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> | |
ae5e8127 | 42 | #include <net/ip6_checksum.h> |
f942dc25 | 43 | |
ca981633 | 44 | #include <xen/xen.h> |
f942dc25 IC |
45 | #include <xen/events.h> |
46 | #include <xen/interface/memory.h> | |
47 | ||
48 | #include <asm/xen/hypercall.h> | |
49 | #include <asm/xen/page.h> | |
50 | ||
e1f00a69 WL |
51 | /* Provide an option to disable split event channels at load time as |
52 | * event channels are limited resource. Split event channels are | |
53 | * enabled by default. | |
54 | */ | |
55 | bool separate_tx_rx_irq = 1; | |
56 | module_param(separate_tx_rx_irq, bool, 0644); | |
57 | ||
2810e5b9 WL |
58 | /* |
59 | * This is the maximum slots a skb can have. If a guest sends a skb | |
60 | * which exceeds this limit it is considered malicious. | |
61 | */ | |
37641494 WL |
62 | #define FATAL_SKB_SLOTS_DEFAULT 20 |
63 | static unsigned int fatal_skb_slots = FATAL_SKB_SLOTS_DEFAULT; | |
64 | module_param(fatal_skb_slots, uint, 0444); | |
65 | ||
66 | /* | |
67 | * To avoid confusion, we define XEN_NETBK_LEGACY_SLOTS_MAX indicating | |
68 | * the maximum slots a valid packet can use. Now this value is defined | |
69 | * to be XEN_NETIF_NR_SLOTS_MIN, which is supposed to be supported by | |
70 | * all backend. | |
71 | */ | |
72 | #define XEN_NETBK_LEGACY_SLOTS_MAX XEN_NETIF_NR_SLOTS_MIN | |
2810e5b9 | 73 | |
2810e5b9 WL |
74 | /* |
75 | * If head != INVALID_PENDING_RING_IDX, it means this tx request is head of | |
76 | * one or more merged tx requests, otherwise it is the continuation of | |
77 | * previous tx request. | |
78 | */ | |
b3f980bd | 79 | static inline int pending_tx_is_head(struct xenvif *vif, RING_IDX idx) |
f942dc25 | 80 | { |
b3f980bd | 81 | return vif->pending_tx_info[idx].head != INVALID_PENDING_RING_IDX; |
f942dc25 IC |
82 | } |
83 | ||
7376419a WL |
84 | static void xenvif_idx_release(struct xenvif *vif, u16 pending_idx, |
85 | u8 status); | |
86 | ||
f942dc25 IC |
87 | static void make_tx_response(struct xenvif *vif, |
88 | struct xen_netif_tx_request *txp, | |
89 | s8 st); | |
b3f980bd WL |
90 | |
91 | static inline int tx_work_todo(struct xenvif *vif); | |
92 | static inline int rx_work_todo(struct xenvif *vif); | |
93 | ||
f942dc25 IC |
94 | static struct xen_netif_rx_response *make_rx_response(struct xenvif *vif, |
95 | u16 id, | |
96 | s8 st, | |
97 | u16 offset, | |
98 | u16 size, | |
99 | u16 flags); | |
100 | ||
b3f980bd | 101 | static inline unsigned long idx_to_pfn(struct xenvif *vif, |
ea066ad1 | 102 | u16 idx) |
f942dc25 | 103 | { |
b3f980bd | 104 | return page_to_pfn(vif->mmap_pages[idx]); |
f942dc25 IC |
105 | } |
106 | ||
b3f980bd | 107 | static inline unsigned long idx_to_kaddr(struct xenvif *vif, |
ea066ad1 | 108 | u16 idx) |
f942dc25 | 109 | { |
b3f980bd | 110 | return (unsigned long)pfn_to_kaddr(idx_to_pfn(vif, idx)); |
f942dc25 IC |
111 | } |
112 | ||
2eba61d5 PD |
113 | /* This is a miniumum size for the linear area to avoid lots of |
114 | * calls to __pskb_pull_tail() as we set up checksum offsets. The | |
115 | * value 128 was chosen as it covers all IPv4 and most likely | |
116 | * IPv6 headers. | |
f942dc25 | 117 | */ |
2eba61d5 | 118 | #define PKT_PROT_LEN 128 |
f942dc25 | 119 | |
ea066ad1 IC |
120 | static u16 frag_get_pending_idx(skb_frag_t *frag) |
121 | { | |
122 | return (u16)frag->page_offset; | |
123 | } | |
124 | ||
125 | static void frag_set_pending_idx(skb_frag_t *frag, u16 pending_idx) | |
126 | { | |
127 | frag->page_offset = pending_idx; | |
128 | } | |
129 | ||
f942dc25 IC |
130 | static inline pending_ring_idx_t pending_index(unsigned i) |
131 | { | |
132 | return i & (MAX_PENDING_REQS-1); | |
133 | } | |
134 | ||
b3f980bd | 135 | static inline pending_ring_idx_t nr_pending_reqs(struct xenvif *vif) |
f942dc25 IC |
136 | { |
137 | return MAX_PENDING_REQS - | |
b3f980bd | 138 | vif->pending_prod + vif->pending_cons; |
f942dc25 IC |
139 | } |
140 | ||
141 | static int max_required_rx_slots(struct xenvif *vif) | |
142 | { | |
143 | int max = DIV_ROUND_UP(vif->dev->mtu, PAGE_SIZE); | |
144 | ||
2810e5b9 | 145 | /* XXX FIXME: RX path dependent on MAX_SKB_FRAGS */ |
82cada22 | 146 | if (vif->can_sg || vif->gso_mask || vif->gso_prefix_mask) |
f942dc25 IC |
147 | max += MAX_SKB_FRAGS + 1; /* extra_info + frags */ |
148 | ||
149 | return max; | |
150 | } | |
151 | ||
7376419a | 152 | int xenvif_rx_ring_full(struct xenvif *vif) |
f942dc25 IC |
153 | { |
154 | RING_IDX peek = vif->rx_req_cons_peek; | |
155 | RING_IDX needed = max_required_rx_slots(vif); | |
156 | ||
157 | return ((vif->rx.sring->req_prod - peek) < needed) || | |
158 | ((vif->rx.rsp_prod_pvt + XEN_NETIF_RX_RING_SIZE - peek) < needed); | |
159 | } | |
160 | ||
7376419a | 161 | int xenvif_must_stop_queue(struct xenvif *vif) |
f942dc25 | 162 | { |
7376419a | 163 | if (!xenvif_rx_ring_full(vif)) |
f942dc25 IC |
164 | return 0; |
165 | ||
166 | vif->rx.sring->req_event = vif->rx_req_cons_peek + | |
167 | max_required_rx_slots(vif); | |
168 | mb(); /* request notification /then/ check the queue */ | |
169 | ||
7376419a | 170 | return xenvif_rx_ring_full(vif); |
f942dc25 IC |
171 | } |
172 | ||
173 | /* | |
174 | * Returns true if we should start a new receive buffer instead of | |
175 | * adding 'size' bytes to a buffer which currently contains 'offset' | |
176 | * bytes. | |
177 | */ | |
178 | static bool start_new_rx_buffer(int offset, unsigned long size, int head) | |
179 | { | |
180 | /* simple case: we have completely filled the current buffer. */ | |
181 | if (offset == MAX_BUFFER_OFFSET) | |
182 | return true; | |
183 | ||
184 | /* | |
185 | * complex case: start a fresh buffer if the current frag | |
186 | * would overflow the current buffer but only if: | |
187 | * (i) this frag would fit completely in the next buffer | |
188 | * and (ii) there is already some data in the current buffer | |
189 | * and (iii) this is not the head buffer. | |
190 | * | |
191 | * Where: | |
192 | * - (i) stops us splitting a frag into two copies | |
193 | * unless the frag is too large for a single buffer. | |
194 | * - (ii) stops us from leaving a buffer pointlessly empty. | |
195 | * - (iii) stops us leaving the first buffer | |
196 | * empty. Strictly speaking this is already covered | |
197 | * by (ii) but is explicitly checked because | |
198 | * netfront relies on the first buffer being | |
199 | * non-empty and can crash otherwise. | |
200 | * | |
201 | * This means we will effectively linearise small | |
202 | * frags but do not needlessly split large buffers | |
203 | * into multiple copies tend to give large frags their | |
204 | * own buffers as before. | |
205 | */ | |
206 | if ((offset + size > MAX_BUFFER_OFFSET) && | |
207 | (size <= MAX_BUFFER_OFFSET) && offset && !head) | |
208 | return true; | |
209 | ||
210 | return false; | |
211 | } | |
212 | ||
33bc801d WL |
213 | struct xenvif_count_slot_state { |
214 | unsigned long copy_off; | |
215 | bool head; | |
216 | }; | |
217 | ||
218 | unsigned int xenvif_count_frag_slots(struct xenvif *vif, | |
219 | unsigned long offset, unsigned long size, | |
220 | struct xenvif_count_slot_state *state) | |
221 | { | |
222 | unsigned count = 0; | |
223 | ||
224 | offset &= ~PAGE_MASK; | |
225 | ||
226 | while (size > 0) { | |
227 | unsigned long bytes; | |
228 | ||
229 | bytes = PAGE_SIZE - offset; | |
230 | ||
231 | if (bytes > size) | |
232 | bytes = size; | |
233 | ||
234 | if (start_new_rx_buffer(state->copy_off, bytes, state->head)) { | |
235 | count++; | |
236 | state->copy_off = 0; | |
237 | } | |
238 | ||
239 | if (state->copy_off + bytes > MAX_BUFFER_OFFSET) | |
240 | bytes = MAX_BUFFER_OFFSET - state->copy_off; | |
241 | ||
242 | state->copy_off += bytes; | |
243 | ||
244 | offset += bytes; | |
245 | size -= bytes; | |
246 | ||
247 | if (offset == PAGE_SIZE) | |
248 | offset = 0; | |
249 | ||
250 | state->head = false; | |
251 | } | |
252 | ||
253 | return count; | |
254 | } | |
255 | ||
f942dc25 IC |
256 | /* |
257 | * Figure out how many ring slots we're going to need to send @skb to | |
258 | * the guest. This function is essentially a dry run of | |
7376419a | 259 | * xenvif_gop_frag_copy. |
f942dc25 | 260 | */ |
7376419a | 261 | unsigned int xenvif_count_skb_slots(struct xenvif *vif, struct sk_buff *skb) |
f942dc25 | 262 | { |
33bc801d | 263 | struct xenvif_count_slot_state state; |
f942dc25 | 264 | unsigned int count; |
33bc801d WL |
265 | unsigned char *data; |
266 | unsigned i; | |
f942dc25 | 267 | |
33bc801d WL |
268 | state.head = true; |
269 | state.copy_off = 0; | |
f942dc25 | 270 | |
33bc801d WL |
271 | /* Slot for the first (partial) page of data. */ |
272 | count = 1; | |
f942dc25 | 273 | |
33bc801d | 274 | /* Need a slot for the GSO prefix for GSO extra data? */ |
f942dc25 IC |
275 | if (skb_shinfo(skb)->gso_size) |
276 | count++; | |
277 | ||
33bc801d WL |
278 | data = skb->data; |
279 | while (data < skb_tail_pointer(skb)) { | |
280 | unsigned long offset = offset_in_page(data); | |
281 | unsigned long size = PAGE_SIZE - offset; | |
4f0581d2 | 282 | |
33bc801d WL |
283 | if (data + size > skb_tail_pointer(skb)) |
284 | size = skb_tail_pointer(skb) - data; | |
6a8ed462 | 285 | |
33bc801d | 286 | count += xenvif_count_frag_slots(vif, offset, size, &state); |
4f0581d2 | 287 | |
33bc801d WL |
288 | data += size; |
289 | } | |
4f0581d2 | 290 | |
33bc801d WL |
291 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
292 | unsigned long size = skb_frag_size(&skb_shinfo(skb)->frags[i]); | |
293 | unsigned long offset = skb_shinfo(skb)->frags[i].page_offset; | |
4f0581d2 | 294 | |
33bc801d | 295 | count += xenvif_count_frag_slots(vif, offset, size, &state); |
f942dc25 IC |
296 | } |
297 | return count; | |
298 | } | |
299 | ||
300 | struct netrx_pending_operations { | |
301 | unsigned copy_prod, copy_cons; | |
302 | unsigned meta_prod, meta_cons; | |
303 | struct gnttab_copy *copy; | |
b3f980bd | 304 | struct xenvif_rx_meta *meta; |
f942dc25 IC |
305 | int copy_off; |
306 | grant_ref_t copy_gref; | |
307 | }; | |
308 | ||
b3f980bd WL |
309 | static struct xenvif_rx_meta *get_next_rx_buffer(struct xenvif *vif, |
310 | struct netrx_pending_operations *npo) | |
f942dc25 | 311 | { |
b3f980bd | 312 | struct xenvif_rx_meta *meta; |
f942dc25 IC |
313 | struct xen_netif_rx_request *req; |
314 | ||
315 | req = RING_GET_REQUEST(&vif->rx, vif->rx.req_cons++); | |
316 | ||
317 | meta = npo->meta + npo->meta_prod++; | |
82cada22 | 318 | meta->gso_type = XEN_NETIF_GSO_TYPE_NONE; |
f942dc25 IC |
319 | meta->gso_size = 0; |
320 | meta->size = 0; | |
321 | meta->id = req->id; | |
322 | ||
323 | npo->copy_off = 0; | |
324 | npo->copy_gref = req->gref; | |
325 | ||
326 | return meta; | |
327 | } | |
328 | ||
33bc801d WL |
329 | /* |
330 | * Set up the grant operations for this fragment. If it's a flipping | |
331 | * interface, we also set up the unmap request from here. | |
332 | */ | |
7376419a WL |
333 | static void xenvif_gop_frag_copy(struct xenvif *vif, struct sk_buff *skb, |
334 | struct netrx_pending_operations *npo, | |
335 | struct page *page, unsigned long size, | |
33bc801d | 336 | unsigned long offset, int *head) |
f942dc25 IC |
337 | { |
338 | struct gnttab_copy *copy_gop; | |
b3f980bd | 339 | struct xenvif_rx_meta *meta; |
f942dc25 | 340 | unsigned long bytes; |
82cada22 | 341 | int gso_type; |
f942dc25 IC |
342 | |
343 | /* Data must not cross a page boundary. */ | |
6a8ed462 | 344 | BUG_ON(size + offset > PAGE_SIZE<<compound_order(page)); |
f942dc25 IC |
345 | |
346 | meta = npo->meta + npo->meta_prod - 1; | |
347 | ||
6a8ed462 IC |
348 | /* Skip unused frames from start of page */ |
349 | page += offset >> PAGE_SHIFT; | |
350 | offset &= ~PAGE_MASK; | |
351 | ||
f942dc25 | 352 | while (size > 0) { |
6a8ed462 | 353 | BUG_ON(offset >= PAGE_SIZE); |
f942dc25 IC |
354 | BUG_ON(npo->copy_off > MAX_BUFFER_OFFSET); |
355 | ||
6a8ed462 IC |
356 | bytes = PAGE_SIZE - offset; |
357 | ||
358 | if (bytes > size) | |
359 | bytes = size; | |
360 | ||
33bc801d | 361 | if (start_new_rx_buffer(npo->copy_off, bytes, *head)) { |
f942dc25 IC |
362 | /* |
363 | * Netfront requires there to be some data in the head | |
364 | * buffer. | |
365 | */ | |
33bc801d | 366 | BUG_ON(*head); |
f942dc25 IC |
367 | |
368 | meta = get_next_rx_buffer(vif, npo); | |
369 | } | |
370 | ||
f942dc25 IC |
371 | if (npo->copy_off + bytes > MAX_BUFFER_OFFSET) |
372 | bytes = MAX_BUFFER_OFFSET - npo->copy_off; | |
373 | ||
374 | copy_gop = npo->copy + npo->copy_prod++; | |
375 | copy_gop->flags = GNTCOPY_dest_gref; | |
b3f980bd WL |
376 | copy_gop->len = bytes; |
377 | ||
43e9d194 WL |
378 | copy_gop->source.domid = DOMID_SELF; |
379 | copy_gop->source.u.gmfn = virt_to_mfn(page_address(page)); | |
f942dc25 | 380 | copy_gop->source.offset = offset; |
f942dc25 | 381 | |
b3f980bd | 382 | copy_gop->dest.domid = vif->domid; |
f942dc25 IC |
383 | copy_gop->dest.offset = npo->copy_off; |
384 | copy_gop->dest.u.ref = npo->copy_gref; | |
f942dc25 IC |
385 | |
386 | npo->copy_off += bytes; | |
387 | meta->size += bytes; | |
388 | ||
389 | offset += bytes; | |
390 | size -= bytes; | |
391 | ||
6a8ed462 IC |
392 | /* Next frame */ |
393 | if (offset == PAGE_SIZE && size) { | |
394 | BUG_ON(!PageCompound(page)); | |
395 | page++; | |
396 | offset = 0; | |
397 | } | |
398 | ||
f942dc25 | 399 | /* Leave a gap for the GSO descriptor. */ |
82cada22 PD |
400 | if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) |
401 | gso_type = XEN_NETIF_GSO_TYPE_TCPV4; | |
402 | else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) | |
403 | gso_type = XEN_NETIF_GSO_TYPE_TCPV6; | |
404 | else | |
405 | gso_type = XEN_NETIF_GSO_TYPE_NONE; | |
406 | ||
407 | if (*head && ((1 << gso_type) & vif->gso_mask)) | |
f942dc25 IC |
408 | vif->rx.req_cons++; |
409 | ||
33bc801d | 410 | *head = 0; /* There must be something in this buffer now. */ |
f942dc25 IC |
411 | |
412 | } | |
413 | } | |
414 | ||
415 | /* | |
416 | * Prepare an SKB to be transmitted to the frontend. | |
417 | * | |
418 | * This function is responsible for allocating grant operations, meta | |
419 | * structures, etc. | |
420 | * | |
421 | * It returns the number of meta structures consumed. The number of | |
422 | * ring slots used is always equal to the number of meta slots used | |
423 | * plus the number of GSO descriptors used. Currently, we use either | |
424 | * zero GSO descriptors (for non-GSO packets) or one descriptor (for | |
425 | * frontend-side LRO). | |
426 | */ | |
7376419a WL |
427 | static int xenvif_gop_skb(struct sk_buff *skb, |
428 | struct netrx_pending_operations *npo) | |
f942dc25 IC |
429 | { |
430 | struct xenvif *vif = netdev_priv(skb->dev); | |
431 | int nr_frags = skb_shinfo(skb)->nr_frags; | |
432 | int i; | |
433 | struct xen_netif_rx_request *req; | |
b3f980bd | 434 | struct xenvif_rx_meta *meta; |
f942dc25 | 435 | unsigned char *data; |
33bc801d | 436 | int head = 1; |
f942dc25 | 437 | int old_meta_prod; |
82cada22 PD |
438 | int gso_type; |
439 | int gso_size; | |
f942dc25 IC |
440 | |
441 | old_meta_prod = npo->meta_prod; | |
442 | ||
82cada22 PD |
443 | if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) { |
444 | gso_type = XEN_NETIF_GSO_TYPE_TCPV4; | |
445 | gso_size = skb_shinfo(skb)->gso_size; | |
446 | } else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) { | |
447 | gso_type = XEN_NETIF_GSO_TYPE_TCPV6; | |
448 | gso_size = skb_shinfo(skb)->gso_size; | |
449 | } else { | |
450 | gso_type = XEN_NETIF_GSO_TYPE_NONE; | |
451 | gso_size = 0; | |
452 | } | |
453 | ||
f942dc25 | 454 | /* Set up a GSO prefix descriptor, if necessary */ |
82cada22 | 455 | if ((1 << skb_shinfo(skb)->gso_type) & vif->gso_prefix_mask) { |
f942dc25 IC |
456 | req = RING_GET_REQUEST(&vif->rx, vif->rx.req_cons++); |
457 | meta = npo->meta + npo->meta_prod++; | |
82cada22 PD |
458 | meta->gso_type = gso_type; |
459 | meta->gso_size = gso_size; | |
f942dc25 IC |
460 | meta->size = 0; |
461 | meta->id = req->id; | |
462 | } | |
463 | ||
464 | req = RING_GET_REQUEST(&vif->rx, vif->rx.req_cons++); | |
465 | meta = npo->meta + npo->meta_prod++; | |
466 | ||
82cada22 PD |
467 | if ((1 << gso_type) & vif->gso_mask) { |
468 | meta->gso_type = gso_type; | |
469 | meta->gso_size = gso_size; | |
470 | } else { | |
471 | meta->gso_type = XEN_NETIF_GSO_TYPE_NONE; | |
f942dc25 | 472 | meta->gso_size = 0; |
82cada22 | 473 | } |
f942dc25 IC |
474 | |
475 | meta->size = 0; | |
476 | meta->id = req->id; | |
477 | npo->copy_off = 0; | |
478 | npo->copy_gref = req->gref; | |
479 | ||
480 | data = skb->data; | |
481 | while (data < skb_tail_pointer(skb)) { | |
482 | unsigned int offset = offset_in_page(data); | |
483 | unsigned int len = PAGE_SIZE - offset; | |
484 | ||
485 | if (data + len > skb_tail_pointer(skb)) | |
486 | len = skb_tail_pointer(skb) - data; | |
487 | ||
7376419a | 488 | xenvif_gop_frag_copy(vif, skb, npo, |
33bc801d | 489 | virt_to_page(data), len, offset, &head); |
f942dc25 IC |
490 | data += len; |
491 | } | |
492 | ||
493 | for (i = 0; i < nr_frags; i++) { | |
7376419a WL |
494 | xenvif_gop_frag_copy(vif, skb, npo, |
495 | skb_frag_page(&skb_shinfo(skb)->frags[i]), | |
496 | skb_frag_size(&skb_shinfo(skb)->frags[i]), | |
497 | skb_shinfo(skb)->frags[i].page_offset, | |
33bc801d | 498 | &head); |
f942dc25 IC |
499 | } |
500 | ||
501 | return npo->meta_prod - old_meta_prod; | |
502 | } | |
503 | ||
504 | /* | |
7376419a | 505 | * This is a twin to xenvif_gop_skb. Assume that xenvif_gop_skb was |
f942dc25 IC |
506 | * used to set up the operations on the top of |
507 | * netrx_pending_operations, which have since been done. Check that | |
508 | * they didn't give any errors and advance over them. | |
509 | */ | |
7376419a WL |
510 | static int xenvif_check_gop(struct xenvif *vif, int nr_meta_slots, |
511 | struct netrx_pending_operations *npo) | |
f942dc25 IC |
512 | { |
513 | struct gnttab_copy *copy_op; | |
514 | int status = XEN_NETIF_RSP_OKAY; | |
515 | int i; | |
516 | ||
517 | for (i = 0; i < nr_meta_slots; i++) { | |
518 | copy_op = npo->copy + npo->copy_cons++; | |
519 | if (copy_op->status != GNTST_okay) { | |
520 | netdev_dbg(vif->dev, | |
521 | "Bad status %d from copy to DOM%d.\n", | |
522 | copy_op->status, vif->domid); | |
523 | status = XEN_NETIF_RSP_ERROR; | |
524 | } | |
525 | } | |
526 | ||
527 | return status; | |
528 | } | |
529 | ||
7376419a WL |
530 | static void xenvif_add_frag_responses(struct xenvif *vif, int status, |
531 | struct xenvif_rx_meta *meta, | |
532 | int nr_meta_slots) | |
f942dc25 IC |
533 | { |
534 | int i; | |
535 | unsigned long offset; | |
536 | ||
537 | /* No fragments used */ | |
538 | if (nr_meta_slots <= 1) | |
539 | return; | |
540 | ||
541 | nr_meta_slots--; | |
542 | ||
543 | for (i = 0; i < nr_meta_slots; i++) { | |
544 | int flags; | |
545 | if (i == nr_meta_slots - 1) | |
546 | flags = 0; | |
547 | else | |
548 | flags = XEN_NETRXF_more_data; | |
549 | ||
550 | offset = 0; | |
551 | make_rx_response(vif, meta[i].id, status, offset, | |
552 | meta[i].size, flags); | |
553 | } | |
554 | } | |
555 | ||
33bc801d WL |
556 | struct skb_cb_overlay { |
557 | int meta_slots_used; | |
558 | }; | |
559 | ||
7376419a | 560 | static void xenvif_kick_thread(struct xenvif *vif) |
b3f980bd WL |
561 | { |
562 | wake_up(&vif->wq); | |
563 | } | |
564 | ||
7376419a | 565 | void xenvif_rx_action(struct xenvif *vif) |
f942dc25 | 566 | { |
f942dc25 | 567 | s8 status; |
e1f00a69 | 568 | u16 flags; |
f942dc25 IC |
569 | struct xen_netif_rx_response *resp; |
570 | struct sk_buff_head rxq; | |
571 | struct sk_buff *skb; | |
572 | LIST_HEAD(notify); | |
573 | int ret; | |
574 | int nr_frags; | |
575 | int count; | |
576 | unsigned long offset; | |
577 | struct skb_cb_overlay *sco; | |
b3f980bd | 578 | int need_to_notify = 0; |
f942dc25 IC |
579 | |
580 | struct netrx_pending_operations npo = { | |
b3f980bd WL |
581 | .copy = vif->grant_copy_op, |
582 | .meta = vif->meta, | |
f942dc25 IC |
583 | }; |
584 | ||
585 | skb_queue_head_init(&rxq); | |
586 | ||
587 | count = 0; | |
588 | ||
b3f980bd | 589 | while ((skb = skb_dequeue(&vif->rx_queue)) != NULL) { |
f942dc25 IC |
590 | vif = netdev_priv(skb->dev); |
591 | nr_frags = skb_shinfo(skb)->nr_frags; | |
592 | ||
593 | sco = (struct skb_cb_overlay *)skb->cb; | |
7376419a | 594 | sco->meta_slots_used = xenvif_gop_skb(skb, &npo); |
f942dc25 | 595 | |
33bc801d | 596 | count += nr_frags + 1; |
f942dc25 IC |
597 | |
598 | __skb_queue_tail(&rxq, skb); | |
599 | ||
600 | /* Filled the batch queue? */ | |
33bc801d WL |
601 | /* XXX FIXME: RX path dependent on MAX_SKB_FRAGS */ |
602 | if (count + MAX_SKB_FRAGS >= XEN_NETIF_RX_RING_SIZE) | |
f942dc25 IC |
603 | break; |
604 | } | |
605 | ||
b3f980bd | 606 | BUG_ON(npo.meta_prod > ARRAY_SIZE(vif->meta)); |
f942dc25 IC |
607 | |
608 | if (!npo.copy_prod) | |
609 | return; | |
610 | ||
b3f980bd WL |
611 | BUG_ON(npo.copy_prod > ARRAY_SIZE(vif->grant_copy_op)); |
612 | gnttab_batch_copy(vif->grant_copy_op, npo.copy_prod); | |
f942dc25 IC |
613 | |
614 | while ((skb = __skb_dequeue(&rxq)) != NULL) { | |
615 | sco = (struct skb_cb_overlay *)skb->cb; | |
616 | ||
617 | vif = netdev_priv(skb->dev); | |
618 | ||
82cada22 PD |
619 | if ((1 << vif->meta[npo.meta_cons].gso_type) & |
620 | vif->gso_prefix_mask) { | |
f942dc25 | 621 | resp = RING_GET_RESPONSE(&vif->rx, |
b3f980bd | 622 | vif->rx.rsp_prod_pvt++); |
f942dc25 IC |
623 | |
624 | resp->flags = XEN_NETRXF_gso_prefix | XEN_NETRXF_more_data; | |
625 | ||
b3f980bd WL |
626 | resp->offset = vif->meta[npo.meta_cons].gso_size; |
627 | resp->id = vif->meta[npo.meta_cons].id; | |
f942dc25 IC |
628 | resp->status = sco->meta_slots_used; |
629 | ||
630 | npo.meta_cons++; | |
631 | sco->meta_slots_used--; | |
632 | } | |
633 | ||
634 | ||
635 | vif->dev->stats.tx_bytes += skb->len; | |
636 | vif->dev->stats.tx_packets++; | |
637 | ||
7376419a | 638 | status = xenvif_check_gop(vif, sco->meta_slots_used, &npo); |
f942dc25 IC |
639 | |
640 | if (sco->meta_slots_used == 1) | |
641 | flags = 0; | |
642 | else | |
643 | flags = XEN_NETRXF_more_data; | |
644 | ||
645 | if (skb->ip_summed == CHECKSUM_PARTIAL) /* local packet? */ | |
646 | flags |= XEN_NETRXF_csum_blank | XEN_NETRXF_data_validated; | |
647 | else if (skb->ip_summed == CHECKSUM_UNNECESSARY) | |
648 | /* remote but checksummed. */ | |
649 | flags |= XEN_NETRXF_data_validated; | |
650 | ||
651 | offset = 0; | |
b3f980bd | 652 | resp = make_rx_response(vif, vif->meta[npo.meta_cons].id, |
f942dc25 | 653 | status, offset, |
b3f980bd | 654 | vif->meta[npo.meta_cons].size, |
f942dc25 IC |
655 | flags); |
656 | ||
82cada22 PD |
657 | if ((1 << vif->meta[npo.meta_cons].gso_type) & |
658 | vif->gso_mask) { | |
f942dc25 IC |
659 | struct xen_netif_extra_info *gso = |
660 | (struct xen_netif_extra_info *) | |
661 | RING_GET_RESPONSE(&vif->rx, | |
662 | vif->rx.rsp_prod_pvt++); | |
663 | ||
664 | resp->flags |= XEN_NETRXF_extra_info; | |
665 | ||
82cada22 | 666 | gso->u.gso.type = vif->meta[npo.meta_cons].gso_type; |
b3f980bd | 667 | gso->u.gso.size = vif->meta[npo.meta_cons].gso_size; |
f942dc25 IC |
668 | gso->u.gso.pad = 0; |
669 | gso->u.gso.features = 0; | |
670 | ||
671 | gso->type = XEN_NETIF_EXTRA_TYPE_GSO; | |
672 | gso->flags = 0; | |
673 | } | |
674 | ||
7376419a WL |
675 | xenvif_add_frag_responses(vif, status, |
676 | vif->meta + npo.meta_cons + 1, | |
677 | sco->meta_slots_used); | |
f942dc25 IC |
678 | |
679 | RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&vif->rx, ret); | |
f942dc25 | 680 | |
b3f980bd WL |
681 | if (ret) |
682 | need_to_notify = 1; | |
683 | ||
f942dc25 IC |
684 | xenvif_notify_tx_completion(vif); |
685 | ||
f942dc25 IC |
686 | npo.meta_cons += sco->meta_slots_used; |
687 | dev_kfree_skb(skb); | |
688 | } | |
689 | ||
b3f980bd | 690 | if (need_to_notify) |
e1f00a69 | 691 | notify_remote_via_irq(vif->rx_irq); |
f942dc25 IC |
692 | |
693 | /* More work to do? */ | |
b3f980bd | 694 | if (!skb_queue_empty(&vif->rx_queue)) |
7376419a | 695 | xenvif_kick_thread(vif); |
f942dc25 IC |
696 | } |
697 | ||
7376419a | 698 | void xenvif_queue_tx_skb(struct xenvif *vif, struct sk_buff *skb) |
f942dc25 | 699 | { |
b3f980bd | 700 | skb_queue_tail(&vif->rx_queue, skb); |
f942dc25 | 701 | |
7376419a | 702 | xenvif_kick_thread(vif); |
f942dc25 IC |
703 | } |
704 | ||
7376419a | 705 | void xenvif_check_rx_xenvif(struct xenvif *vif) |
f942dc25 IC |
706 | { |
707 | int more_to_do; | |
708 | ||
709 | RING_FINAL_CHECK_FOR_REQUESTS(&vif->tx, more_to_do); | |
710 | ||
711 | if (more_to_do) | |
b3f980bd | 712 | napi_schedule(&vif->napi); |
f942dc25 IC |
713 | } |
714 | ||
715 | static void tx_add_credit(struct xenvif *vif) | |
716 | { | |
717 | unsigned long max_burst, max_credit; | |
718 | ||
719 | /* | |
720 | * Allow a burst big enough to transmit a jumbo packet of up to 128kB. | |
721 | * Otherwise the interface can seize up due to insufficient credit. | |
722 | */ | |
723 | max_burst = RING_GET_REQUEST(&vif->tx, vif->tx.req_cons)->size; | |
724 | max_burst = min(max_burst, 131072UL); | |
725 | max_burst = max(max_burst, vif->credit_bytes); | |
726 | ||
727 | /* Take care that adding a new chunk of credit doesn't wrap to zero. */ | |
728 | max_credit = vif->remaining_credit + vif->credit_bytes; | |
729 | if (max_credit < vif->remaining_credit) | |
730 | max_credit = ULONG_MAX; /* wrapped: clamp to ULONG_MAX */ | |
731 | ||
732 | vif->remaining_credit = min(max_credit, max_burst); | |
733 | } | |
734 | ||
735 | static void tx_credit_callback(unsigned long data) | |
736 | { | |
737 | struct xenvif *vif = (struct xenvif *)data; | |
738 | tx_add_credit(vif); | |
7376419a | 739 | xenvif_check_rx_xenvif(vif); |
f942dc25 IC |
740 | } |
741 | ||
7376419a WL |
742 | static void xenvif_tx_err(struct xenvif *vif, |
743 | struct xen_netif_tx_request *txp, RING_IDX end) | |
f942dc25 IC |
744 | { |
745 | RING_IDX cons = vif->tx.req_cons; | |
746 | ||
747 | do { | |
748 | make_tx_response(vif, txp, XEN_NETIF_RSP_ERROR); | |
b9149729 | 749 | if (cons == end) |
f942dc25 IC |
750 | break; |
751 | txp = RING_GET_REQUEST(&vif->tx, cons++); | |
752 | } while (1); | |
753 | vif->tx.req_cons = cons; | |
f942dc25 IC |
754 | } |
755 | ||
7376419a | 756 | static void xenvif_fatal_tx_err(struct xenvif *vif) |
48856286 IC |
757 | { |
758 | netdev_err(vif->dev, "fatal error; disabling device\n"); | |
759 | xenvif_carrier_off(vif); | |
48856286 IC |
760 | } |
761 | ||
7376419a WL |
762 | static int xenvif_count_requests(struct xenvif *vif, |
763 | struct xen_netif_tx_request *first, | |
764 | struct xen_netif_tx_request *txp, | |
765 | int work_to_do) | |
f942dc25 IC |
766 | { |
767 | RING_IDX cons = vif->tx.req_cons; | |
2810e5b9 WL |
768 | int slots = 0; |
769 | int drop_err = 0; | |
59ccb4eb | 770 | int more_data; |
f942dc25 IC |
771 | |
772 | if (!(first->flags & XEN_NETTXF_more_data)) | |
773 | return 0; | |
774 | ||
775 | do { | |
59ccb4eb WL |
776 | struct xen_netif_tx_request dropped_tx = { 0 }; |
777 | ||
2810e5b9 WL |
778 | if (slots >= work_to_do) { |
779 | netdev_err(vif->dev, | |
780 | "Asked for %d slots but exceeds this limit\n", | |
781 | work_to_do); | |
7376419a | 782 | xenvif_fatal_tx_err(vif); |
35876b5f | 783 | return -ENODATA; |
f942dc25 IC |
784 | } |
785 | ||
2810e5b9 WL |
786 | /* This guest is really using too many slots and |
787 | * considered malicious. | |
788 | */ | |
37641494 | 789 | if (unlikely(slots >= fatal_skb_slots)) { |
2810e5b9 WL |
790 | netdev_err(vif->dev, |
791 | "Malicious frontend using %d slots, threshold %u\n", | |
37641494 | 792 | slots, fatal_skb_slots); |
7376419a | 793 | xenvif_fatal_tx_err(vif); |
35876b5f | 794 | return -E2BIG; |
f942dc25 IC |
795 | } |
796 | ||
2810e5b9 | 797 | /* Xen network protocol had implicit dependency on |
37641494 WL |
798 | * MAX_SKB_FRAGS. XEN_NETBK_LEGACY_SLOTS_MAX is set to |
799 | * the historical MAX_SKB_FRAGS value 18 to honor the | |
800 | * same behavior as before. Any packet using more than | |
801 | * 18 slots but less than fatal_skb_slots slots is | |
802 | * dropped | |
2810e5b9 | 803 | */ |
37641494 | 804 | if (!drop_err && slots >= XEN_NETBK_LEGACY_SLOTS_MAX) { |
2810e5b9 WL |
805 | if (net_ratelimit()) |
806 | netdev_dbg(vif->dev, | |
807 | "Too many slots (%d) exceeding limit (%d), dropping packet\n", | |
37641494 | 808 | slots, XEN_NETBK_LEGACY_SLOTS_MAX); |
2810e5b9 WL |
809 | drop_err = -E2BIG; |
810 | } | |
811 | ||
59ccb4eb WL |
812 | if (drop_err) |
813 | txp = &dropped_tx; | |
814 | ||
2810e5b9 | 815 | memcpy(txp, RING_GET_REQUEST(&vif->tx, cons + slots), |
f942dc25 | 816 | sizeof(*txp)); |
03393fd5 WL |
817 | |
818 | /* If the guest submitted a frame >= 64 KiB then | |
819 | * first->size overflowed and following slots will | |
820 | * appear to be larger than the frame. | |
821 | * | |
822 | * This cannot be fatal error as there are buggy | |
823 | * frontends that do this. | |
824 | * | |
825 | * Consume all slots and drop the packet. | |
826 | */ | |
827 | if (!drop_err && txp->size > first->size) { | |
828 | if (net_ratelimit()) | |
829 | netdev_dbg(vif->dev, | |
830 | "Invalid tx request, slot size %u > remaining size %u\n", | |
831 | txp->size, first->size); | |
832 | drop_err = -EIO; | |
f942dc25 IC |
833 | } |
834 | ||
835 | first->size -= txp->size; | |
2810e5b9 | 836 | slots++; |
f942dc25 IC |
837 | |
838 | if (unlikely((txp->offset + txp->size) > PAGE_SIZE)) { | |
2810e5b9 | 839 | netdev_err(vif->dev, "Cross page boundary, txp->offset: %x, size: %u\n", |
f942dc25 | 840 | txp->offset, txp->size); |
7376419a | 841 | xenvif_fatal_tx_err(vif); |
35876b5f | 842 | return -EINVAL; |
f942dc25 | 843 | } |
59ccb4eb WL |
844 | |
845 | more_data = txp->flags & XEN_NETTXF_more_data; | |
846 | ||
847 | if (!drop_err) | |
848 | txp++; | |
849 | ||
850 | } while (more_data); | |
2810e5b9 WL |
851 | |
852 | if (drop_err) { | |
7376419a | 853 | xenvif_tx_err(vif, first, cons + slots); |
2810e5b9 WL |
854 | return drop_err; |
855 | } | |
856 | ||
857 | return slots; | |
f942dc25 IC |
858 | } |
859 | ||
7376419a WL |
860 | static struct page *xenvif_alloc_page(struct xenvif *vif, |
861 | u16 pending_idx) | |
f942dc25 IC |
862 | { |
863 | struct page *page; | |
b3f980bd WL |
864 | |
865 | page = alloc_page(GFP_ATOMIC|__GFP_COLD); | |
f942dc25 IC |
866 | if (!page) |
867 | return NULL; | |
b3f980bd WL |
868 | vif->mmap_pages[pending_idx] = page; |
869 | ||
f942dc25 IC |
870 | return page; |
871 | } | |
872 | ||
7376419a WL |
873 | static struct gnttab_copy *xenvif_get_requests(struct xenvif *vif, |
874 | struct sk_buff *skb, | |
875 | struct xen_netif_tx_request *txp, | |
876 | struct gnttab_copy *gop) | |
f942dc25 IC |
877 | { |
878 | struct skb_shared_info *shinfo = skb_shinfo(skb); | |
879 | skb_frag_t *frags = shinfo->frags; | |
ea066ad1 | 880 | u16 pending_idx = *((u16 *)skb->data); |
2810e5b9 WL |
881 | u16 head_idx = 0; |
882 | int slot, start; | |
883 | struct page *page; | |
884 | pending_ring_idx_t index, start_idx = 0; | |
885 | uint16_t dst_offset; | |
886 | unsigned int nr_slots; | |
887 | struct pending_tx_info *first = NULL; | |
888 | ||
889 | /* At this point shinfo->nr_frags is in fact the number of | |
37641494 | 890 | * slots, which can be as large as XEN_NETBK_LEGACY_SLOTS_MAX. |
2810e5b9 WL |
891 | */ |
892 | nr_slots = shinfo->nr_frags; | |
f942dc25 IC |
893 | |
894 | /* Skip first skb fragment if it is on same page as header fragment. */ | |
ea066ad1 | 895 | start = (frag_get_pending_idx(&shinfo->frags[0]) == pending_idx); |
f942dc25 | 896 | |
2810e5b9 WL |
897 | /* Coalesce tx requests, at this point the packet passed in |
898 | * should be <= 64K. Any packets larger than 64K have been | |
7376419a | 899 | * handled in xenvif_count_requests(). |
2810e5b9 WL |
900 | */ |
901 | for (shinfo->nr_frags = slot = start; slot < nr_slots; | |
902 | shinfo->nr_frags++) { | |
f942dc25 | 903 | struct pending_tx_info *pending_tx_info = |
b3f980bd | 904 | vif->pending_tx_info; |
f942dc25 | 905 | |
b3f980bd | 906 | page = alloc_page(GFP_ATOMIC|__GFP_COLD); |
f942dc25 | 907 | if (!page) |
4cc7c1cb | 908 | goto err; |
f942dc25 | 909 | |
2810e5b9 WL |
910 | dst_offset = 0; |
911 | first = NULL; | |
912 | while (dst_offset < PAGE_SIZE && slot < nr_slots) { | |
913 | gop->flags = GNTCOPY_source_gref; | |
914 | ||
915 | gop->source.u.ref = txp->gref; | |
916 | gop->source.domid = vif->domid; | |
917 | gop->source.offset = txp->offset; | |
918 | ||
919 | gop->dest.domid = DOMID_SELF; | |
920 | ||
921 | gop->dest.offset = dst_offset; | |
922 | gop->dest.u.gmfn = virt_to_mfn(page_address(page)); | |
923 | ||
924 | if (dst_offset + txp->size > PAGE_SIZE) { | |
925 | /* This page can only merge a portion | |
926 | * of tx request. Do not increment any | |
927 | * pointer / counter here. The txp | |
928 | * will be dealt with in future | |
929 | * rounds, eventually hitting the | |
930 | * `else` branch. | |
931 | */ | |
932 | gop->len = PAGE_SIZE - dst_offset; | |
933 | txp->offset += gop->len; | |
934 | txp->size -= gop->len; | |
935 | dst_offset += gop->len; /* quit loop */ | |
936 | } else { | |
937 | /* This tx request can be merged in the page */ | |
938 | gop->len = txp->size; | |
939 | dst_offset += gop->len; | |
940 | ||
b3f980bd | 941 | index = pending_index(vif->pending_cons++); |
2810e5b9 | 942 | |
b3f980bd | 943 | pending_idx = vif->pending_ring[index]; |
2810e5b9 WL |
944 | |
945 | memcpy(&pending_tx_info[pending_idx].req, txp, | |
946 | sizeof(*txp)); | |
2810e5b9 WL |
947 | |
948 | /* Poison these fields, corresponding | |
949 | * fields for head tx req will be set | |
950 | * to correct values after the loop. | |
951 | */ | |
b3f980bd | 952 | vif->mmap_pages[pending_idx] = (void *)(~0UL); |
2810e5b9 WL |
953 | pending_tx_info[pending_idx].head = |
954 | INVALID_PENDING_RING_IDX; | |
955 | ||
956 | if (!first) { | |
957 | first = &pending_tx_info[pending_idx]; | |
958 | start_idx = index; | |
959 | head_idx = pending_idx; | |
960 | } | |
961 | ||
962 | txp++; | |
963 | slot++; | |
964 | } | |
f942dc25 | 965 | |
2810e5b9 WL |
966 | gop++; |
967 | } | |
f942dc25 | 968 | |
2810e5b9 WL |
969 | first->req.offset = 0; |
970 | first->req.size = dst_offset; | |
971 | first->head = start_idx; | |
b3f980bd | 972 | vif->mmap_pages[head_idx] = page; |
2810e5b9 | 973 | frag_set_pending_idx(&frags[shinfo->nr_frags], head_idx); |
f942dc25 IC |
974 | } |
975 | ||
2810e5b9 WL |
976 | BUG_ON(shinfo->nr_frags > MAX_SKB_FRAGS); |
977 | ||
f942dc25 | 978 | return gop; |
4cc7c1cb IC |
979 | err: |
980 | /* Unwind, freeing all pages and sending error responses. */ | |
2810e5b9 | 981 | while (shinfo->nr_frags-- > start) { |
7376419a | 982 | xenvif_idx_release(vif, |
2810e5b9 WL |
983 | frag_get_pending_idx(&frags[shinfo->nr_frags]), |
984 | XEN_NETIF_RSP_ERROR); | |
4cc7c1cb IC |
985 | } |
986 | /* The head too, if necessary. */ | |
987 | if (start) | |
7376419a | 988 | xenvif_idx_release(vif, pending_idx, XEN_NETIF_RSP_ERROR); |
4cc7c1cb IC |
989 | |
990 | return NULL; | |
f942dc25 IC |
991 | } |
992 | ||
7376419a WL |
993 | static int xenvif_tx_check_gop(struct xenvif *vif, |
994 | struct sk_buff *skb, | |
995 | struct gnttab_copy **gopp) | |
f942dc25 IC |
996 | { |
997 | struct gnttab_copy *gop = *gopp; | |
ea066ad1 | 998 | u16 pending_idx = *((u16 *)skb->data); |
f942dc25 | 999 | struct skb_shared_info *shinfo = skb_shinfo(skb); |
2810e5b9 | 1000 | struct pending_tx_info *tx_info; |
f942dc25 IC |
1001 | int nr_frags = shinfo->nr_frags; |
1002 | int i, err, start; | |
2810e5b9 | 1003 | u16 peek; /* peek into next tx request */ |
f942dc25 IC |
1004 | |
1005 | /* Check status of header. */ | |
1006 | err = gop->status; | |
7d5145d8 | 1007 | if (unlikely(err)) |
7376419a | 1008 | xenvif_idx_release(vif, pending_idx, XEN_NETIF_RSP_ERROR); |
f942dc25 IC |
1009 | |
1010 | /* Skip first skb fragment if it is on same page as header fragment. */ | |
ea066ad1 | 1011 | start = (frag_get_pending_idx(&shinfo->frags[0]) == pending_idx); |
f942dc25 IC |
1012 | |
1013 | for (i = start; i < nr_frags; i++) { | |
1014 | int j, newerr; | |
2810e5b9 | 1015 | pending_ring_idx_t head; |
f942dc25 | 1016 | |
ea066ad1 | 1017 | pending_idx = frag_get_pending_idx(&shinfo->frags[i]); |
b3f980bd | 1018 | tx_info = &vif->pending_tx_info[pending_idx]; |
2810e5b9 | 1019 | head = tx_info->head; |
f942dc25 IC |
1020 | |
1021 | /* Check error status: if okay then remember grant handle. */ | |
2810e5b9 WL |
1022 | do { |
1023 | newerr = (++gop)->status; | |
1024 | if (newerr) | |
1025 | break; | |
b3f980bd WL |
1026 | peek = vif->pending_ring[pending_index(++head)]; |
1027 | } while (!pending_tx_is_head(vif, peek)); | |
2810e5b9 | 1028 | |
f942dc25 IC |
1029 | if (likely(!newerr)) { |
1030 | /* Had a previous error? Invalidate this fragment. */ | |
1031 | if (unlikely(err)) | |
7376419a WL |
1032 | xenvif_idx_release(vif, pending_idx, |
1033 | XEN_NETIF_RSP_OKAY); | |
f942dc25 IC |
1034 | continue; |
1035 | } | |
1036 | ||
1037 | /* Error on this fragment: respond to client with an error. */ | |
7376419a | 1038 | xenvif_idx_release(vif, pending_idx, XEN_NETIF_RSP_ERROR); |
f942dc25 IC |
1039 | |
1040 | /* Not the first error? Preceding frags already invalidated. */ | |
1041 | if (err) | |
1042 | continue; | |
1043 | ||
1044 | /* First error: invalidate header and preceding fragments. */ | |
1045 | pending_idx = *((u16 *)skb->data); | |
7376419a | 1046 | xenvif_idx_release(vif, pending_idx, XEN_NETIF_RSP_OKAY); |
f942dc25 | 1047 | for (j = start; j < i; j++) { |
5ccb3ea7 | 1048 | pending_idx = frag_get_pending_idx(&shinfo->frags[j]); |
7376419a WL |
1049 | xenvif_idx_release(vif, pending_idx, |
1050 | XEN_NETIF_RSP_OKAY); | |
f942dc25 IC |
1051 | } |
1052 | ||
1053 | /* Remember the error: invalidate all subsequent fragments. */ | |
1054 | err = newerr; | |
1055 | } | |
1056 | ||
1057 | *gopp = gop + 1; | |
1058 | return err; | |
1059 | } | |
1060 | ||
7376419a | 1061 | static void xenvif_fill_frags(struct xenvif *vif, struct sk_buff *skb) |
f942dc25 IC |
1062 | { |
1063 | struct skb_shared_info *shinfo = skb_shinfo(skb); | |
1064 | int nr_frags = shinfo->nr_frags; | |
1065 | int i; | |
1066 | ||
1067 | for (i = 0; i < nr_frags; i++) { | |
1068 | skb_frag_t *frag = shinfo->frags + i; | |
1069 | struct xen_netif_tx_request *txp; | |
ea066ad1 IC |
1070 | struct page *page; |
1071 | u16 pending_idx; | |
f942dc25 | 1072 | |
ea066ad1 | 1073 | pending_idx = frag_get_pending_idx(frag); |
f942dc25 | 1074 | |
b3f980bd WL |
1075 | txp = &vif->pending_tx_info[pending_idx].req; |
1076 | page = virt_to_page(idx_to_kaddr(vif, pending_idx)); | |
ea066ad1 | 1077 | __skb_fill_page_desc(skb, i, page, txp->offset, txp->size); |
f942dc25 IC |
1078 | skb->len += txp->size; |
1079 | skb->data_len += txp->size; | |
1080 | skb->truesize += txp->size; | |
1081 | ||
7376419a | 1082 | /* Take an extra reference to offset xenvif_idx_release */ |
b3f980bd | 1083 | get_page(vif->mmap_pages[pending_idx]); |
7376419a | 1084 | xenvif_idx_release(vif, pending_idx, XEN_NETIF_RSP_OKAY); |
f942dc25 IC |
1085 | } |
1086 | } | |
1087 | ||
7376419a | 1088 | static int xenvif_get_extras(struct xenvif *vif, |
f942dc25 IC |
1089 | struct xen_netif_extra_info *extras, |
1090 | int work_to_do) | |
1091 | { | |
1092 | struct xen_netif_extra_info extra; | |
1093 | RING_IDX cons = vif->tx.req_cons; | |
1094 | ||
1095 | do { | |
1096 | if (unlikely(work_to_do-- <= 0)) { | |
48856286 | 1097 | netdev_err(vif->dev, "Missing extra info\n"); |
7376419a | 1098 | xenvif_fatal_tx_err(vif); |
f942dc25 IC |
1099 | return -EBADR; |
1100 | } | |
1101 | ||
1102 | memcpy(&extra, RING_GET_REQUEST(&vif->tx, cons), | |
1103 | sizeof(extra)); | |
1104 | if (unlikely(!extra.type || | |
1105 | extra.type >= XEN_NETIF_EXTRA_TYPE_MAX)) { | |
1106 | vif->tx.req_cons = ++cons; | |
48856286 | 1107 | netdev_err(vif->dev, |
f942dc25 | 1108 | "Invalid extra type: %d\n", extra.type); |
7376419a | 1109 | xenvif_fatal_tx_err(vif); |
f942dc25 IC |
1110 | return -EINVAL; |
1111 | } | |
1112 | ||
1113 | memcpy(&extras[extra.type - 1], &extra, sizeof(extra)); | |
1114 | vif->tx.req_cons = ++cons; | |
1115 | } while (extra.flags & XEN_NETIF_EXTRA_FLAG_MORE); | |
1116 | ||
1117 | return work_to_do; | |
1118 | } | |
1119 | ||
7376419a WL |
1120 | static int xenvif_set_skb_gso(struct xenvif *vif, |
1121 | struct sk_buff *skb, | |
1122 | struct xen_netif_extra_info *gso) | |
f942dc25 IC |
1123 | { |
1124 | if (!gso->u.gso.size) { | |
48856286 | 1125 | netdev_err(vif->dev, "GSO size must not be zero.\n"); |
7376419a | 1126 | xenvif_fatal_tx_err(vif); |
f942dc25 IC |
1127 | return -EINVAL; |
1128 | } | |
1129 | ||
a9468587 PD |
1130 | switch (gso->u.gso.type) { |
1131 | case XEN_NETIF_GSO_TYPE_TCPV4: | |
1132 | skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4; | |
1133 | break; | |
1134 | case XEN_NETIF_GSO_TYPE_TCPV6: | |
1135 | skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6; | |
1136 | break; | |
1137 | default: | |
48856286 | 1138 | netdev_err(vif->dev, "Bad GSO type %d.\n", gso->u.gso.type); |
7376419a | 1139 | xenvif_fatal_tx_err(vif); |
f942dc25 IC |
1140 | return -EINVAL; |
1141 | } | |
1142 | ||
1143 | skb_shinfo(skb)->gso_size = gso->u.gso.size; | |
f942dc25 IC |
1144 | |
1145 | /* Header must be checked, and gso_segs computed. */ | |
1146 | skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; | |
1147 | skb_shinfo(skb)->gso_segs = 0; | |
1148 | ||
1149 | return 0; | |
1150 | } | |
1151 | ||
2eba61d5 PD |
1152 | static inline void maybe_pull_tail(struct sk_buff *skb, unsigned int len) |
1153 | { | |
1154 | if (skb_is_nonlinear(skb) && skb_headlen(skb) < len) { | |
1155 | /* If we need to pullup then pullup to the max, so we | |
1156 | * won't need to do it again. | |
1157 | */ | |
1158 | int target = min_t(int, skb->len, MAX_TCP_HEADER); | |
1159 | __pskb_pull_tail(skb, target - skb_headlen(skb)); | |
1160 | } | |
1161 | } | |
1162 | ||
1163 | static int checksum_setup_ip(struct xenvif *vif, struct sk_buff *skb, | |
1164 | int recalculate_partial_csum) | |
f942dc25 | 1165 | { |
2eba61d5 PD |
1166 | struct iphdr *iph = (void *)skb->data; |
1167 | unsigned int header_size; | |
1168 | unsigned int off; | |
f942dc25 | 1169 | int err = -EPROTO; |
f942dc25 | 1170 | |
2eba61d5 | 1171 | off = sizeof(struct iphdr); |
f942dc25 | 1172 | |
2eba61d5 PD |
1173 | header_size = skb->network_header + off + MAX_IPOPTLEN; |
1174 | maybe_pull_tail(skb, header_size); | |
f942dc25 | 1175 | |
2eba61d5 | 1176 | off = iph->ihl * 4; |
f942dc25 | 1177 | |
f942dc25 IC |
1178 | switch (iph->protocol) { |
1179 | case IPPROTO_TCP: | |
2eba61d5 | 1180 | if (!skb_partial_csum_set(skb, off, |
bea89336 JW |
1181 | offsetof(struct tcphdr, check))) |
1182 | goto out; | |
f942dc25 IC |
1183 | |
1184 | if (recalculate_partial_csum) { | |
bea89336 | 1185 | struct tcphdr *tcph = tcp_hdr(skb); |
2eba61d5 PD |
1186 | |
1187 | header_size = skb->network_header + | |
1188 | off + | |
1189 | sizeof(struct tcphdr); | |
1190 | maybe_pull_tail(skb, header_size); | |
1191 | ||
f942dc25 | 1192 | tcph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, |
2eba61d5 | 1193 | skb->len - off, |
f942dc25 IC |
1194 | IPPROTO_TCP, 0); |
1195 | } | |
1196 | break; | |
1197 | case IPPROTO_UDP: | |
2eba61d5 | 1198 | if (!skb_partial_csum_set(skb, off, |
bea89336 JW |
1199 | offsetof(struct udphdr, check))) |
1200 | goto out; | |
f942dc25 IC |
1201 | |
1202 | if (recalculate_partial_csum) { | |
bea89336 | 1203 | struct udphdr *udph = udp_hdr(skb); |
2eba61d5 PD |
1204 | |
1205 | header_size = skb->network_header + | |
1206 | off + | |
1207 | sizeof(struct udphdr); | |
1208 | maybe_pull_tail(skb, header_size); | |
1209 | ||
f942dc25 | 1210 | udph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, |
2eba61d5 | 1211 | skb->len - off, |
f942dc25 IC |
1212 | IPPROTO_UDP, 0); |
1213 | } | |
1214 | break; | |
1215 | default: | |
1216 | if (net_ratelimit()) | |
1217 | netdev_err(vif->dev, | |
2eba61d5 PD |
1218 | "Attempting to checksum a non-TCP/UDP packet, " |
1219 | "dropping a protocol %d packet\n", | |
f942dc25 IC |
1220 | iph->protocol); |
1221 | goto out; | |
1222 | } | |
1223 | ||
f942dc25 IC |
1224 | err = 0; |
1225 | ||
1226 | out: | |
1227 | return err; | |
1228 | } | |
1229 | ||
2eba61d5 PD |
1230 | static int checksum_setup_ipv6(struct xenvif *vif, struct sk_buff *skb, |
1231 | int recalculate_partial_csum) | |
1232 | { | |
1233 | int err = -EPROTO; | |
1234 | struct ipv6hdr *ipv6h = (void *)skb->data; | |
1235 | u8 nexthdr; | |
1236 | unsigned int header_size; | |
1237 | unsigned int off; | |
1238 | bool fragment; | |
1239 | bool done; | |
1240 | ||
1241 | done = false; | |
1242 | ||
1243 | off = sizeof(struct ipv6hdr); | |
1244 | ||
1245 | header_size = skb->network_header + off; | |
1246 | maybe_pull_tail(skb, header_size); | |
1247 | ||
1248 | nexthdr = ipv6h->nexthdr; | |
1249 | ||
1250 | while ((off <= sizeof(struct ipv6hdr) + ntohs(ipv6h->payload_len)) && | |
1251 | !done) { | |
1252 | switch (nexthdr) { | |
1253 | case IPPROTO_DSTOPTS: | |
1254 | case IPPROTO_HOPOPTS: | |
1255 | case IPPROTO_ROUTING: { | |
1256 | struct ipv6_opt_hdr *hp = (void *)(skb->data + off); | |
1257 | ||
1258 | header_size = skb->network_header + | |
1259 | off + | |
1260 | sizeof(struct ipv6_opt_hdr); | |
1261 | maybe_pull_tail(skb, header_size); | |
1262 | ||
1263 | nexthdr = hp->nexthdr; | |
1264 | off += ipv6_optlen(hp); | |
1265 | break; | |
1266 | } | |
1267 | case IPPROTO_AH: { | |
1268 | struct ip_auth_hdr *hp = (void *)(skb->data + off); | |
1269 | ||
1270 | header_size = skb->network_header + | |
1271 | off + | |
1272 | sizeof(struct ip_auth_hdr); | |
1273 | maybe_pull_tail(skb, header_size); | |
1274 | ||
1275 | nexthdr = hp->nexthdr; | |
1276 | off += (hp->hdrlen+2)<<2; | |
1277 | break; | |
1278 | } | |
1279 | case IPPROTO_FRAGMENT: | |
1280 | fragment = true; | |
1281 | /* fall through */ | |
1282 | default: | |
1283 | done = true; | |
1284 | break; | |
1285 | } | |
1286 | } | |
1287 | ||
1288 | if (!done) { | |
1289 | if (net_ratelimit()) | |
1290 | netdev_err(vif->dev, "Failed to parse packet header\n"); | |
1291 | goto out; | |
1292 | } | |
1293 | ||
1294 | if (fragment) { | |
1295 | if (net_ratelimit()) | |
1296 | netdev_err(vif->dev, "Packet is a fragment!\n"); | |
1297 | goto out; | |
1298 | } | |
1299 | ||
1300 | switch (nexthdr) { | |
1301 | case IPPROTO_TCP: | |
1302 | if (!skb_partial_csum_set(skb, off, | |
1303 | offsetof(struct tcphdr, check))) | |
1304 | goto out; | |
1305 | ||
1306 | if (recalculate_partial_csum) { | |
1307 | struct tcphdr *tcph = tcp_hdr(skb); | |
1308 | ||
1309 | header_size = skb->network_header + | |
1310 | off + | |
1311 | sizeof(struct tcphdr); | |
1312 | maybe_pull_tail(skb, header_size); | |
1313 | ||
1314 | tcph->check = ~csum_ipv6_magic(&ipv6h->saddr, | |
1315 | &ipv6h->daddr, | |
1316 | skb->len - off, | |
1317 | IPPROTO_TCP, 0); | |
1318 | } | |
1319 | break; | |
1320 | case IPPROTO_UDP: | |
1321 | if (!skb_partial_csum_set(skb, off, | |
1322 | offsetof(struct udphdr, check))) | |
1323 | goto out; | |
1324 | ||
1325 | if (recalculate_partial_csum) { | |
1326 | struct udphdr *udph = udp_hdr(skb); | |
1327 | ||
1328 | header_size = skb->network_header + | |
1329 | off + | |
1330 | sizeof(struct udphdr); | |
1331 | maybe_pull_tail(skb, header_size); | |
1332 | ||
1333 | udph->check = ~csum_ipv6_magic(&ipv6h->saddr, | |
1334 | &ipv6h->daddr, | |
1335 | skb->len - off, | |
1336 | IPPROTO_UDP, 0); | |
1337 | } | |
1338 | break; | |
1339 | default: | |
1340 | if (net_ratelimit()) | |
1341 | netdev_err(vif->dev, | |
1342 | "Attempting to checksum a non-TCP/UDP packet, " | |
1343 | "dropping a protocol %d packet\n", | |
1344 | nexthdr); | |
1345 | goto out; | |
1346 | } | |
1347 | ||
1348 | err = 0; | |
1349 | ||
1350 | out: | |
1351 | return err; | |
1352 | } | |
1353 | ||
1354 | static int checksum_setup(struct xenvif *vif, struct sk_buff *skb) | |
1355 | { | |
1356 | int err = -EPROTO; | |
1357 | int recalculate_partial_csum = 0; | |
1358 | ||
1359 | /* A GSO SKB must be CHECKSUM_PARTIAL. However some buggy | |
1360 | * peers can fail to set NETRXF_csum_blank when sending a GSO | |
1361 | * frame. In this case force the SKB to CHECKSUM_PARTIAL and | |
1362 | * recalculate the partial checksum. | |
1363 | */ | |
1364 | if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) { | |
1365 | vif->rx_gso_checksum_fixup++; | |
1366 | skb->ip_summed = CHECKSUM_PARTIAL; | |
1367 | recalculate_partial_csum = 1; | |
1368 | } | |
1369 | ||
1370 | /* A non-CHECKSUM_PARTIAL SKB does not require setup. */ | |
1371 | if (skb->ip_summed != CHECKSUM_PARTIAL) | |
1372 | return 0; | |
1373 | ||
1374 | if (skb->protocol == htons(ETH_P_IP)) | |
1375 | err = checksum_setup_ip(vif, skb, recalculate_partial_csum); | |
1376 | else if (skb->protocol == htons(ETH_P_IPV6)) | |
1377 | err = checksum_setup_ipv6(vif, skb, recalculate_partial_csum); | |
1378 | ||
1379 | return err; | |
1380 | } | |
1381 | ||
f942dc25 IC |
1382 | static bool tx_credit_exceeded(struct xenvif *vif, unsigned size) |
1383 | { | |
059dfa6a WL |
1384 | u64 now = get_jiffies_64(); |
1385 | u64 next_credit = vif->credit_window_start + | |
f942dc25 IC |
1386 | msecs_to_jiffies(vif->credit_usec / 1000); |
1387 | ||
1388 | /* Timer could already be pending in rare cases. */ | |
1389 | if (timer_pending(&vif->credit_timeout)) | |
1390 | return true; | |
1391 | ||
1392 | /* Passed the point where we can replenish credit? */ | |
059dfa6a WL |
1393 | if (time_after_eq64(now, next_credit)) { |
1394 | vif->credit_window_start = now; | |
f942dc25 IC |
1395 | tx_add_credit(vif); |
1396 | } | |
1397 | ||
1398 | /* Still too big to send right now? Set a callback. */ | |
1399 | if (size > vif->remaining_credit) { | |
1400 | vif->credit_timeout.data = | |
1401 | (unsigned long)vif; | |
1402 | vif->credit_timeout.function = | |
1403 | tx_credit_callback; | |
1404 | mod_timer(&vif->credit_timeout, | |
1405 | next_credit); | |
059dfa6a | 1406 | vif->credit_window_start = next_credit; |
f942dc25 IC |
1407 | |
1408 | return true; | |
1409 | } | |
1410 | ||
1411 | return false; | |
1412 | } | |
1413 | ||
7376419a | 1414 | static unsigned xenvif_tx_build_gops(struct xenvif *vif) |
f942dc25 | 1415 | { |
b3f980bd | 1416 | struct gnttab_copy *gop = vif->tx_copy_ops, *request_gop; |
f942dc25 IC |
1417 | struct sk_buff *skb; |
1418 | int ret; | |
1419 | ||
b3f980bd WL |
1420 | while ((nr_pending_reqs(vif) + XEN_NETBK_LEGACY_SLOTS_MAX |
1421 | < MAX_PENDING_REQS)) { | |
f942dc25 | 1422 | struct xen_netif_tx_request txreq; |
37641494 | 1423 | struct xen_netif_tx_request txfrags[XEN_NETBK_LEGACY_SLOTS_MAX]; |
f942dc25 IC |
1424 | struct page *page; |
1425 | struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX-1]; | |
1426 | u16 pending_idx; | |
1427 | RING_IDX idx; | |
1428 | int work_to_do; | |
1429 | unsigned int data_len; | |
1430 | pending_ring_idx_t index; | |
1431 | ||
48856286 IC |
1432 | if (vif->tx.sring->req_prod - vif->tx.req_cons > |
1433 | XEN_NETIF_TX_RING_SIZE) { | |
1434 | netdev_err(vif->dev, | |
1435 | "Impossible number of requests. " | |
1436 | "req_prod %d, req_cons %d, size %ld\n", | |
1437 | vif->tx.sring->req_prod, vif->tx.req_cons, | |
1438 | XEN_NETIF_TX_RING_SIZE); | |
7376419a | 1439 | xenvif_fatal_tx_err(vif); |
48856286 IC |
1440 | continue; |
1441 | } | |
1442 | ||
f942dc25 | 1443 | RING_FINAL_CHECK_FOR_REQUESTS(&vif->tx, work_to_do); |
b3f980bd WL |
1444 | if (!work_to_do) |
1445 | break; | |
f942dc25 IC |
1446 | |
1447 | idx = vif->tx.req_cons; | |
1448 | rmb(); /* Ensure that we see the request before we copy it. */ | |
1449 | memcpy(&txreq, RING_GET_REQUEST(&vif->tx, idx), sizeof(txreq)); | |
1450 | ||
1451 | /* Credit-based scheduling. */ | |
1452 | if (txreq.size > vif->remaining_credit && | |
b3f980bd WL |
1453 | tx_credit_exceeded(vif, txreq.size)) |
1454 | break; | |
f942dc25 IC |
1455 | |
1456 | vif->remaining_credit -= txreq.size; | |
1457 | ||
1458 | work_to_do--; | |
1459 | vif->tx.req_cons = ++idx; | |
1460 | ||
1461 | memset(extras, 0, sizeof(extras)); | |
1462 | if (txreq.flags & XEN_NETTXF_extra_info) { | |
7376419a WL |
1463 | work_to_do = xenvif_get_extras(vif, extras, |
1464 | work_to_do); | |
f942dc25 | 1465 | idx = vif->tx.req_cons; |
48856286 | 1466 | if (unlikely(work_to_do < 0)) |
b3f980bd | 1467 | break; |
f942dc25 IC |
1468 | } |
1469 | ||
7376419a | 1470 | ret = xenvif_count_requests(vif, &txreq, txfrags, work_to_do); |
48856286 | 1471 | if (unlikely(ret < 0)) |
b3f980bd | 1472 | break; |
48856286 | 1473 | |
f942dc25 IC |
1474 | idx += ret; |
1475 | ||
1476 | if (unlikely(txreq.size < ETH_HLEN)) { | |
1477 | netdev_dbg(vif->dev, | |
1478 | "Bad packet size: %d\n", txreq.size); | |
7376419a | 1479 | xenvif_tx_err(vif, &txreq, idx); |
b3f980bd | 1480 | break; |
f942dc25 IC |
1481 | } |
1482 | ||
1483 | /* No crossing a page as the payload mustn't fragment. */ | |
1484 | if (unlikely((txreq.offset + txreq.size) > PAGE_SIZE)) { | |
48856286 | 1485 | netdev_err(vif->dev, |
f942dc25 IC |
1486 | "txreq.offset: %x, size: %u, end: %lu\n", |
1487 | txreq.offset, txreq.size, | |
1488 | (txreq.offset&~PAGE_MASK) + txreq.size); | |
7376419a | 1489 | xenvif_fatal_tx_err(vif); |
b3f980bd | 1490 | break; |
f942dc25 IC |
1491 | } |
1492 | ||
b3f980bd WL |
1493 | index = pending_index(vif->pending_cons); |
1494 | pending_idx = vif->pending_ring[index]; | |
f942dc25 IC |
1495 | |
1496 | data_len = (txreq.size > PKT_PROT_LEN && | |
37641494 | 1497 | ret < XEN_NETBK_LEGACY_SLOTS_MAX) ? |
f942dc25 IC |
1498 | PKT_PROT_LEN : txreq.size; |
1499 | ||
1500 | skb = alloc_skb(data_len + NET_SKB_PAD + NET_IP_ALIGN, | |
1501 | GFP_ATOMIC | __GFP_NOWARN); | |
1502 | if (unlikely(skb == NULL)) { | |
1503 | netdev_dbg(vif->dev, | |
1504 | "Can't allocate a skb in start_xmit.\n"); | |
7376419a | 1505 | xenvif_tx_err(vif, &txreq, idx); |
f942dc25 IC |
1506 | break; |
1507 | } | |
1508 | ||
1509 | /* Packets passed to netif_rx() must have some headroom. */ | |
1510 | skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN); | |
1511 | ||
1512 | if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) { | |
1513 | struct xen_netif_extra_info *gso; | |
1514 | gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1]; | |
1515 | ||
7376419a WL |
1516 | if (xenvif_set_skb_gso(vif, skb, gso)) { |
1517 | /* Failure in xenvif_set_skb_gso is fatal. */ | |
f942dc25 | 1518 | kfree_skb(skb); |
b3f980bd | 1519 | break; |
f942dc25 IC |
1520 | } |
1521 | } | |
1522 | ||
1523 | /* XXX could copy straight to head */ | |
7376419a | 1524 | page = xenvif_alloc_page(vif, pending_idx); |
f942dc25 IC |
1525 | if (!page) { |
1526 | kfree_skb(skb); | |
7376419a | 1527 | xenvif_tx_err(vif, &txreq, idx); |
b3f980bd | 1528 | break; |
f942dc25 IC |
1529 | } |
1530 | ||
f942dc25 IC |
1531 | gop->source.u.ref = txreq.gref; |
1532 | gop->source.domid = vif->domid; | |
1533 | gop->source.offset = txreq.offset; | |
1534 | ||
1535 | gop->dest.u.gmfn = virt_to_mfn(page_address(page)); | |
1536 | gop->dest.domid = DOMID_SELF; | |
1537 | gop->dest.offset = txreq.offset; | |
1538 | ||
1539 | gop->len = txreq.size; | |
1540 | gop->flags = GNTCOPY_source_gref; | |
1541 | ||
1542 | gop++; | |
1543 | ||
b3f980bd | 1544 | memcpy(&vif->pending_tx_info[pending_idx].req, |
f942dc25 | 1545 | &txreq, sizeof(txreq)); |
b3f980bd | 1546 | vif->pending_tx_info[pending_idx].head = index; |
f942dc25 IC |
1547 | *((u16 *)skb->data) = pending_idx; |
1548 | ||
1549 | __skb_put(skb, data_len); | |
1550 | ||
1551 | skb_shinfo(skb)->nr_frags = ret; | |
1552 | if (data_len < txreq.size) { | |
1553 | skb_shinfo(skb)->nr_frags++; | |
ea066ad1 IC |
1554 | frag_set_pending_idx(&skb_shinfo(skb)->frags[0], |
1555 | pending_idx); | |
f942dc25 | 1556 | } else { |
ea066ad1 IC |
1557 | frag_set_pending_idx(&skb_shinfo(skb)->frags[0], |
1558 | INVALID_PENDING_IDX); | |
f942dc25 IC |
1559 | } |
1560 | ||
b3f980bd | 1561 | vif->pending_cons++; |
f942dc25 | 1562 | |
7376419a | 1563 | request_gop = xenvif_get_requests(vif, skb, txfrags, gop); |
f942dc25 IC |
1564 | if (request_gop == NULL) { |
1565 | kfree_skb(skb); | |
7376419a | 1566 | xenvif_tx_err(vif, &txreq, idx); |
b3f980bd | 1567 | break; |
f942dc25 IC |
1568 | } |
1569 | gop = request_gop; | |
1570 | ||
b3f980bd | 1571 | __skb_queue_tail(&vif->tx_queue, skb); |
1e0b6eac | 1572 | |
f942dc25 | 1573 | vif->tx.req_cons = idx; |
f942dc25 | 1574 | |
b3f980bd | 1575 | if ((gop-vif->tx_copy_ops) >= ARRAY_SIZE(vif->tx_copy_ops)) |
f942dc25 IC |
1576 | break; |
1577 | } | |
1578 | ||
b3f980bd | 1579 | return gop - vif->tx_copy_ops; |
f942dc25 IC |
1580 | } |
1581 | ||
b3f980bd | 1582 | |
7376419a | 1583 | static int xenvif_tx_submit(struct xenvif *vif, int budget) |
f942dc25 | 1584 | { |
b3f980bd | 1585 | struct gnttab_copy *gop = vif->tx_copy_ops; |
f942dc25 | 1586 | struct sk_buff *skb; |
b3f980bd | 1587 | int work_done = 0; |
f942dc25 | 1588 | |
b3f980bd WL |
1589 | while (work_done < budget && |
1590 | (skb = __skb_dequeue(&vif->tx_queue)) != NULL) { | |
f942dc25 | 1591 | struct xen_netif_tx_request *txp; |
f942dc25 IC |
1592 | u16 pending_idx; |
1593 | unsigned data_len; | |
1594 | ||
1595 | pending_idx = *((u16 *)skb->data); | |
b3f980bd | 1596 | txp = &vif->pending_tx_info[pending_idx].req; |
f942dc25 IC |
1597 | |
1598 | /* Check the remap error code. */ | |
7376419a | 1599 | if (unlikely(xenvif_tx_check_gop(vif, skb, &gop))) { |
f942dc25 IC |
1600 | netdev_dbg(vif->dev, "netback grant failed.\n"); |
1601 | skb_shinfo(skb)->nr_frags = 0; | |
1602 | kfree_skb(skb); | |
1603 | continue; | |
1604 | } | |
1605 | ||
1606 | data_len = skb->len; | |
1607 | memcpy(skb->data, | |
b3f980bd | 1608 | (void *)(idx_to_kaddr(vif, pending_idx)|txp->offset), |
f942dc25 IC |
1609 | data_len); |
1610 | if (data_len < txp->size) { | |
1611 | /* Append the packet payload as a fragment. */ | |
1612 | txp->offset += data_len; | |
1613 | txp->size -= data_len; | |
1614 | } else { | |
1615 | /* Schedule a response immediately. */ | |
7376419a WL |
1616 | xenvif_idx_release(vif, pending_idx, |
1617 | XEN_NETIF_RSP_OKAY); | |
f942dc25 IC |
1618 | } |
1619 | ||
1620 | if (txp->flags & XEN_NETTXF_csum_blank) | |
1621 | skb->ip_summed = CHECKSUM_PARTIAL; | |
1622 | else if (txp->flags & XEN_NETTXF_data_validated) | |
1623 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
1624 | ||
7376419a | 1625 | xenvif_fill_frags(vif, skb); |
f942dc25 | 1626 | |
2eba61d5 | 1627 | if (skb_is_nonlinear(skb) && skb_headlen(skb) < PKT_PROT_LEN) { |
f942dc25 IC |
1628 | int target = min_t(int, skb->len, PKT_PROT_LEN); |
1629 | __pskb_pull_tail(skb, target - skb_headlen(skb)); | |
1630 | } | |
1631 | ||
1632 | skb->dev = vif->dev; | |
1633 | skb->protocol = eth_type_trans(skb, skb->dev); | |
f9ca8f74 | 1634 | skb_reset_network_header(skb); |
f942dc25 IC |
1635 | |
1636 | if (checksum_setup(vif, skb)) { | |
1637 | netdev_dbg(vif->dev, | |
1638 | "Can't setup checksum in net_tx_action\n"); | |
1639 | kfree_skb(skb); | |
1640 | continue; | |
1641 | } | |
1642 | ||
40893fd0 | 1643 | skb_probe_transport_header(skb, 0); |
f9ca8f74 | 1644 | |
f942dc25 IC |
1645 | vif->dev->stats.rx_bytes += skb->len; |
1646 | vif->dev->stats.rx_packets++; | |
1647 | ||
b3f980bd WL |
1648 | work_done++; |
1649 | ||
1650 | netif_receive_skb(skb); | |
f942dc25 | 1651 | } |
b3f980bd WL |
1652 | |
1653 | return work_done; | |
f942dc25 IC |
1654 | } |
1655 | ||
1656 | /* Called after netfront has transmitted */ | |
7376419a | 1657 | int xenvif_tx_action(struct xenvif *vif, int budget) |
f942dc25 IC |
1658 | { |
1659 | unsigned nr_gops; | |
b3f980bd | 1660 | int work_done; |
f942dc25 | 1661 | |
b3f980bd WL |
1662 | if (unlikely(!tx_work_todo(vif))) |
1663 | return 0; | |
1664 | ||
7376419a | 1665 | nr_gops = xenvif_tx_build_gops(vif); |
f942dc25 IC |
1666 | |
1667 | if (nr_gops == 0) | |
b3f980bd WL |
1668 | return 0; |
1669 | ||
1670 | gnttab_batch_copy(vif->tx_copy_ops, nr_gops); | |
f942dc25 | 1671 | |
7376419a | 1672 | work_done = xenvif_tx_submit(vif, nr_gops); |
f942dc25 | 1673 | |
b3f980bd | 1674 | return work_done; |
f942dc25 IC |
1675 | } |
1676 | ||
7376419a WL |
1677 | static void xenvif_idx_release(struct xenvif *vif, u16 pending_idx, |
1678 | u8 status) | |
f942dc25 | 1679 | { |
f942dc25 | 1680 | struct pending_tx_info *pending_tx_info; |
2810e5b9 WL |
1681 | pending_ring_idx_t head; |
1682 | u16 peek; /* peek into next tx request */ | |
1683 | ||
b3f980bd | 1684 | BUG_ON(vif->mmap_pages[pending_idx] == (void *)(~0UL)); |
f942dc25 IC |
1685 | |
1686 | /* Already complete? */ | |
b3f980bd | 1687 | if (vif->mmap_pages[pending_idx] == NULL) |
f942dc25 IC |
1688 | return; |
1689 | ||
b3f980bd | 1690 | pending_tx_info = &vif->pending_tx_info[pending_idx]; |
f942dc25 | 1691 | |
2810e5b9 | 1692 | head = pending_tx_info->head; |
f942dc25 | 1693 | |
b3f980bd WL |
1694 | BUG_ON(!pending_tx_is_head(vif, head)); |
1695 | BUG_ON(vif->pending_ring[pending_index(head)] != pending_idx); | |
f942dc25 | 1696 | |
2810e5b9 WL |
1697 | do { |
1698 | pending_ring_idx_t index; | |
1699 | pending_ring_idx_t idx = pending_index(head); | |
b3f980bd | 1700 | u16 info_idx = vif->pending_ring[idx]; |
f942dc25 | 1701 | |
b3f980bd | 1702 | pending_tx_info = &vif->pending_tx_info[info_idx]; |
2810e5b9 WL |
1703 | make_tx_response(vif, &pending_tx_info->req, status); |
1704 | ||
1705 | /* Setting any number other than | |
1706 | * INVALID_PENDING_RING_IDX indicates this slot is | |
1707 | * starting a new packet / ending a previous packet. | |
1708 | */ | |
1709 | pending_tx_info->head = 0; | |
1710 | ||
b3f980bd WL |
1711 | index = pending_index(vif->pending_prod++); |
1712 | vif->pending_ring[index] = vif->pending_ring[info_idx]; | |
f942dc25 | 1713 | |
b3f980bd | 1714 | peek = vif->pending_ring[pending_index(++head)]; |
2810e5b9 | 1715 | |
b3f980bd | 1716 | } while (!pending_tx_is_head(vif, peek)); |
2810e5b9 | 1717 | |
b3f980bd WL |
1718 | put_page(vif->mmap_pages[pending_idx]); |
1719 | vif->mmap_pages[pending_idx] = NULL; | |
f942dc25 IC |
1720 | } |
1721 | ||
2810e5b9 | 1722 | |
f942dc25 IC |
1723 | static void make_tx_response(struct xenvif *vif, |
1724 | struct xen_netif_tx_request *txp, | |
1725 | s8 st) | |
1726 | { | |
1727 | RING_IDX i = vif->tx.rsp_prod_pvt; | |
1728 | struct xen_netif_tx_response *resp; | |
1729 | int notify; | |
1730 | ||
1731 | resp = RING_GET_RESPONSE(&vif->tx, i); | |
1732 | resp->id = txp->id; | |
1733 | resp->status = st; | |
1734 | ||
1735 | if (txp->flags & XEN_NETTXF_extra_info) | |
1736 | RING_GET_RESPONSE(&vif->tx, ++i)->status = XEN_NETIF_RSP_NULL; | |
1737 | ||
1738 | vif->tx.rsp_prod_pvt = ++i; | |
1739 | RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&vif->tx, notify); | |
1740 | if (notify) | |
e1f00a69 | 1741 | notify_remote_via_irq(vif->tx_irq); |
f942dc25 IC |
1742 | } |
1743 | ||
1744 | static struct xen_netif_rx_response *make_rx_response(struct xenvif *vif, | |
1745 | u16 id, | |
1746 | s8 st, | |
1747 | u16 offset, | |
1748 | u16 size, | |
1749 | u16 flags) | |
1750 | { | |
1751 | RING_IDX i = vif->rx.rsp_prod_pvt; | |
1752 | struct xen_netif_rx_response *resp; | |
1753 | ||
1754 | resp = RING_GET_RESPONSE(&vif->rx, i); | |
1755 | resp->offset = offset; | |
1756 | resp->flags = flags; | |
1757 | resp->id = id; | |
1758 | resp->status = (s16)size; | |
1759 | if (st < 0) | |
1760 | resp->status = (s16)st; | |
1761 | ||
1762 | vif->rx.rsp_prod_pvt = ++i; | |
1763 | ||
1764 | return resp; | |
1765 | } | |
1766 | ||
b3f980bd | 1767 | static inline int rx_work_todo(struct xenvif *vif) |
f942dc25 | 1768 | { |
b3f980bd | 1769 | return !skb_queue_empty(&vif->rx_queue); |
f942dc25 IC |
1770 | } |
1771 | ||
b3f980bd | 1772 | static inline int tx_work_todo(struct xenvif *vif) |
f942dc25 IC |
1773 | { |
1774 | ||
b3f980bd WL |
1775 | if (likely(RING_HAS_UNCONSUMED_REQUESTS(&vif->tx)) && |
1776 | (nr_pending_reqs(vif) + XEN_NETBK_LEGACY_SLOTS_MAX | |
1777 | < MAX_PENDING_REQS)) | |
f942dc25 IC |
1778 | return 1; |
1779 | ||
1780 | return 0; | |
1781 | } | |
1782 | ||
7376419a | 1783 | void xenvif_unmap_frontend_rings(struct xenvif *vif) |
f942dc25 | 1784 | { |
c9d63699 DV |
1785 | if (vif->tx.sring) |
1786 | xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif), | |
1787 | vif->tx.sring); | |
1788 | if (vif->rx.sring) | |
1789 | xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif), | |
1790 | vif->rx.sring); | |
f942dc25 IC |
1791 | } |
1792 | ||
7376419a WL |
1793 | int xenvif_map_frontend_rings(struct xenvif *vif, |
1794 | grant_ref_t tx_ring_ref, | |
1795 | grant_ref_t rx_ring_ref) | |
f942dc25 | 1796 | { |
c9d63699 | 1797 | void *addr; |
f942dc25 IC |
1798 | struct xen_netif_tx_sring *txs; |
1799 | struct xen_netif_rx_sring *rxs; | |
1800 | ||
1801 | int err = -ENOMEM; | |
1802 | ||
c9d63699 DV |
1803 | err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(vif), |
1804 | tx_ring_ref, &addr); | |
1805 | if (err) | |
f942dc25 IC |
1806 | goto err; |
1807 | ||
c9d63699 | 1808 | txs = (struct xen_netif_tx_sring *)addr; |
f942dc25 IC |
1809 | BACK_RING_INIT(&vif->tx, txs, PAGE_SIZE); |
1810 | ||
c9d63699 DV |
1811 | err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(vif), |
1812 | rx_ring_ref, &addr); | |
1813 | if (err) | |
f942dc25 | 1814 | goto err; |
f942dc25 | 1815 | |
c9d63699 | 1816 | rxs = (struct xen_netif_rx_sring *)addr; |
f942dc25 IC |
1817 | BACK_RING_INIT(&vif->rx, rxs, PAGE_SIZE); |
1818 | ||
c9d63699 DV |
1819 | vif->rx_req_cons_peek = 0; |
1820 | ||
f942dc25 IC |
1821 | return 0; |
1822 | ||
1823 | err: | |
7376419a | 1824 | xenvif_unmap_frontend_rings(vif); |
f942dc25 IC |
1825 | return err; |
1826 | } | |
1827 | ||
7376419a | 1828 | int xenvif_kthread(void *data) |
b3f980bd WL |
1829 | { |
1830 | struct xenvif *vif = data; | |
1831 | ||
1832 | while (!kthread_should_stop()) { | |
1833 | wait_event_interruptible(vif->wq, | |
1834 | rx_work_todo(vif) || | |
1835 | kthread_should_stop()); | |
1836 | if (kthread_should_stop()) | |
1837 | break; | |
1838 | ||
1839 | if (rx_work_todo(vif)) | |
7376419a | 1840 | xenvif_rx_action(vif); |
b3f980bd WL |
1841 | |
1842 | cond_resched(); | |
1843 | } | |
1844 | ||
1845 | return 0; | |
1846 | } | |
1847 | ||
f942dc25 IC |
1848 | static int __init netback_init(void) |
1849 | { | |
f942dc25 | 1850 | int rc = 0; |
f942dc25 | 1851 | |
2a14b244 | 1852 | if (!xen_domain()) |
f942dc25 IC |
1853 | return -ENODEV; |
1854 | ||
37641494 | 1855 | if (fatal_skb_slots < XEN_NETBK_LEGACY_SLOTS_MAX) { |
383eda32 JP |
1856 | pr_info("fatal_skb_slots too small (%d), bump it to XEN_NETBK_LEGACY_SLOTS_MAX (%d)\n", |
1857 | fatal_skb_slots, XEN_NETBK_LEGACY_SLOTS_MAX); | |
37641494 | 1858 | fatal_skb_slots = XEN_NETBK_LEGACY_SLOTS_MAX; |
2810e5b9 WL |
1859 | } |
1860 | ||
f942dc25 IC |
1861 | rc = xenvif_xenbus_init(); |
1862 | if (rc) | |
1863 | goto failed_init; | |
1864 | ||
1865 | return 0; | |
1866 | ||
1867 | failed_init: | |
f942dc25 | 1868 | return rc; |
f942dc25 IC |
1869 | } |
1870 | ||
1871 | module_init(netback_init); | |
1872 | ||
b103f358 WL |
1873 | static void __exit netback_fini(void) |
1874 | { | |
b103f358 | 1875 | xenvif_xenbus_fini(); |
b103f358 WL |
1876 | } |
1877 | module_exit(netback_fini); | |
1878 | ||
f942dc25 | 1879 | MODULE_LICENSE("Dual BSD/GPL"); |
f984cec6 | 1880 | MODULE_ALIAS("xen-backend:vif"); |