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 /*
237  * Returns true if we should start a new receive buffer instead of
238  * adding 'size' bytes to a buffer which currently contains 'offset'
239  * bytes.
240  */
241 static bool start_new_rx_buffer(int offset, unsigned long size, int head,
242 				bool full_coalesce)
243 {
244 	/* simple case: we have completely filled the current buffer. */
245 	if (offset == MAX_BUFFER_OFFSET)
246 		return true;
247 
248 	/*
249 	 * complex case: start a fresh buffer if the current frag
250 	 * would overflow the current buffer but only if:
251 	 *     (i)   this frag would fit completely in the next buffer
252 	 * and (ii)  there is already some data in the current buffer
253 	 * and (iii) this is not the head buffer.
254 	 * and (iv)  there is no need to fully utilize the buffers
255 	 *
256 	 * Where:
257 	 * - (i) stops us splitting a frag into two copies
258 	 *   unless the frag is too large for a single buffer.
259 	 * - (ii) stops us from leaving a buffer pointlessly empty.
260 	 * - (iii) stops us leaving the first buffer
261 	 *   empty. Strictly speaking this is already covered
262 	 *   by (ii) but is explicitly checked because
263 	 *   netfront relies on the first buffer being
264 	 *   non-empty and can crash otherwise.
265 	 * - (iv) is needed for skbs which can use up more than MAX_SKB_FRAGS
266 	 *   slot
267 	 *
268 	 * This means we will effectively linearise small
269 	 * frags but do not needlessly split large buffers
270 	 * into multiple copies tend to give large frags their
271 	 * own buffers as before.
272 	 */
273 	BUG_ON(size > MAX_BUFFER_OFFSET);
274 	if ((offset + size > MAX_BUFFER_OFFSET) && offset && !head &&
275 	    !full_coalesce)
276 		return true;
277 
278 	return false;
279 }
280 
281 struct netrx_pending_operations {
282 	unsigned copy_prod, copy_cons;
283 	unsigned meta_prod, meta_cons;
284 	struct gnttab_copy *copy;
285 	struct xenvif_rx_meta *meta;
286 	int copy_off;
287 	grant_ref_t copy_gref;
288 };
289 
290 static struct xenvif_rx_meta *get_next_rx_buffer(struct xenvif_queue *queue,
291 						 struct netrx_pending_operations *npo)
292 {
293 	struct xenvif_rx_meta *meta;
294 	struct xen_netif_rx_request *req;
295 
296 	req = RING_GET_REQUEST(&queue->rx, queue->rx.req_cons++);
297 
298 	meta = npo->meta + npo->meta_prod++;
299 	meta->gso_type = XEN_NETIF_GSO_TYPE_NONE;
300 	meta->gso_size = 0;
301 	meta->size = 0;
302 	meta->id = req->id;
303 
304 	npo->copy_off = 0;
305 	npo->copy_gref = req->gref;
306 
307 	return meta;
308 }
309 
310 /*
311  * Set up the grant operations for this fragment. If it's a flipping
312  * interface, we also set up the unmap request from here.
313  */
314 static void xenvif_gop_frag_copy(struct xenvif_queue *queue, struct sk_buff *skb,
315 				 struct netrx_pending_operations *npo,
316 				 struct page *page, unsigned long size,
317 				 unsigned long offset, int *head,
318 				 struct xenvif_queue *foreign_queue,
319 				 grant_ref_t foreign_gref)
320 {
321 	struct gnttab_copy *copy_gop;
322 	struct xenvif_rx_meta *meta;
323 	unsigned long bytes;
324 	int gso_type = XEN_NETIF_GSO_TYPE_NONE;
325 
326 	/* Data must not cross a page boundary. */
327 	BUG_ON(size + offset > PAGE_SIZE<<compound_order(page));
328 
329 	meta = npo->meta + npo->meta_prod - 1;
330 
331 	/* Skip unused frames from start of page */
332 	page += offset >> PAGE_SHIFT;
333 	offset &= ~PAGE_MASK;
334 
335 	while (size > 0) {
336 		BUG_ON(offset >= PAGE_SIZE);
337 		BUG_ON(npo->copy_off > MAX_BUFFER_OFFSET);
338 
339 		bytes = PAGE_SIZE - offset;
340 
341 		if (bytes > size)
342 			bytes = size;
343 
344 		if (start_new_rx_buffer(npo->copy_off,
345 					bytes,
346 					*head,
347 					XENVIF_RX_CB(skb)->full_coalesce)) {
348 			/*
349 			 * Netfront requires there to be some data in the head
350 			 * buffer.
351 			 */
352 			BUG_ON(*head);
353 
354 			meta = get_next_rx_buffer(queue, npo);
355 		}
356 
357 		if (npo->copy_off + bytes > MAX_BUFFER_OFFSET)
358 			bytes = MAX_BUFFER_OFFSET - npo->copy_off;
359 
360 		copy_gop = npo->copy + npo->copy_prod++;
361 		copy_gop->flags = GNTCOPY_dest_gref;
362 		copy_gop->len = bytes;
363 
364 		if (foreign_queue) {
365 			copy_gop->source.domid = foreign_queue->vif->domid;
366 			copy_gop->source.u.ref = foreign_gref;
367 			copy_gop->flags |= GNTCOPY_source_gref;
368 		} else {
369 			copy_gop->source.domid = DOMID_SELF;
370 			copy_gop->source.u.gmfn =
371 				virt_to_mfn(page_address(page));
372 		}
373 		copy_gop->source.offset = offset;
374 
375 		copy_gop->dest.domid = queue->vif->domid;
376 		copy_gop->dest.offset = npo->copy_off;
377 		copy_gop->dest.u.ref = npo->copy_gref;
378 
379 		npo->copy_off += bytes;
380 		meta->size += bytes;
381 
382 		offset += bytes;
383 		size -= bytes;
384 
385 		/* Next frame */
386 		if (offset == PAGE_SIZE && size) {
387 			BUG_ON(!PageCompound(page));
388 			page++;
389 			offset = 0;
390 		}
391 
392 		/* Leave a gap for the GSO descriptor. */
393 		if (skb_is_gso(skb)) {
394 			if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4)
395 				gso_type = XEN_NETIF_GSO_TYPE_TCPV4;
396 			else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)
397 				gso_type = XEN_NETIF_GSO_TYPE_TCPV6;
398 		}
399 
400 		if (*head && ((1 << gso_type) & queue->vif->gso_mask))
401 			queue->rx.req_cons++;
402 
403 		*head = 0; /* There must be something in this buffer now. */
404 
405 	}
406 }
407 
408 /*
409  * Find the grant ref for a given frag in a chain of struct ubuf_info's
410  * skb: the skb itself
411  * i: the frag's number
412  * ubuf: a pointer to an element in the chain. It should not be NULL
413  *
414  * Returns a pointer to the element in the chain where the page were found. If
415  * not found, returns NULL.
416  * See the definition of callback_struct in common.h for more details about
417  * the chain.
418  */
419 static const struct ubuf_info *xenvif_find_gref(const struct sk_buff *const skb,
420 						const int i,
421 						const struct ubuf_info *ubuf)
422 {
423 	struct xenvif_queue *foreign_queue = ubuf_to_queue(ubuf);
424 
425 	do {
426 		u16 pending_idx = ubuf->desc;
427 
428 		if (skb_shinfo(skb)->frags[i].page.p ==
429 		    foreign_queue->mmap_pages[pending_idx])
430 			break;
431 		ubuf = (struct ubuf_info *) ubuf->ctx;
432 	} while (ubuf);
433 
434 	return ubuf;
435 }
436 
437 /*
438  * Prepare an SKB to be transmitted to the frontend.
439  *
440  * This function is responsible for allocating grant operations, meta
441  * structures, etc.
442  *
443  * It returns the number of meta structures consumed. The number of
444  * ring slots used is always equal to the number of meta slots used
445  * plus the number of GSO descriptors used. Currently, we use either
446  * zero GSO descriptors (for non-GSO packets) or one descriptor (for
447  * frontend-side LRO).
448  */
449 static int xenvif_gop_skb(struct sk_buff *skb,
450 			  struct netrx_pending_operations *npo,
451 			  struct xenvif_queue *queue)
452 {
453 	struct xenvif *vif = netdev_priv(skb->dev);
454 	int nr_frags = skb_shinfo(skb)->nr_frags;
455 	int i;
456 	struct xen_netif_rx_request *req;
457 	struct xenvif_rx_meta *meta;
458 	unsigned char *data;
459 	int head = 1;
460 	int old_meta_prod;
461 	int gso_type;
462 	const struct ubuf_info *ubuf = skb_shinfo(skb)->destructor_arg;
463 	const struct ubuf_info *const head_ubuf = ubuf;
464 
465 	old_meta_prod = npo->meta_prod;
466 
467 	gso_type = XEN_NETIF_GSO_TYPE_NONE;
468 	if (skb_is_gso(skb)) {
469 		if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4)
470 			gso_type = XEN_NETIF_GSO_TYPE_TCPV4;
471 		else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)
472 			gso_type = XEN_NETIF_GSO_TYPE_TCPV6;
473 	}
474 
475 	/* Set up a GSO prefix descriptor, if necessary */
476 	if ((1 << gso_type) & vif->gso_prefix_mask) {
477 		req = RING_GET_REQUEST(&queue->rx, queue->rx.req_cons++);
478 		meta = npo->meta + npo->meta_prod++;
479 		meta->gso_type = gso_type;
480 		meta->gso_size = skb_shinfo(skb)->gso_size;
481 		meta->size = 0;
482 		meta->id = req->id;
483 	}
484 
485 	req = RING_GET_REQUEST(&queue->rx, queue->rx.req_cons++);
486 	meta = npo->meta + npo->meta_prod++;
487 
488 	if ((1 << gso_type) & vif->gso_mask) {
489 		meta->gso_type = gso_type;
490 		meta->gso_size = skb_shinfo(skb)->gso_size;
491 	} else {
492 		meta->gso_type = XEN_NETIF_GSO_TYPE_NONE;
493 		meta->gso_size = 0;
494 	}
495 
496 	meta->size = 0;
497 	meta->id = req->id;
498 	npo->copy_off = 0;
499 	npo->copy_gref = req->gref;
500 
501 	data = skb->data;
502 	while (data < skb_tail_pointer(skb)) {
503 		unsigned int offset = offset_in_page(data);
504 		unsigned int len = PAGE_SIZE - offset;
505 
506 		if (data + len > skb_tail_pointer(skb))
507 			len = skb_tail_pointer(skb) - data;
508 
509 		xenvif_gop_frag_copy(queue, skb, npo,
510 				     virt_to_page(data), len, offset, &head,
511 				     NULL,
512 				     0);
513 		data += len;
514 	}
515 
516 	for (i = 0; i < nr_frags; i++) {
517 		/* This variable also signals whether foreign_gref has a real
518 		 * value or not.
519 		 */
520 		struct xenvif_queue *foreign_queue = NULL;
521 		grant_ref_t foreign_gref;
522 
523 		if ((skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) &&
524 			(ubuf->callback == &xenvif_zerocopy_callback)) {
525 			const struct ubuf_info *const startpoint = ubuf;
526 
527 			/* Ideally ubuf points to the chain element which
528 			 * belongs to this frag. Or if frags were removed from
529 			 * the beginning, then shortly before it.
530 			 */
531 			ubuf = xenvif_find_gref(skb, i, ubuf);
532 
533 			/* Try again from the beginning of the list, if we
534 			 * haven't tried from there. This only makes sense in
535 			 * the unlikely event of reordering the original frags.
536 			 * For injected local pages it's an unnecessary second
537 			 * run.
538 			 */
539 			if (unlikely(!ubuf) && startpoint != head_ubuf)
540 				ubuf = xenvif_find_gref(skb, i, head_ubuf);
541 
542 			if (likely(ubuf)) {
543 				u16 pending_idx = ubuf->desc;
544 
545 				foreign_queue = ubuf_to_queue(ubuf);
546 				foreign_gref =
547 					foreign_queue->pending_tx_info[pending_idx].req.gref;
548 				/* Just a safety measure. If this was the last
549 				 * element on the list, the for loop will
550 				 * iterate again if a local page were added to
551 				 * the end. Using head_ubuf here prevents the
552 				 * second search on the chain. Or the original
553 				 * frags changed order, but that's less likely.
554 				 * In any way, ubuf shouldn't be NULL.
555 				 */
556 				ubuf = ubuf->ctx ?
557 					(struct ubuf_info *) ubuf->ctx :
558 					head_ubuf;
559 			} else
560 				/* This frag was a local page, added to the
561 				 * array after the skb left netback.
562 				 */
563 				ubuf = head_ubuf;
564 		}
565 		xenvif_gop_frag_copy(queue, skb, npo,
566 				     skb_frag_page(&skb_shinfo(skb)->frags[i]),
567 				     skb_frag_size(&skb_shinfo(skb)->frags[i]),
568 				     skb_shinfo(skb)->frags[i].page_offset,
569 				     &head,
570 				     foreign_queue,
571 				     foreign_queue ? foreign_gref : UINT_MAX);
572 	}
573 
574 	return npo->meta_prod - old_meta_prod;
575 }
576 
577 /*
578  * This is a twin to xenvif_gop_skb.  Assume that xenvif_gop_skb was
579  * used to set up the operations on the top of
580  * netrx_pending_operations, which have since been done.  Check that
581  * they didn't give any errors and advance over them.
582  */
583 static int xenvif_check_gop(struct xenvif *vif, int nr_meta_slots,
584 			    struct netrx_pending_operations *npo)
585 {
586 	struct gnttab_copy     *copy_op;
587 	int status = XEN_NETIF_RSP_OKAY;
588 	int i;
589 
590 	for (i = 0; i < nr_meta_slots; i++) {
591 		copy_op = npo->copy + npo->copy_cons++;
592 		if (copy_op->status != GNTST_okay) {
593 			netdev_dbg(vif->dev,
594 				   "Bad status %d from copy to DOM%d.\n",
595 				   copy_op->status, vif->domid);
596 			status = XEN_NETIF_RSP_ERROR;
597 		}
598 	}
599 
600 	return status;
601 }
602 
603 static void xenvif_add_frag_responses(struct xenvif_queue *queue, int status,
604 				      struct xenvif_rx_meta *meta,
605 				      int nr_meta_slots)
606 {
607 	int i;
608 	unsigned long offset;
609 
610 	/* No fragments used */
611 	if (nr_meta_slots <= 1)
612 		return;
613 
614 	nr_meta_slots--;
615 
616 	for (i = 0; i < nr_meta_slots; i++) {
617 		int flags;
618 		if (i == nr_meta_slots - 1)
619 			flags = 0;
620 		else
621 			flags = XEN_NETRXF_more_data;
622 
623 		offset = 0;
624 		make_rx_response(queue, meta[i].id, status, offset,
625 				 meta[i].size, flags);
626 	}
627 }
628 
629 void xenvif_kick_thread(struct xenvif_queue *queue)
630 {
631 	wake_up(&queue->wq);
632 }
633 
634 static void xenvif_rx_action(struct xenvif_queue *queue)
635 {
636 	s8 status;
637 	u16 flags;
638 	struct xen_netif_rx_response *resp;
639 	struct sk_buff_head rxq;
640 	struct sk_buff *skb;
641 	LIST_HEAD(notify);
642 	int ret;
643 	unsigned long offset;
644 	bool need_to_notify = false;
645 
646 	struct netrx_pending_operations npo = {
647 		.copy  = queue->grant_copy_op,
648 		.meta  = queue->meta,
649 	};
650 
651 	skb_queue_head_init(&rxq);
652 
653 	while (xenvif_rx_ring_slots_available(queue, XEN_NETBK_RX_SLOTS_MAX)
654 	       && (skb = xenvif_rx_dequeue(queue)) != NULL) {
655 		RING_IDX max_slots_needed;
656 		RING_IDX old_req_cons;
657 		RING_IDX ring_slots_used;
658 		int i;
659 
660 		queue->last_rx_time = jiffies;
661 
662 		/* We need a cheap worse case estimate for the number of
663 		 * slots we'll use.
664 		 */
665 
666 		max_slots_needed = DIV_ROUND_UP(offset_in_page(skb->data) +
667 						skb_headlen(skb),
668 						PAGE_SIZE);
669 		for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
670 			unsigned int size;
671 			unsigned int offset;
672 
673 			size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
674 			offset = skb_shinfo(skb)->frags[i].page_offset;
675 
676 			/* For a worse-case estimate we need to factor in
677 			 * the fragment page offset as this will affect the
678 			 * number of times xenvif_gop_frag_copy() will
679 			 * call start_new_rx_buffer().
680 			 */
681 			max_slots_needed += DIV_ROUND_UP(offset + size,
682 							 PAGE_SIZE);
683 		}
684 
685 		/* To avoid the estimate becoming too pessimal for some
686 		 * frontends that limit posted rx requests, cap the estimate
687 		 * at MAX_SKB_FRAGS. In this case netback will fully coalesce
688 		 * the skb into the provided slots.
689 		 */
690 		if (max_slots_needed > MAX_SKB_FRAGS) {
691 			max_slots_needed = MAX_SKB_FRAGS;
692 			XENVIF_RX_CB(skb)->full_coalesce = true;
693 		} else {
694 			XENVIF_RX_CB(skb)->full_coalesce = false;
695 		}
696 
697 		/* We may need one more slot for GSO metadata */
698 		if (skb_is_gso(skb) &&
699 		   (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4 ||
700 		    skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6))
701 			max_slots_needed++;
702 
703 		old_req_cons = queue->rx.req_cons;
704 		XENVIF_RX_CB(skb)->meta_slots_used = xenvif_gop_skb(skb, &npo, queue);
705 		ring_slots_used = queue->rx.req_cons - old_req_cons;
706 
707 		BUG_ON(ring_slots_used > max_slots_needed);
708 
709 		__skb_queue_tail(&rxq, skb);
710 	}
711 
712 	BUG_ON(npo.meta_prod > ARRAY_SIZE(queue->meta));
713 
714 	if (!npo.copy_prod)
715 		goto done;
716 
717 	BUG_ON(npo.copy_prod > MAX_GRANT_COPY_OPS);
718 	gnttab_batch_copy(queue->grant_copy_op, npo.copy_prod);
719 
720 	while ((skb = __skb_dequeue(&rxq)) != NULL) {
721 
722 		if ((1 << queue->meta[npo.meta_cons].gso_type) &
723 		    queue->vif->gso_prefix_mask) {
724 			resp = RING_GET_RESPONSE(&queue->rx,
725 						 queue->rx.rsp_prod_pvt++);
726 
727 			resp->flags = XEN_NETRXF_gso_prefix | XEN_NETRXF_more_data;
728 
729 			resp->offset = queue->meta[npo.meta_cons].gso_size;
730 			resp->id = queue->meta[npo.meta_cons].id;
731 			resp->status = XENVIF_RX_CB(skb)->meta_slots_used;
732 
733 			npo.meta_cons++;
734 			XENVIF_RX_CB(skb)->meta_slots_used--;
735 		}
736 
737 
738 		queue->stats.tx_bytes += skb->len;
739 		queue->stats.tx_packets++;
740 
741 		status = xenvif_check_gop(queue->vif,
742 					  XENVIF_RX_CB(skb)->meta_slots_used,
743 					  &npo);
744 
745 		if (XENVIF_RX_CB(skb)->meta_slots_used == 1)
746 			flags = 0;
747 		else
748 			flags = XEN_NETRXF_more_data;
749 
750 		if (skb->ip_summed == CHECKSUM_PARTIAL) /* local packet? */
751 			flags |= XEN_NETRXF_csum_blank | XEN_NETRXF_data_validated;
752 		else if (skb->ip_summed == CHECKSUM_UNNECESSARY)
753 			/* remote but checksummed. */
754 			flags |= XEN_NETRXF_data_validated;
755 
756 		offset = 0;
757 		resp = make_rx_response(queue, queue->meta[npo.meta_cons].id,
758 					status, offset,
759 					queue->meta[npo.meta_cons].size,
760 					flags);
761 
762 		if ((1 << queue->meta[npo.meta_cons].gso_type) &
763 		    queue->vif->gso_mask) {
764 			struct xen_netif_extra_info *gso =
765 				(struct xen_netif_extra_info *)
766 				RING_GET_RESPONSE(&queue->rx,
767 						  queue->rx.rsp_prod_pvt++);
768 
769 			resp->flags |= XEN_NETRXF_extra_info;
770 
771 			gso->u.gso.type = queue->meta[npo.meta_cons].gso_type;
772 			gso->u.gso.size = queue->meta[npo.meta_cons].gso_size;
773 			gso->u.gso.pad = 0;
774 			gso->u.gso.features = 0;
775 
776 			gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
777 			gso->flags = 0;
778 		}
779 
780 		xenvif_add_frag_responses(queue, status,
781 					  queue->meta + npo.meta_cons + 1,
782 					  XENVIF_RX_CB(skb)->meta_slots_used);
783 
784 		RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&queue->rx, ret);
785 
786 		need_to_notify |= !!ret;
787 
788 		npo.meta_cons += XENVIF_RX_CB(skb)->meta_slots_used;
789 		dev_kfree_skb(skb);
790 	}
791 
792 done:
793 	if (need_to_notify)
794 		notify_remote_via_irq(queue->rx_irq);
795 }
796 
797 void xenvif_napi_schedule_or_enable_events(struct xenvif_queue *queue)
798 {
799 	int more_to_do;
800 
801 	RING_FINAL_CHECK_FOR_REQUESTS(&queue->tx, more_to_do);
802 
803 	if (more_to_do)
804 		napi_schedule(&queue->napi);
805 }
806 
807 static void tx_add_credit(struct xenvif_queue *queue)
808 {
809 	unsigned long max_burst, max_credit;
810 
811 	/*
812 	 * Allow a burst big enough to transmit a jumbo packet of up to 128kB.
813 	 * Otherwise the interface can seize up due to insufficient credit.
814 	 */
815 	max_burst = RING_GET_REQUEST(&queue->tx, queue->tx.req_cons)->size;
816 	max_burst = min(max_burst, 131072UL);
817 	max_burst = max(max_burst, queue->credit_bytes);
818 
819 	/* Take care that adding a new chunk of credit doesn't wrap to zero. */
820 	max_credit = queue->remaining_credit + queue->credit_bytes;
821 	if (max_credit < queue->remaining_credit)
822 		max_credit = ULONG_MAX; /* wrapped: clamp to ULONG_MAX */
823 
824 	queue->remaining_credit = min(max_credit, max_burst);
825 }
826 
827 static void tx_credit_callback(unsigned long data)
828 {
829 	struct xenvif_queue *queue = (struct xenvif_queue *)data;
830 	tx_add_credit(queue);
831 	xenvif_napi_schedule_or_enable_events(queue);
832 }
833 
834 static void xenvif_tx_err(struct xenvif_queue *queue,
835 			  struct xen_netif_tx_request *txp, RING_IDX end)
836 {
837 	RING_IDX cons = queue->tx.req_cons;
838 	unsigned long flags;
839 
840 	do {
841 		spin_lock_irqsave(&queue->response_lock, flags);
842 		make_tx_response(queue, txp, XEN_NETIF_RSP_ERROR);
843 		spin_unlock_irqrestore(&queue->response_lock, flags);
844 		if (cons == end)
845 			break;
846 		txp = RING_GET_REQUEST(&queue->tx, cons++);
847 	} while (1);
848 	queue->tx.req_cons = cons;
849 }
850 
851 static void xenvif_fatal_tx_err(struct xenvif *vif)
852 {
853 	netdev_err(vif->dev, "fatal error; disabling device\n");
854 	vif->disabled = true;
855 	/* Disable the vif from queue 0's kthread */
856 	if (vif->queues)
857 		xenvif_kick_thread(&vif->queues[0]);
858 }
859 
860 static int xenvif_count_requests(struct xenvif_queue *queue,
861 				 struct xen_netif_tx_request *first,
862 				 struct xen_netif_tx_request *txp,
863 				 int work_to_do)
864 {
865 	RING_IDX cons = queue->tx.req_cons;
866 	int slots = 0;
867 	int drop_err = 0;
868 	int more_data;
869 
870 	if (!(first->flags & XEN_NETTXF_more_data))
871 		return 0;
872 
873 	do {
874 		struct xen_netif_tx_request dropped_tx = { 0 };
875 
876 		if (slots >= work_to_do) {
877 			netdev_err(queue->vif->dev,
878 				   "Asked for %d slots but exceeds this limit\n",
879 				   work_to_do);
880 			xenvif_fatal_tx_err(queue->vif);
881 			return -ENODATA;
882 		}
883 
884 		/* This guest is really using too many slots and
885 		 * considered malicious.
886 		 */
887 		if (unlikely(slots >= fatal_skb_slots)) {
888 			netdev_err(queue->vif->dev,
889 				   "Malicious frontend using %d slots, threshold %u\n",
890 				   slots, fatal_skb_slots);
891 			xenvif_fatal_tx_err(queue->vif);
892 			return -E2BIG;
893 		}
894 
895 		/* Xen network protocol had implicit dependency on
896 		 * MAX_SKB_FRAGS. XEN_NETBK_LEGACY_SLOTS_MAX is set to
897 		 * the historical MAX_SKB_FRAGS value 18 to honor the
898 		 * same behavior as before. Any packet using more than
899 		 * 18 slots but less than fatal_skb_slots slots is
900 		 * dropped
901 		 */
902 		if (!drop_err && slots >= XEN_NETBK_LEGACY_SLOTS_MAX) {
903 			if (net_ratelimit())
904 				netdev_dbg(queue->vif->dev,
905 					   "Too many slots (%d) exceeding limit (%d), dropping packet\n",
906 					   slots, XEN_NETBK_LEGACY_SLOTS_MAX);
907 			drop_err = -E2BIG;
908 		}
909 
910 		if (drop_err)
911 			txp = &dropped_tx;
912 
913 		memcpy(txp, RING_GET_REQUEST(&queue->tx, cons + slots),
914 		       sizeof(*txp));
915 
916 		/* If the guest submitted a frame >= 64 KiB then
917 		 * first->size overflowed and following slots will
918 		 * appear to be larger than the frame.
919 		 *
920 		 * This cannot be fatal error as there are buggy
921 		 * frontends that do this.
922 		 *
923 		 * Consume all slots and drop the packet.
924 		 */
925 		if (!drop_err && txp->size > first->size) {
926 			if (net_ratelimit())
927 				netdev_dbg(queue->vif->dev,
928 					   "Invalid tx request, slot size %u > remaining size %u\n",
929 					   txp->size, first->size);
930 			drop_err = -EIO;
931 		}
932 
933 		first->size -= txp->size;
934 		slots++;
935 
936 		if (unlikely((txp->offset + txp->size) > PAGE_SIZE)) {
937 			netdev_err(queue->vif->dev, "Cross page boundary, txp->offset: %x, size: %u\n",
938 				 txp->offset, txp->size);
939 			xenvif_fatal_tx_err(queue->vif);
940 			return -EINVAL;
941 		}
942 
943 		more_data = txp->flags & XEN_NETTXF_more_data;
944 
945 		if (!drop_err)
946 			txp++;
947 
948 	} while (more_data);
949 
950 	if (drop_err) {
951 		xenvif_tx_err(queue, first, cons + slots);
952 		return drop_err;
953 	}
954 
955 	return slots;
956 }
957 
958 
959 struct xenvif_tx_cb {
960 	u16 pending_idx;
961 };
962 
963 #define XENVIF_TX_CB(skb) ((struct xenvif_tx_cb *)(skb)->cb)
964 
965 static inline void xenvif_tx_create_map_op(struct xenvif_queue *queue,
966 					  u16 pending_idx,
967 					  struct xen_netif_tx_request *txp,
968 					  struct gnttab_map_grant_ref *mop)
969 {
970 	queue->pages_to_map[mop-queue->tx_map_ops] = queue->mmap_pages[pending_idx];
971 	gnttab_set_map_op(mop, idx_to_kaddr(queue, pending_idx),
972 			  GNTMAP_host_map | GNTMAP_readonly,
973 			  txp->gref, queue->vif->domid);
974 
975 	memcpy(&queue->pending_tx_info[pending_idx].req, txp,
976 	       sizeof(*txp));
977 }
978 
979 static inline struct sk_buff *xenvif_alloc_skb(unsigned int size)
980 {
981 	struct sk_buff *skb =
982 		alloc_skb(size + NET_SKB_PAD + NET_IP_ALIGN,
983 			  GFP_ATOMIC | __GFP_NOWARN);
984 	if (unlikely(skb == NULL))
985 		return NULL;
986 
987 	/* Packets passed to netif_rx() must have some headroom. */
988 	skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
989 
990 	/* Initialize it here to avoid later surprises */
991 	skb_shinfo(skb)->destructor_arg = NULL;
992 
993 	return skb;
994 }
995 
996 static struct gnttab_map_grant_ref *xenvif_get_requests(struct xenvif_queue *queue,
997 							struct sk_buff *skb,
998 							struct xen_netif_tx_request *txp,
999 							struct gnttab_map_grant_ref *gop)
1000 {
1001 	struct skb_shared_info *shinfo = skb_shinfo(skb);
1002 	skb_frag_t *frags = shinfo->frags;
1003 	u16 pending_idx = XENVIF_TX_CB(skb)->pending_idx;
1004 	int start;
1005 	pending_ring_idx_t index;
1006 	unsigned int nr_slots, frag_overflow = 0;
1007 
1008 	/* At this point shinfo->nr_frags is in fact the number of
1009 	 * slots, which can be as large as XEN_NETBK_LEGACY_SLOTS_MAX.
1010 	 */
1011 	if (shinfo->nr_frags > MAX_SKB_FRAGS) {
1012 		frag_overflow = shinfo->nr_frags - MAX_SKB_FRAGS;
1013 		BUG_ON(frag_overflow > MAX_SKB_FRAGS);
1014 		shinfo->nr_frags = MAX_SKB_FRAGS;
1015 	}
1016 	nr_slots = shinfo->nr_frags;
1017 
1018 	/* Skip first skb fragment if it is on same page as header fragment. */
1019 	start = (frag_get_pending_idx(&shinfo->frags[0]) == pending_idx);
1020 
1021 	for (shinfo->nr_frags = start; shinfo->nr_frags < nr_slots;
1022 	     shinfo->nr_frags++, txp++, gop++) {
1023 		index = pending_index(queue->pending_cons++);
1024 		pending_idx = queue->pending_ring[index];
1025 		xenvif_tx_create_map_op(queue, pending_idx, txp, gop);
1026 		frag_set_pending_idx(&frags[shinfo->nr_frags], pending_idx);
1027 	}
1028 
1029 	if (frag_overflow) {
1030 		struct sk_buff *nskb = xenvif_alloc_skb(0);
1031 		if (unlikely(nskb == NULL)) {
1032 			if (net_ratelimit())
1033 				netdev_err(queue->vif->dev,
1034 					   "Can't allocate the frag_list skb.\n");
1035 			return NULL;
1036 		}
1037 
1038 		shinfo = skb_shinfo(nskb);
1039 		frags = shinfo->frags;
1040 
1041 		for (shinfo->nr_frags = 0; shinfo->nr_frags < frag_overflow;
1042 		     shinfo->nr_frags++, txp++, gop++) {
1043 			index = pending_index(queue->pending_cons++);
1044 			pending_idx = queue->pending_ring[index];
1045 			xenvif_tx_create_map_op(queue, pending_idx, txp, gop);
1046 			frag_set_pending_idx(&frags[shinfo->nr_frags],
1047 					     pending_idx);
1048 		}
1049 
1050 		skb_shinfo(skb)->frag_list = nskb;
1051 	}
1052 
1053 	return gop;
1054 }
1055 
1056 static inline void xenvif_grant_handle_set(struct xenvif_queue *queue,
1057 					   u16 pending_idx,
1058 					   grant_handle_t handle)
1059 {
1060 	if (unlikely(queue->grant_tx_handle[pending_idx] !=
1061 		     NETBACK_INVALID_HANDLE)) {
1062 		netdev_err(queue->vif->dev,
1063 			   "Trying to overwrite active handle! pending_idx: %x\n",
1064 			   pending_idx);
1065 		BUG();
1066 	}
1067 	queue->grant_tx_handle[pending_idx] = handle;
1068 }
1069 
1070 static inline void xenvif_grant_handle_reset(struct xenvif_queue *queue,
1071 					     u16 pending_idx)
1072 {
1073 	if (unlikely(queue->grant_tx_handle[pending_idx] ==
1074 		     NETBACK_INVALID_HANDLE)) {
1075 		netdev_err(queue->vif->dev,
1076 			   "Trying to unmap invalid handle! pending_idx: %x\n",
1077 			   pending_idx);
1078 		BUG();
1079 	}
1080 	queue->grant_tx_handle[pending_idx] = NETBACK_INVALID_HANDLE;
1081 }
1082 
1083 static int xenvif_tx_check_gop(struct xenvif_queue *queue,
1084 			       struct sk_buff *skb,
1085 			       struct gnttab_map_grant_ref **gopp_map,
1086 			       struct gnttab_copy **gopp_copy)
1087 {
1088 	struct gnttab_map_grant_ref *gop_map = *gopp_map;
1089 	u16 pending_idx = XENVIF_TX_CB(skb)->pending_idx;
1090 	/* This always points to the shinfo of the skb being checked, which
1091 	 * could be either the first or the one on the frag_list
1092 	 */
1093 	struct skb_shared_info *shinfo = skb_shinfo(skb);
1094 	/* If this is non-NULL, we are currently checking the frag_list skb, and
1095 	 * this points to the shinfo of the first one
1096 	 */
1097 	struct skb_shared_info *first_shinfo = NULL;
1098 	int nr_frags = shinfo->nr_frags;
1099 	const bool sharedslot = nr_frags &&
1100 				frag_get_pending_idx(&shinfo->frags[0]) == pending_idx;
1101 	int i, err;
1102 
1103 	/* Check status of header. */
1104 	err = (*gopp_copy)->status;
1105 	if (unlikely(err)) {
1106 		if (net_ratelimit())
1107 			netdev_dbg(queue->vif->dev,
1108 				   "Grant copy of header failed! status: %d pending_idx: %u ref: %u\n",
1109 				   (*gopp_copy)->status,
1110 				   pending_idx,
1111 				   (*gopp_copy)->source.u.ref);
1112 		/* The first frag might still have this slot mapped */
1113 		if (!sharedslot)
1114 			xenvif_idx_release(queue, pending_idx,
1115 					   XEN_NETIF_RSP_ERROR);
1116 	}
1117 	(*gopp_copy)++;
1118 
1119 check_frags:
1120 	for (i = 0; i < nr_frags; i++, gop_map++) {
1121 		int j, newerr;
1122 
1123 		pending_idx = frag_get_pending_idx(&shinfo->frags[i]);
1124 
1125 		/* Check error status: if okay then remember grant handle. */
1126 		newerr = gop_map->status;
1127 
1128 		if (likely(!newerr)) {
1129 			xenvif_grant_handle_set(queue,
1130 						pending_idx,
1131 						gop_map->handle);
1132 			/* Had a previous error? Invalidate this fragment. */
1133 			if (unlikely(err)) {
1134 				xenvif_idx_unmap(queue, pending_idx);
1135 				/* If the mapping of the first frag was OK, but
1136 				 * the header's copy failed, and they are
1137 				 * sharing a slot, send an error
1138 				 */
1139 				if (i == 0 && sharedslot)
1140 					xenvif_idx_release(queue, pending_idx,
1141 							   XEN_NETIF_RSP_ERROR);
1142 				else
1143 					xenvif_idx_release(queue, pending_idx,
1144 							   XEN_NETIF_RSP_OKAY);
1145 			}
1146 			continue;
1147 		}
1148 
1149 		/* Error on this fragment: respond to client with an error. */
1150 		if (net_ratelimit())
1151 			netdev_dbg(queue->vif->dev,
1152 				   "Grant map of %d. frag failed! status: %d pending_idx: %u ref: %u\n",
1153 				   i,
1154 				   gop_map->status,
1155 				   pending_idx,
1156 				   gop_map->ref);
1157 
1158 		xenvif_idx_release(queue, pending_idx, XEN_NETIF_RSP_ERROR);
1159 
1160 		/* Not the first error? Preceding frags already invalidated. */
1161 		if (err)
1162 			continue;
1163 
1164 		/* First error: if the header haven't shared a slot with the
1165 		 * first frag, release it as well.
1166 		 */
1167 		if (!sharedslot)
1168 			xenvif_idx_release(queue,
1169 					   XENVIF_TX_CB(skb)->pending_idx,
1170 					   XEN_NETIF_RSP_OKAY);
1171 
1172 		/* Invalidate preceding fragments of this skb. */
1173 		for (j = 0; j < i; j++) {
1174 			pending_idx = frag_get_pending_idx(&shinfo->frags[j]);
1175 			xenvif_idx_unmap(queue, pending_idx);
1176 			xenvif_idx_release(queue, pending_idx,
1177 					   XEN_NETIF_RSP_OKAY);
1178 		}
1179 
1180 		/* And if we found the error while checking the frag_list, unmap
1181 		 * the first skb's frags
1182 		 */
1183 		if (first_shinfo) {
1184 			for (j = 0; j < first_shinfo->nr_frags; j++) {
1185 				pending_idx = frag_get_pending_idx(&first_shinfo->frags[j]);
1186 				xenvif_idx_unmap(queue, pending_idx);
1187 				xenvif_idx_release(queue, pending_idx,
1188 						   XEN_NETIF_RSP_OKAY);
1189 			}
1190 		}
1191 
1192 		/* Remember the error: invalidate all subsequent fragments. */
1193 		err = newerr;
1194 	}
1195 
1196 	if (skb_has_frag_list(skb) && !first_shinfo) {
1197 		first_shinfo = skb_shinfo(skb);
1198 		shinfo = skb_shinfo(skb_shinfo(skb)->frag_list);
1199 		nr_frags = shinfo->nr_frags;
1200 
1201 		goto check_frags;
1202 	}
1203 
1204 	*gopp_map = gop_map;
1205 	return err;
1206 }
1207 
1208 static void xenvif_fill_frags(struct xenvif_queue *queue, struct sk_buff *skb)
1209 {
1210 	struct skb_shared_info *shinfo = skb_shinfo(skb);
1211 	int nr_frags = shinfo->nr_frags;
1212 	int i;
1213 	u16 prev_pending_idx = INVALID_PENDING_IDX;
1214 
1215 	for (i = 0; i < nr_frags; i++) {
1216 		skb_frag_t *frag = shinfo->frags + i;
1217 		struct xen_netif_tx_request *txp;
1218 		struct page *page;
1219 		u16 pending_idx;
1220 
1221 		pending_idx = frag_get_pending_idx(frag);
1222 
1223 		/* If this is not the first frag, chain it to the previous*/
1224 		if (prev_pending_idx == INVALID_PENDING_IDX)
1225 			skb_shinfo(skb)->destructor_arg =
1226 				&callback_param(queue, pending_idx);
1227 		else
1228 			callback_param(queue, prev_pending_idx).ctx =
1229 				&callback_param(queue, pending_idx);
1230 
1231 		callback_param(queue, pending_idx).ctx = NULL;
1232 		prev_pending_idx = pending_idx;
1233 
1234 		txp = &queue->pending_tx_info[pending_idx].req;
1235 		page = virt_to_page(idx_to_kaddr(queue, pending_idx));
1236 		__skb_fill_page_desc(skb, i, page, txp->offset, txp->size);
1237 		skb->len += txp->size;
1238 		skb->data_len += txp->size;
1239 		skb->truesize += txp->size;
1240 
1241 		/* Take an extra reference to offset network stack's put_page */
1242 		get_page(queue->mmap_pages[pending_idx]);
1243 	}
1244 	/* FIXME: __skb_fill_page_desc set this to true because page->pfmemalloc
1245 	 * overlaps with "index", and "mapping" is not set. I think mapping
1246 	 * should be set. If delivered to local stack, it would drop this
1247 	 * skb in sk_filter unless the socket has the right to use it.
1248 	 */
1249 	skb->pfmemalloc	= false;
1250 }
1251 
1252 static int xenvif_get_extras(struct xenvif_queue *queue,
1253 				struct xen_netif_extra_info *extras,
1254 				int work_to_do)
1255 {
1256 	struct xen_netif_extra_info extra;
1257 	RING_IDX cons = queue->tx.req_cons;
1258 
1259 	do {
1260 		if (unlikely(work_to_do-- <= 0)) {
1261 			netdev_err(queue->vif->dev, "Missing extra info\n");
1262 			xenvif_fatal_tx_err(queue->vif);
1263 			return -EBADR;
1264 		}
1265 
1266 		memcpy(&extra, RING_GET_REQUEST(&queue->tx, cons),
1267 		       sizeof(extra));
1268 		if (unlikely(!extra.type ||
1269 			     extra.type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
1270 			queue->tx.req_cons = ++cons;
1271 			netdev_err(queue->vif->dev,
1272 				   "Invalid extra type: %d\n", extra.type);
1273 			xenvif_fatal_tx_err(queue->vif);
1274 			return -EINVAL;
1275 		}
1276 
1277 		memcpy(&extras[extra.type - 1], &extra, sizeof(extra));
1278 		queue->tx.req_cons = ++cons;
1279 	} while (extra.flags & XEN_NETIF_EXTRA_FLAG_MORE);
1280 
1281 	return work_to_do;
1282 }
1283 
1284 static int xenvif_set_skb_gso(struct xenvif *vif,
1285 			      struct sk_buff *skb,
1286 			      struct xen_netif_extra_info *gso)
1287 {
1288 	if (!gso->u.gso.size) {
1289 		netdev_err(vif->dev, "GSO size must not be zero.\n");
1290 		xenvif_fatal_tx_err(vif);
1291 		return -EINVAL;
1292 	}
1293 
1294 	switch (gso->u.gso.type) {
1295 	case XEN_NETIF_GSO_TYPE_TCPV4:
1296 		skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1297 		break;
1298 	case XEN_NETIF_GSO_TYPE_TCPV6:
1299 		skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1300 		break;
1301 	default:
1302 		netdev_err(vif->dev, "Bad GSO type %d.\n", gso->u.gso.type);
1303 		xenvif_fatal_tx_err(vif);
1304 		return -EINVAL;
1305 	}
1306 
1307 	skb_shinfo(skb)->gso_size = gso->u.gso.size;
1308 	/* gso_segs will be calculated later */
1309 
1310 	return 0;
1311 }
1312 
1313 static int checksum_setup(struct xenvif_queue *queue, struct sk_buff *skb)
1314 {
1315 	bool recalculate_partial_csum = false;
1316 
1317 	/* A GSO SKB must be CHECKSUM_PARTIAL. However some buggy
1318 	 * peers can fail to set NETRXF_csum_blank when sending a GSO
1319 	 * frame. In this case force the SKB to CHECKSUM_PARTIAL and
1320 	 * recalculate the partial checksum.
1321 	 */
1322 	if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) {
1323 		queue->stats.rx_gso_checksum_fixup++;
1324 		skb->ip_summed = CHECKSUM_PARTIAL;
1325 		recalculate_partial_csum = true;
1326 	}
1327 
1328 	/* A non-CHECKSUM_PARTIAL SKB does not require setup. */
1329 	if (skb->ip_summed != CHECKSUM_PARTIAL)
1330 		return 0;
1331 
1332 	return skb_checksum_setup(skb, recalculate_partial_csum);
1333 }
1334 
1335 static bool tx_credit_exceeded(struct xenvif_queue *queue, unsigned size)
1336 {
1337 	u64 now = get_jiffies_64();
1338 	u64 next_credit = queue->credit_window_start +
1339 		msecs_to_jiffies(queue->credit_usec / 1000);
1340 
1341 	/* Timer could already be pending in rare cases. */
1342 	if (timer_pending(&queue->credit_timeout))
1343 		return true;
1344 
1345 	/* Passed the point where we can replenish credit? */
1346 	if (time_after_eq64(now, next_credit)) {
1347 		queue->credit_window_start = now;
1348 		tx_add_credit(queue);
1349 	}
1350 
1351 	/* Still too big to send right now? Set a callback. */
1352 	if (size > queue->remaining_credit) {
1353 		queue->credit_timeout.data     =
1354 			(unsigned long)queue;
1355 		queue->credit_timeout.function =
1356 			tx_credit_callback;
1357 		mod_timer(&queue->credit_timeout,
1358 			  next_credit);
1359 		queue->credit_window_start = next_credit;
1360 
1361 		return true;
1362 	}
1363 
1364 	return false;
1365 }
1366 
1367 static void xenvif_tx_build_gops(struct xenvif_queue *queue,
1368 				     int budget,
1369 				     unsigned *copy_ops,
1370 				     unsigned *map_ops)
1371 {
1372 	struct gnttab_map_grant_ref *gop = queue->tx_map_ops, *request_gop;
1373 	struct sk_buff *skb;
1374 	int ret;
1375 
1376 	while (skb_queue_len(&queue->tx_queue) < budget) {
1377 		struct xen_netif_tx_request txreq;
1378 		struct xen_netif_tx_request txfrags[XEN_NETBK_LEGACY_SLOTS_MAX];
1379 		struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX-1];
1380 		u16 pending_idx;
1381 		RING_IDX idx;
1382 		int work_to_do;
1383 		unsigned int data_len;
1384 		pending_ring_idx_t index;
1385 
1386 		if (queue->tx.sring->req_prod - queue->tx.req_cons >
1387 		    XEN_NETIF_TX_RING_SIZE) {
1388 			netdev_err(queue->vif->dev,
1389 				   "Impossible number of requests. "
1390 				   "req_prod %d, req_cons %d, size %ld\n",
1391 				   queue->tx.sring->req_prod, queue->tx.req_cons,
1392 				   XEN_NETIF_TX_RING_SIZE);
1393 			xenvif_fatal_tx_err(queue->vif);
1394 			break;
1395 		}
1396 
1397 		work_to_do = RING_HAS_UNCONSUMED_REQUESTS(&queue->tx);
1398 		if (!work_to_do)
1399 			break;
1400 
1401 		idx = queue->tx.req_cons;
1402 		rmb(); /* Ensure that we see the request before we copy it. */
1403 		memcpy(&txreq, RING_GET_REQUEST(&queue->tx, idx), sizeof(txreq));
1404 
1405 		/* Credit-based scheduling. */
1406 		if (txreq.size > queue->remaining_credit &&
1407 		    tx_credit_exceeded(queue, txreq.size))
1408 			break;
1409 
1410 		queue->remaining_credit -= txreq.size;
1411 
1412 		work_to_do--;
1413 		queue->tx.req_cons = ++idx;
1414 
1415 		memset(extras, 0, sizeof(extras));
1416 		if (txreq.flags & XEN_NETTXF_extra_info) {
1417 			work_to_do = xenvif_get_extras(queue, extras,
1418 						       work_to_do);
1419 			idx = queue->tx.req_cons;
1420 			if (unlikely(work_to_do < 0))
1421 				break;
1422 		}
1423 
1424 		ret = xenvif_count_requests(queue, &txreq, txfrags, work_to_do);
1425 		if (unlikely(ret < 0))
1426 			break;
1427 
1428 		idx += ret;
1429 
1430 		if (unlikely(txreq.size < ETH_HLEN)) {
1431 			netdev_dbg(queue->vif->dev,
1432 				   "Bad packet size: %d\n", txreq.size);
1433 			xenvif_tx_err(queue, &txreq, idx);
1434 			break;
1435 		}
1436 
1437 		/* No crossing a page as the payload mustn't fragment. */
1438 		if (unlikely((txreq.offset + txreq.size) > PAGE_SIZE)) {
1439 			netdev_err(queue->vif->dev,
1440 				   "txreq.offset: %x, size: %u, end: %lu\n",
1441 				   txreq.offset, txreq.size,
1442 				   (txreq.offset&~PAGE_MASK) + txreq.size);
1443 			xenvif_fatal_tx_err(queue->vif);
1444 			break;
1445 		}
1446 
1447 		index = pending_index(queue->pending_cons);
1448 		pending_idx = queue->pending_ring[index];
1449 
1450 		data_len = (txreq.size > XEN_NETBACK_TX_COPY_LEN &&
1451 			    ret < XEN_NETBK_LEGACY_SLOTS_MAX) ?
1452 			XEN_NETBACK_TX_COPY_LEN : txreq.size;
1453 
1454 		skb = xenvif_alloc_skb(data_len);
1455 		if (unlikely(skb == NULL)) {
1456 			netdev_dbg(queue->vif->dev,
1457 				   "Can't allocate a skb in start_xmit.\n");
1458 			xenvif_tx_err(queue, &txreq, idx);
1459 			break;
1460 		}
1461 
1462 		if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) {
1463 			struct xen_netif_extra_info *gso;
1464 			gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1];
1465 
1466 			if (xenvif_set_skb_gso(queue->vif, skb, gso)) {
1467 				/* Failure in xenvif_set_skb_gso is fatal. */
1468 				kfree_skb(skb);
1469 				break;
1470 			}
1471 		}
1472 
1473 		XENVIF_TX_CB(skb)->pending_idx = pending_idx;
1474 
1475 		__skb_put(skb, data_len);
1476 		queue->tx_copy_ops[*copy_ops].source.u.ref = txreq.gref;
1477 		queue->tx_copy_ops[*copy_ops].source.domid = queue->vif->domid;
1478 		queue->tx_copy_ops[*copy_ops].source.offset = txreq.offset;
1479 
1480 		queue->tx_copy_ops[*copy_ops].dest.u.gmfn =
1481 			virt_to_mfn(skb->data);
1482 		queue->tx_copy_ops[*copy_ops].dest.domid = DOMID_SELF;
1483 		queue->tx_copy_ops[*copy_ops].dest.offset =
1484 			offset_in_page(skb->data);
1485 
1486 		queue->tx_copy_ops[*copy_ops].len = data_len;
1487 		queue->tx_copy_ops[*copy_ops].flags = GNTCOPY_source_gref;
1488 
1489 		(*copy_ops)++;
1490 
1491 		skb_shinfo(skb)->nr_frags = ret;
1492 		if (data_len < txreq.size) {
1493 			skb_shinfo(skb)->nr_frags++;
1494 			frag_set_pending_idx(&skb_shinfo(skb)->frags[0],
1495 					     pending_idx);
1496 			xenvif_tx_create_map_op(queue, pending_idx, &txreq, gop);
1497 			gop++;
1498 		} else {
1499 			frag_set_pending_idx(&skb_shinfo(skb)->frags[0],
1500 					     INVALID_PENDING_IDX);
1501 			memcpy(&queue->pending_tx_info[pending_idx].req, &txreq,
1502 			       sizeof(txreq));
1503 		}
1504 
1505 		queue->pending_cons++;
1506 
1507 		request_gop = xenvif_get_requests(queue, skb, txfrags, gop);
1508 		if (request_gop == NULL) {
1509 			kfree_skb(skb);
1510 			xenvif_tx_err(queue, &txreq, idx);
1511 			break;
1512 		}
1513 		gop = request_gop;
1514 
1515 		__skb_queue_tail(&queue->tx_queue, skb);
1516 
1517 		queue->tx.req_cons = idx;
1518 
1519 		if (((gop-queue->tx_map_ops) >= ARRAY_SIZE(queue->tx_map_ops)) ||
1520 		    (*copy_ops >= ARRAY_SIZE(queue->tx_copy_ops)))
1521 			break;
1522 	}
1523 
1524 	(*map_ops) = gop - queue->tx_map_ops;
1525 	return;
1526 }
1527 
1528 /* Consolidate skb with a frag_list into a brand new one with local pages on
1529  * frags. Returns 0 or -ENOMEM if can't allocate new pages.
1530  */
1531 static int xenvif_handle_frag_list(struct xenvif_queue *queue, struct sk_buff *skb)
1532 {
1533 	unsigned int offset = skb_headlen(skb);
1534 	skb_frag_t frags[MAX_SKB_FRAGS];
1535 	int i;
1536 	struct ubuf_info *uarg;
1537 	struct sk_buff *nskb = skb_shinfo(skb)->frag_list;
1538 
1539 	queue->stats.tx_zerocopy_sent += 2;
1540 	queue->stats.tx_frag_overflow++;
1541 
1542 	xenvif_fill_frags(queue, nskb);
1543 	/* Subtract frags size, we will correct it later */
1544 	skb->truesize -= skb->data_len;
1545 	skb->len += nskb->len;
1546 	skb->data_len += nskb->len;
1547 
1548 	/* create a brand new frags array and coalesce there */
1549 	for (i = 0; offset < skb->len; i++) {
1550 		struct page *page;
1551 		unsigned int len;
1552 
1553 		BUG_ON(i >= MAX_SKB_FRAGS);
1554 		page = alloc_page(GFP_ATOMIC);
1555 		if (!page) {
1556 			int j;
1557 			skb->truesize += skb->data_len;
1558 			for (j = 0; j < i; j++)
1559 				put_page(frags[j].page.p);
1560 			return -ENOMEM;
1561 		}
1562 
1563 		if (offset + PAGE_SIZE < skb->len)
1564 			len = PAGE_SIZE;
1565 		else
1566 			len = skb->len - offset;
1567 		if (skb_copy_bits(skb, offset, page_address(page), len))
1568 			BUG();
1569 
1570 		offset += len;
1571 		frags[i].page.p = page;
1572 		frags[i].page_offset = 0;
1573 		skb_frag_size_set(&frags[i], len);
1574 	}
1575 	/* swap out with old one */
1576 	memcpy(skb_shinfo(skb)->frags,
1577 	       frags,
1578 	       i * sizeof(skb_frag_t));
1579 	skb_shinfo(skb)->nr_frags = i;
1580 	skb->truesize += i * PAGE_SIZE;
1581 
1582 	/* remove traces of mapped pages and frag_list */
1583 	skb_frag_list_init(skb);
1584 	uarg = skb_shinfo(skb)->destructor_arg;
1585 	/* increase inflight counter to offset decrement in callback */
1586 	atomic_inc(&queue->inflight_packets);
1587 	uarg->callback(uarg, true);
1588 	skb_shinfo(skb)->destructor_arg = NULL;
1589 
1590 	xenvif_skb_zerocopy_prepare(queue, nskb);
1591 	kfree_skb(nskb);
1592 
1593 	return 0;
1594 }
1595 
1596 static int xenvif_tx_submit(struct xenvif_queue *queue)
1597 {
1598 	struct gnttab_map_grant_ref *gop_map = queue->tx_map_ops;
1599 	struct gnttab_copy *gop_copy = queue->tx_copy_ops;
1600 	struct sk_buff *skb;
1601 	int work_done = 0;
1602 
1603 	while ((skb = __skb_dequeue(&queue->tx_queue)) != NULL) {
1604 		struct xen_netif_tx_request *txp;
1605 		u16 pending_idx;
1606 		unsigned data_len;
1607 
1608 		pending_idx = XENVIF_TX_CB(skb)->pending_idx;
1609 		txp = &queue->pending_tx_info[pending_idx].req;
1610 
1611 		/* Check the remap error code. */
1612 		if (unlikely(xenvif_tx_check_gop(queue, skb, &gop_map, &gop_copy))) {
1613 			/* If there was an error, xenvif_tx_check_gop is
1614 			 * expected to release all the frags which were mapped,
1615 			 * so kfree_skb shouldn't do it again
1616 			 */
1617 			skb_shinfo(skb)->nr_frags = 0;
1618 			if (skb_has_frag_list(skb)) {
1619 				struct sk_buff *nskb =
1620 						skb_shinfo(skb)->frag_list;
1621 				skb_shinfo(nskb)->nr_frags = 0;
1622 			}
1623 			kfree_skb(skb);
1624 			continue;
1625 		}
1626 
1627 		data_len = skb->len;
1628 		callback_param(queue, pending_idx).ctx = NULL;
1629 		if (data_len < txp->size) {
1630 			/* Append the packet payload as a fragment. */
1631 			txp->offset += data_len;
1632 			txp->size -= data_len;
1633 		} else {
1634 			/* Schedule a response immediately. */
1635 			xenvif_idx_release(queue, pending_idx,
1636 					   XEN_NETIF_RSP_OKAY);
1637 		}
1638 
1639 		if (txp->flags & XEN_NETTXF_csum_blank)
1640 			skb->ip_summed = CHECKSUM_PARTIAL;
1641 		else if (txp->flags & XEN_NETTXF_data_validated)
1642 			skb->ip_summed = CHECKSUM_UNNECESSARY;
1643 
1644 		xenvif_fill_frags(queue, skb);
1645 
1646 		if (unlikely(skb_has_frag_list(skb))) {
1647 			if (xenvif_handle_frag_list(queue, skb)) {
1648 				if (net_ratelimit())
1649 					netdev_err(queue->vif->dev,
1650 						   "Not enough memory to consolidate frag_list!\n");
1651 				xenvif_skb_zerocopy_prepare(queue, skb);
1652 				kfree_skb(skb);
1653 				continue;
1654 			}
1655 		}
1656 
1657 		skb->dev      = queue->vif->dev;
1658 		skb->protocol = eth_type_trans(skb, skb->dev);
1659 		skb_reset_network_header(skb);
1660 
1661 		if (checksum_setup(queue, skb)) {
1662 			netdev_dbg(queue->vif->dev,
1663 				   "Can't setup checksum in net_tx_action\n");
1664 			/* We have to set this flag to trigger the callback */
1665 			if (skb_shinfo(skb)->destructor_arg)
1666 				xenvif_skb_zerocopy_prepare(queue, skb);
1667 			kfree_skb(skb);
1668 			continue;
1669 		}
1670 
1671 		skb_probe_transport_header(skb, 0);
1672 
1673 		/* If the packet is GSO then we will have just set up the
1674 		 * transport header offset in checksum_setup so it's now
1675 		 * straightforward to calculate gso_segs.
1676 		 */
1677 		if (skb_is_gso(skb)) {
1678 			int mss = skb_shinfo(skb)->gso_size;
1679 			int hdrlen = skb_transport_header(skb) -
1680 				skb_mac_header(skb) +
1681 				tcp_hdrlen(skb);
1682 
1683 			skb_shinfo(skb)->gso_segs =
1684 				DIV_ROUND_UP(skb->len - hdrlen, mss);
1685 		}
1686 
1687 		queue->stats.rx_bytes += skb->len;
1688 		queue->stats.rx_packets++;
1689 
1690 		work_done++;
1691 
1692 		/* Set this flag right before netif_receive_skb, otherwise
1693 		 * someone might think this packet already left netback, and
1694 		 * do a skb_copy_ubufs while we are still in control of the
1695 		 * skb. E.g. the __pskb_pull_tail earlier can do such thing.
1696 		 */
1697 		if (skb_shinfo(skb)->destructor_arg) {
1698 			xenvif_skb_zerocopy_prepare(queue, skb);
1699 			queue->stats.tx_zerocopy_sent++;
1700 		}
1701 
1702 		netif_receive_skb(skb);
1703 	}
1704 
1705 	return work_done;
1706 }
1707 
1708 void xenvif_zerocopy_callback(struct ubuf_info *ubuf, bool zerocopy_success)
1709 {
1710 	unsigned long flags;
1711 	pending_ring_idx_t index;
1712 	struct xenvif_queue *queue = ubuf_to_queue(ubuf);
1713 
1714 	/* This is the only place where we grab this lock, to protect callbacks
1715 	 * from each other.
1716 	 */
1717 	spin_lock_irqsave(&queue->callback_lock, flags);
1718 	do {
1719 		u16 pending_idx = ubuf->desc;
1720 		ubuf = (struct ubuf_info *) ubuf->ctx;
1721 		BUG_ON(queue->dealloc_prod - queue->dealloc_cons >=
1722 			MAX_PENDING_REQS);
1723 		index = pending_index(queue->dealloc_prod);
1724 		queue->dealloc_ring[index] = pending_idx;
1725 		/* Sync with xenvif_tx_dealloc_action:
1726 		 * insert idx then incr producer.
1727 		 */
1728 		smp_wmb();
1729 		queue->dealloc_prod++;
1730 	} while (ubuf);
1731 	wake_up(&queue->dealloc_wq);
1732 	spin_unlock_irqrestore(&queue->callback_lock, flags);
1733 
1734 	if (likely(zerocopy_success))
1735 		queue->stats.tx_zerocopy_success++;
1736 	else
1737 		queue->stats.tx_zerocopy_fail++;
1738 	xenvif_skb_zerocopy_complete(queue);
1739 }
1740 
1741 static inline void xenvif_tx_dealloc_action(struct xenvif_queue *queue)
1742 {
1743 	struct gnttab_unmap_grant_ref *gop;
1744 	pending_ring_idx_t dc, dp;
1745 	u16 pending_idx, pending_idx_release[MAX_PENDING_REQS];
1746 	unsigned int i = 0;
1747 
1748 	dc = queue->dealloc_cons;
1749 	gop = queue->tx_unmap_ops;
1750 
1751 	/* Free up any grants we have finished using */
1752 	do {
1753 		dp = queue->dealloc_prod;
1754 
1755 		/* Ensure we see all indices enqueued by all
1756 		 * xenvif_zerocopy_callback().
1757 		 */
1758 		smp_rmb();
1759 
1760 		while (dc != dp) {
1761 			BUG_ON(gop - queue->tx_unmap_ops > MAX_PENDING_REQS);
1762 			pending_idx =
1763 				queue->dealloc_ring[pending_index(dc++)];
1764 
1765 			pending_idx_release[gop-queue->tx_unmap_ops] =
1766 				pending_idx;
1767 			queue->pages_to_unmap[gop-queue->tx_unmap_ops] =
1768 				queue->mmap_pages[pending_idx];
1769 			gnttab_set_unmap_op(gop,
1770 					    idx_to_kaddr(queue, pending_idx),
1771 					    GNTMAP_host_map,
1772 					    queue->grant_tx_handle[pending_idx]);
1773 			xenvif_grant_handle_reset(queue, pending_idx);
1774 			++gop;
1775 		}
1776 
1777 	} while (dp != queue->dealloc_prod);
1778 
1779 	queue->dealloc_cons = dc;
1780 
1781 	if (gop - queue->tx_unmap_ops > 0) {
1782 		int ret;
1783 		ret = gnttab_unmap_refs(queue->tx_unmap_ops,
1784 					NULL,
1785 					queue->pages_to_unmap,
1786 					gop - queue->tx_unmap_ops);
1787 		if (ret) {
1788 			netdev_err(queue->vif->dev, "Unmap fail: nr_ops %tx ret %d\n",
1789 				   gop - queue->tx_unmap_ops, ret);
1790 			for (i = 0; i < gop - queue->tx_unmap_ops; ++i) {
1791 				if (gop[i].status != GNTST_okay)
1792 					netdev_err(queue->vif->dev,
1793 						   " host_addr: %llx handle: %x status: %d\n",
1794 						   gop[i].host_addr,
1795 						   gop[i].handle,
1796 						   gop[i].status);
1797 			}
1798 			BUG();
1799 		}
1800 	}
1801 
1802 	for (i = 0; i < gop - queue->tx_unmap_ops; ++i)
1803 		xenvif_idx_release(queue, pending_idx_release[i],
1804 				   XEN_NETIF_RSP_OKAY);
1805 }
1806 
1807 
1808 /* Called after netfront has transmitted */
1809 int xenvif_tx_action(struct xenvif_queue *queue, int budget)
1810 {
1811 	unsigned nr_mops, nr_cops = 0;
1812 	int work_done, ret;
1813 
1814 	if (unlikely(!tx_work_todo(queue)))
1815 		return 0;
1816 
1817 	xenvif_tx_build_gops(queue, budget, &nr_cops, &nr_mops);
1818 
1819 	if (nr_cops == 0)
1820 		return 0;
1821 
1822 	gnttab_batch_copy(queue->tx_copy_ops, nr_cops);
1823 	if (nr_mops != 0) {
1824 		ret = gnttab_map_refs(queue->tx_map_ops,
1825 				      NULL,
1826 				      queue->pages_to_map,
1827 				      nr_mops);
1828 		BUG_ON(ret);
1829 	}
1830 
1831 	work_done = xenvif_tx_submit(queue);
1832 
1833 	return work_done;
1834 }
1835 
1836 static void xenvif_idx_release(struct xenvif_queue *queue, u16 pending_idx,
1837 			       u8 status)
1838 {
1839 	struct pending_tx_info *pending_tx_info;
1840 	pending_ring_idx_t index;
1841 	unsigned long flags;
1842 
1843 	pending_tx_info = &queue->pending_tx_info[pending_idx];
1844 	spin_lock_irqsave(&queue->response_lock, flags);
1845 	make_tx_response(queue, &pending_tx_info->req, status);
1846 	index = pending_index(queue->pending_prod);
1847 	queue->pending_ring[index] = pending_idx;
1848 	/* TX shouldn't use the index before we give it back here */
1849 	mb();
1850 	queue->pending_prod++;
1851 	spin_unlock_irqrestore(&queue->response_lock, flags);
1852 }
1853 
1854 
1855 static void make_tx_response(struct xenvif_queue *queue,
1856 			     struct xen_netif_tx_request *txp,
1857 			     s8       st)
1858 {
1859 	RING_IDX i = queue->tx.rsp_prod_pvt;
1860 	struct xen_netif_tx_response *resp;
1861 	int notify;
1862 
1863 	resp = RING_GET_RESPONSE(&queue->tx, i);
1864 	resp->id     = txp->id;
1865 	resp->status = st;
1866 
1867 	if (txp->flags & XEN_NETTXF_extra_info)
1868 		RING_GET_RESPONSE(&queue->tx, ++i)->status = XEN_NETIF_RSP_NULL;
1869 
1870 	queue->tx.rsp_prod_pvt = ++i;
1871 	RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&queue->tx, notify);
1872 	if (notify)
1873 		notify_remote_via_irq(queue->tx_irq);
1874 }
1875 
1876 static struct xen_netif_rx_response *make_rx_response(struct xenvif_queue *queue,
1877 					     u16      id,
1878 					     s8       st,
1879 					     u16      offset,
1880 					     u16      size,
1881 					     u16      flags)
1882 {
1883 	RING_IDX i = queue->rx.rsp_prod_pvt;
1884 	struct xen_netif_rx_response *resp;
1885 
1886 	resp = RING_GET_RESPONSE(&queue->rx, i);
1887 	resp->offset     = offset;
1888 	resp->flags      = flags;
1889 	resp->id         = id;
1890 	resp->status     = (s16)size;
1891 	if (st < 0)
1892 		resp->status = (s16)st;
1893 
1894 	queue->rx.rsp_prod_pvt = ++i;
1895 
1896 	return resp;
1897 }
1898 
1899 void xenvif_idx_unmap(struct xenvif_queue *queue, u16 pending_idx)
1900 {
1901 	int ret;
1902 	struct gnttab_unmap_grant_ref tx_unmap_op;
1903 
1904 	gnttab_set_unmap_op(&tx_unmap_op,
1905 			    idx_to_kaddr(queue, pending_idx),
1906 			    GNTMAP_host_map,
1907 			    queue->grant_tx_handle[pending_idx]);
1908 	xenvif_grant_handle_reset(queue, pending_idx);
1909 
1910 	ret = gnttab_unmap_refs(&tx_unmap_op, NULL,
1911 				&queue->mmap_pages[pending_idx], 1);
1912 	if (ret) {
1913 		netdev_err(queue->vif->dev,
1914 			   "Unmap fail: ret: %d pending_idx: %d host_addr: %llx handle: %x status: %d\n",
1915 			   ret,
1916 			   pending_idx,
1917 			   tx_unmap_op.host_addr,
1918 			   tx_unmap_op.handle,
1919 			   tx_unmap_op.status);
1920 		BUG();
1921 	}
1922 }
1923 
1924 static inline int tx_work_todo(struct xenvif_queue *queue)
1925 {
1926 	if (likely(RING_HAS_UNCONSUMED_REQUESTS(&queue->tx)))
1927 		return 1;
1928 
1929 	return 0;
1930 }
1931 
1932 static inline bool tx_dealloc_work_todo(struct xenvif_queue *queue)
1933 {
1934 	return queue->dealloc_cons != queue->dealloc_prod;
1935 }
1936 
1937 void xenvif_unmap_frontend_rings(struct xenvif_queue *queue)
1938 {
1939 	if (queue->tx.sring)
1940 		xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(queue->vif),
1941 					queue->tx.sring);
1942 	if (queue->rx.sring)
1943 		xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(queue->vif),
1944 					queue->rx.sring);
1945 }
1946 
1947 int xenvif_map_frontend_rings(struct xenvif_queue *queue,
1948 			      grant_ref_t tx_ring_ref,
1949 			      grant_ref_t rx_ring_ref)
1950 {
1951 	void *addr;
1952 	struct xen_netif_tx_sring *txs;
1953 	struct xen_netif_rx_sring *rxs;
1954 
1955 	int err = -ENOMEM;
1956 
1957 	err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(queue->vif),
1958 				     tx_ring_ref, &addr);
1959 	if (err)
1960 		goto err;
1961 
1962 	txs = (struct xen_netif_tx_sring *)addr;
1963 	BACK_RING_INIT(&queue->tx, txs, PAGE_SIZE);
1964 
1965 	err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(queue->vif),
1966 				     rx_ring_ref, &addr);
1967 	if (err)
1968 		goto err;
1969 
1970 	rxs = (struct xen_netif_rx_sring *)addr;
1971 	BACK_RING_INIT(&queue->rx, rxs, PAGE_SIZE);
1972 
1973 	return 0;
1974 
1975 err:
1976 	xenvif_unmap_frontend_rings(queue);
1977 	return err;
1978 }
1979 
1980 static void xenvif_queue_carrier_off(struct xenvif_queue *queue)
1981 {
1982 	struct xenvif *vif = queue->vif;
1983 
1984 	queue->stalled = true;
1985 
1986 	/* At least one queue has stalled? Disable the carrier. */
1987 	spin_lock(&vif->lock);
1988 	if (vif->stalled_queues++ == 0) {
1989 		netdev_info(vif->dev, "Guest Rx stalled");
1990 		netif_carrier_off(vif->dev);
1991 	}
1992 	spin_unlock(&vif->lock);
1993 }
1994 
1995 static void xenvif_queue_carrier_on(struct xenvif_queue *queue)
1996 {
1997 	struct xenvif *vif = queue->vif;
1998 
1999 	queue->last_rx_time = jiffies; /* Reset Rx stall detection. */
2000 	queue->stalled = false;
2001 
2002 	/* All queues are ready? Enable the carrier. */
2003 	spin_lock(&vif->lock);
2004 	if (--vif->stalled_queues == 0) {
2005 		netdev_info(vif->dev, "Guest Rx ready");
2006 		netif_carrier_on(vif->dev);
2007 	}
2008 	spin_unlock(&vif->lock);
2009 }
2010 
2011 static bool xenvif_rx_queue_stalled(struct xenvif_queue *queue)
2012 {
2013 	RING_IDX prod, cons;
2014 
2015 	prod = queue->rx.sring->req_prod;
2016 	cons = queue->rx.req_cons;
2017 
2018 	return !queue->stalled
2019 		&& prod - cons < XEN_NETBK_RX_SLOTS_MAX
2020 		&& time_after(jiffies,
2021 			      queue->last_rx_time + queue->vif->stall_timeout);
2022 }
2023 
2024 static bool xenvif_rx_queue_ready(struct xenvif_queue *queue)
2025 {
2026 	RING_IDX prod, cons;
2027 
2028 	prod = queue->rx.sring->req_prod;
2029 	cons = queue->rx.req_cons;
2030 
2031 	return queue->stalled
2032 		&& prod - cons >= XEN_NETBK_RX_SLOTS_MAX;
2033 }
2034 
2035 static bool xenvif_have_rx_work(struct xenvif_queue *queue)
2036 {
2037 	return (!skb_queue_empty(&queue->rx_queue)
2038 		&& xenvif_rx_ring_slots_available(queue, XEN_NETBK_RX_SLOTS_MAX))
2039 		|| (queue->vif->stall_timeout &&
2040 		    (xenvif_rx_queue_stalled(queue)
2041 		     || xenvif_rx_queue_ready(queue)))
2042 		|| kthread_should_stop()
2043 		|| queue->vif->disabled;
2044 }
2045 
2046 static long xenvif_rx_queue_timeout(struct xenvif_queue *queue)
2047 {
2048 	struct sk_buff *skb;
2049 	long timeout;
2050 
2051 	skb = skb_peek(&queue->rx_queue);
2052 	if (!skb)
2053 		return MAX_SCHEDULE_TIMEOUT;
2054 
2055 	timeout = XENVIF_RX_CB(skb)->expires - jiffies;
2056 	return timeout < 0 ? 0 : timeout;
2057 }
2058 
2059 /* Wait until the guest Rx thread has work.
2060  *
2061  * The timeout needs to be adjusted based on the current head of the
2062  * queue (and not just the head at the beginning).  In particular, if
2063  * the queue is initially empty an infinite timeout is used and this
2064  * needs to be reduced when a skb is queued.
2065  *
2066  * This cannot be done with wait_event_timeout() because it only
2067  * calculates the timeout once.
2068  */
2069 static void xenvif_wait_for_rx_work(struct xenvif_queue *queue)
2070 {
2071 	DEFINE_WAIT(wait);
2072 
2073 	if (xenvif_have_rx_work(queue))
2074 		return;
2075 
2076 	for (;;) {
2077 		long ret;
2078 
2079 		prepare_to_wait(&queue->wq, &wait, TASK_INTERRUPTIBLE);
2080 		if (xenvif_have_rx_work(queue))
2081 			break;
2082 		ret = schedule_timeout(xenvif_rx_queue_timeout(queue));
2083 		if (!ret)
2084 			break;
2085 	}
2086 	finish_wait(&queue->wq, &wait);
2087 }
2088 
2089 int xenvif_kthread_guest_rx(void *data)
2090 {
2091 	struct xenvif_queue *queue = data;
2092 	struct xenvif *vif = queue->vif;
2093 
2094 	if (!vif->stall_timeout)
2095 		xenvif_queue_carrier_on(queue);
2096 
2097 	for (;;) {
2098 		xenvif_wait_for_rx_work(queue);
2099 
2100 		if (kthread_should_stop())
2101 			break;
2102 
2103 		/* This frontend is found to be rogue, disable it in
2104 		 * kthread context. Currently this is only set when
2105 		 * netback finds out frontend sends malformed packet,
2106 		 * but we cannot disable the interface in softirq
2107 		 * context so we defer it here, if this thread is
2108 		 * associated with queue 0.
2109 		 */
2110 		if (unlikely(vif->disabled && queue->id == 0)) {
2111 			xenvif_carrier_off(vif);
2112 			xenvif_rx_queue_purge(queue);
2113 			continue;
2114 		}
2115 
2116 		if (!skb_queue_empty(&queue->rx_queue))
2117 			xenvif_rx_action(queue);
2118 
2119 		/* If the guest hasn't provided any Rx slots for a
2120 		 * while it's probably not responsive, drop the
2121 		 * carrier so packets are dropped earlier.
2122 		 */
2123 		if (vif->stall_timeout) {
2124 			if (xenvif_rx_queue_stalled(queue))
2125 				xenvif_queue_carrier_off(queue);
2126 			else if (xenvif_rx_queue_ready(queue))
2127 				xenvif_queue_carrier_on(queue);
2128 		}
2129 
2130 		/* Queued packets may have foreign pages from other
2131 		 * domains.  These cannot be queued indefinitely as
2132 		 * this would starve guests of grant refs and transmit
2133 		 * slots.
2134 		 */
2135 		xenvif_rx_queue_drop_expired(queue);
2136 
2137 		xenvif_rx_queue_maybe_wake(queue);
2138 
2139 		cond_resched();
2140 	}
2141 
2142 	/* Bin any remaining skbs */
2143 	xenvif_rx_queue_purge(queue);
2144 
2145 	return 0;
2146 }
2147 
2148 static bool xenvif_dealloc_kthread_should_stop(struct xenvif_queue *queue)
2149 {
2150 	/* Dealloc thread must remain running until all inflight
2151 	 * packets complete.
2152 	 */
2153 	return kthread_should_stop() &&
2154 		!atomic_read(&queue->inflight_packets);
2155 }
2156 
2157 int xenvif_dealloc_kthread(void *data)
2158 {
2159 	struct xenvif_queue *queue = data;
2160 
2161 	for (;;) {
2162 		wait_event_interruptible(queue->dealloc_wq,
2163 					 tx_dealloc_work_todo(queue) ||
2164 					 xenvif_dealloc_kthread_should_stop(queue));
2165 		if (xenvif_dealloc_kthread_should_stop(queue))
2166 			break;
2167 
2168 		xenvif_tx_dealloc_action(queue);
2169 		cond_resched();
2170 	}
2171 
2172 	/* Unmap anything remaining*/
2173 	if (tx_dealloc_work_todo(queue))
2174 		xenvif_tx_dealloc_action(queue);
2175 
2176 	return 0;
2177 }
2178 
2179 static int __init netback_init(void)
2180 {
2181 	int rc = 0;
2182 
2183 	if (!xen_domain())
2184 		return -ENODEV;
2185 
2186 	/* Allow as many queues as there are CPUs, by default */
2187 	xenvif_max_queues = num_online_cpus();
2188 
2189 	if (fatal_skb_slots < XEN_NETBK_LEGACY_SLOTS_MAX) {
2190 		pr_info("fatal_skb_slots too small (%d), bump it to XEN_NETBK_LEGACY_SLOTS_MAX (%d)\n",
2191 			fatal_skb_slots, XEN_NETBK_LEGACY_SLOTS_MAX);
2192 		fatal_skb_slots = XEN_NETBK_LEGACY_SLOTS_MAX;
2193 	}
2194 
2195 	rc = xenvif_xenbus_init();
2196 	if (rc)
2197 		goto failed_init;
2198 
2199 #ifdef CONFIG_DEBUG_FS
2200 	xen_netback_dbg_root = debugfs_create_dir("xen-netback", NULL);
2201 	if (IS_ERR_OR_NULL(xen_netback_dbg_root))
2202 		pr_warn("Init of debugfs returned %ld!\n",
2203 			PTR_ERR(xen_netback_dbg_root));
2204 #endif /* CONFIG_DEBUG_FS */
2205 
2206 	return 0;
2207 
2208 failed_init:
2209 	return rc;
2210 }
2211 
2212 module_init(netback_init);
2213 
2214 static void __exit netback_fini(void)
2215 {
2216 #ifdef CONFIG_DEBUG_FS
2217 	if (!IS_ERR_OR_NULL(xen_netback_dbg_root))
2218 		debugfs_remove_recursive(xen_netback_dbg_root);
2219 #endif /* CONFIG_DEBUG_FS */
2220 	xenvif_xenbus_fini();
2221 }
2222 module_exit(netback_fini);
2223 
2224 MODULE_LICENSE("Dual BSD/GPL");
2225 MODULE_ALIAS("xen-backend:vif");
2226