1 /*
2 * Back-end of the driver for virtual network devices. This portion of the
3 * driver exports a 'unified' network-device interface that can be accessed
4 * by any operating system that implements a compatible front end. A
5 * reference front-end implementation can be found in:
6 * drivers/net/xen-netfront.c
7 *
8 * Copyright (c) 2002-2005, K A Fraser
9 *
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public License version 2
12 * as published by the Free Software Foundation; or, when distributed
13 * separately from the Linux kernel or incorporated into other
14 * software packages, subject to the following license:
15 *
16 * Permission is hereby granted, free of charge, to any person obtaining a copy
17 * of this source file (the "Software"), to deal in the Software without
18 * restriction, including without limitation the rights to use, copy, modify,
19 * merge, publish, distribute, sublicense, and/or sell copies of the Software,
20 * and to permit persons to whom the Software is furnished to do so, subject to
21 * the following conditions:
22 *
23 * The above copyright notice and this permission notice shall be included in
24 * all copies or substantial portions of the Software.
25 *
26 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
27 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
28 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
29 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
30 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
31 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
32 * IN THE SOFTWARE.
33 */
34
35 #include "common.h"
36
37 #include <linux/kthread.h>
38 #include <linux/if_vlan.h>
39 #include <linux/udp.h>
40 #include <linux/highmem.h>
41
42 #include <net/tcp.h>
43
44 #include <xen/xen.h>
45 #include <xen/events.h>
46 #include <xen/interface/memory.h>
47 #include <xen/page.h>
48
49 #include <asm/xen/hypercall.h>
50
51 /* Provide an option to disable split event channels at load time as
52 * event channels are limited resource. Split event channels are
53 * enabled by default.
54 */
55 bool separate_tx_rx_irq = true;
56 module_param(separate_tx_rx_irq, bool, 0644);
57
58 /* The time that packets can stay on the guest Rx internal queue
59 * before they are dropped.
60 */
61 unsigned int rx_drain_timeout_msecs = 10000;
62 module_param(rx_drain_timeout_msecs, uint, 0444);
63
64 /* The length of time before the frontend is considered unresponsive
65 * because it isn't providing Rx slots.
66 */
67 unsigned int rx_stall_timeout_msecs = 60000;
68 module_param(rx_stall_timeout_msecs, uint, 0444);
69
70 unsigned int xenvif_max_queues;
71 module_param_named(max_queues, xenvif_max_queues, uint, 0644);
72 MODULE_PARM_DESC(max_queues,
73 "Maximum number of queues per virtual interface");
74
75 /*
76 * This is the maximum slots a skb can have. If a guest sends a skb
77 * which exceeds this limit it is considered malicious.
78 */
79 #define FATAL_SKB_SLOTS_DEFAULT 20
80 static unsigned int fatal_skb_slots = FATAL_SKB_SLOTS_DEFAULT;
81 module_param(fatal_skb_slots, uint, 0444);
82
83 /* The amount to copy out of the first guest Tx slot into the skb's
84 * linear area. If the first slot has more data, it will be mapped
85 * and put into the first frag.
86 *
87 * This is sized to avoid pulling headers from the frags for most
88 * TCP/IP packets.
89 */
90 #define XEN_NETBACK_TX_COPY_LEN 128
91
92
93 static void xenvif_idx_release(struct xenvif_queue *queue, u16 pending_idx,
94 u8 status);
95
96 static void make_tx_response(struct xenvif_queue *queue,
97 struct xen_netif_tx_request *txp,
98 s8 st);
99 static void push_tx_responses(struct xenvif_queue *queue);
100
101 static inline int tx_work_todo(struct xenvif_queue *queue);
102
103 static struct xen_netif_rx_response *make_rx_response(struct xenvif_queue *queue,
104 u16 id,
105 s8 st,
106 u16 offset,
107 u16 size,
108 u16 flags);
109
idx_to_pfn(struct xenvif_queue * queue,u16 idx)110 static inline unsigned long idx_to_pfn(struct xenvif_queue *queue,
111 u16 idx)
112 {
113 return page_to_pfn(queue->mmap_pages[idx]);
114 }
115
idx_to_kaddr(struct xenvif_queue * queue,u16 idx)116 static inline unsigned long idx_to_kaddr(struct xenvif_queue *queue,
117 u16 idx)
118 {
119 return (unsigned long)pfn_to_kaddr(idx_to_pfn(queue, idx));
120 }
121
122 #define callback_param(vif, pending_idx) \
123 (vif->pending_tx_info[pending_idx].callback_struct)
124
125 /* Find the containing VIF's structure from a pointer in pending_tx_info array
126 */
ubuf_to_queue(const struct ubuf_info * ubuf)127 static inline struct xenvif_queue *ubuf_to_queue(const struct ubuf_info *ubuf)
128 {
129 u16 pending_idx = ubuf->desc;
130 struct pending_tx_info *temp =
131 container_of(ubuf, struct pending_tx_info, callback_struct);
132 return container_of(temp - pending_idx,
133 struct xenvif_queue,
134 pending_tx_info[0]);
135 }
136
frag_get_pending_idx(skb_frag_t * frag)137 static u16 frag_get_pending_idx(skb_frag_t *frag)
138 {
139 return (u16)frag->page_offset;
140 }
141
frag_set_pending_idx(skb_frag_t * frag,u16 pending_idx)142 static void frag_set_pending_idx(skb_frag_t *frag, u16 pending_idx)
143 {
144 frag->page_offset = pending_idx;
145 }
146
pending_index(unsigned i)147 static inline pending_ring_idx_t pending_index(unsigned i)
148 {
149 return i & (MAX_PENDING_REQS-1);
150 }
151
xenvif_rx_ring_slots_needed(struct xenvif * vif)152 static int xenvif_rx_ring_slots_needed(struct xenvif *vif)
153 {
154 if (vif->gso_mask)
155 return DIV_ROUND_UP(vif->dev->gso_max_size, XEN_PAGE_SIZE) + 1;
156 else
157 return DIV_ROUND_UP(vif->dev->mtu, XEN_PAGE_SIZE);
158 }
159
xenvif_rx_ring_slots_available(struct xenvif_queue * queue)160 static bool xenvif_rx_ring_slots_available(struct xenvif_queue *queue)
161 {
162 RING_IDX prod, cons;
163 int needed;
164
165 needed = xenvif_rx_ring_slots_needed(queue->vif);
166
167 do {
168 prod = queue->rx.sring->req_prod;
169 cons = queue->rx.req_cons;
170
171 if (prod - cons >= needed)
172 return true;
173
174 queue->rx.sring->req_event = prod + 1;
175
176 /* Make sure event is visible before we check prod
177 * again.
178 */
179 mb();
180 } while (queue->rx.sring->req_prod != prod);
181
182 return false;
183 }
184
xenvif_rx_queue_tail(struct xenvif_queue * queue,struct sk_buff * skb)185 void xenvif_rx_queue_tail(struct xenvif_queue *queue, struct sk_buff *skb)
186 {
187 unsigned long flags;
188
189 spin_lock_irqsave(&queue->rx_queue.lock, flags);
190
191 __skb_queue_tail(&queue->rx_queue, skb);
192
193 queue->rx_queue_len += skb->len;
194 if (queue->rx_queue_len > queue->rx_queue_max)
195 netif_tx_stop_queue(netdev_get_tx_queue(queue->vif->dev, queue->id));
196
197 spin_unlock_irqrestore(&queue->rx_queue.lock, flags);
198 }
199
xenvif_rx_dequeue(struct xenvif_queue * queue)200 static struct sk_buff *xenvif_rx_dequeue(struct xenvif_queue *queue)
201 {
202 struct sk_buff *skb;
203
204 spin_lock_irq(&queue->rx_queue.lock);
205
206 skb = __skb_dequeue(&queue->rx_queue);
207 if (skb)
208 queue->rx_queue_len -= skb->len;
209
210 spin_unlock_irq(&queue->rx_queue.lock);
211
212 return skb;
213 }
214
xenvif_rx_queue_maybe_wake(struct xenvif_queue * queue)215 static void xenvif_rx_queue_maybe_wake(struct xenvif_queue *queue)
216 {
217 spin_lock_irq(&queue->rx_queue.lock);
218
219 if (queue->rx_queue_len < queue->rx_queue_max)
220 netif_tx_wake_queue(netdev_get_tx_queue(queue->vif->dev, queue->id));
221
222 spin_unlock_irq(&queue->rx_queue.lock);
223 }
224
225
xenvif_rx_queue_purge(struct xenvif_queue * queue)226 static void xenvif_rx_queue_purge(struct xenvif_queue *queue)
227 {
228 struct sk_buff *skb;
229 while ((skb = xenvif_rx_dequeue(queue)) != NULL)
230 kfree_skb(skb);
231 }
232
xenvif_rx_queue_drop_expired(struct xenvif_queue * queue)233 static void xenvif_rx_queue_drop_expired(struct xenvif_queue *queue)
234 {
235 struct sk_buff *skb;
236
237 for(;;) {
238 skb = skb_peek(&queue->rx_queue);
239 if (!skb)
240 break;
241 if (time_before(jiffies, XENVIF_RX_CB(skb)->expires))
242 break;
243 xenvif_rx_dequeue(queue);
244 kfree_skb(skb);
245 }
246 }
247
248 struct netrx_pending_operations {
249 unsigned copy_prod, copy_cons;
250 unsigned meta_prod, meta_cons;
251 struct gnttab_copy *copy;
252 struct xenvif_rx_meta *meta;
253 int copy_off;
254 grant_ref_t copy_gref;
255 };
256
get_next_rx_buffer(struct xenvif_queue * queue,struct netrx_pending_operations * npo)257 static struct xenvif_rx_meta *get_next_rx_buffer(struct xenvif_queue *queue,
258 struct netrx_pending_operations *npo)
259 {
260 struct xenvif_rx_meta *meta;
261 struct xen_netif_rx_request req;
262
263 RING_COPY_REQUEST(&queue->rx, queue->rx.req_cons++, &req);
264
265 meta = npo->meta + npo->meta_prod++;
266 meta->gso_type = XEN_NETIF_GSO_TYPE_NONE;
267 meta->gso_size = 0;
268 meta->size = 0;
269 meta->id = req.id;
270
271 npo->copy_off = 0;
272 npo->copy_gref = req.gref;
273
274 return meta;
275 }
276
277 struct gop_frag_copy {
278 struct xenvif_queue *queue;
279 struct netrx_pending_operations *npo;
280 struct xenvif_rx_meta *meta;
281 int head;
282 int gso_type;
283
284 struct page *page;
285 };
286
xenvif_setup_copy_gop(unsigned long gfn,unsigned int offset,unsigned int * len,struct gop_frag_copy * info)287 static void xenvif_setup_copy_gop(unsigned long gfn,
288 unsigned int offset,
289 unsigned int *len,
290 struct gop_frag_copy *info)
291 {
292 struct gnttab_copy *copy_gop;
293 struct xen_page_foreign *foreign;
294 /* Convenient aliases */
295 struct xenvif_queue *queue = info->queue;
296 struct netrx_pending_operations *npo = info->npo;
297 struct page *page = info->page;
298
299 BUG_ON(npo->copy_off > MAX_BUFFER_OFFSET);
300
301 if (npo->copy_off == MAX_BUFFER_OFFSET)
302 info->meta = get_next_rx_buffer(queue, npo);
303
304 if (npo->copy_off + *len > MAX_BUFFER_OFFSET)
305 *len = MAX_BUFFER_OFFSET - npo->copy_off;
306
307 copy_gop = npo->copy + npo->copy_prod++;
308 copy_gop->flags = GNTCOPY_dest_gref;
309 copy_gop->len = *len;
310
311 foreign = xen_page_foreign(page);
312 if (foreign) {
313 copy_gop->source.domid = foreign->domid;
314 copy_gop->source.u.ref = foreign->gref;
315 copy_gop->flags |= GNTCOPY_source_gref;
316 } else {
317 copy_gop->source.domid = DOMID_SELF;
318 copy_gop->source.u.gmfn = gfn;
319 }
320 copy_gop->source.offset = offset;
321
322 copy_gop->dest.domid = queue->vif->domid;
323 copy_gop->dest.offset = npo->copy_off;
324 copy_gop->dest.u.ref = npo->copy_gref;
325
326 npo->copy_off += *len;
327 info->meta->size += *len;
328
329 /* Leave a gap for the GSO descriptor. */
330 if (info->head && ((1 << info->gso_type) & queue->vif->gso_mask))
331 queue->rx.req_cons++;
332
333 info->head = 0; /* There must be something in this buffer now */
334 }
335
xenvif_gop_frag_copy_grant(unsigned long gfn,unsigned offset,unsigned int len,void * data)336 static void xenvif_gop_frag_copy_grant(unsigned long gfn,
337 unsigned offset,
338 unsigned int len,
339 void *data)
340 {
341 unsigned int bytes;
342
343 while (len) {
344 bytes = len;
345 xenvif_setup_copy_gop(gfn, offset, &bytes, data);
346 offset += bytes;
347 len -= bytes;
348 }
349 }
350
351 /*
352 * Set up the grant operations for this fragment. If it's a flipping
353 * interface, we also set up the unmap request from here.
354 */
xenvif_gop_frag_copy(struct xenvif_queue * queue,struct sk_buff * skb,struct netrx_pending_operations * npo,struct page * page,unsigned long size,unsigned long offset,int * head)355 static void xenvif_gop_frag_copy(struct xenvif_queue *queue, struct sk_buff *skb,
356 struct netrx_pending_operations *npo,
357 struct page *page, unsigned long size,
358 unsigned long offset, int *head)
359 {
360 struct gop_frag_copy info = {
361 .queue = queue,
362 .npo = npo,
363 .head = *head,
364 .gso_type = XEN_NETIF_GSO_TYPE_NONE,
365 };
366 unsigned long bytes;
367
368 if (skb_is_gso(skb)) {
369 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4)
370 info.gso_type = XEN_NETIF_GSO_TYPE_TCPV4;
371 else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)
372 info.gso_type = XEN_NETIF_GSO_TYPE_TCPV6;
373 }
374
375 /* Data must not cross a page boundary. */
376 BUG_ON(size + offset > PAGE_SIZE<<compound_order(page));
377
378 info.meta = npo->meta + npo->meta_prod - 1;
379
380 /* Skip unused frames from start of page */
381 page += offset >> PAGE_SHIFT;
382 offset &= ~PAGE_MASK;
383
384 while (size > 0) {
385 BUG_ON(offset >= PAGE_SIZE);
386
387 bytes = PAGE_SIZE - offset;
388 if (bytes > size)
389 bytes = size;
390
391 info.page = page;
392 gnttab_foreach_grant_in_range(page, offset, bytes,
393 xenvif_gop_frag_copy_grant,
394 &info);
395 size -= bytes;
396 offset = 0;
397
398 /* Next page */
399 if (size) {
400 BUG_ON(!PageCompound(page));
401 page++;
402 }
403 }
404
405 *head = info.head;
406 }
407
408 /*
409 * Prepare an SKB to be transmitted to the frontend.
410 *
411 * This function is responsible for allocating grant operations, meta
412 * structures, etc.
413 *
414 * It returns the number of meta structures consumed. The number of
415 * ring slots used is always equal to the number of meta slots used
416 * plus the number of GSO descriptors used. Currently, we use either
417 * zero GSO descriptors (for non-GSO packets) or one descriptor (for
418 * frontend-side LRO).
419 */
xenvif_gop_skb(struct sk_buff * skb,struct netrx_pending_operations * npo,struct xenvif_queue * queue)420 static int xenvif_gop_skb(struct sk_buff *skb,
421 struct netrx_pending_operations *npo,
422 struct xenvif_queue *queue)
423 {
424 struct xenvif *vif = netdev_priv(skb->dev);
425 int nr_frags = skb_shinfo(skb)->nr_frags;
426 int i;
427 struct xen_netif_rx_request req;
428 struct xenvif_rx_meta *meta;
429 unsigned char *data;
430 int head = 1;
431 int old_meta_prod;
432 int gso_type;
433
434 old_meta_prod = npo->meta_prod;
435
436 gso_type = XEN_NETIF_GSO_TYPE_NONE;
437 if (skb_is_gso(skb)) {
438 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4)
439 gso_type = XEN_NETIF_GSO_TYPE_TCPV4;
440 else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)
441 gso_type = XEN_NETIF_GSO_TYPE_TCPV6;
442 }
443
444 /* Set up a GSO prefix descriptor, if necessary */
445 if ((1 << gso_type) & vif->gso_prefix_mask) {
446 RING_COPY_REQUEST(&queue->rx, queue->rx.req_cons++, &req);
447 meta = npo->meta + npo->meta_prod++;
448 meta->gso_type = gso_type;
449 meta->gso_size = skb_shinfo(skb)->gso_size;
450 meta->size = 0;
451 meta->id = req.id;
452 }
453
454 RING_COPY_REQUEST(&queue->rx, queue->rx.req_cons++, &req);
455 meta = npo->meta + npo->meta_prod++;
456
457 if ((1 << gso_type) & vif->gso_mask) {
458 meta->gso_type = gso_type;
459 meta->gso_size = skb_shinfo(skb)->gso_size;
460 } else {
461 meta->gso_type = XEN_NETIF_GSO_TYPE_NONE;
462 meta->gso_size = 0;
463 }
464
465 meta->size = 0;
466 meta->id = req.id;
467 npo->copy_off = 0;
468 npo->copy_gref = req.gref;
469
470 data = skb->data;
471 while (data < skb_tail_pointer(skb)) {
472 unsigned int offset = offset_in_page(data);
473 unsigned int len = PAGE_SIZE - offset;
474
475 if (data + len > skb_tail_pointer(skb))
476 len = skb_tail_pointer(skb) - data;
477
478 xenvif_gop_frag_copy(queue, skb, npo,
479 virt_to_page(data), len, offset, &head);
480 data += len;
481 }
482
483 for (i = 0; i < nr_frags; i++) {
484 xenvif_gop_frag_copy(queue, skb, npo,
485 skb_frag_page(&skb_shinfo(skb)->frags[i]),
486 skb_frag_size(&skb_shinfo(skb)->frags[i]),
487 skb_shinfo(skb)->frags[i].page_offset,
488 &head);
489 }
490
491 return npo->meta_prod - old_meta_prod;
492 }
493
494 /*
495 * This is a twin to xenvif_gop_skb. Assume that xenvif_gop_skb was
496 * used to set up the operations on the top of
497 * netrx_pending_operations, which have since been done. Check that
498 * they didn't give any errors and advance over them.
499 */
xenvif_check_gop(struct xenvif * vif,int nr_meta_slots,struct netrx_pending_operations * npo)500 static int xenvif_check_gop(struct xenvif *vif, int nr_meta_slots,
501 struct netrx_pending_operations *npo)
502 {
503 struct gnttab_copy *copy_op;
504 int status = XEN_NETIF_RSP_OKAY;
505 int i;
506
507 for (i = 0; i < nr_meta_slots; i++) {
508 copy_op = npo->copy + npo->copy_cons++;
509 if (copy_op->status != GNTST_okay) {
510 netdev_dbg(vif->dev,
511 "Bad status %d from copy to DOM%d.\n",
512 copy_op->status, vif->domid);
513 status = XEN_NETIF_RSP_ERROR;
514 }
515 }
516
517 return status;
518 }
519
xenvif_add_frag_responses(struct xenvif_queue * queue,int status,struct xenvif_rx_meta * meta,int nr_meta_slots)520 static void xenvif_add_frag_responses(struct xenvif_queue *queue, int status,
521 struct xenvif_rx_meta *meta,
522 int nr_meta_slots)
523 {
524 int i;
525 unsigned long offset;
526
527 /* No fragments used */
528 if (nr_meta_slots <= 1)
529 return;
530
531 nr_meta_slots--;
532
533 for (i = 0; i < nr_meta_slots; i++) {
534 int flags;
535 if (i == nr_meta_slots - 1)
536 flags = 0;
537 else
538 flags = XEN_NETRXF_more_data;
539
540 offset = 0;
541 make_rx_response(queue, meta[i].id, status, offset,
542 meta[i].size, flags);
543 }
544 }
545
xenvif_kick_thread(struct xenvif_queue * queue)546 void xenvif_kick_thread(struct xenvif_queue *queue)
547 {
548 wake_up(&queue->wq);
549 }
550
xenvif_rx_action(struct xenvif_queue * queue)551 static void xenvif_rx_action(struct xenvif_queue *queue)
552 {
553 s8 status;
554 u16 flags;
555 struct xen_netif_rx_response *resp;
556 struct sk_buff_head rxq;
557 struct sk_buff *skb;
558 LIST_HEAD(notify);
559 int ret;
560 unsigned long offset;
561 bool need_to_notify = false;
562
563 struct netrx_pending_operations npo = {
564 .copy = queue->grant_copy_op,
565 .meta = queue->meta,
566 };
567
568 skb_queue_head_init(&rxq);
569
570 while (xenvif_rx_ring_slots_available(queue)
571 && (skb = xenvif_rx_dequeue(queue)) != NULL) {
572 queue->last_rx_time = jiffies;
573
574 XENVIF_RX_CB(skb)->meta_slots_used = xenvif_gop_skb(skb, &npo, queue);
575
576 __skb_queue_tail(&rxq, skb);
577 }
578
579 BUG_ON(npo.meta_prod > ARRAY_SIZE(queue->meta));
580
581 if (!npo.copy_prod)
582 goto done;
583
584 BUG_ON(npo.copy_prod > MAX_GRANT_COPY_OPS);
585 gnttab_batch_copy(queue->grant_copy_op, npo.copy_prod);
586
587 while ((skb = __skb_dequeue(&rxq)) != NULL) {
588
589 if ((1 << queue->meta[npo.meta_cons].gso_type) &
590 queue->vif->gso_prefix_mask) {
591 resp = RING_GET_RESPONSE(&queue->rx,
592 queue->rx.rsp_prod_pvt++);
593
594 resp->flags = XEN_NETRXF_gso_prefix | XEN_NETRXF_more_data;
595
596 resp->offset = queue->meta[npo.meta_cons].gso_size;
597 resp->id = queue->meta[npo.meta_cons].id;
598 resp->status = XENVIF_RX_CB(skb)->meta_slots_used;
599
600 npo.meta_cons++;
601 XENVIF_RX_CB(skb)->meta_slots_used--;
602 }
603
604
605 queue->stats.tx_bytes += skb->len;
606 queue->stats.tx_packets++;
607
608 status = xenvif_check_gop(queue->vif,
609 XENVIF_RX_CB(skb)->meta_slots_used,
610 &npo);
611
612 if (XENVIF_RX_CB(skb)->meta_slots_used == 1)
613 flags = 0;
614 else
615 flags = XEN_NETRXF_more_data;
616
617 if (skb->ip_summed == CHECKSUM_PARTIAL) /* local packet? */
618 flags |= XEN_NETRXF_csum_blank | XEN_NETRXF_data_validated;
619 else if (skb->ip_summed == CHECKSUM_UNNECESSARY)
620 /* remote but checksummed. */
621 flags |= XEN_NETRXF_data_validated;
622
623 offset = 0;
624 resp = make_rx_response(queue, queue->meta[npo.meta_cons].id,
625 status, offset,
626 queue->meta[npo.meta_cons].size,
627 flags);
628
629 if ((1 << queue->meta[npo.meta_cons].gso_type) &
630 queue->vif->gso_mask) {
631 struct xen_netif_extra_info *gso =
632 (struct xen_netif_extra_info *)
633 RING_GET_RESPONSE(&queue->rx,
634 queue->rx.rsp_prod_pvt++);
635
636 resp->flags |= XEN_NETRXF_extra_info;
637
638 gso->u.gso.type = queue->meta[npo.meta_cons].gso_type;
639 gso->u.gso.size = queue->meta[npo.meta_cons].gso_size;
640 gso->u.gso.pad = 0;
641 gso->u.gso.features = 0;
642
643 gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
644 gso->flags = 0;
645 }
646
647 xenvif_add_frag_responses(queue, status,
648 queue->meta + npo.meta_cons + 1,
649 XENVIF_RX_CB(skb)->meta_slots_used);
650
651 RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&queue->rx, ret);
652
653 need_to_notify |= !!ret;
654
655 npo.meta_cons += XENVIF_RX_CB(skb)->meta_slots_used;
656 dev_kfree_skb(skb);
657 }
658
659 done:
660 if (need_to_notify)
661 notify_remote_via_irq(queue->rx_irq);
662 }
663
xenvif_napi_schedule_or_enable_events(struct xenvif_queue * queue)664 void xenvif_napi_schedule_or_enable_events(struct xenvif_queue *queue)
665 {
666 int more_to_do;
667
668 RING_FINAL_CHECK_FOR_REQUESTS(&queue->tx, more_to_do);
669
670 if (more_to_do)
671 napi_schedule(&queue->napi);
672 }
673
tx_add_credit(struct xenvif_queue * queue)674 static void tx_add_credit(struct xenvif_queue *queue)
675 {
676 unsigned long max_burst, max_credit;
677
678 /*
679 * Allow a burst big enough to transmit a jumbo packet of up to 128kB.
680 * Otherwise the interface can seize up due to insufficient credit.
681 */
682 max_burst = max(131072UL, queue->credit_bytes);
683
684 /* Take care that adding a new chunk of credit doesn't wrap to zero. */
685 max_credit = queue->remaining_credit + queue->credit_bytes;
686 if (max_credit < queue->remaining_credit)
687 max_credit = ULONG_MAX; /* wrapped: clamp to ULONG_MAX */
688
689 queue->remaining_credit = min(max_credit, max_burst);
690 }
691
xenvif_tx_credit_callback(unsigned long data)692 void xenvif_tx_credit_callback(unsigned long data)
693 {
694 struct xenvif_queue *queue = (struct xenvif_queue *)data;
695 tx_add_credit(queue);
696 xenvif_napi_schedule_or_enable_events(queue);
697 }
698
xenvif_tx_err(struct xenvif_queue * queue,struct xen_netif_tx_request * txp,RING_IDX end)699 static void xenvif_tx_err(struct xenvif_queue *queue,
700 struct xen_netif_tx_request *txp, RING_IDX end)
701 {
702 RING_IDX cons = queue->tx.req_cons;
703 unsigned long flags;
704
705 do {
706 spin_lock_irqsave(&queue->response_lock, flags);
707 make_tx_response(queue, txp, XEN_NETIF_RSP_ERROR);
708 push_tx_responses(queue);
709 spin_unlock_irqrestore(&queue->response_lock, flags);
710 if (cons == end)
711 break;
712 RING_COPY_REQUEST(&queue->tx, cons++, txp);
713 } while (1);
714 queue->tx.req_cons = cons;
715 }
716
xenvif_fatal_tx_err(struct xenvif * vif)717 static void xenvif_fatal_tx_err(struct xenvif *vif)
718 {
719 netdev_err(vif->dev, "fatal error; disabling device\n");
720 vif->disabled = true;
721 /* Disable the vif from queue 0's kthread */
722 if (vif->queues)
723 xenvif_kick_thread(&vif->queues[0]);
724 }
725
xenvif_count_requests(struct xenvif_queue * queue,struct xen_netif_tx_request * first,struct xen_netif_tx_request * txp,int work_to_do)726 static int xenvif_count_requests(struct xenvif_queue *queue,
727 struct xen_netif_tx_request *first,
728 struct xen_netif_tx_request *txp,
729 int work_to_do)
730 {
731 RING_IDX cons = queue->tx.req_cons;
732 int slots = 0;
733 int drop_err = 0;
734 int more_data;
735
736 if (!(first->flags & XEN_NETTXF_more_data))
737 return 0;
738
739 do {
740 struct xen_netif_tx_request dropped_tx = { 0 };
741
742 if (slots >= work_to_do) {
743 netdev_err(queue->vif->dev,
744 "Asked for %d slots but exceeds this limit\n",
745 work_to_do);
746 xenvif_fatal_tx_err(queue->vif);
747 return -ENODATA;
748 }
749
750 /* This guest is really using too many slots and
751 * considered malicious.
752 */
753 if (unlikely(slots >= fatal_skb_slots)) {
754 netdev_err(queue->vif->dev,
755 "Malicious frontend using %d slots, threshold %u\n",
756 slots, fatal_skb_slots);
757 xenvif_fatal_tx_err(queue->vif);
758 return -E2BIG;
759 }
760
761 /* Xen network protocol had implicit dependency on
762 * MAX_SKB_FRAGS. XEN_NETBK_LEGACY_SLOTS_MAX is set to
763 * the historical MAX_SKB_FRAGS value 18 to honor the
764 * same behavior as before. Any packet using more than
765 * 18 slots but less than fatal_skb_slots slots is
766 * dropped
767 */
768 if (!drop_err && slots >= XEN_NETBK_LEGACY_SLOTS_MAX) {
769 if (net_ratelimit())
770 netdev_dbg(queue->vif->dev,
771 "Too many slots (%d) exceeding limit (%d), dropping packet\n",
772 slots, XEN_NETBK_LEGACY_SLOTS_MAX);
773 drop_err = -E2BIG;
774 }
775
776 if (drop_err)
777 txp = &dropped_tx;
778
779 RING_COPY_REQUEST(&queue->tx, cons + slots, txp);
780
781 /* If the guest submitted a frame >= 64 KiB then
782 * first->size overflowed and following slots will
783 * appear to be larger than the frame.
784 *
785 * This cannot be fatal error as there are buggy
786 * frontends that do this.
787 *
788 * Consume all slots and drop the packet.
789 */
790 if (!drop_err && txp->size > first->size) {
791 if (net_ratelimit())
792 netdev_dbg(queue->vif->dev,
793 "Invalid tx request, slot size %u > remaining size %u\n",
794 txp->size, first->size);
795 drop_err = -EIO;
796 }
797
798 first->size -= txp->size;
799 slots++;
800
801 if (unlikely((txp->offset + txp->size) > XEN_PAGE_SIZE)) {
802 netdev_err(queue->vif->dev, "Cross page boundary, txp->offset: %u, size: %u\n",
803 txp->offset, txp->size);
804 xenvif_fatal_tx_err(queue->vif);
805 return -EINVAL;
806 }
807
808 more_data = txp->flags & XEN_NETTXF_more_data;
809
810 if (!drop_err)
811 txp++;
812
813 } while (more_data);
814
815 if (drop_err) {
816 xenvif_tx_err(queue, first, cons + slots);
817 return drop_err;
818 }
819
820 return slots;
821 }
822
823
824 struct xenvif_tx_cb {
825 u16 pending_idx;
826 };
827
828 #define XENVIF_TX_CB(skb) ((struct xenvif_tx_cb *)(skb)->cb)
829
xenvif_tx_create_map_op(struct xenvif_queue * queue,u16 pending_idx,struct xen_netif_tx_request * txp,struct gnttab_map_grant_ref * mop)830 static inline void xenvif_tx_create_map_op(struct xenvif_queue *queue,
831 u16 pending_idx,
832 struct xen_netif_tx_request *txp,
833 struct gnttab_map_grant_ref *mop)
834 {
835 queue->pages_to_map[mop-queue->tx_map_ops] = queue->mmap_pages[pending_idx];
836 gnttab_set_map_op(mop, idx_to_kaddr(queue, pending_idx),
837 GNTMAP_host_map | GNTMAP_readonly,
838 txp->gref, queue->vif->domid);
839
840 memcpy(&queue->pending_tx_info[pending_idx].req, txp,
841 sizeof(*txp));
842 }
843
xenvif_alloc_skb(unsigned int size)844 static inline struct sk_buff *xenvif_alloc_skb(unsigned int size)
845 {
846 struct sk_buff *skb =
847 alloc_skb(size + NET_SKB_PAD + NET_IP_ALIGN,
848 GFP_ATOMIC | __GFP_NOWARN);
849 if (unlikely(skb == NULL))
850 return NULL;
851
852 /* Packets passed to netif_rx() must have some headroom. */
853 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
854
855 /* Initialize it here to avoid later surprises */
856 skb_shinfo(skb)->destructor_arg = NULL;
857
858 return skb;
859 }
860
xenvif_get_requests(struct xenvif_queue * queue,struct sk_buff * skb,struct xen_netif_tx_request * txp,struct gnttab_map_grant_ref * gop,unsigned int frag_overflow,struct sk_buff * nskb)861 static struct gnttab_map_grant_ref *xenvif_get_requests(struct xenvif_queue *queue,
862 struct sk_buff *skb,
863 struct xen_netif_tx_request *txp,
864 struct gnttab_map_grant_ref *gop,
865 unsigned int frag_overflow,
866 struct sk_buff *nskb)
867 {
868 struct skb_shared_info *shinfo = skb_shinfo(skb);
869 skb_frag_t *frags = shinfo->frags;
870 u16 pending_idx = XENVIF_TX_CB(skb)->pending_idx;
871 int start;
872 pending_ring_idx_t index;
873 unsigned int nr_slots;
874
875 nr_slots = shinfo->nr_frags;
876
877 /* Skip first skb fragment if it is on same page as header fragment. */
878 start = (frag_get_pending_idx(&shinfo->frags[0]) == pending_idx);
879
880 for (shinfo->nr_frags = start; shinfo->nr_frags < nr_slots;
881 shinfo->nr_frags++, txp++, gop++) {
882 index = pending_index(queue->pending_cons++);
883 pending_idx = queue->pending_ring[index];
884 xenvif_tx_create_map_op(queue, pending_idx, txp, gop);
885 frag_set_pending_idx(&frags[shinfo->nr_frags], pending_idx);
886 }
887
888 if (frag_overflow) {
889
890 shinfo = skb_shinfo(nskb);
891 frags = shinfo->frags;
892
893 for (shinfo->nr_frags = 0; shinfo->nr_frags < frag_overflow;
894 shinfo->nr_frags++, txp++, gop++) {
895 index = pending_index(queue->pending_cons++);
896 pending_idx = queue->pending_ring[index];
897 xenvif_tx_create_map_op(queue, pending_idx, txp, gop);
898 frag_set_pending_idx(&frags[shinfo->nr_frags],
899 pending_idx);
900 }
901
902 skb_shinfo(skb)->frag_list = nskb;
903 }
904
905 return gop;
906 }
907
xenvif_grant_handle_set(struct xenvif_queue * queue,u16 pending_idx,grant_handle_t handle)908 static inline void xenvif_grant_handle_set(struct xenvif_queue *queue,
909 u16 pending_idx,
910 grant_handle_t handle)
911 {
912 if (unlikely(queue->grant_tx_handle[pending_idx] !=
913 NETBACK_INVALID_HANDLE)) {
914 netdev_err(queue->vif->dev,
915 "Trying to overwrite active handle! pending_idx: 0x%x\n",
916 pending_idx);
917 BUG();
918 }
919 queue->grant_tx_handle[pending_idx] = handle;
920 }
921
xenvif_grant_handle_reset(struct xenvif_queue * queue,u16 pending_idx)922 static inline void xenvif_grant_handle_reset(struct xenvif_queue *queue,
923 u16 pending_idx)
924 {
925 if (unlikely(queue->grant_tx_handle[pending_idx] ==
926 NETBACK_INVALID_HANDLE)) {
927 netdev_err(queue->vif->dev,
928 "Trying to unmap invalid handle! pending_idx: 0x%x\n",
929 pending_idx);
930 BUG();
931 }
932 queue->grant_tx_handle[pending_idx] = NETBACK_INVALID_HANDLE;
933 }
934
xenvif_tx_check_gop(struct xenvif_queue * queue,struct sk_buff * skb,struct gnttab_map_grant_ref ** gopp_map,struct gnttab_copy ** gopp_copy)935 static int xenvif_tx_check_gop(struct xenvif_queue *queue,
936 struct sk_buff *skb,
937 struct gnttab_map_grant_ref **gopp_map,
938 struct gnttab_copy **gopp_copy)
939 {
940 struct gnttab_map_grant_ref *gop_map = *gopp_map;
941 u16 pending_idx = XENVIF_TX_CB(skb)->pending_idx;
942 /* This always points to the shinfo of the skb being checked, which
943 * could be either the first or the one on the frag_list
944 */
945 struct skb_shared_info *shinfo = skb_shinfo(skb);
946 /* If this is non-NULL, we are currently checking the frag_list skb, and
947 * this points to the shinfo of the first one
948 */
949 struct skb_shared_info *first_shinfo = NULL;
950 int nr_frags = shinfo->nr_frags;
951 const bool sharedslot = nr_frags &&
952 frag_get_pending_idx(&shinfo->frags[0]) == pending_idx;
953 int i, err;
954
955 /* Check status of header. */
956 err = (*gopp_copy)->status;
957 if (unlikely(err)) {
958 if (net_ratelimit())
959 netdev_dbg(queue->vif->dev,
960 "Grant copy of header failed! status: %d pending_idx: %u ref: %u\n",
961 (*gopp_copy)->status,
962 pending_idx,
963 (*gopp_copy)->source.u.ref);
964 /* The first frag might still have this slot mapped */
965 if (!sharedslot)
966 xenvif_idx_release(queue, pending_idx,
967 XEN_NETIF_RSP_ERROR);
968 }
969 (*gopp_copy)++;
970
971 check_frags:
972 for (i = 0; i < nr_frags; i++, gop_map++) {
973 int j, newerr;
974
975 pending_idx = frag_get_pending_idx(&shinfo->frags[i]);
976
977 /* Check error status: if okay then remember grant handle. */
978 newerr = gop_map->status;
979
980 if (likely(!newerr)) {
981 xenvif_grant_handle_set(queue,
982 pending_idx,
983 gop_map->handle);
984 /* Had a previous error? Invalidate this fragment. */
985 if (unlikely(err)) {
986 xenvif_idx_unmap(queue, pending_idx);
987 /* If the mapping of the first frag was OK, but
988 * the header's copy failed, and they are
989 * sharing a slot, send an error
990 */
991 if (i == 0 && sharedslot)
992 xenvif_idx_release(queue, pending_idx,
993 XEN_NETIF_RSP_ERROR);
994 else
995 xenvif_idx_release(queue, pending_idx,
996 XEN_NETIF_RSP_OKAY);
997 }
998 continue;
999 }
1000
1001 /* Error on this fragment: respond to client with an error. */
1002 if (net_ratelimit())
1003 netdev_dbg(queue->vif->dev,
1004 "Grant map of %d. frag failed! status: %d pending_idx: %u ref: %u\n",
1005 i,
1006 gop_map->status,
1007 pending_idx,
1008 gop_map->ref);
1009
1010 xenvif_idx_release(queue, pending_idx, XEN_NETIF_RSP_ERROR);
1011
1012 /* Not the first error? Preceding frags already invalidated. */
1013 if (err)
1014 continue;
1015
1016 /* First error: if the header haven't shared a slot with the
1017 * first frag, release it as well.
1018 */
1019 if (!sharedslot)
1020 xenvif_idx_release(queue,
1021 XENVIF_TX_CB(skb)->pending_idx,
1022 XEN_NETIF_RSP_OKAY);
1023
1024 /* Invalidate preceding fragments of this skb. */
1025 for (j = 0; j < i; j++) {
1026 pending_idx = frag_get_pending_idx(&shinfo->frags[j]);
1027 xenvif_idx_unmap(queue, pending_idx);
1028 xenvif_idx_release(queue, pending_idx,
1029 XEN_NETIF_RSP_OKAY);
1030 }
1031
1032 /* And if we found the error while checking the frag_list, unmap
1033 * the first skb's frags
1034 */
1035 if (first_shinfo) {
1036 for (j = 0; j < first_shinfo->nr_frags; j++) {
1037 pending_idx = frag_get_pending_idx(&first_shinfo->frags[j]);
1038 xenvif_idx_unmap(queue, pending_idx);
1039 xenvif_idx_release(queue, pending_idx,
1040 XEN_NETIF_RSP_OKAY);
1041 }
1042 }
1043
1044 /* Remember the error: invalidate all subsequent fragments. */
1045 err = newerr;
1046 }
1047
1048 if (skb_has_frag_list(skb) && !first_shinfo) {
1049 first_shinfo = skb_shinfo(skb);
1050 shinfo = skb_shinfo(skb_shinfo(skb)->frag_list);
1051 nr_frags = shinfo->nr_frags;
1052
1053 goto check_frags;
1054 }
1055
1056 *gopp_map = gop_map;
1057 return err;
1058 }
1059
xenvif_fill_frags(struct xenvif_queue * queue,struct sk_buff * skb)1060 static void xenvif_fill_frags(struct xenvif_queue *queue, struct sk_buff *skb)
1061 {
1062 struct skb_shared_info *shinfo = skb_shinfo(skb);
1063 int nr_frags = shinfo->nr_frags;
1064 int i;
1065 u16 prev_pending_idx = INVALID_PENDING_IDX;
1066
1067 for (i = 0; i < nr_frags; i++) {
1068 skb_frag_t *frag = shinfo->frags + i;
1069 struct xen_netif_tx_request *txp;
1070 struct page *page;
1071 u16 pending_idx;
1072
1073 pending_idx = frag_get_pending_idx(frag);
1074
1075 /* If this is not the first frag, chain it to the previous*/
1076 if (prev_pending_idx == INVALID_PENDING_IDX)
1077 skb_shinfo(skb)->destructor_arg =
1078 &callback_param(queue, pending_idx);
1079 else
1080 callback_param(queue, prev_pending_idx).ctx =
1081 &callback_param(queue, pending_idx);
1082
1083 callback_param(queue, pending_idx).ctx = NULL;
1084 prev_pending_idx = pending_idx;
1085
1086 txp = &queue->pending_tx_info[pending_idx].req;
1087 page = virt_to_page(idx_to_kaddr(queue, pending_idx));
1088 __skb_fill_page_desc(skb, i, page, txp->offset, txp->size);
1089 skb->len += txp->size;
1090 skb->data_len += txp->size;
1091 skb->truesize += txp->size;
1092
1093 /* Take an extra reference to offset network stack's put_page */
1094 get_page(queue->mmap_pages[pending_idx]);
1095 }
1096 }
1097
xenvif_get_extras(struct xenvif_queue * queue,struct xen_netif_extra_info * extras,int work_to_do)1098 static int xenvif_get_extras(struct xenvif_queue *queue,
1099 struct xen_netif_extra_info *extras,
1100 int work_to_do)
1101 {
1102 struct xen_netif_extra_info extra;
1103 RING_IDX cons = queue->tx.req_cons;
1104
1105 do {
1106 if (unlikely(work_to_do-- <= 0)) {
1107 netdev_err(queue->vif->dev, "Missing extra info\n");
1108 xenvif_fatal_tx_err(queue->vif);
1109 return -EBADR;
1110 }
1111
1112 RING_COPY_REQUEST(&queue->tx, cons, &extra);
1113 if (unlikely(!extra.type ||
1114 extra.type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
1115 queue->tx.req_cons = ++cons;
1116 netdev_err(queue->vif->dev,
1117 "Invalid extra type: %d\n", extra.type);
1118 xenvif_fatal_tx_err(queue->vif);
1119 return -EINVAL;
1120 }
1121
1122 memcpy(&extras[extra.type - 1], &extra, sizeof(extra));
1123 queue->tx.req_cons = ++cons;
1124 } while (extra.flags & XEN_NETIF_EXTRA_FLAG_MORE);
1125
1126 return work_to_do;
1127 }
1128
xenvif_set_skb_gso(struct xenvif * vif,struct sk_buff * skb,struct xen_netif_extra_info * gso)1129 static int xenvif_set_skb_gso(struct xenvif *vif,
1130 struct sk_buff *skb,
1131 struct xen_netif_extra_info *gso)
1132 {
1133 if (!gso->u.gso.size) {
1134 netdev_err(vif->dev, "GSO size must not be zero.\n");
1135 xenvif_fatal_tx_err(vif);
1136 return -EINVAL;
1137 }
1138
1139 switch (gso->u.gso.type) {
1140 case XEN_NETIF_GSO_TYPE_TCPV4:
1141 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1142 break;
1143 case XEN_NETIF_GSO_TYPE_TCPV6:
1144 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1145 break;
1146 default:
1147 netdev_err(vif->dev, "Bad GSO type %d.\n", gso->u.gso.type);
1148 xenvif_fatal_tx_err(vif);
1149 return -EINVAL;
1150 }
1151
1152 skb_shinfo(skb)->gso_size = gso->u.gso.size;
1153 /* gso_segs will be calculated later */
1154
1155 return 0;
1156 }
1157
checksum_setup(struct xenvif_queue * queue,struct sk_buff * skb)1158 static int checksum_setup(struct xenvif_queue *queue, struct sk_buff *skb)
1159 {
1160 bool recalculate_partial_csum = false;
1161
1162 /* A GSO SKB must be CHECKSUM_PARTIAL. However some buggy
1163 * peers can fail to set NETRXF_csum_blank when sending a GSO
1164 * frame. In this case force the SKB to CHECKSUM_PARTIAL and
1165 * recalculate the partial checksum.
1166 */
1167 if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) {
1168 queue->stats.rx_gso_checksum_fixup++;
1169 skb->ip_summed = CHECKSUM_PARTIAL;
1170 recalculate_partial_csum = true;
1171 }
1172
1173 /* A non-CHECKSUM_PARTIAL SKB does not require setup. */
1174 if (skb->ip_summed != CHECKSUM_PARTIAL)
1175 return 0;
1176
1177 return skb_checksum_setup(skb, recalculate_partial_csum);
1178 }
1179
tx_credit_exceeded(struct xenvif_queue * queue,unsigned size)1180 static bool tx_credit_exceeded(struct xenvif_queue *queue, unsigned size)
1181 {
1182 u64 now = get_jiffies_64();
1183 u64 next_credit = queue->credit_window_start +
1184 msecs_to_jiffies(queue->credit_usec / 1000);
1185
1186 /* Timer could already be pending in rare cases. */
1187 if (timer_pending(&queue->credit_timeout))
1188 return true;
1189
1190 /* Passed the point where we can replenish credit? */
1191 if (time_after_eq64(now, next_credit)) {
1192 queue->credit_window_start = now;
1193 tx_add_credit(queue);
1194 }
1195
1196 /* Still too big to send right now? Set a callback. */
1197 if (size > queue->remaining_credit) {
1198 queue->credit_timeout.data =
1199 (unsigned long)queue;
1200 mod_timer(&queue->credit_timeout,
1201 next_credit);
1202 queue->credit_window_start = next_credit;
1203
1204 return true;
1205 }
1206
1207 return false;
1208 }
1209
1210 /* No locking is required in xenvif_mcast_add/del() as they are
1211 * only ever invoked from NAPI poll. An RCU list is used because
1212 * xenvif_mcast_match() is called asynchronously, during start_xmit.
1213 */
1214
xenvif_mcast_add(struct xenvif * vif,const u8 * addr)1215 static int xenvif_mcast_add(struct xenvif *vif, const u8 *addr)
1216 {
1217 struct xenvif_mcast_addr *mcast;
1218
1219 if (vif->fe_mcast_count == XEN_NETBK_MCAST_MAX) {
1220 if (net_ratelimit())
1221 netdev_err(vif->dev,
1222 "Too many multicast addresses\n");
1223 return -ENOSPC;
1224 }
1225
1226 mcast = kzalloc(sizeof(*mcast), GFP_ATOMIC);
1227 if (!mcast)
1228 return -ENOMEM;
1229
1230 ether_addr_copy(mcast->addr, addr);
1231 list_add_tail_rcu(&mcast->entry, &vif->fe_mcast_addr);
1232 vif->fe_mcast_count++;
1233
1234 return 0;
1235 }
1236
xenvif_mcast_del(struct xenvif * vif,const u8 * addr)1237 static void xenvif_mcast_del(struct xenvif *vif, const u8 *addr)
1238 {
1239 struct xenvif_mcast_addr *mcast;
1240
1241 list_for_each_entry_rcu(mcast, &vif->fe_mcast_addr, entry) {
1242 if (ether_addr_equal(addr, mcast->addr)) {
1243 --vif->fe_mcast_count;
1244 list_del_rcu(&mcast->entry);
1245 kfree_rcu(mcast, rcu);
1246 break;
1247 }
1248 }
1249 }
1250
xenvif_mcast_match(struct xenvif * vif,const u8 * addr)1251 bool xenvif_mcast_match(struct xenvif *vif, const u8 *addr)
1252 {
1253 struct xenvif_mcast_addr *mcast;
1254
1255 rcu_read_lock();
1256 list_for_each_entry_rcu(mcast, &vif->fe_mcast_addr, entry) {
1257 if (ether_addr_equal(addr, mcast->addr)) {
1258 rcu_read_unlock();
1259 return true;
1260 }
1261 }
1262 rcu_read_unlock();
1263
1264 return false;
1265 }
1266
xenvif_mcast_addr_list_free(struct xenvif * vif)1267 void xenvif_mcast_addr_list_free(struct xenvif *vif)
1268 {
1269 /* No need for locking or RCU here. NAPI poll and TX queue
1270 * are stopped.
1271 */
1272 while (!list_empty(&vif->fe_mcast_addr)) {
1273 struct xenvif_mcast_addr *mcast;
1274
1275 mcast = list_first_entry(&vif->fe_mcast_addr,
1276 struct xenvif_mcast_addr,
1277 entry);
1278 --vif->fe_mcast_count;
1279 list_del(&mcast->entry);
1280 kfree(mcast);
1281 }
1282 }
1283
xenvif_tx_build_gops(struct xenvif_queue * queue,int budget,unsigned * copy_ops,unsigned * map_ops)1284 static void xenvif_tx_build_gops(struct xenvif_queue *queue,
1285 int budget,
1286 unsigned *copy_ops,
1287 unsigned *map_ops)
1288 {
1289 struct gnttab_map_grant_ref *gop = queue->tx_map_ops;
1290 struct sk_buff *skb, *nskb;
1291 int ret;
1292 unsigned int frag_overflow;
1293
1294 while (skb_queue_len(&queue->tx_queue) < budget) {
1295 struct xen_netif_tx_request txreq;
1296 struct xen_netif_tx_request txfrags[XEN_NETBK_LEGACY_SLOTS_MAX];
1297 struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX-1];
1298 u16 pending_idx;
1299 RING_IDX idx;
1300 int work_to_do;
1301 unsigned int data_len;
1302 pending_ring_idx_t index;
1303
1304 if (queue->tx.sring->req_prod - queue->tx.req_cons >
1305 XEN_NETIF_TX_RING_SIZE) {
1306 netdev_err(queue->vif->dev,
1307 "Impossible number of requests. "
1308 "req_prod %d, req_cons %d, size %ld\n",
1309 queue->tx.sring->req_prod, queue->tx.req_cons,
1310 XEN_NETIF_TX_RING_SIZE);
1311 xenvif_fatal_tx_err(queue->vif);
1312 break;
1313 }
1314
1315 work_to_do = RING_HAS_UNCONSUMED_REQUESTS(&queue->tx);
1316 if (!work_to_do)
1317 break;
1318
1319 idx = queue->tx.req_cons;
1320 rmb(); /* Ensure that we see the request before we copy it. */
1321 RING_COPY_REQUEST(&queue->tx, idx, &txreq);
1322
1323 /* Credit-based scheduling. */
1324 if (txreq.size > queue->remaining_credit &&
1325 tx_credit_exceeded(queue, txreq.size))
1326 break;
1327
1328 queue->remaining_credit -= txreq.size;
1329
1330 work_to_do--;
1331 queue->tx.req_cons = ++idx;
1332
1333 memset(extras, 0, sizeof(extras));
1334 if (txreq.flags & XEN_NETTXF_extra_info) {
1335 work_to_do = xenvif_get_extras(queue, extras,
1336 work_to_do);
1337 idx = queue->tx.req_cons;
1338 if (unlikely(work_to_do < 0))
1339 break;
1340 }
1341
1342 if (extras[XEN_NETIF_EXTRA_TYPE_MCAST_ADD - 1].type) {
1343 struct xen_netif_extra_info *extra;
1344
1345 extra = &extras[XEN_NETIF_EXTRA_TYPE_MCAST_ADD - 1];
1346 ret = xenvif_mcast_add(queue->vif, extra->u.mcast.addr);
1347
1348 make_tx_response(queue, &txreq,
1349 (ret == 0) ?
1350 XEN_NETIF_RSP_OKAY :
1351 XEN_NETIF_RSP_ERROR);
1352 push_tx_responses(queue);
1353 continue;
1354 }
1355
1356 if (extras[XEN_NETIF_EXTRA_TYPE_MCAST_DEL - 1].type) {
1357 struct xen_netif_extra_info *extra;
1358
1359 extra = &extras[XEN_NETIF_EXTRA_TYPE_MCAST_DEL - 1];
1360 xenvif_mcast_del(queue->vif, extra->u.mcast.addr);
1361
1362 make_tx_response(queue, &txreq, XEN_NETIF_RSP_OKAY);
1363 push_tx_responses(queue);
1364 continue;
1365 }
1366
1367 ret = xenvif_count_requests(queue, &txreq, txfrags, work_to_do);
1368 if (unlikely(ret < 0))
1369 break;
1370
1371 idx += ret;
1372
1373 if (unlikely(txreq.size < ETH_HLEN)) {
1374 netdev_dbg(queue->vif->dev,
1375 "Bad packet size: %d\n", txreq.size);
1376 xenvif_tx_err(queue, &txreq, idx);
1377 break;
1378 }
1379
1380 /* No crossing a page as the payload mustn't fragment. */
1381 if (unlikely((txreq.offset + txreq.size) > XEN_PAGE_SIZE)) {
1382 netdev_err(queue->vif->dev,
1383 "txreq.offset: %u, size: %u, end: %lu\n",
1384 txreq.offset, txreq.size,
1385 (unsigned long)(txreq.offset&~XEN_PAGE_MASK) + txreq.size);
1386 xenvif_fatal_tx_err(queue->vif);
1387 break;
1388 }
1389
1390 index = pending_index(queue->pending_cons);
1391 pending_idx = queue->pending_ring[index];
1392
1393 data_len = (txreq.size > XEN_NETBACK_TX_COPY_LEN &&
1394 ret < XEN_NETBK_LEGACY_SLOTS_MAX) ?
1395 XEN_NETBACK_TX_COPY_LEN : txreq.size;
1396
1397 skb = xenvif_alloc_skb(data_len);
1398 if (unlikely(skb == NULL)) {
1399 netdev_dbg(queue->vif->dev,
1400 "Can't allocate a skb in start_xmit.\n");
1401 xenvif_tx_err(queue, &txreq, idx);
1402 break;
1403 }
1404
1405 skb_shinfo(skb)->nr_frags = ret;
1406 if (data_len < txreq.size)
1407 skb_shinfo(skb)->nr_frags++;
1408 /* At this point shinfo->nr_frags is in fact the number of
1409 * slots, which can be as large as XEN_NETBK_LEGACY_SLOTS_MAX.
1410 */
1411 frag_overflow = 0;
1412 nskb = NULL;
1413 if (skb_shinfo(skb)->nr_frags > MAX_SKB_FRAGS) {
1414 frag_overflow = skb_shinfo(skb)->nr_frags - MAX_SKB_FRAGS;
1415 BUG_ON(frag_overflow > MAX_SKB_FRAGS);
1416 skb_shinfo(skb)->nr_frags = MAX_SKB_FRAGS;
1417 nskb = xenvif_alloc_skb(0);
1418 if (unlikely(nskb == NULL)) {
1419 kfree_skb(skb);
1420 xenvif_tx_err(queue, &txreq, idx);
1421 if (net_ratelimit())
1422 netdev_err(queue->vif->dev,
1423 "Can't allocate the frag_list skb.\n");
1424 break;
1425 }
1426 }
1427
1428 if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) {
1429 struct xen_netif_extra_info *gso;
1430 gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1];
1431
1432 if (xenvif_set_skb_gso(queue->vif, skb, gso)) {
1433 /* Failure in xenvif_set_skb_gso is fatal. */
1434 kfree_skb(skb);
1435 kfree_skb(nskb);
1436 break;
1437 }
1438 }
1439
1440 XENVIF_TX_CB(skb)->pending_idx = pending_idx;
1441
1442 __skb_put(skb, data_len);
1443 queue->tx_copy_ops[*copy_ops].source.u.ref = txreq.gref;
1444 queue->tx_copy_ops[*copy_ops].source.domid = queue->vif->domid;
1445 queue->tx_copy_ops[*copy_ops].source.offset = txreq.offset;
1446
1447 queue->tx_copy_ops[*copy_ops].dest.u.gmfn =
1448 virt_to_gfn(skb->data);
1449 queue->tx_copy_ops[*copy_ops].dest.domid = DOMID_SELF;
1450 queue->tx_copy_ops[*copy_ops].dest.offset =
1451 offset_in_page(skb->data) & ~XEN_PAGE_MASK;
1452
1453 queue->tx_copy_ops[*copy_ops].len = data_len;
1454 queue->tx_copy_ops[*copy_ops].flags = GNTCOPY_source_gref;
1455
1456 (*copy_ops)++;
1457
1458 if (data_len < txreq.size) {
1459 frag_set_pending_idx(&skb_shinfo(skb)->frags[0],
1460 pending_idx);
1461 xenvif_tx_create_map_op(queue, pending_idx, &txreq, gop);
1462 gop++;
1463 } else {
1464 frag_set_pending_idx(&skb_shinfo(skb)->frags[0],
1465 INVALID_PENDING_IDX);
1466 memcpy(&queue->pending_tx_info[pending_idx].req, &txreq,
1467 sizeof(txreq));
1468 }
1469
1470 queue->pending_cons++;
1471
1472 gop = xenvif_get_requests(queue, skb, txfrags, gop,
1473 frag_overflow, nskb);
1474
1475 __skb_queue_tail(&queue->tx_queue, skb);
1476
1477 queue->tx.req_cons = idx;
1478
1479 if (((gop-queue->tx_map_ops) >= ARRAY_SIZE(queue->tx_map_ops)) ||
1480 (*copy_ops >= ARRAY_SIZE(queue->tx_copy_ops)))
1481 break;
1482 }
1483
1484 (*map_ops) = gop - queue->tx_map_ops;
1485 return;
1486 }
1487
1488 /* Consolidate skb with a frag_list into a brand new one with local pages on
1489 * frags. Returns 0 or -ENOMEM if can't allocate new pages.
1490 */
xenvif_handle_frag_list(struct xenvif_queue * queue,struct sk_buff * skb)1491 static int xenvif_handle_frag_list(struct xenvif_queue *queue, struct sk_buff *skb)
1492 {
1493 unsigned int offset = skb_headlen(skb);
1494 skb_frag_t frags[MAX_SKB_FRAGS];
1495 int i, f;
1496 struct ubuf_info *uarg;
1497 struct sk_buff *nskb = skb_shinfo(skb)->frag_list;
1498
1499 queue->stats.tx_zerocopy_sent += 2;
1500 queue->stats.tx_frag_overflow++;
1501
1502 xenvif_fill_frags(queue, nskb);
1503 /* Subtract frags size, we will correct it later */
1504 skb->truesize -= skb->data_len;
1505 skb->len += nskb->len;
1506 skb->data_len += nskb->len;
1507
1508 /* create a brand new frags array and coalesce there */
1509 for (i = 0; offset < skb->len; i++) {
1510 struct page *page;
1511 unsigned int len;
1512
1513 BUG_ON(i >= MAX_SKB_FRAGS);
1514 page = alloc_page(GFP_ATOMIC);
1515 if (!page) {
1516 int j;
1517 skb->truesize += skb->data_len;
1518 for (j = 0; j < i; j++)
1519 put_page(frags[j].page.p);
1520 return -ENOMEM;
1521 }
1522
1523 if (offset + PAGE_SIZE < skb->len)
1524 len = PAGE_SIZE;
1525 else
1526 len = skb->len - offset;
1527 if (skb_copy_bits(skb, offset, page_address(page), len))
1528 BUG();
1529
1530 offset += len;
1531 frags[i].page.p = page;
1532 frags[i].page_offset = 0;
1533 skb_frag_size_set(&frags[i], len);
1534 }
1535
1536 /* Copied all the bits from the frag list -- free it. */
1537 skb_frag_list_init(skb);
1538 xenvif_skb_zerocopy_prepare(queue, nskb);
1539 kfree_skb(nskb);
1540
1541 /* Release all the original (foreign) frags. */
1542 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
1543 skb_frag_unref(skb, f);
1544 uarg = skb_shinfo(skb)->destructor_arg;
1545 /* increase inflight counter to offset decrement in callback */
1546 atomic_inc(&queue->inflight_packets);
1547 uarg->callback(uarg, true);
1548 skb_shinfo(skb)->destructor_arg = NULL;
1549
1550 /* Fill the skb with the new (local) frags. */
1551 memcpy(skb_shinfo(skb)->frags, frags, i * sizeof(skb_frag_t));
1552 skb_shinfo(skb)->nr_frags = i;
1553 skb->truesize += i * PAGE_SIZE;
1554
1555 return 0;
1556 }
1557
xenvif_tx_submit(struct xenvif_queue * queue)1558 static int xenvif_tx_submit(struct xenvif_queue *queue)
1559 {
1560 struct gnttab_map_grant_ref *gop_map = queue->tx_map_ops;
1561 struct gnttab_copy *gop_copy = queue->tx_copy_ops;
1562 struct sk_buff *skb;
1563 int work_done = 0;
1564
1565 while ((skb = __skb_dequeue(&queue->tx_queue)) != NULL) {
1566 struct xen_netif_tx_request *txp;
1567 u16 pending_idx;
1568 unsigned data_len;
1569
1570 pending_idx = XENVIF_TX_CB(skb)->pending_idx;
1571 txp = &queue->pending_tx_info[pending_idx].req;
1572
1573 /* Check the remap error code. */
1574 if (unlikely(xenvif_tx_check_gop(queue, skb, &gop_map, &gop_copy))) {
1575 /* If there was an error, xenvif_tx_check_gop is
1576 * expected to release all the frags which were mapped,
1577 * so kfree_skb shouldn't do it again
1578 */
1579 skb_shinfo(skb)->nr_frags = 0;
1580 if (skb_has_frag_list(skb)) {
1581 struct sk_buff *nskb =
1582 skb_shinfo(skb)->frag_list;
1583 skb_shinfo(nskb)->nr_frags = 0;
1584 }
1585 kfree_skb(skb);
1586 continue;
1587 }
1588
1589 data_len = skb->len;
1590 callback_param(queue, pending_idx).ctx = NULL;
1591 if (data_len < txp->size) {
1592 /* Append the packet payload as a fragment. */
1593 txp->offset += data_len;
1594 txp->size -= data_len;
1595 } else {
1596 /* Schedule a response immediately. */
1597 xenvif_idx_release(queue, pending_idx,
1598 XEN_NETIF_RSP_OKAY);
1599 }
1600
1601 if (txp->flags & XEN_NETTXF_csum_blank)
1602 skb->ip_summed = CHECKSUM_PARTIAL;
1603 else if (txp->flags & XEN_NETTXF_data_validated)
1604 skb->ip_summed = CHECKSUM_UNNECESSARY;
1605
1606 xenvif_fill_frags(queue, skb);
1607
1608 if (unlikely(skb_has_frag_list(skb))) {
1609 if (xenvif_handle_frag_list(queue, skb)) {
1610 if (net_ratelimit())
1611 netdev_err(queue->vif->dev,
1612 "Not enough memory to consolidate frag_list!\n");
1613 xenvif_skb_zerocopy_prepare(queue, skb);
1614 kfree_skb(skb);
1615 continue;
1616 }
1617 }
1618
1619 skb->dev = queue->vif->dev;
1620 skb->protocol = eth_type_trans(skb, skb->dev);
1621 skb_reset_network_header(skb);
1622
1623 if (checksum_setup(queue, skb)) {
1624 netdev_dbg(queue->vif->dev,
1625 "Can't setup checksum in net_tx_action\n");
1626 /* We have to set this flag to trigger the callback */
1627 if (skb_shinfo(skb)->destructor_arg)
1628 xenvif_skb_zerocopy_prepare(queue, skb);
1629 kfree_skb(skb);
1630 continue;
1631 }
1632
1633 skb_probe_transport_header(skb, 0);
1634
1635 /* If the packet is GSO then we will have just set up the
1636 * transport header offset in checksum_setup so it's now
1637 * straightforward to calculate gso_segs.
1638 */
1639 if (skb_is_gso(skb)) {
1640 int mss = skb_shinfo(skb)->gso_size;
1641 int hdrlen = skb_transport_header(skb) -
1642 skb_mac_header(skb) +
1643 tcp_hdrlen(skb);
1644
1645 skb_shinfo(skb)->gso_segs =
1646 DIV_ROUND_UP(skb->len - hdrlen, mss);
1647 }
1648
1649 queue->stats.rx_bytes += skb->len;
1650 queue->stats.rx_packets++;
1651
1652 work_done++;
1653
1654 /* Set this flag right before netif_receive_skb, otherwise
1655 * someone might think this packet already left netback, and
1656 * do a skb_copy_ubufs while we are still in control of the
1657 * skb. E.g. the __pskb_pull_tail earlier can do such thing.
1658 */
1659 if (skb_shinfo(skb)->destructor_arg) {
1660 xenvif_skb_zerocopy_prepare(queue, skb);
1661 queue->stats.tx_zerocopy_sent++;
1662 }
1663
1664 netif_receive_skb(skb);
1665 }
1666
1667 return work_done;
1668 }
1669
xenvif_zerocopy_callback(struct ubuf_info * ubuf,bool zerocopy_success)1670 void xenvif_zerocopy_callback(struct ubuf_info *ubuf, bool zerocopy_success)
1671 {
1672 unsigned long flags;
1673 pending_ring_idx_t index;
1674 struct xenvif_queue *queue = ubuf_to_queue(ubuf);
1675
1676 /* This is the only place where we grab this lock, to protect callbacks
1677 * from each other.
1678 */
1679 spin_lock_irqsave(&queue->callback_lock, flags);
1680 do {
1681 u16 pending_idx = ubuf->desc;
1682 ubuf = (struct ubuf_info *) ubuf->ctx;
1683 BUG_ON(queue->dealloc_prod - queue->dealloc_cons >=
1684 MAX_PENDING_REQS);
1685 index = pending_index(queue->dealloc_prod);
1686 queue->dealloc_ring[index] = pending_idx;
1687 /* Sync with xenvif_tx_dealloc_action:
1688 * insert idx then incr producer.
1689 */
1690 smp_wmb();
1691 queue->dealloc_prod++;
1692 } while (ubuf);
1693 spin_unlock_irqrestore(&queue->callback_lock, flags);
1694
1695 if (likely(zerocopy_success))
1696 queue->stats.tx_zerocopy_success++;
1697 else
1698 queue->stats.tx_zerocopy_fail++;
1699 xenvif_skb_zerocopy_complete(queue);
1700 }
1701
xenvif_tx_dealloc_action(struct xenvif_queue * queue)1702 static inline void xenvif_tx_dealloc_action(struct xenvif_queue *queue)
1703 {
1704 struct gnttab_unmap_grant_ref *gop;
1705 pending_ring_idx_t dc, dp;
1706 u16 pending_idx, pending_idx_release[MAX_PENDING_REQS];
1707 unsigned int i = 0;
1708
1709 dc = queue->dealloc_cons;
1710 gop = queue->tx_unmap_ops;
1711
1712 /* Free up any grants we have finished using */
1713 do {
1714 dp = queue->dealloc_prod;
1715
1716 /* Ensure we see all indices enqueued by all
1717 * xenvif_zerocopy_callback().
1718 */
1719 smp_rmb();
1720
1721 while (dc != dp) {
1722 BUG_ON(gop - queue->tx_unmap_ops >= MAX_PENDING_REQS);
1723 pending_idx =
1724 queue->dealloc_ring[pending_index(dc++)];
1725
1726 pending_idx_release[gop - queue->tx_unmap_ops] =
1727 pending_idx;
1728 queue->pages_to_unmap[gop - queue->tx_unmap_ops] =
1729 queue->mmap_pages[pending_idx];
1730 gnttab_set_unmap_op(gop,
1731 idx_to_kaddr(queue, pending_idx),
1732 GNTMAP_host_map,
1733 queue->grant_tx_handle[pending_idx]);
1734 xenvif_grant_handle_reset(queue, pending_idx);
1735 ++gop;
1736 }
1737
1738 } while (dp != queue->dealloc_prod);
1739
1740 queue->dealloc_cons = dc;
1741
1742 if (gop - queue->tx_unmap_ops > 0) {
1743 int ret;
1744 ret = gnttab_unmap_refs(queue->tx_unmap_ops,
1745 NULL,
1746 queue->pages_to_unmap,
1747 gop - queue->tx_unmap_ops);
1748 if (ret) {
1749 netdev_err(queue->vif->dev, "Unmap fail: nr_ops %tu ret %d\n",
1750 gop - queue->tx_unmap_ops, ret);
1751 for (i = 0; i < gop - queue->tx_unmap_ops; ++i) {
1752 if (gop[i].status != GNTST_okay)
1753 netdev_err(queue->vif->dev,
1754 " host_addr: 0x%llx handle: 0x%x status: %d\n",
1755 gop[i].host_addr,
1756 gop[i].handle,
1757 gop[i].status);
1758 }
1759 BUG();
1760 }
1761 }
1762
1763 for (i = 0; i < gop - queue->tx_unmap_ops; ++i)
1764 xenvif_idx_release(queue, pending_idx_release[i],
1765 XEN_NETIF_RSP_OKAY);
1766 }
1767
1768
1769 /* Called after netfront has transmitted */
xenvif_tx_action(struct xenvif_queue * queue,int budget)1770 int xenvif_tx_action(struct xenvif_queue *queue, int budget)
1771 {
1772 unsigned nr_mops, nr_cops = 0;
1773 int work_done, ret;
1774
1775 if (unlikely(!tx_work_todo(queue)))
1776 return 0;
1777
1778 xenvif_tx_build_gops(queue, budget, &nr_cops, &nr_mops);
1779
1780 if (nr_cops == 0)
1781 return 0;
1782
1783 gnttab_batch_copy(queue->tx_copy_ops, nr_cops);
1784 if (nr_mops != 0) {
1785 ret = gnttab_map_refs(queue->tx_map_ops,
1786 NULL,
1787 queue->pages_to_map,
1788 nr_mops);
1789 BUG_ON(ret);
1790 }
1791
1792 work_done = xenvif_tx_submit(queue);
1793
1794 return work_done;
1795 }
1796
xenvif_idx_release(struct xenvif_queue * queue,u16 pending_idx,u8 status)1797 static void xenvif_idx_release(struct xenvif_queue *queue, u16 pending_idx,
1798 u8 status)
1799 {
1800 struct pending_tx_info *pending_tx_info;
1801 pending_ring_idx_t index;
1802 unsigned long flags;
1803
1804 pending_tx_info = &queue->pending_tx_info[pending_idx];
1805
1806 spin_lock_irqsave(&queue->response_lock, flags);
1807
1808 make_tx_response(queue, &pending_tx_info->req, status);
1809
1810 /* Release the pending index before pusing the Tx response so
1811 * its available before a new Tx request is pushed by the
1812 * frontend.
1813 */
1814 index = pending_index(queue->pending_prod++);
1815 queue->pending_ring[index] = pending_idx;
1816
1817 push_tx_responses(queue);
1818
1819 spin_unlock_irqrestore(&queue->response_lock, flags);
1820 }
1821
1822
make_tx_response(struct xenvif_queue * queue,struct xen_netif_tx_request * txp,s8 st)1823 static void make_tx_response(struct xenvif_queue *queue,
1824 struct xen_netif_tx_request *txp,
1825 s8 st)
1826 {
1827 RING_IDX i = queue->tx.rsp_prod_pvt;
1828 struct xen_netif_tx_response *resp;
1829
1830 resp = RING_GET_RESPONSE(&queue->tx, i);
1831 resp->id = txp->id;
1832 resp->status = st;
1833
1834 if (txp->flags & XEN_NETTXF_extra_info)
1835 RING_GET_RESPONSE(&queue->tx, ++i)->status = XEN_NETIF_RSP_NULL;
1836
1837 queue->tx.rsp_prod_pvt = ++i;
1838 }
1839
push_tx_responses(struct xenvif_queue * queue)1840 static void push_tx_responses(struct xenvif_queue *queue)
1841 {
1842 int notify;
1843
1844 RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&queue->tx, notify);
1845 if (notify)
1846 notify_remote_via_irq(queue->tx_irq);
1847 }
1848
make_rx_response(struct xenvif_queue * queue,u16 id,s8 st,u16 offset,u16 size,u16 flags)1849 static struct xen_netif_rx_response *make_rx_response(struct xenvif_queue *queue,
1850 u16 id,
1851 s8 st,
1852 u16 offset,
1853 u16 size,
1854 u16 flags)
1855 {
1856 RING_IDX i = queue->rx.rsp_prod_pvt;
1857 struct xen_netif_rx_response *resp;
1858
1859 resp = RING_GET_RESPONSE(&queue->rx, i);
1860 resp->offset = offset;
1861 resp->flags = flags;
1862 resp->id = id;
1863 resp->status = (s16)size;
1864 if (st < 0)
1865 resp->status = (s16)st;
1866
1867 queue->rx.rsp_prod_pvt = ++i;
1868
1869 return resp;
1870 }
1871
xenvif_idx_unmap(struct xenvif_queue * queue,u16 pending_idx)1872 void xenvif_idx_unmap(struct xenvif_queue *queue, u16 pending_idx)
1873 {
1874 int ret;
1875 struct gnttab_unmap_grant_ref tx_unmap_op;
1876
1877 gnttab_set_unmap_op(&tx_unmap_op,
1878 idx_to_kaddr(queue, pending_idx),
1879 GNTMAP_host_map,
1880 queue->grant_tx_handle[pending_idx]);
1881 xenvif_grant_handle_reset(queue, pending_idx);
1882
1883 ret = gnttab_unmap_refs(&tx_unmap_op, NULL,
1884 &queue->mmap_pages[pending_idx], 1);
1885 if (ret) {
1886 netdev_err(queue->vif->dev,
1887 "Unmap fail: ret: %d pending_idx: %d host_addr: %llx handle: 0x%x status: %d\n",
1888 ret,
1889 pending_idx,
1890 tx_unmap_op.host_addr,
1891 tx_unmap_op.handle,
1892 tx_unmap_op.status);
1893 BUG();
1894 }
1895 }
1896
tx_work_todo(struct xenvif_queue * queue)1897 static inline int tx_work_todo(struct xenvif_queue *queue)
1898 {
1899 if (likely(RING_HAS_UNCONSUMED_REQUESTS(&queue->tx)))
1900 return 1;
1901
1902 return 0;
1903 }
1904
tx_dealloc_work_todo(struct xenvif_queue * queue)1905 static inline bool tx_dealloc_work_todo(struct xenvif_queue *queue)
1906 {
1907 return queue->dealloc_cons != queue->dealloc_prod;
1908 }
1909
xenvif_unmap_frontend_rings(struct xenvif_queue * queue)1910 void xenvif_unmap_frontend_rings(struct xenvif_queue *queue)
1911 {
1912 if (queue->tx.sring)
1913 xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(queue->vif),
1914 queue->tx.sring);
1915 if (queue->rx.sring)
1916 xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(queue->vif),
1917 queue->rx.sring);
1918 }
1919
xenvif_map_frontend_rings(struct xenvif_queue * queue,grant_ref_t tx_ring_ref,grant_ref_t rx_ring_ref)1920 int xenvif_map_frontend_rings(struct xenvif_queue *queue,
1921 grant_ref_t tx_ring_ref,
1922 grant_ref_t rx_ring_ref)
1923 {
1924 void *addr;
1925 struct xen_netif_tx_sring *txs;
1926 struct xen_netif_rx_sring *rxs;
1927
1928 int err = -ENOMEM;
1929
1930 err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(queue->vif),
1931 &tx_ring_ref, 1, &addr);
1932 if (err)
1933 goto err;
1934
1935 txs = (struct xen_netif_tx_sring *)addr;
1936 BACK_RING_INIT(&queue->tx, txs, XEN_PAGE_SIZE);
1937
1938 err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(queue->vif),
1939 &rx_ring_ref, 1, &addr);
1940 if (err)
1941 goto err;
1942
1943 rxs = (struct xen_netif_rx_sring *)addr;
1944 BACK_RING_INIT(&queue->rx, rxs, XEN_PAGE_SIZE);
1945
1946 return 0;
1947
1948 err:
1949 xenvif_unmap_frontend_rings(queue);
1950 return err;
1951 }
1952
xenvif_queue_carrier_off(struct xenvif_queue * queue)1953 static void xenvif_queue_carrier_off(struct xenvif_queue *queue)
1954 {
1955 struct xenvif *vif = queue->vif;
1956
1957 queue->stalled = true;
1958
1959 /* At least one queue has stalled? Disable the carrier. */
1960 spin_lock(&vif->lock);
1961 if (vif->stalled_queues++ == 0) {
1962 netdev_info(vif->dev, "Guest Rx stalled");
1963 netif_carrier_off(vif->dev);
1964 }
1965 spin_unlock(&vif->lock);
1966 }
1967
xenvif_queue_carrier_on(struct xenvif_queue * queue)1968 static void xenvif_queue_carrier_on(struct xenvif_queue *queue)
1969 {
1970 struct xenvif *vif = queue->vif;
1971
1972 queue->last_rx_time = jiffies; /* Reset Rx stall detection. */
1973 queue->stalled = false;
1974
1975 /* All queues are ready? Enable the carrier. */
1976 spin_lock(&vif->lock);
1977 if (--vif->stalled_queues == 0) {
1978 netdev_info(vif->dev, "Guest Rx ready");
1979 netif_carrier_on(vif->dev);
1980 }
1981 spin_unlock(&vif->lock);
1982 }
1983
xenvif_rx_queue_stalled(struct xenvif_queue * queue)1984 static bool xenvif_rx_queue_stalled(struct xenvif_queue *queue)
1985 {
1986 RING_IDX prod, cons;
1987
1988 prod = queue->rx.sring->req_prod;
1989 cons = queue->rx.req_cons;
1990
1991 return !queue->stalled && prod - cons < 1
1992 && time_after(jiffies,
1993 queue->last_rx_time + queue->vif->stall_timeout);
1994 }
1995
xenvif_rx_queue_ready(struct xenvif_queue * queue)1996 static bool xenvif_rx_queue_ready(struct xenvif_queue *queue)
1997 {
1998 RING_IDX prod, cons;
1999
2000 prod = queue->rx.sring->req_prod;
2001 cons = queue->rx.req_cons;
2002
2003 return queue->stalled && prod - cons >= 1;
2004 }
2005
xenvif_have_rx_work(struct xenvif_queue * queue)2006 static bool xenvif_have_rx_work(struct xenvif_queue *queue)
2007 {
2008 return (!skb_queue_empty(&queue->rx_queue)
2009 && xenvif_rx_ring_slots_available(queue))
2010 || (queue->vif->stall_timeout &&
2011 (xenvif_rx_queue_stalled(queue)
2012 || xenvif_rx_queue_ready(queue)))
2013 || kthread_should_stop()
2014 || queue->vif->disabled;
2015 }
2016
xenvif_rx_queue_timeout(struct xenvif_queue * queue)2017 static long xenvif_rx_queue_timeout(struct xenvif_queue *queue)
2018 {
2019 struct sk_buff *skb;
2020 long timeout;
2021
2022 skb = skb_peek(&queue->rx_queue);
2023 if (!skb)
2024 return MAX_SCHEDULE_TIMEOUT;
2025
2026 timeout = XENVIF_RX_CB(skb)->expires - jiffies;
2027 return timeout < 0 ? 0 : timeout;
2028 }
2029
2030 /* Wait until the guest Rx thread has work.
2031 *
2032 * The timeout needs to be adjusted based on the current head of the
2033 * queue (and not just the head at the beginning). In particular, if
2034 * the queue is initially empty an infinite timeout is used and this
2035 * needs to be reduced when a skb is queued.
2036 *
2037 * This cannot be done with wait_event_timeout() because it only
2038 * calculates the timeout once.
2039 */
xenvif_wait_for_rx_work(struct xenvif_queue * queue)2040 static void xenvif_wait_for_rx_work(struct xenvif_queue *queue)
2041 {
2042 DEFINE_WAIT(wait);
2043
2044 if (xenvif_have_rx_work(queue))
2045 return;
2046
2047 for (;;) {
2048 long ret;
2049
2050 prepare_to_wait(&queue->wq, &wait, TASK_INTERRUPTIBLE);
2051 if (xenvif_have_rx_work(queue))
2052 break;
2053 ret = schedule_timeout(xenvif_rx_queue_timeout(queue));
2054 if (!ret)
2055 break;
2056 }
2057 finish_wait(&queue->wq, &wait);
2058 }
2059
xenvif_kthread_guest_rx(void * data)2060 int xenvif_kthread_guest_rx(void *data)
2061 {
2062 struct xenvif_queue *queue = data;
2063 struct xenvif *vif = queue->vif;
2064
2065 if (!vif->stall_timeout)
2066 xenvif_queue_carrier_on(queue);
2067
2068 for (;;) {
2069 xenvif_wait_for_rx_work(queue);
2070
2071 if (kthread_should_stop())
2072 break;
2073
2074 /* This frontend is found to be rogue, disable it in
2075 * kthread context. Currently this is only set when
2076 * netback finds out frontend sends malformed packet,
2077 * but we cannot disable the interface in softirq
2078 * context so we defer it here, if this thread is
2079 * associated with queue 0.
2080 */
2081 if (unlikely(vif->disabled && queue->id == 0)) {
2082 xenvif_carrier_off(vif);
2083 break;
2084 }
2085
2086 if (!skb_queue_empty(&queue->rx_queue))
2087 xenvif_rx_action(queue);
2088
2089 /* If the guest hasn't provided any Rx slots for a
2090 * while it's probably not responsive, drop the
2091 * carrier so packets are dropped earlier.
2092 */
2093 if (vif->stall_timeout) {
2094 if (xenvif_rx_queue_stalled(queue))
2095 xenvif_queue_carrier_off(queue);
2096 else if (xenvif_rx_queue_ready(queue))
2097 xenvif_queue_carrier_on(queue);
2098 }
2099
2100 /* Queued packets may have foreign pages from other
2101 * domains. These cannot be queued indefinitely as
2102 * this would starve guests of grant refs and transmit
2103 * slots.
2104 */
2105 xenvif_rx_queue_drop_expired(queue);
2106
2107 xenvif_rx_queue_maybe_wake(queue);
2108
2109 cond_resched();
2110 }
2111
2112 /* Bin any remaining skbs */
2113 xenvif_rx_queue_purge(queue);
2114
2115 return 0;
2116 }
2117
xenvif_dealloc_kthread_should_stop(struct xenvif_queue * queue)2118 static bool xenvif_dealloc_kthread_should_stop(struct xenvif_queue *queue)
2119 {
2120 /* Dealloc thread must remain running until all inflight
2121 * packets complete.
2122 */
2123 return kthread_should_stop() &&
2124 !atomic_read(&queue->inflight_packets);
2125 }
2126
xenvif_dealloc_kthread(void * data)2127 int xenvif_dealloc_kthread(void *data)
2128 {
2129 struct xenvif_queue *queue = data;
2130
2131 for (;;) {
2132 wait_event_interruptible(queue->dealloc_wq,
2133 tx_dealloc_work_todo(queue) ||
2134 xenvif_dealloc_kthread_should_stop(queue));
2135 if (xenvif_dealloc_kthread_should_stop(queue))
2136 break;
2137
2138 xenvif_tx_dealloc_action(queue);
2139 cond_resched();
2140 }
2141
2142 /* Unmap anything remaining*/
2143 if (tx_dealloc_work_todo(queue))
2144 xenvif_tx_dealloc_action(queue);
2145
2146 return 0;
2147 }
2148
netback_init(void)2149 static int __init netback_init(void)
2150 {
2151 int rc = 0;
2152
2153 if (!xen_domain())
2154 return -ENODEV;
2155
2156 /* Allow as many queues as there are CPUs if user has not
2157 * specified a value.
2158 */
2159 if (xenvif_max_queues == 0)
2160 xenvif_max_queues = num_online_cpus();
2161
2162 if (fatal_skb_slots < XEN_NETBK_LEGACY_SLOTS_MAX) {
2163 pr_info("fatal_skb_slots too small (%d), bump it to XEN_NETBK_LEGACY_SLOTS_MAX (%d)\n",
2164 fatal_skb_slots, XEN_NETBK_LEGACY_SLOTS_MAX);
2165 fatal_skb_slots = XEN_NETBK_LEGACY_SLOTS_MAX;
2166 }
2167
2168 rc = xenvif_xenbus_init();
2169 if (rc)
2170 goto failed_init;
2171
2172 #ifdef CONFIG_DEBUG_FS
2173 xen_netback_dbg_root = debugfs_create_dir("xen-netback", NULL);
2174 if (IS_ERR_OR_NULL(xen_netback_dbg_root))
2175 pr_warn("Init of debugfs returned %ld!\n",
2176 PTR_ERR(xen_netback_dbg_root));
2177 #endif /* CONFIG_DEBUG_FS */
2178
2179 return 0;
2180
2181 failed_init:
2182 return rc;
2183 }
2184
2185 module_init(netback_init);
2186
netback_fini(void)2187 static void __exit netback_fini(void)
2188 {
2189 #ifdef CONFIG_DEBUG_FS
2190 if (!IS_ERR_OR_NULL(xen_netback_dbg_root))
2191 debugfs_remove_recursive(xen_netback_dbg_root);
2192 #endif /* CONFIG_DEBUG_FS */
2193 xenvif_xenbus_fini();
2194 }
2195 module_exit(netback_fini);
2196
2197 MODULE_LICENSE("Dual BSD/GPL");
2198 MODULE_ALIAS("xen-backend:vif");
2199