Commit | Line | Data |
---|---|---|
f942dc25 IC |
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 | ||
ca981633 | 43 | #include <xen/xen.h> |
f942dc25 IC |
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 | ||
2810e5b9 WL |
50 | /* |
51 | * This is the maximum slots a skb can have. If a guest sends a skb | |
52 | * which exceeds this limit it is considered malicious. | |
53 | */ | |
54 | #define MAX_SKB_SLOTS_DEFAULT 20 | |
55 | static unsigned int max_skb_slots = MAX_SKB_SLOTS_DEFAULT; | |
56 | module_param(max_skb_slots, uint, 0444); | |
57 | ||
58 | typedef unsigned int pending_ring_idx_t; | |
59 | #define INVALID_PENDING_RING_IDX (~0U) | |
60 | ||
f942dc25 | 61 | struct pending_tx_info { |
2810e5b9 | 62 | struct xen_netif_tx_request req; /* coalesced tx request */ |
f942dc25 | 63 | struct xenvif *vif; |
2810e5b9 WL |
64 | pending_ring_idx_t head; /* head != INVALID_PENDING_RING_IDX |
65 | * if it is head of one or more tx | |
66 | * reqs | |
67 | */ | |
f942dc25 | 68 | }; |
f942dc25 IC |
69 | |
70 | struct netbk_rx_meta { | |
71 | int id; | |
72 | int size; | |
73 | int gso_size; | |
74 | }; | |
75 | ||
76 | #define MAX_PENDING_REQS 256 | |
77 | ||
ea066ad1 IC |
78 | /* Discriminate from any valid pending_idx value. */ |
79 | #define INVALID_PENDING_IDX 0xFFFF | |
80 | ||
f942dc25 IC |
81 | #define MAX_BUFFER_OFFSET PAGE_SIZE |
82 | ||
83 | /* extra field used in struct page */ | |
84 | union page_ext { | |
85 | struct { | |
86 | #if BITS_PER_LONG < 64 | |
87 | #define IDX_WIDTH 8 | |
88 | #define GROUP_WIDTH (BITS_PER_LONG - IDX_WIDTH) | |
89 | unsigned int group:GROUP_WIDTH; | |
90 | unsigned int idx:IDX_WIDTH; | |
91 | #else | |
92 | unsigned int group, idx; | |
93 | #endif | |
94 | } e; | |
95 | void *mapping; | |
96 | }; | |
97 | ||
98 | struct xen_netbk { | |
99 | wait_queue_head_t wq; | |
100 | struct task_struct *task; | |
101 | ||
102 | struct sk_buff_head rx_queue; | |
103 | struct sk_buff_head tx_queue; | |
104 | ||
105 | struct timer_list net_timer; | |
106 | ||
107 | struct page *mmap_pages[MAX_PENDING_REQS]; | |
108 | ||
109 | pending_ring_idx_t pending_prod; | |
110 | pending_ring_idx_t pending_cons; | |
111 | struct list_head net_schedule_list; | |
112 | ||
113 | /* Protect the net_schedule_list in netif. */ | |
114 | spinlock_t net_schedule_list_lock; | |
115 | ||
116 | atomic_t netfront_count; | |
117 | ||
118 | struct pending_tx_info pending_tx_info[MAX_PENDING_REQS]; | |
2810e5b9 WL |
119 | /* Coalescing tx requests before copying makes number of grant |
120 | * copy ops greater or equal to number of slots required. In | |
121 | * worst case a tx request consumes 2 gnttab_copy. | |
122 | */ | |
123 | struct gnttab_copy tx_copy_ops[2*MAX_PENDING_REQS]; | |
f942dc25 IC |
124 | |
125 | u16 pending_ring[MAX_PENDING_REQS]; | |
126 | ||
127 | /* | |
128 | * Given MAX_BUFFER_OFFSET of 4096 the worst case is that each | |
129 | * head/fragment page uses 2 copy operations because it | |
130 | * straddles two buffers in the frontend. | |
131 | */ | |
132 | struct gnttab_copy grant_copy_op[2*XEN_NETIF_RX_RING_SIZE]; | |
133 | struct netbk_rx_meta meta[2*XEN_NETIF_RX_RING_SIZE]; | |
134 | }; | |
135 | ||
136 | static struct xen_netbk *xen_netbk; | |
137 | static int xen_netbk_group_nr; | |
138 | ||
2810e5b9 WL |
139 | /* |
140 | * If head != INVALID_PENDING_RING_IDX, it means this tx request is head of | |
141 | * one or more merged tx requests, otherwise it is the continuation of | |
142 | * previous tx request. | |
143 | */ | |
144 | static inline int pending_tx_is_head(struct xen_netbk *netbk, RING_IDX idx) | |
145 | { | |
146 | return netbk->pending_tx_info[idx].head != INVALID_PENDING_RING_IDX; | |
147 | } | |
148 | ||
f942dc25 IC |
149 | void xen_netbk_add_xenvif(struct xenvif *vif) |
150 | { | |
151 | int i; | |
152 | int min_netfront_count; | |
153 | int min_group = 0; | |
154 | struct xen_netbk *netbk; | |
155 | ||
156 | min_netfront_count = atomic_read(&xen_netbk[0].netfront_count); | |
157 | for (i = 0; i < xen_netbk_group_nr; i++) { | |
158 | int netfront_count = atomic_read(&xen_netbk[i].netfront_count); | |
159 | if (netfront_count < min_netfront_count) { | |
160 | min_group = i; | |
161 | min_netfront_count = netfront_count; | |
162 | } | |
163 | } | |
164 | ||
165 | netbk = &xen_netbk[min_group]; | |
166 | ||
167 | vif->netbk = netbk; | |
168 | atomic_inc(&netbk->netfront_count); | |
169 | } | |
170 | ||
171 | void xen_netbk_remove_xenvif(struct xenvif *vif) | |
172 | { | |
173 | struct xen_netbk *netbk = vif->netbk; | |
174 | vif->netbk = NULL; | |
175 | atomic_dec(&netbk->netfront_count); | |
176 | } | |
177 | ||
7d5145d8 MD |
178 | static void xen_netbk_idx_release(struct xen_netbk *netbk, u16 pending_idx, |
179 | u8 status); | |
f942dc25 IC |
180 | static void make_tx_response(struct xenvif *vif, |
181 | struct xen_netif_tx_request *txp, | |
182 | s8 st); | |
183 | static struct xen_netif_rx_response *make_rx_response(struct xenvif *vif, | |
184 | u16 id, | |
185 | s8 st, | |
186 | u16 offset, | |
187 | u16 size, | |
188 | u16 flags); | |
189 | ||
190 | static inline unsigned long idx_to_pfn(struct xen_netbk *netbk, | |
ea066ad1 | 191 | u16 idx) |
f942dc25 IC |
192 | { |
193 | return page_to_pfn(netbk->mmap_pages[idx]); | |
194 | } | |
195 | ||
196 | static inline unsigned long idx_to_kaddr(struct xen_netbk *netbk, | |
ea066ad1 | 197 | u16 idx) |
f942dc25 IC |
198 | { |
199 | return (unsigned long)pfn_to_kaddr(idx_to_pfn(netbk, idx)); | |
200 | } | |
201 | ||
202 | /* extra field used in struct page */ | |
203 | static inline void set_page_ext(struct page *pg, struct xen_netbk *netbk, | |
204 | unsigned int idx) | |
205 | { | |
206 | unsigned int group = netbk - xen_netbk; | |
207 | union page_ext ext = { .e = { .group = group + 1, .idx = idx } }; | |
208 | ||
209 | BUILD_BUG_ON(sizeof(ext) > sizeof(ext.mapping)); | |
210 | pg->mapping = ext.mapping; | |
211 | } | |
212 | ||
213 | static int get_page_ext(struct page *pg, | |
214 | unsigned int *pgroup, unsigned int *pidx) | |
215 | { | |
216 | union page_ext ext = { .mapping = pg->mapping }; | |
217 | struct xen_netbk *netbk; | |
218 | unsigned int group, idx; | |
219 | ||
220 | group = ext.e.group - 1; | |
221 | ||
222 | if (group < 0 || group >= xen_netbk_group_nr) | |
223 | return 0; | |
224 | ||
225 | netbk = &xen_netbk[group]; | |
226 | ||
227 | idx = ext.e.idx; | |
228 | ||
229 | if ((idx < 0) || (idx >= MAX_PENDING_REQS)) | |
230 | return 0; | |
231 | ||
232 | if (netbk->mmap_pages[idx] != pg) | |
233 | return 0; | |
234 | ||
235 | *pgroup = group; | |
236 | *pidx = idx; | |
237 | ||
238 | return 1; | |
239 | } | |
240 | ||
241 | /* | |
242 | * This is the amount of packet we copy rather than map, so that the | |
243 | * guest can't fiddle with the contents of the headers while we do | |
244 | * packet processing on them (netfilter, routing, etc). | |
245 | */ | |
246 | #define PKT_PROT_LEN (ETH_HLEN + \ | |
247 | VLAN_HLEN + \ | |
248 | sizeof(struct iphdr) + MAX_IPOPTLEN + \ | |
249 | sizeof(struct tcphdr) + MAX_TCP_OPTION_SPACE) | |
250 | ||
ea066ad1 IC |
251 | static u16 frag_get_pending_idx(skb_frag_t *frag) |
252 | { | |
253 | return (u16)frag->page_offset; | |
254 | } | |
255 | ||
256 | static void frag_set_pending_idx(skb_frag_t *frag, u16 pending_idx) | |
257 | { | |
258 | frag->page_offset = pending_idx; | |
259 | } | |
260 | ||
f942dc25 IC |
261 | static inline pending_ring_idx_t pending_index(unsigned i) |
262 | { | |
263 | return i & (MAX_PENDING_REQS-1); | |
264 | } | |
265 | ||
266 | static inline pending_ring_idx_t nr_pending_reqs(struct xen_netbk *netbk) | |
267 | { | |
268 | return MAX_PENDING_REQS - | |
269 | netbk->pending_prod + netbk->pending_cons; | |
270 | } | |
271 | ||
272 | static void xen_netbk_kick_thread(struct xen_netbk *netbk) | |
273 | { | |
274 | wake_up(&netbk->wq); | |
275 | } | |
276 | ||
277 | static int max_required_rx_slots(struct xenvif *vif) | |
278 | { | |
279 | int max = DIV_ROUND_UP(vif->dev->mtu, PAGE_SIZE); | |
280 | ||
2810e5b9 | 281 | /* XXX FIXME: RX path dependent on MAX_SKB_FRAGS */ |
f942dc25 IC |
282 | if (vif->can_sg || vif->gso || vif->gso_prefix) |
283 | max += MAX_SKB_FRAGS + 1; /* extra_info + frags */ | |
284 | ||
285 | return max; | |
286 | } | |
287 | ||
288 | int xen_netbk_rx_ring_full(struct xenvif *vif) | |
289 | { | |
290 | RING_IDX peek = vif->rx_req_cons_peek; | |
291 | RING_IDX needed = max_required_rx_slots(vif); | |
292 | ||
293 | return ((vif->rx.sring->req_prod - peek) < needed) || | |
294 | ((vif->rx.rsp_prod_pvt + XEN_NETIF_RX_RING_SIZE - peek) < needed); | |
295 | } | |
296 | ||
297 | int xen_netbk_must_stop_queue(struct xenvif *vif) | |
298 | { | |
299 | if (!xen_netbk_rx_ring_full(vif)) | |
300 | return 0; | |
301 | ||
302 | vif->rx.sring->req_event = vif->rx_req_cons_peek + | |
303 | max_required_rx_slots(vif); | |
304 | mb(); /* request notification /then/ check the queue */ | |
305 | ||
306 | return xen_netbk_rx_ring_full(vif); | |
307 | } | |
308 | ||
309 | /* | |
310 | * Returns true if we should start a new receive buffer instead of | |
311 | * adding 'size' bytes to a buffer which currently contains 'offset' | |
312 | * bytes. | |
313 | */ | |
314 | static bool start_new_rx_buffer(int offset, unsigned long size, int head) | |
315 | { | |
316 | /* simple case: we have completely filled the current buffer. */ | |
317 | if (offset == MAX_BUFFER_OFFSET) | |
318 | return true; | |
319 | ||
320 | /* | |
321 | * complex case: start a fresh buffer if the current frag | |
322 | * would overflow the current buffer but only if: | |
323 | * (i) this frag would fit completely in the next buffer | |
324 | * and (ii) there is already some data in the current buffer | |
325 | * and (iii) this is not the head buffer. | |
326 | * | |
327 | * Where: | |
328 | * - (i) stops us splitting a frag into two copies | |
329 | * unless the frag is too large for a single buffer. | |
330 | * - (ii) stops us from leaving a buffer pointlessly empty. | |
331 | * - (iii) stops us leaving the first buffer | |
332 | * empty. Strictly speaking this is already covered | |
333 | * by (ii) but is explicitly checked because | |
334 | * netfront relies on the first buffer being | |
335 | * non-empty and can crash otherwise. | |
336 | * | |
337 | * This means we will effectively linearise small | |
338 | * frags but do not needlessly split large buffers | |
339 | * into multiple copies tend to give large frags their | |
340 | * own buffers as before. | |
341 | */ | |
342 | if ((offset + size > MAX_BUFFER_OFFSET) && | |
343 | (size <= MAX_BUFFER_OFFSET) && offset && !head) | |
344 | return true; | |
345 | ||
346 | return false; | |
347 | } | |
348 | ||
349 | /* | |
350 | * Figure out how many ring slots we're going to need to send @skb to | |
351 | * the guest. This function is essentially a dry run of | |
352 | * netbk_gop_frag_copy. | |
353 | */ | |
354 | unsigned int xen_netbk_count_skb_slots(struct xenvif *vif, struct sk_buff *skb) | |
355 | { | |
356 | unsigned int count; | |
357 | int i, copy_off; | |
358 | ||
e26b203e | 359 | count = DIV_ROUND_UP(skb_headlen(skb), PAGE_SIZE); |
f942dc25 IC |
360 | |
361 | copy_off = skb_headlen(skb) % PAGE_SIZE; | |
362 | ||
363 | if (skb_shinfo(skb)->gso_size) | |
364 | count++; | |
365 | ||
366 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { | |
9e903e08 | 367 | unsigned long size = skb_frag_size(&skb_shinfo(skb)->frags[i]); |
6a8ed462 | 368 | unsigned long offset = skb_shinfo(skb)->frags[i].page_offset; |
f942dc25 | 369 | unsigned long bytes; |
6a8ed462 IC |
370 | |
371 | offset &= ~PAGE_MASK; | |
372 | ||
f942dc25 | 373 | while (size > 0) { |
6a8ed462 | 374 | BUG_ON(offset >= PAGE_SIZE); |
f942dc25 IC |
375 | BUG_ON(copy_off > MAX_BUFFER_OFFSET); |
376 | ||
6a8ed462 IC |
377 | bytes = PAGE_SIZE - offset; |
378 | ||
379 | if (bytes > size) | |
380 | bytes = size; | |
381 | ||
382 | if (start_new_rx_buffer(copy_off, bytes, 0)) { | |
f942dc25 IC |
383 | count++; |
384 | copy_off = 0; | |
385 | } | |
386 | ||
f942dc25 IC |
387 | if (copy_off + bytes > MAX_BUFFER_OFFSET) |
388 | bytes = MAX_BUFFER_OFFSET - copy_off; | |
389 | ||
390 | copy_off += bytes; | |
6a8ed462 IC |
391 | |
392 | offset += bytes; | |
f942dc25 | 393 | size -= bytes; |
6a8ed462 IC |
394 | |
395 | if (offset == PAGE_SIZE) | |
396 | offset = 0; | |
f942dc25 IC |
397 | } |
398 | } | |
399 | return count; | |
400 | } | |
401 | ||
402 | struct netrx_pending_operations { | |
403 | unsigned copy_prod, copy_cons; | |
404 | unsigned meta_prod, meta_cons; | |
405 | struct gnttab_copy *copy; | |
406 | struct netbk_rx_meta *meta; | |
407 | int copy_off; | |
408 | grant_ref_t copy_gref; | |
409 | }; | |
410 | ||
411 | static struct netbk_rx_meta *get_next_rx_buffer(struct xenvif *vif, | |
412 | struct netrx_pending_operations *npo) | |
413 | { | |
414 | struct netbk_rx_meta *meta; | |
415 | struct xen_netif_rx_request *req; | |
416 | ||
417 | req = RING_GET_REQUEST(&vif->rx, vif->rx.req_cons++); | |
418 | ||
419 | meta = npo->meta + npo->meta_prod++; | |
420 | meta->gso_size = 0; | |
421 | meta->size = 0; | |
422 | meta->id = req->id; | |
423 | ||
424 | npo->copy_off = 0; | |
425 | npo->copy_gref = req->gref; | |
426 | ||
427 | return meta; | |
428 | } | |
429 | ||
430 | /* | |
431 | * Set up the grant operations for this fragment. If it's a flipping | |
432 | * interface, we also set up the unmap request from here. | |
433 | */ | |
434 | static void netbk_gop_frag_copy(struct xenvif *vif, struct sk_buff *skb, | |
435 | struct netrx_pending_operations *npo, | |
436 | struct page *page, unsigned long size, | |
437 | unsigned long offset, int *head) | |
438 | { | |
439 | struct gnttab_copy *copy_gop; | |
440 | struct netbk_rx_meta *meta; | |
441 | /* | |
e34c0246 | 442 | * These variables are used iff get_page_ext returns true, |
f942dc25 IC |
443 | * in which case they are guaranteed to be initialized. |
444 | */ | |
445 | unsigned int uninitialized_var(group), uninitialized_var(idx); | |
446 | int foreign = get_page_ext(page, &group, &idx); | |
447 | unsigned long bytes; | |
448 | ||
449 | /* Data must not cross a page boundary. */ | |
6a8ed462 | 450 | BUG_ON(size + offset > PAGE_SIZE<<compound_order(page)); |
f942dc25 IC |
451 | |
452 | meta = npo->meta + npo->meta_prod - 1; | |
453 | ||
6a8ed462 IC |
454 | /* Skip unused frames from start of page */ |
455 | page += offset >> PAGE_SHIFT; | |
456 | offset &= ~PAGE_MASK; | |
457 | ||
f942dc25 | 458 | while (size > 0) { |
6a8ed462 | 459 | BUG_ON(offset >= PAGE_SIZE); |
f942dc25 IC |
460 | BUG_ON(npo->copy_off > MAX_BUFFER_OFFSET); |
461 | ||
6a8ed462 IC |
462 | bytes = PAGE_SIZE - offset; |
463 | ||
464 | if (bytes > size) | |
465 | bytes = size; | |
466 | ||
467 | if (start_new_rx_buffer(npo->copy_off, bytes, *head)) { | |
f942dc25 IC |
468 | /* |
469 | * Netfront requires there to be some data in the head | |
470 | * buffer. | |
471 | */ | |
472 | BUG_ON(*head); | |
473 | ||
474 | meta = get_next_rx_buffer(vif, npo); | |
475 | } | |
476 | ||
f942dc25 IC |
477 | if (npo->copy_off + bytes > MAX_BUFFER_OFFSET) |
478 | bytes = MAX_BUFFER_OFFSET - npo->copy_off; | |
479 | ||
480 | copy_gop = npo->copy + npo->copy_prod++; | |
481 | copy_gop->flags = GNTCOPY_dest_gref; | |
482 | if (foreign) { | |
483 | struct xen_netbk *netbk = &xen_netbk[group]; | |
484 | struct pending_tx_info *src_pend; | |
485 | ||
486 | src_pend = &netbk->pending_tx_info[idx]; | |
487 | ||
488 | copy_gop->source.domid = src_pend->vif->domid; | |
489 | copy_gop->source.u.ref = src_pend->req.gref; | |
490 | copy_gop->flags |= GNTCOPY_source_gref; | |
491 | } else { | |
492 | void *vaddr = page_address(page); | |
493 | copy_gop->source.domid = DOMID_SELF; | |
494 | copy_gop->source.u.gmfn = virt_to_mfn(vaddr); | |
495 | } | |
496 | copy_gop->source.offset = offset; | |
497 | copy_gop->dest.domid = vif->domid; | |
498 | ||
499 | copy_gop->dest.offset = npo->copy_off; | |
500 | copy_gop->dest.u.ref = npo->copy_gref; | |
501 | copy_gop->len = bytes; | |
502 | ||
503 | npo->copy_off += bytes; | |
504 | meta->size += bytes; | |
505 | ||
506 | offset += bytes; | |
507 | size -= bytes; | |
508 | ||
6a8ed462 IC |
509 | /* Next frame */ |
510 | if (offset == PAGE_SIZE && size) { | |
511 | BUG_ON(!PageCompound(page)); | |
512 | page++; | |
513 | offset = 0; | |
514 | } | |
515 | ||
f942dc25 IC |
516 | /* Leave a gap for the GSO descriptor. */ |
517 | if (*head && skb_shinfo(skb)->gso_size && !vif->gso_prefix) | |
518 | vif->rx.req_cons++; | |
519 | ||
520 | *head = 0; /* There must be something in this buffer now. */ | |
521 | ||
522 | } | |
523 | } | |
524 | ||
525 | /* | |
526 | * Prepare an SKB to be transmitted to the frontend. | |
527 | * | |
528 | * This function is responsible for allocating grant operations, meta | |
529 | * structures, etc. | |
530 | * | |
531 | * It returns the number of meta structures consumed. The number of | |
532 | * ring slots used is always equal to the number of meta slots used | |
533 | * plus the number of GSO descriptors used. Currently, we use either | |
534 | * zero GSO descriptors (for non-GSO packets) or one descriptor (for | |
535 | * frontend-side LRO). | |
536 | */ | |
537 | static int netbk_gop_skb(struct sk_buff *skb, | |
538 | struct netrx_pending_operations *npo) | |
539 | { | |
540 | struct xenvif *vif = netdev_priv(skb->dev); | |
541 | int nr_frags = skb_shinfo(skb)->nr_frags; | |
542 | int i; | |
543 | struct xen_netif_rx_request *req; | |
544 | struct netbk_rx_meta *meta; | |
545 | unsigned char *data; | |
546 | int head = 1; | |
547 | int old_meta_prod; | |
548 | ||
549 | old_meta_prod = npo->meta_prod; | |
550 | ||
551 | /* Set up a GSO prefix descriptor, if necessary */ | |
552 | if (skb_shinfo(skb)->gso_size && vif->gso_prefix) { | |
553 | req = RING_GET_REQUEST(&vif->rx, vif->rx.req_cons++); | |
554 | meta = npo->meta + npo->meta_prod++; | |
555 | meta->gso_size = skb_shinfo(skb)->gso_size; | |
556 | meta->size = 0; | |
557 | meta->id = req->id; | |
558 | } | |
559 | ||
560 | req = RING_GET_REQUEST(&vif->rx, vif->rx.req_cons++); | |
561 | meta = npo->meta + npo->meta_prod++; | |
562 | ||
563 | if (!vif->gso_prefix) | |
564 | meta->gso_size = skb_shinfo(skb)->gso_size; | |
565 | else | |
566 | meta->gso_size = 0; | |
567 | ||
568 | meta->size = 0; | |
569 | meta->id = req->id; | |
570 | npo->copy_off = 0; | |
571 | npo->copy_gref = req->gref; | |
572 | ||
573 | data = skb->data; | |
574 | while (data < skb_tail_pointer(skb)) { | |
575 | unsigned int offset = offset_in_page(data); | |
576 | unsigned int len = PAGE_SIZE - offset; | |
577 | ||
578 | if (data + len > skb_tail_pointer(skb)) | |
579 | len = skb_tail_pointer(skb) - data; | |
580 | ||
581 | netbk_gop_frag_copy(vif, skb, npo, | |
582 | virt_to_page(data), len, offset, &head); | |
583 | data += len; | |
584 | } | |
585 | ||
586 | for (i = 0; i < nr_frags; i++) { | |
587 | netbk_gop_frag_copy(vif, skb, npo, | |
ea066ad1 | 588 | skb_frag_page(&skb_shinfo(skb)->frags[i]), |
9e903e08 | 589 | skb_frag_size(&skb_shinfo(skb)->frags[i]), |
f942dc25 IC |
590 | skb_shinfo(skb)->frags[i].page_offset, |
591 | &head); | |
592 | } | |
593 | ||
594 | return npo->meta_prod - old_meta_prod; | |
595 | } | |
596 | ||
597 | /* | |
598 | * This is a twin to netbk_gop_skb. Assume that netbk_gop_skb was | |
599 | * used to set up the operations on the top of | |
600 | * netrx_pending_operations, which have since been done. Check that | |
601 | * they didn't give any errors and advance over them. | |
602 | */ | |
603 | static int netbk_check_gop(struct xenvif *vif, int nr_meta_slots, | |
604 | struct netrx_pending_operations *npo) | |
605 | { | |
606 | struct gnttab_copy *copy_op; | |
607 | int status = XEN_NETIF_RSP_OKAY; | |
608 | int i; | |
609 | ||
610 | for (i = 0; i < nr_meta_slots; i++) { | |
611 | copy_op = npo->copy + npo->copy_cons++; | |
612 | if (copy_op->status != GNTST_okay) { | |
613 | netdev_dbg(vif->dev, | |
614 | "Bad status %d from copy to DOM%d.\n", | |
615 | copy_op->status, vif->domid); | |
616 | status = XEN_NETIF_RSP_ERROR; | |
617 | } | |
618 | } | |
619 | ||
620 | return status; | |
621 | } | |
622 | ||
623 | static void netbk_add_frag_responses(struct xenvif *vif, int status, | |
624 | struct netbk_rx_meta *meta, | |
625 | int nr_meta_slots) | |
626 | { | |
627 | int i; | |
628 | unsigned long offset; | |
629 | ||
630 | /* No fragments used */ | |
631 | if (nr_meta_slots <= 1) | |
632 | return; | |
633 | ||
634 | nr_meta_slots--; | |
635 | ||
636 | for (i = 0; i < nr_meta_slots; i++) { | |
637 | int flags; | |
638 | if (i == nr_meta_slots - 1) | |
639 | flags = 0; | |
640 | else | |
641 | flags = XEN_NETRXF_more_data; | |
642 | ||
643 | offset = 0; | |
644 | make_rx_response(vif, meta[i].id, status, offset, | |
645 | meta[i].size, flags); | |
646 | } | |
647 | } | |
648 | ||
649 | struct skb_cb_overlay { | |
650 | int meta_slots_used; | |
651 | }; | |
652 | ||
653 | static void xen_netbk_rx_action(struct xen_netbk *netbk) | |
654 | { | |
655 | struct xenvif *vif = NULL, *tmp; | |
656 | s8 status; | |
657 | u16 irq, flags; | |
658 | struct xen_netif_rx_response *resp; | |
659 | struct sk_buff_head rxq; | |
660 | struct sk_buff *skb; | |
661 | LIST_HEAD(notify); | |
662 | int ret; | |
663 | int nr_frags; | |
664 | int count; | |
665 | unsigned long offset; | |
666 | struct skb_cb_overlay *sco; | |
667 | ||
668 | struct netrx_pending_operations npo = { | |
669 | .copy = netbk->grant_copy_op, | |
670 | .meta = netbk->meta, | |
671 | }; | |
672 | ||
673 | skb_queue_head_init(&rxq); | |
674 | ||
675 | count = 0; | |
676 | ||
677 | while ((skb = skb_dequeue(&netbk->rx_queue)) != NULL) { | |
678 | vif = netdev_priv(skb->dev); | |
679 | nr_frags = skb_shinfo(skb)->nr_frags; | |
680 | ||
681 | sco = (struct skb_cb_overlay *)skb->cb; | |
682 | sco->meta_slots_used = netbk_gop_skb(skb, &npo); | |
683 | ||
684 | count += nr_frags + 1; | |
685 | ||
686 | __skb_queue_tail(&rxq, skb); | |
687 | ||
688 | /* Filled the batch queue? */ | |
2810e5b9 | 689 | /* XXX FIXME: RX path dependent on MAX_SKB_FRAGS */ |
f942dc25 IC |
690 | if (count + MAX_SKB_FRAGS >= XEN_NETIF_RX_RING_SIZE) |
691 | break; | |
692 | } | |
693 | ||
694 | BUG_ON(npo.meta_prod > ARRAY_SIZE(netbk->meta)); | |
695 | ||
696 | if (!npo.copy_prod) | |
697 | return; | |
698 | ||
699 | BUG_ON(npo.copy_prod > ARRAY_SIZE(netbk->grant_copy_op)); | |
c571898f | 700 | gnttab_batch_copy(netbk->grant_copy_op, npo.copy_prod); |
f942dc25 IC |
701 | |
702 | while ((skb = __skb_dequeue(&rxq)) != NULL) { | |
703 | sco = (struct skb_cb_overlay *)skb->cb; | |
704 | ||
705 | vif = netdev_priv(skb->dev); | |
706 | ||
707 | if (netbk->meta[npo.meta_cons].gso_size && vif->gso_prefix) { | |
708 | resp = RING_GET_RESPONSE(&vif->rx, | |
709 | vif->rx.rsp_prod_pvt++); | |
710 | ||
711 | resp->flags = XEN_NETRXF_gso_prefix | XEN_NETRXF_more_data; | |
712 | ||
713 | resp->offset = netbk->meta[npo.meta_cons].gso_size; | |
714 | resp->id = netbk->meta[npo.meta_cons].id; | |
715 | resp->status = sco->meta_slots_used; | |
716 | ||
717 | npo.meta_cons++; | |
718 | sco->meta_slots_used--; | |
719 | } | |
720 | ||
721 | ||
722 | vif->dev->stats.tx_bytes += skb->len; | |
723 | vif->dev->stats.tx_packets++; | |
724 | ||
725 | status = netbk_check_gop(vif, sco->meta_slots_used, &npo); | |
726 | ||
727 | if (sco->meta_slots_used == 1) | |
728 | flags = 0; | |
729 | else | |
730 | flags = XEN_NETRXF_more_data; | |
731 | ||
732 | if (skb->ip_summed == CHECKSUM_PARTIAL) /* local packet? */ | |
733 | flags |= XEN_NETRXF_csum_blank | XEN_NETRXF_data_validated; | |
734 | else if (skb->ip_summed == CHECKSUM_UNNECESSARY) | |
735 | /* remote but checksummed. */ | |
736 | flags |= XEN_NETRXF_data_validated; | |
737 | ||
738 | offset = 0; | |
739 | resp = make_rx_response(vif, netbk->meta[npo.meta_cons].id, | |
740 | status, offset, | |
741 | netbk->meta[npo.meta_cons].size, | |
742 | flags); | |
743 | ||
744 | if (netbk->meta[npo.meta_cons].gso_size && !vif->gso_prefix) { | |
745 | struct xen_netif_extra_info *gso = | |
746 | (struct xen_netif_extra_info *) | |
747 | RING_GET_RESPONSE(&vif->rx, | |
748 | vif->rx.rsp_prod_pvt++); | |
749 | ||
750 | resp->flags |= XEN_NETRXF_extra_info; | |
751 | ||
752 | gso->u.gso.size = netbk->meta[npo.meta_cons].gso_size; | |
753 | gso->u.gso.type = XEN_NETIF_GSO_TYPE_TCPV4; | |
754 | gso->u.gso.pad = 0; | |
755 | gso->u.gso.features = 0; | |
756 | ||
757 | gso->type = XEN_NETIF_EXTRA_TYPE_GSO; | |
758 | gso->flags = 0; | |
759 | } | |
760 | ||
761 | netbk_add_frag_responses(vif, status, | |
762 | netbk->meta + npo.meta_cons + 1, | |
763 | sco->meta_slots_used); | |
764 | ||
765 | RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&vif->rx, ret); | |
766 | irq = vif->irq; | |
767 | if (ret && list_empty(&vif->notify_list)) | |
768 | list_add_tail(&vif->notify_list, ¬ify); | |
769 | ||
770 | xenvif_notify_tx_completion(vif); | |
771 | ||
772 | xenvif_put(vif); | |
773 | npo.meta_cons += sco->meta_slots_used; | |
774 | dev_kfree_skb(skb); | |
775 | } | |
776 | ||
777 | list_for_each_entry_safe(vif, tmp, ¬ify, notify_list) { | |
778 | notify_remote_via_irq(vif->irq); | |
779 | list_del_init(&vif->notify_list); | |
780 | } | |
781 | ||
782 | /* More work to do? */ | |
783 | if (!skb_queue_empty(&netbk->rx_queue) && | |
784 | !timer_pending(&netbk->net_timer)) | |
785 | xen_netbk_kick_thread(netbk); | |
786 | } | |
787 | ||
788 | void xen_netbk_queue_tx_skb(struct xenvif *vif, struct sk_buff *skb) | |
789 | { | |
790 | struct xen_netbk *netbk = vif->netbk; | |
791 | ||
792 | skb_queue_tail(&netbk->rx_queue, skb); | |
793 | ||
794 | xen_netbk_kick_thread(netbk); | |
795 | } | |
796 | ||
797 | static void xen_netbk_alarm(unsigned long data) | |
798 | { | |
799 | struct xen_netbk *netbk = (struct xen_netbk *)data; | |
800 | xen_netbk_kick_thread(netbk); | |
801 | } | |
802 | ||
803 | static int __on_net_schedule_list(struct xenvif *vif) | |
804 | { | |
805 | return !list_empty(&vif->schedule_list); | |
806 | } | |
807 | ||
808 | /* Must be called with net_schedule_list_lock held */ | |
809 | static void remove_from_net_schedule_list(struct xenvif *vif) | |
810 | { | |
811 | if (likely(__on_net_schedule_list(vif))) { | |
812 | list_del_init(&vif->schedule_list); | |
813 | xenvif_put(vif); | |
814 | } | |
815 | } | |
816 | ||
817 | static struct xenvif *poll_net_schedule_list(struct xen_netbk *netbk) | |
818 | { | |
819 | struct xenvif *vif = NULL; | |
820 | ||
821 | spin_lock_irq(&netbk->net_schedule_list_lock); | |
822 | if (list_empty(&netbk->net_schedule_list)) | |
823 | goto out; | |
824 | ||
825 | vif = list_first_entry(&netbk->net_schedule_list, | |
826 | struct xenvif, schedule_list); | |
827 | if (!vif) | |
828 | goto out; | |
829 | ||
830 | xenvif_get(vif); | |
831 | ||
832 | remove_from_net_schedule_list(vif); | |
833 | out: | |
834 | spin_unlock_irq(&netbk->net_schedule_list_lock); | |
835 | return vif; | |
836 | } | |
837 | ||
838 | void xen_netbk_schedule_xenvif(struct xenvif *vif) | |
839 | { | |
840 | unsigned long flags; | |
841 | struct xen_netbk *netbk = vif->netbk; | |
842 | ||
843 | if (__on_net_schedule_list(vif)) | |
844 | goto kick; | |
845 | ||
846 | spin_lock_irqsave(&netbk->net_schedule_list_lock, flags); | |
847 | if (!__on_net_schedule_list(vif) && | |
848 | likely(xenvif_schedulable(vif))) { | |
849 | list_add_tail(&vif->schedule_list, &netbk->net_schedule_list); | |
850 | xenvif_get(vif); | |
851 | } | |
852 | spin_unlock_irqrestore(&netbk->net_schedule_list_lock, flags); | |
853 | ||
854 | kick: | |
855 | smp_mb(); | |
856 | if ((nr_pending_reqs(netbk) < (MAX_PENDING_REQS/2)) && | |
857 | !list_empty(&netbk->net_schedule_list)) | |
858 | xen_netbk_kick_thread(netbk); | |
859 | } | |
860 | ||
861 | void xen_netbk_deschedule_xenvif(struct xenvif *vif) | |
862 | { | |
863 | struct xen_netbk *netbk = vif->netbk; | |
864 | spin_lock_irq(&netbk->net_schedule_list_lock); | |
865 | remove_from_net_schedule_list(vif); | |
866 | spin_unlock_irq(&netbk->net_schedule_list_lock); | |
867 | } | |
868 | ||
869 | void xen_netbk_check_rx_xenvif(struct xenvif *vif) | |
870 | { | |
871 | int more_to_do; | |
872 | ||
873 | RING_FINAL_CHECK_FOR_REQUESTS(&vif->tx, more_to_do); | |
874 | ||
875 | if (more_to_do) | |
876 | xen_netbk_schedule_xenvif(vif); | |
877 | } | |
878 | ||
879 | static void tx_add_credit(struct xenvif *vif) | |
880 | { | |
881 | unsigned long max_burst, max_credit; | |
882 | ||
883 | /* | |
884 | * Allow a burst big enough to transmit a jumbo packet of up to 128kB. | |
885 | * Otherwise the interface can seize up due to insufficient credit. | |
886 | */ | |
887 | max_burst = RING_GET_REQUEST(&vif->tx, vif->tx.req_cons)->size; | |
888 | max_burst = min(max_burst, 131072UL); | |
889 | max_burst = max(max_burst, vif->credit_bytes); | |
890 | ||
891 | /* Take care that adding a new chunk of credit doesn't wrap to zero. */ | |
892 | max_credit = vif->remaining_credit + vif->credit_bytes; | |
893 | if (max_credit < vif->remaining_credit) | |
894 | max_credit = ULONG_MAX; /* wrapped: clamp to ULONG_MAX */ | |
895 | ||
896 | vif->remaining_credit = min(max_credit, max_burst); | |
897 | } | |
898 | ||
899 | static void tx_credit_callback(unsigned long data) | |
900 | { | |
901 | struct xenvif *vif = (struct xenvif *)data; | |
902 | tx_add_credit(vif); | |
903 | xen_netbk_check_rx_xenvif(vif); | |
904 | } | |
905 | ||
906 | static void netbk_tx_err(struct xenvif *vif, | |
907 | struct xen_netif_tx_request *txp, RING_IDX end) | |
908 | { | |
909 | RING_IDX cons = vif->tx.req_cons; | |
910 | ||
911 | do { | |
912 | make_tx_response(vif, txp, XEN_NETIF_RSP_ERROR); | |
b9149729 | 913 | if (cons == end) |
f942dc25 IC |
914 | break; |
915 | txp = RING_GET_REQUEST(&vif->tx, cons++); | |
916 | } while (1); | |
917 | vif->tx.req_cons = cons; | |
918 | xen_netbk_check_rx_xenvif(vif); | |
919 | xenvif_put(vif); | |
920 | } | |
921 | ||
48856286 IC |
922 | static void netbk_fatal_tx_err(struct xenvif *vif) |
923 | { | |
924 | netdev_err(vif->dev, "fatal error; disabling device\n"); | |
925 | xenvif_carrier_off(vif); | |
629821d9 | 926 | xenvif_put(vif); |
48856286 IC |
927 | } |
928 | ||
f942dc25 IC |
929 | static int netbk_count_requests(struct xenvif *vif, |
930 | struct xen_netif_tx_request *first, | |
2810e5b9 | 931 | RING_IDX first_idx, |
f942dc25 IC |
932 | struct xen_netif_tx_request *txp, |
933 | int work_to_do) | |
934 | { | |
935 | RING_IDX cons = vif->tx.req_cons; | |
2810e5b9 WL |
936 | int slots = 0; |
937 | int drop_err = 0; | |
f942dc25 IC |
938 | |
939 | if (!(first->flags & XEN_NETTXF_more_data)) | |
940 | return 0; | |
941 | ||
942 | do { | |
2810e5b9 WL |
943 | if (slots >= work_to_do) { |
944 | netdev_err(vif->dev, | |
945 | "Asked for %d slots but exceeds this limit\n", | |
946 | work_to_do); | |
48856286 | 947 | netbk_fatal_tx_err(vif); |
35876b5f | 948 | return -ENODATA; |
f942dc25 IC |
949 | } |
950 | ||
2810e5b9 WL |
951 | /* This guest is really using too many slots and |
952 | * considered malicious. | |
953 | */ | |
954 | if (unlikely(slots >= max_skb_slots)) { | |
955 | netdev_err(vif->dev, | |
956 | "Malicious frontend using %d slots, threshold %u\n", | |
957 | slots, max_skb_slots); | |
48856286 | 958 | netbk_fatal_tx_err(vif); |
35876b5f | 959 | return -E2BIG; |
f942dc25 IC |
960 | } |
961 | ||
2810e5b9 WL |
962 | /* Xen network protocol had implicit dependency on |
963 | * MAX_SKB_FRAGS. XEN_NETIF_NR_SLOTS_MIN is set to the | |
964 | * historical MAX_SKB_FRAGS value 18 to honor the same | |
965 | * behavior as before. Any packet using more than 18 | |
966 | * slots but less than max_skb_slots slots is dropped | |
967 | */ | |
968 | if (!drop_err && slots >= XEN_NETIF_NR_SLOTS_MIN) { | |
969 | if (net_ratelimit()) | |
970 | netdev_dbg(vif->dev, | |
971 | "Too many slots (%d) exceeding limit (%d), dropping packet\n", | |
972 | slots, XEN_NETIF_NR_SLOTS_MIN); | |
973 | drop_err = -E2BIG; | |
974 | } | |
975 | ||
976 | memcpy(txp, RING_GET_REQUEST(&vif->tx, cons + slots), | |
f942dc25 | 977 | sizeof(*txp)); |
03393fd5 WL |
978 | |
979 | /* If the guest submitted a frame >= 64 KiB then | |
980 | * first->size overflowed and following slots will | |
981 | * appear to be larger than the frame. | |
982 | * | |
983 | * This cannot be fatal error as there are buggy | |
984 | * frontends that do this. | |
985 | * | |
986 | * Consume all slots and drop the packet. | |
987 | */ | |
988 | if (!drop_err && txp->size > first->size) { | |
989 | if (net_ratelimit()) | |
990 | netdev_dbg(vif->dev, | |
991 | "Invalid tx request, slot size %u > remaining size %u\n", | |
992 | txp->size, first->size); | |
993 | drop_err = -EIO; | |
f942dc25 IC |
994 | } |
995 | ||
996 | first->size -= txp->size; | |
2810e5b9 | 997 | slots++; |
f942dc25 IC |
998 | |
999 | if (unlikely((txp->offset + txp->size) > PAGE_SIZE)) { | |
2810e5b9 | 1000 | netdev_err(vif->dev, "Cross page boundary, txp->offset: %x, size: %u\n", |
f942dc25 | 1001 | txp->offset, txp->size); |
48856286 | 1002 | netbk_fatal_tx_err(vif); |
35876b5f | 1003 | return -EINVAL; |
f942dc25 IC |
1004 | } |
1005 | } while ((txp++)->flags & XEN_NETTXF_more_data); | |
2810e5b9 WL |
1006 | |
1007 | if (drop_err) { | |
1008 | netbk_tx_err(vif, first, first_idx + slots); | |
1009 | return drop_err; | |
1010 | } | |
1011 | ||
1012 | return slots; | |
f942dc25 IC |
1013 | } |
1014 | ||
1015 | static struct page *xen_netbk_alloc_page(struct xen_netbk *netbk, | |
ea066ad1 | 1016 | u16 pending_idx) |
f942dc25 IC |
1017 | { |
1018 | struct page *page; | |
1019 | page = alloc_page(GFP_KERNEL|__GFP_COLD); | |
1020 | if (!page) | |
1021 | return NULL; | |
1022 | set_page_ext(page, netbk, pending_idx); | |
1023 | netbk->mmap_pages[pending_idx] = page; | |
1024 | return page; | |
1025 | } | |
1026 | ||
1027 | static struct gnttab_copy *xen_netbk_get_requests(struct xen_netbk *netbk, | |
1028 | struct xenvif *vif, | |
1029 | struct sk_buff *skb, | |
1030 | struct xen_netif_tx_request *txp, | |
1031 | struct gnttab_copy *gop) | |
1032 | { | |
1033 | struct skb_shared_info *shinfo = skb_shinfo(skb); | |
1034 | skb_frag_t *frags = shinfo->frags; | |
ea066ad1 | 1035 | u16 pending_idx = *((u16 *)skb->data); |
2810e5b9 WL |
1036 | u16 head_idx = 0; |
1037 | int slot, start; | |
1038 | struct page *page; | |
1039 | pending_ring_idx_t index, start_idx = 0; | |
1040 | uint16_t dst_offset; | |
1041 | unsigned int nr_slots; | |
1042 | struct pending_tx_info *first = NULL; | |
1043 | ||
1044 | /* At this point shinfo->nr_frags is in fact the number of | |
1045 | * slots, which can be as large as XEN_NETIF_NR_SLOTS_MIN. | |
1046 | */ | |
1047 | nr_slots = shinfo->nr_frags; | |
f942dc25 IC |
1048 | |
1049 | /* Skip first skb fragment if it is on same page as header fragment. */ | |
ea066ad1 | 1050 | start = (frag_get_pending_idx(&shinfo->frags[0]) == pending_idx); |
f942dc25 | 1051 | |
2810e5b9 WL |
1052 | /* Coalesce tx requests, at this point the packet passed in |
1053 | * should be <= 64K. Any packets larger than 64K have been | |
1054 | * handled in netbk_count_requests(). | |
1055 | */ | |
1056 | for (shinfo->nr_frags = slot = start; slot < nr_slots; | |
1057 | shinfo->nr_frags++) { | |
f942dc25 IC |
1058 | struct pending_tx_info *pending_tx_info = |
1059 | netbk->pending_tx_info; | |
1060 | ||
2810e5b9 | 1061 | page = alloc_page(GFP_KERNEL|__GFP_COLD); |
f942dc25 | 1062 | if (!page) |
4cc7c1cb | 1063 | goto err; |
f942dc25 | 1064 | |
2810e5b9 WL |
1065 | dst_offset = 0; |
1066 | first = NULL; | |
1067 | while (dst_offset < PAGE_SIZE && slot < nr_slots) { | |
1068 | gop->flags = GNTCOPY_source_gref; | |
1069 | ||
1070 | gop->source.u.ref = txp->gref; | |
1071 | gop->source.domid = vif->domid; | |
1072 | gop->source.offset = txp->offset; | |
1073 | ||
1074 | gop->dest.domid = DOMID_SELF; | |
1075 | ||
1076 | gop->dest.offset = dst_offset; | |
1077 | gop->dest.u.gmfn = virt_to_mfn(page_address(page)); | |
1078 | ||
1079 | if (dst_offset + txp->size > PAGE_SIZE) { | |
1080 | /* This page can only merge a portion | |
1081 | * of tx request. Do not increment any | |
1082 | * pointer / counter here. The txp | |
1083 | * will be dealt with in future | |
1084 | * rounds, eventually hitting the | |
1085 | * `else` branch. | |
1086 | */ | |
1087 | gop->len = PAGE_SIZE - dst_offset; | |
1088 | txp->offset += gop->len; | |
1089 | txp->size -= gop->len; | |
1090 | dst_offset += gop->len; /* quit loop */ | |
1091 | } else { | |
1092 | /* This tx request can be merged in the page */ | |
1093 | gop->len = txp->size; | |
1094 | dst_offset += gop->len; | |
1095 | ||
1096 | index = pending_index(netbk->pending_cons++); | |
1097 | ||
1098 | pending_idx = netbk->pending_ring[index]; | |
1099 | ||
1100 | memcpy(&pending_tx_info[pending_idx].req, txp, | |
1101 | sizeof(*txp)); | |
1102 | xenvif_get(vif); | |
1103 | ||
1104 | pending_tx_info[pending_idx].vif = vif; | |
1105 | ||
1106 | /* Poison these fields, corresponding | |
1107 | * fields for head tx req will be set | |
1108 | * to correct values after the loop. | |
1109 | */ | |
1110 | netbk->mmap_pages[pending_idx] = (void *)(~0UL); | |
1111 | pending_tx_info[pending_idx].head = | |
1112 | INVALID_PENDING_RING_IDX; | |
1113 | ||
1114 | if (!first) { | |
1115 | first = &pending_tx_info[pending_idx]; | |
1116 | start_idx = index; | |
1117 | head_idx = pending_idx; | |
1118 | } | |
1119 | ||
1120 | txp++; | |
1121 | slot++; | |
1122 | } | |
f942dc25 | 1123 | |
2810e5b9 WL |
1124 | gop++; |
1125 | } | |
f942dc25 | 1126 | |
2810e5b9 WL |
1127 | first->req.offset = 0; |
1128 | first->req.size = dst_offset; | |
1129 | first->head = start_idx; | |
1130 | set_page_ext(page, netbk, head_idx); | |
1131 | netbk->mmap_pages[head_idx] = page; | |
1132 | frag_set_pending_idx(&frags[shinfo->nr_frags], head_idx); | |
f942dc25 IC |
1133 | } |
1134 | ||
2810e5b9 WL |
1135 | BUG_ON(shinfo->nr_frags > MAX_SKB_FRAGS); |
1136 | ||
f942dc25 | 1137 | return gop; |
4cc7c1cb IC |
1138 | err: |
1139 | /* Unwind, freeing all pages and sending error responses. */ | |
2810e5b9 WL |
1140 | while (shinfo->nr_frags-- > start) { |
1141 | xen_netbk_idx_release(netbk, | |
1142 | frag_get_pending_idx(&frags[shinfo->nr_frags]), | |
1143 | XEN_NETIF_RSP_ERROR); | |
4cc7c1cb IC |
1144 | } |
1145 | /* The head too, if necessary. */ | |
1146 | if (start) | |
1147 | xen_netbk_idx_release(netbk, pending_idx, XEN_NETIF_RSP_ERROR); | |
1148 | ||
1149 | return NULL; | |
f942dc25 IC |
1150 | } |
1151 | ||
1152 | static int xen_netbk_tx_check_gop(struct xen_netbk *netbk, | |
1153 | struct sk_buff *skb, | |
1154 | struct gnttab_copy **gopp) | |
1155 | { | |
1156 | struct gnttab_copy *gop = *gopp; | |
ea066ad1 | 1157 | u16 pending_idx = *((u16 *)skb->data); |
f942dc25 | 1158 | struct skb_shared_info *shinfo = skb_shinfo(skb); |
2810e5b9 | 1159 | struct pending_tx_info *tx_info; |
f942dc25 IC |
1160 | int nr_frags = shinfo->nr_frags; |
1161 | int i, err, start; | |
2810e5b9 | 1162 | u16 peek; /* peek into next tx request */ |
f942dc25 IC |
1163 | |
1164 | /* Check status of header. */ | |
1165 | err = gop->status; | |
7d5145d8 MD |
1166 | if (unlikely(err)) |
1167 | xen_netbk_idx_release(netbk, pending_idx, XEN_NETIF_RSP_ERROR); | |
f942dc25 IC |
1168 | |
1169 | /* Skip first skb fragment if it is on same page as header fragment. */ | |
ea066ad1 | 1170 | start = (frag_get_pending_idx(&shinfo->frags[0]) == pending_idx); |
f942dc25 IC |
1171 | |
1172 | for (i = start; i < nr_frags; i++) { | |
1173 | int j, newerr; | |
2810e5b9 | 1174 | pending_ring_idx_t head; |
f942dc25 | 1175 | |
ea066ad1 | 1176 | pending_idx = frag_get_pending_idx(&shinfo->frags[i]); |
2810e5b9 WL |
1177 | tx_info = &netbk->pending_tx_info[pending_idx]; |
1178 | head = tx_info->head; | |
f942dc25 IC |
1179 | |
1180 | /* Check error status: if okay then remember grant handle. */ | |
2810e5b9 WL |
1181 | do { |
1182 | newerr = (++gop)->status; | |
1183 | if (newerr) | |
1184 | break; | |
1185 | peek = netbk->pending_ring[pending_index(++head)]; | |
1186 | } while (!pending_tx_is_head(netbk, peek)); | |
1187 | ||
f942dc25 IC |
1188 | if (likely(!newerr)) { |
1189 | /* Had a previous error? Invalidate this fragment. */ | |
1190 | if (unlikely(err)) | |
7d5145d8 | 1191 | xen_netbk_idx_release(netbk, pending_idx, XEN_NETIF_RSP_OKAY); |
f942dc25 IC |
1192 | continue; |
1193 | } | |
1194 | ||
1195 | /* Error on this fragment: respond to client with an error. */ | |
7d5145d8 | 1196 | xen_netbk_idx_release(netbk, pending_idx, XEN_NETIF_RSP_ERROR); |
f942dc25 IC |
1197 | |
1198 | /* Not the first error? Preceding frags already invalidated. */ | |
1199 | if (err) | |
1200 | continue; | |
1201 | ||
1202 | /* First error: invalidate header and preceding fragments. */ | |
1203 | pending_idx = *((u16 *)skb->data); | |
7d5145d8 | 1204 | xen_netbk_idx_release(netbk, pending_idx, XEN_NETIF_RSP_OKAY); |
f942dc25 | 1205 | for (j = start; j < i; j++) { |
5ccb3ea7 | 1206 | pending_idx = frag_get_pending_idx(&shinfo->frags[j]); |
7d5145d8 | 1207 | xen_netbk_idx_release(netbk, pending_idx, XEN_NETIF_RSP_OKAY); |
f942dc25 IC |
1208 | } |
1209 | ||
1210 | /* Remember the error: invalidate all subsequent fragments. */ | |
1211 | err = newerr; | |
1212 | } | |
1213 | ||
1214 | *gopp = gop + 1; | |
1215 | return err; | |
1216 | } | |
1217 | ||
1218 | static void xen_netbk_fill_frags(struct xen_netbk *netbk, struct sk_buff *skb) | |
1219 | { | |
1220 | struct skb_shared_info *shinfo = skb_shinfo(skb); | |
1221 | int nr_frags = shinfo->nr_frags; | |
1222 | int i; | |
1223 | ||
1224 | for (i = 0; i < nr_frags; i++) { | |
1225 | skb_frag_t *frag = shinfo->frags + i; | |
1226 | struct xen_netif_tx_request *txp; | |
ea066ad1 IC |
1227 | struct page *page; |
1228 | u16 pending_idx; | |
f942dc25 | 1229 | |
ea066ad1 | 1230 | pending_idx = frag_get_pending_idx(frag); |
f942dc25 IC |
1231 | |
1232 | txp = &netbk->pending_tx_info[pending_idx].req; | |
ea066ad1 IC |
1233 | page = virt_to_page(idx_to_kaddr(netbk, pending_idx)); |
1234 | __skb_fill_page_desc(skb, i, page, txp->offset, txp->size); | |
f942dc25 IC |
1235 | skb->len += txp->size; |
1236 | skb->data_len += txp->size; | |
1237 | skb->truesize += txp->size; | |
1238 | ||
1239 | /* Take an extra reference to offset xen_netbk_idx_release */ | |
1240 | get_page(netbk->mmap_pages[pending_idx]); | |
7d5145d8 | 1241 | xen_netbk_idx_release(netbk, pending_idx, XEN_NETIF_RSP_OKAY); |
f942dc25 IC |
1242 | } |
1243 | } | |
1244 | ||
1245 | static int xen_netbk_get_extras(struct xenvif *vif, | |
1246 | struct xen_netif_extra_info *extras, | |
1247 | int work_to_do) | |
1248 | { | |
1249 | struct xen_netif_extra_info extra; | |
1250 | RING_IDX cons = vif->tx.req_cons; | |
1251 | ||
1252 | do { | |
1253 | if (unlikely(work_to_do-- <= 0)) { | |
48856286 IC |
1254 | netdev_err(vif->dev, "Missing extra info\n"); |
1255 | netbk_fatal_tx_err(vif); | |
f942dc25 IC |
1256 | return -EBADR; |
1257 | } | |
1258 | ||
1259 | memcpy(&extra, RING_GET_REQUEST(&vif->tx, cons), | |
1260 | sizeof(extra)); | |
1261 | if (unlikely(!extra.type || | |
1262 | extra.type >= XEN_NETIF_EXTRA_TYPE_MAX)) { | |
1263 | vif->tx.req_cons = ++cons; | |
48856286 | 1264 | netdev_err(vif->dev, |
f942dc25 | 1265 | "Invalid extra type: %d\n", extra.type); |
48856286 | 1266 | netbk_fatal_tx_err(vif); |
f942dc25 IC |
1267 | return -EINVAL; |
1268 | } | |
1269 | ||
1270 | memcpy(&extras[extra.type - 1], &extra, sizeof(extra)); | |
1271 | vif->tx.req_cons = ++cons; | |
1272 | } while (extra.flags & XEN_NETIF_EXTRA_FLAG_MORE); | |
1273 | ||
1274 | return work_to_do; | |
1275 | } | |
1276 | ||
1277 | static int netbk_set_skb_gso(struct xenvif *vif, | |
1278 | struct sk_buff *skb, | |
1279 | struct xen_netif_extra_info *gso) | |
1280 | { | |
1281 | if (!gso->u.gso.size) { | |
48856286 IC |
1282 | netdev_err(vif->dev, "GSO size must not be zero.\n"); |
1283 | netbk_fatal_tx_err(vif); | |
f942dc25 IC |
1284 | return -EINVAL; |
1285 | } | |
1286 | ||
1287 | /* Currently only TCPv4 S.O. is supported. */ | |
1288 | if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4) { | |
48856286 IC |
1289 | netdev_err(vif->dev, "Bad GSO type %d.\n", gso->u.gso.type); |
1290 | netbk_fatal_tx_err(vif); | |
f942dc25 IC |
1291 | return -EINVAL; |
1292 | } | |
1293 | ||
1294 | skb_shinfo(skb)->gso_size = gso->u.gso.size; | |
1295 | skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4; | |
1296 | ||
1297 | /* Header must be checked, and gso_segs computed. */ | |
1298 | skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; | |
1299 | skb_shinfo(skb)->gso_segs = 0; | |
1300 | ||
1301 | return 0; | |
1302 | } | |
1303 | ||
1304 | static int checksum_setup(struct xenvif *vif, struct sk_buff *skb) | |
1305 | { | |
1306 | struct iphdr *iph; | |
f942dc25 IC |
1307 | int err = -EPROTO; |
1308 | int recalculate_partial_csum = 0; | |
1309 | ||
1310 | /* | |
1311 | * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy | |
1312 | * peers can fail to set NETRXF_csum_blank when sending a GSO | |
1313 | * frame. In this case force the SKB to CHECKSUM_PARTIAL and | |
1314 | * recalculate the partial checksum. | |
1315 | */ | |
1316 | if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) { | |
1317 | vif->rx_gso_checksum_fixup++; | |
1318 | skb->ip_summed = CHECKSUM_PARTIAL; | |
1319 | recalculate_partial_csum = 1; | |
1320 | } | |
1321 | ||
1322 | /* A non-CHECKSUM_PARTIAL SKB does not require setup. */ | |
1323 | if (skb->ip_summed != CHECKSUM_PARTIAL) | |
1324 | return 0; | |
1325 | ||
1326 | if (skb->protocol != htons(ETH_P_IP)) | |
1327 | goto out; | |
1328 | ||
1329 | iph = (void *)skb->data; | |
f942dc25 IC |
1330 | switch (iph->protocol) { |
1331 | case IPPROTO_TCP: | |
bea89336 JW |
1332 | if (!skb_partial_csum_set(skb, 4 * iph->ihl, |
1333 | offsetof(struct tcphdr, check))) | |
1334 | goto out; | |
f942dc25 IC |
1335 | |
1336 | if (recalculate_partial_csum) { | |
bea89336 | 1337 | struct tcphdr *tcph = tcp_hdr(skb); |
f942dc25 IC |
1338 | tcph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, |
1339 | skb->len - iph->ihl*4, | |
1340 | IPPROTO_TCP, 0); | |
1341 | } | |
1342 | break; | |
1343 | case IPPROTO_UDP: | |
bea89336 JW |
1344 | if (!skb_partial_csum_set(skb, 4 * iph->ihl, |
1345 | offsetof(struct udphdr, check))) | |
1346 | goto out; | |
f942dc25 IC |
1347 | |
1348 | if (recalculate_partial_csum) { | |
bea89336 | 1349 | struct udphdr *udph = udp_hdr(skb); |
f942dc25 IC |
1350 | udph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, |
1351 | skb->len - iph->ihl*4, | |
1352 | IPPROTO_UDP, 0); | |
1353 | } | |
1354 | break; | |
1355 | default: | |
1356 | if (net_ratelimit()) | |
1357 | netdev_err(vif->dev, | |
1358 | "Attempting to checksum a non-TCP/UDP packet, dropping a protocol %d packet\n", | |
1359 | iph->protocol); | |
1360 | goto out; | |
1361 | } | |
1362 | ||
f942dc25 IC |
1363 | err = 0; |
1364 | ||
1365 | out: | |
1366 | return err; | |
1367 | } | |
1368 | ||
1369 | static bool tx_credit_exceeded(struct xenvif *vif, unsigned size) | |
1370 | { | |
1371 | unsigned long now = jiffies; | |
1372 | unsigned long next_credit = | |
1373 | vif->credit_timeout.expires + | |
1374 | msecs_to_jiffies(vif->credit_usec / 1000); | |
1375 | ||
1376 | /* Timer could already be pending in rare cases. */ | |
1377 | if (timer_pending(&vif->credit_timeout)) | |
1378 | return true; | |
1379 | ||
1380 | /* Passed the point where we can replenish credit? */ | |
1381 | if (time_after_eq(now, next_credit)) { | |
1382 | vif->credit_timeout.expires = now; | |
1383 | tx_add_credit(vif); | |
1384 | } | |
1385 | ||
1386 | /* Still too big to send right now? Set a callback. */ | |
1387 | if (size > vif->remaining_credit) { | |
1388 | vif->credit_timeout.data = | |
1389 | (unsigned long)vif; | |
1390 | vif->credit_timeout.function = | |
1391 | tx_credit_callback; | |
1392 | mod_timer(&vif->credit_timeout, | |
1393 | next_credit); | |
1394 | ||
1395 | return true; | |
1396 | } | |
1397 | ||
1398 | return false; | |
1399 | } | |
1400 | ||
1401 | static unsigned xen_netbk_tx_build_gops(struct xen_netbk *netbk) | |
1402 | { | |
1403 | struct gnttab_copy *gop = netbk->tx_copy_ops, *request_gop; | |
1404 | struct sk_buff *skb; | |
1405 | int ret; | |
1406 | ||
2810e5b9 WL |
1407 | while ((nr_pending_reqs(netbk) + XEN_NETIF_NR_SLOTS_MIN |
1408 | < MAX_PENDING_REQS) && | |
f942dc25 IC |
1409 | !list_empty(&netbk->net_schedule_list)) { |
1410 | struct xenvif *vif; | |
1411 | struct xen_netif_tx_request txreq; | |
2810e5b9 | 1412 | struct xen_netif_tx_request txfrags[max_skb_slots]; |
f942dc25 IC |
1413 | struct page *page; |
1414 | struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX-1]; | |
1415 | u16 pending_idx; | |
1416 | RING_IDX idx; | |
1417 | int work_to_do; | |
1418 | unsigned int data_len; | |
1419 | pending_ring_idx_t index; | |
1420 | ||
1421 | /* Get a netif from the list with work to do. */ | |
1422 | vif = poll_net_schedule_list(netbk); | |
48856286 IC |
1423 | /* This can sometimes happen because the test of |
1424 | * list_empty(net_schedule_list) at the top of the | |
1425 | * loop is unlocked. Just go back and have another | |
1426 | * look. | |
1427 | */ | |
f942dc25 IC |
1428 | if (!vif) |
1429 | continue; | |
1430 | ||
48856286 IC |
1431 | if (vif->tx.sring->req_prod - vif->tx.req_cons > |
1432 | XEN_NETIF_TX_RING_SIZE) { | |
1433 | netdev_err(vif->dev, | |
1434 | "Impossible number of requests. " | |
1435 | "req_prod %d, req_cons %d, size %ld\n", | |
1436 | vif->tx.sring->req_prod, vif->tx.req_cons, | |
1437 | XEN_NETIF_TX_RING_SIZE); | |
1438 | netbk_fatal_tx_err(vif); | |
1439 | continue; | |
1440 | } | |
1441 | ||
f942dc25 IC |
1442 | RING_FINAL_CHECK_FOR_REQUESTS(&vif->tx, work_to_do); |
1443 | if (!work_to_do) { | |
1444 | xenvif_put(vif); | |
1445 | continue; | |
1446 | } | |
1447 | ||
1448 | idx = vif->tx.req_cons; | |
1449 | rmb(); /* Ensure that we see the request before we copy it. */ | |
1450 | memcpy(&txreq, RING_GET_REQUEST(&vif->tx, idx), sizeof(txreq)); | |
1451 | ||
1452 | /* Credit-based scheduling. */ | |
1453 | if (txreq.size > vif->remaining_credit && | |
1454 | tx_credit_exceeded(vif, txreq.size)) { | |
1455 | xenvif_put(vif); | |
1456 | continue; | |
1457 | } | |
1458 | ||
1459 | vif->remaining_credit -= txreq.size; | |
1460 | ||
1461 | work_to_do--; | |
1462 | vif->tx.req_cons = ++idx; | |
1463 | ||
1464 | memset(extras, 0, sizeof(extras)); | |
1465 | if (txreq.flags & XEN_NETTXF_extra_info) { | |
1466 | work_to_do = xen_netbk_get_extras(vif, extras, | |
1467 | work_to_do); | |
1468 | idx = vif->tx.req_cons; | |
48856286 | 1469 | if (unlikely(work_to_do < 0)) |
f942dc25 | 1470 | continue; |
f942dc25 IC |
1471 | } |
1472 | ||
2810e5b9 WL |
1473 | ret = netbk_count_requests(vif, &txreq, idx, |
1474 | txfrags, work_to_do); | |
48856286 | 1475 | if (unlikely(ret < 0)) |
f942dc25 | 1476 | continue; |
48856286 | 1477 | |
f942dc25 IC |
1478 | idx += ret; |
1479 | ||
1480 | if (unlikely(txreq.size < ETH_HLEN)) { | |
1481 | netdev_dbg(vif->dev, | |
1482 | "Bad packet size: %d\n", txreq.size); | |
1483 | netbk_tx_err(vif, &txreq, idx); | |
1484 | continue; | |
1485 | } | |
1486 | ||
1487 | /* No crossing a page as the payload mustn't fragment. */ | |
1488 | if (unlikely((txreq.offset + txreq.size) > PAGE_SIZE)) { | |
48856286 | 1489 | netdev_err(vif->dev, |
f942dc25 IC |
1490 | "txreq.offset: %x, size: %u, end: %lu\n", |
1491 | txreq.offset, txreq.size, | |
1492 | (txreq.offset&~PAGE_MASK) + txreq.size); | |
48856286 | 1493 | netbk_fatal_tx_err(vif); |
f942dc25 IC |
1494 | continue; |
1495 | } | |
1496 | ||
1497 | index = pending_index(netbk->pending_cons); | |
1498 | pending_idx = netbk->pending_ring[index]; | |
1499 | ||
1500 | data_len = (txreq.size > PKT_PROT_LEN && | |
2810e5b9 | 1501 | ret < XEN_NETIF_NR_SLOTS_MIN) ? |
f942dc25 IC |
1502 | PKT_PROT_LEN : txreq.size; |
1503 | ||
1504 | skb = alloc_skb(data_len + NET_SKB_PAD + NET_IP_ALIGN, | |
1505 | GFP_ATOMIC | __GFP_NOWARN); | |
1506 | if (unlikely(skb == NULL)) { | |
1507 | netdev_dbg(vif->dev, | |
1508 | "Can't allocate a skb in start_xmit.\n"); | |
1509 | netbk_tx_err(vif, &txreq, idx); | |
1510 | break; | |
1511 | } | |
1512 | ||
1513 | /* Packets passed to netif_rx() must have some headroom. */ | |
1514 | skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN); | |
1515 | ||
1516 | if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) { | |
1517 | struct xen_netif_extra_info *gso; | |
1518 | gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1]; | |
1519 | ||
1520 | if (netbk_set_skb_gso(vif, skb, gso)) { | |
48856286 | 1521 | /* Failure in netbk_set_skb_gso is fatal. */ |
f942dc25 | 1522 | kfree_skb(skb); |
f942dc25 IC |
1523 | continue; |
1524 | } | |
1525 | } | |
1526 | ||
1527 | /* XXX could copy straight to head */ | |
27f85228 | 1528 | page = xen_netbk_alloc_page(netbk, pending_idx); |
f942dc25 IC |
1529 | if (!page) { |
1530 | kfree_skb(skb); | |
1531 | netbk_tx_err(vif, &txreq, idx); | |
1532 | continue; | |
1533 | } | |
1534 | ||
f942dc25 IC |
1535 | gop->source.u.ref = txreq.gref; |
1536 | gop->source.domid = vif->domid; | |
1537 | gop->source.offset = txreq.offset; | |
1538 | ||
1539 | gop->dest.u.gmfn = virt_to_mfn(page_address(page)); | |
1540 | gop->dest.domid = DOMID_SELF; | |
1541 | gop->dest.offset = txreq.offset; | |
1542 | ||
1543 | gop->len = txreq.size; | |
1544 | gop->flags = GNTCOPY_source_gref; | |
1545 | ||
1546 | gop++; | |
1547 | ||
1548 | memcpy(&netbk->pending_tx_info[pending_idx].req, | |
1549 | &txreq, sizeof(txreq)); | |
1550 | netbk->pending_tx_info[pending_idx].vif = vif; | |
2810e5b9 | 1551 | netbk->pending_tx_info[pending_idx].head = index; |
f942dc25 IC |
1552 | *((u16 *)skb->data) = pending_idx; |
1553 | ||
1554 | __skb_put(skb, data_len); | |
1555 | ||
1556 | skb_shinfo(skb)->nr_frags = ret; | |
1557 | if (data_len < txreq.size) { | |
1558 | skb_shinfo(skb)->nr_frags++; | |
ea066ad1 IC |
1559 | frag_set_pending_idx(&skb_shinfo(skb)->frags[0], |
1560 | pending_idx); | |
f942dc25 | 1561 | } else { |
ea066ad1 IC |
1562 | frag_set_pending_idx(&skb_shinfo(skb)->frags[0], |
1563 | INVALID_PENDING_IDX); | |
f942dc25 IC |
1564 | } |
1565 | ||
f942dc25 IC |
1566 | netbk->pending_cons++; |
1567 | ||
1568 | request_gop = xen_netbk_get_requests(netbk, vif, | |
1569 | skb, txfrags, gop); | |
1570 | if (request_gop == NULL) { | |
1571 | kfree_skb(skb); | |
1572 | netbk_tx_err(vif, &txreq, idx); | |
1573 | continue; | |
1574 | } | |
1575 | gop = request_gop; | |
1576 | ||
1e0b6eac AL |
1577 | __skb_queue_tail(&netbk->tx_queue, skb); |
1578 | ||
f942dc25 IC |
1579 | vif->tx.req_cons = idx; |
1580 | xen_netbk_check_rx_xenvif(vif); | |
1581 | ||
1582 | if ((gop-netbk->tx_copy_ops) >= ARRAY_SIZE(netbk->tx_copy_ops)) | |
1583 | break; | |
1584 | } | |
1585 | ||
1586 | return gop - netbk->tx_copy_ops; | |
1587 | } | |
1588 | ||
1589 | static void xen_netbk_tx_submit(struct xen_netbk *netbk) | |
1590 | { | |
1591 | struct gnttab_copy *gop = netbk->tx_copy_ops; | |
1592 | struct sk_buff *skb; | |
1593 | ||
1594 | while ((skb = __skb_dequeue(&netbk->tx_queue)) != NULL) { | |
1595 | struct xen_netif_tx_request *txp; | |
1596 | struct xenvif *vif; | |
1597 | u16 pending_idx; | |
1598 | unsigned data_len; | |
1599 | ||
1600 | pending_idx = *((u16 *)skb->data); | |
1601 | vif = netbk->pending_tx_info[pending_idx].vif; | |
1602 | txp = &netbk->pending_tx_info[pending_idx].req; | |
1603 | ||
1604 | /* Check the remap error code. */ | |
1605 | if (unlikely(xen_netbk_tx_check_gop(netbk, skb, &gop))) { | |
1606 | netdev_dbg(vif->dev, "netback grant failed.\n"); | |
1607 | skb_shinfo(skb)->nr_frags = 0; | |
1608 | kfree_skb(skb); | |
1609 | continue; | |
1610 | } | |
1611 | ||
1612 | data_len = skb->len; | |
1613 | memcpy(skb->data, | |
1614 | (void *)(idx_to_kaddr(netbk, pending_idx)|txp->offset), | |
1615 | data_len); | |
1616 | if (data_len < txp->size) { | |
1617 | /* Append the packet payload as a fragment. */ | |
1618 | txp->offset += data_len; | |
1619 | txp->size -= data_len; | |
1620 | } else { | |
1621 | /* Schedule a response immediately. */ | |
7d5145d8 | 1622 | xen_netbk_idx_release(netbk, pending_idx, XEN_NETIF_RSP_OKAY); |
f942dc25 IC |
1623 | } |
1624 | ||
1625 | if (txp->flags & XEN_NETTXF_csum_blank) | |
1626 | skb->ip_summed = CHECKSUM_PARTIAL; | |
1627 | else if (txp->flags & XEN_NETTXF_data_validated) | |
1628 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
1629 | ||
1630 | xen_netbk_fill_frags(netbk, skb); | |
1631 | ||
1632 | /* | |
1633 | * If the initial fragment was < PKT_PROT_LEN then | |
1634 | * pull through some bytes from the other fragments to | |
1635 | * increase the linear region to PKT_PROT_LEN bytes. | |
1636 | */ | |
1637 | if (skb_headlen(skb) < PKT_PROT_LEN && skb_is_nonlinear(skb)) { | |
1638 | int target = min_t(int, skb->len, PKT_PROT_LEN); | |
1639 | __pskb_pull_tail(skb, target - skb_headlen(skb)); | |
1640 | } | |
1641 | ||
1642 | skb->dev = vif->dev; | |
1643 | skb->protocol = eth_type_trans(skb, skb->dev); | |
f9ca8f74 | 1644 | skb_reset_network_header(skb); |
f942dc25 IC |
1645 | |
1646 | if (checksum_setup(vif, skb)) { | |
1647 | netdev_dbg(vif->dev, | |
1648 | "Can't setup checksum in net_tx_action\n"); | |
1649 | kfree_skb(skb); | |
1650 | continue; | |
1651 | } | |
1652 | ||
40893fd0 | 1653 | skb_probe_transport_header(skb, 0); |
f9ca8f74 | 1654 | |
f942dc25 IC |
1655 | vif->dev->stats.rx_bytes += skb->len; |
1656 | vif->dev->stats.rx_packets++; | |
1657 | ||
1658 | xenvif_receive_skb(vif, skb); | |
1659 | } | |
1660 | } | |
1661 | ||
1662 | /* Called after netfront has transmitted */ | |
1663 | static void xen_netbk_tx_action(struct xen_netbk *netbk) | |
1664 | { | |
1665 | unsigned nr_gops; | |
f942dc25 IC |
1666 | |
1667 | nr_gops = xen_netbk_tx_build_gops(netbk); | |
1668 | ||
1669 | if (nr_gops == 0) | |
1670 | return; | |
f942dc25 | 1671 | |
c571898f | 1672 | gnttab_batch_copy(netbk->tx_copy_ops, nr_gops); |
f942dc25 | 1673 | |
c571898f | 1674 | xen_netbk_tx_submit(netbk); |
f942dc25 IC |
1675 | } |
1676 | ||
7d5145d8 MD |
1677 | static void xen_netbk_idx_release(struct xen_netbk *netbk, u16 pending_idx, |
1678 | u8 status) | |
f942dc25 IC |
1679 | { |
1680 | struct xenvif *vif; | |
1681 | struct pending_tx_info *pending_tx_info; | |
2810e5b9 WL |
1682 | pending_ring_idx_t head; |
1683 | u16 peek; /* peek into next tx request */ | |
1684 | ||
1685 | BUG_ON(netbk->mmap_pages[pending_idx] == (void *)(~0UL)); | |
f942dc25 IC |
1686 | |
1687 | /* Already complete? */ | |
1688 | if (netbk->mmap_pages[pending_idx] == NULL) | |
1689 | return; | |
1690 | ||
1691 | pending_tx_info = &netbk->pending_tx_info[pending_idx]; | |
1692 | ||
1693 | vif = pending_tx_info->vif; | |
2810e5b9 | 1694 | head = pending_tx_info->head; |
f942dc25 | 1695 | |
2810e5b9 WL |
1696 | BUG_ON(!pending_tx_is_head(netbk, head)); |
1697 | BUG_ON(netbk->pending_ring[pending_index(head)] != pending_idx); | |
f942dc25 | 1698 | |
2810e5b9 WL |
1699 | do { |
1700 | pending_ring_idx_t index; | |
1701 | pending_ring_idx_t idx = pending_index(head); | |
1702 | u16 info_idx = netbk->pending_ring[idx]; | |
f942dc25 | 1703 | |
2810e5b9 WL |
1704 | pending_tx_info = &netbk->pending_tx_info[info_idx]; |
1705 | make_tx_response(vif, &pending_tx_info->req, status); | |
1706 | ||
1707 | /* Setting any number other than | |
1708 | * INVALID_PENDING_RING_IDX indicates this slot is | |
1709 | * starting a new packet / ending a previous packet. | |
1710 | */ | |
1711 | pending_tx_info->head = 0; | |
1712 | ||
1713 | index = pending_index(netbk->pending_prod++); | |
1714 | netbk->pending_ring[index] = netbk->pending_ring[info_idx]; | |
f942dc25 | 1715 | |
2810e5b9 WL |
1716 | xenvif_put(vif); |
1717 | ||
1718 | peek = netbk->pending_ring[pending_index(++head)]; | |
1719 | ||
1720 | } while (!pending_tx_is_head(netbk, peek)); | |
1721 | ||
1722 | netbk->mmap_pages[pending_idx]->mapping = 0; | |
f942dc25 IC |
1723 | put_page(netbk->mmap_pages[pending_idx]); |
1724 | netbk->mmap_pages[pending_idx] = NULL; | |
1725 | } | |
1726 | ||
2810e5b9 | 1727 | |
f942dc25 IC |
1728 | static void make_tx_response(struct xenvif *vif, |
1729 | struct xen_netif_tx_request *txp, | |
1730 | s8 st) | |
1731 | { | |
1732 | RING_IDX i = vif->tx.rsp_prod_pvt; | |
1733 | struct xen_netif_tx_response *resp; | |
1734 | int notify; | |
1735 | ||
1736 | resp = RING_GET_RESPONSE(&vif->tx, i); | |
1737 | resp->id = txp->id; | |
1738 | resp->status = st; | |
1739 | ||
1740 | if (txp->flags & XEN_NETTXF_extra_info) | |
1741 | RING_GET_RESPONSE(&vif->tx, ++i)->status = XEN_NETIF_RSP_NULL; | |
1742 | ||
1743 | vif->tx.rsp_prod_pvt = ++i; | |
1744 | RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&vif->tx, notify); | |
1745 | if (notify) | |
1746 | notify_remote_via_irq(vif->irq); | |
1747 | } | |
1748 | ||
1749 | static struct xen_netif_rx_response *make_rx_response(struct xenvif *vif, | |
1750 | u16 id, | |
1751 | s8 st, | |
1752 | u16 offset, | |
1753 | u16 size, | |
1754 | u16 flags) | |
1755 | { | |
1756 | RING_IDX i = vif->rx.rsp_prod_pvt; | |
1757 | struct xen_netif_rx_response *resp; | |
1758 | ||
1759 | resp = RING_GET_RESPONSE(&vif->rx, i); | |
1760 | resp->offset = offset; | |
1761 | resp->flags = flags; | |
1762 | resp->id = id; | |
1763 | resp->status = (s16)size; | |
1764 | if (st < 0) | |
1765 | resp->status = (s16)st; | |
1766 | ||
1767 | vif->rx.rsp_prod_pvt = ++i; | |
1768 | ||
1769 | return resp; | |
1770 | } | |
1771 | ||
1772 | static inline int rx_work_todo(struct xen_netbk *netbk) | |
1773 | { | |
1774 | return !skb_queue_empty(&netbk->rx_queue); | |
1775 | } | |
1776 | ||
1777 | static inline int tx_work_todo(struct xen_netbk *netbk) | |
1778 | { | |
1779 | ||
2810e5b9 WL |
1780 | if ((nr_pending_reqs(netbk) + XEN_NETIF_NR_SLOTS_MIN |
1781 | < MAX_PENDING_REQS) && | |
1782 | !list_empty(&netbk->net_schedule_list)) | |
f942dc25 IC |
1783 | return 1; |
1784 | ||
1785 | return 0; | |
1786 | } | |
1787 | ||
1788 | static int xen_netbk_kthread(void *data) | |
1789 | { | |
1790 | struct xen_netbk *netbk = data; | |
1791 | while (!kthread_should_stop()) { | |
1792 | wait_event_interruptible(netbk->wq, | |
1793 | rx_work_todo(netbk) || | |
1794 | tx_work_todo(netbk) || | |
1795 | kthread_should_stop()); | |
1796 | cond_resched(); | |
1797 | ||
1798 | if (kthread_should_stop()) | |
1799 | break; | |
1800 | ||
1801 | if (rx_work_todo(netbk)) | |
1802 | xen_netbk_rx_action(netbk); | |
1803 | ||
1804 | if (tx_work_todo(netbk)) | |
1805 | xen_netbk_tx_action(netbk); | |
1806 | } | |
1807 | ||
1808 | return 0; | |
1809 | } | |
1810 | ||
1811 | void xen_netbk_unmap_frontend_rings(struct xenvif *vif) | |
1812 | { | |
c9d63699 DV |
1813 | if (vif->tx.sring) |
1814 | xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif), | |
1815 | vif->tx.sring); | |
1816 | if (vif->rx.sring) | |
1817 | xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif), | |
1818 | vif->rx.sring); | |
f942dc25 IC |
1819 | } |
1820 | ||
1821 | int xen_netbk_map_frontend_rings(struct xenvif *vif, | |
1822 | grant_ref_t tx_ring_ref, | |
1823 | grant_ref_t rx_ring_ref) | |
1824 | { | |
c9d63699 | 1825 | void *addr; |
f942dc25 IC |
1826 | struct xen_netif_tx_sring *txs; |
1827 | struct xen_netif_rx_sring *rxs; | |
1828 | ||
1829 | int err = -ENOMEM; | |
1830 | ||
c9d63699 DV |
1831 | err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(vif), |
1832 | tx_ring_ref, &addr); | |
1833 | if (err) | |
f942dc25 IC |
1834 | goto err; |
1835 | ||
c9d63699 | 1836 | txs = (struct xen_netif_tx_sring *)addr; |
f942dc25 IC |
1837 | BACK_RING_INIT(&vif->tx, txs, PAGE_SIZE); |
1838 | ||
c9d63699 DV |
1839 | err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(vif), |
1840 | rx_ring_ref, &addr); | |
1841 | if (err) | |
f942dc25 | 1842 | goto err; |
f942dc25 | 1843 | |
c9d63699 | 1844 | rxs = (struct xen_netif_rx_sring *)addr; |
f942dc25 IC |
1845 | BACK_RING_INIT(&vif->rx, rxs, PAGE_SIZE); |
1846 | ||
c9d63699 DV |
1847 | vif->rx_req_cons_peek = 0; |
1848 | ||
f942dc25 IC |
1849 | return 0; |
1850 | ||
1851 | err: | |
1852 | xen_netbk_unmap_frontend_rings(vif); | |
1853 | return err; | |
1854 | } | |
1855 | ||
1856 | static int __init netback_init(void) | |
1857 | { | |
1858 | int i; | |
1859 | int rc = 0; | |
1860 | int group; | |
1861 | ||
2a14b244 | 1862 | if (!xen_domain()) |
f942dc25 IC |
1863 | return -ENODEV; |
1864 | ||
2810e5b9 WL |
1865 | if (max_skb_slots < XEN_NETIF_NR_SLOTS_MIN) { |
1866 | printk(KERN_INFO | |
1867 | "xen-netback: max_skb_slots too small (%d), bump it to XEN_NETIF_NR_SLOTS_MIN (%d)\n", | |
1868 | max_skb_slots, XEN_NETIF_NR_SLOTS_MIN); | |
1869 | max_skb_slots = XEN_NETIF_NR_SLOTS_MIN; | |
1870 | } | |
1871 | ||
f942dc25 IC |
1872 | xen_netbk_group_nr = num_online_cpus(); |
1873 | xen_netbk = vzalloc(sizeof(struct xen_netbk) * xen_netbk_group_nr); | |
e404decb | 1874 | if (!xen_netbk) |
f942dc25 | 1875 | return -ENOMEM; |
f942dc25 IC |
1876 | |
1877 | for (group = 0; group < xen_netbk_group_nr; group++) { | |
1878 | struct xen_netbk *netbk = &xen_netbk[group]; | |
1879 | skb_queue_head_init(&netbk->rx_queue); | |
1880 | skb_queue_head_init(&netbk->tx_queue); | |
1881 | ||
1882 | init_timer(&netbk->net_timer); | |
1883 | netbk->net_timer.data = (unsigned long)netbk; | |
1884 | netbk->net_timer.function = xen_netbk_alarm; | |
1885 | ||
1886 | netbk->pending_cons = 0; | |
1887 | netbk->pending_prod = MAX_PENDING_REQS; | |
1888 | for (i = 0; i < MAX_PENDING_REQS; i++) | |
1889 | netbk->pending_ring[i] = i; | |
1890 | ||
1891 | init_waitqueue_head(&netbk->wq); | |
1892 | netbk->task = kthread_create(xen_netbk_kthread, | |
1893 | (void *)netbk, | |
1894 | "netback/%u", group); | |
1895 | ||
1896 | if (IS_ERR(netbk->task)) { | |
6b84bd16 | 1897 | printk(KERN_ALERT "kthread_create() fails at netback\n"); |
f942dc25 IC |
1898 | del_timer(&netbk->net_timer); |
1899 | rc = PTR_ERR(netbk->task); | |
1900 | goto failed_init; | |
1901 | } | |
1902 | ||
1903 | kthread_bind(netbk->task, group); | |
1904 | ||
1905 | INIT_LIST_HEAD(&netbk->net_schedule_list); | |
1906 | ||
1907 | spin_lock_init(&netbk->net_schedule_list_lock); | |
1908 | ||
1909 | atomic_set(&netbk->netfront_count, 0); | |
1910 | ||
1911 | wake_up_process(netbk->task); | |
1912 | } | |
1913 | ||
1914 | rc = xenvif_xenbus_init(); | |
1915 | if (rc) | |
1916 | goto failed_init; | |
1917 | ||
1918 | return 0; | |
1919 | ||
1920 | failed_init: | |
1921 | while (--group >= 0) { | |
1922 | struct xen_netbk *netbk = &xen_netbk[group]; | |
1923 | for (i = 0; i < MAX_PENDING_REQS; i++) { | |
1924 | if (netbk->mmap_pages[i]) | |
1925 | __free_page(netbk->mmap_pages[i]); | |
1926 | } | |
1927 | del_timer(&netbk->net_timer); | |
1928 | kthread_stop(netbk->task); | |
1929 | } | |
1930 | vfree(xen_netbk); | |
1931 | return rc; | |
1932 | ||
1933 | } | |
1934 | ||
1935 | module_init(netback_init); | |
1936 | ||
1937 | MODULE_LICENSE("Dual BSD/GPL"); | |
f984cec6 | 1938 | MODULE_ALIAS("xen-backend:vif"); |