?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
xen: Use more current logging styles
[linux-2.6.git] / drivers / net / xen-netback / netback.c
blob64828de25d9a252119a16bb805bfe4fb3ed4b809
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
41 #include <net/tcp.h>
42
43 #include <xen/xen.h>
44 #include <xen/events.h>
45 #include <xen/interface/memory.h>
46
47 #include <asm/xen/hypercall.h>
48 #include <asm/xen/page.h>
49
50 /* Provide an option to disable split event channels at load time as
51 * event channels are limited resource. Split event channels are
52 * enabled by default.
53 */
54 bool separate_tx_rx_irq = 1;
55 module_param(separate_tx_rx_irq, bool, 0644);
56
57 /*
58 * This is the maximum slots a skb can have. If a guest sends a skb
59 * which exceeds this limit it is considered malicious.
60 */
61 #define FATAL_SKB_SLOTS_DEFAULT 20
62 static unsigned int fatal_skb_slots = FATAL_SKB_SLOTS_DEFAULT;
63 module_param(fatal_skb_slots, uint, 0444);
64
65 /*
66 * To avoid confusion, we define XEN_NETBK_LEGACY_SLOTS_MAX indicating
67 * the maximum slots a valid packet can use. Now this value is defined
68 * to be XEN_NETIF_NR_SLOTS_MIN, which is supposed to be supported by
69 * all backend.
70 */
71 #define XEN_NETBK_LEGACY_SLOTS_MAX XEN_NETIF_NR_SLOTS_MIN
72
73 typedef unsigned int pending_ring_idx_t;
74 #define INVALID_PENDING_RING_IDX (~0U)
75
76 struct pending_tx_info {
77 struct xen_netif_tx_request req; /* coalesced tx request */
78 struct xenvif *vif;
79 pending_ring_idx_t head; /* head != INVALID_PENDING_RING_IDX
80 * if it is head of one or more tx
81 * reqs
82 */
83 };
84
85 struct netbk_rx_meta {
86 int id;
87 int size;
88 int gso_size;
89 };
90
91 #define MAX_PENDING_REQS 256
92
93 /* Discriminate from any valid pending_idx value. */
94 #define INVALID_PENDING_IDX 0xFFFF
95
96 #define MAX_BUFFER_OFFSET PAGE_SIZE
97
98 /* extra field used in struct page */
99 union page_ext {
100 struct {
101 #if BITS_PER_LONG < 64
102 #define IDX_WIDTH 8
103 #define GROUP_WIDTH (BITS_PER_LONG - IDX_WIDTH)
104 unsigned int group:GROUP_WIDTH;
105 unsigned int idx:IDX_WIDTH;
106 #else
107 unsigned int group, idx;
108 #endif
109 } e;
110 void *mapping;
111 };
112
113 struct xen_netbk {
114 wait_queue_head_t wq;
115 struct task_struct *task;
116
117 struct sk_buff_head rx_queue;
118 struct sk_buff_head tx_queue;
119
120 struct timer_list net_timer;
121
122 struct page *mmap_pages[MAX_PENDING_REQS];
123
124 pending_ring_idx_t pending_prod;
125 pending_ring_idx_t pending_cons;
126 struct list_head net_schedule_list;
127
128 /* Protect the net_schedule_list in netif. */
129 spinlock_t net_schedule_list_lock;
130
131 atomic_t netfront_count;
132
133 struct pending_tx_info pending_tx_info[MAX_PENDING_REQS];
134 /* Coalescing tx requests before copying makes number of grant
135 * copy ops greater or equal to number of slots required. In
136 * worst case a tx request consumes 2 gnttab_copy.
137 */
138 struct gnttab_copy tx_copy_ops[2*MAX_PENDING_REQS];
139
140 u16 pending_ring[MAX_PENDING_REQS];
141
142 /*
143 * Given MAX_BUFFER_OFFSET of 4096 the worst case is that each
144 * head/fragment page uses 2 copy operations because it
145 * straddles two buffers in the frontend.
146 */
147 struct gnttab_copy grant_copy_op[2*XEN_NETIF_RX_RING_SIZE];
148 struct netbk_rx_meta meta[2*XEN_NETIF_RX_RING_SIZE];
149 };
150
151 static struct xen_netbk *xen_netbk;
152 static int xen_netbk_group_nr;
153
154 /*
155 * If head != INVALID_PENDING_RING_IDX, it means this tx request is head of
156 * one or more merged tx requests, otherwise it is the continuation of
157 * previous tx request.
158 */
159 static inline int pending_tx_is_head(struct xen_netbk *netbk, RING_IDX idx)
160 {
161 return netbk->pending_tx_info[idx].head != INVALID_PENDING_RING_IDX;
162 }
163
164 void xen_netbk_add_xenvif(struct xenvif *vif)
165 {
166 int i;
167 int min_netfront_count;
168 int min_group = 0;
169 struct xen_netbk *netbk;
170
171 min_netfront_count = atomic_read(&xen_netbk[0].netfront_count);
172 for (i = 0; i < xen_netbk_group_nr; i++) {
173 int netfront_count = atomic_read(&xen_netbk[i].netfront_count);
174 if (netfront_count < min_netfront_count) {
175 min_group = i;
176 min_netfront_count = netfront_count;
177 }
178 }
179
180 netbk = &xen_netbk[min_group];
181
182 vif->netbk = netbk;
183 atomic_inc(&netbk->netfront_count);
184 }
185
186 void xen_netbk_remove_xenvif(struct xenvif *vif)
187 {
188 struct xen_netbk *netbk = vif->netbk;
189 vif->netbk = NULL;
190 atomic_dec(&netbk->netfront_count);
191 }
192
193 static void xen_netbk_idx_release(struct xen_netbk *netbk, u16 pending_idx,
194 u8 status);
195 static void make_tx_response(struct xenvif *vif,
196 struct xen_netif_tx_request *txp,
197 s8 st);
198 static struct xen_netif_rx_response *make_rx_response(struct xenvif *vif,
199 u16 id,
200 s8 st,
201 u16 offset,
202 u16 size,
203 u16 flags);
204
205 static inline unsigned long idx_to_pfn(struct xen_netbk *netbk,
206 u16 idx)
207 {
208 return page_to_pfn(netbk->mmap_pages[idx]);
209 }
210
211 static inline unsigned long idx_to_kaddr(struct xen_netbk *netbk,
212 u16 idx)
213 {
214 return (unsigned long)pfn_to_kaddr(idx_to_pfn(netbk, idx));
215 }
216
217 /* extra field used in struct page */
218 static inline void set_page_ext(struct page *pg, struct xen_netbk *netbk,
219 unsigned int idx)
220 {
221 unsigned int group = netbk - xen_netbk;
222 union page_ext ext = { .e = { .group = group + 1, .idx = idx } };
223
224 BUILD_BUG_ON(sizeof(ext) > sizeof(ext.mapping));
225 pg->mapping = ext.mapping;
226 }
227
228 static int get_page_ext(struct page *pg,
229 unsigned int *pgroup, unsigned int *pidx)
230 {
231 union page_ext ext = { .mapping = pg->mapping };
232 struct xen_netbk *netbk;
233 unsigned int group, idx;
234
235 group = ext.e.group - 1;
236
237 if (group < 0 || group >= xen_netbk_group_nr)
238 return 0;
239
240 netbk = &xen_netbk[group];
241
242 idx = ext.e.idx;
243
244 if ((idx < 0) || (idx >= MAX_PENDING_REQS))
245 return 0;
246
247 if (netbk->mmap_pages[idx] != pg)
248 return 0;
249
250 *pgroup = group;
251 *pidx = idx;
252
253 return 1;
254 }
255
256 /*
257 * This is the amount of packet we copy rather than map, so that the
258 * guest can't fiddle with the contents of the headers while we do
259 * packet processing on them (netfilter, routing, etc).
260 */
261 #define PKT_PROT_LEN (ETH_HLEN + \
262 VLAN_HLEN + \
263 sizeof(struct iphdr) + MAX_IPOPTLEN + \
264 sizeof(struct tcphdr) + MAX_TCP_OPTION_SPACE)
265
266 static u16 frag_get_pending_idx(skb_frag_t *frag)
267 {
268 return (u16)frag->page_offset;
269 }
270
271 static void frag_set_pending_idx(skb_frag_t *frag, u16 pending_idx)
272 {
273 frag->page_offset = pending_idx;
274 }
275
276 static inline pending_ring_idx_t pending_index(unsigned i)
277 {
278 return i & (MAX_PENDING_REQS-1);
279 }
280
281 static inline pending_ring_idx_t nr_pending_reqs(struct xen_netbk *netbk)
282 {
283 return MAX_PENDING_REQS -
284 netbk->pending_prod + netbk->pending_cons;
285 }
286
287 static void xen_netbk_kick_thread(struct xen_netbk *netbk)
288 {
289 wake_up(&netbk->wq);
290 }
291
292 static int max_required_rx_slots(struct xenvif *vif)
293 {
294 int max = DIV_ROUND_UP(vif->dev->mtu, PAGE_SIZE);
295
296 /* XXX FIXME: RX path dependent on MAX_SKB_FRAGS */
297 if (vif->can_sg || vif->gso || vif->gso_prefix)
298 max += MAX_SKB_FRAGS + 1; /* extra_info + frags */
299
300 return max;
301 }
302
303 int xen_netbk_rx_ring_full(struct xenvif *vif)
304 {
305 RING_IDX peek = vif->rx_req_cons_peek;
306 RING_IDX needed = max_required_rx_slots(vif);
307
308 return ((vif->rx.sring->req_prod - peek) < needed) ||
309 ((vif->rx.rsp_prod_pvt + XEN_NETIF_RX_RING_SIZE - peek) < needed);
310 }
311
312 int xen_netbk_must_stop_queue(struct xenvif *vif)
313 {
314 if (!xen_netbk_rx_ring_full(vif))
315 return 0;
316
317 vif->rx.sring->req_event = vif->rx_req_cons_peek +
318 max_required_rx_slots(vif);
319 mb(); /* request notification /then/ check the queue */
320
321 return xen_netbk_rx_ring_full(vif);
322 }
323
324 /*
325 * Returns true if we should start a new receive buffer instead of
326 * adding 'size' bytes to a buffer which currently contains 'offset'
327 * bytes.
328 */
329 static bool start_new_rx_buffer(int offset, unsigned long size, int head)
330 {
331 /* simple case: we have completely filled the current buffer. */
332 if (offset == MAX_BUFFER_OFFSET)
333 return true;
334
335 /*
336 * complex case: start a fresh buffer if the current frag
337 * would overflow the current buffer but only if:
338 * (i) this frag would fit completely in the next buffer
339 * and (ii) there is already some data in the current buffer
340 * and (iii) this is not the head buffer.
341 *
342 * Where:
343 * - (i) stops us splitting a frag into two copies
344 * unless the frag is too large for a single buffer.
345 * - (ii) stops us from leaving a buffer pointlessly empty.
346 * - (iii) stops us leaving the first buffer
347 * empty. Strictly speaking this is already covered
348 * by (ii) but is explicitly checked because
349 * netfront relies on the first buffer being
350 * non-empty and can crash otherwise.
351 *
352 * This means we will effectively linearise small
353 * frags but do not needlessly split large buffers
354 * into multiple copies tend to give large frags their
355 * own buffers as before.
356 */
357 if ((offset + size > MAX_BUFFER_OFFSET) &&
358 (size <= MAX_BUFFER_OFFSET) && offset && !head)
359 return true;
360
361 return false;
362 }
363
364 /*
365 * Figure out how many ring slots we're going to need to send @skb to
366 * the guest. This function is essentially a dry run of
367 * netbk_gop_frag_copy.
368 */
369 unsigned int xen_netbk_count_skb_slots(struct xenvif *vif, struct sk_buff *skb)
370 {
371 unsigned int count;
372 int i, copy_off;
373
374 count = DIV_ROUND_UP(skb_headlen(skb), PAGE_SIZE);
375
376 copy_off = skb_headlen(skb) % PAGE_SIZE;
377
378 if (skb_shinfo(skb)->gso_size)
379 count++;
380
381 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
382 unsigned long size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
383 unsigned long offset = skb_shinfo(skb)->frags[i].page_offset;
384 unsigned long bytes;
385
386 offset &= ~PAGE_MASK;
387
388 while (size > 0) {
389 BUG_ON(offset >= PAGE_SIZE);
390 BUG_ON(copy_off > MAX_BUFFER_OFFSET);
391
392 bytes = PAGE_SIZE - offset;
393
394 if (bytes > size)
395 bytes = size;
396
397 if (start_new_rx_buffer(copy_off, bytes, 0)) {
398 count++;
399 copy_off = 0;
400 }
401
402 if (copy_off + bytes > MAX_BUFFER_OFFSET)
403 bytes = MAX_BUFFER_OFFSET - copy_off;
404
405 copy_off += bytes;
406
407 offset += bytes;
408 size -= bytes;
409
410 if (offset == PAGE_SIZE)
411 offset = 0;
412 }
413 }
414 return count;
415 }
416
417 struct netrx_pending_operations {
418 unsigned copy_prod, copy_cons;
419 unsigned meta_prod, meta_cons;
420 struct gnttab_copy *copy;
421 struct netbk_rx_meta *meta;
422 int copy_off;
423 grant_ref_t copy_gref;
424 };
425
426 static struct netbk_rx_meta *get_next_rx_buffer(struct xenvif *vif,
427 struct netrx_pending_operations *npo)
428 {
429 struct netbk_rx_meta *meta;
430 struct xen_netif_rx_request *req;
431
432 req = RING_GET_REQUEST(&vif->rx, vif->rx.req_cons++);
433
434 meta = npo->meta + npo->meta_prod++;
435 meta->gso_size = 0;
436 meta->size = 0;
437 meta->id = req->id;
438
439 npo->copy_off = 0;
440 npo->copy_gref = req->gref;
441
442 return meta;
443 }
444
445 /*
446 * Set up the grant operations for this fragment. If it's a flipping
447 * interface, we also set up the unmap request from here.
448 */
449 static void netbk_gop_frag_copy(struct xenvif *vif, struct sk_buff *skb,
450 struct netrx_pending_operations *npo,
451 struct page *page, unsigned long size,
452 unsigned long offset, int *head)
453 {
454 struct gnttab_copy *copy_gop;
455 struct netbk_rx_meta *meta;
456 /*
457 * These variables are used iff get_page_ext returns true,
458 * in which case they are guaranteed to be initialized.
459 */
460 unsigned int uninitialized_var(group), uninitialized_var(idx);
461 int foreign = get_page_ext(page, &group, &idx);
462 unsigned long bytes;
463
464 /* Data must not cross a page boundary. */
465 BUG_ON(size + offset > PAGE_SIZE<<compound_order(page));
466
467 meta = npo->meta + npo->meta_prod - 1;
468
469 /* Skip unused frames from start of page */
470 page += offset >> PAGE_SHIFT;
471 offset &= ~PAGE_MASK;
472
473 while (size > 0) {
474 BUG_ON(offset >= PAGE_SIZE);
475 BUG_ON(npo->copy_off > MAX_BUFFER_OFFSET);
476
477 bytes = PAGE_SIZE - offset;
478
479 if (bytes > size)
480 bytes = size;
481
482 if (start_new_rx_buffer(npo->copy_off, bytes, *head)) {
483 /*
484 * Netfront requires there to be some data in the head
485 * buffer.
486 */
487 BUG_ON(*head);
488
489 meta = get_next_rx_buffer(vif, npo);
490 }
491
492 if (npo->copy_off + bytes > MAX_BUFFER_OFFSET)
493 bytes = MAX_BUFFER_OFFSET - npo->copy_off;
494
495 copy_gop = npo->copy + npo->copy_prod++;
496 copy_gop->flags = GNTCOPY_dest_gref;
497 if (foreign) {
498 struct xen_netbk *netbk = &xen_netbk[group];
499 struct pending_tx_info *src_pend;
500
501 src_pend = &netbk->pending_tx_info[idx];
502
503 copy_gop->source.domid = src_pend->vif->domid;
504 copy_gop->source.u.ref = src_pend->req.gref;
505 copy_gop->flags |= GNTCOPY_source_gref;
506 } else {
507 void *vaddr = page_address(page);
508 copy_gop->source.domid = DOMID_SELF;
509 copy_gop->source.u.gmfn = virt_to_mfn(vaddr);
510 }
511 copy_gop->source.offset = offset;
512 copy_gop->dest.domid = vif->domid;
513
514 copy_gop->dest.offset = npo->copy_off;
515 copy_gop->dest.u.ref = npo->copy_gref;
516 copy_gop->len = bytes;
517
518 npo->copy_off += bytes;
519 meta->size += bytes;
520
521 offset += bytes;
522 size -= bytes;
523
524 /* Next frame */
525 if (offset == PAGE_SIZE && size) {
526 BUG_ON(!PageCompound(page));
527 page++;
528 offset = 0;
529 }
530
531 /* Leave a gap for the GSO descriptor. */
532 if (*head && skb_shinfo(skb)->gso_size && !vif->gso_prefix)
533 vif->rx.req_cons++;
534
535 *head = 0; /* There must be something in this buffer now. */
536
537 }
538 }
539
540 /*
541 * Prepare an SKB to be transmitted to the frontend.
542 *
543 * This function is responsible for allocating grant operations, meta
544 * structures, etc.
545 *
546 * It returns the number of meta structures consumed. The number of
547 * ring slots used is always equal to the number of meta slots used
548 * plus the number of GSO descriptors used. Currently, we use either
549 * zero GSO descriptors (for non-GSO packets) or one descriptor (for
550 * frontend-side LRO).
551 */
552 static int netbk_gop_skb(struct sk_buff *skb,
553 struct netrx_pending_operations *npo)
554 {
555 struct xenvif *vif = netdev_priv(skb->dev);
556 int nr_frags = skb_shinfo(skb)->nr_frags;
557 int i;
558 struct xen_netif_rx_request *req;
559 struct netbk_rx_meta *meta;
560 unsigned char *data;
561 int head = 1;
562 int old_meta_prod;
563
564 old_meta_prod = npo->meta_prod;
565
566 /* Set up a GSO prefix descriptor, if necessary */
567 if (skb_shinfo(skb)->gso_size && vif->gso_prefix) {
568 req = RING_GET_REQUEST(&vif->rx, vif->rx.req_cons++);
569 meta = npo->meta + npo->meta_prod++;
570 meta->gso_size = skb_shinfo(skb)->gso_size;
571 meta->size = 0;
572 meta->id = req->id;
573 }
574
575 req = RING_GET_REQUEST(&vif->rx, vif->rx.req_cons++);
576 meta = npo->meta + npo->meta_prod++;
577
578 if (!vif->gso_prefix)
579 meta->gso_size = skb_shinfo(skb)->gso_size;
580 else
581 meta->gso_size = 0;
582
583 meta->size = 0;
584 meta->id = req->id;
585 npo->copy_off = 0;
586 npo->copy_gref = req->gref;
587
588 data = skb->data;
589 while (data < skb_tail_pointer(skb)) {
590 unsigned int offset = offset_in_page(data);
591 unsigned int len = PAGE_SIZE - offset;
592
593 if (data + len > skb_tail_pointer(skb))
594 len = skb_tail_pointer(skb) - data;
595
596 netbk_gop_frag_copy(vif, skb, npo,
597 virt_to_page(data), len, offset, &head);
598 data += len;
599 }
600
601 for (i = 0; i < nr_frags; i++) {
602 netbk_gop_frag_copy(vif, skb, npo,
603 skb_frag_page(&skb_shinfo(skb)->frags[i]),
604 skb_frag_size(&skb_shinfo(skb)->frags[i]),
605 skb_shinfo(skb)->frags[i].page_offset,
606 &head);
607 }
608
609 return npo->meta_prod - old_meta_prod;
610 }
611
612 /*
613 * This is a twin to netbk_gop_skb. Assume that netbk_gop_skb was
614 * used to set up the operations on the top of
615 * netrx_pending_operations, which have since been done. Check that
616 * they didn't give any errors and advance over them.
617 */
618 static int netbk_check_gop(struct xenvif *vif, int nr_meta_slots,
619 struct netrx_pending_operations *npo)
620 {
621 struct gnttab_copy *copy_op;
622 int status = XEN_NETIF_RSP_OKAY;
623 int i;
624
625 for (i = 0; i < nr_meta_slots; i++) {
626 copy_op = npo->copy + npo->copy_cons++;
627 if (copy_op->status != GNTST_okay) {
628 netdev_dbg(vif->dev,
629 "Bad status %d from copy to DOM%d.\n",
630 copy_op->status, vif->domid);
631 status = XEN_NETIF_RSP_ERROR;
632 }
633 }
634
635 return status;
636 }
637
638 static void netbk_add_frag_responses(struct xenvif *vif, int status,
639 struct netbk_rx_meta *meta,
640 int nr_meta_slots)
641 {
642 int i;
643 unsigned long offset;
644
645 /* No fragments used */
646 if (nr_meta_slots <= 1)
647 return;
648
649 nr_meta_slots--;
650
651 for (i = 0; i < nr_meta_slots; i++) {
652 int flags;
653 if (i == nr_meta_slots - 1)
654 flags = 0;
655 else
656 flags = XEN_NETRXF_more_data;
657
658 offset = 0;
659 make_rx_response(vif, meta[i].id, status, offset,
660 meta[i].size, flags);
661 }
662 }
663
664 struct skb_cb_overlay {
665 int meta_slots_used;
666 };
667
668 static void xen_netbk_rx_action(struct xen_netbk *netbk)
669 {
670 struct xenvif *vif = NULL, *tmp;
671 s8 status;
672 u16 flags;
673 struct xen_netif_rx_response *resp;
674 struct sk_buff_head rxq;
675 struct sk_buff *skb;
676 LIST_HEAD(notify);
677 int ret;
678 int nr_frags;
679 int count;
680 unsigned long offset;
681 struct skb_cb_overlay *sco;
682
683 struct netrx_pending_operations npo = {
684 .copy = netbk->grant_copy_op,
685 .meta = netbk->meta,
686 };
687
688 skb_queue_head_init(&rxq);
689
690 count = 0;
691
692 while ((skb = skb_dequeue(&netbk->rx_queue)) != NULL) {
693 vif = netdev_priv(skb->dev);
694 nr_frags = skb_shinfo(skb)->nr_frags;
695
696 sco = (struct skb_cb_overlay *)skb->cb;
697 sco->meta_slots_used = netbk_gop_skb(skb, &npo);
698
699 count += nr_frags + 1;
700
701 __skb_queue_tail(&rxq, skb);
702
703 /* Filled the batch queue? */
704 /* XXX FIXME: RX path dependent on MAX_SKB_FRAGS */
705 if (count + MAX_SKB_FRAGS >= XEN_NETIF_RX_RING_SIZE)
706 break;
707 }
708
709 BUG_ON(npo.meta_prod > ARRAY_SIZE(netbk->meta));
710
711 if (!npo.copy_prod)
712 return;
713
714 BUG_ON(npo.copy_prod > ARRAY_SIZE(netbk->grant_copy_op));
715 gnttab_batch_copy(netbk->grant_copy_op, npo.copy_prod);
716
717 while ((skb = __skb_dequeue(&rxq)) != NULL) {
718 sco = (struct skb_cb_overlay *)skb->cb;
719
720 vif = netdev_priv(skb->dev);
721
722 if (netbk->meta[npo.meta_cons].gso_size && vif->gso_prefix) {
723 resp = RING_GET_RESPONSE(&vif->rx,
724 vif->rx.rsp_prod_pvt++);
725
726 resp->flags = XEN_NETRXF_gso_prefix | XEN_NETRXF_more_data;
727
728 resp->offset = netbk->meta[npo.meta_cons].gso_size;
729 resp->id = netbk->meta[npo.meta_cons].id;
730 resp->status = sco->meta_slots_used;
731
732 npo.meta_cons++;
733 sco->meta_slots_used--;
734 }
735
736
737 vif->dev->stats.tx_bytes += skb->len;
738 vif->dev->stats.tx_packets++;
739
740 status = netbk_check_gop(vif, sco->meta_slots_used, &npo);
741
742 if (sco->meta_slots_used == 1)
743 flags = 0;
744 else
745 flags = XEN_NETRXF_more_data;
746
747 if (skb->ip_summed == CHECKSUM_PARTIAL) /* local packet? */
748 flags |= XEN_NETRXF_csum_blank | XEN_NETRXF_data_validated;
749 else if (skb->ip_summed == CHECKSUM_UNNECESSARY)
750 /* remote but checksummed. */
751 flags |= XEN_NETRXF_data_validated;
752
753 offset = 0;
754 resp = make_rx_response(vif, netbk->meta[npo.meta_cons].id,
755 status, offset,
756 netbk->meta[npo.meta_cons].size,
757 flags);
758
759 if (netbk->meta[npo.meta_cons].gso_size && !vif->gso_prefix) {
760 struct xen_netif_extra_info *gso =
761 (struct xen_netif_extra_info *)
762 RING_GET_RESPONSE(&vif->rx,
763 vif->rx.rsp_prod_pvt++);
764
765 resp->flags |= XEN_NETRXF_extra_info;
766
767 gso->u.gso.size = netbk->meta[npo.meta_cons].gso_size;
768 gso->u.gso.type = XEN_NETIF_GSO_TYPE_TCPV4;
769 gso->u.gso.pad = 0;
770 gso->u.gso.features = 0;
771
772 gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
773 gso->flags = 0;
774 }
775
776 netbk_add_frag_responses(vif, status,
777 netbk->meta + npo.meta_cons + 1,
778 sco->meta_slots_used);
779
780 RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&vif->rx, ret);
781
782 xenvif_notify_tx_completion(vif);
783
784 if (ret && list_empty(&vif->notify_list))
785 list_add_tail(&vif->notify_list, &notify);
786 else
787 xenvif_put(vif);
788 npo.meta_cons += sco->meta_slots_used;
789 dev_kfree_skb(skb);
790 }
791
792 list_for_each_entry_safe(vif, tmp, &notify, notify_list) {
793 notify_remote_via_irq(vif->rx_irq);
794 list_del_init(&vif->notify_list);
795 xenvif_put(vif);
796 }
797
798 /* More work to do? */
799 if (!skb_queue_empty(&netbk->rx_queue) &&
800 !timer_pending(&netbk->net_timer))
801 xen_netbk_kick_thread(netbk);
802 }
803
804 void xen_netbk_queue_tx_skb(struct xenvif *vif, struct sk_buff *skb)
805 {
806 struct xen_netbk *netbk = vif->netbk;
807
808 skb_queue_tail(&netbk->rx_queue, skb);
809
810 xen_netbk_kick_thread(netbk);
811 }
812
813 static void xen_netbk_alarm(unsigned long data)
814 {
815 struct xen_netbk *netbk = (struct xen_netbk *)data;
816 xen_netbk_kick_thread(netbk);
817 }
818
819 static int __on_net_schedule_list(struct xenvif *vif)
820 {
821 return !list_empty(&vif->schedule_list);
822 }
823
824 /* Must be called with net_schedule_list_lock held */
825 static void remove_from_net_schedule_list(struct xenvif *vif)
826 {
827 if (likely(__on_net_schedule_list(vif))) {
828 list_del_init(&vif->schedule_list);
829 xenvif_put(vif);
830 }
831 }
832
833 static struct xenvif *poll_net_schedule_list(struct xen_netbk *netbk)
834 {
835 struct xenvif *vif = NULL;
836
837 spin_lock_irq(&netbk->net_schedule_list_lock);
838 if (list_empty(&netbk->net_schedule_list))
839 goto out;
840
841 vif = list_first_entry(&netbk->net_schedule_list,
842 struct xenvif, schedule_list);
843 if (!vif)
844 goto out;
845
846 xenvif_get(vif);
847
848 remove_from_net_schedule_list(vif);
849 out:
850 spin_unlock_irq(&netbk->net_schedule_list_lock);
851 return vif;
852 }
853
854 void xen_netbk_schedule_xenvif(struct xenvif *vif)
855 {
856 unsigned long flags;
857 struct xen_netbk *netbk = vif->netbk;
858
859 if (__on_net_schedule_list(vif))
860 goto kick;
861
862 spin_lock_irqsave(&netbk->net_schedule_list_lock, flags);
863 if (!__on_net_schedule_list(vif) &&
864 likely(xenvif_schedulable(vif))) {
865 list_add_tail(&vif->schedule_list, &netbk->net_schedule_list);
866 xenvif_get(vif);
867 }
868 spin_unlock_irqrestore(&netbk->net_schedule_list_lock, flags);
869
870 kick:
871 smp_mb();
872 if ((nr_pending_reqs(netbk) < (MAX_PENDING_REQS/2)) &&
873 !list_empty(&netbk->net_schedule_list))
874 xen_netbk_kick_thread(netbk);
875 }
876
877 void xen_netbk_deschedule_xenvif(struct xenvif *vif)
878 {
879 struct xen_netbk *netbk = vif->netbk;
880 spin_lock_irq(&netbk->net_schedule_list_lock);
881 remove_from_net_schedule_list(vif);
882 spin_unlock_irq(&netbk->net_schedule_list_lock);
883 }
884
885 void xen_netbk_check_rx_xenvif(struct xenvif *vif)
886 {
887 int more_to_do;
888
889 RING_FINAL_CHECK_FOR_REQUESTS(&vif->tx, more_to_do);
890
891 if (more_to_do)
892 xen_netbk_schedule_xenvif(vif);
893 }
894
895 static void tx_add_credit(struct xenvif *vif)
896 {
897 unsigned long max_burst, max_credit;
898
899 /*
900 * Allow a burst big enough to transmit a jumbo packet of up to 128kB.
901 * Otherwise the interface can seize up due to insufficient credit.
902 */
903 max_burst = RING_GET_REQUEST(&vif->tx, vif->tx.req_cons)->size;
904 max_burst = min(max_burst, 131072UL);
905 max_burst = max(max_burst, vif->credit_bytes);
906
907 /* Take care that adding a new chunk of credit doesn't wrap to zero. */
908 max_credit = vif->remaining_credit + vif->credit_bytes;
909 if (max_credit < vif->remaining_credit)
910 max_credit = ULONG_MAX; /* wrapped: clamp to ULONG_MAX */
911
912 vif->remaining_credit = min(max_credit, max_burst);
913 }
914
915 static void tx_credit_callback(unsigned long data)
916 {
917 struct xenvif *vif = (struct xenvif *)data;
918 tx_add_credit(vif);
919 xen_netbk_check_rx_xenvif(vif);
920 }
921
922 static void netbk_tx_err(struct xenvif *vif,
923 struct xen_netif_tx_request *txp, RING_IDX end)
924 {
925 RING_IDX cons = vif->tx.req_cons;
926
927 do {
928 make_tx_response(vif, txp, XEN_NETIF_RSP_ERROR);
929 if (cons == end)
930 break;
931 txp = RING_GET_REQUEST(&vif->tx, cons++);
932 } while (1);
933 vif->tx.req_cons = cons;
934 xen_netbk_check_rx_xenvif(vif);
935 xenvif_put(vif);
936 }
937
938 static void netbk_fatal_tx_err(struct xenvif *vif)
939 {
940 netdev_err(vif->dev, "fatal error; disabling device\n");
941 xenvif_carrier_off(vif);
942 xenvif_put(vif);
943 }
944
945 static int netbk_count_requests(struct xenvif *vif,
946 struct xen_netif_tx_request *first,
947 struct xen_netif_tx_request *txp,
948 int work_to_do)
949 {
950 RING_IDX cons = vif->tx.req_cons;
951 int slots = 0;
952 int drop_err = 0;
953 int more_data;
954
955 if (!(first->flags & XEN_NETTXF_more_data))
956 return 0;
957
958 do {
959 struct xen_netif_tx_request dropped_tx = { 0 };
960
961 if (slots >= work_to_do) {
962 netdev_err(vif->dev,
963 "Asked for %d slots but exceeds this limit\n",
964 work_to_do);
965 netbk_fatal_tx_err(vif);
966 return -ENODATA;
967 }
968
969 /* This guest is really using too many slots and
970 * considered malicious.
971 */
972 if (unlikely(slots >= fatal_skb_slots)) {
973 netdev_err(vif->dev,
974 "Malicious frontend using %d slots, threshold %u\n",
975 slots, fatal_skb_slots);
976 netbk_fatal_tx_err(vif);
977 return -E2BIG;
978 }
979
980 /* Xen network protocol had implicit dependency on
981 * MAX_SKB_FRAGS. XEN_NETBK_LEGACY_SLOTS_MAX is set to
982 * the historical MAX_SKB_FRAGS value 18 to honor the
983 * same behavior as before. Any packet using more than
984 * 18 slots but less than fatal_skb_slots slots is
985 * dropped
986 */
987 if (!drop_err && slots >= XEN_NETBK_LEGACY_SLOTS_MAX) {
988 if (net_ratelimit())
989 netdev_dbg(vif->dev,
990 "Too many slots (%d) exceeding limit (%d), dropping packet\n",
991 slots, XEN_NETBK_LEGACY_SLOTS_MAX);
992 drop_err = -E2BIG;
993 }
994
995 if (drop_err)
996 txp = &dropped_tx;
997
998 memcpy(txp, RING_GET_REQUEST(&vif->tx, cons + slots),
999 sizeof(*txp));
1000
1001 /* If the guest submitted a frame >= 64 KiB then
1002 * first->size overflowed and following slots will
1003 * appear to be larger than the frame.
1004 *
1005 * This cannot be fatal error as there are buggy
1006 * frontends that do this.
1007 *
1008 * Consume all slots and drop the packet.
1009 */
1010 if (!drop_err && txp->size > first->size) {
1011 if (net_ratelimit())
1012 netdev_dbg(vif->dev,
1013 "Invalid tx request, slot size %u > remaining size %u\n",
1014 txp->size, first->size);
1015 drop_err = -EIO;
1016 }
1017
1018 first->size -= txp->size;
1019 slots++;
1020
1021 if (unlikely((txp->offset + txp->size) > PAGE_SIZE)) {
1022 netdev_err(vif->dev, "Cross page boundary, txp->offset: %x, size: %u\n",
1023 txp->offset, txp->size);
1024 netbk_fatal_tx_err(vif);
1025 return -EINVAL;
1026 }
1027
1028 more_data = txp->flags & XEN_NETTXF_more_data;
1029
1030 if (!drop_err)
1031 txp++;
1032
1033 } while (more_data);
1034
1035 if (drop_err) {
1036 netbk_tx_err(vif, first, cons + slots);
1037 return drop_err;
1038 }
1039
1040 return slots;
1041 }
1042
1043 static struct page *xen_netbk_alloc_page(struct xen_netbk *netbk,
1044 u16 pending_idx)
1045 {
1046 struct page *page;
1047 page = alloc_page(GFP_KERNEL|__GFP_COLD);
1048 if (!page)
1049 return NULL;
1050 set_page_ext(page, netbk, pending_idx);
1051 netbk->mmap_pages[pending_idx] = page;
1052 return page;
1053 }
1054
1055 static struct gnttab_copy *xen_netbk_get_requests(struct xen_netbk *netbk,
1056 struct xenvif *vif,
1057 struct sk_buff *skb,
1058 struct xen_netif_tx_request *txp,
1059 struct gnttab_copy *gop)
1060 {
1061 struct skb_shared_info *shinfo = skb_shinfo(skb);
1062 skb_frag_t *frags = shinfo->frags;
1063 u16 pending_idx = *((u16 *)skb->data);
1064 u16 head_idx = 0;
1065 int slot, start;
1066 struct page *page;
1067 pending_ring_idx_t index, start_idx = 0;
1068 uint16_t dst_offset;
1069 unsigned int nr_slots;
1070 struct pending_tx_info *first = NULL;
1071
1072 /* At this point shinfo->nr_frags is in fact the number of
1073 * slots, which can be as large as XEN_NETBK_LEGACY_SLOTS_MAX.
1074 */
1075 nr_slots = shinfo->nr_frags;
1076
1077 /* Skip first skb fragment if it is on same page as header fragment. */
1078 start = (frag_get_pending_idx(&shinfo->frags[0]) == pending_idx);
1079
1080 /* Coalesce tx requests, at this point the packet passed in
1081 * should be <= 64K. Any packets larger than 64K have been
1082 * handled in netbk_count_requests().
1083 */
1084 for (shinfo->nr_frags = slot = start; slot < nr_slots;
1085 shinfo->nr_frags++) {
1086 struct pending_tx_info *pending_tx_info =
1087 netbk->pending_tx_info;
1088
1089 page = alloc_page(GFP_KERNEL|__GFP_COLD);
1090 if (!page)
1091 goto err;
1092
1093 dst_offset = 0;
1094 first = NULL;
1095 while (dst_offset < PAGE_SIZE && slot < nr_slots) {
1096 gop->flags = GNTCOPY_source_gref;
1097
1098 gop->source.u.ref = txp->gref;
1099 gop->source.domid = vif->domid;
1100 gop->source.offset = txp->offset;
1101
1102 gop->dest.domid = DOMID_SELF;
1103
1104 gop->dest.offset = dst_offset;
1105 gop->dest.u.gmfn = virt_to_mfn(page_address(page));
1106
1107 if (dst_offset + txp->size > PAGE_SIZE) {
1108 /* This page can only merge a portion
1109 * of tx request. Do not increment any
1110 * pointer / counter here. The txp
1111 * will be dealt with in future
1112 * rounds, eventually hitting the
1113 * `else` branch.
1114 */
1115 gop->len = PAGE_SIZE - dst_offset;
1116 txp->offset += gop->len;
1117 txp->size -= gop->len;
1118 dst_offset += gop->len; /* quit loop */
1119 } else {
1120 /* This tx request can be merged in the page */
1121 gop->len = txp->size;
1122 dst_offset += gop->len;
1123
1124 index = pending_index(netbk->pending_cons++);
1125
1126 pending_idx = netbk->pending_ring[index];
1127
1128 memcpy(&pending_tx_info[pending_idx].req, txp,
1129 sizeof(*txp));
1130 xenvif_get(vif);
1131
1132 pending_tx_info[pending_idx].vif = vif;
1133
1134 /* Poison these fields, corresponding
1135 * fields for head tx req will be set
1136 * to correct values after the loop.
1137 */
1138 netbk->mmap_pages[pending_idx] = (void *)(~0UL);
1139 pending_tx_info[pending_idx].head =
1140 INVALID_PENDING_RING_IDX;
1141
1142 if (!first) {
1143 first = &pending_tx_info[pending_idx];
1144 start_idx = index;
1145 head_idx = pending_idx;
1146 }
1147
1148 txp++;
1149 slot++;
1150 }
1151
1152 gop++;
1153 }
1154
1155 first->req.offset = 0;
1156 first->req.size = dst_offset;
1157 first->head = start_idx;
1158 set_page_ext(page, netbk, head_idx);
1159 netbk->mmap_pages[head_idx] = page;
1160 frag_set_pending_idx(&frags[shinfo->nr_frags], head_idx);
1161 }
1162
1163 BUG_ON(shinfo->nr_frags > MAX_SKB_FRAGS);
1164
1165 return gop;
1166 err:
1167 /* Unwind, freeing all pages and sending error responses. */
1168 while (shinfo->nr_frags-- > start) {
1169 xen_netbk_idx_release(netbk,
1170 frag_get_pending_idx(&frags[shinfo->nr_frags]),
1171 XEN_NETIF_RSP_ERROR);
1172 }
1173 /* The head too, if necessary. */
1174 if (start)
1175 xen_netbk_idx_release(netbk, pending_idx, XEN_NETIF_RSP_ERROR);
1176
1177 return NULL;
1178 }
1179
1180 static int xen_netbk_tx_check_gop(struct xen_netbk *netbk,
1181 struct sk_buff *skb,
1182 struct gnttab_copy **gopp)
1183 {
1184 struct gnttab_copy *gop = *gopp;
1185 u16 pending_idx = *((u16 *)skb->data);
1186 struct skb_shared_info *shinfo = skb_shinfo(skb);
1187 struct pending_tx_info *tx_info;
1188 int nr_frags = shinfo->nr_frags;
1189 int i, err, start;
1190 u16 peek; /* peek into next tx request */
1191
1192 /* Check status of header. */
1193 err = gop->status;
1194 if (unlikely(err))
1195 xen_netbk_idx_release(netbk, pending_idx, XEN_NETIF_RSP_ERROR);
1196
1197 /* Skip first skb fragment if it is on same page as header fragment. */
1198 start = (frag_get_pending_idx(&shinfo->frags[0]) == pending_idx);
1199
1200 for (i = start; i < nr_frags; i++) {
1201 int j, newerr;
1202 pending_ring_idx_t head;
1203
1204 pending_idx = frag_get_pending_idx(&shinfo->frags[i]);
1205 tx_info = &netbk->pending_tx_info[pending_idx];
1206 head = tx_info->head;
1207
1208 /* Check error status: if okay then remember grant handle. */
1209 do {
1210 newerr = (++gop)->status;
1211 if (newerr)
1212 break;
1213 peek = netbk->pending_ring[pending_index(++head)];
1214 } while (!pending_tx_is_head(netbk, peek));
1215
1216 if (likely(!newerr)) {
1217 /* Had a previous error? Invalidate this fragment. */
1218 if (unlikely(err))
1219 xen_netbk_idx_release(netbk, pending_idx, XEN_NETIF_RSP_OKAY);
1220 continue;
1221 }
1222
1223 /* Error on this fragment: respond to client with an error. */
1224 xen_netbk_idx_release(netbk, pending_idx, XEN_NETIF_RSP_ERROR);
1225
1226 /* Not the first error? Preceding frags already invalidated. */
1227 if (err)
1228 continue;
1229
1230 /* First error: invalidate header and preceding fragments. */
1231 pending_idx = *((u16 *)skb->data);
1232 xen_netbk_idx_release(netbk, pending_idx, XEN_NETIF_RSP_OKAY);
1233 for (j = start; j < i; j++) {
1234 pending_idx = frag_get_pending_idx(&shinfo->frags[j]);
1235 xen_netbk_idx_release(netbk, pending_idx, XEN_NETIF_RSP_OKAY);
1236 }
1237
1238 /* Remember the error: invalidate all subsequent fragments. */
1239 err = newerr;
1240 }
1241
1242 *gopp = gop + 1;
1243 return err;
1244 }
1245
1246 static void xen_netbk_fill_frags(struct xen_netbk *netbk, struct sk_buff *skb)
1247 {
1248 struct skb_shared_info *shinfo = skb_shinfo(skb);
1249 int nr_frags = shinfo->nr_frags;
1250 int i;
1251
1252 for (i = 0; i < nr_frags; i++) {
1253 skb_frag_t *frag = shinfo->frags + i;
1254 struct xen_netif_tx_request *txp;
1255 struct page *page;
1256 u16 pending_idx;
1257
1258 pending_idx = frag_get_pending_idx(frag);
1259
1260 txp = &netbk->pending_tx_info[pending_idx].req;
1261 page = virt_to_page(idx_to_kaddr(netbk, pending_idx));
1262 __skb_fill_page_desc(skb, i, page, txp->offset, txp->size);
1263 skb->len += txp->size;
1264 skb->data_len += txp->size;
1265 skb->truesize += txp->size;
1266
1267 /* Take an extra reference to offset xen_netbk_idx_release */
1268 get_page(netbk->mmap_pages[pending_idx]);
1269 xen_netbk_idx_release(netbk, pending_idx, XEN_NETIF_RSP_OKAY);
1270 }
1271 }
1272
1273 static int xen_netbk_get_extras(struct xenvif *vif,
1274 struct xen_netif_extra_info *extras,
1275 int work_to_do)
1276 {
1277 struct xen_netif_extra_info extra;
1278 RING_IDX cons = vif->tx.req_cons;
1279
1280 do {
1281 if (unlikely(work_to_do-- <= 0)) {
1282 netdev_err(vif->dev, "Missing extra info\n");
1283 netbk_fatal_tx_err(vif);
1284 return -EBADR;
1285 }
1286
1287 memcpy(&extra, RING_GET_REQUEST(&vif->tx, cons),
1288 sizeof(extra));
1289 if (unlikely(!extra.type ||
1290 extra.type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
1291 vif->tx.req_cons = ++cons;
1292 netdev_err(vif->dev,
1293 "Invalid extra type: %d\n", extra.type);
1294 netbk_fatal_tx_err(vif);
1295 return -EINVAL;
1296 }
1297
1298 memcpy(&extras[extra.type - 1], &extra, sizeof(extra));
1299 vif->tx.req_cons = ++cons;
1300 } while (extra.flags & XEN_NETIF_EXTRA_FLAG_MORE);
1301
1302 return work_to_do;
1303 }
1304
1305 static int netbk_set_skb_gso(struct xenvif *vif,
1306 struct sk_buff *skb,
1307 struct xen_netif_extra_info *gso)
1308 {
1309 if (!gso->u.gso.size) {
1310 netdev_err(vif->dev, "GSO size must not be zero.\n");
1311 netbk_fatal_tx_err(vif);
1312 return -EINVAL;
1313 }
1314
1315 /* Currently only TCPv4 S.O. is supported. */
1316 if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4) {
1317 netdev_err(vif->dev, "Bad GSO type %d.\n", gso->u.gso.type);
1318 netbk_fatal_tx_err(vif);
1319 return -EINVAL;
1320 }
1321
1322 skb_shinfo(skb)->gso_size = gso->u.gso.size;
1323 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1324
1325 /* Header must be checked, and gso_segs computed. */
1326 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1327 skb_shinfo(skb)->gso_segs = 0;
1328
1329 return 0;
1330 }
1331
1332 static int checksum_setup(struct xenvif *vif, struct sk_buff *skb)
1333 {
1334 struct iphdr *iph;
1335 int err = -EPROTO;
1336 int recalculate_partial_csum = 0;
1337
1338 /*
1339 * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy
1340 * peers can fail to set NETRXF_csum_blank when sending a GSO
1341 * frame. In this case force the SKB to CHECKSUM_PARTIAL and
1342 * recalculate the partial checksum.
1343 */
1344 if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) {
1345 vif->rx_gso_checksum_fixup++;
1346 skb->ip_summed = CHECKSUM_PARTIAL;
1347 recalculate_partial_csum = 1;
1348 }
1349
1350 /* A non-CHECKSUM_PARTIAL SKB does not require setup. */
1351 if (skb->ip_summed != CHECKSUM_PARTIAL)
1352 return 0;
1353
1354 if (skb->protocol != htons(ETH_P_IP))
1355 goto out;
1356
1357 iph = (void *)skb->data;
1358 switch (iph->protocol) {
1359 case IPPROTO_TCP:
1360 if (!skb_partial_csum_set(skb, 4 * iph->ihl,
1361 offsetof(struct tcphdr, check)))
1362 goto out;
1363
1364 if (recalculate_partial_csum) {
1365 struct tcphdr *tcph = tcp_hdr(skb);
1366 tcph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
1367 skb->len - iph->ihl*4,
1368 IPPROTO_TCP, 0);
1369 }
1370 break;
1371 case IPPROTO_UDP:
1372 if (!skb_partial_csum_set(skb, 4 * iph->ihl,
1373 offsetof(struct udphdr, check)))
1374 goto out;
1375
1376 if (recalculate_partial_csum) {
1377 struct udphdr *udph = udp_hdr(skb);
1378 udph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
1379 skb->len - iph->ihl*4,
1380 IPPROTO_UDP, 0);
1381 }
1382 break;
1383 default:
1384 if (net_ratelimit())
1385 netdev_err(vif->dev,
1386 "Attempting to checksum a non-TCP/UDP packet, dropping a protocol %d packet\n",
1387 iph->protocol);
1388 goto out;
1389 }
1390
1391 err = 0;
1392
1393 out:
1394 return err;
1395 }
1396
1397 static bool tx_credit_exceeded(struct xenvif *vif, unsigned size)
1398 {
1399 unsigned long now = jiffies;
1400 unsigned long next_credit =
1401 vif->credit_timeout.expires +
1402 msecs_to_jiffies(vif->credit_usec / 1000);
1403
1404 /* Timer could already be pending in rare cases. */
1405 if (timer_pending(&vif->credit_timeout))
1406 return true;
1407
1408 /* Passed the point where we can replenish credit? */
1409 if (time_after_eq(now, next_credit)) {
1410 vif->credit_timeout.expires = now;
1411 tx_add_credit(vif);
1412 }
1413
1414 /* Still too big to send right now? Set a callback. */
1415 if (size > vif->remaining_credit) {
1416 vif->credit_timeout.data =
1417 (unsigned long)vif;
1418 vif->credit_timeout.function =
1419 tx_credit_callback;
1420 mod_timer(&vif->credit_timeout,
1421 next_credit);
1422
1423 return true;
1424 }
1425
1426 return false;
1427 }
1428
1429 static unsigned xen_netbk_tx_build_gops(struct xen_netbk *netbk)
1430 {
1431 struct gnttab_copy *gop = netbk->tx_copy_ops, *request_gop;
1432 struct sk_buff *skb;
1433 int ret;
1434
1435 while ((nr_pending_reqs(netbk) + XEN_NETBK_LEGACY_SLOTS_MAX
1436 < MAX_PENDING_REQS) &&
1437 !list_empty(&netbk->net_schedule_list)) {
1438 struct xenvif *vif;
1439 struct xen_netif_tx_request txreq;
1440 struct xen_netif_tx_request txfrags[XEN_NETBK_LEGACY_SLOTS_MAX];
1441 struct page *page;
1442 struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX-1];
1443 u16 pending_idx;
1444 RING_IDX idx;
1445 int work_to_do;
1446 unsigned int data_len;
1447 pending_ring_idx_t index;
1448
1449 /* Get a netif from the list with work to do. */
1450 vif = poll_net_schedule_list(netbk);
1451 /* This can sometimes happen because the test of
1452 * list_empty(net_schedule_list) at the top of the
1453 * loop is unlocked. Just go back and have another
1454 * look.
1455 */
1456 if (!vif)
1457 continue;
1458
1459 if (vif->tx.sring->req_prod - vif->tx.req_cons >
1460 XEN_NETIF_TX_RING_SIZE) {
1461 netdev_err(vif->dev,
1462 "Impossible number of requests. "
1463 "req_prod %d, req_cons %d, size %ld\n",
1464 vif->tx.sring->req_prod, vif->tx.req_cons,
1465 XEN_NETIF_TX_RING_SIZE);
1466 netbk_fatal_tx_err(vif);
1467 continue;
1468 }
1469
1470 RING_FINAL_CHECK_FOR_REQUESTS(&vif->tx, work_to_do);
1471 if (!work_to_do) {
1472 xenvif_put(vif);
1473 continue;
1474 }
1475
1476 idx = vif->tx.req_cons;
1477 rmb(); /* Ensure that we see the request before we copy it. */
1478 memcpy(&txreq, RING_GET_REQUEST(&vif->tx, idx), sizeof(txreq));
1479
1480 /* Credit-based scheduling. */
1481 if (txreq.size > vif->remaining_credit &&
1482 tx_credit_exceeded(vif, txreq.size)) {
1483 xenvif_put(vif);
1484 continue;
1485 }
1486
1487 vif->remaining_credit -= txreq.size;
1488
1489 work_to_do--;
1490 vif->tx.req_cons = ++idx;
1491
1492 memset(extras, 0, sizeof(extras));
1493 if (txreq.flags & XEN_NETTXF_extra_info) {
1494 work_to_do = xen_netbk_get_extras(vif, extras,
1495 work_to_do);
1496 idx = vif->tx.req_cons;
1497 if (unlikely(work_to_do < 0))
1498 continue;
1499 }
1500
1501 ret = netbk_count_requests(vif, &txreq, txfrags, work_to_do);
1502 if (unlikely(ret < 0))
1503 continue;
1504
1505 idx += ret;
1506
1507 if (unlikely(txreq.size < ETH_HLEN)) {
1508 netdev_dbg(vif->dev,
1509 "Bad packet size: %d\n", txreq.size);
1510 netbk_tx_err(vif, &txreq, idx);
1511 continue;
1512 }
1513
1514 /* No crossing a page as the payload mustn't fragment. */
1515 if (unlikely((txreq.offset + txreq.size) > PAGE_SIZE)) {
1516 netdev_err(vif->dev,
1517 "txreq.offset: %x, size: %u, end: %lu\n",
1518 txreq.offset, txreq.size,
1519 (txreq.offset&~PAGE_MASK) + txreq.size);
1520 netbk_fatal_tx_err(vif);
1521 continue;
1522 }
1523
1524 index = pending_index(netbk->pending_cons);
1525 pending_idx = netbk->pending_ring[index];
1526
1527 data_len = (txreq.size > PKT_PROT_LEN &&
1528 ret < XEN_NETBK_LEGACY_SLOTS_MAX) ?
1529 PKT_PROT_LEN : txreq.size;
1530
1531 skb = alloc_skb(data_len + NET_SKB_PAD + NET_IP_ALIGN,
1532 GFP_ATOMIC | __GFP_NOWARN);
1533 if (unlikely(skb == NULL)) {
1534 netdev_dbg(vif->dev,
1535 "Can't allocate a skb in start_xmit.\n");
1536 netbk_tx_err(vif, &txreq, idx);
1537 break;
1538 }
1539
1540 /* Packets passed to netif_rx() must have some headroom. */
1541 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
1542
1543 if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) {
1544 struct xen_netif_extra_info *gso;
1545 gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1];
1546
1547 if (netbk_set_skb_gso(vif, skb, gso)) {
1548 /* Failure in netbk_set_skb_gso is fatal. */
1549 kfree_skb(skb);
1550 continue;
1551 }
1552 }
1553
1554 /* XXX could copy straight to head */
1555 page = xen_netbk_alloc_page(netbk, pending_idx);
1556 if (!page) {
1557 kfree_skb(skb);
1558 netbk_tx_err(vif, &txreq, idx);
1559 continue;
1560 }
1561
1562 gop->source.u.ref = txreq.gref;
1563 gop->source.domid = vif->domid;
1564 gop->source.offset = txreq.offset;
1565
1566 gop->dest.u.gmfn = virt_to_mfn(page_address(page));
1567 gop->dest.domid = DOMID_SELF;
1568 gop->dest.offset = txreq.offset;
1569
1570 gop->len = txreq.size;
1571 gop->flags = GNTCOPY_source_gref;
1572
1573 gop++;
1574
1575 memcpy(&netbk->pending_tx_info[pending_idx].req,
1576 &txreq, sizeof(txreq));
1577 netbk->pending_tx_info[pending_idx].vif = vif;
1578 netbk->pending_tx_info[pending_idx].head = index;
1579 *((u16 *)skb->data) = pending_idx;
1580
1581 __skb_put(skb, data_len);
1582
1583 skb_shinfo(skb)->nr_frags = ret;
1584 if (data_len < txreq.size) {
1585 skb_shinfo(skb)->nr_frags++;
1586 frag_set_pending_idx(&skb_shinfo(skb)->frags[0],
1587 pending_idx);
1588 } else {
1589 frag_set_pending_idx(&skb_shinfo(skb)->frags[0],
1590 INVALID_PENDING_IDX);
1591 }
1592
1593 netbk->pending_cons++;
1594
1595 request_gop = xen_netbk_get_requests(netbk, vif,
1596 skb, txfrags, gop);
1597 if (request_gop == NULL) {
1598 kfree_skb(skb);
1599 netbk_tx_err(vif, &txreq, idx);
1600 continue;
1601 }
1602 gop = request_gop;
1603
1604 __skb_queue_tail(&netbk->tx_queue, skb);
1605
1606 vif->tx.req_cons = idx;
1607 xen_netbk_check_rx_xenvif(vif);
1608
1609 if ((gop-netbk->tx_copy_ops) >= ARRAY_SIZE(netbk->tx_copy_ops))
1610 break;
1611 }
1612
1613 return gop - netbk->tx_copy_ops;
1614 }
1615
1616 static void xen_netbk_tx_submit(struct xen_netbk *netbk)
1617 {
1618 struct gnttab_copy *gop = netbk->tx_copy_ops;
1619 struct sk_buff *skb;
1620
1621 while ((skb = __skb_dequeue(&netbk->tx_queue)) != NULL) {
1622 struct xen_netif_tx_request *txp;
1623 struct xenvif *vif;
1624 u16 pending_idx;
1625 unsigned data_len;
1626
1627 pending_idx = *((u16 *)skb->data);
1628 vif = netbk->pending_tx_info[pending_idx].vif;
1629 txp = &netbk->pending_tx_info[pending_idx].req;
1630
1631 /* Check the remap error code. */
1632 if (unlikely(xen_netbk_tx_check_gop(netbk, skb, &gop))) {
1633 netdev_dbg(vif->dev, "netback grant failed.\n");
1634 skb_shinfo(skb)->nr_frags = 0;
1635 kfree_skb(skb);
1636 continue;
1637 }
1638
1639 data_len = skb->len;
1640 memcpy(skb->data,
1641 (void *)(idx_to_kaddr(netbk, pending_idx)|txp->offset),
1642 data_len);
1643 if (data_len < txp->size) {
1644 /* Append the packet payload as a fragment. */
1645 txp->offset += data_len;
1646 txp->size -= data_len;
1647 } else {
1648 /* Schedule a response immediately. */
1649 xen_netbk_idx_release(netbk, pending_idx, XEN_NETIF_RSP_OKAY);
1650 }
1651
1652 if (txp->flags & XEN_NETTXF_csum_blank)
1653 skb->ip_summed = CHECKSUM_PARTIAL;
1654 else if (txp->flags & XEN_NETTXF_data_validated)
1655 skb->ip_summed = CHECKSUM_UNNECESSARY;
1656
1657 xen_netbk_fill_frags(netbk, skb);
1658
1659 /*
1660 * If the initial fragment was < PKT_PROT_LEN then
1661 * pull through some bytes from the other fragments to
1662 * increase the linear region to PKT_PROT_LEN bytes.
1663 */
1664 if (skb_headlen(skb) < PKT_PROT_LEN && skb_is_nonlinear(skb)) {
1665 int target = min_t(int, skb->len, PKT_PROT_LEN);
1666 __pskb_pull_tail(skb, target - skb_headlen(skb));
1667 }
1668
1669 skb->dev = vif->dev;
1670 skb->protocol = eth_type_trans(skb, skb->dev);
1671 skb_reset_network_header(skb);
1672
1673 if (checksum_setup(vif, skb)) {
1674 netdev_dbg(vif->dev,
1675 "Can't setup checksum in net_tx_action\n");
1676 kfree_skb(skb);
1677 continue;
1678 }
1679
1680 skb_probe_transport_header(skb, 0);
1681
1682 vif->dev->stats.rx_bytes += skb->len;
1683 vif->dev->stats.rx_packets++;
1684
1685 xenvif_receive_skb(vif, skb);
1686 }
1687 }
1688
1689 /* Called after netfront has transmitted */
1690 static void xen_netbk_tx_action(struct xen_netbk *netbk)
1691 {
1692 unsigned nr_gops;
1693
1694 nr_gops = xen_netbk_tx_build_gops(netbk);
1695
1696 if (nr_gops == 0)
1697 return;
1698
1699 gnttab_batch_copy(netbk->tx_copy_ops, nr_gops);
1700
1701 xen_netbk_tx_submit(netbk);
1702 }
1703
1704 static void xen_netbk_idx_release(struct xen_netbk *netbk, u16 pending_idx,
1705 u8 status)
1706 {
1707 struct xenvif *vif;
1708 struct pending_tx_info *pending_tx_info;
1709 pending_ring_idx_t head;
1710 u16 peek; /* peek into next tx request */
1711
1712 BUG_ON(netbk->mmap_pages[pending_idx] == (void *)(~0UL));
1713
1714 /* Already complete? */
1715 if (netbk->mmap_pages[pending_idx] == NULL)
1716 return;
1717
1718 pending_tx_info = &netbk->pending_tx_info[pending_idx];
1719
1720 vif = pending_tx_info->vif;
1721 head = pending_tx_info->head;
1722
1723 BUG_ON(!pending_tx_is_head(netbk, head));
1724 BUG_ON(netbk->pending_ring[pending_index(head)] != pending_idx);
1725
1726 do {
1727 pending_ring_idx_t index;
1728 pending_ring_idx_t idx = pending_index(head);
1729 u16 info_idx = netbk->pending_ring[idx];
1730
1731 pending_tx_info = &netbk->pending_tx_info[info_idx];
1732 make_tx_response(vif, &pending_tx_info->req, status);
1733
1734 /* Setting any number other than
1735 * INVALID_PENDING_RING_IDX indicates this slot is
1736 * starting a new packet / ending a previous packet.
1737 */
1738 pending_tx_info->head = 0;
1739
1740 index = pending_index(netbk->pending_prod++);
1741 netbk->pending_ring[index] = netbk->pending_ring[info_idx];
1742
1743 xenvif_put(vif);
1744
1745 peek = netbk->pending_ring[pending_index(++head)];
1746
1747 } while (!pending_tx_is_head(netbk, peek));
1748
1749 netbk->mmap_pages[pending_idx]->mapping = 0;
1750 put_page(netbk->mmap_pages[pending_idx]);
1751 netbk->mmap_pages[pending_idx] = NULL;
1752 }
1753
1754
1755 static void make_tx_response(struct xenvif *vif,
1756 struct xen_netif_tx_request *txp,
1757 s8 st)
1758 {
1759 RING_IDX i = vif->tx.rsp_prod_pvt;
1760 struct xen_netif_tx_response *resp;
1761 int notify;
1762
1763 resp = RING_GET_RESPONSE(&vif->tx, i);
1764 resp->id = txp->id;
1765 resp->status = st;
1766
1767 if (txp->flags & XEN_NETTXF_extra_info)
1768 RING_GET_RESPONSE(&vif->tx, ++i)->status = XEN_NETIF_RSP_NULL;
1769
1770 vif->tx.rsp_prod_pvt = ++i;
1771 RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&vif->tx, notify);
1772 if (notify)
1773 notify_remote_via_irq(vif->tx_irq);
1774 }
1775
1776 static struct xen_netif_rx_response *make_rx_response(struct xenvif *vif,
1777 u16 id,
1778 s8 st,
1779 u16 offset,
1780 u16 size,
1781 u16 flags)
1782 {
1783 RING_IDX i = vif->rx.rsp_prod_pvt;
1784 struct xen_netif_rx_response *resp;
1785
1786 resp = RING_GET_RESPONSE(&vif->rx, i);
1787 resp->offset = offset;
1788 resp->flags = flags;
1789 resp->id = id;
1790 resp->status = (s16)size;
1791 if (st < 0)
1792 resp->status = (s16)st;
1793
1794 vif->rx.rsp_prod_pvt = ++i;
1795
1796 return resp;
1797 }
1798
1799 static inline int rx_work_todo(struct xen_netbk *netbk)
1800 {
1801 return !skb_queue_empty(&netbk->rx_queue);
1802 }
1803
1804 static inline int tx_work_todo(struct xen_netbk *netbk)
1805 {
1806
1807 if ((nr_pending_reqs(netbk) + XEN_NETBK_LEGACY_SLOTS_MAX
1808 < MAX_PENDING_REQS) &&
1809 !list_empty(&netbk->net_schedule_list))
1810 return 1;
1811
1812 return 0;
1813 }
1814
1815 static int xen_netbk_kthread(void *data)
1816 {
1817 struct xen_netbk *netbk = data;
1818 while (!kthread_should_stop()) {
1819 wait_event_interruptible(netbk->wq,
1820 rx_work_todo(netbk) ||
1821 tx_work_todo(netbk) ||
1822 kthread_should_stop());
1823 cond_resched();
1824
1825 if (kthread_should_stop())
1826 break;
1827
1828 if (rx_work_todo(netbk))
1829 xen_netbk_rx_action(netbk);
1830
1831 if (tx_work_todo(netbk))
1832 xen_netbk_tx_action(netbk);
1833 }
1834
1835 return 0;
1836 }
1837
1838 void xen_netbk_unmap_frontend_rings(struct xenvif *vif)
1839 {
1840 if (vif->tx.sring)
1841 xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif),
1842 vif->tx.sring);
1843 if (vif->rx.sring)
1844 xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif),
1845 vif->rx.sring);
1846 }
1847
1848 int xen_netbk_map_frontend_rings(struct xenvif *vif,
1849 grant_ref_t tx_ring_ref,
1850 grant_ref_t rx_ring_ref)
1851 {
1852 void *addr;
1853 struct xen_netif_tx_sring *txs;
1854 struct xen_netif_rx_sring *rxs;
1855
1856 int err = -ENOMEM;
1857
1858 err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(vif),
1859 tx_ring_ref, &addr);
1860 if (err)
1861 goto err;
1862
1863 txs = (struct xen_netif_tx_sring *)addr;
1864 BACK_RING_INIT(&vif->tx, txs, PAGE_SIZE);
1865
1866 err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(vif),
1867 rx_ring_ref, &addr);
1868 if (err)
1869 goto err;
1870
1871 rxs = (struct xen_netif_rx_sring *)addr;
1872 BACK_RING_INIT(&vif->rx, rxs, PAGE_SIZE);
1873
1874 vif->rx_req_cons_peek = 0;
1875
1876 return 0;
1877
1878 err:
1879 xen_netbk_unmap_frontend_rings(vif);
1880 return err;
1881 }
1882
1883 static int __init netback_init(void)
1884 {
1885 int i;
1886 int rc = 0;
1887 int group;
1888
1889 if (!xen_domain())
1890 return -ENODEV;
1891
1892 if (fatal_skb_slots < XEN_NETBK_LEGACY_SLOTS_MAX) {
1893 pr_info("fatal_skb_slots too small (%d), bump it to XEN_NETBK_LEGACY_SLOTS_MAX (%d)\n",
1894 fatal_skb_slots, XEN_NETBK_LEGACY_SLOTS_MAX);
1895 fatal_skb_slots = XEN_NETBK_LEGACY_SLOTS_MAX;
1896 }
1897
1898 xen_netbk_group_nr = num_online_cpus();
1899 xen_netbk = vzalloc(sizeof(struct xen_netbk) * xen_netbk_group_nr);
1900 if (!xen_netbk)
1901 return -ENOMEM;
1902
1903 for (group = 0; group < xen_netbk_group_nr; group++) {
1904 struct xen_netbk *netbk = &xen_netbk[group];
1905 skb_queue_head_init(&netbk->rx_queue);
1906 skb_queue_head_init(&netbk->tx_queue);
1907
1908 init_timer(&netbk->net_timer);
1909 netbk->net_timer.data = (unsigned long)netbk;
1910 netbk->net_timer.function = xen_netbk_alarm;
1911
1912 netbk->pending_cons = 0;
1913 netbk->pending_prod = MAX_PENDING_REQS;
1914 for (i = 0; i < MAX_PENDING_REQS; i++)
1915 netbk->pending_ring[i] = i;
1916
1917 init_waitqueue_head(&netbk->wq);
1918 netbk->task = kthread_create(xen_netbk_kthread,
1919 (void *)netbk,
1920 "netback/%u", group);
1921
1922 if (IS_ERR(netbk->task)) {
1923 pr_alert("kthread_create() fails at netback\n");
1924 del_timer(&netbk->net_timer);
1925 rc = PTR_ERR(netbk->task);
1926 goto failed_init;
1927 }
1928
1929 kthread_bind(netbk->task, group);
1930
1931 INIT_LIST_HEAD(&netbk->net_schedule_list);
1932
1933 spin_lock_init(&netbk->net_schedule_list_lock);
1934
1935 atomic_set(&netbk->netfront_count, 0);
1936
1937 wake_up_process(netbk->task);
1938 }
1939
1940 rc = xenvif_xenbus_init();
1941 if (rc)
1942 goto failed_init;
1943
1944 return 0;
1945
1946 failed_init:
1947 while (--group >= 0) {
1948 struct xen_netbk *netbk = &xen_netbk[group];
1949 del_timer(&netbk->net_timer);
1950 kthread_stop(netbk->task);
1951 }
1952 vfree(xen_netbk);
1953 return rc;
1954
1955 }
1956
1957 module_init(netback_init);
1958
1959 static void __exit netback_fini(void)
1960 {
1961 int i, j;
1962
1963 xenvif_xenbus_fini();
1964
1965 for (i = 0; i < xen_netbk_group_nr; i++) {
1966 struct xen_netbk *netbk = &xen_netbk[i];
1967 del_timer_sync(&netbk->net_timer);
1968 kthread_stop(netbk->task);
1969 for (j = 0; j < MAX_PENDING_REQS; j++) {
1970 if (netbk->mmap_pages[j])
1971 __free_page(netbk->mmap_pages[j]);
1972 }
1973 }
1974
1975 vfree(xen_netbk);
1976 }
1977 module_exit(netback_fini);
1978
1979 MODULE_LICENSE("Dual BSD/GPL");
1980 MODULE_ALIAS("xen-backend:vif");
Linus' kernel tree