
1874 | return 0; |
1875 | } |
1876 | |
1877 | static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[]) |
1878 | { |
1879 | char *s, *e, *macstr; |
1880 | int i; |
1881 | |
1882 | macstr = s = xenbus_read(XBT_NIL, dir: dev->nodename, node: "mac" , NULL); |
1883 | if (IS_ERR(macstr)) |
1884 | return PTR_ERR(macstr); |
1885 | |
1886 | for (i = 0; i < ETH_ALEN; i++) { |
1887 | mac[i] = simple_strtoul(s, &e, 16); |
1888 | if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) { |
1889 | kfree(macstr); |
1890 | return -ENOENT; |
1891 | } |
1892 | s = e+1; |
1893 | } |
1894 | |
1895 | kfree(macstr); |
1896 | return 0; |
1897 | } |
1898 | |
1899 | static int setup_netfront_single(struct netfront_queue *queue) |
1900 | { |
1901 | int err; |
1902 | |
1903 | err = xenbus_alloc_evtchn(dev: queue->info->xbdev, port: &queue->tx_evtchn); |
1904 | if (err < 0) |
1905 | goto fail; |
1906 | |
1907 | err = bind_evtchn_to_irqhandler_lateeoi(evtchn: queue->tx_evtchn, |
1908 | handler: xennet_interrupt, irqflags: 0, |
1909 | devname: queue->info->netdev->name, |
1910 | dev_id: queue); |
1911 | if (err < 0) |
1912 | goto bind_fail; |
1913 | queue->rx_evtchn = queue->tx_evtchn; |
1914 | queue->rx_irq = queue->tx_irq = err; |
1915 | |
1916 | return 0; |
1917 | |
1918 | bind_fail: |
1919 | xenbus_free_evtchn(dev: queue->info->xbdev, port: queue->tx_evtchn); |
1920 | queue->tx_evtchn = 0; |
1921 | fail: |
1922 | return err; |
1923 | } |
1924 | |
1925 | static int setup_netfront_split(struct netfront_queue *queue) |
1926 | { |
1927 | int err; |
1928 | |
1929 | err = xenbus_alloc_evtchn(dev: queue->info->xbdev, port: &queue->tx_evtchn); |
1930 | if (err < 0) |
1931 | goto fail; |
1932 | err = xenbus_alloc_evtchn(dev: queue->info->xbdev, port: &queue->rx_evtchn); |
1933 | if (err < 0) |
1934 | goto alloc_rx_evtchn_fail; |
1935 | |
1936 | snprintf(buf: queue->tx_irq_name, size: sizeof(queue->tx_irq_name), |
1937 | fmt: "%s-tx" , queue->name); |
1938 | err = bind_evtchn_to_irqhandler_lateeoi(evtchn: queue->tx_evtchn, |
1939 | handler: xennet_tx_interrupt, irqflags: 0, |
1940 | devname: queue->tx_irq_name, dev_id: queue); |
1941 | if (err < 0) |
1942 | goto bind_tx_fail; |
1943 | queue->tx_irq = err; |
1944 | |
1945 | snprintf(buf: queue->rx_irq_name, size: sizeof(queue->rx_irq_name), |
1946 | fmt: "%s-rx" , queue->name); |
1947 | err = bind_evtchn_to_irqhandler_lateeoi(evtchn: queue->rx_evtchn, |
1948 | handler: xennet_rx_interrupt, irqflags: 0, |
1949 | devname: queue->rx_irq_name, dev_id: queue); |
1950 | if (err < 0) |
1951 | goto bind_rx_fail; |
1952 | queue->rx_irq = err; |
1953 | |
1954 | return 0; |
1955 | |
1956 | bind_rx_fail: |
1957 | unbind_from_irqhandler(irq: queue->tx_irq, dev_id: queue); |
1958 | queue->tx_irq = 0; |
1959 | bind_tx_fail: |
1960 | xenbus_free_evtchn(dev: queue->info->xbdev, port: queue->rx_evtchn); |
1961 | queue->rx_evtchn = 0; |
1962 | alloc_rx_evtchn_fail: |
1963 | xenbus_free_evtchn(dev: queue->info->xbdev, port: queue->tx_evtchn); |
1964 | queue->tx_evtchn = 0; |
1965 | fail: |
1966 | return err; |
1967 | } |
1968 | |
1969 | static int setup_netfront(struct xenbus_device *dev, |
1970 | struct netfront_queue *queue, unsigned int feature_split_evtchn) |
1971 | { |
1972 | struct xen_netif_tx_sring *txs; |
1973 | struct xen_netif_rx_sring *rxs; |
1974 | int err; |
1975 | |
1976 | queue->tx_ring_ref = INVALID_GRANT_REF; |
1977 | queue->rx_ring_ref = INVALID_GRANT_REF; |
1978 | queue->rx.sring = NULL; |
1979 | queue->tx.sring = NULL; |
1980 | |
1981 | err = xenbus_setup_ring(dev, GFP_NOIO | __GFP_HIGH, vaddr: (void **)&txs, |
1982 | nr_pages: 1, grefs: &queue->tx_ring_ref); |
1983 | if (err) |
1984 | goto fail; |
1985 | |
1986 | XEN_FRONT_RING_INIT(&queue->tx, txs, XEN_PAGE_SIZE); |
1987 | |
1988 | err = xenbus_setup_ring(dev, GFP_NOIO | __GFP_HIGH, vaddr: (void **)&rxs, |
1989 | nr_pages: 1, grefs: &queue->rx_ring_ref); |
1990 | if (err) |
1991 | goto fail; |
1992 | |
1993 | XEN_FRONT_RING_INIT(&queue->rx, rxs, XEN_PAGE_SIZE); |
1994 | |
1995 | if (feature_split_evtchn) |
1996 | err = setup_netfront_split(queue); |
1997 | /* setup single event channel if |
1998 | * a) feature-split-event-channels == 0 |
1999 | * b) feature-split-event-channels == 1 but failed to setup |
2000 | */ |
2001 | if (!feature_split_evtchn || err) |
2002 | err = setup_netfront_single(queue); |
2003 | |
2004 | if (err) |
2005 | goto fail; |
2006 | |
2007 | return 0; |
2008 | |
2009 | fail: |
2010 | xenbus_teardown_ring(vaddr: (void **)&queue->rx.sring, nr_pages: 1, grefs: &queue->rx_ring_ref); |
2011 | xenbus_teardown_ring(vaddr: (void **)&queue->tx.sring, nr_pages: 1, grefs: &queue->tx_ring_ref); |
2012 | |
2013 | return err; |
2014 | } |
2015 | |
2016 | /* Queue-specific initialisation |
2017 | * This used to be done in xennet_create_dev() but must now |
2018 | * be run per-queue. |
2019 | */ |
2020 | static int xennet_init_queue(struct netfront_queue *queue) |
2021 | { |
2022 | unsigned short i; |
2023 | int err = 0; |
2024 | char *devid; |
2025 | |
2026 | spin_lock_init(&queue->tx_lock); |
2027 | spin_lock_init(&queue->rx_lock); |
2028 | spin_lock_init(&queue->rx_cons_lock); |
2029 | |
2030 | timer_setup(&queue->rx_refill_timer, rx_refill_timeout, 0); |
2031 | |
2032 | devid = strrchr(queue->info->xbdev->nodename, '/') + 1; |
2033 | snprintf(buf: queue->name, size: sizeof(queue->name), fmt: "vif%s-q%u" , |
2034 | devid, queue->id); |
2035 | |
2036 | /* Initialise tx_skb_freelist as a free chain containing every entry. */ |
2037 | queue->tx_skb_freelist = 0; |
2038 | queue->tx_pend_queue = TX_LINK_NONE; |
2039 | for (i = 0; i < NET_TX_RING_SIZE; i++) { |
2040 | queue->tx_link[i] = i + 1; |
2041 | queue->grant_tx_ref[i] = INVALID_GRANT_REF; |
2042 | queue->grant_tx_page[i] = NULL; |
2043 | } |
2044 | queue->tx_link[NET_TX_RING_SIZE - 1] = TX_LINK_NONE; |
2045 | |
2046 | /* Clear out rx_skbs */ |
2047 | for (i = 0; i < NET_RX_RING_SIZE; i++) { |
2048 | queue->rx_skbs[i] = NULL; |
2049 | queue->grant_rx_ref[i] = INVALID_GRANT_REF; |
2050 | } |
2051 | |
2052 | /* A grant for every tx ring slot */ |
2053 | if (gnttab_alloc_grant_references(NET_TX_RING_SIZE, |
2054 | pprivate_head: &queue->gref_tx_head) < 0) { |
2055 | pr_alert("can't alloc tx grant refs\n" ); |
2056 | err = -ENOMEM; |
2057 | goto exit; |
2058 | } |
2059 | |
2060 | /* A grant for every rx ring slot */ |
2061 | if (gnttab_alloc_grant_references(NET_RX_RING_SIZE, |
2062 | pprivate_head: &queue->gref_rx_head) < 0) { |
2063 | pr_alert("can't alloc rx grant refs\n" ); |
2064 | err = -ENOMEM; |
2065 | goto exit_free_tx; |
2066 | } |
2067 | |
2068 | return 0; |
2069 | |
2070 | exit_free_tx: |
2071 | gnttab_free_grant_references(head: queue->gref_tx_head); |
2072 | exit: |
2073 | return err; |
2074 | } |
2075 | |
2076 | static int write_queue_xenstore_keys(struct netfront_queue *queue, |
2077 | struct xenbus_transaction *xbt, int write_hierarchical) |
2078 | { |
2079 | /* Write the queue-specific keys into XenStore in the traditional |
2080 | * way for a single queue, or in a queue subkeys for multiple |
2081 | * queues. |
2082 | */ |
2083 | struct xenbus_device *dev = queue->info->xbdev; |
2084 | int err; |
2085 | const char *message; |
2086 | char *path; |
2087 | size_t pathsize; |
2088 | |
2089 | /* Choose the correct place to write the keys */ |
2090 | if (write_hierarchical) { |
2091 | pathsize = strlen(dev->nodename) + 10; |
2092 | path = kzalloc(size: pathsize, GFP_KERNEL); |
2093 | if (!path) { |
2094 | err = -ENOMEM; |
2095 | message = "out of memory while writing ring references" ; |
2096 | goto error; |
2097 | } |
2098 | snprintf(buf: path, size: pathsize, fmt: "%s/queue-%u" , |
2099 | dev->nodename, queue->id); |
2100 | } else { |
2101 | path = (char *)dev->nodename; |
2102 | } |
2103 | |
2104 | /* Write ring references */ |
2105 | err = xenbus_printf(t: *xbt, dir: path, node: "tx-ring-ref" , fmt: "%u" , |
2106 | queue->tx_ring_ref); |
2107 | if (err) { |
2108 | message = "writing tx-ring-ref" ; |
2109 | goto error; |
2110 | } |
2111 | |
2112 | err = xenbus_printf(t: *xbt, dir: path, node: "rx-ring-ref" , fmt: "%u" , |
2113 | queue->rx_ring_ref); |
2114 | if (err) { |
2115 | message = "writing rx-ring-ref" ; |
2116 | goto error; |
2117 | } |
2118 | |
2119 | /* Write event channels; taking into account both shared |
2120 | * and split event channel scenarios. |
2121 | */ |
2122 | if (queue->tx_evtchn == queue->rx_evtchn) { |
2123 | /* Shared event channel */ |
2124 | err = xenbus_printf(t: *xbt, dir: path, |
2125 | node: "event-channel" , fmt: "%u" , queue->tx_evtchn); |
2126 | if (err) { |
2127 | message = "writing event-channel" ; |
2128 | goto error; |
2129 | } |
2130 | } else { |
2131 | /* Split event channels */ |
2132 | err = xenbus_printf(t: *xbt, dir: path, |
2133 | node: "event-channel-tx" , fmt: "%u" , queue->tx_evtchn); |
2134 | if (err) { |
2135 | message = "writing event-channel-tx" ; |
2136 | goto error; |
2137 | } |
2138 | |
2139 | err = xenbus_printf(t: *xbt, dir: path, |
2140 | node: "event-channel-rx" , fmt: "%u" , queue->rx_evtchn); |
2141 | if (err) { |
2142 | message = "writing event-channel-rx" ; |
2143 | goto error; |
2144 | } |
2145 | } |
2146 | |
2147 | if (write_hierarchical) |
2148 | kfree(objp: path); |
2149 | return 0; |
2150 | |
2151 | error: |
2152 | if (write_hierarchical) |
2153 | kfree(objp: path); |
2154 | xenbus_dev_fatal(dev, err, fmt: "%s" , message); |
2155 | return err; |
2156 | } |
2157 | |
2158 | |
2159 | |
2160 | static int xennet_create_page_pool(struct netfront_queue *queue) |
2161 | { |
2162 | int err; |
2163 | struct page_pool_params pp_params = { |
2164 | .order = 0, |
2165 | .flags = 0, |
2166 | .pool_size = NET_RX_RING_SIZE, |
2167 | .nid = NUMA_NO_NODE, |
2168 | .dev = &queue->info->netdev->dev, |
2169 | .offset = XDP_PACKET_HEADROOM, |
2170 | .max_len = XEN_PAGE_SIZE - XDP_PACKET_HEADROOM, |
2171 | }; |
2172 | |
2173 | queue->page_pool = page_pool_create(params: &pp_params); |
2174 | if (IS_ERR(ptr: queue->page_pool)) { |
2175 | err = PTR_ERR(ptr: queue->page_pool); |
2176 | queue->page_pool = NULL; |
2177 | return err; |
2178 | } |
2179 | |
2180 | err = xdp_rxq_info_reg(xdp_rxq: &queue->xdp_rxq, dev: queue->info->netdev, |
2181 | queue_index: queue->id, napi_id: 0); |
2182 | if (err) { |
2183 | netdev_err(dev: queue->info->netdev, format: "xdp_rxq_info_reg failed\n" ); |
2184 | goto err_free_pp; |
2185 | } |
2186 | |
2187 | err = xdp_rxq_info_reg_mem_model(xdp_rxq: &queue->xdp_rxq, |
2188 | type: MEM_TYPE_PAGE_POOL, allocator: queue->page_pool); |
2189 | if (err) { |
2190 | netdev_err(dev: queue->info->netdev, format: "xdp_rxq_info_reg_mem_model failed\n" ); |
2191 | goto err_unregister_rxq; |
2192 | } |
2193 | return 0; |
2194 | |
2195 | err_unregister_rxq: |
2196 | xdp_rxq_info_unreg(xdp_rxq: &queue->xdp_rxq); |
2197 | err_free_pp: |
2198 | page_pool_destroy(pool: queue->page_pool); |
2199 | queue->page_pool = NULL; |
2200 | return err; |
2201 | } |
2202 | |
2203 | static int xennet_create_queues(struct netfront_info *info, |
2204 | unsigned int *num_queues) |
2205 | { |
2206 | unsigned int i; |
2207 | int ret; |
2208 | |
2209 | info->queues = kcalloc(n: *num_queues, size: sizeof(struct netfront_queue), |
2210 | GFP_KERNEL); |
2211 | if (!info->queues) |
2212 | return -ENOMEM; |
2213 | |
2214 | for (i = 0; i < *num_queues; i++) { |
2215 | struct netfront_queue *queue = &info->queues[i]; |
2216 | |
2217 | queue->id = i; |
2218 | queue->info = info; |
2219 | |
2220 | ret = xennet_init_queue(queue); |
2221 | if (ret < 0) { |
2222 | dev_warn(&info->xbdev->dev, |
2223 | "only created %d queues\n" , i); |
2224 | *num_queues = i; |
2225 | break; |
2226 | } |
2227 | |
2228 | /* use page pool recycling instead of buddy allocator */ |
2229 | ret = xennet_create_page_pool(queue); |
2230 | if (ret < 0) { |
2231 | dev_err(&info->xbdev->dev, "can't allocate page pool\n" ); |
2232 | *num_queues = i; |
2233 | return ret; |
2234 | } |
2235 | |
2236 | netif_napi_add(dev: queue->info->netdev, napi: &queue->napi, poll: xennet_poll); |
2237 | if (netif_running(dev: info->netdev)) |
2238 | napi_enable(n: &queue->napi); |
2239 | } |
2240 | |
2241 | netif_set_real_num_tx_queues(dev: info->netdev, txq: *num_queues); |
2242 | |
2243 | if (*num_queues == 0) { |
2244 | dev_err(&info->xbdev->dev, "no queues\n" ); |
2245 | return -EINVAL; |
2246 | } |
2247 | return 0; |
2248 | } |
2249 | |
2250 | /* Common code used when first setting up, and when resuming. */ |
2251 | static int talk_to_netback(struct xenbus_device *dev, |
2252 | struct netfront_info *info) |
2253 | { |
2254 | const char *message; |
2255 | struct xenbus_transaction xbt; |
2256 | int err; |
2257 | unsigned int feature_split_evtchn; |
2258 | unsigned int i = 0; |
2259 | unsigned int max_queues = 0; |
2260 | struct netfront_queue *queue = NULL; |
2261 | unsigned int num_queues = 1; |
2262 | u8 addr[ETH_ALEN]; |
2263 | |
2264 | info->netdev->irq = 0; |
2265 | |
2266 | /* Check if backend is trusted. */ |
2267 | info->bounce = !xennet_trusted || |
2268 | !xenbus_read_unsigned(dir: dev->nodename, node: "trusted" , default_val: 1); |
2269 | |
2270 | /* Check if backend supports multiple queues */ |
2271 | max_queues = xenbus_read_unsigned(dir: info->xbdev->otherend, |
2272 | node: "multi-queue-max-queues" , default_val: 1); |
2273 | num_queues = min(max_queues, xennet_max_queues); |
2274 | |
2275 | /* Check feature-split-event-channels */ |
2276 | feature_split_evtchn = xenbus_read_unsigned(dir: info->xbdev->otherend, |
2277 | node: "feature-split-event-channels" , default_val: 0); |
2278 | |
2279 | /* Read mac addr. */ |
2280 | err = xen_net_read_mac(dev, mac: addr); |
2281 | if (err) { |
2282 | xenbus_dev_fatal(dev, err, fmt: "parsing %s/mac" , dev->nodename); |
2283 | goto out_unlocked; |
2284 | } |
2285 | eth_hw_addr_set(dev: info->netdev, addr); |
2286 | |
2287 | info->netback_has_xdp_headroom = xenbus_read_unsigned(dir: info->xbdev->otherend, |
2288 | node: "feature-xdp-headroom" , default_val: 0); |
2289 | if (info->netback_has_xdp_headroom) { |
2290 | /* set the current xen-netfront xdp state */ |
2291 | err = talk_to_netback_xdp(np: info, xdp: info->netfront_xdp_enabled ? |
2292 | NETBACK_XDP_HEADROOM_ENABLE : |
2293 | NETBACK_XDP_HEADROOM_DISABLE); |
2294 | if (err) |
2295 | goto out_unlocked; |
2296 | } |
2297 | |
2298 | rtnl_lock(); |
2299 | if (info->queues) |
2300 | xennet_destroy_queues(info); |
2301 | |
2302 | /* For the case of a reconnect reset the "broken" indicator. */ |
2303 | info->broken = false; |
2304 | |
2305 | err = xennet_create_queues(info, num_queues: &num_queues); |
2306 | if (err < 0) { |
2307 | xenbus_dev_fatal(dev, err, fmt: "creating queues" ); |
2308 | kfree(objp: info->queues); |
2309 | info->queues = NULL; |
2310 | goto out; |
2311 | } |
2312 | rtnl_unlock(); |
2313 | |
2314 | /* Create shared ring, alloc event channel -- for each queue */ |
2315 | for (i = 0; i < num_queues; ++i) { |
2316 | queue = &info->queues[i]; |
2317 | err = setup_netfront(dev, queue, feature_split_evtchn); |
2318 | if (err) |
2319 | goto destroy_ring; |
2320 | } |
2321 | |
2322 | again: |
2323 | err = xenbus_transaction_start(t: &xbt); |
2324 | if (err) { |
2325 | xenbus_dev_fatal(dev, err, fmt: "starting transaction" ); |
2326 | goto destroy_ring; |
2327 | } |
2328 | |
2329 | if (xenbus_exists(XBT_NIL, |
2330 | dir: info->xbdev->otherend, node: "multi-queue-max-queues" )) { |
2331 | /* Write the number of queues */ |
2332 | err = xenbus_printf(xbt, dev->nodename, |
2333 | "multi-queue-num-queues" , "%u" , num_queues); |
2334 | if (err) { |
2335 | message = "writing multi-queue-num-queues" ; |
2336 | goto abort_transaction_no_dev_fatal; |
2337 | } |
2338 | } |
2339 | |
2340 | if (num_queues == 1) { |
2341 | err = write_queue_xenstore_keys(&info->queues[0], &xbt, 0); /* flat */ |
2342 | if (err) |
2343 | goto abort_transaction_no_dev_fatal; |
2344 | } else { |
2345 | /* Write the keys for each queue */ |
2346 | for (i = 0; i < num_queues; ++i) { |
2347 | queue = &info->queues[i]; |
2348 | err = write_queue_xenstore_keys(queue, &xbt, 1); /* hierarchical */ |
2349 | if (err) |
2350 | goto abort_transaction_no_dev_fatal; |
2351 | } |
2352 | } |
2353 | |
2354 | /* The remaining keys are not queue-specific */ |
2355 | err = xenbus_printf(xbt, dev->nodename, "request-rx-copy" , "%u" , |
2356 | 1); |
2357 | if (err) { |
2358 | message = "writing request-rx-copy" ; |
2359 | goto abort_transaction; |
2360 | } |
2361 | |
2362 | err = xenbus_printf(xbt, dev->nodename, "feature-rx-notify" , "%d" , 1); |
2363 | if (err) { |
2364 | message = "writing feature-rx-notify" ; |
2365 | goto abort_transaction; |
2366 | } |
2367 | |
2368 | err = xenbus_printf(xbt, dev->nodename, "feature-sg" , "%d" , 1); |
2369 | if (err) { |
2370 | message = "writing feature-sg" ; |
2371 | goto abort_transaction; |
2372 | } |
2373 | |
2374 | err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4" , "%d" , 1); |
2375 | if (err) { |
2376 | message = "writing feature-gso-tcpv4" ; |
2377 | goto abort_transaction; |
2378 | } |
2379 | |
2380 | err = xenbus_write(xbt, dev->nodename, "feature-gso-tcpv6" , "1" ); |
2381 | if (err) { |
2382 | message = "writing feature-gso-tcpv6" ; |
2383 | goto abort_transaction; |
2384 | } |
2385 | |
2386 | err = xenbus_write(xbt, dev->nodename, "feature-ipv6-csum-offload" , |
2387 | "1" ); |
2388 | if (err) { |
2389 | message = "writing feature-ipv6-csum-offload" ; |
2390 | goto abort_transaction; |
2391 | } |
2392 | |
2393 | err = xenbus_transaction_end(xbt, 0); |
2394 | if (err) { |
2395 | if (err == -EAGAIN) |
2396 | goto again; |
2397 | xenbus_dev_fatal(dev, err, "completing transaction" ); |
2398 | goto destroy_ring; |
2399 | } |
2400 | |
2401 | return 0; |
2402 | |
2403 | abort_transaction: |
2404 | xenbus_dev_fatal(dev, err, "%s" , message); |
2405 | abort_transaction_no_dev_fatal: |
2406 | xenbus_transaction_end(xbt, 1); |
2407 | destroy_ring: |
2408 | xennet_disconnect_backend(info); |
2409 | rtnl_lock(); |
2410 | xennet_destroy_queues(info); |
2411 | out: |
2412 | rtnl_unlock(); |
2413 | out_unlocked: |
2414 | device_unregister(&dev->dev); |
2415 | return err; |
2416 | } |
2417 | |
2418 | static int xennet_connect(struct net_device *dev) |
2419 | { |
2420 | struct netfront_info *np = netdev_priv(dev); |
2421 | unsigned int num_queues = 0; |
2422 | int err; |
2423 | unsigned int j = 0; |
2424 | struct netfront_queue *queue = NULL; |
2425 | |
2426 | if (!xenbus_read_unsigned(dir: np->xbdev->otherend, node: "feature-rx-copy" , default_val: 0)) { |
2427 | dev_info(&dev->dev, |
2428 | "backend does not support copying receive path\n" ); |
2429 | return -ENODEV; |
2430 | } |
2431 | |
2432 | err = talk_to_netback(dev: np->xbdev, info: np); |
2433 | if (err) |
2434 | return err; |
2435 | if (np->netback_has_xdp_headroom) |
2436 | pr_info("backend supports XDP headroom\n" ); |
2437 | if (np->bounce) |
2438 | dev_info(&np->xbdev->dev, |
2439 | "bouncing transmitted data to zeroed pages\n" ); |
2440 | |
2441 | /* talk_to_netback() sets the correct number of queues */ |
2442 | num_queues = dev->real_num_tx_queues; |
2443 | |
2444 | if (dev->reg_state == NETREG_UNINITIALIZED) { |
2445 | err = register_netdev(dev); |
2446 | if (err) { |
2447 | pr_warn("%s: register_netdev err=%d\n" , __func__, err); |
2448 | device_unregister(dev: &np->xbdev->dev); |
2449 | return err; |
2450 | } |
2451 | } |
2452 | |
2453 | rtnl_lock(); |
2454 | netdev_update_features(dev); |
2455 | rtnl_unlock(); |
2456 | |
2457 | /* |
2458 | * All public and private state should now be sane. Get |
2459 | * ready to start sending and receiving packets and give the driver |
2460 | * domain a kick because we've probably just requeued some |
2461 | * packets. |
2462 | */ |
2463 | netif_tx_lock_bh(dev: np->netdev); |
2464 | netif_device_attach(dev: np->netdev); |
2465 | netif_tx_unlock_bh(dev: np->netdev); |
2466 | |
2467 | netif_carrier_on(dev: np->netdev); |
2468 | for (j = 0; j < num_queues; ++j) { |
2469 | queue = &np->queues[j]; |
2470 | |
2471 | notify_remote_via_irq(irq: queue->tx_irq); |
2472 | if (queue->tx_irq != queue->rx_irq) |
2473 | notify_remote_via_irq(irq: queue->rx_irq); |
2474 | |
2475 | spin_lock_bh(lock: &queue->rx_lock); |
2476 | xennet_alloc_rx_buffers(queue); |
2477 | spin_unlock_bh(lock: &queue->rx_lock); |
2478 | } |
2479 | |
2480 | return 0; |
2481 | } |
2482 | |
2483 | /* |
2484 | * Callback received when the backend's state changes. |
2485 | */ |
2486 | static void netback_changed(struct xenbus_device *dev, |
2487 | enum xenbus_state backend_state) |
2488 | { |
2489 | struct netfront_info *np = dev_get_drvdata(dev: &dev->dev); |
2490 | struct net_device *netdev = np->netdev; |
2491 | |
2492 | dev_dbg(&dev->dev, "%s\n" , xenbus_strstate(backend_state)); |
2493 | |
2494 | wake_up_all(&module_wq); |
2495 | |
2496 | switch (backend_state) { |
2497 | case XenbusStateInitialising: |
2498 | case XenbusStateInitialised: |
2499 | case XenbusStateReconfiguring: |
2500 | case XenbusStateReconfigured: |
2501 | case XenbusStateUnknown: |
2502 | break; |
2503 | |
2504 | case XenbusStateInitWait: |
2505 | if (dev->state != XenbusStateInitialising) |
2506 | break; |
2507 | if (xennet_connect(dev: netdev) != 0) |
2508 | break; |
2509 | xenbus_switch_state(dev, new_state: XenbusStateConnected); |
2510 | break; |
2511 | |
2512 | case XenbusStateConnected: |
2513 | netdev_notify_peers(dev: netdev); |
2514 | break; |
2515 | |
2516 | case XenbusStateClosed: |
2517 | if (dev->state == XenbusStateClosed) |
2518 | break; |
2519 | fallthrough; /* Missed the backend's CLOSING state */ |
2520 | case XenbusStateClosing: |
2521 | xenbus_frontend_closed(dev); |
2522 | break; |
2523 | } |
2524 | } |
2525 | |
2526 | static const struct xennet_stat { |
2527 | char name[ETH_GSTRING_LEN]; |
2528 | u16 offset; |
2529 | } xennet_stats[] = { |
2530 | { |
2531 | "rx_gso_checksum_fixup" , |
2532 | offsetof(struct netfront_info, rx_gso_checksum_fixup) |
2533 | }, |
2534 | }; |
2535 | |
2536 | static int xennet_get_sset_count(struct net_device *dev, int string_set) |
2537 | { |
2538 | switch (string_set) { |
2539 | case ETH_SS_STATS: |
2540 | return ARRAY_SIZE(xennet_stats); |
2541 | default: |
2542 | return -EINVAL; |
2543 | } |
2544 | } |
2545 | |
2546 | static void xennet_get_ethtool_stats(struct net_device *dev, |
2547 | struct ethtool_stats *stats, u64 * data) |
2548 | { |
2549 | void *np = netdev_priv(dev); |
2550 | int i; |
2551 | |
2552 | for (i = 0; i < ARRAY_SIZE(xennet_stats); i++) |
2553 | data[i] = atomic_read(v: (atomic_t *)(np + xennet_stats[i].offset)); |
2554 | } |
2555 | |
2556 | static void xennet_get_strings(struct net_device *dev, u32 stringset, u8 * data) |
2557 | { |
2558 | int i; |
2559 | |
2560 | switch (stringset) { |
2561 | case ETH_SS_STATS: |
2562 | for (i = 0; i < ARRAY_SIZE(xennet_stats); i++) |
2563 | memcpy(data + i * ETH_GSTRING_LEN, |
2564 | xennet_stats[i].name, ETH_GSTRING_LEN); |
2565 | break; |
2566 | } |
2567 | } |
2568 | |
2569 | static const struct ethtool_ops xennet_ethtool_ops = |
2570 | { |
2571 | .get_link = ethtool_op_get_link, |
2572 | |
2573 | .get_sset_count = xennet_get_sset_count, |
2574 | .get_ethtool_stats = xennet_get_ethtool_stats, |
2575 | .get_strings = xennet_get_strings, |
2576 | .get_ts_info = ethtool_op_get_ts_info, |
2577 | }; |
2578 | |
2579 | #ifdef CONFIG_SYSFS |
2580 | static ssize_t show_rxbuf(struct device *dev, |
2581 | struct device_attribute *attr, char *buf) |
2582 | { |
2583 | return sprintf(buf, fmt: "%lu\n" , NET_RX_RING_SIZE); |
2584 | } |
2585 | |
2586 | static ssize_t store_rxbuf(struct device *dev, |
2587 | struct device_attribute *attr, |
2588 | const char *buf, size_t len) |
2589 | { |
2590 | char *endp; |
2591 | |
2592 | if (!capable(CAP_NET_ADMIN)) |
2593 | return -EPERM; |
2594 | |
2595 | simple_strtoul(buf, &endp, 0); |
2596 | if (endp == buf) |
2597 | return -EBADMSG; |
2598 | |
2599 | /* rxbuf_min and rxbuf_max are no longer configurable. */ |
2600 | |
2601 | return len; |
2602 | } |
2603 | |
2604 | static DEVICE_ATTR(rxbuf_min, 0644, show_rxbuf, store_rxbuf); |
2605 | static DEVICE_ATTR(rxbuf_max, 0644, show_rxbuf, store_rxbuf); |
2606 | static DEVICE_ATTR(rxbuf_cur, 0444, show_rxbuf, NULL); |
2607 | |
2608 | static struct attribute *xennet_dev_attrs[] = { |
2609 | &dev_attr_rxbuf_min.attr, |
2610 | &dev_attr_rxbuf_max.attr, |
2611 | &dev_attr_rxbuf_cur.attr, |
2612 | NULL |
2613 | }; |
2614 | |
2615 | static const struct attribute_group xennet_dev_group = { |
2616 | .attrs = xennet_dev_attrs |
2617 | }; |
2618 | #endif /* CONFIG_SYSFS */ |
2619 | |
2620 | static void xennet_bus_close(struct xenbus_device *dev) |
2621 | { |
2622 | int ret; |
2623 | |
2624 | if (xenbus_read_driver_state(path: dev->otherend) == XenbusStateClosed) |
2625 | return; |
2626 | do { |
2627 | xenbus_switch_state(dev, new_state: XenbusStateClosing); |
2628 | ret = wait_event_timeout(module_wq, |
2629 | xenbus_read_driver_state(dev->otherend) == |
2630 | XenbusStateClosing || |
2631 | xenbus_read_driver_state(dev->otherend) == |
2632 | XenbusStateClosed || |
2633 | xenbus_read_driver_state(dev->otherend) == |
2634 | XenbusStateUnknown, |
2635 | XENNET_TIMEOUT); |
2636 | } while (!ret); |
2637 | |
2638 | if (xenbus_read_driver_state(path: dev->otherend) == XenbusStateClosed) |
2639 | return; |
2640 | |
2641 | do { |
2642 | xenbus_switch_state(dev, new_state: XenbusStateClosed); |
2643 | ret = wait_event_timeout(module_wq, |
2644 | xenbus_read_driver_state(dev->otherend) == |
2645 | XenbusStateClosed || |
2646 | xenbus_read_driver_state(dev->otherend) == |
2647 | XenbusStateUnknown, |
2648 | XENNET_TIMEOUT); |
2649 | } while (!ret); |
2650 | } |
2651 | |
2652 | static void xennet_remove(struct xenbus_device *dev) |
2653 | { |
2654 | struct netfront_info *info = dev_get_drvdata(dev: &dev->dev); |
2655 | |
2656 | xennet_bus_close(dev); |
2657 | xennet_disconnect_backend(info); |
2658 | |
2659 | if (info->netdev->reg_state == NETREG_REGISTERED) |
2660 | unregister_netdev(dev: info->netdev); |
2661 | |
2662 | if (info->queues) { |
2663 | rtnl_lock(); |
2664 | xennet_destroy_queues(info); |
2665 | rtnl_unlock(); |
2666 | } |
2667 | xennet_free_netdev(netdev: info->netdev); |
2668 | } |
2669 | |
2670 | static const struct xenbus_device_id netfront_ids[] = { |
2671 | { "vif" }, |
2672 | { "" } |
2673 | }; |
2674 | |
2675 | static struct xenbus_driver netfront_driver = { |
2676 | .ids = netfront_ids, |
2677 | .probe = netfront_probe, |
2678 | .remove = xennet_remove, |
2679 | .resume = netfront_resume, |
2680 | .otherend_changed = netback_changed, |
2681 | }; |
2682 | |
2683 | static int __init netif_init(void) |
2684 | { |
2685 | if (!xen_domain()) |
2686 | return -ENODEV; |
2687 | |
2688 | if (!xen_has_pv_nic_devices()) |
2689 | return -ENODEV; |
2690 | |
2691 | pr_info("Initialising Xen virtual ethernet driver\n" ); |
2692 | |
2693 | /* Allow as many queues as there are CPUs inut max. 8 if user has not |
2694 | * specified a value. |
2695 | */ |
2696 | if (xennet_max_queues == 0) |
2697 | xennet_max_queues = min_t(unsigned int, MAX_QUEUES_DEFAULT, |
2698 | num_online_cpus()); |
2699 | |
2700 | return xenbus_register_frontend(&netfront_driver); |
2701 | } |
2702 | module_init(netif_init); |
2703 | |
2704 | |
2705 | static void __exit netif_exit(void) |
2706 | { |
2707 | xenbus_unregister_driver(drv: &netfront_driver); |
2708 | } |
2709 | module_exit(netif_exit); |
2710 | |
2711 | MODULE_DESCRIPTION("Xen virtual network device frontend" ); |
2712 | MODULE_LICENSE("GPL" ); |
2713 | MODULE_ALIAS("xen:vif" ); |
2714 | MODULE_ALIAS("xennet" ); |
2715 | |