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