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 | ||
2810e5b9 WL |
65 | /* |
66 | * If head != INVALID_PENDING_RING_IDX, it means this tx request is head of | |
67 | * one or more merged tx requests, otherwise it is the continuation of | |
68 | * previous tx request. | |
69 | */ | |
b3f980bd | 70 | static inline int pending_tx_is_head(struct xenvif *vif, RING_IDX idx) |
f942dc25 | 71 | { |
b3f980bd | 72 | return vif->pending_tx_info[idx].head != INVALID_PENDING_RING_IDX; |
f942dc25 IC |
73 | } |
74 | ||
7376419a WL |
75 | static void xenvif_idx_release(struct xenvif *vif, u16 pending_idx, |
76 | u8 status); | |
77 | ||
f942dc25 IC |
78 | static void make_tx_response(struct xenvif *vif, |
79 | struct xen_netif_tx_request *txp, | |
80 | s8 st); | |
b3f980bd WL |
81 | |
82 | static inline int tx_work_todo(struct xenvif *vif); | |
83 | static inline int rx_work_todo(struct xenvif *vif); | |
84 | ||
f942dc25 IC |
85 | static struct xen_netif_rx_response *make_rx_response(struct xenvif *vif, |
86 | u16 id, | |
87 | s8 st, | |
88 | u16 offset, | |
89 | u16 size, | |
90 | u16 flags); | |
91 | ||
b3f980bd | 92 | static inline unsigned long idx_to_pfn(struct xenvif *vif, |
ea066ad1 | 93 | u16 idx) |
f942dc25 | 94 | { |
b3f980bd | 95 | return page_to_pfn(vif->mmap_pages[idx]); |
f942dc25 IC |
96 | } |
97 | ||
b3f980bd | 98 | static inline unsigned long idx_to_kaddr(struct xenvif *vif, |
ea066ad1 | 99 | u16 idx) |
f942dc25 | 100 | { |
b3f980bd | 101 | return (unsigned long)pfn_to_kaddr(idx_to_pfn(vif, idx)); |
f942dc25 IC |
102 | } |
103 | ||
2eba61d5 PD |
104 | /* This is a miniumum size for the linear area to avoid lots of |
105 | * calls to __pskb_pull_tail() as we set up checksum offsets. The | |
106 | * value 128 was chosen as it covers all IPv4 and most likely | |
107 | * IPv6 headers. | |
f942dc25 | 108 | */ |
2eba61d5 | 109 | #define PKT_PROT_LEN 128 |
f942dc25 | 110 | |
ea066ad1 IC |
111 | static u16 frag_get_pending_idx(skb_frag_t *frag) |
112 | { | |
113 | return (u16)frag->page_offset; | |
114 | } | |
115 | ||
116 | static void frag_set_pending_idx(skb_frag_t *frag, u16 pending_idx) | |
117 | { | |
118 | frag->page_offset = pending_idx; | |
119 | } | |
120 | ||
f942dc25 IC |
121 | static inline pending_ring_idx_t pending_index(unsigned i) |
122 | { | |
123 | return i & (MAX_PENDING_REQS-1); | |
124 | } | |
125 | ||
ca2f09f2 | 126 | bool xenvif_rx_ring_slots_available(struct xenvif *vif, int needed) |
f942dc25 | 127 | { |
ca2f09f2 | 128 | RING_IDX prod, cons; |
f942dc25 | 129 | |
ca2f09f2 PD |
130 | do { |
131 | prod = vif->rx.sring->req_prod; | |
132 | cons = vif->rx.req_cons; | |
f942dc25 | 133 | |
ca2f09f2 PD |
134 | if (prod - cons >= needed) |
135 | return true; | |
f942dc25 | 136 | |
ca2f09f2 | 137 | vif->rx.sring->req_event = prod + 1; |
f942dc25 | 138 | |
ca2f09f2 PD |
139 | /* Make sure event is visible before we check prod |
140 | * again. | |
141 | */ | |
142 | mb(); | |
143 | } while (vif->rx.sring->req_prod != prod); | |
f942dc25 | 144 | |
ca2f09f2 | 145 | return false; |
f942dc25 IC |
146 | } |
147 | ||
148 | /* | |
149 | * Returns true if we should start a new receive buffer instead of | |
150 | * adding 'size' bytes to a buffer which currently contains 'offset' | |
151 | * bytes. | |
152 | */ | |
153 | static bool start_new_rx_buffer(int offset, unsigned long size, int head) | |
154 | { | |
155 | /* simple case: we have completely filled the current buffer. */ | |
156 | if (offset == MAX_BUFFER_OFFSET) | |
157 | return true; | |
158 | ||
159 | /* | |
160 | * complex case: start a fresh buffer if the current frag | |
161 | * would overflow the current buffer but only if: | |
162 | * (i) this frag would fit completely in the next buffer | |
163 | * and (ii) there is already some data in the current buffer | |
164 | * and (iii) this is not the head buffer. | |
165 | * | |
166 | * Where: | |
167 | * - (i) stops us splitting a frag into two copies | |
168 | * unless the frag is too large for a single buffer. | |
169 | * - (ii) stops us from leaving a buffer pointlessly empty. | |
170 | * - (iii) stops us leaving the first buffer | |
171 | * empty. Strictly speaking this is already covered | |
172 | * by (ii) but is explicitly checked because | |
173 | * netfront relies on the first buffer being | |
174 | * non-empty and can crash otherwise. | |
175 | * | |
176 | * This means we will effectively linearise small | |
177 | * frags but do not needlessly split large buffers | |
178 | * into multiple copies tend to give large frags their | |
179 | * own buffers as before. | |
180 | */ | |
181 | if ((offset + size > MAX_BUFFER_OFFSET) && | |
182 | (size <= MAX_BUFFER_OFFSET) && offset && !head) | |
183 | return true; | |
184 | ||
185 | return false; | |
186 | } | |
187 | ||
f942dc25 IC |
188 | struct netrx_pending_operations { |
189 | unsigned copy_prod, copy_cons; | |
190 | unsigned meta_prod, meta_cons; | |
191 | struct gnttab_copy *copy; | |
b3f980bd | 192 | struct xenvif_rx_meta *meta; |
f942dc25 IC |
193 | int copy_off; |
194 | grant_ref_t copy_gref; | |
195 | }; | |
196 | ||
b3f980bd WL |
197 | static struct xenvif_rx_meta *get_next_rx_buffer(struct xenvif *vif, |
198 | struct netrx_pending_operations *npo) | |
f942dc25 | 199 | { |
b3f980bd | 200 | struct xenvif_rx_meta *meta; |
f942dc25 IC |
201 | struct xen_netif_rx_request *req; |
202 | ||
203 | req = RING_GET_REQUEST(&vif->rx, vif->rx.req_cons++); | |
204 | ||
205 | meta = npo->meta + npo->meta_prod++; | |
82cada22 | 206 | meta->gso_type = XEN_NETIF_GSO_TYPE_NONE; |
f942dc25 IC |
207 | meta->gso_size = 0; |
208 | meta->size = 0; | |
209 | meta->id = req->id; | |
210 | ||
211 | npo->copy_off = 0; | |
212 | npo->copy_gref = req->gref; | |
213 | ||
214 | return meta; | |
215 | } | |
216 | ||
33bc801d WL |
217 | /* |
218 | * Set up the grant operations for this fragment. If it's a flipping | |
219 | * interface, we also set up the unmap request from here. | |
220 | */ | |
7376419a WL |
221 | static void xenvif_gop_frag_copy(struct xenvif *vif, struct sk_buff *skb, |
222 | struct netrx_pending_operations *npo, | |
223 | struct page *page, unsigned long size, | |
33bc801d | 224 | unsigned long offset, int *head) |
f942dc25 IC |
225 | { |
226 | struct gnttab_copy *copy_gop; | |
b3f980bd | 227 | struct xenvif_rx_meta *meta; |
f942dc25 | 228 | unsigned long bytes; |
82cada22 | 229 | int gso_type; |
f942dc25 IC |
230 | |
231 | /* Data must not cross a page boundary. */ | |
6a8ed462 | 232 | BUG_ON(size + offset > PAGE_SIZE<<compound_order(page)); |
f942dc25 IC |
233 | |
234 | meta = npo->meta + npo->meta_prod - 1; | |
235 | ||
6a8ed462 IC |
236 | /* Skip unused frames from start of page */ |
237 | page += offset >> PAGE_SHIFT; | |
238 | offset &= ~PAGE_MASK; | |
239 | ||
f942dc25 | 240 | while (size > 0) { |
6a8ed462 | 241 | BUG_ON(offset >= PAGE_SIZE); |
f942dc25 IC |
242 | BUG_ON(npo->copy_off > MAX_BUFFER_OFFSET); |
243 | ||
6a8ed462 IC |
244 | bytes = PAGE_SIZE - offset; |
245 | ||
246 | if (bytes > size) | |
247 | bytes = size; | |
248 | ||
33bc801d | 249 | if (start_new_rx_buffer(npo->copy_off, bytes, *head)) { |
f942dc25 IC |
250 | /* |
251 | * Netfront requires there to be some data in the head | |
252 | * buffer. | |
253 | */ | |
33bc801d | 254 | BUG_ON(*head); |
f942dc25 IC |
255 | |
256 | meta = get_next_rx_buffer(vif, npo); | |
257 | } | |
258 | ||
f942dc25 IC |
259 | if (npo->copy_off + bytes > MAX_BUFFER_OFFSET) |
260 | bytes = MAX_BUFFER_OFFSET - npo->copy_off; | |
261 | ||
262 | copy_gop = npo->copy + npo->copy_prod++; | |
263 | copy_gop->flags = GNTCOPY_dest_gref; | |
b3f980bd WL |
264 | copy_gop->len = bytes; |
265 | ||
43e9d194 WL |
266 | copy_gop->source.domid = DOMID_SELF; |
267 | copy_gop->source.u.gmfn = virt_to_mfn(page_address(page)); | |
f942dc25 | 268 | copy_gop->source.offset = offset; |
f942dc25 | 269 | |
b3f980bd | 270 | copy_gop->dest.domid = vif->domid; |
f942dc25 IC |
271 | copy_gop->dest.offset = npo->copy_off; |
272 | copy_gop->dest.u.ref = npo->copy_gref; | |
f942dc25 IC |
273 | |
274 | npo->copy_off += bytes; | |
275 | meta->size += bytes; | |
276 | ||
277 | offset += bytes; | |
278 | size -= bytes; | |
279 | ||
6a8ed462 IC |
280 | /* Next frame */ |
281 | if (offset == PAGE_SIZE && size) { | |
282 | BUG_ON(!PageCompound(page)); | |
283 | page++; | |
284 | offset = 0; | |
285 | } | |
286 | ||
f942dc25 | 287 | /* Leave a gap for the GSO descriptor. */ |
82cada22 PD |
288 | if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) |
289 | gso_type = XEN_NETIF_GSO_TYPE_TCPV4; | |
290 | else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) | |
291 | gso_type = XEN_NETIF_GSO_TYPE_TCPV6; | |
292 | else | |
293 | gso_type = XEN_NETIF_GSO_TYPE_NONE; | |
294 | ||
295 | if (*head && ((1 << gso_type) & vif->gso_mask)) | |
f942dc25 IC |
296 | vif->rx.req_cons++; |
297 | ||
33bc801d | 298 | *head = 0; /* There must be something in this buffer now. */ |
f942dc25 IC |
299 | |
300 | } | |
301 | } | |
302 | ||
303 | /* | |
304 | * Prepare an SKB to be transmitted to the frontend. | |
305 | * | |
306 | * This function is responsible for allocating grant operations, meta | |
307 | * structures, etc. | |
308 | * | |
309 | * It returns the number of meta structures consumed. The number of | |
310 | * ring slots used is always equal to the number of meta slots used | |
311 | * plus the number of GSO descriptors used. Currently, we use either | |
312 | * zero GSO descriptors (for non-GSO packets) or one descriptor (for | |
313 | * frontend-side LRO). | |
314 | */ | |
7376419a WL |
315 | static int xenvif_gop_skb(struct sk_buff *skb, |
316 | struct netrx_pending_operations *npo) | |
f942dc25 IC |
317 | { |
318 | struct xenvif *vif = netdev_priv(skb->dev); | |
319 | int nr_frags = skb_shinfo(skb)->nr_frags; | |
320 | int i; | |
321 | struct xen_netif_rx_request *req; | |
b3f980bd | 322 | struct xenvif_rx_meta *meta; |
f942dc25 | 323 | unsigned char *data; |
33bc801d | 324 | int head = 1; |
f942dc25 | 325 | int old_meta_prod; |
82cada22 PD |
326 | int gso_type; |
327 | int gso_size; | |
f942dc25 IC |
328 | |
329 | old_meta_prod = npo->meta_prod; | |
330 | ||
82cada22 PD |
331 | if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) { |
332 | gso_type = XEN_NETIF_GSO_TYPE_TCPV4; | |
333 | gso_size = skb_shinfo(skb)->gso_size; | |
334 | } else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) { | |
335 | gso_type = XEN_NETIF_GSO_TYPE_TCPV6; | |
336 | gso_size = skb_shinfo(skb)->gso_size; | |
337 | } else { | |
338 | gso_type = XEN_NETIF_GSO_TYPE_NONE; | |
339 | gso_size = 0; | |
340 | } | |
341 | ||
f942dc25 | 342 | /* Set up a GSO prefix descriptor, if necessary */ |
a3314f3d | 343 | if ((1 << gso_type) & vif->gso_prefix_mask) { |
f942dc25 IC |
344 | req = RING_GET_REQUEST(&vif->rx, vif->rx.req_cons++); |
345 | meta = npo->meta + npo->meta_prod++; | |
82cada22 PD |
346 | meta->gso_type = gso_type; |
347 | meta->gso_size = gso_size; | |
f942dc25 IC |
348 | meta->size = 0; |
349 | meta->id = req->id; | |
350 | } | |
351 | ||
352 | req = RING_GET_REQUEST(&vif->rx, vif->rx.req_cons++); | |
353 | meta = npo->meta + npo->meta_prod++; | |
354 | ||
82cada22 PD |
355 | if ((1 << gso_type) & vif->gso_mask) { |
356 | meta->gso_type = gso_type; | |
357 | meta->gso_size = gso_size; | |
358 | } else { | |
359 | meta->gso_type = XEN_NETIF_GSO_TYPE_NONE; | |
f942dc25 | 360 | meta->gso_size = 0; |
82cada22 | 361 | } |
f942dc25 IC |
362 | |
363 | meta->size = 0; | |
364 | meta->id = req->id; | |
365 | npo->copy_off = 0; | |
366 | npo->copy_gref = req->gref; | |
367 | ||
368 | data = skb->data; | |
369 | while (data < skb_tail_pointer(skb)) { | |
370 | unsigned int offset = offset_in_page(data); | |
371 | unsigned int len = PAGE_SIZE - offset; | |
372 | ||
373 | if (data + len > skb_tail_pointer(skb)) | |
374 | len = skb_tail_pointer(skb) - data; | |
375 | ||
7376419a | 376 | xenvif_gop_frag_copy(vif, skb, npo, |
33bc801d | 377 | virt_to_page(data), len, offset, &head); |
f942dc25 IC |
378 | data += len; |
379 | } | |
380 | ||
381 | for (i = 0; i < nr_frags; i++) { | |
7376419a WL |
382 | xenvif_gop_frag_copy(vif, skb, npo, |
383 | skb_frag_page(&skb_shinfo(skb)->frags[i]), | |
384 | skb_frag_size(&skb_shinfo(skb)->frags[i]), | |
385 | skb_shinfo(skb)->frags[i].page_offset, | |
33bc801d | 386 | &head); |
f942dc25 IC |
387 | } |
388 | ||
389 | return npo->meta_prod - old_meta_prod; | |
390 | } | |
391 | ||
392 | /* | |
7376419a | 393 | * This is a twin to xenvif_gop_skb. Assume that xenvif_gop_skb was |
f942dc25 IC |
394 | * used to set up the operations on the top of |
395 | * netrx_pending_operations, which have since been done. Check that | |
396 | * they didn't give any errors and advance over them. | |
397 | */ | |
7376419a WL |
398 | static int xenvif_check_gop(struct xenvif *vif, int nr_meta_slots, |
399 | struct netrx_pending_operations *npo) | |
f942dc25 IC |
400 | { |
401 | struct gnttab_copy *copy_op; | |
402 | int status = XEN_NETIF_RSP_OKAY; | |
403 | int i; | |
404 | ||
405 | for (i = 0; i < nr_meta_slots; i++) { | |
406 | copy_op = npo->copy + npo->copy_cons++; | |
407 | if (copy_op->status != GNTST_okay) { | |
408 | netdev_dbg(vif->dev, | |
409 | "Bad status %d from copy to DOM%d.\n", | |
410 | copy_op->status, vif->domid); | |
411 | status = XEN_NETIF_RSP_ERROR; | |
412 | } | |
413 | } | |
414 | ||
415 | return status; | |
416 | } | |
417 | ||
7376419a WL |
418 | static void xenvif_add_frag_responses(struct xenvif *vif, int status, |
419 | struct xenvif_rx_meta *meta, | |
420 | int nr_meta_slots) | |
f942dc25 IC |
421 | { |
422 | int i; | |
423 | unsigned long offset; | |
424 | ||
425 | /* No fragments used */ | |
426 | if (nr_meta_slots <= 1) | |
427 | return; | |
428 | ||
429 | nr_meta_slots--; | |
430 | ||
431 | for (i = 0; i < nr_meta_slots; i++) { | |
432 | int flags; | |
433 | if (i == nr_meta_slots - 1) | |
434 | flags = 0; | |
435 | else | |
436 | flags = XEN_NETRXF_more_data; | |
437 | ||
438 | offset = 0; | |
439 | make_rx_response(vif, meta[i].id, status, offset, | |
440 | meta[i].size, flags); | |
441 | } | |
442 | } | |
443 | ||
8f13dd96 | 444 | struct xenvif_rx_cb { |
33bc801d WL |
445 | int meta_slots_used; |
446 | }; | |
447 | ||
8f13dd96 ZK |
448 | #define XENVIF_RX_CB(skb) ((struct xenvif_rx_cb *)(skb)->cb) |
449 | ||
ca2f09f2 | 450 | void xenvif_kick_thread(struct xenvif *vif) |
b3f980bd WL |
451 | { |
452 | wake_up(&vif->wq); | |
453 | } | |
454 | ||
ca2f09f2 | 455 | static void xenvif_rx_action(struct xenvif *vif) |
f942dc25 | 456 | { |
f942dc25 | 457 | s8 status; |
e1f00a69 | 458 | u16 flags; |
f942dc25 IC |
459 | struct xen_netif_rx_response *resp; |
460 | struct sk_buff_head rxq; | |
461 | struct sk_buff *skb; | |
462 | LIST_HEAD(notify); | |
463 | int ret; | |
f942dc25 | 464 | unsigned long offset; |
11b57f90 | 465 | bool need_to_notify = false; |
f942dc25 IC |
466 | |
467 | struct netrx_pending_operations npo = { | |
b3f980bd WL |
468 | .copy = vif->grant_copy_op, |
469 | .meta = vif->meta, | |
f942dc25 IC |
470 | }; |
471 | ||
472 | skb_queue_head_init(&rxq); | |
473 | ||
b3f980bd | 474 | while ((skb = skb_dequeue(&vif->rx_queue)) != NULL) { |
9ab9831b | 475 | RING_IDX max_slots_needed; |
ca2f09f2 PD |
476 | int i; |
477 | ||
478 | /* We need a cheap worse case estimate for the number of | |
479 | * slots we'll use. | |
480 | */ | |
481 | ||
482 | max_slots_needed = DIV_ROUND_UP(offset_in_page(skb->data) + | |
483 | skb_headlen(skb), | |
484 | PAGE_SIZE); | |
485 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { | |
486 | unsigned int size; | |
487 | size = skb_frag_size(&skb_shinfo(skb)->frags[i]); | |
488 | max_slots_needed += DIV_ROUND_UP(size, PAGE_SIZE); | |
489 | } | |
490 | if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4 || | |
491 | skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) | |
492 | max_slots_needed++; | |
493 | ||
494 | /* If the skb may not fit then bail out now */ | |
495 | if (!xenvif_rx_ring_slots_available(vif, max_slots_needed)) { | |
496 | skb_queue_head(&vif->rx_queue, skb); | |
11b57f90 | 497 | need_to_notify = true; |
9ab9831b | 498 | vif->rx_last_skb_slots = max_slots_needed; |
ca2f09f2 | 499 | break; |
9ab9831b ZK |
500 | } else |
501 | vif->rx_last_skb_slots = 0; | |
f942dc25 | 502 | |
8f13dd96 ZK |
503 | XENVIF_RX_CB(skb)->meta_slots_used = xenvif_gop_skb(skb, &npo); |
504 | BUG_ON(XENVIF_RX_CB(skb)->meta_slots_used > max_slots_needed); | |
f942dc25 IC |
505 | |
506 | __skb_queue_tail(&rxq, skb); | |
f942dc25 IC |
507 | } |
508 | ||
b3f980bd | 509 | BUG_ON(npo.meta_prod > ARRAY_SIZE(vif->meta)); |
f942dc25 IC |
510 | |
511 | if (!npo.copy_prod) | |
ca2f09f2 | 512 | goto done; |
f942dc25 | 513 | |
ac3d5ac2 | 514 | BUG_ON(npo.copy_prod > MAX_GRANT_COPY_OPS); |
b3f980bd | 515 | gnttab_batch_copy(vif->grant_copy_op, npo.copy_prod); |
f942dc25 IC |
516 | |
517 | while ((skb = __skb_dequeue(&rxq)) != NULL) { | |
f942dc25 | 518 | |
82cada22 PD |
519 | if ((1 << vif->meta[npo.meta_cons].gso_type) & |
520 | vif->gso_prefix_mask) { | |
f942dc25 | 521 | resp = RING_GET_RESPONSE(&vif->rx, |
b3f980bd | 522 | vif->rx.rsp_prod_pvt++); |
f942dc25 IC |
523 | |
524 | resp->flags = XEN_NETRXF_gso_prefix | XEN_NETRXF_more_data; | |
525 | ||
b3f980bd WL |
526 | resp->offset = vif->meta[npo.meta_cons].gso_size; |
527 | resp->id = vif->meta[npo.meta_cons].id; | |
8f13dd96 | 528 | resp->status = XENVIF_RX_CB(skb)->meta_slots_used; |
f942dc25 IC |
529 | |
530 | npo.meta_cons++; | |
8f13dd96 | 531 | XENVIF_RX_CB(skb)->meta_slots_used--; |
f942dc25 IC |
532 | } |
533 | ||
534 | ||
535 | vif->dev->stats.tx_bytes += skb->len; | |
536 | vif->dev->stats.tx_packets++; | |
537 | ||
8f13dd96 ZK |
538 | status = xenvif_check_gop(vif, |
539 | XENVIF_RX_CB(skb)->meta_slots_used, | |
540 | &npo); | |
f942dc25 | 541 | |
8f13dd96 | 542 | if (XENVIF_RX_CB(skb)->meta_slots_used == 1) |
f942dc25 IC |
543 | flags = 0; |
544 | else | |
545 | flags = XEN_NETRXF_more_data; | |
546 | ||
547 | if (skb->ip_summed == CHECKSUM_PARTIAL) /* local packet? */ | |
548 | flags |= XEN_NETRXF_csum_blank | XEN_NETRXF_data_validated; | |
549 | else if (skb->ip_summed == CHECKSUM_UNNECESSARY) | |
550 | /* remote but checksummed. */ | |
551 | flags |= XEN_NETRXF_data_validated; | |
552 | ||
553 | offset = 0; | |
b3f980bd | 554 | resp = make_rx_response(vif, vif->meta[npo.meta_cons].id, |
f942dc25 | 555 | status, offset, |
b3f980bd | 556 | vif->meta[npo.meta_cons].size, |
f942dc25 IC |
557 | flags); |
558 | ||
82cada22 PD |
559 | if ((1 << vif->meta[npo.meta_cons].gso_type) & |
560 | vif->gso_mask) { | |
f942dc25 IC |
561 | struct xen_netif_extra_info *gso = |
562 | (struct xen_netif_extra_info *) | |
563 | RING_GET_RESPONSE(&vif->rx, | |
564 | vif->rx.rsp_prod_pvt++); | |
565 | ||
566 | resp->flags |= XEN_NETRXF_extra_info; | |
567 | ||
82cada22 | 568 | gso->u.gso.type = vif->meta[npo.meta_cons].gso_type; |
b3f980bd | 569 | gso->u.gso.size = vif->meta[npo.meta_cons].gso_size; |
f942dc25 IC |
570 | gso->u.gso.pad = 0; |
571 | gso->u.gso.features = 0; | |
572 | ||
573 | gso->type = XEN_NETIF_EXTRA_TYPE_GSO; | |
574 | gso->flags = 0; | |
575 | } | |
576 | ||
7376419a WL |
577 | xenvif_add_frag_responses(vif, status, |
578 | vif->meta + npo.meta_cons + 1, | |
8f13dd96 | 579 | XENVIF_RX_CB(skb)->meta_slots_used); |
f942dc25 IC |
580 | |
581 | RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&vif->rx, ret); | |
f942dc25 | 582 | |
11b57f90 | 583 | need_to_notify |= !!ret; |
b3f980bd | 584 | |
8f13dd96 | 585 | npo.meta_cons += XENVIF_RX_CB(skb)->meta_slots_used; |
f942dc25 IC |
586 | dev_kfree_skb(skb); |
587 | } | |
588 | ||
ca2f09f2 | 589 | done: |
b3f980bd | 590 | if (need_to_notify) |
e1f00a69 | 591 | notify_remote_via_irq(vif->rx_irq); |
f942dc25 IC |
592 | } |
593 | ||
7376419a | 594 | void xenvif_check_rx_xenvif(struct xenvif *vif) |
f942dc25 IC |
595 | { |
596 | int more_to_do; | |
597 | ||
598 | RING_FINAL_CHECK_FOR_REQUESTS(&vif->tx, more_to_do); | |
599 | ||
600 | if (more_to_do) | |
b3f980bd | 601 | napi_schedule(&vif->napi); |
f942dc25 IC |
602 | } |
603 | ||
604 | static void tx_add_credit(struct xenvif *vif) | |
605 | { | |
606 | unsigned long max_burst, max_credit; | |
607 | ||
608 | /* | |
609 | * Allow a burst big enough to transmit a jumbo packet of up to 128kB. | |
610 | * Otherwise the interface can seize up due to insufficient credit. | |
611 | */ | |
612 | max_burst = RING_GET_REQUEST(&vif->tx, vif->tx.req_cons)->size; | |
613 | max_burst = min(max_burst, 131072UL); | |
614 | max_burst = max(max_burst, vif->credit_bytes); | |
615 | ||
616 | /* Take care that adding a new chunk of credit doesn't wrap to zero. */ | |
617 | max_credit = vif->remaining_credit + vif->credit_bytes; | |
618 | if (max_credit < vif->remaining_credit) | |
619 | max_credit = ULONG_MAX; /* wrapped: clamp to ULONG_MAX */ | |
620 | ||
621 | vif->remaining_credit = min(max_credit, max_burst); | |
622 | } | |
623 | ||
624 | static void tx_credit_callback(unsigned long data) | |
625 | { | |
626 | struct xenvif *vif = (struct xenvif *)data; | |
627 | tx_add_credit(vif); | |
7376419a | 628 | xenvif_check_rx_xenvif(vif); |
f942dc25 IC |
629 | } |
630 | ||
7376419a WL |
631 | static void xenvif_tx_err(struct xenvif *vif, |
632 | struct xen_netif_tx_request *txp, RING_IDX end) | |
f942dc25 IC |
633 | { |
634 | RING_IDX cons = vif->tx.req_cons; | |
635 | ||
636 | do { | |
637 | make_tx_response(vif, txp, XEN_NETIF_RSP_ERROR); | |
b9149729 | 638 | if (cons == end) |
f942dc25 IC |
639 | break; |
640 | txp = RING_GET_REQUEST(&vif->tx, cons++); | |
641 | } while (1); | |
642 | vif->tx.req_cons = cons; | |
f942dc25 IC |
643 | } |
644 | ||
7376419a | 645 | static void xenvif_fatal_tx_err(struct xenvif *vif) |
48856286 IC |
646 | { |
647 | netdev_err(vif->dev, "fatal error; disabling device\n"); | |
648 | xenvif_carrier_off(vif); | |
48856286 IC |
649 | } |
650 | ||
7376419a WL |
651 | static int xenvif_count_requests(struct xenvif *vif, |
652 | struct xen_netif_tx_request *first, | |
653 | struct xen_netif_tx_request *txp, | |
654 | int work_to_do) | |
f942dc25 IC |
655 | { |
656 | RING_IDX cons = vif->tx.req_cons; | |
2810e5b9 WL |
657 | int slots = 0; |
658 | int drop_err = 0; | |
59ccb4eb | 659 | int more_data; |
f942dc25 IC |
660 | |
661 | if (!(first->flags & XEN_NETTXF_more_data)) | |
662 | return 0; | |
663 | ||
664 | do { | |
59ccb4eb WL |
665 | struct xen_netif_tx_request dropped_tx = { 0 }; |
666 | ||
2810e5b9 WL |
667 | if (slots >= work_to_do) { |
668 | netdev_err(vif->dev, | |
669 | "Asked for %d slots but exceeds this limit\n", | |
670 | work_to_do); | |
7376419a | 671 | xenvif_fatal_tx_err(vif); |
35876b5f | 672 | return -ENODATA; |
f942dc25 IC |
673 | } |
674 | ||
2810e5b9 WL |
675 | /* This guest is really using too many slots and |
676 | * considered malicious. | |
677 | */ | |
37641494 | 678 | if (unlikely(slots >= fatal_skb_slots)) { |
2810e5b9 WL |
679 | netdev_err(vif->dev, |
680 | "Malicious frontend using %d slots, threshold %u\n", | |
37641494 | 681 | slots, fatal_skb_slots); |
7376419a | 682 | xenvif_fatal_tx_err(vif); |
35876b5f | 683 | return -E2BIG; |
f942dc25 IC |
684 | } |
685 | ||
2810e5b9 | 686 | /* Xen network protocol had implicit dependency on |
37641494 WL |
687 | * MAX_SKB_FRAGS. XEN_NETBK_LEGACY_SLOTS_MAX is set to |
688 | * the historical MAX_SKB_FRAGS value 18 to honor the | |
689 | * same behavior as before. Any packet using more than | |
690 | * 18 slots but less than fatal_skb_slots slots is | |
691 | * dropped | |
2810e5b9 | 692 | */ |
37641494 | 693 | if (!drop_err && slots >= XEN_NETBK_LEGACY_SLOTS_MAX) { |
2810e5b9 WL |
694 | if (net_ratelimit()) |
695 | netdev_dbg(vif->dev, | |
696 | "Too many slots (%d) exceeding limit (%d), dropping packet\n", | |
37641494 | 697 | slots, XEN_NETBK_LEGACY_SLOTS_MAX); |
2810e5b9 WL |
698 | drop_err = -E2BIG; |
699 | } | |
700 | ||
59ccb4eb WL |
701 | if (drop_err) |
702 | txp = &dropped_tx; | |
703 | ||
2810e5b9 | 704 | memcpy(txp, RING_GET_REQUEST(&vif->tx, cons + slots), |
f942dc25 | 705 | sizeof(*txp)); |
03393fd5 WL |
706 | |
707 | /* If the guest submitted a frame >= 64 KiB then | |
708 | * first->size overflowed and following slots will | |
709 | * appear to be larger than the frame. | |
710 | * | |
711 | * This cannot be fatal error as there are buggy | |
712 | * frontends that do this. | |
713 | * | |
714 | * Consume all slots and drop the packet. | |
715 | */ | |
716 | if (!drop_err && txp->size > first->size) { | |
717 | if (net_ratelimit()) | |
718 | netdev_dbg(vif->dev, | |
719 | "Invalid tx request, slot size %u > remaining size %u\n", | |
720 | txp->size, first->size); | |
721 | drop_err = -EIO; | |
f942dc25 IC |
722 | } |
723 | ||
724 | first->size -= txp->size; | |
2810e5b9 | 725 | slots++; |
f942dc25 IC |
726 | |
727 | if (unlikely((txp->offset + txp->size) > PAGE_SIZE)) { | |
2810e5b9 | 728 | netdev_err(vif->dev, "Cross page boundary, txp->offset: %x, size: %u\n", |
f942dc25 | 729 | txp->offset, txp->size); |
7376419a | 730 | xenvif_fatal_tx_err(vif); |
35876b5f | 731 | return -EINVAL; |
f942dc25 | 732 | } |
59ccb4eb WL |
733 | |
734 | more_data = txp->flags & XEN_NETTXF_more_data; | |
735 | ||
736 | if (!drop_err) | |
737 | txp++; | |
738 | ||
739 | } while (more_data); | |
2810e5b9 WL |
740 | |
741 | if (drop_err) { | |
7376419a | 742 | xenvif_tx_err(vif, first, cons + slots); |
2810e5b9 WL |
743 | return drop_err; |
744 | } | |
745 | ||
746 | return slots; | |
f942dc25 IC |
747 | } |
748 | ||
7376419a WL |
749 | static struct page *xenvif_alloc_page(struct xenvif *vif, |
750 | u16 pending_idx) | |
f942dc25 IC |
751 | { |
752 | struct page *page; | |
b3f980bd WL |
753 | |
754 | page = alloc_page(GFP_ATOMIC|__GFP_COLD); | |
f942dc25 IC |
755 | if (!page) |
756 | return NULL; | |
b3f980bd WL |
757 | vif->mmap_pages[pending_idx] = page; |
758 | ||
f942dc25 IC |
759 | return page; |
760 | } | |
761 | ||
8f13dd96 ZK |
762 | |
763 | struct xenvif_tx_cb { | |
764 | u16 pending_idx; | |
765 | }; | |
766 | ||
767 | #define XENVIF_TX_CB(skb) ((struct xenvif_tx_cb *)(skb)->cb) | |
768 | ||
7376419a WL |
769 | static struct gnttab_copy *xenvif_get_requests(struct xenvif *vif, |
770 | struct sk_buff *skb, | |
771 | struct xen_netif_tx_request *txp, | |
772 | struct gnttab_copy *gop) | |
f942dc25 IC |
773 | { |
774 | struct skb_shared_info *shinfo = skb_shinfo(skb); | |
775 | skb_frag_t *frags = shinfo->frags; | |
8f13dd96 | 776 | u16 pending_idx = XENVIF_TX_CB(skb)->pending_idx; |
2810e5b9 WL |
777 | u16 head_idx = 0; |
778 | int slot, start; | |
779 | struct page *page; | |
780 | pending_ring_idx_t index, start_idx = 0; | |
781 | uint16_t dst_offset; | |
782 | unsigned int nr_slots; | |
783 | struct pending_tx_info *first = NULL; | |
784 | ||
785 | /* At this point shinfo->nr_frags is in fact the number of | |
37641494 | 786 | * slots, which can be as large as XEN_NETBK_LEGACY_SLOTS_MAX. |
2810e5b9 WL |
787 | */ |
788 | nr_slots = shinfo->nr_frags; | |
f942dc25 IC |
789 | |
790 | /* Skip first skb fragment if it is on same page as header fragment. */ | |
ea066ad1 | 791 | start = (frag_get_pending_idx(&shinfo->frags[0]) == pending_idx); |
f942dc25 | 792 | |
2810e5b9 WL |
793 | /* Coalesce tx requests, at this point the packet passed in |
794 | * should be <= 64K. Any packets larger than 64K have been | |
7376419a | 795 | * handled in xenvif_count_requests(). |
2810e5b9 WL |
796 | */ |
797 | for (shinfo->nr_frags = slot = start; slot < nr_slots; | |
798 | shinfo->nr_frags++) { | |
f942dc25 | 799 | struct pending_tx_info *pending_tx_info = |
b3f980bd | 800 | vif->pending_tx_info; |
f942dc25 | 801 | |
b3f980bd | 802 | page = alloc_page(GFP_ATOMIC|__GFP_COLD); |
f942dc25 | 803 | if (!page) |
4cc7c1cb | 804 | goto err; |
f942dc25 | 805 | |
2810e5b9 WL |
806 | dst_offset = 0; |
807 | first = NULL; | |
808 | while (dst_offset < PAGE_SIZE && slot < nr_slots) { | |
809 | gop->flags = GNTCOPY_source_gref; | |
810 | ||
811 | gop->source.u.ref = txp->gref; | |
812 | gop->source.domid = vif->domid; | |
813 | gop->source.offset = txp->offset; | |
814 | ||
815 | gop->dest.domid = DOMID_SELF; | |
816 | ||
817 | gop->dest.offset = dst_offset; | |
818 | gop->dest.u.gmfn = virt_to_mfn(page_address(page)); | |
819 | ||
820 | if (dst_offset + txp->size > PAGE_SIZE) { | |
821 | /* This page can only merge a portion | |
822 | * of tx request. Do not increment any | |
823 | * pointer / counter here. The txp | |
824 | * will be dealt with in future | |
825 | * rounds, eventually hitting the | |
826 | * `else` branch. | |
827 | */ | |
828 | gop->len = PAGE_SIZE - dst_offset; | |
829 | txp->offset += gop->len; | |
830 | txp->size -= gop->len; | |
831 | dst_offset += gop->len; /* quit loop */ | |
832 | } else { | |
833 | /* This tx request can be merged in the page */ | |
834 | gop->len = txp->size; | |
835 | dst_offset += gop->len; | |
836 | ||
b3f980bd | 837 | index = pending_index(vif->pending_cons++); |
2810e5b9 | 838 | |
b3f980bd | 839 | pending_idx = vif->pending_ring[index]; |
2810e5b9 WL |
840 | |
841 | memcpy(&pending_tx_info[pending_idx].req, txp, | |
842 | sizeof(*txp)); | |
2810e5b9 WL |
843 | |
844 | /* Poison these fields, corresponding | |
845 | * fields for head tx req will be set | |
846 | * to correct values after the loop. | |
847 | */ | |
b3f980bd | 848 | vif->mmap_pages[pending_idx] = (void *)(~0UL); |
2810e5b9 WL |
849 | pending_tx_info[pending_idx].head = |
850 | INVALID_PENDING_RING_IDX; | |
851 | ||
852 | if (!first) { | |
853 | first = &pending_tx_info[pending_idx]; | |
854 | start_idx = index; | |
855 | head_idx = pending_idx; | |
856 | } | |
857 | ||
858 | txp++; | |
859 | slot++; | |
860 | } | |
f942dc25 | 861 | |
2810e5b9 WL |
862 | gop++; |
863 | } | |
f942dc25 | 864 | |
2810e5b9 WL |
865 | first->req.offset = 0; |
866 | first->req.size = dst_offset; | |
867 | first->head = start_idx; | |
b3f980bd | 868 | vif->mmap_pages[head_idx] = page; |
2810e5b9 | 869 | frag_set_pending_idx(&frags[shinfo->nr_frags], head_idx); |
f942dc25 IC |
870 | } |
871 | ||
2810e5b9 WL |
872 | BUG_ON(shinfo->nr_frags > MAX_SKB_FRAGS); |
873 | ||
f942dc25 | 874 | return gop; |
4cc7c1cb IC |
875 | err: |
876 | /* Unwind, freeing all pages and sending error responses. */ | |
2810e5b9 | 877 | while (shinfo->nr_frags-- > start) { |
7376419a | 878 | xenvif_idx_release(vif, |
2810e5b9 WL |
879 | frag_get_pending_idx(&frags[shinfo->nr_frags]), |
880 | XEN_NETIF_RSP_ERROR); | |
4cc7c1cb IC |
881 | } |
882 | /* The head too, if necessary. */ | |
883 | if (start) | |
7376419a | 884 | xenvif_idx_release(vif, pending_idx, XEN_NETIF_RSP_ERROR); |
4cc7c1cb IC |
885 | |
886 | return NULL; | |
f942dc25 IC |
887 | } |
888 | ||
7376419a WL |
889 | static int xenvif_tx_check_gop(struct xenvif *vif, |
890 | struct sk_buff *skb, | |
891 | struct gnttab_copy **gopp) | |
f942dc25 IC |
892 | { |
893 | struct gnttab_copy *gop = *gopp; | |
8f13dd96 | 894 | u16 pending_idx = XENVIF_TX_CB(skb)->pending_idx; |
f942dc25 | 895 | struct skb_shared_info *shinfo = skb_shinfo(skb); |
2810e5b9 | 896 | struct pending_tx_info *tx_info; |
f942dc25 IC |
897 | int nr_frags = shinfo->nr_frags; |
898 | int i, err, start; | |
2810e5b9 | 899 | u16 peek; /* peek into next tx request */ |
f942dc25 IC |
900 | |
901 | /* Check status of header. */ | |
902 | err = gop->status; | |
7d5145d8 | 903 | if (unlikely(err)) |
7376419a | 904 | xenvif_idx_release(vif, pending_idx, XEN_NETIF_RSP_ERROR); |
f942dc25 IC |
905 | |
906 | /* Skip first skb fragment if it is on same page as header fragment. */ | |
ea066ad1 | 907 | start = (frag_get_pending_idx(&shinfo->frags[0]) == pending_idx); |
f942dc25 IC |
908 | |
909 | for (i = start; i < nr_frags; i++) { | |
910 | int j, newerr; | |
2810e5b9 | 911 | pending_ring_idx_t head; |
f942dc25 | 912 | |
ea066ad1 | 913 | pending_idx = frag_get_pending_idx(&shinfo->frags[i]); |
b3f980bd | 914 | tx_info = &vif->pending_tx_info[pending_idx]; |
2810e5b9 | 915 | head = tx_info->head; |
f942dc25 IC |
916 | |
917 | /* Check error status: if okay then remember grant handle. */ | |
2810e5b9 WL |
918 | do { |
919 | newerr = (++gop)->status; | |
920 | if (newerr) | |
921 | break; | |
b3f980bd WL |
922 | peek = vif->pending_ring[pending_index(++head)]; |
923 | } while (!pending_tx_is_head(vif, peek)); | |
2810e5b9 | 924 | |
f942dc25 IC |
925 | if (likely(!newerr)) { |
926 | /* Had a previous error? Invalidate this fragment. */ | |
927 | if (unlikely(err)) | |
7376419a WL |
928 | xenvif_idx_release(vif, pending_idx, |
929 | XEN_NETIF_RSP_OKAY); | |
f942dc25 IC |
930 | continue; |
931 | } | |
932 | ||
933 | /* Error on this fragment: respond to client with an error. */ | |
7376419a | 934 | xenvif_idx_release(vif, pending_idx, XEN_NETIF_RSP_ERROR); |
f942dc25 IC |
935 | |
936 | /* Not the first error? Preceding frags already invalidated. */ | |
937 | if (err) | |
938 | continue; | |
939 | ||
940 | /* First error: invalidate header and preceding fragments. */ | |
8f13dd96 | 941 | pending_idx = XENVIF_TX_CB(skb)->pending_idx; |
7376419a | 942 | xenvif_idx_release(vif, pending_idx, XEN_NETIF_RSP_OKAY); |
f942dc25 | 943 | for (j = start; j < i; j++) { |
5ccb3ea7 | 944 | pending_idx = frag_get_pending_idx(&shinfo->frags[j]); |
7376419a WL |
945 | xenvif_idx_release(vif, pending_idx, |
946 | XEN_NETIF_RSP_OKAY); | |
f942dc25 IC |
947 | } |
948 | ||
949 | /* Remember the error: invalidate all subsequent fragments. */ | |
950 | err = newerr; | |
951 | } | |
952 | ||
953 | *gopp = gop + 1; | |
954 | return err; | |
955 | } | |
956 | ||
7376419a | 957 | static void xenvif_fill_frags(struct xenvif *vif, struct sk_buff *skb) |
f942dc25 IC |
958 | { |
959 | struct skb_shared_info *shinfo = skb_shinfo(skb); | |
960 | int nr_frags = shinfo->nr_frags; | |
961 | int i; | |
962 | ||
963 | for (i = 0; i < nr_frags; i++) { | |
964 | skb_frag_t *frag = shinfo->frags + i; | |
965 | struct xen_netif_tx_request *txp; | |
ea066ad1 IC |
966 | struct page *page; |
967 | u16 pending_idx; | |
f942dc25 | 968 | |
ea066ad1 | 969 | pending_idx = frag_get_pending_idx(frag); |
f942dc25 | 970 | |
b3f980bd WL |
971 | txp = &vif->pending_tx_info[pending_idx].req; |
972 | page = virt_to_page(idx_to_kaddr(vif, pending_idx)); | |
ea066ad1 | 973 | __skb_fill_page_desc(skb, i, page, txp->offset, txp->size); |
f942dc25 IC |
974 | skb->len += txp->size; |
975 | skb->data_len += txp->size; | |
976 | skb->truesize += txp->size; | |
977 | ||
7376419a | 978 | /* Take an extra reference to offset xenvif_idx_release */ |
b3f980bd | 979 | get_page(vif->mmap_pages[pending_idx]); |
7376419a | 980 | xenvif_idx_release(vif, pending_idx, XEN_NETIF_RSP_OKAY); |
f942dc25 IC |
981 | } |
982 | } | |
983 | ||
7376419a | 984 | static int xenvif_get_extras(struct xenvif *vif, |
f942dc25 IC |
985 | struct xen_netif_extra_info *extras, |
986 | int work_to_do) | |
987 | { | |
988 | struct xen_netif_extra_info extra; | |
989 | RING_IDX cons = vif->tx.req_cons; | |
990 | ||
991 | do { | |
992 | if (unlikely(work_to_do-- <= 0)) { | |
48856286 | 993 | netdev_err(vif->dev, "Missing extra info\n"); |
7376419a | 994 | xenvif_fatal_tx_err(vif); |
f942dc25 IC |
995 | return -EBADR; |
996 | } | |
997 | ||
998 | memcpy(&extra, RING_GET_REQUEST(&vif->tx, cons), | |
999 | sizeof(extra)); | |
1000 | if (unlikely(!extra.type || | |
1001 | extra.type >= XEN_NETIF_EXTRA_TYPE_MAX)) { | |
1002 | vif->tx.req_cons = ++cons; | |
48856286 | 1003 | netdev_err(vif->dev, |
f942dc25 | 1004 | "Invalid extra type: %d\n", extra.type); |
7376419a | 1005 | xenvif_fatal_tx_err(vif); |
f942dc25 IC |
1006 | return -EINVAL; |
1007 | } | |
1008 | ||
1009 | memcpy(&extras[extra.type - 1], &extra, sizeof(extra)); | |
1010 | vif->tx.req_cons = ++cons; | |
1011 | } while (extra.flags & XEN_NETIF_EXTRA_FLAG_MORE); | |
1012 | ||
1013 | return work_to_do; | |
1014 | } | |
1015 | ||
7376419a WL |
1016 | static int xenvif_set_skb_gso(struct xenvif *vif, |
1017 | struct sk_buff *skb, | |
1018 | struct xen_netif_extra_info *gso) | |
f942dc25 IC |
1019 | { |
1020 | if (!gso->u.gso.size) { | |
48856286 | 1021 | netdev_err(vif->dev, "GSO size must not be zero.\n"); |
7376419a | 1022 | xenvif_fatal_tx_err(vif); |
f942dc25 IC |
1023 | return -EINVAL; |
1024 | } | |
1025 | ||
a9468587 PD |
1026 | switch (gso->u.gso.type) { |
1027 | case XEN_NETIF_GSO_TYPE_TCPV4: | |
1028 | skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4; | |
1029 | break; | |
1030 | case XEN_NETIF_GSO_TYPE_TCPV6: | |
1031 | skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6; | |
1032 | break; | |
1033 | default: | |
48856286 | 1034 | netdev_err(vif->dev, "Bad GSO type %d.\n", gso->u.gso.type); |
7376419a | 1035 | xenvif_fatal_tx_err(vif); |
f942dc25 IC |
1036 | return -EINVAL; |
1037 | } | |
1038 | ||
1039 | skb_shinfo(skb)->gso_size = gso->u.gso.size; | |
b89587a7 | 1040 | /* gso_segs will be calculated later */ |
f942dc25 IC |
1041 | |
1042 | return 0; | |
1043 | } | |
1044 | ||
2eba61d5 PD |
1045 | static int checksum_setup(struct xenvif *vif, struct sk_buff *skb) |
1046 | { | |
2721637c | 1047 | bool recalculate_partial_csum = false; |
2eba61d5 PD |
1048 | |
1049 | /* A GSO SKB must be CHECKSUM_PARTIAL. However some buggy | |
1050 | * peers can fail to set NETRXF_csum_blank when sending a GSO | |
1051 | * frame. In this case force the SKB to CHECKSUM_PARTIAL and | |
1052 | * recalculate the partial checksum. | |
1053 | */ | |
1054 | if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) { | |
1055 | vif->rx_gso_checksum_fixup++; | |
1056 | skb->ip_summed = CHECKSUM_PARTIAL; | |
2721637c | 1057 | recalculate_partial_csum = true; |
2eba61d5 PD |
1058 | } |
1059 | ||
1060 | /* A non-CHECKSUM_PARTIAL SKB does not require setup. */ | |
1061 | if (skb->ip_summed != CHECKSUM_PARTIAL) | |
1062 | return 0; | |
1063 | ||
2721637c | 1064 | return skb_checksum_setup(skb, recalculate_partial_csum); |
2eba61d5 PD |
1065 | } |
1066 | ||
f942dc25 IC |
1067 | static bool tx_credit_exceeded(struct xenvif *vif, unsigned size) |
1068 | { | |
059dfa6a WL |
1069 | u64 now = get_jiffies_64(); |
1070 | u64 next_credit = vif->credit_window_start + | |
f942dc25 IC |
1071 | msecs_to_jiffies(vif->credit_usec / 1000); |
1072 | ||
1073 | /* Timer could already be pending in rare cases. */ | |
1074 | if (timer_pending(&vif->credit_timeout)) | |
1075 | return true; | |
1076 | ||
1077 | /* Passed the point where we can replenish credit? */ | |
059dfa6a WL |
1078 | if (time_after_eq64(now, next_credit)) { |
1079 | vif->credit_window_start = now; | |
f942dc25 IC |
1080 | tx_add_credit(vif); |
1081 | } | |
1082 | ||
1083 | /* Still too big to send right now? Set a callback. */ | |
1084 | if (size > vif->remaining_credit) { | |
1085 | vif->credit_timeout.data = | |
1086 | (unsigned long)vif; | |
1087 | vif->credit_timeout.function = | |
1088 | tx_credit_callback; | |
1089 | mod_timer(&vif->credit_timeout, | |
1090 | next_credit); | |
059dfa6a | 1091 | vif->credit_window_start = next_credit; |
f942dc25 IC |
1092 | |
1093 | return true; | |
1094 | } | |
1095 | ||
1096 | return false; | |
1097 | } | |
1098 | ||
10574059 | 1099 | static unsigned xenvif_tx_build_gops(struct xenvif *vif, int budget) |
f942dc25 | 1100 | { |
b3f980bd | 1101 | struct gnttab_copy *gop = vif->tx_copy_ops, *request_gop; |
f942dc25 IC |
1102 | struct sk_buff *skb; |
1103 | int ret; | |
1104 | ||
121fa4b7 | 1105 | while (xenvif_tx_pending_slots_available(vif) && |
10574059 | 1106 | (skb_queue_len(&vif->tx_queue) < budget)) { |
f942dc25 | 1107 | struct xen_netif_tx_request txreq; |
37641494 | 1108 | struct xen_netif_tx_request txfrags[XEN_NETBK_LEGACY_SLOTS_MAX]; |
f942dc25 IC |
1109 | struct page *page; |
1110 | struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX-1]; | |
1111 | u16 pending_idx; | |
1112 | RING_IDX idx; | |
1113 | int work_to_do; | |
1114 | unsigned int data_len; | |
1115 | pending_ring_idx_t index; | |
1116 | ||
48856286 IC |
1117 | if (vif->tx.sring->req_prod - vif->tx.req_cons > |
1118 | XEN_NETIF_TX_RING_SIZE) { | |
1119 | netdev_err(vif->dev, | |
1120 | "Impossible number of requests. " | |
1121 | "req_prod %d, req_cons %d, size %ld\n", | |
1122 | vif->tx.sring->req_prod, vif->tx.req_cons, | |
1123 | XEN_NETIF_TX_RING_SIZE); | |
7376419a | 1124 | xenvif_fatal_tx_err(vif); |
48856286 IC |
1125 | continue; |
1126 | } | |
1127 | ||
d9601a36 | 1128 | work_to_do = RING_HAS_UNCONSUMED_REQUESTS(&vif->tx); |
b3f980bd WL |
1129 | if (!work_to_do) |
1130 | break; | |
f942dc25 IC |
1131 | |
1132 | idx = vif->tx.req_cons; | |
1133 | rmb(); /* Ensure that we see the request before we copy it. */ | |
1134 | memcpy(&txreq, RING_GET_REQUEST(&vif->tx, idx), sizeof(txreq)); | |
1135 | ||
1136 | /* Credit-based scheduling. */ | |
1137 | if (txreq.size > vif->remaining_credit && | |
b3f980bd WL |
1138 | tx_credit_exceeded(vif, txreq.size)) |
1139 | break; | |
f942dc25 IC |
1140 | |
1141 | vif->remaining_credit -= txreq.size; | |
1142 | ||
1143 | work_to_do--; | |
1144 | vif->tx.req_cons = ++idx; | |
1145 | ||
1146 | memset(extras, 0, sizeof(extras)); | |
1147 | if (txreq.flags & XEN_NETTXF_extra_info) { | |
7376419a WL |
1148 | work_to_do = xenvif_get_extras(vif, extras, |
1149 | work_to_do); | |
f942dc25 | 1150 | idx = vif->tx.req_cons; |
48856286 | 1151 | if (unlikely(work_to_do < 0)) |
b3f980bd | 1152 | break; |
f942dc25 IC |
1153 | } |
1154 | ||
7376419a | 1155 | ret = xenvif_count_requests(vif, &txreq, txfrags, work_to_do); |
48856286 | 1156 | if (unlikely(ret < 0)) |
b3f980bd | 1157 | break; |
48856286 | 1158 | |
f942dc25 IC |
1159 | idx += ret; |
1160 | ||
1161 | if (unlikely(txreq.size < ETH_HLEN)) { | |
1162 | netdev_dbg(vif->dev, | |
1163 | "Bad packet size: %d\n", txreq.size); | |
7376419a | 1164 | xenvif_tx_err(vif, &txreq, idx); |
b3f980bd | 1165 | break; |
f942dc25 IC |
1166 | } |
1167 | ||
1168 | /* No crossing a page as the payload mustn't fragment. */ | |
1169 | if (unlikely((txreq.offset + txreq.size) > PAGE_SIZE)) { | |
48856286 | 1170 | netdev_err(vif->dev, |
f942dc25 IC |
1171 | "txreq.offset: %x, size: %u, end: %lu\n", |
1172 | txreq.offset, txreq.size, | |
1173 | (txreq.offset&~PAGE_MASK) + txreq.size); | |
7376419a | 1174 | xenvif_fatal_tx_err(vif); |
b3f980bd | 1175 | break; |
f942dc25 IC |
1176 | } |
1177 | ||
b3f980bd WL |
1178 | index = pending_index(vif->pending_cons); |
1179 | pending_idx = vif->pending_ring[index]; | |
f942dc25 IC |
1180 | |
1181 | data_len = (txreq.size > PKT_PROT_LEN && | |
37641494 | 1182 | ret < XEN_NETBK_LEGACY_SLOTS_MAX) ? |
f942dc25 IC |
1183 | PKT_PROT_LEN : txreq.size; |
1184 | ||
1185 | skb = alloc_skb(data_len + NET_SKB_PAD + NET_IP_ALIGN, | |
1186 | GFP_ATOMIC | __GFP_NOWARN); | |
1187 | if (unlikely(skb == NULL)) { | |
1188 | netdev_dbg(vif->dev, | |
1189 | "Can't allocate a skb in start_xmit.\n"); | |
7376419a | 1190 | xenvif_tx_err(vif, &txreq, idx); |
f942dc25 IC |
1191 | break; |
1192 | } | |
1193 | ||
1194 | /* Packets passed to netif_rx() must have some headroom. */ | |
1195 | skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN); | |
1196 | ||
1197 | if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) { | |
1198 | struct xen_netif_extra_info *gso; | |
1199 | gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1]; | |
1200 | ||
7376419a WL |
1201 | if (xenvif_set_skb_gso(vif, skb, gso)) { |
1202 | /* Failure in xenvif_set_skb_gso is fatal. */ | |
f942dc25 | 1203 | kfree_skb(skb); |
b3f980bd | 1204 | break; |
f942dc25 IC |
1205 | } |
1206 | } | |
1207 | ||
1208 | /* XXX could copy straight to head */ | |
7376419a | 1209 | page = xenvif_alloc_page(vif, pending_idx); |
f942dc25 IC |
1210 | if (!page) { |
1211 | kfree_skb(skb); | |
7376419a | 1212 | xenvif_tx_err(vif, &txreq, idx); |
b3f980bd | 1213 | break; |
f942dc25 IC |
1214 | } |
1215 | ||
f942dc25 IC |
1216 | gop->source.u.ref = txreq.gref; |
1217 | gop->source.domid = vif->domid; | |
1218 | gop->source.offset = txreq.offset; | |
1219 | ||
1220 | gop->dest.u.gmfn = virt_to_mfn(page_address(page)); | |
1221 | gop->dest.domid = DOMID_SELF; | |
1222 | gop->dest.offset = txreq.offset; | |
1223 | ||
1224 | gop->len = txreq.size; | |
1225 | gop->flags = GNTCOPY_source_gref; | |
1226 | ||
1227 | gop++; | |
1228 | ||
b3f980bd | 1229 | memcpy(&vif->pending_tx_info[pending_idx].req, |
f942dc25 | 1230 | &txreq, sizeof(txreq)); |
b3f980bd | 1231 | vif->pending_tx_info[pending_idx].head = index; |
8f13dd96 | 1232 | XENVIF_TX_CB(skb)->pending_idx = pending_idx; |
f942dc25 IC |
1233 | |
1234 | __skb_put(skb, data_len); | |
1235 | ||
1236 | skb_shinfo(skb)->nr_frags = ret; | |
1237 | if (data_len < txreq.size) { | |
1238 | skb_shinfo(skb)->nr_frags++; | |
ea066ad1 IC |
1239 | frag_set_pending_idx(&skb_shinfo(skb)->frags[0], |
1240 | pending_idx); | |
f942dc25 | 1241 | } else { |
ea066ad1 IC |
1242 | frag_set_pending_idx(&skb_shinfo(skb)->frags[0], |
1243 | INVALID_PENDING_IDX); | |
f942dc25 IC |
1244 | } |
1245 | ||
b3f980bd | 1246 | vif->pending_cons++; |
f942dc25 | 1247 | |
7376419a | 1248 | request_gop = xenvif_get_requests(vif, skb, txfrags, gop); |
f942dc25 IC |
1249 | if (request_gop == NULL) { |
1250 | kfree_skb(skb); | |
7376419a | 1251 | xenvif_tx_err(vif, &txreq, idx); |
b3f980bd | 1252 | break; |
f942dc25 IC |
1253 | } |
1254 | gop = request_gop; | |
1255 | ||
b3f980bd | 1256 | __skb_queue_tail(&vif->tx_queue, skb); |
1e0b6eac | 1257 | |
f942dc25 | 1258 | vif->tx.req_cons = idx; |
f942dc25 | 1259 | |
b3f980bd | 1260 | if ((gop-vif->tx_copy_ops) >= ARRAY_SIZE(vif->tx_copy_ops)) |
f942dc25 IC |
1261 | break; |
1262 | } | |
1263 | ||
b3f980bd | 1264 | return gop - vif->tx_copy_ops; |
f942dc25 IC |
1265 | } |
1266 | ||
b3f980bd | 1267 | |
10574059 | 1268 | static int xenvif_tx_submit(struct xenvif *vif) |
f942dc25 | 1269 | { |
b3f980bd | 1270 | struct gnttab_copy *gop = vif->tx_copy_ops; |
f942dc25 | 1271 | struct sk_buff *skb; |
b3f980bd | 1272 | int work_done = 0; |
f942dc25 | 1273 | |
10574059 | 1274 | while ((skb = __skb_dequeue(&vif->tx_queue)) != NULL) { |
f942dc25 | 1275 | struct xen_netif_tx_request *txp; |
f942dc25 IC |
1276 | u16 pending_idx; |
1277 | unsigned data_len; | |
1278 | ||
8f13dd96 | 1279 | pending_idx = XENVIF_TX_CB(skb)->pending_idx; |
b3f980bd | 1280 | txp = &vif->pending_tx_info[pending_idx].req; |
f942dc25 IC |
1281 | |
1282 | /* Check the remap error code. */ | |
7376419a | 1283 | if (unlikely(xenvif_tx_check_gop(vif, skb, &gop))) { |
f942dc25 IC |
1284 | netdev_dbg(vif->dev, "netback grant failed.\n"); |
1285 | skb_shinfo(skb)->nr_frags = 0; | |
1286 | kfree_skb(skb); | |
1287 | continue; | |
1288 | } | |
1289 | ||
1290 | data_len = skb->len; | |
1291 | memcpy(skb->data, | |
b3f980bd | 1292 | (void *)(idx_to_kaddr(vif, pending_idx)|txp->offset), |
f942dc25 IC |
1293 | data_len); |
1294 | if (data_len < txp->size) { | |
1295 | /* Append the packet payload as a fragment. */ | |
1296 | txp->offset += data_len; | |
1297 | txp->size -= data_len; | |
1298 | } else { | |
1299 | /* Schedule a response immediately. */ | |
7376419a WL |
1300 | xenvif_idx_release(vif, pending_idx, |
1301 | XEN_NETIF_RSP_OKAY); | |
f942dc25 IC |
1302 | } |
1303 | ||
1304 | if (txp->flags & XEN_NETTXF_csum_blank) | |
1305 | skb->ip_summed = CHECKSUM_PARTIAL; | |
1306 | else if (txp->flags & XEN_NETTXF_data_validated) | |
1307 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
1308 | ||
7376419a | 1309 | xenvif_fill_frags(vif, skb); |
f942dc25 | 1310 | |
2eba61d5 | 1311 | if (skb_is_nonlinear(skb) && skb_headlen(skb) < PKT_PROT_LEN) { |
f942dc25 IC |
1312 | int target = min_t(int, skb->len, PKT_PROT_LEN); |
1313 | __pskb_pull_tail(skb, target - skb_headlen(skb)); | |
1314 | } | |
1315 | ||
1316 | skb->dev = vif->dev; | |
1317 | skb->protocol = eth_type_trans(skb, skb->dev); | |
f9ca8f74 | 1318 | skb_reset_network_header(skb); |
f942dc25 IC |
1319 | |
1320 | if (checksum_setup(vif, skb)) { | |
1321 | netdev_dbg(vif->dev, | |
1322 | "Can't setup checksum in net_tx_action\n"); | |
1323 | kfree_skb(skb); | |
1324 | continue; | |
1325 | } | |
1326 | ||
40893fd0 | 1327 | skb_probe_transport_header(skb, 0); |
f9ca8f74 | 1328 | |
b89587a7 PD |
1329 | /* If the packet is GSO then we will have just set up the |
1330 | * transport header offset in checksum_setup so it's now | |
1331 | * straightforward to calculate gso_segs. | |
1332 | */ | |
1333 | if (skb_is_gso(skb)) { | |
1334 | int mss = skb_shinfo(skb)->gso_size; | |
1335 | int hdrlen = skb_transport_header(skb) - | |
1336 | skb_mac_header(skb) + | |
1337 | tcp_hdrlen(skb); | |
1338 | ||
1339 | skb_shinfo(skb)->gso_segs = | |
1340 | DIV_ROUND_UP(skb->len - hdrlen, mss); | |
1341 | } | |
1342 | ||
f942dc25 IC |
1343 | vif->dev->stats.rx_bytes += skb->len; |
1344 | vif->dev->stats.rx_packets++; | |
1345 | ||
b3f980bd WL |
1346 | work_done++; |
1347 | ||
1348 | netif_receive_skb(skb); | |
f942dc25 | 1349 | } |
b3f980bd WL |
1350 | |
1351 | return work_done; | |
f942dc25 IC |
1352 | } |
1353 | ||
1354 | /* Called after netfront has transmitted */ | |
7376419a | 1355 | int xenvif_tx_action(struct xenvif *vif, int budget) |
f942dc25 IC |
1356 | { |
1357 | unsigned nr_gops; | |
b3f980bd | 1358 | int work_done; |
f942dc25 | 1359 | |
b3f980bd WL |
1360 | if (unlikely(!tx_work_todo(vif))) |
1361 | return 0; | |
1362 | ||
10574059 | 1363 | nr_gops = xenvif_tx_build_gops(vif, budget); |
f942dc25 IC |
1364 | |
1365 | if (nr_gops == 0) | |
b3f980bd WL |
1366 | return 0; |
1367 | ||
1368 | gnttab_batch_copy(vif->tx_copy_ops, nr_gops); | |
f942dc25 | 1369 | |
10574059 | 1370 | work_done = xenvif_tx_submit(vif); |
f942dc25 | 1371 | |
b3f980bd | 1372 | return work_done; |
f942dc25 IC |
1373 | } |
1374 | ||
7376419a WL |
1375 | static void xenvif_idx_release(struct xenvif *vif, u16 pending_idx, |
1376 | u8 status) | |
f942dc25 | 1377 | { |
f942dc25 | 1378 | struct pending_tx_info *pending_tx_info; |
2810e5b9 WL |
1379 | pending_ring_idx_t head; |
1380 | u16 peek; /* peek into next tx request */ | |
1381 | ||
b3f980bd | 1382 | BUG_ON(vif->mmap_pages[pending_idx] == (void *)(~0UL)); |
f942dc25 IC |
1383 | |
1384 | /* Already complete? */ | |
b3f980bd | 1385 | if (vif->mmap_pages[pending_idx] == NULL) |
f942dc25 IC |
1386 | return; |
1387 | ||
b3f980bd | 1388 | pending_tx_info = &vif->pending_tx_info[pending_idx]; |
f942dc25 | 1389 | |
2810e5b9 | 1390 | head = pending_tx_info->head; |
f942dc25 | 1391 | |
b3f980bd WL |
1392 | BUG_ON(!pending_tx_is_head(vif, head)); |
1393 | BUG_ON(vif->pending_ring[pending_index(head)] != pending_idx); | |
f942dc25 | 1394 | |
2810e5b9 WL |
1395 | do { |
1396 | pending_ring_idx_t index; | |
1397 | pending_ring_idx_t idx = pending_index(head); | |
b3f980bd | 1398 | u16 info_idx = vif->pending_ring[idx]; |
f942dc25 | 1399 | |
b3f980bd | 1400 | pending_tx_info = &vif->pending_tx_info[info_idx]; |
2810e5b9 WL |
1401 | make_tx_response(vif, &pending_tx_info->req, status); |
1402 | ||
1403 | /* Setting any number other than | |
1404 | * INVALID_PENDING_RING_IDX indicates this slot is | |
1405 | * starting a new packet / ending a previous packet. | |
1406 | */ | |
1407 | pending_tx_info->head = 0; | |
1408 | ||
b3f980bd WL |
1409 | index = pending_index(vif->pending_prod++); |
1410 | vif->pending_ring[index] = vif->pending_ring[info_idx]; | |
f942dc25 | 1411 | |
b3f980bd | 1412 | peek = vif->pending_ring[pending_index(++head)]; |
2810e5b9 | 1413 | |
b3f980bd | 1414 | } while (!pending_tx_is_head(vif, peek)); |
2810e5b9 | 1415 | |
b3f980bd WL |
1416 | put_page(vif->mmap_pages[pending_idx]); |
1417 | vif->mmap_pages[pending_idx] = NULL; | |
f942dc25 IC |
1418 | } |
1419 | ||
2810e5b9 | 1420 | |
f942dc25 IC |
1421 | static void make_tx_response(struct xenvif *vif, |
1422 | struct xen_netif_tx_request *txp, | |
1423 | s8 st) | |
1424 | { | |
1425 | RING_IDX i = vif->tx.rsp_prod_pvt; | |
1426 | struct xen_netif_tx_response *resp; | |
1427 | int notify; | |
1428 | ||
1429 | resp = RING_GET_RESPONSE(&vif->tx, i); | |
1430 | resp->id = txp->id; | |
1431 | resp->status = st; | |
1432 | ||
1433 | if (txp->flags & XEN_NETTXF_extra_info) | |
1434 | RING_GET_RESPONSE(&vif->tx, ++i)->status = XEN_NETIF_RSP_NULL; | |
1435 | ||
1436 | vif->tx.rsp_prod_pvt = ++i; | |
1437 | RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&vif->tx, notify); | |
1438 | if (notify) | |
e1f00a69 | 1439 | notify_remote_via_irq(vif->tx_irq); |
f942dc25 IC |
1440 | } |
1441 | ||
1442 | static struct xen_netif_rx_response *make_rx_response(struct xenvif *vif, | |
1443 | u16 id, | |
1444 | s8 st, | |
1445 | u16 offset, | |
1446 | u16 size, | |
1447 | u16 flags) | |
1448 | { | |
1449 | RING_IDX i = vif->rx.rsp_prod_pvt; | |
1450 | struct xen_netif_rx_response *resp; | |
1451 | ||
1452 | resp = RING_GET_RESPONSE(&vif->rx, i); | |
1453 | resp->offset = offset; | |
1454 | resp->flags = flags; | |
1455 | resp->id = id; | |
1456 | resp->status = (s16)size; | |
1457 | if (st < 0) | |
1458 | resp->status = (s16)st; | |
1459 | ||
1460 | vif->rx.rsp_prod_pvt = ++i; | |
1461 | ||
1462 | return resp; | |
1463 | } | |
1464 | ||
b3f980bd | 1465 | static inline int rx_work_todo(struct xenvif *vif) |
f942dc25 | 1466 | { |
9ab9831b ZK |
1467 | return !skb_queue_empty(&vif->rx_queue) && |
1468 | xenvif_rx_ring_slots_available(vif, vif->rx_last_skb_slots); | |
f942dc25 IC |
1469 | } |
1470 | ||
b3f980bd | 1471 | static inline int tx_work_todo(struct xenvif *vif) |
f942dc25 IC |
1472 | { |
1473 | ||
b3f980bd | 1474 | if (likely(RING_HAS_UNCONSUMED_REQUESTS(&vif->tx)) && |
121fa4b7 | 1475 | xenvif_tx_pending_slots_available(vif)) |
f942dc25 IC |
1476 | return 1; |
1477 | ||
1478 | return 0; | |
1479 | } | |
1480 | ||
7376419a | 1481 | void xenvif_unmap_frontend_rings(struct xenvif *vif) |
f942dc25 | 1482 | { |
c9d63699 DV |
1483 | if (vif->tx.sring) |
1484 | xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif), | |
1485 | vif->tx.sring); | |
1486 | if (vif->rx.sring) | |
1487 | xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif), | |
1488 | vif->rx.sring); | |
f942dc25 IC |
1489 | } |
1490 | ||
7376419a WL |
1491 | int xenvif_map_frontend_rings(struct xenvif *vif, |
1492 | grant_ref_t tx_ring_ref, | |
1493 | grant_ref_t rx_ring_ref) | |
f942dc25 | 1494 | { |
c9d63699 | 1495 | void *addr; |
f942dc25 IC |
1496 | struct xen_netif_tx_sring *txs; |
1497 | struct xen_netif_rx_sring *rxs; | |
1498 | ||
1499 | int err = -ENOMEM; | |
1500 | ||
c9d63699 DV |
1501 | err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(vif), |
1502 | tx_ring_ref, &addr); | |
1503 | if (err) | |
f942dc25 IC |
1504 | goto err; |
1505 | ||
c9d63699 | 1506 | txs = (struct xen_netif_tx_sring *)addr; |
f942dc25 IC |
1507 | BACK_RING_INIT(&vif->tx, txs, PAGE_SIZE); |
1508 | ||
c9d63699 DV |
1509 | err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(vif), |
1510 | rx_ring_ref, &addr); | |
1511 | if (err) | |
f942dc25 | 1512 | goto err; |
f942dc25 | 1513 | |
c9d63699 | 1514 | rxs = (struct xen_netif_rx_sring *)addr; |
f942dc25 IC |
1515 | BACK_RING_INIT(&vif->rx, rxs, PAGE_SIZE); |
1516 | ||
1517 | return 0; | |
1518 | ||
1519 | err: | |
7376419a | 1520 | xenvif_unmap_frontend_rings(vif); |
f942dc25 IC |
1521 | return err; |
1522 | } | |
1523 | ||
ca2f09f2 PD |
1524 | void xenvif_stop_queue(struct xenvif *vif) |
1525 | { | |
1526 | if (!vif->can_queue) | |
1527 | return; | |
1528 | ||
1529 | netif_stop_queue(vif->dev); | |
1530 | } | |
1531 | ||
1532 | static void xenvif_start_queue(struct xenvif *vif) | |
1533 | { | |
1534 | if (xenvif_schedulable(vif)) | |
1535 | netif_wake_queue(vif->dev); | |
1536 | } | |
1537 | ||
121fa4b7 | 1538 | int xenvif_kthread_guest_rx(void *data) |
b3f980bd WL |
1539 | { |
1540 | struct xenvif *vif = data; | |
ca2f09f2 | 1541 | struct sk_buff *skb; |
b3f980bd WL |
1542 | |
1543 | while (!kthread_should_stop()) { | |
1544 | wait_event_interruptible(vif->wq, | |
1545 | rx_work_todo(vif) || | |
1546 | kthread_should_stop()); | |
1547 | if (kthread_should_stop()) | |
1548 | break; | |
1549 | ||
ca2f09f2 | 1550 | if (!skb_queue_empty(&vif->rx_queue)) |
7376419a | 1551 | xenvif_rx_action(vif); |
b3f980bd | 1552 | |
ca2f09f2 PD |
1553 | if (skb_queue_empty(&vif->rx_queue) && |
1554 | netif_queue_stopped(vif->dev)) | |
1555 | xenvif_start_queue(vif); | |
1556 | ||
b3f980bd WL |
1557 | cond_resched(); |
1558 | } | |
1559 | ||
ca2f09f2 PD |
1560 | /* Bin any remaining skbs */ |
1561 | while ((skb = skb_dequeue(&vif->rx_queue)) != NULL) | |
1562 | dev_kfree_skb(skb); | |
1563 | ||
b3f980bd WL |
1564 | return 0; |
1565 | } | |
1566 | ||
f942dc25 IC |
1567 | static int __init netback_init(void) |
1568 | { | |
f942dc25 | 1569 | int rc = 0; |
f942dc25 | 1570 | |
2a14b244 | 1571 | if (!xen_domain()) |
f942dc25 IC |
1572 | return -ENODEV; |
1573 | ||
37641494 | 1574 | if (fatal_skb_slots < XEN_NETBK_LEGACY_SLOTS_MAX) { |
383eda32 JP |
1575 | pr_info("fatal_skb_slots too small (%d), bump it to XEN_NETBK_LEGACY_SLOTS_MAX (%d)\n", |
1576 | fatal_skb_slots, XEN_NETBK_LEGACY_SLOTS_MAX); | |
37641494 | 1577 | fatal_skb_slots = XEN_NETBK_LEGACY_SLOTS_MAX; |
2810e5b9 WL |
1578 | } |
1579 | ||
f942dc25 IC |
1580 | rc = xenvif_xenbus_init(); |
1581 | if (rc) | |
1582 | goto failed_init; | |
1583 | ||
1584 | return 0; | |
1585 | ||
1586 | failed_init: | |
f942dc25 | 1587 | return rc; |
f942dc25 IC |
1588 | } |
1589 | ||
1590 | module_init(netback_init); | |
1591 | ||
b103f358 WL |
1592 | static void __exit netback_fini(void) |
1593 | { | |
b103f358 | 1594 | xenvif_xenbus_fini(); |
b103f358 WL |
1595 | } |
1596 | module_exit(netback_fini); | |
1597 | ||
f942dc25 | 1598 | MODULE_LICENSE("Dual BSD/GPL"); |
f984cec6 | 1599 | MODULE_ALIAS("xen-backend:vif"); |