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 | ||
e1f00a69 WL |
50 | /* Provide an option to disable split event channels at load time as |
51 | * event channels are limited resource. Split event channels are | |
52 | * enabled by default. | |
53 | */ | |
54 | bool separate_tx_rx_irq = 1; | |
55 | module_param(separate_tx_rx_irq, bool, 0644); | |
56 | ||
2810e5b9 WL |
57 | /* |
58 | * This is the maximum slots a skb can have. If a guest sends a skb | |
59 | * which exceeds this limit it is considered malicious. | |
60 | */ | |
37641494 WL |
61 | #define FATAL_SKB_SLOTS_DEFAULT 20 |
62 | static unsigned int fatal_skb_slots = FATAL_SKB_SLOTS_DEFAULT; | |
63 | module_param(fatal_skb_slots, uint, 0444); | |
64 | ||
65 | /* | |
66 | * To avoid confusion, we define XEN_NETBK_LEGACY_SLOTS_MAX indicating | |
67 | * the maximum slots a valid packet can use. Now this value is defined | |
68 | * to be XEN_NETIF_NR_SLOTS_MIN, which is supposed to be supported by | |
69 | * all backend. | |
70 | */ | |
71 | #define XEN_NETBK_LEGACY_SLOTS_MAX XEN_NETIF_NR_SLOTS_MIN | |
2810e5b9 | 72 | |
2810e5b9 WL |
73 | /* |
74 | * If head != INVALID_PENDING_RING_IDX, it means this tx request is head of | |
75 | * one or more merged tx requests, otherwise it is the continuation of | |
76 | * previous tx request. | |
77 | */ | |
b3f980bd | 78 | static inline int pending_tx_is_head(struct xenvif *vif, RING_IDX idx) |
f942dc25 | 79 | { |
b3f980bd | 80 | return vif->pending_tx_info[idx].head != INVALID_PENDING_RING_IDX; |
f942dc25 IC |
81 | } |
82 | ||
7376419a WL |
83 | static void xenvif_idx_release(struct xenvif *vif, u16 pending_idx, |
84 | u8 status); | |
85 | ||
f942dc25 IC |
86 | static void make_tx_response(struct xenvif *vif, |
87 | struct xen_netif_tx_request *txp, | |
88 | s8 st); | |
b3f980bd WL |
89 | |
90 | static inline int tx_work_todo(struct xenvif *vif); | |
91 | static inline int rx_work_todo(struct xenvif *vif); | |
92 | ||
f942dc25 IC |
93 | static struct xen_netif_rx_response *make_rx_response(struct xenvif *vif, |
94 | u16 id, | |
95 | s8 st, | |
96 | u16 offset, | |
97 | u16 size, | |
98 | u16 flags); | |
99 | ||
b3f980bd | 100 | static inline unsigned long idx_to_pfn(struct xenvif *vif, |
ea066ad1 | 101 | u16 idx) |
f942dc25 | 102 | { |
b3f980bd | 103 | return page_to_pfn(vif->mmap_pages[idx]); |
f942dc25 IC |
104 | } |
105 | ||
b3f980bd | 106 | static inline unsigned long idx_to_kaddr(struct xenvif *vif, |
ea066ad1 | 107 | u16 idx) |
f942dc25 | 108 | { |
b3f980bd | 109 | return (unsigned long)pfn_to_kaddr(idx_to_pfn(vif, idx)); |
f942dc25 IC |
110 | } |
111 | ||
2eba61d5 PD |
112 | /* This is a miniumum size for the linear area to avoid lots of |
113 | * calls to __pskb_pull_tail() as we set up checksum offsets. The | |
114 | * value 128 was chosen as it covers all IPv4 and most likely | |
115 | * IPv6 headers. | |
f942dc25 | 116 | */ |
2eba61d5 | 117 | #define PKT_PROT_LEN 128 |
f942dc25 | 118 | |
ea066ad1 IC |
119 | static u16 frag_get_pending_idx(skb_frag_t *frag) |
120 | { | |
121 | return (u16)frag->page_offset; | |
122 | } | |
123 | ||
124 | static void frag_set_pending_idx(skb_frag_t *frag, u16 pending_idx) | |
125 | { | |
126 | frag->page_offset = pending_idx; | |
127 | } | |
128 | ||
f942dc25 IC |
129 | static inline pending_ring_idx_t pending_index(unsigned i) |
130 | { | |
131 | return i & (MAX_PENDING_REQS-1); | |
132 | } | |
133 | ||
b3f980bd | 134 | static inline pending_ring_idx_t nr_pending_reqs(struct xenvif *vif) |
f942dc25 IC |
135 | { |
136 | return MAX_PENDING_REQS - | |
b3f980bd | 137 | vif->pending_prod + vif->pending_cons; |
f942dc25 IC |
138 | } |
139 | ||
140 | static int max_required_rx_slots(struct xenvif *vif) | |
141 | { | |
142 | int max = DIV_ROUND_UP(vif->dev->mtu, PAGE_SIZE); | |
143 | ||
2810e5b9 | 144 | /* XXX FIXME: RX path dependent on MAX_SKB_FRAGS */ |
f942dc25 IC |
145 | if (vif->can_sg || vif->gso || vif->gso_prefix) |
146 | max += MAX_SKB_FRAGS + 1; /* extra_info + frags */ | |
147 | ||
148 | return max; | |
149 | } | |
150 | ||
7376419a | 151 | int xenvif_rx_ring_full(struct xenvif *vif) |
f942dc25 IC |
152 | { |
153 | RING_IDX peek = vif->rx_req_cons_peek; | |
154 | RING_IDX needed = max_required_rx_slots(vif); | |
155 | ||
156 | return ((vif->rx.sring->req_prod - peek) < needed) || | |
157 | ((vif->rx.rsp_prod_pvt + XEN_NETIF_RX_RING_SIZE - peek) < needed); | |
158 | } | |
159 | ||
7376419a | 160 | int xenvif_must_stop_queue(struct xenvif *vif) |
f942dc25 | 161 | { |
7376419a | 162 | if (!xenvif_rx_ring_full(vif)) |
f942dc25 IC |
163 | return 0; |
164 | ||
165 | vif->rx.sring->req_event = vif->rx_req_cons_peek + | |
166 | max_required_rx_slots(vif); | |
167 | mb(); /* request notification /then/ check the queue */ | |
168 | ||
7376419a | 169 | return xenvif_rx_ring_full(vif); |
f942dc25 IC |
170 | } |
171 | ||
172 | /* | |
173 | * Returns true if we should start a new receive buffer instead of | |
174 | * adding 'size' bytes to a buffer which currently contains 'offset' | |
175 | * bytes. | |
176 | */ | |
177 | static bool start_new_rx_buffer(int offset, unsigned long size, int head) | |
178 | { | |
179 | /* simple case: we have completely filled the current buffer. */ | |
180 | if (offset == MAX_BUFFER_OFFSET) | |
181 | return true; | |
182 | ||
183 | /* | |
184 | * complex case: start a fresh buffer if the current frag | |
185 | * would overflow the current buffer but only if: | |
186 | * (i) this frag would fit completely in the next buffer | |
187 | * and (ii) there is already some data in the current buffer | |
188 | * and (iii) this is not the head buffer. | |
189 | * | |
190 | * Where: | |
191 | * - (i) stops us splitting a frag into two copies | |
192 | * unless the frag is too large for a single buffer. | |
193 | * - (ii) stops us from leaving a buffer pointlessly empty. | |
194 | * - (iii) stops us leaving the first buffer | |
195 | * empty. Strictly speaking this is already covered | |
196 | * by (ii) but is explicitly checked because | |
197 | * netfront relies on the first buffer being | |
198 | * non-empty and can crash otherwise. | |
199 | * | |
200 | * This means we will effectively linearise small | |
201 | * frags but do not needlessly split large buffers | |
202 | * into multiple copies tend to give large frags their | |
203 | * own buffers as before. | |
204 | */ | |
205 | if ((offset + size > MAX_BUFFER_OFFSET) && | |
206 | (size <= MAX_BUFFER_OFFSET) && offset && !head) | |
207 | return true; | |
208 | ||
209 | return false; | |
210 | } | |
211 | ||
33bc801d WL |
212 | struct xenvif_count_slot_state { |
213 | unsigned long copy_off; | |
214 | bool head; | |
215 | }; | |
216 | ||
217 | unsigned int xenvif_count_frag_slots(struct xenvif *vif, | |
218 | unsigned long offset, unsigned long size, | |
219 | struct xenvif_count_slot_state *state) | |
220 | { | |
221 | unsigned count = 0; | |
222 | ||
223 | offset &= ~PAGE_MASK; | |
224 | ||
225 | while (size > 0) { | |
226 | unsigned long bytes; | |
227 | ||
228 | bytes = PAGE_SIZE - offset; | |
229 | ||
230 | if (bytes > size) | |
231 | bytes = size; | |
232 | ||
233 | if (start_new_rx_buffer(state->copy_off, bytes, state->head)) { | |
234 | count++; | |
235 | state->copy_off = 0; | |
236 | } | |
237 | ||
238 | if (state->copy_off + bytes > MAX_BUFFER_OFFSET) | |
239 | bytes = MAX_BUFFER_OFFSET - state->copy_off; | |
240 | ||
241 | state->copy_off += bytes; | |
242 | ||
243 | offset += bytes; | |
244 | size -= bytes; | |
245 | ||
246 | if (offset == PAGE_SIZE) | |
247 | offset = 0; | |
248 | ||
249 | state->head = false; | |
250 | } | |
251 | ||
252 | return count; | |
253 | } | |
254 | ||
f942dc25 IC |
255 | /* |
256 | * Figure out how many ring slots we're going to need to send @skb to | |
257 | * the guest. This function is essentially a dry run of | |
7376419a | 258 | * xenvif_gop_frag_copy. |
f942dc25 | 259 | */ |
7376419a | 260 | unsigned int xenvif_count_skb_slots(struct xenvif *vif, struct sk_buff *skb) |
f942dc25 | 261 | { |
33bc801d | 262 | struct xenvif_count_slot_state state; |
f942dc25 | 263 | unsigned int count; |
33bc801d WL |
264 | unsigned char *data; |
265 | unsigned i; | |
f942dc25 | 266 | |
33bc801d WL |
267 | state.head = true; |
268 | state.copy_off = 0; | |
f942dc25 | 269 | |
33bc801d WL |
270 | /* Slot for the first (partial) page of data. */ |
271 | count = 1; | |
f942dc25 | 272 | |
33bc801d | 273 | /* Need a slot for the GSO prefix for GSO extra data? */ |
f942dc25 IC |
274 | if (skb_shinfo(skb)->gso_size) |
275 | count++; | |
276 | ||
33bc801d WL |
277 | data = skb->data; |
278 | while (data < skb_tail_pointer(skb)) { | |
279 | unsigned long offset = offset_in_page(data); | |
280 | unsigned long size = PAGE_SIZE - offset; | |
4f0581d2 | 281 | |
33bc801d WL |
282 | if (data + size > skb_tail_pointer(skb)) |
283 | size = skb_tail_pointer(skb) - data; | |
6a8ed462 | 284 | |
33bc801d | 285 | count += xenvif_count_frag_slots(vif, offset, size, &state); |
4f0581d2 | 286 | |
33bc801d WL |
287 | data += size; |
288 | } | |
4f0581d2 | 289 | |
33bc801d WL |
290 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
291 | unsigned long size = skb_frag_size(&skb_shinfo(skb)->frags[i]); | |
292 | unsigned long offset = skb_shinfo(skb)->frags[i].page_offset; | |
4f0581d2 | 293 | |
33bc801d | 294 | count += xenvif_count_frag_slots(vif, offset, size, &state); |
f942dc25 IC |
295 | } |
296 | return count; | |
297 | } | |
298 | ||
299 | struct netrx_pending_operations { | |
300 | unsigned copy_prod, copy_cons; | |
301 | unsigned meta_prod, meta_cons; | |
302 | struct gnttab_copy *copy; | |
b3f980bd | 303 | struct xenvif_rx_meta *meta; |
f942dc25 IC |
304 | int copy_off; |
305 | grant_ref_t copy_gref; | |
306 | }; | |
307 | ||
b3f980bd WL |
308 | static struct xenvif_rx_meta *get_next_rx_buffer(struct xenvif *vif, |
309 | struct netrx_pending_operations *npo) | |
f942dc25 | 310 | { |
b3f980bd | 311 | struct xenvif_rx_meta *meta; |
f942dc25 IC |
312 | struct xen_netif_rx_request *req; |
313 | ||
314 | req = RING_GET_REQUEST(&vif->rx, vif->rx.req_cons++); | |
315 | ||
316 | meta = npo->meta + npo->meta_prod++; | |
317 | meta->gso_size = 0; | |
318 | meta->size = 0; | |
319 | meta->id = req->id; | |
320 | ||
321 | npo->copy_off = 0; | |
322 | npo->copy_gref = req->gref; | |
323 | ||
324 | return meta; | |
325 | } | |
326 | ||
33bc801d WL |
327 | /* |
328 | * Set up the grant operations for this fragment. If it's a flipping | |
329 | * interface, we also set up the unmap request from here. | |
330 | */ | |
7376419a WL |
331 | static void xenvif_gop_frag_copy(struct xenvif *vif, struct sk_buff *skb, |
332 | struct netrx_pending_operations *npo, | |
333 | struct page *page, unsigned long size, | |
33bc801d | 334 | unsigned long offset, int *head) |
f942dc25 IC |
335 | { |
336 | struct gnttab_copy *copy_gop; | |
b3f980bd | 337 | struct xenvif_rx_meta *meta; |
f942dc25 IC |
338 | unsigned long bytes; |
339 | ||
340 | /* Data must not cross a page boundary. */ | |
6a8ed462 | 341 | BUG_ON(size + offset > PAGE_SIZE<<compound_order(page)); |
f942dc25 IC |
342 | |
343 | meta = npo->meta + npo->meta_prod - 1; | |
344 | ||
6a8ed462 IC |
345 | /* Skip unused frames from start of page */ |
346 | page += offset >> PAGE_SHIFT; | |
347 | offset &= ~PAGE_MASK; | |
348 | ||
f942dc25 | 349 | while (size > 0) { |
6a8ed462 | 350 | BUG_ON(offset >= PAGE_SIZE); |
f942dc25 IC |
351 | BUG_ON(npo->copy_off > MAX_BUFFER_OFFSET); |
352 | ||
6a8ed462 IC |
353 | bytes = PAGE_SIZE - offset; |
354 | ||
355 | if (bytes > size) | |
356 | bytes = size; | |
357 | ||
33bc801d | 358 | if (start_new_rx_buffer(npo->copy_off, bytes, *head)) { |
f942dc25 IC |
359 | /* |
360 | * Netfront requires there to be some data in the head | |
361 | * buffer. | |
362 | */ | |
33bc801d | 363 | BUG_ON(*head); |
f942dc25 IC |
364 | |
365 | meta = get_next_rx_buffer(vif, npo); | |
366 | } | |
367 | ||
f942dc25 IC |
368 | if (npo->copy_off + bytes > MAX_BUFFER_OFFSET) |
369 | bytes = MAX_BUFFER_OFFSET - npo->copy_off; | |
370 | ||
371 | copy_gop = npo->copy + npo->copy_prod++; | |
372 | copy_gop->flags = GNTCOPY_dest_gref; | |
b3f980bd WL |
373 | copy_gop->len = bytes; |
374 | ||
43e9d194 WL |
375 | copy_gop->source.domid = DOMID_SELF; |
376 | copy_gop->source.u.gmfn = virt_to_mfn(page_address(page)); | |
f942dc25 | 377 | copy_gop->source.offset = offset; |
f942dc25 | 378 | |
b3f980bd | 379 | copy_gop->dest.domid = vif->domid; |
f942dc25 IC |
380 | copy_gop->dest.offset = npo->copy_off; |
381 | copy_gop->dest.u.ref = npo->copy_gref; | |
f942dc25 IC |
382 | |
383 | npo->copy_off += bytes; | |
384 | meta->size += bytes; | |
385 | ||
386 | offset += bytes; | |
387 | size -= bytes; | |
388 | ||
6a8ed462 IC |
389 | /* Next frame */ |
390 | if (offset == PAGE_SIZE && size) { | |
391 | BUG_ON(!PageCompound(page)); | |
392 | page++; | |
393 | offset = 0; | |
394 | } | |
395 | ||
f942dc25 | 396 | /* Leave a gap for the GSO descriptor. */ |
33bc801d | 397 | if (*head && skb_shinfo(skb)->gso_size && !vif->gso_prefix) |
f942dc25 IC |
398 | vif->rx.req_cons++; |
399 | ||
33bc801d | 400 | *head = 0; /* There must be something in this buffer now. */ |
f942dc25 IC |
401 | |
402 | } | |
403 | } | |
404 | ||
405 | /* | |
406 | * Prepare an SKB to be transmitted to the frontend. | |
407 | * | |
408 | * This function is responsible for allocating grant operations, meta | |
409 | * structures, etc. | |
410 | * | |
411 | * It returns the number of meta structures consumed. The number of | |
412 | * ring slots used is always equal to the number of meta slots used | |
413 | * plus the number of GSO descriptors used. Currently, we use either | |
414 | * zero GSO descriptors (for non-GSO packets) or one descriptor (for | |
415 | * frontend-side LRO). | |
416 | */ | |
7376419a WL |
417 | static int xenvif_gop_skb(struct sk_buff *skb, |
418 | struct netrx_pending_operations *npo) | |
f942dc25 IC |
419 | { |
420 | struct xenvif *vif = netdev_priv(skb->dev); | |
421 | int nr_frags = skb_shinfo(skb)->nr_frags; | |
422 | int i; | |
423 | struct xen_netif_rx_request *req; | |
b3f980bd | 424 | struct xenvif_rx_meta *meta; |
f942dc25 | 425 | unsigned char *data; |
33bc801d | 426 | int head = 1; |
f942dc25 IC |
427 | int old_meta_prod; |
428 | ||
429 | old_meta_prod = npo->meta_prod; | |
430 | ||
431 | /* Set up a GSO prefix descriptor, if necessary */ | |
432 | if (skb_shinfo(skb)->gso_size && vif->gso_prefix) { | |
433 | req = RING_GET_REQUEST(&vif->rx, vif->rx.req_cons++); | |
434 | meta = npo->meta + npo->meta_prod++; | |
435 | meta->gso_size = skb_shinfo(skb)->gso_size; | |
436 | meta->size = 0; | |
437 | meta->id = req->id; | |
438 | } | |
439 | ||
440 | req = RING_GET_REQUEST(&vif->rx, vif->rx.req_cons++); | |
441 | meta = npo->meta + npo->meta_prod++; | |
442 | ||
443 | if (!vif->gso_prefix) | |
444 | meta->gso_size = skb_shinfo(skb)->gso_size; | |
445 | else | |
446 | meta->gso_size = 0; | |
447 | ||
448 | meta->size = 0; | |
449 | meta->id = req->id; | |
450 | npo->copy_off = 0; | |
451 | npo->copy_gref = req->gref; | |
452 | ||
453 | data = skb->data; | |
454 | while (data < skb_tail_pointer(skb)) { | |
455 | unsigned int offset = offset_in_page(data); | |
456 | unsigned int len = PAGE_SIZE - offset; | |
457 | ||
458 | if (data + len > skb_tail_pointer(skb)) | |
459 | len = skb_tail_pointer(skb) - data; | |
460 | ||
7376419a | 461 | xenvif_gop_frag_copy(vif, skb, npo, |
33bc801d | 462 | virt_to_page(data), len, offset, &head); |
f942dc25 IC |
463 | data += len; |
464 | } | |
465 | ||
466 | for (i = 0; i < nr_frags; i++) { | |
7376419a WL |
467 | xenvif_gop_frag_copy(vif, skb, npo, |
468 | skb_frag_page(&skb_shinfo(skb)->frags[i]), | |
469 | skb_frag_size(&skb_shinfo(skb)->frags[i]), | |
470 | skb_shinfo(skb)->frags[i].page_offset, | |
33bc801d | 471 | &head); |
f942dc25 IC |
472 | } |
473 | ||
474 | return npo->meta_prod - old_meta_prod; | |
475 | } | |
476 | ||
477 | /* | |
7376419a | 478 | * This is a twin to xenvif_gop_skb. Assume that xenvif_gop_skb was |
f942dc25 IC |
479 | * used to set up the operations on the top of |
480 | * netrx_pending_operations, which have since been done. Check that | |
481 | * they didn't give any errors and advance over them. | |
482 | */ | |
7376419a WL |
483 | static int xenvif_check_gop(struct xenvif *vif, int nr_meta_slots, |
484 | struct netrx_pending_operations *npo) | |
f942dc25 IC |
485 | { |
486 | struct gnttab_copy *copy_op; | |
487 | int status = XEN_NETIF_RSP_OKAY; | |
488 | int i; | |
489 | ||
490 | for (i = 0; i < nr_meta_slots; i++) { | |
491 | copy_op = npo->copy + npo->copy_cons++; | |
492 | if (copy_op->status != GNTST_okay) { | |
493 | netdev_dbg(vif->dev, | |
494 | "Bad status %d from copy to DOM%d.\n", | |
495 | copy_op->status, vif->domid); | |
496 | status = XEN_NETIF_RSP_ERROR; | |
497 | } | |
498 | } | |
499 | ||
500 | return status; | |
501 | } | |
502 | ||
7376419a WL |
503 | static void xenvif_add_frag_responses(struct xenvif *vif, int status, |
504 | struct xenvif_rx_meta *meta, | |
505 | int nr_meta_slots) | |
f942dc25 IC |
506 | { |
507 | int i; | |
508 | unsigned long offset; | |
509 | ||
510 | /* No fragments used */ | |
511 | if (nr_meta_slots <= 1) | |
512 | return; | |
513 | ||
514 | nr_meta_slots--; | |
515 | ||
516 | for (i = 0; i < nr_meta_slots; i++) { | |
517 | int flags; | |
518 | if (i == nr_meta_slots - 1) | |
519 | flags = 0; | |
520 | else | |
521 | flags = XEN_NETRXF_more_data; | |
522 | ||
523 | offset = 0; | |
524 | make_rx_response(vif, meta[i].id, status, offset, | |
525 | meta[i].size, flags); | |
526 | } | |
527 | } | |
528 | ||
33bc801d WL |
529 | struct skb_cb_overlay { |
530 | int meta_slots_used; | |
531 | }; | |
532 | ||
7376419a | 533 | static void xenvif_kick_thread(struct xenvif *vif) |
b3f980bd WL |
534 | { |
535 | wake_up(&vif->wq); | |
536 | } | |
537 | ||
7376419a | 538 | void xenvif_rx_action(struct xenvif *vif) |
f942dc25 | 539 | { |
f942dc25 | 540 | s8 status; |
e1f00a69 | 541 | u16 flags; |
f942dc25 IC |
542 | struct xen_netif_rx_response *resp; |
543 | struct sk_buff_head rxq; | |
544 | struct sk_buff *skb; | |
545 | LIST_HEAD(notify); | |
546 | int ret; | |
547 | int nr_frags; | |
548 | int count; | |
549 | unsigned long offset; | |
550 | struct skb_cb_overlay *sco; | |
b3f980bd | 551 | int need_to_notify = 0; |
f942dc25 IC |
552 | |
553 | struct netrx_pending_operations npo = { | |
b3f980bd WL |
554 | .copy = vif->grant_copy_op, |
555 | .meta = vif->meta, | |
f942dc25 IC |
556 | }; |
557 | ||
558 | skb_queue_head_init(&rxq); | |
559 | ||
560 | count = 0; | |
561 | ||
b3f980bd | 562 | while ((skb = skb_dequeue(&vif->rx_queue)) != NULL) { |
f942dc25 IC |
563 | vif = netdev_priv(skb->dev); |
564 | nr_frags = skb_shinfo(skb)->nr_frags; | |
565 | ||
566 | sco = (struct skb_cb_overlay *)skb->cb; | |
7376419a | 567 | sco->meta_slots_used = xenvif_gop_skb(skb, &npo); |
f942dc25 | 568 | |
33bc801d | 569 | count += nr_frags + 1; |
f942dc25 IC |
570 | |
571 | __skb_queue_tail(&rxq, skb); | |
572 | ||
573 | /* Filled the batch queue? */ | |
33bc801d WL |
574 | /* XXX FIXME: RX path dependent on MAX_SKB_FRAGS */ |
575 | if (count + MAX_SKB_FRAGS >= XEN_NETIF_RX_RING_SIZE) | |
f942dc25 IC |
576 | break; |
577 | } | |
578 | ||
b3f980bd | 579 | BUG_ON(npo.meta_prod > ARRAY_SIZE(vif->meta)); |
f942dc25 IC |
580 | |
581 | if (!npo.copy_prod) | |
582 | return; | |
583 | ||
b3f980bd WL |
584 | BUG_ON(npo.copy_prod > ARRAY_SIZE(vif->grant_copy_op)); |
585 | gnttab_batch_copy(vif->grant_copy_op, npo.copy_prod); | |
f942dc25 IC |
586 | |
587 | while ((skb = __skb_dequeue(&rxq)) != NULL) { | |
588 | sco = (struct skb_cb_overlay *)skb->cb; | |
589 | ||
590 | vif = netdev_priv(skb->dev); | |
591 | ||
b3f980bd | 592 | if (vif->meta[npo.meta_cons].gso_size && vif->gso_prefix) { |
f942dc25 | 593 | resp = RING_GET_RESPONSE(&vif->rx, |
b3f980bd | 594 | vif->rx.rsp_prod_pvt++); |
f942dc25 IC |
595 | |
596 | resp->flags = XEN_NETRXF_gso_prefix | XEN_NETRXF_more_data; | |
597 | ||
b3f980bd WL |
598 | resp->offset = vif->meta[npo.meta_cons].gso_size; |
599 | resp->id = vif->meta[npo.meta_cons].id; | |
f942dc25 IC |
600 | resp->status = sco->meta_slots_used; |
601 | ||
602 | npo.meta_cons++; | |
603 | sco->meta_slots_used--; | |
604 | } | |
605 | ||
606 | ||
607 | vif->dev->stats.tx_bytes += skb->len; | |
608 | vif->dev->stats.tx_packets++; | |
609 | ||
7376419a | 610 | status = xenvif_check_gop(vif, sco->meta_slots_used, &npo); |
f942dc25 IC |
611 | |
612 | if (sco->meta_slots_used == 1) | |
613 | flags = 0; | |
614 | else | |
615 | flags = XEN_NETRXF_more_data; | |
616 | ||
617 | if (skb->ip_summed == CHECKSUM_PARTIAL) /* local packet? */ | |
618 | flags |= XEN_NETRXF_csum_blank | XEN_NETRXF_data_validated; | |
619 | else if (skb->ip_summed == CHECKSUM_UNNECESSARY) | |
620 | /* remote but checksummed. */ | |
621 | flags |= XEN_NETRXF_data_validated; | |
622 | ||
623 | offset = 0; | |
b3f980bd | 624 | resp = make_rx_response(vif, vif->meta[npo.meta_cons].id, |
f942dc25 | 625 | status, offset, |
b3f980bd | 626 | vif->meta[npo.meta_cons].size, |
f942dc25 IC |
627 | flags); |
628 | ||
b3f980bd | 629 | if (vif->meta[npo.meta_cons].gso_size && !vif->gso_prefix) { |
f942dc25 IC |
630 | struct xen_netif_extra_info *gso = |
631 | (struct xen_netif_extra_info *) | |
632 | RING_GET_RESPONSE(&vif->rx, | |
633 | vif->rx.rsp_prod_pvt++); | |
634 | ||
635 | resp->flags |= XEN_NETRXF_extra_info; | |
636 | ||
b3f980bd | 637 | gso->u.gso.size = vif->meta[npo.meta_cons].gso_size; |
f942dc25 IC |
638 | gso->u.gso.type = XEN_NETIF_GSO_TYPE_TCPV4; |
639 | gso->u.gso.pad = 0; | |
640 | gso->u.gso.features = 0; | |
641 | ||
642 | gso->type = XEN_NETIF_EXTRA_TYPE_GSO; | |
643 | gso->flags = 0; | |
644 | } | |
645 | ||
7376419a WL |
646 | xenvif_add_frag_responses(vif, status, |
647 | vif->meta + npo.meta_cons + 1, | |
648 | sco->meta_slots_used); | |
f942dc25 IC |
649 | |
650 | RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&vif->rx, ret); | |
f942dc25 | 651 | |
b3f980bd WL |
652 | if (ret) |
653 | need_to_notify = 1; | |
654 | ||
f942dc25 IC |
655 | xenvif_notify_tx_completion(vif); |
656 | ||
f942dc25 IC |
657 | npo.meta_cons += sco->meta_slots_used; |
658 | dev_kfree_skb(skb); | |
659 | } | |
660 | ||
b3f980bd | 661 | if (need_to_notify) |
e1f00a69 | 662 | notify_remote_via_irq(vif->rx_irq); |
f942dc25 IC |
663 | |
664 | /* More work to do? */ | |
b3f980bd | 665 | if (!skb_queue_empty(&vif->rx_queue)) |
7376419a | 666 | xenvif_kick_thread(vif); |
f942dc25 IC |
667 | } |
668 | ||
7376419a | 669 | void xenvif_queue_tx_skb(struct xenvif *vif, struct sk_buff *skb) |
f942dc25 | 670 | { |
b3f980bd | 671 | skb_queue_tail(&vif->rx_queue, skb); |
f942dc25 | 672 | |
7376419a | 673 | xenvif_kick_thread(vif); |
f942dc25 IC |
674 | } |
675 | ||
7376419a | 676 | void xenvif_check_rx_xenvif(struct xenvif *vif) |
f942dc25 IC |
677 | { |
678 | int more_to_do; | |
679 | ||
680 | RING_FINAL_CHECK_FOR_REQUESTS(&vif->tx, more_to_do); | |
681 | ||
682 | if (more_to_do) | |
b3f980bd | 683 | napi_schedule(&vif->napi); |
f942dc25 IC |
684 | } |
685 | ||
686 | static void tx_add_credit(struct xenvif *vif) | |
687 | { | |
688 | unsigned long max_burst, max_credit; | |
689 | ||
690 | /* | |
691 | * Allow a burst big enough to transmit a jumbo packet of up to 128kB. | |
692 | * Otherwise the interface can seize up due to insufficient credit. | |
693 | */ | |
694 | max_burst = RING_GET_REQUEST(&vif->tx, vif->tx.req_cons)->size; | |
695 | max_burst = min(max_burst, 131072UL); | |
696 | max_burst = max(max_burst, vif->credit_bytes); | |
697 | ||
698 | /* Take care that adding a new chunk of credit doesn't wrap to zero. */ | |
699 | max_credit = vif->remaining_credit + vif->credit_bytes; | |
700 | if (max_credit < vif->remaining_credit) | |
701 | max_credit = ULONG_MAX; /* wrapped: clamp to ULONG_MAX */ | |
702 | ||
703 | vif->remaining_credit = min(max_credit, max_burst); | |
704 | } | |
705 | ||
706 | static void tx_credit_callback(unsigned long data) | |
707 | { | |
708 | struct xenvif *vif = (struct xenvif *)data; | |
709 | tx_add_credit(vif); | |
7376419a | 710 | xenvif_check_rx_xenvif(vif); |
f942dc25 IC |
711 | } |
712 | ||
7376419a WL |
713 | static void xenvif_tx_err(struct xenvif *vif, |
714 | struct xen_netif_tx_request *txp, RING_IDX end) | |
f942dc25 IC |
715 | { |
716 | RING_IDX cons = vif->tx.req_cons; | |
717 | ||
718 | do { | |
719 | make_tx_response(vif, txp, XEN_NETIF_RSP_ERROR); | |
b9149729 | 720 | if (cons == end) |
f942dc25 IC |
721 | break; |
722 | txp = RING_GET_REQUEST(&vif->tx, cons++); | |
723 | } while (1); | |
724 | vif->tx.req_cons = cons; | |
f942dc25 IC |
725 | } |
726 | ||
7376419a | 727 | static void xenvif_fatal_tx_err(struct xenvif *vif) |
48856286 IC |
728 | { |
729 | netdev_err(vif->dev, "fatal error; disabling device\n"); | |
730 | xenvif_carrier_off(vif); | |
48856286 IC |
731 | } |
732 | ||
7376419a WL |
733 | static int xenvif_count_requests(struct xenvif *vif, |
734 | struct xen_netif_tx_request *first, | |
735 | struct xen_netif_tx_request *txp, | |
736 | int work_to_do) | |
f942dc25 IC |
737 | { |
738 | RING_IDX cons = vif->tx.req_cons; | |
2810e5b9 WL |
739 | int slots = 0; |
740 | int drop_err = 0; | |
59ccb4eb | 741 | int more_data; |
f942dc25 IC |
742 | |
743 | if (!(first->flags & XEN_NETTXF_more_data)) | |
744 | return 0; | |
745 | ||
746 | do { | |
59ccb4eb WL |
747 | struct xen_netif_tx_request dropped_tx = { 0 }; |
748 | ||
2810e5b9 WL |
749 | if (slots >= work_to_do) { |
750 | netdev_err(vif->dev, | |
751 | "Asked for %d slots but exceeds this limit\n", | |
752 | work_to_do); | |
7376419a | 753 | xenvif_fatal_tx_err(vif); |
35876b5f | 754 | return -ENODATA; |
f942dc25 IC |
755 | } |
756 | ||
2810e5b9 WL |
757 | /* This guest is really using too many slots and |
758 | * considered malicious. | |
759 | */ | |
37641494 | 760 | if (unlikely(slots >= fatal_skb_slots)) { |
2810e5b9 WL |
761 | netdev_err(vif->dev, |
762 | "Malicious frontend using %d slots, threshold %u\n", | |
37641494 | 763 | slots, fatal_skb_slots); |
7376419a | 764 | xenvif_fatal_tx_err(vif); |
35876b5f | 765 | return -E2BIG; |
f942dc25 IC |
766 | } |
767 | ||
2810e5b9 | 768 | /* Xen network protocol had implicit dependency on |
37641494 WL |
769 | * MAX_SKB_FRAGS. XEN_NETBK_LEGACY_SLOTS_MAX is set to |
770 | * the historical MAX_SKB_FRAGS value 18 to honor the | |
771 | * same behavior as before. Any packet using more than | |
772 | * 18 slots but less than fatal_skb_slots slots is | |
773 | * dropped | |
2810e5b9 | 774 | */ |
37641494 | 775 | if (!drop_err && slots >= XEN_NETBK_LEGACY_SLOTS_MAX) { |
2810e5b9 WL |
776 | if (net_ratelimit()) |
777 | netdev_dbg(vif->dev, | |
778 | "Too many slots (%d) exceeding limit (%d), dropping packet\n", | |
37641494 | 779 | slots, XEN_NETBK_LEGACY_SLOTS_MAX); |
2810e5b9 WL |
780 | drop_err = -E2BIG; |
781 | } | |
782 | ||
59ccb4eb WL |
783 | if (drop_err) |
784 | txp = &dropped_tx; | |
785 | ||
2810e5b9 | 786 | memcpy(txp, RING_GET_REQUEST(&vif->tx, cons + slots), |
f942dc25 | 787 | sizeof(*txp)); |
03393fd5 WL |
788 | |
789 | /* If the guest submitted a frame >= 64 KiB then | |
790 | * first->size overflowed and following slots will | |
791 | * appear to be larger than the frame. | |
792 | * | |
793 | * This cannot be fatal error as there are buggy | |
794 | * frontends that do this. | |
795 | * | |
796 | * Consume all slots and drop the packet. | |
797 | */ | |
798 | if (!drop_err && txp->size > first->size) { | |
799 | if (net_ratelimit()) | |
800 | netdev_dbg(vif->dev, | |
801 | "Invalid tx request, slot size %u > remaining size %u\n", | |
802 | txp->size, first->size); | |
803 | drop_err = -EIO; | |
f942dc25 IC |
804 | } |
805 | ||
806 | first->size -= txp->size; | |
2810e5b9 | 807 | slots++; |
f942dc25 IC |
808 | |
809 | if (unlikely((txp->offset + txp->size) > PAGE_SIZE)) { | |
2810e5b9 | 810 | netdev_err(vif->dev, "Cross page boundary, txp->offset: %x, size: %u\n", |
f942dc25 | 811 | txp->offset, txp->size); |
7376419a | 812 | xenvif_fatal_tx_err(vif); |
35876b5f | 813 | return -EINVAL; |
f942dc25 | 814 | } |
59ccb4eb WL |
815 | |
816 | more_data = txp->flags & XEN_NETTXF_more_data; | |
817 | ||
818 | if (!drop_err) | |
819 | txp++; | |
820 | ||
821 | } while (more_data); | |
2810e5b9 WL |
822 | |
823 | if (drop_err) { | |
7376419a | 824 | xenvif_tx_err(vif, first, cons + slots); |
2810e5b9 WL |
825 | return drop_err; |
826 | } | |
827 | ||
828 | return slots; | |
f942dc25 IC |
829 | } |
830 | ||
7376419a WL |
831 | static struct page *xenvif_alloc_page(struct xenvif *vif, |
832 | u16 pending_idx) | |
f942dc25 IC |
833 | { |
834 | struct page *page; | |
b3f980bd WL |
835 | |
836 | page = alloc_page(GFP_ATOMIC|__GFP_COLD); | |
f942dc25 IC |
837 | if (!page) |
838 | return NULL; | |
b3f980bd WL |
839 | vif->mmap_pages[pending_idx] = page; |
840 | ||
f942dc25 IC |
841 | return page; |
842 | } | |
843 | ||
7376419a WL |
844 | static struct gnttab_copy *xenvif_get_requests(struct xenvif *vif, |
845 | struct sk_buff *skb, | |
846 | struct xen_netif_tx_request *txp, | |
847 | struct gnttab_copy *gop) | |
f942dc25 IC |
848 | { |
849 | struct skb_shared_info *shinfo = skb_shinfo(skb); | |
850 | skb_frag_t *frags = shinfo->frags; | |
ea066ad1 | 851 | u16 pending_idx = *((u16 *)skb->data); |
2810e5b9 WL |
852 | u16 head_idx = 0; |
853 | int slot, start; | |
854 | struct page *page; | |
855 | pending_ring_idx_t index, start_idx = 0; | |
856 | uint16_t dst_offset; | |
857 | unsigned int nr_slots; | |
858 | struct pending_tx_info *first = NULL; | |
859 | ||
860 | /* At this point shinfo->nr_frags is in fact the number of | |
37641494 | 861 | * slots, which can be as large as XEN_NETBK_LEGACY_SLOTS_MAX. |
2810e5b9 WL |
862 | */ |
863 | nr_slots = shinfo->nr_frags; | |
f942dc25 IC |
864 | |
865 | /* Skip first skb fragment if it is on same page as header fragment. */ | |
ea066ad1 | 866 | start = (frag_get_pending_idx(&shinfo->frags[0]) == pending_idx); |
f942dc25 | 867 | |
2810e5b9 WL |
868 | /* Coalesce tx requests, at this point the packet passed in |
869 | * should be <= 64K. Any packets larger than 64K have been | |
7376419a | 870 | * handled in xenvif_count_requests(). |
2810e5b9 WL |
871 | */ |
872 | for (shinfo->nr_frags = slot = start; slot < nr_slots; | |
873 | shinfo->nr_frags++) { | |
f942dc25 | 874 | struct pending_tx_info *pending_tx_info = |
b3f980bd | 875 | vif->pending_tx_info; |
f942dc25 | 876 | |
b3f980bd | 877 | page = alloc_page(GFP_ATOMIC|__GFP_COLD); |
f942dc25 | 878 | if (!page) |
4cc7c1cb | 879 | goto err; |
f942dc25 | 880 | |
2810e5b9 WL |
881 | dst_offset = 0; |
882 | first = NULL; | |
883 | while (dst_offset < PAGE_SIZE && slot < nr_slots) { | |
884 | gop->flags = GNTCOPY_source_gref; | |
885 | ||
886 | gop->source.u.ref = txp->gref; | |
887 | gop->source.domid = vif->domid; | |
888 | gop->source.offset = txp->offset; | |
889 | ||
890 | gop->dest.domid = DOMID_SELF; | |
891 | ||
892 | gop->dest.offset = dst_offset; | |
893 | gop->dest.u.gmfn = virt_to_mfn(page_address(page)); | |
894 | ||
895 | if (dst_offset + txp->size > PAGE_SIZE) { | |
896 | /* This page can only merge a portion | |
897 | * of tx request. Do not increment any | |
898 | * pointer / counter here. The txp | |
899 | * will be dealt with in future | |
900 | * rounds, eventually hitting the | |
901 | * `else` branch. | |
902 | */ | |
903 | gop->len = PAGE_SIZE - dst_offset; | |
904 | txp->offset += gop->len; | |
905 | txp->size -= gop->len; | |
906 | dst_offset += gop->len; /* quit loop */ | |
907 | } else { | |
908 | /* This tx request can be merged in the page */ | |
909 | gop->len = txp->size; | |
910 | dst_offset += gop->len; | |
911 | ||
b3f980bd | 912 | index = pending_index(vif->pending_cons++); |
2810e5b9 | 913 | |
b3f980bd | 914 | pending_idx = vif->pending_ring[index]; |
2810e5b9 WL |
915 | |
916 | memcpy(&pending_tx_info[pending_idx].req, txp, | |
917 | sizeof(*txp)); | |
2810e5b9 WL |
918 | |
919 | /* Poison these fields, corresponding | |
920 | * fields for head tx req will be set | |
921 | * to correct values after the loop. | |
922 | */ | |
b3f980bd | 923 | vif->mmap_pages[pending_idx] = (void *)(~0UL); |
2810e5b9 WL |
924 | pending_tx_info[pending_idx].head = |
925 | INVALID_PENDING_RING_IDX; | |
926 | ||
927 | if (!first) { | |
928 | first = &pending_tx_info[pending_idx]; | |
929 | start_idx = index; | |
930 | head_idx = pending_idx; | |
931 | } | |
932 | ||
933 | txp++; | |
934 | slot++; | |
935 | } | |
f942dc25 | 936 | |
2810e5b9 WL |
937 | gop++; |
938 | } | |
f942dc25 | 939 | |
2810e5b9 WL |
940 | first->req.offset = 0; |
941 | first->req.size = dst_offset; | |
942 | first->head = start_idx; | |
b3f980bd | 943 | vif->mmap_pages[head_idx] = page; |
2810e5b9 | 944 | frag_set_pending_idx(&frags[shinfo->nr_frags], head_idx); |
f942dc25 IC |
945 | } |
946 | ||
2810e5b9 WL |
947 | BUG_ON(shinfo->nr_frags > MAX_SKB_FRAGS); |
948 | ||
f942dc25 | 949 | return gop; |
4cc7c1cb IC |
950 | err: |
951 | /* Unwind, freeing all pages and sending error responses. */ | |
2810e5b9 | 952 | while (shinfo->nr_frags-- > start) { |
7376419a | 953 | xenvif_idx_release(vif, |
2810e5b9 WL |
954 | frag_get_pending_idx(&frags[shinfo->nr_frags]), |
955 | XEN_NETIF_RSP_ERROR); | |
4cc7c1cb IC |
956 | } |
957 | /* The head too, if necessary. */ | |
958 | if (start) | |
7376419a | 959 | xenvif_idx_release(vif, pending_idx, XEN_NETIF_RSP_ERROR); |
4cc7c1cb IC |
960 | |
961 | return NULL; | |
f942dc25 IC |
962 | } |
963 | ||
7376419a WL |
964 | static int xenvif_tx_check_gop(struct xenvif *vif, |
965 | struct sk_buff *skb, | |
966 | struct gnttab_copy **gopp) | |
f942dc25 IC |
967 | { |
968 | struct gnttab_copy *gop = *gopp; | |
ea066ad1 | 969 | u16 pending_idx = *((u16 *)skb->data); |
f942dc25 | 970 | struct skb_shared_info *shinfo = skb_shinfo(skb); |
2810e5b9 | 971 | struct pending_tx_info *tx_info; |
f942dc25 IC |
972 | int nr_frags = shinfo->nr_frags; |
973 | int i, err, start; | |
2810e5b9 | 974 | u16 peek; /* peek into next tx request */ |
f942dc25 IC |
975 | |
976 | /* Check status of header. */ | |
977 | err = gop->status; | |
7d5145d8 | 978 | if (unlikely(err)) |
7376419a | 979 | xenvif_idx_release(vif, pending_idx, XEN_NETIF_RSP_ERROR); |
f942dc25 IC |
980 | |
981 | /* Skip first skb fragment if it is on same page as header fragment. */ | |
ea066ad1 | 982 | start = (frag_get_pending_idx(&shinfo->frags[0]) == pending_idx); |
f942dc25 IC |
983 | |
984 | for (i = start; i < nr_frags; i++) { | |
985 | int j, newerr; | |
2810e5b9 | 986 | pending_ring_idx_t head; |
f942dc25 | 987 | |
ea066ad1 | 988 | pending_idx = frag_get_pending_idx(&shinfo->frags[i]); |
b3f980bd | 989 | tx_info = &vif->pending_tx_info[pending_idx]; |
2810e5b9 | 990 | head = tx_info->head; |
f942dc25 IC |
991 | |
992 | /* Check error status: if okay then remember grant handle. */ | |
2810e5b9 WL |
993 | do { |
994 | newerr = (++gop)->status; | |
995 | if (newerr) | |
996 | break; | |
b3f980bd WL |
997 | peek = vif->pending_ring[pending_index(++head)]; |
998 | } while (!pending_tx_is_head(vif, peek)); | |
2810e5b9 | 999 | |
f942dc25 IC |
1000 | if (likely(!newerr)) { |
1001 | /* Had a previous error? Invalidate this fragment. */ | |
1002 | if (unlikely(err)) | |
7376419a WL |
1003 | xenvif_idx_release(vif, pending_idx, |
1004 | XEN_NETIF_RSP_OKAY); | |
f942dc25 IC |
1005 | continue; |
1006 | } | |
1007 | ||
1008 | /* Error on this fragment: respond to client with an error. */ | |
7376419a | 1009 | xenvif_idx_release(vif, pending_idx, XEN_NETIF_RSP_ERROR); |
f942dc25 IC |
1010 | |
1011 | /* Not the first error? Preceding frags already invalidated. */ | |
1012 | if (err) | |
1013 | continue; | |
1014 | ||
1015 | /* First error: invalidate header and preceding fragments. */ | |
1016 | pending_idx = *((u16 *)skb->data); | |
7376419a | 1017 | xenvif_idx_release(vif, pending_idx, XEN_NETIF_RSP_OKAY); |
f942dc25 | 1018 | for (j = start; j < i; j++) { |
5ccb3ea7 | 1019 | pending_idx = frag_get_pending_idx(&shinfo->frags[j]); |
7376419a WL |
1020 | xenvif_idx_release(vif, pending_idx, |
1021 | XEN_NETIF_RSP_OKAY); | |
f942dc25 IC |
1022 | } |
1023 | ||
1024 | /* Remember the error: invalidate all subsequent fragments. */ | |
1025 | err = newerr; | |
1026 | } | |
1027 | ||
1028 | *gopp = gop + 1; | |
1029 | return err; | |
1030 | } | |
1031 | ||
7376419a | 1032 | static void xenvif_fill_frags(struct xenvif *vif, struct sk_buff *skb) |
f942dc25 IC |
1033 | { |
1034 | struct skb_shared_info *shinfo = skb_shinfo(skb); | |
1035 | int nr_frags = shinfo->nr_frags; | |
1036 | int i; | |
1037 | ||
1038 | for (i = 0; i < nr_frags; i++) { | |
1039 | skb_frag_t *frag = shinfo->frags + i; | |
1040 | struct xen_netif_tx_request *txp; | |
ea066ad1 IC |
1041 | struct page *page; |
1042 | u16 pending_idx; | |
f942dc25 | 1043 | |
ea066ad1 | 1044 | pending_idx = frag_get_pending_idx(frag); |
f942dc25 | 1045 | |
b3f980bd WL |
1046 | txp = &vif->pending_tx_info[pending_idx].req; |
1047 | page = virt_to_page(idx_to_kaddr(vif, pending_idx)); | |
ea066ad1 | 1048 | __skb_fill_page_desc(skb, i, page, txp->offset, txp->size); |
f942dc25 IC |
1049 | skb->len += txp->size; |
1050 | skb->data_len += txp->size; | |
1051 | skb->truesize += txp->size; | |
1052 | ||
7376419a | 1053 | /* Take an extra reference to offset xenvif_idx_release */ |
b3f980bd | 1054 | get_page(vif->mmap_pages[pending_idx]); |
7376419a | 1055 | xenvif_idx_release(vif, pending_idx, XEN_NETIF_RSP_OKAY); |
f942dc25 IC |
1056 | } |
1057 | } | |
1058 | ||
7376419a | 1059 | static int xenvif_get_extras(struct xenvif *vif, |
f942dc25 IC |
1060 | struct xen_netif_extra_info *extras, |
1061 | int work_to_do) | |
1062 | { | |
1063 | struct xen_netif_extra_info extra; | |
1064 | RING_IDX cons = vif->tx.req_cons; | |
1065 | ||
1066 | do { | |
1067 | if (unlikely(work_to_do-- <= 0)) { | |
48856286 | 1068 | netdev_err(vif->dev, "Missing extra info\n"); |
7376419a | 1069 | xenvif_fatal_tx_err(vif); |
f942dc25 IC |
1070 | return -EBADR; |
1071 | } | |
1072 | ||
1073 | memcpy(&extra, RING_GET_REQUEST(&vif->tx, cons), | |
1074 | sizeof(extra)); | |
1075 | if (unlikely(!extra.type || | |
1076 | extra.type >= XEN_NETIF_EXTRA_TYPE_MAX)) { | |
1077 | vif->tx.req_cons = ++cons; | |
48856286 | 1078 | netdev_err(vif->dev, |
f942dc25 | 1079 | "Invalid extra type: %d\n", extra.type); |
7376419a | 1080 | xenvif_fatal_tx_err(vif); |
f942dc25 IC |
1081 | return -EINVAL; |
1082 | } | |
1083 | ||
1084 | memcpy(&extras[extra.type - 1], &extra, sizeof(extra)); | |
1085 | vif->tx.req_cons = ++cons; | |
1086 | } while (extra.flags & XEN_NETIF_EXTRA_FLAG_MORE); | |
1087 | ||
1088 | return work_to_do; | |
1089 | } | |
1090 | ||
7376419a WL |
1091 | static int xenvif_set_skb_gso(struct xenvif *vif, |
1092 | struct sk_buff *skb, | |
1093 | struct xen_netif_extra_info *gso) | |
f942dc25 IC |
1094 | { |
1095 | if (!gso->u.gso.size) { | |
48856286 | 1096 | netdev_err(vif->dev, "GSO size must not be zero.\n"); |
7376419a | 1097 | xenvif_fatal_tx_err(vif); |
f942dc25 IC |
1098 | return -EINVAL; |
1099 | } | |
1100 | ||
1101 | /* Currently only TCPv4 S.O. is supported. */ | |
1102 | if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4) { | |
48856286 | 1103 | netdev_err(vif->dev, "Bad GSO type %d.\n", gso->u.gso.type); |
7376419a | 1104 | xenvif_fatal_tx_err(vif); |
f942dc25 IC |
1105 | return -EINVAL; |
1106 | } | |
1107 | ||
1108 | skb_shinfo(skb)->gso_size = gso->u.gso.size; | |
1109 | skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4; | |
1110 | ||
1111 | /* Header must be checked, and gso_segs computed. */ | |
1112 | skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; | |
1113 | skb_shinfo(skb)->gso_segs = 0; | |
1114 | ||
1115 | return 0; | |
1116 | } | |
1117 | ||
2eba61d5 PD |
1118 | static inline void maybe_pull_tail(struct sk_buff *skb, unsigned int len) |
1119 | { | |
1120 | if (skb_is_nonlinear(skb) && skb_headlen(skb) < len) { | |
1121 | /* If we need to pullup then pullup to the max, so we | |
1122 | * won't need to do it again. | |
1123 | */ | |
1124 | int target = min_t(int, skb->len, MAX_TCP_HEADER); | |
1125 | __pskb_pull_tail(skb, target - skb_headlen(skb)); | |
1126 | } | |
1127 | } | |
1128 | ||
1129 | static int checksum_setup_ip(struct xenvif *vif, struct sk_buff *skb, | |
1130 | int recalculate_partial_csum) | |
f942dc25 | 1131 | { |
2eba61d5 PD |
1132 | struct iphdr *iph = (void *)skb->data; |
1133 | unsigned int header_size; | |
1134 | unsigned int off; | |
f942dc25 | 1135 | int err = -EPROTO; |
f942dc25 | 1136 | |
2eba61d5 | 1137 | off = sizeof(struct iphdr); |
f942dc25 | 1138 | |
2eba61d5 PD |
1139 | header_size = skb->network_header + off + MAX_IPOPTLEN; |
1140 | maybe_pull_tail(skb, header_size); | |
f942dc25 | 1141 | |
2eba61d5 | 1142 | off = iph->ihl * 4; |
f942dc25 | 1143 | |
f942dc25 IC |
1144 | switch (iph->protocol) { |
1145 | case IPPROTO_TCP: | |
2eba61d5 | 1146 | if (!skb_partial_csum_set(skb, off, |
bea89336 JW |
1147 | offsetof(struct tcphdr, check))) |
1148 | goto out; | |
f942dc25 IC |
1149 | |
1150 | if (recalculate_partial_csum) { | |
bea89336 | 1151 | struct tcphdr *tcph = tcp_hdr(skb); |
2eba61d5 PD |
1152 | |
1153 | header_size = skb->network_header + | |
1154 | off + | |
1155 | sizeof(struct tcphdr); | |
1156 | maybe_pull_tail(skb, header_size); | |
1157 | ||
f942dc25 | 1158 | tcph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, |
2eba61d5 | 1159 | skb->len - off, |
f942dc25 IC |
1160 | IPPROTO_TCP, 0); |
1161 | } | |
1162 | break; | |
1163 | case IPPROTO_UDP: | |
2eba61d5 | 1164 | if (!skb_partial_csum_set(skb, off, |
bea89336 JW |
1165 | offsetof(struct udphdr, check))) |
1166 | goto out; | |
f942dc25 IC |
1167 | |
1168 | if (recalculate_partial_csum) { | |
bea89336 | 1169 | struct udphdr *udph = udp_hdr(skb); |
2eba61d5 PD |
1170 | |
1171 | header_size = skb->network_header + | |
1172 | off + | |
1173 | sizeof(struct udphdr); | |
1174 | maybe_pull_tail(skb, header_size); | |
1175 | ||
f942dc25 | 1176 | udph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, |
2eba61d5 | 1177 | skb->len - off, |
f942dc25 IC |
1178 | IPPROTO_UDP, 0); |
1179 | } | |
1180 | break; | |
1181 | default: | |
1182 | if (net_ratelimit()) | |
1183 | netdev_err(vif->dev, | |
2eba61d5 PD |
1184 | "Attempting to checksum a non-TCP/UDP packet, " |
1185 | "dropping a protocol %d packet\n", | |
f942dc25 IC |
1186 | iph->protocol); |
1187 | goto out; | |
1188 | } | |
1189 | ||
f942dc25 IC |
1190 | err = 0; |
1191 | ||
1192 | out: | |
1193 | return err; | |
1194 | } | |
1195 | ||
2eba61d5 PD |
1196 | static int checksum_setup_ipv6(struct xenvif *vif, struct sk_buff *skb, |
1197 | int recalculate_partial_csum) | |
1198 | { | |
1199 | int err = -EPROTO; | |
1200 | struct ipv6hdr *ipv6h = (void *)skb->data; | |
1201 | u8 nexthdr; | |
1202 | unsigned int header_size; | |
1203 | unsigned int off; | |
1204 | bool fragment; | |
1205 | bool done; | |
1206 | ||
1207 | done = false; | |
1208 | ||
1209 | off = sizeof(struct ipv6hdr); | |
1210 | ||
1211 | header_size = skb->network_header + off; | |
1212 | maybe_pull_tail(skb, header_size); | |
1213 | ||
1214 | nexthdr = ipv6h->nexthdr; | |
1215 | ||
1216 | while ((off <= sizeof(struct ipv6hdr) + ntohs(ipv6h->payload_len)) && | |
1217 | !done) { | |
1218 | switch (nexthdr) { | |
1219 | case IPPROTO_DSTOPTS: | |
1220 | case IPPROTO_HOPOPTS: | |
1221 | case IPPROTO_ROUTING: { | |
1222 | struct ipv6_opt_hdr *hp = (void *)(skb->data + off); | |
1223 | ||
1224 | header_size = skb->network_header + | |
1225 | off + | |
1226 | sizeof(struct ipv6_opt_hdr); | |
1227 | maybe_pull_tail(skb, header_size); | |
1228 | ||
1229 | nexthdr = hp->nexthdr; | |
1230 | off += ipv6_optlen(hp); | |
1231 | break; | |
1232 | } | |
1233 | case IPPROTO_AH: { | |
1234 | struct ip_auth_hdr *hp = (void *)(skb->data + off); | |
1235 | ||
1236 | header_size = skb->network_header + | |
1237 | off + | |
1238 | sizeof(struct ip_auth_hdr); | |
1239 | maybe_pull_tail(skb, header_size); | |
1240 | ||
1241 | nexthdr = hp->nexthdr; | |
1242 | off += (hp->hdrlen+2)<<2; | |
1243 | break; | |
1244 | } | |
1245 | case IPPROTO_FRAGMENT: | |
1246 | fragment = true; | |
1247 | /* fall through */ | |
1248 | default: | |
1249 | done = true; | |
1250 | break; | |
1251 | } | |
1252 | } | |
1253 | ||
1254 | if (!done) { | |
1255 | if (net_ratelimit()) | |
1256 | netdev_err(vif->dev, "Failed to parse packet header\n"); | |
1257 | goto out; | |
1258 | } | |
1259 | ||
1260 | if (fragment) { | |
1261 | if (net_ratelimit()) | |
1262 | netdev_err(vif->dev, "Packet is a fragment!\n"); | |
1263 | goto out; | |
1264 | } | |
1265 | ||
1266 | switch (nexthdr) { | |
1267 | case IPPROTO_TCP: | |
1268 | if (!skb_partial_csum_set(skb, off, | |
1269 | offsetof(struct tcphdr, check))) | |
1270 | goto out; | |
1271 | ||
1272 | if (recalculate_partial_csum) { | |
1273 | struct tcphdr *tcph = tcp_hdr(skb); | |
1274 | ||
1275 | header_size = skb->network_header + | |
1276 | off + | |
1277 | sizeof(struct tcphdr); | |
1278 | maybe_pull_tail(skb, header_size); | |
1279 | ||
1280 | tcph->check = ~csum_ipv6_magic(&ipv6h->saddr, | |
1281 | &ipv6h->daddr, | |
1282 | skb->len - off, | |
1283 | IPPROTO_TCP, 0); | |
1284 | } | |
1285 | break; | |
1286 | case IPPROTO_UDP: | |
1287 | if (!skb_partial_csum_set(skb, off, | |
1288 | offsetof(struct udphdr, check))) | |
1289 | goto out; | |
1290 | ||
1291 | if (recalculate_partial_csum) { | |
1292 | struct udphdr *udph = udp_hdr(skb); | |
1293 | ||
1294 | header_size = skb->network_header + | |
1295 | off + | |
1296 | sizeof(struct udphdr); | |
1297 | maybe_pull_tail(skb, header_size); | |
1298 | ||
1299 | udph->check = ~csum_ipv6_magic(&ipv6h->saddr, | |
1300 | &ipv6h->daddr, | |
1301 | skb->len - off, | |
1302 | IPPROTO_UDP, 0); | |
1303 | } | |
1304 | break; | |
1305 | default: | |
1306 | if (net_ratelimit()) | |
1307 | netdev_err(vif->dev, | |
1308 | "Attempting to checksum a non-TCP/UDP packet, " | |
1309 | "dropping a protocol %d packet\n", | |
1310 | nexthdr); | |
1311 | goto out; | |
1312 | } | |
1313 | ||
1314 | err = 0; | |
1315 | ||
1316 | out: | |
1317 | return err; | |
1318 | } | |
1319 | ||
1320 | static int checksum_setup(struct xenvif *vif, struct sk_buff *skb) | |
1321 | { | |
1322 | int err = -EPROTO; | |
1323 | int recalculate_partial_csum = 0; | |
1324 | ||
1325 | /* A GSO SKB must be CHECKSUM_PARTIAL. However some buggy | |
1326 | * peers can fail to set NETRXF_csum_blank when sending a GSO | |
1327 | * frame. In this case force the SKB to CHECKSUM_PARTIAL and | |
1328 | * recalculate the partial checksum. | |
1329 | */ | |
1330 | if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) { | |
1331 | vif->rx_gso_checksum_fixup++; | |
1332 | skb->ip_summed = CHECKSUM_PARTIAL; | |
1333 | recalculate_partial_csum = 1; | |
1334 | } | |
1335 | ||
1336 | /* A non-CHECKSUM_PARTIAL SKB does not require setup. */ | |
1337 | if (skb->ip_summed != CHECKSUM_PARTIAL) | |
1338 | return 0; | |
1339 | ||
1340 | if (skb->protocol == htons(ETH_P_IP)) | |
1341 | err = checksum_setup_ip(vif, skb, recalculate_partial_csum); | |
1342 | else if (skb->protocol == htons(ETH_P_IPV6)) | |
1343 | err = checksum_setup_ipv6(vif, skb, recalculate_partial_csum); | |
1344 | ||
1345 | return err; | |
1346 | } | |
1347 | ||
f942dc25 IC |
1348 | static bool tx_credit_exceeded(struct xenvif *vif, unsigned size) |
1349 | { | |
1350 | unsigned long now = jiffies; | |
1351 | unsigned long next_credit = | |
1352 | vif->credit_timeout.expires + | |
1353 | msecs_to_jiffies(vif->credit_usec / 1000); | |
1354 | ||
1355 | /* Timer could already be pending in rare cases. */ | |
1356 | if (timer_pending(&vif->credit_timeout)) | |
1357 | return true; | |
1358 | ||
1359 | /* Passed the point where we can replenish credit? */ | |
1360 | if (time_after_eq(now, next_credit)) { | |
1361 | vif->credit_timeout.expires = now; | |
1362 | tx_add_credit(vif); | |
1363 | } | |
1364 | ||
1365 | /* Still too big to send right now? Set a callback. */ | |
1366 | if (size > vif->remaining_credit) { | |
1367 | vif->credit_timeout.data = | |
1368 | (unsigned long)vif; | |
1369 | vif->credit_timeout.function = | |
1370 | tx_credit_callback; | |
1371 | mod_timer(&vif->credit_timeout, | |
1372 | next_credit); | |
1373 | ||
1374 | return true; | |
1375 | } | |
1376 | ||
1377 | return false; | |
1378 | } | |
1379 | ||
7376419a | 1380 | static unsigned xenvif_tx_build_gops(struct xenvif *vif) |
f942dc25 | 1381 | { |
b3f980bd | 1382 | struct gnttab_copy *gop = vif->tx_copy_ops, *request_gop; |
f942dc25 IC |
1383 | struct sk_buff *skb; |
1384 | int ret; | |
1385 | ||
b3f980bd WL |
1386 | while ((nr_pending_reqs(vif) + XEN_NETBK_LEGACY_SLOTS_MAX |
1387 | < MAX_PENDING_REQS)) { | |
f942dc25 | 1388 | struct xen_netif_tx_request txreq; |
37641494 | 1389 | struct xen_netif_tx_request txfrags[XEN_NETBK_LEGACY_SLOTS_MAX]; |
f942dc25 IC |
1390 | struct page *page; |
1391 | struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX-1]; | |
1392 | u16 pending_idx; | |
1393 | RING_IDX idx; | |
1394 | int work_to_do; | |
1395 | unsigned int data_len; | |
1396 | pending_ring_idx_t index; | |
1397 | ||
48856286 IC |
1398 | if (vif->tx.sring->req_prod - vif->tx.req_cons > |
1399 | XEN_NETIF_TX_RING_SIZE) { | |
1400 | netdev_err(vif->dev, | |
1401 | "Impossible number of requests. " | |
1402 | "req_prod %d, req_cons %d, size %ld\n", | |
1403 | vif->tx.sring->req_prod, vif->tx.req_cons, | |
1404 | XEN_NETIF_TX_RING_SIZE); | |
7376419a | 1405 | xenvif_fatal_tx_err(vif); |
48856286 IC |
1406 | continue; |
1407 | } | |
1408 | ||
f942dc25 | 1409 | RING_FINAL_CHECK_FOR_REQUESTS(&vif->tx, work_to_do); |
b3f980bd WL |
1410 | if (!work_to_do) |
1411 | break; | |
f942dc25 IC |
1412 | |
1413 | idx = vif->tx.req_cons; | |
1414 | rmb(); /* Ensure that we see the request before we copy it. */ | |
1415 | memcpy(&txreq, RING_GET_REQUEST(&vif->tx, idx), sizeof(txreq)); | |
1416 | ||
1417 | /* Credit-based scheduling. */ | |
1418 | if (txreq.size > vif->remaining_credit && | |
b3f980bd WL |
1419 | tx_credit_exceeded(vif, txreq.size)) |
1420 | break; | |
f942dc25 IC |
1421 | |
1422 | vif->remaining_credit -= txreq.size; | |
1423 | ||
1424 | work_to_do--; | |
1425 | vif->tx.req_cons = ++idx; | |
1426 | ||
1427 | memset(extras, 0, sizeof(extras)); | |
1428 | if (txreq.flags & XEN_NETTXF_extra_info) { | |
7376419a WL |
1429 | work_to_do = xenvif_get_extras(vif, extras, |
1430 | work_to_do); | |
f942dc25 | 1431 | idx = vif->tx.req_cons; |
48856286 | 1432 | if (unlikely(work_to_do < 0)) |
b3f980bd | 1433 | break; |
f942dc25 IC |
1434 | } |
1435 | ||
7376419a | 1436 | ret = xenvif_count_requests(vif, &txreq, txfrags, work_to_do); |
48856286 | 1437 | if (unlikely(ret < 0)) |
b3f980bd | 1438 | break; |
48856286 | 1439 | |
f942dc25 IC |
1440 | idx += ret; |
1441 | ||
1442 | if (unlikely(txreq.size < ETH_HLEN)) { | |
1443 | netdev_dbg(vif->dev, | |
1444 | "Bad packet size: %d\n", txreq.size); | |
7376419a | 1445 | xenvif_tx_err(vif, &txreq, idx); |
b3f980bd | 1446 | break; |
f942dc25 IC |
1447 | } |
1448 | ||
1449 | /* No crossing a page as the payload mustn't fragment. */ | |
1450 | if (unlikely((txreq.offset + txreq.size) > PAGE_SIZE)) { | |
48856286 | 1451 | netdev_err(vif->dev, |
f942dc25 IC |
1452 | "txreq.offset: %x, size: %u, end: %lu\n", |
1453 | txreq.offset, txreq.size, | |
1454 | (txreq.offset&~PAGE_MASK) + txreq.size); | |
7376419a | 1455 | xenvif_fatal_tx_err(vif); |
b3f980bd | 1456 | break; |
f942dc25 IC |
1457 | } |
1458 | ||
b3f980bd WL |
1459 | index = pending_index(vif->pending_cons); |
1460 | pending_idx = vif->pending_ring[index]; | |
f942dc25 IC |
1461 | |
1462 | data_len = (txreq.size > PKT_PROT_LEN && | |
37641494 | 1463 | ret < XEN_NETBK_LEGACY_SLOTS_MAX) ? |
f942dc25 IC |
1464 | PKT_PROT_LEN : txreq.size; |
1465 | ||
1466 | skb = alloc_skb(data_len + NET_SKB_PAD + NET_IP_ALIGN, | |
1467 | GFP_ATOMIC | __GFP_NOWARN); | |
1468 | if (unlikely(skb == NULL)) { | |
1469 | netdev_dbg(vif->dev, | |
1470 | "Can't allocate a skb in start_xmit.\n"); | |
7376419a | 1471 | xenvif_tx_err(vif, &txreq, idx); |
f942dc25 IC |
1472 | break; |
1473 | } | |
1474 | ||
1475 | /* Packets passed to netif_rx() must have some headroom. */ | |
1476 | skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN); | |
1477 | ||
1478 | if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) { | |
1479 | struct xen_netif_extra_info *gso; | |
1480 | gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1]; | |
1481 | ||
7376419a WL |
1482 | if (xenvif_set_skb_gso(vif, skb, gso)) { |
1483 | /* Failure in xenvif_set_skb_gso is fatal. */ | |
f942dc25 | 1484 | kfree_skb(skb); |
b3f980bd | 1485 | break; |
f942dc25 IC |
1486 | } |
1487 | } | |
1488 | ||
1489 | /* XXX could copy straight to head */ | |
7376419a | 1490 | page = xenvif_alloc_page(vif, pending_idx); |
f942dc25 IC |
1491 | if (!page) { |
1492 | kfree_skb(skb); | |
7376419a | 1493 | xenvif_tx_err(vif, &txreq, idx); |
b3f980bd | 1494 | break; |
f942dc25 IC |
1495 | } |
1496 | ||
f942dc25 IC |
1497 | gop->source.u.ref = txreq.gref; |
1498 | gop->source.domid = vif->domid; | |
1499 | gop->source.offset = txreq.offset; | |
1500 | ||
1501 | gop->dest.u.gmfn = virt_to_mfn(page_address(page)); | |
1502 | gop->dest.domid = DOMID_SELF; | |
1503 | gop->dest.offset = txreq.offset; | |
1504 | ||
1505 | gop->len = txreq.size; | |
1506 | gop->flags = GNTCOPY_source_gref; | |
1507 | ||
1508 | gop++; | |
1509 | ||
b3f980bd | 1510 | memcpy(&vif->pending_tx_info[pending_idx].req, |
f942dc25 | 1511 | &txreq, sizeof(txreq)); |
b3f980bd | 1512 | vif->pending_tx_info[pending_idx].head = index; |
f942dc25 IC |
1513 | *((u16 *)skb->data) = pending_idx; |
1514 | ||
1515 | __skb_put(skb, data_len); | |
1516 | ||
1517 | skb_shinfo(skb)->nr_frags = ret; | |
1518 | if (data_len < txreq.size) { | |
1519 | skb_shinfo(skb)->nr_frags++; | |
ea066ad1 IC |
1520 | frag_set_pending_idx(&skb_shinfo(skb)->frags[0], |
1521 | pending_idx); | |
f942dc25 | 1522 | } else { |
ea066ad1 IC |
1523 | frag_set_pending_idx(&skb_shinfo(skb)->frags[0], |
1524 | INVALID_PENDING_IDX); | |
f942dc25 IC |
1525 | } |
1526 | ||
b3f980bd | 1527 | vif->pending_cons++; |
f942dc25 | 1528 | |
7376419a | 1529 | request_gop = xenvif_get_requests(vif, skb, txfrags, gop); |
f942dc25 IC |
1530 | if (request_gop == NULL) { |
1531 | kfree_skb(skb); | |
7376419a | 1532 | xenvif_tx_err(vif, &txreq, idx); |
b3f980bd | 1533 | break; |
f942dc25 IC |
1534 | } |
1535 | gop = request_gop; | |
1536 | ||
b3f980bd | 1537 | __skb_queue_tail(&vif->tx_queue, skb); |
1e0b6eac | 1538 | |
f942dc25 | 1539 | vif->tx.req_cons = idx; |
f942dc25 | 1540 | |
b3f980bd | 1541 | if ((gop-vif->tx_copy_ops) >= ARRAY_SIZE(vif->tx_copy_ops)) |
f942dc25 IC |
1542 | break; |
1543 | } | |
1544 | ||
b3f980bd | 1545 | return gop - vif->tx_copy_ops; |
f942dc25 IC |
1546 | } |
1547 | ||
b3f980bd | 1548 | |
7376419a | 1549 | static int xenvif_tx_submit(struct xenvif *vif, int budget) |
f942dc25 | 1550 | { |
b3f980bd | 1551 | struct gnttab_copy *gop = vif->tx_copy_ops; |
f942dc25 | 1552 | struct sk_buff *skb; |
b3f980bd | 1553 | int work_done = 0; |
f942dc25 | 1554 | |
b3f980bd WL |
1555 | while (work_done < budget && |
1556 | (skb = __skb_dequeue(&vif->tx_queue)) != NULL) { | |
f942dc25 | 1557 | struct xen_netif_tx_request *txp; |
f942dc25 IC |
1558 | u16 pending_idx; |
1559 | unsigned data_len; | |
1560 | ||
1561 | pending_idx = *((u16 *)skb->data); | |
b3f980bd | 1562 | txp = &vif->pending_tx_info[pending_idx].req; |
f942dc25 IC |
1563 | |
1564 | /* Check the remap error code. */ | |
7376419a | 1565 | if (unlikely(xenvif_tx_check_gop(vif, skb, &gop))) { |
f942dc25 IC |
1566 | netdev_dbg(vif->dev, "netback grant failed.\n"); |
1567 | skb_shinfo(skb)->nr_frags = 0; | |
1568 | kfree_skb(skb); | |
1569 | continue; | |
1570 | } | |
1571 | ||
1572 | data_len = skb->len; | |
1573 | memcpy(skb->data, | |
b3f980bd | 1574 | (void *)(idx_to_kaddr(vif, pending_idx)|txp->offset), |
f942dc25 IC |
1575 | data_len); |
1576 | if (data_len < txp->size) { | |
1577 | /* Append the packet payload as a fragment. */ | |
1578 | txp->offset += data_len; | |
1579 | txp->size -= data_len; | |
1580 | } else { | |
1581 | /* Schedule a response immediately. */ | |
7376419a WL |
1582 | xenvif_idx_release(vif, pending_idx, |
1583 | XEN_NETIF_RSP_OKAY); | |
f942dc25 IC |
1584 | } |
1585 | ||
1586 | if (txp->flags & XEN_NETTXF_csum_blank) | |
1587 | skb->ip_summed = CHECKSUM_PARTIAL; | |
1588 | else if (txp->flags & XEN_NETTXF_data_validated) | |
1589 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
1590 | ||
7376419a | 1591 | xenvif_fill_frags(vif, skb); |
f942dc25 | 1592 | |
2eba61d5 | 1593 | if (skb_is_nonlinear(skb) && skb_headlen(skb) < PKT_PROT_LEN) { |
f942dc25 IC |
1594 | int target = min_t(int, skb->len, PKT_PROT_LEN); |
1595 | __pskb_pull_tail(skb, target - skb_headlen(skb)); | |
1596 | } | |
1597 | ||
1598 | skb->dev = vif->dev; | |
1599 | skb->protocol = eth_type_trans(skb, skb->dev); | |
f9ca8f74 | 1600 | skb_reset_network_header(skb); |
f942dc25 IC |
1601 | |
1602 | if (checksum_setup(vif, skb)) { | |
1603 | netdev_dbg(vif->dev, | |
1604 | "Can't setup checksum in net_tx_action\n"); | |
1605 | kfree_skb(skb); | |
1606 | continue; | |
1607 | } | |
1608 | ||
40893fd0 | 1609 | skb_probe_transport_header(skb, 0); |
f9ca8f74 | 1610 | |
f942dc25 IC |
1611 | vif->dev->stats.rx_bytes += skb->len; |
1612 | vif->dev->stats.rx_packets++; | |
1613 | ||
b3f980bd WL |
1614 | work_done++; |
1615 | ||
1616 | netif_receive_skb(skb); | |
f942dc25 | 1617 | } |
b3f980bd WL |
1618 | |
1619 | return work_done; | |
f942dc25 IC |
1620 | } |
1621 | ||
1622 | /* Called after netfront has transmitted */ | |
7376419a | 1623 | int xenvif_tx_action(struct xenvif *vif, int budget) |
f942dc25 IC |
1624 | { |
1625 | unsigned nr_gops; | |
b3f980bd | 1626 | int work_done; |
f942dc25 | 1627 | |
b3f980bd WL |
1628 | if (unlikely(!tx_work_todo(vif))) |
1629 | return 0; | |
1630 | ||
7376419a | 1631 | nr_gops = xenvif_tx_build_gops(vif); |
f942dc25 IC |
1632 | |
1633 | if (nr_gops == 0) | |
b3f980bd WL |
1634 | return 0; |
1635 | ||
1636 | gnttab_batch_copy(vif->tx_copy_ops, nr_gops); | |
f942dc25 | 1637 | |
7376419a | 1638 | work_done = xenvif_tx_submit(vif, nr_gops); |
f942dc25 | 1639 | |
b3f980bd | 1640 | return work_done; |
f942dc25 IC |
1641 | } |
1642 | ||
7376419a WL |
1643 | static void xenvif_idx_release(struct xenvif *vif, u16 pending_idx, |
1644 | u8 status) | |
f942dc25 | 1645 | { |
f942dc25 | 1646 | struct pending_tx_info *pending_tx_info; |
2810e5b9 WL |
1647 | pending_ring_idx_t head; |
1648 | u16 peek; /* peek into next tx request */ | |
1649 | ||
b3f980bd | 1650 | BUG_ON(vif->mmap_pages[pending_idx] == (void *)(~0UL)); |
f942dc25 IC |
1651 | |
1652 | /* Already complete? */ | |
b3f980bd | 1653 | if (vif->mmap_pages[pending_idx] == NULL) |
f942dc25 IC |
1654 | return; |
1655 | ||
b3f980bd | 1656 | pending_tx_info = &vif->pending_tx_info[pending_idx]; |
f942dc25 | 1657 | |
2810e5b9 | 1658 | head = pending_tx_info->head; |
f942dc25 | 1659 | |
b3f980bd WL |
1660 | BUG_ON(!pending_tx_is_head(vif, head)); |
1661 | BUG_ON(vif->pending_ring[pending_index(head)] != pending_idx); | |
f942dc25 | 1662 | |
2810e5b9 WL |
1663 | do { |
1664 | pending_ring_idx_t index; | |
1665 | pending_ring_idx_t idx = pending_index(head); | |
b3f980bd | 1666 | u16 info_idx = vif->pending_ring[idx]; |
f942dc25 | 1667 | |
b3f980bd | 1668 | pending_tx_info = &vif->pending_tx_info[info_idx]; |
2810e5b9 WL |
1669 | make_tx_response(vif, &pending_tx_info->req, status); |
1670 | ||
1671 | /* Setting any number other than | |
1672 | * INVALID_PENDING_RING_IDX indicates this slot is | |
1673 | * starting a new packet / ending a previous packet. | |
1674 | */ | |
1675 | pending_tx_info->head = 0; | |
1676 | ||
b3f980bd WL |
1677 | index = pending_index(vif->pending_prod++); |
1678 | vif->pending_ring[index] = vif->pending_ring[info_idx]; | |
f942dc25 | 1679 | |
b3f980bd | 1680 | peek = vif->pending_ring[pending_index(++head)]; |
2810e5b9 | 1681 | |
b3f980bd | 1682 | } while (!pending_tx_is_head(vif, peek)); |
2810e5b9 | 1683 | |
b3f980bd WL |
1684 | put_page(vif->mmap_pages[pending_idx]); |
1685 | vif->mmap_pages[pending_idx] = NULL; | |
f942dc25 IC |
1686 | } |
1687 | ||
2810e5b9 | 1688 | |
f942dc25 IC |
1689 | static void make_tx_response(struct xenvif *vif, |
1690 | struct xen_netif_tx_request *txp, | |
1691 | s8 st) | |
1692 | { | |
1693 | RING_IDX i = vif->tx.rsp_prod_pvt; | |
1694 | struct xen_netif_tx_response *resp; | |
1695 | int notify; | |
1696 | ||
1697 | resp = RING_GET_RESPONSE(&vif->tx, i); | |
1698 | resp->id = txp->id; | |
1699 | resp->status = st; | |
1700 | ||
1701 | if (txp->flags & XEN_NETTXF_extra_info) | |
1702 | RING_GET_RESPONSE(&vif->tx, ++i)->status = XEN_NETIF_RSP_NULL; | |
1703 | ||
1704 | vif->tx.rsp_prod_pvt = ++i; | |
1705 | RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&vif->tx, notify); | |
1706 | if (notify) | |
e1f00a69 | 1707 | notify_remote_via_irq(vif->tx_irq); |
f942dc25 IC |
1708 | } |
1709 | ||
1710 | static struct xen_netif_rx_response *make_rx_response(struct xenvif *vif, | |
1711 | u16 id, | |
1712 | s8 st, | |
1713 | u16 offset, | |
1714 | u16 size, | |
1715 | u16 flags) | |
1716 | { | |
1717 | RING_IDX i = vif->rx.rsp_prod_pvt; | |
1718 | struct xen_netif_rx_response *resp; | |
1719 | ||
1720 | resp = RING_GET_RESPONSE(&vif->rx, i); | |
1721 | resp->offset = offset; | |
1722 | resp->flags = flags; | |
1723 | resp->id = id; | |
1724 | resp->status = (s16)size; | |
1725 | if (st < 0) | |
1726 | resp->status = (s16)st; | |
1727 | ||
1728 | vif->rx.rsp_prod_pvt = ++i; | |
1729 | ||
1730 | return resp; | |
1731 | } | |
1732 | ||
b3f980bd | 1733 | static inline int rx_work_todo(struct xenvif *vif) |
f942dc25 | 1734 | { |
b3f980bd | 1735 | return !skb_queue_empty(&vif->rx_queue); |
f942dc25 IC |
1736 | } |
1737 | ||
b3f980bd | 1738 | static inline int tx_work_todo(struct xenvif *vif) |
f942dc25 IC |
1739 | { |
1740 | ||
b3f980bd WL |
1741 | if (likely(RING_HAS_UNCONSUMED_REQUESTS(&vif->tx)) && |
1742 | (nr_pending_reqs(vif) + XEN_NETBK_LEGACY_SLOTS_MAX | |
1743 | < MAX_PENDING_REQS)) | |
f942dc25 IC |
1744 | return 1; |
1745 | ||
1746 | return 0; | |
1747 | } | |
1748 | ||
7376419a | 1749 | void xenvif_unmap_frontend_rings(struct xenvif *vif) |
f942dc25 | 1750 | { |
c9d63699 DV |
1751 | if (vif->tx.sring) |
1752 | xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif), | |
1753 | vif->tx.sring); | |
1754 | if (vif->rx.sring) | |
1755 | xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif), | |
1756 | vif->rx.sring); | |
f942dc25 IC |
1757 | } |
1758 | ||
7376419a WL |
1759 | int xenvif_map_frontend_rings(struct xenvif *vif, |
1760 | grant_ref_t tx_ring_ref, | |
1761 | grant_ref_t rx_ring_ref) | |
f942dc25 | 1762 | { |
c9d63699 | 1763 | void *addr; |
f942dc25 IC |
1764 | struct xen_netif_tx_sring *txs; |
1765 | struct xen_netif_rx_sring *rxs; | |
1766 | ||
1767 | int err = -ENOMEM; | |
1768 | ||
c9d63699 DV |
1769 | err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(vif), |
1770 | tx_ring_ref, &addr); | |
1771 | if (err) | |
f942dc25 IC |
1772 | goto err; |
1773 | ||
c9d63699 | 1774 | txs = (struct xen_netif_tx_sring *)addr; |
f942dc25 IC |
1775 | BACK_RING_INIT(&vif->tx, txs, PAGE_SIZE); |
1776 | ||
c9d63699 DV |
1777 | err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(vif), |
1778 | rx_ring_ref, &addr); | |
1779 | if (err) | |
f942dc25 | 1780 | goto err; |
f942dc25 | 1781 | |
c9d63699 | 1782 | rxs = (struct xen_netif_rx_sring *)addr; |
f942dc25 IC |
1783 | BACK_RING_INIT(&vif->rx, rxs, PAGE_SIZE); |
1784 | ||
c9d63699 DV |
1785 | vif->rx_req_cons_peek = 0; |
1786 | ||
f942dc25 IC |
1787 | return 0; |
1788 | ||
1789 | err: | |
7376419a | 1790 | xenvif_unmap_frontend_rings(vif); |
f942dc25 IC |
1791 | return err; |
1792 | } | |
1793 | ||
7376419a | 1794 | int xenvif_kthread(void *data) |
b3f980bd WL |
1795 | { |
1796 | struct xenvif *vif = data; | |
1797 | ||
1798 | while (!kthread_should_stop()) { | |
1799 | wait_event_interruptible(vif->wq, | |
1800 | rx_work_todo(vif) || | |
1801 | kthread_should_stop()); | |
1802 | if (kthread_should_stop()) | |
1803 | break; | |
1804 | ||
1805 | if (rx_work_todo(vif)) | |
7376419a | 1806 | xenvif_rx_action(vif); |
b3f980bd WL |
1807 | |
1808 | cond_resched(); | |
1809 | } | |
1810 | ||
1811 | return 0; | |
1812 | } | |
1813 | ||
f942dc25 IC |
1814 | static int __init netback_init(void) |
1815 | { | |
f942dc25 | 1816 | int rc = 0; |
f942dc25 | 1817 | |
2a14b244 | 1818 | if (!xen_domain()) |
f942dc25 IC |
1819 | return -ENODEV; |
1820 | ||
37641494 | 1821 | if (fatal_skb_slots < XEN_NETBK_LEGACY_SLOTS_MAX) { |
383eda32 JP |
1822 | pr_info("fatal_skb_slots too small (%d), bump it to XEN_NETBK_LEGACY_SLOTS_MAX (%d)\n", |
1823 | fatal_skb_slots, XEN_NETBK_LEGACY_SLOTS_MAX); | |
37641494 | 1824 | fatal_skb_slots = XEN_NETBK_LEGACY_SLOTS_MAX; |
2810e5b9 WL |
1825 | } |
1826 | ||
f942dc25 IC |
1827 | rc = xenvif_xenbus_init(); |
1828 | if (rc) | |
1829 | goto failed_init; | |
1830 | ||
1831 | return 0; | |
1832 | ||
1833 | failed_init: | |
f942dc25 | 1834 | return rc; |
f942dc25 IC |
1835 | } |
1836 | ||
1837 | module_init(netback_init); | |
1838 | ||
b103f358 WL |
1839 | static void __exit netback_fini(void) |
1840 | { | |
b103f358 | 1841 | xenvif_xenbus_fini(); |
b103f358 WL |
1842 | } |
1843 | module_exit(netback_fini); | |
1844 | ||
f942dc25 | 1845 | MODULE_LICENSE("Dual BSD/GPL"); |
f984cec6 | 1846 | MODULE_ALIAS("xen-backend:vif"); |