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 | ||
ca2f09f2 | 141 | bool xenvif_rx_ring_slots_available(struct xenvif *vif, int needed) |
f942dc25 | 142 | { |
ca2f09f2 | 143 | RING_IDX prod, cons; |
f942dc25 | 144 | |
ca2f09f2 PD |
145 | do { |
146 | prod = vif->rx.sring->req_prod; | |
147 | cons = vif->rx.req_cons; | |
f942dc25 | 148 | |
ca2f09f2 PD |
149 | if (prod - cons >= needed) |
150 | return true; | |
f942dc25 | 151 | |
ca2f09f2 | 152 | vif->rx.sring->req_event = prod + 1; |
f942dc25 | 153 | |
ca2f09f2 PD |
154 | /* Make sure event is visible before we check prod |
155 | * again. | |
156 | */ | |
157 | mb(); | |
158 | } while (vif->rx.sring->req_prod != prod); | |
f942dc25 | 159 | |
ca2f09f2 | 160 | return false; |
f942dc25 IC |
161 | } |
162 | ||
163 | /* | |
164 | * Returns true if we should start a new receive buffer instead of | |
165 | * adding 'size' bytes to a buffer which currently contains 'offset' | |
166 | * bytes. | |
167 | */ | |
168 | static bool start_new_rx_buffer(int offset, unsigned long size, int head) | |
169 | { | |
170 | /* simple case: we have completely filled the current buffer. */ | |
171 | if (offset == MAX_BUFFER_OFFSET) | |
172 | return true; | |
173 | ||
174 | /* | |
175 | * complex case: start a fresh buffer if the current frag | |
176 | * would overflow the current buffer but only if: | |
177 | * (i) this frag would fit completely in the next buffer | |
178 | * and (ii) there is already some data in the current buffer | |
179 | * and (iii) this is not the head buffer. | |
180 | * | |
181 | * Where: | |
182 | * - (i) stops us splitting a frag into two copies | |
183 | * unless the frag is too large for a single buffer. | |
184 | * - (ii) stops us from leaving a buffer pointlessly empty. | |
185 | * - (iii) stops us leaving the first buffer | |
186 | * empty. Strictly speaking this is already covered | |
187 | * by (ii) but is explicitly checked because | |
188 | * netfront relies on the first buffer being | |
189 | * non-empty and can crash otherwise. | |
190 | * | |
191 | * This means we will effectively linearise small | |
192 | * frags but do not needlessly split large buffers | |
193 | * into multiple copies tend to give large frags their | |
194 | * own buffers as before. | |
195 | */ | |
196 | if ((offset + size > MAX_BUFFER_OFFSET) && | |
197 | (size <= MAX_BUFFER_OFFSET) && offset && !head) | |
198 | return true; | |
199 | ||
200 | return false; | |
201 | } | |
202 | ||
f942dc25 IC |
203 | struct netrx_pending_operations { |
204 | unsigned copy_prod, copy_cons; | |
205 | unsigned meta_prod, meta_cons; | |
206 | struct gnttab_copy *copy; | |
b3f980bd | 207 | struct xenvif_rx_meta *meta; |
f942dc25 IC |
208 | int copy_off; |
209 | grant_ref_t copy_gref; | |
210 | }; | |
211 | ||
b3f980bd WL |
212 | static struct xenvif_rx_meta *get_next_rx_buffer(struct xenvif *vif, |
213 | struct netrx_pending_operations *npo) | |
f942dc25 | 214 | { |
b3f980bd | 215 | struct xenvif_rx_meta *meta; |
f942dc25 IC |
216 | struct xen_netif_rx_request *req; |
217 | ||
218 | req = RING_GET_REQUEST(&vif->rx, vif->rx.req_cons++); | |
219 | ||
220 | meta = npo->meta + npo->meta_prod++; | |
82cada22 | 221 | meta->gso_type = XEN_NETIF_GSO_TYPE_NONE; |
f942dc25 IC |
222 | meta->gso_size = 0; |
223 | meta->size = 0; | |
224 | meta->id = req->id; | |
225 | ||
226 | npo->copy_off = 0; | |
227 | npo->copy_gref = req->gref; | |
228 | ||
229 | return meta; | |
230 | } | |
231 | ||
33bc801d WL |
232 | /* |
233 | * Set up the grant operations for this fragment. If it's a flipping | |
234 | * interface, we also set up the unmap request from here. | |
235 | */ | |
7376419a WL |
236 | static void xenvif_gop_frag_copy(struct xenvif *vif, struct sk_buff *skb, |
237 | struct netrx_pending_operations *npo, | |
238 | struct page *page, unsigned long size, | |
33bc801d | 239 | unsigned long offset, int *head) |
f942dc25 IC |
240 | { |
241 | struct gnttab_copy *copy_gop; | |
b3f980bd | 242 | struct xenvif_rx_meta *meta; |
f942dc25 | 243 | unsigned long bytes; |
82cada22 | 244 | int gso_type; |
f942dc25 IC |
245 | |
246 | /* Data must not cross a page boundary. */ | |
6a8ed462 | 247 | BUG_ON(size + offset > PAGE_SIZE<<compound_order(page)); |
f942dc25 IC |
248 | |
249 | meta = npo->meta + npo->meta_prod - 1; | |
250 | ||
6a8ed462 IC |
251 | /* Skip unused frames from start of page */ |
252 | page += offset >> PAGE_SHIFT; | |
253 | offset &= ~PAGE_MASK; | |
254 | ||
f942dc25 | 255 | while (size > 0) { |
6a8ed462 | 256 | BUG_ON(offset >= PAGE_SIZE); |
f942dc25 IC |
257 | BUG_ON(npo->copy_off > MAX_BUFFER_OFFSET); |
258 | ||
6a8ed462 IC |
259 | bytes = PAGE_SIZE - offset; |
260 | ||
261 | if (bytes > size) | |
262 | bytes = size; | |
263 | ||
33bc801d | 264 | if (start_new_rx_buffer(npo->copy_off, bytes, *head)) { |
f942dc25 IC |
265 | /* |
266 | * Netfront requires there to be some data in the head | |
267 | * buffer. | |
268 | */ | |
33bc801d | 269 | BUG_ON(*head); |
f942dc25 IC |
270 | |
271 | meta = get_next_rx_buffer(vif, npo); | |
272 | } | |
273 | ||
f942dc25 IC |
274 | if (npo->copy_off + bytes > MAX_BUFFER_OFFSET) |
275 | bytes = MAX_BUFFER_OFFSET - npo->copy_off; | |
276 | ||
277 | copy_gop = npo->copy + npo->copy_prod++; | |
278 | copy_gop->flags = GNTCOPY_dest_gref; | |
b3f980bd WL |
279 | copy_gop->len = bytes; |
280 | ||
43e9d194 WL |
281 | copy_gop->source.domid = DOMID_SELF; |
282 | copy_gop->source.u.gmfn = virt_to_mfn(page_address(page)); | |
f942dc25 | 283 | copy_gop->source.offset = offset; |
f942dc25 | 284 | |
b3f980bd | 285 | copy_gop->dest.domid = vif->domid; |
f942dc25 IC |
286 | copy_gop->dest.offset = npo->copy_off; |
287 | copy_gop->dest.u.ref = npo->copy_gref; | |
f942dc25 IC |
288 | |
289 | npo->copy_off += bytes; | |
290 | meta->size += bytes; | |
291 | ||
292 | offset += bytes; | |
293 | size -= bytes; | |
294 | ||
6a8ed462 IC |
295 | /* Next frame */ |
296 | if (offset == PAGE_SIZE && size) { | |
297 | BUG_ON(!PageCompound(page)); | |
298 | page++; | |
299 | offset = 0; | |
300 | } | |
301 | ||
f942dc25 | 302 | /* Leave a gap for the GSO descriptor. */ |
82cada22 PD |
303 | if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) |
304 | gso_type = XEN_NETIF_GSO_TYPE_TCPV4; | |
305 | else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) | |
306 | gso_type = XEN_NETIF_GSO_TYPE_TCPV6; | |
307 | else | |
308 | gso_type = XEN_NETIF_GSO_TYPE_NONE; | |
309 | ||
310 | if (*head && ((1 << gso_type) & vif->gso_mask)) | |
f942dc25 IC |
311 | vif->rx.req_cons++; |
312 | ||
33bc801d | 313 | *head = 0; /* There must be something in this buffer now. */ |
f942dc25 IC |
314 | |
315 | } | |
316 | } | |
317 | ||
318 | /* | |
319 | * Prepare an SKB to be transmitted to the frontend. | |
320 | * | |
321 | * This function is responsible for allocating grant operations, meta | |
322 | * structures, etc. | |
323 | * | |
324 | * It returns the number of meta structures consumed. The number of | |
325 | * ring slots used is always equal to the number of meta slots used | |
326 | * plus the number of GSO descriptors used. Currently, we use either | |
327 | * zero GSO descriptors (for non-GSO packets) or one descriptor (for | |
328 | * frontend-side LRO). | |
329 | */ | |
7376419a WL |
330 | static int xenvif_gop_skb(struct sk_buff *skb, |
331 | struct netrx_pending_operations *npo) | |
f942dc25 IC |
332 | { |
333 | struct xenvif *vif = netdev_priv(skb->dev); | |
334 | int nr_frags = skb_shinfo(skb)->nr_frags; | |
335 | int i; | |
336 | struct xen_netif_rx_request *req; | |
b3f980bd | 337 | struct xenvif_rx_meta *meta; |
f942dc25 | 338 | unsigned char *data; |
33bc801d | 339 | int head = 1; |
f942dc25 | 340 | int old_meta_prod; |
82cada22 PD |
341 | int gso_type; |
342 | int gso_size; | |
f942dc25 IC |
343 | |
344 | old_meta_prod = npo->meta_prod; | |
345 | ||
82cada22 PD |
346 | if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) { |
347 | gso_type = XEN_NETIF_GSO_TYPE_TCPV4; | |
348 | gso_size = skb_shinfo(skb)->gso_size; | |
349 | } else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) { | |
350 | gso_type = XEN_NETIF_GSO_TYPE_TCPV6; | |
351 | gso_size = skb_shinfo(skb)->gso_size; | |
352 | } else { | |
353 | gso_type = XEN_NETIF_GSO_TYPE_NONE; | |
354 | gso_size = 0; | |
355 | } | |
356 | ||
f942dc25 | 357 | /* Set up a GSO prefix descriptor, if necessary */ |
a3314f3d | 358 | if ((1 << gso_type) & vif->gso_prefix_mask) { |
f942dc25 IC |
359 | req = RING_GET_REQUEST(&vif->rx, vif->rx.req_cons++); |
360 | meta = npo->meta + npo->meta_prod++; | |
82cada22 PD |
361 | meta->gso_type = gso_type; |
362 | meta->gso_size = gso_size; | |
f942dc25 IC |
363 | meta->size = 0; |
364 | meta->id = req->id; | |
365 | } | |
366 | ||
367 | req = RING_GET_REQUEST(&vif->rx, vif->rx.req_cons++); | |
368 | meta = npo->meta + npo->meta_prod++; | |
369 | ||
82cada22 PD |
370 | if ((1 << gso_type) & vif->gso_mask) { |
371 | meta->gso_type = gso_type; | |
372 | meta->gso_size = gso_size; | |
373 | } else { | |
374 | meta->gso_type = XEN_NETIF_GSO_TYPE_NONE; | |
f942dc25 | 375 | meta->gso_size = 0; |
82cada22 | 376 | } |
f942dc25 IC |
377 | |
378 | meta->size = 0; | |
379 | meta->id = req->id; | |
380 | npo->copy_off = 0; | |
381 | npo->copy_gref = req->gref; | |
382 | ||
383 | data = skb->data; | |
384 | while (data < skb_tail_pointer(skb)) { | |
385 | unsigned int offset = offset_in_page(data); | |
386 | unsigned int len = PAGE_SIZE - offset; | |
387 | ||
388 | if (data + len > skb_tail_pointer(skb)) | |
389 | len = skb_tail_pointer(skb) - data; | |
390 | ||
7376419a | 391 | xenvif_gop_frag_copy(vif, skb, npo, |
33bc801d | 392 | virt_to_page(data), len, offset, &head); |
f942dc25 IC |
393 | data += len; |
394 | } | |
395 | ||
396 | for (i = 0; i < nr_frags; i++) { | |
7376419a WL |
397 | xenvif_gop_frag_copy(vif, skb, npo, |
398 | skb_frag_page(&skb_shinfo(skb)->frags[i]), | |
399 | skb_frag_size(&skb_shinfo(skb)->frags[i]), | |
400 | skb_shinfo(skb)->frags[i].page_offset, | |
33bc801d | 401 | &head); |
f942dc25 IC |
402 | } |
403 | ||
404 | return npo->meta_prod - old_meta_prod; | |
405 | } | |
406 | ||
407 | /* | |
7376419a | 408 | * This is a twin to xenvif_gop_skb. Assume that xenvif_gop_skb was |
f942dc25 IC |
409 | * used to set up the operations on the top of |
410 | * netrx_pending_operations, which have since been done. Check that | |
411 | * they didn't give any errors and advance over them. | |
412 | */ | |
7376419a WL |
413 | static int xenvif_check_gop(struct xenvif *vif, int nr_meta_slots, |
414 | struct netrx_pending_operations *npo) | |
f942dc25 IC |
415 | { |
416 | struct gnttab_copy *copy_op; | |
417 | int status = XEN_NETIF_RSP_OKAY; | |
418 | int i; | |
419 | ||
420 | for (i = 0; i < nr_meta_slots; i++) { | |
421 | copy_op = npo->copy + npo->copy_cons++; | |
422 | if (copy_op->status != GNTST_okay) { | |
423 | netdev_dbg(vif->dev, | |
424 | "Bad status %d from copy to DOM%d.\n", | |
425 | copy_op->status, vif->domid); | |
426 | status = XEN_NETIF_RSP_ERROR; | |
427 | } | |
428 | } | |
429 | ||
430 | return status; | |
431 | } | |
432 | ||
7376419a WL |
433 | static void xenvif_add_frag_responses(struct xenvif *vif, int status, |
434 | struct xenvif_rx_meta *meta, | |
435 | int nr_meta_slots) | |
f942dc25 IC |
436 | { |
437 | int i; | |
438 | unsigned long offset; | |
439 | ||
440 | /* No fragments used */ | |
441 | if (nr_meta_slots <= 1) | |
442 | return; | |
443 | ||
444 | nr_meta_slots--; | |
445 | ||
446 | for (i = 0; i < nr_meta_slots; i++) { | |
447 | int flags; | |
448 | if (i == nr_meta_slots - 1) | |
449 | flags = 0; | |
450 | else | |
451 | flags = XEN_NETRXF_more_data; | |
452 | ||
453 | offset = 0; | |
454 | make_rx_response(vif, meta[i].id, status, offset, | |
455 | meta[i].size, flags); | |
456 | } | |
457 | } | |
458 | ||
33bc801d WL |
459 | struct skb_cb_overlay { |
460 | int meta_slots_used; | |
461 | }; | |
462 | ||
ca2f09f2 | 463 | void xenvif_kick_thread(struct xenvif *vif) |
b3f980bd WL |
464 | { |
465 | wake_up(&vif->wq); | |
466 | } | |
467 | ||
ca2f09f2 | 468 | static void xenvif_rx_action(struct xenvif *vif) |
f942dc25 | 469 | { |
f942dc25 | 470 | s8 status; |
e1f00a69 | 471 | u16 flags; |
f942dc25 IC |
472 | struct xen_netif_rx_response *resp; |
473 | struct sk_buff_head rxq; | |
474 | struct sk_buff *skb; | |
475 | LIST_HEAD(notify); | |
476 | int ret; | |
f942dc25 IC |
477 | unsigned long offset; |
478 | struct skb_cb_overlay *sco; | |
11b57f90 PD |
479 | bool need_to_notify = false; |
480 | bool ring_full = false; | |
f942dc25 IC |
481 | |
482 | struct netrx_pending_operations npo = { | |
b3f980bd WL |
483 | .copy = vif->grant_copy_op, |
484 | .meta = vif->meta, | |
f942dc25 IC |
485 | }; |
486 | ||
487 | skb_queue_head_init(&rxq); | |
488 | ||
b3f980bd | 489 | while ((skb = skb_dequeue(&vif->rx_queue)) != NULL) { |
ca2f09f2 PD |
490 | int max_slots_needed; |
491 | int i; | |
492 | ||
493 | /* We need a cheap worse case estimate for the number of | |
494 | * slots we'll use. | |
495 | */ | |
496 | ||
497 | max_slots_needed = DIV_ROUND_UP(offset_in_page(skb->data) + | |
498 | skb_headlen(skb), | |
499 | PAGE_SIZE); | |
500 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { | |
501 | unsigned int size; | |
502 | size = skb_frag_size(&skb_shinfo(skb)->frags[i]); | |
503 | max_slots_needed += DIV_ROUND_UP(size, PAGE_SIZE); | |
504 | } | |
505 | if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4 || | |
506 | skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) | |
507 | max_slots_needed++; | |
508 | ||
509 | /* If the skb may not fit then bail out now */ | |
510 | if (!xenvif_rx_ring_slots_available(vif, max_slots_needed)) { | |
511 | skb_queue_head(&vif->rx_queue, skb); | |
11b57f90 PD |
512 | need_to_notify = true; |
513 | ring_full = true; | |
ca2f09f2 PD |
514 | break; |
515 | } | |
f942dc25 IC |
516 | |
517 | sco = (struct skb_cb_overlay *)skb->cb; | |
7376419a | 518 | sco->meta_slots_used = xenvif_gop_skb(skb, &npo); |
ca2f09f2 | 519 | BUG_ON(sco->meta_slots_used > max_slots_needed); |
f942dc25 IC |
520 | |
521 | __skb_queue_tail(&rxq, skb); | |
f942dc25 IC |
522 | } |
523 | ||
b3f980bd | 524 | BUG_ON(npo.meta_prod > ARRAY_SIZE(vif->meta)); |
f942dc25 | 525 | |
11b57f90 PD |
526 | vif->rx_queue_stopped = !npo.copy_prod && ring_full; |
527 | ||
f942dc25 | 528 | if (!npo.copy_prod) |
ca2f09f2 | 529 | goto done; |
f942dc25 | 530 | |
ac3d5ac2 | 531 | BUG_ON(npo.copy_prod > MAX_GRANT_COPY_OPS); |
b3f980bd | 532 | gnttab_batch_copy(vif->grant_copy_op, npo.copy_prod); |
f942dc25 IC |
533 | |
534 | while ((skb = __skb_dequeue(&rxq)) != NULL) { | |
535 | sco = (struct skb_cb_overlay *)skb->cb; | |
536 | ||
82cada22 PD |
537 | if ((1 << vif->meta[npo.meta_cons].gso_type) & |
538 | vif->gso_prefix_mask) { | |
f942dc25 | 539 | resp = RING_GET_RESPONSE(&vif->rx, |
b3f980bd | 540 | vif->rx.rsp_prod_pvt++); |
f942dc25 IC |
541 | |
542 | resp->flags = XEN_NETRXF_gso_prefix | XEN_NETRXF_more_data; | |
543 | ||
b3f980bd WL |
544 | resp->offset = vif->meta[npo.meta_cons].gso_size; |
545 | resp->id = vif->meta[npo.meta_cons].id; | |
f942dc25 IC |
546 | resp->status = sco->meta_slots_used; |
547 | ||
548 | npo.meta_cons++; | |
549 | sco->meta_slots_used--; | |
550 | } | |
551 | ||
552 | ||
553 | vif->dev->stats.tx_bytes += skb->len; | |
554 | vif->dev->stats.tx_packets++; | |
555 | ||
7376419a | 556 | status = xenvif_check_gop(vif, sco->meta_slots_used, &npo); |
f942dc25 IC |
557 | |
558 | if (sco->meta_slots_used == 1) | |
559 | flags = 0; | |
560 | else | |
561 | flags = XEN_NETRXF_more_data; | |
562 | ||
563 | if (skb->ip_summed == CHECKSUM_PARTIAL) /* local packet? */ | |
564 | flags |= XEN_NETRXF_csum_blank | XEN_NETRXF_data_validated; | |
565 | else if (skb->ip_summed == CHECKSUM_UNNECESSARY) | |
566 | /* remote but checksummed. */ | |
567 | flags |= XEN_NETRXF_data_validated; | |
568 | ||
569 | offset = 0; | |
b3f980bd | 570 | resp = make_rx_response(vif, vif->meta[npo.meta_cons].id, |
f942dc25 | 571 | status, offset, |
b3f980bd | 572 | vif->meta[npo.meta_cons].size, |
f942dc25 IC |
573 | flags); |
574 | ||
82cada22 PD |
575 | if ((1 << vif->meta[npo.meta_cons].gso_type) & |
576 | vif->gso_mask) { | |
f942dc25 IC |
577 | struct xen_netif_extra_info *gso = |
578 | (struct xen_netif_extra_info *) | |
579 | RING_GET_RESPONSE(&vif->rx, | |
580 | vif->rx.rsp_prod_pvt++); | |
581 | ||
582 | resp->flags |= XEN_NETRXF_extra_info; | |
583 | ||
82cada22 | 584 | gso->u.gso.type = vif->meta[npo.meta_cons].gso_type; |
b3f980bd | 585 | gso->u.gso.size = vif->meta[npo.meta_cons].gso_size; |
f942dc25 IC |
586 | gso->u.gso.pad = 0; |
587 | gso->u.gso.features = 0; | |
588 | ||
589 | gso->type = XEN_NETIF_EXTRA_TYPE_GSO; | |
590 | gso->flags = 0; | |
591 | } | |
592 | ||
7376419a WL |
593 | xenvif_add_frag_responses(vif, status, |
594 | vif->meta + npo.meta_cons + 1, | |
595 | sco->meta_slots_used); | |
f942dc25 IC |
596 | |
597 | RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&vif->rx, ret); | |
f942dc25 | 598 | |
11b57f90 | 599 | need_to_notify |= !!ret; |
b3f980bd | 600 | |
f942dc25 IC |
601 | npo.meta_cons += sco->meta_slots_used; |
602 | dev_kfree_skb(skb); | |
603 | } | |
604 | ||
ca2f09f2 | 605 | done: |
b3f980bd | 606 | if (need_to_notify) |
e1f00a69 | 607 | notify_remote_via_irq(vif->rx_irq); |
f942dc25 IC |
608 | } |
609 | ||
7376419a | 610 | void xenvif_check_rx_xenvif(struct xenvif *vif) |
f942dc25 IC |
611 | { |
612 | int more_to_do; | |
613 | ||
614 | RING_FINAL_CHECK_FOR_REQUESTS(&vif->tx, more_to_do); | |
615 | ||
616 | if (more_to_do) | |
b3f980bd | 617 | napi_schedule(&vif->napi); |
f942dc25 IC |
618 | } |
619 | ||
620 | static void tx_add_credit(struct xenvif *vif) | |
621 | { | |
622 | unsigned long max_burst, max_credit; | |
623 | ||
624 | /* | |
625 | * Allow a burst big enough to transmit a jumbo packet of up to 128kB. | |
626 | * Otherwise the interface can seize up due to insufficient credit. | |
627 | */ | |
628 | max_burst = RING_GET_REQUEST(&vif->tx, vif->tx.req_cons)->size; | |
629 | max_burst = min(max_burst, 131072UL); | |
630 | max_burst = max(max_burst, vif->credit_bytes); | |
631 | ||
632 | /* Take care that adding a new chunk of credit doesn't wrap to zero. */ | |
633 | max_credit = vif->remaining_credit + vif->credit_bytes; | |
634 | if (max_credit < vif->remaining_credit) | |
635 | max_credit = ULONG_MAX; /* wrapped: clamp to ULONG_MAX */ | |
636 | ||
637 | vif->remaining_credit = min(max_credit, max_burst); | |
638 | } | |
639 | ||
640 | static void tx_credit_callback(unsigned long data) | |
641 | { | |
642 | struct xenvif *vif = (struct xenvif *)data; | |
643 | tx_add_credit(vif); | |
7376419a | 644 | xenvif_check_rx_xenvif(vif); |
f942dc25 IC |
645 | } |
646 | ||
7376419a WL |
647 | static void xenvif_tx_err(struct xenvif *vif, |
648 | struct xen_netif_tx_request *txp, RING_IDX end) | |
f942dc25 IC |
649 | { |
650 | RING_IDX cons = vif->tx.req_cons; | |
651 | ||
652 | do { | |
653 | make_tx_response(vif, txp, XEN_NETIF_RSP_ERROR); | |
b9149729 | 654 | if (cons == end) |
f942dc25 IC |
655 | break; |
656 | txp = RING_GET_REQUEST(&vif->tx, cons++); | |
657 | } while (1); | |
658 | vif->tx.req_cons = cons; | |
f942dc25 IC |
659 | } |
660 | ||
7376419a | 661 | static void xenvif_fatal_tx_err(struct xenvif *vif) |
48856286 IC |
662 | { |
663 | netdev_err(vif->dev, "fatal error; disabling device\n"); | |
664 | xenvif_carrier_off(vif); | |
48856286 IC |
665 | } |
666 | ||
7376419a WL |
667 | static int xenvif_count_requests(struct xenvif *vif, |
668 | struct xen_netif_tx_request *first, | |
669 | struct xen_netif_tx_request *txp, | |
670 | int work_to_do) | |
f942dc25 IC |
671 | { |
672 | RING_IDX cons = vif->tx.req_cons; | |
2810e5b9 WL |
673 | int slots = 0; |
674 | int drop_err = 0; | |
59ccb4eb | 675 | int more_data; |
f942dc25 IC |
676 | |
677 | if (!(first->flags & XEN_NETTXF_more_data)) | |
678 | return 0; | |
679 | ||
680 | do { | |
59ccb4eb WL |
681 | struct xen_netif_tx_request dropped_tx = { 0 }; |
682 | ||
2810e5b9 WL |
683 | if (slots >= work_to_do) { |
684 | netdev_err(vif->dev, | |
685 | "Asked for %d slots but exceeds this limit\n", | |
686 | work_to_do); | |
7376419a | 687 | xenvif_fatal_tx_err(vif); |
35876b5f | 688 | return -ENODATA; |
f942dc25 IC |
689 | } |
690 | ||
2810e5b9 WL |
691 | /* This guest is really using too many slots and |
692 | * considered malicious. | |
693 | */ | |
37641494 | 694 | if (unlikely(slots >= fatal_skb_slots)) { |
2810e5b9 WL |
695 | netdev_err(vif->dev, |
696 | "Malicious frontend using %d slots, threshold %u\n", | |
37641494 | 697 | slots, fatal_skb_slots); |
7376419a | 698 | xenvif_fatal_tx_err(vif); |
35876b5f | 699 | return -E2BIG; |
f942dc25 IC |
700 | } |
701 | ||
2810e5b9 | 702 | /* Xen network protocol had implicit dependency on |
37641494 WL |
703 | * MAX_SKB_FRAGS. XEN_NETBK_LEGACY_SLOTS_MAX is set to |
704 | * the historical MAX_SKB_FRAGS value 18 to honor the | |
705 | * same behavior as before. Any packet using more than | |
706 | * 18 slots but less than fatal_skb_slots slots is | |
707 | * dropped | |
2810e5b9 | 708 | */ |
37641494 | 709 | if (!drop_err && slots >= XEN_NETBK_LEGACY_SLOTS_MAX) { |
2810e5b9 WL |
710 | if (net_ratelimit()) |
711 | netdev_dbg(vif->dev, | |
712 | "Too many slots (%d) exceeding limit (%d), dropping packet\n", | |
37641494 | 713 | slots, XEN_NETBK_LEGACY_SLOTS_MAX); |
2810e5b9 WL |
714 | drop_err = -E2BIG; |
715 | } | |
716 | ||
59ccb4eb WL |
717 | if (drop_err) |
718 | txp = &dropped_tx; | |
719 | ||
2810e5b9 | 720 | memcpy(txp, RING_GET_REQUEST(&vif->tx, cons + slots), |
f942dc25 | 721 | sizeof(*txp)); |
03393fd5 WL |
722 | |
723 | /* If the guest submitted a frame >= 64 KiB then | |
724 | * first->size overflowed and following slots will | |
725 | * appear to be larger than the frame. | |
726 | * | |
727 | * This cannot be fatal error as there are buggy | |
728 | * frontends that do this. | |
729 | * | |
730 | * Consume all slots and drop the packet. | |
731 | */ | |
732 | if (!drop_err && txp->size > first->size) { | |
733 | if (net_ratelimit()) | |
734 | netdev_dbg(vif->dev, | |
735 | "Invalid tx request, slot size %u > remaining size %u\n", | |
736 | txp->size, first->size); | |
737 | drop_err = -EIO; | |
f942dc25 IC |
738 | } |
739 | ||
740 | first->size -= txp->size; | |
2810e5b9 | 741 | slots++; |
f942dc25 IC |
742 | |
743 | if (unlikely((txp->offset + txp->size) > PAGE_SIZE)) { | |
2810e5b9 | 744 | netdev_err(vif->dev, "Cross page boundary, txp->offset: %x, size: %u\n", |
f942dc25 | 745 | txp->offset, txp->size); |
7376419a | 746 | xenvif_fatal_tx_err(vif); |
35876b5f | 747 | return -EINVAL; |
f942dc25 | 748 | } |
59ccb4eb WL |
749 | |
750 | more_data = txp->flags & XEN_NETTXF_more_data; | |
751 | ||
752 | if (!drop_err) | |
753 | txp++; | |
754 | ||
755 | } while (more_data); | |
2810e5b9 WL |
756 | |
757 | if (drop_err) { | |
7376419a | 758 | xenvif_tx_err(vif, first, cons + slots); |
2810e5b9 WL |
759 | return drop_err; |
760 | } | |
761 | ||
762 | return slots; | |
f942dc25 IC |
763 | } |
764 | ||
7376419a WL |
765 | static struct page *xenvif_alloc_page(struct xenvif *vif, |
766 | u16 pending_idx) | |
f942dc25 IC |
767 | { |
768 | struct page *page; | |
b3f980bd WL |
769 | |
770 | page = alloc_page(GFP_ATOMIC|__GFP_COLD); | |
f942dc25 IC |
771 | if (!page) |
772 | return NULL; | |
b3f980bd WL |
773 | vif->mmap_pages[pending_idx] = page; |
774 | ||
f942dc25 IC |
775 | return page; |
776 | } | |
777 | ||
7376419a WL |
778 | static struct gnttab_copy *xenvif_get_requests(struct xenvif *vif, |
779 | struct sk_buff *skb, | |
780 | struct xen_netif_tx_request *txp, | |
781 | struct gnttab_copy *gop) | |
f942dc25 IC |
782 | { |
783 | struct skb_shared_info *shinfo = skb_shinfo(skb); | |
784 | skb_frag_t *frags = shinfo->frags; | |
ea066ad1 | 785 | u16 pending_idx = *((u16 *)skb->data); |
2810e5b9 WL |
786 | u16 head_idx = 0; |
787 | int slot, start; | |
788 | struct page *page; | |
789 | pending_ring_idx_t index, start_idx = 0; | |
790 | uint16_t dst_offset; | |
791 | unsigned int nr_slots; | |
792 | struct pending_tx_info *first = NULL; | |
793 | ||
794 | /* At this point shinfo->nr_frags is in fact the number of | |
37641494 | 795 | * slots, which can be as large as XEN_NETBK_LEGACY_SLOTS_MAX. |
2810e5b9 WL |
796 | */ |
797 | nr_slots = shinfo->nr_frags; | |
f942dc25 IC |
798 | |
799 | /* Skip first skb fragment if it is on same page as header fragment. */ | |
ea066ad1 | 800 | start = (frag_get_pending_idx(&shinfo->frags[0]) == pending_idx); |
f942dc25 | 801 | |
2810e5b9 WL |
802 | /* Coalesce tx requests, at this point the packet passed in |
803 | * should be <= 64K. Any packets larger than 64K have been | |
7376419a | 804 | * handled in xenvif_count_requests(). |
2810e5b9 WL |
805 | */ |
806 | for (shinfo->nr_frags = slot = start; slot < nr_slots; | |
807 | shinfo->nr_frags++) { | |
f942dc25 | 808 | struct pending_tx_info *pending_tx_info = |
b3f980bd | 809 | vif->pending_tx_info; |
f942dc25 | 810 | |
b3f980bd | 811 | page = alloc_page(GFP_ATOMIC|__GFP_COLD); |
f942dc25 | 812 | if (!page) |
4cc7c1cb | 813 | goto err; |
f942dc25 | 814 | |
2810e5b9 WL |
815 | dst_offset = 0; |
816 | first = NULL; | |
817 | while (dst_offset < PAGE_SIZE && slot < nr_slots) { | |
818 | gop->flags = GNTCOPY_source_gref; | |
819 | ||
820 | gop->source.u.ref = txp->gref; | |
821 | gop->source.domid = vif->domid; | |
822 | gop->source.offset = txp->offset; | |
823 | ||
824 | gop->dest.domid = DOMID_SELF; | |
825 | ||
826 | gop->dest.offset = dst_offset; | |
827 | gop->dest.u.gmfn = virt_to_mfn(page_address(page)); | |
828 | ||
829 | if (dst_offset + txp->size > PAGE_SIZE) { | |
830 | /* This page can only merge a portion | |
831 | * of tx request. Do not increment any | |
832 | * pointer / counter here. The txp | |
833 | * will be dealt with in future | |
834 | * rounds, eventually hitting the | |
835 | * `else` branch. | |
836 | */ | |
837 | gop->len = PAGE_SIZE - dst_offset; | |
838 | txp->offset += gop->len; | |
839 | txp->size -= gop->len; | |
840 | dst_offset += gop->len; /* quit loop */ | |
841 | } else { | |
842 | /* This tx request can be merged in the page */ | |
843 | gop->len = txp->size; | |
844 | dst_offset += gop->len; | |
845 | ||
b3f980bd | 846 | index = pending_index(vif->pending_cons++); |
2810e5b9 | 847 | |
b3f980bd | 848 | pending_idx = vif->pending_ring[index]; |
2810e5b9 WL |
849 | |
850 | memcpy(&pending_tx_info[pending_idx].req, txp, | |
851 | sizeof(*txp)); | |
2810e5b9 WL |
852 | |
853 | /* Poison these fields, corresponding | |
854 | * fields for head tx req will be set | |
855 | * to correct values after the loop. | |
856 | */ | |
b3f980bd | 857 | vif->mmap_pages[pending_idx] = (void *)(~0UL); |
2810e5b9 WL |
858 | pending_tx_info[pending_idx].head = |
859 | INVALID_PENDING_RING_IDX; | |
860 | ||
861 | if (!first) { | |
862 | first = &pending_tx_info[pending_idx]; | |
863 | start_idx = index; | |
864 | head_idx = pending_idx; | |
865 | } | |
866 | ||
867 | txp++; | |
868 | slot++; | |
869 | } | |
f942dc25 | 870 | |
2810e5b9 WL |
871 | gop++; |
872 | } | |
f942dc25 | 873 | |
2810e5b9 WL |
874 | first->req.offset = 0; |
875 | first->req.size = dst_offset; | |
876 | first->head = start_idx; | |
b3f980bd | 877 | vif->mmap_pages[head_idx] = page; |
2810e5b9 | 878 | frag_set_pending_idx(&frags[shinfo->nr_frags], head_idx); |
f942dc25 IC |
879 | } |
880 | ||
2810e5b9 WL |
881 | BUG_ON(shinfo->nr_frags > MAX_SKB_FRAGS); |
882 | ||
f942dc25 | 883 | return gop; |
4cc7c1cb IC |
884 | err: |
885 | /* Unwind, freeing all pages and sending error responses. */ | |
2810e5b9 | 886 | while (shinfo->nr_frags-- > start) { |
7376419a | 887 | xenvif_idx_release(vif, |
2810e5b9 WL |
888 | frag_get_pending_idx(&frags[shinfo->nr_frags]), |
889 | XEN_NETIF_RSP_ERROR); | |
4cc7c1cb IC |
890 | } |
891 | /* The head too, if necessary. */ | |
892 | if (start) | |
7376419a | 893 | xenvif_idx_release(vif, pending_idx, XEN_NETIF_RSP_ERROR); |
4cc7c1cb IC |
894 | |
895 | return NULL; | |
f942dc25 IC |
896 | } |
897 | ||
7376419a WL |
898 | static int xenvif_tx_check_gop(struct xenvif *vif, |
899 | struct sk_buff *skb, | |
900 | struct gnttab_copy **gopp) | |
f942dc25 IC |
901 | { |
902 | struct gnttab_copy *gop = *gopp; | |
ea066ad1 | 903 | u16 pending_idx = *((u16 *)skb->data); |
f942dc25 | 904 | struct skb_shared_info *shinfo = skb_shinfo(skb); |
2810e5b9 | 905 | struct pending_tx_info *tx_info; |
f942dc25 IC |
906 | int nr_frags = shinfo->nr_frags; |
907 | int i, err, start; | |
2810e5b9 | 908 | u16 peek; /* peek into next tx request */ |
f942dc25 IC |
909 | |
910 | /* Check status of header. */ | |
911 | err = gop->status; | |
7d5145d8 | 912 | if (unlikely(err)) |
7376419a | 913 | xenvif_idx_release(vif, pending_idx, XEN_NETIF_RSP_ERROR); |
f942dc25 IC |
914 | |
915 | /* Skip first skb fragment if it is on same page as header fragment. */ | |
ea066ad1 | 916 | start = (frag_get_pending_idx(&shinfo->frags[0]) == pending_idx); |
f942dc25 IC |
917 | |
918 | for (i = start; i < nr_frags; i++) { | |
919 | int j, newerr; | |
2810e5b9 | 920 | pending_ring_idx_t head; |
f942dc25 | 921 | |
ea066ad1 | 922 | pending_idx = frag_get_pending_idx(&shinfo->frags[i]); |
b3f980bd | 923 | tx_info = &vif->pending_tx_info[pending_idx]; |
2810e5b9 | 924 | head = tx_info->head; |
f942dc25 IC |
925 | |
926 | /* Check error status: if okay then remember grant handle. */ | |
2810e5b9 WL |
927 | do { |
928 | newerr = (++gop)->status; | |
929 | if (newerr) | |
930 | break; | |
b3f980bd WL |
931 | peek = vif->pending_ring[pending_index(++head)]; |
932 | } while (!pending_tx_is_head(vif, peek)); | |
2810e5b9 | 933 | |
f942dc25 IC |
934 | if (likely(!newerr)) { |
935 | /* Had a previous error? Invalidate this fragment. */ | |
936 | if (unlikely(err)) | |
7376419a WL |
937 | xenvif_idx_release(vif, pending_idx, |
938 | XEN_NETIF_RSP_OKAY); | |
f942dc25 IC |
939 | continue; |
940 | } | |
941 | ||
942 | /* Error on this fragment: respond to client with an error. */ | |
7376419a | 943 | xenvif_idx_release(vif, pending_idx, XEN_NETIF_RSP_ERROR); |
f942dc25 IC |
944 | |
945 | /* Not the first error? Preceding frags already invalidated. */ | |
946 | if (err) | |
947 | continue; | |
948 | ||
949 | /* First error: invalidate header and preceding fragments. */ | |
950 | pending_idx = *((u16 *)skb->data); | |
7376419a | 951 | xenvif_idx_release(vif, pending_idx, XEN_NETIF_RSP_OKAY); |
f942dc25 | 952 | for (j = start; j < i; j++) { |
5ccb3ea7 | 953 | pending_idx = frag_get_pending_idx(&shinfo->frags[j]); |
7376419a WL |
954 | xenvif_idx_release(vif, pending_idx, |
955 | XEN_NETIF_RSP_OKAY); | |
f942dc25 IC |
956 | } |
957 | ||
958 | /* Remember the error: invalidate all subsequent fragments. */ | |
959 | err = newerr; | |
960 | } | |
961 | ||
962 | *gopp = gop + 1; | |
963 | return err; | |
964 | } | |
965 | ||
7376419a | 966 | static void xenvif_fill_frags(struct xenvif *vif, struct sk_buff *skb) |
f942dc25 IC |
967 | { |
968 | struct skb_shared_info *shinfo = skb_shinfo(skb); | |
969 | int nr_frags = shinfo->nr_frags; | |
970 | int i; | |
971 | ||
972 | for (i = 0; i < nr_frags; i++) { | |
973 | skb_frag_t *frag = shinfo->frags + i; | |
974 | struct xen_netif_tx_request *txp; | |
ea066ad1 IC |
975 | struct page *page; |
976 | u16 pending_idx; | |
f942dc25 | 977 | |
ea066ad1 | 978 | pending_idx = frag_get_pending_idx(frag); |
f942dc25 | 979 | |
b3f980bd WL |
980 | txp = &vif->pending_tx_info[pending_idx].req; |
981 | page = virt_to_page(idx_to_kaddr(vif, pending_idx)); | |
ea066ad1 | 982 | __skb_fill_page_desc(skb, i, page, txp->offset, txp->size); |
f942dc25 IC |
983 | skb->len += txp->size; |
984 | skb->data_len += txp->size; | |
985 | skb->truesize += txp->size; | |
986 | ||
7376419a | 987 | /* Take an extra reference to offset xenvif_idx_release */ |
b3f980bd | 988 | get_page(vif->mmap_pages[pending_idx]); |
7376419a | 989 | xenvif_idx_release(vif, pending_idx, XEN_NETIF_RSP_OKAY); |
f942dc25 IC |
990 | } |
991 | } | |
992 | ||
7376419a | 993 | static int xenvif_get_extras(struct xenvif *vif, |
f942dc25 IC |
994 | struct xen_netif_extra_info *extras, |
995 | int work_to_do) | |
996 | { | |
997 | struct xen_netif_extra_info extra; | |
998 | RING_IDX cons = vif->tx.req_cons; | |
999 | ||
1000 | do { | |
1001 | if (unlikely(work_to_do-- <= 0)) { | |
48856286 | 1002 | netdev_err(vif->dev, "Missing extra info\n"); |
7376419a | 1003 | xenvif_fatal_tx_err(vif); |
f942dc25 IC |
1004 | return -EBADR; |
1005 | } | |
1006 | ||
1007 | memcpy(&extra, RING_GET_REQUEST(&vif->tx, cons), | |
1008 | sizeof(extra)); | |
1009 | if (unlikely(!extra.type || | |
1010 | extra.type >= XEN_NETIF_EXTRA_TYPE_MAX)) { | |
1011 | vif->tx.req_cons = ++cons; | |
48856286 | 1012 | netdev_err(vif->dev, |
f942dc25 | 1013 | "Invalid extra type: %d\n", extra.type); |
7376419a | 1014 | xenvif_fatal_tx_err(vif); |
f942dc25 IC |
1015 | return -EINVAL; |
1016 | } | |
1017 | ||
1018 | memcpy(&extras[extra.type - 1], &extra, sizeof(extra)); | |
1019 | vif->tx.req_cons = ++cons; | |
1020 | } while (extra.flags & XEN_NETIF_EXTRA_FLAG_MORE); | |
1021 | ||
1022 | return work_to_do; | |
1023 | } | |
1024 | ||
7376419a WL |
1025 | static int xenvif_set_skb_gso(struct xenvif *vif, |
1026 | struct sk_buff *skb, | |
1027 | struct xen_netif_extra_info *gso) | |
f942dc25 IC |
1028 | { |
1029 | if (!gso->u.gso.size) { | |
48856286 | 1030 | netdev_err(vif->dev, "GSO size must not be zero.\n"); |
7376419a | 1031 | xenvif_fatal_tx_err(vif); |
f942dc25 IC |
1032 | return -EINVAL; |
1033 | } | |
1034 | ||
a9468587 PD |
1035 | switch (gso->u.gso.type) { |
1036 | case XEN_NETIF_GSO_TYPE_TCPV4: | |
1037 | skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4; | |
1038 | break; | |
1039 | case XEN_NETIF_GSO_TYPE_TCPV6: | |
1040 | skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6; | |
1041 | break; | |
1042 | default: | |
48856286 | 1043 | netdev_err(vif->dev, "Bad GSO type %d.\n", gso->u.gso.type); |
7376419a | 1044 | xenvif_fatal_tx_err(vif); |
f942dc25 IC |
1045 | return -EINVAL; |
1046 | } | |
1047 | ||
1048 | skb_shinfo(skb)->gso_size = gso->u.gso.size; | |
b89587a7 | 1049 | /* gso_segs will be calculated later */ |
f942dc25 IC |
1050 | |
1051 | return 0; | |
1052 | } | |
1053 | ||
1431fb31 PD |
1054 | static inline int maybe_pull_tail(struct sk_buff *skb, unsigned int len, |
1055 | unsigned int max) | |
2eba61d5 | 1056 | { |
1431fb31 PD |
1057 | if (skb_headlen(skb) >= len) |
1058 | return 0; | |
1059 | ||
1060 | /* If we need to pullup then pullup to the max, so we | |
1061 | * won't need to do it again. | |
1062 | */ | |
1063 | if (max > skb->len) | |
1064 | max = skb->len; | |
1065 | ||
1066 | if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL) | |
1067 | return -ENOMEM; | |
1068 | ||
1069 | if (skb_headlen(skb) < len) | |
1070 | return -EPROTO; | |
1071 | ||
1072 | return 0; | |
2eba61d5 PD |
1073 | } |
1074 | ||
1431fb31 PD |
1075 | /* This value should be large enough to cover a tagged ethernet header plus |
1076 | * maximally sized IP and TCP or UDP headers. | |
1077 | */ | |
1078 | #define MAX_IP_HDR_LEN 128 | |
1079 | ||
2eba61d5 PD |
1080 | static int checksum_setup_ip(struct xenvif *vif, struct sk_buff *skb, |
1081 | int recalculate_partial_csum) | |
f942dc25 | 1082 | { |
2eba61d5 | 1083 | unsigned int off; |
1431fb31 PD |
1084 | bool fragment; |
1085 | int err; | |
1086 | ||
1087 | fragment = false; | |
1088 | ||
1089 | err = maybe_pull_tail(skb, | |
1090 | sizeof(struct iphdr), | |
1091 | MAX_IP_HDR_LEN); | |
1092 | if (err < 0) | |
1093 | goto out; | |
f942dc25 | 1094 | |
1431fb31 PD |
1095 | if (ip_hdr(skb)->frag_off & htons(IP_OFFSET | IP_MF)) |
1096 | fragment = true; | |
f942dc25 | 1097 | |
1431fb31 | 1098 | off = ip_hdrlen(skb); |
f942dc25 | 1099 | |
1431fb31 | 1100 | err = -EPROTO; |
f942dc25 | 1101 | |
7022ef8b WY |
1102 | if (fragment) |
1103 | goto out; | |
1104 | ||
1431fb31 | 1105 | switch (ip_hdr(skb)->protocol) { |
f942dc25 | 1106 | case IPPROTO_TCP: |
d52eb0d4 PD |
1107 | err = maybe_pull_tail(skb, |
1108 | off + sizeof(struct tcphdr), | |
1109 | MAX_IP_HDR_LEN); | |
1110 | if (err < 0) | |
1111 | goto out; | |
1112 | ||
2eba61d5 | 1113 | if (!skb_partial_csum_set(skb, off, |
0c8d087c WY |
1114 | offsetof(struct tcphdr, check))) { |
1115 | err = -EPROTO; | |
bea89336 | 1116 | goto out; |
0c8d087c | 1117 | } |
f942dc25 | 1118 | |
d52eb0d4 | 1119 | if (recalculate_partial_csum) |
1431fb31 PD |
1120 | tcp_hdr(skb)->check = |
1121 | ~csum_tcpudp_magic(ip_hdr(skb)->saddr, | |
1122 | ip_hdr(skb)->daddr, | |
1123 | skb->len - off, | |
1124 | IPPROTO_TCP, 0); | |
f942dc25 IC |
1125 | break; |
1126 | case IPPROTO_UDP: | |
d52eb0d4 PD |
1127 | err = maybe_pull_tail(skb, |
1128 | off + sizeof(struct udphdr), | |
1129 | MAX_IP_HDR_LEN); | |
1130 | if (err < 0) | |
1131 | goto out; | |
1132 | ||
2eba61d5 | 1133 | if (!skb_partial_csum_set(skb, off, |
0c8d087c WY |
1134 | offsetof(struct udphdr, check))) { |
1135 | err = -EPROTO; | |
bea89336 | 1136 | goto out; |
0c8d087c | 1137 | } |
f942dc25 | 1138 | |
d52eb0d4 | 1139 | if (recalculate_partial_csum) |
1431fb31 PD |
1140 | udp_hdr(skb)->check = |
1141 | ~csum_tcpudp_magic(ip_hdr(skb)->saddr, | |
1142 | ip_hdr(skb)->daddr, | |
1143 | skb->len - off, | |
1144 | IPPROTO_UDP, 0); | |
f942dc25 IC |
1145 | break; |
1146 | default: | |
f942dc25 IC |
1147 | goto out; |
1148 | } | |
1149 | ||
f942dc25 IC |
1150 | err = 0; |
1151 | ||
1152 | out: | |
1153 | return err; | |
1154 | } | |
1155 | ||
1431fb31 PD |
1156 | /* This value should be large enough to cover a tagged ethernet header plus |
1157 | * an IPv6 header, all options, and a maximal TCP or UDP header. | |
1158 | */ | |
1159 | #define MAX_IPV6_HDR_LEN 256 | |
1160 | ||
1161 | #define OPT_HDR(type, skb, off) \ | |
1162 | (type *)(skb_network_header(skb) + (off)) | |
1163 | ||
2eba61d5 PD |
1164 | static int checksum_setup_ipv6(struct xenvif *vif, struct sk_buff *skb, |
1165 | int recalculate_partial_csum) | |
1166 | { | |
1431fb31 | 1167 | int err; |
2eba61d5 | 1168 | u8 nexthdr; |
2eba61d5 | 1169 | unsigned int off; |
1431fb31 | 1170 | unsigned int len; |
2eba61d5 PD |
1171 | bool fragment; |
1172 | bool done; | |
1173 | ||
1431fb31 | 1174 | fragment = false; |
2eba61d5 PD |
1175 | done = false; |
1176 | ||
1177 | off = sizeof(struct ipv6hdr); | |
1178 | ||
1431fb31 PD |
1179 | err = maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN); |
1180 | if (err < 0) | |
1181 | goto out; | |
2eba61d5 | 1182 | |
1431fb31 | 1183 | nexthdr = ipv6_hdr(skb)->nexthdr; |
2eba61d5 | 1184 | |
1431fb31 PD |
1185 | len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len); |
1186 | while (off <= len && !done) { | |
2eba61d5 PD |
1187 | switch (nexthdr) { |
1188 | case IPPROTO_DSTOPTS: | |
1189 | case IPPROTO_HOPOPTS: | |
1190 | case IPPROTO_ROUTING: { | |
1431fb31 | 1191 | struct ipv6_opt_hdr *hp; |
2eba61d5 | 1192 | |
1431fb31 PD |
1193 | err = maybe_pull_tail(skb, |
1194 | off + | |
1195 | sizeof(struct ipv6_opt_hdr), | |
1196 | MAX_IPV6_HDR_LEN); | |
1197 | if (err < 0) | |
1198 | goto out; | |
2eba61d5 | 1199 | |
1431fb31 | 1200 | hp = OPT_HDR(struct ipv6_opt_hdr, skb, off); |
2eba61d5 PD |
1201 | nexthdr = hp->nexthdr; |
1202 | off += ipv6_optlen(hp); | |
1203 | break; | |
1204 | } | |
1205 | case IPPROTO_AH: { | |
1431fb31 PD |
1206 | struct ip_auth_hdr *hp; |
1207 | ||
1208 | err = maybe_pull_tail(skb, | |
1209 | off + | |
1210 | sizeof(struct ip_auth_hdr), | |
1211 | MAX_IPV6_HDR_LEN); | |
1212 | if (err < 0) | |
1213 | goto out; | |
1214 | ||
1215 | hp = OPT_HDR(struct ip_auth_hdr, skb, off); | |
1216 | nexthdr = hp->nexthdr; | |
1217 | off += ipv6_authlen(hp); | |
1218 | break; | |
1219 | } | |
1220 | case IPPROTO_FRAGMENT: { | |
1221 | struct frag_hdr *hp; | |
2eba61d5 | 1222 | |
1431fb31 PD |
1223 | err = maybe_pull_tail(skb, |
1224 | off + | |
1225 | sizeof(struct frag_hdr), | |
1226 | MAX_IPV6_HDR_LEN); | |
1227 | if (err < 0) | |
1228 | goto out; | |
1229 | ||
1230 | hp = OPT_HDR(struct frag_hdr, skb, off); | |
1231 | ||
1232 | if (hp->frag_off & htons(IP6_OFFSET | IP6_MF)) | |
1233 | fragment = true; | |
2eba61d5 PD |
1234 | |
1235 | nexthdr = hp->nexthdr; | |
1431fb31 | 1236 | off += sizeof(struct frag_hdr); |
2eba61d5 PD |
1237 | break; |
1238 | } | |
2eba61d5 PD |
1239 | default: |
1240 | done = true; | |
1241 | break; | |
1242 | } | |
1243 | } | |
1244 | ||
1431fb31 | 1245 | err = -EPROTO; |
2eba61d5 | 1246 | |
1431fb31 | 1247 | if (!done || fragment) |
2eba61d5 | 1248 | goto out; |
2eba61d5 PD |
1249 | |
1250 | switch (nexthdr) { | |
1251 | case IPPROTO_TCP: | |
d52eb0d4 PD |
1252 | err = maybe_pull_tail(skb, |
1253 | off + sizeof(struct tcphdr), | |
1254 | MAX_IPV6_HDR_LEN); | |
1255 | if (err < 0) | |
1256 | goto out; | |
1257 | ||
2eba61d5 | 1258 | if (!skb_partial_csum_set(skb, off, |
0c8d087c WY |
1259 | offsetof(struct tcphdr, check))) { |
1260 | err = -EPROTO; | |
2eba61d5 | 1261 | goto out; |
0c8d087c | 1262 | } |
2eba61d5 | 1263 | |
d52eb0d4 | 1264 | if (recalculate_partial_csum) |
1431fb31 PD |
1265 | tcp_hdr(skb)->check = |
1266 | ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr, | |
1267 | &ipv6_hdr(skb)->daddr, | |
1268 | skb->len - off, | |
1269 | IPPROTO_TCP, 0); | |
2eba61d5 PD |
1270 | break; |
1271 | case IPPROTO_UDP: | |
d52eb0d4 PD |
1272 | err = maybe_pull_tail(skb, |
1273 | off + sizeof(struct udphdr), | |
1274 | MAX_IPV6_HDR_LEN); | |
1275 | if (err < 0) | |
1276 | goto out; | |
1277 | ||
2eba61d5 | 1278 | if (!skb_partial_csum_set(skb, off, |
0c8d087c WY |
1279 | offsetof(struct udphdr, check))) { |
1280 | err = -EPROTO; | |
2eba61d5 | 1281 | goto out; |
0c8d087c | 1282 | } |
2eba61d5 | 1283 | |
d52eb0d4 | 1284 | if (recalculate_partial_csum) |
1431fb31 PD |
1285 | udp_hdr(skb)->check = |
1286 | ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr, | |
1287 | &ipv6_hdr(skb)->daddr, | |
1288 | skb->len - off, | |
1289 | IPPROTO_UDP, 0); | |
2eba61d5 PD |
1290 | break; |
1291 | default: | |
2eba61d5 PD |
1292 | goto out; |
1293 | } | |
1294 | ||
1295 | err = 0; | |
1296 | ||
1297 | out: | |
1298 | return err; | |
1299 | } | |
1300 | ||
1301 | static int checksum_setup(struct xenvif *vif, struct sk_buff *skb) | |
1302 | { | |
1303 | int err = -EPROTO; | |
1304 | int recalculate_partial_csum = 0; | |
1305 | ||
1306 | /* A GSO SKB must be CHECKSUM_PARTIAL. However some buggy | |
1307 | * peers can fail to set NETRXF_csum_blank when sending a GSO | |
1308 | * frame. In this case force the SKB to CHECKSUM_PARTIAL and | |
1309 | * recalculate the partial checksum. | |
1310 | */ | |
1311 | if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) { | |
1312 | vif->rx_gso_checksum_fixup++; | |
1313 | skb->ip_summed = CHECKSUM_PARTIAL; | |
1314 | recalculate_partial_csum = 1; | |
1315 | } | |
1316 | ||
1317 | /* A non-CHECKSUM_PARTIAL SKB does not require setup. */ | |
1318 | if (skb->ip_summed != CHECKSUM_PARTIAL) | |
1319 | return 0; | |
1320 | ||
1321 | if (skb->protocol == htons(ETH_P_IP)) | |
1322 | err = checksum_setup_ip(vif, skb, recalculate_partial_csum); | |
1323 | else if (skb->protocol == htons(ETH_P_IPV6)) | |
1324 | err = checksum_setup_ipv6(vif, skb, recalculate_partial_csum); | |
1325 | ||
1326 | return err; | |
1327 | } | |
1328 | ||
f942dc25 IC |
1329 | static bool tx_credit_exceeded(struct xenvif *vif, unsigned size) |
1330 | { | |
059dfa6a WL |
1331 | u64 now = get_jiffies_64(); |
1332 | u64 next_credit = vif->credit_window_start + | |
f942dc25 IC |
1333 | msecs_to_jiffies(vif->credit_usec / 1000); |
1334 | ||
1335 | /* Timer could already be pending in rare cases. */ | |
1336 | if (timer_pending(&vif->credit_timeout)) | |
1337 | return true; | |
1338 | ||
1339 | /* Passed the point where we can replenish credit? */ | |
059dfa6a WL |
1340 | if (time_after_eq64(now, next_credit)) { |
1341 | vif->credit_window_start = now; | |
f942dc25 IC |
1342 | tx_add_credit(vif); |
1343 | } | |
1344 | ||
1345 | /* Still too big to send right now? Set a callback. */ | |
1346 | if (size > vif->remaining_credit) { | |
1347 | vif->credit_timeout.data = | |
1348 | (unsigned long)vif; | |
1349 | vif->credit_timeout.function = | |
1350 | tx_credit_callback; | |
1351 | mod_timer(&vif->credit_timeout, | |
1352 | next_credit); | |
059dfa6a | 1353 | vif->credit_window_start = next_credit; |
f942dc25 IC |
1354 | |
1355 | return true; | |
1356 | } | |
1357 | ||
1358 | return false; | |
1359 | } | |
1360 | ||
10574059 | 1361 | static unsigned xenvif_tx_build_gops(struct xenvif *vif, int budget) |
f942dc25 | 1362 | { |
b3f980bd | 1363 | struct gnttab_copy *gop = vif->tx_copy_ops, *request_gop; |
f942dc25 IC |
1364 | struct sk_buff *skb; |
1365 | int ret; | |
1366 | ||
b3f980bd | 1367 | while ((nr_pending_reqs(vif) + XEN_NETBK_LEGACY_SLOTS_MAX |
10574059 PD |
1368 | < MAX_PENDING_REQS) && |
1369 | (skb_queue_len(&vif->tx_queue) < budget)) { | |
f942dc25 | 1370 | struct xen_netif_tx_request txreq; |
37641494 | 1371 | struct xen_netif_tx_request txfrags[XEN_NETBK_LEGACY_SLOTS_MAX]; |
f942dc25 IC |
1372 | struct page *page; |
1373 | struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX-1]; | |
1374 | u16 pending_idx; | |
1375 | RING_IDX idx; | |
1376 | int work_to_do; | |
1377 | unsigned int data_len; | |
1378 | pending_ring_idx_t index; | |
1379 | ||
48856286 IC |
1380 | if (vif->tx.sring->req_prod - vif->tx.req_cons > |
1381 | XEN_NETIF_TX_RING_SIZE) { | |
1382 | netdev_err(vif->dev, | |
1383 | "Impossible number of requests. " | |
1384 | "req_prod %d, req_cons %d, size %ld\n", | |
1385 | vif->tx.sring->req_prod, vif->tx.req_cons, | |
1386 | XEN_NETIF_TX_RING_SIZE); | |
7376419a | 1387 | xenvif_fatal_tx_err(vif); |
48856286 IC |
1388 | continue; |
1389 | } | |
1390 | ||
d9601a36 | 1391 | work_to_do = RING_HAS_UNCONSUMED_REQUESTS(&vif->tx); |
b3f980bd WL |
1392 | if (!work_to_do) |
1393 | break; | |
f942dc25 IC |
1394 | |
1395 | idx = vif->tx.req_cons; | |
1396 | rmb(); /* Ensure that we see the request before we copy it. */ | |
1397 | memcpy(&txreq, RING_GET_REQUEST(&vif->tx, idx), sizeof(txreq)); | |
1398 | ||
1399 | /* Credit-based scheduling. */ | |
1400 | if (txreq.size > vif->remaining_credit && | |
b3f980bd WL |
1401 | tx_credit_exceeded(vif, txreq.size)) |
1402 | break; | |
f942dc25 IC |
1403 | |
1404 | vif->remaining_credit -= txreq.size; | |
1405 | ||
1406 | work_to_do--; | |
1407 | vif->tx.req_cons = ++idx; | |
1408 | ||
1409 | memset(extras, 0, sizeof(extras)); | |
1410 | if (txreq.flags & XEN_NETTXF_extra_info) { | |
7376419a WL |
1411 | work_to_do = xenvif_get_extras(vif, extras, |
1412 | work_to_do); | |
f942dc25 | 1413 | idx = vif->tx.req_cons; |
48856286 | 1414 | if (unlikely(work_to_do < 0)) |
b3f980bd | 1415 | break; |
f942dc25 IC |
1416 | } |
1417 | ||
7376419a | 1418 | ret = xenvif_count_requests(vif, &txreq, txfrags, work_to_do); |
48856286 | 1419 | if (unlikely(ret < 0)) |
b3f980bd | 1420 | break; |
48856286 | 1421 | |
f942dc25 IC |
1422 | idx += ret; |
1423 | ||
1424 | if (unlikely(txreq.size < ETH_HLEN)) { | |
1425 | netdev_dbg(vif->dev, | |
1426 | "Bad packet size: %d\n", txreq.size); | |
7376419a | 1427 | xenvif_tx_err(vif, &txreq, idx); |
b3f980bd | 1428 | break; |
f942dc25 IC |
1429 | } |
1430 | ||
1431 | /* No crossing a page as the payload mustn't fragment. */ | |
1432 | if (unlikely((txreq.offset + txreq.size) > PAGE_SIZE)) { | |
48856286 | 1433 | netdev_err(vif->dev, |
f942dc25 IC |
1434 | "txreq.offset: %x, size: %u, end: %lu\n", |
1435 | txreq.offset, txreq.size, | |
1436 | (txreq.offset&~PAGE_MASK) + txreq.size); | |
7376419a | 1437 | xenvif_fatal_tx_err(vif); |
b3f980bd | 1438 | break; |
f942dc25 IC |
1439 | } |
1440 | ||
b3f980bd WL |
1441 | index = pending_index(vif->pending_cons); |
1442 | pending_idx = vif->pending_ring[index]; | |
f942dc25 IC |
1443 | |
1444 | data_len = (txreq.size > PKT_PROT_LEN && | |
37641494 | 1445 | ret < XEN_NETBK_LEGACY_SLOTS_MAX) ? |
f942dc25 IC |
1446 | PKT_PROT_LEN : txreq.size; |
1447 | ||
1448 | skb = alloc_skb(data_len + NET_SKB_PAD + NET_IP_ALIGN, | |
1449 | GFP_ATOMIC | __GFP_NOWARN); | |
1450 | if (unlikely(skb == NULL)) { | |
1451 | netdev_dbg(vif->dev, | |
1452 | "Can't allocate a skb in start_xmit.\n"); | |
7376419a | 1453 | xenvif_tx_err(vif, &txreq, idx); |
f942dc25 IC |
1454 | break; |
1455 | } | |
1456 | ||
1457 | /* Packets passed to netif_rx() must have some headroom. */ | |
1458 | skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN); | |
1459 | ||
1460 | if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) { | |
1461 | struct xen_netif_extra_info *gso; | |
1462 | gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1]; | |
1463 | ||
7376419a WL |
1464 | if (xenvif_set_skb_gso(vif, skb, gso)) { |
1465 | /* Failure in xenvif_set_skb_gso is fatal. */ | |
f942dc25 | 1466 | kfree_skb(skb); |
b3f980bd | 1467 | break; |
f942dc25 IC |
1468 | } |
1469 | } | |
1470 | ||
1471 | /* XXX could copy straight to head */ | |
7376419a | 1472 | page = xenvif_alloc_page(vif, pending_idx); |
f942dc25 IC |
1473 | if (!page) { |
1474 | kfree_skb(skb); | |
7376419a | 1475 | xenvif_tx_err(vif, &txreq, idx); |
b3f980bd | 1476 | break; |
f942dc25 IC |
1477 | } |
1478 | ||
f942dc25 IC |
1479 | gop->source.u.ref = txreq.gref; |
1480 | gop->source.domid = vif->domid; | |
1481 | gop->source.offset = txreq.offset; | |
1482 | ||
1483 | gop->dest.u.gmfn = virt_to_mfn(page_address(page)); | |
1484 | gop->dest.domid = DOMID_SELF; | |
1485 | gop->dest.offset = txreq.offset; | |
1486 | ||
1487 | gop->len = txreq.size; | |
1488 | gop->flags = GNTCOPY_source_gref; | |
1489 | ||
1490 | gop++; | |
1491 | ||
b3f980bd | 1492 | memcpy(&vif->pending_tx_info[pending_idx].req, |
f942dc25 | 1493 | &txreq, sizeof(txreq)); |
b3f980bd | 1494 | vif->pending_tx_info[pending_idx].head = index; |
f942dc25 IC |
1495 | *((u16 *)skb->data) = pending_idx; |
1496 | ||
1497 | __skb_put(skb, data_len); | |
1498 | ||
1499 | skb_shinfo(skb)->nr_frags = ret; | |
1500 | if (data_len < txreq.size) { | |
1501 | skb_shinfo(skb)->nr_frags++; | |
ea066ad1 IC |
1502 | frag_set_pending_idx(&skb_shinfo(skb)->frags[0], |
1503 | pending_idx); | |
f942dc25 | 1504 | } else { |
ea066ad1 IC |
1505 | frag_set_pending_idx(&skb_shinfo(skb)->frags[0], |
1506 | INVALID_PENDING_IDX); | |
f942dc25 IC |
1507 | } |
1508 | ||
b3f980bd | 1509 | vif->pending_cons++; |
f942dc25 | 1510 | |
7376419a | 1511 | request_gop = xenvif_get_requests(vif, skb, txfrags, gop); |
f942dc25 IC |
1512 | if (request_gop == NULL) { |
1513 | kfree_skb(skb); | |
7376419a | 1514 | xenvif_tx_err(vif, &txreq, idx); |
b3f980bd | 1515 | break; |
f942dc25 IC |
1516 | } |
1517 | gop = request_gop; | |
1518 | ||
b3f980bd | 1519 | __skb_queue_tail(&vif->tx_queue, skb); |
1e0b6eac | 1520 | |
f942dc25 | 1521 | vif->tx.req_cons = idx; |
f942dc25 | 1522 | |
b3f980bd | 1523 | if ((gop-vif->tx_copy_ops) >= ARRAY_SIZE(vif->tx_copy_ops)) |
f942dc25 IC |
1524 | break; |
1525 | } | |
1526 | ||
b3f980bd | 1527 | return gop - vif->tx_copy_ops; |
f942dc25 IC |
1528 | } |
1529 | ||
b3f980bd | 1530 | |
10574059 | 1531 | static int xenvif_tx_submit(struct xenvif *vif) |
f942dc25 | 1532 | { |
b3f980bd | 1533 | struct gnttab_copy *gop = vif->tx_copy_ops; |
f942dc25 | 1534 | struct sk_buff *skb; |
b3f980bd | 1535 | int work_done = 0; |
f942dc25 | 1536 | |
10574059 | 1537 | while ((skb = __skb_dequeue(&vif->tx_queue)) != NULL) { |
f942dc25 | 1538 | struct xen_netif_tx_request *txp; |
f942dc25 IC |
1539 | u16 pending_idx; |
1540 | unsigned data_len; | |
1541 | ||
1542 | pending_idx = *((u16 *)skb->data); | |
b3f980bd | 1543 | txp = &vif->pending_tx_info[pending_idx].req; |
f942dc25 IC |
1544 | |
1545 | /* Check the remap error code. */ | |
7376419a | 1546 | if (unlikely(xenvif_tx_check_gop(vif, skb, &gop))) { |
f942dc25 IC |
1547 | netdev_dbg(vif->dev, "netback grant failed.\n"); |
1548 | skb_shinfo(skb)->nr_frags = 0; | |
1549 | kfree_skb(skb); | |
1550 | continue; | |
1551 | } | |
1552 | ||
1553 | data_len = skb->len; | |
1554 | memcpy(skb->data, | |
b3f980bd | 1555 | (void *)(idx_to_kaddr(vif, pending_idx)|txp->offset), |
f942dc25 IC |
1556 | data_len); |
1557 | if (data_len < txp->size) { | |
1558 | /* Append the packet payload as a fragment. */ | |
1559 | txp->offset += data_len; | |
1560 | txp->size -= data_len; | |
1561 | } else { | |
1562 | /* Schedule a response immediately. */ | |
7376419a WL |
1563 | xenvif_idx_release(vif, pending_idx, |
1564 | XEN_NETIF_RSP_OKAY); | |
f942dc25 IC |
1565 | } |
1566 | ||
1567 | if (txp->flags & XEN_NETTXF_csum_blank) | |
1568 | skb->ip_summed = CHECKSUM_PARTIAL; | |
1569 | else if (txp->flags & XEN_NETTXF_data_validated) | |
1570 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
1571 | ||
7376419a | 1572 | xenvif_fill_frags(vif, skb); |
f942dc25 | 1573 | |
2eba61d5 | 1574 | if (skb_is_nonlinear(skb) && skb_headlen(skb) < PKT_PROT_LEN) { |
f942dc25 IC |
1575 | int target = min_t(int, skb->len, PKT_PROT_LEN); |
1576 | __pskb_pull_tail(skb, target - skb_headlen(skb)); | |
1577 | } | |
1578 | ||
1579 | skb->dev = vif->dev; | |
1580 | skb->protocol = eth_type_trans(skb, skb->dev); | |
f9ca8f74 | 1581 | skb_reset_network_header(skb); |
f942dc25 IC |
1582 | |
1583 | if (checksum_setup(vif, skb)) { | |
1584 | netdev_dbg(vif->dev, | |
1585 | "Can't setup checksum in net_tx_action\n"); | |
1586 | kfree_skb(skb); | |
1587 | continue; | |
1588 | } | |
1589 | ||
40893fd0 | 1590 | skb_probe_transport_header(skb, 0); |
f9ca8f74 | 1591 | |
b89587a7 PD |
1592 | /* If the packet is GSO then we will have just set up the |
1593 | * transport header offset in checksum_setup so it's now | |
1594 | * straightforward to calculate gso_segs. | |
1595 | */ | |
1596 | if (skb_is_gso(skb)) { | |
1597 | int mss = skb_shinfo(skb)->gso_size; | |
1598 | int hdrlen = skb_transport_header(skb) - | |
1599 | skb_mac_header(skb) + | |
1600 | tcp_hdrlen(skb); | |
1601 | ||
1602 | skb_shinfo(skb)->gso_segs = | |
1603 | DIV_ROUND_UP(skb->len - hdrlen, mss); | |
1604 | } | |
1605 | ||
f942dc25 IC |
1606 | vif->dev->stats.rx_bytes += skb->len; |
1607 | vif->dev->stats.rx_packets++; | |
1608 | ||
b3f980bd WL |
1609 | work_done++; |
1610 | ||
1611 | netif_receive_skb(skb); | |
f942dc25 | 1612 | } |
b3f980bd WL |
1613 | |
1614 | return work_done; | |
f942dc25 IC |
1615 | } |
1616 | ||
1617 | /* Called after netfront has transmitted */ | |
7376419a | 1618 | int xenvif_tx_action(struct xenvif *vif, int budget) |
f942dc25 IC |
1619 | { |
1620 | unsigned nr_gops; | |
b3f980bd | 1621 | int work_done; |
f942dc25 | 1622 | |
b3f980bd WL |
1623 | if (unlikely(!tx_work_todo(vif))) |
1624 | return 0; | |
1625 | ||
10574059 | 1626 | nr_gops = xenvif_tx_build_gops(vif, budget); |
f942dc25 IC |
1627 | |
1628 | if (nr_gops == 0) | |
b3f980bd WL |
1629 | return 0; |
1630 | ||
1631 | gnttab_batch_copy(vif->tx_copy_ops, nr_gops); | |
f942dc25 | 1632 | |
10574059 | 1633 | work_done = xenvif_tx_submit(vif); |
f942dc25 | 1634 | |
b3f980bd | 1635 | return work_done; |
f942dc25 IC |
1636 | } |
1637 | ||
7376419a WL |
1638 | static void xenvif_idx_release(struct xenvif *vif, u16 pending_idx, |
1639 | u8 status) | |
f942dc25 | 1640 | { |
f942dc25 | 1641 | struct pending_tx_info *pending_tx_info; |
2810e5b9 WL |
1642 | pending_ring_idx_t head; |
1643 | u16 peek; /* peek into next tx request */ | |
1644 | ||
b3f980bd | 1645 | BUG_ON(vif->mmap_pages[pending_idx] == (void *)(~0UL)); |
f942dc25 IC |
1646 | |
1647 | /* Already complete? */ | |
b3f980bd | 1648 | if (vif->mmap_pages[pending_idx] == NULL) |
f942dc25 IC |
1649 | return; |
1650 | ||
b3f980bd | 1651 | pending_tx_info = &vif->pending_tx_info[pending_idx]; |
f942dc25 | 1652 | |
2810e5b9 | 1653 | head = pending_tx_info->head; |
f942dc25 | 1654 | |
b3f980bd WL |
1655 | BUG_ON(!pending_tx_is_head(vif, head)); |
1656 | BUG_ON(vif->pending_ring[pending_index(head)] != pending_idx); | |
f942dc25 | 1657 | |
2810e5b9 WL |
1658 | do { |
1659 | pending_ring_idx_t index; | |
1660 | pending_ring_idx_t idx = pending_index(head); | |
b3f980bd | 1661 | u16 info_idx = vif->pending_ring[idx]; |
f942dc25 | 1662 | |
b3f980bd | 1663 | pending_tx_info = &vif->pending_tx_info[info_idx]; |
2810e5b9 WL |
1664 | make_tx_response(vif, &pending_tx_info->req, status); |
1665 | ||
1666 | /* Setting any number other than | |
1667 | * INVALID_PENDING_RING_IDX indicates this slot is | |
1668 | * starting a new packet / ending a previous packet. | |
1669 | */ | |
1670 | pending_tx_info->head = 0; | |
1671 | ||
b3f980bd WL |
1672 | index = pending_index(vif->pending_prod++); |
1673 | vif->pending_ring[index] = vif->pending_ring[info_idx]; | |
f942dc25 | 1674 | |
b3f980bd | 1675 | peek = vif->pending_ring[pending_index(++head)]; |
2810e5b9 | 1676 | |
b3f980bd | 1677 | } while (!pending_tx_is_head(vif, peek)); |
2810e5b9 | 1678 | |
b3f980bd WL |
1679 | put_page(vif->mmap_pages[pending_idx]); |
1680 | vif->mmap_pages[pending_idx] = NULL; | |
f942dc25 IC |
1681 | } |
1682 | ||
2810e5b9 | 1683 | |
f942dc25 IC |
1684 | static void make_tx_response(struct xenvif *vif, |
1685 | struct xen_netif_tx_request *txp, | |
1686 | s8 st) | |
1687 | { | |
1688 | RING_IDX i = vif->tx.rsp_prod_pvt; | |
1689 | struct xen_netif_tx_response *resp; | |
1690 | int notify; | |
1691 | ||
1692 | resp = RING_GET_RESPONSE(&vif->tx, i); | |
1693 | resp->id = txp->id; | |
1694 | resp->status = st; | |
1695 | ||
1696 | if (txp->flags & XEN_NETTXF_extra_info) | |
1697 | RING_GET_RESPONSE(&vif->tx, ++i)->status = XEN_NETIF_RSP_NULL; | |
1698 | ||
1699 | vif->tx.rsp_prod_pvt = ++i; | |
1700 | RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&vif->tx, notify); | |
1701 | if (notify) | |
e1f00a69 | 1702 | notify_remote_via_irq(vif->tx_irq); |
f942dc25 IC |
1703 | } |
1704 | ||
1705 | static struct xen_netif_rx_response *make_rx_response(struct xenvif *vif, | |
1706 | u16 id, | |
1707 | s8 st, | |
1708 | u16 offset, | |
1709 | u16 size, | |
1710 | u16 flags) | |
1711 | { | |
1712 | RING_IDX i = vif->rx.rsp_prod_pvt; | |
1713 | struct xen_netif_rx_response *resp; | |
1714 | ||
1715 | resp = RING_GET_RESPONSE(&vif->rx, i); | |
1716 | resp->offset = offset; | |
1717 | resp->flags = flags; | |
1718 | resp->id = id; | |
1719 | resp->status = (s16)size; | |
1720 | if (st < 0) | |
1721 | resp->status = (s16)st; | |
1722 | ||
1723 | vif->rx.rsp_prod_pvt = ++i; | |
1724 | ||
1725 | return resp; | |
1726 | } | |
1727 | ||
b3f980bd | 1728 | static inline int rx_work_todo(struct xenvif *vif) |
f942dc25 | 1729 | { |
11b57f90 PD |
1730 | return (!skb_queue_empty(&vif->rx_queue) && !vif->rx_queue_stopped) || |
1731 | vif->rx_event; | |
f942dc25 IC |
1732 | } |
1733 | ||
b3f980bd | 1734 | static inline int tx_work_todo(struct xenvif *vif) |
f942dc25 IC |
1735 | { |
1736 | ||
b3f980bd WL |
1737 | if (likely(RING_HAS_UNCONSUMED_REQUESTS(&vif->tx)) && |
1738 | (nr_pending_reqs(vif) + XEN_NETBK_LEGACY_SLOTS_MAX | |
1739 | < MAX_PENDING_REQS)) | |
f942dc25 IC |
1740 | return 1; |
1741 | ||
1742 | return 0; | |
1743 | } | |
1744 | ||
7376419a | 1745 | void xenvif_unmap_frontend_rings(struct xenvif *vif) |
f942dc25 | 1746 | { |
c9d63699 DV |
1747 | if (vif->tx.sring) |
1748 | xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif), | |
1749 | vif->tx.sring); | |
1750 | if (vif->rx.sring) | |
1751 | xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif), | |
1752 | vif->rx.sring); | |
f942dc25 IC |
1753 | } |
1754 | ||
7376419a WL |
1755 | int xenvif_map_frontend_rings(struct xenvif *vif, |
1756 | grant_ref_t tx_ring_ref, | |
1757 | grant_ref_t rx_ring_ref) | |
f942dc25 | 1758 | { |
c9d63699 | 1759 | void *addr; |
f942dc25 IC |
1760 | struct xen_netif_tx_sring *txs; |
1761 | struct xen_netif_rx_sring *rxs; | |
1762 | ||
1763 | int err = -ENOMEM; | |
1764 | ||
c9d63699 DV |
1765 | err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(vif), |
1766 | tx_ring_ref, &addr); | |
1767 | if (err) | |
f942dc25 IC |
1768 | goto err; |
1769 | ||
c9d63699 | 1770 | txs = (struct xen_netif_tx_sring *)addr; |
f942dc25 IC |
1771 | BACK_RING_INIT(&vif->tx, txs, PAGE_SIZE); |
1772 | ||
c9d63699 DV |
1773 | err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(vif), |
1774 | rx_ring_ref, &addr); | |
1775 | if (err) | |
f942dc25 | 1776 | goto err; |
f942dc25 | 1777 | |
c9d63699 | 1778 | rxs = (struct xen_netif_rx_sring *)addr; |
f942dc25 IC |
1779 | BACK_RING_INIT(&vif->rx, rxs, PAGE_SIZE); |
1780 | ||
1781 | return 0; | |
1782 | ||
1783 | err: | |
7376419a | 1784 | xenvif_unmap_frontend_rings(vif); |
f942dc25 IC |
1785 | return err; |
1786 | } | |
1787 | ||
ca2f09f2 PD |
1788 | void xenvif_stop_queue(struct xenvif *vif) |
1789 | { | |
1790 | if (!vif->can_queue) | |
1791 | return; | |
1792 | ||
1793 | netif_stop_queue(vif->dev); | |
1794 | } | |
1795 | ||
1796 | static void xenvif_start_queue(struct xenvif *vif) | |
1797 | { | |
1798 | if (xenvif_schedulable(vif)) | |
1799 | netif_wake_queue(vif->dev); | |
1800 | } | |
1801 | ||
7376419a | 1802 | int xenvif_kthread(void *data) |
b3f980bd WL |
1803 | { |
1804 | struct xenvif *vif = data; | |
ca2f09f2 | 1805 | struct sk_buff *skb; |
b3f980bd WL |
1806 | |
1807 | while (!kthread_should_stop()) { | |
1808 | wait_event_interruptible(vif->wq, | |
1809 | rx_work_todo(vif) || | |
1810 | kthread_should_stop()); | |
1811 | if (kthread_should_stop()) | |
1812 | break; | |
1813 | ||
ca2f09f2 | 1814 | if (!skb_queue_empty(&vif->rx_queue)) |
7376419a | 1815 | xenvif_rx_action(vif); |
b3f980bd | 1816 | |
ca2f09f2 PD |
1817 | vif->rx_event = false; |
1818 | ||
1819 | if (skb_queue_empty(&vif->rx_queue) && | |
1820 | netif_queue_stopped(vif->dev)) | |
1821 | xenvif_start_queue(vif); | |
1822 | ||
b3f980bd WL |
1823 | cond_resched(); |
1824 | } | |
1825 | ||
ca2f09f2 PD |
1826 | /* Bin any remaining skbs */ |
1827 | while ((skb = skb_dequeue(&vif->rx_queue)) != NULL) | |
1828 | dev_kfree_skb(skb); | |
1829 | ||
b3f980bd WL |
1830 | return 0; |
1831 | } | |
1832 | ||
f942dc25 IC |
1833 | static int __init netback_init(void) |
1834 | { | |
f942dc25 | 1835 | int rc = 0; |
f942dc25 | 1836 | |
2a14b244 | 1837 | if (!xen_domain()) |
f942dc25 IC |
1838 | return -ENODEV; |
1839 | ||
37641494 | 1840 | if (fatal_skb_slots < XEN_NETBK_LEGACY_SLOTS_MAX) { |
383eda32 JP |
1841 | pr_info("fatal_skb_slots too small (%d), bump it to XEN_NETBK_LEGACY_SLOTS_MAX (%d)\n", |
1842 | fatal_skb_slots, XEN_NETBK_LEGACY_SLOTS_MAX); | |
37641494 | 1843 | fatal_skb_slots = XEN_NETBK_LEGACY_SLOTS_MAX; |
2810e5b9 WL |
1844 | } |
1845 | ||
f942dc25 IC |
1846 | rc = xenvif_xenbus_init(); |
1847 | if (rc) | |
1848 | goto failed_init; | |
1849 | ||
1850 | return 0; | |
1851 | ||
1852 | failed_init: | |
f942dc25 | 1853 | return rc; |
f942dc25 IC |
1854 | } |
1855 | ||
1856 | module_init(netback_init); | |
1857 | ||
b103f358 WL |
1858 | static void __exit netback_fini(void) |
1859 | { | |
b103f358 | 1860 | xenvif_xenbus_fini(); |
b103f358 WL |
1861 | } |
1862 | module_exit(netback_fini); | |
1863 | ||
f942dc25 | 1864 | MODULE_LICENSE("Dual BSD/GPL"); |
f984cec6 | 1865 | MODULE_ALIAS("xen-backend:vif"); |